Turbomole

Program Package for ab initio

Electronic Structure Calculations

USER’S MANUAL

Turbomole Version 7.6

September 27, 2021
Contents

1 Preface and General Information 15

1.1 Contributions and Acknowledgements . . . . . . . . . . . . . . . . . 15
1.2 Features of Turbomole . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3 How to Quote Usage of Turbomole . . . . . . . . . . . . . . . . . . 17
1.4 Modules and Their Functionality . . . . . . . . . . . . . . . . . . . . 33
1.5 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2 Installation of Turbomole 42
2.1 Install TURBOMOLE command line version . . . . . . . . . . . . . . . . 42
2.1.1 Settings for each user: . . . . . . . . . . . . . . . . . . . . . . 42
2.1.2 Setting system type and $PATH by hand . . . . . . . . . . . 43
2.1.3 Testing the installation . . . . . . . . . . . . . . . . . . . . . . 44
2.2 Installation problems: How to solve . . . . . . . . . . . . . . . . . . . 44

3 How to Run Turbomole 47

3.1 Simple input files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2 The graphical user interface TmoleX . . . . . . . . . . . . . . . . . . 49
3.3 A ‘Quick and Dirty’ Tutorial for the define input generator . . . . . 49
3.3.1 Single Point Calculations: Running Turbomole Modules . . 51
3.3.2 Energy and Gradient Calculations . . . . . . . . . . . . . . . 52
3.3.3 Calculation of Molecular Properties . . . . . . . . . . . . . . . 53
3.3.4 Modules and Data Flow . . . . . . . . . . . . . . . . . . . . . 53
3.4 Parallel Runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.1 Running Parallel Jobs — MPI case . . . . . . . . . . . . . . . 56
3.4.2 Running Parallel Jobs — SMP case . . . . . . . . . . . . . . . 62

3
4 CONTENTS

4 Preparing your input file with Define 67

4.0.3 Universally Available Display Commands in Define . . . . . 68
4.0.4 Specifying Atomic Sets . . . . . . . . . . . . . . . . . . . . . . 68
4.0.5 control as Input and Output File . . . . . . . . . . . . . . . 68
4.0.6 Be Prepared . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.1 The Geometry Main Menu . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1.1 Description of commands . . . . . . . . . . . . . . . . . . . . 72
4.1.2 Internal Coordinate Menu . . . . . . . . . . . . . . . . . . . . 75
4.1.3 Manipulating the Geometry . . . . . . . . . . . . . . . . . . . 80
4.2 The Atomic Attributes Menu . . . . . . . . . . . . . . . . . . . . . . 81
4.2.1 Description of the commands . . . . . . . . . . . . . . . . . . 84
4.3 Generating MO Start Vectors . . . . . . . . . . . . . . . . . . . . . . 86
4.3.1 The MO Start Vectors Menu . . . . . . . . . . . . . . . . . . 86
4.3.2 Assignment of Occupation Numbers . . . . . . . . . . . . . . 89
4.3.3 Orbital Specification Menu . . . . . . . . . . . . . . . . . . . 91
4.3.4 Roothaan Parameters . . . . . . . . . . . . . . . . . . . . . . 91
4.3.5 Start-MOs for broken symmetry treatments ("flip") . . . . . . 92
4.4 The General Options Menu . . . . . . . . . . . . . . . . . . . . . . . 94
4.4.1 Important commands . . . . . . . . . . . . . . . . . . . . . . 95
4.4.2 Special adjustments . . . . . . . . . . . . . . . . . . . . . . . 102
4.4.3 Relax Options . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.4.4 Definition of External Electrostatic Fields . . . . . . . . . . . 110
4.4.5 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

5 Calculation of Molecular Structure and Ab Initio Molecular Dy-

namics 121
5.1 Structure Optimizations using the Jobex Script . . . . . . . . . . . 121
5.1.1 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.1.2 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.2 Program Statpt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.1 General Information . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.2 Hessian matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 124
CONTENTS 5

5.2.3 Finding Minima . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.2.4 Finding transition states . . . . . . . . . . . . . . . . . . . . . 125
5.3 Program Relax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.3.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.3.2 Optimization of General Coordinates . . . . . . . . . . . . . . 126
5.3.3 Force Constant Update Algorithms . . . . . . . . . . . . . . . 128
5.3.4 Definition of Internal Coordinates . . . . . . . . . . . . . . . . 130
5.3.5 Structure Optimizations Using Internal Coordinates . . . . . 130
5.3.6 Structure Optimization in Cartesian Coordinates . . . . . . . 131
5.3.7 Optimization of Basis Sets (SCF only) . . . . . . . . . . . . . 132
5.3.8 Simultaneous Optimization of Basis Set and Structure . . . . 132
5.3.9 Optimization of Structure and a Global Scaling Factor . . . . 132
5.3.10 Conversion from Internal to Cartesian Coordinates . . . . . . 133
5.3.11 Conversion of Cartesian Coordinates, Gradients and Force Con-
stants to Internals . . . . . . . . . . . . . . . . . . . . . . . . 133
5.3.12 The m-Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.3.13 Initialization of Force Constant Matrices . . . . . . . . . . . . 134
5.3.14 Look at Results . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.4 Force Field Calculations . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.4.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.4.2 How to Perform a Uff Calculation . . . . . . . . . . . . . . . 135
5.4.3 The Uff implementation . . . . . . . . . . . . . . . . . . . . 136
5.5 Semiempirical Extended Tight-Binding Calculations . . . . . . . . . 138
5.5.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.5.2 How to Perform a xTB Calculation . . . . . . . . . . . . . . 138
5.6 Molecular Dynamics Calculations . . . . . . . . . . . . . . . . . . . . 139
5.7 Global Structure Optimization – The DoDo Program . . . . . . . . 139
5.7.1 Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 139
5.7.2 How to Perform . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.7.3 The DoDo Input File . . . . . . . . . . . . . . . . . . . . . . 142
5.8 Counterpoise-Corrections using the Jobbsse Script . . . . . . . . . . 145
5.8.1 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6 CONTENTS

5.8.2 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

5.9 Reaction Path Optimization . . . . . . . . . . . . . . . . . . . . . . . 148
5.9.1 Background and Program structure . . . . . . . . . . . . . . . 148
5.9.2 Input Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.9.3 How it works . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

6 Hartree–Fock and DFT Calculations for Molecular Systems 152

6.1 Background Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.2 Exchange-Correlation Functionals Available . . . . . . . . . . . . . . 155
6.2.1 Exchange-Correlation Functionals from LibXC library . . . . 160
6.2.2 Exchange-Correlation Functionals from XCFun library . . . . 162
6.3 Restricted Open-Shell Hartree–Fock . . . . . . . . . . . . . . . . . . 167
6.3.1 Brief Description . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.3.2 One Open Shell . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.3.3 More Than One Open Shell . . . . . . . . . . . . . . . . . . . 170
6.3.4 Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.4 Relativistic effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6.4.1 One- and two-component relativistic methods . . . . . . . . . 174
6.4.2 How to use . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6.5 Finite magnetic fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.6 Dispersion Correction for DFT Calculations . . . . . . . . . . . . . . 180
6.7 Energy Decomposition Analysis (EDA) . . . . . . . . . . . . . . . . . 183
6.7.1 How to perform . . . . . . . . . . . . . . . . . . . . . . . . . . 183

7 DFT Calculations for Molecular and Periodic Systems 185

7.1 Functionalities of Riper . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . 186
7.2.1 Kohn-Sham DFT for Molecular and Periodic Systems . . . . 186
7.2.2 RI-CFMM Approach . . . . . . . . . . . . . . . . . . . . . . . 187
7.2.3 k Point Sampling Scheme . . . . . . . . . . . . . . . . . . . . 188
7.2.4 Metals and Semiconductors: Gaussian Smearing . . . . . . . 189
7.2.5 Low-Memory Iterative Density Fitting Method . . . . . . . . 189
7.2.6 RT-TDDFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
CONTENTS 7

7.3 How to Perform a Calculation . . . . . . . . . . . . . . . . . . . . . . 192

7.3.1 Basis Sets for Periodic Calculations . . . . . . . . . . . . . . . 192
7.3.2 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.3.3 Creating the Input File . . . . . . . . . . . . . . . . . . . . . 192
7.3.4 Single Point Energy and Gradient . . . . . . . . . . . . . . . 196
7.3.5 Structure Optimization . . . . . . . . . . . . . . . . . . . . . 197
7.3.6 Optimization of Cell Parameters . . . . . . . . . . . . . . . . 197
7.3.7 Band Structure Plots . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.8 Calculation of Densities and MOs on Grids . . . . . . . . . . 198
7.3.9 Density of States . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.10 RT-TDDFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

8 Hartree–Fock and DFT Response Calculations: Stability, Dynamic

Response Properties, and Excited States 206
8.1 Functionalities of Escf and Egrad . . . . . . . . . . . . . . . . . . . . 206
8.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.4 How to Perform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.4.2 Polarizabilities and Optical Rotations . . . . . . . . . . . . . 213
8.4.3 Damped Response Calculations . . . . . . . . . . . . . . . . . 214
8.4.4 Dynamic First Hyperpolarizability . . . . . . . . . . . . . . . 215
8.4.5 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.4.6 Vertical Excitation and CD Spectra . . . . . . . . . . . . . . 216
8.4.7 Two-photon absorption . . . . . . . . . . . . . . . . . . . . . . 219
8.4.8 Excited State Geometry Optimizations . . . . . . . . . . . . . 220
8.4.9 Excited State Force Constant Calculations . . . . . . . . . . . 221
8.4.10 Polarizability Derivatives and Raman Spectra . . . . . . . . . 221
8.4.11 State-to-state properties . . . . . . . . . . . . . . . . . . . . . 221
8.4.12 Nuclear spin-spin coupling constants . . . . . . . . . . . . . . 223
8.4.13 Prediciting colors using the Color Prediction Tool cpt . . . . 223
8.4.14 Approximations for Coulomb and Exchange integrals . . . . . 223
8 CONTENTS

9 Second-order Møller–Plesset Perturbation Theory 225

9.1 Functionalities of mpgrad, ricc2, and pnoccsd . . . . . . . . . . . . . 225
9.1.1 How to quote . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
9.2 Some Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
9.3 How to Prepare and Perform MP2 Calculations . . . . . . . . . . . . 229
9.4 General Comments on MP2 Calculations, Practical Hints . . . . . . 232
9.5 RI-MP2-F12 Calculations . . . . . . . . . . . . . . . . . . . . . . . . 234
9.6 LT-SOS-RI-MP2 with O(N 4 ) scaling costs . . . . . . . . . . . . . . . 239
9.7 COSMO-MP2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
9.8 OSV-PNO-MP2 and OSV-PNO-MP2-F12 calculations . . . . . . . . 241

10 Second-Order Approximate Coupled-Cluster (CC2) Calculations 243

10.1 CC2 Ground-State Energy Calculations . . . . . . . . . . . . . . . . 248
10.2 Calculation of Excitation Energies . . . . . . . . . . . . . . . . . . . 250
10.3 First-Order Properties and Gradients . . . . . . . . . . . . . . . . . . 255
10.3.1 Ground State Properties, Gradients and Geometries . . . . . 255
10.3.2 Excited State Properties, Gradients and Geometries . . . . . 257
10.3.3 Visualization of densities and Density analysis . . . . . . . . . 260
10.3.4 Fast geometry optimizations with RI-SCF based gradients . . 261
10.4 Transition Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
10.4.1 Ground to excited state transition moments . . . . . . . . . . 262
10.4.2 Transition moments between excited states . . . . . . . . . . 263
10.4.3 Ground to excited state two-photon transition moments . . . 264
10.4.4 Phosphorescence lifetimes using SOC-PT-CC2 . . . . . . . . . 265
10.5 Ground State Second-order Properties with MP2 and CC2 . . . . . . 266
10.6 Parallel RI-MP2 and RI-CC2 Calculations . . . . . . . . . . . . . . . 267
10.7 Spin-component scaling approaches (SCS/SOS) . . . . . . . . . . . . 267

11 CCSD, CCSD(F12*) and CCSD(T) calculations 270

11.1 Characteristics of the Implementation and Computational Demands 272

12 PNO-based CCSD and CCSD(T) calculations 281

12.1 Selection Thresholds used in the pnoccsd Program . . . . . . . . . . 283
CONTENTS 9

12.2 Characteristics of the Implementation . . . . . . . . . . . . . . . . . 284

12.2.1 Strong pair approximation . . . . . . . . . . . . . . . . . . . . 285

13 Random Phase Approximation Calculations: Energy and First-

Order Properties 286
13.1 Ground State Energy Theory . . . . . . . . . . . . . . . . . . . . . . 287
13.2 Gradients Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
13.3 Generarlized Kohn Sham scheme for RIRPA . . . . . . . . . . . . . . 290
13.4 Further Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 292
13.5 Comments on the Output . . . . . . . . . . . . . . . . . . . . . . . . 293

14 Many body perturbation theory in the GW approximation 295

14.1 Single particle spectra based on the GW approximation . . . . . . . 295
14.1.1 Theoretical background . . . . . . . . . . . . . . . . . . . . . 295
14.1.2 GW features . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
14.1.3 General recipe for G0 W0 , RI-AC-G0 W0 and RI-CD-G0 W0 cal-
culations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
14.2 Excitation energies from BSE . . . . . . . . . . . . . . . . . . . . . . 299
14.2.1 Theoretical background . . . . . . . . . . . . . . . . . . . . . 299
14.2.2 BSE features . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
14.2.3 General recipe for a BSE calculation . . . . . . . . . . . . . . 302

15 Calculation of Vibrational Frequencies and Vibrational Spectra 303

15.1 Analysis of Normal Modes in Terms of Internal Coordinates . . . . . 305
15.2 Calculation of Raman Spectra . . . . . . . . . . . . . . . . . . . . . . 306
15.3 Calculation of VCD Spectra . . . . . . . . . . . . . . . . . . . . . . . 306
15.4 Vibrational frequencies with fixed atoms using NumForce . . . . . . 307
15.5 Interface to hotFCHT . . . . . . . . . . . . . . . . . . . . . . . . . . 308

16 First order electron-vibration coupling 309

16.1 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . . . 309
16.2 evib features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
16.3 General usage of evib . . . . . . . . . . . . . . . . . . . . . . . . . . 310
10 CONTENTS

17 Calculation of NMR Shieldings 311

17.1 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
17.2 How to Perform a SCF or DFT Calculation . . . . . . . . . . . . . . 312
17.3 How to Perform a MP2 calculation . . . . . . . . . . . . . . . . . . . 312
17.4 Chemical Shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
17.5 Other Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
17.6 Known Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314

18 Embedding and Solvation Effects 315

18.1 Charge and multipole embedding . . . . . . . . . . . . . . . . . . . . 315
18.2 Treatment of Solvation Effects with Cosmo . . . . . . . . . . . . . . 317
18.2.1 Iterative COSMO-MP2 . . . . . . . . . . . . . . . . . . . . . 319
18.2.2 COSMO-CC2 for ground-state calculations . . . . . . . . . . 320
18.2.3 Vertical excitations and Polarizabilities for TDDFT, TDA and
RPA: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
18.2.4 The Direct COSMO-RS method (DCOSMO-RS): . . . . . . . 320
18.2.5 Solvation effects on excited states using COSMO in ricc2: . . 322
18.3 Frozen Density Embedding calculations . . . . . . . . . . . . . . . . 325
18.3.1 Background Theory . . . . . . . . . . . . . . . . . . . . . . . 325
18.3.2 Frozen Density Embedding calculations using the FDE script 326
18.3.3 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
18.3.4 FDE with hybrid and orbital-dependent functionals . . . . . . 335
18.4 Periodic Electrostatic Embedded Cluster Method . . . . . . . . . . . 336
18.4.1 General Information . . . . . . . . . . . . . . . . . . . . . . . 336
18.4.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . 336
18.4.3 Calculation Setup . . . . . . . . . . . . . . . . . . . . . . . . . 337
18.5 Polarizable embedding calculations . . . . . . . . . . . . . . . . . . . 344
18.5.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
18.5.2 Computational details: SCF calculations . . . . . . . . . . . . 345
18.5.3 Computational details for post-SCF methods . . . . . . . . . 348

19 Molecular Properties, Wavefunction Analysis, and Interfaces to Vi-

sualization Tools 350
CONTENTS 11

19.1 Molecular Properties, Wavefunction Analysis, and Localized Orbitals 350

19.1.1 Selection of densities . . . . . . . . . . . . . . . . . . . . . . . 351
19.1.2 Electrostatic moments . . . . . . . . . . . . . . . . . . . . . . 351
19.1.3 Relativistic corrections . . . . . . . . . . . . . . . . . . . . . . 351
19.1.4 Population analyses . . . . . . . . . . . . . . . . . . . . . . . 352
19.1.5 Generation of localized MOs . . . . . . . . . . . . . . . . . . . 352
19.1.6 Intrinsic Bond Orbitals Analysis . . . . . . . . . . . . . . . . 354
19.2 Interfaces to Visualization Tools . . . . . . . . . . . . . . . . . . . . . 356

20 Orbital Dependent Kohn-Sham Density Functional Theory 363

20.1 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . 363
20.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
20.2.1 OEP-EXX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
20.2.2 LHF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
20.3 How to Perform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
20.4 How to plot the exchange potential . . . . . . . . . . . . . . . . . . . 370
20.5 How to quote . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371

21 Vibronic absorption and emission spectra 372

21.1 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . 372
21.1.1 Vibronic spectra at zero temperature . . . . . . . . . . . . . . 372
21.1.2 Single Vibronic Level Spectra . . . . . . . . . . . . . . . . . . 373
21.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
21.3 Functionalities of Radless . . . . . . . . . . . . . . . . . . . . . . . . 374
21.3.1 How to use radless . . . . . . . . . . . . . . . . . . . . . . . 374
21.3.2 Output of radless . . . . . . . . . . . . . . . . . . . . . . . . 376

22 Keywords in the control file 379

22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
22.2 Format of Keywords and Comments . . . . . . . . . . . . . . . . . . 379
22.2.1 Keywords for System Specification . . . . . . . . . . . . . . . 379
22.2.2 Keyword for the General Memory Specification . . . . . . . . 381
22.2.3 Other General Keywords . . . . . . . . . . . . . . . . . . . . . 382
12 CONTENTS

22.2.4 Keywords for redundant internal coordinates in $redund_inp 383

22.2.5 Keywords for Module uff . . . . . . . . . . . . . . . . . . . . 385
22.2.6 Keywords for Module tb . . . . . . . . . . . . . . . . . . . . . 389
22.2.7 Keywords for Module woelfling . . . . . . . . . . . . . . . . 390
22.2.8 Keywords for Modules dscf and ridft . . . . . . . . . . . . . 391
22.2.9 Keywords for Point Charge Embedding . . . . . . . . . . . . 414
22.2.10 Keywords for Periodic Electrostatic Embedded Cluster Method 414
22.2.11 Keywords for COSMO . . . . . . . . . . . . . . . . . . . . . . 416
22.2.12 Keywords for Module riper . . . . . . . . . . . . . . . . . . . 422
22.2.13 Keywords for Modules grad and rdgrad . . . . . . . . . . . . 429
22.2.14 Keywords for Module aoforce . . . . . . . . . . . . . . . . . 430
22.2.15 Keywords for Module evib . . . . . . . . . . . . . . . . . . . 433
22.2.16 Keywords for Module escf . . . . . . . . . . . . . . . . . . . 433
22.2.17 Keywords for Module rirpa . . . . . . . . . . . . . . . . . . . 442
22.2.18 Keywords for Module egrad . . . . . . . . . . . . . . . . . . . 443
22.2.19 Keywords for Module mpgrad . . . . . . . . . . . . . . . . . . 444
22.2.20 Keywords for Module ricc2 . . . . . . . . . . . . . . . . . . . 445
22.2.21 Keywords for Module ccsdf12 . . . . . . . . . . . . . . . . . 457
22.2.22 Keywords for Module pnoccsd . . . . . . . . . . . . . . . . . 458
22.2.23 Keywords for Module relax . . . . . . . . . . . . . . . . . . . 462
22.2.24 Keywords for Module statpt . . . . . . . . . . . . . . . . . . 470
22.2.25 Keywords for Module moloch . . . . . . . . . . . . . . . . . . 471
22.2.26 Keywords for wave function analysis and generation of plotting
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
22.2.27 Keywords for Module frog . . . . . . . . . . . . . . . . . . . 484
22.2.28 Keywords for Module mpshift . . . . . . . . . . . . . . . . . 490
22.2.29 Keywords for Parallel Runs . . . . . . . . . . . . . . . . . . . 493

23 Sample control files 497

23.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
23.2 NH3 Input for a RHF Calculation . . . . . . . . . . . . . . . . . . . . 498
23.3 NO2 input for an unrestricted DFT calculation . . . . . . . . . . . . 502
CONTENTS 13

23.4 TaCl5 Input for an RI-DFT Calculation with ECPs . . . . . . . . . . 506

23.5 Basisset optimization for Nitrogen . . . . . . . . . . . . . . . . . . . 513
23.6 ROHF of Two Open Shells . . . . . . . . . . . . . . . . . . . . . . . . 516

24 The Perl-based Test Suite Structure 519

24.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
24.2 Running the tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
24.3 Taking the timings and benchmarking . . . . . . . . . . . . . . . . . 521
24.4 Modes and options of the TTEST script . . . . . . . . . . . . . . . . 521

Bibliography 525

Index 547
14 CONTENTS
Chapter 1

Preface and General Information

1.1 Contributions and Acknowledgements

TURBOMOLE [1] is a development of University of Karlsruhe and Forschungszentrum

Karlsruhe GmbH 1989-2007, TURBOMOLE GmbH, since 2007. The following peo-
ple have made contributions:
Reinhart Ahlrichs, Markus Klaus Armbruster, Rafał A. Bachorz, Hilke
Bahmann, Alexander Baldes, Michael Bär, Hans–Peter Baron, Rüdi-
ger Bauernschmitt, Florian A. Bischoff, Stephan Böcker, Asbjörn M.
Burow, Guo P. Chen, Nathan Crawford, Peter Deglmann, Fabio Della
Sala, Michael Diedenhofen, Sebastian Ehlert, Michael Ehrig, Karin Eich-
korn, Simon Elliott, Yannick J. Franzke, Daniel Friese, Filipp Furche,
James Furness, Sree Ganesh Balasubramani, Tino Gimon, Andreas Glöß,
Nora Graf, Lukáš Grajciar, Robin Grotjahn, Frank Haase, Marco Häser,
Christof Hättig, Arnim Hellweg, Benjamin Helmich, Sebastian Höfener,
Christof Holzer, Hans Horn, Christian Huber, Waldemar Hujo, Uwe
Huniar, Aaron D. Kaplan, Marco Kattannek, Max Kehry, Sascha Kla-
wohn, Wim Klopper, Andreas Köhn, Christoph Kölmel, Markus Koll-
witz, Katharina Krause, Michael Kühn, Roman Łazarski, Toni M. Maier,
Fabian Mack, Klaus May, Nils Middendorf, Paola Nava, Christian Ochsen-
feld, Holger Öhm, Mathias Pabst, Shane M. Parker, Holger Patzelt, Ans-
gar Pausch, Patrik Pollak, Dmitrij Rappoport, Kevin Reiter, Saswata
Roy, Oliver Rubner, Ansgar Schäfer, Gunnar Schmitz, Uwe Schneider,
Tobias Schwabe, Marek Sierka, Jianwei Sun, David P. Tew, Oliver Treut-
ler, Barbara Unterreiner, Malte von Arnim, Vamsee K. Voora, Florian
Weigend, Patrick Weis, Horst Weiss, Nina Winter

15
16 CHAPTER 1. PREFACE AND GENERAL INFORMATION

We acknowledge help from

• Michael Dolg, University of Stuttgart, now: University of Cologne

• Jürgen Gauss, University of Mainz

• Christoph van Wüllen, University of Bochum, now: TU Kaiserslautern

• Stefan Grimme, University of Bonn

• Stefan Brode, BASF AG, Ludwigshafen

• Heinz Schiffer, HOECHST AG, Frankfurt

• the research groups of Ove Christiansen, Aarhus University, and Jacob Kong-
sted, University of Southern Denmark

and financial support by the University of Karlsruhe, BASF AG, BAYER AG,
HOECHST AG, the German Research Foundation (Deutsche Forschungsgemein-
schaft, DFG), Carl-Zeiss-Stiftung and the "Fonds der Chemischen Industrie".
Contact address:
Turbomole GmbH
Litzenhardtstrasse 19
76135 Karlsruhe
Germany
E-mail: support@turbomole.com
Web: https://www.turbomole.org
Support E-mail: support@turbomole.com
1.2. FEATURES OF TURBOMOLE 17

1.2 Features of Turbomole

TURBOMOLE has been specially designed for UNIX workstations and PCs and efficiently
exploits the capabilities of this type of hardware. TURBOMOLE consists of a series of
modules; their use is facilitated by various tools.
Outstanding features of TURBOMOLE are

• semi-direct algorithms with adjustable main memory and disk space require-
ments

• full use of all point groups

• efficient integral evaluation

• stable and accurate grids for numerical integration

• low memory and disk space requirements

1.3 How to Quote Usage of Turbomole

Please quote the usage of the program package under consideration of the version
number:

TURBOMOLE V7.5 2020, a development of University of Karlsruhe and

Forschungszentrum Karlsruhe GmbH, 1989-2007,
TURBOMOLE GmbH, since 2007; available from
https://www.turbomole.org.

A LaTeX template could look like this:

@misc{TURBOMOLE,
title = {{TURBOMOLE V7.5 2020}, a development of {University of Karlsruhe} and
{Forschungszentrum Karlsruhe GmbH}, 1989-2007,
{TURBOMOLE GmbH}, since 2007; available from \\
{\tt https://www.turbomole.org}.}}

In addition, we kindly ask to cite the recent review of the TURBOMOLE project [2]
18 CHAPTER 1. PREFACE AND GENERAL INFORMATION

A LaTeX template for this review could look like this:

@Article{ Balasubramani.Chen.ea:TURBOMOLE.2020,
author = {Balasubramani, Sree Ganesh and Chen, Guo P.
and Coriani, Sonia and Diedenhofen, Michael and
Frank, Marius S. and Franzke, Yannick J. and
Furche, Filipp and Grotjahn, Robin and Harding, Michael E.
and H\"attig, Christof and Hellweg, Arnim and
Helmich-Paris, Benjamin and Holzer, Christof and Huniar, Uwe
and Kaupp, Martin and Marefat Khah, Alireza
and Karbalaei Khani, Sarah and M\"uller, Thomas and Mack, Fabian
and Nguyen, Brian D. and Parker, Shane M. and Perlt, Eva
and Rappoport, Dmitrij and Reiter, Kevin and Roy, Saswata and
R\"uckert, Matthias and Schmitz, Gunnar and Sierka, Marek
and Tapavicza, Enrico and Tew, David P. and van W\"ullen, Christoph
and Voora, Vamsee K. and Weigend, Florian and
Wody{\’n}ski, Artur and Yu, Jason M.},
title = {TURBOMOLE: Modular program suite for \textit{ab initio}
quantum-chemical and condensed-matter simulations},
journal = {J. Chem. Phys.},
volume = {152},
issue = {18},
pages = {184107},
year = {2020},
url = { https://doi.org/10.1063/5.0004635},
DOI = {10.1063/5.0004635}
}

Scientific publications require proper citation of methods and procedures employed.

The output headers of TURBOMOLE modules include the relevant papers. One may also
use the following connections between: method [module] number in the subsequent
list (For module ricc2 see also Section 10).
1.3. HOW TO QUOTE USAGE OF TURBOMOLE 19

• Programs and methods

– general program structure and features: I

– HF-SCF [dscf, ridft]: II
– DFT (quadrature) [dscf, ridft, escf, aoforce]: IV, VI (m grids), VII (a
grids)
– RI-DFT [ridft, aoforce, escf, riper]: V, VI, XXX (marij), X (escf), XXXI
(aoforce)
– periodic DFT [riper]: XLV, XLVII, XLVIII, XLVI
– MP2 [mpgrad]: III
– RI-MP2 [ricc2]: energies and gradients XI, XXXVI, XII, and (static)
polarizabilities XLII
– PNO-MP2 [pnoccsd]: energies XLIII
– stability analysis [escf]: VIII
– electronic excitations with CIS, RPA, TD-DFT [escf]: IX, X, XXVI, XXXIV
– excited state structures and properties with CIS, RPA, TD-DFT [egrad]: XXVII,
XXXIII, XXXIV
– RI-CC2 [ricc2]:
∗ singlet XIX and triplet excitation energies XX
∗ transition moments and first-order properties of excited states XXII
and first-order properties for triplet states XXI
∗ ground state geometry optimizations XXVIII
∗ excited state geometry optimizations and relaxed properties XXIX
∗ parallelization XXXVI
∗ spin-component scaled (SCS) variants XXXVIII
∗ frequency-dependent and static polarizabilities XLII
– RI-ADC(2), RI-CIS(D) and RI-CIS(D∞ ) [ricc2]: XXXV
– SOS variants of MP2, CIS(D), CIS(D∞ ), ADC(2) and CC2 with O(N 4 )-
scaling XXXIX
– analytical second derivatives (force fields) [aoforce]: XXIII, XXIV
– RI-JK [ridft]: XXV
– NMR chemical shifts [mpshift]: XIII, XIV (HF, DFT) XV, XVI (MP2)
– parallel DFT [ridft]: XVII
– geometry optimization in redundant internal coordinates [relax]: XVIII
– RI integral evaluation: XXXII
20 CHAPTER 1. PREFACE AND GENERAL INFORMATION

– explicitly correlated F12 methods for ground state energies [ccsdf12 and pnoccsd]:
MP2-F12 XL, PNO-MP2-F12 XLIV, MP3-F12 XLI, MP4(F12*) XLI,
CCSD(F12) XXXVII, CCSD(F12*) XLI, CCSD(F12)(T) XXXVII, CCSD(F12*)(T)
XLI
– Relativistic approaches: XLIX, L, LI, LII, LIII ([dscf, ridft, etc.]),
LIII, LI,([grad, rdgrad, etc.]), LIV ([mpshift])
– Local hybrid calculations: LV ([ridft]), LVI ([grad, rdgrad]), LVII,
LVIII, LIX, LX ([escf]), LXI ([egrad]),
– Seminumerical and pseudospectral methods: L, LI ([ridft, rdgrad]),
LIX ([escf, egrad, aoforce])

• Orbital and auxiliary basis sets

– basis sets:
∗ SV, SV(P), SVP, DZ (a), TZV, TZVP, TZVPP (b), TZVPP(Rb-Hg)
(f), QZV, QZVP, QZVPP (i)
∗ new balanced basis sets (with smaller ECPs, i.e. the def2 basis sets): j
∗ all-electron basis sets for Rb to Xe (SVPall, SVPPall, TZVPall, TZVP-
Pall): g
∗ references for the correlation consistent basis sets (cc-pVXZ, etc.) can
be found e.g. at
http://tyr0.chem.wsu.edu/˜kipeters/Pages/cc_append.html, or
http://www.grant-hill.group.shef.ac.uk/ccrepo/, or
http://www.emsl.pnl.gov/forms/basisform.html.
Note, that most of the correlation consistent basis sets in the basis
set exchange library of TURBOMOLE have been downloaded from the
latter EMSL web site and therefore users are requested to include
in addition to the original scientific reference an appropriate citation
(see web site for details) in any publications resulting from the use
of these basis sets. See [3] for the current version of the basis set
exchange library and [4] for previous versions. The same applies to
the polarization consistent (pc) basis sets.
∗ property–optimized augmentations: def2-SVPD, def2-TZVPD, def2-
TZVPPD, def2-QZVPD,def2-QZVPPD (m).
∗ basis sets for Dirac–Fock ECPs, i.e. the dhf basis sets: q.
∗ basis sets for relativistic all-electron approaches, i.e. the x2c-XVPall
(X=S, TZ, QZ) basis sets: r, t and their extensions for NMR shielding
constants s, t
– auxiliary basis sets for RI-DFT: c, d, e
– auxiliary basis sets for RI-MP2: f, k, h (for Dunning basis sets)
1.3. HOW TO QUOTE USAGE OF TURBOMOLE 21

Further references of papers not from the TURBOMOLE group are given in the bibliog-
raphy. The following publications describe details of the methodology implemented
in TURBOMOLE:
22 CHAPTER 1. PREFACE AND GENERAL INFORMATION

Methods

I. Electronic Structure Calculations on Workstation Computers: The Program

System TURBOMOLE. R. Ahlrichs, M. Bär, M. Häser, H. Horn and
C. Kölmel; Chem. Phys. Lett., 162, 165 (1989).

II. Improvements on the Direct SCF Method. M. Häser and R. Ahlrichs; J. Com-
put. Chem., 10, 104 (1989).

III. Semi-direct MP2 Gradient Evaluation on Workstation Computers: The MP-

GRAD Program. F. Haase and R. Ahlrichs; J. Comp. Chem., 14, 907 (1993).

IV. Efficient Molecular Numerical Integration Schemes.

O. Treutler and R. Ahlrichs; J. Chem. Phys., 102, 346 (1995).

V. Auxiliary Basis Sets to Approximate Coulomb Potentials. K. Eichkorn, O. Treut-

ler, H. Öhm, M. Häser and R. Ahlrichs; Chem. Phys. Lett., 242, 652 (1995).

VI. Auxiliary basis sets for main row atoms and transition metals and their use to
approximate Coulomb potentials. K. Eichkorn, F. Weigend, O. Treutler and
R. Ahlrichs; Theor. Chem. Acc., 97, 119 (1997).

VII. Error-consistent segmented contracted all-electron relativistic basis sets of double-

and triple-zeta quality for NMR shielding constants.
Y. J. Franzke, R. Treß, T. M. Pazdera and F. Weigend; Phys. Chem. Chem. Phys., 21,
16658–16664 (2019).

VIII. Stability Analysis for Solutions of the Closed Shell Kohn–Sham Equation. R. Bauern-
schmitt and R. Ahlrichs; J. Chem. Phys., 104, 9047 (1996).

IX. Treatment of Electronic Excitations within the Adiabatic Approximation of

Time Dependent Density Functional Theory.
R. Bauernschmitt and R. Ahlrichs; Chem. Phys. Lett., 256, 454 (1996).

X. Calculation of excitation energies within time-dependent density functional the-

ory using auxiliary basis set expansions. R. Bauernschmitt, M. Häser, O. Treut-
ler and R. Ahlrichs; Chem. Phys. Lett., 264, 573 (1997).

XI. RI-MP2: first derivatives and global consistency. F. Weigend and M. Häser;
Theor. Chem. Acc., 97, 331 (1997).

XII. RI-MP2: Optimized Auxiliary Basis Sets and Demonstration of Efficiency. F.

Weigend, M. Häser, H. Patzelt and R. Ahlrichs; Chem. Phys. Lett., 294, 143
(1998).

XIII. Direct computation of second-order SCF properties of large molecules on work-

station computers with an application to large carbon clusters. M. Häser,
 1.3. HOW TO QUOTE USAGE OF TURBOMOLE 23

R. Ahlrichs, H. P. Baron, P. Weis and H. Horn; Theoret. Chim. Acta, 83,

455 (1992).

XIV. Calculation of Magnetic Shielding Constants with meta-GGA Functionals Em-

ploying the Multipole-Accelerated Resolution of the Identity: Implementation
and Assessment of Accuarcy and Efficiency. K. Reiter, F. Mack and F. Weigend;
J.. Chem. Theory Comput., 14, 191 (2018).

XV. A direct implementation of the GIAO-MBPT(2) method for calculating NMR

chemical shifts. Application to the naphthalenium and anthracenium ions.
M. Kollwitz and J. Gauss; Chem. Phys. Lett., 260, 639 (1996).

XVI. Non-Abelian point group symmetry in direct second-order many-body pertur-

bation theory calculations of NMR chemical shifts. M. Kollwitz, M. Häser and
J. Gauss; J. Chem. Phys., 108, 8295 (1998).

XVII. Parallelization of Density Functional and RI-Coulomb Approximation in Tur-

bomole. M. v. Arnim and R. Ahlrichs; J. Comp. Chem., 19, 1746 (1998).

XVIII. Geometry optimization in generalized natural internal Coordinates.

M. v. Arnim and R. Ahlrichs; J. Chem. Phys., 111, 9183 (1999).

XIX. CC2 excitation energy calculations on large molecules using the resolution of
the identity approximation. C. Hättig and F. Weigend; J. Chem. Phys., 113,
5154 (2000).

XX. Implementation of RI-CC2 for triplet excitation energies with an application to

trans-azobenzene. C. Hättig and K. Hald; Phys. Chem. Chem. Phys., 4 2111
(2002).

XXI. First-order properties for triplet excited states in the approximated Coupled
Cluster model CC2 using an explicitly spin coupled basis. C. Hättig, A. Köhn
and K. Hald; J. Chem. Phys., 116, 5401 (2002) and Vir. J. Nano. Sci. Tech.,
5 (2002).

XXII. Transition moments and excited-state first-order properties in the coupled-

cluster model CC2 using the resolution-of-the-identity approximation.
C. Hättig and A. Köhn; J. Chem. Phys., 117, 6939 (2002).

XXIII. An efficient implementation of second analytical derivatives for density func-

tional methods. P. Deglmann, F. Furche and R. Ahlrichs; Chem. Phys. Lett., 362,
511 (2002).

XXIV. Efficient characterization of stationary points on potential energy surfaces.

P. Deglmann and F. Furche; J. Chem. Phys., 117, 9535 (2002).
 24 CHAPTER 1. PREFACE AND GENERAL INFORMATION

XXV. A fully direct RI-HF algorithm: Implementation, optimised auxiliary basis sets,
demonstration of accuracy and efficiency. F. Weigend; Phys. Chem. Chem. Phys., 4,
4285 (2002).

XXVI. An improved method for density functional calculations of the frequency-depen-

dent optical rotation.
S. Grimme, F. Furche and R. Ahlrichs; Chem. Phys. Lett., 361,321 (2002).

XXVII. Adiabatic time-dependent density functional methods for excited state proper-
ties. F. Furche and R. Ahlrichs; J. Chem. Phys. 117, 7433 (2002), J. Chem.
Phys., 121, 12772 (2004) (E).

XXVIII. Geometry optimizations with the coupled-cluster model CC2 using the reso-
lution-of-the-identity approximation. C. Hättig; J. Chem. Phys., 118, 7751,
(2003).

XXIX. Analytic gradients for excited states in the coupled-cluster model CC2 employ-
ing the resolution-of-the-identity approximation. A. Köhn and C. Hättig; J.
Chem. Phys., 119, 5021, (2003).

XXX. Fast evaluation of the Coulomb potential for electron densities using multipole
accelerated resolution of identity approximation. M. Sierka, A. Hogekamp and
R. Ahlrichs;J. Chem. Phys., 118, 9136, (2003).

XXXI. Nuclear second analytical derivative calculations using auxiliary basis set expan-
sion. P. Deglmann, K. May, F. Furche and R. Ahlrichs; Chem. Phys. Lett., 384,
103, (2004).

XXXII. Efficient evaluation of three-center two-electron integrals over Gaussian func-

tions. R. Ahlrichs; Phys. Chem. Chem. Phys., 6, 5119, (2004).

XXXIII. Analytical time-dependent density functional derivative methods within the

RI-J approximation, an approach to excited states of large molecules. D. Rap-
poport and F. Furche; J. Chem. Phys., 122, 064105 (2005).

XXXIV. Density functional theory for excited states: equilibrium structure and elec-
tronic spectra. F. Furche and D. Rappoport; Ch. III of "Computational Pho-
tochemistry", Ed. by M. Olivucci, Vol. 16 of "Computational and Theoretical
Chemistry", Elsevier, Amsterdam, 2005.

XXXV. Structure optimizations for excited states with correlated second-order meth-
ods: CC2, CIS(D∞ ), and ADC(2). C. Hättig; Adv. Quant. Chem., 50, 37-60
(2005).

XXXVI. Distributed memory parallel implementation of energies and gradients for second-
order Møller-Plesset perturbation theory with the resolution-of-the-identity ap-
 1.3. HOW TO QUOTE USAGE OF TURBOMOLE 25

proximation. C. Hättig, A. Hellweg and A. Köhn; Phys. Chem. Chem. Phys.,

8, 1159-1169, (2006).

XXXVII. Quintuple-ζ quality coupled-cluster correlation energies with triple-ζ basis sets.
D. P. Tew, W. Klopper, C. Neiss and C. Hättig; Phys. Chem. Chem. Phys., 9
921–1930 (2007).

XXXVIII. Benchmarking the performance of spin-component scaled CC2 in ground and

electronically excited states. A. Hellweg, S. A. Grün and C. Hättig; Phys.
Chem. Chem. Phys., 10, 4119-4127 (2008).

XXXIX. Scaled opposite-spin CC2 for ground and excited states with fourth order scaling
computational costs. N. O. C. Winter and C. Hättig; J. Chem. Phys., 134,
184101 (2011).
and: Quartic scaling analytical gradients of scaled opposite-spin CC2. N. O.
C. Winter and C. Hättig; Chem. Phys., 401 (2012) 217.

XL. The MP2-F12 Method in the TURBOMOLE Programm Package. R. A. Ba-

chorz, F. A. Bischoff, A. Glöß, C. Hättig, S. Höfener, W. Klopper and D. P.
Tew; J. Comput. Chem., 32, 2492–2513 (2011).

XLI. Accurate and efficient approximations to explicitly correlated coupled-cluster

singles and doubles, CCSD-F12. C. Hättig, D. P. Tew and A, Köhn; J. Chem.
Phys., 132, 231102 (2010).

XLII. Large scale polarizability calculations using the approximate coupled cluster
model CC2 and MP2 combined with the resolution-of-the identity approxima-
tion. D. H. Friese, N. O. C. Winter, P. Balzerowski, R. Schwan and C. Hättig;
J. Chem. Phys., 136, 174106 (2012).

XLIII. A O(N 3 )-scaling PNO-MP2 method using a hybrid OSV-PNO approach with
an iterative direct generation of OSVs. G. Schmitz, B. Helmich and C. Hättig;
Mol. Phys., 111, 2463–2476, (2013).

XLIV. Explicitly correlated PNO-MP2 and PNO-CCSD and its application to the S66
set and large molecular systems. G. Schmitz, C. Hättig and D. P. Tew; Phys.
Chem. Chem. Phys., 16, 22167–22178 (2014).

XLV. Density functional theory for molecular and periodic systems using density fit-
ting and continuous fast multipole methods. R. Łazarski, A. M. Burow and M.
Sierka; J. Chem. Theory Comput., 11, 3029–3041 (2015).

XLVI. Low-memory iterative density fitting. L. Grajciar; J. Comput. Chem., 36,

1521–1535 (2015).
 26 CHAPTER 1. PREFACE AND GENERAL INFORMATION

XLVII. Linear scaling hierarchical integration scheme for the exchange-correlation term
in molecular and periodic systems. A. M. Burow and M. Sierka; J. Chem.
Theory Comput., 7, 3097–3104 (2011).

XLVIII. Resolution of identity approximation for the Coulomb term in molecular and
periodic systems. A. M. Burow, M. Sierka and F. Mohamed; J. Chem. Phys.,
131, 214101 (2009).

XLIX. Self-consistent treatment of spin-orbit interactions with efficient Hartree-Fock

and density functional methods. M. K. Armbruster, F. Weigend, C. van Wüllen
and W. Klopper; Phys. Chem. Chem. Phys., 10, 1748–1756, (2008).

L. Seminumerical exchange and two-component local hybrids. P. Plessow and F.

Weigend; J. Comput. Chem., 33, 810–816 (2012).

LI. Improved SCF treatment of spin-orbit interactions and gradients. A. Baldes

and F. Weigend; Mol. Phys., 111, 2617–2624 (2013).

LII. Relativistic all-electron approaches (BSS, DKH, and X2C). Daoling Peng, Nils
Middendorf, Florian Weigend, Markus Reiher; J. Chem. Phys., 138, 184105
(2013).

LIII. Relativistic all-electron approaches including finite nucleus model and SNSO ap-
proach, and geometry gradients. Y. J. Franzke, N. Middendorf and F. Weigend;
J. Chem. Phys., 148, 104110 (2018).

LIV. Scalar-relativistic (local) X2C for NMR shieldings. Y. J. Franzke and F. Weigend;
J. Chem. Theory Comput., 15, 1028–1043 (2019).

LV. Efficient self-consistent implementation of local hybrid functionals. H. Bah-

mann and M. Kaupp; J. Chem. Theory Comput., 11, 1540–1548, (2015).

LVI. Implementation of molecular gradients for local hybrid density functionals using
seminumerical integration techniques. S. Klawohn, H. Bahmann and M. Kaupp;
J. Chem. Theory Comput., 12, 4254–4262, (2016).

LVII. Efficient semi-numerical implementation of global and local hybrid functionals

for time-dependent density functional theory. T. M. Maier, H. Bahmann and
M. Kaupp; J. Chem. Theory Comput., 11, 4226–4237, (2015).

LVIII. Quasirelativistic two-component core excitations and polarisabilities from a

damped-response formulation of the Bethe–Salpeter equation. M. Kehry, Y.
J. Franzke, C. Holzer and W. Klopper; Mol. Phys., 118, e1755064 (2020).

LIX. An improved seminumerical Coulomb and exchange algorithm for properties

and excited states in modern density functional theory. C. Holzer; J. Chem.
Phys., 153, 184115 (2020).
1.3. HOW TO QUOTE USAGE OF TURBOMOLE 27

LX. Assessing the accuracy of local hybrid density functional approximations for
molecular response properties. C. Holzer, Y. J. Franzke, M. Kehry; J. Chem.
Theory. Comput., 17, 2928–2947 (2021).

LXI. Development and implementation of excited-state gradients for local hybrid

functionals. R. Grotjahn; F. Furche; M. Kaupp, J. Chem. Theory Comput.,
15, 5508–5522, (2019).
28 CHAPTER 1. PREFACE AND GENERAL INFORMATION

Basis sets

The following tables can be used to find the proper citations of the standard orbital
and auxiliary basis sets in the TURBOMOLE basis set library. Recommendations for
applications and a historical overview are provided in the supporting information
of [2]. There, the employed ECPs for heavy elements are listed. ECPs can also be
obtained from the website of the Dolg group together with a detailed bibliography,
please see http://www.tc.uni-koeln.de/PP/clickpse.en.html.
Orbital basis sets, elements H–Kr

H,He Li Be B–Ne Na,Mg Al–Ar K Ca Sc–Zn Ga–Kr

SVP, SV(P) x a a a a a a a a a
TZVP x b b b b b b b b b
TZVPP x f f f f f f f f f
QZVP, QZVPP i
def2-SV(P) j j j j j j j j j j
def2-SVP j j j j j j j j j j
def2-TZVP j j j j j j j j j j
def2-TZVPP j j j j j j j j j j
def2-XVPD/XVPPD, X=S,T,Q m
x2c-XVP (X=S, TZ), PP, -2c j r
x2c-XVP-s (X=S, TZ) s
x2c-QZVP, PP, -2c, -s j t
Note: For H–Kr def-SV(P), def-SVP, ... are identical with the basis sets without def prefix. def2-
QZVPP and def2-QZVP are identical with QZVPP and QZVP. One-component dhf and def2 type
basis sets are identical for the elements up to Kr.
def2-XVPD/XVPPD denotes the property–optimized augmentations def2-SVPD, def2-TZVPD,
def2- TZVPPD, def2-QZVPD, def2-QZVPPD.

Orbital basis sets, elements Rb–Rn

Rb Sr Y–Cd In–Xe Cs Ba La, Hf–Hg Ce–Lu Tl–At Rn

def-SVP, def-SV(P), def-TZVP d - d j
def-TZVPP f d f d f - d j
def2-SV(P) j j j j j j j n j j
def2-SVP j j j j j n j j
def2-TZVP, def2-TZVPP j n j
def2-QZVP, def2-QZVP j n j
def2-XVPD/XVPPD, X=S,T,Q m - m
dhf-XVP (X=S–QZ), PP, -2c q j q – q
x2c-XVP (X=S, TZ), PP, -2c r
x2c-XVP-s (X=S, TZ) s
x2c-QZVP, PP, 2c-, -s t
1.3. HOW TO QUOTE USAGE OF TURBOMOLE 29

Auxiliary basis sets for RI-J in HF/DFT (Coulomb fitting)

H, He Li–Kr Rb–La Ce–Lu Hf–At Rn

(def-)SVP,(def-)SV(P) c c d - d e
(def-)TZVP d d d - d e
def2, universal e n e
x2c-XVP, PP, -2c, -s (X=S,TZ) r
x2c-QZVP, PP, -2c, -s r, t

Auxiliary basis sets for RI-K in HF/DFT (Exchange fitting)

H He B–F Ne Al–Cl Ar Ga-Br Kr-La Ce–Lu Hf–Rn

(def-)TZVPP o p o p o p o -
def2, universal p n p
cc-pVXZ (X = T, Q, 5) o - o - o - o -
30 CHAPTER 1. PREFACE AND GENERAL INFORMATION

Auxiliary basis sets for RI-MP2 and RI-CC, elements H–Ar

H He Li Be B–F Ne Na,Mg Al–Cl Ar

SVP,SV(P) f k f f f k f f k
TZVP,TZVPP f k f f f k f f k
QZVP,QZVPP k
def2-SV(P) f k l f f k l f k
def2-SVP f k l f f k l f k
def2-TZVP,def2-TZVPP f k f l f k l l k
def2-XVPD/XVPPD, X=S,T,Q v
(aug-)cc-pVXZ, X=D–Q h h k k h h k h h
(aug-)cc-pV5Z k k - - k k - k k
cc-pwCVXZ, X=D–5 - - - - k k - k k
Note: the auxiliary basis sets for the (aug-)cc-pV(X+d)Z basis sets for Al–Ar are identical with the
(aug-)cc-pVXZ auxiliary basis sets.

Auxiliary basis sets for RI-MP2 and RI-CC, elements K–Kr

K Ca Sc–Zn Ga–Br Kr
SVP, SV(P) f f f f k
TZVP, TZVPP f f f f k
QZVP, QZVPP k
def2-SV(P) l f f f k
def2-SVP l f l f k
def2-TZVP, def2-TZVPP l f l f k
def2-XVPD/XVPPD, X=S,T,Q v
(aug-)cc-pVXZ, X=D–Q - - - h h
(aug-)cc-pV5Z - - - u u
cc-pCWVXZ, X=D–5 - - - u u
(aug-)cc-pVXZ-PP, X=D–5 - - - u u
cc-pwCVXZ-PP, X=D–5 - - - u u

Auxiliary basis sets for RI-MP2 and RI-CC, elements Rb–Rn

Rb Sr Y–Cd In–Xe Cs Ba La, Hf–Hg Ce–Lu Tl–At Rn

def-SVP,def-SV(P) f - f l
def2-SVP,def2-SV(P) l f f l l f f w l l
def-TZVP,def-TZVPP f - f l
def2-TZVP,def2-TZVPP l w l
def2-QZVP,def2-QZVP l w l
def2-XVPD/XVPPD, X=S,T,Q v - v
aug-cc-pVXZ-PP, X=D–5 - - - u - - - - u u
cc-pwCVXZ-PP, X=D–5 - - - u - - - - u u
1.3. HOW TO QUOTE USAGE OF TURBOMOLE 31

a. Fully Optimized Contracted Gaussian Basis Sets for Atoms Li to Kr. A. Schäfer,
H. Horn and R. Ahlrichs; J. Chem. Phys., 97, 2571 (1992).

b. Fully Optimized Contracted Gaussian Basis Sets of Triple Zeta Valence Quality
for Atoms Li to Kr. A. Schäfer, C. Huber and R. Ahlrichs; J. Chem. Phys., 100,
5829 (1994).

c. Auxiliary Basis Sets to Approximate Coulomb Potentials. K. Eichkorn, O. Treut-

ler, H. Öhm, M. Häser and R. Ahlrichs; Chem. Phys. Lett., 242, 652 (1995).

d. Auxiliary basis sets for main row atoms and transition metals and their use to
approximate Coulomb potentials. K. Eichkorn, F. Weigend, O. Treutler and
R. Ahlrichs; Theor. Chem. Acc., 97, 119 (1997).

e. Accurate Coulomb-fitting basis sets for H to Rn. F. Weigend; Phys. Chem.

Chem. Phys., 8, 1057 (2006).

f. RI-MP2: Optimized Auxiliary Basis Sets and Demonstration of Efficiency. F.

Weigend, M. Häser, H. Patzelt and R. Ahlrichs; Chem. Phys. Lett., 294, 143
(1998).

g. Contracted all-electron Gaussian basis sets for Rb to Xe. R. Ahlrichs and K. May;
Phys. Chem. Chem. Phys., 2, 943 (2000).

h. Efficient use of the correlation consistent basis sets in resolution of the identity
MP2 calculations. F. Weigend, A. Köhn and C. Hättig; J. Chem. Phys., 116,
3175 (2002).

i. Gaussian basis sets of quadruple zeta valence quality for atoms H–Kr. F. Weigend,
F. Furche and R. Ahlrichs; J. Chem. Phys., 119, 12753 (2003).

j. Balanced basis sets of split valence, triple zeta valence and quadruple zeta va-
lence quality for H to Rn: Design an assessment of accuracy. F. Weigend and
R. Ahlrichs; Phys. Chem. Chem. Phys., 7, 3297 (2005).

k. Optimization of auxiliary basis sets for RI-MP2 and RI-CC2 calculation: Core-
valence and quintuple-ζ basis sets for H to Ar and QZVPP basis sets for Li to Kr.
C. Hättig; Phys. Chem. Chem. Phys., 7, 59 (2005).

l. Optimized accurate auxiliary basis sets for RI-MP2 and RI-CC2 calculations
for the atoms Rb to Rn. A. Hellweg, C. Hättig, S. Höfener and W. Klopper;
Theor. Chem. Acc., 117, 587 (2007).

m. Property–optimized Gaussian basis sets for molecular response calculations. D.

Rappoport and F. Furche; J. Chem. Phys., 133, 134105 (2010).

n. Error-Balanced Segmented Contracted Basis Sets of Double-ζ to Quadruple-ζ

Valence Quality for the Lanthanides. R. Gulde, P. Pollak and F. Weigend;
J. Chem. Theory Comput., 18, 4062 (2012).
32 CHAPTER 1. PREFACE AND GENERAL INFORMATION

o. A fully direct RI-HF algorithm: Implementation, optimised auxiliary basis sets,

demonstration of accuracy and efficiency. F. Weigend; Phys. Chem. Chem. Phys., 4,
4285 (2002).

p. Hartree–Fock Exchange Fitting Basis Sets for H to Rn. F. Weigend; J. Com-

put. Chem., 29, 167 (2008).

q. Segmented contracted basis sets for one– and two–component Dirac–Fock effective
core potentials. F. Weigend and A. Baldes; J. Chem. Phys. 133, 174102 (2010).

r. Segmented Contracted Error-Consistent Basis Sets of Double- and Triple-ζ Va-

lence Quality for One- and Two-Component Relativistic All-Electron Calculations.
P. Pollak and F. Weigend; J. Chem. Theory Comput., 13, 3696 (2017).

s. Error-consistent segmented contracted all-electron relativistic basis sets of double-

and triple-zeta quality for NMR shielding constants.
Y. J. Franzke, R. Treß, T. M. Pazdera and F. Weigend; Phys. Chem. Chem. Phys., 21,
16658–16664 (2019).

t. Error-consistent segmented contracted all-electron relativistic basis sets of double-

and triple-zeta quality for NMR shielding constants.
Y. J. Franzke, L. Spiske, P. Pollak and F. Weigend; J. Chem. Theory. Comput., ac-
cepted (2020), DOI: 10.1021/acs.jctc.0c00546.

u. Auxiliary basis sets for density-fitted correlated wavefunction calculations: Weight-

ed core-valence and ECP basis sets for post-d elements. C. Hättig, G. Schmitz
and J. Koßmann; Phys. Chem. Chem. Phys., 14, 6549 (2012).

v. Development of new auxiliary basis functions of the Karlsruhe segmented con-

tracted basis sets including diffuse basis functions (def2-SVPD, def2-TZVPPD,
and def2-QVPPD) for RI-MP2 and RI-CC calculations. A. Hellweg and D. Rap-
poport; Phys. Chem. Chem. Phys., 17, 1010 (2015).

w. Optimized auxiliary basis sets for density fitted post-Hartree-Fock calculations of

lanthanide containing molecules. J. Chmela and M. E. Harding; Mol. Phys., 116,
1523 (2018).

x. unpublished. Orbital basis sets given in the supporting information of j.

1.4. MODULES AND THEIR FUNCTIONALITY 33

1.4 Modules and Their Functionality

For references see Bibliography.

define interactive input generator which creates the input file control.
define supports all basis sets available from the basis set library, espe-
cially the fully atom optimized consistent basis sets of SVP, TZV and
QZV quality [5–9] available for the atoms H–Rn. define determines
the molecular symmetry and internal coordinates which allow efficient
geometry optimization. define allows to perform a geometry optimiza-
tion at a force field level to preoptimize the geometry and to calculate
a Cartesian Hessian matrix. define sets the keywords necessary for
single point calculations and geometry optimizations within a variety
of methods. There are also many features to manipulate geometries of
molecules: just try and see how it works.

uff performs a geometry optimization at a force field level. The Universal

Force Field (UFF) [10] is implemented. Beyond this it calculates an
analytical Hessian (Cartesian) which can be used as a start Hessian for
an ab initio geometry optimization.

dscf for (semi–)direct SCF–HF and DFT calculations (see keywords for func-
tionals supported). dscf supports restricted closed-shell (RHF), spin-
restricted ROHF as well as UHF runs. dscf includes an in-core version
for small molecules.

grad requires a successful dscf run and calculates the gradient of the energy
with respect to nuclear coordinates for all cases treated by dscf.
perform (direct) SCF–HF and DFT calculations—as dscf and grad—
ridft within the very efficient (multipole accelerated) RI–J approximation for
and the interelectronic Coulomb term. These programs also permit to ap-
rdgrad proximate HF exchange within the RI–K approximation or using semi-
numerical techniques. The exchange correlation functionals supported
are specified in define. Relativistic two-component calculations are also
available.
riper performs DFT calculations for molecules and periodic systems using the
RI technique. In addition, a low-memory RI implementation based on
preconditioned conjugate gradient algorithm is available for molecular
systems. Furthermore, a real time-time dependent DFT implementation
based on Magnus expansion for the time evolution operator is available
for molecular systems. Both RHF and UHF runs are supported. Cur-
rently hybrid functionals are not supported.

mpgrad requires a well converged SCF run—by dscf, see keywords—and per-
forms closed-shell RHF or UHF calculations yielding single point MP2
energies and, if desired, the corresponding gradient. Note that mpgrad
34 CHAPTER 1. PREFACE AND GENERAL INFORMATION

performs conventional, i.e. non-RI MP2 calculations only. For real-

life applications it is highly recommended to use RI-MP2 instead (see
module ricc2).

ricc2 calculates electronic ground and excitation energies, transition moments

and properties of ground and excited states at the MP2, CIS, CIS(D),
ADC(2) and CC2 level using either a closed-shell RHF or a UHF SCF
reference function. Calculates R12 basis set limit correction for MP2
energies. Employs the RI technique to approximate two-electron inte-
grals. [11–18].

ccsdf12 calculations of electronic ground state energies beyond MP2/CC2:

RI-MP2-F12, MP3, MP3-F12, MP4, MP4(F12*), CCSD, CCSD(F12),
CCSD(F12*), CCSD(F12)(T), CCSD(F12*)(T) and electronic excita-
tion energies at the CCSD level. [19–22]

pnoccsd calculations of electronic ground state energies with PNO-based meth-

ods starting from MP2 and MP2-F12 up to PNO-CCSD(T). [23, 24]

relax requires a gradient run—by grad, rdgrad, ricc2, egrad, or mpgrad—

and proposes a new structure based on the gradient and the approxi-
mated force constants. The approximated force constants will be up-
dated. This module will not be used by default any more if jobex is
called.

statpt performs structure optimization using the "Trust Radius Image Mini-
mization" algorithm. It can be used to find minima or transition struc-
tures (first order saddle points). Transition structure searches usually
require initial Hessian matrix calculated analytically or the transition
vector from the lowest eigenvalue search.

frog executes one molecular dynamics (MD) step. Like relax, it follows a
gradient run: these gradients are used as classical Newtonian forces to
alter the velocities and coordinates of the nuclei.

aoforce requires a well converged SCF or DFT run—by dscf or ridft, see
keywords—and performs an analytic calculation of force constants, vi-
brational frequencies and IR intensities. aoforce is also able to calculate
only the lowest Hessian eigenvalues with the corresponding eigenvectors
which reduces computational cost. The numerical calculation of force
constants is also possible (see tool NumForce in Section 1.5).

escf requires a well converged SCF or DFT run and calculates time de-
pendent and dielectric properties (spin-restricted closed-shell or spin-
unrestricted open-shell reference):

– static and frequency-dependent polarizabilities within the SCF ap-

proximation
1.4. MODULES AND THEIR FUNCTIONALITY 35

– static and frequency-dependent polarizabilities within the time-

dependent Kohn–Sham formalism, including hybrid functionals
such as B3-LYP
– electronic excitations within the RHF and UHF CI(S) restricted
CI method
– electronic excitations within the so-called SCF-RPA approximation
(poles of the frequency dependent polarizability)
– electronic excitations within the time dependent Kohn–Sham for-
malism (adiabatic approximation). It can be very efficient to use
the RI approximation here, provided that the functional is of non-
hybrid type: we recommend B-P86 (but slightly better results are
obtained for the hybrid functional B3-LYP) [25].
–stability analysis of single-determinant closed-shell wave functions
(second derivative of energy with respect to orbital rotations) [26].
–relativistic two-component calculations [27–29]
–Bethe–Salpeter equation (BSE) [29–33] and the GW method [34–
37]

egrad computes gradients and first-order properties of excited states. Well

converged orbitals are required. The following methods are available
for spin-restricted closed shell or spin-unrestricted open-shell reference
states:

– CI-Singles approximation (TDA)

– Time-dependent Hartree–Fock method (RPA)
– Time-dependent density functional methods

egrad can be employed in geometry optimization of excited states (using

jobex, see Section 5.1), and in finite difference force constant calcula-
tions (using NumForce). Details see [38].

rirpa calculates ground state energies and analytic first-order properties within
the random phase approximation (RPA) and its perturbative correc-
tions, see Section 13. A Kramers-restricted two-component formalism
is implemented for ground state energies.

mpshift requires a converged SCF or DFT run for closed shells. mpshift com-
putes NMR chemical shieldings for all atoms of the molecule at the
SCF, DFT or MP2 level within the GIAO ansatz and the (CPHF) SCF
approximation. From this one gets the NMR chemical shifts by compar-
ison with the shieldings for the standard compound usually employed
for this purpose, e.g. TMS for carbon shifts. Note that NMR shielding
typically requires more flexible basis sets than necessary for geometries
or energies. In molecules with ECP-carrying atoms, chemical shieldings
on all the other atoms can be computed with mpshift [39, 40] in the
way suggested in J. Chem. Phys. 136, 114110 (2012). Alternatively, a
36 CHAPTER 1. PREFACE AND GENERAL INFORMATION

scalar-relativistic all-electron approach (X2C) is available to treat heavy

elements [41]. The (multipole accelerated) RI–J approximation is sup-
ported [42]. mpshift and aoforce can be used to calculate vibrational
circular dichroism (VCD) spectra [43].

freeh calculates thermodynamic functions from molecular data in a control

file; an aoforce or a NumForce run is a necessary prerequisite.

evib calculates the matrix elements of the first order derivative of the Kohn–
Sham operator with respect to atomic displacements and describes the
first order electron-vibration (EV) interaction.

intense calculates Raman scattering cross sections from molecular data in a

control file; an aoforce and an egrad run are a necessary prerequisite.
Please use the Raman script to run these three steps in an automated
way.

woelfling computes a finite number of structures along reaction paths within dif-
ferent interpolation algorithms. It provides an initial path using a mod-
ified Linear Synchronous Transit. See Section 5.9 for details. Please use
the woelfling-job script to run optimizations with it.

proper computes a variety of first-order properties and provides several func-

tionalities to analyse wavefunctions such as orbital localization, popula-
tion analysis, natural transition orbitals, AIM critical points and paths,
etc. and can generate output in a variety of plotting formats (see chapter
19).

1.5 Tools

Note: these tools are very helpful and meaningful for many features of TURBOMOLE.
This is a brief description of additional TURBOMOLE tools. Further information will
be available by running the programs with the argument -help.

actual please use: actual -help

adg adds data group to control file.

E.g.: ’adg scfinstab ucis’ inserts:
$scfinstab ucis

aoforce2g98 usage: aoforce2g98 aoforce.out > g98.out

converts output from the aoforce program to Gaussian 98 style,
which can be interpreted by some molecular viewer (e.g. jmol) to
animate the normal coordinates.

bend example: bend 1 2 3

displays the bending angle of three atoms specified by their number
1.5. TOOLS 37

from the control file. Note that unlike in the TURBOMOLE definition
of internal coordinates the apex atom is the second!

cbasopt optimize auxiliary basis sets for RI-MP2 and RI-CC2 calculations.
Uses ricc2 to calculate the error functional and its gradient and
relax as optimization module. For further details call cbasopt -h.

cc2cosmo manages macro iterations for RI-MP2, RI-CC2 or RI-ADC(2) calcu-

lations in an equilibrated solvent environment described by cosmo(see
Chapter 18.2).

cgnce plots energies as a function of SCF iteration number (gnuplot re-

quired).

cosmoprep sets up control file for a cosmo run (see Chapter 18.2).

cpt The Color Prediction Tool predicts color for calculated and measured
spectra. Simply type cpt in the directory of a finished TD-DFT cal-
culation to obtain absorption and emission colors within a linear color
approximation. However note that cpt is a standalone tool and not
strictly bound to TURBOMOLEİt may be used with data obtained from
various theoretical programs as well as experimental data by reading
in simple ASCII files. A detailed list of functionality is obtained from
cpt –help.

dist example: dist 1 2

calculates atomic distances from TURBOMOLE input files; dist -l 4
gives all interatomic distances to 4 a.u. (5 a.u. is the default).

DRC automates dynamic reaction coordinate calculations forward and back-

ward along the imaginary vibrational mode of a transition state struc-
ture. A transition state optimization with a subsequent frequency
calculation is prerequisite.
For further details call DRC -h.

eiger displays orbital eigenvalues obtained from data group $scfmo. Shows
HOMO-LUMO gap, occupation, checks if there are holes in the occu-
pation, and much more.

evalgrad reads the gradient file and prints the energies of each cycle versus
bond lengths or angles. Five operational modes are possible:
evalgrad prints the energy.
evalgrad 1 prints the coordinate of atom 1.
evalgrad 1 2 prints the distance between atoms 1 and 2.
evalgrad 1 2 3 prints the bending angle as defined in Bend.
evalgrad 1 2 3 4 prints the torsional angle as defined in Tors.

file2control This script copies the content of external data groups (file=) into the
control file. If $file2control is found, the process is reverted.
38 CHAPTER 1. PREFACE AND GENERAL INFORMATION

finit initialises the force constant matrix for the next statpt or relax step.

FDE drives the Frozen Density Embedding calculations.

Fukui automates the calculations of Fukui functions. The density change is

written in dtx files and condensed Fukui functions based on different
population analyses are computed.
For further details call Fukui -h.

gallier converts IR intensities and/or VCD rotational strengths to a spectrum

after the corresponding calculation was performed. Intensities can be
broadened with Gaussian or Lorentzian functions.

hcore prepares the control file for a Hamilton core guess.

jobex usage: see Section 5.1

is the TURBOMOLE driver for all kinds of optimizations.

jobbsse usage: see jobbsse -h

is the driver for counterpoise corrected calculations.

kdg example: kdg scfdiis

kills a data group (here $scfdiis) in the control file.

lhfprep prepares for Localized Hartree-Fock calculations by adjusting param-

eters of the control file.

log2x converts the file logging an MD trajectory into coordinates in frames

appropriate for jmol animation program.

log2egy extracts the energy data (KE, total energy, PE) from an MD log file.

log2int extracts bond lengths or angle from an MD log file.

log2rog computes the radius of gyration, geometric radius and diameter from
an MD log file.

mdprep interactive program to prepare for an MD run, checking in particular

the mdmaster file (mdprep is actually a FORTRAN program).

MECPprep prepares the input for minimum-energy crossing point calculations.

The subdirectories state1 and state2 will be created. Multiplicity
and charge for the two states can be set.
For further details call MECPprep -h.

MECPopt driver for geometry optimizations of minimum-energy crossing points.

The electronic structure calculations are carried out in the subdirec-
tories state1 and state2 and the optimizer step is performed in the
starting directory.
For further details call MECPopt -h.
1.5. TOOLS 39

mp2prep prepares MP2 calculations interactively by adjusting parameters of

the control file according to your system resources.

NumForce calculates numerically force constants, vibrational frequencies, and IR

intensities. (Note that the name of the shell script is NumForce with
capital F.)

outp example: outp 1 2 3 4

displays the out-of-plan angle between atom1 and the plane that is
defined by the last three atoms. atom1 is fixed at atom4.

panama converts energies and oscillator strengths to a spectrum broadened by

Gaussian functions and/or calculates non-relaxed difference densities
of excitations.

past translates and rotates coordinates in the principal axis system and
prints out the rotational constants.

raman calculates vibrational frequencies and Raman intensities. See Sec-

tion 15.2 for explanation.

screwer distorts a molecule along a vibrational mode.

scanprep prepares a series of control files with frozen internal coordinates. The
data group $constraints (e.g. provided by TmoleX) is evaluated.
For further details call scanprep -h.

redox automates the calculation of reduction/oxidation potentials functions

w.r.t. the standard hydrogen electrode by computing the electron
affinities/ionization energies in the gas phase and Gibbs free energy
of solvation with DCOSMO-RS. As a side product electron reorgani-
sation energies are printed.
For further details call redox -h.

vibration distorts a molecule along a vibrational mode or generates a plot of an

IR spectrum (gnuplot required)

sdg shows data group from control file:

for example sdg energy shows the list of calculated energies.

sysname returns the name of your system, used in almost all TURBOMOLE scripts.

stati prepares the control file for a statistics run.

t2x converts TURBOMOLE coordinates to xyz format.

t2aomix creates an input file for the AOMix program. AOMix a software the
analysis of molecular orbitals. For more information
see: (http://www.sg-chem.net/aomix).
Uses tm2molden as described below by automatically adding the $aomix
keyword to the control file.
40 CHAPTER 1. PREFACE AND GENERAL INFORMATION

tm2molden is a versatile tool to create

– molden format input file for the Molden program,

– AOMix input files or
– detailed information about the largest AO contributions to the
MOs.

Molden is a graphical interface for displaying the molecular density,

MOs, normal modes, and reaction paths. For more information about
molden see: http://www.cmbi.ru.nl/molden/molden.html.
This format is also often used as input for other program packages or
property tools.
NOTE: The default normalization of molecular orbitals of d-type (and
beyond) when using tm2molden is different to what Molden expects.
To generate Molden input files with the Molden-own normalization,
please call tm2molden norm, the default name of the resulting file will
be molden_std.input rather than molden.input.
If tm2molden finds the keyword $aomix in the control file, it will write
out an AOMix input file, see: http://www.sg-chem.net/aomix
Finally, tm2molden can be used to print out the largest contributions
of the AO basis functions to the molecular orbitals.
Usage:
tm2molden mostat [molist] [above <threshold>]
e.g.:
tm2molden mostat 230-240,251,255
tm2molden mostat 434-440 above 0.001
tm2molden mostat above 0.02
Only contributions which are larger than a certain percentage (default
is 1%) are printed, this value can be changed with the above option
(as absolute value, so 1% is 0.01). Without a list of orbitals (the
numbering follows the output of eiger) all MOs are printed.

tors is a script to query a dihedral angle in a molecular structure:

e.g. tors 1 2 3 4 gives the torsional angle of atom 4 out of the plane
of atoms 1, 2 and 3.

tbtim is used to convert timings output files from Turbobench calculations

to LATEXtables (for options please type TBTIM –help).

tblist is used to produce summaries of timings from Turbobench calcula-

tions to LATEXformat. (for options please type TBLIST –help).

uhfuse deprecated command, present in the current version for compatibility

reasons only.
Transforms the UHF MOs from a given symmetry to another sym-
metry, which is C1 by default (just enter uhfuse). but can be speci-
fied (e.g. as C2v ) by entering uhfuse -s c2v. Now this functionality
1.5. TOOLS 41

is included in the MO definition menu of define program, see Sec-

tion 4.3.1.

vcd calculates VCD rotational strengths. See Section 15.3 for explanation.

woelfling-job optimizes a reaction path with woelfling.

For further information please type woelfling-job -h.

x2t converts standard xyz files into TURBOMOLE coordinates.

Chapter 2

Installation of Turbomole

2.1 Install TURBOMOLE command line version

Installation requires familiarity with some simple UNIX commands. The TURBOMOLE
package is generally shipped as one tar file. This has to be uncompressed

gunzip turbomole_76.tar.gz

and unpacked

tar -xvf turbomole_76.tar

to produce the whole directory structure.

Important Note: Do NOT install or run TURBOMOLE as root or with root

permissions! The files would have the wrong user- and group-IDs and the per-
missions will not be set correctly!

Unpacking TURBOMOLE has to be done just once, preferably by a user who has the
same group-ID as all other users of the program. The best place is on a net-
work disk which is available on all machines. If only root is allowed to place
files there, please first generate an empty directory as root (e.g. sudo mkdir /nfs-
disk/software/TURBOMOLE-76 ), change the owner of the directory (e.g. sudo
chown user:usergroup /nfs-disk/software/TURBOMOLE-76 ) and then unpack as
non-root user the downloaded file there.

2.1.1 Settings for each user:

The environmental variable $TURBODIR must be set to the directory where TURBOMOLE
has been unpacked, for example:

42
2.1. INSTALL TURBOMOLE COMMAND LINE VERSION 43

TURBODIR=/nfs_disk/software/TURBOMOLE

Then, the most convenient way to extend your path to the TURBOMOLE scripts and
binaries is to source the file Config_turbo_env:

source $TURBODIR/Config_turbo_env

If you have a csh or tcsh as default login shell use

source $TURBODIR/Config_turbo_env.tcsh

instead.
It is recommended to add the two lines given above to your .bashrc (or .profile
or wherever you prefer to add your local settings).
Note: If you do not set $TURBODIR first but use the full path to the Config_turbo_env
file when sourcing, the path should be detected automatically.

2.1.2 Setting system type and $PATH by hand

This section is only needed if

• the automatic setting as described in the sub-chapter before, using the Con-
fig_turbo_env file, does not work. And/or if you:

• want to know what variables and paths have to be set to make TURBOMOLE
ready for use in a detailed manner.

First check that the Sysname tool works on your computer:

$TURBODIR/scripts/sysname

should return the name of your system and this should match a bin/[arch] subdirec-
tory in your TURBOMOLE installation.
If Sysname does not print out a single string matching a directory name in $TURBODIR/bin/,
and if one of the existing binary versions does work, you can force sysname to print
out whatever is set in the environment variable $TURBOMOLE_SYSNAME:

TURBOMOLE_SYSNAME=em64t-unknown-linux-gnu

Please make sure not to append _mpi or _smp to the string when setting $TURBOMOLE_SYSNAME,
even if you intend to run parallel calculations. sysname will append this string au-
tomatically to the system name if $PARA_ARCH is set to MPI or SMP (see chapter 3.4.1
how to set up parallel environment).
You can call TURBOMOLE executables and tools easily from anywhere if you add the
corresponding directories to your path (kornshell or bash syntax):
44 CHAPTER 2. INSTALLATION OF TURBOMOLE

PATH=$PATH:$TURBODIR/scripts
PATH=$PATH:$TURBODIR/bin/‘sysname‘

Note that sysname is set in back quotes which tells the shell to substitute the entry
by the output of sysname.
Now the TURBOMOLE executables can be called from a directory with the required
input files. For example to call dscf and to save the output to a file named dscf.out:

$TURBODIR/bin/‘sysname‘/dscf > dscf.out

or, if the path is OK, simply

dscf > dscf.out

Executable modules are in the bin/[arch] directory (for example, Linux modules
are in bin/em64t-unknown-linux-gnu). Tools (including jobex) are in scripts
and (auxiliary) basis sets are kept in the directories basen, jbasen, jkbasen, cbasen,
xbasen and cabasen. Coordinates for some common chemical fragments are supplied
in structures. The documentation and a tutorial can be found in the folder DOC.

2.1.3 Testing the installation

In addition, some sample calculations are supplied in Turbotest so that the modules
can be tested. Just run TTEST from this directory to run all tests or TTEST -help to
get help on how this works:

cd $TURBODIR/TURBOTEST
TTEST

2.2 Installation problems: How to solve

Please check your user limits!

If one or several tests of the test suite fail, it is very likely that your user limits for
stack size and/or memory are too small.
sh/bash/ksh users: please do a

ulimit -a

to get your actual limits. The output should look like:

core file size (blocks) 0

data seg size (kbytes) unlimited
2.2. INSTALLATION PROBLEMS: HOW TO SOLVE 45

file size (blocks) unlimited

max locked memory (kbytes) unlimited
max memory size (kbytes) unlimited
open files 1024
pipe size (512 bytes) 8
stack size (kbytes) unlimited
cpu time (seconds) unlimited
max user processes 8191
virtual memory (kbytes) unlimited

The most important entries are data size, stack size, max memory size, and virtual
memory. Those should be either unlimited or as big as your total RAM.
To set, e.g. the stack size to the maximum allowed size on your system (the so called
hard limit), do:

ulimit -s hard

csh/tcsh users: please do limit instead of ulimit and check the output.
Again, like given above, the limits should be at least as high as your memory avail-
able. The syntax for changing the limits to unlimited using csh/tcsh is:

limit stacksize hard

Note that on some machines the option hard leads to an error and is not recognized.
In such cases try ulimit -s unlimited or set it to a value like ulimit -s 4019516
(value is in kB, so this example sets it to 4GB).
If you are using a queuing system:
Note that if you are submitting jobs to a queue, the user limits might be different
from what you get when you log in on the machines! To check your limits, you have
to add ulimit or limit in the script that is sent to the queue:

....
ulimit -a > mylimits.out
jobex -ri -c 200 -statpt > jobex.out
...

send it to the queue and check the file mylimits.out to find out which limits are set.
Parallel version:
The parallel binaries are being started by the mpirun command which often uses ssh
to start a process on a remote node. The limits for the stack size can not be set by
the user in such a case, so everything in $HOME/.profile, $HOME/.bashrc, etc. will
not help to get rid of the problem.
To check the limits on a remote node, try (sh/bash/ksh syntax):
46 CHAPTER 2. INSTALLATION OF TURBOMOLE

ssh <hostname> ulimit -a

If the ssh command gives a lower stack size than unlimited or a large number, you
have to change the file

/etc/security/limits.conf

on all nodes where the parallel binaries might run, and add there the line (example
for 4GB limit)

* soft stack 4194303

Redo ssh <hostname>ulimit -a and you should get 4GB stack size limit, as it is set
in limits.conf now.
Chapter 3

How to Run Turbomole

All TURBOMOLE modules need the control file as input file. The control file provides
directly or by cross references the information necessary for all kinds of calculations.
There are different ways to generate the input file, depending on how you want to
use TURBOMOLE.

3.1 Simple input files

For standard non-relativistic DFT (or HF-SCF) ground-state calculations TURBOMOLE

requires only a minimal input:

• generate your atomic coordinates by any tool you are familiar with, save it as
an .xyz file which is a format and use the TURBOMOLE script x2t to convert the
.xyz file into the TURBOMOLE format:
x2t xyzinputfile > coord
or use a conversion tool like babel:
babel -i input-type -o tmol coord
where input-type should be babel’s abbreviation for the format of your input
file.
• generate then the control file with the following content:
$atoms
basis = def2-SV(P)
$coord file=coord
$dft
funtional b-p
grid m3
$end

• you can then start either a single-point calculation with

dscf > dscf.out

47
48 CHAPTER 3. HOW TO RUN TURBOMOLE

or a geometry optimization with

jobex

• if you want to use the RI-J approximation (ridft) add a line with $rij. In
that case single-point calculations should be started with

ridft > ridft.out

and geometry optimizations with

jobex -ri

• for vibrational frequencies and IR intensities invoke after a converged geometry

optimization the program aoforce:
aoforce > aoforce.out
• you can optionally include a title line that will be printed in the outputs

$title
DFT/B-P/def2-SV(P) for MyMolecule

The DFT and HF-SCF programs dscf and ridft will automatically detect the
molecular point group from the provided coordinates and exploit the point group
symmetry. By default the programs will do an Extended Hückel Theory (EHT)
guess for the (neutral) molecule to generate start orbitals and determine the orbital
occupation from the aufbau principle with EHT orbital energies.
If you want to do a calculation on a charged species or for a different number of
unpaired electrons, or, more precisely, a different minor spin quantum number MS ,
this can be specified in the control file with:
$eht charge=n unpaired=m
where n and m have to be integer numbers and m has to be positive.
If a calculation should for any reason not use the full molecular point group, the
point group actually used in the calculation can be set in the control file with:
$symmetry sflies
where sflies should be the Schönflies symbol in lower case letters without using any
special symbol to indicate lower indices, i.e. c2v for C2v . With c1 the use of point
group symmetry is disabled.
Geometry optimizations on molecular (i.e. non-periodic) systems are by default done
in internal redundant coordinates that will be generated automatically by the opti-
mizer statpt when it is invoked for the first time. If you want to set up different
internal coordinates or if you want to freeze internal coordinates you can do this with
the input generator define (vide infra).
The Cartesian positions of atoms can be fixed by adding in the $coord data group
a f behind the atom symbol:
3.2. THE GRAPHICAL USER INTERFACE TMOLEX 49

$coord
0.000 0.000 -0.764 o f
1.500 0.000 0.382 h f
-1.500 0.000 0.368 h

Some basis sets require for heavier atoms the addition of effective core potentials
(ECPs). This can be done as follows:

$atoms
basis = dhf-SVP-2c
pb 2-4,7
ecp = pb dhf-ecp-2c

Some old basis sets from Pople and coworkers (e.g. 6-31G(d)) should be used with 6-
component, i.e. cartesian, d-functions, while TURBOMOLE’s default is to use 5-component
spherical d-functions. The use of 6 component d-functions and the full spherical sets
for higher angular momentum functions can be requested by adding the keyword

$pople CAO

For the additional input needed for post-HF or post-KS calculation please see the
following sections and chapters.

3.2 The graphical user interface TmoleX

An easy way to start as a newbie with TURBOMOLE is to use the free graphical
user interface TmoleX which is part of every TURBOMOLE distribution. Please in-
stall TmoleX on your local desktop computer and avoid running the GUI on re-
mote machines using remote desktop tools like X11. If you do not have a local
version of TmoleX, consider to download and use the free version which is able to
generate input, to run TURBOMOLE jobs on external (Linux) boxes and to visualize
the results - download is available from here: http://www.cosmologic.de/support-
download/downloads/tmolex-client.html
A detailed tutorial for the usage of TURBOMOLE on the command line can be found in
the DOC directory of your TURBOMOLE installation or on the web site of COSMOlogic,
see http://www.cosmologic.de/

3.3 A ‘Quick and Dirty’ Tutorial for the define input gen-
erator

For the generation of input files for more complex calculations TURBOMOLE offers the
interactive input generator define, which guides the user through a series of menues
to set up the required input without the need to know by hard the names of the
keywords and options.
50 CHAPTER 3. HOW TO RUN TURBOMOLE

The define module (program) generates in a step by step manner and interactively
the control file: coordinates, atomic attributes (e.g. basis sets), MO start vectors
and keywords specific for the desired method of calculation. We recommend gen-
erating a set of Cartesian coordinates for the desired molecule using your favourite
molecular builder (e.g. molden) and converting these coordinates into TURBOMOLE for-
mat (see Section 23.2) as input for define. Alternatively the graphical user interface
TmoleX can be used to import and/or build molecules.
A straightforward way to perform even complex TURBOMOLE calculations from scratch
is as follows:

• generate your atomic coordinates by any tool you are familiar with,

• save it as an .xyz file which is a standard output format of all programs, or

use a conversion tool like babel,

• use the TURBOMOLE script x2t to convert your .xyz file to the TURBOMOLE coord
file:
x2t xyzinputfile > coord

• since input files for TURBOMOLE are always called control, each input
has to be placed in a different directory. Create a new directory and
copy the coord file there,

• call
define
after specifying the title, you get the coord menu — just enter
a coord
to read in the coordinates.
Use desy to let define determine the point group automatically.
If you want to do geometry optimizations, we recommend to use generalized
internal coordinates; ired generates them automatically.

• you may then go through the menus without doing anything: just press <Enter>,
* or q—whatever ends the menu, or by confirming the proposed decision of
define again by just pressing <Enter>.
This way you get the necessary specifications for a (SCF-based) run with def-
SV(P) as the default basis set (which is qualitatively similar to 6-31G*).

• for more accurate SCF or DFT calculations choose larger basis sets, e.g. TZVP
by entering b all def-TZVP or b all def2-TZVP in the basis set menu.

• ECPs which include (scalar) relativistic corrections are automatically used be-
yond Kr.

• an initial guess for MOs and occupation numbers is provided by eht

• for DFT you have to enter dft in the last menu and then enter on
3.3. A ‘QUICK AND DIRTY’ TUTORIAL FOR THE DEFINE INPUT GENERATOR51

• for efficient DFT calculations you best choose the RI approximation by entering
ri and then on. For small molecules it can be beneficial to provide additional
memory (with m number ; number in MB), but make sure not to use more than
80% of the memory your computer has available (note that the setting is per
core for parallel jobs!). Auxiliary basis sets are provided automatically. For
medium-sized to larger molecules, additional memory for integral-storage is
not helpful (can even slow down the calculation), but activating the multipole
accelerated RI-J (marij) can speed up the calculation significantly (without
introducing additional errors for RI-J).

• B-P86 is the default functional. It has a good and stable performance through-
out the periodic system.

• for an HF or DFT run without RI, you simply enter:

[nohup] dscf > dscf.out &
or, for a RI-DFT run:
[nohup] ridft > ridft.out &

• for a gradient run, you simply enter:

[nohup] grad > grad.out &
or
[nohup] rdgrad > rdgrad.out &

• for a geometry optimization simply call jobex:

for a standard SCF input:
[nohup] jobex &
for a standard RI-DFT input:
[nohup] jobex -ri &

• many features, such as NMR chemical shifts or vibrational frequencies at SCF

or DFT level, do not require further modifications of the input. Just call e.g.
mpshift or aoforce after the appropriate energy calculation.

• other features, such as post–SCF methods need further action on the input,
using either the last menu of define where one can activate all settings needed
for DFT, TDDFT, MP2, CC2, etc. calculations (this is the recommended way),
or tools like mp2prep.

If that was a too quick and dirty chapter, please read the TURBOMOLE Tutorial in the
DOC directory of your local TURBOMOLE installation. It explains step by step the
generation of input with define and how to run calculations on the command line.

3.3.1 Single Point Calculations: Running Turbomole Modules

All calculations are carried out in a similar way. First you have to run define
to obtain the control file or to add/change the keywords you need for your pur-
pose. This can also be done manually with an editor. Given a bash and a path to
52 CHAPTER 3. HOW TO RUN TURBOMOLE

$TURBODIR/bin/[arch] (see installation, Chapter 2) you call the appropriate module

in the following way (e.g. module dscf):

nohup dscf > dscf.out &

nohup means that the command is immune to hangups, logouts, and quits. & runs
a background command. The output will be written to the file dscf.out. Several
modules write some additional output to the control file. For the required keywords
see Section 22. The features of TURBOMOLE will be described in the following section.

3.3.2 Energy and Gradient Calculations

Energy calculations may be carried out at different levels of theory.

Hartree–Fock–SCF
use modules dscf and grad or ridft and rdgrad to obtain the energy and
gradient. The energy can be calculated after a define run without any previous
runs. dscf and grad need no further keywords ridft and rdgrad only need
the keyword $rij. The gradient calculation however requires a converged dscf
or ridft run.

Density functional theory

DFT calculations are carried out in exactly the same way as Hartree–Fock
calculations except for the additional keyword $dft. For DFT calculations
with the fast Coulomb approximation you have to use the modules ridft and
rdgrad instead of dscf and grad. Be careful: dscf and grad ignore RI–K
flags and will try to do a normal calculation, but they will not ignore RI–J
flags ($rij) and stop with an error message. To obtain correct derivatives of
the DFT energy expression in grad or rdgrad the program also has to consider
derivatives of the quadrature weights—this option can be enabled by adding
the keyword weight derivatives to the data group $dft.
For a semi-direct dscf calculation (Hartree–Fock or DFT) you first have to
perform a statistics run. If you type

stati dscf
nohup dscf > dscf.stat &

the disk space requirement (MB) of your current $thime and $thize combina-
tion will be computed and written to the data group $scfintunit size=integer
(see Section 22.2.8). The requirement of other combinations will be computed
as well and be written to the output file dscf.stat. The size of the integral
file can be set by the user to an arbitrary (but reasonable) number. The file
will be written until it reaches the given size and dscf will continue in direct
mode for the remaining integrals. Note that TURBOMOLE has no 2GB file size
limit.
3.3. A ‘QUICK AND DIRTY’ TUTORIAL FOR THE DEFINE INPUT GENERATOR53

MP2 and MP2-F12

MP2 calculations need well converged SCF runs (the SCF run has to be done
with at least the density convergence $denconv 1.d-7, and $scfconv 7 as
described in Section 22). This applies also to the spin-component scaled (SCS
and SOS) and explicitly-correlated (F12) variants of MP2. For MP2 and MP2-
F12 calculations in the RI approximation use the ricc2 or pnoccsd modules.
The module mpgrad calculates the conventional (non-RI and non-F12) MP2
energy its gradient (only recommended for test calculations). The input can
be prepared with the mp2, cc, or pnocc menu in define.

Excited states with CIS, TDHF and TDDFT (escf)

Single point excited state energies for CIS, TDHF, and TDDFT methods can
be calculated using escf. Excited state energies, gradients, and other first
order properties are provided by egrad. Both modules require well converged
ground state orbitals.

Excited states with second-order wavefunction methods (ricc2)

The module ricc2 calculates beside MP2 and CC2 ground state energies also
CIS (identical to CCS), CIS(D), CIS(D∞ ), ADC(2) or CC2 excitation energies
using the resolution-of-the-identity (RI) approximation. Also available are
spin-component scaled (SCS and SOS) variants of the second-order methods
CIS(D), CIS(D∞ ), ADC(2) or CC2. Excited state gradients are available at
the CCS, CIS(D∞ ), ADC(2), and CC2 levels and the spin-component scaled
variants of the latter three methods. In addition, transition moments and first-
order properties are available for some of the methods. For more details see
Section 10. The input can be prepared using the cc menu of define.

Coupled-Cluster methods beyond CC2: CCSD(F12*)(T) (ccsdf12)

Coupled-Cluster methods beyond CC2 as CCSD and CCSD(T) and Møller-
Plesset perturbation theory beyond MP2 and explicitly-correlated F12 vari-
ants thereof are since Release V7.0 implemented in the ccsdf12 program.
The F12 variants of these methods have a much faster basis set convergence
and are therefore more efficient. We recommend in particular CCSD(F12*)
and CCSD(F12*)(T). Excitation energies are only available for (conventional)
CCSD.

3.3.3 Calculation of Molecular Properties

See Section 1.4 for the functionality and Section 22 for the required keywords of the
modules dscf, ridft, mpshift, escf, and ricc2.

3.3.4 Modules and Data Flow

See Figure 3.1.

 54 CHAPTER 3. HOW TO RUN TURBOMOLE

USER INPUT:
coordinates

input generator  From TURBOMOLE library:

define basis sets

?
vib. frequencies
ground state energy aoforce - thermodynamics
NumForce Freeh
- dscf -

ridft
TDDFT excited states,
- response properties
escf, egrad
 @
R
@
HF/DFT/TDDFT NMR shieldings
J gradient MP2/CC2 -
o ?
energy+gradient mpshift
6
b grad
ricc2
e rdgrad
x mpgrad Raman spectrum
egrad -
Raman

@
@
CC2 excited states and
@
R
@ - response properties
geometry changes ricc2
minimum or transition
state molecular dynamics
(PNO-)CC and
relax frog (PNO-)CC-F12
statpt - calculations
ccsdf12
pnoccsd

Figure 3.1: The modules of TURBOMOLE and the main data flow between them.
3.4. PARALLEL RUNS 55

3.4 Parallel Runs

Some of the TURBOMOLE modules are parallelized using the message passing inter-
face (MPI) for distributed and shared memory machines or with OpenMP or multi-
threaded techniques for shared memory and multi-core machines.
Generally there are two hardware scenarios which determine the kind of paralleliza-
tion that is possible to use:

• On a single node with several CPUs and/or cores using the same memory
(shared memory), the user can run all parallelized modules of TURBOMOLE . For
some modules, both shared-memory and MPI versions are available, but it is
recommended not to use the latter ones for performance reasons.
How to run the parallel TURBOMOLE SMP version on multi-core and/or multi-
CPU systems: Please see chapter 3.4.2.

• On a cluster a parallel calculation can be performed using several distinct

nodes, each one with local memory and disks. This can be done with the MPI
version. It is, however, often more efficient to use the SMP version also on a
cluster by running each individual job on a single node using all cores.
How to run the parallel TURBOMOLE MPI version on clusters: Please see chapter
3.4.1.

The list of programs parallelized includes presently:

• ridft — parallel ground state Hartree-Fock and DFT energies including RI-J
and the multipole accelerated RI (MA-RI-J), SMP and MPI

• rdgrad — parallel ground state gradients from ridft calculations, SMP and
MPI

• dscf — Hartree-Fock and DFT ground state calculations for all available DFT
functionals, without the usage of RI-J approximation, SMP and MPI

• riper — parallel ground state DFT energies for molecular and periodic sys-
tems. SMP only.

• grad — parallel ground state gradients from dscf calculations, SMP and MPI

• ricc2 — parallel ground and excited state calculations of energies and gradi-
ents at MP2 and CC2 level using RI, as well as energy calculations of other
wave function models, see chapter 10.6. SMP and MPI

• ccsdf12 — parallel ground state energies beyond MP2/CC2 and excitation

energies beyond CC2. SMP

• pnoccsd — parallel ground state energies at the OSV-PNO-MP2 and OSV-

PNO-MP2-F12 level, SMP and MPI. For PNO-CCSD and PNO-CCSD(T) only
the SMP version is available.
56 CHAPTER 3. HOW TO RUN TURBOMOLE

• mpgrad — parallel conventional (i.e. non-RI) MP2 energy and gradient cal-
culations. Please note that RI-MP2 is one to two orders of magnitude faster
than conventional MP2, so even serial RI-MP2 will be faster than parallel MP2
calculations. SMP and MPI

• aoforce — parallel Hartree-Fock and DFT analytic 2nd derivatives for vi-
brational frequencies, IR spectra, generation of Hessian for transition state
searches and check for minimum structures. SMP and MPI

• escf — parallel TDDFT, RPA, CIS excited state calculations (UV-Vis and
CD spectra, polarizabilities). SMP and MPI

• egrad — parallel TDDFT, RPA, CIS excited state analytic gradients, including
polarizability derivatives for RAMAN spectra. SMP only.

• mpshift — parallel NMR shielding constants and chemical shifts. SMP only.

• tb — parallel GFN2-xTB energy and gradient calculations. SMP only.

• NumForce — this script can used for a trivial parallelization of the numerical
displaced coordinates. SMP and MPI

See also [2]. Additional keywords necessary for parallel runs with the SMP or MPI
binaries are not needed. When using the parallel version of TURBOMOLE, scripts are
replacing the binaries. Those scripts prepare a usual input, run the necessary steps
and automatically start the parallel programs. The users just have to set environment
variables, see Sec. 3.4.1 below.
To use the OpenMP parallelization, only an environment variable needs to be set.
But to use this parallelization efficiently one should consider a few additional points,
e.g. memory usage, which are described in Sec. 3.4.2.

3.4.1 Running Parallel Jobs — MPI case

The parallel version of TURBOMOLE runs on all supported systems:

• workstation cluster with Ethernet, Infiniband, Myrinet (or other) connection

• SMP systems

• or combinations of SMP and cluster

Setting up the parallel MPI environment

In addition to the installation steps described in Section 2 (see page 42) you just
have to set the variable PARA_ARCH to MPI, i.e. in sh/bash/ksh syntax:

export PARA_ARCH=MPI
3.4. PARALLEL RUNS 57

This will cause sysname to append the string _mpi to the system name and the scripts
like jobex will take the parallel binaries by default. To call the parallel versions of
programs like ridft, rdgrad, dscf, ricc2, etc. from your command line without
explicit path, expand your $PATH environment variable to:

export PATH=$TURBODIR/bin/‘sysname‘:$PATH

The usual binaries are replaced by scripts that prepare the input for a parallel run
and start mpirun (or poe on IBM) automatically. The number of CPUs that shall
be used can be chosen by setting the environment variable PARNODES:

export PARNODES=8

The default for PARNODES is 2.

Finally the user can set a default scratch directory that must be available on all nodes.
Writing scratch files to local directories is highly recommended for I/O intensive
modules like ricc2, otherwise the scratch files will be written over the network to
the same directory where the input is located. The path to the local disk can be set
with

export TURBOTMPDIR=/scratch/username

This setting is automatically recognized by most parallel programs. Note:

• For RI-DFT calculations it is usually neither necessary nor helpful to set TUR-
BOTMPDIR.

• This does not set the path for the integral scratch files for dscf (see section
below about twoint of keyword $scfintunit).

• In MPI parallel runs the programs attach to the name given in $TURBOTMPDIR
node-specific extension (e.g. /scratch/username-001) to avoid clashes be-
tween processes that access the same file system. The jobs must have the per-
missions to create these directories. Therefore one must not set $TURBOTMPDIR
to something like /scratch which would result in directory names like
/scratch-001 — which usually can not be created by jobs running under a
standard user account.
So please set the temporary directory for parallel files to a local file system
at a position you are allowed to generate directories, like /tmp/myname/tm-
jobs/mpifiles

MPI versions, distributions and flavours

TURBOMOLE is using the MPI version which has been utilized to generate the binaries.
To make sure that the parallel version is running, no matter which MPI flavour you
have installed on your machines, TURBOMOLE does include the run-time version of
the MPI flavour it needs.
58 CHAPTER 3. HOW TO RUN TURBOMOLE

Please do not try to use TURBOMOLE with your local MPI version (OpenMPI,
MPICH, ...)! Do not call the parallel MPI binaries directly, just set $PARA_ARCH
as described above and call the modules the same way you use them in the serial
version.

On Linux for PCs and Windows systems either IBM Platform MPI (formerly
known as HP-MPI, now also known as IBM Spectrum MPI) is used and included —
see IBM Platform MPI — or Intel MPI Intel MPI.
COSMOlogic ships TURBOMOLE with a IBM Platform MPI Community Edition or
the full Intel MPI version. TURBOMOLE users do not have to install or license IBM
Platform MPI or Intel MPI themselves. Parallel binaries will run out of the box on
the fastest interconnect that is found - Infiniband, Myrinet, TCP/IP, etc.
Note: most parallel TURBOMOLE modules need an extra server running in addition
to the clients. This server is included in the parallel binaries and it will be started
automatically — but this results in one additional task that usually does not need
any CPU time. So if you are setting PARNODES to N, N+1 tasks will be started.
If you are using a queuing system or if you give a list of hosts where TURBOMOLE
jobs shall run on (see below), make sure that the number of supplied nodes match
$PARNODES — e.g. if you are using 4 CPUs via a queuing system, make sure that
$PARNODES is set to 4.

Starting parallel jobs

After setting up the parallel environment as described in the previous section, parallel
jobs can be started just like the serial ones. If the input is a serial one, it will be
prepared automatically for the parallel run.
For the additional mandatory or optional input for parallel runs with the ricc2
program see Section 10.6.

Running calculations on different nodes

If TURBOMOLE is supposed to run on a cluster, we highly recommend the usage of

a queuing system like PBS, Univa/SGE GridEngine or LFS. The parallel version
of TURBOMOLE will automatically recognise that it is started from within one of the
queuing systems:

• PBS (Torque/Maui)

• LSF

• SLURM

• SGE or Univa Grid Engine

3.4. PARALLEL RUNS 59

and the binaries will run on the machines those queuing systems provide.
Note: The default is to use secure shell (ssh) to start the jobs on the nodes. Please
make sure that ssh works in a passwordless fashion on the cluster. If ssh is not
possible, but srun is enabled, set export MPI_USESRUN=1 to use srun instead of ssh.
Important: Make sure that the input files are located on a network directory like an
NFS disk which can be accessed on all nodes that participate at the calculation.
If parallel jobs are started outside a queuing system, or if you have a non-supported
or a non-default installation of above mentioned queuing systems, the number of
nodes and their names can also be provided by the user. A file that contains a list
of machines has to be created, each line containing one machine name:

node1
node1
node2
node3
node4
node4

And the environment variable $HOSTS_FILE has to be set to that file:

export HOSTS_FILE=/nfshome/username/hostsfile

Note: Do not forget to set $PARNODES to the number of lines in $HOSTS_FILE, unless
you have set in addition OMP_NUM_THREADS (see below).
Note: In general the stack size limit has to be raised to a reasonable amount
of the memory (or to unlimited). In the serial version the user can set this by
ulimit -s unlimited on bash/sh/ksh shells or limit stacksize unlimited on
csh/tcsh shells. However, for the parallel version that is not sufficient if several
nodes are used, and the /etc/security/limits.conf files on all nodes might have
to be changed. See chapter 2.2 of this documentation, page 45

OpenMP/MPI hybrid version

Some TURBOMOLE modules like dscf, grad, aoforce, ricc2, escf or pnoccsd are
parallelized using a hybrid OpenMP/MPI scheme. For those modules it is sufficient
to start just one single process per node. In addition, please set

export OMP_NUM_THREADS=<number of cores per node>

when starting the job. This environment variable will be exported to each node such
that the processes started there will open <number of cores per node> threads.
60 CHAPTER 3. HOW TO RUN TURBOMOLE

Memory for parallel jobs

Since there are several different parallel versions of the individual TURBOMOLE modules
available, the meaning of the keywords to set memory ($ricore and $maxcor) can
be quite confusing. A lot of problems can be avoided if following points are taken
care of:

• for almost all cases increasing $ricore will not speed up the calculation but
increase memory consumption significantly. It is therefore recommended to set
$ricore to a small value like 100 or 500. Except:

• usage of RI-JK does benefit from large $ricore values. Check if $rik is present
in your control file — and if yes, try to increase the memory to a value which
your machines are capable to handle.

• many post-SCF programs read and use $maxcor. There are a couple of options
to this keyword like per_proc to define the amount of memory per process or
per_node for memory settings on one node, etc. A detailed description can be
found in chapter 22.2.2 on page 381.

• if you worry about speed in RI-DFT calculations, make sure to have $marij
in your control file

Testing the parallel binaries

The binaries ridft, rdgrad, dscf, grad, and ricc2 can be tested by the usual test
suite: go to $TURBODIR/TURBOTEST and call TTEST
Note: Some of the tests are very small and will only pass properly if 2 CPUs are
used at maximum. Therefore TTEST will not run any test if $PARNODES is set to a
higher value than 2.
If you want to run some of the larger tests with more CPUs, you have to edit the
DEFCRIT file in TURBOMOLE/TURBOTEST and change the $defmaxnodes option.

Sending additional server task to sleep

Except for the MPI parallel binaries of ridft and rdgrad all modules start an
additional server process which is not participating in the calculation itself, but just
responsible for the task distribution. This server task is waiting for the processes to
ask for new tasks in a way that the response time is as low as possible. This leads
to a noticeable CPU usage.
To send the server task to sleep while waiting for communication, it is possible to
set the environment variable TM_SERVERSLEEP either to a value in microseconds or
to a string like ’on’ to use the default of 500 microseconds:

export TM_SERVERSLEEP=1000
3.4. PARALLEL RUNS 61

#or
export TM_SERVERSLEEP=on

This is only reasonable if you use relatively few CPUs for a calculation. The larger
the number of processes, the more important is a timely reply from the server.

Sample simple PBS start script

#!/bin/sh
# Name of your run :
#PBS -N turbomole
#
# Number of nodes to run on:
#PBS -l nodes=4
#
# Export environment:
#PBS -V

# Set your TURBOMOLE pathes:

######## ENTER YOUR TURBOMOLE INSTALLATION PATH HERE ##########

export TURBODIR=/whereis/TURBOMOLE
###############################################################

export PATH=$TURBODIR/scripts:$PATH

## set locale to C
unset LANG
unset LC_CTYPE

# set stack size limit to unlimited:

ulimit -s unlimited

# Count the number of nodes

PBS_L_NODENUMBER=‘wc -l < $PBS_NODEFILE‘

# Check if this is a parallel job

if [ $PBS_L_NODENUMBER -gt 1 ]; then
##### Parallel job
# Set environment variables for a MPI job
export PARA_ARCH=MPI
export PATH="${TURBODIR}/bin/‘sysname‘:${PATH}"
export PARNODES=‘expr $PBS_L_NODENUMBER‘
else
##### Sequentiel job
# set the PATH for Turbomole calculations
62 CHAPTER 3. HOW TO RUN TURBOMOLE

export PATH="${TURBODIR}/bin/‘sysname‘:${PATH}"
fi

#VERY important is to tell PBS to change directory to where

# the input files are:

cd $PBS_O_WORKDIR

######## ENTER YOUR JOB HERE ##################################

jobex -ri > jobex.out
###############################################################

3.4.2 Running Parallel Jobs — SMP case

The SMP version of TURBOMOLE currently combines three different parallelization

schemes which all use shared memory:

• dscf, grad, ridft, rdgrad, aoforce, escf, egrad, ricc2, ccsdf12, pnoccsd,
mpshift, evib, odft, rirpa and riper are partially parallelized with OpenMP
for applications on shared-memory, in particular multi-CPU and multi-core,
machines.

• aoforce, escf, egrad, ridft and rdgrad are also parallelized as described
in [44]

• ridft and rdgrad are parallelized with MPI using shared memory on SMP
systems. This is also the default version for SMP systems, not just for MPI
runs.

Setting up the parallel SMP environment

In addition to the installation steps described in Section 2 (see page 42) you just
have to set the variable PARA_ARCH to SMP, i.e. in sh/bash/ksh syntax:

export PARA_ARCH=SMP

This will cause sysname to append the string _smp to the system name and the scripts
like jobex will take the parallel binaries by default. To call the parallel versions of
the programs (like ridft or aoforce) from your command line without explicit path,
expand your $PATH environment variable to:

export PATH=$TURBODIR/bin/‘sysname‘:$PATH

The usual binaries are replaced now by scripts that prepare the input for a parallel
run and start the job automatically. The number of CPUs that shall be used can be
chosen by setting the environment variable PARNODES:
3.4. PARALLEL RUNS 63

export PARNODES=8

The default for PARNODES is 2.

NOTE: Depending on what you are going to run, some care has to be taken that
the system settings like memory limits, etc. will not prevent the parallel versions to
run. See the following sections.

OpenMP parallelization of almost all time consuming modules

The OpenMP parallelization does not need any special program startup. The bi-
naries can be invoked in exactly the same manner as for sequential (non-parallel)
calculations. Just set the environment variable PARNODES to the number or threads
that should be used by the programs. The scripts will set OMP_NUM_THREADS to the
same value and start the OpenMP binaries directly. The number of threads is essen-
tially the max. number of CPU cores the program will try to utilize. To exploit e.g.
all eight cores of a machine with two quad-core CPUs set

export PARNODES=8

(for csh and tcsh use setenv PARNODES=8).

Presently the OpenMP parallelization of ricc2 comprises all functionalities apart
from the O(N 4 )-scaling LT-SOS-RI functionalities (which are only parallelized with
MPI) and expectation values for Ŝ 2 (not parallelized). Note that the memory speci-
fied with $maxcor is for OpenMP-parallel calculation the maximum amount of mem-
ory that will be dynamically allocated by all threads together. To use your compu-
tational resources efficiently, it is recommended to set this value to about 75% of the
physical memory available for your calculations.
For Localized Hartree-Fock calculations please use the dscf program which is paral-
lelized using OpenMP. In this case an almost ideal speedup is obtained because the
most expensive part of the calculation is the evaluation of the Fock matrix and of
the Slater-potential, and both of them are well parallelized. The calculation of the
correction-term of the grid will use a single thread.
The OpenMP parallelization of riper covers all contributions to the Kohn-Sham
matrix and nuclear gradient. Hence an almost ideal speedup is obtained.
To use the OpenMP version of ridft and rdgrad instead of the default parallelization
on SMP machines, just set

export TM_PAR_OMP=on

and start the jobs the usual way. Some features like the semi-numerical exchange
(keyword $senex, see section 22.2.8, two-component difference densities, periodic em-
bedding, COSMO, coulex) are parallelized with OpenMP only. Moreover, OpenMP
can have other benefits like the amount of hardware resources used.
Restrictions:
64 CHAPTER 3. HOW TO RUN TURBOMOLE

• In the ricc2 program the parts related to RI-MP2-F12, LT-SOS-RI-MP2 or

calculation of expectation values for Ŝ 2 do not (yet) use OpenMP paralleliza-
tion. If the OpenMP parallelization is switched on (by setting OMP_NUM_THREADS)
these parts will still be executed sequentially.

• In the dscf program the $incore option will be ignored if more than one thread
is used. Semi-direct dscf calculations (i.e. if a size larger than 0 is given two-
electron integral scratch file in $scfintunit) can not be combined with the
OpenMP parallel runs. (The program will then stop with error message in the
first Fock matrix construction.)

Multi-thread parallelization of dscf, grad, aoforce, escf, egrad, ridft and

rdgrad

The parallelization of those modules is described in [44] and is based on fork()

and Unix sockets. Except setting PARNODES which triggers the environment variable
SMPCPUS, the environment variable

export TM_PAR_FORK=on

has to be set. Alternatively, the binaries can be called with -smpcpus <N> command
line option or with the keyword $smp_cpus in the control file.
The efficiency of the parallelization is usually similar to the default version, but for
ridft and rdgrad RI-K is not parallelized. If density convergence criteria ($denconv)
is switched on using ridft and if no RI-K is being used, the multi-threaded version
should be used.

SMP/MPI version of ridft and rdgrad

Since TURBOMOLE version 7.2 the usage of GlobalArrays has been omitted. Instead,
a set of routines which utilize shared memory on a node has been implemented.
Both modules, ridft and rdgrad, start each process as an individual MPI instance.
Processes on the same node are then collected to collectively store and use data
in a shared memory region. This avoids excessive memory usage and reduces the
amount of memory requirements significantly, especially compared to the old MPI
implementation (which has been used by default in former TURBOMOLE versions). It
is nevertheless recommended to

• run jobs in parallel only if the molecules and/or basis set size is large enough
— several hundred basis functions for non-hybrid functionals, or few hundred
for hybrid functional calculations. The RI approximation in combination with
MARI-J is already very fast in the serial version, usage of many cores would
introduce a communication overhead which exceeds the computational time on
a single core.
3.4. PARALLEL RUNS 65

• not ask for too much memory. For medium sized to large molecules adding a
significant amount of $ricore will not speed up the calculation, but eventually
lead to overstressed systems (or even memory swapping) or failure of jobs due
to too large memory requirements. So please do not set large $ricore values
in such cases, a few ten or hundred MB are sufficient (even zero works equally
well).
66 CHAPTER 3. HOW TO RUN TURBOMOLE
Chapter 4

Preparing your input file with

Define

define is the general interactive input generator of TURBOMOLE. During a session

with define, you will create the control file which controls the actions of all other
TURBOMOLE programs. Starting with Version 7.5, the syntax of control (namely the
$atoms data group) has changed and older versions of TURBOMOLE are no longer able
to read the new control file. The old syntax can be enforced by starting define
with the command line option -old. New versions of TURBOMOLE are able to handle
both the old and the new syntax.
During your define session you will be guided through four main menus:

1. The geometry main menu: This first menu allows you to build your molecule,
define internal coordinates for geometry optimizations, determine the point
group symmetry of the molecule, adjust internal coordinates to the desired
values and related operations. Beyond this one can perform a geometry op-
timization at a force field level to preoptimize the geometry and calculate a
Cartesian analytical Hessian. After leaving this menu, your molecule to be
calculated should be fully specified.
2. The atomic attributes menu: Here you will have to assign basis sets and/or
effective core potentials to all atoms. The SV(P) basis is assigned automatically
as default, as well as ECPs (small core) beyond Kr.
3. The occupation numbers and start vectors menu: In this menu you
should choose eht to start from Extended Hückel MO vectors. Then you have
to define the number of occupied orbitals in each irreducible representation.
4. The general menu: The last menu manages a lot of control parameters for
all TURBOMOLE programs.

Most of the menu commands are self-explanatory and will only be discussed briefly.
Typing * (or q) terminates the current menu, writes data to control and leads to
the next while typing & goes back to the previous menu.

67
68 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

4.0.3 Universally Available Display Commands in Define

There are some commands which may be used at (almost) every stage of your define
session. If you build up a complicated molecular geometry, you will find the dis
command useful. It will bring you to the following little submenu:
ANY COMMAND WHICH STARTS WITH THE 3 LETTERS dis IS A
DISPLAY COMMAND. AVAILABLE DISPLAY COMMANDS ARE :
disc <range> : DISPLAY CARTESIAN COORDINATES
dist <real> : DISPLAY DISTANCE LIST
disb <range> : DISPLAY BONDING INFORMATION
disa <range> : DISPLAY BOND ANGLE INFORMATION
disi <range> : DISPLAY VALUES OF INTERNAL COORDINATES
disg <range> : GRAPHICAL DISPLAY OF MOL. GEOMETRY
<range> IS A SET OF ATOMS REFERENCED
<real> IS AN OPTIONAL DISTANCE THRESHOLD (DEFAULT=5.0)
AS AN EXAMPLE CONSIDER disc 1,3-6,10,11 WHICH DISPLAYS
THE CARTESIAN COORDINATES OF ATOMS 1,3,4,5,6,10,and 11 .
HIT >return< TO CONTINUE OR ENTER ANY DISPLAY COMMAND

Of course, you may enter each of these display commands directly without entering
the general command dis before. The option disg needs special adaption to the
computational environment, however, and will normally not be available.

4.0.4 Specifying Atomic Sets

For many commands in define you will have to specify a set of atoms on which that
command shall act. There are three ways to do that:

• You may enter all or none, the meaning of which should be clear (entering
none makes not much sense in most cases, however).
• You may specify a list of atomic indices like 1 or 3,5,6 or 2,4-6,7,8-10 or
similar.
• You may also enter atomic identifiers which means strings of at most eight
characters: the first two contain the element symbol and the remaining six
could be used to distinguish different atoms of the same type. For example, if
you have several carbon atoms in your molecule, you could label some c ring
and others c chain to distinguish them. Whenever you want to enter an atomic
identifier, you have to put it in double quotation marks: "c ring".

You should take into account that define also creates, from the atoms you entered,
all others according to symmetry. If necessary, you will therefore have to lower the
(formal) symmetry before executing a command.

4.0.5 control as Input and Output File

define may be used to update an existing control file, which is helpful if only the
basis set has been changed. In this case just keep all data, i.e. reply with <enter> on
 69

all questions, and only specify new start MOs. The more general usage is described
now.
At the beginning of each define session, you will be asked to enter the name of
the file to be created. As mentioned earlier, all TURBOMOLE programs require their
input to be on a file named control, but it may be useful at this moment to choose
another name for this file (e.g. if you have an old input file control and you do not
want to overwrite it). Next you will be asked to enter the name of an old file which
you want to use as input for this session. This prevents you from creating the new
input from scratch if you want to make only minor changes to an old control file.
It is possible to use the same file as input and output file during a define session
(which means that it will only be modified). This may lead to difficulties, however,
because define reads from the input file when entering each main menu and writes
the corresponding data when leaving this menu. Therefore the input file may be in
an ill-defined status for the next main menu (this will be the case, for example, if
you add or change atoms in the first menu so that the basis set information is wrong
in the second menu). define takes care of most—but not all—of these problems.
For these reasons, it is recommended to use a different filename for the input and
the output file of the define session if you change the molecule to be investigated.
In most cases involving only changes in the last three of the four main menus no
problem should arise when using the same file as input and output.

4.0.6 Be Prepared

Atomic Coordinates

Molecules and their structures are specified by coordinates of its atoms, within the
program invariably by Cartesian coordinates in atomic units (Ångstrøm would also
do). In TURBOMOLE these coordinates are contained in the file coord (see Section 23
“Sample control files” for an example).

Recommendation

We strongly recommend to create the coord file before calling define, only for small
molecules one should use the interactive input feature of define. Set up the molecule
by any program you like and write out coordinates in the xyz-format (XMol format),
which is supported by most programs. Then use the TURBOMOLE tool x2t to convert
it into a TURBOMOLE coord file (see Section 1.5.

Internal Coordinates

Structure optimizations, see jobex, are most efficient if carried out in internal coor-
dinates and TURBOMOLE offers the following choices.

internals based on bond distances and angles, see Section 4.1.2.

70 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

redundant internals
defined as linearly independent combinations of internals (see ref. [45]),
provided automatically by the command ired in the ‘geometry main
menu’ in Section 4.1 below. This works in almost all cases and is efficient.
The disadvantage is, that this is a black box procedure, the coordinates
employed have no direct meaning and cannot be modified easily by the
user.

cartesians
should always work but are inefficient (more cycles needed for conver-
gence). Cartesians are the last resort if other options fail, they are as-
signed as default if one leaves the main geometry menu and no other
internals have been defined.

4.1 The Geometry Main Menu

After some preliminaries providing the title etc. you reach the geometry main menu:

SPECIFICATION OF MOLECULAR GEOMETRY ( #ATOMS=0 SYMMETRY=c1 )

YOU MAY USE ONE OF THE FOLLOWING COMMANDS :
sy <group> <eps> : DEFINE MOLECULAR SYMMETRY (default for eps=3d-1)
desy <eps> : DETERMINE MOLECULAR SYMMETRY AND ADJUST
COORDINATES (default for eps=1d-6)
syndi <eps> : LIKE DESY, BUT FIND ONLY GROUPS WITH NON-
DEGENERATE IRREPS (D2h AND SUBGROUPS)
susy : ADJUST COORDINATES FOR SUBGROUPS
ai : ADD ATOMIC COORDINATES INTERACTIVELY
a <file> : ADD ATOMIC COORDINATES FROM FILE <file>
aa <file> : ADD ATOMIC COORDINATES IN ANGSTROEM UNITS FROM FILE <file>
sub : SUBSTITUTE AN ATOM BY A GROUP OF ATOMS
i : INTERNAL COORDINATE MENU
ired : REDUNDANT INTERNAL COORDINATES
red_info : DISPLAY REDUNDANT INTERNAL COORDINATES
ff : UFF-FORCEFIELD CALCULATION
m : MANIPULATE GEOMETRY
frag : DEFINE FRAGMENTS FOR BSSE CALCULATION
w <file> : WRITE MOLECULAR COORDINATES TO FILE <file>
r <file> : RELOAD ATOMIC AND INTERNAL COORDINATES FROM FILE <file>
name : CHANGE ATOMIC IDENTIFIERS
del : DELETE ATOMS
dis : DISPLAY MOLECULAR GEOMETRY
banal : CARRY OUT BOND ANALYSIS
* : TERMINATE MOLECULAR GEOMETRY SPECIFICATION
AND WRITE GEOMETRY DATA TO CONTROL FILE

IF YOU APPEND A QUESTION MARK TO ANY COMMAND AN EXPLANATION

OF THAT COMMAND MAY BE GIVEN

This menu allows you to build your molecule by defining the Cartesian coordinates
4.1. THE GEOMETRY MAIN MENU 71

interactively (ai) or by reading the coordinates from an external file (a, aa). The
structure can be manipulated by the commands sub, m, name and del. The com-
mand sy allows you to define the molecular symmetry while desy and syndi try to
determine automatically the symmetry group of a given molecule.
There exists a structure library which contains the Cartesian coordinates of selected
molecules, e.g. CH4 . These data can be obtained by typing for example a ! ch4 or
a ! methane. The data files are to be found in the directory $TURBODIR/structures.
The library can be extended.
You can perform a geometry optimization at a force field level to preoptimize the
geometry. For this purpose the Universal Force Field (UFF) developed from Rappé
et al. in 1992 [10] has been implemented in the uff module (see also Section 5.4).
This can also be used to calculate an analytical approximate Cartesian Hessian. If
one does so, the start Hessian for the ab initio geometry optimization is this Hessian
instead of the diagonal one ($forceinit on carthess for relax module).

Recommendation

Here is an easy way to get internal coordinates, which should work.

Have coord ready before calling define. In the main geometry menu proceed as
follows to define redundant internals:

a coord read coord

desy determine symmetry, if you expect a higher symmetry, repeat with in-
creased tolerance desy 0.1 , you may go up to desy 1..

ired get redundant internals

* quit main geometry menu

To define internals:

a coord read coord

desy determine symmetry

i go to internal coordinate menu

iaut automatic assignment of bends etc.

q to quit bond analysis

imet to get the metric, unnecessary internals are marked d now. If #ideg = #k
in the head line you are done. Otherwise this did not work.

<enter> go back to main geometry menu

* quit main geometry menu

72 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

To define cartesians:

a coord read coord

desy determine symmetry

* quit main geometry menu

4.1.1 Description of commands

Main Geometry Menu

In the headline of this menu you can see the current number of atoms and molecular
symmetry (we use an input for PH3 as example). The commands in this menu will
now be described briefly:

sy Definition of the Schönflies symbol of the molecular point group sym-

metry. If you enter only sy, define will ask you to enter the symbol,
but you may also directly enter sy c3v. define will symmetrize the
geometry according to the new Schönflies symbol and will create new
nuclei if necessary. You therefore have to take care that you enter
the correct symbol and that your molecule is properly oriented.
All TURBOMOLE programs require the molecule to be in a standard orien-
tation depending on its point group. For the groups Cn , Cnv , Cnh , Dn ,
Dnh and Dnd the z-axis has to be the main rotational axis, secondary
(twofold) rotational axis is always the x-axis, σv is always the xz-plane
and σh the xy-plane. Oh is oriented as D4h . For Td , the threefold rota-
tional axis points in direction (1,1,1) and the z-axis is one of the twofold
axes bisecting one vertex of the tetrahedron.

desy desy allows you to determine the molecular symmetry automatically.

The geometry does not need to be perfectly symmetric for this command
to work. If there are small deviations from some point group symmetry
(as they occur in experimentally determined structures), desy will rec-
ognize the higher symmetry and symmetrize the molecule properly. If
symmetry is lower than expected, use a larger threshold: <eps> up to
1.0 is possible.

syndi syndi works just like desy, but only recognizes point groups with non-
reducible irreducible representations (that is D2h and its subgroups).
This is designed for some features of TURBOMOLE which can only handle
such point groups. Note that the orientation of the molecule might be
different from what you expect. For example, the xy plane will be a
mirror plane of NH3 (which is assigned Cs symmetry).

susy susy leads you through the complete subgroup structure if you want to
lower symmetry, e.g. to investigate Jahn–Teller distortions. The molecule
4.1. THE GEOMETRY MAIN MENU 73

is automatically reoriented if necessary.

Example: Td → D2d → C2v → Cs .

ai You may enter Cartesian atomic coordinates and atomic symbols inter-
actively. After entering an atomic symbol, you will be asked for Carte-
sian coordinates for this type of atom until you enter *. If you enter
&, the atom counter will be decremented and you may re-define the last
atom (but you surely won’t make mistakes, will you?). After entering
*, define asks for the next atom type. Entering & here will allow you
to re-define the last atom type and * to leave this mode and return to
the geometry main menu. Enter q as atom symbol if you want to use a
dummy center without nuclear charge. Symmetry equivalent atoms are
created immediately after you entered a set of coordinates.
This is a convenient tool to provide e.g. rings: exploit symmetry group
Dnh to create an n-membered planar ring by putting an atom on the
x-axis.

a file You may also read atomic coordinates (and possibly internal coordinates)
from file, where file must have the same format as the data group $coord
in file control.
The Cartesian coordinates and the definitions of the internal coordinates
are read in free format; you only have to care for the keywords $coord
and (optionally) $intdef and (important!) for the $end at the end of
the file. The atomic symbol follows the Cartesian coordinates separated
by (at least) one blank. For a description of the internal coordinate
definitions refer to 4.1.2.
Entering ‘!’ as first character of file will tell define to take file from the
structure library. (The name following the ‘!’ actually does not need to
be a filename in this case but rather a search string referenced in the
structure library contents file, see Section 4.1).

aa file same as a, but assumes the atomic coordinates to be in Å rather than

a.u.

sub This command allows you to replace one atom in your molecule by an-
other molecule. For example, if you have methane and you want to create
ethane, you could just substitute one hydrogen atom by another methane
molecule. The only requirement to be met by the substituted atom is
that it must have exactly one bond partner. The substituting molecule
must have an atom at the substituting site; in the example above it would
not be appropriate to use CH3 instead of CH4 for substitution. Upon
substitution, two atoms will be deleted and the two ones forming the
new bond will be put to a standard distance. define will then ask you
to specify a dihedral angle between the old and the new unit. It is also
possible to use a part of your molecule as substituting unit, e.g. if you
have some methyl groups in your molecule, you can create further ones
74 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

by substitution. Some attention is required for the specification of this

substituting unit, because you have to specify the atom which will be
deleted upon bond formation, too. If you enter the filename from which
the structure is to be read starting with ‘!’, the file will be taken from
the structure library (see Section 4.1). Definitions of internal coordinates
will be adjusted after substitution, but no new internal coordinates are
created.

i This command offers a submenu which contains everything related to

internal coordinates. It is further described in Section 4.1.2.

m This command offers a submenu which allows you to manipulate the

molecular geometry, i.e. to move and rotate the molecule or parts of it.
It is further described in Section 4.1.3.

frag Here, the fragments will be defined as being used by the jobbsse script
in order to do a calculation osing the counter-poise correction scheme.
In this menu, up to three monomers can be defined, together with their
charges and their symmetry. When assigning atom numbers to frag-
ments, if x is entered instead of a number, the program will request
the first and last atoms of a range. This will be useful for very large
fragments.

w file The command w writes your molecular geometry and your internal coor-
dinates to file. Afterwards you will be back in the geometry main menu.
If the filename entered starts with ‘!’, the structure will be written to
the structure library.

name name allows you to change atomic identifiers turning, e.g. oxygen atoms
into sulfur atoms. After entering the identifier to be changed (remember
the double quotation marks : "c ring"), you will be asked to enter the
new one. You can use question marks for characters not to be changed,
e.g. you enter "??ring" to change c chain to c ring. If you do not
enter eight characters, your input will be filled up with trailing blanks.

del The command del allows you to delete one or more atoms. After you
entered the atomic list, define will show you a list of all atoms con-
cerned and will ask you to confirm deleting these atoms. If any internal
coordinate definitions exist, which rely on some of the deleted atoms,
these definitions will be deleted, too.

banal The command banal allows you to perform a bonding analysis, that
is, define will try to decide which atoms are bonded and which are
not (according to a table of standard bond lengths which is included in
the code of define). You must have performed this command before
you can use the display commands disb (display bonding information)
or disa (display bond angle information). The standard bond lengths
(and the bonding analysis available from these) are also needed for the
4.1. THE GEOMETRY MAIN MENU 75

commands sub and iaut (see internal coordinate menu, Section 4.1.2).
If you want to change the standard bond lengths (or define more bond
lengths, because not for all possible combinations of elements a standard
length is available) you can do that by creating your own file with the
non-default values and by specifying its full path name in file .sys.data.
The file has the following simple format:

c - h 2.2
h - h 2.0
. - . ...

The format of the entries is almost arbitrary: the two element symbols
have to be separated by a bar, the new bond distance follows in free
format (in atomic units). If the file cannot be read properly, a warning
message is displayed.

* This command leaves this first main menu and writes all data generated
so far to file. The default output file is the file you choose in the first
question during your define session (usually control). Now the data
groups $coord and $intdef will be written to file. After leaving this
menu, you will enter the atomic attributes menu, which is described in
Section 4.2.

4.1.2 Internal Coordinate Menu

INTERNAL COORDINATE MENU ( #ideg=6 #k=2 #f=0 #d=0 #i=0 )

imet <a> : PROVIDE B-MATRIX FOR ACTIVE INTERNAL COORDINATES

(CHECK COMPLETENESS AND NUMERICAL QUALITY
AND CHANGE REDUNDANT INTERNALS TO display)
idef : SUB-MENU FOR INTERACTIVE DEFINITION OF INTERNAL COORDINATES
ideg <a> : OUTPUT NUMBER OF TOT. SYMMETRIC INTERNAL DEGREES OF FREEDOM
iaut : TRY AUTOMATIC DEFINITION OF INTERNAL COORDINATES
iman <a> : MANIPULATE GEOMETRY BY CHANGING INTERNAL COORDINATE VALUES
imanat <i>: AS iman BUT STARTING AT INTERNAL COORD. NUMBER i
ic <i> <x>: CHANGE STATUS OF INTERNAL COORDINATE <i> TO <x>
e.g. ic 5 d TO MAKE 5TH COORD. display OR ic k d
irem <i> : REMOVE INTERNAL COORDINATE <i>,
e.g. irem d TO REMOVE ALL display COORDS
dis : ANY DISPLAY COMMAND e.g. disi OR disc
disiat <i>: AS disi BUT STARTING AT INTERNAL COORD. NUMBER i

WHERE <a>= OPTIONAL ATOMIC SET (DEFAULT=all)

<i>= INDEX(LIST) OF INTERNAL COORDINATE(S) LIKE 3-6,8 OR <i>=<x>
<x>= STATUS OF INTERNAL COORDINATE = k, f, d OR i
ADDING A QUESTION MARK TO ANY COMMAND MAY PROVIDE EXPLANATIONS

ENTER COMMAND OR HIT >return< TO GET BACK TO GEOMETRY MAIN MENU

76 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

The parameters in the headline of this menu have the following meanings:

#ideg is the total number of symmetry restricted degrees of freedom.

#k is the number of active internal coordinates specified up to now. Only

these coordinates are optimized during a geometry optimization.

#f is the number of fixed internal coordinates specified. These coordinates

will be included in the B-matrix (see command imet), but their values
will not be changed during geometry optimization.

#d is the number of internal coordinates whose values will only be displayed

(e.g. by command disi), but no gradients will be calculated for these
coordinates nor will they be included in the geometry optimization.

#i means the number of coordinates which are defined, but will be com-
pletely ignored, i.e. they are not even displayed on the screen and will
not be used by any program (this is the waste-paper-basket of define).

Note that the #k plus #f must equal the number of degrees of freedom (#ideg)
of your molecule, if you want to perform a geometry optimization. If you have less
coordinates than degrees of freedom, you will have to specify further ones (commands
idef or iaut, see below); if you have more coordinates than degrees of freedom, you
will have to throw away some of them (commands irem or imet, see below).
The commands in this menu allow you to define internal coordinates for your molecule,
adjust your geometry to special values of these internal coordinates and to control the
numeric reliability of the chosen set of internal coordinates. In detail, the commands
act as follows.

Description of commands

imet a This command computes the so-called B-matrix, which is the matrix
of the derivatives of the (active and fixed ) internal coordinates with re-
spect to Cartesian coordinates. This matrix is used in program relax
for the conversion from Cartesian coordinates and gradients to internal
ones and vice versa. If this matrix is singular (or even nearly singular)
this indicates a linear dependency of your internal coordinate set. As
a consequence the geometry update (more exactly the transformation
of the updated internal coordinates into Cartesian ones) will fail. This
may also happen in the course of a geometry optimization if the coordi-
nates run into linear dependency during their optimization. imet checks
the B-matrix and removes linear dependent internal coordinates from
your list (their status is changed from #k or #f to #d). If B is only
near singular, a warning is issued and the lowest eigenvalue(s) as well as
the corresponding eigenvector(s) are displayed. In this case, you should
try to find better internal coordinates (although this may not always be
possible). After the command imet, there may be too few (active plus
4.1. THE GEOMETRY MAIN MENU 77

fixed) internal coordinates, but certainly not too many (because linear
dependencies have been eliminated). Perhaps you will have to add new
ones or—better!—try command iaut or ired in the preceding menu.
Command imet should be used always after creating internal coordinates
with iaut or idef (especially after iaut, because this command creates
usually an overcomplete set of internal coordinates).

idef idef unfolds a little submenu where you can define internal coordinates
manually. The exact procedure of the definition will be described below
in a separate section.

ideg a This command gives you the number of symmetry restricted degrees of
freedom (for the atomic set specified by a). Without symmetry, this is
just 3N − 6, where N is the number of atoms, but if there is symmetry,
some of these degrees of freedom will violate symmetry and therefore
are not valid. For geometry optimizations, only the symmetry allowed
degrees of freedom are needed, because the symmetry requirements are
imposed anyway. In connection with the optional atomic set a this com-
mand can help you to find out, in which part of a complicated molecule
internal coordinates are missing, if you fail to get the full number of
#ideg (which equals the result of ideg all) for the molecule as a whole.

iaut iaut tries an automatic definition of internal coordinates. This command

relies on an recursive procedure which tries to simplify the molecule as
far as possible and then starts the definition of internal coordinates. At
present not all molecular topologies are supported, therefore it may hap-
pen that no internal coordinates can be assigned to your molecule or at
least a part of it. However, for all cases in which an automatic assign-
ment of coordinates is possible, iaut has up to now proved to provide
very good internal coordinates. If iaut works for your molecule (and in
most non-pathological cases it will) we recommend strongly to use these
coordinates, as they may help you to save several cycles in the geometry
optimization procedure. After creating internal coordinates with iaut
you should always use imet (see above), because iaut may provide an
overcomplete set of coordinates. All coordinates which conflict with the
molecular symmetry are set to ignore by iaut.

iman a iman allows you to modify the values of internal coordinates. If you
specify a list of atoms a only those internal coordinates which refer to
only these atoms will be handled. You will get a list of all (active and
fixed) internal coordinates and their current values and you will be able
to enter a new value for each of them if you like. Default (<enter>)
keeps the value shown. Be aware that all distances are given in atomic
units (1 a.u. = 52.9 pm).

ic i x This option allows you to change the status of a coordinate, e.g. from
active to display or every other combination. The syntax is ic 5 d, if
78 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

coordinate no. 5 is to be set to display, or ic k d, if all active coordinates

are to be set to display.
irem i This option allows you to delete definitions of internal coordinates from
your list. The indices of the internal coordinates always refer to the full
list of coordinates including display and ignore coordinates. To make
sure you delete the right ones, use disi. Also the indices will imme-
diately change if you delete coordinates. If you want to delete several
coordinates, this is therefore done most easily if you delete them in order
of descending indices (because deletion of a coordinate has only an effect
on the coordinates with higher indices). After choosing the coordinates
to be deleted, a list of all coordinates involved will be displayed and you
will be asked to confirm deletion.
The syntax is simply irem 5 to delete internal coordinate no. 5 or irem d
to remove all ‘display’ coordinates.

Hitting <return> will bring you back to the geometry main menu.

Interactive Definition of Internal Coordinates

If you choose idef in the internal coordinate menu, you will get the following infor-
mation:

ENTER INTERNAL COORDINATE DEFINITION COMMAND

<x> <type> <indices>
WHERE <x> = k f d i
<type> = stre invr bend outp tors linc linp
comp ring pyrm bipy pris cube octa
THESE COMMANDS WILL BE EXPLAINED IN DETAIL IF YOU ENTER
<x> <type>? FOR SOME CHOICE OF <x> AND <type>, E.G. k stre ?
DEFAULT=GO BACK TO INTERNAL MAIN MENU DISPLAY=dis

The <x> means the status (see page 75) of the internal coordinate entered (k, f, d,
i). The syntax is:

k stre 1 2
d tors 3 6 2 7
f bend 3 4 5
i outp 3 4 7 9

Note that in the third example atom 5 is the central atom of the angle!

Specification of available internal coordinates

The following types of coordinates are available:

stre The stre (for stretch) describes a distance between two atoms. It needs
only two atomic indices to be given, the order of which is arbitrary.
4.1. THE GEOMETRY MAIN MENU 79

invr The invr coordinate (for inverse r) describes an inverse distance. The
declaration is the same as for stre, but in some cases (if you are far away
from the minimum) the use of invr may result in better convergence.

bend bend describes a bond angle. It requires three atoms to be specified, of

which the third one is the atom at the apex.

outp Out-of-plane angle: outp abcd is the angle between bond a − d and plane
b − c − d.

tors Dihedral angle: tors abcd is the angle between the planes a − b − c and
b − c − d.

linc This is a special coordinate type to describe the bending of a near-linear

system. linc abcd describes the collinear bending of a − b − c (where the
angle is defined as for bend: the apex atom appears last) in the plane of
b − c − d (see also below, command linp). The system b − c − d has to
be non-linear, of course.

linp This coordinate is similar to linc, but describes the bending of a − b − c

perpendicular to the plane b − c − d. These two types of coordinates are
in most cases sufficient to describe the bending of near-linear systems.
An example may help you to understand these two coordinate types.
Consider ketene, H2 CCO, which contains a linear system of three atoms.
Without symmetry, this molecule has 9 degrees of freedom. You could
choose the four bond lengths, two CCH angles and the out-of-plane an-
gle of the C–C bond out of the CHH–plane. But then two degrees of
freedom still remain, which cannot be specified using these normal co-
ordinate types. You can fix these by using linc and linp. The two
coordinates linc 1 3 2 4 and linp 1 3 2 4 (where 1=oxygen, 2=car-
bon, 3=carbon, 4=hydrogen) would solve the problem.

comp The type comp describes a compound coordinate, i.e. a linear combina-
tion of (primitive) internal coordinates. This is often used to prevent
strong coupling between (primitive) internal coordinates and to achieve
better convergence of the geometry optimization. The use of linear com-
binations rather than primitive coordinates is especially recommended
for rings and cages (see ref. [46]). Command iaut uses linear combina-
tions in most cases.
After you entered k comp n where n is the number of primitive internal
coordinates to be combined, you will be asked to enter the type of the
coordinate (stre, bend, . . . ). Then you will have to enter the weight (the
coefficient of this primitive coordinate in the linear combination) and the
atomic indices which define each coordinate. The definition of the prim-
itive coordinates is the same as described above for the corresponding
coordinate types. It is not possible to combine internal coordinates of
different types.
80 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

ring This type helps you to define special ring coordinates. You only have to
enter k ring n where n is the ring size. Then you will be asked for the
atomic indices of all atoms which constitute the ring and which must be
entered in the same order as they appear in the ring. The maximum
number of atoms in the ring is 69 (but in most cases the ring size will be
limited by the maximum number of atoms which is allowed for define).

Hitting <return> will bring you back to the internal coordinate menu where you can
see the new number of internal coordinates in the headline.

4.1.3 Manipulating the Geometry

Note that the molecular geometry can be modified with the iman command of the
internal coordinate menu described earlier, if internal coordinates has been defined.
Another option is to select m in the geometry main menu which provides the following
submenu:

CARTESIAN COORDINATE MANIPULATION MENU :

move : TRANSLATE AND/OR ROTATE PART OF THE MOLECULE
align : ROTATE MOLECULE TO ALIGN TWO AXES
inert : MOVE MOLECULE SO THAT COORDINATE AXES BECOME
PRINCIPAL AXES OF INERTIA
mback : RESTORE PREVIOUS MOLECULAR GEOMETRY
dis : DISPLAY MOLECULAR GEOMETRY
YOU MAY APPEND A QUESTION MARK TO ANY OF THESE COMMANDS
FOR FURTHER EXPLANATIONS.
HIT >return< OR USE ANY GEOMETRY COMMAND NOT IN THIS LIST
TO TERMINATE THIS MENU.
UPON TERMINATION THE MOLECULAR SYMMETRY WILL BE ENFORCED
ACCORDING TO SYMMETRY GROUP c3v .

The meaning of the commands inert and mback should be clear; the commands move
and align allow you to manipulate the geometry of your molecule. After entering
move or align, you will be asked to specify a set of atoms on which the command
shall act. You can use this to manipulate only a part of your molecule, e.g. if you are
building a structure from subunits and you want to adjust their relative arrangement.
As long as you stay in this menu, the molecular symmetry needs not be correct (so
that you can try different movements and/or rotations), but as soon as you leave
it, the geometry will be symmetrized according to the present Schönflies symbol.
In move, after you specified the atomic set to be considered, you get the following
information:

INPUT DIRECTION OF MOVEMENT OR LOCATION OF ROTATION AXIS

EITHER AS A COORDINATE TRIPLE SEPARATED BY BLANKS,
OR AS TWO ATOMIC INDICES SEPARATED BY KOMMA, OR x OR y OR z
OR ENTER ANY DISPLAY COMMAND FIRST OR & TO GO BACK

You can thus specify the direction of movement (or the rotational axis) in the form
0. 0. 1. or simply z (which both describes the z-axis) or 1.3256 -3.333 0.2218
for an arbitrary axis. If you want to specify an axis which is related to your molecule,
4.2. THE ATOMIC ATTRIBUTES MENU 81

you may also enter two atomic indices which define it. After having specified the
axis, you have to enter the distance of movement and the angle of rotation. If you
want to perform a simple rotation, enter 0 for the distance of movement and if you
want to simply move your structure, enter 0 for the rotational angle.
In the align submenu, you are asked for two directions (“current orientation” and
“new direction”) and you can use the same syntax as described above. The molecule
will be rotated such that the “current orientation” axis points into the new direction.
For example, first entering 1,2 and then z rotates (a part of) the molecule such that
the bond between atom 1 and atom 2 is parallel to the z axis.
You can leave this menu and return to the geometry main menu by hitting <return>
or by entering any command of the geometry main menu.

4.2 The Atomic Attributes Menu

After you specified the molecular geometry and symmetry and wrote this data to
file, you will encounter the atomic attributes menu, which is the second of the four
main menus. You will enter this menu, if all necessary data cannot be read from your
input file or if you do not use an input file. This menu deals with the specification
of basis sets and other data related to the atom type:

ATOMIC ATTRIBUTE DEFINITION MENU ( #atoms=5 #bas=5 #ecp=0 )

b : ASSIGN ATOMIC BASIS SETS

bb : b RESTRICTED TO BASIS SET LIBRARY
bl : LIST ATOMIC BASIS SETS ASSIGNED
bm : MODIFY DEFINITION OF ATOMIC BASIS SET
bp : SWITCH BETWEEN 5d/7f AND 6d/10f
lib : SELECT BASIS SET LIBRARY
ecp : ASSIGN EFFECTIVE CORE POTENTIALS
ecpb : ecp RESTRICTED TO BASIS SET LIBRARY
ecpi : GENERAL INFORMATION ABOUT EFFECTIVE CORE POTENTIALS
ecpl : LIST EFFECTIVE CORE POTENTIALS ASSIGNED
ecprm: REMOVE EFFECTIVE CORE POTENTIAL(S)
c : ASSIGN NUCLEAR CHARGES (IF DIFFERENT FROM DEFAULTS)
cem : ASSIGN NUCLEAR CHARGES FOR EMBEDDING
m : ASSIGN ATOMIC MASSES (IF DIFFERENT FROM DEFAULTS)
iso : ASSIGN ISOTOPE FOR NUCLEAR COUPLING CALCULATION
dis : DISPLAY MOLECULAR GEOMETRY
dat : DISPLAY ATOMIC ATTRIBUTES YET ESTABLISHED
h : EXPLANATION OF ATTRIBUTE DEFINITION SYNTAX
* : TERMINATE THIS SECTION AND WRITE DATA OR DATA REFERENCES TO control
GOBACK=& (TO GEOMETRY MENU !)

The headline gives you the number of atoms, the number of atoms to which basis
sets have already been assigned and the number of atoms to which effective core
potentials have already been assigned. Most of the commands in this menu deal
with the specification of basis sets and pseudopotentials.
82 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

Basis sets available

The following basis sets are available on $TURBODIR/basen/, which you may inspect
to see which other basis sets are supported automatically. The corresponding publi-
cations can be found here 1.3.

SV(P) or def-SV(P) for routine SCF or DFT. Quality is about 6–31G*.

TZVP or def-TZVP for accurate SCF or DFT. Quality is slightly better than
6–311G**.

TZVPP or def-TZVPP for MP2 or close to basis set limit SCF or DFT. Comparable
to 6–311G(2df).

QZVP and QZVPP for highly correlated treatments; quadruple zeta + 3d2f1g or
4d2f1g (beyond Ne), 3p2d1f for H.

These basis sets are available for atoms H–Kr, and the split-valence (SV) and valence-
triple-ζ (TZV) basis sets types with ECPs also for Rb–Rn, except lanthanides.
For calculations with the programs ricc2, ccsdf12, and pnoccsd optimized auxiliary
basis sets are available for the basis sets SV(P), SVP, TZVP, TZVPP, and QZVPP.
NEW: New sets of basis functions, partly identical with those mention above, de-
noted def2-XYZ are available for atoms H–Rn [9]. The def2 basis sets for 5p and
6p block elements are designed for small core ECPs (ECP-28, ECP-46 and ECP-
60). For each family, SV, TZV, and QZV, we offer two sets of polarisation functions
leading to:

def2-SV(P) and def2-SVP

def2-TZVP and def2-TZVPP

def2-QZVP and def2-QZVPP

We strongly recommended the new def2-basis, since they have been shown to provide
consistent accuracy across the periodic table.

Recommendation

Use the same basis set type for all atoms; use ECPs beyond Kr since this accounts
for scalar relativistic effects.
New basis sets (def2-XYZ): MP2 implies RI-MP2 and RICC2

exploratory MP2: SVP

almost quantitative DFT: SV(P), HF: SVP, MP2: TZVPP; properties (HF and
DFT): TZVPP

quantitative DFT: TZVP, HF: TZVPP, MP2: QZVPP

4.2. THE ATOMIC ATTRIBUTES MENU 83

basis set limit DFT: QZVP, HF: QZVP

If you want a better basis than SV(P), assigned automatically, use b all def2-TZVP
(or another basis). The assignment can be checked by bl.
Diffuse functions should only be added if really necessary. E.g. for small anions or
treatment of excited states use: TZVP instead of SVP + diffuse. This is more accurate
and usually faster. Only for excited states of small molecules or excited states with
(a partial) Rydberg character add additional diffuse functions (e.g. by using the aug-
cc-pVTZ basis) as well as the keyword diffuse, for more information, see page 393
in the keyword section.
[Old basis sets (def-XYZ): For standard correlated calculations (MP2, RI-MP2,
RI-CC2) use the doubly-polarized TZVPP (or def-TZVPP) basis.]

Correlation-Consistent (Dunning) Basis Sets

Dunning basis sets like cc-pVDZ, cc-pVTZ, cc-pVQZ are also supported, e.g. by
b all cc-pVTZ. But these basis sets employ generalized contractions for which TURBOMOLE
is not optimized. This has in particular strong effects on the performance of all pro-
grams which use 4-index electron repulsion integrals, for RI-MP2 and RI-CC2 this
is partially compensated by the RI-approximation.
The following correlation consistent basis sets are available in the TURBOMOLE basis
set library:

cc-pVXZ standard valence X-tuple zeta basis sets (X = D, T, Q, 5, 6); available

for H, He, Li–Ne, Na–Ar, K, Ca, Ga–Kr.
(cc-pV6Z only for H, He, B–Ne, Al–Ar; for Al–Ar also the recom-
mended newer cc-pV(X+d)Z sets are available)

cc-pwCVXZ-PP weighted core-valence x-tuple zeta basis sets (X= D, T, Q, 5) are

available for post-d main group elements Ga–Kr, In–Xe, and Tl–Rn.
(also pure valence basis sets cc-pVXZ-PP are available for these ele-
ments, but it is not recommended to use them)

cc-pwCVXZ weighted core-valence X-tuple zeta basis sets (X = D, T, Q, 5); avail-

able for B–Ne, Al–Ar, and Ga–Kr
(for Al–Ar also the recommended combination of the cc-pV(X+d)Z
sets with the core valence functions (wC), i.e. the cc-pwCV(X+d)Z
basis set are available)

aug- diffuse functions for combination with the basis sets cc-pVXZ, cc-
pV(X+d)Z, cc-pwCVXZ, cc-pV(X+d)Z, cc-pVXZ-PP or cc-pwCVXZ-
PP; available for H, He, B–Ne, Al–Ar with X = D–6 and Ga–Kr,
In–Xe, and Tl–Rn with X = D–5.

cc-pVXZ-F12 with X = D, T, Q for use with the explicitly-correlated F12 variants

of wavefunction methods (MP2-F12, CCSD(F12*), etc.)
84 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

For calculations with the programs that employ the RI approximation with a cor-
related wavefunction optimized auxiliary basis sets are available for most of the
correlation consistent basis set series.

4.2.1 Description of the commands

b With b you can specify basis sets for all atoms in your molecule. After
entering b you will be asked to specify the atoms to which you want
to assign basis sets. You can do this in the usual ways (refer to Sec-
tion 4.0.4), including all and none. Then you will be asked to enter the
nickname of the basis set to be assigned. There are two principal ways
to do this:
1) If you are in the append mode, the nickname you entered will be
appended to the atomic symbol of the element under consideration.
This is especially useful if you want to assign basis sets to different
atoms with one command. For example, if you want to assign basis
sets to hydrogen and oxygen atoms and you enter only DZ, the basis
sets h DZ and o DZ will be read from the basis set library.
2) If you are in the non-append mode, no atomic symbol will be in-
serted in front of the nickname entered. Therefore you have to enter
the full basis set nickname, e.g. h DZ. This mode is advantageous if
you want to assign basis sets to dummy centers (i.e. points without
nuclear charge but with basis functions, e.g. for counterpoise calcu-
lations) or if you want to use the basis set nickname none (which
means no basis functions at this atom).
You can switch between the two modes with ‘+’ (switches to append
mode) and ‘-’ (switches to non-append mode).
Once you have specified your basis set nickname, define will look in the
standard input file (normally control) for this basis set. If it can not
be found there, you can switch to the standard basis set library (if you
did not use a standard input file, the standard library will be searched
immediately). If the basis set cannot be found there, you are asked
either to enter a new standard library (which will be standard only until
you leave this menu) or another input file, where the basis set can be
found. If you do not know the exact nickname of your basis set, you may
abbreviate it by ‘?’, so you could enter h DZ? to obtain basis sets like
h DZ, h DZP, h DZ special, etc. define will give you a list of all basis
sets whose nicknames match your search string and allows you to choose
among them. You may also use the command list to obtain a list of all
basis sets cataloged.
bb bb does essentially the same as b but does not search your default input
file for basis sets. Instead it will look in the basis set library immediately.
bl bl gives you a list of all basis sets assigned so far.
4.2. THE ATOMIC ATTRIBUTES MENU 85

bm This command is used to modify basis sets which are already assigned.
The corresponding submenu is self-explanatory, but we recommend to
change directly the file basis.

bp The TURBOMOLE programs normally work with basis sets of 5d -functions

(which means they delete the s-component of the full 6d -set). bp allows
to switch between the proper 5d /7f -set and the Cartesian 6d /10f -set.

ecp This command allows you to specify effective core potentials for some
atoms. The assignment works exactly like the specification of basis sets
(see above).

ecpb This one does the same as command ecp, but restricted to the basis set
library (the input file will not be used).

ecpi ecpi gives you some general information about what type of pseudopo-
tentials is supported. For more information we refer to [47] and references
therein.

ecpl ecpl gives you a list of all pseudopotentials assigned so far.

ecprm ecprm allows to remove a pseudopotential assignment from the list. This
command is useful if you want to perform an all electron calculation after
an ECP treatment.

c Command c assigns a special nuclear charge to an atom. This is useful

to define dummy centers for counterpoise calculations where you set the
nuclear charge to zero.

m This command allows you to assign non-default atomic masses to an

atom. Use this if you want to analyze isotopic shifts of calculated har-
monic frequencies. The standard masses are those of the natural isotope
mix.

iso This command allows you to assign non-default gyromagnetic ratios to

an atom. For practically all isotopes, it is sufficient to enter the mass
number (sum of protons and neutrons) to obtain the gyromagnetic ratio.

dat dat gives you a list of all data already specified.

* This is again the usual command to leave a menu and write all data to
file control (or any other output file). It is not possible to leave this
menu unless basis sets have been specified for all atoms in your molecule.
If you do not want to use a basis set for one or more atoms, use the basis
set nickname none. On leaving this menu, the data groups $atoms and
$basis will be written to the output file.

After you finished this menu, you will enter the third main menu of define which
deals with start vectors and occupation numbers.
86 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

4.3 Generating MO Start Vectors

4.3.1 The MO Start Vectors Menu

This menu serves to define the occupation numbers, and to generate the start vectors
for HF-SCF and DFT calculations. They may be constructed from earlier SCF
calculations (perhaps employing another basis set, type use), by Hamilton core guess
(hcore), or by an extended Hückel calculation which can be performed automatically
(eht). An occupation of the start orbitals will be proposed and can be modified if
desired.

OCCUPATION NUMBER & MOLECULAR ORBITAL DEFINITION MENU

CHOOSE COMMAND
infsao : OUTPUT SAO INFORMATION
atb : Switch for writing MOs in ASCII or binary format
eht : PROVIDE MOS && OCCUPATION NUMBERS FROM EXTENDED HUECKEL GUESS
use <file> : SUPPLY MO INFORMATION USING DATA FROM <file>
man : MANUAL SPECIFICATION OF OCCUPATION NUMBERS
hcore : HAMILTON CORE GUESS FOR MOS
flip : FLIP SPIN OF A SELECTED ATOM
& : MOVE BACK TO THE ATOMIC ATTRIBUTES MENU
THE COMMANDS use OR eht OR * OR q(uit) TERMINATE THIS MENU !!!
FOR EXPLANATIONS APPEND A QUESTION MARK (?) TO ANY COMMAND

Recommendation

You will normally only need to enter eht. For the EHT-guess, define will ask for
some specifications, and you should always choose the default, i.e. just <enter>. Of
importance is only the molecular charge, specified as 0 (neutral, default), 1 or -1
etc.
Based on the EHT orbital energies define proposes an occupation. If you accept
you are done, if not you get the “occupation number assignment menu” explained
in 4.3.2.

Description of Commands

infsao Command infsao provides information about the symmetry adapted

basis which is used for the SCF-calculation. To exploit the molecular
symmetry as efficiently as possible, TURBOMOLE programs do not use the
simple basis which you specified during your define session. Instead
it builds linear combinations of equal basis functions on different—but
symmetry equivalent—atoms. This basis is then called the SAO-basis
(Symmetry Adapted Orbital). It has the useful property that each basis
function transformed to this basis transforms belongs to one irreducible
representation of the molecular point group (that is, the basis reflects the
full molecular symmetry as specified by the Schönflies symbol). infsao
4.3. GENERATING MO START VECTORS 87

gives you a listing of all symmetry adapted basis functions and their
constituents either on file or on the screen. This may help you if you
want to have a closer look at the SCF vectors, because the vector which
is output by program dscf is written in terms of these SAOs.

atb Molecular orbitals can be written either in ASCII or in binary format.

This command switches from one option to the other, and it is highly
recommended to read which setting is currently active. ASCII format
is portable and allows the usage of TURBOMOLEinput files on different
systems with incompatible binary format. Binary format is faster and
smaller files will be written. The external program atbandbta can be
used to transform existing mos, alpha, and beta files from ASCII to
binary format and vice versa.

eht eht performs an extended Hückel calculation for your molecule. The
orbital energies available from this calculation are then used to provide
occupation numbers for your calculation and the Hückel MOs will be
projected onto the space that is spanned by your basis set. This start-
vectors are not as good as the ones which may be obtained by projection
of an old SCF vector, but they are still better than the core Hamilto-
nian guess that is used if no start vectors are available. When using this
command, you will be asked if you want to accept the standard Hückel
parameters and to enter the molecular charge. Afterwards you will nor-
mally get a list of the few highest occupied and lowest unoccupied MOs,
their energies and their default occupation. If you don’t want to accept
the default occupation you will enter the occupation number assignment
menu, which is described in Section 4.3.2. Note that the occupation
based on the Hückel calculation may be unreliable if the difference of the
energies of the HOMO and the LUMO is less than 0.05 a.u. (you will
get a warning). You will also have to enter this menu for all open-shell
cases other than doublets.

use file With command use you are able to use information about occupied MOs
and start vectors from a former calculation on the same molecule. file
should be the path and name of the control file of this former calcula-
tion, of which all data groups related to occupation numbers and vectors
will be read. As the new generated data will overwrite the existing data if
both resist in the same directory, it is best and in some cases necessary to
have the data of the former calculation in another directory than the one
you started the define session in. Then just type use <path>/control
to construct a new SCF vector from the data of the old calculation, with-
out changing the old data. The data groups $closed shells and $open
shells will be taken for your new calculation and the SCF vector from
the old calculation will be projected onto the space which is spanned by
your present basis set. These start vectors are usually better than the
ones you could obtain by an extended Hückel calculation.
88 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

man man allows you to declare occupation numbers or change a previous dec-
laration manually. After selecting this command, you will get a short
information about the current occupation numbers:

---------------------------------------------------------
actual closed shell orbital selection range
---------------------------------------------------------
a1 # 1- 18
a2 # 1- 1
e # 1- 13
---------------------------------------------------------

any further closed-shell orbitals to declare ? DEFAULT(y)

If you answer this question with y, you enter the orbital specification
menu which will be described in Section 4.3.3.
The same procedure applies to the open-shell occupation numbers after
you finished the closed-shell occupations.

hcore hcore tells programs dscf and ridft to run without a start vector (it
writes the data group $scfmo none to file control). dscf or ridft will
then start from the core Hamiltonian start vector, which is the vector
obtained by diagonalizing the one-electron Hamiltonian. Note that you
still have to specify the occupation numbers. This concerns only the
first SCF run, however, as for the following calculations the converged
vector of the previous iteration will be taken. A SCF calculation with
a core Hamiltonian start vector typically will take 2 − 3 iterations more
than a calculation with an extended Hückel start vector (a calculation
with the converged SCF vector of a previous calculation will need even
less iterations, depending on how large the difference in the geometry
between the two calculations is).

flip flipping of spins at a selected atom. Requirements: converged UHF

molecular orbitals and no symmetry (C1 ). definewill localize the or-
bitals, assign them to the atoms and give the user the possibility to
choose atoms at which alpha-orbitals are moved to beta orbitals, or vice
versa. This is useful for spin-broken start orbitals, but not for spatial
symmetry breaking.

* This command (as well as use and eht) terminates this menu, but with-
out providing a start vector. If the keyword $scfmo exists in your input
file, it will be kept unchanged (i.e. the old vector will be taken), other-
wise $scfmo none will be inserted into your output file, which forces a
calculation without start vector to be performed. When you leave this
menu, the data groups $closed shells, $open shells (optionally) and
$scfmo will be written to file. You will then reach the last of the four
main menus (the General Menu) which is described in Section 4.4.
4.3. GENERATING MO START VECTORS 89

4.3.2 Assignment of Occupation Numbers

If an automatic assignment of occupation numbers is not possible or you do not

except the occupation numbers generated by the EHT, you enter the following menu:

OCCUPATION NUMBER ASSIGNMENT MENU ( #e=60 #c=0 #o=0)

s : CHOOSE UHF SINGLET OCCUPATION

t : CHOOSE UHF TRIPLET OCCUPATION
u <int> : CHOOSE UHF WITH <int> UNPAIRED ELECTRONS
l <list> : PRINT MO’S FROM EHT IN <list>, (DEFAULT=ALL)
p <index> : PRINT MO-COEFFICIENTS OF SHELL <index>
c <list> : CHOOSE SHELLS IN <list> TO BECOME CLOSED SHELLS
o <index> : CHOOSE SHELL <index> TO BECOME AN RHF OPEN SHELL
a <list> : CHOOSE SHELLS IN <list> TO BECOME UHF ALPHA SHELLS
b <list> : CHOOSE SHELLS IN <list> TO BECOME UHF BETA SHELLS
v <list> : CHOOSE SHELLS IN <list> TO BECOME EMPTY SHELLS
& : REPEAT THE EXTENDED HUECKEL CALCULATION
* : SAVE OCCUPATION NUMBERS & GO TO NEXT ITEM
dis : GEOMETRY DISPLAY COMMANDS
e : CALCULATE EHT-ENERGY
f : FURTHER ADVICE
<int> = INTEGER
<index> = INDEX OF MO-SHELL ACCORDING TO COMMAND s
<list> = LIST OF MO-SHELL INDICES (LIKE 1-5,7-8,11)

Recommendation

Enter l to get a list of EHT MO energies. Then make up your mind on what to do:
closed shell, RHF open shell (not allowed for DFT) or UHF. Look at the examples
below.

RHF c 1-41,43,45 to define these levels to be doubly occupied.

UHF a 1-5 alpha levels to be occupied, b 1-3,5 beta levels to be occupied.

Or simply, s, t, or u 1 to get singlet, triplet or doublet occupation
pattern.

ROHF c 1-41,43,45 levels to be doubly occupied; o 42 level 42 should be

partially occupied. You will then be asked to specify the occupation. If
there are more open shells you have to repeat, since only a single open
shell can be specified at a time. Watch the headline of the menu, which
tells you the number of electrons assigned to MOs.

Description of Commands

s list This command gives you a listing of all MOs and their energies as ob-
tained from the extended Hückel calculation. For NH3 in C3v and TZVP
you get, e.g.:
90 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

ORBITAL SYMMETRY ENERGY SHELL CUMULATED CL.SHL OCC. OP.SHL OCC.

(SHELL) TYPE DEGENERACY SHELL DEG. PER ORBITAL PER ORBITAL
1 1a1 -15.63244 2 2 0.0000 0.0000
2 2a1 -0.99808 2 4 0.0000 0.0000
3 1e -0.64406 4 8 0.0000 0.0000
4 3a1 -0.57085 2 10 0.0000 0.0000
5 2e 0.30375 4 14 0.0000 0.0000
6 4a1 0.87046 2 16 0.0000 0.0000
TO CONTINUE, ENTER <return>

p index This allows you to get the linear combination of basis functions which
form the MO-index. Note that this refers not to the basis set you spec-
ified, but to the extended Hückel basis. index must be a single index,
not an index list.
c list This command allows you to specify closed shells. Their occupation will
be 2 per MO, the total occupation the shell degeneracy which you can
obtain by using command s. list is a list of shell indices like 1-13 or
1,3-5,7.
o index This command allows you to specify open shells. index must be a single
shell index, not an index list. You will then be asked for the number of
electrons per MO which shall be contained in this shell. For example,
for a fluorine atom you should choose o n (where n is the index of the
p-shell) and an occupation of 5/3 per MO. You may enter the occu-
pation numbers as simple integers or as integer fractions, e.g. 1 for the
s-occupation in sodium, 5/3 for the p-occupation in fluorine.
v list With this command you can remove an orbital occupation, if you speci-
fied a wrong one. list is again a list of shell indices in usual syntax.
& This command has a different meaning in this menu than in the rest of
define. Here it will repeat the extended Hückel calculation (perhaps
you want to change some Hückel parameters for the next one).
* * will not bring you back to the occupation numbers menu, but will
terminate the whole occupation number and start vector section and
will bring you to the last main menu, which is described in Section 4.4.
If you want to leave this menu without assigning all electrons in your
molecule to shells, define will issue a warning and suggest to continue
defining occupation numbers. You can ignore this warning, if you do not
want to assign all electrons.
e Calculates and displays the extended Hückel total energy of your molecule.
f f will give you some information about the commands in this menu.

You may overwrite occupation numbers once given by just redefining the correspond-
ing shell. For example, if you choose shells 1–10 as closed shells and afterwards shell
no. 9 as open shell (with any occupation number), the open shell will be correctly
assigned.
4.3. GENERATING MO START VECTORS 91

4.3.3 Orbital Specification Menu

define provides the possibility to assign the occupation numbers of the MOs man-
ually, if you like. To do that, use the command man in the occupation number main
menu and you will arrive at the following submenu:

------------- ORBITAL SPECIFICATION MENU --------------

<label> <list> : select orbitals within <list>

-<label> <list> : skip orbitals within <list>
& : ignore input for last label
clear : clear all assignments
p(rint) : print actual orbital selection
for help, type ? or help // for quit, type * or q(uit)

Depending on whether you are in the closed- or in the open-shell section, the com-
mands of this menu refer only to the corresponding type of orbitals. The commands
of this menu do not need much explanation. <label> is the irrep label of one irre-
ducible representation of the molecular point group (e.g. a1, b2, t1g, . . . ). <list> is
a list of orbital indices within this irrep (e.g. 1,2,4 or 1-8,10,11). p or print will
give you the same listing of the orbital occupations as you saw before entering this
menu. After you leave this submenu, you will be back in the occupation numbers
main menu.

4.3.4 Roothaan Parameters

In open-shell calculations within the restricted Hartree–Fock ansatz (ROHF), the
coupling between the closed and the open shells must be specified using two param-
eters a and b, which depend on the type of the open shell, the number of electrons in
it (the electron configuration), but also on the state to be calculated. For example,
there are three states arising from the s2 p2 configuration of an atom (3 P, 1 D, 1 S)
which have different values of a and b. For the definition of these parameters and
their use refer to Roothaan’s original paper [48]. For simple cases, define sets these
parameters automatically. If not, you have to enter them yourself. In this case, you
will get the following message:

ROOTHAAN PARAMETERS a AND b COULD NOT BE PROVIDED ...

TYPE IN ROOTHAAN a AND b AS INTEGER FRACTIONS
OR ENTER val FOR AN AVERAGE OF STATES CALCULATION
OR ENTER & TO REPEAT OCCUPATION NUMBER ASSIGNMENT

Note that not all open shell systems can be handled in this way. It is possible to
specify a and b for atomic calculations with sn , pn , d1 , and d9 configurations and
for calculations on linear molecules with π n and δ n configurations. Furthermore, it
is possible to do calculations on systems with half-filled shells (where a=1, b=2).
In the literature you may find tabulated values for individual states arising from
dn configurations, but these are not correct. Instead, these are parameters for an
average of all states arising from these configurations. You can obtain these values
if you enter val on the above question. For a detailed description see Section 6.3.
92 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

4.3.5 Start-MOs for broken symmetry treatments ("flip")

Broken-symmetry treatments suggested by e.g. Noodleman or Ruiz are a popular

tool for the calculation of spin coupling parameters in the framework of DFT. As
an example one might consider two coupled CuII centers, e.g. for a (hypothetical)
arrangement like this:

$coord
0.0 2.7 0.0 cu
0.0 -2.7 0.0 cu
0.0 -6.1 0.0 f
0.0 6.1 0.0 f
2.4 0.0 0.0 f
-2.4 0.0 0.0 f
$end

The high-spin case, a doublet with an excess alpha electron at each Cu atom, "aa"
in an obvious notation, preserves D2h symmetry, while the low spin state "ba" does
not. For a broken-symmetry treatment, it is advisable to calculate the high-spin
state first, and then broken-symmetry state(s); from the energy difference(s) one
may calculate approximate values for the spin-spin coupling parameters as described
by e.g. the above authors. Access to broken-symmetry states usually is possible by
the choice of appropriate start-MOs, followed by an SCF-procedure. Start MOs may
be obtained by first applying a localization procedure to the MOs of the high-spin
state and then by "moving" localized alpha orbitals to the beta subset.
The preparation of broken-symmetry start-MOs can be done with define (semi-
)automatically. Prerequisite is a converged wave function for the high-spin state in
C1 -symmetry, that fulfills the auf bau principle.
If in this case one enters flip in the orbital definition menu, define selects the
occupied valence orbitals of the system (by an orbital energy criterion, which one
can usually accept, unless the system is highly charged and the orbital energies are
shifted). Next a Boys orbital localization procedure is carried out, which - depending
on the size of the problem - may take some time. Then the user is asked:
ENTER INDICES OF ATOMS OR ELEMENT TO BE MANIPULATED (example: 1,2-3 or
"mn")
In case of our above example one may enter "cu", which immediately leads to the
following output (a def-SV(P) basis and the B-P functional were used for the high-
spin state):

RELEVANT LMOS FOR ATOM 1 cu

ALPHA:
index occupation "energy" s p d f (dxx dyy dzz dxy dxz dyz)
15 1.000 -0.357 0.01 0.00 0.98 0.20 0.27 0.01 0.50 0.00 0.00
18 1.000 -0.357 0.01 0.00 0.98 0.20 0.27 0.01 0.50 0.00 0.00
20 1.000 -0.335 0.00 0.00 1.00 0.00 0.00 0.00 0.00 1.00 0.00
22 1.000 -0.333 0.01 0.00 0.99 0.13 0.03 0.32 0.00 0.00 0.51
4.3. GENERATING MO START VECTORS 93

23 1.000 -0.333 0.01 0.00 0.99 0.14 0.03 0.34 0.00 0.00 0.49

BETA:
39 1.000 -0.326 0.00 0.00 1.00 0.33 0.08 0.09 0.00 0.00 0.50
41 1.000 -0.326 0.00 0.00 1.00 0.33 0.08 0.09 0.00 0.00 0.50
43 1.000 -0.321 0.00 0.00 1.00 0.00 0.00 0.00 0.00 1.00 0.00
46 1.001 -0.318 0.05 0.00 0.95 0.00 0.43 0.51 0.00 0.00 0.00

RELEVANT LMOS FOR ATOM 2 cu

ALPHA:
index occupation "energy" s p d f (dxx dyy dzz dxy dxz dyz)
16 1.000 -0.357 0.01 0.00 0.98 0.20 0.27 0.01 0.50 0.00 0.00
17 1.000 -0.357 0.01 0.00 0.98 0.20 0.27 0.01 0.50 0.00 0.00
19 1.000 -0.335 0.00 0.00 1.00 0.00 0.00 0.00 0.00 1.00 0.00
21 1.000 -0.333 0.01 0.00 0.99 0.13 0.03 0.32 0.00 0.00 0.51
24 1.000 -0.333 0.01 0.00 0.99 0.14 0.03 0.34 0.00 0.00 0.49

BETA:
40 1.000 -0.326 0.00 0.00 1.00 0.33 0.08 0.09 0.00 0.00 0.50
42 1.000 -0.326 0.00 0.00 1.00 0.33 0.08 0.09 0.00 0.00 0.50
44 1.000 -0.321 0.00 0.00 1.00 0.00 0.00 0.00 0.00 1.00 0.00
45 1.001 -0.318 0.05 0.00 0.95 0.00 0.43 0.51 0.00 0.00 0.00

a2b : FLIPPING ALPHA TO BETA (default)

b2a : FLIPPING BETA TO ALPHA
r : repeat atom choice

As evident from the second column, for each Cu atom five localized alpha and four
localized beta orbitals were generated which are of d-type (the sixth column labelled
"d" shows values close to 1, the other columns such close to 0). The six columns at
the right show the individual contributions of the six Cartesian d-functions.
What has to be done to generate start MOs for the "ba"-case? Obviously one of the
five localized alpha spin orbitals from the first Cu atom (atom label 1 cu) has to
become a beta spin orbital. These five orbitals have the indices 15, 18, 20, 22,
23. In order to avoid linear dependencies, it is advisable to take the orbital that
has no beta-analogue. This can be found by comparing the contributions of the six
d-functions. In the present example this is the case for the localized alpha orbitals 15
and 18: in contrast to all localized beta orbitals they show significant contributions
from dxy . One thus enters

a2b
15

and after confirming the replacement of original MOs with the generated start-MOs
one is finally asked

It is advisable to modify damping and orbital shift in the following way:

94 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

$scfdamp start=5.000 step=0.050 min=0.500

$scforbitalshift automatic=1.0
$scfiterlimit 999

Do you want to replace the corresponding entries in the control-file? (y)

which should be confirmed, as otherwise the prepared spin state might be destroyed
during the SCF iterations.
From this input one may start the SCF(HF/DFT)-procedure. For recommended
choices of DFT functionals and formulae to calculate the coupling parameters from
these energy differences please consult the papers of the above-mentioned authors.
For reasons of economy, a pre-optimization by a pure (non-hybrid) DFT-functional
is reasonable.
Important: For the converged wave function one should check, whether the resulting
state is really the desired one. This can quite reliably be done by a Mulliken popu-
lation analysis. For this purpose, add $pop to the control file, type ridft -proper
or dscf -proper, respectively, and check the signs of the calculated numbers of
unpaired electrons in the output.

4.4 The General Options Menu

After you specified all data concerning the molecule you want to examine, you are
on your way to the last of the four main menus. Before reaching it, you will perhaps
get a message like the following:

DO YOU WANT TO DELETE DATA GROUPS LIKE

$energy
$grad
$hessian
$hessian (projected)
$last energy change
$maximum norm of internal gradient
$dipgrad
$vibrational normal modes
$vibrational spectrum
$cartesianforce interspace
LEFT OVER FROM PREVIOUS CALCULATIONS ? DEFAULT(n)

define has scanned your input file for this session and found some data groups which
might have become obsolete. If they are still acceptable depends on the changes
you made during your present define session. They are obviously incorrect if you
changed the molecule under consideration; but any change in the basis sets or the
occupation numbers will make them dangerous, too, because you might not know
some day if they really refer to the basis set which is defined in this control file. As
a rough guide, delete them whenever you have made changes in one of the first three
main menus during your define session.
4.4. THE GENERAL OPTIONS MENU 95

After that you will reach the last main menu of define which helps you to control
the actions of all TURBOMOLE programs. The meanings of the various options are
explained in more detail in the description of the individual programs, therefore only
a short explanation will be given here.
Now have a look at the menu:

GENERAL MENU : SELECT YOUR TOPIC

scf : SELECT NON-DEFAULT SCF PARAMETER
mp2 : OPTIONS AND DATA GROUPS FOR rimp2 and mpgrad
pnocc : OPTIONS AND DATA GROUPS FOR pnoccsd
cc : OPTIONS AND DATA GROUPS FOR ricc2
ex : EXCITED STATE AND RESPONSE OPTIONS
prop : SELECT TOOLS FOR SCF-ORBITAL ANALYSIS
drv : SELECT NON-DEFAULT INPUT PARAMETER FOR EVALUATION
OF ANALYTICAL ENERGY DERIVATIVES
(GRADIENTS, FORCE CONSTANTS)
rex : SELECT OPTIONS FOR GEOMETRY UPDATES USING RELAX
stp : SELECT NON-DEFAULT STRUCTURE OPTIMIZATION PARAMETER
e : DEFINE EXTERNAL ELECTROSTATIC FIELD
dft : DFT Parameters
ri : RI Parameters
rijk : RI-JK-HF Parameters
rirpa : RIRPA Parameters
gw : OPTIONS AND DATA GROUPS FOR GW (escf)
senex : seminumeric exchange parameters
hybno : hybrid Noga/Diag parameters
dsp : DFT dispersion correction
nmr : NMR shift parameters
ncoup : NMR coupling parameters
trunc : USE TRUNCATED AUXBASIS DURING ITERATIONS
marij : MULTIPOLE ACCELERATED RI-J
dis : DISPLAY MOLECULAR GEOMETRY
list : LIST OF CONTROL FILE
& : GO BACK TO OCCUPATION/ORBITAL ASSIGNMENT MENU
* or q : END OF DEFINE SESSION

This menu serves very different purposes. The next subsection deals with commands
required to activate and/or specify specific methods of calculation. The subsequent
subsection describes commands used to select non-default options. Standard SCF
calculations do not require special action, just leave the menu. The final subsection
describes the settings for property calculations.

4.4.1 Important commands

DFT calculations

Command dft leads you to the menu:

STATUS OF DFT_OPTIONS:
96 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

DFT is NOT used

functional b-p
gridsize m3

ENTER DFT-OPTION TO BE MODIFIED

func: TO CHANGE TYPE OF FUNCTIONAL

grid: TO CHANGE GRIDSIZE
on: TO SWITCH ON DFT
Just <ENTER>, q or ’*’ terminate this menu.

To activate DFT input on and then specify the grid for the quadrature of exchange-
correlation terms. TURBOMOLE offers grids 1 (coarse) to 7 (finest), and the multiple
grids m3 to m5 [7]. The latter employ a coarser grid during SCF iterations, and grid
3 to grid 5 in the final SCF iteration and the gradient evaluation. Default is grid m3,
for clusters with more than 50 atoms use m4.
The functionals supported within define are obtained with the command func:
4.4. THE GENERAL OPTIONS MENU 97

SURVEY OF AVAILABLE EXCHANGE-CORRELATION ENERGY FUNCTIONALS

FUNCTIONAL | TYPE | EXCHANGE | CORRELATION | REFERENCES

---------------------------------------------------------------------
s-vwn | LDA | S | VWN(V) | 1-3
s-vwn_Gaussian | LDA | S | VWN(III) | 1-3
pwlda | LDA | S | PW | 1,2,4
b-lyp | GGA | S+B88 | LYP | 1,2,6
b-vwn | GGA | S+B88 | VWN(V) | 1-3,5
b-p | GGA | S+B88 | VWN(V)+P86 | 1-3,5,7
pbe | GGA | S+PBE(X) | PW+PBE(C) | 1,2,4,8
tpss | MGGA | S+TPSS(X) | PW+TPSS(C) | 1,2,4,14
bh-lyp | HYB | 0.5(S+B88) | LYP | 1,2,5,6,9
| | +0.5HF | |
b3-lyp | HYB | 0.8S+0.72B88 | 0.19VWN(V) | 1-3,5,6,10
| | +0.2HF | +0.81LYP |
b3-lyp_Gaussian | HYB | 0.8S+0.72B88 | 0.19VWN(III) | 1-3,5,6,10
| | +0.2HF | +0.81LYP |
pbe0 | HYB | 0.75(S+PBE(X)) | PW+PBE(C) | 1,2,4,8,11
| | +0.25HF | |
tpssh | MHYB | 0.9(S+TPSS(X)) | PW+TPSS(C) | 1,2,4,14,15
| | +0.1HF | |
pw6b95 | MHYB | PW91(X)+0.28HF | PW+B95(C) | 19
m06 | MHYB | M06(X)+0.27HF | M06(C) | 20
m06-l | MGGA | M06-L(X) | M06-L(C) | 20
m06-2x | MHYB | M06-2X(X) | M06-2X(C) | 20
| | +0.54HF | |
lhf | ODFT | E-EXX | | 12,13
oep | ODFT | EXX | | 18
b97-d | GGA | B97 refit | B97 refit | 16
pbeh-3c | HYB | PBE0 refit | PBE0 refit | 21
b97-3c | GGA | B97-D refit | B97 refit | 26
lh07t-svwn | LHYB | S@t-LMF | VWN(V) | 22
lh07s-svwn | LHYB | S@s-LMF | VWN(V) | 23
lh12ct-ssirpw92 | LHYB | S@ct-LMF | sirPW92 | 24
lh12ct-ssifpw92 | LHYB | S@ct-LMF | sifPW92 | 24
lh14t-calpbe | LHYB | S+0.507PBE | PW92+0.451PBE | 25
| | +pig1@t-LMF | |
b2-plyp | DHYB |0.47(SB88) |0.73LYP+0.27PT2 | 17
| | +0.53HF | |
cam-b3lyp | RSH |b(B88)+aHF |0.81LYP+0.19VWN5| 27

This list is far from being complete. Especially with the inclusion of the XCFun and
LibXC libraries, uncounted combinations of exchange and correlation functionals
are available. To get an overview of further functionals and how to use arbitrary
combinations and exact exchange portions, please read section 6.2.
Default is b-p, i.e. B-P86, which is probably best for the whole of Chemistry [49]. For
main group compounds we recommend b3-lyp; note that GAUSSIAN uses partly
different implementations [49].
98 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

The programs dscf and grad are used to carry out conventional DFT treatments,
i.e. J and K are evaluated without approximations.

RI-J calculations

For non-hybrid functionals we strongly recommend the RI-J procedure, which speeds
up calculations by a factor 10 at least (as compared to conventional treatments)
without sacrificing accuracy. Command ri gives:

STATUS OF RI-OPTIONS:
RI IS NOT USED
Memory for RI: 200 MB
Filename for auxbasis: auxbasis

ENTER RI-OPTION TO BE MODIFIED

m: CHANGE MEMORY FOR RI
f: CHANGE FILENAME
jbas: ASSIGN AUXILIARY RI-J BASIS SETS
on: TO SWITCH ON RI
Use <ENTER>, q, end, or * to leave this menu

Activate RI-J with on, and choose with m the memory you can dedicate to store
three-center integrals (Keyword: $ricore), default is 200 MB. The more memory,
the faster the calculation.
A rough guide: put $ricore to about 2/3 of the memory of the computer. Use OS
specific commands (top on most UNIX systems), during an ridft run to find the
actual memory usage and then adjust $ricore, the keyword in control specifying
memory.
If the option jbas is selected, define enters a submenu which allows the assignment
of auxiliary basis sets (for an explanation of the menu items see Section 4.2). Where
available, the program will select by default the auxiliary basis sets optimized for the
orbital basis used. Please note that treatment of systems with diffuse wavefunctions
may also require an extension of the auxiliary basis. For this cases enlarge the sets
of s- and p-functions with diffuse functions.
The RI-J option is only supported by programs ridft and rdgrad, if you use jobex
to optimize molecular geometry, put: nohup jobex -ri ...

MARI-J option

RI-J calculations can be done even more efficiently with the Multipole Accelerated
RI-J (MARI-J ) option, especially for larger molecules where almost linear scaling
is achieved [50].

Parameters:
1) precision parameter: 1.00E-06
2) maximum multipole l-moment: 10
3) maximum number of bins: 8
4.4. THE GENERAL OPTIONS MENU 99

4) minimum separation of bins: 0.00

5) maximum allowed extension: 20.00
6) threshold for multipole neglect: 1.00E-18

Enter the number to change a value or <return> to accept all.

Just rely on the defaults.

Multiple auxiliary basis sets

With the command trunc you can switch on this option. Effect: a reduced auxiliary
(or fitting) basis to represent the electron density is employed during SCF iterations,
the final SCF iteration and the gradient are computed with the full auxiliary basis.

truncated RI ALREADY SWITCHED ON

DO YOU WANT TO SWITCH OFF truncation ? (default=no)

Note: trunc is presently not compatible with marij!

RI in SCF calculations

Considerable savings in CPU times are achieved with the RI technique for both
Coulomb J and exchange K terms in SCF calculations, the RI-JK method [51],
provided large basis sets are employed, e.g. TZVPP, cc-pVTZ, or cc-pVQZ. With rijk
you get:

STATUS OF RI-OPTIONS:
RI IS NOT USED
Memory for RI: 200 MB
Filename for auxbasis: auxbasis

ENTER RI-OPTION TO BE MODIFIED

m: CHANGE MEMORY FOR RI
f: CHANGE FILENAME
jkbas: ASSIGN AUXILIARY RI-JK BASIS SETS
on: TO SWITCH ON RI
Use <ENTER>, q, end, or * to leave this menu

For an explanation of the menu items see Section 4.4.1. RI-JK calculations can be
carried out with the program ridft.

Optimization to minima and transition structures using Statpt

Structure optimizations can be carried out by the program statpt. For minimiza-
tions no additional keywords are required. The default values are assumed, which
work in most of the cases. Structure optimization is performed in internal coordi-
nates if they have been set. Otherwise, Cartesian coordinates are used. One can
switch the optimization in internal coordinates on or off, respectively in internal
100 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

redundant or Cartesian coordinates. For transition structure optimizations the in-

dex of transition vector has to be set to an integer value > 0 (0 means structure
minimization). The value of the index specifies transition vector to follow during
the saddle point search. Note, that Hessian eigenpairs are stored in ascending or-
der of the eigenvalues, i.e. the eigenpair with the smallest eigenvector has the index 1.

The command stp gives:

------------------------------------------------------------------------
CONVERGENCE CRITERIA:

thre 1.000000E-06 thre : threshold for ENERGY CHANGE

thrd 1.000000E-03 thrd : threshold for MAX. DISPL. ELEMENT
thrg 1.000000E-03 thrg : threshold for MAX. GRAD. ELEMENT
rmsd 5.000000E-04 rmsd : threshold for RMS OF DISPL.
rmsg 5.000000E-04 rmsg : threshold for RMS OF GRAD.

defl : set default values.

------------------------------------------------------------------------

OPTIMIZATION refers to

int off int: INTERNAL coordinates

rdn off rdn: REDUNDANT INTERNAL coordinates
crt on crt: CARTESIAN coordinates
NOTE : options int and crt exclude each other

ENTER STATPT-OPTIONS TO BE MODIFIED

itvc 0 itvc : change INDEX OF TRANSITION VECTOR

updte bfgs updte: change method of HESSIAN UPDATE
hsfrq 0 hsfrq: frequency of HESSIAN CALCULATION
kptm 0 kptm : FREEZING transition vector INDEX
hdiag 5.000000E-01 hdiag: change DIAGONAL HESSIAN ELEMENTS
rmax 3.000000E-01 rmax : change MAX. TRUST RADIUS
rmin 1.000000E-04 rmin : change MIN. TRUST RADIUS
trad 3.000000E-01 trad : change TRUST RADIUS
------------------------------------------------------------------------
Just <ENTER>, q or ’*’ terminate this menu.

Excited states, frequency-dependent properties, and stability analysis

Excited state calculations with RPA or CIS (based on HF-SCF) and TDDFT pro-
cedures as well as stability analyses (SCF or DFT) are carried out by the program
escf.
You will need a well converged HF-SCF or DFT calculation that were converged to
4.4. THE GENERAL OPTIONS MENU 101

at least $scfconv=7, see Section 4.4.2.

Details of calculations are specified with the command ex:

MAIN MENU FOR RESPONSE CALCULATIONS

OPTION | STATUS | DESCRIPTION

-------------------------------------------------------------------
rpas | off | RPA SINGLET EXCITATIONS (TDHF OR TDDFT)
ciss | off | TDA SINGLET EXCITATIONS (CI SINGLES)
rpat | off | RPA TRIPLET EXCITATIONS (TDHF OR TDDFT)
cist | off | TDA TRIPLET EXCITATIONS (CI SINGLES)
polly | off | STATIC POLARIZABILITY
dynpol | off | DYNAMIC POLARIZABILITY
single | off | SINGLET STABILITY ANALYSIS
triple | off | TRIPLET STABILITY ANALYSIS
nonrel | off | NON-REAL STABILITY ANALYSIS
bse | off | BETHE-SALPETER EX.
cbse | off | corr-aug. BETHE-SALPETER EX.

ENTER <OPTION> TO SWITCH ON/OFF OPTION, * OR q TO QUIT

If you have selected an option, e.g. rpas, and quit this menu, you will get another
menu:

SELECT IRREP AND NUMBER OF STATES

ENTER ? FOR HELP, * OR Q TO QUIT, & TO GO BACK

This should be self-evident. The input of the excitations is used to estimate the
memory requirements. The GW method is available in a separate gw menu:

NO GW OPTIONS CURRENTLY SET

GW (gw on/off) : off
RI-GW (rigw on/off) : off
RPA (rpa on/off) : off

INPUT MENU FOR GW CALCULATIONS WITH escf:

gw : TURN STANDARD GW USING SPECTRAL FUNCTIONS (N^6) ON/OFF
rigw : TURN RI-GW USING ANALYTIC CONTINUATION (N^4) ON/OFF
rpa : TURN RPA MODULE (N^6) ON/OFF
Just <ENTER>, q or ’*’ terminate this menu.

MP2, MP2-F12, RI-MP2, and PNO-MP2

We recommend to use MP2 for standard applications together with the RI technique
using the ricc2 program. The mpgrad program is supplied for special benchmark
application where the RI approximation needs to be avoided. The pnoccsd program
is meant only for very large (> 100 atoms and > 3000 basis functions) MP2 and
MP2-F12 single point energy calculations. This is more efficient and supports the
frozen core option in the gradient calculation.
102 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

The entry mp2 leads to a submenu which allows to set some keywords for MP2 and
RI-MP2 calculations, e.g. defining frozen orbitals, maximum memory usage, or assign
auxiliary basis sets for RI-MP2 calculations, etc. If you want to use ricc2, you have
to use the entry cc and the submenu ricc2 in order to assign MP2 as wavefunction
model. For the pnoccsd program you have to use the entry pnocc and the submenu
pnoccsd to assign the wavefunction model.
Conventional MP2 calculations with mpgrad require a number of additional settings
for which it is recommended to invoke the interactive tool mp2prep. For geometry
optimizations with jobex use nohup jobex -level mp2 -ri ...

CC2 and CCSD calculations

The entry cc leads to a submenu which allows to set a number of keywords essen-
tial for calculations with the programs ricc2 and ccsdf12. In particular it allows
the assignment of the wavefunction method(s), selection of auxiliary basis sets, the
specification of frozen orbitals, and the definition of a scratch directory and of the
maximum core memory usage.
The ricc2 program must be used for excitation energies and response properties with
second-order methods (MP2, CIS(D), ADC(2), CC2, etc. and their spin-scaled vari-
ants), while the ccsdf12 program must be used for third- and higher-order methods
(MP3, CCSD, CCSD(T), etc.).

2nd analytical derivatives for SCF and DFT

The program aoforce computes force constants and IR and Raman Spectra on SCF
and DFT level. Analytical second derivative calculations can directly be started
from converged SCF or DFT calculations. Note, that the basis is restricted to d-
functions, and ROHF as well as broken occupation numbers are not allowed. For
better efficiency, in case of larger systems, use the keyword $maxcor as described in
Chapter 15 to reduce computational cost. RI will be used if the RI option for DFT
has been specified.

4.4.2 Special adjustments

Adjustments described by the following menus are often better done directly in the
control file; have a look at the keywords in Chapter 22. For common calculations
just start with the defaults, and change keywords directly in control if you encounter
problems with your calculation.

SCF options

ENTER SCF-OPTION TO BE MODIFIED

conv : ACCURACY OF SCF-ENERGY $scfconv

4.4. THE GENERAL OPTIONS MENU 103

thi : INTEGRAL STORAGE CRITERIA $thize $thime

ints : INTEGRAL STORAGE ALLOCATION $scfintunit
iter : MAXIMUM NUMBER OF ITERATIONS $scfiterlimit
diis : DIIS CONVERGENCE ACCELERATION $scfdiis
damp : OPTIONS FOR DAMPING $scfdamp
shift : SHIFTING OF ORBITALS $scforbitalshift
order : ORDERING OF ORBITALS $scforbitalorder
fermi : THERMAL SMEARING OF OCC. NUMBERS $fermi
pdiag : PREDIAGONALIZATION $prediag
soghf : SPIN ORBIT GENERALIZED SCF $soghf
x2c : EXACT 2C HAMILTONIAN $rx2c
rlocal : DLU SCHEME FOR X2C $rlocal
finnuc : FINITE NUCLEUS MODEL $finnuc
rsym : RELATIVISTIC SYMMETRY $rsym
snso : SCREENED NUCLEAR SPIN ORBIT $snso

By the command $fermi you can switch on smearing of occupation numbers, and
thus automatically optimize occupations and spin. The command $soghf guides the
user for the different settings of a two-component calculation. Relativistic effects are
included by ECPs or relativistic all-electron methods, see section 6.4.

Menu nmr and ncoup

This menu sets the flags for NMR shielding and coupling constant calculations. The
nmr menu allows to switch between two different CPHF solvers for mpshiftand to
set the corresponding keywords for both, for details see chapter 17.

ENTER SCF-OPTION TO BE MODIFIED

Old CPHF solver is selected (default up to V7.4)
Convergence is checked for the heaviest atom by default

STATUS OF NMR OPTIONS

The old CPHF solver (up to V7.4) is used
A threshold of 0.1000D-01 ppm is used
Maximum number of iterations is set to 30
NMR shieldings are calculated for all atoms

ENTER OPTION TO BE MODIFIED

oldcphf : to switch on or off old CPHF solver (up to V7.4)
csconv : to select convergence criterion for old solver
shiftconv : to select convergence criterion for default solver
csmaxiter : to select maximum number of iterations
nucsel : to calculate certain nuclei only
gimic : prepare input for GIMIC calculation
Just <ENTER>, q or * to terminate this menu.

The ncoup menue is used to control the input for a NMR coupling constant calculion
with escf.

STATUS OF OPTIONS FOR COUPLING CONSTANTS

104 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

The following terms will be calculated: fc sd pso dso fcsdcross

The threshold for the final output is 0.1000
Wave function convergence criteria is set to 9
Response convergence criteria is set to 6
Maximum number of iterations is set to 25
Response equations will be solved for all atoms
Right-hand side integrals will be calculated for all atoms

ENTER OPTION TO BE MODIFIED

fc/sd/pso/dso on/only/off: set terms to be calculated
fcsdcross/all on/off
fromfile/simple
thr : threshold for final output
scfconv : convergence criterion for wave function (recommended: >= 7)
rpaconv : convergence criterion for response equations (recommended: >= 6)
limit : maximum number of iterations in response equations ($escfiterlimit)
nucsel : nuclei to solve response equations for
nucsel2 : nuclei to calculate right-hand side integrals for
Just <ENTER>, q or * to terminate this menu

The settings are discussed in detail in chapter 8.

Menu drv

The most important of the derivative menus is the first one which tells the programs
which derivatives to calculate. This is only necessary for special purposes and you
should better not change default options.

------------------------------------------------------------------------
derivative data groups ’$drvopt, $drvtol’
------------------------------------------------------------------------
option | status | description :
------------------------------------------------------------------------
crt | T | CARTESIAN 1st derivatives
sec | T | CARTESIAN 2nd derivatives
bas | F | energy derivatives with respect to
| | BASIS SET exponents/scaling factors/
| | contraction coefficients
glb | F | energy derivative with respect to
| | a GLOBAL scaling factor
dip | T | cartesian 1st derivatives of DIPOLE MOMENT
pol | T | nuclear contribution to POLARIZABILITY
fa | F | SPECTROSCOPIC ANALYSIS only
tol 0.100D-06 derivative integral cutoff
------------------------------------------------------------------------
use <opt> for enabling, -<opt> for disabling of logical switches
<&> will bring you back to GENERAL MENU without more changes
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

The handling of these options is very simple. With the exception of tol, all are
logical switches which are either true (or on, active) or false (or off, inactive). You
4.4. THE GENERAL OPTIONS MENU 105

can switch between the two states if you enter, for example, crt (to switch calculation
of Cartesian first derivatives on) or -crt (to switch it off). The options crt, sec and
bas should provide no problems. glb refers to a global scaling factor for all basis
set exponents. Imagine that you would like to replace your basis set, which contains
basis functions

χµ = (x − x0 )l (y − y0 )m (z − z0 )n exp −ηµ (r − r0 )2
 

by another basis set which contains basis functions

χµ = (x − x0 )l (y − y0 )m (z − z0 )n exp −αηµ (r − r0 )2
 

where α is the same for all primitive basis functions χµ . With command glb you are
able to calculate analytical derivatives of the total energy with respect to α and can
thus easily determine the optimum α.
dip enables you to calculate the first derivatives of the electric dipole moment with
respect to nuclear displacements which gives you infrared intensities. pol allows you
to calculate the contribution of the nuclear rearrangement on the electric polariz-
ability. fa finally performs only a frequency analysis which means that aoforce will
read the force constant matrix ($hessian or $hessian (projected)), diagonalize it
and give you the frequencies and normal modes. tol is not a logical switch as the
other options in this menu, but a cutoff threshold for the derivative integrals, i.e.
integrals below this threshold will be neglected in the derivative calculations.
Entering * will bring you to the second derivative submenu.

Debug Options for the Derivative Programs

The following menu deals only with some debug options for grad. Use them with
caution, each of them can produce lots of useless output:
------------------------------------------------------------------------
derivative debug options ’$drvdebug’
------------------------------------------------------------------------
option |status| description :
------------------------------------------------------------------------
disp1e | F | display 1e contributions to desired derivatives
only1e | F | calculate 1e contributions to desired derivatives only
debug1e | F | display 1e shell contributions to desired derivatives
| | (WARNING : this produces large outputs!)
debug2e | F | display 2e shell contributions to desired derivatives
| | (WARNING : this produces VERY large outputs!)
debugvib| F | debug switch for vibrational analysis (force only)
notrans | F | disable transfer relations (gradient only!)
novirial| F | disable virial scaling invariance in basis set
| | optimizations (gradient only)
------------------------------------------------------------------------
use <opt> for enabling, -<opt> for disabling option <opt>
<&> will bring you back to GENERAL MENU without more changes
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU
106 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

As there is no need to use these options normally and the menu text is self-explaining,
no further description will be given. Note that all options are logical switches and
may be enabled and disabled the same way as shown for the last menu. Entering *
will bring you to the last derivative submenu.

4.4.3 Relax Options

Program relax has a huge variety of options to control its actions which in program
define are grouped together in eight consecutive menus. These are only briefly
described in the following sections; for a more detailed discussion of the underlying
algorithms refer to the documentation of program relax (see Section 5.3). Only
experts should try to change default settings.

Optimization Methods

The first of the relax subgenus deals with the type of optimization to be performed:
------------------------------------------------------------------------
optimization options for RELAX
------------------------------------------------------------------------
option | status | description : optimization refers to
------------------------------------------------------------------------
int | F | INTERNAL coordinates
crt | F | CARTESIAN coordinates
bas | F | BASIS SET exponents/scale factors
glb | F | GLOBAL scaling factor
------------------------------------------------------------------------
use <opt> for enabling, -<opt> for disabling option <opt>
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

You can choose between a geometry optimization in the space of internal coordi-
nates (in this case you will need definitions of internal coordinates, of course) or
in the space of Cartesian coordinates (these possibilities are mutually exclusive, of
course). Furthermore optimizations of basis set parameters (exponents, contraction
coefficients and scaling factors) or of a global scaling factor is possible (these options
are also exclusive, but can be performed simultaneous to a geometry optimization).
For the geometry optimization you should normally use internal coordinates as they
provide better convergence characteristics in most cases.

Coordinate Updates

The next submenu deals with the way relax updates the old coordinates. You may
choose a maximum change for the coordinates or you can allow coordinate updates
by means of extrapolation:
4.4. THE GENERAL OPTIONS MENU 107

------------------------------------------------------------------------
coordinate update options for RELAX
------------------------------------------------------------------------
dqmax <real> : coordinates are allowed to change by at most
<real> (DEFAULT : 0.3000 ) a.u.
polish : perform an interpolation or extrapolation of
coordinates (DEFAULT :y)
-polish : disable inter/extrapolation
------------------------------------------------------------------------
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

These options result in better convergence of your optimization in most cases.

Interconversion Between Internal and Cartesian Coordinates

The interconversion between internal and Cartesian coordinates is not possible di-
rectly (in this direction). Instead it is performed iteratively. The following options
control this conversion:
--------------------------------------------------------------------
interconversion options for RELAX
--------------------------------------------------------------------
option | description
--------------------------------------------------------------------
on | switch on interconversion (DEFAULT: off)
qconv <r> | set convergence threshold for interconversion
| of coordinates to <r>. DEFAULT : <r> = .1000E-09
iter <i> | allow at most <i> iterations for interconversion
| of coordinates. DEFAULT : <i> = 25
crtint | transform cartesian into internal coordinates (DEFAULT=n)
intcrt | transform internal into cartesian coordinates (DEFAULT=n)
grdint | transform cartesian into internal gradients (DEFAULT=n)
hssint | transform cartesian into internal hessian (DEFAULT=n)
--------------------------------------------------------------------
use -<opt> for disabling any interconversion option
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

The options qconv and iter are used in each normal relax run to determine the char-
acteristics of the back-transformation of coordinates into the internal space. With
the other options and interconversion switched on, you can force relax to perform
only the specified coordinate transformation and write the transformed coordinates
to file control. To achieve this, enter on to switch to the transformation-only mode,
and one of the last four options, e.g. crtint, to specify the desired transformation.

Updating the Hessian

relax provides a variety of methods to generate an updated Hessian every cycle.

This includes the well known methods such as BFGS, DFP, or MS update methods
as well as some less common procedures:
108 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

--------------------------------------------------------------
OPTIONS FOR UPDATING THE HESSIAN
--------------------------------------------------------------
option | status | description
--------------------------------------------------------------
none | F | NO UPDATE (STEEPEST DESCENT)
bfgs | F | BROYDEN-FLETCHER-GOLDFARB-SHANNO UPDATE
dfp | F | DAVIDON-FLETCHER-POWELL UPDATE
bfgs-dfp | F | COMBINED (BFGS+DFP) UPDATE
ms | F | MURTAGH-SARGENT UPDATE
schlegel | F | SCHLEGEL UPDATE
diagup | F | DIAGONAL UPDATE (AHLRICHS/EHRIG)
multidim | F | RANK > 2 BFGS-TYPE UPDATE
ahlrichs | T | MACRO : AHLRICHS UPDATE (DEFAULT)
--------------------------------------------------------------
USE <opt> FOR ENABLING OPTION <opt> AND THUS DISABLING
ALL OTHER OPTIONS.
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

We recommend to use the default method ahlrichs which provides excellent con-
vergency in most cases.

General Boundary Conditions for Update

The force constant matrix will only be updated if least mingeo cycles exist. The
maximum number of cycles used for the update is specified by the parameter maxgeo.
Normally the default values provided by define need not be changed.

DEFINE BOUNDARY CONDITIONS FOR UPDATE

--------------------------------------------------------------
mingeo <i> | START UPDATE IF THERE ARE AT LEAST <i> CYCLES
| DEFAULT : min 3
maxgeo <i> | USE LAST <i> CYCLES FOR UPDATE, DEFAULT : max 4
--------------------------------------------------------------
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

Special Boundary Conditions for Ahlrichs and Pulay Updates

For the default update method ahlrichs some additional control parameters are
available which can be defined in this menu:

DEFINE BOUNDARY CONDITIONS FOR AHLRICHS OR PULAY UPDATE

--------------------------------------------------------------
option | description
--------------------------------------------------------------
modus <i> | DEFINE MODUS FOR GDIIS PROCEDURE : MINIMIZE
| <dq|dq> IF <i> = 0
| <g|dq> IF <i> = 1
| <g|g> IF <i> = 2
| <dE> IF <i> = 3
| DEFAULT : <i> = 1
4.4. THE GENERAL OPTIONS MENU 109

fail <r> | IGNORE GDIIS IF <g|dq> /| <g|dq> | IS

| LARGER THAN -<r>. DEFAULT : <r> = 0.1
--------------------------------------------------------------
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

For detailed description consult Section 5.3.

--------------------------------------------------------------
OPTIONS FOR MANIPULATING THE HESSIAN
--------------------------------------------------------------
option | description
--------------------------------------------------------------
diagonal | RESTRICT UPDATE TO DIAGONAL-ELEMENTS IF
| METHOD IS BFGS,DFP OR MS. DEFAULT=n
offreset | DISCARD OFF-DIAGONAL ELEMENTS. DEFAULT=n
offdamp <r> | DAMP OFF-DIAGONAL ELEMENTS BY 1/(1+<r>) DEFAULT= 1.000
damp <real> | DAMP UPDATE BY 1/(1+<real>), DEFAULT= .0000E+00
scale <real> | SCALE INPUT HESSIAN BY <real>, DEFAULT= 1.000
allow <real> | SCALE INPUT HESSIAN BY <real>/|DE| IF |DE|,
| THE OBSERVED ABSOLUTE CHANGE IN ENERGY, IS
| OBEYING THE CONDITION |DE| > <real> > 0.
| DEFAULT : NO SCALING
min <real> | DO NOT ALLOW EIGENVALUES OF HESSIAN TO DROP
| BELOW <real>. DEFAULT= .1000E-02
reset <real> | USE <real> AS A RESET VALUE FOR TOO SMALL
| EIGENVALUES (CP. min). DEFAULT= .1000E-02
max <real> | DO NOT ALLOW EIGENVALUES OF HESSIAN TO BECOME
| LARGER THAN <real>. DEFAULT= 1000.
--------------------------------------------------------------
WITH THE EXCEPTION OF min,reset AND max, ALL OPTIONS MAY BE
DISABLED BY ENTERING -<opt>
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

Initialization of the Hessian

Finally there are some options to control the choice of the initial Hessian during your
geometry optimization:
110 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

--------------------------------------------------------------
FORCE CONSTANTS INITIALIZATION OPTIONS FOR RELAX
--------------------------------------------------------------
OPTION | DESCRIPTION
--------------------------------------------------------------
off | switch off initialization (DEFAULT: on)
cart | use analytical cartesian hessian provided by a
| 2nd derivatives calculation. DEFAULT(n)
diag | use diagonal matrix with diagonal elements set
| individually within data groups $intdef or
| $basis or $global. DEFAULT(n)
unit <r> | use multiple of the unit matrix ( H = <r>*E ).
| DEFAULT(n) - DEFAULT <r> = 1.000
--------------------------------------------------------------
NOTE THAT THESE OPTIONS ARE MUTUALLY EXCLUSIVE
<RETURN> OR * OR q(uit) WILL TERMINATE THIS MENU

Option off will be used if you have already a good Hessian from a previous calcu-
lation which may be used. cart describes an even better state where you have a
Hessian from a calculation of the second derivatives available (aoforce). The other
two options describe real procedures for initialization of the Hessian. Default values:
stretches (0.5), angles (0.2).

4.4.4 Definition of External Electrostatic Fields

This submenu allows you to calculate first and second numerical derivatives of the
energy with respect to an external electric field. The first three options should be
clear; 1st and 2nd are logical switches which are turned on and off the usual way
(1st or -1st) and delta is the increment for the numerical differentiation, that is,
the finite value of the external field, which replaces the (ideally) differential field:

--------------------------------------------------------------------
electrostatic field definition menu
--------------------------------------------------------------------
option | status | description
--------------------------------------------------------------------
1st | F | numerical 1st derivative dE/dField
2nd | F | numerical 2nd derivative d2E/dField2
delta <real> | | increment for numerical differentiation
| | DEFAULT = .5000E-02
geofield | F | geometry optimization with external field
man | F | explicit definition of electrostatic field(s)
--------------------------------------------------------------------

geofield gives the possibility to perform a whole geometry optimization under the
influence of a finite external field and thus to obtain the (distorted) minimum geom-
etry in this field. To do this, an external electrostatic field must be defined explicitly
which can be done using command man. Note that geofield must also be switched
on if any properties are to be evaluated in the presence of an electric field. The most
prominent example is the calculation of hyperpolarizabilies.
4.4. THE GENERAL OPTIONS MENU 111

Take Care, due to some inconsistencies in define it is always necessary to switch on

the field calculations manually. Therefore edit the control file after having finished
your define session and enter on after the entries of fields and geofield.

4.4.5 Properties

Most properties can be calculated with the -proper option in the corresponding
modules like dscf or ridft directly. Previously, the program moloch was used for
this purpose and define is still able to use the old menu for moloch (see below).
For a detailed list of the keywords, please see Sec. 22.2.26. If you enter prop in
the general menu, define first will check whether the data group $properties does
already exist in your control file or in a file referenced therein. If this is not the
case you will be asked to specify the file on which $properties shall be written:

CURRENT STATUS OF PROPERTY KEYWORDS:

NO KEYWORDS FOR CALCULATION OF PROPERTIES ARE SET

SELECT ONE OF THE FOLLOWING OPTIONS:

mvd : Calculation of p^4 and Darwin term

pop : population analyses (-> submenu)
loc : calculation of localized orbitals
esp : ESP-fit
plt : Calculation of densities, MOs, etc. on a grid; e.g.
for further use in visualization programs (keqword: $pointval)
old : old menu for moloch
* : leave this submenu

All keywords can be disabled by using the same option again and then selecting to
deactivate the input (*).

Option mvd

The option mvd directly activates the calculation of the relativistic mass-velocity and
the Darwin term based on perturbation theory.

Option esp

The option esp directly activates the calculation of the electrostatic potential (ESP)
fit.
112 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

Option pop

Activating pop drives the calculation of various population analyses.

YOU MAY CHOOSE BETWEEN:

mul: Mulliken PA
low: Loewdin PA
nbo: natural PA
pab: PA basen on occupation numbers
wbi: Wiberg bond indices
all: all of the above PAs
* : continue without activating any PA

After selecting one of the analyses, all atoms and bonds or a subset can be considered.
It is recommended to run the PA for the full molecule as the evaluation is not time-
determing. Note that Wiberg bond indices are always calculated for all bonds.

Option loc

Activating loc allows to setup the input for constructing localized molecular orbitals
according to the Foster–Boys scheme.

CHOOSE ORBITALS FOR LOCALIZATION (default=all)

m <list> : MOs from list are included

ge <thr> : above list is generated automatically.
It will contain MOs with energies
e(HOMO)-thr < e <= e(HOMO)
gn <int> : above list is generated automatically.
It will contain the <int> highest
occupied MOs
lm <no> : list highest <no> occupied MOs
(default: no=200)
* : activate localization and continue

NOTE: for further options see manual

For other localization schemes such as the Pipek-Mezey method or intrinsic bond
orbitals (IBOs) and the required keywords, please see Sec. 22.2.26.

Option plt

Using the option plt inserts the necessary keywords to calculate various quantities
such as MOs, LMOs or the electron localization function on a grid. The default
format is plt, which can be opened by TmoleX or gOpenMol. Other formats can
be selected in define. A post-processing script to convert plt to other frequently
used formats in quantum chemistry is available at the homepage of the TURBOMOLE
project, please see https://www.turbomole.org/turbomole/utilities/.
4.4. THE GENERAL OPTIONS MENU 113

YOU MAY CHOOSE BETWEEN:

m <list> : plot for MOs from <list>

ge <thr> : above list is generated automatically.
It will contain MOs with energies
e(HOMO)-thr < e <= e(HOMO)
gn <int> : above list is generated automatically.
It will contain the <int> highest
occupied MOs
lm <no>,<nv> : list highest <no> occupied and lowest <nv>
virtual MOs (default: no=20, nv=1)
e : plot electron localization function (ELF)
f <fmt> : change format; default is plt; further options:
map, xyz, plv (see also manual)
d : plot densities (default)
* : activate $pointval and continue

NOTE: for further options see manual

Option old

The program moloch was previously used for the calculation of properties and anal-
ysis of the wave function. The subsequent description for an older version may not
work in all cases—sorry for that.
If you enter prop in the general menu and use the option old, define first will check
whether the data group $properties does already exist in your control file or in a
file referenced therein. If this is not the case you will be asked to specify the file on
which $properties shall be written:

data group $properties has not yet been specified

FOR INITIALIZING <moloch> KEYWORDS ENTER
[return] : WRITE TO CONTROL FILE control (DEFAULT), OR
filename : WRITE TO ANOTHER FILE

Afterwards you will get the following submenu which allows you to control all possible
actions of program moloch:
114 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

switch on one or more of the following options <i>

<i> = 1,..., 9
for switching off option <i>, specify -<i>
( 1) trace off
( 2) moments off
( 3) potential off
( 4) cowan-griffin off
( 5) localization off
( 6) population analyses off
( 7) plot off
( 8) firstorder off
selecting an already active option indicates that
suboptions shall be modified
* or q(uit) = quit | for help, type help <integer>

All options in this menu are selected by entering their number as indicated in the
first column. For example, to switch on option trace enter 1. The flag off will
then change to active. To switch off an option enter its negative number, e.g. -1
for trace. Most of the options require additional input and will therefore lead you
to further submenus. These are briefly described below.
Option trace trace will calculate the trace of density times overlap matrix:
N = tr{DS}
If the orbitals are orthonormal, N should yield the total number of electrons in your
molecule. If this is not true, your MO-vector will most probably be erroneous. For
example, the vector might belong to another geometry or basis set. As this is a very
sensitive test for errors like these and the calculation requires almost no time, you
should always switch on this option.
Option moments This option leads you to the following submenu:

add/change options for data group $moments

option | status | description
------------------|--------|-------------------------------
point <x> <y> <z> | T | reference point = (x,y,z)
atom <i> | F | reference point = atom no. <i>
0th | T | compute 0th moment
1st | F | compute 1st moment
2nd | F | compute 2nd moment
3rd | F | compute 3rd moment
------------------|--------|-------------------------------
-<moment> : skip computation of <moment>
* or q(uit) : terminate input

This menu serves to specify the electrostatic moments to be calculated (0th=charge,

1st=dipole moment, 2nd=quadrupole moment, 3rd=octuple moment). The refer-
ence point is the origin of the coordinate system used in the calculation. The value
of any calculated moment will be independent of this reference point, if all lower mo-
ments are zero. The default for the reference point is the origin, i.e. the coordinate
system used for the calculation of the moments will be the same as the one in which
4.4. THE GENERAL OPTIONS MENU 115

the atomic coordinates are specified. The reference point may be changed by typing
point with the three new coordinates appended. Alternatively you may choose the
coordinates of one of the atoms as reference point by entering atom and the atom
index.
Option potential This option collects all possible quantities related to the electro-
static field created by the molecular charge distribution. This includes the following
suboptions:

list of suboptions :
pot - electrostatic potential
fld - electrostatic field
fldgrd - electrostatic field gradient
shld - diamagnetic shielding
file - file reference
* - quit

The meaning of the four suboptions pot, fld, fldgrd and shld will probably present
no problems to you. For each of them, however, you will have to specify at which
point(s) this property should be calculated. This is accomplished by one or more
data groups $points in file control. After you chose one or more of the above op-
tions, you will therefore reach the next submenu which deals with the specification
of these data groups:

there are 1 data groups $points

manipulate data group(s) $points
a - add another data group
m <integer> - modify <integer>th data group
m all - modify all data groups
d <integer> - delete <integer>th data group
d all - delete all data groups
off <integer> - switch off <integer>th data group
off all - switch off all data groups
on <integer> - switch on <integer>th data group
on all - switch on all data groups
s - scan through data groups
* - quit

The first line informs you how many of these data groups already exist in your
control file. Each of these data groups may consist of several points at which
the properties will be calculated. You may now create new data groups, delete old
ones or simply switch on or off individual data groups (without deleting them from
control). The number of different data groups $points as well as the number of
points in each of them are not limited. However, if you use many points, you should
consider specifying them in a separate file. This is most easily done using option
file in the potential menu. This option will create a file for your data groups
$points and will write a reference of this file to file control.
Option cowan-griffin This option activates the computation of the first order
relativistic correction to the energy as given by the expectation value of the Cowan–
116 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

Griffin operator.
Option localization Specifying option localization will switch on a Boys local-
ization of molecular orbitals. define by default chooses a set of MOs to be localized
according to a certain threshold for the orbital energy. Information about these are
displayed like this:

BOYS localization will be performed with respect to x y z

number of sweeps = 10000
subset of molecular orbitals to be localized :
---> all occupied molecular orbitals
with orbital energy above -2.00000 Hartree
----------------------------------------------------------------
shells to be localized
----------------------------------------------------------------
a1 4-5 # 1- 5
e 2 # 1- 2
----------------------------------------------------------------
you are employing default options for localization
do you want to modify them ? DEFAULT(n)

If you want to change the MO selection or other options for the localization enter y
at this point (By default or when typing n you will reach the moloch options menu
again). You will then be asked whether to change the MO selection method. If you
want this, you will enter a little submenu where you can choose one of three possible
selection procedures:

all selects all occupied orbitals

thr selects all occupied orbitals with orbital energy larger than a certain
threshold

man enables you to select the MOs manually later in this section

If the selection method thr is specified you then will be asked for the threshold to be
applied for the selection. Afterwards you have the possibility to change some other
topics concerning the localization:

• specify other localization directions

• switch on utilization of localized orbitals for population analysis and/or prepa-

ration of plot data within the same moloch run

• set the maximum number of sweeps in the localization procedure

• specify a file where localized orbitals shall be written to

Option population analyses When activating this option you first have to specify
whether the population analysis (PA) should be performed in the CAO (default) or
AO basis. Afterwards define will ask you whether you want to perform a Mulliken
4.4. THE GENERAL OPTIONS MENU 117

population analysis. In this case, the following submenu will be displayed:

add or delete one or more special options for a

mulliken population analysis
option | status | description
-------|--------|---------------------------------------
spdf | F | compute MO contributions to atomic
| | brutto populations
molap | F | compute MO contributions to atomic
| | overlap populations
netto | F | compute atomic netto populations
irpspd | F | compute IRREP contributions to atomic
| | brutto populations
irpmol | F | compute IRREP contributions to atomic
| | overlap populations
mommul | F | print electrostatic moments resulting
| | from atomic charges
-------|--------|---------------------------------------
-<option> : switch off <option>
* or q(uit) : leave this menu

Here you can activate several optional quantities to be computed along with the Mul-
liken PA. To switch on one or more of these options you must enter the corresponding
option keywords, e.g. spdf netto for computation of atomic net populations and MO
contributions to atomic gross populations. The status flags for these tasks will then
change from F (false) to T (true). To switch off any option you simply have to enter
the corresponding keyword preceded by a ‘-’, e.g. -netto for disabling calculation
of atomic net populations.
After having left the Mulliken PA section you will be asked whether a population
analysis based on occupation numbers (a modified Roby–Davidson PA) should be
performed by moloch. When typing y you will see the following submenu, where you
can switch on several special options for the PA in the same manner as described
above.
add or delete one or more special options for a
population analysis based on occupation numbers
option | status | description
--------|--------|----------------------------------------
momao | F | compute MO contributions to modified
| | atomic orbital (MAO) occupation numbers
maodump | F | dump all MAOs onto standard output
maofile | F | write MAOs onto a separate file
select | F | write only those MAOs which have been
| | employed in the population analysis
all | F | write all MAOs
--------|--------|----------------------------------------
note that the options select and all are complementary
-<option> : switch off <option>
* or q(uit) : leave this menu
118 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

Afterwards you have the possibility to change the criterion to be applied for the
selection of modified atomic orbitals (MAOs) within the following little submenu:

global criterion for selection of Modified Atomic Orbitals (MAOs) :

-------------------------------------------------------------------
MAOs are employed if ’atomic’ density eigenvalues
exceed a threshold of .1000
-------------------------------------------------------------------
specify the appropriate option if you want to use another
global criterion for selecting MAOs
option | status | description
--------|--------|---------------------------------------
eig <r> | T | select by eigenvalues of the
| | ’atomic’ density matrices
occ <r> | F | select by occupation numbers
--------|--------|---------------------------------------
<r> is the selection threshold (DEFAULT= .1000 )
* or q(uit) : leave this menu

The criterion applied by default is the so-called atomic density eigenvalue with a
threshold of 0.1. You can switch the criterion to occupation numbers by entering
occ. If you also want to change the threshold, you just have to append its new
value to the selection keyword, e.g. occ .2. Finally you can select or disable various
options in connection with the computation of shared electron numbers (SEN) within
the following menu:

actual settings for data group $shared electron numbers

2-center shared electron numbers will be computed;
values are printed if absolute value exceeds .0100
3-center shared electron numbers will be computed;
values are printed if absolute value exceeds .0100
4-center shared electron numbers will be computed;
values are printed if absolute value exceeds .0100
add or delete one or more options for the
computation of Shared Electron Numbers (SEN)
option | status | description
--------|--------|----------------------------------------
2c <r> | T | compute 2-center SEN and print if
| | |SEN| > <r> (DEFAULT = .1000E-01)
3c <r> | T | compute 3-center SEN and print if
| | |SEN| > <r> (DEFAULT = .1000E-01)
4c <r> | T | compute 4-center SEN and print if
| | |SEN| > <r> (DEFAULT = .1000E-01)
--------|--------|----------------------------------------
nosym | F | switch off use of symmetry
orbs | F | compute orbital contributions to SEN
irreps | F | compute irrep contributions to SEN
--------|--------|----------------------------------------
-<option> : switch off <option>
* or q(uit) : leave this menu
4.4. THE GENERAL OPTIONS MENU 119

The procedure for changing the options is the same as described above. By default
calculation of 2-, 3- and 4-center SENs will be enabled with thresholds of 0.01 each.
Option plot This option allows you to prepare the data needed for contour plots
of orbital amplitudes or total electron densities. We do not recommend to prepare
plotting data this way; an easier method—with an easier syntax—is to generate
these data directly by the programs, where densities (also MP2 or excited ones) and
Molecular orbitals are calculated. This is described in Chapter 19. If you nevertheless
want to prepare the input for plotting data as needed by moloch using define, on
activating plot you get the following menu:

there are 1 data groups $grid

manipulate data group(s) $grid
a - add another data group
m <integer> - modify <integer>th data group
m all - modify all data groups
d <integer> - delete <integer>th data group
d all - delete all data groups
off <integer> - switch off <integer>th data group
off all - switch off all data groups
on <integer> - switch on <integer>th data group
on all - switch on all data groups
s - scan through data groups
* - quit

The commands in this menu serve for the manipulation of data groups $grid in an
analogous way as described for $points in the potential section above. $grid data
groups contain the input information necessary to create the plot data by moloch
(one data group for each plot). If you want to add a new data group you will enter
this submenu:

specify the input orbital / input density :

mo <label> - use occupied molecular orbital <label>
mo density - use one electron density built from the
occupied molecular orbitals
lmo <i> - use localized molecular orbital no. <lmo>
mao <i> <k> - use modified atomic orbital no. <i>
centered on atom no. <k>
help - explanation of the syntax for <label>
* - quit

Here you may specify the orbital to be plotted. To plot the amplitude of the fifth
orbital in irrep a1, e.g., you would enter mo 5a1. Equivalently you can use localized
orbitals from a Boys localization procedure or modified atomic orbitals as obtained in
a Roby–Davidson–Ahlrichs–Heinzmann population analysis. In the latter cases you
will not have to enter an irrep label, as these orbitals are necessarily in C1 symmetry.
Instead you will have to enter the index of the orbital to be plotted (and for option
mao the index of the atom at which it is situated). In all cases you will additionally
have to specify the plane in which the amplitudes or densities will be monitored.
To do this, you have to declare two vectors which span that plane and the origin
120 CHAPTER 4. PREPARING YOUR INPUT FILE WITH DEFINE

of this new coordinate system relative to the one in which the atomic coordinates
are given. Furthermore, you will have to create a grid of points on this plane. The
orbital amplitude or electron density will then be calculated for every point in this
grid. The grid is created by telling define the range to be included along both
vectors spanning the plane (where the unit in each direction is the length of the
corresponding basis vector) and the number of points to be calculated in this range.
It is advantageous to use a wide grid while you test the ranges or planes which give
the best results and then to switch to a finer grid for the final calculation. Finally
input (MO vector) and output (plot data) files can be specified.
In case you do not want to add a new data group as described above but to change
an existing one, you will be asked which one of the specifications you want to modify.
Chapter 5

Calculation of Molecular Structure

and Ab Initio Molecular
Dynamics

5.1 Structure Optimizations using the Jobex Script

In its normal mode of operation, the shell script jobex controls and executes au-
tomatic optimizations of molecular geometry parameters. It will cycle through the
direct SCF, gradient and force relaxation programs and stop if either the maximum
number of cycles is reached or the convergence criteria (change in the total energy,
maximum norm of the gradient) are fulfilled. By default, the executable programs
are taken from the load modules library within the TURBOMOLE directory.

5.1.1 Options

Given a shell the usage is:

nohup jobex &

This command invokes structure optimization using the default program statpt.
Structure optimizations using program relax can be performed using -relax flag:

nohup jobex -relax &

nohup means that the command is immune to hangups, logouts, and quits. & runs a
background command. jobex accepts the following arguments controlling the level of
calculation, convergence criteria and many more (for example nohup jobex -gcart 4 &):

-energy integer converge total energy up to

10(−<integer>) Hartree (default: 6)

121
122 CHAPTER 5. STRUCTURE OPTIMIZATIONS

-gcart integer converge maximum norm of Cartesian gradient up to

10(−<integer>) atomic units (default: 3)
-c integer perform up to integer cycles (default: 100)
-dscf begin with a direct SCF step
-grad begin with a gradient step
-statpt begin with a force relaxation step
-relax use the relax program for force relaxation
-trans perform transition state search
-level level define the optimization level, level =scf, mp2, cc2, uff, rirpa
or xtb (default is scf).
-ri use RI modules ridft and rdgrad (fast Coulomb approxi-
mation) instead of dscf and grad as well as rimp2 instead
of mpgrad; obligatory option if -level rirpa
-rijk in connection with ’-level cc2’, the RI-JK versions of HF and
CPHF are switched on
-ex perform excited state geometry optimization using egrad
-l <path> employ programs from directory <path>
-ls <path> load scripts from directory <path>
-md a molecular dynamics (MD) run (using frog instead of relax)
-mdfile file commands for MD run are contained in this file (default:
mdmaster).
-mdscript file option to execute a shell script before the frog step
-keep keep program output from all optimization steps
-help shows a short description of the commands above

5.1.2 Output

There will be an output written to file job.start which informs you about the
current options. The convergence is signalled by the file converged; otherwise, you
should find the file not.converged within your working directory. If jobex finds a
file named stop or STOP in the working directory, jobex will stop after the present
step has terminated. You can create stop by the command touch stop.
The output of the last complete cycle is written to file job.last, while the output
of the running cycle is collected within the file job.<cycle>, where <cycle> is the
index of the cycle. The convergence criteria and their current values are written out
at the bottom of the job.last file.
5.2. PROGRAM STATPT 123

5.2 Program Statpt

5.2.1 General Information

Stationary points are places on the potential energy surface (PES) with a zero gradi-
ent, i.e. zero first derivatives of the energy with respect to atomic coordinates. Two
types of stationary points are of special importance to chemists. These are minima
(reactants, products, intermediates) and first-order saddle points (transition states).
The two types of stationary points can be characterized by the curvature of the PES
at these points. At a minimum the Hessian matrix (second derivatives of energy with
respect to atomic coordinates) is positive definite, that is the curvature is positive in
all directions. If there is one, and only one, negative curvature, the stationary point
is a transition state (TS). Because vibrational frequencies are basically the square
roots of the curvatures, a minimum has all real frequencies, and a saddle point has
one imaginary vibrational “frequency”.
Structure optimizations are most effectively done by so-called quasi-Newton–Raph-
son methods. They require the exact gradient vector and an approximation to the
Hessian matrix. The rate of convergence of the structure optimization depends on
anharmonicity of the PES and of the quality of the approximation to the Hessian
matrix.
The optimization procedure implemented in statpt belongs to the family of quasi-
Newton–Raphsod methods [52]. It is based on the restricted second-order method,
which employs Hessian shift parameter in order to control the step length and direc-
tion. This shift parameter is determined by the requirement that the step size should
be equal to the actual value of the trust radius, tradius, and ensures that the shifted
Hessian has the correct eigenvalue structure, all positive for a minimum search, and
one negative eigenvalue for a TS search. For TS optimization there is another way of
describing the same algorithm, namely as a minimization on the "image" potential.
The latter is known as TRIM (Trust Radius Image Minimization) [53].
For TS optimizations the TRIM method implemented in statpt tries to maximize
the energy along one of the Hessian eigenvectors, while minimizing it in all other
directions. Thus, one “follows” one particular eigenvector, hereafter called the “tran-
sition” vector. After computing the Hessian for your guess structure you have to
identify which vector to follow. For a good TS guess this is the eigenvector with
negative eigenvalue, or imaginary frequency. A good comparison of different TS
optimization methods is given in [54].
Structure optimizations using statpt are controlled by the keyword $statpt to be
present in the control file. It can be set either manually or by using the stp
menu of define. The type of stationary point optimization depends on the value of
itrvec specified as an option within $statpt. By default itrvec is set to 0, which
implies a structure minimization. A value itrvec > 0 implies a transition state
optimization using the eigenvalue-following TRIM algorithm, where the index of the
transition vector is specified by itrvec. Note, that statpt orders eigenvalues (and
eigenvectors) of the Hessian in ascending order, shifting six (or five in the case of
124 CHAPTER 5. STRUCTURE OPTIMIZATIONS

linear molecules) zero translation and rotation eigenvalues to the end.

Note: this order differs from that used for vibrational frequencies in the control
file, where rotational and translational eigenvalues are not shifted.
By default a structure optimization is converged when all of the following criteria
are met:

• the energy change between two optimization cycles drops below the value given
by threchange (default: 10−6 a.u.),
• the maximum displacement element drops below the value given by thrmax\-displ
(default: 10−3 a.u.),
• the maximum gradient element drops below the value given by thrmaxgrad
(default: 10−3 a.u.),
• the root mean square of the displacement elements drops below the value given
by thrrmsdispl (default: 5 · 10−4 a.u.),
• the root mean square of the gradient elements drops below the value given by
thrrmsgrad (defaul:t 5 · 10−4 a.u.).

The default values for the convergence criteria can be changed using the stp menu
of define. The necessary keywords are described in Section 22.2.24 below.
For structure optimization of minima with statpt as relaxation program just use:

jobex &

TS optimizations are performed by the jobex invocation:

jobex -trans &

5.2.2 Hessian matrix

The choice of the initial Hessian matrix has a great effect on the convergence of the
structure optimization. At present, there are three choices for the Hessian matrix in
statpt. For minimization, a diagonal matrix or approximate Hessian matrix from
a force field calculation using uff(see Section 5.4) can be used. For transition state
optimizations you have to provide either the “exact” Hessian or results from the low-
est eigenvalue search (LES, see Section 15). Note also that you can calculate the
Hessian with a smaller basis set and/or at a lower wavefunction level, and
use it for higher level structure optimization. Usually, a Hessian matrix calcu-
lated in a minimal basis using RI-DFT is good enough for all methods implemented
in TURBOMOLE.
statpt automatically takes the best choice of the Hessian from the control file.
For minimizations it first looks for the exact Hessian and then for the UFF Hes-
sian. If none of them is found it takes the scaled unit matrix. For transition state
optimization the exact Hessian has a higher priority than the results of LES.
5.2. PROGRAM STATPT 125

The results of LES can be used to obtain an initial Hessian matrix for transition
state optimizations involving large molecules, where calculation of the full Hessian
is too expensive. Note, that LES calculations for statpt, in addition to the $les
keyword require the following keywords to be added manually in the control file:

$h0hessian
$nomw

The default Hessian update for minimization is bfgs, which is likely to remain pos-
itive definite. The powell update is the default for transition state optimizations,
since the Hessian can develop a negative curvature as the search progresses.

5.2.3 Finding Minima

Simply specify the $statpt keyword in the control file and run jobex as explained
above. You can very often speedup the optimization by calculating the initial Hessian
matrix using uff.

5.2.4 Finding transition states

Locating minima on a PES is straightforward. In contrast, transition state optimiza-

tion requires much more input. The diagonal guess Hessian will almost never work,
so you must provide a computed one. The Hessian should be computed at your best
guess as to what the TS should be.
The real trick here is to find a good guess for the transition state structure. The closer
you are, the better. It is often difficult to guess these structures. One way to obtain a
good guess is to built an approximate TS and to perform a constrained minimization
by freezing internal coordinates that change most during the reaction. Alternatively,
you can generate several structures intermediate to reactants and products, and
compute the energy at each point. The maximum energy structure is usually a good
guess for the true TS.
After obtaining a reasonable initial guess for the TS structure you have to perform
a vibrational analysis (or LES calculation for a large molecule) and to identify the
index of the transition vector to follow during the optimization. Ideally, this is a
vector with a negative eigenvalue, or "imaginary" frequency. The best way to find
the right vector is to use some graphical interface to visualize vibrations. For a
reasonable guess structure there should be one vibration that resembles the reaction
under study. Remember that statpt uses a different ordering of eigenvalues as
compared to the aoforce output—six (five) zero eigenvalues are shifted to the end.
There is an important thing to remember at this point. Even such sophisticated
optimization methods like TRIM will not replace your own chemical intuition about
where transition states may be located. If you need to restart your run, do so
with the coordinates which have the smallest RMS gradient. Note that the energy
does not have necessarily to decrease in a transition state search (as opposed to
126 CHAPTER 5. STRUCTURE OPTIMIZATIONS

minimizations). It is sometimes necessary to do restart several times (including a

recomputation of the Hessian) before the saddle point can be located.
Assuming you do find the TS, it is always a good idea to recompute the Hessian at
this structure. It is fairly common, especially when using symmetry, that at your
“TS” there is a second imaginary frequency. This means that you have not found
the correct TS. The proper procedure is to distort the structure along the “extra”
imaginary normal mode using the tool screwer (see Section 1.5). Very often such a
distortion requires also lowering the point group symmetry. The distortion must be
large enough, otherwise the next run will come back to the invalid structure.

5.3 Program Relax

5.3.1 Purpose

relax drives and controls a non-linear optimization procedure to locate the minimum
(or a stationary point) of a function f (x). In TURBOMOLE f is always the electronic
energy, and the coordinates x will be referred to as general coordinates. They include

• cartesian atomic coordinates

• internal atomic coordinates

• exponents, contraction coefficients and scaling factors of basis functions

• a global scaling factor (a common scaling factor for all basis set exponents)

The optimization employs an iterative procedure based on gradients ∇f of the cur-

rent and, if available, previous iterations. Various procedures can be applied: steep-
est descent, Pulay’s DIIS, quasi–Newton, conjugate gradients, as well as combina-
tions of them. relax carries out:

• update of general coordinates

• update of approximate Hessians if needed

• conversion of coordinates (internal ←→ Cartesian)

The mode of operation is chosen by the keywords $optimize and $interconversion

and the corresponding options, which will be described in the following sections.

5.3.2 Optimization of General Coordinates

After gradients Gk have been calculated for coordinates q k in optimization cycle k,

new coordinates (or basis set exponents) q k+1 can be obtained from the quasi–Newton
update:
q k+1 = q k − F k Gk
5.3. PROGRAM RELAX 127

where F k is the inverse of an approximate force constant matrix H k . This method

would immediately converge to the equilibrium geometry if F k would be the inverse
of the exact force constant matrix and the force field would be quadratic. In real
applications usually none of these requirements is fulfilled. Often only a crude ap-
proximation to the force constant matrix H k is known. Sometimes a unit matrix
is employed (which means coordinate update along the negative gradient with all
coordinates treated on an equal footing).
The optimization of nuclear coordinates in the space of internal coordinates is the
default task performed by relax and does not need to be enabled. Any other opti-
mization task requires explicit specifications in data group $optimize, which takes
several possible options:
$optimize options

internal on/off Structure optimization in internal coordinates.

redundant on/off Structure optimization in redundant coordinates.

cartesian on/off Structure optimization in Cartesian coordinates.

basis on/off Optimization of basis set exponents, contraction coefficients,

scaling factors.

global on/off Optimization of global scaling factor for all basis set expo-
nents.

Note: All options except internal are switched off by default, unless they have
been activated explicitly by specifying on.
Some of the options may be used simultaneously, e.g.

• internal, basis

• internal, global

• cartesian, basis

Other options have to be used exclusively, e.g.

• internal, cartesian

• basis, global

The update of the coordinates may be controlled by special options provided in data
group $coordinateupdate which takes as options:

dqmax=real Maximum total coordinate change (default: 0.3).

interpolate on/off Calculate coordinate update by inter/extrapolation us-

ing coordinates and gradients of the last two optimiza-
tion cycles (default: interpolate on) if possible.
128 CHAPTER 5. STRUCTURE OPTIMIZATIONS

statistics integer /off Display optimization statistics for the integer previous
optimization cycles. Without integer all available in-
formation will be displayed. off suppresses optimiza-
tion statistics.

The following data blocks are used by program relax:

1. Input data from gradient programs grad, rdgrad, egrad, rimp2, mpgrad, etc.:

$grad Cartesian atomic coordinates and their gradients.

$egrad exponents and scale factors and their gradients.
$globgrad global scale factor and its gradient.

2. Input data from force constant program aoforce:

$grad Cartesian atomic coordinates and their gradients.

$globgrad global scale factor and its gradient.
$hessian the force constant matrix in the space of Cartesian coordinates.

3. Output data from program relax:

$coord cartesian atomic coordinates.

$basis exponents and scale factors.
$global global scale factor.

For structure optimizations the use of (redundant) internal coordinates is recom-

mended, see Section 4.0.6. Normally internal coordinates are not used for input or
output by the electronic structure programs (dscf, mpgrad, etc.). Instead the coor-
dinates, gradients, etc. are automatically converted to internal coordinates by relax
on input and the updated positions of the nuclei are written in Cartesian coordinates
to the data group $coord. Details are explained in the following sections.

5.3.3 Force Constant Update Algorithms

In a Newton-type geometry update procedure often only a crude approximation to

the force constant matrix H k is available. What can be done then is to update
F k = (H k )−1 in each iteration using information about previous coordinates and
gradients. This constitutes the quasi–Newton or variable metric methods of which
there are a few variants:

1. Murtagh/Sargent (MS):

Z k−1 (Z k−1 )†
F k = F k−1 +
(Z k−1 )† dGk−1
5.3. PROGRAM RELAX 129

2. Broyden/Fletcher/Goldfarb/Shanno (BFGS):
S(dq k−1 )† dq k−1 − dq k−1 (dGk−1 )† F k−1 − F k−1 dGk−1 (dq k−1 )†
F k = F k−1 +
S1
3. Davidon/Fletcher/Powell (DFP):
(dq k−1 )† dq k−1 F k−1 dGk−1 (dGk−1 )† F k−1
F k = F k−1 + −
S1 (S − 1)S1

4. combined method (BFGS/DFP): If S1 < (S − 1)S1 and S1 > 0 perform DFP

update, otherwise BFGS.

The meaning of the symbols above is as follows:

F k = (H k )−1 approximate inverse force constant matrix in the k-th iteration.s

qk general coordinates in the k-th iteration.
Gk gradients in the k-th iteration.
dq k−1 = q k − q k−1
dg k−1 = g k − g k−1
Z k−1 = dq k−1 − F k−1 dGk−1
S1 = (dq k−1 )† dg k−1
S = 1 + ((dg k−1 )† F k−1 dGk−1 )/(S1)

An alternative is to use update algorithms for the hessian H k itself:

Ehrig, Ahlrichs : Diagonal update for the hessian by means of a least squares fit
q
Hii = Hiik−1 (hi + di )
k

with the new estimate h for the diagonal elements obtained by

dGki dqik
P
hi = Pk k 2
k (dqi )
and the error d obtained by the regression
rP
(dgik )2
Pk k 2 − h2i
k (dqi )
di = .
k−2
Another alternative is to use DIIS-like methods: structure optimization by direct
inversion in the iterative subspace. (See ref. [55] for the description of the algorithm).
The DIIS procedure can often be applied with good success, using static or updated
force constant matrices.
Any of the algorithms mentioned above may be chosen. Recommended is the macro
option ahlrichs, which leads to the following actions (n is the maximum number of
structures to be included for the update, default is n = 4):
130 CHAPTER 5. STRUCTURE OPTIMIZATIONS

ncycles < n: geometry update by inter/extrapolation using the last 2 geometries.

ncycles ≥ n: diagonal update for the hessian as described above; DIIS–like update
for the geometry.

||G|| < thr: BFGS-type update of the hessian and quasi–Newton update of (gener-
alized) coordinates.

References for the algorithms mentioned above: [52, 55–59]

5.3.4 Definition of Internal Coordinates

If structure optimizations are to be performed in the space of internal coordinates

($optimize internal, is the default setting), appropriate internal coordinate defi-
nitions have to be provided on data block $intdef. The types available and their
definitions are described in Section 4.1.2. For recommendations about the choice
of internal coordinates consult ref. [46]. Nevertheless the structure of $intdef will
shortly be described. The syntax is (in free format):

1 k 1.00000000 bend 1 2 3 val=1.9500 fdiag=.6666

The first items have been explained in Chapter 4.

Two additional items val=real, fdiag=real may be supplied for special purposes:

val= serves for the input of values for internal coordinates for the intercon-
version internal → cartesian coordinates; it will be read in by relax if the
flag for interconversion of coordinates has been activated ($interconversion on
), or by the interactive input program define within the geometry spec-
ification menu.

fdiag= serves for the input of (diagonal) force constants for the individual in-
ternal coordinates to initialize $forceapprox.

5.3.5 Structure Optimizations Using Internal Coordinates

This is the default task of relax ($optimize internal on does not need to be
specified!) You need as input the data groups :

$grad Cartesian coordinates and gradients as provided and accumulated in

subsequent optimization cycles by the programs grad, or rdgrad etc.

$intdef definitions of internal coordinates.

$redundant definitions of redundant coordinates.

Output will be the updated coordinates on $coord and the updated force constant
matrix on $forceapprox. If any non-default force constant update option has been
5.3. PROGRAM RELAX 131

chosen, relax increments its counting variables <numgeo>, <numpul> within com-
mand keyword $forceupdate. If the approximate force constant has been initialized
($forceinit on ) relax switches the initialization flag to $forceinit off. Re-
fer also to the general documentation of TURBOMOLE. It is recommended to check
correctness of your definition of internal coordinates:

1. Calculate their values for your Cartesian start coordinates using the relax
program (see Section 5.3.11) or within a define session.

2. Have a look at the eigenvectors of the BmB† -matrix. Set some ‘?’ behind
keyword $intdef, if there are any eigenvalues close to zero (< 10−2 is to be
considered bad for small molecules, but there is no general rule) check those in-
ternal coordinates for consistency which contribute to the corresponding eigen-
vector(s)!

5.3.6 Structure Optimization in Cartesian Coordinates

For this task you have to specify:

$optimize
cartesian on
internal off

These lines switch on the non-default optimization in Cartesian coordinates and

switch off the optimization in internal coordinates (this has to be done explicitly!).
As input data groups you need only $grad as provided by on of the gradient programs.
For the first coordinate update an approximate force constant matrix is needed in
data group $forceapprox. Output will be the updated coordinates on $coord, and
the updated force constant matrix on $forceapprox.
The coordinates for any single atom can be fixed by placing an ’f’ in the third to
eighth column of the chemical symbol/flag group. As an example, the following
coordinates specify acetone with a fixed carbonyl group:

$coord
2.02693271108611 2.03672551266230 0.00000000000000 c
1.08247228252865 -0.68857387733323 0.00000000000000 c f
2.53154870318830 -2.48171472134488 0.00000000000000 o f
-1.78063790034738 -1.04586399389434 0.00000000000000 c
-2.64348282517094 -0.13141435997713 1.68855816889786 h
-2.23779643042546 -3.09026673535431 0.00000000000000 h
-2.64348282517094 -0.13141435997713 -1.68855816889786 h
1.31008893646566 3.07002878668872 1.68840815751978 h
1.31008893646566 3.07002878668872 -1.68840815751978 h
4.12184425921830 2.06288409251899 0.00000000000000 h
$end
132 CHAPTER 5. STRUCTURE OPTIMIZATIONS

5.3.7 Optimization of Basis Sets (SCF only)

For this task you have to specify:

$optimize
basis on
internal off

This example would perform only a basis set optimization without accompanying
geometry optimization. It is possible, of course, to optimize both simultaneously:
Just leave out the last line of the example (internal off). Input data groups are:

$egrad Basis set exponents, contraction coefficients, scaling factors and their re-
spective gradients as provided and accumulated in subsequent optimiza-
tion cycles by one of the programs grad or mpgrad, if $drvopt basis
on has been set.

$basis Description of basis sets used, see Section 4.2.

Output will be the updated basis on $basis, and the updated force constant matrix
on $forceapprox.
For an example, see Section 23.5.

5.3.8 Simultaneous Optimization of Basis Set and Structure

The optimization of geometry and basis set may be performed simultaneously and
requires the specification of:

$optimize
internal on (or: cartesian on)
basis on

and needs as input data groups $grad and $egrad. Output will be on $coord,
$basis, also on $forceapprox (updated).

5.3.9 Optimization of Structure and a Global Scaling Factor

Optimization of a global scaling factor is usually not performed in geometry opti-

mizations. It is a special feature for special applications by even more special users.
As reference see [60].
To optimize the structure and a global scaling factor specify:

$optimize
internal on (or: cartesian on)
global on
5.3. PROGRAM RELAX 133

You need as input data groups $grad and $globgrad, the latter contains the global
scaling factors and their gradients accumulated in all optimization cycles. Out-
put will be on $coord, $global, also on $forceapprox (updated). Note that for
optimization of a global scaling factor a larger initial force constant element is rec-
ommended (about 10.0).

5.3.10 Conversion from Internal to Cartesian Coordinates

Due to translational and rotational degrees of freedom and the non-linear dependence
of internal coordinates upon Cartesian coordinates, there is no unique set of Cartesian
coordinates for a given set of internal coordinates. Therefore an iterative procedure is
employed to calculate the next local solution for a given Cartesian start coordinates.
This task may be performed using the relax program, but it is much easier done
within a define session.

5.3.11 Conversion of Cartesian Coordinates, Gradients and Force

Constants to Internals

To perform this tasks, you have to activate the interconversion mode by

$interconversion on
cartesian --> internal coordinate gradient hessian

Note that any combination of the three options showed is allowed! The default value
is coordinate, the two other have to be switched on explicitly if desired.
You need as input data groups:

intdef Definitions of (redundant) internal coordinates

coord Cartesian coordinates (for option ‘coordinate’)

grad Cartesian coordinates and gradients as provided and accumulated in

subsequent optimization cycles by the various gradient programs (for
coordinate and gradient)

hessian Analytical force constant matrix (as provided by the force constant pro-
gram aoforce) (only if option hessian is specified). The data group
$hessian (projected) may be used alternatively for this purpose.

All output will be written to the screen except for option hessian (output to data
group $forceapprox)

5.3.12 The m-Matrix

The m-matrix serves to fix position and orientation of your molecule during geometry
optimizations. It cannot be used to fix internal coordinates! The m-matrix is a
134 CHAPTER 5. STRUCTURE OPTIMIZATIONS

diagonal matrix of dimension 3n2 (where n is the number of atoms). Normally m

will be initialized as a unit matrix by relax. As an example consider you want to
restrict an atom to the xy-plane. You then set the m(z)–matrix element for this atom
to zero. You can use at most six zero m-matrix diagonals (for linear molecules only
five)—corresponding to translational and rotational degrees of freedom. Note that
the condition of the BmB† -matrix can get worse if positional restrictions are applied
to the m-matrix. m-matrix elements violating the molecular point group symmetry
will be reset to one. Non-default settings for m-matrix diagonals of selected atoms
have to be specified within data group $m-matrix as:

$m-matrix
1 0.0 0.0 0.0
10 1.0 0.0 0.0
11 1.0 1.0 0.0

5.3.13 Initialization of Force Constant Matrices

The most simple initial hessian is a unit matrix. However, better choices are prefer-
able. For structure optimizations using internal coordinates you may use structural
information to set up a diagonal force constant matrix with elements chosen in ac-
cord to the softness or stiffness of the individual modes. For detailed information
refer to ref. [58]. For optimization of basis set parameters less information is avail-
able. When neither data block $forceapprox is available nor $forceinit on is set,
the force constant matrix will be initialized as a unit matrix. Specifying the force
constant initialization key $forceinit on diag=... will lead to:

diag=real Initialization with real as diagonal elements.

diag=default Initial force constant diagonals will be assigned the following

default values:
internal coordinates : stretches 0.50
angles 0.20
scaling factors : s,p 1.50
d 3.00
exponents : uncontracted 0.15
contracted 10.00
contraction coefficients : 100.00
global scaling factor : 15.00
cartesian force constants : 0.50
diag=individual Initial force constant diagonals will be taken from
$intdef fdiag=... or
$global fdiag=...
Similar initialization modes are NOT supported for geometry
optimization in Cartesian space and for the optimization of basis
set parameters!
5.4. FORCE FIELD CALCULATIONS 135

carthess Data group $hessian (projected) is used.

5.3.14 Look at Results

The energy file includes the total energy of all cycles of a structure optimization
completed so far. To get a display of energies and gradients use the UNIX command
grep cycle gradient which yields, e. g. H2 O.

cycle = 1 SCF energy = -76.3432480651 |dE/dxyz| = 0.124274

cycle = 2 SCF energy = -76.3575482860 |dE/dxyz| = 0.082663
cycle = 3 SCF energy = -76.3626983371 |dE/dxyz| = 0.033998
cycle = 4 SCF energy = -76.3633251080 |dE/dxyz| = 0.016404
cycle = 5 SCF energy = -76.3634291559 |dE/dxyz| = 0.010640
cycle = 6 SCF energy = -76.3634910117 |dE/dxyz| = 0.000730

This should be self-evident. To see the current—or, if the optimization is con-

verged, the final—atomic distances use the tool dist. Bond angles, torsional an-
gles etc. are obtained with the tools bend, tors, outp, etc. In the file gradient
are the collected Cartesian coordinates and corresponding gradients of all cycles.
The values of the general coordinates and corresponding gradients are an output of
relax written to job.<cycle> of job.last within jobex. To look at this search for
‘Optimization statistics’ in job.last or job.<cycle>.

5.4 Force Field Calculations

5.4.1 Purpose

uff preoptimizes a structure and calculates an analytical Hessian which can be used
as a start Hessian in a geometry optimization. This will accelerate the convergence
of an optimizations. For optimizations in Cartesian space this will be faster by a
factor of two for any molecule.

5.4.2 How to Perform a Uff Calculation

You have to generate Cartesian coordinates (file coord), nothing else. You can start
an single-point calculation calculation by typing

uff

To start an uff geometry optimization, one has to change the number of cycles
(parameter maxcycle) in the block $uff in the file control. The ouput is the
optimized structure (file coord), the analytical gradient (file uffgradient) and the
analytical cartesian hessian (file uffhessian0-0). Furthermore the control file will
be modified:
136 CHAPTER 5. STRUCTURE OPTIMIZATIONS

$forceinit on
carthess
$uffhessian file=uffhesian0-0

These commands have the effect to initialize the force constant matrix for a geometry
optimization with the hessian one.
In some cases uff cannot recognize the connectivity, then one can specify the connec-
tivity in the file ufftopology. The program will calculate the bond, angle, torsion,
inversion and non-bonded terms (force field terms) based on the connectivity speci-
fied in the topology file.

5.4.3 The Uff implementation

The uff implementation follows the paper by Rappé [10]. The energy expression in
uff is as follows:

NB
X 1
E UF F = · KIJ · (r − rIJ )2 (5.1)
2

KIJK

 4 (1 − cos(2θ)) : linear case
KIJK

NA (1 − cos(3θ)) : trigonal planar case


X  9
KIJK
+ 16 (1 − cos(4θ)) : quadratic planar case
 KIJK


 16 (1 − cos(4θ))
: octahedral case
· C0A + C1A cos θ + C2A cos(2θ)

 KIJK : general case
NT
X 1
+ · Vφ · (1 − cos (nφ0 ) cos(nφ))
2
NI
X
Vω · C0I + C1I cos ω + C2I cos 2ω


+
Nnb   
X xIJ 6  xIJ 12
+ DIJ · −2 +
x x
Nnb
X qI · qJ
+
·x
The Fourier coefficients C0A , C1A , C2A of the general angle terms are evaluated as a
function of the natural angle θ0 :
1
C2A = (5.2)
4 sin2 θ0
C1A = −4 · C2A cos θ0 (5.3)
C0A = C2A 2 cos2 θ0 + 1


(5.4)
The expressions in the energy term are:

NB , NA , NT , NI , Nnb the numbers of the bond-, angle-, torsion-, inversion- and the
non bonded-terms.
5.4. FORCE FIELD CALCULATIONS 137

KIJ , KIJK force constants of the bond- and angle-terms.

r, rIJ bond distance and natural bond distance of the two atoms
I and J.

θ, θ0 angle and natural angle for three atoms I − J − K.

C0A , C1A , C2A Fourier coefficients of the general angle terms.

φ, φ0 torsion angle and natural torsion angle of the atoms I − J −

K − L.

Vφ height of the torsion barrier.

n periodicity of the torsion potential.

ω inversion- or out-of-plane-angle at atom I.

Vω height of the inversion barrier.

C0I , C1I , C2I Fourier coefficients of the inversions terms.

x, xIJ distance and natural distance of two non bonded atoms I

and J.

DIJ depth of the Lennard–Jones potential.

qI ,  partial charge of atoms I and dielectric constant.

One major difference in this implementation concerns the atom types. The atom
types in Rappé’s paper have an underscore "_". In the present implementation
an sp3 C atom has the name "C 3" instead of "C_3". Particularly the bond
terms are described with the harmonic potential and the non-bonded van der Waals
terms with the Lennard–Jones potential. The partial charges needed for electrostatic
non-bonded terms are calculated with the Charge Equilibration Modell (QEq) from
Rappé [61]. There is no cutoff for the non-bonded terms.
The relaxation procedure distinguishes between molecules wih more than 90 atoms
and molecules with less atoms. For small molecules it consists of a Newton step
followed by a linesearch step. For big molecules a quasi-Newton relaxation is done.
The BFGS update of the force-constant matric is done [56, 62–64]. Pulay’s DIIS
procedure is implemented for big molecule to accelerate the optimization [55, 65].
The coordinates for any single atom can be fixed by placing an ’f’ in the third to
eighth column of the chemical symbol/flag group. As an example, the following
coordinates specify acetone with a fixed carbonyl group:

$coord
2.02693271108611 2.03672551266230 0.00000000000000 c
1.08247228252865 -0.68857387733323 0.00000000000000 c f
2.53154870318830 -2.48171472134488 0.00000000000000 o f
138 CHAPTER 5. STRUCTURE OPTIMIZATIONS

-1.78063790034738 -1.04586399389434 0.00000000000000 c

-2.64348282517094 -0.13141435997713 1.68855816889786 h
-2.23779643042546 -3.09026673535431 0.00000000000000 h
-2.64348282517094 -0.13141435997713 -1.68855816889786 h
1.31008893646566 3.07002878668872 1.68840815751978 h
1.31008893646566 3.07002878668872 -1.68840815751978 h
4.12184425921830 2.06288409251899 0.00000000000000 h
$end

5.5 Semiempirical Extended Tight-Binding Calculations

5.5.1 Purpose

tb optimizes a structure using GFN2-xtb as described in the papers of S. Grimme

et. al [66, 67].

5.5.2 How to Perform a xTB Calculation

There are two ways to run xTB calculations:

• Either prepare a usual TURBOMOLE input with arbitrary basis set and method.
Only the coordinates will be used, all other settings are ignored. If you want
to add a charge, please note that you have to use the $tb keyword.

• Or just use a simple Cartesian coordinate file (coord) in TURBOMOLE format.

No control file is needed. Start a geometry optimization by typing

jobex -level xtb

To add options the control file has to be modified:

$tb
charge -1
gfn <method>
accuracy 1.0
etemp 300
broydamp
maxiter <number>

The options are:

charge 0
optionally sets the molecular total charge. If not specified, a neutral
input is assumed.
5.6. MOLECULAR DYNAMICS CALCULATIONS 139

gfn 2
choose method, GFN(1)-xTB or GFN2-xTB, default is 2
accuracy 1.0
accuracy for SCC calculations, lower value means higher accuracy.
Default is 1.0
etemp 300
electronic temperature in Kelvin, default is 300K
broydamp 0.4
Broyden damping for charge mixing, default is 0.4
maxiter 250
maximum number of SCC iterations, default is 250

5.6 Molecular Dynamics Calculations

Ab initio molecular dynamics (MD) can be carried out on the ground and excited
state Born–Oppenheimer potential hypersurface. In addition non-adiabatic Tully-
type Surface Hopping MD can be performed using TDDFT. At the start of an MD
run the user must specify the initial atomic positions and velocities and give some
general instructions for the run. This is managed by running the interactive pro-
gram Mdprep and generating the command file mdmaster. If this is successful, the
MD run itself may be started: jobex -md. Time is then advanced in steps. The
electronic potential energy and its gradients are calculated quantum mechanically at
the required coordinates each time step (as detailed above, e.g. dscf and grad). The
MD program frog uses the Leapfrog Verlet algorithm [68] to turn the gradients into
new atomic positions and velocities. The atoms thus undergo classical Newtonian
dynamics on the ab initio potential hypersurface. Trajectory information is recorded
in a log file (mdlog). It is possible to instruct frog to heat or cool the system, use
a thermostat for canonical dynamics, conserve total energy or read in new positions
or velocities: the appropriate keywords are described in Section 22.2.27 below.

5.7 Global Structure Optimization – The DoDo Program

5.7.1 Genetic Algorithm

The reliable structure prediction of gas-phase clusters is complicated by a large num-

ber of possible structural isomers, often in combination with several low-lying elec-
tronic states. Brut force determination of the most stable configuration by manual
construction of all models and following local structure optimizations is a formidable
task. Instead, the DoDo program allows the automatic determination of the most
stable molecular structures using the genetic algorithm (GA) to locate the global
minimum of the total (electronic) energy. The GA is a search heuristic that employs
principal mechanisms of natural evolution such as inheritance, mutation, selection,
and crossover for exploration of the potential energy surface.
140 CHAPTER 5. STRUCTURE OPTIMIZATIONS

Generally, in GA a population of candidate structures is evolved toward improved

solutions. Within the DoDo program the GA is initialized with a pool of randomly
(automatically) generated structures, with initial structure models supplied by the
user – so called seeds – or a combination of both. Every seed is used one or more
times to create initial structures by randomly adding or removing atoms until the
predefined composition and system size is reached. All initial structures are subse-
quently geometry optimized to the nearest local minimum. Next, two evolutionary
operators – crossover and mutation – are used to exchange structure information
between the members of the current population. In the crossover, pairs of structures
are chosen to act as parents that will produce a new structure for the next genera-
tion. Within each parent pair, random pieces are exchanged to form the new mixed
structure, the child, which is subsequently optimized to the nearest local minimum
as well. The assumption is that the child will combine the good structural features
of the parents and thus will be more stable than either one. Evolutionary pressure
towards improved children is added to bias the search in the right direction. This is
achieved by selecting parents that have a relatively high fitness that is defined as a
function of their total energy:
 
i −Emin
exp −α · EEmax −Emin
fi = n    (5.5)
i −Emin
exp −α · EE
P
max −Emin
i=1

where fi is the fitness of the ith individual, n is the number of individuals, α is a con-
stant scaling factor, and Ei is the energy of that individual relative to the maximum
(Emax ) and minimum (Emin ) energies of structures in the whole population. Since
the child structures not always present a better fitness than their parents, elitism or
natural selection is applied. It is achieved by simply replacing parents with a worse
fitness by children with a better one in the structures pool. In order to maintain
a maximum diversity during the GA runs similarity recognition is used that allows
only distinct structures to be included in the pool. To prevent trapping of the pop-
ulation in local minima mutation is added in which random changes are introduced
to randomly chosen structures in the pool.
The determination of the global minimum structure typically requires, depending on
the system size, several hundreds to few thousands of local structure optimizations
of the children and mutated structures. Therefore, the DoDo program uses a simple
parallelization in which a predefined number (usually 10-100, depending on computer
resources) of local optimizations are performed simultaneously, each preferably run-
ning on a single or just a few CPU cores. When a number of local optimizations
finish, i.e. the number of running/queued geometry optimizations falls below a min-
imum value a, natural selection is applied and new child structures are generated by
crossover or mutation and submitted for local optimization until the maximum num-
ber b of simultaneously running/queued optimizations is reached. This pool-based
GA allows for an optimal utilization of computer resources since the algorithm is not
required to wait for the completion of local optimizations of all children structures
in a particular generation. The progress of the pool-based GA is then measured by
number of locally optimized structures rather than by the generation count as in
5.7. GLOBAL STRUCTURE OPTIMIZATION – THE DODO PROGRAM 141

case of the generation-based GA (a = 1).

More details of the implemented GA and application examples can be found in
Ref. [69] and references therein.

5.7.2 How to Perform

In order to start a global optimization procedure one directory is required containing

at least:

1. DoDo input file that specifies the parameters of the GA run (see also section
5.7.3).

2. Tmole input file that contains setup information for all local optimizations
(see also Tmole documentation).

3. job script for the automatic submission of local optimization jobs to the
queueing system during the GA. Such a script starts a geometry optimization
using tmole20 and, e.g., a Tmole input file named “turbo.in” with the line:

tmole20 -l -c coord turbo.in > turbo.log &> turbo.err

A file containing seed structures can be optionally included in the directory. The
GA is started by calling the DoDo main program with:

nohup genetic --gen_inp=INPUT_FILE >& OUTPUT_FILE &

This initiates a process running in background of the clusters front-end that: i)

supervises the GA run, ii) stores the population and the history of the run, iii)
performs the natural selection algorithm, iv) calls specialized genetic operator mod-
ules, v) performs input/output operations, vi) interacts with the queueing system.
The generated output file summarizes the progress of the run including information
about minimal and average energies of structures in a certain population, crossover
and mutation operations as well as the status of queued jobs.
Each local optimization is performed in a separate directory and as an independent
job (stored in “calculations”). The Tmole input file and job script supplied by
the user are copied into the created working directory. Then, the molecular structure
to be optimized is automatically exported as coord file and the job is submitted to
the queuing system using the command compatible with the given job scheduling
software (e.g. qsub for PBS). The DoDo package stores the unique identifier returned
by the queuing system upon the job submission. Job progress is periodically checked
using the identifier and the proper command (e.g. qstat for PBS). After the job is
finished, the relevant output files created by TURBOMOLE are parsed in order to obtain
the final structure definition and the corresponding energy.
The structures of the current population (with the highest fitness) are stored in a
file named “POPULATION.dodo” while the file “DEAD.dodo” contains all structures
142 CHAPTER 5. STRUCTURE OPTIMIZATIONS

that were removed from population due to the natural selection algorithm. A com-
plete history of previous populations and rejected structure models is stored in the
directory “structures”. These files as well as the files for the initial seed structures
use the internal DoDo format (distance units: Bohr).
For conversion between the DoDo format and other commonly used file formats (e.g.,
car and xyz) one can employ the script dodoconvert along with the following op-
tions:

--inpfile=FILE name of file to dodoconvert (default: POPULATION.dodo)

--inpform=FORMAT file format of base structure: car, xyz or dodo (default: dodo)
--outform=FORMAT output file format: car, arc, xyz, mxyz (xyz file optimized
for molden) or dodo (default: mxyz)
--ab change coordinates from Ångstrom to Bohr (default: False)
--ba change coordinates from Bohr to Ångstrom (default: False)

Should it be necessary to abort a GA run one can use genetic --clear to remove
all submitted jobs from the queue before stopping the main process. In order to
restart the GA, set the number of (random) initial structures to zero and remove
the declaration of the seed file (if used) in the DoDo input file. Then, the latest
“POPULATION.dodo” and “DEAD.dodo” files from the previous run are automati-
cally read when genetic is executed again.

5.7.3 The DoDo Input File

A sample input file used for the DoDo genetic algorithm, “genetic_example.inp”, is
generated by calling:

genetic --tmole_example

Such an input file presents all keywords implemented in the program. For clarity,
these keywords are separated in sections but the file is (almost) free formatted. The
hash sign # starts a comment. The keywords and corresponding default values are
summarized in the list presented below.

General settings

population size SIZE(int)

Defines the maximum population size. SIZE > 1, default: SIZE = 20.
min max queued structures MIN(int) MAX(int)
Defines the minimum and maximum number of the local optimization jobs
present at any moment in the queue. Use MIN = 1 to have a ’standard’
generation-based GA run. MAX ≥ MIN ≥ 1, defaults: MIN = 1, MAX =
10.
5.7. GLOBAL STRUCTURE OPTIMIZATION – THE DODO PROGRAM 143

optimized structures MAX(int)

Defines the total number of structures optimized during the run. Default:
MAX = 100.

Initial structures generator settings

gen ini max tries MAX(int)

Defines the maximum number of tries performed in order to generate a single
initial structure model. Default: MAX = 1000.

initial structures NUMBER(int)

Defines the total number of generated initial structures. Default: NUMBER
= 20.

atomic radii bounds A(float) B(float)

Defines the multipliers for atomic radii used to check whether two atoms are
bound. Defaults: A = 0.7, B = 1.2.

composition
SYMBOL1(string) MIN1(int) MAX1(int)
SYMBOL2(string) MIN2(int) MAX2(int)
SYMBOL3(string) MIN3(int) MAX3(int)
...
end composition
The composition block defines the desired atomic composition of the initial
structures. SYMBOL is a chemical symbol of the given element. MIN and
MAX define minimum and maximum number of atoms for that element. Only
systems with constant composition during the GA run are currently available
(MAX = MIN ).

seed file FILENAME(string)

Declares the file that contains seeds given in the internal DoDo format. Default:
FILENAME = OLD_POPULATION.dodo.

seeds amount NUMBER1(int) NUMBER2(int) NUMBER3(int)...

Defines how many initial species will be generated from each seed given in
the seed file. The number of the values has to be the same as the count of
the seeds given in the seed file, i.e. each seed in the seed file has to be
accounted for.

Crossover and mutation settings

constant composition mode

This line is a sanity check. If this line is present in the input file, MIN has to
be equal to MAX for each SYMBOL in the composition section. Otherwise,
an error is raised. Only this mode is currently available.
144 CHAPTER 5. STRUCTURE OPTIMIZATIONS

crossover max tries MAX(int)

Defines the maximum number of tries performed in order to generate a single
structure model. Default: MAX = 1000.

fitness scaling FACTOR(float)

Defines the scaling factor α used in Eq. (5.5). Default: FACTOR = 1.0.

mutation probability VALUE(float)

Defines the mutation probability of structures within the GA population. De-
fault: VALUE = 0.01.

Selection settings

selection RMS VALUE(float)

Defines the maximum root mean square of the distances between the cor-
responding atoms for which two structures are considered similar. Default:
VALUE = 0.3.

selection dstmax VALUE(float)

Defines the maximum distance [Å] used to seek for the corresponding atoms.
Default: VALUE = 0.3.

Software control settings

queueing system NAME(string)

Defines which job scheduling software interface should be used. NAME = pbs
or NAME = hlrn, default: NAME = pbs.

time interval VALUE(int)

Defines how often [seconds] the GA checks the queuing system for the presence
of submitted optimization jobs. Default: VALUE = 600.

computational platform NAME(string)

Defines the computational platform (TURBOMOLE in combination with Tmole).
NAME = turbomole.

job file NAME(string)

Defines the name of file which will be submitted to the queueing system. De-
fault: NAME = job.run.

tmole file NAME(string)

This file is used by Tmole2.0. NAME given here HAS to agree with data given
in job file.
5.8. COUNTERPOISE-CORRECTIONS USING THE JOBBSSE SCRIPT 145

5.8 Counterpoise-Corrections using the Jobbsse Script

The shell script jobbsse controls and executes the automatic calculation of the
counterpoise correction as it has been formulated by Boys and Bernadi (S. F. Boys
and F. Bernardi, Mol. Phys., 19, 553 (1970)) to estimate the Basis Set Superposition
Error (BSSE). For a dimer, the cp-correction takes the form for the monomers A and
B:
CP
EAB = EAB − (EA(B) − EA ) − (EB(A) − EB )
Where parentheses denote ghost basis sets without electrons or nuclear charges.
For a trimer jobbsse used by default the conventional so-called site-site functional
counterpoise corrections:
CP
EABC = EABC − (EA(BC) − EA ) − (EB(AC) − EB ) − (EC(AB) − EC ) .

jobbsse works similar as the jobex script: it cycles through the SCF/DFT and,
if needed, gradient and force relaxation programs and stops if either the maximum
number of cycles is reached or the convergence criteria (change in the total energy,
maximum norm of the gradient) are fulfilled. It does either only energy calculations
or a full geometry optimization including up to three fragments. By default, the
executable programs are taken from the load modules library within the TURBOMOLE
directory.
Note that you need to set up the fragments (and possibly their symmetries using
define in the geometry menu beforehand. The general structure of a jobbsse cal-
culation is as follows:

1. bsseenergy is invoked to generate input files for define, which is then used
to prepare the control files (including occupation numbers, initial guess MOs,
etc.) for the different “ghost“ and monomer calculations and shell scripts with
commands for calculations on these fragments.

2. jobbsse cycles over the supermolecular complex and the fragments and com-
putes the energies and, if requested, gradients for them. Then the counterpoise-
corrected results are evaluated and written to the standard data groups ($en-
ergy and $grad).

3. For geometry optimizations one of the structure relaxation codes (statpt or

relax) is invoked to update the coordinates and check for convergence. If the
structure optimization is not converged jobbsse continues with the previous
step.

Note, that counterpoise-corrected calculations with jobbsse are NOT as black-box as

ordinary geometry optimizations with jobex. The input generated for the fragments
are based on the default occupation numbers obtained from the EHT guess, default
assignments for the frozen orbitals, memory, etc. Since this might be different from
what is needed (or even fail), it is recommended to let jobbsse stop after the initial
setup step using the flag -setup and to check carefully the assigned basis sets,
146 CHAPTER 5. STRUCTURE OPTIMIZATIONS

occupations number and subsystem symmetries. In particular, for MP2 or CC2

calculations with molecules containing not only the atoms H–Ar also the number of
frozen orbitals should be checked, and if necessary corrected.

5.8.1 Options

Given a shell the usage is:

nohup jobbsse &

This command invokes cp-correction, and, if needed structure optimization using the
default program statpt. Note, that the program needs to know which calculation is
being done. Structure optimizations using program relax can be performed using
-relax flag:

nohup jobbsse -opt -relax &

nohup means that the command is immune to hangups, logouts, and quits. & runs a
background command. jobbsse accepts the following arguments controlling the level
of calculation, convergence criteria and many more (for example nohup jobbsse -gcart 4 &):

-energy integer converge total energy up to

10(−<integer>) Hartree (default: 6)

-gcart integer converge maximum norm of Cartesian gradient up to

10(−<integer>) atomic units (default: 3)

-c integer perform up to integer cycles (default: 100)

-gradient calculate the gradient as well

-opt optimise the structure

-relax use the relax program for force relaxation

-level level define the optimization level, level =scf, dft, mp2, or cc2
(default is scf). Note that the program needs this input! If
the level is DFT, the grid will be automatically set to m4.

-ri use RI modules ridft and rdgrad (fast Coulomb approxi-

mation) instead of dscf and grad as well as rimp2 instead
of mpgrad

-l <path> employ programs from directory <path>

-mem integer Is able to control the memory from outside define Note that
if you did not define any memory, it is automatically set to
1 GB
5.8. COUNTERPOISE-CORRECTIONS USING THE JOBBSSE SCRIPT 147

-trimer calculates, in case we have a trimer:

Energy = ABC - AB(C) + AB - AC(B) + AC - BC(A) +
BC
rather than
Energy = ABC - A(BC) + A - B(AC) + B - C(AB) + C
(note that the first term neglects the BSSE in the dimer)

-setup Interrupt calculation after the initial setup step to check and
possibly correct the control files for the fragments and the su-
permolecule. To continue, start jobbsse without the -setup
option.

-help shows a short description of the commands above

5.8.2 Output

There will be an output written to file bsse_out. In this file, you will find all indi-
vidual energies computed which were used to calculate the last cp-corrected energy.
The same holds true for the last gradients, which are written to grad_out.
The convergence criteria and their current values are written out at the not.converged
file. For the possible options to control convergence check the subsection for the opti-
mization program used (statpt, which is used by default, or relax). Since for weak
complexes the force constants for intra- and intermolecular bonds very strongly in
magnitude, it is recommended to use whenever possible redundant internal coordi-
nates.
148 CHAPTER 5. STRUCTURE OPTIMIZATIONS

5.9 Reaction Path Optimization

5.9.1 Background and Program structure

The goal of self-consistent optimization of the reaction path (RP) is usually to ob-
tain an initial guess for Transition State Search or an approximation to the barrier.
Methods that use reactant and product structure to compute the RP are often re-
ferred to as ’double-ended’ methods or, if the the RP is discretized, ’chain-of-states’
methods. [70–72]
The RP connects reactant and product, its highest point being the transition state.
It is a steepest descent path, which means that its tangent t is always parallel to
the gradient g. The RP is in actual calculations discretized by a finite number of
structures n. The tangents are parallel to the gradients for all structures i = 1, .., n.
Assuming normalized tangents tT i ti = 1, this can be written as:

0 = (1 − ti tT
i )gi (5.6)

Several approximations for the tangents ti exist [72, 73], usually using finite differ-
ence schemes. The most common methods, the Nudged Elastic Band (NEB) [72, 73]
and String Method (SM) [74] prevent the structures from ’sliding’ down the re-
action path towards products and reactants with additional springs or interpola-
tion/redistribution algorithms. The method used here achieves equal spacing via
constrained optimization assuming a quadratic potential. [75] An initial path is pro-
vided using a slight variation of the Linear Synchronous Transit [70].
The structure of the optimization is the same as in other TURBOMOLE structure
optimizations: As the jobex script drives optimizations by calling statpt/relax as
well the SCF and gradient modules.The woelfling-job script drives optimizations
by calling woelfling as well as the SCF and gradient modules. The woelfling-
job scripts reads the current path from file path.xyz which is the output of the
woelfling program. woelfling-job then creates folders to run the calculations of
each structure in, gathers coordinates and gradients, and then calls woelfling again.
The aim of RP optimization is usually not to optimize the RP to some accuracy,
but to obtain an initial guess for a TS optimization. It is in general not possible to
find a convergence criterion (and a corresponding threshold) that guarantees a good
initial guess. The maximum number of iterations and the convergence threshold are
therefore relatively high and tight. One can extract a TS guess also during the course
of the optimization. If the TS search is successful (or not) you can stop (or restart)
the RP optimization. Apart from simply killing the program you can add a ’stop’
file in the (scratch) directory, in which the script runs. It will then terminate at the
end of the current cycle and can easily be restarted.

5.9.2 Input Structure

Options can be modified using keywords in the $woelfling data-group. The most
important options are:
5.9. REACTION PATH OPTIMIZATION 149

$woelfling
ninter 14
riter 0
ncoord 2
align 0
maxit 40
dlst 3.00000000000000
thr 1.000000000000000E-004
method q

The values above are the default values. If $woelfling is missing, it will be added
during the first woelfling run and default values will be set. Most importantly,
ncoord is the number of input structures provided, ninter is the number of struc-
tures to be used for discretization of the path and maxit is the number of cycles
to run. If align 0, structures will be rotated/translated to minimize the cartesian
distance, for align 1 structures will be used as provided. Using method qg instead
of method q, a reaction path will be grown as in the growing string method. To
start a RP optimization you need to provide at least a reactant and a product struc-
ture ncoord ≥ 2. You may provide more structures if you have a guess for the
reaction path. The input structures will be used to compute an initial guess with
ninter structures that is then optimized. Reactant and product structure will stay
fixed throughout the optimization. All structures have to have the same ordering of
atoms!
The input structures are provided in a file coords, which contains merged coord
files. All ncoord structures are given in the right order in the typical TURBOMOLE
coord format.

5.9.3 How it works

Minimum Input/Quick and Dirty

1. Make a usual TURBOMOLE input using the coord file of either reactant of
product structure.

2. Join the coord-files of reactant and product in a file coords.

3. Run woelfling-job

4. Check the output and the path (path.xyz) to extract a TS guess.

It is usually a good idea to check the initial path before starting the calculation. Once
you have prepared the input, simply run woelfling directly and check path.xyz. If
it looks reasonable, just run the woelfling-job script.

Unsuccessful Optimization If there is no reasonable TS guess or a frequency

calculation does not give the correct number of imaginary frequencies, you can:
150 CHAPTER 5. STRUCTURE OPTIMIZATIONS

1. Check if have used the best structure as TS guess, maybe the structure with
the highest energy is not the best.

2. Check if the RP has a reasonable amount of structures (if they are far apart,
it is unlikely that a structure is close to the TS)

3. Check if the RP is reasonably converged (mean of rms(gi⊥ ) in output <1.0d-3;

path is continuous in terms of energy and structure)

(a) If it is not yet converged, converge it.

(b) If it is not going to converge, provide useful structures for the initial guess
or maybe use more structures for the path.

Parallelisation woelfling-job provides a few basic options to allow parallel cal-

culation

-mfile filename
specify a file that contains the names of the machines that should be used for
parallel execution (machinefile format with one entry per CPU core). If the
job is run in a PBS queuing system, the filename defined in the environment
variable $PBS_NODEFILE will be used per default.

-nthreads n
specify the number of threads (CPU cores) that should be used for the indi-
vidual energy and gradient calculations. If not specified, each calculation will
use all the cores speficied in the machinefile for a given machine name, one
calculation will be started per unique machine name.

-scrpath directory
specify a path to a local file system for scratch files. If not specified, the path
defined in the environment variable $TURBOTMPDIR will be used instead, if set.
If no path is given and $TURBOTMPDIR is not set, scratch files will be written
into subdirectories of the job’s working directory.

woelfling-job uses for parallel calculations (exclusively) the SMP-parallel binaries

which will be invoked through ssh on the machines specified in the machinefile. If
CPU cores on more than one machine are used, woelfling-job must be started in
a working directory that is available on all machines. To avoid that I/O into a nfs
filesystem cause a bottleneck, a scratch directory in a local file system (which must
exist with the same name on all machine) should be specified with the -scrpath
option or the $TURBOTMPDIR variable. On machines with a large number of cores the
-nthreads option can be used to obtain a higher parallelization efficiency by starting
more than one energy/gradient calculation (each with n cores) per machine.

Freezing of coordinates The current version of woelfling ignores fixed internal

coordinates, only fixed cartesian coordinates are recognized.
5.9. REACTION PATH OPTIMIZATION 151

Calculations with fixed cartesian coordinates require that the same atoms are marked
as fixed in all start structures (e.g. first and last point of the pathway) and have in
all start structures identical cartesian coordinates.

Restart

1. If you have stopped the calculation adding a ’stop’ file, you can just run
woelfling-job again.

2. If you have run the maximum number of cycles, just increase maxit and run
woelfling-job again. It will then run from the old maxit to the new maxit.

If the calculation crashed ’on its own’ it is likely that the SCF failed to converge -
improve the corresponding options. If the optimization crashed in the middle of the
run, it is most likely that it crashed during a SCF or gradient step, since basically all
CPU time is spent there. In that case, remove at least the folder where the SCF and
gradient program had been running when the program crashed. The files necessary
for woelfling should be intact and you can simply restart it.

Restart - more details If the files are not intact, one can still use the optimized
coordinates in one way or the other. There are basically three phases which are
explicitly indicated by the number of iterations riter =:

0 woelfling reads control, coords to generate an initial guess path.xyz

1 woelfling reads control, gradients to compute an optimized path.xyz. Hes-

sian are initialized and written to hessians-new

>1 woelfling reads control, gradients, oldgradients and hessians to update

Hessians and compute an optimized path.xyz. Hessian are updated and writ-
ten to ’hessians-new’.

Therefore repeated execution of woelfling will yield the same output (unless new
energies/gradients are computed). The woelfling-job script takes care of the file-
handling and should enable restart at any time. If files are damaged it will be hard
to gather gradients and corresponding Hessian- and old gradient information. If
you have an intact gradients or oldgradients-file (necessary condition: number of
lines=(3 + 2 × natoms)×number of structures), name it gradients, set riter in the
control-file to 1 and restart. If those file are not intact, you can extract whatever
structure information you have obtained and use it to provide a better initial guess:
Modify file coords and ncoord in the control-file accordingly, set riter to 0 and
restart.
Chapter 6

Hartree–Fock and DFT

Calculations for Molecular
Systems

Energy and gradient calculations at the Hartree–Fock (HF) and DFT level can be
carried out in two ways: dscf and grad perform conventional calculations based on
four–center two–electron repulsion integrals (ERI’s); ridft and rdgrad employ the
RI–J approximation, as detailed below.
dscf and grad are modules for energy and gradient calculations at the HF and DFT
level, which use an efficient semi–direct SCF algorithm. Calculation of the Coulomb
and HF exchange terms is based on the conventional method employing four–center
two–electron repulsion integrals (ERI’s). These modules should be used for HF
and DFT calculations with exchange-correlation functionals including HF exchange
contribution, e.g. B3–LYP, if further approximations (RI–J) are to be avoided. All
functionalities are implemented for closed–shell RHF and open–shell UHF reference
wavefunctions. Restricted open shell treatments (ROHF) are supported on the HF
level only, i. e. not for DFT.
The most important special features of the dscf and grad modules are:

• Selective storage of the most time consuming and frequently used integrals.
The integral storage is controlled by two threshold parameters, $thize and
$thime, related to integral size and computational cost.

• Efficient convergence acceleration techniques for energy calculations. They in-

clude standard methods for convergence acceleration (DIIS), which reduce the
number of SCF iterations needed as well as methods to reduce the effort within
each iteration when the calculation is almost converged (integral prescreening
and differential density scheme).

152
 153

ridft and rdgrad are modules for very efficient calculation of energy and gradient
at the Hartree–Fock (HF) and DFT level [76]. Both programs employ the Resolution
of the Identity approach for computing the electronic Coulomb interaction (RI–J).
This approach expands the molecular electron density in a set of atom–centered
auxiliary functions, leading to expressions involving three–center ERI’s only. This
usually leads to a more than tenfold speedup for non–hybrid DFT compared to
the conventional method based on four–center ERI’s (for example the dscf or grad
module).
The combination of RI–J for Coulomb–interactions with a case–adapted conventional
exchange treatment reduces the scaling behaviour of the (conventional) exchange
evaluation required in HF–SCF and hybrid DFT treatments. Usage of ridft and
rdgrad for HF and hybrid DFT is of advantage (as compared to dscf and grad) for
larger systems, where it reduces computational costs significantly.
The most important special features of the ridft and rdgrad modules are:

• A very efficient semi-core algorithm for energy calculation. The most expensive
three–center integrals are kept in memory which significantly reduces the com-
putational time for small and middle sized molecules. The amount of stored
integrals is controlled by simply specifying the amount of free memory using
the keyword $ricore.
• Multipole accelerated RI for Coulomb (MARI–J ) linear scaling (O(N )) method
for large molecules. It significantly reduces calculation times for molecules with
more than 1000 basis functions.

All algorithms implemented in dscf, grad, ridft, and rdgrad modules can exploit
molecular symmetry for all finite point groups. Typically, the CPU time is pro-
portional to 1/NG , where NG is the order of the nuclear exchange group. Another
important feature is a parallel implementation using the MPI interface.
Additionally dscf and ridft modules include the following common features:

• An UHF implementation [77] with automatic generation of optimal start vec-

tors by solving the HF instability equations [78] in the AO basis (see the key-
word $scfinstab for detailed information).
• Occupation number optimization using (pseudo-Fermi) thermal smearing.

RI-techniques can also be used for the Hartree–Fock exchange part of the Fock
matrix (RI-HF). This is done by the ridft-module, if the keyword $rik is found in
the control file. In this case ridft performs a Hartree–Fock-SCF calculation using
the RI- approximation for both J and K, if suitable auxiliary basis sets (which differ
from that used for fitting of the Coulomb part only) are specified. This is efficient
only for comparably large basis sets like TZVPP, cc-pVTZ and larger.
HF-exchange can also be calculated semi-numerically [79]. The calculation of 4c-2e-
Integrals is split into an analytical and a numerical part. The latter is evaluated on a
dft-type integration grid. The semi-numerical calculation scales better with system
size than RIK and is suitable for large molecules and large basis sets.
154CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

Prerequisites

Both dscf and ridft require the control file and starting orbitals obtained from
the extended Hückel guess using define.
Energy calculations using dscf can be performed in a direct or semi-direct mode.
In the direct mode all four-center ERI’s are recalculated at each SCF iteration. The
semi-direct mode uses a selective storage of the most time consuming and frequently
used integrals. The amount of integrals stored is controlled by the keywords $thize
and $thime, related to integral size and computational cost. The semi-direct mode
requires a separate dscf statistics run to estimate the disk space needed for integral
storage. The statistics run requires the keyword $statistics dscf to be present in
the control file. It can be set either manually or using the tool Stati.
For ridft and rdgrad following additional prerequisites are required:

1. An auxiliary basis defined in the data group $jbas. This group is created
automatically when using ri menu of define.
2. The maximum core memory the program is allowed to allocate should be de-
fined in the data group $ricore; the recommended value is 75–85% of the
available (physical) core memory.
3. Calculations using MARI-J method require the keyword $marij.
4. For RI-HF-calculations auxiliary bases defined in the data group $jkbas are
needed. This group is created by the rijk menu in define.

How to Perform a Calculation

Single point calculations
Call the dscf or ridft program after running define.
Geometry optimizations and molecular dynamics
For HF or DFT calculations using dscf and grad simply invoke jobex.
For DFT calculations using ridft and rdgrad type jobex -ri; see Sec-
tion 5.1 for additional options and parameters for geometry optimizations
and ab initio molecular dynamics calculations.

6.1 Background Theory

In Hartree–Fock theory, the energy has the form,

EHF = h + J − K + Vnuc , (6.1)
where h is the one-electron (kinetic plus potential) energy, J is the classical Coulomb
repulsion of the electrons, K is the exchange energy resulting from the quantum
(fermion) nature of electrons, and Vnuc is the nuclear repulsion energy.
6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 155

In density functional theory, the exact Hartree–Fock exchange for a single determi-
nant is replaced by a more general expression, the exchange-correlation functional,
which can include terms accounting for both exchange energy and the electron cor-
relation which is omitted from Hartree–Fock theory. The DFT energy is expressed
as a functional of the molecular electron density ρ(r),

EDF T [ρ] = T [ρ] + Vne [ρ] + J[ρ] + Ex [ρ] + Ec [ρ] + Vnuc , (6.2)

where T [ρ] is the kinetic energy, Vne [ρ] is the nuclei-electron interaction, Ex [ρ] and
Ec [ρ] are the exchange and correlation energy functionals.
The exchange and correlation functionals normally used in DFT are integrals of some
function of the density and possibly the density gradient. In addition to pure DFT
methods, dscf and grad modules support hybrid functionals in which the exchange
functional includes the Hartree–Fock exchange, e.g. B3-LYP.

6.2 Exchange-Correlation Functionals Available

The following exchange-correlation functionals are available:

• LDAs: S-VWN, PWLDA

• GGAs: B-VWN, B-LYP, B-P, PBE

• MGGA: TPSS; SCAN; M06 (using XCFun)

• hybrid functionals: BH-LYP, B3-LYP, PBE0, TPSSh; M06-2X (using XCfun)

• range-separated hybrid functionals: CAM-B3LYP (using XCfun, LibXC), wB97(X)(-

V), wB97M-V, LC-ωPBE class (using LibXC)

• double–hybrid functional: B2-PLYP (energy calculations only!)

• local hybrid functionals: Lh07t-SVWN, Lh07s-SVWN, Lh12ct-SsirPW92,

Lh12ct-SsifPW92, Lh14t-calPBE, LH20t, PSTS, mPSTS, LHJ14

For EXX and LHF, see Chapter 20

The XCFun library (Arbitrary-Order Exchange-Correlation Functional Library) by
Ulf Ekström and co-workers has been included [80] and some of the functionals
implemented there can now be utilized. Among them are the empirically fitted
MGGAs M06 and M06-2X from the Truhlar group [81]. XCFun functionals are
available for energy, gradient, vib. frequencies and TDDFT excited state energy
calculations - with and without RI approximation. For details and the license of
XCFun please refer to its web site https://github.com/dftlibs/xcfun
The LibXC 5.0.0 library has been included [82] and some of the functionals imple-
mented there can now be utilized. Among them are the empirically fitted MGGAs
156CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

from the Truhlar group as well as the range-separated wB97(X) group of Head-
Gordon [83]. For details and the license of LibXC please refer to its web site
https://tddft.org/programs/libxc/
See the next chapters for available functionals from LibXC and XCFun. For range-
separated hybrid functionals see the notes and restrictions in the corresponding sec-
tion.
In detail, the Turbomole own functional library consists of:

• The Slater–Dirac exchange functional only (S) [84, 85].

• The 1980 correlation functional (functional V in the paper) of Vosko, Wilk,

and Nusair only (VWN) [86].

• A combination of the Slater–Dirac exchange and Vosko, Wilk, and Nusair 1980
(functional V) correlation functionals (S-VWN) [84–86].

• The S-VWN functional with VWN functional III in the paper. This is the
same functional form as available in the Gaussian program [84–86].

• A combination of the Slater–Dirac exchange and Perdew-Wang (1992) correla-

tion functionals [84, 85, 87].

• A combination of the Slater–Dirac exchange and Becke’s 1988 exchange func-

tionals (B88) [84, 85, 88].

• Lee, Yang, and Parr’s correlation functional (LYP) [89].

• The B-LYP exchange-correlation functional (B88 exchange and LYP correla-

tion functionals) [84, 85, 88, 89].

• The B-VWN exchange-correlation functional (B88 exchange and VWN (V)

correlation functionals) [84–86, 88].

• The B-P86 exchange-correlation functional (B88 exchange, VWN(V) and

Perdew’s 1986 correlation functionals) [84–86, 88, 90].

• The Perdew, Burke, and Ernzerhof (PBE) exchange-correlation functional [84,

85, 87, 91].

• The Tao, Perdew, Staroverov, and Scuseria functional (Slater–Dirac, TPSS

exchange and Perdew-Wang (1992) and TPSS correlation functionals) [84, 85,
87, 92].

• The Strongly Constrained and Appropriately Normed (SCAN) meta-GGA

functional [93].

For the SCAN functional, be advised that the convergence of the total energy and
energy gradients with respect to the radial grid used to integrate the exchange-
correlation energy and potential is slower than previous non-empirical function-
als [94]. A radial grid size of 40 is typically needed to converge the total energy
6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 157

(dscf or ridft) to µH accuracy, while a radial grid size of 50 is typically needed to

converge the energy gradients (grad or rdgrad). Convergence with respect to the
angular grids remains unchanged. The radial grid size can be set in the control file
through the radsize keyword.
For the r2SCAN and r2SCAN-3c functionals used for geometry optimizations a grid-
size of m4 is usually sufficient if the radsize keyword is set to 8.
Additionally, for all four modules (dscf, grad, ridft, and rdgrad) the following
hybrid functionals are available (a mixture of Hartree–Fock exchange with DFT
exchange-correlation functionals):

• The BH-LYP exchange-correlation functional (Becke’s half-and-half exchange

in a combination with the LYP correlation functional) [84, 85, 88, 89, 95].

• The B3-LYP exchange-correlation functional (Becke’s three-parameter

functional) with the form,

0.8S + 0.72B88 + 0.2HF + 0.19V W N (V ) + 0.81LY P (6.3)

where HF denotes the Hartree-Fock exchange [84, 85, 88, 89, 96].

• The B3-LYP exchange-correlation functional with VWN functional V in the

paper. This is the same functional form as available in the Gaussian program.

• The 1996 hybrid functional of Perdew, Burke, and Ernzerhof, with the form,

0.75(S + P BE(X)) + 0.25HF + P W + P BE(C) (6.4)

where PBE(X) and PBE(C) are the Perdew–Burke–Ernzerhof exchange and

correlation functionals and PW is the Perdew–Wang correlation functional [84,
85, 87, 91, 97].

• The TPSSH exchange-correlation functional of Staroverov, Scuseria, Tao and

Perdew with the form,

0.9(S + T P SS(X)) + 0.1HF + P W + T P SS(C) (6.5)

where HF denotes the Hartree–Fock exchange [84, 85, 87, 92, 98].

The Double-Hybrid Functional B2-PLYP can be used for single point energy calcu-
lations. Note that one has to run an MP2 calculation after the DFT step to get the
correct B2-PLYP energy!
B2-PLYP is a so-called double-hybrid density functional (DHDF) [99] that uses in
addition to a non-local exchange contribution (as in conventional hybrid-GGAs)
also a non-local perturbation correction for the correlation part. In the following
options/restrictions in the present version of this method:

• single point calculations only (computed with the dscf/ridft and rimp2/ricc2
modules).
158CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

• UKS treatment for open-shell cases.

• can be combined with resolution-of-identity approximation for the SCF step

(RI-JK or RI-J option).

• can be combined with the dispersion correction (DFT-D method, s6 (B2-PLYP)=0.55).

The non-local perturbation correction to the correlation contribution is given by

second-order perturbation theory. The idea is rooted in the ab initio Kohn-Sham
perturbation theory (KS-PT2) by Görling and Levy [100, 101]. The mixing is de-
scribed by two empirical parameters ax and ac in the following manner:

EXC (DHDF ) = (1 − ax )EX (GGA) + ax EX (HF ) (6.6)

+(1 − ac )EC (GGA) + ac EC (KS − P T 2),

where EX (GGA) is the energy of a conventional exchange functional and EC (GGA)

is the energy of a correlation functional. EX (HF ) is the Hartree-Fock exchange
of the occupied Kohn-Sham orbitals and EC (KS − P T 2) is a Møller-Plesset like
perturbation correction term based on the KS orbitals:

1 X X (ia|jb)[(ia|jb) − (ib|ja)]
EC (KS − P T 2) = . (6.7)
2 ei + ej − ea − eb
ia jb

The method is self-consistent only with respect to the first three terms in Eq. 6.6,
i.e., first a SCF using a conventional hybrid-GGA is performed first. Based on these
orbitals EC (KS − P T 2) is evaluated afterwards and added to the total energy.
For B2-PLYP, B88 exchange [88] and LYP correlation [89] are used with the param-
eters ax = 0.53 and ac = 0.27. Due to the relatively large Fock-exchange fraction,
self-interaction error related problems are alleviated in B2-PLYP while unwanted
side effects of this (reduced account of static correlation) are damped or eliminated
by the PT2 term.
How to use B2-PLYP:

• during preparation of your input with Define select b2-plyp in the DFT
menu.

• carry out a Dscf run. Prepare and run a RI-MP2 calculation with either
rimp2 or ricc2 program modules.

• the RI-MP2 program directly prints the B2PLYP energy if this functional has
been chosen before

Or use the b2plypprep script to setup up the calculation.

• define coord and basis set

• (optional: switch on ri or rijk and define jbasis or jkbasis)

6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 159

• run b2plypprep

• run dscf (or ridft) and ricc2

Local hybrid functionals [102] feature an admixture of the exact-exchange energy

density to a (semi-)local exchange energy density in real space managed by a local
mixing function (LMF). They can therefore be viewed as a more flexible generaliza-
tion of global hybrid functionals.
So far, local hybrid functionals are available in the modules ridft [103], grad, rdgrad
[104], escf [37, 105–108], egrad [109], and mpshift [108, 110]. For the calculation of
non-standard exact-exchange integrals, a semi-numerical integration scheme [111] is
used. The screening procedure is outlined in [107] and the parallelization is described
in [107] and [108]. The usual DFT grids are employed for this task and small grids
are thus recommended if possible. In escf-calculations, a larger amount of main
memory is required for the semi-numerical integration. Therefore, additional main
memory of approximately 0.0004 · NBF 2 MB should be provided (for each CPU).
The following local hybrid functionals are available so far:

• Lh07t-SVWN: [112] Slater-Dirac exchange [84, 85] and VWN [86] with t-LMF
(prefactor of 0.48)

• Lh07s-SVWN: [113] Slater-Dirac exchange [84, 85] and VWN [86] with s-LMF

• Lh12ct-SsirPW92: [114] Slater-Dirac exchange [84,85] and self-correlation-reduced

PW92 correlation with common t-LMF (prefactor of 0.646)

• Lh12ct-SsifPW92: [114] Slater-Dirac exchange [84, 85] and self-correlation-free

PW92 correlation with common t-LMF (prefactor of 0.709)

• Lh14t-calPBE: [115] PBE exchange and correlation [84, 85, 87, 91] with t-LMF
(prefactor of 0.5) and pig1 calibration [116]

• LH20t: [117] PBE exchange and refitted B95 correlation [84,85,87,91,118] with
t-LMF (prefactor of 0.751) and pig2 calibration [116]

• PSTS and mPSTS: [108,119] (modified) functional of Perdew, Staroverov, Tao,

and Scuseria (PSTS-LMF). Use option mpsts-noa2 for closed-shell systems.

• LHJ14: [108, 120] Johnson’s local hybrid functional based on the correlation
length (z-LMF).

D3 parameters for the t-LMF based local hybrid functionals have been implemented.
[121] PSTS and mPSTS use the D3 parameters of TPSSh. Note that Lh14t-calPBE
is not available in egrad and requires larger numerical grids in escf-calculations.
For gradient calculations it is strongly recommended to use the weight derivatives
keyword in the $dft data group to avert numerical inaccuracies.
160CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

6.2.1 Exchange-Correlation Functionals from LibXC library

The LibXC library is taken from https://tddft.org/programs/libxc/

The current TURBOMOLE version uses a LibXC 5.1.0. that has been adapted and
checked for the provided functionals. All listed functionals support up to 3rd deriva-
tives.
For some functionals included in LibXC shortcuts have been set. If a shortcut is
present LibXC functionals can be similar to TURBOMOLE own functionals

$dft
functional wb97x

Shortcuts are available for the following functionals:

wb97 -- wb97 range-separated hybrid GGA

wb97x -- wb97x range-separated hybrid GGA
wb97x-v -- wB97X range-separated hybrid GGA with VV10 non-local correlation
wb97m-v -- wB97X range-separated hybrid metaGGA with VV10 non-local correlation
wb97x-d -- wB97X range-separated hybrid metaGGA with dispersion correction
sogga11 -- sogga11 GGA
sogga-11x -- sogga-11x hybrid GGA
m05 -- M05 hybrid metaGGA
m05-2x -- M05 hybrid metaGGA with doubled exact exchange
m06 -- M06 hybrid metaGGA
m06-2x -- M06 hybrid metaGGA with doubled exact exchange
m06-l -- M06-L metaGGA
m11 -- M11 hybrid metaGGA
m11-l -- M11-L metaGGA
revM11 -- revised M11 hybrid metaGGA
mn12-l -- MN12-L metaGGA
mn12-sx -- MN12-SX short-range-separated metaGGA
mn15 -- MN15 hybrid metaGGA
mn15-l -- MN15-L metaGGA
pkzb -- Perdew-Kurth-Zupan-Blaha metaGGA
task -- ultra-nonlocal metaGGA (TASK exchange + PW92 correlation)
bmk -- BMK hybrid metaGGA
scan-libxc -- SCAN metaGGA functional
scan0 -- hybrid metaGGA functional based on SCAN
cam-b3lyp -- CAM-B3LYP range-separated hybrid GGA
tuned-cam-b3lyp -- CAM-B3LYP range-separated hybrid GGA tuned for excitations
6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 161

cam-qtp-00 -- CAM-QTP-00 functional

cam-qtp-01 -- CAM-QTP-01 functional
cam-qtp-02 -- CAM-QTP-02 functional
hse06-libxc -- HSE06 short-range-separated GGA
lr-wpbe -- range-separated range-separated GGA based on PBE
lrc-wpbeh -- long-range corrected range-separated hybrid GGA based on PBE
mpsts -- modified PSTS functional
mpsts-noa2 -- modified PSTS functional for closed-shell systems
lhj14 -- Johnson’s local hybrid functional

Since TURBOMOLE V7.5 certain range-separated functionals can be modified using the
following functional shortcuts and specifying the according parameters:

lc-wpbe_own omega
cam-b3lyp_own alpha beta omega
lrc-wpbeh_own alpha beta omega
hse_own alpha beta omega

alpha, beta, omega are real numbers. alpha specifies the amount of exact exchange in
the short-range limit, while in the long-range limit alpha+beta is used. The screening
parameter omega determines how fast the long-range limit will be obtained and is
defined as usual. Note that beta may be negative, converting a long-range corrected
into a short-range corrected functional.
Other functionals from the LibXC may be accessed using the index numbers from
LibXC Readme or from the Turbomole user forum. To trigger the usage of LibXC
functionals, use the keyword libxc in the $dft section.

$dft
functional libxc <index_of_main_exchange_functionals>
functional libxc add <no. of add. func.> <index_of_add_funcs>
functional libxc factors fac1 fac2 ... fac <no. of func.>
functional libxc set-rangesep alpha beta mu
functional libxc set-hybrid exx

Apart from the first line all following lines are optional. fac1, fac2... are real numbers
specifying the wheight of each component of the defined functionals in exactly the
order they have been specified. fac1 refers to the main functionals, while fac2 etc.
refer to the additional functionals. If more than one functionals is specified set-
rangesep and set-hybrid always refer to the main functionals given in the first line.
If range-sep is specified the main functional must support range-separation, otherwise
the program will stop.
The M11 hybrid metaGGA for example may be specified by the following lines
162CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

$dft
functional libxc 297
functional libxc add 1 76

In the first line the number (297) specifies the index number of the exchange part of
the M11 functional. As LibXC has stores the definition of a functional (amount of
HF exchange, range-separation parameters etc.) on the exchange part of a functional
it is mandatory to first specify the exchange part and add the correlation part. The
second line then specifies how many and which further functional parts are added.
The first number (1) specifies that one additional part is used. The number (76)
is the index number of the correlation part of the M11 functional. A maximum of
three additional X/C functionals may be specified. If exchange and correlation are
specified together then the second line can be skipped. A example for this case is
the Keal-Tozer 1 functional with index number (176):

$dft
functional libxc 176

The functionals described in this section can be used in all modules of TURBOMOLE
earlier limitations have been removed. This also applies to self-defined functionals.
Upon request a dynamically linked TURBOMOLE version which allows to use your own
version of LibXC may be requested from support@turbomole.com.
Using a dynamically linked version one has to take special care of the second deriva-
tives of metaGGAs, where the Fortran interface of LibXC is inconsistent. In TURBOMOLE
the following ordering of second derivatives in the mGGA case is expected: v2rho2,
v2sigma2, v2lapl2, v2tau2, v2rhosigma, v2rholapl, v2rhotau, v2sigmalapl, v2sigmatau,
v2lapltau. This does only affect second derivatives of metaGGAs and may change
upon future versions.

6.2.2 Exchange-Correlation Functionals from XCFun library

The XCFun library is taken from: https://github.com/dftlibs/xcfun

The current TURBOMOLE version uses XCFun 1.99 and enables the usage of individual
mixtures of the available exchange and correlation functionals.
To trigger the usage of XCFun functionals, use the keyword xcfun in the $dft section:

$dft
functional xcfun set-gga
functional xcfun <name1> <factor1>
functional xcfun <name2> <factor2>
6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 163

In addition to the name of the functional, it is necessary to tell TURBOMOLE whether

the used functional is of GGA or MGGA type. Pure LDA functionals are currently
not supported.
Available settings are:

• functional xcfun set-gga – sets a GGA functional

• functional xcfun set-mgga – sets a meta-GGA functional

• functional xcfun set-hybrid 0.2 – defines a hybrid functional with a por-

tion of 0.2 of Hartree-Fock exchange

Add the switch for either GGA or meta-GGA but not both in the same input!
List of available XCFun functionals (copied from XCFun documentation), in arbi-
trary order:

slaterx -- Slater exchange

beckex -- Becke exchange
beckecorrx -- Becke GGA exchange
ktx -- Keal-Tozer exchange
pbex -- Perdew-Burke-Ernzerhof exchange
tpssx -- TPSS original exchange functional
m05x -- Truhlar M05 exchange
m05x2x -- Truhlar M05-2X exchange
m06x -- Truhlar M06 exchange
m06x2x -- Truhlar M06-2X exchange
m06lx -- Truhlar M06L exchange
m06hfx -- M06-HF Meta-Hybrid Exchange Functional
b97x -- B97 exchange
b97_1x -- B97-1 exchange
b97_2x -- B97-2 exchange
b97_dx -- B97-D exchange ($disp3 in addition required)
optx -- OPTX Handy & Cohen exchange GGA exchange functional
optxcorr -- OPTX Handy & Cohen exchange -- correction part only
brx -- Becke-Roussells exchange with jp dependence
brxc -- Becke-Roussells exchange and correlation with jp dependence
pw86xtot -- Perdew-Wang 86 GGA exchange including Slater part
pw91x -- Perdew-Wang 1991 GGA Exchange Functional
ldaerfx -- Short range exchange LDA functional
cam-b3lyp-xcfun -- range-separated hybrid version of B3LYP
164CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

vwn5c -- VWN5 correlation

vwn3c -- VWN3 correlation
lypc -- LYP correlation
pw91c -- PW91 Correlation
pw92c -- PW92 LDA correlation
pz81c -- PZ81 LDA correlation
pbec -- PBE correlation functional
vwn_pbec -- PBE correlation functional with VWN LDA correlation
spbec -- Simplified PBE correlation functional for use with the SSB functionals
tpssc -- TPSS original correlation functional
revtpssc -- Revised TPSS correlation functional
p86c -- P86C GGA correlation
m05c -- M05 Meta-Hybrid Correlation Functional
m05x2c -- M05-2X Meta-Hybrid Correlation Functional
m06c -- M06 Meta-Hybrid Correlation Functional
m06lc -- M06-L Meta GGA Correlation Functional
m06x2c -- M06-2X Meta-Hybrid Correlation Functional
csc -- Colle-Salvetti correlation functional
brc -- Becke-Roussells correlation with jp dependence
b97_1c -- B97-1 correlation
b97_2c -- B97-2 correlation
b97_dc -- B97-D correlation ($disp3 in addition required)
b97c -- B97 correlation
ldaerfc -- Short range correlation LDA functional

pw91k -- PW91 GGA Kinetic Energy Functional

btk -- Borgoo-Tozer kinetic energy functional
tfk -- Thomas-Fermi Kinetic Energy Functional
vW -- von Weizsaecker kinetic energy

Some common functionals are pre-defined in XCFun and their individual parts do
not have to be set manually. Those aliases can be directly used as names with a
following factor of 1.0:
blyp, pbe, bp86, kt1, kt2, kt3, pbe0, b3lyp, m06, m06-2x, m06L, b3lyp-g, b3p86,
b97, b97d, olyp and some more.
Note that if the functional needs a portion of HF exchange, this has to be added
manually in the control file using functional xcfun set-hybrid <number>
Example for B3-LYP using VWN3 instead of VWN5:
6.2. EXCHANGE-CORRELATION FUNCTIONALS AVAILABLE 165

$dft
functional xcfun set-gga
functional xcfun b3lyp-g 1.0
functional xcfun set-hybrid 0.2

CAM-B3LYP from the XCfun may be used by using the shortcut:

$dft
functional cam-b3lyp

The functionals described in this section can be used for ground state energies, gra-
dients and frequency calculations as well as TDDFT spectra. TDDFT analytic gra-
dients are not yet supported up to TURBOMOLE version 7.4, please use the TURBOMOLE
own functionals instead. Later versions of TURBOMOLE will support TDDFT analytic
gradients from XCfun functionals.

Notes about range-separated hybrid functionals

TURBOMOLE now support RSHs in all modules. With RSHs the use of RI-K is not
supported. Seminumerical ($senex etc.) exchange is fully supported for RSHs.

Notes about VV10 non-local correlation dependent functionals

For functionals using VV10 non-local correlation the following hints should be noted:
VV10 can be included non-self-consistently (default) or self-consistently (keyword
$doscnl). For energies non-self-consistent VV10 (default) usually is sufficient, while
for geometry optimizations VV10 should be included self-consistently. definewill
automatically add $doscnl if the ωb97x-V or ωb97m-V functionals are selected.
escf, egrad, mpshiftand aoforcesimply ignore VV10 non-local contributions which
should be an excellent approximation in most cases. If $doscnl is added we strongly
recommend the use of m-grids (e.g. m3 which is the default).

Notes about DFT-D3, gCP and functionals using those corrections

For details about the options of DFT-D3 please see section 6.6.
In the original TURBOMOLE implementation of the B97-D functional only energy and
gradient calculations are possible due to missing higher derivatives of the functional
itself. Using the XCFun version of B97-D, analytic 2nd derivatives using aoforceand
TDDFT excited state energies are possible. The names in the $dft section are b97-d
for the TURBOMOLE own version and b97d for the XCFun version. However, the total
energies of those two flavours are slightly different due to the fact that the parameters
used are either the originally published ones (TURBOMOLE) or re-computed (XCFun).
166CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

For properties like geometries and frequencies the differences are negligible, but one
should not mix the total energies.
The PBEh-3c functional needs, besides the functional name pbeh-3c also DFT-D3
dispersion correction including the three-body term and geometrical counterpoise
correction method called gCP. For details see: Stefan Grimme, University Bonn. In
order to get the full version of PBEh-3c, your control file has to include:

$dft
functional pbeh-3c
gridsize m4
$disp3 -bj -abc

Note: gcp is automatically added if pbeh-3c functional is used, but the D3 part
has to be switched on manually by adding $disp3 as given above. It is, however,
sufficient to use $disp3 -bj since the abc term is added automatically for PBEh-3c
and B97-3c.
To use HF-3c ( R. Sure, S. Grimme, J. Comput. Chem. 2013, 34, 1672–1685),
an input without DFT functional (and without $dft keyword) but with DFT-D3
correction is required. Calculations with and without RI are possible to perform,
but due to the very small basis set non-RI calculations are usually as efficient as
those with RI. Note that it is important to use the ’Minix’ basis set and to select
hf-3c as functional name for DFT-D3:

$disp3 -bj func hf-3c

The gCP correction (H. Kruse, S. Grimme, J. Chem. Phys. 2012, 136, 154101) will
by default be added to the DFT-D3 correction term if pbeh-3c or hf-3c is selected.
6.3. RESTRICTED OPEN-SHELL HARTREE–FOCK 167

6.3 Restricted Open-Shell Hartree–Fock

6.3.1 Brief Description

The spin-restricted open-shell Hartree–Fock method (ROHF) can always be chosen

to systems where all unpaired spins are parallel. The TURBOMOLE keywords for such
a case (one open shell, triplet e2g ) are:

$open shells type=1

eg 1 (1)
$roothaan 1
a=1 b=2

It can also treat more complicated open-shell cases, as indicated in the tables below.
In particular, it is possible to calculate the [xy]singlet case. As a guide for expert
users, complete ROHF TURBOMOLE input for O2 for various CSFs (configuration state
function) is given in Section 23.6. Further examples are collected below.
The ROHF ansatz for the energy expectation value has a term for interactions of
closed-shells with closed-shells (indices k, l), a term for purely open-shell interactions
(indices m, n) and a coupling term (k, m):
X X
E =2 hkk + (2Jkl − Kkl )
k k,l
X X X
+ f [2 hmm + f (2aJmn − bKmn ) + 2 (2Jkm − Kkm )]
m m,n k,m

where f is the (fractional) occupation number of the open-shell part (0 < f < 1),
and a and b are the Roothaan parameters, numerical constants which depend on the
particular configuration of interest.

6.3.2 One Open Shell

Given are term symbols (up to indices depending on actual case and group) and a
and b coefficients. n is the number of electrons in an irrep with degeneracy nir . Note
that not all cases are Roothaan cases.
All single electron cases are described by:
a=b=0
168CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

Table 6.1: Roothaan-coefficients a and b for cases with de-

generate orbitals.

nir =2: e (div. groups), π, δ (C∞v , D∞h )

n f en πn δn a b
3A 3Σ 3Σ 1 2
2 1/2 1 E∗ 1∆ 1Γ 1/2 0
1A 1Σ 1Σ 0 −2
3 3/4 2E 2Π 2∆ 8/9 8/9
1 nir =3: p (O(3)), t (T , O, I)†
n f pn a b
3P 3/4 3/2
2 1/3 1 D∗∗ 9/20 −3/10
1S 0 −3
4S 1 2
3 1/2 2 D∗∗ 4/5 4/5
2P 2/3 0
3P 15/16 9/8
4 2/3 1 D∗∗ 69/80 27/40
1S 3/4 0
5 5/6 2P 24/25 24/25
only irrep g(I)
(mainly high spin available)

n f gn a b
1 1/8 2G 0 0
2 1/4 †† 2/3 4/3
1A 0 −4
3 3/8 4G 8/9 16/9
4 1/2 5A 1 2
5 5/8 4G 24/25 32/25
6 3/4 †† 26/27 28/27
1A 8/9 4/9
7 7/8 2G 48/49 48/49
continues on next page
6.3. RESTRICTED OPEN-SHELL HARTREE–FOCK 169

Table 6.1: Roothaan-coefficients a and b for cases with de-

generate orbitals (continued).

d(O3), h(I)
(mainly high-spin cases work)

n f dn a b
1 1/10 2D 0 0
2 1/5 3 F+3 P†† 5/8 5/4
1S 0 −5
3 3/10 4 F+4 P†† 5/6 5/3
4 2/5 5 5
D, H 15/16 15/8
5 1/2 6 S, 6 A 1 2
6 3/5 5 D, 5 H 35/36 25/18
7 7/10 4 F+4 P†† 95/98 55/49
8 4/5 3 F+3 P†† 125/128 65/64
1S 15/16 5/8
9 9/10 2 D, 2 H 80/81 80/81
∗ except cases (e.g. D or D ) where e2 gives only one-dimensional
2d 4h
irreps, which are not Roothaan cases.
† only pn given, the state for groups T
d etc. follows from
S → A (T ,O,I) P → T (T ,O,I) D → H (I), E+T (T ,O)
∗∗ This is not a CSF in T or O, (a, b) describes average of states

resulting from E+T

†† (a, b) describes weighted average of high spin states, not a CSF.

Example

The 4d9 5s2 2 D state of Ag, in symmetry I

$closed shells
a 1-5 ( 2 )
t1 1-3 ( 2 )
h 1 ( 2 )
$open shells type=1
h 2 ( 9/5 )
$roothaan 1
a = 80/81 b = 80/81
170CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

6.3.3 More Than One Open Shell

A Half-filled shell and all spins parallel

All open shells are collected in a single open shell and

a=1 b=2

Example: The 4d5 5s1 7 S state of Mo, treated in symmetry I

$roothaan 1
a = 1 b = 2
$closed shells
a 1-4 ( 2 )
t1 1-3 ( 2 )
h 1 ( 2 )
$open shells type=1
a 5 ( 1 )
h 2 ( 1 )

Two-electron singlet coupling

The two MOs must have different symmetries (not required for triplet coupling, see
example 6.3.3). We have now two open shells and must specify three sets of (a, b),
i.e. one for each pair of shells, following the keyword $rohf.

Example: CH2 in the 1 B2 state from (3a1 )1 (1b2 )1 , molecule in (x,z) plane.

$closed shells
a1 1-2 ( 2 )
b1 1 ( 2 )
$open shells type=1
a1 3 ( 1 )
b2 1 ( 1 )
$roothaan 1
$rohf
3a1-3a1 a = 0 b = 0
1b2-1b2 a = 0 b = 0
3a1-1b2 a = 1 b = -2
6.3. RESTRICTED OPEN-SHELL HARTREE–FOCK 171

Two open shells

This becomes tricky in general and we give only the most important case:

shell 1 is a Roothaan case, see 6.3.2

shell 2 is one electron in an a (s) MO (nir = 1)

with parallel spin coupling of shells.

This covers e.g. the p5 s1 3 P states, or the d4 s1 6 D states of atoms. The coupling
information is given following the keyword $rohf. The (a, b) within a shell are taken
from above (6.3.2), the cross term (shell 1)–(shell 2) is in this case:

a =1 always
(2nir )
b =2 if n ≤ nir b= if n > nir
n
where nir and n refer to shell 1.

Example 1: The 4d4 5s1 6 D state of Nb, in symmetry I

$closed shells
a 1-4 ( 2 )
t1 1-3 ( 2 )
h 1 ( 2 )
$open shells type=1
a 5 ( 1 )
h 2 ( 4/5 )
$roothaan 1
$rohf
5a-5a a = 0 b = 0
5a-2h a = 1 b = 2
2h-2h a = 15/16 b = 15/8

Example 2: The 4d5 5s1 7 S state of Mo, symmetry I (see Section 6.3.3) can also
be done as follows.

$roothaan 1
$rohf
5a-5a a = 0 b = 0
5a-2h a = 1 b = 2
2h-2h a = 1 b = 2
172CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

$closed shells
a 1-4 ( 2 )
t1 1-3 ( 2 )
h 1 ( 2 )
$open shells type=1
a 5 ( 1 )
h 2 ( 1 )

The shells 5s and 4d have now been made inequivalent. Result is identical to 6.3.3
which is also more efficient.

Example 3: The 4d9 5s1 3 D state of Ni, symmetry I

$closed shells
a 1-3 ( 2 )
t1 1-2 ( 2 )
$open shells type=1
a 4 ( 1 )
h 1 ( 9/5 )
$roothaan 1
$rohf
4a-4a a = 0 b = 0
1h-1h a = 80/81 b = 80/81
4a-1h a =1 b = 10/9

(see basis set catalogue, basis SV.3D requires this input and gives the energy you
must get)

6.3.4 Miscellaneous

Valence states

Valence states are defined as the weighted average of all CSFs arising from an elec-
tronic configuration (occupation): (MO)n . This is identical to the average energy of
all Slater determinants.
2nir (n − 1)
a=b=
(2nir − 1)n
This covers, e.g. the cases n = 1 and n = 2nir − 1: p1 , p5 , d1 , d9 , etc, since there is
only a single CSF which is identical to the average of configurations.
6.3. RESTRICTED OPEN-SHELL HARTREE–FOCK 173

Totally symmetric singlets for 2 or (2nir -2) electrons

n=2 a=0 b = −nir

nir (nir − 2)
n = (2nir − 2) a=
(nir − 1)2
nir (nir − 3)
b=
(nir − 1)2

This covers the 1 S states of p2 , p4 , d2 , d8 , etc.

Average of high-spin states: n electrons in MO with

degenerate nir .

nir 4k(k + l − 1) + l(l − 1)
a=
(nir − 1)n2


2nir 2k(k + l − 1) + l(l − 1)
b=
(nir − 1)n2

where: k = max(0, n − nir ) , l = n − 2k = 2S (spin)

This covers most of the cases given above. A CSF results only if n = {1, (nir − 1),
nir , (nir + 1), (2nir − 1)} since there is a single high-spin CSF in these cases.
The last equations for a and b can be rewritten in many ways, the probably most
concise form is
n(n − 2) + 2S
a=
(n − 2f )n
n(n − 2) + (2S)2
b= .
(n − 2f )n

This applies to shells with one electron, one hole, the high-spin couplings of half-filled
shells and those with one electron more ore less. For d2 , d3 , d7 , and d8 it represents
the (weighted) average of high-spin cases: 3 F + 3 P for d2 ,d8 , 4 F + 4 P for d3 , d7 .
174CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

6.4 Relativistic effects

Turbomole provides two different possibilities for the treatment of relativistic ef-
fects: Via effective core potentials (ECPs) or via all-electron approaches (X2C, DKH,
BSS). Both techniques can be employed in a one-component (scalar-relativistic) or
two-component (including spin-orbit coupling) framework. The latter is only avail-
able in the modules ridft, rdgrad, escf, and ricc2.

6.4.1 One- and two-component relativistic methods

Incorporation of scalar-relativistic effects leads to additional contributions to the

one-electron integrals (either from ECP or all-electron approach). The program
structure is the same as in non-relativistic theory (all quantities are real). Two-
component treatments allow for self-consistent calculations including spin-orbit in-
teractions. These may be particularly important for compounds containing heavy
elements (additionally to scalar-relativistic effects). Two-component treatments re-
quire the use of complex two-component orbitals (spinors)
 α 
ψi (r)
ψi (x) =
ψiβ (r)
instead of real (non-complex) one-component orbitals in non-relativistic or scalar-
relativistic treatments. The Hartree-Fock and Kohn-Sham equations are now spinor
equations with a complex Fock operator
 αα  α   α 
F̂ F̂ αβ ψi (r) ψi (r)
β = i .
F̂ βα F̂ ββ ψi (r) ψiβ (r)
The wavefunction is no longer eigenfunction of the spin operator, the spin vector is
no longer an observable.
In case of DFT for open-shell systems, rotational invariance of the exchange-correlation
energy is ensured by the non-collinear approach. In this approach, the exchange-
correlation energy is a functional of the particle density and the absolute value of
the spin-vector density m(r)
~ σ are the Pauli matrices)
(~

ψi† (x)~
X
m(r)
~ = σ ψi (x).
i

This quantity replaces the spin-density (difference between density of alpha and beta
electrons) of non- or scalar-relativistic treatments.
For closed-shell species, the Kramers-restricted scheme, a generalization of the RHF-
scheme of one component treatments, is applicable.

Effective core potentials

The most economic way to account for relativistic effects is via effective core po-
tentials by choosing either the one- or the two-component ECP (and for the latter
6.4. RELATIVISTIC EFFECTS 175

additionally setting $soghf in the control file or in define). The theoretical back-
ground and the implementation for the two-component SCF procedure is described
in Ref. [122]. For recommendations concerning specific ECPs and corresponding
basis sets see below.

Relativistic all-electron approaches (X2C, DKH, BSS)

Relativistic calculations are based on the Dirac rather than on the Schrödinger Hamil-
tonian. Since the Dirac Hamiltonian introduces pathological negative-energy states
and requires extensive one-electron basis set expansions, methods have been de-
vised which allow one to calculate a matrix representation of that part of the Dirac
Hamiltonian, which describes electronic states only. For this, a unitary transforma-
tion is employed to block-diagonalize the Dirac Hamiltonian and thus to decouple
the negative-energy states from the electronic states. For reasons of efficiency, this
transformation is carried out only for the one-electron part of the full Hamiltonian
(as a consequence, the two-electron interaction will then be slightly affected by a
so-called picture-change effect). The resulting quantum chemical approach, “exact
two-component” (X2C), was developed by several groups starting with formal work
in the mid-1980s. X2C is related to the step-wise Douglas-Kroll-Hess (DKH) ap-
proach, which achieves decoupling in sequential manner. For the latter, the number
of transformation steps is called the order of DKH. Infinite-order DKH yields iden-
tical results compared to X2C, but – in contrast to the latter – not feasible. Eighth
order DKH usually yields results similar to X2C at similar cost. X2C is also re-
lated to the Barysz-Sadlej-Snijders (BSS) method, that first applies the free-particle
Foldy-Wouthuysen transformation (which is the first mandatory step in DKH), and
then constructs the one-step exact decoupling transformation of X2C. These three
approaches have been reviewed and directly compared in terms of formalism and
results, respectively, in Ref. [123] (see also this reference for a complete bibliography
on exact-decoupling methods).
Essentially, X2C methods change the one-electron Hamiltonian in a basis-set rep-
resentation. The Schrödinger one-electron Hamiltonian (including the external po-
tential of the atomic nuclei) is replaced by the transformed (upper-left block of the)
Dirac Hamiltonian. Since the transformation is carried out in the fully decontracted
primitive basis, all matrix operations needed for the generation of the relativistic
one-electron Hamiltonian can be cumbersome and even prohibitive if the molecule is
large. In order to solve this unfavorable scaling problem, a rigorous local approach,
called DLU, has been devised [124] and is recommended.
X2C, DKH, and BSS exist in full two-component (spin-(same-)orbit coupling includ-
ing) and in a one-component scalar-relativistic form. Both have been implemented
into the Turbomole package and all details on the efficient implementation have
been described in Ref. [125]. Additionally implemented modifications like a finite
nucleus model based on a Gaussian charge distribution [126] and a screened nuclear
spin-orbit (SNSO) approach [127–129] are documented in Ref. [130], together with
the implementation of analytical one- and two-component X2C gradients, also in
their local variant.
176CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

Calculation of analytical energy gradients is available within the (local) X2C ap-
proach. The implementation is described in Ref. [130]. The finite nucleus model
based on a Gaussian charge distribution can be used for energy and gradient calcu-
lations. To model the effect of spin-orbit coupling on the two-electron interaction
a (modified) screened-nuclear-spin-orbit approximation can be applied to the spin-
dependent one-electron integrals. The scalar-relativistic approach can be used with
the modules grad, rdgrad, egrad, ricc2, mpgrad, and rirpa. The two-component
Hamiltonian is only available in rdgrad.
In relativistic all-electron calculations additional contributions from point charges or
COSMO (see chapter 18.2) are not part of the relativistic decoupling. These contri-
butions are evaluated after the X2C, DKH, or BSS step. Thus, they are calculated
without picture-change transformation.

6.4.2 How to use

Scalar-relativistic calculations with ECPs

In case of scalar-relativistic calculations with ECPs, relativity is taken into account

simply by the choice of a relativistic ECP together with a suited basis set. A reason-
able choice are the Dirac-Hartree-Fock effective core potentials by the Stuttgart-Köln
group labelled dhf-ecp in TURBOMOLE, together with optimized basis sets termed dhf-
XVP (X = S, TZ, QZ) [131]. These ECPs and bases are available for Rb to Xe,
except for f-elements. For H to Kr dhf and def2 bases are identical non-relativistic
all-electron basis sets. For lanthanides and actinides at present no dhf ECPs are avail-
able. The usage of Wood-Boring type ECPs [132], for lanthanides together with def2-
bases [133], for actinides together with the original Stuttgart-Köln bases [134, 135]
(labelled def in TURBOMOLE) is recommended here. Note that for def2-bases the uner-
lying ECPs are of different type (due to history). For p-elements, def2-bases are
optimized for Dirac-Hartree-Fock ECPs, for the other elements for Wood-Boring
ECPs.

Two-component calculations (general)

The keyword $soghf enforces the two-component calculations. Keywords for spec-
ification of the method of calculation are the same as for the one-component case.
Additionally, for closed-shell species a Kramers invariant density functional formal-
ism can be switched on with the keyword $kramers. The DIIS scheme for complex
Fock operators (GDIIS) can be activated by $gdiis. For improvements on the SCF
convergence behavior and gradients, please see Ref. [136]. It is recommended to use
exact exchange for double- and triple-zeta bases instead of semi-numerical [111] or
RI-K approximations. These keywords can be inserted into the control file manually
or added in the scf section of define.
As start wavefunctions Hückel or SCF wavefunctions may be used. For open-shell
molecules it is often helpful to increase the value for $scforbitalshift closedshell;
6.4. RELATIVISTIC EFFECTS 177

a value of ca. 1.0 may serve as a rough recommendation. Two-component calcula-

tions generally employ the RI-J approximation. To use the exact Coulomb integrals
within the ridft and rdgrad module, the $coulex flag can be set.

Two-component calculations with ECPs

For spin-orbit treatments, the two-component variants of ECPs (suffix -2c) are re-
quired, the use of extended basis sets accounting for the spatial splitting of inner
p-shells (also suffix -2c) is recommended. ECPs dhf-ecp-2c and bases dhf-XVP-2c
(X = S, TZ, QZ) [131] are available for Rb to Kr (f elements excepted); for refer-
ences concerning ECPs see http://www.tc.uni-koeln.de/PP/clickpse.en.html.
The two-component formalism may be most easily prepared and applied in the fol-
lowing way:

• Run define: choose C1 -symmetry; select ECPs and basis sets with suffixes
-2c for the respective elements. RI-J and RI-JK auxiliary basis sets are the
same for dhf and def2 bases. Moreover, they are of sufficient flexibility for
two-component treatments and provided automatically upon request. Switch
on soghf and further desired options in the scf menu.

• Start the two-component calculation with ridft.

• At the end of the SCF procedure real and imaginary parts of spinors are written
to files spinor.r and spinor.i, eigenvalues and spinor occupations are collected in
the file EIGS, the total energy is added to data group $energy. The data groups
$closed shells ($alpha shells and $beta shells for open shell cases) are
no longer significant, but nevertheless kept in the control file; additionally the
spinor occupations are deposited in data group $spinor.

One- and two-component all-electron calculations

All-electron calculations are prepared similarly to ECP calculations, however, rela-

tivistic all-electron basis sets are necessary. It is recommended to use the x2c-type
basis sets and RI-J auxiliary basis sets x2c-XVPall for one-component and x2c-
XVPall-2c for two-component treatments [137–139]. Presently, they are available for
X = S, TZ, QZ for H to Rn. The keywords $rx2c, $rbss and $rdkh n (where n
stands for the order of DKH) are used to activate the X2C, BSS or DKH Hamil-
tonian. n defaults to 4. It is not recommended to go beyond, but to use X2C
instead. For details on the arbitrary-order DKH Hamiltonians, see Ref. [140] for
details on the infinite-order DKH theory, [141] for the implementation, and [142] for
a conceptual review of DKH theory. The local approach (DLU) can be optionally
activated by $rlocal for all one- and two-component all-electron Hamiltonians. For
symmetric molecules, point-group symmetry is not exploited by default, but can be
used in the one-component case by setting $rsym. Without setting $rsym, symmetry
is only exploited in the rest of the program. Note that $rsym cannot be combined
178CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

with $rlocal. The exploitation of symmetry is also supported in X2C gradient cal-
culations [41]. Therein, nuclear symmetry is exploited without setting $rsym. The
finite nucleus model based on a Gaussian charge distribution is selected by $finnuc.
Two-component calculations are invoked by the keyword $soghf as outlined above.
The SNSO approach may be used to estimate the effect of spin-orbit coupling on
the two-electron interaction by using the keyword $snso. By default, the modified
parameters [128,129] are taken, however, the default parameters [127] can be applied
by setting $snsopara to 0. The latter set of parameters was derived for lower-order
DKH whereas the modified parameters were optimized for X2C and greatly improve
the spinor energies for virtual states. All options can be specified in the scf section of
define. The SNSO, GDIIS and scforbitalshift parameters are set in the soghf part
of the scf section. It is further recommended to use grids with an increased number
of radial points [138] in DFT calculations. These grids are selected by appending
“a” to the gridsize. A picture-change correction for several expectation values in the
proper section is available for relativistic all-electron Hamiltonians ($rdkh, $rbss,
$rx2c, also in their local variant $rlocal)and enforced by $pcc. In NMR shielding
calculations, the use of tailored basis sets, i.e. x2c-SVPall-s, x2c-TZVPall-s [138] and
x2c-QZVPall-s [139] etc., is recommended.
6.5. FINITE MAGNETIC FIELDS 179

6.5 Finite magnetic fields

Calculations of magnetic properties in finite magnetic fields can be carried out in

the 2c framework (see section 6.4). The Hamiltonian in a magnetic field (B) can be
written as:

1X X 1X 2 2
B riO − (B · riO )2


Ĥ = Ĥ0 + B · liO + B · si +
2 8
i i i

Here, Ĥ0 is the zero-field Hamiltonian. The terms dependent on the angular mo-
mentum operator liO and spin operator si are called orbital-Zeeman term and spin-
Zeeman term, respectively. Both are referred to as paramagnetic terms since they
are linearly dependent on the magnetic field. The terms which are quadratically
dependent on the magnetic field are called diamagnetic terms. In order to avoid a
dependence on the gauge origin O of the system, London orbitals can be used.
In Order to perform a Hartree-Fock calculation on molecules in finite magnetic fields,
the 2c framework of the ridft module has to be used by using the $soghf and
$magnetic field keywords. It is necessary to specify the x, y and z coordinates of
the magnetic field as well as the absolute value in atomic units. In the next lines it
is possible to include the word ’Tesla’. If this word is present, the absolute value of
the magnetic field is assumed to be in Tesla and will be converted to atomic units.
Adding ’nospin’ to one of the lines neglects the spin-Zeeman term. An exemplary
input into the control file might look like this:

$soghf
$magnetic field
1 1 2 16
Tesla
nospin

This would refer to a Hartree-Fock calculation with a magnetic field of 16 T applied

in the direction (1|1|2) where the spin-Zeeman term is neglected. Using spin-Zeeman
scaling, the spin-Zeeman term can be multiplied by any number in all iterations but
the last one. This can help to calculate spin states which are currently not the
ground state. Adding ’spin 6’ will scale the spin-Zeeman term by a factor of 6 for
all iterations but the last one. Using negative numbers is also possible, ’spin -3.85’
will therefore scale the spin-Zeeman by a factor of -3.85 which might for instance be
favorable for singlet states.
The calculation can be run using the ridft module. If the additional keyword
$coulex is added, ridft will perform a Hartree-Fock calculation with exact Coulomb
and exchange parts. Specifying auxiliary basis sets and adding the keywords $rij
and/or $rik and then running ridft will perform calculations using RI-J and/or
RI-K. This implementation currently only works with Hartree-Fock, not with DFT.
180CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

6.6 Dispersion Correction for DFT Calculations

Grimme DFT+Dispersion, DFT-D3, DFT-D4

Based on an idea that has earlier been proposed for Hartree-Fock calculations [143,
144], a general empirical dispersion correction has been proposed by Stefan Grimme
for density functional calculations [145]. A modified version of the approach with
extension to more elements and more functionals has been published in ref. [146].
A more recent implementation [147] is less empirical, i.e. the most important pa-
rameters are computed by first principles, and it provides a consistent description
across the whole periodic system. This version, named DFT-D3 with Becke-Johnson
damping, has been the default for dispersion correction ever since.
In 2017 a new model, termed D4, was published [148, 149] and released 2018. For
details see below.

• The first version (DFT-D1) can be invoked by the keyword $olddisp in the
control file.

• The second version (DFT-D2) is used if the keyword $disp is found.

• For the usage of DFT-D3 just add keyword $disp3 to the control file. For
DFT-D3 with Becke-Johnson damping, please add $disp3 -bj.

• The latest version, DFT-D4, can be invoked by the keyword $disp4.

Only one of the four keywords is expected to be present.

DFT-D3

If DFT-D3 is used, the total energy is given by

EDF T −D3 = EKS−DF T − Edisp (6.8)

where EKS−DF T is the usual self-consistent Kohn-Sham energy as obtained from the
chosen functional and Edisp is a dispersion correction given by the sum of two- and
three-body energies
Edisp = E (2) + E (3) , (6.9)
with the dominating two-body term
X X CnAB
E (2) = sn n fd,n (rAB ). (6.10)
rAB
AB n=6,8,10,...

The first sum runs over all atom pair, CnAB denotes the nth-order dispersion coef-
ficient for atom pair AB, rAB is their interatomic distance, and fd,n is a damping
function.
6.6. DISPERSION CORRECTION FOR DFT CALCULATIONS 181

Becke-Johnson (BJ) damping can be invoked by adding the option bj or -bj to the
$disp3 keyword: $disp bj If you use this damping option please also cite [150].
The three-body term can be switched on by adding abc to the $disp3 input line,
i.e. to use it in combination with Becke-Johnson damping just add $disp3 bj abc
It is also possible not to use the functional name given in the control file but to
tell the DFT-D3 routines to use the parameters which have been fitted to a specific
functional. Just as in the original DFT-D3 routines, this can be selected by adding
the func option, for example $disp3 bj func pbe0. It is recommended to use this
option as the last one in the $disp3 input line.

Please have look at DFT-D3 homepage, Grimme group Bonn for more detailed infor-
mation.

DFT-D4

DFT-D4 computes molecular dipole-dipole and dipole-quadrupole dispersion coef-

ficients based on dynamic dipole polarizabilities calculated with TD-PBE38/daug-
def2-QZVP used as atomic reference polarizabilities for elements up to radon (Z=86).
In DFT-D4, all atomic reference polarizabilities are scaled according to Mulliken par-
tial charges which are obtained by semi-empirical quantum mechanical tight binding,
within the 2018 developed GFN2-xTB framework [67]. Alternatively a fallback op-
tion is provided in case GFN2-xTB fails to converge the electronic structure. Either
classical, non-interative EN and CN dependent Gasteiger charges (option ’gasteiger’)
or (externally provided) Hirshfeld-charges optained at least at PBE0/def2-TZVP
level of theory can be used (option ’hirshfeld’).
For the scaling of the atomic reference polarizabilities, a special charge-function was
designed containing one global parameter and the chemical hardnesses as elementspe-
cific parameter to get a smooth scaling behavior. The charge-scaled dipole polariz-
abilities are Gaussian interpolated according to an empirically generated fractional
coordination number to obtain the fully weighted dynamic polarizabilities α(iω) for
all atoms which are then numerically integrated via the Casimir-Polder scheme to
obtain charge- and geometry-dependent dipole-dipole dispersion coefficients. The
two-body energy expression has the usual sum over pair interactions form for dipole-
dipole and dipole-quadrupole interactions.
Based on this expression, a self-consistent dispersion potential has been developed
and implemented into the GFN2-xTB Hamiltonian to circumvent a costly coupled-
perturbed SCF procedure when calculating analytical gradients. Furthermore, dy-
namic polarizabilities α(iω) are used within an RPA-like expression to capture many-
body interactions beyond the two-body terms.
The DFT-D4 energy expression differs for single point energies and gradient calcula-
tions, for reasons of computational efficiency. The gradient of the RPA-like dispersion
energy scales as O(N 4 ) and cannot be screened, therefore the gradient is approxi-
mated by the ATM three-body dispersion expression as used in DFT-D3, which is
computational less demanding and yields approximately similar gradients.
182CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

Parameters for the rational damping scheme introduced by Becke and Johnson are
included for a large number of density functionals (including all standard functionals
implemented in Turbomole) in the DFT-D4 code. The rational damping is the
only scheme available in DFT-D4, also many-body dispersion terms are included by
default.
The DFT-D4 keyword $disp4 can have, similar to $disp3, a couple of options. To
see the list of options, just add -help to the keyword and inspect the output file.
As mentioned above, the gradient of the many body dispersion energy is done by us-
ing the Axilrod-Teller-Muto (ATM) three-body term, which is only an approximation
compared to the term used in the energy evaluation. Hence, for larger molecules,
the DFT-D4 energy and the DFT-D4 gradient do not match exactly. Very tight
convergence criteria for both energy and gradient norm can not be expected in all
cases.

Density-based dispersion corrections of non-local vdW-DF

type

A non-local, electron density dependent dispersion correction which is based on Vy-

drov and Van Voorhis’ VV10 [151] has been implemented by the Grimme group [152]
and is available for ridft and rdgrad. This correction can either be applied in a post-
SCF and non-self-consistent way for energy calculations or self-consistently which is
required to compute the gradients.
To switch on DFT-NL in a non-self-consistent way, just add
$donl
to the control file. For a self-consistent treatment of the dispersion correction add
$doscnl
instead. Note that dispersion corrections of DFT-DN and NL–DFT type must not
be combined. The grid size for the non-local integration is set automatically by
adapting the grid for the quadrature of the functional evaluation.
Currently only C1 symmetry and serial jobs are possible. DFT-NL is an interesting
scientific alternative to DFT-D3, but we recommend to use DFT-D3 for applications
instead.
6.7. ENERGY DECOMPOSITION ANALYSIS (EDA) 183

6.7 Energy Decomposition Analysis (EDA)

The interaction energy between molecules can be calculated with the supermolecu-
lar approach: one performs calculations for the supersystem and for the subsystems
with size-consistent methods and derive the interaction energy ∆E by taking the
energy difference. The energy decomposition analysis (EDA) allow a partitioning of
the Hartree-Fock (HF) or DFT interaction energy in physically meaningful contribu-
tions: the classical electrostatic interaction ∆Eele , the exchange-repulsion ∆Eexrep ,
the orbital relaxation energy ∆Eorb and additionally for DFT the correlation inter-
action ∆Ecor :

∆EHF = ∆Eele + ∆Eexrep + ∆Eorb , (6.11)

∆EDFT = ∆Eele + ∆Eexrep + ∆Eorb + ∆Ecor . (6.12)

Further details and derivations of the different energy contributions can be found
in [153].

6.7.1 How to perform

The EDA scheme is implemented in the module ridft and can be done with RI-
Hartree-Fock and with all local, gradient corrected, hybrid and meta density func-
tionals (please note that the functionals included in the XCFun library are not sup-
ported!).

• Calculation of the subsystems:

In HF and hybrid DFT calculations please insert $scfdenapproxl 0 in the
control file, at least in a second run. The EDA scheme needs the exchange and
Coulomb energies of every system separately. After the subsystem calculation
you will find under $subenergy the different energy contribution to the total
energy of the system: the one electron energy, Coulomb- and exchange energy,
correlation energy in case of DFT calculations, nuclear repulsion energy and
optionally the dispersion energy.

• Preparation of the supersystem control file:

First run define for the supersystem and take for consistency the same basis
set and the same method (i.e. the same functional and the same grid). Please
use in case of DFT calculations not the multiple grids m3 to m5, because this
would lead to erroneous orbital relaxation energies. If the subsystems are open-
shell species the occupation in the EHT submenue of define of the supersystem
must be chosen open-shell, too. For open-shell systems the Fermi-smearing is
recommended. The sequence of the supersystem coordinates must have the
same sequence as the subsystem coordinates. In the case of HF and hybrid
DFT calculation use again $scfdenapproxl 0.
Then please insert in the control file :
184CHAPTER 6. HARTREE–FOCK AND DFT CALCULATIONS FOR MOLECULAR SYSTEMS

$subsystems
molecule#1 file=sub1/control
molecule#2 file=sub2/control

If you use the supermolecular basis set for the calculation of the monomers
please insert after $subsystems the option copo:

$subsystems copo
molecule#1 file=sub1/control
molecule#2 file=sub2/control

It is possible to generate orthogonal product wave functions when you use opro
instead of copo. But with this choice it is not possible to calculate the different
energy contributions of the interaction energy.
You can choose at most ten subsystems.

• Generation of the product molecular wave functions:

The module promowa generates RHF and UHF product molecular wave func-
tions. The new (product) start vectors can be found in the files mos for closed-
shell systems or in alpha and beta for open-shell systems. Please note that
the molecular orbitals of the different subsystems are not orthogonal to each
other.

• Energy decomposition analysis:

After the supersystem ridft calculation you will find the following output with
the different contributions of the interaction energy in Hartree:

----------------------------------------------------
| * Total Interaction energy = -0.0058700698 |
----------------------------------------------------
: * Electrostatic Interaction = -0.0134898233 :
: Nuc---Nuc = 18.2843363825 :
: 1-electron = -36.5372802833 :
: 2-electron = 18.2394540775 :
: * Exchange-Repulsion = 0.0112325934 :
: Exchange Int. = -0.0139477002 :
: Repulsion = 0.0251802936 :
: * Orbital Relaxation = -0.0036128399 :
.....................................................
Chapter 7

DFT Calculations for Molecular

and Periodic Systems

7.1 Functionalities of Riper

The riper module is an implementation of Kohn–Sham DFT with Gaussian-type orbitals

(GTO) as basis functions that treats molecular and periodic systems of any dimension-
ality on an equal footing. Its key component is a combination of resolution of identity
(RI) approximation and continuous fast multipole method (CFMM) applied for the elec-
tronic Coulomb term [154, 155]. This RI-CFMM scheme operates entirely in the direct
space and partitions Coulomb interactions into far-field part evaluated using multipole ex-
pansions and near-field contribution calculated employing density fitting. Evaluation of the
exchange-correlation term is performed using an octree-based adaptive numerical integration
scheme [156]. riper offers computational efficiency and favorable scaling behavior approach-
ing O(N ) for the formation of Kohn–Sham matrix [154] and gradient calculation [155]. In
addition, for calculations on very large molecular systems a low-memory modification of the
RI approximation has been implemented [157]. This low-memory iterative density fitting
(LMIDF) scheme is based on a combination of CFMM and a preconditioned conjugate gra-
dient (CG) solver. Compared with the standard RI implementation, up to 15-fold reduction
of the memory requirements is achieved at a cost of only slight increase in computational
time.
Functionalities of riper:

• Kohn–Sham DFT for molecular systems and systems with 1D, 2D, and 3D periodicity

• closed– and open–shell energies and gradients, structure optimization including opti-
mization of cell parameters

• metals and semiconductors can be treated using fractional occupation numbers scheme
employing Gaussian smearing

185
186CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

• efficient k point sampling scheme for periodic systems allowing for consistent results
across different definitions of unit cells

• sequential and parallel runs (OpenMP parallelization for shared-memory computers,

see Sec. 3.4.2)

• all LDA, GGA and meta-GGA exchange-correlation functionals including interface to

the XCFun library (see Sec. 6.2)

• DFT-D3 dispersion correction for energies and gradients (see Sec. 6.6)

• favorable scaling behavior for the formation of the Kohn–Sham matrix approaching
O(N )

• memory-efficient calculations for very large molecular systems using the LMIDF scheme

• real time-time dependent DFT (RT-TDDFT) for molecular systems

Limitations of riper:

• only C1 symmetry point group for molecules and P1 space group for periodic systems

7.2 Theoretical Background

Detailed description of methods implemented in the riper module is provided in Refs [154–
159] and references therein. Here, only a short summary of the underlying theory is provided.

7.2.1 Kohn-Sham DFT for Molecular and Periodic Systems

k
In periodic systems translational symmetry of solids leads to Bloch orbitals ψpσ and one-
k
particle energies εpσ depending on the band index p, spin σ, and the wave vector k within
the Brillouin zone (BZ), which is the unit cell of reciprocal space. The orbitals

k 1 X ikT L X k
ψpσ (r) = √ e Cµpσ µL (r) (7.1)
NUC L µ

are expanded in GTO basis functions µ(r − Rµ − L) ≡ µL (r) centered at atomic positions
Rµ in direct lattice cells L over all NUC unit cells. This results in unrestricted Kohn-Sham
equations
Fkσ Ckσ = Sk Ckσ εkσ , (7.2)

which may be solved separately for each k in the BZ. The same equations hold for the
molecular case, where only L = k = 0 is a valid choice and NUC is one. Equation (7.2)
contains the reciprocal space Kohn-Sham and the overlap matrices Fkσ and Sk , respectively,
obtained as Fourier transforms of real space matrices
X T X T
k
Fµνσ = eik L Fµνσ
L k
Sµν = eik L Sµν
L
. (7.3)
L L
7.2. THEORETICAL BACKGROUND 187

L L
The elements Fµνσ contain three contributions: elements Tµν of the kinetic energy matrix,
L L
elements Jµν of the Coulomb matrix, and elements Xµνσ of the exchange-correlation matrix,

L L L L
Fµνσ = Tµν + Jµν + Xµνσ . (7.4)

The total energy per unit cell E is calculated as the sum of the kinetic T , Coulomb J, and
exchange-correlation EXC contributions,

E = T + J + EXC . (7.5)

7.2.2 RI-CFMM Approach

The key component of riper is a combination of RI approximation and CFMM applied

for the electronic Coulomb term [154, 155, 158]. In the RI scheme the total crystal electron
density ρcryst is approximated by an auxiliary crystal electron density ρ̃cryst
X
ρcryst ≈ ρ̃cryst = ρ̃L , (7.6)
L

composed of unit cell auxiliary densities ρ̃L with

X
ρ̃L = cT α L , (7.7)
α

where αL denotes the vector of auxiliary basis functions translated by a direct lattice vector
L. The vector of expansion coefficients c is determined by minimizing the Coulomb repulsion
D of the residual density δρ = ρ − ρ̃
ZZ
1 X X X
D= δρ (r) 0
δρL (r0 ) dr dr0 = (δρ | δρL ) = (ρ − ρ̃ | ρL − ρ̃L ) . (7.8)
|r−r |
L L L

The RI approximation allows to replace four-center electron repulsion integrals (ERIs) by

L
two- and three-center ones. In this formalism, elements Jµν of the Coulomb matrix are
defined as X
L
Jµν = (µνL | ρ̃L0 − ρnL0 ) , (7.9)
L0

where ρn denotes the unit cell nuclear charge distribution. The total Coulomb energy in-
cluding the nuclear contribution is
X
L L 1X
J= Dµν Jµν − (ρ̃ + ρn | ρ̃L − ρnL ) , (7.10)
2
µνL L

with the real space density matrix elements obtained by integration

Z
L 1 k T
Dµνσ = Dµνσ eik L dk, (7.11)
Vk BZ
of the reciprocal space density matrix

k
X
k k

∗ k
Dµνσ = fpσ Cµpσ Cνpσ (7.12)
p
188CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

over the BZ with volume Vk .

Equations (7.9) and (7.10) as well as other expressions appearing in the RI scheme require
calculation of infinite lattice sums of the form
X
(ρ1 | ρ2L ) , (7.13)
L

where the distribution ρ1 in the central cell interacts with an infinite number of distributions
ρ2L , i.e., ρ2 translated by all possible direct lattice vectors L. In the RI-CFMM scheme
[154, 155] the sum in Eq. (7.13) is partitioned into crystal far-field (CFF) and crystal near-
field (CNF) parts. The CFF part contains summation over all direct space lattice vectors
L for which the overlap between the distributions ρ1 and ρ2L is negligible. This part is
very efficiently calculated using multipole expansions. The CNF contribution is evaluated
using an octree based algorithm. In short, a cubic parent box enclosing all distribution
centers of ρ1 and ρ2 is constructed that is large enough to yield a predefined number ntarg
of distribution centers per lowest level box. The parent box is successively subdivided in
half along all Cartesian axes yielding the octree. In the next step, all charge distributions
comprising ρ1 and ρ2 are sorted into boxes based on their extents. Interactions between
charges from well-separated boxes are calculated using a hierarchy of multipole expansions.
Two boxes are considered well-separated if the distance between their centers is greater
than sum of their lengths times 0.5×wsicl, where wsicl is a predefined parameter ≥ 2.
The remaining contribution to the Coulomb term is obtained from direct integration. This
approach results in nearly linear scaling of the computational effort with the system size.

7.2.3 k Point Sampling Scheme

The integral in Eq. (7.11) is evaluated approximately using a set of sampling points k. riper
uses a Γ-point centered mesh of k points with weights wk , so that Eq. (7.11) can be written
as X T
L
Dµνσ ≈ wk eik L Dµνσ
k
. (7.14)
k

In 3D periodic systems each sampling point is defined by its components k1 , k2 and k3 along
the reciprocal lattice vectors b1 , b2 and b3 as

k = k1 b1 + k2 b2 + k3 b3 . (7.15)

For 2D periodic systems k3 = 0. In case of 1D periodicity k3 = 0 and k2 = 0. In riper the

three components kj (j = 1, 2, 3) of k are given as

i nj − 1 nj − 1 nj − 1 nj − 1
kj = with i = − ,− + 1, . . . , − 1, . (7.16)
nj 2 2 2 2

with nj (j = 1, 2, 3) as integer numbers. riper reduces the number of k points employed

in actual calculation by a factor of two using time-inversion symmetry, i.e., the vectors k
and −k are symmetry equivalent. The k point mesh can be specified providing the integer
values nj within the data group $kpoints.
7.2. THEORETICAL BACKGROUND 189

The number of k points required in a calculation critically depends on required accuracy.

Generally, metallic systems require considerably more k points than insulators to reach
the same precision. For metals, the number of k point also depends on parameters of the
Gaussian smearing [160] used in riper. Please refer to Ref. [160] for more details.

7.2.4 Metals and Semiconductors: Gaussian Smearing

Achieving reasonable accuracy of DFT calculations for metals requires a higher number of k
points than for semiconductors and insulators. The convergence with respect to the number
of k points can be improved applying partial occupancies [160]. To achieve this, riper uses
the Gaussian smearing method in which occupation numbers fnk are calculated as
    
nk − µ 1 nk − µ
fnk = 1 − erf , (7.17)
σ 2 σ

where nk are band (orbital) energies, µ is the Fermi energy and σ is the width of the
smearing.
When smearing is applied the total energy E has to be replaced a generalized free energy F
X
F =E− σS(fnk ) (7.18)
nk

in order to obtain a variational functional. riper output file reports the values of F as
P
“FREE ENERGY” and the term − nk σS(fnk ) as “T*S”. In addition, the value of E for
σ → 0 is given as “ENERGY (sigma->0)”.
Gaussian smearing can be switched on for riper calculations by simply providing the value
of σ > 0 within the $riper data group using the keyword sigma, e.g., sigma 0.01. The
value of σ should be as large as possible, but small enough to yield negligible value of the
“T*S” term. Note, that the value of sigma has to be provided in atomic units. Please refer
to Ref. [160] for a more detailed discussion.
The use of Gaussian smearing often requires much higher damping and orbital shifting.
Please adjust the values for $scfdamp and $scforbitalshift if you encounter SCF conver-
gence problems.
The optional keyword desnue can be used within the $riper data group to constrain the
number of unpaired electrons. This can be used to force a certain multiplicity in case of an
unrestricted calculation, e.g., desnue 0 for singlet and desnue 1 for doublet.

7.2.5 Low-Memory Iterative Density Fitting Method

For calculations on very large molecular systems a low-memory modification of the RI ap-
proximation has been implemented within the riper module [157]. In the RI approximation
minimization of the Coulomb repulsion of the residual density, Eq. (7.8), yields a system of
linear equations
Vc = γ, (7.19)
190CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

where V is the Coulomb metric matrix with elements Vαβ = (α | β) representing Coulomb
interaction between auxiliary basis functions and vector γ is defined as
X
γα = (α | µν) Dµν . (7.20)
µν

In the LMIDF approach a conjugate gradient (CG) iterative method is used for solution of
Eq. (7.19). In order to decrease the number of CG iterations a preconditioning is employed,
i.e., Eq. (7.19) is transformed using a preconditioner P to an equivalent problem

P−1 V c = P−1 γ


(7.21)

with an improved condition number resulting in faster convergence of the CG method. The
iterative CG solver in riper employs one of the following preconditioners that are formed
from blocks of the V matrix corresponding to the strongest and most important interactions
between the auxiliary basis functions such that P−1 V ≈ I:

• atomic block preconditioner


 αI | β I
, I ∈ A , A are all atoms in molecule
at I I
Pαβ =
0, otherwise

at ss
• ss block preconditioner: Pαβ ∪ Pαβ

(α | β) , α, β ∈ {S}, S are all s auxiliary basis functions
ss
Pαβ =
0, otherwise

at sp
• sp block preconditioner: Pαβ ∪ Pαβ

(α | β) , α, β ∈ {S, P }, P are all p auxiliary basis functions
sp
Pαβ =
0, otherwise

The costly matrix-vector products of the Vc type that need to be evaluated in each CG
iteration are not calculated directly. Instead, the linear scaling CFMM implementation
presented above is applied to carry out this multiplication since the elements of the Vc
vector represent Coulomb interaction between auxiliary basis functions α and an auxiliary
density ρ̃  
X X
(Vc)α = (α | β) cβ = α | cβ β  = (α | ρ̃) . (7.22)
β β

Hence, in contrast to conventional RI neither the V matrix nor its Cholesky factors need to
be stored and thus significant memory savings are achieved.

7.2.6 RT-TDDFT

To investigate the electron dynamics in real time, RT-TDDFT based on Magnus propagator
is implemented in riper module [159]. In RT-TDDFT, the time evolution of electron density
7.2. THEORETICAL BACKGROUND 191

ρ(r, t), represented by the single particle reduced density matrix D(t) with elements
N
X MO
†
Dµν (t) = fm Cµm (t)Cνm (t) (7.23)
m=1

is governed by the von Neumann equation

∂D(t)
i = [F(t), D(t)] (7.24)
∂t
where F(t) is the time-dependent KS matrix in the orthonormal basis of MO. The von
Neumann equation (7.24) is efficiently integrated numerically using the Magnus expansion
which evolves the density matrix in time using a unitary operator U(t+∆t, t) = eΩ1 +Ω2 +Ω3 +···
that conserves the idempotency of D(t). Second and fourth order Magnus expansions are
implemented.
External perturbation is provided in the form of an electric field E which is assumed to be
uniform over the whole molecule. The elctric field contribution to the KS matrix can be
written as X
FEµν = −
j
Mµν Ej , j = x, y, z (7.25)
j=x,y,z

with the electric field vector (Ex , Ey , Ez ) and the dipole moment matrices Mj
Z
j
Mµν = − µ(r)jν(r)dr (7.26)

Two time integration methods have been implemented, the self-consistent field (SCF) pro-
cedure and the predictor-corrector (PC) scheme.
In the SCF procedure, starting from the ground state electron density D(0) and KS F(0)
matrices, a guess for F(t + ∆t
2 ) is made through linearly extrapolation. Next, D(t) is propa-
gated and used to calculate F(t + ∆t), which is followed by a linear interpolation for a better
guess for F(t + ∆t
2 ) until convergence is achieved.
In predictor-corrector scheme, F(t + ∆t/4) is predicted by linear extrapolation from previ-
ous values and this is used to step D forward by ∆t/2. In contrast to the SCF procedure,
which requires multiple KS matrix builds per time step, the PC scheme requires no new KS
builds and is comparatively cheaper. However, it is prone to unstability for longer time steps.

Absorption spectra is calculated using the following expression for the dipole strength func-
tion
1 4πω
S(ω) = · Tr (lm [αij ]) , i, j = x, y, z (7.27)
3 c
where αij is the complex polarizability tensor, given as
R ∞ iωt ind
e µj (t)e−γt dt
αij (ω) = −∞ R∞ (7.28)
−∞
eiωt Ei (t)dt

where γ is the damping factor (typically in the range of 0.003 − 0.005au = 124 − 207ps−1 ),
µind
j (t) is the time-dependent induced dipole moment and Ei is the electric field component
along i direction.
192CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

7.3 How to Perform a Calculation

7.3.1 Basis Sets for Periodic Calculations

In periodic systems too diffuse basis functions result in near-singularity of the overlap matrix.
Therefore, it is recommended that the exponents of the Gaussian basis sets used for periodic
calculations are not smaller than 0.01. Optimally, the exponents should be larger than 0.1.
The easiest way to achieve this is to remove the small exponent basis functions from the
basis sets. Optionally, basis sets optimized for periodic calculations can be used, such as the
ones in Ref. [161]. They are available in the TURBOMOLE basis sets library as pob-TZVP.

7.3.2 Prerequisites

Calculations using riper require the control file and starting orbitals generated using
define. DFT method needs to be specified in the $dft data group. All LDA, GGA and
meta-GGA exchange-correlation functionals including interface to the XCFun library (see
Sec. 6.2) are supported. Moreover, auxiliary basis sets defined in the data group $jbas are
required. For periodic calculations additional keywords specifying the system periodicity
and number of k points (if used) have to be added manually to the control file. Option-
ally, the $riper group containing control options specific to riper (including the LMIDF
keywords) can be added. The input preparation steps are summarized below. More detailed
information about riper keywords are provided in Sec. 22.2.12.

7.3.3 Creating the Input File

• Run define: in the geometry menu choose C1 symmetry. Create data groups $dft
and $jbas using dft and ri, respectively, in the general menu.

• riper performs molecular (0D) calculation by default if no other options are specified.
For periodic calculations (1D, 2D and 3D periodicity) specify system dimensionality
using the keyword $periodic n (n = 1, 2, 3) and unit cell parameters or lattice vectors
using the keyword $cell or $lattice, respectively. By default, the parameters have
to be provided in atomic units and degrees. Alternatively, Å can be employed using
the keyword $cell angs or $lattice angs.
Specification of cell parameters using the $cell keyword depends on the periodicity
of a system:

– For 3D periodic systems six unit cell parameters |a|, |b|, |c|, α, β and γ need to
be defined. Here, |a|, |b| and |c| are lengths of the appropriate cell vectors, α is
the angle between vectors b and c, β is the angle between vectors a and c, and
γ is the angle between vectors a and b. riper assumes that the cell vectors a
and b are aligned along the x axis and on the xy plane, respectively.
– For 2D periodic systems three surface cell parameters |a|, |b| and γ have to be
provided. Here, |a| and |b| are lengths of the appropriate cell vectors and γ is
7.3. HOW TO PERFORM A CALCULATION 193

the angle between a and b. riper assumes that the cell vectors a and b are
aligned along the x axis and on the xy plane, respectively.
– For 1D periodic systems only one parameter specifying the length of the unit
cell has to be provided. riper assumes that periodic direction is along the x
axis.

Alternatively, lattice vectors can be provided using the $lattice keyword:

– For 3D periodic systems three (three-dimensional) lattice vectors have to be

specified.
– For 2D periodic systems two (two-dimensional) lattice vectors have to be pro-
vided. riper assumes that the lattice vectors are aligned on the xy plane.
– For 1D periodic systems only one parameter specifying the length of the lattice
vector has to be provided. riper assumes that periodic direction is along the x
axis.

Example 1: 1D periodic system with a unit cell |a| = 5 bohr:

$periodic 1
$cell
5.0

The same input using the $lattice keyword:

$periodic 1
$lattice
5.000000000000000

Example 2: 2D periodic system with a unit cell |a| = 5 and |b| = 8 bohr, the angle
between a and b of 60◦ :

$periodic 2
$cell
5.0 8.0 60.0

Similar input using the $lattice keyword:

$periodic 2
$lattice
5.000000000000000 0.000000000000000
6.928203230275509 4.000000000000000

Example 3: 3D periodic system with a triclinic unit cell |a| = 5, |b| = 8 and |c| = 6
bohr, the angle between b and c of 70◦ , between a and c of 60◦ , and between a and
b of 90◦ :

$periodic 3
$cell
5.0 8.0 6.0 70.0 60.0 90.0
194CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

Similar input using the $lattice keyword:

$periodic 3
$lattice
5.000000000000000 0.000000000000000 0.000000000000000
6.928203230275509 4.000000000000000 0.000000000000000
4.328764234233443 3.324345345565466 6.354331231232453

• When k points are used specify their number using the keyword $kpoints. You
have to specify the number of k points components along each periodic direction.
Alternatively, a custom set of k points can also be provided.
• Optionally, create the $riper data group and add relevant keywords provided in Sec.
22.2.12.

The following examples illustrate the additions to the control file required for calculations
using the riper module.
Example 1: In this example a 3D periodic system is defined ($periodic 3). The unit
cell is specified using the $cell keyword, with lengths and angles given in atomic units and
degrees, respectively. A uniform grid of 27 (3 · 3 · 3) k points for numerical integration over
the BZ is specified using the keyword $kpoints. The section $riper (see Sec. 22.2.12) is
added with the keyword lenonly set to on. It forces riper to skip the calculation of energy
gradients (by default both energy and gradients are calculated in one run). In addition,
Gaussian smearing is switched on by providing sigma 0.01 (see Sec. 7.2.4).

$periodic 3
$cell
18.5911 16.5747 16.5747 90. 90. 90.
$kpoints
nkpoints 3 3 3
$riper
lenonly on
sigma 0.01

The same input using the $lattice keyword:

$periodic 3
$lattice
18.5911 0.0000 0.0000
0.0000 16.5747 0.0000
0.0000 0.0000 16.5747
$kpoints
nkpoints 3 3 3
$riper
lenonly on
sigma 0.01
7.3. HOW TO PERFORM A CALCULATION 195

Example 2: In this example a 2D periodic system is defined ($periodic 2). The unit
cell is specified using the $cell keyword, with lengths and angles given in Å and degrees,
respectively. A uniform grid of 9 (3 · 3) k-points for numerical integration over the BZ is
defined using the keyword $kpoints. The section $riper (see Sec. 22.2.12) is added with
the keyword lmaxmom 30. This changes the maximum order of multipole expansions used
in CFMM to 30 (default value is 20).

$periodic 2
$cell angs
18.5911 16.5747 90.0
$kpoints
nkpoints 3 3
$riper
lmaxmom 30

The same input using the $lattice keyword:

$periodic 2
$lattice angs
18.5911 0.0000
0.0000 16.5747
$kpoints
nkpoints 3 3
$riper
lmaxmom 30

Example 3: In this example a 1D periodic system is defined ($periodic 1). The dimen-
sion of the periodic direction in atomic units is specified using the $cell keyword. A one
dimensional grid of 3 k-points for numerical integration over the BZ is defined using the
keyword $kpoints.

$periodic 1
$cell
18.5911
$kpoints
nkpoints 3

The same input using the $lattice keyword:

$periodic 1
$lattice
18.5911
$kpoints
nkpoints 3
196CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

Example 4: In this example a molecular system is defined (default if no $periodic is

specified). The section $riper (see Sec. 22.2.12) is added with the keyword lpcg on, which
activates the low-memory modification of the RI approximation for calculations on very
large molecular systems.

$riper
lpcg on

Example 5: In this example a cubic 3D periodic system is defined ($periodic 3). The
unit cell is specified using the $cell keyword, with lengths and angles given in atomic
units and degrees, respectively. A set of 5 custom k points (kx (i), ky (i), kz (i)) along with
their weights (w(i)) for numerical integration over the BZ are supplied using the keyword
$kpoints followed by the option custom nks. One can refer to the literature for a list of
special k points and the corresponding weights for a particular Bravais lattice. Weights are
integers that are renormalized internally such that the sum of all weights is 1. Therefore,
only the relative ratios of the weights matter in a calculation.

$periodic 3
$cell
16.5747 16.5747 16.5747 90. 90. 90.
$kpoints
custom 5
0.00 0.00 0.00 1 #Gamma point
0.25 0.25 0.25 1 #Arbitrary point
0.50 0.50 0.50 1 #R point
0.00 0.50 0.00 1 #X point

7.3.4 Single Point Energy and Gradient

By default riper calculates single point energy and gradient in one run. For this, simply
invoke riper as

nohup riper > riper.out &

For energy calculation only add lenonly on to the $riper section in the control file, i.e.,

$riper
lenonly on

The parallel OpenMP version of riper can be invoked ad described in Sec. 3.4.2.
7.3. HOW TO PERFORM A CALCULATION 197

7.3.5 Structure Optimization

jobex will automatically use riper when the keyword $periodic is present in the control
file. Alternatively, the use of riper can be forced by specifying -riper argument of jobex,
i.e., invoking

nohup jobex -riper > jobex.out &

Note, that at present only structure optimization using Cartesian coordinates is possible
when using periodic boundary conditions. Using internal coordinates for periodic systems
will lead to convergence problems in structure optimizations.

7.3.6 Optimization of Cell Parameters

Simultaneous optimization of atomic positions and cell parameters or lattice vectors can be
performed by specifying the keyword $optcell in the control file and then invoking the
jobex script as described in 7.3.5.
The calculation of energy first derivatives with respect to lattice vectors usually requires
higher accuracy than nuclear gradients. Therefore, in case of convergence issues it is rec-
ommended to decrease the SCF convergence threshold to at least 1.0 × 10−7 by specifying
$scfconv 7. In addition, convergence can often be improved by including weight derivatives
by adding the option weight derivatives in the $dft data group.

7.3.7 Band Structure Plots

Band structure plots show the values of band (orbital) energies for values of k along lines
connecting symmetry points. In riper these lines can be specified by providing their start
and end points as well as the number of k points along the line within the $kpoints section.
They will be calculated at the end of SCF procedure and written to the file bands.xyz.
In the following example band energies are calculated along four lines, as specified by the
keyword kptlines 4. Each line definition starts in a new line with the keyword recipr,
followed by three real numbers defining the start point of the line and three real numbers
defining its end point. Finally, the number of k points along the line is given as an integer
number. Thus, the first line starts at the point (0.500 0.500 0.500), ends at (0.000
0.000 0.000) and contains 40 k points.

$kpoints
kptlines 4
recipr 0.500 0.500 0.500 0.000 0.000 0.000 40
recipr 0.000 0.000 0.000 0.500 0.500 0.000 40
recipr 0.500 0.500 0.000 0.746 0.373 0.373 40
recipr 0.746 0.373 0.373 0.000 0.000 0.000 40
198CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

Each line of the resulting bands.xyz file contains five real numbers: the coordinates k1 , k2
and k3 of the k point, its length |k| and the corresponding band energy nk .

7.3.8 Calculation of Densities and MOs on Grids

How to Perform

Calculation of data on grids for molecular and periodic systems using riper requires the
keyword $pointvalper in the control file. The values are calculated on a 3D grid and
written to appropriate output files. These files can be generated either running a single
point energy riper calculation or invoking

nohup riper -proper > riper.out &

provided that converged orbitals and occupation numbers from previous calculation are
present in the files RIPER.BANDS and RIPER.BANDS.OCCUPATIONS, respectively.
The format of the output files can be specified using the keyword fmt= following $pointvalper
in the same line. Supported formats are:

plt (default) The generated data is written to binary files that can be read by gOpen-
Mol and other external visualization programs. This format uses orthogonal grids.
Therefore, for non-orthogonal unit cells grid data is generated for a rectangular box
that contains the supercell (unit cell and its periodic images). By default, the values
at grid points outside of the supercell are set to zero. For strongly non-orthogonal
systems this may lead to large files. The option full switches off the zeroing of values
on grid points outside the supercell.

cub Gaussian cube format. Can be imported by several external visualization programs.

xyz Coordinates of grid points and calculated values are stored in a text file. Each line
contains Cartesian grid point coordinates and the corresponding value.

upt Values and grid data is output to binary files. First 8 (integer) records are: 1-2) rank
value and descriptor, 3-8) number of grid points and number of periodic unit cell
images in each periodic direction, 9...) Cartesian coordinates and the corresponding
value repeated for all grid points.

Output files in cub format contain information about atomic coordinates. For other formats
additional coord.xyz file containing atomic coordinates is generated.
The following options can be used along with the keyword $pointvalper:

nimg n1 n2 n3
Number of unit cell images n1, n2 and n3 in the periodic directions a, b and c,
respectively, for which plot data is generated.
7.3. HOW TO PERFORM A CALCULATION 199

npts n1 n2 n3
Number of grid points n1, n2 and n3 along each periodic direction. If not specified,
value 100 is used for each n.

eps real
Specifies the distance real in bohr around the system for which plot grid is generated
in aperiodic directions. Default value is 5 bohr.

ngrdpbx n
Number of grid points stored in one octree box during density calculations. Default
value is 50. For very large systems or high resolution it may be necessary to increase
this parameter to avoid memory allocation problems.

full
Only valid for plt output format. Switches off zeroing of values on grid points outside
the supercell.

Electron Density

Plots of electron density require option dens in the $pointvalper data group

$pointvalper
dens

The values of the total density and, in case of UHF calculations also the spin density are
written to files td.fmt and sd.fmt, respectively, where fmt is the selected format.

Molecular Orbitals

Plot files for visualization of molecular orbitals can be generated by setting orbs n in the
$pointvalper data group, where n is the number of orbitals to plot. The way to specify
the orbitals is best explained by some examples:
example 1: An open shell system is considered. Number of k points along the reciprocal
lattice vectors b1 , b2 and b3 is 3, 5, and 7, respectively. Possible k point components
kj calculated from equation 7.16 are therefore k1 ∈ −1 1 −2 −1 1 2



3 , 0, 3 , k2 ∈ 5 , 5 , 0, 5 , 5 , and
k2 ∈ −3 3


7 , ..., 7 The kpoints and $pointvalper groups are:

$kpoints
nkpoints 3 5 7
$pointvalper
orbs 2
k 1 4 6 a 1
k 2 3 4 b 3
200CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

Here, two orbitals will be plotted. They are defined in two lines following the orb 2 option.
Each line starts with k followed by three numbers that specify which of possible components
kj are used. In this example k point components for the first orbital are −1 1 2


3 , 5 , 7 . For the
second one k = (0, 0, 0), i.e., this orbital is at the Γ-point. Subsequently, type and number of
the orbitals are defined (orbitals at each k-point are ordered by increasing energy). Here the
first one is a 1 - the lowest α orbital. Similarly, b 3 denotes the third lowest energy β orbital.
For a non-Γ k point, the calculation will generate three plottable files corresponding to the:
(i) real component of Ψ, (ii) imaginary component of Ψ, and (iii)Ψ2 , for a given orbital.
Note: Only the real components are plotted for the Γ-point.
example 2: In this example a closed shell system with 3, 2 and 4 k points in each direction
is given:

$kpoints
nkpoints 3 2 4
$pointvalper
orbs 3
k 1 2 1 a 1
k 0 0 0 a 2

Possible k point components from equation 7.16 are k1 ∈ −1 , 0, 13 , k2 ∈ −1 , 14 and

3 4
k2 ∈ −3 −1 1 3


8 , 8 , 8 , 8 . In analogy to the previous example, three data files for the real compo-
nent, imaginary component, and square of the wavefunction (MO) will be generated for the
lowest energy orbital at k point −1 1 −3


3 , 4 , 8 . For the second molecular orbital only the real
component will be generated at the Γ-point. Orbitals at the Γ-point are defined by using
‘special’ 0 0 0 component indices, as it is not included in the grid for even number of k
points.
example 3: In this example an input for a Γ-point calculation is shown:

$pointvalper fmt=plt
orbs 2
k 1 1 1 a 5
k 1 1 1 b 5

Here k 1 1 1 is the only valid option. For Γ-point calculation, orbitals are real, thus it is
the only component that is plotted. In this case the fifth lowest energy orbitals with spin
components α and β are plotted. The output format is set to plt using the fmt= keyword.
example 4: In this example an input for a custom k point calculation is shown:

$kpoints
custom 2
0.00 0.00 0.00 1
0.50 0.50 0.25 1
$pointvalper fmt=cub
7.3. HOW TO PERFORM A CALCULATION 201

orbs 2
k 1 a 5
k 2 a 6

Instead of three indices after k (like in the previous examples), only a single index is required,
specifying the index of the custom k point in the list given by the user. The above example
will plot the fifth lowest energy molecular orbital at the first k point (Γ) and the sixth MO at
the second, (0.50 0.50 0.25) k point. Note: If the list of custom k points doesn’t include
a Γ-point, one can still plot the MOs at this point, as a Γ-point with 0 weight is prepended
to the supplied list with an index 0. In such a case, to plot the fifth lowest energy orbital
with spin α at the Γ-point, the input would be k 0 a 5 .

7.3.9 Density of States

A simulated density of states (DOS) is calculated by broadening the discrete energy levels
with Gaussians and superimposing them. The output files (dos for RKS calculations, and
dos_a+b, dos_a-b, dos_alpha, dos_beta in case of UKS) contain energies (first column)
and corresponding total DOS (second column) as well as s-, p-, ... contributions (following
columns). The calculation is controlled by the keyword $dosper:

$dosper width=real emin=real emax=real scal=real npt=integer

The parameters in example above have the following meaning:

width the width of each Gaussian, default value is 0.01 a.u.

emin,emax lower/upper bounds for energy in DOS calculation
scal scaling factor for DOS (total and s-, p-, ... contributions)
npt resolution (number of points)

All parameters are optional, default values are usually sufficient to generate a satisfactory
plot. DOS files can be generated either running a single point energy riper calculation or
using -proper option

nohup riper -proper > riper.out &

provided that converged orbitals and occupation numbers from previous calculation are
present in the files RIPER.BANDS and RIPER.BANDS.OCCUPATIONS, respectively.

7.3.10 RT-TDDFT

How to Perform

RT-TDDFT calculations require data-groups: $fields, $electric field and $rttddft to

provide the information about the type of field, electric field and the RT-TDDFT calculation,
respectively. Additionally, keywords $rtdipol and $rtenergy followed by file names may
202CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

be provided in the control file to print out the dipole moment and the energies respectively
every n number of steps. It is recommended to monitor the values in these files as the
calculation is running to ensure that the simulation is stable and not diverging. To calculate
the absorption spectrum and print it, the keyword $rtspectrum followed by the units is
needed. Possible values of the units are ev, nm, cm-1, and hartree. Example:

$rtspectrum ev

For now, only electric fields are supported. So, the $fields block would look like

$fields
electric on

The above would enable the use of an electric field.

The information pertaining to the electric field is provided within the $electric field
data-group.
There are currently two types of electric fields supported.

• Gaussian
• Static

Static fields only require the Cartesian components of the amplitude, while the Gaussian
field require two more parameters: peak position and peak width.
Note: The absorption spectrum can only be calculated for the Gaussian field.
To define a static field

$electric field
amplitude x=2.0E-5 y=2.0E-5 z=2.0E-5
static

To define a Gaussian ("kick") field

$electric field
amplitude x=2.0E-5 y=2.0E-5 z=2.0E-5
gaussian tzero=3.0 width=0.2

Note: the amplitudes are provided in atomic units (1 au = 514.2 V/nm). The Gaussian
electric field is defined as
(t−t )2
~
E(t) ~ − 2w02
= Ae (7.29)

tzero in the above code snippet corresponds to t0 (center of the peak) and width corresponds
to w (which gives a measure of the width of the pulse). Note: both are in atomic units of
time (1 au = 0.02419 fs)
Finally, the parameters regarding the RT-TDDFT calculation go within the $rttddft data-
group. The meanings are explained below
7.3. HOW TO PERFORM A CALCULATION 203

magnus
Can take values 2 or 4. "2" for second order Magnus expansion and "4" for fourth
order Magnus expansion. Default value is 2.

scf
if on, then SCF procedure is used for the time integration. If off
then Predictor-Corrector scheme is used instead.

iterlim
Max SCF cycles if scf is on. Default value is 15.

time
Specifies the evolution time in au. (1 au= 0.02419 fs)

tstep
The time step for the time evolution in au. 0.1 au is usually a good starting point.

print step
Specifies the number of steps n after which the dipole moments and energies are
printed out if requested. Default value is 100. That means the quantities are printed
out at every 100 steps. To have all the information for post processing, a value of 1
is recommended.

damping
Only valid for absorption spectrum calculation. It is the factor gamma in the equa-
tion to calculate the complex polarizability tensor. Default value is 0.004 au. Recom-
mended values in the range of 0.003 au to 0.005 au.

min energy
Only valid for absorption spectrum calculation. Specifies the minimum value of the
energy range used to perform the Fourier transform from time to frequency space.
Units: au. Default value is 0.15 au.

max energy
Only valid for absorption spectrum calculation. Specifies the maximum value of the
energy range used to perform the Fourier transform from time to frequency space.
Units: au. Default value is 0.625 au.

energy step
Only valid for absorption spectrum calculation. Specifies the step value or energy
interval dE at which to sample the energy values for Fourier transform and absorption
spectrum plotting. Units: au. Default value is 0.005 au.

Consider the following example.

Example 1: Here, an absorption spectrum calculation for a molecule is performed using

RT-TDDFT. The following keywords are expected in the control file.
204CHAPTER 7. DFT CALCULATIONS FOR MOLECULAR AND PERIODIC SYSTEMS

$fields
electric on
$electric field
amplitude x=2.0E-5 y=2.0E-5 z=2.0E-5
gaussian tzero=3.0 width=0.2
$rttddft
magnus 2
scf off
time 1000.0d0
tstep 0.1d0
print step 1
min energy = 0.013d0
max energy = 0.5d0
energy step 0.001d0
$rtdipol
$rtenergy
$rtspectrum ev

The data-group $fields is used to specify that only electric field is needed using electric on.
Here, a Gaussian "kick" pulse is used to excite the system. The parameters for which are
specified in the $electric field section. The pulse has a small amplitude in all three di-
rections. The amplitude is specified in au (1 au = 514.2 V/m). The position of the maximum
of the pulse is at t =3 au, specified using tzero. The width of the pulse is specified using
width parameter Eq. (7.29). The simulation goes on for 1000 au (1 au= 0.02419 fs) in steps
of 0.1 au as specified by the time and tstep keywords in the $rttddft section, respectively.
Second order Magnus expansion is used for the time-evolution operator indicated by magnus
2. Predictor-corrector scheme is used for time-integration since scf is off. Notice here, that
we are also requesting to print out energies and dipole moments using the $rtenergy and
$rtdipol keywords, respectively. The default names of the corresponding files are rtenrgy
and rtdipo. To print out to a custom file name use $rtenergy file filename instead.
It is recommended to have these files printed as they allow to monitor the energies in real
time and an unstable simulation can be detected early. The print step 1 in the $rttddft
section specifies that the dipole moments and the energies are printed at each iteration,
i.e. every step. This is useful if post-processing is needed after the calculation. Default
value of print step is 100. Finally, to calculate the spectrum, we provide the $rtspectrum
keyword followed by the units (in this case electron volts). The energy range for which the
absorption spectrum is calculated is 0.35 eV (0.013 au) to 13.6 eV (0.5 au), as specified using
the min energy and max energy keywords with the step interval of 0.001 au. The spectrum
is printed out in a file called rtspec.
The calculation is performed by running
nohup riper > riper.out &
To keep a track of the progress of the calculation, i.e. how many steps have been completed,
check the rtenrgy file.
The RT-TDDFT absorption spectrum can be plotted from the rtspec file after the calcu-
7.3. HOW TO PERFORM A CALCULATION 205

lation ends.

Example 2: In this example the electron density is plotted at each time step, to visualize
the electron dynamics in real-time. The following keywords are expected in the control
file.

$fields
electric on
$electric field
amplitude x=2.0E-5 y=2.0E-5 z=2.0E-5
gaussian tzero=3.0 width=0.2
$rttddft
magnus 2
scf off
time 1000.0d0
tstep 0.5d0
print step 1
$rtdipol
$rtenergy
$rtdens
$pointvalper fmt=cub
dens

Two new keywords are noticed here compared to example 1. $rtdens keyword tells the
program to print out the difference of the excited state and ground state density. Second,
the $pointvalper section, explained already, is used to specify the format in which the
electron density is printed out.
Chapter 8

Hartree–Fock and DFT Response

Calculations: Stability, Dynamic
Response Properties, and Excited
States

8.1 Functionalities of Escf and Egrad

escf and egrad are designed as efficient tools for response and excited state calculations on
large molecules. escf serves to compute the following properties for HF and KS reference
states:

• Eigenvalues of the electronic Hessian (stability analysis)

• Frequency-dependent polarizabilities and optical rotations

• Frequency-dependent electronic hyperpolarizabilities

• Vertical electronic excitation energies (TD-DFT)

• Vertical electronic excitation energies from the Bethe-Salpeter equation

• Transition moments, oscillator and rotatory strengths of electronic excitations

⇒ UV-VIS and CD spectra

• Two-photon transition moments

⇒ 2PA spectra

• Nuclear spin-spin coupling constants (SSCCs)

206
8.2. THEORETICAL BACKGROUND 207

• Damped response TD-DFT and BSE

Spin-restricted closed-shell and spin-unrestricted ground states (except for stability analysis)
are supported. The RI-J approximation in conjunction with LDA, GGA, and meta-GGA
(MGGA) functionals is implemented for all properties. Excitation energies and transition
moments can be computed either within the full time-dependent HF (TDHF) or time-
dependent DFT (TDDFT) formalisms or within the Tamm-Dancoff approximation (TDA).
Furthermore, two-component relativistic Kramers-restricted closed-shell ground states are
supported (vertical electronic excitation energies as well as transition moments, oscillator
and rotatory strengths). LDA and GGA functionals are implemented in combination with
the RI-J approximation at the TDDFT level. HF exchange is accessible within the TDA.
Excited state first order properties can be evaluated analytically using egrad. They include:

• Gradients of the excited state energy with respect to nuclear positions

⇒ Excited state equilibrium structures (jobex), adiabatic excitation energies, emis-
sion spectra
• Exited state densities ⇒ Charge moments, population analysis
• Excited state force constants by numerical differentiation of gradients (using the script
NumForce)
• First-order derivative couplings between the ground and an excited state as well as
between two excited-states (state-to-state)
• Transition moments, oscillator strengths between two TDHF/TDDFT excited-states
(state-to-state)

Moreover, analytical gradients of static and frequency-dependent polarizabilities are avail-

able from egrad. Together with vibrational normal modes from the aoforce or NumForce
they are used to calculate vibrational Raman intensities. Excited state gradients for MGGA
functionals are presently unavailable.
Again, ground states may be spin-restricted closed-shell or spin-unrestricted, RI-J is avail-
able, and either full TDDFT/TDHF or the TDA can be used. For further details we refer
to a recent review [162].

8.2 Theoretical Background

We briefly state the basic working equations in the following, as far as required to understand
the program output. For a more detailed treatment of the theory see refs. [38, 162–166] and
refs. therein. The following discussion is restricted to the one-component (nonrelativistic)
treatment, for the sake of convenience.
The first-order frequency dependent response of the density matrix can be expanded as
X
γ(x, x0 ) = {Xai ϕi (x)ϕ∗a (x0 ) + Yai ϕa (x)ϕ∗i (x0 )}. (8.1)
ai
208 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

The (real) expansion coefficients Xai and Yai are conveniently gathered in a “super-vector”
!
X
|X, Y i = (8.2)
Y

on L, the linear space of products of occupied and virtual ground state MOs
ϕi (x)ϕ∗a (x0 ) plus their complex conjugates. X and Y describe the first-order change of
the ground state MOs due to an external perturbation which is represented by |P, Qi on
L. For example, if an oscillating electric dipole perturbation along the z axis is applied,
|P, Qi = |µz i, where µ is the electric dipole operator.
Next we define the 2 × 2 “super-matrices”
! !
A B 1 0
Λ= , ∆= , (8.3)
B A 0 −1

where the four-index quantities A and B are the so-called “orbital rotation Hessians”. Ex-
plicit expressions for the standard A and B can be found, e.g., in ref. [38]. For MGGA
functionals, the linear response of the paramagnetic current density leads to additional XC
kernel matrix elements, and subsequently to modified definitions of A and B [167]. The
vector |X, Y i is determined as the solution of the TDHF/TDDFT response problem,

(Λ − ω∆)|X, Y i = −|P, Qi. (8.4)

If |Xα , Yα i arises from an electric dipole perturbation |µα i, the electronic dipole polarizabil-
ity at frequency ω is
ααβ (ω) = −hXα , Yα |µβ i, (8.5)
α, β ∈ {x, y, z}. Similarly, if |mα i is a component of the magnetic dipole moment operator,
the optical rotation is [168]
c
δαβ (ω) = − 3ω ImhXα , Yα |mβ i, (8.6)

where c is the light velocity.

Excitation energies Ωn are the poles of the frequency-dependent density matrix response.
They are thus the zeros of the operator on the left-hand side of Eq. (8.4),

(Λ − Ωn ∆)|Xn , Yn i = 0. (8.7)

The corresponding eigenvectors |Xn , Yn i are the transition density matrices for a given exci-
tation (also called “excitation vectors” in the following). They are required to be normalized
according to
hXn , Yn |∆|Xn , Yn i = 1. (8.8)
Transition moments are evaluated by taking the trace with one-particle operators, e.g.,

µ0n = hXn , Yn |µi (8.9)

for the electric and

m0n = hXn , Yn |mi (8.10)
8.2. THEORETICAL BACKGROUND 209

for the magnetic transition dipole moments.

The full TDHF/TDDFT formalism is gauge-invariant, i.e., the dipole-length and dipole-
velocity gauges lead to the same transition dipole moments in the basis set limit. This can
be used as a check for basis set quality in excited state calculations. The TDA can formally
be derived as an approximation to full TDHF/TDDFT by constraining the Y vectors to
zero. For TDHF, the TDA is equivalent to configuration interaction including all single
excitations from the HF reference (CIS). The TDA is not gauge invariant and does not
satisfy the usual sum rules [163], but it is somewhat less affected by stability problems (see
below). For MGGA functionals, the response of the paramagnetic current density is required
to ensure gauge invariance and is included by default.
Stability analysis of closed-shell electronic wavefunctions amounts to computing the lowest
eigenvalues of the electric orbital rotation Hessian A+B, which decomposes into a singlet and
a triplet part, and of the magnetic orbital rotation Hessian A−B. Note that A−B is diagonal
for non-hybrid and non-MGGA DFT, while A + B generally is not. See refs. [26, 167, 169]
for further details.
The two-component relativistic TDDFT eigenvalue problem for excitations of (Kramers-
restricted) closed-shell systems taking (approximately) into account the effect of spin-orbit
coupling is
M (X + Y )n = Ω2n (X + Y )n . (8.11)

M is a Hermitian (complex) matrix containing spinor energy differences, Coulomb matrix

elements as well as matrix elements of the two-component noncollinear exchange-correlation
kernel. An explicit expression for M can be found in ref. [27]. (X + Y )n are complex
two-component excitation vectors. In the case of HF exchange, one has to resort to the
TDA
AXn = Ωn Xn , (8.12)

see ref. [28].

Properties of excited states are defined as derivatives of the excited state energy with
respect to an external perturbation. It is advantageous to consider a fully variational La-
grangian of the excited state energy [38],


L[X, Y, Ω, C, Z, W ] = EGS + hX, Y |Λ|X, Y i − Ω hX, Y |∆|X, Y i − 1
X X (8.13)
+ Zia Fia − Wpq (Spq − δpq ).
ia pq

Here EGS denotes the ground state energy, F and S are the Fock and overlap matrices,
respectively, and indices p, q run over all, occupied and virtual MOs.
First, L is made stationary with respect to all its parameters. The additional Lagrange
multipliers Z and W enforce that the MOs satisfy the ground state HF/KS equations and
are orthonormal. Z is the so-called Z-vector, while W turns out to be the excited state
energy-weighted density matrix. Computation of Z and W requires the solution of a single
static TDHF/TDKS response equation (8.4), also called coupled and perturbed HF/KS
equation. Once the relaxed densities have been computed, excited state properties are
210 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

obtained by simple contraction with derivative integrals in the atomic orbital (AO) basis.
Thus, computation of excited state gradients is more expensive than that of ground state
gradients only by a constant factor which is usually in the range of 1 . . . 4.
TDHF/TDDFT expressions for components of the frequency-dependent polarizability ααβ (ω)
can also be reformulated as variational polarizability Lagrangians [170]

Lαβ [Xα , Yα , Xβ , Yβ , C, Z αβ , W αβ ](ω)

= hXα , Yα |(Λ − ω∆)|Xβ , Yβ i + hXα , Yα |µβ i + hµα |Xβ , Yβ i
X αβ X (8.14)
αβ
+ Ziaσ Fiaσ − Wpqσ (Spqσ − δpq ).
iaσ pqσ,p≤q

The stationary point of Lαβ (ω) equals to −ααβ (ω). The requirement that Lαβ (ω) be sta-
tionary with respect to all variational parameters determines the Lagrange multipliers Z αβ
and W αβ . All polarizability components αβ are processed simultaneously which allows for
computation of polarizability derivatives at the computational cost which is only 2–3 higher
than for the electronic polarizability itself.
Within TDDFT and TDHF, the X and Y coefficients are normalized as follows:
X
(|Xia |2 − |Yia |2 ) = 1, (8.15)
ia

where i and a label occupied and virtual MOs, respectively. Thus, the squared "coefficient"
of a single electron excitation from orbital i to orbital a can be defined as

|cia |2 = |Xia |2 − |Yia |2 . (8.16)

escf prints out |cia |2 ∗100 starting with the largest coefficient, until the sum of the coefficients
is 0.9 or greater. TDA is contained as special case with Yia =0.
The first hyperpolarizability is computed from
 
βαβγ (ωα , ωβ ) = tr K (αβ) v (γ) − hX (α) , Y (α) |P (αβ) , Q(αβ) i (8.17)

where K (αβ) is the unrelaxed density,

(αβ)
X h (α) (β) (α) (β)
i
Kij = − Xja Yia + Xja Yia , (8.18)
a
(αβ)
Xh (α) (β) (α) (β)
i
Kab = Xia Yib + Xia Yib , (8.19)
i

and P (αβ) and Q(αβ) are second-order right-hand-sides defined in Ref. [166].
Two-photon absorption (2PA) amplitudes are

µ0n n n
(αβ) (ω) = hX , Y |P
(αβ)
, Q(αβ) i (8.20)

where the frequencies in the right-hand side are ωα = ω and ωβ = Ωn − ωα .

State-to-state transition moments can be derived by examining the double residue of
the quadratic response function. [166, 171] The state-to-state 1-particle transition moment
8.3. IMPLEMENTATION 211

(1TDM) is

!
nm,QR − Xn (Xm )T + Yn (Ym )T Xnm
γ = (8.21)
(Ynm )T (Xn )T Xm + (Yn )T Ym

where the off-diagonal blocks require the solution of a dynamic-polarizability-like equation,

−1
|X nm , Y nm i = − (Λ − Ωnm ∆) |P nm , Qnm i. (8.22)

For static polarizabilities and nuclear spin-spin coupling constants (SSCC), a linear
system of equations (LSE) of the following type has to be solved:

X
Gai,bj λbj = −Rai , (8.23)
bj

where G = A + B or G = A − B, depending on the exact type of calculation. Details

can be found in [172] (static polarizabilities) and [106, 173] (SSCC). R are the right-hand
side integrals which can be computed rather easily. There are three (one for each Cartesian
direction) LSE for static polarizability and ten per nucleus for coupling constants. The final
quantity of interest is a tensor obtained (in a simplified manner) as

X X
KKL = λai,K Rai,L = Rai,K λai,L , (8.24)
ai ai

where K and L are nuclei (for static polarizabilities, omit these indices).
For SSCCs, this is by default scaled by the gyromagnetic ratios of the nuclei,

γK γL
JKL = h KKL . (8.25)
2π 2π

8.3 Implementation

Without giving details, we discuss features of the implementation in escf and egrad that
matter for applications. The interested reader is referred to the refs. given in the program
headers as well as ref. [174].

Simultaneous vector iteration. The solutions of Eqs. (8.4) and (8.7) (Eq. (8.11))
are expanded in a subspace of L which is iteratively expanded (Davidson method [175]).
The iteration is stopped when the Euclidean norm of the residual vector is smaller than
10−k . The default for k is 5, which usually gives excitation energies accurate to 8 − 10
digits and properties accurate to 4 − 5 digits (k can be changed by specifying $rpaconv
k ). Several roots, i.e., several excited states or frequencies can be treated simultaneously,
which is very effective and permits the calculation of whole excitation spectra and dispersion
curves. During the iteration, the vectors are kept on scratch files vfile_<IR>,wfile_<IR>,
and/or rhs_<IR>, where IR denotes an IRREP of the point group (see below). Before the
212 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

programs terminate, the converged vectors are written onto formatted files htypeihIRi, where
type is an abbreviation for the type of response calculation performed (cf. $scfinstab).
Given these files in the working directory, escf and egrad calculations can be restarted or
continued, e.g., with a larger number of roots.

Integral direct algorithm. In the iterative method outlined above, the super-matrices
A and B never need to be set up explicitly; only the products of A and B with some suitable
basis vectors are required. These matrix-vector-products are evaluated very efficiently in the
AO basis, because the required four-index integrals can be computed “on the fly” and need
not be transformed or stored on disk. In addition, prescreening techniques based on rigorous
bounds are straightforward to apply. This leads to a low-order scaling O(N 2 ) − O(N ) for
the time-determining steps. Due to the similarity to ground state fock matrix construction,
the same keywords are used to control these steps as in semi-direct SCF, namely $thime,
$thize, $scfintunit, see Chapter 6. The same is true for DFT and RI keywords such as
$dft, $ridft, $ricore.

Point group symmetry. escf and egrad can exploit point group symmetry for all
finite point groups (with up to 99-fold symmetry axes, → $symmetry). The calculation of
SSCCs is only possible for D2h and its subgroups. The response and eigenvalue problems
(eqs. (8.4) and (8.7)) decompose into separate problems for each IRREP that are solved
independently. For excited state and instability calculations, it is thus necessary to specify
the IRREPs to be treated ($soes, see below). For response calculations, the perturbation
is automatically subduced into irreducible components. The overall speedup compared to
C1 symmetry is approximately 1/g, where g denotes the point group order. For spin-
restricted closed-shell ground states, spin symmetry is used to further reduce the dimension
of the response and eigenvalue problems by a factor of 2. Point group symmetry cannot be
exploited in two-component calculations.

Other features. escf and egrad fully support external fields (using the keyword $electrostatic
field; specify geofield on in $fldopt), point charges (using the keyword $point_charges),
and effective core potentials (using $ecp). In escf calculations, occupied and virtual MOs
can be frozen (using $freeze).

8.4 How to Perform

The most convenient way to set up an escf or egrad calculation is to use the ex option of
the last (“general”) define menu, see Chapter 4. define will automatically provide most of
the keywords discussed below.
A large number of (not necessarily realistic) sample inputs is contained in the escf and
egrad subdirectories of the test suite (TURBOTEST directory).
8.4. HOW TO PERFORM 213

8.4.1 Preliminaries

All response calculations require a complete set of converged (occupied and virtual) SCF
MOs. It is strongly recommended to use well converged MOs, since the error in the ground-
state wavefunction enters linearly in all response properties. Thus, before starting escf or
egrad, specify the keywords

$scfconv 7
$denconv 1d-7

in control, perform a dscf statistics run if semi-direct integral processing is to be used (see
Chapter 3), and (re-)run dscf or ridft,

dscf > dscf.out & or

ridft > ridft.out & in case of RI-J.

The above tight convergence criteria are also recommended for excited state geometry op-
timizations. It is also recommended to avoid multiple grids as they negatively influence the
numerical stability (use 3 instead of m3). To perform a two-component TDDFT calcula-
tion, the two-component version of ridft has to be run before (see Chapter 6.4) using the
keywords $soghf and $kramers.

8.4.2 Polarizabilities and Optical Rotations

The calculation of dynamic polarizabilities is controlled by the keyword

$scfinstab dynpol unit
list of frequencies

unit specifies the unit of the following frequencies and may be ev, nm, 1/cm, or a.u. (default).
The frequencies may be either purely real or purely imaginary. For example, to calculate
dynamic polarizabilities at 590 nm and 400i nm (i is the imaginary unit), specify

$scfinstab dynpol nm
590
400 i

and run escf,

escf > escf.out & .

The resulting polarizabilities and rotatory dispersions are given in a.u. in the program output
(escf.out in the above example).
The conversion of the optical rotation in a.u. to the specific rotation [α]ω in deg·[dm·(g/cc)]−1
is given in Eq. (15) of ref. [168].
[α]ω = C · δ(ω) (8.26)
214 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

where C = 1.343 · 10−4 ω 2 /M with M being the the molar mass in g/mol, ω the frequency
in cm−1 , and δ(ω) is 1/3 trace of the electronic rotatory dispersion tensor given in atomic
units.
Please note that δ(ω) has the wrong sign in older TURBOMOLE versions. It has been
corrected in version 6.2.
Note that convergence problems may occur if a frequency is close to an electronic excita-
tion energy. This is a consequence of the (physical) fact that the response diverges at the
excitation energies, and not a problem of the algorithm.
Static polarizabilities are calculated most efficiently by specifying

$scfinstab polly

before starting escf.

8.4.3 Damped Response Calculations

A damped response calculation (adding a complex constant to a polarizability) can be in-

voked by additionally adding the following keywords to the control file:

$scfinstab dynpol nm
590
$damped_response 0.25 eV

The last keyword distinguishes it from a simple polarizability calculation described above.
Defined units for the frequency and the damping factor may be different. Valid units are
eV, cm-1 and nm; if nothing is specified atomic units are assumed. Further it is possible to
specify an evenly spaced list of frequencies:

$scfinstab dynpol eV
2.0 3.0 10
$damped_response 0.25 eV

This will perform a damped response calculation of 10 frequencies between 2.0 eV and 3.0 eV
with a complex damping factor of 0.25 eV. Damped response polarizabilities can be per-
formed for 1c (open and closed shell) and 2c Kramers-symmetric quasi-relativistic references
(ECPs or relativistic all-electron approaches) within the time-dependent DFT (including lo-
cal hybrid functionals) and also using the Bethe-Salpeter equation. The latter is especially
recommended if core states are targeted. Note that in the case of GW -BSE damped response
calculations the targeted core states should be included in the orbital range when the GW
quasiparticle energies are evaluated. A picture-change correction of the dipole moment is
available for relativistic all-electron Hamiltonians ($rdkh, $rbss, $rx2c, also in their local
variant $rlocal)and enforced by $pcc. For details on relativistic effects please, see Sec. 6.4.
8.4. HOW TO PERFORM 215

8.4.4 Dynamic First Hyperpolarizability

Hyperpolarizability calculations are run through escf and can be requested by including

$scfinstab hyperpol <unit>

<freq1>
<freq2>
<pair1_a> <pair2_b>
...

in the control file. A hyperpolarizability calculation will be performed using the pairs given
on each line AND using all combinations of the single frequencies given alone on each line.
Specifying no frequencies initiates a static calculation. The same unit specifications are
accepted as for dynamic polarizability calculations. For example,

$scfinstab hyperpol nm
800
1000
1064 1064

will compute hyperpolarizability tensors with frequency pairs

• 1000 nm, 1000 nm

• 1000 nm, 800 nm
• 800 nm, 800 nm
• 1064 nm, 1064 nm

8.4.5 Stability Analysis

Stability analysis of spin-restricted closed-shell ground states is enabled by

$scfinstab singlet
for singlet instabilities,
$scfinstab triplet
for triplet instabilities (most common), and
$scfinstab non-real
for non-real instabilities.
$scfinstab complex
for general complex instabilities (2c Kramers symmetric reference).

After that, it is necessary to specify the IRREPs of the electronic Hessian eigenvectors
(“orbital rotations”) to be considered. Without additional knowledge of the system one
usually needs to calculate the lowest eigenvalue within every IRREP:
216 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

$soes all 1

Positivity of the lowest eigenvalues in all IRREPs is sufficient for stability of the ground
state solution. If one is interested in, say, the lowest eigenvalues in IRREPs eg and t2g only,
one may specify:

$soes
eg 1
t2g 1

Triplet instabilities in the totally symmetric IRREP indicate open shell diradical states
(singlet or triplet). In this case, start MOs for spin-symmetry broken UHF or UKS ground
state calculation can be generated by specifying

$start vector generation

escf will provide the start MOs (→ $uhfmo_alpha, $uhfmo_beta) as well as occupation
numbers (→ $alpha shells, $beta shells) for a spin-unrestricted calculation with equal
numbers of α and β electrons (pseudo-singlet occupation).

8.4.6 Vertical Excitation and CD Spectra

The calculation of excited states within the TDHF(RPA)/TDDFT approach is enabled by

$scfinstab rpas
for closed-shell singlet excitations,

$scfinstab rpat
for closed-shell triplet excitations, and

$scfinstab urpa
for excitations out of spin-unrestricted reference states.

If it is intended to use the TDA instead, specify

$scfinstab ciss
for closed-shell singlet excitations,

$scfinstab cist
for closed-shell triplet excitations,

$scfinstab ucis
for excitations out of spin-unrestricted reference states, and

$scfinstab spinflip
for spin-flip (z-component of the total spin changes by ±1) excitations out of spin-
unrestricted reference states. For details concerning the theory see ref. [176]. In
8.4. HOW TO PERFORM 217

practice, this functionality can be used for the calculation of triplet-singlet, quartet-
doublet, . . . excitations (see ref. [177] also for further information about the imple-
mentation). It is only available within the TDA in combination with LDA functionals
and the HF exchange. It is strongly recommended to increase $escfiterlimit.

In the two-component case, specify

$scfinstab soghf
for two-component excitation energy calculations on closed-shell systems. [27] This
implementation is only available in combination with LDA and GGA functionals;
since version 7.4 also hybrid functionals are supported [29].

$scfinstab tdasoghf
for two-component excitation energy calculations on closed-shell systems using the
TDA, where in addition HF exchange is accessible. [28]

open-shell systems can be accessed using TD-HF or the Bethe-Salpeter equation $bse

The keywords $soghf and $kramers in case of closed-shell systems also have to be set.
The Bethe-Salpeter equation (BSE) for excited states can be invoked by adding the keyword
$bse. The correlation-augmented Bethe-Salpeter equation (cBSE) for excited states can be
invoked by adding the keyword $cbse [30]. In BSE calculation RI-K is mandatory and
automatically set.
Next, the IRREPs of the excitations need to be defined, which is again accomplished using
$soes. For example, to calculate the 17 lowest excitations in IRREP b1g , the 23 lowest
excitations in IRREP eu , and all excitations in IRREP t2g , use

$soes
b1g 17
eu 23
t2g all

and run escf. Since point group symmetry cannot be exploited in two-component calcula-
tions, there is only the totally symmetric IRREP a.
Note that $soes specifies the IRREP of the excitation vector which is not necessarily iden-
tical to the IRREP of the excited state(s) involved. In general, the IRREP(s) of the exci-
tation(s) from the ground to an excited state is given by the direct product of the IRREPs
of the two states. For example, to calculate the first A2 state in a C2v -symmetric molecule
with a B2 (open-shell) ground state, it is necessary to specify

$soes
b1 1
218 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

The number of excitations that have to be calculated in order to cover a certain spectral
range is often difficult to determine in advance. The total number of excitations within each
IRREP as provided by the define ex menu may give some hint. A good strategy is to
start with a smaller number of excitations and, if necessary, perform a second escf run on
a larger number of states using the already converged excitation vectors as input.
To compute absorption and CD spectra, it is often sufficient to include optically allowed
transitions only. This leads to substantial reduction of computational effort for molecules
with higher symmetry. For example, in the UV-VIS spectrum of an Oh symmetric molecule,
only t1u excitations are optically allowed. The IRREPs of the electric and magnetic dipole
moments as well as of the electric quadrupole moment are displayed automatically in the
define ex menu.
If a large number of states is to be calculated, it is highly recommended to provide extra
memory by specifying

$rpacor m

the integer m being the core memory size in megabytes (default is 20). The larger m, the
more vectors can be processed simultaneously without re-calculation of integrals. As a rule
of thumb, m should be ca. 90% of the available main memory. If RI-J is used ($ridft), it
is recommended to set $ricore to a small value and $rpacor to a large value if the number
of states is large, and vice versa if it is small. Since two-component calculations are more
demanding concerning computation time and required memory it is strongly recommended
to increase $rpacor.
By specifying

$spectrum unit and/or

$cdspectrum unit

a list of excitation energies and oscillator and/or rotatory strengths of the optically allowed
transitions is written onto file spectrum and/or cdspectrum. As above, unit specifies the
energy unit and may be ev, nm, 1/cm, or a.u. (default). The files spectrum and cdspectrum
may conveniently be used for further processing, e.g., using a plotting program such as Gnu-
plot.
Additionally, a spectrum broadened by gaussian functions can be generated by the Peak
ANAlyzing MAchine (panama). [178] It reads in excitation energies and their corresponding
oscillator strengths from the escf output file and prints the resulting spectrum to data.plot
(on an eV axis) or to data.nm.plot (on a nanometer axis) which can be plotted by programs
such as Gnuplot.

Both differential densities and non-relaxed difference densities can be visualized to get
an impression of the character and localization of the excitation(s). Differential densities
(ed.plt) are generated by egrad after setting the keyword $pointval in combination with
the -proper option (see Sec. 19.2). A computational less demanding alternative are non-
8.4. HOW TO PERFORM 219

relaxed difference densities which are directly obtained from the escf output file by first
running panama and subsequently dscf -proper or ridft -proper. [178]

Exact TDDFT transition density can also be plotted by the escf program, see section 19.2.
By specifying

$curswitchdisengage

inclusion of the current-density response for MGGA calculations is disabled. Note that the
results of calculations using this flag will no longer be gauge-invariant and will differ from
results obtained with the standard gauge-invariant implementation.

In general, CD spectra calculated with the length representation of the electric transition
dipole moment are not gauge-invariant. Using GIAOs for the magnetic transition dipole
moment, however, makes CD spectra gauge-invariant even if the length representation is
used for the electric transition dipole moment. Calculating gauge-invariant rotatory strength
tensors and thus CD spectra is possible for closed-shell molecules if the keyword $mgiao is
added. First, an mpshift calculation needs to be run in order to obtain the perturbed density
which is written on disk in a filed called ’umunu’. Then, a regular escf calculation can be
run in order to calculate magnetic transition dipole moments which lead to gauge-invariant
rotatory strength tensors for the length representation.

8.4.7 Two-photon absorption

2PA calculations are run through escf and can be requested with

$scfinstab twophoton <excited state method>

<freq1>
...
$soes
...
$exopt
...

where <excited state method> should be a description of the excited state method (rpas,
ciss, urpa, ucis). A 2PA tensor will be computed for each excited state specified by the
$soes block using the frequencies <freq1> and Ωn −<freq1>. Also, half may be provided
as a frequency, which then uses half the excitation energy for each frequency (this is the
default). 2PA amplitudes for specific states can be computed by specifying the states in
$exopt. $exopt for different irreps can be specified just as for $soes or also as a comma
separated list of indices (e.g., “3”) and ranges (e.g. “5-7”). For example

$scfinstab twophoton rpas

220 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

$soes
a1 6
a2 8
$exopt
a1 1-3, 5
a2 all

will compute

• 6 excited states in the a1 IRREP and 8 excited states in a2 and

• 2PA amplitudes with frequencies equal to Ωn /2 for states 1, 2, 3, and 5 in irrep a1 ,

and all 8 computed states in a2 .

2PA amplitudes require the calculation of dynamic polarization vectors |X (α) , Y (α) i at the
frequencies specified (e.g., Ωn /2). If 2PA amplitudes are requested for many states, these
frequencies can become (nearly) resonant with lower-lying excitation energies, which can
cause severe convergence problems. If such problems are encountered, try limiting the
number of 2PA amplitudes computed in a single pass using $exopt.

8.4.8 Excited State Geometry Optimizations

The input for computing excited state gradients and properties using egrad is exactly the
same as for an excited state calculation using escf, see the previous section. Gradients and
properties are calculated only for one state at a time. By default, this is the highest excitation
specified by $soes (only one IRREP is allowed). Sometimes, e.g. close to excited state
intersections, it may be necessary to include higher excited states in the initial excitation
vector calculation to prevent root flipping. This is accomplished using

$exopt n

which explicitly enforces treatment of the n-th state; n must be less or equal the number of
states specified in $soes.
After the input for the ground and excited state calculations has been set up, an excited
state geometry optimization can be started by issuing the command

nohup jobex -ex &

The option -ex forces jobex to call egrad instead of grad (or rdgrad if -ri is also specified).
In each geometry step, the excitation energy is written on the fourth column in $energy,
and the data group $last excitation energy change is updated. Otherwise, the excited
state optimization proceeds in exactly the same way as a ground state optimization (see
Chapter 3).
8.4. HOW TO PERFORM 221

8.4.9 Excited State Force Constant Calculations

Excited state vibrational frequencies can be calculated by numerical differentiation of an-

alytic gradients using NumForce (see Chapter 15). A NumForce calculation for an excited
state may be started by the command

nohup NumForce -ex n > force.out &

where n is the number of the excited state in C1 symmetry. In order to determine n,

it is recommended to perform an escf calculation in C1 symmetry. Note that numerical
calculation of excited state force constants is likely to fail if there are other states nearby
(in C1 ), because the roots may flip when the molecule is distorted. Note also that it may
be necessary to include higher excited states (using $exopt, see above) in C1 calculations
of molecules with higher symmetry in order to enforce convergence to the correct state. In
any case, it should be checked that the energy change due to the displacements (available
in the numforce/KraftWerk/*.log files) is reasonably small.
For a NumForce run, the convergence criteria should be tightened. It is recommended to use
at least

$scfconv 8

in all NumForce calculations. Other NumForce options such as -central, -d, -np work in
exactly the same way as they do for ground states.

8.4.10 Polarizability Derivatives and Raman Spectra

Calculations of polarizability derivatives by the egrad program use the same specifications
in the $scfinstab data group as polarizability calculations by escf.

$scfinstab polly

specifies derivatives of the static polarizability, while

$scfinstab dynpol unit
frequency
requests derivatives of the dynamical polarizability at the given frequency. Note that, unlike
polarizability calculations, multiple frequencies are not allowed. Polarizability derivatives
have to be projected onto vibrational normal modes to obtain Raman intensities, see Chap-
ter 15 for further details.

8.4.11 State-to-state properties

All state-to-state properties, including transition moments and derivative couplings, are
computed using the state-to-state derivative coupling machinery in egrad, as the state-to-
222 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

state 1TDM is a byproduct of computing the derivative coupling. Only C1 symmetry is

supported.
To trigger the calculation of state-to-state derivative couplings, an option must be given to
the $nacme keyword.

$nacme <full/pseudo or response>

Providing either full or pseudo will compute derivative couplings under the pseudowave-
function approximation. This is the recommended option since pseudowavefunction cou-
plings are well-behaved and stable. Providing response will compute derivative couplings
with quadratic response theory. Couplings computed within response theory can diverge
unphysically, so caution is advised.
For derivative couplings, it is also recommended to neglect the antisymmetric overlap inte-
grals which are responsible for translational variance. This is approximately equivalent to
incorporating electron translation factors (ETF) and is enabled by placing

$do_etf

in the control file.

The $coupled states keyword will control which states are coupled. $coupled states
understands a comma separated list of numbers and ranges. Couplings will be computed
between every pair of states specified on $coupled states. By default, $coupled states
has the value “all”, which computes couplings between every state specified in $soes. For
example,

$coupled states 1-3, 6, 8

will compute couplings between each pair of states in the list 1, 2, 3, 6, 8.

$coupled states all

will compute couplings between all available states.

When using $nacme is found, the $exopt key has added flexibility that allows the simul-
taneous calculation of multiple excited state gradients. It understands the same syntax as
$coupled states such that

$coupled states 1-3, 6, 8

will compute gradients corresponding to states 1, 2, 3, 6, 8.

$coupled states all

will likewise compute gradients for all available states.

8.4. HOW TO PERFORM 223

8.4.12 Nuclear spin-spin coupling constants

All four terms of Ramsey’s theory [179] (Fermi-contact, spin–dipole as well as paramagnetic
and diamagnetic spin–orbit contributions, abbreviated FC, SD, PSO, and DSO, respectively)
can be calculated. The FC/SD cross contributions are included in the full tensor.
The calculation is triggered by the keyword $ncoupling in the control file (no $scfinstab
is needed). This keyword along with several options can be set in the last menu in define.
To get reliable results, a basis set with steep s functions (for example Jensen’s pcJ-n basis
sets [180, 181]) is needed.
The gyromagnetic ratios of the coupling nuclei can easily be changed in the atomic attributes
menu of the define module.
The output will include the isotropic part of the coupling (cf. 106),

iso 1 X
JKL = (JKL )αα , (8.27)
3 α=x,y,z

the anisotropic contribution,

s 
anis γK γL 3 1X
2 
iso )2 ,
JKL =h (KKL )αβ + (KKL )βα − 3(KKL (8.28)
2π 2π 2 4
αβ

and the complete tensor JKL according to 8.25. You can also choose to obtain the corre-
sponding quantities of the reduced coupling tensor K.

8.4.13 Prediciting colors using the Color Prediction Tool cpt

To obtain the calculated emission and absorption colors of a compound after a TD-DFT/GW -
BSE calculation simply execute cpt in the same directory (basically only the exspectrum
or spectrum file needs to be present). In case of response calculations from ricc2 or
pnoccsd the $spectrum keywords needs to be present for the spectrum file to be gen-
erated. escf, starting from version 7.4, will always generate exspectrum files that can
automatically be read in from cpt. Note: if more than one spectrum is present in the
exspectrum/spectrum file cpt will generate the accumulated color. For some examples see
test case TURBOTEST/escf/short/CPT and the included files and CRIT file.

8.4.14 Approximations for Coulomb and Exchange integrals

In hybrid functional TD-DFT (gradient) calculations most of the time is spent in evaluating
the exchange-integrals. Therefore two ways to speed these up have been implemented:

$rick
use the RI-K approximation (recommended if enough RAM available)
224 CHAPTER 8. HF AND DFT RESPONSE CALCULATIONS

$senex
use seminumerical exchange. The default settings of $esenex are recommended and
lead to negligible errors in the excitation energies.

Usage of RI-K is highly recommended if enough RAM and a fast disk can be provided.
RI-K is available for all types of calculations in the modules escf and egrad up to D2h
symmetry. The keyword $rick is exclusive to escf, egrad and aoforce, other programs
will not trigger usage of RI-K even if those parts will. The keyword $rik triggers RI-K in all
modules where it is available, also (but not only) in escf, egrad and aoforce. RI-K is not
available for range-separated hybrids. Seminumerical exchange is supported for all types of
calculations in escf and egrad, also two-component calculations are fully supported. We
recommend to use the default settings as they will yield reliable results in virtually all cases.

$esenex

Further, if one also wants to compute the Coulomb contribution using seminumerical tech-
niques the $pseudospectral keyword may be added to the control file.

$esenex
$pseudospectral

Seminumerical exchange and fully pseudospectral approaches can be used with all point
groups and functionals (global and range-separated hybrids). For a more detailed list of
options please refer to the general keyword section.
Chapter 9

Second-order Møller–Plesset
Perturbation Theory

Preliminary note

TURBOMOLE offers three programs for MP2 energy and gradient calculations. A "conven-
tional" implementation [182], mpgrad, based on the calculation of four-center integrals, an
implementation which uses the resolution-of-the-identity (RI) approximation [183] as part
the RI-CC2 program [11], ricc2, and an implementation which is meant for very large
(> 100 atoms and > 3000 basis functions) single point MP2 energy calculations. The latter
program uses in addition to a local RI also a hybrid OSV-PNO approximation to reduce the
scaling of the computational costs with the system size to ≈ O(N 2 ) but has for small and
medium-sized systems a larger prefactor than ricc2.

9.1 Functionalities of mpgrad, and ricc2, and pnoccsd

Functionalities of mpgrad:

• Calculation of MP2 energies and/or MP2 gradients for RHF and UHF wave functions.

• The frozen core approximation (possibility to exclude low-lying orbitals from the MP2
treatment) is implemented only for MP2 energies.

• Exploitation of symmetry of all point groups.

• Can be used sequentially or MPI-parallel.

• Can be combined with the COSMO solvation model (see section 9.7 and chapter 18.2
for details). (Presently restricted to sequential calculations.)

Functionalities of ricc2 at the MP2 level:

225
226 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

• Calculation of MP2 energies and/or gradients for RHF and UHF wave functions within
the RI approximation (RI-MP2). In geometry optimizations and vibrational frequency
calculations (with NumForce) it can be combined with RI-JK-SCF for the Hartree-
Fock reference calculation.

• The frozen core approximation is implemented for both energies and gradients.

• RI-MP2 needs optimised auxiliary basis sets, which are available for most standard
basis sets as e.g. SVP, TZVP, TZVPP, QZVPP as well as for the (aug-)cc-p(wC)VXZ
(X = D, T, Q, 5) basis set series (for Al–Ar also for the (aug-)cc-p(wC)V(X+d)Z
series and for p-block elements Ga–Rn also the respective ECP basis set series (-pp)).

• Exploitation of symmetry for all point groups for MP2 energies and gradients.

• Can be combined with the COSMO solvation model (see chapter 18.2 for details).

• Runs sequentially and parallel (with MPI, OpenMP and hybrid MPI/OpenMP)

• Contains an implementation of explicitly correlated MP2-F12 methods (presently re-

stricted to energies and the C1 point group).

• Can for open-shell calculations be used with UHF and single-determinant high-spin
ROHF reference wavefunctions. (ROHF-MP2 presently limited to energies.)

• Energies and gradients for the spin-component scaled SCS- and SOS-MP2 approaches,
including a Laplace-transformed implementation of SOS-MP2 with O(N 4 ) scaling
computational costs.

• Static polarizabilities (currently restricted to closed-shell reference wavefunctions and

the sequential and SMP versions; cannot yet be combined with spin-component scal-
ing), see Chapter 10.5 for a description of the input

• See Chapter 10 for further details.

Functionalities of pnoccsd:

• Currently restricted to CCSD, CCSD(T0), CCSD(T), MP2 and DFT double hybrid
(e.g. B2PLYP) single point energy calculations with a RHF or UHF reference deter-
minant and the C1 point group.

• Runs sequentially and parallel (MP2 with MPI, OpenMP and hybrid MPI/OpenMP,
CCSD and beyond with OpenMP).

• Contains an implementation of explicitly correlated PNO-MP2-F12 methods.

• See Section 12 for further details.

9.1.1 How to quote

• For calculations with mpgrad:
Semi-direct MP2 Gradient Evaluation on Workstation Computers: The MPGRAD
Program. F. Haase and R. Ahlrichs; J. Comp. Chem. 14, 907 (1993).
9.2. SOME THEORY 227

• For calculations with ricc2:

CC2 excitation energy calculations on large molecules using the resolution of the
identity approximation. C. Hättig and F. Weigend;

• for MPI parallel calculations with ricc2 in addition:

Distributed memory parallel implementation of energies and gradients for second-
order Møller-Plesset perturbation theory with the resolution-of-the-identity approxi-
mation. Christof Hättig, Arnim Hellweg, Andreas Köhn, Phys. Chem. Chem. Phys.
8, 1159-1169, (2006).

• for MP2-F12 calculations in addition:

The MP2-F12 Method in the TURBOMOLE Programm Package. Rafal A. Bachorz,
Florian A. Bischoff, Andreas Glöß, Christof Hättig, Sebastian Höfener, Wim Klopper,
David P. Tew, J. Comput. Chem. 32, 2492–2513 (2011).

• for O(N 4 )-scaling LT-SOS-MP2 calculations:

Scaled opposite-spin CC2 for ground and excited states with fourth order scaling com-
putational costs. Nina O. C. Winter, Christof Hättig, J. Chem. Phys., 134, 184101
(2011) and Scaled opposite-spin second order Møller–Plesset correlation energy: An
economical electronic structure method. Y., Jung, R.C. Lochan, A.D. Dutoi, and
M. Head-Gordon, J. Chem. Phys., 121, 9793 (2004).

• for SCS-MP2 calculations:

S. Grimme, J. Chem. Phys. 118, 9095 (2003).

• for RI-MP2 polarizabilities:

Large scale polarizability calculations using the approximate coupled cluster model
CC2 and MP2 combined with the resolution-of-the identity approximation. Daniel H.
Friese, Nina O. C. Winter, Patrick Balzerowski, Raffael Schwan, Christof Hättig, J.
Chem. Phys., 136, 174106 (2012).

• for PNO-MP2 calculations: A O(N 3 )-scaling PNO-MP2 method using a hybrid OSV-
PNO approach with an iterative direct generation of OSVs. Gunnar Schmitz, Ben-
jamin Helmich, Christof Hättig, Mol. Phys. 111, 2463–2476, (2013). and Principal
Domains in Local Correlation Theory. David. P. Tew, J. Chem. Theory Comput.
15, 6597 (2019)

• for explicitly correlated PNO-MP2-F12 calculations: Explicitly correlated PNO-MP2

and PNO-CCSD and its application to the S66 set and large molecular systems. Gun-
nar Schmitz, Christof Hättig, David Tew, Phys. Chem. Chem. Phys. 16, 22167–
22178 (2014).

9.2 Some Theory

Second-order Møller–Plesset Perturbation Theory (MP2) corrects errors introduced by the

mean-field ansatz of the Hartree–Fock (HF) theory. The perturbation operator is the dif-
ference the full electronic Hamiltonian and the Fock operator for occupied/occupied and
228 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

virtual/virtual block. The MP2 energy is (in spin orbitals) given by:
1 X ij
EMP2 = tab hij||abi, (9.1)
4
iajb

with the doubles amplitudes

hij||abi
tij
ab = , (9.2)
i + j − a − b
i and j denote occupied, a and b virtual orbitals, p are the corresponding orbital energies,
and hij||abi = hij|abi − hij|bai are four-center two-electron repulsion integrals.
MP2 gradients (necessary for optimisation of structure parameters) are calculated as analyt-
ical derivatives of the MP2 energy with respect to nuclear coordinates. Calulation of these
derivatives also yields the first order perturbed wave function, expressed as “MP2 density
matrix”, in analogy to the HF density matrix. MP2 corrections to first-order properties like
electric moments or atomic populations are obtained from the density matrix in the same
way as for Hartree-Fock.
The “resolution of the identity (RI) approximation” means expansion of products of virtual
and occupied orbitals in a basis of auxiliary functions. The calculation and transformation
of the four-center two-electron integrals (see above) is replaced by that of three-center in-
tegrals, which leads to computational savings of RI-MP2 compared to a conventional MP2
calculations by a factor of ca. 5 (small basis sets like SVP) to ca. 10 (large basis sets like
TZVPP) and more (for quadruple-ζ and larger basis sets). The RI errors—i.e. the errors
due to the RI approximation—are with optimised auxiliary basis sets small and well doc-
umented [184, 185]. The use of the mpgrad program is therefore only recommended for
reference calculations or if suitable auxiliary basis sets are not available.
The PNO-MP2 implementation in the pnoccsd program is meant for large systems with
' 100 atoms where the costs for RI-MP2 calculations become unreasonably large. The
PNO-MP2 implementation uses five additional approximations to reduce the scaling of the
computational costs with the system size to below O(N 2 ):

• The truncation of the pair natural orbital (PNO) basis, φāij = a φa dij
P
aā for each pair
ij
ij of occupied orbitals to PNOs with occupation numbers nā ≥ TPNO :
X ij ij
Dab dbb̄ = nij dij
b̄ ab̄
(9.3)
b

φa diaã for each

P
• The truncation of the orbital specific virtual (OSV) basis, φãi = a
occupied orbital i to OSVs with occupation numbers niã ≥ TOSV :
X
ii i
Dab dbb̃ = nib̃ diab̃ (9.4)
b

The OSV basis is used to expand the PNOs. For a pair ij the PNOs are expanded in
the union of the OSVs for the occupied orbitals i and j.
• The distant pair approximation: only such pairs ij are included in the correlation
treatment for which the contribution to the correlation energy is expected (based on
initially computed estimates) to be ≥ Tpair .
9.3. HOW TO PREPARE AND PERFORM MP2 CALCULATIONS 229

• The local RI approximation: two-electron integrals for a pair ij are expanded in a

pair-specific subset of the full auxiliary basis which is determined based on an overlap
criterium and the selection threshold TRI .

• The principal domain PAO approximation: the OSVs and PNOs are represented using
a domain of projected atomic orbitals which are selected to best represent the density.

9.3 How to Prepare and Perform MP2 Calculations

Prerequisites

Calculations with mpgrad, ricc2 or pnoccsd require

• a converged SCF calculation with the one-electron density convergence threshold set
to $denconv 1.d-7 or less

• the maximum core memory the program is allowed to allocate should be defined in
the data group $maxcor (in MB); the recommended value is ca. 3/4 of the available
(physical) core memory at most.

• orbitals to be excluded from the correlation treatment have to be specified in the data
group $freeze

• for mpgrad the calculation of gradients is omitted by adding the flag $mp2energy to
the control file; in this case only the MP2 energy is calculated.

Calculations with ricc2 and pnoccsd moreover use

• an auxiliary basis defined in the data groups $atoms and $cbas. If not specified before
the start of ricc2 or pnoccsd the programs will assign the default auxiliary basis sets
which have the same names as the one-electron basis sets and will try to extract
them from the TURBOMOLE basis set library. If this succeeds the assignment and
auxiliary basis sets are stored in the control file else the programs will stop.

This is not needed for mpgrad, but here one needs

• a specification for scratch files and their size in data group $mointunit (see Sec-
tion 22.2.19)

• and the number of passes for integral evaluations and transformations in data group
$traloop

For explicitly-correlated MP2-F12 calculations one needs—depending the details of the ap-
plied approximations—additionally a so-called complementary auxiliary basis set (CABS,
defined in $cabs) and a RI-SCF auxiliary basis set defined in $jkbas.
230 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

Calculations with ricc2 and pnoccsd

1. RI-MP2 calculations require the specification of auxiliary basis sets ($cbas) and a
converged SCF calculation with the one-electron density convergence threshold set to
$denconv 1.d-7 or less. In addition, the options $freeze (frozen core approximation)
and $maxcor (maximum core memory usage) should be set. All these settings can be
done during the input generation with the program define under the entries mp2/cc
or pnocc of the last main menu.

2. The ricc2 program requires the data group:

$ricc2
mp2
geoopt model=mp2
(Where the last line should only be included if the calculation of gradients is needed
e.g. in geometry optimizations.) This can be prepared with define in the menu cc.

3. The pnoccsd program does not require any special input for a MP2 calculation with
default thresholds but it is recommended to specify the PNO truncation threshold
explicitly in the data group:
$pnoccsd
mp2
tolpno= 1.0E-7
This can be prepared with define in the menu pnocc.

4. For explicitly-correlated MP2-F12 calculations with ricc2 or pnoccsd also the data
groups $rir12 and $lcg are needed.

5. Start a single ricc2 and pnoccsd calculations respectively with the commands ricc2
and pnoccsd.

6. For optimisation of structure parameters at the RI-MP2 level with ricc2 use the com-
mand jobex -level cc2. For geometry optimizations with RI-JK-SCF as reference
for RI-MP2 with the ridft and ricc2 binaries the additional option -rijk has to be
given.

7. The combination of RI-MP2 with RI-JK-SCF can lead to significant computational

savings in particular for geometry optimizations for small and medium sized molecules
with large basis sets (quadruple-ζ and beyond) or basis sets with diffuse functions (e.g.
the aug-cc-pVXZ basis set families). For large molecules with TZVPP or similar basis
sets, conventional direct SCF calculations are usually more efficient.

8. With ricc2 and pnoccsd spin-component scaled SCS- or SOS-RI-MP2 calculations

can be carried out by adding in the $ricc2 (for ricc2) or $pnoccsd (for pnoccsd
data group the line

scs cos=1.2d0 css=0.3333d0

where the two parameters are the scaling factors for, respectively, the opposite- and
same-spin contribution. The specification of the scaling factors is optional; the default
9.3. HOW TO PREPARE AND PERFORM MP2 CALCULATIONS 231

values are cos=6/5 and css=1/3 as recommended by S. Grimme in J. Chem. Phys.

118 (2003) 9095. The abbreviation sos can be used for SOS-MP2 calculations with
cos=1.3 and css=0.0 (Y., Jung, R.C. Lochan, A.D. Dutoi, and M. Head-Gordon, J.
Chem. Phys. 121 (2004) 9793.). SOS-MP2 is in ricc2 implemented with O(N 4 )
scaling costs. For such calculations the data group $laplace has to be added.
9. For technical recommendations and additional options for parallel RI-MP2 and PNO-
MP2 calculations with the ricc2 and pnoccsd programs see Secs. 3.4 and 10.6 and
12.

MP2 calculations with a ROHF reference state

With the program ricc2 it is possible to compute MP2 (and spin-component scaled) MP2
energies with single-determinant restricted open-shell reference wavefunctions. No additional
input is required apart from the usual ROHF input for the dscf and ridft programs and
a standard MP2 input for ricc2.
TURBOMOLEs Hartree-Fock codes can handle within the ROHF framework many cases,
which include beside common high- and low-spin configuration state functions also weighted
averages of high-spin CSFs (see Sec. 6.3 for further details). The Møller-Plesset perturbation
theory and coupled cluster functionalities implemented in ricc2 require a single-determinant
reference state and can thus only deal with high-spin open-shell cases (not averaged):

• The spins of all nopen unpaired electrons are parallel (α spin will be assumed) so that
the ROHF reference state has the spin multiplicity nopen + 1.
• There must be only one type of open shells and all orbitals in this shell must have the
occupation number 1.
• For the single electron case (i.e. doublets) the Roothaan parameters are a = b = 0,
for high-spin cases with more than one unpaired electron the Roothaan parameters
must be set to a = 1 and b = 2.

For non-Abelian points groups this implies that shells with degenerate orbitals (as e.g. t1
in point group I) must be half-filled. An average over the different (symmetry-equivalent or
inequivalent) high-spin determinants that are obtained when a shell of degenerate orbitals
is less than or more than half-filled is not possible with single-point ricc2 calculations.
For states with less than or more than half-filled shells of degenerate orbitals the calculations
half to be done in a point group that lifts the degeneracy such that it becomes possible to
assign integer occupation numbers. A symmetry-breaking of the orbitals can be avoided by
doing the Hartree-Fock calculation in the full point group. The input (and MO coefficients)
can then be transformed to a lower point group using define only for the ricc2 calculation.
When using an ROHF reference, two definitions of the frozen core orbitals are possible: The
frozen core orbitals can either be spin restricted, or spin unrestricted. In the first case the
alpha and beta core orbitals are those on the mos file generated by the ROHF dscf calcu-
lation and are the same, but the Fock matrix elements between core and valence orbitals
232 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

are non-zero. In the second case, the alpha and beta orbitals semi-canonicalised prior to
selecting the frozen orbitals with lowest orbital eigenvalues. Here the frozen alpha and beta
orbitals differ (slightly), but the core-valence Fock matrix elements are zero. Both options
are available in ricc2 calculations by specified restricted res or (semi-)canonical can $ricc2
core res/can

Calculations with mpgrad

1. Add $denconv 1.d-7 to the control file and perform a dscf run.

2. If any orbitals are decided to be excluded from MP2 treatment, add data group
$freeze manually to the control file, see also Section 22.2.19.

3. For preparation of an mpgrad run use the script Mp2prep:

mp2prep -e/g -m memory -p discspace [scratch file directory]
As an example, with the command
mp2prep -e -m 100 -p 1000 /work
an MP2-energy calculation is prepared, the amount of available core memory is re-
stricted to 100 MB, the MOs are blocked, so that integral scratch files—located in
the directory /work—do not need more than 1000 Mb. The number of blocks, i.e.
the number of passes with repeated integral evaluations, is written to the control
file ($traloop) as well as the specification of scratch files ($mointunit, see Sec-
tion 22.2.19). Note: less disc space means more passes and thus lower efficiency
of mpgrad, but due the technical limitations discspace should be limited to values <
16Gb to avoid integer overflow errors. Settings obtained by mp2prep may be changed
manually. You may change the number of passes in $traloop by editing the control
file (e.g. if the originally intended disc space is not available). To adapt the size of
scratch files add $statistics mpgrad to control file and start an mpgrad statistics
run with the command mpgrad.

4. Start a single mpgrad calculation with the command mpgrad.

5. For optimisation of structure parameters at the (non-RI-) MP2 level use the command
jobex -level mp2. Note, that the frozen core approximation is ignored in this case.

9.4 General Comments on MP2 Calculations, Practical

Hints

Recommendations
• It is well-known, that perturbation theory yields reliable results only, if the perturba-
tion is small. This is also valid for MP2, which means, that MP2 improves HF results
only, if HF already provides a fairly good solution to the problem. If HF fails, e.g. in
9.4. GENERAL COMMENTS 233

case of partially filled d-shells, MP2 usually will also fail and should not be used in
this case.

• MP2 results are known to converge very slowly with increasing basis sets, in particular
slowly with increasing l-quantum number of the basis set expansion. Thus for reliable
results the use of TZVPP basis sets (or higher) is recommended. When using SVP
basis sets at most a qualitative trend can be expected. Basis sets much larger than
TZVPP usually do not significantly improve geometries of bonded systems, but still
can improve the energetic description. For non–bonded systems larger basis sets
(especially, with more diffuse functions) are needed.

• It is recommended to exclude all non-valence orbitals from MP2 calculations, as nei-

ther the TURBOMOLE standard basis sets SVP, TZVPP, and QZVPP nor the cc-pVXZ
basis set families (with X=D,T,Q,5,6) are designed for correlation treatment of inner
shells (for this purpose polarisation functions for the inner shells are needed). The
default selection for frozen core orbitals in define (orbitals below -3 a.u. are frozen)
provides a reasonable guess. If core orbitals are included in the correlation treatment,
it is recommended to use basis sets with additional tight correlation functions as e.g.
the cc-pwCVXZ and cc-pCVXZ basis set families.

• RI-MP2: We strongly recommend the use of auxiliary basis sets optimized for the
corresponding orbital basis sets.

RI-MP2 calculations with the ricc2 program: All what is needed for a RI-
MP2 gradient calculation with the ricc2 program is a $ricc2 data group with the entry
geoopt model=mp2. If you want only the RI-MP2 energy for a single point use as option just
mp2. To activate in MP2 energy calculations the evaluation of the D1 diagnostic (for details
see Sec. 10.1). use instead mp2 d1diag. (Note that the calculation of the D1 diagnostic
increases the costs compared to a MP2 energy evaluation by about a factor of three.)

Comments on the Output

• Most important output for ricc2, pnoccsd, and mpgrad are of course the MP2(+HF)
energies (written to standard output and additionally to the file energy) and MP2(+HF)
gradients (written to the file gradient).

• In case of MP2 gradient calculations the modules also calculate the MP2 dipole mo-
ment from the MP2 density matrix (note, that in case of mpgrad a frozen core orbital
specification is ignored for gradient calculations and thus for MP2 dipole moments).

Further output contains indications of the suitability of the (HF+MP2) treatment.

• As discussed above, MP2 results are only reliable if the MP2 corrections to the
Hartree-Fock results are small. One measure for size of MP2 corrections to the wave-
function are the doubles amplitudes, tij ab , as is evident from the above equations.
mpgrad by default prints the five largest amplitudes as well as the five largest norms
234 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

of amplitude matrices tij for fixed i and j. The number of printed amplitudes can be
changed by setting the data group $tplot n where n denotes the number of largest
amplitudes to be plotted. It is up to the user to decide from these quantities, whether
the HF+MP2 treatment is suited for the present problem or not. Unfortunately,
it is not possible to define a threshold, which distinguishes a "good" and a "bad"
MP2-case, since the value of individual amplitudes or amplitudes matrices tij are not
orbital-invariant, but depend on the orbital basis and thereby under certain circum-
stances on the orientation, the point group, or the start guess for the MOs. Example:
the largest norm of t-amplitudes for the Cu-atom (d10 s1 , "good" MP2-case) amounts
to ca. 0.06, that of the Ni-atom (d8 s2 , "bad" MP2 case) is ca. 0.14.

• A more reliable criterion is obtained from the MP2 density matrix. Its eigenvalues
reflect the changes in occupation numbers resulting from the MP2 treatment, com-
pared to the Hartree-Fock level, where occupation numbers are either one (two for
RHF) or zero. Small changes mean small corrections to HF and thus suitability of the
MP2 method for the given problem. In case of gradient calculations ricc2 displays
by default the largest eigenvalue of the MP2 density matrix, i.e. the largest change in
occupation numbers (in %). If $cc2_natocc is set the full set of natural occupation
numbers and orbitals will be save to the control file. For main group compounds
largest changes in occupation numbers of ca. 5 % or less are typical, for d10 metal
compounds somewhat higher values are tolerable.

• A similar idea is pursued by the D2 and D1 diagnostics [186,187] which is implemented

in ricc2. D2 is a diagnostic for strong interactions of the HF reference state with
doubly excited determinants, while D1 is a diagnostic for strong interactions with
singly excited determinants.

9.5 RI-MP2-F12 Calculations

To obtain the F12 correction to the MP2 energy, the data group $rir12 must be added to
the control file. A typical run will include the input:

$ricc2
mp2 energy only
$rir12

The MP2-F12 ground-state energy is

EMP2-F12 = EMP2 + EF12 , (9.5)

where EMP2 is the conventional MP2 energy and EF12 the correction from explicitly-correlated
theory. The second term contains contributions from explicitly-correlated geminal basis
functions of the form

Q̂12 f12 |iji, (9.6)

9.5. RI-MP2-F12 CALCULATIONS 235

where |iji is a two-electron determinant of occupied (semi-)canonical Hartree–Fock spin or-

bitals, f12 is a correlation factor, which can be either linear r12 (in this case, the approach
is denoted MP2-R12 instead of MP2-F12) or a function of r12 , and Q̂12 defines the dou-
bles excitation space covered by the geminals (it also ensures strong orthogonality to the
occupied orbitals). Usually Q̂12 is chosen to be Q̂12 = (1 − Ô1 )(1 − Ô2 ) − V̂1 V̂2 , where
P
Ôµ = k |ϕk (µ)ihϕk (µ)| is the projection operator onto the space spanned by the occupied
P
spin orbitals ϕk and V̂µ = a |ϕa (µ)ihϕa (µ)| is the projector onto the virtual spin orbitals.
The F12 correction is obtained by minimizing the functional
X
FF12 = cTij Bij cij + 2cTij vij (9.7)
i<j

with respect to the amplitudes collected in the vector cij . The vectors vij and the matrices
Bij are defined as

−1
vij (kl) = hkl|f12 Q̂12 r12 |iji, (9.8)
Bij (kl, mn) = hkl|f12 Q̂12 (f1 + fˆ2 − εi − εj )Q̂12 f12 |mni,
ˆ (9.9)

in the spin-orbital formalism (m, n denote spin orbitals and |mni is a two-electron deter-
minant). fˆµ is the Fock operator for electron µ and εk is a (semi-)canonical Hartree–Fock
orbital energy.
The F12 implementation is compatible with ECP, DKH and X2C scalar relativistic treat-
ments. The additional terms that arise in the Fock matrix are included in the evaluation of
the F12 contributions. [188, 189]
A MP2-F12 calculation is defined through a number of choices concerning the nature of the
geminals (f12 and Q̂12 ), the geminal excitation space (ijkl or ijij) and approximations in
computing the B matrix (GBC, EBC, [T̂ , f12 ]). These choices correspond to keywords in
the $rir12 data group, explained below.
To run a MP2-F12 calculation, one has to select the auxiliary basis sets cbas, cabs and
optionally jkbas. The ricc2 program uses the robust fitting techniques of Ref. [190] for
the F12 integrals and the cbas basis is used for both the F12 and the usual MP2 Coulomb
integrals. For the density fitting of the Coulomb and exchange matrices of the Fock matrix,
the jkbas will be used instead of the cbas basis if it is included in the control file (this is
recommended and is achieved using the rijk menu in define). For the RI approximation
of the 3- and 4-electron integrals as sums of products of 2-electron integrals, intrinsic to the
F12 method, the complementary auxiliary basis (CABS) approach is used [191]. If define
is used to set up the cabs basis, the library cabasen is searched. This library contains the
optimised cabs basis sets [192] for the cc-pVXZ-F12 basis sets of Peterson et al. [193]. For
other basis sets, the auxilliary basis in the library cabasen is identical with the auxilliary
basis in the library cbas.
The $rir12 data group may be set by choosing the f12 option in the cc menu when running
define. This command activates the f12 menu, where the default options may be changed
if desired:
236 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

Orbital basis : cc-pVTZ-F12

Cardinal number : T
Recommended exponent: 1.0000
Actual exponent: 1.0000

INPUT MENU FOR MP2-F12 CALCULATIONS

ansatz : CHOOSE ANSATZ 2 [1,2*,2]

r12model : CHOOSE MODEL B [A,B]
comaprox : COMMUTATOR APPROXIMATION F+K [F+K,T+V]
cabs : CABS ORTHOGONALIZATION svd 1.0D-08 [cho,svd]
examp : CHOOSE FORMULATION fixed noflip [inv,fixed,noinv, flip,noflip]
r12orb : CHOOSE GEMINAL ORBITALS hf [hf,rohf,boys,pipek]
corrfac : CHOOSE CORRELATION FACTOR LCG [R12,LCG]
cabsingles: CABS SINGLE EXCITATIONS on [on,off]
pairenergy: PRINT OUT PAIRENERGIES off [on,off]
slater : SLATER EXPONENT 1.0000

* / end : write $rir12 to file and leave the menu

& : go back - leaving $rir12 unchanged...

ansatz corresponds to the choice of Q̂12 . Almost all modern MP2-F12 calcula-
tions use ansatz 2 (default), which gives much improved energies over
ansatz 1 (see Ref. [194] for details). The principal additional cost of
using ansatz 2 over ansatz 1 is concerned with the coupling between the
F12 and conventional amplitudes. This is avoided by choosing 2*, which
corresponds to neglecting EBC (Extended Brillouin Condition) terms in
the Fock matrix elements.

r12model is the method of computing the matrices Bij (see Ref. [194] for details).
The cost and accuracy increases from A to B. It is recommended to use
B (default). The energies computed using A are then also printed out in
the output.

comaprox is the method for approximately computing the integrals for the operator
[T̂ , f12 ], where the matrix representations of F+K or T+V are used. F+K
(the core Hamiltonian plus Coulomb term) is recommended and is the
default.

cabs refers to the method of orthogonalising the orbitals in the complemen-

tary auxiliary basis. Singular-value decomposition (svd) or Cholesky
decomposition (cho) are available. svd is recommended and is the de-
fault, with a threshold of 1.0d-08. The basis set used for CABS is set
from the cc menu.
9.5. RI-MP2-F12 CALCULATIONS 237

examp refers to the choice of excitation space. inv is the orbital-invariant

merhod of Ref. [195], with amplitudes cij (kl). noinv is the original
orbital-dependent diagonal "ijij" method of Ref. [195], with amplitudes
cij (ij) (not recommended, unless in combination with localised orbitals).
fixed is the (diagonal and orbital-invariant) rational generator approach
of Ref. [196], where the F12 amplitudes are not optimised but predeter-
mined using the coalescence conditions (default). An additional keyword
noflip suppresses the use of spin-flipped geminals in open-shell calcula-
tions; by default spin-flipped geminals are used as described in Ref. [22].

r12orb controls which orbitals are used in the F12 energy contribution. hf
means that (semi-)canonical Hartree–Fock orbitals are used (default).
rohf means that ROHF orbitals are used (any frozen orbitals will then
also implicitly be ROHF). For calculations on closed-shell systems, lo-
calised orbitals may be used. Both the Boys [197] and Pipek–Mezey [198]
methods are available for localisation of the orbitals.

corrfac corresponds to the choice of correlation factor f12 in the geminal basis
functions. R12 results in a calculation using linear-r12 and LCG results in
a calculation using the Slater-type correlation factor with exponent 1.4
a−1
0 , represented as a linear combination of six Gaussians (see Ref. [199]).
Note that the exponents 0.9, 1.0 and 1.1 a−1 0 are recommended for use
with the cc-pVXZ-F12 basis sets [193].

cabsingles switches on/off the calculation of a second-order correction to the Hartree–

Fock energy by accounting for single excitations into the complementary
auxiliary basis set (CABS). The single excitations into the CABS basis
can be computed without extra costs if the CABS Fock matrix elements
are required anyway for the F12 calculation (i.e., for ansatz 2, approx-
imation B or comaprox F+K). The computation of CABS singles cannot
be switched off if it is free of costs.

pairenergy controls whether or not the F12 contribution to the MP2 pair energies
appear in the output (default off).

Further options:
corrfac LCG refers to a further data group for the definition of the correlation factor. When
define is used, the default is

$lcg
nlcg 6
slater 1.4000

The nature of the LCG correlation factor may be changed by editing this data group in
the control file. For example, to use a Slater-type correlation factor with exponent 1.0 a−1
0 ,
represented as a linear combination of three Gaussians, use
238 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

$lcg
nlcg 3
slater 1.0000

Alternatively, the exponents and coefficients of the fit may be given explicitly:

$lcg
nlcg 3
expo1 coef1
expo2 coef2
expo3 coef3

MP2-F12 calculations may be combined with Grimme’s SCS approach (S. Grimme, J. Chem.
Phys. 118 (2003) 9095) by inserting scs in $ricc2,

$ricc2
mp2 energy only
scs

In this case, the SCS parameters cos=6/5 and css=1/3 are used. Also individual scaling
factors for the same-spin and opposite-spin contributions may be defined, see Section 10.7.
For open-shell calculations, two choices of the examp fixed noflip method are available.
These are controled by a keyword in the $rir12 data group

ump2fixed full [diag,full]

These differ in the treatment of the αβ block, where either only the diagonal excitations
enter (with amplitude 0.5) diag, or the equivalent of the spin-adapted singlet and triplet
pair excitations enter (as far as possible) full. Note that the diag method with UMP2-
F12 yields a result different to that of fixed MP2-F12, even for identical RHF and UHF
determinants. However, the diag method is somewhat less expensive than the full method.
Recommendations for orbital and auxiliary basis sets:
The best orbital basis sets to use for MP2-F12 calculations are probably the cc-pVXZ-F12
basis sets, specially optimised for MP2-F12 calculations [193] for the atoms H, He, B–Ne
and Al–Ar. In conjunction with these cc-pVXZ-F12 basis sets, we recommend to use the
optimised cc-pVXZ-F12 sets of Yousaf and Peterson [192] as cabs. Furthermore, cbas and
jkbas basis sets can be selected from the cbasen and jkbasen libraries, respectively, using
the alias cc-pVXZ-F12 (a jkbas is currently not available for He, Ne and Ar). This alias
points to the corresponding aug-cc-pwCV(X+1)Z cbas and aug-cc-pV(X+1)Z jkbas. These
recommendations are on the side of caution and are likely to be refined as more experience
is gained [188, 200, 201].
For atoms other than H, He, B–Ne and Al–Ar, optimised F12 basis sets are not yet available.
In this case, basis sets must be selected and/or optimised carefully. It is advised to contact
the Theoretical Chemistry Group in Karlsruhe for support (e-mail to: klopper@kit.edue).
9.6. LT-SOS-RI-MP2 WITH O(N 4 ) SCALING COSTS 239

9.6 Laplace-transformed SOS-RI-MP2 with O(N 4 ) scal-

ing costs

The ricc2 module contains an implementation of SOS-MP2 which exploits the RI approx-
imation and a Laplace transformation of the orbital energy denominators
Z ∞ nL
1 X
= e−(a +b −i −j )t dt ≈ wα e−(a +b −i −j )tα , (9.10)
a + b − i − j 0 α=1

to achieve an implementation with O(N 4 ) scaling costs, opposed to the conventional O(N 5 )
scaling implementation. In particular for large molecules the Laplace-transformed imple-
mentation can reduce a lot the computational costs of SOS-MP2 calculations without loss
in accuracy.
The Laplace-transformed implementation for SOS-MP2 calculations is activated with the
input

$laplace
conv=5

where the parameter conv is a convergence threshold for the numerical integration in Eq.
(9.10). A value of conv=5 means that the numerical integration will be converged to a root
mean squared error of ≈ 10−5 a.u.
Whether the conventional or the Laplace-transformed implementation will be more efficient
depends firstly on the system size (the number of occupied orbitals) and secondly on the
required accuracy (the number of grid points for the numerical integration in Eq. (9.10))
and can be understood and estimated from the following considerations:

• The computational costs for the most expensive step in (canonical) RI-MP2 energy cal-
culations for large molecules requires 12 O2 V 2 Nx floating point multiplications, where
O and V are, respectively, the number occupied and virtual orbitals and Nx is the
number of auxiliary functions for the RI approximation. For the LT-SOS-RI-MP2
implementation the most expensive step involves nL OV Nx2 floating point multiplica-
tions, where nL is the number of grid points for the numerical integration. Thus, the
ratio of the computational costs is approximately
1 2 2
2 O V Nx OV
conv : LT ≈ = ≈ O : 6nL ,
nL OV Nx2 2nL Nx

where for the last step Nx ≈ 3V (typical for valence TZ basis sets) has been assumed.
Thus, the Laplace-transformed implementation will be faster than the conventional
implementation if O > 6nL .

The number of grid points nL depends on the requested accuracy and the spread of the
orbital energy denominators in Eq. (9.10). The efficiency of Laplace-transformed SOS-
RI-MP2 calculations can therefore (in difference to conventional RI-MP2 calculations) be
240 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

enhanced significantly by a carefull choice of the thresholds, the basis set, and the orbitals
included in the correlation treatment:

• The threshold conv for the numerical integration is by default set to the value of conv
specified for the ground state energy in the data group $ricc2 (see Sec. 22.2.20), which
is initialized using the threshold $denconv, which by default is set conservatively to
the tight value of 10−7 .

– For single point energy calculations conv in $laplace can savely be set to 4,
which gives SOS-MP2 energies converged within ≈ 10−4 a.u. with computational
costs reduced by one third or more compared to calculations with the default
settings for these thresholds.
– For geometry optimizations with SOS-MP2 we recommend to set conv in $laplace
to 5.

• The spread of the orbital energy denominators depends on the basis sets and the
orbitals included in the correlation treatment. Most segmented contracted basis sets
of triple-ζ or higher accuracy (as e.g. the TZVPP and QZVPP basis sets) lead to
rather high lying “anti core” orbitals with orbital energies of 10 a.u. and more.

– For the calculation of SOS-MP2 valence correlation energies it is recommended

to exclude such orbitals from the correlation treatment (see input for $freeze
in Sec. 22).
– Alternatively one can use general contracted basis sets, as e.g. the correlation
consistent cc-pVXZ basis sets. But note that general contracted basis sets in-
crease the computational costs for the integral evaluation in the Hartree-Fock
and, for gradient calculations, also the CPHF equations and related 4-index
integral derivatives.
– Also for the calculation of all-electron correlation energies with core-valence basis
sets which include uncontracted steep functions it is recommended to check if
extremely high-lying anti core orbitals can be excluded.

Note that for large molecules it is recommended to disable for geometry optimizations (or for
gradient or property calculations in general) the preoptimization for the Z vector equations
with the nozpreopt option in the $response data group (see Sec. 22.2.20).

Restrictions:

• It is presently not compatible with the calculation of the D1 and D2 diagnostics. The
respective options will be ignored by program if the Laplace-transformed implemen-
tation is used.
9.7. COSMO-MP2 241

9.7 COSMO-MP2

In TURBOMOLE MP2 can be combined in two ways with COSMO solvation model, which
are known in the literature as “Perturbation Theory on Energy” (PTE) and “Perturbation
Theory on Energy and Density” (PTED) approaches. The PTE approach is obtained by
truncating the free energy of the solute strictly at second-order in the difference between
the Hartree-Fock mean field and the actual electron-electron interaction, where the solvent-
mediated electron-electron interaction at the same footing direct electron-electron interac-
tion. In the PTED approach the polarizable environment is self-consistently equilibrated
with the correlated MP2 density. Thereby some higher-order contributions are included.
These higher-order terms should be small since supposition underlying MP2 is that the dif-
ference between the Hartree-Fock and the MP2 wavefunction should be small — if this is
not fulfilled, MP2 is not adequate.
The PTE-COSMO-MP2 energy is obtained by adding to the COSMO-HF free energy the
MP2 correlation energy evaluated with the MOs and the Fock matrix from a COSMO-
HF calculation. The PTE-COSMO-MP2 approach is implemented in mpgrad for energies
and in ricc2 for energies, first-order properties and gradients with closed-shell RHF or
UHF reference wavefunctions. In ricc2 PTE-COSMO-MP2 calculations can be combined
with spin-component scaling (SCS or SOS) and for SOS with the O(N 4 )-scaling LT-based
implementation.
Current restrictions for PTE-COSMO-MP2 are:

• Not available for ROHF reference wavefunctions

• Not available for second-order properties (polarizabilities)

• Not available for vibrational frequencies (see Sec. 18.2 for the complications that arise
for the calculation of vibrational frequencies in solution)

The PTED approach in implemented in both mpgrad and ricc2 only for energies. No
gradients and no other properties are available in the two programs.

9.8 Low-scaling MP2 and MP2-F12 calculations with a

hybrid OSV-PNO approximation

For a PNO-MP2 calculation with default thresholds and standard basis sets the pnoccsd
program does not require any special input apart from $freeze and $maxcor and the input
needed for the Hartree-Fock calculation. It is, however, recommended to specify the PNO
truncation threshold in the data group $pnoccsd. For further details see Secs. 12 and
22.2.22.

Running pnoccsd parallel The MP2 part of the pnoccsd program is parallelized with
OMP for shared-memory and with MPI for distributed memory architectures. Important
242 CHAPTER 9. 2ND-ORDER MØLLER–PLESSET PERTURB. THEORY

for the performance of the parallel pnoccsd calculations are the settings for the core memory
usage and for the directories where large integral and scratch files are stored.
The keyword $maxcor defines for pnoccsd (as for ricc2 and ccsdf12) the core memory
usage. Note, however, that $maxcor defines only the memory controlled by the electronic
structure code. Additional memory can be allocated by the math and MPI libraries linked
into the program and by the operating and I/O systems. It is therefore recommended to
set $maxcor not higher than to 75% of the physical core memory that is available for the
calculation.
For MPI parallel pnoccsd calculations it is very important to set with the $tmpdir keyword
the path to directories in fast local file systems to avoid that large integral and scratch files
are stored in the NFS file system where the calculation is started.
Chapter 10

Second-Order Approximate
Coupled-Cluster (CC2)
Calculations

ricc2 is a module for the calculation of excitation energies and response properties at a
correlated second-order ab initio level, in particular the second-order approximate coupled-
cluster model CC2 [202], but also the MP2, CIS(D), CIS(D∞ ), and ADC(2) levels. All
calculations employ the resolution-of-the-identity (RI) approximation for the electron repul-
sion integrals used in the correlation treatment and the description of excitation processes.
At present the following functionalities are implemented:

ground state energies for MP2 and CC2 and spin-component scaled variants thereof; the
MP2 results are identical with those obtained with rimp2 (but usually the calculations
are somewhat faster).
excitation energies for the models CIS/CCS, CIS(D), CIS(D∞ ), ADC(2), and CC2 in-
cluding spin-component scaled SCS and SOS version of of the latter four methods
transition moments for ground state—excited and excited—excited state transitions for
the models CCS and CC2; for ADC(2) only moments for ground state—excited state
transitions are available
two-photon transition moments for ground state—excited state transitions for the mod-
els CCS and CC2
induced transition moments for ground state—excited state transitions for the models
CCS and CC2 for the computation of spin-orbit induced oscillator strengths for tran-
sitions from the ground state to excited triplet states and phosphorescence lifetimes
with SOC-PT
first-order properties for the ground state with SCF (CCS), MP2, and CC2 and for
excited states with CCS, CC2, ADC(2) and CIS(D∞ )

243
244 CHAPTER 10. RI-CC2

geometric gradients for the electronic ground state at the MP2 and the CC2 level; for
electronically excited states at the CIS(D∞ ), ADC(2), and CC2 level

second-order properties for the ground state with MP2 and CC2 and a closed-shell RHF
reference wavefunction (currently restricted to the sequentical and SMP parallel ver-
sions)

gradients for auxiliary basis sets for RI-MP2, -CC2, etc. calculations based on the RI-
MP2 error functional

F12 corrections to RI-MP2; MP2 ground-state energies can be computed (in C1 symme-
try) using explicitly-correlated two-electron basis functions in the framework of the
MP2-F12 model [200, 203].

solvent effects for the methods and states for which (orbital–relaxed) densities are avail-
able equilibrium solvent effects can be included in the framework of the cosmomode
(for details see Chapter 18.2).

All functionalities at the MP2 and CC2 level are implemented for closed-shell RHF and
open-shell UHF reference wavefunctions (with the exception of induced transition moments
using SOC-PT, which are only available for a closed-shell RHF reference). Ground state
energies for MP2, MP2-F12 and CC2 and excited state energies for CC2 are also implemented
for single determinant restricted open-shell Hartree-Fock (ROHF) reference wavefunctions
(cmp. Sec. 9.3). (Note, that no gradients are available for MP2 and CC2 with ROHF
reference wavefunctions.) For a two-component GHF reference wavefunction energies for
the CCS, MP2/ADC(2), CIS(D∞ ) and CC2 methods as well as ground state—excited state
transtition moments for ADC(2) and CC2 are available.
The second-order models MP2, CIS(D), CIS(D∞ ), ADC(2) and CC2 can be combined with
a spin-component scaling (SCS or SOS). (Not yet available for second-order properties,
two-photon and induced transition moments.) For the SOS variants one can switch to an
implementation with O(N 4 )-scaling costs by setting the keyword for the numerical Laplace
transformation (LT) ($laplace) .
As listed above, some functionalities are, as a side-produce, in ricc2 also implemented at
the uncorrelated HF-SCF, CIS, and CCS levels. There are only made available in ricc2 for
easier test calculations and comparisons, without that the code in ricc2 has optimized for
them.
For calculations with CCSD, CCSD(T) and other higher-order models beyond CC2 see
Chapter 11.

Prerequisites

Calculations with the ricc2 module require (almost) the same prerequisites as RI-MP2
calculations:
 245

1. a converged SCF calculation with the one-electron density convergence threshold set
to $denconv 1.d-5 or less

2. if non-standard basis sets used: an auxiliary basis defined in the data group $cbas
(for standard basis sets, where a corresponding auxiliary basis set is found in the basis
set library, the program will automatically use this if $cbas is not set)

3. if orbitals should be excluded from the correlation treatment (and excitation processes)
the data group $freeze has to be set

4. the maximum core memory which the program is allowed to allocate should be de-
fined in the data group $maxcor; the recommended value is 66–75% of the available
(physical) core memory.

5. depending on the type of calculations that should be carried out, additionally the
data groups $ricc2, $excitations, $response, $laplace, $rir12 and $lcg have to
be set (see below and Section 22.2.20).

For calculations with the ricc2 program it is recommended to use the cc2 submenu of the
define program to set the data groups $denconv, $freeze, $cbas, $maxcor. MP2-F12
calculations require in addition the data groups $rir12, $cabs, $jkbas and $lcg. The
exponent of the Slater function in the interelectronic distance r12 , which appears in the
geminals used MP2-F12 is defined in the data group $lcg and should be adapted to the
one-electron basis set which is used.
Note, that the implementation of non-Abelian point groups in ricc2 is limited to the elec-
tronic ground state (but comprises most of the RI-MP2 functionality included in ricc2).
In the present version ricc2 can for excited states only deal with real Abelian point groups
(C1 , Cs , C2 , Ci , C2h , C2v , D2 , D2h ). The F12 correction can only be calculated in the C1
point group.

How To Perform a Calculation

Single point calculations:
Call the ricc2 program after a converged SCF calculation, which can be carried
either with the dscf or the ridft program.

Geometry optimizations and molecular dynamics:

Invoke jobex with the -level cc2 option; see Section 5.1 for additional options
and parameters of the jobex script that might be needed or useful for geometry
optimizations and ab initio molecular dynamics calculations.

Force constants and vibrational frequencies:

Force constants can be calculated by numerical differentiation of the gradients.
Invoke for this NumForce with the -level cc2 option; see Chapter 15 for details
about NumForce. The usage of the NumForce interface for excited states is
restricted to C1 symmetry.
246 CHAPTER 10. RI-CC2

Note: using ricc2 in connection with jobex or NumForce requires that the method and
the electronic state, for which the gradient should be calculated and written to the interface
files, is specified in the option geoopt (see Section 10.3.1) in datagroup $ricc2 (see Sec-
tion 22.2.20). For calculations on excited states this state has in addition to be included in
the input for excitation energies in datagroup $excitations.
RI-SCF reference wavefunctions: The ricc2 program can be used in combination with
conventional SCF or with the RI-J and RI-JK approximations for SCF, with the exception
that the calculation of gradients for reference wavefunctions which employ only the RI-
J approximation for the Coulomb matrix but 4-index integrals for the exchange matrix
is presently not supported. The implementation of gradients in ricc2 assumes that the
reference wavefunction has either been calculated without RI-J approximation (using dscf)
or with the RI-JK approximation (using ridft).
See Chapter 6 for a discussion of the RI approximations in SCF calculations and 22.2.8 for
the required input. In geometry optimizations with jobex and for the calculation of force
constants and vibrational spectra with NumForce, the ricc2 program is used in combination
with the RI-JK approximation for the Hatree-Fock calculation (using ridft) if jobex and
NumForce are invoked with the -rijk option.

How to quote

If results obtained with the ricc2 program are used in publications, the following citations
should be included if you have used the methods, program parts, auxiliary basis sets, or
results reported in therein:

Methods:

• for the approximate coupled-cluster singles-and-doubles model CC2:

O. Christiansen, H. Koch, P. Jørgensen, Chem. Phys. Lett., 243 (1995) 409–418.
• for CI singles with a perturb. correct. for connected double excitations, CIS(D):
M. Head-Gordon, R. J. Rico, M. Oumi and T. J. Lee, Chem. Phys. Lett., 219
(1994) 21.
and for the iterative CIS(D∞ ) variant:
M. Head-Gordon, M. Oumi and D. Maurice, Mol. Phys. 96 (1999) 593.
• for the algebraic diagrammatic construction through second order ADC(2):
J. Schirmer, Phys. Rev. A. 26 (1981) 2395. A. B. Trofimov and J. Schirmer, J.
Phys. B. 28 (1995) 2299.
• for MP2-F12:
W. Klopper and C. C. M. Samson, J. Chem. Phys. 116 (2002) 6397–6410.
D. P. Tew and W. Klopper, J. Chem. Phys. 123 (2005) 074101.
• for the SCS and SOS variants of MP2:
S. Grimme, J. Chem. Phys. 118 (2003) 9095 (SCS) or Y., Jung, R.C. Lochan,
A.D. Dutoi, M. Head-Gordon, J. Chem. Phys. 121 (2004) 9793 (SOS).
 247

• for the SCS and SOS variants of CC2 and ADC(2):

A. Hellweg, S. Grün, C. Hättig, Phys. Chem. Chem. Phys. 10 (2008) 4119-
4127.
• for the two-component CCS, ADC(2), CIS(D∞ ) and CC2 methods:
K. Krause and W. Klopper, J. Chem. Phys. 142 (2015) 104109.

Implementation:

• please, include always a reference to the publication reporting the implementa-

tion of the core part of the ricc2 program:
C. Hättig and F. Weigend, J. Chem. Phys. 113 (2000) 5154.
• for transition moments and excited state first order properties:
C. Hättig and A. Köhn, J. Chem. Phys. 117 (2002) 6939.
• for triplet excited states include:
C. Hättig and K. Hald, Phys. Chem. Chem. Phys. 4 (2002) 2111. C. Hättig,
A. Köhn and K. Hald, J. Chem. Phys. 116 (2002) 5401.
• for ground state geometry optimizations include:
C. Hättig, J. Chem. Phys. 118 (2003) 7751.
• for geometry optimizations for excited states include:
A. Köhn and C. Hättig, J. Chem. Phys. 119 (2003) 5021.
• for calculations with RI-ADC(2), RI-CIS(D), RI-CIS(D∞ ) include:
C. Hättig, Adv. Quant. Chem. 50 (2005) 37.
• if the parallel version of ricc2 is used include a reference to:
C. Hättig, A. Hellweg, A. Köhn, Phys. Chem. Chem. Phys. 8 (2006) 1159.
• for transition moments between excited states:
M. Pabst and A. Köhn, J. Chem. Phys. 129 (2008) 214101.
• for RI-MP2-F12 calculations:
R. A. Bachorz, F. A. Bischoff, A. Glöß, C. Hättig, S. Höfener, W. Klopper,
D. P. Tew, J. Comput. Chem. 32 (2011) 2492.
• for O(N 4 )-scaling calculations using the Laplace transformation:

– ground-state and excitation energies:

N. O. C. Winter, C. Hättig, J. Chem. Phys. 134 (2011) 184101.
– transition moments, first-order properties and gradients:
N. O. C. Winter, C. Hättig, Chem. Phys. 401 (2012) 217.
• for second-order properties (relaxed or unrelaxed):
D. H. Friese, N. O. C. Winter, P. Balzerowski, R. Schwan, C. Hättig, J. Chem.
Phys. 136 (2012) 174106.
• for two-photon transition moments:
D. H. Friese, C. Hättig, K. Rudd, Phys. Chem. Chem. Phys. 14 (2012) 1175–
1184.
248 CHAPTER 10. RI-CC2

• for phosphorescence live times with SOC-PT-CC2:

B. Helmich-Paris, C. Hättig, C. van Wüllen, J. Chem. Theory Comput. 12
(2016) 1892–1904.

Auxiliary basis sets:

• the appropriate reference for the auxiliary SVP, TZVP and TZVPP basis sets
(for calculations with RI-MP2, RI-CC2 and related methods) is:
F. Weigend, M. Häser, H. Patzelt, R. Ahlrichs, Chem. Phys. Lett. 294 (1998)
143.
• for the auxiliary cc-pVXZ (cc-pV(X+d)Z), aug-cc-pVXZ (aug-cc-pV(X+d)Z)
basis sets with X = D, T, or Q cite:
F. Weigend, A. Köhn, C. Hättig, J. Chem. Phys. 116 (2001) 3175.
• for the auxiliary cc-pV5Z (cc-pV(5+d)Z), aug-cc-pV5Z (aug-cc-pV(5+d)Z), cc-
pwCVXZ with X = D, T, Q, 5 and QZVPP basis sets the reference is:
C. Hättig, Phys. Chem. Chem. Phys. 7 (2005) 59–66.
This reference should also be included if you employ the analytic basis set gra-
dients implemented in the ricc2 program for the optimization of your own
auxiliary basis set(s).
• for the auxiliary def2-basis sets from Rb to Rn the reference is:
A. Hellweg, C. Hättig, S. Höfener, and W. Klopper, Theor. Chem. Acc. 117
(2007) 587–597.

• for the auxiliary cc-pVXZ-PP, aug-cc-pVXZ-PP, cc-pwCVXZ-PP, and aug-cc-

pwCVXZ-PP basis sets for Ga–Kr, In–Xe, and Tl–Rn:
C. Hättig, G. Schmitz, J. Koßmann, Phys. Chem. Chem. Phys. 14 (2012)
6549–6555.

(For more details on the references for the basis sets included in the basis set libraries
of the TURBOMOLE distribution see Sec. 1.3 and the library files.)

10.1 CC2 Ground-State Energy Calculations

The CC2 ground-state energy is—similarly to other coupled-cluster energies—obtained from

the expression

ECC = hHF|H|CCi = hHF|H exp(T )|HFi , (10.1)

X h ij ih i
= ESCF + tab + tia tjb 2(ia|jb) − (ja|ib) , (10.2)
iajb

where the cluster operator T is expanded as T = T1 + T2 with

X
T1 = tia τai (10.3)
ai
1 X ij
T2 = tab τaibj (10.4)
2
aibj
10.1. CC2 GROUND-STATE ENERGY CALCULATIONS 249

(for a closed-shell case; in an open-shell case an additional spin summation has to be in-
cluded). The cluster amplitudes tia and tij ab are obtained as solution of the CC2 cluster
equations [202]:

Ωµ1 = hµ1 |Ĥ + [Ĥ, T2 ]|HFi = 0 , (10.5)

Ωµ2 = hµ2 |Ĥ + [F, T2 ]|HFi = 0 , (10.6)

with

Ĥ = exp(−T1 )H exp(T1 ),

where µ1 and µ2 denote, respectively, the sets of all singly and doubly excited determinants.
The residual of the cluster equations Ω(tai , taibj ) is the so-called vector function. The rec-
ommended reference for the CC2 model is ref. [202], the implementation with the resolution-
of-the-identity approximation, RI-CC2, was first described in ref. [11].

Advantages of the RI approximation: For RI-CC2 calculations, the operation

count and thereby the CPU and the wall time increases—as for RI-MP2 calculations—
approximately with O(O2 V 2 Nx ), where O is the number of occupied and V the number
of virtual orbitals and Nx the dimension of the auxiliary basis set for the resolution of the
identity. Since RI-CC2 calculations require the (iterative) solution of the cluster equations
(10.5) and (10.6), they are about 10–20 times more expensive than MP2 calculations. The
disk space requirements are approximately O(2V + N )Nx + Nx2 double precision words.
The details of the algorithms are described in ref. [11], for the error introduced by the RI
approximation see refs. [185, 204].

Required input data: In addition to the above mentioned prerequisites ground-state

energy calculations with the ricc2 module require only the data group $ricc2 (see Sec-
tion 22.2.20), which defines the methods, convergence thresholds and limits for the number
of iterations etc. If this data group is not set, the program will carry out a CC2 calculation.
With the input

$ricc2
mp2
cc2
conv=6

the ricc2 program will calculate the MP2 and CC2 ground-state energies, the latter con-
verged to approximately 10−6 a.u. The solution for the single-substitution cluster amplitudes
is saved in the file CCR0--1--1---0, which can be kept for a later restart.

Ground-State calculations for other methods than CC2: The MP2 equations
and the energy are obtained by restricting in the CC2 equations the single-substitution
250 CHAPTER 10. RI-CC2

amplitudes tai to zero. In this sense MP2 can be derived as a simplification of CC2. But it
should be noted that CC2 energies and geometries are usually not more accurate than MP2.
For CCS and CIS the double-substitution amplitudes are excluded from the cluster expansion
and the single-substitution amplitudes for the ground state wavefunction are zero for closed–
shell RHF and open–shell UHF reference wavefunctions and thus energy is identical to the
SCF energy.
For the Methods CIS(D), CIS(D∞ ) and ADC(2) the ground state is identified with the MP2
ground state to define is total energy of the excited state, which is needed for the definition
of gradients and (relaxed) first-order properties which are obtained as (analytic) derivatives
the total energy.

Diagnostics: Together with the MP2 and/or CC2 ground state energy the program
evaluates the D1 diagnostic proposed by Janssen and Nielsen [186], which is defined as:
s  hX i hX i
D1 = max λmax tai tbi , λmax tai taj (10.7)
i a

where λmax [M] is the largest eigenvalue of a positive definite matrix M. (For CC2 the
D1 diagnostic will be computed automatically. For MP2 is must explictly be requested
with the d1diag option in the $ricc2 data group, since for RI-MP2 the calculation of D1
will contribute significantly to the computational costs.) Large values of D1 indicate a
multireference character of the ground-state introduced by strong orbital relaxation effects.
In difference to the T1 and S2 diagnostics proposed earlier by Lee and coworkers, the D1
diagnostic is strictly size-intensive and can thus be used also for large systems and to compare
results for molecules of different size. MP2 and CC2 results for geometries and vibrational
frequencies are, in general, in excellent agreement with those of higher-order correlation
methods if, respectively, D1 (MP2) ≤ 0.015 and D1 (CC2) ≤ 0.030 [14, 186]. For D1 (MP2) ≤
0.040 and D1 (CC2) ≤ 0.050 MP2 and/or CC2 usually still perform well, but results should
be carefully checked. Larger values of D1 indicate that MP2 and CC2 are inadequate to
describe the ground state of the system correctly!
The D2 diagnostic proposed by Nielsen and Janssen [187] can also be evaluated. This
analysis can be triggered, whenever a response property is calculated, e.g. dipole moment,
with the keyword $D2-diagnostic. Note that the calculation of D2 requires an additional
O(N 5 ) step! D2 (MP2/CC2) ≤ 0.15 are in excellent agreement with those of higher-order
correlation methods, for D2 (MP2/CC2) ≥ 0.18 the results should be carefully checked.

10.2 Calculation of Excitation Energies

With the ricc2 program excitation energies can presently be calculated with the RI variants
of the methods CCS/CIS, CIS(D), CIS(D∞ ), ADC(2) and CC2. The CC2 excitation ener-
gies are obtained by standard coupled-cluster linear response theory as eigenvalues of the
10.2. CALCULATION OF EXCITATION ENERGIES 251

Jacobian, defined as derivative of the vector function with respect to the cluster amplitudes.
!
dΩ µ hµ 1 |[[Ĥ + [ Ĥ, T2 ], τν ]|HFi hµ 1 |[Ĥ, τν ]|HFi
ACC2
µν = = 1 2
(10.8)
dtν hµ2 |[Ĥ, τν1 ]|HFi hµ2 |[F, τν2 ]|HFi

Since the CC2 Jacobian is a non-symmetric matrix, left and right eigenvectors are different
and the right (left) eigenvectors Eνi (Ēµi ) are not orthogonal among themselves, but form a
biorthonormal basis (if properly normalized):

Ē i E j = Ēµi 1 Eνj1 + Ēµi 2 Eνj2 = δij . (10.9)

To obtain excitation energies only the right or the left eigenvalue problem needs to be
solved, but for the calculation of transition strengths and first-order properties both, left
and right, eigenvectors are needed (see below). A second complication that arises from
the non-symmetric eigenvalue problem is that in the case of close degeneracies within the
same irreducible representation (symmetry) it can happen that instead of two close lying
real roots a degenerate complex conjugated pair of excitation energies and eigenvectors is
obtained. CC2 (and also other standard coupled-cluster response methods) are thus not
suited for the description of conical intersections etc. For the general theory behind coupled
cluster response calculations see e.g. ref. [205, 206] or other reviews.
The ricc2 program exploits that the doubles/doubles block of the CC2 Jacobian is diagonal
and the (linear) eigenvalue problem in the singles and doubles space can be reformulated as
a (non-linear) eigenvalue problem in single-substitution space only:

Aef f
µ1 ν1 (t, ω) = ACC2 CC2 CC2
µ1 ν1 (t) − Aµ1 γ2 (t)(Aγ2 γ2 − ω)Aγ2ν1 (t)

Aef f
µ1 ν1 (t
CC2
, ω CC2 )Eν1 = ω CC2 Eν1

This allows to avoid the storage of the double-substitution part of the eigen- or excitation
vectors Eν2 , Ēν2 . The algorithms are described in refs. [11, 12], about the RI error see
ref. [204].
The solution of the CC2 eigenvalue problem can be started from the solutions of the CCS
eigenvalue problem (see below) or the trial vectors or solutions of a previous CC2 excita-
tion energy calculation. The operation count per transformed trial vector for one iteration
for the CC2 eigenvalue problem is about 1.3–1.7 times the operation count for one itera-
tion for the cluster equations in the ground-state calculation—depending on the number of
vectors transformed simultaneously. The disk space requirements are about O(V + N )Nx
double precision words per vector in addition to the disk space required for the ground state
calculation.
CCS excitation energies are obtained by the same approach, but here double-substitutions
are excluded from the expansion of the excitation or eigenvectors and the ground-state
amplitudes are zero. Therefore the CCS Jacobian,
dΩµ
ACCS
µν = = hµ1 |[H, τν1 ]|HFi , (10.10)
dtν
is a symmetric matrix and left and right eigenvectors are identical and form an orthonormal
basis. The configuration interaction singles (CIS) excitation energies are identical to the
252 CHAPTER 10. RI-CC2

CCS excitation energies. The operation count for a RI-CIS calculation is O(ON 2 Nx ) per
iteration and transformed trial vector.
The second-order perturbative correction CIS(D) to the CIS excitation energies is calculated
from the expression

ω CIS(D) = ω CIS + ω (D) = ECIS Aef f (tMP1 , ω CIS )ECIS (10.11)

(Note that tMP1 are the first-order double-substitution amplitudes from which also the MP2
ground-state energy is calculated; the first-order single-substitution amplitudes vanish for a
Hartree–Fock reference due to the Brillouin theorem.) The operation count for a RI-CIS(D)
calculation is similar to that of a single iteration for the CC2 eigenvalue problem. Also disk
space requirements are similar.

Running excitation energy calculations: The calculation of excitation energies

is initiated by the data group $excitations in which at least the symmetries (irreducible
representations) and the number of the excited states must be given (for other options see
Section 22.2.20). With the following input the ricc2 program will calculate the lowest two
roots (states) for the symmetries A1 and B1 of singlet multiplicity ∗ at the CIS, CIS(D) and
CC2 level with default convergence thresholds. Ground-state calculations will be carried
out for MP2 (needed for the CIS(D) model and used as start guess for CC2) and CC2.

$ricc2
cis
cis(d)
cc2
$excitations
irrep=a1 nexc=2
irrep=b1 nexc=2

The single-substitution parts of the right eigenvectors are stored in files named CCRE0-s--m-xxx,
where s is the number of the symmetry class (irreducible representation), m is the multiplic-
ity, and xxx the number of the excitation within the symmetry class. For the left eigenvectors
the single-substitution parts are stored in files named CCLE0-s--m-xxx. These files can be
kept for later restarts.

Trouble shooting: For the iterative second-order methods CIS(D∞ ), ADC(2), and CC2
the solution of the nonlinear partitioned eigenvalue problem proceeds usually in three steps:

1. solution of the CCS/CIS eigenvalue problem to generate reasonable start vectors; the
eigenvectors are converged in this step only to a remaining residual norm < preopt
∗
Provided that it is not an unrestricted open shell run. In this case the wavefunctions will not
be spin eigenfunctions and multiplicities are not well defined.
10.2. CALCULATION OF EXCITATION ENERGIES 253

2. pre-optimization of the eigenvectors by a robust modified Davidson algorithm (see

ref. [11]) using the LINEAR CC RESPONSE SOLVER until the norm of all residuals are
below preopt, combined with a DIIS extrapolation for roots assumed to be converged
below the threshold thrdiis.
3. solution of the nonlinear eigenvalue problem with a DIIS algorithm using the DIIS
CC RESPONSE SOLVER until the norm of the residuals are below the required threshold
conv

This procedure is usually fairly stable and efficient with the default values for the thresholds.
But for difficult cases it can be necessary to select tighter thresholds. In case of convergence
problems the first thing do is to verify that the ground state is not a multireference case
by checking the D1 diagnostic. If this is not the case the following situations can cause
problems in the calculation of excitation energies:

• almost degenerate roots in the same symmetry class

• complex roots (break down of the CC approximation close to conical intersections)
• large contributions from double excitations

The first two reasons can be identified by running the program with a print level ≤ 3. It
will then print in each iteration the actual estimates for the eigenvalues. If some of these are
very close or if complex roots appear, you should make sure that the DIIS procedure is not
switched on before the residuals of the eigenvectors are small compared to the differences in
the eigenvalues. For this, thrdiis (controlling the DIIS extrapolation in the linear solver)
should be set about one order of magnitude smaller than the smallest difference between two
eigenvalues and preopt (controlling the switch to the DIIS solver) again about one order of
magnitude smaller then thrdiis.
Tighter thresholds or difficult situations can make it necessary to increase the limit for the
number of iterations maxiter.
In rare cases complex roots might persist even with tight convergence thresholds. This can
happen for CC2 and CIS(D∞ ) close to conical intersections between two states of the same
symmetry, where CC response can fail due to its non-symmetric Jacobian. In this case
one can try to use instead the ADC(2) model. But the nonlinear partitioned form of the
eigenvalue problem used in the ricc2 program is not well suited to deal with such situations.

Diagnostics for double excitations: As pointed out in ref. [13], the %T1 diagnostic
(or %T2 = 100 − %T1 ) which is evaluated directly from the squared norm of the single and
double excitation part of the eigenvectors %T1 = 100·T1 /(T1 +T2 ) with Ti = µi Eµ2i where
P

the excitation amplitudes are for spin-free calculations in a corresponding spin-adapted basis
(which is not necessarily normalized) has the disadvantage that the results depend on the
parameterization of the (spin-adapted) excitation operators. This prevents in particular
a simple comparison of the results for singlet and triplet excited states if the calculations
are carried out in a spin-free basis. With the biorthogonal representation for singlet spin-
coupled double excitations [205] results for %T1 also differ largely between the left and right
254 CHAPTER 10. RI-CC2

eigenvectors and are not invariant with respect to unitary transformations of the occupied
or the virtual orbitals.
The ricc2 module therefore uses since release 6.5 an alternative double excitation diagnostic,
2 2
P P P
which is defined by %T1 = 100 ∗ T1 /(T1 + T2 ) with T1 = ai Eai and T2 = i>j a>b Eaibj
with Eai and Eaibj in the spin-orbital basis. They are printed in the summaries for excitation
energies under the headings %t1 and %t2. For spin-adapted excitation amplitudes T1 and T2
have to be computed from respective linear combinations for the amplitudes which reproduce
the values in the spin-orbital basis. For ADC(2), which has a symmetric secular matrix with
identical left and right normalized eigenvectors T1 and T2 are identical with the contributions
from the singles and doubles parts for the eigenvectors to the trace of the occupied or virtual
block of the (orbital unrelaxed) difference density between the ground and the excited state,
i.e. the criterium proposed in ref. [13]. Compared to the suggestion from ref. [13] T1 and T2
have the additional advantage of that they are for all methods guaranteed to be positive and
can be evaluated with the same insignificantly low costs as T1 and T2 . They are invariant
with respect to unitary transformations of the occupied or the virtual orbitals and give
by construction identical results in spin-orbital and spin-free calculations. For CC2 and
CIS(D∞ ) the diagnostics T1 and T2 agree for left and right eigenvectors usually with a few
0.01%, for CIS(D) and ADC(2) they are exactly identical. For singlet excitations in spin-
free calculations, %T2 is typically by a factors of 1.5–2 larger than %T2 . The second-order
methods CC2, ADC(2), CIS(D∞ ) and CIS(D) can usually be trusted for %T2 ≤ 15%.
For compatibility, the program can be switched to use of the old %T1 and %T2 diagnostics
(printed with the headers ||T1|| and ||T2||) by setting the flag oldnorm in the data group
$excitations. Note that the choice of the norm effects the individual results left and right
one- and two-photon transition moments, while transition strengths and all other observable
properties independent of the individual normalization of the right and left eigenvectors.
The %T2 and %T2 diagnostics can not be monitored in the output of the (quasi-) linear
solver. But it is possible to do in advance a CIS(D) calculation. The CIS(D) results for
the %T2 and %T2 correlate usually well with the results for this diagnostic from the iterativ
second-order models, as long as there is clear correspondence between the singles parts of the
eigenvectors. Else the DIIS solver will print the doubles diagnostics in each iteration if the
print level is set > 3. States with large double excitation contributions converge notoriously
slow (a consequence of the partitioned formulation used in the ricc2 program). However,
the results obtained with second-order methods for doubly excited states will anyway be
poor. It is strongly recommended to use in such situations a higher-level method.

Visualization of excitations: An easy way to visualize single excitations is to plot

the natural transition orbitals that can be obtained from a singular value decomposition of
the excitation amplitudes. See Sec. 19.1.6 for further details.
Another, but computational more involved possibility is plot the difference density between
the ground and the respective excited state. This requires, however, a first-order property
or gradient calculation for the excited state to obtain the difference density. For further
details see Sec. 10.3.3.
10.3. FIRST-ORDER PROPERTIES AND GRADIENTS 255

10.3 First-Order Properties and Gradients

For the ground state first-order properties (expectation values) are implemented at the SCF,
MP2 and CC2 level. Note that for the ground state CCS and CIS are equivalent to SCF.
For excited states first-order properties are implemented only at the CCS and CC2 level.
Gradients are presently only available for the ground state at the MP2 and the CC2 and for
excited states only at the CC2 level.

10.3.1 Ground State Properties, Gradients and Geometries

For CC2, one distinguishes between orbital-relaxed and unrelaxed properties. Both are
calculated as first derivatives of the respective energy with respect to an external field
corresponding to the calculated property. They differ in the treatment of the SCF orbitals.
In the orbital-relaxed case the external field is (formally) already included at the SCF stage
and the orbitals are allowed to relax in the external field; in the orbital-unrelaxed case the
external field is first applied after the SCF calculation and the orbitals do not respond to
the external field. Orbital-unrelaxed CC2 properties are calculated as first derivatives of the
real part of the unrelaxed Lagrangian [202]
X
Lur CC2 (t, t̄, β) = hHF|H|CCi + t̄µ1 hµ1 |Ĥ + [Ĥ, T2 ]|HFi (10.12)
µ1
X
+ t̄µ2 hµ2 |Ĥ + [F0 + β V̂ , T2 ]|HFi
µ2

with H = H0 + βV —where V is the (one-electron) operator describing the external field, β

the field strength, and H0 and F0 are the Hamiltonian and Fock operators of the unperturbed
system—by the expression:
 ur CC2 
ur CC2 ∂L (t, t̄, β) X
ur
hV i = < = Dpq Vpq , (10.13)
∂β 0 pq
 X
= < hHF|V̂ |HFi + t̄µ1 hµ1 |V̂ + [V, T2 ]|HFi (10.14)
µ1
X 
+ t̄µ2 hµ2 |[V̂ , T2 ]|HFi ,
µ2

where < indicates that the real part is taken. Relaxed CC2 properties (and gradients) are
calculated from the the full variational density including the contributions from the orbital
response to the external perturbation, which are derived from the Lagrangian [14, 206]
X
Lrel CC2 (t, t̄) = hHF|H|CCi + t̄µ1 hµ1 |Ĥ + [Ĥ, T2 ]|HFi (10.15)
µ1
X X
+ t̄µ2 hµ2 |Ĥ + [F, T2 ]|HFi + κ̄µ0 Fµ0 ,
µ2 µ0
256 CHAPTER 10. RI-CC2

where F is the Fock operator corresponding to the Hamiltonian of the perturbed system
H = H0 + βV . One-electron properties are then obtained as:
 X
hV irel CC2 = < hHF|V̂ |HFi + t̄µ1 hµ1 |V̂ + [V, T2 ]|HFi (10.16)
µ1
X X 
+ t̄µ2 hµ2 |[V, T2 ]|HFi + κ̄µ0 Vµ0 ,
µ2 µ0
X
rel
= Dpq Vpq . (10.17)
pq

The calculation of one-electron first-order properties requires that in addition to the clus-
ter equations also the linear equations for the Lagrangian multipliers t̄µ are solved, which
requires similar resources (CPU, disk space, and memory) as the calculation of a single
excitation energy. For orbital-relaxed properties also a CPHF-like linear equation for the
Lagrangian multipliers κ̄µ0 needs to be solved and the two-electron density has to be build,
since it is needed to set up the inhomogeneity (right-hand side). The calculation of re-
laxed properties is therefore somewhat more expensive—the operation count for solving
the so-called Z-vector equations is similar to what is needed for an SCF calculation—and
requires also more disk space to keep intermediates for the two-electron density—about
O(2V + 2N )Nx + Nx2 in addition to what is needed for the solution of the cluster equations.
For ground states, orbital-relaxed first-order properties are standard in the literature.
The calculation of the gradient implies the calculation of the same variational densities as
needed for relaxed one-electron properties and the solution of the same equations. The
construction of the gradient contributions from the densities and derivative integrals takes
about the same CPU time as 3–4 SCF iterations and only minor extra disk space. For details
of the implementation of CC2 relaxed first-order properties and gradients and a discussion
of applicability and trends of CC2 ground-state equilibrium geometries see ref. [14]. The
following is in example input for a MP2 and CC2 single point calculation of first-order
properties and gradients:

$ricc2
mp2
cc2
$response
static relaxed operators=diplen,qudlen
gradient

A different input is required for geometry optimizations: in this case the model for which the
geometry should be optimized must be specified in the data group $ricc2 by the keyword
geoopt:

$ricc2
mp2
cc2
geoopt model=cc2
10.3. FIRST-ORDER PROPERTIES AND GRADIENTS 257

For CC2 calculations, the single-substitution part of the Lagrangian multipliers t̄µ are saved
in the file CCL0--1--1---0 and can be kept for a restart (for MP2 and CCS, the single-
substitution part t̄µ vanishes).
For MP2 only relaxed first-order properties and gradients are implemented (unrelaxed MP2
properties are defined differently than in CC response theory and are not implemented). For
MP2, only the CPHF-like Z-vector equations for κ̄µ0 need to be solved, no equations have
to be solved for the Lagrangian multipliers t̄µ . CPU time and disk space requirements are
thus somewhat smaller than for CC2 properties or gradients.
For SCF/CIS/CCS it is recommended to use the modules grad and rdgrad for the calcula-
tion of, ground state gradients and first-order properties.

10.3.2 Excited State Properties, Gradients and Geometries

Also for excited states presently unrelaxed and relaxed first-order properties are available in
the ricc2 program. These are implemented for CCS and CC2. Note, that in the unrelaxed
case CIS and CCS are not equivalent for excited-states first-order properties and no first-
order properties are implemented for CIS in the ricc2 program.

Orbital-unrelaxed first-order properties

The unrelaxed first-order properties are calculated from the variational excited states La-
grangian [207], which for the calculation of unrelaxed properties is composed of the unrelaxed
ground state Lagrangian, Eq. (10.12), and the expression for the excitation energy:
X
Lur CC2, ex (E, Ē, t, t̄(ex) , β) = hHF|H|CCi + Ēµ Aµν (t, β)Eν (10.18)
µν
X
+ t̄(ex)
µ1 hµ1 |Ĥ + [Ĥ, T2 ]|HFi
µ1
X
+ t̄(ex)
µ2 hµ2 |Ĥ + [F0 + β V̂ , T2 ]|HFi
µ2

P
where it is assumed that the left and right eigenvectors are normalized such that µν Ēµ hµ|τν iEν =
1 and H = H0 + βV . The first-order properties are calculated as first derivatives of
Lur CC2, ex (E, Ē, t, t̄(ex) , β) with respect to the field strength β and are evaluated via a den-
sity formalism:
 ur,ex
(E, Ē, t, t̄(ex) , β)

ur,ex ∂L X
ur,ex
hV i = R = Dpq Vpq , (10.19)
∂β 0 pq

(Again < indicates that the real part is taken.) The unrelaxed excited-state properties
obtained thereby are related in the same way to the total energy of the excited states as
the unrelaxed ground-state properties to the energy of the ground state and the differences
between excited- and ground-state unrelaxed properties are identical to those identified
from the second residues of the quadratic response function. For a detailed description of
258 CHAPTER 10. RI-CC2

the theory see refs. [206, 207]; the algorithms for the RI-CC2 implementation are described
in refs. [13, 204]. ref. [204] also contains a discussion of the basis set effects and the errors
introduced by the RI approximation.
The calculation of excited-state first-order properties thus requires the calculation of both
the right (Eµ ) and left (Ēµ ) eigenvectors and of the excited state Lagrangian multipliers
(ex) (ex)
t̄µ . The disk space and CPU requirements for solving the equations for Ēµ and t̄µ are
about the same as those for the calculation of the excitation energies. For the construction
of the density matrices in addition some files with O(nroot N 2 ) size are written, where nroot
is the number of excited states.
(ex)
The single-substitution parts of the excited-states Lagrangian multipliers t̄µ are saved in
files named CCNL0-s--m-xxx.
For the calculation of first-order properties for excited states, the keyword exprop must be
added with appropriate options to the data group $excitations; else the input is same as
for the calculation of excitation energies:

$ricc2
cc2
$response
fop unrelaxed_only operators=diplen,qudlen
$excitations
irrep=a1 nexc=2
exprop states=all operators=diplen,qudlen

Orbital-relaxed first-order properties and gradients

To obtain orbital-relaxed first-order properties or analytic derivatives (gradients) the La-

grange functional for the excited state in Eq. (10.18) is—analogously to the treatment of
ground states—augmented by the equations for the SCF orbitals and the perturbations is
also included in the Fock operator:
X
Lrel CC2, ex (E, Ē, t, t̄(ex) , β) = hHF|H|CCi + Ēµ Aµν (t, β)Eν (10.20)
µν
X
+ t̄(ex)
µ1 hµ1 |Ĥ + [Ĥ, T2 ]|HFi
µ1
X X
+ t̄(ex)
µ2 hµ2 |Ĥ + [F, T2 ]|HFi + κ̄(ex)
µ0 Fµ0 .
µ2 µ0

Compared to unrelaxed properties, the calculation of relaxed properties needs in addition

(ex)
for each excited state the solution of a CPHF equations for the Lagrangian multipliers κ̄µ0 ,
for which the computational costs are similar to those of a Hartree-Fock calculation.
Orbital-relaxed properties are requested by adding the flag relaxed to the input line for
the exprop option. The following is an example for a CC2 single point calculation for
orbital-relaxed excited state properties:
10.3. FIRST-ORDER PROPERTIES AND GRADIENTS 259

$ricc2
cc2
$excitations
irrep=a1 nexc=2
exprop states=all relaxed operators=diplen,qudlen

Note that during the calculation of orbital-relaxed excited-state properties the corresponding
unrelaxed properties are also automatically evaluated at essentially no additional costs.
Therefore, the calculation of unrelaxed properties can not be switched off when relaxed
properties have been requested.
Again the construction of gradients requires the same variational densities as needed for
relaxed one-electron properties and the solution of the same equations. The construction of
the gradient contributions from one- and two-electron densities and derivative integrals takes
approximately the same time as for ground states gradients (approx. 3–4 SCF iterations)
and only minor extra disk space. The implementation of the excited state gradients for the
RI-CC2 approach is described in detail in Ref. [15]. There one can also find some information
about the performance of CC2 for structures and vibrational frequencies of excited states.
For the calculation of an excited state gradient with CC2 at a single point (without geometry
optimization and if it is not a calculation with NumForce) one can use the input:

$ricc2
cc2
$excitations
irrep=a1 nexc=2
xgrad states=(a1 1-2)

For geometry optimizations or a numerical calculation of the Hessian with NumForce the
wavefunction model and the excited state for which the geometry should be optimized have
to be specified in the data group $ricc2 with the keyword geoopt:

$ricc2
geoopt model=cc2 state=(a1 2)
$excitations
irrep=a1 nexc=2

If the geometry optimization should carried out for the lowest excited state (of those
for which an excitation energy is requested in $excitation), one can use alternatively
state=(s1).
Since the calculation of unrelaxed and relaxed first-order properties can be combined gra-
dient calculations without significant extra costs, a request for excited state gradients will
automatically enforce the calculation of the relaxed and unrelaxed dipole moments. If the
keyword geoopt is used, the relaxed dipole moment for the specified excited state and wave-
function model will be written to the control file and used in calculations with NumForce
for the evaluation of the IR intensities.
260 CHAPTER 10. RI-CC2

10.3.3 Visualization of densities and Density analysis

As most other programs which allow for the calculation of wavefunctions and densities also
the ricc2 module is interfaced to wavefunction analysis and visualization toolbox described
in chapter 19. From ricc2 module this interface can used in two different ways

1. If through the geoopt keyword in $ricc2 a unique method and state has been specified
for which the density, gradient and properties are evaluated, the density analysis and
visualization routines will called by default with the (orbital-relaxed) density for this
state and method similar as in dscf, ridft, mpgrad, etc.

2. The ricc2 program can be called in a special analysis mode which allows to analyse
densities and combination (e.g. differences) of densities evaluated in preceeding ricc2
calculations.

Default density analysis and visualization:

As in a single calculations with the ricc2 program one–electron densities can be calculated
for more than one method and/or electronic state, the interface to the analysis and visual-
ization routines require the specification of a unique level of calculation and a unique state.
This is presently done through the geoopt flag which determines the method/state for which
results are written to interface files (e.g. control, gradient, or xxx.map).
In ground state calculations ricc2 will pass to the density analysis routines the correlated
total (and for UHF based calculations also the spin) density and the canonical SCF orbitals
from which the SCF (spin) density is constructed. All options described in chapter 19 are
available from within the ricc2 program apart from the evaluation of electrostatic moments,
which would interfere with the calculation of expectation values requested through the fop
option in $response.
In excited state calculation ricc2 will pass the excited state total (and for UHF based
calculation in addition the spin) density. But no ground state densities and/or uncorrelated
densities or orbitals. Thus, for excited states the ricc2 program does, in difference to
egrad not print out a comparison with the ground state SCF density. Also, all some
options which require orbitals (as e.g. the generation and visualization of localized orbitals
or some population analysis options) and not available for excited states in ricc2.
As other modules, also ricc2 provides the -proper flag to bypass a re-calculation of the
density and gradient to enter immediately the density analysis routines with a previously
calculated density. The ricc2 program will then pass the densities found on the interface
file for the density analysis routines without further check on the method and state for which
they have been evaluated. If both, ground and excited state densities are found on file, both
will be passed to the density analysis, thereby providing a shortcut to the -fanal and the
$anadens keyword for the analysis of differences between ground and excited state densities.

The general density analysis option:

In general ricc2 saves by default all relaxed densities generated during a calculation in
10.3. FIRST-ORDER PROPERTIES AND GRADIENTS 261

files named <method>-<type>-<mult><irrep>-<number>-total.cao <method> denotes the

level of theory, like mp2, cc2, or adcp2. <type> is one of gsdn (ground state) or xsdn (excited
state) the other entries specify multiplicity, irreducible representation and the number of the
state. Having specified the calculation of relaxed densities—e.g. by requesting relaxed one-
electron properties or as a by-product of a gradient calculation—you will end up with two
files named like

cc2-gsdn-1a1-001-total.cao
cc2-xsdn-3a2-001-total.cao

In case of open shell molecules, additional files with names ...-spndn.cao (for one-electron
spin-densities) will be generated.
These files are (currently) in a binary format, similar as the files dens, mdens and edens.
Therefore be aware that a transfer between different computer architectures may result in
trouble.
The densities on these files can be analysed with the tools and interfaces provided by Moloch
(see Section 19.2). This can be done by calling ricc2 with the option -fanal which by-
passes the usual wavefunction calculation and triggers the program into an analysis mode
for densities. In this mode the program interpretes $anadens and the keywords described in
Section 19.2. To plot, for example, the difference density of the two above mentioned total
densities you have to add the following lines in your control file

$anadens
calc my_favourite_diffden from
1d0 cc1td-cc2-xs-3a2-001
-1d0 cc1td-cc2-gs-1a1-001
$pointval

and invoke

ricc2 -fanal

This will generate the files my_favourite_diffden and my_favourite_diffden.map. The

latter can be converted into gOpenMol format as described in Section 19.2.

10.3.4 Fast geometry optimizations with RI-SCF based gradients

If geometry optimizations on MP2 or CC2 level are performed with large basis set, especially
with diffuse basis functions, the N 4 –steps might become the dominant part of the overall
timings. In these cases, the integral screening in the Hartree–Fock part often becomes
inefficient. The resolution–of–the–identity can be applied here to speed up the calculation
of the HF reference wavefunction, as well as the solution of the coupled–perturbed Hartree–
Fock (CPHF) equations in the MP2 or CC2 gradient calculation.
262 CHAPTER 10. RI-CC2

An additional auxiliary basis (denoted jkbas) set has to be assigned via the General Options
Menu in the define program. In the submenu rijk choose on and select your auxiliary basis
set. Then, run the jobex script the additional rijk-flag:

> jobex -level cc2 -rijk

10.4 Transition Moments

Transition moments (for one-photon transitions) are presently implemented for excitations
out of the ground state and for excitations between excited states for the coupled cluster
models CCS and CC2. Transition moments for excitations from the ground to an excited
state are also available for ADC(2), but use an additional approximation (see below). Note,
that for transition moments (as for excited-state first-order properties) CCS is not equivalent
to CIS and CIS transition moments are not implemented in the ricc2 program.
Two-photon transition moments are only available for CC2.

10.4.1 Ground to excited state transition moments

In response theory, transition strengths (and moments) for transitions from the ground to
excited state are identified from the first residues of the response functions. Due to the
non-variational structure of coupled cluster different expressions are obtained for the CCS
V
and CC2 “left” and “right” transitions moments M0←f and MfV←0 . The transition strengths
0f
SV1 V2 are obtained as a symmetrized combinations of both [208]:

1 n V1  ∗ o
SV0f1 V2 = M0←f MfV←0
2 V2
+ M0←f MfV←0
1
(10.21)
2

Note, that only the transition strengths SV0f1 V2 are a well-defined observables but not the
V
transition moments M0←f and MfV←0 . For a review of the theory see refs. [206, 208]. The
transition strengths calculated by coupled-cluster response theory according to Eq. (10.21)
have the same symmetry with respect to an interchange of the operators V1 and V2 and with
respect to complex conjugation as the exact transition moments. In difference to SCF (RPA),
(TD)DFT, or FCI, transition strengths calculated by the coupled-cluster response models
CCS, CC2, etc. do not become gauge-independent in the limit of a complete basis set, i.e.,
for example the dipole oscillator strength calculated in the length, velocity or acceleration
gauge remain different until also the full coupled-cluster (equivalent to the full CI) limit is
reached.
For a description of the implementation in the ricc2 program see refs. [14, 204]. The
calculation of transition moments for excitations out of the ground state resembles the
calculation of first-order properties for excited states: In addition to the left and right
eigenvectors, a set of transition Lagrangian multipliers M̄µ has to be determined and some
transition density matrices have to be constructed. Disk space, core memory and CPU time
requirements are thus also similar.
10.4. TRANSITION MOMENTS 263

The single-substitution parts of the transition Lagrangian multipliers N̄µ are saved in files
named CCME0-s--m-xxx.
To obtain the transition strengths for excitations out of the ground state the keyword
spectrum must be added with appropriate options (see Section 22.2.20) to the data group
$excitations; else the input is same as for the calculation of excitation energies and first-
order properties:

$ricc2
cc2
$excitations
irrep=a1 nexc=2
spectrum states=all operators=diplen,qudlen

For the ADC(2) model, which is derived by a perturbation expansion of the expressions for
exact states, the calculation of transition moments for excitations from the ground to an
excited state would require the second-order double excitation amplitudes for the ground
state wavefunction, which would lead to operation counts scaling as O(N 6 ), if no further
approximations are introduced. On the other hand the second-order contributions to the
transition moments are usually not expected to be important. Therefore, the implementa-
tion in the ricc2 program neglects in the calculation of the ground to excited state transition
moments the contributions which are second order in ground state amplitudes (i.e. contain
second-order amplitudes or products of first-order amplitudes). With this approximation
the ADC(2) transition moments are only correct to first-order, i.e. to the same order to
which also the CC2 transition moments are correct, and are typically similar to the CC2
results. The computational costs for the ADC(2) transition moments are (within this ap-
proximation) much lower than for CC2 since the left and right eigenvectors are identical
and no Lagrangian multipliers need to be determined. The extra costs (i.e. CPU and wall
time) for the calculations of the transitions moments are similar to the those for two or
three iterations of the eigenvalue problem, which reduces the total CPU and wall time for
the calculation of a spectrum (i.e. excitation energies and transition moments) by almost a
factor of three.

10.4.2 Transition moments between excited states

For the calculation of transition moments between excited states a set of Lagrangian multi-
pliers N̄µ has to be determined instead of the M̄µ for the ground state transition moments.
From these Lagrangian multipliers and the left and right eigenvectors one obtaines the “right”
transition moment between two excited states i and f as
X
MfV←i = ξ
Dpq (N̄ f i ) + Dpq
A
(Ē f , E i ) V̂pq . (10.22)
pq

where V̂ are the matrix elements of the perturbing operator. A similar expression is ob-
tained for the “left” transition moments. The “left” and “right” transition moments are then
264 CHAPTER 10. RI-CC2

combined to yield the transition strength

1 n V1  ∗ o
SVif1 V2 = Mi←f MfV←i
2 V2
+ Mi←f MfV←i
1
(10.23)
2

As for the ground state transitions, only the transition strengths SVif1 V2 are a well-defined
V
observables but not the transition moments Mi←f and MfV←i .
The single-substitution parts of the transition Lagrangian multipliers N̄µ are saved in files
named CCNE0-s--m-xxx.
To obtain the transition strengths for excitations between excited states the keyword tmexc
must be added to the data group $excitations. Additionally, the initial and final states
must be given in the same line; else the input is same as for the calculation of excitation
energies and first-order properties:

$ricc2
cc2
$excitations
irrep=a1 nexc=2
irrep=a2 nexc=2
tmexc istates=all fstates=(a1 1) operators=diplen

10.4.3 Ground to excited state two-photon transition moments

For closed-shell restricted and high-spin open-shell unrestricted Hartree-Fock reference states
two-photon transition moments for one-electron operators can be computed at the CCS
and the CC2 level. The implementation is restricted to real Abelian point groups (D2h
and its subgroups) and for CC2 without spin-component scaling. The response theory for
the calculation of two-photon transition moments with coupled cluster methods has been
described in [209, 210], the RI-CC2 implementation in [211].
Similar as for one-photon transitions also for two-photon transitions the non-variational
structure of coupled cluster theory leads to different expressions for “left” and “right” tran-
sition moments M0←fV1 V2
and MfV←0
3 V4
. Only the transition strengths SV0f1 V2 ,V3 V4 which are
obtained as symmetrized combination of both [208, 210],
1 n V1 V2  ∗ o
SV0f1 V 2,V3 V4 (ω) = M0←f (−ω)MfV←0
3 V4 V3 V4
(ω) + M0←f (−ω)MfV←0
1 V2
(ω) (10.24)
2
is an observable quantity.
The ricc2 program prints in the output in addition to the transition moments in atomic
units also the rotationally averaged transition strengths in atomic units and transition rates
in Göppert-Maier (GM) units for linear, perpendicular and circular polarized beams.
In addition to the input for the excitation energies, the computation of two-photon transition
moments requires the keyword twophoton in the data group $excitations and the data
group for the numerical Laplace transformation $laplace.
10.4. TRANSITION MOMENTS 265

ricc2
cc2
$laplace
conv=6
$excitations
irrep=a1 nexc=1
irrep=b2 nexc=1
twophoton states=all operators=(diplen,diplen) freq=0.1d0

The syntax for the input for states is the same as for one-photon transition moments (option
spectrum). Frequencies ω2 for the photon associated in the “right” transition moment,
MfV←01 V2
(ω2 ), with the second operator in each operator pair can be given in atomic units with
the suboption freq. It corresponds to the frequency ω in Eq. 10.24. The frequency associated
with the first operator is automatically determined from the sum rule ω1 = ω0f − ω2 ,
V1 V2
where ω0f is the frequency for the transition f ← 0. The “left” transition moments M0←f
are computed for the frequencies with the opposite sign. If not specified, the two-photon
transition moments are computed for the case that both photons have the same frequency,
i.e. their energies are set for each state to 1/2 of the transition energy, which is the most
common case.

10.4.4 Phosphorescence lifetimes using SOC-PT-CC2

For the calculation of oscillator strengths for triplet excited states and phosphorescence
lifetimes for molecules without heavy atoms the ricc2 program provides as alternative to
the relativistic two-component variant a perturbative SOC-PT-CC2 approach which works
with one-component wavefunctions. In SOC-PT-CC2 the oscillator strengths for triplet
excited states are calculated as first derivatives of the (spin-forbidden) oscillator strengths
for one-component wavefunctions with respect to the strength of the spin-orbit coupling
(SOC) in the limit of zero spin-orbit coupling. The spin-orbit coupling is treated within the
effective spin-orbit mean field approximation, where the mean field two-electron contribution
is computed from the Hartree-Fock density. The SOC-PT-CC2 approach is about an order of
magnitude faster than the computation of oscillator strengths with the two-component RI-
CC2 variant. Scalar relativistic effects from the spin-free X2C Hamiltonian can be included
by using the spin-free X2C Hamiltonian for the Hartree-Fock reference wavefunction. The
theory and the implementation are described in [212].
The implementation is restricted to closed-shell Hartree-Fock reference wavefunctions and
CC2 without spin-component scaling. For carrying out such calculations one needs in addi-
tion to the input for the excitation energies for triplet states the keyword momdrv and the
data group for the numerical Laplace transformation $laplace.

$ricc2
cc2
$laplace
266 CHAPTER 10. RI-CC2

conv=6
$excitations
irrep=b2 multiplicity=3 nexc=1
momdrv states=(b2{3} 1) operators=(diplen,soc) freq=0.0d0

For phosphorescence lifetimes the frequency has to be set to zero with the option freq=0.0d0
and the operator pair has to be set to (diplen,soc) for the dipole operator and the SOC
operator in the SOMF approximation, so that the input frequency 0.0d0 is associated with
the SOC operator and the ground to excited state transition frequency ω0f with the dipole
operator.
The program prints in the output in addition to the first-order induced transition moments
for the operator pair (diplen,soc) the induced oscillator strengths and the life times.

10.5 Ground State Second-order Properties with MP2

and CC2

Second-order properties for one-electron perturbation can be computed at the MP2 and the
CC2 level. For MP2, second-order properties are computed as derivatives of the SCF+MP2
total energy. This approach includes the relaxation of the SCF orbitals in the presence of
the perturbation and is restricted to the static (i.e. frequency-independent) limit.
For coupled-cluster model CC2, second-order properties can, similar as the first-order prop-
erties, calculated in orbital-unrelaxed or orbital-relaxed approach as derivatives of the of the
Lagrange functions in Eqs. 10.12 and 10.15. As for MP2, the orbital-relaxed calculations
are restricted to the static limit. Frequency-dependent second-order properties as e.g. dipole
polarizabilities can be computed with the orbital-unrelaxed approach.
Since V7.2 second-order properties are also available in the MPI parallel version and also for
unrestricted high-spin open-shell Hartree-Fock refrences. Note, that second-order properties
not available for spin-component scaled variants of MP2 and CC2 or for restricted open-
shell references. Furthermore, non-Abelian point groups and point groups with complex
irreducible representations are not implemented for second-order properties.
In addition to the standard input, second-order properties require that the data group for
the numerical Laplace transformation $laplace and that the sops option in the data group
$response is set. Frequency-dependent dipole polarizabilities with the CC2 model are
obtained with the input:

$ricc2
cc2
$laplace
conv=4
$response
sop operators=(diplen,diplen) freq=0.077d0
10.6. PARALLEL RI-MP2 AND RI-CC2 CALCULATIONS 267

The frequency has to be given in atomic units. Static orbital-relaxed polarizabilities are
obtained with

$response
sop operators=(diplen,diplen) relaxed

10.6 Parallel RI-MP2 and RI-CC2 Calculations

All functionalities of the ricc2 program are available in the OpenMP-based parallel version
for shared memory (SMP) architectures. Most functionalities of the ricc2 program are also
parallelized for distributed memory architectures (e.g. clusters of Linux boxes) based on the
message passing interface (MPI) standard.
While in general the parallel execution of ricc2 works similar to that of other parallelized
TURBOMOLE modules as e.g. dscf and grad, there are some important difference con-
cerning in particular the handling of the large scratch files needed for RI-CC2 (or RI-MP2).
As the parallel version dscf also the parallel version of ricc2 assumes that the program is
started in a directory which is readable (and writeable) on all compute nodes under the same
path (e.g. a NFS directory). The directory must contain all input files and will at the end of
a calculation contain all output files. Large scratch files (e.g. for integral intermediates) will
be placed under the path specified in the control file with $tmpdir (see Section 22.2.20)
which should point to a directory in a file system with a good performance. All large files
will be placed on the nodes in these file systems. (The local file system must have the same
name on all nodes.) Note that at the end of a ricc2 run the scratch directories specified
with $tmpdir are not guaranteed to be empty. To avoid that they will fill your file system
you should remove them after the ricc2 calculation is finished.
Another difference to the parallel HF and DFT (gradient) programs is that ricc2 will com-
municate much larger amounts of data between the compute nodes. With a fast network
interconnection (Gigabit or better) this should not cause any problems, but with slow net-
works the communication might become the limiting factor for performance or overloading
the system. If this happens the program can be put into an alternative mode where the
communication of integral intermediates is replaced by a reevaluation of the intermediates
(at the expense of a larger operation count) wherever this is feasible. Add for this in the
control the following data group:

$mpi_param
min_comm

10.7 Spin-component scaling approaches (SCS/SOS)

By introducing individual scaling factors for the same–spin and opposite–spin contributions
to the correlation energy most second–order methods can be modified to achieve a (hopefully)
268 CHAPTER 10. RI-CC2

better performance. SCS-MP2 has first been proposed by S. Grimme and SOS-MP2 by
Y. Jung et al. (see below). The generalization of SCS and SOS to CC2 and ADC(2) for
ground and excited states is described in [17]. It uses the same scaling factors as proposed for
the original SCS- and SOS-MP2 approaches (see below). In the ricc2 program we have also
implemented SCS and SOS variants of CIS(D) for excitation energies and of CIS(D∞ ) for
excitation energies and gradients, which are derived from SCS-CC2 and SOS-CC2 in exactly
the same manner as the unmodified methods can be derived as approximations to CC2 (see
Sec. 10.2 and Ref. [213]). Please note, that the SCS-CIS(D) and SOS-CIS(D) approximations
obtained in this way and implemented in ricc2 differ from the spin-component scaled SCS-
and SOS-CIS(D) methods proposed by, respectively, S. Grimme and E. I. Ugorodina in [214]
and Y. M. Rhee and M. Head–Gordon in [215].
A line with scaling factors has to be added in the $ricc2 data group:

$ricc2
scs cos=1.2d0 css=0.3333d0

cos denotes the scaling factor for the opposite–spin component, css the same–spin compo-
nent.
As an abbreviation
scs
can be inserted in $ricc2. In this case, the SCS parameters cos=6/5 and css=1/3 proposed
S. Grimme (S. Grimme, J. Chem. Phys. 118 (2003) 9095.) are used. These parameters are
also recommended in [17] for the SCS variants of CC2, CIS(D), CIS(D∞ ), and ADC(2) for
ground and excited states.
Also, just
sos
can be used as a keyword, to switch to the SOS approach proposed by the Head-Gordon
group for MP2 with scaling factors of cos=1.3 and css=0.0 (Y., Jung, R.C. Lochan, A.D. Du-
toi, and M. Head-Gordon, J. Chem. Phys. 121 (2004) 9793.), which are also recommended
for the SOS variants of CC2, CIS(D), CIS(D∞ ), and ADC(2). The Laplace-transformed
algorithm for the SOS variants are activated by the additional data group $laplace:

$laplace
conv=4

For further details on the Laplace-transformed implementation and how one can estimated
whether the O(N 4 )-scaling Laplace-transformed or O(N 5 )-scaling conventional RI imple-
mentation is efficient see Sec. 9.6.
Since Version 6.6 the O(N 4 )-scaling Laplace-transformed implementation is available for
ground and excited state gradients with CC2 and ADC(2).

Restrictions:
10.7. SPIN-COMPONENT SCALING APPROACHES (SCS/SOS) 269

• the spin (S 2 ) expectation value for open-shell calculation can not be evaluated in the
SCS or SOS approaches

• for LT-SOS-CC2 (and the related CIS(D) and ADC(2) versions) the following further
limitations apply:

– incompatible with the calculation of the D1 and D2 diagnostics

– second-order properties, two-photon transitions moments and induced transition
moments are not implemented for the spin-component scaled variants.
Chapter 11

CCSD, CCSD(F12*) and

CCSD(T) calculations

Since release V7.0 the coupled-cluster singles-and-doubles method CCSD and its explicitly-
correlated variants CCSD(F12) and CCSD(F12*) are implemented in the ccsdf12 program.
CCSD and the F12 variants can be combined with a perturbative correction for connected
triple excitations, CCSD(T).∗ The Brueckner variants BCCD and BCCD(T) with and with-
out explicit correlation are also available. As perturbative approximations beyond MP2,
also the approximations MP3, MP3(F12), MP4, and MP4(F12*) are available. Presently
the implementation of the F12 variants and of connected triple excitations is restricted to
ground state energies. The CCSD implementation without F12 is limited to ground–state
and excitation energies. Closed-shell (RHF), unrestricted (UHF) or single determinant re-
stricted (ROHF) open-shell reference wavefunctions can be used for CCSD and CCSD(T),
but no gradients or properties are (yet) available for these wavefunction models. The same
is true for BCCD and BCCD(T). The MP3 and MP4 approximations can currently not be
combined with ROHF reference wavefunctions.
Further limitations:

no MPI parallelization: calculations at these levels can presently only carried out on a
single compute node, only the OpenMP (see Sec. 3.4.2) parallelization is available

use of symmetry restricted: only D2h and its subgroups can be used for conventional
(i.e. not F12) calculations; no symmetry can be used for the F12 methods

Please note that calculations with MP3, MP4, CCSD, BCCD and methods beyond CCSD
require considerably more disc space and core memory than MP2 or CC2 calculations. (See
section below for more details and recommendations.)
∗
Note that for the explicitly correlated CCSD variants the explicitly-correlated double excitations
are neglected for the calculation of the triples corrections as described in Ref. [216].

270
 271

Prerequisites

MP3, MP4, CCSD and CCSD(T) calculations with the ccsdf12 module require the same
prerequisites as RI-CC2 calculations:

1. a converged SCF calculation with the one-electron density threshold set to $denconv
1.d-5 or less

2. an auxiliary basis defined in the data group $cbas

3. if orbitals should be excluded from the correlation treatment the data group $freeze
has to be set

4. the maximum core memory which the program is allowed to allocate should be de-
fined in the data group $maxcor; the recommended value is 66–75% of the available
(physical) core memory. (see Sec. 22.2.2 for details)

5. the data group $ricc2 with a specification of the coupled-cluster model

The same is true for Brueckner methods. Calculations with the CCSD(F12*), CCSD(F12)
and BCCD(F12∗) methods require in addition:

• the data group $rir12 with the definition of the standard approximations for the
explicitly-correlated contributions (see Sec. 9.5 for details)

• the data group $lcg, which define the correlation function (here it is in particular
important to choose for F12 calculations the exponent; recommended values are 0.9
for cc-pVDZ-F12, 1.0 for cc-pVTZ-F12 and 1.1 for cc-pVQZ-F12 basis sets)

• a complementary auxiliary (CABS) basis set

Furthermore, it is recommended to select in addition an auxiliary JK basis set for the

evaluation of the Fock matrix elements. (The rijk menu of define can be used for this.)

How To Perform a Calculation

As presently no gradients are available, only single–point calculations are possible:

1. Select in define within the menu cc

the wavefunction model (submenu ricc2),

frozen core options (submenu freeze),
an auxiliary basis for $cbas (submenue cbas),
the amount of main memory (option memory),

and for F12 calculations in addition

the F12 options (submenu f12), and

272 CHAPTER 11. CCSD, CCSD(F12*) AND CCSD(T)

a CABS basis (submenu cabs).

By default a CCSD(F12*) with ansatz 2 and geminal amplitudes fixed by the cusp
conditions is performed.†

ccsdapprox ccsd(f12*)

The auxiliary JK basis must be chosen in menu rijk and the exponent for the corre-
lation function must set by editing the $lcg data group of the control file.

2. Do an SCF calculation using either the dscf or the ridft module.

3. Invoke the ccsdf12 program on the command line or with a batch script.

How to quote:
• for all F12 calculations cite the implementation of RI-MP2-F12 in TURBOMOLE:
The MP2-F12 Method in the TURBOMOLE Programm Package. Rafal A. Bachorz,
Florian A. Bischoff, Andreas Glöß, Christof Hättig, Sebastian Höfener, Wim Klopper,
David P. Tew, J. Comput. Chem. 32, 2492–2513 (2011).

• for MP3(F12) and CCSD(F12):

Quintuple-ζ quality coupled-cluster correlation energies with triple-ζ basis sets. David
P. Tew, Wim Klopper, Christian Neiss, Christof Hättig, Phys. Chem. Chem. Phys.
9 921–1930 (2007).

• for MP4(F12*), CCSD(F12*) and CCSD(F12*)(T):

Accurate and efficient approximations to explicitly correlated coupled-cluster singles
and doubles, CCSD-F12. Christof Hättig, David P. Tew, Andreas Köhn, J. Chem.
Phys. 132 231102 (2010).

• for BCCD(F12∗) and BCCD(T)(F12∗) :

Explicitly correlated coupled-cluster theory with Brueckner orbitals. David P. Tew,
J. Chem. Phys., 145, 074103 (2016).

11.1 Characteristics of the Implementation and Compu-

tational Demands

In CCSD the ground–state energy is (as for CC2) evaluated as

ECC = hHF|H|CCi = hHF|H exp(T )|HFi , (11.1)

†
For other avaible approximation and the corresponding input options see Sec. 22.2.20.
11.1. COMPUTATIONAL DEMANDS 273

where the cluster operator T = T1 + T2 consist of linear combination of single and double
excitations:
X
T1 = tai τai , (11.2)
ai
1 X ij
T2 = tab τaibj . (11.3)
2
aibj

In difference to CC2, the cluster amplitudes tai and taibj are determined from equations
which contain no further approximations apart from the restriction of T to single and double
excitations:
ˆ + [H̃,
Ωµ1 = hµ1 |H̃ ˆ T ]|HFi = 0 , (11.4)
2
ˆ + [H̃,
Ωµ2 = hµ2 |H̃ ˆ T ] + [[Ĥ, T ], T ]|HFi = 0 , (11.5)
2 2 2

where again
ˆ = exp(−T )Ĥ exp(T ),
H̃ 1 1

and µ1 and µ2 are, respectively, the sets of all singly and doubly excited determinants. For
(1)
MP3 the energy is computed from the first-order amplitudes (tµ ) as

EMP3,tot =EHF + EMP2 + EMP3 (11.6)

(1) (1)
X
=hHF|Ĥ + [Ĥ, T2 ]|HFi + t(1)
µ2 hµ2 |[Ŵ , T2 ]|HFi (11.7)
µ2

with Ŵ = Ĥ − F̂ . To evaluate the fourth-order energy one needs in addition to the first-order
also the second-order amplitudes, which are obtained from the solution of the equations
(2) (1)
hµ1 |[F̂ , T1 ] + [Ŵ , T2 ]|HFi = 0 (11.8)
(2) (1)
hµ2 |[F̂ , T2 ] + [Ŵ , T2 ]|HFi = 0 (11.9)
(2) (1)
hµ3 |[F̂ , T3 ] + [Ŵ , T2 ]|HFi =0 (11.10)

From these the fourth-order energy correction is computed as:

(2) (2) (2) (1) (1)
X
EMP4 = t(1)
µ2 hµ2 |[Ŵ , T1 + T2 + T3 ] + [[Ŵ , T2 ], T2 ]|HFi . (11.11)
µ2

Eqs. (11.5) and (11.7) – (11.11) are computational much more complex and demanding
than the corresponding doubles equations for the CC2 model. If N is a measure for the
system size (e.g. the number of atoms), the computational costs (in terms of floating point
operations) for CCSD calculations scale as O(N 6 ). If for the same molecule the number
of one-electron basis functions N is increased the costs scale with O(N 4 ). (For RI-MP2
and RI-CC2 the costs scale with the system size as O(N 5 ) and with the number of basis
functions as O(N 3 ).) The computational costs for an MP3 calculations are about the same
as for one CCSD iteration. For MP4 the computational costs are comparable to those for
two CCSD iteration plus the costs for the perturbation triples correction (see below).
274 CHAPTER 11. CCSD, CCSD(F12*) AND CCSD(T)

Explicitly-correlated CCSD(F12) methods: In explicitly-correlated CCSD cal-

culations the double excitations into products of virtual orbitals, described by the operator
T2 = 12 aibj tij
P
ab τaibj , are augmented with double excitations into the explicitly-correlated
pairfunctions (geminals) which are described in Sec. 9.5:

T = T1 + T2 + T20 (11.12)
1 X kl
T20 = cij τkilj (11.13)
2
ijkl

where τkilj |iji = Q̂12 f12 |kli (for the definition Q̂12 and f12 see Sec. 9.5). This enhances
dramatically the basis set convergence of CCSD calculations ( [20]). Without any fur-
ther approximations than those needed for evaluating the necessary matrix elements, this
extension of the cluster operator T leads to the CCSD-F12 method. CCSD(F12) is an ap-
proximation ( [20, 216]) to CCSD-F12 which neglects certain computationally demanding
higher-order contributions of T̂20 . This reduces the computational costs dramatically, while
the accuracy of CCSD(F12) is essentially identical to that of CCSD-F12 [217, 218]. In the
CCSD(F12) approximation the amplitudes are determined from the equations:

Ωµ1 = hµ1 |H̃ ˆ T + T 0 ]|HFi = 0 ,

ˆ + [H̃, (11.14)
2 2

Ωµ2 = hµ2 |H̃ ˆ T + T 0 ] + [[Ĥ, T + 2T 0 ], T ]|HFi = 0 ,

ˆ + [H̃, (11.15)
2 2 2 2 2
ˆ + [H̃,
Ωµ20 = hµ20 |[F̂ , T20 ] + H̃ ˆ T ]|HFi = 0 . (11.16)
2

Similar as for MP2-F12, also for CCSD(F12) the coefficients for the doubles excitations into
the geminals, ckl
ij can be determined from the electronic cusp conditions using the rational
generator (also known as SP or fixed amplitude) approach. In this case Eq. (11.16) is not
solved. To account for this, the energy is then computed from a Lagrange function as:
X
ECCSD(F12)−SP = LCCSD(F12) = hHF|H|CCi + cµ20 Ωµ20 (11.17)
µ20

This is the recommended approach which is used by default if not any other approach
has been chosen with the examp option in $rir12 (see Sec. 9.5 for further details on the
options for F12 calculations; note that the examp noinv option should not be combined
with CCSD calculations). CCSD(F12)-SP calculations are computationally somewhat less
expensive that CCSD(F12) calculations which solve Eq. (11.16), while both approaches are
approximately similar accurate for energy differences.
The SP approach becomes in particular very efficient if combined with the neglect of cer-
tain higher-order explicitly-correlated contributions which have a negligible effect on the
energies but increase the costs during the CC iterations. The most accurate and recom-
mended variant is the CCSD(F12*) approximation [21], which gives essentially iden-
tical energies as CCSD(F12)-SP. Also available are the CCSD[F12] (Ref. [21]), CCSD-F12a
(Ref. [219]) and CCSD-F12b (Ref. [220]) approximations as well as the perturbative correc-
tions CCSD(2)F12 and CCSD(2*)F12 (see Refs. [21, 221, 222]) and CCSD(F12∗) (Ref. [189]).
These approximations should only be used with ansatz 2 and the SP approach (i.e. fixed
geminal amplitudes). Note that the CCSD-F12c method which is available in the molpro
11.1. COMPUTATIONAL DEMANDS 275

package is the same as CCSD(F12*). Since CCSD(F12*) is the original name that was
introduced when the method was proposed for the first time in Ref. [21] users are asked to
use in publications this abbreviation and cite the original reference [21].
For MP3 the approximations (F12*), and (F12) to a full F12 implementation become iden-
tical: they include all contributions linear in the coefficients ckl
ij . The explicitly-correlated
MP4 method MP4(F12*) is defined as fourth-order approximation to CCSD(F12*)(T). Note
that MP4(F12*) has to be used with the SP or fixed amplitude approach for the geminal
coefficients ckl
ij . MP3(F12*) and MP4(F12*) are currently only available for closed-shell or
unrestricted Hartree-Fock reference wavefunctions.
The CPU time for a CCSD(F12) calculation is approximately the sum of the CPU time for an
MP2-F12 calculation with the same basis sets plus that of a conventional CCSD calculation
multiplied by (1 + NCABS /N ), where N is the number of basis and NCABS the number of
complementary auxiliary basis (CABS) functions (typically NCABS ≈ 2−3N ). If the geminal
coefficients are determined by solving Eq. (11.16) instead of using fixed amplitudes, the costs
per CCSD(F12) iteration increase to ≈ (1 + 2NCABS /N ) times the costs for a conventional
CCSD iteration. Irrespective how the geminal coefficients are determined, the disc space
for CCSD(F12) calculations are approximately a factor of ≈ (1 + 2NCABS /N ) larger than
the disc space required for a conventional CCSD calculation. Note that this increase in the
computational costs is by far outweighted by the enhanced basis set convergence.
In combination with the CCSD(F12*) approximation (and also CCSD[F12], CCSD-F12a,
CCSD-F12b, CCSD(2)F12 , CCSD(2*)F12 and CCSD(F12∗) ) the CPU time for the SP ap-
proach is only about 20% or less longer than for a conventional CCSD calculation within
the same basis set.

CC calculations with restricted open-shell (ROHF) references: The MP2

and all CC calculations for ROHF reference wavefunctions are done by first transforming to
a semi-canonical orbital basis which are defined by the eigenvectors of the occupied/occupied
and virtual/virtual blocks of the Fock matrices of alpha and beta spin. No spin restrictions
are applied in the cluster equations. This approach is sometimes also denoted as ROHF-
UCCSD.
Note that if a frozen-core approximation is used, the semicanonical orbitals depend on
whether the block-diagonalization of the Fock matrices is done in space of all orbitals or only
in the space of the correlated valence orbitals. The two approaches lead thus to slightly dif-
ferent energies, but neither one is more valid or more accurate than the other. Both schemes
are available through the specification of core res/can. The default is core can where
the full occupied block is diagonalised. The same scheme is used e.g. in the CFOUR program
suite, but other codes as e.g. the implementation in MOLPRO use a block-diagonalization re-
stricted to the active valence space. The semi-canonical orbitals are written to the $sc_alph
and $sc_bet data groups.
276 CHAPTER 11. CCSD, CCSD(F12*) AND CCSD(T)

Brueckner Calculations and KS-CCSD(T): Brueckner orbital optimisation in

place of single excitations is available, which is useful when the HF reference relaxes signif-
icantly in the presence of correlation (for example for calculations on heavy elements). The
default way a calculation proceeds is by first building the Fock matrix for the input starting
orbitals, semi-canonicalising these orbitals and then optimising the (non-frozen) orbitals at
the same time as solving for the doubles amplitudes, such that the singles residual van-
ishes at convergence. The default is therefore UBCCD for open-shell references. The final
orbitals are written in the $bcc_mos or $bcc_alph, $bcc_bet data groups. It is possible
through options in the $ricc2 data group to skip the initial semi-canonicalisation and to
control the nature of the frozen orbitals. It is also possible to request restricted open-shell
β
BCCD where the Brueckner optimisation applies only to tα 1 + t1 such that the alpha and
beta doubly occupied orbitals of a starting ROHF reference remain identical to each other
β
during the optimisation, and tα 1 − t1 is no longer zero at convergence. Since it is necessary
to semi-canonicalise the converged orbitals in order to perform a (T) calculation, the final
orbitals are always in semi-canonical form if (T) is requested. Since the F12 integrals depend
on the orbitals, the iterative BCCD(F12*) approximation is very expensive. Instead, it is
recommended to use the perturbative equivalent BCCD(T)(F12∗) .
As an alternative to BCCD theory, it is possible (but not recommended) to used Kohn-Sham
orbitals to define the reference. When using orbitals other than converged HF orbitals, it is
necessary to add the $non-canonical MOs data group to the control file. When performing
an F12 calculation, the option r12orb arb should be set in the $rir12 data group.

Perturbative triples corrections: To achieve ground state energies with a high ac-
curacy that systematically surpasses the accuracy MP2 and DFT calculations for reaction
and binding energies, the CCSD model has to be combined with a perturbative correction
for connected triples. The recommended approach for the correction is the CCSD(T) model
(4) (5)
ECCSD(T) = ECCSD + EDT + EST (11.18)

which includes the following two terms:

(4) (2)
X
EDT = tCCSD
µ2 hµ2 |[H, T3 ]|HFi (11.19)
µ2
(5) (2)
X
EST = tCCSD
µ1 hµ2 |[H, T3 ]|HFi (11.20)
µ1

where the approximate triples amplitudes are evaluated as:

ijk
(2) h abc |[Ĥ, T2 ]|HFi
taibjck = − (11.21)
a − i + b − j + c − k

In the literature one also finds sometimes the approximate triples model CCSD[T] (also
(4)
denoted as CCSD+T(CCSD)), which is obtained by adding only EDT to the CCSD energy.
Usually CCSD(T) is slightly more accurate than CCSD[T], although for closed-shell or spin-
unrestricted open-shell reference wavefunctions the energies of both models, CCSD(T) and
CCSD[T] model, are correct through 4.th order perturbation theory. For a ROHF reference,
11.1. COMPUTATIONAL DEMANDS 277

(5)
however, EST contributes already in 4.th order and only the CCSD(T) model is correct
through 4.th order perturbation theory.

Integral-direct implementation and resolution-of-the-identity approxima-

tion: The computationally most demanding (in terms floating point operations) steps of
a CCSD calculation are related to two kinds of terms. One of the most costly steps is the
contraction X ij
ΩB
aibj = tcd (ac|bd) (11.22)
cd

where a, b, c, and d are virtual orbitals. For small molecules with large basis sets or basis sets
with diffuse functions, where integral screening is not effective, it is the time-determing step
and can most efficiently be evaluated with a minimal operation count of 41 O2 V 2 (where O and
V are number of, respectively occupied and virtual orbitals), if the 4-index integrals (ac|bd)
in the MO are precalculated and stored on file before the iterative solution of the coupled-
cluster equation, 11.4 and 11.5, and the full permutational symmetries of tij cd , (ac|bd), and
Ωaibj are exploited. For larger systems, however, the storage and I/O of the integrals (ac|bd)
leads to bottlenecks. Alternatively, this contribution can be evaluated in an integral-direct
way as
X ij X ij X
tij
κλ = tcd Cκc Cλd , ΩBµiνj = tκλ (µκ|νλ), ΩBaibj = ΩBµiνj Cµa Cνb (11.23)
cd κλ µν

which, depending on the implementation and system, has formally a 2–3 times larger oper-
ation count, but allows to avoid the storage and I/O bottlenecks by processing the 4-index
integrals on-the-fly without storing them. Furthermore, integral screening techniques can
be applied to reduce the operation count for large systems to an asymptotic scaling with
O(N 4 ).
In TURBOMOLE only the latter algorithm is presently implemented. (For small systems
other codes will therefore be faster.)
The other class of expensive contributions are so-called ring terms (in some publications
denoted as C and D terms) which involve contractions of the doubles amplitudes taibj with
several 4-index MO integrals with two occupied and two virtual indices, partially evaluated
with T1 -dependent MO coefficients. For these terms the implementation in TURBOMOLE
employs the resolution-of-the-identity (or density-fitting) approximation (with the cbas aux-
iliary basis set) to reduce the overhead from integral transformation steps. Due this approx-
imation CCSD energies obtained with TURBOMOLE will deviate from those obtained with
other coupled-cluster programs by a small RI error. This error is usually in the same order
or smaller than the RI error for a RI-MP2 calculation for the same system and basis sets.
The RI approximation is also used to evaluate the 4-index integrals in the MO basis needed
for the perturbative triples corrections.

Disc space requirements: In difference to CC2 and MP2, the CCSD model does no
longer allow to avoid the storage of double excitation amplitudes (tij
ab ) and intermediates of
278 CHAPTER 11. CCSD, CCSD(F12*) AND CCSD(T)

with a similar size. Thus, also the disc space requirements for CCSD calculations are larger
than for RI-MP2 and RI-CC2 calculations for the same system. For a (closed-shell) CCSD
ground state energy calculation the amount of disc space needed can be estimated roughly
as  
Ndisc ≈ 4N 3 + (4 + mDIIS )O2 N 2 /(128 × 1000) MBytes , (11.24)

where N is the number of basis functions, O the number of occupied orbitals and mDIIS the
number of vectors used in the DIIS procedure (by default 10, see Sec. 22.2.20 for details).
For (closed-shell) CCSD(T) calculations the required disc space is with
 
Ndisc ≈ 5N 3 + 5O2 N 2 + ON 3 /(128 × 1000) MBytes , (11.25)

somewhat larger.
For calculations with an open-shell (UHF or ROHF) reference wavefunction the above esti-
mates should be multiplied by factor of 4.

Memory requirements: The CCSD and CCSD(T) implementation in TURBOMOLE

uses multi-pass algorithms to avoid strictly the need to store arrays with a size of O(N 3 ) or
O(O2 N 2 ) or larger as complete array in main memory. Therefore, the minimum memory
requirements are relatively low—although it is difficult to give accurate estimates for them.
On should, however, be aware that, if the amount of memory provided to the program
in the data group $maxcor becomes too small compared to O2 N 2 /(128 × 1000) MBytes,
loops will be broken in many small batches at the cost of increased I/O and a decrease
in performance. As mentioned above, it is recommended to set $maxcor to 66–75% of the
physical core memory available for the calculation.

Important options: The options to define the orbital and the auxiliary basis sets, the
maximum amount of allocatable core memory ($maxcor), and the frozen-core approximation
($maxcor) have been mentioned above and described in the chapters on MP2 and CC2
calculations. Apart from this, CCSD and CCSD(T) calculations require very little additional
input.
Relevant are in particular some options in the $ricc2 data group:

$ricc2
ccsd
ccsd(t)
conv=7
oconv=6
mxdiis=10
maxiter=25

The options ccsd and ccsd(t) request, respectively, CCSD and CCSD(T) calculations.
Since CCSD(T) requires the cluster amplitudes from a converged CCSD calculation, the
11.1. COMPUTATIONAL DEMANDS 279

option ccsd(t) also implies ccsd. bccd(t) requests the Brueckner coupled cluster variant
BCCD(T).
The number given for mxdiis defines the maximum number of vectors included in the DIIS
procedure for the solution of the cluster equations. As mentioned above, it has some impact
on the amount of disc space used by a CCSD calculation. Unless disc space becomes a
bottleneck, it is not recommended to change the default value.
With maxiter one defines the maximum number of iterations for the solution of the cluster
equations. If convergence is not reached within this limit, the calculation is stopped. Usu-
ally 25 iterations should be sufficient for convergence. Only in difficult cases with strong
correlation effects more iterations are needed. It is recommended to increase this limit only
if the reason for the strong correlation effects is known. (Since one reason could also be
an input error as e.g. unreasonable geometries or orbital occupations or a wrong basis set
assignment. With diffuse basis functions it is sometimes neccessary to tighten the integral
screening threshold in $scftol.) If oscillatory convergence is observed, adding a level shift
can help convergence.
The two parameters conv and oconv define the convergence thresholds for the iterative
solution of the cluster equations. Convergence is assumed if the change in the energy (with
respect to the previous iteration) is smaller than 10−conv and the euclidian norm of the
residual (the so-called vector function) smaller than 10−oconv . If conv is not given in the data
group $ricc2 the threshold for changes in the energy is set to the value given in $denconv
(by default 10−7 ). If oconv is not given in the data group $ricc2 the threshold for the
residual norm is by default set to 10 × conv. With the default settings for these thresholds,
the energy will thus be converged until changes drop below 10−7 Hartree, which typically
ensures an accuracy of about 1 µH. These setting are thus rather tight and conservative
even for the calculation of highly accurate reaction energies. If for your application larger
uncertainties for the energy are tolerable, it is recommended to use less tight thresholds, e.g.
conv=6 or conv=5 for an accuracy of, respectively, at least 0.01 mH (0.03 kJ/mol) or 0.1 mH
(0.3 kJ/mol). The settings for conv (and oconv) have not only an impact on the number of
iterations for the solution of the cluster equations, but as they determine the thresholds for
integral screening also on the costs for the individual iterations.

CCSD(T) energy with a second-order correction from the interference-

corrected MP2-F12: The error introduced from a CCSD(T) calculation with a finite
basis set can be corrected from second-order corrections of the the interference-corrected
MP2-F12 (INT-MP2-F12) (Ref. [223]). The approximate CCSD(T)-INT-F12 at the basis
set limit is given from
X
ECCSD(T)/CBS ≈ ECCSD(T) + F ij [eMP2−F12
ij − eMP2
ij ]. (11.26)
i<j

From define, in the submenu $ricc2 select the ccsd(t) method and add the keyword
intcorr

$ricc2
280 CHAPTER 11. CCSD, CCSD(F12*) AND CCSD(T)

ccsd(t)
intcorr

Then, switch on the f12 method (approximation A or B, inv or fixed). The corrected
CCSD(T)-INT-F12 energy will be printed in the end of the calculation. It is highly recom-
mended to start the CCSD(T)-INT-F12 calculation from a converged SCF calculation with
symmetry, which is transformed to C1 . It is furthermore recommended to use Boys localized
orbitals in the $rir12 submenu. A table with the corrected second-order pair-electron en-
ergies and the corresponding interference factors can also be printed in the output by using
the keyword intcorr all instead of intcorr.

Excitation energies with CCSD: At the (conventional) CCSD level also electronic
excitation energies can be computed. For this the data group $excitations has to be
added (the same keyword as for CC2 applies). The implementation is currently restricted
to vertical excitation energies (no transition moments or properties available) and in the
closed–shell case to singlet excited states.
Note that for single-excitation dominated transitions CCSD is as CC2 correct through
second-order and is not necessarily more accurate than CC2. It is, however, for double
excitations still correct through first-order, while CC2 describes double excitations only in a
zero-order approximation. Therefore, CCSD results are more robust with respect to double
excitation contributions to transitions and are thus usefull to check if CC2 is suitable for a
certain problem.
Chapter 12

PNO-based CCSD and CCSD(T)

calculations

The pnoccsd program provides a PNO-based implementations of CCSD including the per-
turbative triples corrections (T0) and (T) using RHF and UHF references. Presently the
implementation is restricted to ground state energies.
Further limitations:

no MPI parallelization: calculations beyond PNO-MP2 can presently only carried out on
a single compute node, only the OpenMP (see Sec. 3.4.2) parallelization is available.

no use of symmetry: no point group symmetry can be used; the calculations can only be
done in C1

F12: in the current release version only PNO-MP2 can be combined with explicit correlation
and the use of PAOs has to be disabled for PNO-MP2-F12 calculations.

Please note that one of the main strategies in PNO-CCSD calculations is to precompute
most of the required two-electron integrals in the local (LMO, PNO, OSV, etc.) basis
sets prior to the CCSD iterations. Therefore PNO-CCSD calculations on small molecules
or with very tight thresholds can require considerably more disc space and core memory
than the canonical implementation in ccsdf12. In general, the performance of PNO-CCSD
calculations depend crucially on the speed of the disc I/O. It is therefore recommended to
machines with fast RAID systems.

Prerequisites

CCSD and CCSD(T) calculations with the pnoccsd module require the same prerequisites
as calculations with ccsdf12:

281
282 CHAPTER 12. PNO-BASED CCSD AND CCSD(T)

1. a converged SCF calculation with the one-electron density threshold set to $denconv
1.d-5 or less

2. an auxiliary basis defined in the data group $cbas

3. if orbitals should be excluded from the correlation treatment the data group $freeze
has to be set

4. the maximum core memory which the program is allowed to allocate should be de-
fined in the data group $maxcor; the recommended value is 66–75% of the available
(physical) core memory. (see Sec. 22.2.2 for details)

5. the data group $pnocsd with a specification of the method and the PNO threshold.

In principle the the data group $pnoccsd does not require more input than a specification of
the wavefunction model (available are currently: mp2, mp3, ccsd, ccsd(t0), and ccsd(t)).
The program will then use by default a PNO threshold tolpno of 10−7 . All other thresholds
are, if not explicitly specified, coupled to tolpno (see below).
For CCSD(T) calculations one should in addition include the Laplace data group and specify
the threshold for the Laplace transformation used in the PNO-based implementation of the
perturbative triples correction (T). By default the program will use a threshold of conv=2. In
many cases the PNO-CCSD(T0) approximation provides an economic alternative to PNO-
CCSD(T).

How To Perform a Calculation

As presently no gradients are available, only single–point calculations are possible:

1. Select in define within the menu pnocc

the wavefunction model (submenu pnoccsd),

frozen core options (submenu freeze),
an auxiliary basis for $cbas (submenu cbas),
the amount of main memory (option memory),

2. Do an SCF calculation using either the dscf or the ridft module.

3. Invoke the pnoccsd program on the command line or with a batch script.

The minimal input for the $pnoccsd data group needed for a PNO-CCSD(T) calculation is
just:

$pnoccsd
tolpno 1.d-7
ccsd(t)
 12.1. SELECTION THRESHOLDS USED IN THE PNOCCSD PROGRAM 283

How to quote:
• for PNO-MP2 calculations: A O(N 3 )-scaling PNO-MP2 method using a hybrid OSV-
PNO approach with an iterative direct generation of OSVs. Gunnar Schmitz, Ben-
jamin Helmich, Christof Hättig, Mol. Phys. 111, 2463–2476, (2013).

• for PNO-CCSD calculations: Explicitly correlated PNO-MP2 and PNO-CCSD and

its application to the S66 set and large molecular systems. Gunnary Schmitz, Christof
Hättig, David P. Tew, Phys. Chem. Chem. Phys. 16, 22167–22178, (2014).

• for PNO-based triples (T0) or (T) calculations: Perturbative triples correction for local
pair natural orbital based explicitly correlated CCSD(F12*) using Laplace transfor-
mation techniques. Gunnar Schmitz, Christof Hättig, J. Chem. Phys. 145, 234107
(2016)

• for calculations using PAOs (default) Principal Domains in Local Correlation Theory.
David. P. Tew, J. Chem. Theory Comput. 15, 6597 (2019)

12.1 Selection Thresholds used in the pnoccsd Program

The accuracy and the computational costs for calculations with the pnoccsd program de-
pend crucially on a number of selection thresholds. Their default values are coupled to the
PNO truncation threshold tolpno, such that the errors from all other approximations are
smaller then the PNO truncation error. Depending on the application it might be that, in
particular for energy differences, certain errors cancel out to a large extent and significant
computational saving might be obtained by loosening specific thresholds.
The following thresholds can be set in the data group $pnoccsd:

tolpno The PNO truncation threshold. The default value is 10−7 and gives typically errors in
the (total) correlation energy < 1% which decrease as a rule of thumb approximately
√
with tolpno.

tolosv The OSV truncation threshold (cmp. Ref. [23]) which is by default set to < 0.1 ×
tolpno. This threshold mainly influences the time for the PNO-MP2 step.

tolri The threshold for selecting the local RI basis (cmp. Ref. [23]), which is by default set
√
to 107/12 × tolpno. This threshold is important for the performance and accuracy
of the integral evaluation.

tolpair The energy threshold for selecting the significant LMO pairs. If not specified in
$pnoccsd it is set to (0.1 × tolpno)2/3 .

toltno The threshold for the TNO truncation. The default is toltno = tolpno. Since the
TNOs are restricted through the PNOs this is the tightest useful value. Setting it
tighter than tolpno will usually not improve the accuracy. It only affects the triples
corrections and is important for their costs and accuracy.
 284 CHAPTER 12. PNO-BASED CCSD AND CCSD(T)

toltrip The selection threshold for LMO triples included for the (T0) and (T) corrections.
By default toltrip is set to toltno. It only affects the triples corrections and is
important for their costs and accuracy.
tolopao The selection threshold for PAO domains for OSVs (cmp. Ref. [224]).
tolppao The selection threshold for PAO domains for PNOs (cmp. Ref. [224]).
toltpao The selection threshold for PAO domains for TNOs.

For further options and thresholds see Sec. 22.2.22.

12.2 Characteristics of the Implementation

The pnoccsd program works∗ with localized molecular occupied orbitals (LMOs). Their
local nature is exploited to truncate for each LMO pair the number of interacting virtual
orbitals and thereby the number of required two-electron integrals. PNO-MP2 belongs to the
family of local correlation methods which have in common the key idea that the correlation
energy can be written as a sum of pair contributions
X 1 X ij
MP2
Ecorr = eij = tab hij||abi (12.1)
4
i≤j iajb

and that with localized molecular orbitals the pair contributions eij decrease fast with the
distance |rij | between the LMOs i and j (approximately with 1/|rij |6 ) so that for large
molecules the number of pairs that need to be included to compute the correlation energies
within a given accuracy increases only linearly with the number of atoms. Furthermore,
only such virtual orbitals contribute significantly to the correlation energy eij of a pair ij
which are spatially close to i and j.
The OSV-PNO hybrid approach implemented in the pnoccsd program starts with the de-
termination of truncated subsets of projected atomic orbitals (PAOs) and orbital-specific
virtual (OSV) orbitals for each LMO i as the eigenvectors of diagonal density contributions
Dii computed from (approximate) MP2 doubles amplitudes:
X X
ii
Dab =2 tii ii
ac tcb = dii ii ii
ac̃ nc̃ dbc̃ (12.2)
c c̃

with occupation numbers nii

c̃ not smaller than a predefined threshold tolosv. The OSVs are
first used to calculate for each pair ij estimates for the pair correlation energy contribution
eij and to set up a list of pairs ij for which significant contributions to the total correlation
energy are expected. Only for this list of pairs the OSVs of the two LMOs are merged
together to an orthonormal set of pre-PNOs c̃ij which are used to compute the contribution
of the pair ij to the density matrix via approximated MP2 amplitudes and as eigenvectors
of these densities the coefficients of the pair natural orbitals:
X ij X ij ij ij
Dãij b̃ = 2 tãij c̃ij tij
c̃ b̃
= dac̄ nc̄ dbc̄ . (12.3)
ij ij ij ij
c̃ij c̄

∗
The program can also run with canonical orbitals but will then not be efficient.
12.2. CHARACTERISTICS OF THE IMPLEMENTATION 285

Similar as for the OSVs also the PNOs with occupation numbers nij
c̄ below the PNO trunc-
tion threshold tolpno are discarded.
For OSV-PNO-MP2 the MP2 equations are then solved with the amplitudes and two-electron
integals projected onto the so determined set of pair natural orbitals. For the explicitly-
correlated variant OSV-PNO-MP2-F12 similar sets of orbital and pair specific auxiliary
orbitals (OSX) are determined for the occupied and the complementary virtual orbital spaces
which appear in the calculation of the three-electron integrals for the additional contributions
from the geminals.
The thresholds for discarding PNOs and OSVs and the subset of auxiliary orbitals for the
local RI approximation for computing two-electron integrals play an important role for the
accuracy and the computational costs of calculations with the pnoccsd program. Per default
all thresholds are coupled to the PNO truncation threshold tolpno such that the errors due
to the OSV, local RI and pair truncation approximations and due to the screening of (AO)
integrals and contributions to the residual are about an order of magnitude smaller than the
PNO truncation error. The default value for tolpno is 10−7 which gives PNO truncation
errors in the correlation energy < 1%. Test calculations showed that the relative errors in
√
the correlation energy decrease typically linear with tolpno.
All approximations in the CCSD energy are equivalent for UHF and RHF references such
that a UHF-based calculation on a closed shell species gives the same energy as the RHF-
based calculation to within numerical precision. In V7.4.2, the approximations involved in
the (T) energy lead to small differences between UHF and RHF based calculations of a
closed shell system.

12.2.1 Strong pair approximation

Beyond MP2 the PNO implementation makes per default use of a strong pair approximation:

• The 3.rd order CCD ladder terms and corresponding exchange terms:
X 
˜|lj) + ˜|cd) − ˜|bc)tkj + (kj˜|ac)tki
X X ij
tkl
ab (ki t cd (ac (ki ac bc (12.4)
kl cd ck

decay together as R−6 with the distance between the centers of the LMOs i and j
and for the first term also with the distance between the centers of the LMOs k and
l. These terms are only evaluated if ij is a strong pair. For the last two terms in
addition ki and kj and for the first term kl, ki, and lj are required to be strong pairs.

• The Coulomb integrals (ki˜|ca) in the ring terms are neglected if ki is a distant or
neglected pair.
√
Per default the threshold for Tstrg = 10 × TPNO and Tweak = Tpair × Tstrg . Above (pq˜|rs)
2/3 1/3

denote t1 -dressed two-electron integrals.

Chapter 13

Random Phase Approximation

Calculations: Energy and
First-Order Properties

Ground state energies and analytic first-order properties (e.g., ’gradients’ for structure opti-
mizations) can be computed within the random phase approximation (RPA) using the rirpa
module. Theory and development of the rirpa module is published in references [225, 226]
for the energy and reference [227] for the first-order properties. See also [228] for a sur-
vey of applications of RIRPA. For two-component relativistic RPA energy calculations, see
reference [229]. For perturbative beyond-RPA calculations, see references [230, 231]. For
orbital-self-consistent approach called the generalized Kohn–Sham semi-canonical projected
RPA (GKS-spRPA), see reference [232]. For energy and gradients, the resolution-of-the-
identity (RI) approximation is used to approximate the two-electron repulsion integrals in
the correlation treatment and is combined with an imaginary frequency integration. The
RI approximation is also employed by default for the computation of the Coulomb integrals
for the HF energy. For the energy, it is optional to use RI for the Fock exchange integrals
(’RI-K’), while RI-K for the gradients is not available yet. Open shell systems and the
frozen core approximation may be used in RPA energy calculations but are not presently
available in gradient calculations. Two-component RPA energy calculations are only pos-
sible for Kramers-restricted closed-shell systems. ECPs are presently not compatible with
RIRPA gradients. Neither RPA energy nor gradients support symmetry at the moment.
The gradients may be used together with the scripts jobex (for structure optimizations)
and NumForce (for numerical harmonic vibrational frequencies).

286
13.1. GROUND STATE ENERGY THEORY 287

13.1 Ground State Energy Theory

The RPA energy

E RPA = E HF + E C RPA (13.1)

consists of the Hartree-Fock exact exchange energy E HF and a correlation energy piece
E C RPA . rirpa computes Eq. (13.1) non-selfconsistently from a given set of converged input
orbitals. The correlation energy
1 X RPA
E C RPA = Ωn − ΩTDARPA


n (13.2)
2 n

is expressed in terms of RPA excitation energies at full coupling ΩRPA

n and within the
TDARPA
Tamm-Dancoff approximation Ωn . The further discussion is restricted to the one-
component (nonrelativistic) treatment, for the sake of convenience. For the derivation of
the two-component RPA theory see ref. [229]. The excitation energies are obtained from
time-dependent DFT response theory and are eigenvalues of the symplectic eigenvalue prob-
lem [233, 234]
(Λ − Ω0n ∆)|X0n , Y0n i = 0. (13.3)

The super-vectors X0n and Y0n are defined on the product space Locc ×Lvirt and Locc ×Lvirt ,
respectively, where Locc and Lvirt denote the one-particle Hilbert spaces spanned by occupied
and virtual static KS molecular orbitals (MOs). The super-operator
!
A B
Λ= (13.4)
B A

contains the so-called orbital rotation Hessians,

(A + B)iajb = (a − i )δij δab + 2(ia|jb), (13.5)

(A − B)iajb = (a − i )δij δab . (13.6)

i and a denote the energy eigenvalues of canonical occupied and virtual KS MOs. rirpa
computes so-called direct RPA energies only, i.e. no exchange terms are included in Eqs. (13.5)
and (13.6).
In RIRPA the two-electron integrals in Eqs (13.5) are approximated by the resolution-of-
the-identity approximation. In conjunction with a frequency integration this leads to an
efficient scheme for the calculation of RPA correlation energies [225]
Z ∞
C RIRPA dω C
E = F (ω), (13.7)
−∞ 2π

where the integrand contains Naux × Naux quantities only,

1
F C (ω) = tr (ln (Iaux + Q(ω)) − Q(ω)) . (13.8)
2
Naux is the number of auxiliary basis functions. The integral is approximated using Clenshaw-
Curtiss quadrature.
288 CHAPTER 13. RANDOM PHASE APPROXIMATION

Prerequisites

Calculations with rirpa require

• a converged SCF calculation

• rirpa-options may be included by adding them in the lines below the keyword $rirpa
in the control file. Possible options are:

– npoints hintegeri - Number of frequency integration points (default is 60).

– nohxx - HF energy is skipped, (HXX = Hartree + eXact (Fock) eXchange).
– rpaprof - Generates profiling output.

• the maximum core memory the program is allowed to allocate should be defined in
the data group $maxcor (in MB); the recommended value is ca. 3/4 of the available
(physical) core memory at most.

• orbitals to be excluded from the correlation treatment have to be specified in data

group $freeze

• an auxiliary basis defined in the data group $cbas

• an auxiliary basis defined in the data group $jbas for the computation of the Coulomb
integrals for the Hartree-Fock energy. This is optional if the nohxx option is set.

• (optional) an auxiliary basis defined in the data group $jkbas for the computation
of the exchange integrals for the Hartree-Fock energy. $rik should be added to the
control file for RI-JK to be effective.

To perform a two-component relativistic RIRPA calculation [229] on (Kramers-

restricted) closed-shell systems taking into account spin-orbit coupling, the two-component
version of ridft has to be run before (see Chapter 6.4) using the keywords $soghf and
$kramers. The implementation is currently only available in combination with the nohxx
option.

13.2 Gradients Theory

All details on the theory and results are published in [227]. The RI-RPA energy is a function
of the MO coefficients C and the Lagrange multipliers  and depends parametrically (i)
on the interacting Hamiltonian Ĥ, (ii) on the AO basis functions and the auxiliary basis
functions. All parameters may be gathered in a supervector X and thus

E RIRPA ≡ E RIRPA (C, |X) . (13.9)

C and  in turn depend parametrically on X, the exchange-correlation matrix VXC , and

the overlap matrix S through the KS equations and the orbital orthonormality constraint.
First-order properties may be defined in a rigorous and general fashion as total derivatives
13.2. GRADIENTS THEORY 289

of the energy with respect to a “perturbation” parameter ξ. However, the RI-RPA energy is
not directly differentiated in our method. Instead, we define the RI-RPA energy Lagrangian
LRIRPA (C, E, D∆ , W|X, VXC , S)
X
= E RIRPA (C, E|X) + Dσ∆ (CσT Fσ Cσ − Eσ ) − Wσ (CσT SCσ − 1) .


(13.10)
σ

C, E, D∆ , and W are independent variables. LRIRPA is required to be stationary with respect

to C, E, D∆ , and W. D∆ and W act as Lagrange multipliers enforcing that C and E satisfy
the KS equations and the orbital orthonormality constraint,
 RIRPA 
∂L
= CσT Fσ Cσ − Eσ = 0 , (13.11)
∂Dσ∆ stat
 RIRPA 
∂L
= CσT SCσ − 1 = 0 . (13.12)
∂Wσ stat

D∆ and W are determined by the remaining stationarity conditions,

 RIRPA 
∂L
=0, (13.13)
∂E stat
and
∂LRIRPA
 
=0. (13.14)
∂C stat
It turns out from eqs (13.13) and (13.14) that the determination of D∆ and W requires the
solution of a single Coupled-Perturbed KS equation. Complete expressions for D∆ and W
are given in [227]. At the stationary point “stat = (C = C, E = , D∆ = D∆ , W = W)”,
first-order RI-RPA properties are thus efficiently obtained from
dE RIRPA (C, |X)
 RIRPA    RIRPA 
∂VXC
  
∂L dX ∂L
= +
dξ ∂X stat dξ ∂VXC stat ∂ξ stat
 RIRPA  
∂L dS
+ . (13.15)
∂S stat dξ
Finally, the RPA energy gradients may be explicitly expanded as follows:
 * (4)
+ 
dE RIRPA (C, |X) ∂VXC [D]
   
RIRPA dh (4) dΠ ∆
= D + Γ + D
dξ dξ dξ ∂ξ stat
* + * + 
dΠ(3) dΠ(2)

dS
+ Γ(3) + Γ(2) − W . (13.16)
dξ dξ dξ

where DRIRPA is the KS ground state one-particle density matrix D plus the RI-RPA differ-
ence density matrix D∆ which corrects for correlation and orbital relaxation effects. h is the
one-electron Hamiltonian; Π(2/3/4) are 2-, 3-, and 4-centre electron repulsion integrals and
the Γ(2/3/4) are the corresponding 2-, 3-, and 4-index relaxed 2-particle density matrices;
W may be interpreted as the energy-weighted total spin one-particle density matrix.
This result illustrates the key advantage of the Lagrangian method: Total RI-RPA energy
derivatives featuring a complicated implicit dependence on the parameter X through the
variables C and  are replaced by partial derivatives of the RI-RPA Lagrangian, whose
computation is straightforward once the stationary point of the Lagrangian has been fully
determined.
290 CHAPTER 13. RANDOM PHASE APPROXIMATION

Gradients Prerequisites

Geometry optimizations and first order molecular property calculations can be executed by
adding the keyword rpagrad to the $rirpa section in the control file. RPA gradients also
require

• an auxiliary basis defined in the data group $jbas for the computation of the Coulomb
integrals for the Hartree-Fock energy

• an auxiliary basis defined in the data group $cbas for the ERI’s in the correlation
treatment.

• zero frozen core orbitals; RIRPA gradients are not compatible with the frozen core
approximation at this time.

The following gradient-specific options may be further added to the $rirpa section in the
control file

• drimp2 - computes gradients in the DRIMP2 limit.

• niapblocks hintegeri - Manual setting of the integral block size in subroutine rirhs.f;
for developers.

In order to run a geometry optimization, jobex must be invoked with the level set to rirpa,
and the -ri option (E.g. jobex -ri -level rirpa).
In order to run a numerical frequency calculation, NumForce must be invoked with the level
set to rirpa, e.g., NumForce -d 0.02 -central -ri -level rirpa.

13.3 Generarlized Kohn Sham scheme for RIRPA

The RIRPA energy, discussed in Sec. 13.1, is non-self-consistently computed using canonon-
ical KS orbitals obtained from a prior KS-DFT based calculation. If the KS-DFT densities
are poor, the errors can carry over to post-KS RIRPA energetics resulting in poor inter-
action energies in some cases. Such density-driven errors can be alleviated by an orbital
self-consistent approach called the generalized Kohn–Sham semi-canonical projected RPA
(GKS-spRPA). In the GKS-spRPA method, the spRPA energy functional

E spRPA [D, H̃KS

0 [D]] = E
HF
[D] + E C spRPA [D, H̃KS
0 [D]] (13.17)

is minimized with respect to the non-interacting GKS density matrix

X
D= Pλ nλλ0 Pλ . (13.18)
λ

D is constrained to be normalized to N electrons and have eigenvalues between 0 and 1. Pλ

denotes orthogonal projectors belonging to blocks of KS orbitals with degenerate occupation
numbers, and nλλ0 = nλ δλλ0 is diagonal, with nλ denoting occupation number matrices. For
13.3. GENERARLIZED KOHN SHAM SCHEME FOR RIRPA 291

integer KS occupations nλ has eigenvalues nλ = 1, 0. The semicanonical projected (sp) KS

Hamiltonian X
H̃KS
0 = Pλ HKS
0 Pλ , (13.19)
λ

contains only the diagonal (λ = λ0 ) blocks of the KS Hamiltonian

X
HKS
0,ij = hij +
XC
Vipjq Dpq + Vij [D] . (13.20)
pq

h is the one-electron Hamiltonian, the second term denotes the Hartree or Coulomb potential
and VXC is the exchange-correlation potential. The eigenvalues of sp KS Hamiltonian are
denoted by ε̃KS
i . V is the matrix of two-electron integrals

φ∗p (r1 )φ∗q (r2 )φr (r1 )φs (r2 )

Z Z
Vpqrs = d3 r1 d3 r2 . (13.21)
|r1 − r2 |
The subscripts i, j, .. denote orbital indices. The spRPA energy functional, eq. 13.17, thus
generalizes the post-KS RPA energy functional for arbitrary density matrices. A GKS energy
minimization in the space of density matrices leads to a set of one-particle equations,

HspRPA [D]φp = εGKS-spRPA

p φp , (13.22)

which are solved self-consistently. The one-particle spRPA Hamiltonian is the functional
derivative of the RPA energy functional, and contains contributions from the HF potential
and RPA correlation potential,

δE spRPA [D]
HspRPA [D] = = HHF [D] + VC spRPA [D] . (13.23)
δD
The self-consistent procedure used in eq. 13.22 is similar to that used in Hartree-Fock and
KS-DFT. At the stationary point of the minimization procedure, the GKS-spRPA proce-
dure yields a total many-body energy and one-particle orbital energies, εGKS-spRPA
p . The
GKS-spRPA
latter are related to approximate ionization potentials and electron affinities. εp
provide consistently accurate estimates of IPs and EAs due to the inclusion of static Hartree-
exchange effects, via HHF contribution, and orbital-correlation and orbital-relaxation effects,
via VC spRPA [D] contribution. εGKS-spRPA
p provide accurate estimates of valence ionization
potentials for neutral and anionic systems [232]. The balanced inclusion of orbital-correlation
and -relaxation effects leads to accurate estimates of absolute and relative core-ionization
energies [235].
The computation of the complete one-particle spectrum currently has a scaling of O(N 6 ).
To reduce the computational costs, we carry out the minimization in two steps: (i) We
approximate the occupied-occupied and virtual-virtual blocks of HspRPA ≈ HHF for the
self-consistent procedure and obtain the converged total energy, and then (ii) evaluate the
one-particle eigenvalues at the final converged stationary point only. This two step procedure
effects the rate of energy-convergence only and not the final total energy while maintaining
a computational scaling of log(N )O(N 4 ) for energy evaluation in each iteration of step
(i). Thus the cost of GKS-spRPA total energy is number of iterations times the cost for
the computation of right-hand side vector, which has a scaling log(N )O(N 4 ). (Note: The
292 CHAPTER 13. RANDOM PHASE APPROXIMATION

computation of the one-particle eigenspectrum, i.e. the step (ii), will be available in a future
release.)
We note that the GKS-spRPA energy functional has a parametric dependence on the KS
potential. However, the variational GKS minimization of the spRPA energy functional
reduces the dependency; the quality of results, for energy differences and IPs/EAs, is similar
for different KS potentials. We also note that GKS energy minimization of small-gap systems
may have poor or no convergence. For such cases, the convergence can be improved by a
(small) artificial-shift of the KS orbital energy differences.

Prerequisites

GKS-spRPA total energy calculation requires

• a converged KS-DFT calculation which provides the starting guess MOs for GKS-
spRPA. scheme.

• relevant rirpa-options:

– npoints hintegeri - Number of frequency integration points (default is 60).

– iter hintegeri - Turns on GKS-spRPA self-consistent iterations. hintegeri is the
number of GKS-spRPA iterations (default is 0).
– ldiis - Turns on DIIS algorithm which speeds up energy convergence.
– eigshift hreali - (optional) Introduces a shift to the non-interacting gap to
aid in the convergence of GKS iterations. This may be required for small-gap
systems. (Note: Carefully check your results if eigshift is being used. After the
final iteration, rerun a rirpa energy calculation without the iter a nd eigshift
keywords. The resulting energy should be used as the final GKS-spRPA energy.)
– output hstringi - (optional) A condensed version of the output relevant to GKS-
spRPA energetics is written to the file hstringi (default filename is gksrpa.dat).

• the convergence criterion for total energy is controlled by the $scfconv keyword.

13.4 Further Recommendations

• The direct RPA correlation energy is defined in a Kohn-Sham context without inclu-
sion of exchange integrals and therefore the use of self-consistent KS orbitals obtained
from (semi-)local functionals is recommended. HF-orbitals or KS-orbitals obtained
form hybrid functionals lead to inferior results.

• Experience has demonstrated that the difference in RPA correlation energies obtained
from different (semi-)local functionals is very small (much smaller than the inherent
error of the method).
13.5. COMMENTS ON THE OUTPUT 293

• Like MP2, RIRPA results are known to converge very slowly with increasing basis
set size, in particular slowly with increasing l-quantum number of the basis set. For
reliable results the use of QZVP basis sets (or higher) is recommended. For non-
covalently bound systems larger basis sets (especially, with more diffuse functions)
are needed.

• It is recommended to exclude all non-valence orbitals from RIRPA calculations, as

neither the TURBOMOLE standard basis sets SVP, TZVPP, and QZVPP nor the cc-pVXZ
basis set families (with X=D,T,Q,5,6) are designed for correlation treatment of inner
shells (for this purpose polarisation functions for the inner shells are needed). The
default selection for frozen core orbitals in Define (orbitals below -3 a.u. are frozen)
provides a reasonable guess. If core orbitals are included in the correlation treatment,
it is recommended to use basis sets with additional tight correlation functions as e.g.
the cc-pwCVXZ and cc-pCVXZ basis set families.

• We recommend the use of auxiliary basis sets optimized for the corresponding (MO)
basis sets. The auxiliary basis sets optimized for RI-MP2 and RI-CC2 are suitable for
rirpa [225] correlation energy calculations.

• Within the two-component relativistic implementation, RIRPA total energies (HF@KS

+ correlation) must be computed in two steps. RIRPA correlation energies can be
obtained using the nohxx option, and the HF energy can then be computed sepa-
rately, e.g., in ridft if the RI-J approximation is used for the Coulomb integrals. To
compute the HF@KS energy, compute the KS orbitals first; then disable $dft and set
$scfiterlimit 1 in the control file to perform a single SCF iteration. Finally add
the total HF@KS energy from ridft to the correlation energy from the nohxx-rirpa
calculation to obtain the total RIRPA energy. Note: the molecular orbitals are altered
by ridft after a single-iteration, so the HF@KS energy must be computed after the
RIRPA correlation energy.

• Tight SCF ($scfconv 7) and one-electron density matrix ($denconv 1d-7) conver-
gence criteria, large basis sets (QZVP), and large frequency grids which ensure a sen-
sitivity measure of no more than 1d-4 should be used in combination with rpagrad
for accurate results.

13.5 Comments on the Output

• The most important output for rirpa are the Hartree-Fock (HXX) energy and the
RIRPA correlation energy which are written to the standard output.

• The optimal scaling parameter for the quadrature grid is printed together with a sen-
sitivity parameter. The sensitivity parameter provides a numerical estimate for the
error in the numerical integration used to evaluate E C RIRPA . Experience demon-
strates that the sensitivity parameter correlates well with the condition number of
the matrix Q. Small gap systems have large condition numbers and therefore require
294 CHAPTER 13. RANDOM PHASE APPROXIMATION

large grids. An estimate of the number of eigenvalues smaller than 0.05 H is given
and if necessary, a warning to increase the grid size is printed to standard output.

• The molecular dipole and quadrupole moments are written to the standard output
whenever a gradient calculation is carried out.

• Rotational constants are provided in the rirpa output below the coordinate section.

• Enabling the option rpaprof will output additional timings information.

Chapter 14

Many body perturbation theory in

the GW approximation

DFT estimates of both single-particle excitation energies and vertical excitation spectra can
be systematically improved upon by the GW - and Bethe–Salpeter methods, which are both
based on a single-particle Green’s function.

14.1 Single particle spectra based on the GW approxi-

mation

14.1.1 Theoretical background

A method to systematically improve upon DFT estimates of single-particle excitation spec-

tra, that is, ionization potentials and electron affinities, is the GW method. Its central object
is the single-particle Green’s function G; its poles describe single-particle excitation ener-
gies and lifetimes. In particular, the poles up to the Fermi-level correspond to the primary
vertical ionization energies. The GW -approach is based on an exact representation of G in
terms of a power series of the screened Coulomb interaction W , which is called the Hedin
equations. The GW -equations are obtained as an approximation to the Hedin-equations, in
which the screened Coulomb interaction W is calculated neglecting so called vertex correc-
tions. In this approximation the self–energy Σ, which connects the fully interacting Green’s
function G to a reference non-interacting Green’s function G0 , is given by Σ = GW .
This approach can be used to perturbatively calculate corrections to the Kohn–Sham spec-
trum. To this end, the Green’s function is expressed in a spectral representation as a sum

295
296 CHAPTER 14. MBPT CALCULATIONS

of quasi particle states.

X Ψr,n (r, z)Ψ†l,n (r0 , z)

G(r, r0 ; z) = . (14.1)
n
z − εn (z) + iη sgn(εn − µ)

Under the approximation that the KS states are already a good approximation to these
quasi–particle states Ψl,n the leading order correction can be calculated by solving the
zeroth order quasi–particle equation:

εn = n + hn|Σ[GKS ](εn ) − Vxc |ni (14.2)

An approximation to the solution of this equation can be obtained by linearizing it:

εn = n + Zn hn| Σ(n ) − Vxc |ni (14.3)

here, Zn is given by:

" #−1
∂Σ(E)
Zn = 1 − hn| |ni (14.4)
∂E E=n

reducing the computational effort to a single iteration.

The self–energy Σ appearing in Eqn. (14.2) is calculated in the GW approximation from
the KS Green’s function and screening. This is the so-called G0 W0 approximation. The
Self–energy splits in an energy independent exchange part Σx and a correlation part Σc (E)
that does depend on energy. Their matrix elements are given by:
X
hn| Σx |n0 i = = − (ni|in0 ) , (14.5)
i

and
2
XX |(nn|ρm )|
hn| Σc (n ) |ni = . (14.6)
m n
n − n − Zm sgn(n − µ)

Where Zm = Ωm − iη are the excitation energies shifted infinitesimally into the complex
plane. The ρm are the corresponding excitation densities. More details, tests and benchmark
calculations are can be found in Ref. 34.

14.1.2 GW features

The GW method is implemented in TURBOMOLE in the escf module supporting the following
features:

• LDA, GGA, meta-GGA and their Hybrid functionals can be used for the underlying
DFT calculation.

• Single-shot G0 W0

• Quasiparticle selfconsistent GW (qsGW)

• Eigenvalue-only selfconsistent GW (evGW)

14.1. SINGLE PARTICLE SPECTRA BASED ON THE GW APPROXIMATION297

• In G0 W0 , the linearized, Eqn. (14.3), and solved, Eqn. (14.2), quasiparticle equation.

• Both RPA and TDDFT response functions can be used to screen the Coulomb inter-
action in constructing W , although we only recommend to use RPA response.

• Closed shell and open shell references are supported in all calculations. Addition-
ally, Kramers-restricted closed shell and also Kramers-unrestricted closed and open
shell systems within the two-component relativistic framework (inclusion of spin-orbit
coupling) can be treated using $soghf. For open shell two-component calculations
collinear and non-collinear approaches are implemented.

14.1.3 General recipe for G0 W0 calculations

Since TURBOMOLE 7.5 two fast GW options are available: Just add $g0w0 OR $evgw to the
control file and run escf with the -gw flag: escf -gw. This automatically sets all parameters
to curated sensible values and will provide very good results and starting points for a BSE
calculation in the majority of all cases.
The general recipe for a G0 W0 and evGW calculations with dRPA response using RI is as
follows:

1. define session

2. Choose reasonable cbas auxiliary basis sets (define)

3. dscf or ridft calculation

4. Provide $gw OR $rigw flags and keywords

5. To trigger the fast RI algorithm add $rick

6. escf calculation

Ad 1) The def2-TZVPP basis seems to be the most useful, it comes for all tested systems
within 0.1 eV of the def2-QZVP result with about half the number of basis functions. A cbas
must be set, the use of jbas-type fitting bases is discontinued for GW since Turbomole 7.3.
In the final define menu select the gw menu to set options for the actual GW calculation.
The gw menu in define will set all needed variables for a gw/rigw calculation. The according
$scfinstab and $soes flags will be set, and also the $gw or $rigw flags are written to the
control file with the chosen options. Also the $rick flag is set. The control file provided
by definecan usually be used for a calculation without further modifications; which are
nevertheless described below of one wishes to modify it manually.
Ad 3) Symmetry up to D2h is available for both closed shell (rpas) and open shell system
(urpa) for GW , for $gw and $rigw. Especially in the gw module the use of symmetry leads to
large speedups and the user is encouraged to exploit symmetry if possible. Two-component
calculations can only be performed in C1 . The $gw module can also handle open-shell 2c
calculations, while $rigw is formally limited to Kramers-restricted closed shell molecules in
298 CHAPTER 14. MBPT CALCULATIONS

the 2c case. Moreover, the simplified methods xGW and sGW , which neglect contributions
from excitation vectors are available for all point groups implemented in Turbomole.
Ad 4) In the last define step the gw menu can be selected to set up a G0 W0 (and also evGW
and qsGW ) calculation. There are three distinct versions available for G0 W0 and evGW :

1. $gw uses spectral representations just as the previous version, which scales as N 6 with
system size. This version is to be preferred when full quasiparticle spectra are desired
(i.e.: QP energies are also needed for non-valence orbitals, or generally for all). It is
compatible with nearly all available starting points:

(a) scalar/non-relativistic closed-shell systems (up to D2h symmetry)

(b) scalar/non-relativistic open-shell systems (up to D2h symmetry)
(c) two-component (2c) Kramers-restricted systems (C1 only)
(d) two-component (2c) open-shell systems (C1 only)

2. $rigw; RI-AC-G0 W0 and RI-AC-evGW variants construct the self-energy ΣC on the

imaginary axis using numerical integration. The obtained result is then analytically
continued to the real axis using Pade approximants. This ansatz scales roughly as
N 4 , and yields reliable results for valence orbitals, especially for HOMO and LUMO
quasiparticle energies. RI-AC-GW is the default variant for $rigw calculations. It is
available for the following starting points:

(a) scalar/non-relativistic closed-shell systems (up to D2h symmetry)

(b) scalar/non-relativistic open-shell systems (up to D2h symmetry)
(c) two-component (2c) Kramers-restricted systems (C1 only)

3. $rigw; RI-CD-G0 W0 and RI-CD-evGW variants construct the self-energy ΣC using

contour deformation. This ansatz scales roughly as N 4 (valence orbitals) - N 5 (core
orbitals), and yields reliable results for most orbitals. RI-CD-GW is invoked by the
contour keyword within the $rigw datagroup. It is available for the following starting
points:

(a) scalar/non-relativistic closed-shell systems (C1 only (future: up to D2h symme-

try))
(b) scalar/non-relativistic open-shell systems (C1 only (future: up to D2h symme-
try))
(c) two-component (2c) Kramers-restricted systems (C1 only)

Especially for 2c calculations the prefactor of 256 for a $gw calculation is reduced to
4-8 in a $rigw calculation, making calculations feasible also for large systems. This
prefactor reduction is valid for both $rigw variants, RI-AC-GW and RI-CD-GW .

Ad 5) $rick is a synonym for the keyword $rik in escf, however $rick does not interfere with
other programs for easier batch processing.
14.2. EXCITATION ENERGIES FROM BSE 299

Ad 6) The GW module integrated in escf is (since Turbomole 7.3) largely decoupled from
the iterative solver used for normal excited state calculations. Therefore the user is dis-
couraged to only calculate partial spectra. Always ALL excitations should be included in
the $gw approach, as the self-energy usually converges rather slowly with the number of
included excitations. If a system is too large, and this is no longer possible one can resort
to the $rigw approach. If the HOMO-LUMO GAP, or IP/EA of a system are the primary
interest of the user usage of the $rigw approach is always recommended. The output of
both approaches ($gw and $rigw) is processed by the same routines, and therefore they are
compatible in further production steps such as e.g. BSE calculations for properties.
Ad 7) Freezing orbitals with $freeze in combination with GW is only possible with $gw in
C1 symmetry, but not with $rigw.
Performance & Accuracy: Since Turbomole 7.3 specialized GW algorithms are implemented
in escf, leading to large speedups. The analytic continuation ($rigw) variant usually yields
self-energies with meV accuracy compared to standard $gw for HOMO/LUMO orbitals, and
is much more conservative with memory and CPU requirements. For two-component calcu-
lations (including spin-orbit interactions) also the prefactor of G0 W0 and evGW is largely
reduced in $rigw. The standard $gw variant is however accurate and reliable throughout
all quasiparticle states, also describing non-valence orbitals well and therefore recommended
whenever possible.
Possible source of errors: The DFT options used in dscf or ridft should not be altered
before starting escf. Otherwise erratic quasiparticle energies may be obtained. Also any
files containing excitations (sing, unrs) from other escf runs should be removed prior to the
GW run. Also a cbas must be set before starting escf.

14.2 Excitation energies from the Bethe–Salpeter equa-

tion

14.2.1 Theoretical background

The Bethe–Salpeter (BS) excitation energies ωn are obtained as solutions of a pseudo-

Hermitian eigenvalue equation, which has the same form as in time-dependent Hartree–Fock
(TD-HF) theory. For the sake of simplicity, only the equations for general references are
given below. The orbitals are assumed to be complex.
! ! ! !
A B Xn 1 0 Xn
= ωn . (14.7)
B∗ A∗ Yn 0 −1 Yn

In the following, the indices a, b, . . . (i, j, . . .) refer to unoccupied (occupied) spinors, the
indices p, q, r, . . . include occupied and unoccupied spinors. The definitions of the matrices
A and B differ in the one-particle energies, which are the orbital energies in case of TD-HF
and the GW quasi-particle energies εQP in case of BS. Also, the exchange interaction is
300 CHAPTER 14. MBPT CALCULATIONS

screened in case of BS,

∆εQP ¯
Aia,jb = ia,jb + (ai|jb) − (ji|ab) , (14.8)
Bia,jb = (ai|bj) − (ja¯|ib) . (14.9)

The exchange interaction (pq¯|rs), which is defined as

(pq¯|rs) =
X
(−1 )pq,tu (tu|rs) ⇔ W = −1 v , (14.10)
tu

is screened by the inverse of the dielectric function,

pq,rs = δpr δqs − (pq|rs) [χ0 (ω = 0)]rs,rs , (14.11)

X δra δsi + δri δsa
[χ0 (ω = 0)]rs,rs = , (14.12)
ia εQP
i − εQP
a

which depends on the GW quasi-particle energies. It can be expressed in terms of an

infinite-order perturbation expansion in the Coulomb interaction,

W =−1 v
(14.13)
=v + vχ0 (ω = 0)v + vχ0 (ω = 0)vχ0 (ω = 0)v + . . . .

To obtain W, the RI-approximation is introduced for the two-electron integrals. As a direct

result, the inversion  is replaced by the inversion of a matrix in the auxiliary basis:

W ≈ b> ĩ b =b> 1 + i + i2 + i3 + . . . b



−1 (14.14)
=b> 1 − i b,
X bP,ai bQ,ai
i = 2bχ0 (ω = 0)b> ⇔ iP Q = 2 . (14.15)
ia εQP
i − εQP
a

For further details, we refer to Ref. [33].

14.2.2 BSE features

The BSE method is implemented in the escf module. It is activated by the keyword $bse.
Its features and different variants are discussed in the following.

Reference determinant

The BSE method supports both Hartree–Fock and Kohn–Sham references, that is, orbitals
obtained from a dscf or ridft computation. Note that symmetry is only supported up to
D2h and its subgroups. In case of wave functions of higher symmetry, the orbitals have to
be converted using define’s option susy to a subgroup of D2h . Two-component references
($soghf) including spin-orbit interactions can be used.
14.2. EXCITATION ENERGIES FROM BSE 301

Excitations

Singlet, triplet and spin-unrestricted excitation energies based on the Bethe–Salpeter equa-
tion are available in escf. Similar to the RPA and TDA variants in TD-HF, the eigenvalues
of the full 2 × 2 “super-matrix” can be computed, or of A only. To define these options,
the same input blocks as for a TD-HF computation are read in: $scfinstab and $soes.
See Section 8 for details. For two-component calculations only the TDA variant is imple-
mented. Note that 2c-BSE is, compared to TD-DFT, not limited to closed shell cases but
also implemented for open shell cases.

Quasiparticle energies

The BSE method reads in quasiparticle energies obtained from a preceding GW calculation.
It supports all different GW methods available in escf. Since Turbomole 7.3 the number of
frozen orbitals can be different in the GW and BSE step. However qpenergies.dat files from
Turbomole 7.2 and earlier are not directly compatible with Turbomole 7.3 and higher, and
it is recommended to redo the GW calculation. The reason for this is that the quasiparticle
states for frozen orbitals are also projected since Turbomole 7.3, and always all orbitals are
therefore updated and reported in the qpenergies.dat file.

Quasiparticle energies in A and W

By default, the matrices A and W are set up using quasiparticle energies. To set up A
using KS/HF orbital energies instead, i.e.

Aia,jb = ∆εia,jb + (ai|jb) − (ji¯|ab) , (14.16)

the option noqpa has to be added to the $bse block in control. Similarly, if the denominator
in χ0 and, consequently, the static screened interaction should be constructed using KS/HF
orbital energies instead of GW quasiparticle energies, the option noqpw has to be set. The
quasiparticle energies are read in from the file qpenergies.dat, which has to be created
in a preceding GW calculation using escf. A different file name can be defined using the
option file in the $bse block.

Computation of the static screened interaction

The static screened interaction is obtained from the auxiliary matrix ĩ. By default, this is
computed by Cholesky decomposition and inversion of (1 − i)−1 . An alternative iterative
procedure,
ĩ(i+1) = 1 + i ī(i) , (14.17)

is activated by the option iterative. The convergence threshold (default: 1.0d-12) and
maximum number of iterations (default: 100) can be adjusted by the options thrconv and
iterlim, respectively.
302 CHAPTER 14. MBPT CALCULATIONS

Auxiliary basis sets

The static screened interaction is computed using the RI approximation, therefore, an auxil-
iary basis set is required. Since the integrals are similar to those in MP2, the cbas auxiliary
basis sets are recommended. They can be conveniently defined using the cc sub-menu of
define.

14.2.3 General recipe for a BSE calculation

1. dscf or ridft calculation to converge reference orbitals

2. Provide $gw or $rigw control flags and perform a GW calculation

3. Choose reasonable cbas auxiliary basis sets if not done in the GW step

4. escf calculation to obtain qpenergies.dat

5. Remove $gw and $rigw control flags

6. Set BSE control flags

7. Add the keyword $rik or $rick if not already present

8. escf calculation to obtain excitation energies

Chapter 15

Calculation of Vibrational
Frequencies and Vibrational
Spectra

Calculation of second derivatives of total energies leads to the molecular Hessian, which
enables prediction of vibrational frequencies and infrared spectra (within the harmonic ap-
proximation) as well as the application of improved algorithms for geometry optimization
and transition state search.
The aoforce module calculates analytically harmonic vibrational frequencies within the HF-
or (RI)DFT-methods for closed-shell- and spin-unrestricted open-shell-systems. Broken oc-
cupation numbers would lead to results without any physical meaning. Note, that RI is only
used partially, which means that the resulting Hessian is only a (very good) approximation
to exact second derivatives of the RIDFT-energy expression. Apart from a standard force
constant calculation which predicts all (symmetry allowed and forbidden) vibrational tran-
sitions, it is also possible to specify certain irreps for which the calculation has to be done
exclusively or to select only a small number of lowest eigenvalues (and eigenvectors) that
are generated at reduced computational cost.
Furthermore, the NumForce script allows the calculation of second derivatives for all methods
for which a program for analytic gradients is available in TURBOMOLE, i.e. the main use of
this script is the prediction of vibrational spectra at the MP2 level and for excited states
using RI-CC2 or TDDFT.
If force constant calculations result in imaginary frequencies, molecular distortions along
these normal modes should lower the energy. To distort the molecule, use the interactive
module vibration, output of the new coordinates is done to the general input file on
$newcoord.
Vibrational frequencies also enable calculation of the molecular partition function and thus
prediction of thermodynamic functions at temperatures other than 0 K and finite pressure

303
304 CHAPTER 15. VIBRATIONAL FREQUENCIES

(within the assumption of an ideal gas and no coupling between degrees of freedom). These
functions can be obtained with the interactive module Freeh, results are printed to standard
I/O.

Prerequisites
1. Both aoforce and even more NumForce require well converged SCF-/DFT-calculations
(e.g. $scfconv 8 and jobex [-ri] -gcart 4).
2. The maximum core memory the program aoforce is allowed to allocate should be
defined in the data group $maxcor; the recommended value is about 50% of the
available (physical) core memory (in case of RI-calculations subtract the memory
specified in $ricore).
3. To start aoforce in the lowest eigenvalue search mode, use the keyword $les. For
its use as well as other keywords dealing with the calculation of only some irreps, see
the reference guide part of this manual.
4. NumForce additionally requires the file gradient and will not work, if the calculation
is not done at a stationary point of the molecular total energy. For reliable results,
always use NumForce with the option -central (i.e. central differences) and be aware
of effects due to the step length (option -d real ;, default value is 0.02 a.u.). It is
strongly recommended to use NumForce in DFT calculations only with the option
weight derivatives in $dft, since this provides more accurate gradients and thus
frequencies, see Section 22.2.13.
5. The NumForce script can be run for different levels of theory, which means that the
binaries it calls have to be specified additionally. To perform calculations using the
RI approximation, call NumForce with the option -ri. MP2 and CC2 calculations
are requested via the options -level mp2 and -level cc2, respectively. NumForce
works also on the RI-RPA level with -level rirpa (note: the -ri option must be
used in this case). To select the correct option(s), use the explanations you get by
calling NumForce -h.

For a review of theory and implementation see refs. [236, 237].

Limitations

The aoforce code has presently a number of limitations one should be aware of:

• It can only handle basis sets up to at most g functions.

• Point groups with reducible E-representations (such as Cn and Cnh with n ≥ 3, Sn
with n ≥ 5, or T and Td )
• Frozen internal or Cartesian coordinates are not recognized. aoforce will always
evaluate the full Hessian matrix.
15.1. ANALYSIS OF NORMAL MODES IN TERMS OF INTERNAL COORDINATES305

15.1 Analysis of Normal Modes in Terms of Internal Co-

ordinates

A note in advance: The analysis of normal modes can (at nearly no computational cost)
always be redone as long as you keep a copy of the file hessian.
A general prerequisite for this option is that you have defined a set of non-redundant co-
ordinates for all 3N-6 (3N-5) degrees of freedom of your molecule. To make sure that this
is the case, you should switch off redundant coordinates (currently, this is only possible by
manually removing the data group $redundant and also removing the entry redundant on
in $optimize). Run define to generate non-redundant coordinates by using the iaut
command in the internal coordinate menu (or by creating them manually via idef). We
recommend to use the irem command first to delete all previous definitions of internal co-
ordinates. See Section 4 for further details. If the molecules point group is not C1 , define
will set some of the coordinate to status d (display) or i (ignore). Use the ic command to
change all coordinates to k. You can also achieve this by editing in the $intdef data-group
manually.
The analysis in internal coordinates is switched on by adding a line in the data-group
$drvopt that has the following syntax:

analysis [only] intcoord [print print-level ]

Keywords in square brackets are optional. If only is added, the program assumes that the
file hessian exists and runs only the analysis part of aoforce. The program will give the
following output (controlled by the print level given in parenthesis):

• diagonal elements of the Hessian in internal coordinates (force constants of bonds,

angles, etc.) (print level 0)

• complete force constant matrix in internal coordinates (print level 2)

• normal modes in terms of internal coordinates (print level 1)

• Potential energy contributions Ṽijn , defined as

Ṽijn = Lni Lnj Fij /ω n

where Lni are the elements of the normal coordinate belonging to mode n and Fij are
the elements of the force constant matrix, both expressed in the internal coordinate
basis; ω is the related eigenvalue. The program will list the diagonal contributions
Ṽiin (print level 1), the off-diagonal contributions Ṽijn + Ṽjin = 2Ṽijn (print level 2 for up
to 10 atoms, else print level 10) and the gross contributions i Ṽijn (print level 1).
P

• Based on these quantities, the program will give an assignment of normal modes by
listing all internal coordinates with large diagonal or gross contributions (print level
0).
306 CHAPTER 15. VIBRATIONAL FREQUENCIES

Note that for large molecules or complicated topologies the B-matrix (that is used to trans-
form from Cartesian coordinates into internal coordinates and vice versa) may become sin-
gular. In this case only the normal modes in the internal coordinate basis can be listed.

15.2 Calculation of Raman Spectra

Vibrational Raman scattering cross sections are computed in the approximation of the po-
larizability theory from derivatives of the frequency-dependent polarizability tensor with
respect to normal modes of vibration,
 
dσ
= kω ci α02 (ω) + ca γ 02 (ω) .


dΩ

Here, α02 (ω) and γ 02 (ω) denote the isotropic part and the anisotropy of the differentiated po-
larizability tensor, respectively. The coefficients ci and ca depend on the scattering geometry
and the polarization of the incident and scattered radiation. The factor

~ (ω − ωv )4 gv
kω =
4π20 c4 2ωv

includes the frequency ωv and the degeneracy gv of the vibration. c is speed of light and 0
stands for the dielectric constant of vacuum.
Computation of Raman spectra with TURBOMOLE is a three-step procedure. First, vibrational
frequencies and normal modes are calculated by aoforce. Cartesian polarizability deriva-
tives are computed in the second step by egrad, see Section 8.4.10. Finally, the program
intense is used to project the polarizability derivatives onto vibrational normal modes and
to compute Raman scattering cross sections which are written out along with vibrational
frequencies and normal modes. The script Raman can be used to perform all these steps
automatically.

15.3 Calculation of VCD Spectra

VCD intensities are proportional to the rotational strengths, which are defined as scalar prod-
uct between the electric and magnetic transition dipole moments. The rotational strength
for the transition of the nth vibrational normal mode in the electronic ground state is

Rn = Im(µel,n · µmag,n ).

Within the harmonic approximation, the electric and magnetic transition dipole moments
can be written as r
~ X λ λ
(µel,n )β = Pαβ Sα,n
ωn
λα
p X
λ λ
(µmag,n )β = − 2~3 ωn Mαβ Sα,n .
λα
15.4. VIBRATIONAL FREQUENCIES WITH FIXED ATOMS USING NUMFORCE307

λ λ
Pαβ is the so called atomic polar tensor (APT), Mαβ the so called atomic axial tensor (AAT)
λ
and Sα,n the transformation matrix from Cartesian coordinates to normal coordinates.
ωn is the frequency of the nth vibrational normal mode, α and β describe Cartesian coordi-
nates and λ labels the atom nuclei.

Both the APT and the AAT are divided in an electronic and a nuclear contribution. Accord-
ing to Chem. Phys. Lett. 252, 221 (1996), the electronic contribution of the AAT within
the coupled perturbed Hartree-Fock (CPHF) formalism is given by

N Nbf
occ X  N
Rλ B B Rλ
X X
λ
Iαβ = cµi cνi hχµ α |χν β i + cµi cνp Uipβ hχµ α |χν i
i=1 µ,ν=1 p=1
N N 
Rλ B Rλ B
X X
+ cµp cνi Uipα hχµ |χν β i + cµp cνq Uipα Uiq β hχµ |χν i .
p=1 p,q=1

To compute VCD spectra with TURBOMOLE, mpshift and subsequently aoforce have to be
B
called. First, if $vcd is set, mpshift calculates the disturbed Uip . In the second step vibra-
tional frequencies, normal modes and VCD rotational strengths are calculated by aoforce.
At present, calculation of VCD spectra is possible at Hartree-Fock and DFT level (for LDA,
GGA and hybrid-GGA functionals) in C1 symmetry. Since the angle between the electric
and magnetic transition dipole moments is not gauche invariant, it will only be printed out
if the distance between the center of mass and the origin is smaller than 1e-03 a0. The script
VCD can be used to perform these two steps automatically.

15.4 Vibrational frequencies with fixed atoms using NumForce

The NumForce script provides with the option -frznuclei a possibility to do a vibrational
analysis with fixed atoms. The atoms for which the Cartesian coordinates should frozen
have to be marked in $coord with a "f" behind the atom type. The frozen coordinates
will be skipped during the numerical evaluation of the force constant matrix; instead all off-
diagonal elements of the force constant matrix which refer to one or two frozen coordinates
will be set to zero, while the diagonale elements for the frozen coordinates will be set to an
arbitrarily chosen large value.
This feature is mainly intended to allow for a vibrational analysis in embbeded cluster
calculations e.g. for defects in ionic crystals. The vibrational analysis uses a kind of “frozen
phonon” approximation which corresponds to setting the masses of the fixed atoms to infinity,
i.e. decoupling the fixed atoms mechanically from the “mechanically active” subsystem. The
resulting vibrational frequencies will thus only provide good approximations to the true
(harmonic) frequencies for such modes for which the mechanical coupling to the embedding
environment is negligible. In particular the frequencies of stretch modes which involve
bonds between the “mechanically active” subsystem and atoms with frozen coordinates will
be strongly affected by this approximation.
308 CHAPTER 15. VIBRATIONAL FREQUENCIES

Note:

• The -frznuclei is not compatible with the polyhedral difference algorithm. It can
only be used with central differences which should be enforced with the -central
option.

• If the option -frznuclei is switched on, the program assumes that the constraints
enforced by fixing coordinates remove the six external degrees of freedom for on overall
rotation or translation of the system and therefore the hessian matrix is not projected
onto the subspace of internal coordinates. Fixing the coordinates of only one or two
atoms might does lead to some artificial small, but non-zero frequencies.

• Zero-point vibrational energies calculated with the -frznuclei option are only mean-
ingful for comparison of systems with the same mechanically active atoms and similar
embedding, as the contributions from the frozen coordinates are not included.

15.5 Interface to hotFCHT

aoforce supports the generation of input files for the hotFCHT code (version 2.0 and later)
of R. Berger and co-workers, see hotFCHT, which allows for the calculation of Franck-
Condon factors. Just include the keyword $hotfcht in the control file. The option is also
active in analysis mode, that is as long as you still have the data group $hessian (in the
control file or in a file referenced in the control file) you can always use aoforce (in analysis
mode) to quickly generate the hotFCHT input. The program will write three files. The first
one, hotfcht_header.inp contains a collection of the most important keywords of hotFCHT
(set to some default values, please adapt to your needs) and list of all atomic masses (either
TURBOMOLE’s default masses or the ones given in the $atoms data group). The other
two, hotfcht_data_i.inp and hotfcht_data_f.inp contain the vibrational frequencies,
normal modes and the names of the irreducible representations of the normal modes. In the
former file, these data are associated with the hotFCHT keywords for the initial state, while
the latter file contains the same data, but associated with the keywords for the final state.
In order to run a hotFCHT calculation, you need to optimize the structures of two electronic
states (usually the electronic ground state and an excited or ionized state) and obtain the
harmonic force fields for both, using either aoforce or NumForce. In order to generate the
hotFCHT input, just concatenate the hotfcht_header.inp file (from any of the two calcu-
lations) and the hotfcht_data_i.inp file from the calculation that refers to the initial state
(e.g. the ground state in case of an absorption spectrum) and the hotfcht_data_f.inp file
from the calculation of the final state (the excited state in case of an absorption spectrum).
Carefully edit the keywords in the header of the resultant file and run hotFCHT (please,
refer to the hotFCHT documentation for further information).
Chapter 16

First order electron-vibration

coupling

16.1 Theoretical background

At the effective single-particle level, the Hamiltonian of the coupled system of electrons and
vibrations is given by [238]

Ĥ = Ĥ e + Ĥ v + Ĥ ev , (16.1)

where the first term Ĥ e describes the electronic system and the second term Ĥ v the vibra-
tional degrees of freedom. The last term in the Hamiltonian
XX
Ĥ ev = dˆ†µ λα ˆ †
µν dν (b̂α + b̂α ) (16.2)
µν α

describes the first order electron-vibration (EV) interaction. The EV coupling constants are
given as
1/2 X
dĤ1e

α ~
λµν = µ ν Aαχ, (16.3)
2ωα χ
dχ

where χ = (k, u) is a shorthand notation that refers both to the displacement of atom k
from the equilibrium value of the position R ~ k along the Cartesian component Rk,u with
√
u = x, y, z as well as the index pair itself. Furthermore, Aα α
χ = Cχ / Mk are the mass-
normalized normal modes, obtained from the eigenvectors Cχα of the dynamical matrix as
calculated from the aoforce module [238].

309
310 CHAPTER 16. EVIB CALCULATIONS

16.2 evib features

The evib module, implemented in TURBOMOLE, allows to calculate the matrix elements of
the first order derivative of the Kohn-Sham operator with respect to atomic displacements
χ
e dĤ1e
Hµν,χ = µ ν , (16.4)
dχ
which are required to obtain the first order EV coupling constant, as given in Eq. (16.3) [238].
Limitations:

• only c1 symmetry at the moment,

• RI-approximation only partely implemented.

16.3 General usage of evib

Calculating the matrix elements given in Eq. (16.4) consists of two steps. First a force
constant calculation using aoforce is performed, where the following control flags have to
be added:

$nosalc ,

$sijuai_out

this will save the derivative of the density matrix, which are necssary for the subsequent
evib run.
e
The matrix elements of Hµν,χ (Eq. (16.4)) are stored by default in binary format in the file
dfdxi.dat (dfdxi_a.dat and dfdxi_b.dat for UHF), using formatted Fortran output with a
e
record length of 8 bytes for each matrix element. The matrix elements Hµν,χ with χ = (k, u)
are in the atomic orbital (AO) and ordered as follows: (i) Cartesian component u = x, y, z,
(ii) atom number k, and (iii) (Nao + 1) ∗ Nao /2 matrix elements of the upper part of the
triangular matrix for the µν indices.
Optionally,

$dfdxi textout

can be used to generate text output of the matrix elements.

Additonally the first derivatives of the orbital energies (Kohn-Sham eigenvalues) with respect
to the atomic displacements di /dχ are calculated and stored in the text file dEidR.dat
(dEidR_a.dat and dEidR_b.dat for UHF).
Chapter 17

Calculation of NMR Shieldings

The program mpshift calculates nuclear magnetic resonance (NMR) shielding constants
using the GIAO (Gauge-Including Atomic Orbital) method.
At present the following methods are implemented:

HF-SCF the coupled-perturbed Hartree–Fock (CPHF) equations in the AO basis are

solved using a semi-direct iterative algorithm [39] similar to dscf. The (multipole-
accelerated-) resolution-of-the-identity-fitting approximation is available for the
Coulomb term (MARI-J) [42]. Due to the effective screening approach the ex-
change part shows a low-order scaling.

DFT using either non-hybrid functionals where no iterations are needed [239] or
hybrid functionals where the same algorithm as at the HF-SCF level is used.
Meta-GGAs and the XCfun library can be further used [42]. Moreover, LibXC
and range-separated functionals are further supported, see also chapter 6.

MP2 semi-direct method, see ref. [40] and [240].

The following Hamiltonians are available in addition to the usual non-relativistic one:

ECP In molecules with ECP-carrying atoms, chemical shieldings on all the other
atoms can be computed with mpshift in the way suggested by van Wüllen [241].
ECPs can be used to treat scalar-relativistic effects on neighboring atoms [43].

X2C A scalar-relativistic or spin-free all-electron exact two-component Hamiltonian

can be used to calculate the NMR shielding tensor of heavy elements [41]. A
finite nucleus model based on a Gaussian charge distribution is available for
the scalar potential and the vector potential. This model can be also used with
the DLU-X2C Hamiltonian below. Grids with an increased number of radial
points [138] (e.g. gridsize 4a) should be used. Moreover, it is recommended

311
312 CHAPTER 17. SHIELDINGS

to use NMR-tailored basis sets, i.e. the x2c-XVPall-s (X=S,TZ) type bases.
These employ additional tight functions to accurately sample the density in
the vicinity of the nuclei.

DLU-X2C A local X2C Hamiltonian is further available and recommended. The diag-
onal local approximation to the unitary decoupling transformation (DLU) is
employed. This results in a very efficient algorithm. The error introduced by
the DLU scheme is negligible. For details on the theory, implementation and
application please see ref. [41].

17.1 Prerequisites
1. mpshift needs converged MO vectors from a SCF or DFT run (dscf or ridft)

2. For SCF or DFT calculations, no specifications have to be made in the control file
but currently only closed shell cases are implemented in mpshift.

3. To perform an MP2 calculation of the NMR shieldings, you have to prepare the input
with mp2prep -c

17.2 How to Perform a SCF or DFT Calculation

All you have to do for running mpshift is typing mpshift at the shell level.
The results of a SCF or DFT calculation (the trace of the total shielding tensors, its
anisotropy and the CPHF contribution for each symmetry distinct atom) are written into
the control file after the keyword $nmr <rhf/dft> shielding constants.
This data group is write-only for mpshift, but you can utilize it for graphical rendering of the
calculated NMR spectra and for a quick overview of the results. A more detailed output with
the complete shielding tensors can be found in the output of mpshift, so it is recommended
to put the output in a file when calling the program. For version 7.5 the solver for the
CPHF equations has been reimplemented based on the Davidson algorithm used by escf
and aoforce. This utilizes the keyword $shiftconv (default 7, i.e. a residuum threshold
of 10−7 ) to check for the convergence of the perturbed orbitals and density. Keywords for
the calculation with non-default settings are given in Sec. 22.2.28.

17.3 How to Perform a MP2 calculation

To perform an MP2 calculation of the NMR shieldings you have to prepare the input with
mp2prep -c.
mpshift will then calculate both the SCF and MP2 shielding constants. The result is written
into the control file after the keyword $nmr mp2 shielding constants. In addition, the
HF shielding constants are given after $nmr rhf shielding constants.
17.4. CHEMICAL SHIFTS 313

The script mp2prep will create the keywords

$csmp2
$thize .10000000E+10
$mointunit
type=intermed unit=61 size=0 file=halfint
type=1112 unit=63 size=0 file=moint#1
type=1122 unit=64 size=0 file=moint#j
type=1212 unit=65 size=0 file=moint#k
type=1212a unit=70 size=0 file=moint#a
type=gamma#1 unit=71 size=0 file=gamma#1
type=gamma#2 unit=72 size=0 file=gamma#2
type=dtdb#1 unit=76 size=0 file=dtdb#1
type=dtdb#2 unit=77 size=0 file=dtdb#2
$traloop 1
$statistics mpshift

and starts a statistics run of mpshift (by calling mpshift). If the resulting disk space re-
quirement exceeds the automatically detected free disk space on your system, it will increase
$traloop and run a statistics run again. This will be done as long as your free disk space
is not sufficient for the calculation.
If the mp2prep script fails to run on your system, try to use the -p option or do the procedure
described above by hand. Call mp2prep -h for more information about mp2prep.

17.4 Chemical Shifts

NMR shifts are obtained by comparing nuclear shieldings of your test compound with a
reference molecule (δsubst = δref + σref − σsubst ). Therefore you have to choose a reference
molecule with a well-known shift for which you can easily calculate the absolute shielding
constant. This implies a certainty about the geometry, too. Furthermore you have to use
the very same basis set for corresponding atoms to minimize the basis set influence.

Keywords for the module Mpshift

A list of keyword for the module mpshift can be found in Section 22.2.28.

17.5 Other Features

• The mpshift program can be restarted with the old CPHF solver at any stage of
computing, since all intermediate results are written into the file restartcs. In
case of an external program abort you have to remove the $actual step flag (by
314 CHAPTER 17. SHIELDINGS

the command actual -r or using an editor). mpshift analyses this file and decides
where to continue. The new solver also allows to restart calculations.

• Vibrational circular dichroism (VCD) spectra can be calculated using the gallier
script utilzing the aoforce module [43].

• The conductor-like screening model (COSMO) to account for counterions and solva-
tion effects can be selected [42].

• NMR shielding tensors can be calculated for given nuclei only by setting $nucsel in
the control file.

• The default maximum number of iterations for the CPHF procedure is set to 30. It
can be increased by adapting $csmaxiter in the control file.

• mpshift is parallelized by OpenMP. The corresponding environment variables have

to be set previously to running mpshift.

• Nucleus-independent chemical shieldings (NICS) can be calculated be setting $nics

in the control file and then listing the Cartesian coordinates in atomic units just as
the coordinates of the molecule itself. For details, see Section 22.2.28.

• The (un)perturbed density matrix can be stored on disk by setting $gimic in the con-
trol file. This matrices are required as input for the gauge-including magnetically in-
duced currents (GIMIC) method [242,243], see https://github.com/qmcurrents/gimic/.
This method allows to study electron delocalization pathways and to estimate the de-
gree of aromaticity [244].

17.6 Known Limitations

• Only closed shell calculations are currently supported.

• Molecular point groups that contain reducible e representations are not supported
(Cn , Cnh with n > 2).

• Spin–orbit coupling can not be considered yet.

Chapter 18

Embedding and Solvation Effects

TURBOMOLE provides provides a number of different embedding schemes to model the inter-
action of the quantum system with environments. They reach from a simple point charge
embedding to the continuum solvation model cosmo and the atomistic polarizable embed-
ding.
Most of these schemes ignore each other. So they should not be used in combination unless
it has been tested that the chosen combination works indeed correctly together.
The embedding schemes differ also in the extent to which they have been implemented in
the different programs and for different functionalities.

18.1 Charge and multipole embedding

Point Charge Embedding: The embedding of the quantum system in a electric po-
tential of an (non-periodic) set of charges and multipole moments is driven by the data
group $point_charges. In the simplest case it has the structure:

$point_charges
<x> <y> <z> <q>

where <x>, <y>, <z> are the coordinates and <q> the value of the point charge.
The point charge embedding is implemented in the dscf, ridft, grad, rdgrad, escf, ricc2,
ccsdf12, and pnoccsd programs and can be used essentially with every method and for all
properties including gradients with respect to nuclear coordinates. Exceptions are (auxiliary)
basis set gradients, analytic second derivatives (aoforce (but they can be computed semi-
numerical with NumForce.
For QM/MM applications it is also possible to compute the forces that the quantum system
exerts on the point charges. For further details about available options and the input see
Sec. 22.2.9.

315
316 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

Point Mutipole Embedding: In addition to point charges one can also use point
multipoles up to octupole moments. The input uses generalization of the point charge
input:

$point_charges mxrank=3
<x> <y> <z> <q> <qx> <qy> <qz> <qxx> <qyy> <qzz> <qxy> <qxz> <qyz> ...

The value of keyword mxrank defines the maximum multipole rank (0=charge, 1=dipole,
2=quadrupole, 3=octupole) and the tensor components are given for each multipole site
after the coordinates in canonical order. For the components of the octupole moment the
canonical order is:

xxx yyy zzz xxy xxz xyy yyz xzz yzz xyz

The multipole embedding (beyond charges) is only implemented for (excitation) energies.
Gradients are not (yet) available and symmetry can not be used.

Gaussian Smeared Charges: Instead of point charges one use Gaussian charge dis-


tributions of the form G(~r) = N exp − α(~r − ~r0 )2 . The input format is in this case:

$point_charges gaussians
<x> <y> <z> <q> <alpha>

The normalization factor N is determined internally such that the total (integrated) charge
R
of the distribution is q = R3 G(~r)dτ .
For Gaussian charge distributions gradients are available, but symmetry can not be used.
18.2. TREATMENT OF SOLVATION EFFECTS WITH COSMO 317

18.2 Treatment of Solvation Effects with Cosmo

The Conductor-like Screening Model [245] (Cosmo) is a continuum solvation model (CSM),
where the solute molecule forms a cavity within the dielectric continuum of permittivity ε
that represents the solvent. The charge distribution of the solute polarizes the dielectric
medium. The response of the medium is described by the generation of screening charges
on the cavity surface.
CSMs usually require the solution of the rather complicated boundary conditions for a
dielectric in order to obtain the screening charges. Cosmo instead uses the much simpler
boundary condition of vanishing electrostatic potential for a conductor,

Φtot = 0.

This represents an electrostatically ideal solvent with ε = ∞. The vector of the total
electrostatic potential on the cavity surface segments is determined by the solute potential
Φsol , which consist of the electronic and the nuclear part, and the vector of the screening
charges q,

Φtot = Φsol + Aq = 0.

A is the Coulomb matrix of the screening charge interactions. For a conductor, the boundary
condition Φtot = 0 defines the screening charges as

q = −A−1 Φsol .

To take into account the finite permittivity of real solvents, the screening charges are scaled
by a factor.
ε−1
f (ε) =
ε + 12
q? = f (ε)q

The deviation between the Cosmo approximation and the exact solution is rather small.
For strong dielectrics like water it is less than 1%, while for non-polar solvents with ε ≈ 2 it
may reach 10% of the total screening effects. However, for weak dielectrics, screening effects
are small and the absolute error therefore typically amounts to less than one kcal/mol. As
shown in [246] ions can be described more accurately by using the scaling factor f (ε) = ε−1
ε+0 .
This also leads to results (e.g. by comparing solvation free energies) more or less identical to
IEFPCM or SS(V)PE. This is possible since TURBOMOLE version 7.1 by adding the keyword
ions to the epsilon option in the $cosmo section (see 22.2.11).
The dielectric energy, i.e. the free electrostatic energy gained by the solvation process, is
half of the solute-solvent interaction energy.
1
Ediel = f (ε)q† Φsol
2
The total free energy of the solvated molecule is the sum of the energy of the isolated system
calculated with the solvated wave function and the dielectric energy

E = E(Ψsolv ) + Ediel .
318 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

A Cosmo energy calculation starts with the construction of the cavity surface grid. Within
the SCF procedure, the screening charges are calculated in every cycle and the potential
generated by these charges is included into the Hamiltonian. This ensures a variational
optimization of both the molecular orbitals and the screening charges and allows for the
evaluation of analytic gradients.

Radii based Cavity Construction: In order to ensure a sufficiently accurate and

efficient segmentation of the molecular shaped cavity the COSMO implementation uses a
double grid approach and segments of hexagonal, pentagonal, and triangular shape. The
cavity construction starts with a union of spheres of radii Ri + RSOLV for all atoms i. In
order to avoid problems with symmetric species, the cavity construction uses de-symmetrized
coordinates. The coordinates are slightly distorted with a co-sinus function of amplitude
AMPRAN and a phase shift PHSRAN. Initially a basis grid with NPPA segments per atom
is projected onto atomic spheres of radii Ri + RSOLV . In order to avoid the generation of
points in the problematic intersections, all remaining points, which are not in the interior of
another sphere, are projected downwards onto the radius Ri . In the next step a segment grid
of NSPH segments per H atom and NSPA segments for the other atoms is projected onto
the surface defined by Ri . The basis grid points are associated to the nearest segment grid
centers and the segment coordinates are re-defined as the center of area of their associated
basis grid points, while the segment area is the sum of the basis grid areas. Segments
without basis grid points are discarded. In order to ensure nearest neighbor association for
the new centers, this procedure is repeated once. At the end of the cavity construction the
intersection seams of the spheres are paved with individual segments, which do not hold
associated basis grid points.

Density based Cavity Construction: Instead of using atom specific radii the cavity
can be defined by the electron density. In such an isodensity cavity construction one can
use the same density value for all atoms types or the so-called scaled isodensity values. In
the later approach different densities are used for the different atom types. The algorithm
implemented in Turbomole uses a marching tetrahedron algorithm for the density based
cavity construction. In order to assure a smooth density change in the intersection seams
of atoms with different isodensity specification, this areas are smoothened by a radii based
procedure.

Radii based Isosurface Cavity: A cavity construction algorithm based on the tri-
angulation of an iso-surface is available as an alternative to the radii or density based con-
struction. It overcomes deficiencies which have become apparent for the original COSMO
standard cavity, especially in concave regions of the molecular shaped cavity. The new
construction, called FINE Cavity, is described in details in [247].
To enable the new radii based isosurface cavity the keyword $cosmo_isorad has to be added
to the control file.
18.2. TREATMENT OF SOLVATION EFFECTS WITH COSMO 319

A-Matrix Setup: The A matrix elements are calculated as the sum of the contributions
of the associated basis grid points of the segments k and l if their distance is below a certain
threshold, the centers of the segments are used otherwise. For all segments that do not
have associated basis grid points, i.e. intersection seam segments, the segment centers are
used. The diagonal elements Akk that represent the self-energy of the segment are calculated
√
via the basis grid points contributions, or by using the segment area Akk ≈ 3.8 ak , if no
associated basis grid points exist.

Outlying charge correction: The part of the electron density reaching outside the
cavity causes an inconsistency that can be compensated by the "outlying charge correction".
This correction will be performed at the end of a converged SCF or an iterative MP2
calculation and uses an outer surface for the estimation of the energy and charge correction
[248]. The outer surface is constructed by an outward projection of the spherical part of the
surface onto the radius Ri + ROU T F ∗ RSOLV . It is recommended to use the corrected
values.

Numerical Frequency Calculation: The calculation of harmonic frequencies raises

the problem of non-equilibrium solvation in the Cosmo framework, because the molecular
vibrations are on a time scale that do not allow a re-orientation of the solvent molecules.
Therefore, the total response of the continuum is split into a fast contribution, described by
the electronic polarization, and a slow term related to the orientational relaxation. As can
be shown [249] the dielectric energy for the disturbed state can be written as

d 1 1
Ediel = f (ε)q(P0 )Φ(P0 ) + f (n2 )q(P∆ )Φ(P∆ ) + f (ε)q(P0 )Φ(P∆ ),
2 2
where P∆ denotes the density difference between the distorted state and the initial state with
density P0 . The interaction is composed of three contributions: the initial state dielectric
energy, the interaction of the potential difference with the initial state charges, and the
electronic screening energy that results from the density difference. The energy expression
can be used to derive the correspondent gradients, which can be applied in a numerical
frequency calculation. Because the Cosmo cavity changes for every distorted geometry the
initial state potential has to be mapped onto the new cavity in every step. The mapped
potential of a segment of the new cavity is calculated from the distance-weighted potentials
of all segments of the old cavity that fulfill a certain distance criterion. The mapped initial
state screening charges are re-calculated from the new potential.

18.2.1 Iterative COSMO-MP2

For ab initio MP2 calculations within the CSM framework three alternatives can be found
in the literature [250]. The first approach, often referred to as PTE, performs a normal
MP2 energy calculation on the solvated HF wave function. The response of the solvent,
also called reaction field, is still on the HF level. It is the only of the three approaches
that is formally consistent in the sense of second-order perturbation theory [251,252]. In the
320 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

so-called PTD approach the vacuum MP2 density is used to calculate the reaction field. The
third approach, often called PTED, is iterative so that the reaction field reflects the density
of the first-order wave function. In contrast to the PTE approach the reaction field, i.e. the
screening charges, change during the iterations until self consistency is reached. Gradients
are available on the formally consistent PTE level only [253].

18.2.2 COSMO-CC2 for ground-state calculations

The ground state energy and gradient is available for COSMO-CC2 within the post-SCF
[254] reaction-field scheme with a CCS-like approximation for the density that used to eval-
uate the reaction field. In the post-SCF scheme, the ground-state reaction field is first deter-
mined self-consistently at the COSMO-HF level. In the subsequent correlation treatment,
the correlation effect to the reaction-field is calculated from the correlation contribution to
the unrelaxed density and included in the equations for the ground-state wavefunction pa-
rameters. In order to set the input file for the calculation of ground-state gradient, energy
and relaxed properties the following data groups must be included in the control file.

$reaction_field
post-SCF
ccs-like
$cosmo
epsilon= 50.000
rsolv= 1.30
$cosmo_atoms
...

18.2.3 Vertical excitations and Polarizabilities for TDDFT, TDA

and RPA:

The escf program accounts for the Cosmo contribution to the excitation energies and
polarizabilities. The Cosmo settings have to be defined for the underlying Cosmo dscf or
ridft calculation. In case of the excitation energies the solvent response will be divided
into the so-called slow and fast term [249, 255]. The screening function of the fast term
depends on the refractive index of the solvent which can be defined in the input. If only
the Cosmo influence on the ground state should be taken into account we recommend
to perform a normal Cosmo calculation (dscf or ridft) and to switch off Cosmo (i.e.
deactivate $cosmo) before the escf calculation.

18.2.4 The Direct COSMO-RS method (DCOSMO-RS):

In order to go beyond the pure electrostatic model a self consistent implementation of

the COSMO-RS model the so-called "Direct COSMO-RS" (DCOSMO-RS) [256] has been
implemented in ridft and dscf.
18.2. TREATMENT OF SOLVATION EFFECTS WITH COSMO 321

COSMO-RS (COSMO for Real Solvents) [257,258] is a predictive method for the calculation
of thermodynamic properties of fluids that uses a statistical thermodynamics approach based
on the results of COSMO SCF calculations for molecules embedded in an electric conductor,
i.e. using f (ε) = 1. The liquid can be imagined as a dense packing of molecules in the perfect
conductor (the reference state). For the statistical thermodynamic procedure this system
is broken down to an ensemble of pair wise interacting surface segments. The interactions
can be expressed in terms of surface descriptors. e.g. the screening charge per segment
area (σt = qt /at ). Using the information about the surface polarity σ and the interaction
energy functional, one can obtain the so-called σ-potential (µS (σ; T )). This function gives
a measure for the affinity of the system S to a surface of polarity σ. The system S might
be a mixture or a pure solvent at a given temperature T . Because the parabolic part of
the potential can be described well by the Cosmo model, we subtract this portion form the
COSMO-RS potential:

µ̃S (σ; T ) = µS (σ; T ) − (1 − f (ε))c0 σ 2 .

The parameter c0 can be obtained from the curvature of a COSMO-RS σ-potential of a

nonpolar substance, e.g. hexane. Thus, the remaining part of the chemical potential of a
compound i with mole fraction xi in the mixture Si can be expressed as:
m
µ ∼
X
i
= fpol at µ̃S (σ; T ) + µiC,S + kT ln(xi ).
t=1

where the combinatorial term µiC,S accounts for effects due to the size and shape differences
of the molecules in the mixture and at denotes the area of segment t. The kT ln(xi ) can be
skipped for infinite dilution. The factor fpol has been introduced to account for the missing
solute-solvent back polarization. The default value is one in the current implementation.
The free energy gained by the solvation process in the DCOSMO-RS framework is the sum
of the dielectric energy of the Cosmo model and the chemical potential described above:
1
Ediel,RS = f (ε)q† Φsol + µi = Ediel + µi
2
From the above expression the solvent operator V̂ RS can be derived by functional derivative
with respect to the electron density:
m m
X f (ε)qt + q ∆RS t
X qt∆RS
V̂ RS = − = V̂ cos − .
t=1
|rt − r| t=1
|rt − r|

Thus, the solvation influence of the COSMO-RS model can be viewed as a correction of the
Cosmo screening charges qt . The additional charges denoted as qt∆RS can be obtained from
q∆RS = −A−1 Φ∆RS , where the potential Φ∆RS arises from the chemical potential of the
solute in the solvent:
 
δ µ̃S
φ∆RS
t = a t .
δq q=qt

In order to get a simple and differentiable representation of the COSMO-RS σ-potential

µS (σ; T ), we use equally spaced cubic splines. An approximate gradient of the method has
322 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

been implemented. DCOSMO-RS can be used in SCF energy and gradient calculations (ge-
ometry optimizations) with dscf, ridft, grad, and rdgrad. Please regard the restriction of
the DCOSMO-RS energy explained in the keyword section 22.2.11. Because the DCOSMO-
RS contribution can be considered as a slow term contribution in vertical excitations it does
not have to be taken into account in response calculations. For the calculation of vertical
excitation energies it is recommended to use the mos of a DCOSMO-RS calculation in a
Cosmo response calculation (see above).

18.2.5 Solvation effects on excited states using COSMO in ricc2:

The COSMO approach has been recently implemented into the ricc2 module of TURBO-
MOLE for excitation energies with CC2 and ADC(2). The ADC(2) method has been impl-
mented in combination with the iterative PTED and the more economical post-SCF reaction
field schemes. CC2 is currently only available in combination with the post-SCF reaction
field scheme.

Iterative COSMO-ADC(2) within the PTED scheme In the framework of the

old implementation of COSMO-ADC(2) within the PTED reaction field (RF) scheme it is
possible to equilirate the solvent charges for the ground state at MP2 or any excited state.
Using the methods CCS/CIS or ADC(2) the implementation is complete, for CC2 or higher
methods, however it still has to be proven if there are terms missing.
The PTED implementation of COSMO-ADC(2) contains contributions to the off-diagonal el-
ements of the one-electron density. Furthermore the energy contributions for non-equilibrated
states can be calculated. Non-equilibrated means in this sense, that the slow part of the
solvent charges (described by f (ε)) are still equilibrated with a given initial state, while
the fast electronic part of the solvent charges (described by f (n2 )) are in equilibrium with
the target state. To handle this one has to do a macro iteration, like in MP2. This macro
iteration can be managed with the script ’cc2cosmo’ which is the same as ’mp2cosmo’ but
using ricc2 instead of rimp2 or mpgrad. To set up the basic settings one can reuse the
cosmoprep module, note that the refractive index must be specified ’refind’ when observing
excited states. To specify the state to which the solvent charges should be equilibrated one
inserts the keyword ’cosmorel state=(x)’, where the ground state (x) is used normally but
can be replaced by any requested excited state. Make sure to request relaxed properties for
any desired state, otherwise the COSMO macro iteration will not work in ricc2. The off-
diagonal contributions mentioned above can be switched off by setting the keyword ’nofast’
in $cosmo. A typical input might look like:

$ricc2
adc(2)
$excitations
irrep=a’ multiplicity= 1 nexc= 1 npre= 1 nstart= 1
irrep=a" multiplicity= 1 nexc= 1 npre= 1 nstart= 1
exprop relaxed states=all
18.2. TREATMENT OF SOLVATION EFFECTS WITH COSMO 323

$response
fop relaxed
$cosmo
epsilon= 50.000
rsolv= 1.30
refind= 3.0000
cosmorel state=(a" 1)
# nofast
$cosmo_correlated
$cosmo_atoms
...

This would deliver an excited state calculation for the lowest singlet A’ and A" excitations
using the ADC(2) method. The solvent charges are equilibrated to state 11 A" and the non-
equilibrium energy contributions for the MP2 ground state and the 11 A’ excited state are
calculated furthermore. All contributions to the one-electron density are included since the
proper keyword is commented out. Note: when doing solvent relaxations with the CCS/CIS
model, no request of relaxed ground–state properties are needed, since the relaxed ground
state is identical to the HF ground state.

COSMO-ADC(2) within the post-SCF scheme: The new implementation of

COSMO-ADC(2) within the framework of post-SCF reaction–field scheme enables the cal-
culations of vertical excitations energies,transition moments (TM),excited state first-order
properties, and analytic gradients. [259] Unlike the PTED scheme, the COMSO-ADC(2)
method with post-SCF does not include the macro iterations to calculate vertical excitation
energies, meaning that after solving the ground–state reaction field self-consistently and de-
termining the polarized molecular orbitals at the COSMO-HF level the program computes
the correlation contribution to the reaction field without coupling back to the ground–state
HF.

COSMO-CC2 within the post-SCF scheme: The implementation of COSMO-

CC2 within the post-SCF reaction field enables the calculations of vertical excitation ener-
gies, transition moments and Faraday B terms for the simulation of UV-Vis and magnetic
circular dichroism (MCD) absorption spectrum. [260]
A typical input for COSMO-CC(2) and COSMO-ADC(2) any excited-state calculations with
the post-SCF scheme include following data groups in addition to the typical COSMO data
groups:

$reaction_field
post-SCF
ccs-like
$cosmo
epsilon= 50.000
324 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

rsolv= 1.30
refind= 3.0000
$cosmo_atoms
...

The input for $ricc2, $excitations, $laplace, and $response data groups is the same
as without COSMO.
The Following table shows the avaliability of COSMO-ADC(2) method with the PTED (old
implementation) and post-SCF (new implementation), and COSMO-CC2 with the post-SCF
scheme for different calculations.

COSMO-ADC(2) COSMO-ADC(2) COSMO-CC2

PTED post-SCF post-SCF
Excitation Energy X X X
Transition Moments X X X
Excited-state Prop. X X -
Excited-state Gradient - X -
Magnetic Circular Dichroism B - - X
Two-photon Absorption - - X

The combination of COSMO-CC2 and COSMO-ADC(2) with SCS and SOC are available
in the ricc2 module. Furthermore, the calculation of all the features available at COSMO-
CC2 and COSMO-ADC(2) with post-SCF scheme are possible with open-shell systems and
also with the SMP (OpenMP) and MPI parallelizations.
A short summary of the COSMO input is given at the beginning of the ricc2 output as well
as a summary of the energy contributions is given at the end of the ricc2 output.
18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 325

18.3 Frozen Density Embedding calculations

18.3.1 Background Theory

In the subsystem formulation of the density-functional theory a large system is decomposed

into several constituting fragments that are treated individually. This approach offers the
advantage of focusing the attention and computational cost on a limited portion of the
whole system while including all the remaining environmental effects through an effective
embedding potential. Here we refer in particular to the (fully-variational) Frozen Density
Embedding (FDE) [261] with the Kohn-Sham Constrained Electron Density (KSCED) equa-
tions [262, 263].
In the FDE/KSCED method the embedding potential required by an embedded subsystem
with density ρA to account for the presence of another (frozen) subsystem with density ρB
is:
nadd
δT nadd [ρA ; ρB ] δExc [ρA ; ρB ]
B
vemb (r) = vext (r) + vJ [ρB ](r) + s + (18.1)
δρA (r) δρA (r)
B
where vext (r) and vJ [ρB ](r) are the electrostatic potentials generated by the nuclei and
electron density of the subsystem B, respectively, and

Tsnadd [ρA ; ρB ] = Ts [ρA + ρB ] − Ts [ρA ] − Ts [ρB ] , (18.2)

nadd
Exc [ρA ; ρB ] = Exc [ρA + ρB ] − Exc [ρA ] − Exc [ρB ] (18.3)

are the non-additive non-interacting kinetic energy and exchange-correlation energy func-
tionals, respectively. In the expressions above Ts [ρ] is the (unknown) non-interacting kinetic
energy density functional and Exc [ρ] is the exchange-correlation energy functional. Note
that, while the first two terms in Eq. (18.1) refer to classical electrostatics (and could be de-
scribed by e.g. external point-charges), the last two terms are related to quantum-mechanical
effects.
Using freeze-and-thaw [264] cycles, the role of the frozen and the embedded subsystem is
iteratively exchanged, till convergence. If expressions (18.2) and (18.3) are computed exactly,
then the density ρA + ρB will coincide with the exact density of the total system.
Because the FDE/KSCED was originally developed in the Kohn-Sham framework, using
standard GGA approximations for Exc [ρ], the non-additive exchange-correlation potential
nadd
(δExc /δρA (r)) can be computed exactly as a functional of the density, leaving the ex-
pression of the non-additive kinetic energy term as the only approximation (with respect
the corresponding GGA calculation of the total system), because the exact explicit density
dependence of Ts from the density is not known. Using GGA approximations for the kinetic
energy functional (Ts ≈ T̃sGGA ) we have:

Tsnadd [ρA ; ρB ] ≈ T̃sGGA [ρA + ρB ] − T̃sGGA [ρA ] − T̃sGGA [ρB ] (18.4)

and
δT nadd [ρA ; ρB ]
≈ ṽTGGA [ρA + ρB ](r) − ṽTGGA [ρA ](r). (18.5)
δρA (r)
326 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

where ṽTGGA (r) = δ T̃sGGA /δρ(r).

The FDE total energy of total system is:

E F DE [ρ̃A , ρ̃B ] = Ts [ρ̃A ] + Ts [ρ̃B ] + Tsnadd [ρ̃A ; ρ̃B ]

+ Vext [A + B] + J[ρ̃A + ρ̃B ] + Exc [ρ̃A + ρ̃B ] . (18.6)

Note that this energy differs from the KS total energy of the total system due to the approx-
imation in Eq. (18.4) as well as the approximated kinetic potential (see Eq. 18.5) which lead
to approximated embedded densities (ρ̃A ≈ ρA and ρ̃B ≈ ρB ). With the current state-of-
the-art GGA kinetic approximations, the error in the binding energy for weakly interacting
systems is close to chemical accuracy.
Using the Generalized Kohn-Sham (GKS) theory, also hybrid exchange-correlation function-
als can be used in embedding calculations. To obtain a practical computational method,
the obtained embedding potential must be approximated by a local expression as shown in
Ref. [265]. This corresponds to performing for each subsystem hybrid calculations including
the interaction with other subsystems through an embedding potential derived at a semilocal
level of theory. When orbital dependent exchange-correlation functionals (e.g. hybrid func-
tional and LHF) are considered within the FDE method, the embedding potential includes
a non-additive exchange-correlation term of the form
nadd
Exc [ρA ; ρB ] = Exc [ΦA+B [ρA + ρB ]] − Exc [ΦA [ρA ]] − Exc [ΦB [ρB ]] (18.7)

where ΦA+B [ρA + ρB ] denotes the Slater determinant which yields the total density ρA +
ρB . Since such a determinant is not easily available, the non-additive exchange-correlation
contribution cannot be determined directly and the non-additive exchange-correlation term
can be approximated as [266]
nadd GGA GGA GGA
Exc [ρA ; ρB ] ≈ Exc [ρA + ρB ] − Exc [ρA ] − Exc [ρB ] . (18.8)

18.3.2 Frozen Density Embedding calculations using the FDE script

The shell script FDE controls and executes automatically FDE calculations. The script FDE
prepares the input files (running define), runs the calculations (only dscf is supported in
the present versiob), and combines the results (running fdetools). Because the FDE equa-
tions are coupled sets of one-electron equations (one for each subsystem), full relaxation
of the electron densities of both subsystems is obtained by using a freeze-and-thaw [264]
procedure until convergence.
The converged FDE calculations are store in the subdirectories STEPN/SUBSYSTEM_A and
STEPN/SUBSYSTEM_B, where N is the number of the FDE iteration. The subdirectory
ISOLATED_SUBSYSTEM_A and ISOLATED_SUBSYSTEM_B contain instead the calculations for
isolated subsystems (see also Section 18.3.3).
Current functionalities and limitations of FDE are:

• only C1 point group;

18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 327

• only for closed-shell systems that consist of two closed-shell subsystems (e.g. weakly
interacting closed-shell dimers);

• only total and binding energy calculations (no gradients);

• serial and OMP dscf runs (no MPI);

• monomolecular and supermolecular basis set approach;

• LDA/GGA kinetic energy functionals (for weakly interacting systems);

• full or pure electrostatic embedding;

• LDA/GGA, hybrid or orbital-dependent exchange-correlation potentials;

• multilevel FDE calculation;

• energy-decomposition;

• FDE calculation with subsystem B taken frozen.

In order to perform a FDE calculation, the files coord and control for the total system
are necessary to take informationson atomic coordinates and basis sets. The input file for
the total system can be generated, as usual, with define but no calculation on the total
system is required. $denconv 1.d-7 option should be defined in file control in order to
better converge the embedded densities and better describe the dipole moment.
Given a closed-shell supramolecular system with a GGA/LDA exchange-correlation func-
tional, the command

FDE -p 3

invokes an iterative resolution of the KSCED equations with revAPBEk [267, 268] as ap-
proximation of the non-additive kinetic potential (see Eq. 18.5) in the monomolecular basis
set approach. The two subsystems are defined via an integer m = 3 in the example above
which identifies the first atom of the subsystem B in the file coord of the supramolecular
system with n atoms, where the atoms 1 . . . m − 1 belong to the subsystem A while the
atoms m . . . n to the B one. Thus the file coord must contains first all the atoms of the
system A and then all the atoms of the system B.
As an example we report here the FDE -p 3 output for the HF dimer:

FDE Version 1.02

Frozen Density Embedding Main Driver

Scf-like procedure for

closed-shell interacting systems (dimers)

program development: Savio Laricchia

Eduardo Fabiano
328 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

Fabio Della Sala

S. Laricchia, E. Fabiano, L. A. Constantin, F. Della Sala,

J. Chem. Theory Comp. (2011)
S. Laricchia, E. Fabiano, F. Della Sala,
J. Chem. Phys. 133, 164111 (2010)
L. A. Constantin, E. Fabiano, S. Laricchia, F. Della Sala,
Phys. Rev. Lett. 106, 186406 (2011)
S. Laricchia, E. Fabiano, F. Della Sala,
Chem. Phys. Lett. 518, 114 (2011)

Sun Mar 25 23:00:01 CEST 2012

Monomolecular basis set approach...

Serial calculation will be performed...

running /home/fabiods/REDO/branch64/TURBOMOLE/bin/em64t-unknown-linux-gnu/dscf

b-lyp exchange-correlation potential in KS supermolecular calculation...

revapbek kinetic energy approximation will be used...
Default convergence criterion on the system dipole: 0.005
Default value of starting damping parameter is 0.45
Default value of step damping parameter is 0.10
Default value of maximum damping parameter is 0.90
Default value of maximum fde iterations is 20
Saving options in fde.input

+-----------------------------------------------------------+
| Subsystem A atomic coordinates and basis set information |
| x y z atom basis set ecp |
+-----------------------------------------------------------+

2.5015 -0.1705 -0.0000 f def2-TZVP none

3.2889 1.3859 0.0000 h def2-TZVP none

+-----------------------------------------------------------+
| Subsystem B atomic coordinates and basis set information |
| x y z atom basis set ecp |
+-----------------------------------------------------------+

-2.7537 0.0364 -0.0000 f def2-TZVP none

-1.0191 -0.1789 0.0003 h def2-TZVP none
18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 329

Running Isolated subsystems:

************************
* ISOLATED SUBSYSTEM A *
************************
Done!

************************
* ISOLATED SUBSYSTEM B *
************************
Done!

Saved isolated subsystems data in:

isolated_energy.ks
mos_A.ks
mos_B.ks

*********************
* FDE - step 1 *
*********************

FDE ENERGY (TOTAL SYSTEM): -200.96417090754 Ha

FDE BINDING ENERGY: 5.865327 mHa
3.680548 kcal/mol
Dipole convergence: 0.138071, Damping: 0.45

*********************
* FDE - step 2 *
*********************

FDE ENERGY (TOTAL SYSTEM): -200.96418098234 Ha

FDE BINDING ENERGY: 5.875401 mHa
3.686870 kcal/mol
Dipole convergence: 0.009246, Damping: 0.35

*********************
* FDE - step 3 *
*********************

FDE ENERGY (TOTAL SYSTEM): -200.96418289036 Ha

FDE BINDING ENERGY: 5.877309 mHa
3.688067 kcal/mol
Dipole convergence: 0.004395, Damping: 0.25
330 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

See embedded susbsystems calculations in:

STEP3/SUBSYSTEM_A
STEP3/SUBSYSTEM_B
See total system in:
STEP3/ENERGY_SYSTEM

Sun Mar 25 23:00:21 CEST 2012

Total time: 20 secs.

The final energies are stored in the file fde_energy. The directory STEPN/ENERGY_SYSTEM
contains the total system with density ρA + ρB ; this directory can (only) be used for density
analysis.

18.3.3 Options

All the options for the FDE can be specified as command lines, and are described below.
The options can be also be specified in file fde.input, which is read by the FDE script. If
fde.input is not present it is created by the FDE script. Command lines options overwrites
options found in the fde.input file.

Subsystem definition

The flag -p integer is required or it must be present in the fde.input file.

Equivalent command: --pos-cut integer
fde.input option: pos-cut= integer

Kinetic-energy functionals

In order to use different GGA approximations of the non-additive kinetic potential, the flag
-k string must be used. Here string is the acronym used to identify a given GGA kinetic
energy approximation, that can be selected among the following functionals:

• string=revapbek: generalized gradient approximation with a PBE-like enhancement

factor, obtained using the asymptotic expansions of the semiclassical neutral atom as
reference [267, 268] (revAPBEk). This is the default choice;

• string=lc94: Perdew-Wang (PW91) exchange functional reparametrized for kinetic

energy by Lembarki and Chermette [269] (LC94);

• string=t-f: gradient expansion truncated at the zeroth order (GEA0), corresponding

to the Thomas-Fermi functional.

For example, the command

18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 331

FDE -p 3 -k lc94

approximates the non-additive kinetic contribution to the embedding potential through the
functional derivative of LC94 kinetic energy functional.
A pure electrostatic embedding can be also performed with FDE script, where the embedding
potential required by a subsystem A to account for the presence of the B one will be merely:
B
vemb (r) = vext (r) + vJ [ρB ](r) (18.9)
B
with vext (r) and vJ [ρB ](r) the electrostatic potentials generated respectively by the nuclei
and electron density of the subsystem B. To perform an electrostatic embedding calculation
use

FDE -p 3 -k electro

and can be performed for both Kohn-Sham (only for LDA/GGA exchange-correlation func-
tionals) and Hartree-Fock methods.
The electrostatic embedding is implemented only for testing purpose. It resembles an elec-
trostatic embedding with external point-charges and/or point-dipoles, but it is “exact” as it
is based on the whole densities (i.e. it considers all multipole moments of the density and
the polarizabilies at all orders).
Equivalent command: --kin string
fde.input option: kin= string

FDE charged subsystems

FDE can perform calculations for charged closed-shell systems whose charge is localized on
one or both subsystems. To localize the charge on a given subsystem, --chargeA= integer
must be used for the subsystem A and --chargeB= integer for the B one. Here integer
denotes the charge added to the neutral subsystem. For example, the command

FDE -p 3 --chargeA= 2

performs a FDE calculation for a negative charged closed shell system (for example Zn(
H2 O)2+
2 ) whose subsystem B has charge 2. Note that in this case the starting control file
must have a charge +2.
fde.input option: chargeA= integer , chargeB= integer

FDE with subsystem B taken frozen

FDE can perform embedding calculations where the subsystem B is taken frozen, i.e. without
scf calculation on it using an embedding potential. Therefore only one step will be performed
if the flag --frozen will be used
332 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

FDE -p 3 --frozen

The frozen embedding calculation is store in the subdirectory STEP1/SUBSYSTEM_A. The

control file is modified with the following keywords:

$fde read
$fde-input zj file=fde_ZJ.mat
$fde-input kxc file=fde_KXC.mat

The program dscf will read the submatrices fde_ZJ.mat and fde_KXC.mat and add them
to the Hamiltonian.
fde.input option: frozen=1

Parallel calculations

If PARA_ARCH=SMP and OMP calculation will be performed. The flag -nth nthreads can be
used to specify the number of threads. For example, with the following command

FDE -p 3 -nth 4

will use 4 threads.

Equivalent command: --nthreads integer
fde.input option: nthreads= integer

Monomolecular and supermolecular basis set approach

The ρA and ρB densities can be expanded using the supermolecular or monomolecular basis
set. In a supermolecular basis set expansion the basis functions {χ} of both subsystems
are employed to expand the subsystem electron densities. In a monomolecular basis set
expansion, instead, only basis functions {χ` } centered on the atoms in the `-th subsystem
are used to expand the corresponding density.
Both monomolecular and supermolecular basis set expansion of the electron densities are
implemented in FDE: with the flag -m a monomolecular expansion is performed, while for a
supermolecular one -s is used. In the absence of both flags a monomolecular expansion is
performed by default.
For an accurate calculation of binding-energies of weakly interacting molecular systems a
supermolecular basis set is required (to avoid the basis-set superposition error). Otherwise
a very large monomolecular basis set is necessary.
NOTE: The FDE script supports only basis-set in the TURBOMOLE library.
Equivalent command: --mono or --super
fde.input option: method=mono or method=super
18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 333

Convergence of the freeze-and-thaw cycles

The script FDE runs a self-consistent calculation when a convergence criterion is fulfilled. The
convergence criterion is the change in the total dipole moment. This is a tight convergence
criterion, as the dipole moment is highly sensitive to small changes in electron density. The
convergence parameter εj for the j-th step in the freeze-and-thaw procedure is computed by
means the following expression

|∆µjA | + |∆µjB |
εj = (18.10)
2
where

|∆µji | = |µji | − |µij−1 | i = A, B

is the difference between the dipole moments of two consecutive steps for the i-th subsystem.
Eq. (18.10) allows to consider changes in both subsystems or one of them because of the
relaxation of their electron densities. By default, FDE stops when εj ≤ 0.005 a.u.. The
default value for the convergence criteria can be changed using the flag --epsilon= real
where real is a decimal number.
The maximum number of freeze-and-thaw cycles can be specified by --max-iter= integer,
and the default value is 20.
In order to make easy the convergence of the iterative solution of the KSCED coupled
equations, a damping factor η must be used for the matrix elements of the embedding


potential vemb ij as perturbation to a given subsystem

j
j
j−1
d vemb ik
= (1 − η) vemb ik
+ η vemb ik
(18.11)

j
for the j-th iteration. Here d vemb ik is the matrix element effectively used in the j-th itera-
tion after the damping. In FDE the starting value of η can be changed using --start-damp=
real (default value is 0.45) where real is a decimal number. The damping parameter can
also dynamically change at each iterative step (according to the convergence process) of
a quantity set by --step-damp= real (default value is 0.10). The minimum value set by
--max-damp= real (default value is 0.90).
fde.input options:
epsilon= real
max-iter= integer
start-damp= real
max-damp= real
step-damp= real

Embedding energy error

The embedding error in the total energy is computed as

∆E = E FDE [ρ̃A ; ρ̃B ] − E DFT [ρ] (18.12)

334 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

where E DFT is the DFT total energy of total system with density ρ(r). In order to compute
∆E as well as its components, the flag --err-energy must be used. This flag will required
also the DFT calculation on the total system. In this case the converged SCF output file
must be named output.dscf.
An example of session output for the computation of embedding energy and energy error
decomposition, when --err-energy flag is present, is the following:

FDE ENERGY (TOTAL SYSTEM): -200.99720391651 Ha

FDE BINDING ENERGY: 4.960885 mHa
3.113002 kcal/mol
FDE ENERGY ERROR: 2.003352 mHa
ERROR ENERGY DECOMPOSITION
coulomb contribution: -0.693026 mHa
nuclear contribution: -3.136544 mHa
exchange-correlation contribution: -1.156390 mHa
kinetic contribution: 6.989320 mHa

where the FDE energy (E FDE ), the FDE binding energy, the embedding energy error (∆E)
and the error energy decomposition in its coulomb, nuclear, exchange-correlation and kinetic
contributions are reported. This output is present at each FDE iteration.
fde.input option: err-energy=1

Restarting

The script FDE checks in the current directory for previous FDE calculations. If these are
present, then the FDE calculation will be restarted from the last iteration found. The
directories ISOLATED_SUBSYSTEM_A and ISOLATED_SUBSYSTEM_B will be overwritten by the
converged calculations from previous run. The energy and the orbital from the isolated
systems are saved in the current directory in the files: isolated_energy.ks, mos_A.ks and
mos_B.ks.
Note that a restart is possible only if the same subsystem definition and the same basis
set are used (e.g. the same -p flag and the -s or -m flag). Other flags, e.g. kinetic and
xc-functionals and convergence parameters, can be instead modified.
As all the options are saved in the fde.input file, to restart a FDE calculation the FDE
script can be invoked without any parameters.
To force a calculation from scratch use:

FDE -p 3 --scratch
18.3. FROZEN DENSITY EMBEDDING CALCULATIONS 335

Table 18.1: Other options in the shell script FDE

-d or --dipole dipole=1 dipole moment each step

-v or --verbose shows more informations
--save-mos save-mos=1 save the MOs of both subsystems
for each step
--save-matrix save-matrix=1 save fde matrices
in each step
--help list all commands

18.3.4 FDE with hybrid and orbital-dependent functionals

In order to use local approximations (18.1) and (18.8) with FDE, the flag -f string must
be add to the options of the script. Here string denotes the local/semilocal approximation
to hybrid or orbital-dependent exchange-correlation potentials in vemb (r). All LDA/GGA
functionals in TURBOMOLE can be considered as approximations.
For example, the command

FDE -p 3 -f b-lyp

can be used to approximate bh-lyp or b3-lyp hybrid non-additive potentials, while the
command

FDE -p 3 -f pbe

approximates the pbe0 hybrid non-additive potentials. Other combinations of functionals

are not recommended (meta-GGA are not supported).
Finally, also calculations with the Local Hartree-Fock (LHF) potential can be performed.
In this case the command

FDE -p 3 -f becke-exchange

can be used to approximate the LHF non-additive potential [266].

Equivalent command: --func string
fde.input option: func= string
336 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

18.4 Periodic Electrostatic Embedded Cluster Method

18.4.1 General Information

The Periodic Electrostatic Embedded Cluster Method (PEECM) functionality [270] provides
electronic embedding of a finite, quantum mechanical cluster in a periodic, infinite array of
point charges. It is implemented within HF and DFT energy and gradient TURBOMOLE
modules: dscf, grad, ridft, rdgrad, and escf. Unlike embedding within a finite set of
point charges the PEEC method always yields the correct electrostatic (Madelung) potential
independent of the electrostatic moments of the point charges field. It is also significantly
faster than the traditional finite point charges embedding.

18.4.2 Theoretical Background

Generally, the PEEC method divides the entire, periodic and infinite system into two parts,
the inner (I) part, or so called cluster, and the outer (O) part which describes its environ-
ment. Thus, unlike "true" periodic quantum mechanical methods, PEECM primarily aims
at calculations of structure and properties of localized defects in dominantly ionic crystals.
The innermost part of the cluster is treated quantum mechanically (QM), whereas in the
remaining cluster part cations are replaced by effective core potentials (ECPs) and anions
by ECPs or by simply point charges. Such an "isolating" outer ECP shell surrounding the
actual QM part is necessary in order to prevent artificial polarization of the electron density
by cations which would otherwise be in a direct contact with the QM boundary. The outer
part or the environment of the cluster is described by a periodic array of point charges,
representing cationic and anionic sites of a perfect ionic crystal.
The electronic Coulomb energy term arising from the periodic field of point charges sur-
rounding the cluster has the following form
∈UC X
X NX ∞ Z
µ(~r)ν(~r)
J= Dµν qk d~r,
|~r − R ~ k − L|
~
µν k ~
L∈O

where UC denotes the unit cell of point charges, Dµν are elements of the density matrix, µ, ν
~ k denote charges and positions of point charges, and L
are basis functions, qk , R ~ denote direct
lattice vectors of the outer part O. It is evaluated using the periodic fast multipole method
(PFMM) [271] which, unlike the Ewald method [272], defines the lattice sums entirely in
the direct space. In general, PFMM yields a different electrostatic potential than the Ewald
method, but the difference is merely a constant shift which depends on the shape of external
infinite surface of the solid (i.e. on the way in which the lattice sum converges toward the
infinite limit). However, this constant does not influence relative energies which are the
same as obtained using the Ewald method, provided that the total charge of the cluster
remains constant. Additionally, since the electrostatic potential within a solid is not a well
defined quantity, both the absolute total energies and orbital energies have no meaning (i.e.
you cannot compare energies of neutral and charged clusters!).
18.4. PERIODIC ELECTROSTATIC EMBEDDED CLUSTER METHOD 337

18.4.3 Calculation Setup

There are three key steps in setting up a PEECM calculation. In the first step the periodic
field of point charges has to be defined by specifying the point charges unit cell. Next
step is the definition of the part infinite of point charges field that will be replaced by the
explicit quantum mechanical cluster. Finally, the quantum mechanical cluster together with
surrounding ECPs representing cationic sites as well as point charges representing anions is
defined and put in place of the point charges. The input preparation steps can be summarized
as follows

1. Dimensionality of the system is specified by the keyword periodic in the $embed

section: periodic 3 means a bulk three-dimensional system, periodic 2 denotes a
two-dimensional surface with an aperiodic z direction.
2. Definition of the unit cell of periodic point charges field is specified in the subsections
cell and content of the $embed section.
3. Definition of the values of the point charges by specifying a charge value per species,
using the subsection charges, or a charge value for each point charge, using the
subsection ch_list. Note that only one of the subsections can be defined.
4. Definition of the part of point charges field that will be replaced by the QM cluster
together with the isolating shell (ECPs, explicit point charges) is specified in the
subsection cluster of the $embed section.
5. Definition of the quantum mechanical cluster as well as the surrounding ECPs and
anionic point charges is included in the usual $coord section.

The following two examples show the definition of the point charges unit cells.
Example 1. Ca4 F19 cluster embedded in bulk CaF2
In this example a QM cluster with the composition Ca4 F19 , surrounded by 212 ECPs and 370
explicit point charges, representing Ca2+ cations and F− anions is embedded in a periodic
field of point charges (+2 for Ca and -1 for F) corresponding to the CaF2 fluorite lattice.
First, the program has to know that this is a three-dimensional periodic system. This
is specified by the keyword periodic 3, meaning periodicity in three dimensions. The
dimensions of the unit cell for bulk CaF2 are given in the subsection cell of the $embed
keyword. By default, the unit cell dimensions are specified in atomic units and can be
changed to Å using cell ang. The positions of the point charges in the unit cell are
specified in the subsection content. In this example positions are given in fractional crystal
coordinates (content frac). You can change this by specifying content for Cartesian
coordinates in atomic units or content ang for Cartesian coordinates in Å. The values of
point charges for Ca and F are given in the subsection charges.

$embed
periodic 3
cell
338 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

10.47977 10.47977 10.47977 90.0 90.0 90.0

content frac
F 0.00 0.00 0.00
Ca -0.25 -0.75 -0.75
F 0.50 -0.50 0.00
F 0.50 0.00 -0.50
F 0.00 -0.50 -0.50
F 0.50 -0.50 -0.50
F 0.00 0.00 -0.50
F 0.50 0.00 0.00
F 0.00 -0.50 0.00
Ca -0.25 -0.25 -0.25
Ca 0.25 -0.75 -0.25
Ca 0.25 -0.25 -0.75
end
...
charges
F -1.0
Ca 2.0
end

The above input defines a periodic, perfect, and infinite three-dimensional lattice of point
charges corresponding to the bulk CaF2 structure. In order to use this lattice for PEECM
calculation we have to make “space” for our QM cluster and the isolating shell. This is done
by specifying the part of the lattice that is virtually removed from the perfect periodic array
of point charges to make space for the cluster. The positions of the removed point charges
are specified in the subsection cluster of the $embed keyword. Note, that the position
of the QM cluster and the isolating shell must exactly correspond to the removed part of
the crystal, otherwise positions of the cluster atoms would overlap with positions of point
charges in the periodic lattice, resulting in a “nuclear fusion”.

cluster
F 0.00000000000000 0.00000000000000 0.00000000000000
Ca -2.61994465796043 -2.61994465796043 -2.61994465796043
Ca 2.61994465796043 -2.61994465796043 2.61994465796043
Ca 2.61994465796043 2.61994465796043 -2.61994465796043
Ca -2.61994465796043 2.61994465796043 2.61994465796043
F -5.23988931592086 0.00000000000000 0.00000000000000
F 0.00000000000000 0.00000000000000 -5.23988931592086
F 5.23988931592086 0.00000000000000 0.00000000000000
F 0.00000000000000 -5.23988931592086 0.00000000000000
F 0.00000000000000 0.00000000000000 5.23988931592086
F 0.00000000000000 5.23988931592086 0.00000000000000
18.4. PERIODIC ELECTROSTATIC EMBEDDED CLUSTER METHOD 339

F -5.23988931592086 -5.23988931592086 0.00000000000000

F -5.23988931592086 0.00000000000000 -5.23988931592086
F -5.23988931592086 0.00000000000000 5.23988931592086
F -5.23988931592086 5.23988931592086 0.00000000000000
F 5.23988931592086 -5.23988931592086 0.00000000000000
...

repeated for Ca216 F389

end
By default, the positions of point charges are specified in atomic units as Cartesian coordi-
nates. You can change this by specifying cluster frac for fractional crystal coordinates or
cluster ang for Cartesian coordinates in Å.
Finally, you have to specify the coordinates of the QM cluster along with the surrounding
ECPs representing cationic sites and explicit point charges representing anions. This is done
in the usual way using the $coord keyword.

$coord
0.00000000000000 0.00000000000000 0.00000000000000 f
-2.86167504097169 -2.86167504097169 -2.86167504097169 ca
2.86167504097169 2.86167504097169 -2.86167504097169 ca
-2.86167504097169 2.86167504097169 2.86167504097169 ca
2.86167504097169 -2.86167504097169 2.86167504097169 ca
0.00000000000000 -5.24009410923923 0.00000000000000 f
-5.24009410923923 0.00000000000000 0.00000000000000 f
0.00000000000000 5.24009410923923 0.00000000000000 f
0.00000000000000 0.00000000000000 -5.24009410923923 f
5.24009410923923 0.00000000000000 0.00000000000000 f
0.00000000000000 0.00000000000000 5.24009410923923 f
0.00000000000000 -5.24009410923923 -5.24009410923923 f
-5.24009410923923 -5.24009410923923 0.00000000000000 f
5.24009410923923 -5.24009410923923 0.00000000000000 f
0.00000000000000 -5.24009410923923 5.24009410923923 f
0.00000000000000 5.24009410923923 -5.24009410923923 f
...

repeated for Ca216 F389

$end
This is the standard TURBOMOLE syntax for atomic coordinates. The actual distinction be-
tween QM cluster, ECP shell, and explicit point charges is made in the $atoms section.

$atoms
f 1,6-23 \
340 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

basis =f def-TZVP
ca 2-5 \
basis =ca def-TZVP
ca 24-235 \
basis =none \
ecp =ca ecp-18 hay & wadt
f 236-605 \
basis =none \
charge= -1.00000000

In the example above the F atoms 1 and 6-23 as well Ca atoms 2-5 are defined as QM atoms
with def-TZVP basis sets. The Ca atoms 24-235 are pure ECPs and have no basis functions
(basis =none) and F atoms 236-605 are explicit point charges with charge -1, with no basis
functions and no ECP.
This step ends the input definition for the PEECM calculation.
Example 2. Al8 O12 cluster embedded in α-Al2 O3 (0001) surface
In this example a QM cluster with the composition Al8 O12 , surrounded by 9 ECPs repre-
senting Al3+ cations is embedded in a two-dimensional periodic field of point charges (+3
for Al and -2 for O) corresponding to the (0001) surface of α-Al2 O3 .
As in the first example, the program has to know that this is a two-dimensional periodic
system and this is specified by the keyword periodic 2. The dimensions of the unit cell for
the (0001) α-Al2 O3 surface are given in the subsection cell of the $embed keyword. The
aperiodic direction is always the z direction, but you have to specify the unit cell as if it was
a 3D periodic system. This means that the third dimension of the unit cell must be large
enough to enclose the entire surface in this direction. The unit cell dimensions are specified
in Å using cell ang. The positions of the point charges in the unit cell are specified as
Cartesian coordinates in Å (content ang). The values of point charges for Al and O are
given in the subsection charges.

$embed
periodic 2
cell angs
4.8043 4.8043 24.0000 90.0000 90.0000 120.0000
content ang
Al 2.402142286 1.386878848 5.918076515
Al -0.000013520 -0.000003382 7.611351967
Al -0.000008912 2.773757219 8.064809799
Al 2.402041674 1.386946321 0.061230399
Al -0.000005568 -0.000003223 10.247499466
Al 2.402137518 1.386872172 9.977437973
Al 0.000000070 2.773757935 5.390023232
Al 0.000006283 -0.000005607 3.696748018
18.4. PERIODIC ELECTROSTATIC EMBEDDED CLUSTER METHOD 341

Al 2.402151346 1.386879444 3.243290186

Al 0.000100868 2.773690462 11.246870041
Al -0.000001982 -0.000005796 1.060600400
Al 0.000004853 2.773764610 1.330662251
O -0.731205344 1.496630311 6.749288559
O 0.743527174 1.296469569 8.957922935
O 1.588027477 0.104536049 11.127140045
O 1.471626759 2.779079437 6.749288559
O 3.309734344 -0.004341011 8.957920074
O 3.919768333 1.323050499 11.127141953
O -0.740424335 4.045563698 6.749289513
O -1.651123047 2.868478537 8.957910538
O 1.698525310 2.733071804 11.127161026
O 3.133347750 2.664006472 4.558811665
O 1.658615232 2.864167213 2.350177050
O 0.814115047 4.056100845 0.180959582
O 0.930515707 1.381557465 4.558811188
O 1.494558096 0.004332162 2.350180149
O -1.517625928 2.837586403 0.180958077
O 3.142566681 0.115072958 4.558810234
O -0.751034439 1.292158127 2.350189686
O 0.703617156 1.427564979 0.180938885
end
...
charges
O -2.0
Al 3.0
end

The above input defines a periodic, perfect, and infinite two-dimensional lattice of point
charges corresponding to the (0001) α-Al2 O3 surface. In order to use the lattice for PEECM
calculation we have to make “space” for our QM cluster and the surrounding ECP shell. This
is done by specifying the part of the lattice that is virtually removed from the perfect periodic
array of point charges to make space for the cluster. The positions of the removed point
charges are specified in the subsection cluster of the $embed keyword. Note, that the
position of the QM cluster must exactly correspond to the removed part of the crystal,
otherwise positions of the cluster atoms would overlap with positions of point charges in the
periodic lattice, resulting in a “nuclear fusion”.

cluster ang
Al -0.000012482 5.547518253 9.977437973
Al 2.402141094 6.934402943 8.064809799
Al 2.402144432 4.160642624 10.247499466
342 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

Al 4.804287434 5.547518253 9.977437973

Al 2.402250767 6.934336185 11.246870041
Al -0.000005568 8.321288109 10.247499466
Al 2.402137518 9.708164215 9.977437973
Al 4.804294586 8.321288109 10.247499466
O 0.907584429 4.156304836 8.957920074
O 1.517618299 5.483696461 11.127141953
O -0.703624666 6.893717766 11.127161026
O 3.145677090 5.457115650 8.957922935
O 3.990177393 4.265182018 11.127140045
O 0.751026928 7.029124260 8.957910538
O 4.100675106 6.893717766 11.127161026
O 0.743527174 9.617761612 8.957922935
O 1.588027477 8.425827980 11.127140045
O 3.309734344 8.316950798 8.957920074
O 3.919768333 9.644342422 11.127141953
O 5.555326939 7.029124260 8.957910538
Al 4.804400921 11.094982147 11.246870041
Al -0.000008912 2.773757219 8.064809799
Al -2.402049065 6.934336185 11.246870041
Al 4.804400921 2.773690462 11.246870041
Al 2.402136564 4.160642624 7.611351967
Al -0.000013520 8.321288109 7.611351967
Al -0.000008912 11.095048904 8.064809799
Al 7.206440926 6.934402943 8.064809799
Al 4.804286480 8.321288109 7.611351967
end

The positions of point charges are specified in Å as Cartesian coordinates.

Finally, you have to specify the coordinates of the QM cluster along with the surrounding
ECPs. This is done in the usual way using the $coord keyword.

$coord
-0.00002358760000 10.48329315900000 18.85463057110000 al
4.53939007480000 13.10412613690000 15.24028611330000 al
4.53939638280000 7.86247730390000 19.36497297520000 al
9.07879006320000 10.48329315900000 18.85463057110000 al
4.53959732680000 13.10399998250000 21.25351019750000 al
-0.00001052200000 15.72496001430000 19.36497297520000 al
4.53938331720000 18.34577677080000 18.85463057110000 al
9.07880357850000 15.72496001430000 19.36497297520000 al
1.71508649490000 7.85428007030000 16.92802041340000 o
2.86788376470000 10.36268741690000 21.02725683720000 o
18.4. PERIODIC ELECTROSTATIC EMBEDDED CLUSTER METHOD 343

-1.32965829240000 13.02724227310000 21.02729288000000 o

5.94446987180000 10.31245694970000 16.92802581990000 o
7.54034461170000 8.06002818410000 21.02725323160000 o
1.41923561090000 13.28312353520000 16.92800239300000 o
7.74915508620000 13.02724227310000 21.02729288000000 o
1.40506312580000 18.17494056150000 16.92802581990000 o
3.00093786570000 15.92251179600000 21.02725323160000 o
6.25449323900000 15.71676368210000 16.92802041340000 o
7.40729073370000 18.22517102690000 21.02725683720000 o
10.49804944110000 13.28312353520000 16.92800239300000 o
9.07900452260000 20.96648359440000 21.25351019750000 al
-0.00001684120000 5.24164297480000 15.24028611330000 al
-4.53921616520000 13.10399998250000 21.25351019750000 al
9.07900452260000 5.24151682240000 21.25351019750000 al
4.53938151440000 7.86247730390000 14.38337475740000 al
-0.00002554910000 15.72496001430000 14.38337475740000 al
-0.00001684120000 20.96660974680000 15.24028611330000 al
13.61820356690000 13.10412613690000 15.24028611330000 al
9.07878826040000 15.72496001430000 14.38337475740000 al
$end

This is the standard TURBOMOLE syntax for atomic coordinates. The actual distinction be-
tween QM cluster and ECP shell is made in the $atoms section.

$atoms
al 1-8 \
basis =al def-SV(P)
o 9-20 \
basis =o def-SV(P)
al 21-29 \
basis =none \
ecp =al ecp-10 hay & wadt

In the example above the Al atoms 1-8 and O atoms 9-20 are defined as QM atoms with
def-SV(P) basis sets. The Al atoms 21-29 are pure ECPs and have no basis functions
(basis =none).
This step ends the input definition for the PEECM calculation.
344 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

18.5 Polarizable embedding calculations

Realizable embedding (PE) calculations are a based on a hybrid model of quantum me-
chanics and molecular mechanics (QM/MM) in which the classical region is represented by
an electrostatic potential with up to octupole moments and induced point dipole moments.
The main improvement over the more common QM/MM approaches without polarizable
MM sites can be found for the description of electronic excitations but also for any other
process which causes a significant change in the QM density and which is accompanied by
a fast response of the environment.
In TURBOMOLE, ground state energies computed with the dscf, ridft, and ricc2 module
and electronic excitation properties based on RI-CC2 and RI-ADC(2) are implemented.
The excited state analytic gradients are also available at the RI-ADC(2) level. The general
theory is presented in ref. [273] and [274,275], the PERI-CC2 model and the TURBOMOLE
implementation is described in ref. [254].

18.5.1 Theory

In the following, only the most important ideas are presented and discussed with a focus on
the PERI-CC2 model. The essential concept is the introduction of an environment coupling
operator Ĝ(DCC )

Ĝ(DCC ) = Ĝes + Ĝpol (DCC ) (18.13)

with the electrostatic contribution

M X
X K X
Ĝes = Θ(k) (k)
m,pq Qm Êpq (18.14)
m=1 k=0 pq

and the polarization contribution

U X
X
Ĝpol (DCC ) = Θ(1) ind
u,pq µu (D
CC
)Êpq . (18.15)
u=1 pq

(k)
Here, Θm,pq are multipole interaction integrals of order k and µind
u are the induced dipoles
which can be obtained from the electric field Fu and the polarizability αu at a site u:

µind
u = Fu α u (18.16)

Because the induced dipoles depend on the electron density and vice versa, their compu-
tations enter the self-consistent part of the HF cycle. Introducing Ĝ(DCC ) into standard
18.5. POLARIZABLE EMBEDDING CALCULATIONS 345

equations for the HF reference state and the CC2 equations leads to a general PE-CC2 for-
mulation. To maintain efficiency, a further approximation has been introduced which makes
the operator only dependent on a CCS-like density term. These general ideas define the
PERI-CC2 model and allow to formulate the corresponding Lagrangian expression

1
LPERI-CC2 (t, t̄) = EPE-HF + hHF|Ŵ (T̂1 + T̂2 + T̂12 )|HFi +
X 2
t̄µ1 hµ1 |W̃ + [F̂ PE , T̂1 ] + [W̃ , T̂2 ]|HFi +
µ1
X
t̄µ2 hµ2 |W̃ + [F̂ PE , T̂2 ]|HFi
µ2
1 X elec ∆0
− F (D )Ruv Fvelec (D∆0 ) (18.17)
2 uv u

from which all PERI-CC2 equations including the linear response terms may be derived.
Note that the dependency on the density couples the CC amplitude and multiplier equa-
tions for the ground state solution vector.

This coupling is avoided by the simplified polarizable embedding method (sPE) described
in ref. [276].

LsPERI-CC2 (t, t̄) =E PE-HF

(18.18)
+ hΛ|ĝN + F̂NPE + Ĝpol
N (D
∆CC
)|CCi .

The subscript N indicates that the operator is normal ordered with respect to the Hartree–
Fock state. Here, a polarization operator Ĝpol (D) was introduced,
M
pol ∆CC 1 X  > ∆CC
Ĝ (D )=− F̂u Ruv Fel
v (D ). (18.19)
2 uv

which depends on the elements of the difference density matrix D∆CC . These are defined as
∆CC
Dpq = hHF|â†p âq |CCi − Dpq
HF
, (18.20)

hence, they do not depend on the Lagrangian multipliers.

18.5.2 Computational details: SCF calculations

To carry out a PE-SCF calculation with the DSCF or RIDFT module, you have to specify
the following in the control file:

$point_charges pe [options]
<length unit>
<no. MM sites> <order k> <order pol> <length exclude list>
<list of MM sites: exclude list, xyz coords, multipole mom., pol. tensor>
346 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

length unit specifies the unit for the MM site coordinates (use AA or AU)

no. MM sites the amount of MM sites (length of the list)

order k the order of multipoles used (0: point charges, 1: dipole moments, 2: quadrupole
moments, 3: octupole moments)

order pol the treatment of polarizabilities (0: none, 1: isotropic, 2: anisotropic)

length exclude list number of elements in the exclude list

list of MM sites each MM sites is described on one line, entries separated by blanks; first
entry is the exclude list of with as much elements as defined in the head line (If the first
element in the exclusion list of one site occurs in the exclude list of another site, they
do not contribute to each others polarization); next follows the MM site coordinates
in (x,y,z positions), the point charge, the dipole moment (for k ≥ 1, x,y,z component),
the quadrupole moment (for k ≥ 2, xx, xy, xz, yy, yz, zz component), the octupole
moment (for k = 3, xxx, xxy, xxz, xyy, xyz, xzz, yyy, yyz, yzz, zzz component), the
polarizability ( one component for pol-order 1, xx, xy, xz, yy, yz, zz component for
pol-order 2)

An example for a polarizable embedding with coordinates given in Å, point charges and
isotropic polarizabilities:

$point_charges pe
AA
6 0 1 1
39 -0.2765102481 2.5745845304 3.5776314866 0.038060 15.217717
39 1.3215071687 2.3519378014 2.8130403183 -0.009525 14.094642
39 -0.5595582934 1.2645007691 4.7571719292 -0.009509 14.096775
39 -1.5471918244 2.5316479230 2.3240961995 -0.009519 14.096312
39 -0.3207417883 4.1501938400 4.4162313889 -0.009507 14.096476
41 -1.1080691595 4.9228723099 -1.6753825535 0.038060 15.217717
41 -0.9775910525 6.5274614891 -2.4474576239 -0.009525 14.094642
41 -2.5360480539 4.8923046027 -0.6040781123 -0.009509 14.096775
41 0.3630448878 4.6028736791 -0.7155647205 -0.009519 14.096312
41 -1.2817317422 3.6689143712 -2.9344225518 -0.009507 14.096476

All values are given in atomic units (except coordinates if stated otherwise). These data are
mandatory. An alternative input format can also be used by specifying daltoninp as option
on the $point_charges line. The format is completely compatible with the current Dalton
2015 input format (The definition can be found in the manual of the Dalton program at
https://daltonprogram.org/documentation/. See there for more information).
In addition, you can specify further options on the same line as the $point_charges flag.
These are:
18.5. POLARIZABLE EMBEDDING CALCULATIONS 347

• rmin=<float>: minimum distance between an active MM site and any QM center (in
a.u.), treatment is handle by option iskip, (DEFAULT: 0.00 a.u.)

• iskip=(1,2) : treatment of too close MM sites

– (1) zeroing all contributions

– (2) distribute values to nearest non-skipped MM site (DEFAULT)

• rmax=<float>: maximum distance between an active MM site and QM center of

coordinates (in a.u.), sites too far away are skipped (zeroed) (DEFAULT: 1000.00
a.u.)

• nomb: no treatment of many body effects between induced dipoles (all interaction
tensors on the off-diagonal of the response matrix are set to Zero); works best with
isotropic polarizabilities, speeds up calculations (especially for large response matri-
ces), has reduced accuracy, not well tested so far

• longprint=(1,2,3): sets a flag for additional output

– (1) print all MM site input information

– (2) additionally: print all induced dipoles due to nulcei/multipole/electron elec-
tric filed
– (3) additionally: print response matrix

• file=<input file>: specifies a file from which the data group $point_charges is
read. Note that all options which are following on the line in the control file are then
ignored because reading continues in the input file (But here, further options can be
specified after the $point_charges flag). The file has to start with $point_charges
as top line and should be finished with $end

• ccdens: activates the simplified polarizable embedding method

Limitations with respect to standard SCF computations:

• In PE-SCF computations, symmetry cannot be exploited.

• PE-SCF computations do not work in parallel (MPI parallelization).

• For two-component all-electron calculations, the decoupling of the embedding poten-

tial is neglected, i.e. it is affected by a picture-change error just like the two-electron
integrals.

The energy of a PE-SCF calculation printed in the output contains the following terms:

EPE-SCF = EQM + EQM/MM,es + Epol (18.21)

Here, EQM is the energy of the quantum mechanical method of your choice, EQM/MM,es
the electrostatic interaction energy between the QM and the MM region, and Epol the
348 CHAPTER 18. EMBEDDING AND SOLVATION EFFECTS

energy gain due to the total of induced dipole moments. If necessary, missing terms can be
computed without knowledge of the electron distribution.
At the moment, TURBOMOLE does not offer the possibility to generate the necessary
potentials or to create a potential file from a set of coordinates. Embedding potentials can
be obtained from literature or generated by approaches like the LoProp method. [277] Atom
centered polarizabilities are also available from other methods or from experiment. Finally,
there are some polarizable force fields which, in principle, can be used for the PE method
(for example, the AMOEBA force field).

18.5.3 Computational details for post-SCF methods

PERI-CC2 calculations:

Apart from the definition of the embedding described above, the input for PERI-CC2 cal-
culations is the same as without polarizable embedding.
There are several limitations for the use of PERI-CC2:

• only ground state energies, excitation energies and transition moments are supported
(no other properties or gradients and so on)

• no use of symmetry

• no MPI parallelization is available (but SMP binaries work)

• open-shell systems are not covered (exception: two-component references)

PE-MP2 within PTED Reaction-Field Scheme:

In addition to the PERI-CC2 method, the calculation of the ground-state energy is available
at PE-MP2 level within the framework of PTE and PTED reaction-field schemes. The
implementation of PE-MP2 using the PTED approach is inspired by the work published by
Lunkenheimer and Köhn. [278] For the detail about the PTED reaction-field scheme read
section 18.2.5. For the calculation of the ground-state energy the following data groups must
be included in the control in addition to the PE data groups given in section 18.5.2 and
the $ricc2 data group:

$response
fop relaxed
$reaction_field
PTED
cycle

The cycle flag in the $reaction_field data group controls the PTED macro-iterations
between the SCF and post-SCF calculations. In order to invoke the PTED-PE-MP2 calcu-
lations, the pecc2 script must be used. The combination of PTED-PE-MP2 with SCS and
18.5. POLARIZABLE EMBEDDING CALCULATIONS 349

SOC is also available. Furthermore, the calculation of ground-state energy is possible for
closed-shell and open-shell systems with the SMP (OpenMP) and MPI parallelizations.

PE-ADC(2) within post-SCF Reaction-Field Scheme

:The new PE-ADC(2) method [279] implemented in the framework of the post-SCF reaction
field scheme makes the following calculations available in the ricc2 module for both closed-
shell and open-shell systems:

• Vertical excitation energy and transition moments

• Excited-state analytic gradients

• Excited-state properties

For these calculations, in addition to the typical data groups $ricc2, $excitations and
$point_charges, the following keywords should be added to the control file:

$reaction_field
post-SCF
ccs-like
Chapter 19

Molecular Properties,
Wavefunction Analysis, and
Interfaces to Visualization Tools

19.1 Molecular Properties, Wavefunction Analysis, and

Localized Orbitals

Molecular properties (electrostatic moments, relativistic corrections, population analyses

for densities and MOs, construction of localized MOs, NTOs, etc.) can be calculated with
proper. This program is menu-driven and reads the input that determines which prop-
erties are evaluated from standard input (i.e. the terminal or an input file if started as
proper < inputfile). The control file and files referenced therein are only used to deter-
mine the molecular structure, basis sets, and molecular orbitals and to read results computed
before with other programs.
proper is a post-processing tool, mainly intended for interactive use which reads (almost
all) its input from the terminal so that it is (usually) not necessary to modify the control
file.
Several functionalities are also integrated in the programs that generate MOs or densities and
can be invoked directly from the modules dscf, ridft, rimp2, mpgrad, ricc2 and egrad, if
corresponding keywords are set in the control file. If one wants to skip the MO- or density
generating step for dscf, ridft, rimp2, mpgrad, ricc2 it is possible to directly jump to the
routine that carries out the analyses by starting the program with "<program> -proper".
(For ricc2 it is, however, recommended to use instead ricctools -proper.) Currently, the
respective keywords have to be inserted by hand (not with define) in the control file.

350
19.1. MOLECULAR PROPERTIES, WAVEFUNCTION ANALYSIS, AND LOCALIZED ORBITALS351

Here we briefly present the functionalities. A detailed description of the keywords that can
be used in combination with the -proper flag is found Section 22.2.26.

19.1.1 Selection of densities

The proper program tries on start to read all densities that have been pre-calculated with
any of the other programs of the TURBOMOLE package, prints a list with the densities that
have been found and selects one of them for the calculation of properties. The default choice
can be changed in the menu that is entered with the option dens. Here one can also list
or edit additional attributes of the densities or build linear combinations of the available
densities to form differences and superposition of densities.
The feature can thus be used to evaluate difference densities between the ground and excited
electronic states or differences between densities calculated with different electronic structure
methods.
The selected density can then be used in the subsequent menues to evaluate a variety of
properties as expectation values or on a grid of points to generate interface files for visual-
ization.

19.1.2 Electrostatic moments

Use the mtps option in the eval menu in proper, or add the $moments keyword control
file when you start a program with the -proper flag to evaluate electrostatic moments.
Up to quadrupole moments are calculated by default, on request also octupole moments
are available. By default unnormalized traced Cartesian moments are calculated which are
defined as Z
(n)
Qαβ...ν = drρ(r)rα rβ . . . rν ,

where ρ(r) is the charge density as position r. With the option Buckingham one can request
in addition the computation of Cartesian traceless (Buckingham) multipole moments defined
as:
(−1)n dn
Z
(n) 1
Mαβ...ν = drρ(r)r2n
n! drα drβ . . . drν r

19.1.3 Relativistic corrections

The option relcor in the eval menu of proper or the keyword $mvd (when starting a
program with the -proper flag) initiate the calculation of relativistic corrections. With the
-proper flag they are calculated for the SCF or DFT total density in case of dscf and ridft,
for the SCF+MP2 density in case of rimp2 and mpgrad and for that of the calculated excited
state in case of egrad. Quantities calculated are the expectation values hp2 i , hp4 i and the
P
Darwin term h A 1/ZA ∗ ρ(RA )i. Note, that at least the Darwin term requires an accurate
description of the cusp in the wave function, thus the use of basis sets with uncontracted
352 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

steep basis functions is recommended. Moreover note, that the results for these quantities
are not really reasonable if ECPs are used (a respective warning is written to the output).

19.1.4 Population analyses

For population analyses enter the pop menu of proper. The available options and parameters
that can be specified are the same as those for the $pop keyword (vide infra). If an electronic
structure program is started with the -proper flag the population analyses is requested with
the keyword $pop.
mulliken or $pop without any extension start a Mulliken population analysis (MPA). For
-proper the analysis is carried out for all densities present in the respective program, e.g.
total (and spin) densities leading to Mulliken charges (and unpaired electrons) per atom in
RHF(UHF)-type calculations in dscf or ridft, SCF+MP2 densities in rimp2 or mpgrad,
excited state densities in egrad. Suboptions (see Section 22.2.26) also allow for the calcula-
tion of Mulliken contributions of selectable atoms to selectable MOs including provision of
data for graphical output (simulated density of states).
With $pop nbo a Natural Population Analysis (NPA) [280] is done. Currently only the
resulting charges are calculated.
With $pop paboon a population analyses based on occupation numbers [281] is performed
yielding "shared electron numbers (SENs)" and multicenter contributions. For this method
always the total density is used, i.e. the sum of alpha and beta densities in case of UHF, the
SCF+MP2-density in case of MP2 and the GHF total density for (two-component-)GHF.
Note that the results of such an analysis may depend on the choice of the number of modified
atomic orbitals ("MAOs"). By default, the number of MAOs is chosen such that they are
reasonable in most cases (see Section 22.2.26). Nevertheless it is recommended to read
carefully the information concerning MAOs given in the output before looking at the results
for atomic charges and shared electron numbers. For different ways of selecting MAOs see
Section 22.2.26.
With $pop wiberg Wiberg bond indices (WBI) [282] are calculated.

19.1.5 Generation of localized MOs

The option lmos in the mos menu of proper and the keyword $localize with the -proper
flag trigger the calculation of localized molecular orbitals. Per default a Boys localization
including all occupied MOs is carried out (i.e. the squared distance of charge centers of
different LMOs is maximized). Alternative localization methods are Pipek-Mezey, Intrinsic
Bond Orbitals (IBOs), and minimizations of (powers) of the second or fourth orbital moment.

pm Pipek-Mezey localization, maximizes the sum of the squared orbital charges on the
nuclei:
X atoms
X
2
LPM = qiA
i A
19.1. MOLECULAR PROPERTIES, WAVEFUNCTION ANALYSIS, AND LOCALIZED ORBITALS353

It has the lowest operation count of all the implemented localization procedures, but
gives only well-localized orbitals for basis sets without diffuse functions. In difference
to Foster-Boys (see below) it conserves the πσ separation.
boys Foster-Boys localization, minimizes the average orbital spread:
X
LFB = hi|(r − ri )2 |ii
i

where ri = hi|r|ii is the center of the LMO. It is almost as fast as Pipek-Mezey,

but stable with the basis set. Its disadvantage relative to Pipek-Mezey and IBO
localization is that it breaks the π − σ separation for double bonds.
sm second moment localization, minimizes a power of the sum of orbital spreads:
X
LSM,n = hi|(r − ri )2 |iin
i

The exponent is defined with the additional option exp=n where n must be an integer
> 0. For n = 1 SM localization is identical to Foster-Boys. Values n > 1 cause a larger
penalty for LMOs with larger spreads and thereby reduce the variance of the LMO
spreads at the price of a small increase of the average spread and a small increase in
the computational costs.
fm fourth moment localization [283], mimimizes a power of the sum of the fourth central
moments: X
LFM,n = hi|(r − ri )4 |iin
i
The exponent is again defined with the additional option exp=n where n must be
an integer > 0. Fourth moment localization produces in contrast to PM, IBO, FB,
and SM localization LMOs with smaller tails Values n > 1 cause a larger penalty for
LMOs with a larger Kurtosis and thereby reduce the variance of the LMO Kurtosis
at the price of a small increase of the average Kurtosis. The computational costs for
FM localization are 3–5 times larger than for SM localization.
ibo intrinsic bond orbital localization [284], maximizes the sum of the fourth power of the
orbital charges on the nuclei:
X atoms
X
4
LIBO = qiA
i A

The IBO localization is similar to Pipek-Mezey localization, but the localization is

carried out in in a minimal basis of intrinsic atomic orbitals (IAOs, see below) which
make the procedure stable with respect to basis basis extension. The exponent of 4
reduces the problem of (nearly) degenerate minima of the PM functional. The cost of
the IBO localization is for large systems and/or basis sets dominated by the costs for
the construction of the IAOs.

As output one gets localized MOs (written to files lmos or lalp/lbet in UHF cases). In
addition information about the dominant contributions of canonical MOs to the LMOs and
a population analysis and, if requested, also about the centers, the spread and the Kurtosis
of the LMOs is written to standard output.
354 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

19.1.6 Intrinsic Bond Orbitals Analysis

The option ibos in the mos menu of proper initiates the computation of intrinsic bond
orbitals [284] (IBOs). Instead of a Mulliken PA in the AO basis the contributions a popu-
lation analysis in the basis of the intrinsic atomic orbitals [284] (IAOs) and an IAO atomic
charge analysis is done. The start guess for the IBO optimization is slightly different then
the one used with the option lmos, which in the case of degenerate solutions (core orbitals,
symmetry-degenerate LMOs located at the same centers) can lead to slightly different IBOs.
As proposed in Ref. [284] the IAOs are obtained by projection of the occupied orbitals onto
pre-computed atomic orbitals from Hartree-Fock calculations on the isolated atoms in the
correlation consistent triple-ζ basis (cc-pVTZ and cc-pVTZ-PP for pseudo-potential basis
sets). These are available for most main group and transition metal atoms, for atoms beyond
Kr with the standard pseudo-potential cores. If additional definitions are needed, they can
be added to the basis set library.
Available reference orbitals for IAOs:
Atoms all electron default core large core
H–He 1s / cc-pVTZ
Li–Be 2s / cc-pVTZ
B–Ne 2s1p / cc-pVTZ
Na–Mg 3s1p / cc-pVTZ
Al–Ar 3s2p / cc-pVTZ
K 4s2p / TZV 2s1p / LANL2DZ ECP 1s / ecp-18-sdf
Ca 4s2p / cc-pVTZ
Sc–Ni 4s2p1d / cc-pVTZ
Cu–Zn 4s2p1d / cc-pVTZ 2s1p1d /cc-pVTZ-PP
Ga–Kr 4s3p1d / cc-pVTZ 2s2p1d /cc-pVTZ-PP
Rb —
Y–Cd — 2s1p1d / cc-pVTZ-PP
In–Xe — 2s2p1d / cc-pVTZ-PP
Cs–Lu —
Hf–Hg — 2s1p1d / cc-pVTZ-PP
Tl–Rn — 2s2p1d / cc-pVTZ-PP
For transition (earth) alkali and transition metal atoms the inclusion of the highest s orbital
(2s for Li–Be, 3s for Na–Mg, and 4s for K–Zn) can cause problems for systems that contain
the respect cations. In the cations these s orbitals are (essentially) unoccupied and lead the
unphysical delocalized IAOs. To avoid this problem it is possible to restrict the reference
basis for the construction of the IAOs by adding to the control file data group $iaoopts:

$iaoopts
subset=3s2p1d <list of atom indes>
19.1. MOLECULAR PROPERTIES, WAVEFUNCTION ANALYSIS, AND LOCALIZED ORBITALS355

Natural transition orbitals For excited states calculated at the CIS (or CCS) level
the transition density between the ground and an excited state

Eia = hΨex |a†i aa |Ψex i (19.1)

can be brought to a diagonal form through a singular value decomposition (SVD) of the
excitation amplitudes Eia :
√
[O† EV]ij = δij λi (19.2)

The columns of the matrices O and V belonging to a certain singular value λi can be
interpreted as pairs of occupied and virtual natural transition orbitals [285, 286] and the
singular values λi are the weights with which this occupied-virtual pair contributes to the
excitation. Usually electronic excitations are dominated by one or at least just a few NTO
transitions and often the NTOs provide an easier understanding of transition than the
excitation amplitudes Eia in the canonical molecular orbital basis.
From excitation amplitudes computed with the ricc2 program NTOs and their weights (the
singular values) can be calculated with the ntos option in the mos menu of proper or with
ricctools. E.g. using the right eigenvectors for the second singlet excited state in irrep 1
with:

ricctools -ntos CCRE0-2--1---1

Both programs store the results for the occupied and virtual NTOs in files named, respec-
tively, ntos_occ and ntos_vir. The option nto in the grid menu of the proper program
can used to evaluate NTOs for visualization on a grid of points.
Note that the NTO analysis ignores for the correlated methods (CIS(D), ADC(2), CC2,
CCSD, etc.) the double excitation contributions and correlation contributions to the ground
state. This is no problem for single excitation dominated transition out of a “good” single
reference ground state, in particular if only a qualitative picture is wanted, but one has
to be aware of these omissions when using NTOs for states with large double excitation
contributions or when they are used for quantitative comparisons.

Difference densities based on natural transition orbitals If the excitation vec-

tors have been obtained starting from a GHF reference, the NTOs are complex and contain
contributions from both spin function. Moreover, the transitions are usually dominated by
two NTOs at least. Thus, the interpretation of 2c-NTOs may become difficult. To get a sim-
ple picture of the transition at hand still, approximate difference densities can be computed
according to
Nvir Nocc Nocc N vir 
† †
Cia ∗ Cib − Cia ∗ Cja .
X X X X
ρ(r)n = < φa (r) φb (r) φi (r) φj (r) (19.3)
ab i ij a

The first term corresponds to the increase of the occupation of the virtual NTOs, while the
second term corresponds to the decrease of the occupation of the occupied NTOs.
356 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

This approximate difference density is available for excitation vectors obtained with the
following methods: CCS/CIS, CIS(D∞), ADC(2) and CC2. Symmetry other than C1 is
currently not supported. Note that the approximate difference densities are based on the
same approximations as the NTOs, namely ignoring correlation and double excitation con-
tributions.

From excitation amplitudes computed with the ricc2 program the approximate difference
densities are computed with ricctools. E.g. using the right eigenvectors for the second
singlet excited state in irrep 1:

ricctools -diffden CCRE0-1--1---2

This resulting density file can be visualized using the analysis mode of the ricc2 program
as described in Section 10.3.3, e.g. by adding the following lines to the control file

$anadens
calc my_approx_diffden from
1d0 cc2-1a-002-approxdiffden.cao
$pointval

and running

ricc2 -fanal

Fit of charges due to the electrostatic potential: $esp_fit fits point charges
at the positions of nuclei to electrostatic potential arising from electric charge distribution
(for UHF cases also for spin density, also possible in combination with $soghf). For this
purpose the ("real") electrostatic potential is calculated at spherical shells of grid points
around the atoms. By default, Bragg-Slater radii, rBS , are taken as shell radii.
A parametrization very close to that suggested by Kollman (a multiple-shell model with
shells of radii ranging from 1.4*rvdW to 2.0*rvdW , rvdW is the van-der-Waals radius; U.C.
Singh, P.A. Kollman, J. Comput. Chem. 5(2), 129-145 (1984)) is used if the keyword is
extended:
$esp_fit kolman

19.2 Interfaces to Visualization Tools

There are several possibilities to visualize structures, vibrational frequencies, molecular or-
bitals, densities and a large number of properties.
The easiest one is to use the graphical user interface of TURBOMOLE TmoleX. TmoleX is either
included in your TURBOMOLE distribution or can be downloaded for free from the COSMOlogic
web site TmoleX client version. Use the COSMObuild part of TmoleX to read in the 3D
data from file (plt or plv format).
19.2. INTERFACES TO VISUALIZATION TOOLS 357

Visualization of Molecular Geometries

The tool t2x can be used to convert the atomic coordinates stored in the $grad and $coord
data groups into the xyz-format, which is supported by most viewers, e.g. jmol. Typing

t2x > opt.xyz

in a directory containing the control file generates a series of frames using the information
in $grad. Note t2x writes to standard output which here is redirected to a file. If you are
only interested in the most recent structure, type

t2x -c > str.xyz

which only extracts the information on $coord.

Visualization of Densities, MOs, Electrostatic Potentials and Fields

Note that the easiest way to visualize orbitals, densities and electrostatic properties is to
use the graphical user interface TmoleX.
There are several other possibilities to visualize molecular orbitals or densities.
The molden option in the export menu of proper converts the MO and geometry informa-
tion to the molden format. Alternatively one can use the conversion program tm2molden,
which is interactive and self-explanatory. The generated file can be read in by the molden
program (see molden web site).
For larger systems this may become very time-consuming, as plotting data (values on grids)
are calculated by the respective programs (e.g. molden). It is more efficient to calculate the
data for plots (MO amplitudes, densities, etc.) with TURBOMOLE and to use a visualization
tool afterwards, a way, that is described in the following.

Calculation of data on grids to be used for plots with visualization tools (e.g. gOpen-
Mol, available via gopenmol download) can be generated with the options in the grid menu
of proper. Alternatively one can use the keyword $pointval in combination with the
-proper option. This keyword is obeyed by all TURBOMOLE program that generate densities
as dscf, ridft, rimp2 mpgrad, ricc2 (see Section 10.3.3) and egrad. Note, that with
-proper and $pointval all of the following quantities may be calculated simultaneusly, and
that for programs dscf, ridft, rimp2 and mpgrad the density matrix generating steps may
be skipped by typing "<program> -proper".

Electron densities The option dens in the grid menu of proper or with the above
mentioned programs setting of the keyword
$pointval dens
or simply
358 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

$pointval
results in the calculation of densities
X
~P ) =
ρ(R ~ P )φµ (R
Dνµ φν (R ~P ) (19.4)
νµ

~ P . The size of the grid is automatically adjusted to the size

on an orthogonal grid of point R
of the molecule and the resolution is adjusted to yield acceptable plots (for specification of
non-default grid types (planes, lines) and non-default output formats see Section 22.2.26).
The names of the output files are:

td.plt total density (UHF: α density plus β density )

sd.plt spin density (α density minus β density )
mp2d.plt MP2 density
mp2sd.plt MP2 spin density
ed.plt differential density for excited state
esd.plt differential spin density for excited state
<myname>.plt general density passed e.g. by the ricc2 program.

The .plt files may be visualized with gOpenMol; the file coord.xyz, which is also necessary
for gOpenMol, is generated by the above programs, if $pointval is set in the control-file.
For two-component wavefunctions (only module ridft with $soghf is set): Total density is
on file td.plt like for one-component wave functions; this is also true for all other quantities
depending only on the density matrix (electrostatic potential etc.). sd.plt contains the
absolute value of the spin vector density, which is the absolute value of the following vector:
!
  Ψ α
si (r) = Ψ∗α Ψ∗β σ i i = x, y, z
Ψβ

$pointval fmt=txt
leads to a file containing the spin density vectors, which can be used by gOpenMol. It is
advisable to choose ca. one Bohr as the distance between two grid points. The options for
visualizing two-component wavefunctions are not yet available in the proper program.

TDDFT Transition densities The escf program can plot the TDDFT transition
densities. For the excited state n the transition density is
√ X n
ρn (r) = 2 (X + Y n )ia φi (r)φa (r) (19.5)
ia
√ Xp n
= 2 ωia /ωn Zia φi (r)φa (r) (19.6)
ia

where i (a) are occupied (virtual) orbitals, ωn is the excitation energy, ωia = i − a , and
Z n contains of the TDDFT vector coefficients (for singlet closed-shell). Note that TDDFT
19.2. INTERFACES TO VISUALIZATION TOOLS 359

transition density are different from the differential and non-relaxed densities. Transition
density can be plotted setting:
$pointval transdens
By default the first excited state (of the first irrep) will be printed. To print other states
add the group

$tdlist
irrep1 num1 col1
irrep2 num2 col2
....

where irrep is the number of the irrep considered in the $soes data group, num is the number
of the state, col is the column number in the case of degenerate irreps.
The plot are stored in files named:

vcao_td_<irrep>_<num>.<col>.plt

or with different extension for the different formats.

Available output formats: To get a list of the available output formats of the proper
program invoke the format option with the flag help. For the direct generation of graphics
output different formats are available for 3D and 2D plots. They depend furthermore on
the graphics libraries linked to proper. Use the format option with the help flag for a 2D
and a 3D grid to get the list of formats available for the two cases. For the output formats
available in combination with -proper keyword to export data on grids see Section 22.2.26.

Electrostatic potentials In an analogous way electrostatic potentials can be calculated

on grids. The option ($pointval) pot leads to the calculation of the electrostatic potential
of electrons and nuclei (and external constant electric fields and point charges Q if present).
Z
~ ρ(~r) 3 X ZA 
~P E~+
X Q 
V (RP ) = − d ~r + + R (19.7)
rP r RP A RP Q
A Q

In order to prevent the calculation of singularities at the positions of nuclei, for grid points
that are closer to a nucleus than 10−6 a.u. the charge of the respective nucleus is omitted
in the calculation of the electrostatic potential for these points. The output files are termed
tp.plt, sp.plt, etc.

Electric fields (as derivatives of potentials) are calculated with ($pointval) fld. The
absolute values of electric fields are written to files tf.plt, sf.plt, etc. For non-default
grid types and outputs that allow also for displaying of components of electric fields see
Section 22.2.26.
360 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

Exchange-correlation potentials (Only for DFT) Computation of the Kohn-Sham

exchange-correlation potential on a grid is requested with
$pointval xc
This functionality is not (yet) available in the proper program.

Canonical molecular orbitals. Visualization of molecular orbitals, i.e. generation of

.plt-files containing amplitudes of MOs i
X
~P ) =
Ai ( R ~P )
ciν φν (R (19.8)
ν

or in the two-component case

X
~P ) =
AΓi (R ~P )
cΓiν φν (R (19.9)
ν

with Γ as a part of the coefficient matrix (Re(α), Im(α), Re(β), Im(β)), is achieved e.g. by
($pointval) mo 10-12,15. This triggers the calculation of amplitudes for the MOs/spinors
10-12 and 15 on the grid. The numbering of MOs refers to that you get from the first column
of the output of the tool Eiger, the one for spinors refers to the file EIGS. The filenames
contain the type of the irreducible representation (irrep) of the MO, the current number
within this irrep and in case of UHF calculations also the spin, e.g. 2a1g_a.plt contains
amplitudes for the second alpha-spin MO of a1g type. For more-dimensional irreps columns
are written to separate files, e.g. 1t2g1_a.plt, 1t2g2_a.plt and 1t2g3_a.plt contain the
amplitutes of the three columns of the first irrep (alpha spin) of type t2g .
Two-component wavefunctions (only module ridft and only if $soghf is set): By default
only the density of the chosen spinors is written in files named e.g. 10a_d.plt. Visualization
of the amplitudes of the different spinor parts is achieved e.g. by
$pointval mo 10-12,15 minco real,
where real is a plotting threshold that may take values between zero and one. The corre-
sponding part Γ of the spinor (Re(α), Im(α), Re(β), Im(β)) will be written to file, if N Γ
(see below) is larger than that threshold.

NΓ = tr(DΓ S)
X
Γ
Dµν = cΓ∗ Γ
iν ciµ
i

The filenames consist of the number of the spinor according to file EIGS and an additional
number for the respective part Γ of the spinor (1 for Re(α), 2 for Im(α), 3 and 4 for the
corresponding β-parts) e.g. 10a_4.plt for the Im(β) of spinor 10.

Localised molecular orbitals If one has generated localized molecular orbitals (LMOs,
see above) they can also be visualized.
$pointval lmo 3-6,8
19.2. INTERFACES TO VISUALIZATION TOOLS 361

as an example, leads to calculation of amplitudes for LMOs 3-6 and 8. The coefficients are
read from file lmos (UHF: lalp and lbet), the numbering refers to the output from the
localization section. For an UHF case the β spin orbitals get an offset of NMO , where NMO
is the total number molecular orbitals for each spin case. If one has e.g. 22 orbitals per spin
case and is interested in plotting the first 3 β-type LMOs only, one have to type
$pointval lmo 23-25

Natural molecular orbitals for two-component wavefunctions (only module

ridft and only if $soghf is set): In two-component calculations it is often useful to visu-
alize natural molecular orbitals. In contrast to one-component calculations the occupation
numbers are no longer close to zero, one or two, but can take any value between zero and
two. Therefor
$natural orbitals file=natural
$natural orbital occupation file=natural
has to be set additionally to $soghf (also possible via define).
By setting
$pointval nmo 9
in control-file a gOpenMol-compatible file named nmo_9.plt is written which can also be
visualized with TmoleX.

Natural atomic orbitals If one has generated natural molecular orbitals (NAOs, see
above) they can be visualized with the following command in the control file:
$pointval nao 7-9,12
where the numbers of the NAOs are in the output of the population analysis.

Natural transition orbitals If natural transition orbitals (NTOs) for electronic ex-
citations are available in files named nto_nocc and nto_vir for, respectively, the occupied
and virtual NTOs, plot files for visualizing them can be generated by setting
$pointval nto 1-5
This will generate plot files for the first five occupied and virtual NTOs. The plot file are
named nto_vir_n.plt, where n is the NTO index.

Non-default grids are described in detail in Sections 22.2.26. In the proper program
non-default grids can be specified with the grid option. Calculation of the above quantities
at single points is needed quite often, thus an example is given here.

$pointval geo=point
7 5 3
0 0 7
362 CHAPTER 19. PROPERTIES AND ANALYSIS, AND GRAPHICS

1 2 3

calculates densities at points (7,5,3), (0,0,7) and (1,2,3). Output is (x,y,z, density), output
file suffix is .xyz.
We note in passing that calculation of electrostatic potential at positions of nuclei may be
used as an efficient tool to distinguish atoms of similar atomic numbers thus providing a
complement to X-Ray Structure Analysis (for details, see ref. [287]).

Electron localization function (ELF) [288] is calculated with ($pointval) elf, the
ELF is defined as

−1
ELF(~r) = 1 + χσ (~r)2 , (19.10)

where the index σ describes the spin and χ is a dimensionless localization index defined with
respect to the uniform electron gas,

χσ (~r) = Dσ (~r)/Dσ0 (~r) (19.11)

3
Dσ0 (~r) = (6π 2 )2/3 ρσ (~r)5/3 . (19.12)
5
Possible values of the ELF range from 0 to 1. ELF = 1 corresponds to perfect localization
and ELF = 0.5 to electron-gas like pair probability.
Chapter 20

Orbital Dependent Kohn-Sham

Density Functional Theory

20.1 Theoretical Background

Approximations to the exchange-correlation (XC) functional of the Kohn-Sham (KS) Density

Functional Theory (DFT) can be classified by the so-called “Jacob’s ladder.” The ground on
which the ladder lies is the Hartree approximation (XC energy is zero), and the first rung
is the local density approximation (LDA) in which the XC energy density is a simple local
function of the density. The second rung of the Jacob’s ladder is the generalized gradient
approximation (GGA): in this case the XC energy density depends also on the gradient of the
density. In the third rung (meta-GGA) an additional variable is used, the Kohn-Sham kinetic
energy density which allows, e.g., to construct self-correlation-free functionals. Functionals
in the above rungs can have high accuracy for different class of problems in chemistry and
solid-state physics, but their main limitation is the self-interaction error (SIE) [289–292]. To
avoid the SIE the exchange must be treated exactly and this can be achieved by functionals
in the fourth rung which depend explicitly on all the occupied KS orbitals. In the KS
formalism the EXX (exact-exchange) energy is (for closed-shell systems ns = 2) [289–292]:
occ. occ. Z Z KS 0 KS 0
ns X X φKS (r)φKS
b (r)φa (r )φb (r )
ExEXX = − drdr0 a . (20.1)
2 a kr − r0 k
b

i.e. the same functional form of the Hartree-Fock (HF) exchange but computed with KS
orbitals which are obtained using a self-consistent local EXX potential. At this point we
should recall that hybrid DFT functionals (including HF exchange), doesn’t belong to the
KS formalism: in hybrid DFT, in fact, the non-local HF exchange operator v̂xNL (r, r0 ) =
Pocc. (r)φa (r0 )
− a φakr−r 0k is employed in the self-consistent (Generalized Kohn-Sham) equations
determining the orbitals.

363
364 CHAPTER 20. ORBITAL DEPENDENT DFT

While LDA, GGA, meta-GGA and hybrid functionals are implemented (for ground-state
calculations) in the dscf and ridft, the odft module considers functionals of the fourth
rung. Currently exchange-only orbital-dependent approaches are implemented in the odft
module. The EXX KS local potential (vxEXX (r)) can be obtained using the optimized effective
potential (OEP) method (in each self-consistent step): [290–293]:
Z occ. X
vir.
X φs (r)φa (r)
dr0 χs (r, r0 )vxEXX (r) = 2ns hφa | v̂xNL |φs i (20.2)
a s
a − s
Pocc. Pvir. φa (r)φs (r)φs (r0 )φa (r0 )
where χs (r, r0 ) = a 2ns s a −s is the non-interacting density response.
An effective approximation to the OEP-EXX potential is given by the Localized Hartree–
Fock (LHF) potential [289] which is given by
occ.
φi (r0 )φj (r0 )
Z
X φi (r)φj (r)
vxLHF (r) = − ns dr0
ij
ρ(r) kr − r0 k
occ.
X φi (r)φj (r)
+ ns hφi | vxLHF − v̂xNL |φj i (20.3)
ij
ρ(r)

where the first term is called Slater potential and the second term correction term. If terms
i 6= j are neglected in the correction term, the Krieger-Li-Iafrate (KLI) potential [294] is
obtained. Note that the Eq. (20.3) depends only on occupied orbitals, whereas Eq. (20.2)
depends also on virtual orbitals. The LHF total energy is assumed to be the EXX total
energy, even if LHF is not variational (although the deviation from the EXX energy is
very small, usually below 0.01%). The LHF potential is equivalent to the Common Energy
Denominator Approximation (CEDA) [295] and to the Effective Local Potential (ELP) [296].
Both OEP-EXX and LHF (in contrast to functionals of the first three rungs) satisfy the
HOMO condition [294]

hφHOM O | vx |φHOM O i = hφHOM O | v̂xNL |φHOM O i , (20.4)

and the asymptotic relation [297, 298]

r →∞ 1
l
vx (rl ) −→ hφM | vx − v̂xNL |φM i − . (20.5)
rl
where φM is the highest occupied orbital which do not have a nodal surface in the asymptotic
region along direction rl . Considering together with condition (20.4), we finally obtain that:

- vx (r) will approach −1/r along all directions where φHOM O (r) does not have a nodal
surface in the asymptotic region (e.g. this is the case of atoms);

- on directions which belong to the nodal surface of the HOMO, the vx (r) will approach
hφM | vx − v̂xNL |φM i − 1/r.

Both OEP-EXX and LHF gives total energies very close to the Hartree-Fock one (actually
ELHF > EEXX > EHF ), thus, without an appropriate correlation functional, these methods
are not suitable for thermochemistry. On the other hand OEP-EXX and LHF give very good
20.2. IMPLEMENTATION 365

KS orbital spectra. In fact the eigenvalues of the HOMO is very close to the Hartree-Fock
and to exact ionization potential (I.P): this is in contrast to functional of the first three
rungs which underestimate the HOMO energy by several eVs. In addition a continuum set
of bound unoccupied orbitals are obtained. Thus OEP-EXX or LHF KS orbitals are very
good input quantities for computing NMR shielding constants [299], energy-levels in hybrid
interfaces [300] and TD-DFT excitation energies [301] (the latter using LDA/GGA kernels,
not the hybrid ones).

20.2 Implementation

Both the OEP-EXX and LHF methods can be used in spin–restricted closed–shell and spin–
unrestricted open–shell ground state calculations. Both OEP-EXX and LHF are parallelized
in the OpenMP mode.

20.2.1 OEP-EXX

In the present implementation the OEP-EXX local potential is expanded as [293]:

X Z gp (r0 )
EXX
vx (r) = cp dr0 , (20.6)
p
|r − r0 |

where gp are gaussian functions, representing a new type of auxiliary basis-set (see directory
xbasen). Inserting Eq. (20.6) into Eq. (20.2) a matrix equation is easily obtained for the
coefficient cp . Actually, not all the coefficients cp are independent each other as there are
other two conditions to be satisfied: the HOMO condition, see Eq. (20.4), and the Charge
condition Z X
cp gp (r)dr = −1 , (20.7)
p

which ensures that vxEXX (r) approaches −1/r in the asymptotic region. Actually Eq. (20.6)
violates the condition (20.5) on the HOMO nodal surfaces (such condition cannot be achieve
in any simple basis-set expansion).
Note that for the computation of the final KS Hamiltonian, only orbital basis-set matrix
elements of vxEXX are required, which can be easily computes as three-index Coulomb in-
tegrals. Thus the present OEP-EXX implementation is grid-free, like Hartree-Fock, but in
contrast to all other XC-functionals.

20.2.2 LHF

In the LHF implementation the exchange potential in Eq. (20.3) is computed on each grid-
point and numerically integrated to obtain orbital basis-sets matrix elements. In this case
the DFT grid is needed but no auxiliary basis-set is required. The Slater potential can be
computed numerically on each grid point (as in Eq. 20.3) or using a basis-set expansion
as [289]:
366 CHAPTER 20. ORBITAL DEPENDENT DFT

occ.
ns X T
vxslat (r) = u χ(r)χT (r)S −1 K ua . (20.8)
ρ(r) a a

Here, the vector χ(r) contains the basis functions, S stands for the corresponding overlap
matrix, the vector ua collects the coefficients representing orbital a, and the matrix K
represents the non-local exchange operator v̂xNL in the basis set. While the numerical Slater
is quite expensive but exact, the basis set method is very fast but its accuracy depends on
the completeness of the basis set.
Concerning the correction term, Eq. (20.3) shows that it depends on the exchange potential
itself. Thus an iterative procedure is required in each self-consistent step: this is done using
the conjugate-gradient method.
Concerning conditions (20.4) and (20.5), both are satisfied in the present implementation.
KS occupied orbitals are asymptotically continued [297] on the asymptotic grid point r
according to:
 (Q+1)/βi −1
|r|
φ̃i (r) = φi (r0 ) e−βi (|r|−|r0 |) , (20.9)
|r0 |
√
where r0 is the reference point (not in the asymptotic region), β = −2i and Q is the
molecular charge. A surface around the molecule is used to defined the points r0 .

20.3 How to Perform

OEP-EXX

To run OEP-EXX calculations select:

$dft
functional oep

As the computation of the OEP functional is completely analytic and grid free, any selection
of a grid type or size will not influence the OEP calculation in contrast to other density
functionals.
Particular care is instead required to orbital and auxiliary basis set. An arbitrary combi-
nation of them can lead to very good total energy (i.e. very close to the Hartree-Fock one)
but unphysical OEP potential. In the present release we strongly recommend to use the
d-aug-cc-pVTZ-oep basis set and the corresponding auxiliary basis set (directory xbasen).
The following options can modify the quality, time and output of an OEP calculation. All
the options can be set by define.
Every option has a reasonable default value so the user does not need to select any of the
options below to run a proper OEP calculation.
20.3. HOW TO PERFORM 367

$oep options
Listing of all possible options for the flag $oep.

charge vector integer

The Charge condition expansion coefficients in auxiliary basis set representa-
tion can be calculated in different kinds.
The selection of integer = 1 will use the following ansatz to calculate the coef-
ficients: (
− N 01 · G1p if GP 6= 0
cP1 = aux

0 if GP = 0 .
0
GP is the integral over a normalized Gaussian auxiliary basis function. Naux
is the number of auxiliary basis functions with GP 6= 0.
The selection of integer = 2 will use the following ansatz to calculate the
coefficients: 
 − P 1G if GP 6= 0
P
cP2 = P
 0 if GP = 0 .

The variable integer must have an integer value. The default value is 2.
condition [string2 ] string
In the OEP method two constraints can be applied in the OEP equation. This
is the HOMO condition and the Charge condition. The variable string can
have the values none, HOMO, Charge and both. No condition is chosen when
none is elected. The HOMO condition is chosen when HOMO is elected. The
Charge condition is chosen when Charge is elected. The HOMO condition and
the Charge condition are chosen when both is elected.
The variable string2 is optional and only electable if a spin–unrestric-ted calcu-
lation is performed. The variable string2 can have the values alpha and beta.
If string2 = alpha then the condition is defined for the alpha spin channel. If
string2 = beta then the condition is defined for the beta spin channel. Both
spin channels can have different values.

Example:

$oep
condition alpha HOMO
condition beta Charge

If only one spin channel is defined the other spin channel uses the same condi-
tion automatically. The default value in any case is string = both.
core memory integer
Core memory is the amount of main memory given to the OEP calculation
to store the three index integrals calculated during the OEP calculation. The
core memory amount is given MB. The calculation runs as fast as possible if
368 CHAPTER 20. ORBITAL DEPENDENT DFT

all three index integrals can be stored in the core memory. The variable integer
must have an integer value. The default value is 200.
debug
Print further information about the OEP calculation especially matrices and
vectors used during the OEP calculation. Use this option carefully since a lot
of data is written. The default value is .false..
eigenvalue difference integer
Two molecular orbitals are considered as degenerated (due to symmetry or
incidentally), if the difference between them is smaller then 10−integer . The
variable integer must have an integer value. The default value is 6.
plot coefficient string
The expansion coefficients for the auxiliary basis functions which build the lo-
cal exact exchange potential are written to the file oepcVx.dat or in case of a
spin–unrestricted calculation to the files oepcVxa.dat and oepcVxb.dat.
If string is cartesian the expansion coefficients are given for a Cartesian
atomic orbital auxiliary basis, if string equals spheric the expansion coeffi-
cients are given for a spherical atomic orbital auxiliary basis. In any case the
expansion coefficients are given for the single atomic orbital auxiliary basis
function and contain no information about the symmetry of the system (c1
case). The default value is cartesian.
reference potential
Use the reference potential constructed by the applied conditions to the OEP
calculation as exchange potential. The solution of the OEP equation is skipped.
The default value is .false..

LHF

To run a LHF calculations select:

$dft
functional lhf
gridsize 3

This can be done using define (modified grid are not supported) and then run odft.
A more suitable procedure is the following:

1) Do a Hartree–Fock calculation using dscf.

2) Use the script lhfprep to prepare the control file (the old control file will be saved in
control.hf and the molecular orbitals in mos.hf or in alpha.hf and beta.hf for the
spin-unrestricted case). See lhfprep -help for options. Actually LHF can be started
from any guessed orbitals, but if HF orbitals are used, a much faster convergence is
expected. By default the script lhfprep will add/modify the control file with:
20.3. HOW TO PERFORM 369

$dft
functional lhf
gridtype 6
gridsize 3
radsize 3
$lhf
off-diag on
num-slater off
asymptotic dynamic=1.d-3
conj-grad conv=1.d-6 maxit=20 output=1 asy=1
slater-dtresh 1.d-9
slater-region 7.0 0.5 10.0 0.5
corrct-region 10.0 0.5
$scfdump
$scfiterlimit 30
$scfconv 6
$scfdamp start=0.000 step=0.500 min=0.50
$scforbitalshift noautomatic
$correction matrix-elements file=lhfcg
$correction alpha matrix-elements file=lhfcg_alpha
$correction beta matrix-elements file=lhfcg_beta

3) Run odft.

With the LHF potential Rydberg series of virtual orbitals can be obtained. To that end,
diffuse orbital basis sets have to be used and special grids are required.
gridtype 4 is the most diffuse with special radial scaling; gridtype 5 is for very good
Rydberg orbitals; gridtype 6 (default in Lhfprep) is the least diffuse, only for the first
Rydberg orbitals.
Only gridsize 3–5 can be used, no modified grids.
Use test-integ to check if the selected grid is accurate enough for the employed basis-set,
see page 410.
The options in the $lhf group are:

off-diag on
The LHF exchange potential is computed (default);
off-diag off
The KLI exchange potential is computed (can be selected by lhfprep -kli).
num-slater on
the Slater potential is calculated numerically everywhere: this is more accurate
but quite expensive. When ECPs are used, turn on this option. It can be
selected by lhfprep -num.
370 CHAPTER 20. ORBITAL DEPENDENT DFT

num-slater off
the Slater potential is computed using basis-sets. This leads to very fast cal-
culations, but accurate results are obtained only for first-row elements or if
an uncontracted basis set or a basis set with special additional contractions is
used. This is the default.
asymptotic
for asymptotic treatment there are three options:
asymptotic off
No asymptotic-treatment and no use of the numerical Slater. The total
exchange potential is just replaced by −1/r in the asymptotic region.
This method is the fastest one but can be used only for the density-
matrix convergence or if Rydberg virtual orbitals are of no interest.
asymptotic on
Full asymptotic-treatment and use of the numerical Slater in the near
asymptotic-region. It can be selected by lhfprep -asy.
asymptotic dynamic=1.d-3
Automatic switching on (off) to the special asymptotic treatment if the
differential density-matrix rms is below (above) 1.d-3. This is the default.
pot-file save
the converged Slater and correction potentials for all grid points are saved in
the files slater.pot and corrct.pot, respectively. Using pot-file load, the
Slater potential is not calculated but read from slater.pot (the correction
potential is instead recalculated). For spin unrestricted calculations the corre-
sponding files are slaterA.pot, slaterB.pot, corrctA.pot and correctB.pot.
homo
allows the user to specify which occupied orbital will not be included in the
calculation of correction potential: by default the highest occupied orbital is
selected. This option is useful for those systems where the HOMO of the
starting orbitals (e.g. EHT, HF) is different from the final LHF HOMO. homob
is for the beta spin.
correlation func=functional
a correlation functional can be added to the LHF potential: use func=lyp for
LYP, or func=vwn for VWN5 correlation.

For other options see 20.3.

20.4 How to plot the exchange potential

It is recommended to check plots of the exchange potential for both OEP-EXX and LHF
potential, to avoid spurious numerical oscillations (which usually originates from too small
or too large basis-set). To plot the LHF potential over a line, add to the control file (e.g.
for a 2000 points along the z axis):
20.5. HOW TO QUOTE 371

$pointval xc geo=line
grid1 vector 0 0 1 range -10,10 points 2000
origin 0 0 0

and run odft -proper. The plotting subroutine reads the file lhfcg, containing the matrix
elements of the correlation potential is already generated by a previous run. The file tx.vec
will be generated with four columns (distance, LHF potential, Slater potential, Correction
potential).
The procedure to plot the OEP-EXX potential is the same. In this case the expansion
coefficients (see Eq. 20.6) are read from the file oepcVx.dat (Cartesian format). The file
tx.vec will be generated with four columns (distance, EXX potential, EXX potential, zero).

20.5 How to quote

• For LHF calculations with odft:
Efficient localized Hartree–Fock methods as effective exact-exchange Kohn–Sham meth-
ods for molecules Fabio Della Sala and Andreas Görling J. Chem. Phys. 115, 5718
(2001) and The asymptotic region of the Kohn–Sham exchange potential in molecules
Fabio Della Sala and Andreas Görling J. Chem. Phys. 116, 5374 (2002)

• For OEP-EXX calculations with odft:

Numerically stable optimized effective potential method with balanced Gaussian basis
sets Andreas Heßelmann, Andreas W. Götz, Fabio Della Sala, and Andreas Gr̈ling
J. Chem. Phys. 127, 054102 (2007).
Chapter 21

Vibronic absorption and emission

spectra

21.1 Theoretical Background

21.1.1 Vibronic spectra at zero temperature

The accurate prediction of absorption and emission spectra often requires a quantum me-
chanical treatment of nuclear vibration [302]. In addition, the geometrical differences be-
tween ground and excited state structures, and the differences in vibrational spectra and
normal modes lead to mode mixing, making the prediction of the vibrational structure in
electronic spectra non-trivial. Under the neglect of anharmonicity, these effects can be
described by the Duschinsky rotation [303]

Qi = D + JQf , (21.1)

where Qi and Qf denote initial and final state vibrational coordinates, respectively. D is
the geometric displacement between ground and excited state vibrational structures and J
is the Duschinsky rotation matrix. In case of absorption, the initial state is the electronic
ground state and the final state is the electronically excited state; in case of emission, the
opposite order applies.
Within the harmonic oscillator approximation, the absorption and emission spectra are given
by:

4π 2 ω X
σabs (ω) = |µif |2 |hθ0 (Qi )|θvf (Qf )i|2 δ(Evf − E0 − ω) (21.2)
3c v
f

372
21.1. THEORETICAL BACKGROUND 373

and
4ω 3 X
σem (ω) = 3
|µif |2 |hθ0 (Qi )|θvf (Qf )i|2 δ(E0 − Evf − ω), (21.3)
3c v f

respectively. Here, E0 denotes the absolute energy of the initial state where all quantum
numbers are zero; Evf denotes the final state with quantum numbers vf . Both cases assume
that absorption and emission occurs from the lowest vibrational level of the initial state (zero
temperature approximation). The absorption spectrum (σabs ) is given as the absorption
cross section in atomic units (Bohr2 ); the emission spectrum (σem ) is given as the emission
rate in inverse atomic time units.
Writing the delta function in Eqs. 21.2 and 21.3 as a Fourier transform and applying Mehler’s
formula [304,305], the infinite sum in equations 21.2 and 21.3 can be eliminated and written
in terms of a generating function G(t) in the time domain, which for absorption reads
Z ∞
X
2
 1X f 
|hθ0 (Qi )|θvf (Qf )i| δ(Evf −E0 −ω) = dt exp −it(∆Eif − ωj −ω) G(t), (21.4)
v −∞ 2 j
f

with ∆Eif being the adiabatic excitation energy. The generating function is given by
1
det(S−1 Ωi Ωf ) 2

N
exp DT (Ωf BJM−1 JT Ωf B − Ωf B)D , (21.5)


G(t) = 2 2
det(L) det(M)

where Ωi , Ωf , S, B are diagonal matrices with (Ωi )kk = ωki , (Ωf )kk = ωkf , Skk = sinh(iωkf t),
and Bkk = tanh(iωkf t/2). T denotes the transpose of the matrix. ωki and ωkf denote vibra-
tional frequencies of initial and final state, respectively. Matrices L and M are obtained as
M = JT Ωf BJ + Ωi and L = JT Ωf B−1 J + Ωi . Hence, knowledge of ground and excited state
structures and their vibrational spectra allows the construction of all matrices appearing in
Eq. 21.5 and G(t) can be propagated in time. In practice, G(t) has to be truncated after a
maximum time tmax . In addition, G(t) is multiplied by a damping function exp(−t/τ ) with
lifetime τ , which leads to a Lorentzian broadening of the spectral lines. More details can be
found in references [302, 306].

21.1.2 Single Vibronic Level Spectra

Radless allows the calculation of spectra originating from vibrationally singly excited ini-
tial states, i.e. single vibronic level (SVL) emission spectra and Vibrationally Promoted
Electronic Resonance (VIPER) spectra [306]. In the frequency domain the SVL emission
spectrum for a vibronic initial state |θk1 i, where mode k is singly-excited, reads
4ω 3 X
σ 1 em,k (ω) = 3
|µif |2 |hθk1 |θvf i|2 δ(∆Eif + E0i + ωki − Evf − ω). (21.6)
3c v
f

Analogously, the absorption spectrum from a singly excited vibrational state reads
4π 2 ω X
σ 1 abs,k (ω) = |µif |2 |hθk1 |θvf i|2 δ(∆Eif − E0i − ωki + Evf − ω). (21.7)
3c v f
374 CHAPTER 21. VIBRONIC SPECTRA

Using Mehler’s formula and recursive harmonic oscillator recursive relationships, Eq. (21.6)
can be formulated in time-domain for emission
Z ∞
1 4ω 3 2 1
dt exp it ∆Eif + E0i + ωki − ω G1k (t), (21.8)


σ em,k (ω) = |µif |
3c3 2π −∞

with the generating function

h i
G1k (t) = G(t) × ωki 2(M −1 )kk + 4D† Ωf BJRk J† Ωf BD − 2(L−1 )kk , (21.9)

k
with Rij = (M −1 )ik (M −1 )kj . For absorption, the time-domain expression reads
∞
4π 2 ω
Z
1
σ 1 abs,k (ω) |µif |2 dt exp −it ∆Eif − E0i − ωki − ω G1k (t).(21.10)



=
3c 2π −∞

Here, E0i denotes the zero point vibrational energy of the initial state. The generating
function G1k (t) in Eq. (21.10) is identical in structure to Eq. (21.9), but initial state index
i refers to the electronic ground state and state index f refers to the electronically excited
state.

21.2 Implementation

Any electronic structure method can be used with radless if the necessary input, i.e.
structures and vibrational data for the electronic ground and excited state, is provided.
After successful calculation of the ground and excited state structures and their vibrational
spectra, matrices of Eq. 21.5 are constructed at different times t to obtain the generating
function. Fourier transform of the generating function together with the appropriate pref-
actor gives the absorption or emission spectra. Matrix products are evaluated using basic
linear algebra subroutines (BLAS) [307] for real and complex matrices. The spectra are
obtained by discrete Fast Fourier transformation [308] of G(t).

21.3 Functionalities of radless

21.3.1 How to use radless

Radless allows the calculation of vibronic absorption and emission spectra. Absorption
and emission spectra must be computed in separate jobs. Calculations with radless can
only be done within C1 symmetry. The following steps are necessary to prepare the required
input:

1. Optimize ground state structure and compute vibrational spectrum with aoforce or
NumForce.

2. Optimize excited state structure and compute vibrational spectrum using NumForce.
21.3. FUNCTIONALITIES OF RADLESS 375

3. In a new directory, copy the S1-geometry coord and its control file (control), which
points to the coordinate file coord.

4. In the same directory, copy the S0-geometry coord-gs and its control file (control-gs)
pointing to coord-gs.

5. Make sure the excited state control file (control) points to vibrational modes, fre-
quencies, and reduced masses of the excited state, whereas the ground state control file
(control-gs) points to vibrational frequencies, normal modes, and reduced masses of
the ground state. This can be achieved either by renaming ground state vibrational
files (e.g. vib_normal_modes-gs, vibspectrum-gs etc.) or by directly adding the
vibrational data to the control-gs file.

6. Add necessary keywords to excited state control file (control):

(a) $fluorescence or $absorption, to specify what kind of spectrum will be cal-

culated; the two keywords are exclusive.
(b) $gsenergy= number ground state total energy (in a.u.) of the excited state
structure
(c) $esenergy= number excited state total energy (in a.u.) of the excited state
structure
(d) $spectral width integer; spectral width for fluorescence (for absorption, spec-
tral width is given in control-gs) spectrum in atomic units. Default: adiabatic
excitation energy of the current system plus four times the inverse lifetime. The
radiative rate will be computed by integration from 0 to integer. Too high
integer can lead to artifacts of the radiative rate due to the numerical error of
the discrete Fourier transform in the high frequency region. It is recommended
to set integer by approximately two times the full width at half maximum higher
than the adiabatic excitation energy. The fluorescence spectrum will be written
from 0 to integer onto the file FLUORESCENCE_spectrum.dat.
(e) $transdip number: magnitude of transition dipole (length representation) in
atomic units (obtained from escf or egrad output). This is only required for
fluorescence spectra calculations; for absorption spectra the transition dipole is
given in control-gs. Default: 1.0 au.
(f) $max time integer: total integration time in atomic units optional keywords.
Due to the fast Fourier transform, integer should be a power of 2.

7. Add optional keywords to excited state control file (control):

(a) $broadening: employ lifetime broadening

(b) $lifetime integer: if broadening is used, integer is the lifetime of the damping
function given in atomic units
(c) $dhoa: Displaced harmonic oscillator approximation is applied. Duschinsky
matrix is set to unity (no mode mixing).
376 CHAPTER 21. VIBRONIC SPECTRA

(d) $delta_t number: use an integration time step of number (default: 1) in atomic
units
(e) $zerofil number: increase the number of data points in spectrum by adding
zeros to the generating function to obtain number times the number of data
points in time domain; zerofilling is recommended to increase accuracy in peak
positions.
(f) $deuterium: replace all hydrogen atoms by deuterium
(g) $vib1_mode number: calculate SVL spectrum of initial state where mode number
is singly excited; numbering refers to aoforce/NumForce output. Depending
on the keyword used ($absorption /$emission ), this will either be the SVL
absorption or the SVL emission spectrum.
(h) $vib1_separate: calculate SVL spectra of all possible singly excited initial
states. Depending on the keyword used ($absorption /$emission ), these
will either be the SVL absorption or the SVL emission spectra.
(i) $debug_radless : print out generating function
(j) $scaling number: scales excited state frequencies by number

8. Add necessary keywords to control-gs

(a) $gsenergy= number ground state total energy of the ground state structure
(b) $esenergy= number excited state total energy of the ground state structure
(c) $transdip number: magnitude of transition dipole (length representation) in
atomic units (obtained from escf or egrad output). (This is only required for
absorption spectra calculations). Default: 1.0 au.
(d) $spectral width integer; spectral width for absorption spectrum in atomic
units. Default: twice the adiabatic excitation energy of the current system. The
total frequency integrated absorption will be computed by integration from 0 to
integer. The absorption spectrum will be written from 0 to integer onto the
file ABSORPTION_spectrum.dat.

9. Add optional keywords to control-gs

(a) $scaling number: scales ground state frequencies by number

Start a Radless calculation with radless > radless.out.

21.3.2 Output of radless

1. Absorption and emission spectra are written to files, ABSORPTION_spectrum.dat and
FLUORESCENCE_spectrum.dat, respectively. The first column gives the frequency (ω)
in atomic units. In case of absorption, the second column gives the absorption cross
section in atomic units (Bohr2 ). In case of emission, the second column contains the
emission rate in inverse atomic time units.
21.3. FUNCTIONALITIES OF RADLESS 377

2. If $vib1_separate or $vib1_mode are specified, SVL absorption or emission spectra

are written to files, ABSORPTION_spectrum_vib1_number.dat and
FLUORESCENCE_spectrum_vib1_number.dat, respectively. The first column gives the
frequency (ω) in atomic units. In case of absorption, the second column gives the
absorption cross section in atomic units (Bohr2 ).

3. If $debug_radless is specified, additional output files are generated. These include

Genfunc_fluor.dat and Genfunc_abs.dat, containing the generating functions for
emission and absorption, respectively. Real part is given in the first column, imaginary
part is given in the second column.

4. DUSCHINSKY.dat contains the Duschinsky rotation matrix

5. Geometric displacements projected on vibrational modes are given in the standard

output.

6. Radiative rates are given in the standard output. The static radiative rate refers to
3
4∆Eif 2
3c3 |µif | , whereas the vibrational radiative rate is the integral over the fluorescence
spectrum.
R
7. The frequency-integrated absorption coefficient [309] ( σabs (ω)dω) is given in the
standard output.
378 CHAPTER 21. VIBRONIC SPECTRA
Chapter 22

Keywords in the control file

22.1 Introduction

The file control is the input file for TURBOMOLE which directly or by cross references provides
the information necessary for all kinds of runs and tasks. control is usually generated by
define, the input generator. This chapter provides a short-hand documentation: a list of
the most important key words, the possible parameters for each keyword, default values,
and a brief explanation.

22.2 Format of Keywords and Comments

TURBOMOLE input is keyword-directed. Keywords start with a ’$’, e.g. $title. Comments
may be given after $dummy, or by a line starting with #; these lines are ignored by TURBOMOLE.
Blank lines are also ignored. Keywords may be in any order unless stated otherwise below.
The sample inputs given below should help to give an idea how the keywords are to be used.
Complete control files are provided in Chapter 23. An alphabetical list of all keywords is
given in the index.
The control file and other input files references therein must end with this keyword:
$end

22.2.1 Keywords for System Specification

General information defining the molecular system: nuclear coordinates, symmetry, basis
functions, number of occupied MOs, etc. which are required by every module.

379
380 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$title
give title of run or project here.

$symmetry d4h
Schönflies symbol of the point group. All point groups are supported with the ex-
ception of NMR shielding and force constant calculations etc. which do not work for
groups with complex irreps (C3 , C3 h, T , etc). Use a lower symmetry group in this
case.

$atoms
Example:

$atoms
cu 1-4 \
basis =cu ecp-18 arep \
jbas =cu ecp-18 \
ecp =cu ecp-18 arep
se 5-6 \
basis =se ecp-28 arep dzp \
jbas =se ecp-28 \
ecp =se arep

note the backslash \ : this is necessary. For each type of atom, one has to specify

- the basis set

- and the auxiliary (fitting) basis for RIDFT calculations
- the ECP if this is used.

The files basis, ecp and jbas must provide the necessary information under the
labels specified in $atoms.

$pople char
This data group specifies the number of Cartesian components of basis functions (i.e.
5d and 7f in AO-Basis, 6d and 10f in CAO-Basis) for which the SCF calculation
should be performed. Possible values for char are AO (default) or CAO. If CAO is
used—which is not recommended—a core guess must be used instead of a Hückel
guess (see $scfmo).

RHF

$closed shells
Specification of MO occupation for RHF, e.g.

a1g 1-4 ( 2 )
a2g 1 ( 2 )
22.2. FORMAT OF KEYWORDS AND COMMENTS 381

$open shells type=1

MO occupation of open shells and number of open shells. type=1 here means that
there is only a single open shell consisting e.g. of two MOs:

b2g 1 ( 1 )
b3g 1 ( 1 )

$roothaan 1
a = 1 b = 2

$roothaan
Roothaan parameters for the open shell, here a triplet case. define recognizes most
cases and suggests good Roothaan parameters.
For further information on ROHF calculations, see the sample input in Section 23.6
and the tables of Roothaan parameters in Section 6.3.

UHF

$uhf directs the program to carry out a UHF run,e.g.

$alpha shells
a1g 1-4 ( 1 )
a2g 1 ( 1 )
$beta shells
a1g 1-4 ( 1 )
a2g 1 ( 1 )

The specification of MO occupation for UHF, $uhf overwrites closed-shell occupation

specification.

22.2.2 Keyword for the General Memory Specification

Most post-SCF programs (aoforce, ccsdf12, egrad, escf, evib, mpgrad, mpshift, pnoccsd,
ricc2, rirpa) in TURBOMOLE read the data group
$maxcor 500 mib per_core
to control the amount of dynamically allocated memory. The integer number (above “500”)
one can (optionally) give a SI prefix (kb, mb, gb) or an IEC prefix (kib, mib, gib) to specify
the unit. If no unit is specified mib (binary megabytes, i.e. 10242 bytes) are used.
In addition one can specify a reference for parallel calculations:

per_proc indicates that the memory is specified per process

per_node indicates that the memory is specified per computer node, i.e. shared by all
processes of the calculation that run on the machine with the same host name
382 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

per_core indicates that the memory is specified per core, i.e. per thread

total means the specified memory should be divided by all processes and nodes

If not given the specified memory will be used per thread. Note that in release versions before
V7.2 the memory was specified per process. For sequential calculations this sub-option has
no effect.
If $maxcor is not given 500 MiB per thread will be used as limit for dynamically allocated
memory.
Note:

• $maxcor defines only the memory controled by the electronic structure code. Addi-
tional memory can be allocated by the math and MPI libraries linked into the program
and by the operating and I/O systems. It is therefore recommended to set $maxcor
not higher than to 75% – 85% of the physical core memory that is available for the
calculation.

• Some programs use additional keywords ($incore, $ricore, $ricore_slave, $rpacor)

to specify limits for dynamically allocated memory for certain tasks.

22.2.3 Other General Keywords

$operating system unix

$path
$lock off
$suspend off

The four keywords above are set by define, but are not necessary.
$statistics dscf
or
$statistics mpgrad
Only a statistics run will be performed to determine file space requirements as specified for
dscf or mpgrad. On return the statistics option will be changed to $statistics off.
$actual step dscf
means current step. Keyword and data group (as e.g. dscf) is set by every program and
removed on successful completion.
$last step relax
Keyword and data group (as e.g. relax) set by every program on successful completion.
General file cross-references:

$coord file=coord
$intdef file=coord
22.2. FORMAT OF KEYWORDS AND COMMENTS 383

$user-defined bonds file=coord

$basis file=basis
$ecp file=basis
$jbas file=auxbasis
$scfmo file=mos
$uhfmo_alpha file=alpha
$uhfmo_beta file=beta
$natural orbitals file=natural
$natural orbital occupation file=natural
$energy file=energy
$grad file=gradient
$forceapprox file=forceapprox

It is convenient not to include all input in the control file directly and to refer instead to
other files providing the corresponding information. The above cross references are default
settings from define; you may use other file names. define will create most of these files.
Examples of these files are given below in the samples.

$coord (and $intdef and $userdefined bonds)

contains atom specification—type and location—and the bonds and internal coordi-
nates convenient for geometry optimizations.
$basis
specification of basis sets.
$ecp specification of effective core potentials.
$jbas
auxiliary (fitting) basis for the Coulomb terms in ridft.
$scfmo, $uhfmo_alpha, $uhfmo_beta
MO vectors of SCF or DFT calculations for RHF or UHF runs.
$natural orbitals, $natural orbital occupation
keywords and data groups set by unrestricted dscf or ridft runs. Contain natural
MO vector and orbital occupation.
$energy, $grad
energies and gradients of all runs, e.g. for documentation in a geometry optimizations.
$forceapprox
approximate force constant for geometry optimizations.

22.2.4 Keywords for redundant internal coordinates in $redund_inp

With the parameters in $redund_inp the generation of redundant internal coordinates can
be modified. All entries have to be made in the control file before invoking the ired option.
Important options are:
384 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

iprint n
print parameter for debug output: The larger n is, the more output is printed
n ≥ 0, n ≤ 5 (default: 0)

metric n
method for generating and processing of redundant internal coordinates
n ≥ −3, n ≤ 3, n 6= 0 (default: 3)

Values for the metric option:

n = 1 “Delocalized Coordinates”
The BmBt matrix is diagonalized for the complete set of redundant internal
coordinates, matrix m is a unit matrix.
n = -3 Delocalized Coordinates obtained with a modified matrix m, the values of
m can be defined by user input (see below).
n = -1 “Hybrid Coordinates”
Natural internal coordinates are defined as in the old iaut option. If a cage
remains, delocalized coordinates (as for n=1) are defined for the cage.
n = -2 Very similar to the n = 1 option, but for the remaining cage delocalized
coordinates with modified matrix m are defined as for n = −3.
n = 2 “Decoupled coordinates”
The redundant coordinates are divided into a sequence of blocks. These are
expected to have decreasing average force constants, i.e. stretches, angle
coordinates, torsions and “weak” coordinates. The BBt matrix is
diagonalized for each block separately after the columns of B were orthog-
onalized against the columns of B of the the preceding blocks.
n = 3 “Generalized natural coordinates”
Natural internal coordinates are defined first, for the remaining cage decou-
pled coordinates are defined.

type r
a positive real number, which is an approximate “force constant”, can be read in for
each type of coordinate (see below). The force constants are used for the definition
of the matrix m in BmBt .

Types of internal coordinates for the definition of m

The matrix m is assumed to be a diagonal matrix. For each type of coordinate a different
value for the force constants mii can be read in. Types of coordinates are:

stre bond stretch (default: 0.5)

invr inverse bond stretch (default: 0.5)

22.2. FORMAT OF KEYWORDS AND COMMENTS 385

bend bond angle (default: 0.2)

outp Out of plane angle (default: 0.2)

tors dihedral or “torsional” angle (default: 0.2)

linc Special angle coordinate for collinear chains, bending of the chain a–b–c in the
plane of b–c–d (default: 0.2)

linp bending of the chain a–b–c perpendicular to the plane of b–c–d

(default: 0.2)

wstr stretch of a “weak” bond, i.e. the bond is assumed to have a very low force constant,
e.g. a “hydrogen bond” or a “van der Waals bond”
(default: 0.05)

winv inverse stretch of a weak bond (default: 0.05)

wbnd bond angle involving at least one weak bond (default: 0.02)

wout Out of plane angle for weak bonds (default: 0.02)

wtor dihedral angle for weak bonds (default: 0.02)

wlnc linc coordinate for weak bonds (default: 0.02)

wlnp linp coordinate for weak bonds (default: 0.02)

22.2.5 Keywords for Module uff

One has to specify only the Cartesian coordinates (data group $coord) to start a uff
run. The program uff requires the data groups $uff, $ufftopology, $uffgradient and
$uffhessian. If these keywords do not exist in the control file the program will generate
these data groups.
The data group $uff contains the parameters described below. The default values—in the
control file—are:

1 1 0 ! maxcycle,modus,nqeq
111111 ! iterm
0.10D-07 0.10D-04 ! econv,gconv
0.00 1.10 ! qtot,dfac
0.10D+03 0.10D-04 0.30 ! epssteep,epssearch,dqmax
25 0.10 0.00 ! mxls,dhls,ahls
1.00 0.00 0.00 ! alpha,beta,gamma
F F F ! transform,lnumhess,lmd

The explanation of the variables are as follows:

maxcycle
number of max. optimization cycles (maxcycle=1: single-point calculation).
386 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

modus
can have the values +1 or -1. If modus = -1 only the topology will be calculated.

nqeq each nqeq cycle the partial charges will be calculated. If nqeq = 0, then the partial
charges are calculated only in the first cycle, if the file ufftopology does not exist.

iterm
switch for the different types of force field terms:

100000 bond terms will be calculated.

010000 angle terms will be calculated.
001000 torsion terms will be calculated.
000100 inversion terms will be calculated.
000010 non bonded van der Waals terms will be calculated.
000001 non bonded electrostatic terms will be calculated.

econv, gconv
convergence criteria for energy and gradient.

qtot total charge of the molecule.

dfac distance parameter to calculate the topology. If the distance between the atoms I
and J is less than the sum of the covalent radii of the the atoms multiplied with dfac,
then there is a bond between I and J.

epssteep
if the norm of the gradient is greater than epssteep, a deepest-descent-step will be
done.

epssearch
if the norm of the gradient is smaller than epssearch, no line-search step will be done
after the Newton step.

dqmax
max. displacement in a.u. for a coordinate in a relax step.

mxls, dhls, ahls

parameters of linesearch:

ahls start value

dhls increment
mxls number of energy calculations

alpha, beta, gamma

modification parameter for the eigenvalues of the Hessian (see below): f (x) = x ∗
(alpha + beta ∗ exp(−gamma ∗ x)).

transform
a switch for the transformation in the principal axis system.
22.2. FORMAT OF KEYWORDS AND COMMENTS 387

lnumhess
a switch for the numerical Hessian.

lmd a switch for a MD calculation.

Input Data Blocks Needed by Uff

$coord
cartesian coordinates of the atoms (default: $coord file=coord)

$ufftopology
contains a list of the next neighbours of each atom (see Section 22.2.5). Sometimes it is
useful to enter the connectivity (in the input block nxtnei12 in the file ufftopology)
by hand (not always necessary; default: $ufftopology file=ufftopology).

Beyond this uff reads the force field parameters for the atoms from the file parms.in. If
this file exists in the directory from which one starts an uff calculation the program will use
this file, if not the program reads the data from the file $TURBODIR/uff/parms.in. If one
wants own atom types, one has to add these atoms types in the file parms.in. For each new
atom type one has to specify the natural bond distance, the natural bond angle, the natural
non-bond distance, the well depth of the Lennard-Jones potential, the scaling factor ζ, the
effective charge, torsional barriers invoking a pair of sp3 atoms, torsional barriers involving
a pair of sp2 atoms, generalized Mulliken–Pauling electronegativities, the idem potentials,
characteristic atomic size, lower bound of the partial charge, upper bound of the partial
charge. Distances, energies and charges are in atomic units and angles are in rad.

Uff Output Data Blocks

$coord
contains the (updated) Cartesian coordinates of the atoms (default: $coord file=coord).

$ufftopology
contains the full information of the topology of the molecule and the whole force field
terms (see below; default: $ufftopology file=ufftopology).

$uffgradient
contains the accumulated Cartesian analytical gradients (default: $uffgradient file=uffgradient).

$uffhessian
contains the Cartesian analytical Hessian;
(default: $uffhessian file=uffhessian0-0).

The file ufftopology

The topology file ufftopology contains the blocks nxtnei12, nxtenei13, nxtnei14, connec-
tivity, angle, torsion, inversion, nonbond and qpartial. It starts with $ufftopology and
388 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

ends with $end. The first three blocks (nxtnei12, nxtnei13, nxtnei14) have the same form:
they start with the atom number and the number of its neighbours, in the next line are
the numbers of the neighbour atoms. Then the connectivity-block follows starting with the
number of bond terms. Each line contains one bond term:

I J d BO.

Here are I and J the number of the atoms, d the distance in a.u. and BO is the bond order.
The angle terms follow, starting with the number of the angle terms. In each line is one
angle term:
J I K wtyp θ nrJI nrIK .

Here are J, I and K the atoms number, where atom I is in the apex. “wtyp” is the angle
type and has the following values:

wtyp = 1 linear case

wtyp = 2 trigonal planar case

wtyp = 3 quadratic planar case

wtyp = 6 octahedral case

wtyp = 9 all other cases.

θ is the angle value in degree. nrJI and nrIK are the number of the bonds between J and
I and the bond between I and K. The hybridization of atom I determines “wtyp”.
Then the torsion terms follow, starting with the number of the torsion terms. Each line
contains one torsion term:

I J K L nrJK ttyp φ θIJK θJKL .

Here are I, J, K and L the atom numbers. nrJK is the number of the bond between J and
K. “ttyp” is the torsion type:

ttyp = 1 J (sp3 )–K (sp3 )

ttyp = 11 like ttyp=1, but one or both atoms are in Group 16

ttyp = 2 J (sp2 )–K (sp3 ) or vice versa

ttyp = 21 like ttyp=2, but one or both atoms are in Group 16

ttyp = 22 like ttyp=2, but J or K is next a sp2 atom

ttyp = 3 J (sp2 )–K (sp2 )

ttyp = 9 all other cases.

φ is the value of the torsion angle in degree. θIJK is the angle value of (I − J − K) and
θJKL is the cwone for J − K − L. The hybridizations of J and K determine “ttyp”.
22.2. FORMAT OF KEYWORDS AND COMMENTS 389

The inversion terms follow starting with the number of inversion terms (e.g. the pyramidal-
isation of NH3 ). In each line is one inversion term:

I J K L ityp1 ityp2 ityp3 ω1 ω2 ω3 .

I, J, K and L are the atom numbers. Atom I is the central one. ityp1, ityp2, ityp3 are the
types of the inversions:

ityp = 10 atom I is C and atom L is O

ityp = 11 like ityp=10, but L is any atom

ityp = 2 I is P

ityp = 3 I is As

ityp = 4 I is Sb

ityp = 5 I is Bi

ityp = 9 all other cases.

ω1 , ω2 and ω3 are the values of the inversion angles in degree.

The nonbond terms follow starting with the number of the non-bonded terms. In each line
is one nonbond term:
I J d.

Here I and J are the atom numbers, d the distance in a.u. Then the partial charges follow.
If the determination of the molecule connectivity failed, you can specify the block nxtnei12
in the file ufftopology. Then the program calculates the other blocks.
Based on the numbers of the next neighbours (block nxtnei12 in the file ufftopology) the
program tries to determine the UFF type of an atom. The following rules are implemented: If
the atom has three next neighbours and it is in the nitrogen group, then it has a hybridization
three. If it is not in the nitrogen group, it has hybridization two. If the atom has four next
neighbours and it is in the carbon group, it has hybridization three. If it is not in the carbon
group, it becomes hybridization four. If the number of next neighbours is six, then it gets
the hybridization six.
Since the smallest eigenvalues λi of the Hessian has the greatest influence on the convergence
of the geometry optimization, one can shift these values with

λ̃i = λi · α + β · e−γx

and calculates a new Hessian with these modified eigenvalues.

22.2.6 Keywords for tb

Module tb which performs GFN-xTB or GFN2-xTB extended tight binding calculations

reads in the keyword $tb
390 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$tb
charge <number>
gfn <method-number>
accuracy <number>
etemp <number>
broydamp <number>
maxiter <number>

The options are:

charge -1
sets the molecular total charge. ...,-2, -1, 0, 1, 2,... If not specified, a neutral
input is assumed: charge 0

gfn 2
choose method, GFN-xTB (1) or GFN2-xTB (2), default is 2

accuracy 1.0
accuracy for SCC calculations, lower value means higher accuracy. Default
is 1.0

etemp 300
electronic temperature in Kelvin, default is 300K

broydamp 0.4
Broyden damping for charge mixing, default is 0.4

maxiter 250
maximum number of SCC iterations, default is 250

22.2.7 Keywords for woelfling

Module Woelfling reads options from data group

$woelfling

The below values of the options are default values with the following meaning:

ninter 14

Number of interpolated structures for optimization.

ncoord 2

Number of input structures provided by user.

align 0
22.2. FORMAT OF KEYWORDS AND COMMENTS 391

Align input structures by translation/rotation 0=yes, 1=no.

maxit 40

Maximum number of iterations.

dlst 3.00000000000000

Threshold for accuracy of LST-interpolation.

thr 1.000000000000000E-004

Threshold for mean of norms of projected gradients.

method q

Use standard optimization from initial LST-path (method q) or grow reaction path (method
qg).
Furthermore,

riter 0

counts the number of completed iteration (no option).

22.2.8 Keywords for Modules dscf and ridft

$denconv real
Convergency criterion for the root mean square of the density matrix. If you want to
calculate an analytical MP2 gradient (program mpgrad) real should be 1.d-7 or less.

$dft options
Listing of all possible sub-keywords for $dft (cross-references are given).
The user normally has to choose only the functional and the grid size, see below. All
other parameters have proven defaults.

functional name
Specification of the functional, default is BP86, printed as functional b-p.
For all possible—and useful—functionals, please refer to page 406 and for def-
inition of the functionals the section 6.2 on page 155.
Example (default input):

$dft
functional b-p
392 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

gridsize integer or minteger

Specification of the spherical grid (see section 22.2.8 on page 406). Default is
gridsize m3.
Example:

$dft
gridsize m3

gridtype integer —not recommended for use—

Specification of the mapping of the radial grid.
Possible values for gridtype are 1, . . . , 6, but gridtype 4 to 6 is only for the
use with functional lhf (see page 410). For the definition of gridtype 1 – 3,
please refer to Eq. (16), (17) and, (19) in Ref. [310]. In relativistic all-electron
calculations grids with an increased number of radial points are suggested.
These can be selected by appending the character a after the gridsize, e.g.
gridsize 4a.
Example (default value):

$dft
gridtype 3

debug integer
Flag for debugging. debug 0 means no debug output (default), debug 1 means
some output, debug 2 means a lot more output. Be careful!
nkk integer
Specification of the sharpness of the partition function as proposed by Becke
[311], default is nkk 3. The usage of nkk makes sense only in the range 1 ≤
nkk ≤ 6.
Example:

$dft
nkk 3
ntheta integer
—not recommended for use—
nphi integer
Only for user-specified Lobatto grids, i.e. gridsize 9, ntheta specifies the
number of θ points and nphi specifies the number of φ points. For the fixed
Lobatto grid, i.e. gridtype 8, the default value is ntheta 25 and nphi 48.
When gridsize 9 is given, you have to specify both, ntheta and nphi (see be-
low), otherwise the program will crash. The restriction for user-defined Lobatto
grids is the number of grid points, which must not exceed 2000 grid points.
Example:

$dft
gridsize 9
ntheta 30
nphi 60
22.2. FORMAT OF KEYWORDS AND COMMENTS 393

old_RbCs_xi
Original grids had not been carefully optimized for all atoms individually. This
has now been done, which let to changes of ξ for Rb and Cs only resulting in mi-
nor improvements. If you have ongoing projects, which have been started with
the old grids, you should continue using them with the keyword old_RbCs_xi.
Example:

$dft
old_RbCs_xi

radsize integer
Specifies the number of radial grid points. Default values depend on type of
atom and grid (see keyword gridsize). The formula for the radial gridsize is
given as,

number of radial grid points = ioffrad + (radsize − 1) ∗ 5 .

ioffrad is atom-dependent, the more shells of electrons, the larger ioffrad:

elements ioffrad elements ioffrad

for H–He 20 for K–Kr 40
for Li–Ne 25 for Rb–Xe 45
for Na–Ar 30 for Cs–Lw 50

The radial grid size increases further for finer grids:

gridsize 1 2 3 4 5 6 7 8 9
radsize 1 2 3 6 8 10 14 9 3

If you want to converge results with respect to radial grid size, increase radsize
in steps of 5, which is convenient (see equation above).

diffuse integer
Serves to increase quadrature grids; this is recommended in case of very diffuse
wavefunctions. With the keyword diffuse grids are modified by changing the
linear scaling factor ξ of radial grid points and the radial gridsize:

radsize =⇒ radsize + incr

ξ =⇒ ξ * scal

diffuse integer 1 2 3 4 5 6
incr 1 2 3 4 5 6
scal 1.5 2.0 2.8 4.0 5.0 6.0

For information about the linear scaling parameter ξ, see Eq. (16)–(19) and
Table 1 in Ref. [310].
394 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

In addition, the reduction of spherical grid points near nuclei is suppressed, i.e.
fullshell on is set (see page 395).
Note: the keyword radsize integer overrules the setting of incr ; for more
information, see p. 393.

Recommendation: For diffuse cases use gridsize m4 (or larger) in combination

with diffuse 2 and check the number of electrons; for more difficult cases use
diffuse 4. In case of doubt, verify the calculation with a larger grid, i.e.
gridsize 7.
The test suite example $TURBODIR/TURBOTEST/dscf/short/H3PO4.DSCF.DIFFUSE
provides an example of usage; this also gives reasonable values for damping and
orbitalshift to reach convergence in this and similar cases, see $scfdamp and
$scforbitalshift (p. 400 and p. 403).
Example (Recommendation):
$dft
gridsize m4
diffuse 2
rhostart integer
—for developers only—
rhostop integer
Radial grid points have a linear scaling parameter ξ, see Eq.(16)–(19) and Table
1 in Ref. [310]. With the following input,
$dft
rhostart 50
rhostop 200
one performs a numerical integration for the density and the exchange corre-
lation term for ξ = 0.5, (0.01), 2.0 for given MOs and functional. NOTE: only
molecules with a single atom type can be used. The results serve to establish
stable, optimal ξ values, see Figure 1 in Ref. [310]. Program stops after this
testing.
reference
Usage of the reference grid, which is a very fine grid with very tight thresholds.
The default values for the different variables are:
gridsize 7
radsize 14
fullshell 1
dgrenze 16
fgrenze 16
qgrenze 16
fcut 16
Please refer to the corresponding sub-keywords for explanation.
If you want to use the reference grid, you have to skip the keyword gridsize,
and type instead reference. Example:
22.2. FORMAT OF KEYWORDS AND COMMENTS 395

$dft
functional b-p
reference

test-integ
Check if the selected grid is accurate enough for the employed basis-set by
performing a numerical integration of the norm of all (occupied and virtual)
orbitals. Useful for LHF. 410.
batchsize integer
Grid points are sorted into batches, which are then processed. This increases
efficency. This should be changed only by developers. Default is batchsize
100.
fullshell
Standard grids have reduced number of spherical grid points near nuclei. With
the keyword fullshell this reduction is suppressed. Reference grid (see key-
word reference) always has full spherical grids with 1202 points. Should be
used to checked the influence of spherical grid reduction.
Example for the usage of fullshell:

$dft
functional b-p
gridsize m4
fullshell

symblock1 real
—for developers only—
symblock2 real
Values of real effects efficiency of the quadrature, default is symblock1 0.001
and symblock2 0.001, one can try higher or smaller values.
xparameter integer —not recommended for use—
Where xparameter (default) can be: sgrenze (8), fgrenze (10), qgrenze (12),
dgrenze (12) and, fcut (14). These parameters control neglect of near zeros of
various quantities. With xparameter integer one changes the default. integer
larger than defaults will increase the numerical accuracy. Tighter threshold are
set automatically with keyword $scfconv (see section 22.2.8 on page 399).
weight derivatives
Includes the derivatives of quadrature weights to get more accurate results.
Default is that the derivatives of quadrature weights will be not considered, see
section 22.2.13 on page 430.
gridordering
Grid points are ordered into batches of neighbouring points. This increases
efficiency, since now zeros can be skipped for entire batches. gridordering
is default for serial version, not for the parallel one. You cannot use weight
derivatives and gridordering together.
Example for switching off gridordering:
396 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$dft
gridordering 0
s-junc integer
p-junc integer
Within the semi-numerical integration scheme employed for local hybrid func-
tionals, S- and P-junctions feature pre-screenings of the analytical two-center
integrals with respect to shell pair overlap and shell pair conjunction through
the density matrix, respectively. This allows the neglection of certain shell pairs
and thus accelerates the calculation. s-junc and p-junc set the corresponding
thresholds to 10−integer . Smaller values increase the number of neglected shell
pairs and thus reduce the calculation cost, but also decrease the accuracy of the
results. Note that for gradient and scf calculations the P-junctions also depend
on the grid weights. If larger grids are employed seeking more accurate results,
the threshold for P-junctions should thus be adjusted (increasing the value
for p-junc 6). The defaults settings are s-junc 6 and p-junc 6 for grad,
rdgrad, and ridft. Too small values for s-junc and p-junc might result in
non-converging calculations or deteriorated results. This holds particularly for
excited state calculations. For escf the default values are therefore s-junc 8
and p-junc 8.

$electrostatic field
Specification of external electrostatic field(s). The specification may take place either
by Ex, Ey, Ez or by x, y, z, |E|. See also $fldopt.
Example:

$electrostatic field
0.1000E-03 0.000 0.000

Electrostatic fields are also possible for 2c calculations, turning on geofield is manda-
tory in these calculations. $pcc activates picture change correction for fields integrals
in 1c and 2c (local) BSS/DKH/X2C calculations.

$fermi tmstrt=<300.0> tmend=<100.0> tmfac=<0.9> hlcrt=<1.0E-01> stop=<1.0E-03> nue=<N>

Requests calculation of occupation numbers at a finite temperature T . For an or-

bital with the energy εi the occupation number ni ∈ [0, 1] is calculated as
 
1 εi − µ
ni = erfc ,
2 fT
√
where µ is the Fermi level. The factor f = 4k/ π is chosen to yield the same slope
at the Fermi level as the Fermi distribution.
Calculation of the fractional occupation numbers starts when the current HOMO-
LUMO gap drops below the value given by hlcrit (default: 0.1). The initial tem-
perature given by tmstrt (default: 300 K) is reduced at each SCF cycle by the factor
tmfac (default: 1.0) until it reaches the value tmend (default: 300 K). Note that the
22.2. FORMAT OF KEYWORDS AND COMMENTS 397

default values lead to occupation numbers calculated at a constant T = 300 K. Cur-

rent occupation numbers are frozen if the energy change drops below the value given
by stop (default: 1·10−3 ). This prevents oscillations at the end of the SCF procedure.
Calculation of fractional occupation numbers often requires much higher damping and
orbital shifting. Please adjust the values for $scfdamp and $scforbitalshift if you
encounter convergence problems.
In UHF runs this option can be used to automatically locate the lowest spin state.
In order to obtain integer occupation numbers tmend should be set to relatively low
value, e.g. 50 K.
Calculation of fractional occupation numbers should be used only for single point
calculations. When used during structure optimizations it may lead to energy oscil-
lations.
The optional nue value (number of unpaired electrons) can be used to force a certain
multiplicity in case of an unrestricted calculation. nue=0 is for singlet, nue=1 for
dublet, etc.
The option addTS adds the entropic contribution of the smearing to the total energy
which is important for very high temperatures only.
Finally the option noerf will use the full correct Fermi statistics rather than the term
given above.
$firstorder
Perform first-order SCF-calculation, i.e. perform only one SCF-iteration with the start
MOs (which should be the orthogonalized MOs of two independent subsystems as is
explained in detail in Chapter 19).
$fldopt options
Specification of options related with external electrostatic fields. The following options
are available:

1st derivative on/off

Calculate numerical 1st derivative of SCF energy with respect to electrostatic
field (default: off), increment for numerical differentiation is edelt (see below).
2nd derivative on/off
Calculate numerical 2nd derivative of SCF energy with respect to electrostatic
field (default: off), increment for numerical differentiation is edelt.
edelt= real
Increment for numerical differentiation (default: 0.005).
fields on/off
Calculate SCF energy for non-zero external electrostatic
fields defined in $electrostatic field.
geofield on/off
Calculate SCF energy for one external field definition and dump disturbed
MOs onto $scfmo. This enables to evaluate properties or perform geometry
optimizations in the presence of an external field.
398 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

Caution: don’t use the RI approximation for all these calculations since this will
lead to non-negligible errors!!

$incore integer
By using this option the two-electron integrals are kept in RAM; integer specifies
how many megabytes should be allocated. If the integrals exceed the RAM allocated
the program reverts to the standard mode. Supports all methods which process two-
electron integrals, i.e. SCF and DFT (including hybrid functionals); RHF and UHF.
The following condition must be met:
$scfdenapproxl 1
and rhfshells 1 or 2. It is advisable to set $thize as small as possible (e.g. $thize
0.1d-08) and to remove the keyword $scfdump.

Note: this keyword does not work for parallel runs.

$mo-diagram only nirreps=integer

If this keyword is set the energies and symmetry labels of all occupied MOs will be
dumped to this data group. This may be helpful to draw mo-diagrams. If only has
been set only the start MOs are dumped and the program quits.
nirreps will hold the total number of displayed orbitals after the successful run.

$mom
This keyword enables the use of the maximum overlap method in unrestricted ridft
calculations to access excited states self-consistently. [312]

$moprint
If this keyword is present all occupied orbitals are dumped to standard output. Be
careful about this option as it can create huge output files in case of many basis
functions.

$mo output format format

If this line is present, the dscf program is forced to output the MOs using the new
FORTRAN format format regardless of the format-option in data group $scfmo.
Otherwise the input format will be used.
Example: $mo output format(3(2x,d15.8))

$natural orbitals
This data group will be written after an UHF calculation (together with $natural
orbital occupation) and contains the natural space orbitals (same syntax as $scfmo).

$natural orbital occupation

This data group will be written after an UHF calculation (together with $natural
orbitals) and contains the occupation of natural orbitals (syntax as any data group
related with orbital occupation information, e.g. $closed shells), e.g.
22.2. FORMAT OF KEYWORDS AND COMMENTS 399

a 1-5 ( 2.00000000000000 )
a 6 ( 1.99949836999366 )
a 7 ( 1.99687490286069 )
a 8 ( 1.00000000000000 )
a 9 ( .00312509713931 )
a 10 ( .00050163000634 )

$prediag
concerns the first SCF iteration cycle if start MOs from an EHT guess are used.
The SCF iteration procedure requires control mechanisms to ensure (fast) conver-
gence, in TURBOMOLE these are based on orbital energies i of the preceeding iteration
used for level shifting and damping (besides DIIS, see below). This feature cannot be
used in the first iteration if EHT MOs are employed as start, since i are not avail-
able. The keyword $prediag provides ’i of the zeroth iteration’ by diagonalization
of occ–occ and virt–virt part of the first Fock matrix, to allow for level shifting etc..
See $scfdiis below.
$restart dscf twoint
Try a dscf restart. The program will read the data group $restartd (which must
exist, also $scfmo has to exist!) and continue the calculation at the point where it
ended before. If the additional option twoint is appended, the program will read the
two-electron integrals from the files specified in $scfintunit, so there will be almost
no loss of cpu-time.
All this information is normally provided by the previous dscf run if the keyword
$scfdump (see there) was given.
$restartd data
Data provided by a previous dscf run that has been interrupted. This keyword is
created when $scfdump was given.
$rundimensions data
is set by define so usually no changes are necessary. The dimensions must be greater
or equal to those actually required, i.e. you can delete basis functions and keep rundi-
mensions. This keyword is not necessary for small cases.
Example:

dim(fock,dens)=6072
natoms=6
nshell=34
nbf(CAO)=108
nbf(AO)=98
dim(trafo[SAO<-->AO/CAO])=256
rhfshells=1

$scfconv integer
SCF convergency criterion will be 10−integer for the energy. Gradients will only be
evaluated if integer > 6.
400 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$scfdamp start=<.500> step=<.050> min=<.100>

Damping parameters for SCF iterations in order to reduce oscillations. The old
Fock-operator is added to the current one with weight 0.5 as start; if convergence is
good, this weight is then reduced by the step 0.05 in each successive iteration until the
minimum of 0.1 is reached. (These are the default settings of define for closed-shell
RHF). DSCF automatically tries to adjust the weight to optimize convergence but in
difficult cases it is recommended to start with a large weight, e.g. 1.5, and to set the
minimum to a larger value, e.g. 0.5.

$scfdebug options
Flags for debugging purposes. Following options are available:

vectors integer
Output level concerning molecular orbitals. integer =0 (default) means minimal
output, >1 will output all start MOs and all MOs in each iteration.
density integer
Output level concerning difference density matrices.
debug integer
integer > 0 will dump a lot of information—be careful!

$scfdenapproxl integer
Direct SCF procedures build the Fock matrix by exploiting information from previous
iterations for better efficiency. With this keyword information from the last integer
iterations will be used. This feature is switched on with the default value 20, even
if the keyword is absent. The user may reduce the number of iterations employed to
smaller values (e. g. 10) in cases were numerical stability could become an issue. With
the value 0 this feature is switched off; the Fock matrix is constructed from scratch
in each iteration.

$scfdiis options
Control block for convergence acceleration via Pulay’s DIIS ∗ .
Options are:

errvec=char specifies the kind of error vector to be used (two different kind of DIIS
algorithms)
char=’FDS’ or ’SDF’ or ’FDS-SDF’
uses (FDS − SDF) as error vector.
char= none
no DIIS
char= sFDs
use S −1/2 F DS 1/2 − transposed

Further suboptions:
∗
P.Pulay; Chem. Phys. Lett., 73, 393 (1980), P.Pulay; J. Comput. Chem., 4, 556 (1982)
22.2. FORMAT OF KEYWORDS AND COMMENTS 401

maxiter=integer
maximum number of iterations used for extrapolation.
debug=integer
debug level (default: 0)
integer=1 print applied DIIS coefficients
integer=2 print DIIS matrix and eigenvalues, too
qscal=real
scaling factor in DIIS procedure: qscal > 1 implies some damping,
qscal = 1.0: straight DIIS.
thrd=real
directs the reduction of qscal to qscal = 1.0 (no damping in DIIS), done
if ||errvec|| < thrd.

Defaults for $prediag (see above) and $scfdiis

errvec=FDS-SDF, maxiter=5, qscal=1.2, thrd=0.0, this implies DIIS damping in all
iterations, prediag is switched of.
Recommended:
errvec=sFDs leads to the following defaults:
qscal=1.2, for SCF runs: maxiter=6 and thrd=0.3, prediag is off; for DFT runs:
maxiter=5 and thrd=0.1 prediag is on. If you want to switch off prediag put
$prediag none.

$scfdump
Dump SCF restart information onto data group $restartd and dump SCF MOs in
each iteration onto $scfmo (scfdump = iter). Additionally, a data block $scfiterinfo
will be dumped containing accumulated SCF total-, one- and two-electron energies of
all previous SCF iterations. Information that will allow you to perform a restart if
your calculation aborts will be dumped on data group $restartd (see also $restart).

$scfintunit options
Disc space specification for two-electron integrals. The following suboptions are avail-
able (and necessary):

unit=integer
A legacy suboption which is ignored.
size=integer
Filespace in megabytes for this file. size=0 leads to a fully direct run. size is
set by a statistics run, see $statistics. DSCF switches to direct mode if the
file space is exhausted.
file=char
Filename. This may also be a complete path name, if you want to store the
integrals in a special directory. Make sure the file is local, otherwise integrals
are transmitted over the network.
402 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

Thus your data group $scfintunit may look like this:

$scfintunit
unit=30 size=35 file=twoint1
unit=31 size=35 file=/users/work/twoint2

Maximal 30 files may be specified in this way.

$scfiterlimit integer
Maximum number of SCF iterations (default: 30).

$scfmo none file=char

Input/output data group for SCF MOs. You can specify

none
To perform a calculation without a start vector (i.e. use a core Hamiltonian
guess).
file=char
The file where the MOs are written on output (default: mos).

These two options can also be used for $uhfmo_alpha and $uhfmo_beta to use a core
guess and write the molecular orbitals to file.
After running define or a TURBOMOLE calculation additional options may appear
specifying the origin of the MOs:

expanded
These MOs were obtained by projection form another basis set. They should
not be used for wavefunction analysis.
scfconv=integer
The MOs are converged SCF MOs , the convergence criterion applied was
10−integer
scfdump=integer
The MOs are unconverged SCF MOs which were written on this data group
after iteration integer. The latter three options are mutually exclusive.
format(format string)
This specifies the FORTRAN format specification which was used for MO out-
put. The standard format is (4d20.14). (See data group $mo output format.)

Example:
Your data group $scfmo could look like this after a successful TURBOMOLE run :

$scfmo scfconv=7 format(3(1x,d19.13))

1 a1 eigenvalue=-.524127 nsao=6
.1234567890123d+01 -.1234567890123d+00 .1234567890123d-01
.1234567890123d+01 -.1234567890123d+00
3 a2 eigenvalue=-.234810
...
22.2. FORMAT OF KEYWORDS AND COMMENTS 403

$scforbitalorder on/off
Order SCF MOs with respect to their energies (default: on)

$scforbitalshift options
To assist convergence, either the energies of unoccupied MOs can be shifted to higher
energies or, in open-shell cases, the energies of closed-shell MOs to lower energies. In
general a large shift may help to get better convergence.
Options are:

noautomatic
Automatic virtual shell shift switched off.
automatic real
Automatic virtual shell shift switched on; the energies of virtual orbitals will
be shifted if the HOMO-LUMO gap drops below real such that a gap of real is
sustained. This is the default setting if the keyword is missing with real=0.1.
closedshell=real
Option for open-shell cases. Closed shells are shifted to lower energies by real.
The default shift value is closedshell=0.4.
Note: Normally this will disable the automatic shift of energies of virtual
orbitals! To override this, you should append an exclamation mark to the
’automatic’ switch, i.e. specify ’automatic! real’.
individual
Set shifts for special occupied MOs. To use this option, start the line with
the symmetry label and the list of MOs within this symmetry and append the
desired shift in brackets as in the following example:

a1 1,2,4-6 (-.34)
b1 8 (+.3)

$scftol real
Integral evaluation threshold. Integrals smaller than real will not be evaluated. Note
that this threshold may affect accuracy and the convergence properties if it is chosen
too large. If $scftol is absent, a default value will be taken obtained from $scfconv
−(scf conv+1)
by real = 10 3·#bf (#bf = number of basis functions).

$scratch files
The scratch files allocated by dscf can be placed anywhere in your file systems in-
stead of the working directory by referencing their path names in this data group. All
possible scratch files are listed in the following example:
404 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$scratch files
dscf dens path1/file1
dscf fock path2/file2
dscf dfock path3/file3
dscf ddens path4/file4
dscf statistics path7/file7
dscf errvec path8/file8
dscf oldfock path9/file9
dscf oneint path10/file10

The first column specifies the program type (dscf stands for SCF energy calculations,
i. e. the dscf program), the second column the scratch file needed by this program
and the third column the path name of the file to be used as scratch file.
$statistics options
The following options are allowed

off Do not perform integrals statistics

dscf Perform integrals statistics for dscf
mpgrad see mpgrad
dscf parallel see PARALLEL PROCESSING

Options dscf parallel, grad, mpgrad will be described in the related chapters.
If $statistics dscf has been given integral prescreening will be performed (which is
an n4 -step and may therefore be time-consuming) and a table of the number of stored
integrals as a function of the two parameters $thize and $thime will be dumped.
Afterwards, the files pace needed for the current combination of $thize and $thime
will be written to the data group ($scfintunit) and $statistics dscf will be
replaced by $statistics off.
$thime integer
Integral storage parameter, which is related to the time needed to calculate the inte-
gral. The larger integer the less integrals will be stored. The default value is integer
= 5. (see also $thize, $statistics)
$thize real
Integral storage parameter, that determines, together with $thime, the number of
integrals stored on disc. Only integrals larger than real will be stored. The default
value is real = 0.100E-04.

RHF/ROHF

$closed shells
Specification of MO occupation for RHF, e.g.

a1g 1-4 ( 2 )
a2g 1 ( 2 )
22.2. FORMAT OF KEYWORDS AND COMMENTS 405

$open shells type=1

MO occupation of open shells and number of open shells. ’type=1’ here means that
there is only a single open shell consisting e.g. of two MOs:

b2g 1 ( 1 )
b3g 1 ( 1 )

$rohf
This data group is necessary for ROHF calculations with more than one open shell.
Example:

$rohf 1
a -a a=0 b=0
h -h a=1 b=2
a -h a=1 b=2

This example is for the 7S state of chromium (3d5 4s1 ) in symmetry group I. Note that
for this option being activated, $roothaan also has to be specified in your control file,
although its parameter has no meaning in this case. For more details see Section 6.3.

$roothaan
For ROHF-calculations with only one open shell the Roothaan parameters† a and b
have to be specified within this data group (see also $rohf). Example:

$roothaan
a = 3/4 b = 3/2

This example is for the 3P ground state of carbon (2p2 ) in symmetry group I. define
recognizes most cases and suggests good Roothaan parameters.
For further information on ROHF calculations (e.g. with more than one open shell),
see the sample input in Section 23.6 and the tables of Roothaan parameters in Sec-
tion 6.3.
Note that this keyword toggles the ROHF mode also for more than one open shell. If
it is not given, the open-shell electrons are simply ignored.

UHF

$alpha shells and $beta shells

these two data groups specify the occupation of alpha and beta spin UHF MOs (syntax
as any data group related with orbital occupation information, e.g. $closed shells)
Example:

$alpha shells
a 1-8 ( 1 )
b 1-2 ( 1 )
†
C. C. J. Roothaan: Rev. Mod. Phys. 32 (1960) 179.
406 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$beta shells
a 1-7 ( 1 )
b 1-3 ( 1 )

$uhf
directs the program to carry out a UHF run. $uhf overwrites closed-shell occupation
specification.

$spin constraint {real}

The presence of $uhf activates the UHF mode. An additional constraint can be
imposed for the (spin)2 exceptation value by using $spin constraint {real}. A
non-zero real value fixes the (spin)2 expectation value by using [313]
Fα (SAO) − 2 ∗ τ ∗ SDβ S
Fβ (SAO) − 2 ∗ τ ∗ SDα S
instead of Fα , Fβ ; in the limiting case (real → INFINITY) the ROHF (spin)2 expec-
tation value would result. Default spin constraint 0, this is also used if the key is not
set. We suggest to increase the constraint in the sequence 0.01, 0.1, 1.0, and 10.0 to
check the impact on the (spin)2 expectation value.

$uhfmo_alpha and $uhfmo_beta

These two data groups contain the UHF MO vectors for alpha and beta spin respec-
tively (same syntax as $scfmo).

$uhfmo_beta
see $uhfmo_alpha

DFT

$dft
functional b-p
gridsize m3

for DFT calculations one has to specify the functional and the grid (for the quadrature of
the exchange correlation part). The settings above are default: both lines can be left out if
the B-P86 functional and grid m3 are required. Other useful functionals supported are:

b-lyp
b3-lyp
b3-lyp_Gaussian (equivalent to the Gaussian98 keyword B3LYP with VWNIII)
bh-lyp
s-vwn
s-vwn_Gaussian (equivalent to the Gaussian98 keyword SVWN with VWNIII)
tpss
tpssh
22.2. FORMAT OF KEYWORDS AND COMMENTS 407

Possible grids are 1–5 and m3–m5 where grid 1 is coarse (least accurate) and 5 most dense.
We recommend however the use of so-called multiple grids m3–m5: SCF iterations with
grid 1–3, final energy and gradient with grid 3–5. Usually m3 is fine: for large or delicate
systems, try m4. For a reference calculation with a very fine grid and very tight thresholds
use ’reference’ as grid specification instead of ’gridsize xy’.
Note: the functionals b3-lyp_Gaussian and s-vwn_Gaussian are made available only for
comparability with Gaussian. The functional VWNIII is much less well founded than VWN5
and the TURBOMOLE team does not recommend the use of VWNIII.

RI

Dscf does not run with the keyword $rij: you must call the RI modules Ridft and Rdgrad
for energy and gradient calculations. However, it does run with the keyword $rik, but it will
ignore all RI settings and do a conventional non-RI Hartree–Fock or DFT calculation.

$rij
Enforces an RI-J calculation if module ridft is used, can be used for Hartree-Fock
as well as for DFT calculations with pure or hybrid functionals.

$ridft
Obsolete keyword - use $rij instead!

$rik
Enforces a RI-JK calculation if module ridft is used, can be used for Hartree-Fock
as well as for DFT calculations with pure or hybrid functionals.

$ricore integer
Choose the memory core available (in megabyte) for special arrays in the RI calcula-
tion (the less memory you give the more integrals are treated directly, i.e. recomputed
on the fly in every iteration)

$jbas file=auxbasis
Cross reference for the file specifying the auxiliary basis as referenced in $atoms. We
strongly recommend using auxbasis sets optimized for the respective MO basis sets,
e.g. use SVP (or TZVP) for the basis and the corresponding auxbasis as provided by
define (default: file=auxbasis).

$ripop
Calculation of atomic charges according to the s partial wave and atomic dipole mo-
ments according to the p partial wave as resulting from the auxbasis representation
of the density
408 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

RI-JK

If the keyword $rik is found in the control file, ridft performs a Hartree–Fock–SCF
calculation using the RI-approximation for both Coulomb and HF-exchange (efficient for
large basis sets). For this purpose needed (apart from $ricore):

$jkbas file=auxbasis
Cross reference for the file specifying the JK-auxiliary basis as referenced in $atoms.
This group is created by the rijk menu in define.

MARI-J

Multipole-Accelerated-Resolution-of-Identity-J. This method partitions the Coulomb inter-

actions in the near- and far-field parts. The calculation of the far-field part is performed by
application of the multipole expansions and the near-field part is evaluated employing the
RI-J approximation. It speeds up calculation of the Coulomb term for large systems. It can
only be used with the ridft module and requires setting of the $ridft keyword.

$marij
precision 1.0D-06
lmaxmom 10
nbinmax 8
wsindex 0.0
extmax 20.0
thrmom 1.0D-18

The following options are available:

precision specifies precision parameter for the multipole expansions. Low-precision

MARI-J calculations require 1 · 10−6 , which is the default. For higher
precision calculations it should be set to 1 · 10−8 –1 · 10−9 .
lmaxmom maximum l-moment of multipole expansions. It should be set to a value
equal at least twice the maximum angular momentum of basis functions.
Default value is 10 and it should probably never be set higher than 18.
thrmom Threshold for moment summation. For highly accurate calculations it
should be set to 1 · 10−24 .
nbinmax number of bins per atom for partitioning of electron densities. Default
value is 8 and hardly ever needs to be changed.
wsindex minimum separation between bins. Only bins separated more than the
sum of their extents plus wsindex are considered as far-field. Default is
0.0 and should be changed only by the experts.
extmax maximum extent for charge distributions of partitioned densities. Ex-
tents with values larger then this are set to extmax. Hardly ever needs
to be changed.
22.2. FORMAT OF KEYWORDS AND COMMENTS 409

Seminumeric HF-Exchange

If the keyword $senex is found in the control file, ridft performs a Hartree–Fock– SCF
calculation using the seminumerical approximation for HF-exchange. Standard dft-grids can
be used for the numerical integration. Smaller grids (-1,0) and the corresponding m-grids
(m1,m2) have been defined as well. For high precision energies the use of de-aliasing is rec-
ommended (do_sfit, C1 symmetry only) which will yield reliable energies with the m1 grid
in nearly all cases. Alternatively grids of at least size m3 are recommended for heavy atoms.
The gridsize can be modified just like in dft-calculations. We introduced three different key-
words to finetune where you want to employ the seminumerical exchange approximations.
The keyword $senex activates seminumerical calculations in energies (ridft), excitation en-
ergies (escf), vibrational frequencies (aoforce) and chemical shifts (mpshift). The keyword
$esenex activates seminumerical exchange in escf, aoforce, mpshift, egrad(excitation en-
ergy part only) but not in ridft. Further it resets the default grid to -1 in these modules;
which was found to be sufficient in most cases. The keyword $dsenex activates seminumer-
ical gradient calculations in ground (rdgrad) and excited states (egrad; excitation energy
AND gradient part). An example using the default grid and de-aliasing for SCF (m1) and
grid m2 for gradients looks like this:

$senex
do_sfit
$dsenex
gridsize m2

Seminumeric Coulomb+Exchange: The pseudospectral approach

The modules escf, aoforceand egradcan make use of the pseudospectral approximation to
calculate Coulomb contribution seminumerically on a grid. To do so simply add the keyword
$pseudospectral additionally if $senex or $esenex are present. This is especially valuable
when no RI-Fitting basis is available for the chosen basis set (e.g. "cbas" for Sapporo-type
basis sets, ANO-type basis sets, x2c-TZVPPall etc.) as the pseudospectral approach does
not need any auxiliar/fitting basis sets at all. Excitation energies, harmonic frequencies
and ZPEs obtained from the pseudospectral approximation are extremely accurate (usually
considerably lower errors than RI) and still orders of magnitude faster than using classic
Coulomb integrals. For ground states (e.g. ridft) the pseudospectral approximation is
not yet available. As the RI approximation provides true upper bounds in this case it will
outperform the pseudospectral approximation. Further auxiliary basis sets of type "jbas"
are, unlike "cbas" auxiliary basis sets, widely available for the whole periodic table. In C1-
symmetry we recommend the use of de-aliasing using the do_sfit keyword. An example
for applying the de-aliased pseudospectral approximation looks like this:

$senex
do_sfit
$pseudospectral
410 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

LHF

Use the Localized Hartree–Fock (LHF) method to obtain an effective Exact-Exchange Kohn–
Sham potential (module dscf). The LHF method is a serial implementation for spin-
restricted closed-shell and spin-unrestricted ground states.

$dft
functional lhf
gridsize 6

With the LHF potential Rydberg series of virtual orbitals can be obtained. To that end,
diffuse orbital basis sets have to be used and special grids are required.
gridtype 4 is the most diffuse with special radial scaling; gridtype 5 is for very good
Rydberg orbitals; gridtype 6 (default in Lhfprep) is the least diffuse, only for the first
Rydberg orbitals.
Only gridsize 3–5 can be used, no multiple grids.
Use test-integ to check if the selected grid is accurate enough for the employed basis-set.

How to do LHF runs

1) Do a Hartree–Fock calculation using dscf.

2) Use the script lhfprep to prepare the control file (the old control file will be saved
in control.hf and the molecular orbitals in mos.hf or in alpha.hf and beta.hf for
the spin-unrestricted case). See lhfprep -help for options.

3) Run again dscf.

Otherwise the LHF functional can be selected in define: in this case default options are
used.
Options for the LHF potential can be specified as follows ( see also lhfprep -help)

$lhf
off-diag on
numerical-slater off
pot-file save
asymptotic dynamic=1.d-3
homo 1b1u
homob 1b1u # ONLY UNRESTRICTED
conj-grad conv=1.d-7 maxit=20 output=1 cgasy=1
slater-dtresh 1.d-9
slater-region 7.0 0.5 10.0 0.5
corrct-region 10.0 0.5
slater-b-region 7.0 0.5 10.0 0.5 # ONLY UNRESTRICTED
22.2. FORMAT OF KEYWORDS AND COMMENTS 411

corrct-b-region 10.0 0.5 # ONLY UNRESTRICTED

correlation func=lyp

off-diag off
calculation of the KLI exchange potential. By default the LHF exchange po-
tential is computed (off-diag on).
numerical-slater on
the Slater potential is calculated numerically everywhere: this is more accurate
but much more expensive. When ECP are used, turn on this option.
numerical-slater off
leads to accurate results only for first-row elements or if an uncontracted basis
set or a basis set with special additional contractions is used: in other cases
numerical-slater on has to be used (this is default).
asymptotic
for asymptotic treatment there are three options:
asymptotic off
No asymptotic-treatment and no use of the numerical Slater. The total
exchange potential is just replaced by −1/r in the asymptotic region.
This method is the fastest one but can be used only for the density-
matrix convergence or if Rydberg virtual orbitals are of no interest.
asymptotic on
Full asymptotic-treatment and use of the numerical Slater in the near
asymptotic-region.
asymptotic dynamic=1.d-3
Automatic switching on (off) to the special asymptotic treatment if the
differential density-matrix rms is below (above) 1.d-3. This is the default.
pot-file save
the converged Slater and correction potentials for all grid points are saved in
the files slater.pot and corrct.pot, respectively. Using pot-file load, the
Slater potential is not calculated but read from slater.pot (the correction
potential is instead recalculated). For spin unrestricted calculations the corre-
sponding files are slaterA.pot, slaterB.pot, corrctA.pot and correctB.pot.
homo
allows the user to specify which occupied orbital will not be included in the
calculation of correction potential: by default the highest occupied orbital is
selected. This option is useful for those systems where the HOMO of the
starting orbitals (e.g. EHT, HF) is different from the final LHF HOMO. homob
is for the beta spin.
correlation func=functional
a correlation functional can be added to the LHF potential: use func=lyp for
LYP, or func=vwn for VWN5 correlation.
412 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

For expert users

Options for the conjugate-gradient algorithm for the computation of the correction poten-
tial: rms-convergence (conj-grad conv=1.d-7), maximum number of iteration (maxit=20),
output level output=0-3, asymptotic continuation in each iteration (cgasy=1).
With slater-dtresh= 1.d-9 (default) the calculations of the numerical integrals for the
Slater potential is performed only if it changes more than 1.d-9.
Asymptotic regions specification:

corrct-region RF ∆F
0 . . . RF − ∆F : basis-set correction potential
RF − ∆F . . . RF + ∆F : smooth region
RF + ∆F . . . + ∞ : asymptotic correction
Defaults: RF = 10, ∆F = 0.5
slater-region RN ∆N RF0 ∆0F
0 . . . RN − ∆N : basis-set Slater potential
RN − ∆N . . . RN + ∆N : smoothing region
RN + ∆N . . . RF0 − ∆0F : numerical Slater
RF0 − ∆0F . . . RF0 + ∆0F : smoothing region
RF0 + ∆0F . . . + ∞ : asymptotic Slater
Note: RF0 − ∆0F ≤ RF − ∆F
Defaults: RN = 7, ∆N = 0.5, RF0 = 10, ∆0F = 0.5
Use correct-b-region and slater-b-region for the beta spin.

Two-component SCF (GHF)

Self-consistent two-component calculations (e.g. for spin-orbit interactions) can be carried

out using the module ridft . The following keywords are valid:

$soghf
enforces two-component-SCF calculations; this option is compatible with $rij, $rik
and $dft.
$kramers
switches on Kramers-restricted formalism
$collinear
switches on collinear two-component formalism (not rotational invariant)
$gdiis
enforces DIIS for complex Fock operator.

All-electron relativistic approaches (X2C, BSS, DKH)

Relativistic all-electron calculations can be done employing the X2C, the BSS or the DKH
Hamiltonian. Implemented for modules dscf and ridft. Note that gradients are only
22.2. FORMAT OF KEYWORDS AND COMMENTS 413

available with X2C in the modules grad and rdgrad.

$rx2c
switches on X2C calculation.
$rbss
switches on BSS calculation.
$rdkh integer
switches on DKH calculation of order integer.
$dkhparam integer
selects parameterization of the DKH Hamiltonian. Valid values are 1 (=default), 2,
3, 4, and 5.

$dkhparam 1: Optimum parametrization (OPT)

$dkhparam 2: Exponential parametrization (EXP)
$dkhparam 3: Square-root parametrization (SQR)
$dkhparam 4: McWeeny parametrization (MCW)
$dkhparam 5: Cayley parametrization (CAY)

Note in particular that the parametrization does not affect the Hamiltonian up to
4th order. Therefore, as long as you run calculations with DKH Hamiltonians be-
low 5th order you may use any symbol for the parametrization as they would all
yield the same results. Higher-order DKH Hamiltonians depend slightly on the cho-
sen paramterization of the unitary transformations applied in order to decouple the
Dirac Hamiltonian, but this effect can be neglected. For details on the different
parametrizations of the unitary transformations see [314].
$rlocal
switches on local approach. Default is DLU (see $rlocpara).
$rlocpara integer
selects parameterization of the local approximation. Valid values are 0 or 1. Default
is the DLU (0). 1 refers to the DLH approximation. For details on the different
parametrizations see [124]. Gradients are restricted to the DLU approximation.
$finnuc
switches on finite nucleus model based on a Gaussian charge distribution with param-
eters taken from [315].
$snso
selects screened-nuclear-spin-orbit (SNSO) approach for the spin-dependent integrals
in two-component calculations.
$snsopara integer
selects parametrization of SNSO. Valid values are 0 and 1. 0 refers to the original
set of parameters originally used in low-order DKH theory [127], while 1 refers to the
modified parameters [128, 129] for exact decoupling methods. Default is 0.
414 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

All of these keywords are compatible with $soghf employing the RI-J approximation.

22.2.9 Keywords for Point Charge Embedding

$point_charges
Specification of position and magnitude of point charges to be included in the Hamil-
tonian. Each point charge is defined in the format

<x> <y> <z> <q>

with <x>, <y>, <z> being the coordinates and <q> its charge,e.g.

$point_charges thr=<real> selfenergy nocheck list

2. 2. 2. 5.
5. 0. 0. 2.5

In addition the following optional arguments may be given:

thr=real
distance threshold for discarding redundant point charges, default value 10−6 .

selfenergy
if given, the selfenergy of the point charge array will will be included in the energy
and the gradient

nocheck
switches off the check for redundant point charges and the default symmetrization.
This option can significantly speed up the point charge treatment if many of them
are involved - use only if the point charges are well distributed and symmetry is C1 ,
e.g. when they stem from proper MM runs

list
print all point charges in the output (default is to print the point charges only if less
than 100 charges given)

22.2.10 Keywords for Periodic Electrostatic Embedded Cluster Method

The Periodic Electrostatic Embedded Cluster Method (PEECM) functionality provides elec-
tronic embedding of a finite, quantum mechanical cluster in a periodic, infinite array of point
charges. It is implemented within HF and DFT energy and gradient TURBOMOLE modules:
dscf, grad, ridft, rdgrad, and escf. Unlike embedding within a finite set of point charges
the PEEC method always yields the correct electrostatic (Madelung) potential independent
of the electrostatic moments of the point charges field. It is also significantly faster than the
traditional finite point charges embedding.
22.2. FORMAT OF KEYWORDS AND COMMENTS 415

The basic PEECM settings are defined in the $embed block. It can be redirected to an
external file using $embed file=<file_name>.
Following keywords are used for the PEECM calculation setup:

periodic

Specifies the number of periodic directions. Allowed values for number are 3 for a bulk
three-dimensional system, 2 for a two-dimensional surface slab, and 1 for a one-dimensional
system. Default value is 3.

cell

Unit cell parameters in a form of six real values |a|, |b|, |c|, α, β, γ, where |a|, |b|, |c| are
lengths of the appropriate cell vectors, α is the angle between vectors b and c, β is the angle
between vectors a and c, and γ is the angle between vectors a and b. Default are atomic
units and degrees. You can specify unit cell parameters in Å and degrees using cell ang.

content
label x y z
end

Content of the unit cell, where label is the label of the point charge Content of the unit cell,
where label is the label of the point charge type and x y z are corresponding Cartesian
or fractional crystal coordinates. Defaults are Cartesian coordinates and atomic units. You
can specify Cartesian coordinates in Å using content ang and fractional coordinates using
content frac. Note that Cartesian coordinates assume that the cell vector a is aligned
along the x axis, and the vector b on the xy plane.

cluster
label x y z
end

Atomic coordinates of the piece of the crystal to be replaced by the QM cluster and sur-
rounding isolation shell (ECPs and explicit point charges), where label is the point charge
label and x y z are corresponding Cartesian or fractional crystal coordinates. Defaults are
Cartesian coordinates and atomic units. You can specify Cartesian coordinates in Å using
cluster ang and fractional coordinates using cluster frac.

charges
label charge
end

Values of point charges (for each atom-type) , where label is the point charge label and
charge specifies charge in atomic units.
416 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

ch_list
label charge
end

Values of point charges (for each atom), where label is the point charge label and charge
specifies charge in atomic units.
Note that charges and ch_list are mutually exclusive. An integer number n can also be
appended to charges or ch_list to set the tolerance for charge neutrality violation to 10−n
(default n = 5).
Additionally, the following keywords control the accuracy of PEECM calculation:

lmaxmom
Maximum order of the multipole expansions in periodic fast multipole method (PFMM).
Default value is 25.

potval
Electrostatic potential at the lattice points resulting from periodic point charges field
will be output if this keyword is present. Default is not to output.

wsicl
Well-separateness criterion for PFMM. Default is 3.0.

epsilon
Minimum accuracy for lattice sums in PFMM. Default is 1.0d-8.

22.2.11 Keywords for COSMO

The Conductor-like Screening Model (Cosmo) is a continuum solvation model, where the
solute molecule forms a cavity within the dielectric continuum of permittivity epsilon that
represents the solvent. A brief description of the method is given in chapter 18.2. The
model is currently implemented for SCF energy and gradient calculations (dscf/ridft and
grad/rdgrad), MP2 energy calculations (rimp2 and mpgrad) and MP2 gradients (rimp2),
and response calculations with escf. The ricc2 implementation is described in section
18.2.5.

For simple HF or DFT single point calculations or optimizations with standard settings,
we recommend to add the $cosmo keyword to the control file and to skip the rest of this
section.

Please note: due to improvements in the A matrix and cavity setup the Cosmo energies
and gradients may differ from older versions (5.7 and older). The use_old_amat option can
be used to calculate energies (not gradients) using the old cavity algorithm of Turbomole
5.7.
22.2. FORMAT OF KEYWORDS AND COMMENTS 417

The basic Cosmo settings are defined in the $cosmo and the $cosmo_atoms block.
Example with default values:

$cosmo
epsilon=infinity
nppa= 1082
nspa= 92
disex= 10.0000
rsolv= 1.30
routf= 0.85
cavity closed
ampran= 0.1D-04
phsran= 0.0
refind= 1.3
# the following options are not used by default
allocate_nps= 140
use_old_amat
use_contcav
no_oc

epsilon=real
defines a finite permittivity used for scaling of the screening charges. If the option
ε−1
ion is added to the same input line, the scaling factor for ions f (ε) = ε+x with
x = 0 will be used. Alternatively the x value can be set by adding ion=x with x
as a real value.

allocate_nps=integer
skips the Cosmo segment statistics run and allocates memory for the given num-
ber of segments.

no_oc
skips the outlying charge correction.

All other parameters affect the generation of the surface and the construction of the A
matrix:

nppa= integer
number of basis grid points per atom
(allowed values: i = 10 × 3k × 4l + 2 = 12, 32, 42, 92...)

nspa= integer
number of segments per atom
(allowed values: i = 10 × 3k × 4l + 2 = 12, 32, 42, 92...)

disex= real
distance threshold for A matrix elements (Ångstrom)
418 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

rsolv= real
distance to outer solvent sphere for cavity construction (Ångstrom)

routf= real
factor for outer cavity construction in the outlying charge correction

cavity closed
pave intersection seams with segments

cavity open
leave untidy seams between atoms

ampran= real
amplitude of the cavity de-symmetrization

phsran= real
phase of the cavity de-symmetrization

refind= real
refractive index used for the calculation of vertical excitations and num. frequen-
cies (the default 1.3 will be used if not set explicitly)

use_old_amat
uses A matrix setup of TURBOMOLE 5.7

use_contcav
in case of disjunct cavities only the largest contiguous cavity will be used and the
smaller one(s) neglected. This makes sense if an unwanted inner cavity has been
constructed e.g. in the case of fullerenes. Default is to use all cavities.

point_charges
Allows the COSMO response to point charges. By default it is disables, so that
point charges interacts only with the molecule. Note that when point charges are
used together with COSMO, the COSMO surface is still constructed considering
only the atoms.

If the $cosmo keyword is given without further specifications the default parameter are used
(recommended). For the generation of the cavity, Cosmo also requires the definition of
atomic radii. User defined values can be provided in Ångstrom units in the data group
$cosmo_atoms, e.g. for a water molecule:

$cosmo_atoms
# radii in Angstrom units
o 1 \
radius= 1.7200
h 2-3 \
radius= 1.3000
22.2. FORMAT OF KEYWORDS AND COMMENTS 419

If this section is missing in the control file, the default values defined in the radii.cosmo
file (located in $TURBODIR/parameter) are used. A user defined value supersedes this de-
faults. $cosmo and $cosmo_atoms can be set interactively with the Cosmo input program
cosmoprep after the usual generation of the TURBOMOLE input.
The Cosmo energies and total charges are listed in the result section. E.g.:

SCREENING CHARGE:
cosmo : -0.003925
correction : 0.003644
total : -0.000282
ENERGIES [a.u.]:
Total energy = -76.0296831863
Total energy + OC corr. = -76.0297567835
Dielectric energy = -0.0118029468
Diel. energy + OC corr. = -0.0118765440
The following value is included for downward compatibility
Total energy corrected = -76.0297199849

The dielectric energy of the system is already included in the total energy. OC corr denotes
the outlying charge correction. The last energy entry gives the total outlying charge cor-
rected energy in the old definition used in TURBOMOLE 5.7 and older versions. The Cosmo
result file, which contains the segment information, energies, and settings, can be set using:
$cosmo_out file= filename.cosmo

Isodensity Cavity: This option can be used in HF/DFT single point calculations only.
The $cosmo_isodens section defines the settings for the density based cavity setup (see
also chapter 18.2). If the $cosmo_isodens keyword is given without suboptions, a scaled
iosodensity cavity with default settings will be created. Possible options are:

$cosmo_isodens
activates the density based cavity setup. The default values of nspa and nsph
are changed to 162 and 92, respectively. This values are superseded by the user
defined nspa value of the $cosmo section. By default the scaled density method
is used. The atom type dependent density values are read from the radii.cosmo
file (located in $TURBODIR/parameter).

dx=real
spacing of the marching tetrahedron grid in Å (default: 0.3Å).

all_dens=real
use one isodensity value for all atom types (value in a.u.)

The outlying charge correction will be performed with a radii based outer cavity. Therfore,
and for the smoothing of the density changes in the intersection areas of the scaled density
method, radii are needed as for the standard Cosmo cavity. Please note: The isodensity
420 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

cavity will be constructed only once at the beginning of the SCF calculation. The density
constructed from the initial mos will be used (file mos or alpha/beta in case of unrestricted
calculations). Because the mos of an initial guess do not provide a good density for the
cavity construction, it is necessary to provide mos of a converged SCF calculation (e.g.
a Cosmo calculation with standard cavity). We recommend the following three steps:
perform a standard Cosmo calculation, add the isodensity options afterwards, and start
the calculation a second time.

Radii based Isosurface Cavity: The $cosmo_isorad section defines the radii defined
isosurface cavity construction. The option uses the algorithm of the isodensity cavity con-
struction but the objective function used depends on the cosmo radii instead of the electron
density. The default values of nspa and nsph are changed to 162 and 92, respectively. This
values are superseded by the user defined nspa value of the $cosmo section. The resulting
surface exhibits smoother intersection seams and the segment areas are less diverse than the
ones of the standard radii bases cavity construction.

$cosmo_isorad

dx=real
spacing of the marching tetrahedron grid in Å (default: 0.3Å).

COSMO in MP2 Calculations: The iterative Cosmo PTED scheme (see chapter
18.2) can be used with the mp2cosmo script. Options are explained in the help message
(mp2cosmo -h). Both MP2 modules rimp2 and mpgrad can be utilized. The control file
can be prepared by a normal Cosmo SCF input followed by a rimp2 or mpgrad input. The
PTE gradients can be switched on by using the

$cosmo_correlated

keyword (rimp2 only). Again a normal SCF Cosmo input followed by a rimp2 input has to
be generated. The $cosmo_correlated keyword forces dscf to keep the Cosmo information
needed for the following MP2 calculation and toggles on the Cosmo gradient contribution.

COSMO in Numerical Frequency Calculations: NumForce can handle two types

of Cosmo frequency calculations. The first uses the normal relaxed Cosmo energy and
gradient. It can be performed with a standard dscf or ridft Cosmo input without further
settings. This is the right method to calculate a Hessian for optimizations. The second
type, which uses the approach described in chapter 18.2, is implemented for ridft only.
The input is the same as in the first case but NumForce has to be called with the -cosmo
option. If no solvent refractive index refind=real is given in the $cosmo section of the
control file the program uses the default (1.3).
22.2. FORMAT OF KEYWORDS AND COMMENTS 421

COSMO in vertical excitations and polarizabilities: Cosmo is implemented in

escf and will be switched on automatically by the Cosmo keywords of the underlying SCF
calculation. The refractive index, used for the fast term screening of the vertical excitations,
needs to be defined in the cosmo section of control file (refind=real).

COSMO in CC2 and ADC(2) calculations: For the calculation of ground-state

energy at COSMO-CC2, vertival excitation energy at COSMO-CC2 and COSMO-ADC(2)
and excited-state analytic gradient at COSMO-ADC(2), the post-SCF reaction-field scheme
will be switched on by the $reaction_field keyword as:

$reaction_field
post-SCF
ccs-like

DCCOSMO-RS: The DCOSMO-RS model (see chapter 18.2) has been implemented
for restricted and unrestricted DFT and HF energy calculations and gradients (programs:
dscf/ridft and grad/rdgrad). In addition to the Cosmo settings defined at the beginning
of this section, the $dcosmo_rs keyword has to be set.

$dcosmo_rs file=filename.pot
activates the DCOSMO-RS method. The file defined in this option contains the
DCOSMO-RS σ-potential and related data (examples can be found in the defaut
potentials in the $TURBODIR/parameter directory).

If the potential file cannot be found in the local directory of the calculation, it will be
searched in the $TURBODIR/parameter directory. The following σ-potential files for pure
solvents at 25 ◦ C are implemented in the current Turbomole distribution (see parameter
subdirectory):

Water: h2o_25.pot
Ethanol: ethanol_25.pot
Methanol: methanol_25.pot
Tetrahydrofurane: thf_25.pot
Acetone: propanone_25.pot
Chloroform: chcl3_25.pot
Tetrachloromethane: ccl4_25.pot
Acetonitrile: acetonitrile_25.pot
Nitromethane: nitromethane_25.pot
Dimethylsulfoxide: dimethylsulfoxide_25.pot
Diethylether: diethylether_25.pot
Hexane: hexane_25.pot
Cyclohexane: cyclohexane_25.pot
Benzene: benzene_25.pot
Toluene: toluene_25.pot
Aniline: aniline_25.pot
422 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

The DCOSMO-RS energies and total charges are listed in the Cosmo section of the output:

SCREENING CHARGE:
cosmo : -0.012321
correction : 0.011808
total : -0.000513
(correction on the COSMO level)
ENERGIES [a.u.]:
Total energy = -76.4841708454
Outlying charge corr. (COSMO) = -0.0006542315
Outlying charge corr. (DCOSMO-RS)= -0.0011042856
Combinatorial contribution of the solute = -0.0017627889
(at inf. dil. in the mixture/pure solvent. Not included in the total energy above)

The outlying charge correction cannot be defined straight forward like in the normal Cosmo
model. Therefore, the output shows two corrections that can be added to the Total energy.
The first one is the correction on the Cosmo level (COSMO) and the second is the difference
of the DCOSMO-RS dielectric energy calculated form the corrected and the uncorrected
Cosmo charges, respectively (DCOSMO-RS). The charges are corrected on the Cosmo level
only. The Total energy includes the Ediel,RS defined in section 18.2.4. Additionally the
combinatorial contribution at infinute dilution of the COSMO-RS model is given in the out-
put. The use of this energy makes sense if the molecule under consideration is different than
the used solvent or not component of the solvent mixture, respectively. To be consistent one
should only compare energies containing the same contributions, i.e. same outlying charge
correction and with or without combinatorial contribution. Please note: the COSMO-RS
contribution of the DCOSMO-RS energy depends on the reference state and the COSMO-RS
parameterization (used in the calculation of the chosen COSMO-RS potential). Therefore,
the DCOSMO-RS energies should not be used in a comparision with the gas phase energy,
i.e. the calculation of solvation energies.

22.2.12 Keywords for Module riper

riper shares most of the relevant keywords of the dscf and ridft modules. The $dft
data group (see 22.2.8) and auxiliary basis sets defined using the keyword $jbas are always
required.
For periodic calculations two additional keywords are necessary:

$periodic n
Specifies the number of periodic directions: n = 3 for a 3D periodic bulk solid, n = 2
for a 2D periodic surface slab and n = 1 for a 1D periodic system. The default value
is 0 for a molecular system.
$cell
Specifies the unit cell parameters. The number of cell parameters depends on the
periodicity of the system:
22.2. FORMAT OF KEYWORDS AND COMMENTS 423

For 3D periodic systems six unit cell parameters |a|, |b|, |c|, α, β and γ need to be
provided. Here, |a|, |b| and |c| are lengths of the appropriate cell vectors, α is the
angle between vectors b and c, β is the angle between vectors a and c, and γ is the
angle between vectors a and b. riper assumes that the cell vectors a and b are
aligned along the x axis and on the xy plane, respectively.
For 2D periodic systems three surface cell parameters |a|, |b| and γ have to be pro-
vided. Here, |a| and |b| are lengths of the appropriate cell vectors and γ is the angle
between a and b. riper assumes that the cell vectors a and b are aligned along the
x axis and on the xy plane, respectively.
For 1D periodic systems only one parameter specifying the length of the unit cell has
to be provided. riper assumes that periodic direction is along the x axis.

$lattice
Alternatively, lattice vectors can be provided. The number of cell parameters depends
on the periodicity of the system:
For 3D periodic systems three (three-dimensional) lattice vectors need to be provided.
For 2D periodic systems two (two-dimensional) lattice vectors have to be provided.
riper assumes that the lattice vectors are aligned on the xy plane.
For 1D periodic systems only one parameter specifying the length of the lattice vector
has to be provided. riper assumes that periodic direction is along the x axis.

Optionally, for periodic systems a k points mesh can be specified:

$kpoints
nkpoints n1 n2 n3

Specifies components along each reciprocal lattice vector of a Γ centered mesh of

k points. In 3D periodic systems each k point is defined by its components k1 , k2
and k3 along the reciprocal lattice vectors b1 , b2 and b3 as

k = k1 b1 + k2 b2 + k3 b3 . (22.1)

For 2D periodic systems k3 = 0. In case of 1D periodicity k3 = 0 and k2 = 0. The

three components kj (j = 1, 2, 3) of k are given as

i nj − 1 nj − 1 nj − 1 nj − 1
kj = with i = − ,− + 1, . . . , − 1, . (22.2)
nj 2 2 2 2

with nj (j = 1, 2, 3) as integer numbers. For 3D periodic systems n1 , n2 and n3 have

to be specified. The component n3 can be omitted for 2D periodic systems. In case
of 1D periodicity only the n1 has to be specified.

In addition, the options kptlines and recipr within the $kpoints group can be
used to obtain band structure plots:

kptlines Specifies the number of reciprocal space lines for band structure plots.
424 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

recipr Specifies lines in the reciprocal space using three real numbers defining
the start point of the line and three real numbers defining its end point.
Finally, the number of k points along the line is given as an integer
number.

In the following example band energies are calculated along four lines, as specified by
the keyword kptlines 4. Each line definition starts in a new line with the keyword
recipr, followed by three real numbers defining the start point of the line and three
real numbers defining its end point. Finally, the number of k points along the line
is given as an integer number. Thus, the first line starts at the point (0.500 0.500
0.500), ends at (0.000 0.000 0.000) and contains 40 k points.

$kpoints
kptlines 4
recipr 0.500 0.500 0.500 0.000 0.000 0.000 40
recipr 0.000 0.000 0.000 0.500 0.500 0.000 40
recipr 0.500 0.500 0.000 0.746 0.373 0.373 40
recipr 0.746 0.373 0.373 0.000 0.000 0.000 40

The calculated band structure is written to the file bands.xyz. Each line of the file
contains five real numbers: the coordinates k1 , k2 and k3 of the k point, its length |k|
and the corresponding band energy nk .

Keywords within the section $riper can be used to control precision and parameters of
algorithms implemented in riper. If the $riper group is absent, the following default
values are used:

$riper
# general keywords
thrints 1.0d-12
lenonly off
lchgprj on (for periodic systems)
lchgprj off (for molecular systems)
northol 5
pqmatdiag off
pqsingtol 1.0d-8
# CFMM control options
lmaxmom 20
nctrgt 10 (for periodic systems)
nctrgt 1 (for molecular systems)
wsicl 3.0
epsbext 1.0d-9
locmult on (for periodic systems)
locmult off (for molecular systems)
locmomadd 2
22.2. FORMAT OF KEYWORDS AND COMMENTS 425

# LMIDF control options

lpcg on
lcfmmpcg on
lmxmompcg 20
pcgtol 1.0d-9
pcgtyp sp
# Gaussian smearing options
sigma 0.0d0
desnue 0.0d0

The following options are available:

# general keywords
thrints Threshold for integrals neglect and for differential overlap when screen-
ing basis functions pairs. Probably never needs to be changed.
lenonly Flag for energy calculation only, no gradients.
lchgprj If set to on charge projection of the auxiliary electron density [158]
is performed for molecular systems during calculation of the Coulomb
term. The charge projection constraints the charge of auxiliary density
exactly to the number of electrons in the system. It is required for pe-
riodic systems, otherwise the Coulomb energy would be infinitely large.
For molecular systems charge projection leads to a slight increase of the
RI fitting error. It may be useful in some cases but we have so far not
identified any.
northol Forces orthonormalization of orbital coefficients every northol SCF it-
eration.
pqmatdiag If set to on full diagonalization of the Coulomb metric matrix [158]
is performed and used to solve density fitting equations. When diffuse
auxiliary basis functions are used the default Cholesky decomposition of
the Coulomb metric matrix may fail due to small negative eigenvalues.
In this case the slower method based on a full diagonalization of the
metric matrix is necessary.
pqsingtol If pqmatdiag is used pqsingtol sets threshold for neglect of small eigen-
values of the Coulomb metric matrix.
# CFMM control options
lmaxmom Maximum l-moment of multipole expansions used for calculation of the
Coulomb term. The default value hardly ever needs to be changed.
nctrgt Target number of charge distributions per lowest level box of the octree
[154]. The default value hardly ever needs to be changed.
wsicl Sets the well-separateness criterion [154]. Octree boxes with centers sep-
arated more than sum of their lengths times 0.5×wsicl are considered
426 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

as well-separated. The default hardly ever needs to be changed. wsicl

makes sense only for values ≥ 2.0. For wsicl ≤ 3.0 increasing lmaxmom
may be necessary for reasonable accuracy.
epsbext Precision parameter used to determine basis function extents [154].
locmult If set to on, an additional acceleration method employing local multi-
pole expansions is used. For periodic systems this leads to a significant
speedup of calculations, especially for small unit cells and/or diffuse ba-
sis functions. Default value is off and on for molecular and periodic
systems, respectively.
locmomadd For locmult set to on the order of local multipole expansions is increased
by locmomadd. The default value probably never needs to be changed.
# LMIDF control options
lpcg If set to on the low-memory iterative density fitting (LMIDF) scheme is
used for solving the RI equations [157] using the preconditioned conju-
gate gradient (PCG) method. It is implemented for molecular systems
only. Default value is off.
lcfmmpcg If lpcg is used, lcfmmpcg specifies whether the CFMM is applied for
evaluation of the matrix-vector products needed for the PCG solver. Not
employing CFMM speeds up the calculations but significantly increases
memory demand since the full Coulomb metric matrix has to be stored.
Default value is on.
lmxmompcg Maximum l-moment of multipole expansions for calculations of Coulomb
interactions within the PCG algorithm. It should be set to the same or
larger value than lmaxmom.
pcgtol Sets the threshold parameter controlling accuracy of the PCG solver
(see [157] for details). Default value is 1.0 · 10−9 . For lower-precision
calculations it can be set to 1.0 · 10−8 but values larger than 1.0 · 10−7
are not allowed as these lead to large errors in Coulomb energies and
occasionally to SCF convergence problems.
pcgtyp char
char = at, ss or sp
Specifies the type of preconditioner used in the PCG algorithm. Three
types of preconditioners are implemented and are defined explicitly in
Sec. 7. The sp preconditioner is a default one performing consistently
the best among the preconditioners considered. The at preconditioner
is less efficient in decreasing the number of CG iterations needed for
convergence. However, it has negligible memory requirements and more
favorable scaling properties, albeit with a large prefactor. The ss pre-
conditioner represents a middle ground between the sp and at precon-
ditioners both in terms of the efficiency and memory requirements.
# Gaussian smearing options
22.2. FORMAT OF KEYWORDS AND COMMENTS 427

sigma Width of the Gaussian smearing in hartree. See ref. [160] for more
information. Note, that [160] uses eV as the unit for the width of the
Gaussian smearing.
desnue Specifying desnue along with sigma forces occupancy leading to the
number of unpaired electrons equal to desnue.

Keywords within the section $pointvalper can be used to evaluate quantities (electron
density/molecular orbitals) for visualization on grid points:

$pointvalper fmt=plt
dens
orbs 2
k 3 2 1 a 1 i
k 0 0 0 b 2 r
nimg 2 2 1
npts 100 100 100
eps 5.0
ngrdpbx 50
full

The following options are available:

fmt Specifies output format. Currently .plt, .cub, .xyz and .upt exten-
sions are supported (see 7.3.8).
dens If present, total (and spin for UHF) density is calculated.
orbs Specifies the number of plotted orbitals. For further details see subsec-
tion 7.3.8.
nimg Number of unit cell images n1, n2 and n3 in the periodic directions a,
b and c, respectively, for which plot data is generated.
npts Number of grid points n1, n2 and n3 along each periodic direction. If
not specified, value 100 is used for each n.
eps Specifies the distance real in bohr around the system for which plot
grid is generated in aperiodic directions. Default value is 5 bohr.
ngrdpbx Number of grid points stored in one octree box during density calcula-
tions. Default value is 50. For very large systems or high resolution it
may be necessary to increase this parameter to avoid memory allocation
problems.
full Only valid for plt output format. This format uses orthogonal grids.
Therefore, for non-orthogonal unit cells grid data is generated for a
rectangular box that contains the supercell (unit cell and its periodic
428 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

images). By default, the values at grid points outside of the supercell

are set to zero. For strongly non-orthogonal systems this may lead to
large files. The option full switches off the zeroing of values on grid
points outside the supercell.

To calculate a simulated density of states (DOS) keyword $dosper is necessary:

$dosper width=real emin=real emax=real scal=real npt=integer

The following options are available:

width The width of each Gaussian, default value is 0.01 a.u.

emin,enmax Lower/upper bounds for energy in DOS calculation.
scal Scaling factor for DOS (total and s-, p-, ... contributions).
npt Resolution (number of points).

Keywords within the section $rttddft can be used to specify the parameters for RT-
TDDFT. Example:

$rttddft
magnus 2
scf off
time 1000.0d0
tstep 0.1d0
min energy = 0.013d0
max energy = 0.5d0
energy step 0.001d0

The options are explained below.

magnus Can take values 2 or 4. "2" for second order Magnus expansion and "4"
for fourth order Magnus expansion. Default value is 2.
scf if on, then SCF procedure is used for the time integration. If off
then Predictor-Corrector scheme is used instead.
iterlim Max SCF cycles if scf is on. Default value is 15.
time Specifies the evolution time in au. (1 au= 0.02419 fs)
tstep The time step for the time evolution in au. 0.1 au is usually a good
starting point.
print step Specifies the number of steps n after which the dipole moments and
energies are printed out if requested. Default value is 100. That means
the quantities are printed out at every 100 steps. To have all the infor-
mation for post processing, a value of 1 is recommended.
22.2. FORMAT OF KEYWORDS AND COMMENTS 429

damping Only valid for absorption spectrum calculation. It is the factor gamma
in the equation to calculate the complex polarizability tensor. Default
value is 0.004 au. Recommended values in the range of 0.003 au to 0.005
au.
min energy Only valid for absorption spectrum calculation. Specifies the minimum
value of the energy range used to perform the Fourier transform from
time to frequency space. Units: au. Default value is 0.15 au.
max energy Only valid for absorption spectrum calculation. Specifies the maximum
value of the energy range used to perform the Fourier transform from
time to frequency space. Units: au. Default value is 0.625 au.
energy step Only valid for absorption spectrum calculation. Specifies the step value
or energy interval dE at which to sample the energy values for Fourier
transform and absorption spectrum plotting. Units: au. Default value
is 0.005 au.

Keywords within the section $electric field can be used to specify the parameters for
electric field. To specify a Gaussian electric field pulse use

$electric field
amplitude x=2.0E-5 y=2.0E-5 z=2.0E-5
gaussian tzero=3.0 width=0.2

Options are explained below.

amplitude x y z Cartesian components of the amplitude of the Gaussian pulse (in

Electric field au).
gaussian Sets the electric field type to Gaussian.
tzero Position of Gaussian pulse maximum (in time au).
width Width of Gaussian pulse (in time au).

22.2.13 Keywords for Modules grad and rdgrad

Many of the dscf and ridft keywords are also used by grad and rdgrad.

$drvopt

This keyword and corresponding options are required in gradient calculations only in special
circumstances. Just $drvopt is fine, no options needed to compute derivatives of the energy
with respect to nuclear coordinates within the method specified: SCF, DFT, RIDFT.
430 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

If running a DFT gradient calculation, it is possible to include the derivatives of the quadra-
ture weights, to get more accurate results. In normal cases however those effects are
marginal. An exception is numerical calculation of frequencies by NumForce, where it is
strongly recommended to use the weight derivatives option. The biggest deviations from
the uncorrected results are to be expected if doing gradient calculations for elements heavier
than Kr using all electron basis sets and very small grids. To use the weight derivatives
option, add
weight derivatives
in $dft.
The option
point charges
in $drvopt switches on the evaluation of derivatives with respect to coordinates of point
charges. The gradients are written to the file $point_charge_gradients old gradients will
be overwritten.

22.2.14 Keywords for Module aoforce

This module calculates analytically harmonic vibrational frequencies within the HF- or
(RI)DFT-methods for closed-shell and spin-unrestricted open-shell-systems. Broken occu-
pation numbers would lead to results without any physical meaning. Note, that RI is only
used partially, which means that the resulting Hessian is only a (very good) approximation
to exact second derivatives of the RIDFT-energy expression. Apart from a standard force
constant calculation which predicts all (allowed and forbidden) vibrational transitions, it is
also possible to specify certain irreps for which the calculation has to be done exclusively or
to select only a small number of lowest eigenvalues (and eigenvectors) that are generated at
reduced computational cost.

General keywords

$drvopt
is the keyword for non-default options of gradient and second derivative calculations.
Possibilities in case of the module aoforce are:

frequency analysis only

analysis only
to read a complete Hessian from the input file $hessian and perform only the
frequency analysis
analysis [only] intcoord [print printlevel ]
to perform an analysis of normal modes in terms of internal coordinates. Details
about this option and the effect of the printlevel (default is 0) are given in
Section 15. The effect of the keyword only is the same as described above.
22.2. FORMAT OF KEYWORDS AND COMMENTS 431

$maxcor 50
fixes the amount of core memory to be used for dynamically allocated arrays (here 50
MiB), about 70% of available memory should be fine, because $maxcor specifies only
the memory used to store derivatives of density and Fock matrices as well as right
hande side vectors for the CPHF equations. For further details see subsection 22.2.2.

$forceconv 7
sets the convergence criterion for the CPHF-equations to a residual norm of 1.0d-
7. Normally the default value of 1.0d-5 already provides an accuracy of vibrational
frequencies of 0.01 cm−1 with respect to the values obtained for the convergence limit.

$forceiterlimit 10
fixes the maximum number of Davidson iterations for the solution of the CPHF-
equations to a value of ten. Normal calculations should not need more than eight
iterations, but as a precaution the default value is 25.

$nosalc
forces the program in case of molecules with C1 symmetry not to use 3N − 6(5) sym-
metry adapted but all 3N cartesian nuclear displacement vectors. This option may
lead to a moderate speed-up for molecules notedly larger than 1000 basis functions
and 100 atoms.

$noproj
forces the program not to project out translations and rotations when forming a
basis of symmetry adapted molecular displacements. This option may be needed if
a Hessian is required, that contains translation- and rotation-contributions, e.g. for
coupling the system with low cost methods. Output of the unprojected hessian is done
on $nprhessian; format is the same as for conventional $hessian. Output of the
corresponding eigenvalues and eigenvectors is done analogously on $nprvibrational
spectrum and $nprvibrational normal modes.

$nomw
causes the program to diagonalize a not mass weighted hessian. Output is on
$nprhessian, $nprvibrational spectrum and $nprvibrational normal modes,
because projection of rotations is not possible in this case.

$isosub
This keyword allows to trace back the effects of isotopic substitution on vibra-
tional frequencies. The atom(s) for which isotopic substitution is to be investigated
are specified in subsequent lines of the form (atom index) (mass in special isotope), e.g.

$isosub
3 2.001
5 13

The interpolation then takes place between the mass(es) specified in $atoms (or
the default mass(es), if none specified) and the mass(es) in $isosub. Take care of
432 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

symmetry equivalent atoms, otherwise symmetry analysis will fail. This feature can
not be used in a lowest eigenvalue search (keyword $les).

$isopts 6
Sets the number of points for interpolation between the two isotopes compared by
the $isosub option to six. Default value is 21.

Keywords for the treatment of only selected nuclear displacement vectors:

$ironly
CPHF-iteration is done only for distortions, that are IR active.

$ramanonly
CPHF-iteration is done only for distortions, that are Raman active.

$vcd
Calculation of VCD rotational strenghts after preceding mpshiftrun.

$les
This causes a lowest Hessian eigenvalue search to be performed instead of a complete
force constant calculation. The lowest eigenvalue search consists of the calculation
of a guess-Hessian and macro-iterations to find the solution vector(s) for the lowest
eigenvalue(s). In each macro-iteration the CPHF-equations are solved for the present
search vector(s). $les all 1 means that one lowest eigenvalue for each irrep will be
determined, other numbers of lowest eigenvalues per irrep are admissible too.
Different numbers of lowest eigenvalues for different irreps are requested by e.g.
$les
a1 3
a2 all
b2 1
The convergence criterion of the Davidson iterations for the solution of the CPHF-
equations as well as the maximal residual norm for the lowest Hessian eigenvalue in
the macro-iteration are specified by $forceconv as explained above.
The maximum number of macro-iterations is specified by $lesiterlimit x
with the default x=25. The maximum number of iterations for each solution of the
CPHF-equations is again determined by $forceiterlimit as shown above.
The convergence of the macro-iterations is strongly influenced by the size of the
starting search-subspace. Generally all guess-Hessian eigenvectors corresponding to
imaginary frequencies and at least two real ones are used as starting search-subspace.
However it proved to be necessary to use even more vectors in the case of guess-
Hessians with very large conditioning numbers.
$hesscond 8.0d-5
means that all eigenvalues with the quotient (eigenvalue)/(max. eigenvalue) lower
than 0.00008 are added to the starting search-subspace. Default is 1.0d-4.
22.2. FORMAT OF KEYWORDS AND COMMENTS 433

$hotfcht
Triggers the generation of input files for hotFCHT (program to calculate Franck-
Condon-factors by R. Berger and co-workers). See 15.5.

$sijuai_out
Save the derivative of the density matrix for subsequent use in the module evib. See
16

Force constant calculations on the DFT level prove to be numerically reliable only with
large integration grids or if one includes the effects of quadrature weights. This is done by
default—to prevent this, insert
no weight derivatives
in $dft.

22.2.15 Keywords for Module evib

$dfdxi textout
can be used to generate text output of the matrix elements of the derivative of the
Fock-operator. For bigger systems this can however generate very large output files.
See 16

22.2.16 Keywords for Module escf

TDHF and TDDFT calculations

To perform an escf calculation converged molecular orbitals from a HF, DFT or RIDFT
calculation are needed. The HF, DFT or RIDFT method is chosen according to the $dft
or $ridft keywords, specified above. It is recommended to use well-converged orbitals,
specifying $scfconv 7 and $denconv 1d-7 for the ground-state calculation. The input
for an escf calculation can be conveniently generated using the ex menu in define, see
Section 4.
During an escf run, a system-independent formatted logfile will be constructed for each
IRREP. It can be re-used in subsequent calculations (restart or extension of eigenspace
or of $rpaconv). An escf run can be interrupted by typing “touch stop” in the working
directory.
In an escf run one of the following properties can be calculated: (please note the ’or’ in the
text, do only one thing at a time.)
1. RPA and time-dependent DFT singlet or triplet or spin-unrestricted excitation energies
(HF+RI(DFT))

$scfinstab rpas or

$scfinstab rpat or
434 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$scfinstab urpa

2. TDA (for HF: CI singles) singlet or triplet or spin-unrestricted or spin-flip excitation

energies (HF+RI(DFT))

$scfinstab ciss or

$scfinstab cist or

$scfinstab ucis or

$scfinstab spinflip

3. Two-component TDDFT/TDA excitation energies of Kramers-restricted closed-shell sys-

tems

$scfinstab soghf or

$scfinstab tdasoghf

4. Eigenvalues of singlet or triplet or non-real stability matrices (HF+RI(DFT), RHF) or

complex stability matrices (HF+RI(DFT), Kramers-GHF)

$scfinstab singlet or

$scfinstab triplet or

$scfinstab non-real or

$scfinstab complex

5. Static polarizability and rotatory dispersion tensors (HF+(RI)DFT, RHF+UHF)

$scfinstab polly

6. Dynamic polarizability and rotatory dispersion tensors (HF+(RI)DFT, RHF+ UHF)

$scfinstab dynpol unit
list of frequencies
where unit can be eV, nm, rcm; default is a.u. (Hartree). For example, to calculate dynamic
polarizabilities at 590 nm and 400 i nm (i is the imaginary unit):

$scfinstab dynpol nm
590
400 i

The number and symmetry labels of the excited states to be calculated is controlled by the
data group $soes. Example:
22.2. FORMAT OF KEYWORDS AND COMMENTS 435

$soes
b1g 17
eu 23
t2g all

will yield the 17 lowest excitations in IRREP b1g, the 23 lowest excitations in IRREP eu,
and all excitations in IRREP t2g. Specify $soes alln; to calculate the n first excitations in
all IRREPS. If n is not specified, all excitations in all IRREPS will be obtained. Both static
and dynamic polarizabilites can be calculated with (local) all-electron relativistic methods
including the picture-change correction of the corresponding dipole moment in one and two-
component calculations. The picture-change correction is invoked by the keyword $pcc and
the two-component framework by $soghf.
Adding $damped_response will trigger a damped response calculations at the give frequen-
cies. Example:

$damped_response 0.25 eV

The damping factor must additionaly be specified in eV, cm-1 or nm.

7. Two-photon absorption

$scfinstab twophoton rpas or

$scfinstab twophoton urpa or

$scfinstab twophoton ciss or

$scfinstab twophoton ucis

8. Nuclear spin-spin coupling constants

The data group $ncoupling can be set using the ncoup section in define. A sample input
might look like this:

$ncoupling
fc sd pso dso nofcsdcross or
simple or
fromfile
reduced
thr=0.1
$nucsel 1,3,5-8
$nucsel2 "c","h"

You can specify the contributions to be calculated by indicating the appropriate abbrevia-
tion. If both fc and sd are specified, the FC/SD cross term is calculated, unless manually
switched off.
436 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

The simple method only calculates the FC and FC/SD cross terms, thus yielding the most
important contributions to the isotropic and anisotropic part. Be warned that this might
yield qualitatively wrong results in some cases! This method is quite fast because only
response equations for the FC term are solved. However, this means that it is incompatible
with $nucsel2.
If the option fromfile is set, SSCCs from the data group $coupling_reduced from a
previous calculation are used. This option can be used to obtain SSCC for a different
isotope without redoing the expensive part of the calculation.
If none of the keywords mentioned above is given, all terms (equivalent to fc sd pso dso)
are calculated.
By default, the output is multiplied by the gyromagnetic ratios, yielding J in Hertz. With
reduced, the reduced coupling constants K are printed in units of 1019 T2 /J. The content
of $coupling_reduced is not influenced by this setting.
In the final output, couplings where both the isotropic and the anisotropic part are smaller
than the threshold thr (in Hertz or 1019 T2 /J, default: 0.1) will not be printed.
You can specify for which atoms coupling constants shall be calculated by the data groups
$nucsel and $nucsel2. As shown above, the syntax is either a list of atomic indices
or of element symbols. The default is to calculate all atoms. Response equations (the
most expensive step) are solved for the atoms in $nucsel and right-hand side integrals are
calculated for atoms in any of the data groups. This means that a coupling tensor is obtained
if at least one of the partner is in $nucsel. Notice that the very same data group $nucsel
is used by the module mpshift.
It is furthermore recommended to set $rpaconv to at least 6.
The file referenced in $coupling_reduced (default file name: coupling.reduced) is suitable
for automated post-processing. It consists of the atomic indices, J iso , γK · γL , and the nine
elements of the matrix K̃ = J/γK γL (see eqs. 8.24 and 8.25 for an explanation of these
quantities). The gyromagnetic ratios γK are given in 107 rad s−1 T−1 . All couplings with
K ≥ L (ignoring $nucsel and thr) are printed.

general keywords

$rpacor n
The maximum amount of core memory to be allocated for the storage of trial vectors
is restricted to n MB. If the memory needed exceeds the threshold given by $rpacor,
a multiple pass algorithm will be used. However, especially for large cases, this will
increase computation time significantly. The default is 200 MB.

$spectrum unit
The calculated excitation energies and corresponding oscillator strengths are ap-
pended to a file named ’spectrum’. Possible values of unit are eV, nm and cm−1
22.2. FORMAT OF KEYWORDS AND COMMENTS 437

or rcm. If no unit is specified, excitation energies are given in a.u.

$cdspectrum unit
The calculated excitation energies and corresponding rotatory strengths are appended
to a file named ’cdspectrum’. unit can have the same values as in $spectrum.

$start vector generation e

Flag for generation of UHF start MOs in a triplet instability calculation. The option
will become effective only if there are triplet instabilities in the totally symmetric
IRREP. The optional real number e specifies the approximate second order energy
change in a.u. (default: 0.1).

$velocity gauge
Enables calculation of dipole polarizability/rotatory dispersion in the velocity gauge.
Active only for pure DFT (no HF exchange).
$sum rules unit
list of frequencies
Enable calculation of oscillator and rotatory strength sum rules at frequencies specified
by list of frequencies in unit unit (see $scfinstab dynpol). Note that the sums will
be taken only over the states specified in $soes.

$rpaconv n
The vectors are considered as converged if the Euclidean residual norm is less than
10−n . Larger values of n lead to higher accuracy. The default is a residual norm less
than 10−5 . For SSCCs, it is recommended to increase n to 6.

$escfiterlimit n
Sets the upper limit for the number of Davidson Iterations to n. Default is n = 25.

GW Keywords

$gw

The main keyword that switches on a GW calculation using full spectral representations.
This keyword will perform a standard G0 W0 calculation with default values for the other
flags.

$rigw

The main keyword that switches on a GW calculation using analytic continuation of the
self-energy from imaginary to real space.
There are several options which can be added to the $gw or $rigw keyword. The recom-
mended settings for a quick G0 W0 run are:
Full quasiparticle spectrum using spectral representations:
438 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$gw
rpa
eta 0.001

HOMO-LUMO gap using analytic continuation:

$rigw
rpa
ips+1
gap

Orbital 3-6 using contour deformation and iterative solution of the QP equation:

$rigw
rpa
eta 0.001
contour start=3 end=6
qpeiter 10

With the optional entries:

rpa

Default: false (not set). If added as option pure rpa response function is calculated. If
not added, the TDDFT response function is calculated and used to screen the coulomb
interaction. In combination with $rigw this keyword is mandatory, with $gw it should also
always be set and only be deactivated by experts. The gw menu in define always sets this
as default.

qpeiter <integer>

Default: 0. Switches between linearized quasiparticle equation for 0 and iterate quasiparticle
equation for > 0. eta is changed from 0.2 to the supplied value to obtain smooth convergence
during the iteration. Only with $gw.

gw0

Default: false (not set). A GW0 calculation is performed instead of G0 W0 . The number of
GW0 iterations performed is set by qpeiter.

evgw

Default: false (not set). An eigenvalue-only selfconsistent GW calculation is performed

instead of G0 W0 .
22.2. FORMAT OF KEYWORDS AND COMMENTS 439

scgw

Default: false (not set). A quasiparticle selfconsistent GW calculation is performed instead

of G0 W0 . Only with $gw and not compatible with two-component calculations.

offpq <real>

Default: 0.03. Infinitesimal complex energy shift in Fock-matrix contribution of self-energy

in scGW calculation. Only in combination with scgw keyword, ignored otherwise.

fdamp <real>

Default: 0.3. Damping of new Fock-matrix in scGW calculation. Only in combination with
scgw keyword, ignored otherwise.

unlimitz

Default: false (not set). In linearized G0 W0 the linearization Factor Z is allowed to take any
value instead of values between 0.5 - 1.0. Ignored if calculation is not a linearized G0 W0
calculation.

fixz

Default: false (not set). In linearized G0 W0 the linearization Factor Z is fixed to 1.0. Ignored
if calculation is not a linearized G0 W0 calculation.

csf <integer>

Default: false (not set). Calculate self-energy on an energy grid for the specified orbital.
Results are saved on file.

nl <integer>

Default: nocc +5. Number of orbitals to calculate gw for. It is set to the number of occupied
orbitals + 5 if set smaller than the number of occupied orbitals.

eta <real>

Default: 0.001. Infinitesimal complex energy shift. Negative value switches to calculating
at that value but extrapolating to 0 in linear approximation.

output <filename>
440 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

Default: qpenergies.dat. Output filename for the quasiparticle energies.

For $rigw there are some special keywords that control the cost and efficiency of the module.
The defaults chosen by define (or escfitself) are suitable for most systems. They were
selected rather to be rather tight to yield reliable results without much knowledge of the
system.

gap

Default: not set. Only the HOMO and LUMO quasiparticle states are calculated, all other
orbitals are shifted by the HOMO-LUMO gap. In conjunction with the "ips+<integer>"
keyword the N highest occupied and N lowest unoccupied states are calculated, making
$rigw calculations feasible for many systems.

ips+ <integer>

Default: not set. When $rigw is set then the the quasiparticle states of all occupied + the
N lowest virtual orbitals are calculated. Only with $rigw, ignored with $gw.

contour start=<integer> end=<integer> spin=<integer>

Default: not set. start, end and spin are optional and may be defined in arbitrary order.
start defines the starting point for the orbital range and end the ending of the orbital range
to be treated with RI-CD-GW . If neither start nor end are defined then the definition from
the ips+ and gap keywords is used. In open shell systems spin=1 calculates the self-energy
only for alpha shells, spin=2 for alpha and beta shells. For closed shell or any 2c calculation
spin=1 is always the correct option and does not need to be set by the user.

npade <integer>

Default: 128. Number of Pade approximants used in the $rigw module for RI-AC-GW .
Recommended to be the same as npoints in a dual-grid ansatz. Due to numerical instabilities
not more than 256 Pade approximants should be used!

npoints <integer>

Default: 128. Number of imaginary frequency integration points used in the $rigw module
for RI-AC-GW , RI-AC-GW and RPA energies. Recommended to be the same as npade in
a dual-grid ansatz in RI-AC-GW .

rpoints <integer>

Default: 16. Number of pseudo Fermi-levels in RI-AC-GW calculation. Not used if "con-
tour" keyword is present.
22.2. FORMAT OF KEYWORDS AND COMMENTS 441

rshift <real>

Default: 0.3 eV. Initial shift for first Fermi level when RI-AC-GW is used in eV. For in-
sulators 0.3 is recommended, for lower gap systems (< 3-4 eV HOMO-LUMO gap) 0.2 is
recommended. Not used if "contour" keyword is present.

width <real>

Default: rshift. stepwidth between subsequent Fermi levels when RI-AC-GW is used in eV.
Recommended to be set such that eHOM O + rshif t + rpoints ∗ width < eLU M O . Usually val-
ues from 0.05-0.20 eV are reasonable if the given condition is fulfilled. Not used if "contour"
keyword is present.

ccgrid

Default: false (not set). Use Clenshaw-Curtis instead of Gauss-Legendre grids. NOT rec-
ommended. If it is used npoints should be chosen much larger than the default.

BSE Keywords

$bse

The main keyword that switches on a BSE calculation. Provided that the response function
is calculated setting this keyword will perform a standard BSE calculation with default
values for the other flags.

$cbse

The main keyword that switches on a correlation augmented BSE (=cBSE) calculation. A
cBSE calculation with default values for the other flags will be performed.
The following optional entries can be added to the $bse or $cbse keyword:

noqpa
noqpw

Default: Not set. If set, the matrices A and W, respectively, are constructed using KS/HF
orbital energies instead of GW quasi-particles energies. if $cbse is used noqpw is set auto-
matically.

file <filename>

Default: qpenergies.dat. This option defines the file name of GW quasi-particle energies
which are read in.
442 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

iterative

Default: Not set. Compute the auxiliary matrix for the static screened interaction iteratively
instead of by Cholesky decomposition.

thrconv <real>

Default: 1.0d-12. Threshold for convergence in case of the iterative computation of the
static screened interaction.

iterlim <integer>

Default: 100. Maximum number of iterations in case of the iterative computation of the
static screened interaction.

$keep_fxc

Default: not set. Appears outside of the $bse data group. It adds the correlation part of
the underlying XC Kernel to the BSE response. Automatically set if $cbse is set.

22.2.17 Keywords for Module rirpa

The keyword $rirpa allows to specify the following options,

npoints n
Number of frequency quadrature points, n (default is 60).

maxitergrid n
Maximum number of nested Clenshaw–Curtis quadrature iterations. During each
iteration, the number of quadrature points doubles. The default is 1; namely, nested
quadrature rule is turned off.

gridtol 
Convergence criterion for the energy difference between successive nested quadrature
iterations. Because of the exponentially convergent Clenshaw–Curtis rule, the quadra-
ture error should be on the order of 2 once the nested rule converges. The default is
1d-4.

riaxk
Performs perturbative corrections to RPA such as AXK, ACSOSEX, or bare second-
order exchange; see [231]. The implementation is full parallelized using OpenMP.

acsosex
By default, the riaxk option triggers an RI-AXK calculation. Adding the acsosex
option triggers an RI-ACSOSEX calculation instead.
22.2. FORMAT OF KEYWORDS AND COMMENTS 443

acsox
trigers an bare second-order exchange correction instead of RI-AXK.

o4riaxk
employs the AO based O(N 4 ln N ) RI-AXK algorithm instead of the default MO
based O(N 5 ln N ) algorithm. The AO based algorithm only becomes more efficient
than the MO based algorithm for very large systems with small basis sets.

axktol 
basis shell quadruple screening threshold for the AO based RI-AXK algorithm.  is
defined according to equation (27) in [231]. The default value of  is 10−scfconv .

nohxx
HF energy calculation is skipped, (HXX = Hartree + eXact (Fock) eXchange). The
HXX energy computation in the rirpa module is not parallelized for now. For parallel
RI-RPA or RI-AXK energy calculations on large systems, it is recommended to use
the nohxx option for rirpa, and then compute the HXX energy separately using the
parallel version of ridft or dscf (by removing the $dft keyword block and setting
$scfiterlimit 1 in the control file).

rpaprof
Generates profiling output.

rpagrad
Switches on the gradients calculation for RI-RPA.

drimp2
Computes gradients in the direct RI-MP2 limit.

iter n
Number of GKS iterations, n (default is 0).

ldiis
Turns on DIIS algorithm for faster convergence of GKS iterations (default is off).

eigshift r
Adds a shift of r Hartree to the Kohn–Sham gap to aid in the convergence of difficult
small-gap systems (default is 0.0 Hartree).

output f ilename
Prints a condensed version of GKS-spRPA relevant output to the f ilename (default
filename is gksrpa.dat).

22.2.18 Keywords for Module egrad

egrad uses the same general keywords as escf and grad, see Sections 22.2.13 and 22.2.16.
The state to be optimized is by default the highest excited state specified in $soes. Note
that only one IRREP can be treated at the same time in contrast to escf calculations. When
the desired excited state is nearly degenerate with another state of the same symmetry it
444 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

may be necessary to include higher states in the initial calculation of the excitation energy
and vector in order to avoid root flipping. This is accomplished by means of the additional
keyword
$exopt n
which explicitly enforces that n-th excited state is optimized. n must not be larger than the
number of states specified in $soes.
$nacme
flag to compute Cartesian non-adiabatic coupling vectors between the excited state of in-
terest and the ground state [316]. This option requires the use of weight derivatives in
section dft. It is only implemented for C1 symmetry.

22.2.19 Keywords for Module mpgrad

If an MP2 run is to be performed after the SCF run, the SCF run has to be done with at
least

1) density convergence $denconv 1.d-7

2) energy convergence $scfconv 6

$maxcor n
The data group $maxcor adjusts the maximum size of core memory (n in MB) which
will be allocated during the MP2 run. Recommendation: 3/4 of the actual main
memory at most. For further details see subsection 22.2.2.

$mp2energy
Calculation of MP2 gradient is omitted, only MP2 energy is calculated.

$freeze
a1g 1-2
t1u 1

The data group $freeze specifies frozen orbitals, in the above syntax by irreducible
representations. The symmetry-independent and for standard-applications recom-
mended syntax is

$freeze
implicit core=5 virt=2

This will freeze the 5 lowest occupied and 2 highest virtual orbitals (alpha and beta
count as one in UHF cases). Note that degenerate orbitals count twice (e representa-
tions), thrice (t representations) etc. Note: In case of gradient calculations frozen core
orbitals are not regarded by mpgrad, moreover freezing of virtual orbitals is generally
not supported by mpgrad.
22.2. FORMAT OF KEYWORDS AND COMMENTS 445

All essential data groups for mpgrad may be generated by the preparation tool mp2prep,
apart from $maxcor (see above) these are the following:

$traloop n
specifies the number of loops (or ’passes’) over occupied orbitals, n, performed in the
mpgrad run: the more passes the smaller file space requirements—but CPU time will
go up.

$mointunit
type=intermed unit=61 size=0 file=halfint
type=1111 unit=62 size=0 file=moint#0
type=1112 unit=63 size=0 file=moint#1
type=1122 unit=64 size=0 file=moint#j
type=1212 unit=65 size=0 file=moint#k
type=1212a unit=70 size=0 file=moint#a
type=gamma#1 unit=71 size=0 file=gamma#1
type=gamma#2 unit=72 size=0 file=gamma#2
type=1212u unit=73 size=0 file=moint#u
type=1112u unit=74 size=0 file=moint#v
type=gamma#1u unit=75 size=0 file=gamma#1u

The data group $mointunit specifies:

• which scratch files are needed,

• where they are located (path name) and

• (after a statistics run, see below) an estimated file size.

$statistics mpgrad
statistics run (estimation of disc space needed) for the adjustment of the file sizes will
be performed.

$mp2pair
calculation of MP2 pair correlation energies.

22.2.20 Keywords for Module ricc2

Note that beside the keywords listed below the outcome of the ricc2 program also depends
on the settings of most thresholds that influence the integral screening (e.g. $denconv,
$scfconv, $scftol) and for the solution of Z vector equation with 4-index integrals (for
relaxed properties and gradients) on the settings for integrals storage in semi-direct SCF
runs (i.e. $thime, $thize, $scfintunit). For the explanation of these keywords see Sec-
tion 22.2.8.
446 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$cbas file=auxbasis
Auxiliary basis set for RI approximation.

$freeze
Freeze orbitals in the calculation of correlation and excitation energies. For details
see Section 22.2.19.

$printlevel 1
Print level. The default value is 1.

$tmpdir /work/thisjob
Specify a directory for large intermediate files (typically three-index coulomb integrals
and similar intermediates), which is different from the directory where the ricc2
program is started.

$maxcor n unit reference

The data group $maxcor adjusts the maximum size of core memory which will be
allocated during the ricc2 calculation. $maxcor has a large influence on computation
times. It is recommended to set $maxcor to ca. 75–85% of the available physical core
memory. For further details see subsection 22.2.2.

$spectrum unit
The calculated excitation energies and corresponding oscillator strengths are ap-
pended to a file named ’spectrum’. Possible values of unit are eV, nm and cm−1
or rcm. If no unit is specified, excitation energies are given in a.u.

$cdspectrum unit
The calculated excitation energies and corresponding rotatory strengths are appended
to a file named ’cdspectrum’. unit can have the same values as in $spectrum.

$laplace

conv = 5

The purpose of this data group is twofold: It activates the Laplace-transformed imple-
mentation of SOS-MP2 in the ricc2 module (if the sos option has been specified in
$ricc2) and it provides the options to specify the technical details for the numerical
Laplace-transformation.

conv
Threshold for the numerical integration used for the Laplace transformation
of orbital energy denominators. The grid points for the numerical integration
are determined such that is the remaining root mean squared error (RMSE) of
the Laplace transformation is < 10−conv . By default the threshold is set to the
value of conv given in $ricc2 (see below).

$ricc2

ccs
cis
22.2. FORMAT OF KEYWORDS AND COMMENTS 447

mp2 d1diag
cis(d) energy only
cis(dinf)
adc(2)
cc2
restart
norestart
hard_restart
nohard_restart
conv = 8
oconv = 7
lindep = 15
maxiter = 25
mxdiis = 10
maxred = 100
shift = 0.d0
iprint = 1
fmtprop = f15.8
geoopt model=cc2 state=(a" 2)
scs cos=1.2d0 css=0.3333d0
sos
gsonly
d1diag
intcorr

specifies the ab initio models (methods) for ground and excited states and the most
important parameters and thresholds for the solution of the cluster equations, linear
response equations or eigenvalue problems. If more than one model is given, the
corresponding calculations are performed successively. Note: The CCS ground state
energy is identical with the SCF reference energy, CCS excitation energies are identical
to CIS excitation energies. The MP2 results is equivalent to the result from the rimp2
module. cis(dinf) denotes the iterative CIS(D) variant CIS(D∞ ).

mp2 d1diag
Request the calculation of the D1 diagnostic in MP2 energy calculations (ig-
nored in MP2 gradient calculations). Note that the evaluation of the D1 di-
agnostic increases the computational costs of the RI-MP2 energy calculation
roughly by a factor of 3.
cis(d) energy only
If the energy only flag is given after the cis(d) keyword, it is assumed that
only excitation energies are requested. This switches on some shortcuts to avoid
the computation of intermediates needed e.g. for the generation of improved
start vectors for CC2.
(no)restart
448 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

If the restart flag is set, the program will try to restart the CC2 calcula-
tions from previous solution vectors on file. If the norestart flag is set no
restart will be done. Default is to do a restart for CC2 if and only if the file
CCR0--1--1---0 exists. Note: There is no restart possibility for CCS/CIS or
MP2/CIS(D).
(no)hard_restart
If the hard_restart flag is set, the program will try to reuse integrals and
intermediates from a previous calculation. This requires that the restart.cc
file has been kept, which contains check sums and some other information
needed. The hard_restart flag is switched on by default, if the restart.cc
file is present.
conv The conv parameter gives the convergence threshold for the ground state energy
for the iterative coupled-cluster methods as 10−conv . The default value is taken
from the data group $deneps.
oconv
The oconv parameter gives an additional threshold for the residual of the clus-
ter equations (vector function). If this parameter is given, the iterations for the
cluster equations are not stopped before the norm of the residual is < 10−oconv .
By default the threshold is set to oconv =conv−1, or 10× deneps if no input
for conv is given.
lindep
If the norm of a vector is smaller than 10−lindep , the vector is assumed to be
zero. This threshold is also used to test if a set of vectors is linear dependent.
The default threshold is 10−15 .
maxiter
gives the maximum number of iterations for the solution of the cluster equa-
tions, eigenvalue problems or response equations (default: 25).
mxdiis
is the maximum number of vectors used in the DIIS procedures for ground
state or excitation energies (default: 10).
maxred
the maximum dimension of the reduced space in the solution of linear equations
(default: 100).
iprint
print level, by default set to 1 or (if given) the the value of the $printlevel
data group.
fmtprop
Fortran print format used to print several results (in particular one-electron
properties and transition moments) to standard output.
geoopt
specify wavefunction and electronic state for which a geometry optimization is
22.2. FORMAT OF KEYWORDS AND COMMENTS 449

intended. For this model the gradient will be calculated and the energy and
gradient will be written onto the data groups $energy and $grad. Required
for geometry optimizations using the jobex script. Note, that in the present
version gradients are only available for ground states at the MP2 and CC2 and
for excited states at the CC2 level and not for ROHF based open-shell calcu-
lations. Not set by default. The default model is CC2, the default electronic
state the ground state. To obtain gradients for the lowest excited state (of those
included in the excitation energy calculation, but else of arbitrary multiplicity
and symmetry) the short cut s1 can be used. x is treated as synonym for the
ground state.
scs
the opposite–spin scaling factor cos and the same–spin scaling factor css can
be chosen. If scs is set without further input, the SCS parameters cos=6/5 and
css=1/3 are applied. This keyword can presently only be used in connection
with MP2.
sos
the SOS parameters cos=1.3 and css=0.0 are applied. This keyword can
presently only be used in connection with MP2.
d1diag
request the calculation of the D1 diagnostic for the ground state wavefunction.
Only needed for MP2 (see above for the alternative input option mp2 d1diag).
For all other correlated methods the D1 diagnostic is evaluated by default
(without significant extra costs).
intcorr
calculates the second-order corrections to the CCSD(T) energy from the interference-
corrected MP2-F12 (INT-MP2-F12) if $rir12 is switched on. It can be com-
bined either with the mp2 or the ccsd(t) methods. In the latter case, the
CCSD(T)-INT-F12 energy is printed. The intcorr all keyword writes on
the output all pair energies.

$rir12

ansatz
r12model
comaprox
cabs
examp
r12orb
pairenergy
corrfac
cabsingles
f12metric
450 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

ansatz char
char =1, 2* or 2
The ansatz flag determines which ansatz is used to calculate the RI-MP2-F12
ground state energy.
(Ansatz 2 is used if ansatz is absent.)
r12model char
char =A, A’ or B
The r12model flag determines which approximation model is used to calculate
the RI-MP2-F12 ground state energy.
(Ansatz B is used if r12model is absent.)
comaprox char
char =F+K or T+V
The comaprox flag determines the method used to approximate the commutator
integrals [T, f12 ].
(Approximation T+V is used if comaprox is absent.)
cabs char val
char =svd or cho
The cabs flag determines the method used to orthogonalize the orbitals of the
CABS basis. val is the threshold below which CABS orbitals are removed from
the calculation.
(svd 1.0d-08 is used if cabs is absent.)
examp char
char =noinv, fixed or inv with flip or noflip
The examp flag determines which methods are used to determine the F12 am-
plitudes. For inv the amplitudes are optimized using the orbital-invariant
method. For fixed and noinv only the diagonal amplitudes are non-zero and
are either predetermined using the coalescence conditions (fixed), or optimized
(noinv—not orbital invariant). If char =inv, the F12 energy contribution is
computed using all three methods. For open-shell calculations noflip supresses
the use of spin-flipped geminal functions.
(The fixed flip method is used if examp is absent.)
pairenergy char
char =off or on
If char =off (default), the print out of the standard and F12 contributions to
the pair energies is suppressed. The summary of the RI-MP2-F12 correlation
energies is always printed out.
corrfac char
char =LCG or R12
The corrfac flag determines which correlation factor is used for the gemi-
nal basis. LCG requires the data group $lcg, which contains the information
regarding exponents and coefficients of the linear combination of Gaussians.
22.2. FORMAT OF KEYWORDS AND COMMENTS 451

cabsingles char
char =off or on or *
The cabsingles flag determines whether or not the single excitations into the
CABS basis are computed. Neglect of the EBC (virtual-CABS) Fock matrix
elements is activated by *.
The default is to always compute the CABS singles correction if the CABS Fock
matrix elements are anyway available as a byproduct of the F12 calculation
(i.e., for ansatz 2 or r12model B or comaprox F+K).
r12orb char
char =hf, rohf, boys pipek or arb
The r12orb flag controls which orbitals are used for the F12 geminal basis func-
tions. With hf the (semi)-canonical Hartree–Fock orbitals are used (default).
For ROHF-based UMP2 calculations rohf orbitals can be used, which also
implies that the $freeze data group options refer to ROHF rather than semi-
canonical orbitals. For closed-shell species, localised orbitals can be used with
either the Boys or Pipek-Mezey method. For the non-(semi)-canonical options,
the r12orb noinv F12 energy correction is evaluated using active occupied or-
bitals transformed to the same basis as the orbitals in the geminal function. The
option arb uses the semi-canonicalised reference orbitals for the F12 correction,
but makes no assumption about whether or not these orbitals are converged
HF orbitals and is appropriate for computing F12 corrections using e.g. DFT
or Brueckner orbitals (Note, that the data group $non-canonical MOs should
be set in this case).
no_f12metric
f12metric
If no_f12metric is selected the coulomb metric is used in the density fitting
2
scheme to calculate the four index integrals over the operators f12 , f12 g12 , f12
and f12 r12 . If f12metric is selected the operator’s own metric is used. The
default for the ricc2 program is no_f12metric, while the pnoccsd program
can only be used with f12metric, where it is therefore the default.

$excitations

irrep=au multiplicity=1 nexc=4 npre=6 nstart=8

irrep=bg multiplicity=3 nexc=2 npre=4 nstart=5
spectrum states=all operators=diplen,dipvel
tmexc istates=all fstates=all operators=diplen,dipvel
exprop states=all operators=qudlen
twophoton states=all operators=(diplen,diplen) freq=0.1d0
momdrv states=all operators(diplen,soc) freq=0.0d0
xgrad states=(ag{3} 1)
conv = 6
thrdiis = 2
452 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

preopt = 3
leftopt
bothsides
oldnorm

In this data group you have to give additional input for calculations on excited states:

irrep
the irreducible representation.
multiplicity
spin multiplicity (1 for singlet, 3 for triplet); default: singlet, not needed for
UHF.
nexc the number of excited states to be calculated within this irrep and for this
multiplicity.
npre the number of roots used in preoptimization steps (default: npre = nexc).
nstart
the number of start vectors generated or read from file (default: nstart =
npre).
spectrum
This flag switches on the calculation of oscillator strengths for excited state—
ground state transitions. Setting the parameter states=all is mandatory for
the calculation of transition properties in the present version. The operators
flag can be followed by a list of operators (see below) for which the transition
properties will be calculated. Default is to compute the oscillator strengths for
all components of the dipole operator.
tmexc
This flag switches on the calculation of oscillator strengths for excited state—
excited state transitions. Specifying the initial and final states via istates=all
and fstates=all is mandatory for the calculation of transition properties in
the present version. The operators flag can be followed by a list of operators
(see below) for which the transition properties will be calculated. Default is to
compute the oscillator strengths for all components of the dipole operator.
twophoton
request the calculation of two-photon transition moments between ground and
excited states. It recognizes the optional suboptions states for specifying
the states for which the two-photon transition moments should be computed,
operators for specifying a list of pairs of one-electron transition operators,
and freq for a list of frequencies for one of the photons. If these suboptions
are specified, two-photon transition moments are computed per default for all
requested singlet states, dipole operators (in the length representation) and
photon energies which are equal 1/2 of the transition energies.
momdrv
request the calculation of derivatives of transition moments between ground
22.2. FORMAT OF KEYWORDS AND COMMENTS 453

and excited states. It recognizes the optional suboptions states for specifying
the states for which the two-photon transition moments should be computed,
operators for specifying a list of pairs of one-electron transition operators.
For phosphorescence lifetimes operators should be set to (diplen,soc) and
freq to 0.0d0.
exprop
requests the calculation of first-order properties for excited states. For the
states option see spectrum option above; for details for the operators input
see below.
xgrad
request calculation of the gradient for the total energy of an excited state. If
no state is specified, the gradient will be calculated for the lowest excited state
included in the calculation of excitation energies. The simultaneous calculation
of gradients for several state is possible.
conv convergence threshold for norm of residual vectors in eigen-
value problems is set to 10−conv . If not given, a default
value is used, which is chosen as max(10−conv , 10−oconv , 10−6 ),
where conv refers to the values given in the data group $ricc2.
preopt
convergence threshold used for preoptimization of CC2 eigenvectors is set to
10−preopt (default: 3).
thrdiis
threshold (10−thrdiis ) for residual norm below which DIIS extrapolation is
switched on in the modified Davidson algorithm for the non-linear CC2 eigen-
value problem (default: 2).
leftopt
If the flag leftopt is set the left eigenvectors are computed (default is to
compute the right eigenvectors, for test purposes only).
bothsides
The bothsides flag enforces the calculation of both, the left and the right
eigenvectors (for test purposes only).
bothsides
The oldnorm flag switches the program to the old normalization of the eigen-
vectors and %T1 and %T2 diagnostics which were identical with those used in
the CC response code of the Dalton program.

$response

fop unrelaxed_only operators=diplen

sop operators=(diplen,diplen) freq=0.077d0
gradient
conv = 6
454 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

zconv = 6
semicano
nosemicano
thrsemi = 3

In this data group you have to give additional input for the calculation of ground
state properties and the solution of response equations:

fop This flag switches on the calculation of ground state first-order properties (ex-
pectation values). The operators flag can be followed by a list of operators
(see below) for which the first-order properties will be calculated. Default is
to compute the components of the dipole and the quadrupole moment. The
option unrelaxed_only suppress the calculation of orbital-relaxed first-order
properties, which require solution the CPHF-like Z-vector equations. Default
is the calculation of unrelaxed and orbital-relaxed first-order properties. The
unrelaxed_only option will be ignored, if the calculation of gradients is re-
quested (see gradient option below and geoopt in data group $ricc2).
sop requests the calculation of ground state second-order properties as e.g. dipole
polarizabilities. The operators flag has to be followed by a comma seper-
ated pair of operators. (If more pairs are needed they have to be given with
additional sop commands.) Default is to compute all symmetry-allowed ele-
ments of the dipole-dipole polarizability. With the freq flag on can specify
a frequency (default is to compute static polarizabilities). The relaxed flag
switched from the unrelaxed approach, which is used by default, to the orbital-
relaxed approach. Note that the orbital-relaxed approach can not only be used
in the static limit (freq=0.0d0). For further restrictions for the computation
of second-order properties check Chapter 10.5.
gradient
require calculation of geometric gradients. In difference to the geoopt keyword
in the data group $ricc2 this can be used to compute gradients for several
methods within a loop over models; but gradients and energies will not be
written to the data groups $grad and $energy as needed for geometry opti-
mizations. Note, that in the present version gradients are only available for
MP2 and CC2 and only for a closed-shell RHF reference.
conv convergence threshold for norm of residual vectors in linear response equations
is set to 10−conv . If not given in the $response data group, a default value is
used, which is chosen as max(10−conv ,
10−oconv , 10−6 ), where conv and oconv refer to the values given in the data
group $ricc2.
zconv
convergence threshold for the norm of the residual vector in the solution of the
Z vector equations will be set to 10−zconv .
semicano
use semi-canonical formulation for the calculation of (transition) one-electron
22.2. FORMAT OF KEYWORDS AND COMMENTS 455

densities. Switched on by default. The semi-canonical formulation is usually

computationally more efficient than the non-canonical formulation. Exceptions
are systems with many nearly degenerate pairs of occupied orbitals, which have
to be treated in a non-canonical way anyway. (See also explanation for thrsemi
below).
nosemicano
use non-canonical formulation for the calculation of (transition) one-electron
densities. Default is to use the semi-canonical formulation.
thrsemi
the threshold for the selection of nearly degenerate pairs of occupied orbitals
which (if contributing to the density) have to be treated in a non-canonical
fashion will be set to 10−thrsemi . If set to tight the semi-canonical algorithm
will become inefficient, if the threshold is to large the algorithm will become
numerical unstable
zpreopt
threshold for preoptimizating the so-called Z vector (i.e. the Lagrangian mul-
tipliers for orbital coefficients) with a preceding RI-CPHF calculation with the
cbas auxiliary basis. The RI-CPHF equations will be converged to a residual
error < 10−zpreopt . Default is zpreopt=4. This preoptimization can reduce
significantly the computational costs for the solution of the Z vector equa-
tions for large basis sets, in particular if they contain diffuse basis functions.
For calculations on large molecules with small or medium sized basis sets the
preoptimization becomes inefficient compared to the large effects of integral
screening for the conventional CPHF equations and should be disabled. This
option is automatically disabled for ricc2 calculations based on foregoing RI-
JK Hartree-Fock calculation.
nozpreopt
disable the preoptimization of the Z vector by a preceding RI-CPHF calcula-
tion with the cbas basis set. (Note that the preoptimization is automatically
deactivated if the ricc2 calculation is based on a foregoing RI-JK Hartree-Fock
calculation.)

Common options for keywords in the data groups $ricc2, $response, and $excitations:

operators=diplen,dipvel
input of operator labels for first-order properties, transition moments, etc. Currently
implemented operators/labels are

overlap overlap (charge) operator: the integrals evaluated in the AO basis are
hµ|νi
diplen dipole operator in length gauge: hµ|riO |νi with i = x, y, z; the index
O indicates dependency on the origin (for expectation values of charged
molecules), which in the present version is fixed to (0, 0, 0)
456 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

(all three components, individual components can be specified with the

labels xdiplen, ydiplen, zdiplen).
dipvel dipole operator in velocity gauge: hµ|∇i |νi
(all three components, individual components can be specified with the
labels xdipvel, ydipvel, zdipvel).
qudlen quadrupole operator hµ|riO rjO |νi
(all six components, individual components can be specified with the
labels xxqudlen, xyqudlen, xzqudlen, yyqudlen, yzqudlen, zzqudlen).
If all six components are present, the program will automatically give
the electronic second moment tensor (which involves only the electronic
contributions) Mij , the isotropic second moment α = 13 trM and the
anisotropy
v
u z z
u1 X X
β= t 2
(Mii − Mi+1,i+1 ) + 3 2
Mi,i+1 .
2 i=x i=x

Furthermore the traceless quadrupole moment

1
Θij = h3ri rj − r2 δij i
2
(including nuclear contributions) is given.
angmom angular momentum hµ|LO i |νi
(all three components, individual components can be specified with the
labels xangmom, yangmom, zangmom).
Z rI
nef electronic force on nuclei hµ| rII 3i |νi, where ZI is the charge of the nucleus
I and rI is the position vector of the electron relative to the nucleus
(all three components for all nuclei: the labels are xnef001, ynef001,
znef001, xnef002, etc. where the number depends on the order in the
coord file).

states=all
specification of states for which transition moments or first-order properties are to be
calculated. The default is all, i.e. the calculations will be done for all excited states for
which excitation energies have been calculated. Alternatively, one can select a subset
of these listed in parentheses, e.g. states=(ag{3} 1,3-5; b1u{1} 1-3; b2u4). This
will select the triplet ag states no. 1, 3, 4, 5 and the singlet b1u states no. 1, 2, 3 and
the singlet (which is default if no {} is found) b2u state no. 4.

istates=all fstates=all
The specification of initial and final states for transition properties between excited
states is mandatory. The syntax is analog to the states option, i.e. either all or a
list of of states is required.

$D2-diagnostic
Calculate the double-substitution-based diagnostics D2 .
22.2. FORMAT OF KEYWORDS AND COMMENTS 457

$cc2_natocc
Write MP2/CC2 natural occupation numbers and natural orbitals to a file.

$cgrad 1000
Calculate the error functional δRI for the RI approximation of (ai|bj) integrals

1 |hab||ijiexact − hab||ijiRI |2
δRI =
4 a − i + b − j

and its gradients with respect to exponents and coefficients of the auxiliary basis set
as specified in the data group $cbas. The results are written to $egrad scaled by the
factor given with the keyword $cgrad and can be used to optimize auxiliary basis
sets for RI-MP2 and RI-CC2 calculations (see Section 1.5).

22.2.21 Keywords for Module ccsdf12

The ccsdf12 program uses a subset of the data groups of the ricc2 program. For F12
calculations it uses in addition the data groups $rir12, $lcg, $cabs, and $jkbas. See the
respective sections for further details.
The ccsdf12 program recognizes in the $ricc2 data group the following options to specify
wavefunction methods:

$ricc2

ccsd
mp3
mp4
ccsd(t)
bccd(t)
rohf-bccd(t)
shift 0.0
core can/res/bru
no_sc

The options mp3, mp4, and ccsd request, respectively, MP3, MP4 and CCSD calcula-
tions. The option ccsd(t) request a CCSD calculation with the perturbative triples
correction, CCSD(T), and as a side result also the CCSD[T] energy will be printed.
The option bccd requests a BCCD calculation, which is UBCCD for open-shell sys-
tems by default. The option rohf-bccd requests a ROHF-BCCD calculation, where
the alpha and beta Brueckner orbitals are restricted to be the same for the doubly
occupied orbitals of an ROHF open shell initial reference state.

shift
sets a level shift (> 0) for the amplitude update in the CC iterations, useful to
improve convergence for CCSD or BCCD calculations on e.g. transition metal
complexes.
458 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

core
specifies the nature of the core orbitals in a frozen core calculation as either the
(semi-)canonicalised reference mos (can) or the unmodified reference mos (res)
or Brueckner core orbitals (bru) that are relaxed during a BCCD calculation.
no_sc
suppresses the default semi-canonicalisation of a ROHF reference prior to a
CCSD or BCCD calculation, such that the output amplitudes and pair energies
correspond to the restricted orbitals. Note that a semi-canonicalisation step is
always performed immediately before the evaluation of the (T) energy.

In the data group $rir12 it recognizes ccsdapprox as additional option:

ccsdapprox label
defines the approximation to CCSD-F12 which will be used if the MP2-F12 calculation
is followed by a CCSD or CCSD(T) calculation. The available approximation and
corresponding labels are

CCSD(F12) ccsd(f12)
CCSD(F12*) ccsd(f12*)
CCSD[F12] ccsd[f12]
CCSD-F12b ccsd-f12b
CCSD(2*)F12 ccsd(2)_/f12
CCSD(2)F12 ccsd(2)_/f12
CCSD(F12∗) ccsd_(f12*)

It is recommended that these approximations are only used in combination with ansatz
2 and the SP approach (i.e. geminal coefficients fixed by the cusp conditions). For
CCSD-F12b calculations also the CCSD-F12a energies are calculated as a byprod-
uct. By default a CCSD(F12*) calculation is carried out, because it gives the best
cost/accuracy ration among the approximations. CCSD(F12∗) is the perturbative vari-
ant of CCSD(F12*) and should be used when performing a Brueckner calculation.

22.2.22 Keywords for Module pnoccsd

Note that beside the keywords listed below the outcome of the pnoccsd program also depends
on the settings of most thresholds that influence the integral screening (e.g. $denconv,
$scfconv, $scftol). For the explanation of these keywords see Section 22.2.8.

$cbas file=auxbasis
Auxiliary basis set for RI approximation.

$freeze
Freeze orbitals in the calculation of correlation and excitation energies. For details
see Section 22.2.19.
22.2. FORMAT OF KEYWORDS AND COMMENTS 459

$printlevel 1
Print level. The default value is 1.

$tmpdir /work/thisjob
Specify a directory for large intermediate files (typically three-index coulomb integrals
and similar intermediates), which is different from the directory where the program
is started.

$maxcor n unit reference

The data group $maxcor adjusts the maximum size of core memory which will be
allocated during the pnoccsd run. $maxcor has a large influence on computation
times. It is recommended to set $maxcor to ca. 75–85% of the available physical core
memory. For further details see subsection 22.2.2.

$laplace

conv = 1

Only needed for the (T) perturbative triples correction; for other methods only for test
purposes. It sets the accuracy for the numerical Laplace-transformation to 10−conv
used for the (T) correction and the iterative OSV generation. For the (T0) cor-
rection and methods without triples his threshold is only used for the approximate
doubles amplitudes from which the OSVs are computed when the iterative algorithm
is enabled. The default value is 10−1 for the OSV generation and 10−2 for the (T)
correction. If specified the input value will currently be used in both cases.

$pnoccsd

mp2
localize ibo
osvmode paos
paos tnos
mxrdim 800
tolpno= 1.00E-7
tolosv= 1.00E-9
tolri= 1.21E-3
tolpair= 4.64E-6
opnos on
tolcapno= 1.00E-9 3.16E-8
tolosc= 1.00E-10 3.15E-9
tolopno= 1.00E-9
toloso= 1.00E-10
conv = 7
oconv = 3
lindep = 12
maxiter = 25
mxdiis = 10
460 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

maxred = 100
scs cos=1.2d0 css=0.3333d0
sos

mp2 specifies the ab initio model (method). The current release version is restriced
to MP2, CCSD, CCSD(T0) and CCSD(T). MP2 is the default model.
localize specifies the localization method; possible choices are boys for Foster-Boys,
pm for Pipek-Mezey, ibo for intrinsic bond orbitals (IBOs) and none for canon-
ical (deprecated, only meant for testing) orbitals. Default are (since V7.3)
IBOs, which are recommended if e.g. for aromatic systems the seperation of σ-
and π-type orbitals is important. (Pipek-Mezey orbitals become very delocal
with diffuse basis sets.)
osvmode switch between to algorithms for the generation of OSV coefficients. Possi-
ble choices are full for an O(N 4 ) scaling direct diagonalization (cmp. [317]),
davidson for an O(N 3 ) scaling iterative diagonalization (cmp. [23]) and paos
for sub-O(N 2 ) scaling implementation. paos is the default choice and rec-
ommended but is not currently compatible with gradients, excited states or
explicit correlation, for which cases davidson is the default.
mxrdim the maximal dimension of trial vectors in the iterative OSV generation (osvmode
davidson). For PNO-MP2 the default is 800. The dimension is bounded by the
number of active virtual orbitals and except for small systems a much smaller
value as the number of virutals is sufficient. Smaller dimensions increase the
performance, but than the iterative scheme might not converge and the pro-
gram must be restarted with an adjusted dimension. For PNO-MP2-F12 the
Default is 4000 since a larger reduced space is required to construct the OSX
(cmp. [317]). Usually there is no need to touch this parameter.
tolpno specifies the PNO truncation threshold. Default value: 10−7 .
tolosv specifies the OSV truncation threshold. If no given tolosv is set set such
that the OSV truncation error is 10 times smaller than the PNO truncation
error.
toltno specifies the TNO truncation threshold. If no given toltno is set set equal
to tolpno.
paos (de)-activates PAO selection at OSV, PNO and TNO steps. Possible choices
are off, osv, pno, tn0 and tno, which turn on PAO selection at the specified
level and all earlier levels. If PAOs are used and no PAO selection is performed,
the domains are constructed by merging the domains one level below (eg pair
domains are formed by merging OSV domains). The default is tno, which
activates PAO selection at every step and is recommended.
tolopao specifies the PAO truncation threshold for OSVs, the default is linked to
tolpno.
tolppao specifies the PAO truncation threshold for PNOs, the default is linked to
tolpno.
22.2. FORMAT OF KEYWORDS AND COMMENTS 461

tolt0pao specifies the PAO truncation threshold for T0-TNOs, the default is linked
to toltno.
toltpao specifies the PAO truncation threshold for (T)-TNOs, the default is linked
to toltno.
tolri specifies the threshold for selecting orbital and pair-specific auxiliary basis sets
√
for the local RI approximation. If not given tolri is set to 107/12 × tolpno,
which is the recommended value.
tolpair specifies the energy threshold for selecting the significant pairs. If not given
tolpair is set to (0.1 × tolpno)2/3
toltrip specifies the energy threshold for selecting the significant triples. If not
given toltrip is set equal to tolpair.
opnos enables (on) or disables (off) for F12 calculations the use of OPNOs for the oc-
cupied orbital spaces in the projectors for the three-electron integrals. Default:
on
tolcapno sets for F12 calculations the truncation thresholds for the complementary
auxiliary PNOs for the virtual spaces (CAPNOs) in the projectors for the
three-electron integrals. If not specified, default values are calculated from the
threshold tolpno.
tolosc sets for F12 calculations the truncation thresholds for the orbital specific
complementary auxiliary virtuals from which the CAPNOs are generated. If
not specified, default values are chosen as 0.1 × the tolcapno thresholds, which
is the recommended choice.
tolopno sets for F12 calculations the truncation thresholds for the selection of PNOs
for the occupied orbital spaces (OPNOs) in the projectors for the three-electron
integrals. If not specified, default values are calculated from the threshold
tolpno.
toloso set for F12 calculations the truncation thresholds for the orbital specific aux-
iliary occupied orbitals from which the OPNOs are generated. If not specified,
default values are chosen as 0.1 × the tolopno threshold, which is the recom-
mended choice.
conv The conv parameter gives the convergence threshold for the ground state energy
as 10−conv . The default threshold is 10−7 .
oconv The oconv parameter gives an additional threshold for the residual of the
ground state equations as < 10−oconv . The default threshold is 10−3 .
lindep If the norm of a vector is smaller than 10−lindep , the vector is assumed to be
zero. This threshold is also used to test if a set of pre-PNOs is linear dependent.
The default threshold is 10−10 if PAOs are used and 10−12 otherwise.
maxiter gives the maximum number of iterations for the solution of the cluster equa-
tions, eigenvalue problems or response equations (default: 25).
mxdiis is the maximum number of vectors used in the DIIS procedures for the ground
state equations (default: 10).
462 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

maxred not used in the current release.

scs the opposite–spin scaling factor cos and the same–spin scaling factor css can
be chosen. If scs is set without further input, the SCS parameters cos=6/5
and css=1/3 are applied.
sos the SOS parameters cos=1.3 and css=0.0 are applied.

$rir12 for the description of this data group see Sec. 22.2.20.

22.2.23 Keywords for Module relax

$optimize options
define what kind of nonlinear parameters are to be optimized by relax and specify
some control variables for parameter update.
Available options are:

internal on/off
optimize molecular structures in the space of internal coordinates using defi-
nitions of internal coordinates given in $intdef as described in Section 4.1 (
default: on).
redundant on/off
optimize molecular structures in redundant internal coordinates using defini-
tions of redundant internal coordinates given in $redundant. For an optimiza-
tion in redundant internal coordinates option internal has to be switched on
too, and option cartesian has to be switched off (default: on).
Cartesian on/off
optimize molecular structures in the space of (symmetry-distinct) Cartesian
coordinates (default: off).
basis on/off suboptions
optimize basis set exponents (default=off).
Available suboptions are:
logarithm
exponents of uncontracted basis functions will be optimized after con-
version into their logarithms (this improves the condition of the approx-
imate force constant matrix obtained by variable metric methods and
the behavior of the optimization procedure); scale factors of contracted
basis functions will not be affected by the logarithm suboption
scale
ALL basis set exponents will be optimized as scale factors (i.e. contracted
blocks and single functions will be treated in the same way); if both
suboptions (scale and logarithm) are given the logarithms of the scale
factors will be optimized
22.2. FORMAT OF KEYWORDS AND COMMENTS 463

global on/off
optimize a global scaling factor for all basis set exponents (default: off).
NOTES: • basis and global have to be used exclusively!

• if $optimize has been specified but $forceapprox is absent, the

option $forceinit on is switched on by default.

• specification of the option $interconversion on will override

$optimize!

$coordinateupdate options
define some variables controlling the update of coordinates.
Available options are:

dqmax real
maximum allowed total change for update of coordinates. The maximum
change of individual coordinate will be limited to dqmax /2 and the collective
change dq will be damped by dqmax /hdq | dqi if hdq | dqi > dqmax q
(default: 0.3)
interpolate on/off
calculate geometry update by inter/extrapolation of geometries of the last two
cycles (the interpolate option is always switched on by default, but it is only
active ANY time if steepest descent update has been chosen, i.e. $forceupdate
method=none; otherwise it will only be activated if the DIIS update for the
geometry is expected to fail)
statistics on/integer /off
provide a statistics output in each optimization cycle by displaying all (the last
integer, default setting by define is 5) subsequent coordinates, gradient and
energy values (default: on).

$gdiishistory file=char
the presence of this keyword forces relax to provide informational output about the
usage of DIIS for the update of the molecular geometry.

$interconversion options default=off

special input related to the transformation of atomic coordinates between cartesian
and internal coordinate spaces (default: off).
Available options are:

maxiter=n
maximum number of iterations for the iterative conversion procedure internal
→ cartesian coordinates (default: 25).
qconv
convergence criterion for the coordinate conversion (default: 1.d-10).
on/off options
this switch activates special tasks: transform coordinates/gradients/ hessians
464 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

between spaces of internal/cartesian coordinates using the definitions of inter-

nal coordinates given in $intdef:

available suboptions are:

cartesian –> internal coordinate gradient hessian
cartesian <– internal coordinatethe direction of the transformation is in-
dicated by the direction of the arrow

Note: specification of $interconversion on will override $optimize!

$forceupdate method options

this data group defines both the method for updating the approximate force constant
matrix and some control variables needed for the force constant update.

Options for method:

none no update (steepest descent)

ms suboptions Murtagh–Sargent update
dfp suboptions Davidon–Fletcher–Powell update
bfgs suboptions Broyden–Fletcher–Goldfarb–Shanno update
dfp-bfgs suboptions combined (bfgs+dfp) update
schlegel suboptions Schlegel update
ahlrichs suboptions Ahlrichs update (macro option)

suboptions if method=ms, dfp, bfgs, schlegel, ahlrichs

numgeo=integer number of structures used
maxgeo=integer maximum number of geometries (= rank of the up-
date procedure, for ahlrichs only)
ingeo=integer minimum number of geometries needed to start up-
date

if method =ms, dfp, bfgs:

maxgeo=2, mingeo=1 as default

additional suboptions if method=ahlrichs

modus= char fmode for an explanation see suboptions pulay given be-
low e.g. ahlrichs numgeo=7 mingeo=3 maxgeo=4
modus=<g|dg> dynamic
22.2. FORMAT OF KEYWORDS AND COMMENTS 465

NOTES: if the macro option ahlrichs has been chosen and

n=numgeo, ncycl=‘number of geometries available’

• if ncycl < n: geometry update by inter/extrapolation

using the last two geometries
• if ncycl ≥ n: diagonal update for the hessian by
least mean squares fit; pulay update for the geometry
(using specified modus, fmode (see pulay below))
• if (ncycl ≥ max(5, n + 3) and max(| g |) < 0.01 and
ḡ < 0.001 ) or Hij 6= 0 ∀ i 6= j : diagonal update is
replaced by multidimensional BFGS (rank n) update
for the hessian

pulay suboptions
try to find an optimal linear combination of the coordinates of the numpul previous
optimization cycles as specified by modus (see below).

Available suboptions are:

numpul=integer
number of geometries to be utilized
maxpul=integer
maximum number of geometries
minpul=integer
minimum number of geometries needed to start update
modus=char fmode
char =<g|g> or <g|dq> or <dq|dq> defines the quantity to be minimized (dq
= internal coordinate change).
fmode specifies the force constants to be used (only if char =<g|dq> or <dq|dq>!)
fmode=static: use static force constants
fmode=dynamic: use updated force constants
fail=real
real defines the threshold for the quantity g ∗ dq/|g| ∗ |dq| which defines the
angle between gradient vector and coordinate change (default: 0.1). If pulay is
used in connection with a multidimensional BFGS update for the hessian than
g·dq
the default is real =0.0. If |g|∗|dq| > −real the pulay update for the geometry
is expected to fail and will be ignored. For example:
pulay numpul=4 maxpul=4 minpul=3 modus=<dq|dq> static fail=0.2

options for $forceupdate

diagonal
update only the diagonal force constants (update for off-diagonals will be sup-
pressed) (only active if method =ms, dfp, bfgs)
466 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

offdamp real
this allows to damp off-diagonal force constants by 1/real (compare offreset,
which discards off-diagonals completely). Only values > 1.0 will be accepted.
This option is active only within one relax run and will be disabled automati-
cally by relax. This is useful in difficult cases, where the non-diagonal update
has lead to too large non-diagonal elements of the hessian.
offreset
reset off-diagonal force constants to zero. This option will be active for the
current optimization cycle only, i.e. it will be removed by relax after having
discarded off-diagonals!
allow=real
optimization cycle specification of a maximum energy change allowed (given in
mHartree) which will be accepted using the actual approximate force constant
matrix from $forceapprox; if this energy change will be exceeded, the force
constants will be scaled appropriately
(The default: 0.0 means NO action)
scale=real
scaling factor for the input hessian (default: 1.0).
threig=real
lower bound for eigenvalues of the approximate hessian (default: 0.005); if any
eigenvalue drops below threig, it will be shifted to a reasonable value defined
by:
reseig=realdefault: texttt0.005.
thrbig=real
upper bound for eigenvalues of the hessian; if any eigenvalue exceeds thrbig,
it will limited to this value (default: 1000.0).
damping=real
damp the variable metric update for the hessian by 1/(1+ real ) (default: 0.0).

$forceinit option
specify initialization of the (approximate) force constant matrix.
Available options are:

on/off
this activates or deactivates initialization; if on has been set, relax will pro-
vide an initial force constant matrix as specified by one of the possible ini-
tialization options as described below and will store this matrix in data group
$forceapprox; after initialization relax resets $forceinit to off!
diag=suboptions
provide a diagonal force constant matrix with:
available suboptions are:
22.2. FORMAT OF KEYWORDS AND COMMENTS 467

real
this will lead to an assignment of diagonal elements (default: 1.0)).
default
this will lead to an assignment of initial force constant diagonals depend-
ing on the coordinate type.
individual
Provide individual defined force constant diagonals for
• internal coordinates (supplied in $intdef ... fdiag=..)
• a global scale factor ( $global ... fdiag=..)
This does not work for basis set optimization. For the correct syntax of
‘fdiag=..’ see descriptions of $intdef, $global
carthess
read a cartesian (e.g. analytical) hessian from $hessian and use it as a
start force constant matrix; if $optimize internal has been set: use
its transform in internal coordinate space. If large molecules are to be
optimized, it may be necessary (large core memory requirements!) to
deactivate the numerical evaluation of the derivative of the B-matrix
with respect to cartesian coordinates, which is needed to transform
H(cart) → H(int) exactly by specifying no dbdx.

$last SCF energy change = real

$last MP2 energy change = real

These keywords depend on the optimization task to be processed and are updated by
the corresponding program (i. g. SCF energy).

$m-matrix options
This data block contains non-default specifications for the m-matrix diagonals. This
is of use if some Cartesian atomic coordinates shall be kept fixed during optimization.
Available options are:

integer real real real

atomic index followed by diagonal elements of the m-matrix for this atom

$scratch files
The scratch file ftmp allocated by relax can be placed anywhere in your file systems
instead of the working directory by referencing its path name in this data group as
follows:

$scratch files
relax ftmp path/file

The first column specifies the program, the second column the scratch file and the
third column the path name of the file to be used as scratch file.
468 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

Input Data Blocks Needed by Relax

$intdef or $redundant
Definitions of internal coordinates and, optionally, values of internal coordinates
(val=..., given in a.u. or degrees) or force constant diagonal elements (fdiag=...).

$grad
Cartesian coordinates and gradients calculated in subsequent optimization cycles.
Entries are accumulated by one of the gradient programs (grad, mpgrad, rimp2,
ricc2, egrad, etc.).

$egrad
Basis set exponents scale factors and their gradients as calculated in subsequent op-
timization cycles. Entries are accumulated by one of the gradient programs.

$globgrad
Global scale factors and gradients as calculated in subsequent optimization cycles.
Entries are accumulated by the grad or aoforce program.

$corrgrad
Allows to augment internal SCF gradients by approximate increments obtained from
treatments (e.g. correlation or relativistic) on higher level. See the example below.

$corrgrad
# coordinate increment
1 0.0600
8 -0.0850

$forceapprox options
Approximate force constant matrix (as needed for geometry optimization tasks). The
storage format may be specified by the
available options:

format=format
the default format is format=(8f10.5), but other 10-digit f10.x formats (e.g.
x=4,6,..) are possible and will be used, after being manually specified within
$forceapprox. See the example below:

$forceapprox format=(8f10.4)
0.9124
-.0108 0.3347
0.2101 0.0299 1.3347
0.0076 0.1088 0.0778 0.6515

$hessian (projected)
this data block contains the analytical Cartesian force constant matrix (with transla-
tional and rotational combinations projected out) as output by the aoforce program
22.2. FORMAT OF KEYWORDS AND COMMENTS 469

and may be used to supply a high quality force constant matrix $forceapprox for ge-
ometry optimizations (specifying $forceinit on carthess, or $interconversion
cartesian –> internal hessian).

Relax Output Data Groups

$coord
either updated Cartesian coordinates if a successful coordinate update has been per-
formed, or Cartesian coordinates for input internal coordinates if only a conversion
from internal to Cartesian coordinates has been performed.

$basis
updated basis set exponents, basis sets contraction coefficients or scaling factors, if
$optimize basis on has been specified.

$global
updated global scaling factor for all basis set exponents, if $optimize global on has
been specified.

$forceapprox
an approximate force constant matrix to be used in quasi-Newton type geometry
optimizations; this matrix will be improved in subsequent optimization cycles if one
of the variable-metric methods ($forceupdate) has been chosen. See 5.3.13 and
22.2.23.

$forcestatic
a static (i.e. never updated) approximate force constant matrix to be used in DIIS-type
geometry optimizations. It will be initialized by relax specifying: $forceupdate pulay
. . . modus=<dq|dq> static.

The next data groups are output by relax (depending on the optimization subject) in order
to control the convergence of optimization procedures driven by the shell script jobex.

$maximum norm of cartesian gradient = real

$maximum norm of internal gradient = real

$maximum norm of basis set gradient = real

real is the absolute value of the maximum component of the corresponding gradient.

Other Input/Output data used by Relax

In order to save the effort for conversion of accumulated geometry and gradient data (as
needed for the force constant update or the DIIS update of the geometry) to the optimization
space, within which the geometry has to be optimized, one may specify the keyword
470 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$oldgrad
Then the relax program accumulates all subsequent coordinates and gradient as used
in optimization in this data group (or a referenced file). This overrides the input of
old coordinate and gradient data from data blocks $grad, $egrad, . . . as accumulated
by the grad program.

degrees

22.2.24 Keywords for Module statpt

$statpt
itrvec 0
hssupdate bfgs
hssfreq 0
keeptmode
hssidiag 0.5
radmax 0.3
radmin 1.0d-4
tradius 0.3
threchange 1.0d-6
thrmaxdispl 1.0d-3
thrmaxgrad 1.0d-3
thrrmsdispl 5.0d-4
thrrmsgrad 5.0d-4

Only non-default values are written in the control file except:

$statpt
itrvec 0

Following options are available:

itrvec
Index of the Hessian eigenvector to follow for transition structure search (transition
vector). Eigenpairs are sorted in ascending order, i.e. with increasing eigenvalues and
start with index 1. The eigenpairs corresponding to translations and rotations are
shifted to the end. For minimization the value 0 has to be specified.

hssupdate
Method of hessian update. For minimization default is BFGS, for TS search default
is Powell and none is for no update.

hssfreq
Recompute the full Hessian every N’th step during a transition state search. The
default is zero and the Hessian is read in or computed in the first step only. If the
22.2. FORMAT OF KEYWORDS AND COMMENTS 471

standard Hessian update methods fail, it can help to use this keyword. Warning: This
will make the calculation much more time demanding!

keeptmode
Freezing transition vector index.

hssidiag
diagonal hessian elements for diagonal Hessian guess (default: 0.5).

radmax
Maximum allowed value for trust radius (default: 0.3).

radmin
Minimum allowed value for trust radius (default: 1.0d-4).

tradius
Initial value for trust radius (default tradius: radmax = 0.3).

Convergence criteria

threchange threshold for energy change (default: 1.0d-6).

thrmaxdispl threshold for maximal displacement element (default: 1.0d-3).

thrmaxgrad threshold for maximal gradient element (default: 1.0d-3).

thrrmsdispl threshold for RMS of displacement (RMS = root mean square)

(default = 5.0d-4)

thrrmsgrad threshold for RMS of gradient (default: 5.0d-4).

All values are in atomic units.

22.2.25 Keywords for Module moloch

$properties specifies the global tasks for program moloch by virtue of the following options

$properties
trace off
moments active
potential off
cowan-griffin off
localization off
population analyses off
plot off
firstorder off
fit off
472 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

a missing option or a option followed by the flag off will not be taken into account. The flag
active may be omitted. For most of these options (with the only exceptions of trace and
cowan-griffin), there are additional data groups allowing for more detailed specifications,
as explained below.

moments
if moment is active you need

$moments
0th 1st 2nd 3rd
point .0 .0 .0

to compute the 0th, 1st, 2nd and 3rd moment at the reference point 0 0 0.

potential
if potential is active you need

$points #1
pot fld fldgrd shld
point .0 .0 .0

to compute the electrostatic potential (pot) and/or electrostatic field (fld) and/or
electrostatic field gradient (fldgrd) and/or the zeroth order contribution to the dia-
magnetic shielding (shld) at reference point 0 0 0.

localization
if localization is active you need $boys to perform a boys-localization of orbitals
with orbital energies ≥ thresholad=-2 Hartrees; localize with respect to locxyz=x,
y and z and write resulting orbitals to lmofile= ’lmo’. At the most sweeps=10000
orbital rotations are performed. Non-defaults may be specified using the following
suboptions:

lmofile= filename
locxyz dir1 dir2 dir3
threshold real
sweeps integer

population analyses
if population analyses is active you need

$mulliken
spdf molap netto irpspd irpmol mommul

to perform a Mulliken population analysis. The options specify the output data:

spdf print molecular orbital contributions to atomic s, p, d,. . . -populations

molap print molecular orbital contributions to overlap populations
22.2. FORMAT OF KEYWORDS AND COMMENTS 473

netto print atomic net populations

irpspd print contributions of (irreducible) representations to atomic s,p,d,. . . -
populations
irpmol print contributions of (irreducible) representations to overlap populations

or
$loewdin
to perform a Löwdin population analysis (options are invalid here). A Löwdin popu-
√ √
lation analysis is based on decomposing SD S instead of DS in case of a Mulliken
PA.
or

$paboon
momao maodump maofile=mao all

to perform a population analysis based on occupation numbers (the options are not
necessary and produce some output data concerning the modified atomic orbitals):

momao print MO contributions to occupation numbers of modified atomic orbitals

(MAOs).
maodump print all MAOs on standard output
maofile=mao all
print all MAOs to file mao.

This kind of population analysis basically aims at so-called shared electron numbers
(SEN) between two or more atoms. By default 2-, 3- and 4-center contributions to
the total density are plotted if they are larger than 0.01 electrons. Thresholds may be
individually chosen, as well as the possibility to compute SENs for molecular orbitals:
$shared electron numbers
orbitals
2-center threshold = real
3-center threshold = real
4-center threshold = real
Results of this kind of PA depend on the choice of MAOs. By default, all MAOs with
eigenvalues of the atomic density matrices larger than 0.1 will be taken into account.
This is a reasonable minimal basis set for most molecules. If modified atomic orbitals
shall not be selected according to this criterion, the data group $mao selection has to
be specified
$mao selection threshold =real ;
The default criterion for the selection of MAOs is the occupation number, for which a
global threshold can be specified within the same line as the keyword $maoselection.
If the global criterion or threshold is not desirable for some atoms, lines of the following
syntax have to be added for each atom type of these.
474 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

atom symb list nmao=i method=meth threshold=r

The parameters in this definition have the following meaning:

symb atom symbol

list list of all atoms for which this definition should apply. The syntax for this list
is as usual in TURBOMOLE, e.g. 2,3,8-10,12
nmao=i
means number of MAOs to be included
method=meth
means selection criterion for MAOs. meth can be occ (default), eig, or man
string, where occ denotes selection of MAOs by occupation numbers, eig se-
lection by eigenvalues and man allows manual selection. In the latter case the
string (max. 8 characters) appended to man serves as nickname for the defini-
tion of the MAOs to be chosen. This nickname is expected to appear as the
leftmost word in a line somewhere within data group $mao selection and is
followed by the indices of the modified atomic orbitals which are to be selected.
threshold=r
means the threshold to be applied for the selection criteria occ or eig (default:
0.1).

Example:

$mao selection threshold= 0.09

atom c 1,3-5 nmao= 5 method= eig threshold= 0.1
atom o 2 nmao= 3 method= man olabel
olabel 3-5

plot
option plot is out of fashion; to plot quantities on a grid, rather use $pointval in
connection with dscf, ridft, rimp2 or egrad, as described below. If nevertheless
plot is active you need

$grid #1
mo 4a1g
origin .000000 .000000 .000000
vector1 1.000000 .000000 .000000
vector2 .000000 1.000000 .000000
grid1 range -5.000000 5.000000 points 100
grid2 range -5.000000 5.000000 points 100
outfile = 4a1g

to obtain two-dimensional plot data of mo 4a1g (the plane is specified by origin and
two vectors with grid range and number of grid points) which is written to file 4a1g.
22.2. FORMAT OF KEYWORDS AND COMMENTS 475

Several plots may be obtained (#1, #2 etc.) at the same time. Use tool ’konto’ to
visualize the plot.
Note: This is the old-fashioned way to plot MOs and densities. A new—and easier—
one is to use $pointval, as described below.
fit
if fit is active you need

$vdw_fit
shell number_of_gridpoints distance_from_vdW_surface
refine value_of_potential

shell Each line refers to all atoms, the line specifies a spherical layer of grid
points around the atoms. The number of points and their distance from
the van der Waals surface [Bohr] are given (the default is 1.0).
refine one line only, smoothing of the layers of grid points around the molecule:
the real number is used to define isopotential surfaces on which the points
of the layers have to lie.

$vdw_radii
element_symbol van_d_waals_radius

One line per element has to be specified, it contains the name of the element and the
van der Waals radius in [Bohr].

22.2.26 Keywords for wave function analysis and generation of plot-

ting data

Properties of RHF, UHF and (two-component) GHF wave functions as well as those of
SCF+MP2 densities or such from excited state DFT-calculations can be directly analyzed
within the respective programs (dscf, ridft, mpgrad, rimp2 and egrad). In case of spin-
unrestricted calculations results are given for total densities (Dα + Dβ ) and spin densities
(Dα − Dβ ). If not explicitly noted otherwise, in the following "D" is the SCF density,
D(SCF), in case of dscf and ridft, the MP2-corrected density, D(SCF)+D(MP2), for
mpgrad and rimp2 and the entire density of the excited state in case of egrad. For modules
dscf and ridft the analysis of properties may be directly started by calling dscf -proper
(or ridft -proper). In case of mpgrad and rimp2 this is possible only, if the MP2 density
has already been generated, i.e. after a complete run of mpgrad or rimp2.
Functionalities of analyses are driven by the following keywords.

$mvd
leads to calculation of relativistic corrections for the SCF total density in case of dscf
and ridft, for the SCF+MP2 density in case of rimp2 and mpgrad and for that of
the calculated excited state in case of egrad. Quantities calculated are expectation
values < p2 >, < p4 > and the Darwin term ( 1/ZA ∗ ρ(RA )).
P
476 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$moments
yields calculation of electrostatic moments arising from nuclear charges and total
electron densities. Also without setting this keyword moments up to quadrupole are
calculated, with respect to reference point (0,0,0). Supported extensions:

$moments <i>
x1 y1 z1
x2 y2 z2
.
.

By integer i ; the maximum order of moments is specified, maximum and default is

i=3 (octopole moments), real numbers x, y, z allow for the specification of one or more
reference points.

$pop
drives the options for population analyses. By default a Mulliken PA in the basis
of Cartesian atomic orbitals (CAOs) is performed for the total density (Dα + Dβ )
leading to Mulliken (gross) charges and, in case of spin-unrestricted calculations also
for the spin density (Dα − Dβ ) leading to Mulliken (gross) numbers for unpaired
electrons. Besides total numbers also contributions from s-, p-, . . . functions are listed
separately.
Two-component wavefunctions (only module ridft and only if $soghf is set): In
two-component calculations instead of Sz |(Sx , Sy , Sz )| is written to the output. Ad-
ditionally a vector-file named spinvec.txt is written, which includes the resulting
spin vector for each atom in the system (also the direction).
The following modifications and extensions are supported, if the respective commands
are written in the same line as $pop:

lall Additional information about px , py , pz (and analogous for d and f functions)

is displayed (lengthy output).
atoms list of atoms
Contributions are plotted only if arising from atoms selected by list.
thrpl=real
Contributions smaller than thrpl are not displayed (default: 0.01).
overlapMulliken atomic overlap matrix is displayed.
netto Mulliken net populations (diagonal elements of Mulliken overlap matrix) are
calculated.
mosum list of MOs
Summed Mulliken contributions for a group of molecular orbitals defined by
numbers referring to the numbering obtained e.g. from the tool eiger. Note
that occupancy of MOs is ignored, i.e. all orbitals are treated as occupied.
22.2. FORMAT OF KEYWORDS AND COMMENTS 477

mo list of MOs
Mulliken contributions for single MOs defined by numbers (independent of
whether they are occupied or not). If this option is valid, one may additionally
set
dos width=real points=integer
to calculate a (simulated) density of states by broadening the discrete
energy levels with Gaussians and superimposing them. The width of
each Gaussian may be set by input (default: 0.01 a.u.). The resolution
(number of points) may be chosen automatically (default values are usu-
ally sufficient to generate a satisfactory plot) or specified by hand. The
output files (dos in case of RHF wave functions, and dos_a+b, dos_a-b,
dos_alpha, dos_beta; for UHF cases) contain energies (first column),
resulting DOS for the respective energy (second column) as well as s-,
p-, d-contributions for the respective energy (following columns).

Example:

$pop mo 23-33 dos atoms 2,3,7-8

leads to Mulliken PA (CAO-basis) for each of the eleven MOs 23-33, regarding only
contributions from atoms 2-3 and 7-8 (results are written to standard output) and
generation of file(s) with the respective simulated density of states.

$pop nbo
to perform a natural population analyses [280]. The possible options (specified in the
same line) are

AO must be provided, the CAO case is not implemented.

tw=real Threshold tw to circumvent numerical difficulties in computing Ow
(default: tw=1.d-6).
idbgl=integer Debug level
(default: idbgl=0).
ab For UHF cases: Print alpha and beta density results.
short Print only natural electron configuration and summary.

Example:

$pop nbo AO ab short atoms 1,2,6

leads to a natural population analysis (AO-basis) with printing the results of alpha
and beta densities (only the electron configuration and the summary) for the atoms
1,2 and 6.
To change the NMB set for atoms, one has to add a $nbonmb-block in the control
file. Example:
478 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$nbonmb
ni s:4 p:2 d:1
o s:2 p:1

leads to a NMB set for Ni of 4 s-, 2 p- and 1d-functions and for O of 2 s- and 1
p-functions.
$pop paboon
to perform a population analyses based on occupation numbers [281] yielding "shared
electron numbers (SENs)" and multicenter contributions. For this method always the
total density is used, i.e. the sum of alpha and beta densities in case of UHF, the
SCF+MP2-density in case of MP2 and the GHF total density for (two-component-
)GHF.
The results of such an analysis may depend on the choice of the number of modified
atomic orbitals ("MAOs"), which can be specified by an additional line; without
further specification their number is calculated by the method "mix", see below.
Note: One should carefully read the information concerning MAOs given in the output
before looking at the numbers for atomic charges and shared electron numbers.

$mao selection options

to specify how MAOs are selected per atom.
Available options are:
a) for the way of sorting MAOs of each atom:
eig
MAOs are sorted according to their eigenvalue (those with largest EW
finally are chosen). This is the default.
occ
MAOs are sorted according to their occupation; note that the number
of all occupation is NOT the number of electrons in the system. This
option is kept rather for historical reasons.
b) for the determination of the number of MAOs:
fix
A fixed number of MAOs is taken for each atom; usually this is the
number of shells up to the complete valence shell, e.g. 5 for B-Ne, 6
for Na-Mg, etc. Exceptions are Elements Sc (Y, La), Ti (Zr, Hf), V
(Nb, Ta) for which not all five d-shells are included, but only 2, 3 or 4,
respectively. This modification leads to better agreement with partial
charges calculated by an ESP-fit.
thr <real>
All MAOs with an eigenvalue larger than <real> are chosen; default
is <real>=0.1. This and the following two options are not valid in
connection with occ.
max
Maximum of numbers calculated from fix and thr=0.1 is taken.
22.2. FORMAT OF KEYWORDS AND COMMENTS 479

mix
2:1 mixture of fix and thr=0.1. This choice gives best agreement (sta-
tistical) with charges from an ESP-fit. It is the default choice.
c) for additional information about MAOs:
info
Eigenvalues and occupations for each MAO are written to output.
dump
Entire information about each MAO is written to output. Lengthy.
Further for each atom the number of MAOs and the sorting mode can be
specified individually in lines below this keyword. Example:
atom 1,3-4 eig 7
leads to choice of the 7 MAOs with largest eigenvalue at atoms 1, 3-4.

$localize
enables the generation of localized molecular orbitals (LMOs) using Boys localization.
By default, all occupied orbitals are included, localized orbitals are written (by de-
fault in the AO basis) to file(s) lmo in case of RHF and lalp and lbet in case of UHF
orbitals. Note, that LMOs usually break the molecular symmetry; so, even for sym-
metric cases the AO (not the SAO) basis is used for the output. The localized orbitals
are sorted with respect to the corresponding diagonal element of the Fock matrix in
the LMO basis. In order to characterize these orbitals, dominant contributions of
(canonical) MOs are written to standard output as well as results of a Mulliken PA
for each LMO (for plotting of LMOs see option $pointval).
The keyword allows for following options (to be written in the same line):

mo list of MOs
Include only selected MOs (e.g. valence MOs) in localization procedure (num-
bering as available from Eiger).
sweeps=integer
maximum number of orbital rotations to get LMOs; default value is 10000
(sometimes not enough, in particular for highly delocalised systems).
thrcont=real
lower threshold for displaying MO and Mulliken contributions (default: 0.1).
CAO
LMOs are written to file in the CAO basis (instead of AO).
pipmez or pm or pipek-mezey
Pipek-Mezey localization is used.
boys
Foster-Boys localization is used.
ibo
Intrinsic Bond Orbitals are used.
480 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

center
Compute LMO centers, i.e. the diagonal matrix elements of the position oper-
ator ~ri = hi|~r|ii
spread
p p
Compute LMO spreads defined as σi = hi|(~r − ~ri )2 |ii = hi|~r 2 |ii2 − ~ri2 ,
where ~ri is the LMO center
spatial_sorting
sort LMOs such that LMOs with spatially close centers are also close in in-
dex space, by default the LMOs are sorted according to the orbital energy
expectation values
timing
print timings for localization procedures

$wfn
triggers the generations of a wfn file. It can be used in dscf/ridft single-point calcu-
lations or in ricc2/egrad gradient calculations.

$esp_fit
fits point charges at the positions of nuclei to electrostatic potential arising from
electric charge distribution (also possible for two-component calculations, for UHF
cases also for spin density). For this purpose the ("real") electrostatic potential is
calculated at spherical shells of grid points around the atoms. By default, Bragg-
Slater radii, rBS , are taken as shell radii, for each atom the number of points is given
2
by 1000 · rBS , the total number of points is the sum of points for each atom reduced
by the number of points of overlapping spheres. Non-default shells (one or more) can
be specified as follows:
$esp_fit
shell i1 s1
shell i2 s2
..
.
Integer numbers i define the number of points for the respective shell, real numbers
s constants added to radii (default corresponds to one shell with s=1.0).
A parameterization very close to that by Kollman (U.C. Singh, P.A. Kollman, J.
Comput. Chem. 5(2), 129-145 (1984)) may be obtained by
$esp_fit kollman

Here five shells are placed around each atom with r=1.4*rvdW + k, k=0pm, 20pm,
40pm, 60pm, 80pm, and rvdW are the van-der-Waals radii of the atoms.

$pointval
drives the calculation of space-dependent molecular quantities at 3D grids, planes,
lines or single points. Without further specifications the values of densities are plot-
ted on a three-dimensional grid adapted to the molecular size. Data are deposed to
22.2. FORMAT OF KEYWORDS AND COMMENTS 481

output files (suffix ṗlt) that can be visualized directly with the gOpenMol program.
In case of RHF-dscf/ridft calculations you get the total density on file td.plt, for
UHF-dscf/ridft calculations one gets both values for the total density (Dα + Dβ )
on td.plt and the "spin density" (Dα − Dβ ) on sd.plt. For mpgrad/rimp2 cal-
culations one gets in the RHF case the total density (D(SCF+MP2)) on td.plt
and the MP2 contribution on mp2d.plt and in the UHF case one obtains the to-
tal density (Dα (SCF + M P 2) + Dβ (SCF + M P 2)) on td.plt, the "spin density"
(Dα (SCF + M P 2) − Dβ (SCF + M P 2)) on td.plt, and the respective MP2 contri-
butions on files mp2d.plt and mp2sd.plt. For egrad it is similar, just replace in the
filenames mp2 by e.
Integration of density (if absolute value greater than eps) within a sphere (origin
x, y, z, radius r) is performed for

$pointval integrate x y z r eps

By default the origin is at (0,0,0), the radius is chosen large enough to include the
whole 3D box and all contributions are regarded (eps=0).
Data different from total and spin densities are generated by following (combinable)
settings (to be written in the same line as statement $pointval):

pot leads to calculation of electrostatic potential arising from electron densities,

nuclei and—if present—constant electric fields and point charges. The densities
used for calculation of potentials are the same as above; the respective filenames
are generated from those of densities by replacement of the "d" (for density)
by a "p" (for potential). By "pot eonly" only the electronic contribution to
the electrostatic potential is calculated.
fld calculation of electric field. Note, that for 3D default output format (.plt, see
below) only norm is displayed. Densities used are the same as above, filenames
are generated from those of densities by replacement of "d" (for density) by
"f" (for field).
mo list of MO numbers
calculation of amplitudes of MOs specified by numbers referring to the num-
bering obtained e.g. from the tool eiger in the same format. The respective
filenames are self-explanatory and displayed in the output. Note, that also
in MP2 and excited state calculations the HF/DFT ground state orbitals are
plotted (and not natural MP2/excited orbitals).
Two-component cases: The density of the spinors specified by numbers re-
ferring to the numbering obtained e.g. from the file EIGS are visualized. By
setting the keyword minco also the amplitudes of the spinor-parts are calcu-
lated, whose weights (the probability of finding the electron in this part) lie
above the threshold.
lmo list of LMO numbers
calculation of amplitudes of LMOs (previously generated by $localize) or-
482 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

dered by the corresponding diagonal element of the Fock matrix in the LMO
basis.
nmo list of NMO numbers
calculation of amplitudes of NMOs (previously generated by
$natural orbitals file=natural and
$natural orbital occupation file=natural
elf calculation of the electron localization function (ELF). [288]
dens has to be set, if additionally to one of the above quantities also the density is
to be computed.
xc calculation of the Kohn-Sham exchange-correlation potential. It is only valid
for DFT calculations and it works for all exchange-correlation functionals, in-
cluding LHF. Note that for hybrid functionals, only the Kohn-Sham part of
the potential will be computed (the HF part is a non-local-operator and can’t
be plotted). For GGA functional the full potential will be computed (local and
non-local terms)
For line plots the output file is tx.vec . For UHF calculations the output files
are tx.vec (alpha-spin potential) and sx.vec (beta-spin potential).
For a line plot the file has three columns: 1: total potential 2: local term ( or
Slater-potential for LHF) 3: non-local terms or Correction term for LHF

Output formats may be specified by e.g. fmt=xyz if written to the same line as
$pointval. Supported are:

xyz in case of scalars (density, (L)MO amplitudes, electrostatic potential) this for-
mat is: (x, y, z, f (x, y, z)). In case of vectors components of the vector and its
norm are displayed. This format is valid for all types of grid (3D, plane, line,
points, see below), it is the default format in case of calculation of values at
single points. Output file suffix is .xyz.
plt only for 3D, default in this case. Data are written to binary files that can
be directly read by gOpenMol. Note, that this output is restricted to scalar
quantities; thus in case of vectors (E-field) only the norm is plotted. Output
file suffix is .plt.
map only for 3D. Data are written to ASCII files that can be imported by e.g.
gOpenMol. Note, that this output is restricted to scalar quantities; thus in
case of vectors (E-field) only the norm is plotted. Output file suffix is .map.
txt a format compatible with gOpenMol for visualization of vectors v. The format
is x, y, z, vx , vy , vz .
vec for planes and lines (default in these cases). In case of a line specified by α · ~v
(see below) output is α, f (x, y, z) for scalars, for vectors components and norm
are displayed. vectors. Analogously, in case of planes it is α, β, f (x, y, z). The
output (file suffix .vec) may be visualized by plotting programs suited for two-
dimensional plots. A command file (termed gnuset) to get a contour plot by
gnuplot is automatically generated.
22.2. FORMAT OF KEYWORDS AND COMMENTS 483

cub only for 3D, writes out data in Cube format which can be imported by many
external visualization programs.

For 3D grids non-default boundarys, basis vector directions, origin and resolutions
may be specified as follows:

$pointval
grid1 vector 0 3 0 range -2,2 points 200
grid2 vector 0 0 -7 range -1,4 points 300
grid3 vector 1 0 0 range -1,1 points 300
origin 1 1 1

Grid vectors (automatically normalized) now are (0, 1, 0),(0, 0, −1),(1, 0, 0), the grid is
centered at (1, 1, 1), and e.g. for the first direction 200 points are distributed between
-2 and 2.
In particular for 2D plots it is often useful to shift and rotate the grid plane such that
some particular atoms are located in the plot plane. This can be achieved with the
orient option, which accepts as additional input a list of atoms, e.g. the atoms 4-5
and 9 in the following example:

$pointval
grid1 vector 0 3 0 range -2,2 points 200
grid2 vector 0 0 -7 range -1,4 points 300
orient 4-5,9
rotate 25 3

This will shift the origin of the plot into the center of mass of the specified set of
atoms and align the grid axes with their principal axes: If two or more atoms are
specified which lie on one line grid axis 1 is rotated into this line. If the specified
subset consists of three atoms located in a plane the grid axis 1 and 2 are rotated into
this plane. If more than three atoms are specified the grid axis 1 and 2 are rotated
into a plane which fitted to the position of all specified atoms. With the rotate
option the grid can be rotated around a grid axis. It accepts as input the rotation
angle the index (1/2/3) of the grid axis around which the grid will be rotated.
Grids of lower dimensionality may be specified (in the same line as $pointval) by
typing either geo=plane or geo=line or geo=point The way to use is best explained
by some examples:

$pointval geo=plane
grid1 vector 0 1 0 range -2,2 points 200
grid2 vector 0 0 1 range -1,4 points 300
origin 1 1 1

Values are calculated at a plane spanned by vectors (0,1,0) and (0,0,1) centered at
(1,1,1).
484 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$pointval geo=line
grid1 vector 0 1 0 range -2,2 points 50
origin 0 0 1

Values are calculated at a line in direction (0,1,0) centered at (0,0,1). Output format
as above.

$pointval geo=point
7 5 3
0 0 7

Values are calculated at the two points (7.0, 5.0, 3.0) and (0.0, 0.0, 7.0).
Plane-averaged density can be computed by

$pointval dens averagez fmt=vec

grid1 vector 1 0 0 range -10,10 points 100
grid2 vector 0 1 0 range -10,10 points 100
grid3 vector 0 0 1 range -20,20 points 200
origin 0 0 0

The generated file td.vec will contain the quantity

Z Z
ρ(z) = dxdy ρ(x, y, z) (22.3)

22.2.27 Keywords for Module frog

The ab initio molecular dynamics (MD) program frog needs a command file named mdmaster.
The interactive Mdprep program manages the generation of mdmaster and associated files.
It is always a good idea to let Mdprep check over mdmaster before starting an MD run.
Mdprep has online-help for all menus.
In this implementation of ab initio MD, time is divided into steps of equal duration ∆t.
Every step, the energy and its gradient are calculated and these are used by the frog to
work out the new coordinates for the next step along the dynamical trajectory. Both the
accuracy of the trajectory and the total computation time thus depend crucially on the time
step chosen in Mdprep. A bad choice of timestep will result in integration errors and cause
fluctuations and drift in the total energy. As a general rule of thumb, a timestep ∆t should
be chosen which is no longer than one tenth of the shortest vibrational period of the system
to be simulated.
Note that Mdprep will transform velocities so that the total linear and angular momentum
is zero. (Actually, for the Leapfrog algorithm, initial velocities are ∆t/2 before the starting
time).
The following keywords are vital for frog:
22.2. FORMAT OF KEYWORDS AND COMMENTS 485

$nsteps 75
Number of MD time steps to be carried out. $nsteps is decreased by 1 every time
frog is run and JOBEX -md stops when $nsteps reaches 0.

$natoms 9
Number of atoms in system.

$current file=mdlog.aa
The file containing the current position, velocity, time and timestep, that is, the input
configuration. During an MD run the $current information is generally kept at the
end of the $log file.

$log file=mdlog.ZZ
The file to which the trajectory should be logged, i.e. the output: t=time (a.u.);
atomic positions x,y,z (Bohr) and symbols at t;
timestep (au) ∆t;
atomic symbols and velocities x,y,z (au) at t − (∆t/2);
kinetic energy (H) interpolated at t, ab initio potential energy (H) calculated at t,
and pressure recorded at the barrier surface (atomic units, 1 au = 29.421 TPa) during
the corresponding timestep;
ab initio potential energy gradients x,y,z (H/Bohr) at t.
This file can be manipulated with log2? tools after the MD run (Section 1.5).

$turbomole file=control
Where to look for TURBOMOLE keywords $grad etc.

$md_status
The status of the MD run is a record of the action carried out during the previous MD
step, along with the duration of that step. The format matches that of $md_action
below.
Canonical dynamics is supported using the Nosé-Hoover thermostat. This option can
be enabled in Mdprep or by the following syntax:

$md_status
canonical T=500 t=100
from t= -25.0000000000 until t= 0.00000000000

Here, T specifies the temperature of the thermostat in K (500 K in the example) and
t specifies the thermostat relaxation time in a.u. (100 a.u. in the example). It is
advisable to choose the thermostat relaxation 2-10 times larger than the time step.
Note that user-defined actions are presently not supported in canonical dynamics
mode.

These are optional keywords:

$seed -123
Integer random number seed
486 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

$title
Arbitrary title

$log_history
100 mdlog.P
71 mdlog.Q

$ke_control
length 50
response 1

To determine the trends in kinetic energy and total energy (average values and overall drifts)
it is necessary to read the history of energy statistics over the recent MD steps. The number
of MD steps recorded so far in each log file are therefore kept in the $log_history entry: this
is updated by the program each step. The length of records needed for reliable statistics and
the number of steps over which changes are made to kinetic energy (response) are specified
in $ke_control.

$barrier angstroms
type elps
limits 5.0 10.0 7.5
constant 2.0
springlen 1.0
temperature 300.0

$barrier specifies a virtual cavity for simulating condensed phases. The optional flag,
angstroms, can be used to indicate that data will be entered in Ångstrøms rather
than Bohr.

type
can be one of orth, elps, or none, for orthorhombic, ellipsoidal, or no barrier
(the default) respectively.
limits
are the +x,y,z sizes of the cavity. In this case, an ellipsoid with a major axis
of 20 Å along y, semi-major of 15 Å on z, and minor of 10 Å on x.
constant
is the Hooke’s Law force constant in atomic units of force (H/Bohr) per length
unit. Here, it is 2.0 H/Bohr/Ångstrøm, a bastard combination of units.
springlen
is the effective limit to the restorative force of the barrier. For this system, an
atom at 5 Å into the barrier will feel the same force as at 1.0 Å.
22.2. FORMAT OF KEYWORDS AND COMMENTS 487

temperature
denotes the temperature of the cavity walls in Kelvin. If the system quasi-
temperature is below this setpoint, particles will be accelerated on their re-
turn to the interior. Alternately, they will be retarded if the system is too
warm. A temperature of 0.0 K will turn off wall temperature control, returning
molecules to the system with the same momentum as when they encountered
the barrier.

$constraints angstroms
tolerance 0.05
adjpercyc 0.25
type H O 0.9 1.2
type F C 0.0 1.7
type H C -1.0 1.2
2 1 0.0
3 1 1.54
4 1 -1.0

$constraints
specifies and/or automatically generates atomic distance constraints. The optional
flag, angstroms, can be used to indicate that data will be entered in Ångstrøms rather
than Bohr.

tolerance
is the convergence criterion for application of constraints. All distances must
be within +/- tolerance of the specified constraint. Additionally, the RMS
deviation of all constrained distances must be below 2/3 of tolerance.
adjpercyc
is the fraction of the total distance correction to be applied on each constraint
iteration.
type X A const rmax
commands frog to find the closest A atom to each atom X that is closer than
rmax and apply const. The first type line above examines each H atom and
looks for any O atoms within 1.2 Å. The shortest distance, if any, is then fixed
at 0.9 Å. Similarly, the second type line binds each F to the closest C within
1.7 Å, but since const=0.0, that distance is fixed at the current value. The
third type line attaches H atoms to the appropriate nearby C, but at the current
average H-C distance multiplied by the absolute value of const.

Explicitly specified constraints are listed by atom index and supercede auto-generated
constraints. A positive third number fixes the constraint at that value, while zero fixes
the constraint at the current distance, and a negative number unsets the constraint.
The output of frog contains the full list of constrained atom pairs and their current
constraints in explicit format.
488 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

User-defined instructions allow the user to tell frog to change some aspect of the MD run
at some point in time t=real number. The same format is used for $md_status above. Here
is an example:

$md_action
fix total energy from t=2000.0
anneal from t=2500.0
free from t=3000.0

In this example, starting from the time 2000.0 a.u., velocities are to be scaled every step to
keep average total energy constant. Then, from 2500.0 a.u., gradual cooling at the default
rate (annealing) is to occur until the time 3000.0 a.u., when free Newtonian dynamics will
resume.
Here are all the possible instructions:

$md_action
fix temperature from t=<real>
fix total energy from t=<real>
These commands cause velocities to be scaled so as to keep the average kinetic energy
(i.e. quasi-temperature) or the average total energy approximately constant. This is
only possible once enough information about run history is available to give reliable
statistics. (Keywords $log_history, $ke_control).

$md_action
set temperature at t=<real> to x=<real> K
set total energy at t=<real> to x=<real> H
set kinetic energy at t=<real> to x=<real> H
set position file=<filename> at t=<real>
set velocity file=<filename> at t=<real>
set velocity at t=<real> random
set velocity at t=<real> zero
At some time during the ab initio MD run the user can specify a new value for one of
the dynamical variables. The old value is discarded. Single values are given by x=real
number. Vectors must be read in frog format from file=file.

$md_action
anneal from t=<real>
anneal from t=<real> x=<real>
quench from t=<real>
quench from t=<real> x=<real> file=<file>
relax at t=<real>
22.2. FORMAT OF KEYWORDS AND COMMENTS 489

In Simulated Annealing MD, the temperature of a run is lowered so as to find

minimum-energy structures. Temperature may be lowered gradually by a small factor
each step (anneal; default factor 0.905 over 100 steps) or lowered rapidly by reversing
all uphill motion (quench; default factor -0.8 each step). The cooling factors may be
changed from the default using x=. Another option allows the quenching part of the
run to be logged to a separate file. Alternatively, a standard non-dynamical geometry
optimization can be carried out in a subdirectory (relax).

$md_action
free from t=<real>
Finally, this instruction turns off any previous action and resumes free dynamics. This
is the default status of an MD run.

$surface_hopping
This keyword allows to carry out Tully-type fewest switches surface hopping (SH)
[318]. This option is only available in combination with TDDFT. For the TDDFT
surface hopping see Tapavicza et al. 2007 [319]; for the current implementation see
Tapavicza et al. 2011 [320]. In the current implementation the surface hopping
algorithm only allows switches between the first excited singlet state and the ground
state. However, total energies of higher excited states can be computed during the
MD simulation. The proper functioning of SH has only been tested for the option

$md_action
fix total energy from t= 0.00000000000

To carry out SH dynamics simulations, the keyword $surface_hopping has to be

added to the control and mdmaster file. In addition several keywords are required in
the control file:

$currentstate
1

$currentstate
keyword needed to ensure dynamics starting in S1

$nacme
needed to compute non-adiabatic couplings; this keyword requires the use of weight derivatives
in section dft
The excitations, excited state gradients and nonadiabatic coupling vectors are gener-
ated from a TDDFT calculation and stores by egrad. The gradients and the nona-
diabatic vectors are stored under the data group and the excitations under . These
data groups are automatically generated in an MD simulation.

$sh_coeffs file=sh_coeffs
collects amplitudes of the adiabatic states along the trajectory . This keyword is
automatically generated in the first MD step, with the entire electronic population
490 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

on the current active state. However, the user may specify the initial coefficients to
be something different by manually creating the file.

$population_el
stores the electronic populations at each time step. This is generated by frog and
the user need not specify.

$prob_fssh
stores the probability of hopping at each time step. This is generated by frog and
the user need not specify.

$active
stores the active state for surface hopping at each time step. This is generated by
frog and the user need not specify if is specified.
Special caution has to be taken to control problems related to conical intersections
[321, 322]. At geometries where conical intersections between the ground and excited
state are present, DFT often exhibits singlet instabilities, which leads to imaginary
excitation energies in linear response TDDFT; in this case the MD run is terminated.
This problem can be circumvented by the use of the Tamm-Dancoff approximation
(TDA) to TDDFT (see 8). In addition an optional keyword for the md_master file
can be used:

$gap_threshold <real>
enforces a switch to the ground state in case the S1 -S0 energy gap drops
below <real> eV. As default a switch to S0 is enforced if the S1 TDDFT-TDA
excitation energy becomes negative.

Often times if a switch is enforced due to a negative TDA excitation energy the
potential energy surface is discontinuous and limited numerical precision of the nuclear
forces may lead to a loss of total energy conservation. In this case the nuclear velocities
are rescaled to obtain a conserved total energy.

22.2.28 Keywords for Module mpshift

In order to control the program execution, you can use the following keywords within the
control file:

$csmp2
Switches on the calculation of the MP2 NMR shieldings. The required SCF shielding
step will be performed in the same run. This flag will be set by the script mp2prep.

$traloop n
specifies the number of loops (or ’passes’) over occupied orbitals n when doing an
MP2 calculation: the more passes the smaller file space requirements—but CPU time
will go up. This flag will be set by the script mp2prep.
22.2. FORMAT OF KEYWORDS AND COMMENTS 491

$mointunit
Scratch file settings for an MP2 calculation. Please refer to Section 22.2.19 for a
description of the syntax. This flag will be set by the script mp2prep.

$shiftconv integer
Residuum convergence threshold for the solution of the CPHF equations. Default is
7.

$oldcphf
Uses the old CPHF solver, which was the default up to Version 7.4, the calculation
can be restarted at any stage with this solver.

$csconv real
Sets the convergence threshold for the shielding constant of succeeding CPHF itera-
tions. The unit is ppm and the default value is 0.01. Only used with the old solver.

$csconvatom integer
This selects the atom number for convergence check after each cphf iteration. After
this convergence is reached all other atoms are checked, too (default: 1). Only used
with the old solver.

$thime, $thize, $scftol, $scfintunit, $scfmo

have the save meaning as in dscf (see Section 22.2.8);
Since mpshift works ’semi-direct’ it uses the same integral storage.

$scratch files
The scratch files allocated by mpshift can be placed anywhere in your file systems
instead of the working directory by referencing their pathnames in this data group.
All possible scratch files are listed in the following example:

$scratch files
mpshift csssmat path1/file1
mpshift cshsmat path2/file2
mpshift csdgsmat path3/file3
mpshift csusmat path4/file4
mpshift dens path5/file5
mpshift fock path6/file6
mpshift dfock path7/file7
mpshift idvds1 path8/file8
mpshift idvds2 path9/file9
mpshift idvds3 path10/file10
mpshift jdvds1 path11/file11
mpshift jdvds2 path12/file12
mpshift jdvds3 path13/file13
mpshift cshmmat path14/file14

$trast , $trand traloop-number

stands for traloop start and traloop end. Each loop or pass in MP2 chemical shift
492 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

calculations can be done individually by providing the keywords $trast and $trand.
This can be used to do a simple parallelization of the run:
Create separate inputs for each traloop. Add

$trast <number>
$trand <number>

in the control files, number goes from 1 to the number of $traloops. Each calcu-
lation will create a restart file called restart.mpshift. To collect all steps and to
do the remaining work, copy all restart files to one directory and rename them to
restart.mpshift.number, add $trast -1 and $trand number_of_traloops to the
control file and start mpshift.

$vcd
B
Uaiβ will be written to file for a subsequent calculation of VCD rotational strengths
by aoforce.

$nucsel
Selection of nuclei whose NMR shielding tensor has to be computed. The nuclei can
be specified by the element symbol (i.e. $nucsel "c","h") or by the corresponding
number as set by the coordinates (i.e. $nucsel 1,3,5-8). Without setting $nucsel,
the shielding tensors of all nuclei are calculated. You can reuse this keyword for
nuclear spin-spin coupling constants.

$csmaxiter
Set the maximum number of CPHF iterations, default 30.

$nics
Calculation of nucleus-independent chemical shieldings. The coordinates have to be
listed just as the molecular coordinates of the molecule. A sample input for NH3
would look like this:

$nics
0.00000000000000 0.00000000000000 0.12917062194577
1.75612466223131 0.00000000000000 -0.59831613718919
-0.87806233111566 1.52084856970468 -0.59831613718919
-0.87806233111566 -1.52084856970468 -0.59831613718919
0.00000000000000 0.00000000000000 0.00000000000000
$coord
0.00000000000000 0.00000000000000 0.12917062194577 n
1.75612466223131 0.00000000000000 -0.59831613718919 h
-0.87806233111566 1.52084856970468 -0.59831613718919 h
-0.87806233111566 -1.52084856970468 -0.59831613718919 h

$gimic
Prepare input for a GIMIC calcuation, see https://github.com/qmcurrents/gimic/.
22.2. FORMAT OF KEYWORDS AND COMMENTS 493

The density and the perturbed density matrix are written to disk. GIMIC allows
to calculate the magnetically induced current density to estimate the degree of aro-
maticity and to study electron delocalization pathways.

22.2.29 Keywords for Parallel Runs

On all systems the parallel input preparation is done automatically. Details for the parallel
installation are given in Section 3.4.1. The following keywords are necessary for all parallel
runs:
$parallel_platform architecture

Currently the following parallel platforms are supported:

SMP for systems with very fast communication; all CPUs are used for the linear
algebra part. Synonyms for SMP are:
HP V-Class, SP3-SMP and HP S/X-Class

MPP for systems with fast communication like Fast-Ethernet, the number of CPUs
that will be taken for linear algebra part depends on the size of the matrices.
Synonyms for MPP are:
SP3 and linuxcluster

cluster for systems with slow communication, the linear algebra part will be done on
one single node. Synonyms for cluster are:
HP Cluster and every platform that is not known by TURBOMOLE

SGI similar to SMP, but here the server task is treated differently: the MPI imple-
mentation on the SGIs would cause this task to request too much CPU time
otherwise.

If you want to run mpgrad, $traloop has to be equal to or a multiple of the number of
parallel workers.
For very large parallel runs it may be impossible to allocate the scratch files in the working
directory. In this case the $scratch files option can be specified; an example for a dscf
run is given below. The scratch directory must be accessible from all nodes.

$scratch files
dscf dens /home/dfs/cd00/cd03_dens
dscf fock /home/dfs/cd00/cd03_fock
dscf dfock /home/dfs/cd00/cd03_dfock
dscf ddens /home/dfs/cd00/cd03_ddens
dscf xsv /home/dfs/cd00/cd03_xsv
dscf pulay /home/dfs/cd00/cd03_pulay
dscf statistics /home/dfs/cd00/cd03_statistics
494 CHAPTER 22. KEYWORDS IN THE CONTROL FILE

dscf errvec /home/dfs/cd00/cd03_errvec

dscf oldfock /home/dfs/cd00/cd03_oldfock
dscf oneint /home/dfs/cd00/cd03_oneint

For all programs employing density functional theory (DFT) (i.e. dscf/gradand ridft/rdgrad)
$pardft can be specified:

$pardft
tasksize=1000
memdiv=0

The tasksize is the approximate number of points in one DFT task (default: 1000) and
memdiv says whether the nodes are dedicated exclusively to your job (memdiv=1) or not
(default: memdiv=0).
For dscf and grad runs you need a parallel statistics file which has to be generated in
advance. The filename is specified with
$2e-ints_shell_statistics file=DSCF-par-stat
or
$2e-ints’_shell_statistics file=GRAD-par-stat
respectively.
The statistics files have to be generated with a single node dscf or grad run. For a dscf
statistics run one uses the keywords:

$statistics dscf parallel

$2e-ints_shell_statistics file=DSCF-par-stat
$parallel_parameters
maxtask=400
maxdisk=0
dynamic_fraction=0.300000

and for a grad statistics run:

$statistics grad parallel

$2e-ints’_shell_statistics file=GRAD-par-stat
$parallel_parameters
maxtask=400

maxtask is the maximum number of two-electron integral tasks,

maxdisk defines the maximum task size with respect to mass storage (MBytes) and
dynamic_fraction is the fraction of two-electron integral tasks which will be allocated
dynamically.
In the parallel version of ridft, the first client reads in the keyword $ricore from the
control file and uses the given memory for the additional RI matrices and for RI-integral
22.2. FORMAT OF KEYWORDS AND COMMENTS 495

storage. All other clients use the same amount of memory as the first client does, although
they do not need to store any of those matrices. This leads to a better usage of the available
memory per node. But in the case of a big number of auxiliary basis functions, the RI
matrices may become bigger than the specified $ricore and all clients will use as much
memory as those matrices would allocate even if that amount is much larger than the given
memory. To omit this behavior one can use:
$ricore_slave integer
specifying the number of MBs that shall be used on each client.
For parallel jobex runs one has to specify all the parallel keywords needed for the different
parts of the geometry optimization, i.e. those for dscf and grad, or those for ridft and
rdgrad, or those for dscf and mpgrad.
496 CHAPTER 22. KEYWORDS IN THE CONTROL FILE
Chapter 23

Sample control files

23.1 Introduction

The file control is the input file for TURBOMOLE which directly or by cross references provides
the information necessary for all kinds of runs and tasks. control is usually generated by
define, the input generator. The following sample control files cover a variety of methods
and systems. The keywords themselves are explained in Chapter 22.

497
498 CHAPTER 23. SAMPLE CONTROL FILES

23.2 NH3 Input for a RHF Calculation

Main File control

$title
NH3 c3v SVP
$operating system unix
$symmetry c3v
$coord file=coord
$intdef file=coord
$atoms
n 1 \
basis =n def-SVP
h 2-4 \
basis =h def-SVP
$pople AO
$basis file=basis
$rundimensions
dim(fock,dens)=495
natoms=4
nshell=15
nbf(CAO)=30
nbf(AO)=29
dim(trafo[SAO<-->AO/CAO])=69
rhfshells=1
$scfmo file=mos
$closed shells
a1 1-3 ( 2 )
e 1 ( 2 )
$scfiterlimit 30
$scfconv 7
$thize .10000000E-04
$thime 5
$scfdamp start= .500 step= .050 min= .100
$scfdump
$scfintunit
unit=30 size=0 file=twoint
$scfdiis start=0.5
$drvopt
cartesian on
basis off
global off
hessian on
23.2. NH3 INPUT FOR A RHF CALCULATION 499

dipole on
nuclear polarizability
$interconversion off
qconv=1.d-10
maxiter=25
$optimize
internal on
cartesian off
global off
basis off logarithm
$coordinateupdate
dqmax=0.3
interpolate on
statistics 5
$forceupdate
ahlrichs numgeo=0 mingeo=3 maxgeo=4 modus=<g|dq> dynamic fail=0.1
threig=0.005 reseig=0.005 thrbig=3.0 scale=1.00 damping=0.0
$forceinit on
diag=default
$energy file=energy
$grad file=grad
$forceapprox file=force
$lock off
$last step define
$end

File coord

$coord
.00000000000000 .00000000000000 .54561506935122 n
-.87806233111566 1.52084856970468 -.18187168978374 h
-.87806233111566 -1.52084856970468 -.18187168978374 h
1.75612466223131 .00000000000000 -.18187168978374 h
$intdef
# definitions of internal coordinates
1 k 1.0000000000000 stre 4 1 val= 1.90084
2 k 1.0000000000000 bend 4 3 1 val= 106.27756
1.0000000000000 bend 3 2 1
1.0000000000000 bend 2 4 1
$end
500 CHAPTER 23. SAMPLE CONTROL FILES

File basis

$basis
*
n def-SVP
# n (7s4p1d) / [3s2p1d] {511/31/1}
*
5 s
1712.8415853 -.53934125305E-02
257.64812677 -.40221581118E-01
58.458245853 -.17931144990
16.198367905 -.46376317823
5.0052600809 -.44171422662
1 s
.58731856571 1.0000000000
1 s
.18764592253 1.0000000000
3 p
13.571470233 -.40072398852E-01
2.9257372874 -.21807045028
.79927750754 -.51294466049
1 p
.21954348034 1.0000000000
1 d
1.0000000000 1.0000000000
*
h def-SVP
# h (7s) / [3s] {511}
*
3 s
13.010701000 .19682158000E-01
1.9622572000 .13796524000
.44453796000 .47831935000
1 s
.12194962000 1.0000000000
1 p
.80000000000 1.0000000000
*
$end

File mos

$scfmo expanded format(4d20.14)

23.2. NH3 INPUT FOR A RHF CALCULATION 501

1 a1 eigenvalue=-.15633041862301D+02 nsaos=10
.98699003163455D+00-.47221435341751D-01 .55873125006179D-02-.48016374887169D-02
.26746008768233D-02 .20823779196149D-03 .14270460008808D-01 .90849517503597D-02
.58676121352806D-03 .29091871198884D-03
2 a1 eigenvalue=-.99896275238736D+00 nsaos=10
.26412162337482D+00 .51846472345768D+00 .37623729061179D+00-.77139882704089D-02
-.47252329287316D-02-.21494050853221D-02 .11795673774658D+00 .83316086019184D-01
-.11229203933488D-01-.27038186251429D-02
3 a1 eigenvalue=-.57101279949392D+00 nsaos=10
-.35584199011701D-01-.96938258881594D-01-.70254605702716D-01 .65569041318341D+00
-.44746149963029D+00 .40094287741992D-03 .51691151834284D-01 .47722350097160D-01
.19189122068531D-02 .56638497851180D-03
1 e eigenvalue=-.64374209294851D+00 nsaos=9
-.49313475446075D+00 .33757893447603D+00-.76142296567409D-04-.74524664248740D-04
-.26407572210452D+00-.22619038902975D+00-.50035170531670D-05-.12199166245418D-03
.63021657663245D-04
$end
502 CHAPTER 23. SAMPLE CONTROL FILES

23.3 NO2 input for an unrestricted DFT calculation

Main File control

$title
NO2 c2v UKS SVP
$operating system unix
$symmetry c2v
$coord file=coord
$intdef file=coord
$atoms
n 1 \
basis =n def-SVP
o 2-3 \
basis =o def-SVP
$pople AO
$basis file=basis
$rundimensions
dim(fock,dens)=1098
natoms=3
nshell=18
nbf(CAO)=45
nbf(AO)=42
dim(trafo[SAO<-->AO/CAO])=85
rhfshells=2
$uhfmo_alpha none file=alpha
$uhfmo_beta none file=beta
# none : hamilton core guess will be made
# files alpha and beta will be generated by the program
$uhf
$alpha shells
a1 1-6 ( 1 )
a2 1 ( 1 )
b1 1-4 ( 1 )
b2 1 ( 1 )
$beta shells
a1 1-5 ( 1 )
a2 1 ( 1 )
b1 1-4 ( 1 )
b2 1 ( 1 )
$scfiterlimit 30
$scfconv 7
$thize .10000000E-04
23.3. NO2 INPUT FOR AN UNRESTRICTED DFT CALCULATION 503

$thime 5
$scfdamp start=1.500 step= .050 min= .100
$scfdump
$scfintunit
unit=30 size=2 file=/work/user/twoint
$scfdiis start=0.5
$scforbitalshift closedshell=.3
$drvopt
cartesian on
basis off
global off
hessian on
dipole on
nuclear polarizability
$interconversion off
qconv=1.d-10
maxiter=25
$optimize
internal on
cartesian off
global off
basis off logarithm
$coordinateupdate
dqmax=0.3
interpolate on
statistics 5
$forceupdate
ahlrichs numgeo=0 mingeo=3 maxgeo=4 modus=<g|dq> dynamic fail=0.1
threig=0.005 reseig=0.005 thrbig=3.0 scale=1.00 damping=0.0
$forceinit on
diag=default
$energy file=energy
$grad file=grad
$forceapprox file=force
$lock off
$dft
functional b-p
gridsize m3
$last step define
$end
504 CHAPTER 23. SAMPLE CONTROL FILES

File coord

$coord
.00000000000000 .00000000000000 -1.00494155217173 n
1.85766051386774 .00000000000000 .50247077608587 o
-1.85766051386774 .00000000000000 .50247077608587 o
$intdef
# definitions of internal coordinates
1 k 1.0000000000000 stre 2 1 val= 2.39232
2 d 1.0000000000000 stre 3 1 val= 2.39232
3 k 1.0000000000000 bend 2 3 1 val= 101.88429
$end

File basis

$basis
*
n def-SVP
# n (7s4p1d) / [3s2p1d] {511/31/1}
*
5 s
1712.8415853 -.53934125305E-02
257.64812677 -.40221581118E-01
58.458245853 -.17931144990
16.198367905 -.46376317823
5.0052600809 -.44171422662
1 s
.58731856571 1.0000000000
1 s
.18764592253 1.0000000000
3 p
13.571470233 -.40072398852E-01
2.9257372874 -.21807045028
.79927750754 -.51294466049
1 p
.21954348034 1.0000000000
1 d
1.0000000000 1.0000000000
*
o def-SVP
# o (7s4p1d) / [3s2p1d] {511/31/1}
*
5 s
23.3. NO2 INPUT FOR AN UNRESTRICTED DFT CALCULATION 505

2266.1767785 -.53431809926E-02
340.87010191 -.39890039230E-01
77.363135167 -.17853911985
21.479644940 -.46427684959
6.6589433124 -.44309745172
1 s
.80975975668 1.0000000000
1 s
.25530772234 1.0000000000
3 p
17.721504317 .43394573193E-01
3.8635505440 .23094120765
1.0480920883 .51375311064
1 p
.27641544411 1.0000000000
1 d
1.2000000000 1.0000000000
*
$end
506 CHAPTER 23. SAMPLE CONTROL FILES

23.4 TaCl5 Input for an RI-DFT Calculation with ECPs

Main File control

$title
$operating system unix
$symmetry d3h
$coord file=coord
$intdef file=coord
$atoms
ta 1 \
jbas=ta def-SVP \
basis =ta def-SVP \
ecp =ta def-ecp
cl 2-6 \
jbas=cl def-SVP \
basis =cl def-SVP
$pople AO
$basis file=basis
$ecp file=basis
$rundimensions
dim(fock,dens)=7662
natoms=6
nshell=51
nbf(CAO)=122
nbf(AO)=115
dim(trafo[SAO<-->AO/CAO])=346
$scfmo none file=mos
# none : hamilton core guess will be made
# file mos will be generated by the program
$scfiterlimit 30
$scfconv 6
$thize .10000000E-04
$thime 5
$scfdamp start= .900 step= .050 min= .100
$scfdump
$scfintunit
unit=30 size=0 file=twoint
$scfdiis start=0.5
$drvopt
cartesian on
basis off
global off
23.4. TACL5 INPUT FOR AN RI-DFT CALCULATION WITH ECPS 507

hessian on
dipole on
nuclear polarizability
$interconversion off
qconv=1.d-10
maxiter=25
$optimize
internal on
cartesian off
global off
basis off logarithm
$coordinateupdate
dqmax=0.3
interpolate on
statistics 5
$forceupdate
ahlrichs numgeo=0 mingeo=3 maxgeo=4 modus=<g|dq> dynamic fail=0.1
threig=0.005 reseig=0.005 thrbig=3.0 scale=1.00 damping=0.0
$forceinit on
diag=default
$energy file=energy
$grad file=grad
$forceapprox file=force
$lock off
$dft
functional b-p
gridsize m3
$last step define
$ricore 20
$ridft
$jbas file=auxbasis
$closed shells
a1’ 1-11 ( 2 )
a2’ 1-2 ( 2 )
e’ 1-10 ( 2 )
a2" 1-8 ( 2 )
e" 1-4 ( 2 )
$end

File coord

$coord
.00000000000000 .00000000000000 .00000000000000 ta
508 CHAPTER 23. SAMPLE CONTROL FILES

2.19392179448315 -3.79998401587749 .00000000000000 cl

2.19392179448315 3.79998401587749 .00000000000000 cl
-4.38784358896629 .00000000000000 .00000000000000 cl
.00000000000000 .00000000000000 4.46615918865523 cl
.00000000000000 .00000000000000 -4.46615918865523 cl
$intdef
# definitions of internal coordinates
1 k 1.0000000000000 stre 1 2 val= 4.38784
2 k 1.0000000000000 stre 1 5 val= 4.46616
$end

File basis

$basis
*
ta def-SVP
# ta (7s6p5d) / [6s3p2d] {211111/411/41}
*
2 s
14.400000000 .99343296745
12.000000000 -1.6510077975
1 s
5.0701477302 1.0000000000
1 s
.86033356487 1.0000000000
1 s
.37158938894 1.0000000000
1 s
.10745336254 1.0000000000
1 s
.39142776556E-01 1.0000000000
4 p
7.4188720000 .26979695152
5.6984100000 -.46968874449
1.1777211960 .50905100155
.54478533555 .52298161137
1 p
.22309270117 1.0000000000
1 p
.43100000000E-01 1.0000000000
4 d
3.9738796278 -.52799310714E-01
1.4528884813 .18558319471
23.4. TACL5 INPUT FOR AN RI-DFT CALCULATION WITH ECPS 509

.61042908544 .42959071631
.24216276510 .43497228232
1 d
.87909318337E-01 1.0000000000
*
cl def-SVP
# cl (7s5p) / [6s2p] {211111/41}
*
5 s
10449.827566 .19708362484E-02
1571.7365221 .14754727977E-01
357.12065523 .66679112875E-01
100.25185935 .17228924084
30.812727554 .15883786100
3 s
51.923789434 -.10009298909
5.7045760975 .60841752753
2.3508376809 .54352153355
1 s
.44605124672 1.0000000000
1 s
.16848856190 1.0000000000
5 p
307.66790569 -.87801484118E-02
72.102015515 -.63563355471E-01
22.532680262 -.24016428276
7.8991765444 -.47798866557
2.8767268321 -.38515850005
1 p
.77459363955 1.0000000000
1 p
.21037699698 1.0000000000
1 d
.65000000000 1.0000000000
*
$ecp
*
ta def-ecp
*
ncore = 60 lmax = 3
# coefficient r^n exponent
f
12.0179609 2 2.0178811
510 CHAPTER 23. SAMPLE CONTROL FILES

s-f
1345.8806470 2 14.5464077
36.7668062 2 7.2732038
-12.0179609 2 2.0178811
p-f
378.4253015 2 9.9355653
22.2930909 2 4.9677824
-12.0179609 2 2.0178811
d-f
104.8839557 2 6.3473769
8.7558481 2 3.1736885
-12.0179609 2 2.0178811
*
$end

File auxbasis

$jbas
*
ta def-SVP
*
3 s
15.521335 -.493702989D+00
7.555743 .259256574D+01
3.699576 -.523168657D+01
1 s
1.820141 .262393615D+01
1 s
0.898838 .157711902D+01
1 s
0.445062 .200789711D+00
1 s
0.220729 .185974307D+00
1 s
0.109530 .765184411D-01
1 p
1.5024958 1.0
1 p
0.5629855 1.0
1 p
0.2281880 1.0
1 p
0.09507835 1.0
23.4. TACL5 INPUT FOR AN RI-DFT CALCULATION WITH ECPS 511

2 d
1.337006 .190072032D-01
0.599535 -.155214344D-01
1 d
0.280427 -.138946250D-01
1 d
0.133078 -.895263676D-02
1 f
1.1428211 1.0
1 f
0.4395465 1.0
1 f
0.1758186 1.0
3 g
1.630421 .100251139D+00
0.747093 .737448223D-01
0.349040 .276219913D-01
1 g
0.164143 .546316580D-02
*
*
cl def-SVP
*
8 s
4097.080409 .198054511D+01
1203.083193 .530973450D+01
386.280948 .132352655D+02
135.337690 .107149960D+02
51.567046 -.132565114D+01
21.261034 .271180364D+01
9.420135 .754640511D+01
4.445228 .173603618D+01
1 s
2.209399 -.140197496D+01
1 s
1.141575 .982719736D+00
1 s
0.604182 .464178589D+00
1 s
0.322378 .369336889D+00
4 p
51.8499902611 .359335506D-01
17.5847835188 -.869599318D-01
512 CHAPTER 23. SAMPLE CONTROL FILES

6.49227239618 .721211200D-01
2.55889114714 -.634201864D-01
1 p
1.05118767781 .264152293D-01
1 p
.437994865757 -.197670692D-01
4 d
34.705550 -.548703710D-01
10.704427 -.619019402D-02
3.568067 .337450480D-01
1.249848 -.905232209D-01
1 d
0.445360 .418680075D-01
1 f
1.1872146118 1.0000000
1 g
1.30000000 1.0000000
*
$end
23.5. BASISSET OPTIMIZATION FOR NITROGEN 513

23.5 Basisset optimization for Nitrogen

Main File control

$title
Basisset-optimization for nitrogen SV(P)
$operating system unix
$symmetry oh
#--- uncomment following line to clean the basis-file after optimization ----
#$dump basis set
$coord file=coord
$user-defined bonds file=coord
$pople AO
$basis file=basis
$rundimensions
dim(fock,dens)=141
natoms=1
nshell=6
nbf(CAO)=15
nbf(AO)=14
dim(trafo[SAO<-->AO/CAO])=17
rhfshells=2
$scfmo none file=mos
$roothaan 1
a = 1 b = 2
$scfiterlimit 60
$scfconv 10
$thize 0.10000000E-04
$thime 5
$scfdamp start=1.500 step=0.050 min=0.100
$scfdump
$scfintunit
unit=30 size=90 file=twoint
$scfdiis start=0.5
$scforbitalshift closedshell=.4
$drvopt
cartesian off
#---- optimize basis! -> basis on ----
basis on
global off
hessian on
dipole on
nuclear polarizability
514 CHAPTER 23. SAMPLE CONTROL FILES

$interconversion off
qconv=1.d-7
maxiter=25
$optimize
internal off
cartesian off
global off
#---- optimize basis! -> basis on logarithm ----
basis on logarithm
$coordinateupdate
dqmax=0.3
interpolate on
statistics 5
$forceupdate
ahlrichs numgeo=0 mingeo=3 maxgeo=4 modus=<g|dq> dynamic fail=0.6
threig=0.005 reseig=0.005 thrbig=3.0 scale=1.00 damping=0.0
$forceinit on
diag=default
$energy file=energy
$grad file=gradient
#---- optimize basis! -> $egrad file=egradient ----
$egrad file=egradient
$forceapprox file=forceapprox
$lock off
$atoms
n 1 \
basis =n def-SV(P)
$closed shells
a1g 1-2 ( 2 )
$open shells type=1
t1u 1 ( 1 )
$end

File coord

$coord
0.00000000000000 0.00000000000000 0.00000000000000 n
$user-defined bonds
$end

File basis

*
23.5. BASISSET OPTIMIZATION FOR NITROGEN 515

n def-SV(P)
# n (7s4p1d) / [3s2p1d] {511/31/1}
# use expopt to optimize exponents and contopt to optimize contractions
*
5 s expopt contopt
1712.8415853 0.53934125305E-02
257.64812677 0.40221581118E-01
58.458245853 0.17931144990
16.198367905 0.46376317823
5.0052600809 0.44171422662
1 s expopt
0.58731856571 1.0000000000
1 s expopt
0.18764592253 1.0000000000
3 p expopt contopt
13.571470233 0.40072398852E-01
2.9257372874 0.21807045028
0.79927750754 0.51294466049
1 p expopt
0.21954348034 1.0000000000
# 1 d
# 1.0000000000 1.0000000000
*

File mos

$scfmo scfconv=10 format(4d20.14)

# SCF energy is -54.3329250250 a.u. (virial theorem = 2.000000001)
#
1 a1g eigenvalue=-.15623888057347D+02 nsaos=3
-.99166890864040D+00-.28420294406651D-010.91519592317893D-02
2 a1g eigenvalue=-.92524548524703D+00 nsaos=3
0.30506869715453D+00-.65051761026701D+00-.44610487551870D+00
3 a1g eigenvalue=0.74881229854801D+00 nsaos=3
0.30759302935434D+00-.16295969601691D+010.16126161147521D+01
1 t1u eigenvalue=-.56865046629517D+00 nsaos=2
0.67926397018841D+000.46005039868410D+00
2 t1u eigenvalue=0.96169069264790D+00 nsaos=2
-.95675659621171D+000.10794148212163D+01
$end
516 CHAPTER 23. SAMPLE CONTROL FILES

23.6 ROHF of Two Open Shells

Extracts from control for O2 in D3d Symmetry

# HF-SCF/SVP

# Reference: triplet-sigma in D3d

# This is a Roothaan case (as is D-infinity-h).
#
$coord
0.0 0.0 1.08597397921317 o
0.0 0.0 -1.08597397921317 o
$symmetry d3d
$closed shells
a1g 1-3 (2)
a2u 1-2 (2)
eu 1 (2)
$open shells type=1
eg 1 (1)
$roothaan 1
a = 1 b = 2
$energy SCF SCFKIN SCFPOT
1 -149.4774402753 149.4799190239 -298.9573592992

# Reference: singlet-delta in D3d

# This is a Roothaan case (as is D-infinity-h).
#
$coord
0.0 0.0 1.08597397921317 o
0.0 0.0 -1.08597397921317 o
$symmetry d3d
$closed shells
a1g 1-3 (2)
a2u 1-2 (2)
eu 1 (2)
$open shells type=1
eg 1 (1)
$roothaan 1
a = 1/2 b = 0
$energy SCF SCFKIN SCFPOT
1 -149.4297623470 149.4298692899 -298.8596316369
23.6. ROHF OF TWO OPEN SHELLS 517

Extracts from control for O2 in D2h Symmetry

# HF-SCF/SVP

# Triplet-sigma in D2h
#
$coord
0.0 0.0 1.08597397921317 o
0.0 0.0 -1.08597397921317 o
$symmetry d2h
$closed shells
ag 1-3 ( 2 )
b1u 1-2 ( 2 )
b2u 1 ( 2 )
b3u 1 ( 2 )
$open shells type=1
b2g 1 ( 1 )
b3g 1 ( 1 )
$roothaan 1
a = 1 b = 2
$energy SCF SCFKIN SCFPOT
1 -149.4774402750 149.4798706643 -298.9573109393

# Singlet-delta in D2h : xx-yy component

# where x = b2g and y = b3g. In D-infinity-h, b2g and b3g combine to eg.
#
$coord
0.0 0.0 1.08597397921317 o
0.0 0.0 -1.08597397921317 o
$symmetry d2h
$closed shells
ag 1-3 ( 2 )
b1u 1-2 ( 2 )
b2u 1 ( 2 )
b3u 1 ( 2 )
$open shells type=1
b2g 1 ( 1 )
b3g 1 ( 1 )
$roothaan 2
$rohf
1b2g-1b3g a = 0 b = 2
518 CHAPTER 23. SAMPLE CONTROL FILES

1b2g-1b2g a = 1 b = 0
1b3g-1b3g a = 1 b = 0
$energy SCF SCFKIN SCFPOT
1 -149.4297623516 149.4298351805 -298.8595975321

# Singlet-delta in D2h : xy+yx component

# (an example of the general type: [xy]-singlet)
# where in D2h x = b2g and y = b3g are of different symmetry.
# In D-infinity-h, b2g and b3g combine to eg; see the reference
# calculation in D3d above.
#
$coord
0.0 0.0 1.08597397921317 o
0.0 0.0 -1.08597397921317 o
$symmetry d2h
$closed shells
ag 1-3 ( 2 )
b1u 1-2 ( 2 )
b2u 1 ( 2 )
b3u 1 ( 2 )
$open shells type=1
b2g 1 ( 1 )
b3g 1 ( 1 )
$roothaan 2
$rohf
1b2g-1b3g a = 1 b = -2
1b2g-1b2g a = 0 b = 0
1b3g-1b3g a = 0 b = 0
$energy SCF SCFKIN SCFPOT
1 -149.4297623501 149.4298391833 -298.8596015334
Chapter 24

The Perl-based Test Suite

Structure

24.1 General

Testing the Turbomole modules for correctness and speed is the first task once the coding
is completed. It is subject to automatization and thus requires a structure which is as
simple and flexible as possible. In the Perl-based test suite this is implemented by a Perl
script TTEST which performs all the testing and benchmarking tasks and resides in the
central scripts directory of the Turbomole installation. The test examples are located in
subdirectories of the TURBOTEST directory, grouped according to the modules modules to be
tested and a rough short/long classification. The benchmark suite shows the same directory
structure and is rooted in the TURBOBENCH directory.
The central idea of the Perl-based test suite is that only the specific information about an
individual test example is included in its local directory along with the input and reference
files. This information is stored in the criteria file CRIT which contains the program calls,
test criteria, and specific reference timings. Running the test script creates a new test subdi-
rectory, usually called like TESTDIR.i786-pc-linux-gnu, where the Turbomole programs
are run and the results are summarized in the protocol file TESTPROTOKOLL.

24.2 Running the tests

Starting a single test example is simple. Change to the test example of your choice and
call the TTEST script without arguments. The test is started in a subdirectory named
TESTDIR.sysname, where sysname is the current platform name as returned by the Sysname

519
520 CHAPTER 24. PERL-BASED TEST SUITE

script. The tested executable, a short description, and the test summary are output to the
screen. Detailed information about the performed commands and results of all test criteria
are found in the TESTPROTOKOLL file in the test subdirectory.
The default location for the binaries and scripts used for testing is the $TURBODIR directory.
If you like to test some other, e.g., your local version of the Turbomole binaries or scripts,
you can specify the loading paths by the -l or -ls options for the binaries and scripts,
respectively,

TTEST -l /usr/local/TURBOMOLE/bin/i786-pc-linux-gnu \
-ls /usr/local/TURBOMOLE/scripts.

A specific executable can be chosen by the -x option,

TTEST -x /usr/local/TURBOMOLE/bin/i786-pc-linux-gnu/dscf.

If a test output is already present, e.g., in the TESTDIR directory, you may wish to check the
results. This is accomplished by calling TTEST in check mode,

TTEST --check TESTDIR,

which compares the results in TESTDIR with the reference and writes the results to the
CHECKPROTOKOLL file in the test directory.
Testing parts of the TURBOTEST directory structure or the entire test suite at once is per-
formed by calling the TTEST script from the appropriate place. The test script works recur-
sively, executing all test examples underneath its starting directory. This requires that the
test examples be arranged in a TURBOTEST-like directory structure,

progname/short|long/example (e.g., dscf/short/H2O.SCF.E1),

and the TURBOTEST directory contain a DEFCRIT file with general test suite settings. If
TTEST is started in the central TURBOTEST without any options, all available test examples
are executed. By giving the list of module names (for full list, check TTEST –help) as
argument to the script, the test can be restricted to these modules. The –short and –long
options allow the user to select only the short or long test examples, respectively. Some
examples of usage are given in the following table:

TTEST dscf called in the TURBOTEST directory, performs

only the tests for Dscf module.
TTEST called in the TURBOTEST/dscf directory, does
the same.
TTEST –long executes long examples for all modules.
TTEST ridft –short performs all short examples from the ridft
directory.
24.3. TAKING THE TIMINGS AND BENCHMARKING 521

Recursive testing creates some additional files in the central TURBOTEST directory. The global
protocol file TESTPROTOKOLL.sysname contains short result messages for all test and a list
of errors occurred. The list of failed tests is also written to the PROBLEMS.sysname file and
can be rerun by calling the test script with the -r option,

TTEST -r PROBLEMS.i786-pc-linux-gnu.

The -r may also be useful to create any user-defined selection of test examples. The full list
of available examples is obtained by the TTEST –list command.
Once you are done with testing, you may wish to clean up afterwards. To do it, use the
–clean and –realclean options of the TTEST script. The difference between these two is
that TTEST –clean deletes only the test directories and protocols that were created for the
current computer architecture as returned by Sysname. In contrast, the TTEST –realclean
wipes out all test directories and protocols that get in its way.

24.3 Taking the timings and benchmarking

Benchmarking differs from testing only in that program timings are computed and compared
with reference timings. Calling the script as

TTEST --timings

performs the test, calculates the CPU and wall clock timings, and writes the raw results
to the TESTTIMINGS.sysname.nodename file. Auxiliary scripts Tbtim and Tblist help to
convert this data to a more readable form and produce summaries as LATEX tables. The
Tbtim script creates a summary of benchmark results for a given computer platform from
the original timings file. Tblist produces benchmark comparisons of different platforms.
The corresponding timings files must be provided as arguments to the Tblist script. For
more details and options, see TBTIM –help and TBLIST –help.

24.4 Modes and options of the TTEST script

The TTEST script knows several operation modes: "run", "check", "list", "clean", "real-
clean", and "validate", controlled by its options. The "run" mode is default and means
that the test calculations are performed and the results are written to the TESTPROTOKOLL
file. The "check" mode differs only in that the programs are not executed, but the existing
program output is checked against the reference. The results of the check are written to
the CHECKPROTOKOLL file. Calling the test script in the "list" mode simply lists the test
examples that are currently available. This allows the user to save the full list to file, edit,
and re-use it with the -r option. The "clean" and "realclean" options are for cleaning up
the test directories and protocols. Finally, the "validate" mode is mainly of use for writing
the CRIT files. It helps to verify the match patterns provided in the test criteria and shows
522 CHAPTER 24. PERL-BASED TEST SUITE

if it extracts the expected data for comparison with the reference. For every output file
used for testing, the "validate" option produces a copy with an additional .val extension.
The match strings evaluated for test criteria are highlighted in the output by ««< and »»>
marks.
There is a lot of options controlling the behavior of TTESTṪesting specific versions of Tur-
bomole modules is provided by loading path options, -l for binaries, -ls for scripts, and
-x for a specific executable. For benchmarking, you need the –timings option to produce
the timing summaries, and the –newref option to save the current program timings as the
new reference. The module specifications and –short, –long, and -r options can be used for
selecting the test examples. The more specialized options are summarized in the following
table. Note that most of these options can also be set in the DEFCRIT file (see below).

Operation modes
–help Prints out the help message and exits.
-h
-?
–list Lists the available test examples.
–clean Deletes the test directories and summary files for
the current architecture (given by Sysname, see
Chapter 1.5).
–realclean Deletes all test directories and protocols.
–check dir Checks the correctness of an existing program test
in the directory dir (default: TESTDIR.sysname).
Useful if new criteria or new references are
established.
–validate dir Examines the output files in the directory dir
-val dir (default: TESTDIR.sysname) and highlights the
positions of the retrieved matches.
Loading path and naming options
–loaddir dir Loading path for the Turbomole binaries
-l dir (default: $TURBODIR/bin/sysname).
–scriptdir dir Loading path for the Turbomole scripts
-ls dir (default: $TURBODIR/scripts).
–testprog prog Tests the given executable prog.
-x prog
–dir dir Name for the local test directory
(default: TESTDIR.sysname).
–critfile file Name for the local criteria file
(default: CRIT).
–defcritfile file Name for the test suite settings file
(default: DEFCRIT).
24.4. MODES AND OPTIONS OF THE TTEST SCRIPT 523

–protfile file Name for the local protocol file

–output file (default: TESTPROTOKOLL).
–gprotfile file Name for the global protocol file
(default: TESTPROTOKOLL.sysname).
–checkfile file Name for the check protocol file
(default: CHECKPROTOKOLL).
–errfile file Name for the local error output file
(default: output.err).
–probfile file Name for the failed tests list
(default: PROBLEMS.sysname).
–timfile file Name for the timings file
(default: TIMINGS.sysname).
–valfile file Name for the validation file for ’run’
criteria (default: RUNCRIT.val).
Execution options
–short Only short / long subdirectories of the
–long test suite will be tested (default: –short –long).
–restart file The list of test examples for execution will
-r file be read in from file
(default: PROBLEMS.sysname).
–newref string Produces new reference timings and writes them
to the CRIT file. A short description of the refer-
ence platform is provided by string.
–fileref Produces new reference files.
–batchmode Running in batch mode, no screen output.
–errorstop Stops / Does not stop after the first error.
–noerrorstop (default: –noerrorstop).
–timings Writes / Does not write the timings on file for
–notimings further processing. (default: –notimings).
–runopts Sets the conditions under which the test is run
-o (default: "sequential, parallel")
524 CHAPTER 24. PERL-BASED TEST SUITE
Bibliography

[1] R. Ahlrichs; M. Bär; M. Häser; H. Horn; C. Kölmel. Electronic structure calcula-

tions on workstation computers: The program system Turbomole. Chem. Phys. Lett.,
162(3), 165–169, (1989).

[2] S. G. Balasubramani; G. P. Chen; S. Coriani; M. Diedenhofen; M. S. Frank; Y. J.

Franzke; F. Furche; R. Grotjahn; M. E. Harding; C. Hättig; A. Hellweg; B. Helmich-
Paris; C. Holzer; U. Huniar; M. Kaupp; A. Marefat Khah; S. Karbalaei Khani; T.
Müller; F. Mack; B. D. Nguyen; S. M. Parker; E. Perlt; D. Rappoport; K. Reiter; S.
Roy; M. Rückert; G. Schmitz; M. Sierka; E. Tapavicza; D. P. Tew; C. van Wüllen;
V. K. Voora; F. Weigend; A. Wodyński; J. M. Yu. Turbomole: Modular program suite
for ab initio quantum-chemical and condensed-matter simulations. J. Chem. Phys.,
152, 184107, (2020).

[3] B. P. Pritchard; D. Altarawy; B. Didier; T. D. Gibson; T. L. Windus. New basis

set exchange: An open, up-to-date resource for the molecular sciences community. J.
Chem. Inf. Model., 59(11), 4814–4820, (2019).

[4] K. L. Schuchardt; B. T. Didier; T. Elsethagen; L. Sun; V. Gurumoorthi; J. Chase;

J. Li; T. L. Windus. Basis set exchange: A community database for computational
science. J. Chem. Inf. Model., 47(3), 1045–1052, (2007).

[5] A. Schäfer; H. Horn; R. Ahlrichs. Fully optimized contracted Gaussian basis sets for
atoms Li to Kr. J. Chem. Phys., 97(4), 2571–2577, (1992).

[6] A. Schäfer; C. Huber; R. Ahlrichs. Fully optimized contracted Gaussian basis sets of
triple zeta valence quality for atoms Li to Kr. J. Chem. Phys., 100(8), 5829–5835,
(1994).

[7] K. Eichkorn; F. Weigend; O. Treutler; R. Ahlrichs. Auxiliary basis sets for main row
atoms and transition metals and their use to approximate Coulomb potentials. Theor.
Chem. Acc., 97(1–4), 119–124, (1997).

[8] F. Weigend; F. Furche; R. Ahlrichs. Gaussian basis sets of quadruple zeta valence
quality for atoms H–Kr. J. Chem. Phys., 119(24), 12753–12762, (2003).

[9] F. Weigend; R. Ahlrichs. Balanced basis sets of split valence, triple zeta valence and
quadruple zeta valence quality for H to Rn: Design and assessment of accuracy. Phys.
Chem. Chem. Phys., 7(18), 3297–3305, (2005).

525
526 BIBLIOGRAPHY

[10] A. K. Rappé; C. J. Casewit; K. S. Colwell; W. A. Goddard III; W. M. Skiff. UFF, a full

periodic table force field for molecular mechanics and molecular dynamics simulations.
J. Am. Chem. Soc., 114(25), 10024–10035, (1992).
[11] C. Hättig; F. Weigend. CC2 excitation energy calculations on large molecules using
the resolution of the identity approximation. J. Chem. Phys., 113(13), 5154–5161,
(2000).
[12] C. Hättig; K. Hald. Implementation of RI-CC2 for triplet excitation energies with an
application to trans-azobenzene. Phys. Chem. Chem. Phys., 4(11), 2111–2118, (2002).
[13] C. Hättig; A. Köhn; K. Hald. First–order properties for triplet excited states in the
approximated coupled cluster model CC2 using an explicitly spin coupled basis. J.
Chem. Phys., 116(13), 5401–5410, (2002).
[14] C. Hättig. Geometry optimizations with the coupled-cluster model CC2 using the
resolution-of-the-identity approximation. J. Chem. Phys., 118(17), 7751–7761, (2003).
[15] A. Köhn; C. Hättig. Analytic gradients for excited states in the coupled-cluster model
CC2 employing the resolution-of-the-identity approximation. J. Chem. Phys., 119(10),
5021–5036, (2003).
[16] C. Hättig; A. Hellweg; A. Köhn. Distributed memory parallel implementation of
energies and gradients for second-order Møller-Plesset perturbation theory with the
resolution-of-the-identity approximation. Phys. Chem. Chem. Phys., 8(10), 1159–
1169, (2006).
[17] A. Hellweg; S. Grün; C. Hättig. Benchmarking the performance of spin–component
scaled CC2 in ground and electronically excited states. Phys. Chem. Chem. Phys.,
10, 1159–1169, (2008).
[18] N. O. C. Winter; C. Hättig. Scaled opposite-spin CC2 for ground and excited states
with fourth order scaling computational costs. J. Chem. Phys., 134, 184101, (2011).
[19] R. A. Bachorz; F. A. Bischoff; A. Glöß; C. Hättig; S. Höfener; W. Klopper; D. P. Tew.
The mp2-f12 method in the turbomole programm package. J. Comput. Chem., 32,
2492–2513, (2011).
[20] D. P. Tew; W. Klopper; C. Neiss; C. Hättig. Quintuple-ζ quality coupled-cluster
correlation energies with triple-ζ basis sets. Phys. Chem. Chem. Phys., 9, 1921–1930,
(2007).
[21] C. Hättig; D. P. Tew; A. Köhn. Accurate and efficient approximations to explicitly
correlated coupled-cluster singles and doubles, CCSD-F12. J. Chem. Phys., 132,
231102, (2010).
[22] D. P. Tew; W. Klopper. Open-shell explicitly correlated f12 methods. Mol. Phys.,
108, 315–325, (2010).
[23] G. Schmitz; B. Helmich; C. Hättig. A O(N 3 )-scaling PNO-MP2 method using a hybrid
OSV-PNO approach with an iterative direct generation of OSVs. Mol. Phys., 111,
2463–2473, (2013).
BIBLIOGRAPHY 527

[24] G. Schmitz; C. Hättig; D. Tew. Explicitly correlated PNO-MP2 and PNO-CCSD and
its application to the S66 set and large molecular systems. Phys. Chem. Chem. Phys.,
16, 22167–22178, (2014).

[25] R. Bauernschmitt; R. Ahlrichs. Treatment of electronic excitations within the adia-

batic approximation of time dependent density functional theory. Chem. Phys. Lett.,
256(4–5), 454–464, (1996).

[26] R. Bauernschmitt; R. Ahlrichs. Stability analysis for solutions of the closed shell
Kohn-Sham equation. J. Chem. Phys., 104(22), 9047–9052, (1996).

[27] M. Kühn; F. Weigend. Implementation of Two-component Time-Dependent Den-

sity Functional Theory in TURBOMOLE. J. Chem. Theory Comput., 9, 5341–5348,
(2013).

[28] M. Kühn; F. Weigend. Two-Component Hybrid Time-Dependent Density Functional

Theory within the Tamm-Dancoff Approximation. J. Chem. Phys., 142, 034116,
(2015).

[29] C. Holzer; W. Klopper. Ionized, electron-attached, and excited states of molecular

systems with spin–orbit coupling: Two-component gw and bethe–salpeter implemen-
tations. J. Chem. Phys., 150(20), 204116, (2019).

[30] C. Holzer; W. Klopper. Communication: A hybrid bethe–salpeter/time-dependent

density-functional-theory approach for excitation energies. J. Chem. Phys., 149(10),
101101, (2018).

[31] X. Gui; C. Holzer; W. Klopper. Accuracy assessment of gw starting points for cal-
culating molecular excitation energies using the bethe–salpeter formalism. J. Chem.
Theory Comput., 14(4), 2127–2136, (2018).

[32] C. Holzer; X. Gui; M. E. Harding; G. Kresse; T. Helgaker; W. Klopper. Bethe–salpeter

correlation energies of atoms and molecules. J. Chem. Phys., 149(14), 144106, (2018).

[33] K. Krause; W. Klopper. Implementation of the bethe-salpeter equation in the turbo-

mole program. J. Comput. Chem., 38, 383–388, (2017).

[34] M. J. van Setten; F. Weigend; F. Evers. The gw-method for quantum chemistry
applications: Theory and implementation. J. Chem. Theory Comput, 9, 232, (2013).

[35] M. Kühn; F. Weigend. One-Electron Energies from the Two-Component GW Method.

J. Chem. Theory Comput., 11, 969–979, (2015).

[36] C. Holzer; A. M. Teale; F. Hampe; S. Stopkowicz; T. Helgaker; W. Klopper. Gw

quasiparticle energies of atoms in strong magnetic fields. J. Chem. Phys., 150(21),
214112, (2019).

[37] M. Kehry; Y. J. Franzke; C. Holzer; W. Klopper. Quasirelativistic two-component

core excitations and polarisabilities from a damped-response formulation of the bethe–
salpeter equation. Mol. Phys., 118(21-22), e1755064, (2020).
528 BIBLIOGRAPHY

[38] F. Furche; R. Ahlrichs. Adiabatic time-dependent density functional methods for

excited state properties. J. Chem. Phys., 117(16), 7433–7447, (2002).

[39] M. Häser; R. Ahlrichs; H. P. Baron; P. Weis; H. Horn. Direct computation of second-

order SCF properties of large molecules on workstation computers with an application
to large carbon clusters. Theor. Chim. Acta, 83(5–6), 455–470, (1992).

[40] M. Kollwitz; J. Gauss. A direct implementation of the GIAO-MBPT(2) method for

calculating NMR chemical shifts. Application to the naphthalenium and and anthrace-
nium ions. Chem. Phys. Lett., 260(5–6), 639–646, (1996).

[41] Y. J. Franzke; F. Weigend. NMR shielding tensors and chemical shifts in scalar-
relativistic local exact two-component theory. J. Chem. Theory Comput., 15(2), 1028–
1043, (2019).

[42] K. Reiter; F. Mack; F. Weigend. Calculation of magnetic shielding constants with

meta-gga functionals employing the multipole-accelerated resolution of the identity:
Implementation and assessment of accuracy and efficiency. J. Chem. Theory Comput.,
14(1), 191–197, (2018).

[43] K. Reiter; M. Kühn; F. Weigend. Vibrational circular dichroism spectra for large
molecules and molecules with heavy elements. J. Chem. Phys., 146(5), 054102, (2017).

[44] C. van Wüllen. Shared-memory parallelization of the TURBOMOLE programs AO-

FORCE, ESCF, and EGRAD: How to quickly parallelize legacy code. J. Comput.
Chem., 32, 1195–1201, (2011).

[45] M. von Arnim; R. Ahlrichs. Geometry optimization in generalized natural internal

coordinates. J. Chem. Phys., 111(20), 9183–9190, (1999).

[46] P. Pulay; G. Fogarasi; F. Pang; J. E. Boggs. Systematic ab initio gradient calculation of

molecular geometries, force constants, and dipole moment derivatives. J. Am. Chem.
Soc., 101(10), 2550–2560, (1979).

[47] M. Dolg; U. Wedig; H. Stoll; H. Preuß. Energy-adjusted ab initio pseudopotentials for

the first row transition elements. J. Chem. Phys., 86(2), 866–872, (1986).

[48] C. C. J. Roothaan. Self-consistent field theory for open shells of electronic systems.
Rev. Mod. Phys., 32(2), 179–185, (1960).

[49] R. Ahlrichs; F. Furche; S. Grimme. Comment on “Assessment of exchange correlation

functionals”. Chem. Phys. Lett., 325(1–3), 317–321, (2000).

[50] M. Sierka; A. Hogekamp; R. Ahlrichs. Fast evaluation of the Coulomb potential for
electron densities using multipole accelerated resolution of identity approximation. J.
Chem. Phys., 118(20), 9136–9148, (2003).

[51] F. Weigend. A fully direct RI-HF algorithm: Implementation, optimised auxiliary

basis sets, demonstration of accuracy and efficiency. Phys. Chem. Chem. Phys., 4(18),
4285–4291, (2002).
BIBLIOGRAPHY 529

[52] R. Fletcher. Practical Methods of Optimization. Unconstrained Optimization. Band 1.

Wiley: New York, 1980.

[53] T. Helgaker. Transition-state optimizations by trust-region image minimization.

Chem. Phys. Lett., 182(5), 503–510, (1991).

[54] F. Jensen. Locating transition structures by mode following: A comparison of six

methods on the Ar8 Lennard-Jones potential. J. Chem. Phys., 102(17), 6706–6718,
(1995).

[55] P. Császár; P. Pulay. Geometry optimization by direct inversion in the iterative sub-
space. J. Mol. Struct., 114, 31–34, (1984).

[56] R. Fletcher. A new approach to variable metric algorithms. Comput. J., 13(3), 317–
322, (1970).

[57] H. B. Schlegel. Optimization of equilibrium geometries and transition structures. J.

Comput. Chem., 3(2), 214–218, (1982).

[58] H. B. Schlegel. Estimating the hessian for gradient-type geometry optimizations.

Theor. Chim. Acta, 66(5), 333–340, (1984).

[59] M. Ehrig. Diplomarbeit. Master’s thesis, Universität Karlsruhe, 1990.

[60] T. Koga; H. Kobayashi. Exponent optimization by uniform scaling technique. J.

Chem. Phys., 82(3), 1437–1439, (1985).

[61] A. K. Rappé; W. A. Goddard III. Charge equilibration for molecular dynamics simu-
lations. J. Phys. Chem., 95(8), 3358–3363, (1991).

[62] C. G. Broyden. The convergence of a class of double-rank minimization algorithms 1.

General considerations. J. Inst. Math. Appl., 6(1), 76–90, (1970).

[63] D. Goldfarb. A family of variable-metric methods derived by variational means. Math.

Comput., 24(109), 23–26, (1970).

[64] D. F. Shanno. Conditioning of quasi-newton methods for function minimization. Math.

Comput., 24(111), 647–656, (1970).

[65] P. Pulay. Convergence acceleration of iterative sequences. the case of SCF iteration.
Chem. Phys. Lett., 73(2), 393–398, (1980).

[66] S. Grimme; C. Bannwarth; P. Shushkov. A robust and accurate tight-binding quantum

chemical method for structures, vibrational frequencies, and noncovalent interactions
of large molecular systems parametrized for all spd-block elements (z = 1–86). J.
Chem. Theory Comput., 13, 1989–2009, (2017).

[67] C. Bannwarth; S. Ehlert; S. Grimme. Gfn2-xtb—an accurate and broadly

parametrized self-consistent tight-binding quantum chemical method with multipole
electrostatics and density-dependent dispersion contributions. J. Chem. Theory Com-
put., 15, 1652–1671, (2019).
530 BIBLIOGRAPHY

[68] M. P. Allen; D. J. Tildesley. Computer Simulation of Liquids. Oxford University Press:

Oxford, 1987.

[69] M. Sierka. Synergy between theory and experiment in structure resolution of low-
dimensional oxides. Prog. Surf. Sci., 85, 398–434, (2010).

[70] T. Halgren; W. Lipscomb. Synchronous-transit method for determining reaction path-

ways and locating molecular transition-states. Chem. Phys. Lett., 49(2), 225–232,
(1977).

[71] R. Elber; M. Karplus. A method for determining reaction paths in large molecules -
application to myoglobin. Chem. Phys. Lett., 139(5), 375–380, (1987).

[72] M. G; H. Jonsson. Quantum and thermal effects in h-2 dissociative adsorption -

evaluation of free-energy barriers in multidimensional quantum-systems. Phys. Rev.
Lett., 72(7), 1124–1127, (1994).

[73] G. Henkelman; H. Jonsson. Improved tangent estimate in the nudged elastic band
method for finding minimum energy paths and saddle points. J. Chem. Phys., 113(22),
9978–9985, (2000).

[74] E. Weinan; W. Ren; E. Vanden-Eijnden. String method for the study of rare events.
Phys. Rev. B, 66(5), 052301, (2002).

[75] P. Plessow. Reaction Path Optimization without NEB Springs or Interpolation Algo-
rithms. J. Chem. Theory Comput., 9(3), 1305–1310, (2013).

[76] K. Eichkorn; O. Treutler; H. Öhm; M. Häser; R. Ahlrichs. Auxiliary basis sets to ap-
proximate Coulomb potentials (erratum, 1995, 242, 283). Chem. Phys. Lett., 242(6),
652–660, (1995).

[77] J. A. Pople; R. K. Nesbet. Self-consistent orbitals for radicals. J. Chem. Phys., 22(3),
571–572, (1954).

[78] J. Čižek; J. Paldus. Stability conditions for solutions of Hartree-Fock equations for
atomic and molecular systems. application to pi-electron model of cyclic plyenes. J.
Chem. Phys., 47(10), 3976–3985, (1967).

[79] F. Neese; F. Wennmohs; A. Hansen; U. Becker. Efficient, approximate and parallel

Hartree–Fock and hybrid DFT calculations. A ’chain-of-spheres’ algorithm for the
Hartree–Fock exchange. Chem. Phys., 356, 98–109, (2009).

[80] U. Ekström; L. Visscher; R. Bast; A. J. Thorvaldsen; K. Ruud. Arbitrary-order density

functional response theory from automatic differentiation. J. Chem. Theory Comput.,
6, 1971–1980, (2010).

[81] Y. Zhao; D. G. Truhlar. The M06 suite of density functionals for main group ther-
mochemistry, thermochemical kinetics, noncovalent interactions, excited states, and
transition elements: two new functionals and systematic testing of four M06-class
functionals and 12 other functionals. Theor. Chem. Acc., 120, 215–241, (2008).
BIBLIOGRAPHY 531

[82] S. Lehtola; C. Steigemann; M. J. T. Oliveira; M. A. L. Marques. Recent developments

in LIBXC - A comprehensive library of functionals for density functional theory. Soft-
wareX, 7, 1–5, (2018).

[83] J.-D. Chaia; M. Head-Gordon. Systematic optimization of long–range corrected hybrid

density functionals. J. Chem. Phys., 128, 084106, (2008).

[84] P. A. M. Dirac. Quantum mechanics of many-electron systems. Proc. Royal Soc.

(London) A, 123(792), 714–733, (1929).

[85] J. C. Slater. A simplification of the Hartree-Fock method. Phys. Rev., 81(3), 385–390,
(1951).

[86] S. Vosko; L. Wilk; M. Nusair. Accurate spin-dependent electron-liquid correlation

energies for local spin density calculations: a critical analysis. Can. J. Phys., 58(8),
1200–1211, (1980).

[87] J. P. Perdew; Y. Wang. Accurate and simple analytic representation of the electron-gas
correlation energy. Phys. Rev. B, 45(23), 13244–13249, (1992).

[88] A. D. Becke. Density-functional exchange-energy approximation with correct asymp-

totic behaviour. Phys. Rev. A, 38(6), 3098–3100, (1988).

[89] C. Lee; W. Yang; R. G. Parr. Development of the Colle-Salvetti correlation-energy

formula into a functional of the electron density. Phys. Rev. B, 37(2), 785–789, (1988).

[90] J. P. Perdew. Density-functional approximation for the correlation-energy of the in-

homogenous electron gas. Phys. Rev. B, 33(12), 8822–8824, (1986).

[91] J. P. Perdew; K. Burke; M. Ernzerhof. Generalized gradient approximation made

simple. Phys. Rev. Lett., 77(18), 3865–3868, (1996).

[92] J. Tao; J. P. Perdew; V. N. Staroverov; G. E. Scuseria. Climbing the density functional

ladder: Nonempirical meta–generalized gradient approximation designed for molecules
and solids. Phys. Rev. Lett., 91(14), 146401, (2003).

[93] J. Sun; A. Ruzsinszky; J. P. Perdew. Strongly constrained and appropriately normed

semilocal density functional. Phys. Rev. Lett., 115, 036402, (2015).

[94] J. G. Brandenburg; J. E. Bates; J. Sun; J. P. Perdew. Benchmark tests of a strongly

constrained semilocal functional with a long-range dispersion correction. Phys. Rev.
B, 94, 115144, (2016).

[95] A. D. Becke. A new mixing of Hartree-Fock and local density-functional theories. J.

Chem. Phys., 98(2), 1372–1377, (1993).

[96] A. D. Becke. Density-functional thermochemistry. III. The role of exact exchange. J.

Chem. Phys., 98(7), 5648–5652, (1993).

[97] J. P. Perdew; M. Ernzerhof; K. Burke. Rationale for mixing exact exchange with
density functional approximations. J. Chem. Phys., 105(22), 9982–9985, (1996).
532 BIBLIOGRAPHY

[98] V. N. Staroverov; G. E. Scuseria; J. Tao; J. P. Perdew. Comparative assessment of a

new nonempirical density functional: Molecules and hydrogen-bonded complexes. J.
Chem. Phys., 119(23), 12129–12137, (2003).

[99] S. Grimme. Semiempirical hybrid density functional with perturbative second-order

correlation. J. Chem. Phys., 124, 034108, (2006).

[100] A. Görling; M. Levy. Correlation-energy functional and its high-density limit obtained
from a coupling-constant perturbation expansion. Phys. Rev. B, 47, 13105, (1993).

[101] A. Görling; M. Levy. Exact Kohn-Sham scheme based on perturbation theory.

Phys. Rev. A, 50, 196, (1994).

[102] J. Jaramillo; G. E. Scuseria; M. Ernzerhof. Local hybrid functionals. J. Chem. Phys.,

118, 1068–1073, (2003).

[103] H. Bahmann; M. Kaupp. Efficient self-consistent implementation of local hybrid func-

tionals. J. Chem. Theory Comput., 11, 1540–1548, (2015).

[104] S. Klawohn; H. Bahmann; M. Kaupp. Implementation of molecular gradients for

local hybrid density functionals using seminumerical integration techniques. J. Chem.
Theory Comput., 12, 4254–4262, (2016).

[105] T. M. Maier; H. Bahmann; M. Kaupp. Efficient semi-numerical implementation of

global and local hybrid functionals for time-dependent density functional theory. J.
Chem. Theory Comput., 11, 4226–4237, (2015).

[106] F. Mack; C. J. Schattenberg; M. Kaupp; F. Weigend. Nuclear spin–spin couplings:

Efficient evaluation of exact exchange and extension to local hybrid functionals. J.
Phys. Chem. A, 124(41), 8529–8539, (2020).

[107] C. Holzer. An improved seminumerical coulomb and exchange algorithm for properties
and excited states in modern density functional theory. J. Chem. Phys., 153(18),
184115, (2020).

[108] C. Holzer; Y. J. Franzke; M. Kehry. Assessing the accuracy of local hybrid density
functional approximations for molecular response properties. J. Chem. Theory Com-
put., 17(5), 2928–2947, (2021).

[109] R. Grotjahn; F. Furche; M. Kaupp. Development and implementation of excited-state

gradients for local hybrid functionals. J. Chem. Theory Comput., 15, 5508–5522,
(2019).

[110] C. J. Schattenberg; K. Reiter; F. Weigend; M. Kaupp. An efficient coupled-perturbed

kohn–sham implementation of nmr chemical shift computations with local hybrid func-
tionals and gauge-including atomic orbitals. J. Chem. Theory Comput., 16(2), 931–
943, (2020).

[111] P. Plessow; F. Weigend. Seminumerical calculation of the Hartree-Fock exchange

matrix: Application to two-component procedures and efficient evaluation of local
hybrid density functionals. J. Comput. Chem., 33, 810, (2012).
BIBLIOGRAPHY 533

[112] H. Bahmann; A. Rodenberg; A. V. Arbuznikov; M. Kaupp. A thermochemically

competitive local hybrid functional without gradient corrections. J. Chem. Phys.,
126, 011103, (2007).

[113] A. V. Arbuznikov; M. Kaupp. Local hybrid exchange-correlation functionals based on

the dimensionless density gradient. Chem. Phys. Lett., 440, 160–168, (2007).

[114] A. V. Arbuznikov; M. Kaupp. Importance of the correlation contribution for local

hybrid functionals: Range separation and self-interaction corrections. J. Chem. Phys.,
136, 014111, (2012).

[115] A. V. Arbuznikov; M. Kaupp. Towards improved local hybrid functionals by calibra-

tion of exchange-energy densities. J. Chem. Phys., 141, 204101, (2014).

[116] T. M. Maier; M. Haasler; A. V. Arbuznikov; M. Kaupp. New approaches for the cali-
bration of exchange-energy densities in local hybrid functionals. Phys. Chem. Chem.
Phys., 18, 21133–21144, (2016).

[117] M. Haasler; T. M. Maier; R. Grotjahn; S. Gückel; A. V. Arbuznikov; M. Kaupp. A local

hybrid functional with wide applicability and good balance between (de)localization
and left–right correlation. jctc, 16, 5645–5657, (2020).

[118] A. D. Becke. Density-functional thermochemistry. IV. A new dynamical correlation

functional and implications for exact-exchange mixing. J. Chem. Phys., 104, 1040–
1046, (1996).

[119] J. P. Perdew; V. N. Staroverov; J. Tao; G. E. Scuseria. Density functional with full

exact exchange, balanced nonlocality of correlation, and constraint satisfaction. Phys.
Rev. A, 78, 052513, (2008).

[120] E. R. Johnson. Local-hybrid functional based on the correlation length. J. Chem.

Phys., 141(12), 124120, (2014).

[121] K. Theilacker; A. V. Arbuznikov; M. Kaupp. Gauge effects in local hybrid functionals

evaluated for weak interactions and the gmtkn30 test set. Mol. Phys., 114, 1118,
(2016).

[122] M. K. Armbruster; F. Weigend; C. van Wüllen; W. Klopper. Self-consistent treatment

of spin-orbit interactions with efficient hartree-fock and density functional methods.
Phys. Chem. Chem. Phys., 10, 1748–1756, (2008).

[123] D. Peng; M. Reiher. Exact decoupling of the relativistic fock operator. Theor. Chem.
Acc., 131, 1081, (2012).

[124] D. Peng; M. Reiher. Local relativistic exact decoupling. J. Chem. Phys., 136, 244108,
(2012).

[125] D. Peng; N. Middendorf; F. Weigend; M. Reiher. An efficient implementation of two-

component relativistic exact-decoupling methods for large molecules. J. Chem. Phys.,
138, 184105, (2013).
534 BIBLIOGRAPHY

[126] L. Visscher; K. G. Dyall. Dirac-fock atomic electronic structure calculations using

different nuclear charge distributions. Atom. Data Nucl. Data Tabl., 67, 207, (1997).

[127] J. C. Boettger. Approximate two-electron spin-orbit coupling term for density-

functional-theory dft calculations using the douglas-kroll-hess transformation. Phys.
Rev. B, 62, 7809–7815, (2000).

[128] M. Filatov; W. Zou; D. Cremer. Spin-orbit coupling calculations with the two-
component normalized elimination of the small component method. J. Chem. Phys.,
139, 014106, (2013).

[129] W. Zou; M. Filatov; D. Cremer. Analytical energy gradient for the two-component
normalized elimination of the small component method. J. Chem. Phys., 142, 214106,
(2015).

[130] Y. J. Franzke; N. Middendorf; F. Weigend. Efficient implementation of one- and

two-component analytical energy gradients in exact two-component theory. J. Chem.
Phys., 148, 104110, (2018).

[131] F. Weigend; A. Baldes. Segmented contracted basis sets for one- and two-component
Dirac-Fock effective core potentials. J. Chem. Phys., 133, 174102, (2010).

[132] M. Dolg; H. Stoll; H. Preuss. Energy-adjusted ab initio pseudopotentials for the rare
earth elements. J. Chem. Phys., 90, 1730, (1989).

[133] R. Gulde; P. Pollak; F. Weigend. Error-Balanced Segmented Contracted Basis Sets

of Double-Zeta to Quadruple-Zeta Valence Quality for the Lanthanides. J. Chem.
Theory Comput., 8, 4062, (2012).

[134] W. Küchle; M. Dolg; H. Stoll; H. Preuß. Energy-adjusted pseudopotentials for the

actinides. parameter sets and test calculations for thorium and thorium monoxide. J.
Chem. Phys., 100, 7535, (1994).

[135] X. Cao; M. Dolg; H. Stoll. Valence basis sets for relativistic energy-consistent small-
core actinide pseudopotentials. J. Chem. Phys., 118, 487, (2003).

[136] A. Baldes; F. Weigend. Efficient two-component self-consistent field procedures and

gradients: implementation in turbomole and application to Au20-. Mol. Phys., 111,
2617–2624, (2013).

[137] P. Pollak; F. Weigend. Segmented contracted error-consistent basis sets of double-

and triple-ζ valence quality for one- and two-component relativistic all-electron calcu-
lations. J. Chem. Theory Comput., 13, 3696 – 3705, (2017).

[138] Y. J. Franzke; R. Treß; T. M. Pazdera; F. Weigend. Error-consistent segmented

contracted all-electron relativistic basis sets of double- and triple-zeta quality for NMR
shielding constants. Phys. Chem. Chem. Phys., 21, 16658–16664, (2019).

[139] Y. J. Franzke; L. Spiske; P. Pollak; F. Weigend. Segmented contracted error-consistent

basis sets of quadruple-ζ valence quality for one- and two-component relativistic all-
electron calculations. J. Chem. Theory Comput., 16, 5658–5674, (2020).
BIBLIOGRAPHY 535

[140] M. Reiher; A. Wolf. Exact decoupling of the Dirac Hamiltonian. I. General theory. J.
Chem. Phys., 121, 2037–2047, (2004).

[141] M. Reiher; A. Wolf. Exact decoupling of the Dirac Hamiltonian. II. The generalized
Douglas-Kroll-Hess transformation up to arbitrary order. J. Chem. Phys., 121, 10945–
10956, (2004).

[142] M. Reiher. Douglas-Kroll-Hess Theory: a relativistic electrons-only theory for chem-

istry. Theor. Chem. Acc., 116, 241–252, (2006).

[143] J. Hepburn; G. Scoles; R. Penco. A simple but reliable method for the prediction of
intermolecular potentials. Chem. Phys. Lett., 36, 451–456, (1975).

[144] R. Ahlrichs; R. Penco; G. Scoles. Intermolecular forces in simple systems. Chem.

Phys., 19, 119–130, (1977).

[145] S. Grimme. Accurate Description of van der Waals Complexes by Density Func-
tional Theory Including Empirical Corrections. J. Comput. Chem., 25(12), 1463–1473,
(2004).

[146] S. Grimme. Semiempirical GGA-type density functional constructed with a long-range

dispersion contribution. J. Comput. Chem., 27(15), 1787–1799, (2006).

[147] S. Grimme; J. Antony; S. Ehrlich; H. Krieg. A consistent and accurate ab initio

parametrization of density functional dispersion correction (DFT-D) for the 94 ele-
ments H-Pu. J. Chem. Phys., 132, 154104, (2010).

[148] E. Caldeweyher; C. Bannwarth; S. Grimme. Extension of the d3 dispersion coefficient

model. J. Chem. Phys., 147, 034112, (2017).

[149] E. Caldeweyher; S. Ehlert; A. Hansen; H. Neugebauer; S. Spicher; C. Bannwarth; S.

Grimme. A generally applicable atomic-charge dependent london dispersion correc-
tion. J. Chem. Phys., 150, 154122, (2019).

[150] S. Grimme; S. Ehrlich; L. Goerigk. Effect of the damping function in dispersion

corrected density functional theory. J. Comput. Chem., 32, 1456–1465, (2011).

[151] O. A. Vydrov; T. V. Voorhis. Nonlocal van der waals density functional: The simpler
the better. J. Chem. Phys., 133, 244103, (2010).

[152] W. Hujo; S. Grimme. Comparison of the performance of dispersion-corrected density

functional theory for weak hydrogen bonds. Phys. Chem. Chem. Phys., 13, 13942–
13950, (2011).

[153] P. Su; H. Li. Energy decomposition analysis of covalent bonds and intermolecular
interactions. J. Chem. Phys., 131, 014102, (2009).

[154] R. Łazarski; A. M. Burow; M. Sierka. Density functional theory for molecular and
periodic systems using density fitting and continuous fast multipole methods. J. Chem.
Theory Comput., 11, 3029–3041, (2015).
536 BIBLIOGRAPHY

[155] R. Łazarski; A. M. Burow; L. Grajciar; M. Sierka. Density functional theory for

molecular and periodic systems using density fitting and continuous fast multipole
method: Analytical gradients. J. Comput. Chem., 37(28), 2518–2526, (2016).
[156] A. M. Burow; M. Sierka. Linear scaling hierarchical integration scheme for the
exchange-correlation term in molecular and periodic systems. J. Chem. Theory Com-
put., 7, 3097–3104, (2011).
[157] L. Grajciar. Low-memory iterative density fitting. J. Comput. Chem., 36, 1521–1535,
(2015).
[158] A. M. Burow; M. Sierka; F. Mohamed. Resolution of identity approximation for the
coulomb term in molecular and periodic systems. J. Chem. Phys., 131, 214101, (2009).
[159] C. Müller; M. Sharma; M. Sierka. Real-time time-dependent density functional theory
using density fitting and the continuous fast multipole method. Journal of Computa-
tional Chemistry, 41(30), 2573–2582, (2020).
[160] G. Kresse; J. Furthmüller. Efficiency of ab-initio total energy calculations for metals
and semiconductors using a plane-wave basis set. Comput. Mat. Sci., 6, 15–50, (1996).
[161] M. F. Peintinger; D. V. Oliveira; T. Bredow. Consistent gaussian basis sets of triple-
zeta valence with polarization quality for solid-state calculations. J. Comput. Chem.,
34, 451–459, (2013).
[162] F. Furche; D. Rappoport. Density functional methods for excited states: equilibrium
structure and electronic spectra. In M. Olivucci, Ed., Computational Photochemistry,
Band 16 von Computational and Theoretical Chemistry, Kapitel III. Elsevier, Amster-
dam, 2005.
[163] F. Furche. On the density matrix based approach to time-dependent density functional
theory. J. Chem. Phys., 114(14), 5982–5992, (2001).
[164] F. Furche; K. Burke. Time-dependent density functional theory in quantum chemistry.
Annual Reports in Computational Chemistry, 1, 19–30, (2005).
[165] D. Rappoport; F. Furche. Excited states and photochemistry. In M. A. L. Marques;
C. A. Ullrich; F. Nogueira; A. Rubio; K. Burke; E. K. U. Gross, Eds., Time-Dependent
Density Functional Theory, Kapitel 22. Springer, 2005.
[166] S. M. Parker; D. Rappoport; F. Furche. Quadratic Response Properties from TDDFT:
Trials and Tribulations. J. Chem. Theory Comput., 14(2), 807–819, (2018).
[167] J. E. Bates; F. Furche. Harnessing the meta-generalized gradient approximation for
time-dependent density functional theory. J. Chem. Phys., 137, 164105, (2012).
[168] S. Grimme; F. Furche; R. Ahlrichs. An improved method for density functional cal-
culations of the frecuency-dependent optical rotation. Chem. Phys. Lett., 361(3–4),
321–328, (2002).
[169] H. Weiss; R. Ahlrichs; M. Häser. A direct algorithm for self-consistent-field linear
response theory and application to C60 : Excitation energies, oscillator strengths, and
frequency-dependent polarizabilities. J. Chem. Phys., 99(2), 1262–1270, (1993).
BIBLIOGRAPHY 537

[170] D. Rappoport; F. Furche. Lagrangian approach to molecular vibrational raman inten-

sities using time-dependent hybrid density functional theory. J. Chem. Phys., 126(20),
201104, (2007).

[171] S. M. Parker; S. Roy; F. Furche. Multistate hybrid time-dependent density functional

theory with surface hopping accurately captures ultrafast thymine photodeactivation.
Phys. Chem. Chem. Phys., 21, 18999–19010, (2019).

[172] R. Bauernschmitt. Statische und dynamische Aspekte des Kohn-Sham-Formalismus:

Stabilität, statischer Response, Anregungsenergien. PhD thesis, Universität Karlsruhe,
1997.

[173] Calculation of NMR and EPR Parameters: Theory and Applications. M. Kaupp; M.
Bühl; V. G. Malkin, Eds. Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, 2004.

[174] F. Furche. Dichtefunktionalmethoden für elektronisch angeregte Moleküle. Theorie–

Implementierung–Anwendung. PhD thesis, Universität Karlsruhe, 2002.

[175] E. R. Davidson. The iterative calculation of a few of the lowest eigenvalues and
corresponding eigenvectors of large real-symmetric matrices. J. Comp. Phys., 17(1),
87–94, (1975).

[176] F. Wang; T. Ziegler. Time-dependent density functional theory based on a noncollinear

formulation of the exchange-correlation potential. J. Chem. Phys., 121(24), 12191–
12196, (2004).

[177] M. Kühn; F. Weigend. Phosphorescence energies of organic light-emitting diodes

from spin-flip Tamm-Dancoff approximation time-dependent density functional the-
ory. Chem. Phys. Chem., 12, 3331–3336, (2011).

[178] M. Kühn; F. Weigend. Phosphorescence lifetimes of organic light-emitting diodes from

two-component time-dependent density functional theory. J. Chem. Phys., 141(22),
224302, (2014).

[179] N. F. Ramsey. Electron coupled interactions between nuclear spins in molecules. Phys.
Rev., 91, 303–307, (1953).

[180] F. Jensen. The optimum contraction of basis sets for calculating spin–spin coupling
constants. Theor. Chem. Acc., 126(5), 371–382, (2010).

[181] P. A. Aggelund; S. P. A. Sauer; F. Jensen. Development of polarization consistent

basis sets for spin-spin coupling constant calculations for the atoms Li, Be, Na, and
Mg. J. Chem. Phys., 149(4), 044117, (2018).

[182] F. Haase; R. Ahlrichs. Semidirect MP2 gradient evaluation on workstation computers:

The MPGRAD program. J. Comp. Chem., 14(8), 907–912, (1993).

[183] F. Weigend; M. Häser. RI-MP2: first derivatives and global consistency. Theor. Chem.
Acc., 97(1–4), 331–340, (1997).

[184] F. Weigend; M. Häser; H. Patzelt; R. Ahlrichs. RI-MP2: Optimized auxiliary basis

sets and demonstration of efficiency. Chem. Phys. Letters, 294(1–3), 143–152, (1998).
538 BIBLIOGRAPHY

[185] F. Weigend; A. Köhn; C. Hättig. Efficient use of the correlation consistent basis sets
in resolution of the identity MP2 calculations. J. Chem. Phys., 116(8), 3175–3183,
(2001).

[186] C. L. Janssen; I. M. B. Nielsen. New diagnostics for coupled-cluster and Møller-Plesset

perturbation theory. Chem. Phys. Lett., 290(4–6), 423–430, (1998).

[187] I. M. B. Nielsen; C. L. Janssen. Double-substitution-based diagnostics for coupled-

cluster and Møller-Plesset perturbation theory. Chem. Phys. Lett., 310(5–6), 568–576,
(1999).

[188] F. A. Bischoff; S. Höfener; A. Glöß; W. Klopper. Explicitly correlated second-order

perturbation theory calculations on molecules containing heavy main-group elements.
Theor. Chem. Acc., 121(1), 11–19, (2008).

[189] D. P. Tew. Explicitly correlated coupled-cluster theory with brueckner orbitals. J.

Chem. Phys., 145(7), 074103, (2016).

[190] F. R. Manby. Density fitting in second-order linear-r12 Møller-Plesset perturbation

theory. J. Chem. Phys., 119(9), 4607–4613, (2003).

[191] E. F. Valeev. Improving on the resolution of the identity in linear R12 ab initio
theories. Chem. Phys. Lett., 395(4-6), 190–195, (2004).

[192] K. E. Yousaf; K. A. Peterson. Optimized auxiliary basis sets for explicitly correlated
methods. J. Chem. Phys., 129(18), 184108, (2008).

[193] K. A. Peterson; T. B. Adler; H.-J. Werner. Systematically convergent basis sets for
explicitly correlated wavefunctions: The atoms H, He, B–Ne, and Al–Ar. J. Chem.
Phys., 128(8), 084102, (2008).

[194] W. Klopper; C. C. M. Samson. Explicitly correlated second-order Møller-Plesset meth-

ods with auxiliary basis sets. J. Chem. Phys., 116(15), 6397–6410, (2002).

[195] W. Klopper; W. Kutzelnigg. Møller-Plesset calculations taking care of the correlation

cusp. Chem. Phys. Lett., 134(1), 17–22, (1987).

[196] S. Ten-no. Explicitly correlated second order perturbation theory: Introduction of

a rational generator and numerical quadratures. J. Chem. Phys., 121(1), 117–129,
(2004).

[197] S. F. Boys. Localized orbitals and localized adjustment functions. In P.-O. Löwdin,
Ed., Quantum Theory of Atoms, Molecules and the Solid State, Page 253. Academic
Press, New York, 1966.

[198] J. Pipek; P. G. Mezey. A fast intrinsic localization procedure applicable for ab initio
and semiempirical linear combination of atomic orbital wave functions. J. Chem.
Phys., 90(9), 4916–4926, (1989).

[199] D. P. Tew; W. Klopper. New correlation factors for explicitly correlated electronic
wave functions. J. Chem. Phys., 123(7), 074101, (2005).
BIBLIOGRAPHY 539

[200] W. Klopper; B. Ruscic; D. P. Tew; F. A. Bischoff; S. Wolfsegger. Atomization energies

from coupled-cluster calculations augmented with explicitly-correlated perturbation
theory. Chem. Phys., 356(1–3), 14–24, (2009).

[201] S. Höfener; F. A. Bischoff; A. Glöß; W. Klopper. Slater-type geminals in explicitly-

correlated perturbation theory: application to n-alkanols and analysis of errors and
basis-set requirements. Phys. Chem. Chem. Phys., 10(23), 3390–3399, (2008).

[202] O. Christiansen; H. Koch; P. Jørgensen. The second-order approximate coupled cluster

singles and doubles model CC2. Chem. Phys. Lett., 243(5–6), 409–418, (1995).

[203] W. Klopper; F. R. Manby; S. Ten-no; E. F. Valeev. R12 methods in explicitly cor-

related molecular electronic structure theory. Int. Rev. Phys. Chem., 25(3), 427–468,
(2006).

[204] C. Hättig; A. Köhn. Transition moments and excited state first-order properties
in the second-order coupled cluster model CC2 using the resolution of the identity
approximation. J. Chem. Phys., 117(15), 6939–6951, (2002).

[205] T. Helgaker; P. Jørgensen; J. Olsen. Molecular Electronic-Structure Theory. Wiley:

New York, 2000.

[206] O. Christiansen; P. Jørgensen; C. Hättig. Response functions from Fourier compo-

nent variational perturbation theory applied to a time-averaged quasienergy. Int. J.
Quantum Chem., 68(1), 1–52, (1998).

[207] C. Hättig; P. Jørgensen. Derivation of coupled cluster excited states response func-
tions and multiphoton transition moments between two excited states as derivatives
of variational functionals. J. Chem. Phys., 109(21), 9219–9236, (1998).

[208] C. Hättig; O. Christiansen; P. Jørgensen. Multiphoton transition moments and absorp-

tion cross section in coupled cluster response theory employing variational transition
moment functionals. J. Chem. Phys., 108(20), 8331–8354, (1998).

[209] H. Koch; P. Jørgensen. Coupled cluster response functions. J. Chem. Phys., 93, 3333,
(1990).

[210] C. Hättig; O. Christiansen; P. Jørgensen. Coupled cluster response calculations of

twophoton transition probability rate constants for helium, neon and argon. J. Chem.
Phys., 108, 8355–8359, (1998).

[211] D. H. Friese; C. Hättig; K. Ruud. Calculation of two-photon absorption strengths with

the approximate coupled cluster singles and doubles model cc2 using the resolution-
of-identity approximation. Phys. Chem. Chem. Phys., 14, 1175–1184, (2012).

[212] B. Helmich-Paris; C. Hättig; C. van Wüllen. Spin-free cc2 implementation of induced

transitions between singlet ground and triplet excited states. J. Chem. Theory Com-
put., 12(4), 1892–1904, (2016).

[213] C. Hättig. Structure optimizations for excited states with correlated second-order
methods: CC2, CIS(D∞), and ADC(2). Adv. Quant. Chem., 50, 37–60, (2005).
540 BIBLIOGRAPHY

[214] S. Grimme; E. Ugorodina. Calculation of 0-0 excitation energies of organic molecules

by CIS(D) quantum chemical methods. Chem. Phys., 305, 223–230, (2004).
[215] Y. M. Rhee; M. Head-Gordon. Scaled second–order perturbation corrections to config-
uration interaction singles: Efficient and reliable excitation energy methods. J. Phys.
Chem. A, 111, 5314–5326, (2007).
[216] H. Fliegl; C. Hättig; W. Klopper. Coupled–cluster theory with simplified linear-r12
corrections: The CCSD(R12) model. J. Chem. Phys., 122, 084107, (2005).
[217] T. Shiozaki; M. Kamiya; S. Hirata; E. F. Valeev. Explicitly correlated coupled-cluster
singles and doubles method based on complete diagrammatic equations. J. Chem.
Phys., 129, 071101, (2008).
[218] A. Köhn; G. W. Richings; D. P. Tew. Implementation of the full explicitly correlated
coupled-cluster singles and doubles model CCSD-F12 with optimally reduced auxiliary
basis dependence. J. Chem. Phys., 129, 201103, (2008).
[219] T. B. Adler; G. Knizia; H.-J. Werner. A simple and efficient ccsd(t)-f12 approximation.
J. Chem. Phys., 127, 221106, (2007).
[220] G. Knizia; T. B. Adler; H.-J. Werner. Simplified ccsd(t)-f12 methods: Theory and
benchmarks. J. Chem. Phys., 130, 054104, (2009).
[221] M. Torheyden; E. F. Valeev. Variational formulation of perturbative explicitly-
correlated coupled-cluster methods. Phys. Chem. Chem. Phys., 10, 3410–3420, (2008).
[222] E. F. Valeev; D. Crawford. Simple coupled-cluster singles and doubles method with
perturbative inclusion of triples and explicitly correlated geminals: The ccsd(t)r12
¯
model. J. Chem. Phys., 128, 244113, (2008).
[223] K. D. Vogiatzis; E. C. Barnes; W. Klopper. Interference-corrected explicitly-correlated
second-order perturbation theory. Chem. Phys. Lett., 503(1-3), 157–161, (2011).
[224] D. P. Tew. Principal domains in local correlation theory. J. Chem. Theory Comput.,
15(12), 6597–6606, (2019).
[225] H. Eshuis; J. Yarkony; F. Furche. Fast computation of molecular random phase ap-
proximation correlation energies using resolution of the identity and imaginary fre-
quency integration. J. Chem. Phys., 132, 234114, (2010).
[226] H. Eshuis; J. E. Bates; F. Furche. Electron correlation methods based on the random
phase approximation. Theor. Chem. Acc., 131, 1084, (2012).
[227] A. M. Burow; J. E. Bates; F. Furche; H. Eshuis. Analytical first-order molecular
properties and forces within the adiabatic connection random phase approximation.
J. Chem. Theory Comput., 10(1), 180–194, (2014).
[228] G. P. Chen; V. K. Voora; M. M. Agee; S. G. Balasubramani; F. Furche. Random-phase
approximation methods. Annu. Rev. Phys. Chem., 68, 421–445, (2017).
[229] M. Kühn. Correlation Energies from the Two-component Random Phase Approxima-
tion. J. Chem. Theory Comput., 10, 623–633, (2014).
BIBLIOGRAPHY 541

[230] J. E. Bates; F. Furche. Random phase approximation renormalized many-body per-

turbation theory. J. Chem. Phys., 139(17), 171103, (2013).

[231] G. P. Chen; M. M. Agee; F. Furche. Performance and scope of perturbative corrections

to random-phase approximation energies. J. Chem. Theory Comput., 14(11), 5701–
5714, (2018).

[232] V. K. Voora; S. G. Balasubramani; F. Furche. Variational generalized kohn-sham

approach combining the random-phase-approximation and green’s-function methods.
Phys. Rev. A, 99, 012518, (2019).

[233] F. Furche. Molecular tests of the random phase approximation to the exchange-
correlation energy functional. Phys. Rev. B, 64, 195120, (2001).

[234] F. Furche. Developing the random phase approximation into a practical post-Kohn–
Sham correlation model. J. Chem. Phys., 129, 114105, (2008).

[235] V. K. Voora; R. Galhenage; J. C. Hemminger; F. Furche. Effective one-particle energies

from generalized kohn–sham random phase approximation: A direct approach for
computing and analyzing core ionization energies. The Journal of Chemical Physics,
151(13), 134106, (2019).

[236] P. Deglmann; F. Furche; R. Ahlrichs. An efficient implementation of second analytical

derivatives for density functional methods. Chem. Phys. Lett., 362(5–6), 511–518,
(2002).

[237] P. Deglmann; F. Furche. Efficient characterization of stationary points on potential

energy surfaces. J. Chem. Phys., 117(21), 9535–9538, (2002).

[238] M. Bürkle; J. Viljas; T. Hellmuth; E. Scheer; F. Weigend; G. Schön; F. Pauly. Influence

of vibrations on electron transport through nanoscale contacts. Phys. Status Solidi B,
250, 2468, (2013).

[239] T. Ziegler; G. Schreckenbach. Calculation of NMR shielding tensors using gauge-

including atomic orbitals and modern density functional theory. J. Phys. Chem.,
99(2), 606–611, (1995).

[240] M. Kollwitz; M. Häser; J. Gauss. Non-abelian point group symmetry in direct second-
order many-body perturbation theory calculations of nmr chemical shifts. J. Chem.
Phys., 108(20), 8295–8301, (1998).

[241] C. van Wüllen. On the use of effective core potentials in the calculation of magnetic
properties, such as magnetizabilites and magnetic shieldings. J. Chem. Phys., 136(11),
114110, (2012).

[242] J. Jusélius; D. Sundholm; J. Gauss. Calculation of current densities using gauge-

including atomic orbitals. J. Chem. Phys., 121(9), 3952–3963, (2004).

[243] H. Fliegl; S. Taubert; O. Lehtonen; D. Sundholm. The gauge including magnetically

induced current method. Phys. Chem. Chem. Phys., 13, 20500–20518, (2011).
542 BIBLIOGRAPHY

[244] D. Sundholm; H. Fliegl; R. J. F. Berger. Calculations of magnetically induced current

densities: theory and applications. Wiley Interdiscip. Rev.: Comput. Mol. Sci., 6(6),
639–678, (2016).
[245] A. Klamt; G. Schüürmann. COSMO: A new approach to dielectric screening in solvents
with explicit expressions for the screening energy and its gradient. J. Chem. Soc.
Perkin Trans.2, (5), 799–805, (1993).
[246] A. Klamt; C. Moya; J. Palomar. A comprehensive comparison of the iefpcm and
ss(v)pe continuum solvation methods with the cosmo approach. J. Chem. Theory
Comput., 11(9), 4220–4225, (2015).
[247] A. Klamt; M. Diedenhofen. A refined cavity construction algorithm for the conductor-
like screening model. J. Comput. Chem., 39(21), 1648–1655, (2018).
[248] A. Klamt; V. Jonas. Treatment of the outlying charge in continuum solvation models.
J. Chem. Phys., 105(22), 9972–9981, (1996).
[249] A. Klamt. Calculation of UV/Vis spectra in solution. J. Phys. Chem., 100(9), 3349–
3353, (1996).
[250] F. J. Olivares del Valle; J. Tomasi. Electron correlation and solvation effects. I. Basic
formulation and preliminary attempt to include the electron correlation in the quan-
tum mechanical polarizable continuum model so as to study solvation phenomena.
Chem. Phys., 150(2), 139–150, (1991).
[251] J. G. Ángyán. Rayleigh-Schrödinger perturbation theory for nonlinear Schrödinger
equations with linear perturbation. Int. J. Quantum Chem., 47(6), 469–483, (1993).
[252] J. G. Ángyán. Choosing between alternative MP2 algorithms in the self-consistent
reaction field theory of solvent effects. Chem. Phys. Lett., 241(1–2), 51–56, (1995).
[253] R. Cammi; B. Mennucci; J. Tomasi. Second-order Møller-Plesset analytical derivatives
for the polarizable continuum model using the relaxed density approach. J. Phys.
Chem. A, 103(45), 9100–9108, (1999).
[254] T. Schwabe; K. Sneskov; J. M. Olsen; J. Kongsted; O. Christiansen; C. Hättig. PERI-
CC2: A polarizable embedded RI-CC2 method. J. Chem. Theory Comput., 8, 3274–
3283, (2012).
[255] G. Scalmani; M. J. Frisch; B. Mennucci; J. Tomasi; R. Cammi; V. Barone. Geometries
and properties of excited states in the gas phase and in solution: Theory and applica-
tion of a time-dependent density functional theory polarizable continuum model. J.
Chem. Phys., 124(9), 094107, (2006).
[256] S. Sinnecker; A. Rajendran; A. Klamt; M. Diedenhofen; F. Neese. Calculation of sol-
vent shifts on electronic g-tensors with the Conductor-like Screening Model (COSMO)
and its self-consistent generalization to real solvents (Direct COSMO-RS). J. Phys.
Chem. A, 110, 2235–2245, (2006).
[257] F. Eckert; A. Klamt. Fast solvent screening via quantum chemistry: COSMO-RS
approach. AICHE Journal, 48, 369–385, (2002).
BIBLIOGRAPHY 543

[258] A. Klamt; V. Jonas; T. Bürger; J. C. W. Lohrenz. Refinement and parametrization

of COSMO-RS. J. Phys. Chem. A, 102, 5074–5085, (1998).

[259] S. K. Khani; A. M. Khah; C. Hättig. Cosmo-ri-adc(2) excitation energies and excited

state gradients. Phys. Chem. Chem. Phys., 20(24), 16354–16363, (2018).

[260] S. K. Khani; R. Faber; F. Santoro; S. Coriani; C. Hättig. Uv absorption and magnetic

circular dichroism spectra of purine, adenine, and guanine: A Coupled Cluster study in
vacuo and in aqueous solution. J. Chem. Theory Comput., 15(2), 1242–1254, (2019).

[261] P. Cortona. Self-consistently determined properties of solids without band-structure

calculations. Phys. Rev. B, 44, 8454, (1991).

[262] T. A. Wesolowski; A. Warshel. Frozen density functional approach for ab initio calcu-
lations of solvated molecules. J. Phys. Chem., 97, 8050, (1993).

[263] T. A. Wesolowski. In J. Leszczynski, Ed., Chemistry: Reviews of Current Trends,

Band 10, Page 1. World Scientific: Singapore, 2006, Singapore, 2006.

[264] T. A. Wesolowski; A. Warshel. Kohn–Sham equations with constrained electron

density: an iterative evaluation of the ground-state electron density of interacting
molecules. Chem. Phys. Lett., 248, 71, (1996).

[265] S. Laricchia; E. Fabiano; F. Della Sala. Frozen density embedding with hybrid func-
tionals. J. Chem. Phys., 133, 164111, (2010).

[266] S. Laricchia; E. Fabiano; F. Della Sala. Frozen density embedding calculations with the
orbital-dependent localized Hartree–Fock Kohn–Sham potential. Chem. Phys. Lett.,
518, 114, (2011).

[267] L. A. Constantin; E. Fabiano; S. Laricchia; F. Della Sala. Semiclassical neutral atom

as a reference system in density functional theory. Phys. Rev. Lett., 106, 186406,
(2011).

[268] S. Laricchia; E. Fabiano; L. A. Constantin; F. Della Sala. Generalized gradient ap-

proximations of the noninteracting kinetic energy from the semiclassical atom theory:
Rationalization of the accuracy of the frozen density embedding theory for nonbonded
interactions. J. Chem. Theory Comput., 7, 2439, (2011).

[269] A. Lembarki; H. Chermette. Obtaining a gradient-corrected kinetic-energy functional

from the Perdew-Wang exchange functional. Phys. Rev. A, 50, 5328, (1994).

[270] M. Sierka; A. Burow; J. Döbler; J. Sauer. Point defects in CeO2 and CaF2 investi-
gated using periodic electrostatic embedded cluster method. J. Chem. Phys., 130(17),
174710, (2009).

[271] K. N. Kudin; G. E. Scuseria. A fast multipole method for periodic systems with
arbitrary unit cell geometries. Chem. Phys. Lett., 283, 61–68, (1998).

[272] P. Ewald. Die Berechnung optischer und elektrostatischer Gitterpotentiale. Ann.

Phys., 64, 253–287, (1921).
544 BIBLIOGRAPHY

[273] J. M. Olsen; K. Aidas; J. Kongsted. Excited states in solution through polarizable

embedding. J. Chem. Theory Comput, 6, 3721–3734, (2010).

[274] K. Sneskov; T. Schwabe; J. Kongsted; O. Christiansen. The polarizable embedding

coupled cluster method. J. Chem. Phys., 134, 104108, (2011).

[275] K. Sneskov; T. Schwabe; J. Kongsted; O. Christiansen. Scrutinizing the effects of

polarization in QM/MM excited state calculations. Phys. Chem. Chem. Phys., 13,
18551–18560, (2011).

[276] K. Krause; W. Klopper. Communication: A simplified coupled-cluster lagrangian for

polarizable embedding. J. Chem. Phys., 144, 041101, (2016).

[277] L. Gagliardi; R. Lindh; G. Karlström. Local properties of quantum chemical systems:

The LoProp approach. J. Chem. Phys., 121, 4494–5000, (2004).

[278] B. Lunkenheimer; A. Köhn. Solvent effects on electronically excited states using

the conductor- like screening model and the second-order correlated method adc(2).
J. Chem. Theory Comput., 9, 977–994, (2015).

[279] A. M. Khah; S. K. Khani; C. Hättig. Analytic excited state gradients for the qm/mm
polarizable embedded second-order algebraic diagrammatic construction for the po-
larization propagator pe-adc(2). J. Chem. Theory Comput., 14, 4640–4650, (2018).

[280] A. E. Reed; R. B. Weinstock; F. Weinhold. Natural population analysis. J. Chem.

Phys., 83(2), 735–746, (1985).

[281] C. Ehrhardt; R. Ahlrichs. Population Analysis Based on Occupation Numbers. II. Re-
lationship between Shared Electron Numbers and Bond Energies and Characterization
of Hypervalent Contributions. Theor. Chim. Acta, 68(3), 231–245, (1985).

[282] K. Wiberg. Application of the pople-santry-segal cndo method to the cyclopropyl-

carbinyl and cyclobutyl cation and to bicyclobutane. Tetrahedron, 24(3), 1083 – 1096,
(1968).

[283] I.-M. Hoyvik; B. Jansik; P. Jorgensen. Orbital localization using fourth central moment
minimization. J. Chem. Phys., 137(22), (2012).

[284] G. Knizia. Intrinsic atomic orbitals: An unbiased bridge between quantum theory and
chemical concepts. J. Chem. Theory Comput., 9, 4834–4843, (2013).

[285] A. V. Luzanov; A. A. Sukhorukov; V. E. Úmanskii. Application of transition density

matrix for analysis of excited states. Theor. Exp. Chem., 10, 354, (1976).

[286] R. L. Martin. Natural transition orbitals. J. Chem. Phys., 118, 4775, (2003).

[287] F. Weigend; C. Schrodt. Atom-type assignment in molecule and clusters by pertu-

bation theory— A complement to X-ray structure analysis. Chem. Eur. J., 11(12),
3559–3564, (2005).

[288] A. D. Becke; K. E. Edgecombe. A simple measure of electron localization in atomic

and molecular systems. J. Chem. Phys., 92(9), 5397–5403, (1990).
BIBLIOGRAPHY 545

[289] F. D. Sala; A. Görling. Efficient localized Hartree-Fock methods as effective exact-

exchange Kohn-Sham methods for molecules. J. Chem. Phys., 115(13), 5718–5732,
(2001).

[290] A. Görling. Orbital- and state-dependent functionals in density-functional theory. J.

Chem. Phys., 123(6), 062203, (2005).

[291] S. Kümmel; L. Kronik. Orbital-dependent density functionals: Theory and applica-

tions. Rev. Mod. Phys., 80(1), 3, (2008).

[292] F. Della Sala. Orbital-dependent exact-exchange mmethods in density functional the-

ory. In M. Springborg, Ed., Chemical Modelling: Applications and Theory, Band 7,
Pages 115–161. Royal Society of Chemistry, 2010.

[293] A. Heßelmann; A. W. Götz; F. Della Sala; A. Görling. Numerically stable optimized

effective potential method with balanced gaussian basis sets. J. Chem. Phys., 127(5),
054102, (2007).

[294] J. B. Krieger; Y. Li; G. J. Iafrate. Construction and application of an accurate local

spin-polarized kohn-sham potential with integer discontinuity: Exchange-only theory.
Phys. Rev. A, 45, 101, (1992).

[295] O. V. Gritsenko; E. J. Baerends. Orbital structure of the kohn-sham exchange po-

tential and exchange kernel and the field-counteracting potential for molecules in an
electric field. Phys. Rev. A, 64, 042506, (2001).

[296] A. F. Izmaylov; V. N. Staroverov; G. E. Scuseria; E. R. Davidson; G. Stoltz; E. Cancès.

The effective local potential method: Implementation for molecules and relation to
approximate optimized effective potential techniques. J. Chem. Phys., 126(8), 084107,
(2007).

[297] F. Della Sala; A. Görling. The asymptotic region of the Kohn-Sham exchange potential
in molecules. J. Chem. Phys., 116(13), 5374–5388, (2002).

[298] F. Della Sala; A. Görling. Asymptotic behavior of the Kohn-Sham exchange potential.
Phys. Rev. Lett., 89, 033003, (2002).

[299] W. Hieringer; F. Della Sala; A. Görling. Density-functional calculations of NMR

shielding constants using the localized Hartree–Fock method. Chem. Phys. Lett.,
383(1-2), 115–121, (2004).

[300] E. Fabiano; M. Piacenza; S. D’Agostino; F. Della Sala. Towards an accurate descrip-

tion of the electronic properties of the biphenylthiol/gold interface: The role of exact
exchange. J. Chem. Phys., 131(23), 234101, (2009).

[301] F. Della Sala; E. Fabiano. Accurate singlet and triplet excitation energies using the
Localized Hartree-Fock Kohn-Sham potential. Chem. Phys., 391(1), 19 – 26, (2011).

[302] E. Tapavicza; F. Furche; D. Sundholm. Importance of vibronic effects in the uv-vis

spectrum of the 7, 7, 8, 8-tetracyanoquinodimethane anion. J. Chem. Theory Comput.,
12(10), 5058–5066, (2016).
546 BIBLIOGRAPHY

[303] F. Duschinsky. The importance of the electron spectrum in multi atomic molecules.
concerning the Franck-Condon principle. Acta Physicochim. URSS, 7, 551, (1937).

[304] F. G. Mehler. Ueber die Entwicklung einer Function von beliebig vielen Variabeln
nach Laplaceschen Functionen höherer Ordnung. J. Reine Angew. Math., 66, 161–
176, (1866).

[305] M. Etinski; J. Tatchen; C. M. Marian. Time-dependent approaches for the calculation

of intersystem crossing rates. J. Chem. Phys., 134(15), 154105, (2011).

[306] E. Tapavicza. Generating function approach to single vibronic level fluorescence spec-
tra. J. Phys. Chem. Lett., 10, 6003–6009, (2019).

[307] J. J. Dongarra; J. Du Croz; S. Hammarling; I. S. Duff. A set of level 3 basic linear

algebra subprograms. ACM Trans. Math. Software, 16, 1–17, (1990).

[308] W. H. Press; S. A. Teukolsky; W. T. Vetterling; B. P. Flannery. Numerical recipes in

C. Band 2. Cambridge university press: Cambridge, 1996.

[309] R. C. Hilborn. Einstein coefficients, cross sections, f values, dipole moments, and all
that. Am. J. Phys., 50(11), 982–986, (1982).

[310] O. Treutler; R. Ahlrichs. Efficient molecular numerical integration schemes. J. Chem.

Phys., 102(1), 346–354, (1995).

[311] A. D. Becke. A multicenter numerical integration scheme for polyatomic molecules.

J. Chem. Phys., 88(4), 2547–2553, (1988).

[312] A. T. B. Gilbert; N. A. Besley; P. M. W. Gill. Self-Consistent Field Calculations of

Excited States Using the Maximum Overlap Method (MOM). J. Phys. Chem. A, 112,
13164–13171, (2008).

[313] J. S. Andrews; D. Jayatilaka; R. G. Bone; N. C. Handy; R. D. Amos. Spin con-

tamination in single-determinant wavefunctions. Chem. Phys. Lett., 183(5), 423–431,
(1991).

[314] A. Wolf; M. Reiher; B. Hess. The generalized Douglas-Kroll transformation. J. Chem.

Phys., 117, 9215–9226, (2002).

[315] L. Visscher; K. G. Dyall. Dirac-fock atomic electronic structure calculations using

different nuclear charge distributions. At. Data Nucl. Data Tables, 67, 207–224, (1997).

[316] R. Send; F. Furche. First-order nonadiabatic couplings from time-dependent hybrid

density functional response theory: Consistent formalism, implementation, and per-
formance. J. Phys. Chem., 132, 044107, (2010).

[317] C. Hättig; D. P. Tew; B. Helmich. Local explicitly correlated second- and third-order
møller–plesset perturbation theory with pair natural orbitals. J. Chem. Phys., 146,
204105, (2012).

[318] J. C. Tully. Molecular dynamics with electronic transitions. J. Chem. Phys., 93, 1061,
(1990).
BIBLIOGRAPHY 547

[319] E. Tapavicza; I. Tavernelli; U. Rothlisberger. Trajectory surface hopping within lin-

ear response time-dependent density-functional theory. Phys. Rev. Lett., 98, 023001,
(2007).

[320] E. Tapavicza; A. M. Meyer; F. Furche. Unravelling the details of vitamin D photosyn-

thesis by non-adiabatic molecular dynamics simulations. Phys. Chem. Chem. Phys.,
13, 20986, (2011).

[321] B. G. Levine; C. Ko; J. Quenneville; T. J. Martinez. Conical intersections and double

excitations in density functional theory. Mol. Phys., 104, 1039, (2006).

[322] E. Tapavicza; I. Tavernelli; U. Rothlisberger; C. Filippi; M. E. Casida. Mixed time-

dependent density-functional theory/classical trajectory surface hopping study of oxi-
rane photochemistry. J. Chem. Phys., 129(12), 124108, (2008).
Index

(non)-append mode, 82 $cbas, 227, 228, 243, 269, 280, 286, 288, 444,
*, 65 455, 456
-central, 306 $cbse, 215, 439
-fanal, 259 $cc2_natocc, 232, 455
-frznuclei, 305, 306 $cdspectrum, 216, 435, 444
-level rirpa, 120 $cell, 190, 192–194, 420
-mfile, 148 $cell angs, 190
-nthreads, 148 $cgrad, 455
-old, 65 $closed shells, 85, 86, 175, 378, 396, 402,
-relax, 119, 144 403
-ri, 288 $collinear, 410
-ri -level rirpa, 288 $constraints, 485
-scrpath, 148 $coord, 71, 73, 126, 128–131, 305, 335, 337,
...-spndn.cao, 259 340, 355, 381, 383, 385, 467
.map, 259, 480 $coordinateupdate, 125, 461
.sys.data, 73 dqmax, 461
$2e-ints_shell_statistics,
´ 492 interpolate, 461
$2e-ints_shell_statistics, 492 statistics, 461
$D2-diagnostic, 454 $corrgrad, 466
$TURBODIR/uff/parms.in, 385 $cosmo, 415–418
$actual step, 311 allocate_nps, 415
$alpha shells, 175, 214, 379, 403 ampran, 415
$anadens, 258, 259 cavity, 415
$atoms, 83, 227, 337, 341, 378, 405, 406, 429 closed, 415
$barrier, 484 open, 415
$basis, 83, 126, 130, 381, 467 disex, 415
$beta shells, 175, 214, 379, 403 epsilon, 415
$boys, 470 nppa, 415
$bse, 215, 298, 299, 439 nspa, 415
$bse file, 439 phsran, 415
$bse iterative, 440 point_charges, 415
$bse iterlim, 440 refind, 415
$bse noqpa, 439 routf, 415
$bse noqpw, 439 rsolv, 415
$bse thrconv, 440 use_contcav, 415
$cabs, 227, 243, 455 use_old_amat, 415

548
INDEX 549

$cosmo_atoms, 415–417 rhostop, 392

$cosmo_isodens, 417 s-junc, 394
$cosmo_isorad, 316, 418 sgrenze, 393
$cosmo_out file=, 417 symblock1, 393
$coulex, 175, 177 symblock2, 393
$coupled states, 220 test-integ, 393, 408
$coupling_reduced, 434 weight derivatives, 393
$csconv, 489 $disp3, 178, 180
$csconvatom, 489 $disp4, 178, 180
$csmaxiter, 312, 490 $dkhparam, 411
$csmp2, 311, 488 $dosper, 199, 426
$current, 483 emax, 426
$currentstate, 488 emin, 426
$curswitchdisengage, 217 npt, 426, 427
$damped_response, 433 scal, 426
$dcosmo_rs, 419 width, 426
potential file definition, 419 $drvopt, 102, 303, 427, 428
$denconv, 52, 211, 227, 228, 230, 238, 243, basis on, 130
269, 277, 280, 389, 431, 443, 456 $drvtol, 102
$denconv 1d-7, 291 $dsenex, 407
$dfdxi textout, 308, 431 $ecp, 210, 381
$dft, 52, 157, 190, 195, 210, 291, 302, 364, $egrad, 126, 130, 455, 466, 468
389, 404, 408, 410, 428, 431, 441 $egradmon, 487
batchsize, 393 $electric field, 199, 200, 202, 427
functional, 389 amplitude, 427
debug, 390 gaussian, 427
dgrenze, 393 tzero, 427
diffuse, 391 width, 427
fgrenze, 393 $electrostatic field, 210, 394, 395
fullshell, 393 $embed, 335, 336, 338, 339, 413, 414
functional, 404 cell, 414
gridordering, 393 charges, 414
gridsize, 390, 404 cluster, 414
gridtype, 390 content, 414
nkk, 390 epsilon, 414
nphi, 390 lmaxmom, 414
ntheta, 390 periodic, 414
old_RbCs_xi, 391 potval, 414
p-junc, 394 wsicl, 414
qgrenze, 393 $end, 71, 345, 377
radsize, 391 $energy, 175, 218, 381, 447, 452
reference, 392 $escfiterlimit, 215, 435
rhostart, 392 $esenex, 222, 407
550 INDEX

$esp_fit, 478 default, 465

$excitation, 257, 487 individual, 465
$excitations, 243, 244, 250, 252, 256, 261, off, 129, 464
262, 278, 449, 450, 453 on, 69, 129, 132, 461, 464, 467
bothsides, 450 carthess, 69, 133, 467
conv, 450 diag, 132
exprop, 256, 450 $forceiterlimit, 429, 430
irrep, 450 $forcestatic, 467
leftopt, 450 $forceupdate, 129, 462, 467
momdrv, 263, 450 ahlrichs, 462
oldnorm, 450 indgeo, 462
preopt, 450 maxgeo, 462
spectrum, 261, 450 numgeo, 462
thrdiis, 450 allow, 464
tmexc, 262, 450 bfgs, 462
twophoton, 262, 450 damping, 464
xgrad, 450 dfp, 462
$exopt, 218–220, 442 dfp-bfgs, 462
$fermi, 101, 394 diagonal, 463
addTS, 394 ms, 462
hlcrt, 394 offdamp, 464
noerf, 394 offreset, 464
nue, 394 pulay, 463, 467
stop, 394 fail, 463
tmend, 394 maxpul, 463
tmfac, 394 minpul, 463
tmstrt, 394 modus, 463
$fields, 199, 200, 202 numpul, 463
$finnuc, 176, 411 reseig, 464
$firstorder, 395 scale, 464
$fldopt, 210, 394, 395 schlegel, 462
1st derivative, 396 thrbig, 464
2nd derivative, 396 threig, 464
edelt, 396 $freeze, 210, 227, 228, 230, 238, 239, 243,
fields, 396 269, 280, 286, 297, 442, 444, 456
geofield, 396 $gap_threshold, 488
$forceapprox, 128–132, 381, 461, 464, 466, $gdiis, 174, 410
467 $gdiishistory, 461
format, 466 $gimic, 312, 490
$forceconv, 429, 430 $global, 126, 131, 465, 467
$forceinit, 69, 464, 467 $globgrad, 126, 131, 466
diag, 464 $grad, 126, 128–131, 355, 381, 447, 452, 466,
carthess, 465 468
INDEX 551

$grid, 117, 472 444, 457

$gw, 435 conv, 444, 457
$gw ccgrid, 439 $last MP2 energy change, 465
$gw csf, 437 $last SCF energy change, 465
$gw eta, 437 $last excitation energy change, 218
$gw evgw, 436 $last step
$gw fdamp, 437 relax, 380
$gw fixz, 437 $lattice, 190–193, 421
$gw gap, 438 $lattice angs, 190
$gw gw0, 436 $lcg, 228, 243, 269, 270, 448, 455
$gw nl, 437 $les, 123, 302, 430
$gw npade, 438 all, 430
$gw npoints, 438, 439 $lesiterlimit, 430
$gw offpq, 437 $lhf, 367, 409
$gw output, 438 $localize, 350, 477, 479
$gw qpeiter, 436 mo, 478
$gw rpa, 436 sweeps, 478
$gw scgw, 437 thrcont, 478
$gw unlimitz, 437 $lock off, 380
$h0hessian, 123 $loewdin, 471
$hessian, 103, 126, 131, 133, 428, 429, 465 $log, 483
$hessian (projected), 103, 466 $log_history, 484, 486
$hotfcht, 431 $m-matrix, 132, 465
$iaoopts, 352 $magnetic field, 177
$incore, 63, 380, 396 $mao, 471
$intdef, 71, 73, 128, 129, 303, 381, 460, 462, $mao selection, 476
465, 466 $marij, 152, 406
$interconversion, 124, 461 extmax, 406
maxiter, 461 lmaxmom, 406
on, 128, 461, 462 nbinmax, 406
qconv, 461 precision, 406
$ironly, 430 thrmom, 406
$isopts, 430 wsindex, 406
$isosub, 429, 430 $maxcor, 59
$jbas, 152, 190, 286, 288, 381, 405, 420 $maxcor, 63, 100, 227, 228, 239, 240, 243,
$jkbas, 152, 227, 243, 286, 406, 455 269, 276, 280, 286, 302, 379, 380,
$ke_control, 484, 486 429, 442–444, 457
$kpoints, 186, 192–195, 421 $maximum norm of
kptlines, 421 basis set gradient, 467
nkpoints, 421 cartesian gradient, 467
recipr, 421 internal gradient, 467
$kramers, 174, 211, 215, 286, 410 $md_action, 486, 487
$laplace, 229, 238, 242, 243, 262–264, 266, $md_status, 483, 486
552 INDEX

$mgiao, 217 $optimize, 124, 125, 303, 460–462

$mo output format, 396, 400 basis, 125, 460
$mo-diagram, 396 logarithm, 460
$mointunit, 227, 230, 311, 443, 489 scale, 460
$mom, 396 cartesian, 125, 460
$moments, 349, 470, 474 global, 125, 461
$moprint, 396 internal, 125, 128, 460, 465
$mp2energy, 227, 442 redundant, 125, 460
$mp2pair, 443 $paboon, 471
$mulliken, 470 $parallel_parameters, 492
$mvd, 349, 473 $parallel_platform, 491
$nacme, 220, 442 $pardft, 492
$natoms, 483 $path, 380
$natural orbital $pcc, 176, 212, 394, 433
occupation, 381 $periodic, 190, 194, 195, 420
$periodic 1, 193
$natural orbital occupation file=natural,
480 $periodic 2, 193
$natural orbitals, 381, 396 $periodic 3, 192, 194
occupation, 396 $pnoccsd, 228, 239, 280, 281, 457, 458
$natural orbitals file=natural, 480 mp2, 458
$ncoupling, 221, 433 $pnocsd, 280
$newcoord, 301 $point_charges, 210, 313, 344, 345, 412
$nics, 312, 490 $points, 113, 117, 470
$nmr, 310 $pointval, 216, 259, 355–358, 360, 478
dft, 310 dens, 480
mp2, 310 elf, 360, 480
rhf, 310 fld, 357, 480
shielding constants, 310 fmt, 480
$nomw, 123, 429 cub, 481
$noproj, 429 map, 481
$nosalc, 308, 429 plt, 481
$nprhessian, 429 txt, 481
$nprvibrational normal modes, 429 vec, 481
$nprvibrational spectrum, 429 xyz, 481
$nsteps, 483 geo, 360, 482
$nucsel, 312, 434, 490 line, 482
$nucsel2, 434 plane, 482
$oep, 365 point, 482
$oldcphf, 489 integrate, 479
$oldgrad, 468 lmo, 358, 480
$open shells, 85, 86, 379, 403 mo, 358, 480
$operating system, 380 nao, 359
$optcell, 195 nmo, 480
INDEX 553

nto, 359 zpreopt, 452

pot, 357, 480 $response, 238, 243, 255, 258, 264, 451, 453
xc, 358, 480 $restart, 399
$pointvalper, 196, 197, 203, 425 $restartd, 397, 399
dens, 425 $ricc2, 228, 231, 236, 238, 243, 244, 247,
eps, 425 248, 250, 255, 258, 266, 269, 276,
fmt, 425 277, 444, 445, 452, 453, 455
full, 425 adc(2), 445
ngrdpbx, 425 bccd(t), 455
nimg, 425 cc2, 445
npts, 425 ccs, 445
orbs, 425 ccsd, 455
$pop, 350, 474, 475 ccsd(t), 455
atoms, 475 cis, 445
dos, 475 cis(d), 445
lall, 475 cisdinf, 445
mo, 475 conv, 445
netto, 475 core, 455
overlap, 475 d1diag, 445
thrpl, 475 fmtprop, 445
$pop nbo, 475 geoopt, 255, 445
$pop paboon, 476 gsonly, 445
$pople, 378 hard_restart, 445
$prediag, 397, 399 intcorr, 277, 445
$printlevel, 444, 446, 457 iprint, 445
$properties, 109, 111, 470 lindep, 445
$pseudospectral, 222, 407 maxiter, 445
$ramanonly, 430 maxred, 445
$rbss, 175, 176, 212, 411 mp2, 445
$rdkh, 175, 176, 212, 411 mp3, 455
$reaction_field, 346, 419 mp4, 455
$redund_inp, 381 mxdiis, 445
$redundant, 128, 303, 460, 466 no_sc, 455
$response, 452 nohard_restart, 445
conv, 452 norestart, 445
fop, 452 oconv, 445
gradient, 452 restart, 445
nosemicano, 452 rohf-bccd(t), 455
nozpreopt, 452 scs, 445
semicano, 452 shift, 445, 455
sop, 452 sos, 445
thrsemi, 452 $rick, 222
zconv, 452 $ricore, 59
554 INDEX

$ricore, 96, 151, 152, 210, 216, 302, 380, $rirpa, 286, 288, 440
405, 406, 492, 493 $rlocal, 175, 176, 212, 411
$ricore_slave, 380, 493 $rlocpara, 411
$ridft, 210, 216, 405, 406 $rohf, 403
$rigw contour, 438 $roothaan, 379, 403
$rigw ips+, 438 $rpaconv, 209, 431, 434, 435
$rij, 177, 405, 410 $rpacor, 216, 380, 434
$rik, 151, 177, 222, 286, 405, 406, 410 $rsym, 175
$riper, 187, 190, 192–194, 422, 423 $rtdens, 203
desnue, 423 $rtdipol, 199, 202
epsbext, 423 $rtenergy, 199, 202
lcfmmpcg, 423 $rtspectrum, 200, 202
lchgprj, 423 $rttddft, 199, 200, 202, 426
lenonly, 423 damping, 426
lmaxmom, 423 energy step, 426
lmxmompcg, 423 magnus, 426
locmomadd, 423 max energy, 426
locmult, 423 min energy, 426
lpcg, 423 print step, 426
nctrgt, 423 scf, 426
northol, 423 time, 426
pcgtol, 423 tstep, 426
pcgtyp, 423 $rundimensions, 397
pqmatdiag, 423 $rx2c, 175, 176, 212, 411
pqsingtol, 423 $scfconv, 52, 99, 211, 290, 397, 401, 431,
sigma, 423 443, 456
thrints, 423 settings for
wsicl, 423 Aoforce, 302
$ripop, 405 Numforce, 302
$rir12, 228, 232, 233, 236, 243, 269, 272, $scfconv 7, 195, 291
447, 455, 456, 460 $scfdamp, 187
ansatz, 447 $scfdenapproxl, 396, 398
cabs, 447 $scfdenapproxl 0, 181
cabsingles, 447 $scfdiis, 397, 399
ccsdapprox, 456 $scfdump, 396, 397, 399
comaprox, 447 $scfinstab, 151, 210, 299, 435
corrfac, 447 ciss, 432
examp, 447 cist, 432
f12metric, 447 complex, 432
no_f12metric, 447 dynpol, 432
pairenergy, 447 non-real, 432
r12model, 447 polly, 432
r12orb, 447 rpas, 432
INDEX 555

rpat, 432 $snso, 176, 411

singlet, 432 $snsopara, 176, 411
soghf, 432 $soes, 210, 215, 218, 220, 299, 432, 435, 441,
spinflip, 432 442
tdasoghf, 432 $soghf, 101, 173, 174, 176, 177, 211, 215,
triplet, 432 286, 354, 410, 412, 433
twophoton, 433 $spectrum, 216, 221, 434, 444
ucis, 432 $spin constraint {real}, 404
urpa, 432 $spin constraint {real} , 404
$scfintunit, 57, 63, 210, 397, 399, 402, 443, $spinor, 175
489 $start vector
file, 399 generation, 214, 435
size, 399 $statistics, 399, 402
unit, 399 dscf, 152, 380, 402
$scfiterinfo, 399 dscf parallel, 402, 492
$scfiterlimit, 400, 441 grad parallel, 492
$scfiterlimit 1, 291 mpgrad, 230, 380, 402, 443
$scfmo, 86, 378, 381, 396, 397, 399, 400, 404, mpshift, 311
489 off, 380, 402
expanded, 400 $statpt, 121, 123, 469
file, 400 bfgs, 123
format, 400 hssfreq, 469
none, 86, 400 hssidiag, 469
scfconv, 400 hssupdate, 469
scfdump, 400 itrvec, 122, 469
$scfmo none, 86 keeptmode, 469
$scforbitalshift, 187 powell, 123
$scforbitalorder, 401 radmax, 469
$scforbitalshift, 401 radmin, 469
automatic, 401 threchange, 122
closedshell, 401 thrmax-displ, 122
individual, 401 thrmaxgrad, 122
noautomatic, 401 thrrmsdispl, 122
$scftol, 277, 401, 443, 456, 489 thrrmsgrad, 122
$scratch tradius, 121, 469
$scratch $subenergy, 181
files, 489, 492 $subsystems, 182
$scratch files, 401, 465, 491 $sum
$seed, 483 rules, 435
$senex, 63, 407 $surface_hopping, 487
$sh_coeffs, 487 $suspend off, 380
$shiftconv, 310, 489 $symmetry, 210, 378
$sijuai_out, 308, 431 $tb, 136, 387
556 INDEX

$thime, 150, 152, 210, 402, 443, 489 aoforce, 19, 20, 34, 36, 48, 51, 54, 56, 59,
$thize, 150, 152, 210, 311, 396, 402, 443, 62, 64, 100, 103, 108, 123, 126, 131,
489 163, 205, 222, 301–308, 310, 313,
$title, 378, 484 379, 407, 428, 466, 490
$tmpdir, 240, 265, 444, 457 aoforce2g98, 36
$tplot, 232
$traloop, 227, 230, 311, 443, 488, 491 B-matrix, 74
$trand, 489 babel, 47, 50
$trast, 489 Bend, 37
$turbomole, 483 bend, 36
$uff, 133, 383 Boys localization, 478
maxcycle, 133 Broken symmetry, 90
$uffgradient, 383, 385 bsse_out, 145
$uffhessian, 383, 385 bsseenergy, 143
$ufftopology, 383, 385 Buckingham, 349
$uhf, 379, 404
cbasopt, 37
$uhfmo_alpha, 214, 381, 400, 404
CC2, 34
$uhfmo_beta, 214, 381, 400, 404
RI-, 443
$userdefined bonds, 381
cc2-gsdn-1a1-001-total.cao, 259
$vcd, 305, 430, 490
cc2-xsdn-3a2-001-total.cao, 259
$vdw_fit, 473
cc2cosmo, 37
$velocity gauge, 435
CCL0–m-ss-xxx, 255
$wfn, 478
CCLE0-s–m-xxx, 250
&, 65
CCME0-s–m-xxx, 260
ṗlt, 479
CCNE0-s–m-xxx, 262
<method>-<type>-<mult><irrep>-<number>-total.cao,
CCNL0-s–m-xxx, 256
259
CCRE0-s–m-xxx, 250
NumForce
CCS
-frznuclei, 305
RI-, 443
-frznuclei
CCSD, 443
NumForce, 305
CCSD(F12∗) , 272
1d-4, 291
CCSD(2)F12 , 272
actual, 36 CCSD(2*)F12 , 272
actual step CCSD(F12), 272
dscf, 380 CCSD(F12*), 272
ADC(2) CCSD(T), 443
RI-, 443 CCSD-F12a, 272
adg, 36 CCSD-F12b, 272
analysis of normal modes CCSD-F12c, 272
internal coordinate, 303 CCSD[F12], 272
Aoforce CCSD[T], 443
keywords, 428 ccsdf12, 20, 34, 53, 55, 62, 80, 100, 240,
268–270, 279, 313, 379, 455
INDEX 557

cgnce, 37 298, 300, 303, 348, 359, 364, 366,

charge vector, 365 377, 379–381, 400, 403, 405, 406,
CIS, 34 408, 431, 436, 461, 495
RI-, 443 -old, 65
CIS(D), 34 degrees of freedom, 74
RI-, 443 dens, 259, 355
condition, 365 desnue, 187
conjugate gradients, 124 desnue 0, 187
control, 33, 36, 47, 49–52, 65, 67, 82, 85, 92, desnue 1, 187
122, 232, 258, 265, 291, 343, 345– DIIS, 124, 461
348, 354 dist, 37
converged, 120 do_sfit, 407
coord, 50 dos_a+b, 475
coordinates dos_a-b, 475
frozen, 305 dos_alpha, 475
copo, 182 dos_beta, 475
core memory, 365 DRC, 37
cos, 266, 447, 460 Dscf
COSMO keywords, 389
keywords, 414 Dscf, 405
cosmo, 37, 242, 313 dscf, 19, 20, 33, 34, 44, 48, 51–57, 59, 60, 62–
COSMO- 64, 85, 86, 96, 109, 120, 126, 137,
MP2, 239 144, 150–153, 155, 211, 229, 230,
PTE, 239 243, 244, 258, 265, 270, 280, 295,
PTED, 239 297, 298, 300, 309, 310, 313, 318,
cosmoprep, 37, 417 320, 334, 342, 348–350, 355, 362,
counterpoise calculation, 82 366, 380, 381, 397, 401, 402, 408,
CP-corrections, 143 410, 412, 414, 418–420, 427, 472,
CPHF, 309 473, 489, 491–493
cpt, 37, 221 dummy center, 71
cpt –help, 37
css, 266, 447, 460 edens, 259
custom, 198 Egrad
custom nks, 194 keywords, 441
cycle, 346 egrad, 19, 20, 34–36, 53, 54, 56, 62, 64, 120,
126, 157, 163, 174, 204, 205, 209–
debug, 366 211, 216, 218, 219, 222, 258, 304,
Define, 291 348–350, 355, 379, 407, 441, 466,
define, 33, 41, 48–54, 65–67, 69–74, 78, 82– 472, 473, 487
86, 88–90, 92–94, 96, 104, 106, 109– eigenvalue difference, 366
111, 114, 117, 118, 121, 122, 128, Eiger, 358, 477
129, 131, 143, 144, 152, 163, 173– eiger, 37
176, 190, 210, 216, 228, 231, 233, elf, 360
235, 243, 260, 269, 277, 280, 295, energy, 133
558 INDEX

environment grad, 20, 33, 34, 52, 54, 55, 59, 60, 62, 64, 96,
variable 103, 120, 126, 128, 130, 137, 144,
OMP_NUM_THREADS, 62 150–153, 155, 157, 174, 218, 255,
PARA_ARCH, 62 265, 313, 320, 334, 394, 411, 412,
PARNODES, 62 414, 419, 427, 441, 466, 468, 492,
escf 493
keywords, 431 grad_out, 145
escf, 19, 20, 34, 53, 54, 56, 59, 62, 64, 98, gradient, 258
101, 157, 163, 172, 204, 209–219, gradients
221, 222, 294, 295, 297–300, 310, excited states, 255
313, 318, 334, 356, 379, 394, 407, ground state, 253
412, 414, 419, 431, 438, 441 grid, 355, 359
evalgrad, 37 Grimme Dispersion, 178
evib
keywords, 431 hcore, 38
evib, 36, 62, 308, 379, 431
ibos, 352
export, 355
Infrared Spectra, 301
extended Hückel calculation, 85
intcorr, 277
FDE, 38 intense, 36, 304
file2control, 37 internal coordinates
finit, 38 linear combination of, 77
fld, 357 manual definition of, 76
fmt=, 198 types of, 76
format, 357 intersections
Freeh, 54, 302 conical, 249
freeh, 36
jmol, 38
Frog
job.<cycle>, 120
keywords, 482
job.last, 120
frog, 34, 54, 120, 137, 482, 483, 485, 486,
job.start, 120
488
jobbsse, 38, 72, 143–145
frozen coordinates, 305
Jobex, 152
Fukui, 38
jobex, 34, 35, 38, 44, 51, 67, 96, 100, 119,
full, 220
120, 122, 123, 133, 143, 146, 195,
gallier, 38 205, 218, 243, 244, 284, 288, 447,
geofield, 394 467, 493
geometries -c, 119, 144
excited states, 255 -dscf, 119
ground state, 253 -energy, 119, 144
geometry -ex, 119
manipulation of, 78 -gcart, 119, 144
Grad -grad, 119
keywords, 427 -gradient, 144
INDEX 559

-keep, 119 mdprep, 38

-l, 119, 144 MECPopt, 38
-level, 119, 144 MECPprep, 38
-ls, 119, 144 memory, 59
-md, 119 menu
-mdfile, 119 atomic attributes, 79, 82
-mdmaster, 119 general, 92, 93
-mem, 144 geometry main, 68
-opt, 144 geometry menu, 70
-relax, 119, 144 internal coordinate, 73, 74
-ri, 119, 144 occupation number assignment, 87
-rijk, 119 start vectors, 84
-setup, 144 mo 10-12,15, 358
-statpt, 119 molden, 50, 355
-trans, 119 molecular dynamics, 137, 482
-trimer, 144 molecular orbitals
binary format, 85
kdg, 38 Moloch
kinetic energy, 486 keywords, 469
moloch, 109, 111, 114, 115, 117, 469
lalp, 477
mos, 353
lbet, 477
MP2
Leapfrog Verlet algorithm, 137, 482
COSMO, 239
lenonly on, 194
COSMO-
Lhfprep, 367, 408
PTE-, 239
lhfprep, 38, 366, 408
PTED-, 239
lhfprep -asy, 368
RI-, 443
lhfprep -kli, 367
Mp2prep, 230
lhfprep -num, 367
mp2prep, 39, 51, 230
lmo, 477
MP3, 443
lmos, 350, 352
MP4, 443
log2?, 483
mpgrad
log2egy, 38
keywords, 442
log2int, 38
mpgrad, 19, 33, 34, 53–55, 99, 100, 120, 126,
log2rog, 38
130, 144, 174, 223, 224, 226, 227,
log2x, 38
230, 231, 239, 258, 348–350, 355,
LT-SOS-RI-MP2, 63, 237
379, 380, 389, 402, 414, 418, 442,
mdens, 259 443, 466, 473, 491, 493
mdlog, 137 Mpshift
mdmaster, 482 keywords, 488
mdmaster, 137 mpshift, 19, 20, 35, 36, 51, 53, 54, 56, 62,
Mdprep, 137, 482, 483 101, 157, 163, 217, 305, 309–312,
mdprep, 482 379, 407, 430, 490
560 INDEX

mulliken, 350 plot coefficient, 366

multi-core, 62 plotting data
keywords, 473
NAO, 359 PNO-MP2
nao, 359 keywords, 456
natural pnoccsd
orbitals keywords, 456
atomic, 359 pnoccsd, 19, 20, 34, 52, 55, 59, 62, 80, 99,
transition, 359 100, 221, 224, 226–229, 231, 239,
nbo, 350 240, 279–283, 313, 379, 449, 456,
no weight derivatives, 431 457
nohxx, 286, 291 population analysis, 475
not.converged, 120, 145 pot, 357
NTO, 359 proper, 36, 348–350, 353, 355–359
nto, 359 properties
nto, 353 excited states, 255
ntos, 353 ground state, 253
NumForce, 34–36, 39, 54, 56, 205, 219, 243, pseudo, 220
244, 257, 284, 288, 301, 302, 305, PTE-
313, 418, 428 COSMO-
NumForce -d 0.02 -central -ri -level rirpa, MP2, 239
288 PTED-
COSMO-
odft, 62, 362, 366, 367, 369
MP2, 239
OMP_NUM_THREADS, 62
OpenMP, 62 q, 65
opro, 182 quasi–Newton, 124
orient, 481
outp, 39 Raman, 36, 54, 304
raman, 39
paboon, 350 Raman spectra, 304
panama, 39, 216, 217 Rdgrad
PARA_ARCH, 62 keywords, 427
parallelization Rdgrad, 405
multi-core, 62 rdgrad, 20, 33, 34, 52, 54–56, 60, 62–64, 96,
OpenMP, 62 120, 126, 128, 144, 150–152, 155,
SMP, 62 157, 172, 174, 175, 180, 218, 255,
threads, 62 313, 320, 334, 394, 407, 411, 412,
parms.in, 385 414, 419, 427, 492, 493
PARNODES, 62 redox, 39
past, 39 reference potential, 366
PEECM Relax
keywords, 412 keywords, 460
Plane-averaged, 482
INDEX 561

relax, 19, 34, 37, 38, 54, 69, 74, 104, 105, rotate, 481
119, 120, 124–126, 128, 129, 131– rpagrad, 288, 291
133, 143–146, 460, 461, 464, 465, rpaprof, 292
467, 468
response, 220 scanprep, 39
restart.cc, 446 screwer, 39, 124
RI-ADC(2), 34, 443 scs, 236, 266, 447, 460
RI-CC2, 34, 443 SCS-ADC(2), 34
keywords, 443 SCS-CC2, 34
RI-CCS, 443 SCS-MP2, 265
RI-CIS, 34, 443 sdg, 39
RI-CIS(D), 34, 443 sh_coeff, 488
RI-MP2, 443 sigma, 187
RI-MP2-F12, 63 sigma 0.01, 187
ricc2 Simulated Annealing, 487
keywords, 443 SMP, 62
ricc2, 18, 19, 34, 37, 52–60, 62, 63, 80, 99, soghf, 175
100, 172, 174, 221, 223, 225, 227– sos, 266
233, 237, 239–252, 255, 258–261, 265, SOS-ADC(2), 34
266, 313, 342, 347, 348, 353–356, SOS-CC2, 34
379, 414, 443, 444, 449, 453, 455, SOS-MP2, 265
466 SOS-RI-MP2, 63
ricctools, 348 fourth-order scaling, 237
Ridft spectra
keywords, 389 Raman, 304
Ridft, 405 VCD, 304
ridft, 19, 20, 33, 34, 48, 52–56, 60, 62– spin
64, 86, 96, 97, 109, 120, 144, 150– flipping spins on atoms, 86
152, 155, 157, 172, 175, 177, 180, Stati, 152
211, 228, 229, 243, 244, 258, 270, stati, 39
280, 286, 291, 295, 297, 298, 300, Statpt
310, 313, 318, 320, 334, 342, 348– keywords, 468
350, 355, 356, 358, 359, 362, 381, statpt, 34, 38, 48, 54, 97, 119, 121–123, 143–
394, 405–407, 410, 412, 414, 418– 146
420, 427, 472–474, 492, 493 steepest descent, 124
rimp2, 120, 126, 144, 241, 348–350, 355, 414, STOP, 120
418, 445, 466, 472, 473 stop, 120
Riper structure library, 71
keywords, 420 structure optimization, 119
riper, 19, 33, 55, 62, 63, 183–188, 190–192, substitution, 71
194–196, 199, 420–422 syndi, 70
rirpa, 35, 62, 174, 284–286, 290–292, 379 Sysname, 43, 517, 519
Roothaan parameters, 89 sysname, 39, 43
562 INDEX

t2aomix, 39 VCD, 305

t2x, 39, 355 vcd, 41
tb VCD spectra, 304
keywords, 387 vector
tb, 56, 136 function, 247
Tblist, 519 velocity, 486
tblist, 40 vibration, 39, 301
Tbtim, 519 Vibrational Frequencies, 301
tbtim, 40
temperature, 486 wave function analysis
time, 137, 482 keywords, 473
timestep, 482 weight derivatives, 157
tm2molden, 39, 40, 355 wiberg, 350
Tors, 37 woelfling, 36, 41, 146, 147, 149
tors, 40 woelfling-job, 36, 41
transformation
x2t, 41, 47, 50, 67
Laplace, 237
xxx.map, 258
TTEST, 517–520
TURBOMOLE, 15, 17, 18, 20–22, 28, 36–39, 41–
44, 47, 49–59, 62, 64, 65, 67, 70, 81,
83–85, 93, 94, 110, 119, 122, 124,
129, 136, 143, 158–161, 163, 165,
174, 223, 231, 239, 246, 291, 294,
295, 301, 304, 305, 308, 313, 315,
334, 337, 341, 349, 354, 355, 377,
379, 397, 400, 405, 412, 416, 417,
472, 483, 491, 495
Turbomole
installation, 42
modules, 33
quotation of, 17
tools, 36
Turbotest, 44
twoint, 57

Uff
keywords, 383
uff, 33, 69, 122, 123, 133, 134, 383, 385
uffgradient, 133
uffhessian0-0, 133
ufftopology, 134, 385, 387
nxtn12, 385
uhfuse, 40