Dpartement de Chimie e Universit de Fribourg (Suisse) e

Modeling the properties of open d-shell molecules with a multi-determinantal DFT.

THESE

Prsente ` la Facult des Sciences de lUniversit de Fribourg (Suisse) e e a e e pour lobtention du grade de Doctor rerum naturalium Cdrick RAUZY e de TOULOUSE (FRANCE)

Th`se Nr. 1465 e Atelier mcanographique, Universit de Fribourg e e 2005

Accepte par la Facult des Sciences de lUniversit de Fribourg (Suisse) sur la proposition e e e des membres du jury suivant: Prof. Thomas Bally, Prsident du jury, Universit de Fribourg e e Prof. Prof. Prof. Prof. Claude Auguste Daul, Directeur de th`se, Universit de Fribourg e e Jacques Weber, Universit de Gen`ve (Suisse) e e Annick Goursot, Universit de Montpellier (France) e Mihail Atanasov, Universit de Soa (Bulgarie) e

Fribourg, le 23 mars 2005 Le directeur de th`se: e Prof. Claude Auguste Daul

Le Doyen: Prof. Marco Celio

Ce travail a t soutenu par le Fond National Suisse ee pour la Recherche Scientique.

Acknowledgements
This work was accomplished under the direction of Prof. Dr. Claude A. Daul at the Departement of Chemistry of the University of Fribourg, from April 2001 until December 2004. I would like to express my sincere gratitude to him, rstly for accepting me in his research group and also for his support and invaluable guidance throughout this work. I wish to thank Prof. Dr. Mihail Atanasov, Prof. Dr. Annick Goursot and Prof. Dr. Jacques Weber for accepting to evaluate my thesis and to be members of the jury. I would like also to thank the president of the jury: Prof. Dr. Thomas Bally. I would like to thank again Prof. Dr. Mihail Atanasov for the long and nice scientic discussions we had. I would like to thank Prof. Dr. Carl-Wilhelm Schlpfer and Prof. Dr. Svetozar Niketic a for the interesting dicussions and the good advices they gave during the weekly seminars. I thank Prof. Dr. Franzpeter Emmenegger, Prof. Dr. Peter Belser and Prof. Dr. Thomas Bally for the nice challenges they proposed me in web design. I would like to thank all my colleagues: Fabio, Valry, Pio, Mohamed, Sahnoun, Tamil, e Emmanuel. To all the institute sta, a big thank you! And in particular to: Emerith, Hubert, Michel and Philippe. Jaimerais aussi ddier ce travail ` toute ma famille, mes parents : Dani`le et Grard, e a e e mais aussi Jacques, Patricia, Rgis, Aurlia, Franck et Georay. e e And of course, last but not least, I want to thank Cris, the best colleague I had in and outside the institute.

Abstract
Ligand eld theory has been used along decades with success to describe ground and excited electronic states originating from dn transition metals complexes. Experimental chemists use such a theory to interpret spectra. On the opposite side, computational chemists can describe with good accuracy the ground states properties but models to calculate excited states properties are still being developed. The Ligand Field -Density Functional Theory, which is the method presented in this thesis, proposes a link between the density functional theory applied to ground state and the determination of excited states properties through the ligand eld theory. To achieve this, we compute within the DFT formalism the energies of all the Slater determinants originating of a dn conguration taken as reference an average of conguration to satisfy the requirement of the Ligand Field Theory. In a rst step, the method is applied to well known compounds to test the ligand eld and Racahs parameterization in comparison to values tted from experimental UV-Vis spectra. Then we use a Ligand eld program to predict the multiplet structure. Next, extension of the method is proposed to determine ESR parameters and relativistic eect within the same formalism. At each step, the results are compared to data which are well known for many decades by the chemists. We will also show the ability of the method to give more informations than usually expected.

Rsum e e
La thorie du champ des ligands a t utilise avec succ`s durant des dcennies pour dcrire e ee e e e e ltat fondamental et les tats excits des complexes. Les chimistes utilisent cette thorie e e e e an dinterprter des spectres UV-Vis essentiellement. Dun autre ct, les chimistes e oe computationels peuvent dcrire assez prcisment les proprits correspondant ` ltat e e e ee a e fondamental mais les mod`les permettant de dcrire les proprits des tats excits sont e e ee e e encore en voix de dveloppement. La mthode Champ des Ligands - Thorie de la Fonce e e tionnelle de la Densit, qui est la mthode prsente dans cette th`se, propose un lien e e e e e entre la Thorie de la Fonctionnelle de la Densit (TFD) applique ` ltat fondamental e e e a e et la modlisation des proprits des tats excits par lintermdiaire de la thorie du e ee e e e e champ des ligands. Pour ce faire, nous calculons grce ` la TFD les nergies de tous a a e les dterminants de Slater due ` une conguration dn en rfrence a un tat correspone a ee e dant ` une conguration moyenne (rpartition gale des lectrons d dans les 5 orbitales a e e e molculaires correspondant aux orbitales d de llment de transition) an de satisfaire e ee aux exigences de la thorie du champ des ligands. Dans un premier temps, la mthode e e est applique ` des composs connus an de tester sa validit. Dans un deuxi`me temps, e a e e e le champ dapplication de la mthode est tendue ` la modlisation des tenseurs g et A. e e a e Tout au long de cette th`se, les rsultats obtenus sont compars aux donnes exprimentales e e e e e obtenues par les chimistes. Nous montrons aussi que la mthode donne plus dinformations e que lon ne pouvait esprer, en particuliers, lors du traitement des eets relativistes. e

Contents
Contents List of Figures List of Tables Useful notations 1 General Introduction 2 Theoretical Concepts 2.1 The Crystal Field Theory . . . . . . . . . . . 2.1.1 A Single d-electron in a Cubic Field . . 2.1.2 Two electrons in a Cubic Field . . . . 2.1.3 Many Electrons in a Cubic Field . . . 2.1.4 Reduced Electrostatic Matrix Elements 2.1.5 Angular Overlap Model (AOM) . . . . 2.1.6 Spin-Orbit Coupling . . . . . . . . . . 2.1.7 Qualitative Considerations of LFT . . 2.2 DFT . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Basic Theory . . . . . . . . . . . . . . 2.2.2 The Thomas-Fermi Model . . . . . . . 2.2.3 The Hohenberg and Kohn Theorems . 2.2.4 The Constrained Search Approach . . 2.2.5 The Kohn-Sham Approach . . . . . . . 2.2.6 Hole Functions . . . . . . . . . . . . . 2.2.7 Exchange and Correlation Functional . 2.2.8 Scaling Relations in DFT . . . . . . . 2.2.9 Janaks Theorem . . . . . . . . . . . . 2.2.10 From Ground to Excited States . . . . 3 LFDFT 3.1 DFT Calculations . . . . . . . . . . . . 3.1.1 General Settings . . . . . . . . 3.1.2 Geometry Determination . . . . 3.1.3 AOC . . . . . . . . . . . . . . . 3.1.4 Slater Determinant Calculation ix ix xiii xv xvii 1 5 5 6 9 11 12 15 17 17 18 18 21 22 23 24 26 27 31 33 34 35 35 37 37 39 40

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (Wigner-Eckart) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

3.2

3.3

3.4 3.5

3.6

Extraction of Data . . . . . . . . . . . . . . . . . . . . 3.2.1 Eigenvectors and Eigenvalues of the Kohn-Sham 3.2.2 Energies of Slater Determinants . . . . . . . . . LFDFT: Theory . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . 3.3.2 Theory . . . . . . . . . . . . . . . . . . . . . . . LFDFT: Program . . . . . . . . . . . . . . . . . . . . . LFDFT: Applications . . . . . . . . . . . . . . . . . . . 3.5.1 Cubic d2 Systems . . . . . . . . . . . . . . . . . 3.5.2 Cubic dn Systems . . . . . . . . . . . . . . . . . 3.5.3 Application to Model the Nephelauxetic Eect . 3.5.4 The Eect of Covalency on 10Dq . . . . . . . . 3.5.5 Applications to a Low Symmetrical Compound General Conclusions . . . . . . . . . . . . . . . . . . .

. . . . . Orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

40 41 43 43 43 44 45 47 47 48 50 51 52 52 53 53 53 54 55 55 55 57 59 59 61 62 63 66 67 67 67 70 70 71 73 73 74 75 78 79

4 Multiplet Structure 4.1 Theory of Multiplet . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Multiplet Energy as a Function of ESD . . . . . . . . . . . . . 4.1.2 Multiplet Energy as Solution of H . . . . . . . . . . . . . . . 4.2 LF Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Matlab Program . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 AOMX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Program Gener ln . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Octahedral CrIII d3 and CoII d7 Complexes . . . . . . . . . . 4.3.2 Octahedral Cyano MnIII (d4 ), CoIII and FeII (d6 ) Complexes 4.3.3 Applications to Tetrahedral d5 MnCl2 and FeCl1 Complexes 4 4 4.3.4 Tetrahedral d2 Complexes . . . . . . . . . . . . . . . . . . . . 4.4 General Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

5 Relativistic Eect within LFDFT 5.1 Spin-Orbit Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Theoretical Description . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Computational Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Geometry Optimizations . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Calculation of Reduced Matrix Elements of Spin-Orbit Coupling from ZORA-DFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 The Jahn-Teller Eect and the Geometries of NiX2 (X = F , Cl , 4 Br , I ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Spin-Orbit Coupling in Td Symmetry . . . . . . . . . . . . . . . . . 5.3.3 Spin-orbit Coupling in D2d Symmetry . . . . . . . . . . . . . . . . . 5.3.4 Ligand Field-Parameters, Ground and Excited States Energies of NiX2 (X=Cl , Br , I ) and Comparison with Experimental Data 4 5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

6 ESR Parameters within LFDFT 6.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.1 LFDFT: a New Way to Calculate ESR Parameters 6.1.2 Spin-Orbit ZORA Approach . . . . . . . . . . . . . 6.2 Computational Details . . . . . . . . . . . . . . . . . . . . 6.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . 6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

83 84 84 86 86 88 91

7 Conclusions and Outlook 93 7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 7.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Bibliography A Mathematical background A.1 Expansion of 1/rij . . . . . . . . . . . A.2 Spherical Harmonic Properties . . . . . A.3 Spherical Harmonic Addition Theorem A.4 Wigner-Eckart Theorem . . . . . . . . A.5 Clebsch-Gordan Coecient . . . . . . . A.6 Laplace Expansion . . . . . . . . . . . 97 105 . 105 . 107 . 107 . 109 . 109 . 110 111 . 111 . 112 . 112 . 114 . 114 . 114 115 115 116 116 121 125 125 129

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

B LFDFT Scripts B.1 Creation of the SD calculation input . . B.2 Extraction of data . . . . . . . . . . . . B.2.1 SD energies . . . . . . . . . . . . B.2.2 Eigenvalues of KS orbitals . . . . B.2.3 Eigenvectors of KS orbitals . . . . B.2.4 Evaluation of Standard Deviation

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

C Gener ln Functions C.1 GENERSD Function . . . . . . . . . . . . . . C.2 Function to Determine Matrix Elements in the C.2.1 GET2EI4A . . . . . . . . . . . . . . . C.2.2 GETLS . . . . . . . . . . . . . . . . . C.3 Functions corresponding to the Slaters rules . C.3.1 One-electron Slaters rule . . . . . . . . C.3.2 Two-electrons Slaters rule . . . . . . .

. . . . . . . . . . . . . Basis of Spin-Orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

D SOC: supplement 133 D.1 Spin-orbit coupling elements in symmetry D2d . . . . . . . . . . . . . . . . 133 D.2 DFT treatment of JT activity in the case of mixing of electronic states . . 134 E ESR: Programs 137 E.1 Gener ln pogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 E.2 LF program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Curriculum Vitae xi 141

List of publications

143

xii

List of Figures
2.1 The Metal surrounded by the 4 ligands in the plane. . . . . . . . . . . . . 2.2 Scheme of the crystal eld splitting in an octahedral environment . . . . 2.3 Representation of a - and a -bond and illustration of the resulting MO splitting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Ligand positions for the demonstrative case. . . . . . . . . . . . . . . . . 2.5 Scheme of the AOM splitting in an octahedral and in a tetrahedral environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 A shortened spectro-chemical series. . . . . . . . . . . . . . . . . . . . . . . . 6 9

. 15 . 16 . 16 . 18 36 38 38 38 39 40 41 42 46 47 51 51

3.1 Scheme of the LFDFT calculations procedure . . . . . . . . . . . . . . . . 3.2 Example of the geometry optimization input for the [CoCl4 ]2 complex. . . 3.3 Example of the AOC calculation input for the [CoCl4 ]2 complex. . . . . . 3.4 Part of the KS molecular orbital diagram of a geometry optimization output 3.5 Part of the KS orbital diagram of an AOC calculation output . . . . . . . 3.6 Example of a SD calculation input for [CoCl4 ]2 . . . . . . . . . . . . . . . 3.7 Part of the B U I L D section of the AOC calculation output for the MO E:1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Part of the SFO MO coecients section of the AOC calculations output. 3.9 Source code of the LFDFT program. . . . . . . . . . . . . . . . . . . . . . 3.10 Comparison between SD energies from DFT and from LFDFT calculation. 3.11 Evolution of the nephelauxetic reduction factor for a series of compounds. . 3.12 10Dq values versus the metal-ligand distance in the ionic crystal eld approximation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.1 Source code of the LF matlab program. . . . . . . . . . . . . . . . . . . . . 56 4.2 Source code of the Gener ln program. . . . . . . . . . . . . . . . . . . . . . 57 4.3 Structure of the Gener ln program. . . . . . . . . . . . . . . . . . . . . . . 58 5.1 5.2 5.3 5.4 The angle describing the tetrahedral distortion due to Jahn-Teller activity in NiX2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Relative energies of Kohn-Sham orbitals with dominant 3 d character from a ZORA spin-orbit calculation of [NiX4 ]2 (X=F , Cl , Br , I ). . . . . Orbital level splittings and notations for a symmetry based on description of the D2d distorted (elongated) NiCl2 . . . . . . . . . . . . . . . . . . . 4 The dependence of the KS MO energies from non-relativistic calculation with AOC occupancies on the geometrical angle for NiCl2 . . . . . . 4 xiii . 73 . 74 . 75 . 76

5.5 5.6 6.1 6.2

Reduced matrix elements of the SOC operator from ZORA-ADF calculations in the Td and D2d geometries of NiCl2 . . . . . . . . . . . . . . . . . 77 4 The splitting of the 3 T1 ground state in the Td geometry due to the spinorbit coupling and in the lower D2d symmetry. . . . . . . . . . . . . . . . . 80 The axial coordinates of the system in the discussion (x, y, z) and in the ADF calculations (x , y , z ) are represented along with the [Co(acacen)]. . 87 KS MO diagram for the ground state conguration. . . . . . . . . . . . . . 89

A.1 Position of the electrons i and j. . . . . . . . . . . . . . . . . . . . . . . . . 105 B.2 Source code of the script get.x. . . . . . . . . . . . . . . . . . . . . . . . 112 B.1 Snapshot of the program chem.x. . . . . . . . . . . . . . . . . . . . . . . . 113 B.3 Source code of the script eigen.x. . . . . . . . . . . . . . . . . . . . . . . 114 C.1 Source C.2 Source C.3 Source C.4 Source C.5 Source C.6 Source C.7 Source C.8 Source C.9 Source C.10 Source C.11 Source C.12 Source C.13 Source code code code code code code code code code code code code code of of of of of of of of of of of of of the the the the the the the the the the the the the GENERSD function. . YLM1 function. . . . . CET2EI4A function. . CGR function. . . . . GETJ function. . . . . GETLS function. . . . LMAT function. . . . ZAB function. . . . . . IWAB function. . . . . LFAB function. . . . . GAB function. . . . . GIJKL function. . . . IROW function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 117 119 120 121 122 124 126 127 128 130 131 131

E.1 Source code of the Gener ln program modied to enable us to calculate the g- and the A-tensors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 E.2 Source code of the hfab function. . . . . . . . . . . . . . . . . . . . . . . . 139 E.3 Source code of the LF program modied to enable us to calculate the g and the A-tensors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

xiv

List of Tables
2.1 The non-redundant interelectronic repulsion integrals express in function of the Racahs parameters A, B and C. . . . . . . . . . . . . . . . . . . . . 13 2.2 Overlap contributions due to each ligand with the ve d-orbitals of the TM. 16 3.1 3.2 3.3 3.4 3.5 3.6 3.7 Acronyms of GGA functional used. . . . . . . . . . . . . . . . . . . . . . . Number of SD originating from an open dn shell. . . . . . . . . . . . . . . Racahs parameters obtained for CrO4 with an optimized geometry and 4 the experimental geometry and compare to ones tted from UV-Vis spectra. Energy expressions for the DFT distinguishable Slater determinants of a d2 -conguration in a cubic ligand eld. . . . . . . . . . . . . . . . . . . . . Racahs parameters determined by LFDFT for various complexes and compare to ones tted from UV-Vis spectra. . . . . . . . . . . . . . . . . . . . B and C parameters determined by LFDFT for the free ions Cr4+ , Mn5+ and Fe6+ and compare to ones tted from atomic spectra. . . . . . . . . . Racahs parameters obtained for a series of oxide and a series of halogenes complexes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 40 48 49 49 50 50

4.1 Electronic transition energies of CoX4 , X = {Cl , Br } . . . . . . . . . 6 4.2 Electronic transition energies of CrX3 , X = F ,Cl ,Br and CN . . . . 6 4.3 Electronic transition energies of of the low-spin Mn(CN)3 octahedral d4 6 complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Electronic transition energies of low-spin Fe(CN)4 and Co(CN)3 octahe6 6 dral d6 complexes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Electronic transition energies of high-spin tetrahedral MnCl2 and FeCl1 4 4 d5 complexes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Electronic transition energies in CrO4 , MnO3 and FeO2 oxo-anions . 4 4 4 4.7 Electronic transition energies of CrX4 , X = F , Cl , Br and I . . . . . 5.1

. 59 . 60 . 61 . 62 . 63 . 64 . 65 71 72

The selecting scheme for ZORA spin-orbit coupling eigenvalues, 3d-eigenfunctions and their occupations for tetrahedral NiCl2 as a model example. 4 5.2 Ligand eld matrix elements and reduced spin-orbit coupling matrix elements for Td and D2d geometries of NiX2 (X = F , Cl , Br , I ) complexes. 4 5.3 Bond lengths, the extent of Jahn-Teller elongation of the tetrahedral into the D2d ground state energy minima and the Jahn-Teller stabilization energy EJT for NiX2 (X=F ,Cl ,Br and I ). . . . . . . . . . . . . . . . . . 4 5.4 Ligand eld matrix elements and Racahs parameters B and C for Td and D2d NiX2 (X = F , Cl , Br , I ) complexes . . . . . . . . . . . . . . . . 4 xv

73 76

5.5 5.6 5.7 6.1

The Racahs parameters determined with LFDFT method for [NiCl4 ]2 using the exchange and correlation functionals available in ADF2003.01 . . 78 Orbital interpretations of ligand eld energies from LFDFT calculations in terms of the AOM parameterization scheme . . . . . . . . . . . . . . . . . 79 Multiplet energies for NiX2 (X=Cl ,Br ,I ) . . . . . . . . . . . . . . . . . 81 4 the . . . . . . . . . . . . 87 88 90 90

All non-empirically determined parameters used in the calculation of g- and A-tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Multiplet splitting energies determine by LFDFT for Co(acacen) . . . 6.3 g-tensor values of Co(acacen) . . . . . . . . . . . . . . . . . . . . . . 6.4 A-tensor values of Co(acacen) . . . . . . . . . . . . . . . . . . . . . . D.1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

xvi

Useful notations
Orbitals labels: : dx2 y2 , : dz2 , : dyz , : dxz , : dxy 1 eV = 8066 cm1 ADF: Amsterdam Density Functional AO: Atomic Orbital LFT: Ligand Field Theory AOC: Average Of Conguration MCD: Magnetic Circular Dichroism AOM: Angular Overlap Model MO: Molecular Orbital CFs: Core Functions SCF: Self Consistent Field CFT: Crystal Field Theory SD: Slater Determinant CI: Conguration Interaction SFO: Symmetrized Fragment Orbitals DF: Density Functional SOC: Spin-Orbit Coupling DFT: Density Functional Theory EPR: Electron Paramagnetic Resonance TD-DFT: Time Dependent - Density Functional Theory

LDA: Local Density Approximation LF: Ligand Field LFDFT: Ligand Field - Density Functional Theory

GGA: Generalized Gradient Approxima- TF: Thomas-Fermi tion TFD: Thomas-Fermi-Dirac HF: Hartree-Fock TM: Transition Metal HK: Hohenberg-Kohn TZP: Triple- STO basis sets plus one poirrep: irreducible representation larization function JT: Jahn-Teller KS: Kohn-Sham xvii ZORA: Zero-Order Regular Approximation

Chapter 1 General Introduction

This thesis is about: Modeling the properties of open d-shell molecules with multi-determinantal DFT where Modeling means to propose a model that can explain experimental data with the aim to describe and predict properties of molecular systems. In this thesis I restrict the investigation to open d-shell molecules, i.e. essentially to complexes with an open d-shell transition metal. The model proposed is based on multi-determinantal DFT calculation which means that we calculate the energies of Slater determinants (SD) within the Density Functional Theory (DFT) formalism. Finally, these energies are used in a Ligand Field (LF) approach to parameterize the LF matrix and the Racahs parameters. So the main objective of this thesis is to describe the model called LFDFT for Ligand Field - Density Functional Theory. Why LFT? The Ligand Field Theory (LFT) has been and still is a useful theory for interpreting optical and magnetic properties of TM complexes. The predominantly ionic picture, based on metal ions perturbed by surrounding ligands provides an adequate description of the optical d d transition, paramagnetism, magnetic interactions and electronic spin resonance (ESR). LFT is an empirical approach gaining its parameters by comparison with existing experimental data and its justication lies in its ability to reproduce these data. Hence LF approach is a method for interpreting rather than a method for predicting electronic phenomena in TM complexes. The demand for a predictive theory in this respect is mostly tangible for systems, like active sites of enzymes for which no structural and spectroscopic data are known. The need for a model which is parameter free on one side but compatible with the usual ligand eld formalism on the other side becomes increasingly pronounced in view of the new developments in bio-inorganic chemistry.1, 2 Why DFT? On the other hand, in the computational chemistry eld, Density Functional Theory is very popular. This popularity comes from the good balance between computational eort and accuracy of the results for both ground state3 as well as excited states47 properties. 1

CHAPTER 1. GENERAL INTRODUCTION Moreover, it has been shown in previous studies811 that the multiplet theory of TM ions and complexes is compatible with the DFT formalism. In turn, recipes have been proposed (see1214 ) to link DFT, in the form of SD energies, with the multiplets energies of Transition Metal (TM) ions or complexes. Why combining DFT and LFT? The work presented in this thesis can be considered as an extension of these previous works. Originally, the calculation of multiplet energies was based on a multi Slater determinants method approach: the wave function of the multiplet being expressed as a linear combination of Slater determinants. If we restrict our investigation to multiplet arising from an open d-shell system and if we use the symmetry,11 it can be then proven that only a limited numbers of them (named non-redundant) are necessary to express the wave function of all multiplets considered. But we know that within, the Kohn-Sham (KS) formalism, the exchange-correlation term of the Hamiltonian is calculated in an approximated fashion so the spin-density exhibits a lower symmetry than the electronic Hamiltonian for an open shell system: the DFT symmetry dilemma.15, 16 One of the consequences is that SDs redundancy conditions are not veried within the DFT approach. A simple way to cure this problem (in an approximated fashion) is to consider all the SDs originating of an open d-shell and then reduce the error due to approximate exchangecorrelation in a least square sense (more elaborated methods can be proposed: to use a better functional or to adopt the symmetrized Kohn-Sham approach17 and its later developments18 ). Then comes the main idea of this work that is to introduce the LFT instead of expressing multiplets as linear combination of SDs as it was done previously.11, 12, 19 In doing so we introduce the following approximations: the electrons of the valence shell of the TM ions dont interact with the electrons of the ligands, we just consider an electrostatic potential exerted by the ligands on them, the bi-electronic part of the Hamiltonian is not treated in agreement with the symmetry of the system but reduced to 3 (for example: via symmetry rule we can reduce the number of bi-electronic integrals to 10 for 0h symmetry, but within the LFT, their numbers is reduced to 2 like in the group of the sphere independently of the symmetry considered), We can summarize the situation in saying that the mono-electronic part of the Hamiltonian in LFT respects the symmetry of the system whereas the bi-electronic part is treated like in the spherical group. One of the consequences of the introduction of LFT is that we can not treat systems which are too covalent or where Metal to Ligand Charge Transfer (MLCT) or Ligand to Metal Charge Transfer (LMCT) contribute with an appreciable weight to the ground state wavefunction. Advantages and Inconveniences of the LFDFT method? In bringing together the DFT and the LFT, we are taking advantages and inconveniences of both methods but nevertheless adding few advantages. The main advantage 2

lies in the fact that we are making a bridge between two communities: the experimentalists and the theoreticians because we convert DFT data in variables which are familiar to experimentalists. Moreover the values we get are directly comparable to experimental ones. The second main advantage is that we can get from DFT multiplets energies but we get more freedom from LFT by the fact we can modify the Hamiltonian to take into account other smaller perturbations like Spin-Orbit coupling, Zeeman eects and Hyperne coupling to cite only the ones treated in this thesis. The other advantages come from the fact we can approach virtual systems, we have a model which is parameter free (non-empirical method). On the other side, there is no real inconvenience but we have to be aware that we are doing some approximations. These approximations dont come from the fact we associate DFT and LFT but are intrinsic to each method. From DFT, the approximation is due to the fact that exchange-correlation is not calculated exactly: the consequences are that kinetic term and bi-electronic term are approximated and that the self-interaction is not null. From LFT, the approximations are due to the fact that bi-electronic integrals are considered like in a spherical symmetry and to the fact that we neglect some contributions of the ligands. This thesis presents the LFDFT method: a general theory applied to transition metal (TM) complexes which is based on DFT calculations of the manifold of all Slater determinants (SD), originating from a given dn conguration, to parameterize the ligand eld parameters. The description of the theories (DFT and LFT) lying under this method are presented as well as the method by itself (including source code of the programs) and also the tools developed to help along the procedure. The thesis is structured in the following way. The next chapter consists in an introduction to the theories lying under the LFDFT method; it is dedicated to the presentation of the Ligand Field Theory and to the presentation of the Density Functional Theory. In Chapter 3, a full description of the LFDFT is proposed including: the procedure to run the DFT calculations and extract the data from the output les, all the details of the method, the source code of the LFDFT program, and at the end, the LFDFT method is applied to some examples and compared to values tted from experimental spectra. Chapter 4 is dedicated to the determination of the multiplet structure once the ligand eld matrix and the Racahs parameters are determined by LFDFT. And in Chapter 5, the introduction of relativistic eects in LFDFT is proposed. In Chapter 6, we applied the LFDFT method to calculate the g- and A-tensors to a complex with a Schi base which is particularly interesting because its ESR parameters present a large anisotropy. In order to test the LFDFT method on this particular aspect, the values are compared with experiments and with values obtaind by other models implemented in the DFT program used. Finally conclusions are summarized in the last chapter along with an outlook for further extensions of the LFDFT model.

CHAPTER 1. GENERAL INTRODUCTION

Chapter 2 Theoretical Concepts

This chapter gives an introduction to the two theories which are at the basis of the LFDFT method. We can distinguish two parts. Firstly, an introduction to ligand eld theory (LFT) is presented over the Crystal Field Theory (CFTa )2023 and the Angular Overlap Model (AOM).24, 25 All the demonstrations are given for a dn conguration as this work treats only complexes with open d-shell, but has been extended to compounds with open f -shell.2628 Secondly, a description of the Density Functional Theory (DFT) is also presented in this chapter.

2.1

The Crystal Field Theory

Within the Crystal Field Theory we treat inorganic complex as an ionic molecule where the central atom, generally a transition metal (TM) cation, is subjected to a perturbation due to the surrounding ligands. The ligands can be considered as point charges (overlap between the ligands and the TM orbitals is neglected) which provides a constant electric eld having the symmetry of the ligand nuclei arrangement. Thus the crystalline eld leads to a lowering of the spherical symmetry of the free ion. Van Vleck,29 working on the magnetism of complex salts, showed that the Mulliken methods30 of molecular orbital gives a justication to the above point of view. The Hamiltonian of the system can be written as: H = HF + V (2.1) where HF is the Hamiltonian of the free ion and V , the potential provided by the ligands. V is considered as a perturbation which inuences the motions of the metal electrons. If we decompose the Hamiltonian, we can write it down in terms of various contributions: H =
2

2m
T

2 i i

Ze2 1 + ri 2
VN e

i=j Vee

e2 + rij

li si
i

+V

(2.2)

Spin-orbit coupling

where i and j are indices for electrons in the d-shell, T represent the kinetic energy term, N e the nuclear-electron attraction term, Vee the inter-electronic repulsion term, Z is V
CFT is a particular case of LFT because we just take into account the electric eld generated by the ligands and not the polarizability. Within the CFT, the eects of the second sphere of coordination are negligible according to the work of Bethe.
a

CHAPTER 2. THEORETICAL CONCEPTS the eective nuclear charge of the TM cation and the spin-orbit constant of the TM. The order of magnitude of V is dierent depending on the metal considered and we can distinguish the four following cases: (r)l s < e2 /rij V (r)l s e2 /rij V V < (r)l s e2 /rij V (r)l s e2 /rij : complexes of the rst- and second-transition series : complexes of third transition series : complexes of lanthanides : complexes of actinides

If V is less important than the electronic repulsion, the case is called a weak crystal eld otherwise it is called a strong crystal eld case. If we focus on 3d TM, the spin-orbit coupling term of the Hamiltonian in Eq 2.2 can be neglected in a rst approximation. Then to present the CFT in a didactic way, it is useful to consider the strong eld case where the interelectronic repulsion (Vee ) contribution is neglected and consider the TM cation as an hydrogen-like atom (one electron system).

2.1.1

A Single d-electron in a Cubic Field

L

The Hamiltonian of the free-ion with one single electron is equal to: H0 =
2 2

2m

Ze r

(2.3)
r L M R L

where Z is the eective nuclear charge. If we consider the free-ion (TM cation) surrounded by six ligands in an octahedral environment, the Hamiltonian is then H = H0 + V with V being the electrostatic perturbation generated by the six ligands and dened by: V = e (R) |R r| dR (2.4)

where R and r are dened according to Figure 2.1. 2.1.1.1 Expansion of V

Figure 2.1: The Metal surrounded by the 4 ligands in the plane.

The potential V is expanded in terms of Legendre polynomials (cf. Appendix A.1) and reads as Equation 2.5 where r< and r> are respectively r and R the metal to ligand distanceb . i is the angle between Ri and r where i runs over the point charges.

Vi (r) =

(R)
k

1 r R R

Pk (cos i )dR

(2.5)

To proceed, we have to expand Pk (cos i ) using the spherical harmonic addition theorem (cf. Appendix A.3). Moreover it is inconvenient to have complex quantities occurring in
Here we have to make the statement that the electrons of the ions are passive, it means that the energies associated with the electrons of the ligands are low enough compared to the energy of the d-electron of the cations so their role is just to produce a shielding eect of the ligand nuclear charge.
b

2.1 The Crystal Field Theory the potential and so we have to replace the complex spherical harmonic, Ylm , by the set of real quantities as it is explained in Appendix A.2. So the spherical harmonic addition theorem reads: k 4 Pk (cos i ) = ykm ()ykm (i i ) (2.6) 2k + 1 m=k where ykm ()s are real spherical harmonics and (r, ) and (a, i i ) are the polar coordinates of r and Ri , respectively. From Equations 2.5 & 2.6, Vi (for one point charge) reads: k (R) e4 k Vi (r) = r ykq ()ykq (i i )dR (2.7) 2k + 1 Rk+1 k=0 q=k As we can see we have a summation over k from 0 to . But there is no need to investigate a value of k higher than 2l, since the integrals with k > 2l of the product of two spherical harmonics of order l must vanish. In the same way, it is not required to calculate the integrals for the odd values of k as a plane of symmetry containing the z-axis is present so the terms cancel. Equation 2.7 can be written as: V (r) =
i

e4 k r 2k + 1 k=0 q=k

2l

(R) ykq ()ykq (i i )dR Rk+1

(2.8)

This equation can be generalized by:

2l k

V (r) =
k=0 q=k

hkq ykq ()

(2.9)

where hkq is dened by: hkq = =

i m,m

4e rk 2k + 1
m

(R)ykq (i i ) Rk+1 k q (2.10) lm|V |lm

l l (1) (2k + 1) m m

where 2.1.1.2

j1 j2 j3 m1 m2 m3

represents the 3j-Wigners symbol (cf. Appendix A.4).

Evaluation of the Ligand Field Splitting for Oh

As an example, we can consider in a rough approximation ligands as point charges which create an electric eld around the central atom (which is a good approximation because d-orbitals of a TM are well contracted around the TM and the mixing with ligands orbitals is small), so equation 2.4 reads if we consider an octahedral environment:
6

V =
i=1

qi e |Ri r|

(2.11)

CHAPTER 2. THEORETICAL CONCEPTS where r and Ri are respectively the electron position vector and the position vector of the ith point charge (the norm of the Ri being equal to a). Equations 2.11 & 2.4 are equivalent since we express (R) as: (R) = L qL (R RL ) where L is an index for the ligands and qL is the corresponding charge. To determine the ligand eld matrix for a cation in an octahedral environment, we solve Equation 2.12 which is equivalent to Equation 2.8:
6

V (r) =
i=1

Ze

1 4 rk k+1 ykq ()ykq (i i ) 2k + 1 a k=0 q=k

2l

(2.12)

As we are concerned by 3d-electron (n = 3 and l = 2), we have to make a sum over k = {0, 2, 4}. Starting with k = 0, the real spherical harmonic y00 is well known and is equal to: y00 = 21/2 (2)1/2 (2.13) So the potential V(i;x,y,z) from k = 0 is equal to: Vi,k=0 = Ze 4 Ze (21/2 (2)1/2 )2 = a 20+1 a (2.14)

Since there are six ligands around the central atom, V(x,y,z) from k = 0 is equal to 6Ze/a. In the same way, the term for k = 2 is equal to 0 and the term for k = 4 gives: Vi,k=4 = 7 3 Zer4 a5 Y40 + 5 (Y 4 + Y44 ) 14 4 (2.15)

So the potential V due to the ligand is equal to the sum: Vk=0 + Vk=4 . To evaluate the energy shift on the orbital due to the ligand eld, we have to evaluate the following integrals: nlm |V |nlm (2.16) For Vk=0 we can write: E = (e 6Ze/a)nlm = 6Ze2 /a nlm = nlm nlm (2.17)

6Ze2 /a if m = m 0 if m = m

So the energy shift due to Vk=0 is independent of the angular momentum of the orbital considered and is equal to: 6Ze2 /a (cf. Figure 2.2). Doing the same for Vk=4 , we nd the matrix elements equal to: nd+2 |e Vk=4 |nd+2 nd2 |e Vk=4 |nd2 nd1 |e Vk=4 |nd1 nd0 |e Vk=4 |nd0 where: 35Ze2 D= 4a5 = 4Dq = 6Dq = 4Dq = 6Dq 2 4 r 105

(2.18)

and q = 8

nd

(2.19)

2.1 The Crystal Field Theory and: rk

nd

dr r2+k |Rnd (r)|2

(2.20)

We can notice that D depends on the ligands positions and q reects the properties of the central ion electrons. Now, the perturbed energies of the electrons are given the following secular matrix which is arranged by m = 2, 1, 0, 1, 2: 0 4Dq 0 0 0 0 3 0 0 3 4Dq 0 0 0 0 0 0 + 6Dq 0 0 =0 3 0 0 0 0 0 3 4Dq 0 0 0 0 0 3 + 6Dq (2.21) This system has two roots which can be easily determined, the ve degenerate orbitals (eg ) splits into one doubly and one triply degenerate level having for energy eg and t2g dened by: 6Dq xy yz z2 xz 2 x y2 eg = 0 + 6Dq 3 t2g = 0 4Dq 3 (2.22)
0 3

We can see from Figure 2.2 that the rst term is just elevating the energies of all dorbitals and the other ones are symmetry dependent; they split the 5 d-orbitals in two levels where 10Dq is the amount of energy separating them. But for TM with more than one d-electron, neglecting the interelectronic repulsion, except for a didactic purpose, is not a realistic model.

6Ze2 a

4Dq (t2g )

Figure 2.2: Scheme of the crystal eld splitting in an octahedral environment

2.1.2

Two electrons in a Cubic Field

Neglecting spin-orbit coupling, the Hamiltonian of a system with two d-electrons can be written as: H = f1 + f2 + g12 (2.23) 1 Ze2 fi = i + V (ri ) (i {1, 2}) 2 ri is the one-electron operator acting on the ith electron and g12 : g12 = 1/r12 where: (2.24)

(2.25)

is the two-electron operator representing the Coulomb interaction between electrons 1 and 2. It is to note that we can use the same development than in Section 2.1.1.1 to nd the matrix elements of g12 and a program to determine the matrix elements is presented in Section C.2.1. With the Hamiltonian given in Eq 2.23, the Schrdinger equation is o written as: H (r1 1 , r2 2 ) = E(r1 1 , r2 2 ) (2.26) 9

CHAPTER 2. THEORETICAL CONCEPTS where ri and i are respectively the position and the spin ( or ) of the ith electron. The Hamiltonian can be split into two terms: H = H0 + H 1 (2.27)

where H0 is the one-electron Hamiltonian (f1 + f2 ) and H1 is the interelectronic repulsion (g12 ). The eigenfunctions and the eigenvalues of H0 are already known: H0 0 (r1 1 , r2 2 ) = E0 0 (r1 1 , r2 2 ) (2.28)

separating f1 and f2 and looking for the eigenvalues of the one-electron operator which can be written in a general equation: k = k |f |k (2.29)

where f is a spinless operator and k (r) is the product of an orbital function and a spin-function called spin-orbital. Then if we dene: 0kl (r1 1 , r2 2 ) = k (r1 1 )l (r2 2 ) the energy E0kl is equal to: E0kl = k + l (2.31) According to the Pauli principle, wavefunctions for electrons should be antisymmetric with respect to the exchange of electrons. To obey to the antisymmetric condition, the wavefunction can be written as a Slater determinant: 1 k (r1 1 ) l (r1 1 ) 0kl (r1 1 , r2 2 ) = 2 k (r2 2 ) l (r2 2 ) |k l | (2.32) (2.30)

If we now consider two electrons conned into the t2g orbitals, there are 15 possibilities to distribute these two electrons over the 6 spin-orbitals. Here we know that orbitals are degenerate so we can conclude that the one electron operator gives the same results for: |f | = |f | = |f | = t2g (2.33)

The expectation values of the integrals dened in Eq 2.33 are given within the LFDFT method by the energies of the Kohn-Sham orbitals for an Average Of Conguration (AOC) calculation (the calculation details are reported in section 3.1.3). As long as we neglect the coulomb interactions, the fteen -states have the same energies. This high degeneracy is partially removed by the introduction of the interelectronic repulsion. To account for this, we have to calculate the matrix elements of H1 : d1 d2 |l1 m1 s1 (1)l2 m2 s2 (2)| g12 |l1 m1 s1 (1)l2 m2 s2 (2)|
1 2

(2.34)

for the chosen example li represents the t2g orbitals, mi its component and si the spin function. After the expansion of the single determinants, we get the expression: l1 m1 s1 (1)l2 m2 s2 (2)|g12 |l1 m1 s1 (1)l2 m2 s2 (2) l1 m1 s1 (1)l2 m2 s2 (2)|g12 |l2 m2 s2 (1)l1 m1 s1 (2) (2.35) 10

2.1 The Crystal Field Theory Since g12 is independent from the spin coordinate, this equation can be expressed in terms of orbital functions lm (r) rather than spin-orbital functions lms (r) = lm (r) 1 s ():
2

l1 m1 s1 (1)l2 m2 s2 (2)|g12 |l1 m1 s1 (1)l2 m2 s2 (2) = (s1 s1 )(s2 s2 ) l1 m1 (1)l2 m2 (2)|g12 |l1 m1 (1)l2 m2 (2) = (s1 s1 )(s2 s2 ) l1 m1 (1)l2 m2 (2)|l1 m1 (1)l2 m2 (2) We use the following notation for the Coulomb integral: J(l1 m1 l2 m2 ) = l1 m1 (1)l2 m2 (2)|g12 |l1 m1 (1)l2 m2 (2) = l1 m1 (1)l2 m2 (2)|l1 m1 (1)l2 m2 (2) = [l1 m1 (1)l1 m1 (1)|g12 |l2 m2 (2)l2 m2 (2)] and the following for the exchange integral: K(l1 m1 l2 m2 ) = l1 m1 (1)l2 m2 (2)|g12 |l2 m2 (1)l1 m1 (2) = l1 m1 (1)l2 m2 (2)|l2 m2 (1)l1 m1 (2) = [l1 m1 (1)l2 m2 (1)|g12 |l1 m1 (2)l2 m2 (2)] (2.38) (2.37) (2.36)

Then the expression for the diagonal matrix element (l1 = l1 , l2 = l2 , m1 = m1 , m2 = m2 ) of Eq 2.35 can be written as: d1 d2 |l1 m1 s1 (1)l2 m2 s2 (2)| g12 |l1 m1 s1 (1)l2 m2 s2 (2)|
1 2

(2.39)

= J(l1 m1 l2 m2 ) (s1 s2 )K(l1 m1 l2 m2 )

2.1.3

Many Electrons in a Cubic Field

n n

For the many electrons cases, the Hamiltonian of the Eq 2.23 can be generalized to: H =
i=1

fi +
i>j

gij

(2.40)

then if we dene the two wavefunctions and by: = |k1 k2 . . . kn | = |k1 k2 . . . kn | (2.41)

We can calculate separately the matrix elements corresponding to the one-electron Hamiltonian:
N

|H0 | =
i=1

d |k1 k2 . . . kN | fi |k1 k2 . . . kN |

(2.42)

and the matrix elements corresponding to the two-electron Hamiltonian:

N

|H1 | =
i>j=1

dt|k1 k2 . . . kn | g|k1 k2 . . . kn |

(2.43)

Then if we use Laplaces expansion (cf. Appendix A.6), we can demonstrate the Slaters rules which are proposed in Appendix C.3. 11

CHAPTER 2. THEORETICAL CONCEPTS

2.1.4
2.1.4.1

Reduced Electrostatic Matrix Elements (Wigner-Eckart)

Slater-Condon Parameters

One way to reduce the number of two-electron integrals is to assume a simple angular dependence for the t2g and eg wavefunctions. As d-electrons in the crystal are relatively localized around the TM cation and the deformation of atomic d-orbitals is not preponderant, it is reasonable to assume, as rst approximation, that the t2g and eg wavefunctions have pure d-characters and if we consider that the radial part is the same for all electrons occupying the d-orbitals, we can then express molecular orbital wavefunction as: dm (r) = Rd (r)Y2m () m {2, 1, 0, 1, 2} (2.44)

where Y2m () is a spherical harmonic (cf. Appendix A.2). Let us take a two electron integral: 1 dm1 (1)dm2 (2)| |dm1 (1)dm2 (2) (2.45) r12 Then we can expand 1/r12 (cf. Appendix A.1) and using Eq 2.6: 1 = r12
2l k r< 4 k+1 2k + 1 r> k

(1)m Ykm (1 , 1 )Ykm (2 , 2 )

m=k

(2.46)

k=0

where (r1 , 1 , 1 ) and (r2 , 2 , 2 ) are the polar coordinates of the electrons 1 and 2 and where r< and r> are the lesser and the greater of r1 and r2 respectively. If we use real spherical harmonics as we are concerned by real quantities, Eqs 2.45 & 2.46 read: dm1 (1)dm2 (2)|
2l

1 1 |dm1 (1)dm2 (2) = [dm1 (1)dm1 (1)| |dm2 (2)dm2 (2)] r12 r12
k

=
k=0 2l

4 2k + 1 4 2k + 1

ydm1 (1)ydm2 (2)ykm (1)ykm (2)ydm1 (1)ydm2 (2)d1 d2 F k (dd)

m=k k

=
k=0

ydm1 (1)ykm (1)ydm1 (1)d1

m=k

ydm2 (2)ykm (2)ydm2 (2)d2 F k (dd) (2.47)

where (1), (2), d1 and d2 mean respectively (1 , 1 ), (2 , 2 ), d1 d1 sin 1 and d2 d2 sin 2 and where: k r< 2 2 2 2 F k (dd) = r1 dr1 r2 dr2 Rd (r1 )Rd (r2 ) k+1 (2.48) r> 0 0 To simplify Eq 2.47, we can use Clebsch-Gordan coecients, ck 1 ,m (cf. Appendix A.5), m 1 dened as: ck 1 ,m1 = m ylm1 ykm ylm1 dd sin 12 (2.49)

2.1 The Crystal Field Theory Eq 2.47 reads: dm1 (1)dm2 (2)|
2l k

1 |dm1 (1)dm2 (2) r12

1 (1)k ck 1 ,m1 ck 2 ,m2 (m + m1 , m1 )(m + m2 , m2 ) F k (dd) m m 2k + 1 k=0 m=k

2l m1 m1 k=0

= (m1 + m2 , m1 + m2 )(1)

1 ck ck 1 ,m2 F k (dd) m 2k + 1 m1 ,m2 (2.50)

The Slater integrals or the Slater-Condon parameters, Fk , are often used and are dened as: 1 1 4 F0 = F 0 (dd), F2 = F 2 (dd), F4 = F (dd) (2.51) 49 441 2.1.4.2 Racahs Parameters

Another parameterization of the interelectronic repulsion parameters has been proposed by Racah. Eq 2.52 gives the expression of the A, B and C Racahs parameters in terms of the three Slater-Condon parameters: A = F0 49F4 , B = F2 5F4 , C = 35F4 (2.52)

We notice that Racahs parameters are linked linearly to the Slater-Condon parameters so we can use the less physical but more commonly used Racahs parameters for our further studies. | | | | | | | | | | | | | | | | | = = = = = = = = = = = = = = = = = 4B + 3C + A 3B + C 2B + C + A 2B + C + A 4B + C + A 3B 4B + C + A 3B 3B + C 4B + 3C + A 2B + C + A 3B 2B + C + A 3B 2 3B 2B + C + A 3B + C | | | | | | | | | | | | | | | | | = = = = = = = = = = = = = = = = = 4B + 3C + A 2B + C + A 2 3B 2 3B 2B + C + A C 3B + C 3B + C 4B + 3C + A 3B 3B 4B + C + A 4B + C B+C B+C 4B + C 4B + 3C + A

Table 2.1: The non-redundant interelectronic repulsion integrals express in function of the Racahs parameters A, B and C. The adopted notation is: a(1)b(2)|c(1)d(2) .

13

CHAPTER 2. THEORETICAL CONCEPTS Moreover the main advantage of Racahs parameterization is that the dierences in energy between multiplets having the same spin multiplicity are expressed only in term of B. The non-redundant interelectronic integrals in spherical symmetry can be expressed as a function of the three Racahs parameters. Table 2.1 gives the linear relation of each non-redundant interelectronic integral as a function of the three parameters. The Slater determinant energies can be easily calculated within the DFT formalism and for a d2 conguration, we obtain from DFT calculation the 45 energies of all SD. Each of these energies can be expressed in terms of the Racahs parameters A, B and C. The result is an overdetermined system of linear equations which can be solved by least square t. From those, only B and C can be obtained from the t, the parameter A being responsible for a shift of all atomic multiplets by the same amount. 2.1.4.3 Griths Parameters

A more rened treatment of the two electron term of the Hamiltonian H is to use Grifths parameterization.31 In this case, the number of interelectronic repulsion parameters is not limited to three but to a series of ten since it refers to octahedral symmetry and to MO which includes ligand AO. The result found by this approach are not given in this thesis but can be found in the article published in Chem. Phys. Lett..32 However, the details of the Criths parameterization are proposed for d-TM in an octahedral environment. If we consider the Eq 2.47, we see that the electrostatic interaction can be expressed a a sum of products of one electron function. In Eq 2.12, we can replace the real spherical harmonic, ykm , by linear combinations which are components of bases for irreducible representation: G= fiM (1)fiM (2) (2.53)

where fiM is a one electron function which forms the component M of a basis for the irrep and parameter i labels dierent set of functions forming basis for the same irreducible representation . So the two-electron integrals ab|G|cd reads: ab|G|cd = = ab|fiM (1)fiM (2)|cd a|fiM (1)|c b|fiM (2)|d

(2.54)

and depends only on the matrix elements of the one electron operator fiM . If we replace fiM by giM where giM has the same matrix elements as fiM within the ve e and t2 functions of interest but is equal to zero between these functions and any others and, also, between any other functions. Then G reads: G=
<

G() =
iM <

giM ()giM ()

(2.55)

with giM real and having their components correctly connected in phase. A one electron integral a|giM |c can only be non-zero if occurs in the product of the representation 1 and 3 to which |a and |c belong or, if 1 = 3 , to the symmetrized 14

2.1 The Crystal Field Theory square of 1 . So if we write: G=

i

Gi () (2.56) giM ()giM ()

M <

and Gi () =

The product of the two functions e and t2 lead to a sum over four contribution for , one from each irreducible representation A1 , E, T1 and T2 . The derivation of this sum lead to the parameterization of the two-electron integrals in terms of the ten Grith parameters. In Eq 2.57, the relations between the Racahs and the Griths parameters are given if instead of MO, we use pure d-functions: a = e = A + 4B + 3C b = d = A 2B + C c = 2B 3 f = 4B + C g =B+C h=i=B 3 j = 3B + C

(2.57)

2.1.5

Angular Overlap Model (AOM)

More than forty years ago, the developments of LFT by Jrgensen and Schaeer25 give rise to the Angular Overlap Model. In this model, the perturbation of the free ion takes into account the overlap between the metal and the ligands contrary to CFT. For a transition metal with open d-shell surrounded by an octahedral arrangement of ligands, we can distinguish the and the bond (cf. Fig. 2.3): an example of bond is the overlap between the orbital dz2 of the TM and the orbital pz from the ligand, bond is for example the overlap between the orbital dxz of the TM and the px orbital from the ligM and. dz 2 dxz These two interactions can be modeled by two paramepx pz L ters e and e . e is generally positive because -bond has a strong destabilizing eect (for a given metal to ligand distance, the overlap between orbital of the ligand and orbital of the metal is larger for a -bond than for a -bond ) while e can be either positive if the ligand Figure 2.3: Representation of a is -donor or negative if the ligand is -acceptor. - and a -bond and illustration
of the resulting MO splitting.

15

CHAPTER 2. THEORETICAL CONCEPTS Metal AO dz2 dx2 y2 dxz dyz 1 0 0 0 0 0 1 1 1/4 3/4 0 0 0 0 1 0 1/4 3/4 0 0 0 0 0 1 1/4 3/4 0 0 0 0 1 0 1/4 3/4 0 0 0 0 0 1 1 0 0 0 0 0 1 1 3 3 0 0 0 0 4 4

Ligand 1 2 3 4 5 6 sum

dxy 0 0 0 1 0 1 0 1 0 1 0 0 0 4

Figure 2.4: Ligand positions for the demonstrative case.

e Td
d-orbitals of free ion

Oh t Oh

Td

Table 2.2: Overlap contributions due to each ligand with the ve d-orbitals of the TM.

Figure 2.5: Scheme of the AOM splitting in an octahedral and in a tetrahedral environment.

Let us take as example an octahedral complex where the ligands are labelled as in the Figure 2.4. The contributions of each ligand to the and the perturbations are listed in the Table 2.2. Then if we do the overall sum, the energy of the MO eg and t2g are then formulated as eg = 3e and t2g = 4e leading to a diagonal hab matrix with 10Dq = 3e 4e . The great advantage of using overlap arguments when we parameterize hab lies in the fact, that overlap energies may be expressed in terms of parameters e and e which reect the strength of the metal-ligand bond. The angular dependence is given by the rotational transformation properties of the d-orbitals since ligand-ligand interaction is neglected. The AOM matrix, hab , can be expressed in terms of angular factors and the e and e parameters. Thus the model has ability to treat systems with little or no symmetry on an equal footing as the higher symmetric ones. Its drawback lies in the large number of model parameters to be determined from experiment for such systems, resulting in a complex parameterization. We can make two remarks on the AOM theory. 1) Knowing that e is bigger than e (in general 4 to 5 times, cf. MO diagram in Fig 2.3), the Figure 2.5 can also be explained in a qualitative way if we consider the nature of the bond in a complex, let say if the bond is composed by a -interaction or by a -interaction. From the values of e and e , we can say that the MO splitting in an octahedral environment is bigger than the MO splitting in a tetrahedral one. 2) From Table 2.2, we can remark that for each ligand, the sum over and the sum over angular factors are respectively 1 and 2. We can link these numbers to the well known fact that a ligand is linked to the transition metal by one (usually pz ) and two (px , py ) bonds. 16

2.1 The Crystal Field Theory

2.1.6

Spin-Orbit Coupling

Spin-orbit coupling is an essential constituent to the Hamiltonian for electronic states originating from dn -congurations of transition metals in ligand elds (LF). It governs the ne structure of the electronic multiplets and, for the ground state it is mainly responsible for the zero-eld splitting and the anisotropy aecting the spectroscopic and magnetic behavior of TM compounds with open d-shells. The ligand eld and the interelectronic repulsion dominate over the spin-orbit coupling in complexes of 3d transition metal series. This is the reason to neglect the latter thus far. However magnetism, EPR spectroscopy and hyperne-coupling are largely aected by spin-orbit coupling. Moreover, for 2nd and 3rd row transition metals as well as for the f -elements even qualitative accounts of electronic absorption spectra cannot neglect spin-orbit eects. Similarly, for molecular orbitals calculations in which ligands play an important role, spin-orbit coupling due to ligands like e.g. iodide or bromide must be considered. The nature and origin of spin-orbit coupling have been discussed in many places.33 Misetich and Buch34 have shown that the spin-orbit Hamiltonian of a molecule can be reasonably well approximated as HSO =
N,i

N li,N si =
i

ui,N si

(2.58)

where N , the spin-orbit coupling constant of nucleus N, is incorporated into the molecular operator ui for electron i. In order to carry out a spin-orbit calculation it is necessary to relate the resultant splitting of many electron states and also the interaction of dierent states to one-electron spin-orbit coupling matrix elements. This can be done most conveniently in the basis of micro-states (single determinants) as shown in references11, 32, 35 for the calculation of multiplets, using Slaters rules.

2.1.7

Qualitative Considerations of LFT

The use of LFT over decades led to the formulation of some general rules. Here, a brief introduction to two of them is presented: the notion of spectro-chemical series and the notion of nephelauxetic eect. In fact, to test our LFDFT method, we will compare quantitatively the values obtained to the ones tted from experiments but also qualitatively to see if the evolution of the values respects this two rules.

2.1.7.1

The Spectro-chemical Series

Chemists remarked that if we consider one metal, the energy splitting between orbitals change in function of the ligands. Jrgensen36 studied this eect and classied the ligands in order of their ability to cause orbital separations. This classication is called the spectro-chemical series and in Figure 2.6, a shortened list is given. 17

CHAPTER 2. THEORETICAL CONCEPTS Weak Field I < Br < SCN Cl < F < OH ONO < C2O2 < H2 O 4 < NCS < EDTA4 < NH3 pyr en < bipy < phen < CN CO Strong Field
Figure 2.6: A shortened spectro-chemical series.

The negatively charged ligands, such as I < Br < Cl < F , give rise to smaller 10Dq values than neutral molecules, like H2 O < NH3 < pyridine, CN and CO being the strongest ligands due to their ability for back bonding (in AOM theory 10Dq = 3e 4e , epi < 0). 2.1.7.2 The Nephelauxetic Eect

Using LFT theory many years ago, Scher and Jrgensen37 discovered that the intera electronic parameters B and C in complexes are smaller than those for the free ions. This phenomenon has been rationalized in terms of the electronic cloud expansion of the d-orbitals when going from free TM ions to complexes and has been called the nephelauxetic eect. The more reducing and softer ligands show a stronger reduction than the more oxidizing and harder ones. The nephelauxetic reduction factor is dened by the relation: Bcomplex (2.59) xBcomplex = B0 where B0 is the value of B for the free ion, Bcomplex the value of B for the complex considered and xBcomplex the nephelauxetic reduction factor of B for the complex considered. Another way to see the nephelauxetic eect is that depending on the ligand, the complex becomes more covalent and the electrons are to some extent spread over the ligands so the d-orbitals are in some way large, we call it cloud expansion. So the space at the electron disposal is bigger and the electron-electron repulsion is reduced.

2.2

DFT

The DFT which is well described in many text books3841 is the method we choose to determine the ligand eld parameters. This choice is justied, based on the pragmatic observation, that DFT is computationally less intensive than other methods with similar accuracy. In this section, an introduction to DFT is reported. The main fact in DFT is that the energy of a molecule can be determined from the electron density instead of a wavefunction (Hohenberg and Kohn theorems). From the DFT calculations, in our LFDFT model, we will use the Kohn-Sham (KS) molecular orbitals (MO) to model the hab ligand eld matrix and the energies of Slater determinants to t the Racahs parameters.

2.2.1

Basic Theory

Within the Born-Oppenheimer approximation, we can separate the electronic Hamiltonian, Helec , from the nuclear part. For a N -electrons and M -nuclei system without 18

2.2 DFT relativistic eects, Helec is dened by: Helec = or in a short hand notation: Helec = T + VN e + Vee (2.61) T , VN e and Vee are the kinetic energy term, the electron-nuclear attraction term and the electron-electron repulsion term, respectively. The total energy of the system can be found by solving the following Schrdinger equation: o Helec elec = Eelec elec Eelec = elec |Helec |elec (2.62) 1 2
N 2 i i=1 N

+
i=1

ZA + riA A=1

i=1 j>i

1 rij

(2.60)

and then adding the eect of the constant interatomic repulsion potential: VN N (VN N is constant due to the Born-Oppenheimer approximation): Etot = Eelec + Enuc where: Enuc = ZA ZB rAB A=1 B>A
M M

(2.63)

(2.64)

The attractive potential exerted by the nuclei on the electrons is often called the external potential, Vext , because it is not limited to the nuclear eld but it may include some other perturbations like eects of magnetic or electric elds. The wavefunction is not an observable and a physical interpretation is relevant only if we consider the square of the wavefunction: |(x1 , x2 , . . . , xN )|2 dx1 dx2 . . . dxN (2.65) which represents the probability that electrons 1, 2, . . . , N are found simultaneously in the volume elements dx1 dx2 . . . dxN . As the electrons are indistinguishable, the following equality can be written: |(x1 , x2 , . . . , xi , xj , . . . , xN )|2 = |(x1 , x2 , . . . , xj , xi , . . . , xN )|2 Then the electronic density (r) is easily introduced as: (r) = N ... |(r1 , x2 , . . . , xN )|2 ds1 dx2 . . . dxN (2.67) (2.66)

where ri = {xi , yi , zi }, xi = {xi , yi , zi , si } and (r) determines the probability to nd one of the N -electrons within the volume element dr1 but with arbitrary spin while the others N 1 electrons have arbitrary positions and spin in the state represented by . We can then make the two following statements:
r

lim (r) = 0 (r) dr = N 19

(2.68) (2.69)

CHAPTER 2. THEORETICAL CONCEPTS This notion of the electron density can then be extended to the notion of the pair density and instead to have the density of probability to nd one electron with an arbitrary spin within a particular volume element while other electrons are anywhere, we deal with the probability to nd a pair of electrons with spins 1 and 2 simultaneously within two volume elements dr1 and dr2 . This quantity is dened by the Eq 2.70: 2 (r1 , r2 ) = 2 (r1 r2 , r1 r2 ) N (N 1) = ... 2

|(r1 s1 , r2 s2 , x3 , . . . , xN )|2 ds1 ds2 dx3 . . . dxN

(2.70)

It should be easily calculated if the electrons cannot interact with themselves, but the electrons are charged and the movement of one is correlated with the movements of the others. Then 2 (x1 , x2 ) contains electron correlation. To describe 2 , we have to introduce the reduced density matrix 2 : 2 (r1 r2 , r1 r2 ) = N (N 1) ... (r1 s1 , r2 s2 , x3 , . . . , xN )(r1 s1 , r2 s2 , x3 , . . . , xN )dx3 . . . dxN

(2.71) 2 is called the second order density matrix. The rst order density matrix can be dened in the same way as: 1 (r1 , r1 ) = N ... (r1 s1 , x2 , . . . , xN )(r1 s1 , x2 , . . . , xN )dx2 . . . dxN (2.72)

If we consider in Eq 2.70 the case where r1 = r2 then: 2 (r1 , r1 ) = 2 (r1 , r1 ) (2.73)

This means that the quantity previously dened is 0, this implies that the probability to nd two electrons at the same place vanishies according to the Paulis principle. If we integrate upon the spin component, we thus get the density (r) whose eigenvectors are called natural orbitals. All the interactions described by the Hamiltonian are either one- or two-particle interactions. Therefore just 1st and 2nd order density matrices (often called geminal functions) are necessary to describe a system of N interacting electrons. An equivalent formulation of the Hamiltonian of Eq 2.60 in term of density matrices is: |H | = where: 1 (r1 , r1 ) = N ... (r1 s1 , x2 , . . . , xN )(r1 s1 , x2 , . . . , xN )d1 dx2 . . . dxN (2.75) ... |(r1 s1 , r2 s2 , x3 , . . . , xN )| ds1 ds2 dx3 . . . dxN
2

1 2

2 1 1 (r1 , r2 ) r1 =r2

dr1 +

Vext (r)dr +

1 2 (r1 , r2 )dr1 dr2 (2.74) r12

N (N 1) 2 (r1 , r2 ) = 2

Thomas and Fermi were the rst to use the density as a variable to describe the ground state with the Thomas-Fermi model. 20

2.2 DFT

2.2.2

The Thomas-Fermi Model

Thomas and Fermi,4244 before the development of the DFT due to Hohenberg and Kohn,45 introduced a model close to density functional formalism where the energy is expressed in function of : E[] = |H| = (r)V (r) dr + T [] + Vee [] (2.76)

and they proposed to substitute the kinetic energy function with the one derived for a uniform electron gas TT F : TT F [(r)] = 3 (3 2 )2/3 10 5/3 (r)dr (2.77)

while solving the two other terms in a classical way so Eq 2.76 reads: ET F [(r)] = 3 (3 2 )2/3 10 5/3 (r)dr Z (r) 1 dr + r 2 (r1 )(r2 ) dr1 dr2 r12 (2.78)

This equation was the rst to introduce the density rather than the wavefunction in order to solve the Hamiltonian. But the importance of this equation is more historical because it cannot really calculate the energy of an atom, this is due to the fact that Eq 2.78 completely neglects the exchange and correlation eect. The only part of the interelectronic interaction taken into account is the Coulomb repulsion, therefore the chemical bonds cannot be well described.46 Dirac proposed the Thomas-Fermi-Dirac (TFD) model47 that corrects the TF model by the introduction of exchange of electron gas but TFD model does not constitute a great improvement. A successive correction was given by Weizsacker48 who proposed a correction of the kinetic energy term with the introduction of TT F which takes into account the inhomogeneity of the density: TW [] = 1 8 | (r)|2 dr (r) (2.79)

which leads to the total energy of the T F D W model: ET F DW = TT F [] TW [] +

x where ED [] is the Dirac correction: x (r)V (r) dr + J[] + ED []

(2.80)

x ED []

3 = 4

1/3

(r)4/3 dr

(2.81)

In a rst time, Weizsacker took = 1 but this value changed in many developments.4952 The advantages of the T F DW model are that the density is now nite at the nuclei and has a correct asymptotic behavior. Moreover, the T F D W model is able to describe the chemical bond. But the real development of DFT comes from the two theorems of Hohenberg and Kohn. 21

CHAPTER 2. THEORETICAL CONCEPTS

2.2.3
2.2.3.1

The Hohenberg and Kohn Theorems

The First Hohenberg and Kohn Theorem

The rst Hohenberg and Kohn theorem45 states, in case of non degenerate ground state, that the density (r) completely determines the external potential Vext . The proof is given by reductio ab absurdum supposing the existence of two dierent external potentials Vext and Vext being part of two dierent Hamiltonians: H = T + Vee + Vext and H = T + Vee + Vext . Then the two Hamiltonians belong to two dierent ground state wavefunctions and corresponding to two ground state energies E0 and E0 . If we assume that the two wavefunctions give rise to the same electron density and if we dene E0 = |H | then: E0 < |H | = |H | + |H H | H and H just dier by the external potential so: E0 < E0 + |T + Vee + Vext T Vee Vext | E0 < E0 + (r) {Vext Vext } (2.83) (2.82)

If we proceed in the same manner with interchanging the unprimed and the primed quantities we obtain: E0 < E0 + Eq 2.83 & Eq 2.84 give the absurd result: E0 + E0 < E0 + E0 (2.85) (r) {Vext Vext } (2.84)

This is the proof that two dierent external potentials, Vext and Vext , can not yield to the same ground state electron density. 2.2.3.2 The Second Hohenberg and Kohn Theorem

The second Hohenberg and Kohn theorem45 introduces the variational principle in terms of the density. Choosing an arbitrary density , the corresponding unique wavefunction leads using the variational principle to: |H| = Vext dr + T [] + Vee [] = E[] > E[] (2.86)

where is the true ground state density. Then the variational principle requires that the ground-state density satisfy the stationary principle: E[] (r)dr N = 0 (2.87)

where the integration of the density is imposed to be equal to N . Then using EulerLagrange equation: F [] E[] = Vext (r) + (2.88) = (r) [r] 22

2.2 DFT where F [] = T [] + Vee [] is called the universal functional and is independent of the system. The latter equation shows that knowing the universal function, the best density can be obtained trough a variational procedure. All properties of the system dened by an external potential Vext are determined by the ground state density.

2.2.4

The Constrained Search Approach

E0 = min |T + VN e + Tee |
N

The variational principle applied to a wavefunction is dened by: (2.89)

But if we want to connect the variational principle to the DF formalism, we have rst to dene the v- and N -representability of the density. A density is v-representable if it is associated to the wavefunction of a Hamiltonian dened in Eq 2.61 where the external potential Vext (r) is not necessarily a Coulomb potential. A density is N -representable if it can be obtained by squaring the antisymmetric potential wavefunction. Then the N -representability is a weaker condition than the v-representability. The HK theorems demonstrate a one-to-one correspondence between 0 (r) and 0 (r) (the subscript 0 referring to ground state). But we want to determine the wavefunction in terms of a density, this is not trivial because an innite number of antisymmetric wavefunctions give rise to the same density. How to determine the real wavefunctions 0 and non 0 ? The Levy constrained search5355 is the answer because the minimum energy principle gives: 0 |H |0 So we can write: 0 |H |0 = 0 |T + Tee |0 + 0 |T + Tee |0 + which is equivalent to: 0 |T + Tee |0 0 |T + Tee |0 (2.92) (r)Vext (r)dr (2.91) (r)Vext (r)dr 0 |H |0 = E0 (2.90)

If we rewrite the second HK theorem for a v-representable density we obtain: E[] FHK [] + Therefore we have: FHK [0 ] = 0 |T + Vee |0 = min |T + Vee |
0

(r)Vext dr

Ev [0 ]

(2.93)

(2.94)

The denition of the constrained search comes from the fact that only the wavefunctions of 0 density are taken into account. If we look at Eq 2.94, we can notice that we are not obliged to restrict to a v-representable 0 ground state but to any N -representable density. Therefore we can extend the denition of FHK [0 ] to any N -representable density: F [] = min |T + Vee |

(2.95)

23

CHAPTER 2. THEORETICAL CONCEPTS where F [] = FHK [0 ]. The minimization procedure can be performed in two steps: F [] = min min
N

|T + Vee | +

(r)Vext dr

(2.96)

which can be written in a more compact form as: E0 = min F [] +

N N

(r)Vext dr (2.97)

= min E[]

2.2.5

The Kohn-Sham Approach

Kohn and Sham3 suggested to take into account a system of N non-interacting electrons (Vee = 0) moving in an external potential Vext . For this approximation the kinetic energy is well known and corresponds to the kinetic energy, TS , dened as:
N

TS =
i=1

i |

1 2

|i

(2.98)

but the true kinetic energy, T , is not equal to TS . Then Kohn and Sham introduce this dierence in the functional thus from Eq 2.88: = Vext (r) + T [] Vee [] TS [] + = Vext (r) + + Vef f (r) (r) (r) (r) (2.99)

where Vext (r) is a ctitious external potential which should be applied to the non-interacting system to get the density : Vext (r) = Then the total energy is: E[] = TS [] + J[] + T [] TS [] + where: J[] = If we set: EXC [] = (T [] TS []) + (Vee [] J[]) then the energy of the system can be expressed as: E[] = TS [] + J[] + EXC [] + Vext (r)dr (2.104) (2.103) 1 2 Vext (r)dr + Vee [] J[] (2.101) T [] TS [] Vee [] + (r) (r) (r) (2.100)

(r1 )(r2 ) dr1 dr2 |r1 r2 |

(2.102)

where EXC is called the exchange-correlation energy. The exchange-correlation energy functional can be derived with respect to the density to dene the exchange-correlation potential VXC []: EXC [] (2.105) VXC [] = (r) 24

2.2 DFT From Eq 2.101, the KS equations can be derived: where: (r) =
i=1

1 2

(r ) + Vext (r) + VXC (r) i = i i |r r |

N

(2.106)

|i (r, )|2

(2.107)

and the total energy is equal to:

N

E=
i=1

1 2

(r)(r ) dr dr + EXC [] |r r |

VXC (r)(r)dr

(2.108)

We can underline that EXC does not contain only the exchange and correlation energy but also the correction of the kinetic energy: correlation kinetic energy. The main advantage of the DFT approach is the substitution of the non-local non-multiplicative HF term by a local and multiplicative exchange and correlation term. On the other hand, the functional form of EXC is unknown. Therefore all DFT applications are based on an approximate EXC (cf. next sections). As underlined before, Vext can not only consider the potential of the nuclei but can also take into account, for example, a magnetic eld which acts only on the spin of the electrons. For this case the total density (r) and the electron density ((r) , (r) ) should be consider as variables. The Hamiltonian in presence of an external eld, B(r), reads: Helec 1 = 2
N 2 i i=1 N

+
i=1

ZA + riA A=1

i=1 j>i

1 + 2 e rij

B(r) Si
i=1

(2.109)

where e = e /2mc and the interaction of orbital angular momentum of electron with the magnetic eld has been neglected. It can be shown38 that the constrained search formulation can be rewritten as: F , = min
,

|T + Vee |

(2.110)

Therefore following the KS method we obtain: F , = TS , + J , + EXC , and nally, we obtain the spin-polarized set of KS equations: 1 h f = ef 2 1 h f = ef 2 and:
Vef f (r) 2 i 2 i + Vef f (r) i (r) = i i (r)

(2.111)

(2.112) +
Vef f (r)

i (r) = i i (r)

(r ) EXC [ , ] = Vext (r) + e b(r) + dr + |r r | (r) (r ) EXC [ , ] Vef f (r) = Vext (r) + e b(r) + dr + |r r | (r) 25

(2.113)

CHAPTER 2. THEORETICAL CONCEPTS where b(r) is the magnetic eld in the z direction. Actually, the spin-polarized extension can also be applied in absence of an external eld to give a better description of the EXC for open shell systems and to allow dierent spins to have dierent densities. The latter point is equivalent to the concept of UnrestrictedHF and should account for molecules at the near dissociation limit and for processes of spontaneous localization/magnetisation.

2.2.6

Hole Functions

From the Eq 2.74, we can formulate the Hamiltonian in terms of rst and second order density matrices and express the energy as: |H | = 1 2
2 r 1 (r, r

)
r=r

dr +

Vext 1 (r)dr +

1 2 (r1 , r2 )dr1 dr2 (2.114) r12

In the case we have independent electrons: 1 2 (r1 , r2 )dr1 dr2 = (r1 )(r2 ) 2 In the case we have interacting electrons: 1 2 (r1 , r2 )dr1 dr2 = (r1 )(r2 ) [1 + h(r1 r2 )] 2 (2.116) (2.115)

where the last terms incorporates all the interaction between the electrons. If we now recall that: 2 (r1 ) = 2 (r1 , r2 )dr2 (2.117) N 1 We can dene the exchange-correlation hole as: xc (r1 , r2 ) = (r2 )h(r1 , r2 ) From Eq 2.116 & 2.117 we can write: 2 N 1 (r1 ) = (r1 ) + (r1 ) N 1 2 2 and: xc (r1 , r2 )dr2 = Therefore from Eq 2.114, we derive: Vee = = 1 (r1 , r2 )dr1 dr2 r12 (r1 )xc (r1 , r2 ) dr1 dr2 r12 (2.121) (r2 )h(r1 , r2 )dr2 = 1 (2.120) xc (r1 , r2 )dr2 (2.119) (2.118)

1 1 (r1 )(r2 )dr1 dr2 + 2 2 = J[] + Exc []

If now we separate the spin-like pairs of electrons from the opposite spins pairs, we have: Exc [] = 1 2 (r1 ) (r1 , r2 ) 1 x dr1 dr2 + r12 2 26 (r1 ) (r1 , r2 ) c dr1 dr2 r12 (2.122)

2.2 DFT where (r1 r2 ) is the Exchange Hole or Fermi Hole and it contains both the exchange x contribution to the total energy and the correlation of electrons of the same spin, and (r1 r2 ) is the Correlation Hole or Coulomb Hole and it describes the correlation energy c for electrons of opposite spin. The general condition for the exchange-correlation hole function can be rewritten in terms of Exchange and Correlation Hole functions as: (r1 , r2 )r12 dr1 dr2 = 1 x (2.123) (r1 , r2 )r12 dr1 dr2 c =0

Let us consider the exchange correlation hole. Dening the position of electron 2 as a function of the position of electron 1 ( r2 = r1 + s), we have:
s0

lim (r1 , r1 + s) = (r1 ) x

(2.124)

This condition is equivalent to Pauli principle and it exactly compensates the self interaction energy contained in the J[].

2.2.7

Exchange and Correlation Functional

With the aim to nd functionals which can give a better description of chemical systems, several classes of EXC have been proposed. Unfortunately, a straightforward way in which the functional can be improved does not exist. Nevertheless two main routes have been followed: the rst is to start from a model for which an exact solution can be found and try to design functionals which satisfy the theoretical conditions; the second is, starting from a given functional form, to t a number of parameters in order to reproduce the chemical properties of a bunch of compounds taken as reference. The last approach can yield better results but it introduces a lack of generality in the EXC and therefore this concept of a functional optimized for a given set of molecules certainly drives DFT towards the semi-empirical approaches. Although the number of functional is constantly increasing, we can recast most of the current functionals into three dierent classes: the Local, the Gradient Corrected and the Hybrid ones. It is important to stress that all the functionals are local, even if sometimes the generalized gradient approximation was misleadingly referred as non-local approximation, since they depend only upon the density or gradient of density at the given position. Finally a short overview of the so-called meta-GGA is given in section 2.2.7.4. 2.2.7.1 Local Density Approximation

A simple reference model for which an exact solution can be found is the uniform electron gas. A uniform electron gas can be thought as a system of N electrons in a cube of volume V = a3 where a uniform charge has been spread out in order to maintain the overall system neutral, when N , V while the density remains constant = N . V In this case the ground state energy reads: E[] = TS [] + (r)Vext (r)dr + J[] + EXC [] + Eb 27 (2.125)

CHAPTER 2. THEORETICAL CONCEPTS with Eb being the electrostatic energy of the positive background. This electrostatic energy does compensate the sum of the external potential generated by the electrons and the coulombic interaction between electrons. Therefore the second, third and fth terms of Equation 2.125 add to zero. In this case, we can thus write: E[] = TS [] + EXC [] = TS [] + EX [] + EC [] (2.126)

It can be shown that, using plane waves, the rst order spinless density matrix can be expressed as: 1 (r1 , r2 ) = 3(r)
2

sin t t cos t t3 t = kF (r)s

1/3

(2.127)

kF (r) = [3 (r)] s = r1 r 2 Consequently, we have: TS [] = CF EX [] = Cx

(r)5/3 dr (r)4/3 dr (2.128)

3 CF = (3 2 )2/3 10 1/3 3 1 Cx = 4 3

From Eq 2.128 the TFD formulation can be recognized. In case of a system with a dierent number of and electrons with the corresponding density and , the kinetic energy and the exchange energy can be expressed as: TS [ , ] = 22/3 CF EX [ , ] = 2
1/3

( (r))5/3 + ( (r))5/3 dr (2.129) ( (r))

4/3

Cx

+ ( (r))

4/3

dr

The correlation functional for a uniform electron gas, was derived from the numerical Monte-Carlo simulations of Ceperley and Alder,56 subtracting from the total energy the kinetic and the exchange ones obtained through Eq 2.126. And then Vosko, Wilk and Nusair57 formulated the correlation energy per particle, c , in an analytical way. If rs is dened as the radius of a sphere whose volume is the eective volume of an electron: 4 3 1 r = 3 s The expression of c (rs ) reads: c (rs ) = A x2 2b Q ln + arctan 2 X(x) Q 2x + b 2 bx0 (x x0 ) 2(b + 2x0 ) Q ln + arctan X(x0 ) X(x) Q 2x + b 28 (2.130)

(2.131)

2.2 DFT where x = rs , X(x) = x2 + bx + c, Q = (4c b2 ) and A = 0.0621814, 0.0310907; x0 = 0.409286, 0.743294; b = 13.0720, 20.1231; c = 42.7198, 191.578 for the non spinpolarized and for the spin-polarized cases respectively. The LDA exchange is often referred as S-VWN to stress the two components of the exchange (Slater-Dirac) and correlation (VWN) parts respectively. Most of the functionals developed contain the LDA exchange and a correction to it. Therefore we expect that for LDA functional, the more inhomogeneous the system is, the worst the description should be. c 2.2.7.2 Generalized Gradient Approximation
1/2

One of the most striking failure of LDA is its wrong asymptotic behavior. Becke58 starting from the denition of exchange and correlation hole (Eq 2.118) has dened exchange energy as a function of the exchange energy density (x (r1 )) as: Ex = 1 x (r1 ) = 2 (r1 )(r1 )dr1 xc (r1 , r2 ) dr2 r12 (2.132)

The following asymptotic condition should be fulllled:

r1

lim (r1 ) =

1 2r1

(2.133)

In the case of LDA functional, the asymptotic condition is equivalent to:

r1

lim (r) exp(r)

(2.134)

so Becke dened the asymptotic condition as:

r1

lim xLDA exp(r/3)

(2.135)

Therefore an additional correction should be added. The one proposed by Becke in order to recover the correct asymptotic behavior contains the gradient of the density and has the following functional form: xBecke = 1/3 x= 4/3 x2 1 + 6x sinh1 x

(2.136)

This expression contains one adjustable parameter () which was chosen so that the sum of the LDA and Becke exchange terms reproduces the exchange energy of six noble gas atoms ( = 0.0042). The Becke exchange energy functional has the correct asymptotic behavior but the derived potential not (i.e. it behaves like r2 instead of r1 ). On the other hand the introduction of the gradient of the density enables a better treatment
We are, within the LFDFT method, using LDA functional to do geometry optimization if we dont have the experimental structure, especially for small coumpounds with negative charge. Sometimes we use it for all the DFT calculations of the LFDFT procedure.
c

29

CHAPTER 2. THEORETICAL CONCEPTS of inhomogeneous systems such as the molecular systems we are normally dealing with. As a typical example of Gradient corrected correlation functional we can consider the Lee-Yang-Parr correlation potential. Starting from the work of Colle and Salvetti59 who formulated and approximated expression of the correlation energy of He as a function of second order HF density matrix, Lee, Yang and Parr60 expressed this formula in term of , its gradient and its Laplacian. After integration by part61 the Laplacian terms are eliminated and the nal expression for a closed shell system reads: EcLY P = a where w= exp(c1/3 ) 11/3 1 + d1/3 d1/3 = c1/3 + 1 + d1/3 dr ab 1 + d1/3 w2 CF 8/3 + | |2 5 7 12 12 11 2 | |2 dr 24 (2.137)

(2.138)

and a = 0.04918, b = 0.132, c = 0.2533, d = 0.349 are the Colle-Salvetti parameters for He. Contrary to the Becke functional for exchange, the LYP functional is not derived from a uniform electron gas model but from correlated wavefunction for a two electrons system. Several other GGA exchange and correlation functionals have been developed from the late 80s until now. 2.2.7.3 Hybrid Functional

As well as pure-GGA, an increasing number of exchange and correlation functionals include a percentage of exact exchange. This latter class of functionals recasts the socalled Hybrid-Functionals. We can suppose that a system of density can be described as a sum of ctitious systems, all with density but with dierent electron-electron interaction. This interaction can be tuned via a parameter such that = 1 corresponds to full interacting electrons and = 0 corresponds to non-interacting electrons. Therefore we can write: F [] = min |T + Vee | (2.139) where minimizes the Hamiltonian T + Vee and has an exact density . From the expression of Exc given in Eq 2.103 we have: Exc = T [] TS [] J[] + Vee [] = F1 [] F0 [] J[]
1

=
0

F d J[]

(2.140)

Using the Hellman-Feynman theorem we can write:

1

Exc =
0

d |Vee | J[] 30

(2.141)

2.2 DFT If now we evaluate the integral using a two-point quadrature, we just need the value of the integral for = 0 and = 1, the rst being the HF-exchange and the latter the full exchange-correlation energy. This leads to the so-called half-and-half formula where 50 % of HF-exchange has been introduced. Successively, Becke62 proposed a dierent (empirical) mixing of exact exchange and exchange-correlation energy derived from dierent exchange-correlation functionals, leading to the family of the three parameters hybrid functional, the most known being B3LYP:
Dirac HF V LY B3LY P AEx + (1 A)Ex + BBecke + (1 C)Ec W N + CEc P x

(2.142)

2.2.7.4

meta-GGA

With the denition of meta-GGA a new family of exchange-correlation functionals was introduced by Perdew.63 The aim of this new class of functionals is to overcome the shortcomings of LDA and GGA approaches due to their foundation on a strictly local formulation. Therefore, to reach chemical accuracy, non local terms are included in the exchange-correlation functional formulation. In particular we can see, that at a LDA level, EXC is only a function of the density () and at the GGA level, only a function of the density and its gradient ( and ). Instead of proposing a fully non local functional, Perdew63 rstly proposed an exchange and correlation functional which depends on the density and the gradient of the density ( and ) as a common GGA but also on the Laplacian of the density ( 2 ) or of the kinetic energy density of the occupied orbitals. Dierent meta-GGA functionals have been developed6466 but we are not able to use them, up to now, in our LFDFT approach. It comes from the fact that meta-GGA functional are implemented in the ADF program as a post-SCF procedure. So we can calculate energies in regards of these kinds of functionals but the density matrix is not optimized according to them. As one can see in the next chapter, we need to extract eigenvectors of the Kohn-Sham molecular orbitals (KS MO) from DFT calculations in our LFDFT approach.

2.2.8

Scaling Relations in DFT

Let us consider a change of the electron coordinates in the system, in order to expand ( > 1) or to shrink ( < 1) the electron cloud: r = r. The scaling factor is dened as the operation on the wavefunction or on the density which changes all the coordinates but keeping the normalization condition. Thus we have: = 3N/2 (r) = 3 (r) (2.143)

If we express the kinetic energy and the electron-electron interaction as a functional of the wavefunction, they scale homogeneously as 2 and : T [ ] = 2 T [] Vee [ ] = Vee [] (2.144)

If, now, we express the T and Vee as a functional of the density, the scaling relations are no more the same. In particular, recalling the Levy constrained search formulation, we 31

CHAPTER 2. THEORETICAL CONCEPTS have: F [(r)] = min

(r) T (r) + Vee [r] (r)dr (r) T (r) + Vee [r] (r)dr (2.145) (r) T (r) + Vee [r] (r)dr

= min3N

= min2

= min2 |T + Vee | 3N/2 min (r) gives the density and minimizes the expectation value of 2 T + Vee , while min gives and minimizes T + Vee . If we dene min the wavefunction with the corresponding density which minimizes T + Vee , the following inequality arises: 3N/2 min (r) = min (2.146)

As a consequence T [] and Vee [] do not scale homogeneously. The following scaling relations have been derived by Perdew and Levy:67 Vee [] < Vee [] ; Vee [] > Vee [] ; T [] > 2 T [] for < 1 T [] < 2 T [] for > 1 (2.147)

In case of non-interacting electrons, the KS kinetic energy scales homogeneously: TS [] = 2 T [] as well as the exchange-energy functional and the classical Coulomb potential: Ex [] = Ex [] J[] = J[] (2.149) (2.148)

What does not scale homogeneously is Tc [], the dierence between the real kinetic energy and the kinetic energy of a non-interacting electron system: Tc [] = T [] Ts [] The following inequality can be demonstrated:4, 68 T [] Therefore: Tc Correspondingly the correlation energy reads: Ec [] = Vee [] J[] Ex [] + Tc [] (2.153) 0 (2.152) Ts [] (2.151) (2.150)

Therefore Ec [] always contains a positive contribution due to kinetic energy and particular care should be taken in direct comparison of correlation energy derived from correlated 32

2.2 DFT calculations and the correlation energy derived from DFT. As previously seen for Ex , scaling inequalities can be derived also for the correlation energy:67 Ec [ ] < Ec [] for < 1 Ec [ ] > Ec [] for > 1 (2.154)

The scaling relations in DFT have been reviewed by Levy.69 As a nal remark, it is important to point out that most of the current functionals do not satisfy several of these inequalities or the asymptotic conditions, which in principle, should be taken as a rigourus guide in the design of new functionals.

2.2.9

Janaks Theorem

Looking at the KS Equation 2.108, the choice of expanding the density by using N -orbitals and integer occupation numbers is arbitrary. Janak70 proposed to dene the kinetic energy of the non-interacting electron as: TJ [] = min ni
i

(x) i

1 2

2 i i (x)d(x)

(2.155)

where ni is 0 ni 1. The number of functions in Janak derivation is arbitrary but subject to the constrain that i ni = N . The total density is then dened as: (r) =
i

ni

|(x)|2

(2.156)

and the corresponding total energy reads: E[] = TJ [] + Vee [] + = 1 2 ni

i

(r)Vext (r)dr (2.157)

2 i i (x)d(x)

(x) i

+ J[] + Exc [] +

(r)Vext (r)dr

We can minimize E[] with respect to i and ni . For a given set of ni , we have: 1 ni 2
2

+ ni Vef f (r) i = i i

(2.158)

The standard KS-like equation can be obtained considering the substitution: i = nii . The dierentiation of the energy E with respect to ni gives Eq 2.159 that is usually referred as the Janaks theorem: E 1 = ni 2 1 = 2 = i (x) i (x) i
2 i i (x)d(x) 2 i i (x)d(x)

+ +

J[] + Exc [] + (r) Vef f (r) (x)i (x)dr i

(r)Vext (r)dr

(r) ni

(2.159) 33

CHAPTER 2. THEORETICAL CONCEPTS From a practical point of view, the KS implementation is denitely easier because it just xes the occupation number of the rst N orbitals to 1 and the rest to 0, but several and important are the information that can be derived from Janak theorem such as the Electron Anity or the Ionization Potential in terms of orbitals eigenvalues.d

2.2.10

From Ground to Excited States

As we saw in the previous demonstration, the density provides information about the ground state density but in many works46, 38, 7173 the DFT was extended to describe the properties of excited states. These methods are still being developed and we can cite for example the TD-DFT or the CI. In our LFDFT approach, we can consider that we do a mini-CI on the MOs with dominant d-character.

In the LFDFT procedure, we use the Janaks theorem when we do the AOC calculation. In fact to satisfy the LFT requirement, we have to occupy all the MOs with dominant d-character with the same amount of electrons.

34

Chapter 3 LFDFT
The LFDFT method is a LFT based on DFT calculations. The procedure can be divided in three main parts. The rst part consists of a set of DFT calculations and the extraction of the necessary data from its outputs. The second part consists in a matlab program, which uses the Slater determinant energies (calculated by DFT) and the eigenvectors of the Kohn-Sham orbitals with dominant d-character (extracted from the AOC calculation), to calculate the ligand eld matrices and the Racahs parameters. The third and nal step corresponds to the utilization of a LF program which can determine, so far, the multiplet structure (discussed in Chapter 4) and the ESR parameters (subject of Chapter 6). The extension of the method to other properties is currently being developed by our group. Figure 3.1 gives an overview of all the procedure. This Chapter is dedicated to the description of the rst two parts: the DFT calculations and the LFDFT method while the analysis of the LF program is proposed in Chapter 4. For DFT, the calculations process is described through an example. Afterwards, the theory of LFDFT is presented, then the source code of the LFDFT program, and nally some applications are proposed to test the accuracy of the method. At the end of the chapter, some general conclusions about the LFDFT method are given.

3.1

DFT Calculations

All DFT calculations have been performed using the ADF program package.74 The DFT part of the LFDFT method is, in its turn, divided in three steps: (i) the rst consists either in a single point calculation (if we have the X-ray crystallographic data) or in a geometry optimization, (ii) the second step, called an AOC calculation, is necessary to build the Kohn-Sham orbital diagram in agreement with the LFT requirement, (i.e., the KS MO diagram of an AOC calculation comply with the spherical symmetry inherent to LFT), (iii) the last step, called Slater Determinant (SD) calculations, determine, within the DFT formalism, the energies of all SDs. An overview of these three steps is proposed through an example: the LFDFT method is applied to the [CoCl4 ]2 complex (the input les and some extracts of the output les are given). Before a review of the general settings used to perform the DFT calculations is presented.

35

CHAPTER 3. LFDFT

(chem.x)

DFT

Geometry optimization or Single point Calculation

AOC calculation

SD calculation

eigenvectors of MOs with dominant d-character

energies of all SDs (get.x)

LFDFT

GENER LN

4 matrices gdata, lfdata, ir, lsdata

LFDFT program

Racahs parameters & LF matrix

LF

LF program

Figure 3.1: Scheme of the LFDFT calculations procedure. We can see that the DFT part needs three steps, the LFDFT part two and the nal step consists in using a LF program.

ADF version Acronyms PW91 PBE 2003.01 RPBE revPBE Blyp 2000.01 PW91

Exchange PW91x75 PBEx76 RPBEx77 revPBEx78 Becke58 Becke58 PW91x75

Correlation PW91c75 PBEc76 PBEc76 PBEc76 LYP60, 79, 80 Perdew81 PW91c75

Table 3.1: Listed in the second column the acronyms of the GGA functional we use depending on the version of ADF. In the third and fourth column, names and references are given on the the exchange and the correlation part used.

36

3.1 DFT Calculations

3.1.1

General Settings

For each calculation we have to choose the basis set and the functional but, we can say that we use for almost all the calculations the same settings that are summarized here.a In general, the atoms are described through triple- STO basis sets plus one polarization function given in the program database (basis set IV: ADF release 2000, or TZP: ADF release 2003) and the frozen core approximation is used up to 3p for the TM and up to 1s for second row elements, 2p for third row elements and so on. The choice of the functional for the geometry optimization calculation is discussed in the Section 3.1.2. Within one LFDFT calculation, the functional for the AOC calculations and the SD calculation must be the same. In fact Figure 3.1 shows that data have to be extracted from these two calculations so the two calculations have to refer to the same approximation, i.e. the same functional. We use either the local density approximation (LDA) where exchange-correlation potential and energies have been computed according to the Vosko, Wilk, and Nusairs (VWN)57 parameterization of the electron gas data or the generalized gradient approximation (GGA) where exchange-correlation and energies have been computed according to the PW91 parametrisation (cf. Table 3.1). To test the sensitivity of our method to the functional, we, for some examples, use every functional listed in Table 3.1 (where acronyms are used to dene in one stroke the exchange and the correlation part of the functional). We can already underline that if we use the functionals: PW91, PBE, RPBE and revPBE, we get similar results. While the results obtained with the BLYP functional are worst. And for the LB9482 functional, as it is written in ADF User Guide:The energy expression underlying the LB94 functional is very inaccurate, so we test and we verify that this functional should not be used in a LFDFT treatment.

3.1.2

Geometry Determination

Two cases can be considered: we know the geometry of the compound from X-ray crystallographic data or not. In the rst case, we just have to run a single point calculation on the structure given by the X-ray crystallographic data to know how to specify the occupations for the AOC calculations. Otherwise, we have to do a geometry optimization and then, we choose LDA functional (cf. Fig. 3.2) if we have to deal with a small (a tenth of atoms) and negatively charged system because GGA functional overestimate the metal to ligand bond length83 but GGA functional could be used for cases like porphyrin. In fact, our experience is that for small and negatively charged compounds, geometry optimization with LDA functional gives a very close bond distance compared to experimental data. The Figure 3.2 corresponds to the input for a geometry optimization run. Whatever the case, the Kohn-Sham orbital diagram is examined to determine which are the molecular orbitals (MO) with a dominant d-character. This KS-MO diagram gures in the output le under the section List of all MOs, ordered by energy, with the most signicant SFO gross populations and Figure 3.4 shows a part of the KS-MO diagram for the [CoCl4 ]2 complex.
However, for each chapter or new calculations, a footnote like this one will summarize the settings: the frozen core approximation, the type of basis sets and the functional used. I will also specify if the geometry comes from either an optimization of geometry or X-ray crystallographic data.
a

37

CHAPTER 3. LFDFT
TITLE CoCl4 ATOMS cartesian Co 0.000000 0.000000 0.000000 Cl 1.316090 1.316090 1.316090 Cl -1.316090 -1.316090 1.316090 Cl -1.316090 1.316090 -1.316090 Cl 1.316090 -1.316090 -1.316090 END BASIS Co /atomicdata/TZP/Co.3p Cl /atomicdata/TZP/Cl.2p END XC LDA vwn END CHARGE -2 EPRINT ORBPOPER -1000 1000 END GEOMETRY OPTIM cartesian END END INPUT TITLE CoCl4 AOC ATOMS cartesian Co 0.000000 0.000000 0.000000 Cl 1.316090 1.316090 1.316090 Cl -1.316090 -1.316090 1.316090 Cl -1.316090 1.316090 -1.316090 Cl 1.316090 -1.316090 -1.316090 END BASIS Co /atomicdata/TZP/Co.3p Cl /atomicdata/TZP/Cl.2p END XC GGA PW91 END RESTRICTED CHARGE -2 OCCUPATIONS A1 4 T1 6 T2 18 4.2 E 4 2.8 END END INPUT

Figure 3.2: Example of the geometry optimization input for the [CoCl4 ]2 complex.
E(eV) 3.748 3.748 4.041 Occ 2.00 2.00 1.00 MO 2 E:1 2 E:2 4 T2:1 % 84.10% 15.77% 84.10% 15.77% 73.04% 22.49% 2.72% 1.47% 73.04% 22.49% 2.72% 1.47% 73.04% 22.49% 2.72% 1.47%

Figure 3.3: Example of the AOC calculation input for the [CoCl4 ]2 complex.
SFO (rst member) 1 D:z2 1 P:x 1 D:x2-y2 1 P:x 1 D:yz 1 P:x 1 P:y 1 P:x 1 D:xz 1 P:y 1 P:x 1 P:y 1 D:xy 1 P:z 1 P:x 1 P:z E(eV) -8.771 -8.688 -8.771 -8.688 -8.771 -8.688 -8.688 -0.498 -8.771 -8.688 -8.688 -0.498 -8.771 -8.688 -8.688 -0.498 Occ 1.40 1.67 1.40 1.67 1.40 1.67 1.67 0.00 1.40 1.67 1.67 0.00 1.40 1.67 1.67 0.00 Fragment 1 Co 2 Cl 1 Co 2 Cl 1 Co 2 Cl 2 Cl 1 Co 1 Co 2 Cl 2 Cl 1 Co 1 Co 2 Cl 2 Cl 1 Co

4.041

1.00

4 T2:2

4.041

1.00

4 T2:3

Figure 3.4: Part of the KS molecular orbital diagram (after removing MOs which are not involving the TM d-orbitals) of a geometry optimization output. This occupation corresponds to the ground state conguration.

38

3.1 DFT Calculations In Figure 3.4, the rst column corresponds to the eigenvalues of the MOs, the second to the electronic occupation, the third to the number and the label of the MO considered, the fourth and the fth contain information about the construction of the MO: the MO are build in ADF from SFO, the fourth column gives the square of the coecient and the fth the label of the SFO constituting the MO, the sixth gives the eigenvalues of the SFO, the seventh the electronic occupations of the SFO, nally the last column species from which atom the SFO comes from (number and label of the atom).

3.1.3

AOC

This step consists in a spin-restricted DFT calculation with an equal electronic occupation on each MO with dominant d-character. In our example the cobalt cation has a d7 conguration so we have to occupy each of the 5 MO with dominant d-character by 7/5 = 1.4 electrons. In a tetrahedral environment, like in our example, the crystal eld splitting has to be considered, it means 3 1.4 = 4.2 electrons on the t2 MO and 2 1.4 = 2.8 electrons on the e MO (cf. Fig. 3.3). As a result, the Kohn-Sham orbitals obtained are best suited for a ligand eld treatment (cf. Fig. 3.5) because they comply with the spherical symmetry inherent to ligand eld theory. This recipe is consistent with the prerequisites of the LF approach, where orbital relaxation is only taken into account at the level of averaging the electron density to provide proper LF orbitals, while all SD energies for latter LF treatment are calculated without SCF iteration.84 Comparing Fig. 3.4 & Fig. 3.5, we can underline the fact that the occupations are different (second column) and the eigenvectors (fourth column represents its square) and eigenvalues (rst column) are also dierent.

E(eV) Occ 3.396 1.40 3.396 1.40 3.913 1.40 3.913 1.40 3.913 1.40

MO 2 E:1 2 E:2 4 T2:1 4 T2:2 4 T2:3

% 80.92% 19.04% 80.92% 19.04% 71.73% 23.69% 2.89% 1.50% 71.73% 23.69% 2.89% 1.50% 71.73% 23.69% 2.89% 1.50%

SFO (rst member) 1 D:z2 1 P:x 1 D:x2-y2 1 P:x 1 D:yz 1 P:x 1 P:y 1 P:x 1 D:xz 1 P:y 1 P:x 1 P:y 1 D:xy 1 P:z 1 P:x 1 P:z

E(eV) -8.771 -8.688 -8.771 -8.688 -8.771 -8.688 -8.688 -0.498 -8.771 -8.688 -8.688 -0.498 -8.771 -8.688 -8.688 -0.498

Occ 1.40 1.67 1.40 1.67 1.40 1.67 1.67 0.00 1.40 1.67 1.67 0.00 1.40 1.67 1.67 0.00

Fragment 1 Co 2 Cl 1 Co 2 Cl 1 Co 2 Cl 2 Cl 1 Co 1 Co 2 Cl 2 Cl 1 Co 1 Co 2 Cl 2 Cl 1 Co

Figure 3.5: KS orbital diagram of the AOC calculation (after removing parts which are not involving the MO without dominant d-character). We can remark that the occupations are dierent than in Fig 3.4.

39

CHAPTER 3. LFDFT

3.1.4

Slater Determinant Calculation

Title Cocl4 SD calculation

n number of SD

2 or 8 3 or 7 45 120

4 or 6 5 210 252

Table 3.2: Number of SD originating from an open dn shell.

The SD calculation consists in a spinunrestricted calculation of all Slater determinants originating from the dn conguration (cf. Table 3.2). Figure 3.6 shows an input le for our example just for one single determinant: since the cobalt has a d7 conguration, the normal input has 120 SD. We use as we can see in the input (cf. Fig. 3.6), the AOC calculation as a fragment, in doing so, the KS-MO diagram is frozen and during the calculation no SCF procedure occurs. At this step the program changes only the occupations on the d-shell according to the SLATERDETERMINANTS keyword list in the input. The program chem.x, presented in Appendix B.1, is designed to create the input of SD calculation. The output consists then in a list of all the SD energies. It is important to make two remarks concerning this input le. Firstly, the congurations of some Slater determinants are not compatible with the real symmetry of the compound, as a consequence a lower symmetry must be imposed (Nosym in Fig 3.6). Secondly the SD calculation is an unrestricted calculation but the net total spin polarization value doesnt have to be precise for the CHARGE keyword because this value changes for each SD.

Atoms cartesian Co 0.00000 0.00000 0.00000 f=Av Cl 1.31609 1.31609 1.31609 f=Av Cl -1.31609 -1.31609 1.31609 f=Av Cl -1.31609 1.31609 -1.31609 f=Av Cl 1.31609 -1.31609 -1.31609 f=Av End Charge -2 Unrestricted Fragments Av /home/rauzy/cocl4 aoc.TAPE21 End XC GGA PW91 End Symmetry Nosym SLATERDETERMINANTS SD 1 A1 2 // 2 E:1 1 0 // 1 0 E:2 1 0 // 1 1 T1:1 1 // 1 T1:2 1 // 1 T1:3 1 // 1 T2:3 3 1 // 3 1 T2:2 3 1 // 3 1 T2:1 3 1 // 3 1 SUBEND END End Input

Figure 3.6: Example of a SD calculation input for [CoCl4 ]2 .

3.2

Extraction of Data

In LFDFT, we have to extract from the AOC DFT calculations the eigenvectors of the KS molecular orbitals with dominant d-character and get the list of SD energies from the SD calculations. 40

3.2 Extraction of Data

3.2.1

Eigenvectors and Eigenvalues of the Kohn-Sham Orbitals

In this section we explain how to extract eigenvectors and eigenvalues of the KS orbitals for a non relativistic calculation. The extraction of eigenvalues is not necessary for the LFDFT program but it is useful to compare them to the LF matrix after LFDFT treatment. 3.2.1.1 Eigenvectors of KS Orbitals

The determination of the MO eigenvectors is a key point in the LFDFT method. This process can not be easily programmed and we have to do it manually. When the eigenvectors are extracted, they are arranged in a 55 matrix format where the columns represent the MOs and the lines represent the d-AOs. The line should be arranged from m = l to m = l so for a d-shell from m = 2 to m = 2 which corresponds to the following orbital order: dxy , dyz , dz2 , dxz , dx2 y2 (order xed by the program gener ln). The order of the column is xed by the program chem.x and to be consistent with the order of the line, the rst column should correspond to the MO which is dominated by dxy , the second to the MO dominated by dyz and so on until m = 2 for d-shell (in this way we have always the bigger coecients on the diagonal). To determine the eigenvectors we have rst to note, in the List of all MOs section of the AOC calculation output, the numbers and the labels of the MOs with dominant d-character (for our example: 2 E:1, 2 E:2, 4 T2:1, 4 T2:2 and 4 T2:3, cf. Fig 3.5). Then in the B U I L D section of the AOC calculation output, we have to look for the subsections of each MO previously dened: Figure 3.7 corresponds to E:1 MO for our example.
=== E:1 === Nr. of SFOs : 8 Cartesian basis functions that participate in this irrep (total 38 36 37 61 59 60 84 82 83 107 105 106 15 10 13 21 16 19 27 22 25 44 42 43 67 65 66 90 88 89 113 111 112 47 45 46 70 68 69 93 91 92 116 114 115 50 48 49 73 71 72 96 94 95 119 117 118 56 51 54 79 74 77 102 97 100 125 120 123 53 55 52 76 78 75 99 101 98 122 124 121 SFO (index Fragment Generating indx incl.CFs) Occup Orb.Energy FragmentType 1 1.400 -0.322 au Co 13 ( -8.771 eV) 2 14 0.530 au Co ( 14.418 eV) 3 15 5.236 au Co ( 142.490 eV)

number = 81) :

Expansion in Fragment Orbitals Coe. Orbital on Fragment 1.00 1 D:z2 1 1.00 1.00 2 D:z2 3 D:z2 1 1

Figure 3.7: Part of the B U I L D section of the AOC calculation output for the MO E:1.

41

CHAPTER 3. LFDFT In the column index incl.CFs, we have to obtain the number corresponding to the line where the occupation is not null and where the orbital corresponds to a d-AO of the Cobalt cation (for our example: 13). Once we know the irrep numbers (2 for the E orbitals for our example) and the SFO index (13), under the E:1 part of the SFO MO coecients section of the AOC calculation output, which is a matrix where the columns represent the number of the MOs and the lines represent the SFO index (cf. Figure 3.8), we take the value corresponding to the line with the label 13 and the column with the label 2.

=== E:1 === MOs expanded in CFs+SFOs ======================== MOs : 1 2 3 4 13 14 15 16 17 18 19 20 0.3813 -0.0343 0.0016 -0.8563 0.0417 -0.0013 0.0146 -0.0187 -0.9260 0.0396 -0.0009 -0.4882 0.0537 -0.0024 -0.0341 0.0303 0.0859 -0.0683 -0.0037 -0.0913 -0.9665 -0.0127 0.0720 0.0800 0.1467 0.4135 0.0041 -0.1476 0.1099 -0.0017 -0.5129 0.6107

5 -0.0288 -0.0173 0.0014 0.0027 0.1487 -0.0089 0.7460 0.6596

6 -0.1414 1.1269 -0.0163 -0.3832 0.2568 0.0506 0.5180 -0.5025

7 0.0499 -0.1916 0.0187 0.0856 -0.0701 1.0138 -0.0405 0.0921

8 0.0609 -0.1877 -1.0282 0.1910 -0.1372 0.0665 -0.1238 0.1297

Figure 3.8: Part of the SFO MO coecients section of the AOC calculations output.

We do the same for the ve MOs, then we can build the matrix U (in a matlab format) which represents the eigenvectors of the MO with dominant d-character respecting the order previously dened (Eq. 3.1). U = [ 0.8863 0 0 0 0 0 0.8863 0 0 0 0 0 0.9260 0 0 0 0 0 0.8863 0 0 0 0 0 0.9260 ; ; ; ; ];

(3.1)

We can underline that for highly symmetric compounds the matrix is diagonal but if we go down in symmetry then we generally have a mixing between MOs belonging to the same irrep, in this case the matrix is no more diagonal. With C1 symmetry, we can have each matrix elements dierent than zero depending on the mixing.

3.2.1.2

Eigenvalues of KS Orbitals

It is relevant but not necessary to get the eigenvalues of MOs from the AOC calculation output and compare them to the LF matrix. The eigenvalues correspond to the rst column of the Figure 3.5. 42

3.3 LFDFT: Theory

3.2.2

Energies of Slater Determinants

In the output of Slater determinant calculations gures the list of all the SD energies. It is possible to extract them directly in a vector format with the script based on awk language given in Appendix B.2.1. This script acts on the logle, summary of the output for an ADF calculation, and can be modied to list either the LDA or the GGA SD energies. The execution of this script on the logle of a SD calculation creates a le esd which can be directly loaded by the LFDFT program.

3.3

LFDFT: Theory

As we know how to extract all the data necessary to run the LFDFT program. In this section, we describe the LFDFT method and then give the source code of the LFDFT program.

3.3.1

Introduction

If we consider an octahedral complex, the LF matrix is reduced to one parameter: 10Dq. We thus have for each SD energy the simple linear expression in terms of B, C and 10Dq (Eq. 3.2):
d E(SD ) = E(det|di(,1) i(,1) di(,2) i(,2) . . . di(,n) i(,n) |)

=
i

di |hLF |di +
i<j

(Jij Kij i j )

(3.2)

3m 2n 10Dq + B + C + E0 5

The single determinants SDd are labeled with the subscript = 1, . . ., 10 and with the n superscript d to refer to pure d-spinorbitals. The values of m and n specify the elecn m tronic conguration t2g eg while the and are coecients obtained after substituting standard expressions for the Coulomb Jij and exchange Kij integrals in terms of d-only orbitals di and spin functions i (cf. Table 2.1). Having obtained energy expressions of each SDd , we can determine the ligand eld parameters 10Dq, B and C plus a shift parameter E0 which is due to the dierent gauge origin between the DFT and the LF approach. We obtain the Equation 3.3 in which X stores the ligand eld parameters we should determine plus the shift energy E0 , E is an array composed by the SD energies calculated by DFT and A is a matrix which stores the coecients of the Equation 3.2 (cf. Table 3.4): E = AX (3.3) Eq. 3.3 represents a system of linear equations which is overestimated. We can solve it by least square t (Eq. 3.4) to obtain the LF parameters: X = (AT A)1 AT E (3.4)

Then we are able to compare SD energies from DFT with those calculated using the LF parameter values. The determination of the mean square deviation allows us to estimate the accuracy of the t. 43

CHAPTER 3. LFDFT

3.3.2

Theory

On one side from the eective Hamiltonian theory,85, 86 we can approximate the hab LF matrix by using the ve KS orbitals with dominant d-character. On the other side, according to the work of Ziegler et al.,10 it is possible to use the single determinant energies obtained by DFT in a LF scheme. From the rst remark, we can extract as it is shown in Section 3.2.1.1, the eigenvectors of the MOs, which correspond to the ve MOs orbitals with dominant d-character, and store them in the matrix U . Then we introduce the overlap matrix S (Eq. 3.5): S = UUT (3.5)

Since U is in general not orthogonal, we use the Lwdins symmetric orthogonalisation o scheme to obtain an equivalent set of orthogonal eigenvectors (C): C = S 1/2 U (3.6)

We identify now these vectors as the eigenfunctions of the eective LF Hamiltonian hef f LF we seek as:
5

i =
i=1

ci d

(3.7)

Thus, the tting procedure described in the previous section enable us to estimate hii = i |hef f |i and hence the full representation matrix of hef f as: LF LF
5

h =

d |hef f |d LF

=
i=1

ci hii ci

(3.8)

The energy of a single determinant becomes thus:

E(SDk ) = E(det|i(k,1) i(k,1) i(k,2) i(k,2) . . . i(k,n) i(k,n) |)

=
i

i |hLF |i +
i<j

(Jij Kij i j )

(3.9)

where SD is composed of the spinorbitals. k In order to calculate the electrostatic contribution (2nd term in Eq. 3.9), it is useful to consider basis transformation. So rst C is transformed from the orbital basis to the spin-orbital basis: C =C (3.10) and then we consider the following transformation from the basis of SD to the one of k SDd . Using basic linear algebra, we get:
|SDk = d Tk |SD

(3.11)

where Tk = det|ci(k,1:n),j(,1:n) | i.e. the determinant of a n n sub-matrix of C: ci(k,1),j(,1) ci(k,1),j(,2) ci(k,2),j(,1) ci(k,2),j(,2) ... ... ci(k,n),j(,1) ci(k,n),j(,2) 44 . . . ci(k,1),j(,n) . . . ci(k,2),j(,n) ... ... . . . ci(k,n),j(,n)

(3.12)

3.4 LFDFT: Program with the indices of the spinorbitals i(k,j) i(k,j) and di(,j) i(,j) respectively. Note that these indices are in fact a two-dimensional array of (number of SD) (number of electrons or holes) integers. Finally the energy of a SD in Eq. 3.9 can be rewritten as:

Ek = E(SDk ) = i

i |hLF |i +
,

d Tk Tk SD |

1 d |SD r12

(3.13)

where G =1/r12 i.e. the electrostatic repulsion of all electron pairs in the LF manifold. The matrix elements are obtained using the Slaters rules (cf. Section 2.1.3) and the resulting electrostatic two-electron integrals ab|cd in terms of Racahs parameters. Thus the nal equation to estimate the ligand eld matrix hab and the B and C Rachahs parameters from the DFT energies, Ek of all the SD within the LF manifold, has the same shape than Eq 3.3 where X = (h11 , . . . , h55 , B, C) and A are calculated as indicated above (Eq. 3.13). The Matlab script, the LFDFT program, to carry out these calculations for all dn congurations is proposed in the next section.

3.4

LFDFT: Program

Now, the source code of the LFDFT program is presented in Figure 3.9 with comments referring to previous equations. We have to specify as input U , ne and esd which are, respectively, the eigenvectors of MO with dominant d character (structured as Eq 3.1, the order of d-orbitals being from m = 2 to m = 2), the number of electron on the d-shell and the list of single determinant energy. One way to do it is to put U and ne in a le matrice.m and change the name of the le produce by the script get.x (cf. Appendix B.2.1) from esd to esd.mat (the energies should be expressed in eV ). In the second line of the program, the rst one cleaning the memory and the screen, this two les are loaded in the same time than data.mat generated by the Gener ln program (presented in section 4.2.3). The data.mat le contains 4 matrices ir, gdata, lfdata and lsdata and we can remark that for the LFDFT program, only the two matrices ir and gdata are used. A plot function at the end of the program produces for our example the Figure 3.10 where the x-axis refers to SD and the y-axis corresponds to the energy: for each SD, there are two marks: one referring to the energy calculated by DFT () and the second to the LFDFT values ((o), SD energies calculated using the B, C and 10Dq parameters previously tted). If for each SD we compare the two marks, we can then appreciate the accuracy of the t. This kind of picture plus the calculation of the mean square deviation are the two possibilities we haveto appreciate the accuracy of the t.

45

CHAPTER 3. LFDFT
clc, clear all, matrice, load data, load esd if (ne>5), ne=10-ne, end kphi=ir;[nsd nf]=size(kphi); kkp=0; for k=1:nsd for kp=1:k kkp=kkp+1; gb(k,kp)=gdata(kkp,2); gc(k,kp)=gdata(kkp,3); if(k =kp) gb(kp,k)=gdata(kkp,2); gc(kp,k)=gdata(kkp,3); end end end % Lwdin orthogonalisation: Eqs 3.5 & 3.6 o S = U*U; Shalf=S^(-0.5); S = Shalf*U; % from orbital basis to spin-orbital basis: Eq 3.10 e=[1 0; 0 1]; c=[S(1,1)*e S(1,2)*e S(1,3)*e S(1,4)*e S(1,5)*e; S(2,1)*e S(2,2)*e S(2,3)*e S(2,4)*e S(2,5)*e; S(3,1)*e S(3,2)*e S(3,3)*e S(3,4)*e S(3,5)*e; S(4,1)*e S(4,2)*e S(4,3)*e S(4,4)*e S(4,5)*e; S(5,1)*e S(5,2)*e S(5,3)*e S(5,4)*e S(5,5)*e]; c=c; % get t: |i = Ti, | : Eq 3.12 t=zeros(nsd,nsd); for i=1:nsd for mu=1:nsd t(i,mu)=det(c(kphi(i,:),kphi(mu,:))); end end % Setup LF-model ivar=[1 1 2 2 3 3 4 4 5 5]; a=zeros(nsd,7); for i=1:nsd for j=1:nf k=ivar(ir(i,j)); a(i,k)=a(i,k)+1; end end % electrostatic part a(:,6)=diag(t*gb*t); a(:,7)=diag(t*gc*t); % Least square fit: Eq 3.4 lf=inv(a*a)*a*esd; % from eV to cm-1 and calculation of Racahs parameter B and C lf=lf*8065; B=lf(6); C=lf(7); fprintf(Racah B= %8.3f cm-1 n,B), fprintf(Racah C= %8.3f cm-1 % predicted energies of SD f=a*lf; % Standard deviation: Eq B.2.4 diff=esd*8065-f; mean=(sqrt(sum(diff.^2))/nsd); fprintf(Standard deviation: %8.3f cm-1 n,mean) % plot function to make figure as the Fig. 3.10 X=1:1:nsd; f=f/8065; plot(X,esd,*,X,f,o) % ligandfield matrix in cm-1 h=S*diag(lf(1:5))*S

n,C)

Figure 3.9: Source code of the LFDFT program.

46

3.5 LFDFT: Applications

Figure 3.10: Comparison between SD energies from DFT calculation () and from LFDFT calculation (o). The x-axis refers to SD and the y-axis corresponds to the energy in eV. In fact the 0 eV energy corresponds to the reference which is the energy of the AOC calculation.

3.5

LFDFT: Applications

In this section, the LFDFT method is applied to some compounds whose UV-Vis spectra can be found in the literature. During the course of this Ph.D., many systems have been studied and the results obtained have been published in two articles.32, 35 Here, only a selection of these results is presented.

3.5.1

Cubic d2 Systems

The main idea behind our approach is presented using a simple example: CrO4 , a d2 4 (CrIV ) TM in a tetrahedral coordination environment.c We solve the Equation 3.4 where E is an array of the 45 SD energies from the DFT calculation (originating from d2 open shell system) and A stores the coecients of the 45 linear equations corresponding to Eq 3.2.
The AOC calculation was performed with e4/5 t2 using LDA VWN functional (for AOC and SD calculations) and TZP basis sets (IV for ADF v2000.01) with a frozen core approximation up to 3p for Co and up to 1s for O.
c 6/5

47

CHAPTER 3. LFDFT

The Racahs parameters B, C and 10Dq are then Ropt Rexp Exp b determined by least square t. We did the maR = dCrO 1.868 1.7787 nipulation for a geometry optimized using LDA B 436 427 555 functional and for the experimental geometry.87 C 2250 2274 2331 In Table 3.3, we compare the results with those 10Dq 7315 8847 8950 tted from a UV-Vis spectra. We can see that the (10Dq)M O 7363 8831 results obtained with experimental bond length StDev 290 314 are really close with the results tted from the spectra, in spite of the fact that B values are Table 3.3: Comparison between the underestimated by the LFDFT procedure. From Racahs parameters obtained from the list of single determinant energies (E), we can LFDFT calculation for an optimized Cr-O bond length (LDA) and the experunderline that only 15 SD energies are dierent at imental Cr-O bond length with values the DFT level (cf. Table 3.4), however we take tted from a LF spectra for CrO4 4 all 45 equations into account. In doing so, we (values in cm1 ). provide a better statistical weight for those SD which have the same energy. In Table 3.4, the 15 non redundant SD energies are given as well as the energy expression in the LF model (the energy of the |+ + | ground state is taken to zero and the other energies are expressed relatively to this one). A comparison between the SD energies calculated by DFT and the SD energies calculated using Equation 3.2 by replacing B, C and 10Dq with the values determined by LFDFT shows a nice t with a rather small standard deviation (0.039 eV). This result illustrates the consistency between the DFT formalism and the LF parameterization. Moreover, one can make two remarks about the 10Dq: - the value of 10Dq for the experimental geometry is really close to the one t from the UV-Vis spectra as we know that 10Dq is sensible to the metal to ligand bond length. So if the experimental structure is known, one should use it to run LFDFT calculation, - the tted 10Dq value is very close to the 10DqMO .

3.5.2

Cubic dn Systems

The procedure described above has been extended to dn TM complexes (the number of SD resulting from a dn conguration is stored in Table 3.2) and the ligand eld parameters 10Dq, B, C have been obtained as for the previous cubic d2 example. The Table 3.5 is a summary of the results published in the article of the Structure and Bonding volume dedicated to Jrgensen.32 In the same time, we compared the values obtained by LFDFT to the values tted directly to experimental LF transition energies (which we call LFT). We can underline, that when we use an experimental bond length, the results are very close to the values tted from the experimental data as we know that 10Dq depends strongly on the metal to ligand bond length. The calculations were performed using LDA functional (VWN for AOC and SD calculations) and TZP basis sets with the frozen core approximation; up to 2p for TM, 1s for C, O, N and F, 2p for Cl, and 3d for Br.

The abbreviation Exp means values tted from the LF spectra.

48

3.5 LFDFT: Applications

Table 3.4: Energy expressions for the DFT distinguishable Slater determinants of a d2 conguration in a cubic ligand eld and their DFT energies for the tetrahedral CrO4 4 complex. The energy expressions of each determinant within the conventional LFT model (B, C, 10Dq parameterization) are also included.

SD |+ | |+ | |+ + | |+ | |+ + | |+ | |+ + | |+ | |+ + | |+ | |+ + | |+ | | + | | + + | | + |

DFT Energies CrO4 4 1.571 1.557 0 0.379 1.564 1.905 1.154 1.506 1.288 1.633 1.690 2.031 3.573 2.471 2.811

LF Model (B, C, 10Dq) 12B + 3C 12B + 3C 0 4B + C 9B + 10Dq 10B + C + 10Dq 10Dq 4B + C + 10Dq 3B + 10Dq 6B + C + 10Dq 12B + 10Dq 12B + C + 10Dq 12B + 3C + 20Dq 3B + 20Dq 6B + C + 20Dq

Complex CrO4 Ropt 4 d2 Rexp LFT 3 MnO4 Ropt d2 Rexp LFT FeO2 Ropt 4 2 d Rexp LFT CrCl4 Ropt 2 d LFT MnCl4 Ropt d5 LFT

B 436 427 555 347 347 430 242 242 375 355 376 548 516

C 2250 2274 2331 1928 1936 2600 1637 1645 1388 1903 1579 2339 3363

10Dq 7315 8847 8950 9831 10872 10515 11259 11952 12938 7008 7250 3298 2661

Complex d CrCl3 6 d3 CrBr3 6 d3 Cr(CN)3 6 d3 Mn(CN)3 6 d4 Co(CN)3 6 d6 Fe(CN)4 6 d6 CoCl4 6 d7 CoBr4 6 d7 CrF3 6 3 Ropt LFT Ropt LFT Ropt LFT Ropt LFT Ropt LFT Ropt LFT Ropt LFT Ropt Rexp LFT Ropt Rexp LFT

Table 3.5: Values of B, C and 10Dq for various tetrahedral (left) and octahedral (right) complexes, deduced from LFDFT energies (Ropt means LFDFT procedure using an LDA optimized structure and Rexp means LFDFT procedure using experimental structure) and compared to values obtained from experimental ligand eld spectra (LFT).

B 605 734 484 550 427 543 452 554 444 630 387 456 427 411 573 548 795 492 460 808

C 2694 3482 2403 3450 2395 3296 1919 2559 2361 2598 2573 3184 2420 3566 2540 2436 3108 2444 2371 3159

10Dq 13598 15297 10911 12800 9816 12400 30760 26595 34085 36900 37180 34944 35421 33678 3952 7436 7206 3782 6387 6384

49

CHAPTER 3. LFDFT

3.5.3

Application to Model the Nephelauxetic Eect

The nephelauxetic eect is modeled using LFDFT on complexes with Cr4+ , Mn5+ and Fe6+ cations which have a d2 conguration. To evaluate the nephelauxetic eect, we have in a rst time to calculate the B0 and C0 parameters (Racahs parameters for the freeion) for each free-ion in a gas phase and the results are stored in the Table 3.6. In this table we compare these values to the ones tted from atomic spectra (LFT). In a second time, we determine, using the LFDFT method, Racahs parameters for the oxide of these cations, i.e. CrO4 , MnO3 , FeO2 and for the halogen series with the cromium cation, 4 4 4 i.e. [CrF4 ], [CrCl4 ], [CrBr4 ] and [CrI4 ]. The B and C parameters for these complexes are given in Table 3.7. LFDFT 927 4508 1044 5194 1153 5799 LFT 1015 4263 1160 4930 1300 5525 Complex B C 4 CrO4 436[555] 2250[2331] 3 MnO4 347[430] 1928[2600] FeO2 242[375] 1637[1388] 4 CrF4 476 2452 CrCl4 355 1903 CrBr4 347 1855 CrI4 323 1758
Table 3.7: Racahs parameters values (in cm1 ) for tetrahedral d2 (MOz : M(z): 4 CrIV (-4) ; MnV (-3); FeV I (-2); CrX4 , X = F , Cl , Br , I ) calculated using a LDA functional. Values for the oxo-anions, deduced from a direct t to LF spectra are included in square brackets.

Cr4+ Mn5+ Fe6+

B0 C0 B0 C0 B0 C0

Table 3.6: B and C parameters (in cm1 ) for the Cr4+ , Mn5+ and Fe6+ cations derived from LFDFT (using LDA VWN functional and TZP basis sets with frozen core approximation up to 3p) calculations versus values deduced from a t to atomic spectra (LFT).

The DFT calculations yield values of B which are by 10% smaller and values of C which are by 5 to 12% larger than experiment. In Fig 3.11 we plot the B/Bo and C/Co ratios for the series of tetrahedral d2 complexes. We can remark that B is more reduced than C which is in agreement with the study of Ferguson and Wood88 on Co and Cr complexes. Moreover they claim that the parameters B and C probe inuences from dierent regions in a complex; while B mirrors outer properties and is therefore more aected by covalency, C is a more inner parameter and thus less inuenced by the TM-ligand bond. We fully conrm this statement here. The nephelauxetic reduction increases with increasing TM oxidation state from CrO4 to MnO3 to FeO2 and with increasing covalency from F 4 4 4 to Cl to Br to I, as it is seen from the series of compounds CrX4 (X = F, Cl, Br, I). It is interesting to note that the F and O in CrO4 and CrF4 are characterized by similar 4 reduction factors, oxygen being slightly more covalent than F. Keeping in mind that the B is more reduced than C in a complex, the ratio B/C, which is usually assumed to be the same as that for the free ion (Co /Bo = 4.2 0.2), might deviate strongly from it; it is found to increase from CrO4 (C/B = 5.16) to MnO3 (5.55) to FeO2 (6.76), i.e. 4 4 4 with increasing covalency. We should note, that is this assumption which has led to a denition of variable B values B33 , B35 , B55 .36

50

Reduction factor

Figure 3.11: The nephelauxetic reduction factors for B/Bo and C/Co for the d2 tetrahedral MOz : 4 M(z) : CrIV (-4); MnV (-3); FeV I (-2); CrX4 , X = F , Cl , Br , I .

CrF4 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20

3.5.4

The Eect of Covalency on 10Dq

We have performed a series of calculations of the (10Dq)CF T values resulting to Cr3+ surrounded octahedrally by six (-1) point charges. (10Dq)CF T values are plotted against the Cr3+ -point charge distance in Fig. 3.12 along with the values of (10Dq)LF DF T that result from LDA geometry optimized CrF3 , CrCl3 and CrBr3 complexes. Comparing 6 6 6 the values of (10Dq)CF T corresponding to the Cr-F, Cr-Cl, Cr-Br distances (1.933, 2.419 and 2.588 ) 0.321, 0.109 and 0.078 eV with the (10Dq)LF DF T 1.727, 1.336 and 1.193 eV A respectively. We conclude that crystal eld eect do not exceed 18% from the values of (10Dq)LF DF T , the latter values being in good agreement with experiment.

Figure 3.12: 10Dq values 10Dq (eV) (in eV) versus the metalligand distance in the ionic 2.0 (crystal eld) approximation for Cr3+ cation sur1.5 rounded octahedrally by six (-1) point charges at different distances (full line). 1.0 DFT calculations of 10Dq for the LDA optimized geometries of CrX3 (X = F, 0.5 6 Cl, Br) are also plotted ().
0.0 1.2

1.4

The next step is now to generalize the tting procedure to treat systems with symmetry lower than cubic or even without any symmetry (C1 ). 51

3.5 LFDFT: Applications
CrO4 4 B/B0 C/C0 CrBr4 MnO3 4 CrCl4 CrI4 FeO2 4
CrF3 6 * CrCl3 6 * * CrBr3 6 1.6 1.8 2.0 2.2 2.4 2.6 RM L

CHAPTER 3. LFDFT

3.5.5

Applications to a Low Symmetrical Compound

As an application of a complex with low symmetry, we study a Cr(NH3 )3+ chromophore 6 which has a Cs symmetry. The GGA optimized structure with RCrN = 2.166 to 2.284 yields a value of 10Dq resulting from averaging over the Cs split components of t2g A (a+a+a) and eg (a+a) of 19082 cm1 , in reasonable agreement with the experiment (21550 cm1 ).36 An electrostatic calculation taking six (-3) point charges for N and eighteen (+1) for H with coordinates from the Cr(NH3 )3+ geometry optimization yields a 6 (10Dq)CF T value of 0.435 eV, again reproducing only 18% of the total (10Dq)LF DF T value. We conclude again, that the bulk of the values 10Dq is due to covalency, in agreement with Jrgensens original suggestion.36

3.6

General Conclusions

After the discussion of some results obtained by the LFDFT method, we can make some general comments on the method. The Figure 3.10 conrms that the results obtained by least square tting are accurate enough to be used because the dierence of energies for each single determinant is small. In fact, the standard deviations (cf. Appendix B.2.4) between DFT-SD energies and their LFDFT values for the examples considered were calculated between 0.016 and 0.124 eV, i.e. between 130 and 1000 cm1 which are in the order of the DFT calculation precision (0.2 eV ). From Table 3.3d , we can remark that 10Dq is almost equal to the dierence |t2 e | referred as 10DqM O . From this statement and the eective Hamiltonian theory,85, 86 we justify the construction of hab LF matrix using the ve KS orbitals with dominant d-character. Up to now we also remark, with the exception of CN complexes where 10Dq (LFDFT) values are higher than experimental, that LDA and GGA functionals are accurate enough in calculating ligand eld matrix elements (hab ) not only for cubic but also for complexes with lower symmetries. This is not the case for B and C. In Table 3.5 we collected LFDFT (LDA) values of B and C of dierent complexes. For the sake of comparison we also list values of B and C deduced using a t to LF transitions from experiment. When comparing the two sets of data one should be cautious, because experimental uncertainties prevent an accurate t of B and C in many cases. In spite of this, we note that the LFDFT B and C values are systematically lower than experimental ones.

This remark is valid for all the compounds treated by LFDFT in this thesis.

52

Chapter 4 Multiplet Structure

Generally, it is more useful to compare multiplet transition energies with experimental data than to compare ligand eld (10Dq) and Racahs parameters to those tted from UV-Vis spectra. Figure 3.1 shows that the last step of the LFDFT method consists in the execution of a LF program to calculate multiplets energies. Two programs are at our disposal: a matlab program (the source code is given in section 4.2.1: a modied version to calculate magnetic properties is presented in Appendix E.2) and the AOMX program.89 As in the previous chapter we have shown how to calculate Racahs parameters and the (2l +1)(2l +1) hab matrix; these parameters can be given as input for the two programs. After execution, the programs list the energies of all multiplets.a In a rst time, we will give details about the theory, next the LF matlab program is proposed with the needed Gener ln program and also a brief introduction to AOMX is presented. At the end, applications are proposed to test the theory.

4.1

Theory of Multiplet

There are two ways to calculate the energies of multiplet if we know the energies of SD or the LF and Racahs parameters. One is to express the wavefunction of multiplets arising from a given conguration in a linear combination of SD and, the other one, is to use an Hamiltonian and the ligand eld parameters. Thus, a presentation of the two methods is given but all calculations done in this thesis are based on the second one.

4.1.1

Multiplet Energy as a Function of ESD

Following,8, 9 the energy of a multiplet can be expanded in rst order as a weighted sum of single determinant energies. And according to the work of Ziegler et al.,10 the energy of a single determinant can be replaced by the corresponding energy obtained by the DFT. In order to illustrate the method, as a rst example, we consider the singlet and triplet energies arising from an a1 b1 conguration. While spin-unrestricted calculations with two up (+) spins, |a+ b+ |, yields directly the energy of the S = 1, Ms = 1 state, a calculation with one spin-up (+) and one spin-down (), |a+ b | (or equivalently |a b+ |) gives the
The output of AOMX is more complete than the matlab program one because it species the label of the multiplets.
a

53

CHAPTER 4. MULTIPLET STRUCTURE average energy between the S = 0 and S = 1 with Ms = 0 for both states: E(|a+ b+ |) = E(ab 3 B, Ms = 1) E(|a+ b |) = 1 E(ab 3 B, Ms = 0) + E(ab 1 B) 2 (4.1) (4.2)

From these equations the energy of ES=0 (Ms = 0) can be extracted as the weighted sum: E(ab 1 B) = 2E(|a+ b |) E(|a+ b+ |) (4.3)

The situation becomes more involved when going to TM complex with highly degenerate orbitals. However we fully exploit the symmetry in order to simplify the relation between the multiplet splitting and SD energies. In general, we can write the multiplet wavefunction as: i = |m Sms > (4.4) where is the label of the irreducible representation of the space part of the wavefunction, m refers to its component in case of degeneracy, S is the spin part of the wavefunction with component ms in case of spin multiplicity larger than 1. The relation between multiplet 1st order energies, E(k ) and the energies of some symmetry independent (nonredundant) SD E( ) is given by Eq. 4.5, where the coecient Fk are the symmetry dependent weights: E(k ) =
k

Fk E( )

(4.5)

In two articles published by Daul et al.,11, 90 the full description of the method is given.

4.1.2

Multiplet Energy as Solution of H

Another way to determine the multiplet structure is to consider that the multiplet wavefunctions, i , are solution of an Hamiltonian H : H =
i

fi +
i j>i

1 + rij

i li si
i

(4.6)

where fi , 1/rij are respectively the one-electron and the two-electrons operators. This method is the one implemented in the LF matlab program and the AOMX program: the only dierence between them is that the one-electron operator is not dened in the same manner. When we introduce LFT (cf. section 2.1.3), we show how to determine the matrices equivalent operator by using the Slaters rules and next, how to parameterize the two-electrons operator in function of Racahs parameters. The last term to consider is the spin orbit coupling contribution and since our study is limited to d-electrons (l = 2 and s = 1/2), the spin orbit coupling matrix element can be easily obtained. So from the LFDFT program, we know the B and C Racahs parameters and the LF matrix and, if in a rst time, we consider (the spin-orbit constant) equal to zero, we can determine the eigenvalues of the Hamiltonian. 54

4.2 LF Programs

4.2
4.2.1

LF Programs
Matlab Program

The output of the LF matlab program is less elaborated than the AOMX one because it does not give information about the label of the multiplet term but the big advantage is that it is very short and easily modiable. We have just to enter the B and C Racahs parameters in r, the ligand eld matrix elements in lf par where the columns and the lines are ordered from m = 2 to m = 2 and dierent than zero if we want to consider spin-orbit coupling. This program like the LFDFT program loads the le data.mat, generated by the program Gener ln which is given in the Section 4.2.3, which contains the gdata matrix (2 electrons electrostatic matrix elements in the basis of microstates), the lfdata matrix (1 electron ligand eld matrix elements in the basis of microstates) and the lsdata matrix (spin and angular momentum 1-electron matrix elements in the basis of microstates). The Figure 4.1 consists in the source code of the program. We can see that the multiplet energies are determined by calculating the eigenvalues of the Hamiltonian (Eq 4.6), subtracting the lower one and ordering them. The number of time the energy appears is equal to (2S + 1) m where (2S + 1) is the spin multiplicity and m the number of time the level is degenerate. At the end, the multiplet energies are printed. All the inputs: lf par (ligand eld matrix in a vector form), r (B and C Racah parameters) and (the spin-orbit constant) has to be entered in kK. To determine , two options can be considered: we use the value given for free ion in the Griths book or we determine the value by DFT using a Relativstic ZORA calculation on the free ion. Then we reduce the obtained value by the orbital reduction factor, k, which is dened by: 1 k= 2l + 1
2l+1 2l+1

(U (i, ))2
i=1 =1

(4.7)

where U (i, ) are the matrix elements of Eq 3.1 and l = 2 for open d-shell TM. The matlab program does not present any diculties and the more interesting part is contained in the Gener ln program.

4.2.2

AOMX

The last version of the AOMX program89 reads the ligand eld matrix so the inputs are the same than the ones used in the previous matlab program. If we use an older version, we need to derive AOM parameters from the ligand eld matrix obtained by the LFDFT program. A set of 1, 3 or 5 AOM parameters e can be precised: e , es , ec , es and ec and one has to refer to the AOMX manual to know how to determine them. For the interelectronic repulsion, the Racahs parameters can be enter directly.

55

CHAPTER 4. MULTIPLET STRUCTURE load data % Input of Racahs parameter: r=[A B C] r=[0; B; C]; % Input of LF mat. el. <di|LF|dj> lfpar=[ h(1,1) ; h(2,1) ; h(2,2) ; h(3,1) ; h(3,2) ; h(3,3) ; h(4,1) ; h(4,2) ; h(4,3) ; h(4,4) ; h(5,1) ; h(5,2) ; h(5,3) ; h(5,4) ; h(5,5)]; % Spin-orbit coupling constant zeta=0 % ge=2.0023; [nsd ne]=size(ir); ij=0; % h = er + lf + so: Eq 4.6 for i=1:nsd for j=1:i ij=ij+1; h(i,j)=gdata(ij,:)*r+lfdata(ij,:)*lfpar+lsdata(ij,7)*zeta; h(j,i)=conj(h(i,j)); end end % get Eigenvalues(e) and Eigenvectors(c) of h [c,e]=eig(h); % sort eigenvectors and create an array of number: ie [e,ie]=sort(real(diag(e))); e(:)=e(:)-e(1); c=c(:,ie); % find the multiplicity of one energy i0=1; i1=0; for i=2:nsd if abs(e(i)-e(i-1))>0.001 i1=i1+1; w(i1)=e(i-1); mul(i1)=i-i0; i0=i; end end fprintf(----------------------------- n) fprintf((2S+1)*M(gamma) E n) fprintf(----------------------------- n) for i=1:i1 fprintf( %3i %12.3f n,mul(i),w(i)) end fprintf(----------------------------- n)
Figure 4.1: Source code of the LF matlab program.

56

4.2 LF Programs

4.2.3

Program Gener ln

In the two matlab programs presented, the LFDFT and the LF one, a le data.mat is loaded, it contains four matrices which are generated by the Gener ln matlab program. For this program, we just must specify the type of shell: lq should be equal to 2 for d-electrons, and the number of electrons on this shell: ne. It produces the ir, gdata, lfdata and lsdata matrices for the given conguration. The script is short, the source code gures in Figure 4.2, and the essential parts are contained in the called functions. From the Figure 4.3, two parts can be distinguished in the program: rstly, the two functions GET2EI4A and GETLS are calculating matrix elements in the basis of spin orbitals, and in a second part, the three functions ZAB, GAB and LFAB correspond to the implementation of the Slaters rule to determine the matrix elements of the one- and two-electron operator in the basis of -states. All the functions used by this program are given in Appendix C. clear all clc global iabcd vabcd lq lx ly lz sx sy sz ls % Type of shell lq: s(0), p(1), d(2) , f(3) and number of electrons: ne lq=[2]; ne=3; % Generate single determinants or microstates for lq^ne ir=GENERSD(ne,1,2*sum(lq+lq+1)); % Get <ab|cd> [vabcd,iabcd]=GET2EI4A(lq); % Get l, s and l*s 1-e matrices (s=1/2) [lx,ly,lz,sx,sy,sz,ls]=GETLS(lq,1/2); % nsd=length(ir); ij=0; for i=1:nsd for j=1:i ij=ij+1; lfdata(ij,:)=LFAB(ir(i,:),ir(j,:)); lsdata(ij,:)=ZAB(ir(i,:),ir(j,:)); gdata(ij,:)=GAB(ne,ir(i,:),ir(j,:)); end end gdata racah(:,1)= gdata(:,1); gdata racah(:,2)= (49.0/5.0)*gdata(:,2); gdata racah(:,3)= (7.0/5.0) *(gdata(:,1)+gdata(:,2)+gdata(:,3)); gdata(:,:)=gdata racah(:,:); save data ir lfdata lsdata gdata
Figure 4.2: Source code of the Gener ln program.

57

CHAPTER 4. MULTIPLET STRUCTURE

GENERSD SD order (lq, ne > ir)

GET2EI4A Non-redundant two electrons integrals (lq > iabcd, vabcd)

CGR Clebsh-Gordon coef.

YLM1 spherical harmonic

Matrices generated in the basis of spin orbitals

GETJ Total Angular momentum operator Pauli spin matrices (s = 1/2 > Ji ) GETLS SOC matrix in the basis of Spin-Orbitals (lq, s = 1/2 > li , si , ls) LMAT matrix of angular momentum ZAB SOC matrix in the basis of SD (li , si , ls > lsdata)

GENER LN (lq,ne > ir,gdata, lsdata,lfdata)

Matrices generated in the basis of microstates

LFAB one electron matrix operator (Eq 2.46) (lfdata)

IWAB

GAB two electrons matrix operator (Eq 2.51) (iabcd, vabcd > gdata)

GIJKL

IROW

Figure 4.3: The structure of the Gener ln matlab program, the lines represent the subroutines called in the program. In each frame, there is the name of the function, a short explanation of the functions role and between parenthesis the input variables needed then the sign > and the outputs variables (i {x, y, z}).

58

4.3 Applications

4.3
4.3.1

Applications
Octahedral CrIII d3 and CoII d7 Complexes

In Tables 4.1 and 4.2 we list, respectively, multiplet energies for the d7 CoX4 (X = Cl , 6 Br ) and the CrX3 (X = F , Cl , Br , CN ) complexes. We use a LDA functional and 6 the optimized CoII -X and CrIII -X bond lengths for the DFT calculations. The multiplet energies obtained are compared to the ones from a LF calculation utilizing the values of B, C and 10Dq tted to the UV-Vis spectra. In lines with section 3.1.2, in Table 4.1 one can remark that the LDA bond lengths are too long and so, the values of 10Dq are too small compared to experiment. Hence, the energies obtained are shifted compared to the experimental ones. The situation improves if instead of optimized bond lengths, experimental ones are taken into account for the calculation (cf. Table 4.1). Even in this case, spin-forbidden transitions come out by 3000-4000 cm1 too low in energy compared to experiment. For the CrX3 (X = F , Cl , Br , CN ) complexes (cf. Table 4.2), the results are 6 less accurate even if for CrF3 and Cr(CN)3 the optimized bond lengths are close to 6 6 experiment. Moreover for the Cr(CN)3 we note a LDA value of 10Dq which is by 6 4000 cm1 higher than the experimental one. This is unusual and not expected. For these complexes with a 4 A2 ground state, spin-forbidden transitions deviate from reported experimental energies by about 6000 cm1 ; a result due to a drastic lowering of the B and C energies. We can conclude that for the latter systems, existing DFT functionals do not perform properly. In addition standard deviations comparing DFT and LFDFT numerical values are too high. Term T1 T2 2 E 4 A2 2 T1 2 T2 4 T1 R(M X) B C 10Dq StDev
4 4

CoCl4 6 Ropt Rexp LFT-t 0 0 0 3384 6613 6261 7772 3980 7794 7335 14050 13467 10688 10134 13542 10804 10479 13783 11412 14008 17241 2.684 2.414 2.414 573 548 795 2540 2436 3108 3952 7436 7206 0.108 0.110 -

Exp. 660091 1330091 1725091 -

CoBr4 6 Ropt Rexp LFT-t 0 0 0 3258 5687 5506 7171 4307 8817 7041 12074 11890 9960 9481 13789 10124 9836 13969 10113 11886 16748 2.821 2.589 492 460 808 2444 2371 3159 3782 6387 6384 0.122 0.124 -

Exp. 570092 1180092 1675092 -

Table 4.1: Theoretical and experimental electronic transition energies of high-spin CoX4 , 6 X = {Cl , Br } octahedral d7 complexes for LDA optimized and experimental geometries. Theoretical values are obtained using LFDFT taken as reference an AOC corresponding to 2t4.2 4e2.8 conguration. Fitted B, C and 10Dq parameters from experimental transition energies g 2g are also included.

59

Term
4 3 A2g (t2g ) 2 3 Eg (t2g ) 2 3 T1g (t2g ) 2 3 T2g (t2g ) 4 1 1 T2g (t2g eg ) 4 1 1 T1g (t2g eg ) 2 1 1 A1g (t2g eg ) 2 1 1 T1g (t2g eg ) 2 1 1 T2g (t2g eg ) 2 1 1 Eg (t2g eg ) 4 1 1 T1g (t2g eg ) R(M-X) B C 10Dq StDev (10Dq)orb

CHAPTER 4. MULTIPLET STRUCTURE

3 CrF6 LDA LFT-t 0 0 12497 15802 13044 16461 18628 23260 13569 15298 19443 22262 24071 28709 26348 31473 25959 30970 27819 33341 30339 34636 1.957 605 734 2694 3492 13598 15297 0.113 13928 -

Exp. 0 1630093 1630093 2300093 1520093 2180093 3500093 1.93396 -

3 CrCl6 LDA LFT-t 0 0 10756 14426 11180 14873 15918 21037 10911 12800 15618 18198 20056 25351 21878 27421 21568 27079 23147 29098 24375 28455 2.419 484 550 2403 3450 10911 12800 0.105 10775 -

Exp. 0 1443094 1280094 1820094 2.33512 -

3 CrBr6 LDA LFT-t Exp. 0 0 0 10333 13900 1390094 10694 14348 15185 20281 9816 12400 1240094 13992 17700 1770094 18709 24459 20316 26503 20047 26159 21530 28126 21861 27643 2.588 2.4712 427 543 2395 3296 9816 12400 0.113 9622 -

3 Cr(CN)6 LDA LFT-t Exp. 0 0 0 9413 12034 1246095 9682 12425 1307095 15234 19084 1837095 30760 26595 2670095 35910 32743 3268095 38324 36488 40022 38569 39983 38438 41084 39952 63147 55352 2.071 2.07797 452 554 1919 2559 30760 26595 0.105 30953 -

3 Table 4.2: Electronic transition energies of CrX6 , X = F ,Cl ,Br and CN complexes with geometries optimized using LDA functionals calculated using values of B,C and 10Dq from least square t to DFT energies (LDA) of the Slater determinants and to experiment (LFT t 1.8 1.2 to exp.). The values of (10Dq)orb as deduced from the eg t2g Kohn-Sham orbital energy dierence taken from the t2g eg SCF Kohn-Sham energies are also listed. Experimental transition energies are also listed.

60

4.3 Applications
Table 4.3: Theoretical and experimental electronic transition energies of the low-spin Mn(CN)3 octahedral d4 complex. Theoreti6 cal values are obtained using geometries from a LDA geometry optimization utilizing values of B, C and 10Dq resulting from least square t to the energies of Slater determinants constructed from 2t2.4 4e1.6 SCF Kohn-Sham orbitals. Best g 2g tted B, C and 10Dq parameters from experimental transition energies are also included.

Electronic state 3 T1g 1 T2g 1 Eg 1 A1g 5 Eg 3 Eg 3 T1g 3 T2g 3 A1g 3 A2g 3 Eg B C 10Dq StDev

LDA 0 6796 7242 16028 20013 31746 32354 32935 33453 34591 35499 444 2361 34085 0.082

LF-t Exp. 0 7948 8710 18519 20699 34326 34966 35713 36761 38130 39630 630 2598 36900

798499 871099 1847099 2070099

4.3.2

Octahedral Cyano MnIII (d4 ), CoIII and FeII (d6 ) Complexes

In this section we are interested in the three following octahedral cyano complexes: Mn(CN)3 (d4 ), Fe(CN)4 (d6 ) and Co(CN)3 (d6 ). 6 6 6 Mn(CN)3 is a low-spin complex with a 3 T1g ground state. The LDA optimized bond 6 distance (1.99 ) matches perfectly the experimental one (1.98 )98 and calculated mulA A tiplet energies agree reasonably with the experimental spectrum (Table 4.3) - the highest deviation (2400 cm1 ) being met for the 3 T1g 1 A1g transition. Fe(CN)4 and Co(CN)3 possess a 1 A1g ground state. The multiplet enegies deter6 6 mined by LFDFT using LDA optimized bond lengths (dFeC = 1.930 and dCoC = 1.899 A ) are listed in Table 4.4. Owing to the over estimatation of the 10Dq value by 1740 and A 2230 cm1 for the Fe and Co complex respectively, DFT energies of 1 A1g 1 T1g , 1 T2g transition are higher in energy than the experimental values, values of B from LFDFT and experiment being very close in these complexes. Dierences in the C parameter are just the opposite (this is due to the least square t procedure), they would lead to lowering of energies of spin-forbidden transitions. However this is overcompensated by the larger DFT value of the parameter 10Dq.

61

CHAPTER 4. MULTIPLET STRUCTURE Electronic State A1g T1g 3 T2g 1 T1g 1 T2g 5 T2g 3 T1g 3 T2g 3 Eg B C 10Dq StDev
3 1

LDA 0 28516 31669 33499 39842 50080 62079 62543 63939 427 2420 35421 0.082

Fe(CN)4 6 LF-t Exp. 0 23698 23700100 26664 30969 30970100 37001 37000100 38067 54723 55010 56352 411 3566 33678

LDA 0 29845 32721 35107 40900 52577 65030 65391 66667 387 2573 37180 0.112

Co(CN)3 6 LF-t Exp. 0 25969 25969100 29276 32500 32500100 39194 39200100 43240 58538 58948 60430 456 3184 34944 -

Table 4.4: Theoretical and experimental electronic transition energies of low-spin Fe(CN)4 6 and Co(CN)3 octahedral d6 complexes. Theoretical values are obtained using geometries from 6 a LDA geometry optimization utilizing values of B, C and 10Dq resulting from least square t to the energies of Slater determinants constructed from 2t3.6 4e2.4 SCF Kohn-Sham orbitals. Best g 2g tted B, C and 10Dq parameters from experimental transition energies are also included.

4.3.3

Applications to Tetrahedral d5 MnCl2 and FeCl1 Com4 4 plexes

In the discussions so far, we noted, that because B and C deduced using DFT data are smaller than parameters from a direct t to experiment, electronic transitions with change of the spin multiplicity are calculated at lower energies than experiment. In this respect high-spin d5 complexes with tetrahedral geometry and a 6 A1 ground state, such as MnCl2 4 and FeCl provide a stringent test of the ability of the up-to-date functionals to calculate 4 multiplet structures for such cases. In Table 4.5 we list such energies and compare them with experiment and with energies using a direct t of B, C and 10Dq to experiment. We base our treatment on LDA optimized Mn-Cl (2.385 ) and Fe-Cl (2.207 ) bond A A distances. For both systems we obtain the remarkable result that all transitions match with the observed transition energies with a shift of about 6000 cm1 . The origin of this correction is possibly due to the fact that DFT leads to lower values in particular of the parameter C.

62

4.3 Applications Electronic state A1 T1 4 T2 4 A1 , 4 E 4 T2 4 E 4 T1 4 A2 4 T1 4 T2 B C 10Dq StDev

4 6

LDA 0 15253 16720 17175 20359 21011 21922 28429 28638 29536 548 2339 3298 0.061

MnCl2 4 LFT-t 0 20954 21720 21975 25198 25587 28051 34883 34998 35514 516 3363 2661 -

Exp.

101

21250 22235 23020 26080 26710 27770 33300 34500 36650

LDA 0 8875 11060 12940 14554 15705 18436 21276 22255 24307 395 1798 5395 0.051

FeCl1 4 LFT-t Exp.101 0 14835 15600 16836 16300 18230 18800 20271 20100 21366 22400 25114 29106 29774 32917 448
b

5320

Table 4.5: Theoretical and experimental electronic transition energies of high-spin tetrahedral (MnCl2 and FeCl1 ) d5 complexes. Theoretical values are obtained using geometries from a 4 4 LDA geometry optimization utilizing values of B, C and 10Dq resulting from least square t to the energies of Slater determinants constructed from 2t3 4e2 SCF Kohn-Sham orbitals. Best t 2 B, C and 10Dq parameters from experimental transition energies are also included.

4.3.4

Tetrahedral d2 Complexes

Tetrahedral d2 complexes possess a 3 A2 (e2 ) ground state and 3 A2 3 T2 and 3 A2 3 T1 , e t2 singly excited states. They give rise to broad d d transitions in the optical spectra. In addition, spin-ip transitions within the e2 conguration lead to sharp line excitations. In this section we consider the three oxoanions of CrIV , MnV , FeV I complexes and the CrX4 (X = F , Cl , Br , I ) complexes. For the three oxanions, multiplet energies from LDA LFDFT agree within 2000 cm1 with experimental data (cf. Table 4.6) supposed that we use experimental rather than of LDA optimized M-O bond lengths. In particular the 3 A2 3 T2 transition energy and thus 10Dq nicely agrees with experiment. LDA optimized M-O bond lengths are larger than the experimental values. This becomes more pronounced with increasing anionic charge from FeO2 (0.02 ) MnO3 (0.05 ) A A 4 4 and CrO4 (0.1 ). This is reected in the values of 10Dq which are distinctly smaller A 4 than 10Dq from experiment (by 1680, 680 and 1600 cm1 , respectively). B and C parameters from DFT (LDA) are by 20-35% smaller than those deduced from a t to observed LF transitions. Thus the 3 A2 3 T1 and the spin-forbidden 3 A2 1 E, 1 A1 transitions, being sensitive to B and C, respectively, are underestimated by DFT (LDA) by 15-20 %. In order to study the eect of the adopted functional we calculated multiplet energies of CrX4 (X = F , Cl , Br , I ) (Table 4.7). Going from the LDA functional to the gradient corrected GGA functional we get a lengthening of the geometry optimized Cr-X bond distances accompanied with a decrease of the value of 10Dq. On the other hand the value of B increases by 15-20 % going from the LDA to the GGA functional.
b

calculated using a C/B ratio equal to 6.14

63

Term
3 A2 (e2 ) 1 E(e2 ) 1 A1 (e2 ) 3 1 T2 (e1 t2 ) 3 1 T1 (e1 t2 ) 1 1 T2 (e1 t2 ) 1 1 T1 (e1 t2 ) 1 2 A1 (t2 ) 1 2 E(t2 ) 3 2 T1 (t2 ) 1 2 T2 (t2 ) R(M-O) B C 10Dq StDev (10Dq)orb

CHAPTER 4. MULTIPLET STRUCTURE

R Rexp opt 0 0 7838 7844 13136 13412 7315 8847 11125 12892 15019 16580 17047 18519 36567 39265 23203 26204 17359 20053 23337 26314 1.868 1.7787 436 427 2250 2274 7315 8847 0.036 0.039 7363 8831

CrO4 4 AOM t Exp. 0 8904 15276 14730104 8949 9087104 13714 13887104 17677 20271 42471 27754 21459 27930 555 2331 8950 -

MnO3 4 Ropt Rexp AOM t Exp. 0 0 0 6560 6583 8537 8432102 11546 11698 14633 9837 10872 10515 10380102 13408 14512 14790 14266102 16329 17393 18956 17857 18908 20875 37278 39268 44416 26726 28804 30202 21309 23309 23203 23430102 26794 28866 30298 1.746 1.700102 347 347 430 1928 1936 2600 9831 10872 10515 0.025 0.027 9984 11025 -

Ropt 0 5179 9369 11259 13936 16408 17437 36109 28000 23470 28030 1.674 242 1637 11259 0.016 11436

FeO2 4 R AOM t Exp. exp 0 0 5197 5712 6215103 9446 10605 9118103 11952 12939 12940103 14645 16936 17700103 17121 18589 18146 20215 37491 40514 29401 32092 24841 27505 29429 32154 1.65c 242 375 1645 1388 11952 12938 0.016 12130 -

4 3 2 Table 4.6: Electronic transition energies in CrO4 , MnO4 and FeO4 oxo-anions calculated using values of B, C and 10Dq as obtained from a t to DFT energies for LDA geometrically optimized clusters (Ropt ), for Td geometries with metal-ligand distances from experiment(Rexp ), from an AOM t to experimental transition energies(AOM t) and experiment (Exp.). Values of B, C and 10Dq and mean square deviations (StDev) between DFT and tted energies (for Ropt and Rexp ) are listed for each entity. The values of (10Dq)orb as deduced from the 4t2 -2e 1.2 Kohn-Sham orbital energy dierence taken from the 2e0.8 4t2 SCF Kohn-Sham energies are also listed.
c

The distance is estimated as a sum of i = 1 ionic radii of FeV I (0.25 A) and OII (1.40 A),105

64

Term

65

A2 (e2 ) E(e2 ) 1 A1 (e2 ) 3 T2 (e1 t1 ) 2 3 T1 (e1 t1 ) 2 1 T2 (e1 t1 ) 2 1 T1 (e1 t1 ) 2 1 A1 (t2 ) 2 1 E(t2 ) 2 3 T1 (t2 ) 2 1 T2 (t2 ) 2 R(M-X) B C 10Dq StDev (10Dq)orb

CrF4 LDA GGA 0 0 8583 9092 14804 15630 10259 9677 14825 14571 18723 18608 20875 21141 43965 44352 29834 29354 23091 22681 29953 29516 1.710 1.735 476 548 2452 2444 10259 9678 0.034 0.043 10549 9928 LDA 0 6542 11114 7008 10316 13454 15074 32099 21121 16033 21217 2.104 355 1903 7008 0.030 7258 LDA 0 6373 10698 6163 9269 12434 14037 30120 19271 14424 19373 2.264 347 1855 6162 0.030 6371

CrCl4 GGA LF-t Exp.106 0 0 0 7101 6089 11971 10586 6524 7010 7250 10062 10440 10000 13489 12991 12000 15456 14718 32891 30599 20878 20716 15795 16229 16666 21014 20822 2.144 419 376 1952 1579 6524 7250 0.039 6686 -

CrBr4 GGA Exp.106 0 0 6832 6666 11364 10869 5605 8820 12295 14215 30531 18777 14043 13258 18920 2.318 403 1887 5605 0.038 5726 -

CrI4 LDA GGA 0 0 5986 6514 9921 10705 5267 4781 8054 7663 8054 7663 12659 13021 27416 28198 17070 16819 12591 12486 17172 16969 2.489 2.547 323 387 1758 1798 5266 4782 0.032 0.033 5605 4984 4.3 Applications

Table 4.7: Electronic transition energies of CrX4 , X = F , Cl , Br and I ions with geometries optimized using LDA and GGA functionals calculated using values of B, C and 10Dq from least square t to DFT energies of the Slater determinants. The values of (10Dq)orb as deduced from the 4t2 -2e Kohn-Sham orbital energy dierence taken from the 2e0.8 4t1.2 SCF Kohn-Sham energies are also listed. Experimental 2 transition energies for CrCl4 and CrBr4 as well as values of B, C and 10Dq deduced from a t to experiment for CrCl4 are also included.

CHAPTER 4. MULTIPLET STRUCTURE

4.4

General Conclusions

In conclusion, our LFDFT allows to calculate multiplet energies in TM complexes, similar to the empirical CI-DFT procedure described by Grimme et al.107, 108 In this approach, KS potentials and energies have been used to approximate the exact CI Hamiltonian by introducing scaling factors, thus avoiding double counting of electron correlation in the o-diagonal CI matrix elements. Though, principally able to treat open-shell TM complexes as well, the method has been applied thus far only to closed shell ground state systems. Considering the LFDFT, we can make the same remark than in the previous chapter: if we have the experimental structure, it is better to use this one than the optimized, as the 10Dq value depend strongly on the bond length. In the set of compounds we study in this chapter, we saw that sometimes we were in perfect agreement with the expimental data which means that the theory is valid, but sometimes we had a constant shift in energy between experiment and LFDFT results or a less accurate description of the multiplet structure and this can be due to the fact that the functional used does not describe well the compounds considered.

66

Chapter 5 Relativistic Eect within LFDFT

In this chapter we extend the LFDFT with the inclusion of spin-orbit coupling. A good opportunity to achieve this consist in using for the DFT calculations the Zero-Order Regular Approximation109 (ZORA) which permits to include spin-orbit coupling eects variationally. ZORA has been implemented into the Amsterdam Density Functional (ADF) program74 and has been proven to work well. This chapter is organized as follows: in Section 5.1 spin-orbit coupling in Td symmetry and in its subgroup D2d is described, in Section 5.2 the implementation of the formalism along the lines of LFDFT is presented. It is, thus, possible to get symmetry consistent spin-orbit coupling parameters without recourse to implicit use of orbital reduction factors. We intend to show that orbital dependence of that kind can be larger than one might expect. We will show that it is possible to apply the formalism using data from DFT-ZORA calculations. Finally, we apply the theory to the NiX2 (X = F , Cl , Br , 4 I ) series of compounds and compare the results with experimental data from literature. An outlook towards an extension of the theory to systems with little or no symmetry will also be given.

5.1
5.1.1

Spin-Orbit Coupling
Theoretical Description

The spin-orbit operator in Eq 2.58 enables the expression of matrix elements of spin-orbit coupling in and between sub-shells in terms of molecular orbitals. Thus, any arbitrary one-electron spin-orbit coupling matrix element can be written as: s ms , a | HSO |s ms , b = s ms , a | u s |s ms , b =
k=x,y,z

s ms | sk |s ms a | uk |b

(5.1)

where s is the spin of a single electron with component ms , a and b are the irreducible representations (irreps.) of the molecular orbitals and and the corresponding components in case of degeneracy. The 1st term on the right hand side of Eq 5.1 is nothing but the Pauli matrices and the 2nd term can be further reduced using Wigner-Eckhardts theorem: a | uk |b = a |b , t1 k a u b (5.2) 67

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT where a |b , t1 k are coupling coecients for the tetrahedral group and a u b are reduced matrix elements. Combining Eqs 5.1 & 5.2 enables us to express any arbitrary spin-orbit matrix element as a product of symmetry coecients and reduced matrix elements: s ms , a |HSO |s ms , b = a||u||b
k=x,y,z

s, ms |k |s ms a |b , t1 k s

(5.3)

In order to estimate the yet unknown reduced spin-orbit coupling matrix elements a u b we shall map Eq 5.3 onto a ZORA-DFT calculation and adjust the reduced matrix elements in order to reproduce the calculated ligand eld levels as done previously in LFDFT. This task requires a symmetry adaptation of the Kohn-Sham molecular orbitals to the double-group Td * i.e.: e 6 (, ) = 8 : |8 (e) : |8 (e) : |8 (e) : |8 (e) : t2 6 (, ) = 8 : i 2 1 |8 (t2 ) : = |t2 , |t2 , + |t2 , ; 6 6 6 1 i |8 (t2 ) : = |t2 , + |t2 , ; 2 2 1 i |8 (t2 ) : = |t2 , + |t2 , ; 2 2 i 2 1 |8 (t2 ) : = |t2 , |t2 , + |t2 , 6 6 6 t2 6 (, ) = 7 : 1 |7 (t2 ) : = |t2 , + 3 1 |7 (t2 ) : = |t2 , 3 i 1 |t2 , + |t2 , ; 3 3 i 1 |t2 , |t2 , ; 3 3 = |e , ; = |e , ; = |e , ; = |e , ;

(5.4)

where the notation of the double-group representation is according to Bethe and their components are Griths.31 Using this basis transformation along with spin-orbit coupling elements (cf. Appendix D.1), the one-electron spin-orbit coupling and LF matrix reduces to: 8 (e) 8 (e) 8 (t2 ) 7 (t2 ) hee
3 t1 i 2 et2

8 (t2 )
3 t1 i 2 et2

7 (t2 ) 0 (5.5)

1 ht2 t2 1 tt2 t2 0 2 1 0 ht2 t2 + tt2 t2

68

5.1 Spin-Orbit Coupling where hee and ht2 t2 are the one-electron ligand eld matrix elements, the ht2 t2 - hee diert1 1 ence being the cubic eld splitting and et2 and tt2 t2 the reduced matrix elements:
t1 et2 = e||su(t1 )||t2
1 tt2 t2 = t2 ||su(t1 )||t2

(5.6)

which we derive in a form to be directly compared to the free Ni2+ spin-orbit coupling constant (630cm1 ). In section 5.3.3, we describe a procedure of getting these parameters from ZORA-DFT calculations. The 3 T1 ground state of Ni2+ (d8 ) in a tetrahedral (Td ) ligand eld is Jahn-Teller unstable and distorts towards tetragonal D2d symmetry with elongation along the S4 axis of the tetrahedron. For this point group, the symmetry species t2 (, , ) and e (, ) split into e (, ) + b2 () and a1 () + b1 () respectively. Symmetry adaptation of the Kohn-Sham molecular orbitals to the D double group yields: 2d a1 6 (, ) = 6 : |6 (a1 ), = |a1 , |6 (a1 ), = |a1 , e 6 (, ) = 6 : i 1 |6 (e), = |e, |e, 2 2 1 i |6 (e), = |e, + |e, 2 2 b1 6 (, ) = 7 : |7 (b1 ), |7 (b1 ), b2 6 (, ) = 7 : |7 (b2 ), |7 (b2 ), e 6 (, ) = 7 : 1 i = |e, |e, 2 2 1 i |7 (e), = |e, + |e, 2 2 where again double-group representation are according to Bethe. Using this basis transformation along with the spin-orbit coupling matrix elements (cf. Appendix D.1) we obtain the spin-orbit coupling and LF matrix for this symmetry: |7 (e), 6 (a1 ) 6 (a1 ) 6 (e) 7 (b1 ) 7 (b2 ) 7 (e) ha1 a1
3 e i 2 a1 e

= |b1 , = |b1 , = |b2 , = |b2 ,

(5.7)

6 (e)
3 e i 2 a1 e 1 a2 hee 2 ee 0 0 0

7 (b1 ) 7 (b2 ) 0 0 hb1 b1 a ib12b2 i e 2 b1 e 0

7 (e) 0 (5.8)

0 0 0

0 0 a2 i e ib1 b2 2 b1 e 1 e hb2 b2 2 b2 e 1 a2 e hee + 1 ee 2 2 b2 e

69

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT where ha1 a1 , hee , hb1 b1 , hb2 b2 are the (diagonal in this case as well) one-electron ligand eld k matrix elements and ij are the reduced matrix elements:
e a1 e = a1 ||su(e)||e a b12b2 = b1 ||su(a2 )||b2 e b1 e = b1 ||su(e)||e a2 ee = e||su(a2 )||e e b2 e = b2 ||su(e)||e

(5.9)

t1 1 We note that in D symmetry the cubic quantities et2 and tt2 t2 split into three and two 2d dierent reduced matrix elements thus yielding a total of ve independent parameters. In ligand eld studies, thus far the variation of spin-orbit coupling between the various symmetries of the involved ligand eld orbitals have been approximated in terms of orbital reduction factors. In the next section, we derive a rigorous procedure allowing us to deduce these quantities from DFT-ZORA calculations.

5.2

Computational Procedure

The DFT calculations have been performed with the aid of the ADF program code (release 2003.01).74 For the exchange-correlation functionals, both, the local density approximation (LDA, for geometry optimizations) and the generalized gradient approximation (GGA , for energies of electronic states) have been used. For LDA we adopt an X functional for exchange ( = 0.7)110 and Vosko, Wilk and Nusair functional for correlation.57 The GGA has been introduced in the form given by Perdew-Wang.111 The frozen core approximation was used for inner core electrons. The orbitals up to 3p for Ni, 1s for uorine, 2p for chlorine, 3d for bromine and up to 4d for iodine were kept frozen. The valence shells were described by triple zeta plus one polarization function (TZP basis set). Spin-restricted relativistic ZORA calculations have been done adopting the ZORA basis set (TZP). Using basis functions of increasing quality from TZP to TZ2P to TZ2P+ does not change results signicantly.

5.2.1

Geometry Optimizations

Geometry optimizations of the NiX2 species have been done in non-relativistic spin4 unrestricted (Ms =1) formalism, using the LDA-only functional, which we know from experience35 to yield TM-ligand bond distances in good agreement with experiment. In order to study the Jahn-Teller activity within the 3 T1 ground state, separate optimizations imposing a D2d geometry have been done following the guidelines of accounting for the Jahn-Teller eect within DFT.112 It should be noted that, being a single determinant method, DFT is not able to yield optimized geometries in the case where two or more congurations mix with each other. This is the case in our study, where two 3 A2 states (originating from 3 T1 in tetrahedral symmetry) - corresponding to the ground conguration (e4 t4 ) and to the excited conguration (e3 t5 ) - mix and reduce the extent of the 2 2 structural distortion and of the Jahn-Teller stabilization (EJT ).113 A procedure to solve 70

5.2 Computational Procedure this problem is briey outlined in Appendix D.2 and is used here to get ground state D2d geometries for all NiX2 species. 4 The geometries optimized, the LFDFT method is used to determine the B and C Racahs parameters and the 5 5 LF matrix for the two symmetries, which in our cases take a diagonal form with two dierent energies (e and t2 ) for Td and 4 dierent energies (ha1 a1 , hb1 b1 , hb2 b2 and hee ) for D2d .

5.2.2

Calculation of Reduced Matrix Elements of Spin-Orbit Coupling from ZORA-DFT

Let us consider the Kohn-Sham eigenvalues in the basis of the double group symmetry adapted fragment orbitals (SFO) and the symmetry adapted fragment spin-orbitals (see Eq 5.4 & Eq 5.7). To be more specic let us take as an example NiCl2 . Focusing on the 4 Td complex, the 5t1 , 16t2 and 8e spin-orbitals give rise to a total of 298 and 167 KSorbitals, from which we identify the 68 , 78 and 47 ones with dominant 3d character. We correspondingly occupy these evenly in the ZORA input with occupation numbers and ZORA eigenvalues taken from the output listed in Table 5.1.

Double group KS-orbital 68 78 4 7

Occupancy KS (eV)

3.2 3.2 1.6

2.940 3.458 3.509

Symmetry and number 8E 5T1 3d Nr.1 no 3d-species c(1,6)=-0.842i X c(1,7)=-0.135i X 0 0

of SFO 16T2 3d Nr.2 c(15,6)=-0.104 c(15,7)= 0.789 c2(2,4)=-0.816

Table 5.1: The selecting scheme for ZORA spin-orbit coupling eigenvalues, 3d-eigen-functions and their occupations for tetrahedral NiCl2 as a model example. 4

We further nd from the table of the SFO 8, 5 and 16 species of E, T1 and T2 symmetry, to yield a total of 29 basis functions for 8 and 16 T2 species to give rise to the same number of 7 species. It is further important that the 3d orbitals, whose identity is sought, are the rst and the second in the list for the E and T2 , representation, respectively. These yield the coecients of the 68 , 78 and 47 KS-ZORA eigenvectors (Table 5.1, to be extracted from the TAPE15 output after saving and converting into ASCII). We thus get truncated ZORA-KS eigenvectors and (diagonal) eigenvalues matrices U and (Eq 5.10) which after 68 78 47 e 0.842 0.135 0.0 68 2.940 0.0 0.0 = 78 0.0 0.0 3.458 0.0 (5.10) U = t2 0.104 0.789 t2 0.0 0.0 0.816 47 0.0 0.0 3.509 S = U.U T h = {h } = S (1/2) U U T S (1/2) 71 (5.11) (5.12)

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT e(8 ) t2 (8 ) t2 (7 ) (5.13)

e(8 ) 2.951 0.074i 0.0 0.0 h = t2 (8 ) 0.074i 3.447 t2 (7 ) 0.0 0.0 3.509

manipulations as described by Eq 5.11 and Eq 5.12 yields the one-electron Hamiltonian matrix (Eq 5.13). The comparison with Eq 5.5 yields directly the reduced matrix elements t1 1 tt2 t2 and et2 as well as the cubic ligand splitting (333, 487 and 4166 cm1 , respectively). In a similar way, the matrices of the spin-orbit coupling plus the ligand eld for NiCl2 4 in D2d symmetry are derived from matrix 5.14 to give using denition Eq 5.9 the ligand eld and 6 (a1 ) 2.991 0.072i 0.0 0.0 0.0 6 (e) 0.072i 3.582 0.0 0.0 0.0 0.0 7 (b1 ) 0.0 2.950 0.063i 0.040i (5.14) 0.0 7 (b2 ) 0.0 0.063i 3.301 0.029 7 (e) 0.0 0.0 0.040i 0.029 3.623 spin-orbit coupling matrix elements listed in Table 5.2. Thus, from a single ZORA calculation both the ligand eld and spin-orbit coupling matrices are obtained. In order to facilitate the analysis of the ADF output a series of MATLAB scripts are used as interfaces. Td hee ht2 t2
t1 et2

1 tt2 t2

D2d ha1 a1 (dz2 ) hb1 b1 (dx2 y2 ) hb2 b2 (dxy ) hee (dxz , dyz ) e a1 e e b1 e a b12b2 a2 ee e b2 e

[NiF4 ]2 Td D2d -1904 -2622 -3600 -990 1748 3247 576 588 566 598 516 518 516

[NiCl4 ]2 Td D2d -2410 -2500 -2736 95 1667 2526 474 488 461 510 326 332 329

[NiBr4 ]2 Td D2d -2310 -2277 -2386 178 1518 2259 337 373 300 452 -242 -242 -253

[NiI4 ]2 Td D2d -2150 -2037 -2151 310 1358 1945 222 375 198 410 -824 -832 -837

Table 5.2: Ligand eld (diagonal) matrix elements and reduced spin-orbit coupling matrix elements for Td and elongated Td D2d DFT optimized (spin-unrestricted, LDA-functional) [NiX4 ]2 (X = F , Cl , Br , I ) geometries from ZORA (spin-restricted, LDA+PW91 functional) calculations.

72

5.3 Results and Discussion

5.3
5.3.1

Results and Discussion

The Jahn-Teller Eect and the Geometries of NiX2 (X = 4 F , Cl , Br , I ).

NiX2 (X = Cl , Br and I ) are well stud4 ied structurally114, 115 and/or spectroscopically116119 and found to exist as slightly distorted (NiCl4 ) or almost regular tetrahedra (NiBr4 and NiI4 ). NiF2 is not known yet, Ni2+ -F complexes tend4 ing invariably to adopt a regular octahedral geometry. In line with these observations, our geometry optimizations (Table 5.3) show, that the extent of the Jahn-Teller elongation (Fig. 5.1) and the stabilization energy is strongly reduced due to mixing between the e4 t4 and e3 t5 con2 2 gurations (cf. Appendix D.2), the latter conguration being Jahn-Teller stabilized by a D2d compression. D2d D2d In this respect, the e3 t5 conguration resemTd 2 2+ compression elongation bles very much the Jahn-Teller activity in Cu with a single hole in the t2 -shell, which readily Figure 5.1: The angle describing the explains the dierent stereochemistries of these tetrahedral distortion due to Jahn-Teller 2 two ions.113 In Table 5.3, we also list geome- activity in NiX4 . 2 tries and EJT values of NiX4 neglecting the 3 T1 (e4 t4 )-3 T1 (e3 t5 ) mixing. It is this geometry with more pronounced distortions, which 2 2 we use in order to explore the eect of symmetry lowering on the anisotropy of spin-orbit coupling. However in the discussion of electronic transitions and comparison with experiment we make use of the correct geometry.

X F Cl Br I

R min 1.94 -5.2 [-8.3] 2.29 -3.1 [-5.5] 2.44 -2.9 [-5.3] 2.64 0.0 [-4.6]

EJT -335 [-825] -83 [-260] -62 [-198] 0 [-49]

Table 5.3: Bond lengths (in ), the extent of Jahn-Teller elongation of the tetrahedral into A the D2d ground state energy minima (min in o ) and the Jahn-Teller stabilization energy EJT (in cm1 ) for NiX2 (X=F ,Cl ,Br and I ). Spin unrestricted DFT geometry optimizations 4 (VWN-LDA functional) with and without (in square brackets) taking mixing between 3 T1 (e4 t4 ) 2 and 3 T1 (e3 t5 ) into account (still described by single determinants) have been performed, using 2 a strategy described in Appendix D.2.

73

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT

5.3.2

Spin-Orbit Coupling in Td Symmetry

The (,) - e, t2 (3d) spin-orbit coupling in Td symmetry gives rise to 8 (e) and 8 + 7 (t2 ) spinor levels, whose ZORA energies we plot in Fig. 5.2.

KS (in eV) 7 (t2 ) 8 (t2 + e)

0.3 0.1 0.1 0.3

8 (e + t2 ) 0.5

[NiF4]2

[NiCl4]2

[NiBr4]2

[NiI4]2

Figure 5.2: Relative energies of Kohn-Sham orbitals with dominant 3 d character from a ZORA spin-orbit calculation of [NiX4 ]2 (X=F , Cl , Br , I ).

While spin-orbit coupling leads to splitting of t2 (,) in rst order. It causes a mixing of 8 (t2 ) and 8 (e) to second order. Ligand eld splitting (Td ) is calculated in the range of 4370 (NiF2 ) to 3395 (NiI2 ) and dominates over the spin-orbit coupling: ( Ni2+ )= 4 4 630 cm1 ; this is accounted for in Fig. 5.2, where dominant contributions from e and t2 are underlined. We notice that when moving from NiF2 to NiCl2 the 8 (t2 +e)-7 (t2 ) 4 4 1 energy separation (which equals (3/2) tt2 t2 , see Eq 5.5) gets smaller and it is lowered further going to NiBr2 and NiI2 . It is important to observe that the sign of the splitting 4 4 changes in the latter two complexes. The analysis of the energy levels (Fig. 5.2) in terms t1 1 of the parameters ht2 t2 -hee , et2 and tt2 t2 (cf. Eq 5.5, calculated values are listed Table 5.2) t1 shows indeed that t2 t2 becomes negative in the case of NiBr2 and NiI2 . As has been 4 4 pointed out previously,119 ligand spin-orbit coupling (as large as 5000 cm1 for I !) in combination with the metal(3d)-ligand(np) mixing can strongly modify the eective spinorbit coupling constant; this contribution can be of dierent sign and can even outweigh the spin-orbit coupling due to the 3d electrons; this turns the sign of the 8 (t2 +e)-7 (t2 ) splitting. Our ZORA calculations lend full support of this proposition, initially observed by MCD data on NiI2 .119 Contrary to earlier interpretations120, 121 our results indicate 4 that spin-orbit coupling undergoes a stronger decrease with increasing metal-ligand covalency than Stevens orbital reduction factors in the magnetic moment operator (amenable from a t to magnetic susceptibilities). Indeed, with increasing covalency from F to Cl t1 1 to Br to I , et2 and tt2 t2 show a much stronger reduction than deduced from orbital reduction factors as the squared MO coecients c2 for 3d in the e and t2 MOs and their 3d combinations [0.82, 0.72, 0.68, 0.66 (e) and 0.75, 0.63, 0.60, 0.56 (t2 ) for F , Cl , Br , I complexes, respectively]. The order of values for the reduced matrix elements in a 74

5.3 Results and Discussion

t1 1 given complex et2 > tt2 t2 (obeyed for all cases in Table 5.3) reects also subtle changes in the metal-ligand overlap (dierential covalency) being larger for the more strongly ( + )-antibonding t2 , compared to the only weakly () - antibonding e orbital.

5.3.3

Spin-orbit Coupling in D2d Symmetry

Going to a D2d distorted Td complex, the e(dz2 ,dx2 y2 ) and t2 (dxy , dxz , dyz ) orbitals split into a1 (dz2 )+b1 (dx2 y2 ) and b2 (dxy )+e(dxz ,dyz ) species and the ligand eld matrix becomes fully dened in terms of 10Dq plus the t2 and e splitting parameters 32 and 21 , respectively. This is illustrated on Figure 5.3 with parameter values pertaining to NiCl2 . 4
Figure 5.3: Orbital level splittings and notations for a symmetry based on description of the D2d distorted (elongated) NiCl2 complex 4 for a geometry obtained without correction for 3 T (e4 t4 )-3 T (e3 t5 ) missing. 1 1 2 2
e (dxz ,dyz ) 4805 +2 4047 t2

22 2531

b2 (dxy ) 10Dq

a1 (dz 2 ) 0e b1 (dx2 y2 )

+1 1

+173 -173

In parallel with this increase of the level of parameterization, the matrix of spin-orbit coupling becomes dependent on ve reduced matrix elements (eq 5.8). In Figure 5.4 we show their variation with the angular geometry for NiCl2 , changing the -angle in wide 4 range from elongated to compressed D2d structures. It is striking that the variations k of the spin-orbit coupling parameters ij follow the same trends as the energies of the correspondingly involved orbitals i and j (Fig. 5.5 & 5.4); the stronger the extend of k anti-bonding is (increasing the energy of the involved orbitals i, j) the stronger the ij reduction is. k Thus, the ij parameters reect, in attenuated way, the angular dependence of the ligand eld matrix. In classical LF theory, the parameter has been deemed to be of atomic nature, being scaled by some reduction factor, occasionally accounting for axial anisotropy as well. In view of our results we suggest that such treatments are incomplete. The strong interrelation between the ligand eld and molecular spin-orbit forces can even invalidate the common opinion of the two physical eects being opposite to each other.122 In cases of very low-symmetry (causing non-zero o-diagonal elements in the matrix h ) it can 75

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT

KS (in eV)

4.4 4.2 4.0 3.8 3.6 3.4 3.2 3.0 2.8 2.6 e e [2 e (D4h )] b1 [4 e (D4h )] a1 [e -4esd (D4h )] t2 D2d elong.

Td

D2d compr.

b2 [3 e (D4h )]

45 50 55 60 65 70 (in 0 ) Figure 5.4: The dependence of the Kohn-Sham orbital energies from non-relativistic cal1.6 culation with average-of-conguration occupancies (Td : e3.2 t4.8 ; D2d : a1.6 b1.6 b2 e3.2 ) on the 2 1 1 A geometrical angle for NiCl2 (R = 2.29 , PW91-functional, TZP basis). Orbital energy 4 expressions on the right hand side refer to the limiting case of a compression - square planar geometry and angular overlap model expressions (see Section 5.3.4).

Td hee ht2 t2 B C

D2d ha1 a1 (dz2 ) hb1 b1 (dx2 y2 ) hb2 b2 (dxy ) hee (dxz , dyz ) B C

[NiF4 ]2 Td D2d -1696 -2494 -3528 -999 1662 3111 715 713 2732 2685

[NiCl4 ]2 Td D2d -2320 -2428 -2666 146 1619 2420 521 521 2136 2126

[NiBr4 ]2 Td D2d -2321 -2274 -2363 247 1516 2218 462 458 1944 1948

[NiI4 ]2 Td D2d -2226 -2034 -2032 445 1356 1906 401 399 1804 1844

Table 5.4: Ligand eld (diagonal) matrix elements and interelectronic repulsion B and C energies (in cm1 ) for Td and elongated Td-D2d DFT-optimized (spin-unrestricted, LDA-functional) [NiX4 ]2 (X = F , Cl , Br , I ) geometries from LFDFT (LDA+PW91 functional) calculations.

76

5.3 Results and Discussion

k ij (in cm1 )

e b1 e

550

D2d elong.

Td

D2d compr.

500

t1 et2

e a1 e

450

400

a b12b2

a2 ee

350

1 tt2 t2

300

e eb2

250

45

50

55

60

65

70

(in 0 )

Figure 5.5: Reduced matrix elements of the spin-orbit coupling operator from ZORA-ADF calculations in the Td and D2d geometries of the NiCl2 and their dependence in the geometrical 4 anlge (see g. 5.1).

77

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT B 810 cm1 521 cm1 518 cm1 535 cm1 530 cm1 521 cm1 359 cm1 C 3150 cm1 2136 cm1 2196 cm1 2218 cm1 2212 cm1 1576 cm1 1596 cm1 3500 cm1 4155 cm1 4147 cm1 4106 cm1 4115 cm1 4220 cm1 3041 cm1

Experiment

116

GGA

PW91 PBE RPBE revPBE BLYP LB94

Table 5.5: The Racahs parameters determined with LFDFT method for [NiCl4 ]2 using the exchange and correlation functionals available in ADF2003.01

even prevent the separation of the one-electron Hamiltonian, in parts due to ligand eld and spin-orbit coupling. This is fortunately not the case here (h is fully diagonal) to enable a neat analysis of the two eects separately.

5.3.4

Ligand Field-Parameters, Ground and Excited States Energies of NiX2 (X=Cl , Br , I ) and Comparison with 4 Experimental Data

A list of ligand eld parameters - the (diagonal) ligand eld matrix and the B and C values for NiX2 halide complexes in Td and D2d symmetry are listed in Table 5.4. There 4 is a good agreement between the LFDFT values of 10Dq for NiCl2 (4150 cm1 ) and 4 the one deduced from a direct t to the spectrum (3500 cm1 ,116 referred to hereafter as experimental). However, values of B and C are correspondingly 64% and 68% o from the experimental ones (810 and 3150 cm1 , respectively116 ). This can be traced back to the functionals in use which overestimate 3d-electron delocalization. A list of LF parameters in dependence of the functionals oered by the ADF code (Table 5.5) shows a weak sensitivity with respect to this choice. Likewise, use of more sophisticate basis functions, such as a quadruple 3d basis for 2+ Ni (TZ2P+) does not alter the values of (causing an increase of s (Table 5.2) by 3 4%). Excited state energies (Table 5.7) for geometries, corresponding to the Td and D2d energy minima (in this case taking 3 T1 (e4 t4 ) 3 T1 (e3 t5 ) mixing into account) shows a good 2 2 agreement between LFDFT and experimental values for the energies of the transitions 3 T1 3 T2 and 3 T1 3 A2 , while the energies to the spin-forbidden transitions to 1 T2 , 1 E and 1 T2 ,1 T1 ,1 A1 , as well as the spin-allowed one to 3 T1 are found to be by about 4000 cm1 and 5500 cm1 smaller than the experimental ones. This is in accordance with the smaller B and C values and the stronger dependence of the energies of the mentioned states on B and C. It is interesting that LF matrices from a non-relativistic LFDFT calculation and from ZORA dier from each other. These are compared in Table 5.6 taking the b1 a1 , b1 b2 and b1 e energy dierences. One can easily show when taking the angular geometry into account that these dierences can be translated into an angular overlap parameterization to yield values of the parameters for and -bonding and for the mixing 78

5.4 Conclusions [NiF4 ]2 LFDFT ZORA 1832 1696 2529 2610 6639 6847 6488 6688 2612 2690 2277 2629 4353 4597 4110 4237 [NiCl4 ]2 LFDFT ZORA 346 326 2812 2831 5086 5262 4980 5350 1594 1857 3402 4094 4155 4289 2274 2431 [NiBr4 ]2 LFDFT ZORA 42 76 2610 2564 4581 4645 4457 4735 1379 1622 3910 4382 3903 3913 1971 2081 [NiI4 ]2 LFDFT ZORA -194 10 2477 2461 3938 4096 3734 4231 1012 1397 4359 4562 3548 3550 1461 1635

E(b1 a1 ) E(b1 b2 ) E(b1 e) e e esd 10 Dq 32

Table 5.6: Orbital interpretations of ligand eld energies from LFDFT and ZORA calculations in terms of the AOM parameterization scheme (e , e , esd ) along with values of 10Dq and the t2 -splitting 32 for tetragonally (D2d ) elongated [NiX4 ]2 (X = F , Cl , Br , I ).

of dz2 and 4s (which, being of the same symmetry, i.e. a1 , in D2d can mix with each other)yielding e , e and esd , respectively. Remarkably, ZORA results reect a distinctly larger extent of - and -antibonding and esd -mixing compared to the non-relativistic LFDFT result. The eect of sd-mixing has a crucial inuence via the Fermi contact terms on the hyperne structure: A-tensor (treated in Section 6.1.1.2). We also notice that the value t1 of the parameter t2t2 for NiCl2 we deduce from the ZORA results (332 cm1 , Table 5.2) 4 is found in excellent agreement with the one deduced from a t to magnetic susceptibility data (380 cm1 ).120, 121 It is this parameter (in combination with the 3 T1 e ground state Jahn-Teller activity, see Appendix D.2) which aects the 3 T1 ground state splitting. These are shown in Figure 5.6 for NiX2 (X = Cl , Br and I ). In accordance with large 4 t1 negative t2t2 value for NiI2 a inverted zero-eld splitting pattern for the ground state is 4 calculated with a T2 ground state and an E excited state 33 cm1 higher in energy. In NiBr2 an intermediate coupling scheme is realized with an E ground state and T2 next 4 in energy. A D2d distortion leads to an A1 ground state, as it is the case as in NiCl2 , 4 however originating from the tetrahedral E, rather than from the tetrahedral A1 state 1 as in NiCl2 . Therefore it follows that the sign and magnitude of the parameter tt2 t2 is 4 crucial importance for the ground state splitting. This, as well of Jahn-Teller activity in the 3 T1 ground state is a further experimental challenge for these systems. For more conclusive results, these compounds should be studied with more modern tools, such as resonance Raman and high-eld - high-frequency EPR. We hope this study will motivate further experimental work in this direction.

5.4

Conclusions

In this chapter, we extended our LFDFT proposed with spin-orbit coupling, developing a procedure allowing one to extract spin-orbit coupling matrix elements from spin-restricted ZORA calculations. We show, using a symmetry analysis, that a set spin-orbit coupling parameters, rather than a single value or values scaled for anisotropy using (Stevens) orbital reduction factors are needed in order to properly describe a realistic situation. In particular, highly covalent ligands such as Br and I and their own spin orbit coupling 79

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT contributions can lead to unexpected, unprecedented splitting patterns of the ground state (zero-eld splitting) which deserve further theoretical justication and testing and motivates further experimental work. In following contributions we will extend the formalism to systems with lower or no symmetry, calculating zero-eld splitting for systems well characterized by EPR, thus extending the applicability of our LFDFT approach to ne structure tensor and hyperne coupling eects.
B2 (1516) B1 (1407) E (1329)

A1 (1059) A2 (970) E (948) B2 (940) B1 (934) T2 (782) A2 (922) E (667) A1 (908) A1 (712) T1 (598)

T1 (234) E (91) A1 (0)

T1 (40) A1 (37) T2 (18) E (0) E (16) A1 (0) E (33) T2 (0)

NiCl2 NiBr2 NiI2 4 4 4 Figure 5.6: The splitting of the 3 T1 ground state in the Td geometry due to the spin-orbit coupling and in the lower D2d symmetry minimum [for NiCl2 and NiBr2 , 3 T1 (e4 t4 )-3 T1 2 4 4 (e3 t5 ) mixing is taken into account]. Data for NiBr2 are plotted schematically, not following 2 4 the energy scale used for NiCl2 and NiI2 . 4 4

80

NiBr2 4

NiI2 4

81

Electronic State Td (D2d ) 3 T1 (3 A2 , 3 E) 3 T2 (3 E, 3 B2 ) 3 A2 (3 B1 ) 1 T2 (1 B2 , 1 E) 1 E (1 B1 , 1 A1 ) 3 T1 (3 E, 3 A2 ) 1 T2 (1 E, 1 B2 ) 1 T1 (1 E, 1 A2 ) 1 A1 (1 A1 ) 1 E (1 B1 , 1 A1 ) 1 A1 (1 A1 ) LFDFT Exp.117 0.0 (0.0, 1.65) 3.28 (3.86, 5.80) 7.06 (8.41) 6.40 6.89 (6.92, 8.44) 7.19 (7.03, 9.02) 9.69 (10.32, 12.01) 11.50 (12.26, 13.75) (16.50, 17.44) 12.71 (13.23, 15.24) 18.13 12.47 (14.03) 14.99 (16.25, 16.42) 19.00 27.29 (28.56)

NiCl2 4 LFDFT Exp.118 0.0 (0.0, 1.87) 3.48 (4.20, 6.21) 3.84 - 4.76 7.53 (9.03) 6.90 7.64 (7.67, 9.40) 11.69 7.99 (7.80, 10.08) 12.22 10.74 (11.44, 13.57) 14.70 12.69 (13.61, 15.10) 18.18 14.01 (14.55, 17.06) 19.48 13.79 (15.49) 16.38 (17.78, 18.09) 22.08 30.12 (31.55) -

LFDFT Exp.119 0.0 (0.0, 1.24) 2.93 (3.39, 4.99) 6.32 (7.47) 6.22 (6.34, 7.46) 6.47 (6.44, 7.98) 8.49 (9.09, 10.19) 10.30 (10.94, 12.22) 11.35(11.97, 13.28) (13.60, 14.64) 11.15 (12.45) 14.08 13.43 (14.59, 14.66) (15.75, 15.62) 24.58 (25.89)

Table 5.7: Multiplet energies (in kK=103 cm1 ) for NiX2 (X=Cl ,Br ,I ) complexes from LFDFT calculations (zero spin-orbit coupling) 4 in their Td and D2d -elongated geometries (from DFT geometry optimizations taking mixing between 3 T1 (e4 t4 ) and 3 T1 (e3 t5 ) into account) 2 2 in comparison to experimental data from literature.

5.4 Conclusions

CHAPTER 5. RELATIVISTIC EFFECT WITHIN LFDFT

82

Chapter 6 ESR Parameters within LFDFT

A wealth of information is encoded in the electron paramagnetic resonance (EPR) spectra of a given molecule. The EPR spectroscopy provides unique insights into the structural and electronic features of both organic and inorganic paramagnetic compounds. In organic chemistry, the technique is mostly used to get access to information about free radicals, since it provides a direct experimental measure for the distribution of the unpaired spin density. In the eld of inorganic, or bio-inorganic, chemistry, the method is applied to examine the splitting of orbital and spin levels and the molecular symmetry and environment. Consequently, it is of great interest to compute and analyze the relevant quantities (g- and A-tensors) of an EPR spectrum from rst principle calculations. The area of theoretical and computational chemistry has in the last couple of years shown an increasing interest in the calculation of magnetic coupling parameters. The application of Density Functional Theory (DFT) to EPR spectroscopy is relatively recent excepting the pioneering work of Daul and Weber123 based on multiple-scattering X. Schreckenbach and Ziegler124 have presented energy derivative calculations for the g-tensor with the usage of gauge including atomic orbitals (GIAO). Baerends and al. as well as Bruyndonckx125 have published results of DFT calculations for the g- and Atensors of TiF3 by means of second order perturbation theory.126 Furthermore, with the recent developments to include relativistic eects in modern DFT calculations, spin-orbit coupling (SOC) can be taken into account variationally using the zeroth-order regular approximation (ZORA) to the Dirac equation.109, 127129 To obtain then the g-tensor, the eect of the external homogeneous magnetic eld has only to be treated with rst order perturbation theory.130, 131 In this chapter, we present the result of calculations of the g- and A-tensors components for a transition metal complex: [Co(acacen)] or N,N-ethylenebis(acetylacetoneiminato)cobalt (II) , as obtained from two methods: by the LFDFT and by calculating the Zeeman splitting of the ground Kramers doublet obtained by using ZORA calculation and second-order perturbation theory implemented in ADF for the hyperne interaction.130, 132 The choice of this complex was motivated by two reasons: rstly, the [Co(acacen)] complex belongs to a series of complexes with tetradentate Schi bases which received much attention during the last decades,133, 134 largely because of their ability to reversibly absorb molecular oxygen under certain conditions. Secondly, it is a planar molecule (ligands are present only on a square planar environment) so it can be expected to show a 83

CHAPTER 6. ESR PARAMETERS WITHIN LFDFT large anisotropy of the ESR parameters. In the next sections, a description of the methodological and computational details is given. We show then, that with two methods, LFDFT and ZORA, this anisotropy is nicely reproduced. However, an analysis of the various contributions to the g- and Atensors is only possible using the LFDFT method.

6.1
6.1.1

Methods
LFDFT: a New Way to Calculate ESR Parameters

As we know well the LFDFT method now, we will just give the details of the calculation of the g- and A-tensor components. The program which contains the implementation of these calculations is a modied version of the LF matlab program and is proposed in Appendix E.2. 6.1.1.1 Calculation of g-tensor

For TM complexes with light atoms such as carbon, nitrogen and oxygen, the spin-orbit coupling on the ligand can be completely neglected. The methodology we consider in the preceding section is based on LF theory, it is therefore justied to express the spin-orbit interaction of the whole d- or f-manifold by a single parameter nl , i.e.: s ms , a | HSO |s ms , b nl s ms , m (a, )| l s |s ms , m (b, ) (6.1)

where: n = Rn | 1 dV |Rn and |lml > are real spherical harmonics. In practice n is r dr evaluated from the SO splitting obtained by a DFT spin-orbit ZORA caculation on the free ion ( = 598cm1 ) and reduce further it by the orbital reduction factor (k = 0.77, cf. Eq 4.7). The ground state Kramers doublet |0 is obtained by diagonalization d d of the full conguration interaction matrix: SD hLF + gER + hSO SD , where hLF and hSO have been dened previously32 and where gER represents the electrostatic inter 32 electronic repulsion. The g-tensor can be calculated from the equation for the Zeeman matrix elements 0| k L + ge S |0 equating them with those of the spin-hamiltonian ef f | g S | , i.e.: |+
1 (g 2 x 1 g 2 z

|
1 (g + igy ) 2 x 1 (gz ) 2

igy )

|0+ 0+| k L + ge S |0+ 0| k L + ge S |0+

|0 0+| k L + ge S |0 0| k L + ge S |0 (6.2)

or gz = 0+|k L + ge S |0+ 0|k L + ge S |0 gx = 0+|k L + ge S |0 + 0|k L + ge S |0+ gy = i ( 0| k L + ge S |0+ 0+| k L + ge S |0 ) 84

(6.3)

6.1 Methods where: k is the orbital reduction factor used to scale the spin-orbit coupling constant of the free ion, = x, y, z and L and S are the orbital and spin-angular momentum operators. For n-electrons, we have:
n

L =
i=1 n

li

(6.4)

S =
i=1

si

(6.5)

6.1.1.2

Calculation of A-tensor

The ligand eld description of the hyperne interaction is already well described in literature,31, 135 but we summarize it briey here. The interaction between the nuclear and electrons angular momenta of a many-electron system is described by the hyperne coupling Hamiltonian: HHF = HF I (6.6) where HF , the hyperne coupling operator is given by equations 6.7 and 6.8, summation being carried out over all electrons (see Ref 31 for more details):
n

HHF = P
i=1

1 li + ai si I 7

(6.7)

and ai = 4si (li si )li li (li si ) (6.8) The rst term corresponds to the interaction of the nuclear spin with the orbital angular momentum of the electron, the second term to the interaction of the nuclear spin with the electronic spin and the last term is the Fermi contact term. P in equation 6.7 is the electron-nuclear dipolar coupling constant dened as: P = ge N r3
3d

(6.9)

where N is the giromagnetic ratio of the nucleus N (for Co: N = 0.63171 104 G1 ),136 the bohr magneton and r3 3d the expectation value of the 1/r3 operator over the 3d wavefunction. The Parameter is related with the Fermi hyperne coupling constant aF (ge = 2.0023): 8 aF = ge N [i (0) i (0)] (6.10) 3 i as = Direct substitution yields: = 8 3 [i (0) i (0)]
i

aF P

(6.11)

1 r3
3d

(6.12)

85

CHAPTER 6. ESR PARAMETERS WITHIN LFDFT with N in G1 , P in cm1 and r3

3d

in atomic units, we have explicitly:

3d

P = ge N 332.5258 109 r3

(6.13)

Finally, the A-tensor can be calculated, similarly to the g-tensor, from the aforementioned ground Kramers doublet |0 evaluating the hyperne matrix elements 0| hf |0 as: Az = 0+|hf |0+ 0|hf |0 Ax = 0+|hf |0 + 0|hf |0+ Ay = i 0|hf |0+ 0+|hf |0 (6.14)

where = x, y, z and k is the orbital reduction factor used to scale the spin-orbit coupling constant of the free ion and is determined according to Eq 4.7. We now consider the numerical evaluations of P and which are used in the calculation of the A-tensor components. The value of P has been calculated using TZP exponents given by the ADF data base and a numerical integration resulting to r3 3d = 5.83 cm3 for the free ion: Co2+ . The corresponding value: P = 245 104 cm1 has been further reduced according to Eq 4.7 to P = 188 104 cm1 . The value of aF needed to obtain (Eq 6.11) is more subtle to evaluate and deserves a bit of explanation. In fact there is no Fermi interaction within dn conguration. However, the analysis of the hyperne structure of supposedly d-electrons clearly requires an isotropic contribution denoted here as aF . This term has two well-known contributions: (i) the spin polarization of the closed |ns shells through the unpaired electron occupying the ground Kramers-doublet and (ii) the direct admixture of |4s into the ground Kramers doublet through spin-orbit coupling with the low-lying |d12 , A1 excited state that is contaminated z by orbital interaction with |4s of Co because of the low symmetry of [Co(acacen)]. Indeed, this mixing amounts to 4%. The rst contribution to aF is easily obtained using the DENSF utility program of ADF to compute i (0) i (0) (Eq 6.10) from an all-electron spin-polarized KS-calculation of [Co(acacen)]. The second contribution to aF is obtained from the weight of all SD that contain dz2 present in the ground Kramers doublet. The resulting parameters aF = 28 104 and P = 188 104 cm1 lead to = 0.147 which was used in the calculation of the A-tensor. Table 6.1 summarizes all non-empirically determined parameters used in the calculation of the ESR and multiplet ne structure.

6.1.2

Spin-Orbit ZORA Approach

We use the ZORA implementation available in the ADF code to calculate g- and Atensors in order to compare the results with those from LFDFT and with experiment. This method, developed by van Lenthe et al.,130, 132 uses GIAO, where the g- and A-tensors are calculated in a spin-orbit relativistic calculation using ZORA Hamiltonian with a spin restricted wavefunction.

6.2

Computational Details

The DFT calculations were performed with the Amsterdam Density Functional (ADF) program package (release 2003.01).74 Both the local density approximation (LDA) and 86

6.2 Computational Details

Table 6.1: All nonempirically determined parameters used in the calculation of the g- and Atensors and in the calculation of multiplet ne structure

Racahs parameters Ligand Field matrix elements

spin-orbit coupling constant orbital reduction factor electron-nuclear dipolar coupling constant

B C x y |hLF |x y y z |hLF |y z z 2 |hLF |z 2 x 2 y 2 |hLF |x 2 y 2 z 2 |hLF |x 2 y 2 x z |hLF |x z k P

512 cm1 3118 cm1 1071 cm1 6308 cm1 5052 cm1 3731 cm1 2771 cm1 24003 cm1 460 cm1 0.77 188 104 cm1

0.147

the generalized gradient approximation (GGA) for exchange-correlation functionals were used. The LDA was applied with the Vosko, Wilk, and Nusair functional57 and the GGA by using the exchange-correlation Perdew-Wang 91 functional.111 The atoms were described by a triple- Slater Type Orbital (STO) basis sets plus one polarization function and the frozen core (FC) approximation was used up to 3p for cobalt and up to 1s for carbon, nitrogen and oxygen for the g-Tensor calculation and the LFDFT calculations. For the calculation of A-tensor, we used all electrons basis sets.

Figure 6.1: The axial coordinates of the system in the discussion (x, y, z) and in the ADF calculations (x , y , z ) are represented along with the [Co(acacen)].

87

CHAPTER 6. ESR PARAMETERS WITHIN LFDFT

Table 6.2: Multiplet splitting energies determine by LFDFT method using GGA functional and frozen core approximation and compare to experiment (in cm1 ).

A2 A1 2 B1 2 A1 4 B1 4 A1 4 B1
2

LFDFT Exp. 0.0 4665 7036 4000 10885 8000 13021 12835 14694 -

The geometry of the complex has been adopted from X-ray crystallographic data determined by Cariati et al.137 Since the deviation from C2v symmetry is not signicant we have chosen to impose this symmetry in our calculations. The coordinate system of Co(acacen) which by convention, has always been used to discuss this type of complexes is shown in Figure 6.1. In this way the molecule belongs actually to the point group C2v (x), with x as the principal symmetry axis instead of the more conventional z-axis. The d-orbitals have hereby the following symmetry labels: dz2 and dx2 y2 , a1 ; dxy , b2 ; dxz , b1 ; dyz , a2 . In the ADF calculations another orientation has been adopted (x , y , z : Fig 6.1) with two-fold axis along z. However, results have been always converted back to the traditional one (x, y, z).

6.3

Results and Discussion

The Co(acacen) with a d7 -conguration for CoII has low-spin S=1/2 ground state. It shows a large anisotropy of both the g- and A-tensors (Tables 6.3 & 6.4) related to the low (C2v ) symmetry of the Co2+ coordination centers. A MO-diagram comprising MOs dominated by the 3d-atomic orbitals (Fig. 6.2) shows the typical splitting for square planar coordination with the -antibonding dxy (b2 )-orbital, separated by about 27.4 kK from the weaker - dz2 , the in-plane - dx2 y2 (both of a1 symmetry) and the outof-plane yz (a2 ) and xz (b1 ) orbitals. The latter ones are much less separated in energy, covering a narrow range of energies (7.4 kK), as shown in Fig. 6.2. The dxz and dyz orbitals which are degenerate in a square planar complex (eg , D4h symmetry) are found in [Co(acacen)] to be considerably split (7.0 kK), this being responsible for large in-plane anisotropy of the main values of the g- and A-tensors. The underlying cause of this large splitting can be understood in the context of the molecular orbital model. Restricting to the highest occupied (a2 ) (HOMO) and lowest unoccupied *(b1 ) (LUMO) ligand orbitals, we notice that the corresponding 3d-orbitals with the same symmetry, dyz (dxz ) become destabilized (stabilized) by the interplay of -donation (-back donation). The consequences of this anisotropic -bonding eects have been accounted for in rened LF models, such as the phase coupling concept of Ceulemans et al.,138140 quantifying an early idea of Orgel.141 However, quantitatively, a large number of ill dened model parameters had to be introduced in order to account for this eect. Our LFDFT results lend full support of the -anisotropy, as reected by the calculated LF transitions (Table 6.2) and the calculated g- and A-tensors (Tables 6.3 & 6.4). 88

6.3 Results and Discussion

kK
30

17b2

25

20

15

10

7a2
5

19a1 7b1 18a1

Figure 6.2: KS MO diagram for the ground state conguration.

89

CHAPTER 6. ESR PARAMETERS WITHIN LFDFT ZORA LDA GGA 2.85 2.76 1.89 1.93 1.91 1.92 2.22 2.20 LFDFT GGA EXP134 A B 3.21 2.80 2.92/3.26 1.87 1.94 1.90(0.03) 1.87 2.11 2.00(0.02) 2.28 2.32

gxx gyy gzz giso

Table 6.3: g-tensor values of [Co(acacen)] determined by spin-orbit restricted spin-orbit ZORA calculation and the LFDFT approaches and compared to a set of experimental data. LFDFT column A corresponds to the two states model: 97% |d1 d2 , 2 A2 + 3% |d12 d2 ,2 A1 , and yz xy z xy column B to the full calculation. For the expermimental data, we give a range of values because of a strong dependency upon the host lattice.

Axx Ayy Azz

ZORA LDA GGA 151 108 25 28 66 71

LFDFT GGA EXP134 A B 95 55 100/128 12 14 32/40 39 19 29/34

Table 6.4: A-tensor values of [Co(acacen)] determined by spin-orbit restricted spin-orbit ZORA calculation and the LFDFT approaches and compared to a set of experimental data. LFDFT column A corresponds to the two states model: 97% |d1 d2 , 2 A2 + 3% |d12 d2 ,2 A1 , and yz xy z xy column B to the full calculation. For the expermimental data, we give a range of values because of a strong dependency upon the host lattice.

With 7 electrons on the closely spaced a2 , a1 , b1 and a1 -orbitals we have a |d1 ,2 A2 yz ground state. Mixing with excited states via spin-orbit coupling then leads to the observed anisotropic g- and A-tensors. These are given in Tables 6.3 & 6.4, where we also include results of ZORA spin-orbit calculations. Both the ZORA and the LFDFT show largest values for gxx and Axx . However, ner details in the g-tensor anisotropy, in particular the gzz > gxx relationship are better reproduced the LFDFT method. As far as the overall agreement between theoretical and experimental A-tensor components goes, both the ZORA and the LFDFT calculations are of comparable moderate quality. However, we must keep in mind that experimental values of Axx , Ayy and Azz are less accurately determined. Thus, coordination to more distant atoms in the solid seems to make important contributions.134 However, this is beyond the scope of the present study. Finally, we should mention that models of g and A-tensors, conned to one or two excited states (resulting in two-state model |d1 , 2 A2 -|d12 , 2 A1 ,134 or three state model |d1 , 2 A2 yz yz z |d12 , 2 A1 -|d1 , 2 B1 ,139, 140 respectively) reect essential features of the physical origin xz z of the anisotropy. In Tables 6.3 & 6.4 we list the results from a consideration using a two-state model. The overall features are reasonably reproduced using this simplied approach. However, our present calculation enables a more detailled description of the experimental ndings, particularly the dierence between gyy and gzz for the g-tensor, and important contributions to Axx from other excited states, ignored by the two or three states models are taken into account. 90

6.4 Conclusions

6.4

Conclusions

In this chapter we have extended our LFDFT approach with ESR ne structures, demonstrating that the model works, even in such complicated situations as the Co(acacen) complex. No doubt, the method can be rened including anisotropic covalent reduction factors and anisotropic spin-orbit coupling and inter-electronic repulsion. However we have found that calculating all these quantities using an average-of-conguration concept, thus introducing non-empirical atomic like B, C and parameters and further, a single overall parameter k-in order to account for covalent reduction is a reasonable approximation, capable of describing electronic transitions and multiplet ne structures.

91

CHAPTER 6. ESR PARAMETERS WITHIN LFDFT

92

Chapter 7 Conclusions and Outlook

7.1 Conclusions

We have developed a DFT based LF model which utilizes SD energies in order to determine ligand eld parameters from rst principles. The formalism has been implemented using the following recipe: 1. make a geometry choice; for neutral complexes we recommend a GGA geometry optimization. For complexes with negative charges use of experimental bond lengths or LDA geometry optimization is preferable, 2. construct KS-LF orbitals with the average of conguration (AOC) providing n/5 occupancy of each d-MO for a TM with a dn conguration, 3. calculate the set of all SD energies using these orbitals without allowing for orbital relaxation (no SCF iteration) 4. determine B,C and the 5x5 LF matrix hab from these data, 5. introduce these parameters into a favourite LF -program to get all multiplet energies. The rather consistent t for cubic symmetry reproducing the total manifold of symmetry independent SD shows, that DFT and LF theory are compatible. A theoretical justication (analysis) of this result is still lacking (but see the discussion in Ref.84 ). Comparing LF parameters deduced from DFT calculations with those resulting from t to experimental spectra in high resolution, we can conclude that existing functionals are able to describe properly not only cubic but also low-symmetric ligand elds. In contrast parameters of interelectronic repulsion are calculated systematically smaller than values from spectral data. Similar observations have been reported by Solomon et al..142 Moreover, in Chapter 3, comparing results for both 10Dq and the parameters of interelectronic repulsion we can conclude, based on DFT considerations, that both the eect of the ligands in regards of spectrochemical series and the nephelauxetic eect are well reproduced by LFDFT. On the basis of these results we can conclude that DFT provides a rigorous interpretation of the LF parameters and leads to a justication of the parametric structure of the classical LF theory.

93

CHAPTER 7. CONCLUSIONS AND OUTLOOK Thus ne structure tensors in EPR became a good candidate for application using this approach. In Chapter 6, we proove the ability of the method to determine the value of g and A-tensors. Modern functionals which are all based on quantum Monte-Carlo treatment of a homogeneous electron gas contain most of the dynamical correlation. The non-dynamical or near degeneracy correlation however is missing. The key feature of our approach is the explicit treatment of near degeneracy correlation using adhoc Conguration Interaction (CI) within the active space of Kohn-Sham (KS) orbitals with dominant d-character. The calculation of the CI-matrices is based on a ligandeld analysis of the energies of all single determinants (micro-states) calculated according to Density Functional Theory (DFT) for frozen KS-orbitals corresponding to the averaged congurationof the d-orbitals. This procedure yields multiplet energies with an accuracy within 2000 cm1 . The accuracy of this approximation can be judged from the agreement between calculated and observed transition energy.32, 35

7.2

Outlook

We can use these results in order to motivate further work aimed at developing functionals and/or basis sets for spectroscopic purposes. But another aim is to extend the domain of applications of this method and we can cite for examples: 1. treat complexes with Lanthanides or Actinides (f -shell): progress in this direction are already done since rst results are published or are in press.2628 But further work should be done because up to now, real description of the LF for these complexes can be obtained essentially by using the Kohn-Sham Equations with Constrained Electron Density (KSCED) method.143 2. model magnetic susceptibility: magnetic susceptibility has been described by Grifth and to model it should be a good extension to the LFDFT method. 3. use meta-GGA functional: up to now, the meta-GGA functional can only be used as a post-SCF procedure in ADF, as we need to run SCF procedure on the AOC, we can not use this functional for the LFDFT but it will be interesting to see, if in the future we are able to use it in an SCF procedure, the results we will obtain for our LFDFT treatment. 4. treat TM complexes including more than one TM: up to now, we focus on complexes with one TM but more often, we have to model properties for crystals and extend the LFDFT to, in the rst time, dinuclear complex can be a good way to extend step by step the eld of application of this method. 5. model MCD spectra of biological metal complexes: the program Gener ln accept in the version which is presented in this thesis for the l value an integer like we used it along this thesis but can also accept an array, so up to now, we can generate the matrices we need in our method to model MCD spectra. So one has to extend the LFDFT matlab program to t values of LF parameters for system which have more than one value of l. 94

7.2 Outlook These are only some suggestions and the list is not exhausive.

95

CHAPTER 7. CONCLUSIONS AND OUTLOOK

96

Bibliography
[1] Solomon E.I. and Hodgson K.O. Spectroscopic Methods in Bioinorganic Chemistry. Oxford University Press, 1998. [2] Bertini I., Gray H.B., Lippard S.J., and Valentine J.S. Bioinorganic chemistry. University Science, 1994. [3] Kohn W. and Sham L.J. Self-consistent equations including exchange and correlation eects. Phys. Rev., 140, p:A1133, 1965. [4] Theophilou A. The energy density functional formalism for excited states. J. Phys. C., 12, p:5419, 1979. [5] Gross E.K.U., Oliveira L.N., and Kohn W. Rayleigh-ritz variational principle for ensembles of fractionally occupied states. Phys. Rev. A, 37, p:2805, 1988. [6] Gross E.K.U., Oliveira L.N., and Kohn W. Density-functional theory for ensembles of fractionally occupied states. i. basic formalism. Phys. Rev. A, 37, p:2809, 1988. [7] Gross E.K.U., Oliveira L.N., and Kohn W. Density-functional theory for ensembles of fractionally occupied states. ii. application to the he atom. Phys. Rev. A, 37, p:2821, 1988. [8] Slater J.C. Adv. Quantum Chem., 6, p:1, 1972. [9] Messmer R.P. and Salahub D.R. J. Chem. Phys., 65, p:779, 1976. [10] Ziegler T., Rauk A., and Baerends E.J. Theor. Chim. Acta, 43, p:261, 1977. [11] Daul C.A. Int. J. Quantum Chem., 52, p:867, 1994. [12] Gilardoni F., Weber J., Bellafrouh K., Daul C.A., and Guedel H.U. J. Chem. Phys., 104, p:7624, 1996. [13] Aramburu J.A., Moreno M., Doclo K., Daul C.A., and Barriuso M.T. J. Chem. Phys., 110, p:1497, 1999. [14] Deghoul F., Chermette H., Rogemond F., Moncorge R., Stueckl C., and Daul C.A. Phys. Rev. B, 60, p:2404, 1999. [15] Dunlap B.I. Adv. Chem. Phys., 69, p:287, 1987. [16] Lwdin P.-O. Adv. Chem. Phys., 14, p:283, 1969. o 97

BIBLIOGRAPHY [17] Goerling A. Symmetry in density-functional theory. Phys. Rev. A, 47, p:2783, 1993. [18] Sala F.D. and Goerling A. Open-shell localized hartree-fock approach for an ecient eective exact-exchange kohn-sham treatment of open-shell atoms and molecules. J. Chem. Phys., 118, p:10439, 2003. [19] Mineva T., Goursot A., and Daul C.A. Atomic multiplet energies from density functional calculations. Chem. Phys. Lett., 350, p:147154, 2001. [20] Bethe H. Ann. Phys., 3, p:133, 1929. [21] Ilse F.E. and Hartmann H. Z. Phys., 197, p:239, 1951. [22] Hartmann H. and Schlaefer H.L. Z. Naturforsch., 6a, p:751, 1951. [23] Hartmann H. and Schlaefer H.L. Z. Phys., 197, p:239, 1951. [24] Jrgensen C.K., Pappalardo R., and Schmidtke H. J. Chem. Phys., 39, p:1422, 1963. [25] Schaeer C.E. and Jrgensen C.K. Mol. Phys., 9, p:401, 1965. [26] Borel A., Helm L.E., and Daul C.A. Hybrid ligand-eld theory/quantum chemical calculation of the ne structure and zfs in lanthanide(iii) complexes. Chem. Phys. Lett., 383, p:584591, 2004. [27] Zbiri M., Atanasov M., Daul C.A., Garcia-Lastra J., and Wesolowski T.A. Chem. Phys. Lett. in press. [28] Atanasov M., Daul C.A., Guedel H.U., Wesolowski T.A., and Zbiri M. Inorg. Chem. in press. [29] van Vleck J.H. Valence strength and the magnetism of complex salts. J. Chem. Phys., 3, p:807, 1935. [30] Mulliken R.S. Phys. Rev., 40, p:55, 1932. [31] Grith J.S. The Theory of Transition-Metal Ions. Cambridge University Press, 1971. [32] Atanasov M., Daul C.A., and Rauzy C. Struct. Bond., 106, p:97125, 2004. [33] Slater J.C. Quantum Theory of Atomic Structure, Vol. 2. McGraw-Hill, London, 1960. [34] Misetich A.A. and Buch T. J. Chem. Phys., 41, p:2524, 1964. [35] Atanasov M., Daul C.A., and Rauzy C. Chem. Phys. Lett., 367, p:737746, 2003. [36] Jrgensen C.K. Struct. Bond., 1, p:331, 1966. [37] Schaeer C.E. and Jrgensen C.K. J. Inorg. Nucl. Chem., 8, p:143, 1958. 98

BIBLIOGRAPHY [38] Parr R.G. and Yang W. Density Functional Theory of Atoms and Molecules. New York, Oxford University Press, 1998. [39] Koch W. and Holthausen M. A chemists Guide to Density Functional Theory. New York, Wiley-VCH, 2000. [40] Dreizler R.M. and Gross E.K.U. Density Functional Theory, an approach to Quantum Many-Body Problem. New York, Springer-Verlag, 1990. [41] Labanowsky J.K. and Andzelm J.W. Density Functional Methods in Chemistry. Springer-Verlag, 1991. [42] Fermi E. Rend. Accad. Lincei, 6, p:602, 1927. [43] Thomas H. Z. Phys., 48, p:73, 1928. [44] Thomas H. Proc. Camb. Phil. Soc., 23, p:542, 1927. [45] Hohenberg P. and Kohn W. Phys. Rev. B, 136, p:864, 1964. [46] Teller E. Rev. Mod. Phys., 34, p:627, 1962. [47] Dirac P.A.M. Proc. Camb. Phil. Soc., 26, p:376, 1930. [48] von Weizsacker C.F. Z. Phys., 96, p:431, 1935. [49] Yonei K. and Tomishima Y. J. Phys. Soc. Japan, 20, p:1051, 1965. [50] Lieb E.H. Rev. Mod. Phys., 53, p:603, 1981. [51] Yang W. Phys. Rev. A, 34, p:4575, 1986. [52] Brack M. Density Functional Methods in Physics, page 331. demic/Plenum Publishers, 1985. [53] Levy M. Proc. Natl. Acad. Sci, 76, p:6062, 1879. [54] Levy M. Phys. Rev. A, 26, p:1200, 1982. [55] Perdew J. Phys. Rev. B, 31, p:6264, 1985. [56] Ceperley D.M. and Alder B.J. Phys. Rev. Lett., 45, p:566, 1980. [57] Vosko S.H., Wilk L., and Nusair M. Can. J. Chem., 58, p:1200, 1980. [58] Becke A.D. Phys. Rev. A, 38, p:3098, 1988. [59] Colle R. and Salvetti O. Theor. Chim. Acta, 37, p:329, 1975. [60] Lee C., Yang W., and Parr R.G. Phys. Rev. B, 37(2), p:785, 1988. [61] Miehlich B., Savin A., Stoll H., and Preuss H. Chem. Phys. Lett., 157, p:200, 1989. [62] Becke A.D. J. Chem. Phys., 98, p:5648, 1993. 99 Kluwer Aca-

BIBLIOGRAPHY [63] Perdew J.P. and Kurth S. Phys. Rev. Lett., 82, p:2544, 1999. [64] Proynov E., Chermette H., and Salahub D.R. J. Chem. Phys., 113, p:10013, 2000. [65] Ernzerhof M. and Scuseria G.E. J. Chem. Phys., 111, p:911, 1999. [66] van Voorhis T. and Scuseria G.E. J. Chem. Phys., 109, p:400, 1998. [67] Levy M. and Perdew J. Phys. Rev. A, 32, p:2010, 1985. [68] Levy M. and Perdew J. Density Functional Methods in Physics, page 11. Kluwer Academic/Plenum Publishers, 1985. [69] Levy M. Single-Particle Density in Physics and in Chemistry, page 45. London, Academic Press, 1987. [70] Janak J.F. Phys. Rev. B, 18, p:7165, 1978. [71] Goerling A. Density functional theory for excited states. Phys. Rev. A, 54, p:3912, 1996. [72] Nagy A. Excited states in density functional theory. Int. J. Quantum Chem., 70, p:681, 1998. [73] Nagy A. Kohn-sham equations for multiplets. Phys. Rev. A, 57, p:1672, 1998. [74] te Velde G., Bickelhaupt F.M., Baerends E.J., Fonseca Guerra C., van Gisbergen S.J.A, Snijders J.G., and Ziegler T. J. Comput. Chem., 22, p:931967, 2001. [75] Perdew J.P., Chevary J.A., Vosko S.H., Jackson K.A., Pederson M.R., Singh D.J., and Fiolhais C. Atoms, molecules, solids, and surfaces: Applications of the generalized gradient approximation for exchange and correlation. Phys. Rev. B, 46, p:6671, 1992. [76] Perdew J.P., Burke K., and Ernzerhof M. Phys. Rev. Lett., 77, p:3865, 1996. [77] Hammer B., Hansen L.B., and Norskov J.K. Phys. Rev. B, 59, p:7413, 1999. [78] Zhang Y. and Yang W. Phys. Rev. Lett., 80, p:890, 1998. [79] Johnson B.G., Gill P.M.W., and Pople J.A. J. Chem. Phys., 98(7), p:5612, 1993. [80] Russo T.V., Martin R.L., and Hay P.J. J. Chem. Phys., 101(9), p:7729, 1994. [81] Perdew J.P. Density-functional approximation for the correlation energy of the inhomogeneous elctron gas. Phys. Rev. B, 33, p:8822, 1986. [82] van Leeuwen R. and Baerends E.J. Exchange-correlation potential with correct asymptotic behaviour. Phys. Rev. A, 49(4), p:24212431, 1994. [83] Ziegler T. Chem. Rev., 91, p:651, 1991. [84] Bridgeman A.J. and Gerloch M. Chem. Phys. Lett., 45, p:179, 1997. 100

BIBLIOGRAPHY [85] Durand P.H. and Malrieu J.P. Adv. Chem. Phys., 67, p:321, 1987. [86] Des Cloizeaux J. Extension dune formule de Lagrange ` des probl`mes de valeurs a e propres. Nucl. Phys., 20, p:321, 1960. [87] Shigekazu U., Kazuyori U., and Keiichi K. Semento Gijetsu Nenpo, 29, p:32, 1975. [88] Ferguson J. and Wood D.L. Crystal eld spectra of d3,7 ions. vi the weak eld formalism and covalency. Austr. J. Chem., 23, p:861, 1970. [89] Adamsky H. PhD thesis, University of Dsseldorf, 1995. AOMX is available in the u Internet at the address http://www.aomx.de This site provides selected examples and an AOM parameter database, and will be further developed to a forum for scientic exchange about Angular Overlap Model applications to TM complexes. [90] Daul C.A., Doclo K.G., and Stueckl C. Recent Advances in Density Functional Methods (Part II), page 61. World Scientic, 1997. [91] Ferguson J., Wood D.L., and Knox K. J. Chem. Phys., 39, p:881, 1963. [92] Ludi A. and Feitknecht W. Helv. Chim. Acta, 46, p:2226, 1963. [93] Allen G.C., El-Sharkawy A.M., and Warren K.D. Inorg. Chem., 10, p:2538, 1971. [94] Schwartz R.W. Inorg. Chem., 15, p:2817, 1976. [95] Witzke H. Theor. Chim. Acta, 20, p:171, 1971. [96] Knox K. and Mitchell D.W. J. Inorg. Nucl. Chem., 21, p:253, 1961. [97] Jagner S., Ljungstroem E., and Vannerberg N.G. Acta Chem. Scand., A28, p:623, 1974. [98] Gupta M.P., Milledge H.J., and McCarthy A.E. Acta Cryst. B., 30, p:656, 1974. [99] Mukherjee R.K. and Chowdhury M. Chem. Phys. Lett., 34, p:178, 1975. [100] Alexander J.J. and Gray H.B. J. Am. Chem. Soc., 90, p:4260, 1968. [101] Lever A.B.P. Inorganic Electronic Spectroscopy, page 450. Elsevier, 2 edition, 1984. [102] Cohen S., Mayer I., and Reinen D. J. Solid State Chem., 107, p:218, 1993. [103] Brunold T.C., Hauser A., and Guedel H.U. J. Luminescence, 59, p:321, 1994. [104] Hazenkamp M.F., Guedel H.U., Atanasov M., Kesper U., and Reinen D. Phys. Rev. B, 53, p:2357, 1996. [105] Shannon R.D. Acta Cryst, A32, p:751, 1976. [106] Studer P. PhD thesis, University of Fribourg, 1975. [107] Grimme S. Chem. Phys. Lett., 259, p:128, 1996. 101

BIBLIOGRAPHY [108] Grimme S. and Waletzke M. J. Chem. Phys., 111, p:5645, 1999. [109] van Lenthe E., Baerends E.J., and Snijders J.G. J. Chem. Phys., 99, p:4597, 1993. [110] Slater J.C. A simplication of the hartree-fock method. Phys. Rev., 81, p:385, 1951. [111] Perdew J.P., Chevary J.A., Vosko S.H., Jackson K.A., Pederson M.R., Singh D.J., and Fiolhais C. Phys. Rev. B, 46, p:66716687, 1992. [112] Bruyndonckx R., Daul C.A., Manoharan P.T., and Deiss E. Inorg. Chem., 36, p:42514256, 1997. [113] Reinen D., Atanasov M., Nikolov G.S., and Steens F. Inorg. Chem., 27, p:1678 1686, 1988. [114] Pauling P. Inorg. Chem., 5, p:14981505, 1966. [115] Stucky C.D., Folkers J.B., and Kistenmacher T.J. Acta Cryst, 23, p:1064, 1967. [116] Liehr A.D. and Ballhausen C.J. Ann. Phys., 6, p:134, 1959. [117] Mooney A., Nuttal R.H., and Smith W.E. J. Chem. Soc., Dalton Trans., page 1920, 1973. [118] Koester V.J. and Dunn T.M. Inorg. Chem., 14, p:1811, 1975. [119] Collingwood J.C., Day P., and Dennning R.G. J. Chem. Soc. Faraday Trans. II, 69(4), p:591607, 1973. [120] Gerloch M. and Slade R.C. Paramagnetic susceptibilities and anisotropies of nickel(II) ions in tetrahedral environments. I. general behavior of a spin-triplet model. J. Chem. Soc. (A), 6, p:10121022, 1969. [121] Gerloch M. and Slade R.C. Paramagnetic susceptibilities and anisotropies of nickel(II) ions in tetrahedral environments. II. covalency and distortion in tetraethylammonium tetrachloronickelate(II), and bis(nisopropylsalicylaldiminato)nickel(II). J. Chem. Soc. (A), 6, p:10221028, 1969. [122] Liehr A.D. J. Phys. Chem., 64, p:4351, 1960. [123] Daul C. and Weber J. Helv. Chim. Acta, 65, p:2486, 1982. [124] Schreckenbach G. and Ziegler T. J. Phys. Chem. A, 101, p:3388, 1997. [125] Bruyndonckx R. Density Functional Theory as a Practical Tool to Calculate Molecular Properties: EPR and Jahn-Teller parameters. PhD thesis, University of Fribourg (Switzerland), 1999. [126] Belanzoni P., Baerends E.J., van Asselt S., and Langewen P.B. J. Chem. Phys., 99, p:13094, 1995. [127] van Leeuwen R., van Lenthe E., Baerends E.J., and Snijders J.G. J. Chem. Phys., 101, p:1272, 1994. 102

BIBLIOGRAPHY [128] van Lenthe E., Baerends E.J., and Snijders J.G. J. Chem. Phys., 101, p:9783, 1994. [129] van Lenthe E., Snijders J.G., and Baerends E.J. J. Chem. Phys., 105, p:6505, 1996. [130] van Lenthe E., Wormer P.E.S., and van der Avoird A. J. Chem. Phys., 107, p:2488, 1997. [131] Belanzoni P., Van Lenthe E., and Baerends E.J. J. Chem. Phys., 114, p:4421, 2001. [132] van Lenthe E., van der Avoird A., and Wormer P.E.S. J. Chem. Phys., 108, p:4783, 1998. [133] Calvin M. and Barkelew C. H. J. Am. Chem. Soc., 68, p:2267, 1946. [134] Daul C., Schlaepfer C.W., and A. von Zelewsky. Struct. Bond., 36, p:128, 1978. [135] Bencini A. and Gatteschi D. J. Magn. Reson., 34, p:653, 1979. [136] Wertz J.E. and Bolton J.R. Electron Spin Resonance, Elementary Theory and Practical Applications. McGraw-Hill, New York, 1972. [137] Cariati F., Morazzoni F., Busetto C., Del Piero G., and Zazzetta A. J. Chem. Soc., Dalton Trans., page 342, 1976. [138] Ceulemans A., Dendooven M., and Vanquickenborne L.G. Inorg. Chem., 24, p:1153, 1985. [139] Ceulemans A., Dendooven M., and Vanquickenborne L.G. Inorg. Chem., 24, p:1159, 1985. [140] Ceulemans A., Debuyst R., Dejehet F., King G.S.D., and Vanhecke M. J. Phys. Chem., 94, p:105, 1990. [141] Orgel L.E. J. Chem. Soc. (A), 6, p:3683, 1961. [142] Solomon E.I., Chen P., Metz M., Lee Angew. Chem. Int. Ed. Engl., 40, p:4570, 2001. S.-K., and Palmer A.E.

[143] Wesolowski T.A. and Weber J. Chem. Phys. Lett., 248, p:71, 1996. [144] Welch B.B., Jones K., and Hobbs J. Pratical programming in Tcl and Tk. Prentice Hall PTR, 4 edition, 2003. [145] Rose M.E. Elementary Theory of Angular Momentum. Dover Publications, Inc., 1995.

103

BIBLIOGRAPHY

104

Appendix A Mathematical background

A.1 Expansion of 1/rij

The ligand eld potential is expressed in function of j 1/|R r| where R dened the ligand position and r rj dened the electron position. The expansion of this rij term is analog to the expansion of 1/rij where i and j represent the positions of the electrons i and j respectively. We will proceed to the expansion of 1/|rij | in 0 i ri order to give the intrinsic details of ligand eld theory. So the norm of |rij | is equal to: Figure A.1: Position of the electrons i and j.

|rij | = |ri rj | =

2 ri

2 rj

2ri rj cos

(A.1)

where is the angle between the two vectors. If we dene x = rj /ri , Equation A.1 reads: |ri rj | = ri 1 + x2 2x cos (A.2)

So if we consider the inverse and we expand it in terms of Legendre polynomials we get the relation A.3: 1 1 = |rij | ri 1 = ri 1 1 + x2 2x cos an Pn (cos )
n

If we square the both side of this equation and multiply by sin d and integrate over from 0 to , we obtained 1 1+x 2 log = a2 (A.4) x 1x 2n + 1 n n But, by an expansion, we have: 1 1+x log = x 1x 2 x2n 2n + 1 (A.5)

105

(A.3)

APPENDIX A. MATHEMATICAL BACKGROUND so we conclude that an = xn and the expansion of 1/|rij | in terms of Pn (cos ) is: 1 1 = |rij | ri xn Pn (cos )
n

(A.6)

We must now express Pn (cos ) as a function of s and s of two particles and we may |m| expand it in terms of the orthogonal functions Pl (cosi )eimi . Pn (cos ) is a solution of the equation: 1 sin i i sin i Pn i + 1 2 Pn + n(n + 1)Pn = 0 sin2 i 2 i (A.7)

since this equation remains unchanged under any rotation. The general solution of this |m| equation is a linear combination of the functions Pn (cosi )eimi , so that we may express Pn (cos ) as:
n

Pn (cos ) =
m=n

|m| Anm Pn (cos i )eimi

(A.8)

where Anm =

2n + 1 (n |m|)! 4 (n + |m|)!

|m| Pn (cos )Pn (cos i )eimi d |k|

(A.9)

we also can expand Pn (cos i )eimi in terms of a function Pn (cos )eik

n |m| Pn (cos i )eimi = k=n |k| Bnk Pn (cos )eik

|m|

(A.10)

where Bnk =

2n + 1 (n |k|)! 4 n + |k|)!

|m| |k| Pn (cos i )eimi Pn (cos )eik d

(A.11)

Equation A.10 must hold for = 0, that is, for i = j and i = j . Then
n |m| Pn (cos j )eimj

=
k=n

|k| 0 Bnk Pn (1)eik = Bn0 Pn (1)

(A.12)

k 0 since Pn (1) = 0 for k = 0. Pn (1) = 1 from Eq A.11 & A.12 we have:

2n + 1 4 so that

|m| |m| Pn (cos i )eimi Pn (cos )d = Pn (cos j )eimi

(A.13)

Anm =

(n |m|)! |m| P cos(j )eimj (n + |m|)! n

(A.14)

Combining Eqs A.6, A.9 & A.14, we have the nal result: 1 = |rij |
n (n |m|)! rj |m| im(i j ) n+1 Pn cos(j )e (n + |m|)! ri n=0 m=n n

(A.15)

106

A.2 Spherical Harmonic Properties Eq. A.15 may alternatively be expressed as: 1 = |rij |
n rj 4 n+1 Yn,m (i , i )Yn,m (j , j ) 2n + 1 ri n=0 m=n n

(A.16)

A.2

Spherical Harmonic Properties

For chemical problem, we are concerned by real potential quantities so it is more convenient to work with real spherical harmonics, yl,m , which are dened in regards to complex spherical harmics as: 1 Yl,m (, ) + Yl,m (, ) if m > 0 2 yl,m (, ) = Yl0 (, ) i Yl,|m| (, ) Yl,|m| (, ) 2 if m = 0 if m < 0

(A.17)

Moreover, the real spherical harmonics can be also dened as: yl,m (, ) = N Pl,m (cos ) sin |m| if m < 0 cos m if m 0 (A.18)

with N such as yl,m (, ) = sqrt4. The complex spherical harmonics Yl,m and its complex conjugate Yl,m are linked by the following relation
Yl,m (, ) = (1)m Yl,m (, )

(A.19)
2 yl,m (, )d = 1.

And the normalization of the real spherical harmonics yl,m is such that If we look at the d-wavefunctions, the real forms will be: dz2 y2,0 (, ) = Y2,0 (, ) i dyz y2,1 (, ) = [Y2,1 (, ) + Y2,1 (, )] 2 1 dxz y2,1 (, ) = [Y2,1 (, ) Y2,1 (, )] 2 i dxy y2,2 (, ) = [Y2,2 (, ) Y2,2 (, )] 2 1 dx2 y2 y2,2 (, ) = [Y2,2 (, ) + Y2,2 (, )] 2

(A.20)

A.3

Spherical Harmonic Addition Theorem

We will attempt to demonstrate the spherical harmonic theorem. For this, we have to recall the denition of a rotation: (x , y , z ) = R(x, y, z) and more particulary a rotation through the Euler angles , and : Rjm =
m j Dm m ()jm

(A.21)

(A.22)

107

APPENDIX A. MATHEMATICAL BACKGROUND Now if we consider the expression: g=

m Ylm (1 , 1 )Ylm (2 , 2 )

(A.23)

where (1 , 1 ) and (2 , 2 ) are the spherical coordinates of two points P1 and P2 on the unit sphere and we rotate the coordinate system through the Euler angles, we can write: g=
m [R1 Ylm (1 , 1 )] R1 Ylm (2 , 2 )

(A.24)

where the primes indicates the spherical coordinates in the rotated frame. Because j Dmm () is unitary, the Eq A.24 can be rewritten as R1 jm =
m j Dmm (, , )jm

(A.25)

which means that: g=

m1 m2 m l l Dmm2 Dmm1 Ylm1 (1 , 1 )Ylm2 (2 , 2 )

(A.26)

But the sum over m is, by the unitary nature of the rotation matrix, simply m1 m2 . Therefore we have shown that g is independant of coordinate system: g=
m Ylm1 (1 , 1 )Ylm2 (2 , 2 ) = m Ylm1 (1 , 1 )Ylm2 (2 , 2 )

(A.27)

i.e., g is invariant under rotations. We can evaluate g in any coordinate system, and the one we now choose is such that P1 is on the z-axis and 2 = 0; i.e., the xz-plane is dened as the plane containing P1 , P2 and the origin. Thus 1 = 0, and 2 = , 2 = 0, being the angle between the radius vectors of the two points. Since: Ylm (, 1 ) = m0 So g= 2l + 1 4
1/2

2l + 1 4

1/2

(A.28)

Yl0 (, 0)

(A.29)

Finally, the spherical harmonic addition theorem is obtained: Yl0 (, 0) = 4 2l + 1

1/2 Ylm (1 , 1 )Ylm (2 , 2 ) m

(A.30)

But we use more often the expression: Yl0 (, 0) = 4 2l + 1 108

1/2

Pl (cos )

(A.31)

A.4 Wigner-Eckart Theorem where Pl is the Legendre polynomial, and write the spherical harmonic addition theorem as: 4 Pl (cos ) = (1)m Yl,m (1 , 1 )Yl,m (2 , 2 ) 2l + 1 m=l 4 = Yl,m (1 , 1 )Yl,m (2 , 2 ) 2l + 1 m=l = Pl (cos 1 )Pl (cos 2 ) + 2
m=l l l

(l m)! m P (cos 1 )Plm (cos 2 ) cos[m(1 2 )] (l + m)! l (A.32)

where is dened by: cos cos 1 cos 2 + sin 1 sin 2 cos(1 2 ) (A.33)

A.4

Wigner-Eckart Theorem

A theorem of fundamental importance in spectroscopy and angular momentum theory which provides both (1) an explicit form for the dependence of all matrix elements of irreducible tensors on the projection quantum numbers and (2) a formal expression of the conservation laws of angular momentum (Rose 1995). The theorem states that the dependence of the matrix element j m |TLM |jm on the projection quantum numbers is entirely contained in the Wigner 3j-symbol (or, equivalently, the Clebsch-Gordan coecient), given by: j m |TLM |jm = C(jLj ; mM m ) j ||TL ||j (A.34) where C(jLj ; mM m ) is a Clebsch-Gordan coecient and TLM is a set of tensor operators. The quantity j ||TL ||j is called the reduced matrix element of the set of tensor operators TLM .

A.5

Clebsch-Gordan Coecient

Clebsch-Gordan coecients are mathematical symbol used to integrate products of three spherical harmonics. Clebsch-Gordan coecients commonly arise in applications involving the addition of angular momentum in quantum mechanics. If products of more than three spherical harmonics are desired, then a generalization known as Wigner 6j-symbols or Wigner 9j-symbols is used. The Clebsch-Gordan coecients are variously written as j j1 j2 j Cm1 m2 , Cm1 m2 m , (j1 j2 m1 m2 |j1 j2 jm), j1 j2 m1 m2 |j1 j2 m . The Clebsch-Gordon coecients are dened by: j (A.35) jm = Cm1 m2 m1 m2
m=m1 +m2

where j j1 + j2 and satisfy (j1 j2 m1 m2 |j1 j2 jm) = 0 109 (A.36)

APPENDIX A. MATHEMATICAL BACKGROUND for m1 + m2 = m. The coecients are subject to the restrictions that (j1 , j2 , j) be positive integers or half-integers, (m1 , m2 , m) be positive or negative integers or half-integers, and j1 + j 2 j 0 j1 j2 + j 0 j1 + j2 + j 0 j1 + j2 + j N and |j1 | |j2 | |j| m1 |j1 | m2 |j2 | m |j|

(A.37)

(A.38)

A.6

Laplace Expansion

For a given matrix A, the determinant, det A, is a scalar that depends upon the elements of A. Also the determinant is dened only for a square matrix. The Laplace expansion allow us to calculate the determinant of an order-N matrix as a weighted sum of N order-2 determinants which simplify its evaluation. At this level we have to introduce the notion of minor and cofactor. The minor Mij of the element Aij of an N N matrix A is the determinant of the (N 1) (N 1) matrix obtained by removing all the elements of the ith row and the j th column of A; the associated cofactor is found by multiplying the minor by (1)i+j . If we dene the matrix A as: A11 A12 A13 (A.39) A = A21 A22 A23 A31 A32 A33 det A is then dened as: det A = |A| = A21 C21 + A22 C22 + A23 C23 = A21 (1)2+1 M21 + A22 (1)2+2 M22 + A23 (1)2+3 M23 = A21 A12 A13 A A13 A A12 + A22 11 A23 11 A32 A33 A31 A33 A31 A32

(A.40)

In the same way, the determinant of the 2 2 matrix can be calculate: A12 A13 = A12 (1)1+1 |A33 | + A13 (1)1+2 |A32 | A32 A33 = A12 A33 A13 A32

(A.41)

110

Appendix B LFDFT Scripts

B.1 Creation of the SD calculation input

The inputs for the geometry optimization or the AOC calculation correspond to a normal ADF input but for the SD calculation, we have to list all the Slater determinants in a specic order. I wrote a program which is a graphical interface (GUI) to generate the SD calculation inputs because as we can see in section 3.1.4 this input can be very long and it can be time consuming and source of errors to write it. This program is written in Tcl/Tk (Tool Command Language/Tool Kit)144 and it calls a small Fortran program. I will describe here how to use it. Under the create menu, two options are available, full input and SD part, the rst option opens a window like in Figure B.1 while the second option opens a window corresponding to the part delimited by the black rectangle of Figure B.1. For the full input creation mode you need: 1. to precise the xyz lea of your compound. For this, you have to click on the Load xyz le button and a selector of le appears. 2. to precise the title you want to give to your calculation. 3. to give the charge of your compound. 4. you have to precise the kind of approximation you want to use: LDA or GGA. 5. in you choose GGA, you have to precise which functional you want to use (cf. Table 3.1 to have the full description of each name). 6. to precise which is the TAPE21 le of the AOC calculations. For this, you have to click on the button Precise which is the tape21 le of your AOC and a selector of le appears. As soon as you had entered all these data, you arrived to the part corresponding to the part delimited by the black rectangle in the Figure B.1 which is common for the two
The xyz le is constructed in the following fashion: the rst line contains the number of atoms in the compound, the second line is normally dedicated to the name of the compound and the following line gives (one line per atom) give the label of the atom and its x, y, z coordinates.
a

111

APPENDIX B. LFDFT SCRIPTS creation modes: full input and SD part. Thus the following description is valid for the two creation modes. Now the data you have to enter now refer to the SD part of the input, you need: 7. to precise the number of electrons on the d-shell. 8. to precise the total number of electrons in your compound.b 9. to precise the symmetry you use in the AOC calculation. 10. to precise, for each irrep, the number of time it appears with a full electronic congurations in the KS MO diagram of the average of conguration output. 11. to precise, for each of the ve d-AO (since the TM has an open d-shell), which are the corresponding MO in the AOC calculation output. It means, for each d-AO, to select the MO which corresponds to the maximum contribution of this d-AO. As soon as the user gives all the data, some routines check if the input is valid based on the total number of electrons, the number of electrons on the d-shell and the occupations of each irreps, and then if all data entered are valid, the program proposed to save the le with the desired lename. The advantage of the full input creation mode is that the input created can directly be executed by the ADF program.

B.2
B.2.1

Extraction of data
SD energies

The very short and very useful script get.x reads the logle of the single determinant calculation and stores all the energies of SD in a new le esd. This new le contains only an array of numbers which is directly loadable by the LFDFT program. The command to run get.x is: get.x < aoc-le.logfile The source code of this script is contained in Figure B.2.
#! /bin/bash # By Rauzy Cdrick e # this script is to generate the list of SD energies # from a SD calculations logfile. # grep "GGA-XC" $1 | grep "eV" | awk {print $5} > esd

Figure B.2: Source code of the script get.x.

You can nd it in the AOC calculation output under the section S Y M M E T R Y , E L E C T R O N S and then Total. Since you use frozen-core approximation, this number is not equal to the sum of electron per real atom.
b

112

B.2 Extraction of data

Figure B.1: Snapshot of the program chem.x. The number on each side refer to the item number in the previous two lists.

113

APPENDIX B. LFDFT SCRIPTS

B.2.2

Eigenvalues of KS orbitals

The script to extract the eigenvalues of the MOs is not fully optimized, it works for cases where the compound has a high symmetry like Td or Oh but for lower symmetry where there is mixing between d-AO in the MOs, this script does not work. A script valid for every symmetry can be written with a more complex code by including some line of code to compare the MO compositions. The source code of the actual version of eigen.x is presented in Figure B.3.
#! /bin/tcsh # By Rauzy Cdrick e # -----# This script is to run on the average of configuration # with the command eigen.x aoc-calculation.OUT # it gives the matrix of the eigen values of MO with # dominant d-character. It act just on transition metal with # and for a mono metallic compound # grep -A1000 "List of all MOs" $1 | grep "1 D:z2" | awk { if print "L = [" $1 " 0.000 0.000 0.000 0.000 ;"} grep -A1000 "List of all MOs" $1 | grep "1 D:x2-y2" | awk { print " 0.000 " $1 " 0.000 0.000 0.000;"} grep -A1000 "List of all MOs" $1 | grep "1 D:xy" | awk { if print " 0.000 0.000 " $1" 0.000 0.000;"} grep -A1000 "List of all MOs" $1 | grep "1 D:xz" | awk { if print " 0.000 0.000 0.000 " $1" 0.000 ;"} grep -A1000 "List of all MOs" $1 | grep "1 D:yz" | awk { if print " 0.000 0.000 0.000 0.000 " $1 "];"}

d-orbitals

($10 =="1") if ($10 =="1") ($10 =="1") ($10 =="1") ($10 =="1")

Figure B.3: Source code of the script eigen.x.

The command to run eigen.x is: eigen.x < aoc-le.OUT

B.2.3

Eigenvectors of KS orbitals

There is no script written yet and the better way up to now to extract the eigenvectors is to do it manually as it is explained in Section 3.2.1.1.

B.2.4

Evaluation of Standard Deviation

The LFDFT method accuracy is evaluated through the calculation of the Standard deviation between the SD energies calculated by DFT and the SD energies from LFDFT: we use the Racahs parameters and the ligand eld matrix determined by the LFDFT program in equation 3.2 to calculate these SD energies. Then to calculate the standard deviation, we use the Equation B.1:
NSD i=1 (Ei (f it)

StDev =

Ei (DF T ))2

(NSD 1)

(B.1)

114

Appendix C Gener ln Functions

In this appendix, each functions of the Gener ln program are given. First, GENERSD is depicted, then the functions to calculate matrix elements in the basis of spin-orbitals are presented, at the end, the ZAB, LFAB and GAB functions which are corresponding to the Slaters rule applied to the one-electron operator for the two rst ones and to the Slaters rule applied to the two-electron operator for the latter one. This structure of this appendix corresponds to the Figure 4.3.

C.1

GENERSD Function

This function is done to create a matrix, ir, which by associating one number to each spin orbitals, list all the congurations of each Slater determinants in a specic order. So the number of column in ir corresponds to the number of electron on the d-shell and the number of line to the number of SD for the given conguration. If we have, for example, a d3 conguration and we consider the rst SD: we x arbitrarily that the rst electron is on the orbital l = 2 with spin down, the second on the obital l = 2 with spin up and the third on the orbital l = 1 whith spin down. So the rst SD corresponds to the conguration d d+ d , in the matrix ir, it corresponds to a line xy xy yz with the numbers 1 2 3. The second SD is 1 2 4, the third 1 2 5 and so on until 8 9 10 with the following correspondence: 1 d , 2 d+ , 3 d , 4 d+ and so on up to 10 xy xy yz yz with m going from 2 to 2 and having the spin down before the spin up (the spin moving faster than the value of m). This order is the same as the one used to generate the ADF SD calculation input. The Figure C.1 corresponds to the source code of the GENERSD function. 115

APPENDIX C. GENER LN FUNCTIONS function ir=genersd(ne,ka,kb) % generate ir corresponding to Slater determinants configuration % ordered in a specific way. nf=kb-ka+1; nsd=factorial(nf)/factorial(nf-ne)/factorial(ne); % initialize ir nr=1; ir(nr,:)=[ka:ka+ne-1]; % generate SD for nr=2:nsd ir(nr,1:ne)=ir(nr-1,1:ne); loop=1;j=0; while loop==1 & j<=ne j=j+1; ip=ne-j+1; np=kb-j+1; if ir(nr,ip)<np ir0=ir(nr,ip)-ip+1; ir(nr,ip:ne)=[ir0+ip:ir0+ne]; loop=0; end end end
Figure C.1: Source code of the GENERSD function.

C.2
C.2.1

Function to Determine Matrix Elements in the Basis of Spin-Orbitals

GET2EI4A

This function calculates the non-redundant electrostatic 2-e integrals for atomic orbitals with angular momenta lq. The Slater-Condon parameter are ordered by: Fk (lq1 lq1 |lq1 lq1 ), Fk (lq1 lq2 |lq1 lq1 ), . . . , Fk (lq1 lqn |lq1 lq1 ), Fk (lq2 lq1 |lq1 lq1 ), . . . , Fk (lqn lqn |lqn lqn ). The way to obtain the Slater condon parametererization is explained in Section 2.1.4.1 so the function can be easily understand; the source code is given in Figure C.3. First, the Clebsh Gordan coecients are calculated in the CGR function and they are stored as < lm|m1 m2 >, where: abs(l1 l2 ) l l1 l2 l l1 + l2 m l m1 l1 m2 l2

Only even l values if la + lb is even or odd l values if la + lb is odd contribute. Then, the one-electrons product is calculated so the loop runs over (2l + 1) ((2l + 1) + 1)/2 = 15 elements for l = 2. Finally the two electrons products is achieved and the loop runs 116

C.2 Function to Determine Matrix Elements in the Basis of Spin-Orbitals over (15 (15 + 1)/2 elements. As the result, iabcd matrix contains the left hand of Table 2.1 where orbitals are represented by numbers as it is explained in Appendix C.1 and vabcd matrix contains the coecients corresponding to the right hand of Table 2.1 but for Slater Condon parameterisation. One can remark at the end of the Gener ln program that Racahs parameterization is obtained by linear transformation. CGR This function calculates the real Clebsch-Gordan coecients, lm|m1 m2 , of full rotation group. They are stored in a ((2l + 1) (2l + 1))2 matrix, v, where lm| corresponds to the lines and m1 m2 | corresponds to the columns. For lq = 2, the m1 m2 | values corresponding to the columns are ordered as: -2 -2, -2 -1, . . . , -2 2, -1 -2, -1 -1, . . . 2, 2, and the the lm| values corresponding to the lines are ordered as: 0 0, 1 -1, 1 0, 1 1, 2 -2, . . . , 4 4. This function is equivalent to the Eq 2.49. We can see that Clebsch-Gordan coecients represent an integration over the product of three spherical harmonics, thus we have rst to calculate this three spherical harmonics. Depending on the value of l1 , l2 and l, the function YLM1 is called 1, 2 or 3 times. The spherical harmonic function are calculated numerically using a Lebedev grid which is chosen depending on the values of l1 and l2 . The source code of the function is given in Figure C.4. YLM1 function yl=ylm1(l,z,phi) % get associated legendre function normalised to 1 plm=legendre(l,z,sch); plm(1,:)=plm(1,:)*sqrt(0.5*(l+l+1)); plm(2:l+1,:)=plm(2:l+1,:)*sqrt(0.25*(l+l+1)); % get azimuthal part for m=-l:l if m<0, fm(l+m+1,:)=sin(-m*phi); end if m==0, fm(l+1,:)=0.70710678118655; end if m>0, fm(l+m+1,:)=cos(m*phi); end end fm=fm/sqrt(pi); % multiply both to get yl(-l:l) for m=-l:l yl(l+m+1,:)=plm(abs(m)+1,:).*fm(l+m+1,:); end
Figure C.2: Source code of the YLM1 function.

This function determine the values of the real spherical harmonic function, ylm (, ), within a Lebedev grid. The source code is given in Figure C.2. We have then to dene the expression of Ylm (, ): Ylm (, ) = (1)m 2l + 1 (l m)! Plm (cos )eim 4 (l + m)! (C.1)

In a rst time, the Schmidt semi-normalized associated Legendre functions is computed 117

APPENDIX C. GENER LN FUNCTIONS so we can dene it in terms of the Legendre function dened by: Pnm (x) = (1)m (1 x2 )m/2 d Pn (x) dxm (C.2)

where Pn (x) is the Legendre polynomial of degree n dened by: Pn (x) = dn 2 1 (x 1)n 2n n! dx (C.3)

The Schmidt seminormalized associated Legendre function are related to the normalized m associated Legendre functions Pn (x) by: Slm (x) = (1)m Thus Eq C.1 read as: Ylm (, ) = (1)m 2l + 1 Slm (cos )eim 2 (C.5) 2(n m)! Plm (x) (n + m)! (C.4)

As we consider real spherical harmonic, we can write: ylm (, ) = (1)m sin |m| if m < 0 2l + 1 Slm (cos ) 2 cos m if m 0 (C.6)

118

C.2 Function to Determine Matrix Elements in the Basis of Spin-Orbitals

function [vabcd,iabcd]=get2ei4a(lq) ndl=length(lq); nf=0; for kl=1:ndl for km=1:lq(kl)+lq(kl)+1 nf=nf+1; lf(nf)=lq(kl); mf(nf)=km-lq(kl)-1; end end nvx=(2*max(lf)+1)^2; v=zeros(nvx,nvx,ndl*(ndl+1)/2); kk=0; for k=1:ndl mulk=lq(k)+lq(k)+1; for kp=k:ndl mulkp=lq(kp)+lq(kp)+1; kk=kk+1; nv(kk)=mulk*mulkp; v(1:nv(kk),1:nv(kk),kk)=CGR(lq(kp),lq(k)); lab(kk,:)=[lq(kp) lq(k)]; end end kkp=0; ip0(1)=0; for k=1:kk la=lab(k,1); lb=lab(k,2); for kp=1:k lc=lab(kp,1); ld=lab(k,2); k1=abs(la-lb):2:la+lb; k2=abs(lc-ld):2:lc+ld; k12=find(k1==k2); kx=k2(k12); kkp=kkp+1; ip0(kkp+1)=ip0(kkp)+length(kx); end end % 1-e products nab=0; k=0; for ia=1:nf for ib=1:ia nab=nab+1; iab(nab,:)=[ia ib]; la=lf(ia); lb=lf(ib); ma=mf(ia); mb=mf(ib); % get index of matrix where Clebsch-Gordan coeff. are stored lv(nab) = find( lab(:,1)==la & lab(:,2)==lb ); % get column of matrix where Clebsch-Gordan coeff. are stored mv(nab)=(la+ma)*(lb+lb+1)+lb+mb+1; end end % non-redundant 2-e integrals for j1=1:nab for j2=1:j1 k=k+1; ia=iab(j1,1); ib=iab(j1,2); ic=iab(j2,1); id=iab(j2,2); % iabcd contains the 4 indices of the non-redundant 2-e integrals iabcd(k,:)=[iab(j1,:) iab(j2,:)]; la=lf(ia); lb=lf(ib); lc=lf(ic); ld=lf(id); k1=abs(la-lb):2:la+lb; k2=abs(lc-ld):2:lc+ld; k12=find(k1==k2); kx=k2(k12); for ix=1:length(kx) mu1=(kx(ix)+1)^2-min(k1)^2; mu2=(kx(ix)+1)^2-min(k2)^2; ml1=mu1-kx(ix)-kx(ix); ml2=mu2-kx(ix)-kx(ix); a(ix)=sum( v(ml1:mu1,mv(j1),lv(j1)).*v(ml2:mu2,mv(j2),lv(j2)) ); end ip0(lv(j1)*(lv(j1)-1)/2+lv(j2)); vabcd(k,ip0(lv(j1)*(lv(j1)-1)/2+lv(j2))+1:ip0(lv(j1)*(lv(j1)-1)/2+lv(j2))+ length(kx))=a(1:length(kx)); end end

Figure C.3: Source code of the CET2EI4A function.

119

APPENDIX C. GENER LN FUNCTIONS function v=cgr(l1,l2) % test if l1 + l2 are out of range if l1+l2 > 29 , disp(l1+l2 are out of range), pause, end % get grid load lebedev lmx=[1 3 5 7 9 11 15 17 19 23 29 35 41 47 53 59]; igrid=min(find(lmx>=2*(l1+l2))); grid=LG(LLLG(igrid):ULLG(igrid),:); x=grid(:,1);y=grid(:,2);z=grid(:,3);w=grid(:,4); % get azimuthal angle phi phi=atan2(y,x); % get yl1 and yl2 yl1=sqrt(4*pi)*YLM1(l1,z,phi); if l1==l2, yl2=yl1; else yl2=sqrt(4*pi)*YLM1(l2,z,phi); end % start quadrature loop mu=0; for l=abs(l1-l2):l1+l2 % get yl if l==l1, yl=yl1; elseif l==l2, yl=yl2; else yl=sqrt(4*pi)*YLM1(l,z,phi); end % get < l m | l1 m1 , l2 m2 > as sum(Yl m*Yl1 m1*Yl2 m2) for m=-l:l mu=mu+1; nu=0; for m1=-l1:l1 for m2=-l2:l2 nu=nu+1; v(mu,nu)=sum(w.*yl(l+m+1,:).*yl1(l1+m1+1,:).*yl2(l2+m2+1,:)); end end end end
Figure C.4: Source code of the CGR function.

120

C.2 Function to Determine Matrix Elements in the Basis of Spin-Orbitals

C.2.2

GETLS

This function creates the lx, ly, lz, sx, sy, sz and ls matrices in the basis of the spin-orbitals (10 10 matrix for lq = 2). We can devide this function in three parts. Firstly, the function lmat calculates the equivalent matrix operator of li in the basis of orbitals. Secondly, the function getj is called to calculate the Pauli matrices for the electron. And nally, the li and si matrices are transformed in the basis of spin orbitals. In this function, the spin-orbit coupling matrix is also calculated: l, ml , s, ms |l s|l, ml , s , ms >. The order of the basis functions is: |ml , ms = | l, s , | l, +s , | l + 1, s , | l + 1, +s , ..., |l, s , |l, +s . The source code of the function is given in Figure C.6. GETJ J is the total angular momentum which is the vector sum of the spin angular momentum S and the orbital angular momentum L: J S+L (C.7)

This operator is quantized by two quantum numbers j and mj and this function has the role to compute: ji (m : m ) = j, mj |Ji |j, mj which in fact corresponds to the Pauli spin matrices. The complete description of the construction of these matrices can be found in the Roses book.145 function [jx,jy,jz]=getj(j)% j=1/2 half=0.5; mul=j+j+1;% mul=2 jx=zeros(mul,mul)+i*zeros(mul,mul); jy=zeros(mul,mul)+i*zeros(mul,mul); jz=zeros(mul,mul)+i*zeros(mul,mul); % jz mj=-j-1; for k=1:mul mj=mj+1; jz(k,k)=jz(k,k)+mj; end % jx and jy mj=-j-1; for k=1:mul-1 mj=mj+1; jx(k+1,k)=jx(k+1,k)+half*sqrt(j*(j+1)-mj*(mj+1)); jx(k,k+1)=jx(k+1,k); jy(k+1,k)=jy(k+1,k)-half*i*sqrt(j*(j+1)-mj*(mj+1)); jy(k,k+1)=conj(jy(k+1,k)); end
Figure C.5: Source code of the GETJ function.

121

APPENDIX C. GENER LN FUNCTIONS function [alx,aly,alz,asx,asy,asz,ls]=getls(l,s) nl=l+l+1; lx=zeros(nl,nl)+i*zeros(nl,nl); ly=zeros(nl,nl)+i*zeros(nl,nl); lz=zeros(nl,nl)+i*zeros(nl,nl); % get l-matrices in basis of real spherical harmonics for k=1:nl% 1 to 5 for lq=2 ml=k-l-1;% ml going from -2 to 2 for lq=2 mls=0; if ml<0 mls=1; end for kp=1:nl mlp=kp-l-1; mlsp=0; if mlp<0 mlsp=1; end lx(k,kp)=lx(k,kp)+i*LMAT(1,l,abs(ml),mls,l,abs(mlp),mlsp); ly(k,kp)=ly(k,kp)+i*LMAT(2,l,abs(ml),mls,l,abs(mlp),mlsp); lz(k,kp)=lz(k,kp)+i*LMAT(3,l,abs(ml),mls,l,abs(mlp),mlsp); end end % get s-matrices [sx,sy,sz]=GETJ(s); ns=s+s+1; % get al=<ml,ms|l|mlp,msp> and as=<ml,ms|s|mlp,msp> us=eye(ns,ns); ul=eye(nl,nl); for kl=1:nl for ks=1:ns for klp=1:nl for ksp=1:ns alx(ns*(kl-1)+ks,ns*(klp-1)+ksp)=lx(kl,klp)*us(ks,ksp); aly(ns*(kl-1)+ks,ns*(klp-1)+ksp)=ly(kl,klp)*us(ks,ksp); alz(ns*(kl-1)+ks,ns*(klp-1)+ksp)=lz(kl,klp)*us(ks,ksp); asx(ns*(kl-1)+ks,ns*(klp-1)+ksp)=ul(kl,klp)*sx(ks,ksp); asy(ns*(kl-1)+ks,ns*(klp-1)+ksp)=ul(kl,klp)*sy(ks,ksp); asz(ns*(kl-1)+ks,ns*(klp-1)+ksp)=ul(kl,klp)*sz(ks,ksp); end end end end ls=alx*asx+aly*asy+alz*asz;
Figure C.6: Source code of the GETLS function.

122

C.2 Function to Determine Matrix Elements in the Basis of Spin-Orbitals LMAT The function LMAT calculates matrix elements of the angular momentum between atomic orbitals on one center. The angular momentum is dened by: L = i (r ) (C.8)

Squaring the expression, we can write out the Lx component: Lx = i y z z y (C.9)

By using spherical coordinates representations, Eq C.9 reads: Lx = i sin cos tan (C.10)

Then applying the angular momentum operator to atomic wavefunctions, the components of the angular momentum have the following expectation values: Lz |nlm = m |nlm 1 1 Lx |nlm = (l m)(l + m + 1)|n,l,m+1 + (l + m)(l m + 1)|n,l,m1 2 2 1 1 Lx |nlm = (l m)(l + m + 1)|n,l,m+1 + i (l + m)(l m + 1)|n,l,m1 2 2 (C.11) And we have to realise that in fact when lmat is called, we compute: n2 ,l2 ,m2 |Li |n1 ,l1 ,m1 (C.12)

In the program the value of m is decomposed in m and ms, respectively the m absolute value and the m sign so the atomic orbitals are dened with quantum numbers (l,m,ms) which corresponds to a real normalized combination of spherical harmonics Y(l,m) and Y(l,-m) (ms=0 is the cosine, ms=1 the sine). In the last line of this function, we can see the same expression than in equation C.11 for couple (l2, m2, ms2; l1, m1, ms1) which are not vanishing (two wavefunctions which are not orthogonal). The source code of the function is given in Figure C.7

123

APPENDIX C. GENER LN FUNCTIONS function al=lmat(ix,l2,m2,ms2,l1,m1,ms1) % mm=[1 1 0]; mms=[1 0 1]; m=mm(ix); ms=mms(ix); if (l2 =l1)|(abs(m2-m1) =m)|(abs(ms2-ms1) =ms), al=0; else if m2 =m1, if m2<m1, msign=-1; ma=m2; msa=ms2; end if m2>m1, msign=1; ma=m1; msa=ms1; end if ma==0, sign=msign*(1-2*ms); squ=(l1)*(l1+1)/2; al=sign*sqrt(squ); else sign=msign*(1-2*ms*(1-msa)); squ=(l1-ma)*(l1+ma+1); al=sign*sqrt(squ)/2; end else al=(1-2*ms1)*m1; end end
Figure C.7: Source code of the LMAT function.

124

C.3 Functions corresponding to the Slaters rules

C.3
C.3.1

Functions corresponding to the Slaters rules

One-electron Slaters rule

Considering Eq 2.42 and using Laplaces expansion (cf. Appendix A.6) on SD, we get:

|H0 | =
i=1 j=1

(1)i+j ki |f |kj Sij

(C.13)

where:

Sij =

dt|k1 k2 . . . ki1 ki+1 . . . kn | |k1 k2 . . . kj1 kj+1 . . . kn |

(C.14)

where cases:

dt is carried out for electrons 2, 3, . . . , N . Then we can distinguish three

if ki = ki for all i if ki = ki for only one i if k1 = k1 , k2 = k2

|H0 | =
i

ki |f |ki (C.15)

|H0 | = (1)P ki |f |ki |H0 | = 0

where P is the number of permutations needed to put the functions of in the same order of those of . Eq C.15 corresponds to the Slater rules for evaluating the matrix elements of one-electron operators between Slater determinants written in an orthonormal basis of orbitals.

C.3.1.1

ZAB

This subroutine is reponsible of the creation of the matrix lsdata which is the sin-orbit coupling matrix in the basis of the microstates. It uses the li , si and ls matrices in the basis of spin-orbitals to create one matrix containing the li , si and ls matrix element in the basis of -states. The function corresponds to the application of the Slater rules for the one-electron operator (cf. Section 2.1.3) to perform the transformation of basis. The matrix created has the size: (nsd (nsd + 1))/2 7 where the seven columns corresponds to lx , ly , lz , sx , sy , sz , ls and the line corresponds to each element of the lower symmetric matrix. The source code of the function is given in Figure C.8 125

APPENDIX C. GENER LN FUNCTIONS function z=zab(la,lb) global lx ly lz sx sy sz ls % Get <A|z(1)|jB> ne=max(size(la)); ie=0; id=0; sgn=1; for k=1:ne ld=find(la(k)==lb); if isempty(ld) id=id+1; kd(id)=la(k); if rem(k,2)==1, sgn=-sgn; end else ie=ie+1; ke(ie)=la(k); end end if id == 0 z=zeros(1,7)+i*zeros(1,7); for k=1:ne z=z+[lx(ke(k),ke(k)) ly(ke(k),ke(k)) lz(ke(k),ke(k)) sx(ke(k),ke(k)) sy(ke(k),ke(k)) sz(ke(k),ke(k)) ls(ke(k),ke(k))]; end elseif id==1 iv=0; for k=1:ne ld=find(lb(k)==la); if isempty(ld) iv=iv+1; kdp(iv)=lb(k); if rem(k,2)==1, sgn=-sgn; end end end z=sgn*[lx(kd(1),kdp(1)) ly(kd(1),kdp(1)) lz(kd(1),kdp(1)) sx(kd(1),kdp(1)) sy(kd(1),kdp(1)) sz(kd(1),kdp(1)) ls(kd(1),kdp(1))]; else z=zeros(1,7)+i*zeros(1,7); end
Figure C.8: Source code of the ZAB function.

126

C.3 Functions corresponding to the Slaters rules C.3.1.2 LFAB

The role of LFAB is to generate the ligand eld matrix elements which are stored in lfdata. lfdata has a sized (nsd(nsd + 1)/2) (lq 2 ), it is in fact the lower ligand eld diagonal matrix in the basis of microstates. The LFAB function uses the Slater rules for the one electron matrix (cf. Section 2.1.3). The diagonal elements of LF matrix, i.e. the lines 1, 3, 6, 10, 15, . . . of lfdata, corresponds to the occupations of the corresponding SD in the basis of orbitals. The source code of the function is given in Figure C.10. IWAB The source code of this function is given in Figure C.9. The role of this function is just to determine the index iw in the basis of orbitals from the indexes of i and ip in the basis of spin orbitals. function iw=iwab(i,ip) % i and ip in the basis of spin orbitals % ia and ib in the basis of orbitals ia=fix((i+1)/2); ib=fix((ip+1)/2); % if rem(i,2) =rem(ip,2) iw=0; else if ia<=ib, iw=ia+ib*(ib-1)/2;end if ib<=ia, iw=ib+ia*(ia-1)/2;end end
Figure C.9: Source code of the IWAB function.

127

APPENDIX C. GENER LN FUNCTIONS function lf=lfab(la,lb) global lq; % Get <A|lf(1:(lq+lq+1)*(lq+1))|jB> ne=max(size(la)); lf=zeros(1,(lq+lq+1)*(lq+1)); ie=0; id=0; sgn=1; for k=1:ne ld=find(la(k)==lb); if isempty(ld) id=id+1; kd(id)=la(k); if rem(k,2)==1 sgn=-sgn; end else ie=ie+1; ke(ie)=la(k); end end if id == 0 for k=1:ne if IWAB(ke(k),ke(k))>0 lf(IWAB(ke(k),ke(k)))=lf(IWAB(ke(k),ke(k)))+1; end end elseif id==1 iv=0; for k=1:ne ld=find(lb(k)==la); if isempty(ld) iv=iv+1; kdp(iv)=lb(k); if rem(k,2)==1 sgn=-sgn; end end end if iwab(kd(1),kdp(1))>0, lf(iwab(kd(1),kdp(1)))=sgn; end end
Figure C.10: Source code of the LFAB function.

128

C.3 Functions corresponding to the Slaters rules

C.3.2

Two-electrons Slaters rule

Considering Eq 2.43 and if we use Laplace expansion to write SDs in terms of twodimensional SDs: |k1 k2 . . . kn | = 1 [N (N 2)/2]
j=1

N (1)i+j+3 |ki kj |(1,2) (C.16)

|k1 k2 . . . ki1 ki+1 . . . kj1 kj+1 . . . kN |(3,4,...,N ) Therefore Eq 2.43 reads:

N N

|H1 | =
q>p=1 s>r=1

(1)p+q+r+s

kp kq |g|kr ks kp kq |g|ks kr

Spq,rs (C.17)

where: Spq,rs =

dt|k1 k2 . . . kp1 kp+1 . . . kq1 kq+1 . . . kN |

(C.18)

|k1 k2 . . . kr1 kr+1 . . . ks1 ks+1 . . . kN | where

dt is carried out for electrons 3, 4, . . . , N . Then we can distinguish four cases:

N

if ki = ki for all i

|H1 | =
j>i=1

ki kj |g|ki kj ki kj |g|kj ki
N P j=2

if ki = ki only for i = 1

|H1 | = (1)

k1 kj |g|k1 kj k1 kj |g|kj k1 k1 k2 |g|k1 k2 k1 k2 |g|k2 k1 (C.19)

if ki = ki only for i {1, 2} if ki = ki for i {1, 2, 3}

|H1 | = (1)P |H1 | = 0

where P is the number of permutations needed to put the functions of in the same order of those of . Eq C.19 corresponds to the Slater rule for evaluating the matrix elements of two-electrons operators between Slater determinants written in an orthonormal basis of orbitals. GAB The GAB function transforms the gdata matrix in the basis of SDs and corresponds to the Slater rule for a two electrons operator (Eq C.19). So it calculates the electrostatic repulsion matrix elements: GAB = la(1), la(2), ..., la(ne)|G|lb(1), lb(2), ..., lb(ne) between two Slater determinantals. The resulting elements GAB are expressed as a sum of reduced two-electron integrals p(ip) i, j||k, l expressed in terms of Slater-Condons parameters. The source code of the function is given in Figure C.11.

129

APPENDIX C. GENER LN FUNCTIONS function g=gab(ne,la,lb) % get < A|G|B > ie=0; id=0; sgn=1; for k=1:ne if any(la(k)==lb), ie=ie+1; ke(ie)=la(k); else id=id+1; kd(id)=la(k); if mod(k,2)==1, sgn=-sgn; end end end if id==1 | id==2 iv=0; for k=1:ne if any(lb(k)==la), iv=iv+1; kdp(iv)=lb(k); if mod(k,2)==1, sgn=-sgn; end end end end % id>2, more than 2 spinorbitals differ g=zeros(size(GIJKL(1,1,1,1))); % id=2, 2 spinorbitals differ if id==2 g=sgn*GIJKL(kd(1),kd(2),kdp(1),kdp(2)); end % id=1, 1 spinorbital differ if id==1 for k=1:ne g=g+sgn*GIJKL(kd(1),la(k),kdp(1),la(k)); end end % id=0, no spinorbitals differ, i.e. la=lb if id==0 for k1=1:ne-1 for k2=k1+1:ne g=g+GIJKL(ke(k1),ke(k2),ke(k1),ke(k2)); end end end
Figure C.11: Source code of the GAB function.

130

C.3 Functions corresponding to the Slaters rules GIJKL This function determine: g = i(1), j(2)|g|ip(1), jp(2) i(1), j(2)|g|jp(1), ip(2) where g are the reduced 2-e integrals obtained in GET2EI4A and stored in the vabcd matrix. IROW being a function to nd the good line in vabcd. function g=gijkl(i,j,ip,jp) global iabcd vabcd % get orbital indices ia=fix((i+1)/2); ib=fix((ip+1)/2); ic=fix((j+1)/2); id=fix((jp+1)/2); % initialise g=zeros(size(vabcd(1,:))); % if mod(i,2)==mod(ip,2) & mod(j,2)==mod(jp,2) g=vabcd(IROW(ia,ib,ic,id),:); end if mod(i,2)==mod(jp,2) & mod(j,2)==mod(ip,2) g=g-vabcd(IROW(ia,id,ic,ib),:); end
Figure C.12: Source code of the GIJKL function.

IROW The source code of this function is given in Figure C.13. The role of this function is to determine the index of the line, k in the matrix gdtata in function of the values of ia, ib, ic and id which represent the orbitals considered for a bi-electronic integral. function k=irow(ia,ib,ic,id) global iabcd vabcd if ia>=ib, j1=ib+ia*(ia-1)/2; if ia<ib, j1=ia+ib*(ib-1)/2; if ic>=id, j2=id+ic*(ic-1)/2; if ic<id, j2=ic+id*(id-1)/2; if j1>=j2, k =j2+j1*(j1-1)/2; end if j1<j2, k =j1+j2*(j2-1)/2; end end end end end

Figure C.13: Source code of the IROW function.

131

APPENDIX C. GENER LN FUNCTIONS

132

Appendix D SOC: supplement

D.1 Spin-orbit coupling elements in symmetry D2d

In Eq D.1, the spin-orbit coupling elements in terms of reduced matrix elements for D2d symmetry are given (the notations in square brakets refer to the tetrahedral e and t2 species).
e a1 [e]||su(e)||e[t2 ] = a1 e e b1 [e]||su(e)||e[t2 ] = b1 e a b1 [e]||su(a2 )||b2 [t2 ] = b12b2

(D.1)

e[t2 ]||su(a2 )||e[t2 ] = b2 [t2 ]||su(e)||e[t2 ] =

a2 ee e b2 e

Next, spin-orbit coupling elements are given in the basis of spin-orbitals, the symmetry notations for the real 3d orbitals in D2d are written on the left hand side along with conventions for the e-components given in the ADF code.
+ b+ 2 e : 2+ e : 1+ b+ 1 a+ 1 b 2 e : 2 e : 1 b 1 a 1 + + + +
+

+
i a2 2 ee

+ a ib12b2

i e 2 b2 e 1 e 2 b2 e

i e 2 b2 e

1 e 2 b2 e

1 e 2 b1 e i e 2 b1 e

3 e 2 a1 e e 23 ia1 e

a ib12b2

i a2 2 ee

i e 2 b2 e i e 2 b2 e 1 e 2 b2 e

1 e 2 b2 e 1 e 2 b1 e i e - 2 b1 e

1 e 2 b1 e e 23 a1 e

i e 2 b1 e 3 e 2 ia1 e i a2 2 ee

e 1 b1 e 2 3 e 2 a1 e

i e 2 b1 e 3 e 2 ia1 e

e 23 a1 e e 23 ia1 e

a ib12b2

a ib12b2

i a2 2 ee

In Eq D.2, the spin-orbit coupling elements in terms of reduced matrix elements for Td symmetry are given.
t1 a e e e||su(t1 )||t2 = et2 = a1 e = b1 e = b12b2 a2 e t2 ||su(t1 )||t2 = ee = b2 e

(D.2)

133

APPENDIX D. SOC: SUPPLEMENT

D.2

DFT treatment of JT activity in the case of mixing of electronic states

H = (1/2)Ke (Q2 + Q2 ).I + Ve Q .C + Ve Q .C (D.3)

The Hamiltonian of the 3 T e Jahn-Teller problem in a linear approximation is:

where I is the identity matrix and T1 , and wave-functions: 1 0 2 0 1 C = 2 0 0

C and C are 3 3 matrices taken in the basis of the 0 0 ; 1 0 C = 0

3 2

0 0

3 2

0 0 0

(D.4)

Q and Q are the higher and lower symmetric components of the e-vibration which distorts the tetrahedron into D2d and D2 geometries, respectively. If we restrict to distortions of Q type and dene Q as being positive for tetragonal compression Eq D.3 simplies to Eq D.5 for the non-degenerate ground state (3 A2 in D2d ) and to Eq D.6 for the degenerate higher energy state (3 E).
3 3

A2 : E(T1 , ) = (1/2)Ke Q2 Ve Q

(D.5) (D.6)

E : E(T1 , , ) = (1/2)Ke Q2 + (1/2)Ve Q

Minimization of Eq D.5 with respect to Q yields the equilibrium geometry Qo and the Jahn-Teller stabilization energy EJT : Qo = Ve /Ke EJT = (1/2)Ve2 /Ke (D.7) (D.8)

If we take (in o ) to be the angle between the S4 axis and the TM-ligand bond (for the tetrahedron we have T d = 54.73o ). Qo (in radians) can be calculated from Eq D.9: A Q = R(o Td )(/180) In tetrahedral NiX2 complexes the 3 T1 wavefunction (3 T1 ) is given by: 4 (3 T1 ) = c1 (3 T1 , e4 t4 ) + c2 (3 T1 , e3 t5 ) 2 2 (A.2.8) (D.9)

where c1 and c2 (Table D.1) are given by the ground state eigenvector diagonalizing the matrix. 3 1 4 4 T (e t2 ) 3 T1 (e3 t5 ) 2 0 6B (A.2.9) 6B 9B + 10Dq The 3 T1 (e4 t4 ) and 3 T1 (e3 t5 ) states, before the mixing, are described by single determinants 2 2 and their equilibrium geometries (elongation, Table 5.3 and compression, Table D.1) and energies EJT can be calculated from separate DFT geometry optimizations to yield corresponding parameters Ke and Ve (via Eqs D.7 & D.8). Let us denote these parameters by Ke , Ve and Ke , Ve , respectively. Than Ke and Ve after mixing are given by: Ke = c 2 Ke + c 2 Ke 1 2 134 (A.2.10)

D.2 DFT treatment of JT activity in the case of mixing of electronic states Ve = c2 Ve + c2 Ve 2 1 (A.2.11)

Substitution into Eq. D.7 yields the equilibrium distortion Qo and utilizing Eq. D.9 we get the equilibrium angle o . This is given in Table 5.3. Parameter values summarizing the logical steps of this procedure are listed in Table D.1. Species NiF2 4 NiCl2 4 NiBr2 4 NiI2 4 c1 -0.944 -0.953 -0.955 -0.956 c2 0.330 0.303 0.297 0.292 34.4 15.5 13.3 10.1 EJT -8655 -2599 -1907 -1247 Ke Ve Ke Ve Ke Ve 12792 14883 22234 -6049 21206 -3770 13721 8447 10549 -2334 10840 -1344 11948 6752 7686 -1747 8064 -998 11517 5360 2122 -453 2921 43

Table D.1: Mixing coecients c1 and c2 of 3 T1 (e4 t4 ) and 3 T1 (e3 t5 ) in the 3 T1 ground state 2 2 function, DFT optimized angles o = o Td and EJT values for 3 T1 (e3 t5 ), force eld constants 2 and vibronic coupling parameters Ke , Ve ; Ke , Ve and Ke , Ve of 3 T1 (e3 t5 ); 3 T1 (e4 t4 ) and the 2 2 3 T ground state (taking mixing into account) for NiX2 (X=F , Cl , Br , I ) . For values 1 4 of o and EJT for the 3 T1 (e4 t4 ) DFT optimized single determinant energy, see the entries in 2 Table 5.3. Ke , Ve and EJT parameters have been expressed in cm1 /2 , cm1 / and cm1 , A A respectively, in o .

135

APPENDIX D. SOC: SUPPLEMENT

136

Appendix E ESR: Programs

To calculate the g and Atensors in Chapter 6, the two programs, Gener ln and LF, have been modied and in this appendix the complete version of both are proposed.

E.1

Gener ln pogram

Figures E.1& E.2 depict the source code of the modied Gener ln program and the source code of the HFAB function, respectively. In the rst time, we can see that three lines are added in the LF program to compute a : a = 4 s ls l l ls (E.1)

which is the part of the hyperne operator and in a second time, the hf ab function is called to apply the Slater rules to a in order to transform it from the basis of spin-orbitals to the basis of Slater determinants.

E.2

LF program

To calculate the g- and A-tensors, the LF program proposed in section 4.2.1 has been modied. The structure is pratically the same but we are obiged to introduce three constant: , and P which are respectively the spin-orbit constant, the Fermi contact term and the hyperne parameters. The determination of these paramameters is discussed in Chapter 6. Then in the source code, the parts which are added in the calculation are the determination of equivalent operator matrix elements of li and si for i {x, y, z} and the calculation of the Zeeman and the HFS matrices determined according to the Equation 6.2 & Equation 6.7 respectively. Then the g- and A-tensors are determined respectively to Equation 6.3 & Equation 6.14. We can underline that the spin-orbit constant is determined by an atomic relativistic ZORA DFT calculation and is reduced by the orbital reduction factor, k, which should be equal to one if we dont want to take into account the reduction twice (since we already reduce the spin-orbit constant, ). (k l kappa s + 1 a ) 7 (E.2)

The source code of this modied LF program is proposed in Figure E.3. 137

APPENDIX E. ESR: PROGRAMS global lx ly lz sx sy sz ls ax ay az t0=cputime; %type de couche s, p, d(2) , f(3) lq=[2]; ne=3; % generate single determinants or microstates for lq^ne ir=genersd(ne,1,2*sum(lq+lq+1)); % get ab|cd [vabcd,iabcd]=get2ei4a(lq); % get l, s and l*s 1-e matrices: l=2 & s=1/2 [lx,ly,lz,sx,sy,sz,ls]=getls(lq,1/2); % a k (Griffith eq. 12.22) ax=4*sx-ls*lx-lx*ls; ay=4*sy-ls*ly-ly*ls; az=4*sz-ls*lz-lz*ls; nsd=length(ir); ij=0; for i=1:nsd for j=1:i ij=ij+1; lfdata(ij,:)=lfab(ir(i,:),ir(j,:)); lsdata(ij,:)=zab(ir(i,:),ir(j,:)); hfdata(ij,:)=hfab(ir(i,:),ir(j,:)); gdata(ij,:)=gab(ne,ir(i,:),ir(j,:)); end end gdata racah(:,1)= gdata(:,1); gdata racah(:,2)= (49.0/5.0)*gdata(:,2); gdata racah(:,3)= (7.0/5.0) *(gdata(:,1)+gdata(:,2)+gdata(:,3)); Elapsed time=cputime-t0 save data d3 ir lfdata lsdata hfdata gdata gdata racah
Figure E.1: Source code of the Gener ln program modied to enable us to calculate the gand the A-tensors.

138

E.2 LF program function a=hfab(la,lb) global ax ay az % Get A|hf (1)|jB a part of hyperfine interaction (cf. Griffith eq. 12.22) ne=max(size(la)); ie=0; id=0; sgn=1; for k=1:ne ld=find(la(k)==lb); if isempty(ld) id=id+1; kd(id)=la(k); if rem(k,2)==1, sgn=-sgn; end else ie=ie+1; ke(ie)=la(k); end end if id == 0 a=zeros(1,3)+i*zeros(1,3); for k=1:ne a=a+[ax(ke(k),ke(k)) ay(ke(k),ke(k)) az(ke(k),ke(k))]; end elseif id==1 iv=0; for k=1:ne ld=find(lb(k)==la); if isempty(ld) iv=iv+1; kdp(iv)=lb(k); if rem(k,2)==1, sgn=-sgn; end end end a=sgn*[ax(kd(1),kdp(1)) ay(kd(1),kdp(1)) az(kd(1),kdp(1))]; else a=zeros(1,3)+i*zeros(1,3); end
Figure E.2: Source code of the hfab function.

139

APPENDIX E. ESR: PROGRAMS

load data r=[0; B; C]; r=r./1000; lfpar=[ h(1,1) ; h(2,1) ; h(2,2) ; h(3,1) ; h(3,2) ; h(3,3) ; h(4,1) ; h(4,2) ; h(4,3) ; h(4,4) ; h(5,1) ; h(5,2) ; h(5,3) ; h(5,4) ; h(5,5)]; lfpar=lfpar./1000; zeta=. . . ; P=. . . ; kappa=. . . ; ge=2.0023; [nsd ne]=size(ir); ij=0; for i=1:nsd for j=1:i ij=ij+1; h(i,j)=gdata(ij,:)*r+lfdata(ij,:)*lfpar+lsdata(ij,7)*zeta; h(j,i)=conj(h(i,j)); % Get orbital and spin angular momentum matrices lx(i,j)=lsdata(ij,1); sx(i,j)=lsdata(ij,4); lx(j,i)=conj(lx(i,j)); sx(j,i)=conj(sx(i,j)); ly(i,j)=lsdata(ij,2); sy(i,j)=lsdata(ij,5); ly(j,i)=conj(ly(i,j)); sy(j,i)=conj(sy(i,j)); lz(i,j)=lsdata(ij,3); sz(i,j)=lsdata(ij,6); lz(j,i)=conj(lz(i,j)); sz(j,i)=conj(sz(i,j)); kx(k,kp)=hfdata(kkp,1);kx(kp,k)=conj(kx(k,kp)); ky(k,kp)=hfdata(kkp,2);ky(kp,k)=conj(ky(k,kp)); kz(k,kp)=hfdata(kkp,3);kz(kp,k)=conj(kz(k,kp)); end end [c,e]=eig(h); [e,ie]=sort(real(diag(e))); e(:)=e(:)-e(1); c=c(:,ie); i0=1; i1=0; for i=2:nsd if abs(e(i)-e(i-1))>0.001 i1=i1+1; w(i1)=e(i-1); mul(i1)=i-i0; i0=i; end end fprintf(----------------------------- n) fprintf((2S+1)*M(gamma) E n) fprintf(----------------------------- n) for i=1:i1, fprintf( %3i %12.3f n,mul(i),w(i)), end fprintf(----------------------------- n) % Get Zeeman matrices zx=c(:,1:2)*(orf*lx+ge*sx)*c(:,1:2);zy=c(:,1:2)*(orf*ly+ge*sy)*c(:,1:2); zz=c(:,1:2)*(orf*lz+ge*sz)*c(:,1:2); % Get HFS matrices ax=c(:,1:2)*p*(orf*lx-kappa*sx+0.14285714285714*kx)*c(:,1:2); ay=c(:,1:2)*p*(orf*ly-kappa*sy+0.14285714285714*ky)*c(:,1:2); az=c(:,1:2)*p*(orf*lz-kappa*sz+0.14285714285714*kz)*c(:,1:2); % Get g-tensor g(1,3)=real(zx(1,1))-real(zx(2,2));g(2,3)=real(zy(1,1))-real(zy(2,2)); g(3,3)=real(zz(1,1))-real(zz(2,2));g(1,1)=real(zx(1,2))+real(zx(2,1)); g(2,1)=real(zy(1,2))+real(zy(2,1));g(3,1)=real(zz(1,2))+real(zz(2,1)); g(1,2)=imag(zx(1,2))-imag(zx(2,1));g(2,2)=imag(zy(1,2))-imag(zy(2,1)); g(3,2)=imag(zz(1,2))-imag(zz(2,1)); gg=g*g; gtensor=sqrt(eig(gg)) % Get A-tensor a(1,3)=real(ax(1,1))-real(ax(2,2));a(2,3)=real(ay(1,1))-real(ay(2,2)); a(3,3)=real(az(1,1))-real(az(2,2));a(1,1)=real(ax(1,2))+real(ax(2,1)); a(2,1)=real(ay(1,2))+real(ay(2,1));a(3,1)=real(az(1,2))+real(az(2,1)); a(1,2)=imag(ax(1,2))-imag(ax(2,1));a(2,2)=imag(ay(1,2))-imag(ay(2,1)); a(3,2)=imag(az(1,2))-imag(az(2,1)); A=a*a; Atensor=sqrt(eig(A))

Figure E.3: Source code of the LF program modied to enable us to calculate the g and the A-tensors. 140

Curriculum Vitae
2001-04 Ph.D. in computational chemistry University of Fribourg

2000a 1999 DEA Chemistry of atmospheric pollution and physics of environment (eq. Master) University Paris 7 & University of Grenoble

1998

Ma trise (four year university degree) of chemistry University of Toulouse Licence (three year university degree) of chemistry University of Toulouse DEUG (two year university degree) in Life Sciences University of Toulouse Bac C (equivalent A-level) in mathematics and physics Toulouse

1997

1996

1994

Military service done from February to November 2000

141

List of publications
2002
1. A Novel Density Functional Study of the Ground State Properties of a localized trinuclear Copper (II,II,III) Mixed-Valence System. Daul C., Fernandez-Ceballos S., Cioni I., Rauzy C. and Schlpfer C. W. a Chem. Eur. J., 2002, 8, 4392-4401.

2003
2. New insights into the eects of covalency on the ligand eld parameters : a DFT study. Atanasov M., Daul C. and Rauzy C. Chem. Phys. Lett., 2003, 367, 737-746. 3. General chemistry for students enrolled in a life sciences curriculum. Bunzli J.-C. G., Fernandes E., Imbert D., Chauvin A.-S., Emmenegger F., Rauzy C., Piguet C., Ouali N., Koeller S., Suss-Fink G. and Cherioux F. Chimia, 2003, 57(3), 99-104.

2004
4. A DFT Based Ligand Field Theory Atanasov M., Daul C.A. and Rauzy C. Structure and Bonding, 2004, 106, 97-125. 5. Polytopal Rearrangement of [Ni(acac)2 (py)]: A New Square Pyramid Bipyramid Twist Mechanism Daul C.A., Niketic S., Rauzy C. and Schlpfer C.W. a Chem. Eur. J., 2004, 10, 721-727. Trigonal

6. Calculation of spin-orbit coupling within the LFDFT: applications to [NiX4 ]2 (X = F , Cl , Br , I ) Atanasov M., Rauzy C., Baettig P., Daul C.A. Int. J. Quant. Chem., 2005, 102, 119-131. 7. A DFT study of mixed valent Mn(II/III) hexacyanide clusters 143

Daul C.A., Rauzy C., Decurtins S., Franz P. Int. J. Quant. Chem., 2004, 101, 753-760. 8. Inuence of the ice growth rate on the incorporation of gaseous HCl Domin F. and Rauzy C. e Atm. Chem. Phys., 2004, 4, 2513-2519. 9. The calculation of ESR parameters by density functional theory: the g- and A-tensors of Co(acacen) Atanasov M., Baerends E.J., Baettig P., Bruyndonckx R., Daul C., Rauzy C. and Zbiri M. Chem. Phys. Lett., 2004, 399(4-6), 433-439. 10. DFT Study of Equilibrium Mo-Isotope Fractionations From Vibrational Spectroscopy Rauzy C., Wille M., Daul C. and Naegler T. Chem. Phys. Lett., to be submitted.