2022 August 15

Lectures on modular forms and strings

Eric D’Hoker(a) and Justin Kaidi(b)1
(a)
Mani L. Bhaumik Institute for Theoretical Physics
arXiv:2208.07242v1 [hep-th] 15 Aug 2022

Department of Physics and Astronomy

University of California, Los Angeles, CA 90095, USA
(b)
Simons Center for Geometry and Physics
Stony Brook University
Stony Brook NY 11794, USA

dhoker@physics.ucla.edu, jkaidi@scgp.stonybrook.edu

Abstract

The goal of these lectures is to present an informal but precise introduction to a

body of concepts and methods of interest in number theory and string theory revolv-
ing around modular forms and their generalizations. Modular invariance lies at the
heart of conformal field theory, string perturbation theory, Montonen-Olive duality,
Seiberg-Witten theory, and S-duality in Type IIB superstring theory. Automorphic
forms with respect to higher arithmetic groups as well as mock modular forms enter in
toroidal string compactifications and the counting of black hole microstates. After in-
troducing the basic mathematical concepts including elliptic functions, modular forms,
Maass forms, modular forms for congruence subgroups, vector-valued modular forms,
and modular graph forms, we describe a small subset of the countless applications to
problems in Mathematics and Physics, including those mentioned above.

1
Address after 1 September 2022 : Department of Physics, University of Washington, Seattle, WA, 98195
Contents
1 Introduction 7

I Modular forms and their variants 12

2 Elliptic functions 13
2.1 Periodic functions of a real variable . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Periodic functions of a complex variable . . . . . . . . . . . . . . . . . . . . 16
2.3 Elliptic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 The Weierstrass elliptic function . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Jacobi elliptic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Jacobi ϑ-functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Uniformization of cubics and quartics . . . . . . . . . . . . . . . . . . . . . . 32
2.8 Elliptic curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.9 Addition formulas and group structure of elliptic curves . . . . . . . . . . . . 35
2.10 Meromorphic functions and Abelian differentials . . . . . . . . . . . . . . . . 36
2.11 Abelian and elliptic integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Modular forms for SL(2, Z) 43

3.1 Automorphisms of lattices and the modular group . . . . . . . . . . . . . . . 43
3.2 The fundamental domain for SL(2, Z) . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Modular functions, modular forms, and cusp forms . . . . . . . . . . . . . . 48
3.4 Modular forms as differential forms . . . . . . . . . . . . . . . . . . . . . . . 49
3.5 Eisenstein series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 Dimension and generators of the ring of modular forms . . . . . . . . . . . . 54
3.7 Modular functions and the j-function . . . . . . . . . . . . . . . . . . . . . . 57
3.8 Modular transformations of Jacobi ϑ-functions . . . . . . . . . . . . . . . . . 59
3.9 The Dedekind η-function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4 Variants of modular forms 65

4.1 Quasi-modular and almost-holomorphic modular forms . . . . . . . . . . . . 65
4.2 Non-holomorphic Eisenstein series . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Maass forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Spectral decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5 Maass forms of general weight . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.6 Mock modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.7 Quantum modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

2
5 Quantum fields on a torus 86
5.1 Quantum fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2 The bc system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3 Correlators of the bc system on the torus . . . . . . . . . . . . . . . . . . . . 90
5.4 The scalar field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.5 The scalar Green function on the torus . . . . . . . . . . . . . . . . . . . . . 94
5.6 Scalar determinant and the first Kronecker limit formula . . . . . . . . . . . 97
5.7 Spinor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.8 Spinor Green functions on the torus . . . . . . . . . . . . . . . . . . . . . . . 101
5.9 Spinor determinant and the second Kronecker limit formula . . . . . . . . . 103

6 Congruence subgroups and modular curves 107

6.1 Definition of congruence subgroups of SL(2, Z) . . . . . . . . . . . . . . . . . 107
6.2 The classic congruence subgroups . . . . . . . . . . . . . . . . . . . . . . . . 107
6.3 Computing the order of SL(2, ZN ) . . . . . . . . . . . . . . . . . . . . . . . 109
6.4 Modular curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.5 Compactification of the modular curves . . . . . . . . . . . . . . . . . . . . . 112
6.6 The genus of X(Γ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.7 Formulas for d, ε2 , ε3 and ε∞ for Γ(N ), Γ1 (N ) and Γ0 (N ) . . . . . . . . . . . 115
6.8 Explicit expressions for the genus . . . . . . . . . . . . . . . . . . . . . . . . 118

7 Modular forms for congruence subgroups 120

7.1 Modular forms and cusp forms with respect to Γ . . . . . . . . . . . . . . . . 120
7.2 Holomorphic Eisenstein series for Γ(N ) . . . . . . . . . . . . . . . . . . . . . 122
7.3 Holomorphic Eisenstein series for Γ1 (N ) and Γ0 (N ) . . . . . . . . . . . . . . 125
7.4 Jacobi’s theorem on sums of four squares . . . . . . . . . . . . . . . . . . . . 127

8 Modular differential equations and vector-valued

modular forms 130
8.1 Modular covariant derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.2 Modular differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.3 Vector-valued modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

9 Modular graph functions and forms 141

9.1 One-loop modular graph functions are Eisenstein series . . . . . . . . . . . . 141
9.2 Maass operators and Laplacians . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.3 Two-loop modular graph functions . . . . . . . . . . . . . . . . . . . . . . . 144
9.4 General modular graph functions and forms . . . . . . . . . . . . . . . . . . 146
9.5 Relating modular graph functions to holomorphic forms . . . . . . . . . . . . 150
9.6 Modular graph functions and iterated modular integrals . . . . . . . . . . . . 152

3
II Extensions and Applications 156
10 Hecke Theory 157
10.1 Definition of Hecke operators . . . . . . . . . . . . . . . . . . . . . . . . . . 157
10.2 Explicit parametrization of equivalence classes SL(2, Z)\Mn . . . . . . . . . 158
10.3 Hecke operators map Mk to Mk . . . . . . . . . . . . . . . . . . . . . . . . 160
10.4 Fourier expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
10.5 Example: the Ramanujan tau function . . . . . . . . . . . . . . . . . . . . . 161
10.6 Multiplicative properties of Hecke operators . . . . . . . . . . . . . . . . . . 162
10.7 Hecke eigenforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
10.8 Hecke operators acting on Maass forms . . . . . . . . . . . . . . . . . . . . . 165
10.9 Hecke operators on vvmfs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.10Physics applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

11 Singular moduli and complex multiplication 174

11.1 Conditions for complex multiplication . . . . . . . . . . . . . . . . . . . . . . 174
11.2 Elliptic functions at complex multiplication points . . . . . . . . . . . . . . . 175
11.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
11.4 j(τ ) as an algebraic integer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
11.5 ϑ-functions at complex multiplication points . . . . . . . . . . . . . . . . . . 180
11.6 The values of E2 , E4 , E6 at the points τ = i, ρ . . . . . . . . . . . . . . . . . 184

12 String amplitudes 186

12.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
12.2 String amplitudes as integrals over moduli space . . . . . . . . . . . . . . . . 188
12.3 Conformal invariance and decoupling negative norm states . . . . . . . . . . 190
12.4 String amplitudes in terms of vertex operators . . . . . . . . . . . . . . . . . 193
12.5 Superstring amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
12.6 Ultraviolet finiteness from modular invariance . . . . . . . . . . . . . . . . . 198
12.7 Effective interactions from the four-graviton amplitude . . . . . . . . . . . . 199
12.8 Genus one in terms of modular graph functions . . . . . . . . . . . . . . . . 201
12.9 Genus two in terms of modular graph functions . . . . . . . . . . . . . . . . 202
12.10Integration over the genus-one moduli space . . . . . . . . . . . . . . . . . . 204
12.11Integration over the genus-two moduli space . . . . . . . . . . . . . . . . . . 210

13 Toroidal compactification 214

13.1 Conformal field theory on flat manifolds . . . . . . . . . . . . . . . . . . . . 214
13.2 Lattices and tori of dimension d . . . . . . . . . . . . . . . . . . . . . . . . . 215
13.3 Fields taking values in a torus . . . . . . . . . . . . . . . . . . . . . . . . . . 216
13.4 T-duality on a circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
13.5 T-duality on a torus Td . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

4
 13.6 Rationality and complex multiplication . . . . . . . . . . . . . . . . . . . . . 223

14 S-duality of Type IIB superstrings 226

14.1 Type IIB supergravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
14.2 From SL(2, R) to SL(2, Z) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
14.3 Low energy effective interactions . . . . . . . . . . . . . . . . . . . . . . . . . 229
14.4 SL(2, Z) duality in Type IIB superstring theory . . . . . . . . . . . . . . . . 230
14.5 Eisenstein series from supersymmetry and S-duality . . . . . . . . . . . . . . 231

15 Dualities in N = 2 super Yang-Mills theories 238

15.1 Super Yang-Mills: states and fields . . . . . . . . . . . . . . . . . . . . . . . 238
15.2 Super Yang-Mills: Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . 241
15.3 Low-energy Lagrangian on the Coulomb branch . . . . . . . . . . . . . . . . 244
15.4 BPS states, monopoles, and dyons . . . . . . . . . . . . . . . . . . . . . . . . 245
15.5 SL(2, Z)-duality of the N = 4 theory . . . . . . . . . . . . . . . . . . . . . . 247
15.6 The Seiberg-Witten Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
15.7 The N = 2∗ theory for gauge group SU (N ) . . . . . . . . . . . . . . . . . . 254
15.8 The N = 2∗ theory for gauge group SU (2) . . . . . . . . . . . . . . . . . . . 257
15.9 Linear quiver chains from a limit of N = 2∗ . . . . . . . . . . . . . . . . . . 261
15.10N = 2 dualities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

16 Basic Galois Theory 272

16.1 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
16.2 Field Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
16.3 Field automorphisms and the Galois Group . . . . . . . . . . . . . . . . . . 274
16.4 The Fundamental Theorem of Galois Theory . . . . . . . . . . . . . . . . . . 277
16.5 Solvability by Radicals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
16.6 Galois Theory in rational conformal field theory . . . . . . . . . . . . . . . . 279

III Appendix 282

A Some arithmetic 283
A.1 Arithmetic mod N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
A.2 Chinese remainder theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
A.3 Solving polynomial equations . . . . . . . . . . . . . . . . . . . . . . . . . . 284
A.4 Quadratic residues and quadratic residue symbols . . . . . . . . . . . . . . . 285
A.5 Gauss sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
A.6 Quadratic reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
A.7 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
A.8 Dirichlet characters and L-functions . . . . . . . . . . . . . . . . . . . . . . . 291

5
B Riemann surfaces 295
B.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
B.2 Metrics and complex structures . . . . . . . . . . . . . . . . . . . . . . . . . 300
B.3 Uniformization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
B.4 Fuchsian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
B.5 Construction of Riemann surfaces via Fuchsian groups . . . . . . . . . . . . 306

C Line bundles on Riemann surfaces 309

C.1 Holomorphic line bundles on a Riemann surface . . . . . . . . . . . . . . . . 309
C.2 Holomorphic sections and the Riemann-Roch theorem . . . . . . . . . . . . . 313
C.3 Vanishing Theorem and dimension formulas . . . . . . . . . . . . . . . . . . 313
C.4 Tensors and spinors on Σ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
C.5 Proof of the Riemann-Roch theorem . . . . . . . . . . . . . . . . . . . . . . 316
C.6 Proof of the vanishing theorem . . . . . . . . . . . . . . . . . . . . . . . . . 317
C.7 The dimension of moduli space . . . . . . . . . . . . . . . . . . . . . . . . . 318

D Higher genus ϑ-functions and meromorphic forms 320

D.1 The Siegel half space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
D.2 The Riemann theta function . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
D.3 The Jacobian, Abel map, and Riemann vanishing theorem . . . . . . . . . . 322
D.4 The prime form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.5 Holomorphic differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.6 Meromorphic differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.7 The bc system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326

6
1 Introduction
Integers form a group under addition but not under multiplication. However, matrices with
integer entries can form groups under multiplication. For example 2 × 2 matrices of unit
determinant and integer entries a, b, c, d ∈ Z,
 
a b
ad − bc = 1 (1.0.1)
c d

form the group SL(2, Z) under multiplication, referred to as the modular group. Functions
and differential forms that are invariant under SL(2, Z) are referred to as modular functions
and modular forms, respectively. Modular functions generalize periodic functions and elliptic
functions, to which they are intimately related. In turn, modular functions are special cases
of automorphic functions which are invariant under more general arithmetic groups.
In Mathematics, the study of elliptic integrals goes back to Euler, Legendre, and Abel,
while the theory of elliptic functions was developed by Gauss, Jacobi, and Weierstrass and
was motivated in part by questions ranging from number theory to the solvability of algebraic
equations by radicals. Riemann developed the theory of general Riemann surfaces and
generalized elliptic functions to higher rank ϑ-functions on surfaces of higher genus.
The development of modular forms dates back to Eisenstein, Kronecker, and Hecke.
Automorphic functions and forms were studied by Fuchs, Fricke, Klein, and Poincaré. In
modern times, amongst many other developments, the Taniyama-Shimura-Weil conjecture
ultimately led to the proof of Fermat’s Last Theorem by Wiles and Taylor, and to a proof
of the Modularity Theorem by Breuil, Conrad, Taylor, and Diamond. A fundamental role
was played by modular forms and quasi-modular forms in the solution to the sphere packing
problem in eight dimensions by Viazovska, work for which she was awarded the Fields Medal
in 2022.
In Physics, elliptic integrals arose already in the simplest classical mechanical problems,
such as the pendulum, and were studied in this context as far back as the 18th century.
The solution of various boundary problems in electrostatics and fluid mechanics via elliptic
functions, or more generally automorphic functions, further stimulated the mathematical
development of these subjects a century ago. More recently, Riemann ϑ-functions were
found to provide the solution to a host of completely integrable mechanical systems.
The modular group SL(2, Z) made its appearance in string theory in 1972 when Shapiro
identified it as a symmetry of the integrand that defines the one-loop closed bosonic string
amplitude. Shapiro defined the amplitude as the integral over the quotient of the Poincaré
upper half-plane by SL(2, Z) and argued that the amplitude thus obtained is free of the
short-distance divergences that arise in quantum field theory. This fundamental observation
extends to the five perturbative superstring theories, and to all loop orders, provided that the
group SL(2, Z) is replaced by the modular group Sp(2h, Z) for genus h Riemann surfaces.

7
The absence of UV divergences is principally responsible for advocating string theory as the
uniquely viable candidate for a consistent quantum theory of gravity.
A second context in which the modular group SL(2, Z) arose in high-energy physics is
Yang-Mills gauge theory in four space-time dimensions. The Standard Model of Particle
Physics is a Yang-Mills theory with gauge group SU (3) × SU (2) × U (1) in which the Higgs
mechanism generates the masses of quarks, leptons, and the gauge bosons W ± and Z via
spontaneous symmetry breaking. A particular class of Yang-Mills theories are those with
simple or semi-simple gauge groups that are spontaneously broken to a subgroup containing
at least one U (1)-factor. This effective U (1) gauge theory contains a photon and various
electrically charged particles, as standard Maxwell theory does. However, unlike standard
Maxwell theory, it also contains magnetically charged particles that arise as ‘t Hooft Polyakov
magnetic monopoles. Goddard, Nuyts, and Olive conjectured in 1977 that, under certain
conditions, such theories exhibit an electric-magnetic duality which swaps electric particles
and magnetic monopoles. Generalizing the construction to dyons which carry both elec-
tric and magnetic charges shows that the duality is actually captured by the duality group
SL(2, Z), or a congruence subgroup thereof. A concrete realization of electric-magnetic dual-
ity, referred to as Montonen-Olive duality, is provided by Yang-Mills theories with extended
supersymmetry, and led to Seiberg-Witten theory in 1994.
A third context in which SL(2, Z) arises takes us back to superstring theory. The five
superstring theories that admit a perturbative description in ten-dimensional Minkowski
space-time are the Type IIA and Type IIB theories with the maximal number of 32 super-
symmetries, and the Type I and two Heterotic string theories with 16 supersymmetries. The
massless states of each one of these superstring theories are described, at tree-level, by an as-
sociated supergravity theory, which is an extension of the Einstein-Hilbert theory of general
relativity in ten dimensions. For example, Type IIB supergravity contains, in addition to the
space-time metric field, a complex axion-dilaton scalar field τ , anti-symmetric tensor fields
of rank 2 and 4, as well as their fermionic partners. The imaginary part of τ is related to the
inverse string coupling and must be positive on physical grounds. Möbius transformations
on the field τ ,

aτ + b
τ→ ad − bc = 1 (1.0.2)
cτ + d
with real parameters a, b, c, d form the group SL(2, R) and are a symmetry of Type IIB
supergravity, when accompanied by suitable transformations on the other fields. The field
τ takes values in the coset SL(2, R)/SO(2). However, the SL(2, R) symmetry of tree-level
supergravity has an anomaly and is broken to its discrete subgroup SL(2, Z), the so-called
S-duality symmetry of Type IIB superstring theory. Type IIB string backgrounds whose
fields are related by an SL(2, Z) transformation are really identical, so that the space of
inequivalent theories is given by the double coset SL(2, Z)\SL(2, R)/SO(2). The implica-

8
tions of this conjectured symmetry were fully appreciated only with the discovery of NS- and
D-brane solutions in the 1990s.
A fourth context where SL(2, Z) and higher arithmetic groups emerge is when string
theory is considered on a space-time of the form R10−d × Td where Td is a d-dimensional flat
torus. Such a setup is often referred to as toroidal compactification. While the Fourier anal-
ysis of supergravity fields on a torus produces only momentum modes, the compactification
of a string theory produces both momentum and winding modes. The winding modes are re-
sponsible for quintessentially string-theoretic discrete symmetries, referred to as T-dualities,
that have no counterpart in quantum field theory. For example, Type IIB superstring theory
on a circle of radius R is T -dual to Type IIA superstring theory on a circle of radius α0 /R.
Toroidal compactification converts some of the components of the bosonic fields, such as
the metric, into scalar fields. These new scalar fields combine with the axion-dilaton field
τ of Type IIB to live on a larger coset space which enjoys a larger arithmetic symmetry
group. For increasing dimensions d of the toroidal compactification, we have the following
arithmetic symmetry groups G(Z) and corresponding coset spaces G(R)/K(R) starting with
ten-dimensional Type IIB superstring theory for d = 0,

d=0 SL(2, Z) SL(2, R)/SO(2)

d=1 SL(2, Z) SL(2, R)/SO(2)
d=2 SL(2, Z)×SL(3, Z) SL(2, R)×SL(3, R)/(SO(3)×SO(2))
d=3 SL(5, Z) SL(5, R)/SO(5)
d=4 SO(5, 5, Z) SO(5, 5, R)/(SO(5)×SO(5))
d=n−1 En,n (Z) En,n (R)/SO(2n)

On the last line En,n for n = 6, 7, 8 denotes a particular real form of the corresponding
complexified Lie groups E6 , E7 , E8 . Compactification on Calabi-Yau manifolds or orbifolds
exhibit similarly quintessential string theoretic relations that go under the name of mirror
symmetry, which will not be discussed in these lectures.
A fifth context where modular symmetry is of great importance is in two-dimensional
conformal field theory. Cardy linked S-duality in a conformal field theory to its unitarity
properties, while Erik Verlinde provided a powerful restriction on the operator product ex-
pansion coefficients of conformal primary fields in terms of representations of the modular
group. More recently, Hecke operators were used to relate different conformal field theories to
one another. Besides describing the worldsheet theory of strings, two-dimensional conformal
field theory models the critical behavior of a number of simple, experimentally realizable,
systems such as the critical Ising and n-states Potts models. Experimental realizations of
these systems tend to be on lattices of trivial topology, and one might naively expect that
such real-world systems do not possess any manifest modular properties. However, it turns
out that modular symmetry gives rise to important constraints even for topologically trivial

9
systems. The use of modular symmetry to constrain the basic data of a conformal theory
is a program referred to as the modular bootstrap, and has led to a classification of certain
simple families of conformal field theories. Thanks to the gauge-gravity correspondence,
these constraints on conformal field theory also give rise to constraints on theories of gravity
in three-dimensional anti-de Sitter space, which indeed was one of the original motivations
for the modular bootstrap program.
The goal of these lecture notes is to provide the reader with some of the mathematical
tools that are of most immediate use in addressing the various directions of research in physics
that are related to modular forms. We have attempted to make the notes as self-contained
as possible as far as the mathematics is concerned. We shall assume familiarity with little
more than complex analysis and some basic group theory and differential geometry. These
lecture notes are not a review: in no way have we attempted to present a complete survey of
the literature either in Mathematics or in Physics. The task would be insurmountable: each
of the Physics directions mentioned above counts thousands of research papers, while the
corresponding directions in Mathematics often date back more than 150 years. In particular,
many beautiful and important directions of inquiry have not been addressed at all here. They
include the modular and arithmetic properties of orbifold and Calabi-Yau compactifications,
F-theory, Moonshine, three-dimensional topological field theory, topological modular forms,
black-hole microstate counting, and the structure of super moduli space and related topics.
Perhaps the discussion of some of these topics will be included in later, revised, versions.

Organization
These lecture notes are organized in three parts. The first part provides an introduction to
elliptic functions and modular forms, as well as to numerous generalizations such as quasi-
modular forms, almost-holomorphic modular forms, non-holomorphic modular forms, mock
modular forms, and quantum modular forms. Full sections are dedicated to a discussion of
modular forms for congruence subgroups, vector-valued modular forms, and modular graph
functions. In the second part, various mathematical and physical applications and extensions
of the material of the first part are provided. The mathematical extensions include Hecke
theory, complex multiplication, and Galois theory, while the physical applications include
string perturbation theory, S-duality in Type IIB superstring theory, dualities in super-
Yang-Mills theories with extended supersymmetry, and Seiberg-Witten theory. Finally, the
third part consists of four appendices of material that is central to the material of the
core sections, but may be read independently thereof. This includes an introduction to
arithmetic, Riemann surfaces, line bundles on Riemann surfaces, and higher rank ϑ-functions
and holomorphic and meromorphic forms on Riemann surfaces of higher genus. Each part
is preceded by a one-page organizational summary.
Bibliographical notes are provided at the end of each section and appendix. For Mathe-

10
matics references, many excellent textbooks are available on elliptic functions, ϑ-functions,
modular forms, Riemann surfaces, and Galois theory. We shall provide references to the
expositions that we have found particularly useful. For Physics references, we shall refer as
much as possible to the textbooks, review papers, and lecture notes that we believe will be
useful to the reader. We shall refer to original research papers whenever they are classics,
or when the material is not readily available in textbooks, reviews, and lecture notes.

Acknowledgments
We are very happy to acknowledge collaborations on a variety of related topics with Sri-
ram Bharadwaj, Gordon Chan, Bill Duke, Thomas Dumitrescu, Nick Geiser, Jan Gerken,
Michael Green, Ömer Gürdogan, Michael Gutperle, Martijn Hidding, Axel Kleinschmidt,
Zohar Komargodski, Per Kraus, Igor Krichever, Ying-Hsuan Lin, Carlos Mafra, Julio Parra-
Martinez, Mario Martone, Emily Nardoni, Kantaro Ohmori, Eric Perlmutter, D.H. Phong,
Boris Pioline, Oliver Schlotterer, Shu-Heng Shao, Sahand Seifnashri, Pierre Vanhove, Gabi
Zafrir, and Yunqin Zheng.
We are grateful to Oliver Schlotterer for helpful comments on the manuscript. We wish to
thank Oliver Schlotterer and Terry Tomboulis for much appreciated encouragement during
this project, and Yi Li, John Miller, and Raphael Rouquier for helpful discussions at the
earliest stages of this project.
These notes are an outgrowth of lectures presented for the 2015 workshop “Supermoduli”
at the Simons Center for Geometry and Physics; for the 2017 UCLA graduate course on
Physics and Geometry; for the “Amplitudes 2018 Summer School” at the Center for Quantum
Mathematics and Physics at UC Davis; and for the 2019 conference “Arithmetic, Geometry,
and Modular Forms in honor of Bill Duke” at the ETH in Zurich.
The research of ED was supported in part by the National Science Foundation under
research grants PHY-16-19926 and PHY-19-14412, and by a Fellowship from the Simons
Foundation during the academic year 2017-18. ED’s visits to the KITP in Santa Barbara in
2017 and 2022 were supported in part by the NSF grant PHY-17-48958, and funding from
the Simons Foundation, which are gratefully acknowledged.

11
Part I
Modular forms and their variants
In the first part of these lecture notes, we introduce basic concepts, properties, and applica-
tions of holomorphic and non-holomorphic modular functions and forms and their variants.
In section 2 we review periodic functions, the Poisson resummation formula, and appli-
cations to analytic continuations that will be needed in the sequel. We relate Weierstrass
and Jacobi elliptic functions to Jacobi ϑ-functions and exhibit their connection with elliptic
curves, Abelian differentials, and Abelian integrals.
In section 3, we introduce the modular group SL(2, Z), discuss its structure, and review
its action on the Poincaré upper half-plane. We proceed to holomorphic and meromorphic
modular functions, modular forms, and cusp forms, and realize them in terms of holomorphic
Eisenstein series. The valence formula is used to provide the dimensions and the generators
of the ring of modular forms of a given weight. The discriminant cusp form ∆, the j-function,
and the Dedekind η-function are introduced and related to Jacobi ϑ-functions.
In section 4 we introduce variants of modular forms for SL(2, Z), including quasi-modular
forms, almost-holomorphic modular forms, non-holomorphic modular forms (such as non-
holomorphic Eisenstein series and Maass forms), mock modular forms, and quantum modular
forms. We conclude section 4 by presenting the spectral decomposition of square-integrable
modular functions into non-holomorphic Eisenstein series and cusp forms.
In section 5, we illustrate the use of elliptic functions and modular forms in the physical
context of two-dimensional conformal field theory. The construction of Green functions,
correlation functions, and functional determinants is presented for scalar and spinor fields
on a torus and related to the Kronecker limit formulas.
In section 6 the classic congruence subgroups of SL(2, Z) and the associated modular
curves are reviewed. The elliptic points and cusps are counted and used to derive formulas
for the genera of modular curves. In section 7 we construct modular forms for the classic
congruence subgroups, and apply the results to proving the theorems of Lagrange and Jacobi
on representing positive integers by sums of four squares. In section 8 we further generalize
to modular forms that transform under multi-dimensional representations of SL(2, Z) but
whose individual components are invariant under a congruence subgroup. Such vector-valued
modular forms satisfy modular differential equations, a notion which is reviewed.
In section 9 we introduce modular graph functions and forms. Physically, they arise
in the low energy expansion of superstring amplitudes where they emerge in integrands
on the moduli space of compact Riemann surfaces. Mathematically, they provide non-
trivial generalizations of non-holomorphic Eisenstein series and multiple zeta values, and are
intimately related with elliptic polylogarithms and iterated modular integrals.

12
2 Elliptic functions
In this section, we shall begin with a review of periodic functions of a real variable, Poisson
resummation, the unfolding trick, and a simple applications to analytic continuation of the
Riemann ζ-function. Next, we review simply periodic functions of a complex variable from
the stand point of summation over image charges, show how their differential equation and
addition formulas arise as a consequence, and provide a simple application to evaluating
the Riemann ζ-function at even positive integers. Elliptic functions are defined and their
general properties regarding zeros and poles are obtained. The classic constructions of elliptic
functions, based on the Weierstrass ℘ function, the Jacobi elliptic functions sn, cn, dn and
Jacobi ϑ-functions are presented. Finally, the elliptic function theory developed here is
placed in the framework of elliptic curves, Abelian differentials, and Abelian integrals.

2.1 Periodic functions of a real variable

A function f : R → C is periodic with period 1 if for all x ∈ R it satisfies,
f (x + 1) = f (x) (2.1.1)
Equivalently, f is a function of the circle S 1 = R/Z → C. The function f is completely
specified by the values it takes in the unit interval [0, 1[. There are various other useful
systematic methods for constructing periodic functions, some of which we shall now review.
If a function g : R → C decays sufficiently rapidly at ∞ then a periodic function may be
constructed by the method of images,
X
f (x) = g(x + n) (2.1.2)
n∈Z
2πimx
The functions e for m ∈ Z are all periodic with period 1, and in fact form a basis for
the square-integrable periodic functions L2 (S 1 ). Their orthogonality and completeness are
manifest from the following relations (in the sense of distributions),
Z 1
0
dx e2πimx e−2πim x = δm,m0
0
X X
e2πimx e−2πimy = δ(x − y + n) (2.1.3)
m∈Z n∈Z
R
where δ(x) is the Dirac δ-function normalized to R dy δ(x − y)f (y) = f (x). Any periodic
function f with period 1 may be decomposed into a Fourier series,
X Z 1
f (x) = fm e2πimx
fm = dx f (x) e−2πimx (2.1.4)
m∈Z 0

We note that in the mathematics literature, one often uses the notation e(x) = e2πix .

13
2.1.1 Unfolding and Poisson summation formulas
A simple but extremely useful tool, which will be substantially generalized later on in the
form of trace formulas, is the unfolding trick. If a function g : R → C decays sufficiently
rapidly at ∞ to make its integral over R converge absolutely, then the unfolding trick gives,
XZ 1 Z ∞
dx g(x + n) = dx g(x) (2.1.5)
n∈Z 0 −∞

Combining the method of images and Fourier decomposition, we construct a periodic func-
tion f from a non-periodic function function g using (2.1.2), and then calculate the Fourier
coefficients fm using the unfolding trick,
Z 1 X Z ∞
−2πimx
fm = dx e g(x + n) = dx g(x) e−2πimx (2.1.6)
0 n∈Z −∞

The Fourier transform of g will be denoted by ĝ and is given by,

Z ∞
ĝ(y) = dx g(x) e−2πixy (2.1.7)
−∞

so that fm = ĝ(m). Therefore, the function f may be expressed in two different ways,
X X
f (x) = g(x + n) = ĝ(m) e2πimx (2.1.8)
n∈Z m∈Z

Setting x = 0 (or any integer) gives the Poisson summation formula.

Theorem 2.1 A function g and its Fourier transform ĝ, defined in (2.1.7) obey the Poisson
summation formula,
X X
g(n) = ĝ(m) (2.1.9)
n∈Z m∈Z

assuming both sums are absolutely convergent.

An immediate application is to the case where g is a Gaussian. It will be convenient to
normalize Gaussians as follows,
2 1 2
g(x) = e−πtx ĝ(y) = √ e−πy /t (2.1.10)
t
The formulas are valid as long as Re (t) > 0, and may be continued to Re (t) = 0. The
corresponding Poisson summation formula then reads,
X 2 1 X −πm2 /t
e−πtn = √ e (2.1.11)
n∈Z
t m∈Z

We shall see later on that this relation admits an important generalization to Jacobi ϑ-
functions, and corresponds to a special case of a modular transformation.

14
2.1.2 Application to analytic continuation
The Riemann ζ-function, known already to Euler, is defined by the following series,
∞
X 1
ζ(s) = (2.1.12)
n=1
ns

which converges absolutely for Re (s) > 1 and thus defines a holomorphic functions of s
in that region. Its arithmetic significance derives from the fact that it admits a product
representation, due again to Euler,
Y  −1
1
ζ(s) = 1− s (2.1.13)
p prime
p

The Riemann ζ-function is just one example of the family of zeta functions that may be
associated with certain classes of self-adjoint operators whose spectra are discrete, free of
accumulation points, and bounded from below (we shall take them to be bounded from
below by zero without loss of generality). Consider such a self-adjoint operator H, and
its associated spectrum λn of discrete real eigenvalues with n ∈ N and λ1 > 0.1 We may
associate a ζ-function to the operator H as follows,
∞
X 1 −s


ζH (s) = = Tr H (2.1.14)
λs
n=1 n

provided λn grows with n sufficiently fast, such as λn ∼ nα for α > 0 at large n. If the growth
is λn ∼ n for large n then the series is absolutely convergent for Re (s) > 1, and defines a
holomorphic function in s in that region. The Riemann ζ-function may be associated in
this way with the Hamiltonian of either the harmonic oscillator or the free non-relativistic
particle in an interval with periodic boundary conditions.
It is a famous result of Riemann that the function ζ(s) may be analytically continued
to the entire complex plane, has a single pole at s = 1, and admits a functional reflexion
relation. The result may be derived using the Poisson summation relation (2.1.11). Consider
the function ξ(s) defined by the integral,
Z ∞ !
dt s X −πtn2
ξ(s) = t2 e −1 (2.1.15)
0 t n∈Z

For large t the integral is convergent for any s in the complex plane, but for small t it is
convergent only for Re (s) > 1, as may be seen by using (2.1.11) for t → 0. For Re (s) > 1,

1
Throughout N = {1, 2, 3, · · · } denotes the set of positive integers.

15
the integral defining ξ may be evaluated by integrating term by term, where the contribution
of the n = 0 is cancelled by the subtraction of 1, and we find,
s
s
ξ(s) = 2π − 2 Γ ζ(s) (2.1.16)
2
To construct a form that may be analytically continued, we partition the integral into the
intervals [0, 1] and [1, ∞]. We leave the integral over [1, ∞] unchanged, but change variables
t → 1/t in the integral over [0, 1], transforming it into an integral over [1, ∞],
Z ∞ ! Z !
∞
dt − s X −πn2 /t dt s X −πtn2
ξ(s) = t 2 e −1 + t2 e −1 (2.1.17)
1 t n∈Z 1 t n∈Z

In the first integral, we rewrite the sum with the help of the Poisson relation (2.1.11),
Z ∞ ! Z !
∞
dt 1−s X −πtn2 1 dt s X −πtn2
ξ(s) = t 2 e −√ + t2 e −1 (2.1.18)
1 t n∈Z
t 1 t n∈Z

The pure power integrals may be performed exactly and admit an obvious analytic contin-
uation with poles at s = 0 and s = 1, so that the final result is given by,
Z ∞  !
2 dt s 1−s
 X 2
ξ(s) = − + t2 + t 2 e−πtn − 1 (2.1.19)
s(1 − s) 1 t n∈Z

The integral gives a function that is holomorphic throughout C, and the entire expression
manifestly admits the functional relation,

ξ(s) = ξ(1 − s) (2.1.20)

which in turn gives the functional relation of the Riemann ζ-function. Immediate conse-
quences are that the residue of ζ(s) at s = 1 is one, while ζ(s) is holomorphic elsewhere
throughout C. It is also immediate that ζ(−2n) = 0 for all n ∈ N.

2.2 Periodic functions of a complex variable

Before turning to elliptic functions, we look at how complex analytic periodic functions and
trigonometric functions arise naturally from applying the method of images. We start with
a meromorphic function g : C → C that has a single simple pole g(z) = 1/z and use the
method of images to construct a periodic meromorphic function,
X 1
f (z) = (2.2.1)
n∈Z
z+n

16
The sum is not absolutely convergent, but it is conditionally convergent, which means that
its value depends on the order in which we choose to sum the terms. In quantum field theory,
we refer to this process as “regularizing” the infinite sum. A natural way to do this is by
requiring f (−z) = −f (z) and taking a symmetric limit of a finite sum with this property.
Equivalently we may group terms with n together under the summation,
N ∞  
X 1 1 X 1 1
f (z) = lim = + + (2.2.2)
N →∞
n=−N
z+n z n=1 z+n z−n

The meromorphic nature of f allows us to evaluate the sum by finding a known function
with identical poles with unit residue at every integer, identical asymptotic behavior at ∞,
and identical symmetry property f (−z) = −f (z). The result is the following formula,

cos πz 1 + e2πiz
f (z) = π = −iπ (2.2.3)
sin πz 1 − e2πiz
Integrating both expressions for f and determining the integration constant by matching the
behavior of z = 0 gives Euler’s product formula for the sin-function,
∞ 
z2
Y 
sin πz = πz 1− 2 (2.2.4)
n=1
n

Linearity of the method of images allows us to treat any function g that admits a partial
fraction decomposition into poles in z in an analogous manner.
We can also proceed by deriving a differential equation for f directly from its series
representation in (2.2.2). For |z| < 1, we may expand the second representation in (2.2.2) in
a convergent power series in z,
∞
1 X
f (z) = − 2 z 2m−1 ζ(2m) (2.2.5)
z m=1

Its derivative is given by,

∞
0 1 X
f (z) = − 2 − 2 (2m − 1)z 2m−2 ζ(2m) (2.2.6)
z m=1

The combination f 0 + f 2 is free of poles, bounded in C, and therefore constant by Liouville’s

theorem. The constant is computed by evaluating the series near z = 0, and we find,

f 0 + f 2 = −6 ζ(2) = −π 2 (2.2.7)

17
The differential equation may be solved by quadrature as it is equivalent to,
Z Z Z
df df
− = −2πi dz (2.2.8)
f − iπ f + iπ

Interpreting z(f ) as the inverse function of f (z) we see that z(f ) must be multiple-valued
as a function of f , since the line integrals in df are logarithms and therefore multiple-valued
around the poles at f = ±iπ. Since z(f ) is multiple-valued with period 1, its inverse
function f (z) must be periodic with period 1. The integral relation of (2.2.8), along with the
symmetry condition f (−z) = −f (z) used to fix the integration constant, reproduces (2.2.3).
One application is to the derivation of the addition formulas for trigonometric functions.
Using only the fact f (z) of (2.2.3) has a simple pole at z = 0 and is periodic with period 1,
we can derive the addition formulas for trigonometric functions. Indeed, for fixed w, the
function f (z + w) has simple poles at z = −w (mod 1) with unit residue, which match those
poles in f 0 (z)/(f (z) + f (w)). However, the latter also has simple poles at z ∈ Z which may
be matched by the function f (z). Implementing also the symmetry under swapping z and w,
we arrive at the following identity,

f 0 (z) + f 0 (w)
2f (z + w) − − f (z) − f (w) = 0 (2.2.9)
f (z) + f (w)

from which every addition theorem for elliptic functions may be deduced.

2.2.1 Application to calculating ζ(s) at special values

As another application, we evaluate the Riemann ζ-function at even positive integers by
equating the representations for f (z) given in (2.2.3) and (2.2.5),
∞
1 + e2πiz 1 X
−iπ = − 2 z 2m−1 ζ(2m) (2.2.10)
1 − e2πiz z m=1

It follows that ζ(2m) is π 2m times a rational number. To see this, we express the power series
in z of the left side in terms of Bernoulli numbers Bm ∈ Q, which are defined as follows,
∞
w X Bm m
= w (2.2.11)
ew − 1 m=0 m!

Clearly, we have B0 = 1, B1 = −1/2, and B2m+1 = 0 for all m ∈ N. The next few non-zero
values for the Bernoulli numbers are as follows,
1 1 1 5 691
B2 = B4 = B8 = − B6 = B10 = B12 = − (2.2.12)
6 30 42 66 20730
18
An equivalent formula for Bernoulli numbers of even index is as follows,
∞
w ew + 1 X B2m 2m
× w = w (2.2.13)
2 e − 1 m=0 (2m)!
Setting w = 2πiz and identifying term-by-term in the series (2.2.10) gives,
B2m
ζ(2m) = −(2πi)2m (2.2.14)
2(2m)!
which, in turn, gives the following low order values,
π2 π4 π6 π8 π 10
ζ(2) = ζ(4) = ζ(6) = ζ(8) = ζ(10) = (2.2.15)
6 90 945 9450 93555
By using the functional relation, we may also compute the ζ-function at other values of
interest in physics, such as
1 1 1
ζ(0) = − , ζ 0 (0) = − ln(2π), ζ(−1) = − (2.2.16)
2 2 12
The last formula is key to the argument that superstrings live in ten space-time dimensions !

2.3 Elliptic functions

Periodic functions of one real variable may be generalized to functions of several real variables
by considering functions that are periodic in each real variable separately. Elliptic functions
are periodic functions in two real variables, or equivalently doubly periodic functions in a
complex variable z, supplemented with the requirement of complex analyticity in z.
Denoting the two periods in the complex variable z by ω1 , ω2 ∈ C,2 we require that
the periods be linearly independent as vectors in R2 so that ω2 /ω1 6∈ R. The periods then
generate a two-dimensional lattice Λ defined by (see Figure 1),
Λ = Zω1 + Zω2 (2.3.1)
The lattice Λ is an Abelian group under addition. The quotient C/Λ of the complex plane
by the lattice Λ is torus of dimension two over R, and also an Abelian group. The torus
C/Λ may be represented in C by a parallelogram Pµ anchored at a point µ ∈ C on which
opposite sides are pairwise identified with one another (see Figure 1),
Pµ = {µ + t1 ω1 + t2 ω2 , 0 ≤ t1 , t2 ≤ 1} (2.3.2)
A function of z which is doubly periodic with periods ω1 , ω2 is periodic under the lattice Λ,
or equivalently is a well-defined function on the torus C/Λ. Collecting these ingredients, we
obtain the following definition.
2
In Bateman [1], the periods are denoted by 2ω and 2ω 0 respectively, and are related to the periods
defined here by ω1 = 2ω and ω2 = 2ω 0 . Our notation for the periods ω1 and ω2 is not to be confused with
the half-periods ωα used in equation (18) in section 13.12 of [1].

19
 • •
• •
• • •
• •
ω2 • • •
• • •
•
• • •
• •
Pµ •
• • •
• • •o ω1
• • • •
• µ •
• •
• •

Figure 1: The points of the lattice Λ are represented in the complex plane by black dots.
The lattice Λ is generated by the periods ω1 , ω2 so that Λ = Zω1 ⊕ Zω2 . The boundary ∂Pµ
of the fundamental parallelogram Pµ is drawn in red, where µ has been chosen so that ∂Pµ
contains neither periods, nor half periods represented by blue dots.

Definition 2.2 A meromorphic function f : C → C is elliptic with periods ω1 , ω2 ∈ C with

ω2 /ω1 6∈ R if any one of the following equivalent conditions holds,

• f is periodic with periods ω1 , ω2 , namely f (z + ω1 ) = f (z + ω2 ) = f (z) for all z ∈ C;

• f is invariant under translations by the lattice Λ, namely f (z + ω) = f (z) for all ω ∈ Λ
and for all z ∈ C;
• f is single-valued on the torus C/Λ.

It is often as single-valued functions on a torus that elliptic functions appear in physics.

Sums, products, and inverses of elliptic functions for a given lattice Λ are elliptic func-
tions for the same lattice Λ, and the constant function with value 1 is the identity under
multiplication, so that elliptic functions for a given Λ form a field.
As we shall establish shortly, elliptic functions are essentially determined by their zeros
and poles, subject to certain conditions. For example, Liouville’s theorem (which states that
a holomorphic function on C which is bounded on C must be constant), implies that an
elliptic function without poles must be a constant function. To obtain the relations between
zeros and poles of elliptic functions more generally, we proceed by defining the residue
function resf (w) to be the residue of f at the point w, and the order function ordf (w) to be
the order of f at the point w, namely the integer such that f (z)(z − w)−ordf (w) is a finite
non-zero constant as z → w.

20
Theorem 2.3 The residues and orders of an arbitrary elliptic function f with lattice Λ and
corresponding fundamental parallelogram Pµ satisfy the following relations,
X
resf (w) = 0
w∈Pµ
X
ordf (w) = 0
w∈Pµ
X
w · ordf (w) = 0 (mod Λ) (2.3.3)
w∈Pµ

These relations are proven by integrating the functions f (z), f 0 (z)/f (z), and zf 0 (z)/f (z)
over the boundary ∂Pµ , respectively. When considering a function f on a torus C/Λ, it will
often be convenient to choose µ such that no zeros or poles of f lie on ∂Pµ .
In two-dimensional electro-statics, the electric potential Φ may be viewed as a function
of z and z̄, and is a harmonic function away from the support of the electric charges. The
electric field E = −∇Φ may be decomposed into derivatives with respect to z and z̄, so that
Ez = −∂z Φ is a holomorphic function away from the charges. Considering a distribution
of point-like charges (without accumulation points) on a torus C/Λ, the electric field is an
elliptic function for the lattice Λ. The first condition in (2.3.3) requires the sum of all the
charges to vanish, and must hold on any compact space. The second and third conditions are
specific to the torus. The conditions (2.3.3) imply that there exists no elliptic function with
just one simple pole in Pµ , corresponding to the fact that the torus does not support a single
point charge, but there do exist elliptic functions with two simple poles at arbitrary positions
and opposite residues corresponding in electro-statics with two opposite point charges. There
also exist elliptic functions with a single double pole (which may be considered as a limit of
the case with two simple poles) corresponding to the electric field of an electric dipole.

2.3.1 Construction by the method of images

Just as periodic functions may be constructed using the method of images in (2.1.2), an
elliptic function may be constructed by using the method of images for the lattice Λ with
periods ω1 , ω2 and ω2 /ω1 6∈ R. If a meromorphic function g : C → C decays sufficiently
rapidly at ∞ then an elliptic function f may be constructed by the method of images,
X X
f (z) = g(z + ω) = g(z + mω1 + nω2 ) (2.3.4)
ω∈Λ m,n∈Λ

Any function g(z) that is bounded by |g(z)| ≤ |z|−2−ε for ε > 0 satisfies this criterion. The
most famous example is given by the Weierstrass function, in terms of which all other elliptic
functions may be expressed, as will be explained in the subsequent subsection.

21
2.4 The Weierstrass elliptic function
All elliptic functions for a lattice Λ may be built up from the Weierstrass elliptic function,
which has one double pole (mod Λ) and no other poles. The Weierstrass elliptic function
℘(z|Λ) for the lattice Λ may be constructed with the method of images. While it may be
natural to consider the sum over ω ∈ Λ of translates (z +ω)−2 of the double pole z −2 , actually
this series converges only conditionally. We shall learn how to handle such series later on.
Here we shall define ℘ by the manifestly convergent series over the lattice Λ0 = Λ \ {0},

1 X 1 1

℘(z|Λ) = 2 + − (2.4.1)
z ω∈Λ0
(z + ω)2 ω 2

The function ℘ is even in z since the lattice summation is invariant under Λ0 → −Λ0 .

2.4.1 The field of elliptic functions in terms of ℘ and ℘0

We start by describing the field of elliptic functions for a fixed lattice Λ in terms of ℘(z|Λ) for
even functions in z. The elliptic function ℘(z|Λ) − ℘(w|Λ), viewed as a function of z for fixed
w 6∈ Λ, has a double pole at z = 0 and no other poles, and therefore must have two zeros in
view of the second relation in (2.3.3). For generic w, the two zeros z = ±w are distinct and
must be the only two zeros. However, for the non-generic case where w = −w (mod Λ), the
zero is double. There are exactly four points in Pµ with w ≡ −w (mod Λ), namely the half
periods 0, ω1 /2, ω2 /2, and ω3 /2 (mod Λ), where ω3 = ω1 + ω2 . At the three non-zero half
periods w, the function ℘(z|Λ) − ℘(w|Λ) has a double zero in z.
The first key result is that every even elliptic function f for the lattice Λ is a rational
function of ℘(z|Λ), given by,
Y  ordf (w)
f (z) = ℘(z|Λ) − ℘(w|Λ) (2.4.2)
w∈Pµ

which may be established by matching poles and zeros. To represent elliptic functions which
are odd under z → −z, or to describe functions without definite parity, we differentiate ℘,
X 1
℘0 (z|Λ) = −2 (2.4.3)
ω∈Λ
(z + ω)3

This series is absolutely convergent for z 6∈ Λ and we have ℘0 (−z|Λ) = −℘0 (z|Λ). The
function ℘0 (z|Λ)2 is even and has a single pole of order 6 at z = 0, and is therefore a
polynomial of degree 3 in ℘(z|Λ). To determine this polynomial, we proceed by analogy
with the trigonometric case of subsection 2.2. We expand both ℘ and ℘0 in powers of z near

22
z = 0 in convergent series,
∞
1 X
℘(z|Λ) = 2 + (k + 1)Gk+2 (Λ) z k
z k=1
∞
2 X
℘0 (z|Λ) = − 3
+ k(k + 1)Gk+2 (Λ) z k−1 (2.4.4)
z k=1

The coefficients Gk are given by the convergent series,

X 1
Gk (Λ) = k≥3 (2.4.5)
ω∈Λ0
ωk

The reflection symmetry −Λ = Λ implies,

G2k+1 (Λ) = 0 k∈N (2.4.6)

so that the expansions to low order are as follows,

1
℘(z|Λ) = + 3 G4 (Λ)z 2 + 5 G6 (Λ)z 4 + O(z 6 )
z2
2
℘0 (z|Λ) = − 2 + 6 G4 (Λ)z + 20 G6 (Λ)z 3 + O(z 5 ) (2.4.7)
z
The pole in z of order six, which is present in both ℘0 (z|Λ)2 and ℘(z|Λ)3 , is cancelled in
the combination ℘0 (z|Λ)2 − 4℘(z|Λ)3 , whose remaining pole is of order two. By matching all
terms of positive or zero power in z and then using Liouville’s theorem, we find the relation,

℘0 (z|Λ)2 = 4℘(z|Λ)3 − g2 (Λ)℘(z|Λ) − g3 (Λ) (2.4.8)

where it is conventional to introduce the notation,

g2 (Λ) = 60 G4 (Λ) g6 (Λ) = 140 G6 (Λ) (2.4.9)

We may assemble these results in the following theorem.

Theorem 2.4 Every elliptic function f (z) for a lattice Λ may be expressed as a rational
function of the Weierstrass functions ℘(z|Λ) and ℘0 (z|Λ) for lattice Λ, which equivalently
may be reduced to a linear function in ℘0 (z) using the relation (2.4.8).

To prove the theorem, we decompose f (z) into its even and odd parts, f (z) = f+ (z) + f− (z)
with f± (−z) = ±f± (z). Since f+ (z) and f− (z)/℘0 (z|Λ) are both even elliptic functions,
we use the result derived in (2.4.2) that every even elliptic function may be expressed as a
rational function of ℘(z|Λ).

23
2.4.2 The discriminant and the roots of the cubic
The roots of the cubic polynomial in (2.4.8) must all produce double zeros as a function of z,
since the left side of (2.4.8) is a perfect square. But we had seen earlier that double zeros
of ℘(z|Λ) − ℘(w|Λ) occur if and only if w 6∈ Λ and w ≡ −w (mod Λ), namely at the three
non-zero half-periods. Denoting the values of ℘ at the half periods by eα with α = 1, 2, 3,
1


eα (Λ) = ℘ ω |Λ
2 α
(2.4.10)

we obtain the following factorized form of the cubic polynomial,3

℘0 (z|Λ)2 = 4 ℘(z|Λ) − e1 ℘(z|Λ) − e2 ℘(z|Λ) − e3

(2.4.11)

where the three symmetric functions of ei may be identified with the coefficients g2 and g3
of the cubic (2.4.8) as follows,
g2 g3
e1 + e2 + e3 = 0 e1 e2 + e2 e3 + e3 e1 = − e1 e2 e3 = (2.4.12)
4 4
The discriminant ∆ of the cubic 4x3 − g2 x − g3 is defined as follows,

∆ = 16(e1 − e2 )2 (e2 − e3 )2 (e3 − e1 )2 (2.4.13)

In terms of the coefficients g2 and g3 , it is given by,

∆ = g23 − 27g32 (2.4.14)

The discriminant is non-zero if and only if all roots are distinct. When two of the roots or
all three roots coincide, the discriminant vanishes and the cubic is singular. To obtain ei in
terms of g2 and g3 , it suffices to find the roots of the cubic polynomial 4x3 − g2 x − g3 , which
are given in Lagrange’s presentation of the roots by,
1   √ √
eα = √ ρα δ + ρ2α g2 δ −1 δ3 = 27 g3 + −∆ (2.4.15)
2 3

for α = 1, 2, 3 and ρ = e2πi/3 a cube root of unity.

2.4.3 The Weierstrass ζ-function

Since ℘(z|Λ), viewed as a doubly periodic function on C, has only double poles, it is the
derivative in z of a single-valued meromorphic function ζ(z|Λ) on C, referred to as the

3
Henceforth, we shall suppress the Λ-dependence of g2 , g3 , and eα .

24
Weierstrass ζ-functions (not to be confused with the Riemann ζ-function). More precisely,
ζ(z|Λ) is defined uniquely by,

℘(z|Λ) = −ζ 0 (z|Λ) ζ(−z|Λ) = −ζ(z|Λ) (2.4.16)

and has simple poles in z with unit residue at every point in Λ. As a result, it has a single
pole in any fundamental parallelogram Pµ and therefore cannot be double periodic in the
lattice Λ. Since its derivative is doubly periodic the monodromies of ζ(z|Λ) are constant,

ζ(z + ω1 |Λ) = ζ(z|Λ) + η1 (Λ) η1 (Λ) = 2 ζ( 21 ω1 |Λ)

ζ(z + ω2 |Λ) = ζ(z|Λ) + η2 (Λ) η2 (Λ) = 2 ζ( 21 ω2 |Λ) (2.4.17)

The expressions for η1 , η2 have been obtained by setting z equal to − 21 ω1 , − 12 ω2 respectively

and using the odd parity of ζ(z|Λ). By integrating z℘(z|Λ) along ∂Pµ , we obtain the relation,

η1 ω2 − η2 ω1 = 2πi (2.4.18)

Since the monodromies of ζ(z|Λ) are independent of z, the difference ζ(z − a|Λ) − ζ(z − b|Λ)
is an elliptic function for the lattice Λ with simple poles at a and b with residues +1 and −1
respectively. One also defines the Weierstrass σ-function by ζ(z|Λ) = ∂z ln σ(z|Λ) normalized
so that σ(z|Λ) = z + O(z 2 ).

2.4.4 Rescaling the periods

From the definition in (2.4.1), it is manifest that the Weierstrass functions ζ, ℘, and ℘0 are
homogeneous of degree −1, −2, and −3 respectively under simultaneous rescaling of z, the
periods ω1 , ω2 , and the lattice Λ,

ζ(αz|αΛ) = α−1 ζ(z|Λ)

℘(αz|αΛ) = α−2 ℘(z|Λ)
℘0 (αz|αΛ) = α−3 ℘0 (z|Λ) (2.4.19)

As a result, the coefficients g2 , g3 , and Gm are homogeneous of degree 4, 6, and −m respec-

tively, while the roots ei have degree −2 and the discriminant ∆ has degree −12. These
rescaling properties show that the overall scale of the periods is an “easy” degree of freedom
which in many, but not all, applications in mathematics and physics is immaterial. The free-
dom to rescale allows us to choose a canonical value for one of the periods which is usually
taken to be ω1 = 1 and ω2 /ω1 = τ . In this normalization the lattice is given by,

Λ = Z ⊕ τZ (2.4.20)

25
 Im(z) τ τ +1
• •

−
B = = B

• •

−
C/Λ 0
A A 1 Re(z)

Figure 2: The torus C/Λ is represented in the plane by a parallelogram with complex coor-
dinates z, z̄ and opposite sides pairwise identified and is shown with a choice of canonical
homology cycles A and B corresponding to the cycles [0, 1] and [0, τ ] in C.

and the torus C/Λ may be represented in C by a parallelogram with vertices 0, 1, τ , and τ +1
and opposite sides identified pairwise, as illustrated in Figure 2. For this normalization of
the lattice the Weierstrass function will be denoted as follows by,
℘(z) = ℘(z|τ ) = ℘(z|Λ) (2.4.21)
The left most form ℘(z) is reserved for when the lattice Λ and the value of τ are fixed and
clear from the context. Out of the coefficients g2 and g3 , one may construct one combination
that is invariant under scaling and, by construction, depends only on the modulus τ ,
j = 1728 g23 /∆ (2.4.22)
The normalization factor is chosen in such a way that j(τ ) ≈ e−2πiτ as τ → i∞ with unit
coefficients. This so-called j-function will play a fundamental role in the theory of modular
forms, to be studied in subsequent sections.

2.5 Jacobi elliptic functions

For a given lattice Λ, the elliptic function ℘(z|Λ) − ei associated with a root ei has a unique
double pole at z = 0 and a unique double zero at z = ωi /2 (mod Λ). The p latter follows
from the fact that its derivative ℘0 (z|Λ) vanishes at the half periods. Hence ℘(z|Λ) − ei
is a holomorphic function which is doubly periodic, up to sign factors, in z with periods ω1
and ω2 . The Jacobi elliptic functions sn(u|k), cn(u|k), and dn(u|k) may be defined in terms
these square roots as follows,
r s
e1 − e3 ℘(z|Λ) − e2
sn(u|k) = dn(u|k) =
℘(z|Λ) − e3 ℘(z|Λ) − e3
s
℘(z|Λ) − e1 e2 − e3
cn(u|k) = k2 = (2.5.1)
℘(z|Λ) − e3 e1 − e3

26
The variables u and z, and the periods are related as follows,
√ √
u = e1 − e3 z Kα = ωα e1 − e3 (2.5.2)

When the value of k is clear from context, one uses the abbreviated notations sn(u) =
sn(u|k), cn(u) = cn(u|k), and dn(u) = dn(u|k). While the functions sn(u)2 , cn(u)2 , dn(u)2 ,
√
and sn(u)cn(u)dn(u) are doubly periodic under translations of u by the lattice Λ e1 − e3
generated by the periods K1 , K2 , the functions sn(u), cn(u), and dn(u) are doubly periodic
only up to signs in view of the square roots in their definition. Under u → −u the function
sn is odd while cn, dn are even. Their zeros and poles are simple and may be read off from
the definition. The modulus k 2 plays a role for Jacobi elliptic functions analogous to the
role played by the j-function for Weierstrass functions, and the two may be related to one
another by expressing the roots eα in the definition of k 2 in (2.5.1) using (2.4.15), giving,
(k 4 − k 2 + 1)3
j = 256 (2.5.3)
k 4 (k 2 − 1)2
The following identities follow from the definitions of the Jacobi elliptic functions,

sn(u)2 + cn(u)2 = 1 sn0 (u) = −cn(u) dn(u)

k 2 sn(u)2 + dn(u)2 = 1 cn0 (u) = −sn(u) dn(u)
dn(u)2 − k 2 cn(u)2 = 1 − k 2 dn0 (u) = −k 2 sn(u) cn(u) (2.5.4)

where the prime denotes the derivative with respect to u. The differential relations were
obtained with the help of the following relation,
3 cn(u) dn(u)
℘0 (z) = 2(e1 − e3 ) 2 (2.5.5)
sn(u)3
where the prime on the left side denotes the derivative with respect to z, and on the right
side with respect to u. Upon taking the square of each differential relation and using the
algebraic equations to express both sides in terms of the same function, we obtain the
following differential equations,

sn0 (u)2 = 1 − sn(u)2 1 − k 2 sn(u)2

cn0 (u)2 = 1 − cn(u)2 1 − k 2 + k 2 cn(u)2

dn0 (u)2 = 1 − dn(u)2 k 2 − 1 + dn(u)2

(2.5.6)

An immediate corollary for Jacobi elliptic functions following from Theorem 2.4 is as follows.
√
Corollary 2.5 An arbitrary elliptic function f (u) with a lattice of periods Λ e1 − e3 may
be expressed as a rational function of either sn(u)2 , cn(u)2 , or dn(u)2 which is linear in the
product sn(u) cn(u) dn(u).

27
2.6 Jacobi ϑ-functions
In the Weierstrass approach to elliptic functions, the basic building block is the meromorphic
℘-function and its derivative ℘0 , in terms of which every elliptic function is a rational function.
Instead, the Jacobi ϑ-function approach produces elliptic functions in terms of Jacobi ϑ-
functions, which are holomorphic, at the cost of being multiple-valued on C/Λ. Scaling the
lattice Λ so that ω1 = 1 and ω2 = τ with Im (τ ) > 0, the Jacobi ϑ-function is defined by,
2
X
ϑ(z|τ ) = eiπτ n +2πinz (2.6.1)
n∈Z

and is often denoted simply by ϑ(z) when the τ -dependence is clear. The series is absolutely
convergent for Im (τ ) > 0 and defines a holomorphic function in z ∈ C. The ϑ-function
satisfies a complexified version of the heat equation,
∂z2 ϑ(z|τ ) = 4πi∂τ ϑ(z|τ ) (2.6.2)
The function ϑ(z|τ ) is even in z → −z and transforms under shifts in the lattice Λ by,
ϑ(z + 1|τ ) = ϑ(z|τ )
ϑ(z + τ |τ ) = ϑ(z|τ ) e−iπτ −2πiz (2.6.3)
Thus, ϑ(z|τ ) is not an elliptic function in z. Indeed, it could never be, as a holomorphic
doubly periodic function must be constant. We shall later see how ϑ can be naturally viewed
as a holomorphic section of a line bundle on C/Λ, but for the time being we shall just work
with it as a multiple-valued function.

2.6.1 The number of zeros of ϑ

To find the number of zeros of ϑ(z|τ ) as a function of z, we integrate its logarithmic derivative
along the closed boundary of the fundamental parallelogram P0 (with vertices 0, 1, τ, 1 + τ ),
and decompose the integration as a sum of the line integrals along the four edges,
I Z 1 Z 1+τ Z τ Z 0
dz ∂z ln ϑ(z|τ ) = + + + dz ∂z ln ϑ(z|τ ) (2.6.4)
∂P0 0 1 1+τ τ

By periodicity of ϑ(z|τ ) under z → z + 1, the contributions from the second and fourth
integrals on the right side cancel one another. The contribution of the third integral is just
a translate and opposite of the first integral, so that we have,
I Z 1  
dz ∂z ln ϑ(z|τ ) = dz ∂z ln ϑ(z|τ ) − ∂z ln ϑ(z + τ |τ ) (2.6.5)
∂P0 0

Using the second relation in (2.6.3), this integral is readily evaluated, and we find 2πi which
implies that ϑ(z|τ ) has exactly one zero in P0 and, since it is holomorphic, no poles. To
determine the position of this zero, it will be useful to introduce characteristicss.

28
2.6.2 ϑ-functions with characteristics
A convenient variant of the ϑ-function introduced above is the ϑ-function with characteristics
α, β ∈ C, defined by,
2
X
ϑ [ αβ ] (z|τ ) = eiπτ (n+α) +2πi(n+α)(z+β) (2.6.6)
n∈Z

It is related to the original ϑ-function by translation of z and an exponential factor,

2 +2πiα(z+β)
ϑ [ αβ ] (z|τ ) = ϑ(z + ατ + β|τ ) eiπτ α (2.6.7)

The original ϑ(z|τ )-function corresponds to α = β = 0. Under translations of z by the

lattice Λ, and under a reflection z → −z, we have,

ϑ [ αβ ] (z + 1|τ ) = ϑ [ αβ ] (z|τ ) e2πiα

ϑ [ αβ ] (z + τ |τ ) = ϑ [ αβ ] (z|τ ) e−2πiβ−iπτ −2πiz
ϑ −α α
 
−β (−z|τ ) = ϑ [ β ] (z|τ ) (2.6.8)

Under integer shift in the characteristics, the transformations are as follows,

ϑ α+1 (z|τ ) = ϑ [ αβ ] (z|τ )

 
β
α
ϑ [ β+1 ] (z|τ ) = ϑ [ αβ ] (z|τ ) e2πiα (2.6.9)

For a ϑ-function with characteristics to have definite parity under z → −z, we must have
−α ≡ α (mod 1) and −β ≡ β (mod 1), which requires α, β ∈ {0, 21 } and corresponds to the
four half-periods. It is traditional to give these four ϑ-functions special names,
h1i h1i
ϑ1 (z|τ ) = − ϑ 21 (z|τ ) ϑ2 (z|τ ) = ϑ 02 (z|τ )
2
h i
0 0
ϑ3 (z|τ ) = ϑ [ 0 ] (z|τ ) ϑ4 (z|τ ) = ϑ 21 (z|τ ) (2.6.10)

The minus sign for ϑ1 is introduced for later convenience. From the above discussion, it
is manifest that ϑ1 is odd under z → −z while ϑ2 , ϑ3 , ϑ4 are even. One designates the
corresponding half-integer characteristics as odd and even characteristics respectively.
The parity of ϑ1 means that we know the location of its zero, namely at z = 0, and by
periodicity we have zeros at every z ∈ Λ. Similarly, the zeros of ϑ2 , ϑ3 , and ϑ4 are respectively
at 21 + Λ, 12 + τ2 + Λ, and τ2 + Λ. The function ϑ1 is particularly useful, and we record here
its monodromy properties in z for later use,

ϑ1 (z + 1|τ ) = −ϑ1 (z|τ )

ϑ1 (z + τ |τ ) = −ϑ1 (z|τ ) e−iπτ −2πiz (2.6.11)

29
2.6.3 The Riemann relations for Jacobi ϑ-functions

There are four basic Riemann relations on Jacobi ϑ-functions, given by,

X 4
Y 4
Y
hκ|λi ϑ[κ](ζi ) = 2 ϑ[λ](ζi0 ) (2.6.12)
κ i=1 i=1

where κ and λ are half-integer characteristics,

 0  0
κ λ
κ = 00 λ = 00 (2.6.13)
κ λ

namely κ0 , κ00 , λ0 , λ00 ∈ {0, 12 }. The sign factor hκ|λi is given by,

hκ|λi = exp{4πi(κ0 λ00 − κ00 λ0 )} (2.6.14)

and the relation between ζi and ζi0 is given by,

 0    
ζ1 ζ1 1 1 1 1
ζ20  ζ2  1 1 1 −1 −1
 0 = M   M=   (2.6.15)
ζ3  ζ3  2 1 −1 1 −1
0
ζ4 ζ4 1 −1 −1 1

The proof of these Riemann relations is left as an exercise. An important special case is
obtained by setting ζ1 = ζ2 = z and ζ3 = ζ4 = w so that ζ10 = z + w, ζ20 = z − w, and
ζ30 = ζ40 = 0 for i = 2, 3, 4. For this special case, the Riemann relation for λ0 = λ00 = 0
becomes an addition formula,
4
X
ϑi (z|τ )2 ϑi (w|τ )2 = 2ϑ3 (z + w|τ )ϑ3 (z − w|τ )ϑ3 (0|τ )2 (2.6.16)
i=1

and for z = w becomes a duplication formula,

ϑ2 (z|τ )4 + ϑ3 (z|τ )4 + ϑ4 (z|τ )4 = 2ϑ3 (2z|τ )ϑ3 (0|τ )3 (2.6.17)

Further setting z = 0, we obtain the famous Jacobi formula,

ϑ3 (0|τ )4 − ϑ2 (0|τ )4 − ϑ4 (0|τ )4 = 0 (2.6.18)

whose validity guarantees the cancellation of the one-loop contribution to the cosmological
constant in Type II and Heterotic string theories in flat Minkowski space-time.

30
2.6.4 Constructions of elliptic functions in terms of zeros and poles
There are various classic constructions of elliptic functions in terms of Weierstrass elliptic
functions, Jacobi elliptic functions, and Jacobi ϑ-functions, in addition to the construction
given in Theorem 2.4. The most important ones of those will be reviewed here. The relations
between these different constructions have direct physical applications, for example to Bose-
Fermi equivalence in two-dimensional quantum field theory.
• The first construction is convenient when the zeros and poles of an elliptic function f
are specified (subject to the conditions of Theorem 2.3) and proceeds by taking ratios of
products of ϑ1 -functions. An elliptic function f must have the same number of zeros and
poles in any fundamental parallelogram Pµ by the second condition in Theorem 2.3. We
then have the following theorem.
Theorem 2.6 A meromorphic function f (z|τ ) with N poles bi and N zeros ai in z modulo
Λ, with i = 1, · · · , N may be expressed in the following general form,
N N
Y ϑ1 (z − ai |τ ) X
f (z) = (ai − bi ) ∈ Λ (2.6.19)
i=1
ϑ1 (z − bi |τ ) i=1

for ai , bi ∈ C, not necessarily mutually distinct.

To prove the theorem, we observe that the poles and zeros of both sides clearly match,
the right side is manifestly periodic under z → z + 1, while periodicity in z → z + τ may be
verified to hold using (2.6.11) if and only if the condition on ai , bi of the theorem holds.
• The second construction is convenient when the poles of an elliptic function and their
residues are given (subject to the conditions of Theorem 2.3) and proceeds by using the
Weierstrass ζ-function and ℘-function, or equivalently by taking the logarithmic derivative
of ϑ-functions. The equivalence is made clear by the following relations, presented here for
the rescaled lattice Λ = Z ⊕ τ Z with periods ω1 = 1, ω2 = τ ,
℘(z|τ ) = −η1 (τ ) − ∂z2 ln ϑ1 (z|τ )
ζ(z|τ ) = η1 (τ )z + ∂z ln ϑ1 (z|τ )
ln σ(z|τ ) = ln ϑ1 (z|τ ) − ln ϑ1 (0|τ ) + 21 η1 (τ )z 2 (2.6.20)
The first relation follows from the fact that both sides have a double pole 1/z 2 and no other
poles and are doubly periodic. The constant η1 may be determined in terms of ϑ-functions
by using the fact that ϑ1 (z|τ ) is odd in z and that ℘(z|τ ) = z −2 + O(z 2 ), and we find,
1 ϑ000
1 (0|τ )
η1 (τ ) = − = 2ζ( 21 |τ ) (2.6.21)
3 ϑ01 (0|τ )
The second relation results from integrating the first and requiring ζ(z|τ ) to be odd in z.
Using (2.4.17), (2.4.18), and (2.6.3) one verifies that both sides have the same monodromy.
The last equation follows from the second by integration and normalizing σ(z) = z + O(z 3 ).

31
Theorem 2.7 An elliptic function f (z|τ ) with Ns poles bi,s of order s and generalized residue
ri,s for s = 1, · · · , S and i = 1, · · · Ns may be represented as a sum of Weierstrass ζ-functions
and its derivatives, by identifying poles and generalized residues,
S X
X Ns N1
X
(s−1)
f (z|τ ) = r0 + ri,s ζ (z − bi,s |τ ) ri,1 = 0 (2.6.22)
s=1 i=1 i=1

where r0 and ri,s may depend on τ but are independent of z, and ζ 0 (z|τ ) = −℘(z|τ ).
Finally, expressing ℘(z|τ ) as an infinite sum in (2.4.1), integrating this formula twice
in z, and matching the integration constants, gives the product formula for the ϑ1 -function
and, by translations by half periods, for the remaining ϑα functions,
∞
1
Y
ϑ1 (z|τ ) = 2q sin πz
8 (1 − q n e2πiz )(1 − q n e−2πiz )(1 − q n )
n=1
∞
1
Y
ϑ2 (z|τ ) = 2q 8 cos πz (1 + q n e2πiz )(1 + q n e−2πiz )(1 − q n )
n=1
∞
1 1
Y
ϑ3 (z|τ ) = (1 + q n− 2 e2πiz )(1 + q n− 2 e−2πiz )(1 − q n )
n=1
∞
1 1
Y
ϑ4 (z|τ ) = (1 − q n− 2 e2πiz )(1 − q n− 2 e−2πiz )(1 − q n ) (2.6.23)
n=1

with q = e2πiτ . We shall leave the derivation as an exercise.

2.7 Uniformization of cubics and quartics

Trigonometric functions may be viewed as parametrizing the points on the unit circle, spec-
ified by the quadratic equation x2 + y 2 = 1, by the functions x = sin t and y = sin0 t = cos t.
This process is referred to as uniformization.
Elliptic functions analogously parametrize all solutions to cubic and quartic equations.
The Weierstrass function ℘(z|Λ) uniformizes the following cubic equation,

y 2 = 4x3 − g2 (Λ)x − g3 (Λ) = 4(x − e1 )(x − e2 )(x − e3 ) (2.7.1)

by x = ℘(z|Λ) and y = ℘0 (z|Λ) in view of (2.4.8), where g2 (Λ) and g3 (Λ) are associated
with the lattice Λ via (2.4.9) and (2.4.5). Analogously, the Jacobi elliptic function sn(u|k)
uniformizes the following quartic equation,

y 2 = (1 − x2 )(1 − k 2 x2 ) (2.7.2)

32
by x = sn(u|k) and y = sn0 (u|k) in view of the first equation in (2.5.6). The Jacobi elliptic
functions cn and dn similarly uniformize different, but equivalent quartics. More generally,
elliptic functions may be used to uniformize an arbitrary quartic of the form,

w2 = (v − v1 )(v − v2 )(v − v3 )(v − v4 ) = v 4 + c3 v 3 + c2 v 2 + c1 v + c0 (2.7.3)

for arbitrary points v1 , · · · , v4 ∈ C or, equivalently, for arbitrary coefficients c0 , · · · , c3 ∈ C.

To show this, we make use of the fact that the group SL(2, C) acts conformally and bijectively
on the compactified complex plane C ∪ {∞} by Möbius transformations. If the action of an
arbitrary SL(2, C) transformation on the variables v and vi is given by,
 
ax + b axi + b a b
v= vi = ∈ SL(2, C) (2.7.4)
cx + d cxi + d c d

then w and the differences v − vi obey simple transformation rules,

w0 y x − xi
w= v − vi = (2.7.5)
(cx + d)2 (cx + d)(cxi + d)

which map the general quartic (2.7.3) in (v, w) into a quartic in (x, y) given by,

w12 y 2 = (x − x1 )(x − x2 )(x − x3 )(x − x4 )

w12 = w02 (cx1 + d)(cx2 + d)(cx3 + d)(cx4 + d) (2.7.6)

Here w0 is an arbitrary constant in terms of which w1 is given by the second formula above.
The choice (x1 , x2 , x3 , x4 ) = (1, −1, k, −k) with w1 = 1 reproduces the quartic (2.7.2). Taking
the limit x4 → ∞ for (x1 , x2 , x3 ) = (e1 , e2 , e3 ) and 4cw02 3i=1 (cei + d) = −1 reproduces the
Q
cubic (2.7.1). This completes the proof that an arbitrary quartic may be uniformized either
by the Weierstrass or the Jacobi elliptic functions.

2.8 Elliptic curves

Since an arbitrary lattice Λ is an Abelian subgroup of C the lattice Λ is a normal subgroup
of C and the torus C/Λ is an Abelian group. Since Λ acts on C without fixed points, the
torus C/Λ is also a compact Riemann surface. Representing the torus in C by a fundamental
parallelogram Pµ , the group law may be realized by addition in C modulo the lattice Λ. The
group structure requires an identity element which is just 0 ∈ C modulo Λ, and an additive
inverse of z which is just −z modulo Λ.
The procedure of uniformization, discussed in the preceding subsection, shows that elliptic
functions naturally parametrize arbitrary cubics and quartics. In particular, the pair of
coordinates (x, y) = (℘(z|Λ), ℘0 (z|Λ)) parametrizes the cubic curve y 2 = 4x3 −g2 (Λ)x−g3 (Λ)

33
 B • e3 B • v3
e2 • v2 •
P∞ •v4
e1• v1•
A A P+∞
P−∞

Figure 3: The elliptic curve E is represented in terms of a double cover of the Riemann
sphere: with one branch point P∞ at ∞ in the left panel and with all four branch points
finite in the right panel and two distinct points P±∞ added at ∞ to compactify the elliptic
curve. A choice of canonical homology cycles A, B is indicated.

in terms of the points in the torus z ∈ C/Λ. The group structure of C/Λ maps to an Abelian
group structure on the corresponding elliptic curve, as we shall now explain in detail.
An elliptic curve E for given g2 , g3 ∈ C is defined as follows,
n o
E = (x, y) ∈ C such that y 2 = 4x3 − g2 x − g3 ∪ {P∞ } (2.8.1)

The point at infinity P∞ = (∞, ∞) is added to obtain a compact space, just as the torus C/Λ
is compact. While y 2 is determined in terms of x by the equation, the sign of y is not, which
is why the second entry y is required. For every generic value of x where y 6= 0, ∞, there are
two points in E corresponding to ±y. For the points x = e1 , e2 , e3 , e4 = ∞, however, we have
y = 0, ∞ and therep is only a single point in E. These points correspond to branch points of
the function y = 4x3 − g2 x − g3 where the two copies of C intersect (see the left panel of
Figure 3). Alternatively, an elliptic curve may be defined in terms of a quartic,
n 4
Y o
2
E = (v, w) ∈ C such that w = (v − vi ) ∪ {P+∞ , P−∞ } (2.8.2)
i=1

The points v = v1 , v2 , v3 , v4 correspond to the branch points of the function w where the two
copies of C intersect (see the right panel of Figure 3). In this representation two points at
infinity P±∞ are required, one on each sheet.

2.8.1 Embedding into CP2

Two complex-dimensional projective space CP2 is a compact space defined by an equivalence
relation ≈ under rescaling coordinates in (C3 )∗ = C3 \ {(0, 0, 0)},

CP2 = (x, y, z) ∈ (C3 )∗ such that (λx, λy, λz) ≈ (x, y, z) for λ ∈ C∗

(2.8.3)

34
Actually, CP2 is a Kähler space isomorphic to SU (3)/S(U (2) × U (1)). Every compact
(orientable) Riemann surface may be embedded in CP2 . In particular, the elliptic curve E
may be embedded in CP2 by rendering its defining equation homogeneous in three variables,
namely the original x, y as well as the additional variable z, to obtain,
n o
E = (x, y, z) ∈ CP2 such that y 2 z = 4x3 − g2 xz 2 − g3 z 3 (2.8.4)

Since E is hereby defined as a closed subset of the compact space CP2 , it is compact without
the need to add points at infinity. The original presentation of the elliptic curve E is recovered
in the coordinate patch z = 1, while in the other two patches it is given by,

x=1 y 2 z = 4 − g2 z 2 − g3 z 3
y=1 z = 4x3 − g2 xz 2 − g3 z 3 (2.8.5)

The last equation shows that the point (0, 1, 0) ∈ CP2 belongs to every elliptic curve E. In
fact the point (0, 1, 0) corresponds to the point (x, y) = P∞ in (2.8.1). This may be seen by
considering the large x, y behavior in the z = 1 patch and allowing for a suitable rescaling
(x, y, 1) ≈ (xy −1 , 1, y −1 ) → (0, 1, 0) as x, y → ∞ subject to y 2 = 4x3 − g2 x − g3 .

2.9 Addition formulas and group structure of elliptic curves

The additive group structure of the torus C/Λ allows one to associate with two arbitrary
points P, Q ∈ C/Λ a third point R ∈ C/Λ by requiring the sum of their complex coordinates
zP , zQ , zR to vanish zP +zQ +zR = 0 (mod Λ). The addition of the points P and Q gives −R.
To obtain the corresponding operation on the elliptic curve in the Weierstrass presentation,
we begin by obtaining the addition formula for the Weierstrass function, which generalizes
the addition theorem of (2.2.9) for trigonometric functions,
2
℘0 (zP ) − ℘0 (zQ )

1
℘(zP + zQ ) + ℘(zP ) + ℘(zQ ) − =0 (2.9.1)
4 ℘(zP ) − ℘(zQ )

To prove it, one begins by showing that the left side has no poles, so that it must be
constant by Liouville’s theorem. One then shows that the constant vanishes by evaluating
the expression at special points. Specifically, for fixed generic zQ , the function ℘(zP + zQ ) is
an elliptic function in zP with one double pole at zP = −zQ (mod Λ). Thus, it is a rational
function of ℘(zP ) and ℘0 (zP ). Now ℘(zP ) − ℘(zQ ) has a simple zero at zP = −zQ , so it
should occur to the power −2, but it also has a simple zero at zP = zQ , which is cancelled
by multiplying by the square of ℘0 (zP ) − ℘0 (zQ ). The combination has a double pole in zP at
zP = 0 which is cancelled by the addition of ℘(zP ) and, by symmetry under the interchange
of zP and zQ , also ℘(zQ ). Thus the left side has no poles and is therefore constant. To

35
 y
•
R
•
Q
x
•
P

Figure 4: The real section of the elliptic curve y 2 = 4x3 − 4x + 4 in black, the straight line
9y = 10x + 14.72 in red, and their three intersection points P, Q, R.

evaluate the constant, we set zP = ω1 /2 and zQ = ω2 /2, use the fact that ℘0 vanishes at the
non-zero half-periods, and then use the first formula in (2.4.12).
On the elliptic curve E we represent the points P, Q, R by the coordinates (xP , yP ),
(xQ , yQ ) and (xR , yR ) respectively and, for simplicity, we assume that the points P and Q
are distinct. A straight line y = αx + β through the two distinct points P, Q on E intersects
the cubic E at exactly one further point R. Therefore, its coordinates (xR , yR ) satisfy,
yP − yQ xP y Q − xQ y P
yR = xR + yR2 = 4x3R − g2 xR − g3 (2.9.2)
xP − xQ xP − xQ
This system of equations may be solved for xR as follows,
 2
1 yP − yQ
xR + xP + xQ − =0 (2.9.3)
4 xP − xQ
with yR given by the second equation of (2.9.2). Parametrizing the coordinates by x = ℘(z)
and y = ℘0 (z) evaluated at z = zP , zQ , zR , we recover the addition law (2.9.1) for the
Weierstrass function. The point P∞ plays the role of the unit element of the additive group E.
This may be seen by letting xQ , yQ → ∞ which implies (xR , yR ) = (xP , yP ).

2.10 Meromorphic functions and Abelian differentials

Much of the work on elliptic functions in the 19-th century was motivated by the study of
elliptic integrals, namely definite or indefinite integrals involving the square root of a cubic or
quartic polynomial. In this subsection we shall consider meromorphic functions and Abelian
differentials before addressing their integrals in the subsequent subsection. For a discussion
of differentials and their properties on arbitrary Riemann surfaces, see appendix B.

36
 The space of meromorphic functions on the Riemann sphere Ĉ = C ∪ {∞} is the field
C(x) of rational functions of a holomorphic coordinate x ∈ C. Meromorphic functions on an
elliptic curve E correspond to elliptic functions on the torus C/Λ. Since elliptic functions on
C/Λ are rational functions of ℘(z) = ℘(z|Λ) and ℘0 (z) = ℘0 (z|Λ), the space of meromorphic
functions on an elliptic curve E is the field C(x, y) of rational functions of x and y. The field
C(x, y) may be viewed as a quadratic extension of C(x) by y, since y 2 ∈ C(x) on an elliptic
curve E. The Galois group of this extension is Z2 , the group of involutions of E that maps
(x, y) → (x, −y) and swaps the two sheets in the elliptic representation of Figure 3. On the
torus C/Λ, the Z2 involution simply corresponds to z → −z (mod Λ).

2.10.1 Perspective of the torus C/Λ

To investigate meromorphic differentials we first consider the perspective of the torus. On
a torus C/Λ with local complex coordinate z, there exists a translation invariant (1, 0)-form
dz which is unique up to a multiplicative constant. The differential dz is holomorphic and
nowhere vanishing. As a result, the vector spaces of meromorphic (n, 0)-forms for n ∈ Z
are all isomorphic to the space of meromorphic (0, 0)-forms, i.e. elliptic functions. Thus,
the most general meromorphic differential (n, 0)-form on C/Λ is given by f (z)(dz)n where
f (z) is an elliptic function on C/Λ. Using the general construction of elliptic functions in
terms of the Weierstrass functions, Theorems 2.4 and 2.7 allow us to express an arbitrary
meromorphic (n, 0)-form $ = f (z)(dz)n as a rational function in ℘ and ℘0 which may be
chosen to be linear in ℘0 or, equivalently, as a sum of derivatives of the Weierstrass ζ-function,

$ = A ℘(z) (dz)n + B ℘(z) ℘0 (z)(dz)n

(2.10.1)
S X
X Ns N1
X
(s−1)
$ = r0 dz + ri,s ζ (z − bi,s ) dz ri,1 = 0 (2.10.2)
s=1 i=1 i=1

where A and B are rational functions of their argument, while the poles bi,s and the coeffi-
cients r0 , ri,s are arbitrary aside from the vanishing of the sum of ri,1 . We shall denote the
vector space of meromorphic (n, 0)-forms Ωn = Ωn (C/Λ).
The space dΩ0 of (1, 0)-forms obtained by taking the differential of elliptic functions is a
subspace dΩ0 ⊂ Ω1 . This may be seen in the representation (2.10.1) by taking the differential
of a general elliptic function f ∈ Ω0 and then using the differential equation (2.4.8) and its
derivative to recast the result again in the form (2.10.1). In the representation of (2.10.2)
the differential kills the first term and simply increases the order of the derivatives in the
terms under the double sum, thereby producing a (1, 0)-form of the type (2.10.2).
However, not every meromorphic (1, 0)-form is the differential of an elliptic function.
Constructing the space Ω1 /dΩ0 of meromorphic (1, 0) modulo differentials of elliptic functions
is a cohomology problem whose solution is fundamental in the theory of Riemann surfaces

37
and in string theory. To obtain the generators of Ω1 /dΩ0 we express an arbitrary f ∈ Ω0
and $ ∈ Ω1 in terms of the representation (2.10.2), and then take the differential of f ,

S̃ X
X Ñs Ñ1
X
(s)
df = r̃i,s ζ (z − b̃i,s ) dz r̃i,1 = 0
s=1 i=1 i=1
S X
X Ns N1
X
$ = r0 dz + ri,s ζ (s−1) (z − bi,s ) dz ri,1 = 0 (2.10.3)
s=1 i=1 i=1

All terms in $ involving ζ (s−1) (z − bi,s ) for s ≥ 3 may be matched by terms in df provided
we choose the poles b̃i,s−1 = bi,s and coefficients r̃i,s−1 = ri,s . For s = 2, this would also be
the case if it were not for the constraint on the sum of the r̃i,1 , which means that a single
ζ (1) (z − b|τ ) is, manifestly, not the differential of an elliptic function. The term r0 dz and the
terms proportional to ζ(z − bi,1 ) are not differentials of elliptic functions but, in view of the
vanishing sum of residues, always occur as differences ζ(z − bi,1 ) − ζ(z − bj,1 ). In summary,
Ω1 /dΩ0 is generated by three type of (1, 0)-forms, referred to as Abelian differentials.

1. Abelian differential of the first kind : holomorphic and given by a constant multiple of
the coordinate differential dz;
2. Abelian differentials of the second kind : have one double pole at an arbitrary point
zi ∈ C/Λ and no other poles and are given by a constant multiple of

℘(z − zi )dz = −ζ (1) (z − zi )dz (2.10.4)

3. Abelian differentials of the third kind : have two simple poles with residues ±1 at
arbitrary points zi 6= zj ∈ C/Λ and no other poles and are given in terms of the
Weierstrass ζ-function by,


ζ(z − zi ) − ζ(z − zj ) + ζ(zi − zj ) dz (2.10.5)

We note that the third term inside the parentheses is included to make the differential
invariant under Λ not only in z but also in the points zi and zj .

The characterization of Abelian differentials of the second and third kind in terms of their
poles does not specify them uniquely as one may add an arbitrary multiple of dz to either.

2.10.2 Perspective of the elliptic curve E

For an elliptic curve E with local coordinates (x, y) given in terms of g2 , g3 ∈ C by,

y 2 = 4x3 − g2 x − g3 (2.10.6)

38
the (1, 0) form dz on C/Λ may be expressed in terms of (x, y) = (℘(z|Λ), ℘0 (z|Λ)) by using
the relation (2.4.8) and we find,

dx dx
dz = =p (2.10.7)
y 4x3 − g2 x − g3

The differential dx/y is single-valued and holomorphic on E since the square root at one of
the branch points ei is well-defined in a suitable local coordinate ξ with ξ 2 = x − ei , in which
the differential is given by dx/y = c−1 2
i dξ + O(ξ) where ci = (ei − ej )(ei − ek ) for j, k 6= i.
The coordinate differential dx vanishes at the branch points ei and has a triple pole at ∞.
The differential dz may also be obtained for an arbitrary quartic of the form (2.7.3),

dv
w2 = (v − v1 )(v − v2 )(v − v3 )(v − v4 ) (2.10.8)
w
The differential dx/y is mapped to dv/w by the SL(2, C) Möbius transformation of (2.7.4)
which maps (x, y) to (v, w), xi to vi , and the differential dx to dv = dx/(cx + d)2 .
An arbitrary meromorphic differential ω on the torus C/Λ, expressed in terms of ℘ and
℘0 in (2.10.1), translates to the following expression on the corresponding elliptic curve E,

dx
ω = A(x) + B(x)dx (2.10.9)
y
where A and B are rational functions of x. Under the Z2 involution of E which maps
(x, y) → (x, −y), the first term in ω is odd while the second term is even. We leave it as an
exercise to show that the Abelian differentials map to the elliptic curve E as follows,

1. Abelian differential of the first kind : a constant multiple of dx/y;

2. Abelian differentials of the second kind : have one double pole at an arbitrary point
(xi , yi ) ∈ E and no other poles and, using the addition formula of (2.9.1), are given by,

1 (y + yi )2 dx
 
−x − xi + (2.10.10)
4 (x − xi )2 y

3. Abelian differentials of the third kind : have two simple poles with opposite residues ±1
at arbitrary points (xi , yi ) 6= (xj , yj ) and no other poles and are given by,
 
1 y + yi yj + y yi + yj dx
+ + (2.10.11)
2 x − xi xj − x xi − xj y

The last term inside the parentheses arises from the ζ(zi − zj )dz contribution in the
torus representation, and is proportional to the Abelian differential of the first kind.

39
2.11 Abelian and elliptic integrals
The decomposition of the space of meromorphic (1, 0)-forms Ω1 into a sum of the space of
exact differentials of elliptic functions dΩ0 plus the three kinds of basic Abelian differentials
greatly simplifies the problem of integration of (1, 0)-forms. Integration of exact differentials
$ ∈ dΩ0 is straightforward, and we shall now concentrate on the integration of the three
kinds of Abelian differentials. A canonical basis of first homology generators of the elliptic
curve E will be denoted by A, B and was shown already in Figures 2 and 3. Parametrizing
the lattice Λ = ω1 Z ⊕ ω2 Z by the periods ω1 and ω2 satisfying τ = ω2 /ω1 with τ 6∈ R, the
homology basis may be represented on the torus C/Λ by the following simple cycles,

A : z → z + ω1
B : z → z + ω2 (2.11.1)

We shall study integrals of Abelian differentials along open and closed paths in E and C/Λ.
We consider open paths between arbitrary points z1 , z2 ∈ C/Λ corresponding to the points
(x1 , y1 ), (x2 , y2 ) ∈ E, related by the Weierstrass function xi = ℘(zi ), yi = ℘0 (zi ) for i = 1, 2.

2.11.1 Abelian integrals of the first kind

The integral of the Abelian differential dz = dx/y of the first kind is given as follows,
Z (x2 ,y2 ) Z (x2 ,y2 ) Z z2
dx dx
p = = dz = z2 − z1 (2.11.2)
(x1 ,y1 ) 4x3 − g2 x − g3 (x1 ,y1 ) y z1

Viewed as taking values in C, the integral depends on the path of integration. Indeed, while
holomorphicity of dz = dx/y guarantees independence under small deformations of the path
of integration, the integral will depend on how many times the path circles the cycles A and
B. The integrals of dz = dx/y around A, B give the periods of the lattice Λ,
I I I I
dx dx
= dz = ω1 = dz = ω2 (2.11.3)
A y A B y B

Thus, while the value of the integral of (2.11.2) depends on the path of integration when its
range is C, the integral becomes single-valued as a map to C/Λ.

2.11.2 Abelian integrals of the second kind

The integral of the Abelian differential of the second kind,

1 (y + yi )2 dx
 
$ = ℘(z − zi )dz = −x − xi + (2.11.4)
4 (x − xi )2 y

40
is, by definition, given by the Weierstrass ζ-function in view of ζ 0 (z) = −℘(z),
Z z2
$ = ζ(z1 − zi ) − ζ(z2 − zi ) (2.11.5)
z1

The integral depends on the path of integration and the number of times the path winds
around the cycles A and B, which may be evaluated using (2.4.17),
I I
$ = −η1 $ = −η2 (2.11.6)
A B

These periods do not belong to the lattice Λ, as is clear from the relation (2.4.18). In fact,
it is not possible to form a linear combination of $ and the Abelian differential of the first
kind dz to obtain a single-valued integral taking values in C/Λ. But it is possible to form
combinations with either vanishing A period or vanishing B period. For example $+η1 dz/ω1
has vanishing A period while its B period is 2π/ω1 .

2.11.3 Abelian integrals of the third kind

The integral of the Abelian differential of the third kind,
 
$ = ζ(z − zi ) + ζ(zi − zj ) + ζ(zj − z) dz
 
1 y + yi yi + yj yj + y dx
= + + (2.11.7)
2 x − xi xi − xj x j − x y

is given by the logarithm of the Weierstrass σ-function in view of σ(z)0 /σ(z) = ζ(z),
Z z2
σ(z1 − zi )σ(z2 − zj )
$ = ln (2.11.8)
z1 σ(z2 − zi )σ(z1 − zj )

Its period integrals are given by,

I I
$ = ω1 η1 (zi − zj ) $ = ω2 η1 (zi − zj ) (2.11.9)
A B

For the canonical periods ω1 = 1 and ω2 = τ , we may convert this expression in terms of
Jacobi ϑ-functions, using (2.6.20), and we obtain,
Z z2
ϑ1 (z1 − zi )ϑ1 (z2 − zj )
$ = ln + η1 (z2 − z1 )(zi − zj ) (2.11.10)
z1 ϑ1 (z2 − zi )ϑ1 (z1 − zj )

We conclude that all elliptic integrals can be evaluated in terms of Weierstrass and Jacobi
ϑ-functions.

41
• Bibliographical notes
The early history of the development of elliptic functions and modular forms, with empha-
sis on the contributions of Eisenstein and Kronecker, is discussed in the delightful book by
Weil [2]. Classic treatises on Weierstrass elliptic functions, Jacobi elliptic functions, and
Jacobi ϑ-functions may be found in Whittaker and Watson [3] as well as in the Bateman
manuscript [1], where many practical definitions and useful formulas are collected. Mum-
ford [4] emphasizes Jacobi ϑ-functions, while Lang [5] deals with elliptic functions as well as
modular forms. An account of applications of elliptic functions to physics and technology
may be found in a classic book by Oberhettinger and Magnus [6].

42
3 Modular forms for SL(2, Z)
In the preceding section, we studied the dependence of elliptic functions on the points in
the torus C/Λ. In this section, we shall investigate the dependence on the lattice Λ. We
begin by discussing the equivalence relations which lead to the transformation laws of elliptic
functions, Eisenstein series, and ϑ-functions under the modular group. We shall also discuss
the Poincaré upper half-plane and the fundamental domain of the modular group.

3.1 Automorphisms of lattices and the modular group

A lattice Λ in the complex plane C may be generated by a pair of periods ω1 , ω2 which
are linearly independent over R, namely such that ω2 /ω1 6∈ R. We shall be interested in
orientable surfaces C/Λ, and choose a definite orientation so that the pair of periods (ω1 , ω2 )
is ordered by choosing Im (ω2 /ω1 ) > 0.
The lattice Λ, and thus the torus C/Λ, may be generated equivalently by other choices
of periods, such as for example by ω1 and ω2 + nω1 for any n ∈ Z. More generally, Λ is
equivalently generated by any pair of periods ω10 , ω20 given by a linear combination of ω1 , ω2
with integer coefficients,
 0    
ω2 ω2 a b
=γ γ= a, b, c, d ∈ Z (3.1.1)
ω10 ω1 c d

provided γ has an inverse with integer entries. This condition is equivalent to det γ = ±1, or
more formally γ ∈ GL(2, Z). On the ratio τ of periods, which is referred to as the modulus
of the lattice Λ or the torus C/Λ, a transformation γ ∈ GL(2, Z) acts as follows,

ω2 ω20 aτ + b
τ= τ0 = γτ = τ 0 = (3.1.2)
ω1 ω10 cτ + d

A transformation γ ∈ GL(2, Z) will preserve the orientation Im (ω2 /ω1 ) = Im (τ ) > 0 of the
lattice provided the new periods enjoy the same orientation Im (ω20 /ω10 ) = Im (τ 0 ) > 0. Using
the above transformation rule, the imaginary parts are found to transform as follows,

det γ
Im (τ 0 ) = Im (τ ) (3.1.3)
|cτ + d|2

The orientation of the lattice Λ will be preserved provided γ is restricted to the SL(2, Z)
subgroup of GL(2, Z) for which det γ = 1, promoting SL(2, Z) to the group of orientation-
preserving automorphisms of Λ. The transformation −I ∈ SL(2, Z) maps the periods into
their opposites, but leaves their ratio τ invariant. Thus, the faithful action on the lattice Λ
is by SL(2, Z), but on τ it is by the group P SL(2, Z) = SL(2, Z)/Z2 where Z2 = {±I}.

43
3.1.1 Structure of the modular group
The group SL(2, Z) is an infinite discrete non-Abelian group, referred to as the modular
group. It is generated by two of its elements,
   
0 −1 1 1
S= T = (3.1.4)
1 0 0 1
These generators satisfy three fundamental relations,
S 2 = (ST )3 = (T S)3 = −I (3.1.5)
Any γ ∈ SL(2, Z) can be written as a finite word in the letters S, T . By using the first of the
above relations, words with two consecutive S letters may be simplified and omitted when
listing independent elements. Thus, any γ has the following product decomposition,
γ = ±T α1 ST α2 S · · · ST αn (3.1.6)
for αi ∈ Z and some positive integer n. Furthermore, any word with three consecutive
combinations of ST or T S may also be omitted by the last two relations.
To prove the validity of this decomposition, we consider an arbitrary γ ∈ SL(2, Z),
 
a b
γ= a, b, c, d ∈ Z, ad − bc = 1 (3.1.7)
c d
If c = 0, then ad = 1, and γ = aT ab . If c < 0 we multiply γ by −I to reduce this case to the
case c > 0. For c ≥ 1 we proceed by induction on c. If c = 1, then ad − b = 1 and we have
γ = T a ST d . Thus, the proposition holds true for c = 0, 1. If c ≥ 2 we shall assume that the
proposition holds true for all matrices whose lower left entry is positive and less than c, and
the proposition remains to be proven for all matrices whose left lower entry equals c. Since
c ≥ 2, c cannot divide a, leaving the following two possibilities:
• If |a| > c, then there exists a non-zero integer n such that a = nc + r with 0 < r < c,
and the matrix γ 0 = ST −n γ has lower left entry equal to r < c.
• If |a| < c, then the matrix γ 0 = sign(a)Sγ has lower left entry |a| < c.
In either case, by applying a product of S and T generators we have constructed from γ a
new matrix with strictly smaller value of the lower left entry. The validity of the proposition
follows by induction on the value of c.

3.2 The fundamental domain for SL(2, Z)

In this subsection, we begin with a brief review of the geometry of the Poincaré upper half-
plane, and then discuss the construction of the fundamental domain for SL(2, Z) as well as
its fixed points and cusps.

44
 p H
• H D
α
β

a b c R R

Figure 5: In the left figure the geodesics of the Poincaré upper half-plane H are shown as the
semi-circles centered on R with arbitrary radius; the geometry is non-Euclidean as several
geodesics through a given point p, such as b and c, are parallel to the geodesic a. In the right
figure, the area enclosed by the semi-infinite triangle D is area(D) = π − α − β.

3.2.1 The Poincaré upper half-plane

We denote the Poincaré upper half-plane by,
H = {τ ∈ C, Im (τ ) > 0} (3.2.1)
Under the action of SL(2, R) by Möbius transformations on τ , the upper half-plane H is
mapped to itself and its boundary ∂H = R ∪ {∞} is mapped to itself. The isotropy group
of an arbitrary point in H is isomorphic to SO(2) and hence H is the symmetric space,
H = SL(2, R)/SO(2) (3.2.2)
For example, the elements which leave τ = i invariant obey i = (ai + b)/(ci + d) and
ad − bc = 1 with a, b, c, d ∈ R and may be parametrized by b = −c, d = a, and a2 + b2 = 1 to
form a group isomorphic to SO(2). The Poincaré metric on H, defined by the line element,
|dτ |2
ds2 = (3.2.3)
(Im τ )2
is an SL(2, R)-invariant Riemannian metric of constant negative curvature. The Poincaré
upper half-plane equipped with the Poincaré metric is a model of non-Euclidean geometry.
Its geodesics (or straight lines) are the half-circles of arbitrary radius centered on R. Given
a geodesic a and a point p not on this geodesic, there are an infinite number of geodesics
through p that do not intersect a and are thus parallel to a, as illustrated in Figure 5.
The area of a polygon whose sides are geodesics may be obtained by decomposing the
area into sums and differences of semi-infinite triangles with one vertex at the cusp and two
corners with opening angles α and β, as shown in Figure 5,
d2 τ
Z
i
area(D) = 2
=π−α−β d2 τ = dτ ∧ dτ̄ (3.2.4)
D (Im τ ) 2
The area of a finite triangle with opening angles α, β, γ is then π − α − β − γ, and is clearly
always less than π, which is a well-known characteristic of hyperbolic geometry.

45
3.2.2 The fundamental domain for SL(2, Z)
Since the modular group SL(2, Z) is a subgroup of SL(2, R), it also maps H to H and ∂H
to ∂H. Every oriented lattice Λ, or equivalently oriented torus C/Λ, may be specified by a
point τ ∈ H up to overall rescaling of the lattice. Since the lattices corresponding to two
points τ and τ 0 are equivalent to one another if τ 0 = γτ with γ ∈ SL(2, Z), the space of
inequivalent lattices or inequivalent tori is given by the quotient,

SL(2, Z)\H = SL(2, Z)\SL(2, R)/SO(2) (3.2.5)

Just as we represented the torus C/Λ by a fundamental parallelogram in C, we can represent

the coset SL(2, Z)\H by a fundamental domain F in H, which is a connected subset of H.
The fundamental domain is not unique since, for example, if F is a fundamental domain
then so if γF . The standard choice is given by the following theorem.
Theorem 3.1 The following F is a fundamental domain for SL(2, Z),

F = τ ∈ H, |τ | ≥ 1, |Re (τ )| ≤ 21

(3.2.6)

where it is understood that the boundary components at Re (τ ) = ± 12 are identified under

τ → τ + 1 while the boundary components at |τ | = 1 for Re (τ ) positive and negative are
identified with one another under τ → −1/τ (see Figure 6). The area of F in the Poincaré
metric is given by area(F ) = π3 in view of (3.2.4).

To prove the theorem we show that for an arbitrary point τ ∈ H there exists a transfor-
mation γ ∈ SL(2, Z) such that γτ ∈ F . To do so, we fix τ and make use of the relation,
Im (τ )
Im (γτ ) = (3.2.7)
|cτ + d|2
to construct a γ ∈ SL(2, Z) with the largest possible value of Im (γτ ). This is done by
minimizing |cτ + d| for fixed τ as a, b, c, d run over the integers satisfying ad − bc = 1.
Taking γ to be the identity gives the bound |cτ + d| ≤ 1 which can be satisfied by only a
finite number of integer pairs (c, d). We construct γ as a product of translations by T and
inversions by S using an iterative process that terminates. Starting from an arbitrary τ ∈ H
which is not in F , apply a translation T α for α ∈ Z to map τ → τ 0 = T α τ into the vertical
strip τ 0 ∈ {|Re (τ )| ≤ 1/2}. Under this transformation, the value of the imaginary part is
unchanged Im (τ 0 ) = Im (τ ). If τ 0 ∈ F , we are done. If τ 0 6∈ F then it must satisfy |τ 0 | < 1.
Its image under S given by τ 00 = Sτ 0 has |τ 00 | > 1 as well as Im (τ 00 ) > Im (τ 0 ) = Im (τ ) in
view of the fact that c = 1 and d = 0 for S and that |τ 0 | < 1. If this image is in F we are
done, and if not, we repeat the process. Since only a finite number of pairs (c, d) satisfy the
bound |cτ + d| ≤ 1, and the iterative process strictly increases the value of Im (γτ ) at each
inversion S, the process must terminate, which completes the proof.

46
 i∞

F T (F )

ρ i ρ0
•
• •
S(F ) T S(F )

H
−1 − 12 0 1
2 1

Figure 6: The standard fundamental domain F for SL(2, Z) in H; its images T (F ), S(F ),
T S(F ); the orbifold points i, ρ, ρ0 in F ; and the cusp at i∞.

3.2.3 Elliptic points

Every point τ ∈ H is invariant under the Z2 = {±I} subgroup of SL(2, Z). An elliptic point
is a point in H that is invariant under a subgroup of SL(2, Z) that is larger than Z2 . We
shall now show that the only elliptic points in F are i, ρ, and ρ0 = ρ + 1, where ρ3 = 1 with
Im (ρ) > 0. Specifically, the point i is invariant under the subgroup {±I, ±S} of order four.
The point ρ is invariant under the subgroup {±I, ±(ST ), ±(ST )2 } of order six, while the
point ρ0 is invariant under the subgroup {±I, ±(T S), ±(T S)2 }, also of order six.
To prove the assertion we consider a point τγ ∈ H that is invariant under γ ∈ SL(2, Z),
 
a b
γτγ = τγ γ= (3.2.8)
c d

and therefore satisfies the quadratic equation cτγ2 − (a − d)τγ − b = 0 subject to Im (τγ ) > 0.
For c = 0 there are no solutions that satisfy both conditions, while for c 6= 0 we may assume
c > 0 without loss of generality. The point τγ lies in H provided |a + d| < 2 in which case τγ
is given by the solution with positive imaginary part,
p
a−d 4 − (a + d)2
τγ = +i (3.2.9)
2c 2c
47
A transformation γ → αγα−1 with α ∈ SL(2, Z) maps τγ → ατγ and a suitable choice of α
maps τγ to the standard fundamental√ domain F for SL(2, Z).2 The 2choice τγ ∈ F imposes
the condition Im (τγ ) ≥ Im (ρ) = 3/2 which requires (a + d) + 3c ≤ 4. Its solutions are
c = 1 and either a + d = 0 or a + d = ±1. By a suitable translation of τγ to the fundamental
domain F we may set a, d ∈ {0, 1} corresponding to the following matrices,
     
0 −1 0 −1 1 −1
S= ST = TS = (3.2.10)
1 0 1 1 1 0
for which τS = i and τST = ρ and τT S = ρ0 . Since we have S 2 = (ST )3 = −I, their orders
are 4 and 6 respectively. These are all the elliptic points of SL(2, Z).

3.2.4 Cusps
Topologically, F is a sphere with one puncture at τ = i∞, referred to as the cusp of the
fundamental domain of SL(2, Z). The cusp does not belong to H. It is invariant under the
infinite Borel subgroup of translations and sign reversal,
Γ∞ = {±T n , n ∈ Z} (3.2.11)
Under γ ∈ Γ∞ , the cusp in F is mapped to the cusp of the corresponding fundamental
domain γF . Under an arbitrary γ ∈ SL(2, Z) with c 6= 0, the cusp is mapped to a rational
number, and more precisely,
SL(2, Z) i∞ = Q ∪ {i∞} (3.2.12)
so that each rational number is a cusp of SL(2, Z) in the corresponding fundamental domain.

3.3 Modular functions, modular forms, and cusp forms

We shall now define modular functions, modular forms, and cusp forms. Let f (τ ) be a
meromorphic function on the upper half-plane H, which obeys the following transformation
law for some integer k ∈ Z, and for all γ ∈ SL(2, Z),
 
k a b
f (γτ ) = (cτ + d) f (τ ) γ= (3.3.1)
c d
Since f is a periodic function under τ → τ + 1, it has a Fourier expansion given by,
X
f (τ ) = an q n q = e2πiτ (3.3.2)
n∈Z

1. f is a modular function if f is meromorphic at i∞, namely has only a finite number

of non-vanishing Fourier coefficients an with negative n;
2. f is a modular form of weight k if it is holomorphic in H and at i∞, namely its Fourier
coefficients an vanish for all negative n, and thus f is finite at i∞;
3. f is a cusp form of weight k if it is a modular form of weight k and vanishes at i∞.

48
3.3.1 The ring of modular forms
The space Mk of modular forms of weight k for the group SL(2, Z) is a vector space over C.
Since the product of two modular forms of weight k and ` is a modular form of weight k + `,
the space of modular forms of arbitrary weight forms a ring M graded by the weight,
M
M= Mk Mk M` ⊂ Mk+` (3.3.3)
k

The ratio of two modular forms, in general, will not be a modular form, as dividing by a cusp
form will produce an object that is not holomorphic at ∞. In the next few subsections, we
shall show that there are no modular forms of negative weight and that M is a polynomial
ring generated by the modular forms G4 and G6 .

3.4 Modular forms as differential forms

It is useful to view modular forms as SL(2, Z)-invariant differential forms. Using the following
transformation rule for γ ∈ SL(2, Z) for the differential of τ ,
dτ
d(γτ ) = (3.4.1)
(cτ + d)2
we see that a modular form f of weight k is naturally mapped to a meromorphic differential
form of weight ( k2 , 0),
k
f = f (τ )(dτ ) 2 (3.4.2)

which is strictly invariant under γ ∈ SL(2, Z). This correspondence will be very useful in
both mathematics and physics applications. The mismatch by a factor of 2 between the
modular weight k of f and the weight k2 of the associated differential form f is conventional.

3.5 Eisenstein series

Eisenstein series Gk were already encountered when studying elliptic functions in (2.4.5).
In this subsection, we shall show that they are modular forms of weight k and obtain their
Fourier series decomposition. In the next subsection we shall show that they actually gen-
erate the complete polynomial ring of all modular forms.
We begin by considering the Eisenstein series defined in (2.4.5),4
X 1 0
X 1
Gk (ω1 , ω2 ) = Gk (Λ) = k
= (3.5.1)
ω∈Λ0
ω m,n∈Z
(mω1 + nω2 )k

4
Throughout, the prime superscript on the sum instructs us to omit the term with m = n = 0.

49
The sums that define Gk are absolutely convergent for k ≥ 3, and vanish for odd k in
view of the lattice symmetry Λ = −Λ. The resulting Gk (ω1 , ω2 ) is invariant under SL(2, Z)
transformations of the periods ω1 , ω2 since Gk depends only on the lattice Λ and not on
the specific periods chosen to represent Λ. The functions Gk (ω1 , ω2 ) are also manifestly
homogeneous in the periods of degree −k. In summary, under an arbitrary transformation
γ ∈ SL(2, Z), which maps Λ to Λ and the periods ω1 , ω2 into periods ω10 , ω20 given in terms
of γ in (3.1.1), and an arbitrary complex scaling factor λ ∈ C∗ , we have,
Gk (ω10 , ω20 ) = Gk (ω1 , ω2 )
Gk (λω1 , λω2 ) = λ−k Gk (ω1 , ω2 ) (3.5.2)
Scaling allows us to define a function of τ = ω2 /ω1 on the upper half-plane H by factoring
out a power of ω1 or, equivalently, by setting ω1 = 1,
0
X 1
Gk (τ ) = Gk (1, τ ) = ω1k Gk (ω1 , ω2 ) = (3.5.3)
m,n∈Z
(m + nτ )k

Combining the lattice automorphism and the scaling transformation of (3.5.2) with the
definition of Gk in (3.5.3) gives the modular transformation law for Gk (τ ),
 0 k
ω1
Gk (γτ ) = Gk (τ ) = (cτ + d)k Gk (τ ) (3.5.4)
ω1
This transformation may also be inferred directly from the sum in (3.5.3). Thus, Gk (τ ) is a
modular function. To show that it is a modular form we study its Fourier decomposition.

3.5.1 Fourier decomposition of Eisenstein series

Since the Eisenstein series Gk (τ ) is a holomorphic function for τ ∈ H as well as a periodic
function of τ with period 1 it admits a Fourier expansion of the form,
X
Gk (τ ) = aν (k) q ν q = e2πiτ (3.5.5)
ν∈Z

To evaluate the Fourier series, we decompose the double sum of (3.5.3) that defines Gk (τ )
by isolating the contributions from n = 0,
X 1 ∞ X
X 1
Gk (τ ) = + 2 (3.5.6)
m6=0
mk n=1 m∈Z
(m + nτ )k

As τ approaches the cusp τ → i∞, the double sum in (3.5.6) tends to zero. As a result, all
expansion coefficients aν (k) vanish for ν < 0 and the first sum in (3.5.6) gives a0 (k),
a0 (k) = 2ζ(k) 4 ≤ k ∈ 2N (3.5.7)

50
Thus, Gk (τ ) has a finite limit at the cusp and is a modular form of weight k. To evaluate
the double sum in (3.5.6), we use the following formula,
∞
X 1 1 + e2πiz X
= −iπ 2πiz
= −iπ − 2πi e2πi`z (3.5.8)
m∈Z
z + m 1 − e `=1

which may be established by using the fact that both sides are meromorphic in z with simple
poles of residue 1 at all integers and are periodic with period 1. Actually, we shall need the
derivative of order k − 1 of this formula,
∞
X 1 (2πi)k X k−1 2πi`z
k
= ` e (3.5.9)
m∈Z
(z + m) Γ(k) `=1

Setting z = nτ and including the sum over n, we obtain the expression for the second sum
in (3.5.6). Putting all together gives the following Fourier series decomposition for Gk (τ ),
∞ ∞
(2πi)k X X k−1 n`
Gk (τ ) = 2ζ(k) + 2 ` q (3.5.10)
Γ(k) n=1 `=1

Changing summation variables from (n, `) to (N, `) with N = n`, the sum over ` for given
N may be rearranged in terms of the sums of divisors functions, defined for any α ∈ C by,
X
σα (N ) = `α (3.5.11)
`|N

The sum is over all positive divisors ` of N , including the divisors 1 and N . In terms of sums
of divisor functions, we obtain our final expression for the Fourier series of Gk (τ ),
∞
(2πi)k X
Gk (τ ) = 2ζ(k) + 2 σk−1 (N )q N (3.5.12)
Γ(k) N =1

Factoring out 2ζ(k), and using the relation between Bernoulli numbers and ζ(k) for k even
(2πi)k Bk = −2k!ζ(k) we obtain the normalized Eisenstein series Ek , defined by,
∞
Gk (τ ) X 2k
Ek (τ ) = = 1 + νk σk−1 (N ) q N νk = − (3.5.13)
2ζ(k) N =1
Bk

With this normalization, we manifestly have Ek (τ ) → 1 as τ → i∞ for any even k ≥ 4. The

first few values for νk are given as follows,
65520
ν4 = 240 ν6 = −504 ν8 = 480 ν10 = −264 ν12 = (3.5.14)
691
51
In particular, the modular forms g2 , g3 , G4 , G6 , and the discriminant ∆ of (2.4.14) for the
lattice Λ = Z ⊕ τ Z are given in terms of the normalized Eisenstein series by,
4π 4
g2 (τ ) = 60 G4 (τ ) = E4 (τ )
3
8π 6
g3 (τ ) = 140 G6 (τ ) = E6 (τ )
27
(2π)12
∆(τ ) = g23 − 27g32 = 3 2


E 4 (τ ) − E 6 (τ ) (3.5.15)
123
Since Ek (i∞) = 1 the discriminant vanishes at the cusp and we have ∆(τ ) = (2π)12 q +O(q 2 ).
While the functions Gk are modular forms of weight k for even k ≥ 4, the discriminant ∆ is
a cusp form of weight 12.

3.5.2 Poincaré series representation

One construction of elliptic functions for a lattice Λ is through the method of images, exhib-
ited in all generality in (2.3.4) and shown in detail for the case of the Weierstrass function in
(2.4.1). One may view the construction by the method of images as performing a sum over
the Abelian group Λ under which every elliptic function is invariant.
To obtain modular functions and modular forms, which are defined to be invariant or
covariant under the group SL(2, Z), an analogous method is available through the so-called
Poincaré series. This method may be applied to obtain meromorphic modular forms, as
will be the case here, but also non-holomorphic modular functions and forms, as will be the
case in subsection 4.2 and section 9. To construct a modular form by a Poincaré series one
starts with a meromorphic function g or a meromorphic differential form g of weight k and
aims to sum over its images under the modular group. In many practical cases, however, the
function g or the form g is invariant under a “stability subgroup” Γstab of SL(2, Z). When
this subgroup has an infinite number of elements, as will often be the case, the sum over all
images under SL(2, Z) would diverge. The Poincaré series is instead defined by summing
over the images under the cosets Γstab \SL(2, Z). For the case of a meromorphic function g
we obtain a meromorphic modular function f given by the Poincaré series,
X
f (τ ) = g(γτ ) (3.5.16)
γ ∈ Γstab \SL(2,Z)

The function g is usually referred to as the seed of the Poincaré series. The corresponding
formula defines the Poincaré series for a modular form f from a seed g. As noted originally
by Poincaré, the seed function for a given modular function is not unique. In particular, the
Poincaré sum of certain functions may vanish.
By way of example, we may take the seed to be simply g = (dτ )k for some positive
integer k. This seed is invariant under −I ∈ SL(2, Z) because τ is invariant, as well as

52
under the following Borel subgroup of SL(2, Z) since dτ is invariant under shifts τ → τ + b,
   
1 b
Γ∞ = ± ,b∈Z (3.5.17)
0 1
so that Γstab = Γ∞ . The remaining Poincaré series for even k ≥ 4 is given by,
X (dτ )k 1 X (dz)k
f= = (3.5.18)
(cτ + d)k 2 (cτ + d)k
γ∈Γ∞ \SL(2,Z) c,d∈Z,
gcd(c,d)=1

The restriction gcd(c, d) = 1 implicitly assumes that (c, d) 6= (0, 0). The form f is closely
related to the Eisenstein series defined in (3.5.3), as may be seen by factoring out the greatest
common divisor p = gcd(m, n) ∈ N from the summation variables by parametrizing them as
follows (m, n) = (pd, pc) with gcd(c, d) = 1,
0
X 1 X 1
Gk (τ ) = k
= ζ(k) (3.5.19)
m,n∈Z
(m + nτ ) c,d∈Z,
(cτ + d)k
gcd(c,d)=1

so that f = 2ζ(k)Gk (τ )(dz)k for k > 2 even.

3.5.3 The polynomial ring of Eisenstein series

The Eisenstein series G2k is a modular form of weight 2k for any integer k ≥ 2. We shall
now show that the Eisenstein series G2k may be expressed as a polynomial in G4 and G6
with rational coefficients, so that the vector space generated by the Eisenstein series G2k is
a polynomial ring generated by G4 and G6 .
To show this, we differentiate the differential equation for the Weierstrass ℘-function
0 2
(℘ ) = 4℘3 − 60G4 ℘ − 140G6 to obtain the following second order differential equation,
℘00 = 6℘2 − 30G4 (3.5.20)
which is even in z. The Laurent expansion for ℘, obtained earlier in (2.4.4) for an arbitrary
lattice Λ, may be expressed as follows for the lattice Λ = Z ⊕ τ Z,
∞
1 X
℘(z) = + (2k + 1)G2k+2 z 2k (3.5.21)
z 2 k=1

Substituting this expression and its double derivative into (3.5.20), we see that the terms in
z −4 , z 0 , and z 2 automatically match. For the remaining terms, after some simplifications,
we obtain the following recursion relation for k ≥ 4,
k−3
X 3(2` + 1)(2k − 2` − 3)
G2k = G2`+2 G2k−2`−2 (3.5.22)
`=1
(k − 3)(2k − 1)(2k + 1)

53
which shows recursively that G2k for k ≥ 4 is a polynomial in G2m+4 with 0 ≤ m ≤ k − 4,
of homogeneous weight 2k. Therefore, G2k for k ≥ 4 is a polynomial in G4 and G6 . For
example, to the lowest few orders we have,

G8 = 37 G24 G14 = 30
143
G24 G6
5
G10 = 11 G4 G6 G16 = 9
221
G44 + 2431
300
G4 G26
G12 = 18
143
G34 + 25
143
G26 G18 = 3915
46189
G34 G6 + 4199
125
G36 (3.5.23)

Thus, the subspace of the graded ring of all modular forms M that is generated by the
Eisenstein series G2k for all k ≥ 4 is a polynomial ring generated by G4 and G6 . While the
modular forms G4 , G6 , G8 , G10 , and G14 are the only Eisenstein series of weight 4, 6, 8, 10,
and 14 respectively, we see that at weight 12 there are two independent modular forms,
namely G34 and G26 , whose difference involves the discriminant ∆.
Equivalently, we may express G2k in terms of the normalized Eisenstein series E2k , to
obtain the following formula,
k−3
X 6(2` + 1)(2k − 2` − 3)ζ(2` + 2)ζ(2k − 2` − 2)
E2k = E2`+2 E2k−2`−2 (3.5.24)
`=1
(k − 3)(2k − 1)(2k + 1)ζ(2k)

in terms of which the relations (3.5.23) become,

E8 = E24 E14 = E24 E6

1617 4
E10 = E4 E6 E16 = E + 2000
3617 4
E E2
3617 4 6
E12 = 441
691
E34 + 250
691
E26 E18 = 38367 3 5500
E E + 43867
43867 4 6
E36 (3.5.25)

These relations, together with the Fourier series expansions of the Ek in (3.5.13), imply an
infinite number of identities between sums of divisor functions. The identities corresponding
to the relations E8 = E24 and E10 = E4 E6 are given as follows,
N
X −1
σ7 (N ) = σ3 (N ) + 120 σ3 (N − M )σ3 (M )
M =1
N
X −1
11 σ9 (N ) = 21σ5 (N ) − 10σ3 (N ) + 5040 σ3 (N − M )σ5 (M ) (3.5.26)
M =1

More generally, the divisor functions σ2k−1 may be expressed as polynomials in σ3 and σ5 .

3.6 Dimension and generators of the ring of modular forms

To obtain the dimension of the space Mk of all modular forms of weight k, we begin by
proving the following theorem on modular functions of weight k. We shall denote by ordf (p)

54
 − 12 + iT 1
2
+ iT

Freg

•
ρ • i • ρ0

H
− 12 0 1
2

Figure 7: Integration contour of the regularized fundamental domain Freg .

the order of a zero (counted positively) or pole (counted negatively) of f at a point p, and
by ordf (∞) the exponent of the leading term in the q-expansion of f (τ ). Meromorphicity of
modular functions guarantees that ordf (p) are integers, including for the cusp p = i∞ and
for the elliptic points p = i and p = ρ = e2πi/3 .

Theorem 3.2 A non-zero modular function f of weight k satisfies the valence formula,
1 1 X k
ordf (i∞) + ordf (i) + ordf (ρ) + ordf (p) = (3.6.1)
2 3 12
p∈F \{i∞,i,ρ}

where F is the standard fundamental domain of the group SL(2, Z) given in (3.2.6).

To prove the theorem, we count the number of zeros and poles of f by integrating the
logarithmic derivative f 0 /f over a suitable contour. Doing so presents complications at the
points i∞, i, and ρ, so we regularize the domain F to a domain Freg in which we cut off F
near those points, as shown in figure 7. The small circular arcs near i, ρ, and ρ0 have radius
1/T . For simplicity, we shall assume that f (τ ) has neither zeros nor poles on ∂Freg , and
refer to standard mathematics textbooks when this is not the case.
On the one hand, the residue theorem gives,
f 0 (τ )
I
1 X
lim dτ = ordf (p) (3.6.2)
T →∞ 2πi ∂F f (τ )
reg p∈F reg

55
On the other hand, the integral along ∂Freg may be evaluated by using the relations under
modular transformations that identify the different line segments of ∂Freg with one another.
First, the contributions from the two vertical edges cancel one another, as their integrals
are opposite. Second, we evaluate the contributions at the point i∞ by using the dominant
behavior at Im (τ ) = T  1 given by f (τ ) ∼ q ordf (i∞) = e2πiτ ordf (i∞) to obtain the contri-
bution −ordf (i∞). The minus sign arises from the orientation of the contour. Analogous
contributions are collected at i, but are counted with a factor of 12 since the integration
contour is only half a circle. Similarly, at ρ, the contribution is only 61 of a full circle, but
there are identical contributions from ρ and ρ0 which accounts for the final factor of 13 . The
above contributions account for the left side of (3.6.1). Finally, we collect the contribution
along the |τ | = 1 segments of the contour. The contributions to the left and to the right
of i are related to one another by the transformation S : τ → −1/τ , which acts on f (τ ) by
f (Sτ ) = τ k f (τ ), so that taking the differential of the logarithm we have,

f 0 (Sτ ) f 0 (τ ) dτ
d(Sτ ) = dτ +k (3.6.3)
f (Sτ ) f (τ ) τ
1
The integration arc goes from τ = ρ to τ = i and spans 12 of a full circle, reproducing the
right side of the (3.6.1) and thereby completing the proof of the theorem.

3.6.1 Dimension and generators of Mk

While Theorem 3.2 applies to any modular function of weight k, whether holomorphic or
not, we shall now specialize to modular forms, which are holomorphic, and use Theorem 3.2
to compute the dimensions of the spaces Mk . Modular forms f (τ ) can have no poles, so
that ordf (p) ≥ 0, including at the points p = i∞, i, ρ.

Theorem 3.3 Theorem 3.2 implies the following properties of Mk :

1. For k odd, k < 0, and k = 2 there are no modular forms, i.e. dim Mk = 0;
2. For k even and k ≥ 0 the space Mk is finite-dimensional;
3. For k = 0, 4, 6, 8, 10, 14, we have dim Mk = 1 and Mk = CGk ;
4. The subspace Sk of weight k cusp forms is empty for k < 12, while S12 = C∆;
5. For k ≥ 16, we have Sk = ∆Mk−12 and Mk = CGk ⊕ Sk .

For low weight these results are consistent with the results of (3.5.23). Moreover, the
theorem implies that M is precisely the polynomial ring freely generated by G4 and G6 .
To prove Theorem 3.3, we use the valence formula (3.6.1) of Theorem 3.2. Considering
the relation (3.6.1) multiplied by 12, and then mod 2, we see that there can be no solutions
for k odd. Since modular forms have ordf (p) ≥ 0, including at the points p = i, ρ, and i∞,

56
the left side of (3.6.1) is non-negative so that there can be no solutions to (3.6.1) for k < 0.
Clearly, there is also no solution when k = 2, which concludes the proof of point 1. For k
even and positive, there can only be a finite number of solutions since the sum on the left
of (3.6.1) is over positive or zero numbers only, proving point 2. To proceed further, it is
helpful first to prove two auxiliary statements,

G4 (ρ) = G6 (i) = 0 (3.6.4)

They follows from inspection of the sums,

0 0
X 1 1 X 1
G4 (ρ) = 3 4
= 4 = ρ2 G4 (ρ)
(m ρ + n ρ) ρ (m ρ + n)4
2
(m,n) (m,n)
0 0
X 1 1 X 1
G6 (i) = 4 6
= 6 = −G6 (i) (3.6.5)
(m i + n i) i (m i + n)6
3
(m,n) (m,n)

To recast the second sum on the first line, we use mρ2 + n = m(−1 − ρ) + n = n − m − mρ
so that the sum over (m, n) gives G4 (ρ). To recast the second sum on the second line, we
use m i3 + n = −im + n, so that the sum over (m, n) gives G6 (i).
By inspection of (3.6.1), we must have ordf (i∞) = 0 for k = 4, 6, 8, 10 since the right side
of (3.6.1) is less than 1, as well as for k = 14 since the remaining 1/6 cannot be accounted
for by non-vanishing ordf (i) or ordf (ρ). Therefore, the solutions to (3.6.1) for these values of
k are unique and respectively given by (ordf (i), ordf (ρ)) = (0, 1), (1, 0), (0, 2), (1, 1), (1, 2).
They are generated by G4 , G6 , G8 ∼ G24 , G10 ∼ G4 G6 , G14 ∼ G24 G6 , which proves point
3. To prove point 4, we note that a cusp form has ordf (i∞) ≥ 1, which requires k ≥ 12,
and uniqueness for k = 12. Finally for k ≥ 16, the solutions manifestly satisfy the recursion
relation Sk = ∆Mk−12 and Mk = CGk ⊕ Sk by inspecting (3.6.1), proving point 5.

3.7 Modular functions and the j-function

Having classified all modular forms under SL(2, Z) of weight k, we now classify all mero-
morphic modular functions of weight k. Given an arbitrary meromorphic modular function
of weight k, we may always divide by a modular form of weight k, such as Gk , to obtain a
meromorphic modular function of weight 0. Thus, it will suffice to understand the space of
modular functions of weight 0. Sums, products, and ratios of modular functions of weight 0
produce again modular functions of weight 0, so that the corresponding space forms the field
of modular functions of weight 0. We seek to obtain a complete description of this space.
Clearly, the only holomorphic modular function of weight 0 is constant in τ . We begin
by constructing a modular function with a single pole at the cusp and no other poles.

57
Normalizing the residue of the pole to unity, we define the famous j-function by,
123 g23 123 E34 1
j(τ ) = = 3 2
= + O(q 0 ) (3.7.1)
∆ E4 − E6 q
Inspecting (3.6.1), we have k = 0 since we are dealing with a modular function of weight zero;
ord∞ (j) = −1 since j has a simple pole at ∞; ordi (j) = 0 since E4 (i) 6= 0; and ordρ (j) = 3
since j has a triple zero at ρ in view of the identity G4 (ρ) = E4 (ρ) = 0 established in (3.6.4).
These values indeed satisfy (3.6.1), and we have,5
j(i∞) = ∞ j(i) = 1 j(ρ) = 0 (3.7.2)
The classification of modular functions is given as follows.

Theorem 3.4 Every rational function of j(τ ) is a modular function of SL(2, Z). Con-
versely, every modular function f (τ ) of SL(2, Z) of weight 0 is a rational function of j.
In terms of the zeros zn and poles pn (both of which are to be counted with multiplicities,
including at i and ρ), we have the explicit formula,
N
Y j(τ ) − j(zn )
f (τ ) = (3.7.3)
n=1
j(τ ) − j(pn )
This result completes the construction of the field of modular functions of weight 0 and,
combining it with our earlier general construction of modular forms of arbitrary weight, also
gives the construction for all meromorphic SL(2, Z)-covariant functions of arbitrary weight.

3.7.1 Modular forms with prescribed zeros

A final comment is in order about modular forms with prescribed zeros. From (3.6.1) we
see that a zero at a regular point in F requires k ≥ 12, and the addition of one regular zero
requires shifting k → k + 12. To construct, for example, the modular forms of weight 12
with one zero at an arbitrary point p ∈ F \ {i∞, i, ρ}, we take a linear combination of the
two modular forms available, namely E34 and E26 , and arrange the coefficients so that a zero
emerges at p,
 
fp (τ ) = f0 E4 (τ )3 E6 (p)2 − E4 (p)3 E6 (τ )2 (3.7.4)

The normalization factor f0 may be fixed by evaluation at any other point, including i∞.
One proceeds analogously for modular forms with more zeros at prescribed points.
5
The first 19 orders in the q-expansion of j are as follows: j(τ ) = q −1 + 744 + 196884 q + 21493760 q 2 +
864299970 q 3 + 20245856256 q 4 + 333202640600 q 5 + 4252023300096 q 6 + 44656994071935 q 7 +
401490886656000 q 8 + 3176440229784420 q 9 + 22567393309593600 q 10 + 146211911499519294 q 11 +
874313719685775360 q 12 + 4872010111798142520 q 13 + 25497827389410525184 q 14 +
126142916465781843075 q + 593121772421445058560 q + 2662842413150775245160 q 17 + O(q 18 ).
15 16

58
3.8 Modular transformations of Jacobi ϑ-functions
The Jacobi ϑ-functions are subject to more delicate modular transformation laws, which we
shall now derive. Recall the definition of the basic ϑ-function,
2
X
ϑ(z|τ ) = eiπτ n +2πinz (3.8.1)
n∈Z

Setting z = 0 and τ = it for t real and positive, we may apply the Poisson summation
formula (2.1.11) to obtain the transformation rule for S ∈ SL(2, Z),
1
ϑ(0| − τ1 ) = (−iτ ) 2 ϑ(0|τ ) (3.8.2)
Although originally derived for t real, the formula remains valid for τ ∈ H provided we
choose the proper branch cut for the square root. Under the modular transformation T ,
however, the ϑ-function does not transform into itself, but instead we have,
ϑ(z|τ + 1) = ϑ4 (z|τ ) (3.8.3)
where ϑ4 is a ϑ-function with half-integer characteristics defined in (2.6.10). The set of four
ϑ-functions with half characteristics transforms into itself, and we have,
1 2 /τ
ϑ1 (z|τ + 1) = ε ϑ1 (z|τ ) ϑ1 ( τz | − τ1 ) = −i(−iτ ) 2 eiπz ϑ1 (z|τ )
1
iπz 2 /τ
ϑ2 (z|τ + 1) = ε ϑ2 (z|τ ) ϑ2 ( τz | − τ1 ) = (−iτ ) e 2 ϑ4 (z|τ )
1
iπz 2 /τ
ϑ3 (z|τ + 1) = ϑ4 (z|τ ) ϑ3 ( τz | − τ1 ) = (−iτ ) e 2 ϑ3 (z|τ )
1
iπz 2 /τ
ϑ4 (z|τ + 1) = ϑ3 (z|τ ) ϑ4 ( τz | − τ1 ) = (−iτ ) e 2 ϑ2 (z|τ ) (3.8.4)
where ε = e2πi/8 . One may collect these transformation laws in a single expression for each
transformation, valid for all half-characteristics,
   
α −iπα(1+α) α
ϑ (z|τ + 1) = e ϑ (z|τ )
β β + α + 21
    
α z 1 1
2πiαβ+iπz 2 /τ β
ϑ − τ = (−iτ ) 2 e ϑ (z|τ ) (3.8.5)
β τ −α
Combining the two, one finds that under a generic element γ ∈ SL(2, Z), the ϑ1 function
transforms as follows,
1 2 /(cτ +d)
ϑ1 (z 0 |τ 0 ) = εϑ (γ)(cτ + d) 2 eπicz ϑ1 (z|τ ) (3.8.6)
from which the transformation of the ϑ-function with arbitrary characteristic may be deduced
using (2.6.7), and we find,
 0 1
α + 21
 
0 0 0 1
πicz 2 /(cτ +d) α +
ϑ (z |γτ ) = ε(α, β α , β , γ)(cτ + d) 2 e ϑ 0 12 (z|τ ) (3.8.7)
β + 12 β +2

59
where,
0 0
ε(α, β α0 , β 0 , γ) = εϑ (γ) eπi{α(β+1)−α (β +1)} (3.8.8)
and z and the characteristics transform as follows,
 0    
0 z α aα + cβ a b
z = = γ= (3.8.9)
cτ + d β0 bα + dβ c d
Finally, εϑ (γ) and ε(α, β α0 , β 0 , γ) are both eight roots of unity and εϑ (γ) is independent of
the characteristics α, β. Its evaluation is in terms of the generalized Legendre symbol and
will be given in subsection 3.9 on the Dedekind η function.
Setting z = 0 in the theta functions gives so-called theta-constants, whose modular trans-
formation laws are special cases of the general transformation laws presented above. None
of these are modular forms under SL(2, Z), though we shall see later that they are modular
forms under certain congruence subgroups of SL(2, Z). Furthermore, the derivative θ10 (0|τ )
is closely related to the discriminant ∆ and various sums of eight powers of ϑ2 , ϑ3 , and ϑ4
are modular forms. We now discuss both of these constructions in turn.

3.8.1 The discriminant ∆: ϑ-constants and product formula

The modular transformation law for ϑ01 (0|τ ) is readily obtained from (3.8.4) and is given
below along with its eighth power,
3
ϑ01 (0|τ + 1) = ε ϑ01 (0|τ ) ϑ01 (0| − τ1 ) = (−iτ ) 2 ϑ01 (0|τ )
ϑ01 (0|τ + 1)8 = ϑ01 (0|τ )8 ϑ01 (0| − τ1 )8 = (−iτ )12 ϑ01 (0|τ )8 (3.8.10)
While ϑ01 itself is not a modular form, its eight power is a modular form of weight 12, and
therefore must be a linear combination of E34 and E26 . To obtain a precise relation, we appeal
to the infinite product formula for ϑ1 (z|τ ) given in (2.6.23), and find the following expression,
∞
1
Y
ϑ01 (0|τ ) = πϑ2 (0|τ )ϑ3 (0|τ )ϑ4 (0|τ ) = 2πq 8 (1 − q n )3 (3.8.11)
n=1
1
This expression indeed fails to be a modular form in view of the q 8 prefactor. However, its
eight power produces a factor of q which is not only consistent with being a genuine modular
form, but also shows that it is a cusp form proportional to the discriminant ∆. The constant
of proportionality is easily fixed by matching the term of order q in the Fourier expansion of
both. As a result, we find not only a relation between θ10 (0|τ )8 and the discriminant ∆ but
also a product formula for ∆,
∞
Y
∆(τ ) = (2π)4 ϑ01 (0|τ )8 12
= (2π) q (1 − q n )24 (3.8.12)
n=1

60
The product formula for the discriminant shows that ∆(τ ) is nowhere vanishing for τ ∈ H,
and has a simple zero in q at the cusp.
The Fourier coefficients τ (n) of ∆, defined by,
∞
X
12
∆(τ ) = (2π) τ (n)q n (3.8.13)
n=1

are referred to as the Ramanujan τ -function, and have special arithmetic significance. The
lowest orders are given as follows,

τ (1) = 1 τ (4) = −1472 τ (7) = −16744 τ (10) = −115920

τ (2) = −24 τ (5) = 4830 τ (8) = 84480 τ (11) = 534612
τ (3) = 252 τ (6) = −6048 τ (9) = −113643 τ (12) = −370944 (3.8.14)

The relation 1728∆ = (2π)12 (E43 −E62 ) along with E42 = E8 allows us to express the τ -function
in terms of sums of divisor functions as follows,
n−1
X
200


τ (n) = 3
σ3 (m)σ7 (n − m) − 147 σ5 (m)σ5 (N − m)
m=1
5 7 5
+ 36 σ3 (n) + 12
σ5 (n) + 18
σ7 (n) (3.8.15)

Eliminating σ7 in terms of σ3 using the first formula in (3.5.26) confirms that τ (n) may be
expressed solely in terms of σ3 and σ5 . Ramanujan conjectured that τ (n) is a multiplicative
arithmetic function, which satisfies,

τ (mn) = τ (m)τ (n) gcd(m, n) = 1 (3.8.16)

where gcd(m, n) stands for the greatest common divisor of m and n. Ramanujan’s function is,
however, not completely multiplicative (which would mean that the first equation in (3.8.16)
would hold for arbitrary m, n, not necessarily relatively prime) and instead satisfies,
X  mn 
τ (m)τ (n) = d11 τ (3.8.17)
d2
d | gcd(m,n)

Both conjectures on τ (n) were proven by Mordell in 1917 using the theory of Hecke operators
applied to the cusp form ∆(τ ), and we shall provide this proof in section 10.

3.8.2 Sums of eight powers

Another way to construct a modular form using the ϑ-constants is by sums of eighth powers.
The modular transformation properties show right away that the following sums,

T4k (τ ) = ϑ2 (0|τ )8k + ϑ2 (0|τ )8k + ϑ2 (0|τ )8k (3.8.18)

61
are modular forms of weight 4k. For k = 1, 2, the corresponding spaces of modular forms
are one-dimensional, so that we must have T4 = 2E4 and T8 = 2E8 = 2E24 by matching
asymptotics at the cusp. In particular, we may express the modular j-function in terms of
eighth powers of ϑ-constants,

3
g2 (τ )3 ϑ2 (0|τ )8 + ϑ3 (0|τ )8 + ϑ4 (0|τ )8
j(τ ) = 1728 = 32 (3.8.19)
∆(τ ) ϑ2 (0|τ )8 ϑ3 (0|τ )8 ϑ4 (0|τ )8

It may be verified directly from the asymptotics of the ϑ-constants at the cusp that j(q) has
a simple pole at the cusp with unit residue j(q) ∼ q −1 .

3.9 The Dedekind η-function

The Dedekind η-function is defined as follows,
∞
1
Y
η(τ ) = q 24 (1 − q n ) (3.9.1)
n=1

It is not a modular form of SL(2, Z), but its 24th power is a modular form of weight 12 and
related to the discriminant as follows,

∆(τ ) = (2π)12 η(τ )24 (3.9.2)

Under the generators T and S of the modular group, η transforms as follows,

η(τ + 1) = e2πi/24 η(τ )

1
η(− τ1 ) = (−iτ ) 2 η(τ ) (3.9.3)

The first relation may be read off from the definition of η, while the second relation may be
derived first for τ ∈ iR+ using Poisson resummation, and then be extended to τ ∈ H.6
The transformation of η(τ ) under an arbitrary γ ∈ SL(2, Z) may be expressed as follows,
 
1 a b
η(γτ ) = εη (γ) (cτ + d) 2 η(τ ) γ= (3.9.4)
c d

with εη (γ)24 = 1, or ε : SL(2, Z) → Z24 . The function εη (γ) depends on γ, but not on τ .
Its values may be determined from the transformation rules given in (3.9.3). Note that the
actions of the transformations γ and −γ coincide on τ but differ on the square root and on
6
The square root is defined via the argument function arg(τ ) for τ ∈ C∗ , in terms of which τ =
1 1
|τ | exp(i arg(τ )) and −π ≤ arg(τ ) < π by τ 2 = |τ | 2 exp(i arg(τ )/2) (see for example [7]). This definition is
compatible with the fact that for τ = it with t ∈ R+ we have η(it), η(i/t) > 0.

62
 1
εη giving, in particular, the value εη (−I) = (−1) 2 . In view of this property, εη (γ) is not a
proper homomorphism of SL(2, Z) → Z24 but instead is referred to as a multiplier system.
Special values may be deduced from the basic transformation laws (3.9.3),

εη (T n ) = e2πin/24 εη (S) = e−2πi/8 εη (−I) = e2πi/4 = i (3.9.5)

In particular, if γ is such that either c > 0, or c = 0 and d > 0, then ε(−γ) = iε(γ). The
Petersson formula for εη (γ) for a γ with either c > 0, or c = 0 and d > 0 is given as follows,
(see [7])

εη (γ) = (d|c) exp 2πi (a + d)c + bd(1 − c2 ) − 3c



c odd: 24
(3.9.6)
εη (γ) = sign(d)(c|d) exp 24 (a + d)c + bd(1 − c2 ) + 3d + 3 − 3cd
 2πi

c even:

We recall from appendix A, and in particular (A.4.3), that the Jacobi symbol (n|N ) for odd
denominator NQ(which is always the case here)
Q with decomposition into distinct primes pi
αi
given by N = i pi is defined by (n|N ) = i (n|pi ) where (n|pi ) is the Legendre symbol for
an odd prime pi . The definition is extended to include n = 0 by (0|1) = (0| − 1) = 1 and
negative N by (n|N ) = (n| − N ).
Knowledge of εη (γ) allows us to produce a formula for the multiplier εϑ (γ) encountered
in the modular transformation law of the ϑ-functions in (3.8.7) by using the relation,

ϑ01 (0|τ ) = 2πη(τ )3 (3.9.7)

which is obtained by combining (3.8.11) with the definition of the η-functions. The modular
transformation of the left side may be derived from (3.8.6) while the modular transformation
of the right side is given by (3.9.4), and we obtain,

εϑ (γ) = εη (γ)3 (3.9.8)

consistent with the fact that εϑ is an 8-th root of unity while εη is a 24-th root of unity.
Finally, we note that the fourth power ε4η (γ) no longer suffers from the subtleties of sign
reversal of γ and the square root, and is a genuine homomorphism ε4η : SL(2, Z) → Z6 with
ε4η (−I) = 1 and ε4η (γ1 γ2 ) = ε4η (γ1 )ε4η (γ2 ).

63
• Bibliographical notes
The books by Apostol [8] and Lang [9] and the lecture notes by Zagier [10] offer classic
introductions to the subject of modular forms. The books by Shimura [11] and Koblitz [12]
provide advanced and more abstract presentations aimed at arithmetic. A marvelously clear
and detailed presentation, to which we shall refer often in these notes, is given in the fairly
recent treatise by Diamond and Shurman [13]. A general perspective from the point of view
of automorphic functions is given in Bateman [14] and Iwaniec [15]. A detailed exposition of
identities involving the Dedekind η-function and the role they play in the theory of modular
forms may be found in the book by Köhler [7]. Presentations with an emphasis on L-functions
and q-series expansions may be found in the books by Bump [16] and Ono [17], respectively.

64
4 Variants of modular forms
In this section, we discuss a variety of objects that are closely related to modular forms,
including quasi-modular forms, almost-holomorphic modular forms, non-holomorphic Eisen-
stein series, Maass forms, mock modular forms, and quantum modular forms. Other gener-
alizations, including modular forms for congruence subgroups, vector-valued modular forms,
and modular graph forms, will be introduced in sections 7, 8, and 9 respectively.

4.1 Quasi-modular and almost-holomorphic modular forms

Quasi-modular forms and almost-holomorphic modular forms are generalizations of the mod-
ular forms introduced earlier, which are obtained by relaxing certain parts of the definition
of modular forms. We begin by reviewing the most famous of these objects, referred to
as E2 (τ ) and E∗2 (τ ) respectively, before proceeding to the definition and presentation of the
general properties of quasi-modular forms and almost-holomorphic modular forms.

4.1.1 E2 and E∗2

We introduced the holomorphic Eisenstein series Gk in (2.4.5) and (3.5.1), or equivalently
Ek in (3.5.13), which are given for even k ≥ 4 in terms of absolutely convergent sums over a
lattice Λ = Z ⊕ τ Z. For k = 2, the corresponding sum,
X 1 X 1
2
= (4.1.1)
ω∈Λ0
ω m,n∈Z
(m + nτ )2
(m,n)6=(0,0)

fails to be absolutely convergent. Instead, it is conditionally convergent and its value will
depend on how the infinite sums are arranged. Physicists would say that the sums need to
be “regularized”. There may be one or several choices that are deemed “natural” because
they preserve one property or another of modular forms.
One choice of regularization is to adopt the Eisenstein summation which consists in
cutting off the infinite sums in a symmetrical way, as was already used in (2.2.2),
N M
1 X X 1
E2 (τ ) = lim lim (4.1.2)
N →∞ M →∞ 2ζ(2) n=−N m=−M
(m + nτ )2
(m,n)6=(0,0)

For finite M and N the sum is holomorphic in τ and the limit M, N → ∞ produces a
finite holomorphic function E2 (τ ). However, E2 (τ ) fails to transform as a modular form
under SL(2, Z) transformations. This is not surprising as Theorem 3.3 tells us there are no
modular forms of weight 2. Instead the transformation law of E2 (τ ) is as follows,
12
E2 (γτ ) = (cτ + d)2 E2 (τ ) + c(cτ + d) (4.1.3)
2πi
65
To show this, we decompose the sum of (4.1.2) into contributions from n = 0 and n 6= 0,
N M
1 X X 1
E2 (τ ) = 1 + lim lim (4.1.4)
N →∞ M →∞ ζ(2)
n=1 m=−M
(m + nτ )2

and perform the sum over m using (3.5.9) in the convergent limit M → ∞. The remaining
sum over n admits a finite limit as N → ∞ and evaluates as follows,
∞ X
X ∞ ∞
X
E2 (τ ) = 1 − 24 ` e2πi`nτ = 1 − 24 σ1 (`)q ` (4.1.5)
n=1 `=1 `=1

Alternatively, the sum over n in (4.1.5) may be performed first and the remaining sum over
` may be rearranged in terms of the discriminant ∆(τ ),
1
E2 (τ ) = ∂τ ln ∆(τ ) (4.1.6)
2πi
Using the modular transformation law of the weight 12 modular form ∆ under SL(2, Z), we
deduce the transformation law of E2 (τ ) by differentiating ln(∆) to obtain (4.1.3). Remarkable
relations to the Weierstrass ζ and ℘ functions include the following,

π2 π2
ζ( 12 |τ ) = E2 (τ ) ℘(z|τ ) = −∂z2 ln ϑ1 (z|τ ) − E2 (τ ) (4.1.7)
6 3
The latter demonstrates how E2 (τ ) acts as a modular connection to guarantee the proper
modular transformation of ℘ given that of the ϑ1 -function.
Another choice of regularization was introduced by Siegel by defining the following sum,
1 X 1
E∗2 (τ ) = lim+ (4.1.8)
ε→0 2ζ(2) ω∈Λ0 ω 2 |ω|ε

Finite ε > 0 renders the sum over Λ0 absolutely convergent and the limit ε → 0 turns out to
be finite. The result is a weight 2 modular form under SL(2, Z),

E∗2 (γτ ) = (cτ + d)2 E∗2 (τ ) (4.1.9)

However, E∗2 (τ ) fails to be holomorphic in τ . The regularization factor |ω|ε introduces non-
holomorphic behavior for ε > 0 which leaves a non-holomorphic remnant as ε → 0,7
3
E∗2 (τ ) = E2 (τ ) − (4.1.10)
πτ2
7
Throughout, we shall use the notation τ = τ1 + iτ2 with τ1 , τ2 ∈ R.

66
where the holomorphic E2 (τ ) was given in (4.1.6). One verifies that the modular transfor-
mation laws of E2 and E∗2 map into one another under this relation.
In summary, while the Eisenstein series Ek for k ≥ 4 are holomorphic modular forms of
weight k, the regularization of the infinite sum for k = 2 allows for two “natural” choices:
one is holomorphic but not modular, while the other is modular but not holomorphic. One
cannot enforce both modularity and holomorphicity simultaneously. The incompatibility
between holomorphicity and modular invariance may be viewed as perhaps the first example
of what physicists refer to as an anomaly. Eisenstein’s result dates back to before 1850 !

4.1.2 The rings of almost-holomorphic modular and quasi-modular forms

A function f : H → C is an almost-holomorphic form of weight k with respect to the full
modular group SL(2, Z) if it transforms as follows under γ ∈ SL(2, Z),

f (γτ ) = (cτ + d)k f (τ ) (4.1.11)

and is a polynomial in τ2−1 whose coefficients fn (q) are holomorphic functions of q = e2πiτ ,
N
X
f (τ ) = fn (q)τ2−n (4.1.12)
n=0

By this definition, every modular form of weight k is an almost-holomorphic modular form of

weight k, and E∗2 (τ ) is an almost-holomorphic modular form of weight 2. Clearly, any linear
combination of almost-holomorphic modular forms of weight k is an almost-holomorphic
modular form of weight k, and the product of two almost-holomorphic modular forms of
weights k and ` is an almost-holomorphic modular form of weight k + `, so that the space
of almost-holomorphic forms a ring graded by the weight.
A function g : H → C is a quasi-modular form of weight k if it is the holomorphic part
f0 (q) of some almost-holomorphic form f of weight k, as defined above. By this definition,
every modular form of weight k is a quasi-modular form of weight k, and E2 (τ ) is a quasi-
modular form of weight 2. The space of quasi-modular forms is a ring graded by the weight.
More precisely, we have the following theorem, presented here without proof.

Theorem 4.1 The space of almost-holomorphic modular forms is a polynomial ring, graded
by the weight k, and generated by E∗2 , E4 , and E6 . The space of quasi-modular forms is a
polynomial ring, graded by the weight k, and generated by E2 , E4 , and E6 .

4.1.3 Differential equations for quasi-modular forms

The inhomogeneous nature of the transformation law of E2 (τ ) under SL(2, Z) suggests that
E2 (τ ) may be viewed geometrically as a connection in the space of modular forms M. The

67
connection E2 produces a covariant derivative Dk acting on Mk ,
1 d k
Dk : Mk → Mk+2 Dk = − E2 (τ ) (4.1.13)
2πi dτ 12
To show this, we use the fact that a modular form f of weight k, namely f ∈ Mk , and its
τ -derivative f 0 satisfy the following transformation rules,

f (γτ ) = (cτ + d)k f (τ )

f 0 (γτ ) = (cτ + d)k+2 f 0 (τ ) + kc(cτ + d)k+1 f (τ ) (4.1.14)

Combining (4.1.3) with the above relation, and the definition of Dk , we obtain,

Dk f (γτ ) = (cτ + d)k+2 Dk f (τ ) (4.1.15)

so that Dk f is a modular form of weight k + 2, as announced. The Leibnitz formula for the
product of two modular forms f1 and f2 of respective weights k1 and k2 is given as follows,

Dk1 +k2 (f1 f2 ) = (Dk1 f1 ) f2 + f1 (Dk2 f2 ) (4.1.16)

When the space Mk+2 is one-dimensional, one easily derives a differential equation for mod-
ular forms of weight k. The precise normalizations may be deduced from the asymptotic
behavior at the cusp. For example, we have,

D4 E4 = − 31 E6 D8 E8 = − 23 E10 D12 ∆ = 0
D6 E6 = − 12 E8 D12 E12 = −E14 (4.1.17)

Equivalent expressions may be obtained by using E8 = E24 , E10 = E4 E6 , and E14 = E24 E6
together with Leibnitz’s rule. The derivative of E10 is, however, more involved since the
space M12 is two-dimensional.
Given that E2 may naturally be interpreted as a connection in Mk there should be a
natural notion of curvature. The associated curvature should transform tensorially, namely
it should be a modular form, and should be obtained from the derivative and the square of
the connection. Given the transformation law (4.1.3) for E2 , we obtain the transformations
for its derivative with respect to τ and its square,
12c2
E02 (τ 0 ) = (cτ + d)4 E02 (τ ) + 2c(cτ + d)3 E2 (τ ) + 2πi
(cτ + d)2
12c 2
E2 (τ 0 )2 = (cτ + d)4 E2 (τ )2 + 24
d)3 E2 (τ ) + 2πi + d)2


2πi
c(cτ + (cτ (4.1.18)

As a result, the following combination transforms as a modular form of weight 4,

 
1 0 0 1 0 2 4 1 0 1 2
E (τ ) − E2 (τ ) = (cτ + d) E (τ ) − E2 (τ ) (4.1.19)
2πi 2 12 2πi 2 12

68
Since this combination is both holomorphic in τ and a modular form of weight 4, it must
belong to M4 . But by Theorem 3.3, the space M4 is dimension 1 and is generated by E4 .
Matching the value at q = 0, we obtain the relation,
1 d 1 1
E2 (τ ) − E2 (τ )2 = − E4 (τ ) (4.1.20)
2πi dτ 12 12
It is natural to interpret E4 as the curvature, given indeed by a linear combination of the
derivative and the square of E2 . We note that the coefficient multiplying the square is half
of what one would have expected if the derivative had been applied to a modular form of
weight 2. This is as usual when obtaining the curvature from the connection.

4.2 Non-holomorphic Eisenstein series

We next introduce non-holomorphic Eisenstein series, which may be viewed as generalizing
the Riemann ζ-function, modular forms, and almost-holomorphic modular forms. Given a
lattice Λ = Z + τ Z, a non-holomorphic Eisenstein series may be defined by the following
lattice sum,8 or so-called Kronecker-Eisenstein sum,
0
X τ2s X τ2s
Es (τ ) = = (4.2.1)
ω∈Λ0
π s |ω|2s m,n∈Z π s |m + nτ |2s

where τ2 = Im (τ ) and the prime on the sum indicates that the contribution m = n = 0 is
to be omitted from the sum. The sum is absolutely convergent for Re (s) > 1. The lattice
sum definition of Es (τ ) may be used to show that Es is modular invariant,
 
a b
Es (γτ ) = Es (τ ) γ= ∈ SL(2, Z) (4.2.2)
c d
It may also be used to factor the non-trivial common divisors of m and n in the double sum,
0
X τ2s
Es (τ ) = ζ(2s) (4.2.3)
c,d∈Z
π s |cτ + d|2s
gcd(c,d)=1

where ζ(2s) is the Riemann ζ-function and the remaining sum is over pairs (c, d) ∈ Z2 that
have no common divisors strictly greater than 1. This expression in turn allows us to write
Es (τ ) as a Poincaré series over the coset Γ∞ \SL(2, Z) as follows,
ζ(2s) X
Es (τ ) = (Im γτ )s (4.2.4)
πs
γ∈Γ∞ \SL(2,Z)

8
The notation Es (τ ) used for the non-holomorphic Eisenstein series should not be confused with the
notation Es (τ ) used for the holomorphic Eisenstein series.

69
This relation implies that Es (τ ) vanishes for all τ whenever 2s is a zero of the Riemann
ζ-function, including at its zeros on the negative real axis, as well as at its non-trivial zeros
when Re (4s) = 1. Finally, the non-holomorphic Eisenstein series Es (τ ) is an eigenfunction
of the Laplace equation,

∆Es (τ ) = s(s − 1)Es (τ ) ∆ = 4τ22 ∂τ ∂τ̄ = τ22 (∂τ21 + ∂τ22 ) (4.2.5)

as may be verified by applying the Laplace operator term-wise in the sum of (4.2.1).

4.2.1 Analytic continuation of Es (τ )

While Es (τ ) is defined by the series (4.2.1) for Re (s) > 1, it may be analytically continued
in s throughout the complex s-plane to produce a meromorphic function, just as was possible
for the Riemann ζ-function. To exhibit the analytic structure of Es (τ ) in s, we use the
following integral representation for the summands,
Z ∞ !
dt s X 2
Γ(s)Es (τ ) = t e−πt|m+nτ | /τ2 − 1 (4.2.6)
0 t m,n∈Z

The subtraction of 1 eliminates the zero mode. The non-analyticity in s arises from the
small t regime. As we did in the case of the Riemann ζ-function, we split the integral at
t = 1 and Poisson re-sum the integrand for 0 < t < 1 in both m and n to obtain,
Z ∞ !
dt s X 2
Γ(s)Es (τ ) = t e−πt|m+nτ | /τ2 − 1
1 t m,n∈Z
Z 1 !
dt s 1 X −π|m+nτ |2 /(tτ2 )
+ t e −1 (4.2.7)
0 t t m,n∈Z

Next, we change variable in the second integral t → 1/t, collect the resulting integrals for
t ∈ [1, ∞[, and analytically continue a trivial remaining integral. The result is as follows,
Z ∞  !
1 dt s  X 2
Γ(s)Es (τ ) = − + t + t1−s e−πt|m+nτ | /τ2 − 1 (4.2.8)
s(1 − s) 1 t m,n∈Z

The first term on the right side is meromorphic in s ∈ C, while the integral is manifestly
holomorphic throughout C. As a result, Es (τ ) admits a meromorphic analytic continuation
to s ∈ C with a simple pole at s = 1 with residue 1. Moreover, the expression is manifestly
invariant under s → 1 − s which implies the functional equation,

Γ(s)Es (τ ) = Γ(1 − s)E1−s (τ ) (4.2.9)

70
In particular, the functional relation requires Es (τ ) = 0 whenever s = −1, −2, · · · in view of
the fact that the right side is finite for these values and that Γ(s) on the left side is infinite.
This observation reproduces our earlier observation that Es (τ ) = 0 whenever 2s is a zero of
the Riemann ζ-function.

4.2.2 Fourier series of Es (τ )

The Fourier series of Es (τ ) for all Re (s) > 1 may be derived from the integral representation
in (4.2.6), this time by Poisson re-summation in the variable m only, using the formula,
r
X
−πt|m+nτ |2 /τ2 τ2 X 2πimnτ1 −πm2 τ2 /t−πn2 tτ2
e = e e (4.2.10)
m,n∈Z
t m,n∈Z

In the mathematics literature this procedure is referred to as the Chowla-Selberg method.

Substituting this result into (4.2.6) and splitting the sum over m, n into contributions for
mn 6= 0 and mn = 0, we have

Γ(s)Es (τ ) = As (τ ) + Bs (τ ) (4.2.11)

where,
∞
√ X X 2πimnτ1
Z
dt s− 1 −πτ2 (m2 /t+n2 t)
As (τ ) = τ2 e t 2e
m6=0 n6=0 0 t
∞ r
Z ∞ r !
dt s X τ2 −πn2 tτ2 X τ2 −πm2 τ2 /t
Bs (τ ) = t −1 + 2 e + e (4.2.12)
0 t n=1
t m∈Z
t

The integral As (τ ) is manifestly holomorphic in s throughout C, and may be computed by

performing the t-integral in terms of a K-Bessel function,

√ X  m s− 21 2πimnτ1
As (τ ) = 2 τ2 e Ks− 1 (2πτ2 |mn|) (4.2.13)
n
  2
m,n6=0

The integral Bs (τ ) may be computed by Poisson re-summing the m-sum in the integrand,
X Z ∞ dt  τ2
r 
s −πn2 tτ2 −πn2 t/τ2
Bs (τ ) = t e +e (4.2.14)
n6=0 0 t t

which is readily evaluated in terms of the Riemann ζ-function or associated ξ-function,

1
Bs (τ ) = 2Γ(s)π −s τ2s ζ(2s) + 2Γ(s − 12 )π 2 −s τ21−s ζ(2s − 1)
= τ2s ξ(2s) + τ21−s ξ(2s − 1) (4.2.15)

71
where ξ(s) is the function computed in (2.1.16). We may now verify the functional relation
(4.2.9) directly in terms of the transformation properties of As (τ ) and Bs (τ ). For As (τ ) this
is manifest upon using the symmetry under interchange of m and n, while for Bs (τ ) it is a
consequence of the relation ξ(1 − s) = ξ(s). Assembling both contributions, and recasting
As (τ ) in terms of a summation over a single variable N = mn, we obtain the celebrated
Fourier series decomposition for Es (τ ),
1
Γ(s)Es (τ ) = 2Γ(s)π −s ζ(2s) τ2s + 2Γ(s − 12 )π 2 −s ζ(2s − 1) τ21−s
√ X 1
+4 τ2 |N | 2 −s σ2s−1 (|N |) e2πiN τ1 Ks− 1 (2πτ2 |N |) (4.2.16)
2
N 6=0

where σα (n) is defined to be the sum σα (n) = d|n dα of positive divisors d of n. One verifies
P

that each Fourier mode, including the power-behaved terms on the first line, individually
satisfy the Laplace eigenvalue equation, as required by the fact that ∆ commutes with the
translation operator ∂τ1 .
For integer s = k the Bessel function in the Fourier series becomes a spherical Bessel
function Kk− 1 (x), which is a combination of an exponential in x and a polynomial in 1/x.
2
As a result, the Fourier series of Ek (τ ) simplifies and may be recast as follows,
B2k k 4 (2k − 3)! ζ(2k − 1)
Ek (τ ) = (−)k+1 y +
(2k)! (k − 2)! (k − 1)! y k−1
∞
2 X
N k−1 σ1−2k (N )Pk (N y) q N + q̄ N


+ (4.2.17)
(k − 1)! N =1

where q = e2πiτ and y = 4πτ2 . Furthermore, B2k are the Bernoulli numbers and Pk (x) is a
polynomial in 1/x given by,
k−1
X (k + m − 1)!
Pk (x) = (4.2.18)
m=0
m! (k − m − 1)! xm

Non-holomorphic Eisenstein series with integer s will play a crucial role in the theory of
modular graph functions and forms, to be developed in section 9.

4.3 Maass forms

The non-holomorphic Eisenstein series discussed in the previous subsection are special ex-
amples of a more general class of non-holomorphic objects known as Maass forms. A non-
holomorphic function f (τ ) is called a Maass form if it is SL(2, Z)-invariant, satisfies a Laplace
eigenvalue equation,

∆f (τ ) = λf (τ ) ∆ = 4τ22 ∂τ ∂τ̄ = τ22 (∂τ21 + ∂τ22 ) (4.3.1)

72
for λ ∈ C, and has at most polynomial growth at the cusps. Without loss of generality,
we will write λ = s(s − 1) and denote the space of such Maass forms by N (s), where
N (s) = N (1 − s). From (4.2.5) we see that the non-holomorphic Eisenstein series Es (τ )
are elements of N (s). Indeed, below we will see that for generic s ∈ C the non-holomorphic
Eisenstein series are the only elements of N (s).
The Fourier expansions of generic Maass forms admit surprisingly simple closed form ex-
pressions. In general, non-holomorphic modular forms admit Fourier expansions in separate
powers of q and q̄, of the type,
X
f (τ ) = cM,N (τ2 )q M q̄ N
e (4.3.2)
M,N ∈Z

with e cM,N (τ2 ) being functions of τ2 but not of τ1 , since the latter would violate invariance
under T : τ → τ + 1. It is often useful to absorb all τ2 dependence, including that of q and
q̄, into the Fourier coefficients, giving an alternative form for the Fourier expansion as,
X
f (τ ) = cN (τ2 ) e2πiN τ1 (4.3.3)
N ∈Z

We will now see that, for Maass forms, the coefficients cN (τ2 ) can be fixed uniquely, up to
overall constants. Indeed, inverting the Fourier expansion gives,
Z 1
cN (τ2 ) = dτ1 e−2πiN τ1 f (τ ) (4.3.4)
0

and we may make use of the eigenvalue equation to obtain,

Z 1
1
cN (τ2 ) = dτ1 e−2πiN τ1 ∆f (τ )
s(s − 1) 0
τ22 h 2 2
i
= (2πiN ) cN (τ2 ) + ∂τ2 cN (τ2 ) (4.3.5)
s(s − 1)
This gives a second order differential equation for the coefficients cN (τ2 ),
 
2 2 2
τ2 ∂τ2 cN (τ2 ) = s(s − 1) + (2πN τ2 ) cN (τ2 ) (4.3.6)

For N = 0 the general solution is of the form,

c0 (τ2 ) = a τ2s + b τ21−s (4.3.7)

For N 6= 0 we obtain a modified Bessel equation with general solutions Is−1/2 and Ks−1/2 .
Discarding the solutions Is−1/2 since they grow exponentially at the cusp, we obtain,
√
cN (τ2 ) = αN τ2 Ks−1/2 (2πτ2 |N |) N 6= 0 (4.3.8)

73
We conclude that a general element of N (s) has a Fourier expansion of the form,
X √
f (τ ) = a τ2s + b τ21−s + αN τ2 e2πiN τ1 Ks−1/2 (2πτ2 |N |) (4.3.9)
N 6=0

Note that the Fourier expansions of the non-holomorphic Eisenstein series were obtained
already in (4.2.16), and match with the above result upon substituting,
1 1
2ζ(2s) 2Γ(s − 12 )π 2 −s ζ(2s − 1) 4|N | 2 −s σ2s−1 (|N |)
a= b= αN = (4.3.10)
πs Γ(s) Γ(s)

The constant part of the Fourier expansion, i.e. the term a τ2s + b τ21−s , is known as the
Laurent polynomial of the Maass form. Though it may at first sight seem that the space
of Laurent polynomials is two-dimensional and spanned by a and b, in fact this space is
only one-dimensional. This may be shown by considering Maass forms on the truncated
fundamental domain,
 
1
FL = τ ∈ H, |τ | ≥ 1, |Re(τ )| ≤ , τ2 ≤ L (4.3.11)
2
Indeed, consider two Maass forms f, g ∈ N (s) with Fourier expansions
X X
f (τ ) = cN (τ2 )e2πiN τ1 g(τ ) = dN (τ2 )e2πiN τ1 (4.3.12)
N ∈Z N ∈Z

∂
Green’s theorem on the region FL states that if ∂n is the normal derivative on ∂FL and ds
is the differential arc length, then we have
Z   Z
∂g ∂f dτ1 dτ2
ds f −d = (f ∆g − g∆f ) = 0 (4.3.13)
∂FL ∂n ∂n FL τ22
where the last equality follows from the eigenvalue equation. The left side may be evaluated
in terms of the Fourier coefficients. Noticing that modular invariance reduces the boundary
integral to an integral over τ2 = L (since the remaining boundaries of the fundamental
domain give equal and opposite contributions), we have
Z   Z 1
∂g ∂f 2 X
ds f −f = dτ1 (cN (L)d0M (L) − c0N (L)dM (L)) e2πi(N +M )τ1
∂FL ∂n ∂n − 12 N,M ∈Z
X
cN (L)d0−N (L) − c0N (L)d−N (L)

= (4.3.14)
N ∈Z

where the derivative is defined by,

∂cN (τ2 ) 
c0N (L) = (4.3.15)
∂τ2 τ2 =L


74
and similarly for d0−N (L). In particular, this implies that c0 d00 − c00 d0 = 0, and if we write,

c0 (τ2 ) = a τ2s + c τ21−s d0 (τ2 ) = b τ2s + d τ21−s (4.3.16)

then for s 6= 21 we obtain ad − bc = 0, which indeed means that the space of Laurent
polynomials has dimension one. In the case that s = 12 this space is also obviously of
dimension one, since then the two factors of τ2 have the same power.

4.3.1 The space of Maass forms

As was mentioned above, at generic values of s ∈ C the space of Maass forms N (s) consists
only of non-holomorphic Eisenstein series Es (τ ). More generally, at non-generic values of s,
the space of Maass forms can include cusp forms. In the current context, cusp forms are
defined to be the subset of elements of N (s) such that the Laurent polynomial vanishes,
that is to say a = b = 0 in (4.3.9). We will denote the space of cusp forms by S(s). The
non-holomorphic Eisenstein series are clearly not cusp forms for any s. For generic values
of s, the set S(s) will be empty.
The structure of N (s) can be summarized as follows:

Theorem 4.2 The space of Maass forms N (s) is given by


 CEs (τ ) Re s > 21 and s ∈/ [0, 1]
N (s) = CEs (τ ) ⊕ S(s) Re s = 12 or s ∈ [0, 1) (4.3.17)
s=1

C

The proof of this theorem follows almost immediately from the fact, proven above, that
the space of Laurent polynomials has dimension 1. This means that for any f (τ ) ∈ N (s),
there exists a constant c such that the combination f (τ ) − cE(τ, s) has trivial Laurent
polynomial, i.e. is a cusp form.
To complete the proof, we need only prove that the space of cusp forms is empty for all
values of s which satisfy neither Re s = 21 nor s ∈ [0, 1). The non-existence for s = 1 is
immediate: in this case the cusp form would be harmonic, i.e. satisfy a Laplace eigenvalue
equation with zero eigenvalue, and hence would have to vanish by the maximum modulus
principle. More generally we may argue as follows. Above we have seen that it is possible
to find a constant c such that for any f (τ ) ∈ N (s), the combination h(τ ) := f (τ ) − cEs (τ )
is a cusp form. Since cusp forms are square-integrable over the fundamental domain F =
SL(2, Z)\H equipped with the Poincaré measure d2 τ /τ22 ,

d2 τ
Z
2
|h(τ )|2 < ∞ (4.3.18)
F τ2

75
we may manipulate the following absolutely convergent integral by using the differential
equation ∆h = s(s − 1)h satisfied by h, and Green’s theorem,
Z 2 Z 2 Z
dτ dτ
s(s − 1) 2
2
|h(τ )| = 2
∆h(τ ) h(τ ) = −4 d2 τ |∂τ h|2 (4.3.19)
F τ2 F τ2 F

Since the right side is real and non-positive, cusp forms must have real s(s − 1) ≤ 0, which
restricts us to s ∈ [0, 1] or Re s = 21 .
In closing, let us note that the triviality of the space of cusp forms for Re s 6= 21 and s 6∈ R
implies that for these values the space N (s) is one-dimensional and hence there must be a
functional relation relating Es (τ ) and E1−s (τ ), since both have Laplace eigenvalue s(s − 1).
Indeed, precisely such a relation was identified in (4.2.9).

4.3.2 Cusp forms

We have just shown that the space S(s) of cusp forms is empty unless Re s = 21 or s ∈ [0, 1).
Actually, one can further show that S(s) vanishes for s ∈ [0, 1), and hence that all non-trivial
cusp forms must exist at Re s = 12 and away from the real axis: we refer to [18] for the proof.
Denoting s = 21 + it for t ∈ R, the corresponding eigenvalue is λ = s(s − 1) = − 41 + t2 .

For any t ∈ R, the space S 12 + it
is finite-dimensional. To show this, begin with M

elements v1 (τ ), . . . , vM (τ ) ∈ S 21 + it . Given an additional element ve(τ ), it is possible to

choose coefficients ci ∈ C such that the linear combination,
M
X
v(τ ) := ve(τ ) − ci vi (τ ) (4.3.20)
i=1

has vanishing Fourier coefficients up to order M . We will now show that for M sufficiently
large but finite, the vanishing of Fourier coefficients up to order M implies vanishing to all

orders. In other words we will have v(τ ) = 0 exactly, and hence all elements ve(τ ) ∈ S 21 + it
can be written as a linear combination of a finite number M of other elements.
To prove the desired result, begin by recalling that a cusp form v(τ ) is bounded on
the upper-half plane, and hence that |v(τ )| must have a maximum at some location τ (0) =
(0) (0)
τ1 +iτ2 . From the form of the Fourier expansion (4.3.9) and orthogonality of exponentials,
we have,
Z 1/2 √ (0)
!
|N |)
q
(0) (0) (0) 2K it (2πτ
αN τ2 Kit (2πτ2 |N |) = dτ1 v(τ1 + iτ2 /2) e−2πiN τ1 (0)
2
(4.3.21)
−1/2 Kit (πτ2 |N |)
It then follows that,
 (0)  √   X Kit (2πτ2(0) N )
v(τ ) < 2 2 v(τ (0) ) (4.3.22)
(0)
|N |>M Kit (πτ2 N )

76
If M is sufficiently large (depending on t), we may estimate the Bessel functions by expo-
nentials, giving the inequality,
(0)
v(τ ) ≤ v(τ ) Ce−πM τ2
 (0)   (0) 
(4.3.23)

(0)
for some proportionality constant C. Hence if M ≥ (ln C)/(πτ2 ) then we conclude that
|v(τ (0) )| = 0, and hence that v(τ ) is identically zero.
We have just shown that S 12 + it is finite-dimensional for any t. Furthermore, though

we will not reproduce the proof here, it has been shown that there exist an infinite number
of values of t for which the space of cusp forms is non-empty. However, there is no exact
value of t for which this is known to be the case. The smallest approximate values of t for
which cusp forms are believed to exist are,

t ≈ 9.53369, 12.17300, 13.77975, 14.35850, 16.13807, 16.64425, 17.73856, . . . (4.3.24)

In particular, we see that there is only a discretum of such values.

4.3.3 Comments on the physical interpretation of the spectrum

The fundamental domain for SL(2, Z) is non-compact as it has a puncture at the cusp. From
the point of view of scattering theory, one may view the cusp as an asymptotic region from
which a free wave in the continuous spectrum can be sent in. This wave is indeed oscillatory
1
+it
and of the form τ22 for real t, from the first constant term in the Fourier expansion. Since
the fundamental domain is otherwise bounded, the wave hits a wall at |τ | = 1 and must be
1
−it
reflected out through the puncture and is asymptotically of the form τ22 . This phenomenon
is analogous to one-dimensional scattering on a semi-infinite line and explains in physical
terms why the Eisenstein solutions have E 1 +it (τ ) = E 1 −it (τ ).
2 2

The cusp forms are wave functions of bound states, as is confirmed by the fact that they
decay exponentially towards the cusp. In most quantum mechanics problems, the number
of bound states in the presence of a wall would be finite. But there are exceptions, such
as the Coulomb problem. Here, it is the special properties of the hyperbolic metric on the
fundamental domain of SL(2, Z) that produce an infinite number of cusp forms, or bound
states, with a spectrum that rises all the way to infinity.

4.4 Spectral decomposition

We have now discussed various aspects of the spectrum of the Laplacian on the fundamental
domain SL(2, Z)\H. It is possible to extract from this data the following orthonormal basis
of eigenfunctions of the Laplacian:

77
 • Discrete series: The constant function v0 := Area(F ) = 3/π (which is trivially
p p

an eigenfunction of the Laplacian) together with an orthonormal basis of cusp forms

1 2


vn (τ ) with corresponding eigenvalues sn (sn − 1) = − 4 + tn for tn ∈ R. The first few
values of tn are given in (4.3.24).

• Continuous series: The non-holomorphic Eisenstein series Es (τ ) with s = 12 + it and

t ∈ R taking a continuum of values.

Orthogonality here is defined relative to the Petersson inner product with the Poincaré
measure d2 τ /τ22 on the fundamental domain, i.e.
Z 2
dτ
hf, gi = 2
f (τ )g(τ ) (4.4.1)
F τ2

Let us use this inner product to verify the orthogonality claimed above. Consider the inner
product between a non-holomorphic Eisenstein series Es (τ ) and a cusp form vn (τ ),
Z 2
dτ
hEs (τ ), vn (τ )i = 2
Es (τ )vn (τ ) (4.4.2)
F τ2

This inner product converges since cusp forms have exponential decay at the cusps, whereas
non-holomoprhic Eisenstein series have at most polynomial growth. For Re s > 1, we may
proceed by using the Poincaré series for Es (τ ) given in (4.2.4), which gives
Z 2
ζ(2s) dτ X
hEs (τ ), vn (τ )i = (Im γτ )s vn (τ )
π s F τ22
γ∈Γ∞ \SL(2,Z)
Z 1 Z ∞
ζ(2s) 2 dτ2 s
= dτ1 τ vn (τ ) (4.4.3)
π s
− 12 0 τ22 2

where in the second equality we have used modular invariance to unfold the integral over
the fundamental domain, i.e. we have replaced the integral over SL(2, Z) orbits in F with
an integral over the strip Γ∞ \H,
Z X Z
· = · (4.4.4)
SL(2,Z)\H Γ∞ \SL(2,Z) Γ∞ \H

This is the analog of the unfolding trick introduced for periodic functions in section 2.1.1.
The integral over τ1 may now be performed and isolates the Laurent polynomial of vn (τ ),
which is vanishing by definition. This gives the desired orthogonality for Re s > 1. The
analogous result for Re s < 1 may be obtained by analytic continuation.
An arbitrary square integrable modular function can now be decomposed into this or-
thogonal basis. We state without proof the following spectral decomposition theorem,

78
 p
Theorem 4.3 (Roelcke-Selberg) Let v0 = 3/π be the normalized constant function
and {vn (τ )}n≥1 be an orthonormal basis of cusps forms. Then any square integrable modular
function f (τ ) admits the following decomposition,
Z ∞  
X 1
f (τ ) = hf (τ ), vn (τ )i vn (τ ) + dt f (τ ), E 1 +it (τ ) E 1 +it (τ ) (4.4.5)
n≥0
4π −∞ 2 2

The reader may be worried about the convergence properties of the continuous portion of
this decomposition. Indeed, Es (τ ) is not square-integrable because of its Laurent polynomial.
√
However, when Re s = 21 , one can show that |Es (τ )|  τ2 when τ2 → ∞, and the extra
integration over t in the decomposition formula saves a factor of log τ2 , which is sufficient to
turn the right side into a square-integrable function.

4.5 Maass forms of general weight

Thus far we have focused on Maass forms of weight (0, 0). We now discuss Maass forms of
arbitrary (integer) weight k, i.e. those satisfying
   
aτ + b k a b
f = (cτ + d) f (τ ) ∈ SL(2, Z) (4.5.1)
cτ + d c d
The Maass property in this case must be recast in terms of a weight-k Laplacian

∆k = 4τ22−k ∂τ τ2k ∂τ̄ = ∆ − 2ikτ2 ∂τ̄ (4.5.2)

in order to get a modular covariant object.

Of particular interest are the harmonic Maass forms of weight k, i.e. those satisfying an
eigenvalue equation ∆k f = 0. Unlike in the case of weight 0, for non-zero weight the space
of harmonic Maass forms is non-trivial. A simple example of a weight-2 harmonic Maass
form is the almost-holomorphic Eisenstein series E∗2 (τ ) introduced in section 4.1. Indeed,
from (4.1.10) and the fact that E2 (τ ) is holomorphic we have
3 3
∆2 E∗2 (τ ) = − (∆ − 4iτ2 ∂τ̄ )τ2−1 = − (2τ2−1 − 2τ2−1 ) = 0 (4.5.3)
π π
The space of weight-k harmonic Maass forms Ik admits an interesting decomposition
which will feature prominently in the subsequent section. This decomposition can be cap-
tured by the following exact sequence,

0 → M!k → Ik → M!2−k → 0 (4.5.4)

with M!k the space of weight-k modular functions. The content of this exact sequence is
two-fold. First, it says that the space M!k of weight-k modular functions is a subset of

79
the space Ik of weight-k harmonic Maass forms. This follows trivially from the fact that
modular functions are annihilated by ∆k . Second, it says that the space M!2−k of weight-
(2 − k) modular functions is a quotient of Ik by M!k . This follows from the fact that, given
an element f (τ ) ∈ Ik , the function g(τ ) := τ2k ∂τ f (τ ) is holomorphic and of weight 2 − k,
and vanishes if and only if f (τ ) is in M!k .

4.6 Mock modular forms

In section 4.1 we saw that there exist two variants of the weight-two holomorphic Eisenstein
series, namely the function E2 (τ ) defined in (4.1.2) which is holomorphic but not modular,
and the function E∗2 (τ ) defined in (4.1.8) which is modular but not holomorphic. The former
is an example of a quasi-modular form, whereas the latter is an example of an almost-
holomorphic modular form. Section 4.1.2 described an isomorphism between the sets of
quasi-modular forms and almost-holomorphic modular forms.
Having introduced non-holomorphic modular forms in the previous subsections, it is
natural to ask if their holomorphic parts lead to interesting analogs of quasi-modular forms.
In particular, we will consider the holomorphic parts of harmonic “weak” Maass forms,9
which are known as mock modular forms.10 Given a harmonic Maass form F ∗ (τ ), we may
extract its holomorphic part as follows. We begin by defining,
∂ ∗
g(τ ) = τ2k F (τ ) (4.6.1)
∂τ
By the condition that F ∗ (τ ) be harmonic, it follows that g(τ ) is holomorphic, since ∆k =
τ22−k ∂τ τ2k ∂τ . If one can identify a new function g ∗ (τ ) such that its image under τ2k ∂τ is g(τ ),
then it is clear that,

F (τ ) := F ∗ (τ ) − g ∗ (τ ) (4.6.2)

will be holomorphic. Such a function g ∗ (τ ) can be obtained by simply inverting the differ-
ential operator τ2k ∂τ , giving the expression,
 k−1 Z i∞
∗ i g(−z)
g (τ ) = dz (4.6.3)
2 −τ (z + τ )k

9
The adjective “weak” indicates the potential for exponential growth at the cusps; in contrast, a “Maass
form” is required to have at most polynomial growth at the cusp.
10
Mock modular forms made their first appearance in a letter of Ramanujan to Hardy, in which they
were referred
P to as mock theta functions. In the modern definition, a mock theta function is a q-series
H(q) = n≥0 an q n such that for some λ ∈ Q the series q λ H(q) gives a mock modular form of weight 21 ,
2
with the shadow (defined below) being a weight- 32 function of the form n∈Z (n)nq |κ|n with κ ∈ Q and 
P
an odd periodic function.

80
known as a non-holomorphic Eichler integral. This integral is independent of the path since
the integrand is holomorphic in z.
The function F (τ ) is holomorphic, but will in general fail to be modular, since g ∗ (τ )
is not modular. The function g(τ ) in terms of which F (τ ) is defined is referred to as the
shadow, and is a weight-(2 − k) modular function. The pair of F (τ ) and g(τ ) are often
together referred to as a mock modular form, though in some contexts the function F (τ )
alone is given that name.
Given a harmonic Maass form F ∗ (τ ), it is always possible to extract from it a unique
mock modular form by taking the holomorphic part in the way described above. Conversely,
given a mock modular form, it is always possible to complete it to a harmonic Maass form by
taking the sum of F (τ ) with the function g ∗ (τ ) obtained from the shadow. This establishes
an isomorphism between the two spaces, exactly analogous to the isomorphism between the
spaces of almost-holomorphic modular forms and quasi-modular forms given in section 4.1.
The space of mock modular forms then fits into an exact sequence of the form in (4.5.4). By
nature of it being an exact sequence, the map from the space of mock modular forms to the
space of shadows (known as the shadow map) is surjective.

4.6.1 Examples of mock modular forms

We now give some concrete examples of mock modular forms.
A first example is provided by E2 (τ ). In the previous section we showed that E∗2 (τ ) is
a weight-2 harmonic Maass form, and hence by definition the holomorphic part of E∗2 (τ ),
namely E2 (τ ), is a weight-2 mock modular form. The shadow g(τ ) in this case is simply a
constant, namely g(τ ) = − 12 π
. It is easy to check that this constant gives g ∗ (τ ) = − πτ32 ,
reproducing the result in (4.1.10). Thus we see that E2 (τ ) is both quasi-modular and mock
modular. This is not in general the case: quasi-modular forms are generically not mock
modular. For example, (E2 (τ ))2 is a quasi-modular form, but fails to be mock modular since
its modular completion is not a harmonic Maass form.
A second example, which appears rather unexpectedly in physics via the elliptic genus
of strings propagating on a K3 manifold, is the following function,
Z ∞
∗ 6 η(−z)3
F (τ ) = F (τ ) + √ dz (4.6.4)
i −τ (z + τ )1/2
with F (τ ) defined as,
 
1 X
r rs/2
F (τ ) = − E2 (τ ) + 24 (−1) s q (4.6.5)
η(τ )3 r>s>0
r−s=1 mod 2

It is known that F ∗ (τ ) is modular with weight 21 . Hence its holomorphic part F (τ ) must be
√
a mock modular function. The shadow in this case is g(τ ) = 3 2 η(τ )3 .

81
 A curious fact is that the first few terms in the Fourier expansion of F (τ ),
1
F (τ ) = q − 8 −1 + 45q + 231q 2 + 770q 3 + 2277q 4 + 5796q 5 + . . .


(4.6.6)
give dimensions of irreducible representations of the Mathieu group M24 , a sporadic simple
group of order 244,823,040. This correspondence goes under the name of Mathieu moonshine,
and has been the subject of much research.
As a third and final example, we mention with a classic result of Zagier, which states
that the following two functions are harmonic with k = 23 ,
−1/2 2
X X
F0∗ (τ ) = 3 H(4n) q n + 6τ2 β(4πn2 τ2 ) q −n
n≥0 n∈Z
1 2
n− 41
β 4π(n + 1/2)2 τ2 q −(n+ 2 )
−1/2
X X
F1∗ (τ ) = 3


H(4n − 1) q + 6τ2 (4.6.7)
n>0 n∈Z

where β(t) is an incomplete gamma function

Z ∞
1
β(t) = du u−3/2 e−ut (4.6.8)
16π 1
and H(n) are the Hurwitz class number of imaginary quadratic fields. The latter are defined
for n > 0 to be the number of P SL(2, Z) equivalence classes of integral binary quadratic
forms of discriminant −n, weighted by the reciprocal of the number of their automorphisms.
The first few non-zero entries are given by
1 1
H(3) = H(4) = H(7) = H(8) = H(11) = 1 (4.6.9)
3 2
1
For n = 0 we have H(0) = − 12 and for n < 0 we have H(n) = 0.
∗
The holomorphic parts F0,1 (τ ) of F0,1 (τ ) are easy to read off,
X
F0 (τ ) = 3 H(4n)q n
n≥0
1
X
F1 (τ ) = 3 H(4n − 1)q n− 4 (4.6.10)
n>0

and have corresponding shadows

3 3
g0 (τ ) = − ϑ3 (0|2τ ) g1 (τ ) = − ϑ2 (0|2τ ) (4.6.11)
4π 4π
It is important to note that these functions are not mock modular forms in the strict sense
∗
discussed above, since under τ → −1/τ the original harmonic functions F0,1 (τ ) are not
individually invariant, but rather transform into one another as
 ∗  ∗ 
(−iτ )3/2 1 1
 
F0 (−1/τ ) F0 (τ )
∗
=− √ (4.6.12)
F1 (−1/τ ) 2 1 −1 F1∗ (τ )

82
 ∗
There are two ways of interpreting this fact. The first is to think of F0,1 (τ ) as forming
a vector-valued harmonic Maass form. In this interpretation the functions F0,1 (τ ) likewise
form a vector-valued mock modular form. Vector-valued modular forms will be discussed in
more detail in section 8.
The second approach is to notice that F0∗ (τ ) is invariant under T and F1∗ (τ ) is invariant
under T 4 . Hence we can say that F0∗ (τ ) individually is modular invariant under the sub-
group of SL(2, Z) generated by T and ST 4 S. This subgroup is denoted by Γ0 (4), and is
one of the so-called congruence subgroups of SL(2, Z), discussed in detail in section 6. In
this interpretation one would say that F0∗ (τ ) is a harmonic Maass form for the congruence
subgroup Γ0 (4), and that F0 (τ ) is likewise a mock modular form for this subgroup. Modular
forms for congruence subgroups will be discussed in detail in section 7.

4.7 Quantum modular forms

Finally, let us briefly mention the notion of a quantum modular form.11 Unlike for standard
modular forms for which the domain is the upper half-plane, quantum modular forms are
defined on the space of cusps Q ∪ {i∞} (or some subset thereof). Since this domain is
a discrete space, quantum modular forms do not obey the usual analyticity properties of
modular forms. Furthermore, they are not modular in the standard sense. Instead, a weight-
k quantum modular form f (τ ) is defined such that for all γ ∈ SL(2, Z), the difference
 
k aτ + b
hγ (τ ) := f (τ ) − γ (cτ + d) f (4.7.1)
cτ + d
is an analytic function on the upper half-plane. Here we have allowed for a multiplier phase
γ , similar to that appearing in the transformation rules for the Dedekind η-function in
section 3.9.
It is simplest to illustrate this definition by proceeding directly to an example. Consider
the following q-expansion,
∞ n−1
X Y
F (q) = (1 − q i+1 ) (4.7.2)
n=0 i=0

known as Kontsevich’s strange function. This function does not converge on any open subset
of C, but is well-defined at roots of unity, i.e. when τ ∈ Q. We now define the closely related
object,
1
φ(τ ) = q 24 F (q) (4.7.3)

11
As far as we are aware, there is no direct connection between the term “quantum” used here and the
usual notion of “quantum” in Physics.

83
It follows from this definition that φ(τ + 1) = e2πi/24 φ(τ ).
To understand the behavior under S, we first note that the T transformation just given
implies φ(n) = e2πin/24 φ(0) = e2πin/24 for n ∈ Z. On the other hand, Kontsevich and Zagier
have shown that for τ = − n1 , one has,
  ∞  j
1 (3+n)
3/2 2πi 24
X 2πi
φ − ∼n e + cj (4.7.4)
n j=0
n

as n → ∞, with the first few coefficients in the expansion being,

23 1681
c0 = 1 c1 = c2 = (4.7.5)
24 1152
Using the T transformation, the first term above can be rewritten as,
2πi
n3/2 e 8 φ(n) = (−in)3/2 φ(n) (4.7.6)

and hence we conclude that the function

hS (n) := φ(n) + (−in)3/2 φ(−1/n) (4.7.7)

has a well-defined Taylor expansion at τ = 0, and so is smooth there. Indeed, the n-th
derivative of hS (n) at τ = 0 is given by cn (2πi/n)n . Though we do not give details here, the
function hS (n) can furthermore be extended to a real-analytic function on C (except for at
τ = 0), so φ(τ ) is an example of a quantum modular form.

84
• Bibliographical notes
Classic introductions to spectral analysis on symmetric spaces with applications to Maass
forms and non-holomorphic Eisenstein series may be found in the books by Kubota [19],
Terras [20, 21], and Iwaniec [22, 23]. Group-theoretic aspects of the spectral decomposition
of automorphic forms are expounded on in the book by Gel’fand, Graev, and Pyatetskii-
Shapiro [24]. Faddeev’s approach, which is based on operator perturbation theory, and per-
haps closer to physics, is reviewed in detail and commented upon in Lang [18]. A construction
of cusp forms which are not Maass forms was given in [25]. The spectral decomposition the-
orem has recently been used in a physics context in [26, 27]. Generalizations to automorphic
forms under the group GL(n, R) are developed in the book by Goldfeld [28].
A fundamental role was played by quasi-modular forms in Viazovska’s proof of the sphere
packing problem in eight dimensions, namely that no packing is more dense than the one
corresponding to the root lattice of E8 . The original paper may be found in [29], while the
generalization to sphere packing in dimension 24 was settled in [30].
Reviews on mock modular forms include the lecture notes by Zagier [31] in the math-
ematics literature and [32, 33] in the physics literature. The second example of a mock
modular form, given in (4.6.5), appears in physics as the elliptic genus of a K3 manifold
[34, 35]. The connection with dimensions of representations of the Monster group M24 was
observed in [36], beginning the program of Mathieu moonshine. Reviews of moonshine aimed
at physicists may be found in [33, 37, 38, 39] and references therein. The third example of
mock modular forms, given in (4.6.10), also arises in at least one physics context, as the
partition function of N = 4 super Yang-Mills theory with gauge group SO(3) on CP2 [40].
Quantum modular forms were introduced in mathematics in [41, 42, 43] but their role in
physics remains unexplored to date, despite the word “quantum” in their name.

85
5 Quantum fields on a torus
In this section, we shall present some immediate applications of the theory of elliptic and
modular functions to problems of physical interest, including the construction of the Green
functions and the functional determinants for the bc system, the scalar, and the spinor fields
on the torus. In particular, we shall show how the singular terms in the operator product
expansion of holomorphic fields for the bc system essentially determine arbitrary correlation
functions on the torus in terms of elliptic functions.

5.1 Quantum fields

A quantum field is a map from a manifold Σ to a linear operator acting on a vector space F,
or Fock space, which is equipped with a Hermitian inner product. If the inner product is
positive definite then F is a Hilbert space familar from standard quantum theory, but we shall
consider quantum field theories where the inner product is not necessarily definite as well.
We shall be interested in the special case where the manifold Σ is a (connected) Riemann
surface which may be compact or non-compact, with or without a boundary. Quantum fields
on the complex plane Σ = C describe the continuum limit and critical phenomena of two-
dimensional statistical mechanics systems, such as the Ising, Potts, XY, and non-linear sigma
models. Quantum fields on a general Riemann surface Σ provide the proper setting for string
perturbation theory, where probability amplitudes are obtained via an expansion in powers
of the string coupling constant. The expansion is obtained by summing contributions from
Riemann surfaces of all genera and, for each genus, by integrating over the corresponding
moduli space of Riemann surfaces, as will be explained in more detail in section 12.

5.1.1 Conformal fields

A conformal field φ on Σ may be defined as follows. We consider two arbitrary coordinate
charts U, U 0 ⊂ Σ with local complex coordinate systems (z, z̄) and (z 0 , z̄ 0 ). Assuming that
these coordinate charts have a non-empty intersection U ∩ U 0 6= ∅, the coordinate systems in
the overlap U ∩ U 0 are related to one another by a holomorphic transition function z 0 = f (z)
whose derivative f 0 (z) is non-zero throughout U ∩ U 0 . Any such transformation of the form
z → z 0 = f (z) is a locally conformal transformation, namely it preserves all angles. For
more details see appendix B. In the coordinate systems (z, z̄) and (z 0 , z̄ 0 ) the field φ is given
by φ(z, z̄) and φ0 (z 0 , z̄ 0 ), respectively. The field φ is a conformal (primary) field of conformal
weight (h, h̃), provided φ and φ0 are related as follows,
φ0 (z 0 , z̄ 0 )(dz 0 )h (dz̄ 0 )h̃ = φ(z, z̄)(dz)h (dz̄)h̃ (5.1.1)
or in view of the holomorphicity of f (z),
h̃
f 0 (z)h f 0 (z) φ0 (z 0 , z̄ 0 ) = φ(z, z̄) (5.1.2)

86
Equivalently, a conformal field may be viewed as a section of an operator-valued holomorphic
line bundle over Σ. Clearly, h and h̃ must be the same for all charts U ⊂ Σ.
Local conformal transformations form an infinite-dimensional Lie algebra, referred to as
the Virasoro algebra. Its generators Lm for m ∈ Z obey the following structure relations,
c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n,0 (5.1.3)
12
Here c is a real-valued constant referred to as the central charge of the Virasoro algebra. The
last term is proportional to the identity operator in the representation space of the Virasoro
algebra, and therefore commutes with all Lm . When c 6= 0, the Virasoro algebra is a central
extension of the Witt algebra, which has c = 0.
Actually, conformal transformations may be allowed to act independently on the holo-
morphic and anti-holomorphic coordinates, in which case we have two copies of the Virasoro
algebra, whose generators are denoted Ln acting on z and L̃n acting on z̄. These two Virasoro
algebras commute so that [Lm , L̃n ] = 0 for all m, n ∈ Z. The conformal weights h, h̃ ∈ R
characterize the representation of the Virasoro algebra under which the field φ transforms,
[Lm , φ(z, z̄)] = z m+1 ∂z φ(z, z̄) + h(m + 1)z m φ(z, z̄)
[L̃m , φ(z, z̄)] = z̄ m+1 ∂z̄ φ(z, z̄) + h̃(m + 1)z̄ m φ(z, z̄) (5.1.4)
The minimal models correspond to the irreducible representations of the Virasoro algebra
for 0 < c < 1, for which the spectrum of allowed conformal weights h (of primary conformal
fields) is finite. Minimal models have rational conformal weights h, h̃ and a rational central
charge c. They are special cases of the more general class of “rational conformal field theo-
ries,” discussed further in section 13.6. For 0 < c < 1, the unitary minimal models constitute
a further subclass of conformal field theories that are so-called unitary, and in particular are
associated with F being a genuine Hilbert space. Many conformal field theories may be
realized in terms of free fields.

5.1.2 Free conformal fields

We shall limit our examples in the sequel to free conformal field theories. A free quantum
field satisfies a linear partial differential equation on Σ, given by a differential operator D.
The order of the operator D is usually of order no larger than 2. All correlation functions
of a free quantum field may be evaluated in terms of the Green function D−1 and the
functional determinant of D. For free conformal fields on an arbitrary genus-one Riemann
surface Σ, the correlators and determinants may be expressed in terms of elliptic functions
and modular forms. It turns out that the quantization of free conformal fields already
provides a sufficiently rich structure for all of string perturbation theory on an arbitrary
flat space-time manifold, including Minkowski space-time with or without partial toroidal
compactification, as will be further detailed in sections 12 and 13.

87
5.2 The bc system
An important example of a free conformal field theory on a Riemann surface Σ, that has an
immediate connection with elliptic functions and modular forms, is provided by a system of
two quantum fields b and c governed by a first order action S[b, c],
Z
1 i
S[b, c] = d2 z b(z, z̄)∂z̄ c(z, z̄) d2 z = dz ∧ dz̄ (5.2.1)
2π Σ 2
Invariance of S[b, c] under local conformal transformations requires both fields to have h̃ = 0
and their remaining conformal weights to be conjugate: if b has conformal weight (h, 0)
then c must have conformal weight (1 − h, 0). The field equations for b and c are obtained
using the variational principle for S[b, c], and are given by,
∂z̄ b = ∂z̄ c = 0 (5.2.2)
so that b and c are locally holomorphic functions of z on Σ. The quantum field theory with
these assignments is referred to as the bc system of weight h. Below we shall consider the bc
system on the annulus and on the torus. The bc system on a Riemann surface of arbitrary
genus will be discussed in appendix D.7.

5.2.1 The bc system on an annulus

An annulus is conformally isomorphic to a cylinder. Choosing the local complex coordinate
z 0 = x + iy on the cylinder with x, y ∈ R and the periodic identification x ≈ x + 1,
0
the coordinate z on the annulus is given by the conformal map z = e2πiz = e2πi(x+iy) .
Holomorphicity and periodicity of the fields b and c on the cylinder produce the following
Fourier expansion in the coordinate z 0 = x + iy,12
X X
b0 (z 0 ) ∼ bn e−2πin(x+iy) c0 (z 0 ) ∼ cn e−2πin(x+iy) (5.2.3)
n∈Z n∈Z

If the fields b and c have conformal weights (h, 0) and (1−h, 0) respectively, then the fields on
the annulus are obtained by the conformal transformation (5.1.2), and the Fourier expansions
0
become Laurent expansions in z = e2πiz given as follows,
X X
b(z) = bn z −n−h c(z) = cn z −n−1+h (5.2.4)
n∈Z n∈Z

Considering the fields b and c as quantum operators, the coefficients bn and cn are operators.
The closest connection with holomorphic and meromorphic functions will be obtained by
requiring the operators to satisfy anti-commutation relations,
{bm , bn } = {cm , cn } = 0 {bm , cn } = δm+n,0 (5.2.5)
12
We use proportionality signs here in order to leave the precise normalization to be fixed on the annulus.

88
where {A, B} = AB + BA is the anti-commutator and δm+n,0 is the Kronecker symbol. We
choose a polarization in which bn and cn are annihilation operators for n > 0 while b†n = b−n
and c†n = c−n are creation operators. The ground state |0i is annihilated by all bn , cn for
n > 0. The anti-commutation relations for b0 , c0 define a Clifford-Dirac algebra whose unique
irreducible representation has dimension 2. We shall define the ground state by,

b0 |0i = h0|c0 = 0 bn |0i = 0 cn |0i = 0 n≥1 (5.2.6)

The Fock space F is then generated by applying arbitrary polynomials in b†n and c†n for all
possible values of n > 0 to the ground state |0i.

5.2.2 The operator product expansion

The fields b(z) and c(z) are holomorphic in z on the annulus. The product c(z)b(w) of
operators, however, is singular as z → w. This may be seen by applying the operator to the
ground state and using the fact that |0i is annihilated by all bn with n > 0,
X X
c(z)b(w)|0i = cm z −m−1+h bn w−n−h |0i (5.2.7)
m∈Z n≤0

The contributions from m ≤ 0 are non-singular as z → w while for m > 0 we may replace the
product cm bn by the anti-commutator {cm , bn } = δm+n,0 in view of cm |0i = 0. The resulting
series is geometric in w/z, absolutely convergent for |z/w| < 1, and may be analytically
continued in z and w throughout the annulus,
X 1
c(z)b(w)|0i = z −m−1+h wm−h |0i + regular = |0i + regular (5.2.8)
m>0
z−w

where “regular” stands for all contributions that have a finite limit as z → w. The pole in
z − w arises due to the infinite nature of the series and is universal in the following sense. If
we apply the operator c(z)b(w) to an arbitrary Fock space state, obtained by acting on |0i
with an arbitrary polynomial in b†n and c†n with n > 0, then only a finite number of terms in
the infinite series (5.2.8) will be affected, which leaves the singularity unchanged. For this
reason, one promotes the singularity to an operator statement,
IF
c(z)b(w) = + regular (5.2.9)
z−w
where IF is the identity operator in the Fock space F. The Laurent series for the operator
b(z)c(w) is absolutely convergent for |w/z| < 1, produces the pole IF /(z − w), and may
be analytically continued to arbitrary z, w in the annulus. Comparing with the pole in
c(z)b(w), we observe that the sign reversal is consistent with the fact that b and c satisfy
anti-commutation relations. Actually, the pole term is only the first in an infinite Laurent

89
series expansion of the operator c(z)b(w) in powers of z − w. The general expansion is
referred to as the operator product expansion of the operators c(z) and b(w),
∞
IF X
c(z)b(w) = + (z − w)n j (n+1) (w) (5.2.10)
z − w n=0

The coefficients j (n+1) (w) for n > 0 are conformal primary fields of weight (n + 1, 0). The
transformation law for j (1) is more complicated and will be discussed later. The operators
b(z)b(w) and c(z)c(w) similarly have operator product expansions. In view of the anti-
commutation relations (5.2.5) we have b(z)2 = c(z)2 = 0 and,

b(z)b(w) = (z − w)(∂w b(w)) b(w) + O (z − w)2

c(z)c(w) = (z − w)(∂w c(w)) c(w) + O (z − w)2

(5.2.11)

so that the operator product expansion is entirely regular.

5.3 Correlators of the bc system on the torus

We now consider the bc system on an arbitrary torus C/Λ with Λ = Z ⊕ τ Z and Im (τ ) > 0.
The fields b and c are meromorphic and doubly periodic, and are thus elliptic functions. On
the torus the bc systems for different integer values of h are all isomorphic to one another, in
view of the fact that the torus has a holomorphic differential dz which is nowhere vanishing.
As a result, a field φ(z)(dz)h of conformal weight (h, 0) is isomorphic to a scalar field φ(z)
of weight (0, 0). This isomorphism allows us for a direct evaluation of all correlators of b, c
fields on the torus. A correlator is the matrix element in the vacuum state |0i of a product
of b and c fields evaluated at distinct points on Σ,

A = h0|c(z1 ) · · · c(zN )b(w1 ) · · · b(wM )|0i (5.3.1)

The correlators are meromorphic in zn for all n = 1, · · · , N and in wm for all m = 1, · · · , M .

We shall now show that the positions of their zeros and poles in each variable are known from
the operator product expansions discussed earlier. In view of (5.2.11), the correlator has a
simple zero when two points zn and zn0 coincide, as well as when two points wm and wm0
coincide. In view of (5.2.9), the correlator has a simple pole whenever a point zn approaches
a point wm , whose residue is given by a correlator with one fewer b and one fewer c,

(−)N −1+m−n [ \
A= h0|c(z1 ) · · · c(zn ) · · · c(zN )b(w1 ) · · · b(wm ) · · · b(wM )|0i + regular (5.3.2)
zn − wm
The sign factor arises from permuting the positions of the fields, the hatted operators are to
be omitted, and “regular” includes all terms that are non-singular as zn → wm .

90
 Considered as a function of zn , the correlator has precisely M simple poles at w1 , · · · , wM .
In view of Theorem 2.3 it must have M zeros in zn . These zeros must include the N − 1
points zn0 with n0 6= n, so we must have M ≥ N − 1. Similarly, as a function of wm , the
correlator has precisely N simple poles at z1 , · · · , zN . In view of Theorem 2.3 it must have
N zeros in wm which must include the M − 1 points wm0 with m0 6= m, so we must also have
N ≥ M − 1. Combing both inequalities allows for either M = N or M − N = ±1. The latter
is excluded by symmetry under reversing the sign of both b and c. Thus only M = N gives
a non-vanishing correlator. We shall now evaluate the correlators A for M = N in terms of
ϑ-functions using Theorems 2.3, 2.6, and 2.7.

5.3.1 The 4-point correlator

We begin with the correlator h0|c(z1 )c(z2 )b(w1 )b(w2 )|0i. As a function of z1 , it has a simple
zero at z2 , simple poles at w1 and w2 , and no other poles. In view of the second relation in
Theorem 2.3, it must have a second zero, whose position is determined by the third relation
in Theorem 2.3 to be given by −z2 + w1 + w2 (mod Λ). Proceeding analogously for z2 using
Theorem 2.6 we obtain the following expression for the correlator,
ϑ1 (z1 + z2 − w1 − w2 )ϑ1 (z1 − z2 )ϑ1 (w1 − w2 )
h0|c(z1 )c(z2 )b(w1 )b(w2 )|0i = A0 (5.3.3)
ϑ1 (z1 − w1 )ϑ1 (z1 − w2 )ϑ1 (z2 − w1 )ϑ1 (z2 − w2 )
Here, A0 is independent of z1 , z2 , w1 , w2 and all dependence on τ has been suppressed. We
may use this expression to obtain the correlator h0|c(z1 )b(w2 )|0i by letting z2 → w1 ,
1 A0
h0|c(z1 )c(z2 )b(w1 )b(w2 )|0i = + regular (5.3.4)
z2 − w1 ϑ01 (0)
The two-point correlator is constant and saturated by the zero modes of the c and b fields,
A0
h0|c(z1 )b(w2 )|0i = (5.3.5)
ϑ01 (0)
consistently with the fact that no function on a torus can have a single simple pole. Al-
ternatively, we may use Theorem 2.7 to express the correlator in terms of the Weierstrass
ζ-function and its derivatives. Since the poles in z1 , z2 of the correlator at w1 , w2 are simple,
and must have opposite residue, we have,
2
X 
h0|c(z1 )c(z2 )b(w1 )b(w2 )|0i = A3 + Ai ζ(zi − w1 ) − ζ(zi − w2 ) (5.3.6)
i=1

where A1 , A2 , A3 are independent of z1 , z2 , w1 , w2 . Matching the residues with (5.3.4), we

find A1 = −A2 = −A0 /ϑ01 (0). This condition also shows that the correlator has zeros at
z1 = z2 and w1 = w2 provided A3 = 0. We leave it as an exercise to the reader to express the
correlator in terms of the Weierstrass ℘ function and to show that these three expressions
may be related to one another with explicit identities.

91
5.3.2 Arbitrary correlators
Correlators for arbitrary N may be evaluated iteratively. As a function of z1 the correlator
has N simple poles at w1 , · · · , wN , andP
N − 1 known
PN zeros at z2 , · · · , zN . The missing zero
is given by Theorem 2.4 by the sum − N z
n=2 n + n=1 wn . Putting all together, we have,

QN N
ϑ1 (D) n=2 ϑ1 (z1 − zn ) X
A = Â1 QN D= (zn − wn ) (5.3.7)
n=1 ϑ1 (z1 − wn ) n=1

where Â1 is independent of z1 , but depends on all other zn and wn . Again, all dependence
on τ has been suppressed. Proceeding iteratively on the points zn we find that the factor
ϑ1 (D) always accounts for the missing zero. Putting all together, we obtain,
Q
ϑ1 (D) m<n ϑ1 (zm − zn )ϑ1 (wm − wn )
A = A0 Q (5.3.8)
m,n ϑ1 (zm − wn )

where A0 depends only on τ . We leave it as an exercise for the reader to express the correlator
also in terms of Weierstrass ℘- and ζ-functions using Theorems 2.4 and 2.7 respectively.

5.4 The scalar field

A real-valued scalar field ϕ is governed by a second order differential operator on the torus,
namely the Laplace-Beltrami operator. In local complex coordinates (z, z̄) on the torus
Σ = C/Λ with Λ = Z ⊕ τ Z, its action may be expressed as follows,
Z
1
S[ϕ] = d2 z ∂z ϕ ∂z̄ ϕ (5.4.1)
2π Σ
The corresponding field equation ∂z ∂z̄ ϕ = 0 implies that ∂z ϕ is a locally holomorphic field
while ∂z̄ ϕ is a locally anti-holomorphic field. On an annulus, these quantum fields admit
Laurent expansions,
X X
i∂z ϕ(z, z̄) = ϕm z −m−1 i∂z̄ ϕ(z, z̄) = ϕ̃m z̄ −m−1 (5.4.2)
m∈Z m∈Z

where the coefficient operators ϕm and ϕ̃m satisfy commutation relations,

[ϕm , ϕn ] = mδm+n,0 [ϕ̃m , ϕ̃n ] = mδm+n,0 (5.4.3)

We choose a polarization such that ϕ−n = ϕ†n and ϕ̃−n = ϕ̃†n , and define the ground state
|0i to be annihilated by ϕn |0i = ϕ̃n |0i = 0 for n ≥ 0. The Fock space is constructed by
applying arbitrary polynomials in ϕ†n for n > 0 to |0i. Each such state has positive norm so
that the Fock space for a scalar field is actually a Hilbert space, which we denote by F.

92
 The scalar field theory is conformal and has two commuting Virasoro algebras generated
by Lm and L̃m whose expressions in terms of ϕm and ϕ̃m are given as follows,13
∞
1 1 2 X 1X
L0 = − + ϕ0 + ϕ−n ϕn Lm = ϕm−n ϕn for m 6= 0 (5.4.4)
24 2 n=1
2 n∈Z

and similarly for L̃m . The operators ∂z ϕ and ∂z̄ ϕ are conformal primary fields of weights
(1, 0) and (0, 1) respectively, and satisfy the corresponding transformations given in (5.1.4).
The operator product expansion of two holomorphic fields is given by,
1
∂z ϕ(z)∂w ϕ(w) = − 2 T (w) + O(z − w) (5.4.5)
(z − w)2

where the Laurent expansion of the stress tensor T (w) for the holomorphic component of
the scalar field is given in terms of the Virasoro generators Lm ,
X
T (w) = Ln w−n−2 (5.4.6)
n∈Z

and similarly for the anti-holomorphic field ∂z̄ ϕ and the generators L̃m .
We may attempt to use the operator product expansion (5.4.5) to evaluate the correlator
of holomorphic fields ∂z ϕ(z), just as we had in the case of the bc system. For simplicity, we
choose matrix elements between the ground state |0i = |0, 0i on the torus Σ,

h0|∂z ϕ(z1 ) · · · ∂z ϕ(zN )|0i (5.4.7)

The correlator vanishes for odd N and is meromorphic in all its points zn for even N . As a
function of zn , the correlator has a double pole at every point zm with m 6= n. For example,
for the 2-point correlator, the double pole shows that the correlator must be of the form,

h0|∂z ϕ(z1 )∂z ϕ(z2 )|0i = ℘(z1 − z2 ) + A4 (5.4.8)

where A4 is independent of z1 , z2 . In contrast with the bc system, the zeros of the correlator
are not manifest now, and the constant A4 cannot be fixed by using known zeros. It may
be determined, however, by using not just the pole term in the operator product expansion,
but also the term proportional to the stress tensor T , and we find A4 = 0. For higher
point correlators, however, the corresponding constant terms will not be so easy to evaluate.
Instead we shall turn to evaluating the full correlators of the field ϕ itself.
13 1
The offset − 24 in L0 arises from the conformal transformation of the expression on the cylinder (which
has no such offset) to the annulus. It may also be viewed as the sum of the zero-point energies of the
1
harmonic oscillators, using the fact that the sum of all positive integers m gives ζ(−1) = − 12 .

93
5.5 The scalar Green function on the torus
The scalar Laplace-Beltrami operator on the torus has a zero mode and is not invertible on
all scalar functions. However, an inverse may be defined on its range, namely on functions
orthogonal to constant functions on Σ. To construct this inverse, we shall present two
different approaches, the first based on Fourier series, and the second using ϑ-functions.

5.5.1 The Laplacian on the torus

An arbitrary Riemannian metric on the torus Σ may be rescaled to a flat metric using a Σ-
dependent Weyl transformation, as shown in more detail in appendix B. A further rescaling
by a constant Weyl transformation allows us to obtain a flat metric with unit area. This
reduced metric depends on a single complex modulus parameter τ ∈ H, and we represent
the torus by the quotient Σ = C/Λ for the lattice Λ = Z + Zτ . The points on Σ may be
parametrized by a complex coordinate z = x + yτ for x, y ∈ R/Z, for example in the range
0 ≤ x, y ≤ 1 (see Figure 2). In these coordinates, the flat metric g and associated volume
form dµg of unit area are given by,

|dz|2 |dx + τ dy|2 i dz ∧ dz̄

g= = dµg = = dx ∧ dy (5.5.1)
τ2 τ2 2 τ2

We define an inner product on the space of complex-valued functions f1 , f2 : Σ 7→ C by,

Z
hf1 |f2 i = dµg f¯1 f2 (5.5.2)
Σ

making the space of bounded functions on Σ into an L2 (Σ) Hilbert space. The Cauchy-
Riemann operators in these coordinates take the form,

1 1
∂z = (∂y − τ̄ ∂x ) ∂z̄ = − (∂y − τ ∂x ) (5.5.3)
2iτ2 2iτ2

while the Laplace-Beltrami operator, or simply Laplacian, is given by,

1
∆g = −4τ2 ∂z̄ ∂z = − (∂y − τ ∂x )(∂y − τ̄ ∂x ) (5.5.4)
τ2

The Laplacian is self-adjoint with respect to the inner product (5.5.2) on L2 (Σ) and positive.
It is invariant under translations in x, y and thus commutes with the operators i∂x and i∂y
that generate infinitesimal translations in the x and y directions respectively. For fixed τ , the
Laplacian ∆g has a discrete spectrum with a single zero mode, namely the constant function
on Σ, so that Ker∆g = C. Its range is the set of functions orthogonal to constant functions.

94
5.5.2 The scalar Green function via Fourier series
Since the self-adjoint operators i∂x and i∂y mutually commute and commute with ∆g , the
Laplacian is diagonal in a basis where i∂x and i∂y are simultaneously diagonal. Such a basis
is given by the following orthonormal basis of doubly periodic functions on Σ,

fm,n (x, y) = e2πi(nx−my)

hfm0 ,n0 |fm,n i = δm0 ,m δn0 ,n m, m0 , n, n0 ∈ Z (5.5.5)

They also satisfy the completeness relation,

X
fm,n (u, v)fm,n (x, y) = δ(x − u)δ(y − v) (5.5.6)
m,n∈Z

where the Dirac δ functions on the right are periodic in their argument with period 1. One
verifies that the Laplace operator is indeed diagonal in this basis, and we find,

|m + nτ |2
∆g fm,n = λm,n (τ ) fm,n λm,n (τ ) = 4π 2 (5.5.7)
τ2
The zero eigenvalue corresponds to the constant function with m = n = 0, while all other
eigenvalues are strictly positive, as expected. The Green function is customarily defined as
the inverse of ∆g /4π on the space of functions orthogonal to constants. In view of translation
invariance, we have G(z, w|τ ) = G(z − w|τ ) with z = x + yτ , and G(z|τ ) is given by,
X τ2
G(z|τ ) = e2πi(nx−my) (5.5.8)
π|m + nτ |2
(m,n)6=(0,0)

Applying the Laplacian, and using the completeness relation (5.5.5), one verifies,

∆g G(z − w|τ ) = 4πτ2 δ (2) (z − w) − 4π (5.5.9)

where the 2-dimensional δ-function, with w = u + τ v, is defined by,

τ2 δ (2) (z − w) = δ(x − u)δ(y − v) (5.5.10)

By construction, the right side of (5.5.9) is orthogonal to constant functions. Actually, the
Green function obtained in (5.5.8) itself is orthogonal to constant functions,
Z
dµg (z) G(z − w|τ ) = 0 (5.5.11)
Σ

since the zero mode (m, n) = (0, 0) has been removed from the sum in (5.5.8).

95
5.5.3 The scalar Green function via ϑ-functions
We may take equations (5.5.9) and (5.5.11) as defining G and express both in terms of the
complex variables z, w,
π
−∂z ∂z̄ G(z − w|τ ) = πδ (2) (z − w) − (5.5.12)
τ2
The solution for G is easy to construct in terms of ϑ-functions, as we shall now show. This
method will generalize to constructing the Green function on Riemann surfaces of arbitrary
genus, whereas the method of Fourier series of the preceding section does not generalize since
higher genus Riemann surfaces have no continuous isometries.
We proceed as follows for fixed τ ∈ H. For |z − w|  1 and |z − w|  τ2 , the global
structure of the torus is negligible and the Green function must approach its expression on
C which is given by G(z − w|τ ) ≈ − ln |z − w|2 . The function ϑ1 (z − w|τ )/ϑ01 (0|τ ) behaves
as z − w for small |z − w| and is therefore a reasonable candidate to play the role of z − w on
the torus. However, ϑ1 (z − w|τ ) cannot be a doubly periodic function as it is holomorphic
and would then have to be constant by Liouville’s theorem. Still, given the transformation
laws of ϑ1 (z − w|τ ) under z → z + 1 and z → z + τ given in (2.6.8), one readily finds a
simple additive term to restore double periodicity, and G(z, w|τ ) = G(z − w|τ ) is given by,
 ϑ1 (z|τ ) 2 2
 
G(z|τ ) = − ln  0  + 2π (Im z) + G0 (τ ) (5.5.13)
ϑ1 (0|τ )  τ2
Upon applying the Laplace operator, the first term produces the δ-function, while the second
term produces the constant on the right side of (5.5.12). The z-independent term G0 (τ ) may
be determined by imposing the normalization condition (5.5.11).
To compute G0 (τ ) it is convenient to choose the range of integration − 21 ≤ x, y ≤ 12 . The
integrals of the last two terms in G are readily evaluated. To compute the integral of the
first term in G we use the product formula for ϑ1 given in (2.6.23), expand the logarithm of
the factors (1 − q n e±2πiz ) in powers of q, observe that their integral vanishes term-by-term,
and finally compute the integral of ln | sin(πz)|2 by the same expansion method. Putting all
together, we obtain the properly normalized scalar Green function,
 ϑ1 (z|τ ) 2 2
 
G(z|τ ) = − ln   + 2π (Im z) (5.5.14)
η(τ )  τ2
The expression (5.5.8) for G may be related to (5.5.14) by carrying out the sum over m first
in (5.5.8) using the following Lemma.
Lemma 5.1 The Lipschitz formula for Re (z) > 0, Re (s) > 0 and 0 < α ≤ 1 is given by,
X e2πinα ∞
(2π)s X
s
= (m + α)s−1 e−2πz(m+α) (5.5.15)
n∈Z
(z + in) Γ(s) m=0

The proof may be found in Rademacher [44].

96
5.5.4 Modular properties
A modular transformation γ ∈ SL(2, Z) leaves the metric g and the Laplacian ∆g invariant
provided τ and the complex coordinate z transform according to the following rules,
 
0 aτ + b 0 z a b
τ = γτ = z = γz = γ= (5.5.16)
cτ + d cτ + d c d

The transformations on the real coordinates are as follows,

 0   
x a −b x
0 = (5.5.17)
y −c d y

The spectrum of the Laplacian is invariant under SL(2, Z), as the eigenvalues λm,n (τ ) of ∆g
transform as follows,
 0   
0 0 m a −b m
λm,n (x, y) = λm0 ,n0 (x , y ) = (5.5.18)
n0 −c d n

in view of the relation n0 x0 − m0 y 0 = nx − my. It readily follows that also the normalized
Green function, given in (5.5.8), is modular invariant,
  
z − w  aτ + b
G = G(z − w|τ ) (5.5.19)
cτ + d  cτ + d

Modular invariance may be verified directly on the expression for the Green function in terms
of ϑ-functions in (5.5.14) by using the transformation laws of ϑ1 given in (3.8.5).

5.6 Scalar determinant and the first Kronecker limit formula

The functional determinant of ∆ = ∆g /4π, considered as an operator on the subspace
of functions orthogonal to constants, is formally given by the product of the eigenvalues
λm,n (τ )/4π. However, this product does not converge. To define the determinant we intro-
duce the following ζ-function associated with ∆,
X  4π s
ζ∆ (τ, s) = = Es (τ ) (5.6.1)
λm,n
(m,n)6=(0,0)

where Es (τ ) is the non-holomorphic Eisenstein series introduced in section 4.2. The series
for Es (τ ) is absolutely convergent for Re (s) > 1. Its analytic continuation to s ∈ C has a
simple pole at s = 1, is modular invariant, and satisfies the functional relation,

Γ(s)Es (τ ) = Γ(1 − s)E1−s (τ ) (5.6.2)

97
Taking the limit s → 0 of (4.2.16) gives ζ∆ (τ, 0) = −1, while the determinant is given by,

0 0
ln Det ∆ = −ζ∆ (τ, 0) = −∂s Es (τ ) (5.6.3)

s=0

To compute the determinant, we use the expressions for Es (τ ) in terms of the functions
As (τ ) and Bs (τ ) defined in (4.2.12) and (4.2.13),
0 πτ2
ζ∆ (τ, 0) = A0 (τ ) + − ln τ2 − ln(4π) (5.6.4)
3
The last three terms on the right side arise from differentiating Bs (τ )/Γ(s) at s = 0. To
compute A0 (τ ), we simplify the summand in (4.2.13) by using the
p expression for the modified
−x
Bessel function in terms of elementary functions, K− 1 (x) = π/2x e . We discover that
2
A0 (τ ) becomes a sum over holomorphic and anti-holomorphic dependence on τ ,
X 1 n o
A0 (τ ) = exp 2πimnτ1 − 2π|mn|τ2 (5.6.5)
m,n6=0
|m|

which may be compactly expressed as follows,

∞
Y
A0 (τ ) = −2 ln (1 − q m )(1 − q̄ m ) (5.6.6)
m=1

We may express the final combined result in terms of the Dedekind η(τ ) function,
0
(τ, 0) = − ln 4πτ2 |η(τ )|4


ζ∆ (5.6.7)
Combining this with ζ∆ (τ, 0) = −1 and (5.6.3), we obtain,
Det0 ∆ = 4πτ2 |η(τ )|4 = 4πDet0 ∆g (5.6.8)
which is manifestly modular invariant.

5.6.1 The first Kronecker limit formula

The first Kronecker limit formula gives the residue and the finite part of ζ∆ (τ, s) at its pole
s = 1 by the following expression,
a−1
ζ∆ (τ, s) = + a0 + O(s − 1) (5.6.9)
s−1
Using the functional relation (5.6.2), it is immediate that a−1 = −1. From the same formula
we get also a0 by,
d 
= − ln 4πτ2 |η(τ )|4 + 2γ


a0 = (s − 1)Es (τ )  (5.6.10)

ds s=1

where γ is Euler’s constant as it arises in Γ0 (1) = −γ.

98
5.6.2 The scalar partition function
The scalar determinant is intimately related to the partition function for the free scalar field
in statistical mechanics. A partition function Z(β) for a Hamiltonian H acting in a Hilbert
space F, evaluated at an inverse temperature β is given by,

Z(β) = TrF e−βH

(5.6.11)

where TrF is the trace evaluated in the Hilbert space F. The scalar field has two independent
sets of oscillators, namely ϕm and ϕ̃m , with independent Virasoro generators. In particular,
the generators L0 and L̃0 are given by a sum of decoupled harmonic oscillator Hamiltonians
plus the term due to the zero mode. Instead of the real-valued inverse temperature β > 0, we
shall complexify β and let β → −2πiτ whose real part is still positive in view of Im (τ ) > 0.
The partition function for L0 is then given as follows,
∞
1 1 2 Y 1 1 2
Tr q L0 = q − 24 + 2 ϕ0 ϕ
2 0 η(τ )−1


n
= q (5.6.12)
n=1
(1 − q )

where the trace is taken only over the oscillator part of the Hilbert space F. Proceeding
similarly for L̃0 and complexifying its inverse temperature β → 2πiτ̄ , we obtain the combined
partition function,
  1 2 1 2 1
Tr q L0 q̄ L̃0 = q 2 ϕ0 q̄ 2 ϕ̃0 (5.6.13)
|η(τ )|2
The zero mode operators ϕ0 and ϕ̃0 are self-adjoint and have a continuous spectrum of
real eigenvalues. Physically, these eigenvalues are the momenta of left- and right-movers.
Periodicity in τ1 → τ1 + 1 requires ϕ20 − ϕ̃20 ∈ 2Z. When the field ϕ takes values in R, we
require the eigenvalues of ϕ0 and ϕ̃0 to be equal to one another, and to range over R. The
full trace is then given by,
Z

L0 L̃0
 1 p2 1 0 − 21
TrH q q̄ = dp |q| = √ = (Det ∆g ) (5.6.14)
|η(τ )|2 R τ2 |η(τ )|2

which is precisely the modular invariant partition function evaluated using the functional
integral.

5.7 Spinor fields

The orthogonal Lie algebra SO(d) has spinor representations in addition to its tensorial
representations. A spinor representation may be viewed either as a double-valued represen-
tation of the group SO(d) or as a single-valued representation of the double cover Spin(d).
A manifold supports spinor fields provided it is orientable and its second Stiefel-Whitney

99
class vanishes. These conditions are satisfied in d = 2 for Riemann surfaces. In terms of
bundles, one decomposes the cotangent bundle T ∗ Σ into a direct sum of its holomorphic
and anti-holomorphic part T ∗ Σ = K ⊕ K ∗ where K is referred to as the canonical bundle.
A spin bundle S is a holomorphic line bundle whose square is isomorphic to the canonical
bundle S ⊗2 ≈ K, and a spinor field is then an operator-valued section of S. For more details
on Riemann surfaces see appendix B, and for the general construction of line bundles on an
arbitrary Riemann surface see appendix C.
Different realizations of spinor fields, corresponding to different spin bundles, on a given
Riemann surface Σ, are referred to as spin structures. They are labeled by the first homology
class of the surface with Z2 coefficients, namely H1 (Σ, Z2 ). For example, the homology group
of a compact genus g Riemann surface is isomorphic to Z2g and therefore the different spin
structures are labeled by Z2 on each one of the 2g independent homology cycle.
For the annulus, there is a single homology cycle, and thus two different spin structures,
referred to as Neveu-Schwarz (NS) or Ramond (R). Denoting a holomorphic spinor field of
conformal weight (h, 0) by ψ, its Laurent series on the cylinder and on the annulus, with
0
coordinates z 0 = x + iy and z = e2πiz respectively, are given as follows,
X X
NS ψ 0 (z 0 ) = ψr e−2πir(x+iy) ψ(z) = ψr z −r−h
r∈Z+ 21 r∈Z+ 21
X X
R ψ 0 (z 0 ) = ψn e−2πin(x+iy) ψ(z) = ψn z −r−h (5.7.1)
n∈Z n∈Z

A free spinor field on a Riemann surface is governed by a first order differential operator,
namely by a generalized bc system in which the fields b and c are allowed either NS or R
identifications. An important special case is that of a spinor field of conformal weight ( 21 , 0),
in which case the fields b and c have the same conformal weight ( 12 , 0) and are quantized
using anti-commutation relations as in the bc system we discussed earlier.14 In this case
the NS spin structure makes the field ψ(z) single-valued on the annulus, while the R spin
structure makes the field ψ(z) double-valued on the annulus.
For the torus, a canonical choice of homology cycles A and B is given by the identifications
z → z + 1 and z → z + τ respectively. The four spin structures are distinguished according
to whether the field is periodic or anti-periodic around each cycle. The signs are inherited
from the cylinder,

ψ(z + 1) = −e2πiα ψ(z)

ψ(z + τ ) = −e2πiβ ψ(z) (5.7.2)

14
Another important case is the superghost βγ system where b = β has weight ( 23 , 0) and c = γ has weight
(− 12 , 0) and the oscillators obey commutation relations.

100
for α, β ∈ {0, 21 } (mod 1) with the values 0 and 12 corresponding to NS and R respectively.
Under the modular transformations S and T the four spin structures transform as follows,
       
α β α α
S: → T : → (5.7.3)
β α β α + β + 21

where the addition α + β + 12 is evaluated mod 1. The spin structure α = β = 12 is

referred to as the odd spin structure, while the three other spin structures are referred
to as even spin structures. Under modular transformations, the odd spin structure maps
to itself, while the three even spin structures are permuted amongst one another. This
transformation rule precisely corresponds to the transformation rule for Jacobi ϑ-functions
with half-integer characteristics given in (3.8.5). This is no accident, as will be further
clarified below. Furthermore, in view of the existence of the holomorphic and nowhere
vanishing differential dz, spinor fields on the torus for a given spin structure but different
conformal weights (h, 0) are all isomorphic to one another. Therefore, we may focus attention
on the case h = 12 . For two functions f1 and f2 with the same spin structure (or more
accurately for two sections of a given spin bundle S), we define a Hermitian inner product,
Z
hf1 |f2 i = dµg f¯1 f2 (5.7.4)
Σ

which has a positive definite norm and produces a Hilbert space L2 (Σ, S).

5.8 Spinor Green functions on the torus

In this subsection, we shall derive the Green function for the spinor field on a torus Σ = C/Λ
with Λ = Z ⊕ τ Z for each one of the four spin structures. Just as we did for the scalar field,
we offer two perspectives: a calculation via Fourier analysis and a calculation directly in
terms of ϑ-functions.

5.8.1 Odd spin structure Green function

For the odd spin structure α = β = 12 , the Green function may be obtained from the results
of the scalar case, because the spinor field is doubly periodic and has the constant spinor as
a zero mode. We define the Green function S1 (z|τ ) that satisfies the following equation,
π
∂z̄ S1 (z − w|τ ) = πδ (2) (z − w) − (5.8.1)
τ2
by analogy with the Laplace equation satisfied by the scalar Green function G(z − w|τ ) in
(5.5.12). The solution is given in terms of the scalar Green function by,
2πi
S1 (z − w|τ ) = −∂z G(z − w|τ ) = ∂z ln ϑ1 (z − w|τ ) + Im (z − w) (5.8.2)
τ2

101
up to an additive constant which amounts to adding a term proportional to the Dirac zero
mode. Since S1 (z −w|τ ) has a single pole in z at w, and is doubly periodic, it cannot possibly
be meromorphic as no elliptic functions with a single simple pole exist. The second term
on the right is present to compensate. Alternatively, a term-by-term differentiation of the
Kronecker-Eisenstein series of G(z − w|τ ) in (5.5.8) gives the following expression for S1 ,
X e2πi(nx−my)
S1 (z − w|τ ) = (5.8.3)
m + nτ
(m,n)6=(0,0)

The sum is only conditionally convergent and requires regularization. An alternative Green
function S̃1 (z; w1 , w2 |τ ) allows for two poles in z at w1 and w2 with opposite residue. Re-
quiring anti-symmetry in w1 , w2 produces a unique Green function,

S̃1 (z; w1 , w2 |τ ) = S1 (z − w1 |τ ) + S1 (w1 − w2 |τ ) + S1 (w2 − z|τ )

= ζ(z − w1 |τ ) + ζ(w1 − w2 |τ ) + ζ(w2 − z|τ ) (5.8.4)

where the expression is recast in terms of the Weierstrass ζ-function on the second line above.

5.8.2 Even spin structure Green functions via Fourier analysis

The chiral Dirac operator ∂z̄ is not self-adjoint. However, it is a linear combination with
constant coefficients of two self-adjoint operators i∂x and i∂y ,
1
∂z̄ = (i∂y − τ i∂x ) (5.8.5)
2τ2
Since the operators i∂x and i∂y mutually commute, the operator ∂z̄ may be diagonalized
in the basis in which i∂x and i∂y are diagonal. However, since the coefficients of the linear
combination in (5.8.5) are complex, the eigenvalues of ∂z̄ are complex.
For a given even spin structure [κ] = [ αβ ], the operators i∂x and i∂y are diagonalized in
the following basis,
n o
f [κ]m,n (x, y) = exp 2πi((n + β + 12 )x − (m + α + 21 )y) (5.8.6)

The eigenvalues of the operator ∂z̄ are given by,

π 
λ[κ]m,n = m + α + 12 + (n + β + 12 )τ (5.8.7)
τ2
Since Im (τ ) > 0 and α, β do not both equal 12 mod 1, none of the eigenvalues vanishes. The
Green function for the operator τ2 ∂z̄ is then given by,

1 X f [κ]m,n (x, y) X e2πi{(n+β+ 12 )x−(m+α+ 12 )y}

S[κ](z|τ ) = = (5.8.8)
τ2 m,n∈Z λ[κ]m,n m,n∈Z
m + α + 12 + (n + β + 21 )τ

102
so that S[κ] satisfies,

∂z̄ S[κ](z − w|τ ) = πδ(z − w) (5.8.9)

where the integral of δ(z − w) over Σ is normalized to τ2 .

5.8.3 Even spin structure Green functions via ϑ-functions

We seek a meromorphic function S[κ](z − w|τ ), which has a single simple pole in a funda-
mental parallelogram for Λ, is doubly periodic with periods in 2Λ, and behaves as follows
under translations in Λ/(2Λ),

S[κ](z + 1|τ ) = −e2πiα S[κ](z|τ ) [κ] = [ αβ ]

S[κ](z + τ |τ ) = −e2πiβ S[κ](z|τ ) (5.8.10)

For even spin structures κ, the simple pole in C/Λ and the above translations are realized
uniquely by the ratio ϑ[κ](z|τ )/ϑ1 (z|τ ), and the normalization of the residue at z = w gives,
ϑ[κ](z − w|τ ) ϑ01 (0|τ )
S[κ](z − w|τ ) = (5.8.11)
ϑ[κ](0|τ ) ϑ1 (z − w|τ )
As an elliptic function in C/(2Λ), the Szegö kernel S[κ](z − w|τ ) has four simple poles,
namely at z ≡ 0, τ + 1 with residue 1, and at z ≡ 1, τ with residue −1.

5.9 Spinor determinant and the second Kronecker limit formula

The eigenvalues of the chiral Dirac operator ∂z̄ are complex, rendering the ζ-function ap-
proach to evaluating its determinant inapplicable. Instead we shall make use of the fact that
the dependence of the operator τ2 ∂z̄ on the modulus τ is holomorphic and we shall seek to
construct an object Det(∂z̄ ) that is holomorphic in τ .
What we do know how to evaluate using ζ-function methods is the determinant of the
Laplace operator ∆κ = −4τ2 ∂z ∂z̄ on a spinor with even spin structure κ. The Laplace
operator is self-adjoint, has positive spectrum, and no zero modes,
4π 2 
1 
2 + α
λ[κ]m,n = m + α + (n + β)τ  κ = 21 (5.9.1)
τ2 2
+β

In fact we may generalize the problem by taking the characteristics α, β to be arbitrary with
the sole restriction that (α, β) 6≡ (0, 0) mod 1. The corresponding ζ-function is defined by
the Kronecker-Eisenstein sum,
X τ2s
ζ∆κ (τ, s) = (5.9.2)
m,n∈Z
π s |m + α + (n + β)τ |2s

103
which is absolutely convergent for Re (s) > 1. We shall show below that ζ∆κ (τ, s) may be
analytically continued in s throughout the complex plane. The determinant is obtained from
this analytic continuation as usual,
0
ln Det∆κ = −ζ∆ κ
(τ, 0) (5.9.3)

We recast the Kronecker-Eisenstein sum in terms of the following heat-kernel representation,

Z ∞  
dt s X πt  2
Γ(s)ζ∆κ (τ, s) = t exp − m + α + (n + β)τ  (5.9.4)
0 t m,n∈Z τ2

and Poisson re-sum either in the summation variable m if β 6= 0 or in n if α 6= 0 (but not in

both). We shall work out the case β 6= 0, the other case being analogous, using the formula,
r
X
−πt|m+α+(n+β)τ |2 /τ2 τ2 X 2πim(α+(n+β)τ1 −πm2 τ2 /t−π(n+β)2 tτ2
e = e e (5.9.5)
m,n∈Z
t m,n∈Z

Substituting this result into (5.9.4) and splitting the sum over m, n into contributions for
m 6= 0 and m = 0, we have Γ(s)ζ∆κ (τ, s) = As (τ ) + Bs (τ ) with,
√ X X 2πim(α+(n+β)τ1 ) ∞ dt s− 1 −πτ2 (m2 /t+(n+β)2 t)
Z
As (τ ) = τ2 e t 2e
m6=0 n 0 t
Z ∞
√ dt s− 1 X −π(n+β)2 tτ2
Bs (τ ) = τ2 t 2 e (5.9.6)
0 t n

The integral As (τ ) is manifestly holomorphic in s throughout C, and may be evaluated by

performing the t-integral in terms of a Bessel function,
1
√X X  m s− 2
2πim(α+(n+β)τ1 )


As (τ ) = 2 τ2   e K 1 2πτ2 |m(n + β)| (5.9.7)
n + β  s− 2
m6=0 n

To evaluate the determinant, we only need A0 (τ ), which may be read off from thepabove for-
mula by setting s = 0 and using the expression for the spherical Bessel K− 1 (x) = π/2x e−x ,
2

∞ X
X 1 −2πτ2 m|n+β| 2πim(α+(n+β)τ1 )
+ e−2πim(α+(n+β)τ1 )


A0 (τ ) = e e (5.9.8)
m=1 n
m

Without loss of generality we may take 0 < β < 1 to carry out the sum and obtain,
∞ ∞
n+β 2πiα 2
X X 2
ln 1 − q n−β e−2πiα 
  
A0 (τ ) = − ln 1 − q
 e − (5.9.9)
n=0 n=1

104
Finally, we compute Bs (τ ) by performing the integrals under the sum,
1
X 1
Bs (τ ) = Γ(s − 12 )π 2 −s τ21−s (5.9.10)
n
|n + β|2s−1

The sum is absolutely convergent for Re (s) > 1 and may be evaluated in terms of Hurwitz
ζ-functions. The analytic continuation of the Hurwitz ζ-function is well-known, and the final
result is given by,

 ϑ[α β](0|τ ) 2
 
0
ζ∆κ (τ, 0) = − ln   (5.9.11)
η(τ ) 

This result is a version of the second Kronecker limit formula.

The determinant obtained above is for the bc system in which both b and c are conformal
spinor fields of weight ( 21 , 0), along with their complex conjugate fields. The natural candidate
for the determinant of the chiral operator ∂z̄ is obtained by retaining only the part with
holomorphic dependence on τ ,
 ϑ[α β](0|τ )
Det ∂z̄  = (5.9.12)

bc η(τ )

The spinor fields that are required in string theory for the worldsheet matter fermions are
further restricted to ψ = b = c, so that the b and c fields are actually identical fields.
The functional chiral determinant of the operator ∂z̄ in this case is the square root of the
determinant of the bc system,
1
 ϑ[α β](0|τ ) 2
Det ∂z̄  = (5.9.13)

1
ψ η(τ ) 2

Neither determinant is modular invariant or even covariant. Physicists say that there is a
global anomaly, which manifests the incompatibility between maintaining holomorphicity
and modular invariance. We may try a combination similar to the one we established in the
non-chiral case,
4 N N N N
X ϑi (0|τ ) 2 ϑ[00](0|τ ) 2 ϑ[0 12 ](0|τ ) 2 ϑ[ 12 0](0|τ ) 2
N = N + N + N (5.9.14)
i=2 η(τ ) 2 η(τ ) 2 η(τ ) 2 η(τ ) 2

The smallest value of N for which we have good modular transformations is N = 16, and
more generally any multiple of 16. In those cases, the sum of the numerators transforms as
a modular form of weight N/4. The first modular invariant is encountered as N = 24 for 48
chiral fermions.

105
• Bibliographical notes
A systematic introduction to quantum field theory and more specifically conformal field the-
ory in two dimensions is presented in the book by Di Francesco, Mathieu, and Sénéchal [45],
which also has an extensive bibliography. The lecture notes by Ginsparg [46] provide an
excellent introduction.
The development of two-dimensional conformal field theory and the use of radial quanti-
zation date back to [47]. Modern conformal field theory, based on the representation theory
of the Virasoro algebra, was developed in [48], where minimal models were also introduced.
The bc system was discussed in the context of string perturbation theory in [49]. Functional
determinants of Laplace and Cauchy-Riemann operators on the torus are evaluated there as
well as in the classic paper by Polchinski [50]. The Kronecker limit formulas are discussed
in detail in the lecture notes by Siegel [51], as well as in [22].
On higher genus Riemann surfaces, functional determinants of Laplace operators were
considered in [52] in connection with Ray-Singer torsion, in [53], and in [54] in connection with
Reidemeister torsion, and were evaluated in terms of the Selberg zeta function in [55, 56]. An
overview is presented in the booklet by Jorgensen and Lang [57]. Functional determinants of
Laplace and Cauchy-Riemann operators were evaluated in terms of Riemann ϑ-functions via
chiral bosonization in [58] and [59] by building on the properties of the holomorphic anomaly
established in [60] and [61].

106
6 Congruence subgroups and modular curves
The modular group SL(2, Z) possesses an infinite number of non-Abelian subgroups. Various
objects encountered in the previous sections, such as the Jacobi ϑ-functions, ϑ-constants,
and the Dedekind η-function, do not transform as modular forms under SL(2, Z), but they
are modular forms under certain subgroups of SL(2, Z). These subgroups belong to the
general class of congruence subgroups of SL(2, Z), which we shall now define and study.

6.1 Definition of congruence subgroups of SL(2, Z)

The principal congruence subgroup of level N ≥ 1 is defined as follows,
  
a b
Γ(N ) = , a, d ≡ 1 (mod N ) , b, c ≡ 0 (mod N ) (6.1.1)
c d

Clearly, we have Γ(1) = SL(2, Z) and Γ(N ) is a subgroup of SL(2, Z) for any N . Actually,
Γ(N ) is a normal subgroup of SL(2, Z) so that the quotient is itself a group. The quotient
group is isomorphic to SL(2, ZN ) (to physicists, isomorphism is synonymous with equality),

SL(2, Z)/Γ(N ) = SL(2, ZN ) (6.1.2)

namely the group of 2 × 2 matrices whose entries are integers mod N and whose determi-
nant is 1 mod N . Equivalently, the group Γ(N ) may be identified with the kernel of the
homomorphism Z → ZN of reduction mod N . The index [SL(2, Z) : Γ(N )] of Γ(N ) equals
the order of SL(2, ZN ), which is finite. It will be determined below shortly.
By definition, a group Γ is a congruence subgroup of SL(2, Z) of level N provided there
exists an integer N ≥ 1 such that Γ(N ) ⊂ Γ ⊂ SL(2, Z). Since the index of Γ(N ) is finite,
it follows that the index [SL(2, Z) : Γ] is finite for any congruence subgroup Γ. Note that Γ
need not be a normal subgroup of SL(2, Z) so that SL(2, Z)/Γ need not be a group.

6.2 The classic congruence subgroups

Besides the principal congruence subgroups Γ(N ), two other fundamental congruence sub-
groups of level N are defined as follows,15
  
a b
Γ1 (N ) = , a, d ≡ 1 (mod N ) , c ≡ 0 (mod N )
c d
  
a b
Γ0 (N ) = , c ≡ 0 (mod N ) (6.2.1)
c d

15
The groups of transposes are denoted by, Γ1 (N ) = {γ|γ t ∈ Γ1 (N )} and Γ0 (N ) = {γ|γ t ∈ Γ0 (N )}.

107
with the following manifest subgroup inclusions,

Γ(N ) ⊂ Γ1 (N ) ⊂ Γ0 (N ) ⊂ SL(2, Z) (6.2.2)

For N ≥ 2, none of the groups Γ(N ), Γ1 (N ), Γ0 (N ) contains the generator S of SL(2, Z)

and, while Γ1 (N ) and Γ0 (N ) contain the translation generator T , the principal congruence
subgroup Γ(N ) contains T N but not T .
We recall that Γ(N ) is a normal subgroup of SL(2, Z) induced by the homomorphism
Z → ZN on all its entries. Similarly, Γ(N ) is a normal subgroup of Γ1 (N ) induced by
Z → ZN on the entry b whose quotient is isomorphic to ZN . Finally, Γ1 (N ) is a normal
subgroup of Γ0 (N ) whose quotient is isomorphic to the group Z∗N of all multiplicatively
invertible elements in ZN . In summary, we have the following quotients and isomorphisms,

SL(2, Z)/Γ(N ) = SL(2, ZN )

Γ1 (N )/Γ(N ) = ZN
Γ0 (N )/Γ1 (N ) = Z∗N (6.2.3)

The order of SL(2, ZN ) is denoted 2dN and will be evaluated in section 6.3. The order of ZN
is N . The order of Z∗N equals the Euler function φ(N ), which counts the number of integers
in ZN that are relatively prime to N , and is given by the following formula,

Y 1
 Y
|Z∗N | = φ(N ) = N 1− N= pei i for ei ≥ 1 (6.2.4)
i
pi i

where the products are over distinct primes pi . For the special case where N is prime, we
have φ(N ) = N − 1. The indices of these subgroups are related to the orders of the quotient
groups as follows,

[SL(2, Z) : Γ(N )] = |SL(2, ZN )| = 2dN

[Γ1 (N ) : Γ(N )] = |ZN | = N
[Γ0 (N ) : Γ1 (N )] = |Z∗N | = φ(N ) (6.2.5)

We obtain the remaining indices,

[SL(2, Z) : Γ0 (N )] = 2dN /(N φ(N ))

[SL(2, Z) : Γ1 (N )] = 2dN /N (6.2.6)

using the multiplicative property [G : H] × [H : K] = [G : K] for groups K ⊂ H ⊂ G.

108
6.3 Computing the order of SL(2, ZN )
We shall now evaluate the order 2dN of SL(2, ZN ), following [13]. The order will enter
crucially in the valence formulas for congruence subgroups. To compute dN , we shall make
use of the Chinese remainder theorem (reviewed and proven in appendix A), which provides
a fundamental isomorphism for ZN in terms of its prime decomposition factors.
Let N have the following prime number decomposition N = i pei i where the pi are dis-
Q

tinct primes and ei are positive integer exponents. Then by the Chinese remainder theorem,
we have the following isomorphism,
Y
SL(2, ZN ) = SL(2, Zpei i ) (6.3.1)
i

Thus, the order |SL(2, ZN )| of the group SL(2, ZN ) is given by,

Y
|SL(2, ZN )| = |SL(2, Zpei i )| (6.3.2)
i

To compute the order of each component SL(2, Zpe ) we proceed by induction on e. We begin
by computing the order of GL(2, Zp ). To do so, consider the set of all matrices γ,
 
a b
γ= (6.3.3)
c d

whose entries a, b, c, d ∈ Zp may be parametrized by a, b, c, d ∈ {0, 1, · · · , p − 1}. To belong

to GL(2, Zp ), the entries must satisfy ad − bc 6= 0 (mod p). To enumerate all its elements
we proceed by listing the number of elements in the four subsets of possible values of the
doublet (a, b). No matrices with a = b = 0 belong to GL(2, Zp ); there are p − 1 doublets
with a 6= 0, b = 0 and for each non-zero value of a the p − 1 non-zero values of d and the
p arbitrary values c give (p − 1)2 p elements γ ∈ GL(2, Zp ). For the case a = 0, b 6= 0 the
roles of c and d are reversed and we obtain another (p − 1)2 p elements. Finally, for each one
of the (p − 1)2 doublets with a, b 6= 0, the entry c may take all p values, while d may take
all p values except the one that gives zero determinant, producing (p − 1)3 p contributions.
Putting all together, we find,
 
3 2 3 1
|GL(2, Zp )| = (p − 1) p + 2(p − 1) p = (p − 1)p 1 − 2 (6.3.4)
p

Any γ ∈ GL(2, Zp ) satisfies det γ 6= 0, and taking the quotient by the determinant normal
subgroup isomorphic to the multiplicative group (Zp )∗ with p − 1 elements, we obtain,
 
3 1
|SL(2, Zp )| = p 1 − 2 (6.3.5)
p

109
To proceed by induction on the exponent e, we consider an arbitrary matrix γ 0 (mod pe+1 )
and parametrize it uniquely (mod pe ),
 0 0  
0 e a b a b
γ =γ+p γ= (6.3.6)
c0 d 0 c d

where a, b, c, d ∈ Zpe and a0 , b0 , c0 , d0 ∈ Zp . Requiring det γ ≡ 1 (mod pe+1 ) requires ad − bc ≡

1 (mod pe ) so that γ ∈ SL(2, Zpe ). Now for an arbitrary matrix γ ∈ SL(2, Zpe ), we obtain
γ 0 ∈ SL(2, Zpe+1 ) if and only if,

ad0 + da0 − bc0 − cb0 ≡ 0 (mod p) (6.3.7)

At least two entries in γ are non-zero (mod p) since ad − bc ≡ 1 (mod p). If a 6= 0, we

let a0 , b0 , c0 take independent arbitrary values in Zp and determine the number d0 uniquely
by solving equation (6.3.7). If a = 0, then we must have b 6≡ 0 (mod p), take the values of
a0 , b0 , d0 arbitrary and solve uniquely for c0 using (6.3.7). In each case, to every γ ∈ SL(2, Zpe )
there correspond p3 elements γ 0 ∈ SL(2, Zpe+1 ). Hence we have,
 
3e 1
|SL(2, Zpe )| = p 1− 2 (6.3.8)
p
Putting all together, for any N ≥ 2 with prime number decomposition N = i pei i we have,
Q

Y 1

3
2dN = |SL(2, ZN )| = N 1− 2 (6.3.9)
i
pi

The order of the simplest non-trivial group SL(2, Z2 ) is 6 by the above formula. In fact, we
can write its generators in terms of the generators S and T of SL(2, Z) and we find,

SL(2, Z2 ) = I, S, T, ST, T S, ST S (mod 2) (6.3.10)

Opposite elements are not included in the set since I ≡ −I (mod 2).

6.4 Modular curves

Recall that the quotient SL(2, Z)\H of the upper half-plane H by SL(2, Z) is topologically
a sphere with one puncture corresponding to the unique cusp. The quotient SL(2, Z)\H
may be represented by a choice of fundamental domain in H with boundary sides identified
pairwise under generators of SL(2, Z). The standard choice F of fundamental domain for
SL(2, Z) was defined in (3.2.6) and depicted in Figure 6.
For an arbitrary congruence subgroup Γ the quotient is the modular curve Y (Γ) = Γ\H
which is again a Riemann surface. However, Y (Γ) may be a sphere or a surface of genus
greater than zero and may have several punctures corresponding to the several cusps of Γ.

110
6.4.1 Fundamental domain for a congruence subgroup Γ
The modular curve Y (Γ) may again be represented by a choice of fundamental domain FΓ
in H with opposite sides identified pairwise under generators of Γ. Given the fundamental
domain F for SL(2, Z), the fundamental domain FΓ may be obtained by applying the cosets
of SL(2, Z)/Γ to F ,

SL(2, Z)/Γ : F → FΓ (6.4.1)

Therefore FΓ is the union of a number of copies of F equal to the index [SL(2, Z) : Γ]. The
elliptic points of Γ are the images of the elliptic points i and ρ of SL(2, Z) under the cosets
SL(2, Z)/Γ and, similarly, the cusps of Γ are the images of the cusp i∞ of F under the
cosets SL(2, Z)/Γ. Thus, the number of cusps in Y (Γ) equals the index [SL(2, Z) : Γ]. For
a non-trivial congruence subgroup Γ 6= SL(2, Z), the index is strictly greater than 1, as may
be verified for the index for Γ = Γ(N ), Γ1 (N ), Γ0 (N ) evaluated in (6.2.5) and (6.2.6). Since
Y (Γ) has at least one puncture, it is always non-compact.
There is a useful alternative way of looking at the cusps of a congruence subgroup Γ of
SL(2, Z). Recall that the Borel subgroup Γ∞ of SL(2, Z) is defined by Γ∞ = {±T n , n ∈ Z}
and that SL(2, Z)/Γ∞ acts transitively on Q ∪ {∞}. This property results from the fact
that every rational number r ∈ Q is the image of i∞ under a unique coset in SL(2, Z)/Γ∞ .
To show this, we distinguish two cases: r = 0 is the image of i∞ under the transformation
S ∈ SL(2, Z)/Γ∞ , while for r 6= 0 we write r = a/c uniquely with a, c ∈ Z∗ , gcd(a, c) = 1,
and c > 0. By Bézout’s theorem (reviewed in appendix A) there exists a particular solution
(b0 , d0 ) for b and d to the equation ad − bc = 1. The general solution is given by b = b0 + ka
and d = d0 + kc, which is equivalent to the particular solution by right multiplication by T k ,
which proves the assertion. The cusps of Y (Γ) may now be viewed as those points in Q∪{∞}
which are inequivalent under the action of the congruence group Γ, or equivalently,16


{cusps of Γ} = Γ\SL(2, Z)/Γ∞ {∞} (6.4.2)

The number of cusps, counted with multiplicity, is again given by the index [SL(2, Z) : Γ].
For the example Γ = Γ(2) with SL(2, Z)/Γ(2) = SL(2, Z2 ), the fundamental domain
FΓ(2) is the union of the 6 images of F under the elements of SL(2, Z2 ) listed in (6.3.10).
The resulting fundamental domain FΓ(2) is depicted in Figure 8. Using the area formula
(3.2.4) we have area(FΓ(2) ) = 2π or area(FΓ(2) ) = 6 area(F ), given by the order of SL(2, Z2 ),
as expected. The Γ(2)-inequivalent elliptic points in FΓ(2) are i, i0 , i00 , i000 , ρ, ρ0 and its Γ(2)-
inequivalent cusps are 0, 1, i∞. The elliptic points i and i0 have multiplicity 2; i00 and i000 have
multiplicity 1; ρ and ρ0 have multiplicity 6, while the cusps 0, 1, i∞ all have multiplicity 2.
16
Throughout, it will be convenient to use the notation Γz = {γz for all γ ∈ Γ} for the set of all images
of the point z under the action of the group Γ.

111
 i∞

TF F TF H

i
i0 i0
• ρ • ρ0 •
TSF
• • TSF
SF

STSF STF STF STSF

• •
i000 i00

−1 − 12 0 1
2 1

Figure 8: The fundamental domain ΓΓ(2) for the congruence subgroup Γ(2), its elliptic points
i, i0 , i00 , i000 , ρ, ρ0 , and its cusps 0, ±1, i∞ are obtained as the images of the fundamental domain
F , the elliptic points i, ρ, and the cusp i∞ of SL(2, Z), respectively, under the action of the
group of cosets SL(2, Z2 ) = {I, S, T, ST, T S, ST S}.

6.5 Compactification of the modular curves

To compactify Y (SL(2, Z)), one adjoins the point {i∞} to the fundamental domain F of
SL(2, Z). To compactify Y (Γ) for an arbitrary congruence subgroup Γ, one similarly adjoins
{cusp of Γ} = Γ\SL(2, Z)/Γ∞ {i∞} to Y (Γ). For an arbitrary congruence subgroup Γ, the
resulting modular curve X(Γ) is a connected and compact Riemann surface.
An alternative but equivalent way to obtain the compact modular curve X(Γ) is to start
from the union H∗ = H ∪ {i∞} ∪ Q of the upper half-plane with the rational numbers and
infinity. One chooses a topology for H∗ whose open sets include all the standard open sets of
H (which are generated by the open coordinate discs whose closure lies entirely in H), as well
as the open sets generated by discs that are open in H and intersect R at a rational number.
The latter provide an open neighborhood for every point in {i∞} ∪ Q and are depicted in
green in Figure 9 for the discs at infinity Ui∞ and the discs U±1 and U0 intersecting the real
axis at the rational points ±1 and 0. The topological space H∗ obtained this way is not
Haussdorff. Recall that a topological space is Haussdorff if for any two distinct points of H∗
there exist neighborhoods of each point that are disjoint from each other.

112
 i∞

Ui∞

H
−i
U−1 U0 U1

R
• • •
−1 0 1

Figure 9: The green rectangle Ui∞ represents a neighborhood of the cusp i∞ of SL(2, Z)
while the green discs U−1 , U0 , U1 represent the neighborhoods of the additional cusps of Γ(2).

The modular curve X(Γ) may be obtained from H∗ by taking the extended quotient,
which includes the action of Γ on Q ∪ {i∞},

X(Γ) = Γ\H∗ (6.5.1)

The modular curve X(Γ) is Haussdorff even though H∗ is not. The compact modular curves
for the classic congruence subgroups are denoted as follows,


X(N ) = X Γ(N )


X1 (N ) = X Γ1 (N )


X0 (N ) = X Γ0 (N ) (6.5.2)

A modular form f (τ ) of weight k for a congruence subgroup Γ ⊂ SL(2, Z) corresponds to

k
a Γ-invariant differential form f = f (τ )dτ 2 on H∗ , and thus pulls back to a well-defined
differential on X(Γ) = Γ\H∗ . To obtain the dimensions of the spaces Mk (Γ) and Sk (Γ) of
modular forms and cusp forms of weight k, we can now use the machinery of line bundles
on the compact Riemann surfaces X(Γ), reviewed in appendix C.

6.6 The genus of X(Γ)

To study X(Γ), we shall view it as a branched covering of the sphere X(SL(2, Z)) by a
non-constant holomorphic function f : X(Γ) → X(SL(2, Z)). We provide a summary of
various relevant topological aspects of Riemann surfaces in appendix B.

113
6.6.1 The Hurwitz formula for branched coverings
Consider a non-constant holomorphic map f : X → Y between two compact Riemann
surfaces X, Y . The degree of f is the cardinality of f −1 denoted d = |f −1 (y)| for all but
finitely many points y ∈ Y . A formula valid for all points in Y is obtained in terms of
the degree of ramification ex of f at a point x ∈ X which, in local coordinates, is given by
f (z) − f (x) ∼ (z − x)ex . The degree of f may then be defined for all y ∈ Y by the formula,
X
d= ex (6.6.1)
x∈f −1 (y)

valid for an arbitrary point y ∈ Y .

Theorem 6.1 The Riemann-Hurwitz formula relates the genera gX and gY of X and Y to
the degree of f and the deviations from 1 of the degree of ramification,
X
2gX − 2 = d(2gY − 2) + (ex − 1) (6.6.2)
x∈X

A proof of the Riemann-Hurwitz formula is given in appendix B.

6.6.2 Calculating the genus of X(Γ)

Returning now to the map f : X(Γ) → X(SL(2, Z)) we obtain an alternative formula for its
degree in terms of the index of the congruence subgroup Γ ⊂ SL(2, Z),

[SL(2, Z) : Γ] if −I ∈ Γ
d= 1 (6.6.3)
2
[SL(2, Z) : Γ] if −I 6∈ Γ

We shall denote the elliptic points of SL(2, Z) of order 2h by yh ∈ SL(2, Z){eπi/h } for h = 2, 3
and the cusps by y∞ ∈ SL(2, Z){i∞}. The number of elliptic points of order 2h in X(Γ)
will be denoted by εh , and the number of cusps in X(Γ) by ε∞ .

Theorem 6.2 The genus g of X(Γ) is given by,

d ε2 ε3 ε∞
g =1+ − − − (6.6.4)
12 4 3 2
where d is given in (6.6.3), εh denotes the number of elliptic points of order 2h, and ε∞
denotes the number of cusps in X(Γ).
To prove this result using the Riemann-Hurwitz formula (6.6.2), we evaluate the contri-
butions from the points x ∈ X(Γ) for which ex > 1, namely from the points f −1 (yh ) and the
cusps of X(Γ). If f −1 (yh ) is a ramification point of X(Γ) then eyh = 1, while if it is not a

114
ramification point of X(Γ) then eyh = h. Using the degree formula (B.1.3) at the points yh ,
we have,
X
ex = εh + h(|f −1 (yh )| − εh ) (6.6.5)
x∈f −1 (yh )

where the first term arises from those points yh that are ramification points of X(Γ) while
the second term arises from those points in f −1 (yh ) that are not ramification points of X(Γ),
in each case weighted by their respective ramification degrees 1 and h. Using the degree
formula again to eliminate the number |f −1 (yh )|, one obtains,
X h−1
(ex − 1) = (d − εh ) (6.6.6)
h
x∈f −1 (yh )

The formula holds for the cusps by setting h = ∞. Using the Riemann-Hurwitz formula
(6.6.2), we set gY = 0, g = gX(Γ) , and substitute the contributions to the sum of (ex − 1)
obtained in (6.6.6) to obtain (6.6.4).

6.7 Formulas for d, ε2 , ε3 and ε∞ for Γ(N ), Γ1 (N ) and Γ0 (N )

In order to obtain formulas for the genus of X(Γ), and later on the dimensions of the spaces of
modular and cusp forms on X(Γ), we need to obtain the degree d, and formulas for ε2 , ε3 , ε∞
for an arbitrary congruence subgroup Γ. But first, it will be useful to obtain these formulas
for the special cases of the congruence subgroups Γ(N ), Γ1 (N ), and Γ0 (N ). To do so, we
begin by assembling all the partial results that are required.

6.7.1 The absence of elliptic points

Recall that the elliptic points of SL(2, Z) are the points of order four, namely SL(2, Z){i},
and the points of order six, namely SL(2, Z){e2πi/6 }. In this subsection and the next, we
shall count the number of elliptic points of orders four and six of the congruence subgroups
Γ(N ), Γ1 (N ), and Γ0 (N ). We begin with the following theorem.

Theorem 6.3 The following congruence groups have no elliptic points,

1. Γ(N ) for N ≥ 2;
2. Γ1 (N ) for N ≥ 4;
3. Γ0 (N ) for any N divisible by a prime p which obeys p ≡ −1 (mod 12).
To prove 1, we parametrize γ ∈ Γ(N ) as follows,

1 + N a0 N b0
   
a b
γ= = (6.7.1)
c d N c0 1 + N d0

115
For an elliptic γ we must have |a + d| < 2, following the analysis of subsection 3.2.3. The
condition |a + d| = |2 + N (a0 + d0 )| < 2 allows for three possibilities, namely N (a0 + d0 ) =
−1, −2, −3, which are all ruled out for all N ≥ 4. For both N = 2 and N = 3, the only
solution is given by a0 + d0 = −1. Combining this condition with det γ = 1, we find that we
must have N (a0 d0 − b0 c0 ) = 1 which has no solutions for N ≥ 2. To prove 2, we parametrize
γ ∈ Γ1 (N ) as follows,
1 + N a0
 
b
γ= (6.7.2)
N c0 1 + N d0
The condition |a + d| = |2 + N (a0 + d0 )| < 2 for γ to be elliptic again allows for three
possibilities, namely N (a0 + d0 ) = −1, −2, −3, which are all ruled out for N ≥ 4. To prove 3,
we parametrize γ ∈ Γ0 (N ) as follows,
 
a b
γ= (6.7.3)
N c0 d
The condition |a + d| < 2 again leaves three possibilities, namely a + d = 0, ±1. For
a + d = 0, we furthermore have −a2 − N bc0 = 1, so that a2 ≡ −1 (mod N ). For a + d = ±1,
we have −a2 ± a − N bc0 = 1 so that a2 ± a ≡ −1 (mod N ). Now if p ≡ −1 (mod 12)
then p ≡ −1 (mod 3) and p ≡ −1 (mod 4). But a2 ≡ −1 (mod p) has no solutions if
p ≡ −1 (mod 4) in view of (A.4.6) while a2 + a ≡ −1 (mod p) has no solutions in view
(A.4.8), which completes the proof of point 3 and of the theorem.

6.7.2 Counting elliptic points for Γ0 (N )

We now count elliptic points for Γ0 (N ), at the values of N for which they exist. There are
two cases, namely elliptic points of orders 4 and 6, for which the transformation in SL(2, Z)
is conjugate to either,
   
0 −1 0 −1
S= ST = (6.7.4)
1 0 1 1

with S 4 = (ST )6 = I, tr(S) = 0, and tr(ST ) = 1. Given the conditions on the traces, the
representatives of S and ST in Γ0 (N ) are of the form,
   
a b a b
γ4 = γ6 = (6.7.5)
N c0 −a N c0 1 − a

with −a2 − bc0 N = 1 and a − a2 − bc0 N = 1 respectively. We shall now assume the prime
decomposition N = i pαi i for distinct primes pi and αi ≥ 1. For the elliptic points of order 4
Q
the congruence equation mod N ,

a2 ≡ −1 (mod N ) (6.7.6)

116
is equivalent to the congruence equations a2 ≡ −1 (mod pαi i ). Using the results of subsection
A.4.1 the number of solutions is given by (A.4.6),
(
 0 4|N
ε2 (Γ0 (N )) = Q (6.7.7)

i 1 + (−1|pi ) 4-N

where (−1|p) = ±1 for p ≡ ±1 (mod 4) and vanishes for p = 2. For the elliptic points of
order 6, the congruence equation, a2 + a ≡ −1 (mod N ) is equivalent to,

(2a + 1)2 + 3 ≡ 0 (mod 4N ) (6.7.8)

The number of solutions was obtained in subsection A.4.2, and is given by,
(
 0 9|N
ε3 (Γ0 (N )) = Q (6.7.9)

i 1 + (−3|pi ) 9-N

where (−3|p) = ±1 for p ≡ ±1 (mod 3) and vanishes for p = 3.

6.7.3 Counting the number of cusps

The cusps of a modular curve X(Γ) for a congruence subgroup Γ ⊂ SL(2, Z) are given by
the cusps {∞} ∪ Q = SL(2, Z)∞ modulo equivalence under Γ. In this subsection, we shall
count the number ε∞ of cusps of X(Γ) for Γ = Γ(N ), Γ1 (N ), and Γ0 (N ).
We begin by considering the case Γ = Γ(N ). Two cusps r = m/n and r0 = m0 /n0 with
gcd(m, n) = gcd(m0 , n0 ) = 1 are equivalent under Γ(N ) provided there exists an γ ∈ Γ(N )
such that r0 = γr, or more explicitly,
m0 (1 + N a0 )m + N b0 n 1 + N a0 N b0
 
= γ= (6.7.10)
n0 N c0 m + (1 + N d0 )n N c0 1 + N d0

The equation on the left side is equivalent to mn0 ≡ m0 n (mod N ), whose solution is given
by (m0 , n0 ) = ±(m, n) (mod N ) in view of the fact that both pairs (m, n) and (m0 , n0 ) are
relatively prime. Therefore, the cusps for Γ(N ) are labelled by all pairs (m, n) (mod N ) with
gcd(m, n) = 1.
An alternative perspective on this problem uses double cosets and will lend itself more
directly to the cases Γ1 (N ) and Γ0 (N ). Double cosets will be used in the construction of
Hecke operators in section 10, where a more extensive discussion will be provided. Suffice
it here to say that if G is a group and H1 , H2 are subgroups of G, then a double coset is
an orbit under the left action of H1 and the right action of H2 containing a given element
g ∈ G, and is of the form H1 gH2 . The set of all double cosets is denoted,

H1 \G/H2 = {H1 g H2 , g ∈ G} (6.7.11)

117
Applying the construction to computing the cusps of Γ(N ), we set G = SL(2, Z), H1 = Γ(N ),
H2 = Γ∞ = {± ( 10 1b ) , b ∈ Z}. The simple coset SL(2, Z)/Γ∞ provides a one-to-one map
from ∞ to all the cusps of SL(2, Z) while the further left cosets identified cusps that are
equivalent under Γ(N ). Hence we have,

ε∞ (Γ(N )) = |Γ(N )\SL(2, Z)/Γ∞ | (6.7.12)

Since Γ(N ) is a normal subgroup of SL(2, Z) and Γ(N )\SL(2, Z) = SL(2, ZN ), we have,

Γ(N )\SL(2, Z)/Γ∞ = SL(2, ZN )/Γ̄∞ (6.7.13)

where Γ̄∞ = {± ( 10 1b ) , b ∈ ZN }, and therefore the number of cusps is given by,

dN
ε∞ (Γ(N )) = |SL(2, ZN )|/|Γ̄∞ | = (6.7.14)
N
in view of (6.3.9) and the fact that |Γ̄∞ | = 2N .
The case of an arbitrary congruence subgroup Γ may similarly be treated with the double
coset construction, and we have,

ε∞ (Γ) = |Γ\SL(2, Z)/Γ∞ | (6.7.15)

The difference with the case of Γ(N ) is that an arbitrary congruence subgroup Γ is not
a normal subgroup of SL(2, Z), and this is indeed the case for Γ1 (N ) and Γ0 (N ). The
calculation of the number of cusps in these cases is more involved and we refer to [11, 13]
for a detailed exposition and here only quote the result,
1 X
N ≥5 ε∞ (Γ1 (N )) = φ(a)φ(d)
2 ad=N
X

N ≥3 ε∞ (Γ0 (N )) = φ gcd(a, d) (6.7.16)
ad=N

where the sums are over a, d > 0 and furthermore,

ε∞ (Γ1 (2)) = ε∞ (Γ1 (3)) = ε∞ (Γ0 (2)) = 2 ε∞ (Γ1 (4)) = 3 (6.7.17)

This concludes our calculation of d, ε2 , ε3 and ε∞ , from which the genus of X(Γ) may be
computed using (6.6.4).

6.8 Explicit expressions for the genus

The formulas for the genus of X(Γ) are particularly simple for the cases when no elliptic
points occur, as specified in Theorem 6.1. In these cases the genus is obtained by combining

118
 N 2 3 4 5 6 7 8 9 10 11 12 13
3
dN /N
2
4 6 12 12 24 24 36 36 60 48 84
g X(N )
0 0 0 0 1 3 5 10 13 26 25 50
g X1 (N )
0 0 0 0 0 0 0 0 0 1 0 2
g X0 (N ) 0 0 0 0 0 0 0 0 0 1 0 0
ε2 , ε3 1, 0 0, 1 0, 0 2, 0 0, 0 0, 2 0, 0 0, 0 2, 0 0, 0 0, 0 2, 2

N 14 15 16 17 18 19 20 21 22 23 24 25
dN /N
72 96 96 144 108 180 144 192 180 264 192 300
g X(N )
49 73 81 133 109 196 169 241 241 375 289 476
g X1 (N )
1 1 2 5 2 7 3 5 6 12 5 12
g X0 (N ) 1 1 0 1 0 1 1 1 2 2 1 0
ε2 , ε3 0, 0 0, 0 0, 0 2, 0 0, 0 0, 2 0, 0 0, 2 0, 0 0, 0 0, 0 2,0

Table 1: Genera for X(N ), X1 (N ), X0 (N ) for 2 ≤ N ≤ 25 and the number of elliptic points
ε2 and ε3 in X0 (N ).

(6.6.4) with the formula for the degree in (6.6.1) with the expressions of the indices given in
(6.2.5), (6.2.6), and the formulas for the number of cusps in (6.7.14), (6.7.16), (6.7.17),

dN (N − 6)
N ≥3: g(X(N )) = 1 + (6.8.1)
12N
dN 1 X
N ≥5: g(X1 (N )) = 1 + − φ(a)φ(d)
12N 4 ad=N
dN ε2 (Γ0 (N )) ε3 (Γ0 (N )) 1 X

N ≥2: g(X0 (N )) = 1 + − − − φ gcd(a, d)
6N φ(N ) 4 3 2 ad=N

where ε2 (Γ0 (N )) and ε3 (Γ0 (N )) were given in (6.7.7) and (6.7.9) respectively. For low values
of N , we obtain the expressions for dN and the genera given in Table 1.

• Bibliographical notes
The books by Shimura [11] and by Diamond and Shurman [13] stand out for their clear
and useful introductions to congruence subgroups and modular curves. Our summary here
closely follows their presentation.

119
7 Modular forms for congruence subgroups
In this section, we shall discuss the structure and the counting of modular forms for an
arbitrary congruence subgroup Γ, and present the corresponding Eisenstein series for Γ.
These modular forms are maps from the modular curves X(Γ) to spaces of meromorphic
differentials. The modular curves are compact Riemann surfaces of arbitrary genus, and
the spaces of meromorphic differentials may be viewed as spaces of meromorphic sections
of holomorphic line bundles of the Riemann surface. We refer to appendix C for a review
of holomorphic line bundles on compact Riemann surfaces, the Riemann-Roch theorem,
various vanishing theorems, and dimension formulas for the spaces of meromorphic sections.
We shall then give, without proof, the general formula for the dimensions of modular forms
and cusp forms for arbitrary Γ and provide concrete examples for the cases of the standard
congruence subgroups.

7.1 Modular forms and cusp forms with respect to Γ

One defines modular functions, modular forms, and cusp forms for an arbitrary congruence
subgroup Γ by analogy with SL(2, Z). Let f (τ ) be a meromorphic function on the upper
half-plane H which obeys the following transformation law,
 
k a b
f (γτ ) = (cτ + d) f (τ ) γ= (7.1.1)
c d
for some integer k ∈ Z, and all γ ∈ Γ. Since Γ is a congruence subgroup, there exists an
integer N such that Γ(N ) ⊂ Γ, and hence a minimum integer h which divides N such that
T h ∈ Γ. As a result, f is invariant under τ → τ + h and admits a Fourier expansion,
X
f (τ ) = an qhn qh = e2πiτ /h (7.1.2)
n∈Z

1. f is a modular function of weight k if f is meromorphic at every cusp of Γ\H, namely

if the Fourier expansion of f (γτ ) has only a finite number of non-vanishing Fourier
coefficients an with negative n for every γ ∈ SL(2, Z);
2. f is a modular form of weight k if it is holomorphic in H and at every cusp of Γ\H,
namely the Fourier coefficients an of f (γτ ) vanish for all negative n for every γ ∈
SL(2, Z), and thus f is finite at all cups;
3. f is a cusp form of weight k if it is a modular form and vanishes at all cusps.
The spaces of modular and cusp forms of weight k are denoted Mk (Γ) and Sk (Γ) respectively.
Clearly, the smaller the subgroup, the more modular forms it will have,
Mk (Γ) ⊂ Mk (Γ0 )

0
Γ ⊂Γ (7.1.3)
Sk (Γ) ⊂ Sk (Γ0 )

120
In particular, a congruence subgroup Γ of level N contains Γ(N ) so that Mk (Γ) ⊂ Mk (Γ(N ))
and similarly for cusp forms.

7.1.1 Modular forms of odd weight

For the full modular group SL(2, Z) there are no non-zero modular forms of odd weight k
because the transformation law (7.1.1) for the element −I ∈ SL(2, Z) shows that f (τ ) = 0.
The same conclusion holds for any congruence subgroup Γ that contains −I. The congruence
subgroups Γ(2), Γ1 (2) = Γ0 (2) and Γ0 (N ) for every N contain −I, and thus have no modular
forms of odd weight k, while Γ = Γ1 (N ), Γ0 (N ) for N > 2 do not contain −I so that the
corresponding spaces Mk (Γ) and Sk (Γ) for k odd are not necessarily trivial.

7.1.2 Comments on terminology

The terminology for automorphic functions and automorphic forms and its relation with
the terminology for modular functions and modular forms is not entirely standard in the
Mathematics literature, and may lead to some confusion, which we shall eliminate here with
the help of precise definitions.
Automorphic functions and forms may be defined generally in terms of functions with
specific transformation properties under a discrete group acting on a complex manifold.
Here, we shall limit attention to the case where the complex manifold is the Poincaré upper
half-plane H and the discrete group Γ is a Fuchsian subgroup of SL(2, R), acting on H by
Möbius transformations (see appendix B for the definition of Fuchsian groups).
In the complex analytic category, following Shimura [11] and Diamond and Shurman [13],
an automorphic function f with respect to a Fuchsian group Γ is a meromorphic function
on H that is invariant under all γ ∈ Γ,
 
a b
f (γτ ) = f (τ ) for all γ = (7.1.4)
c d
and is meromorphic at the cusps of Γ in H. An automorphic form of weight k with respect to
Γ is a meromorphic function on H that instead transforms under all γ ∈ Γ with a non-trivial
automorphy factor (cτ + d)k ,

f (γτ ) = (cτ + d)k f (τ ) (7.1.5)

Equivalently, we may view an automorphic form f of weight k as a differential k/2-form

k
f = f (τ )(dτ ) 2 (7.1.6)

which is invariant under Γ, in analogy with the approach taken in subsection 3.4. The
nomenclature of a form is particularly suited here, and familiar to a physics audience.

121
 In the category of non-holomorphic functions, following Terras [20], we may define an
automorphic function as a complex-valued function of H that is invariant under Γ, thereby
generalizing the notion from the meromorphic case. From the point of view of a physicist, it
is natural to generalize the notion of an automorphic form to the non-holomorphic category
by defining an automorphic form of weight (k, `) to transform under all γ ∈ Γ by,

f (γτ ) = (cτ + d)k (cτ̄ + d)` f (τ ) (7.1.7)

corresponding to a differential form f of weight ( k2 , 2` ),

k `
f = f (τ )(dτ ) 2 (dτ̄ ) 2 (7.1.8)

which is strictly invariant under Γ. In these notes, unless the meaning is clear from the
context, we shall add qualifiers, such as holomorphic automorphic forms, meromorphic au-
tomorphic forms, or more generally non-holomorphic automorphic forms.
In the special case where Γ = SL(2, Z) or a congruence subgroup thereof, one refers
to modular functions and forms instead of automorphic functions or forms. Again, un-
less the meaning is clear from the context, these terms will be accompanied by qualifiers,
holomorphic modular forms such as the holomorphic Eisenstein series Ek (τ ), meromorphic
modular functions such as the j(τ )-function, meromorphic modular forms such as the inverse
discriminant ∆(τ )−1 , and non-holomorphic modular functions such as the non-holomorphic
Eisenstein series Es (τ ) and modular graph functions.

7.2 Holomorphic Eisenstein series for Γ(N )

The holomorphic Eisenstein series of integer weight k ≥ 3 was defined in (2.4.5) by,
0
X 1
Gk (τ ) = (7.2.1)
m,n∈Z
(mτ + n)k

where the prime on the sum instructs us to omit any contribution that makes the summand
singular, in this case just (m, n) = (0, 0). Clearly Gk (τ ) vanishes for odd k, and is a modular
form of weight k for even k, transforming as follows,
 
k a b
Gk (γτ ) = (cτ + d) Gk (τ ) γ= ∈ SL(2, Z) (7.2.2)
c d

The ring of modular forms of SL(2, Z) is polynomial and generated by G4 and G6 .

To construct modular forms for congruence subgroup Γ(N ) with N ≥ 2 we parametrize
the elements of the group Γ(N )\SL(2, Z) = (ZN )2 by a row vector v = (µ ν) whose entries

122
satisfy µ, ν ∈ {0, 1, · · · N −1} = ZN , and introduce the following Kronecker-Eisenstein sum,17
0 0
X 1 X 1
Gk (v|τ ) =
k = (7.2.3)
m,n∈Z (mN + µ)τ + nN + ν (mτ + n)k
(m,n)≡v (N )

For v 6≡ 0 (mod N ) none of the denominators vanish and the prime on the sum may be
omitted. The double sum is absolutely convergent for integer k ≥ 3, while for k = 1, 2 it
may be defined by the Eisenstein summation convention,
0
X 1 X X 1
Gk (v|τ ) = + (7.2.4)
(µτ + n)k m6=0
(mτ + n)k
n≡ν (N ) n≡ν (N )
m≡µ (N )

Under a transformation γ ∈ SL(2, Z), we have,

Gk (vγ|γτ ) = (cτ + d)k Gk (v|τ ) (7.2.5)

where the ingredients are defined as follows with a, b, c, d ∈ Z and ad − bc = 1,

 
a b aτ + b
γ= γτ = vγ = (aµ + cν, bµ + dν) (7.2.6)
c d cτ + d
Under the action of SL(2, Z) a row vector v may be mapped into itself or into different row
vector. For example, (0 0) maps to itself under any γ ∈ SL(2, Z), while the row (0 ν0 ) for
ν0 6≡ 0 (mod N ) maps to (µ ν) = (cν0 dν0 ) (mod N ). Conversely, given a row v = (µ ν)
with gcd(gcd(µ, ν), N ) = 1, then there exists a γ ∈ SL(2, Z) which maps v to (0, ν0 ) with
ν0 = gcd(µ, ν) (mod N ).

7.2.1 Invariance under Γ(N )

A transformation γ ∈ SL(2, Z) maps Gk (v|τ ) to a multiple of itself provided v satisfies
vγ − v ≡ 0 (mod N ), or more explicitly,

(a − 1)µ + cν ≡ 0 (mod N )
bµ + (d − 1)ν ≡ 0 (mod N ) (7.2.7)

This condition is realized for an arbitrary γ ∈ Γ(N ) and an arbitrary pair (µ ν) ∈ (ZN )2 .
A given pair v = (µ ν) may be invariant under a subgroup of SL(2, Z) that is larger than
Γ(N ). For example (0 0) is invariant under all of SL(2, Z). But if we require that all
v ∈ (ZN )2 be invariant then the largest invariance group is Γ(N ). To see this, it suffices
to choose (µ ν) = (1, 0) (mod N ) and (µ ν) = (0, 1) (mod N ) to obtain the conditions
a, d ≡ 1 (mod N ) and b, c ≡ 0 (mod N ) so that γ ∈ Γ(N ).
17
When no confusion is expected to arise we shall sometimes abbreviate (mod N ) as simply (N ).

123
7.2.2 Asymptotics near the cusps
The cusps of Γ(N ) are given by the double coset,

Γ(N )\SL(2, Z)/Γ∞ (7.2.8)

acting on i∞, namely by all the images under SL(2, Z) of the cusp at infinity modulo
equivalence under Γ(N ). The number of cusps is denoted ε∞ (Γ(N )) and was given in (6.7.14).
Since every cusp may be mapped to ∞ under a suitable γ ∈ SL(2, Z), under which Gk (v|τ )
maps to Gk (vγ|γτ ), it suffices to examine the asymptotics of Gk (v|τ ) at the cusp ∞, the
asymptotics at the other cusps being deduced by mapping back with γ −1 .
The asymptotics of Gk (v|τ ) is qualitatively different for the cases where µ ≡ 0 (mod N )
or µ 6≡ 0 (mod N ). For µ 6≡ 0 (mod N ) the limit at the cusp τ → i∞ vanishes,

lim Gk (v|τ ) = 0 (7.2.9)

τ →∞

Of course, Gk (v|τ ) will in general be non-vanishing at cusps other than i∞. If µ ≡ 0 (mod N )
and ν 6≡ 0 (mod N ), then the limit at the cusp τ → i∞ is given by,
X 1 1  
lim Gk (v|τ ) = = ζ(k, Nν ) + (−)k ζ(k, 1 − ν
N
) (7.2.10)
τ →i∞ nk Nk
n≡ν (mod N )

It is shown in [13] that the Eisenstein series form a basis for the Eisenstein space, which is
defined to be the quotient,




Ek Γ(N ) = Mk Γ(N ) /Sk Γ(N ) (7.2.11)

where Mk (Γ(N )) is the vector space of all holomorphic modular forms of weight k and
Sk (Γ(N )) is the sub-space of cusp forms. For even k, the dimension of the Eisenstein space
is given by the number of cusps ε∞ (Γ(N )),



dim Ek Γ(N ) = ε∞ Γ(N ) (7.2.12)

with ε∞ (Γ(N )) given in section 6.7.3. One may consider linear combinations of the forms
Gk (v|τ ) which produce normalized Eisenstein series for Γ(N ), defined by,
X 1
Ek (v|τ ) = (7.2.13)
(c,d)≡v (N )
(cτ + d)k
gcd(c,d)=1

For even k the limit of Ek (v|τ ) vanishes at all the cusps of Γ(N ) except for the cusp corre-
sponding to v.

124
7.3 Holomorphic Eisenstein series for Γ1 (N ) and Γ0 (N )
The construction of holomorphic Eisenstein series for arbitrary congruence subgroups uses
Dirichlet characters to symmetrize the Eisenstein series for Γ(N ). As discussed in more
detail in appendix A.8, a Dirichlet character χa for a ∈ N is a map χa : Z → C which is
periodic (i.e. χa (n) = χa (n + a)), completely multiplicative (i.e. χa (nm) = χa (n)χa (m)),
and which is non-vanishing only when a and the argument n are coprime.
For any positive integer N and integer k ≥ 3, we define the set CN,k of triples (χa1 , χ
ea2 , t)
with χa1 and χ̃a2 (primitive) Dirichlet characters and t ∈ N,

χa2 (−1) = (−1)k and ta1 a2 |N

CN,k := (χa1 , χ
ea2 , t)  χa1 (−1)e (7.3.1)

In terms of Dirichlet characters, we also define the generalized divisor sums

χ ,e
χ
X
σma1 a2 (n) = χa2 (d)dm
χa1 (n/d)e (7.3.2)
d|n

as well as the delta function and Dirichlet L-series,

 ∞
1 χa1 = 1 , X χ
ea2 (n)
δ(χa1 ) = L(e
χa2 , s) = (7.3.3)
0 χa1 6= 1 , nsn=1

In terms of these ingredients, we may then construct the following functions

∞
χ ,χ ,t 2 X χa ,eχa
Ek a1 a2 (q) = δ(χ1 ) + σ 1 2 (n)q tn (7.3.4)
χa2 , 1 − k) n=1 k−1
L(e

When χa1 = χ ea2 = 1 and k ≥ 4, the generalized divisor sums reduce to the usual divisor
sums, and the L-series reduces to the Riemann-zeta function. In this case the functions
χ ,χ ,t
Ek a1 a2 (q) reduce to the usual Eisenstein series for SL(2, Z) at shifted values of q, i.e.
Ek1,1,t (q) = Ek (q t ). It is clear that for any t|N the functions Ek (q t ) should be Eisenstein series
for Γ1 (N ). What is less obvious is the following theorem, which we quote without proof,

Theorem 7.1 The space of weight-k holomorphic Eisenstein series for Γ1 (N ) is generated
χ ,χ ,t
by Ek a1 a2 (q) for (χa1 , χ
ea2 , t) ∈ CN,k , i.e.
D E
χa1 ,e
χa2 ,t
Ek (Γ1 (N )) = Ek ea2 , t) ∈ CN,k
for all (χa1 , χ (7.3.5)

for k ≥ 3. For k = 2, we have

( t
D hE2 (q) − tE2 (q ) for all t|N i E χa1 = χ
ea2 = 1
E2 (Γ1 (N )) = χ ,eχ ,t (7.3.6)
E2 a1 a2 for all (χa1 , χ
ea2 , t) ∈ CN,2 otherwise

125
 The proof can be found in [13, 62]. Note the qualitative distinction between the cases of
k = 2 and k > 2, which results from the fact that there is no weight-2 modular function
for SL(2, Z). The first line of (7.3.6) is indeed consistent with the triviality of the space of
weight-2 modular forms for N = 1 (and hence χa1 = χ ea2 = t = 1).
We now briefly discuss the case of Γ0 (N ). First note that if χa1 and χ
ea2 are two Dirichlet
characters and A is a common multiple of a1 and a2 , then it is possible to define a modulo
(A) (A)
A Dirichlet character χ
bA = χa1 χea2 , where

χa (n) (n, A) = 1
χ(A)
a (n) := (7.3.7)
0 (n, A) > 1

We may now further define the following subgroups of Ek (Γ1 (N )),

D E
χ ,χ ,t
Ek (N, χ
bN ) = Ek a1 a2 ea2 , t) ∈ CN,k and
for all (χa1 , χ χ(N ) (N )
a1 χea2 =χ
bN (7.3.8)

L
such that Ek (Γ1 (N )) = χbN Ek (N, χ
bN ). The quantity χ
bN is referred to as the “nebentypus.”
The space of modular forms of Γ0 (N ) is then simply obtained by restricting to the sector
with trivial nebentypus, i.e.

Ek (Γ0 (N )) = Ek (N, 1) (7.3.9)

Let us close with a note for the practical researcher. Holomorphic Eisenstein series for
various congruence subgroups, along with many of their properties, can be straightforwardly
implemented in the computer program SageMath [63]. For example, the simple code

> E = EisensteinForms(Gamma0(9), 4, prec = 10)

> E.eisenstein series()

can be used to output the Fourier expansions to order O(q 10 ) for all generators of weight-4
Eisenstein series of Γ0 (9). The triplets (χa1 , χ
ea2 , t) labelling each generator can be further
obtained using

> e = E.eisenstein series()

> for e in E.eisenstein series() :
print e.parameters()

In practice, using such computer algebra programs to generate the spaces of modular forms
at a given level can be much simpler than working through the definitions above.

126
7.4 Jacobi’s theorem on sums of four squares
We now provide a classic application of modular forms for congruence subgroups, here as-
sociated with Γ0 (4). For any positive integer n, we denote by rk (n) the number of ways in
which n can be written as the sum of k positive, zero, or negative integers squared,

rk (n) = # m1 , · · · , mk ∈ Z such that n = m21 + · · · + m2k


(7.4.1)

The generating function for rk (n) in powers of q = e2πiτ is the function ϑ(0|2τ )k ,
∞
m21 +···+m2k
X X
k
ϑ(0|2τ ) = q = rk (n)q n (7.4.2)
m1 ,··· ,mk ∈Z n=0

Note that the argument of ϑ has been taken to be 2τ for convenience, so that the expansion
is in terms of q raised to integer powers. The function ϑ(0|2τ )k is invariant under τ → τ + 1
but it is not invariant under τ → −1/τ for k 6= 0 and, therefore, is not a modular form under
SL(2, Z). However, it is a modular form under a certain congruence subgroup of SL(2, Z).
We specialize to the case k = 4, and propose to prove Jacobi’s theorem,
X
r4 (n) = 8 d (7.4.3)
d | n, 4 - d

with the help of modular forms. The sum is over positive divisors d of n which are not
divisible by 4. Jacobi’s theorem implies the famous theorem of Lagrange that every positive
integer can be written as the sum of four squares, but Lagrange did not evaluate the number
of ways in which this can be done.
We begin by identifying the modular subgroup under which ϑ(0|2τ )4 is a modular form.
One generator may be taken to be T : τ → τ + 1. Under τ → −1/τ we have,

ϑ(0| − 2/τ )4 = (−iτ /2)2 ϑ(0|τ /2)4 (7.4.4)

so that ϑ(0|2τ )4 clearly is not a modular form under SL(2, Z). Instead, we look for more
general modular transformations γ = ( a0 db ) ∈ SL(2, Z) under which we obtain good trans-
formation laws. A general γ ∈ SL(2, Z) may be simplified by using the fact that we have
invariance under T , and this allows us to set b = 0. But then we have ad = 1 which is solved
by a = d = 1 and a = d = −1. These cases are equivalent to one another upon reversal of
the sign of c, so that we consider the modular transformation,
τ
τ 0 = γτ = (7.4.5)
cτ + 1
under which we have,
 2   4
0 4 cτ + 1  c 1
ϑ (0|2τ ) = i ϑ 0 − − (7.4.6)
2τ 2 2τ

127
The shift by −c/2 cancels in view of T -invariance if and only if c ≡ 0 (mod 4), which
may be established by recalling that ϑ(z|τ + 1) = ϑ2 (z|τ ) from (3.8.3). Assuming now
that c ≡ 0 (mod 4) the remaining argument −1/(2τ ) of ϑ is then independent of c. Using
ϑ(0| − 1/(2τ ))4 = −4τ 2 ϑ(0|2τ )4 , we obtain,

ϑ(0|2τ 0 )4 = (cτ + 1)2 ϑ(0|2τ )4 (7.4.7)

We have now identified two transformations under which ϑ(0|2τ )4 transforms as a modular
form of weight 2, namely,
τ
T :τ →τ +1 S̃ : τ → (7.4.8)
4τ + 1
These transformations may be used to generate the group of their compositions, which is
Γ0 (4). Therefore ϑ(0|2τ )4 is a weight-2 modular form under the congruence subgroup Γ0 (4).
Of course there are no modular forms of weight two under the full modular group SL(2, Z).
So this is a simple example of the fact that, the “smaller” the modular subgroup of SL(2, Z)
is, the larger the number of modular forms it will admit for a given weight.
Now we would like to identify ϑ(0|2τ )4 with an Eisenstein series whose Fourier expansion
we can compute directly in terms of divisor functions σα (n). What are the modular forms of
weight two? Let’s start from E2 (τ ) which is not a modular form under SL(2, Z), but rather
a modular connection. Since the difference of two connections is a tensor, we investigate how
E2 (nτ ) transforms for various values of n. Under T : τ → τ + 1, the form E2 (nτ ) is invariant
for any n ∈ N. Next, examine its transformation properties under the transformation
τ
τ → τ0 = (7.4.9)
Nτ + 1
Since E2 is a derivative of ln ∆ given by (4.1.6), we examine the transformation law of ∆,
 12  
0 Nτ + 1 N 1
∆(nτ ) = − ∆ − − (7.4.10)
nτ n nτ
When n|N , the combination −N/n in the argument of ∆ cancels by T -symmetry. Taking
the transform of this simplified result, we obtain after some simplifications,

∆(nτ 0 ) = (N τ + 1)12 ∆(nτ ) (7.4.11)

Thus, for n|N , the function ∆(nτ ) is a weight 12 modular form under Γ0 (N ). Applying this
general result to the sum of four squares problem we set N = 4, so that ∆(τ ), ∆(2τ ), ∆(4τ )
are modular forms of Γ0 (4). Taking the derivatives of the logarithms on both sides, and
expressing the result in terms of E2 , we find,
4 12
E2 (nτ 0 ) = (4τ + 1)2 E2 (nτ ) + × (4τ + 1) (7.4.12)
n 2πi
128
It follows that the two linear combinations,

E2 (τ ) − 2 E2 (2τ ) E2 (2τ ) − 2 E2 (4τ ) (7.4.13)

are linearly independent modular forms under Γ0 (4) of weight 2. This matches with the
results from (7.3.9) and Theorem 7.1.
We can alternatively count the number of modular forms of weight two under Γ0 (4) by
the same arguments we used for SL(2, Z). The difference is that the fundamental domain
for Γ0 (4) is four times larger than the fundamental domain for SL(2, Z), so we would need
to derive an extension of our previous result (3.6.1) to Γ0 (4). We shall do so soon, but for
the time being will simply invoke Theorem 7.1 to claim that the space has dimension two.
The above combinations generate this space completely, and we must have,
   
ϑ(0|2τ )4 = α E2 (τ ) − 2 E2 (2τ ) + β E2 (2τ ) − 2 E2 (4τ ) (7.4.14)

for some rational coefficients α, β. Using the q-series representation for E2 ,

∞
X
E2 (nτ ) = 1 − 24 σ1 (`)q n` (7.4.15)
`=1

and matching the q 0 and q 1 terms requires 1 = −α − β and 1 = −3α respectively so that,
∞ ∞
4 1 4 X
`
X
ϑ(0|2τ ) = − E2 (τ ) + E2 (4τ ) = 1 + 8 σ1 (`)q − 32 σ1 (`)q 4` (7.4.16)
3 3 `=1 `=1

Jacobi’s theorem (7.4.3) now readily follows.

• Bibliographical notes
The books by Shimura [11] and by Diamond and Shurman [13] provide useful introductions
for modular forms for congruence subgroups. We also refer to the book by Miyake [62] and
the software data base [63]. A clear account of Jacobi’s sum of four squares theorem may be
found in Apostol’s book [8]. Our summary closely follows their presentation. Essays geared
towards applications to the proof of Fermat’s Last Theorem are collected in [64].

129
8 Modular differential equations and vector-valued
modular forms
In this section, we construct differential equations in the modular parameter τ , and find
solutions to these equations in simple cases. The solutions can generically be assembled into
“vector-valued modular forms,” which have proven fruitful in recent works in Mathematics
and Physics. In fact, we will see that in general the components of a vector-valued modular
forms are modular forms for congruence subgroups, discussed in the previous section.

8.1 Modular covariant derivatives

In section 4.1 we encountered the holomorphic function E2 (τ ), whose transformation under
SL(2, Z) is that of a modular connection, as given in (4.1.3). We may use this object to
define a modular covariant differential operator, acting on modular forms of weight k,
1 d k
Dk = − E2 (τ ) (8.1.1)
2πi dτ 12
The operator Dk maps a modular form of weight k to a modular form of weight (k + 2),

Dk : Mk → Mk+2 (8.1.2)

as was already shown in section 4.1.3.

Note that the first order differential operators introduced above can be composed to
define higher-order modular covariant derivatives,
d
Y
(d) (d)
Dk = Dk+2s−2 , Dk : Mk → Mk+2d (8.1.3)
s=1

For example, the first order differential operators on Eisenstein series give,
1 1
D4 E4 (τ ) = − E6 D6 E6 (τ ) = − E24 (8.1.4)
3 2
while for the second order differential operators we have

(2) 1 (2) 1
D4 E4 (τ ) = E24 D6 E6 (τ ) = E4 E6 (8.1.5)
6 3
The second order derivatives follow directly from the first order ones. To prove the former,
we note that the spaces of weight 6 and 8 modular forms are generated by E6 and E24
respectively, so the results are fixed up to normalization. To determine the normalization,
one can compare the first terms in the q-expansion using (3.5.13).

130
8.2 Modular differential equations
Consider now an arbitrary linear differential equation of order d in τ on a modular form f (τ )
of weight k, referred to as a modular invariant differential equation (MDE),
" d
#
(d)
X (d−r)
Dk + hr (τ )Dk f (τ ) = 0 (8.2.1)
r=1

In the above equation, hr (τ ) is an arbitrary meromorphic modular form of weight 2r and

f (τ ) has weight k. The simplest MDEs are obtained by taking the hr (τ ) to be holomorphic,
in which case they are given by polynomials in E4 (τ ) and E6 (τ ).

8.2.1 First order MDE

The generic first order MDE for modular weight k takes the form,
Dk f + h1 f = 0 (8.2.2)
for a modular function h1 of weight 2 and a modular function f of weight k. Using the
relation,
1
E2 (τ ) = ∂τ ln ∆(τ ) (8.2.3)
2πi
we obtain a simplified equation for the modular function f0 defined by,
k 1 df0
f (τ ) = ∆(τ ) 12 f0 (τ ) + h1 (τ )f0 (τ ) = 0 (8.2.4)
2πi dτ
Furthermore, since h1 is a meromorphic function of weight 2 it may be expressed as h1 (τ ) =
E6 (τ )
h (j), where h0 is a modular function of τ which may be expressed as a function of j(τ ).
E4 (τ ) 0
Using the definition of j(τ ) in terms of E4 and E6 given by j = 123 E34 /(E34 − E26 ), together
with (8.1.4), we obtain the derivative of j(τ ) respect to τ ,
1 dj E6
=− j (8.2.5)
2πi dτ E4
which by the chain rule also gives
1 df0 E6 df0 (j)
=− j (8.2.6)
2πi dτ E4 dj
Cancelling a common factor of E6 /E4 , the equation for f0 now becomes,
d ln f0 (j)
j = h0 (j) (8.2.7)
dj
which may be solved by quadrature. It remains to ensure that the local solution f (τ )
obtained via this procedure is a proper modular form of weight k. This requirement may
place conditions on the allowed values of k.

131
8.2.2 Second order MDE
Consider now the example of a second order MDE with holomorphic coefficient functions,
acting on a weight-zero modular function. The most general such MDE takes the form,

[D2 D0 + γE4 (τ )]f (τ ) = 0 (8.2.8)

for some free constant parameter γ, or more explicitly,

" 2 #
1 d 1 d
− E2 (τ ) + γE4 (τ ) f (τ ) = 0 (8.2.9)
2πi dτ 12πi dτ

The general solution to this equation may be obtained as follows. We begin by trading the
derivatives with respect to τ for derivatives with respect to the j-function, as we had done
for the first order MDE, using (8.2.5). It follows that

df E4 1 df
=− (8.2.10)
dj j E6 2πi dτ
and by similar means,
" 2 2 ! #
d2 f
  
E4 1 d E2 1 d 7j − 6912 d
= − − f (8.2.11)
dj 2 j E6 2πi dτ 6 2πi dτ 6j(j − 1728) dj

With these results the equation (8.2.9) can be rewritten as,

d2
 
1 d
j(j − 1728) 2 + (7j − 6912) + γ f (τ (j)) = 0 (8.2.12)
dj 6 dj

In terms of the variable j(τ )/1728, this equation is of standard hypergeometric type. For
generic choice of γ, there exist two distinct solutions given in closed form by,18
(1+µ)
f1 (τ ) = j(τ )− 1+µ 5+µ µ
; 1728 j(τ )−1


12 2 F1 , 12 ; 1+
12 6
(1−µ)
f2 (τ ) = j(τ )− 12 2 F1 , 12 ; 1 − µ6 ; 1728 j(τ )−1
1−µ 5−µ


12
(8.2.14)
√
where we have defined µ = 1 − 144γ.

18
When µ ∈ 6 N, the second solution is divergent, since the third argument of the hypergeometric function
is a negative integer. In this case, the second solution should be replaced by the Meijer G-function,
 2 
1 

2,0 3 −1
f2 (τ ) = G2,2 1−µ 1+µ 1728 j(τ ) (8.2.13)
12 12

132
8.2.3 Third order MDE

We may similarly analyze the case of a third order modular differential equation with holo-
morphic coefficient functions and with f of weight 0. In this case the most general modular
differential equation depends on two free parameters a and b,

[D4 D2 D0 + aE4 (τ )D0 + bE6 (τ )] f (τ ) = 0 (8.2.15)

By switching to the local coordinate j(τ ) as before, we can again re-express this as a hy-
pergeometric equation. For generic choices of parameters, there are three distinct solutions
given by
µ1 +1
f1 (τ ) = j − µ1 +1 µ1 +3 µ1 +5 µ1 −µ2
, 6 , 6 ; 6 + 1, µ1 −µ + 1; 1728 j −1


6 3 F2 3
6 6
µ2 +1
µ2 +1 µ2 +3 µ2 +5 µ2 −µ1
f2 (τ ) = j − , 6 , 6 ; 6 + 1, µ2 −µ + 1; 1728 j −1


3 F2
6 3
6 6
µ +1
− 36 µ3 +1 µ3 +3 µ3 +5 µ3 −µ2 µ3 −µ1 −1


f3 (τ ) = j 3 F2 6
, 6
, 6
; 6
+ 1, 6
+ 1; 1728 j (8.2.16)

Here we have defined µi as

µ1 = 4x1 − 2x2 − 2x3

µ2 = 4x2 − 2x3 − 2x1
µ3 = 4x3 − 2x1 − 2x2 (8.2.17)

and xi , i = 1, 2, 3 as the three roots of the cubic equation

x3 − 12 x2 + a + 1


18
x+b=0 (8.2.18)

8.2.4 Example

One interesting example is when the parameters in (8.2.15) are taken to be

107 23
a=− b= (8.2.19)
2304 55296

In this case the exact solutions given above have the following q-expansions
23
f1 (τ ) = q 48 1 + q + q 2 + q 3 + 2q 4 + 2q 5 + . . .


1
f2 (τ ) = q − 48 1 + q 2 + q 3 + 2q 4 + 2q 5 + . . .


1
f3 (τ ) = q 24 1 + q + q 2 + 2q 3 + 2q 4 + 3q 5 + . . .


(8.2.20)

133
To all order in q, these match with the following alternative closed-form expressions,
s s !
1 ϑ3 (0|τ ) ϑ4 (0|τ )
f1 (τ ) = −
2 η(τ ) η(τ )
s s !
1 ϑ3 (0|τ ) ϑ4 (0|τ )
f2 (τ ) = +
2 η(τ ) η(τ )
s
ϑ2 (0|τ )
f3 (τ ) = (8.2.21)
η(τ )

Indeed, these can be checked to be the solutions to the MDE in question.

8.2.5 Modular invariance of solution spaces

Because MDEs are by definition modular invariant, so too are their solution spaces. This
means that under SL(2, Z) the solutions must transform into linear combinations of them-
selves. It is useful to organize the solutions of an order-d MDE into a d-vector, in which case
the S and T generators of SL(2, Z) can be represented by d × d matrices. For example, in
the concrete case of (8.2.21) we can organize the solutions into a three-dimensional vector
f (τ ) = (f1 (τ ), f2 (τ ), f3 (τ ))t , and then from the known transformation properties of ϑi (0|τ )
and η(τ ) we obtain T and S as
 23πi
  
e 24 0 0 1 1 −1
πi 1
T = 0 e− 24 0  S= 1 1 1 (8.2.22)
 
πi 2
0 0 e 12 −2 2 0

This leads us to a fruitful generalization of the notion of modular forms, known as vector-
valued modular forms.

8.3 Vector-valued modular forms

A d-dimensional vector-valued modular form f (τ ) = (f1 (τ ), . . . , fd (τ ))t of weight k is defined
by its modular transformation properties,
 
k a b
f (γτ ) = (cτ + d) ρ(γ) f (τ ) γ= ∈ SL(2, Z) (8.3.1)
c d

where ρ : SL(2, Z) → GL(d, C) is a d × d matrix representation. f (τ ) is further required to

be holomorphic in the upper-half plane and to have at most finite-order poles at the cusps.

134
8.3.1 Examples
We begin with some simple examples of vector-valued modular functions. We have already
seen that solutions to an order-d MDE form a d-dimensional vector-valued modular form,
and in the section 8.3.3 we will show that the converse is also true, i.e. the components of a
d-dimensional vector-valued modular form are always solutions to an order-d MDE. In this
subsection however we will momentarily ignore the connection to MDEs.
The Jacobi ϑ-functions at z = 0 form a 3-dimensional vector-valued modular form
(ϑ2 (0|τ ), ϑ83 (0|τ ), ϑ84 (0|τ )). We have taken the eighth power to eliminate potential phases
8

from the modular transformations given in (3.8.7). Under S and T transformations one has,

ϑ82 (0|τ + 1) = ϑ82 (0|τ ) ϑ82 (0| − 1/τ ) = τ 4 ϑ84 (0|τ )

ϑ83 (0|τ + 1) = ϑ84 (0|τ ) ϑ83 (0| − 1/τ ) = τ 4 ϑ83 (0|τ )
ϑ84 (0|τ + 1) = ϑ83 (0|τ ) ϑ84 (0| − 1/τ ) = τ 4 ϑ82 (0|τ ) (8.3.2)

Clearly these act as permutations of the three elements. In the usual permutation notation,
the T transformation corresponds to the transformation (132), while the S transformation
corresponds to (321). Together, these generate the group S3 of permutations of three ele-
ments. One way of saying this is that there is a surjective homomorphism ρ : SL(2, Z) → S3 .
It is interesting to ask what the kernel of the homomorphism ρ is. That is to say, we are
interested in the group defined by,
   
a b

 8 4 8
Ker(ρ) = γ = ∈ SL(2, Z)  ϑi (0|γτ ) = (cτ + d) ϑi (0|τ ), i = 2, 3, 4 (8.3.3)
c d
 1/2 
Recalling the
 0  notation in terms of half-integer characteristics ϑ2 = ϑ 0
, ϑ3 = ϑ [ 00 ] , and
ϑ4 = ϑ 1/2 , from (3.8.7) we see that elements of Ker(ρ) must have,

aα + cβ ≡ α (mod 1) bα + dβ ≡ β (mod 1) (8.3.4)

for any choice (α, β) ∈ (0, 0), 12 , 0 , 0, 12 . For α = β = 0 the equations are trivially


satisfied. Taking α = 12 and β = 0, the two equations above require,

a ≡ 1 (mod 2) b ≡ 0 (mod 2) (8.3.5)

1
while taking α = 0 and β = 2
require,

c ≡ 0 (mod 2) d ≡ 1 (mod 2) (8.3.6)

We thus conclude that Ker(ρ) is given by

      
a b  a b 1 0
Ker(ρ) = ∈ SL(2, Z) ≡ (mod 2) (8.3.7)
c d c d 0 1


135
In other words, we find that Ker(ρ) is equal to the principal congruence subgroup Γ(2).
Thus while the vector (ϑ82 (0|τ ), ϑ83 (0|τ ), ϑ84 (0|τ )) transforms as a vector-valued modular form
in a non-trivial permutation representation of SL(2, Z), we see that each of the individual
components of this vector is itself a genuine modular form under Γ(2).
Proceeding to a four-dimensional example, we consider the vector f (τ ) with components,
 τ 24
f1 (τ ) = 312 η(3τ )24 f2 (τ ) = η
3
 24  24
τ +1 τ +2
f3 (τ ) = η f4 (τ ) = η (8.3.8)
3 3

Under S and T modular transformations, we have the following

f1 (τ + 1) = f1 (τ ) f1 (−1/τ ) = τ 12 f2 (τ )
f2 (τ + 1) = f3 (τ ) f2 (−1/τ ) = τ 12 f1 (τ )
f3 (τ + 1) = f4 (τ ) f3 (−1/τ ) = τ 12 f4 (τ )
f4 (τ + 1) = f2 (τ ) f4 (−1/τ ) = τ 12 f3 (τ ) (8.3.9)

The generators S and T again permute elements of the vector; in permutation notation,
they correspond to (1342) and (2143), respectively. However, now these transformations do
not generate the full group S4 of permutations of four elements. Instead, we only obtain
the even permutations, i.e. we generate the alternating group A4 . So in this case we obtain
a surjective homomorphism ρ : SL(2, Z) → A4 . We may again ask about the kernel of ρ.
Demanding invariance of f1 (τ ) under ( ac db ) ∈ SL(2, Z) is equivalent to requiring,
   −1  
3 0 a b 3 0 a 3b
= ∈ SL(2, Z) (8.3.10)
0 1 c d 0 1 c/3 d

which in particular requires that c ∈ 3Z. Similarly, demanding invariance of f2 (τ ) requires,

   −1  
1 0 a b 1 0 a b/3
= ∈ SL(2, Z) (8.3.11)
0 3 c d 0 3 3c d

which requires b ∈ 3Z. Any permutation in A4 which fixes f1 (τ ) and f2 (τ ) also fixes f3 (τ )
and f4 (τ ). We conclude that the kernel consists of all matrices for which b ≡ c ≡ 0 (mod 3).
This includes in particular the element −I ∈ SL(2, Z). Hence the kernel is not quite Γ(3),
but rather ±Γ(3), where we use the shorthand ±Γ(N ) = Γ(N ) × {±I}. So we have found
that while the vector f (τ ) transforms as a vector-valued modular form in an alternating
representation of SL(2, Z), each of the individual components of this vector is itself modular
under the group ±Γ(3).

136
 For completeness, let us record here some of the isomorphisms which we have seen, as
well as some we have not yet seen,

SL(2, Z)/Γ(2) ∼
= S3 SL(2, Z)/ ± Γ(3) ∼
= A4
SL(2, Z)/ ± Γ(4) ∼
= S4 SL(2, Z)/ ± Γ(5) ∼
= A5 (8.3.12)

These identities can be checked to be consistent with the index [SL(2, Z); Γ(N )], or equiva-
lently with the order |SL(2, ZN )|, as given in (6.2.5) and (6.3.9). Indeed, from those formulas
one finds

[SL(2, Z); Γ(2)] = 6 [SL(2, Z); Γ(3)] = 24

[SL(2, Z); Γ(4)] = 48 [SL(2, Z); Γ(5)] = 120 (8.3.13)

which can be compared with the orders of the relevant permutation and alternating groups,

|S3 | = 6 |A4 | = 12 |S4 | = 24 |A5 | = 60 (8.3.14)

For N = 2 we see a match, whereas for N > 2 there is a factor of two difference, stemming
from the fact that we are quotienting by ±Γ(N ) as opposed to Γ(N ).

8.3.2 Integrality and Γ(N )

In both of the above examples, we found that the components of the vector-valued modular
forms were themselves modular under Γ(N ) for some appropriate N . This is not true in
general, and it is interesting to ask when it holds. The following result has been proven
for 2- and 3-dimensional vector-valued modular forms, and is conjectured to hold more
generally:19

Conjecture 8.1 (Integrality Conjecture) Consider a component fi (τ ) of a vector-valued

modular form with Fourier expansion
(i) X
fi (τ ) = q n0 a(i)
n q
n
(8.3.15)
n≥0

(i) (i)
with n0 = pi /Ni and (pi , Ni ) = 1. If all coefficients an are algebraic integers, then fi (τ ) is
a modular form for Γ(Ni ).

The Integrality Conjecture states that a vector-valued modular form transforming in a

representation ρ is integral (i.e. each component of the vector has integer Fourier coefficients)

19
In the context of vector-valued modular functions realized by the characters of RCFTs, which is the case
of principal interest in Physics, this conjecture has been proven in [65].

137
only if Ker(ρ) contains a principal congruence subgroup Γ(N ). Indeed, the two examples
given above had integer Fourier coefficients, and are directly in line with this conjecture.
As another example, the three-dimensional vector-valued modular function given in (8.2.21)
has all integer Fourier coefficients, and indeed the components can be seen to be modular
functions for Γ(48).
One common place in Physics where vector-valued modular forms appear is as characters
of CFTs. Since the Fourier coefficients in that case are interpretable as physical degeneracies,
they must of course be algebraic integers (and in fact rational integers). The conjecture then
tells us that CFT characters are always modular functions for some Γ(N ). Going back to
the example of (8.2.21), the functions in this case are actually characters for the Ising CFT,

f1 (τ ) = χ (τ ) f2 (τ ) = χ1 (τ ) f3 (τ ) = χσ (τ ) (8.3.16)

8.3.3 Relation to MDEs

We now show that the components of a d-dimensional vector-valued modular form are always
solutions to an order-d MDE. To see this, we begin by constructing the Wronskians Wr
associated to f (τ ),
 
f1 ... fd
 D(1) f ...
(1)
Dk fd 
 k 1 
 .. .. 

 . . 

D(r−1) f ...
(r−1)
Dk fd 
1
Wr = det  k(r+1) (8.3.17)
 
(r+1) 
Dk f1 ... Dk fd 
 .. .. 

 . . 

 (d−1) (d−1) 
Dk f1 ... Dk fd 
(d) (d)
Dk f1 ... Dk fd
 
Wd−1
for r = 1, . . . , d. From the trivial identity Wd−1 − Wd
Wd = 0, we obtain the following
equation,
 
f1 ... fd
(1) (1)

 Dk f1 ... Dk fd 

 .. .. 
det 
 . . =0
 (8.3.18)
(d−2) (d−2)

 Dk f1 ... Dk fd 

(d) Wd−1 (d−1) (d) Wd−1 (d−1)
Dk f1 − Wd Dk f1 ... Dk fd − Wd Dk fd

138
For this determinant to vanish, the bottom row must be a linear combination of the other
rows, which tells us that
d
(d) Wd−1 (d−1) X (d−r)
D k fi − Dk fi + hr (τ )Dk fi = 0 (8.3.19)
Wd r=2

This is of the usual MDE form (8.2.1), where in particular we have the coefficient function

Wd−1
h1 (τ ) = − (8.3.20)
Wd
(d)
In fact, by plugging the expression for Dk fi obtained from (8.3.19) into (8.3.17), we may
express all of the coefficient functions in terms of Wronskians, giving
Wd−r
hr (τ ) = (−1)r (8.3.21)
Wd

Say that the function fi (τ ) is bounded in the interior of the fundamental domain, and
has zeroes at locations p of order ordfi (p). We define the so-called Wronskian index by
!
1 1 X
` := 6 ordWd (i) + ordWd (ρ) + ordWd (p) (8.3.22)
2 3 p∈F

so that `/6 is the number of zeroes of Wd . The Wronskian index can take any non-negative
integer value except 1. We note for future use that the valence formula (3.6.1) implies

` d(k + d − 1)
ordWd (i∞) + = (8.3.23)
6 12
where we have used the fact that the Wronskian has modular weight d(k + d − 1).
For ` = 0, the coefficient functions hr (τ ) are holomorphic, which is the case analyzed
in detail for d = 2, 3 in the previous section. For ` > 0 on the other hand, the coefficient
functions will generically be singular. For example, consider the case for which d = ` = 2.
By definition (8.3.22) we conclude that ordW2 (ρ) = 1, and hence W2 ∼ E4 (τ ). The coefficient
functions then take the form hr (τ ) ∼ E41(τ ) W2−r . The remaining τ dependence can be fixed
by simply requiring that one has the correct modular weight, giving h1 (τ ) ∼ E4 (τ ) and
h2 (τ ) ∼ E6 (τ )
E4 (τ )
.
Finding a solution to an MDE with meromorphic coefficients is often significantly more
difficult than for its holomorphic counterpart (though for d = 2 and ` < 6 it turns out that
the solutions can again be written as hypergeometric functions). As we discuss in section
10.9, Hecke operators can be used to generate such solutions.

139
• Bibliographical notes
Solving modular differential equations by recasting them as hypergeometric equations has
been discussed in [66, 67, 68]. Vector-valued modular forms grew out of work of Gunning
[69], as well as Selberg [70], who used them as a way of studying Fourier coefficients of
modular forms for non-congruence subgroups. For a modern review of various properties of
vector-valued modular forms see [71], and for a discussion of modular forms and differential
operators see [72]. The integrality conjecture given in section 8.3.2 was first put forward
in [73], and has since been proven for two-dimensional and three-dimensional vector-valued
modular forms in [74, 75] and [76, 77], respectively. It has also been proven for vector-valued
modular forms whose entries can be interpreted as the characters for RCFTs [65]. Its validity
in a more general context has been discussed in e.g. [67, 78].
Finally, we note that modular differential equations have been applied to the classification
of rational conformal field theories, starting with the seminal work in [79, 80], together
with follow-up papers [81, 82] shortly thereafter. In subsequent works modular differential
techniques were applied to the classification of bosonic rational conformal fields theories with
two characters at Wronskian index ` = 2, 4 [83, 84, 85], and to the classification of theories
with three characters at ` = 0 [86]; for a recent review of some of this progress, see [87]. As of
the writing of these lecture notes, the state-of-the-art for the classification of bosonic rational
conformal field theories is for theories with up to five characters (subject to restrictions on
the Wronskian index), obtained in [88]. For fermionic theories on the other hand, there is
a complicating factor that the relevant modular differential equations need not be invariant
under SL(2, Z), but only an appropriate index 2 congruence subgroup. Works towards the
classification of fermionic theories with small numbers of characters have appeared in [89, 90].

140
9 Modular graph functions and forms
Modular graph functions and forms map certain (decorated) graphs to complex-valued func-
tions on the Poincaré upper half-plane H with definite transformation properties under
SL(2, Z). The terminology of modular graph functions versus forms follows the terminol-
ogy adopted for automorphic functions and forms in the holomorphic and non-holomorphic
categories discussed in subsection 7.1.2. Thus, modular graph functions may be viewed
as SL(2, Z)-invariant functions on H, while modular graph forms may be identified with
SL(2, Z)-invariant differential forms.
To be more precise, we introduce the spaces Mk,` of complex-valued smooth functions
f : H → C with the following SL(2, Z) transformation properties,
   
k ` a b
Mk,` = f (γτ ) = (cτ + d) (cτ̄ + d) f (τ ) for all γ = ∈ SL(2, Z) (9.0.1)
c d
for k, ` ∈ C, subject to the condition k − ` ∈ Z. This condition is imposed to ensure that the
functions are single-valued. The functions of interest to us here will have at most polynomial
growth at the cusp τ2 → ∞, and we shall impose this property also on the functions in Mk,` .
A modular graph form of modular weight (k, `) is then a map from a certain (decorated)
graph to Mk,` . For the special case k = `, one refers to such maps as modular graph
functions. For k, ` ∈ Z, we may equivalently view the functions in Mk,` as differential forms
of weight ( k2 , 2` ).
We shall establish that the non-holomorphic Eisenstein series introduced in section 4.2
may be associated with simple one-loop graphs and thus represent a special class of modular
graph functions. Eisenstein series and modular graph functions and forms beyond Eisenstein
series occur naturally and pervasively in the study of the low energy expansion of superstring
amplitudes, as will be made clear in section 12. Here we shall present a purely mathematical
approach with only minimal reference to Physics.

9.1 One-loop modular graph functions are Eisenstein series

We begin by considering the scalar Green function G(z − w|τ ) on the torus Σ = C/Λ for a
lattice Λ = Z+τ Z with modulus τ ∈ H and z, w ∈ Σ. The scalar Green function was defined
as the inverse of the scalar Laplacian on the torus on the space of functions orthogonal to
constant functions in (5.5.12), solved in terms of Jacobi ϑ-functions in (5.5.14), and shown
to be invariant under SL(2, Z) in (5.5.19). Here we shall use its expression in terms of a
double Fourier series or Kronecker-Eisenstein series, given in (5.5.8) and expressed here in
terms of a sum over the lattice Λ0 = Λ \ {0},
X τ2
G(z|τ ) = e2πi(nx−my) (9.1.1)
π|m + nτ |2
(m,n)6=(0,0)

141
where z = x+τ y and x, y are real “co-moving coordinates” valued in R/Z. By concatenating
the Green function repeatedly using convolution, we may construct an infinite family of
modular invariant functions, which may be defined recursively as follows,
Z 2
du
Gs (z|τ ) = G(z − u|τ )Gs−1 (u|τ ) (9.1.2)
Σ τ2

where we set G1 = G and s is an arbitrary positive integer. To the Green function G and its
generalizations Gs for positive integer s it is natural to associate a graph, which physicists
refer to as a Feynman graph,

G(z − w|τ ) =
z w

Gs (z − w|τ ) =
z u1 u2 u3 us−1 w (9.1.3)

The black dots represent points u1 , · · · us−1 that are integrated over Σ, while the white
dots represent given points in Σ, that are not integrated. Upon Fourier transformation,
convolution acts by multiplication, and we obtain the following representation for Gs ,
X τ2s
Gs (z|τ ) = e2πi(nx−my) (9.1.4)
π s |m + nτ |2s
(m,n)6=(0,0)

where z = x + τ y. As long as s ≥ 2, we may set the point z = 0 which graphically is

equivalent to closing the open chain in (9.1.3) into the closed loop of (9.1.5),

u3 u2

u4 u1
Es (τ ) =

us−1 us
(9.1.5)

Comparing to (4.2.1), we see that Gs (0|τ ) = Es (τ ), the non-holomorphic Eisenstein series.

Actually, the double sum in (9.1.4) converges absolutely for any s ∈ C with Re (s) > 1, but
the graphical interpretation no longer makes sense when s is not a positive integer.
When s is integer and satisfies s ≥ 2, the number s counts the number of edges in the
graph (9.1.5) and is referred to as the weight of the graph (not to be confused with the
modular weight of the corresponding modular graph function). In subsection 12.7 we shall
identify this definition of the weight with that of transcendental weight.

142
9.2 Maass operators and Laplacians
Before studying higher-loop modular graph functions in detail, let us give a brief intermission
on differential operators on the space Mk,` . There exist several natural maps between the
spaces Mk,` which we shall now exhibit.
One noteworthy element of M−1,−1 is the imaginary part τ2 of the modulus τ , whose
powers satisfy τ2s ∈ M−s,−s . Multiplication by a suitable power of τ2 therefore provides a
map between the different spaces Mk,` . Since a function f (τ ) and its image τ2s f (τ ) under this
map are essentially equivalent to one another, we may introduce the equivalence relations,

Mk,` ≈ Mk−`,0 ≈ M0,`−k (9.2.1)

This equivalence is particularly convenient when it comes to the action of partial differential
operators. The Cauchy-Riemann operator ∂τ̄ acts covariantly on the spaces Mk,0 without
the need for a connection and maps to the space Mk,2 . Similarly, the operator ∂τ maps
M0,` covariantly to the space M2,` without the need for a connection. One may, of course,
include a k- and `-dependent connection to obtain the Maass operators (see appendix C.4).
We shall circumvent doing so here by applying ∂τ̄ to Mk,0 and ∂τ to M0,` only, using the
equivalence of (9.2.1). Actually, we can multiply the Cauchy-Riemann operators by suitable
powers of τ2 so that the operators ∇ = 2iτ22 ∂τ and ∇¯ = −2iτ 2 ∂τ̄ map the spaces as follows,
2

¯ : Mk,0 → Mk−2,0
∇
∇ : M0,` → M0,`−2 (9.2.2)

Laplace operators may be defined as follows,20

−k k−2 ¯
∆+
k : Mk,0 → Mk,0 ∆+
k = τ2 ∇ τ2 ∇
−` ¯ `−2
∆−
` : M0,` → M0,`
−
∆ = τ ∇τ
` 2 ∇2 (9.2.3)
−
On modular forms of weight (0, 0) the Laplace operators ∆+ 0 and ∆0 coincide and are denoted
by ∆ = 4τ22 ∂τ ∂τ̄ . Famously, the Eisenstein series satisfies the Laplace eigenvalue equation,

∆Es = s(s − 1)Es (9.2.4)

Its expansion at the cusp consists of two power behaved terms, plus an infinite number of
exponentially suppressed terms, given in (4.2.16), and which we recall here for convenience,

ζ(2s) s Γ(s − 12 )
Es (τ ) = 2 τ + 2 ζ(2s − 1) τ21−s + O(e−2πτ2 ) (9.2.5)
πs 2 s− 12
π Γ(s)

20
Here we define the Laplacians as negative operators, while in appendix C.4 they are positive operators.

143
In terms of the independent variables τ, τ̄ and the real co-moving coordinates x, y, the Laplace
operator ∆ also has Gs (x + τ y|τ ) as an eigenfunction with eigenvalue independent of x, y,

∆Gs (x + τ y|τ ) = s(s − 1)Gs (x + τ y|τ ) (9.2.6)

Application of the operator ∇k to Es produces modular graph forms of weight (0, −2k),

k Γ(s + k) X τ2s+k
∇ Es (τ ) = (9.2.7)
Γ(s) π s (m + nτ )s+k (m + nτ̄ )s−k
(m,n)6=(0,0)

These functions coincide with Zagier’s elliptic polylogarithm, up to an overall normalization.

9.3 Two-loop modular graph functions

We now investigate the special class of modular graph functions in M0,0 associated with
arbitrary (decorated) two-loop graphs. A connected two-loop graph has two trivalent vertices
and an arbitrary number of bivalent vertices, distributed on the three concatenated edges
connecting the trivalent vertices at z and w according to three positive integers a1 , a2 , a3 .
The most general such graph is depicted in (9.3.1),
a1
• • •
z a2 w
Ca1 ,a2 ,a3 (τ ) = • • • • •
a3
• • •
(9.3.1)

The associated modular graph function may be defined in terms of a double integral over Σ
of a triple product of concatenated Green functions Gs ,
Z 2 Z 2
dz dw
Ca1 ,a2 ,a3 (τ ) = Ga1 (z − w|τ )Ga2 (z − w|τ )Ga3 (z − w|τ ) (9.3.2)
Σ τ2 Σ τ2

Translation invariance on the torus renders one of the integrals redundant, and we have the
following equivalent expression,
Z 2
dz
Ca1 ,a2 ,a3 (τ ) = Ga1 (z|τ )Ga2 (z|τ )Ga3 (z|τ ) (9.3.3)
Σ τ2

Upon Fourier transforming, the corresponding modular graph function is given by the fol-
lowing Kronecker-Eisenstein sums,
0 3 3 3  ar
X X  X Y τ2
Ca1 ,a2 ,a3 (τ ) = δ mr δ nr (9.3.4)
mr ,nr ∈Z r=1 r=1 r=1
π|mr + nr τ |2
r=1,2,3

144
The sums are absolutely convergent provided ar + as > 1 for all pairs r 6= s. The weight
is given by w = a1 + a2 + a3 and is required to satisfy w ≥ 3 for convergence. The
Kronecker-Eisenstein sums are absolutely convergent for complex values of a1 , a2 , a3 as long
as Re (a1 ), Re (a2 ), Re (a3 ) > 1, but contrarily to the case of the Eisenstein series, the problem
of the existence of analytic continuations in these variables remains unexplored.
Two key properties of the non-holomorphic Eisenstein series, namely that they satisfy
a differential equation, and have a simple expansion at the cusp, have more complicated
counterparts in two-loop modular graph functions.
Theorem 9.1 Two-loop modular graph functions Ca1 ,a2 ,a3 of weight w = a1 + a2 + a3 obey
a system of differential equations of uniform weight w

2∆Ca1 ,a2 ,a3 = 2a1 a2 Ca1 +1,a2 −1,a3 + a1 a2 Ca1 +1,a2 +1,a3 −2 − 4a1 a2 Ca1 +1,a2 ,a3 −1
+a1 (a1 − 1) Ca1 ,a2 ,a3 + 5 permutations of (a1 , a2 , a3 )

where ∆ = 4τ22 ∂τ ∂τ̄ . When a1 , a2 , a3 ∈ N, the system of differential equations for a given
weight truncates to a finite-dimensional linear system of inhomogeneous equations where the
inhomogeneous part consists of a linear combination of Ew and Ew1 Ew2 with w = w1 + w2
and w1 , w2 ≥ 2, as a consequence of the following relations,

Cw1 ,w2 ,0 = Ew1 Ew2 − Ew

Cw1 +1,w2 ,−1 = Ew1 Ew2 + Ew1 +1 Ew2 −1 (9.3.5)

The eigenvalues of the linear system are of the form s(s − 1) where 1 ≤ s ≤ w − 2 and
w − s ∈ 2N and have multiplicity [(s + 2)/3].
For example, up to weight 5, the relations of Theorem 9.1 are given as follows,

∆C1,1,1 = 6E3
(∆ − 2)C2,1,1 = 9E4 − E22
(∆ − 6)C3,1,1 − 3C2,2,1 = 16E5 − 4E2 E3
∆C2,2,1 = 8E5 (9.3.6)

Note that the first and last identities imply C1,1,1 − E3 is constant, as must be 5C2,2,1 − 2E5 .
These constants may be determined from the asymptotic behavior of these functions at the
cusp, a problem to which we now turn.

Theorem 9.2 The expansion of Ca1 ,a2 ,a3 (τ ) near the cusp τ → i∞ is given by a Laurent
polynomial in τ2 of degree (w, 1 − w)
w−1
c2−w X cw−2k−1 ζ(2k + 1)
w
Ca1 ,a2 ,a3 (τ ) = cw (−4πτ2 ) + w−2
+ 2k+1−w
+ O(e−2πτ2 ) (9.3.7)
(4πτ2 ) k=1
(4πτ2 )

145
where cw , cw−2k−1 ∈ Q and c2−w is a linear combination with integer coefficients of products
of odd zeta-values,
w−2
X
c2−w = γk ζ(2k + 1)ζ(2w − 2k − 3) γk ∈ Z (9.3.8)
k=1

A consequence of this theorem is that no multiple zeta-values occur in the Laurent polyno-
mial of two-loop modular graph functions. It was shown by Zerbini that higher loop modular
graph functions do involve irreducible multiple zeta values.
At every odd weight, there is one linear combination of the above differential identities
in the subspace of vanishing eigenvalue whose inhomogeneous part does not involve bilinears
in the Eisenstein series. This property may be verified in the identities for C1,1,1 and C2,2,1
in (9.3.6), and holds for all odd weights. Each one of these identities may be integrated up
to an additive constant which may be fixed by the asymptotics near the cusp, and we find,
3 ζ(7)
C1,1,1 = E3 + ζ(3) C3,3,1 + C3,2,2 = E7 +
7 252
2 ζ(5) ζ(9)
C2,2,1 = E5 + 9C4,4,1 + 18C4,3,2 + 4C3,3,3 = 4E7 + (9.3.9)
5 30 240
The equation for C1,1,1 was derived by Zagier in an unpublished note by direct summation
of the Kronecker-Eisenstein series which defines C1,1,1 .

9.4 General modular graph functions and forms

At higher loops, it is no longer true that the Laplace operator in τ , (acting on the immediate
generalization of two-loop modular graph functions to higher loops) closes onto functions
defined by Kronecker-Eisenstein series in which the exponents of holomorphic momenta
mr + nr τ and their complex conjugate coincide. It is natural to also lift the restriction to
modular functions, and consider more generally modular graph forms which have non-trivial
modular weights. The required generalization may be defined as follows.
The connectivity matrix Γ of an arbitrary graph has components Γv r where the index
v = 1, · · · , V labels the vertices of the graph and the index r = 1, · · · , R labels its edges. No
edge is allowed to begin and end on the same vertex. When edge r contains vertex v we have
Γv r = ±1, while otherwise we have Γv r = 0. We shall view the pair of exponents (ar , br )
of the momenta pr = mr + nr τ and its complex conjugate p̄r through edge r as providing a
decoration of the graph. The decorated graph is then specified by the connectivity matrix Γ
and the pair of arrays (A, B) defined by,

A = [a1 , · · · , , aR ] a = a1 + · · · + aR
B = [b1 , · · · , , bR ] b = b1 + · · · + bR (9.4.1)

146
where ar , br ∈ C with ar − br ∈ Z for all r = 1, · · · , R. The pair of total exponents (a, b)
generalizes the weight to the full decorated graph.
To a decorated graph (Γ, A, B) we associate a complex-valued function on H, defined by
the following Kronecker-Eisenstein sum, whenever this sum is absolutely convergent,

R V R
!
 
A  τ  21 a+ 12 b 1
2
X Y Y X
CΓ (τ ) = δ Γv s ps (9.4.2)
B π (p r )ar (p̄ )br
r
p1 ,...,pR ∈Λ0 r=1 v=1 s=1

The Kronecker δ-symbol equals 1 when its argument vanishes as an element of the lattice
Λ, and equals 0 otherwise. The number of loops L is the number of independent momenta,
given by L = R − V + 1. For any given decorated graph (Γ, A, B), the domain of absolute
convergence of the sums in (9.4.2) is given by a system of inequalities on the combinations
Re (ar + br ). Outside of the domain of convergence, it is an open question as to whether
and how the functions C of (9.4.2) may be defined by analytic continuation in the variables
ar + br . The function C in (9.4.2) vanishes whenever the integer a − b is odd, or whenever
it is associated with a graph Γ which becomes disconnected upon severing a single edge. A
function C associated with a graph Γ which is the union of two graphs Γ = Γ1 ∪ Γ2 , such
that the intersection of Γ1 and Γ2 consists of a single vertex, factorizes into the product of
functions C for Γ1 and Γ2 , with the corresponding partitions of the exponents.

9.4.1 Modular properties

Under SL(2, Z) the functions defined in (9.4.2) transform as follows,

     1 a− 1 b  
A aτ + b cτ + d 2 2 A
CΓ = CΓ (τ ) (9.4.3)
B cτ + d cτ̄ + d B

where a, b, c, d ∈ Z and ad − bc = 1. One defines the modular weight of C by the pair

( 12 (a − b), 21 (b − a)) which is integer-valued for a − b even (C vanishes for odd a − b). As
we have explained above, we refer to C as a modular graph form when a 6= b and a modular
graph function when a = b. For a 6= b the normalization factor of τ2 is not canonical, and
we introduce the forms C + and C − of modular weight (0, b − a) and (a − b, 0) respectively
as follows,
   
A ± a−b A
CΓ± = (τ2 ) 2 CΓ (τ ) (9.4.4)
B B

CΓ+ transforms with a factor (cτ̄ + d)−a+b while CΓ− transforms with a factor (cτ + d)a−b .

147
9.4.2 Examples: Eisenstein series and dihedral graphs
In this newfound notation, we may represent and generalize some of the examples of modular
graph functions and forms encountered earlier. The Eisenstein series and its successive
derivatives take the form,
   
s 0 k Γ(s + k) + s + k 0
Es = CΓ ∇ Es = CΓ (9.4.5)
s 0 Γ(s) s−k 0
where Γ is a connected one-loop graph with only bivalent vertices. The two-loop modular
graph functions Ca1 ,a2 ,a3 may be generalized to (n − 1)-loop modular graph functions,
 
a1 · · · an
Ca1 ,··· ,an = CΓ (9.4.6)
a1 · · · an
where the “dihedral” graph Γ contains two vertices of valence n + 1 in addition to the
remaining vertices which are all bivalent. For n = 3 we recover the two-loop modular graph
functions studied earlier.

9.4.3 Examples: holomorphic modular graph forms

When all anti-holomorphic exponents bi vanish identically the corresponding modular graph
form reduces to the following expression,
  R V R
!
a · · · a X Y 1 Y X
CΓ− 1 n
(τ ) = δ Γ v s ps (9.4.7)
0 ··· 0 0
(p r )ar
1p ,...,p ∈Λ r=1
R v=1 s=1

The multiple sum is absolutely convergent provided ar ≥ 2 for all r = 1, · · · , R ≥ 2. In this

case the result is holomorphic in τ and a modular form of weight (a, 0) with a = a1 +· · · , +aR .
When CΓ− is a modular form of modular weight (a, 0), it may be expressed as a polynomial in
the holomorphic modular forms E4 and E6 , and therefore produces a map from graphs Γ to
holomorphic modular forms, or equivalently to polynomials in E4 and E6 . The nature of this
map, which is much simpler than the map to non-holomorphic modular forms, remains to be
investigated. Under weaker conditions of conditional convergence the sum must be suitably
regularized. Maintaining holomorphicity leads to a violation of full modular invariance, such
as is the case with a1 = 2, a2 = 0 for R = 2 which yields the quasi-modular form E2 .

9.4.4 Algebraic relations

Under complex conjugation, any of these functions have their arrays of holomorphic and
anti-holomorphic exponents swapped,
   
A ∗ B
CΓ (τ ) = CΓ (τ ) (9.4.8)
B A

148
In addition, modular graph forms obey momentum conservation identities,
R   XR  
X A − Sr A
Γk r CΓ = Γ k r CΓ =0 (9.4.9)
B B − Sr
r=1 r=1

where the R-dimensional row-vector Sr is defined to have zeroes in all slots except for the
r-th, which instead has value 1,

Sr = [0, . . . , 0, 1, 0, . . . , 0] (9.4.10)
| {z } | {z }
r−1 R−r

These identities simply express the fact that the lattice momenta entering each vertex v
add up to zero or, as physicists would state, that momenta is conserved at each vertex as a
consequence of translation invariance on the torus.

9.4.5 Differential relations

The action of the Cauchy-Riemann operator ∇ = 2iτ22 ∂τ on modular graph forms C + defined
and normalized in (9.4.4) for arbitrary exponents A, B is given by,
  XR  
A + A + Sr
∇CΓ+ = ar C (9.4.11)
B B − Sr
r=1

where Sr was defined in (9.4.10). The action of the Laplace operator ∆ = 4τ22 ∂τ̄ ∂τ on
modular graph functions (for which a = b) is given by,
  XR  
A A + Sr − Ss
(∆ + a) CΓ = ar b s C (9.4.12)
B B − Sr + Ss
r,s=1

It is similarly possible to define the action of the Laplace operator on modular graph forms
of arbitrary modular weight with a 6= b, but we shall not need them here.

9.4.6 Higher loop algebraic identities

Higher-loop modular graph functions and forms satisfy systems of algebraic and differential
identities which generalize those found for two-loops. For example, one has the following
algebraic identities involving modular graph functions of weight 4 and 5,

C1,1,1,1 = 24C2,1,1 + 3E22 − 18E4

C1,1,1,1,1 = 60C3,1,1 + 10E2 C1,1,1 − 48E5 + 16ζ(5)
40C2,1,1,1 = 300C3,1,1 + 120E2 E3 − 276E5 + 7ζ(5) (9.4.13)

149
These identities were proven using holomorphic subgraph reduction and a sieve algorithm.
We shall discuss the techniques used in the proof of these identities in the next subsection
for the example of the first identity. We refer to the bibliographical notes for the papers
where all algebraic identities between modular graphs of weight up to six, and some selected
identities of weight 7 are proven, and where a Mathematica package is now available.

9.5 Relating modular graph functions to holomorphic forms

The Maass operator ∇ = 2iτ22 ∂τ maps a modular graph form of modular weight (0, `) to a
modular graph form of modular weight (0, ` − 2), as was shown in subsection 9.2. When the
derivative is applied to a non-holomorphic Eisenstein series En , for example, and repeated
n times, one obtains the following form, using (9.4.5),
 
n Γ(2n) + 2n 0 Γ(2n) 2n
∇ En = CΓ = τ G2n (τ ) (9.5.1)
Γ(n) 0 0 Γ(n)π n 2

where G2n (τ ) is the holomorphic Eisenstein series of modular weight (2n, 0).
Applying the Maass operator to more complicated modular graph functions is not as
simple as in the case of non-holomorphic Eisenstein series, but nonetheless a similar pattern
emerges: ∇ decreases the anti-holomorphic exponents and increases the holomorphic expo-
nents. Let’s see how this works in the case of the modular graph function C1,1,1,1 . Using the
rules of (9.4.11) for ∇, we obtain for the first derivative,
   
1111 + 2 1 1 1
∇C1,1,1,1 = ∇C = 4C (9.5.2)
1111 0111

The form on the right has modular weight (0, −2). Applying ∇ again gives,
   
2 + 2 2 1 1 + 3 1 1 1
∇ C1,1,1,1 = 12 C − 24 C (9.5.3)
0011 0011

where to obtain the last term we have a momentum conservation identity to convert a
negative entry on the lower row,
   
+ 3 111 + 3 1 1 1
C = −3 C (9.5.4)
−1 1 1 1 0011

Further application of ∇ to (9.5.3), in its present form, would yield modular graph functions
with some negative anti-holomorphic exponents that cannot be removed simply by using
momentum conservation identities. However, the subgraph corresponding to the first two
entries forms a closed loop with only holomorphic momenta. The momentum summations in
such holomorphic subgraphs may be evaluated explicitly in terms of holomorphic modular

150
forms. This procedure is dubbed holomorphic subgraph reduction and, in the present case,
leads to the following identity,
         
+ 2 2 1 1 + 3 1 1 1 + 4 1 1 + 4 0 + 2 0
C − 2C = 2C + 3C C (9.5.5)
0011 0011 011 00 2 0

Collecting all contributions to ∇2 C1,1,1,1 and proceeding analogously for the entries C2,1,1
and E22 in the first identity in (9.4.13), we obtain,
     
2 + 4 1 1 + 4 0 20
∇ C1,1,1,1 = 24 C + 36 C C+
011 00 20
     
2 + 4 1 1 + 3 2 1 + 6 0
∇ C2,1,1 = 6 C + 6C + 4C
011 101 20
 2    
30 40 + 20
∇2 E22 = 8 C + + 12 C + C (9.5.6)
10 00 20

Applying ∇ to either ∇2 C1,1,1,1 or to E22 produces modular graph functions with negative
entries in the lower row that cannot be removed by using momentum conservation identities.
This occurs when ∇ is applied to one of the modular graph functions that is proportional
to a holomorphic modular form, such as the first factors in the last terms on the right of
∇2 C1,1,1,1 and ∇2 E22 . However, C2,1,1 and the linear combination C1,1,1,1 − 3E22 (that cancels
the last terms discussed above) produces the following ∇3 derivatives,
     
3 2 + 7 0 + 4 0 30
∇ (C1,1,1,1 − 3E2 ) = 432 C − 288 C C+
10 00 10
     
70 40 + 30
∇3 C2,1,1 = 108 C + − 12 C + C (9.5.7)
10 00 10
Proceeding analogously for the fourth ∇-derivative, we obtain,
 
4 2 + 8 0
∇ (C1,1,1,1 − 24C2,1,1 − 3E2 ) = −15120 C (9.5.8)
00

The right side is proportional to τ28 G8 , where G8 is the holomorphic Eisenstein series of
weight (8, 0). But this form is proportional to ∇4 E4 , so that we obtain,

∇4 C1,1,1,1 − 24C2,1,1 − 3E22 + 18E4 = 0

(9.5.9)

Next, we invoke the following lemma.

Lemma 9.3 Let F be a non-holomorphic modular function with polynomial growth near the
cusp as τ2 → ∞. If F satisfies the differential equation,

∇n F = 0 (9.5.10)

151
for an arbitrary integer n ≥ 1, then F is constant as a function of τ .
A simple proof of the lemma proceeds by making use of the following formula,
n 
Y 
¯ n (τ2 )−2n ∇n =
∇ ∆ − s(s − 1) (9.5.11)
s=1

This formula immediately implies that F must satisfy the weaker condition,
n 
Y 
∆ − s(s − 1) F = 0 (9.5.12)
s=1

Being a modular graph function, F has at most polynomial growth at the cusp. The solution
to the equation ∆fs = s(s − 1)fs for positive integer s is either a constant for s = 1 or an
Eisenstein series Es for s ≥ 2. This implies that F must be a linear combination of a constant
and the Eisenstein series Es for s = 2, · · · , n. Returning now to the original equation (9.5.10)
in the formulation of the lemma and using the asymptotics of the Eisenstein series at the
cusp of (4.2.16), we conclude that the coefficients of the Eisenstein series in F must all vanish
so that F is constant. This completes the proof of the lemma.
Applying the lemma to (9.5.9), it follows that,

F4 = C1,1,1,1 − 24C2,1,1 − 3E22 + 18E4 (9.5.13)

must be constant. To calculate the constant, we investigate the asymptotics at the cusp.
Clearly, all τ -dependent terms must cancel on the right side as one may verify using the
expansions of C1,1,1,1 and C2,1,1 in terms of the variable y = πτ2 ,

y4 2ζ(3)y 10ζ(5) 3ζ(3)2 9ζ(7)

C1,1,1,1 (τ ) = + + − + + O(e−2y )
945 3 y y2 4y 3
2y 4 ζ(3)y 5ζ(5) ζ(3)2 9ζ(7)
C2,1,1 (τ ) = + + − + + O(e−2y ) (9.5.14)
14175 45 12y 4y 2 16y 3

and the expansion of the Eisenstein series in (4.2.16). The constant y 0 term has vanishing co-
efficient in each of the the modular graph functions in F4 , so that F4 = 0. This completes the
proof of the first identity in (9.4.13). The other identities in (9.4.13) may be established with
the same methods, and holomorphic subgraph reduction formulas that generalize (9.5.5).

9.6 Modular graph functions and iterated modular integrals

The Fourier expansion of non-holomorphic Eisenstein series Ek (τ ) for integer k ≥ 2, pre-
sented in (4.2.17), shows that the Fourier non-zero modes are of the form (q N + q̄ N ) times a
polynomial in 1/τ2 . Such a simple form is no longer shared by modular graph functions and

152
forms at higher loops since in that case terms of the form q M q̄ N also occur. Nonetheless,
modular graph forms are related to holomorphic modular forms, even at higher loop, as
demonstrated by the fact that certain repeated derivatives ∇ map to holomorphic modular
forms. Another way to exhibit this connection concretely is via iterated modular integrals.
The basic building blocks of iterated modular integrals are the functions E(k1 , · · · , kr |τ )
and E 0 (k1 , · · · , kr |τ ) defined iteratively by the following relations,
Z i∞
dτr
E(k1 , · · · , kr |τ ) = 2πi Gkr (τr ) E(k1 , · · · , kr−1 |τ )
τ (2πi)kr
Z i∞
0 dτr
E (k1 , · · · , kr |τ ) = 2πi k
G0kr (τr ) E 0 (k1 , · · · , kr−1 |τ ) (9.6.1)
τ (2πi) r

The holomorphic modular form Gk (τ ) of weight k is defined for k ≥ 3 by the by now familiar
absolutely convergent Kronecker-Eisenstein series,
X 1
Gk (τ ) = k
= 2ζ(k) + G0k (τ ) (9.6.2)
m,n∈Z
(m + nτ )
(m,n)6=(0,0)

The function Gk (τ ) vanishes for all odd k ≥ 3 and evaluates to 2ζ(k) at the cusp, while G0k (τ )
consists of the non-constant Fourier modes of Gk (τ ), and decays exponentially at the cusp.
We extend the definition to k = 0 and supply initial conditions for the iterated integrals,
G0 (τ ) = G00 (τ ) = −1
E(·|τ ) = E 0 (·|τ ) = 1 (9.6.3)
In view of the exponential decay of G0k (τ ) at the cusp for k ≥ 4, the corresponding inte-
grals are convergent, while all other integrations are defined via the tangential base-point
prescription at the cusp, which essentially amounts to the following prescription,
Z i∞
dτr → −τ (9.6.4)
τ

The iterated integrals satsify the following differential equations,

4πτ22
∇E(k1 , · · · , kr |τ ) = Gkr (τ ) E(k1 , · · · , kr−1 |τ )
(2πi)kr
4πτ22
∇E 0 (k1 , · · · , kr |τ ) = G0 (τ ) E 0 (k1 , · · · , kr−1 |τ ) (9.6.5)
(2πi)kr kr
For example, we have,

0 4πτ22
∇E (2k|τ ) = 2k
G02k (τ )
(2πi)
∇E (2k, 0n |τ ) = −4πτ22 E 0 (2k, 0n−1 |τ )
0
(9.6.6)

153
where the notation 0n stands for an array of n zeros.
We begin by showing how the simplest Eisenstein series E2 may be expressed in terms of
these iterated integrals. The starting point is the differential equation obtained in (9.5.1),

2π 4 τ24
π 2 ∇2 E2 = 6τ24 G4 = + 6τ24 G04 (9.6.7)
15
Using the relation ∇τ2n = nτ2n+1 and eliminating G04 in favor of ∇E 0 (4|τ ) using the first
equation in (9.6.6), we obtain,

2 2 2π 4
π ∇ E2 = ∇τ33 + 24π 3 τ22 ∇E 0 (4|τ ) (9.6.8)
45
Successive use of the second relation in (9.6.6) for n = 1, 2 allows us to recast the right side in
the form of a total derivative in terms of the iterated integrals E 0 (4|τ ), E 0 (4, 0|τ ), E 0 (4, 02 |τ ),
 2π 3 τ 3 
2
∇(π∇E2 ) = ∇ + 24π 2 τ22 E 0 (4|τ ) + 12πτ2 E 0 (4, 0|τ ) + 3E 0 (4, 02 |τ ) (9.6.9)
45
The equation is readily solved, and the general solution is given as follows,

2π 3 τ23
π∇E2 = + 24π 2 τ22 E 0 (4|τ ) + 12πτ2 E 0 (4, 0|τ ) + 3 E 0 (4, 02 |τ ) + f1 (τ ) (9.6.10)
45
where f1 (τ ) is holomorphic. Making use anew of the equations (9.6.6) we can integrate this
equation once more, and find,

π 2 τ22 3 0 f1 (τ )
E2 (τ ) = − 6 E 0 (4, 0|τ ) − E (4, 02 |τ ) − + f2 (τ ) (9.6.11)
45 πτ2 πτ2
where f2 (τ ) is another holomorphic function. The reality of E2 determines f1 and f2 uniquely
up to an additive constant in f1 , which may be fixed by the known asymptotics at the cusp.
Using again the shorthand y = πτ2 , we have,

y 2 ζ(3)  6 
− 12 Re E 0 (4, 0|τ ) − Re E 0 (4, 02 |τ )
 
E2 (τ ) = + (9.6.12)
45 y y
The above expression reflects the structure of the Fourier decomposition of the Eisenstein
series mentioned at the beginning of this section.
Next, we provide an example of the decomposition into iterated integrals for the simplest
modular graph function that is not an Eisenstein series, namely C2,1,1 . The differential
equations for C2,1,1 , given in the second line of (9.5.7), may be recast in the following form,

π 2 ∇3 C2,1,1 − 10
9
E4 = −6τ22 G4 (τ )∇E2


(9.6.13)

154
The subtraction of the term in E4 is convenient as it readily cancels the first term on the
right side of the second line in (9.5.7). One may begin by using expression given in (9.6.10)
to eliminate ∇E2 from the above equation in favor of iterated integrals. Successive use of
(9.6.5) again allows us to proceed by the same methods as used for E2 , and one obtains,
  
y4 ζ(3)2
+ ζ(3)y 5ζ(5) 2y 3ζ(3)
Re E 0 (4, 02 |τ )

C2,1,1 = − 20250 45
+ 12y
− 4y 2
− 15
− y2
(9.6.14)
2
− y92 Re E 0 (4, 02 |τ ) − 72 Re E 0 (42 , 02 |τ ) − 15 Re E 0 (4, 03 |τ )
    

− 36 Re E 0 (4, 0, 4, 02 |τ ) − 108 Re E 0 (42 , 03 |τ ) − 10y

1
Re E 0 (4, 04 |τ )
     
y y

In this formulation, the first term on the second line reveals the presence of terms of the
form q M q̄ N in the Fourier series of C2,1,1 , as expected.

• Bibliographical notes
The study of modular graph functions originated with explorations into the low energy
effective interactions produced by one-loop superstring amplitudes [91], as will be described
in sections 12 and 14. Systematic investigations of modular graph functions were initiated
in [92, 93, 94]. The proof of Theorem 9.1 was given in [92], while the proof of Theorem 9.2
was given in [95], where explicit expressions for all the coefficients in the expansion were also
provided. The Poincaré series for Γ∞ \SL(2, Z), and Fourier series were computed in [96].
The relation between modular graph functions and single-valued elliptic polylogarithms was
exhibited in [97]. Introductions to elliptic polylogarithms may be found in [98, 99, 100, 101].
The systematic construction of the algebraic and differential identities satisfied by modu-
lar graph functions and forms was carried out in [102, 103, 104] using the method of holomor-
phic subgraph reduction. A variety of other methods were used in [105, 106, 107, 108, 109] to
obtain identities within special classes of modular graph functions. An efficient Mathematica
package for algebraic and differential relations was developed in [110, 111]. A class of related
modular functions is presented in [112, 113].
Fourier series expansions were obtained in [95, 114, 96, 115], while Poincaré series were de-
rived in [114, 96, 115]. The role of modular graph functions in genus-one amplitudes for open
and Heterotic strings was elucidated in [116, 117]. The solutions to the differential identities
obtained from holomorphic subgraph reduction may be obtained in terms of iterated elliptic
integrals and systematically represented in terms of generating functions [116, 118, 119].
The generalization of modular graph functions to higher genera was initiated in [120] and
is directly motivated by the study of the low energy expansion of superstring amplitudes,
as will be discussed extensively in section 12. The natural generalization of modular graph
forms to higher genus is provided by modular graph tensors, which were introduced in [121].
Various degenerations of the higher genus Riemann surface, studied in [122], naturally lead
one to introduce elliptic modular graph functions, whose study was begun in [123].

155
Part II
Extensions and Applications
In the second part of these lecture notes, we consider mathematical and physical extensions
and applications of the material discussed in the first part. We shall also exhibit the con-
nections with the second part of the title, namely various topics in string theory and related
fields, wherever appropriate.
We start off in section 10 with a review of Hecke operators, Hecke eigenforms, and the
action of Hecke operators on Maass forms and vector-valued modular forms. We close with
a physical application to the study of two-dimensional conformal field theory. In section 11,
we consider the topic of complex multiplication and singular moduli. We prove that the
modular j-function is an algebraic integer at complex multiplication points τ and evaluate
elliptic and ϑ-functions at these points.
In section 12, we present a quick introduction to string amplitudes. We review their
perturbative expansion in the genus which corresponds to the loop expansion in quantum
field theory, and the importance of conformal symmetry in decoupling of negative norm
states. We summarize results for amplitudes to tree-level, one-loop, and two-loop orders
without derivation. We emphasize the crucial role played by modular invariance in the UV
finiteness of string theory, and by modular graph functions in the low energy expansion in
effective interactions. As an application, we analyze toroidal compactifications in section 13.
In section 14 we review Type IIB supergravity and superstring theory, discuss its S-
duality symmetry under SL(2, Z), spell out the modular properties of the low energy effective
interactions, and make contact with the low energy effective interactions predicted earlier
from superstring amplitudes in section 12.
In section 15 we present a brief introduction to Yang-Mills theories with extended su-
persymmetry in four space-time dimensions. We review Goddard-Nuyts-Olive duality, its
concrete realization as Montonen-Olive dualities in N = 4 theories, and then discuss the
Seiberg-Witten solution with emphasis on the role played by SL(2, Z). We close with a
discussion of dualities of N = 2 superconformal Yang-Mills theory.
In the last section 16, we address Galois theory, along with some applications to rational
conformal field theory.

156
10 Hecke Theory
A natural set of operators acting on the space of modular forms are the Hecke operators Tn ,
labeled by positive integers n. They map holomorphic functions to holomorphic functions,
weight-k modular forms to weight-k modular forms, and weight-k cusp forms to weight-k
cusp forms. For the group SL(2, Z) the Hecke operators are endomorphisms that map the
space of holomorphic modular forms Mk into itself, and map the subspace of cusp forms Sk
into itself. For congruence subgroups, the Hecke operators map weight-k modular forms of
one congruence subgroup into those of another congruence subgroup. Hecke operators com-
mute with the Laplace-Beltrami operator on the upper half-plane so that non-holomorphic
Eisenstein series and cusp forms are simultaneous eigenfunctions of all Hecke operators. Fi-
nally, given a modular form with positive integer Fourier coefficients, the Hecke transforms
also have positive integer Fourier coefficients. For this reason, Hecke operators are relevant
in a number of physical problems, as we shall see shortly.

10.1 Definition of Hecke operators

Given a lattice Λ = Zω1 + Zω2 with modulus τ = ω2 /ω1 ∈ H, a modular form f (τ ) of weight
k under SL(2, Z) may be represented by a degree k homogeneous function of the lattice Λ,
denoted by the same letter, f (Λ) = ω1−k f (ω2 /ω1 ). Modularity of f (τ ) is then equivalent to
the fact that f (Λ) is intrinsic and depends only on the lattice Λ, but not on the particular
basis vectors ω1 , ω2 used to represent Λ. The Hecke operator Tn : Mk → Mk may now be
defined on f (Λ) as a sum over all index n sub-lattices Λn ⊂ Λ,
X
Tn f (Λ) = nk−1 f (Λn ) (10.1.1)
Λn ⊂Λ
[Λ:Λn ]=n

where the sub-lattice Λn is understood to be a subgroup of Λ. The definition of Tn given

above is intrinsic, in that it depends only on the lattices Λ and Λn , and not on any particular
choice of the basis vectors for each lattice. An arbitrary multiplicative normalization factor,
which depends only on k and n, is allowed in the definition of Tn , and has been chosen here
for later convenience.
To make contact with modular forms expressed as functions of τ ∈ H, we now choose
specific basis vectors for both Λ and Λn . In particular, the set of index n sub-lattices Λn
(n) (n)
may be parametrized by choosing basis vectors ω1 , ω2 for Λn which are integer linear
combinations of the basis vectors ω1 and ω2 of Λ,
(n) (n)
ω2 = a ω2 + b ω1 ω1 = c ω2 + d ω1 (10.1.2)

where a, b, c, d ∈ Z. Since Λ and Λn are Abelian groups, Λn is a normal subgroup of Λ and

the quotient Λ/Λn is a finite group. One may represent Λ/Λn as the set of the points in Λ

157
modulo Λn , equipped with addition modulo Λn . The cardinality of Λ/Λn is the index [Λ : Λn ]
which is given by the number of points in Λ that belong in the fundamental parallelogram
of Λn , and this number is n = ad − bc, a number that is positive assuming that Λn inherits
its orientation from Λ. The above parametrization of Λn motivates us to introduce the set
Mn of all 2 × 2 matrices with integer coefficients of determinant n > 0,
  
a b 
Mn =  a, b, c, d ∈ Z; ad − bc = n > 0 (10.1.3)
c d

Given two matrices A, B ∈ Mn , one introduces an equivalence relation A ∼ B whenever

A = µB for some element µ ∈ SL(2, Z). The quotient SL(2, Z)\Mn is the set of equivalence
classes of elements of Mn , subject to the equivalence relation. This relation is precisely the
equivalence relation between lattices Λn under SL(2, Z) transformations of the underlying
lattice Λ. Therefore, the cosets in SL(2, Z)\Mn are in one-to-one correspondence with the
lattices Λn and the Hecke operator Tn may equivalently be defined to act as follows,
X
Tn f (τ ) = nk−1 (cτ + d)−k f (γτ ) (10.1.4)
γ∈SL(2,Z)\Mn

The above definition makes it clear that Tn maps holomorphic functions to holomorphic
functions. Below we shall show that it maps modular forms of SL(2, Z) to modular forms.

10.2 Explicit parametrization of equivalence classes SL(2, Z)\Mn

The action of the Hecke operators Tn can be made even more explicit by using the following
convenient explicit parametrization of the equivalence classes SL(2, Z)\Mn .

Lemma 10.1 Every equivalence class in SL(2, Z)\Mn contains an upper triangular repre-
sentative of the form ( a0 db ) ∈ Mn , with d > 0. A complete set of inequivalent elements of
SL(2, Z)\Mn is given by b ∈ {0, 1, · · · , d − 1}.
To prove the first part of the Lemma, we show that for every matrix A = ( ac db ) ∈ Mn
there exists a µ ∈ SL(2, Z) such that µA is of the form given in the Lemma. To do so we
parameterize µ as follows,
   
p q pa + qc pb + qd
µ= µA = (10.2.1)
r s ra + sc rb + sd

If c = 0 then d 6= 0 in view of ad − bc = n > 0 by the definition of Mn and we set r = q = 0

and p = s = sign(d) to obtain a representative of the desired form. For the remaining cases
c 6= 0 we choose mutually prime r and s such that s/r = −a/c to render the matrix µA
upper triangular. By Bézout’s theorem (reviewed in appendix A) there exist integers p and q
such that ps − qr = 1 guarantees µ ∈ SL(2, Z), and either µ or −µ gives the desired matrix.

158
 To prove the second part of the Lemma, we show that the set given by b ∈ {0, 1, · · · , d−1}
is complete and that all such elements are inequivalent

under
SL(2, Z). To see that the set
is complete, we take two matrices A1 = 0 d and A2 = a0 bd2 with b1 ≡ b2 (mod d). Then
a b1

if b1 = b2 + kd we define µ = ( 10 k1 ) ∈ SL(2, Z) so that A1 = µA2 , and hence A1 ∼ A2 .

Conversely, to see that all elements are inequivalent we proceed as follows. If A1 and A2 are
such that A1 ∼ A2 , there exists µ = ( pr qs ) ∈ SL(2, Z) such that A1 = µA2 , namely,
      
a1 b 1 p q a2 b 2 pa2 pb2 + qd2
= = (10.2.2)
0 d1 r s 0 d2 ra2 rb2 + sd2

Since a2 d2 = n, clearly a2 6= 0, and hence by equating the bottom-left entries on both sides
we conclude that r = 0. Then since pq − rs = 1, we have pq = 1 and thus p = q = ±1.
We make the choice p = q = +1, which we can always do by replacing µ with −µ if needed.
Then equating entries on both sides, we see that a1 = a2 , d1 = d2 , and b1 ≡ b2 (mod d).
Hence if we restrict to b ∈ {0, . . . , d − 1}, there are no equivalences.
We shall also make use of the following simple but important lemma.

Lemma 10.2 For every pair (A1 , µ1 ) with A1 ∈ Mn and µ1 ∈ SL(2, Z), there exists a pair
(A2 , µ2 ) with A2 ∈ Mn and µ2 ∈ SL(2, Z) such that A1 µ1 = µ2 A2 . As a result, the left and
right cosets SL(2, Z)\Mn and Mn /SL(2, Z) are isomorphic.
To prove this lemma, we may choose A1 and A2 of the form provided by Lemma 10.1
without loss of generality, and parametrize the matrices A1 , A2 , µ1 , µ2 involved as follows,
       
a1 b 1 a2 b 2 p1 q1 p2 q2
A1 = A2 = µ1 = µ2 = (10.2.3)
0 d1 0 d2 r1 s1 r2 s2

Given µ1 and A1 , one proceeds to solve for A2 and µ2 using µ2 A2 = A1 µ1 in components,

p2 a2 = a1 p1 + b1 r1 p 2 b 2 + q 2 d 2 = a1 q 1 + b 1 s 1
r2 a2 = d1 r1 r2 b2 + s2 d2 = d1 s1 (10.2.4)

From the equations on the left one obtains p2 /r2 = (a1 p1 +b1 r1 )/(d1 r1 ), which determines the
pair (p2 , r2 ) uniquely up to a common sign in view of the fact that p2 , r2 must be relatively
prime. The entry a2 is then determined by a2 = d1 r1 /r2 . Having determined the pair p2 , r2 ,
Bézout’s theorem guarantees the existence of a solution q2 , s2 to p2 s2 − q2 r2 = 1. In terms
of these solutions, we readily solve the equations on the right to obtain,

b 2 = s 2 a1 q 1 + s 2 b 1 s 1 − q 2 d 1 s 1
d2 = p2 d1 s1 − r2 a1 q1 − r2 b1 s1 (10.2.5)

159
10.3 Hecke operators map Mk to Mk
Lemma 10.1 allows us to recast the action of the Hecke operators more explicitly as follows,
X 1 X d−1  
k−1 aτ + b
Tn f (τ ) = n k
f (10.3.1)
ad=n
d b=0
d
d>0

or by solving for a = n/d,

X 1 X d−1  
k−1 nτ + bd
Tn f (τ ) = n k
f (10.3.2)
d|n
d b=0
d2
d>0

In terms of these expressions for Tn , it is now straightforward to prove the modularity of

Tn f (τ ). We shall use the notations and parametrizations of Lemma 10.2. Indeed, considering
a transformation by µ1 ∈ SL(2, Z), we have,
X 1
Tn f (µ1 τ ) = nk−1 f (A1 µ1 τ )
A1
dk1
X 1
= nk−1 k
f (µ2 A2 τ )
A2
d 1
X 1
= nk−1 k
(r2 A2 τ + s2 )k f (A2 τ ) (10.3.3)
A2
d 1
P
where we have introduced a shorthand notation A for the summations in (10.3.1), and have
used the fact that if A1 runs through a complete set of inequivalent elements of SL(2, Z)\Mn ,
so too does A2 . Substituting the two equations on the second line of (10.2.4), we then obtain
X 1
Tn f (µ1 τ ) = nk−1 (r τ + s1 )k f (A2 τ ) = (r1 τ + s1 )k Tn f (τ )
k 1
(10.3.4)
A
d 2
2

which is the desired result.

10.4 Fourier expansions

For many applications in Physics and Mathematics, we will often be interested in the Fourier
expansions of Hecke
P transforms. Given a weight-k modular function f (τ ) with Fourier ex-
m
pansion f (τ ) = m am q , the Fourier expansion of the Hecke transform can be computed
explicitly as follows. We begin by noting that,
X 1 Xd−1 X  n k−1
nm nm
2πi bm
X X
k−1
Tn f (τ ) = n a m q d2 e d = am q d 2 (10.4.1)
dk b=0 m∈Z m∈Z
d
d|n d| gcd(n,m)

160
 Pd−1 2πi bm
The second equality follows from the fact that b=0 e d equals d if md ∈ Z, and vanishes
otherwise. Then we simply make repeated changes to the summation index to get,
X X  n k−1 n`
XX
Tn f (τ ) = a`d q d = tk−1 a`n/t q `t
`∈Z d|n
d `∈Z t|n
X X
k−1 m
= t amn/t2 q (10.4.2)
m∈Z t|gcd(m,n)

We thus conclude that the Fourier expansion of the Hecke transform is given by,
X X
Tn f (τ ) = b(n)
m q
m
b(n)
m = dk−1 amn/d2 (10.4.3)
m∈Z d|gcd(m,n)

If the Fourier coefficients am of f (τ ) are integers, then by construction so are the Fourier
coefficients of the Hecke transform Tn f for all n > 0. Furthermore, we see from (10.4.3) that
(n)
if a0 = 0 then b0 = 0 as well, and thus Tn acts as an endomorphism on the space Sk of
weight-k cusp forms.
(n)
In the particular case of n = p being prime, the Fourier coefficients bm of the Hecke
transform take a particularly simple form,

(p) pk apm p-m
bm = k−1 (10.4.4)
p (p apm + am/p ) p|m

(p)
If k ≥ 1, we see that if am ∈ Z, then bm ∈ Z as well. For k = 0 though this is no longer the
case. For this reason, in the case of k = 0 one usually drops the conventional factor of nk−1
in (10.1.4).

10.5 Example: the Ramanujan tau function

We illustrate the utility of this machinery by means of a simple example. In particular, let
us prove the product formulas (3.8.16) and (3.8.17) for the Ramanujan tau function, defined
in (3.8.13) of section 3.8.1, and repeated here for convenience,
∞
X
12
∆(τ ) = (2π) τ (n)q n (10.5.1)
n=1

To do so, we apply the Hecke operators to the discriminant ∆(τ ) to obtain,

X X  nm 
Tn ∆(τ ) = b(n)
m q m
b (n)
m = (2π)12
d 11
τ (10.5.2)
m
d2
d|gcd(m,n)

161
Since ∆(τ ) and Tn ∆(τ ) are both holomorphic cusp forms of weight 12, and the space S12 is
one-dimensional, we conclude that we must have,

Tn ∆(τ ) = αn ∆(τ ) (10.5.3)

for some sequence of constants αn ∈ C. Equating Fourier coefficients using (10.4.3), we

obtain the following relation,
X  nm 
αn τ (m) = d11 τ (10.5.4)
d2
d|gcd(m,n)

For the particular case of m = 1 we have τ (1) = 1 and only the term with d = 1 contributes
to the sum on the right side, so that αn = τ (n). Substituting this expression for αn into
(10.5.4) readily reproduces (3.8.17).

10.6 Multiplicative properties of Hecke operators

An important property of Hecke operators is their behavior under multiplication,
X
Tn Tm = dk−1 Tmn/d2 (10.6.1)
d|(n,m)

The full proof of this property is straightforward and standard (see e.g. [124]), but somewhat
lengthy. Here we settle for a proof of a special case of this relation,

Tn Tm = Tnm gcd(m, n) = 1 (10.6.2)

To prove this, we begin by explicitly applying the Hecke operators on a weight-k modular
form f (τ ) to obtain

δ−1 X
d−1  
k−1
X X X 1 αaτ + (αb + βd)
Tn Tm f (τ ) = (nm) f (10.6.3)
αδ=n ad=m β=0 b=0
(dδ)k dδ
δ>0 d>0

Now we make use of the fact that (m, n) = 1, which tells us that since d and δ run through
positive divisors of n and m respectively, dδ should run through the positive divisors of
the product nm. Furthermore, since b (mod d) runs through a complete set of repre-
sentatives for SL(2, Z)\Mm and β (mod δ) runs through a full set of representatives for
SL(2, Z)\Mn , then αb + βd (mod dδ) runs through a complete set of inequivalent represen-
tatives for SL(2, Z)\Mmn . Hence by the definition given in (10.1.4), the right side of (10.6.3)
is exactly the action of Tmn on f (τ ).

162
10.7 Hecke eigenforms
The multiplication identity (10.6.1) implies that the Hecke operators commute. As such, we
may consider modular forms f (τ ) ∈ Mk which are simultaneous eigenfunctions of all Hecke
operators,
Tn f (τ ) = λn f (τ ) ∀n ∈ N (10.7.1)
Such forms are called Hecke eigenforms, or just “eigenforms” for short.21 Comparing the
O(q) terms in the Fourier expansions on both sides of (10.7.1), the Fourier coefficients of
eigenforms are seen to satisfy,
b(n)
m = λ n am (10.7.2)
using (10.4.3). Considering in particular m = 1, we see that an = λn a1 , and thus that any
non-trivial eigenform must have a1 6= 0. One often finds it useful to normalize a1 = 1 so
that the Hecke eigenvalues coincide with the Fourier coefficients of f (τ ),
Tn f (τ ) = an f (τ ) (10.7.3)
It is a general fact that normalized Hecke eigenforms have Fourier coefficients which are real
(by Hermiticity of Tn , as will be discussed momentarily) and multiplicative,
X
am an = dk−1 amn/d2 (10.7.4)
d|gcd(m,n)

The existence of eigenforms is obvious for certain weights. For example, for weights
k = 4, 6, 8, 10, and 14, we saw in Theorem 3.3 that the space Mk is one-dimensional, being
generated by the relevant Eisenstein series Ek (τ ). Hence for such values of k the Hecke trans-
form Tn Ek (τ ) must clearly be a multiple of Ek (τ ) for any n, making Ek (τ ) a Hecke eigenform.
More surprisingly, the Eisenstein series are Hecke eigenforms even when dim Mk > 1, which
is shown as follows. Consider a generic non-cuspidal modular function satisfying
T n f = λn f (10.7.5)
with f having even weight k. For non-cuspidal f the Fourier coefficient a0 is non-zero, and
(n)
by extension b0 is non-zero as well. In particular, for m = 0, 1 we have from (10.4.3) that
(n)
X
b0 = dk−1 a0 = a0 σk−1 (n)
d|n
(n)
X
b1 = dk−1 an/d2 = an (10.7.6)
d|gcd(1,n)

21
In some literature the term eigenform refers to eigenfunctions of only a single Hecke operator, whereas
our definition of eigenforms would be referred to as “simultaneous eigenforms.”

163
with σ∗ (n) the sum of divisors function. Comparing the first of these to (10.7.2) allows us to
conclude that λn = σk−1 (n), and then from the second of these we obtain an = a1 σk−1 (n).
Choosing normalization a0 = 1, the Fourier expansion of the Hecke eigenform f (τ ) is required
to take the following form,

∞
X
f (τ ) = 1 + a1 σk−1 (N ) q N (10.7.7)
N =1

The normalized Eisenstein series Ek (τ ) take exactly this form, as in (3.5.13). Incidentally,
note that the divisor function indeed satisfies the multiplicative property (10.7.4).
Besides Eisenstein series, all Hecke eigenforms are cuspidal. We have already encoun-
tered one example of a cuspidal eigenform in section 10.5, where we saw that the modular
discriminant ∆(τ ) is an eigenfunction of the Hecke operators, with the eigenvalues being
precisely the values of the Ramanujan tau function. We also saw that the Ramanujan tau
function satisfied a multiplicativity property consistent with (10.7.4).
The fact that ∆(τ ) was an eigenform followed simply from the fact that dim S12 = 1. The
same argument holds for cusp forms of weight k = 16, 18, 20, 22, and 26, where dim Sk = 1,
and is generated by ∆(τ )Ek (τ ). However, it is a remarkable fact, due to Hecke, that there
exist cuspidal eigenforms for other k as well. In fact, Hecke showed that eigenforms provide
a basis of Mk for every k. In order to prove this, he found it necessary to introduce an inner
product on the space of cusp forms, as we now briefly review.

10.7.1 Petersson inner product

Weight-k cusp forms admit an inner product known as the Petersson inner product,
Z
dτ1 dτ2
hf, gi = (τ2 )k f (τ )g(τ ) (10.7.8)
SL(2,Z)\H τ22

which is convergent due to the vanishing of f or g towards the cusp. A key fact is that Hecke
operators are self-adjoint under this inner product,

hTn f, gi = hf, Tn gi (10.7.9)

This together with commutativity of Tn implies that a generic f ∈ Sk can be decomposed

into a basis of Hecke eigenforms, which is the result quoted before. This also tells us that
the eigenvalues of Tn (and hence the Fourier coefficients of normalized eigenforms) are all
real—another result quoted earlier.

164
10.8 Hecke operators acting on Maass forms
We now review the properties of Hecke operators acting on the space of Maass forms of weight
0, which was introduced in section 4.3. Recall that a Maass form f (τ ) is a complex-valued
function of τ that satisfies a Laplace eigenvalue equation,

∆f (τ ) = s(s − 1)f (τ ) (10.8.1)

is invariant under SL(2, Z), and has at most polynomial growth at the cusp. The Fourier
series for general Maass forms was obtained in (4.3.9), and we repeat it here for convenience,
X √
f (τ ) = a τ2s + b τ21−s + αN τ2 Ks−1/2 (2π|N |τ2 ) e2πiN τ1 (10.8.2)
N 6=0

The space of all Maass forms for the full modular group SL(2, Z) and with eigenvalue s(s−1)
is denoted N (s) = N (SL(2, Z), s) with the understanding that it satisfies N (1 − s) = N (s).
Non-holomorphic Eisenstein series Es (τ ), defined and normalized in (4.2.1), are Maass forms
whose Fourier series were obtained in (4.2.16),
1
2ζ(2s) 2Γ(s − 12 )ζ(2s − 1) 4|N | 2 −s σ2s−1 (|N |)
a= b= 1 αN = (10.8.3)
πs Γ(s) π s− 2 Γ(s)

A Maass form f (τ ) is a cusp form provided the constant Fourier mode in (10.8.2) vanishes,
a = b = 0. Cusp forms decay exponentially at the cusp and have finite norm with respect
to the Petersson inner product. The space of cusp forms is denoted S(s). The space S(s)
is empty except for an infinite number of discrete eigenvalues of the form s ∈ 21 + iR.
Theorem 4.2 provides a basis for N (s) in terms of Eisenstein series and cusp forms.
Hecke operators are linear operators that may be defined on Maass forms in N (s) similarly
to how they were defined on holomorphic modular forms,

1 X
Tn f (τ ) = √ f (γτ ) (10.8.4)
n
γ∈SL(2,Z)\Mn

In contrast to (10.1.4), there is no power of (cτ +d) multiplying the summand as the form f (τ )
has weight 0. The definition allows for an arbitrary normalization prefactor that depends only
√
on n and is completely multiplicative and has been chosen to be 1/ n for later convenience.
Using Lemma 10.2, and the fact that a Maass form is invariant under SL(2, Z), it follows
immediately that Tn f (τ ) is also invariant under SL(2, Z). Furthermore, since f (τ ) has at
most polynomial growth at the cusp, so must Tn f (τ ). This is because it is defined to be
a finite sum of terms, each of which has at most polynomial growth. By manipulations

165
analogous to those used for the holomorphic case, one may use an explicit parametrization
of the coset SL(2, Z)\Mn to recast the action of the Hecke operators in the following form,
 
1 X aτ + b
Tn f (τ ) = √ f (10.8.5)
n ad=n,d>0 d
b=0,··· ,d−1

The Laplace-Beltrami operator ∆ = 4τ22 ∂τ ∂τ̄ is invariant under Tn because the transforma-
tion τ → τ 0 = (aτ + d)/d implies τ2 → τ20 = aτ2 /d and ∂τ → ∂τ 0 = d/a ∂τ . These results may
be summarized and further extended in the following theorem [20]:
Theorem 10.3 The Hecke operators Tn satisfy the following properties.
1. Tn map N (s) to N (s) and S(s) to S(s).
2. The Fourier series of a Maass form f (τ ) in N (s), given in (10.8.2), and its Hecke
image Tn f (τ ) expressed in the form given in (10.8.2) with a → a0 , b → b0 , αN → αN 0

are related as follows,

 0
a = n1/2−s σ2s−1 (n) a 0
X
α = αN n/d2 (10.8.6)
b0 = ns−1/2 σ1−2s (n) b N
d|gcd(N,n)

3. The product of operators Tm and Tn on N (s) is commutative and given as follows,

X
Tm Tn = Tmn/d2 (10.8.7)
d|gcd(m,n)

4. The operators Tn are self-adjoint linear operators with respect to the Petersson inner
product on the space of cusp forms, and may be simultaneously diagonalized on S(s).
5. If f ∈ N (s) is a simultaneous eigenfunction of all Hecke operators with eigenvalues
Tn f = λn f then the coefficients αN in its Fourier expansion (10.8.2) satisfy,
α±N = λN α±1 N >0 (10.8.8)
Therefore, up to the normalizations α±1 , the coefficients of the non-constant Fourier
modes are the eigenvalues of the Hecke operators.
Item 1 was already established in the preamble to the theorem, while Item 2 readily
follows by identifying terms in the Fourier series for f and Tn f . Item 3 may be established
along the same lines as the product formula (10.6.1) was for holomorphic modular forms.
The proof of Item 4 is given in [20]. Finally, Item 5 is established using the second relation
0
in Item 2 for αN = λn αN , which implies,
X
λn αN = αN n/d2 (10.8.9)
d|gcd(N,n)

for all n > 0 and all N 6= 0. Setting N = ±1 collapses the sum on the right to just the term
d = 1, which produces the result announced.

166
10.9 Hecke operators on vvmfs
In section 8 we introduced modular differential equations (MDEs), and discussed how they
were usefully organized according to the so-called Wronskian index `, defined in (8.3.22),
which gives a rough measure of the number of poles allowed in the coefficient functions. For
the case of ` = 0 the coefficient functions must be holomorphic, and for low orders the MDEs
can be recast as hypergeometric equations and solved exactly, as was done in sections 8.2.2
and 8.2.3 for the cases of second- and third order, respectively.
Obtaining exact solutions to MDEs for general ` is more difficult. One way to obtain (a
subset of) solutions to MDEs with non-zero ` is to make use of Hecke operators. Here we will
give only a flavor of how this is done. Above we have defined the action of Hecke operators on
modular forms for SL(2, Z). In fact, there exists a straightforward generalization to modular
forms for congruence subgroups. Assuming the Integrality Conjecture of section 8.3.2, this
then leads to a component-wise action of the Hecke operators on integral vector-valued
modular forms. For a vector-valued modular form f (τ ) with components fi (τ ) transforming
covariantly under Γ(N ), the action of Tp for p prime such that (p, N ) = 1 is given by
p−1  
X X τ + bN
(Tp f )i (τ ) = ρij (σp )fj (pτ ) + fi (10.9.1)
j b=0
p

where ρ(σp ) is a matrix representation of

σp = T p S −1 T p ST p S (10.9.2)

and p is the modular inverse of p, i.e. pp ≡ 1 (mod N ). This definition of the Hecke operators
is chosen to lead to a simple action on the Fourier coefficients. Indeed, if we write the Fourier
expansion of the components of a vector-valued modular function as,
X
fi (τ ) = an(i) q n/N (10.9.3)
n

then one has that

X
(Tp f )i (τ ) = bn(i) (p)q n/N (10.9.4)
n

with
(
(i)
papn p-n
b(i)
n (p) = (i) P (j) (10.9.5)
papn + j ρij (σp )an/p p| n

This action of the Hecke operator gives a new vector-valued modular form with components
which have the same weight k under Γ(N ) as before, but now with different leading exponents

167
 (i) P (i)
n0 in the notation of (8.3.15). Because ordWd (i∞) = i n0 , we expect from (8.3.23) that
the Hecke image Tp f (τ ) generically satisfies an MDE with different Wronskian index `. The
Hecke operators can thus be used to transform solutions to MDEs with ` = 0 to solutions to
those with ` > 0.
We content ourselves with a single example here. Consider the three-dimensional vector-
valued modular function made up of characters of the Ising CFT, given in (8.2.21). From
(8.2.20), the leading exponents are seen to be
(1) 23 (2) 1 (3) 1
n0 = n0 = − n0 = (10.9.6)
48 48 24
in the notation of (8.3.15). Upon application of the Hecke operator Tp , these change to

(1) 23p mod 48 (2) p (3) 2p mod 48

ñ0 = ñ0 = − ñ0 = (10.9.7)
48 48 48
from (10.9.5). Then using the formula (8.3.23), we conclude that for e.g. T49 , T53 , and T55
we have ` = 6, and hence application of these Hecke operators to the Ising characters gives
solutions to an MDE of order 3 with meromorphic coefficient functions. These solutions are
exact in the sense that their full Fourier expansions can be obtained, though they are not
known to be expressible in terms of any standard functions.
Another interesting case is that of T47 acting on the Ising characters. In that case we
obtain
47
(T47 χ1 )(τ ) = q − 48 1 + 96256 q 2 + 9646891 q 3 + 366845011 q 4 + . . .


(10.9.8)
25
(T47 χ )(τ ) = q 48 4371 + 1143745 q + 64680601 q 2 + 1829005611 q 3 + . . .


23
(T47 χσ )(τ ) = q 24 96256 + 10602496 q + 420831232 q 2 + 9685952512 q 3 + . . .

These turn out to be solutions to an order 3 MDE with ` = 0, and can be shown to
be precisely the characters of the so-called Baby Monster CFT, which has as its global
symmetry group the Baby Monster B, the second-largest finite sporadic group. Hence the
Hecke operator T47 maps from the Ising CFT to the Baby Monster CFT, in some appropriate
sense.

10.10 Physics applications

We now give a physical application of Hecke theory. Before doing so, we will need to recall the
basic notion of Virasoro characters. In section A.7, we introduced characters for finite groups.
Characters can be introduced for Lie groups as well, in which case, given a representation
R of the group, we have,

χR (g) = TrR (g) (10.10.1)

168
with the trace TrR taken in representation space of R. The general theory of characters is
rich, but largely unnecessary for our purposes. The particular case of importance to us here
is that of the characters of the Virasoro group. As we have discussed in section 5.1.1, the
Virasoro group is an infinite-dimensional Lie group which is the universal central extension
of the group of diffeomorphisms of the circle. The corresponding Lie algebra is the Virasoro
algebra, whose generators Ln satisfy the structure relations given in (5.1.3), which we repeat
here for convenience,
c 3
[Ln , Lm ] = (n − m)Ln+m + (n − n)δn,−m (10.10.2)
12
The parameter c here is known as the central charge.
A quantum field theory with Virasoro symmetry is known as a conformal field theory
(CFT). By “symmetry,” one means that the state space of the theory is organized into
Virasoro representations. In a unitary CFT, these representations can be constructed by
beginning with a highest weight state |hi satisfying

L0 |hi = h|hi Ln |hi = 0 n>0 (10.10.3)

and then generating other states in the representation by taking linear combinations of states
of the form,

L−n1 L−n2 . . . L−nr |hi ni > 0 (10.10.4)

Such states are known as descendants, and will be collectively denoted by |h, {n}i. The
character associated to a representation labelled by h is then defined to be,
c
X
χh (τ ) = q h− 24 qN (10.10.5)
{n}

with q = e2πiτ and N =

P
i ni . This is sometimes also written as,
 c

χh (τ ) = TrVh q L0 − 24 (10.10.6)

with Vh known as the weight-h Verma module.

Characters play an important role in the study of 2d CFTs. In particular, the torus
partition function of many CFTs can be decomposed into a sum over conformal families and
a sum within each family, giving
X
Z(τ, τ̄ ) = Nh,h̃ χh (τ )χh̃ (τ ) (10.10.7)
h,h̃

for some gluing matrix Nh,h̃ . Here we are assuming that the indices h, h̃ are discrete, but
not necessarily that the sums over conformal families are finite.

169
 The partition function is by definition a weight-0 modular function. But conversely, not
every modular function corresponds to a valid CFT partition function. For example, since
partition functions count degeneracies of states, any physically sensible partition function
must have positive integer Fourier coefficients. The same is true for the constituent charac-
ters. In this context, it is useful to introduce a variant of the Hecke operators with alternative
normalization,

Tn := n Tn (10.10.8)

which has the virtue that if our original function χ(τ ) has positive integer Fourier coefficients
(n)
am ∈ Z≥0 , the Hecke transform Tn χ(τ ) has positive integer Fourier coefficients n bm ∈ Z
as well; c.f. the discussion after (10.4.4). The application of Hecke operators Tn to a CFT
character thus produces another function, which itself has some properties expected of a CFT
character. In this sense, we might hope that the Hecke operators provide a “map between
CFTs,” as in the example of the Ising and Baby Monster CFTs mentioned above. We now
study this phenomenon in the simplest examples of holomorphically factorizable CFTs, i.e.
those for which the partition function is given simply as a product Z(τ, τ̄ ) = Z(τ )Z(τ ).
Let us consider a class of two-dimensional CFTs arising as the holographic duals to
three-dimensional gravity in anti-de Sitter space AdS3 . We will consider theories of pure
gravity, which in three dimensions are known to not have any propagating gravitational
waves. Indeed, three-dimensional gravity can be rewritten as a topological theory, namely
an SL(2, R) × SL(2, R) Chern-Simons theory specified by levels kL , kR . Here we work with
the case of kL = kR = κ. By comparing the Chern-Simons formulation of the theory with
the Einstein-Hilbert formulation, one can show that the curvature radius of AdS3 is related
to the level κ by ` = 16κGN , where GN is Newton’s constant. In particular, we see that ` is
quantized.
What do the tentative dual CFTs look like? By the Brown-Henneaux formula [125], the
CFTs should have central charges,

3`
c= = 24κ (10.10.9)
2GN

The vacuum energy is given by − 24c = −κ. Additional states can be obtained from the
vacuum state by application of the generators of the Virasoro algebra. In particular, the
generators Ln for n ≥ −1 annihilate the vacuum, while the generators L−n for n > 1 act
on the vacuum to give new states with energy −κ + n. For given energy −κ + n, the naive
degeneracy would be given by p(n), the number of integer partitions of n. For example, for
n = 3 we have p(3) = 3, corresponding to the three states,

L3−1 |0i L−1 L−2 |0i L−3 |0i (10.10.10)

170
This would give total vacuum contribution to the chiral partition function,
∞ ∞
?
X Y 1
Zκvac (τ ) = q −κ p(n)q n = q −κ (10.10.11)
n=0 n=1
1 − qn

as per the discussion in section 5.6.2. However, in the current case L−1 annihilates the
vacuum, and the corresponding degeneracies must be removed. This is accomplished by
including a factor of (1 − q), giving the actual vacuum contribution
∞
−κ
Y 1
Zκvac (τ ) =q (10.10.12)
n=2
1 − qn

The vacuum contribution to the partition function is not itself modular invariant, which
means that there must be additional states beyond the vacuum and its descendants. Indeed,
there must be some states in the CFT which capture the known black holes states of the
holographic dual. These black holes are expected on general grounds to appear at order
O(q 1 ), giving a full partition function of the form
∞
−κ
Y 1
Zκ (τ ) = q + O(q) (10.10.13)
n=2
1 − qn

Remarkably, there is a unique such partition function for every κ.

For κ = 1, i.e. pure AdS3 gravity with the smallest value of the cosmological constant,
the only possible candidate is

Z1 (τ ) = j(τ ) − 744
= q −1 + 196884 q + . . . (10.10.14)

with j(τ ) the j-function introduced in section 3.7. Physically, the 196884 states at level
1 can be interpreted as 1 descendant of the vacuum, and an additional 196883 black hole
states. The 2d CFT with this partition function was first constructed by Frenkel, Lepowsky,
and Merman and is known as the “Monster CFT,” due to it having global symmetry group
given by the Fischer-Greiss monster M, the largest finite sporadic group. The smallest non-
trivial representation of M has dimension 196883, and the black hole states indeed lie in this
representation.
To obtain partition functions for higher κ > 1, we now make use of Hecke operators.
Thus far in this section we have focused on Hecke operators acting on holomorphic modular
forms, but the extension to the meromorphic case is straightforward. The q-expansions of
Hecke transforms can be obtained using (10.4.3), giving e.g.

T2 J(τ ) = q −2 + 42987520 q + . . .
T3 J(τ ) = q −3 + 2592899910 q + . . . (10.10.15)

171
where we have defined J(τ ) = j(τ ) − 744. For any κ, it is easy to show that one has

Tκ J(τ ) = q −κ + O(q) (10.10.16)

Using this, we can now write down closed form expressions for Zκ (τ ) as follows. Say that
∞
X
Zκvac (τ ) = cr q r (10.10.17)
r=−κ

Of course, using (10.10.12) all cr are known. Then, the full chiral partition function is given
simply by
κ
X
Zκ (τ ) = c−r Tr J (10.10.18)
r=0

where we take T0 J(τ ) = 1. Using the formulas above, we get explicitly

Z2 (τ ) = q −2 + 1 + 42987520 q + . . .
Z3 (τ ) = q −3 + q −1 + 1 + 2593096794 q + . . . (10.10.19)

and so on. Though no concrete construction of CFTs with these partition functions are
known (assuming they actually exist), there exists much literature on the subject. Such
tentative theories are referred to as “extremal CFTs.” We have thus seen that the Hecke
operators act as maps between extremal CFTs.
Furthermore, by using Hecke operators we have obtained the exact number of quantum
black hole microstates in theories of pure gravity. In particular, this number is obtained by
subtracting the number of vacuum descendants from the coefficient of the O(q 1 ) term. It is
interesting to compare these results to those predicted by the Bekenstein-Hawking entropy,
p
SBH = 4π κL0 (10.10.20)

For example, for κ = L0 = 1 one finds SBH = 4π ≈ 12.57, to be compared to the exact result
Sexact = log 196883 ≈ 12.19. The agreement between the Bekenstein-Hawking entropy and
the exact microstate count improves in the semiclassical limit κ, L0 → ∞ with L0 /k fixed.
The origin of this agreement is in fact a classic result of Petersson and Rademacher. Writing
the Fourier expansion of j(τ ) as
∞
X
j(τ ) − 744 = cn q n (10.10.21)
n=−1

with c0 = 0, Petersson and Rademacher proved that as n → ∞, one has

√ 3 1
log cn ≈ 4π n − ln m − ln 2 + . . . (10.10.22)
4 2
172
We now combine this with the results for the Fourier coefficients of Hecke transforms. Since
we are working in the limit κ, L0 → ∞, it suffices to consider only the Tκ J contribution
(κ)
to (10.10.18). The dominant contribution to the corresponding Fourier coefficients bn are
(κ)
bn ≈ κcκn , which gives
√ 1 3 1
log b(κ)
n ≈ 4π κn + ln κ − ln n − ln 2 + . . . (10.10.23)
4 4 2
The first term reproduces the Bekenstein-Hawking result, while the others give the leading-
order quantum corrections.

• Bibliographical notes
The formal theory of Hecke operators was developed by Hecke in [126, 127], but the operators
had been introduced two decades earlier when Mordell proved the multiplication law of the
Ramanujan τ -function [128], reproduced in section 10.5. Most of the material discussed
above is by now standard, and may be found in the books by Apostol [8], Shimura [11],
Diamond and Shurman [13], and Serre [124]. The theory of Hecke operators acting on Maass
forms is described in the book by Terras [20]. The double coset construction is described in
detail in the book by Diamond and Shurman [13] and the book by Iwaniec [15]. The action
of Hecke operators on vector-valued modular forms was developed recently in both physics
[129, 130] and mathematics [131] literature.
The discussion of three-dimensional gravity given in section 10.10 follows closely the work
of [132]. For standard introductions to two-dimensional quantum field theory and conformal
field theory we refer again to [46, 45], already mentioned in section 5. The basic tools of
AdS/CFT are reviewed in [133, 134] as well as in the book by Ammon and Erdmenger [135].
The non-existence of several extremal CFTs was recently proven in [136], using the theory of
topological modular forms. The asymptotic form of the coefficients of the j-function quoted
in (10.10.22) was obtained by Petersson in [137], and independently by Rademacher [138].

173
11 Singular moduli and complex multiplication
In previous sections we introduced SL(2, Z) as the automorphism group for any lattice
Λ = Zω1 + Zω2 with Im (ω2 /ω1 ) > 0. Singular moduli correspond to special values of
τ = ω2 /ω1 for which the lattice has an automorphism group that is larger than SL(2, Z).
The mechanism by which this happens is referred to as complex multiplication.
A first rather obvious example is when τ = i, in which case the lattice has an extra
symmetry group {±I, ±S} of order four. A second obvious example is when τ = ρ = e2πi/3 ,
in which case the extra symmetry group {±I, ±(ST ), ±(ST )2 } is of order six. One way
of looking at these automorphisms is by complex multiplication, namely in the first case
we may take the periods to be ω1 = 1 and ω2 = i, and the extra non-trivial symmetries
correspond mapping ω1 → iω1 = ω2 and ω2 → iω2 = −ω1 , leaving the lattice invariant.

11.1 Conditions for complex multiplication

We seek the lattices for which there exists a complex number α ∈ C \ Z such that αΛ ⊂ Λ.
The condition may be written out explicitly on the periods as follows,

αω2 = aω2 + bω1

αω1 = cω2 + dω1 (11.1.1)

with a, b, c, d ∈ Z and ad − bc 6= 0. Note that we are not requiring ad − bc = 1 since we only

demand that αΛ be contained in Λ without necessarily being equal to Λ. Taking the ratio
gives a condition on τ , while given such a solution τ , the value of α is then determined,
aτ + b
τ= α = cτ + d (11.1.2)
cτ + d
The condition on τ is that it satisfies a quadratic polynomial equation,

cτ 2 + (d − a)τ − b = 0 (11.1.3)

with integer coefficients. The discriminant of this equation is denoted by D,

D = (a − d)2 + 4bc = (a + d)2 − 4(ad − bc) (11.1.4)

The restriction Im (τ ) 6= 0 imposes the condition D < 0, which in turn requires

√ bc < 0 and
ad − bc ≥ 1. Without loss of generality
√ we may assume that c > 0 and define D to be the
square root of D satisfying Im ( D) > 0, so that we have,
√ √
a−d+ D a+d+ D
τ= α= |α|2 = ad − bc (11.1.5)
2c 2
Extending the field of the rationals Q by τ gives the quadratic extension field Q(τ ).

174
11.2 Elliptic functions at complex multiplication points
When αΛ ⊂ Λ and α ∈ C \ Z, elliptic functions and modular forms at τ satisfy remarkable
properties. A first key result is for elliptic functions. It suffices to investigate the Weierstrass
function ℘(z) = ℘(z|Λ) for the lattice Λ. Since αΛ ⊂ Λ, the function ℘(αz) is an elliptic
function for the original lattice Λ and may therefore be expressed as a rational function of
℘(z). More specifically, we have,

A(℘(z))
℘(αz) = (11.2.1)
B(℘(z))

with A and B relatively prime polynomials of degrees related by,

deg A = 1 + deg B = |α|2 = ad − bc (11.2.2)

The only assertion above that needs proof is the relation between the degrees. First, since
℘(αz) has a double pole at z = 0 and since A and B will have poles at z = 0 of respective
degrees 2 deg A and 2 deg B, we must have the first equality above between the degrees.
Second, the lattice αΛ has a fundamental parallelogram which is scaled up from Λ by a
factor of |α|, so that the area of the fundamental parallelogram is |α|2 times that of the
fundamental parallelogram of Λ. Thus, the index is given by [Λ : αΛ] = |α|2 . While ℘(z) has
one double pole in the fundamental parallelogram of Λ, the total number of poles of ℘(αz)
in the fundamental parallelogram of Λ is 2|α|2 , which is also its number of zeros. It follows
that deg A = |α|2 giving the second equality.
Constructing the polynomials A, B is achieved by matching the Laurent expansions of
both sides at z = 0 using the Laurent series of ℘(z),
∞
1 X
℘(z) = + (2m + 1)G2m+2 z 2m (11.2.3)
z 2 m=1

where we have taken into account the fact that Gm = 0 for m odd. We shall also use the
fact that G2m for m ≥ 4 may be expressed as a polynomial in the generators G4 , G6 of the
ring of modular forms of SL(2, Z) of (3.5.22),
m−1
X 3(2k + 1)(2m − 2k + 1)
G2m+4 = G2k+2 G2m−2k+2 (11.2.4)
k=1
(m − 1)(2m + 3)(2m + 5)

Of course, the relation between ℘(αz) and ℘(z), and therefore the matching of the Laurent
series, will only be possible at the singular modulus τ for which we have the corresponding
complex conjugation factor α. Therefore, the matching will impose a relation between the
Eisenstein series G4 and G6 , and this will give a formula for the value of j(τ ).

175
11.3 Examples
√
Consider the simple example α = i 2 for which a + d = 0,√the discriminant is given by
D = −8, and |α|2 = 2. We may take the representative τ = i 2. In this case A has degree
two and B degree one, so that we may express the relation (11.2.1) as,
√ c3
℘(i 2 z) = c1 ℘(z) + c2 + (11.3.1)
℘(z) + c4
√
Since ℘(i 2 z) = −1/(2z 2 ) + O(z 2 ) we must have c1 = − 21 and c2 = 0. Substituting the
Laurent expansion we obtain the following expression,
−1
√ G26 2
  
1 2 1 G6 2 4
℘(i 2 z) + ℘(z) = − +5 + 25 2 z − 5G4 z + O(z )
2 9G4 z 2 G4 G4
2
 
1 G6 2 2 4
= ℘(z) + c4 + 25 2 z − 8G4 z + O(z ) (11.3.2)
c3 G4
To match the Laurent expansions we have set,
9 G6
c3 = − G4 c4 = 5 (11.3.3)
2 G4
But the relation must be exact to all orders in z by our global analysis, so the terms of
order z 2 and higher in the above parentheses must vanish. This can be achieved only at the
singular modulus associated with the complex multiplication factor α, and this gives us a
relation between G4 and G6 ,

25 G26 = 8G34 (11.3.4)

It may be checked that all higher order terms in the parentheses on the second line also
vanish. We thus find that the j-invariant at the complex multiplication point is given by,
−1
√ (12)3 g23 49 G26

3
j(i 2) = 3 = (12) 1 − = (20)3 = 8000 (11.3.5)
g2 − 27g32 20G34
√
Remarkably, j(i 2) is an integer, and in fact the cube of an integer !

11.3.1 Heegner points

A related set of points are those labelled by the so-called Heegner numbers. The Heegner

1
√ 
numbers can be defined as numbers d such that j 2 (1 + i d) ∈ Z. There exist precisely
eight such numbers,

1, 3, 7, 11, 19, 43, 67, 163 (11.3.6)

176
a fact first conjectured by Gauss and proven by Heegner, Baker, and Stark over a century
later. It is a remarkable fact that the j-invariant evaluates to a cube on each of them,

1

3

1
√ 
j 2 (1 + i) = 12 j 2 (1 + i 3) = 0
 √   √ 
j 12 (1 + i 7) = −153 j 12 (1 + i 11) = −323
 √   √ 
j 21 (1 + i 19) = −963 j 12 (1 + i 43) = −9603
 √  3
 √ 
1
j 2 (1 + i 67) = −5280 j 2 (1 + i 163) = −6403203
1
(11.3.7)

These are precisely the discriminants D = −d for which the class number is one and the
corresponding ring of integers has unique factorization.
Incidentally, the Heegner numbers are related to some well-known examples of “almost
integers.” Perhaps the most famous almost integer is Ramanujan’s constant,
√
eπ 163
= 262537412640768743.99999999999925 . . . (11.3.8)
1
√

whose near-integrality can be explained by using the value of j 2
(1 + i 163) given above
together with the q-expansion of j(τ ) to write
√ √
eπ 163
= 6403203 + 744 + O(e−π 163
) (11.3.9)

11.4 j(τ ) as an algebraic integer

An algebraic integer is a complex number that is a root of a monic polynomial with integer
coefficients, namely of a monic polynomial in Z[x]. Recall that a monic polynomial is
one whose leading monomial has coefficient 1. A general result regarding the arithmetic
properties of j(τ ) at points τ of complex multiplication is the following:

Theorem 11.1 For a lattice Z+τ Z with complex multiplication, j(τ ) is an algebraic integer.

To prove this result, we define the following monic polynomial in x for given τ ,
Y
Pn (x|τ ) = (x − j(γτ )) (11.4.1)
γ∈SL(2,Z)\Mn

where Mn is the set of 2×2 matrices with determinant n, introduced in (10.1.3). Since j(τ ) is
invariant under SL(2, Z) the product in γ is taken over the left cosets SL(2, Z)\Mn discussed
above (10.1.4). The function Pn (x|τ ) is invariant under τ → γτ for all γ ∈ SL(2, Z) because
the right action of SL(2, Z) on the left cosets SL(2, Z)\Mn simply permutes the cosets.

177
 Because the set of matrices,
 ab 
( 0 d )  ad = n, 0 ≤ b ≤ d − 1 (11.4.2)

provides a complete set of representatives of the quotient SL(2, Z)\Mn , as was explained in
section 10.1, we obtain the following equivalent product formula,
d−1   
Y Y aτ + b
Pn (x|τ ) = x−j (11.4.3)
ad=n b=0
d

This expression confirms that Pn (x|τ + 1) = Pn (x|τ ) since the shift in τ amounts to a shift
b → b + a (mod d) under which the product over b is invariant. The degree of Pn is given by,

deg(Pn ) = #(SL(2, Z)\Mn ) = σ1 (n) (11.4.4)

P
where σ1 (n) = d|n d is the divisor sum. The full structure of the τ -dependence of Pn (x|τ )
is given by the following Lemma.

Lemma 11.2 Pn (x|τ ) is a polynomial in x and j(τ ) with integer coefficients, so that there
exists a polynomial Qn (x, y) ∈ Z[x, y] which satisfies Pn (x|τ ) = Qn (x, j(τ )).

To prove the Lemma, we begin by focussing on the product over b for given d. Writing
the q-expansion of j(τ ) as j(τ ) = q −1 + 744 + k≥1 ck q k with ck ∈ N, we have,
P

!
− 2πib − ad 2πikb ka
Y Y X
aτ +b
= (−1)d



x−j d
e d q + 744 + ck e d q d −x
0≤b≤d−1 0≤b≤d−1 k≥1
`a
X
d −a
= (−1) q + a` (x)q d (11.4.5)
`≥1−d

By construction, the coefficients a` (x) are polynomials in x and the root of unity e2πi/d ,
2πi
namely a` (x) ∈ Z(x, e d ). Actually, (11.4.5) implies that the a` (x) are invariant under
2πi 2πir
all transformations σr : e d → e d for 0 ≤ r ≤ d − 1; that is, the so-called “Galois
2πi
automorphisms” of Q(e d )/Q, which will be further discussed in section 16. As a result,
2πi
a` (x) must be independent of e d and therefore belongs to Z[x]. Furthermore, the left side
of (11.4.5) is invariant under τ 7→ τ + 1, so that we must have a` (x) = 0 for all d - `.
The function Pn (x|τ ) is obtained by taking the product of (11.4.5) over ad = n, which we
reorganize as follows,
" # deg(Pn )
Y X X
Pn (x|τ ) = (−1)d q −a + amd (x)q ma = hr (τ )xr (11.4.6)
ad=n m≥0 r=0

178
 P (r) `
where the coefficient functions hr (τ ) admit a Fourier expansion hr (τ ) = ` a` q with
(r)
a` ∈ Z. Recall that any meromorphic modular function with integer Fourier coefficients
can be written as a polynomial in j(τ ) with integer coefficients; that is, we can write hr (τ ) =
P (r) ` (r)
` ã` j(τ ) with ã` ∈ Z. This allows us to conclude that the same is true of Pn (x|τ ),
thereby proving that Pn (x|τ ) is a polynomial in x and j(τ ) with integer coefficient. This
result may be summarized by the statement that there exists a polynomial Qn (x, y) with
integer coefficients such that,

Pn (x|τ ) = Qn (x, j(τ )) (11.4.7)

Actually Qn satisfies Qn (y, x) = ±Qn (x, y) and is of degree σ1 (n) in x and thus also in y.
To complete the proof of Theorem 11.1, we now specialize to a point τ = τCM ∈ H of
complex multiplication of order n, namely satisfying γ0 τCM = τCM for a matrix γ0 = ( ac db ) ∈
Mn with determinant n = ad − bc. The polynomial Pn (x|τCM ) = Qn (x, j(τCM ) must vanish
at the point x = j(τCM ) since, up to an SL(2, Z) transformation, γ0 will coincide with one of
the γ-matrices in the product in (11.4.1). Thus, the polynomial Qn (x, x) ∈ Z[x] with integer
coefficients vanishes at x = j(τCM ).
To complete the proof that j(τCM ) is an algebraic integer, it remains to show that it is
a root of a non-vanishing polynomial with integer coefficients. To this end, we evaluate the
leading behavior at the cusp of the polynomial Qn (x, x) ∈ Z[x] from (11.4.3) with the help
of the factorization formula,
d−1
Y
(x − e2πib/d y) = xd − y d (11.4.8)
b=0

and we obtain,
Y 
Qn (j(τ ), j(τ )) = q −d − q −a + lower order terms (11.4.9)
ad=n

There are now two cases to be considered, depending on whether n is a perfect square or
not. We first consider the case when n is not a perfect square. Every factor in the product
has a 6= d since otherwise n = ad = a2 is a perfect square and the leading behavior at the
cusp is given by the power of q obtained by summing the maxima of (d, a),
+ X
Qn (j(τ ), j(τ )) = ±q −σ1 (n) + lower orders σ1+ (n) = max(a, d) (11.4.10)
ad=n

Therefore, the polynomial Qn (x, x) is non-vanishing and monic (up to a sign) when n is not
a perfect square. When n = ν 2 is a perfect square, then one element in SL(2, Z)\Mn is of
the form νI so that a = d = ν and Qn (j(τ ), j(τ )) = 0. The vanishing is caused by the fact

179
that Qn (x, y) has a factor (x − y). Adapting the above arguments to this case, one shows
that the polynomial,
Qn (x, y)
Q̃n (x, y) = (11.4.11)
(x − y)
evaluated at x = y = j(τ ) is an integer multiple of a monic polynomial of degree σ1+ (n) − n.
This completes the proof of the Theorem.

11.4.1 The examples of n = 2, 3, 4, 5

Zagier worked out the example of n = 2 in detail. The cases n = 3, 4, 5 may be obtained
analogously, and we find the following factorizations of Qn (x, x),

Q2 (x, x) = −(x − 123 )(x − 203 )(x + 153 )2 (11.4.12)

Q3 (x, x) = −x(x − 203 )(x + 323 )(x − 54000)
Q̃4 (x, x) = 2(x + 153 )2 (x − 54000)2 (x − 663 )(x2 + 191025x − 121287375)2
Q5 (x, x) = (x − 123 )2 (x + 323 )2 (x − 663 )2 (x + 963 )2 (x2 − 1264000x − 681472000)
The second order polynomials factor as follows,
√ √
x2 + 191025x − 121287375 = (x − 21 (−191025 + 85995 5))(x − 12 (−191025 − 85995 5))
√ √
x2 − 1264000x − 681472000 = (x − 632000 − 282889 5)(x − 632000 + 282889 5) (11.4.13)

In addition to the entries listed in (11.3.7) we also recognize,

√
j(i) = 123 j(i 3) = 54000
√ √
j(2i) = 663 j(i 5) = 632000 + 282880 5
√  √  √

j(i 2) = 203 j 12 (1 + i 15) = − 12 191025 + 85995 5 (11.4.14)

11.5 ϑ-functions at complex multiplication points

√
We now examine Jacobi theta functions at some arguments τ = i n with n ∈ N. To do so,
we begin by defining the function,
Z π  
2 dx π 1 1 2
K(k) = p = 2 F1 , ; 1; k (11.5.1)
0 1 − k 2 sin2 x 2 2 2

known as a complete elliptic integral of the first kind. A valuable result involving elliptic
integrals and Jacobi theta-functions is that
r
2
ϑ3 (0|τ ) = K(k) (11.5.2)
π

180
where the parameter τ is defined in terms of the elliptic integrals as,
√ 1 1 2
K( 1 − k 2 ) 2 F1 ( 2 , 2 ; 1; 1 − k )
τ = iF (k) , F (k) = = 1 1 2
(11.5.3)
K(k) 2 F1 ( 2 , 2 ; 1; k )

In this presentation, the action of the generators T and S of SL(2, Z) corresponds to,

k2
T : τ →τ +1 k2 → −
1 − k2
1
S: τ →− k2 → 1 − k2 (11.5.4)
τ
which may be readily verified using the transformation formulas for the hypergeometric
function (see Bateman [1], Vol II page 318).
As we will now describe, the above relations between ϑ-functions and hypergeometric
functions can be used to evaluate ϑ3 (0|τ ) at special values of τ . Having obtained ϑ3 (0|τ ),
one can then obtain the values of other theta functions by means of,
√
ϑ2 (0|τ ) = k ϑ3 (0|τ ) ϑ4 (0|τ ) = (1 − k 2 )1/4 ϑ3 (0|τ ) (11.5.5)

11.5.1 The point τ = i

Let us begin by evaluating ϑ3 (0|i). In this case, (11.5.3) tells us to consider F (k) = 1, i.e.
√
K( 1 − k 2 ) = K(k) (11.5.6)

This is solved by k = √1 . Using this in (11.5.2) then gives

2

s  
2 1
ϑ3 (0|i) = K √ (11.5.7)
π 2

In order to evaluate the remaining elliptic integral, we convert it to a hypergeometric function

as per the second equality of (11.5.1) and make use of the following standard result,
√
π Γ( 12 (a + b + 1))
 
1 1
F
2 1 a, b; (a + b + 1); = (11.5.8)
2 2 Γ( 21 (1 + a))Γ( 12 (1 + b))

valid for arbitrary a, b. Using this, we conclude that,

1
π4
ϑ3 (0|i) = 3 (11.5.9)
Γ( 4 )

181
11.5.2 The point τ = 2i
Next consider ϑ3 (0|2i). We begin by searching for k such that F (k) = 2. To do so, we make
use of the elliptic integral identity,
√ !
2 k
F (k) = 2F (11.5.10)
1+k

which follows from a similar functional identity involving hypergeometric functions, (see for
example Bateman [139], Vol I, page 110)
 
1 1 4x 1 1 2


2 F1 , ; 1;
2 2
= (1 + x) 2 F 1 2
, 2
; 1; x (11.5.11)
(1 + x)2
Using this identity, we may recast our equation for k as follows,
√ !
2 k
F =1 (11.5.12)
1+k

The solution to this equation was identified in the previous subsection—we require,
√
2 k 1 √
=√ ⇒ k = ( 2 − 1)2 (11.5.13)
1+k 2
and thus we have
s
√
 
1 1
ϑ3 (0|2i) = 2 F1 , ; 1; ( 2 − 1)4 (11.5.14)
2 2

Making use of duplication and reflection identities of hypergeometric functions, one shows,

1 1
√ 4
 1
√ 2 F1 12 , 12 ; 1; 12


2 F1 2 , 2 ; 1; ( 2 − 1) = (11.5.15)
4−2 2
and then using (11.5.8) we obtain,
1
√
q
π4
ϑ3 (0|2i) = 2+ 2 (11.5.16)
2Γ( 43 )
√
11.5.3 The point τ = 2 i
√ √
Finally we consider ϑ3 (0| 2i). We would like to solve F (k) = 2, or equivalently,
√
2 2 F1 12 , 12 ; 1; k 2 = 2 F1 21 , 12 ; 1; 1 − k 2



(11.5.17)

182
 √
Comparing this to (11.5.11), we see that k = 2 − 1 is a solution. Thus we have,

√ √
r  
1 1
θ3 (0| 2i) = 2 F1 , ; 1; (
2 2
2 − 1)2 (11.5.18)

To proceed, we again use duplication and reflection identities of hypergeometric functions

to show that,

1 1
√ 2
 1 1 3
, ; 1; 12


2 F1 , ; 1; (
2 2
2 − 1) =p √ 2 F1 4 4
(11.5.19)
4−2 2

from which we can evaluate,

s
√ Γ( 9 ) Γ( 14 )
ϑ3 (0| 2i) = 85 (11.5.20)
Γ( 4 ) 21/4 π

11.5.4 The points τ = in and τ = i/n with n ∈ N

To close, we quote without proof the results for some other ϑ-values at imaginary complex
multiplication points ϑ3 (0|in). It is a general fact that one can write,

ϑ3 (0|i)
ϑ3 (0|in) = (11.5.21)
n1/4 hn

with ϑ3 (0|i) given in (11.5.9). The first few values of hn are found to be,

h1 = 1
√
q
h2 = 2 2−2
√
h3 = (2 3 − 3)1/4
23/4
h4 = = 1/4
q2 + 1
√
h5 = 5−2 5
√ √
23/4 31/8 (( 2 − 1)( 3 − 1))1/6
h6 = √ √ (11.5.22)
(−4 + 3 2 + 35/4 + 2 3 − 33/4 + 23/2 33/4 )1/3

From the above results, one can also obtain expressions for ϑ3 (0|i/n) by means of the fol-
lowing useful relation,
√
√
ϑ3 0|i/ n = n1/4 ϑ3 (0|i n) (11.5.23)

183
11.6 The values of E2 , E4 , E6 at the points τ = i, ρ
The quasi-modular transformation property of E2 (τ ),
12
E2 (γτ ) = (cτ + d)2 E2 (τ ) + c(cτ + d) (11.6.1)
2πi
allows one to evaluate E2 (τ ) at the points τ = i and τ = ρ = e2πi/3 . These points are fixed
points of the modular transformations γ = S, for which (c, d) = (1, 0), and γ = ST , for
which (c, d) = (1, 1), respectively. Substituting these expressions into the transformation
law (11.6.1), we readily obtain,
√
3 2 3
E2 (i) = E2 (ρ) = (11.6.2)
π π
The modular forms E4 and E6 vanish at the points ρ and i respectively, as was already
shown in (3.6.4) in terms of the closely related modular forms G4 and G6 . Evaluating E4
and E6 at the points i and ρ respectively requires expressing elliptic integrals in terms of
hypergeometric functions, and we shall quote here the result without proof,

48 Γ( 45 )4 729 Γ( 43 )6
E4 (i) = 2 3 4 E4 (ρ) = 0 E6 (i) = 0 E6 (ρ) = 3 5 6 (11.6.3)
π Γ( 4 ) 2π Γ( 6 )

For a derivation of the first and fourth relations, see Diamond and Shurman [13].

184
• Bibliographical notes
The algebraic nature of j(τ ) at complex multiplication points is a classic result; the discussion
given in section 11.4 follows the one given by Zagier in section 6.1 of [10]. Another useful
account may be found in the beautiful book by Cox [140].
The method used to evaluate ϑ-functions at special complex multiplication points in
section 11.5 was pioneered by Ramanujan, and can be found in his third notebook [141]; see
Chapter 17, Entry 6. It was systematized by Chowla and Selberg in [142, 143]. The explicit
values of hn listed in (11.5.22) were obtained in [144].
Generally speaking, points of enhanced symmetry tend to play a special role in physics,
and singular moduli are precisely the values of the moduli where such enhancements occur.
As such, singular moduli and complex multiplication have appeared in various places in
physics. Concrete examples can be found in Seiberg-Witten theory [145, 146, 147, 148, 149],
where points on the Coulomb branch corresponding to singular moduli generically have
extra massless degrees of freedom, as well as in string theory where the additional discrete
symmetries at singular values of the axion-dilaton can be gauged to give so-called “S-folds”
[150, 151, 152, 153, 154]. In section 13.6 we will introduce another physical application:
namely we will show that a non-linear sigma model CFT with T 2 target space is rational if
and only if the torus admits complex multiplication [155].
We close by noting that another context in which singular moduli appear is in the study
of orbifolds, in which one quotients by the emergent symmetry at such points. In the
context of perturbative string theory on a target space torus, orbifolds were first introduced
in [156, 157], and were further developed in [158] among many others publications. A large
number of references (with a particular emphasis on string phenomenology) can be found
in the bibliography of [159]. The study of singular moduli at genus-2 was carried out in
the 1960s by Gottschling in [160, 161, 162], though the application to genus-2 orbifolds is
fairly recent [163]. Some more general comments at arbitrary genus, from a two-dimensional
conformal field theory perspective, can be found in [164].

185
12 String amplitudes
We now finally turn to the topic of string theory. The starting point for string theory is
the assumption that elementary particles are one-dimensional objects, namely strings, in
contrast to quantum field theory where elementary particles are assumed to be point-like.
The size of strings is set by the Planck length `P ,
p
`P = GN ~/c3 ≈ 10−33 cm (12.0.1)

where GN is Newton’s gravitational constant, ~ is Planck’s constant, and c is the speed of

light. The Planck length is approximately 10−19 times the size of a proton. All distance
scales accessible to us to date are enormously larger than `P . At long distance scales or,
equivalently, at low energy scales strings effectively behave as point-like particles.

12.1 Overview
Being a one-dimensional object, a connected string may have two different topologies: that of
a circle or closed string, or that of an interval or open string. The time evolution of a closed
string sweeps out a two-dimensional surface with the topology of a cylinder, represented
in the left panel of Figure 10, while an open string sweeps out a rectangle. The surface
swept out by a string is referred to as the worldsheet. One of the most striking features of
string theory is that all interactions are governed by the joining and splitting of strings, as
represented in the right panel of Figure 10 for closed strings, without the need for the point-
like interactions required in quantum field theory. The interaction by joining and splitting
is unique once the free propagation of the string is known, since locally on the worldsheet
there is no way to distinguish free propagation from interaction.

Figure 10: Time evolution of a free closed string is represented in the left panel. Interaction
by joining and splitting of closed strings is represented in the right panel.

String theory may be promoted into a consistent quantum theory provided certain con-
ditions are met on the space-time in which strings propagate and interact. Superstrings can
propagate in flat space-time provided the dimension of the space-time is ten. The spectrum

186
of superstring theory then automatically contains a graviton, and string theory is auto-
matically a theory of quantum gravity, albeit in space-time dimension greater than four.
Supersymmetry, which swaps boson and fermion states, is a key ingredient in making su-
perstring theory mathematically and physically consistent. The center of attention in these
lectures is modular invariance which is responsible for making superstring theory UV finite.
The ultimate goal of string theory is the unification of gravity with the Standard Model
of Particle Physics into a single theory that is consistent with quantum mechanics and gen-
eral relativity. Successful unification will require principles and mechanisms for selecting our
four-dimensional universe from all the solutions of string theory, for breaking space-time su-
persymmetry, and for maintaining a small cosmological constant in the process. Satisfactory
answers to these questions remain largely outstanding today.
In this section we shall review some of the string amplitudes derived from superstring
perturbation theory and use those to obtain corrections to supergravity and super-Yang-
Mills theory in the form of local effective interactions. Superstring perturbation theory
uses a series expansion in powers of the string coupling gs to evaluate quantum mechanical
probability amplitudes, or simply amplitudes. The absolute value squared of an amplitude
gives the quantum mechanical probability for a given process to occur. A process is specified
by the data of the incoming and the outgoing string states, and the amplitude for the process
is schematically given by the Feynman functional integral prescription of summing over all
possible configurations of the string worldsheet given the initial and final state data.
The only topological information that remains, once the initial and final string data have
been fixed by specifying the process, is the genus h ≥ 0 of the surface, namely the number of
handles on the worldsheet. The string coupling gs provides the weight given to each topology,
so that the perturbative string amplitude A for a given process is provided by the following
topological expansion in powers of gs ,
∞
X
A= gs2h−2 A(h) + non-perturbative effects (12.1.1)
h=0
(h)
where A is the genus h (or h-loop) contribution to the string amplitude A. The expansion
is schematically represented in Figure 11. The prescription of the initial and final data for
the incoming and/or outgoing string states will be discussed in subsection 12.4.
The construction of string amplitudes that we shall sketch below contains only bosonic
states and the corresponding string theory is referred to as the closed bosonic string. To
include fermions introduces many technical complications and has led to the elaboration of
several possible different formulations, which are mostly beyond the scope of these lectures.
Instead we shall focus on the bosonic string, and some mild extensions thereof, to illustrate
the questions and results that relate to modular invariance of string perturbation theory
in this section, to T-duality for toroidal compactifications in the next section 13, and to
S-duality in Type IIB superstring theory in section 14.

187
 gs−2 + gs0 + gs2 +···

Figure 11: Schematic representation of the perturbative expansion in the string coupling gs
of the string amplitude A for a process of four incoming and/or outgoing string states.

12.2 String amplitudes as integrals over moduli space

Space-time is assumed to be a manifold M (or an orbifold) with a space-time metric G.
Physically, the signature of G is Minkowskian. However, as is familiar from quantum field
theory, it will often be convenient to construct string amplitudes in terms of a metric of
Euclidean signature, so that M is a Riemannian manifold, and then analytically continue
to Minkowski signature. String amplitudes are formulated in terms of two-dimensional ori-
entable surfaces Σ and continuous maps X : Σ → M . The Riemannian metric G on M
induces a Riemannian metric X ∗ (G) on Σ under the pull-back X ∗ of the map X. Since Σ is
orientable and carries a Riemannian metric it is automatically a Riemann surface.
The Polyakov formulation of string theory also invokes an intrinsic Riemannian metric g
on Σ, which is independent of the induced metric X ∗ (G). The map X and the intrinsic
metric g on Σ are governed by a worldsheet action. String amplitudes are required to be
independent of the coordinates used to parametrize Σ and M , which leads us to require
invariance under the diffeomorphism groups Diff(Σ) and Diff(M ). A natural candidate
satisfying these conditions is given by the following action,22
Z
1
IG [X, g] = 0
dµg g mn ∂m X µ ∂n X ν Gµν (X) (12.2.1)
4πα Σ
Feynman’s functional integral prescription for probability amplitudes gives the amplitudes
as integrals over the spaces Map(Σ) of maps X and Met(Σ) of metrics g. The corresponding
functional measures of integration Dg and DX may be constructed from suitable L2 norms
on the tangent spaces to Map(Σ) and Met(Σ). Denoting their tangent vectors by δX and
δg respectively, the norms are dictated by Diff(Σ) and Diff(M ) invariance,
Z Z
2 µ ν 2
kδXk = dµg Gµν (x) δX δX kδgk = dµg g mn g pq δgmp δgnq (12.2.2)
Σ Σ

22
Throughout, ξ m with m = 1, 2 are real local coordinates on Σ and ∂m = ∂/∂ξ m are the derivatives with
m
respect to ξ√ . The Riemannian metric on Σ takes the form g = gmn dξ m ⊗dξ n while the volume form is given
1
by dµg = 2 det g mn dξ m ∧ dξ n where mn = −nm specifies the orientation of Σ with 12 = 1. Throughout,
the Einstein summation convention on a pair of matching upper and lower indices will be implied, both for
the worldsheet indices m, n = 1, 2 as well as for the space-time indices µ, ν = 1, · · · , dim(M ).

188
The partial integral over Map(Σ) is given as follows,
Z
−WG [g]
e = DX e−IG [X,g] (12.2.3)
Map(Σ)

The two-dimensional quantum field theory defined by IG [X, g], for fixed metrics g and G,
is referred to as a non-linear sigma model. The quantum field theory on Σ defined by the
functional integral of DX depends on the space-time metric G, which is a function of X and
therefore may depend on an infinite number of couplings. This may be seen explicitly by
(0)
Taylor expanding G in powers of the field X in the neighborhood of a flat metric Gµν ,

1 (2)
Gµν (X) = G(0) (1) ρ ρ σ 3
µν + Gµν;ρ X + Gµν;ρσ X X + O(X ) (12.2.4)
2
(0) (1) (2)
The sets of coefficients Gµν , Gµν;ρ , Gµν;ρσ , · · · may be viewed as independent coupling pa-
rameters of the quantum field theory corresponding to the non-linear sigma model. This
theory is renormalizable in the sense of Friedan namely that, upon insisting on invariance
under Diff(M ), the number of counter-terms is finite at each order in α0 .
The genus h contribution A(h) to the string amplitude for metric G at genus h is formally
given by the functional integral over Met(Σ),
Z
(h)
A ∼ Dg e−WG [g] (12.2.5)
Met(Σ)

where the symbol ∼ indicates that this relation is formal. Taken literally, the integral over g
would be divergent as the measure and the integrand are both invariant under Diff(Σ) so
that Met(Σ) contains an infinite number of images of each metric g under Diff(Σ). Factoring
out Diff(Σ) amounts to a reduction of the integration to the quotient Met(Σ)/Diff(Σ) of
orbits, and may be carried out using the Faddeev-Popov gauge fixing procedure familiar
from Yang-Mills gauge theory. The Faddeev-Popov determinant produced in the process
may be represented by a functional integral over ghost fields b and c with action Igh [b, c, g],
Z Z
−Wgh [g] −Igh [b,c,g] 1
e = D(bc)e Igh [b, c, g] = dµg (b∂z̄ c + c.c.) (12.2.6)
2π Σ

The resulting amplitude is then given by,

Z
A =(h)
Dg e−WG [g]−Wgh [g] (12.2.7)
Met(Σ)/Diff(Σ)

The reduction removes two of the three functional degrees of freedom of a two-dimensional
metric, leaving a single real field. Such theories are referred to as non-critical strings.

189
 Critical string theories are obtained by requiring that the combination of the measure
Dg and the integrand are invariant also under Weyl transformations. The group of Weyl
transformations Weyl(Σ) leaves the metric G and the map X invariant, and multiplies the
metric g by a positive scalar function, gmn (ξ) → e2σ(ξ) gmn (ξ), where σ : Σ → R. The quotient
of the space Met(Σ) of metrics by the semi-direct product of Diff(Σ) and Weyl(Σ) may be
identified with the moduli space Mh of conformal structures on Σ, and is isomorphic to the
space of complex structures on a Riemann surface Σ of genus h,
 
Mh = Met(Σ)/ Diff(Σ) n Weyl(Σ) (12.2.8)

Thus, our final expression for the genus h contribution A(h) to the amplitude for the critical
string is given by a finite-dimensional integral over moduli space,
Z
(h)
A = Dĝ e−WG [g]−Wgh [g] (12.2.9)
Mh

where Dĝ is the integration measure on moduli induced on the quotient. The resulting
amplitude A(h) is a function of the metric G on the space-time M .

12.3 Conformal invariance and decoupling negative norm states

The classical actions IG [X, g] and Igh [b, c, g] are Weyl-invariant, but the measures DX, Dg,
and D(bc) are not Weyl invariant, thereby producing Weyl anomalies. In particular, the
functional WG [g], defined in (12.2.1), is not invariant under Weyl(Σ). For general metric G,
the Weyl transformation law of WG is complicated and not known explicitly. However, when
the transformation law of WG takes the following special form,
Z
c
δWG [g] = dµg Rg δσ (12.3.1)
24π Σ

where Rg is the scalar curvature of the metric g and c is a constant, then IG produces a
conformal field theory with central charge c. For example, when G is the flat metric on
RD , then the action IG is quadratic in X, and defines a conformal field theory with central
charge c = D, and Igh is a conformal field theory with central charge c = −26. In the critical
dimension D = 26 the combined field theory of X, b, c is therefore Weyl-invariant.
When the metric G is not flat, the Weyl transformation of WG may be evaluated in an
expansion in powers of α0 , which is equivalent to an expansion in powers of the Riemann
tensor of the metric G and its derivatives. To leading order in α0 , one finds,
Z  
1 D 1 mn
δWG [g] = dµg Rg + g ∂m X̄ ∂n X̄ Rµν (X̄) δσ + O(α0 )
µ ν
(12.3.2)
4π Σ 6 2

190
The first term is familiar from (12.3.1) for c = D. The second term involves the Ricci
tensor R of the metric G and an arbitrary background field X̄ used to evaluate the quantum
corrections in the background field method. The term in Rµν cancels when the metric G is
Ricci flat. Thus we find the remarkable fact that the requirement of Weyl symmetry on the
worldsheet metric g imposes Einstein’s equations on the space-time metric G.
Conformal symmetry plays a fundamental role in decoupling the negative norm and null
states which arise in the Lorentz-covariant formulation of string theory in flat Minkowski
space-time M = R26 . The maps X : Σ → M satisfy the Laplace equation ∂z ∂z̄ X µ = 0, so
that the field ∂z X µ is holomorphic, while the field ∂z̄ X µ is anti-holomorphic. Both fields
may be expanded in a Laurent series,
X X
i∂z X µ = z −m−1 Xm
µ
− i∂z̄ X µ = z̄ −m−1 X̃m
µ
(12.3.3)
m∈Z m∈Z

which upon canonical quantization satisfy,

µ † µ µ
(Xm ) = X−m [Xm , Xnν ] = mδm+n,0
µ † µ µ
(X̃m ) = X̃−m [X̃m , X̃nν ] = mδm+n,0 (12.3.4)

while Xm µ
and X̃nν commute with one another.23 The operators X0µ and X̃0µ are self-adjoint.
They commute with one another and with all other modes and correspond to left and right
moving momentum operators. (They will play a key role in toroidal compactification and will
be discussed in greater detail in section 13.) Thus the free fields ∂z X µ and ∂z̄ X µ decompose
into X0µ , X̃0µ in addition to an infinite number of decoupled harmonic oscillators Xm µ µ
, X̃m for
m 6= 0, each of which satisfies a Heisenberg algebra.
The Fock space of the closed bosonic string in flat Minkowski space-time R26 is con-
structed as follows. The operators X0µ and X̃0µ may be diagonalized simultaneously; their
eigenvalues are real and are denoted k µ and k̃ µ ; their eigenvectors |0; ki and |0; k̃i are ground
states of the string excitation spectrum of momenta k µ and k̃ µ respectively. In flat Minkowski
space-time, the left and right momenta must be equal k̃ µ = k µ . (This condition will be re-
laxed for toroidal compactifications in order to accommodate winding modes, as will be
discussed in section 13.) The Fock space is the sum over string momenta k ∈ R26 of the
tensor product of left and right chiral Fock spaces,
M
Fclosed = Fk ⊗ F̃k (12.3.5)
k∈R26

23
We note that the fields ∂z X µ and ∂z̄ X µ are treated as independent of one another, and not as complex
conjugates of one another, a property that is inherited from the worldsheet with Minkowskian signature
where the fields correspond to the data on the two light-cone directions.

191
The chiral Fock space Fk is the infinite tensor product of the Fock spaces of the harmonic
oscillators Xnµ for n 6= 0. Concretely, we define the ground state |0; ki to satisfy,

X0µ |0; ki = k µ |0; ki Xnµ |0; ki = 0 n>0 (12.3.6)

The ground state is then a scalar under the Lorentz group and Fk is obtained as follows,
M n o
Fk = xµ−n
1
1
· · · x µ`
−n` |0; ki (12.3.7)
`≥0
n1 ,n2 ,··· ,n` >0

with the analogous construction for F̃k . We define a norm on these spaces by normalizing
the string ground state for given momentum by k|0; kik2 = 1. The lowest excited state in
µ
Fk with polarization vector εµ (k) is given by εµ (k)X−1 |0; ki and has the following norm,
µ
kεµ (k)X−1 |0; kik2 = εµ (k)εµ (k) (12.3.8)

The norm is respectively positive, null, or negative when εµ (k) is space-like, light-like, or
time-like. Negative norm states are inconsistent with the principles of quantum mechanics,
but they are unavoidable byproducts of the Lorentz-covariant quantization of the string, just
as they are unavoidable in the covariant quantization of gauge theories.
In flat 26-dimensional space-time negative norm states and null states may be decoupled
with the help of conformal symmetry, whose Virasoro generators take the form,
X1 1 X
Lm = Xm−n · Xn L0 = X02 + X−n · Xn (12.3.9)
n∈Z
2 2 n∈N

The conformal transformations by the Virasoro algebras act as follows on the oscillators,
µ
[Lm , Xnµ ] = −nXm+n (12.3.10)

and similarly for the tilde operators. One defines the subspace of physical states Fphys
k ⊂ Fk
by imposing the following conditions on the states |ψi ∈ Fk ,

(L0 − 1)|ψi = Lm |ψi = 0 m>0 (12.3.11)

All negative norm states are eliminated by these physical state conditions, and all null states
decouple, just as is the case when imposing Gauss’s law in gauge theory. The ground state is
physical provided k 2 = 2 which makes it a tachyon. The tachyon renders the bosonic string
vacuum unstable but it is eliminated in the superstring, and we shall omit its discussion in
µ
the sequel. Returning to the example of the first excited state, |ψi = εµ (k)X−1 |0; ki, we see
that the physical state conditions Lm |ψi = 0 for m ≥ 2 are automatically satisfied, while
those for m = 1, 0 impose the constraints k · ε(k) = 0 and the massless condition k 2 = 0

192
on the momenta. The resulting physical states constitute the spectrum of the open bosonic
string and include a massless vector particle, such as the photon.
The above construction of the Fock space also shows that the closed bosonic string
µ ν
spectrum contains the massless states εµ (k)ε̃ν (k)X−1 X̃−1 |0; ki where k 2 = 0 and k · ε =
k · ε̃ = 0. The traceless symmetric part of εµ (k)ε̃ν (k) corresponds to the graviton state as
expected from the presence of the metric G in the worldsheet action. However, there is
also a trace-part which corresponds to the scalar dilaton state, and an anti-symmetric part
which corresponds to a rank-two anti-symmetric state. These additional states and particles
are not required by general relativity, but their presence in the spectrum will, in fact, be
mandated by supergravity, as we shall see in section 14.
In view of the presence of the extra states of the dilaton and rank two anti-symmetric
tensor field, we may generalize the non-linear sigma model action (12.2.1) to include a field
Φ for the dilaton as well as a rank two anti-symmetric field Bµν ,24
Z 
1 
IG,Φ,B [X, g] = 0
dµg g Gµν (X)− iε Bµν (X) ∂m X µ ∂n X ν
mn mn
4πα Σ

0
+2α Rg Φ(X) (12.3.12)

where Rg is the scalar curvature of the metric g, already encountered in (12.3.1). In the
presence of the fields G, Φ, B conformal invariance requires, up to lowest non-trivial order in
the α0 -expansion, a generalization of Einstein’s equations that includes the dynamics of the
fields Φ and Bµν . Those field equations may be derived from the following action,
Z  
1 −2Φ µ 1 µνρ
S[G, Φ, B] = 2 dµG e RG + 4Dµ ΦD Φ − Hµνρ H (12.3.13)
2κ 12
where κ is a constant, Dµ is the covariant derivative, and Hµνρ = Dµ Bνρ + Dν Bρµ + Dρ Bµν .
In summary, conformal symmetry is the local gauge symmetry that eliminates negative
norm states and decouples null states. In this section we have illustrated this mechanism
when space-time is flat, but the result is expected to hold in space-times which are asymp-
totically flat or anti-de Sitter.

12.4 String amplitudes in terms of vertex operators

Conformal symmetry also allows us to simplify the formulation of string scattering pro-
cesses. Instead of representing each incoming and outgoing string in a scattering process by
a boundary of the string worldsheet, as was presented in Figure 11, we can encode all the
24
The factor of i multiplying the Bµν term is present for worldsheet metric g of Euclidean signature, but
is absent when Minkowski signature is used.

193
physical information of the incoming or outgoing strings in terms of vertex operators. The
data for these vertex operators is specified at vertex insertion points on the worldsheet, as
illustrated in Figure 12. One may view these insertion points as the limit in which the radii
of the boundary discs of Figure 11 are shrunk to zero by a conformal transformation.

gs−2 + gs0 + gs2 +···

Figure 12: Schematic representation of the perturbative expansion in the string coupling gs
formulated in terms of vertex operator insertions on the worldsheet.

Consider the scattering of N on-shell gravitons in flat space-time R26 , with momenta ki
and polarization tensors εi;µν (k) for i = 1, · · · , N . A graviton of momentum ki introduce
a small ripple on the metric of space-time, given in terms of a plane monochromatic wave.
Adding up the contributions of the N gravitons the space-time metric is perturbed as follows
to lowest order in the amplitude of the wave given by the polarizations tensors εi ,
N
X
Gµν (X) = ηµν + εi;µν (k) eiki ·X + O(ε2i ) (12.4.1)
i=1
Conformal invariance requires G to satisfy Einstein’s equations, as was discussed in the
second paragraph of subsection 12.3. Einstein’s equations apply in linearized form, because
we consider the metric only to first order in each εi , and impose the conditions ki2 = 0
and kiµ εi;µν (k) = 0. Substituting the expression (12.4.1) for the metric fluctuations into
the action IG [X, g] of (12.2.1) and the functional integral of (12.2.9), and retaining the
contribution which is linear in each εi gives the genus h contribution to the amplitude for
the scattering of N gravitons in terms of vertex operators,
Z Z
(h)
A = Dĝ DX V1 [X, g] · · · VN [X, g] e−IG [X,g]−Wgh [g] (12.4.2)
Mh Map(Σ)

The vertex operators are read off the expansion in εi to linear order and are given by,
Z
Vi [X, g] = εi;µν (k) dµg g mn ∂m X µ ∂n X ν eiki ·X (12.4.3)
Σ
The physical state conditions for the graviton, namely its massless condition ki2 = 0, and the
transversality of its polarization tensors kiµ εi;µν (k) = 0, and εi;µν εµν
i = 0, guarantee that the
unintegrated vertex operator g mn ∂m X µ ∂n X ν eiki ·X has conformal dimension (1, 1) and may
be integrated over Σ consistently with conformal symmetry. One may proceed analogously
for the fields Φ and Bµν .

194
12.5 Superstring amplitudes
Scattering amplitudes for the superstrings may similarly be formulated in terms of functional
integrals over the string degrees of freedom, such as X µ , b, c, involving vertex operators.
Unlike for the case of the bosonic string, several different formulations have been developed
over time for the case of the superstring, each with certain advantages and disadvantages. Of
central concern to us here is not so much how the amplitudes have been obtained but rather
what their structure is. Thus, we shall proceed below by simply quoting the final expressions
for the amplitudes we need, and refer to the literature for their detailed derivations, which
are often quite involved. We shall concentrate on Type II theories here, but a similar analysis
may be carried out for Type I and Heterotic superstrings.
For definiteness, we shall consider the scattering process that involves four gravitons,
whose momenta we denote by ki and whose polarization tensors are chosen in factorized
form εµi (k)ε̃νi (k) with ki2 = ki · εi = ki · ε̃i = 0 and i = 1, 2, 3, 4. It will be useful to introduce
the dimensionless kinematic Lorentz-invariants defined by,
α0
sij = − (ki + kj )2 (12.5.1)
4
These variables satisfy a number of kinematic relations as a result of momentum conservation
and the massless conditions ki2 = 0, which may be solved as follows,

s = s12 = s34
t = s14 = s23
u = s13 = s24 s+t+u=0 (12.5.2)

Closed superstring perturbation theory produces the following on-shell four-graviton ampli-
tude in Type II superstring theory,
∞
X
A(εi ; ki ) = κ210 R4 gs2h−2 A(h) (sij ) (12.5.3)
h=0

where R stands for the on-shell linearized Riemann tensor, R4 for a particular scalar invariant
constructed out of R, and A(h) is a Lorentz scalar function that depends only on s, t, u. The
fact that a single kinematic combination R4 is shared by the amplitudes at all loop order is a
consequence of the space-time supersymmetry of the Type II superstring, and does not hold
for other superstring theories such as Type I or Heterotic strings. The particular structure
of the Lorentz contractions needed to form the scalar R4 is also a direct consequence of
supersymmetry. In the factorized basis for the polarization tensors for the gravitons, the
combination R4 is itself factorized,

26 R4 = K K̃ (12.5.4)

195
where K depends only on the polarization vectors εµi and K̃ depends only on the polarization
vectors ε̃µi . In terms of the linearized field strengths,

fiµν = kiµ ενi − kiν εµi

f˜iµν = kiµ ε̃νi − kiν ε̃µi (12.5.5)

the factor K is given as follows,

K = (f1 f2 )(f3 f4 ) + (f1 f4 )(f2 f3 ) + (f1 f3 )(f2 f4 )

−4(f1 f2 f3 f4 ) − 4(f1 f2 f4 f3 ) − 4(f1 f3 f2 f4 ) (12.5.6)

where the parentheses are defined by (fi fj ) = fiµν fjνµ and (fi fj fk f` ) = fiµν fjνρ fkρσ f`σµ . The
expression for K̃ is given by the above expression with f → f˜.
Explicit formulas for A(h) have been established from first principles for tree-level h = 0,
one-loop h = 1, and two loops h = 2, and are given as follows,
1 Γ(1 − s)Γ(1 − t)Γ(1 − u)
A(0) (sij ) =
stu Γ(1 + s)Γ(1 + t)Γ(1 + u)
|dτ |2
Z
(1) π
A (sij ) = B (1) (s, t, u|τ )
16 M1 (Im τ )2
|dΩ|2
Z
(2) π
A (sij ) = 3
B (2) (s, t, u|Ω) (12.5.7)
64 M2 (det Im Ω)

The integrands B (1) and B (2) are dimensionless Lorentz scalar functions of s, t, u given by,
4
( )
d2 zi
Z Y X
(1)
B (s, t, u|τ ) = exp sij G(zi , zj |τ )
Σ4 i=1 Im τ i<j
( )
2
|Y|
Z X
B (2) (s, t, u|Ω) = 2
exp sij G(zi , zj |Ω) (12.5.8)
Σ4 (det Im Ω) i<j

Here, zi are the vertex insertion points on the surface Σ, τ is the modulus used here as a
local complex coordinate for the genus-one moduli space M1 , and Ω is the period matrix
used here as a set of local coordinates for genus-two moduli space M2 , both of which are
defined in appendix B. Furthermore, in the genus-two amplitude, Y is a holomorphic (1, 0)
form in each one of the vertex points zi given by,

Y = (t − u)∆(1, 2)∆(3, 4) + (s − t)∆(1, 3)∆(4, 2) + (u − s)∆(1, 4)∆(2, 3) (12.5.9)

where ∆ is a holomorphic (1, 0) form in its entries defined by,

∆(i, j) = ω1 (zi )ω2 (zj ) − ω2 (zi )ω1 (zj ) (12.5.10)

196
and ωI (z) are the canonically normalized holomorphic Abelian differentials on a Riemann
surface of genus 2 (see appendix B). In B (1) (s, t, u|τ ), the Green function G(xi , zj |τ ) is the
scalar Green function on the torus of modulus τ defined in (5.5.9) and given explicitly
either in terms of Kronecker-Eisenstein series in (5.5.8) or Jacobi ϑ-functions in (5.5.14). In
B (2) (s, t, u|Ω), the Green function G(zi , zj |Ω) is the Arakelov Green function on the genus-two
Riemann surface Σ with period matrix Ω, which we shall now construct.
Analogous formulas hold for amplitudes with more than four external states, as well as
for Type I and Heterotic strings.

12.5.1 The Arakelov Green function for genus 2 Riemann surfaces

To construct the Arakelov Green function on a genus-two Riemann surface Σ, we begin by
defining the canonical Kähler form κ on Σ which is given by the pull-back under the Abel
map of the canonical Kähler form on the Jacobian J(Σ) (see appendix D),
2 Z
i X
κ= Y −1 )IJ ωI ∧ ω̄J κ=1 (12.5.11)
4 I,J=1 Σ

where Y = Im Ω. In terms of local complex coordinates z, z̄ on Σ, we have κ = 2i κzz̄ dz ∧ dz̄,

and the Arakelov Green function is defined by,
Z
∂z̄ ∂z G(z, w|Ω) = −πδ(z, w) + πκzz̄ (z) κ(z)G(z, w|Ω) = 0 (12.5.12)
Σ

It may be constructed explicitly in terms of the prime form and Abelian integrals,

G(z, w|Ω) = G(z, w|Ω) − γ(z|Ω) − γ(w|Ω) + γ 1 (Ω)

Z z  Z z 
2 −1 IJ
G(z, w|Ω) = − ln |E(z, w|Ω)| + 2πIm ωI (Y ) Im ωJ (12.5.13)
w w

where,
Z Z
1
γ(z|Ω) = κ(w)G(z, w|Ω) γ (Ω) = κ(z)γ(z|Ω) (12.5.14)
Σ Σ

12.5.2 Physical singularity structure of the four-graviton amplitudes

The singularity structure of the genus-one and genus-two amplitudes is as follows. For
fixed moduli, the singularities of B (h) as a function of s, t, u are governed by the operator
product expansion of the graviton vertex operators, namely the conformal field theory of free
scalar fields on Σ. The Green functions G(z, w|τ ) and G(z, w|Ω) are smooth throughout Σ,
except for a logarithmic singularity at z = w where they both behave as − ln |z − w|2 plus

197
regular terms. As a result, the integrations which define B (h) are absolutely convergent for
Re (sij ) < 1. The analytic continuation of B (h) may be carried out to the complex plane in
each variable and produces simple poles at positive integers s, t, u ∈ N.
The further integration over moduli, required to obtain the physical amplitude A(h)
for h = 1, 2, is absolutely convergent only when Re (sij ) = 0. Analytic continuation in
sij to the complex plane may be carried out using a decomposition of moduli space into
subregions in each of which the domain of convergence is enlarged. This procedure has been
carried out explicitly only for genus one. The resulting analytic continuation has branch cuts
in sij starting at any non-negative integer, which signals two-particle intermediate states.
The analytic continuation process will be illustrated in the low energy expansion, near the
massless branch cut starting at sij = 0 below.

12.6 Ultraviolet finiteness from modular invariance

An alternative presentation of the genus-one and genus-two amplitudes may be given by
introducing internal loop momenta. We shall illustrate this here for the case of genus one.
One begins by choosing a canonical homology basis of cycles A, B on the torus Σ, and
chooses the loop momentum p to flow through the cycle A. The amplitude is then given in
terms of a Hermitian pairing between left and right chiral amplitudes,
Z Z Z
(1)
A (sij ) = dp F(zi , ki , p|τ ) F(zi , −ki , p|τ ) (12.6.1)
R10 M1 Σ4

where we have suppressed the dependence on the polarization tensors. The chiral amplitude
F(zi , ki , p|τ ) is locally holomorphic in τ and in zi for i = 1, · · · , 4. Its explicit form is given
by the following top holomorphic differential form on M1 × Σ4 ,
4
iπτ p2 +2πi
P Y Y
ki zi −sij
F(zi , ki , p|τ ) = e i ϑ1 (zi − zj |τ ) dτ dzi (12.6.2)
i<j i=1

The price to pay for the local holomorphicity is that F has non-trivial monodromy when a
point z` is taken around one of the homology cycles of the surface, and we have,

F(zi + δi,` A, ki , p|τ ) = e2πik` ·p F(zi , ki , p|τ )

F(zi + δi,` B, ki , p|τ ) = F(zi , ki , p + k` |τ ) (12.6.3)

When the point z` is taken around the point zi with ` 6= i, the chiral amplitude F is multiplied
by a phase factor e−2πisij . Modular transformations of F involve a change of homology basis
A, B and thus a change of momentum routing through the torus. The Hermitian pairing of
F and F̄ is familiar from two-dimensional conformal field theory where the loop momentum
p labels the conformal blocks of ten copies of a c = 1 theory.

198
 Thanks to modular invariance, all string amplitudes are ultraviolet (UV) finite. The
result was obtained by Shapiro for the bosonic string at genus one, but holds for all modular
invariant superstrings to all genera.
We can illustrate the mechanism by which the genus-one string amplitude is UV finite by
inspection of the chiral amplitude derived in the previous paragraph. For any string theory,
the chiral amplitude, with loop momentum p chosen to flow through the cycle A, has the
following universal pre-factor which is Gaussian in the loop momentum,
2
F(zi , εi , ki , p|τ ) = eiπτ p × exponentials and powers in p (12.6.4)
The Gaussian factor is universal in the sense that it is independent of the momenta ki
and polarization tensors εi of the external states and actually independent of the specific
string theory under consideration. Modular invariance allows one to choose the fundamental
domain M1 for H/SL(2, Z) that is adapted to the particular choice for the flow of loop
momentum. For our choice of loop momentum traversing the A-cycle, the choice of funda-
mental domain is the standard one, M1 = {τ ∈ H, |τ | ≥ 1, |Re (τ )| ≤ 21 }. The high energy
behavior of the loop momentum is governed by the magnitude of Im (τ ), which is uniformly
bounded away from zero by our choice of fundamental domain. The uniform Gaussian sup-
pression at large loop momenta implies UV finiteness as all other factors are either only
exponential or polynomial in p. Higher genus amplitudes have a generalization of the above
factor which involves the period matrix of the corresponding higher genus surface, but the
structure is otherwise analogous to the one-loop case.

12.7 Effective interactions from the four-graviton amplitude

The tree-level, one-loop, and two-loop string amplitudes listed in (12.5.7) for four gravitons
have important applications to the evaluation of string theory corrections to supergravity in
the low energy limit. Type IIB supergravity and its relation to Type IIB superstring theory
will be reviewed in section 14. In preparation for those discussions and the comparison
of perturbative results with the implications from supersymmetry and SL(2, Z)-duality of
Type IIB string theory, we shall here evaluate the behavior of the partial amplitudes in
(12.5.8) and the full amplitudes in (12.5.7) in the low energy limit.
To illustrate the emergence of low energy effective interactions from superstring am-
plitudes, we recall the expansion in powers of the string coupling gs of the four-graviton
amplitude, given in (12.5.3) and repeated here for convenience,
∞
X
A(εi ; ki ) = κ210 R4 gs2h−2 A(h) (s, t, u) (12.7.1)
h=0

where R stands for the on-shell linearized Weyl tensor, whose expression was given in (12.5.4)
and (12.5.6), while the expressions for A(0) , A(1) , A(2) were given in (12.5.7).

199
12.7.1 Low energy expansion at tree-level
We concentrate first on the tree-level contribution A(0) and use the following series expansion
for the ratio of Γ-functions,
(∞ )
Γ(1 − s)Γ(1 − t)Γ(1 − u) X 2ζ(2n + 1)
= exp (s2n+1 + t2n+1 + u2n+1 ) (12.7.2)
Γ(1 + s)Γ(1 + t)Γ(1 + u) n=1
2n + 1

The argument of the exponential must clearly be odd in s, t, u, since the left side is mapped to
its inverse under simultaneous sign reversal of s, t, u. The linear term cancels by s+t+u = 0.
The amplitude A(0) is a symmetric function of s, t, u and may therefore be expanded in powers
of the two remaining symmetric polynomials in s, t, u,

σ2 = s2 + t2 + u2 σ3 = s3 + t3 + u3 = 3stu (12.7.3)

The first few terms in the expansion in powers of σ2 and σ3 are given by,
1 2 1
A(0) = + 2ζ(3) + σ2 ζ(5) + σ3 ζ(3)2 + σ22 ζ(7) + O(sij )5 (12.7.4)
stu 3 2
The first term on the right side in the expansion is non-analytic at low energy and corresponds
to the exchange of massless gravitons and other massless bosons in Type IIB supergravity.
The remaining terms on the right side are all local and successively correspond to effective
interactions which are schematically of the form D2k R4 for k = 0, 2, 3 and 4. The effective
interaction D2 R4 is missing in this list because of the relation s + t + u = 0.

12.7.2 Transcendental weight

An important notion that has arisen over the past few years, starting in the study of quantum
field theory amplitudes, is that of transcendental weight. A natural starting point for its
definition is with zeta-values and multiple zeta-values,
∞
X 1 X 1
ζ(k) = ζ(k1 , · · · , kp ) = k
(12.7.5)
n=1
nk k1
n1 ≥n2 ≥···≥np ≥1 n1 · · · npp

for integer k, k1 ≥ 2 and k2 , · · · , kp ≥ 1. When k is even, ζ(k) ∈ π k Q. Transcendental weight

provides a grading upon multiplication and one assigns weight 0 to any algebraic number,
weight 1 to π, and thus weight k to ζ(k) for k even. By extension, one assigns weight k to
ζ(k) for all k and weight k = k1 + · · · + kp to the multiple zeta value ζ(k1 , · · · , kp ).
With these assignments of transcendental weight, we observe that the argument of the
exponential on the right of (12.7.2) has transcendental weight 0 provided we assign weight
−1 to s, t, and u. Each term in the expansion of the tree-level amplitude in (12.7.4) thus
has weight 3. The amplitude is said to exhibit uniform transcendentality.

200
 Remarkably, string theory has non-trivial transcendentality properties already at tree-
level. This is in contrast with the case of quantum field theory where tree-level amplitudes
have trivial transcendentality and it is only through the expansion of the Γ-functions that
arise in dimensional regularization at loop level that non-trivial transcendentality arises.
Uniform transcendentality, observed here for tree-level Type II string amplitudes, is shared
in quantum field theory by the maximally supersymmetric N = 4 Yang-Mills theory in four
dimensions. A thorough understanding of how this property arises, and for precisely which
correlators it holds, remains an open problem.

12.8 Genus one in terms of modular graph functions

The genus-one amplitude A(1) is obtained in (12.5.7) by integrating the partial amplitude
B (1) (s, t, u|τ ), which is a modular function in τ , over the genus-one moduli space M1 . The
partial amplitude in turn is given by the integration over four copies of the torus Σ with
modulus τ in (12.5.8). For fixed τ ∈ H, the scalar Green function G(z|τ ) on the torus
is smooth throughout the torus except for the logarithmic singularity at z = 0 given by
G(z|τ ) ≈ − ln |z|2 . As a result, for fixed τ , the integrals which define B (1) (s, t, u|τ ) as
a function of s, t, u are absolutely convergent for Re (s), Re (t), Re (u) < 1, and may be
expanded in a Taylor series in powers of s, t, u with radius of convergence |s|, |t|, |u| < 1.
Since the function B (1) (s, t, u|τ ) is symmetric in s, t, u, we may organize this expansion in
powers of the symmetric polynomials σ2 and σ3 in s, t, u as we did for the tree-level amplitude,
∞
X (1) σ2p σ3q
B (1) (s, t, u|τ ) = B(p,q) (τ ) (12.8.1)
p,q=0
p! q!

(1)
Since B (1) (s, t, u|τ ) is a modular function in τ each coefficient B(p,q) (τ ) is a modular function
of τ . The coefficients may be computed by expanding the exponential in the integrand of
(12.5.8) in powers of its argument to a given order w, which is the overall degree in the
variables s, t, u, given by w = 2p + 3q, so that we have,
4 Z
!w
p q 2
X (1) σ σ 1 Y d zi
X
B(p,q) (τ ) 2 3 = sij G(zi − zj |τ ) (12.8.2)
p,q≥0
p! q! w! i=1 Σ Im τ i<j
2p+3q=w

(1)
The coefficients B(p,q) (τ ) for different values of p, q satisfying w = 2p + 3q may then be sorted
out by further expanding the w-th power on the right side. The expansion may be presented
graphically, and the modular function corresponding to a given graph is precisely one of the
modular graph functions discussed in section 9. Each graph has four vertices, namely the
points zi for i = 1, 2, 3, 4, and w edges. An edge may connect any pair of distinct vertices,
but is not allowed to begin and end on the same vertex.

201
 The expansion is simplified by the following observations. Any modular graph function
corresponding to

• a graph containing a vertex on which only a single edge begins or ends vanishes;
• a graph which becomes disconnected upon removing one edge vanishes;
• a graph which becomes disconnected upon removing a single vertex factorizes into the
modular graph functions of the resulting connected components.

The graphs corresponding to the modular graph functions which contribute to the modular
functions B(p,q) (τ ) up to weight w = 5 are presented in Figure 13. Instead of organizing the
graphs by weight w, they may alternatively be organized by loop order, which is indicated
in Figure 13 by the colored rectangular boxes.
The weight w of the graphs contributing to B(p,q) (τ ) for w = 2p + 3q may now be seen
to coincide with the definition of transcendental weight given in subsection 12.7.2, provided
we continue to assign weight −1 to s, t, u and weight 0 to τ .
By expanding (12.8.2) and using identities such as the one in (9.3.9), one finds,
(1)
B(0,0) = 1
(1)
B(1,0) = E2
(1) 5
B(0,1) = E
3 3
+ 31 ζ(3)
(1)
B(2,0) = 2C2,1,1 + E22 − E4
(1) 7
B(1,1) = C
3 3,1,1
+ 35 E2 E3 + 31 ζ(3)E3 − 34
E
15 5
+ 15 ζ(5) (12.8.3)

up to weight w = 5 included. The graphs that contribute are shown in Figure 13.

12.9 Genus two in terms of modular graph functions

The genus-two amplitude A(2) is obtained in (12.5.7) by integrating B (2) (s, t, u|Ω) over the
genus-two moduli space M2 . The partial amplitude B (2) (s, t, u|Ω) in turn is given by the
integration over four copies on the genus-two surface Σ with moduli Ω in (12.5.8). For fixed
Ω ∈ H2 , the scalar Arakelov Green function G(z, w|Ω) is smooth throughout Σ except for the
logarithmic singularity at z = w given by G(z, w|Ω) ≈ − ln |z − w|2 . As a result, for fixed Ω,
the integrals that define B (2) (s, t, u|Ω) as a function of s, t, u are absolutely convergent for
Re (s), Re (t), Re (u) < 1 and may be expanded in a Taylor series in powers of s, t, u with
radius of convergence |s|, |t|, |u| < 1. This situation is completely analogous to the genus-one
case. The presence of the additional measure factor |Y|2 is inconsequential for these results
to hold as Y is holomorphic on Σ.

202
 (1)
B(1,0) D4 R4 • •

•
(1)
B(0,1) D6 R4 • •
• •

• • •
(1)
B(2,0) D8 R4
• • • •
• •

• • •
(1)
B(1,1) D10 R4 • • • •
• •

• • •

• •
• •

one-loop two-loops three-loops

Figure 13: Modular graph functions contributing up to the genus-one four-graviton amplitude,
up to weight w = 5 included.

Since the function B (2) (s, t, u|Ω) is symmetric in s, t, u, we may organize its Taylor expan-
sion in powers of s, t, u via the symmetric polynomials σ2 and σ3 as we did for the tree-level
amplitude and the genus-one partial amplitude,

∞
(2)
X (2) σ2p σ3q
B (s, t, u|Ω) = B(p,q) (Ω) (12.9.1)
p,q=0
p! q!

Since B (2) (s, t, u|Ω) is a modular function of Ω, namely it is invariant under the genus-two
(2)
modular group Sp(4, Z), each coefficient B(p,q) (Ω) is itself a modular function of Ω. The
coefficients may be computed by expanding the exponential in the integrand of (12.5.8) in
powers of its argument to a given order w, which is related to the overall degree in the

203
variables s, t, u, given by w + 2 = 2p + 3q, so that we have,
!w
σ2p σ3q |Y|2
Z
X (2) 1 X
B(p,q) (Ω) = sij G(zi , zj |Ω) (12.9.2)
p,q≥0
p! q! w! Σ4 (det Im Ω)2 i<j
2p+3q=w+2

Note that the shift to w + 2 included in the range of summation on the left side is due to
the fact that Y is linear in the variables s, t, u.

12.9.1 Contributions to low weight

Linearity of Y in s, t, u immediately implies that the genus-two contribution to the R4
(2)
interaction without derivatives must vanish B(0,0) = 0. As was already mentioned earlier,
the D2 R4 contribution vanishes in view of the momentum conservation identity s+t+u = 0.
The leading low energy contribution is the D4 R4 interaction obtained from the zero-th
order expansion term of the exponential in (12.8.2). The remaining integrations required
(2)
to evaluate B(1,0) may be carried out using the Riemann bilinear relations, reviewed in
appendix B, and yield an Ω-independent contribution.
(2)
The coefficient B(0,1) has non-trivial dependence on Ω, denoted by 64 ϕ(Ω). Assembling
these results to order D6 R4 we have,

B (2) (s, t, u|Ω) = 32(s2 + t2 + u2 ) + 192 stu ϕ(Ω) + O(s4ij ) (12.9.3)

The genus-two modular function ϕ(Ω) is given in terms of the Arakelov Green function by,
Z
1  
ϕ(Ω) = ωI (x)ωJ (y) ω̄ I (x)ω̄ J (y) − 2ω̄ J (x)ω̄ I (y) G(x, y|Ω) (12.9.4)
8 Σ2

where ω̄ I = (Y −1 )IJ ωJ . This object was identified with the spectral invariant introduced
by Kawazumi and Zhang. Higher order contributions to B (2) (s, t, u|Ω) may similarly be
expressed in terms of integrals of products of Arakelov Green functions, and may be viewed
as higher genus generalizations of the genus-one modular graph functions investigated in
section 9. Their general structure has, however, been less well-studied and is much less
well-understood than that of their genus-one counterparts.

12.10 Integration over the genus-one moduli space

When considering one-loop or two-loop amplitudes we have, so far, focused attention on
the structure of the partial amplitudes B (1) (s, t, u|τ ) and B (2) (s, t, u|Ω), for which we have
obtained a systematic low energy expansion in terms of modular graph functions, both at
genus one and genus two. For fixed τ and Ω, this expansion is absolutely convergent for

204
Re (s), Re (t), Re (u) < 1, and admits an analytic continuation in s, t, u whose sole singulari-
ties are simple poles at positive integers in s, t, u, corresponding to the exchanges of massive
string states, as may be seen from the operator product expansion.
Obtaining the corresponding physical string amplitudes A(1) (sij ) and A(2) (sij ) further
requires the integrations over the moduli spaces of genus-one and genus-two Riemann surfaces
shown in (12.5.7). These integrations are absolutely convergent only when s, t, u are purely
imaginary, in which case B (1) (s, t, u|τ ) and B (2) (s, t, u|Ω) may be uniformly bounded on their
respective moduli spaces. For the case of genus one, the lack of convergence originates from
the integration in the region near the cusp. This may be seen by expressing the genus-
one Green function G(z|τ ) in the co-moving coordinates z = x + yτ for x, y ∈ R/Z first
introduced in section 5.5.1 and given by (5.5.14),
 ϑ1 (x + yτ |τ ) 2
 
G(x + yτ |τ ) = − ln   + 2πτ2 y 2 (12.10.1)
η(τ ) 

For τ2 → ∞ and fixed x and |y| < 21 , the Green function behaves as follows,

G(x + yτ |τ ) ≈ 2πτ2 (y 2 − |y|) (12.10.2)

which diverges as τ2 → ∞.
The analytic continuation in s, t, u of the integrals over moduli now produces double and
additional simple poles at all positive integers, as well as branch cuts in s, t, u that originate
at every non-negative integer, including zero. These singularities are fully expected on
physical grounds. The poles produce mass renormalization and non-zero decay widths of
massive string states, while the branch cuts correspond to the decay of a single string into
a two-string state. The presence of the branch cuts implies that the string amplitudes A do
not admit a convergent low energy Taylor expansion. We now review the derivation of the
branch cuts that arise in the low energy expansion for the case of one-loop amplitudes.

12.10.1 Partitioning the genus-one moduli space

To proceed with the analysis of the branch cuts produced in the low energy limit, we partition
the moduli space M1 of genus-one Riemann surfaces into a region near the cusp MR and
its complement ML , as illustrated in the Figure 14, and given as follows for L > 1,

MR = M1 ∩ {τ2 > L} M L ∩ MR = ∅
ML = M1 ∩ {τ2 ≤ L} M L ∪ M R = M1 (12.10.3)

We shall think of L as a free parameter, but in practice take L  1 so that MR is a

small neighborhood of the cusp. The amplitude A(1) is a sum of the corresponding integrals,
(1) (1)
A(1) (sij ) = AL (sij ) + AR (sij ) (12.10.4)

205
 i∞

MR = M1 ∩ {τ2 > L}
iL
H1

ML = M1 ∩ {τ2 < L}

i
•
• •

R
−1 − 12 0 1
2 1

Figure 14: Partition of the genus-one moduli space M1 into MR and ML .

where
|dτ |2
Z
(1) π
AL,R (sij ) = B (1) (s, t, u|τ ) (12.10.5)
16 ML,R (Im τ )2

The region ML is compact and the integrand B (1) (s, t, u|τ ) admits a uniform low energy
expansion in the region of convergence |s|, |t|, |u| < 1. Therefore, the resulting contribution
(1)
AL (sij ) will be analytic in s, t, u but dependent on L.
The region MR is not compact and B (1) (s, t, u|τ ) cannot be handled with a uniformly
convergent low energy expansion. Instead we need a treatment of this contribution that is
(1)
exact up to contributions exponentially suppressed in L. The resulting AR (sij ) will exhibit
(1)
branch cuts in s, t, u and its L-dependence will cancel the L-dependence of AL (sij ).

12.10.2 Integrals involving Eisenstein series and modular graph functions

(1)
To carry out the integrations over the fundamental domain of the partial amplitudes B(p,q) (τ ),
one invariably encounters integrals of Eisenstein series, products of Eisenstein series, and
other modular graph functions, over the truncated fundamental domain ML . These integrals
are of interest in their own right and we dedicate this subsection to deriving some general

206
results and applying these results to the simplest cases, and then close by quoting the integral
of the product of three Eisenstein series from Zagier’s work. We begin with the following
straightforward application of Stokes’s theorem to an arbitrary modular graph function C of
weight (0, 0), or to an arbitrary modular graph form C + of weight (0, 2),
1
d2 τ
Z Z 
∆C = dτ1 ∂τ2 C(τ1 , τ2 )

ML τ22 0 τ2 =L
1
d2 τ
Z Z 
∇C + = dτ1 C + (τ1 , τ2 ) (12.10.6)

ML τ22 0 τ2 =L

where the result on the right side may be read off from the constant Fourier mode part of
C and C + . The result may be readily used to evaluate the integral of a non-holomorphic
Eisenstein series or products thereof. It turns out that the simplest expressions are obtained
in terms of the normalized Riemann zeta function ζ ∗ (s) and the normalized non-holomorphic
Eisenstein series Es∗ (τ ), defined by,

Γ(s/2)ζ(s)
ζ ∗ (s) = s Es∗ (τ ) = 21 Γ(s)Es (τ ) (12.10.7)
π2
The normalization guarantees the functional equations ζ ∗ (1 − s) = ζ ∗ (s) and E1−s
∗
(τ ) =
∗
Es (τ ), in addition to the following simple Fourier expansion,

Es∗ (τ ) = ζ ∗ (2s) τ2s + ζ ∗ (2s − 1) τ21−s + O(e−2πτ2 ) (12.10.8)

In terms of these functions, we have the following result.

Lemma 12.1 The following integrals involving Eisenstein series are given, up to exponen-
tially decaying terms O(e−2πτ2 ) at the cusp, by,

d2 τ ∗ Lx−1
Z X
E = ζ ∗ (2x) (12.10.9)
ML τ22 s x=s,1−s
x−1
d2 τ ∗ ∗ Lx+y−1
Z X X
E E = ζ ∗ (2x)ζ ∗ (2y)
ML τ22 s t x=s,1−s y=t,1−t
x+y−1
Z 2
dτ ∗ ∗ ∗
2
Es Et Eu = ζ ∗ (w − 1)ζ ∗ (w − 2s)ζ ∗ (w − 2t)ζ ∗ (w − 2u)
ML τ2
X X X Lx+y+z−1
+ ζ ∗ (2x)ζ ∗ (2y)ζ ∗ (2z)
x=s,1−s y=t,1−t z=u,1−u
x+y+z−1

where w = s + t + u in the last line.

207
 The first line in Lemma 12.1 may be proven by substituting Es∗ = ∆Es∗ /s(s − 1) in the
integrand and then using the first line on (12.10.6). The second line in Lemma 12.1 may be
proven by expressing the combination ∆Es∗ Et∗ − Es∗ ∆Et∗ in two different ways,
 
∆Es Et − Es ∆Et = s(s − 1) − t(t − 1) Es∗ Et∗
∗ ∗ ∗ ∗
(12.10.10)
 ∗¯ ∗ ¯ ∗
Es ∇Et − Et∗ ∇E ∗ ∗ ∗ ∗
  
= ∇ s
+ ¯ Es ∇Et − Et ∇Es
∇
2τ22 2τ22
and then using the expression on the right side of the first line to eliminate Es∗ Et∗ from
the integrand of 12.1 in favor of the total derivative terms in the second line. Finally, to
integrate the total derivative terms, we use the second line in (12.10.6) and its complex
conjugate relation along with the asymptotics (12.10.8) of the Eisenstein series. The third
line in Lemma 12.1 was proven by Zagier using different methods, as will be indicated in the
bibliographical notes.

(1)
12.10.3 The non-analytic contribution AR
To evaluate B (1) (s, t, u|τ ) for τ ∈ MR , we cannot use (12.8.1) since this expansion is not
uniformly convergent throughout MR . Instead, our starting point is the original expression
in (12.5.7), which we shall evaluate exactly up to terms that are exponentially suppressed
in L. To do so, we use co-moving coordinates, zi = xi +yi τ with i = 1, 2, 3, 4 and partition the
integration region Σ4 into the six possible orderings of y1 , y2 , y3 , y4 up to cyclic permutations.
(1)
Doing so allows us to decompose AR (sij ) into a sum of six more elementary contributions,
which are pairwise equal to one another,
(1)
AR (sij ) = 2A∗ (L; s, t) + 2A∗ (L; t, u) + 2A∗ (L; u, s) s + t + u = 0 (12.10.11)

The function A∗ (L; s, t) may be computed exactly, up to exponentially suppressed terms,

∂ 
A∗ (L; s, t) = C(s, t; 0) ln(−4πLs) + C(s, t; ε) + O(e−2πL ) (12.10.12)

∂ε ε=0

where C is conveniently expressed as the sum of two integral representations,

Z 1 Z 1
8πs4
C(s, t; ε) = − dx dy x5+2ε y 2+ε (1 − y)2+ε W (s, −sxy + t(1 − x))
Γ(3 + ε) 0 0
∞ 1  2
xk+3+ε (1 − x)k+3+ε ∂ k
X Z 
7 k k
−4πs s t dx W (s, µ − sx) (12.10.13)

2Γ(k + 4 + ε) k! ∂µ k µ=0
k=0 0

The function W is closely related to the Virasoro-Shapiro amplitude,

 
1 Γ(1 − s)Γ(1 − t)Γ(1 − u)
W (s, t) = −1 u = −s − t (12.10.14)
stu Γ(1 + s)Γ(1 + t)Γ(1 + u)

208
Although the above expression may look daunting at first, it analytical structure is manifest.
The function W (s, t) is meromorphic throughout s, t ∈ C, holomorphic in discs of unit radius
near the origin, and has only simple poles at positive integer values of s, t, u, and thus admits
a Taylor expansion in s, t. Substituting this Taylor expansion into the integrals defining C,
we see that the integrals preserve the holomorphicity in s, t, u near the origin, so that C itself
has a convergent Taylor expansion near the origin, which starts at order s4 . To this lowest
order we have,

Γ(3 + ε)
C(s, t; ε) = −16πζ(3)s4 + O(s5 , s4 t) (12.10.15)
Γ(7 + 2ε)

The coefficients of the expansion to all orders inherit contributions linear and bilinear in odd
zeta-values from the expansion of the Virasoro-Shapiro amplitude, with rational coefficients
in the coefficient
 C(s, t; 0) of the ln(−4πLs) term, and additional harmonic sums in the
∂ε C(s, t; ε) ε=0 analytic contribution to A∗ (L; s, t). The discontinuity of ln(−4πLs) near the

origin reproduces the square of the tree-level amplitude by unitarity.

(1)
12.10.4 The analytic contribution AL
(1)
The contribution AL (sij ) is analytic in s, t, u near the origin and we may use the expansion
(1)
of B (1) (s, t, u|τ ) in terms of modular graph functions of (12.8.1) to evaluate AL (sij ) order
by order in powers of s, t, u,
∞
(1)
X (1) σ2p σ3q
AL (sij ) = A(p,q) (12.10.16)
p,q=0
p! q!

(1)
Since the non-analytic part AR (sij ) starts contributing to A(1) (sij ) only at order s4 , all
(1)
its lower orders are given entirely by AL (sij ). The corresponding amplitudes are readily
evaluated, and we find,

(1) π (1) π (1) (1) 29π

A(0,0) = A(0,1) = ζ(3) A(1,0) = 0 A(1,1) = ζ(5) (12.10.17)
3 9 540
The remaining coefficients at weight four are given as follows,

ζ 0 (4) ζ 0 (3) 1
 
(1) 4πζ(3)
A(2,0) = ln(2L) + − − (12.10.18)
45 ζ(4) ζ(3) 4

12.10.5 Assembling analytic and non-analytic contributions

(1) (1)
Assembling the contributions to AL in (12.10.18) and to AR in (12.10.15) gives A(1) (sij ),
the total amplitude, to order O(s4ij ) included. We see explicitly that all L-dependence

209
cancels as required for the consistency of the calculation. It is instructive to rearrange the
total amplitude as the sum of “analytic” and “non-analytic” pieces,
 
(1) (1)
A(1) (sij ) = 2π AL (L; sij ) + AR (L; sij ) = Aan (sij ) + Anon−an (sij ) (12.10.19)

The reason for the quotation marks on analytic and non-analytic is that the non-analytic
piece actually contains also analytic contributions, so that the nomenclature is natural and
suggestive but not entirely precise. The “analytic” piece Aan (sij ) is given by,

2π 2
 
ζ(3)σ3 5
Aan (sij ) = 1+ + O(sij ) (12.10.20)
3 3
and the “non-analytic” piece Anon−an (sij ) has the form,

2π 2
Asugra + A4 + O(s5ij )


Anon−an (sij ) = (12.10.21)
3
The lowest-order term Asugra is a regularized version of the ten-dimensional one-loop super-
gravity amplitude. The contribution A4 is given by,
4s4 ζ 0 (4) ζ 0 (3)
 
63
A4 = ζ(3) − ln(−2πs) + − −γ+ + cycl (s, t, u) (12.10.22)
15 ζ(4) ζ(3) 20
where γ is the Euler constant. The discontinuity of Anon−an (sij ), namely the coefficient of
the ln(−2πs) terms in (12.10.22), reproduce those obtained in [91].

12.11 Integration over the genus-two moduli space

(2)
The function B(1,0) (Ω) needed to obtain the coefficient of the D4 R4 interaction was found to
be independent of moduli Ω in subsection 12.9.1. Thus, its integral over moduli space M2
is given by the volume of M2 with the Siegel metric on the Siegel upper half space,
X
−1 −1
ds2 = YIJ YKL dΩ̄JK dΩIL (12.11.1)
I,J,K,L

of constant negative curvature. The volume form dµ2 and the value of the volume, computed
by Siegel, are given by,
8π 3 |dΩ11 ∧ dΩ12 ∧ dΩ22 |2
Z
dµ2 = dµ2 = (12.11.2)
M2 270 (det Im Ω)3
As a result, the low energy effective interaction to this order is given by,

(2) 4π 4 2
A(1,0) = (s + t2 + u2 )R4 (12.11.3)
270
210
 (2)
The function B(0,1) (Ω) is proportional to the Kawazumi-Zhang invariant ϕ. Its integral
(2)
over M2 is convergent and yields the contribution A(0,1) to the low energy effective action.
Evaluation of the integral is made possible by the fact that ϕ satisfies an inhomogeneous
Laplace eigenvalue equation,

(∆ − 5)ϕ = −2πδSN (12.11.4)

where ∆ is the Laplace-Beltrami operator on the Siegel upper half space for the Siegel metric,
X 1 ∂
∆= 4 YIK YJL ∂¯IJ ∂KL ∂IJ = (1 + δIJ ) (12.11.5)
I,J,K,L
2 ∂ΩIJ

and δSN is the Dirac δ-function supported on the separating degeneration node. The coeffi-
(2)
cient B(0,1) is then given by the integral of the Kawazumi-Zhang invariant over M2 ,

2π 3
Z Z
(2) π  
B(0,1) =π dµ2 ϕ = dµ2 ∆ϕ + 2πδSN = (12.11.6)
M2 5 M2 45

The integral may be evaluated using Stokes’s theorem and the asymptotic behavior for the
Kawazumi-Zhang invariant, and we find,

(2) 2π 3 3
A(0,1) = (s + t3 + u3 )R4 (12.11.7)
45
For possible verifications to genus-three order, we refer to the references of this section.

• Bibliographical notes
Classic books by Green, Schwarz, and Witten [165, 166] and by Polchinski [167, 168] present
comprehensive perspectives on string theory. A pedagogical introduction, aimed at under-
graduate and graduate students, is given in the book by Zwiebach [169]. Other accounts
may be found in the books by Johnson [170], by Becker, Becker, and Schwarz [171], by
Blumenhagen, Lüst, and Theissen [172], and by Kiritsis [173].
The early development of string theory was dominated by the study of string amplitudes.
The Ramond and Neveu-Schwarz sectors of the RNS formulation were introduced in [174]
and [175], respectively. The different spin structure sectors and the Gliozzi-Scherk-Olive
(GSO) projection onto supersymmetric string spectra were introduced in [176]. The Polyakov
formulation of string theory, along with many other key problems in modern theoretical
physics, is presented in the book by Polyakov [177]. Friedan’s renormalization of the non-
linear σ-model and the relation between Einstein’s equations and conformal invariance may
be found in [178]. Generalizations with worldsheet fermions are given in [179], and including

211
torsion in [180] and the dilaton in [181]. The α0 expansion of the effective action including
the metric, anti-symmetric tensor field, and dilaton is presented in detail in the book of
Princeton Institute for Advanced Study lecture notes [182].
The standard treatment of the decoupling of negative norm states for both the bosonic
and superstrings is clearly reviewed in [165]. The modern BRST-based approach to the
covariant quantization of superstrings in the RNS formulation was developed in [49], and
reviewed in [165, 166] and [167, 168]. The structure of gauge and gravitational anomalies,
relevant to supergravity and string theories in various dimensions, was obtained in [183].
The cancellation of anomalies in the Type I theory with gauge group SO(32) was famously
discovered in [184]. Useful lecture notes on anomalies may be found in [185].
Reviews of superstring perturbation theory may be found in [186, 187] for the RNS
formulation and in [188] for the pure spinor formulation. Lectures accessible to a more
mathematically oriented readership are in [182], while an elaboration of the mathematical
underpinning of the RNS formulation is provided in [189, 190]. A broad overview of the
current status of superstring perturbation theory, and open problems, is given in [191],
which contains an extensive bibliography.
The tree-level and one-loop four graviton amplitudes were derived in the original papers
in which the Type II [192] and Heterotic strings [193, 194] were discovered. The existence of
the analytic continuation of the one-loop four graviton amplitude was shown in [195, 196].
The fact that the low energy expansion of string theory leads to N = 4 super-Yang-Mills
theory and N = 8 supergravity in four space-time dimensions was discovered in [197]. The
systematic analysis of the low energy effective interactions induced by the tree-level four-
graviton amplitude was initiated in [198, 199], while the effects at genus-one were obtained
in [200, 201, 91]. The low energy expansion, up to order D8 R4 , including the logarithmic
branch cuts and the transcendentality properties of the amplitude discussed in section 12.7.2
were obtained in [202].
The original Rankin-Selberg method for integrating cusp forms over the fundamental
domain for SL(2, Z) was developed in [203, 204], and was extended to automorphic functions
of non-rapid decay in [205]. The proof of the last integral identity in Lemma 12.1 is in [205],
while methods to carry out integrals of modular graph functions may be found in [206] and
references therein.
Considerable progress has been made recently towards evaluating tree-level and genus-
one string amplitudes with arbitrary numbers of external states in [207, 208, 209, 210, 211]
and references therein. The role of motivic and multiple zeta-values played in the form of
the amplitudes was examined in [212, 213, 214]. Genus-one amplitudes have also recently
been reproduced from N = 4 supersymmetric Yang–Mills theory by considering a flat-space
limit of AdS5 × S 5 [215].
For genus-two amplitudes, the measure on supermoduli space was derived in the RNS

212
formulation in [216, 217, 218] and reproduced using algebraic-geometric methods in [219].
The full amplitude for four gravitons was obtained in [220] in the RNS formulation, and
extended to include external fermion states in [221, 222] using the pure spinor formulation.
The overall normalization of the amplitude was obtained in [223, 224].
Recently, the genus-two amplitudes with five massless states were constructed in [225, 226]
using an amalgam of methods from the RNS and pure spinor formulations and. For external
NS boson states and even spin structure, these results were confirmed from first principles
in the RNS formulation in [227].
The generalization of modular graph functions to genus two and higher genus was devel-
oped in [120] while the relation between the coefficient of the D6 R4 term and the invariant of
Kawazumi [228] and Zhang [229] was identified in [230] and used to derive the corresponding
differential equation in [231]. Degenerations of genus-two modular forms naturally produce
non-holomorphic versions of Jacobi forms, systematic treatment of which may be found in
the book by Eichler and Zagier [232]. A ϑ-lift representation of the genus-two Kawazumi-
Zhang invariant was obtained in [233]. For the derivation of the results in section 12.11 we
refer to [231].
For amplitudes at genus three and beyond, no first principle derivations are available
at the time of this writing. A proposal for the measure on supermoduli space in the RNS
formulation was advanced in [234], building on some earlier unsuccessful attempts in [235].
The proposal of [234] was critiqued, however, in the appendix of [236]. The leading low energy
effective interaction in Type II string theory was obtained using the pure spinor formulation
in [237]. The corresponding expression for the amplitude, however, diverges beyond the
leading low energy limit so that no independent checks on the unitarity of the amplitude
is available. A proposal for the full genus-three four graviton amplitude has been advanced
in [238], but details of the derivation are not yet available. Proposals for the measure on
supermoduli space in the RNS formulation at genus four may be found in [239, 240].

213
13 Toroidal compactification
In this section, we discuss the modular properties of quantum field theory of scalar fields
that take values in a d-dimensional torus with a flat metric and constant anti-symmetric
tensor. The problem is of great interest in quantum field theory and string theory in view of
the fact that such toroidal compactifications admit solutions using free field theory methods
on the worldsheet, preserve Poincaré supersymmetries, and may be used to relate different
perturbative string theories via T-duality. Toroidal compactifications produce large duality
groups that generalize the modular group SL(2, Z) considered thus far.

13.1 Conformal field theory on flat manifolds

In section 5 we considered the functional integral for a single free scalar field ϕ that takes
values in R. The situation is readily generalized to the case of d free scalar fields taking
values in a flat Euclidean Rd . As discussed in section 12, the generalization to d scalar
fields taking values in (the local coordinate systems of) an arbitrary space-time manifold in
general leads to fully interacting quantum field theories on the worldsheet that cannot be
solved exactly.
There is one class of theories, however, where an exact quantization may be obtained,
namely when all the fields G, B, Φ are constant, and the manifold M is flat. Although flat,
M may have non-trivial topology as in the case of a d-dimensional torus M = Td , which is
the case we shall consider here. Constant Φ gives rise to a term proportional to the Euler
number of Σ and weighs the functional integral by a power of the string coupling which is
independent on the quantum field X. Finally, the term involving Bµν is a topological term
when B is constant. It does not contribute when the topology of M is trivial, but when M
is a torus, it does have very interesting physical effects which we shall now examine. For
constant G, B, Φ, the X-dependent part of the action thus reduces to,
Z
1  
I[X, g] = 0
dµg g Gµν − i ε Bµν ∂m X µ ∂n X ν
mn mn
(13.1.1)
4πα Σ

Henceforth, we shall restrict to the case where the worldsheet surface is a torus of modulus τ ,
so that Σ = C/Λ with Λ = Z + τ Z. We shall choose a system of complex coordinates z, z̄
on Σ in which the metric g takes the form,

m n|dz|2
gmn dξ dξ = (13.1.2)
τ2

or in components we have gzz = gz̄z̄ = 0 and gzz̄ = 1/(2τ2 ). Furthermore, the space-time
torus Td is given by Rd /L for some d-dimensional lattice L.

214
13.2 Lattices and tori of dimension d
A lattice L of rank d, also referred to as a lattice of dimension d, may be represented in
terms of a basis of d vectors v1 , · · · , vd in Rd ,

n d
X o
L= `= ni vi ni ∈ Z (13.2.1)
i=1

It will be useful to view the assignment of basis vectors as obtained by a GL(d, R) trans-
formation applied to a standard basis of d vectors, such as for example v10 = (1, 0, · · · , 0),
v20 = (0, 1, 0, · · · , 0), · · · , vn0 = (0, · · · , 0, 1). Decomposing GL(d, R) = SL(d, R) × R+ , the
factor R+ corresponds to overall rescaling of the lattice. An arbitrary set of d vectors vi0 ∈ L
for i = 1, · · · , d may be decomposed in the basis vi ,
d
X
vi0 = Mij vj Mij ∈ Z (13.2.2)
j=1

and generates a sub-lattice L0 ⊂ L. Lattices may be viewed as Abelian groups so that L0 is

a subgroup of L. The index of L0 as a subgroup of L is given as follows,

[L : L0 ] = |det M | (13.2.3)

Equivalently, the unit lattice cell of L0 contains |det M | copies of the unit lattice cell of L.
In the special case where det M = 1, we have L0 = L and therefore M is an automorphism
of L. The group of all automorphisms of L is SL(d, Z), thereby generalizing to d dimensions
the result familiar from the two-dimensional torus. When det M = −1 the lattice L0 is the
mirror image of L, reversing the orientation of the lattice.
Introducing the equivalence relation between two lattices that are related to one another
by an overall rescaling in R+ , an overall rotation in SO(d), and a lattice automorphism in
SL(d, Z), we may identify the space of all inequivalent lattices as follows,

space of inequivalent lattices = SL(d, Z)\SL(d, R)/SO(d, R) (13.2.4)

In the case of a two-dimensional lattice L with d = 2, we may identify SL(2, R)/SO(2) with
the Poincaré upper half-plane H and the entire coset SL(2, Z)\SL(2, R)/SO(2, R) with the
moduli space of toroidal Riemann surfaces. The dynamics of point particles propagating
on the torus Td = Rd /L is sensitive to the lattice geometry of this quotient. We shall
see below that the dynamics of strings propagating on the torus Rd /L is sensitive to a
different equivalence class of lattices, which is “smaller” than SL(d, Z)\SL(d, R) thanks to
the quintessentially string-theoretic phenomenon of T -duality.

215
13.3 Fields taking values in a torus
We begin by reviewing the construction of the space of d scalar fields X(z) that take values
in the torus Rd /L and are functions on a worldsheet Σ = C/Λ with the topology of a
two-dimensional torus, whose complex structure is given by its modulus τ ∈ H so that
Λ = Z ⊕ τ Z. The space of maps,

X : Σ = C/Λ → Td = Rd /L (13.3.1)

is disconnected and may be decomposed into the union of an infinite number of connected
components which are labelled by two copies of the lattice L. In a sector labelled by a pair
(`A , `B ) ∈ L × L, the field X µ satisfies the following monodromy conditions,

X(z + 1) = X(z) + `A
X(z + τ ) = X(z) + `B `A , `B ∈ L (13.3.2)

The field X may be decomposed into a purely periodic part Y , which satisfies,

Y (z + 1) = Y (z + τ ) = Y (z) (13.3.3)

and a special solution to the monodromy conditions. The latter is arbitrary but it will
be convenient to choose it linear in z and z̄. Since the field X and the vectors `A , `B are
real-valued, we have the following decomposition,
`B − τ̄ `A `B − τ `A
X`A ,`B (z) = Y (z) + z − z̄ (13.3.4)
τ − τ̄ τ − τ̄
Such a configuration is sometimes referred to as a worldsheet instanton. The lattice elements
`A , `B are related to the momentum and the winding modes of strings on Td , as we shall
clarify below. The integration over all X : Σ → Td decomposes into a sum over sectors
labelled by L × L. The functional integral takes the following form,
Z X Z
−I[X,g]
DXe = Vol(Td ) D0 Y e−I[X`A ,`B ,g] (13.3.5)
maps(Σ→Td ) `A ,`B ∈L maps(Σ→Rd )

For given worldsheet metric g, the measures DX and D0 Y are evaluated using the norm
of (12.2.2). The factor Vol(Td ) arises from the normalizable zero mode of the field Y , and
the integration D0 Y is over all non-zero modes, i.e. all functions Y orthogonal to constants.
The contribution to the action of Y decouples from the contribution of the momentum and
winding modes, which may be evaluated explicitly,
1
I[X`A ,`B , g] = I[Y, g] + 0
(`µB − τ̄ `µA )(`νB − τ `νA )(Gµν + Bµν ) (13.3.6)
4πα τ2

216
The integral over Y factorizes, the dependence on Bµν cancels out in this integral and reduces
to the functional integrals computed earlier in section 5. The remaining factor is given by
the following lattice sum,
 
X 1 µ µ ν ν
Z(L, τ ) = Vol(Td ) exp − 0τ
(`B − τ̄ `A )(`B − τ `A )(Gµν + Bµν ) (13.3.7)
` ,` ∈L
4πα 2
A B

The dependence on Bµν reduces to −iBµν `µA `νB /(2πα0 ) and is independent of τ , as is expected
for a topological term. Before investigating the properties for a general torus Td , we first
look at the much simpler case of d = 1.

13.4 T-duality on a circle

In this subsection, we study the simplest case of toroidal compactification when d = 1 and
exhibit the phenomenon of T -duality for a circle. We take G = 1, and introduce the lattice
L associated with a circle of radius R, namely L = 2πR Z. We label the monodromy vectors
`A and `B , which are one-dimensional in this case, as follows,

`A = 2πR n
`B = 2πR m m, n ∈ Z (13.4.1)

The sum over momentum and winding modes then simplifies to,
r
πR2
 
R2 X 2
Z(R, τ ) = exp − 0 |m − τ n| (13.4.2)
α0 m,n∈Z α τ2
p
The prefactor in R2 /α0 arises from the zero mode of the field Y which, for a circle of radius
R, gives a finite contribution. The above sum Z(R, τ ) is a rather familiar looking object. To
exhibit T -duality, we perform a Poisson resummation on both m, n. To do so we compute
the Fourier transform in both variables,
πR2 α0 πα0
Z    
−2πi(mx+ny) 2 2
dm dn e exp − 0 |m − τ n| = 2 exp − 2 |y + τ x| (13.4.3)
R2 α τ2 R R τ2
so that Poisson resummation gives us,
r
α0 X πα0
 
2
Z(R, τ ) = exp − 2 |m + τ n| (13.4.4)
R2 m,n∈Z R τ2

Noticing that Poisson resummation has inverted the dependence on R2 /α0 , we have,
 0 
α
Z(R, τ ) = Z ,τ (13.4.5)
R

217
Thus, a free bosonic string on a circle of radius R has exactly the same partition function
as a free bosonic string on a circle of radius α0 /R. While we have proven this above for the
free string on a genus-one worldsheet only, the effect actually is valid on surfaces of arbitrary
genus, and thus holds for fully interacting string theory. The effect is referred to as T -duality.
We may organize the double sum in a slightly different way which makes the decompo-
sition into momentum and winding modes transparent. Performing a Poisson resummation
on (13.4.2) in m only, we obtain the representation,
1 2 2
X
Z(R, τ ) = (τ2 ) 2 e2πiτ pL e−2πiτ̄ pR (13.4.6)
m,n∈Z

where the left and right momenta are given in terms of the integers m, n by,
√ ! √
1 α0 R α0
pL = m+ √ n pL + p R = m
2 R α0 R
√ !
1 α0 R R
pR = m− √ n pL − pR = √ n (13.4.7)
2 R α0 α0
√
The lattice points pL + pR ∈ α0 /R Z are clearly
√ associated with the total momentum of the
string, while the lattice points pL − pR ∈ R/ α0 Z are associated with the winding modes.
The space of all possible compactifications on the circle is labeled by the radius R modulo
the
√ identification R ≡√ α0 /R, a space we may denote by R+ /Z2 . Clearly, the lattices L =
+
α0 /R Z and L = R/ α0 Z are dual to one another, a fact that is reflected in the relation
(pL + pR )(pL − pR ) = mn ∈ Z. Thus, the lattice sum may be written alternatively as,
1 2 2
X X
Z(R, τ ) = (τ2 ) 2 eπiτ (p+w) /2 e−πiτ̄ (p−w) /2 (13.4.8)
p∈L w∈L+
√
At the radius R = α0 the conformal field theory is self-dual √ and enjoys an enhanced
symmetry to the SU (2) Kac-Moody algebra. The radius R = 2α0 is also special in a
different sense: it is at this point that the c = 1 Virasoro algebra representation of the
conformal field theory becomes reducible to two c = 12 fermions.
T -duality has several important consequences for string theory, which include,

1. The fact that (bosonic) string theory on a circle of radius R is physically equivalent
to a string theory on a circle
√ of radius α0 /R means that we can never physically probe
distances smaller than α0 . Analogous arguments exist for the energy of scattering
processes, namely probing a string with energy E and with energy 1/(α0 E) are physi-
cally equivalent. It means that string theory has built in its dynamics a smallest string
√
length scale α0 .

218
 2. Actually, the above picture is modified in the case of superstring theories. In Type II
string theories, compactification of Type IIA on a circle of radius R is physically equiv-
alent to Type IIB theory on a circle of radius α0 /R. In Heterotic string theories, the
E8 × E8 theory is invariant under R ↔ α0 /R but the Spin(32)/Z2 theory is exchanged
with the theory of open and closed strings Type I. In both cases, T -duality provides a
mechanism for a partial unification of the five 10-dimensional superstring theories.
3. T -duality extends to string theories on curved manifolds under certain conditions. For
example, on Calabi-Yau spaces, it is promoted to mirror symmetry.

13.5 T-duality on a torus Td

For a general torus Td = Rd /L with d ≥ 2 new effects arise. We consider the lattice sum of
(13.3.7) and introduce a basis {v1µ , · · · , vdµ } with µ = 1, · · · , d for the lattice L. Generalizing
the conventions of (13.4.1) used for the circle, we have,
d
X
`µA = 2π ni viµ ni ∈ Z
i=1
Xd
`µB = 2π mi viµ mi ∈ Z (13.5.1)
i=1

What enters Z(L, τ ) is the combination,

1
(Gµν + Bµν )viµ vjν = Gij + Bij (13.5.2)
α0
where Gij and Bij are respectively symmetric and anti-symmetric under the interchange of
the indices i, j. Using matrix notation for m, n, G, B, we have,
 
d
X π †
Z(L, τ ) = Vol(T ) exp − (m + τ n) (G + B)(m + τ n) (13.5.3)
d
τ2
m,n∈Z

Since vi forms a basis of L, the volume of the torus is given by,

1
Vol(Td ) = (det G) 2 (13.5.4)
up to some factor of 2π which will be of no interest here.

13.5.1 A first look at T -duality

The argument of the exponential may be decomposed as follows,
π π
− (m + τ n)† (G + B)(m + τ n) = − (m + τ1 n)t G(m + τ1 n)
τ2 τ2
−πτ2 nt Gn − 2πimt Bn (13.5.5)

219
As a result, each term in the exponential is invariant under a shift in B by an arbitrary
anti-symmetric matrix with integer entries, or in components,

Bij → Bij + bij bij = −bji ∈ Z (13.5.6)

This invariance is an immediate consequence of the fact that, for constant B, the B-term
is topological and its values on the field configurations for the torus Td are quantized. The
corresponding symmetry is the analogue of the T -transformation τ1 → τ1 + 1 for SL(2, Z),
and will provide some of the generators of the full T -duality group for Td .
Next, we wish to obtain the analogue of the S-transformation. To do so, we Poisson
resum in both m, n. It is helpful to express the quadratic form combining m, n into a single
column matrix, so that,
 t  
π † m m
− (m + τ n) (G + B)(m + τ n) = −π M (13.5.7)
τ2 n n

where M is given by,

 
G τ1 G + iτ2 B
1 
M=  (13.5.8)
τ2
τ1 G − iτ2 B (τ12 + τ22 )G

The inverse matrix may be expressed as follows,

 t   
0 I G̃ τ1 G̃ + iτ2 B̃ 0 I
1
M−1 =      (13.5.9)
τ2
−I 0 τ1 G̃ − iτ2 B̃ (τ12 + τ22 )G̃ −I 0

where G̃ and B̃ are given by the S-transformation,

 −1
t −1
G̃ = G + B G B B̃ = G −1 B G̃ (13.5.10)

The action of S 2 is obtained by repeating the transformation,

 −1
t −1
Ĝ = G̃ + B̃ G̃ B̃ B̂ = G̃ −1 B̃ Ĝ (13.5.11)

and eliminating B̃ and G̃. The result is B̂ = B and Ĝ = G, so that the action of S 2 on G, B is
the identity.25 We shall show shortly that S and the integer translations of B generate the
duality group SO(d, d, Z).

25
Although in SL(2, Z) the matrix S squares to S 2 = −I, its action on τ reduces to the identity; the above
statement is the direct analogue thereof.

220
13.5.2 Holomorphic block decomposition
As in the case of the scalar field theory valued in the circle, it is instructive to perform a
Poisson resummation only on the integers m, and expose the holomorphic structure in τ . To
do so, we start from the decomposition of the quadratic form given earlier, and we need the
following Fourier transform,
Z  
d −2πixt m π t t
d xe exp − (x + τ1 n) G(x + τ1 n) − 2πix Bn
Rd τ2
d
(τ2 ) 2 n
2 2
o
= 1 exp 2πiτ pL − 2πiτ̄ pR (13.5.12)
(det G) 2
where we obtain after minimal simplifications,

4p2L = (m + Bn)t G −1 (m + Bn) + nt Gn + 2mt n

4p2R = (m + Bn)t G −1 (m + Bn) + nt Gn − 2mt n (13.5.13)

Note that while the integers n were defined with an upper index, as were the old m, by
contrast the Poisson resummation variable m have lower indices. So, it is instructive to
write out the above relations in terms of indices,
 0
  0

4p2L = mi + Bii0 ni (G −1 )ij mj + Bjj 0 nj + ni Gij nj + 2mi ni
 0
  0

4p2R = mi + Bii0 ni (G −1 )ij mj + Bjj 0 nj + ni Gij nj − 2mi ni

In evaluating the partition function, the factor Vol(T d ) cancels out, and we find,
d
X n o
Z(L, τ ) = (τ2 ) 2 exp 2πiτ p2L − 2πiτ̄ p2R (13.5.14)
m,n∈Zd

with p2L , p2R functions of m, n as defined above. The invariance under integer shifts of the
matrix B is now manifested by the fact that such a shift may be compensated in the sum
by a shift in the integers m.

13.5.3 T -duality in terms of the lattice

The description of the torus with both an arbitrary lattice L and an arbitrary metric G is
in fact redundant. By general linear transformation, we may always make the metric equal
to the identity. Geometrically, this is equivalent to expressing the metric Gij in terms of an
orthonormal frame eµ a and its inverse ea µ by

Gij = ei a ej b δab (13.5.15)

221
where i are “Einstein indices” and a are frame indices. The lattice L with identity metric is
then generated by the vector ea i , while the dual lattice L+ is generated by the dual vector
ei a . Defining also the B-field in frame indices,

Bab = Bij ea i eb j (13.5.16)

and defining the lattice and dual lattice vectors,

µa = mi ea i
ν a = ni ei a (13.5.17)

we find that p2L and p2R may be interpreted as square of vectors, respectively given by,
1 
pL = µ + Bν + ν µ ∈ L+
2
1 
pR = µ + Bν − ν ν∈L (13.5.18)
2
with,

pL · pL − pR · pR = µ · ν = mt n ∈ Z (13.5.19)

To parametrize the space of all lattices, we proceed as follows. Given a lattice, it may be
deformed continuously provided we maintain the condition pL · pL − pR · pR ∈ Z. This
condition is invariant under the group SO(d, d; R). But rotations under SO(d) on both pL
and pR are physically indistinguishable from the original lattice. Therefore, any lattice and
its dual may be parametrized by a point in the coset space,

SO(d, d; R)/(SO(d; R) × SO(d; R)) (13.5.20)

The dimension of this space is,

1 1
(2d)(2d − 1) − 2 × d(d − 1) = d2 (13.5.21)
2 2
This dimension is easily understood in the original formulation with arbitrary metric G and
B-field in a square torus: together they have d2 components.
Two different points in SO(d, d; R)/(SO(d; R) × SO(d; R)) may correspond to the same
lattice. We already know that this is the case when Bij are shifted by integers, and when
the S transformation is applied to both G and B. The shift is realized in terms of an
anti-symmetric matrix K with integer entries,
     
I + K −K p̃L p
γ= =γ L (13.5.22)
K I −K p̃R pR

222
whose effect is to shift B → B + 2K. The matrix γ is easily seen to belong to SO(d, d; R),
and to be a translation or parabolic element, such that,
 
n I + nK −nK
γ = (13.5.23)
nK I − nK

The S-transformation is given by,

 
0 I
S= (13.5.24)
−I 0

The space of inequivalent tori Td from the string point of view are thus parametrized by the
double coset space,

SO(d, d; Z)\SO(d, d; R)/SO(d; R) × SO(d; R) (13.5.25)

In the case of heterotic string, the result extends to include the left-moving 16 dimensional
torus, and the duality group is SO(d, d + 16; Z).

13.6 Rationality and complex multiplication

In section 10.10 we introduced the torus partition function of a two-dimensional conformal
field theory, e.g. in (10.10.7). A CFT is said to be rational if the sum over characters is
finite. In a sense, a rational CFT is one which is exactly solvable, and great effort has been
expended towards obtaining a classification of such theories. In this section, we ask when a
T2 non-linear sigma model is rational.

13.6.1 The case T1

To begin, let us consider the simpler case of an S 1 non-linear sigma model, with the circle
having radius R. In this case, the partition function may be computed to be
1 X 1 m nR 2 1 m nR 2
Z(R, τ ) = q 2 ( R + 2 ) q̄ 2 ( R − 2 ) (13.6.1)
|η(τ )|2 m,n∈Z

This comes from (13.4.6) by choosing α0 = 2 and dressing with a factor of |η(τ )|2 to account
for Virasoro descendants. To address the question of rationality, we must ask when this can
be written as a sum over a finite number of characters. It turns out that this is possible
if R2 ∈ Q. Indeed, say that R2 = 2k` for k, ` ∈ Z. In this case we can write the partition
function as,
r !
2k 1 X 1 2 1 2
Z ,τ = 2
q 4k` (m`+nk) q 4k` (m`−nk) (13.6.2)
` |η(τ )| m,n∈Z

223
It will be useful to write
m = µ + 2kr n = ν + 2`s (13.6.3)
where µ ∈ {0, . . . , 2k − 1}, ν ∈ {0, . . . , 2` − 1}, and r, s ∈ Z. We may then split the sums as
! 2k−1
X 2`−1
r
2k 1 XX 1 1
(2k`(r+s)+µ`+νk)2 4k` (2k`(r−s)+µ`−νk)2
Z ,τ = 2
q 4k` q (13.6.4)
` |η(τ )| µ=0 ν=0 r,s∈Z

We now again change summation variables to,

µ̃ = µ` + νk ν̃ = µ` − νk ρ=r+s σ =r−s (13.6.5)
By definition we have,
µ̃ + ν̃ ≡ 0 (mod 2`) µ̃ − ν̃ ≡ 0 (mod 2k) (13.6.6)
Hence we can rewrite the partition function as a finite sum,
r ! 0
2k X
Z ,τ = χµ̃ (τ )χν̃ (τ̄ ) (13.6.7)
` µ̃,ν̃=0,...,2k`−1

where the primed sum denotes a sum subject to restrictions (13.6.6). The result has been
written in terms of the characters,
1 X k`(ρ+ 2k`
µ̃ 2
)
χµ̃ (τ ) = q (13.6.8)
η(τ ) ρ∈Z

That these are the correct characters for T1 is confirmed by noting that the compact boson
at radius R2 = 2k` is equivalent to a Z` orbifold of the U (1)k current algebra theory.

13.6.2 The case T2

We now proceed to the case of M = T2 . For the moment we will consider only tori that are
trivial products of circles, with radii R1 and R2 . Then the CFT factorizes into two S 1 CFTs
studied above, and the total CFT is rational if and only if the two components CFTs are.
2
In other words, we must take R1,2 ∈ Q. Say concretely that,
2k1 2k2
R12 = R22 = ki , `i ∈ Z (13.6.9)
`1 `2
The complex structure of the spacetime torus, which we will denote as ρ to avoid confusion
with the complex structure τ of the worldsheet torus, is given by ρ = iR1 /R2 up to SL(2, Z)
transformations. Hence ρ satisfies a quadratic equation with integer coefficients,
`1 k2 ρ2 + `2 k1 = 0 (13.6.10)

224
This is precisely the condition that the spacetime T2 admit complex multiplication. Though
we have restricted ourselves to tori which are trivial products of circles here, one can argue
by similar methods that for generic elliptic curves C/L, rationality of the CFT is tantamount
to the presence of complex multiplication on C/L.

• Bibliographical notes
Toroidal compactification of higher space-time dimensions goes back to [241] and was gener-
alized to the Heterotic string in [242, 243]. It was used to map out the moduli space of c = 1
unitary conformal field theories in [244]. Comprehensive overviews of toroidal compactifica-
tion of the bosonic string may be found in the first volume of Polchinski’s book [167], while
toroidal compactification of the superstring, T-duality, and related topics are discussed in
detail in the second volume of Polchinski’s book [168] as well as in the book by Kiritsis [173].
Further useful references on correlation functions for compact scalars, representations of cur-
rent and Kac-Moody algebras and related topics may be found in [245] and [246]. For the
contents of section 13.6, we refer again to [155], and to section 11 on complex multiplication
of these lecture notes.

225
14 S-duality of Type IIB superstrings
In this section, we shall draw together a number of different strands of inquiry addressed
earlier in these lecture notes. We shall study the interplay between superstring amplitudes,
their low energy effective interactions, Type IIB supergravity, and the S-duality symmetry
of Type IIB superstring theory. We begin with a brief review of Type IIB supergravity
which, in particular, provides the massless sector of Type IIB superstring theory. We then
discuss how the SL(2, R) symmetry of Type IIB supergravity is reduced to the SL(2, Z)
symmetry of Type IIB superstring theory via an anomaly. We conclude with a discussion
of how the low energy effective interactions induced by string theory on supergravity may
be organized in terms of modular functions and forms under this SL(2, Z) symmetry, and
match the predictions provided by perturbative calculations of section 12.

14.1 Type IIB supergravity

In section 12 we discussed the expansion of bosonic and superstring amplitudes in powers
of the string coupling constant gs . While this expansion produces only an asymptotic series
in gs , it has the advantage of being valid for all energy ranges. A different, and comple-
mentary, expansion of string theory is in powers of α0 , also referred to as the low energy
expansion. This is the expansion that was used for the string amplitudes in subsection 12.7.
To leading order in this approximation, one omits the effects of all the massive string states,
whose mass-squares are set by the string scale 1/α0 . The remaining massless states precisely
make up the contents of the corresponding supergravity.
To each 10-dimensional perturbative string theory, namely Type I, Type IIA, Type IIB,
and Heterotic E8 × E8 and Spin(32)/Z2 there is a corresponding supergravity. The Type II
theories have the maximal number of 32 supersymmetries, while the other three have 16
supersymmetries. We shall focus here on Type IIB as its corresponding Type IIB superstring
theory has SL(2, Z) duality symmetry, which is the focus of these lecture notes.
The contents of Type IIB supergravity consist of the following fields,

G the metric, whose excitations are the graviton

C2 the complex 2-form potential
τ the complex axion/dilaton field
C4 the real 4-form potential, whose field strength is self-dual
ψ chirality + Weyl spinor gravitino
λ chirality − Weyl spinor dilatino (14.1.1)

The fields G, C2 , τ , and C4 are bosonic fields, while ψ, λ are fermions. Here we shall concen-
trate on the properties and field equations of the bosonic fields. The field strength of C2 is

226
F3 = dC2 , while the field strengths of τ and C4 are given as follows,

P = f 2 dB B = (1 + iτ )/(1 − iτ )
2
Q = f Im (BdB̄) f 2 = (1 − |B|2 )−1
i


F5 = dC4 + 16
C2 ∧ F̄3 − C̄2 ∧ F3 ?F5 = F5 (14.1.2)

The axion-dilaton field τ is valued in the hyperbolic upper half-plane τ2 = Im τ > 0. The field
equation for F5 coincides with its Bianchi identity dF5 = 8i F3 ∧ F̄3 in view of its self-duality
relation. In terms of the components of the fields P, Q, F5 , and K = f (F3 − B F̄3 ),

P = Pµ dxµ K= 1
K dxµ ∧ dxν ∧
3! µνρ
dxρ
Q = Qµ dxµ F5 = 1
F
5! 5µνρστ
dxµ ∧ dxν ∧ dxρ ∧ dxσ ∧ dxτ (14.1.3)

the remaining field equations are conveniently expressed as follows,

0 = ∇µ Pµ − 2iQµ Pµ + 1
K K µνρ
24 µνρ
0 = ∇ρ Kµνρ − iQρ Kµνρ − P ρ K̄µνρ + 2i3 F5µνρστ K ρστ
0 = Rµν − Pµ P̄ν − P̄µ Pν − 16 (F52 )µν − 14 Re (Kµ ρσ K̄νρσ ) + 1
G K ρτ σ K̄ρτ σ
48 µν
(14.1.4)

where Rµν is the Ricci tensor for the metric Gµν . The axion-dilaton field τ may be related
to the axion field χ and the dilaton field Φ,

τ = χ + ie−Φ (14.1.5)

In view of the self-duality condition on F5 , there does not exist a Lorentz-covariant action
from which the field equations of Type IIB supergravity may be deduced using the standard
variational principle. Instead one may define the following action,26

∂µ τ ∂ µ τ̄ (F31 − τ F32 ) · (F31 − τ̄ F32 ) 1

Z  
1
SIIB = dµG R − − − F5 · F5
2κ210 2τ22 2 τ2 4
Z
2
− 2 C4 ∧ F31 ∧ F32 (14.1.6)
κ10

Here dµG is the invariant volume form for the metric G; κ210 is related to the 10-dimensional
Newton constant; and we have F3 = F31 + iF32 and τ = τ1 + iτ2 for real field components
F31 , F32 , τ1 , τ2 . Importantly, the action SIIB should be considered for a field F5 that is un-
constrained by the self-duality condition. The field equations for Type IIB supergravity may
be obtained from this action by varying the unconstrained field F5 , along with all the other
fields, and only subsequently imposing the self-duality condition ?F5 = F5 .
26
Our notation for the contraction of two rank n anti-symmetric tensors is Fn · Fn0 = 1 0µ1 ···µn
n! Fnµ1 ···µn Fn .

227
14.2 From SL(2, R) to SL(2, Z)
The Type IIB supergravity field equations are invariant under the group SL(2, R), under
which the metric G and the anti-symmetric fields C4 and F5 are invariant, and the fields
P, Q, K, τ and F3 = F31 + iF32 transform as follows,

P → P 0 = e2iθ P τ → τ 0 = (aτ + b)/(cτ + d)

Q → Q0 = Q + dθ F31 → (F31 )0 = aF31 + bF32
K → K 0 = eiθ K F32 → (F32 )0 = cF31 + dF32 (14.2.1)

The variable θ is given by,

! 21
(τ + i)(aτ̄ − icτ̄ + b − id)
eiθ =
(14.2.2)
(τ̄ − i) aτ + icτ + b + id

The axion-dilaton takes values in the coset SL(2, R)/U (1) and the phase eiθ performs a
field-dependent U (1) rotation. The gravitino field ψ and the dilatino field λ transform with
U (1) phase factors under SL(2, R), given by ψ → ψ 0 = eiθ/2 ψ and λ → λ0 = e3iθ/2 λ.
The SL(2, R) symmetry discussed above is a symmetry of the classical supergravity
field equations, as well as of the action SIIB and the self-duality condition ?F5 = F5 . A
classical symmetry cannot always be promoted to a symmetry of the corresponding quantized
field theory, in which case the symmetry is said to suffer an anomaly. Phase rotations on
chiral fermions often suffer anomalies, and this is also the case here with the field-dependent
U (1) transformations acting on the gravitino and dilatino chiral fermions. A treatment of
anomalies is beyond the scope of this paper, so we shall just state the result.
It turns out that the continuous SL(2, R) symmetry is broken by the U (1)-anomaly, but
the quantum theory retains a discrete symmetry, under which the axion field is shifted by an
integer χ → χ0 = χ + b with b ∈ Z. The discrete nature of the remaining symmetry is due to
the fact that the axion field couples to the topological charge density of the D−1 instantons
in Type IIB superstring theory, whose total charge is quantized. The shift in the axion field
χ → χ0 = χ + b with b ∈ Z that survives quantization also shifts the complex axion-dilaton
field τ → τ 0 = τ + b, and is supplemented by the transformation τ → −1/τ . Together
these transformations generate the discrete group SL(2, Z). This symmetry extends to a
symmetry of the entire Type IIB superstring theory.
This phenomenon is actually familiar from an extension of the Standard Model in which
an axion field is included. Continuous shifts in the axion field again suffer an anomaly but
discrete shifts survive. In this case the discrete nature of the shift is due to the coupling of
the axion field to the topological charge density of Yang-Mills instantons whose topological
charge is quantized in terms of the second Chern class of the gauge group.

228
14.3 Low energy effective interactions
Supergravity, as a classical theory, is valid for all values of the string coupling. Thus, it
permits investigations into certain strong coupling phenomena such as solitonic states, NS
branes and D-branes. Supergravity has been an invaluable tool in the search for semi-realistic
compactifications of superstring theory, such as on tori, oribifolds, and Calabi-Yau spaces.
However, supergravity alone is not a consistent quantum theory. In 10 space-time di-
mensions, it exhibits UV divergences starting at one loop. The UV convergence situation
is somewhat improved by lowering the dimension of space-time. In four space-time dimen-
sions the four-graviton amplitude is UV convergent up to five loops, but it is likely that
non-renormalizable UV divergences start occurring at seven loops, rendering the quantum
theory inconsistent, or more accurately, incomplete. Its UV completion is precisely super-
string theory. From the vantage point of superstring theory, supergravity should be thought
of as an effective low energy field theory that captures the dynamics of the massless states
of Type IIB string theory, to leading order in the α0 expansion.
The validity of an effective field theory may be extended beyond its leading order contri-
butions. Viewed as an effective field theory of Type IIB superstring theory, the corresponding
supergravity may be supplemented with contributions of higher order in α0 . Such contribu- √
tions are due to the effects of massive string states whose mass, we recall, is of order 1/ α0
and are referred to as effective interactions In the approximation
√ where the momenta used
0
to probe string amplitudes are small compared to 1/ α , the effective interactions are local,
as is illustrated in Figure 15 for the exchange of a massive string state. Although these string
induced effective interactions are highly suppressed they provide systematic corrections to
supergravity in an expansion in powers of α0 .

Figure 15: The exchange of a massive string state, indicated by the thick line in the left
panel, induces a local effective interaction indicated by a thick dot in the right panel.

The space-time we observe has dimension four and is approximately flat when its cur-
vature is measured in units of the Planck scale. Therefore, if the space-time of superstring
theory is to be 10 then the six extra dimensions must have radii that are smaller than
the smallest length scales accessible to experiment today. Physically realistic superstring
theories are often based on space-times of the form R10−d × Md where Md is a compact
manifold or orbifold and the physically observed dimension of space time corresponds to
d = 6. For the case where Md is a flat torus, superstring perturbation theory continues
to be well-understood. When Md is an orbifold of a torus, the space Md is flat away from

229
isolated point-like singularities, and string theory on toroidal orbifold spaces still lends itself
to reasonably explicit solutions. However, when Md is an arbitrary curved manifold, the
predictions of superstring theory are much more difficult to obtain. This is the case even
when space-time is a curved maximally symmetric space such as AdS5 × S 5 , for which the
quantization of the superstring is still largely an unresolved problem. An important excep-
tion is when the curvature of Md is uniformly small compared to the Planck scale, a case
that is referred to as the large radius expansion, and that we shall discuss next.
Restricting, for example, to purely gravitational effects, the supergravity Lagrangian
reduces to the Einstein-Hilbert term given by the Ricci scalar for the space-time metric.
In Type II superstring theory, for example, the lowest order α0 correction to the Einstein-
Hilbert action is given by a term which we symbolically represent by R4 where R stands for
the Riemann tensor, and the contribution to the action is obtained via a special contraction
of the four factors, consistent with space-time supersymmetry, as will be explained below.
This effective interaction arises at order (α0 )3 . Higher order effective interactions involve
more derivatives and higher powers of α0 and are symbolically represented by D2k R4 where
D represents a covariant derivative, again suitably contracted.

14.4 SL(2, Z) duality in Type IIB superstring theory

Type IIB superstring theory exhibits SL(2, Z) duality symmetry which leaves the space-time
metric GE in Einstein frame and the self-dual five form invariant, and transforms the 3-form
flux field strengths linearly into one another,
 1    1  
F3 a b F3 a b
→ ∈ SL(2, Z) (14.4.1)
F32 c d F32 c d

As for the axion-dilaton field obtained by combining the dilaton field Φ and the axion field
χ into a complex scalar field τ = χ + ie−Φ , transformations under SL(2, Z) occur as follows,

aτ + b
τ→ (14.4.2)
cτ + d
The transformation properties of the fermionic fields are more complicated, and will not
be reproduced here. In fact, when one accounts for the fermionic fields the true duality
symmetry of Type IIB is not SL(2, Z), but rather an extension of it known as the meta-
linear group M L+ (2, Z); see [247] for details. This subtlety will not affect the purely bosonic
analysis below.
Being a symmetry of Type IIB string theory, SL(2, Z) must be a symmetry of the low
energy effective action of the theory. The effective interactions accessible from the four-
graviton amplitude are of the form D2k R4 , as explained above, but they have coefficients
that depend upon the vacuum expectation value of the dilaton gs = eΦ and the axion χ.

230
Expressing the effective action in terms of the Einstein frame metric GE , with associated
Riemann tensor RE , and the vacuum expectation value τ , we find to order (α0 )7 ,
Z  
dµG E0 (τ )R4E + E4 (τ )D4 R4E + E6 (τ )D6 R4E + E8 (τ )D8 R4E + O((α0 )8 ) (14.4.3)

where the coefficients E2k (τ ) are real-valued scalar functions of τ . Derivatives of τ may occur
as well but will be systematically omitted here for simplicity. Since the action must be
SL(2, Z)-invariant and, in Einstein frame, GE and RE are invariant as well, we see that each
coefficient E2k (τ ) must be invariant,
   
aτ + b a b
E2k = E2k (τ ) ∈ SL(2, Z) (14.4.4)
cτ + d c d

a property which makes them into non-holomorphic modular functions.

String perturbation theory calculations are carried out in the string frame metric which
we shall denote by Gµν and which is related to the Einstein frame metric by Gµν = eΦ/2 GEµν .
Converting the expression for the effective action of (14.4.3) to string frame may be done by
using the following relations (for constant values of τ ),
p √
GE E2k (τ ) DE2k R4E = e(k−1)Φ/2 G E2k (τ ) D2k R4 (14.4.5)

where D and R now stand respectively for the covariant derivative and Riemann tensor in
the string frame. Now let us carry out the following exercise: express combinations of the
non-holomorphic Eisenstein series with effective interactions DE2k R4E in the Einstein frame
such that the leading perturbative contribution in string frame is tree-level,

1 3/2 p π2
π GE E 3 (τ )R4E = e−2Φ ζ(3)R4 + R4 + · · ·
2 2 3
1 5/2 p 2π 4 2Φ 4 4
π GE E 5 (τ )DE4 R4E = e−2Φ ζ(5)D4 R4 + e D R + ··· (14.4.6)
2 2 135
Comparing the terms in the above table which have the Φ-dependence of tree-level pertur-
bative contributions with the known result of (12.10.17) at genus one and (12.11.3) and
(12.11.7) at genus two, we see that their dependence on odd zeta-values is precisely repro-
duced. This observation would suggest that the coefficients E2k (τ ) in (14.4.3) may be given
by the non-holomorphic Eisenstein series.

14.5 Eisenstein series from supersymmetry and S-duality

We will now show how supersymmetry and S-duality can be used to prove the identifications
between E2k (τ ) and the appropriate Eisenstein series. We will begin with the case of E0 (τ ),

231
namely the coefficient of R4 . Because the metric G is in the same supermultiplet as all of the
other supergravity fields reviewed in section 14.1, supersymmetry requires that the R4 term
in the low energy effective action be accompanied by a number of analogous terms related
by supersymmetry. To make this more concrete, we first introduce some notation. As we
have discussed above, the complex 2-form potential C2 , and consequently the corresponding
field strength F3 , transforms as a doublet under SL(2, Z). It is useful to package these field
strengths into an SL(2, Z) singlet via,
eiφ
M=√ (F31 − τ F32 ) (14.5.1)
−2iτ2
where the superscript on F3 is a doublet index. We further define the “super-covariantized”
combination,
cµνρ = Mµνρ − 3ψ [µ γνρ] λ − 6iψ ∗[µ γν ψρ]
M (14.5.2)
obtained by dressing M with appropriate combinations of the gravitino and dilatino fields
ψ and λ. The combination M c has the property that under supersymmetry transformations
it does not contain derivatives of the transformation parameter.
In terms of this supercovariant field strength, the supersymmetric completion to the R4
term can be written as [248],
Z 
(3) (12,−12) 16 (11,−11) c 14 (8,−8) c8
S = dµG E0 λ + E0 M λ + · · · + E0 M

(0,0) (−12,12) ∗ 16
+ · · · + E0 R4 + · · · + E0 (λ ) (14.5.3)

where λ16 and Mcλ14 (which will be the main terms of interest to us below) are short for the
following contractions,
1
λ16 = a ...a λa1 . . . λa16
16! 1 16
cλ14 1 c
M = Mµνρ (γ µνρ γ 0 )a15 a16 a1 ...a16 λa1 . . . λa14 (14.5.4)
14!
(w,−w)
The SL(2, Z) symmetry of Type IIB demands that the coefficient functions E0 (τ ) be
modular forms of holomorphic weight w and anti-holomorphic weight −w. The coefficient
(0,0)
function E0 (τ ) = E0 (τ ) is the original modular invariant coefficient function of R4 . The
coefficient functions are furthermore required by supersymmetry to satisfy
(w,−w)
E0 (τ ) = Dw−1 . . . D0 E0 (τ ) (14.5.5)
 
with Dw = iτ2 ∂
∂τ
− i 2τw2 the covariant derivative mapping weight (w, w0 ) modular forms to
(w,−w)
weight (w + 1, w0 − 1) modular forms. Because of these relations, constraints on E0 for

232
any w can imply non-trivial constraints on E0 . We will now see how to use supersymmetry
(12,−12) (11,−11)
to obtain differential constraints on E0 and E0 in particular, which will suffice to
prove that E0 is, up to normalization, equivalent to E 3 (τ ).
2
Let us begin by schematically writing the low-energy effective action as
1  (0)
S + (α0 )3 S (3) + (α0 )4 S (4) + (α0 )5 S (5) + . . .

S= 0 4
(14.5.6)
(α )

with S (3) in particular given in (14.5.3) above. Likewise, a generic SUSY transformation δ
can be expanded in powers of α0 as

δ = δ (0) + (α0 )3 δ (3) + (α0 )4 δ (4) + (α0 )5 δ (5) + . . . (14.5.7)

Requiring that the action has supersymmetry implies that we must have

δ (0) S (0) = δ (0) S (3) + δ (3) S (0) = δ (0) S (5) + δ (5) S (0) = · · · = 0 (14.5.8)

on the equations of motion.

We now consider the first two terms in the action,
(12,−12) 16 (11,−11) c 14
L(3) = E0 λ + E0 Mλ (14.5.9)

A crucial feature of these two terms is that they are related by a subset of the SUSY trans-
formations that do not mix with any of the other terms at this order, though we will have to
keep track of variations of the lowest order action S (0) by δ (3) . It is a straightforward exercise
to find the lowest-order δ (0) supersymmetry variation of L(3) . Denoting the supersymmetry
parameter by  and keeping only terms proportional to λ16 ψµ∗  gives
 
(0) (3) (12,−12) (11,−11)
δ L |λ16 ψµ∗  = −8i E0 (τ ) + 108D11 E0 (τ ) (14.5.10)

On the other hand, the δ (3) variation of S (0) cannot produce any such term. Hence we require
δ (0) L(3) |λ16 ψµ∗  = 0 alone, which demands

(11,−11) 1 (12,−12)
D11 E0 (τ ) = − E (τ ) (14.5.11)
108 0
Further constraints can be obtained by considering the term in the supersymmetry variation
proportional to λ16 λ∗ ∗ . In this case the computation is slightly more involved since now
there is a contribution from the δ (3) variation of S (0) . The analysis will not be reproduced
here, but the result is the following constraint
(12,−12) (11,−11)
D−12 E0 (τ ) + 3240 E0 (τ ) − 90g = 0 (14.5.12)

233
where g(τ, τ ) is an unknown function. A final constraint may be obtained by demanding
closure of the SUSY algebra on λ∗ , giving
(12,−12)
32D11 g = E0 (τ ) (14.5.13)
We have now obtained three constraints (14.5.11), (14.5.12), and (14.5.13). Combining them,
(12,−12)
we may derive a Laplace eigenvalue equation for E0 ,
 
(12,−12) 3 (12,−12)
∆ E0 (τ ) = −132 + E0 (τ ) (14.5.14)
4
where we have noted that, when acting on weight-(12, −12) modular functions, the Laplacian
is given by ∆ = 4D11 D−12 .
(12,−12)
It remains only to translate this differential constraint on E0 (τ ) to a differential
constraint on E0 (τ ), which can be done using the relation (14.5.5) between the two. This
gives the following Laplace eigenvalue equation for E0 (τ ),
3
∆E0 (τ ) = E0 (τ ) (14.5.15)
4
Together with the asymptotic expansion obtained from string perturbation theory, this equa-
tion uniquely fixes E0 (τ ) to be the non-holomorphic Eisenstein series E 3 (τ ). Indeed, we saw
2
already in (4.2.5) that the Eisenstein series satisfy precisely such eigenvalue equations.
We next briefly outline the analogous calculation for the D4 R4 term. As before, the first
step is to identify a subset of the terms appearing in S (5) which mix with each other, but not
with other terms, under some subset of the supersymmetry transformations. In the current
case, a judicious choice is the following,
(14,−14) (13,−13) (13,−13)
L(5) = λ16 M
c4 E4 + λ15 γ µ ψµ∗ M
c4 E4 + λ16 M
c2 M cρ1 ρ2 ρ3 Ee4
cρ1 ρ2 ρ3 M (14.5.16)
(w,−w)
As before, modular invariance demands that the coefficient functions E4 be modular
(0,0) 4 4
forms of weight-(w, −w), with E4 = E4 the coefficient of D R itself. Note that there are
(13,−13)
two coefficient functions of weight-(13, −13) above, which we have denoted by E4 and
(13,−13)
E4
e . We now proceed by again checking invariance under supersymmetry. Considering
the term proportional to λ16 ψµ∗ , we find only the contribution from δ (0) L(5) , and demanding
that this vanishes gives the constraint
(13,−13) (14,−14)
2D13 E4 − 11 E4 =0 (14.5.17)
Considering instead the term proportional to λ16 λ∗ ∗ , we obtain both a contribution from
the δ (0) variation of L(5) as well as the δ (5) variation of S (0) . Demanding the vanishing of the
combination gives the constraint
(14,−14) (13,−13) 9i e(13,−13) 3
2D−14 E4 + 15 E4 − E4 − 1080i g1 − i g2 = 0 (14.5.18)
16 4
234
with g1 (τ, τ ) and g2 (τ, τ ) undetermined functions. Closure of the supersymmetry algebra on
λ∗ further gives rise to the following three equations,

(14,−14) (13,−13) 1 (13,−13)

−192i D13 g1 = E4 − 108 g1 = Ee4 i g2 + 191ig1 = E4 (14.5.19)
2
(14,−14)
These constraints can then be combined to obtain a Laplace eigenvalue equation for E4 ,
namely
 
(14,−14) 15 (14,−14)
∆E4 = −182 + E4 (14.5.20)
4
(14,−14)
Recalling that E4 = D13 . . . D0 E4 , we conclude that E4 satisfies the equation
15
∆E4 (τ ) = E4 (τ ) (14.5.21)
4
This eigenvalue equation, together with the asymptotics at the cusp, fix E4 to be equal to
E5/2 (τ ) up to an overall constant factor.
Finally, for the coefficient of the D6 R4 term, analogous steps can again be carried out,
though one now obtains an inhomogeneous Laplace eigenvalue equation,

(∆ − 12)E6 (τ ) = −6π 3 E3/2 (τ )2 (14.5.22)

the solution to which is no longer an Eisenstein series. No closed form solution is known to
this differential equation, but it is possible to get a closed form expression for the Laurent
(0)
polynomial E6 of E6 , namely the portions of E6 which are power-law in τ2 and independent of
τ1 . Fortunately, the Laurent polynomial pieces are the most interesting for the present pur-
poses, since they are the ones which can be compared to calculations in string perturbation
theory.
(0)
To obtain the Laurent polynomial E6 (τ ) we begin by inserting the Fourier series of
E3/2 (τ ), given in (4.2.16), into (14.5.22) and keeping only the Laurent polynomial pieces,
 2
(0) 3/2 2 2 −1/2
(∆ − 12)E6 (τ ) = −6 2ζ(3)τ2 + π τ2
3
!
X
+64π 2 τ2 |N |−2 σ2 (|N |)2 K1 (2πτ2 |N |)2 (14.5.23)
N 6=0

Because the Laplacian is given by ∆ = τ22 (∂τ21 + ∂τ22 ), and hence the derivatives ∂τ22 are
(0)
dressed with factors of τ22 , the function E6 must be a polynomial in τ2 with powers being
those appearing on the right side, namely τ23 , τ2 , and τ2−1 , together with the powers solving

235
the homogeneous equation (∆ − 12)f = 0, namely τ2−3 and τ24 . The term τ24 can be ruled out
on physical grounds, since there should be no term more singular than the tree-level τ23 term.
On the other hand, the τ2−3 contribution corresponds to a three-loop contribution which can
be (and indeed is) non-zero.
(0)
The general Ansatz for E6 (τ ) is then,
(0)
E6 (τ ) = a1 τ23 + a2 τ2 + a3 τ2−1 + a4 τ2−3 (14.5.24)

By plugging into (14.5.22) we may uniquely fix a1 , a2 , a3 , giving,

(0) 4 4π 4 −1
E6 (τ ) = 4ζ(3)2 τ23 + π 2 ζ(3)τ2 + τ + a4 τ2−3 (14.5.25)
3 15 2
These match with the expectations from string perturbation given in (12.7.4), (12.10.17), and
(12.11.7) at tree-level, one-loop, and two-loops, respectively. As for the three-loop coefficient
a4 , this cannot be fixed by simply inserting the Ansatz into (14.5.22), since this term already
solves the homogenous equation. Instead, this coefficient was fixed in [249] by multiplying
both sides of the Laplace equation by E4 (τ ) and integrating over the fundamental domain,
using the results presented in Lemma 12.1. The final result is

8π 6
a4 = (14.5.26)
8505
We note in closing that beyond D6 R4 , the terms in the low energy effective action are no
longer protected by supersymmetry, and thus we do not expect to be able to extend the
above analysis to study them.

14.5.1 Non-renormalization Theorems

Perhaps the most striking implications of the modular structure of the R4 , D4 R4 , and
D6 R4 effective interactions is that they make predictions for the amplitudes at all orders
of perturbation theory, i.e. all genera. From the Fourier series expansion of the functions
E 3 (τ ) we see that, aside from the leading perturbative contribution, there is only one other
2
−1
perturbative contribution, given by the second term, and behaving as τ2 2 . In the string
frame, this gives a contribution of order τ20 which corresponds to genus one. This means that
from superstring perturbation theory, the R4 effective interaction receives contributions from
genus zero and genus one only. There are no contributions to genus 2 and higher. Similarly,
for the D4 R4 effective interaction, there is a genus zero contribution and the only other
−3
contribution behaves as τ2 2 in the Einstein frame, or τ2−2 in the string frame. Hence there
is no genus one correction to the D4 R4 effective interaction but there is a non-zero genus
two contribution. There are no higher contributions.

236
• Bibliographical notes
The existence of a web of dualities between different supergravities and string theories was
proposed in [250], and led to the discovery of M-theory in [251]. The action of SL(2, Z) on
branes in Type IIB was developed in [252], while evidence of the duality between Type I and
Heterotic string theory was given in [253]. Useful lecture notes may be found in [254], and
an early account of the role played by automorphic functions in dualities is in [255, 256].
The textbooks by Polchinski [167, 168], Johnson [170], and by Becker, Becker and Schwarz
[171] provide excellent and comprehensive accounts of string dualities.
The use of S-duality to constrain the low-energy effective action goes back to [257],
where the coefficient of the R4 term was related to the non-holomorphic Eisenstein series
E3/2 (τ ) using results from tree-level string theory and D-instantons. A derivation using
supersymmetry, in the way reviewed above, was provided in [248] and extended in [258].
That the same techniques could also be applied to the D4 R4 term was proposed in [248],
and used to provide a full derivation in [259]. As we have mentioned in the text, there is no
closed form expression for the D6 R4 term, though it is known to satisfy a Laplace eigenvalue
equation; this property and others were discussed in [260, 261, 249, 262].
As mentioned in section 14.5.1, the results quoted in the previous paragraph imply a
number of non-renormalization theorems, some of which were anticipated from superstring
perturbation theory in [263, 188]. For the R4 term, the S-duality and supersymmetry analysis
predict the absence of perturbative corrections beyond one-loop, and indeed the vanishing
at genus two was proven in [220]. Conversely, for the D4 R4 term there is expected to be no
genus one correction, as was verified in [201], but there is a non-zero genus-two contribution
which was successfully computed and matched in [223].
The systematic study of genus-one contributions to effective interactions of the form
2` 4
D R with ` ≥ 3 was initiated in [91, 92]. From this grew the notion of modular graph
functions [97, 102], introduced in section 9.
One of the hallmarks of superstring theory is the infinite tower of BPS states, which in-
cludes black hole microstates. If the string background under consideration contains a torus,
then the functions which count these microstates often enjoy nice modular or automorphic
properties. A particularly well-studied setup is string theory compactified to 4d with N = 4
supersymmetry. This can be engineered by considering Type II superstrings on K3 × T 2 or
Heterotic strings on T 6 . Either way, the theory is expected to have an SL(2, Z)×SO(22, 6, Z)
duality symmetry, with the corresponding counting functions expressible as Siegel modular
forms [264]. The contribution of multi-centered black holes was studied in [265], while the
single-centered black holes were shown to have interesting connections with mock modular
and Jacobi forms in [32]. Recent connections have been made with Hurwitz class numbers
[266] and class groups [267] as well. More generally, the tools of Rademacher sums and Farey
tail expansions have been applied to the study of microstate counting in [268, 269, 270].

237
15 Dualities in N = 2 super Yang-Mills theories
In this penultimate section, we shall discuss dualities in Yang-Mills theories with extended
supersymmetry in four-dimensional Minkowski space-time. We briefly review supersymmetry
multiplets of states and fields and the construction of supersymmetric Lagrangian theories
with N = 1, 2, 4 Poincaré supersymmetries. We then discuss the SL(2, Z) Montonen-Olive
duality properties of the maximally supersymmetric N = 4 theory and the low energy
effective Lagrangians for N = 2 theories via the Seiberg-Witten solution. We shall close this
section with a discussion of dualities of N = 2 superconformal gauge theories, which possess
interesting spaces of marginal gauge couplings. In some cases these spaces of couplings can
be identified with the moduli spaces for Riemann surfaces of various genera.

15.1 Super Yang-Mills: states and fields

The quantum field theories we shall consider here are invariant under the Poincaré algebra,
whose generators of translations Pµ and Lorentz transformations Lµν satisfy the following
structure relations,27
[Lµν , Lρσ ] = ηνρ Lµσ + ηµσ Lνρ − ηµρ Lνσ − ηνσ Lµρ
[Lµν , Pρ ] = ηνρ Pµ − ηµρ Pν
[Pµ , Pν ] = 0 (15.1.1)
In supersymmetric theories the Poincaré algebra is extended to a super Poincaré algebra by
supplementing Pµ and Lµν with N supercharges QIα and their adjoints Q̄α̇I = (QIα )† for spinor
index α = 1, 2 and I = 1, · · · , N . The supercharges commute with Pµ , are Weyl spinors
under the Lorentz generators Lµν , and obey the following anti-commutation relations,
{QIα , Q̄β̇J } = 2 σαµβ̇ Pµ δJI I, J = 1, · · · , N
{QIα , QJβ } = 2 αβ Z IJ (15.1.2)
The generators Z IJ = −Z JI can arise only when N ≥ 2. They are central generators in the
Poincaré superalgebra since they commute with one another and with all other generators.
When Z IJ = 0, the supersymmetry algebra is invariant under a U (N )R automorphism group
which, in Physics, is referred to as the R-symmetry group.
27
Our conventions are as follows. The non-zero components of the Minkowski metric ηµν are −η00 =
η11 = η22 = η33 = 1. The Clifford-Dirac algebra is {γ µ , γ ν }
= −2η µν I4 , where Id is the d × d
identity
µ
matrix. In a basis where the chirality matrix γ5 = I02 −I0 2 is diagonal, we have γ µ = σ̄0µ σ0 where
σ 0 = σ̄ 0 = −I2 and σ i = −σ̄ i are the standard Pauli matrices for i = 1, 2, 3. Two-component spinor indices
α, α̇ are raised and lowered with the help of the anti-symmetric symbols αβ , α̇β̇ and αβ , α̇β̇ normalized to
12 = 1̇2̇ = −12 = −1̇2̇ = 1. We use the notations ψχ = ψ α χα , ψ̄ χ̄ = ψ̄α̇ χ̄α̇ , and χσ µ ψ̄ = χα σαµα̇ ψ̄ α̇ . For
more details on spinor index notation, see for example [271].

238
 Applying a supercharge to a state changes the spin by ± 21 so that the values of the
spin in the multiplets of the Poincaré superalgebra must increase with N . When N > 4, the
multiplets necessarily involve states of spin greater than 1, which cannot be accommodated in
Yang-Mills theory. For N ≤ 8 they can, however, be accommodated in supergravity, but this
requires states of spin 2. Thus N = 4 is the maximal number of Poincaré supersymmetries
allowed for supersymmetric Yang-Mills theories in four dimensions.

15.1.1 States
The spectrum of states in a supersymmetric theory exhibits a characteristic pattern. All
states have positive or zero energy. If the state of lowest energy in the spectrum—namely
the ground state—has exactly zero energy then supersymmetry is said to be unbroken or
manifest. In this case all positive energy states occur in boson-fermion pairs of equal mass,
momentum, and internal quantum numbers. If the ground state has strictly positive energy,
however, the ground state is not supersymmetric and supersymmetry is said to be sponta-
neously broken. Bosons and fermions of identical internal quantum numbers do not generally
have the same mass.
In this section we shall consider only the case where supersymmetry is manifest. For
the case of N = 1, this is the complete story. For N ≥ 2, there is an additional interplay
between the mass of the state and its central charge.
For N = 2 there is a single complex-valued central charge Z = Z 12 , which we decom-
pose into its absolute value and its phase, Z = eiϕ |Z| for ϕ ∈ R. We consider a state
with mass M 6= 0 and use a Poincaré transformation to the rest frame of this state with
momentum P µ = (M, 0, 0, 0). In terms of the combinations Q± 1 1
α = 2 (Qα ∓ e
−iϕ 0
σαβ̇ Q2β̇ ) the
anti-commutation relations of (15.1.2) are equivalent to the relations,

Qα , (Q±
 ± †
= δαβ (M ± |Z|)

β) (15.1.3)

while all other anti-commutators vanish identically. Since the operators Q± ± †


α , (Qα ) are
self-adjoint and positive in a unitary theory, we obtain the following so-called BPS bound
between the mass M and the central charge Z of any state,

M ≥ |Z| (15.1.4)

In the limit where the state becomes massless, its central charge must vanish. Of particular
interest are the massive states for which the BPS bound is saturated, say by M = |Z|. In
these states, which are referred to as BPS states, the supersymmetry algebra reduces to,
†
Qα , (Q−
 − †
Qα , (Q+
 +
= 2M δαβ

β) β) =0 (15.1.5)

while all other anti-commutators vanish. The operators Q− − †

α and (Qα ) both annihilate the
BPS states which means that BPS states are invariant under half of the total number of

239
Poincaré supersymmetries. More specifically they are referred to as 12 -BPS states, and they
make up a supermultiplet that is half as long as the non-BPS supermultiplets.
For N = 4, the matrix of central charges Z IJ = −Z JI has two complex-valued eigenvalues
Z1 and Z2 . The BPS bound is now M ≥ |Z1 |, |Z2 |. When |Z1 | = 6 |Z2 | one obtains 21 -BPS
states by setting M equal to the maximum of |Z1 | and |Z2 |. But when |Z1 | = |Z2 |, a further
shortening of the supersymmetry multiplet takes place to 41 -BPS states. It is a general result
of CPT invariance in four space-time dimensions that supersymmetric Yang-Mills theories
with N = 3 automatically enjoy the full N = 4 supersymmetry, so that the case N = 3
need not be considered separately.

15.1.2 Fields
A super Yang-Mills theory with gauge group G and associated Lie algebra g is constructed in
terms of the customary fields of four-dimensional quantum field theory absent gravity, namely
spin-1 gauge fields Aµ in the adjoint representation of G, together with spin- 21 left Weyl
fermion fields ψα , λα and spin-0 scalar fields φ, h transforming under various representations
of G. We recall that, in even dimensions, a Dirac spinor is the direct sum of two Weyl spinors
of opposite chirality and that, in four dimensions, a Majorana spinor is equivalent to a Weyl
spinor. The field multiplets for N = 1, 2, 4 are as follows:

• N = 1 Gauge multiplet (Aµ , λα ) in the adjoint representation of G, where Aµ is the

gauge field and λα is a left Weyl fermion referred to as the gaugino;
• N = 1 Chiral multiplet (ψα , φ) in a representation R of G, where ψα is a left Weyl
fermion and φ is a complex scalar field;

• N = 2 Gauge multiplet (Aµ , λα , λ̃α , φ) in the adjoint representation of G, where Aµ

and φ are singlets under SU (2)R while (λα , λ̃α ) is a doublet;
• N = 2 Hypermultiplet (h, h̃, ψα , ψ̃α ) in a representation R of G, where (h, h̃) is a dou-
blet under SU (2)R while ψα , ψ̃α are singlets;

• N = 4 Gauge multiplet (Aµ , λA I A

α , φ ) in the adjoint representation of G where λα with
I
A = 1, · · · , 4 are left Weyl fermions in the 4 of SU (4)R and φ with I = 1, · · · , 6 are
complex scalars in the 6 of SU (4)R while Aµ is a singlet.

The fields of an N = 1 theory may be collected in fully off-shell superfields with linear
transformations under the action of the super Poincaré algebra, as will be made explicit in
the next subsection. The construction of superfields for higher numbers of supersymmetries
is, however, much more complicated and will not be addressed here. For this reason it will
be useful to decompose the field content of the N = 2 and N = 4 theories in terms of N = 1

240
superfields. In this spirit, the N = 2 gauge multiplet is the direct sum of an N = 1 gauge
multiplet and an N = 1 chiral multiplet, both in the adjoint representation of G. The N = 2
hypermultiplet is a sum of two N = 1 chiral multiplets in complex conjugate representations
of one another R ⊕ R̄. Similarly, the N = 4 gauge multiplet is the direct sum of an N = 2
gauge multiplet and an N = 2 hypermultiplet, both in the adjoint representation of G.
Note that for N = 4 theories the U (1)R factor of the automorphism group U (N )R of the
supersymmetry algebra is anomalous and fails to be a quantum symmetry. The charge
assignments of the U (1)R factor in the case of N = 1 and N = 2 will not be important in
the sequel, and we shall not spell them out here.
Before moving on, we remark that when R is a pseudoreal representation, and hence ad-
mits an antisymmetric invariant tensor IJ , the fields hI and h̃J in the N = 2 hypermultiplet
(with indices I, J in the representation R shown explicitly) can be constrained to satisfy,
hI = IJ h̃J (15.1.6)
This constraint is compatible with N = 2 supersymmetry, and halves the number of degrees
of freedom in the multiplet. The resulting multiplet is referred to as a half-hypermultiplet.
More generally, given a complex representation R there is a standard pseudo-real structure
on R ⊕ R, and hence we may define half-hypermultiplets in the representation R ⊕ R. These
are precisely the same as full hypermultiplets in the complex representation R.

15.2 Super Yang-Mills: Lagrangians

Two types of supersymmetric Lagrangians will be of interest to us: those that correspond
to renormalizable quantum field theories, and those that collect the low energy effective
interactions of some arbitrary Lagrangian or non-Lagrangian quantum field theory. In either
case, we shall restrict attention to Lagrangians in which bosonic fields have a total of at
most two derivatives while fermions have a total of at most one derivative acting on any
field. Furthermore, the Lagrangians for theories with N = 2 and N = 4 supersymmetries
will be expressed in terms of N = 1 chiral and vector superfields, using the decomposition of
the corresponding multiplets given in the penultimate paragraph of the preceding subsection.
Throughout we shall assume that the gauge group G is compact with Lie algebra g, and of
the form of a semi-simple group times a certain number of U (1) factors.
To introduce N = 1 superfields, we supplement the coordinates xµ of Minkowski space-
time with Grassmann spinor variables θα and θ̄α̇ of odd grading. The superfield Φ for the
N = 1 chiral multiplet containing the fields (φ, ψα ) is given by,
Φ = φ + 2ψα θα + F θα θα (15.2.1)
where F is a non-dynamical complex-valued scalar auxiliary field. The chiral superfield Φ is
generally matrix-valued with components Φi and transforms under an arbitrary representa-
tion R of the gauge algebra g with i = 1, · · · , dim R. The superfield V for the N = 1 gauge

241
multiplet satisfies a reality condition V † = V and transforms in the adjoint representation
of the gauge algebra g. The general decomposition of the superfield V is more complicated
than that of the chiral multiplet and we present it here in Wess-Zumino gauge only,

1
V = −θσ µ θµ + iθθθ̄λ̄ − iθ̄θ̄θλ + θθθ̄θ̄D (15.2.2)
2
where D is a real-valued scalar auxiliary field. The vector superfield V is generally matrix-
P a a
valued and may be decomposed V = aT V onto the generators T a of g, with a =
1, · · · , dim g, in the adjoint representation of the gauge algebra g.

15.2.1 N = 1 Lagrangians
In terms of the N = 1 gauge superfield V and chiral superfield Φ, the most general La-
grangian density, under the assumptions spelled out earlier, is of the following form,
Z Z  
4 V † 2 a b
L = d θ K(e Φ, Φ ) + Re d θ U (Φ) + τab (Φ)W W (15.2.3)

in superspace notation. Here K is the real-valued Kähler potential, U is the superpotential,

and τab is the matrix of couplings and mixings of the field strengths Wαa of the gaugeP
fields V a .
The functions U (Φ) and τab (Φ) are locally holomorphic in Φ. The superfield Wα = a Wαa T a
is obtained from the vector superfield V by taking,

1 ∂
Wα = − D̄D̄ e−V Dα eV + iσαµα̇ θ̄α̇ ∂µ


Dα = (15.2.4)
4 ∂θα
The field Wα is a chiral superfield since D̄α̇ Wβ = 0 by construction.

15.2.2 Renormalizable N = 2 Lagrangians

A general four-dimensional N = 2 Lagrangian theory is specified by a gauge algebra g which
determines the vector multiplets, as well as a representation R of g for the hypermultiplets,
or half-hypermultiplets if R is pseudo-real.28 The N = 2 gauge multiplet consists of an
N = 1 gauge multiplet V and an N = 1 chiral multiplet Φ in the adjoint representation
of g, while the N = 2 hypermultiplet consists of two N = 1 chiral multiplets H, H̃ in the
representations R and R̄ of g, respectively. The N = 2 half-hypermultiplet consists of a
single N = 1 chiral multiplet H in a pseudo-real representation R.

28
This data is not quite complete: in general, additional global data such as the spectrum of line operators
is necessary to fully specify the theory. We will discuss such additional data in the context of N = 4 Yang-
Mills in section 15.5, but will otherwise largely ignore this subtlety. As long as one is concerned only with
the local operator spectrum of the theory, no harm arises from this omission.

242
 Let us consider first a renormalizable N = 2 theory without hypermultiplets. We begin
by arguing that the superpotential U (Φ) must vanish. Indeed, for the theory to be renor-
malizable, one requires that τab be independent of Φ and that the superpotential U (Φ) be at
most cubic in the chiral superfield Φ. Upon integrating over superspace, the final two terms
of the Lagrangian (15.2.3) give,
Z n o
2 a a b
Re d θ U (Φ) = Re F ∂a U (φ) + ψ ψ ∂a ∂b U (φ) (15.2.5)
Z   
i a bµν 1 a bµν a
Re d2 θ τab W a W b = −Re τab Fµν F − Fµν F̃ + λ σ µ Dµ λb
2 2
where ∂a = ∂/∂φa . To obtain a Lagrangian with N = 2 supersymmetry, the combined
contributions of the N = 1 gauge and chiral multiplets must be invariant under the SU (2)R
symmetry rotating ψα into λα , which is possible only if U (Φ) is linear in Φ. Upon integrating
out the auxiliary field F , a linear U (Φ) would add a constant |∂U |2 to the energy density and
spontaneously break supersymmetry. Therefore, in a theory with supersymmetric vacua, the
superpotential U is a constant that may set to zero without loss of generality.
What this means is that the only potential for the scalar fields φa comes from the Kähler
potential. To proceed, we shall assume for simplicity that the gauge group G is a simple Lie
group, so that there is a single N = 1 vector superfield V transforming under the adjoint
representation of g. For a
renormalizable theory, the Kähler potential must be canonical i.e.
K(eV Φ, Φ† ) = Tr Φ† eV Φ . In that case the auxiliary field D may be integrated out using
its field equations and the resulting scalar potential takes the form,

2
Vgauge (φ, φ̄) = Tr i[φ, φ̄] (15.2.6)

The structure of this potential will be of central importance in disentangling the vacuum
structure of the N = 2 theories.
Next, we include an N = 2 hypermultiplet, which consists of N = 1 chiral superfields H =
(h, ψα ) and H̃ = (h̃, ψ̃α ) in representations R and R̄ of g, respectively. The representation
R ⊕ R̄ may be reducible into a direct sum of irreducible representations RI and R̄I of g,
M X
R= nI RI ⊕ R̄I nI = Nf (15.2.7)
I I

where nI is the number of hypermultiplets in representation RI , and Nf is the total number

of hypermultiplets in all representations RI . We decompose the hypermultiplet accordingly
into hypermultiplets HI and H̃I in the representations RI and R̄I respectively. The scalar
fields h and h̃ in the hypermultiplets produce the following addition to the scalar potential,
XR
dim nI
XX
a
Vhyper (h, h̃, φ) = hi (φa T )ij h̃j + MI,s,t hI,s h̃I,t (15.2.8)
i,j=1 I s,t=1

243
Here s, t are flavor indices running from 1, . . . , nI in each irreducible representation RI ,
while i, j are the indices in the representation R, and φ is the scalar component of the
superfield Φ. The matrix MI,s,t contains parameters that can be interpreted as masses and
mixings of the hypermultiplets in a given representation RI of g. Gauge invariance precludes
mixings between hypermultiplets RI and RJ with I 6= J. Combining this scalar potential
with that in (15.2.6) allows for interesting structure for the moduli space of vacua.

15.3 Low-energy Lagrangian on the Coulomb branch

In a supersymmetric Yang-Mills theory, the energy is bounded from below by zero. In the
sequel we shall assume that the gauge group of the UV theory is a compact semi-simple
Lie group G with Lie algebra g. In particular, the potential energy Vgauge of (15.2.6) is non-
negative since the Cartan-Killing form of g is negative definite. Any supersymmetric vacuum
state has exactly zero energy, which requires vanishing fermion fields, Dµ φ = Fµν = 0 and,
for the solutions to these equations in the gauge Aµ = 0, a constant field φ satisfying,
[φ, φ̄] = 0 (15.3.1)
Decomposing φ = φ1 + iφ2 into Hermitian matrices φ1 , φ2 gives the equivalent condition
[φ1 , φ2 ] = 0 which guarantees that φ1 and φ2 may be simultaneously diagonalized, and gives
the following general solution to (15.3.1) for the vacuum expectation value of the field φ,
r
X
φ= aI hI r = rank(g) (15.3.2)
I=1

where hI for I = 1, · · · , r = rank(g) are the mutually commuting generators of the Cartan
subalgebra of g. For generic values of aI 6= 0 the vacuum is referred to as the Coulomb branch.
The standard Higgs mechanism breaks the gauge symmetry G to the Cartan subgroup U (1)r .
Actually, the Cartan subgroup is invariant under the Weyl group W(g) group which is the
residual symmetry remaining from the non-Abelian part of G. Thus, the precise symmetry
breaking pattern is as follows,
G −→ U (1)r /W(g) (15.3.3)
Therefore, the low energy effective theory consists of r copies of N = 2 electro-magnetism
with compact U (1) gauge groups, up to identification under the Weyl group W(g).
The field contents of the low energy theory consists of N = 2 gauge multiplets with
component fields (AIµ , λIα , λ̃Iα , φI ) for gauge group U (1)r and I = 1, · · · , r which may be
decomposed into N = 1 gauge superfields V I and chiral superfields ΦI . The theory is
governed by the Lagrangian constructed in (15.2.3) for gauge group U (1)r ,
Z Z
4 †
d2 θ τIJ W I W J


L = d θ K Φ, Φ + Re (15.3.4)

244
The field strength superfields W I of the Abelian gauge multiplets simplifies,
1
WαI = − D̄D̄Dα V I (15.3.5)
4
and τIJ (Φ) is holomorphic in the fields ΦI . Since the chiral fields ΦI are neutral under the
gauge group U (1)r the factor eV multiplying Φ in (15.2.3) is absent here. No superpotential
U (ΦI ) is allowed in view of N = 2 supersymmetry, as may be shown by the same argument
used earlier in the case of renormalizable Lagrangians. Expressed in terms of component
fields the Lagrangian takes the form,29

¯ J¯ 
L = −gI J¯(φ, φ̄) Dµ φI Dµ φJ + iλ̃ σ̄ µ Dµ λ̃I
  
i I Jµν 1 I Jµν I µ J
−Re τIJ (φ) F F − Fµν F̃ + λ̄ σ̄ Dµ λ (15.3.6)
2 µν 2

The SU (2)R symmetry of N = 2 rotates λI into λ̃I so that invariance of L requires the
following relation between their coefficients τIJ and the Kähler metric gI J¯.
∂ 2 K(φ, φ̄)
Im τIJ (φ) = gI J¯(φ, φ̄) = (15.3.7)
∂φI ∂φJ¯
Since τIJ (φ) is holomorphic, this relation implies that the Kähler potential K must be of the
special Kähler type, so that τIJ (φ) and K are given by,
∂ 2 F(φ) 1 I ∂F(φ)
φ̄ φDI − φI φ̄DI


τIJ (φ) = K(φ, φ̄) = φDI = (15.3.8)
∂φI ∂φJ 2i ∂φI
where F(φ) is a locally holomorphic function of φI referred to as the pre-potential. Since the
pre-potential F depends only on the field φ and not on any derivatives of φ, its expression is
entirely determined by evaluating F on the vacuum expectation values of φ, given in (15.3.2).

15.4 BPS states, monopoles, and dyons

In addition to the massless states described by the canonical fields of the N = 2 gauge
multiplet, the theory also contains the massive gauge boson states as well as the ‘t Hooft-
Polyakov magnetic monopoles produced by the Higgs mechanism g → u(1)r . These states
are BPS states and carry non-zero electric and/or magnetic charges under the various u(1)
fields. In this subsection we shall review the properties of these states.
29 ¯
We use the standard notation for coordinates on Kähler manifolds where φI = φ̄I so that the metric is
¯
gI J¯ dφI ⊗ dφJ . The Levi-Civita connection Γ for the metric gI J¯ is compatible with the complex structure
so that its mixed components ΓIJ¯K̄ , ΓIJ K̄ and their complex conjugates vanish. The corresponding covariant
derivatives are given by Dµ f I = ∂µ f I + ΓIJK ∂µ φJ f K for the fields f I = φI , λI , λ̃I .

245
 The massive vector bosons, and their superpartners, that arise via the Higgs mechanism
from the spontaneous symmetry breaking g → u(1)r correspond to the root generators
eα in the decomposition30 of the generators of g while the massless gauge bosons of u(1)r
correspond to the Cartan generators hI identified in (15.3.2). For each root α there is a
unique massive vector boson multiplet in the spectrum. Counting the number of states for
each spin, we see that these massive vector bosons must belong to BPS multiplets. As a
result, the BPS formula gives their masses in terms of their central charges,

MαW = |α · a| a = (a1 , · · · , ar ) (15.4.1)

a formula that is a direct consequence of the Higgs mechanism by a scalar φ in the adjoint
representation of the gauge algebra g.
The massive magnetic monopole states, and their superpartners, also arise via the Higgs
mechanism g → u(1)r as ‘t Hooft-Polyakov magnetic monopoles. In the classical limit, they
arise as static solutions of finite mass that saturate the Bogomolnyi bound,
Z 
2  3

I 
M = 2  d x ∇ · tr φ B ) (15.4.2)
g
where BI is the magnetic field component of the field strength Fµν
I
and the integral gives the
I
magnetic charge of the field Fµν . Precisely one magnetic monopole arises for each possible
embedding of su(2) into g, i.e. for each root β of g. This construction will give us a
BPS formula (at the semi-classical level) for the mass of the ‘t Hooft-Polyakov magnetic
monopole in terms of the vacuum expectation value of the dual gauge scalar φDI . Applying
this construction for each SU (2) subgroup of G, namely for each root of g, we find the
semi-classical magnetic monopole BPS mass formula,

MβM = |β · aD | aD = (aD1 , · · · , aDr ) . (15.4.3)

Here, we have also included the effect of the θ angle, which is to produce an electric charge
on the magnetic monopole and make it into a dyon.
More generally, a dyon may be specified by two roots: one root α for its electric charge
as in the construction of vector boson masses, and one (co)root β for its magnetic charge as
in the construction of the magnetic monopole masses. The full quantum BPS mass formula
for such a dyon is,
D
 
Mα,β = α · a + β · aD  (15.4.4)

30
Our notation for Cartan generators and roots for a rank r Lie algebra g in the Cartan basis is as
follows. The Cartan generators will be denoted hI with I = 1, · · · , r, as we have already used in (15.3.2)
when parametrizing the vacua of the Coulomb branch. The root generators are denoted eα , where the
corresponding root vector is an r-dimensional vector with components P αI . Their commutators may now be
expressed as follows, [hI , hJ ] = 0, [hI , eα ] = αI eα , and [eα , e−α ] = I αI hI .

246
The theory further contains an infinite number of states that may be neutral or carry arbi-
trary electric and magnetic charges, but that do not saturate the BPS bound and are thus
non-BPS states. No simple formula for their mass in terms of the other quantum numbers
is known to exist. Henceforth, we restrict attention to the BPS spectrum only.

15.5 SL(2, Z)-duality of the N = 4 theory

For the N = 4 theory, the pre-potential is given by,

1 X I I
F(φ) = τ φ φ (15.5.1)
2 I

so that aDI = τ aI and τIJ = τ δIJ with τ independent of φ. The spectrum of vector boson,
magnetic monopole, and dyon BPS states is then given by,
D
 
Mα,β (τ ) = α · a + τ β · a (15.5.2)

The roots α and β may be decomposed onto a basis of simple roots αI ,

X X
α= nI αI β= mI αI (15.5.3)
I I

where the coefficients mI , nI are integers. The masses of the BPS states are then given by,
  X
Mm,n (τ ) = Z  Z= (nI + mI τ ) αI · a (15.5.4)
I

One of the original indications of the SL(2, Z) Montonen-Olive duality symmetry of the
maximally supersymmetric N = 4 Yang-Mills theory was that the spectrum of BPS states
is invariant under SL(2, Z) transformations of the complexified coupling τ ,

aτ + b a
Mm0 ,n0 (τ 0 ) = Mm,n (τ ) τ → τ0 = a → a0 = (15.5.5)
cτ + d cτ + d
provided the states are mapped as follows,

mI = am0I + cn0I
nI = bm0I + dn0I (15.5.6)

We note, however, that SL(2, Z) does not in general map N = 4 Yang-Mills with gauge
group G to itself. Instead, as we shall now explain, it transforms between theories with the
same Lie algebra g but different global structures.

247
15.5.1 The global structure of N = 4 theories
The Lie algebra g of an N = 4 Yang-Mills theory does not uniquely specify the theory.
Rather there remain different choices for the global structure of the theory, including (but
not limited to) the particular choice of gauge group G corresponding to the algebra g.31 The
various Lie groups associated with a Lie algebra g may be obtained by first identifying the
unique simply-connected group G whose Lie algebra is g, and then identifying the center
Z(G) of G. All of the other gauge groups corresponding to the algebra g are then of the form
GH = G/H for H a subgroup of Z(G). Whenever H is non-trivial, i.e. contains at least
one element other than the identity, the group GH is non-simply connected and a standard
argument shows that its first homotopy group π1 (GH ) is isomorphic to H.
The weight lattice Λw of a Lie algebra g is the lattice spanned by the weights of all
possible finite-dimensional representations of g, and coincides with the weight lattice of the
simply connected Lie group G so that Λw = ΛGw . The weight lattice of a non-simply connected
subgroup GH ⊂ G will be denoted ΛG w . Every representation of GH is a representation of G
H

G/H G
so that Λw ⊂ Λw but the converse is not true, so that the inclusion is a strict one whenever
G/H
H is non-trivial. The root lattice of g is a sub-lattice of Λw for every H ⊂ Z(G).
We shall now consider the N = 4 Yang-Mills theories with gauge algebra g and gauge
group G = GH for the different choices of H. An SL(2, Z) transformation maps the gauge
algebra g to itself, and maps the gauge group GH to a gauge group GH 0 where H 0 is not
necessarily equal to H. This means that the allowed representations of GH and GH 0 will
differ from one another when H 0 6= H and is most easily seen by analyzing the spectrum of
line operators in each theory. These line operators come in the following types:

• Wilson lines: To each center subgroup H and corresponding gauge group GH is

associated a set of Wilson lines, which correspond physically to the worldlines of heavy
probe particles that transform under a representation R of GH . The Wilson line may
be expressed as the holonomy of the gauge field A in the representation R of GH ,
 n I o
WR (γ) = TrR P exp i A (15.5.7)
γ

with γ a 1-cycle in the four-dimensional spacetime. Defined in this way, each Wilson
line is manifestly invariant under H since A is invariant, but the selection of the
representation R of GH depends on H, as explained earlier. Note that since Wilson
lines are worldlines of probe particles, their existence does not depend on the dynamical
matter content of the theory: once the gauge group GH is fixed, Wilson lines will exist
31
Throughout this section, we shall restrict attention to gauge groups that are connected and compact.
An example in which the global structure of the gauge group alone is not enough to fully specify the global
structure is provided by the case of SO(3) super Yang-Mills, which will be discussed below.

248
 for every representation R of GH . In terms of the weight lattice Λw of g, the Wilson
lines correspond to points in ΛG GH
w /W(g), with Λw ⊂ Λw the sub-lattice of weights of
H

GH and W(g) the Weyl group of g.

• ’t Hooft lines: Gauge theories also come equipped with ’t Hooft lines, which are
the magnetic counterparts of the Wilson lines. Unlike Wilson lines, they cannot be
written as local functionals of the fields of the gauge theory, but they are labelled in
an analogous way. In particular, let us denote the magnetic dual (also referred to as
the Langlands or Goddard-Nuyts-Olive (GNO) dual) algebra of g by g∨ . The algebra
g∨ is dual to g in the sense that the weight lattice of g∨ , which we denote by Λ∨w ,
is the dual of the root lattice of g. For simply-laced g we have g = g∨ , whereas for
non-simply-laced algebras one has

b∨r = cr c∨r = br g∨2 = g2 f∨4 = f4 (15.5.8)

in the standard Cartan notation in which br , cr , g2 , f4 are the Lie algebras of the Lie
groups SO(2r + 1), Sp(2r), G2 , F4 , respectively. At the level of groups the correspon-
dence is more subtle, with the results collected in Table 2. Note that,

W(g∨ ) = W(g) Z(G ∨ ) = Z(G) (15.5.9)

The ’t Hooft lines are then labelled by a particular subset of the points in Λ∨w /W(g),
specified by the global structure of the magnetic gauge group G∨ . The way to determine
this subset will be explained below.

• Dyonic lines: Finally, we can consider mixed Wilson and ’t Hooft lines, labelled by
a subset of the electric and magnetic charges,

(qe , qm ) ∈ Λw × Λ∨w (15.5.10)

subject to the identification,

(qe , qm ) ∼ (wqe , wqm ) w ∈ W(g) (15.5.11)

Note that this set can be larger than the set of pairs of representations of g and g∨ ,
which is labelled instead by (Λw /W(g)) × (Λ∨w /W(g)).

We now explain how to determine the subset of allowed ’t Hooft and dyonic lines. Having
fixed a global form GH for the gauge group, we know that the spectrum of Wilson lines is the
set of all lines (qe , 0) with qe ∈ ΛG
w /W(g). The spectrum of ’t Hooft and dyonic lines is then
H

given by the full spectrum of lines which are mutually local to the Wilson lines. By mutual

249
 G G∨
SU (N M )/ZN SU (N M )/ZM
SO(2N ) SO(2N )
Spin(2N ) N odd Spin(2N )/Z4
Spin(2N ) N even SO(2N )/Z2
SO(2N + 1) U Sp(2N )
Spin(2N + 1) U Sp(2N )/Z2
E6 E6 /Z3
E7 E7 /Z2

Table 2: Langlands or GNO dual groups G∨ for various gauge groups G, with (G∨ )∨ = G.
The gauge groups which do not appear in this table, namely G2 , F4 , and E8 , have trivial
center and hence only a single global form, meaning that G∨ = G.

locality, we mean that if one computes the correlation function of two lines supported on γ
and γ 0 , and then moves γ 0 in a loop around γ, the correlation function is unchanged.
To make this more concrete, note that independent of the particular global group GH ,
Wilson lines labelled by roots of g must always be present, since roots are elements of ΛG w
H

for any choice of subgroup H. Likewise, even without solving the mutual locality constraint,
it is clear that we must have at least the ’t Hooft lines labelled by roots of g∨ . Because the
charge lattice is closed under addition and inversion—i.e. if (qe , qm ) is present then so is
(−qe , −qm ), and if furthermore (qe0 , qm
0
) is present then so is (qe + qe0 , qm + qm
0
)—it is useful to
organize lines into families labelled by the weight lattice modulo the root lattice. The weight
lattice modulo the root lattice is nothing but the center, and recalling that Z(G) = Z(G ∨ )
we conclude that we may label families of lines by pairs
(ze , zm ) ∈ Z(G) × Z(G) (15.5.12)
If one element in such a family exists, so do all of the other elements, obtained by adding
arbitrary root vectors.
For charges valued in the center, the mutual locality condition mentioned above takes
a simple form. For Z(G) = ZN , which is the case for all simple Lie groups except for
SO(4N + 2),32 then for lines (ze , zm ) and (ze0 , zm
0
) the mutual locality constraint reads [272],
0
ze zm − zm ze0 = 0 mod N (15.5.13)
We note that even upon fixing the global form of the gauge group GH , and hence the lattice of
allowed electric charges ze , there can be multiple solutions to the mutual locality constraint,
32
In the case of SO(4N + 2) the center is Z2 × Z2 . This case is more complicated and will not be discussed
here; see [272] for details.

250
 qm qm qm

qe qe qe

Figure 16: Lattice of line operators for the SU (2), SO(3)+ , and SO(3)− theories. The red
dots represent occupied lattice points. The grey square represents the sublattice of elements
(ze , zm ) ∈ Z(SU (2)) × Z(SU (2)) valued in the center, which can be used to obtain the entire
lattice by adding arbitrary root vectors.

giving different lattices of dyonic lines. To fully specify the N = 4 theory, one must specify
all of this additional global data.

15.5.2 Example: su(2) SYM

Let us illustrate the above through the example of g = su(2). In this case G = SU (2) and
Z(G) = Z2 . If we take H ⊂ Z(G) to be trivial, then GH = SU (2) itself. Aside from the
identity line (ze , zm ) = (0, 0), the spectrum of Wilson lines is generated by elements of the
non-trivial center, which is just (ze , zm ) = (1, 0). The mutual locality condition requires that
any other family of lines labelled by (ze0 , zm
0 0
) satisfies zm = 0 mod 2, and hence that no other
non-trivial representation of the center is allowed. The lattice of charges for GH = SU (2) is
shown in the gray square in the leftmost part of Figure 16—the full lattice is obtained from
this by adding elements of the root lattice.
We can instead choose H = Z2 , in which case GH = SU (2)/Z2 = SO(3). The Wilson lines
are now labelled by representations of SO(3), which have even electric charges, and hence
which, when reduced to the center, are all trivial (ze , zm ) = (0, 0). Because of this, demanding
mutual locality with Wilson lines does not impose any constraints on the spectrum of ’t
Hooft and dyonic lines operators, and we may choose to have either (ze , zm ) = (0, 1) or (1, 1)
occupied (but not both, since the two are not mutually local). This gives the middle and
right lattices in Figure 16. We note here that even upon fixing the gauge group GH = SO(3),
there are still two distinct choices for the spectrum of line operators, and one must specify
the full lattice of lines before one completely specifies the theory. The two theories in this
case are typically denoted by SO(3)+ and SO(3)− .
Let us point out here that the only ingredient entering in the above analysis was the
center of G. We may thus conclude that the case of g = e7 , for which the center is again Z2 ,

251
is exactly identical, again with three distinct global variants. For groups with larger centers
the analysis is more involved, but conceptually straightforward. On the other hand, when
the center is trivial as for g = e8 , there is only a single variant of the theory.

15.5.3 SL(2, Z) on line operators

Having understood the spectrum of line operators, we may now return to the question of
the action of SL(2, Z). The S and T transformations act in a straightforward way on the
spectrum of line operators,

S : (qe , qm ) → (qm , qe ) T : (qe , qm ) → (qe + qm , qm ) (15.5.14)

The first of these is the famous result that the S transformation implements electro-magnetic
duality or so-called Montonen-Olive duality on the theory, while the second is the effect by
which magnetic monopoles become dyonic upon shift of the theta angle.
In the context of the example of g = su(2), examining the action of S and T on the
charge lattices in Figure 16 reveals that

S: SU (2) → SO(3)+ T : SU (2) → SU (2)

SO(3)+ → SU (2) SO(3)+ → SO(3)−
SO(3)− → SO(3)− SO(3)− → SO(3)+ (15.5.15)

Thus we see that no individual theory is actually self-dual under SL(2, Z). Instead, one can
think of the set of three theories as transforming as a three-dimensional vector-valued mod-
ular function of SL(2, Z). Alternatively, each individual theory can be said to be invariant
under an appropriate congruence subgroup. For example, for GH = SU (2) the theory is
invariant under T and ST 2 S, which together generate the subgroup Γ0 (2) ⊂ SL(2, Z). Thus
the SU (2) theory is self-dual under Γ0 (2).
Let us close by mentioning that, in the cases in which there is only a single global variant
of the theory, e.g. for g = e8 mentioned above, then the full SL(2, Z) really does act as a
duality group for the theory.

15.6 The Seiberg-Witten Solution

The Seiberg-Witten solution for an N = 2 supersymmetric Yang-Mills theory in four space-
time dimensions with gauge group G of rank r, is given in terms of the vacuum expectation
values of the complex scalars aI of the gauge multiplet and their magnetic duals aDI in
the Coulomb phase. When hypermultiplets are present in a representation R ⊕ R̄ of the
gauge group G, we shall denote their G-invariant masses by mf where the index f runs over
the various irreducible components Rf of R. The prepotential F and the matrix τIJ for

252
I, J = 1, · · · , r of effective couplings and mixings of the unbroken U (1)r gauge multiplets
are related to these expectation values by,

∂F ∂aDI ∂ 2F
aDI = τIJ = = (15.6.1)
∂aI ∂aJ ∂aI ∂aJ

The original Seiberg-Witten solution was obtained for gauge group SU (2), but the results
were quickly generalized to other semi-simple compact gauge groups G.
The building blocks of the Seiberg-Witten solution are a curve C and a differential λ
on C. More precisely, C = C(u, m) is a family of genus r Riemann surfaces parametrized by
G-invariant local complex coordinates u1 , · · · ur and hypermultiplet masses mf . A convenient
definition of the uI may be given in terms of the gauge-invariant traces tr(φn ) of the gauge
scalar φ. The differential λ is meromorphic on C and its partial derivatives with respect to
the parameters uI are holomorphic Abelian differentials, possibly up to the addition of exact
differentials of meromorphic functions on C. In the presence of hypermultiplets, λ has poles
whose residues are given by the masses mf .
The Seiberg-Witten solution gives the vacuum expectation values of the gauge scalars
aI and their magnetic duals aDI in terms of the periods of the differential λ on a canonical
basis of homology cycles AI and BI of the curve C,
I I
2πiaI = λ 2πiaDI = λ (15.6.2)
AI BI

By the very construction of this set-up, the matrix τIJ has positive imaginary part as required
on physical grounds by the fact that it provides the couplings for the U (1)r gauge field
I
strengths Fµν , which must be positive. To see this, we combine the following formulas,
X ∂aDI ∂uK I I
∂aI ∂λ ∂aDI ∂λ
τIJ = 2πi = 2πi = (15.6.3)
K
∂uK ∂aJ ∂uK AI ∂uK ∂uK BI ∂uK

to observe that τIJ is the period matrix of the curve C(u), whose imaginary part is positive
by the Riemann bilinear relations; see appendix B.1.5.
To illustrate the general construction outlined above, we present the Seiberg-Witten
curves and differentials for the gauge group G = SU (N ) in two examples. The first has Nf <
2N hypermultiplets with masses mf in the defining representation of G and is asymptotically
free. The second has one hypermultiplet with mass m in the adjoint representation of G
and is referred to as the N = 2∗ theory. The latter is UV finite and reduces to the N = 4
super Yang-Mills theory upon setting the mass m of the adjoint hypermultiplet to zero, and
to the pure N = 2 theory without hypermultiplets by sending the mass m to infinity while
suitably scaling the gauge-coupling to zero.

253
15.6.1 The Nf < 2N theory for gauge group SU (N )
For the theory with Nf < 2N hypermultiplets in the fundamental representation of SU (N )
with masses mf , the differential λ and the curve C are given by,

B0
 
0 1 xdx
λ= A − (A − y) y 2 = A(x)2 − B(x) (15.6.4)
2 B y

where the polynomials A and B are given as follows,

N Nf
Y Y
2N −Nf
A(x) = (x − āi ) B(x) = Λ (x − mf ) (15.6.5)
i=1 f =1

and Λ is the renormalization scale of the asymptotically free theory. The gauge-invariant
coordinates uI are given in terms of symmetric
P polynomials in the variables āi which, since
the gauge group is SU (N ), must satisfy i āi = 0. The curve C is hyperelliptic of genus N −1.
The differential λ has simple poles with residue mf at the points (x, y) = (mf , −A(mf )) on
the second sheet of the hyper-elliptic curve, but is regular at the points (x, y) = (mf , A(mf ))
on the first sheet of C.

15.7 The N = 2∗ theory for gauge group SU (N )

The N = 2∗ theory possesses a rich modular structure, in part because it may be obtained as
a mass deformation from the N = 4 theory, shares the same ultraviolet finiteness properties,
and has a marginal deformation controlled by the complex gauge coupling τ . Its Seiberg-
Witten theory may be developed in terms of integrable mechanical systems, of the Hitchin
or elliptic Calogero-Moser type. It is the latter approach that we shall follow here.

15.7.1 The Seiberg-Witten solution

To formulate the Seiberg-Witten solution for the N = 2∗ theory for gauge group SU (N ), we
introduce a torus C/Λ for a lattice Λ = ω1 Z ⊕ ω2 Z where the modulus τ = ω2 /ω1 is related
to the gauge coupling g and the θ angle of the non-Abelian gauge theory by,

θ 4πi
τ= + 2 (15.7.1)
2π g

Denoting the mass of the adjoint hypermultiplet by m, the differential λ and the curve C are
constructed in terms of a local complex coordinate z on C/Λ,

λ = k dz RN (k, z|Λ) = 0 (15.7.2)

254
The function RN (k, z|Λ) is given by the characteristic polynomial for the Lax operator of
the elliptic Calogero-Moser system for the root system aN −1 of the Lie algebra su(N ),


RN (k, z|Λ) = det kI − L(z|Λ) (15.7.3)

The components of the matrix L(z|Λ) and its conjugate Lax matrix M (z|Λ) are given by,

Lij (z|Λ) = pi δij − m(1 − δij )Φ(xi − xj , z|Λ)

X
Mij (z|Λ) = m(1 − δij )Φ0 (xi − xj , z|Λ) + mδij ℘(xi − xk |Λ) (15.7.4)
k6=i

where ℘ is the Weierstrass function and Φ is the Lamé function,

σ(z − x|Λ) xζ(z|Λ)

Φ(x, z|Λ) = e (15.7.5)
σ(z|Λ)σ(x|Λ)

given in terms of the Weierstrass ζ- and σ-functions for the lattice Λ; see subsection 2.4.3
for definitions of ζ and σ. The variables xi and pi are the positions and momenta for the
Calogero-Moser system. The integrability condition L̇ = [M, L] guarantees that RN (k, z) is
a conserved quantity for all values of z, k, and Λ.

15.7.2 The Seiberg-Witten curve

The curve R(k, z|Λ) admits an explicit presentation directly in terms of ϑ-functions,
  −1    
z  z ∂  
RN (k, z|Λ) = ϑ1 τ ϑ1 − m τ P (k̃)  (15.7.6)
ω1  ω1 ∂ k̃  
k̃=k+mh1 (z|Λ)

Here P (k) is a monic polynomial in k of degree N with vanishing k N −1 term, reflecting the
fact that the gauge group is SU (N ) as opposed to U (N ),

N
X −2
P (k) = k N + un k n (15.7.7)
n=0

The N − 1 remaining independent coefficients un are the moduli of the Coulomb branch.
The functions hn are defined by,
  −1 n   
z  ∂ z 
hn (z|Λ) = ϑ1 τ ϑ1 τ (15.7.8)
ω1  ∂z n ω1 

255
One may think of the roots of the polynomial as the vacuum expectation values of the gauge
scalars aI when m = 0. The corresponding curves for N = 2, 3, 4 are given by,33
R2 (k, z|Λ) = k 2 + u0 − m2 ℘ (15.7.9)
R3 (k, z|Λ) = k 3 + (u1 − 3m2 ℘)k + u0 + m3 ℘0
R4 (k, z|Λ) = k 4 + (u2 − 6m2 ℘)k 2 + (u1 + 4m3 ℘0 )k + u0 − m4 ℘00 + 3m4 ℘2
The SW curve may thus be constructed from N copies, or sheets, of the underlying torus
C/Λ, namely one for each solution kI , glued together wherever two sheets I 6= J intersect
one another. A schematic representation of the underlying torus C/Λ and the curve C(u)
for gauge group SU (3) is presented in Figure 17.
A canonical choice for A, B cycles on the torus C/Λ may be uplifted to a canonical
homology basis of cycles AI , BI for I = 1, · · · , N − 1 on the sheet corresponding to the
solution kI , as illustrated for N = 3 in Figure 17. The modular group SL(2, Z) acts on the
cycles A, B by,
   0     
B B a b B a b
→ = ∈ SL(2, Z) (15.7.10)
A A0 c d A c d
and this transformation may be lifted to a transformation on the homology cycles AI and
BI on the curves C(u) by a subgroup SL(2, Z) of the full modular group of the curve
Sp(2N − 2, Z). However, independent of this induced action of SL(2, Z), we may act on the
cycles AI , BI of the curve C(u) by the full modular group Sp(2N − 2, Z). Thus, the action of
SL(2, Z) on A, B and the action of Sp(2N − 2, Z) on AI , BI may be viewed as independent
of one another, producing a larger modular structure governed by the group,
SL(2, Z) n Sp(2N − 2, Z) (15.7.11)
The corresponding construction gives rise to the notion of bi-modular forms, and may be
extended to the superconformal N = 2 theory with gauge group SU (N ) and 2N hypermul-
tiplets in the fundamental representation of SU (N ).

15.7.3 Gauge scalar vacuum expectation values and pre-potential

In terms of these data, the expectation values aI and their magnetic duals aDI are given by,
I I I I
2πiaI = k dz = kI dz 2πiaDI = k dz = kI dz (15.7.12)
AI A BI B

33
The Weierstrass function ℘(z|τ ) differs from h2 (z|Λ) − h1 (z|Λ)2 by a z-independent term given by
℘(z|Λ) = −∂z2 ln ϑ1 (z/ω1 |τ ) − E2 (τ )/(3ω12 ), as may be established using (4.1.7). In the final expressions given
in (15.7.9) we have used the freedom to redefine the moduli parameters un by Λ-dependent but z-independent
shifts, such as for example u0 → u0 − m2 E2 /(3ω12 ) for N = 2. These shifts are immaterial as the un provide
an arbitrary coordinate system for the moduli of the Coulomb branch. The resulting shifted versions of
(15.7.9) have the advantage of enjoying transparent modular transformation properties.

256
 B3

A3
B2 C(u)

A2
B1

A1

B C/Λ
A

Figure 17: Schematic representation of the Seiberg-Witten curve C(u) and the underlying
torus C/Λ for gauge group SU (3). Canonical homology cycles A1 , A2 , B1 , B2 on C(u) may
be provided by uplifting the cycles A and B of a canonical homology basis on C/Λ. The
identifications between the different sheets are not shown here.

The pre-potential F satisfies the following renormalization-type differential equation in τ ,

 N −1 I I N −1
∂F  1 X 2 1 X
= dz k = dz kI2 (15.7.13)
∂τ m,aI 4πi I=1 AI 4πi A I=1

The partial derivative is considered at fixed m and aI . Note that ∂F/∂τ receives no pertur-
bative contributions, as those are independent of τ . The advantage of this formula is that we
can now evaluate aI and ∂F/∂τ entirely in terms of integrals over A-periods without having
to carry out the rather complicated analysis of turning points and regularization required
when evaluating the B-periods.
In the limit m → 0, we recover the N = 4 theory and we have R(k, z|τ ) = P (k), so that
the roots of P (k) correspond to the vacuum
H expectation values of the gauge scalar of the
N = 2 gauge multiplet. Since we have A dz = ω1 by definition of these periods, we also
have 2πiaI = ω1 kI in the m → 0 limits, and thus we set ω1 = 2πi.

15.8 The N = 2∗ theory for gauge group SU (2)

It is instructive to analyze the action of the duality group SL(2, Z) for the case of gauge
group SU (2) where many formulas and transformation laws can be made completely explicit.

257
The curve C of (15.7.9) and differential λ may be parametrized as follows,
p
k 2 = κ2 + m2 ℘(z|Λ) λ = κ2 + m2 ℘(z|Λ) dz (15.8.1)
where we have set u0 = −κ2 for later convenience. As z → 0, we have k ∼ ±m/z which
produces the pole in the differential λ whose residue is the hypermultiplet mass, as expected
on the basis of the general construction outlined above. To evaluate the periods a and aD ,
we begin by evaluating the following elliptic integrals, using the relation ℘(z|Λ) = −ζ 0 (z|Λ)
and the monodromy relations of ζ(z|Λ) given in (2.4.17),
2 ζ( 21 |τ )
I
1 1
A1 = dz ℘(z|Λ) = − 2
= E2 (τ )
2πi A (2πi) 12
2 ζ( τ2 |τ )
I
1 τ 1
B1 = dz ℘(z|Λ) = − 2
= E2 (τ ) + (15.8.2)
2πi B (2πi) 12 2πi
where the last equality was obtained by combining the formulas (2.6.21), (3.9.2), (3.9.7), and
(4.1.6). This immediately allows us to compute the right side of the RG equation (15.7.13),
m2
 I I
∂F  1 2 1 
2 2
 1
2
= dzk+ = dz κ + m ℘(z|Λ) = κ + E2 (15.8.3)
∂τ 
m,a 4πi A 4πi A 2 24

15.8.1 Expansion in powers of the mass m

To obtain the periods a and aD , we expand the SW differential λ for SU (2) in powers of m,
∞ n
Γ(n − 21 ) m2
X 
λ=κ 1 − 2 ℘(z|Λ)n dz (15.8.4)
n=0
Γ(− 2
)n! κ

This expansion will be uniformly convergent for small enough m provided z is kept away
from 0, which can always be achieved when we integrate over suitably chosen A and B cycles.
To evaluate the periods a and aD , we compute the integrals,
I I
1 n 1
An = dz ℘(z|Λ) Bn = dz ℘(z|Λ)n (15.8.5)
2πi A 2πi B
Clearly, we have A0 = 1 and B0 = τ while A1 and B1 were already evaluated in (15.8.2).
For n ≥ 2, we make use of the following recursion relation,

0
(8n − 4)℘n = (2n − 3)g2 ℘n−2 + (2n − 4)g3 ℘n−3 + 2 ℘0 ℘n−2 (15.8.6)
which is readily established using the defining equation (℘0 )2 = 4℘3 − g2 ℘ − g3 . Eliminating
g2 , g3 in favor of E4 , E6 for the period normalization ω1 = 2πi,
4π 4 E4 E4 8π 6 E6 E6
g2 = 4
= g3 = 6
=− (15.8.7)
3ω1 12 27ω1 216

258
and integrating the relation (15.8.6), we obtain the following recursion relation, satisfied by
both An and Bn with n ≥ 2,

2n − 3 n−2
(8n − 4)An = E4 An−2 − E6 An−3 (15.8.8)
12 108
The coefficients of this recursion relation are modular forms, but the initial conditions for
A1 and B1 involve the quasi-modular form E2 . Linearity of (15.8.8) guarantees that the
solutions An , Bn are linear in E2 and may be cast in the following form,

E2 Ln−1
An = Kn + Ln−1 Bn = τ An + (15.8.9)
12 2πi
where Kn and Ln are modular forms of weight 2n. The function An satisfies (15.8.8) given
for An with initial conditions A0 and A1 , while Ln satisfies (15.8.8) for Bn − τ An with initial
conditions B0 − τ A0 = 0 and B1 − τ A1 = (2πi)−1 . Substituting these expressions into the
expansion in powers of m of the period integrals, we obtain the a-period,
∞ n 
Γ(n − 21 ) m2
 
X E2
a=κ 1 − 2 Kn + Ln−1 (15.8.10)
n=0
Γ(− 2
)n! κ 12

and the following result for the combination aD − τ a of the aD and a periods,
∞ n
κ X Γ(n − 12 ) m2

aD − τ a = − 2 Ln−1 (15.8.11)
2πi n=0 Γ(− 12 )n! κ

We shall establish below that the combination aD −τ a of the periods transforms as a modular
form in the modulus τ .

15.8.2 Low orders and perturbative contribution

The Eisenstein series E2n (τ ) for n ≥ 1 are normalized to 1 at the cusp τ → i∞ and admit a
series expansion at the cusp in powers of q = e2πiτ . The following coefficients vanish,

K1 = L−1 = L1 = 0 (15.8.12)

while the non-vanishing low order coefficients are given by,

E4 E6 5 E24 E4 E6
K0 = 1 K2 = K3 = − K4 = K5 = −
144 2160 48384 77760
E4 E6 7 E24 29 E4 E6
L0 = 1 L2 = L3 = − L4 = L5 = − (15.8.13)
80 1512 34560 1330560

259
To solve for the prepotential, we may either integrate the RG equation, or integrate aD =
∂F/∂a. In either case, we need to express κ as a function of τ, m, a, which can be done order
by order in m by inverting the series given in (15.8.10) for a as a function of τ, m, κ,
"
E2 m2 E4 − 2E22 m4 20E32 − 11E2 E4 − 4E6 m6

8
κ=a 1− + − + O(m ) (15.8.14)
6 (2a)2 72 (2a)4 216 (2a)6

The expression for F derived from the RG equation is given by,

1 E2 m4 E22 m6
   
1 2 E4
F(τ, m, a) = τ a + F0 (m, a) − + + + O(m8 )(15.8.15)
2 2πi 24 (2a)2 720 144 (2a)4
where F0 is independent of τ and undetermined by the RG equation. However, it may be
determined from the expressions for aD and a that we have already obtained, and we have,
m2 a
F0 (m, a) = cm2 + ln (15.8.16)
4πi m
In the weak coupling limit, where q → 0 and E2 , E4 , E6 → 1 we evaluate the effective gauge
coupling τIR as follows,
∞ n
1 X 1 m2 m2
  
1
τIR = τ − + O(q) = τ + ln 1 − 2 + O(q) (15.8.17)
πi n=1 n 4a2 πi 4a

Thus, the perturbative logarithm at the mass of the massive vector boson m2 = 4a2 is the
result of re-summing an infinite series of genuine modular form contributions. Note that the
logarithmic term produced by F0 is required to obtain the n = 1 term needed to complete
the series for the logarithm.

15.8.3 Modular properties

Exceptionally, in this subsection only, we shall parametrize SL(2, Z) transformations with
Greek letters in order to avoid a clash with the notation for the periods a and aD . Under
modular modular transformations,
      
ατ + β ω̃2 α β ω2 α β
τ̃ = = ∈ SL(2, Z) (15.8.18)
γτ + δ ω̃1 γ δ ω1 γ δ
we have the following transformations of Eisenstein series for n ≥ 1,
12
E2n (τ̃ ) = (γτ + δ)2n E2n (τ ) + γ(γτ + δ) δn,1 (15.8.19)
2πi
and the Weierstrass function, in which we keep the period ω1 = 2πi fixed,
z
℘(z̃|τ̃ ) = (γτ + δ)2 ℘(z|τ ) z̃ = (15.8.20)
γτ + δ

260
The transformation laws of the SW data may be deduced as follows. First of all, modular
invariance of the SW differential implies the transformation law for k 2 ,

λ = k dz = k̃ dz̃ k̃ 2 = (γτ + δ)2 k 2 (15.8.21)

Using the relation between z, k, and κ, we readily obtain,

κ̃ = (γτ + δ)κ (15.8.22)

Here, the sign ambiguity introduced by taking the square root of κ2 amounts to implementing
the transformation −I ∈ SL(2, Z) on κ. The transformation laws on aD , a may be deduced
from their explicit expressions in terms of κ and τ . We begin by observing in (15.8.11) that
the combination aD − τ a transforms as a modular form of weight (−1, 0),

ãD − τ̃ ã = (γτ + δ)−1 (aD − τ a) (15.8.23)

Next, we use the explicit expression for a from (15.8.10) along with the transformation laws
of κ, K2n , and L2n−2 as modular forms of weight (1, 0), (2n, 0), and (2n − 2, 0) respectively,
as well as the transformation law of E2 , and we obtain,

ã = (γτ + δ)a + γ(aD − τ a) = γ aD + δ a (15.8.24)

where the second term in the middle equality arises from the inhomogeneous transformation
term of E2 . Assembling both transformation laws we obtain,
    
ãD α β aD
= (15.8.25)
ã γ δ a

which is precisely the transformation law induced by the action of the modular group on the
cycles A, B for a modular invariant SW differential.

15.9 Linear quiver chains from a limit of N = 2∗

We now return to the N = 2∗ theory for the gauge group SU (N ), whose independent free
parameters are the UV gauge coupling τ , the hypermultiplet mass m, and the vacuum ex-
pectation values aI of the gauge scalars, or equivalently the roots of the polynomial P (k).
The masses of the gauge boson BPS states are |aI − aJ | while the masses of the BPS hyper-
multiplet states are |aI − aJ − m|. Several interesting limits may be taken in this set-up.
A first interesting limit is obtained by keeping the vacuum expectation values aI fixed
while sending the hypermultiplet mass m → ∞ and letting τ → i∞ at the rate dictated
by the one-loop renormalization group β-function with renormalization scale Λ (not to be
confused with the lattice Λ), Λ2N = (−)N m2N q where q = e2πiτ . This gives the pure N = 2
theory with gauge group SU (N ).

261
 SU (N1 ) SU (N2 ) SU (N3 ) SU (N4 ) SU (N5 )
• • • • •
v1 v2 v3 v4 v5 k

Figure 18: The distribution of vacuum expectation values for a breaking from SU (N ) to
SU (N1 ) × · · · × SU (N5 ) with N = N1 + · · · + N5 where the vp are separated from their nearest
neighbors by ±m plus a term Λp that is held fixed as m, vp → ∞. The green rectangles
schematically represent the ranges in which the xpip can vary.

A second interesting limit is obtained by letting m → ∞ and τ → ∞, but now also

sending some of the vacuum expectation values aI to infinity. Upon carefully tuning the
correlations between these limits, one can send some of the hypermultiplet masses and some
of the gauge boson masses to ∞. This limit will break the gauge symmetry to SU (N1 )×· · ·×
SU (NP ) with N = N1 + · · · + NP and Np ≥ 2 for all p = 1, · · · , P , while any U (1) factors
in the gauge group that arise may be shown to decouple. To obtain the above breaking
pattern, we arrange the classical values for the expectation values in a linear chain. We
partition all the order parameters kI for I = 1, · · · , N into P groups with I = i1 = 1, · · · , N1
labeling the indices in the first group, I = N1 + i2 with i2 = 1, · · · , N2 those in the second
group; I = N1 + · · · + Np−1 + ip with ip = 1, · · · , Np those in group p, and so on until
I = N1 + · · · NP −1 + iP with iP = 1, · · · , NP . We now set,
kN1 +···+Np +ip = vp + xpip p = 1, · · · , P (15.9.1)
The decomposition is not unique as it stands since vp and all xpip may be shifted by an
ip -independent amount such that the corresponding kI is unchanged. To make the decom-
position unique, we impose the following further conditions,
P Np
X X
Np vp = 0 xpip = 0 (15.9.2)
p=1 ip =1

We then order the values of vp such that,

vp+1 − vp = m + Λp (15.9.3)
and then send vp , m → ∞ along with τ → i∞ while keeping all Λp and xpip fixed. The result
is a linear quiver chain, depicted in Figure 18. The gauge group is SU (N1 ) × · · · × SU (NP ),
and the theory contains hypermultiplets in bi-fundamental representations,
P −1 n
M o
(Np , Np+1 ) ⊕ (Np , Np+1 ) (15.9.4)
p=1

Depending on the values of Np , hypermultiplets in the fundamental of SU (N ) may also arise.

262
15.10 N = 2 dualities
In the remainder of this section we discuss a set of dualities between N = 2 theories which
makes interesting contact with the moduli spaces of Riemann surfaces. The theories in ques-
tion will be superconformal, i.e. theories with both supersymmetry and conformal symmetry.
We begin by briefly reviewing these notions.

15.10.1 Conformal symmetry

The conformal algebra in four dimensions is given by so(4, 2), generated by Lorentz transfor-
mations Lµν , translations Pµ , dilations D, and special conformal transformations Kµ . These
generators satisfy the following commutation relations,
[Lµν , Lρσ ] = ηνρ Lµσ + ηµσ Lνρ − ηµρ Lνσ − ηνσ Lµρ [Kµ , Pν ] = 2(ηµν D − Lµν )
[Lµν , Pρ ] = ηνρ Pµ − ηµρ Pν [D, Pµ ] = Pµ
[Lµν , Kρ ] = ηνρ Kµ − ηµρ Kν [D, Kµ ] = −Kµ (15.10.1)
while [Pµ , Pν ] = [Kµ , Kν ] = [D, Lµν ] = 0. Dilations D generate an Abelian subalgebra
so(1, 1) ⊂ so(4, 2) that commutes with the Lorentz subalgebra so(1, 3) generated by Lµν .
Therefore, we may choose the Cartan subalgebra of so(4, 2) to be generated by D and by the
Cartan subalgebra of so(1, 3). The dilation generator can be used to assign weights to the
other generators. The eigenvalue of D is customarily denoted by ∆ and referred to as the
conformal weight. The final two commutation relations are the statement that the momenta
Pµ and special conformal generators Kµ have conformal weights 1 and −1, respectively.
Furthermore, they imply that Pµ is not in the Cartan subalgebra, meaning that operators
or states in a conformal field theory cannot be labelled by an energy/mass—indeed, such
quantities would introduce a dimensionful scale to the theory and thereby violate conformal
symmetry.
Conformal fields have already been introduced in section 5.1.1, where they were labelled
by weights (h, h), related to the conformal weight ∆ via ∆ = h + h. An important difference
between the discussion here and in section 5.1.1 is that in the latter case the discussion
was specific to two spacetime dimensions, for which the “global” conformal group discussed
in this section is enhanced to (two copies of) an infinite-dimensional Virasoro symmetry,
given in (5.1.3). In higher dimensions such an enhancement does not occur, though the
notion of conformal primary introduced in section 5.1.1 persists. In the current language, a
state |[j1 , j2 ], ∆i labelled by its conformal weight ∆ and representation of the Lorentz group
so(3, 1) ∼
= su(2) ⊕ su(2) is a conformal primary if it satisfies,
Kµ |[j1 , j2 ], ∆i = 0 (15.10.2)
Starting from this highest weight state, a series of conformal descendants can be obtained
via repeated application of Pµ .

263
15.10.2 Superconformal symmetry
When N supercharges transforming in the Weyl spinor representation of so(3, 1) are added
to the conformal algebra so(4, 2), one obtains the superalgebra su(2, 2|N ). In the current
section we will be mainly interested in the case of N = 2, for which the maximal bosonic
subalgebra of the superconformal algebra is,

so(4, 2) ⊕ u(2)R ⊂ su(2, 2|2) (15.10.3)

The u(2)R factor is referred to as the R-symmetry, and its generators will be denoted by RJI .
They satisfy the standard commutation relations,

[RJI , RLK ] = δJK RLI − δLI RJK (15.10.4)

As for the fermionic generators, in addition to the Poincaré supercharges QIα , Qα̇I , we now
I
have the superconformal partners SαI , S α̇ . The anti-commutation relations involving these
generators take the form,
I
{QIα , QJβ } = {SαI , SβJ } = {QIα , S β̇ } = 0
{QIα , Qβ̇J } = 2σαµβ̇ Pµ δJI
J
{SαI , S β̇ } = 2σαµβ̇ Kµ δIJ
1 µν
{QIα , SβJ } = αβ (δJI D + RJI ) + δJI Lµν σαβ (15.10.5)
2
Note that unlike in the non-conformal case, the superconformal algebra does not admit a
deformation by central charges Z, since these would introduce a scale and thus would break
conformal symmetry.
The states of an N = 2 superconformal field theory can be labelled by their conformal
weight ∆ and Lorentz quantum numbers [j1 , j2 ], as well as additional labels (R, r) for the
u(2)R ∼
= su(2)R × u(1)r R-symmetry. We will denote such states by |[j1 , j2 ], ∆i(R,r) . In
analogy to a conformal primary, one defines a superconformal primary via
I
SαI |[j1 , j2 ], ∆i(R,r) = S α̇ |[j1 , j2 ], ∆i(R,r) = 0 (15.10.6)

Note that these two conditions, together with the anti-commutation relation in (15.10.5),
imply that all superconformal primaries satisfy Kµ |[j1 , j2 ], ∆i(R,r) = 0, and hence are partic-
ular cases of conformal primaries in the definition of (15.10.2). A superconformal multiplet is
obtained by beginning with a superconformal primary and applying arbitrary combinations
of QIα , Qα̇I . By the anti-commutation relations in (15.10.5), this also includes all applications
of the momentum operator Pµ , and hence includes all conformal descendants.

264
15.10.3 Superconformal field theories
We may construct N = 2 superconformal Lagrangians by combining the N = 2 supermulti-
plets discussed in subsection 15.1.2 in such a way as to cancel any potential non-conformality.
Setting all masses for the hypermultiplets to zero, the only parameter left in an N = 2 La-
grangian is the coupling τ , which is classically dimensionless. Thus any such Lagrangian
would naively seem to have conformal invariance. While this is true at the classical level,
quantum mechanically the coupling τ —or rather only g, since θ is topologically protected—
can receive radiative corrections making it dimensionful. The dependence of the quantum-
corrected coupling on the energy scale µ is captured by the β-function,
d 8π 2
 
β(g) = µ (15.10.7)
dµ g 2
which, for a generic four-dimensional Yang-Mills theory with Nf Weyl fermions in a repre-
sentation Rf and Ns scalars in a representation Rs , is given by,
11 2 1
β(g) = C(g) − Nf C(Rf ) − Ns C(Rs ) + O(g) (15.10.8)
3 3 3
Here C(R) is the Dynkin index of the representation R, defined via TrR (T a T b ) = C(R)δ ab
with T a the generators of the gauge algebra. By a slight abuse of notation, we shall denote
the adjoint representation of the algebra g by the same symbol g. In particular, C(g) = h∨
is the dual Coxeter number.
For an N = 1 theory the vector multiplet, which contains one vector and one Weyl
fermion in the adjoint representation, contributes a factor of,
11 2
C(g) − C(g) = 3 h∨ (15.10.9)
3 3
to the β-function. The contribution for a chiral multiplet, containing a Weyl fermion and a
scalar both in the representation R, is given by,
2 1
− C(R) − C(R) = −C(R) (15.10.10)
3 3
For example, an N = 1 super-Yang-Mills theory with gauge group SU (Nc ) (for which
h∨ = Nc ) with Nf chiral multiplets in the representation R has the following β-function,
β(g) = 3Nc − Nf C(R) + O(g) (15.10.11)
For conformality, we see that we must require Nf C(R) = 3Nc .
We now move on to the main case of interest to us, namely N = 2 theories. The
contribution to the β-function of an N = 2 vector multiplet is,
11 2 1
C(g) − 2 × C(g) − C(g) = 2 h∨ (15.10.12)
3 3 3
265
while the contribution from a hypermultiplet in the representation R is,
2 1
−2 × C(R) − 2 × C(R) = −2 C(R) (15.10.13)
3 3
We thus see that for an N = 2 SU (Nc ) gauge theory with Nf hypermultiplets in the
representation R,
β(g) = 2Nc − 2Nf C(R) + O(g) (15.10.14)
Taking in particular the case in which all hypermultiplets are in the fundamental represen-
tation for which C() = 12 , we see that conformality can be achieved only if Nf = 2Nc .
The theories that we will study in the rest of this section will be exactly of this type,
namely N = 2 theories with SU (Nc ) gauge group and Nf fundamental hypermultiplets. The
data of such a theory can be captured in a so-called “quiver diagram”—a diagram whose
internal nodes are gauge groups, whose external nodes are fundamental hypermultiplets, and
whose internal edges are bifundamental hypermultiplets, with the structure of the quiver
chosen to be compatible with the vanishing of all beta functions.34 For example, (N ) − [2N ]
represents SU (N ) gauge theory with 2N fundamentals, and has a manifestly vanishing beta
function. Likewise the quiver
[N ] − (N ) − (N ) − [N ] (15.10.15)
represents SU (N )×SU (N ) gauge theory with one bifundamental hypermultiplet correspond-
ing to the internal edge, and with N additional fundamentals for each copy of SU (N ). Since
the internal edge can be interpreted as giving N fundamental hypers to the left gauge node
and N anti-fundamental hypers to the right gauge node, the beta functions for both gauge
couplings indeed vanish.
We will focus on the case of N = 2, for which something rather special happens. In
particular, note that in this case each bifundamental has two SU (2) gauges indices and
one SU (2) flavor index, i.e. we can write them as Q̂ijs with indices as defined above. We
can imagine weakly gauging the flavor symmetry, or conversely ungauging one of the gauge
symmetries, which makes it clear that all three indices should be treated democratically. In
other words, it is useful to think of the elementary building blocks as trifundamental matter,
which we will represent pictorially by
s
Q̂ijs ⇒
i j

34
One could ask why we allow for only bifundamental edges in the graph. It turns out that for more
complicated edges, it is impossible to satisfy both anomaly cancellation and vanishing of the beta functions.
The only exception is the case of trifundamentals in theories with SU (2) gauge nodes, which we will return
to below.

266
With this in mind, it is possible to replace any SU (2) quiver diagram with a trivalent network,
for example

[2] − (2) − [2] ⇔ (2) = (2) ⇔

[2] − (2) − (2) − [2] ⇔

Conversely, any trivalent graphs can be given an interpretation as a SCFT with SU (2) nodes.

15.10.4 SU (2) quivers and Riemann surfaces

Consider a trivalent graph Γ with g faces and n external legs. We will denote the theories
obtained from such graphs by Tg,n . There is one gauge SU (2) per internal edge, and it can
be shown that the number of such internal edges is 3g − 3 + n. To each of these internal
nodes is associated a gauge coupling τ . The quantity 3g − 3 + n appearing here may seem
familiar—indeed, if n = 0 then this precisely the dimension of the moduli space of genus g
Riemann surfaces, as given in (C.7.9). For non-zero n, this gives the dimension of the moduli
space of genus g Riemann surfaces with n marked points.
The mapping between the moduli space of 4d N = 2 SU(2) quiver theories and that of
punctured Riemann surfaces can be made more precise as follows. We identify each trivalent
vertex with a three-punctured sphere,
s ×
⇔
× ×
i j

Gauging involves gluing two such trivalent vertices together, which in other words corre-
sponds to plumbing together two three-punctured spheres.
Indeed, let us consider gluing two trivalent vertices together to obtain the SU (2) Nf = 4
theory. In the corresponding Riemann surface picture, we would like to plumb together two
spheres along their punctures. Say that the two spheres have complex coordinates w and z,
with the three punctures being located at w, z = 0, 1, ∞. We will glue the point at w = ∞
to the point at z = 0,

× 1 × 1

0× × × ×∞

267
 −1 − 12 0 1
2 1

Figure 19: The fundamental domain for a sphere with four distinguishable marked points.

1
To glue them, we begin by changing to the local coordinate w̃ = w
which describes the
region around the w = ∞ puncture, and then we identify

z w̃ = q (15.10.16)

with q some complex parameter. This parameter is identified with the coupling in the field
theory via q = eiπτ .
From this picture, it is clear that the conformal manifold of the SU (2) Nf = 4 theory—
that is to say the space in which τ can take values—can be identified with the moduli space
of a four-punctured sphere. The latter is none other than the usual fundamental domain,
shown in Figure 6. The cusp at τ = i∞, i.e. q → 0, corresponds to the separating limit in
which the four-punctured sphere pinches off into two three-punctured spheres.
It is also useful to consider the moduli space of a four-punctured sphere when the four
punctures are taken to be distinguishable. In that case, the moduli space takes the form
shown in Figure 19. There are now three cusps, at τ = 0, 1, i∞, which correspond to
configurations in which different pairs of marked points collide. It is interesting to consider
the behavior of the field theory at each of these three cusps.
Let us begin with the weakly-coupled cusp at τ = i∞. At this point the theory is well-
described by the usual SU (2) Nf = 4 theory discussed before. This theory has an SO(8)
flavor symmetry, under which the hypermultiplets transform in the 8v representation. It will
be useful to consider the following subgroup of SO(8),

SO(8) ⊃ SO(4) × SO(4) ⊃ SU (2)a × SU (2)b × SU (2)c × SU (2)d (15.10.17)

This is the subgroup which is manifest in the trivalent diagram. In particular, each of the

268
four external legs carries one copy of SU (2). Let us assign them as follows,

b c
τ → i∞ :
a d

In the corresponding four-punctured sphere, the points labelled by a and b are close to one
another, and separated by a long thin tube from the points c and d. We now move towards
the point in moduli space with τ = 0. In this case the four-punctured sphere changes to a
configuration in which the points a and c approach one another, and are separated by a long
tube from b and d. Likewise as we approach the cusp at τ = 1, the points a and d approach
one another, and are separated by a long tube from c and b. In terms of trivalent diagrams,
this is
c b d c
τ →0: τ →1:
a d a b
We see that in all cases, the quiver is of the same general form as before, suggesting that
the strong-coupling regime of SU (2) Nf = 4 is again described by SU (2) Nf = 4. However,
distinguishing between the marked points allows us to make an important observation about
these strong-coupling dualities. In particular, let us recall that in the original weakly-coupled
theory, the hypermultiplets transformed in the 8v of SO(8). Decomposing this representation
into those of the SU (2) subgroups, we have

8v = 4 ⊕ 4 = (2a ⊗ 2b ) ⊕ (2c ⊗ 2d ) (15.10.18)

On the other hand, one has the following decompositions,

8s = (2a ⊗ 2c ) ⊕ (2b ⊗ 2d )
8c = (2a ⊗ 2d ) ⊕ (2b ⊗ 2c ) (15.10.19)

with 8s and 8c the spinor and conjugate spinor representations of SO(8). Thus swapping
SU (2)b with SU (2)c has the effect of switching from the vector to the spinor representation
of SO(8), and likewise for swapping SU (2)b with SU (2)d . We thus see that upon approach-
ing the strongly-coupled cusps in moduli space, we reobtain SU (2) Nf = 4, but with the
hypermultiplets transforming in spinor representations of the flavor symmetry group.
Similar statements hold for the [2] − (2) − (2) − [2] quiver whose trivalent diagram was
given above. Again, approaching any cusp gives rise to a quiver in the same form, but with
a rearrangement of the SU (2) subgroups of the total flavor symmetry. In general, though,
the story is far richer. Indeed, it is often possible that the theory emerging at a cusp has
a completely distinct quiver than that of the original theory. This for example happens in
the case of the three node quiver [2] − (2) − (2) − (2) − [2]. In particular, if we consider the

269
strong-coupling limit of the middle node, while keeping the outer nodes weakly coupled, the
quiver transforms as

which gives a quiver of an entirely new type.

We close this subsection by noting that a physical origin of the correspondence between
4d N = 2 theories and Riemann surfaces can be obtained via M-theory. In particular, if we
wrap an M5-brane on the relevant Riemann surfaces, with the insertion of appropriate defect
operators at the marked points, then reducing the worldvolume theory from six- to four-
dimension gives precisely the 4d N = 2 theories in question. Due to their six-dimensional
origins, the 4d N = 2 theories constructed in this way are referred to as “Class S” theories.

• Bibliographical notes
Comprehensive expositions of supersymmetric quantum field theories and supersymmetric
Yang Mills theories may be found in the books by Wess and Bagger [271], Weinberg [273],
Freedman and Van Proeyen [274], and Shifman [275].
Goddard-Nuyts-Olive (GNO) duality was introduced in [276] and realized in N = 4
Yang-Mills theory as Montonen-Olive duality in [277, 278]. The effect of topological charges
on the central charge of the supersymmetry algebra was obtained in [279], while the effect by
which magnetic monopoles in the presence of a θ-angle become dyons was derived in [280].
Subsequent tests of S-duality in N = 4 were carried out in [40]. The relation between electric-
magnetic duality and the geometric Langlands program was established in [281]. Integrated
correlators in the superconformal phase of N = 4 at fixed gauge coupling provided explicit
realizations of S-duality [282, 283, 284]. For a SAGEX overview of modular invariance in
N = 4 Yang-Mills theory and Type IIB superstring theory, see [285].
Seiberg-Witten theory was originally formulated for gauge group SU (2) in [145, 146].
Generalizations to other gauge groups with various hypermultiplet contents were soon there-
after produced, and are reviewed in the lecture notes [286, 287, 288, 289] and in the book
by Tachikawa [290], where extensive bibliographical references may also be found.
Construction of the prepotential using localization techniques was initiated for the in-
stanton part in [291], and generalized to include also the perturbative contributions in [292].
An overview of localization techniques is provided by the collection of essays in [293].
The relation between integrable systems and the Seiberg-Witten solution for the N = 2∗
theory was put forward in [294, 295]. The implementation using the Calogero-Moser system
for SU (N ) gauge group was carried out in [296], and for other gauge groups in [297, 298]
where the behavior under SL(2, Z) transformations was also analyzed. The derivations in

270
section 15.8 closely follow [296, 299], and reproduce the hypermultiplet mass expansion of
[300]. The modular properties of the Seiberg-Witten solution for gauge group SU (2) and
Nf ≤ 4 fundamentals were analyzed in detail in [301, 302] and for SU (3) in [303]. The role
of bi-modular forms was discussed in those references, in [304], and in the mathematical
literature in [305]. The Seiberg-Witten curve for linear quivers and gauge group SU (N ) was
constructed from M-theory in [306].
The dualities of N = 2 superconformal Yang-Mills theories were developed in [307] and
are reviewed in [308]. The global structure and spectrum of line operators in N = 4 Yang-
Mills is clearly presented in [272]. Useful lecture notes on the dynamics of four-dimensional
supersymmetric gauge theories may be found in [309].
Besides the beautiful Mathematics which arose from Seiberg-Witten theory, various
exciting physical results also emerged, including the discovery of a number of strongly-
coupled, non-Lagrangian theories known as Argyres-Douglas theories [147], as well as the
later-discovered Minahan-Nemeschansky theories [148, 149]. These bear close relation to
complex-multiplication, as well as to Kodaira’s theory of the degeneration of elliptic fibers.
Finally, let us mention that in the last decade an interesting connection between four-
dimensional N = 2 theories and 2d vertex operator algebras (VOAs) has been uncovered
[310], which has given even more direct connections between 4d N = 2 theories and modu-
larity. Indeed, by studying the modular differential equations satisfied by the corresponding
VOAs [311, 312], it has been possible to provide a partial classification for low-rank 4d N = 2
theories. The full power of these techniques remains to be explored.

271
16 Basic Galois Theory
We close these notes with a brief introduction to Galois theory, largely following the discus-
sions of [313, 314]. After having introduced the basic Mathematical ideas, we will illustrate
their use in Physics through examples from conformal field theory.

16.1 Fields
In Mathematics, a field35 is a set F equipped with two binary operations F × F → F on
elements a, b ∈ F , referred to as addition a + b and multiplication ab = a · b. Under addition
F forms a group with unit element 0, while under multiplication F \ {0} forms a group
with unit 1. The two operations are commutative and are intertwined by the property of
distributivity which requires a · (b + c) = a · b + a · c for all a, b, c ∈ F .
A field F is said to have characteristic 0 if there exists no integer n such that the n-fold
sum of the unit 1 vanishes. If there does exist a positive integer n such that the n-fold sum
of the unit 1 vanishes, then the smallest such integer may be shown to be a prime number p
and the characteristic of F is defined to be p. If F has characteristic p then we have p · a = 0
for all a ∈ F . The simplest finite fields of characteristic p prime are Fp = Z/(pZ).
Well-known examples of fields of characteristic 0 include the rational numbers Q, the real
numbers R, the complex numbers C, and the algebraic numbers (namely complex numbers
that satisfy a polynomial equation with rational coefficients). Rational functions of a single
variable x form fields of characteristic 0, namely Q(x), R(x), and C(x) depending on whether
the coefficients in the rational function are in Q, R, or C respectively.
A subset F ⊂ K of a field K is referred to as a subfield if F is a field with respect to
the binary operations of the field K. In the above examples, we have the subfield inclusions
Q ⊂ R ⊂ C, and Q(x) ⊂ R(x) ⊂ C(x), as well as Q ⊂ Q(x), R ⊂ R(x), and C ⊂ C(x).

16.2 Field Extensions

A field K is said to be an extension of a field F if F is a subfield of K, namely F ⊂ K. The
degree of the extension, denoted [K : F ], is the dimension of K viewed as a vector space
over the field F . If F is a subfield of a field K which itself is a subfield of a field L, then the
following relation holds between the degrees of extension,

F ⊂K⊂L [L : F ] = [L : K][K : F ] (16.2.1)

Equivalently, these relations hold when L is an extension of K which in turn is an extension

of F . For example, C is a field extension of R by the element i, and the degree of the

35
Not to be confused with classical or quantum fields in Physics, such as the electro-magnetic field.

272
extension is [C : R] = 2 since {1, i} is a basis of C over R. On the other hand, R is a field
extension of Q of infinite order.
An element y ∈ K is said to be algebraic over F if there exists a polynomial relation,

an y n + an−1 y n−1 + · · · + a0 = 0 an , . . . , a 0 ∈ F (16.2.2)

with an 6= 0 for some n ≥ 1. Given an extension K containing an element y that is algebraic

over F , there exists a unique monic irreducible polynomial f (x) ∈ F [x] that has y as a
root. This polynomial is known as the minimal polynomial of y over F . The uniqueness is
a non-trivial result—it implies that once we have found any monic irreducible polynomial in
F [x] with y as a root, it must be the minimal polynomial.
√
As an example, consider K =√ Q( 3) and F = Q. The polynomial f (x) = x2 − 3 is√monic
and irreducible in Q[x], and has 3 as a root. It is thus the minimal polynomial of 3 over
Q. This simple example illustrates an important fact: the minimal polynomial is dependent
on the choice of base field F . Indeed,
√ if we had
√ chosen F = R, then f (x) would have √ been
√ = (x − 3)(x + 3). Thus the minimal polynomial of 3 over
reducible, splitting to f (x)
R is instead f˜(x) = x − 3.

16.2.1 Simple extensions

Let K be an extension of F and y be an element in K. We denote by F (y) the intersection

of all subfields of K which contain both F and y. This intersection is a field, since the
intersection of subfields of K is itself a field. Since F (y) is contained in all subfields containing
F and y, it is by definition the smallest subfield containing F and y. We refer to such a
minimal extension F (y) as a simple extension of F by y. In fact, any finite extension of a
field F of characteristic 0 is a simple extension. For algebraic y, namely y satisfying f (y) = 0
for the minimal polynomial f of degree n with coefficients in F , one has the following result.

Theorem 16.1 If y is algebraic over F , and K is a field extension of F by the element y,

then the simple extension F (y) does not depend on K, and satisfies

F (y) ∼
= F [x]/f (x) [F (y) : F ] = n (16.2.3)

Thus, F (y) is completely determined by F [x] and the minimal polynomial f of y.

An analogous statement holds for finitely generated extensions F (y1 , y2 , . . . , yN ), as long
as K is an algebraic extension of F , i.e. if every element in K is algebraic over F .
An immediate consequence of Theorem 16.1 is that if y and w have the same minimal
polynomial f (x) in F [x], then F (y) is isomorphic to F (w).

273
16.2.2 Splitting fields
So far we have considered extension fields of F containing a root of a polynomial f (x) ∈ F [x].
We now consider extension fields that contain all the roots of f (x). In particular, let K be
an extension field of F and f (x) be a non-constant polynomial of degree n in F [x]. If f (x)
factors in K[x] as f (x) = α(x − y1 )(x − y2 ) . . . (x − yn ), then we say that f (x) splits in K.
If K is the smallest extension containing all the roots of f (x), then we refer to K as the
splitting field for f (x).
By Theorem 16.1, any two splitting fields of a polynomial in F [x] are isomorphic. Fur-
thermore, every polynomial f (x) ∈ F [x] admits a splitting field over F .
Theorem 16.2 Let F be a field and f (x) be a non-constant polynomial of degree n in F (x).
Then there exists a splitting field K of f (x) over F such that [K : F ] ≤ n!.

Of course, we already know that, if K is a simple extension and the polynomial f (x) has n
distinct roots, then [K : F ] = n by Theorem 16.1.
We have just noted that every polynomial admits a splitting field; it is then natural to ask
if there exists a field extension of F over which every polynomial splits. If such a universal
splitting field exists, the field F is said to be algebraically closed. A well-known example is
F = C. Another is the field A of all algebraic numbers over Q.

16.2.3 Normal extensions

We close this subsection by introducing the concept of a normal extension. An algebraic
extension K of F is said to be normal provided that whenever an irreducible√polynomial has
one root in K, it has all roots in K (i.e.√it splits over K). For example, Q( 3) is a normal
√
extension of Q since both of the roots√± 3 of the minimal polynomial x2 − 3 are in Q( 3).
On the other hand, the extension Q( 3 2) is not a normal extension of Q since√ two roots of
3 3
the minimal polynomial x − 2 are complex and hence not contained in Q( 2).

16.3 Field automorphisms and the Galois Group

An automorphism of a field K is a bijection σ : K → K such that for all a, b ∈ K we have,
σ(a + b) = σ(a) + σ(b)
σ(ab) = σ(a) σ(b) (16.3.1)
Two automorphisms σ, σ̃ are distinct if σ(a) 6= σ̃(a) for at least one element a ∈ K. Distinct
automorphisms are linearly independent—that is, if σ1 , . . . , σn are distinct automorphisms
of K, then a linear relation valid for all y ∈ K,
a1 σ1 (y) + · · · + an σn (y) = 0 (16.3.2)

274
implies that a1 = · · · = an = 0.
Let K be an extension field of F . An F -automorphism of K is an automorphism of the
field K, namely σ : K → K, that fixes each element of F . The set of all F -automorphisms
will be denoted Gal(K, F ) and is called the Galois group of K over F . The use of the
word “group” is justified since Gal(K, F ) can be endowed with group structure via function
composition. Indeed, if σ1 , σ2 ∈ Gal(K, F ) then clearly σ1 ◦ σ2 is also an automorphism of
K leavings F invariant. We begin with the following simple result,

Theorem 16.3 Consider a polynomial f (x) ∈ F [x]. If y ∈ K is a root of f (x) and

σ ∈ Gal(K, F ), then σ(y) is also a root of f (x).

To prove this theorem, let f (x) be an arbitrary polynomial of arbitrary degree n,

f (x) = cn xn + · · · + c1 x + c0 (16.3.3)

Since f (y) = 0, and σ is a homomorphism with σ(ci ) = ci for each ci ∈ F , we have,

0 = σ(0) = σ (cn y n + . . . c1 y + c0 )
= σ(cn )σ(y n ) + · · · + σ(c1 )σ(y) + σ(c0 )
= cn σ(y)n + · · · + c1 σ(y) + c0 (16.3.4)

and hence σ(y) is a root of f (x).

Theorem 16.3 shows that every image of a root y of f (x) under an automorphism of
the Galois group is also a root of f (x). Conversely, we could ask if every root of f (x) can
be obtained as the image of some root y under an automorphism σ ∈ Gal(K, F ). In some
instancess, this is indeed the case:

Theorem 16.4 If K is a splitting field for a polynomial in F [x] and y1 , y2 ∈ K are two
roots, then there exists an element σ ∈ Gal(K, F ) such that σ(y1 ) = y2 if and only if y1 and
y2 have the same minimal polynomial over F .

The forward direction follows immediately from Theorem 16.3. For the converse, note from
Theorem 16.1 that if y1 and y2 have the same minimal polynomial, then there is an iso-
morphism σ : F (y1 ) ∼ = F (y2 ). This isomorphism can be chosen such that σ(y1 ) = y2 , while
fixing F elementwise. Since K is a splitting field of some polynomial in F [x], it is also
a splitting field for the same polynomial over F (y1 ) and F (y2 ), and thus σ extends to an
F -automorphism of K, i.e. σ ∈ Gal(K, F ).
A important fact of which we make repeated use is the following,

275
Theorem 16.5 If K = F (y1 , . . . , yn ) is an algebraic extension of F , then automorphisms
in Gal(K, F ) are determined completely by their action on y1 , . . . , yn . That is, if σ1 , σ2 ∈
Gal(K, F ) satisfy σ1 (yi ) = σ2 (yi ) for all i = 1, . . . , n, then σ1 = σ2 .

Together with Theorem 16.3, this result can be used to at least partially √ √ construct con-
crete Galois groups. For example, consider the Galois group of Q( 3, 5) over Q. √ By
Theorem
√ 16.5 the Galois automorphisms √ are2 determined completely by their action on 3
and
√ 5. The
√ minimal polynomial of x − 3, and hence by Theorem 16.4 we√must
3 is √ √ have
3 → ± 3. Similar statements hold for 5. Hence the Galois group Gal(Q( 3, 5), Q)
has at most four elements,
√ 1 √ √ α √ √ β √ √ γ √
3→− 3 3−→− 3 3→− 3 3→− − 3
√ √ √ √ √ √ √ √
5→− 5 5→− 5 5→− − 5 5→− − 5 (16.3.5)
√ √
In fact, it turns out that in this case there are exactly four elements, and Gal(Q( 3, 5), Q) ∼
=
Z2 × Z2 , though we will not prove this here.
√In the√ above example, every automorphism of the Galois group permuted the four roots
± 3, ± 5. This is a particular case of a more general result,
Theorem 16.6 If K is the splitting field of a separable polynomial f (x) of degree n in F [x],
then Gal(K, F ) is isomorphic to a subgroup of Sn .

That is, every element of Gal(K, F ) produces a permutation of the√roots

√ of f (x). Of course,
the converse√is not necessarily
√ √ true.√For example, in the case of Q( 3, 5), the permutation
mapping ± 3 → ± 5, ± 5 → ± 3 does not correspond to an F -automorphism of K.

16.3.1 Intermediate Fields

Let K be an extension field of F . A field E such that F ⊆ E ⊆ K is called an intermediate
field of the extension. It can be shown that the Galois group Gal(E, F ) associated to the
intermediate field is a subgroup of Gal(K, F ). Hence to every intermediate field we can
associate a subgroup of the Galois group. Conversely, if H is a subgroup of the Galois
group, one can associate to it an intermediate field EH such that,
EH = {k ∈ K | σ(k) = k ∀ σ ∈ H} (16.3.6)
known as the fixed field of H. That EH is indeed a field is easy to check, since for example
the equalities σ(0) = 0 and σ(1) = 1 imply that 0 and 1 are contained in EH , and since,
σ(c + d) = σ(c) + σ(d) = c + d σ(cd) = σ(c)σ(d) = cd (16.3.7)
imply that EH is closed under addition and multiplication. The presence of inverses follow
simply as well. A somewhat stronger result is the following,

276
Theorem 16.7 Let K be a finite-dimensional extension field of F . If H is a subgroup of
Gal(K, F ) and EH is the fixed field of H, then H = Gal(K, EH ) and |H| = [K : E].
√ √ √
We may now return to the example
√ √ of Gal(Q( 3, 5), Q). In this
√ √case Q( 3) is an interme-
diate field of the extension Q( 3, 5) of Q. Recall
√ √that Gal(Q(
√ 3, 5), Q) = {1, α, β, γ} as
defined in (16.3.5). On the other hand, Gal(Q(
√ 3, 5), Q( 3)) should consist of only the
transformations leaving the base field Q( 3) invariant. Hence we have that
√ √ √
Gal(Q( 3, 5), Q( 3)) = {1, β} (16.3.8)
√ √
which is clearly a subgroup
√ of Gal(Q( 3, 5), Q). Conversely, the fixed field of H = {1, β}
is precisely EH = Q( 3).

16.4 The Fundamental Theorem of Galois Theory

Consider a finite-dimensional extension field K of F . Let S be the set of all intermediate
fields, and T be the set of all subgroups of the Galois group Gal(K, F ). We introduce a
function ϕ : S → T such that for each intermediate field E,

ϕ(E) = Gal(K, E) (16.4.1)

The function ϕ is referred to as a Galois correspondence. In this correspondence K is mapped

to the identity subgroup of Gal(K, F ), while F is mapped to the entire group Gal(K, F ).
By Theorem 16.7, the Galois correspondence is surjective.
While surjectivity of the Galois correspondence
√ is guaranteed, injectivity is not. For
3
example, consider
√ the extension field Q( 2) over Q. Every automorphism in the Galois √
3
group maps 3 2 to a root of the minimal polynomial
√ x − 2 by Theorem 16.3. Since 3
2 is
3
the
√ only real root, and since all elements of Q( 2) are real, every
√ automorphism must map
3 3
2 to itself. Thus by Theorem 16.5 we conclude√that Gal(Q( 2), Q) is trivial. This means
that under the Galois correspondence, both Q( 3 2) and Q must be associated to the same
group, i.e. the trivial group {1}. Therefore the Galois correspondence is not injective.
In the cases in which the Galois correspondence is injective, the extension K of F is
called a Galois extension. Given a Galois extension K, if E1 and E2 are intermediate fields
such that Gal(K, E1 ) = Gal(K, E2 ), then we can conclude that E1 = E2 . Note that in
characteristic 0, a Galois extension is simply a splitting field.
√ √
As an example, consider the extension Q( 3, 5) over Q. This extension is Galois be-
cause it is the splitting field of f (x) = (x2 −3)(x2 −5). In this case the Galois correspondence

277
ϕ must be bijective, and indeed we have
√ √ √ √ √ √
ϕ : Q( 3, 5) 7→ Gal(Q( 3, 5), Q( 3, 5)) = 1
√ √ √ √
Q( 3) 7→ Gal(Q( 3, 5), Q( 3)) = {1, β}
√ √ √ √
Q( 5) 7→ Gal(Q( 3, 5), Q( 5)) = {1, α}
√ √ √ √
Q( 15) 7→ Gal(Q( 3, 5), Q( 5)) = {1, γ}
√ √
Q 7→ Gal(Q( 3, 5), Q) = {1, α, β, γ} (16.4.2)

In this example we observe that √ all intermediate fields appearing are themselves Galois
extensions of Q. For example, Q( 3) is a Galois extension since it is the splitting field of
x2 − 3. Furthermore, the corresponding subgroups of the Galois group are all normal. The
general version of this statement is given by the Fundamental Theorem of Galois Theory,

Theorem 16.8 Let K be a Galois extension of F . Then

1. The map ϕ : S → T between the set S of intermediate fields and the set T of subgroups
of the Galois group Gal(K, F ) defined by ϕ(E) = Gal(K, E) is a bijection. One has

[K : E] = |Gal(K, E)| [E : F ] = [Gal(K, F ) : Gal(K, E)] (16.4.3)

2. An intermediate field E is a normal extension of F if and only if the correspond-

ing group Gal(K, E) is a normal subgroup of Gal(K, F ), in which case Gal(E, F ) ∼
=
Gal(K, F )/Gal(K, E).

16.5 Solvability by Radicals

The most famous application of Galois theory, and the reason for its initial development,
is to the study of roots of polynomial equations. In particular, Galois theory provides a
criterion for a polynomial to be solvable by radicals, as we now discuss. Let us first define
the notion of solvability by radicals. A field K is said to be a radical extension of a field F
if there exists a chain of fields,

F = F0 ⊆ F1 ⊆ F1 ⊆ F2 ⊆ · · · ⊆ Fr = K (16.5.1)

such that for each i = 1, 2, . . . , r we have Fi = Fi−1 (yi ) and some power of yi is in Fi−1 . Now
let f (x) ∈ F [x]. The equation f (x) = 0 is said to be solvable by radicals if there exists a
radical extension of F that contains a splitting field of f (x).
On the other hand, a group G is said to be solvable if there exists a chain of subgroups,

G = G0 ⊇ G1 ⊇ G2 ⊇ · · · ⊇ Gn−1 ⊇ Gn = {1} (16.5.2)

278
such that Gi is a normal subgroup of the preceding group Gi−1 and the quotient group
Gi−1 /Gi is abelian. For example, every Abelian group G is solvable since G ⊇ {1} is a chain
of the required form. As a more non-trivial example, consider the cyclic subgroup h(123)i of
order 3 in S3 . The chain,

S3 ⊇ h(123)i ⊇ h(1)i (16.5.3)

shows that S3 is solvable. On the other hand, it can be shown that for n ≥ 5 the group Sn
is not solvable.
We now state the famous Galois Criterion for solvability by radicals.
Theorem 16.9 Let F be a field of characteristic 0 and f (x) ∈ F [x]. Then f (x) = 0 is
solvable by radicals if and only if the Galois group of f (x) is solvable.

As a simple example, consider the polynomial f (x) = 2x5 − 10x + 5 ∈ Q[x]. By Theo-
rem 16.6, the Galois group must be a subgroup of S5 . We will now show that it is in fact
equal to S5 . To do so, note that the function f (x) has a maximum at x = −1, a minimum
at x = 1, and an inflection point at x = 0. This means that f (x) crosses the x-axis exactly
three times, and hence that f (x) has exactly three real roots. If K is a splitting field of f (x)
over Q, then |Gal(K, Q)| = [K : Q] by Theorem 16.8. Taking any root r of f (x), we have

[K : Q] = [K : Q(r)][Q(r) : Q] = 5[K : Q(r)] (16.5.4)

by Theorem 16.1. From this it follows that 5 divides |Gal(K, Q)|, and thus that Gal(K, Q)
contains an element of order 5. In other words, Gal(K, Q) contains a 5-cycle of S5 . In
addition, Gal(K, Q) also contains complex conjugation, which fixes the three real roots of
f (x) but interchanges the two non-real roots. Thus Gal(K, Q) contains a transposition. But
in fact, the only subgroup of S5 containing both a 5-cycle and a transposition is S5 itself.
We thus conclude that Gal(K, Q) = S5 . Then using the fact that Sn is not solvable for
n ≥ 5, we conclude from Theorem 16.9 that f (x) does not admit a solution in radicals. This
implies the famous result that there is no universal formula (involving only field operations
and roots) for the solutions of fifth-degree polynomial equations.

16.6 Galois Theory in rational conformal field theory

We close with an application of Galois theory to Physics. As discussed in sections 8.3.2
and 10.10, the torus partition of a rational conformal field theory (RCFT) can be expressed
in terms of a finite number of characters χ(τ ) = (χ1 (τ ), . . . , χd (τ )), transforming in a vec-
tor representation of the modular group. The action of S : τ → −1/τ in particular is
implemented by a matrix known as the modular S-matrix,

χ(−1/τ ) = S χ(τ ) (16.6.1)

279
which, by slight abuse of notation, will also be denoted by S. A crucial constraint on S is
that it satisfy the Verlinde formula,
†
X Sp,s Sq,s Sr,s
r
Npq = (16.6.2)
s∈I
S0,s
r
where s ∈ I labels the finite set of primary fields φs of the theory and Npq are fusion
coefficients of primaries, namely they satisfy,
X
r
φp × φq = Npq φr (16.6.3)
r
r
In particular, the operator product coefficients Npq are integers. Treating Np with entries
r r
(Np )q = Npq as matrices, we may write their characteristic polynomials as follows,

det (λp 1 − Np ) = 0 (16.6.4)

The polynomial in each λp is a monic with integer coefficients. From the Verlinde formula,
it follows that the roots are given by,
Sp,q
λ(q)
p = (16.6.5)
S0,q
(q)
Whereas the matrix entries of Np are integers, the ratios of S-matrix elements λp are
generically not even rational. Consider the extension field Q(Sp,q ) over Q, where Q(Sp,q )
denotes extension by the matrix elements of the modular S-matrix. By Theorem 16.3, the
(q)
elements σ ∈ Gal(Q(Sp,q ), Q) act on the λp by permuting them amongst themselves. In
particular, we have,
 
Sp,q σ(Sp,q ) Sp,σ(q)
σ = = (16.6.6)
S0,q σ(S0,q ) S0,σ(q)
(q)
Note that here we have associated the permutation of the roots λp to a permutation of
the underlying primary fields. Indeed, it is possible to prove that there exist a set of signs
σ (q) ∈ {±1} such that

σ(Sp,q ) = σ (q)Sp,σ(q) (16.6.7)

i.e. Galois conjugation acts on the S-matrix by signed permutation.

This so-called Galois symmetry of the RCFT leads to the following physical constraints.
Recall that the partition function of a RCFT is obtained by gluing together the chiral and
anti-chiral characters via,
X
Z(τ, τ̄ ) = Np,p̄ χp (τ ) χp̄ (τ̄ ) (16.6.8)
(p,p̄) ∈ I × I

280
By modular invariance the gluing matrix Np,p̄ is required to satisfy (N S)p,q̄ = (SN )p,q̄ . We
now apply a Galois automorphism σ to this equation, giving,
X X
Np,m̄ ¯σ (m̄)S σ(m̄),q̄ = σ (p) Sσ(p),m Nm,q̄
m̄ m
X
= σ (p) Nσ(p),m̄ S m̄,q̄
m̄
X
= σ (p) Nσ(p),σ(m̄) S σ(m̄),q̄ (16.6.9)
m̄

Using invertibility of S, we then obtain,

Nσ(p),σ(q) = σ (p)¯σ (q̄)Np,q̄ (16.6.10)

Because N ≥ 0, we conclude that Np,q̄ can only be non-zero when σ (p) = ¯σ (q̄) for all σ.
This turns out to be a very useful result, leading to powerful selection rules for possible
modular invariant combinations of characters. In the literature on rational conformal field
theory, it is often referred to as the “parity rule”.

• Bibliographical notes
An all-time classic introduction to Galois theory may be found in the book by Artin [315],
while a general treatment of modern algebra pertinent to Galois theory is given in the book
by van der Waerden [316]. The books by Escofier [313] and Cox [317] provide clear and
systematic introductions with many examples, problems sets, and historical notes. We also
refer to the recent book by Hungerford [314].
In the context of rational conformal field theory, Galois symmetry first appeared in
[318], with numerous follow-up works including [319, 320, 321]. The connection to Hecke
operators was discussed in [129, 130]. Galois symmetry also plays a role in the study of
three-dimensional topological quantum field theories, as discussed in e.g. [322, 323, 324].

281
Part III
Appendix
Four relatively short appendices collect basic mathematical material that, while central to
various parts of the core material in these lecture notes, is not concerned directly with either
modular forms or string theory.
Appendix A is primarily included for the benefit of the reader with a physics training, and
is presumably familiar to every reader with a mathematics training. It includes a basic review
of arithmetic modulo N , the Chinese remainder theorem, solving polynomial equations in
modular arithmetic, quadratic residue calculations, and the quadratic reciprocity theorem.
We conclude with a brief introduction to Dirichlet characters and L-functions.
Appendix B provides a review of some of the basic definitions, theorems, and general
results on compact Riemann surfaces needed in the main text. This material is directly
relevant to the physical applications to conformal field theory discussed in section 5; the
structure of the modular curves in section 6; the perturbative string amplitudes in section 12;
the topic of Seiberg-Witten theory in section 15; and to the subsequent appendices C and D.
The topological, metrical, and complex structural aspects of Riemann surfaces are reviewed
as are their uniformization and construction in terms of Fuchsian groups.
Appendix C provides a review of holomorphic line bundles on compact Riemann surfaces,
the Riemann-Roch theorem, the vanishing theorems, and their relation to the dimension of
moduli space and to the construction of the spaces of tensors and spinor fields on Riemann
surfaces. This material is directly relevant to the physical applications discussed in section 5,
to perturbative string amplitudes in section 12, and to the material in appendix D.
Appendix D provides an overview of higher rank Riemann ϑ-functions, the Riemann
relations on ϑ-functions, modular geometry of the Siegel upper half space, the Abel map,
Jacobian varieties, and their relation to higher genus Riemann surfaces via the Riemann
vanishing theorem. Using these ingredients, the construction of holomorphic and meromor-
phic differentials of arbitrary weight on Riemann surfaces of arbitrary genus is carried out,
using the prime form as an essential building block. As a Physics application, the correlation
functions for the bc system, which were already discussed in the special case of genus one
Riemann surfaces in section 5, are evaluated explicitly using the properties of meromorphic
differentials and the operator product expansion.

282
A Some arithmetic
In this appendix we collect some basic results in number theory, including the Chinese
remainder theorem, its application to solving polynomial equations, the Legendre and Jacobi
symbols, quadratic reciprocity, its application to solving quadratic equations mod N , and a
brief introduction to Dirichlet characters and L-functions. Throughout the set of positive
integers is denoted N = {1, 2, 3, · · · } and does not include 0.

A.1 Arithmetic mod N

We denote by ZN = Z/N Z = {0, 1, 2, · · · , N − 1} the additive group of integers modulo N .
Addition of two elements a, b ∈ ZN is defined by a + b (mod N ) ∈ ZN , the identity is 0, and
the inverse of a ∈ ZN is −a (mod N ). The group ZN is isomorphic to the multiplicative
group of N -th roots of unity {1, ε, ε2 , · · · , εN −1 } where ε = e2πi/N and multiplication is in C.
For p ∈ N a prime number, the set Z∗p = {1, 2, · · · , p − 1} forms a group under multipli-
cation of integers (mod p), and Fp = {0, 1, 2, · · · , p − 1} forms a finite field.
For arbitrary N ∈ N, we denote by Z∗N the maximal multiplicative subgroup of ZN .
When N is prime, the definition coincides with the one provided above. When N is not
prime, however, there are elements in ZN other than 0 which are not invertible, and have
to be removed to produce a multiplicative group. Consider an element a ∈ ZN which is not
relatively prime with N , namely such that gcd(a, N ) > 1, then it follows that the element
b = N/ gcd(a, N ) ∈ ZN satisfies ab = N so that ab ≡ 0 (mod N ) and therefore the element
a is not invertible in ZN . Every invertible element a ∈ ZN must be relatively prime to N ,
so that the multiplicative subgroup is given by,

Z∗N = {a ∈ ZN , gcd(a, N ) = 1} |Z∗N | = φ(N ) (A.1.1)

where φ(N ) is Euler’s totient function counting precisely the number of integers less than N
that are relatively prime to N , and |S| denotes the cardinality of the set S. For example,
we may represent the group Z∗15 by listing its elements in pairs of opposites modulo 15 as
follows Z∗15 = {±1, ±2, ±4, ±7}. The elements ±1 and ±4 square to 1 (mod 15), while
7 × (−2) ≡ (−7) × 2 ≡ 1 (mod 15).

A.2 Chinese remainder theorem

Theorem A.1 Let n1 , · · · , nr be positive integers which are relatively prime in pairs so that
they satisfy gcd(ni , nj ) = 1 for all 1 ≤ i 6= j ≤ r and let b1 , · · · br be arbitrary integers. Then
the system of congruences,

x ≡ bi (mod ni ) for all i = 1, · · · , r (A.2.1)

283
has a unique solution modulo N = n1 · · · nr . An equivalent statement is that the map,


x (mod N ) 7→ x (mod n1 ), · · · , x (mod nr ) (A.2.2)

is a ring isomorphism,

ZN ∼
= Zn1 × · · · × Znr (A.2.3)

To prove the theorem, we begin by proving that the solution is unique, so that the map
is injective. If we had two solutions x, y satisfying the same congruence conditions then
x − y ≡ 0 (mod ni ) for all i. Since the ni are pairwise co-prime their product N also divides
x − y so that x − y ≡ 0 (mod N ). Since the cardinality of the left and right sides are equal
to one another, the map must be surjective as well, and is thereby an isomorphism.

A.3 Solving polynomial equations

Theorem A.2 Let f (x) be a polynomial in x with integer coefficients f ∈ Z[x], let n1 , · · · nr
be positive integers which are relatively prime in pairs gcd(ni , nj ) = 1 for all 1 ≤ i 6= j ≤ r,
and let N = n1 · · · nr . Then the congruence,

f (x) ≡ 0 (mod N ) (A.3.1)

has a solution if and only if each one of the congruences,

f (x) ≡ 0 (mod ni ) i = 1, · · · , r (A.3.2)

has a solution. The number of solutions ν(N ) to (A.3.1) is given in terms of the number of
solutions ν(ni ) for each of the congruences by,

ν(N ) = ν(n1 ) · · · ν(nr ) (A.3.3)

To prove the theorem we prove both implications. Clearly if x satisfies (A.3.1) then it
satisfies (A.3.2). Conversely, let xi (mod ni ) be a solution to each congruence (A.3.2), which
exists by the assumption of the theorem, then there exists x (mod N ) which solves,

x ≡ xi (mod ni ) (A.3.4)

by the Chinese remainder theorem.

Applying Theorem A.2 Q toαipolynomial equations modulo an integer N whose prime factor-
ization is given by N = i pi for pi all distinct primes and αi ≥ 1, we have the equivalence,

f (x) ≡ 0 (mod pαi i ) for all i


f (x) ≡ 0 (mod N ) ⇔ (A.3.5)

The individual congruences for each prime power are given as follows.

284
Theorem A.3 Assume p prime, α ≥ 2, and let r be a solution of the congruence,
f (x) ≡ 0 (mod pα−1 ) (A.3.6)
with 0 ≤ r < pα−1 . The existence of solutions proceeds as follows.
(a) If f 0 (r) 6≡ 0 (mod p) then r lifts in a unique way from (mod pα−1 ) to (mod pα ).
(b) If f 0 (r) ≡ 0 (mod p)
• f (r) ≡ 0 (mod pα ) then r lifts from (mod pα−1 ) to (mod pα ) in p different ways.
• f (r) 6≡ 0 (mod pα ), then r cannot be lifted from (mod pα−1 ) to (mod pα ).
To prove the theorem, we consider all possible lifts of r from (mod pα−1 ) to (mod pα ),
which are given by r + qpα−1 (mod pα ). Taylor expanding the polynomial f of degree N ,
1 1 (N )
f (r + qpα−1 ) = f (r) + f 0 (r)qpα−1 + f 00 (r)q 2 p2α−2 + · · · + f (r)q N pN α−N (A.3.7)
2 N!
and reducing this equation (mod pα ) we obtain,
f (r + qpα−1 ) ≡ f (r) + f 0 (r)qpα−1 (mod pα ) (A.3.8)
Since f (r) ≡ 0 (mod pα−1 ), we must have f (r) = kpα−1 for some integer k. Therefore the
number of solutions to f (x) ≡ 0 (mod pα ) equals the number of solutions to the congruence
k + qf 0 (r) ≡ 0 (mod p). The statements of the Theorem now readily follows.

A.4 Quadratic residues and quadratic residue symbols

If m, N ∈ N and n ∈ Z with gcd(n, N ) = 1, then n is an m-th power residue mod N if
xm ≡ n (mod N ) admits a solution, and a non-residue if it admits no solutions. In the
special case of quadratic residues for m = 2 with prime N = p an odd prime, an integer
n 6≡ 0 (mod p) is a quadratic residue if the equation,
x2 ≡ n (mod p) (A.4.1)
has a solution, and a quadratic non-residue if the equation has no solutions. If n 6≡ 0 (mod p)
is a quadratic residue and x is a solution to (A.4.1), then −x is also a solution to the
same equation. When n ≡ 0 (mod p) then equation (A.4.1) has one solution, namely x ≡
0 (mod p).
The Legendre symbol (n|p) for p prime combines all these cases and is defined as follows,36

+1 if n is a quadratic residue mod p
(n|p) = −1 if n is a quadratic non-residue mod p (A.4.2)
0 if n ≡ 0 (mod p)


36
Frequently used alternative notations for the Legendre symbol are ( np ) and (n/p).

285
The number of solutions x to equation (A.4.1) is 1 + (n|p) in all cases.

Proposition A.4 Further properties of quadratic residues and the Legendre symbol for an
arbitrary odd prime p are as follows.
1. There are (p − 1)/2 residues and (p − 1)/2 non-residues in Z∗p ;
2. For any n ∈ Z we have (n|p) = n(p−1)/2 (mod p);
3. The first supplement is (−1|p) = (−)(p−1)/2 ;
2 −1)/8
4. The second supplement is (2|p) = (−)(p ;
5. For arbitrary m, n ∈ Z we have (mn|p) = (m|p)(n|p);
The proof proceeds as follows. Item 1 follows from the fact that the map x 7→ x2
is two-to-one in Z∗p . For Item 2, we first note that Fermat’s little theorem, namely that
np−1 ≡ 1 (mod p) for any n not divisible by p, implies n(p−1)/2 ≡ ±1 (mod p). For (n|p) = 1
there exists an x so that x2 ≡ n (mod p) and hence n(p−1)/2 = xp−1 ≡ 1 = (n|p) (mod p). For
(n|p) = −1 the degree (p − 1)/2 polynomial congruence equation x(p−1)/2 − 1 ≡ 0 (mod p)
has (p − 1)/2 solutions. The quadratic residues already provide (p − 1)/2 solutions, so the
non-residues cannot be solutions. As a result, we have x(p−1)/2 6≡ 1 (mod p) and hence
x(p−1)/2 ≡ −1 = (n|p) (mod p). Finally, Items 3, 4, and 5 follow from Item 2. Item 3
may also be derived from Fermat’s theorem that a prime p is the sum of the squares of two
integers if and only if p ≡ 1 (mod 4).
The JacobiQ symbol (n|N ) generalizes the Legendre symbol to an arbitrary positive odd
αi
integer N = i pi where pi are distinct primes and αi ≥ 1 and any n ∈ Z by,
Y
(n|N ) = (n|pi )αi (A.4.3)
i

where in each factor (n|pi ) is the Legendre symbol. The Kronecker symbol further generalizes
the Jacobi symbolQby lifting the restrictions that N be odd and positive, and is defined as
follows. If N = u i pαi i where u = ∓1 and the product runs over all distinct primes including
2, then the Kronecker symbol is defined by,
Y
(n|N ) = (n|u) (n|pi )αi (A.4.4)
i

where (n|pi ) stands for the Legendre symbol when pi is an odd prime, while the remaining
data are supplied by,

 0 if n ≡ 0 (mod 2)
1 if n ≥ 0
(n| − 1) = (n|2) = +1 if n ≡ ±1 (mod 8) (A.4.5)
−1 if n < 0
−1 if n ≡ ±3 (mod 8)


The Legendre, Jacobi, and Kronecker symbols are referred to as quadratic residue symbols.

286
A.4.1 Quadratic residues x2 ≡ −1 (mod N )
We now analyze the particular example of quadratic residues of −1 modulo N . This example,
as well as that in the following subsection, will be used extensively when we discuss the
dimensions and genera of congruence subgroups.
To obtain the number of solutions of the
Q congruence x2 ≡ −1 (mod N ) for an arbitrary
αi
integer N with prime factorization N = i pi we consider the polynomial f (x) = x2 + 1
with f 0 (x) = 2x. If p is an odd prime, and 0 ≤ r < p solves r2 + 1 ≡ 0 (mod p) then
r 6≡ 0 (mod p). As a result, f 0 (r) = 2r 6≡ 0 (mod p) so that there is a unique lift of r to
(mod p2 ), and to any (mod pα ). If p = 2 then r ≡ 1 (mod 2) so there is one solution (mod 2).
Now using Theorem 3, f 0 (r) ≡ 0 (mod 2), and the fact that every lift f (1 + 2q) ≡ 1 (mod 4)
implies that there are no solutions (mod 4), and hence no solutions (mod 2α ) for all α ≥ 2.
Hence the number of solutions to x2 ≡ −1 (mod N ) is given by,
(
 0 4|N
n  2 o
# x ∈ ZN  x ≡ −1 (mod N ) = Q (A.4.6)

i 1 + (−1|pi ) 4-N
p−1
where the Legendre symbol (−1|p) evaluates to (−1) 2 for an odd prime p. For p = 2, the
number of solutions is 1, so we set (−1|2) = 0.

A.4.2 Quadratic residues x2 + x ≡ −1 (mod N )

To obtain the number of solutions of theQcongruence x2 + x ≡ −1 (mod N ) for an arbitrary
integer N with prime factorization N = i pαi i we consider the polynomial f (x) = x2 + x + 1
with f 0 (x) = 2x + 1. Solving f (x) ≡ 0 (mod N ) is equivalent to solving,

(2x + 1)2 + 3 ≡ 0 (mod 4N ) (A.4.7)

If 9|N , then the congruence y 2 + 3 ≡ 0 (mod 9) has no solutions as the quadratic residues
mod 9 are 0, 1, 4, 7, giving the residues of y 2 + 3 to be 3, 4, 7, 1 but not 0. For an odd prime
p, the congruence y 2 + 3 ≡ 0 (mod p) has 1 + (−3|p) solutions. By quadratic reciprocity, if
p 6= 3 then (−3|p) = (−1|p)(3|p) = (p|3) which evaluates to ±1 when p ≡ ±1 (mod 3).
For p ≡ −1 (mod 3) there are no solutions y ∈ Z and hence no solution of the form
y ≡ 2x + 1 (mod p). For p ≡ 1 (mod 3) there are two solutions y (mod p). For any
given y, there is then a unique solution x to y ≡ 2x + 1 (mod p). The number of solutions
is thus given by,
(
 0 9|N
n  2 o
# x ∈ ZN  x + x ≡ −1 (mod N ) = Q (A.4.8)

i 1 + (−3|pi ) 9-N

For p = 3 the number of solutions is 1, so we set (−3|3) = 0.

287
A.5 Gauss sums
For an odd prime p, and an arbitrary a ∈ Z, the Gauss sum ga is defined by,
p−1
X
ga = (k|p) εka ε = e2πi/p (A.5.1)
k=0

Gauss sums are directly related to the Legendre symbol,

ga = (a|p)g g = g1 (A.5.2)

Indeed, for a divisible by p we have ga = 0 in view of the fact that the number of residues in
Z∗p equals the number of non-residues. Hence the relation holds trivially when p|a as both
sides vanish. For p - a, we use the fact that (a|p)2 = 1, the definition of ga , and the complete
multiplication property of the Legendre symbol,
p−1 p−1
X X
ka
ga = (a|p) (a|p)(k|p) ε = (a|p) (ka|p) εka = (a|p)g (A.5.3)
k=0 k=0

The last equality follows, for p - a, from the fact that the sum over k is identical to the sum
over ka. Furthermore, we have the following relation,

g 2 = (−)(p−1)/2 p (A.5.4)

which follows from ga g−a = (−1|p)g 2 and expressing its sum over a in two different ways,
p−1 p−1 p−1
X X X
2 (k−`)a
ga g−a = (p − 1)(−1|p)g = (k`|p)ε =p (k`|p)δk,` = p(p − 1) (A.5.5)
a=0 a,k,`=0 k,`=0

where δk,` = 1 when k = ` and zero otherwise. The result (A.5.4) follows upon using the
first supplement (−1|p) = (−)(p−1)/2 .

A.6 Quadratic reciprocity

Quadratic reciprocity states that for p, q distinct odd primes, we have,

(p|q)(q|p) = (−)(p−1)(q−1)/4 (A.6.1)

Many proofs exist of the law of quadratic reciprocity, of which Gauss produced at least 8.
Here we shall use Gauss’s sums as defined in the preceding subsection. For any odd prime q
not equal to p, we begin by using (A.5.4) to write,

g q−1 = (g 2 )(q−1)/2 = (g 2 |q) = (−1|q)(p−1)/2 (p|q) = (−)(p−1)(q−1)/2 (p|q) (A.6.2)

288
where we used the relation (n|p) ≡ n(p−1)/2 (mod p) of property 2. of A.4 in the second
equality. Next, we use the fact that the prime power of a multiple sum evaluates as follows,

(a1 + · · · + an )q = aq1 + · · · aqn (mod q) (A.6.3)

This relation is valid for a1 , · · · , an ∈ Z, but we generalize it here to the ring of algebraic
integers Zp (ε). As a result we have,
p−1
!q p−1 p−1
X X X
q
g = (k|p)εk = q kq
(k|p) ε = (k|p)εkq = gq (mod q) (A.6.4)
k=0 k=0 k=0

By (A.5.2), we have gq = (q|p)g, and upon combining with (A.6.2), we get,

g q−1 = (−)(p−1)(q−1)|2 (p|q) = (q|p) (A.6.5)

The last equality implies the law of quadratic reciprocity upon using (p|q)2 = 1.

A.7 Characters
Consider a finite Abelian group G. A function χ : G → C∗ is said to be a character if it is
a group homomorphism, i.e. χ(gh) = χ(g)χ(h) for all g, h ∈ G. Characters form a group
Ĝ = Hom(G, C∗ ), known as the Poincaré dual of G. We will denote the identity element
of this group as χ0 , such that χ0 (g) = 1 for all g. The inverse elements are obtained by
complex conjugation χ−1 (g) = χ(g).
As an example, consider the case of a cyclic group G = Zn . If we denote the order-n
generator by v, then it is clear that χ(v)n = χ(v n ) = 1 and hence we can take χ(v) to be an
n-th root of unity. Conversely, every n-th root of unity defines a character of Zn . Thus we
conclude that Ĝ is in fact isomorphic to the original group, Ĝ ∼
= Zn .
Characters satisfy the following relations

X |G| χ = χ0
χ(g) =
0 otherwise
g∈G

X |Ĝ| g=1
χ(g) = (A.7.1)
0 otherwise
χ∈Ĝ

The first line of each of these is obvious. To prove the second line of the first relation, begin
by choosing h ∈ G such that χ(h) 6= 1. Then we have
X X X
χ(h) χ(g) = χ(hg) = χ(g) (A.7.2)
g∈G g∈G g∈G

289
where we have used the fact P
that χ is a homomorphism, as well as closure of the group G.
Since χ(h) 6= 1, we conclude g∈G χ(g) = 0 as per the claim. The second relation is proven
exactly analogously—namely we multiply by χ0 (g) 6= 1 and used the fact that Ĝ is a group.
From these results we may obtain the following Schur orthogonality relations

|G| χ = χ0
X 
0
χ(g)χ (g) =
0 otherwise
g∈G

X |Ĝ| g=h
χ(g)χ(h) = (A.7.3)
0 otherwise
χ∈Ĝ

Indeed, for the former note that if χ, χ0 ∈ Ĝ then χ(χ0 )−1 ∈ Ĝ as well. We may thus use the
relation (A.7.1) with χ replaced by χ(χ0 )−1 . Then noting that,
X X
(χ(χ0 )−1 )(g) = χ(g)χ0 (g) (A.7.4)
g∈G g∈G

we obtain the desired result. Likewise for the latter relation, we use the fact that,
X X
χ(gh−1 ) = χ(g)χ(h) (A.7.5)
χ∈Ĝ χ∈Ĝ

This completes the proof of both relations in (A.7.1).

290
A.8 Dirichlet characters and L-functions
A Dirichlet character for a ∈ N is a function χa : Z → C satisfying the following properties:
1. Periodicity: χa (n) = χa (n + a) for all n ∈ Z.
2. χa (n) 6= 0 if and only if gcd(n, a) = 1.
3. Multiplicativity: χa (nm) = χa (n)χa (m) for all n, m ∈ Z.
Dirichlet characters can be thought of as characters for G = Z∗a , the multiplicative group
of invertible elements of Za . This is an Abelian group with |G| = φ(a), where φ is Euler’s
totient function. In this case our orthogonality relations read as follows,
a−1
φ(a) χa = χ0a
X 
0
χa (g)χa (g) =
0 otherwise
g=0

X φ(a) g ≡ h (mod a)
χa (g)χa (h) = (A.8.1)
0 otherwise
χa ∈Ĝ

We now introduce the Dirichlet L-function,

∞
X χa (n)
La (s, χ) = (A.8.2)
n=1
ns

For χa not equal to the identity character, La (s, χ) converges and is analytic for Re s > 1
[23, 124]. In this region, we have the following Euler product formula,
Y 1
La (s, χ) = χa (p)
(A.8.3)
p 1 − ps

the product being over prime numbers. To prove this, note that for each prime p, we have,
∞
X ∞
X
−s −1 n −ns
(1 − χa (p)p ) = χa (p) p = χa (pn )p−ns (A.8.4)
n=0 n=0

Hence for fixed prime q, we have,

Y 1 X χa (n)
χa (p)
= (A.8.5)
1− ns
p≤q ps n∈S q

with Sq the set of all natural numbers whose prime factors are less than or equal to q. Then
for any N ∈ N, we conclude that,
N
X χa (n) Y 1 X χa (n)
= χa (p)
− (A.8.6)
ns 1− ns
n=1 p≤q ps n∈S q
n≥N

291
with q the largest prime less than or equal to N . Taking N → ∞ gives (A.8.3).
Using this product formula, it is possible to show that La (s, χ0 ) extends to a meromorphic
function on Res s > 0, with the only pole at s = 1. We do not reproduce the proof here; it
can be found in e.g. [23, 124].

A.8.1 Dirichlet’s Theorem on Arithmetic Progressions

We now use Dirichlet L-functions to prove Dirichlet’s Theorem on arithmetic progressions,

Theorem 1 Let m and a be relatively prime natural numbers. Then there are infinitely
many prime numbers p such that p ≡ m (mod a).

In order to prove this, we aim to prove the following stronger result,

X 1
is divergent (A.8.7)
p
p≡m (mod a)

which clearly implies the theorem. To do so, begin by considering the logarithm of the
Dirichlet L-function, which by the product formula takes the form
X 
χa (p)

log La (s, χ) = − log 1 − s
(A.8.8)
p
p

valid for Re s > 1. Taylor expanding the logarithm, this gives

∞
X χa (p) XX χa (p)n
log La (s, χ) = + (A.8.9)
p
ps p n=2
npns

where, for reasons to be clear in a moment, we have split the n = 1 term from the rest.
Using the orthogonality of Dirichlet characters in (A.8.1), we find that
X 1 1 X X χa (p)
= χ a (m) (A.8.10)
ps φ(a) p
ps
p≡m (mod a) χa ∈Ĝ
 
∞ n
1 X X X X χa (p)
= χa (m) log La (s, χ) − χa (m) ns

φ(a) p n=2
np
χa ∈Ĝ χa ∈Ĝ

where (a, m) = 1 lest χa (m) = 0. To prove the theorem, we must now show that the right-
χa (p)n
hand side is divergent as s → 1. To do so, we begin by showing that p ∞
P P
n=2 npns is

292
bounded as s → 1, and hence that the second term in the difference does not diverge. To
see this, note that since |χa (p)| = 0 or 1, we have |χa (p)p−s | ≤ |p−s | ≤ 21 . Then we have
  
∞ 2 ∞  n−2   ∞
χa (p)n   χa (p)  X 1  χa (p)   χa (p) 2 X  χa (p) 2
 
X 1 1 1
≤ s  ≤ s  =  s  ≤ 2 (A.8.11)
   
ns
 n=2 np  p n p s p 22n−2 p p
  
n=2 n=2
Hence we conclude that
 
∞ ∞
X X χa (p)n  X 1 X 1
≤ < (A.8.12)

ns
 p n=2 np  p 2 n2

p n=1

the latter of which is convergent.

Having shown that the second term on the right-hand side of (A.8.10) is not divergent
as s → 1, we are now tasked with showing that the first term is divergent. At the end of
appendix A.8 we quoted the result that La (s, χ0 ) → ∞ as s → 1, and hence log La (s, χ0 )
diverges in this limit as well. To complete the proof it thus suffices to show that La (1, χ) 6= 0
for all χ 6= χ0 .
To show this, first consider a character χa such that χ2a is not the identity character.
Then we can consider the function
λ(s) = La (s, χ0 )3 La (s, χ)4 La (s, χ2 ) (A.8.13)
By (A.8.9) we have
 " ∞
#
n 2 n
 X X 3 + 4χa (p ) + χ a (p ) 
|λ(s)| = exp (A.8.14)
 
npns

 
p n=1
n
Letting θn,p = Arg χa (p ), we then obtain
 
∞
X X 3 + 4 cos θn,p + cos 2θn,p 
|λ(s)| = exp  (A.8.15)

npn Re s

 p n=1 
But for any real θ, we have
3 + 4 cos θ + 2 cos θ = 2(1 + cos θ)2 ≥ 0 (A.8.16)
Thus we conclude that
|λ(s)| ≥ e0 = 1 (A.8.17)
This gives the desired result. Indeed, if we had La (1, χ) = 0, then since La (s, χ0 ) has a
simple pole at s = 1 but La (s, χ2 ) does not (because χ2a 6= χ0 by assumption) the degree
≥ 4 zero wins against the degree 3 pole, and we would have λ(1) = 0, in contradiction to
the inequality just obtained. Thus we must have La (1, χ) 6= 0.
For the case of χ2a = χ0 , a similar argument can be run using λ(s) = La (s, χ)La (s, χ0 ). In
this case one aims to show that log λ(s) ≥ log ζ(2s). In total, one concludes that La (1, χ) 6= 0
for all χ 6= χ0 , completing the proof of the theorem.

293
A.8.2 Perfect squares and quadratic residues
We close by using the various tools of this section, including Dirichlet’s theorem of arithmetic
progressions, to prove the following statement:

Theorem 2 A number N ∈ N is a perfect square if and only if it is a quadratic residue for

all primes p coprime to N .

One direction of this statement is obvious: if N can be written as a perfect square, it is

clear that it is also a perfect square modulo any prime. We now focus on the other direction.
Assume that N is a quadratic residue modulo all coprime p. Say that N is not a perfect
square. We may then write N in the form N = j 2 p1 . . . pr for distinct primes p1 , . . . pr .
Since any odd power of 2 is a quadratic non-residue mod 3, we may assume that there is
at least one odd factor. Take p1 to be odd and consider non-zero integers m and ni such
that (m|p1 ) = −1 and (ni |pi ) = 1 for i = 2, . . . , r. By the Chinese remainder theorem, there
exists a solution to the following set of congruences

x ≡ 1 (mod 4) x ≡ m (mod p1 ) x ≡ ni (mod pi ) pi 6= 2 (A.8.18)

If one of the prime factors pi = 2, then we replace the final congruence with x ≡ 1 (mod 8).
Given a particular solution x0 to these congruences, the general solution takes the form
x = x0 + 4kp1 . . . pr for integer k. By Dirichlet’s theorem of arithmetic progressions this
takes an infinite number of prime values, and we choose one, say x = p, such that p - j. By
definition, we have

(p|p1 ) = (m|p1 ) = −1 (p|pi ) = (ni |pi ) = 1 i = 2, . . . , r (A.8.19)

Then using quadratic reciprocity, together with the first congruence of (A.8.18), we conclude
that

(p1 |p) = −1 and (pi |p) = 1 i = 2, . . . , r (A.8.20)

Thus (N |p) = (p1 |p) . . . (pr |p) = −1, and we have identified a prime p coprime to N for
which N is not a quadratic residue, contrary to hypothesis. We conclude that N must be a
perfect square.

• Bibliographical notes
An excellent overview of some classic problems in arithmetic is given in the book by Cox [140],
where a historical perspective is also provided. Two other classics are the books by Serre [124]
and Hecke [325]. Useful references for results in analytic number theory, including L-
functions, may be found in the book by Iwaniec [23], Rademacher [44], and Siegel [51].

294
B Riemann surfaces
A Riemann surface is a connected complex manifold of two real dimensions or equivalently a
connected complex manifold of one complex dimension, also referred to as a complex curve.
In this appendix, we shall review the topology of Riemann surfaces, their homology groups,
homotopy groups, uniformization, construction in terms of Fuchsian groups, as well as their
emergence from two-dimensional orientable Riemannian manifolds.

B.1 Topology
The topology of a Riemann surface Σ is the topology of the underlying orientable two-
dimensional manifold Σ. Throughout, we shall restrict to Riemann surfaces whose boundary
∂Σ is the union of a finite number b of one-dimensional components, and a finite number n
of points referred to as punctures of Σ. Given a triangulation of Σ with FΣ faces, EΣ edges,
and VΣ vertices, the Euler characteristic χ(Σ) is defined by,

χ(Σ) = FΣ − EΣ + VΣ (B.1.1)

The Euler characteristic is a topological invariant in the sense that its value is independent
of the triangulation chosen for a given Σ. In suitably chosen triangulations, the operation
of adding a puncture to Σ (namely removing a point from Σ) leaves FΣ and EΣ unchanged
but diminishes VΣ by one while adding a one-dimensional boundary component leaves EΣ
and VΣ unchanged but diminishes FΣ by one. The remaining contribution to χ(Σ) derived
from the genus g, which is the number of handles on Σ, is,

χ(Σ) = 2 − 2g − n − b (B.1.2)

A Riemann surface with b 6= 0 may be studied by considering its double, which is obtained
by gluing Σ and its mirror image Σ0 into a Riemann surface Σ̂ without one-dimensional
boundary components. The surface Σ̂ is endowed with an involution ι which swaps Σ and
Σ0 . Henceforth, we shall restrict attention to surfaces with b = 0.
Compact Riemann surfaces play a central role in both Mathematics and Physics. It will
often be useful to consider the generalization to a compact Riemann surface with n punctures
defined to be a compact Riemann surface with a finite number n of points removed, namely
with n punctures. Their study may be initiated by first studying the underlying compact
surface and then removing n points.

B.1.1 The Riemann-Hurwitz formula

Holomorphic maps between Riemann surfaces allow us to relate different Riemann surfaces.
We begin by considering a non-constant holomorphic map f : X → Y between two compact

295
Riemann surfaces X and Y . The degree of f is the multiplicity of the function f −1 given
by the cardinality d = |f −1 (y)| for all but finitely many points y ∈ Y . A formula valid for
all points in Y is obtained in terms of the degree of ramification ex of f at a point x ∈ X
which, in local coordinates, is given by f (z) − f (x) ∼ (z − x)ex . The degree of f may then
be defined for all y ∈ Y by the following formula,
X
d= ex (B.1.3)
x∈f −1 (y)

and d is independent of y. The Riemann-Hurwitz formula relates the genera gX and gY of the
surfaces X and Y to the degree of f and the deviations from 1 of the degree of ramification,
X
2gX − 2 = d(2gY − 2) + (ex − 1) (B.1.4)
x∈X

A hyper-elliptic Riemann surface X, viewed as a double cover of the Riemann sphere Y ,

provides an illuminating example of the Riemann-Hurwitz formula for d = 2 since the pres-
ence of 2n branch points at which ex = 2 readily gives gX = n − 1. The formula (B.1.4) may
be proven by using the expression for the Euler characteristic χ(X) in (B.1.1) of X in terms
of a triangulation by FX faces, EX edges, and VX vertices (including all the ramification
points), and similarly for χ(Y ). Given that the degree of the map is d, we have FX = dFY
P
and EX = dEY but VX = dVY − x∈X (ex − 1) since a vertex with ramification degree ex
counts for only one vertex of X. Expressing the result in terms of the Euler numbers we
P
have χ(X) = dχ(Y ) − x∈X (ex − 1). For a compact Riemann surface without boundary or
punctures, the Euler numbers are related to the genus by χ(X) = 2−2gX and χ(Y ) = 2−2gY
from which (B.1.4) follows.37

B.1.2 Homotopy
The homotopy groups π0 (Σ) and π2 (Σ) for a compact Riemann surface are trivial, and the
interesting homotopy resides entirely in the first homotopy group π1 (Σ), which we now define.
A closed curve C : [0, 1] → Σ is a continuous map such that C(1) = C(0). Two closed
curves C0 , C1 are said to be homotopic (to one another) iff there exists a family of curves
F(s, t) parametrized by t ∈ [0, 1], such that F(s, t) ⊂ Σ and is continuous in s, t for all
s, t ∈ [0, 1] and interpolates as follows: F(s, 0) = C0 (s) and F(s, 1) = C1 (s). The homotopy
relation amongst closed curves is clearly an equivalence relation, whose classes are the first
homotopy classes. To make the set of classes into a group under composition of maps, we
choose a base-point P ∈ Σ and restrict to curves in each class that pass through P . Thus,

37
For the special case of an un-ramified covering the formula χ(X) = dχ(Y ) for χ < 0 the formula readily
follows from the fact that the Euler number is proportional to the area for unit negative curvature metrics.

296
we may choose the parametrization of each closed curve C that passes through P so that
C(0) = C(1) = P . The composition C1 ◦ C0 of two curves C0 , C1 is given by,

for s ∈ [0, 12 ]

C0 (2s)
C1 ◦ C0 (s) = (B.1.5)
C1 (2s − 1) for s ∈ [ 12 , 1]

The composition of closed curves induces a composition of first homotopy classes of curves,
where the identity element e is the class of contractile curves, and the inverse is the class
of curves with opposite orientation. The resulting group of classes is independent of the
base-point P 38 and is referred to as the first homotopy group, or fundamental group π1 (Σ).

Proposition B.1 The Riemann sphere Ĉ, the complex plane C, the Poincaré upper half-
plane H, and the unit disc D are simply connected Riemann surfaces, as is any compact
Riemann surface Σ0 of genus 0, so that their first homotopy groups are given by,

π1 (Ĉ) = π1 (C) = π1 (H) = π1 (D) = π1 (Σ0 ) = {e} (B.1.6)

For a compact surface Σ of genus 1, namely a torus, we have π1 (Σ) = Z2 .

For surfaces of genus g ≥ 2, the fundamental group is non-Abelian, and will be discussed
concretely in appendix B.3.2 in terms of representations of SL(2, R).

P
B1
•
B2
A1 A2
Σ

Figure 20: A choice of homotopy generators AI and BI with common base point P is
illustrated for a compact genus-two Riemann surface Σ.

B.1.3 Homology
The top and bottom homology groups of a Riemann surface are trivial. All the interesting
homology resides in the first homology group which, for a compact Riemann surface of genus
g with n punctures, is given by H1 (Σ, Z) = Z2g+n . The primitive homology cycle associated
with each puncture is homologous to a small circle centered at the puncture. Having chosen
38
Independence of P requires Σ to be path-connected as we shall assume throughout.

297
an orientation for the Riemann surface Σ, we introduce a binary intersection pairing J(C1 , C2 )
on oriented 1-cycles C1 and C2 , which counts the number of intersections of the 1-cycles C1
with C2 weighed by ±1 factors for the relative orientation of the cycles at their intersection
points. The pairing J(C1 , C2 ) is odd under swapping C1 , C2 .
Specializing to the case of a compact Riemann surface Σ without punctures, the inter-
section form J is non-degenerate, and we may choose a basis for H1 (Σ, Z) = Z2g consisting
of homology cycles AI , BJ for I, J = 1, · · · , g such that,
J(AI , AJ ) = J(BI , BJ ) = 0
J(AI , BJ ) = −J(BI , AJ ) = δIJ (B.1.7)

Two different canonical bases (A, B) and (Ã, B̃) are related by a linear transformation that
may be represented by a matrix M with integer entries,
   
B̃ B
=M (B.1.8)
à A
Here, A and B are column matrices with entries AI and BI , respectively. The matrix M is
an element of the rank g modular group Sp(2g, Z) which preserves the intersection matrix J,
   
t 0 −Ig A B
M JM = J J= M= (B.1.9)
Ig 0 C D
where A, B, C, D are g ×g matrices with integer entries. An important subgroup of Sp(2g, Z)
is the group GL(g, Z) which consists of those M that transform A-cycles into linear com-
binations of A-cycles and B-cycles into linear combinations of B-cycles. It is obtained by
setting B = C = 0 and D−1 = At .

B.1.4 Cohomology
By the de Rham theorem, the first de Rham cohomology group is given by H 1 (Σ, R) = R2g
and is generated by 2g real-valued harmonic 1-forms on Σ. Since Σ is a complex manifold,
harmonic one-forms may be decomposed into holomorphic and anti-holomorphic one-forms,
and the cohomology may be decomposed into the Dolbeault cohomology groups,
H 1 (Σ, R) = H (1,0) (Σ) ⊕ H (0,1) (Σ) (B.1.10)
generated by holomorphic (1, 0) forms and anti-holomorphic (0, 1) forms respectively. A
canonical basis of the Dolbeault cohomology group H (1,0) (Σ, Z) = Zg consists of holomorphic
(1, 0)-forms ωI with I = 1, · · · , g whose periods are normalized on the cycles AI of the
canonical homology basis (A, B), as follows,
I
ωJ = δIJ (B.1.11)
AI

298
The periods on the remaining 1-cycles are then completely determined and give the matrix
elements of the period matrix Ω of Σ,
I
ωJ = ΩIJ (B.1.12)
BI

Under modular transformations M ∈ Sp(2g, Z), whose parametrization in terms of g × g

matrices A, B, C, D is given in (B.1.9), the matrix of holomorphic Abelian differentials ω,
the period matrix Ω, and its imaginary part Y , transform as follows,
ω̃ = ω(CΩ + D)−1
Ω̃ = (AΩ + B)(CΩ + D)−1
Ỹ = (ΩC t + Dt )−1 Y (CΩ∗ + D) (B.1.13)
Clearly, the transformation formula for Ω reduces to the standard Möbius transformation
formula for the special case of g = 1.

B.1.5 Riemann bilinear relations

A useful formula relates the integral over Σ of a wedge product of closed forms $1 and $2
to their line integrals over canonical homology cycles,
Z g I I I I 
X
$1 ∧ $2 = $1 $2 − $1 $2 (B.1.14)
Σ K=1 AK BK BK AK

The formula may be established by cutting the surface Σ along homology cycles AK and BK
for K = 1, · · · , g, chosen so as to have a point in common, into a simply-connected domain
Σcut in the plane, as illustrated in Figure 21. Expressing one of the closed differentials as
an exact differential of a local function in Σcut , for example, $2 = df2 , the formula may be
derived using Green’s theorem for an arbitrary 1-form $,
Z I
d$ = $ (B.1.15)
Σcut ∂Σcut

where the boundary consists of the union of cycles,

[ 
∂Σcut = AI ∪ BI ∪ A−1I ∪ B−1
I (B.1.16)
I

as illustrated in Figure 21 for genus g = 2.

The period matrix Ω is symmetric by the Riemann bilinear relations obtained by setting
$1 = ωI and $2 = ωJ in (B.1.14). Furthermore, the imaginary part of the period matrix,
Y = Im Ω (B.1.17)
is positive definite by the RiemannR relations obtained by setting $1 = ωI and $2 = ω̄J in
(B.1.14), and using the fact that 2i Σ $ ∧ $∗ is positive definite for holomorphic 1-forms $.

299
 P A−1
1
B−1
1
P

P
B1
P
P
B1 • A2
Σcut A1
B2
P P
A1 A2
Σ
B2
B−1
2

P
A−1
2
P

Figure 21: Representation of the Riemann surface Σ in terms of a simply connected domain
Σcut in C obtained by cutting Σ along the cycles AI and BI with common base point P in
common, with inverse cycles pairwise identified under the dashed arrows, illustrated here for
a compact genus-two Riemann surface Σ.

B.2 Metrics and complex structures

A different point of view on Riemann surfaces, with or without punctures, is obtained by
considering connected two-dimensional orientable manifolds endowed with a Riemannian
metric. This point of view is important in Physics and is widely regarded as the starting
point for string perturbation theory in the Polyakov formulation. Consider a two-dimensional
oriented manifold Σ endowed with a Riemannian metric g. In a system of local real coordi-
nates ξ m = (ξ 1 , ξ 2 ) the metric takes the form,39

g = gmn (ξ)dξ m dξ n (B.2.1)

We review the following notions:

• A diffeomorphism is a smooth map ξ → ξ 0 (ξ) under which g is invariant, and the

components gmn transform as follows,
0m
0 0 ∂ξ ∂ξ 0 n
gmn (ξ ) p = gpq (ξ) (B.2.2)
∂ξ ∂ξ q

Diffeomorphisms form an infinite-dimensional group Diff(Σ).

39
Throughout, the Einstein convention is used to sum over a pair of repeated upper and lower indices.

300
 • A Weyl transformation rescales the metric g by an arbitrary positive function λ on
Σ, namely gmn (ξ) → g0mn (ξ) = λ(ξ)gmn (ξ). Weyl transformations form an infinite-
dimensional group Weyl(Σ).

• Generally, a conformal transformation preserves all angles. All Weyl transformations

are conformal transformations in this sense. A conformal coordinate transformation,
or conformal diffeomorphism is a diffeomorphism which is conformal, which requires
that g0mn (ξ 0 ) = µ(ξ)gmn (ξ) for some positive real function µ.

• An isometry of g preserves all distances. A diffeomorphism is an isometry of g provided

g0mn (ξ 0 ) = gmn (ξ). Every conformal diffeomorphism is an isometry, but there exist
conformal transformations, such as scaling by a constant, that are not isometries.

We have the following theorem,

Theorem B.2 An orientable connected two-dimensional surface Σ endowed with a Rieman-

nian metric is automatically a Riemann surface.
To establish this result, we show that the combination of orientability and the presence of
a Riemannian metric induces a unique complex structure. The orientation may be specified
by choosing an oriented volume form,
p
dµg = det g εmn dξ m ∧ dξ n εmn = −εnm , ε12 = +1 (B.2.3)

A reversal of orientation corresponds to reversing the sign of ε. We denote by Tp (Σ) the

tangent space to Σ at the point p ∈ Σ, and by Tp∗ (Σ) the cotangent space. A convenient
basis is given by the partial derivatives ∂m = ∂/∂ξ m for Tp (Σ) and by the differentials dξ m
for Tp∗ (Σ). The tangent space Tp (Σ) may be viewed as the space of vectors v m ∂m at p, while
the cotangent space may be viewed as the space of forms ωm dξ m at p. An almost complex
structure J is a map from Tp (Σ) to itself, or equivalently from Tp∗ (Σ) to itself, whose square
is minus the identity map,

J : Tp (Σ) → Tp (Σ) vm → J mn vn J 2 = −I
J : Tp∗ (Σ) → Tp∗ (Σ) ωm → Jm n ωn (B.2.4)

Having a metric and an orientation, it is straightforward to exhibit this map,

p
J mn = det g εrn grm
p
Jm n = det g εmr grn (B.2.5)

The metric is a covariantly constant tensor with respect to the affine connection and its
associated covariant derivative ∇s , so that we have ∇s gmn = 0. As a result, J obeys

301
∇s Jm n = 0 and is therefore also a covariantly constant tensor acting on Tp (Σ) and Tp∗ (Σ).
Furthermore, J is invariant under arbitrary Weyl rescaling and therefore depends only on
the conformal class of gmn . Finally, J is integrable in the sense that the system of equations,

Jm n ∂n f = i ∂m f (B.2.6)

is integrable and thereby defines locally holomorphic functions f on Σ which endow Σ with a
complex structure.40 We may use the solutions to (B.2.6) to define a system of local complex
coordinates z, z̄ in which the metric g is locally conformally flat,

g = gzz̄ |dz|2 (B.2.7)

for a real positive function gzz̄ . The conformally flat form of the metric is invariant under
locally holomorphic diffeomorphisms z → z 0 (z). The complex manifold Σ may be constructed
by covering Σ with coordinate charts Uα in each of which we have local complex coordinates
(zα , z̄α ). In the intersection Uα ∩ Uβ of the two coordinate charts, the local coordinates are
related by transition functions ϕαβ (z) which are holomorphic and nowhere vanishing via the
relations zα (z) = ϕαβ (z)zβ (z). Thus Σ is a Riemann surface.

B.2.1 The Euler characteristic in terms of a metric

The Riemann tensor on a connected two-dimensional Riemannian manifold has only one
independent component, which may be expressed in terms of the Gaussian curvature Rg .
The Euler characteristic of a compact Riemann surface Σ with metric g may be expressed
in terms of Rg by the following classic formula,
Z
1
χ(Σ) = dµg Rg (B.2.8)
2π Σ

The normalization of the Gaussian curvature Rg is such that the round sphere S 2 of radius
1 has area 4π and Rg = 1, so that (B.2.8) indeed corresponds to genus zero. The Euler
characteristic is a topological invariant. Indeed, the integral in (B.2.8) is independent of the
metric g because dµg Rg is locally an exact (1, 1)-form.

B.3 Uniformization
In this subsection, we present the uniformization theorem for simply connected Riemann
surfaces, and then use its results to produce concrete constructions for arbitrary compact
Riemann surfaces with punctures.

40
In dimensions higher than two, the system (B.2.6) has non-trivial integrability conditions, but in two
dimensions there are none since the operator J − iI has a one-dimensional null space.

302
Theorem B.3 (The uniformization theorem) A simply connected Riemann surface is
conformally isomorphic to either one of the following Riemann surfaces,

1. The Riemann sphere Ĉ which is the one-point compactification of C (also denoted S 2 );

2. The complex plane C;
3. The complex upper half-plane H which is isomorphic to the unit disc D.

For an arbitrary Riemannian metric g on Σ in each case, there exists a Weyl transformation
to a metric ĝ such that Rĝ = 1 for Ĉ; Rĝ = 0 for C; and Rĝ = −1 for H and D. The
corresponding metrics are given as follows,

4|dz|2 |dz|2 4|dz|2

ĝS 2 = ĝC = |dz|2 ĝH = ĝD = (B.3.1)
(1 + |z|2 )2 (Im z)2 (1 − |z|2 )2

where z ∈ Ĉ, z ∈ C, Im (z) > 0, and |z| < 1 respectively.

Each one of these spaces is maximally symmetric and may be expressed as the quotient
of its isometry group by the isotropy group,

Ĉ = SU (2)/U (1) C = ISO(2)/SO(2)

H = SL(2, R)/SO(2) D = SU (1, 1)/U (1) (B.3.2)

The isometry groups P SL(2, R) = SL(2, R)/Z2 and P SU (1, 1) = SU (1, 1)/Z2 are isomor-
phic to one another. The action on the coordinate z in either case is given by,

az + b
z→ (B.3.3)
cz + d

where a, b, c, d ∈ R with ad − bc = 1 for H and c = b∗ , d = a∗ and |a|2 − |b|2 = 1 for D. In

both cases the elements I and −I leave z invariant and are identified under the center Z2 .
The geodesics of Ĉ are the great circles; those of C are straight lines; and those of H
are half-circles centered on the real axis. The geodesics of D are the circle segments in D,
centered at any α ∈ C with |α| > 1, satisfying |z − α|2 = |α|2 − 1. They intersect the unit
circle |z| = 1 orthogonally to it. The system of geodesics of H and D makes these spaces
into standard models for two-dimensional hyperbolic (or no-Euclidean) geometry.
The uniformization theorem provides us with a powerful tool to classify and construct
Riemann surfaces even when they are not simply connected. The key to the construction
is the first homotopy group π1 (Σ), which is trivial for simply connected Σ but non-trivial
when Σ is not simply-connected.

303
 i
w∈D
z∈H

0
i −1 • • •1
•
∞
•
−1 0 1
i

Figure 22: Geodesics in the upper half-plane H are semi-circles centered on R depicted in the
left figure, with their corresponding images under the conformal map w = (1 + iz)/(1 − iz)
from H into the unit disc D depicted in the right figure.

B.3.1 Genus one

Any compact Riemann surface Σ of genus 1 has π1 (Σ) = Z2 whose generators may be chosen
to be the A and B cycles of the first homology group. The universal simply connected
covering of a compact genus one surface Σ is the complex plane C. The fundamental group
is Abelian and acts on C by translations via a representation ρ : Z2 7→ C which defines a
lattice Λ. To specify the representation, and the lattice Λ, it suffices to assign the periods
ρ(1, 0) = ω1 and ρ(0, 1) = ω2 so that the lattice is given by Λ = ω1 Z ⊕ ω2 Z. The lattice is
two-dimensional (over R) provided ω2 /ω1 = τ ∈ H. The action of the fundamental group,
via the representation ρ on C, is transitive and the quotient C/Λ is the original genus one
compact Riemann surface Σ. Since translations in C are isometries of the flat Euclidean
metric |dz|2 on C, the genus-one surface C/Λ inherits the flat metric |dz|2 subject to the
identifications z ≈ z + ω1 and z ≈ z + ω2 .

B.3.2 Higher genus g ≥ 2

Theorem B.4 Every compact Riemann surface Σ of genus g ≥ 2 admits a Riemannian
metric ĝ of constant negative curvature Rĝ = −1.
The curvature Rg of an arbitrary metric g on Σ is related to the curvature Rĝ of a metric
ĝ related to g by a Weyl transformation g = e2σ ĝ by the Liouville equation,
1 p
Rg = e−2σ Rĝ + ∆g σ ∆g σ = − √ ∂m ( det g gmn ∂n σ) (B.3.4)
det g
where ∆g is the Laplace-Beltrami operator on scalar functions for the metric g. To prove
the theorem one shows that, for an arbitrary metric g, the Liouville equation has a solution

304
for Rĝ = −1. The Liouville equation first arose precisely in this context. Before presenting
the construction of higher genus surfaces we introduce the concept of a Fuchsian group.

B.4 Fuchsian groups

A systematic approach to the classification and explicit construction of Riemann surfaces
is through the use of Fuchsian groups. A Fuchsian group is defined to be isomorphic to a
discrete subgroup of SL(2, R). Fuchsian groups are special cases of Kleinian groups which
are discrete subgroups of SL(2, C). The elements γ 6= ±I of a Fuchsian group Γ ⊂ SL(2, R)
belong to one of the following three types,

γ elliptic |tr(γ)| < 2

γ parabolic tr(γ) = ±2
γ hyperbolic |tr(γ)| > 2 (B.4.1)

Since det (γ) = 1, γ must be conjugate, under SL(2, R), to one of the following matrices,
   
λ 0 ±1 1
(B.4.2)
0 λ−1 0 ±1

The left matrix is for elliptic type with |λ| = 1 and hyperbolic type with λ ∈ R, both with
λ 6= ±1. The right matrix is for parabolic type.

B.4.1 Action of Γ on H and R ∪ {∞}

Elements γ ∈ SL(2, R) acts on H by Möbius transformation,
 
az + b a b
γz = γ(z) = γ= (B.4.3)
cz + d c d

Since the element γ = −I leaves every point of H invariant, the action is effectively by the
normal subgroup P SL(2, R) = SL(2, R)/{±I}. The action of SL(2, R) on H is transitive
since an arbitrary point z = x + iy is the image of the point i ∈ H, namely z = γz (i), for a
group element γz whose entries satisfy c = 0, d = a−1 and x = ab, y = a2 . The isotropy group
of the point i is the SO(2) subgroup of SL(2, R) obtained by setting c = −b and d = a. The
isotropy subgroup of an arbitrary point z is the conjugate of SO(2) by the element γz given
by γz SO(2) γz−1 . As a result, H may be represented by the coset space,

H = SL(2, R)/SO(2) (B.4.4)

The action of SL(2, R) on H extends to a transitive action on R ∪ {∞} so that every point
x ∈ R ∪ {∞} may be expressed as x = γx (∞) for some γx ∈ SL(2, R).

305
 An important role will be played by the fixed points of various elements γ ∈ SL(2, R).
The classification of the elements of SL(2, R) into elliptic, parabolic, and hyperbolic given
in (B.4.1) is equivalent to the following classification in terms of their fixed points,

elliptic element one fixed point z ∈ H, the other at z̄

parabolic element one fixed point, which is in R ∪ {∞}
hyperbolic element two fixed points, which are in R ∪ {∞} (B.4.5)

In particular, the classifications of (B.4.1) and (B.4.5) apply to the elements of an arbitrary
discrete subgroup Γ ⊂ SL(2, R). One defines,

elliptic point of Γ is a point z ∈ H such that there exists

an elliptic γ ∈ Γ with γ(z) = z
cusp of Γ is a point x ∈ R ∪ {∞} such that there exists
a parabolic γ ∈ Γ with γ(x) = x (B.4.6)

If z is an elliptic point of Γ, then γ(z) is an elliptic point of Γ for all γ ∈ Γ. If x is a cusp of

Γ, then γ(x) is a cusp of Γ for every γ ∈ Γ.

Proposition B.5 (a) If z ∈ H is an elliptic point of Γ then the set Γz = {γ ∈ Γ | γ(z) = z}

is a finite cyclic group. (b) The set of elements of finite order in Γ consist of the elliptic
elements of Γ together with ±I.
To prove (a) we note that Γz is a subgroup of Γ. Representing z = γz (i) we see that Γz
is the set of γ ∈ Γ for which γz−1 γγz (i) = i. The isotropy group of the point i was earlier
identified as SO(2) so that Γz ⊂ SO(2). Since Γz is discrete and a subgroup of the compact
group SO(2), Γz must be finite. Since all finite subgroups of SO(2) are cyclic, Γz must also
be cyclic. To prove (b), we use the fact that an element γ ∈ Γ is conjugate under SL(2, C)
to a diagonal matrix with diagonal entries λ, λ−1 ∈ C as in (B.4.2). Finite order implies that
there exists a positive integer n such that λn = 1. When n = 1, 2 we have γ = ±I while for
n > 2, γ must be elliptic. Conversely, every elliptic element is of finite order by (a).

B.5 Construction of Riemann surfaces via Fuchsian groups

The fundamental group π1 (Σ) for an arbitrary compact Riemann surface Σ of genus g ≥ 2
is discrete and non-Abelian. The universal (i.e. simply connected) covering space Σ̂ of Σ
may be realized as a fiber bundle over Σ with structure group isomorphic to π1 (Σ). Since
Σ admits a metric ĝ of constant negative curvature Rĝ = −1 by the above theorem, the
universal covering surface admits a metric of constant negative curvature as well. Therefore,
by the uniformization theorem, we conclude that Σ̂ = H or the conformaly isomorphic D.
Conversely, the fundamental group π1 (Σ) must act on Σ̂ by an isometry of the constant

306
curvature metric. Since we have Σ̂ = H, whose isometry group is P SL(2, R), the action of
π1 (Σ) on H must be by a representation ρ,

ρ : π1 (Σ) 7→ P SL(2, R) (B.5.1)

which assigns an element of P SL(2, R) to each equivalence class of closed curves on Σ with a
specified, though arbitrary, base point P . To realize these assignments concretely, we choose
a base point P ∈ Σ, as well as a set of homology generators AI and BJ for I, J = 1, · · · , g
passing through P and intersecting one another no-where else, as shown in Figure 20.
We now cut the surface Σ along each one of the closed curves AI and BJ to unfold
Σ into a polygonal simply connected domain in C, as shown schematically for genus 2 in
Figure 23. To reconstruct the surface from the fundamental domain represented in Figure
23, we pairwise identify the curves corresponding to AI and A−1 I by the transformation
ρ(BI ) thereby closing the curves BI . Similarly, we identify the curves BI and B−1
I by the
−1
transformation ρ(AI ). Clearly, the composition of all commutators must give the identity,
so we must have,

ρ(C1 )ρ(C2 ) · · · ρ(Cg ) = I ρ(CI ) = ρ(AI )ρ(BI )ρ(AI )−1 ρ(BI )−1 (B.5.2)

Each homotopy class of π1 (Σ) contains a unique geodesic, or closed curve of minimal length
in the hyperbolic metric. Choosing the closed curves AI and BJ to be geodesics, we obtain
a representation of Σ in D by a domain ΣD whose boundary is the union of circular arcs
each of which is a geodesic of D, as represented in Figure 23. The hyperbolic length of the
closed curve AI is identical to the length of A−1
I and similarly the lengths of the segments
BI and B−1I are equal to one another.

B.5.1 Compactification of Γ\H

We choose the standard topology of H generated by open sets given by open metric discs.
Consider a Fuchsian group Γ ⊂ SL(2, R) whose cusps form a set C ⊂ R ∪ {∞}. To Γ we
associate the space,

H̄ = H ∪ C (B.5.3)

To specify the topology of H̄ we add open sets that contain the cusps in C to the topology
of H. For the cusp x = ∞ and the cusps x 6= ∞, these open sets may be chosen as follows,

{∞} ∪ z ∈ H such that Im (z) > y > 0 for y ∈ R+


x=∞
{x} ∪ z ∈ H such that |z − x − iy| < y for y ∈ R+

x 6= ∞ (B.5.4)

307
 P A−1
1
B−1
1
P

P
B1
ρ(A−1
1 )
ρ(B1 ) P

A2
A1
P ρ(B2 ) P

B2 ρ(A−1
2 ) B−1
2

A−1 P D
2

Figure 23: A fundamental domain for a compact genus-two Riemann surface Σ in the hyper-
bolic disc D. The surface Σ is recovered by identifying the curves AI with A−1
I and identifying
−1
the curves BI with BI .

The space H̄ equipped with this topology is clearly not compact. In fact, it is not even locally
compact.41 The group Γ consistently acts of H̄ which allows us to define the quotient,

X(Γ) = Γ\H̄ (B.5.5)

The quotient space Γ\H̄ is locally compact. Finally, we state the following proposition here
without proof.

Proposition B.6 If the space Γ\H̄ is compact then the number of Γ-inequivalent cusps
(resp. the number of elliptic points) is finite.

• Bibliographical notes
Classic references on Riemann surfaces are the books by Gunning [326] and by Farkas and
Kra [327]. Comprehensive lecture notes by Bost may be found in the collection [328]. A
detailed treatment of modular curves and their compactification may be found in Chapters
2 and 3 of [13], where a proof of Proposition B.6 is also given.
41
A topological space X is locally compact if every point in X has an open neighborhood which is contained
in a compact subset of X.

308
C Line bundles on Riemann surfaces
In this appendix, we shall present a brief review of complex line bundles over Riemann
surfaces. We begin by giving a general mathematical definition of a complex line bundle over
a Riemann surface Σ, the topological classification of line bundles, divisors, the Riemann-
Roch theorem, and various dimension formulas including the dimension of moduli space. We
then discuss the same topic of line bundles from a more physics-oriented point of view as
spaces of vector fields, forms, and spinors.

C.1 Holomorphic line bundles on a Riemann surface

A complex line bundle L on a Riemann surface Σ provides an assignment of a one-dimensional
complex vector space Lp to each point p ∈ Σ. A local description of the line bundle L may
S
be given by using a covering of Σ with open sets Uα so that α Uα = Σ. Introducing local
complex coordinates zα , z̄α in each open set makes Uα into a coordinate chart. Locally in
each coordinate chart, the bundle is a direct product Uα × C and each section f of L reduces
to a smooth C-valued function fα on Uα . In the intersection Uα ∩ Uβ the functions fα and
fβ are related by transition functions ϕαβ ,

fα = ϕαβ fβ in Uα ∩ Uβ (C.1.1)

The transition functions ϕαβ are nowhere vanishing and, for a holomorphic line bundle, must
be holomorphic functions. Considering now the intersection of three coordinate charts, we
obtain a condition on products of transition functions,

ϕαβ ϕβγ = ϕαγ in Uα ∩ Uβ ∩ Uγ (C.1.2)

Note that, for a holomorphic line bundle L, the transition functions are holomorphic, but
the sections f and the local functions fα need not be holomorphic.

C.1.1 Topological classification

To obtain the topological classification of line bundles over Σ, we solve the non-vanishing
condition on the transition functions ϕαβ in terms of exponentials, ϕαβ = exp(2πiψαβ ), where
ψαβ is defined by ϕαβ up to the addition of an arbitrary integer nαβ . The product relation
(C.1.2) is then equivalent to the fact that the combinations,

cαβγ = ψαβ + ψβγ − ψαγ (C.1.3)

are integers since their exponential equals 1 by (C.1.2). The transformations ψαβ → ψαβ +nαβ
imply transformations of cαβγ by an exact cocyle nαβ + nβγ − nαγ ,

cαβγ → cαβγ + nαβ + nβγ − nαγ (C.1.4)

309
By construction in terms of ψαβ , the integers cαβγ satisfy the following closed cocycle relation
on the intersection of four coordinate charts,
cαβγ − cβγδ + cγδα − cδαβ = 0 (C.1.5)
The space of closed cocycles cαβγ satisfying (C.1.5) modulo exact cocycles nαβ + nβγ − nαγ
is the second Čech cohomology group of Σ with integer coefficients, denoted H 2 (Σ, Z).
One may translate the Čech cohomology formulation of line bundles into the language
of differential forms, de Rham cohomology, and gauge fields, which is more familiar to
physicists. It will be convenient here to assume that Σ is compact so that one may extract
from the covering of Σ by open sets Uα a covering with only a finite number of open sets,
and avoid issues of convergence.
Each line bundle L on Σ corresponds to a cohomology class of H 2 (Σ, Z), labeled by the
first Chern class c1 (L). In turn, the first Chern class c1 (L) may be represented in terms of a
de Rham cohomology class [F ] where F is the curvature 2-form of a U (1) connection 1-form.
In each coordinate chart Uα a representative F of the class [F ] may be expressed in terms of
a local connection form Aα by F = dAα . The transition functions for the connection are,42
Aα − Aβ = dλαβ (C.1.6)
with cαβγ = λαβ + λβγ − λαγ . The first Chern class, defined by,
Z
i
c1 (L) = F (C.1.7)
2π Σ
takes integer values, so that H 2 (Σ, Z) = Z. For later use, we note that the first Chern class
of the tensor product of two line bundles L1 and L2 over Σ is given as follows,
c1 (L1 ⊗ L2 ) = c1 (L1 ) + c1 (L2 ) (C.1.8)
We conclude that the topological classification of line bundles L over a compact Riemann
surface Σ is in terms of a single integer d, referred to as the degree of L, which labels the
first Chern class c1 (L) of L.

C.1.2 Holomorphic classification: divisors

Holomorphic line bundles in a given topological class characterized by the degree c1 (L) = d
are not necessarily equivalent holomorphically. The space of all holomorphic line bundles for
given degree d is referred to as the Picard variety,
Picd (Σ) = {holomorphic line bundles L on Σ with c1 (L) = d} (C.1.9)
42
To be precise the de Rham theorem states an isomorphism between the de Rham cohomology group
2
HdR (Σ) and the Cech cohomology group with real coefficients H 2 (Σ, R).

310
The special case of d = 0 corresponds to the Jacobian variety J(Σ) = Pic0 (Σ).
A convenient way to characterize a holomorphic line bundle is by its divisor, which is a
complex codimension-one subset of Σ given by a formal sum over distinct points pα ∈ Σ,
N
X
D= nα pα (C.1.10)
α=1

The integer nα denotes the order of the point pα in the divisor D and may take positive or
negative values, while points for which nα = 0 are usually omitted from the sum.
To construct the line bundle L associated with a divisor D, we introduce coordinate
charts Uα with pα ∈ Uα and a local complex coordinate zα in Uα that vanishes at pα for
each α = 1, · · · , N . By choosing each Uα small enough we can make them mutually disjoint
Uα ∩ Uβ = ∅ for α 6= β. We denote the open complement to all the points pα by U∞ =
S
Σ \ {p1 , · · · , pN } so that U∞ α Uα = Σ. The line bundle L corresponding to the divisor D
is then constructed by taking transition functions in the intersections Uα ∩ U∞ to be zαnα ,
the integers nα corresponding to the order of the point pα in the divisor D in (C.1.10). A
section f of L, described locally by functions fα on Uα and f∞ in U∞ , satisfies fα = zαnα f∞ .
Conversely, given a line bundle L and a section f , the corresponding divisor may be recovered
uniquely from the zeros and poles of f . Since the open sets Uα are mutually disjoint, all
intersections with more than two U coordinate charts are empty. As a result, the cocycle
conditions (C.1.2) and (C.1.5) are automatically satisfied.
Adding the contributions from each transition function to the first Chern class, we find,
N
X
d = c1 (L) = nα (C.1.11)
α=1

In other words, the degree d of L is the total number of zeros (counted with their orders nα
when nα > 0) minus the total number of poles (counted with their orders −nα when nα < 0).
A topologically trivial bundle L with d = 0 has an equal number of zeros and poles (counted
with their orders) and its meromorphic sections are the meromorphic functions on Σ. One
may introduce an equivalence relation between divisors such that D1 and D2 are equivalent
to one another if they differ by a divisor of degree zero. When this is the case, the ratio f2 /f1
of any two sections f1 and f2 of D1 and D2 respectively, has degree zero. The equivalence
class with representative divisor D is denoted by [D].

C.1.3 Examples
A familiar example in Physics is provided by the Dirac magnetic monopole, which may be
thought of having a constant field strength F on the Riemann sphere Ĉ. Denoting the North
and South poles by p± , we introduce coordinate charts U± = Ĉ \ {p∓ } and local coordinates

311
 n
z± which vanish at p± . Choosing f− = 1 and f+ = z++ gives ψ+− = n+ ln(z+ )/(2πi) and
A+ − A− = n+ dz+ /(2πiz+ ), so that c1 (L) = n+ for a Dirac monopole of charge n+ .
The tangent bundle T Σ over a Riemann surface Σ is a bundle of rank 2 over the reals.
Using the complex structure of Σ it may be decomposed into the direct sum of its holomorphic
component T Σ(1,0) and its complex conjugate T Σ(0,1) . The cotangent bundle T ∗ Σ may be
decomposed analogously,

T Σ = T Σ(1,0) ⊕ T Σ(0,1)
T ∗ Σ = T ∗ Σ(1,0) ⊕ T ∗ Σ(0,1) (C.1.12)

The bundles T Σ(1,0) and T ∗ Σ(1,0) are holomorphic line bundles over Σ. The holomorphic
cotangent bundle T ∗ Σ(1,0) is also referred to as the canonical bundle over Σ and denoted K,
while T Σ(1,0) is isomorphic to K −1 .
This decomposition may be rendered explicit on the sections of the bundles. In terms
of real local coordinates ξ m on Σ a section of T Σ is a vector field and may be expressed as
v m ∂m while a section of T ∗ Σ is a differential one-form that takes the form ωm dξ m . In local
complex coordinates z, z̄ they may be decomposed as follows,

v m ∂m = v z ∂z + v z̄ ∂z̄
ωm dξ m = ωz dz + ωz̄ dz̄ (C.1.13)

where v z ∂z is a section of T Σ(1,0) and ωz dz is a section of T ∗ Σ(1,0) and similarly for their
complex conjugates. Their transition functions are given as follows,

(v z )α (zα ) = (v z )β (zβ ) (∂zα /∂zβ )

(ωz )α (zα ) = (ωz )β (zβ ) (∂zα /∂zβ )−1 (C.1.14)

where ∂zα /∂zβ is a holomorphic nowhere vanishing transition function in Uα ∩ Uβ .

Finally, a spin bundle S on Σ is a holomorphic line bundle whose tensor square is iso-
morphic to the canonical bundle K,

S⊗S ≈K c1 (S) = 21 c1 (K) (C.1.15)

The relation between the Chern numbers follows from (C.1.8). By a slight abuse of notation,
1
a spin bundle is some times denoted by K 2 . For g = 0 the spin bundle is unique. However,
for g ≥ 1, the relation to the canonical bundle does not specify the spin bundle S uniquely.
The different spin bundles are referred to as spin structures. On a surface of genus g, there
are 22g independent spin structures.

312
C.2 Holomorphic sections and the Riemann-Roch theorem
As emphasized earlier, while the transition functions of a holomorphic line bundle are holo-
morphic functions, the sections of a holomorphic line bundle are not generally holomorphic.
The Cauchy-Riemann operator ∂¯L for a holomorphic line bundle L acts on the sections of L
by the ordinary Cauch-Riemann operator ∂¯ = dz̄ α ∂z̄α with respect to the local coordinate
zα in each open set Uα . This action is covariant without the need for a connection, as may
be seen by applying ∂¯ to local sections in the overlap of open sets Uα given by (C.1.1),
¯ α = ϕαβ ∂f
∂f ¯ β in Uα ∩ U β (C.2.1)

The subspace of all holomorphic sections of a holomorphic line bundle plays a special role
and contains a lot of information on the line bundle. It may be defined as follows. To a
holomorphic line bundle L we associate the vector space of its holomorphic sections,

Ker(∂¯L ) = {holomorphic sections of L} (C.2.2)

These spaces are finite-dimensional and their dimensions are related by the Riemann-Roch
theorem, which will be proven, in a slightly different guise, in appendix C.5.

Theorem C.1 The Riemann-Roch theorem for a holomorphic line bundle L states
1
dim Ker ∂¯L − dim Ker ∂¯K⊗L−1 = c1 (L) + χ(Σ)



(C.2.3)
2
where K is the canonical bundle, χ(Σ) is the Euler characteristic of Σ, and ∂¯L is the Cauchy
Riemann operator acting on L.

C.3 Vanishing Theorem and dimension formulas

In this subsection, we shall state a vanishing theorem and combine its implications with
the Riemann-Roch theorem to obtain general expressions for the dimensions of the spaces
of homomorphic sections of line bundles. We shall specialize to the case of compact Σ and
make use of the fact that the Euler number is then given in terms of the genus g of Σ by
χ(Σ) = 2 − 2g.
We begin by obtaining some immediate consequences of the Riemann-Roch theorem.
Since the only holomorphic functions on Σ are the constant functions, the corresponding
bundle L is the trivial bundle with c1 (L) = 0 and we have Ker(∂¯L ) = C and dim Ker(∂¯L ) = 1.
Using the Riemann-Roch theorem for the trivial bundle L, we find

dim Ker(∂¯K ) = g (C.3.1)

For g = 0 there are no holomorphic 1-forms. For g = 1 we recover the observation, made long
ago, that the space of holomorphic 1-forms is one-dimensional. For g ≥ 1, there are g linearly

313
independent holomorphic 1-forms on Σ already identified in appendix B.1.4. Setting L = K,
and using the fact that K ⊗ K −1 is the trivial bundle, the Riemann-Roch theorem implies
c1 (K) = 2g − 2. For g = 1 we recover the fact, observed long ago, that the holomorphic
1-form dz is nowhere vanishing, while for g ≥ 2 we obtain the result that every holomorphic
one-form has 2g − 2 zeros (counted with multiplicities).
More generally, representing the divisor class [L] of an arbitrary line bundle L by a tensor
1
power of the canonical bundle [K n ] for n ∈ Z/2, where it is understood that K 2 = S is a
spin bundle with specified spin structure, we have the following vanishing theorem:

Theorem C.2 The following holds


g=0 and n>0
dim Ker(∂¯L ) = 0 (C.3.2)
g≥2 and n<0

The entry for g = 0 states the absence of holomorphic forms or spinors of degree greater
than 0 on the sphere. The entry for g ≥ 2 states that there are no holomorphic vector fields
on a higher genus Riemann surface. The proof of this theorem will be giving in appendix
C.6, in a slightly different guise.
Combining the results of the vanishing theorem with the results obtained from the
Riemann-Roch theorem for arbitrary divisor class [L] = [K n ] and using the relations c1 (L) =
c1 (K n ) = nc1 (K) for n ∈ Z/2 we obtain for n 6= 21 ,


 (2n − 1)(g − 1) for g ≥ 2 and n ≥ 32
g g ≥0 n=1



¯
dim Ker(∂L ) = 1 g ≥0 n=0 (C.3.3)
1 g =1 n 6= 21




1 − 2n g =0 n ≤ − 12


For n a half-odd-integer, these dimensions are independent of the spin structure of L. But
for n = 21 the dimension of Ker(∂¯L ) depends on the spin structure of L. Spin structures are
partitioned into even and odd depending on whether dim Ker(∂¯L ) is even or odd,

1 (mod 2) g≥1 n = 21 (odd spin structures)
dim Ker(∂¯L ) = (C.3.4)
1
0 (mod 2) g≥1 n = 2 (even spin structures)


The dimensions in (C.3.4) equal 1 and 0 respectively throughout moduli space for g = 1, 2
and at generic points in moduli space for arbitrary g ≥ 3. For g ≥ 3, the dimensions may
increase by a multiple of 2 at certain sub-varieties of moduli space, including at the locus of
hyper-elliptic Riemann surfaces.

314
C.4 Tensors and spinors on Σ
In Physics, we will deal with scalar, vector, tensor, and spinor fields on a Riemann surface Σ
viewed as a two real dimensional surface equipped with a Riemannian metric,

g = gmn (ξ) dξ m dξ n (C.4.1)

in a system of real coordinates ξ m . We shall now analyze complex line bundles, covariant
derivatives and Laplace-Beltrami operators on line bundles in a tensorial formulation, and
use it to prove the Riemann-Roch and vanishing theorems.
An arbitrary tensor field tm1 ...mn of rank n may be decomposed into a direct sum of
symmetric traceless tensors and scalars. Indeed, any anti-symmetric pair of indices may be
√
contracted with the tensor g εmn to produce a scalar, while the metric gmn may be used
to eliminate all traces. The resulting symmetric traceless tensor may be further decomposed
into the eigenspaces of the complex structure J . The convention v z ∂z = v z (dz)−1 may be
used to identify a form of negative weight with a vector field of opposite weight. Henceforth,
we shall suppress the z and z̄ indices on these reduced tensors, so that t = tz,···z and t̄ = tz̄··· ,z̄ .
Thus, all tensor fields decompose into a direct sum of the following spaces,
n o
∗
T(n,0) = t(z) : Σ → C such that t0 (z 0 )(dz 0 )n = t(z)(dz)n (C.4.2)

∗
The space T(1,0) may be identified with the space of sections of the canonical bundle K,
∗
while the spaces T(n,0) for n > 0 may be identified with the space of sections of the n-th
n
tensor power K of the canonical bundle K. Finally, we include spinors by allowing n to
take half-integer values which further requires specifying a spin structure.

C.4.1 Covariant derivatives and Laplace-Beltrami operators

∗
On each space of sections T(n,0) we define an L2 (Σ) inner product as follows,
Z
ht1 |t2 i = dµg (gzz̄ )−n t1 (z) t2 (z) (C.4.3)
Σ

The Cauchy-Riemann operator ∂¯ = (dz̄)∂z̄ acts covariantly on the spaces T(n,0)

∗
, without the
need for a Christoffel connection,

∂¯ : T(n,0)
∗ ∗
with t(z)(dz)n → ∂z̄ t(z) (dz)n dz̄


→ T(n,1) (C.4.4)

It is often more convenient to use a covariant derivative which maps a space into a space of
∗
the same type. This may be achieved by the Cauchy-Riemann operator acting on T(n,0) ,
∗ ∗
∇z(n) : T(n,0) t(z)(dz)n → ∇z(n) t(z) = g zz̄ ∂z̄ t(z) (dz)n−1 (C.4.5)


→ T(n−1,0) with

315
 (n−1)
To define the adjoint operator −∇z of ∇z(n) with respect to the L2 (Σ) inner product does
require a Christoffel connection and is given by,
∗ ∗
∇(n) n n+1


z : T(n,0) → T(n+1,0) with t(z)(dz) → ∇z t(z) (dz) (C.4.6)

The adjoint operators, and the associated Laplace-Beltrami operators, are as follows,

†
∇z(n) = −∇(n−1)
z ∆+ z (n)
(n) = −2∇(n+1) ∇z

†
∇(n)
z = −∇z(n+1) ∆− (n−1) z
(n) = −2∇z ∇(n) (C.4.7)

The Laplace-Beltrami operators both map T(n,0) → T(n,0) . The non-zero eigenvalues and
their corresponding eigenfunctions are related. To see this, we observe that,

∆+ ∆− (n) (n)



(n) ψ = λψ =⇒ (n+1) ∇z ψ = λ ∇z ψ

∆− ∆+ z z



(n) ϕ = µϕ =⇒ (n−1) ∇(n) ϕ = µ ∇(n) ϕ (C.4.8)

When λ, µ 6= 0, these maps are invertible. As a result, the spectra of non-vanishing eigen-
−
values of ∆+(n) and ∆(n+1) coincide and the one-to-one map between their eigenfunctions is
provided by the above relations. No such correspondence follows for the kernels of these
operators. Since the Laplace-Beltrami operators are positive, their kernels are related to the
kernels of the Cauchy-Riemann operators as follows,

Ker ∆+ (n)
(n) = Ker ∇z
Ker ∆− z
(n) = Ker ∇(n) (C.4.9)

In the subsequent sections, we shall relate the dimensions of these kernels to one another by
the Riemann-Roch and vanishing theorems.

C.5 Proof of the Riemann-Roch theorem

The kernel of ∇z(n) is the space of holomorphic differentials of rank n for n ≥ 1, and holomor-
phic vector fields of rank −n for n ≤ −1. In this formulation, the Riemann-Roch theorem
for a compact Riemann surface Σ may be stated as follows,

dim Ker ∇z(n+1) − dim Ker ∇z(n) = − n + 1

2
χ(Σ) = (2n + 1)(h − 1) (C.5.1)

where χ(Σ) = 2−2h is the Euler number of Σ. The Riemann-Roch theorem may equivalently
be recast in terms of Laplace-Beltrami operators,

dim Ker ∆− +
(n+1) − dim Ker ∆(n) = (2n + 1)(h − 1) (C.5.2)

316
To prove this result, we use the fact established earlier that the non-zero eigenvalues of ∆+
(n)
and ∆−(n+1 are identical to obtain the following relation,

−s∆−
   
−s∆+
dim Ker ∆−(n+1) − dim Ker ∆ +
(n) = Tr e (n+1) − Tr e (n) (C.5.3)

valid for all 0 < s ∈ R. In particular, the right side may be evaluated in the limit s → 0 where
the short-time asymptotics of the heat-kernel may be used to derive an explicit formula. The
short-time expansion, or Bott-Seeley expansion, is given as follows,
1 ± 3n
Z Z

−s∆±
 1
Tr f e (n) = dµg f + dµg Rg f + O(s) (C.5.4)
4πs Σ 12π Σ
Setting f = 1, substituting the expressions into (C.5.3), and using the integral expression
for the Euler number, we establish (C.5.2), which proves the Riemann-Roch theorem.

C.6 Proof of the vanishing theorem

In terms of the differential operators ∇z(n) and ∇z(n) the vanishing theorem read as follows,

dim Ker ∇z(n) = 0 g = 0 and n > 0

dim Ker ∇z(n) = 0 g ≥ 2 and n < 0 (C.6.1)

The proof of these theorems uses the following equations from differential geometry,

[∇m , ∇n ] tp = −tq Rq pmn

[∇m , ∇n ] tp = +tq Rp qmn (C.6.2)

where Rq pmn is the Riemann tensor, and tq is a rank one tensor, i.e. a one-form. Adapted
to the case of two dimensions, and working in complex coordinates, the formulas reduce to,

[∇z , ∇z̄ ] tz = −gzz̄ Rg tz (C.6.3)

where Rg is the Gaussian curvature (normalized to be one for the round sphere of unit
∗
radius). Formulas for tensors and spinors tz···z = t ∈ T(n,0) of arbitrary rank n ∈ Z/2 may be
derived by taking the tensor product and the square root, and we find,

[∇z , ∇z̄ ] t = −n gzz̄ Rg t (C.6.4)

We are now ready to prove the vanishing theorem. We use the following rearrangement
formula, obtained by integrating by parts (no boundary terms arise since Σ has no boundary),
Z Z !
dµg (gzz̄ )n−1 |∇z̄ t|2 = dµg (gzz̄ )n−1 |∇z̄ t̄|2 + n(gzz̄ )n Rg |t|2 (C.6.5)
Σ Σ

317
 • For the case g = 0, we may choose a metric of constant positive curvature Rg = 1 on
the sphere. Since we have n > 0 the right side of (C.6.5) is positive definite in t so that the
left side cannot vanish unless t = 0. As a result there are no holomorphic differential forms
of any positive weight n > 0 on the sphere which proves the result on the first line of (C.6.1).
• For the case g ≥ 2, we may choose a metric of constant negative curvature Rg = −1.
Since we have n < 0 the right side of (C.6.5) is again positive definite so that the left side
cannot vanish unless t = 0. As a result there are no holomorphic vector fields of any negative
weight n < 0 on a higher genus Riemann surface. In particular, this implies that there are no
conformal Killing vectors or conformal Killing spinors on a Riemann surface of genus g ≥ 2,
and thus no continuous symmetries.
• For the case g = 1, and a flat metric Rg = 0 on the torus, the relation (C.6.5) imposes
no restrictions.

C.7 The dimension of moduli space

The moduli space Mg of compact Riemann surfaces of arbitrary genus g is the space of
inequivalent complex structures or, equivalently, the space of inequivalent conformal struc-
tures. Complex structures and conformal structures may be parametrized with the help
√
of the Weyl-invariant tensor Jm n = g εmp gpn . Two tensors that are related by a diffeo-
morphsim of Σ are equivalent to one another, so that the moduli space is given by,

Mg = {J }/Diff(Σ) (C.7.1)

To compute the dimension of Mg , we evaluate the dimension of its tangent space at a given
complex structure with local complex coordinates z, z̄ in terms of which the metric is given
by g = 2gzz̄ |dz|2 , and the complex structure tensor J has the following components,

Jz z̄ = −Jz̄ z = 0 Jz z = −Jz̄ z̄ = i (C.7.2)

Next, we compute the variation of the complex structure tensor, as the metric g is varied,
leaving the system of complex coordinates z, z̄ unchanged. The variation of the metric is
used to first order,

δg = 2δgzz̄ |dz|2 + δgzz dz 2 + δgz̄z̄ dz̄ 2 (C.7.3)

The effect of the infinitesimal Weyl transformation δgzz̄ cancels out of the variation of J
since the complex structure is invariant under Weyl transformations. Thus, we obtain the
following variations, to first order in δgzz and δgz̄z̄ ,
∗
δJz z = 0 δJz z̄ = +i gzz̄ δgz̄z̄ ∈ T(1,−1)
∗
δJz̄ z̄ = 0 δJz̄ z = −i gz̄z δgzz ∈ T(−1,1) (C.7.4)

318
Moduli deformations, however, are variations of complex structure modulo diffeomorphisms.
The action of an infinitesimal diffeomorphism v m ∂m on the metric is given as follows,
δv gmn = ∇m vn + ∇n vm (C.7.5)
Under this variation, the complex structure transforms as follows,
δv Jz z̄ = −i ∇(1)
z v
z̄

(−1) z
δv Jz̄ z = +i ∇z̄ v (C.7.6)
Vectors in the tangent space Tg Mg to moduli space Mg at a given metric g, may be identified
with complex structure deformations that are in the orthogonal complement to the ranges
of these operators, evaluated at the metric g,
 †  †
(−1)
Tg Mg = Range ∇(1) z ⊕ Range ∇ z̄ (C.7.7)

But the orthogonal complement to the range of any operator is the kernel of the adjoint of
(1)
that operator. Since we have (∇z )† = −∇z(2) , we find,

Tg Mg = Ker∇z(2) ⊕ Ker∇z̄(−2) (C.7.8)

First of all, we note that the tangent space to Mg is split according to holomorphic and
anti-holomorphic directions, so that Mg is a complex manifold (actually, more precisely
a complex orbifold). The dimension of moduli space may be read off from the table of
dimensions given in the preceding subsection,

 0 g=0
dimC Mg = 1 g=1 (C.7.9)
3g − 3 g ≥ 2


The combination δJz̄ z and its complex conjugate are often referred to as Beltrami differen-
tials. The elements of the dual space Ker∇z(2) are the holomorphic quadratic differentials.

• Bibliographical notes
A summary of the ingredients of holomorphic line bundles on compact Riemann surfaces,
accessible to physicists, was given in the review paper on string perturbation theory [186]
and in the lecture notes [328]. A fundamental and useful reference is the book by Fay [329].
Helpful discussions of line bundles characterized by divisors may be found in the book by
Farkas and Kra [327], while the deformation theory of complex structures for Riemann
surfaces is treated quite explicitly in the book by Schiffer and Spencer [330]. More general
treatments of vector bundles on complex and Kähler manifolds may be found in the books
by Gunning [331] and Kodaira [332]. The Bott-Seeley expansion in the context of two-
dimensional Riemannian manifolds is discussed and proven in [333] and [186].

319
D Higher genus ϑ-functions and meromorphic forms
In this final appendix, we shall review the modular geometry of the Siegel half-space at
higher rank, Riemann ϑ-functions of higher rank, the embedding of higher-genus Riemann
surfaces into the Jacobian variety via the Abel map, and use these ingredients to construct
the prime form, the Szegö kernel, and other meromorphic differential forms on higher-genus
Riemann surfaces.

D.1 The Siegel half space

The rank g Siegel half space Hg may be defined as the space of complex g × g symmetric
matrices with positive definite imaginary part,
n o
g×g t
Hg = Ω ∈ C , Ω = Ω, Y = Im (Ω) > 0 (D.1.1)

Its dimension is dim Hg = 21 g(g + 1). Alternatively, Hg is given as the coset space
Hg = Sp(2g, R)/U (g) (D.1.2)
The group Sp(2g, R) acts on Ω by
Ω → Ω0 = (AΩ + B)(CΩ + D)−1 (D.1.3)
where the g × g real matrices A, B, C, D are given in terms of M ∈ Sp(2g, R) by,
   
A B t 0 −I
M= M JM = J J= (D.1.4)
C D I 0
The unitary subgroup U (g) is generated by setting C = −B and D = A subject to the
conditions At A + B t B = I and At B − B t A = 0. Since the group U (g) contains a U (1) factor,
Hg is a Kähler manifold, with the following Sp(2g, R)-invariant Kähler metric,
g
X
ds2g = (Y −1 )IJ dΩIK (Y −1 )KL dΩ̄JL (D.1.5)
I,J,K,L=1

The associated Laplace-Beltrami operator ∆ on scalar functions of Hg is given by,

g
X
∆= 4 YIJ YKL ∂ IK ∂¯JL (D.1.6)
I,J,K,L=1

where the derivatives with respect to the components of ΩIJ are defined as follows,
∂ 1 ∂
∂ II = ∂ IJ = for J 6= I (D.1.7)
∂ΩII 2 ∂ΩIJ
For genus one, H1 = H coincides with the Poincaré upper half-plane, ds21 with the Poincaré
metric on H1 , and ∆ with the Laplace-Beltrami operator on scalars functions on H1 .

320
D.2 The Riemann theta function
The Riemann ϑ-functions are complex-valued functions on Hg × Cg that generalize the
Jacobi ϑ-functions on H × C to arbitrary rank g. Just as for Jacobi ϑ-functions, Riemann ϑ-
functions may be considered for arbitrary complex characteristics δ = [δ 0 |δ 00 ] with δ 0 , δ 00 ∈ C2g .
In terms of local coordinates Ω on Hg and ζ = (ζ1 , · · · , ζg )t on Cg , the Riemann ϑ-function
for characteristic δ may be defined by the following absolutely convergent series,
X  
0 t 0 0 t 00
ϑ[δ](ζ|Ω) ≡ exp iπ(n + δ ) Ω(n + δ ) + 2πi(n + δ ) (ζ + δ ) . (D.2.1)
n∈Zg

Upon shifting by m, n ∈ Zg we have the periodicity relations,

t t 0 t 00
ϑ[δ](ζ + m + Ωn|Ω) = e−iπn Ωn−2πin (ζ+δ )+2πim δ ϑ[δ](ζ|Ω)
t 0
ϑ[δ 0 + n|δ 00 + m](ζ|Ω) = e2πim δ ϑ[δ 0 |δ 00 ](ζ|Ω) (D.2.2)

The ϑ-function without characteristics is often denoted by ϑ(ζ|Ω) = ϑ[0](ζ|Ω).

We shall focus on half-integer characteristics δ with values δ 0 , δ 00 ∈ (Z/2Z)g corresponding
to a spin structure. The parity 4δ 0 ·δ 00 (mod 2) of a ϑ-function with half-integer characteristics
depends on δ and, at a point (ζ|Ω) where ϑ[δ](ζ|Ω) 6= 0, is defined by,
0 00
ϑ[δ](−ζ|Ω) = (−1)4δ ·δ ϑ[δ](ζ|Ω) (D.2.3)

The spin structure δ is referred to as an even or odd spin structure according to whether the
integer 4δ 0 · δ 00 is even or odd.

D.2.1 Modular transformations

The discrete (or arithmetic) subgroup Sp(2g, Z) of Sp(2g, R) is the group of modular trans-
formations, whose action on the Siegel upper half-space Hg is given by (D.1.3), but now
with M ∈ Sp(2g, Z) with the entries of the matrices A, B, C, D being integers. Under a
modular transformation M ∈ Sp(2g, Z), a half-integer characteristic δ = [δ 0 | δ 00 ] transforms
into another half-integer characteristic δ̃ = [δ̃ 0 | δ̃ 00 ] given as follows,
 0   0    
δ̃ D −C δ 1 CDt A B
= + diag M= (D.2.4)
δ̃ 00 −B A δ 00 2 AB t C D

Under joint modular transformations of Ω and δ, the ϑ-function transforms as follows,

  1
ϑ[δ̃] {(CΩ + D)−1 }t ζ|Ω̃ = ε(δ, M ) det (CΩ + D) 2 ϑ[δ](ζ|Ω) (D.2.5)

where ε(δ, M )8 = 1 and the precise dependence of ε(δ, M ) on its arguments is given in
[329, 334].

321
D.2.2 Riemann relations on ϑ-functions
For each spin structure λ, there exists a Riemann relation which may be expressed as a
quadrilinear sum over all spin structures κ,
X
hκ|λiϑ[κ](ζ1 )ϑ[κ](ζ2 )ϑ[κ](ζ3 )ϑ[κ](ζ4 ) = 4 ϑ[λ](ζ10 )ϑ[λ](ζ20 )ϑ[λ](ζ30 )ϑ[λ](ζ40 ) (D.2.6)
κ

The signature symbol for the pairing of two spin structures is defined by,
0 00 −κ00 ·λ0 )
hκ|λi = e4πi(κ ·λ (D.2.7)

and the relation between the vectors ζ and ζ 0 is given in terms of a matrix Λ,
 0    
ζ1 ζ1 1 1 1 1
ζ20  1
 0  = Λ ζ2  1 1 −1 −1
  
Λ = (D.2.8)
ζ3  ζ3  2 1 −1 1 −1
0
ζ4 ζ4 1 −1 −1 1

which satisfies Λ2 = I. In the special case where ζ = ζ 0 = 0, only even spin structures κ = δ
contribute to the sum and we recover a Riemann identity for each odd spin structure λ = ν,
X
hν|δiϑ[δ]4 (0, Ω) = 0 (D.2.9)
δ

More generally, if ζ1 + ζ2 + ζ3 + ζ4 = 0 then ζ10 = 0, the right side of (D.2.6) vanishes for any
odd spin structure λ, and the Riemann relations become,
X
hκ|λiϑ[κ](ζ1 )ϑ[κ](ζ2 )ϑ[κ](ζ3 )ϑ[κ](ζ4 ) = 0 (D.2.10)
κ

For genus one, there is only a single odd spin structure and the Riemann relation reduces to
the famous Jacobi identity ϑ42 − ϑ43 + ϑ44 = 0.

D.3 The Jacobian, Abel map, and Riemann vanishing theorem

We recall from appendix B that a canonical basis may be chosen for the holomorphic (1, 0)-
forms ωI with I = 1, · · · , g on a compact Riemann surface of genus g by normalizing the
integrals of ωI on the A-cycles of a canonical basis (A, B) of generators for the homology
group H1 (Σ, Z). The integrals on the B-cycles then give the period matrix Ω of the Riemann
surface Σ in the homology basis (A, B),
I I
ωJ = δIJ ωJ = ΩIJ (D.3.1)
AI BI

322
Under a symplectic change of canonical homology basis, the holomorphic (1, 0)-forms ω and
period matrix Ω transform as given in (B.1.13). In view of the Riemann relations, we have
Ωt = Ω and Im (Ω) > 0, so that the period matrix of a compact Riemann surface takes values
in the Siegel half space Hg . The set of A and B periods defines a lattice Λ = Zg + ΩZg in
Cg , whose quotient is an Abelian variety referred to as the the Jacobian variety,
J(Σ) = Cg /{Zg + ΩZg } (D.3.2)
In other words, the Jacobian is a g-dimensional complex torus. For genus g = 1, this torus
gives an equivalent representation of the Riemann surface itself. For higher genus g ≥ 2,
this simple correspondence no longer holds, but is replaced by a more subtle identification
produced by the Abel map.
Given a base point z0 ∈ Σ, the Abel map sends a divisor D of n points zi ∈ Σ with
weights qi ∈ Z for i = 1, · · · , n, formally denoted by D = q1 z1 + · · · qn zn , into Cg by,
Xn Z zi
D = q1 z1 + · · · + qn zn → qi (ω1 , · · · , ωg ) (D.3.3)
i=1 z0

where the g-tuple (ω1 , · · · , ωg ) stands for the vector of holomorphic

Rz (1, 0)-forms ωI . The
Abel map into Cg is multiple valued because each integral z0i ωI is multiple-valued in C, but
the Abel map is single valued as a map into the Jacobian J(Σ).

Theorem D.1 The Riemann vanishing Theorem states that ϑ(ζ|Ω) = 0 if and only if there
exist g − 1 points p1 , · · · , pg−1 ∈ Σ such that,
g−1 Z zi
X
ζ = p1 + · · · pg−1 − ∆(z0 ) ζI = ωI − ∆I (z0 ) (D.3.4)
i=1 z0

where ∆I (z0 ) are the components of the Riemann vector for base-point z0 , given by,
XI Z z
1 1
∆I (z0 ) = − − ΩII + ωJ (z) ωI (D.3.5)
2 2 J6=I AJ z0

The combination ζ is independent of the base-point.

The proof of the Riemann vanishing theorem may be found in any book on ϑ-functions,
or in the physics literature in [186]. For the special case of genus g = 1, the set of points
pi is empty, and the Riemann vector reduces to ∆ = − 21 − 12 τ which merely states that
ϑ( 12 + 12 τ |τ ) = 0, a fact we established long ago.
An important application of the Riemann vanishing theorem is to the existence of the
following holomorphic (1, 0)-forms, defined for any odd spin structure ν by
X
ων (z) = ωI (z)∂ I ϑ[ν](0|Ω) (D.3.6)
I

323
Its 2(h − 1) zeros are all double zeros at points pi such that ζI = νI in (D.3.4). Therefore,
ων (z) admits a holomorphic square root hν (z). Since the square of hν is ων (z), which is a
section of the canonical bundle, hν is a section of a spin bundle with spin structure ν. For
each odd spin structure ν, the holomorphic ( 21 , 0)-form hν is unique up to a sign.

D.4 The prime form

To construct the prime form we consider an arbitrary odd spin structure ν. Its associated
holomorphic section hν (z) of the spin bundle with spin structure ν has g − 1 zeros. The
prime form is a (− 21 , 0)-form in both variables z and w, defined by,

ϑ[ν](z − w|Ω)
E(z, w|Ω) = (D.4.1)
hν (z)hν (w)

The argument z − w of the ϑ-function stands for the Abel map of (D.3.3) with divisor
D = z − w. The form E(z, w|Ω) defined this way is independent of ν. It is holomorphic in
z and w and has a unique simple zero at z = w. It is single valued when z is moved around
AI cycles, but has non-trivial monodromy when z → z + BI is moved around BI cycles,

E(z + AI , w|Ω) = E(z, w|Ω)

 Z z 
E(z + BI , w|Ω) = − exp −iπΩII + 2πi ωI E(z, w|Ω) (D.4.2)
w

Therefore, it is properly defined on the simply connected covering space of Σ × Σ, or it may

be defined in a fundamental domain for Σ in the complex plane.

D.5 Holomorphic differentials

Combining the Riemann-Roch and vanishing theorems, the dimension of the space of holo-
(n)
morphic (n, 0)-forms is denoted by Υ(n) = dim Ker ∂z̄ and is given by,


 0 n < 0 , and n = 1/2 even spin structure
1 n = 0 , and n = 1/2 odd spin structure

Υ(n) = (D.5.1)

 g n=1
(2n − 1)(g − 1) n ≥ 3/2, and g ≥ 2


The dimensions listed for n = 1/2 are for generic moduli and are valid for exceptional moduli
(n)
mod 2. A basis of holomorphic differentials is denoted by φa , a = 1, · · · , Υ(n). They are
holomorphic sections of the line bundles K n , the n-th power of the canonical bundle K, for
which the number of zeros is given by,

# zeros φa(n) = c1 (K n ) = 2n(g − 1) (D.5.2)

324
For a given a, there are Υ(n) − 1 zeros fixed to be zb with b 6= a, and another g zeros due to
the Riemann vanishing theorem, giving a total of 2n(g − 1), in agreement with the value of
the first Chern class. For n = 0, they are constants, for n = 1/2 and ν odd they are denoted
by hν (z), while for n = 1 they are the holomorphic Abelian differentials ωI , for I = 1, · · · , g.
(n)
Given any set of Υ(n) points z1 , · · · , zΥ(n) on the surface, we may choose a basis φ̂a for
the holomorphic n-differentials normalized at the points zb by,
φ̂(n) b
a (zb ) = δa (D.5.3)
The holomorphic differentials with this normalization may be exhibited explicitly in terms
of the prime form E(z, w), the h/2 differential σ(z) and ϑ-functions. For n ≥ 3/2, we have,
Q 2n−1
ϑ[δ](z − za + zb − (2n − 1)∆) b6=a E(z, zb ) σ(z)
P 
(n)
φ̂a (z) = P Q (D.5.4)
ϑ[δ]( zb − (2n − 1)∆) b6=a E(za , zb ) σ(za )

Here, σ(z) is a holomorphic ( h2 , 0)-form without zeros or poles defined, up to a constant, by

the following ratio,
P h
σ(z) ϑ(z − pi + ∆) Y E(w, pi )
= P (D.5.5)
σ(w) ϑ(w − pi + ∆) i=1 E(z, pi )
where pi , i = 1, · · · , h are arbitrary points on the surface. Note that σ(z) is single valued
around AI cycles but multivalued around BI cycles in the following way

σ(z + BI ) = σ(z) exp −iπ(h − 1)ΩII + 2πi∆Iz (D.5.6)
(n)
Besides the Υ(n) − 1 zeros zb , b 6= a, the differential φ̂a (z) has h additional zeros. The
(n)
differential φ̂a (z) is an (n, 0)-form in z, a (−n, 0)-form in za and a (0, 0)-form zb with b 6= a.
For n = 1, we have,
P
(1) ϑ(z − za + zb − w0 − ∆) Y E(z, zb ) E(za , w0 ) σ(z)
φ̂a (z) = P . (D.5.7)
ϑ( zb − w0 − ∆) E(z, w) b6=a E(za , zb ) E(z, w0 ) σ(za )

D.6 Meromorphic differentials

Meromorphic differentials play the role of Green functions for the Cauchy-Riemann operators
acting on holomorphic line bundles. The Green function Gn (z, w) = Gn (z, w; z1 , · · · , zΥ(n) )
(n)
for the operator ∂z̄ is a meromorphic (n, 0) form in z and (1 − n, 0)-form in w. When
n ≥ 3/2 for general spin structure δ and n = 1/2 for even spin structure δ, the Green
function Gn may be defined by the following relations,
(n)
∂z̄ Gn (z, w) = +2πδ(z, w) (D.6.1)
Υ(n)
(1−n)
X
∂w̄ Gn (z, w) = −2πδ(z, w) + 2π φ̂(n)
a (z)δ(w, za ) (D.6.2)
a=1

325
 (n)
The properly normalized holomorphic n-differentials φ̂a are defined in (D.5.3) and (D.5.4).
(1−n)
Setting z = za , we have ∂w̄ Gn (za , w) = 0, so that Gn (za , w) = 0. Explicit expressions for
the Green’s function are
P
Q 2n−1
ϑ[δ] z − w + zb − (2n − 1)∆

a E(z, za ) σ(z)
Gn (z, w) = P
Q (D.6.3)
ϑ[δ] zb − (2n − 1)∆ E(z, w) a E(w, za ) σ(w)

For n = 1/2, this reduces to the standard form of the Szegö kernel, usually denoted by


ϑ[δ] z − w
Sδ (z, w) =
(D.6.4)
ϑ[δ] 0 E(z, w)

For n = 1, the Green’s function G1 (z, w) = G1 (z, w; z1 , · · · , zh , w0 ) is the Abelian differential

of the third kind, satisfying
(1)
∂z̄ G1 (z, w) = +2πδ(z, w) − 2πδ(z, w0 ) (D.6.5)
Xh
(0)
∂w̄ G1 (z, w) = −2πδ(z, w) + 2π φ̂(1)
a (z) δ(w, za ) (D.6.6)
a=1

and explicitly given by the following expression

P
Q
ϑ z − w − w0 + zb − ∆ E(z, za )E(w, w0 ) σ(z)
G1 (z, w) = P
Qa . (D.6.7)
ϑ − w0 + zb − ∆ E(z, w) a E(w, za )E(z, w0 ) σ(w)

The combination ∂z ∂w ln E(z, w) is a meromorphic differential (Abelian of the second kind)

with a single double pole at z = w. Its integrals around homology cycles are given by
I
dz∂z ∂w ln E(z, w) = 0
AI
I
dz∂z ∂w ln E(z, w) = 2πiωI (w) (D.6.8)
BI

and will be of use throughout.

D.7 The bc system

A useful unification of the formulas for holomorphic and meromorphic forms is provided by
the conformal field theory correlators of anti-commuting fields b and c of weights (n, 0) and
(1 − n, 0). This quantum field theory was already introduced in section 5.2 for a Riemann
surface of genus 1, namely a torus. Here, we shall generalize the correlators to arbitrary
genus g ≥ 2. Since the roles of b and c are swapped by letting n → 1 − n, we shall restrict
to considering the cases n ≥ 21 without loss of generality.

326
 As in the case of the torus, the fields b, c are locally holomorphic,

∂z̄ b = ∂z̄ c = 0 (D.7.1)

However, as b and c fields approach one another poles may develop. The poles may be
represented schematically by the singular contribution to the operator product expansion,
given by a simple pole with unit residue,
1
b(z) c(w) ∼ (D.7.2)
z−w
All poles must arise from these operator coincidences. In view of the anti-commuting nature
of the fields, the operator product of two like fields produces a simple zero,

b(z1 )b(z2 ) ∼ (z1 − z2 ) ∂b(z2 ) b(z2 )

c(w1 )c(w2 ) ∼ (w1 − w2 ) ∂c(w2 ) c(w2 ) (D.7.3)

In the case of the torus we used Fourier analysis to solve for the fields and their correlators.
Since higher-genus Riemann surfaces have no continuous isometries, such as translation
symmetry, Fourier analysis is no longer applicable.
Now consider the correlator of an arbitrary number of the b and c fields of weight (n, 0)
and (1 − n, 0) respectively,

hb(z1 ) · · · b(zN ) c(w1 ) · · · c(wM )i (D.7.4)

Viewed as a function of z1 , the correlator has simple zeros at z2 , · · · , zN and simple poles
at w1 , · · · , wM . But since all poles arise from operator coincidences, and no other poles can
appear, the number of poles in z1 is exactly M , and thus the n-form must have M +2n(g −1)
zeros. Now N −1 of these zeros are specified by the positions of the points z2 , · · · , zN , leaving
an extra g zeros due to the Riemann vanishing theorem, for a total of N −1+g zeros. Equating
the number of zeros computed in these two different ways gives,

N = M + (2n − 1)(g − 1) (D.7.5)

The correlator must vanish unless the number of b and c fields satisfies this relation.

1
D.7.1 The case n = 2
with even spin structure
In the case n = 21 with even spin structure δ and generic moduli, neither the b nor the c field
has zero modes. By using Wick contractions, the correlator is given by,

hb(z1 ) · · · b(zM ) c(wM ) · · · c(w1 )i = det Sδ (zi , wj ) (D.7.6)

327
where Sδ is the Szegö kernel for even spin structure δ, the determinant is taken of the M ×M
matrix S(zi , wj ), and the numbering of the points wj has been chosen in descending order
so that the sign multiplying the determinant on the right side is positive. An alternative
formula is obtained by matching poles and zeros in each variable, and we obtain,
Q
ϑ[δ](D|Ω) i<j E(zi , zj )E(wi , wj )
hb(z1 ) · · · b(zM ) c(wM ) · · · c(w1 )i = Q (D.7.7)
ϑ[δ](0|Ω) i,j E(zi , wj )

where the divisor is D = (z1 − w1 ) + · · · + (zM − wM ). This formula is often referred to as

the bosonized correlator as it may be obtained by representing the spin- 12 fermion fields b, c
in terms of the chiral half of a complex scalar field.

1
D.7.2 The case n = 2
with odd spin structure
In the case n = 12 with odd spin structure ν and generic moduli, there is a single zero mode
hν of the Dirac equation. Clearly, the two-point correlator hb(z)c(w)i must be saturated by
the zero modes and be proportional to hν (z)hν (w). For higher point functions, we need a
Green function. The Green function may be defined by an equation that is orthogonal to
the Dirac zero mode. Such an equation is not unique, but different definitions will produce
the same correlator. Preserving meromorphicity, we choose the following definition,
hν (w)
∂z̄ Sν (z, w) = 2πδ(z, w) − 2πδ(z, w0 ) (D.7.8)
hν (w0 )
which is orthogonal to hν (z) by construction. The general correlator is then given by,
M
X 
hb(z1 ) · · · b(zM ) c(wM ) · · · c(w1 )i = (−)i+j hb(zi )c(wj )i det Sν (zk , zl ) k6=i, (D.7.9)

i,j=1 l6=j

The constant of proportionality in hb(z)c(w)i ∼ hν (z)hν (w) is given by the functional deter-
minant of the Dirac operator on n = 21 sections with odd spin structure ν.

D.7.3 The case n = 1

In any non-vanishing correlator for n = 1, the b field has g zero modes, while the c field has
one zero mode, namely the constants. We shall normalize the correlator so that the simplest
non-zero amplitude is given by,
G0 (z1 , · · · , zg ; w) = hb(z1 ) · · · b(zg )c(w)i = det ωI (zi ) (D.7.10)
where the determinant is taken of the g × g matrix with entries ωI (zi ) for i, I = 1, · · · , g.
The correlator with an arbitrary number of b, c fields subject to the condition (D.7.5),
GM (z1 , · · · wM +1 ) = hb(z1 ) · · · b(zg+M ) c(w1 ) · · · c(wM +1 )i (D.7.11)

328
with this normalization is given by
Q Q Q
ϑ (D) i<j E(zi , zj ) a<b E(wa , wb ) i σ(zi )
GM (z1 , · · · , wM +1 ) = Q Q (D.7.12)
Z3 i,a E(zi , wa ) a σ(wa )

where i, j = 1, · · · , g + M and a, b = 1, · · · , M + 1 and D is defined by

g+M M +1
X X
D= zi − wa − ∆ (D.7.13)
i=1 a=1

The factor Z is the chiral boson partition function determined by setting M = 0,

Q Q
ϑ (p1 + · · · pg − q − ∆) i<jE(pi , pj ) i σ(pi )
det ωI (pi ) = Q (D.7.14)
Z3 i E(pi , q)σ(q)

for arbitrary points q and pi with i = 1, · · · , g. With this normalization the correlator
transforms under modular transformations by,

G̃M (z1 , · · · , wM +1 ) = det (CΩ + D)−1 GM (z1 , · · · , wM +1 ) (D.7.15)

3
D.7.4 The case n ≥ 2

For n ≥ 32 , the b field has Υ = (2n − 1)(g − 1) zero modes, while the c field has no zero
modes. We have the following bosonization formula,
Q Q Q
ϑ[δ](ζ|Ω) a<b E(za , zb ) i<j E(wi , wj ) a σ(za )
hb(z1 ) · · · b(zM +Υ ) c(w1 ) · · · c(wM )i = Q Q
Z i,a E(za , wi ) i σ(wi )
(D.7.16)

where a, b = 1, · · · , M + Υ, while i, j = 1, · · · , M and ζ is given by,

M
X +Υ M
X
ζ= za − wi − (2n − 1)∆ (D.7.17)
a=1 i=1

In the special case where M = 0, we recover a purely holomorphic correlator, which is

entirely saturated by the zero modes of the field b,

ϑ[δ](ζ|Ω) Y Y
hb(z1 ) · · · b(zΥ )i = E(za , zb ) σ(za ) (D.7.18)
Z a<b a

329
• Bibliographical notes
Classic introductions to Siegel modular forms and higher rank modular geometry may be
found in the books by Siegel [335, 336], Igusa [334], and Klingen [337]. Useful lecture notes
on higher rank modular forms are those by van der Geer [338]. The connection between
higher rank ϑ-functions and Riemann surfaces is spelled out in the classic book by Fay [329].
The connection between higher rank ϑ-functions and non-linear equations may be found in
the survey article by Dubrovin [339]. The solution of the bc quantum system in terms of
ϑ-functions, using the work of [329], was given in [58].

330
References
[1] A. Erdèlyi, ed., Higher transcendental functions, vol. 2 of Bateman Manuscript
Project, Robert E. Krieger Publishing Company (1981).

[2] A. Weil, Elliptic functions according to Eisenstein and Kronecker, vol. 88, Springer
Science & Business Media (1999).

[3] E.T. Whittaker and G.N. Watson, A course of modern analysis, Cambridge
University press (1920).

[4] D. Mumford, Tata lectures on Theta. I (modern Birkhäuser classics), Birkhäuser

Boston Incorporated (2007).

[5] S. Lang, Elliptic functions, in Elliptic functions, pp. 5–21, Springer (1987).

[6] F. Oberhettinger and W. Magnus, Anwendung der elliptischen Funktionen in Physik

und Technik, vol. 55, Springer-Verlag (2013).

[7] G. Köhler, Eta products and theta series identities, Springer Monographs in
Mathematics, Springer Science & Business Media (2011).

[8] T.A. Apostol, Modular Functions and Dirichlet Series in Number Theory, Graduate
Texts in Mathematics, number 41, Springer Verlag (1976).

[9] S. Lang, Introduction to Modular Forms, vol. 222 of Grundlehren der mathematischen
Wissenschaften, Springer-Verlag (1976).

[10] D. Zagier, Elliptic modular forms and their applications, in The 1-2-3 of modular
forms, pp. 1–103, Springer (2008).

[11] G. Shimura, Introduction to the arithmetic theory of automorphic functions, vol. 1,

Princeton university press (1971).

[12] N.I. Koblitz, Introduction to elliptic curves and modular forms, vol. 97, Springer
Science & Business Media (2012).

[13] F. Diamond and J.M. Shurman, A first course in modular forms, vol. 228 of
Graduate Texts in Mathematics, Springer (2005).

[14] A. Erdèlyi, ed., Higher transcendental functions, vol. 3 of Bateman Manuscript

Project, Robert E. Krieger Publishing Company (1981).

[15] H. Iwaniec, Topics in classical automorphic forms, vol. 17 of Graduate Studies in

Mathematics, American Mathematical Soc. (1997).

331
[16] D. Bump, Automorphic Forms and Representations, vol. 55 of Cambridge Studies in
Advanced Mathematics, Cambridge University Press (1998).

[17] K. Ono, The Web of Modularity: Arithmetic of the Coefficients of Modular Forms
and q-series, vol. 102 of CBMS, American Mathematical Society (2004).

[18] S. Lang, SL(2,R), vol. 105, Springer Science & Business Media (2012).

[19] T. Kubota, Elementary theory of Eisenstein series, Halsted Press (1973).

[20] A. Terras, Harmonic analysis on symmetric spaces and applications I, Springer

Science & Business Media (1985).

[21] A. Terras, Harmonic analysis on symmetric spaces and applications II, Springer
Science & Business Media (1988).

[22] H. Iwaniec, Spectral methods of automorphic forms, vol. 53 of Graduate Studies in

Mathematics, American Mathematical Soc. (2002).

[23] H. Iwaniec and E. Kowalski, Analytic number theory, vol. 53, American
Mathematical Soc. (2021).

[24] I. Gel’land, M. Graev and I. Pyatetskii-Shapiro, Representation theory and

automorphic functions, (Transl. from Russian) Philadelphia: Saunders (1969) .

[25] F. Brown, A class of non-holomorphic modular forms iii: real analytic cusp forms for
sl2 (Z), arXiv preprint arXiv:1710.07912 (2017) .

[26] N. Benjamin, S. Collier, A.L. Fitzpatrick, A. Maloney and E. Perlmutter, Harmonic

analysis of 2d CFT partition functions, JHEP 09 (2021) 174 [2107.10744].

[27] S. Collier and E. Perlmutter, Harnessing S-Duality in N = 4 SYM & Supergravity as

SL(2, Z)-Averaged Strings, 2201.05093.

[28] D. Goldfeld, Automorphic Forms and L-functions for the Group GL(2, R), vol. 99 of
Cambridge Studies in Advanced Mathematics, Cambridge University Press (2006).

[29] M. Viazovska, The sphere packing problem in dimension 8, Annals of Mathematics

185 (2017) 991 .

[30] H. Cohn, A. Kumar, S. Miller, D. Radchenko and M. Viazovska, The sphere packing
problem in dimension 24, Annals of Mathematics 185 (2017) 1017 .

[31] D. Zagier, Ramanujan’s mock theta functions and their applications, Séminaire
BOURBAKI (2007) .

332
[32] A. Dabholkar, S. Murthy and D. Zagier, Quantum Black Holes, Wall Crossing, and
Mock Modular Forms, 1208.4074.

[33] V. Anagiannis and M.C.N. Cheng, TASI Lectures on Moonshine, PoS TASI2017
(2018) 010 [1807.00723].

[34] T. Eguchi, H. Ooguri, A. Taormina and S.-K. Yang, Superconformal Algebras and
String Compactification on Manifolds with SU(N) Holonomy, Nucl. Phys. B 315
(1989) 193.

[35] T. Eguchi and K. Hikami, Superconformal Algebras and Mock Theta Functions 2.
Rademacher Expansion for K3 Surface, Commun. Num. Theor. Phys. 3 (2009) 531
[0904.0911].

[36] T. Eguchi, H. Ooguri and Y. Tachikawa, Notes on the K3 Surface and the Mathieu
group M24 , Exper. Math. 20 (2011) 91 [1004.0956].

[37] T. Gannon, Monstrous moonshine: The First twenty five years, math/0402345.

[38] T. Gannon, Moonshine beyond the Monster: The bridge connecting algebra, modular
forms and physics, Cambridge University Press (2006).

[39] S.M. Harrison, J.A. Harvey and N.M. Paquette, Snowmass White Paper: Moonshine,
2201.13321.

[40] C. Vafa and E. Witten, A Strong coupling test of S duality, Nucl. Phys. B 431 (1994)
3 [hep-th/9408074].

[41] D. Zagier, Vassiliev invariants and a strange identity related to the dedekind
eta-function, Topology, Elsevier 40 (2001) 945.

[42] D. Zagier, Quantum modular forms, Quanta of maths, American Mathematical

Society Providence 11 (2010) 659.

[43] K. Ono, Harmonic maass forms, mock modular forms, and quantum modular forms,
Notes (2013) 347.

[44] H. Rademacher, Topics in Analytic Number Theory, vol. 169 of Die Grundlehren der
mathematischen Wissenschaften, Springer (1970).

[45] P. Francesco, P. Mathieu and D. Sénéchal, Conformal field theory, Springer Science &
Business Media (2012).

[46] P. Ginsparg, Applied conformal field theory, arXiv preprint hep-th/9108028 (1988) .

333
[47] S. Fubini, A.J. Hanson and R. Jackiw, New approach to field theory, Phys. Rev. D 7
(1973) 1732.

[48] A.A. Belavin, A.M. Polyakov and A.B. Zamolodchikov, Infinite Conformal Symmetry
in Two-Dimensional Quantum Field Theory, Nucl. Phys. B 241 (1984) 333.

[49] D. Friedan, E.J. Martinec and S.H. Shenker, Conformal Invariance, Supersymmetry
and String Theory, Nucl. Phys. B 271 (1986) 93.

[50] J. Polchinski, Evaluation of the One Loop String Path Integral, Commun. Math.
Phys. 104 (1986) 37.

[51] C.L. Siegel, On advanced analytic number theory, Tata Institute of Fundamental
Research, 1961, .

[52] D.B. Ray and I.M. Singer, R-torsion and the laplacian on riemannian manifolds,
Advances in Mathematics 7 (1971) 145.

[53] D.A. Hejhal, The Selberg Trace Formula for SL (2, R): Volumes 1 and 2, vol. 1001,
Springer (1983).

[54] D. Fried, Analytic torsion and closed geodesics on hyperbolic manifolds, Inventiones
mathematicae 84 (1986) 523.

[55] E. D’Hoker and D.H. Phong, On Determinants of Laplacians on Riemann Surfaces,

Commun. Math. Phys. 104 (1986) 537.

[56] P. Sarnak, Determinants of laplacians, Communications in mathematical physics 110

(1987) 113.

[57] J. Jorgenson and S. Lang, Basic Analysis of Regularized Series and Products, Lecture
Notes in Mathematics 1564 (1993) .

[58] E.P. Verlinde and H.L. Verlinde, Chiral Bosonization, Determinants and the String
Partition Function, Nucl. Phys. B 288 (1987) 357.

[59] L. Alvarez-Gaume, J.B. Bost, G.W. Moore, P.C. Nelson and C. Vafa, Bosonization
on Higher Genus Riemann Surfaces, Commun. Math. Phys. 112 (1987) 503.

[60] A.A. Belavin and V.G. Knizhnik, Algebraic Geometry and the Geometry of Quantum
Strings, Phys. Lett. B 168 (1986) 201.

[61] J.B. Bost and T. Jolicoeur, A Holomorphy Property and Critical Dimension in String
Theory From an Index Theorem, Phys. Lett. B 174 (1986) 273.

334
[62] T. Miyake, Modular forms, Springer Science & Business Media (2006).

[63] S. Developers, Sagemath, the sage mathematics software system,

https://www.sagemath.org.

[64] J. Coates and S. Yau, eds., Elliptic Curves, Modular Forms and Fermat’s Last
Theorem, International Press (1997).

[65] F. Calegari, V. Dimitrov and Y. Tang, The Unbounded Denominators Conjecture,

2109.09040.

[66] M. Kaneko and M. Koike, On modular forms arising from a differential equation of
hypergeometric type, The Ramanujan Journal 7 (2003) 145.

[67] P. Bantay and T. Gannon, Vector-valued modular functions for the modular group
and the hypergeometric equation, arXiv preprint arXiv:0705.2467 (2007) .

[68] C. Franc and G. Mason, Hypergeometric series, modular linear differential equations
and vector-valued modular forms, The Ramanujan Journal 41 (2016) 233.

[69] R. Gunning, The eichler cohomology groups and automorphic forms, Transactions of
the American Mathematical Society 100 (1961) 44.

[70] A. Selberg, On the estimation of fourier coefficients of modular forms, in Proc. Symp.
Pure Math., vol. 8, pp. 1–15, AMS, 1965.

[71] T. Gannon, The theory of vector-modular forms for the modular group, arXiv
preprint arXiv:1310.4458 (2013) .

[72] D. Zagier, Modular forms and differential operators, Proc. Ind. Acad. Sciences 104
(1994) 57.

[73] A.O.L. Atkin and H.P.F. Swinnerton-Dyer, Modular forms on noncongruence

subgroups, Combinatorics 19 (1971) 1.

[74] G. Mason, On the fourier coefficients of 2-dimensional vector-valued modular forms,

Proceedings of the American Mathematical Society 140 (2012) 1921.

[75] C. Franc and G. Mason, Fourier coefficients of vector-valued modular forms of

dimension 2, Canadian Mathematical Bulletin 57 (2014) 485.

[76] C. Marks, Fourier coefficients of three-dimensional vector-valued modular forms,

arXiv preprint arXiv:1201.5165 (2012) .

335
[77] C. Franc and G. Mason, Three-dimensional imprimitive representations of the
modular group and their associated modular forms, Journal of Number Theory 160
(2016) 186.

[78] C. Franc and G. Mason, Constructions of vector-valued modular forms of rank four
and level one, International Journal of Number Theory 16 (2020) 1111.

[79] S.D. Mathur, S. Mukhi and A. Sen, On the Classification of Rational Conformal
Field Theories, Phys. Lett. B 213 (1988) 303.

[80] S.D. Mathur, S. Mukhi and A. Sen, Reconstruction of Conformal Field Theories
From Modular Geometry on the Torus, Nucl. Phys. B 318 (1989) 483.

[81] E.B. Kiritsis, Fuchsian Differential Equations for Characters on the Torus: A
Classification, Nucl. Phys. B 324 (1989) 475.

[82] S.G. Naculich, Differential equations for rational conformal characters, Nucl. Phys. B
323 (1989) 423.

[83] H.R. Hampapura and S. Mukhi, On 2d Conformal Field Theories with Two
Characters, JHEP 01 (2016) 005 [1510.04478].

[84] H.R. Hampapura and S. Mukhi, Two-dimensional RCFT’s without Kac-Moody

symmetry, JHEP 07 (2016) 138 [1605.03314].

[85] A.R. Chandra and S. Mukhi, Towards a Classification of Two-Character Rational

Conformal Field Theories, JHEP 04 (2019) 153 [1810.09472].

[86] S. Mukhi, R. Poddar and P. Singh, Rational CFT with three characters: the
quasi-character approach, JHEP 05 (2020) 003 [2002.01949].

[87] S. Mukhi, Classification of RCFT from Holomorphic Modular Bootstrap: A Status

Report, in Pollica Summer Workshop 2019: Mathematical and Geometric Tools for
Conformal Field Theories, 10, 2019 [1910.02973].

[88] J. Kaidi, Y.-H. Lin and J. Parra-Martinez, Holomorphic modular bootstrap revisited,
JHEP 12 (2021) 151 [2107.13557].

[89] J.-B. Bae, Z. Duan, K. Lee, S. Lee and M. Sarkis, Fermionic Rational Conformal
Field Theories and Modular Linear Differential Equations, 2010.12392.

[90] J.-B. Bae, Z. Duan, K. Lee, S. Lee and M. Sarkis, Bootstrapping Fermionic Rational
CFTs with Three Characters, 2108.01647.

336
 [91] M.B. Green, J.G. Russo and P. Vanhove, Low energy expansion of the four-particle
genus-one amplitude in type II superstring theory, JHEP 02 (2008) 020 [0801.0322].

[92] E. D’Hoker, M.B. Green and P. Vanhove, On the modular structure of the genus-one
Type II superstring low energy expansion, JHEP 08 (2015) 041 [1502.06698].

[93] E. D’Hoker, M.B. Green and P. Vanhove, Proof of a modular relation between 1-, 2-
and 3-loop Feynman diagrams on a torus, J. Number Theor. 196 (2019) 381
[1509.00363].

[94] F. Zerbini, Single-valued multiple zeta values in genus 1 superstring amplitudes,

Commun. Num. Theor. Phys. 10 (2016) 703 [1512.05689].

[95] E. D’Hoker and W. Duke, Fourier series of modular graph functions, Journal of
Number Theory 192 (2018) 1.

[96] E. D’Hoker and J. Kaidi, Modular graph functions and odd cuspidal functions.
Fourier and Poincaré series, JHEP 04 (2019) 136 [1902.04180].

[97] E. D’Hoker, M.B. Green, O. Gürdogan and P. Vanhove, Modular Graph Functions,
Commun. Num. Theor. Phys. 11 (2017) 165 [1512.06779].

[98] D. Zagier, The Bloch-Wigner-Ramakrishnan polylogarithm function, Math. Annalen

286 (1990) 613.

[99] F. Brown and A. Levin, Multiple elliptic polylogarithms, arXiv preprint

arXiv:1110.6917 (2011) .

[100] B. Enriquez, Analogues elliptiques des nombres multizétas, Bull. Soc. Math. France
144 (2016) 395 [1301.3042].

[101] J. Broedel, C. Duhr, F. Dulat and L. Tancredi, Elliptic polylogarithms and iterated
integrals on elliptic curves. Part I: general formalism, JHEP 05 (2018) 093
[1712.07089].

[102] E. D’Hoker and M.B. Green, Identities between Modular Graph Forms, J. Number
Theor. 189 (2018) 25 [1603.00839].

[103] E. D’Hoker and J. Kaidi, Hierarchy of Modular Graph Identities, JHEP 11 (2016)
051 [1608.04393].

[104] J.E. Gerken and J. Kaidi, Holomorphic subgraph reduction of higher-point modular
graph forms, JHEP 01 (2019) 131 [1809.05122].

337
[105] A. Basu, Poisson equation for the Mercedes diagram in string theory at genus one,
Class. Quant. Grav. 33 (2016) 055005 [1511.07455].

[106] A. Basu, Poisson equation for the three loop ladder diagram in string theory at genus
one, Int. J. Mod. Phys. A 31 (2016) 1650169 [1606.02203].

[107] A. Basu, Proving relations between modular graph functions, Class. Quant. Grav. 33
(2016) 235011 [1606.07084].

[108] A. Kleinschmidt and V. Verschinin, Tetrahedral modular graph functions, JHEP 09

(2017) 155 [1706.01889].

[109] A. Basu, Eigenvalue equation for the modular graph Ca,b,c,d , JHEP 07 (2019) 126
[1906.02674].

[110] J.E. Gerken, Basis Decompositions and a Mathematica Package for Modular Graph
Forms, J. Phys. A 54 (2021) 195401 [2007.05476].

[111] J.E. Gerken, Modular Graph Forms and Scattering Amplitudes in String Theory,
Ph.D. thesis, Humboldt U., Berlin, Humboldt U., Berlin, 2020. 2011.08647.
10.18452/21829.

[112] F. Brown, A class of non-holomorphic modular forms I, 7, 2017 [1707.01230].

[113] F. Brown, A Class of Nonholomorphic Modular Forms ii: Equivariant Iterated

Eisenstein integrals, Forum Math. Sigma 8 (2020) e31.

[114] O. Ahlén and A. Kleinschmidt, D6 R4 curvature corrections, modular graph functions

and Poincaré series, JHEP 05 (2018) 194 [1803.10250].

[115] D. Dorigoni and A. Kleinschmidt, Modular graph functions and asymptotic expansions
of Poincaré series, Commun. Num. Theor. Phys. 13 (2019) 569 [1903.09250].

[116] J. Broedel, O. Schlotterer and F. Zerbini, From elliptic multiple zeta values to
modular graph functions: open and closed strings at one loop, JHEP 01 (2019) 155
[1803.00527].

[117] J.E. Gerken, A. Kleinschmidt and O. Schlotterer, Heterotic-string amplitudes at one

loop: modular graph forms and relations to open strings, JHEP 01 (2019) 052
[1811.02548].

[118] J.E. Gerken, A. Kleinschmidt and O. Schlotterer, All-order differential equations for
one-loop closed-string integrals and modular graph forms, JHEP 01 (2020) 064
[1911.03476].

338
[119] J.E. Gerken, A. Kleinschmidt and O. Schlotterer, Generating series of all modular
graph forms from iterated Eisenstein integrals, JHEP 07 (2020) 190 [2004.05156].

[120] E. D’Hoker, M.B. Green and B. Pioline, Higher genus modular graph functions,
string invariants, and their exact asymptotics, Commun. Math. Phys. 366 (2019) 927
[1712.06135].

[121] E. D’Hoker and O. Schlotterer, Identities among higher genus modular graph tensors,
Commun. Num. Theor. Phys. 16 (2022) 35 [2010.00924].

[122] E. D’Hoker, M.B. Green and B. Pioline, Asymptotics of the D8 R4 genus-two string
invariant, Commun. Num. Theor. Phys. 13 (2019) 351 [1806.02691].

[123] E. D’Hoker, A. Kleinschmidt and O. Schlotterer, Elliptic modular graph forms. Part
I. Identities and generating series, JHEP 03 (2021) 151 [2012.09198].

[124] J.-P. Serre, A course in arithmetic, vol. 7, Springer Science & Business Media (2012).

[125] J.D. Brown and M. Henneaux, Central Charges in the Canonical Realization of
Asymptotic Symmetries: An Example from Three-Dimensional Gravity, Commun.
Math. Phys. 104 (1986) 207.

[126] E. Hecke, Über modulfunktionen und die dirichletschen reihen mit eulerscher
produktentwicklung. i, Mathematische Annalen 114 (1937) 1.

[127] E. Hecke, Über modulfunktionen und die dirichletschen reihen mit eulerscher
produktentwicklung. ii, Mathematische Annalen 114 (1937) 316.

[128] L.J. Mordell, On Mr. Ramanujan’s empirical expansion of modular functions, in

Proc. Cambridge Philos. Soc, vol. 19, pp. 117–124, 1917.

[129] J.A. Harvey and Y. Wu, Hecke Relations in Rational Conformal Field Theory, JHEP
09 (2018) 032 [1804.06860].

[130] J.A. Harvey, Y. Hu and Y. Wu, Galois Symmetry Induced by Hecke Relations in
Rational Conformal Field Theory and Associated Modular Tensor Categories, J.
Phys. A 53 (2020) 334003 [1912.11955].

[131] V. Bouchard, T. Creutzig and A. Joshi, Hecke Operators on Vector-Valued Modular

Forms, SIGMA 15 (2019) 041 [1807.07703].

[132] E. Witten, Three-Dimensional Gravity Revisited, 0706.3359.

[133] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri and Y. Oz, Large N field
theories, string theory and gravity, Phys. Rept. 323 (2000) 183 [hep-th/9905111].

339
[134] E. D’Hoker and D.Z. Freedman, Supersymmetric gauge theories and the AdS / CFT
correspondence, in Theoretical Advanced Study Institute in Elementary Particle
Physics (TASI 2001): Strings, Branes and EXTRA Dimensions, pp. 3–158, 1, 2002
[hep-th/0201253].

[135] M. Ammon and H. Erdmenger, Gauge/Gravity Duality, Foundations and

Applications, Cambridge University Press (2015).

[136] Y.-H. Lin and D. Pei, Holomorphic CFTs and topological modular forms, 2112.10724.

[137] H. Petersson, Über die entwicklungskoeffizienten der automorphen formen, Acta

mathematica 58 (1932) 169.

[138] H. Rademacher, The fourier coefficients of the modular invariant j (τ ), American

Journal of Mathematics 60 (1938) 501.

[139] A. Erdèlyi, ed., Higher transcendental functions, vol. 1 of Bateman Manuscript

Project, Robert E. Krieger Publishing Company (1981).

[140] D.A. Cox, Primes of the form x2 + ny 2 : Fermat, class field theory, and complex
multiplication, vol. 34, John Wiley & Sons (2011).

[141] B.C. Berndt, Ramanujan’s notebooks: Part III, Springer Science & Business Media
(2012).

[142] S. Chowla and A. Selberg, On epstein’s zeta function (i), Proceedings of the National
Academy of Sciences 35 (1949) 371.

[143] A. Selberg and S. Chowla, On epstein’s zeta-function., .

[144] J. Yi, Theta-function identities and the explicit formulas for theta-function and their
applications, Journal of Mathematical Analysis and Applications 292 (2004) 381.

[145] N. Seiberg and E. Witten, Electric - magnetic duality, monopole condensation, and
confinement in N=2 supersymmetric Yang-Mills theory, Nucl. Phys. B 426 (1994) 19
[hep-th/9407087].

[146] N. Seiberg and E. Witten, Monopoles, duality and chiral symmetry breaking in N=2
supersymmetric QCD, Nucl. Phys. B 431 (1994) 484 [hep-th/9408099].

[147] P.C. Argyres and M.R. Douglas, New phenomena in SU(3) supersymmetric gauge
theory, Nucl. Phys. B 448 (1995) 93 [hep-th/9505062].

[148] J.A. Minahan and D. Nemeschansky, An N=2 superconformal fixed point with E(6)
global symmetry, Nucl. Phys. B 482 (1996) 142 [hep-th/9608047].

340
[149] J.A. Minahan and D. Nemeschansky, Superconformal fixed points with E(n) global
symmetry, Nucl. Phys. B 489 (1997) 24 [hep-th/9610076].

[150] I.n. Garcı́a-Etxebarria and D. Regalado, N = 3 four dimensional field theories, JHEP
03 (2016) 083 [1512.06434].

[151] O. Aharony and Y. Tachikawa, S-folds and 4d N=3 superconformal field theories,
JHEP 06 (2016) 044 [1602.08638].

[152] I.n. Garcı́a-Etxebarria and D. Regalado, Exceptional N = 3 theories, JHEP 12

(2017) 042 [1611.05769].

[153] M. Evtikhiev, N = 3 SCFTs in 4 dimensions and non-simply laced groups, JHEP 06

(2020) 125 [2004.03919].

[154] J. Kaidi, M. Martone and G. Zafrir, Exceptional moduli spaces for exceptional N = 3
theories, 2203.04972.

[155] S. Gukov and C. Vafa, Rational conformal field theories and complex multiplication,
Commun. Math. Phys. 246 (2004) 181 [hep-th/0203213].

[156] L.J. Dixon, J.A. Harvey, C. Vafa and E. Witten, Strings on Orbifolds, Nucl. Phys. B
261 (1985) 678.

[157] L.J. Dixon, J.A. Harvey, C. Vafa and E. Witten, Strings on Orbifolds. 2., Nucl. Phys.
B 274 (1986) 285.

[158] R. Dijkgraaf, C. Vafa, E.P. Verlinde and H.L. Verlinde, The Operator Algebra of
Orbifold Models, Commun. Math. Phys. 123 (1989) 485.

[159] M. Cvetic, J. Halverson, G. Shiu and W. Taylor, Snowmass White Paper: String
Theory and Particle Physics, 2204.01742.

[160] E. Gottschling, Über die fixpunkte der siegelschen modulgruppe, Mathematische

Annalen 143 (1961) 111.

[161] E. Gottschling, Über die fixpunktuntergruppen der siegelschen modulgruppe,

Mathematische Annalen 143 (1961) 399.

[162] E. Gottschling, Die uniformisierbarkeit der fixpunkte eigentlich diskontinuierlicher

gruppen von biholomorphen abbildungen, Mathematische Annalen 169 (1967) 26.

[163] H.P. Nilles, S. Ramos-Sanchez, A. Trautner and P.K.S. Vaudrevange, Orbifolds from
Sp(4,Z) and their modular symmetries, Nucl. Phys. B 971 (2021) 115534
[2105.08078].

341
[164] D. Robbins and T. Vandermeulen, Modular Orbits at Higher Genus, JHEP 02 (2020)
113 [1911.06306].

[165] M.B. Green, J.H. Schwarz and E. Witten, Superstring Theory Vol. 1: 25th
Anniversary Edition, Cambridge Monographs on Mathematical Physics, Cambridge
University Press (11, 2012), 10.1017/CBO9781139248563.

[166] M.B. Green, J.H. Schwarz and E. Witten, Superstring Theory Vol. 2: 25th
Anniversary Edition, Cambridge Monographs on Mathematical Physics, Cambridge
University Press (11, 2012), 10.1017/CBO9781139248570.

[167] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic string, Cambridge
Monographs on Mathematical Physics, Cambridge University Press (12, 2007),
10.1017/CBO9780511816079.

[168] J. Polchinski, String theory. Vol. 2: Superstring theory and beyond, Cambridge
Monographs on Mathematical Physics, Cambridge University Press (12, 2007),
10.1017/CBO9780511618123.

[169] B. Zwiebach, A first course in string theory, Cambridge university press (2004).

[170] C.V. Johnson, D-branes, Cambridge university press (2006).

[171] K. Becker, M. Becker and J.H. Schwarz, String theory and M-theory: A modern
introduction, Cambridge university press (2006).

[172] R. Blumenhagen, D. Lüst and S. Theisen, Basic concepts of string theory, vol. 17,
Springer (2013).

[173] E. Kiritsis, String theory in a nutshell, vol. 21, Princeton University Press (2019).

[174] P. Ramond, Dual theory for free fermions, Physical Review D 3 (1971) 2415.

[175] A. Neveu and J.H. Schwarz, Factorizable dual model of pions, Nuclear Physics B 31
(1971) 86.

[176] F. Gliozzi, J. Scherk and D. Olive, Supersymmetry, supergravity theories and the dual
spinor model, Nuclear Physics B 122 (1977) 253.

[177] A.M. Polyakov, Gauge fields and strings, Taylor & Francis (1987).

[178] D.H. Friedan, Nonlinear Models in Two + Epsilon Dimensions, Annals Phys. 163
(1985) 318.

342
[179] L. Alvarez-Gaume, D.Z. Freedman and S. Mukhi, The Background Field Method and
the Ultraviolet Structure of the Supersymmetric Nonlinear Sigma Model, Annals
Phys. 134 (1981) 85.

[180] E. Braaten, T.L. Curtright and C.K. Zachos, Torsion and Geometrostasis in
Nonlinear Sigma Models, Nucl. Phys. B 260 (1985) 630.

[181] H. Osborn, Weyl consistency conditions and a local renormalization group equation
for general renormalizable field theories, Nucl. Phys. B 363 (1991) 486.

[182] E. D’Hoker, String Theory, in Quantum Fields and Strings: A Course for
Mathematicians, P. Deligne and et al., eds., vol. 2, pp. 807–1012, American
Mathematical Society and Institute for Advanced Study, 1999.

[183] L. Alvarez-Gaume and E. Witten, Gravitational Anomalies, Nucl. Phys. B 234

(1984) 269.

[184] M.B. Green and J.H. Schwarz, Anomaly Cancellation in Supersymmetric D=10
Gauge Theory and Superstring Theory, Phys. Lett. B 149 (1984) 117.

[185] J.A. Harvey, TASI 2003 lectures on anomalies, in TASI 2003, 9, 2005
[hep-th/0509097].

[186] E. D’Hoker and D.H. Phong, The geometry of string perturbation theory, Reviews of
Modern Physics 60 (1988) 917.

[187] E. D’Hoker and D. Phong, Lectures on two-loop superstrings, in Superstring Theory,

S.-T.Y. K. Liu and C. Zhu, eds., Advanced Lectures in Mathematics, Vol 1,
pp. 85–123, International Press, 1, 2002 [hep-th/0211111].

[188] N. Berkovits, Multiloop amplitudes and vanishing theorems using the pure spinor
formalism for the superstring, JHEP 09 (2004) 047 [hep-th/0406055].

[189] E. Witten, Notes On Super Riemann Surfaces And Their Moduli, Pure Appl. Math.
Quart. 15 (2019) 57 [1209.2459].

[190] E. Witten, Superstring Perturbation Theory Revisited, 1209.5461.

[191] N. Berkovits, E. D’Hoker, M.B. Green, H. Johansson and O. Schlotterer, Snowmass

White Paper: String Perturbation Theory, in 2022 Snowmass Summer Study, 3, 2022
[2203.09099].

[192] M.B. Green and J.H. Schwarz, Supersymmetrical String Theories, Phys. Lett. B 109
(1982) 444.

343
[193] D.J. Gross, J.A. Harvey, E.J. Martinec and R. Rohm, Heterotic String Theory. 1.
The Free Heterotic String, Nucl. Phys. B 256 (1985) 253.

[194] D.J. Gross, J.A. Harvey, E.J. Martinec and R. Rohm, Heterotic String Theory. 2.
The Interacting Heterotic String, Nucl. Phys. B 267 (1986) 75.

[195] E. D’Hoker and D.H. Phong, Momentum analyticity and finiteness of the one loop
superstring amplitude, Phys. Rev. Lett. 70 (1993) 3692 [hep-th/9302003].

[196] E. D’Hoker and D.H. Phong, The Box graph in superstring theory, Nucl. Phys. B
440 (1995) 24 [hep-th/9410152].

[197] M.B. Green, J.H. Schwarz and L. Brink, N=4 Yang-Mills and N=8 Supergravity as
Limits of String Theories, Nucl. Phys. B 198 (1982) 474.

[198] D.J. Gross and E. Witten, Superstring Modifications of Einstein’s Equations, Nucl.
Phys. B 277 (1986) 1.

[199] M.T. Grisaru, A.E.M. van de Ven and D. Zanon, Four Loop beta Function for the
N=1 and N=2 Supersymmetric Nonlinear Sigma Model in Two-Dimensions, Phys.
Lett. B 173 (1986) 423.

[200] M.B. Green, M. Gutperle and P. Vanhove, One loop in eleven-dimensions, Phys. Lett.
B 409 (1997) 177 [hep-th/9706175].

[201] M.B. Green and P. Vanhove, The Low-energy expansion of the one loop type II
superstring amplitude, Phys. Rev. D 61 (2000) 104011 [hep-th/9910056].

[202] E. D’Hoker and M.B. Green, Exploring transcendentality in superstring amplitudes,

JHEP 07 (2019) 149 [1906.01652].

[203] R. Rankin, Contributions to the theory of Ramanujan’s function τ (n) and similar
arithmetic functions, Proc. of the Cambridge Philosophical Society 35 (1939) 351.

[204] A. Selber, Bemerkungen über eine Dirichletsche Reihe, die mit der Theorie der
Modulformen nahe verbunden ist, Arch. Math. Naturvid. 43 (1940) 47.

[205] D. Zagier, The Rankin-Selberg method for automorphic functions which are not of
rapid decay, Journal of the Faculty of Science, the University of Tokyo, Section 1A
Mathematics 28 (1982) 415.

[206] E. D’Hoker and N. Geiser, Integrating three-loop modular graph functions and
transcendentality of string amplitudes, JHEP 02 (2022) 019 [2110.06237].

344
[207] C.R. Mafra, O. Schlotterer and S. Stieberger, Complete N-Point Superstring Disk
Amplitude I. Pure Spinor Computation, Nucl. Phys. B 873 (2013) 419 [1106.2645].

[208] T. Azevedo, M. Chiodaroli, H. Johansson and O. Schlotterer, Heterotic and bosonic

string amplitudes via field theory, JHEP 10 (2018) 012 [1803.05452].

[209] C.R. Mafra and O. Schlotterer, Towards the n-point one-loop superstring amplitude.
Part I. Pure spinors and superfield kinematics, JHEP 08 (2019) 090 [1812.10969].

[210] C.R. Mafra and O. Schlotterer, Towards the n-point one-loop superstring amplitude.
Part II. Worldsheet functions and their duality to kinematics, JHEP 08 (2019) 091
[1812.10970].

[211] C.R. Mafra and O. Schlotterer, Towards the n-point one-loop superstring amplitude.
Part III. One-loop correlators and their double-copy structure, JHEP 08 (2019) 092
[1812.10971].

[212] J. Broedel, O. Schlotterer and S. Stieberger, Polylogarithms, Multiple Zeta Values

and Superstring Amplitudes, Fortsch. Phys. 61 (2013) 812 [1304.7267].

[213] S. Stieberger, Closed superstring amplitudes, single-valued multiple zeta values and
the Deligne associator, J. Phys. A 47 (2014) 155401 [1310.3259].

[214] O. Schlotterer and S. Stieberger, Motivic Multiple Zeta Values and Superstring
Amplitudes, J. Phys. A 46 (2013) 475401 [1205.1516].

[215] L.F. Alday, A. Bissi and E. Perlmutter, Genus-One String Amplitudes from
Conformal Field Theory, JHEP 06 (2019) 010 [1809.10670].

[216] E. D’Hoker and D.H. Phong, Two loop superstrings. 1. Main formulas, Phys. Lett. B
529 (2002) 241 [hep-th/0110247].

[217] E. D’Hoker and D.H. Phong, Two loop superstrings. 2. The Chiral measure on moduli
space, Nucl. Phys. B 636 (2002) 3 [hep-th/0110283].

[218] E. D’Hoker and D.H. Phong, Two loop superstrings 4: The Cosmological constant
and modular forms, Nucl. Phys. B 639 (2002) 129 [hep-th/0111040].

[219] E. Witten, Notes On Holomorphic String And Superstring Theory Measures Of Low
Genus, 1306.3621.

[220] E. D’Hoker and D.H. Phong, Two-loop superstrings VI: Non-renormalization

theorems and the 4-point function, Nucl. Phys. B 715 (2005) 3 [hep-th/0501197].

345
[221] N. Berkovits, Super-Poincare covariant two-loop superstring amplitudes, JHEP 01
(2006) 005 [hep-th/0503197].

[222] N. Berkovits and C.R. Mafra, Equivalence of two-loop superstring amplitudes in the
pure spinor and RNS formalisms, Phys. Rev. Lett. 96 (2006) 011602
[hep-th/0509234].

[223] E. D’Hoker, M. Gutperle and D.H. Phong, Two-loop superstrings and S-duality, Nucl.
Phys. B 722 (2005) 81 [hep-th/0503180].

[224] H. Gomez and C.R. Mafra, The Overall Coefficient of the Two-loop Superstring
Amplitude Using Pure Spinors, JHEP 05 (2010) 017 [1003.0678].

[225] E. D’Hoker, C.R. Mafra, B. Pioline and O. Schlotterer, Two-loop superstring

five-point amplitudes. Part I. Construction via chiral splitting and pure spinors,
JHEP 08 (2020) 135 [2006.05270].

[226] E. D’Hoker, C.R. Mafra, B. Pioline and O. Schlotterer, Two-loop superstring

five-point amplitudes. Part II. Low energy expansion and S-duality, JHEP 02 (2021)
139 [2008.08687].

[227] E. D’Hoker and O. Schlotterer, Two-loop superstring five-point amplitudes. Part III.
Construction via the RNS formulation: even spin structures, JHEP 12 (2021) 063
[2108.01104].

[228] N. Kawazumi, Johnson’s homomorphisms and the arakelov-green function, arXiv

preprint arXiv:0801.4218 (2008) .

[229] S.-W. Zhang, Gross–schoen cycles and dualising sheaves, arXiv preprint
arXiv:0812.0371 (2008) .

[230] E. D’Hoker and M.B. Green, Zhang-Kawazumi Invariants and Superstring

Amplitudes, J. Number Theor. 144 (2014) 111 [1308.4597].

[231] E. D’Hoker, M.B. Green, B. Pioline and R. Russo, Matching the D6 R4 interaction at
two-loops, JHEP 01 (2015) 031 [1405.6226].

[232] M. Eichler and D. Zagier, The Theory of Jacobi Forms, Progress in Math. Vol 55
(1985), Birkhäuser-Verlag, .

[233] B. Pioline, A Theta lift representation for the Kawazumi-Zhang and Faltings
invariants of genus-two Riemann surfaces, J. Number Theor. 163 (2016) 520
[1504.04182].

346
[234] S.L. Cacciatori, F. Dalla Piazza and B. van Geemen, Modular Forms and Three Loop
Superstring Amplitudes, Nucl. Phys. B 800 (2008) 565 [0801.2543].

[235] E. D’Hoker and D.H. Phong, Asyzygies, modular forms, and the superstring measure
II, Nucl. Phys. B 710 (2005) 83 [hep-th/0411182].

[236] E. Witten, The Super Period Matrix With Ramond Punctures, J. Geom. Phys. 92
(2015) 210 [1501.02499].

[237] H. Gomez and C.R. Mafra, The closed-string 3-loop amplitude and S-duality, JHEP
10 (2013) 217 [1308.6567].

[238] Y. Geyer, R. Monteiro and R. Stark-Muchão, Superstring Loop Amplitudes from the
Field Theory Limit, Phys. Rev. Lett. 127 (2021) 211603 [2106.03968].

[239] S.L. Cacciatori, F. Dalla Piazza and B. van Geemen, Genus four superstring
measures, Lett. Math. Phys. 85 (2008) 185 [0804.0457].

[240] S. Grushevsky, Superstring scattering amplitudes in higher genus, Commun. Math.

Phys. 287 (2009) 749 [0803.3469].

[241] J. Scherk and J.H. Schwarz, How to Get Masses from Extra Dimensions, Nucl. Phys.
B 153 (1979) 61.

[242] K.S. Narain, New Heterotic String Theories in Uncompactified Dimensions < 10,
Phys. Lett. B 169 (1986) 41.

[243] K.S. Narain, M.H. Sarmadi and E. Witten, A Note on Toroidal Compactification of
Heterotic String Theory, Nucl. Phys. B 279 (1987) 369.

[244] R. Dijkgraaf, E.P. Verlinde and H.L. Verlinde, C = 1 Conformal Field Theories on
Riemann Surfaces, Commun. Math. Phys. 115 (1988) 649.

[245] P. Di Francesco, H. Saleur and J.B. Zuber, Modular Invariance in Nonminimal

Two-dimensional Conformal Theories, Nucl. Phys. B 285 (1987) 454.

[246] P. di Francesco, H. Saleur and J.B. Zuber, Relations between the Coulomb gas picture
and conformal invariance of two-dimensional critical models, .

[247] Y. Tachikawa and K. Yonekura, Why are Fractional Charges of Orientifolds

Compatible with Dirac Quantization?, 1805.02772.

[248] M.B. Green and S. Sethi, Supersymmetry constraints on type IIB supergravity, Phys.
Rev. D 59 (1999) 046006 [hep-th/9808061].

347
[249] M.B. Green and P. Vanhove, Duality and higher derivative terms in M theory, JHEP
01 (2006) 093 [hep-th/0510027].

[250] C.M. Hull and P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438
(1995) 109 [hep-th/9410167].

[251] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995)
85 [hep-th/9503124].

[252] J.H. Schwarz, An SL(2,Z) multiplet of type IIB superstrings, Phys. Lett. B 360
(1995) 13 [hep-th/9508143].

[253] J. Polchinski and E. Witten, Evidence for heterotic - type I string duality, Nucl.
Phys. B 460 (1996) 525 [hep-th/9510169].

[254] J.H. Schwarz, Lectures on superstring and M theory dualities: Given at ICTP Spring
School and at TASI Summer School, Nucl. Phys. B Proc. Suppl. 55 (1997) 1
[hep-th/9607201].

[255] N.A. Obers and B. Pioline, U duality and M theory, Phys. Rept. 318 (1999) 113
[hep-th/9809039].

[256] N.A. Obers and B. Pioline, Eisenstein series and string thresholds, Commun. Math.
Phys. 209 (2000) 275 [hep-th/9903113].

[257] M.B. Green and M. Gutperle, Effects of D instantons, Nucl. Phys. B 498 (1997) 195
[hep-th/9701093].

[258] B. Pioline, A Note on nonperturbative R**4 couplings, Phys. Lett. B 431 (1998) 73
[hep-th/9804023].

[259] A. Sinha, The G(hat)**4 lambda**16 term in IIB supergravity, JHEP 08 (2002) 017
[hep-th/0207070].

[260] M.B. Green, S.D. Miller and P. Vanhove, SL(2, Z)-invariance and D-instanton
contributions to the D6 R4 interaction, Commun. Num. Theor. Phys. 09 (2015) 307
[1404.2192].

[261] M.B. Green, J.G. Russo and P. Vanhove, Automorphic properties of low energy string
amplitudes in various dimensions, Phys. Rev. D 81 (2010) 086008 [1001.2535].

[262] M.B. Green and C. Wen, Modular Forms and SL(2, Z)-covariance of type IIB
superstring theory, JHEP 06 (2019) 087 [1904.13394].

348
[263] E.J. Martinec, Nonrenormalization Theorems and Fermionic String Finiteness, Phys.
Lett. B 171 (1986) 189.

[264] R. Dijkgraaf, E.P. Verlinde and H.L. Verlinde, Counting dyons in N=4 string theory,
Nucl. Phys. B 484 (1997) 543 [hep-th/9607026].

[265] F. Denef, S. Kachru, Z. Sun and A. Tripathy, Higher genus Siegel forms and
multi-center black holes in N=4 supersymmetric string theory, 1712.01985.

[266] S. Kachru and A. Tripathy, Black Holes and Hurwitz Class Numbers, Int. J. Mod.
Phys. D 26 (2017) 1742003 [1705.06295].

[267] N. Benjamin, S. Kachru, K. Ono and L. Rolen, Black holes and class groups,
1807.00797.

[268] R. Dijkgraaf, J.M. Maldacena, G.W. Moore and E.P. Verlinde, A Black hole Farey
tail, hep-th/0005003.

[269] J. de Boer, M.C.N. Cheng, R. Dijkgraaf, J. Manschot and E. Verlinde, A Farey Tail
for Attractor Black Holes, JHEP 11 (2006) 024 [hep-th/0608059].

[270] J. Manschot and G.W. Moore, A Modern Farey Tail, Commun. Num. Theor. Phys. 4
(2010) 103 [0712.0573].

[271] J. Wess and J. Bagger, Supersymmetry and supergravity, Princeton University Press,
Princeton, NJ, USA (1992).

[272] O. Aharony, N. Seiberg and Y. Tachikawa, Reading between the lines of

four-dimensional gauge theories, JHEP 08 (2013) 115 [1305.0318].

[273] S. Weinberg, The quantum theory of fields. Vol. 3: Supersymmetry, Cambridge

University Press (6, 2013).

[274] D.Z. Freedman and A. Van Proeyen, Supergravity, Cambridge Univ. Press,
Cambridge, UK (5, 2012).

[275] M. Shifman, Advanced Topics in Quantum Field Theory, Cambridge University Press
(4, 2022), 10.1017/9781108885911.

[276] P. Goddard, J. Nuyts and D.I. Olive, Gauge Theories and Magnetic Charge, Nucl.
Phys. B 125 (1977) 1.

[277] C. Montonen and D.I. Olive, Magnetic Monopoles as Gauge Particles?, Phys. Lett. B
72 (1977) 117.

349
[278] H. Osborn, Topological Charges for N=4 Supersymmetric Gauge Theories and
Monopoles of Spin 1, Phys. Lett. B 83 (1979) 321.

[279] E. Witten and D.I. Olive, Supersymmetry Algebras That Include Topological Charges,
Phys. Lett. B 78 (1978) 97.

[280] E. Witten, Dyons of Charge e theta/2 pi, Phys. Lett. B 86 (1979) 283.

[281] A. Kapustin and E. Witten, Electric-Magnetic Duality And The Geometric Langlands
Program, Commun. Num. Theor. Phys. 1 (2007) 1 [hep-th/0604151].

[282] S.M. Chester, M.B. Green, S.S. Pufu, Y. Wang and C. Wen, New modular invariants
in N = 4 Super-Yang-Mills theory, JHEP 04 (2021) 212 [2008.02713].

[283] D. Dorigoni, M.B. Green and C. Wen, Exact expressions for n-point maximal
U (1)Y -violating integrated correlators in SU (N ) N = 4 SYM, JHEP 11 (2021) 132
[2109.08086].

[284] D. Dorigoni, M.B. Green and C. Wen, Exact properties of an integrated correlator in
N = 4 SU(N) SYM, JHEP 05 (2021) 089 [2102.09537].

[285] D. Dorigoni, M.B. Green and C. Wen, The SAGEX Review on Scattering Amplitudes,
Chapter 10: Modular covariance of type IIB string amplitudes and their N = 4
supersymmetric Yang-Mills duals, 2203.13021.

[286] K.A. Intriligator and N. Seiberg, Lectures on supersymmetric gauge theories and
electric-magnetic duality, Nucl. Phys. B Proc. Suppl. 45BC (1996) 1
[hep-th/9509066].

[287] L. Alvarez-Gaume and S.F. Hassan, Introduction to S duality in N=2 supersymmetric

gauge theories: A Pedagogical review of the work of Seiberg and Witten, Fortsch.
Phys. 45 (1997) 159 [hep-th/9701069].

[288] E. D’Hoker and D.H. Phong, Lectures on supersymmetric Yang-Mills theory and
integrable systems, in 9th CRM Summer School: Theoretical Physics at the End of
the 20th Century, pp. 1–125, 12, 1999 [hep-th/9912271].

[289] M. Martone, The constraining power of Coulomb Branch Geometry: lectures on

Seiberg-Witten theory, in Young Researchers Integrability School and Workshop 2020:
A modern primer for superconformal field theories, 6, 2020 [2006.14038].

[290] Y. Tachikawa, N=2 supersymmetric dynamics for pedestrians (12, 2013),

10.1007/978-3-319-08822-8, [1312.2684].

350
[291] N.A. Nekrasov, Seiberg-Witten prepotential from instanton counting, Adv. Theor.
Math. Phys. 7 (2003) 831 [hep-th/0206161].

[292] V. Pestun, Localization of gauge theory on a four-sphere and supersymmetric Wilson

loops, Commun. Math. Phys. 313 (2012) 71 [0712.2824].

[293] V. Pestun et al., Localization techniques in quantum field theories, J. Phys. A 50

(2017) 440301 [1608.02952].

[294] R. Donagi and E. Witten, Supersymmetric Yang-Mills theory and integrable systems,
Nucl. Phys. B 460 (1996) 299 [hep-th/9510101].

[295] E.J. Martinec, Integrable structures in supersymmetric gauge and string theory, Phys.
Lett. B 367 (1996) 91 [hep-th/9510204].

[296] E. D’Hoker and D.H. Phong, Calogero-Moser systems in SU(N) Seiberg-Witten

theory, Nucl. Phys. B 513 (1998) 405 [hep-th/9709053].

[297] E. D’Hoker and D.H. Phong, Calogero-Moser Lax pairs with spectral parameter for
general Lie algebras, Nucl. Phys. B 530 (1998) 537 [hep-th/9804124].

[298] E. D’Hoker and D.H. Phong, Spectral curves for superYang-Mills with adjoint
hypermultiplet for general Lie algebras, Nucl. Phys. B 534 (1998) 697
[hep-th/9804126].

[299] J. Manschot, G.W. Moore and X. Zhang, Effective gravitational couplings of

four-dimensional N = 2 supersymmetric gauge theories, JHEP 06 (2020) 150
[1912.04091].

[300] J.A. Minahan, D. Nemeschansky and N.P. Warner, Instanton expansions for mass
deformed N=4 superYang-Mills theories, Nucl. Phys. B 528 (1998) 109
[hep-th/9710146].

[301] J. Aspman, E. Furrer and J. Manschot, Cutting and gluing with running couplings in
N=2 QCD, Phys. Rev. D 105 (2022) 025021 [2107.04600].

[302] J. Aspman, E. Furrer and J. Manschot, Four flavors, triality, and bimodular forms,
Phys. Rev. D 105 (2022) 025017 [2110.11969].

[303] J. Aspman, E. Furrer and J. Manschot, Elliptic Loci of SU(3) Vacua, Annales Henri
Poincare 22 (2021) 2775 [2010.06598].

[304] J. Manschot and G.W. Moore, Topological correlators of SU (2), N = 2∗ SYM on

four-manifolds, 2104.06492.

351
[305] J. Stienstra and D. Zagier, Bimodular forms and holomorphic anomaly equation,
https://www.birs.ca/workshops/2006/06w5041/report06w5041.pdf.

[306] E. Witten, Solutions of four-dimensional field theories via M theory, Nucl. Phys. B
500 (1997) 3 [hep-th/9703166].

[307] D. Gaiotto, N=2 dualities, JHEP 08 (2012) 034 [0904.2715].

[308] M. Akhond, G. Arias-Tamargo, A. Mininno, H.-Y. Sun, Z. Sun, Y. Wang et al., The
Hitchhiker’s Guide to 4d N = 2 Superconformal Field Theories, 12, 2021
[2112.14764].

[309] S.S. Razamat, E. Sabag, O. Sela and G. Zafrir, Aspects of 4d supersymmetric

dynamics and geometry, 2203.06880.

[310] C. Beem, M. Lemos, P. Liendo, W. Peelaers, L. Rastelli and B.C. van Rees, Infinite
Chiral Symmetry in Four Dimensions, Commun. Math. Phys. 336 (2015) 1359
[1312.5344].

[311] C. Beem and L. Rastelli, Vertex operator algebras, Higgs branches, and modular
differential equations, JHEP 08 (2018) 114 [1707.07679].

[312] J. Kaidi, M. Martone, L. Rastelli and M. Weaver, Needles in a haystack. An

algorithmic approach to the classification of 4d N = 2 SCFTs, JHEP 03 (2022) 210
[2202.06959].

[313] J.-P. Escofier, Galois theory, vol. 204, Springer Science & Business Media (2012).

[314] T.W. Hungerford, Abstract algebra: an introduction, Cengage Learning (2012).

[315] E. Artin and A.N. Milgram, Galois theory, vol. 2, Courier Corporation (1998).

[316] B.L. Van der Waerden, Algebra, vol. 2, Springer Science & Business Media (2003).

[317] D.A. Cox, Galois theory, vol. 61, John Wiley & Sons (2011).

[318] J. De Boer and J. Goeree, Markov traces and ii1 factors in conformal field theory,
Communications in mathematical physics 139 (1991) 267.

[319] A. Coste and T. Gannon, Remarks on Galois symmetry in rational conformal field
theories, Phys. Lett. B 323 (1994) 316.

[320] A. Coste and T. Gannon, Congruence subgroups and rational conformal field theory,
math/9909080.

352
[321] P. Bantay, The Kernel of the modular representation and the Galois action in RCFT,
Commun. Math. Phys. 233 (2003) 423 [math/0102149].

[322] M. Buican and R. Radhakrishnan, Galois conjugation and multiboundary

entanglement entropy, JHEP 12 (2020) 045 [1912.04937].

[323] M. Buican and R. Radhakrishnan, Galois orbits of TQFTs: symmetries and

unitarity, JHEP 01 (2022) 004 [2109.02766].

[324] J. Kaidi, Z. Komargodski, K. Ohmori, S. Seifnashri and S.-H. Shao, Higher central
charges and topological boundaries in 2+1-dimensional TQFTs, 2107.13091.

[325] E. Hecke, J.R. Goldman and R. Kotzen, Lectures on the theory of algebraic numbers,
vol. 77, Springer (1981).

[326] R.C. Gunning, Lectures on riemann surfaces, in Lectures on Riemann Surfaces,

Princeton University Press (2015).

[327] H.M. Farkas and I. Kra, Riemann surfaces, in Riemann surfaces, pp. 9–31, Springer
(1992).

[328] J.-B. Bost, Introduction to compact riemann surfaces, jacobians, and abelian
varieties, in From number theory to physics, pp. 64–211, Springer (1992).

[329] J.D. Fay, Theta functions on Riemann surfaces, vol. 352, Springer (2006).

[330] M. Schiffer and D.C. Spencer, Functionals of finite Riemann surfaces, Courier
Corporation (2014).

[331] R.C. Gunning, Lectures on vector bundles over Riemann surfaces, vol. 6, Princeton
university press (1967).

[332] K. Kodaira, Complex manifolds and deformation of complex structures, Springer

(2006).

[333] O. Alvarez, Theory of Strings with Boundaries: Fluctuations, Topology, and

Quantum Geometry, Nucl. Phys. B 216 (1983) 125.

[334] J.-i. Igusa, Theta functions, vol. 194, Springer Science & Business Media (2012).

[335] C.L. Siegel, Topics in Complex Function Theory, Volume 2: Automorphic Functions
and Abelian Integrals, vol. 16, John Wiley & Sons (1989).

[336] C.L. Siegel, Topics in Complex Function Theory, Volume 3: Abelian Functions and
Modular Functions of Several Variables, vol. 16, John Wiley & Sons (1989).

353
[337] H. Klingen, Introductory lectures on siegel modular forms, Cambridge Stud. Adv.
Math. 20 (1990) .

[338] G. Van Der Geer, Siegel modular forms and their applications, in The 1-2-3 of
modular forms, pp. 181–245, Springer, 2008.

[339] B.A. Dubrovin, Theta functions and non-linear equations, Russian mathematical
surveys 36 (1981) 11.

354