Lecture Notes on

Geometric Quantization

Based on the Course Given in 2021/22

December 16, 2024

Martin Schottenloher
Mathematisches Institut der Universität München

Version 15.0

(Under Construction)
 Objective
These Lecture Notes were intended to support the course ”Geometric Quantization”
held in the winter term 2021/2022 at the LMU München. Its content has been written
down step by step on the basis of the actual development of the course in the lectures
and exercises, and, in particular, in interaction with the comments and questions of
the participating students. Some of the contributions of the participants are included
in the Lecture Notes.
The result of these efforts (as of February 2022, end of the course) is roughly the
content of the first 10 chapters (with about 130 pages at Febr. 22) and the 6 appendices
(with about 90 pages at Febr. 22).
Now, after the course has been completed, while many questions remain open, the
Lecture Notes will be developed further. Among others with the aim to present further
approaches to Geometric Quantization and provide more examples. The main goal is to
add several main achievements of Geometric Quantization such as half-density quan-
tization, half-form quantization, pairing, metalinear structure, applications to field
theory, etc. as well as to complement and improve the first 10 chapters and the appen-
dices.

2
CONTENTS i

Contents

Introduction 1

1 Hamiltonian Mechanics 1
1.1 A Simple Hamiltonian System . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Canonical Equations . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Symplectic Involution . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Symplectic Form . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.4 Poisson Bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Symplectic Manifolds and Hamiltonian Systems . . . . . . . . . . . . . 6
1.2.1 Notations for Manifolds . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Symplectic Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 Hamiltonian Systems . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.4 Hamiltonian Vector Fields . . . . . . . . . . . . . . . . . . . . . 17
1.3 Examples of Hamiltonian Systems . . . . . . . . . . . . . . . . . . . . . 20
1.3.1 Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.2 Reduction with respect to first integrals . . . . . . . . . . . . . 21
1.3.3 Kepler Problem (Hydrogen Atom) . . . . . . . . . . . . . . . . . 22
1.3.4 Particle in a Field . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.5 Coadjoint Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4 Lagrangian Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Ansatz Prequantization 33
2.1 Canonical Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Ansatz: Prequantization . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Line Bundles 43
3.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Cocycles Generating Line Bundles . . . . . . . . . . . . . . . . . . . . . 45
3.3 The Tautological Line Bundle . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
ii CONTENTS

4 Connections 61
4.1 Local Connection Form . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Global Connection Form . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Horizontal Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5 Parallel Transport and Curvature 74

5.1 Parallel Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2 Horizontal Section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4 Integrality Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

6 Hermitian and Holomorphic Line Bundles 87

6.1 Hermitian Line Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 Holomorphic Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

7 Prequantization 94
7.1 Quantizable Phase Space . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.2 Prequantum Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.3 Prequantum Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . . 102

8 Integrality 106
8.1 Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
8.2 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
8.3 Flat Line Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

9 Polarization 119
9.1 Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
9.2 Real polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.3 The Complex Linear Case . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.4 Complex Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9.5 Kähler Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
CONTENTS iii

10 Representation Space 137

10.1 Polarized Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
10.2 Kähler Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
10.3 Directly Quantizable Observables . . . . . . . . . . . . . . . . . . . . . 151
10.4 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.5 Sketch of Correction Scheme . . . . . . . . . . . . . . . . . . . . . . . . 157

11 Existence of Polarized Sections and Holonomy 161

11.1 Bohr-Sommerfeld Variety . . . . . . . . . . . . . . . . . . . . . . . . . . 161
11.2 Distributional Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

12 Densities and Their Derivatives 168

12.1 Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
12.2 Integration of Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
12.3 Partial Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
12.4 Partial Lie Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

13 Half-Density Quantization of the Momentum Phase Space 186

13.1 Descend of Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
13.2 Representation Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
13.3 Quantum Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
13.4 The Case of a Real Polarization . . . . . . . . . . . . . . . . . . . . . . 196

14 Half-Density Quantization in General 199

14.1 Descend of Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
14.2 Representation Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
14.3 Quantum Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
14.4 Other Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
14.5 Half-Density Pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

15 Half-Form Quantization 219

15.1 Canonical Bundle of a Vector Bundle . . . . . . . . . . . . . . . . . . . 220
15.2 Need for a Square Root of a Line Bundle . . . . . . . . . . . . . . . . . 222
15.3 Descend of Half-Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
15.4 Representation Space and Quantum Operator . . . . . . . . . . . . . . 228
15.5 Square Root: Existence and Uniqueness . . . . . . . . . . . . . . . . . 231
iv CONTENTS

16 Metalinear Structure 235

16.1 Metalinear Frame Bundle . . . . . . . . . . . . . . . . . . . . . . . . . 235
16.2 Half-Form Quantization Based on the Metalinear Group . . . . . . . . 240
16.3 Metalinear Structure: Existence and Uniqueness . . . . . . . . . . . . . 242

17 Metaplectic Structure 246

17.1 Metaplectic Frame Bundle . . . . . . . . . . . . . . . . . . . . . . . . . 246
17.2 General Metastructures . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
17.3 Square Roots, Metalinear und Metaplectice Structures . . . . . . . . . 252
17.4 Half-Form Pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
17.5 Mteplectic Representation . . . . . . . . . . . . . . . . . . . . . . . . . 256

Appendix: Mathematical Supplements 257

A Manifolds 257
A.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
A.2 Tangent and Cotangent Bundle . . . . . . . . . . . . . . . . . . . . . . 260
A.3 Vector Fields and Dynamical Systems . . . . . . . . . . . . . . . . . . . 267
A.4 Quotient Manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
A.4.1 Topological Quotients . . . . . . . . . . . . . . . . . . . . . . . 270
A.4.2 Differentiable Quotients . . . . . . . . . . . . . . . . . . . . . . 274
A.5 Operations on Differential Forms . . . . . . . . . . . . . . . . . . . . . 275

B Complex Analysis 282

B.1 Cauchy Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
B.2 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
B.3 Hartogs’ Extension Theorem . . . . . . . . . . . . . . . . . . . . . . . . 286
B.4 Sequences of Holomorphic Functions . . . . . . . . . . . . . . . . . . . 288
B.5 Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

C Lie Theory 299

C.1 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
C.2 Extensions of Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . 304
C.3 Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
C.4 The Lie Algebra of a Lie Group . . . . . . . . . . . . . . . . . . . . . . 310
C.5 Lie Group Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
CONTENTS v

D Fibre Bundles 316

D.1 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
D.2 Principal Fibre Bundles and Frame Bundles . . . . . . . . . . . . . . . 317
D.3 Associated bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
D.4 Principal Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.5 Connection on a Vector Bundle . . . . . . . . . . . . . . . . . . . . . . 330

E Cohomology 333
E.1 Čech Cohomology with Values in an Abelian Group . . . . . . . . . . . 333
E.2 DeRham Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
E.3 Sheaf Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

F Quantum Mechanics 345

F.1 Four Postulates of Quantum Mechanics . . . . . . . . . . . . . . . . . . 345
F.2 Self-Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
F.3 Canonical Commutation Relations . . . . . . . . . . . . . . . . . . . . . 366

References 372

Sign Conventions 375

Index 376
vi CONTENTS

Leitfaden
Geometric Quantization begins with a set of observables of a model of Classical Me-
chanics. This model is realized as a symplectic manifold (M, ω) and the observables
form a subset o of the space E(M, R) of real-valued functions on M . To describe this
background these notes start with a mathematical exposition of Classical Mechanics
in chapter 1. They proceed with a first attempt to construct a quantum model on the
basis of (M, ω) and o ⊂ E(M, R) in chapter 2.
This attempt shows that a Hermitian complex line bundle L with connection ∇
on M is required for the program of Geometric Quantization in such a way that the
curvature Curv(L, ∇) of the connection is the symplectic form ω. In order to formulate
this requirement the basic notions of a complex line bundle (ch. 3), a connection (ch. 4),
a curvature (ch. 5) and a Hermitian structure (ch. 6) are developed.
With these ingredients at hand the first step of Geometric Quantization – prequan-
tization – is carried through in chapter 7: For a given symplectic manifold (M, ω) and
and a Hermitian linde bundle (L, ∇, H) satisfying Curv(L, ∇) = ω – called a prequan-
tum line bundle – a complex Hilbert space H (generated by a subspace of sections of
L) and a map q : E(M, R) → S(H) (using the connection ∇) is constructed, where
S(H) is the set of linear operators on H, such that the so called Dirac Conditions are
satisfied:

(D1) q(1) = idH ,

i
(D2) [q(F ), q(G)] = 2π
q({F, G}) , for all F, G ∈ E(M, R).

Here, ”{ , }” is the Poisson bracket of the symplectic manifold. Note that prequantiza-
tion works for arbitrary subsets o of observables. But prequantization fails to provide
good quantum models for several reasons. For instance, let M be the spaceM = R2n
with the standard symplectic form ω = dq j ∧ dpj , and let L = M × C be the trivial
line bundle with connection ∇ given by the form −pj dq j . Then the Hilbert space H is
L2 (R2n , C), the space of square integrable functions ϕ : R2n → C, and the observables
q i , pj have as their prequantization operators

i ∂
q qj = + q j =: Qj ,


2π ∂pj
i ∂
q (pj ) = − =: Pj .
2π ∂q j
CONTENTS vii

Although (D1), i.e. q(1) = id, and (D2), i.e.

i i i i i
[Qi , Pj ] = δj = {q , pj } = q({q i , pj })
2π 2π 2π
are satisfied, this result is not in accordance with the usual quantum model. In particu-
lar, the prequantization q is not irreducible. However, if we restrict the operators Qi , Pj
to the smaller Hilbert space H = L2 (Rn , C) ⊂ L2 (R2n , C) of functions only depending
on the variables q i , the usual quantum model is achieved.
This procedure of cutting down the number of variables can be generalized by
introducing polarizations on M (see ch. 9) along which the sections generating the
Hilbert space have to be constant.
Before the presentation of polarizations in chapter 8 the question is discussed under
which condition a symplectic manifold (M, ω) admits a Hermitian line bundle (L, ∇, H)
with connection such that Curv(L, ∇) = ω. It turns out that this condition is a purely
topological condition on M and ω which can be expressed best by cohomology. We
call symplectic manifolds quantizable if this condition is satisfied.
Chapter 10 is devoted to the construction of the first version of a full Geometric
Quantization. The construction is based on the geometric data of a prequantum line
bundle (L, ∇, H) on a symplectic manifold and a polarization P . The representation
space HP is then a suitable Hilbert completion of polarized sections. Here, a polarized
section is a section s of L satisfying ∇X s = 0. Moreover, we need the notion of a
directly quantizable observable F in order to confirm that for a polarized section s
the derivative ∇XF s is polarized as well. Finally, the prequantum operator determines
the quantum operator, which will be denoted as q(F ) as well, and which satisfies the
Dirac conditions with respect to a smaller representation space H and the directly
quantizable observables.
Several elementary examples are presented in detail in this chapter in order to illus-
trate the impact of Geometric Quantization. Among others, the geometric quantization
of the harmonic oscillator is calculated, leading to a reasonable result. This result has,
however, a shift in the eigenvalues in comparison to the known results from Quantum
Mechanic. By a modification of Geometric Quantization this defect can be removed.
The content of the first 10 chapters of these notes covers the development of the
course given in winter 21/22 which constitutes essentially half the notes. The second
half deals with various improvements, modifications and generalizations.

The second part of the notes begins in ch. 11 with an analysis of the existence
of enough polarized sections and quantizable observables. For instance, when some
the leaves of the quotient M/P of M induced by the polarization P on M are not
simply connected it can happen that there do not exist nontrivial polarized sections and
holonomy comes into play (Bohr-Sommerfeld condition). One way to obtain reasonable
representation spaces nevertheless is to consider generalized sections in the sense of
distributions.
viii CONTENTS

The next step in Geometric Quantization is Half-Density Quantization. In addition

to the prequantum bundle L on M and the polarization P a half-density bundle S on
M is considered as an additional geometric structure together with a partial connection
induced on S by the polarization. The bundle L now is replaced with the line bundle
L ⊗ S and the quantization is then based on the polarized sections of L ⊗ S to obtain
the Half-Density Quantization roughly in the same way as in ch. 10. As a preparation
a detailed exposition on r-densities on a manifold and their integration theory is given
in chapter 12. Chapter 13 and 14 deal with the quantization, first for cotangent spaces
M = T ∗ Q (sometimes called momentum spaces) and then in general.
Chapter 15 deals with Half-Form Quantization which is similar to Half-Density
Quantization but the additional line bundle now is a half-form bundle S, i.e. a line
bundle S with the property S ⊗ S ∼ = Λn P ∨ (Λn P ∨ is the line bundle of n-forms in
P , the so-called canonical bundle of P ). We discuss the topological condition on P
and M under which such a half-bundle exists. Note, that, in contrast to the half-form
case, the half-density bundle always exists as a trivial line bundle. The topological
condition which ensures the existence of a half-form bundle is needed for the existence
of a so-called metalinear structure on P (Chapter 16) and the existence of a metaplectic
structure (Chapter 17).
 1

1 Hamiltonian Mechanics

Hamiltonian Mechanics is the study of conservative systems of Classical Mechanics.

These systems are modeled by Hamiltonian systems. The purpose of this chapter is to
introduce step by step the concept of a Hamiltonian system, first on an open subset of
Rn as configuration space, then on the cotangent bundle of a general manifold as phase
space and finally on a general symplectic manifold as phase space. A Hamiltonian
system is a special case of a dynamical system induced by a function H on phase
space.
Since we need general manifolds as phase spaces this chapter also serves to recall
the basic notions related to a manifold and the notation used throughout these lecture
notes.

1.1 A Simple Hamiltonian System

1.1.1 Canonical Equations

We begin with a special case of a system of Hamiltonian Mechanics where the config-
uration space (”Ortsraum”) is an open subset U ⊂ Rn of Rn (n ∈ N, n > 0):

• U ⊂ Rn open subset of Rn , the Configuration Space,

• M := U × Rn ∼= T ∗ U , the (Momentum) Phase Space, and the smooth func-
tions f : M → R or f : M → C are the Observables,
• H ∈ C ∞ (M ), the Hamiltonian Function.

The Equations of Motion are

∂H ∂H
q̇ = , ṗ = − , (1)
∂p ∂q
i.e.
∂H ∂H
q̇ j = , ṗj = − ,
∂pj ∂q j
for j = 1, . . . n.
Here, q = (q 1 , . . . , q n ) are the Position Coordinates in U (”Ortskoordinaten”)
and p = (p1 , . . . , pn ) are the Momentum Coordinates of the cotangent space Tq∗ U ∼
=
Rn (”Impulskoordinaten”).
The equations of motion are also called Canonical Equations or Hamilto-
nian Equations, and (M, H) will be called a Simple Hamiltonian System with
n degrees of freedom.
In many cases of simple Hamiltonian systems the function H has the interpretation
of the energy as we see in the case of the harmonic oscillator:
2 1. Hamiltonian Mechanics

Example 1.1. The harmonic oscillator in n dimensions can be modeled as a simple

Hamiltonian System in the following way (disregarding constants):

• U = Rn , and

• H(q, p) := 12 (∥p∥2 + ∥q∥2 ) = 1

Pn
2 j=1 ((pj )2 + (q j )2 ) , (q, p) ∈ M = U × Rn ,

with canonical equations

q̇ = p, ṗ = q.

In general, the canonical equations can be written in the form

 
∂H ∂H
(q̇, ṗ) = ,− (2)
∂p ∂q

In this form they look like a dynamical system on the phase space M of the type

ż = A(z),

with a vector field A : M → T M and z = (q, p), where the vector field A is similar to
a gradient
∂F
A = ∇F =
∂z
of a C ∞ -function F on M . This is, in fact, true up to a ”twist”, the symplectic twist!

1.1.2 Symplectic Involution

To explain the symplectic twist, we define the Symplectic Structure on the phase
M = T ∗U ∼= U × Rn by the linear map σ on the tangent space Tz M ∼
= Rn × Rn , z ∈ M ,

σ : Rn × Rn → Rn × Rn , (q, p) 7−→ (p, −q), (3)

given by the block matrix σ

 
0 1
σ= (4)
−1 0

acting as follows:

      
q 0 1 q p
σ: 7−→ = .
p −1 0 p −q
1.1 A Simple Hamiltonian System 3

Definition 1.2. We define the symplectic gradient ∇σF of a function F ∈ C ∞ (M )

to be
 
σ ∂F ∂F
∇ F = σ ◦ ∇F = ,− .
∂p ∂q
The vector field ∇σH will often be denoted by XH , XH := ∇σH, and XH will be
called the Hamiltonian Vector Field associated with H.

With these notations, the canonical equations obtain the form

ż = ∇σH(z) for z = (q, p) ∈ M ,

or

ż = XH (z). (5)

Because of σ 2 = σ ◦ σ = −idR2n the map σ and ist matrix is called the Symplectic
Involution.

1.1.3 Symplectic Form

The symplectic structure on M = T ∗ U can also be given by the symplectic form or by

the Poisson bracket, as will be explained in the following.
Definition 1.3. The Symplectic Form ω on the tangent space Tz P = Tz (T ∗ U ) ∼
=
n n
R × R at z ∈ M is
n
X
ω := dq j ∧ dpj = dq j ∧ dpj
j=1

(we use Einstein summation in the following).

Hence, the bilinear and alternating map

ω : Rn × Rn → R

is given by
ω(X, X̄) = X j Ȳj − X̄ j Yj ,
when

X = (X 1 , . . . , X n , Y1 , . . . , Yn ) , X̄ = (X̄ 1 , . . . , X̄ n , Ȳ1 , . . . , Ȳn ) ∈ Rn × Rn .

with respect to the standard coordinates of Rn × Rn .

4 1. Hamiltonian Mechanics

The corresponding standard vector space basis B = {a1 , . . . , an , b1 , . . . , bn } of Rn ×

R determines the coordinates X = (X 1 , . . . , X n , Y1 , . . . , Yn ) = X j aj + Yj bj which we
n

have just used. ω satisfies ω(ai , bj ) = δij = −ω(bj , ai ) and for all other basis vectors
v, w ∈ B: ω(v, w) = 0. Such a basis is called a symplectic frame. The induced matrix
representing the symplectic form ω is given by the coefficients ω(v, w), v, w ∈ B. It is
the symplectic involution σ(cf. (3), (4)). Therefore, the symplectic form can also be
described by matrix multiplication

ω(X, X̄) = X ⊤ σ X̄.1

1.1.4 Poisson Bracket

The Poisson Bracket {F, G} of two functions (i.e. observables) F, G ∈ C ∞ (M ) is

defined by
{F, G} := ω(XF , XG ) ,
which is also given by the well-known expression

∂F ∂G ∂F ∂G ∂F ∂G ∂F ∂G
{F, G} = − = j − .
∂q ∂p ∂p ∂q ∂q ∂pj ∂pj ∂q j

Remark 1.4. Other sign convention are used in the literature, for example dpj ∧ dq j
which is −ω in our notation. See Table 17.5 for more conventions.

A straightforward and remarkable property of the Poisson bracket is the following

result
Proposition 1.5 (Equations of motion in Poisson form). A curve z : I → M in M
(i.e. z ∈ E(I, M )) is a solution of the canonical equations ż = XH (z) if and only if

Ḟ = {F, H}

for all observables F ∈ E(M ), i.e.

d
F (z(t)) = {F (z(t)), H(z(t))} , t ∈ I.
dt
Proof. If z(t) = (q(t), p(t)) is a solution of ż = XH (z) one obtains for every F ∈ E(M )
d ∂F ∂F ∂F ∂H ∂F ∂H
Ḟ = F (z(t)) = q̇ + ṗ = − .
dt ∂q ∂p ∂q ∂p ∂p ∂q
Hence,
Ḟ = {F, H} .
1
The vectors X, X̄ are tangent vectors written as column vectors and ’⊤ ’ denotes transposition.
1.1 A Simple Hamiltonian System 5

The converse follows by choosing q j and pj for F:

∂H ∂H
q̇ j = {q j , H} = and ṗj = {pj , H} = − .
∂pj ∂q j

Corollary 1.6. F ∈ E(M ) is a first integral or Constant of Motion (”Be-

wegungskonstante”) if and only if

{F, H} = 0 .

Observation 1.7. The symplectic structure of M = T ∗ U ∼

= U × Rn is given either

1. by the symplectic involution σ, or

2. by the symplectic form ω,

3. by the Poisson bracket { , }.

What we have described so far in this chapter presents only the local models of
conservative classical mechanics where a configuration space can be detected as an
open subset U of Rn . For global considerations which are, in particular, needed for the
program of Geometric Quantization one has to redefine the above concepts for general
manifolds.
Note, that in Classical Mechanics the reduction of degrees of freedom by first in-
tegrals, by constraints or by symmetry considerations leads to general manifolds in a
natural way (cf. Subsection 1.3.2).

Example 1.8. The reduction of a Simple Hamiltonian System M = T ∗ U with Hamil-

tonian function H with respect to a first integral H (e.g. F = H): As a first step the
”surface” Sc := F −1 (c) for a c ∈ F (M ) is considered. When the gradient of F does not
vanish on Sc , the surface will be a (2n − 1)-dimensional manifold. Identifying points
in Sc which lie on the same solution of the canonical equations we get an equivalence
relation ∼ on Sc . The quotient Sc / ∼ is the space of orbits (= motions) with F = c. If
this quotient space is also a differentiable quotient the study of the Hamiltonian system
continues by investigating the reduced space Sc / ∼ of dimension 2n − 2. In general,
the reduced space will not be of the form T ∗ Rn−1 or an open subset thereof. But the
reduced space obtains a natural symplectic form (pushforward of ω) and so generates
a general Hamiltonian systems, as we explain in the next section. Concrete examples
of reduction are given in Section 1.3.
6 1. Hamiltonian Mechanics

1.2 Symplectic Manifolds and Hamiltonian Systems

In these lecture notes, a manifold will always be a differentiable (i.e. C ∞ -) real manifold
with countable topology and finite dimension. In most cases the manifold is also
assumed to be connected. Later we consider also complex manifolds.

Remark 1.9. In physics there appear also infinite dimensional manifolds having their
local models in a fixed Hilbert, Banach, or Fréchet space (see e.g. [AM78, Put93]).
In this course, however, to simplify matters, we concentrate on the finite dimensional
case.

Relevant manifolds in geometry and physics are

• Open subsets U ⊂ Rm

• Tangent and cotangent bundles T M , T ∗ M over a manifold M

• The rotation group SO(3) and other Lie groups

• Products M1 × M2 of manifolds M1 , M2

• Submanifolds of the above like the spheres Sn ⊂ Rn+1 (of radius 1), or matrix
groups like SO(3) ⊂ R3 × R3 × R3 ∼= R9

• Quotients of the above like the projective spaces Pn (R), Pn (C)

Exercise 1.10. Describe the projective spaces Pn (R), resp. Pn (C) as quotient mani-
folds (cf. A.10) of Rn+1 \ {0}, resp. Cn+1 \ {0} and of Sn ⊂ Rn , resp. S2n+1 ⊂ Cn+1 by
explicitly presenting suitable charts and confirming the universal property.

1.2.1 Notations for Manifolds

Let us recall some notations for manifolds and related basic concepts. (The notion of
a differentiable (or smooth) manifold as well as related concepts are collected together
in the Appendix, Section A on Manifolds, particularly in A.16, A.19, ff.):

Notation 1.11 (Local view of tangent vectors). Let M be an n-dimensional manifold.

1. The Charts on M defining the differentiable structure of M will often be denoted

as follows
q = (q 1 , q 2 , . . . , q n ) : U → V ,
where U ⊂ M is an open subset in M , V ⊂ Rn is an open subset of Rn and q
is differentiable with differentiable inverse. The q j : U → R, j = 1, . . . , n are the
(local) Coordinates given by the chart q.
1.2 Symplectic Manifolds and Hamiltonian Systems 7

2. Moreover, such a chart q provides for each a ∈ U a natural vector space basis
 
∂ ∂ ∂
(a), 2 (a) . . . , n (a)
∂q 1 ∂q ∂q
of the Tangent Space Ta M at a, where
∂
(a) := q −1 ((q(a) + tej )) a
 
∂q j

is the tangent vector of the curve q −1 (q(a) + tej ) through a ∈ U and where
(e1 , . . . , en ) is the standard unit vector space basis of Rn . Sometimes
∂ ∂
j
(a) is abbreviated as , ∂j (a) or ∂j ,
∂q ∂q j
whenever it is clear from the context for which chart q resp. for which point
a ∈ M the expressions are employed.

3. The corresponding vector fields

∂ ∂
j
: U → T U , a 7→ j (a) ,
∂q ∂q
can be used to represent every vector field X : U → T U over U through the
uniquely determined coefficients X j :
∂
X = Xj .
∂q j

4. For a vector field X over U the action of X on f (the Directional Derivative;

”Richtungsableitung”) is
d
LX f (a) := f ◦ x(t)|t=t0 ,
dt
where the curve x represents X(a) at the point a = x(t0 ): X(a) = [x]a .
With the abbrevation
 
∂f ∂ d
j
(a) = j
f (a) = f ◦ q −1 (q(a) + tej )|t=0
∂q ∂q dt
we obtain the formula for the action of X in local coordinates
∂f
LX f = X j .
∂q j
LX is often called Lie Derivative in the direction of X. Sometimes the notation
Xf instead of LX f is used.
8 1. Hamiltonian Mechanics

Notation 1.12 (Local view of cotangent vectors). Let M be again an n-dimensional

manifold.

1. A chart q : U → V provides for each a ∈ U a natural vector space basis

dq 1 (a), dq 2 (a) . . . , dq n (a)

of the Cotangent Space Ta∗ M = (Ta M )∗ of M at a, where

 
j d j
dq (a) ([x]a ) := (q ◦ x)|t=t0 , (6)
dt

when a = x(t0 ). For convenience, dq j (a) is often abbreviated as dq j when it is

clear for which point a the expressions are employed.
The basis is dual to the above basis if Ta M :
∂
dq k ( ) = δjk .
∂q j

2. Note that
dq j : U → T ∗ U , a 7→ dq j (a) ,
is a 1-Form on U . Moreover, every 1-form α : U → T ∗ U over U can be described
uniquely by
α = αj dq j ,
where the coefficients αj are smooth. αj can be obtained by
 
∂
αj (a) = α(a) (a) = α(∂j )(a) .
∂q j

And dqj (X) = X j for a vector field X ∈ V(U ).

3. A smooth function f : W → R on an open subset W ⊂ M induces a 1-form df

in the same way as dq j (cf. formula (6)):
 
d
df (a) ([x]a ) := (f ◦ x(t))|t=t0 .
dt
In local coordinates
∂f j
df = dq .
∂q j
When the 1-form df is applied to the vector field X the result will be:
∂f j
df (X) = X .
∂q j
In particular, df (X) = LX (f ).
1.2 Symplectic Manifolds and Hamiltonian Systems 9

Notation 1.13 (Global view of fields and forms).

1. For manifolds M, N

E(M, N ) = {f : M → N | f smooth }

denotes the set of smooth mappings from M to N . And

E(M ) = E(M, K)

denotes the corresponding set of functions, where K ∈ {R, C}. Pointwise addition
and multiplication defines on E(M ) the structure of a commutative K-algebra
over K (i.e. a K-vector space with commutative ring multiplication).
From the point of physics, the smooth functions f ∈ E(M, K) are the Classical
Observables.

2. For a commutative algebra R, a Derivation is a K-linear map D : R → R with

D(f g) = D(f )g + f D(g) for all f, g ∈ R. The set Der(R) of derivations of R
is a natural R-module by pointwise addition and multiplication. Moreover, with
respect to the commutator [D, D′ ] : D ◦ D′ − D′ ◦ D for D, D′ ∈ Der(R) this
R-module is also a Lie-algebra over K.
Applied to R = E(M ) we see that every vector field X can be interpreted to be a
derivation since the Lie derivative LX is, in fact, a derivation. Conversely, every
derivation D ∈ Der((M )) is induced by a unique vector field X, i.e. D = LX .
As a result, the Lie algebra Der(E(M )) can be identified with the space of vector
fı́elds, i.e. V(M ) = Der(E(M ). In this way V(M ) obtains the structure of a Lie
algebra: [X, Y ] is the unique vector field satisfying

[X, Y ] = LX ◦ LY − LY ◦ LX

for X, Y ∈ V(M ).

3. The s-forms (Differential Forms of degree s, s ∈ N) on M are the maps η :

(V(M ))s → E(M ) which are s-multilinear over the ring E(M ) and alternating2 .
Finally, the E(M )-module of all s-forms is denoted by

As (M ) = {η : (V(M ))s → E(M ) | s-multilinear over E(M ) and alternating }

Particular cases:
A0 (M ) = E(M ) are the 0-forms or functions, and
A1 (M ) =: A(M ) are the 1-forms.
2
In the same way one can define the (r, s)-tensor fields.
10 1. Hamiltonian Mechanics

4. The Wedge Product α ∧ β of two 1-forms α, β ∈ A(M ) is given by

α ∧ β := α ⊗ β − β ⊗ α ,

i.e. α ∧ β(X, Y ) = α(X)β(Y ) − β(X)α(Y ) for X, Y ∈ V(M ). Similarly one

obtains the general wedge product

∧ : Ar (M ) × As (M ) → Ar+s (M ) .

5. The Exterior Derivative

d = ds : As (M ) → As+1 (M )

is globally given by
s
X  
j
dη(X0 , X1 , . . . , Xs ) := (−1) LXj η(X0 , . . . Xj , . . . , Xs ) +
c
j=0

X
+ (−1)i+j η([Xi , Xj ], X0 , . . . X
ci , . . . , X
cj , . . . , Xs ).
i<j

Here, X
cj means that Xj has to be deleted.
For a 1-form α ∈ A(M ) the definition leads to

dα(X, Y ) = LX (α(Y )) − LY (α(X)) − α ([X, Y ]) (7)

for X, Y ∈ V(M ).
In local coordinates q : U → V , η ∈ As (U ) has the presentation
X
η= ηj1 j2 ...js dq j1 ∧ . . . ∧ dq js
j1 <j2 <...<js

and the exterior derivative of η is then

n
X X ∂ηj 1 j2 ...js
dη = dxj ∧ dxj1 ∧ . . . ∧ dxjs . (8)
j1 <j2 <...<js i=1
∂xj

1.2.2 Symplectic Manifolds

Definition 1.14. A Symplectic Form (”Symplektische Form”) on a manifold M

is a 2-form ω ∈ A2 (M ) which is non-degenerate, i.e. ω(a) : Ta M × Ta M → R is
non-degenerate for all a ∈ M , and which is closed, i.e. dω = 0.
1.2 Symplectic Manifolds and Hamiltonian Systems 11

Recall from Linear Algebra that a bilinear map g : V × V → R on a finite dimen-

sional real vector space is non-degenerate if for each v ∈ V the condition g(v, w) = 0
for all w ∈ W implies v = 0. Or, equivalently if and only if the induced map
g ♭ : V → V ∨ , v 7→ (w 7→ g(v, w)) , v, w ∈ V,
is an isomorphism of vector spaces. Here, V ∨ denotes the dual of a vector space, the
space of linear forms, also denoted by V ∗ .
Also equivalent when we describe g with respect to a basis ei by g(ei , ej ) =: gij is the
condition that the matrix (gij ) has non-zero determinant.
In the case of g being alternating, this implies that the dimension of V has to be
even. In fact,
det(gij ) = det (gij )T = det(−(gij )) = (−1)d det(gij )

if dim V = d, hence 1 = (−1)d when det(gij ) ̸= 0.

Observation 1.15. As a result, if ω is a symplectic form on M the dimension of M
is even, as we have seen in the fundamental example of the special symplectic form
ω = dq j ∧ dpj on the phase space T ∗ U (see 1.3).
Definition 1.16. A symplectic manifold (”Symplektische Mannigfaltigkeit”) is a
manifold M together with a symplectic form ω.
The corresponding maps which preserve the symplectic structure are the Canoni-
cal Transformations (”Kanonische Transformation”) or Symplectomorphisms
(”Symplektomorphismus”), i.e. the diffeomorphisms Φ : M → M ′ between symplectic
manifolds (M, ω) and (M ′ , ω ′ ) which satisfy Φ∗ ω ′ = ω. Recall (cf. Definition A.33:
Φ∗ ω ′ (X, Y )|a = ω ′ (a)(Ta Φ(X), Ta Φ(Y )) , for X, Y ∈ Ta (M ) , a ∈ M .

For a symplectic manifold (M, ω) we know that the dimension of M ,

dim M = dimR Ta M ∈ 2N ,
is even and we write, in general, dim M = 2n.
Examples are the phase spaces M = T ∗ U for open U ⊂ Rn as considered in Section
1.1. We call such an example a Simple Phase Space. Slightly more general examples
are the cotangent bundles M = T ∗ Q of general manifolds Q as will be explained below
in the next example. Further examples occur as quotients of M = T ∗ Q in the process
of reduction of degrees of freedom. Other examples are coadjoint orbits (see Section
1.3.5) and Kähler manifolds (see 9.28).
Construction 1.17 (Cotangent bundle). Let Q be a manifold of dimension n. Recall
that the Cotangent Bundle T ∗ Q is given (as a set) by
[
T ∗ Q := Ta∗ Q
a∈Q
12 1. Hamiltonian Mechanics

Ta∗ Q := (Ta Q)∗ = HomR (Ta Q, R) , with projection τ ∗ : T ∗ Q → Q , τ ∗ (Ta∗ Q) = {a} , a ∈

Q. The structure of a 2n-dimensional manifold on the cotangent bundle M := T ∗ Q is
defined by the bundle charts (cf. A.19)
q̃ = (q 1 , . . . q n , p1 , . . . , pn ) : T ∗ U → V × Rn ,
induced by the charts q : U → V of the manifold Q, U ⊂ Q open, where
 
∂
pj (µ) = µ , µ ∈ T ∗U .
∂q j

The Liouville form on M = T ∗ Q is, by definition,

λ := pj dq j
in local bundle coordinates.
It can be defined globally: For X ∈ Tµ M and µ ∈ Ta∗ Q ⊂ M we define λµ : Tµ M →
R by
λµ (X) := µ (Tµ τ ∗ (X)) ,
where Tµ τ ∗ : Tµ M → Ta Q is the tangent map (derivative) of τ ∗ at µ ∈ M (see
Observation A.14). The same formula is well-defined for vector fields X ∈ V(M ) and
1-forms µ ∈ A(Q) providing a map λ : V(M ) → E(M ) , µ 7→ λµ (X). Since λ is
E(M )-linear, it is a 1-form λ ∈ A(M ).
With respect to bundle charts (q, p) : T ∗ U → V × Rn one sees
λ|T ∗ U := pj dq j .

In fact, µ ∈ A(Q) and X ∈ V(M ) have locally the following representations:

∂ ∂
µ = µk dq k and X = X j j
+ Xk .
∂q ∂pk
Hence,
∂
Tµ τ ∗ (X)(τ ∗ (µ)) = [τ ∗ ◦ X]τ ∗ (µ) = X j |τ ∗ (µ) ,
∂q j
with respect to bundle coordinates, which implies λµ (X) = µ (Tµ τ ∗ (X)) = µj X j =
(pj dq j )µ (X). Therefore, λ|U = pj dq j .
Note, that λ can be characterized by the following property: λ ∈ A(T ∗ Q) is the
unique one form on T ∗ Q such that for any one form α ∈ A(Q) on Q the pullback of λ
via α gives back λ:
α∗ λ = λ .

The corresponding natural Symplectic Form ω on the cotangent bundle is ω :=

−dλ. In local bundles charts ω has the form
ω|T ∗ U = dq j ∧ dpj .
1.2 Symplectic Manifolds and Hamiltonian Systems 13

The local expression shows that ω is indeed non-degenerate and closed. Therefore,
(T ∗ Q, ω) is a symplectic manifold, often called Momentum Phase Space.

Remark 1.18. In the case of the cotangent bundle T ∗ Q the symplectic form is exact:
dω = −ddλ = 0. The 1-form −λ is called the Symplectic Potential.

In general, since a symplectic form ω is closed by definition, ω has a symplectic

potential locally, i.e. (by the Lemma of Poincaré) for each point a ∈ M there exists a
neighbourhood U and a 1-form α ∈ A(U ) with dα = ω|U . A global potential always
exists if H2dR (M, R) = 0. But many naturally defined symplectic manifolds do not have
a symplectic potential, this holds, for instance, for all compact symplectic manifolds.
Locally all symplectic manifolds look like open subspaces of the simple phase spaces
T U∼
∗
= U × Rn with ω = dq j ∧ dpj as the following result confirms.

Theorem 1.19 (Darboux’s Theorem). Every point a ∈ M of a symplectic manifold

(M, ω) has an open neighbourhood U ⊂ M and a chart

φ = (q, p) = (q 1 , . . . , q n , p1 , . . . , pn ) : U → V ⊂ T ∗ Rn ∼
= Rn × Rn ,

such that in these coordinates

ω|U = dq j ∧ dpj .

The proof of the theorem can be found e.g. in [LM87] or [Put93].

The q j , pj are called (local) Canonical Coordinates. Note, that the chart
Φ = (q, p) : U → V is a canonical transformation from (U, ω|U ) to (V, ω ′ |V ), where
ω ′ is the standard symplectic form on T ∗ Rn , ω ′ = dq j ∧ dpj or ω ′ (X, Y ) = X ⊤ JσY
(cf. Definition 1.3). As a consequence, (U, ω|U ) is symplectomorphically equivalent to
the open subspace (V, ω ′ ) of T ∗ Rn with the standard symplectic form.

Remark 1.20. 1. This result is in sharp contrast to Riemannian geometry: In case of

a semi-Riemannian manifold (M,P g) in every point one can find a chart q such that the
metric tensor g(a) has the form nj=0 ηj dq j ⊗ dq j with ηj ∈ {+1, −1}. In general, this
cannot be achieved in a full neighbourhood of a. The measure of this deviation from
the ”flat” case is the curvature of the Riemannian manifold at a ∈ M . In this sense a
symplectic manifold has no curvature, it is locally flat.
2. Moreover, this result allows to transfer the notions of Hamiltonian vector fields
XH and that of Poisson brackets { , } locally to a symplectic manifold. That these
notions have a global description will be shown in Subsection 1.2.3 below.

3. As we will see in the next subsection, a symplectic manifold serves as a general

phase space for Hamiltonian Mechanics. The case of a cotangent bundle M = T ∗ Q
will be called Momentum Phase Space with respect to the configuration space
Q. In M = T ∗ Q we have a special class of canonical coordinates, the bundle charts
14 1. Hamiltonian Mechanics

generated by the charts of Q: To each chart q : U → V we have the bundle chart

(q 1 , . . . , q n , p1 , . . . , pn ) : T ∗ U → V × Rn , where
 
∂
pj (µ) := µ , µ ∈ T ∗U .
∂q j
The pj are called Generalized Momenta, and this explains why T ∗ Q is called
momentum phase space. For a general symplectic manifold we have local canonical
coordinates q, p according the the theorem of Darboux. But these q and p are inter-
changeable and neither of the two can be regarded to describe momenta.
Since in the case of a symplectic manifold the symplectic form ω is assumed to be
non-degenerate, at each point a ∈ Q we obtain vector space isomorphisms according
to Definiton 1.14:

ω ♭ (a) : Ta M → Ta∗ M , X 7→ (Y 7→ ω(X, Y )) , X, Y ∈ Ta M ,

and its inverses ω ♯ := (ω ♭ )−1 . These isomorphisms induce a vector bundle isomorphisms
ω ♭ : T M → T ∗ M and ω ♯ : T ∗ M → T M .
The following proposition is easy to show
Proposition 1.21. Let ω ∈ A2 (M ) be a 2-form. The following conditions are equiva-
lent:

1. ω is non-degenerate

2. For every X ∈ V(M ): X vanishes everywhere ⇐⇒ ω(X, Y ) = 0 for all Y ∈

V(M ).

3. ω ♭ : T M → T ∗ M is a vector bundle isomorphism.

4. ω n = ω ∧ ω ∧ . . . ∧ ω is a nowhere vanishing 2n-form. Thus it is a volume form.

1.2.3 Hamiltonian Systems

Definition 1.22. Let (M, ω) be a symplectic manifold. To every observable H ∈ E(M )

there corresponds the Hamiltonian Vector Field

XH := ω ♯ ◦ dH.

The diagram
M
XH
dH
 $
T ∗M / TM
ω♯

is commutative and illustrates the definition of XH .

1.2 Symplectic Manifolds and Hamiltonian Systems 15

Observation 1.23. XH := ω ♯ ◦ dH implies ω ♭ ◦ XH = dH. Hence, XH is also deter-

mined as being the unique vector field satisfying

ω(XH , Y ) = dH(Y ) for all Y ∈ V(M ).

Proposition 1.24. In local canonical coordinates (q, p) : U → V the Hamiltonian

vector field XH can be written as

∂H ∂ ∂H ∂
XH |U = k
− j . (9)
∂pk ∂q ∂q ∂pj

Proof. In these local canonical coordinates ω|U has the form dq j ∧ dpj (cf. Theorem
1.19). With the use of the representation
∂ ∂
Y =Yj + Ȳ j
∂q j ∂pj
of vector fields Y on U we deduce
   
♭ ∂ ∂
ω (Y ) = ω , Y = Ȳj ,
∂q j ∂q j
which implies  
♭ ∂
ω = dpj .
∂q j
Similarly, using  
♭ ∂
ω (Y ) = −Y j
∂pj
one deduces  
♭ ∂
ω = −dq j .
∂pj
As a consequence, ω ♯ acts with respect to the basis (dq j , dpk ) of Ta∗ U as
∂ ∂
ω ♯ (dq j ) = − ; ω ♯ (dpk ) = k .
∂pj ∂q

Thus, in local canonical coordinates ω ♯ has the form

∂ ∂
ω ♯ (αj dq j + ᾱk dpk ) = ᾱk k
− αj . (10)
∂q ∂pj

(This resembles the symplectic involution σ(α, ᾱ) = (ᾱ, −α), cf. formula (3)). Now,
∂H j ∂H
dH = dq + dpk
∂q j ∂pk
16 1. Hamiltonian Mechanics

gives the desired result

 
♯ ∂H j ∂H ∂H ∂ ∂H ∂
XH |U = ω dq + dpk = − j .
∂q j ∂pk ∂pk ∂q k ∂q ∂pj

Definition 1.25. (X, ω, H) is called a Hamiltonian System whenever ω is a sym-

plectic form. A Motion of the Hamiltonian system is a curve z ∈ E(I, U ) on an open
interval I satisfying

ż = XH (z) ,

where ż(t) := [z]z(t) is the tangent vector given by the curve z in the point z(t).

Using
dq j ∂ dpk ∂
ż(t) = (q̇(t), ṗ(t)) = j
+
dt ∂q dt ∂pk
in local canonical coordinates the preceding Proposition 1.24 immediately implies:
Corollary 1.26. In local canonical coordinates (q, p) the equations of motion for
z(t) =: (q(t), p(t)) have the form

∂H ∂H
q̇ = , ṗ = − .
∂p ∂q

As before in the simple case where M = T ∗ U , U ⊂ Rn , the symplectic form induces

Poisson brackets on a general symlpectic manifold (M, ω):
Definition 1.27. The Poisson Bracket on E(M ), given by the symplectic form
ω ∈ A2 (M ), is defined as
{F, G} := ω(XF , XG ) , F, G ∈ E(M ).
Proposition 1.28 (Equations of motion in Poisson form). The equations of motion
in case of a Hamiltonian system can again be written in the so called Poisson Form

Ḟ = {F, H}.

Proof. In local canonical coordinates (q, p) : U → V ⊂ Rn × Rn (cf. theorem 1.19) the

Poisson bracket has the form

∂F ∂G ∂F ∂G ∂F ∂G ∂F ∂G
{F, G}|U = − = j − (11)
∂q ∂p ∂p ∂q ∂q ∂pj ∂pj ∂q j
1.2 Symplectic Manifolds and Hamiltonian Systems 17

which is well-known from the simple case. Hence, the proof reduces to the proof of
Proposition 1.5.
Corollary 1.29. An observable F ∈ E(M ) is a first integral of the Hamiltonian system
(M, ω, H) if and only if {F, H} = 0.

1.2.4 Hamiltonian Vector Fields

We study relations between the Poisson brackets and Hamiltonian vector fields. The
fact that the Poisson bracket satisfies the Jacobi identity and therefore induces on
E(M ) the structure of a Lie algebra is of fundamental importance for the program of
Geometric Quantization. We provide two proofs of this result.
Theorem 1.30. The Poisson bracket { , } : E(M ) × E(M ) → E(M ) of a symplectic
manifold (M, ω) is a Lie bracket, in other words E(M ) with the Poisson bracket is a
Lie algebra over R3 , the Poisson Algebra, that is

1. { , } is bilinear over R.
2. {F, G} = −{G, F } for F, G ∈ E(M ), i.e. { , } is alternating.
3. {F, {G, H}} + {G, {H, F }} + {H, {F, G}} = 0 for F, G, H ∈ E(M ). ( Jacobi
Identity)

In addition:
4. {F, GH} = G{F, H} + {F, G}H = G{F, H} + H{F, G} ( Product Rule).
Equivalently, {F, } is a derivation on E(M ).
5. For connected M : G ∈ E(M ) is constant, iff {F, G} = 0 for all F ∈ E(M )
( Completeness).
In general: {F, G} = 0 for all F ∈ E(M ) iff dG = 0.

Proof. 1. and 2. are immediately clear.

First proof of 3.: We apply formula (11) to obtain in local canonical coordinates
∂F ∂ ∂F ∂
{F, {G, H}} = {G, H} − {G, H}
∂q ∂p ∂p ∂q
   
∂F ∂ ∂G ∂H ∂G ∂H ∂F ∂ ∂G ∂H ∂G ∂H
= − − −
∂q ∂p ∂q ∂p ∂p ∂q ∂p ∂q ∂q ∂p ∂p ∂q
 2 2 2 2

∂F ∂ G ∂H ∂G ∂ H ∂ G ∂H ∂G ∂ H
= + − −
∂q ∂p∂q ∂p ∂q ∂p2 ∂p2 ∂q ∂p ∂p∂q
 2 2
∂ G ∂H ∂G ∂ 2 H
2

∂F ∂ G ∂H ∂G ∂ H
− + − −
∂p ∂q 2 ∂p ∂q ∂q∂p ∂q∂p ∂q ∂p ∂q 2
3
the Lie algebra properties are 1.-3.
18 1. Hamiltonian Mechanics

In the same way we get expressions for {G, {H, F }} and {H, {F, G}}. Summing up all
the terms one sees that the Jacobi identity is satisfied.
Second proof of 3.: A more conceptual proof which, moreover, does not P use local
canonical coordinates, P
is the following: We introduce the cyclic summation ijk Tijk
of summable terms as ijk Tijk := Tijk + Tjki + Tkij . In particular, he Jacobi identity
has the form X
F, G, H = 0 .
F GH
The exterior derivative of ω vanishes, hence, from
0 = dω(XF , XG , XH )
= LXF (ω(XG , XH )) − LXG (ω(XF , XH )) + LXH (ω(XF , XG ))
− ω(([XF , XG ] , XH ) + ω(([XF , XH ] , XG ) − ω(([XG , XH ] , XF )
we obtain
0 = LXF (ω(XG , XH )) − ω(([XF , XG ] , XH )
+LXG (ω(XH , XF )) − ω(([XG , XH ] , XF )
+LX (ω(XF , XG )) − ω(([XH , XF ] , XG )
X H
= LXF (ω(XG , XH )) − ω(([XF , XG ] , XH ) .
F GH

Now,
LXF (ω(XG , XH )) = {ω(XG , XH ), F } = {{G, H}, F } = −{F, {G, H}}
since, in general,
LXF I = dI(XF ) = ω(XI , XF ) = {I, F } ,
for a function I ∈ E(M ). Applying this again we obtain
−ω ([XF , XG ] , XH ) = L[XF ,XG ] H = LXF LXG H − LXG LXF H
= {{H, G}, F } − {{H, F }, G}
= {F, {G, H}} + {G, {H, F }} (12)
Implementing these identities in the above cyclic sum gives
X
0 = LXF (ω(XG , XH )) − ω(([XF , XG ] , XH )
XF GH
= −{F, {G, H} + {F, {G, H}} + {G, {H, F }}
XF GH X
= {G, {H, F } = {F, {G, H}.
F GH F GH

And this is the Jacobi identity!

The statement in 4. follows immediately from the chain rule d(GH) = HdG+GdH.
To show 5., observe that for G ∈ E(M ) the condition {F, G} = ω(XF , XG ) = 0 for
all F ∈ E(M ) is equivalent to XG = 0 by the non-degeneracy of ω and this is in turn
equivalent to dG = 0.
1.2 Symplectic Manifolds and Hamiltonian Systems 19

The statement of 3. in the preceding theorem is essentially equivalent to the next

proposition.
Corollary 1.31. The mapping
Φ : E(M ) → V(M ) , F 7→ −XF ,
is a Lie algebra homomorphism, i.e. Φ is R-linear and satisfies
Φ({F, G}) = −X{F,G} = [XF , XG ] = [Φ(F ), Φ(G)] .

Proof. For F, G, H ∈ E(M ) we just have shown in formula (12):

L[XF ,YG ] H = {F, {G, H}} + {G, {H, F }}.
By the Jacobi identity this is −{H, {F, G}} and we conclude
[XF , XG ] H = −X{F,G} H.

Corollary 1.32. The Lie bracket of two first integrals is again a first integral.

Proof. Let F, G be first integrals of (M, ω, H). Then {G, H} = {F, H} = 0 (cf. Corol-
lary 1.29). By the Jacobi identity,
{{F, G}, H} = −{H, {F, G}} = {F, {G, H}} + {G, {H, F }} = 0.
As a consequence, {F, G} is a first integral by Corollary 1.29.
Observation 1.33. The second proof of the Jacobi identity (cf. proof of 3. in The-
orem 1.30) yields more than merely the identity. Note, that for a non-degenerate
(and not necessarily closed) two-form ω ∈ A2 (M ) on a manifold M the generation of
Hamiltonian vector fields XH and the introduction of the Poisson bracket is possible
in the same way as it is done in the preceding subsections. The above mentioned
second proof of 3. in Theorem 1.30 now shows that for F, G, H ∈ E(M ) the state-
ment
P dω(XF , XG , XH ) = 0 is equivalent to F, G, H satisfying the Jacobi identity, i.e.
F GH {F, {G, H}} = 0. Since the Hamiltonian vector fields generate the tangent
spaces Ta M we have proven the following remarkable result:
Proposition 1.34. A manifold M with a non-degenerate ω ∈ A2 (M ) is symplectic if
and only if the Poisson bracket induced by ω satisfies the Jacobi identity.
Observation 1.35. As a result, the Hamiltonian vector fields
Ham(M ) := {XF | F ∈ E(M )}
form a Lie subalgebra of the Lie algebra V(M ) of vector fields. The kernel Ker Φ ⊂
E(M ) consists of the locally constant functions. Hence, for connected manifolds M one
has R = Ker Φ and one obtains the following exact sequence of Lie algebras

Φ
0 −→ R −→ E(M ) −→ Ham(M ) −→ 0, .
20 1. Hamiltonian Mechanics

This exact sequence exhibits the Lie algebra E(M ) as a central extension of
the Lie algebra Ham(M ) of Hamiltonian vector fields.

Locally Hamiltonian vector fields form a Lie algebra (subalgebra of V(M )) as well
and have Ham(M ) as an ideal.

1.3 Examples of Hamiltonian Systems

1.3.1 Harmonic Oscillator

In this case the phase space reads:

M = T ∗ Rn ∼
= Rn × Rn = R2n .

The symplectic form is given by:

ω = dq j ∧ dpj ,

The hamiltonian (total energy) takes the form:

1 ∂ ∂
||q||2 + ||p||2 , XH = pj j − q j


H(q, p) = .
2 ∂q ∂pj

The equations of motion are well known:

q̇ = p, ṗ = −q.

H is a first integral, that is, every motion (q, p) : I → M with H(q(t0 ), p(t0 )) = E ≥ 0
satisfies
H(q(t), p(t)) = E ∀t ∈ I.
Hence, the points (q(t), p(t)) of the solution (q, p) remain in the hypersurface

ΣE = H −1 (E)

for all t ∈ I. ΣE is in fact a submanifold of dimension 2n − 1 since ∇H = (q, p) ̸= 0

for E > 0. We see
√ 
ΣE = (q, p) | ||q||2 + ||p||2 = 2E = S2n−1

2E ,

where
Sk−1 (r) := x ∈ Rk | ||x||2 = r2


denotes the (k − 1)-sphere of radius r in Rk . This example is a reduction in the sense

of the following subsection.
1.3 Examples of Hamiltonian Systems 21

1.3.2 Reduction with respect to first integrals

Let F be a first integral of a Hamiltonian system (M, ω, H), i.e. F ∈ E(M ) and
{F, H} = 0 (c.f. Corollary 1.29). Let c ∈ R be a value with
Σc := F −1 (c) ̸= ∅ ,
i.e. c ∈ F (M ). Assume that the level set Σc is a smooth hypersurface, this holds e.g. if
∇F ̸= 0 on Σc . Then the space of orbits with F = c is the quotient
Oc := Σc / ∼,
with respect to the equivalence relation
a ∼ b ⇐⇒ ∃ motion x : I → Σc with x(t1 ) = a and x(t2 ) = b for t1 , t2 ∈ I.

Assume, moreover, that the orbit space Oc has a differentiable structure, that is, the
differentiable quotient exists as a manifold (cf. A.10). Then it is a (2n−2)-dimensional
manifold. Furthermore, assume that ω|Σc induces on Oc a natural symplectic form
ωc ∈ A2 (Oc ) (such that ω|Oc = π ∗ (ωc ) for the projection π : Σc → Oc ).
Since the hamiltonian H is constant on the orbits it descends to Oc as Hc ∈ E(Oc )
with H = Hc ◦ π on Σc .
As a result, the original system (M, ω, H) has been reduced (by one degree of freedom)
to (Oc , ωc , Hc ). In general, this procedure can be repeated. In good cases (”completely
integrable systems”) one can go down to n-dimensional reductions, which then gives
the solution.
In case of the harmonic oscillator of dimension n (see above), the orbit space OE 4
√ 
2n−1
ΣE / ∼ = S 2E / ∼

is isomorphic to the complex projective space Pn−1 (C) of all complex lines going through
the origin 0 ∈ Cn in Cn . And the symplectic form ωE is the usual Kähler form on
Pn−1 (C).
Here, we introduce complex coordinates (resp. the structure of a complex vector
space on Rn × Rn ) by defining z := p + iq, z j := pj + iq j . Observe that the canonical
equations (q̇, ṗ) = (p, −q) are now
ż = iz.
Note, that multiplication by i is the symplectic involution of Subsection 1.1: i = σ .
The motions are z(t) = eit z0 t ∈ R , where z0 = z(0) ∈ Cn . Moreover, the observ-
ables
1 2
1 j 2 1 j j
Hj := pj + (q j )2 = z = z z̄ , j = 1, . . . , n, (no summation!)
2 2 2
4
The equivalence relation given by the orbits is in this case also given by the action of the group
U(1) ∼
= S1 , so that OE = S2n+1 /U(1).
22 1. Hamiltonian Mechanics

on R2n are first integrals:

d 1
Hj (z(t)) = (ż j z̄ j + z j z̄˙ j )
dt 2
1
= (iz j z̄ j + z j (−iz̄ j )) = 0.
2

With the values E =

P ⃗ := (E1 , . . . , En ),
Ej , Ej = Hj (z(t0 )), for some t0 ∈ I, and E
we obtain the level set for (H1 , . . . , Hn ), an n-dimensional manifold,
n
\ n
Y p 
ME⃗ = Hj−1 (Ej ) = S1 2Ej .
j=1 j=1

which is an n dimensional torus. Again, the level ME⃗ set is invariant in the sense that
the motion z(t) remains in ME⃗ , if z(t0 ) ∈ ME⃗ .
This ”reduction”
p
gives a complete solution: Every motion z = z(t) satisfies z j (t) ∈
S 2Ej for j = 1, . . . , n and all t ∈ I. It is determined by z(t0 ), and if t0 = 0 it will
be of the form

z j (t) = eit z j (0) = cos t pj (0) − sin t q j (0) + i(sin t pj (0) + cos(t) q j )

or
q j (t), pj (t) = cos t q j (0) + sin t pj (0), cos t pj (0) − sin t q j (0) .

This is a rather simple example of reduction of a completely integrable system. A

completely integrable system is a hamiltonian system of dimension 2n with n
first integrals in involution, i.e. {Fj , Fk } = 0 for 1 ≤ j, k ≤ n which are independent,
i.e. the map F = (F1 , . . . , Fn ) : M → Rn is of rank n. The theorem of Arnold-Liouville
says that in case the level sets are compact they are finite unions of tori.

1.3.3 Kepler Problem (Hydrogen Atom)

In this case the configuration and phase space is given by:

Q = R3 \ {0}, M = T ∗ Q = Q × R3 .

The symplectic form is the usual one:

ω = dq j ∧ dpj .

And the hamiltonian reads:

1 k
H(q, p) = ||p||2 − , m, k > 0.
2m ||q||
1.3 Examples of Hamiltonian Systems 23

We have  
kq p
∇H = 3
,− ̸= 0
||q|| m
in all of M . Hence, the energy hypersurface

ΣE = H −1 (E)

is a smooth submanifold of dimension 5 for all E ∈ R.

Let E ∈ ] − ∞, 0 [. The orbits in ΣE are ellipses and one can show (cf. Example
7.7), that the orbit space OE = ΣE / ∼ is isomorphic (as a differentiable manifold) to
S2 (mk) × S2 (mk). The symplectic form ω descends to a form ωE on ΣE / ∼. And on
SE := S2 (mk) × S2 (mk) it has the form
 
1 dx1 ∧ dx2 dy1 + dy2
ωE = + ,
2ρ x3 y3
√
with respect to the chart with x3 ̸= 0 ̸= y3 on SE . Here ρ = −2mE.
It is interesting to ask which energy values occur if we quantize the system (SE , ωE )
according to the program of Geometric Quantization.
In Example 7.7 we show: After adjusting the constants the energy levels – predicted by
Geometric Quantization – are EN = −2π 2 mℏ2 N −2 , N ∈ N, N ≥ 1, the values known
for the hydrogen atom from experiments!

1.3.4 Particle in a Field

Twist of the Cotangent Bundle

In this general example the phase space is the cotangent bundle

M = T ∗ Q.

of an n-dimensional manifold Q. However, the symplectic structure is not given by the

previously considered standard form

ω0 = dq j ∧ dpj ,

but by a Twist
ωF = ω := ω0 + τ ∗ F ,
where F ∈ A2 (Q) is a closed two-form and τ ∗ F is the pull-back with respect to the
natural projection τ : T ∗ Q ∼
= Q × Rn → Q. This change of the symplectic structure is
sometimes called Deformation of the Symplectic Structure.
As a special case we describe a Relativistic Charged Particle in the following
example
24 1. Hamiltonian Mechanics

Example 1.36. The configuration space is a spacetime Q with Lorentzian metric g,

for instance, an open subset Q of the Minkowski space R4 . The symplectic manifold
(T ∗ Q, ω0 ) is the phase space for a relativistic particle. The function
1
H :=g(p, p) , p ∈ T ∗ Q ,
2
determines the dynamics of a relativistic particle by its Hamiltonian equations.
Let us assume that in addition to the above structure an electromagnetic field in
form of a closed 2-form F ∈ A2 (Q) on Q is present. Then the Hamiltonian dynamics is
given by the Hamiltonian vector field XH with the same Hamiltonian function H but
with respect to the modified symplectic form
ω := ω0 + eτ ∗ (F ) ,
where e is the charge of the particle.

The configuration space Q can also be a Riemannian manifold with metric tensor
g, and for F one can take a geometrically induced closed 2-form.
Since F is closed, τ ∗ (F ) is closed as well, and consequently ωF is closed. This shows
one part of the following assertion.
Proposition 1.37. For a two-form F ∈ A2 (Q) the twisted two-form ωF := ω0 + τ ∗ F
is a symplectic form on T ∗ Q = M if and only if F is closed.

Proof. We just have seen, that ωF is closed, whenever F is closed. And, of course,
when ωF is closed, τ ∗ (F ) has to be closed and, in turn, F is closed.
To investigate the non-degeneracy we use local coordinates q in an open subset U
of Q and see
F |U = Fjk dq j ∧ dq k ,
where Fjk ∈ E(U ) are suitable functions. On V := τ −1 (U ) = U × Rn this implies with
respect to the canonical coordinates (q, p) (of the bundle chart tu q)
ωF |V = ω0 |V + τ ∗ (F )|V = dq j ∧ dpj + τ ∗ (Fjk dq j ∧ dq k )
= dq j ∧ dpj + τ ∗ Fjk dq j ∧ dq k = dq j ∧ dpj + (Fjk ◦ τ )dq j ∧ dq k

The action of ω0 on the tangent space Ta M ∼

= Rn × Rn at a ∈ V is given by the
block matrix  
0 1
−1 0
and, by the above local expression of ωF , the corresponding action of ωF at a is given
by the block matrix  
F (a) 1
,
−1 0
where F (a) is the matrix (Fjk (τ (a))). This expression shows that ωF is non-degenerate
and the proposition is proven.
1.3 Examples of Hamiltonian Systems 25

Twisted Canonical Coordinates

By Darboux’s theorem every symplectic form can locally be written as dq̃ k ∧dp̃k with
respect to local canonical coordinates q̃ j , p̃j . Let us find local canonical coordinates in
the case of the twisted symplectic form ωF on M = T ∗ Q. Locally, on suitable open
subsets U ⊂ Q the closed form F can be expressed as F |U = dA, where A ∈ A1 (U ) is
a one-form with A = Aj dq j . Then, in the coordinates (q, p) of the bundle chart, which
are canonical coordinates of ω0 , we have:

ωF |U = dq k ∧ dpk + τ ∗ dA = dq k ∧ dpk + τ ∗ d(Ak dq k )

= dq k ∧ dpk + τ ∗ (dAk ∧ dq k + 0) = dq k ∧ dpk + τ ∗ (dAk ) ∧ τ ∗ dq k
= dq k ∧ dpk + d(τ ∗ Ak ) ∧ d(τ ∗ q k ) = dq k ∧ dpk + d(τ ∗ Ak ) ∧ dq k
= dq k ∧ dpk − dq k ∧ d(τ ∗ Ak ) = dq k ∧ d(pk − τ ∗ Ak ).
This result implies, that the following definition yields canonical coordinates for the
twisted symplectic structure.

q̃ k = q k , p̃k = pk − τ ∗ Ak . (13)

The formulas can be interpreted as describing a particle in a generalized magnetic

field, where A corresponds to the vector potential of electrodynamics.

Twisted Hamiltonian Vector Field

We want to determine the Hamiltonian vector fields corresponding to ωF = ω and
to give a physical interpretation of the twist. We use local canonical coordinates (q, p)
with respect to the standard 2-form ω0 = dq j ∧ dpj . Let H ∈ E(M ). The Hamiltonian
0
vector field related to ω0 will be denoted by XH and the one related to ω will be denoted
0
by XH . According to Proposition 1.24 XH takes the form

0 ∂H ∂ ∂H ∂
XH U
= − . (14)
∂pk ∂q k ∂q j ∂pj
ω can be expanded locally as:

ω|U = dq j ∧ dpj + (Fjk ◦ τ )dq j ∧ dq k .

In the following, we omit ”◦ τ ” in order to have simpler formulas and understand Fjk as
functions on open subsets U ⊂ M . Similarly, we write Ak instead of τ ∗ (Ak ) = Ak ◦ τ .
The (local) transformation

(q̃, p̃) 7→ (q, p) = G(q̃, p̃) = (q̃, p̃ + A) ,

which is the inverse of (13), leads to the following identity for functions f ∈ E(U ):
∂f ∂f ∂Gj ∂f ∂Gj ∂f ∂ q̃ j ∂f ∂(p̃j + Aj ) ∂f ∂f ∂Aj
k
= j k
+ k
= j k
+ k
= k+
∂ q̃ ∂q ∂ q̃ ∂pj ∂ q̃ ∂q ∂ q̃ ∂pj ∂ q̃ ∂q ∂pj ∂q k
26 1. Hamiltonian Mechanics

since
∂Aj ∂Aj
k
= .
∂ q̃ ∂q k
As a consequence,
∂ ∂ ∂Aj ∂
k
= k+ k .
∂ q̃ ∂q ∂q ∂pj
In the same way, we obtain
∂ ∂
= .
∂ p̃k ∂pk
Applying these identities, the Hamiltonian XH can be expressed in the canonical
coordinates related to ω0 :

∂H ∂ ∂H ∂
XH = k
− j
∂ p̃k ∂ q̃ ∂ q̃ ∂ p̃j
   
∂H ∂ ∂Aj ∂ ∂H ∂Ak ∂H ∂
= k
+ k − j
+ j
∂pk ∂q ∂q ∂pj ∂q ∂q ∂pk ∂pj
   
∂H ∂ ∂H ∂ ∂H ∂Aj ∂Ak ∂H ∂
= − + −
∂pk ∂q k ∂q j ∂pj ∂pk ∂q k ∂q j ∂pk ∂pj
0 ∂H ∂
= XH + (Fkj − Fjk ) .
∂pk ∂pj
We have shown:

Lemma 1.38.
0 ∂H ∂
XH − XH = (Fkj − Fjk ) .
∂pk ∂pj

And the equations of motion in (M, ωF , H) with respect to the coordinates of the
bundle chart have the following form:

Proposition 1.39.
∂H ∂H ∂H
q̇ k = , ṗk = − k + (Fjk − Fkj ) .
∂pk ∂q ∂pj

Proof. It is easy to see that

∂H ∂H
q̇ k = q̃˙k = = .
∂ p̃k ∂pk
The second set of equations reads
∂H ∂H ∂Aj ∂H
p̃˙k = − k = − k − k .
∂ q̃ ∂q ∂q ∂pj
1.3 Examples of Hamiltonian Systems 27

Moreover,
d ∂Ak j ∂Ak ∂H
p̃˙k = ṗk − Ak = ṗk − j
q̇ = ṗk − .
dt ∂q ∂q j ∂pj
As a consequence,
∂Ak ∂H ∂H ∂Aj ∂H ∂H ∂H
ṗk = j
− k− k = − k + (Fjk − Fkj ) .
∂q ∂pj ∂q ∂q ∂pj ∂q ∂pj

Observation. As we have seen in the previous section, the Hamiltonian vector fields
0
determine the classical equations of motion. Therefore, the difference Φ := XH − XH
can be interpreted as a force. A possible physical interpretation: Φ looks like the
magnetic part of a generalized Lorentz force. In dimension 3 it can rewritten as v × B,
as we see in the special example below.

Charged Particle in R3
Let B = B j dq j ∈ V(R3 ) the divergence-free vector field representing the magnetic
field, where q j , j = 1, 2, 3, are the cartesian coordinates. The classical equations of
motion for a particle with charge e and mass m are given by the Lorentz Force
Law (we set c = 1):
dv
m = e(v × B). (15)
dt
We recover this law as the equations of motion of a suitable Hamiltonian system
which is given by the following following twisted symplectic form on M = T ∗ R3 ∼=
R3 × R3 :
ω := ω0 + eF , (16)
where
1
F := iB⃗ (dq 1 ∧ dq 2 ∧ dq 3 ) ,
2
such that
F = B 1 dq 2 ∧ dq 3 + B 2 dq 3 ∧ dq 1 + B 3 dq 1 ∧ dq 2 .

Note, that F is closed, since B is assumed to be divergence-free.

The Hamiltonian system is (M, ω, H) with H as the kinetic energy. The kinetic
energy is:
1
p21 + p22 + p23 ,


H=
2m
where p = mv.
According to Proposition 1.39 the equations of motion include in particular
∂H ∂H
ṗk = − k
+ e (Fjk − Fkj ) .
∂q ∂pj
28 1. Hamiltonian Mechanics

which is, in our case,

1
ṗk = e (Fjk − Fkj ) pj .
m
These three equations are equivalent to (15). We check the case k = 1:

mv̇1 = ṗ1 = e(Fj1 − F1j )vj = evj 2Fj1 = e(v2 B 3 − v3 B 2 ) = e(v × B)1 .

Remark 1.40. F has to be closed, which in this special case means that F is exact.
On more general spaces, not every two-form, which is closed, need be exact. F = dA
holds true only locally and, in general, not globally. However, we can understand A as
being a connection one-form on a U(1)-bundle or on a line bundle, and F as being the
curvature 2-form of the connection. This already is the topic of gauge theory treated
in the Chapters 3, 4 and 5.

1.3.5 Coadjoint Orbits

This class of examples of symplectic manifolds yields a close, but not obvious connec-
tion between the representation theory of Lie groups and Lie algebras and Geometric
Quantization.
In the following (see Appendix, Chapter C):

• G is a connected Lie group of finite dimension (for instance a closed matrix group
G ⊂ GL(k, R)).

• g = Lie G is the associated Lie algebra.

• the conjugation with respect to g ∈ G yields the smooth map

τg : G → G, x 7→ gxg −1 , x ∈ G,

• and the adjoint representation Ad : G → GL(g) of the group G is defined as the

derivative Te τg of τg at the unit e of G;

Adg := Te τg : Te G = g → g = Te G.

Adg is the map X 7→ gXg −1 in case of a matrix group G.

From Adgh = Adg ◦ Adh for g, h ∈ G, one deduces that

Ad : G → GL(g)

is a Lie group homomorphism.

1.4 Lagrangian Mechanics 29

Definition 1.41. The Coadjoint Representation is the ”dual” or ”adjoint” of

the adjoint representation:
Ad∗ : G → GL(g∗ ),
which is given by:
Ad∗g : g∗ → g∗ , µ → Ad∗g (µ) ∈ g∗ ,
with
Ad∗g (µ)(X) := µ (Adg−1 (X))
for µ ∈ g∗ = {ν : g → R | R − linear} and X ∈ g.

It is easy to check that A∗gh = Ad∗g ◦ Ad∗h , g, h ∈ G, i.e. A∗ is again Lie group
homomorphism.
As a result, we have an action of the Lie group G on the dual of its own Lie algebra
∗
g

G × g∗ → g∗ , (g, µ) 7→ Ad∗g (µ).

This is called the Coadjoint Action and has the orbits:

Mµ := Ad∗g µ | g ∈ G , µ ∈ g∗ .


One can show:

1. Mµ is a smooth submanifold of g∗ ∼
= Rm with a natural symplectic form ωµ and
with symmetry group G.

2. Every symplectic manifold M on which G acts transitively by symplectomor-

phisms looks locally like an open part of smooth orbit Mµ . More precisely,M can
be realized as a covering M → Mµ of Mµ .

Here ϕ : (M, ω) → (M ′ , ω ′ ) is a Symplectomorphism (or canonical transformation),

if ϕ is a diffeomorphism preserving the symplectic structures, i.e. with ϕ∗ ω ′ = ω.

1.4 Lagrangian Mechanics

In many situations a system of Classical Mechanics is given as a Lagrangian system.

In the following we discuss when a Lagrangian system induces a Hamiltonian system
in a natural way and vice versa.
The ingredients of a Lagrangian System (M, L) are:
An n-dimensional manifold Q as the Configuration Space.
The tangent bundle M = T Q, called the Velocity Phase Space.
And a function L ∈ E(M ), called the Lagrangian.
30 1. Hamiltonian Mechanics

Example 1.42. A Natural System5 is a Lagrangian system with a Lagrangian of

the form L(v) = 12 g(v, v) − U (τ (v)) , v ∈ T Q = M , where g is a Riemannian metric on
Q and U ∈ E(Q) is a so-called potential.

In local coordinates
∂
v = vj and g(v, w) = gjk v j wk , gjk = gjk (τ (v)) = gjk (q) ,
∂q j
v, w ∈ Tτ (v) Q. Hence,
1 1
g(v, v) = gjk v j v k .
2 2
Definition 1.43. A curve q : I → Q on an interval I ⊂ R is called a Motion of the
Lagrangian system (M, L), if q satisfies the Euler-Lagrange Equations, that is:
 
d ∂L ∂L
(q̇) = (q̇) (in bundle coordinates) .
dt ∂v ∂q

Here,
dq
q̇ = ∈ TQ.
dt
In the case of a natural system the equations have the form
d
∂U
gjk (q)q̇ k = j (q)
dt ∂q
or
∂gjk i k k ∂U
q̇ q + gjk q̈ = (q) .
∂q i ∂q j
Fact 1.44. In case of a natural system without potential, i.e. L(v) = 21 g(v, v) the
geodesics of the Riemannian manifold (Q, g) are essentially the motions in the level set
L−1 ( 12 ).
Construction 1.45. Given a chart

ϕ = q 1 , . . . , q n : U → V ⊂ Rn

with associated bundle chart

ϕ̃ = q 1 , . . . , q n , v 1 , . . . , v n : T U → V × Rn

the local 1-form, called Liouville Form induced by L is:

 
∂L k ∂L
λL := k dq , ”generalized momenta”
∂v ∂v k
5
called ”mechanical” system in [Arn89]
1.4 Lagrangian Mechanics 31

defining a 2-form

ωL := −dλL
∂ 2L ∂ 2L
= − j k dq j ∧ dq k − j k dv j ∧ dq k .
∂q ∂v ∂v ∂v
ωL is well-defined on all of M and it is closed. Therefore it is a symplectic form if it is
non-degenerate.

Let us call L (or the Lagrangian system (T Q, L)) regular when ωL is non-degenerate.
For a regular Lagrangian L, (M, ωL ) is a symplectic manifold and with
∂L
HL (v) := v k (v) − L(v) ,
∂v k
the Hamiltonian system
(M, ωL , HL )
has the same motions as (M, L).

Example 1.46. For a natural system the induced Hamiltonian HL is of the form
1
HL (v) = g(v, v) + U (τ (v)) ,
2
v ∈ M = T Q.

To understand regular Lagrangian systems from the Hamiltonian viewpoint the

so-called fibre derivative of L is helpful: For L ∈ E(T Q, R) let Lq := L|Tq Q : Tq Q → R.

Definition 1.47. The map

F L : T Q → T ∗Q ,
given by F L(v)(w) := T Lq (v)(w) for q ∈ Q , v, w ∈ Tq Q, is called the fibre derivative
of L

By direct calculations we can show:

Proposition 1.48. The fibre derivative is a smooth map F L : T Q → T ∗ Q preserving

the fibres.6
ωL = F L∗ (ω) where ω ∈ A2 (T ∗ Q) is the standard 2-form on the momentum phase
space T ∗ Q, see 1.17.
The following statements are equivalent:

1. ωL is non-degenerate, i.e. L is regular.

6
F L is not a homomorphism of vector bundles, in general.
32 1. Hamiltonian Mechanics

2. The fibre derivative F L is non-degenerate, i.e. its differential T (F L) : T (T Q) →

T (T ∗ Q) is an isomorphism in all points v ∈ T Q. In particular, F L is a local
diffeomorphism.
 2 
3. det ∂v∂j ∂v
L
k ̸= 0 in local bundle coordinates.

As a consequence, for a regular L the fibre derivative

F L : (T Q, ωL ) → (T ∗ Q, ω)
is a symplectomorphism, i.e. F L respects the symplectic structure.
Proposition 1.49. Moreover, if F L is a diffeomorphism the Hamiltonian systems
(T Q, ωL , HL ) and (T ∗ Q, ω, H) are equivalent, where H := HL ◦ (F L)−1 . Note, that in
local bundle charts H has locally the familiar form
∂L(v)
H(q, p) = v k pk − L(q, v) , where pk = .
∂q k
The last identity can be solved for v: v = v(p), so that H(q, p) = v k (p)pk − L(q, v(p)).
F L can be called Legendre transformation in this situation. Classically the name
is reserved for the transformation which takes L into H:
L(q, v) 7→ H(q, p) = vp − L(q, v) .
Observe, that the Legendre transformation is not a mere coordinate transformation
since L and H live on different spaces.
It is possible to describe the Hamiltonian systems which arise in this way:
Proposition 1.50. Let H ∈ E(T ∗ Q, R) such that the fibre derivative F H : T ∗ Q → T Q
is a diffeomorphism. Then L := (λ(XH ) − H) ◦ (F H)−1 : T Q → R is a Lagrangian
such that F H = (F L)−1 .

Summary:
This chapter introduces the concept of a Hamiltonian system as the mathematical
model of a conservative system of Classical Mechanics. In order to formulate the
general manifold case the basic notations for manifolds have been recalled. Moreover,
several examples of Hamiltonian systems are presented, and it is shown in which way
a Lagrangian system generates a corresponding Hamiltonian system.
For the program of Geometric Quantization the notion of a symplectic manifold
with its Hamiltonian vector fields and its Poisson bracket is crucial. In particular, one
needs the Lie algebra homomorphism
E(M ) → V(M ) , F 7→ −XF ,
which satisfies
[XF , XG ] = −X{F,G} .
 33

2 Ansatz Prequantization

We begin this chapter with some comments about quantization in general and proceed
by presenting canonical quantization in some details in the first section.
Quantization can be viewed to be nothing more than a large set of methods, princi-
ples and procedures to ”construct” quantum systems by using classical systems. Com-
mon feature: Little rigor, great freedom. Main objective: Arrive at a useful quantum
system.
In particular: Quantization is not physics. There are no physical ideas or principles,
which support quantization on a rigorous level. The process

classical system 7→ quantum system

is speculation, even if it generates remarkable examples. Only the reverse process

quantum system 7→ classical system

by taking classical or semi-classical limits can be justified by physical considerations.

Nevertheless, many important quantum systems have been obtained by quantization.
Let us concentrate on quantum mechanics. The main quantum mechanical systems
have been obtained by ”canonical quantization” of models of Classical Mechanics, in
particular, of Hamiltonian systems.

2.1 Canonical Quantization

To quantize a classical system which is given by a Hamiltonian system (X, ω, H) re-

quires according to Dirac to fix a collection o ⊂ E(M, R) of observables, and find a
complex Hilbert space H together with an R-linear map (the Quantization Map)

q : o → ”End”(H)7

such that the following so called Dirac Conditions are satisfied:

(D1) q(1) = λidH ,

(D2) [q(F ), q(G)] = cq({F, G}) , for all F, G ∈ o.

where λ ̸= 0 ̸= c are suitable constants.

In addition, all q(F ) should be self-adjoint (possibly unbounded) linear operators
on H.
7
”End” means, that for F ∈ o, q(F ) is an operator q(F ) : D → H on a dense subspace D ⊂ H. See
Chapter F in the Appendix for operators in Hilbert spaces.
34 2. Ansatz Prequantization

The fact that q should be R-linear says that o can be assumed to be a real vector
space. So we require o to be a real vector subspace of E(M, R).
Similarly, the first condition (D1) supposes that 1 ∈ o. And the second condition
(D2) means that o ⊂ E(M ) can be assumed to be a real Lie subalgebra of the Poisson
algebra E(M, R). Furthermore, the Hamiltonian H should be in the Lie algebra o.
The constant λ is taken as 1 in most cases. However, 2π or similar can be found
in the first papers on geometric quantization. With this choice several formulas in
1
geometry become simpler insofar that they need not the factor 2π . But in these lecture
notes, from now on, λ = 1.
The constant c is mostly ℏi or −i or similar, depending on the conventions. From the
mathematical point of view the value of c is irrelevant except that the self-adjointness
of the operators should not be overlooked: The requirement of self-adjointness implies
that c has to be purely imaginary (c.f. Proposition 2.4 below). In our approaches later
on in these lecture notes, where the assignment F 7→ XF is used in a crucial manner
and certain conventions concerning connections and their curvatures on line bundles,
1
the constant will be c = 2πi = − 2πi .
An important additional property of q is that all q(F ) , F ∈ o, can be recovered
from a common dense domain D ⊂ H. This means, that for the domains of definition
D(q(F )) ⊂ H of the operators q(F ) the condition
\
D ⊂ {D(q(F )) | F ∈ o}

is satisfied. By this condition, it is possible to form the addition q(F ) + q(G) for
F, G ∈ o and rq(F ) for r ∈ R.
Moreover, the image q(F )(D) should be contained in D so that the composition
q(F ) ◦ q(G) can be formed. Only in this way it is possible to define the commutator

[q(F )|D , q(G)|D ] ∈ End D

and it makes sense that it coincides with

cq({F, G})|D .

This is the meaning of ”End”(H).

In many contexts a quantization satisfying the above axioms (D1),(D2) is called
a canonical quantization. But there is no prescription on how to obtain the Hilbert
space H or the map q. Also additional axioms are common, e.g. requirements of the
quantization for special observables like q j or pk for a simple phase space T ∗ U , U ⊂ Rn .
In particular, in several cases irreducibility is required. However, such requirements
can be in conflict with the purpose to formulate quantization in an invariant way.
Geometric quantization is a canonical quantization in the sense above and gives
a well-defined procedure how to find H and q in many cases. The first step in this
2.2 Ansatz: Prequantization 35

procedure is Prequantization and the purpose of this section is to motivate this

concept. In Chapter 7 we pursue this issue further.
Discussion: Essentially different (even controversial) usage of the term ”canonical”:

1. in physics: a special choice, in a standard or common way,

2. in mathematics: a natural way, not dependent on any choice, functorial.

We now come to the subject announced in the title of this section.

2.2 Ansatz: Prequantization

First of all, we have to complexify the whole machinery replacing every R-vector space
W , which occurred so far by the complexification W C i.e. W C := W ⊗R C ∼ = W ⊕ iW .
∼
In particular, the space of observables is now E(M, C) = E(M ) ⊗R C. More examples
are Ta M ⊗C instead of Ta M , As (M )⊗C (∼ = {η : E(M, C)s → E(M, C) | η s-multilinear
over E(M, C) and alternating}) instead of As (M ), g ⊗ C instead of g, etc.
Afterwards, we omit C and in the following Ta M , As (M ), E(M ), g, . . . etc. shall denote
the complexified versions.
Let (M, ω) be a symplectic manifold. We have a natural representation Φ of the
Poisson algebra (E(M ), {, }) by forming the Hamiltonian vector field of a function
F ∈ E(M ) (cf. Proposition 1.31):

Φ : E(M ) → V(M ) (∼
= Der(E(M )) ⊂ End(E(M ))

F 7→ Φ(F ) := −XF .
Recall that Φ is linear (meaning now C-linear) and that for F, G ∈ E(M )

Φ ({F, G}) = −X{F,G} = [XF , XG ] = [Φ(F ), Φ(G)] , (17)

i.e. the respective Lie brackets are respected.

Let us now come to the ”ansatz”:
Attempt 1: As a first try to find an operator for a given F ∈ E(M ), let us consider

q̃(F ) := −cXF = cLXF : E(M ) → E(M ) ,

although E(M ) is not a Hilbert space. In the ansatz we weaken the conditions insofar
as we try to obtain a construction for q with H replaced by the complex vector space
E(M ).

Proposition 2.1. q̃(F ) : E(M ) → E(M ) is C-linear and satisfies

[q̃(F ), q̃(G)] = cq̃({F, G}), F, G ∈ E(M )

36 2. Ansatz Prequantization

Proof.
[q̃(F ), q̃(G)] = (−c)2 [XF , XG ] = c2 −X{F,G} = cq̃({F, G})

Evaluation of attempt 1: (D2) is satisfied (at least for E(M )) and with o = E(M ).
But for 1 ∈ E(M ), we have q̃(1) = 0, hence the first Dirac condition (D1) is not
satisfied.
Attempt 2:
In order to satisfy (D1), we can replace q̃ by q̂ with q̂(F ) = F + q̃(F ) : g 7→ F g −cLXF g.
Evaluation: Now, the first axiom (D1) is satisfied q̂(1) = idE(M ) , but the second (D2)
is not. In the case M = T ∗ R for example, with the coordinate functions q, p : M ∼ =R
∂ ∂
we have {q, p} = 1:Xq = ∂p and Xp = ∂q , and therefore: ω(Xq , Xp ) = 1. Consequently,
∂ ∂
q̃({q, p}) = 1. But q̃(q) = c ∂p + q , q̃(p) = −c ∂q + p with

[q̃(q), q̃(p)] ̸= 1.

We make a further adaption and arrive at

Attempt 3:
q(F ) := F − cLXF + α(XF ), F ∈ E(M ),
with a suitable 1-form α ∈ A1 (M ).

Proposition 2.2. q : E(M ) → E(M ) is C-linear and satisfies the first axiom (D1) of
Dirac quantization (disregarding the fact that we do not have the Hilbert space yet).
Moreover, it fulfills the second axiom (D2) for o = E(M ) if and only if dα = ω.

Proof. Evidently, q is C-linear, and q(1) = 1 since X1 = 0. So, (D1) is fulfilled.

We now check (D2). For F ∈ E(M ) define µ(F ) := F + α(XF ) ∈ E(M ) to be
the multiplication operator µ(F ) : E(M ) → E(M ) , H →
7 µ(F )H. Then q(F ) =
q̃(F ) + µ(F ).

[q(F ), q(G)] = [q̃(F ) + µ(F ), q̃(G) + µ(G)]

= [q̃(F ), q̃(G)] + [q̃(F ), µ(G)] + [µ(F ), q̃(G)] + [µ(F ), µ(G)] .

Hence,
[q(F ), q(G)] = cq̃({F, G}) + [q̃(F ), µ(G)] + [µ(F ), q̃(G)]
according to first attempt (c.f. Proposition 2.1) and because of [µ(F ), µ(G)] = 0.
For every H ∈ E(M )

[q̃(F ), µ(G)] (H) = −cLXF (µ(G)H) + cµ(G)LXF H = −cLXF (µ(G))H

2.2 Ansatz: Prequantization 37

and we obtain
[q̃(F ), µ(G)] = −cLXF µ(G) = c{F, µ(G)}.
In the same way
[µ(F ), q̃(G)] = cLXG µ(F ) = c{µ(F ), G}.

Altogether,

[q(F ), q(G)] = cq̃({F, G}) + c{F, µ(G)} + c{µ(F ), G}

= c (q̃({F, G}) + 2{F, G} + ({F, α(XG )} + {α(XF ), G}))


= c q̃({F, G}) + {F, G} + α(X{F,G} ) +


+ c {F, G} + {F, α(XG )} + {α(XF ), G} − α(X{F,G} )

and we obtain the following: Condition (D2), i.e.

[q(F ), q(G)] = cq({F, G}),

holds if and only if the term in the second bracket vanishes, i.e. iff

{F, G} = {α(XG ), F } − {α(XF ), G} + α(X{F,G} ). (18)

Now, from the formula for dα (cf. 10), we know

dα(XF , XG ) = LXF α(XG ) − LXG α(XF ) − α([XF , XG ])

= {α(XG ), F } − {α(XF ), G} + α(X{F,G} )

and we can finally deduce:

(18) holds ⇐⇒ dα(XF , XG ) = {F, G} = ω(XF , XG ) ∀F, G ∈ E(M )

and
dα(XF , XG ) = ω(XF , XG ) ∀F, G ∈ E(M ) ⇐⇒ dα = ω,
since locally the Hamiltonian vector fields generate the E(U )-module of vector fields.

Because of this result, let us assume for the moment, that ω has a potential α, i.e.
dα = ω. To proceed further, we observe that the 2n- form

ωn = ω ∧ ω ∧ · · · ∧ ω

is a volume form. Let H := L2 (M, ω n ) be the complex Hilbert space, which we obtain
by completing the space of square integrable functions on M , i.e. the space:
Z
D := {ϕ ∈ E(M, C) | |ϕ|2 ω n < ∞}.
M
38 2. Ansatz Prequantization

qR
with respect to the norm ∥ϕ∥ = M
|ϕ|2 dω n . H is a complex Hilbert space with the
inner product defined by Z
⟨ϕ, ψ⟩ := ϕψω n
M
for ϕ, ψ ∈ D and for general ϕ, ψ ∈ H by continuation.
Then the q(F ) , F ∈ E(M ), which are defined on H (or on suitable subspaces)
satisfy both the first and second Dirac condition. But there are generic defects of this
approach:

1. There are important cases without symplectic potential. For example for RN ×S2
(”spin”) or S2 ×S2 (hydrogen atom). Note, that for compact symplectic manifolds
(M, ω) there never exists a potential α ∈ A1 (M ), i.e. dα = ω. To manage this
problem, one has to generalize the ansatz, by replacing E(M ) by sections of a
complex line bundle over M , which will be explained in the next chapter.
2. Even with a symplectic potential α for ω the definition of q depends on the choice
of the potential in an essential manner. To remedy this one can try to make use
of the way two potentials α, α′ differ: since d(α − α′ ) = 0 there exists locally
functions g such that α′ = α + dg. With these local ”gauges” one can build a
quantization map which is no longer dependent on the choice of the potentials.
But this quantization is no longer defined on proper functions ϕ ∈ E(M ), it is
defined on generalized functions which can be given as families of local functions
(ϕi )i∈I , ϕi ∈ E(Ui ), for an open cover (Ui )i∈I of M transforming suitably. This
concept of generalized functions is explained in a reasonable manner by the con-
cept of sections in a complex line bundle where the derivatives are now replaced
by connections on this complex line bundle, see the next chapters.
3. The Hilbert space of wavw functions H is generated by functions in 2n vari-
ables. This should be reduced to n variables. This can be done by introducing
polarizations. We will discuss this in a later chapter.

Apart from these issues, one has to find the appropriate Hilbert space H on the
basis of these sections and then one has to check for which observables F the q(F ) are
self-adjoint, among other considerations.
We conclude this Chapter by restricting to the special case of a simple phase space
M = T ∗ Rn to figure out what properties follow when we use the prequantum oper-
ator
q(F ) := −cLXF + F + α(XF ) ,
with dα = ω as a first step in the program of geometric quantization..
Proposition 2.3. Let H := L2 (M, ω n ) be the Hilbert space introduced above in case of
M = T ∗ Rn ∼
= R2n . For F ∈ E(R2n ) define D := {ϕ ∈ H ∩ E(R2n , C) | LXF ϕ ∈ H}.
Then LXF : D → H is skew symmetric if (and only if ) F is real-valued i.e. F ∈
E(M, R).
2.2 Ansatz: Prequantization 39

Proof. We use the fact that XF is divergence free: In standard coordinates y j =

q j , y n+j = pj for j = 1, 2, . . . , n of R2n let XF = Y k ∂y∂ k . (Summing over k = 1, . . . , 2n.)
Then
X ∂Y k X ∂ ∂F ∂ ∂F
divXF = k
= − = 0,
∂y ∂q ∂pm ∂pj ∂q m
j

now summing over j, m = 1, . . . , n and using the expression (9) for XF . For ϕ, ψ ∈ D:
Z
⟨ϕ, LXF ψ⟩ = ϕ̄XF (ψ) ω n
R 2n
XZ ∂ψ
= ϕ̄Y k k ω n
k R2n ∂y
Z
X ∂
ϕ̄Y k ψ ω n


=− k
k R2n ∂y

by partial integration. Hence,

!
XZ ∂ ϕ̄
Z X ∂Y k
⟨ϕ, LXF ψ⟩ = − Y k k ψ ωn − ϕ̄ ψ ωn
R2n ∂y R2n ∂y k
Zk k

=− LXF ϕψ ω n = −⟨LXF ϕ, ψ⟩,

R2n
P ∂Y k
where ∂y k
= 0 and Y k = Yk where used.

Note, that integration with respect to ω n is the same as Lebesgue integration over
R2n – up to a constant –, since ω n is a constant multiple of the volume form induced
by Lebesgue integration.
Proposition 2.4. Let α ∈ A(R2n ) be a real potential, i.e. dα = ω and all coefficients
αk of α real-valued., and let F ∈ E(M ) be real-valued. Defined D := {ϕ ∈ H | q(F )ϕ ∈
H, LXF ϕ ∈ H}. Then the prequantum operator

q(F ) = −cLXF + F + α(XF )

is symmetric on D if (and only if ) c is purely imaginary.

Proof. We use ⟨ϕ, cLXF ψ⟩ = c⟨ϕ, LXF ψ⟩ = −c⟨LXF ϕ, ψ⟩ by Proposition 2.3, hence, for
purely imaginary c, ⟨ϕ, cLXF ψ⟩ = ⟨cLXF ϕ, ψ⟩, i.q. −cLXF is symmetric. Now, q(F )
is symmetric, since T (F )ϕ := (F + XF )ϕ can be proven to be symmetric: We have
T (F ) = T (F ), since F and α are real. Therefore, for ϕ, ψ ∈ D:
Z Z
n
⟨ϕ, T (F )ψ⟩ = ϕT (F )ψ ω = T (F )ϕψ ω n = ⟨T (F )ϕ, ψ⟩ .
R2n R2n
40 2. Ansatz Prequantization

Corollary 2.5. If the vector field XF is complete, q(F ) will be self-adjoint.

This will be proven in a slightly more general situation in Chapter 7.

Examples 2.6. We determine some q(F ) in the simple phase space M = T ∗ Q , Q ⊂ Rn

open, with the symplectic form ω = dq j ∧ dpj . The canonical coordinates are q j , pj .
1. We choose the negative of the Liouville potential α = −pj dq j = −λ and set
i
c=− ,
2π
so that
i
q(F )ϕ = F ϕ − LX − pj dq j (XF ) .
2π F
We know
∂ ∂
Xq j = − , Xpk = k
∂pj ∂q
α(Xqk ) = 0 , α(Xpk ) = −pk .
So, we have:
i i ∂
q(q j ) = q j − LXqj + 0 = q j + =: Qj
2π 2π ∂pj
i i ∂
q(pj ) = pj − LXpj − pj = − =: Pj , and see
2π 2π ∂q j
i
Q , Pk = cq {q j , pk } = + δkj .
 j 

2π

Note, that with the symplectic potential α = q j dpj the resulting operators Qj , Pj
are quite different, see below.
2. By replacing the Hilbert space H = L2 (T ∗ Q, ω n ) introduced above with its
subspace HP of all square integrable functions ϕ of the form ϕ = g ◦ τ, g : Q → C,
for suitable g we arrive at the function space with the correct dependencies, namely
HP = L2 (Q). From 1. we conclude:

i ∂
Qj = q j , Pj = − .
2π ∂q j
These operators turn out to be self-adjoint, which says that for the simple phase space
and the algebra o of observables generated by 1, pj and q j we have recovered the
canonical commutation relations, CCR.
Moreover, if we include the observable H = 12 p2j into the discussion, we see that
P

∂
XH = −pk
∂q k
2.2 Ansatz: Prequantization 41

and
i ∂
pk k − H.
q(H) =
2π ∂q
This is not the result we expect. The expected quantum operator
P is a multiple of the
1
Laplacian ∆. It can be obtained by simply defining it as /2 Pk Pk . But this adhoc
definition is not in the spirit of geometric quantization.
3. By replacing the potential −λ of ω in 1. by the potential α = q j dpj (note, that
α − λ = d(q j pj )) we obtain

α(Xqk ) = −q k , α(Xpk ) = 0 .

And
i i ∂
q(q j ) = q j − LXqj − qj = =: Qj
2π 2π ∂pj
i i ∂
q(pj ) = pj − LXpj + 0 = pj − =: Pj .
2π 2π ∂q j

Restricting the operations to the Hilbert space HQ of functions ϕ = ϕ(p) depending

only on the verical p ∈ T ∗ Rn this approach leads to the self-adjoint operators
i ∂
Qj = , P j = pj .
2π ∂pj
with the same commutation relations.
4. The two quantizations in 2. and 3. are closely related from a the viewpoint of
representation theory. Both yield representations of the Lie algebra generated by the
pj , q k , the so-called Canonical Commutation Relations (CCR) , and these representa-
tions are equivalent. The equivalence is given by the Fourier transform F : HQ → HP .
In fact, let Z
F(ϕ)(q) := ϕ(p) exp(2πipq)dp 8 ,

for differentiable functions ϕ ∈ E(Rn ) with compact support. F can be extended

uniquely to a unitary map HQ → HP which will be denoted again by F.
Let us denote Q, P the quantizations of q, p in 2. and Q′ , P ′ the quantizations of
q, p in 3. Then for ϕ = ϕ(p) in 3. with
i ∂
P ◦ F(ϕ) = − F(ϕ)
2π ∂q
leads to
Z Z
i ∂
P ◦ F(ϕ) = − ϕ(p) e2πipq dp = pϕ(p)e2πipq dp = F(pϕ) = F ◦ P ′ (ϕ) ,
2π ∂q
8
written for n = 1 or in a schematic notation q = (q 1 , . . . , q n ) , dq = dq 1 ∧ . . . ∧ dq n , etc.
42 2. Ansatz Prequantization

and
Z   Z
′ i ∂ 2πipq
F ◦ Q (ϕ) = ϕ(p) e dp = qϕ(p)e2πipq dp = qF(ϕ) = Q ◦ F(ϕ) ,
2π ∂p

where we have used partial integration

Z Z Z
∂ 2πipq ∂ϕ(p) 2πipq
dp + ϕ(p)2πiqe2πipq dp .


0= ϕ(p)e dp = e
∂p ∂p

We illustrate these relations in a commutative diagram:

HQ
P / HQ
Q
F F
 
HP
P / HP
Q

As a result, the two representations are the same up to unitary equivalence. This
result is a special case of the pairing described in Chapter 14.5, in particular in Propo-
sition 14.20.

Summary: In this Chapter attempts are made for obtaining a canonical quanti-
zation by using the operation F 7→ −XF for observables F on a symplectic manifold
(M, ω) and the fundamental fact that it respects the Poisson bracket. As a result, to
pursue these attempts in full generality one has to use complex line bundles on M
which will be studied in the next chapter. Insofar, Chapter 2 serves as a motivation to
study the geometry of complex line bundles.
Later we will see that starting with a symplectic manifold (M, ω) and a suitable
complex line bundle L the ansatz developed in this chapter (attempt 3) leads to a
quantum model. This process is called prequantization, see Chapter 7. However, the
manifold M has to satisfy an integrality condition (discussed in Chapter 8), which
sounds familiar regarding the principles of quantum theory.
 43

3 Line Bundles

A line bundle over a manifold is a complex vector bundle of rank 1, i.e. with typical
fibre isomorphic to C. This chapter provides an elementary and detailed exposition of
the fundamental properties of line bundles including the description of line bundles by
cocycles. Moreover, the concrete examples of tautological line bundles on the complex
projective spaces are studied.

3.1 Basic Definitions

Definition 3.1. A Line bundle (”Geradenbündel”) over a given manifold M is a

manifold L (the Total space, ”Totalraum”) together with a map9

π:L→M

with the following properties:

1. The fibres of π are Lines: Every fibre La := π −1 (a) , a ∈ M , has the structure
of a one dimensional vector space over C.

2. π is Locally trivial (”lokaltrivial”), i.e. the total space locally looks like a
product U × C up to isomorphisms: To each point a ∈ M there corresponds an
open neighbourhood U ⊂ M of a and a diffeomorphism

ψ : LU := L|π−1 (U ) → U × C

such that

(a) the diagram

ψ|LU
LU / U ×C
π|π−1 (U )
pr1
 {
U
is commutative: pr1 ◦ ψ|LU = π|π−1 (U ) 10 ,
(b) for all b ∈ U , the following induced map
ψ|L pr2
ψb : Lb −→b {b} × C −→ C , ψb := pr2 ◦ ψ|Lb ,

is a homomorphism (in fact an isomorphism) of vector spaces over C.

9
assumed to be smooth, as usual
10
pr1 , pr2 denote the natural projections pr1 : W × V → W , (x, y) 7→ x resp. pr2 : W × V →
V , (x, y) 7→ y for a product W × V .
44 3. Line Bundles

A line bundle is Trivial (”trivial”) if L = M × C with π = pr1 , and Lb = {b} × C

obtains its vector space structure through the bijection pr2 : Lb → C.
However, by abuse of language, the line bundles which are isomorphic to the trivial
line bundle are also called trivial, although they should better be called Trivializ-
able. To understand ”isomorphic” we have to introduce the notion of a homomorphism
of line bundles.
Definition 3.2. A Line Bundle Homomorphism (homomorphism of line bundles)
from L → M to L′ → M is a map11 ψ : L → L′ such π = π ′ ◦ ψ and such that for
each a ∈ M the restriction ψa := ψ|La : La → L′a is a (vector space) homomorphism.
In particular, the diagram
ψ
L / L′

π
~ π′
M
is commutative.

An isomorphism of line bundles is a homomorphism ψ of line bundles which is

bijective such that ψ −1 is again a homomorphism of line bundles. In particular, an
isomorphism is a homeomorphism.
Definition 3.3. A Section of a line bundle π : L → M over an open subset U ⊂ M
is a map12
s:U →L
such that π ◦ s = id|U .

The set of sections over U is denoted by Γ(U, L). By pointwise addition and multi-
plication Γ(U, L) becomes a vector space over C and an E(U )-module: For s, t ∈ Γ(U, L)
and f ∈ E(U ) we set

(f s + t) (a) := f (a)s(a) + t(a) , a ∈ U,

and we see f s + t ∈ Γ(U, L).

In case of the trivial bundle L = M × C the space of sections Γ(U, L) over an open
U ⊂ M is naturally isomorphic to E(U ): Let s1 (a) := (a, 1) , a ∈ U , be the 1-section,
s1 ∈ Γ(U, L). For every f ∈ E(U ) one has

f s1 (a) = f (a)s1 (a) = f (a)(a, 1) = (a, f (a)) , a ∈ U.

Since every section s ∈ Γ(U, L) is of the form s(a) = (a, f (a)) , a ∈ U, for some
f ∈ E(U ), the map
E(U ) → Γ(U, L) , f 7→ f s1 ,
11
which is again smooth
12
which is again smooth
3.2 Cocycles Generating Line Bundles 45

is an E(U )-module isomorphism.

As a consequence, in a general line bundle π : L → M with a local trivialisation
ψ : π −1 (U ) → U × C for every open subset W ⊂ U the space of sections Γ(W, L|W ) is
isomorphic to E(W ) as an E(W )-module.
Proposition 3.4. A line bundle π : L → M is trivial (-izable) if and only if there
exists a global nowhere vanishing section of L, i.e. a section s ∈ Γ(M, L) with s(a) ̸= 0
for all a ∈ M .

Proof. Let s be such a section. It is enough to show that

ψ : M × C → L , (a, λ) 7→ λs(a) ,

for (a, λ) ∈ M × C is a diffeomorphism and linear in the fibres. Of course, ψ is smooth

and satisfies π ◦ ψ = pr1 : π ◦ ψ(a, λ) = π(λs(a)) = a = pr1 (a, λ). And for each a ∈ M

ψa = ψ|{a}×C : {a} × C → La , (a, λ) 7→ λs(a) ,

is an isomorphism of vector spaces, since s(a) ̸= 0.

The fact that for trivial line bundles L over M the sections of L are essentially the
same as functions, Γ(M, L) ∼ = E(M ), strengthens the viewpoint that for general line
bundles L the sections Γ(M, L) of L are generalized functions in M . Another such
interpretation will be given after the next step in describing line bundles using cocycles
(cf. Observation 3.8).

3.2 Cocycles Generating Line Bundles

The second condition in our Definition 3.1 yields an open cover (Uj )j∈I of M with local
trivializations
ψj : L|Uj → Uj × C.
In particular, ψj is a diffeomorphism with π = pr1 ◦ ψj , and ψa : La → {a} × C
is an isomorphism. In addition, for each j ∈ I one obtains a distinguished section
sj ∈ Γ(Uj , L) by
sj (a) := ψj−1 (a, 1) , a ∈ Uj ,
with the property

ψj (zsj (a)) = (a, z) , or ψj−1 (a, z) = zsj (a),

when (a, z) ∈ Uj × C.
The sj und sk satisfy the identity:

sj = gkj sk ,
46 3. Line Bundles

on the intersection Uj ∩Uk 13 where the ”transition functions” (”Übergangsfunktionen”)

gkj : Uj ∩ Uk → C are defined as
sj
gkj := , j, k ∈ I :
sk

For each a ∈ M and j, k ∈ I one has sj (a) ̸= 0 ̸= sk (a). Hence, gkj (a) ∈ C is
well-defined with sj (a) = gkj (a)sk (a).
Because if the importance of this simple identity, we give another description of
gkj : Ujk := Uj ∩ Uk → C: The composition

ψk ◦ ψj−1 : Ujk × C → Ujk × C

where ψj−1 (a, z) = zsj (a) and ψk (wsk (a)) = (a, w) with wsk (a) = zsj (a), acts as
 
sj (a)
(a, z) 7→ a, z = (a, gkj (a).z) . (19)
sk (a)

Proposition 3.5. The transition functions (gjk )j,k∈I , gjk ∈ E(Ujk ), satisfy the follow-
ing Cocycle condition (”Kozyklus-Bedingung”):

gjj = 1

(C) gjk gkj = 1

gij gjk gki = 1

Proof. Nearly trivial: gjj comes by definition from ψj ◦ ψj−1 = id|{a}×C . when a ∈ Uj ,
and the identity id : C → C is given by the multiplication with 1 = gjj : z 7→ 1.z. In the
same way gjk gkj comes from ψj ◦ψk−1 ◦ψk ◦ψj−1 = id|{a}×C , a ∈ Ujk , inducing gjk gkj = 1.
And again, because of ψi ◦ ψj−1 ◦ ψj ◦ ψk−1 ◦ ψk ◦ ψk−1 = id|{a}×C , a ∈ Uijk := Ui ∩ Uj ∩ Uk ,
one obtains gij gjk gki = 1.

The transition functions describe how the various products Uj × C glue together to
form the total space14 .
The transition functions (gij ) of a line bundle L describe the sections of that bundle,
as explained in the following. Each section s ∈ Γ(M, L) yields the local function

fj := pr2 ◦ ψj ◦ s|Uj ,
13
Here and in the following the case of an empty intersection Uj ∩ Uk = ∅ contains no information,
so it is, in general, correct.
14
This feature holds for the more general cases of vector bundles of rank k > 1 and principal bundles
as well
3.2 Cocycles Generating Line Bundles 47

i.e. the diagram

ψj
L|Uj / Uj ×C
O
s|Uj pr2

Uj /C
fj

is commutative.

Lemma 3.6. For all j, k ∈ I the following equations hold:

s|Uj = fj sj , on Uj ,

fk = gkj fj , on Ujk .

Proof. For a ∈ Uj :

s(a) = ψj−1 ◦ ψj (s(a)) = ψj−1 (a, pr2 ◦ ψj ◦ s(a))

= ψj−1 (a, fj (a)) = fj (a)ψ −1 (a, 1) = fj (a)sj (a)

which shows the first identity. On Ujk ̸= ∅ we obtain

s|Ujk = fk sk |Ujk = fj sj |Ujk = fj gkj sk |Ujk ,

hence fk = fj gkj .

The second condition is called Section Condition:

(S) fk = gkj fj

Proposition 3.7. Let π : L → M be a line bundle over M with local trivializations

ψj : L|Uj → Uj × C, j ∈ I, such that (Uj )j∈I is an open cover of M .

1. Then every global section s ∈ Γ(M, L) defines a collection fj ∈ E(Uj ) of local

functions with (S).

2. Conversely, every collection (fj )j∈I , fj ∈ E(Uj ), of local functions satisfying (S)
yields a global section s ∈ Γ(M, L) with s|Uj = fj sj .

Proof. The first statement has just been shown. The data (fj ) in the second statement
have the property fj sj |Ujk = fk sk |Ujk by (S) and thus define a global section s through

s(a) := fj (a)sj (a) , a ∈ Uj .

48 3. Line Bundles

Observation 3.8. Note, that the result of Proposition 3.7 provides another interpre-
tation of sections s ∈ Γ(M, L) as generalized functions. A generalized function under
this viewpoint is a collection of local functions (fj ) which satisfy fk = gkj fj on Ujk for
all j, k ∈ I, i.e. it is a section. This generalization is adapted to our problem of not
having a global potential for a given symplectic form, in general.

The transition functions of a line bundle determine the line bundle completely up
to isomorphism: Similar to the reconstruction of a section from local functions (fj )
with condition (S) any collection of (gjk ) with condition (C) allows to construct a line
bundle L with the (gjk ) as its transition functions:
Proposition 3.9. Let (Uj )j∈I be an open cover of the manifold M , and let gkj ∈
E(Ujk , C× ) , j, k ∈ I , be a collection of functions forming a cocycle, i.e. such that (C)
is satisfied. Then the data (M, (Uj ), (gjk )) induce a complex line bundle π : L → M
over M with local trivializations

ψ j : L Uj → U j × C

such that for (a, z) ∈ Ujk × C:

ψk ◦ ψj−1 (a, z) = (a, gkj (a).z).

Proof. On the disjoint union R := ˙ j∈I Uj × C, which is an (n + 2)-dimensional real

S
manifold, we consider the equivalence relation

(aj , zj ) ∼ (ak , zk ) : ⇐⇒ aj = ak and zj = gkj (a).zk , ,

where (aj , zj ) ∈ Uj × C, (ak , zk ) ∈ Uk × C.

The quotient manifold L = R/ ∼ exists15 , see Definition A.10 ff. for the concept of
a differentiable quotient manifold. The quotient map π : L → M is smooth, and the
maps
ψj : LUj → Uj × C , [(a, z)] 7→ (a, z) ,
for (a, z) ∈ Uj × C turn out to be local trivializations generating fibrewise the vector
space structure on La through

pr2 ◦ ψj |La : [(a, z)] 7→ z.

The transition functions in this construction are the (gkj ). In fact, for a = aj = ak ∈
Ujk :

ψk ◦ ψj−1 (a, zj ) = ψk ([(aj , zj )]) = ψk ([(ak , gkj (a).zk )]) = (a, gkj (a).zk ),

i.e. the (gkj ) are the transition functions of the constructed line bundle.

15
One has to check that M is a Hausdorff space and that L is well-defined with the stated properties.
3.2 Cocycles Generating Line Bundles 49

Thus, we see that a line bundle can essentially be recaptured by its transition
functions.
With respect to this description of line bundles by cocycles, the trivial bundle is
given by U1 = M, (Uj )j∈{1} and g11 (a) = 1. But the data: (Uj )j∈I an open cover,
gjk : Ujk → C , gjk (a) = 1, yield also a trivial bundle L, more precisely an isomorphism
Θ : L → M × C of line bundles.
In general, a homomorphisms Θ of line bundles has a description using the transition
functions (i.e. the cocycles) of the two line bundles which are involved which is similar
to the descriptions of sections:

Proposition 3.10. For a homomorphism of line bundles Θ : L → L′ , where L resp.

L′ have local trivializations ψj resp. ψj′ with respect to an open cover (Uj )j∈I 16 the local
mappings hj : Uj → C given by

hj := pr2 ◦ ψj′ ◦ Θ|LUj ◦ sj

ψj′
LUj / L′Uj / Uj × C
O Θ
sj pr2

Uj /C
hj

satisfy the following identity:

′
gkj
hk = hj ,
gkj

j, k ∈ I . Conversely, (hj ) with this identity determines a homomorphism Θ by locally

defining
−1
Θ|Uj (ψj−1 (a, z)) := ψj′ (a, hj (a).z).

Proof. The definition of hj can also be read as follows:

(a, hj (a).z) = ψj′ ◦ Θ ◦ ψj−1 (a, z) = ψ ′ ◦ Θ ◦ zsj (a) , (a, z) ∈ Uj × C.

In the original definition hj := pr2 ◦ ψj′ ◦ Θ|LUj ◦ sj , on Ujk one can replace sj by gkj sk
and ψj′ by gjk
′
ψk′ . These replacements yield
′
hj = pr2 ◦ gjk ψk′ ◦ Θ|LUjk ◦ gkj sk = gjk
′
gkj pr2 ◦ ψk′ ◦ Θ|LUjk ◦ sk = gjk
′
gkj hk

′ ′
which is the required identity (note that gjk is the inverse of gkj ). The converse is
clear.
16
Given two line bundles there always exists such a cover.
50 3. Line Bundles

Corollary 3.11. Under the assumption of the preceding proposition the line bundles
L and L′ are isomorphic, if and only if there are non-vanishing functions hj : Uj → C
with

′ hk
(I) gkj = gkj
hj

on Ujk and for all j, k ∈ I. The isomorphism Θ : L → L′ given by (hj ) is locally defined
as Θ(sj (a)) = hj (a)s′j (a).

Observation 3.12. These results show that the isomorphism classes of line bundles on
′
a manifold are essentially isomorphism classes of cocycles. Here, two cocycles gkj , gkj
are defined to be equivalent if there exists hj ∈ E(Uj ) satisfying (I). This resembles
cohomology. We come back to this fact later.

But are there nontrivial line bundles at all? If not, the introduction of the notion
of a line bundles would give very little sense.
Presumably, the case of the tangent bundle T S2 is known in the form of the ”Hairy
Ball Theorem” (”Satz vom Igel”): There is no non-vanishing (smooth) vector field on
S2 . Therefore, the tangent bundle cannot be trivial. It is easy to see that the tangent
bundle is, in fact, a complex line bundle by observing that S2 has the interpretation of
being the Riemann sphere, the complex projective space P1 (C).
The question of the existence of nontrivial line bundles has to do with the classi-
fication of all line bundles (determining the set of isomorphism classes) which could
be achieved with the help of cocycles. But we want, first of all, to investigate all line
bundles on a rather typical example, the complex projective space Pn (C) of complex
dimension n.

3.3 The Tautological Line Bundle

The aim of this subsection is to present concrete line bundles and analyse their sec-
tions. This will be started for line bundles over the complex projective spaces where
the tautological bundles are introduced and will be continued by determining all line
bundles over the complex projective spaces.
The Case P1 (C)

Example 3.13 (The projective line P1 (C)). Let P1 := P1 (C) be the Riemann sphere
resp. the one dimensional projective space over C: P1 is the space of lines in C2 through
the origin 0 ∈ C2 . The best way to make this precise is to introduce the following
equivalence relation in C2 \ {0}:

z ∼ w :⇐⇒ ∃ λ ∈ C : z = λw.
3.3 The Tautological Line Bundle 51

Then P1 is the quotient C2 \{0}/ ∼ and is endowed with a natural projection

γ : C2 \ {0} → P1 ,
mapping each line ℓ ∈ C2 to its corresponding equivalence class ℓ \ {0} in P1 (see
Definition A.10 ff. for the concept of a differentiable quotient).
The points of P1 are represented by the so called Homogeneous Coordinates
(”Homogene Koordinaten”) induced from C2 \ {0}: For z = (z0 , z1 ) ∈ C2 \ {0} we set
(z0 : z1 ) := γ(z)
(= [z], the equivalence class of z = (z0 , z1 )). We conclude from γ(λz) = λγ(z), for
λ ̸= 0, that the homogeneous coordinates fulfill
(z0 : z1 ) = (λz0 : λz1 ) , for λ ∈ C× := C \ {0}.

P1 obtains its topological and complex manifold resp. differentiable structure as the
quotient C2 \ {0}/ ∼. Hence, U ⊂ P1 is open if and only if γ −1 (U ) is open, and a
map f : U → C is holomorphic (resp. smooth) if and only if f ◦ γ : γ −1 (U ) → C is
holomorphic (resp. smooth).
P1 can be covered by two holomorphic charts ψj : Uj → C, where Uj := {(z0 : z1 ) |
zj ̸= 0}, j ∈ {0, 1} and
ψ0 : U0 → C, (1 : z) 7→ z , z ∈ C,
ψ1 : U1 → C, (w : 1) 7→ w, w ∈ C.
On the intersection U01 = U0 ∩ U1 = {(z0 : z1 ) | z0 ̸= 0 ̸= z1 } we have for w ∈ U01 :
1 1
ψ0 ◦ ψ1−1 (w) = ψ0 (w : 1) = ψ0 (1 :
)= .
w w
×
This is a holomorphic function on C = C \ {0} with holomorphic inverse. Therefore,
the complex structure on P1 (C) (and the differentiable structure as well) is given by
these two charts.17
U0 can be understood as the complex plane C with (1 : 0) as 0, (1 : z) ∼ = z as the
coordinate and U1 adds only the point (0 : 1) = ∞ to U0 ∼ = C, thus obtaining the
2 ∼ ∼
sphere S = U0 ∪ {(0 : 1)} = C ∪ {∞}.
Construction 3.14 (Tautological Bundle). The product P1 × C2 is a 3-dimensional
complex manifold and a 6-dimensional real manifold. It also has the structure of the
trivial complex vector bundle over P1 of rank 2. Define
T := {(a, w) ∈ P1 × C2 | ∃λ ∈ C : w = (λa0 , λa1 ) if a = (a0 : a1 )}
= {(a, w) ∈ P1 × C2 | w = 0 or γ(w) = a}
[ [
{a} × {0} ∪ γ −1 (a) =


= {a} × ℓa
a∈P1 a∈P1
17
One has to check that the quotient structure exists, which can be done, by proving that P1 (C)
with the structure given by the two charts satisfies indeed the universal property of the quotient.
52 3. Line Bundles

and π := pr1 : T → P1 , (a, w) 7→ a. Here, ℓa := {0} ∪ γ −1 (a) is the complex line in C2

represented by a ∈ P1 .
T is a complex submanifold of P1 ×C2 of dimension 2, since it also has the description

T = {((a0 : a1 ), (w0 , w1 )) ∈ P1 × C2 | a0 w1 − a1 w2 = 0 }.

And π = pr1 |T : T → P1 defines a smooth (even holomorphic) projection . For each

a ∈ P1 the fibre Ta = π −1 (a) = {a} × ℓa obtains its vector space structure from the
vector subspace ℓa ⊂ C2 . Hence, the fibre Ta = {a} × ℓa can be viewed to be precisely
the line given by the equivalence class a. This property is the reason why T is called
the Tautological Bundle.
Moreover, to see that T is indeed a line bundle, we consider, for j = 0, 1, the
diffeomorphisms
φj : TUj → Uj × C , (a, (w0 , w1 )) 7→ (a, wj )
with inverses  
a1
φ−1
0 : ((a0 : a1 ), z)) 7→ (a0 : a1 ), (z, z )
a0
 
a0
φ−1
1 : ((a0 : a1 ), z)) 7→ (a0 : a1 ), (z , z)
a1
It is easy to see that the φj are local trivializations which respect the open Uj . The
diagram
TUj φj / Uj × C
π
pr1
 {
Uj

is commutative and the φj are linear in the fibres.

To calculate the transition functions with respect to φ0 , φ1 we determine the action
of φ1 ◦ φ−1
0 : U01 → U01 as
   
−1 a1 a1
φ1 ◦ φ0 ((a0 : a1 ), z) = φ1 (a0 : a1 ), (z, z ) = (a0 : a1 ), (z )
a0 a0

for a = (a0 : a1 ) ∈ U01 and z ∈ C. Hence, the corresponding transition function

g10 : U10 → C× (defined by φ1 ◦ φ−1
0 (a, z) = (a, g10 (a).z), c.f. formula (19)) is simply

a1 a1
g10 (a) = , i.e. g10 (a)(z) = .z.
a0 a0

Analogously,
a0
g01 (a) = .
a1
3.3 The Tautological Line Bundle 53

Note that T is the tangent bundle with respect to the differentiable as well to the
complex structure, i.e. T ∼ = T P1 as holomorphic line bundles, and the differentiable
structure induced by T → P1 (as a complex holomorphic line bundle over P1 just
described) agrees with the differentiable structure of the tangent bundle T S2 . In other
words, the natural identification map F : T → T S2 is a diffeomorphism.

Proposition 3.15. The tautological bundle T over P1 has no holomorphic section other
than the zero section: Γhol (P1 , T ) = {0}.

Proof. Any holomorphic section s : P1 → T is given by holomorphic functions fj :

Uj → C satisfying (S):
z0
f0 (z) = g01 (z)f1 (z) , z = (z0 : z1 ) ∈ U01 , i.e. f0 (z0 : z1 ) = f1 (z0 : z1 )
z1

according to the calculation above. The two functions Fj on γ −1 (Uj ) given by Fj (z) :=
1
f (γ(z)) , z = (z0 , z1 ) ∈ γ −1 (Uj ) are well-defined and they agree on γ −1 (U01 ):
zj j

1 1 z0 1
g0 (z) = f0 (γ(z)) = f1 (γ(z)) = f1 (γ(z)) = g1 (z).
z0 z0 z1 z1
As a consequence, the given holomorphic section s induces a holomorphic function F
on C2 \ {0} (= F1 = F0 on γ −1 (U01 )) with the property

F (λw) = λ−1 F (w) , λ ∈ C× , w ∈ C2 \ {0}.

But such a holomorphic function F : C2 \ {0} → C is the zero function18 and in turn
the section s has to be zero.

Corollary 3.16. T is not isomorphic to the trivial bundle P1 × C as a holomorphic

bundle.

This result is no surprise regarding the Hairy Ball Theorem.

The Case Pn (C)
The whole consideration of the above example can be generalized to the n-
dimensional projective space Pn := Pn (C), the space of complex lines through 0 in
Cn+1 . We define
Pn := Pn (C) := Cn+1 \ {0} / ∼

with respect to the equivalence relation

w ∼ z :⇐⇒ ∃ λ : w = λz,
18
One has to show that F defines a holomorphic function on all of C2 , since there exist no isolated
singularities for holomorphic functions in more than one variable, cf. Appendix, Corollary B.12.
54 3. Line Bundles

and obtain the projection

γ : Cn+1 \ {0} → Pn , γ(z) = [z] = (z0 : z1 : . . . : zn ),

where z = (z0 , z1 , . . . , zn ) ∈ Cn+1 .

Pn obtains its topological, differentiable and complex structure as the quotient
(Cn+1 \ {0})/ ∼. Convenient holomorphic charts are the following: φj : Uj → Cn on
Uj := {(z0 : z1 : . . . : zn ) ∈ Pn | zj ̸= 0} defined by

1
φj (z0 : z1 , : . . . : zn ) := (z0 , z1 , . . . zˆj , . . . , zn ),
zj

j = 0, 1, . . . n, with biholomorphic φj ◦ φ−1 19

k : Ujk → Ujk .

Construction 3.17 (Tautological Bundle). As before, we start with the trivial com-
plex vector bundle Pn × Cn+1 over Pn and define

T := {(z, w) ∈ Pn × Cn+1 | ∃λ ∈ C : w = λ(z0 , . . . , zn ) if z = (z0 : . . . : zn )}

[ [
{z} × {0} ∪ γ −1 (z) =


= {z} × ℓz
z∈Pn z∈P1
n n+1
= {(z, w) ∈ P × C | zi wj − zj wi = 0 for i, j = 0, 1, . . . n}

with projection π = pr1 . T ⊂ Pn × Cn+1 is a complex submanifold of dimension n + 1

and it is the total space of a differentiable and holomorphic line bundle π : T → Pn ,
the tautological bundle over Pn . T is not trivializable as a holomorphic bundle and
also not as a differentiable line bundle.
Local trivializations are given by

ψj : TUj → Uj × C , (z, w) 7→ (z, wj )

with the corresponding transition functions

zj
gjk (z) = , z = (z0 : z1 : . . . : zn ) ∈ Ujk = {z ∈ Pn | zj ̸= 0 ̸= zk }.
zk

In fact,
 
λ zj
ψj ◦ ψk−1 (z, λ) = ψj z, (z0 , . . . , zn ) = (z, λ) ,
zk zk
i.e.
zj
gjk (z) = .
zk
19
Exercise: Check!
3.3 The Tautological Line Bundle 55

Construction 3.18 (Hyperplane Bundle and More). In order to generate further line
bundles on Pn we modify the transition functions to
 m
m zk
gjk (z0 : . . . : zn ) := , z ∈ Ujk ,
zj
m
where m ∈ Z. gjk : Ujk → C is holomorphic, in particular smooth. Furthermore, for
m
each m ∈ Z the collection (gjk | i, j ∈ I) satisfies the cocycle condition (C). Therefore,
m
by Proposition 3.9 the cocycle (gjk ) determines a smooth line bundle, which will be
denoted by H(m). A slight modification of Proposition 3.9 yields, that H(m) is a
holomorphic line bundle.
Note that our tautological bundle is T = H(−1). Obviously, H(0) is the trivial
bundle. H := H(1) is called the Hyperplane Bundle, it is dual to T = H(−1).
H can alternatively defined as the space of all hyperplanes in Cn+1 , the n-dimensional
complex vector subspaces of Cn+1 .

As before, one can show

Proposition 3.19. H(m) is not trivial as a holomorphic line bundle over Pn for m ∈
Z, m ̸= 0.

See below, Corollary 3.23.

In order to determine the sections of each of the line bundles H(m) → Pn over Pn
we use the section condition (S) which has to be satisfied for the local functions which
represent a given section.
We introduce for each m ∈ Z the m-homogeneous functions Em (V ) on a saturated20
open subset V = γ −1 (γ(V )) ⊂ Cn+1 as follows

Em (V ) := {F ∈ E(V, C) | ∀ λ ∈ C× ∀ z ∈ V : F (λz) = λm F (z)}.

Now, let U ⊂ Pn be open and V := γ −1 (U ): Every section s : U → H(m) deter-

mines a function s̃ = Fs ∈ Em (V ) in the following way: With respect to the open cover
(Uj )j=0,...n and to the transition functions
 m
zk
gjk (z) = , z = (z0 : . . . : zn ) , zj ̸= 0 ̸= zk ,
zj
the given section s ∈ Γ(U, H(m)) determines fj ∈ E(U ∩ Ujk ) such that (according to
(S))
fj = gjk fk on Ujk .
Recall s|U ∩Uj = fj sj where sj (a) = ψj−1 (a, 1). We define

Fj (w) := wjm fj (γ(w)) , w ∈ γ −1 (U ∩ Uj ) = V ∩ γ −1 (Uj ) .

20
V is saturated iff λV = V for all complex numbers λ ∈ C.
56 3. Line Bundles

For all w ∈ γ −1 (U ∩ Ujk ) we obtain

 m
wk
Fj (w) = wjm gjk (γ(w))fk (γ(w)) = wjm fk (γ(w)) = Fk (w).
wj

As a consequence, Fj and Fk agree on V ∩ γ −1 (Ujk ) and

Fs (w) := Fj (w) for w ∈ V ∩ γ −1 (Uj )

is a well-defined (smooth) function s̃ = Fs ∈ E(V ). Morover, it is m-homogeneous:

Fs (λw) = (λwj )m fj (γ(w)) = λm wjm fj (γ(w)) = λm gs (w) , (w, λ) ∈ V ∩ γ −1 (Uj ) × C.

Therefore, s̃ = Fs ∈ Em (V ). The induced map

e : Γ(U, H(m)) → Em (V ) , s 7→ s̃ ,

is linear over C (it is even E(U )-linear) and injective. We have shown the first part of
the following result:

Theorem 3.20. For every open U ⊂ Pn and V := γ −1 (U ),

e : Γ(U, H(m)) → Em (V ) ,

is an isomorphism.

Proof. It remains to prove that e is surjective. Let F ∈ Em (V ) und set for z ∈ U ∩ Uj

fj (z) := wj−m F (w) , if γ(w) = z .

In case of w′ = λw, λ ∈ C× , we have

(wj′ )−m g(w′ ) = λ−m wj−m λm g(w) = wj−m g(w) .

Therefore, fj ∈ E(U ∩ Uj ) is well-defined. Moreover (fj ) satisfies (S):

 m  m
−m wk −m wk
fj (z) = wj g(w) = wk g(z) = fk (z) , z ∈ U ∩ Ujk .
wj wj

As a consequence, (fj ) defines a section s ∈ Γ(U, H(m)) such that s̃ = g.

We are also interested in determining the holomorphic sections of the bundles H(m).
In general, a line bundle over a complex manifold M (i.e. a manifold with enough holo-
morphically compatible charts) which has enough local trivializations which are biholo-
morphic (so that the transition functions are holomorphic) is called a Holomorphic
Line Bundle. However, when it is clear from the context, one simply denotes it as a
line bundle, as before. Let us introduce the following notations:
3.3 The Tautological Line Bundle 57

• O(U ) is the C-algebra of holomorphic functions f : U → C on an open subset

U ⊂ M of a complex manifold M .
• Γhol (U, L) is the O(U )-module of holomorphic sections s : U → L of a (holomor-
phic) line bundle L on U .
• Om (V ) is the O(U )-module of the m-homogeneous holomorphic functions on a
saturated V ⊂ Cn+1 .
• Cm [z0 , z1 , . . . , zn ] is the C-vector space of m-homogeneous complex polynomials
in n + 1-variables for m ≥ 0. Cm [z0 , z1 , . . . , zn ] is generated over C by the
monomials zi1 zi2 . . . zim , ij ∈ {0, 1, . . . , n} of degree m.

The arguments for the result of the preceding proposition also yield:
Corollary 3.21. For every open U ⊂ Pn and V := γ −1 (U ),
e : Γhol (U, H(m)) → Om (V ) ,
is an isomorphism.

With this result it is possible to determine all holomorphic sections in H(m) rather
explicitly:
Proposition 3.22.
Γhol (Pn (C), H(m)) ∼
= {0} for m < 0 , m ∈ Z , (20)
Γhol (Pn (C), H(m)) ∼
= Cm [z0 , z1 , . . . , zn ] for m ≥ 0 , m ∈ Z . (21)

Proof. For m ≥ 1 one has to check that

Om (Cn+1 \ {0}) ∼
= Cm [z0 , z1 , . . . , zn ] .
Since each g ∈ O(Cn+1 \ {0}) has a unique extension to all of Cn+1 as a holomorphic
function (for n ≥ 1 there are no isolated singularities for holomorphic functions of
n + 1 variables, see Corollary B.12 in the Appendix) it is enough to observe that m-
homogeneous holomorphic functions g : Cn+1 → C are m-homogeneous polynomials.
For m = 0 the m-homogeneous functions are the constants. The case m < 0 is left
as an exercise.

As immediate consequences we obtain the following results:

Corollary 3.23.

• The C-vector space Γhol (Pn (C), H(m)) of holomorphic sections of the holomorphic
line bundle H(m) is finite dimensional for all m ∈ Z.21
21
In general, the space of holomorphic sections of a holomorphic vector bundle over a compact
complex manifold is finite dimensional.
58 3. Line Bundles

• All H(m), m ̸= 0, are nontrivial as holomorphic line bundles.

• The bundles H(m) and H(k), m ̸= k, are not isomorphic as holomorphic line
bundles.

Proof. The first statement follows from the fact dimC Cm [z0 , z1 , . . . , zn ] < ∞. For
the second, observe that none of the above sections can be non-vanishing and apply
the holomorphic version of Proposition 3.4. For the third: If H(m) and H(k) were
isomorphic then, according to Corollary 3.11, there would exist hj ∈ O(Uj ) such that
gijm = hhkj gijk on Uij i.e.
 m  k  m−k
zj hj zj zj hj
= , or = ,
zi hi zi zi hi

which would imply that H(m−k) is trivial in contradiction to the second statement.

3.4 Classification

We conclude this chapter with a short description of how cohomology is used to classify
all line bundles on a given manifold M .
Observe, that two cocycles (gjk ), (hjk ) with respect to an open cover (Uj )j∈I can be
multiplied to yield another cocycle

fjk := gjk hjk ∈ E(Ujk ) .

In this way, a composition on the set

Picdiff (M )

of isomorphism classes of line bundles on M is defined. It is denoted [Lg ] ∗ [Lh ] when

Lg resp. Lh is the line bundle given by g = (gjk ), resp. h = (hjk ), and []Lg ] is the
isomorphism class of the bundles Lh . The composition is associative and commutative
with the class of the trivial bundle as unit. The inverse of a class [L] ∈ Picdiff (M ),
where g = (gjk ) generates L = Lg is the class generated by the so-called Dual Line
−1
Bundle L∨ associated to the cocycle (gjk ).
In this way Picdiff (M ) is endowed with the structure of an abelian group. It is called
the (differentiable) Picard Group.
This composition can alternatively be described by the tensor product L ⊗ L′ of
line bundles L and L′22 , by

([L], [L′ ]) 7→ [L ⊗ L′ ] = [L] ∗ [L′ ].

22
The tensor product of vector bundles is explained in the Appendix E.
3.4 Classification 59

The Picard group Picdiff (M ) can be identified with the first (sheaf) cohomology
group
Picdiff (M ) ∼
= Ȟ 1 (M, E × )
with respect to the sheaf E × of germs of nowhere vanishing (smooth) functions. (See
Section E.3 for a short introduction to sheaf cohomology.) In fact, the equivalence
relation for transition functions gij , gij′ representing line bundles L, L′ coincides with
the equivalence relation of the Čech cocycles (gij ), (gij′ ) as is explained in Example
E.16!
Furthermore, Ȟ 1 (M, E × ) is isomorphic to Ȟ 2 (M, Z). This can be shown using
the connecting homomorphism δ : Ȟ 1 (M, E × )→Ȟ 2 (M, Z), induced by the short exact
sequence of sheaves
e
0 −→ Z −→ E −→ E × −→ 0
with e(ϕ) := exp(2πiϕ) , ϕ ∈ E.
Note, that c1 (L) = δ[L] is called the Chern class of the line bundle L:
Since it is known that Ȟ 2 (Pn (C), Z) ∼
= Z (c.f. [?]), we conclude

Picdiff (Pn ) ∼
= Z.

In the case of a complex manifold M the ”holomorphic” Picard group Pic(M ) is

introduced analoguously:

Pic(M ) := {[L] | L a holomorphic line bundle} ,

where [L] is the class of holomorphic line bundles which are (holomorphically) isomor-
phic to L. The multiplication in the group is again given by
 ′ 
 ′

[(gjk )] , (gjk ) 7−→ (gjk gjk ) resp.

([L] , [L′ ]) 7−→ [L ⊗ L′ ] .

Similar to the differentiable case, the group Pic(M ) is essentially Ȟ 2 (M, O× ) where
O× is the sheaf of nowhere vanishing holomorphic functions on M .
In particular, the multiplication in Pic(Pn (C)) on the classes [H(m)] takes the form
[H(m)]∗[H(k)] = [H(m+k)] = [H(m)⊗H(k)] and [H(m)]∗[H(−m)] = [H(0)] = unit .
As a consequence, the results in 3.23 imply, that the [H(m)], m ∈ Z, provide a subgroup
{[H(m)] | m ∈ Z} ⊂ Pic(M ) of the Picard group Pic(M ) isomorphic to Z .
Moreover, by elementary cohomological methods it can be shown that this subgroup
is the full Picard group, i.e. Pic(Pn (C)) ∼
= {[H(m)] | m ∈ Z} ∼
= Z. As a consequence,

Picdiff (Pn ) ∼
= Pic(Pn ) ∼
=Z∼
= {[H(m)] | m ∈ Z}. (22)
60 3. Line Bundles

Summary:
In this chapter the basic notions and properties of line bundles and its cocycles are
presented in order to be able to develop the geometry of line bundles which encompasses
connections, parallel transport, curvature and Hermitian structure, the subject of the
next four chapters. In addition, the example of the tautological line bundle T over
the complex projective space Pn (C) is studied in detail, thereby applying the cocycle
representation in order to determine the space of holomorphic sections Γhol (Pn (C)), L
for the line bundle L = T and its companions L = T ⊗m (= H(−m)) for all m ∈ Z.
There is a general aspect present in the programme of Geometric Quantization:
Whenever holomorphic structure is available, i.e. when M can be viewed as to be a
complex manifold, it is worthwhile to investigate the holomorphic case besides the
smooth case, since, in general, for holomorphic functions there are strong results avail-
able for applications in Geometric Quantization. Note, that a symplectic manifold is
not far from being a complex manifold, as well.
Because of the need of holomorphic functions in Geometric Quantization we present
basic notions and results of complex analysis in the Appendix.
 61

4 Connections

The concept of a connection on a line bundle and - more generally - on a vector bundle
is fundamental for the geometry of such bundles: A connection is the basic geometric
structure on a bundle. Moreover, connections on line bundles are needed in the program
of Geometric Quantization.
A connection on a vector bundle induces a connection form on the corresponding
frame bundle, - which is L× in the case of a line bundle L. Conversely, a connection form
on a principal fibre bundle P over M with structure group G induces a connection on
all associated vector bundles Eρ = P ×ρ Cr , where ρ : G → GL(r, C) is a representation
of Lie groups. In this chapter we focus on line bundles and present several equivalent
descriptions of the notion of a connection. The general case of vector bundles and
principal fibre bundles is treated in the Appendix D.
In the ”ansatz” of geometric quantization presented in Chapter 2 together with our
understanding of sections in a line bundle we are guided to replace E(M ) with the
space Γ(M, L) of sections of a line bundle L. This leads immediately to the question
of what the directional derivatives should be in the generalized context of line bundles.
Geometry gives the answer: the Lie derivatives LX on E(M ) are replaced by covariant
derivatives (also called connections) ∇X for vector fields X ∈ V(M ).

4.1 Local Connection Form

Definition 4.1. A Connection (”Zusammenhang”) on a line bundle π : L → M

over a manifold M is a collection of maps

∇ : V(U ) → End C (Γ(U, L)) , X 7→ ∇X ,

indexed23 by the open subsets U ⊂ M , which are compatible with restrictions to open
subsets V ⊂ U 24 , such that the following properties are satisfied:

(K1) ∇ is an E(U ) module homomorphism with respect to X, i.e.

∇f X+Y = f ∇X + ∇Y , ∀X, Y ∈ V(U ), ∀f ∈ E(U );

(K2) ∇X ∈ End C (Γ(U, L)) for X ∈ V(U ) with the derivative-like property:

∇X (f s) = (LX f )s + f ∇X s , for all s ∈ Γ(U, L) , f ∈ E(U ) , X ∈ V(U ).

23
We omit the index ”U ” attached to ∇X to make formulas not too complicated.
24
That ∇X has to be compatible with restrictions means essentially the following if V is an open
subset of U : For X ∈ V(U ) and s ∈ Γ(U, L) the restrictions satisfy (∇X (s))|V = ∇X|V (s|V )
62 4. Connections

A connection in the form above is also called a Koszul connection. The operator
∇X is called Covariant Derivative (”Kovariante Ableitung”).

The Case of a Trivial Bundle:

We observe immediately that on the trivial bundle

L = M × C,

where we have Γ(U, L) ∼ = E(U ) for open U ⊂ M , the Lie derivative LX : E(U ) → E(U )
is an example of a connection, since LX+Y = LX + LY as well as Lf X = f LX , and
LX (f g) = (LX f )g + f (LX g).
Are there more connections on M × C? How do they look?
Every section s ∈ Γ(M, M × C) has the form s(a) = (a, B(a)), a ∈ M , with a
uniquely defined function B ∈ E(M ). With respect to the previously defined section
s1 ∈ Γ(M, L) , s1 (a) = (a, 1), a ∈ M , we have s = Bs1 . A given connection ∇ defines,
in particular, a map
B : V(M ) → E(M )
by the unique function B(X) ∈ E(M ) with ∇X s1 = B(X)s1 , or ∇X s1 (a) =
(a, B(X)(a)) , a ∈ M . By (K1) B is E(M )-linear:

B(f X +Y )s1 = ∇f X+Y s1 = f ∇X s1 +∇Y s1 = f B(X)s1 +B(Y )s1 = (f B(X)+B(Y ))s1 .

Therefore, B is a 1-form B ∈ A1 (M ).

Lemma 4.2. Every connection ∇ on M × C = L → M is of the form:

∇X s = (LX f + 2πiA(X)f ) s1 , if s = f s1 |U , f ∈ E(U )

 
LX f
= + 2πiA(X) s,
f

where s = f s1 |U , f ∈ E(M ) and X ∈ V(M ), and where A ∈ A1 (M ) is a one form25 .

Conversely, for each A ∈ A1 (M ) the above formula defines a connection on L.

1
Proof. ∇X s = ∇X (f s1 ) = (LX f )s1 + f B(X)s1 according to (K2), and with A = 2πi
B
we get the corresponding form.
Conversely, for any A ∈ A1 (M ):

∇X s := (LX f + 2πiA(X)f ) s1 , s = f s1 ∈ Γ(U, L), X ∈ V(U ),

25
The choice of the factor 2πi in the formula fits to several expressions in the analysis of connections
and their curvatures. It can be replaced by other constants.
4.1 Local Connection Form 63

defines a connection on L. (K1) is evidently satisfied. For (K2), take g ∈ E(U ). Then
for s = f s1 , f ∈ E(U ):

∇X (gs) = ∇X (gf )s1 = (LX (gf ) + 2πiA(X)(gf ))s1

= ((LX g)f + gLX f + gA(X)f )s1 = (LX g)s + g∇X s.

Note, that

Γ(U, T ∗ M ⊗ L) ∼
= (V(U ))∗ ⊗ Γ(U, L) ∼
= HomE(M ) (V(U ), Γ(U, L))

And a connection according to Definition 4.1 can also be described by a C-linear map

∇ : Γ(U, L) → Γ(U, T ∗ M ⊗ L)

with the property

∇(f s) = df ⊗ s + f ∇s, s ∈ Γ(U, L), f ∈ E(U )

and compatible with restrictions.

Local Connection Form:

We now want to describe an arbitrary connection on a non-trivial line-bundle L →
M locally by using Lemma 4.2. Let (Uj ) be an open cover of M with local trivializations
ψj : LUj → Uj × C and corresponding transition functions gjk ∈ E(Ujk , C× ). For each
Uj × C there exists Aj ∈ A1 (Uj ) such that

∇X |Uj (f sj ) = (LX f + 2πiAj (X)f )sj , (23)

where sj (a) = ψj−1 (a, 1), as before.

We want to discover the interrelations of Aj and Ak on an intersection Ujk = Uj ∩Uk .
Given a section s ∈ Γ(U, L), we describe it locally by s|U ∩Uj = fj sj , s|U ∩Uk = fk sk ,
with suitable fj ∈ E(U ∩ Uj ) , fk ∈ E(U ∩ Uk ). We obtain

s|U∩ Ujk = fj sj = fk sk , on Ujk ∩ U

Consequently, on Ujk the local expressions for ∇X s fulfill

∇X s = (LX fj + 2πiAj (X)fj ) sj = (LX fk + 2πiAk (X)fk ) sk

We insert sj = gkj sk , fj = gjk fk , and obtain

(LX fk + 2πiAk (X)fk ) sk = (LX (gjk fk ) + 2πiAj (X)gjk fk ) gkj sk

= ((LX gjk )fk + gjk LX fk + 2πiAj (X)gjk fk ) gkj sk
= (LX fk + 2πiAj (X)fk ) sk + (LX gjk )fk gkj sk
64 4. Connections

Hence,
dgjk (X)
2πiAk (X) = 2πiAj (X) + dgjk (X)gkj = 2πiAj (X) + ,
gjk
which implies

1 dgjk
(Z) Ak = Aj + (24)
2πi gjk

on Ujk .
The Aj are the Local Connection Forms of the connection ∇. We have shown
the first part of the following result:

Proposition 4.3. Let (Uj ) be an open cover of M such that the line bundle L →
M of M has trivializations over Uj with transition functions gjk ∈ E(Ujk , C× ). Any
connection ∇ on L determines uniquely a collection (Aj ) of 1-forms Aj ∈ A1 (Uj ) with
(Z). Conversely, (Aj ) with (Z) induces a connection on L.

Proof. Indeed, given s ∈ Γ(U, L) with s|U ∩Uj = fj sj , fj ∈ E(U ∩ Uj ), by

∇X fj sj := (LX fj + 2πiAj (X)fj )sj

a connection is defined.

The local 1-forms Aj are called the Local Connection Forms (”Zusammen-
hangsform”) of the connection ∇.

4.2 Global Connection Form

In the next step, we want to understand how every line bundle connection is induced
by a global connection on the corresponding frame bundle L× ⊂ L.
Let π : L → M be a line bundle. The Frame Bundle (”Reperbündel”) L×
belonging to L is the bundle

L× := {p ∈ L | p ̸= 0a , a = π(p)} = L \ {0a ∈ La | a ∈ M } = L \ z(M ) ,

where z : M → L , a → 0a , is the zero section. L× is a principal fibre bundle with struc-

ture group C× , the multiplicative group of nonzero complex numbers (cf. Appendix,
Construction D.4). The projection

π : L× → M,

is the restriction of the projection π : L → M . Moreover, the right action of C× is

Ψ : L× × C× → L× , (p, z) 7→ pz = Ψ(p, z),

4.2 Global Connection Form 65

where pz = Ψ(p, z) is simply the multiplication in the fibre La , a = π(p). The frame
bundle L× has the local trivializations

ψj : L× ×
Uj → Uj × C ,

the restrictions of the trivializations ψj : LUj → Uj × C of the original line bundle

L → M . ψj respects the right action (the multiplication) of C× on L× : ψj (pz) = ψj (p)z
for all (p, z) ∈ L× × C× .
Let us consider a connection ∇ on L given by the one-forms Aj ∈ A1 (Uj ) satisfying
(Z) with respect to a suitable open cover (Uj ) and corresponding transition functions
gij , according to Proposition 4.3. Then the forms Aj can be lifted to L× Uj as the
pullbacks by π (c.f. Appendix, Definition A.33). We obtain

π ∗ (Aj ) ∈ A1 (L×
Uj ), j ∈ I.

By Proposition 4.3 the local forms Aj satisfy (Z), hence

 
∗ ∗ ∗ 1 dgjk
π Ak = π Aj + π .
2πi gjk

Moreover, one can show the following lemma:

Lemma 4.4.      
dz dz dgjk
ψj∗ − ψk∗ =π ∗
z z gjk

Here, dz
z
is an abbreviation of pr2∗ ( dz
z
) on Uj resp. Uk , or simply z1 dz on Uj × C× .

Proof. Exercise!

As a consequence of these technical results, the two expressions

   
∗ 1 ∗ dz ∗ 1 ∗ dz
π Aj + ψ , π Ak + ψ
2πi j z 2πi k z

agree on L× 1 ×
Ujk and define a global 1-form on α on A (L ) by:
 
∗ 1 ∗ dz
α|L× := π Aj + ψ , j ∈ I. (25)
Uj 2πi j z

This 1-form α ∈ A(L× ) is the Global Connection Form of the connection! It

is independent of the choice of local forms.
In order to investigate the main properties of the global connection form α let us
introduce the fundamental field induced by the action of C× on L× . This is a special
case of the notion of a fundamental field on a principal fibre bundle as defined in D.11.
66 4. Connections

Definition 4.5. For ξ ∈ C = Lie C× define the Fundamental Field Yξ : L× → T L×

by:
d
Yξ (p) := (p exp(2πiξt)) |t=0 = [p exp(2πiξt)]p ∈ Tp L× .
dt
It is easy to see that T π(Yξ ) = 0 and Yξ (p) is a tangent vector along the fibre L× a
for p ∈ L× with a = π(p): Yξ (p) ∈ Tp L× a . More is true: The fundamental vector
fields generate the so called Vertical Bundle V := Ker T π. Since the map C →
Tp L×a , ξ 7→ Yξ (p) is a linear isomorphism (Yξ (p) ̸= 0 for ξ ̸= 0!), we have

Vp = Tp L×
a = {Yξ (p) | ξ ∈ C}. (26)

In particular, a vertical vector field X, i.e. X ∈ V(L× ) , X(p) ∈ Vp for all p ∈ L× ,

can be described by X(p) = Yξ(p) where ξ(p) = σp (X(p)) with σp : Vp → C the inverse
of ξ 7→ Yξ (p). Note, that Yξ is not the same as Yξ(p) , the above defined fundamental
field Yξ is Yξ(p) with constant map ξ(p) = ξ!
Lemma 4.6. The global connection form α ∈ A(L× ) of a connection ∇ on L satisfies

(I1) α (Yξ ) = ξ for all ξ ∈ C,

(I2) Ψ∗c α = α for all c ∈ C× .

Proof. Let p ∈ L× ×
Uj with ψj (p)) = (a, z) ∈ Uj × C . Then π(p exp(2πiξt)) = π(p),
hence π ∗ Aj (Yξ ) = Aj (0) = 0. Furthermore,

ψj∗ (dz) (Yξ )(p) = dz[(a, z exp(2πiξt))](a,z) = dz[z exp(2πiξt)]z = 2πiξz ,

hence  
1 ∗ dz
ψ (Yξ ) = ξ.
2πi j z
Using the definition of the global connection form α (cf. (25)), we deduce condition
(I1).
To show (I2) let X = [p(t)]p be a tangent vector at p = p(0) ∈ L× |Uj . By definition
we have
 
∗ 1 dz 
Ψc α(X)(p) = α([p(t)c]pc ) = Aj ([π(p(t)c)]π(pc) ) + [ψj (p(t)c))]ψ(pc) . (27)
2πi z

Because of π(p(t)c) = π(p(t)), the first term in (27) satisfies

Aj [π(p(t)c)]π(pc) = Aj [π(p(t))]π(p) .

For the second term of (27) let

ψj (p(t)) = (a(t), z(t)) with ψj (p) = (a(0), z(0)) =: (a, z) .

4.2 Global Connection Form 67

Then ψj (p(t)c) = (a(t), z(t)c)
implies dz([ψj (p(t)c))]ψj (pc) ) = dz([(a(t), z(t)c)](a,zc) ) =
ż(0)c and dz [ψj (p(t)))]ψj (p) = dz([(a(t), z(t))](a,z) ) = ż(0). Hence, the second term
of (27) is also independent of c:
   
1 dz
1 ż(0)c 1 ż(0) 1 dz

[ψj (p(t)c))]ψ(pc) = = = [ψj (p(t))]ψj (p)
2πi z 2πi z(0)c 2πi z(0) 2πi z

This proves (I2).

Proposition 4.7. Let α ∈ A1 (L× ) be the global connection form of a connection,

defined as above. Then α satisfies the conditions (I1), (I2), and the covariant derivative
has the form

∇X s = 2πis∗ α(X)s, (28)

for X ∈ V(U ) and s ∈ Γ(U, L× ) , where U ⊂ M is open in M .

Conversely, each α ∈ A1 (L× ) with (I1) and (I2) defines a connection on L by (28).

Proof. We confirm the formula ∇X s = 2πis∗ α(X)s and leave the rest as an exercise:
We can restrict the consideration to the case s ∈ Γ (Uj , L). We have s = f sj , where
sj (a) := ψj−1 (a, 1) and f ∈ E (Uj ). We know from Proposition 4.3

∇X s = (LX f + 2πiAj (X)f ) sj ,

where Aj is the local connection form. Now,

   
∗ ∗ ∗ 1 ∗
∗ dz ∗ ∗ 1 dz
s α = s π Aj + s ψ = (π ◦ s) Aj + (ψj ◦ s) .
2πi j z 2πi z

We know π ◦ s = idUj and ψj ◦ s(a) = (a, f (a)) , a ∈ Uj . We assume Uj to be a

coordinate neighbourhood, so that a vector field X ∈ V (Uj ) has the form X : Uj →
Uj × Rn , a 7−→ (a, V (a)) , V (a) ∈ Rn . We conclude
  
∗ dz dz
(ψj ◦ s) (X)(a) = (T (ψj ◦ s).V (a)) .
z f (a)

The derivative of ψj ◦ s is
 
En
T (ψj ◦ s) = ,
∂1 f, . . . ∂n f

with the Jacobi matrix as a block matrix (En is the identity matrix in Rn×n ). As a
consequence,
T (ψj ◦ s) V (a) = ∂j f (a)V j (a) = LX f (a). (29)
68 4. Connections

Therefore  
∗ 1 LX f
2πis α(X) = 2πi Aj (X) + , i.e
2πi f
2πis∗ α(X)f sj = (2πiAj (X)f + LX f ) sj ,
and finally,
2πis∗ α(X)s = ∇X s.

4.3 Horizontal Bundle

There are more possibilities to describe a connection on a line bundle π : L → M .

With the global connection form α one defines the Horizontal Bundle H := Ker α
over L× . H obtains its manifold structure from the inclusion H ⊂ T L× and its linear
structure from the inclusion Hp ⊂ Tp L× , p ∈ L× . Because of property (I1), α(Yξ ) =
ξ , ξ ∈ C, the linear map αp : Tp L× → C is surjective, so that Ker αp = Hp is a n-
dimensional subspace of the tangent space Tp L× of real dimension n + 2. Consequently,
H is a real vector bundle of rank n.

Proposition 4.8. For a global connection form α on L× the horizontal bundle H =

Ker α satisfies

(H1) H ⊕ V = T L× (H has the vertical bundle as a complement),

(H2) T Ψc (Hp ) = Hpc (H is invariant).

Conversely, such a horizontal bundle defines a connection on L.

Proof. The property α(Yξ ) = ξ (I1) implies Hp ∩ Vp = {0} since Vp is generated by the
Yξ (p) , ξ ∈ C. Therefore, Hp ⊕ Vp = Tp L× , which is (H1). To prove (H2) it is enough to
show T Ψc (Hp ) ⊂ Hpc . But for X ∈ Hp the condition (I2) implies Ψ∗c α(X) = α(X) = 0,
hence α(T Ψc (X)) = 0, i.e. T Ψc (X) ∈ Hpc .
The converse follows from the next proposition.

For a subbundle H with (H1) and (H2) let v : T L× → T L× the projection with
Ker v = H and Im v = V . Then T Ψc ◦ v = v ◦ T Ψc can be deduced from (H2).

Proposition 4.9. Let v : T L× → T L× a homomorphism of vector bundles with

(V1) v ◦ v = v (v is a projection) and Im v = V

(V2) T Ψc ◦ v = v ◦ T Ψc (v is equivariant)

Then α = σ ◦ v defines a connection form.

4.3 Horizontal Bundle 69

Proof. (I1) holds immediately because of v(Yξ ) = Yξ by (V1): α(Yξ ) = σYξ = ξ (recall
that σp : Vp → C is the inverse of the map ξ 7→ [p exp(2πiξt] = Yξ (p)). (I2) is clear for
X ∈ Ker vp : Ψ∗c α(X) = σ ◦ v ◦ T Ψc (X) = σ ◦ T Ψc ◦ v(X) = 0 = α(X). And for Yξ we
obtain

Ψ∗c α(Yξ )(p) = (σpc ◦ vpc ◦ Tp Ψc )(Yξ )(p) = (σpc ◦ vpc (Yξ )(pc) = α(Yξ )(pc),

because of Tp Ψc )(Yξ )(p) = Yξ (pc), which confirms (I2).

For the formulation of the quantum operator in the context of half-density quantiza-
tion and half-form quantization we need another equivalent description of a connection,
which uses the frame bundle L× of L and where the operation ∇X on sections in L is
essentially reduced to a Lie derivative on functions on L× .
We begin with the fact, that the sections of L can be described by invariant functions
on L× . Every section s ∈ Γ(M, L) of the line bundle induces a unique function s♯ ∈
E(L× ) which satisfies
s(a) = ps♯ (p) , p ∈ π −1 (a) ,
for p ∈ L× .

Lemma 4.10. s♯ : L× → C is a well-defined smooth function with s♯ ◦ s = 1. Further-

more,
s♯ (pc) = s♯ (p)c−1
for all p ∈ L× and c ∈ C× . With the notation

E−1 (L× ) := {g ∈ E(L× ) | g(pc) = c−1 g(p) = g(p)c−1 for all (p, c) ∈ L× × C× }

the map
♯
: Γ(M, L) → E−1 (L× ) , s 7→ s♯ ,
is an isomorphism of E(M )-modules.

Proof. Of course, s♯ is well-defined and smooth, which can be seen using a local triv-
ialization. The identity s♯ ◦ s = 1 follows directly from s(a) = s(a)s♯ (s(a)). Because
of s(a) = ps♯ (p) = pcs♯ (pc) the invariance s♯ (pc) = c−1 s♯ (p) holds. It is easy to see,
that ♯ is E(M )-linear and injective. Finally, this map is surjective: For g ∈ E−1 (M ) the
definition s(a) := pg(p), p ∈ π −1 (a), yields a section s of L with s♯ = g.

The action of the fundamental field Yc on the invariant functions g ∈ E−1 (L× ) is
simply multiplication:

Lemma 4.11. LYc g = −2πig for g ∈ E−1 (L× ).

70 4. Connections

Proof.
d
LYc g(p) = g(e2πict p)|t=0
dt
d
= e−2πict g(p)|t=0
dt
= − 2πicg(p)

Now, let a connection on L be given by the global connection form α and its
horizontal bundle H ⊂ T L× . Since for p ∈ L×
a , a ∈ M , the restriction

Tp π|Hp : Hp → Ta M

of Tp π is bijective, each X ∈ Ta M has a unique lift X ♯ ∈ Hp such that Tp π(X ♯ ) = X.

Definition 4.12. X ♯ ∈ Hp is called the horizontal lift of X ∈ Ta M .

The horizontal lift X ♯ is determined by the two conditions Tp π(X ♯ ) = X and

α(X ♯ ) = 0, in other words: {X ♯ } = (Tp π)−1 (X) ∩ Ker αp .
The following result shows that the connection ∇X can be defined as a Lie derivative
on the level of L× .
Proposition 4.13. Let ∇ be a connection on L. For sections s ∈ Γ(M, L) and vector
fields X ∈ V(M ) the following formula holds

LX ♯ s♯ = (∇X s)♯ .

Proof. First of all, we show that X ♯ (p) = Ta s(X(a)) − Ys∗ α(X) (s(a)), π(p) = a, by
establishing the above two conditions. Because of αp (Ta s(X(a)) = (s∗ αp )(X(a)), we
have
α(Ta s(X(a)) − Ys∗ α(X) (s(a))) = s∗ α(X(a)) − s∗ α(X(a)) = 0 .
Moreover, Tp π(Yc ) = 0, which implies

Tp π(Ta s(X(a)) − Ys∗ α(X) (s(a))) = Tp πTa s(X(a)) = X(a).

We now determine LT s(X) s♯ and LYs∗ α(X) s♯ to obtain LX ♯ s♯ . Let γ(t) represent X(a),
i.e. X(a) = [γ(t)], where γ(0) = a.
d ♯
LT s(X) s♯ (p) = s (s(γ(t)))|t=0 = 0 ,
dt
since s♯ ◦ s = 1. The outcome of the other term is

LYs∗ α(X) s♯ = −2πis∗ α(X)s♯ ,

4.3 Horizontal Bundle 71

according to Lemma 4.11.

Altogether, LX ♯ s♯ = 2πis∗ α(X)s♯ and the last term is (∇X s)♯ since 2πis∗ α(X)s =
∇X s saccording to Proposition 4.7

The last result gives another formulation of the concept of a connection:

Proposition 4.14. The horizontal lifts define a bundle homomorphism

Γ : T M ×M L× → T L× , (X, p) 7→ X ♯ ,

on the fibre product T M ×M L× (cf. Definition A.8), which is equivariant, i.e. such
that
Γ(X, pc) = T Ψc (Γ(X, p)).
Conversely, an eqivariant bundle homomorphism Γ : M ×M L× → T L× defines a
connection.
Remark 4.15. The horizontal lift Γ : M ×M L× → T L× is sometimes called an
Ehresmann connection. And this expression is also used for the definition of
a connection on a bundle in terms of the (equivariant) horizontal bundle H or the
(equivariant) projection v.

We collect the results on the 6 different ways to introduce a connection on a line

bundle:
Proposition 4.16. A connection on a line bundle L → M is given by one of the
following five equivalent data:

1. A Covariant Derivative ∇X : Γ(U, L) → Γ(U, L) satisfying (K1) and (K2)

of Definition 4.1 (c.f. D.5 for the vector bundle case).

2. A collection of 1-forms, the Local Connection Forms, (Aj )j∈I , Aj ∈

A1 (Uj ), satisfying (Z), with respect to an open cover (Uj )j∈I , where LUj has
local trivializations ψj , see Proposition 4.3 and D.14.

3. A Global Connection Form α on the frame bundle L× with the two proper-
ties (I1), (I2) in Lemma 4.6 and D.18.

4. A vector bundle H, the so called Horizontal Bundle H ⊂ T L× with (H1)

(i.e. H is a complement to the vertical bundle V ) and (H2) (i.e. H is equivariant),
see Proposition 4.8 and Proposition D.15.

5. A vector bundle homomorphism v : T L× → T L× with (V1) (i.e. v is a projection

onto the Vertical Bundle and (V2) (i.e. v is equivariant), see Proposition 4.9
and Proposition D.16.

6. An equivariant lifting Γ : T M ×M L× → T L× .
72 4. Connections

To finish the chapter, we illustrate the 6 equivalent data on the trivial line bundle
U × C in a local situation, i.e. U is an open subset of Rn .

1. ∇ is given by A = Ai dq i ∈ A1 (U ) , Ai ∈ E(U, C) as a local connection form, in

such a way that ∇X for X = X i ∂i is

∇X f s1 = (LX f + 2πiA(X)f ) s1
= X i ∂i f + 2πiAi X i f s1

or
∇X (a, f (a)) = a, X i (∂i f + 2πiAi f )

for f ∈ E(U ), where s1 (a) = (a, 1), a ∈ U .

2. This is essentially the same, since Uj = U .

3. The global connection form α on L× ∼

= U × C× is
1 dz 1 dz
α=A+ = Ai dq i +
2πi z 2πi z

4. The horizontal subspace Hp ⊂ Tp L× ∼

= Rn × C in p = (a, z) ∈ L× , is given by
Ai ∈ E(U ) as
 
n i ξ
Hp = (X, ξ) ∈ R × C | 2πiAi (a)X + = 0 .
z

5. This is essentially the same as in 4., but now H being defined by the corresponding
projection v : T L× → T L× onto the vertical bundle V : In p = (a, z) ∈ L× given
by
vp (X, ξ) = (0, ξ + 2πiAi (a)X i z) ,
with p = (a, z).

= U × Rn × C× and T L× ∼
6. With T U ×U L× ∼ = (U × C× ) × (Rn × C) the lift
n × × n
Γ : U × R × C −→ (U × C ) × (R × C) takes the form

Γ(X, z) = (X, −2πiAj X j z) ,

where we drop the base points a, p = (a, z). The equation has the meaning that
for a given local connection potential A = Ai dq i the lift Γ is of the described
form. It also has the meaning, that an equivariant lifting Γ determines A.

In order to recognize the geometric nature of the notion of connection, we present

– in the Appendix – the concept of a connection on a principal fibre bundle and its
associated vector bundles, and we summarize some properties on principal connections.
In this way, we can regard connections on a line bundle with its principal fibre bundle
4.3 Horizontal Bundle 73

L× in the framework of general connections. Some parts become more complicated,

but others look simpler in the general case.

Summary: The description of a connection on a line bundle is now complete.

The geometric nature of this concept has not been emphasized so far except for the
decomposition of the tangent bundle of the frame bundle L× into its vertical and
horizontal subbundles. The geometry of connections on a line bundle is the subject
of the next chapters where we investigate parallel transport, curvature and Hermitian
structure.
74 5. Parallel Transport and Curvature

5 Parallel Transport and Curvature

We introduce and study parallel transport (also called horizontal transport) along a
curve γ : I → M in the base manifold M induced by a connection ∇ on a line bundle
π : L → M . Furthermore, in order to determine whether the parallel transport is inde-
pendent of the curves, horizontal sections and the curvature of (L, ∇) are investigated.
An important result is the integrality condition for a 2-form on M to be the curvature
of a line bundle on M which is needed for the program of Geometric Quantization.

5.1 Parallel Transport

Definition 5.1. Let π : L → M be a line bundle with connection ∇, and let p ∈ L× ⊂

L be a point in the frame bundle of L with base point a = π(p).

1. A Horizontal (or Parallel) Lift (”horizontale Liftung”)26 of a tangent vec-

tor X ∈ Ta M at p is a tangent vector X ♯ ∈ Tp L with:

(a) Tp π(X ♯ ) = X (X ♯ is a Lift)

(b) X ♯ ∈ Hp (X ♯ is Horizontal)

2. Let γ be a (smooth) curve γ : I → M in M (where I ⊂ R is an open interval).

A horizontal lift of γ (through p0 ∈ Lγ(0) ) is a smooth curve λ : I → L with
γ(t0 ) = p0 , t0 ∈ I , such that

(a) γ = π ◦ λ (λ is a lift of γ through p0 ), and

(b) λ̇(t) ∈ Hλ(t) for all t ∈ I (the lifted curve λ is horizontal)27 .

Remark 5.2. Recall, that a horizontal lift X ♯ of X exists, since Tp π : Tp L× → Ta M

is surjective, and it is unique, since Tp π|Hp : Hp → Ta M is an isomorphism. Also,
if λ is a horizontal lift of γ then each λ̇(t) is the horizontal lift of γ̇(t). As a result,
the horizontal lift of a curve is unique if it exists (and it exists as we will see soon in
Proposition 5.3).

In order to explain the definition, the notion of horizontal subbundle H ⊂ T L×

induced by the connection ∇ on L will be recalled (c.f. Section 4.3): For a point a ∈ M
there exists a trivialization
ψ : LU → U × L
26
known already from Definition 4.12
27
A remark on the notation λ̇(t) or γ̇(t) seems to be appropriate: γ̇(t) is the tangent vector at the
point γ(t) = a ∈ M given by the curve s 7→ γ(t + s), i.e. γ̇(t) = [γ(t + s)]a ∈ Ta M. Also, with
1 ∈ Tt I ∼
= R we can write γ̇(t) = Tt γ(1) ∈ Ta M .
5.1 Parallel Transport 75

of the line bundle LU = π −1 (U ) → U over an open neighbourhood U of a. With respect

to this trivialization the connection ∇ has the form
∇X f s1 = (LX f + 2πiA(X)f )s1 , f ∈ E(U ), X ∈ V(U ),
where s1 (a) := ψ −1 (a, 1) as before, and A ∈ A1 (U ) is a one-form, the local connection
form (or local gauge potential), uniquely defined by ∇: A(X) ∈ E(U ) is the function
with ∇X s1 = 2πiA(X)s1 . The horizontal space Hp ⊂ Tp L× , ψ(p) = (a, z) ∈ U × C× ,
is now given by
 
−1 × × ζ
Hp := Y = T(a,z) ψ (X, ζ) ∈ Tp L | (X, ζ) ∈ Ta U ×Tz C : 2πiA(X)+ = 0 . (30)
z

This digression shows again that every X ∈ Ta M has a unique horizontal lift
X ∈ Tp M through a point p = ψ −1 (a, z) ∈ L×
♯ ♯ −1
a : X = T(a,z) ψ (X, −2πiA(X)z).
♯
Moreover, the map Γp : Ta M → Hp , X 7→ X is an isomorphism.
The collection of these lifts Γp yield the Ehresmann connection Γ : T M ×M L× →
T L× introduced in Proposition 4.14.
If q 1 , . . . , q n are local coordinates in U around a with A = Ak dq k , then the
Yk = Tψ(p) ψ −1 (∂k , −2πizAk )
k = 1, . . . , n span the horizontal subspace Hp , so they provide a vector space basis of
Hp .
Proposition 5.3. Let ∇ be a connection on the line bundle L → M , and let γ : I → M
be a (smooth) curve with γ (t0 ) = a. For every point p ∈ L× a there exists a uniquely
×
defined horizontal lift γ̂ : I → L through p : γ̂ (t0 ) = p. In particular, the horizontal
lift of the curve γ through pc is the curve γ̂c for c ∈ C× .

Proof. Let ψ(p) = (a, z). In the above local situation one looks for a curve ζ : I → C× ,
ζ(t0 ) = z such that γ̂ := ψ −1 (γ, ζ) is a lift of γ with γ̂ (t0 ) = ψ −1 (γ (t0 ), ζ (t0 )) = p.
In order that this lift γ̂ through p is, moreover, horizontal, the two curves γ, ζ have to
satisfy
ζ̇(t)
2πiA(γ̇(t)) + = 0,
ζ(t)
according to (30), which amounts to the differential equation
ζ̇(t) = −2πiA(γ̇(t))ζ(t)
And this differential equation has a unique solution on all of I with ζ (t0 ) ∈ C× (initial
value problem for linear ordinary equations). The solution can be written in the form
 Z t 
ζ(t) = c exp −2πi A(γ̇(τ ))dτ ,
t0

with c = ζ(t0 ).
76 5. Parallel Transport and Curvature

Corollary 5.4. From the proof of the proposition we obtain the following characteri-
zation for a lift to be horizontal in a local situation: A lift ψ −1 (γ, ζ) of γ is horizontal
if and only if

ζ̇(t) + 2πiA(γ̇(t))ζ(t) = 0,

for t ∈ I, which is - in slightly greater detail ζ̇(t) + 2πiAj (γ(t))(γ̇ j (t))ζ(t) = 0, or, in
a very short form:

∇γ̇ ζ̇ = 0 .

Observation 5.5. This result allows it to extend the horizontal lifting through all
points of the fibre La , i.e. also through p ∈ L \ L× . However, if the horizontal lift γ̂
of a curve γ once is zero (γ̂(t1 ) = 0γ(t1 ) ) over a point b = γ(t1 ) it vanishes completely:
γ̂(t) = 0γ(t) for all t ∈ I.

Remark 5.6. Note, that the definitions and results extend directly to connections on
a vector bundle E of rank k. Such a connection ∇ is locally given by

∇X f = LX f + A(X)f, X ∈ V(U ), f ∈ E (U, Kr )

where A ∈ A1 (U, End Kr ) is a g = End Kr -valued 1-form. Hence a horizontal lift of a

curve γ : I → M, γ(t0 ) = a, looks locally like γ̂ = ψ −1 (γ, η), with η ∈ E(I, Kr ) and

η̇ + A(γ̇)η = 0 .

Proposition 5.3 leads to the concept of ”parallel transport” or ”horizontal trans-

port”.

Definition 5.7. With the notation of the last proposition and the choice of another
t1 ∈ I let γ̂ = γˆp be the horizontal lift of γ through γ̂ (t0 ) = p ∈ L×
a . Then the map

p 7→ γˆp (t1 ) , L× ×
γ(t0 ) → Lγ(t1 ) ,

is bijective: According to the last statement in Proposition 5.3 pc is mapped to γˆp (t1 ) c.
The mapping can be continued to all of Lγ(t0 ) , Lγ(t1 ) by 0γ(t0 ) 7→ 0γ(t1 ) , or using the fact,
mentioned above in Observation 5.5, that the zero lift ν(t) := 0γ(t) over γ can also be
viewed as a horizontal lift of γ. In this way one obtains an isomorphism (of C-vector
spaces) Lγ(t0 ) → Lγ(t1 . This isomorphism is called Parallel Transport Along γ
(also called horizontal transport along γ) and will be denoted by

Pγt1 ,t0 : Lγ(t0 ) → Lγ(t1 ) .

5.2 Horizontal Section 77

The parallel transport Pγt1 ,t0 (also called parallel operator) describes a linear shift
of vectors over γ (t0 ) to those over γ (t1 ). This shift depends in general on the curve
from γ (t0 ) to γ(t1 ) (see below).
The parallel operators Pγs,t have many natural properties like

Pγs,t ◦ Pγt,s = idLγ (s) and Pγr,s ◦ Pγs,t = Pγr,t for r, s, t ∈ I (31)

One can reconstruct the connection ∇ from the family Pγt0 ,t1


γ,t0 ,t1
(c.f. Proposition
12.3).

5.2 Horizontal Section

Definition 5.8. A section s ∈ Γ (U, L× ) over an open subset U ⊂ M is called hori-

zontal if
Ta s (Ta M ) ⊂ Hs(a)
holds for all a ∈ U .

In case of a horizontal section s ∈ Γ (U, L× ) one even has Ta s (Ta M ) = Hs(a) , and
Ta s is the inverse of the restriction Ts(a) π H : Hs(a) → Ta M for all a ∈ U .
s(a)

The inclusion Ta s (Ta M ) ⊂ Hs(a) for a horizontal section s over U implies that for
each curve γ : I → U, γ(0) = a, the composition λ := s ◦ γ satisfies

λ̇ = Tγ(t) s(γ̇(t)) ∈ Hs(γ(t)) ,

i.e. λ = s ◦ γ is a horizontal lift of γ. Hence, with the notation s ◦ γ = ψ −1 (γ, ζ) in a

local trivialization ψ : L× |U ′ → U ′ × C× (where U ′ ⊂ U is an open subset of U ), the
second component ζ(t) = pr2 ψ(s ◦ γ(t)) satisfies

ζ̇ + 2πiA(γ̇)ζ = 0,

and we conclude that ∇X s = 0 for all X ∈ V(U ): When X(a) = [γ(t)]a we obtain,
using s(γ(t)) = ζ(t)s1 (γ(t)) = ψ −1 (γ(t), ζ(t)), that

∇X s(a) = ∇X ζs1 (γ)|(t0 )

= (LX ζ + 2πiA(X)) ζ)|t0 s1 (a)
 
= ζ̇(t0 ) + 2πiA(γ̇(t0 ))ζ(t0 ) s1 (a)
= 0 s1 (a) = 0 .

We have essentially shown:

Proposition 5.9. Let L → M be a line bundle with connection ∇. Then s ∈ Γ(U, L× )
is a horizontal section if and only if ∇X s = 0 for all X ∈ V(U ).
78 5. Parallel Transport and Curvature

Examples 5.10.

1. ln the trivial case L = M × C, let A be the zero potential A = 0, i.e. ∇X f s1 =

LX f s1 . In this case: s = f s1 is horizontal if and only if f (and hence s) is locally
constant.

2. Again in the trivial case L = M × C with M = R2 let us consider the connection

given by the non-zero local connection form A = q 2 dq 1 − q 1 dq 2 . (Note, that
dA ̸= 0 , i.e. the connection has non-zero curvature, see below.) We show that
for this connection there is no horizontal section. If s(a) = (a, f (a)), a ∈ U ,
would be a horizontal section on U ⊂ M open, U ̸= ∅, with f (a0 ) ̸= 0 at
one point a0 ∈ U we can assume f (a) ̸= 0 throughout U (by possibly taking
a smaller neighbourhood of a0 ). Then Proposition 5.9 implies ∇X s = 0, i.e.
LX f + 2πiA(X)f = 0. Hence,
∂f ∂f
1
+ 2πiA1 f = 1 + 2πiq 2 = 0
∂q ∂q
∂f ∂f
2
+ 2πiA2 f = 2 − 2πiq 1 = 0
∂q ∂q
and this leads to the contradiction
∂ 2f
−2πi = + = 2πi.
∂q 1 ∂q 2

One can prove the following direct relation between ∇ and the corresponding par-
allel transport:
Proposition 5.11. For every curve γ : I → U and every section s ∈ Γ(U, L)
1 γ

∇X s(a) = lim Pt0 ,t0 +t (s ◦ γ(t0 + t)) − s(a) (32)
t→0 t

where γ represents X(a): X = γ̇(t0 ) = [γ]a , γ(t0 ) = a.

This result leads to an interesting Geometric Interpretation of the covariant

derivative: The covariant derivative ∇X measures along the curve γ to what extent a
given section s of the line bundle deviates infinitesimally from being horizontal.
The connection can be reconstructed knowing the action of sufficiently many of its
parallel operators:
Proposition 5.12. Let π : L → M be a line bundle over M . Let (Pγs,t ) a collection of
linear operators Lγ(t) → Lγ(s) assigned to all curves γ : I → M and s, t ∈ I. Assume
that (31) is satisfied, that Pγs,t depends differentiably on t and that the Pγs,t do not depend
on the parametrization of the curves γ. Then (32) defines a connection with the given
Pγs,t as parallel operators.
5.3 Curvature 79

This construction is not needed in the sequel. The rather involved proof will not
presented.
This result yields another way to characterize the notion of a connection. In [Poo81]
a comprehensive study of the various appearances of the concept of connections on
general bundles can be found.
Under which conditions does there exist a horizontal section, at least locally? We
have seen, that in case of a horizontal section s ∈ Γ (U, L× ) for each curve γ in U
its horizontal lift through s (γ (t0 )) has the form s ◦ γ. Consequently, for any two
points a, b ∈ U and any curve γ in U with γ (t0 ) = a, γ (t1 ) = b, parallel transport
of p = s(a) = s (γ (t0 )) ∈ La to Lb along γ is s ◦ γ (t1 ) = s(b) : Pγt1 ,t0 (s(a)) = s(b)
independently of γ (as long as the curves stay in U ). (For p′ ∈ L× ′ ×
a , p = pc, with c ∈ C ,
and s′ (a) = s(a)c is a horizontal section transporting p′ to s(b)c, again independently
of the curve.) We have shown one direction of the following equivalence.
Proposition 5.13. Let L → M be a line bundle with connection ∇ and U ⊂ M open.
Then U admits a horizontal section s ∈ Γ (U, L× ) if and only if the parallel transport
from a point a ∈ U to b ∈ U is independent of the curves in U connecting a and b.

Proof. Assume that parallel transport is independent of the curves. Without loss of
generality we assume furthermore, that U is connected. We obtain to each a ∈ U
and p ∈ L× ×
a a unique horizontal section s : U → L with s(a) = p by the following
γ
prescription: s(b) := Pt1 ,t0 (p), where γ is a curve γ : I → U with γ (t0 ) = a and
γ(t1 ) = b : s(b) is well-defined since the value does not depend on γ, s is smooth since
all the curves γ are smooth, and s is horizontal, since, by definition s ◦ γ(t) is the
horizontal lift of γ.

The question of whether or not parallel transport is independent of the curve con-
necting the points in M is essentially related to the notion of curvature which is the
subject of the next section.

5.3 Curvature

We continue the discussion under which conditions on a given line bundle with con-
nection (L, ∇) there exist horizontal sections

s : U → L×

over an open subset U of M .

Assume, that U is a coordinate neighbourhood with coordinates q 1 , · · · q n , i.e. we
have a diffeomorphism φ = (q 1 , . . . q n ) : U → V ⊂ Rn with φ = (q 1 , . . . , q n ). Let
a ∈ U with φ(a) = 0 and assume V = φ(U ) = I1 × . . . × In where Ik = ]− r, r [ are
equal open intervals around 0. Then the vector fields Xk = ∂k ∈ V(U ) yield a basis
of the E(U )-module of vector fields V(U ). We want to construct a horizontal section
80 5. Parallel Transport and Curvature

s ∈ Γ(U, L× ), and we see, that to check whether s is horizontal it us enough to show

∇Xk s = 0 for k = 1, . . . , n.
We start with the curve γ1 (t) = φ−1 (te1 ) through a representing X1 = ∂1 , t ∈ I1
and choose a point p0 ∈ L× a . There exists the horizontal lift (c.f. Proposition 5.3) γ̂1 of
γ1 through p0 . In particular, γ1 (0) = a and γ̂1 (0) = p0 . We set s (γ1 (t)) := γ̂1 (t), t ∈ I1
with s(a) = p0 . For each t1 ∈ I1 the curve γ2t1 (t) := φ−1 (t1 e
1 + te2 ) , t ∈ I2 ⊂ R, has
again a horizontal lift γ̂2t1 (t) through γ̂1 (t). We set s γ2t1 (t2 ) := γ̂2t1 (t2 ). In the same
way one can proceed with 3, . . . , n. But let us stick to the case n = 2. (The case n > 2
is completely analogous.) Then s as above defines a section s : U = φ−1 (I1 × I2 ) →
L× , s (φ−1 (t1 e1 + t2 e2 )) = γ̂2t1 (t2 ).
Let us check whether s is a horizontal section. s is horizontal if and only if ∇X s = 0
for all vector fields X ∈ V(U ), i.e. if and only if ∇Xi s = 0 for i = 1, 2 in our
special
situation. Now, ∇X2 s = 0 is evident by the definition of s, since t 7→ s γ2 (t) = γ̂2t1 (t)
t1

is horizontal for each t1 ∈ I1 .

If now ∇X1 and ∇X2 commute, we have ∇X2 ∇X1 s = ∇X1 ∇X2 s = 0 (because of
∇2 s = 0) and it follows that (for fixed t1 )

η(t) = ∇X1 s(φ−1 (t1 , t))

is a horizontal lift of γ2 (t1 , t) with η(0) = 0. (η(0) = ∇X1 γˆ1 (t) = 0 since γˆ1 is a
horizontal lift of γ1 ). η is horizontal because of ∇X2 η = 0. Eventually, since the
horizontal lift is unique, we have η(t) = 0 and:

∇X1 s(φ−1 (t1 , t)) = η(t) = 0 ,

which implies ∇X1 s = 0

We have shown that 3. implies 2. of the following theorem:

Theorem 5.14. For a line bundle L → M with connection ∇ the following properties
are equivalent for each open U ⊂ M :

1. Parallel transport over U is independent of the curves.

2. There exists a horizontal section s ∈ Γ(U, L× ) .

3. [∇X , ∇Y ] − ∇[X,Y ] = 0 for X, Y ∈ V(U ).

Proof. That the first and second conditions are equivalent is the content of 5.13. That
3. implies 2. has been shown just before this proposition. To prove the converse,
i.e. 2. implies 3., let s ∈ Γ(U, L× ) be a horizontal section in a neighbourhood of a. Any
other section U → L has the form f s with f ∈ E(U ). Now, by (K2),

∇X (f s) = (LX f )s + f ∇X s = (LX f )s ,
5.3 Curvature 81

since ∇X s = 0 for a horizontal section. Hence,

[∇X , ∇Y ](f s) = ([LX , LY ]f )s = (L[X,Y ] f )s = ∇[X,Y ] (f s),

which had to be proven.

Definition 5.15 (Curvature). Let π : L → M a line bundle with connection ∇ :

1. The Curvature Operator is, for U ⊂ M open:

F = F∇ : V(U ) × V(U ) −→ End Γ(U, L)

1

(X, Y ) 7−→ [∇X , ∇Y ] − ∇[X,Y ]
2πi

2. The Curvature Form Ω = Curv(L, ∇) ∈ A2 (M ) is defined as follows: If

(Uj )j∈I is an open cover of M with trivializations ψj : LUj → Uj × C and local
connection forms Aj ∈ A1 (Uj ) for ∇ then

Ω|Uj := dAj , j∈I

The last expression is well-defined, since we know from Proposition 4.3 that:

1 dgjk
(Z) Ak = Aj +
2πi gjk

on Ujk = Uj ∩ Uk i.e. dAj = dAk .

F∇ can also interpreted as an operator with values in the endomorphism bundle

End L or even in E(M ) by the first result in the following proposition which also treats
the interrelations between Ω, the global connection form α and sections s ∈ Γ(U, L× :

Proposition 5.16. For a connection ∇ on a line bundle L we have

1. F∇ (X, Y ) = Ω(X, Y ) for X, Y ∈ V(M ), in the sense of: For s ∈ Γ(U, L):
F∇ (X, Y )s = Ω(X, Y )s.

2. π ∗ Ω = dα, where α ∈ A1 (L× ) is the global connection form of ∇ on L× .

3. s∗ dα = Ω|U for any section s ∈ Γ (U, L× ).

Proof. 1. First of all, for each section s ∈ Γ(U, L) there exists a function β(X, Y ) ∈
E(U ) such that F∇ (X, Y )s = β(X, Y )s. It is easy to show that β(X, Y ) is independent
of s (because F∇ (X, Y ) is linear over E(U ) ) and it is bilinear over E(U ) and alternating.
Hence, it is a 2-form. The main point of the statement is, that this form is the curvature
form Ω.
82 5. Parallel Transport and Curvature

This can be checked by showing it over each Uj , i.e. we need to show it only for trivial
bundles with ∇X f s1 = (LX f + 2πiA(X)f ) s1 , s1 (a) = (a, 1) and s = f s1 = (a, f (a)),
where A ∈ A1 (U ) is the local gauge potential:

[∇X , ∇Y ] f s1 = (LX LY f − LY LX f + 2πi (A(X)LY f − A(Y )LX f )

+2πiLX (A(Y )f ) − 2πiLY (A(X)f )) s1


= L[X,Y ] f + 2πi (LX A(Y ) − LY A(X)) f s1


∇[X,Y ] f s1 = L[X,Y ] f + 2πiA([X, Y ])f s1
Therefore
dA(X,Y )

z }| {
[∇X, ∇Y ] − ∇[X,Y ] f s1 = 2πi (LX A(Y ) − LY A(X) − A([X, Y ])) f s1
F∇ (X, Y )s = dA (X, Y ) s = Ω(X, Y )s
⇒ F∇ = Ω.

2. From Section 4.2 we know

 
∗ 1 ∗ dz
α|L× = π Aj + ψ , j ∈ I,
Uj 2πi j z
hence,  
∗ ∗
dα|L× = π dAj = π Ω|Uj , j ∈ I.
Uj

3. The same relation between α and Aj yields for s ∈ Γ (U, L)

   
∗ ∗ ∗ 1 ∗ ∗ dz 1 ∗ dz
s α = s π Aj + s ψj = Aj + (ψj ◦ s)
2πi z 2πi z
on U ∩ Uj and
s∗ dα = dAj = Ω on U ∩ Uj .

5.4 Integrality Condition

In the following we want to show how the parallel transport for a line bundle with
connection (L, ∇) can be expressed by a suitable integral over the curvature form
Ω = Curv(L, ∇).
Let L(a) be the set of all loops (closed smooth curves28 ), γ : [t0 , t1 ] → M , which
start and end in a fixed point a ∈ M : γ(t0 ) = γ(t1 ) = a. Then the parallel transport

Pγt1 ,t0 : La = Lγ(t0 ) → La = Lγ(t1 ) , γ (t0 ) = γ (t1 ) = a,

28
one can allow also continuous curves which are piecewise smooth
5.4 Integrality Condition 83

is an isomorphism of 1-dimensional complex vector spaces. Therefore, it is determined

by a complex number Q(γ) ∈ C× ∼ = GL(1, C):

Pγt1 ,t0 = Pγ : La → La , p 7→ Q(γ) p = p Q(γ) .

Proposition 5.17. Let (L, ∇) be a line bundle with connection and Ω = Curv(L, ∇)
its curvature.

1. Let γ ∈ L(a) be a closed curve being contained in a coordinate neighbourhood U

of a. The parallel transport Pγ : La → La along γ is given by
 Z   Z t1 
Q(γ) = exp −2πi A = exp −2πi A(γ̇(s))ds .
γ t0

with A a local connection form of ∇.

2. Let S ⊂ M be an oriented compact surface in M with boundary ∂S parametrized

by γ ∈ L(a). The parallel transport Pγ : La → La along γ is given by
 Z 
Q(γ) = exp −2πi Ω .
S

Proof. Ad 1.: We can assume LU = U × C. The horizontal lift of γ(t) ∈ M has the
form
γ̂(t) = (γ(t), ζ(t)), t ∈ [t0 , t1 ] ,
with ζ(t) = ζ(t0 )ρ(t) ∈ C, where ρ(t) := ζ(t)ζ(t0 )−1 . By definition: Q(γ) = ρ(t1 ).
According to Proposition 5.4 we have

ζ̇ + 2πiA(γ̇)ζ = 0 , and consequently ρ̇ + 2πiA(γ̇)ρ = 0 ,

where A is a local connection form ∇. Hence, ρ(t) can be expressed as the Integral
 Z t 
ρ(t) = exp −2πi A(γ̇(s))ds .
t0

In particular,  Z t1 
ρ(t1 ) = exp −2πi A(γ̇(s))ds .
t0

Ad. 2: It is enough to show the result locally, i.e. we can assume the line bundle
to be trivial: L = M × C. This is evident if the surface is contained in one of the open
subsets Uj where there is the trivialization ψj : LUj → Uj × C. Otherwise S will be
contained in finitely many of the Uj ’s, since S is compact, and the surface will be cut
into finitely many surfaces with boundary, each of them contained in one Uk 29
29
Exercise! Give a detailed proof of the cutting and gluing.
84 5. Parallel Transport and Curvature

The integral along the full curve is

Z t1 Z Z Z Z
A(γ̇(s))ds = A = A= dA = Ω
t0 γ ∂S S S

by Stokes’ theorem. Therefore,

 Z   Z 
Q(γ) = ρ(t1 ) = exp −2πi A = exp −2πi Ω .
γ S

Instead of restricting to smooth curves one often uses the more general class of
piecewise smooth curves with the same results for lifting to horizontal curves and for
parallel transport.
The last result in the preceding proposition leads to an Integrality Condition
for the curvature Ω = Curv(L, ∇) ∈ A2 (M ) of the line bundle (L, ∇) with connection,
which is of a topological nature and which is of great importance from the point of
view of quantization.
Let us explain this in some detail:
Let Σ ⊂ M be an oriented, compact surface smoothly embedded into M . Assume,
moreover, that Σ is closed, i.e. Σ has empty boundary. Then Σ is a 2-dimensional
oriented and compact submanifold of M (see the example in the illustration below).
We can find a simple closed smooth curve γ dividing the surface Σ into two parts S, S ′
such that S is an oriented compact surface with boundary ∂S parametrized by γ, and
S ′ is another oriented compact surface with boundary ∂S ′ = ∂S parametrized by γ −
(i.e. the curve γ paramatrized in the opposite direction). We have S ∪ S ′ = Σ, and
S ∩ S ′ = ∂S = ∂S ′ = |γ| (as sets without orientation), where |γ| := {γ(t) | t ∈ I} is
the support of γ. For example, as in the following illustration:

Σ
S S′
γ

Closed oriented surface Σ divided into 2 oriented surfaces S, S ′ with boundary |γ|

Let a ∈ ∂S be the initial and end point of γ. And suppose that Σ is contained in
a an open Uj ⊂ Σ with trivialization ψj : LUj → Uj × C. Then the parallel transport
along γ is given by the number
 Z   Z 
Q = exp −2πi Aj = exp −2πi Ω
γ S
5.4 Integrality Condition 85

and the parallel transport along γ − , where γ − (t) := γ (t1 − t + t0 )) , t ∈ [t0 , t1 ] ,, is

given correspondingly by:
 Z   Z 
−
Q = exp −2πi Aj = exp −2πi Ω
γ− S′

The formulas  Z   Z 
−
Q = exp −2πi Ω , Q = exp −2πi Ω
S S′

hold true for general compact and closed oriented Σ in M which are not necessarily
contained in a Uj by cutting Σ into suitable pieces, such that the pieces are in suitable
Uj ’s.
Since Q− is the inverse of Q (because Q− describes the parallel transport in the
opposite direction) we have
 Z   Z 
−
1 = Q Q = exp −2πi Ω exp −2πi Ω
S′
 Z Z   S Z 
= exp −2πi Ω+ Ω = exp −2πi Ω
S S′ Σ

As a consequence, Z
Ω ∈ Z,
Σ

which is the Integrality Condition (”Ganzheitsbedingung”).

Proposition 5.18. Let (L, ∇) be a line bundle with connection. Then the curvature
Ω = Curv(L, ∇) satisfies the following integrality condition:

Z
(G) Ω∈Z
Σ

for every oriented closed compact surface Σ ⊂ M in M .

Another description of the integrality is given by cohomology:

Proposition 5.19. A closed two form Ω ∈ A2 (M ) on a manifold M satisfies the above

integrality condition (G) if and only if the deRham cohomology class30 [Ω] ∈ HdR
2
(M, C)
2 2 2
is in the image of ι : H (M, Z)→H (M, C):

(D) [Ω] ∈ Im ι2

30
see Section E.2 in the Appendix
86 5. Parallel Transport and Curvature

The homomorphism ι2 is induced as part of the long exact sequence coming from
the short exact sequence
ι exp
0 → Z → C −→ C× → 1,
where ι(n) = n ∈ C, n ∈ Z, is the inclusion and exp(z) = e2πiz , z ∈ C. In this
situation H ∗ (M, G) is any of the cohomology theories like singular cohomology or
Čech cohomology, which are equivalent on a paracompact manifold M .
The condition (D) can be understood without general knowledge of cohomology
by using Čech cohomology and its equivalence with deRham cohomology as presented
in Chapter E of the Appendix. In particular, there is a natural isomorphism of the
groups Ȟ 2 (M, C) ∼ 2
= HdR (M, C) and ι2 : Ȟ 2 (M, Z)→Ȟ 2 (M, C) is directly induced by
the inclusion homomorphism ι : Z → C.
With this information the condition (D) for a two form Ω can be reformulated in the
following way within deRham cohomology: Recall (see Section E.1), that a cohomology
2
class in HdR (M, C) is given by a 2-cochain c = (cijk ) with respect to an open covering
(Uj ) of M , where the coefficients cijk are in C: Z = [cijk ]. In particular, a closed two
form Ω defines the following class [Ω] (see Remark E.10) if Uj , Ujk , Uijk are contractible:
Because of dΩ = 0 there are one forms αj on Uj with dαj = Ω|Uj . Since d(αk − αj ) = 0,
there are functions fjk ∈ E(Uj ∩ Uk ) with dfjk = αj − αk . Now, cijk = fij + fjk + fki
is constant on Uijk = Ui ∩ Uj ∩ Uk because of d(fij + fjk + fki ) = 0. The class of Ω is
[Ω)] = [cijk ].

Definition 5.20. The 2-form Ω ∈ A2 (M ) is called Entire if it fulfills condition (E):

(E) The cijk ∈ C can be chosen to be entire, i.e. cijk ∈ Z.

The above discussion amounts to

Proposition 5.21. In more concrete terms, condition (D) for a 2-form Ω is equivalent
to condition (E).

The proof of the equivalence of (G) and (E), a purely topological result attributed
to A. Weil, will not be discussed here. But we come back to the integrality condition
and prove parts of the equivalence in Chapter 8 on Integrality.

Summary: This chapter’s study of parallel transport and curvature of connections

∇ on line bundles culminates in the integrality condition which is the basis of the
concepts of a Quantizable Manifold and Prequantum Line Bundle in Chapter
7. The impact of the integrality condition and the question of existence and uniqueness
of prequantum bundles will be investigated in Chapter 8.
 87

6 Hermitian and Holomorphic Line Bundles

The Hilbert space we eventually want to determine for our program of Geometric
Quantization will be a generated by a subspace of the space of sections Γ(M, L) of a line
bundle. In order to obtain on such a subspace a Hermitian scalar product we need the
notion of a Hermitian structure on the line bundle in question which we now introduce
and study in the first part of the chapter. In the second part we investigate another
additional structure on a line bundle, the holomorphic structure, and we consider the
question how the connection, the Hermitian structure and the holomorphic structure
fit together. Of course, the notion of a holomorphic structure on a complex line bundle
gives only sense if the base manifold is a complex manifold, i.e. has a holomorphic
structure.

6.1 Hermitian Line Bundles

As before, M is a smooth manifold and π : L → M denotes a complex line bundle over

M.

Definition 6.1. A Hermitian Line Bundle is a line bundle π : L → M , for which

every fibre La ∈ M has a Hermitian metric or Hermitian form depending smoothly on
the points p ∈ L. The Hermitian metric will be given by a map
[
H: La × La → C31
a∈M

such that H|La ×La : La × La → C is a Hermitian scalar product32 for all a ∈ M . And
the smoothness condition simply means that H is smooth. We denote:

H(p, p′ ) =: ⟨p, p′ ⟩, p, p′ ∈ L.

Remark 6.2. Such a Hermitian metric H on L will induce a smooth function h :

Lt imes → R+ by h(p) := H(p, p) , p ∈ L× with h(λp) = |λ|2 h(p) for all λ ∈ C×
and p ∈ L×. Conversely, h with these properties defines a Hermitian metric by:
H(zpa , wpa ) := z̄wh(pa ) , where z, w ∈ C and pa ∈ L×
a.

Example 6.3. ln case of the trivial line bundle L = M × C we obtain a natural

Hermitian metric H0 by defining

H0 ((a, z), (a, w)) = ⟨(a, z), (a, w)⟩ := z̄w ,

z, w ∈ C, a ∈ M . H0 is called the Constant Hermitian Metric.

31
which is the same as a map H : L ×M L → C on the fibre product L ×M L, see A.8 for the
definition of a fibre product
32
The properties of a Hermitian metric are recalled in Appendix F, (89) ff.
88 6. Hermitian and Holomorphic Line Bundles

We see that any other Hermitian metric on L = M × C is given by a smooth

h : M → R, h(a) > 0, for all a ∈ M by H (p, p′ ) := h(a) ⟨p, p′ ⟩ = h(a)H0 (p, p′ ), i.e.

H ((a, z), (a, w)) = h(a)z̄w.

Lemma 6.4. A Hermitian line bundle (L, H), whose underlying line bundle is the
trivial line bundle M ×C is isomorphic to the trivial line bundle with constant Hermitian
metric H0 .

Proof. The general case is of the form

H((a, z), (a, w)) = h(a)z̄w,

p
and Φ : M × C → M × C, (a, z) 7→ (a, 2 h(a)z), defines an isomorphism of Hermitian
line bundles (M × L, H), (M × L, H0 ) : Φ is an isomorphism of line bundles with

H (p, p′ ) = H0 (Φ(p)), Φ (p′ )) , p, p′ ∈ M × L.

Using the local existence of a Hermitian metric, we can conclude that on every line
bundle over a paracompact manifold there exists a Hermitian metric H such that it
turns our line bundle into a Hermitian line bundle. This can be proven in the same
way as the proof of existence of connection, c.f. Proposition D.19.
To a Hermitian line bundle (L, H) one associates the Circle Bundle L1 → M ,
where
L1 := {p ∈ L : H(p, p) = 1}
This is a principal fibre bundle with the circle group U(1) ∼ = S1 = {z ∈ C : |z| = 1}
as its structure group. Conversely, if P → M is a principal fibre bundle with structure
group S1 and ρ : S1 → C× = GL(1, C) is the natural representation ρ(z) = z : C → C,
w 7→ zw, then the associated vector bundle (see Section D.5 for the concept of associated
bundle) L = P ×ρ C is a line bundle, where S1 acts on the fibers of P × C by scalar
multiplication. The Hermitian metric H on L is then given by:

H([x, z], [y, w]) := z̄w

where x, y ∈ P, z, w ∈ C.
Proposition 6.5. The group of isomorphism classes of Hermitian line bundles (L, H)
over M is isomorphic to Ȟ 1 (M, U(1)).

Here, Ȟ 1 (M, U(1)) is Čech cohomology with respect to the U(1)-valued locally
constant functions on M (see Chapter E).
We now study line bundles on which there exists a connection together with a
Hermitian metric.
6.1 Hermitian Line Bundles 89

Definition 6.6. Given a Hermitian line bundle (L, H) over M , a connection ∇ on

L is called compatible with H if for all sections s, t ∈ Γ(U, L) and all vector fields
X ∈ V(U ), U ⊂ M open, we have

LX ⟨s, t⟩ = ⟨∇X s, t⟩ + ⟨s, ∇X t⟩

(recall ⟨s, t⟩ = H(s, t)). Such a connection is also called a Hermitian Connection.

The notion of a Hermitian connection is similar to the notion of a Levi-Civita

connection in Riemann Geometry: A connection ∇ on the tangent bundle of a semi-
Riemannian manifold M with Riemannian metric g is a Levi-Civita connection if it
respects the metric g:

LZ g(X, Y ) = g(∇Z X, Y ) + g(X, ∇Z Y ) ,

X, Y, Z ∈ V(M ).
Proposition 6.7. A connection ∇ on L is compatible with a Hermitian metric H if
and only if the local gauge potentials (A
Sj )j∈I with respect to local trivializations ψj :
LUj → Uj × C , (where the Uj cover M : j Uj = M ) can be chosen to be real one-forms
Aj ∈ A1 (Uj , R).

Proof. A collection of trivializations ψj : LUj → Uj ×C can be chosen in such a way that

ψj is an isomorphism (LUj , H|Uj ) → (Uj ×C, H0 ) of Hermitian line bundles with respect
to the constant Hermitian metric H0 on Uj × C, see Lemma 6.4. Now, s, t ∈ Γ(Uj , L)
have the form
s = f sj , t = gsj , g, f ∈ E(Uj ) ,
where sj (a) = ψj−1 (a, 1). Hence, we have for geneeral local sections s, t at a ∈ Uj :
⟨s, t⟩(a) = ⟨(a, f (a)), (a, g(a))⟩0 = f¯(a)g(a) , It follows:

LX ⟨s, t⟩ = LX f¯ g + f¯LX g.

⟨∇X s, t⟩ = ⟨(a, LX f (a) + 2πiAj (X)f (a)) , (a, g(a))

= LX f − 2πiĀj (X̄)f¯(a) g(a)

⟨s, ∇X t⟩ = f¯(a) (LX g(a) + 2πiAj (X)g(a)) .

Compatibility is therefore equivalent to:

LX f¯ g + f¯LX g = LX f g + f¯LX g − 2πi Āj (X) − Aj (X) f¯g .

for all f, g ∈ E(Uj ). If we restrict this equation to real vector fields X and evaluate it
for all f, g the equation amounts to:

0 = Āj (X) − Aj (X).

Hence, Aj is a real form. The converse can be read off the above formulas.
90 6. Hermitian and Holomorphic Line Bundles

Another characterization is the following:

Proposition 6.8. A connection ∇ on L is compatible with a Hermitian metric H if

and only if for all non-zero sections s ∈ Γ(U, L× ), U ⊂ M open, we have:
 
∇s
d(H(s, s)) = 2H(s, s)Re ,
s
∇s
where s
denotes the one-form β ∈ A1 (U, C) given by

∇X s = β(X)s , X ∈ V(U ).

The compatibility is therefore also equivalent to:

∇s
Corollary 6.9. If s ∈ Γ(U, L× ) is of length 1, i.e. H(s, s) = 1, then s
is purely
imaginary.

Concerning the existence of compatible connections, we conclude:

Corollary 6.10. Let (L, H) be a Hermitian line bundle. Then there exists a compat-
ible connection ∇. Given such a compatible connection ∇, the set of all connections
compatible with H is the affine space

∇ + A1 (M, R).

Proof. In the existence discussion in Proposition D.19, one only has to make sure to
choose the one-forms on the trivializations LUj → Uj × C as real one-forms. The
second statement follows from the description of all connections on L as the affine
space ∇ + A1 (M, C) (c.f. Proposition D.20).

Remark 6.11.
1. A given connection ∇ on L, on the other hand, may not be compatible to any
Hermitian metric H on L33 .
2. The curvature Ω = Curv(∇, L) of a connection on L compatible with a given
Hermitian metric is always a real two-form Ω ∈ A2 (M, R).

6.2 Holomorphic Case

In this section M is supposed to be a complex manifold. For a complex line bundle

π : L → M the following additional structures will be considered:

1. connections ∇ on L,
33
Exercise: Give an example!
6.2 Holomorphic Case 91

2. Hermitian metric H on L,

3. holomorphic structure on L.

For holomorphic functions and complex manifolds, in particular for holomorphic

vector bundles, we refer to Chapter B. Recall, that a complex line bundle π : L → M
over a complex manifold M is a Holomorphic Line Bundle if L is a complex
manifold, π : L → M is a holomorphic map and there exists an open cover (Uj )j∈I of
M with trivializations
ψj : LUj → Uj × C,
which are holomorphic maps.
Similar to our results on general complex line bundles over a manifold, the holo-
morphic line bundles are given by transition functions (gjk )j,k∈I , but now they are
holomorphic functions:

gjk : Ujk → C× , or gjk ∈ O× (Ujk ).

The group of isomorphism classes of holomorphic line bundles over the complex mani-
fold is H 1 (M, O× ).
A section s : U → L of a holomorphic line bundle L over an open subset U ⊂ M
is called a Holomorphic Section if s is a holomorphic map. Γhol (U, L) ⊂ Γ(U, L)
denotes the subspace of holomorphic sections

Γhol (U, L) := {s ∈ Γ(U, L) | s holomorphic}.

Γhol (U, L) is a complex vector space and a module over the ring O(U ) of holomorphic
functions on U ⊂ M .
Definition 6.12. Let π : L → M be a holomorphic line bundle with connection ∇ on
L.
1◦ ∇ is a Holomorphic Connection if for any holomorphic section s ∈
Γhol (U, L) the map

X 7→ ∇X s, X ∈ V(U ) with X holomorphic,

is a holomorphic one-form (with values in Γ(U, L)), i.e. in local holomorphic coordinates
φ = (z 1 , . . . z n ) : U → V ⊂ Cn
∇s = fj dz j s
with holomorphic fj : U → C. Hence, ∇X s = fj X j s for holomorphic vector fields
X = X j ∂z∂ j .
2◦ ∇ is Compatible with the holomorphic structure on L if for any local holo-
morphic section s ∈ Γhol (U, L), the one-form

X 7→ ∇X s
92 6. Hermitian and Holomorphic Line Bundles

is of pure type (1, 0), i.e. in local holomorphic coordinates

∇s = fj dz j s,

where fj ∈ E(U ).
Proposition 6.13. Every holomorphic line bundle L over the complex manifold M
admits a connection ∇ compatible with the holomorphic structure on the line bundle L.
Proposition 6.14. Let L be a holomorphic line bundle and let ∇ be a connection
on L, which is compatible with the holomorphic structure on L. Then the curvature
Curv(∇, L) = Ω has components only of type (2, 0) and (1, 1), i.e. in local coordinates
φ = (z 1 , . . . , z n ) : U → V ⊂ Cn :

Ω = ωjk dz j ∧ dz k + ρjk dz j ∧ dz̄ k , ωjk , ρjk ∈ E(U ).

Without proof state we the following result:

Proposition 6.15. Let L → M be a holomorphic line bundle, which is also endowed
with a Hermitian metric H. Then there exists a unique connection ∇ , which is compat-
ible both with the holomorphic structure and the Hermitian structure. This connection
is called the Chern connection of the Hermitian holomorphic line bundle. The
curvature is of type (1, 1).
Example 6.16. The tautological line bundle

T = H(−1) −→ Pn (C)

is a holomorphic line bundle (c.f. Construction 3.17 for the definition). Given an
open subset U ⊂ Pn (C) a local section s ∈ Γhol (U, T ) yields for every a ∈ U a point
s(a) ∈ T ⊂ Pn × Cn+1 , s(a) = (a, z0 (a), . . . , zn (a)) ∈ Pn × Cn+1 , and the value s(a)
describes the point a ∈ Pn in homogeneous coordinates: a = (z0 (a) : . . . : zn (a)). A
natural connection is defined by:

z̄j dz j
∇s := P s,
|z j |2

where zj = z j for notational reasons.

∇ is compatible with the holomorphic structure of the line bundle (but it is
not a holomorphic connection!): Indeed, for a local holomorphic section s(a) =
(a, z0 (a), . . . , zn (a)) , a ∈ U , the connection is
X z̄j j
X
∇s := dz s = fj dz j s ,
k |2
P
j k |z j

fj dz j ∈ A(1,0) (U ) as required.
P
with
6.2 Holomorphic Case 93

∇ is also compatible with the Hermitian metric

n
X
H(s, s′ ) = z̄ j z ′j .
j=0

H is induced from the standard Hermitian metric on Cn+1 . To check the compatibility:

dH(s, s) = zj dz̄ j + z̄j dz j = 2(pj dq j + q j dpj )

and
Re z̄j dz j = pj dq j + q j dpj

lead to
z̄j dz j
 
dH(s, s) = 2H(s, s)Re P j2 ,
|z |
which is the criterium of Proposition 6.8.

Proposition 6.17. Let L → M be a smooth complex line bundle over the complex
manifold M equipped with a connection whose curvature is purely of type (1, 1). Then
there exists a unique holomorphic structure on L for which a local section s of L is
holomorphic if and only if X 7→ ∇X s is a one-form of type (1, 0).

Proofs can be found in Brylinski [Bry93], for example.

94 7. Prequantization

7 Prequantization

In the preceding chapters we have collected all the ingredients which we need in order
to continue the discussion started in Section 2 about obtaining a canonical quantization
by using derivatives on functions or covariant derivatives on sections of line bundles.
We now can describe the process of prequantization properly.

7.1 Quantizable Phase Space

Definition 7.1. A symplectic manifold (M, ω) is said to be Quantizable if there

exists a complex line bundle π : L → M with connection ∇ such that Curv(L, ∇) = ω.
Definition 7.2. A Prequantum Line Bundle (L, ∇, H) on a symplectic manifold
(M, ω) is a Hermitian line bundle (L, H) together with a compatible connection ∇ such
that Curv(L, ∇) = ω.

Evidently, when (L, H, ∇) is a prequantum bundle, the base space (M, ω) has to
be quantizable. Conversely, on a quantizable symplectic manifold there always exist
prequantum bundles: since the connection ∇ with Curv(L, ∇) = ω can be chosen to
be real, we can find with the help of a partition of unity a Hermitian metric H such
that ∇ is compatible with respect to H.
We have seen that for a symplectic manifold (M, ω) the condition to be quantizable
is a topological condition on M and ω: The cohomology class induced by ω has to be
an entire cohomology class (condition (E)) or, equivalently, ω has to satisfy integrality
condition (G) (see Section 5.4), i.e.:
Z
ω∈Z
S

for all compact, oriented and closed surfaces S ⊂ M .

We come back to these conditions in the subsequent chapter. In particular, we will
construct a line bundle L with connection ∇ such that Curv(L, ∇) = ω when ω fulfills
(E). And we discuss the uniqueness of this construction.
Before we come to these matters, in this chapter we want to present examples and
we describe the prequantization process.
Example 7.3 (Simple Phase Space). Let M = T ∗ Q be a cotangent bundle for an
open subset Q ⊂ Rn with the standard symplectic form ω = −dλ = dq j ∧ dpj , and its
symplectic potential A = −λ = −pj dq j . The trivial line bundle L = M × C with the
connection
∇X f s1 = (LX f + 2πiA(X)f )s1 ,
s1 (a) = (a, 1), has as its curvature
dA = ω
7.1 Quantizable Phase Space 95

Since for every compact, oriented and closed surface S ⊂ T ∗ Q one has
Z Z
ω= A=0
S ∂S

by Stokes’ theorem (∂S = ∅), the symplectic manifold (T ∗ Q, ω) is quantizable.

Remark 7.4. In the same way, any symplectic manifold (M, ω) for which ω is exact,
i.e. ω = dα , for a suitable α ∈ A1 (M ), is quantizable.
Proposition 7.5 (Twisted Case). Let M = T ∗ Q be a cotangent bundle for a manifold
Q with the twisted symplectic form

ωF := ω0 + τ ∗ (F ) ,

where ω0 = −dλ is the standard symplectic form on the momentum phase space M =
T ∗ Q and F ∈ A2 (Q) is a closed 2-form on the configuration space Q (see Subsection
1.3.4). In general, F will not be exact and so ωF will not beR exact. But the following
result holds true: (T ∗ Q, ωF ) is quantizable if and only if S F = 0 for all surfaces
S ⊂ Q, i.e. F satisfies the integrality condition on Q.

Proof. If T ∗ Q is quantizable, the integrality condition

R
ω = 0 implies that, in par-
S RF
∗
R
ticular, for all surfaces S ⊂ Q ⊂ T Q we have S ωF = S F = 0. Conversely, if Q
satisfies the integrality condition with respect to F there exists a line bundle LQ on
Q with connection ∇Q such that Curv(LQ , ∇Q ) = F . On T ∗ Q we have the trivial
line bundle L with connection ∇ with Curv(L, ∇) = ω0 locally. Now; L ⊗ τ ∗ LQ with
∇ ⊗ τ + ∇Q is a prequantum bundle since

Curv(L ⊗ τ ∗ LQ , ∇ ⊗ τ ∗ ∇Q = Curv(L, ∇) + Curv(τ ∗ LQ , τ ∗ ∇Q ) = ω0 + τ ∗ F .

Example 7.6. Let M be the two sphere M = S2 of radius 1 with the symplectic form
ωC = Cvol , for some constant C ∈ R \ {0}, whereR vol is the standard volume form on
S2 (= sin θdθ∧dϕ in polar coordinates). Since S2 ωC = Cvol(S2 ) = C4π the symplectic
manifold (S2 , ω) is quantizable in the sense of Definition 7.1 if and only if

4πC = Z \ {0},
1
i.e. C = 4π
N, N ∈ Z \ {0}.
Example 7.7. The hydrogen atom. We recall from Example 1.3.3:
∗ 3 ∼ 3 3
P jM = T (R \ 0) = (R \ {0}) × R
The classical system is given by the manifold
with standard symplectic form ω = d(−λ) = j q ∧ pj and hamiltonian

1 k
H(q, p) = |p|2 − ,
2m |q|
96 7. Prequantization

where m, k ∈ R, m > 0, k > 0. Since dH ̸= 0, the ”energy surface” for E ∈ (−∞, 0)

ΣE := H −1 (E) ⊂ M
is a 5-dimensional submanifold of M (hypersurface in M ).
Identifying points x, y ∈ ΣE on a joint orbit leads to the orbit space
ME := ΣE / ∼
as a quotient. Here, the equivalence relation ∼ is given by
x ∼ y ⇐⇒ ∃ solution of γ̇ = XH (γ) with γ(0) = x and γ(t) = y
for a suitable t in the domain of definition of γ.
The orbit space ME has the structure of a quotient manifold for which the natural
quotient map π : Σ → ME is a submersion. In order to show this, we consider the map
Ψ : ΣE → S2 (mk) × S2 (mk)
into the product of two spheres of radius r = mk given by
Ψ(a) = (ρI(a) + R(a), ρI(a) − R(a))
with a = (q, p) ∈ ΣE and where
I(a) = q × p angular momentum
R(a) = I(a) × p Runge-Lenz vector
√
ρ = −2mE
The map Ψ is constant on the orbits, since the observables I and R are constants of
motion. Moreover, the fibres Ψ−1 (s), s ∈ Smk := S2 (mk) × S2 (mk), are the orbits in
ΣE . Hence, there is a unique bijection Φ : ME → Smk such that Ψ = Φ ◦ π: The
following diagram is commutative.

ΣE
Ψ / S
< mk
π
 Φ
ME

Of course, Ψ is smooth, and it can be shown that Ψ has maximal rank in all points
of ΣE . Hence, Ψ → Smk is a differential quotient according to Proposition A.29. As a
consequence, the differentiable structure on ME induced by the bijection Φ : ME → Smk
makes π : Σ → ME to a differentiable quotient.
Quantizing the classical hydrogen atom with energy E is therefore equivalent to
quantizing the symplectic manifold
(S2 (mk) × S2 (mk), ωE′ )
7.2 Prequantum Operator 97

It can be shown that ωE′ , induced on Smk from ωE := ω|ΣE , has the form
 
′ 1 dx1 ∧ dx2 dy1 ∧ dy2
ωE = + , x3 ̸= 0 ̸= y3 ,
2ρ x3 y3

where (x1 , x2 , x3 ) : S2 (mk) → R3 are the standard coordinates of R3 and similarly

(y1 , y2 , y3 ) for the second sphere S2 (mk).
Because of
dx1 ∧ dx2
Z
r = 4π(r)2
S2 (r) x3
we obtain for S := S2 (mk) × {y} ⊂ S2 (mk) × S2 (mk) the quantization condition
dx1 ∧ dx2
Z Z
′ 1 4π
ωE = = mk = N ∈ Z !
S 2ρ S x3 2ρ
2
(mk)
As a consequence, (M, ωE ) is quantizable only if 2π mk
ρ
= N ∈ Z, i.e. if 4π 2 −2mE = N 2.
Hence, −2mE = N12 4π 2 m2 k 2 and we conclude that only for the energy values

2π 2
EN = − 2
mk 2
N

the symplectic manifolds (S2 (mk) × S2 (mk), ωE′ ) are quantizable, i.e. (MEN , ωE′ N ) is
quantizable. We know this from experimental physics, this is the well known Balmer
Series!

7.2 Prequantum Operator

We are now in the position to define the prequantum operator of geometric quantization
on the space of sections of a prequantum line bundle, thus generalising the preliminary
ansatz in Section 2.2.

Theorem 7.8. Let (L, ∇, H) be a prequantum line bundle over the symplectic manifold
(M, ω). Then the C-linear map q : E(M, C) → End C (Γ(M, L))

i
q(F ) := − ∇XF + F
2π

satisfies the Dirac conditions:

(D1) q(1) = idΓ(M,L) ,

i
(D2) [q(F ), q(G)] = 2π
q ({F, G}) for all F, G ∈ E(M, C).
98 7. Prequantization

q(F ) is called the prequantum operator.

Proof. Evidently, q is C-linear, and q(1) = id. To show (D2) we start with the quan-
tizibilty condition of the symplectic manifold (M, ω):

1

[∇X, ∇Y ] − ∇[X,Y ] = Curv(∇, L)(X, Y ) = ω(X, Y ).
2πi
Applied to X = XF , Y = YG and using [XF , XG ] = −X{F,G} this reads:

[∇XF , ∇XG ] = 2πi {F, G} − ∇X{F,G} .

Hence,
 
i i
[q(F ), q(G)] = − ∇XF + F, − ∇XG + G
2π 2π
 2
i i i i i
= [∇XF , ∇XG ] − F ∇XG + ∇XG ◦ F + G∇XF − ∇XF ◦ G
2π 2π 2π 2π 2π
 2 
i  i
= −∇X{F,G} + 2πi {F, G} − (LXF G − LXG F )
2π 2π
 
i i i
= − ∇X{F,G} − {F, G} + 2 {F, G}
2π 2π 2π
i
= q ({F, G}) ,
2π
where we have used among others the identities: ∇XF ◦ G = (LXF G) + G∇XF and
LXF G = {G, F }.

Remark 7.9. One might not be content with the factor 2πi in front of q({F, G}) (in
1 h
(D2)), preferring 2πi or ℏi , where ℏ := 2π for some h > 0. This can be achieved by
changing the quantization condition Curv(L, ∇) = ω to a new condition Curv(L, ∇) =
−ω or Curv(L, ∇) = kω for a suitable k ∈ C.
The special outcome in D2 in our case is reflected by the choices (see Subsection
17.5 for sign conventions in the literature):

1. XH defined by iXH ω := dH (and not −dH),

2. {f, g} = ω(Xf , Xg ) (and not −ω(Xf , Xg )),

1



3. Curv(L, ∇) = 2πi [∇X , ∇Y ] − ∇[X,Y ] and not [∇X , ∇Y ] − ∇[X,Y ] ,

4. k = 1.
7.2 Prequantum Operator 99

In general, to obtain an arbitrary c in (D2) we have to choose in our conventions

1.-3. k = 2πi in the quantization condition of Definition 7.1: Curv(L, ∇) = kω.
h
In particular, for c = ℏ
i
= 2πi
the quantization condition reads Curv(L, ∇) = − h1 ω.

Before we discuss how to obtain a suitable representation space, the Hilbert space
on which the prequantum operators act, we want to describe an alternative way to
introduce the prequantum operator which is more in the spirit of the description of a
connection on a line bundle L given in Proposition 4.13 and where the connection is
induced by a Lie derivative on invariant functions on L× .
Let F ∈ E(M ) a classical real observable. The hamiltonian vector field XF ∈ V(M )
has a unique horizontal lift XF♯ = (XF )♯ ∈ Γ(L× , H) ⊂ V(L× ) (cf. Definition 4.12).
Adding the vertical field −YF ◦π ∈ Γ(L× , V ), where Yc is the fundamental field we
obtain
ZF := XF♯ − YF ◦π ,
a vector field on L× , which we call the natural lift of XF .
The natural lift ZF can also be defined by the following two conditions

T π ◦ ZF = XF ◦ π and α(ZF ) = F ◦ π ,

where α is the global connection form.

Proposition 7.10. The prequantum operator q(F ) can be obtained by using the Lie
derivative LZF on E−1 (L× ) in the following way: For s ∈ Γ(M, L) the section q(F )s ∈
Γ(M, L) is given by

i
(q(F )s)♯ = − LZF s♯ .
2π

Moreover, ZF preserves the global connection form α, i.e. LZF α = 0, and any real
vector field Z ∈ Γ(M, T L× ) preserving α is of the form Z = ZF for a suitable F ∈
E(M ).

Proof. We concentrate on the formula: We have LX ♯ s♯ = (∇XF s)♯ according to Propo-

F
sition 4.13 and we know LYF s♯ = −2πiF s♯ by Lemma 4.11. Therefore,

LZF s♯ = LX ♯ s♯ − LYF s♯ = (∇XF s)♯ − LYF s♯ = (∇XF s + 2πiF s)♯ = 2πi(q(F )s)♯ ,
F

which is essentially the formula.

In order to understand the relationship between XF and its natural lift ZF we study
their local flows.
100 7. Prequantization

♯
Lemma 7.11. Let Φt = ΦFt : Mt → M−t be the flow of XF and Φ♯t = ΦFt : L× t → L× −t
the flow of the natural lift ZF of XF . Then

1. π ◦ Φ♯t = Φt ◦ π for all t ∈ R. More explicitly, let Mt be the maximal domain on

which the local flow Φt is defined (see Proposition A.20), and correspondingly, let
L× t be the maximal domain for the local flow Φ♯t , then π −1 (Mt ) = L× t and the
following diagram is commutative

Φ♯t
L× t / L× −t

π π
 
Mt / M−t
Φt

2. XF is complete if and only if ZF is complete.

3. Φ♯t commutes with the action of c.

Proof. For p ∈ L× with π(p) = a the curve φ(t) := Φt (a) is the unique solution of

d
φ = XF (φ) , φ(0) = a ∈ M , (33)
dt

and, correspondingly, the curve φ♯ (t) := Φ♯t (p) is the unique solution of

d ♯
φ = ZF (φ♯ ) , φ♯ (0) = p ∈ L× .
dt
Because of
d d
(π ◦ φ♯ ) = T π( φ♯ ) = T π ◦ ZF (φ♯ ) = XF ◦ π(φ♯ )
dt dt
and π ◦ φ (0) = a we deduce π ◦ φ♯ = φ by the uniqueness of the solutions of (33).
♯

This implies, that the maximal intervals on which the solutions φ and φ♯ are defined,
coincide. Hence, π −1 (Mt ) = L× t for all t ∈ R. Now 1. follows since π ◦Φ♯t (p) = Φt ◦π(p)
is nothing else then π ◦ φ♯ (p) = φ(a).
Moreover, 2. holds, since the maximal intervals coincide.
Finally, 3. is again a consequence of the uniqueness of solutions of integral curves,
here for the equation dtd γ = ZF (γ): Set ψ := Ψc φ♯ = φ♯ c, where φ♯ (t) = Φ♯t (p). Then
ψ(0) = pc and

d d
ψ = T Ψc ( φ♯ ) = T Ψc (ZF (φ♯ )) = ZF (Ψc (φ♯ )) = ZF (ψ) ,
dt dt

and therefore, ψ = Φ♯t (pc) which implies Φ♯t (pc) = Φ♯t (p)c. We have used T Ψc ◦ ZF =
ZF ◦ Ψc .
7.2 Prequantum Operator 101

The last property guarantees that for sections s ∈ Γ(Mt , L) the function s♯ ◦ Φ♯−t :
L× −t → C is invariant, i.e. in E−1 (L× −t ). As a consequence there exists a unique
ρFt (s) ∈ Γ(M−t , L) satisfying
(ρFt (s))♯ = s♯ ◦ Φ♯−t .
It is easy to show

Lemma 7.12. ρFt : Γ(Mt , L) → Γ(M−t , L) is an isomorphism of E(M )-modules. When

XF is complete, all spaces Mt coincide with M and the maps ρFt : Γ(M, L) → Γ(M, L)
satisfy
ρFt+t′ = ρFt ◦ ρFt′ = ρFt′ ◦ ρFt
for all t, t′ ∈ R.

The last equality holds true in the general case for suitable restrictions of the ρt
and if the |t|, |t′ | are small enough.

Proposition 7.13. The infinitesimal generator of (ρFt ) is q(F ):

i d F
q(F )s = ρ (s)|t=0 . (34)
2π dt t

Proof. We use
 ♯
d F d  ♯ ♯

ρ (s) = s ◦ Φ−t
dt t dt
 
♯ d ♯
= ds Φ
dt −t
  
= −ds♯ ZF Φ♯−t

to obtain
 ♯
i d F i i
ρt (s) |t=0 = − ds♯ (ZF ) = − LZF s♯ = (q(F )s)♯ ,
2π dt 2π 2π

where the last equality is the result of Proposition 7.10.

Remark 7.14. The definition of the prequantum operator we gave in Theorem 7.8 is
inspired by the considerations made in Chapter 2 and the properties of connections
on line bundles. The description of q(F ) in Equation (34) can be used as a definition
as well. This definition is more geometric in nature and it can be generalized to the
half-density quantization and the half-form quantization.
102 7. Prequantization

7.3 Prequantum Hilbert space

In quantum models one wants to represent the observables as operators in a Hilbert

space (see the Chapter F on Quantum Mechanics in the Appendix), in the so-called
representation space of the model. It is not difficult in the Geometric Quantization
program to replace the complex vector space Γ(M, L), on which our prequantum op-
erators q(F ) act, with a natural Hilbert space of sections: The symplectic form ω of
a symplectic manifold (M, ω) always induces a volume form (the so-called Liouville
measure)
1 1
vol := (−1) 2 n(n−1) ω n ∈ A2n (M ),
n!
where ω n = ω ∧ ω · · · ∧ ω (n times) and dim M = 2n. We obtain
R the Hilbert space
L (M, vol) by completing the prehilbert space {f ∈ E(M, C) | M |f |2 dvol < ∞}. But
2

we are interested in the corresponding Hilbert space of square integrable sections in

the prequantum bundle.
Let (L, H) be a Hermitian line bundle over a symplectic manifold (M, ω). We define
Z
Hpre := {s ∈ Γ(M, L) | |s|2 dvol < ∞} ⊂ Γ(M, L)
M

the space of square integrable smooth sections where |s(a)|2 = H(s(a), s(a)) , a ∈
M . This space is a subspace of Γ(M ; L) and it is a prehilbert space with respect to
the scalar product Z
⟨s, t⟩ := H(s, t)dvol ,
M

s, t ∈ Hpre . The completion of Hpre with respect to the norm

sZ
∥s∥ := |s|2 dvol
M

is the Hilbert space H = H(M, L) of square integrable sections This Hilbert space will
be called the prequantum Hilbert space.
The prequantum operator q(F ) is defined on the the space Γ0 = Γ0 (M, L) ⊂ H of
sections having compact support in M . And q(F )(Γ0 ) ⊂ Γ0 . Therefore, the prequan-
tum operator q(F ) induces in any case a linear operator in H with domain D(q(F ))
containing Γ0 . Thus, q(F ) is densely defined. One can show, that for real F the pre-
quantum operator q(F ) will be symmetric on Γ0 in the same way as in Proposition 2.4.
Thus:

Proposition 7.15. For real F ∈ E(M, R), the prequantum operator q(F ) is a densely
defined symmetric operator in the prequantum Hilbert space H = H(M, L) of square
integrable sections of L.
7.3 Prequantum Hilbert space 103

However, given an observable F ∈ E(M ) it is, in general, difficult, to decide whether

the operator induced by q(F ) is essentially self-adjoint as an operator of H. The
problem is, to extend the symmetric and densely defined operator q(F ) : Γ0 → Γ0 to a
unique self-adjoint operator on a suitable domain D(q(F )) with Γ0 ⊂ D(q(F )) ⊂ H.
For a the class of observables F with complete hamiltonian vector field XF we have
the following positive result.

Proposition 7.16. Let (M, ω) be a quantizable symplectic manifold and let (L, ∇, H)
be a prequantum bundle. For every F ∈ E(M, R) for which XF ∈ V(M ) is a complete
vector field on M the prequantum operator q(F ) is an essentially self-adjoint operator34
in H(M, L).

Proof. Let ZF ∈ V(L× ) be the natural lift of XF as before. Since XF is complete

by assumption, ZF is complete as well, and the local flows are global flows. As a
consequence ρFt is defined on Γ(M, L) and leads a one parameter group of linear maps.
It is easy to see that ρFt (Γ0 ) ⊂ Γ0 ⊂ H, and that ρFt |Γ0 is bounded as an operator
from Γ0 to H, with a bounded continuation to all of H. This operator, which will
be denoted by ρFt again, turns out to be unitary, so that Ut := −2πρFt defines a one
parameter group of unitary operators (Ut ). According to the Theorem of Stone F.5 the
infinitesimal generator
1
As = i lim (Ut s − s) , s ∈ D(A),
t→0 t

is a self-adjoint operator A : D(A) → H. From the preceding proposition we know

d F i
ρ (s) =− q(F )(s)
dt t t=0 2π

for s ∈ Γ0 ⊂ D(A). Hence, q(F ) has A a self-adjoint extension, and therefore is

essentially self-adjoint.

Observation 7.17. In the case of a compact symplectic manifold all q(F ) are essen-
tially self-adjoint for F ∈ E(M, R).

Example 7.18 (Simple Phase Space). Let us recall the example at the end of Chapter
2, Examples 2.6, i.e. the case M = T ∗ Q of the momentum phase space, Q ⊂ Rn open
(see also Example 7.3). This example explains why we have to introduce and study
polarizationsPlater. On M we have the standard symplectic structure given by the
2-form ω = j dq j ∧ dpj = d(−λ), λ = pj dq j with respect to the canonical coordinates
q j , pj on T ∗ Q ⊂ Rn × Rn . Let L = M × C the trivial line bundle with connection

∇X f s1 = (LX f − 2πiλ(X)f )s1 ,

34
i.e. the closure of q(F ) is a self-adjoint operator, see Section F.2 in the Appendix F for the basics
of self-adjoint operators on a Hilbert space
104 7. Prequantization

f ∈ E(M ) and s1 (a) := (q, 1) as before. Then Curv(L, ∇) = ω. To determine the

1
operators q(F ) = 2πi ∇XF + F for F = pj ,F = q j , we use
∂F ∂ ∂F ∂
XF = k
− k ,
∂pk ∂q ∂q ∂pk
to obtain
∂ ∂
Xq j = − and Xpj = j .
∂pj ∂q
Hence,   
j i j i ∂ ∂
+ qj ,


q q = − ∇Xqj + q = − − + 2πiλ
2π 2π ∂pj ∂pj
i ∂
q qj = + q j =: Qj . Analogously,


2π ∂pj
i ∂
q (pj ) = − =: Pj .
2π ∂q j
As a result:
i j i
q q j , q (pk ) = Qj , Pk = q q j , pk .

   

δk = (35)
2π 2π
We see that the Dirac conditions are confirmed on the space Γ(M, L) ∼ = E(M ) (which
is a consequence of Proposition 7.8), and also on the space

H = H(M, L) = L2 (T ∗ Q) .

In particular, they are satisfied for the unbounded operators Qj and Pk which are self-
adjoint. That is, the canonical commutation relations (CCR) (cf. Definition F.38) are
satisfied.
However, in comparison to the usual canonical quantization in this situation, we
observe that too many variables are involved: the wave function ψ ∈ H should depend
on n variables and not 2n. In a more abstract wording, the representation of the Qj , Pk
is not irreducible. By definition, the representation of the (Lie algebra generated by)
Qj , Pk is called irreducible, if no proper closed subspace H0 of H is invariant under the
action of the Qj , Pk . But H0 := {f ∈ H | f = g ◦ π} is invariant! And this H0 is a
good candidate for a better representation space as we show in the following.
By replacing the Hilbert space H(M, L) by its closed subspace HP (M, L) of all
functions f of the form f = g ◦ π , g : Q → C , for suitable g, i.e. HP (M, L) = H0 , we
arrive at a prequantization with the correct dependencies and moreover
i ∂
Qj := q j and Pj := − (36)
2π ∂q j
This quantization is called the Schrödinger Representation of the Qj , Pk with
the commutation rules
i
Pj , Qk = − δkj .
 
2π
7.3 Prequantum Hilbert space 105

The replacement of H(M, L) by HP (M, L) is one of the possibilities of arriving at

a correct representation space (space of wave functions). This procedure is not limited
to the simple case M = T ∗ Q , Q ⊂ Rn open, but can be generalized to all symplec-
tic manifolds by introducing the notion of a polarization. We study polarizations in
Chapter 9 and present their applications to geometric quantization in the subsequent
chapters.

Summary: The concept of prequantization is completed in this chapter as an impor-

tant step towards Geometric Quantization. It yields results in great generality as well
as in specific examples.
In the examples one sees that the condition for the existence of a prequantum bundle
i.e. the integrality condition allows only discrete values: The underlying symplectic
manifold is quantized. This is familiar from elementary quantum theory.
The general result is, that on the basis of a prequantum bundle on a symplectic
manifold the prequantum operators q(F ) satisfy the Dirac conditions (D1) and (D2)
and a natural Hilbert space is determined on which all prequantum operators are
densely defined and symmetric. In many cases, namely if the hamiltonian vector field
XF is complete, they are even self-adjoint. In particular, the canonical commutations
relations are rediscovered. The question of whether enough prequantum bundles exist
and how many there are leads to a closer inverstigation of the integrality condition in
the next chapter.
106 8. Integrality

8 Integrality

We now study in detail the question of existence and uniqueness of prequantum line
bundles on a given symplectic manifold (M, ω). In particular, we construct a prequan-
tum bundle for a given symplectic manifold (M, ω) when ω is entire in the sense of (E)
and we show that the equivalence classes of such prequantum bundles are in one-to-one
corespondence to Ȟ 1 (M, U(1)). Finally, we bring this classification in connection with
flat line bundles and there classification.

8.1 Existence

Recall, from Chapter 5, Definition 5.20, that condition (E) can be stated as follows:
A closed ω ∈ A2 (M ) is entire (or ω respectively its deRham class [ω] sat-
isfies the condition (E)) if for an open cover U = (Uj )j∈I of M the Čech class
[ω] ∈ Ȟ 2 ((Uj )j∈I , C) ∼
= Ȟ 2 (M, C) ∼ 2
= HdR (M, C) induced by the deRham cohomol-
ogy class of ω contains a cocycle z = (zijk ), i.e. [ω] = [z], with

(E) zijk ∈ Z

for all i, j, k ∈ I with Uijk ̸= ∅.

Proposition 8.1. Let ω be a closed two form ω ∈ A2 (M, C) satisfying the inte-
grality condition (E). Then there exists a line bundle with connection ∇ such that
Curv(L, ∇) = ω.

Proof. As before, we work with an open cover U = (Uj )j∈I of M where all intersections
Uj0 j1 ...jp = Uj0 ∩ Uj1 ∩ · · · ∩ Ujp , j0 , j1 , . . . jp ∈ I, are empty or they are contractible
(e.g. diffeomorphic to convex open subsets of Rn ), so that one can apply the Lemma
of Poincaré repeatedly.
We start the construction with the possible local connection forms αj ∈ A1 (Uj )
without having determined the line bundle yet. Since ω is closed, there exist, in fact,
αj ∈ A1 (Uj ) so that dαj = ω by Poincaré’s Lemma in Uj for each j ∈ I. On Ujk ̸= ∅
the one-forms αk − αj are closed:

dαk − dαj = ω − ω = 0.

Hence, there exist fjk ∈ E(Ujk ) with dfjk = αk − αj by Poincaré’s Lemma. We can
choose fjk so that fjk + fkj = 0 for all j, k ∈ I. Because of

d(fjk + fki + fij ) = 0

8.1 Existence 107

we finally obtain constants cijk ∈ C defined by

cijk := fij + fjk + fki , on Uijk ̸= ∅.

(cijk ) is a cocycle associated with ω, (cijk ) ∈ Ž 2 (U, C), where cijk ∈ C, in general. (cijk )
determines the Čech cohomology class

[(cijk )] ∈ Ȟ 2 (U, C) ∼ 2
= HdR (M, C)

of the two form ω.

Now we need the integrality condition (E) in order to go on with the construction.
(E) implies that there are entire numbers zijk ∈ Z which forms a cocycle (zijk ) such
that (zijk ) is equivalent to the cocycle (cijk ). That is, there exist xij ∈ C forming a
cocycle (xjk ) ∈ Ž 1 (U, C) such that

zijk = cijk + xij + xjk + xki ∈ Z, if Uijk ̸= ∅.

In the case of Ujk ̸= ∅, we now set

gjk := exp(2πifjk + 2πixjk )

and obtain gjk ∈ E(Ujk , C× ). We immediately conclude that

gij gjk gki = exp (2πi(cijk + xij + xjk + xki )) = exp (2πi(zijk )) = 1

on Ujk (here, the integrality condition (E) is essential).

As a result, the smooth functions

gjk : Ujk → C× , j, k ∈ I, Ujk ̸= ∅

satisfy (C) and therefore, define a complex line bundle L over M according to Propo-
sition 3.9.
The forms αj define a connection over each Uj since the condition (Z) is fulfilled
1 dgjk
αk − αj = dfjk = .
2πi gjk

Therefore, the αj are local gauge potentials (local connection forms) of a connection
∇ on L. Because of ω|Uj = dαj the curvature of ∇ is ω: Curv(L, ∇) = ω.
Observation 8.2. Proposition 8.1 is formulated for general non-degenerate closed two
forms with (E).

In the real case, i.e. in case of a real form ω, it is clear that the αj -s can be chosen
to be real valued as well, and hence the constructed connection allows a compatible
Hermitian structure.
108 8. Integrality

Corollary 8.3. Let (M, ω) be a symplectic manifold where ω satisfies (E). Then the
line bundle with connection, which has been constructed in the proof of Proposition 8.1,
has a compatible Hermitian structure H. As a result, according to Proposition 6.7, we
obtain a prequantum line bundle (L, ∇, H).

Remark 8.4. As we mentioned before, in Chapter 5, the integrality conditions (G)

and (E) are equivalent according to A. Weil’s Theorem. We have given a complete
proof for the implication (E) =⇒ (G): In fact, assuming (E) we obtain by Proposition
8.1 a line bundle L over M with connection ∇ such that Curv(L, ∇)R = ω. Now our
previous result in Proposition 5.18 assures that ω satisfies (G), i.e. S ω ∈ Z for all
oriented compact closed surfaces S ⊂ M .
The converse of the implication ”ω satisfies (E)” =⇒ ”there exists a prequantum
bundle with curvature ω” will be discussed at the end of the next section.

8.2 Uniqueness

After the question of existence, we now discuss the uniqueness of line bundles with
connection with given curvature form ω ∈ A2 (M ). In other words:
What freedom do we have in constructing (L, ∇, H)? How many inequivalent pre-
quantum line bundles exist on (M, ω)? Under which conditions is the prequantum
bundle essentially unique?
In order to answer these questions, we look at the construction in the proof of
Proposition 8.1 which establishes the existence of a line bundle with connection (L, ∇)
with Curv(L, ∇) =ω, and check step by step what freedom we have in the choice of
the transition functions gjk and the local connection forms αj . Before we start this
program, we make precise what the equivalence of line bundles with connections should
be.
We will study pairs (L, ∇) of line bundles L with connection ∇.

Definition 8.5. Two line bundles with connection (L, ∇) , (L′ , ∇′ ) will be called
Equivalent (or isomorphic) (denoted by (L, ∇) ∼ (L′ , ∇′ )), if there exists an iso-
morphism F : L → L′ of line bundles such that for all local sections s ∈ Γ(U, L):

F ◦ (∇s) = ∇′ (F ◦ s) .

In form of a commutative diagram:

Γ(U, L)
F◦ / Γ(U, L′ )
∇ ∇′
 
Γ(U, L)
F◦ / Γ(U, L′ )
8.2 Uniqueness 109

Recall (cf. Corollary 3.11), that with respect to local trivializations ψj , ψj′ for L, L′
related to an open cover U = (Uj ) of M , an isomorphism F : L → L′ is given by
functions hj ∈ E × (Uj ) such that they satisfy condition (I), i.e.

′ hj
gjk = gjk ,,
hk
′
for all j, k ∈ I with respect to the respective transition functions gjk , gjk of the line
′
bundles L and L .
Proposition 8.6. Let (L, ∇) , (L′ , ∇′ ) line bundles with connection and assume an
isomorphism F : L → L′ is given by (hj ). Then (L, ∇) ∼ (L′ , ∇′ ) with this isomorphism
F if and only if
1 dhj
αj − αj′ = . (37)
2πi hj

Proof. With respect to the sections sj (a) = (ψj )−1 (a, 1) , and s′j (a) = (ψj′ )−1 (a, 1) , a ∈
Uj , the isomorphism is given by F ◦ sj = hj s′j . Using ∇′ (F sj ) = ∇′ hj s′j = (dhj +
2πiαj′ hj )s′j and F (∇sj ) = F (2πiαj sj ) = 2πiαj hj s′j the equivalence of the pair implies

dhj + 2πiαj′ hj = 2πiαj hj ,

and immediately the result (37). Conversely, if (37) holds then the calculation above
shows that F (∇s) = ∇′ (F (s)) for local section, i.e. the pair is equivalent.
Example 8.7. (Trivial Line Bundle) As an example, let be L = L0 = M × C the
trivial line bundle. The trivial connection ∇0 is ∇0 (f s1 ) = df s1 , where as before,
s1 (a) = (a, 1). Hence ∇0 could also be denoted by d.

1. A general connection ∇ on L0 is of the form ∇(f s1 ) = (df + 2πiαf )s1 , where

α ∈ A1 (M, C). and will be denoted by ∇α .

2. A general connection ∇ with Curv(L0 , ∇) = 0 is of the form ∇α with dα = 0.

3. A general isomorphism F : L0 → L0 is of the form F (p) = h(a)p for π(p) = a

where h(a) := pr2 (F (s1 (a))) , a ∈ M . This can be confirmed using the general
h
description of ismorphisms through (hj ) with the condition (I), i.e. gjk = gjk hkj ,
which implies that hj = hk glue together to global functions h.

4. For an arbitrary function g ∈ E(M ) the one form α = dg is closed, hence the corre-
sponding connection yields a line bundle (L0 , ∇) with curvature Curv(L0 , ∇α ) =
0, and all these zero curvature bundles (i.e. flat line bundles in the language of the
next section) are equivalent to the trival bundle (L0 , d) with trivial connection.
The equivalence is given by h := exp(−2πig) as follows from
1 dh
0 − α = −dg = .
2πi h
110 8. Integrality

5. The remaining closed one forms α ∈ A1 (M, C) defining a connection ∇α with

curvature Curv(L0 , ∇α ) = 0 are the closed one forms which are not exact. This
situation will be studied in detail in the next section.
We see, that it is possible that line bundles L and L′ are equivalent as line bundles,
but not as line bundles with connection.

We now come the the uniqueness result:

Theorem 8.8. Let ω ∈ A2 (M ) satisfy the integrality condition (E). Then the set of
equivalence classes of line bundles with connection (L, ∇) such that Curv(L, ∇) = ω is
in one-to-one correspondence with Ȟ 1 (M, U(1)), the first Čech cohomology group with
values in the circle group U(1) ∼
= S1 .

Proof. As mentioned before, we follow the construction of the possible pair (L, ∇) with
Curv(L, ∇) = ω presented in the proof of Proposition 8.1. The proof will be carried
through in 3 steps. In the first step the possible changes in the αj -s of the construction
are considered, the second step treats the possible choices of the fjk and the third
and last step is devoted to the possible changes in the xjk which makes the cocycle
cijk induced by ω entire. In the first two variations we remain in the class given by
the original construction, while the third step yields the isomorphism of the group
Ȟ 1 (M, U(1)) with the set of classes of pairs (L, ∇) with Curv(L, ∇) = ω.
1. Step: First of all, in the construction of the proof of Proposition 8.1 we choose,
using dω = 0, a one form αj ∈ A1 (Uj )35 with dαj = ω|Uj . Any other one form
αj′ ∈ A1 (Uj ) with dαj′ = ω|Uj satisfies d(αj − αj′ ) = 0, and there are gj ∈ E(Uj ) with

αj = αj′ + dgj .

The fjk ∈ E(Ujk ) with αk − αj = dfjk in the above construction will be replaced by
′
fjk = fjk + gj − gk ,

and we obtain
′
dfjk = d(fjk + gj − gk ) = αk − αj + dgj − dgk = αk′ − αj′

′
The fjk lead to the same cocycle cijk as before:

cijk = fij + fjk + fki

= (fij + gi − gj ) + (fjk + gj − gk ) + (fki + gk − gi ) .
= fij′ + fjk
′ ′
+ fki .
35
We have again an open cover (Uj ) with all intersections diffeomorphic to contractible open subsets
of Rn
8.2 Uniqueness 111

′
As a result, the corresponding transition functions gjk (instead of gjk in the above
construction) are
′
gjk := exp (2πi (fjk + gj − gk ) + 2πixjk )
where the xjk ∈ R make (cijk ) entire: zijk := cijk + xij + xjk + xki ∈ Z as before. In
′ ′ ′
particular, gjk gki gij = 1 and
′ hj
gjk = gjk ,
hk
where hj = exp (2πigj ) .
As a consequence, the line bundle L′ defined by the cocycle (gjk
′
) is isomorphic to the
line bundle L defined by the cocycle (gjk ) according to condition (I) in Corollary 3.11:
the cocycles are equivalent. (They are also equivalent in the sense of Čech cohomology
and provide a class in Ȟ 1 (U, E × )).
Moreover, if ∇ is the connection on L given by (αj ) and ∇′ is the connection on L′
given by (αj′ ), then (L, ∇) is equivalent to (L′ , ∇′ ) since
1 dhj
αj − αj′ = dgj = .
2πi hj
′
2. Step: Secondly, fixing αj with dαj = ω|Uj , we can replace each fjk by fjk := fjk +bjk
where bjk ∈ C are constants satisfying bjk + bkj = 0. There are no further possibilities
in changing fjk if one wants to follow the above construction. Then, we get a new
cocycle c′ijk = cijk + bij + bjk + bki representing ω and a new cocycle x′jk in order to
′
achieve that zijk = c′ijk + x′ij + x′jk + x′ki becomes entire. With the choice x′jk := xjk − bjk
the cocycle given by c′ijk + x′ij + x′jk + x′ki is indeed entire, since it agrees with zijk . Since
′
exp(2πi(fjk + x′jk )) = exp(2πi(fjk + bjk + xjk − bjk )) = exp(2πi(fjk + xjk)) the possible
′ ′
new line bundle with transition funcions gjk = exp(2πi(fjk + x′ jk)) is the same. So
again we stay in the same equivalence class of line bundles with connection.
3. Step: Finally, we fix αj and fjk . We can replace the constants xjk by xjk + yjk ,
where yjk ∈ R with
yij + yjk + yki ∈ Z
for all i, j, k ∈ I, with Uijk ̸= ∅ and yjk + ykj = 0. To maintain the construction there
are no other possibilities for changing (xjk ).
Let tjk := exp(2πiyjk ) ∈ U(1). In particular, tjj = exp 0 = 1 = tjk tkj . Then
tij tjk tki = exp(2πi(yij + yjk + yki )) = 1 because of yij + yjk + yki ∈ Z, which implies
that the collection t = (tjk ) is a cocycle in Ž 1 (U, U(1)). We replace gjk by
t
gjk := exp (2πi(fjk + xjk + yjk )) = gjk tjk ,
t
in accordance to the construction. (gjk ) defines a line bundle Lt with connection ∇t
given by the same αj . In fact, we have
t
1 dgjk 1 d(gjk tjk ) 1 dgjk
t
= = = dfjk = αk − αj ,
2πi gjk 2πi gjk tjk 2πi gjk
112 8. Integrality

since tjk is constant. Therefore, the αj determine a connection ∇t on Lt .

The cocycle t = (tjk ) induces a cohomology class

[ t ] = [(tjk )] ∈ Ȟ 1 (U, U(1)) = Ȟ 1 (M, U(1)) .

Let t′ = (t′jk ) be another cocycle t′ ∈ Ž 1 (U, U(1)).

′ ′
Claim: (Lt , ∇t ) ∼ (Lt , ∇t ) if and only if t ∼ t′ as Čech cocycles. The latter equivalence
means that there are tj ∈ U(1) with
tj
t′jk = tjk .
tk
′
In fact, for t ∼ t′ the tj induce an isomorphism Lt → Lt since (I) is satisfied.
′
Conversely, if Lt and Lt are equivalent as line bundles with the same connection deter-
t′
mined by the local connection forms αj , there are hj : Uj → U(1) with gjk t
= gjk hj h−1
k ,
which describe the isomorphism and preserve the connection forms αj . Consequently
p
1 dhj
0 = αj − αj = .
2πi hj
Hence, the hj are constant and yield the equivalence of t and t′ as Čech cocycles.

As a first application:

Proposition 8.9. Let M be simply connected and let ω ∈ A2 (M ) satisfy (E). Then
there exists exactly one line bundle L with connection, such that Curv(L, ∇) = ω up to
equivalence.

Here, we use
π1 (M ) = 0 =⇒ Ȟ 1 (M, U(1)) = 0 .

In fact, the abelianization of the fundamental group π1 (M ) is isomorphic to the

homology group H1 (M ): π1 (M )/[π(M ), π1 (M )] ∼= H 1 (M ) = H 1 (M, Z). Ȟ 1 (M, Z) ∼
=
H (M, Z) is naturally isomorphic to the dual of H1 (M, Z). This implies Ȟ 1 (M, U(1)) =
1

0 in case of π1 (M ) = 0. See also Observation 8.23 at the end of the next section.

Corollary 8.10. A simply connected quantizable symplectic manifold (M, ω) has ex-
actly one prequantum line bundle up to equivalence.

Example 8.11. We continue the Example 7.6 of the sphere S2 which is simply con-
1
nected. We have seen in 7.6 that with the form ω := 4π vol the symplectic manifold
2
(S , Cω) is quantizable if and only if C ∈ Z , N = C ̸= 0. With the preceding re-
sult we conclude that for each N ∈ Z , N ̸= 0, there is exactly one prequantum line
bundle (LN , ∇N ) up to isomorphism on (S2 , N ω). We determine the line bundles with
connection (LN , ∇N ) at the end of the next section.
8.3 Flat Line Bundles 113

Remark 8.12 (Chern Class). Let M be a manifold and let U = (Uj )j∈I be an open
cover such that all Uj0 ...jn are contractible. The transition functions (gjk ) of any complex
line bundle L over M with respect to the cover U always exist and define a cohomology
class in
Ȟ 1 (U, E × ) = Ȟ 1 (M, E × ) ,
and an entire cohomology class in Ȟ 2 (U, Z). In fact, let
1
zijk = (log gij + log gjk + log gki ) on Uijk .
2πi
Locally, zijk is well-defined and integer-valued, since e2πizijk = gij gjk gki = 1. Therefore,
(zijk ) defines an element in Ž 2 (U, Z). There is a problem in getting a global definition
of zijk , due to the fact that the logarithm is ambiguous in C × , but the corresponding
Čech cohomology class [(zijk )] in Ȟ 2 (U, Z) = H 2 (M, Z) is well-defined: Any other
′
choices of branches of the logarithms lead to another cocycle zijk ∈ Ž 2 (U, Z) such that
′ ′
zijk = zijk + mij + mjk + mki with entire mjk ∈ Z. Hence, zijk ∼ zijk .

The class c(L) := [(zijk )] ∈ Ȟ 2 (M, Z) is called the Chern Class of the line bundle
L. c(L) is an important invariant of the equivalence class of the line bundle. In case
of a prequantum bundle (L, ∇), H) on a symplectic manifold (M, ω) the Chern class
of L is given by the symplectic form ω: c(L) = [ω].
As an immediate consequence, we obtain:

Assertion 8.13. A symplectic manifold is quantizable if and only if the symplectic

form satisfies the integrality condition (E).

We know this result already by using Weil’s Theorem (i.e. (G) ⇐⇒ (E)), but the
assertion can now be deduced from the above results in the following way without
referring to Weil’s theorem:

1. ”(E) =⇒ prequantum bundle exists” according to Proposition 8.1,

2. The converse ”prequantum bundle exists =⇒ (E)” follows from the last remark.

8.3 Flat Line Bundles

Flat bundles on a manifold are used to present an additional approach to understand

the variety of non equivalent line bundles with connection whose curvature form is
prescribed.
The basic idea comes from the fact that the set of equivalence classes of line bundles
on M is an abelian group where the group multiplication is induced by the multiplica-
tion of the transition functions or, equivalently, is induced by the tensor product L ⊗ L′
as we have seen in Section 3.4 of Chapter 3. For two line bundles L, L′ with transition
114 8. Integrality

′
functions gjk , gjk the tensor product L ⊗ L′ is a line bundle with transition functions
′
gjk gjk .
For connections ∇, ∇′ on the line bundles L, L′ the corresponding tensor product
′
connection ∇L⊗L is given by
′
∇L⊗L (s ⊗ s′ ) = ∇s ⊗ s′ + s ⊗ ∇′ s′ ,

for local sections s of L and s′ of L′ . It is denoted by ∇ + ∇′ for reasons which becomes

clear regarding the result of the next Lemma. For a common open cover (Uj ) of M
which allows local trivializations ψj resp. ψj′ of L resp. L′ , let αj , αj′ be the local
connection forms of the respective connections.
Lemma 8.14. The local connection forms of ∇ + ∇′ on L ⊗ L′ are αj + αj′ .

Proof. Let sj (a) = (ψj )−1 (a, 1) , s′j (a) = (ψ ′ j )−1 (a, 1) the standard non vanishing sec-
tions defined on Uj and let f sj , f ′ s′j local sections. Then

(∇ + ∇′ )(f sj ⊗ f ′ s′j ) = (df + 2πiαj f )sj ⊗ f ′ s′j + f sj ⊗ (df ′ + 2πiαj′ f ′ )s′j

= (df + 2πiαj f )f ′ + f (df ′ + 2πiαj′ f ′ (sj ⊗ s′j )

= d(f f ′ ) + 2πi(αj + αj′ )f f ′ s′′j ,

where s′′j is the corresponding standard section of L ⊗ L′ over Uj .

For a line bundle with connection (L, ∇) with transition functions gjk let ∇∨ be the
−1
”dual” connection on the dual line bundle L∨ associated to the cocycle (gjk ). Similar
to the proof of the preceding lemma one can show for the local connection forms αj of
∇:
Lemma 8.15. The local connection forms of ∇∨ on L∨ are −αj .

As a result, the local connection forms of ∇ + ∇∨ on L ⊗ L′ vanish. In particular,

the curvature Curv(L ⊗ L∨ , ∇ + ∇∨ ) is zero.
Observation. The set of all line bundles with connection on a given manifold form a
group with respect to the composition

((L, ∇), (L′ )∇′ ) 7→ (L ⊗ L′ , ∇ + ∇′ )) .

Definition 8.16. A line bundle with connection (L, ∇) is called Flat if the curvature
vanishes. The connection ∇ is called flat as well.
Lemma 8.17. The curvatures satisfy

Curv (L ⊗ L′ , ∇ + ∇′ )) = Curv (L, ∇) + Curv (L′ , ∇′ )) .

As a consequence, the flat connections form a subgroup of the group of all line bundles
with connection.
8.3 Flat Line Bundles 115

Proposition 8.18. Any flat line bundle with connection (L, ∇) is equivalent to a
suitable line bundle with connection (L′ , ∇0 ), where L′ has constant transition functions
and ∇0 is the trivial connection with local connection forms αj = 0.

Proof. Let αj be the local connection forms of the flat connection ∇. Because of
flatness, dαj = 0 and there exist functions gj ∈ E(Uj ) with dgj = αj . These gj
induce an equivalent line bundle L′ with transition functions gjk
′
:= gjk hj h−1
k where
hj := exp(2πigj ). We have
′
1 dgjk 1 dgjk 1 dhj 1 dhk
= + −
2πi gjk 2πi gjk 2πi hj 2πi hk
(38)
= αk − αj + dgj − dgk
= αk − αj + αj − αk = 0 .
′
Hence gjk is constant and αj′ = 0 defines a flat connection ∇′ on L′ . The isomorphism
′
L → L is given by hj := exp(2πigj ):

′ hj
gjk = gjk , αj − 0 = dgj .
hk

As a consequence of the above lemmata, the tensor product (L ⊗ L∨ , ∇ + ∇∨ ) is

equivalent to the trivial flat bundle with trivial connection: L0 = M × C is isomorphic
to L⊗L∨ and (L0 , d) is the trivial line bundle with trivial connection d with connection
form α = 0. Evidently, (L0 , d) is flat.
Given a closed two form ω ∈ A2 (M, C) satisfying (E) we want to determine the
set Eω of equivalence classes [(L, ∇)] of line bundles with connections (L, ∇) with
Curv(L, ∇) = ω. Eω is an abelian group induced by the tensor product of line bundles
with connection. Given (L, ∇), for an equivalent line bundle with connection (L′ , ∇′ )
the tensor product (L ⊗ L′∨ , ∇ + ∇′∨ ) is equivalent to the trivial flat bundle with
connection ∇ + ∇′∨ determined by the connection forms αj − αj = 0. The connection
is flat.
Since for (L, ∇) with Curv(L, ∇) = ω and a flat line bundle with connection (L′ , ∇′ )
the tensor product (L ⊗ L′ , ∇ + ∇′ ) has the same curvature ω we conclude:

Lemma 8.19. Eω is in one-to-one correspondence with E0 .

Furthermore, E0 is a group with respect to the tensor product as composition. In
fact, E0 is the quotient group of the group of all flat connections (cf. Lemma 8.17).

Therefore, in order to determine Eω it is enough to determine E0 . This is a major

point of investigating flat bundles and their equivalence classes.
116 8. Integrality

Proposition 8.20. The equivalence classes of flat line bundles are in one-to-one cor-
respondence with the Čech classes of Ȟ 1 (U, U(1)):

E0 ∼
= Ȟ 1 (U, U(1)) ∼
= Ȟ 1 (M, U(1)) .

Proof. This result is known already from the preceding section (Theorem 8.8). Here is
a proof using flat line bundles:
Each equivalence class of E0 contains a line bundle L with connection having con-
′
stant transition functions gjk ∈ U(1) (cf. Lemma 8.16). Two such bundles Lg , Lg ,
given by g = (gjk ), g ′ = (gjk′
), are equivalent as flat line bundles, if and only if there are
hj ∈ U(1) with gjk = gjk hj h−1
′
k (according to condition (I)). But this means exactly that
(gjk ) defines a class [(gjk )] in Ȟ 1 (U, U(1)). Furthermore, it follows that the assignment
E0 → Ȟ 1 (U, U(1)) , Lg 7→ [g], is well-defined and bijective.

Example 8.21 (2-Sphere). We continue the Example 8.11 of the sphere S2 and want to
describe the line bundles (LN , ∇N ) on (S2 , N ω) explicitly. For LN one can take the line
bundle H(N ) which we have introduced in Section 3.3 in the context of line bundles over
the Riemann sphere P1 which is diffeomorphic to S2 . The [H(N )] describe the group
of equivalence classes of line bundles on S2 which is isomorphic to Ȟ 1 (S2 , E × ) ∼
= Z. In
the section 3.4 the bundles H(N ) have been described as N -fold tensor products of the
hyperplane bundle H(1) which in turn is the dual of the tautological bundle H(−1)
(which is also the tangent bundle of the sphere S2 ).
In the Example 6.16 we have equipped H(−1) with a connection ∇(−1) compatible
with the Hermitian structure. And from the above we know the following: If this
connection ∇(−1) is given by the local one forms (αj ) with respect to an open cover then
the bundle H(N ) obtains the connection ∇N determined by the one forms (−N αj ).
Hence, (LN , ∇N ) ∼ (H(N ), ∇N ).
With the notation of Example 6.16 the connection ∇N is

z̄j dz j
∇N s = −N P s.
|z j |2

Another elementary example is the cylinder:

Example 8.22 (Cylinder). Let M = T ∗ S1 be the symplectic phase space with the
circle S = S1 = {z ∈ C | |z| = 1} as the configuration space. This is a special case of
the cotangent bundle T ∗ Q with configuration space Q. The cotangent bundle T ∗ S is
trivial (as a real 1-dimensional line bundle): There exists a nowhere vanishing vector
field Z on S generated by the curve γ

γ : R → S , γ(t) = exp 2πit , t ∈ R ,

8.3 Flat Line Bundles 117

Z(a) := γ̇(t0 ) = [γ(t0 + t)]a , when γ(t0 ) = a. The dual one form β on S is defined
by β(Z) = 1 and also nowhere vanishing. Hence, the cotangent bundle is trivial36
and looks like a cylinder S × R. The form β is locally given as dq where q is a local
coordinate of S given by the angle
1
q = γ −1 (t) = log exp(2πit) = t ,
2πi
t in an open interval of length ≤ 1.
On any cotangent bundle there is the natural (Liouville) one form λ (cf. Construc-
tion 1.17) given in local canonical coordinates as λ = pdq. −λ serves for the cotangent
bundle as symplectic potential with symplectic form ω = d(−λ) = dq ∧ dp. Globally:
ω = β ∧ dp. In particular, ω is exact whereas β is not exact.
As for general cotangent bundles, a natural first choice for a prequantum line bundle
is the trivial bundle L0 = L with the connection

∇f s1 = (df − 2πiλf ) s1 ,

and the Hermitian structure H induced from L = M × C.

The inequivalent flat connections we expect according to Proposition 8.20 (because
of E0 ∼= Ȟ 1 (M, U(1)) ∼= U(1)) are given by ∇s1 = +2πi(κ)dqs1 , where κ ∈ R. Al-
though the one form κdq on S1 looks like an exact one form, it is not so, because q is
′
not globally defined, as we mentioned above. As a consequence, the ∇κ , ∇κ (resp. the
one forms κdq, κ′ dq) are pairwise inequivalent for κ, κ′ ∈ ]0, 1], and they are equivalent
for general κ, κ′ ∈ R if and only if κ − κ′ ∈ Z. This describes the result of Proposition
8.20 explicitly:

E0 → Ȟ 1 (T ∗ S1 , U(1)) = U(1) , [(L0 , ∇κ )] 7→ exp(2πiκ) ∈ U(1) .

Therefore we have the connection

∇κ (f s1 ) = (df − 2πiλf + 2πiκdqf ) s1

with local connection form −λ + κdq , κ ∈ R. And we obtain a family of inequivalent

prequantum line bundles (L, ∇κ , H).
The corresponding prequantum operators Pκ assigned to the observable F = p are
i ∂
Pκ := q(p) = − + κ.
2π ∂q
This leads to different quantizations which are not equivalent and describe different
physics as one can read off the respective spectra of the Pκ . As Hilbert space we
36
In general, the tangent bundle of a Lie group is trivial, see Proposition C.21, and hence the
cotangent bundle is trivial as well.
118 8. Integrality

choose L2 (S1 ), thereby reducing the number of variables (choosing the vertical real
polarization in the sense of the notions of the next chapter.)
Then the spectrum of P κ is σ(P κ ) = {n + κ | n ∈ Z}. As a consequence, there
′
exists no unitary operator U : H → H with U ◦ P κ = P κ ◦ U unless κ − κ′ ∈ Z,
Observation 8.23 (Holonomy). The group E0 of equivalence classes of flat line bun-
dles has a physical interpretation as the Moduli Space of Flat Connections mod-
ulo Gauge Transformations. The gauge transformations are the isomorphisms of
flat line bundles respecting the Hermitian structure, they are given by the multiplica-
tion by functions h ∈ E(M, U(1)) as before: L → L , p 7→ h(a)p when p ∈ La , a ∈ M .
This approach leads to the Holonomy Representation of a flat bundle (L, ∇).
We can introduce a Hermitian metric which is compatible with ∇ since the local con-
nection forms are real. Therefore we can introduce the corresponding U(1)-bundle
L1 = S = {p ∈ L | |p| = 1} → M (also being the frame bundle) and the induced
connection there. The associated parallel transport defines, for closed curves γ start-
ing and ending in a fixed point a ∈ M the parallel transport map Q(γ) : L1a → L1a
which can be called the Holonomy (see Section 5.4 for the definition of Q(γ)). Q(γ)
is given as a multiplication with a complex number of norm 1, which we again denote
by Q(γ). Since the curvature is zero, the parallel transport is locally independent of
the respective curves. As a consequence, for homotopic closed curves γ ∼ γ ′ starting
and ending in a ∈ M the holonomies agree: Q(γ) = Q(γ ′ ). This defines the map

Hol∇ : π1 (M ) → U(1) , Hol∇ ([γ]) := Q(γ) ,

which is a group homomorphism. In our Example 8.22 above the local connection
R flat connection ∇ determined by the real parameter κ is κdq. And Q(γ) =
form of the
exp(−2πi γ A) with A = κdq. For γ(t) = exp(2πit) we obtain Q(γ) = exp(−2πiκ)
Now, π1 (T ∗ S) = π1 (S) = {m[γ] | m ∈ Z} ∼
= Z and the connection ∇ given by κ yields
Hol∇ (m[γ]) = exp(−2πiκm).
As a consequence, there is a natural mapping

Hol : E0 → Hom (π1 (M ), U(1)) , ∇ 7→ Hol∇ ,

which turns out to be an injective group homomorphism. Now Hom (π1 (M ), U(1)) ∼ =
Hom (H1 (M ), U(1)) since U(1) is abelian. And Hom (H1 (M ), U(1)) ∼
= Ȟ 1 (M, U(1)).
We are back in the situation where we have worked before: Each class (tjk ) ∈
Ȟ 1 (U, U(1)) defines a line bundle with the transition functions tjk . And the line bundle
with tjk is isomorphic with the line bundle with t′jk if and only if the cocycles tjk and
t′jk are equivalent. This shows that Hol : E0 → Hom (π1 (M ), U(1)) is surjective, hence
an isomorphism and all the groups are isomorphic:

E0 ∼
= Hom (π1 (M ), U(1)) ∼
= Hom (H1 (M ), U(1)) ∼
= Ȟ 1 (M, U(1)) .

Summary:
 119

9 Polarization

In Quantum Mechanics one can represent the Hilbert space of states as the space
of square integrable complex functions on the spectrum of a given complete set of
commuting observables. In the classical situation a natural choice of a ”complete set
of commuting observables” on a given symplectic manifold (M, ω) of dimension 2n is a
set of n = 21 dim M functions F1 , F2 , . . . , Fn ∈ E(M, R), which are independent at each
point of M , commuting in the sense of

{Fi , Fj } = 0, ∀i, j ∈ {1, . . . , n} ,

and such that the corresponding Hamiltonian vector fields XF1 , XF2 , . . . XFn are all
complete. We thus arrive at the notion of a completely integrable system.
In general, however, such classical observables do not exist. As a consequence, one
has to relax the condition of globally defined Fi and one considers instead distributions
which locally describe the above situation with a complete set of commuting variables.
Such distributions have to be integrable, i.e. they are foliations, and they have to be
adapted to the symplectic structure. These requirements lead to the notion of a real
polarization. However, some symplectic manifolds do not admit real polarizations, for
instance the 2-sphere S2 with its natural volume form. Therefore, one generalizes this
concept to complex polarizations, i.e. complex Lagrangian subbundles of the complex-
ification T M C of the tangent bundles which are involutive.
In this chapter we cannot present much more than the formal definitions of the
different types of real and complex polarizations leaving aside the rich geometric theory
of foliations and polarizations. The goal is to prepare the way to the correct Hilbert
representation space by ”cutting down the number of variables” in the direction of
polarizations (which is discussed in Chapter 7), and, moreover, to show that the notion
of a polarization is well adapted to the symplectic structure. In the special case of a
Kähler polarization the interplay of the symplectic structure with the polarization leads
to a complex (holomorphic) structure on the manifold.

9.1 Distribution

Definition 9.1. A Distribution D on a manifold M is a subbundle D of the tangent

bundle T M : D ⊂ T M .

A distribution is therefore a real vector bundle (cf. Appendix Section D.1) D such
that Da is a real vector subspace of the tangent space Ta M at each point. Moreover,
at each point a ∈ M there exist an open neighbourhood U ⊂ M and k smooth vector
fields X1 , . . . , Xk ∈ V(U ) with

Db = spanR {Xj (b) | 1 ≤ j ≤ k}, b ∈ U. (39)

120 9. Polarization

In this description, the natural number k can be chosen to be the rank of the vector
bundle D.
If the distribution D is only given by a collection X of vector fields at each point
by Db := spanR {X(b) | X ∈ X defined on U }, b ∈ U , then in addition one has to
require that dimR Da is constant. Or, equivalently, that the generating vector fields
(Xj ) in (39) are linear independent.
Definition 9.2. A distribution D is called Integrable if for each a ∈ M there exists
a k-dimensional submanifold N in an open neighbourhood U of a so that for each
b ∈ N : Tb N = Db .
Any such submanifold N ⊂ U is called an Integral Manifold of the distribution.

Distributions D ⊂ T M with dimR Da = 1 = rank D are always integrable, since

locally Da = RXa , a ∈ U , for a (local) nowhere vanishing vector field X and the Na
are given by the integral curves of X.
Example 9.3. Let M = S2 be the 2-sphere with the the volume form as symplectic
form ω. There is no 1-dimensional distributions D on S2 . This property can be deduced
from the fact that real line bundles on any manifold M are classified by H 1 (M, Z/2Z),
and H 1 (S2 , Z/2Z) = 0 since S2 is simply connected. Hence, every real line bundle is
trivial. As a consequence, a distribution D ⊂ T S2 would be a trivial subbundle of T S2
which would have a nowhere vanishing section. Therefore T S2 would have a nowhere
vanishing section in contradiction to the ”Hairy Ball Theorem”.

Without proof, we state the fundamental theorem of Frobenius.

Theorem 9.4. A necessary and sufficient condition for a distribution to be integrable
is that the global sections of D form a Lie subalgebra of V(M ):
X, Y ∈ Γ(M, D) =⇒ [X, Y ] ∈ Γ(M, D).

An integrable distribution is also called a Foliation. The maximal connected

integral manifolds, i.e. N ⊂ M with Tb N = Db for all b ∈ N , are called the Leaves of
the foliation.
Let M/D be the space of leaves.
Definition 9.5. A foliation is called Reducible or Admissible if M/D exists as
a quotient manifold (see Section A.4) and the canonical map π : M → M/D is a
submersion.
Remark 9.6. Recall, that the last condition is equivalent to π having maximal rank
at all points of M . Using a result concerning quotient manifolds (cf. Proposition A.29)
we know for concrete instances the following criterium in the case that the quotient
topology on M/D is Hausdorff: If M/D has a differentiable structure with respect to
the quotient topology and if there exists a surjective submersion p : M → M/D with
respect to this differentiable structure, then M/D with this differentiable structure is
the quotient manifold and p is the quotient map (up to isomorphism).
9.1 Distribution 121

Examples 9.7.

1. The vertical distribution: Let Q ⊂ Rn be an open subset of Rn with the stan-

dard coordinates q 1 , q 2 , . . . , q n . The cotangent bundle T ∗ Q ∼ = Q × Rn with
standard (and canonical with respect to ω = dq k ∧ dpk = −dλ) coordinates
q 1 , . . . q n , p1 , . . . , pn has the vertical distribution D spanned by
 
∂ ∂
,..., , D ⊂ T (T ∗ Q).
∂p1 ∂pn
D is integrable and the leaves are given by
{(q, p) ∈ Q × Rn : q = c} = {c} × Rn = τ −1 (c) .
Here, c ∈ Q is constant and τ : T ∗ Q → Q is the canonical projection. Moreover,
T ∗ Q/D ∼
= Q. Thus, D is integrable and reducible.
In this example, the distribution D is generated by the hamiltonian vector fields of
the globally defined functions q 1 , . . . , q n ∈ E(T ∗ Q, R) in involution ({q i , q k } = 0):
Xqk := ∂p∂k and
D = spanE(T ∗ Q,R) {Xqk : k = 1, 2, . . . , n}
This relation shows that this special distribution and the symplectic structure fit
together; in the terms of the following section, D is a polarization. Also, we have
the global definiton of D as described in the beginning of this chapter.
2. The cotangent bundle M = T ∗ Q with respect to a general manifold Q of di-
mension n has again the vertical distribution D given locally by ∂p∂ 1 , . . . , ∂p∂n in
suitable canonical bundle charts (q, p) and leaves τ −1 (c) ∼
= Rn , c ∈ Q. D is inte-
grable and reducible. However, in general, we do not have global F1 , . . . , Fn which
Poisson-commute and whose Hamiltonian vector fields XF1 , . . . , XFn generate D.
3. The horizontal distribution: Let Q ⊂ Rn be an open subset of Rn and let M =
T ∗Q ∼
= Qn× Rn as before o in the first example. The horizontal distribution D is
∂ ∂
given by ∂q1 , . . . , ∂qn . The leaves are

{(q, p) ∈ Q × Rn | p = c} = Q × {c} = pr2−1 (c) , c ∈ Rn

and M/D ∼
= Rn 37 . So, D is integrable and reducible.
4. The radial distribution: M = R2 \ {0} with coordinates (q, p). Let D be the
distribution generated by
∂ ∂
q −p .
∂p ∂q
D is a distribution whose leaves are the circles {p2 + q 2 = r2 } , r ∈ R, r > 0. The
quotient exists as a manifold and satisfies M/D ∼ = R+ . So, D is integrable and
reducible.
37
As a reminder, we assume our manifolds to be connected
122 9. Polarization

Note, that the vertical distribution D on M = R2 \ {0} ⊂ R2 induced by the

projection pr1 : M → R is not reducible. The leaves are the vertical lines through
(0, p), p ̸= 0 and the two rays

{(q, 0) : q > 0} and {(q, 0) : q < 1}.

The quotient M/D is not Hausdorff in the quotient topology.

9.2 Real polarization

Definition 9.8. Let (M, ω) be a symplectic manifold. A real polarization on M

is a foliation (i.e. an integrable distribution) D ⊂ T M on M , which is maximally
isotropic,i.e. for all a ∈ M :

ωa (X, Y ) = 0 for X, Y ∈ Da ,

and no strictly larger subspace of Ta M which contains Da has this property. A real
polarization is called reducible or admissible if it is reducible as a distribution.

This definition is modeled after the notion of a completely integrable system:

Example 9.9 (Completely Integrable System). Let F1 , F2 , . . . , Fn ∈ E(M, R) be n in-
dependent functions on a symplectic manifold (M, ω) of dimension 2n which commute,
i.e. {Fi , Fj } = 0, and whose corresponding Hamiltonian vector fields XF1 , . . . XFn are
all complete. In particular, D := spanR(M,R) {XF1 , . . . XFn } ⊂ T M is a real polariza-
tion.

Conversely, any real polarization looks locally like this example:

Proposition 9.10. Let (M, ω) be a symplectic manifold of dimension 2n. Then a given
smooth distribution D ⊂ T M is a real polarization if and only if for each a ∈ M there
exists an open neighbourhood U of a and n independent smooth functions F1 , . . . , Fn ∈
E(U, R) such that

1. For each a ∈ U , Da = spanR {XF1 (a), . . . , XFn (a)};

2. {Fj , Fk } = 0, i.e. ω|U (XFj , XFk ) = 0 for all j, k ∈ {1, 2, . . . , n}.

Proof. Let D be a real polarization. Since D is in particular an integrable distribution,

there exist locally n independent smooth functions F1 , . . . , Fn ∈ E(U, R) such that the
integral manifolds are of D are locally of the form

{a ∈ U : F1 (a) = c1 , . . . , Fn (a) = cn }

with suitable constants c1 , . . . , cn ∈ R. For each vector field X ∈ Γ(U, D) we have

LX Fi = X(Fi ) = 0 ,
9.2 Real polarization 123

for i = 1, 2, . . . , n. Hence
ω(XFi , X) = dFi (X) = LX (Fi ) = 0.
It follows that XFi ∈ Γ(U, D), since D is maximally isotropic. As a consequence, the
hamiltonian vector fields XF1 , . . . , XFn span D locally. Note, that the F1 , . . . Fn are
independent, i.e. dF1 (a), . . . , dFn (a) are linearly independent for each a ∈ U . By
isotropy, we have ω(XFi , XFj ) = 0, hence
{Fi , Fj } = 0 ,
thus we have shown that the 2 properties are satisfied.
Conversely, by conditions 1. a distribution D is well-defined. The local conditions
”{Fi , Fj } = 0 ⇐⇒ ω(Xi , Xj ) = 0 on U ” imply that D is isotropic. Since the Fi are
independent, dimR Da = n. Finally, again by the independence, the distribution D is
integrable whose integral manifolds on U are the following
{a ∈ U |F1 (a) = c1 , . . . , Fn (a) = cn } .

Concrete examples are the first three out of the four examples in 9.7: They are
reducible real polarizations. Moreover:
Example 9.11. On a two-dimensional symplectic manifold (M, ω) any distribution
of rank 1 is automatically integrable and Lagrangian, hence a real polarization. In
particular, the fourth example in 9.7 of a radial distribution on M = R2 \ {0} is a
reducible real polarization. It is the distribution spanned by the hamiltonian vector
field XH defined by the energy H(q, p) = 12 (p2 + q 2 ) of the harmonic oscillator. Note,
that M is diffeomorphic to S1 × R+ by the diffeomorphism
 
(q, p) p
(q, p) 7→ , r , r := q 2 + p2 .
r
Closely related is the cylinder T ∗ S1 ∼
= S1 × R with the symplectic form ω = dq ∧ dp
as in Example 8.22 and with the horizontal distribution D given by the non vanishing
∂
vector field ∂q . The leaves of D are the circles {(exp 2πit, p) | t ∈ R} , p ∈ R. Thus,
T S can be seen as an extension of M : S1 × R+ ⊂ T ∗ S1 .
∗ 1

Horizontal distributions on T ∗ Q for general Q are quite special from the point of
view of polarizations. In fact, the (connected components of the) leaves have to be of
the form of open subsets of (S1 )k ×Rn−k , due to general results for completely integrable
systems, which does not give much freedom.
In general, real polarizations need not exist on a given symplectic manifold, in
particular, when there does not exist any distribution of rank 12 dimR M , as we have
seen for the 2-sphere M = S2 (see Example 9.3).
For the purpose of geometric quantization we thus need a generalization of the
notion of a real polarization: the complex polarization!
124 9. Polarization

9.3 The Complex Linear Case

Before introducing the definition of a complex polarization on a symplectic manifold,

we first study the linear case, i.e. we consider a 2n-dimensional symplectic vector space
(V, ω) as the prototype of the tangent space Ta M , a ∈ M , of a symplectic manifold
(M, ω). Here, ω : V × V → R is bilinear over R, antisymmetric and non-degenerate.
We have studied this structure in the beginning (see Section 1.1).
Let V C := V ⊗R C ∼= V ⊕iV be the complexification of V with the obvious C-bilinear
extension of ω : V × V → R to V C
ω = ω C : V C × V C → C,
ω(v + iw, v ′ + iw′ ) := ω(v, v ′ ) − ω(w, w′ ) + i(ω(v, w′ ) + ω(w, v ′ )),
and with conjugation
v + iw := v − iw , v, w ∈ V .
Definition 9.12. A complex linear subspace P ⊂ V C is called a Complex La-
grangian Subspace, if P is

• isotropic (i.e. for all z, w ∈ P : ω(z, w) = 0) and

• maximally isotropic, i.e. whenever Q ⊂ V C is a complex isotropic subspace con-
taining P , P ⊂ Q, it follows that P = Q.

Our symplectic form ω defines a sesquilinear and skew-symmetric form on V C by:

1
⟨z, z ′ ⟩P := − iω(z̄, z ′ ) , (40)
2
for z, z ′ ∈ V C . In particular, for v, w ∈ V we have
⟨v + iw, v + iw⟩P := ω(v, w) .

Let P be a complex Lagrangian subspace. Then, in general, the form ⟨ , ⟩P fails to

be non-degenerate on P . The null space of ⟨ , ⟩P |P ×P will be denoted by N :
N = {z ∈ P | ⟨z, w⟩P = 0 ∀ w ∈ P } = P ⊥ .
It is easy to see that N = P ∩ P where P := {z̄ | z ∈ P }. Therefore, the form is
non-degenerate if and only if N = P ∩ P = {0} which is the case if P + P = V C .
In case of N ̸= {0} the form ⟨ , ⟩P projects to a non-degenerate sesquilinear form
on P/N . This form is, in general, not positive definite. Let (r, s) be the signature of
this form. Then its matrix with respect to a suitable basis of P/N is
 

diag 1, 1, . . . , 1, −1, . . . , −1 , 0 ≤ r + s = n − dimC N = dimC P/N . (41)

| {z } | {z }
r s

The complex Lagrangian subspace P is said to be

9.3 The Complex Linear Case 125

• of Type (r, s) if the above signature is (r, s).

• positive in the case of s = 038 .
• real in case of r = s = 0. Then P = DC for a real Lagrangian subspace D ⊂ V .
Then ⟨ , ⟩P = 0 and N = P .
• Kähler39 in case of r + s = n, i.e. N = P ∩ P = {0}. In that case, the form
⟨ , ⟩P is non-degenrate and N = P . In some places the term Kähler is reserved
for complex Lagrangian subspaces which are positive and satisfy N = {0}. Then
the complex Lagrangian subspaces P with merely N = {0} are called pseudo
Kähler or of Kähler type.
Remark 9.13. The choice of sign in 40 is arbitrary, in many places in the literature
1
2
iω(z̄, z) is used as the corresponding form induced by P . As a consequence, with
the alternative choice of sign ”positive” and ”negative” interchanges for the complex
Lagrangian subspaces (and later Lagrangian subbundles). For the results this is not
much of a difference. In particular, what can be proved for positive Lagrangians mostly
can be proved for negative Lagrangian, as well.
The subspaces
D := P ∩ P ∩ V and E := (P + P ) ∩ V
are of special interest in the following. Note, that DC = P ∩ P = N and E C = P + P .
The number k := n−(r +s) = dimC P ∩P = dimR D is sometimes called the number
of ”real directions” in P .
In the following we illustrate the definitions and the different spaces D, E, P, P , N
by an example which uses a special basis of V . A basis (u1 , u2 , . . . , un ; v1 , v2 . . . , vn ) =
(u; v) is called a symplectic frame of the symplectic vector space (V, ω) if the matrix
of the 2-form ω with respect to this basis is the block matrix
 
0 1
J= ,
−1 0
where 1 is the unit n × n-matrix.
Example 9.14. Let (V, ω) be a symplectic vector space with a symplectic frame (u; v).
We define a complex Lagrangian subspace P of V C of type (r, s) with dimC N = k =
n − (r + s) by
D = spanR {u1 , . . . , uk } , DC = P ∩ P = N ,
P = spanC ({u1 , . . . , uk } ∪ {uj + ivj | k < j + r ≤ n} ∪ {uj − ivj | k + r < j ≤ n}) ,
P = spanC ({u1 , . . . , uk } ∪ {uj − ivj | k < j + r ≤ n} ∪ {uj + ivj | k + r < j ≤ n}) ,
E = spanR ({u1 , . . . , un } ∪ {vk+1 , . . . vn | k < j ≤ n}) , E C = P + P ,
C := spanR {v1 , . . . , vk } , with E + C = V C .
38
see the Remark 9.13 below
39
also called purely complex
126 9. Polarization

Note, that ⟨uj + ivj , uj + ivj ⟩P = ω(uj , vj ) and ⟨uj − ivj , uj − ivj ⟩P = −ω(uj , vj ). As a
consequence, the form induced by ⟨ , ⟩P on P/N has the following diagonal matrix as
its matrix with respect to the basis of P/N induced by (uj + ivj | k + 1 ≤ j ≤ n)
 

diag 1, . . . , 1, −1, . . . , −1 , r + s = n − k.

| {z } | {z }
r s

Analogously, on P /N the corresponding matrix is

 

diag −1, −1, . . . , −1, 1, 1, . . . , 1 .

| {z } | {z }
r s

The Kähler case we study in a later section in some detail.

9.4 Complex Polarization

We now come to the definition of a general complex polarization on a symplectic

manifold.

Definition 9.15. Let (M, ω) be a symplectic manifold of dimension 2n. A Complex

Polarization P of (M, ω) is a complex vector subbundle P ⊂ T M C of complex
dimension n such that

1. For all X, Y ∈ Γ(M, P ) we have [X, Y ] ∈ Γ(M, P ) (P is Involutive),

2. Pa ⊂ Ta M C is maximally isotropic for all a ∈ M (P is Lagrangian), i.e. a

complex Lagrangian subspace ,

3. Da := Pa ∩ Pa ∩ Ta M has constant rank k ∈ {0, 1, . . . , n}: dimR Da = k , a ∈ M .

By 3. D = P ∩ P ∩ T M is a vector bundle and hence a distribution. When P is a

complex polarization then P is a complex polarization as well. It follows that D ⊂ T M
is an integrable distribution.
Since we have the constant rank condition in 3. the assertion concerning the non-
degenerate form on Pa /Na (see (41)) leads to a type (r, s) for all a ∈ M . (Here we need
M to be connected which has been assumed in general.)
The complex polarization P is said to be

• of Type (r, s) if Pa /Na is of type (r, s) for all a ∈ M .

• Positive if s = 0 (but see Remark 9.13).

9.4 Complex Polarization 127

• Real if Pa = DaC for at least one a ∈ M and then for all a ∈ M . This is
equivalent to P = P .

• Kähler if Da = {0} for all a ∈ M (sometimes Kähler polarization means D = 0

and positive and the condition D = 0 is called pseudo Kähler).

• Strongly Integrable if the distribution E ⊂ T M (Ea := (Pa + P a ) ∩ Ta M )

is integrable.

• Reducible if the orbit space M/D exists as a differentiable manifold and the
projection M → M/D is a submersion.

Note, that the vertical distribution on T ∗ Rn as well as the horizontal distribution

induce the complex polarizations P := DC , which are real polarizations in this termi-
nology.
A special and well-known example of a complex polarization is the following.

Example 9.16 (Simple Phase Space). Let M = (T ∗ Rn , ω) be the symplectic manifold

with the standard symplectic form ω = dq j ∧ dpj , where q j , pj , 1 ≤ j ≤ n are the
standard canonical coordinates of M . The vector fields
∂ ∂
, j , 1 ≤ j ≤ n,
∂pj ∂q

form a basis for each tangent space Ta M ∼= R2n . They form a basis of the complexified
Ta M C (see below) as well, now over the complex numbers C.
Let zj := pj + iq j complex coordinates, hence we understand M ∼ = Rn × Rn as
n
a complex vector space and as a complex manifold M = C . As a reminder (cf.
Proposition B.23), the following vector fields form a basis of Ta M C as well
 
∂ 1 ∂ ∂
:= −i j ,
∂zj 2 ∂pj ∂q
 
∂ 1 ∂ ∂
:= +i j .
∂ z̄j 2 ∂pj ∂q
The polarization we want to introduce in this example is defined as
 
∂
P := spanE(M ) 1≤j≤n .
∂ z̄j

(In Chapter B on Complex Analysis P is denoted by T (0,1) , cf. B.24.) Using the
identities
∂ ∂
Xzj = −2i , Xz̄j = 2i , 1 ≤ j ≤ n,
∂ z̄j ∂zj
128 9. Polarization

we see that P can be defined as the span of the corresponding hamiltonian vector fields

P := spanE(M ) Xzj | 1 ≤ j ≤ n . (42)

It is easy to show that P ⊂ T M C is involutive: For V, W ∈ Γ(M, P ) there are

w , v j ∈ E(M ) such that
j

∂ ∂
V = vj , W = wj .
∂ z̄j ∂ z̄j
Now for f ∈ E(M )
   
j ∂ k ∂ k ∂ j ∂
[V, W ] f = v w f −w v f
∂ z̄j ∂ z̄k ∂ z̄k ∂ z̄j
k j
 
j ∂w ∂ k ∂v ∂ ∂ ∂ ∂ ∂
=v f −w f since f= f
∂ z̄j ∂ z̄k ∂ z̄k ∂ z̄j ∂ z̄j ∂ z̄k ∂ z̄k ∂ z̄j
!
k k
∂w ∂v ∂
= vj − wj j f.
∂ z̄j ∂ z̄j ∂ z̄k

And !
∂wk ∂v k ∂
vj − wj j ∈P.
∂ z̄j ∂ z̄j ∂ z̄k

Moreover, dzj ∧ dz̄j = d(pj + iq j ) ∧ d(pj − iq j ) = idq j ∧ dpj − idpj ∧ dq j = 2idq j ∧ dpj .
Therefore, our standard symplectic form ω = dq j ∧ dpj can be written as:
n n
1 X 1 X
ω= dzj ∧ dz̄j = − dz̄j ∧ dzj .
2i j=1 2i j=1

Using this form of ω we can immediately conclude that

 
∂ ∂
ω , = 0,
∂ z̄k ∂ z̄l
i.e. P is isotropic (which can be deduced from (42) as well).
P is maximally isotropic (Lagrangian) because of dimC P = n.
Since, furthermore
 
∂
P a := spanC :1≤j≤n ,
∂zj a

it follows that
Pa ∩ P a = {0}, Pa ⊕ P a = Ta M C .
As a result, P is a complex polarization. Because of the last identity P ∩P a = D = {0},
P is a Kähler polarization. Moreover P is positive.
9.4 Complex Polarization 129

Note, that P defines a Kähler polarization, too. But this Kähler polarization is
negative (see Remark9.13). P is called the Holomorphic Polarization and P the
antiholomorphic polarization40 .
Examples 9.17. We present three further elementary examples:
1. The two-sphere S2 has no real polarization as we know by Example 9.3. But it
has, similar to the cotangent bundle T ∗ Rn in the preceding example, the holomorphic
d
polarization P . It is defined locally by dz̄ for local complex coordinates z. P is Kähler
and positive.
2. The same holds true for the cylinder M = T ∗ S1 .
3. Let M = T ∗ R2 with the usual symplectic form, the standard canonical coordi-
nates p1 , p2 , q 1 , q 2 and the complex coordinates zj = pj + iq j as before. Define
 
∂ ∂ ∂
P := spanE(M ) +i 1 , .
∂p1 ∂q ∂p2
Then P is Lagrangian with
 
∂
D = P ∩ P ∩ T M = spanE(M,R) ,
∂p2

with dimR Da = 1 , dimR Ea = 3. Moreover, P and P + P are involutive. Hence, P is

a complex polarization. P is reducible and positive.
The example can be generalized to n > 1 by looking at Example 9.14. For instance,
a positive polarization P ⊂ T M C , M = T ∗ Rn , is given by
   
∂ ∂ ∂ ∂
P = spanE(M,C) ,..., ∪ +i j |k <j ≤n ,
∂p1 ∂pk ∂pj ∂q
   
∂ ∂ ∂ ∂
P = spanE(M,C) ,..., ∪ −i j |k <j ≤n ,
∂p1 ∂pk ∂pj ∂q
 
∂ ∂
D = spanE(M,R) ,..., ,
∂p1 ∂pk
   
∂ ∂ ∂ ∂
E = spanE(M,R) ,..., ∪ ..., n .
∂p1 ∂pn ∂q k+1 ∂q
Remark 9.18. The product of complex polarizations P of (M, ω) and P ′ of (M ′ , ω ′ )
is defined as
P ⊕ P ′ ⊂ T MC ⊕ T M′ ∼ = T (M × M ′ )C .
C

where the symplectic form on M × M ′ is ω ⊕ ω ′ : P ⊕ P ′ is a complex polarization

of the symplectic manifold (M × M ′ , ω ⊕ ω ′ ). For instance, the generalized example
P of T ∗ Rn just described is essentially the product of the vertical polarization Pk of
M = T ∗ Rk and the holomorphic polarization Pn−k of T ∗ Rn−k .
40
In some texts, P is called the holomorphic, and P the antiholomorphic polarization.
130 9. Polarization

The contribution of complex polarizations to the reduction of the prequantum

Hilbert space to obtain a correct Hilbert space as the representation space will be
treated in the next chapters.

9.5 Kähler Polarization

Kähler polarizations P are those complex polarizations on a symplectic manifold (M, ω)

which satisfy P ∩ P = {0}. They obtained this name since every Kähler manifold,
i.e. a complex manifold M with a its symplectic (Kähler) form ω, has the holomorphic
polarization as a natural polarization which is Kähler (and positive). This will be
explained in this section in some detail, together with the converse, namely the fact,
that the existence of a Kähler polarization on a symplectic manifold (M, ω) induces a
complex structure on M such that M is Kähler with Kähler form ω.
We first study the structure of a complex vector space on a real vector space V of
dimension 2n.

Definition 9.19. Any R-linear map J : V → V with J 2 = −1 is called an Almost

Complex Structure 41 .

Note, that this definition appears also as Definition B.21 in the Appendix about
Complex Analysis.
It is easy to see that if J is an almost complex structure, then by

(α + iβ)(v) := αv + βJ(v), v ∈ V, α, β ∈ R

a scalar multiplication C × V → V is defined, introducing on V the structure of a

complex vector space. On the other hand, if V is the underlying real vector space of
a complex vector space, then the multiplication with i (or −i): I : V → V , v 7→ iv is
an almost complex structure: I is R-linear and I(I(v)) = I(iv) = i2 v = −v , v ∈ V .
Let ω be a symplectic form on V (i.e. R-bilinear, non-degenerate and alternating).
ω is called to be Compatible with J if

ω(v, w) = ω(Jv, Jw) ,

for all v, w ∈ V , i.e. if J is a (linear) canonical transformation.

Remark 9.20. If Jω is the matrix representing ω with respect to a basis of V and if

the matrix representing J will be denoted be the same symbol J, then the compatibilty
condition is equivalent to: J ◦ Jω = Jω ◦ J. In fact: ω(v, w) = v ⊤ Jω w. Thus, because of
ω(Jv, Jw) = (Jv)⊤ Jω Jw = v ⊤ J ⊤ Jω Jw and J −1 = J ⊤ the compatibility is equivalent
to J −1 Jω J = Jω or J ◦ Jω = Jω ◦ J.
41
For example, the symplectic involution σ introduced in Section 1.1 is an almost complex strucutere.
9.5 Kähler Polarization 131

In particular, we can choose the basis of V such that Jω is the block matrix
 
0 1
σ= ,
−1 0

where 1 is the unit n × n-matrix. (Such a basis is called a symplectic frame.)

We conclude that the complex structures J such that ω is compatible with J can
be parametrized by certain elements of the (real) symplectic group Sp(2n) (Sp(2n) =
{S ∈ GL(2n, R) | S ⊤ ◦ σ ◦ S = σ}, see (84)), namely those S ∈ Sp(n) with S 2 = −1.
For example, if we choose a real n × n-matrix Y which is invertible and symmetric,
the block matrix
0 Y −1
 
J=
−Y 0
satisfies J 2 = −1 and J ⊥ σJ = σ and thus defines an almost complex structure com-
patible with the symplectic structure. In particular we see that there exist many
compatible almost complex structures.
Another set of compatible almost complex structures is given by the block matrices
 
−X −1
J= ,
1 + X2 X

where X is an arbitrary real symmetric n × n-matrix. Moreover, any compatible J has

as its matrix the block matrix

−Y −1 X −Y −1
 
J= ,
Y + XY −1 X XY −1

X, Y as above.

A compatible almost complex structure on V induces a symmetric bilinear form

g : V × V → R and a sesquilinear form h : V × V → C on V defined by

g(v, w) := ω(v, Jw)

h(v, w) := g(v, w) + iω(v, w)

g is non-degenerate, but in general not positive definite. One could have defined
g (v, w) = ω(Jv, w) to obtain g = −g ′ (see Remark 9.13).
′

Let us now consider these structures on the complexification V C of V as in Section

9.3. The R-linear map J can be linearly extended to V C as a C-linear map J C by
J C (v + iw) := Jv + iJw with (J C )2 = −1. The 4n-dimensional space V C carries now
two almost complex structures, namely the extension J C of J and the almost complex
structure coming from the complexification: v + iw 7→ i(v + iw) = −w + iv.
132 9. Polarization

The extension J C can be diagonalized with respect to the two eigenvalues i, −i: In
fact, for u ∈ V :
J C (u − iJu) = J(u) + iu = i(u − iJu)
and
J C (u + iJu) = J(u) − iu = −i(u + iJu)
with the eigenspaces

V (1,0) := {z − iJ C z | z ∈ V C } = {z ∈ V C | J C z = +iz} = Ker (J C − i)

(43)
V (0,1) := {z + iJ C z | z ∈ V C } = {z ∈ V C | J C z = −iz} = Ker (J C + i).

Note, that V C = V (1,0) ⊕ V (0,1) : For each v ∈ V C

1 1
v = (v − iJ C v) + (v + iJ C v) .
2 2

By definition, the two almost complex structures agree on V (1,0) (J C v = iv , v ∈

(1,0)
V ) and differ on V (0,1) by a minus sign (J C v = −iv , v ∈ V (0,1) ).
The notation V (1,0) , V (0,1) is due to the notation of the direct sum representation of
the complexified tangent bundle T M C = T (1,0) M ⊕T (0,1) M in order to define differential
forms of degree (r, s), among others (see Definition B.26).

Proposition 9.21. Let (V, ω) be a symplectic vector space.

1. Assume, in addition, that J is an almost complex structure on V which is com-

patible with ω. Then the eigenspace P := V (0,1) of J C with eigenvalue −i is a
complex Lagrangian subspace with P ∩ P = {0}, hence Kähler. Analogously for
P.

2. Conversely, a complex Lagrangian subspace P of V C , which is Kähler, defines a

compatible almost complex structure J on V such that P is the eigenspace of J C
with respect to i or −i.

Proof. To show 1., we see that P is isotropic since for v, w ∈ P

ω(v, w) = ω(J C v, J C w) = ω(−iv, −iw) = (−i)2 ω(v, w) = −ω(v, w) ,

i.e. ω(v, w) = 0. P is maximally isotropic since dimC P = n. Finally, for v ∈ P ∩ P =

{0} we have iz = J C z = −iz, hence z = 0.
To show the converse, i.e. 2., we use the decomposition V C = P ⊕ P and the
corresponding projections π , π̄ : V C → V C onto the subspaces P = Im π , P = Im π̄.
Any v ∈ V C has the representation v = πv ⊕ π̄v. When v ∈ V this decomposition
implies
v = v̄ = πv ⊕ π̄v = πv ⊕ π̄v ,
9.5 Kähler Polarization 133

hence πv = π̄v , π̄v = πv . As a consequence:

iπ(v) − iπ̄(v) = iπ(v) − iπ̄(v) ,

i.e. iπ(v) − iπ̄(v) ∈ V Therefore, the map J : V → V

Jv := iπ̄(v) − iπ(v) , v ∈ V,

(which is roughly the ”imaginary part” of πv) is well-defined and R-linear. Moreover,
J satisfies J 2 = −1 = −idV :

iπ̄(iπ̄v − iπv) − iπ(iπ̄v − iπv) = −(π̄)2 v − (π)2 v = −π̄v − πv = −v .

P is the eigenspace Ker (J C + i) since for z ∈ V C

J C z = −iz ⇐⇒ iπ̄(z) − iπ(z) = −i(π(z) + π̄(z)) ⇐⇒ π̄(z) = 0 ⇐⇒ z ∈ P .

Finally, J is compatible with ω by the isotropy of P and P .

As a result, for a given symplectic vector space (V, ω) there is a natural bijective
correspondence between the set of complex Lagrangians on V C (with respect to ω),
which are Kähler, and the set of compatible almost complex structures on (V, ω).
And there is a natural bijective correspondence between the set of positive Kähler
polarization and the set of positive almost complex structures J (i.e. where g, g(v, w) =
ω(v, Jw), is positive definite). Moreover, by Remark 9.20 the set of compatible almost
complex structures can be identified with a subset of {J | J ◦ σ = σ ◦ J} ∼
= Sp(2n), the
symplectic group (see (84)).
Before we discuss the relation between Kähler polarizations and complex manifold
structure we present a generalization of Example 9.16 which shows the effect of changing
the almost complex structure on a given symplectic vector space. In particular, this
example indicates in which way non-positive Kähler polarizations occur.

Example 9.22. We start with the simple phase space M = T ∗ Rn with its standard
symplectic form ω) = dq k ∧ dpknwith respect
o to the usual canonical coordinates (q, p)
of M . The corresponding basis ∂q∂ j , ∂p∂k is a symplectic frame, i.e. the matrix repre-
senting ω is the block matrix  
0 1
σ= .
−1 0
Let J be the almost complex structure on M = R2n given by a real symmetric and
invertible n × n-matrix Y through the block matrix

0 Y −1
 
J= .
−Y 0
134 9. Polarization

We introduce complex coordinates z k := g jk pj + iq k , where Y = (gjk ) , Y −1 = (g jk ).

In this way multiplication by i reflects the almost complex structure J. Note, that
the complex coordinates in Example 9.16 are given by Y = 1, i.e. z k = pk + iq k . In
particular, the new coordinates determine the holomorphic structure in the sense that
the holomorphic function are the functions f such that for all k = 1, ...n
∂
f = 0.
∂ z̄ k
Here,  
∂ 1 ∂ ∂
k
:= gjk +i k .
∂ z̄ 2 ∂pj ∂q
The symplecic form can be written as
1
ω = dq j ∧ dpj = igjk dz̄ j ∧ dz k .
2
The symplectic potential is
1
α = igjk z̄ j dz k .
2
The holomorpic polarization P = PJ determined by the compatible almost complex
structure J is generated by the hamiltonian vector fields ∂∂z̄k . The potential α is adapted
in the sense that α(X) = 0 for X ∈ Γ(M, P ).
Finally, let (r, s) be the signature of the matrix Y = (gjk ) then this signature is the
signature of the polarization. In particular, PJ is positive if and only if s = 0 which is
the same as Y being positive definite.

Kähler Manifolds
After this digression about the linear case we come back to manifolds. We have
seen, that on a real vector space V an almost complex structure is essentially the
same as a complex structure, in the sense that J determines on V the structure of
a complex vector space. An analogous property for manifolds is no longer true. A
complex structure on a manifold is given by an atlas of holomorphically compatible
charts. These charts induce an almost complex structures in the tangent spaces, as is
explained in Section B.21, but the converse need not hold.

Definition 9.23. An Almost Complex Structure on a manifold M is a section (a

tensor field) J ∈ Γ(M, End(T M )) with J 2 = −idT M . In particular, for each a ∈ M the
map Ja : Ta M → Ta M is an almost complex structure in the linear sense (cf. Definition
9.19).

Any complex manifold M has a natural almost complex structure, namely the
multiplication by i in the tangent spaces Ta M , a ∈ M , induced by the holomorphic
9.5 Kähler Polarization 135

charts (see Example 3. in B.22 for the description in local coordinates). However,
there exist examples of differentiable manifolds with an almost complex structure which
cannot be induced by a complex manifold structure.
Definition 9.24. When an almost complex structure J on a manifold comes from
a complex structure, i.e. from a holomorphic atlas as required above, J it is called
Integrable.

The following result of Newlander and Nirenberg can be found, e.g., in [Huy05].
Theorem 9.25 (Newlander-Nirenberg). An almost complex structure on a differen-
tiable manifold M  is integrable if the induced T (0,1) M = T M (0,1) is involutive, i.e.
(0,1) (0,1) (0,1)
T M, T M ⊂T M.

Note, that an almost complex structure J induces a direct sum decomposition

(1,0)
T M ⊕T (0,1) M = T M C of the complixified tangent bundle T M C into the eigenspaces
(1,0)
T M = Ker (J C − i) and T (0,1) M = Ker (J C + i) of the complexification J C of J.
Proposition 9.26. A symplectic manifold (M, ω) with a Kähler polarization P in-
duces a compatible almost complex structure, which is integrable. In particular, M is a
complex manifold in a natural way and P = T (0,1) M is the holomorphic polarization.

Proof. We have to transfer the above results for symplectic vector spaces to the mani-
fold case. For each a ∈ M , there is a natural and compatible almost complex structure
(0,1)
Ja : Ta M → Ta M whose complexification satisfies Ta M = Ker (JaC +i) = Pa ⊂ T M C
according to Proposition 9.21. The induced section J ∈ Γ(M, End (T M )) is an almots
complex structure. Since P = T (0,1) is involutive as a complex polarization, by the
theorem of Newlander-Nirenberg J is integrable.
Definition 9.27. A symplectic manifold (M, ω) with a positive Kähler polarization P
is called a Kähler manifold.

This is a definiton in the spirit of symplectic manifolds and polarizations.

By the last result a Kähler manifold is, in particular, a complex manifold such
that the complex structure is compatible with ω and such that P is the holomorphic
polarization. Moreover, since the Kähler polarization is positive, the symmetric form
g(X, Y ) = ω(X, JY ) , X, Y ∈ V(M ), is positive and therefore a Riemannian metric g
on M .
The usual definition of a Kähler manifold is the following.
Definition 9.28. A Kähler manifold is a complex manifold M with a Riemannian
metric g such that the almost complex structure J is compatible with g:

g(X, Y ) = g(JX, JY ) ∀X, Y ∈ Ta M, a ∈ M.

Moreover, the induced form ω(X, Y ) := g(JX, Y ) is closed.

136 9. Polarization

The two definitions are equivalent. Eventually, a Kähler manifold (M, J, g) carries
the structure of a symplectic manifold (M.ω) and a positive polarization P such that
all the structures J, g, ω, P are compatible with each other.
 137

10 Representation Space

With all the ingredients:

• a symplectic manifold (M, ω);

• a prequantum line bundle (L, ∇, H)

• a complex polarization P ⊂ T M C

developed so far, one now can construct – as an essential part of the programme
of Geometric Quantization – the Representation Space, i.e. the Hilbert space of
the quantum model, on which the quantum observables, which correspond to classical
observables of a given subset o ⊂ E(M ), act as self-adjoint operators42 in an irreducible
way. Of course, this will be done on the basis of the prequantum operator

q(F ) : H → H , F ∈ E(M ) ,

constructed for (L, ∇, H) in Chapter 7, where H = H(M, L) is the prequantum Hilbert

space of the square integrable global sections of L.
To construct the reduced representation space with respect to a complex Polar-
ization P one considers polarized sections s of L (i.e. ∇X s = 0 for all vector fields
X ∈ Γ(M, P )) as the starting point. Intuitively, polarized sections are the sections of
L which are constant along the leaves of the polarization P , more precisely, along the
leaves of the induced distribution D.
The construction of the representation space can be quite complicated. Among
other problems43 , one needs to know how to integrate polarized sections. In general,
the natural volume induced by ω n on M does not work any more. In the case of a
Kähler polarization, however, – which we investigate in Section 2 – it is clear how to
integrate and a general programme of Geometric Quantization can be carried through
directly. Moreover, for the vertical polarization on the simple phase space M = T ∗ U ,
U open in Rn , the integration is no problem, as well as for the horizontal polarization.
In the general case, we assume for this chapter, that there is a natural volume form
d vol on the quotient space M/D44 where D is the integrable real distribution with
DC = P ∩ P . Now, the representation space HP can be defined as the completion of
the prehilbert space
Z
{s ∈ Γ(M, L) | s polarized and H(s, s)d vol < ∞} .
M/D

42
or at least as symmetric operators
43
In some cases there are no non-zero polarized sections at all, cf. the next chapter.
44
In general one should use a density on the quotient M/D, see Chapter 12.
138 10. Representation Space

Finally, for a first full version of Geometric Quantization one needs the concept
of a directly quantizable observable, that is an real function F ∈ E(M ) such that for
all polarized sections s the derivative ∇XF s is also polarized. For such an observable
F the prequantum operator q(F ) has a natural restriction to HP yielding a quantum
operator, which is denoted again by q(F ) ∈ S(HP ), in such a way that the Dirac
conditions (see Chapter 2) are satisfied.

10.1 Polarized Sections

The purpose of introducing polarizations is to reduce the set of wave functions (i.e.the
elements ψ) in the prequantum Hilbert space H of prequantization (cf. Chapter 7)
in order to make the representation irreducible. This is done by restricting to those
functions, sections, vector fields, which are parallel to the given polarization, or, in
other words, which are polarized.

Definition 10.1. For a complex polarization P on a symplectic manifold (M, ω) the

Polarized Functions are the functions f ∈ E(M, C) with

LX f = 0 for all X ∈ Γ(M, P ) .

The Polarized sections in a line bundle L over M with connection ∇ are the
sections s ∈ Γ(M, L) with

∇X s = 0 for all X ∈ Γ(M, P ).

The basic idea is to consider only those sections of a prequantum bundle (L, ∇, H)
on (M, ω) as possible ”wave functions” that are ”constant along the directions of P ”
with respect to a polarization P ⊂ T M C . This idea is made precise by using the
definition of a polarized section and by constructing the corresponding Hilbert space
of ”wave functions” based on the space

Γ∇,P (M, L) := {s ∈ Γ(M, L) | ∇X s = 0 for all X ∈ Γ(M, P )}

of polarized sections. Γ∇,P (M, L) is a vector space over C. However, Γ∇,P (M, L) is
not a module over E(M ). For a section s ∈ Γ∇,P (M, L) and a function f ∈ E(M ) the
covariant derivative ∇f s need not be polarized, in general. It is polarized, whenever
∇X f s = (LX f )s + f ∇X s = 0 for all X ∈ Γ(M, P ). This holds true if f is a polarized
function. Therefore, Γ∇,P (M, L) is a module over the ring EP (M ) of polarized functions.
Note, that ∇X s is in general not polarized for polarized s and X ∈ V(M ). This is a
serious problem when it comes to determine the quantum operator q(F ) appropriately,
which we will address in the third section of this chapter.
In order to construct the Hilbert space using the space of polarized sections, we
have to describe what kind of scalar product on Γ∇,P (M, L) is reasonable. One can
10.1 Polarized Sections 139

not simply integrate along the volume form ω n on M which is the right choice for
the prequantum Hilbert space. For instance, with respect to the vertical polarization
P = DC on M = T ∗ Rn , a non-zero polarized section s in the trivial bundle is constant
along the fibres of the canonical projection τ : T ∗ Rn → Rn , the leaves of the vertical
distribution
R D. These fibres are {(p, q) | p ∈ Rn } , q ∈ Rn fixed. As a consequence,
M
⟨s, s⟩d vol = ∞ for s ̸= 0. In this case, one can integrate along Rn which is essentially
M/D with respect to the natural measure on M/D ∼ = Rn , the Lebesgue measure.
In the following elementary example we show that the original purpose to reduce the
number of variables by using the vertical polarization on T ∗ Rn and using the natural
measure is successful insofar that it leads to the expected results and gives the right
representation spaces. The same holds for the horizontal polarization.
Examples 10.2 (Simple Phase Space). Let M = T ∗ U , U ⊂ Rn open, with the
standard symplectic form dq j ∧ dpj and let L = M × C be the trivial complex line
bundle with the global section s1 ∈ Γ(M, L) , s1 (a) = (a, 1), for a ∈ M , and with the
induced Hermitian structure H on L.

1. Consider the vertical polarization P := DC associated to the vertical distribution

D (see Example 9.7) which is
 
∂
P := spanE(M ) j = 1, . . . , n ,
∂pj
and consider the connection ∇ on L given by the negative of the Liouville form: −λ =
−pk dq k . (L, ∇, H) is a prequantum line bundle. Recall

∇X f s1 = LX f − 2πipk dq k (X)f s1 , X ∈ V(M ).

The polarized sections are

∂
Γ∇,P (M, L) := {f s1 ∈ Γ(M, L) | f ∈ E(M ) with f = 0 , j = 1, . . . , n} ,
∂pj

since −pk dq k (X) = 0 for X ∈ Γ(M, P ).

This space of polarized sections can be identified with the space E(U ) of functions
f = f (q, p) on M which only depend on the variable q. We integrate over U with
respect to the Lebesgue measure and get the Hilbert space HP = L2 (U, dλ(q)) of
square integrable functions on U as the reduced representation space.
The quantum operators q(F ) for F = pj , q k , have the form
 
j i ∂
q(q ) = − − − 0 + q j = q j =: Qj , on E(U ) ,
2π ∂pj
 
i ∂ i ∂
q(pj ) = − j
− 2πipj + pj = − =: Pj , on E(U ) ,
2π ∂q 2π ∂q j
140 10. Representation Space

as we expect it from elementary quantum mechanics in the Schrödinger representation.

2. Now let P ′ := DC be the horizontal polarization associated to the horizontal
distribution D (see Example 9.7). Then
 
′ ∂
P := spanE(M ) j = 1, . . . , n .
∂q j

Consider the connection ∇′ given by the 1-form: α = qk dpk . Recall

∇′X f s1 = LX f + 2πiqk dpk (X)f s1 .

The polarized sections are

∂
Γ∇′ ,P ′ (M, L) := {f s1 ∈ Γ(M, L) | f ∈ E(M ) with f = 0 , j = 1, . . . , n}
∂qj

since α(X) = 0 for X ∈ Γ(M, Q).

This space of polarized sections can be identified with the space E(Rn ) of functions
f = f (q, p) on M which only depend on the variable p. The corresponding representa-
tion space is HP ′ = L2 (Rn ).
The quantum operators q(F ) for F = pj , q k have the form
 
j i ∂ i ∂
q(q ) = − − + 2πiqj + q j = =: Qj ,
2π ∂pj 2π ∂pj
 
i ∂
q(pj ) = − + 0 + pj = pj := Pj ,
2π ∂q j
as we expect it from elementary quantum mechanics in the Heisenberg representation.

Remark 10.3. The prequantum operators satisfy the canonical commutation relations

i j
[Qj , Pk ] = δ
2π k
in both examples. See Section F.3 in the Appendix for the relevance of the canonical
commutation relations.

Remark 10.4. The two connections used in the preceding example are not the same,
but they induce representations which are unitarily equivalent. This can be seen di-
rectly by using the Fourier transform T : HP → HQ as intertwining operator satisfying

T ◦ Qj = Qj ◦ T , T ◦ Pj = Pj ◦ T .

Unitary equivalence follows also by applying the Theorem of Stone-von Neumann F.47.
10.1 Polarized Sections 141

What happens when we use the connection ∇ defined by −λ also for the horizontal
polarization P ′ ? We obtain another representation, which is unitarily equivalent in a
natural way to the second representation: s = f s1 is polarized if and only if

∂
f − 2πipj f = 0 , j = 1 . . . , n.
∂qj

A general solution of this system of partial differential equations is f = g(p) exp(2πiqp),

where g = g(p) ∈ E(Rn ) and qp := q j pj . Hence, the space Γ∇,P ′ (M, L) of polarized
sections can be identified with the space {g(p) exp(2πiqp) | g = g(p) ∈ E(Rn )} and the
corresponding representation space is

H = {g exp(2πiqp) | g = g(p) ∈ L2 (Rn )} ∼

= HP ′ .

The quantum operators q(F ) for F = pj , q k are

 
j i ∂
q(q ) (g exp(2πiqp)) = g exp(2πiqp) ,
2π ∂pj

q(pj ) = pj .

H together with q is unitary equivalent to HP and HP ′ together with its quantum

operators q.

The preceding considerations show, that in the case of a simple phase space T ∗ U ,
U ⊂ Rn open, the reduction by polarization leads to the right representation space
known from elementary quantum mechanic. Before we confirm this result also in the
case of a Kähler polarization, let us consider the general case.
For a reducible complex polarization P we have, as before, the induced integrable
distribution D with P ∩ P = DC and the quotient M/D, the space of leaves M/P =
M/D. We assume that M/D is endowed with some natural volume form vol.45 .

Definition 10.5 (Representation Space). The representation space H = HP = H∇,P

is the completion of
 Z 
Hpre := s ∈ Γ∇,P (M, L) ⟨s, s⟩ := H(s, s)d vol < ∞ .
M/D

Here, for a polarized section s ∈ Γ(M, L) the term H(s, s) in the integral is the following
function on M/D. H(s, s)(x) = H(s(a), s(a)) for a ∈ x. This is well-defined since the
polarized s is constant on the leaves x of D, i.e. s(a) = s(a′ ) for a, a′ ∈ x.
45
According to Chapter 12 we can relax this condition and integrate instead along a density on
M/D which is always possible, and which leads to representation spaces unitarily equivalent to each
other in a natural way.
142 10. Representation Space

Remark 10.6. The representation space may R be trivial in the sense that HP = 0
because for no polarized section s the integral M/D H(s, s)d vol is finite, see for insance
in the coming Example 10.13.

Before we describe Geometric Quantization using the new representation space HP

we study the construction of the reduced representation space in the relatively easy
case of a Kähler polarization, where M/D = M and the symplectic form ω induces a
natural volume form.

10.2 Kähler Quantization

Let P be a Kähler polarization P ⊂ T M C on our symplectic manifold (M, ω), that is

P ∩ P = {0}. In this situation, there exists a unique complex structure on the manifold
M (i.e. a structure of a complex manifold, or in other words an almost complex struc-
ture which is integrable) such that P is the holomorphic polarization (cf. Proposition
9.26), P = T (0,1) M . Thus, for all local holomorphic charts
φ = z = (z1 , . . . zn ) : U → V ⊂ Cn , U, V open,
of the complex structure we have
 
∂ ∂
Pa = spanC ,..., ⊂ Ta M C = Ta M ⊕ iTa M, a ∈ U.
∂ z̄1 ∂ z̄n

In addition, let (L, ∇, H) be a prequantum bundle on the symplectic manifold

(M, ω).
Observation 10.7. The polarization P induces on the complex line bundle L a natural
structure of a holomorphic line bundle.46 Hence, in the following, we regard L is a
holomorphic line bundle. The polarized sections s : M → L are nothing else than the
holomorphic sections.

Proof. For each a ∈ M there exists an open neighbourhood U ⊂ M and a nowhere

zero polarized section s ∈ Γ(U, L). Let (Uj ) be an open cover of M together with
nowhere zero polarized sections sj ∈ Γ(Uj , L). On Ujk = Uj ∩ Uk ̸= ∅ there exists
gjk ∈ E(Ujk ) with sj = gkj sk since sj , sk are nowhere zero. The functions gjk are
holomorphic because of the fact that sj , sk are polarized: For X ∈ Γ(M, P ):
0 = ∇X sj = (LX gkj )sk + gkj ∇X sk = (LX gkj )sk ,
hence LX gkj = 0.
Now, by
ψj : L|Uj → Uj × C , zsj (a) 7→ (a, z) , (a, z) ∈ Uj × C ,
46
The holomorphic line bundle structure is compatible with ∇.
10.2 Kähler Quantization 143

a system of local trivializations of L is defined whose transition functions are holomor-

phic. In fact,

ψk ◦ ψj−1 (a, z) = ψk (zsj (a)) = ψk (zgkj sk (a)) = (a, zgkj (a)) = (a, gkj .z) , (a, z) ∈ Ujk .

and it follows that ψk ◦ ψj−1 : Ujk → Ujk is biholomorphic. As a consequence, the local
trivializations (ψj ) determine the structure of holomorphic line bundle on L. It is clear,
that this structure is inedependent of the choice of the cover (Uj ) and the choice of the
sections (sj ).

The symplectic form induces a natural volume form vol = Cω n on M (C > 0 some
constant) and the prehilbert space we are looking for is:
Z
Hpre := {s ∈ Γhol (M, L) | H(s, s)d vol < ∞}.
M

Hpre can be completed in order to yield a separable quantum Hilbert space HP , the
reduced representation space. HP is a closed subspace of the full prequantum Hilbert
space H(M, L) (see Section 7.3) of square integrable smooth sections of L considered
previously. One can even prove, that Hpre is already closed in H ([Woo80] and Propo-
sition B.14) and there is no need of completing Hpre .
How can the prequantum operators act on the reduced representation space HP ? In
general, it might happen that for a polarized section s the covariant derivative ∇XF s is
no longer polarized. In this situation the operator q(F ) might not be well-defined on a
suitable dense subspace of HP . The natural approach to overcome this difficulty is to
focus on a subset o of the Poisson algebra E(M ) of classical observables such that for
all F ∈ o there is a dense subspace DF ⊂ HP such that q(F )(DF ) ⊂ HP . These F are
called directly quantizable. We will not discuss these matters further at the moment
(see, however, the next section), but rather continue to present elementary examples
after the following remark.

Remark 10.8 (Kähler Quantization). Summarizing the above considerations, the sub-
space HP of holomorphic sections in the prequantum representation space H is the new
representation space of the Kähler polarization P and the restrictions of the prequan-
tum operators q(F ) to HP for the directly quantizable observables F yield a geometric
quantization satisfying (D1) aund (D2). This construction is sometimes called Kähler
quantization.

Example 10.9 (Simple Phase Space. Holomorphic Polarization). We come back to

the example of M = T ∗ Rn with the standard symplectic form ω = dq j ∧ dpj and
introduce, as before, the holomorphic coordinates

zj := pj + iq j ,
144 10. Representation Space

(which is essentially introducing a complex vector space structure on M ∼

= Rn × Rn
∼ n
such that M = C ). Let
 
∂ ∂
P = spanE(M ) ,..., ⊂ T M C.
∂ z̄1 ∂ z̄n
be the holomorphic polarization. The line bundle L → M is trivial (there are only
trivial line bundles on M ) and the connection is unique up to isomorphism as well
1
(since HdR (M, C) = 0). A potential α of
iX
ω= dz̄j ∧ dzj ,
2
which is adapted to P is given by
iX
α= z̄j dzj .
2 j

α is adapted in the sense that α(X) = 0 for all vector fields X ∈ Γ(M, P ) .
Instead of defining the connection of our prequantum bundle (L, ∇, H) with respect
to the non-vanishing section s1 , where s1 (a) = (a, 1) , a ∈ M , as before, we choose the
section !!
n
X π π
se (a) = a, exp − z̄j zj = exp(− z̄z)s1 (a) ,
j=1
2 2

(with z̄z := ∥z∥2 ) and define the connection ∇ by

∇X se := 2πiα(X)se ,

X ∈ V(M ).
Note, that the polarized sections depend on the choice of the connection resp. on the
choice of the potential α of ω and the non-vanishing section s with ∇s = 2πiαs. Even
equivalent connections have different polarized sections, see the subsequent observation
and the remark above. The spaces of polarized sections, however, are isomorphic, and
the resulting representation spaces with its prequantum operators will be unitarily
equivalent.
Each section s ∈ Γ(M, L) is of the form s = f se with f ∈ E(M ). Because of

∇X f se = (LX f + 2πiα(X)f )se ,

the section f se will be polarized if and only if LX f = 0 for all X ∈ Γ(M, P ) (recall
α(X) = 0) and this in turn is equivalent to f being a holomorphic function47 . As a
result
Γ∇,P (M, L) = {f se | f ∈ O(Cn )} ∼
= O(Cn ) .
47
f is partially holomorphic because of LX f = 0 for the generators X = ∂/∂ z̄j ∈ P .
10.2 Kähler Quantization 145

The natural scalar product is

Z
⟨f, g⟩ = f¯g exp(−π∥z∥2 )dz
Cn

for those f, g ∈ O(Cn ) for which the integral is defined (dz is Lebesgue integration
or integration with respect to ω n , which is the same up to a constant). The space of
polarized sections with the scalar product is essentially the Bargmann-Fock space
 Z 
n 2 2
F := f ∈ O(C ) |f | exp(−π ∥z∥ )dz < ∞ .
Cn

Note, that this space of holomorphic functions is already complete in the norm
given by the scalar product (cf. Proposition B.14). Thus, F is already a Hilbert space.
It is a proper subspace of O(Cn ) as a vector space and a proper and closed subspace of
the Hilbert space L2 (Cn , exp(−πz z̄)dz) of functions which are square integrable with
respect to exp(−πz z̄)dz.
We have constructed the reduced representation space HP = F for the simple phase
space T ∗ Rn ∼
= Cn and the holomorphic polarization P . This space will be denoted by
HP := F in the following. In comparison, the unreduced Hilbert space in the simple
case is the prequantum Hilbert space H = L2 (R2n , dλ) of square integrable functions
on R2n ∼=M ∼ = Cn with respect to Lebesgue integration dλ.
Let us finish the example by determining the prequantum operators
i
q(F ) = − ∇XF + F
2π
for F = z and F = z̄ 48 :
The Hamiltonian vector field XF of an observable F ∈ E(M ) has the following
expression in the complex coordinates
X  ∂F ∂ ∂F ∂

XF = 2i − . (44)
j
∂ z̄j ∂zj ∂zj ∂ z̄j

Hence,
∂ ∂
Xzj = −2i , Xz̄j = 2i ,
∂ z̄j ∂zj

For holomorphic f we have ∇Xzj f se = 0 since Xzj ∈ Γ(Cn , P ). Therefore,

q(zj ) = zj

is the multiplication operator f 7→ zj f on HP .

48
We allow complex classical observables in this example.
146 10. Representation Space

Furthermore,  
∂ ∂
∇Xz̄j f s1 = 2i f + 2πiα(2i )f s1
∂zj ∂zj
 
∂ i
= 2i f + 2πi (2iz̄j )f s1
∂zj 2
 
∂
= 2i f − 2πiz̄j f s1 ,
∂zj
hence
1 ∂ 1 ∂
q(z̄j ) = − z̄j + z̄j = .
π ∂zj π ∂zj
With the notations
j 1 ∂
Zk := q(zk ) = zk , Z := q(z̄j ) = , 1 ≤ j, k ≤ n ,
π ∂zj

we obtain
j 1 j 1 ∂
Z Zk ϕ = δk ϕ + zk ϕ,
π π ∂zj
when k = j, and finally
j 1 j
[Z , Zk ] = δ .
π k
These are the canonical commutation relations (CCR).
Further
Pn classical observables F will be discussed later, for instance, the energy
1
H = 2 j=1 zj z̄j of the harmonic oscillator in the example below.

Remark 10.10. With a∗j := Zj , aj := Z j we see that the Zj , Z j act as the familiar
raising and lowering operators. According to Remark B.18 the operator a∗j is a closed
operator with the space P of complex polynomials as its domain. Moreover, as the
notation suggests, a∗j is the adjoint of aj . The ”position” and ”momentum” operators
in this context are the self-adjoint operators49
 
1 1 ∂ 1
Aj := √ zj + = √ (a∗ + aj ),
2 π ∂zj 2
 
i 1 ∂ i
Bj := √ zj − = √ (a∗ − aj ).
2 π ∂zj 2
The CCR are satisfied in the following form

1
[Aj , Bk ] = δjk .
π
49
In Section F.2 closed, adjoint and self-adjoint operators are treated.
10.2 Kähler Quantization 147

Observation. The covariant derivative ∇s1 of the section s1 = exp( π2 ∥z∥2 )se = gse is
 
iX
∇s1 = ∇gse = dg + 2πi z̄j dzj g se
2
π X 
= (z̄j dzj + zj dz̄j ) − πz̄j dzj s1
2 
iX ¯
= 2πi z̄j dzj − zj dzj s1
4

In other words, when the connection ∇ shall be

P defined with respect to the section
¯
s1 this has to be done with the potential β = /4( z̄j dzj − zj dzj ) of ω instead of α in
i

order to obtain the same connection. Let us denote by ∇′ the connection defined using
α and s1 :
∇′ s1 = 2πiαs1 .
The polarized sections are

Γ∇′ ,P (M, L) = {f s1 | f ∈ O(Cn )} ,

Γ∇,P (M, L) = {f se | f ∈ O(Cn )} .

Remark 10.11. The representation space in the preceding example is called the
Bargmann-Fock Representation50 . Similar to this example we have presented
the examples with the vertical resp. horizontal distribution (and real polarization) in
M = T ∗ Rn in 1. and 2. of Examples 10.2: Here the reduced Hilbert space is L2 (Rn ).

Examples 10.12 (Simple Phase Space With Different Polarizations). We collect our
results of geometric quantization of position and momentum in the classical case of a
simple phase space M = T ∗ Rn and the three main natural polarizations

1. The vertical distribution D on T ∗ Rn induces the vertical polarization P = DC .

It will be used to reduce the prequantum representation space H = H(M, L)
to obtain as the reduced Hilbert space the Schrödinger Representation
HP = L2 (Rn , dλ) of square integrable functions ϕ = ϕ(q) with respect to Lebesgue
integration dλ = dλ(q) on the configuration space Rn . The prequantum operators
corresponding to position and momentum are (cf. 1. in Examples 10.2)

i ∂
Qj := q j and Pj := − .
2π ∂q j

They satisfy the following canonical commutation relations

 j  i j
Q , Pk = δ .
2π k
50
also called Segal-Bargmann representation or simply Bargmann representation
148 10. Representation Space

2. The horizontal distribution D on T ∗ Rn leads to the horizontal polarization

Q = DC ). The reduced Hilbert space is sometimes called the Heisenberg
Representation HP = L2 (Rn , dλ) with sλ = dλ(p) the Lebesgue measure but
now consisting of wave functions depending on the momentum variables pj only.
One obtains the quantized operators (cf. 2. in Examples 10.2)
i ∂
Qj := and Pk := pk ,
2π ∂pj
and therefore, the canonical commutation relations
 j  i j
Q , Pk = δ .
2π k

3. Using the holomorphic polarization P on T ∗ Rn in order to reduce the prequantum

representation space one obtains as the reduced Hilbert space the Bargmann
Space HP = OL of the preceding example. The prequantum operators corre-
sponding to the ”classical” observables zj , z̄j are
j 1 ∂
Zk = zk , Z = , 1 ≤ j, k ≤ n ,
π ∂zj
They satisfy the following canonical commutation relations
j 1 j
[Z , Zk ] = δ .
π k

All three representations are unitarily equivalent to each other as is explained in

Section F.3.

The Bargmann-Fock representation is useful for quantization of the harmonic os-

cillator in n dimensions (see below) and of a simplified model of a Bose-Einstein field.
Further Kähler quantizations are given by a family of almost complex structures J
on our simple phase space T ∗ Rn = Rn × Rn generalizing Example 10.9:
Example 10.13 (Simple Phase Space. Variants of Holomorphic Polarization). We
come back to the example of M = T ∗ Rn ∼ = Rn × Rn = V with the standard symplectic
form ω = dq j ∧ dpj . But in contrast to Example 10.9 we do not introduce the holomor-
phic polarization of the standard almost complex structure q k 7→ pk , pk 7→ −q k but
the holomorphic polarization P = PJ determined by the almost complex structure J
(see Proposition 9.21) given by the block matrix

0 Y −1
 
J= ,
−Y 0

where Y = (gjk ) is a symmetric and real n × n-matrix with inverse Y −1 = (g jk ). In

this way Example 9.22 will be continued.
10.2 Kähler Quantization 149

The corresponding holomorphic coordinates with respect to J are

z k := g jk pj + iq k ,

The symplectic form is

1
ω = dq j ∧ dpj = igjk dz̄ j ∧ dz k ,
2
with symplectic potential
1
α = igjk z̄ j dz k .
2
The holomorphic polarization is P = PJ = ker(J C + i) ⊂ T V C is generated by the
vector fields ∂∂z̄k .
The prequantum line bundle L → V is trivial. The connection ∇ on L will be
defined with respect to the non-vanishing section se
π π
se (a) = (a, exp(− gjk z̄j zk )) = (a, exp(− g(z̄, z))) ,
2 2
(g(z̄, z) := gjk z̄ j z k ) by
∇X se := 2πiα(X)se ,
X ∈ V(M ), and the hermitian structure is the one induced from L = V × C.
Each section s ∈ Γ(V, L) is of the form s = f se with f ∈ E(V ). Because of

∇X f se = (LX f + 2πiα(X)f )se ,

the section f se will be polarized if and only if LX f = 0 for all X ∈ Γ(V, P ) (recall
α(X) = 0) and this in turn is equivalent to f being a holomorphic function with
respect to J. Let O(V ) denote the space of these holomorphic functions, a more
precise notation is O(VJ ) when VJ denotes the complex vector space with underlying
V and almost complex structure J. As a result
π
= {f exp(− g(z̄, z)) | f ∈ O(V )} ∼
Γ∇,P (M, L) = {f se | f ∈ O(V )} ∼ = O(V ) .
2
The natural scalar product on {f exp(− π2 g(z̄, z)) | f ∈ O(V )} is
Z
π
′
⟨f, f ⟩ = f¯f ′ exp(− g(z̄, z))dz
V 2
for those f, f ′ ∈ O(V ) for which the integral is defined (dz is Lebesgue integration
or integration with respect to ω n , which is the same up to a constant). The space of
polarized sections with the scalar product is essentially the Bargmann space
 Z 
2 π
FJ = F := f ∈ O(V ) |f | exp(− g(z̄, z)))dz < ∞ .
V 2
150 10. Representation Space

This space of holomorphic functions is already complete in the norm given by the
scalar product (cf. Proposition B.14), i.e. F is already a Hilbert space. It is a proper
subspace of O(V ) as a vector space and a proper and closed subspace of the Hilbert
space L2 (Cn , exp(− π2 g(z̄, z))dz) of functions which are square integrable with respect
to exp(− π2 g(z̄, z))dz.
Note, that the representation space FJ is trivial whenever Y is not positive definite
since then only the holomorphic f , f = 0, has a finite integral
Z
π
|f |2 exp(− g(z̄, z))dz .
V 2

Therfore, we mostly restrict to the positive case.

We have constructed the reduced representation space HP = FJ for the simple
phase space T ∗ Rn ∼
= R2n and the holomorphic polarization PJ .

Example 10.14 (2-sphere). We continue the example of the 2-sphere M = S2 . Let

1
ω be the symplectic form such that ω = 4π vol for the natural volume form. We know
2
from Example 7.6 that (S , Cω) is quantizable if and only if C ∈ Z , N = C ̸= 0,
and from Example 8.11 that up to equivalence each (S2 , N ω) has only the prequantum
bundles (H(N ), ∇N ), N ∈ Z, with ∇N as in Example ??. Here, H(1) is the hyperplane
line bundle and H(−1) the tautological line bundle. H(N ) for N ∈ N is the N -fold
tensor product of H(1)
As a differentiable manifold S2 is naturally diffeomorphic to the projective line P1 .
With respect to the holomorphic resp. Kähler polarization on P1 , the polarized sections
of H(N ) are the holomorphic sections. In Proposition 3.22 it is shown, that the space
of holomorphic sections VN = Γhol (P1 , H(N )) is finite dimensional of dimension N + 1
for N ≥ 0 and 0 for N < 0. For the Kähler quantization this implies, that the space
of polarized sections of the line bundle H(N ) is VN ∼ = CN +1 . In particular, no special
integration is necessary, all Hermitian structures on VN are equivalent. And, once
q(F ) for F ∈ E(S2 ) can be implemented at all on VN (see next section) then q(F )
is self-adjoint. The reduced representation space is VN . This result agrees with the
spin- 21 -representations, including the dimension of the eigenspaces.
It can be shown, that on each HN an irreducible representation of SU(2) is induced,
thereby generating all irreducible representations of SU(2).

Examples 10.15. Now the door is open to investigate Kähler manifolds in general
and use the complex structure as well as the Kähler geometry as powerful tools. We
list some interesting cases:
1. Cn and all its open complex submanifolds.
2. M = T ∗ S1 ∼
= C∗
3. M = S1 × S1 and all other 2-dimensional compact oriented manifolds without
boundary.
10.3 Directly Quantizable Observables 151

4. P1 × P1 (Kepler problem) and other compact complex surfaces.

5. The projective spaces Pn introduced earlier and all its closed complex submani-
folds M ⊂ Pn . These are the so-called projective manifolds.

10.3 Directly Quantizable Observables

We come back to the case of a general complex polarization: As before, let (L, ∇, H) be
a prequantum bundle on a symplectic manifold (M, ω). Let P be a reducible complex
polarization P on M with its integrable distribution D = P ∩ P ∩ T M . Moreover, let
µ be a measure on M/D.
The question arises, for which F ∈ E(M ) the operator q(F ) leads to an operator in
the representation space HP . The following example illustrates the possible problems
for q(F ) becoming an operator in HP .
Example 10.16 (Harmonic Oscillator). It is rather a counterexample showing that
even with a natural measure on M/D a naive implementation of geometric quantization
is not always possible. Let M = T ∗ Rn be as before with the form ω = dq j ∧ dpj . The
energy of the harmonic oscillator is H = 21 (p2 + q 2 ), (q, p) ∈ M (we restrict to the case
n = 1 in the following).
We try to use the vertical distribution D with P = DC in order to quantize H.
In this case the quotient manifold M/D ∼ = R is the configuration space and has the
Lebesgue measure as a familiar measure.
The Hamiltonian vector field is
∂ ∂
XH = q −p .
∂p ∂q

The covariant derivative with respect to the potential −pdq = −λ reads

 
∂ ∂
∇XH f s1 = q −p f s1 − 2πip(−p)f s1 .
∂p ∂q

and the operator q(H) for ϕ ∈ HP is

 
1 ∂
q(H)ϕ = −p ϕ + p2 ϕ + Hϕ .
2πi ∂q

This will not be a function in HP , in general. A change in the potential (for instance
α = qdp) will not help.
As a result, although the vertical distribution looks good from the point of the
space of leaves M/D ∼= R, the energy H of the harmonic oscillator cannot be quantized
directly when the representation space is reduced to HP . To come around this problem
one can consider another polarization, e.g. the holomorphic polarization or the radial
152 10. Representation Space

polarization, see below, or one has to introduce other methods to incorporate classical
observables like H in the geometric quantization program, as will be done in later
chapters.

In order to verify whether the prequantum operator q(F ) leads to an operator in

HP the following fundamental question has to be answered: When s is a polarized
section of L, will q(F )s be polarized as well?
Definition 10.17. Let P be a complex polarization on a symplectic manifold. An
observable F ∈ E(M ) will be called Directly Quantizable (with respect to P ) if for
all X ∈ Γ(M, P ) the Lie bracket [XF , X] remains in Γ(M, P ). In another description:
adXF (X) := [XF , X] fulfills adXF (Γ(M, P )) ⊂ Γ(M, P ).
RP = RP (M ) denotes the set of all directly quantizable classical observables. Note,
that we consider complex-valued observables F although in physics only real-valued F
are relevant.
Lemma 10.18. RP is a Lie algebra with respect to the Poisson bracket on E(M ).

Proof. Let F, G ∈ RP and X ∈ Γ(M, P ). Directly from

[X{F,G} ], X] = [−[XF , XG ], X] = [[XG , X], XF ] + [[X, XF ], XG ]
one can read off that [X{F,G} , X] ∈ Γ(M, P ), hence {F, G} ∈ RP .
Remark 10.19. The condition adXF (Γ(M, P )) ⊂ Γ(M, P ) means that adXF preserves
the polarization. This invariance property has a nice interpretation using the Lie
derivative LX of a vector field Y . It is defined as
d
LX Y := ((Φ−t )∗ Y )|t=0 ,
dt
where Φt = ΦX
t is the local flow of the vector field X on M . In Proposition A.24 it is
shown that LX Y = [X, Y ], hence adX = LX .
Proposition 10.20. For a prequantum bundle (L, ∇, H) on a symplectic manifold
(M, ω) and a complex polarization P we have: When F ∈ RP and s ∈ Γ∇,P (M, L),
then q(F )s is polarized.

Proof. Let X ∈ Γ(M, P ). We have to show ∇X (q(F )s) = 0 for s ∈ Γ∇,P (M, L). Now
 
i
∇X (q(F )s) = ∇X − ∇XF s + F s
2π
i
=− (∇X ∇XF s) + ∇X (F s)
2π
i
= − (∇X ∇XF s) + LX F s + ∇X s
2π
i
= − (∇X ∇XF s) + LX F s
2π
10.4 Main Result 153

since s is polarized. By definition of the curvature

1

ω(X, Y ) = [∇X , ∇Y ] − ∇[X,Y ]
2πi
we obtain
i i i
− ∇X ∇XF s = − ∇XF ∇X s + ω(X, XF )s − ∇[X,XF ] s = ω(X, XF )s
2π 2π 2π
since ∇XF ∇X s = 0 and ∇[X,XF ] s = 0. We complete the above list of equations and
obtain

i
∇X (q(F )s) = − (∇X ∇XF s) + LX F s
2π
= +ω(X, XF )s + LX F s
= −LX f s + LX f s = 0

10.4 Main Result

As the main result so far we obtain the first complete geometric quantization: The set
of classical observables RP – defined in the preceding section – together with the other
ingredients yields a full geometric quantization scheme in the following sense:

Theorem 10.21. Let (L, ∇, H) be a prequantum bundle on a symplectic manifold

(M, ω) together with a reducible complex polarization P on M . Assume, that there
exists a measure µ on the quotient Q = M/D.51
Each F ∈ RP induces a quantum operator
i
q(F ) = − ∇XF + F
2π
in the representation space HP (defined in 10.5) such that the following results hold
true

1. RP is a Lie subalgebra of the Poisson algebra E(M ).

2. q(F ) ∈ S(HP )52 for F ∈ RP , and the quantization map q : RP → S(HP ) is

R-linear and satisfies (D1) and (D2) for o = RP .

3. If XF is complete, F ∈ RP , then q(F ) is self-adjoint.

51
The existence of such a measure is not needed, integrals can be taken with respect to a density
on M/D, see Chapter 12.
52
S(H) denotes the symmetric and densely defined operators on a Hilbert space H
154 10. Representation Space

Proof. By Proposition 10.20 the map q(F ) : Γ(M, L) → Γ(M, L) maps polarized sec-
tions to polarized sections. Hence, it is well-defined on DF := {s ∈ Γ∇,P (M, L) ∩ HP |
q(F )s ∈ HP }. DF is dense n HP since it contains the sections with compact support
in M/D.
Now 1. has been shown in Lemma 10.18. 2. Symmetry of q(F ) can be proved as
in the proof of Proposition 2.4. The Dirac conditions and 3. follow in the same way as
for the prequantization (cf. Chapter 7).
Proposition 10.22. Determination of RP in special cases of the simple phase space:
1. In case of the vertical distribution / polarization P on M = T ∗ Q , Q ⊂ Rn open:

RP = {A(q) + B j (q)pj | A, B j ∈ E(Q)} .

2. Analogously, for the horizontal polarization P on M = T ∗ Rn , swapping the roles

of q and p.
3. In case of the holomorphic polarization P on M = T ∗ Rn = Cn :

RP = {A(z) + B j (z)z̄j | A, B j ∈ O(Cn )} .

Proof. 1. It is enough to restrict to local coordinates q j of Q and to investigate

∂
[ , XF ].
∂pj
Recall that
∂F ∂ ∂F ∂
XF = j
− .
∂q ∂pj ∂pj ∂q j
The two equations
∂ 2F ∂
 
∂ ∂F ∂ ∂F ∂ ∂ ∂F ∂ ∂
, k = + k − k
∂pj ∂q ∂pk ∂pj ∂qk ∂pk ∂q ∂pj ∂pk ∂q ∂pk ∂pj
and
∂ 2F ∂
 
∂ ∂F ∂ ∂F ∂ ∂ ∂F ∂ ∂
,− k
= − − k
+
∂pj ∂pk ∂q ∂pj ∂pk ∂qk ∂pk ∂pj ∂q ∂pk ∂q k ∂pj
show that
∂
[ , XF ] ∈ P
∂pj
if and only if
∂ 2F
= 0.
∂pj ∂pk
Hence, F ∈ RP if and only if
∂ 2F
=0
∂pj ∂pk
10.4 Main Result 155

for all 1 ≤ j, k ≤ n. As a consequence, F is of the form A(q) + B j (q)pj with suitable

A, B j ∈ E(Q).
2. and 3. are shown in the same way.

The last result in the preceding proposition asserts that the energy H = 12 zj z̄j ∈
P
E(Cn ) of the harmonic oscillator is a directly quantizable observable with respect to
the holomorphic polarization P :
H ∈ RP .
We determine the quantum operator q(H) in the next example.

Example 10.23 (Harmonic Oscillator). This example can be viewed as a continuation

of Example 10.9. Consider the Hamiltonian
n
1X 2 1X
H= pj + (q j )2 = zk z̄k
2 2 k=1

of the harmonic oscillator with phase space M = T ∗ Rn ∼

= Cn . The Hamiltonian vector
field
∂ ∂
XH = q j − pj j
∂pj ∂q
does not fit to the vertical distribution as we have seen before. Hence, it is reasonable
to use the holomorphic polarization P with respect to the holomorphic coordinates
zj = pj + iq j (No summation!). P is the polarization generated by the vector fields
Xzk :  
∂
P = span .
∂ z̄j
Using (44) or the above form of XH we obtain
X 
∂ ∂
XH = i zj − z̄j .
∂zj ∂ z̄j

With α = 2i j z̄j dzj and ω = i

P P
2
z̄j ∧ dzj as before one has for holomorphic
functions f : Cn → C:
!
i 1 X ∂ iX
− ∇XH f s1 = − i zj f + 2πi z̄j izj f s1
2π 2πi j
∂z j 2 j
1 X ∂ 1X
= zj f s1 − zj z̄j f s1 .
2π j ∂zj 2 j

Thus, the geometric quantization for the holomorphic polarization P yields

156 10. Representation Space

n
1 X ∂
q(H) = zk
2π k=1 ∂zk
– essentially the Euler Operator – as the quantized Hamiltonian H on the
Bargmann space HP .
The eigenvalues E will be determined by the equation
1 X ∂
zk ϕ = Eϕ .
2π ∂zk

Claim: For an entire holomorphic function f : Cn → C this equality can be satisfied

1
if and only if f is a N -homogeneous complex polynomial and E = EN = 2π P N , N ∈ N.
To show this claim let us recall, that f has the power series expansion f = PN f with
suitable N -homogeneous polynomials PN converging uniformly on all compact subsets
of Cn to f (cf. Proposition B.5). PN f has the form
X 1 ∂ j1 +···+jn f (0) j1
P N f (z) = z · · · znjn .
j1 +...+jn =N
j1 ! · · · jn ! ∂z1 j1 · · · ∂zn jn 1

For a monomial cµ1 µ2 ...µn z1µ1 z2µ2 . . . znµn (no summation!) of degree N = µ1 +µ2 +. . .+µn
it is clear that
X ∂
zk cµ1 ...µn z1µ1 . . . znµn = (µ1 + . . . + µn ) cµ1 ...µn z1µ1 . . . znµn = N cµ1 ...µn z1µ1 . . . znµn ,
k
∂z k

hence,
X ∂
zk (PN f ) = N PN f
k
∂zk
and eventually
X ∂ XX ∂ X
zk f= zk PN f = N PN f .
k
∂zk N k
∂zk N

The claim follows from the last equality. So the eigenvalues of our operator q(H) on HP
1
are EN = 2π N and the eigenspaces VN are the spaces of N -homogenous polynomials
(note, that the homogeneous complex polynomials in N variables
L are in the domain of
q(H)). And we obtain a complete decomposition HP = VN into eigenspaces. By
the way, this shows that q(H) is self-adjoint.
From the physical side this result is not correct. The observed eigenvalues are
1
2π
(N + n/2) instead. So the term n2 is missing, which is related to the zero point
energy.
By comparison, we see that as a correct quantized operator q(H) one should take
!
1 X ∂ n
q(H) := zj + . (45)
2π j
∂zj 2
10.5 Sketch of Correction Scheme 157

This can be achieved by replacing L with the line bundle L ⊗ S where S → M is a

geometrically induced complex line bundle over M respecting the symplectic geometry
of (M, ω) and the polarization (see Chapter 15 and the correction scheme in the next
section).
Let us recall how in conventional canonical quantization thePquantum operator (45)
1
can be obtained. The energy is written in the form H = 4 j (zj z̄j + z̄j zj ) and the
quantum operators
1 ∂
q(zj ) = zj , q(z̄j ) =
π ∂zj
(cf. Example 10.9) are used by replacing the classical coordinates zj , z̄j directly. As a
result we get for f ∈ O(Cn ):

    !
1 X ∂ ∂ 1 X ∂ 1 X ∂ n
q(H)f = zj + zj f = 2zj f +f = zj + f.
4π j ∂zj ∂zj 4π j ∂zj 2π j
∂zj 2

Evidently, this ”conventional correction” has to do with operator ordering.

Remark 10.24. Let us consider the simple phase space T ∗ Rn ∼ = Cn with the
holomorphic polarization P . The directly quantizable observable are the functions
A(z) + B j (z)z̄j with A, B j holomorphic. Hence, the real-valued directly quantizable
observables are the functions
F (z, z̄) = A + D̄j zj + Dj z̄j + sumC j zj z̄j ,
where A, , C j are real constants and Dj are complex constants. We conclude from the
calculatios of the preceding examples that we know the quantum operator q(F ) for all
real F ∈ RP :
1 ∂
q(A) = A idH , q(D̄k zk + Dk z̄k ) = D̄k zk + Dk
π ∂zk
and
X 1X j ∂
q( C j zj z̄j ) = C zj .
π ∂zj

10.5 Sketch of Correction Scheme

Let (L, ∇, H) a prequantum bundle on a symplectic manifold (M, ω) with a complex

Kähler polarization P . The ”polar” P 0 := {µ ∈ T ∗ M C | µ|P = 0} is a complex vector
bundle of rank n over M , subbundle of the complexified cotangent bundle T ∗ M C of
M.
The n-fold wedge product Λn P 0 is a complex line bundle which is called the Canon-
ical Line Bundle53 of P and denoted by KP .
53
For general complex vector bundles V an alternative definition is used: The canonical bundle is
K(V ) := Λr (V ∨ ),where r is the rank of the bundle V (see 15.1) and 15.7.
158 10. Representation Space

Let P ′ be another complex polarization on M with the condition P ∩ P ′ = {0} (for

instance P ′ = P in case of a Kähler polarization). The line bundle KP ⊗KP ′ is naturally
isomorphic to the complex line bundle Λ2n T ∗ M C over M . This line bundle is trivial
having ω n as a nowhere vanishing section. For sections α ∈ Γ(M, KP ) , β ∈ Γ(M, KP ′ )
the wedge ᾱ∧β is a scalar multiple of ω n : ⟨α, β⟩ω n = ᾱ∧β, which provides a sesquilinear
pairing (α, β) 7→ ⟨α, β⟩.
KP has the natural (partial) connection ∇X α = iX α, X ∈ Γ(M, P ), which is flat.54
Now, let us assume that KP has a square root S in the sense that for a line bundle
S over M there exists an isomorphism KP ∼ = S ⊗ S. (The existence of such a line
1/2
bundle S depends on a topological property of M .) We set KP := S and denote the
corresponding induced (partial) connection by ∇1/2 .
Our prequantum bundle will be replaced by
1/2 1 1
Lcorr = L ⊗ KP , ∇corr := ∇ + ∇ /2 = ∇ ⊗ 1 + 1 ⊗ ∇ /2 .

1/2
The polarized sections are the sections s ⊗ α ∈ Γ(M, L ⊗ KP ) satisfying
1
∇ + ∇ /2 (s ⊗ α) = 0

which is the same as ∇s = 0 and ∇1/2 α = 0.

The representation space HP is the completion of the space of polarized sections
1/2
s ⊗ α ∈ Γ(M, L ⊗ KP ) with respect to the scalar product by
Z Z
n
H(s, s)⟨α, α⟩ω = H(s, s)ᾱ ∧ α < ∞.

The quantum operator is defined by

i  1/2
 i 1/2
q corr (F )(s ⊗ α) := − ∇X s ⊗ α + s ⊗ LXF α + F = q(F )s ⊗ α − s ⊗ LXF α
2π 2π
1/2
for directly quantizable observables F ∈ E(M ), where s ⊗ α ∈ Γ(M, L ⊗ KP ) are
polarized sections. In this way HP and q corr (F ) yields a full geometric quantization.
This modification is called half-form correction and can be extended to more general
cases with P ∩ P ̸= {0}. This will be the subject of Chapter 15.

Example 10.25. Applied to the harmonic oscillator on M = T ∗ R = C with Hamil-

tonian H(z) = 21 |z|2 and holomorphic polarization P we get the following: The po-
larized sections of the (trivial) line bundle L are essentially the holomorphic functions
1/2
f : C → C. KP is trivial and generated by dz. KP is generated by a section denoted by
54
See Chapter 15
10.5 Sketch of Correction Scheme 159

∂ 1/2
dz 1/2 . From XH = i(z ∂z − z̄ ∂∂z̄ ) we deduce LXH dz = idz and from this LXH dz 1/2 = 2i dz 1/2
using
1 1/2 1
LXH dz = 2dz /2 LXH dz /2 .
1 d
With the result q(H) = 2π z dz (from Example 10.23) and inserting the last formula
in q (H) the quantum operator q corr (H) we obtain
corr

1 1 d 1 i i 1 1 d 1 1
q corr (H)(f ⊗ dz /2 ) = z f ⊗ dz /2 − f ⊗ dz /2 = (z + )(f ⊗ dz /2 ) .
2π dz 2π 2 2π dz 2
Hence q corr (H) acts on the space O(C) of holomorphic functions f : C → C in the
following way
 
corr 1 d 1
q (H) := z + . (46)
2π dz 2

Therefore, the corrected quantum operator yields the right eigenvalues with the
right multiplicities.

Note, that the very sketchy description of the representation space HP is not
needed for the calculation of the quantum operator q corr (H) acting on the space
ΓP (T ∗ R, Lcorr ) ∼
= O(C). We have mentioned the definition of the Hilbert space HP ,
since in half-density and half-form quatization it is important to obtain general and
corrected representation spaces and since it opens the way to pairing the spaces HP
and HP ′ for different polarizations.

Summary: By implementing a complex polarization P as an additional geometric

data complementing the prequantum bundle on the symplectic manifold the reduced
representation space HP can now be constructed as the completion of the space of po-
larized and square integrable sections. To complete the first basic version of Geometric
Quantization the concept of a directly quantizable observable F has to be introduced in
order to confirm that for a polarized section s the derivative ∇XF s is polarized as well.
For such F it is easy to define the quantum operator q(F ) in the new representation
space HP to achieve a full geometric quantization.
Several elementary examples are presented in detail in this chapterbb in order to
illustrate the action of Geometric Quantization. In particular, for the simple case of
M = T ∗ Rn three different polarizations yield the well-known Schrödinger, Heisenberg
and Bargmann representations. Also, the quantization of the harmonic oscillator is
described, leading to a slightly incorrect result. A sketch of a correction is added
which is presented in greater generality in Chapter 15.
In many situations the first version of Geometric Quantization developed so far is
not satisfying. Nevertheless, the previous 10 chapters cover a rather complete picture
of the basic principles of Geometric Quantization and they describe essentially the
content of the course given in winter 2021/22.
160 10. Representation Space

Among the weaknesses are, for example, that the representation space can be zero
since there do not exist non-trivial polarized sections, a phenomenon which we treat in
the next chapter. Moreover, eigenvalues are not correct as in the case of the harmonic
oscillator. In addition, there is a need the quantize many more classical observables
than merely the directly quantizable ones. As a consequence one has to modify or
complement the basic Geometric Quantization presented in the first 10 chapters. This
is the subject of the next chapters.
 161

11 Existence of Polarized Sections and Holonomy

So far we have seen polarizations in action only for the following two cases: For the
Kähler polarizations and for the vertical and the horizontal polarizations of the simple
phase space M = T ∗ Rn . In these two cases it was no question of whether or not there
exist polarized sections at all (except possibly for the compact Kähler manifolds as
phase spaces).

11.1 Bohr-Sommerfeld Variety

The following example exhibits a general obstacle to the existence of global polarized
sections different from zero55 .

Example 11.1. Let M = T ∗ S the phase space with the circle S as configuration space
(cf. Example 8.22) and with the Horizontal Polarization. The cotangent bundle
T ∗ S = M is trivial and we write M = S × R. As before, as prequantum line bundle
we take the trivial bundle L = M × C with connection ∇ given by the Liouville form
λ = pdq, i.e.
∇f s1 = (df + 2πi(−λ)f ) s1 ,
where s1 (a) = (a, 1) , a ∈ M and f ∈ E(M ), and H induced from L = M × C.
Differently from the previous discussion of this example in 8.22 we now study the
horizontal polarization instead of the vertical polarization. The horizontal polarization
P is generated by the vector field
∂
,
∂q
where q is essentially the angle variable. Consequently a general section s = f s1 is
polarized if LX f − 2πiλ(X)f = 0 for all X ∈ Γ(M, P ), i.e. if

∂
f = 2πipf
∂q

in local coordinates. This differential equation has the general solution

f (q, p) = g(p) exp 2πipq , (47)

with an arbitrary function g ∈ E(R). Moreover, f has to be periodic in the variable q

with period 1. Therefore, only for p ∈ Z there is a non-zero solution of this equation,
depending on (q, p) ∈ S × {p} for fixed p ∈ Z. It follows, by the continuity of f , that
f = 0. Altogether there do not exist nontrivial global polarized sections.
55
If one asks for local sections which are polarized: In the situation of a general complex polarization,
there exist non-zero local sections which are polarized. We do not need this result, we are interested
in this section only in global polarized sections.
162 11. Existence of Polarized Sections and Holonomy

There is another way to arrive at this result which exhibits a general pattern behind
this result and which generalizes to arbitrary reducible polarizations: The submanifolds
S × {p} =: Sp , p ∈ R, are the leaves (integral manifolds) of the horizontal distribution.
The restriction ∇|Sp of ∇ to a leave Sp is a flat connection on the line bundle L|Sp → Sp ,
the restriction of L to Sp . Let a = (1, p) ∈ Sp . Parallel transport Q(γ) : La → La , v 7→
Q(γ)v, along the curve γ(t) = (exp 2πit, p) , t ∈ [0, 1], is given by the integral (cf.
Proposition 5.17)
 Z   Z 1 
Q(γ) = exp 2πi pdq = exp 2πip dt = exp(2πip) . (48)
γ 0

Assume that s ∈ Γ(M, L) is a global polarized section. Then the restriction of s to Sp

is a horizontal section Sp → L|Sp and therefore determines the parallel transport given
by ∇|Sp . In particular, if s(a) ̸= 0, the parallel transport is s(γ(0)) 7→ s(γ(1)) with
γ(0) = γ(1) = a and therefore it is trivial in the sense that it is the identity La → La .
This implies Q(γ) = 1. But then it follows from (48) that p ∈ Z. As a result, s has to
be 0 outside the so called Bohr-Sommerfeld Variety
[
S := {Sp | p ∈ Z} = S × Z ,

which implies that s = 0 by continuity.

With respect to a different connection ∇, given by the one form (κ−p)dq , κ ∈ ]0, 1[
(see Example 8.22), one obtains essentially the same result. The corresponding parallel
transport is given by the integral (cf. Proposition 5.17)
 Z 
κ
Q (γ) = exp 2πi (p − κ)dq = exp(2πi(p − κ) . (49)
γ
S
A global polarized section s on M = S × R has to be zero outside of S := {Sp |
p − κ ∈ Z}, and hence is the zero section.
The above considerations generalize directly to the case M = T ∗ T where T = (S)n
is the n-dimensional torus, n > 0.

It turns out that the non-existence of polarized sections is not a singular phe-
nomenon of this example but rather is a general property whenever the leaves of the
induced distribution are not simply connected. Before we present this result in the
next proposition below, we want to give a different interpretation of the above example
Remark 11.2 (Energy Represention). Consider the example M := R2 \ {(0, 0)} ⊂
T ∗R ∼
= R × R with the usual symplectic form ω = dq ∧ dp. Let the prequantization line
bundle be L = M × C with connection ∇ given by the connection form −pdq = −λ as
before.
Now, let D be the radial distribution given by the circles p2 + q 2 = 2E , E > 0,
i.e. the distribution with the circles around (0, 0) as its leaves. This distribution can
11.1 Bohr-Sommerfeld Variety 163

be understood as the distribution generated by the energy function of the harmonic

oscillator
1
H = (p2 + q 2 ) .
2
In fact, D is generated by the Hamiltonian vector field
∂ ∂
XH = p −q .
∂q ∂p
Let P = DC be the corresponding complex polarization.
There is a natural embedding
1
Φ : M → T ∗ S = S × R , (q, p) = p + iq = re2πit 7→ (e2πit , r2 ) ,
2
with Φ(M ) = {(q, p) ∈ T ∗ S = S × R | p > 0} = S × R+ . The map Φ preserves
the symplectic forms, the polarizations and the connections. Hence, the results of the
preceding example show that there are no nonzero polarized sections on M .

As mentioned before we can transfer the arguments of the example to the case
of general complex reducible distributions P : There exist severe restrictions to the
existence of polarized sections, if the leaves of the induced distribution D = P ∩P ∩T M
are not simply connected.
Holonomy Group
To explain this general result, we first recall some facts about parallel transport of
a connection ∇ on a line bundle L over a general connected manifold X: Given a point
a ∈ X and a closed curve γ in X starting and ending in a the parallel transport along
γ is an isomorphism Q(γ) : La → La given by a complex number which we denote with
Q(γ) (cf. Proposition 5.17). This number can be expressed as the integral
 Z 
Q(γ) = exp −2πi A ,
γ

where A is a local connection form of the connection. We have Q(γ) ∈ U(1).

The collection of all these Q(γ) forms a group G(a), a subgroup of U(1). Note, that
G(a) is a quotient of the loop group L(a), see Section 5.4. This group G(a) is called
the Holonomy Group of the connection at a. Since we assume X to be connected
the holonomy groups are isomorphic to each other: G(a) ∼ = G(b) for a, b ∈ X. In fact,
they are even conjugate subgroups of U(1).
We have seen already in Observation 8.23, that in the case of a flat connection ∇
we obtain a natural group homomorphism

Hol∇ : π1 (X) → G(a) , Hol∇ ([γ]) := Q(γ) ,

since parallel transport is locally independent. Hol∇ is surjective by definition of G(a).

164 11. Existence of Polarized Sections and Holonomy

Observation 11.3. Let ∇ be a flat connection on the line bundle L → X and let
X be simply connected. Then G(a) is the trivial group {1} for all a ∈ X, since the
fundamental group π1 (X) of X is trivial and Hol∇ is surjective. This can also be
seen directly by using the fact that each closed curve γ in X can be contracted to the
constant curve γ0 (t) = a and that the parallel transport La → La along γ and along
γ0 coincide since the parallel transport of a flat connection is locally independent.

We come now to the existence of global polarized sections in the general case:
Let (M, ω) be a symplectic manifold with a prequantum line bundle (L, ∇, H). Let
P a reducible complex polarization, i.e. P is a complex polarization such that with
D = P ∩ P ∩ T M the quotient manifold M/D exists and the projection π : M → M/D
is a submersion.
We fix a leaf (integral manifold) Λ ⊂ M of the distribution D, i.e. Λ = π −1 (x) for a
suitable x ∈ M/D. The restriction of the connection ∇ to Λ induces a flat connection
∇|Λ on the line bundle L|Λ → Λ.

Definition 11.4 (Bohr-Sommerfeld Variety). Let P be a reducible complex polariza-

tion and D its induced real distribution. For each leaf Λ of D and a ∈ Λ let GΛ (a)
denote the holonomy group of the restricted connection ∇|Λ on the restriction L|Λ → Λ
of the line bundle L. Then
[
S := {a ∈ M | GΛ (a) = {1}}

denotes the Bohr-Sommerfeld Variety.

In particular, let Λ be a leaf which is simply connected. Since the connection ∇|Λ
is flat, the holonomy group GΛ (a) has to be trivial: GΛ (a) = {1} which implies Λ ⊂ S.
As a result we obtain the inclusion

{a ∈ M | Λ(a) is simply connected} ⊂ S .

where Λ(a) denotes the leaf through a: a ∈ Λ(a). It follows S = M , when all the leaves
Λ are simply connected. However, the condition GΛ (a) = {1} does not imply that Λ
is simply connected as the Example 11.1 shows.
With the same arguments as in the above Example 11.1 we can deduce the following
result.

Proposition 11.5. Any polarized smooth section s ∈ Γ(M, L) vanishes outside the
Bohr-Sommerfeld variety.

Proof. Let s be a polarized section and a ∈ M with s(a) ̸= 0. Let Λ be the leaf of
the distribution D with a ∈ Λ. Then s|Λ ∈ Γ(Λ, L|Λ ) is a horizontal section of the
restriction ∇|Λ and thus determines the parallel transport La → Lb of ∇|Λ over Λ.
Recall that this means the following. If γ : [0, 1] → Λ is a curve in Λ starting in a and
11.2 Distributional Sections 165

ending in b ∈ Λ, the parallel transport of s(a) = s(γ(0)) ∈ La along γ is determined as

s(γ(1)) = s(b) ∈ Lb . In case of a = b which we consider here, the parallel transport is
the identity, hence Q(γ) = 1. We conclude that the group GΛ (a) is trivial and therefore
a ∈ S.

As a consequence of this result, for a general reducible complex polarization there

exist non-zero polarized global sections only if the Bohr-Sommerfeld variety has a non-
empty interior. For instance, if all leaves are simply connected. Therefore, in many
cases the space of global polarized sections is zero, and cannot be used to determine a
meaningful representation space.

11.2 Distributional Sections

To overcome this difficulty one studies generalized sections, which could be defined in
the same way as distributions56 .
Let us describe the introduction of distributional sections in the case of the example
of the cylinder.

Example 11.6 (Continuation of Example 11.1). The differential equation (47) allows
the solutions
n
ϕn (q) := exp(2πiqn) = e2πiq , q ∈ [0, 1[ ,
defined on Sn = S × {n}. For fixed n ∈ Z these solutions are unique up to a multiplica-
tive complex constant. We consider the Hilbert space ℓ2 (Z) := {(zn ) ∈ CZ | |zn |2 <
P
2
∞} and understand an element (zn ) ∈ ℓ (Z) as a generalized function
X
zn ϕn ,

which is zero outside of the Bohr-Sommerfeld variety S and where zn ϕn represents the
corresponding polarized section in Γ(Sn , LSn ) on the circle Sn . In this interpretation
we denote the Hilbert space by HP = ℓ2 (Z) (with respect to the horizontal polarization
P ) and we regard HP as the representation space of the model.
The height function h := pr2 : S × R → R , (q, p) → p, is a classical observable
∂
whose Hamiltonian vector field Xh = ∂q generates the horizontal distribution. As a
consequence, the operator ∇Xh vanishes on polarized sections and the prequantum
operator ĥ := q(h) is simply the multiplication ϕ 7→ hϕ on the space of polarized
sections. With respect to the representation space HP this implies that ĥ defines a
self-adjoint operator in HP with domain of definition
X
D(ĥ) := {(zn ) ∈ ℓ2 (Z) | n2 |zn |2 < ∞} ,
56
in the sense of generalized functions which are linear functionals, not to mix up with the notion
of a distribution in the geometrical sense as a subbundle of the tangent bundle
166 11. Existence of Polarized Sections and Holonomy

P P
by ĥ((zn )) = h(q, n)zn ϕn = nzn ϕn = (nzn ). Therefore, the representation space
HP decomposes into one dimensional eigenspaces En := Cϕn = Ker (ĥ − n) with
eigenvalues n: ĥ(ϕn ) = nϕn .
Remark 11.7 (Continuation of Remark 11.2). Applied to the Example in Remark 11.2
(energy representation) the introduction of distributional sections leads to the Hilbert
space HP = ℓ2 (N+ ), with N+ := {n ∈ N | n > 0}. In this situation, ϕn = en , n ∈ N+
stands for a special distributional polarized section with support in the circle C√2n :=
√ √
H −1 (n)P= {(q, p) | q 2 + p2 = ( 2n)2 } ⊂ M = T ∗ R× of radius 2n. And each
(zn ) = zn ϕn ∈ ℓ2 is a sum of such sections. Notice, that H = h ◦ Φ. Moreover, since
XH generates the horizontal distribution P in M = T ∗ R× , the program of geometric
quantization yields the prequantum operator q(H) in HP as the multiplication operator
ϕ 7→ Hϕ as before. This multiplication operator, with H(q, p) = n for (q, p) ∈ C√2n ,
i.e. q 2 + p2 = 2n, becomes a densely defined self-adjoint operator q(H) in ℓ2 (N+ ) = HP :
X
q(H) : (zn ) 7→ nzn ϕn = (nzn ).

The eigenspaces of q(H) are C{ϕn }, n ∈ N+ , with eigenvalues n ∈ N+ . In particular,

the eigenspaces of q(H) are one dimensional which is in accordance with the known
quantum mechanical model, but the eigenvalues are again not correct with the same
defect as in the previous Example 10.23. The eigenvalues should be n − 12 instead of
n. This can be corrected with the use of half form quantization as is described in the
preceding Section 10.5.
However, another correction is possible, which is closer in spirit to the geometry of
the Bohr-Sommerfeld condition by choosing an alternative connection for the prequan-
tum bundle. We have seen in Example 8.22 that the non-vanishing of Ȟ 1 (T ∗ S, U(1))
allows non equivalent connection forms −λ + κdq depending on the real parameter
κ ∈ ]0, 1]. With respect to the new connection ∇κ over T ∗ S and the corresponding
new prequantum bundle there appears a shift by −κ in the formation of the Bohr-
Sommerfeld variety. Indeed, by (49) it follows as in the case of κ = 0 that any global
S s on M = S × R has to be zero outside of the new Bohr-Sommerfeld
polarized section
variety S := {Sp | p − κ ∈ Z} ⊂ T ∗ S.
The result of the correction in case of κ = 12 now reads as follows: The representation
space is essentially the same as above, it is ℓ2 (N+ ), but now with the interpretation
that the polarized sections ϕn have their support in the circles C√2n−1 := H −1 (n − 21 ).
As a result, since H = n − 21 on C√2n−1 , we obtain
    
1 1
q(H)(ϕn ) = n − ϕn , or q(H)(zn ) = n− zn
2 2
for n ∈ N+ . This result leads again to a decomposition of the representation space HP
into one dimensional eigenspaces En , now with the correct spectrum
1
σ(q(H)) = {n − | n ∈ N+ }.
2
11.2 Distributional Sections 167

Summary:
168 12. Densities and Their Derivatives

12 Densities and Their Derivatives

The search for an appropriate representation space is not over. We have introduced
complex polarizations P ⊂ T M C on a symplectic manifold (M, ω) in Chapter 9 in order
to reduce the prequantum Hilbert space but we have seen the effect of this reduction
in the preceding chapter so far only in case of the vertical resp. horizontal polarization
on the simple phase space M = T ∗ Rn (simple phase space) and in case of a Kähler
polarization, i.e. a purely complex polarization.
In the following we use the concept of densities on a polarization P in order to
obtain a representation space also for the non-Kähler case. In that case the newly
constructed representation space is essentially a space of half-densities defined on the
quotient manifold M/D, where D = P ∩ P ∩ T M .
We have decided to include in these Lecture Notes a rather detailed exposition of
densities and of the application of densities to geometric quantization, although this
application, the so-called half-density quantization, leaves many problems of geometric
quantization open. One reason for this decision is, that the new structure of r-densities
on a vector bundle deserves an extra attention, and in the literature the treatment is
mostly rather short. Another reason for presenting half-density quantization in detail is,
that it can be understood as a preparation for the more involved half-form quantization
which we develop later in Chapter 15.
The basic idea of half-density quantization for a quantizable symplectic manifold
(M, ω) with prequantum bundle (L, ∇, H) and reducible polarization P is the following:
In order to find a scalar product for polarized sections s, s′ ∈ Γ(M, L) which we use in
order to construct the representation space HP we observe that the expression H(s, s′ )
is a function on M which is constant on the leaves of the distribution D (recall: D =
P ∩ P ∩ T M ). Hence, one could try to integrate H(s, s′ ) over the quotient manifold
M/D. However, there is no natural measure or volume form on M/D. So 1-densities on
M/D come into play since a 1-density on any manifold X can be naturally integrated
over X, as we explain in Section 12.2. To use this fact, the expression H(s, s′ ), which
can be viewed as to be a function on M/D, has to be transformed into a 1-density on
M/D. This can be done by altering the original approach of geometric quantization in
such a way that

• the line bundle L will be replaced by a line bundle L ⊗ δ where δ → M is a

suitable line bundle of densities induced by P 57 which descend to half-densities
on M/D and
• the connection ∇ will be replaced by ∇ ⊗ ∇δ , with a flat (partial) connection ∇δ
on δ (see Section 12.3).

If this is done properly, one considers the space HP of sections ψ = s ⊗ ρ of L ⊗ δ

of compact support, which are polarized, i.e. which satisfy (∇X ⊗ ∇δX )(s ⊗ ρ) = 0 for
57
more precisly, δ is the line bundle δ−1/2 (P ) of −1/2-densities on P .
12.1 Densities 169

all X ∈ Γ(M, P ). For two such sections ψ = s ⊗ ρ, ψ ′ = s′ ⊗ ρ′ the quantity H(s, s′ )ρ̄ρ′
descends to a unique 1-density on M/D. Integrating this 1-density yields a scalar
product on HP , and the completion of HP with respect to the induced norm is the
representation space HδP = Hδ (M, L, P ) one is looking for.
In this chapter a detailed exposition of the concept of a general r-density is pre-
sented. We treat the integration of a 1-density on a manifold and we introduce the
partial connection as well as the partial Lie derivative of r-densities on a polarization
P as a preparation for the half-density quantization in the following to chapters.

12.1 Densities

Let X be an m-dimensional manifold and let π : V → X be a complex vector bundle

of rank k. (In our intended application the manifold X is, in one of the situations, the
space M/D of leaves where D = P ∩ P ∩ T M is the real distribution induced by the
complex polarization P and V is the complexified tangent bundle T X C , and in another
situation X = M and V is the polarization P ⊂ T M C , the distribution DC , or the
quotient bundle Z D = T M C /DC .)
The vector bundle V induces a frame bundle R(V ) → X (see Construction D.4):
R(V ) is a principal fibre bundle over X with structure group G = GL(k, C) whose fibre
Rx (V ) over a point x ∈ X is the set of all ordered bases (frames) of the complex vector
space Vx = π −1 (x), and whose transition functions are the gij , the transition functions
determining the vector bundle V . The right action of G = GL(k, C) on R(V )

R(V ) × G → R(V )

is given by
(bg)α := bβ gαβ , (1 ≤ α, β ≤ k) ,
for a matrix g = (gαβ ) ∈ G and for b = (b1 , b2 , . . . , bk ) ∈ Rx (V ). Then bg =
((bg)1 , . . . , (bg)k ) ∈ Rx (V ) is another frame of Vx . Once a frame b ∈ Rx (V ) at x ∈ X
has been chosen, the map G → Rx (V ), g 7→ bg, is bijective.
To motivate the notion of a density let us consider the top differential forms on
a manifold X of dimension m. Any complex m-form η ∈ Am (X) is a section of
the complex line bundle Λm (T ∗ X C ) and it can be evaluated at a complex frame b ∈
Rx (T X C ) , x ∈ X, to yield the value ηx (b) ∈ C. In this sense η induces a (smooth) map
η ♯ : R(T X C ) → C , b → ηx (b). This map η ♯ has the following transformation property:

η ♯ (bg) = (det g)η ♯ (b) , b ∈ R(T X C ) , g ∈ GL(m, C) . (50)

Conversely, a map w : R(T X C ) → C with

w(bg) = (det g)w(b) , b ∈ R(T X C ), g ∈ GL(m, C) ,

defines an m-form η on X with η ♯ = w.

170 12. Densities and Their Derivatives

Definition 12.1 (Density). Let π : V → X be a complex vector bundle of rank k

and let r be a real number. A Density58 of weight r, also called r-density, on V is a
function
u : R(V ) → C ,
u ∈ E(R(V ), C)59 , which transforms under GL(k, C) according to

u(bg) = | det g|r u(b) , g ∈ GL(k, C) , b ∈ R(V ) .

The E(M )-module of r-densities u : R(V ) → C on V will be denoted by ∆r (V )60 .

An r-density on T X C is also called an r-density on X and we write ∆r (X) :=
∆r (T X C ).

A slightly different but equivalent definition in case of r = 1 will be given at the be-
ginning of the following Section 12.2 where the integration of 1-densities is introduced.

k ∨ ∼ Vk example is the 1-density∨ |η| on V induced by a k-form η ∈

A straightforward
Γ(M, Λ (V )) = (Γ(M, V ), E(M )), where V is the dual bundle of V and where
k ∨
Λ (V ) is the vector bundle of k-linear alternating forms on V . For an alternating and
k-linear η : Γ(M, V )k → E(M ) and a basis (b1 , b2 , . . . bk ) of Vx , x ∈ X, one sets

|η|(b1 , b2 , . . . , bk ) := |ηx (b1 , b2 , . . . , bk )| .

Then |η| is a 1-density on V because of ηx ((bg)1 , . . . , (bg)k ) = (det g) ηx (b1 , b2 , . . . , bk ),

and thus
|η|(bg) = |η(bg)| = | det gη(b)| = | det g||η|(b) .

This example will be used, in particular, in the case of V = T X C where

Γ(M, Λk (T ∗ X C )) ∼
= Ak (X) (k = m = dim M ) is the E(X)-module of differential k-
forms.
Moreover, |η|r is an r-density on V for r ∈ R.
Densities can be multiplied, conjugated, dualized, and more. These properties of
densities can be best understood by transforming densities into sections of suitable line
bundles.

Definition 12.2. Let π : V → X be a complex vector bundle and let r be a real

number. The line bundle R(V )×ρr C over X associated to the frame bundle R(V ) → X
with respect to the representation ρr : GL(k, C) → GL(1, C) = C× , ρr (g) := | det g|−r ,
is called the bundle of r-densities on V and denoted by δr (V ).
58
in physics a density is sometimes called a pseudoscalar
59
In the following, we mostly omit C in expressions like E(X, C) and write E(X), as before.
60
This definition describes complex densities. In the same way one defines real r-densities on a real
vector bundle. Moreover, by choosing X to be the manifold consisting of one point, we obtain the
definition of an r-density on a vector space.
12.1 Densities 171

The concept of an associated vector bundle is explained in Section D.3. Recall, that
the line bundle R(V ) ×ρr C is the quotient of R(V ) × C with respect to the equivalence
relation
(b, z) ∼ (bg, | det g|r z) = (bg, ρr (g −1 )z) , for g ∈ GL(k, C) ,
where (b, z) ∈ R(V ) × C.
Remark 12.3. In particular, by Proposition D.7, we know that the sections s ∈
Γ(U, δr (V )) on an open subset U ⊂ X are equivalently described by the functions
u : R(V )U → C transforming according to
u(bg) = | det g|r u(b) ,
b ∈ R(V )U , g ∈ GL(k, C). Hence, ∆r (V ), the E(M )-module of r-densities on V , can
be identified with the E(M )-module of global sections σ ∈ Γ(X, δr (V )) of the line
bundle δr (V ), and it is justified to call δr (V ) the line bundle of r-densities. To simplify
formulas, we often write Γr (V ) instead of Γ(X, δr (V )) in the following.
We recall the isomorphism ∆r (V ) ∼ = Γr (V ) = Γ(X, δr (V )) in detail by using the
construction of the associated line bundle: Every u ∈ ∆r (V ) induces a section
su : X → δr (V ) by su (a) := [(b, u(b))] , b ∈ Ra (V ) , a ∈ X.
The map su is well-defined: The elements of the fibre π −1 (a) of the line bundle π :
R(V ) ×ρr C → X over a ∈ X are the equivalence classes [(b, z)] of the form [(b, z)] =
{(bg, | det g|r z) | g ∈ GL(k, C)} with b ∈ Ra (V ). Hence,
(bg, u(bg)) = (bg, | det g|r u(b)) ∼ (b, u(b)) ,
which shows that the assignment su (a) = [(b, s(u))] is independent of the choice of
b ∈ Ra (V ). Therefore, su is a well-defined global section of δr (V ). Moreover, u 7→ su ,
is E(X)-linear and injective.
Conversely, a section s ∈ Γ(X, δr (V )) defines an r-density s♯ : Let b ∈ R(V ) with
π(b) = a ∈ X. Then s(a) ∈ δr (V )a as the equivalence class s(a) = [(b′ , z ′ )] contains
exactly one pair of the form (b, z) ∈ s(a). Define s♯ (b) := z. Then s♯ : R(V ) → C is well-
defined and satisfies s♯ (bg) = | det g|r s♯ (b), since the unique ζ ∈ C with (bg, ζ) ∈ s(a)
is ζ = | det g|r z: (b, z) ∼ (bg, | det g|r z)], hence, s♯ (bg) = ζ = | det g|r z. Clearly,
(su )♯ = u and (s♯ )u = s. Moreover, s 7→ s♯ is E(X)-linear, and hence an isomorphism
Γ(X, δr (V ) ∼
= ∆r (V ).

From Proposition D.9 we obtain

Fact 12.4. Let gjk be transition function for the vector bundle V with respect to an
open cover (Uj ) of X. Then | det gjk |−r are suitable transition functions for δr (V ).
Lemma 12.5. Any line bundle δr (V ) of r-densities on a complex vector bundle V
over X is trivial. There exists a positive r-density which induces a a vector bundle
isomorphism δr (V ) → X × C.
172 12. Densities and Their Derivatives

Proof. It is enough to determine a nowhere vanishing section s = sw ∈ Γ(X, δr (V )

given by a non-vanishing r-density w : R(V ) → C: Let (b1 , . . . , bk ) be a fixed basis of
Ck . There is an open cover (Uj ) of X with local trivializations ψj : VUj → Uj ×Ck of the
vector bundle V . On Uj the local sections bκ (x) := ψj−1 (x, bκ ) , x ∈ Uj , κ = 1, . . . , k ,
form a frame field b̂j : Uj → R(V )Uj given by b̂j (x) = (b1 (x), . . . , bk (x)) ∈ Rx V |Uj . Now,
wj (b̂j g) := | det g|r , g ∈ GL(k, C), defines a positive r-density wj on the restriction VUj
satisfying wj (b̂j ) = 1. Hence, with a smooth partition of unity (hj ) subordinated to
the cover (Uj ) we obtain a positive r-density
X
w := hj wj : R(V ) → C .

As a result, all density line bundles on a vector bundle V over X are the same from
the viewpoint of isomorphism classes of line bundles.

Observation 12.6. Up to scaling, the line bundle δr (V ) is determined completely

by choosing one nowhere vanishing r-density u ∈ ∆r (V ): Then every w ∈ ∆r (V ) ∼
=
Γ(X, δr (V )) is of the form w = λu with a unique λ ∈ E(X).
In particular, ∆r (V ) is an E(M )-module of rank 1 and is isomorphic to E(M ).
Using this result we obtain, by choosing a nowhere vanishing and positive 1-density
u on V , an isomorphism of complex line bundles ∆1 (V ) → ∆r (V ) , f u → f ur , resp.
δ1 (V ) ∼
= δr (V ). This isomorphism is not of great interest, it depends on the choice of
u.

Nevertheless, certain natural isomorphisms are of interest. The rest of this section
is not directly needed in the sequel, it is only an elementary and detailed investigation
around the new notion of r-density bundles and gives an impression how to work with
them:

Proposition 12.7. For complex vector bundles V, W, Z over the manifold X and r, s ∈
R there exist the following natural isomorphisms of density bundles:

1. δr (V ) ⊗ δs (V ) ∼
= δr+s (V )

2. (δr (V ))∨ ∼
= δr (V ∨ ) ∼
= δ−r (V )

3. δr (V ⊕ Z) ∼
= δr (V ) ⊗ δr (Z)

4. δr (W ) ⊗ δ−r (V ) ∼
= δr (Z) and, equivalently, δr (W ) ∼
= δr (V ) ⊗ δr (Z) if

0 −→ V −→ W −→ Z −→ 0

is an exact sequence of vector bundles.

12.1 Densities 173

The isomorphisms, and hence their ”naturality”61 are described in the proofs.

Proof. Short proofs can be given by using the transition functions. For instance, in
case 1., the transition functions of δt (V ) for general t are | det gij |−t when gij are
the transition functions of the vector bundle V . Hence, the transition functions of
δr (V ) ⊗ δs (V ) are | det gij |−r | det gij |−s = | det gij |−(r+s) . This equality implies that
δr (V ) ⊗ δs (V ) and δr+s (V ) are isomorphic as line bundles.
However, we present proofs which use the properties of densities directly in order to
give a detailed impression about how one can work with r-densities on vector bundles
which are interrelated.
Ad 1.: For a basis b of Vx , x ∈ X, i.e. b ∈ Rx V , and for σ ∈ Γr (V ) , τ ∈ Γs (V ) we
define
(σ.τ )♯ (b) := σ ♯ (b)τ ♯ (b).
For g ∈ GL(k, C), k = rk V , the following transformation property is satisfied:

(σ.τ )♯ (bg) = σ ♯ (bg)τ ♯ (bg) = | det g|r σ ♯ (b)| det g|s τ ♯ (b) = | det g|r+s (σ.τ )♯ (b) .

Hence, σ.τ is a well-defined r +s-density on V . And it is easy to check that the induced
map σ ⊗ τ 7→ σ.τ is an isomorphism of line bundles.
Ad 2.: For b ∈ Rx V let b∨ be the dual basis b∨ ∈ Rx V ∨ with b∨j (bk ) = δjk . Then
δr (V ∨ ) ∼
= δ−r (V ) , τ 7→ στ , is given by

(στ )♯ (b) := τ ♯ (b∨ )

for τ ∈ δr (V ∨ ). Indeed, for g ∈ GL(k, C) we have

(στ )♯ (bg) = τ ♯ ((bg)∨ ) = τ ♯ (b∨ g −1 ) = τ ♯ (b)| det g −1 |r = (στ )♯ (b)| det g|−r .

Hence, στ ∈ Γ−r (V ).
Ad 3.: If c ∈ Rx V , d ∈ Rx Z then b = (c, d) ∈ Rx (V ⊕ Z). The isomorphism
δr (V ) ⊗ δr (Z) → δr (V ⊕ Z) is given by σ ⊗ τ 7→ σ · τ where

(σ · τ )♯ (b) := σ ♯ (c)τ ♯ (d).

This is well defined, since for another choice of (c′ , d′ ) (instead of (c, d) = b) there are
unique g ∈ GL(k, C) and h ∈ GL(m, C) (m = dim Z) such that c′ = cg , d′ = dh.
Therefore, with  
g 0
G=
0 h
61
The results of this proposition may be described by stating that the δr (or ∆r ) , r ∈ R define a
family of functors satisfying certain natural properties.
174 12. Densities and Their Derivatives

we obtain
(σ · τ )♯ (bG) = (σ · τ )♯ (cg, dh)
= σ ♯ (cg)τ ♯ (dh)
= σ ♯ (c)| det g|r τ ♯ (d)| det h|r
= (σ ♯ (c)τ ♯ (d))| det G|r
Now σ·τ has to be extended to all of Rx (V ⊕Z) by the transformation rule (σ·τ )♯ (bG) =
(σ · τ )♯ (b)| det G|r for all G ∈ GL(k + m, C) to become an r-density on V . The induced
map is an isomorphism of line bundles.
Moreover, 3. is a special case of 4.
Ad 4.: We cannot apply 3. directly, since W is not necessarily isomorphic to the
direct sum V ⊕ Z, although this is locally true. We omit the base point ”x ” in the
following.
Let (Z1 , . . . , Zm ) be a basis of Z, let c = (X1 , . . . , Xk ) be a basis of V and let
b = (X1 , . . . , Xk , Y1 , . . . Ym ) a basis of W such that p(Yj ) = Zj with respect to the
projection p : W → Z in the exact sequence. Let σ ∈ δ−r (V ) , ε ∈ δr (W ). Define
ν = ν(σ, ε) by
ν ♯ (Z1 , . . . , Zm ) := ε♯ (X1 , . . . , Xk , Y1 , . . . Ym )σ ♯ (X1 , . . . Xk ) = ε♯ (b)σ ♯ (c) .
The choice of another basis b′ = (X1′ , . . . , Xk′ , Y1′ , . . . Ym′ ) such that (X1′ , . . . , Xk′ ) is in
R(V ) and p(Yj′ ) = Zj , j = 1, . . . , m, satisfies c′ = ch with a unique h ∈ GL(k, C) and
Yj′ = Yj + Xβ dβj for suitable dβj ∈ C, d = (dβj ). As a consequence, b′ is of the form
b′ = bg, where  
h d
g=
0 1
with det g = det h. Hence,
ε♯ (b′ )σ ♯ (c′ ) = | det g|r | det h|−r ε♯ (b)σ ♯ (c) = ε♯ (b)σ ♯ (c) ,
and ν ♯ (Z1 , . . . , Zm ) is well-defined. In order to show that ν is an r-density
let g ∈ GL(m, C). With the notation z := (Z1 , . . . , Zk ) the following trans-
formation property holds true: ν ♯ (zg) = ε♯ (X1 , . . . , Xk , (Y1 , . . . Ym )g)σ ♯ (c) =
| det g|r ε♯ (X1 , . . . , Xk , Y1 , . . . Ym )σ ♯ (c) = | det g|r ν ♯ (z), so that ν ∈ δr (Z). Finally, it
is straightforward to check that the map
δr (W ) ⊗ δ−r (V ) → δr (Z) , σ ⊗ ε 7→ ν = ν(σ, ε)
is an isomorphism of line bundles.
Remark 12.8. If we choose a fixed nowhere vanishing r-density ϵ ∈ Γr (W ) in the last
part of the proof above, the definition of ν ∈ Γr (Z) leads directly to the definition of
a line bundle isomorphism ν : δ−r (V ) → δr (Z) , σ 7→ ν(σ), by setting
ν(σ)♯ (z) := ε♯ (b)σ ♯ (c) .
This kind of isomorphisms of density bundles will be used later in several occasions.
12.2 Integration of Densities 175

12.2 Integration of Densities

Densities occur in the context of integration on a manifold X. A traditional definition

is the following. Let X be a manifold of dimension m. A density is a rule that assigns
to each chart q : U → V of a differentiable atlas of X a function gq ∈ E(V, R), such
that for any other chart q̄ : Ū → V̄ with the following transition property is satisfied:
U ∩ Ū ̸= ∅  
∂ q̄
gq = det gq̄ .
∂q

The concept agrees essentially with the notion of a 1-density u ∈ ∆1 (T X C ) in the

sense of Definition 12.1 whenever u is real-valued, i.e. u(b) ∈ R for b ∈ R(T X)62 . For
such a density u, the assignment
   
u ∂ ∂ ∂
gq = u 1
, . . . , n =: u ∈ E(V )
∂q ∂q ∂q

for a chart q : U → V defines a density in the traditional way, since

     
∂ ∂ q̄ ∂ ∂ q̄ ∂
u =u = det u ,
∂q ∂q ∂ q̄ ∂q ∂ q̄

by the definition of a 1-density u.

Conversely, a traditional density (gq ) defines a 1-density u through

u(x) := gq (q(x))|dq 1 ∧ . . . ∧ dq m | , x ∈ U .

u is well-defined because of the transition property, and the map u 7→ (gqu ) is one-to-one.
In the following we recall how integral of a density on a general manifold X.

Definition 12.9 (Integration). LetR w : R(T X C ) → C be a 1-density. Leaving aside

convergence problems the integral X w ∈ C is defined
R as follows. With respect to a
chart q : U → V of X , V ⊂ Rm open, the integral U w of w over U is
Z Z  
∂
w := w dq ,
U V ∂q

where dq denotes Lebesgue measure on V ⊂ Rm and

     
∂ ∂ ∂ ∂ ∂
w (q) = w , . . . , m (q) = w |x , . . . , m |x
∂q ∂q 1 ∂q ∂q 1 ∂q

with q(x) = q ∈ V , x ∈ U . The result is independent of the chart because of the change
of variables formula for the Lebesgue measure dq and the transformation property of
62
which means that u is a (real) 1-density on T X.
176 12. Densities and Their Derivatives

w: Both change with the absolute value of the Jacobian, but in an inverse manner to
each other. When q̄ : U → V̄ is another chart and q = q(q̄) : V → V̄ the induced
change of coordinates, we have (schematically)
 
∂q
dq = det dq̄
∂ q̄
and
            −1
∂ ∂ ∂ q̄ ∂ ∂ q̄ ∂ ∂q
w =w =w det =w det .
∂q ∂ q̄ ∂q ∂ q̄ ∂q ∂ q̄ ∂ q̄
Hence, Z   Z    
∂ ∂ ∂q
w dq = w det dq̄
V ∂q V̄ ∂q ∂ q̄
Z    −1  
∂ ∂q ∂q
= w det det dq̄
∂ q̄ ∂ q̄ ∂ q̄
ZV̄  
∂
= w dq̄ .
V̄ ∂ q̄
The integral is extended to all of X by a partition of unity subordinated to a
covering (Uj ) with charts qj : Uj → Vj .

The integral is well-defined, even if X is not oriented. Recall, that the integration
of a volume form η on an oriented manifold X is similarly defined: In that case an
orientation is chosen meaning that we have an atlas with positive coordinate changes
fij i.e. with det Dfij > 0. Then locally
Z Z  
∂ ∂
η= η , . . . , m dq .
Uj Vj ∂q 1 ∂q

As a consequence Z Z
η= |η| .
X X
Remark 12.10. With the aid of densities we not only have a natural integration on
a manifold X but we also can define natural Lp -spaces Lp (X) of 1/p-densities, where
p ∈ R , 1 ≤ p < ∞: On the space ∆1/p (T X C )c of 1/p-densities u on X with compact
support in X one defines the norm
Z  p1
p
∥u∥p := |u| ,
X

and Lp (X) as the completion of the normed space (∆1/p (T X C )c , ∥ ∥p ).

In particular, we obtain the natural HilbertR space L2 (X) as the (completion of
2
R of half-densities u on X such that X |u| < ∞. The scalar product is
the) space
⟨u, v⟩ = X ūv. This Hilbert space is sometimes called the canonical Hilbert space of
the manifold X.
12.2 Integration of Densities 177

We conclude the section with some remarks and results on orientable manifolds.
These results are not needed for for the half-density quantization.

Definition 12.11 (Orientation Bundle). Let X be an m-dimensional manifold with the

frame bundle R(T X) which is a principal fibre bundle with structure group GL(m, R).
The Orientation Bundle is the real(!) line bundle O = O(X) associated to R(T X)
using the representation σ : GL(m, R) → GL(1, R) = R× , g 7→ sgn det g, i.e. with
respect to the left action GL(m, R) × C → C , (g, z) 7→ (sgn det g)z.

Another description of the orientation bundle is the following: Let (Uj ) be an open
covering of X with differentiable charts φj : Uj → Vj . Denote fij := φi ◦ φ−1 j :
φj (Uij ) → φi (Uij ) the change of coordinates. Let O′ be the line bundle which arises
by glueing the Ui × C with respect to the cocycle sij := sgn det Dfij . O′ is isomorphic
to O. Indeed, with the aid of the given charts φj the frame bundle R(T X) has as
transition functions the gij := Dfij 63 . Hence, the transition functions of the associated
line bundle O (see Section D.3) are σ(gij ) = sgn det gij = sij , and the two bundles are
naturally equivalent (note, that s−1ij = sij ).

Proposition 12.12. X is orientable if and only if O is trivial.

Proof. If X is orientable, then, by definition, there exists an atlas of the differentiable

structure such that the changes of charts (coordinate changes) fij are all positive in
the sense that there Jacobians gij := Dfij have positive determinants. Therefore the
sij = sgn det gij are all 1. This implies that O is trivial, a trivialization is induced by
the section t given by local functions tj (x) = 1 , x ∈ Uj , j ∈ I, respecting the section
condition (S): ti = sij tj . Conversely, if O has a trivialization there is an atlas such that
the transition functions can be chosen to be sij = 1 = sgn det Dfij and this means that
the Dfij have positive determinants.

Notice, that for an orientable manifold X an orientation is given by a global nowhere

vanishing section t ∈ Γ(X, O).

Remark 12.13. If, in the case of r = 1, one changes the above definition for the bundle
of r-densities by replacing the representation ρ1 (g) = | det g|−1 with the representation
τ (g) = (det g)−1 , one obtains the line bundle Λm T ∗ X C of complex m-forms on X up to
isomorphism (cf. (50)). This line bundle is also called the determinant bundle of the
cotangent bundle T ∗ X or the canonical bundle of X (see Section 15.1 for the concept
of a canonical line bundle).

Proposition 12.14. δ1 (T X) ∼
= Λm (T ∗ X) ⊗ O as real vector bundles.

Proof. T ∗ X is the dual T X ∨ of T X. Let (Uj ) be an open covering of X with charts

qj : Uj → Vj of the differentiable structure. The Jacobians gij := Dfij of the coordinate
63
since the gij are the transition functions of the tangent bundle T X.
178 12. Densities and Their Derivatives

changes fij := qi ◦ qj−1 |qj (Ui ∩Uj ) are transition functions defining the structure on the
tangent bundle. Hence, the transition functions of Λm (T ∗ X) are the det gij−1 , those
of δ1 (X) are the | det gij |−1 , and those of the orientation bundle are sgn det gij . The
isomorphism now follows from

| det gij |−1 = det gij−1 · sgn det gij .

Corollary 12.15. X is oriented if and only if there exists a volume form, i.e. a nowhere
vanishing m-form.

12.3 Partial Connection

We need the concept of partial connection on the line bundle of densities δr (P ) on a

polarization P . We formulate in this section a general approach suitable for densities
on polarizations on an arbitrary symplectic manifold which can be generalized to half-
forms (cf. Chapter 15). The partial connection will be used in Chapter 14. In the next
chapter, Chapter 13, we need only a simplified version of the partial connection.
Let (M, ω) be a symplectic manifold of dimension 2n, and let P ⊂ T M C be a
complex polarization. We want to define a partial connection on the line bundle δr (P )
of r-densities on P (and similarly on δr (DC ), cf. end of this section, where D = P ∩ P ∩
T M ). The partial connection ∇X will only be defined for vector fields X ∈ Γ(M, P ) ⊂
V(M ), and not for general vector fields on M .
Let U ⊂ M be an open subset U in M and let ξj ∈ Γ(U, P ) , 1 ≤ j ≤ n, local vector
fields such that (ξ1 (a), . . . , ξn (a)) is a basis of Pa for all a ∈ U . Then ξ := (ξ1 , . . . , ξn )
yields a local section ξ : U → R(P ) of the frame bundle R(P ) of P , which is also called
a frame field.
For instance, ξj could be generated by n independent functions F1 , F2 , . . . , Fn ∈
E(U, C) as the Hamiltonians ξj = XFj . In that case we call ξ a Hamiltonian frame
field. It is clear that to every a ∈ M there exists an open neighbourhood U ⊂ M such
that there exists a Hamiltonian frame field ξ ∈ Γ(U, R(P )).
Given a frame field ξ we define an r-density

σξ ∈ Γ(U, δr (P )) by σξ (a) := [(ξ(a), 1)] , a ∈ U .

It is easy to see that σξ is the unique section U → δr (P ) with σξ♯ ◦ ξ = 1. In particular,

σξ is nowhere vanishing. Any other section σ ∈ Γ(U, δr (P )) has a presentation as

σ = (σ ♯ ◦ ξ) σξ ,

since for σ = f σξ and a ∈ U :

σ ♯ (ξ)(a) = (f σξ )♯ (ξ)(a) = f (a)σξ♯ (ξ)(a) = f (a) .

12.3 Partial Connection 179

Now let ξ be a Hamiltonian frame field. The connection to be defined shall have
the property that the covariant derivatives of r-densities σ ∈ Γ(U, δr (P )) should vanish
when the function σ ♯ ◦ ξ : U → C is covariantly constant along P in the sense that
LX (σ ♯ ◦ ξ) = 0 for all X ∈ Γ(U, P ). This requirement leads to the following definition.
Definition 12.16. Let let ξ be a Hamiltonian frame field. For a vector field X ∈
Γ(U, P ) and σ = (σ ♯ ◦ ξ) σξ ∈ Γ(U, δr (P )) one defines

∇X σ := LX (σ ♯ ◦ ξ) σξ .

In particular, ∇X σξ = 0.
Lemma 12.17. The definition ∇X σ, where X ∈ Γ(U, P ) and σξ ∈ Γ(U, δr (P )), is
independent of the Hamiltonian frame field ξ and extends to all of M .

Proof. Any other Hamiltonian frame field ξ ′ : U → R(P ) is of the form ξ ′ = ξg where
g : U → GL(n, C) is a smooth map. Because of
♯
σξg = σξg (ξ)σξ and 1 = σξg (ξg) = | det g|r σξg (ξ)

we have σξg = | det g|−r σξ . Moreover, σ ♯ (ξg) = | det g|r σ ♯ (ξ), since σ ♯ is an r-density.
Comparing now LX (σ ♯ ◦ ξ) σξ with

LX (σ ♯ ◦ ξg) σξg = LX (σ ♯ ◦ ξg) | det g|−r σξ

= LX ((σ ♯ ◦ ξ)| det g|r ) | det g|−r σξ
= LX (σ ♯ ◦ ξ) σξ + (σ ♯ ◦ ξ)LX (| det g|r ) | det g|−r σξ

it remains to show LX (| det g|r ) = 0 in order to obtain

LX (σ ♯ ◦ ξ) σξ = LX (σ ♯ ◦ ξg) σξg .

It is enough to show LX gαβ = 0 for all 1 ≤ α, β ≤ n, and X ∈ Γ(U, P ), i.e. the

components gαβ of the map g = (gαβ ) : U → GL(n, C) are ”covariantly constant along
P ”. For the moment, let us assume that X is Hamiltonian: X = XF ∈ Γ(U, P ). Then

[X, ξα′ ] = 0 and [X, ξβ ] = 0.

Using the Lie derivative64 of vector fields, this is the same as

LX ξα′ = 0 and LX ξβ = 0. (51)

As a consequence

0 = LX ξα′ = LX (ξβ gαβ ) = (LX ξβ )gαβ + ξβ (LX gαβ ) = ξβ (LX gαβ ) .

64
see Proposition A.24 for the Lie derivative LX Y = [X, Y ] of a vector field Y along a vector field
X and some of its properties.
180 12. Densities and Their Derivatives

This implies LX gαβ = 0, since ξ is basis. Since a general X is locally a sum X = hκ X κ

P
with Hamiltonian X κ we finally obtain
X
LX gαβ = hκ LX κ gαβ = 0 .

We have shown, that the local definition of ∇X s is independent of the frame field
ξ. Therefore, the local definitions on U and U ′ with U ∩ U ′ ̸= ∅ agree on U ∩ U ′ and
define ∇X on U ∪ U ′ , and furthermore, on all of M .

The partial connection has all the usual linearity and derivation properties a con-
nection should have. The following properties are easy to prove.

Lemma 12.18. Whenever σ ∈ Γ(M, δr (P )) and X, Y ∈ Γ(M, P ), the following equa-

tions hold for functions f ∈ E(M )
1. ∇X (f σ) = (LX f )σ + f ∇X σ.
2. ∇f X+Y σ = f ∇X σ + ∇Y σ.
3. ∇X ∇Y σ + ∇Y ∇X σ = ∇[X,Y ] σ.

In particular, the last equation means that the partial connection is flat.

The case D
In an analogous way there is a partial connection on DC : For σ ∈ Γr (DC ) and
X ∈ Γ(M, P ) we obtain ∇X σ ∈ Γr (DC ). The definition works for vector fields X ∈
Γ(M, P + P ) = Γ(M, E C ):
Let U ⊂ M be an open subset for which there exists a Hamiltonian frame field ξ ∈
Γ(U, DC ). Define σξ ∈ Γ(U, δr (DC ) by the property σξ♯ (ξ) = 1. Any σ ∈ Γ(U, δr (DC ))
is of the form σ = σ ♯ (ξ)σξ .

Definition 12.19. For a vector field X ∈ Γ(U, E C ) and σ ∈ Γ(U, δr (DC )) one defines

∇X σ := LX (σ ♯ ◦ ξ)σξ .

Lemma 12.20. The definition ∇X σ, where X ∈ Γ(U, E C ) and σξ ∈ Γ(U, δr (DC )), is
independent of the Hamiltonian frame field ξ and extends to all of M .

Proof. This Lemma has essentially the same proof as Lemma 12.17.

Observation 12.21. The connection is partial, because the condition (51) is crucial
for the construction.
12.4 Partial Lie Derivative 181

12.4 Partial Lie Derivative

We want to compare the partial connection just defined with a partial Lie derivative on
r-densities σ ∈ Γr (P ). We need both derivatives for the formulation of the Half-Density
Quantization.
The Lie derivative LX : Γr (P ) → Γr (P ) will be defined only for those vector fields
X ∈ V(M ) which preserve P , i.e. for which [X, Y ] = LX Y ∈ Γ(M, P ) holds for all
Y ∈ Γ(M, P ). Notice, that the vector fields X ∈ Γ(M, P ) preserve P since P is
involutive. But in general, the set of P -preserving vector fields is strictly larger than
Γ(M, P ) as the following example shows.

Example 12.22. Let us have a look at the harmonic oscillator with phase space
M = T ∗ Rn = Cn and holomorphic polarization P generated by the vector fields
∂
∂ z̄j

(see Example 10.23). The Hamiltonian vector field XH of the energy H = 12 zj z̄j is
preserving P but is not in P :
X   
∂ ∂ ∂ ∂
XH = i zj − z̄j ∈
/ Γ(M, P ) and XH , =i ∈ Γ(M, P ) .
∂zj ∂ z̄j ∂ z̄k ∂ z̄k

Moreover, the set of P -preserving vector fields is not an E(M )-module it is merely a
module over the ring of polarized functions. In fact, z̄k XH does not preserve P .

As a consequence the partial Lie derivative LX will be defined for a larger class of
vector fields X than the previously considered vector fields X ∈ Γ(M, P ) for which the
partial connection ∇X has been defined. This is important when defining the quantum
operator for observables F , since the directly quantizable observables F are precisely
those for which XF preserves P (cf. Definition 10.17), and this implies that LXF σ is
well-defined for σ ∈ Γr (P ).
We know that X preserves P if and only the local flow (ΦX
t )t∈R of X satisfies

T ΦX
t (P ) = P , t ∈ R .

Let us abbreviate Φt = ΦX t and Ta Φt (b) = (Ta Φt (b1 ), . . . , Ta Φt (bn )), where b =

(b1 , . . . , bn ) ∈ Ra (P ) is a frame. For σ ∈ Γ(M, δr (P )) the pull-back Φ∗t σ of σ with
respect to Φt is given by (Φ∗t σ)♯ (b) = σ ♯ (Ta Φt (b)).

Definition 12.23. For X ∈ V(M ) preserving P the (partial) Lie derivative LX :

Γr (P ) → Γr (P ) is defined for σ ∈ Γr (P ) and b ∈ Ra (P ) by

d d
(LX σ)♯ (b) := (Φ∗t σ)♯ (b) = σ ♯ (Ta Φt (b)) .
dt t=0 dt t=0
182 12. Densities and Their Derivatives

Notice, that the Lie derivative is defined in analogy to the Lie derivative of a
differential form (cf. (71)) using the pullback of forms, see Definition A.33.

Proposition 12.24. Let σξ ∈ Γ(U, δr (P )) be the local section corresponding to a Hamil-

tonian frame field ξ ∈ Γ(U, R(P )) with (σξ )♯ (ξ) = 1, as above. For each Hamiltonian
X ∈ Γ(M, P ) the Lie derivative LX σξ vanishes.

Proof. There exists a unique gt ∈ E(U, GL(n, C)) such that T Φt (ξ) = ξgt , in the sense
of
(Ta Φt )ξ((a) = ξ(Φt (a))gt (a) , a ∈ U ,
for the local flow ΦX
t = Φt of X. For a section σ ∈ Γ(U, δr (P )) this definition of gt
leads to
σ ♯ (T Φt (ξ)) = σ ♯ (ξ(Φt )gt ) = | det gt |r σ ♯ (ξ(Φt )).
Therefore, LX σ is given by

d
(LX σ)♯ (ξ(a)) = σ ♯ (Ta Φt (ξ(a)))
dt t=0
d
= | det gt (a)|r σ ♯ (ξ(Φt (a))
dt t=0
d d
= | det gt (a)|r σ ♯ (ξ(a)) + | det g0 (a)|r σ ♯ (ξ(Φt (a)) .
dt t=0 dt t=0

In the case of σ = σξ the second term vanishes immediately, since σξ♯ (ξ) = 1. The first
term vanishes, as well: Differentiating T Φt (ξ) = ξgt gives
   
d d d d
T Φt (ξ) = ξ gt + ξ gt = ξ gt ,
dt dt dt dt

which implies
 
d d
ξ gt = T Φt (ξ) = ([X, ξ1 ], . . . , [X, ξn ]) = 0 .
dt t=0 dt t=0

As a consequence
d d
gt = 0 and, hence |gt |r = 0,
dt t=0 dt t=0

so that the first term vanishes.

The partial Lie derivative has all the usual linearity and derivation properties a Lie
derivative should have. In particular,

LX (f σ) = LX f σ + f LX σ

for a scalar function f ∈ E(M ).

12.4 Partial Lie Derivative 183

Corollary 12.25. For a Hamiltonian vector field X ∈ Γ(M, P ) the partial connection
∇X and the partial Lie derivative LX agree on Γr (M, δr (P )):

∇X σ = LX σ .

Proof. A general section σ has locally the form σ|U = f σξ ∈ Γ(U, δr (P )). By definition
of the partial connection ∇X σ = (LX f )σξ . And LX σ = (LX f )σξ + f LX σξ = (LX f )σξ ,
since LX σξ = 0.

This is not true for all vector fields X ∈ Γ(M, P ) as the following example shows.

Example 12.26. In case of the vertical polarization P on M = T ∗ R the vector field

∂
X=p ∈ Γ(M, P )
∂p

satisfies LX |dp| = |dp| but ∇X |dp| = 0.

∂
Proof. ∇X |dp| = 0 by definition, since σξ = |dp| for ξ = ∂p . To determine LX |dp| we
t
see that the flow of X is Φt = (q, pe ). Now, a basis b of P is b = {ξ} , i.e. b ∈ R(P ).
Inserting
    
♯ ∂ t ∂
|dp| (T Φt (b) = |dp| T Φt = |dp| e = |et | = et
∂p ∂p

into
d d
LX |dp|(b) = |dp|(T Φt )(b) = et = 1 = |dp| ,
dt t=0 dt t=0

finally yields LX |dp| = |dp|.

Note, that for the Hamiltonian vector field

∂ ∂
XF = q +p
∂q ∂p

of F = pq we obtain the same result

LXF |dp| = |dp|

with the same calculation now using the flow Φt = (et q, et p).
The following example will be used to determine the Half-Density Quantization of
the harmonic oscillator.
184 12. Densities and Their Derivatives

Example 12.27. In case of the holomorphic polarization P on M = T ∗ R ∼

= C the
Hamiltonian vector field  
∂ ∂
XH = i z − z̄
∂z ∂ z̄
of the energy of the harmonic oscillator (see Example 12.22) satisfies

LXH |dz̄| = 0 .

Morever, for the vector field

∂
X = iz̄
∂ z̄
we obtain the same result LX |dz̄| = 0 .

Proof. The flow is Φt (z) = eit z, and for the basis

 
∂
∂ z̄

of P we obtain
     
∂ ∂
|dz̄| T Φt = |dz̄| eit = |eit | = 1 .
∂ z̄ ∂ z̄

As a consequence
    
∂ d ∂ d
LXH |dz̄| = |dz̄| T Φt = 1 = 0,
∂ z̄ dt t=0 ∂ z̄ dt t=0

and LX |dz̄| = 0.

Not all Lie derivatives LX annihilate |dz̄|:

Example 12.28. Again in case of the holomorphic polarization P on M = T ∗ R ∼

=C
the vector field
∂
X = +z̄ ∈ Γ(M, P )
∂ z̄
satisfies
LX |dz̄| = |dz̄| .

Proof. With the flow Φt (z) = et z we get

      
∂ d ∂ d t ∂ d
LX |dz̄| = |dz̄| T Φt = |dz̄| e = |et | = 1 ,
∂ z̄ dt t=0 ∂ z̄ dt t=0 ∂ z̄ dt t=0

hence LX |dz̄| = |dz̄| .

12.4 Partial Lie Derivative 185

Observation 12.29. The various partial connections ∇r := ∇ on δr (P ), r ∈ R, as

well as the partial Lie derivatives LX are compatible with each other- For instance,
given σ ∈ Γr (P ) , τ ∈ Γs (P ) and στ = σ ⊗ τ ∈ Γr+s (P ) the following holds:

∇r+s r s
X (στ ) = (∇X σ)τ + σ(∇X τ ) ,

LX (στ ) = (LX σ)τ + σ(LX τ ) ,

In particular ∇2s 2 s 2
X (τ ) = 2∇X τ and LX (τ ) = 2LX τ .

Remark 12.30. The partial connection ∇X on δr (DC ) for X ∈ Γ(M, E C ) agrees with
the Lie derivative LX if X is Hamiltonian as in the case of P , see Corollary 12.25.
However, the Lie derivative LX can be extended to vector fields X preserving E C as in
the case of P .

Summary
186 13. Half-Density Quantization of the Momentum Phase Space

13 Half-Density Quantization of the Momentum Phase Space

On the way to the half-density quantization in general we consider in this chapter the
special case of a cotangent bundle M = T ∗ Q with the vertical polarization P . This
case is sufficiently important to deserve a special attention and, moreover, it is less
complicated than the general case, so that the procedure is easier to explain.
So let M = T ∗ Q be the cotangent bundle of an n-dimensional manifold Q, with
its standard symplectic form ω = −dλ = dq ∧ dp and with the vertical polarization
P ⊂ T M C . Recall that the vertical polarization is P = DC where D = Ker T τ ∗ is the
vertical bundle (or distribution) with respect to the natural projection τ ∗ : T ∗ Q → Q.
Of course, P is reducible, we have M/D ∼ = Q. We shall denote the projection onto the
quotient manifold M/D by π : M → M/D, which is essentially τ ∗ : T ∗ Q → Q with
fibres ∼
= Rn . With respect to local coordinates q = (q 1 , . . . , q n ) : U → V ⊂ Rn on an
open subset U ⊂ Q we have the bundle chart

(q 1 , . . . , q n , p1 , . . . pn ) : T ∗ U → V × Rn

on T ∗ U = T ∗ Q|U ∼
= V × Rn with its generalized momenta pj . A basis (over E(U )) for
the vertical polarization PU = P |U is given by
 
∂ ∂
,..., = (Xq1 , . . . , Xqn ) .
∂p1 ∂pn

In particular, the above basis is a basis of the distribution DU , now over E(U, R).
The prequantum line bundle L is trivializable and we set L = M ×C. The Hermitian
structure H on L is the induced structure given in terms of the standard section
s1 (a) = (a, 1) , a ∈ M , through H(s1 , s1 ) = 1. Finally, we choose the connection
determined by the Liouville form λ = pj dqj , i.e.

∇X f s1 = (LX f − 2πiλ(X)f ) s1 ,

for a general section s = f s1 of L with f ∈ E(M ) and X a vector field on M 65 .

The local connection form α = −λ = −pj dq j is adapted to P : α(X) = 0 for every
X ∈ P . In particular, s1 is polarized, i.e. ∇X s1 = 0 for all X ∈ Γ(M, P ).
Observe, that the covariant derivative of f s1 along P , i.e. ∇X f s1 , X ∈ Γ(M, P ),
is essentially the usual directional derivative LX on the scalar f on M :
   
∂f
∇Xj ∂ f s1 = LXj ∂ f s1 = Xj s1 ,
∂pj ∂pj ∂pj

since ∇X s1 = 0.
65
other connections are possible by choosing other potentials α of ω leading to the same results.
13.1 Descend of Densities 187

As a consequence, the space of polarized sections is

∂f
Γ∇,P (M, L) = {f s1 | = 0 , j = 1, . . . , n} ,
∂pj

and can be viewed to be the space E(Q) of functions on Q. The representation space can
be constructed on the basis of these functions. Since, in general, there does not exist a
natural volume form on Q we work with 1-densities on Q. Here the preceding chapter
comes into action! The representation space turns out to be the square integrable
1-densities on the configuration space Q!
After these preliminaries, we describe the half-density quantization of T ∗ Q with
respect to the vertical polarization in the following three sections.

13.1 Descend of Densities

In order to obtain suitable densities on T QC we relate the densities on T QC with

densities on P = DC . We use the quotient bundle Z D := T M C /P → M and see that
the tangent map T π : T M C → T QC satisfies P = Ker T π and therefore induces a map
T : Z D → T QC which is fibrewise an isomorphism. Z D is the pull-back of T QC with
respect to π, Z D = π ∗ (T QC ), and it is essentially the complexified tangent bundle T QC
but lifted to a bundle over T ∗ Q = M . The following commutative diagram illustrates
these properties
0 /P / T MC / ZD /0

Tπ
 | T
T QC

The natural map T : Z D → T QC enables us to lift the densities ν on T QC to

densities τ on Z D , as we explain in the following.
Every r-density ν ∈ Γ(Q, δr (T QC )) can be lifted to become an r-density τ = τν ∈
Γ(M, δr (Z D )) = Γr (Z D ): In terms of the functions τν ♯ and ν ♯ one sets

τν ♯ (Y1 , . . . , Yn )(a) := ν ♯ (Ta π(X1 ), . . . , Ta π(Xn ))

for any basis (Y1 , . . . , Yn ) ∈ Ra (Z D ) and X1 , . . . , Xn ∈ Ta M C with Xj + P = Yj , j =

1, 2, . . . , n. It is easy to check that τν transforms appropriately and thus tν is an
r-density. In particular, we see that ν 7→ τν defines a homomorphism, the lift L :
Γr (T (Q)C ) → Γr (Z D ).
But not every r-density on Z D is such a lift, in fact, Γr (Z D ) = Γ(M, δr (Z D )) is
a module over E(M ) while Γr (T QC ) = Γ(Q, δr (T QC )) ∼= ∆r (T QC ) is a module over
E(Q). With respect to any nowhere vanishing ν1 ∈ Γr (T QC ) and its lift τ1 ∈ Γr (Z D )
each τ ∈ Γr (Z D ) has the unique representation τ = λτ1 with a factor λ ∈ E(M ) which
may depend on p.
188 13. Half-Density Quantization of the Momentum Phase Space

We are essentially interested to determine those densities τ on Z D which descend to

a density ν on T QC , in other words, which are a lift of a ν. To obtain a characterizing
property we introduce the partial connection ∇δ on the line bundle δ := δr (Z D ).

Definition 13.1. For τ1 as above (lift of non-vanishing ν1 ∈ Γr (T QC )) every τ ∈

Γr (Z D ) = Γ(M, δ) has the form τ = f τ1 with a unique f ∈ E(M ) and we set

∇δX τ = ∇δX (f τ1 ) := (LX f )τ1

for X ∈ Γ(M, P ).

Lemma 13.2. This definition is independent of τ1 .

Proof. For another τ2 which is the lift of a non-vanishing ν2 there is a λ ∈ E(Q)

such that ν1 = λν2 . It follows that τ1 = hτ2 with h := λ ◦ π ∈ E(M ). Now, for
τ = f τ1 = f hτ2 we have

(LX (f h))τ2 = ((LX f )h + f LX h) τ2 = (LX f )hτ2 = (LX f )τ1 ,

because of LX h = 0 for X ∈ Γ(M, P ), which shows the assertion.

∇δ is called partial connection, since it is only defined for X ∈ Γ(M, P ).

The partial connection is used to formulate the following simple criterion.

Lemma 13.3. τ ∈ Γr (Z D ) is the lift of an r-density ν ∈ Γr (T QC ) on T QC , and hence

descends to ν, if and only if ∇δX τ = 0 for all X ∈ Γ(M, P ), i.e. if τ is covariantly
constant along DC = P .

Proof. Let σ1 be a lift of a nowhere vanishing ν1 , as above. A general σ ∈ ∆r (Z D ) can

be expressed as σ = f σ1 with f ∈ E(M ). Then ∇δX σ = LX f σ1 . We conclude
∇δX σ = 0 for all X ∈ Γ(M, P )
⇐⇒ LX f = 0 for all X ∈ Γ(M, P )
⇐⇒ f does not depend on the vertical direction
⇐⇒ f has the form f = λ ◦ π with λ ∈ E(Q)
⇐⇒ σ is the lift of λν1 , where f = λ ◦ π with λ ∈ E(Q)

13.2 Representation Space

We now come to the representation space of the half-quantization of M = T ∗ Q using

densities. As explained at the beginning of the preceding chapter, the main aspect is
to replace the line bundle L with a line bundle L ⊗ δ, as we describe in the following.
13.2 Representation Space 189

Construction 13.4 (Representation Space of T ∗ Q). The new line bundle on M = T ∗ Q

is L ⊗ δ, where δ := δ1/2 (Z D )66 and where the corresponding new connection is ∇ ⊗ ∇δ ,
with ∇δ the (partial) connection just defined.
A section ψ = s ⊗ τ ∈ Γ(M, L ⊗ δ)67 is covariantly constant along D (or P ) (also
called a polarized section) if (∇ ⊗ ∇δX )(s ⊗ τ ) = 0 for all X ∈ Γ(M, D), and this is
equivalent to
∇X s = 0 and ∇δX τ = 0 .

In particular, when s ⊗ τ is covariantly constant along P , τ descends to a 1/2-

density ν = ν(τ ) on T QC according to Lemma 13.3. As a result we obtain for two such
polarized sections ψ = s ⊗ τ, ψ ′ = s′ ⊗ τ ′ , the 1-density

H(s, s′ )ν̄ν ′ = H(s, s′ )ν̄(τ )ν(τ ′ )

on T QC ∼
= T (M/D)C as a section of δ1 (T QC ), where ν ′ = ν(τ ′ ) lifts to τ ′ . In this
way one defines the scalar product through integration of the density H(s, s′ )ν̄ν ′ over
Q = M/D:
Z
′ ′ ′
⟨ψ, ψ ⟩ = ⟨s ⊗ τ, s ⊗ τ ⟩ := H(s, s′ )ν̄ ♯ ν ′♯ (52)
M/D

on the vector space HP of polarized sections with compact support. Finally, the com-
pletion HδP = Hδ (M, L, P ) of HP is the representation space we wanted to construct.
Note that we could have used the line bundle δ−1/2 (P ) instead of δ1/2 (Z D ), see
Remark 13.10, as it will be done in the general case in the next chapter.
In a more concrete picture of the representation space the basic elements of the
Hilbert space are functions on Q which are integrated by suitable half-densities on the
space Q, as we show in the following:
Fix a nowhere vanishing positive 1/2 -density ν1 on Q which exists according to
Lemma 12.5. Then ν := ν1 ν1 = ν12 is a positive p
1-density on Q. For instance, when Q
has a natural volume form ε, we can take ν1 = |ε| and it follows ν = |ε|.
Let τ1 be the uniquely defined lift of ν1 in Γ1/2 (M, Z D ). τ1 is a positive half-density
on Z D . Any polarized section s ⊗ τ in L ⊗ δ, δ = δ1/2 (Z D ), can be expressed as

s ⊗ τ = f s1 ⊗ gτ1 = ϕs1 ⊗ τ1

with f, g ∈ E(Q) and ϕ = f g 68 .

66
δ = δ−1/2 (P ) is an alternative choice, as we will see below and in the next chapter.
67
Notice, that every global section ψ ∈ Γ(M, L ⊗ δ) has the form ψ = s ⊗ τ for suitable s ∈ Γ(M, L)
and τ ∈ Γ(M, δ), since δ is trivial.
68
More precisely, there are f, g ∈ E(Q) with ϕ = f g such that τ = (f ◦π)s1 ⊗(g◦π)τ1 = ϕ◦πs1 ⊗τ1 =
ϕ.
190 13. Half-Density Quantization of the Momentum Phase Space

Therefore, the scalar product for polarized sections ψ = s ⊗ τ = ϕs1 ⊗ τ1 , ψ ′ =

s ⊗ τ ′ = ϕ′ s1 ⊗ τ1 ∈ Γ(M, L ⊗ δ), as defined in (52) can be written as
′

Z Z
′ ′ ′ ′ ♯ ♯
⟨ψ, ψ ⟩ = ⟨s ⊗ τ, s ⊗ τ ⟩ = ϕ̄ϕ ν1 ν1 = ϕ̄ϕ′ ν ♯ ,
Q Q

which is Z
′ ′
⟨ψ, ψ ⟩ = ⟨ϕ, ϕ ⟩ = ϕ̄ϕ′ ν ♯ .
Q

It is easy to show that this integral is independent of the choice of ν1 .

AsR a result, we can identify HδP with the Hilbert space of functions ϕ on Q such
that Q |ϕ|2 ν ♯ < ∞, or – which is essentially the same – the canonical Hilbert space
L2 (Q) of Q (cf. Remark 12.10).

13.3 Quantum Operators

To finish the case of the momentum phase space T ∗ Q = M the main step is to de-
termine the quantum operators associated to the quantizable classical observables.
Regarding the prequantization – as developed in Chapter 7 – one is tempted to define
the quantum operator in HδP assigned to a quantizable classical observable F ∈ E(M )
by the preliminary definition
 
δ i δ i
q (F )ψ := − (∇ ⊗ ∇ )XF + F ψ = (q(F )s) ⊗ τ − s ⊗ ∇δXF τ , (53)
2π 2π

where ψ = s ⊗ τ is a polarized section of L ⊗ δ, and q(F ) is the prequantum operator.

However, the partial connection operator ∇δX is defined only for vector fields X ∈
Γ(M, P ). Let us have a look at these vector fields. When XF ∈ P we know that (∇ ⊗
∇δ )XF ψ = 0, since ψ is polarized. In this way we only get multiplication operators as
quantum operators, and the whole effort of introducing connections on line bundles and
polarizations looks superfluous, in particular, there would be no geometric quantization.
To overcome this problem, we shall define a partial Lie derivative operator LX :
Γr (Z D ) → Γr (Z D ) which for vector fields X on M such that LX is well-defined at
least for the case of X = XF , F ∈ E(M ), where, in addition, F is directly quantizable
(cf. Section 10.3). Under these assumption, LX can be used to define the quantum
operator. The notation LX for this operator is justified since it is related to the Lie
derivative, although it is not defined directly as a Lie derivative. As a reminder, F
is called directly quantizable or simply quantizable, if ∇XF s is polarized for polarized
sections s. This is equivalent to XF preserving P , i.e. to the property that [XF , Y ] ∈
Γ(M, P ) for all Y ∈ P . The Lie algebra of all quantizable classical observables is
denoted by RP .
13.3 Quantum Operators 191

In order to define the operator LX we recall the concept of a Lie derivative of n-

forms on an n-dimensional manifold Q and its divergence: The Lie derivative LX ε of
an n-form ε ∈ Γ(Q, Λn T Q) is given by

LX ε = diX ε + iX dε = diX ε ,

since dε = 0. The divergence of X with respect to ε as the unique divε (X) ∈ E(Q)
with LX ε = divε (X)ε where ε ̸= 0, otherwise divε (X) = 0. For instance, locally, for
coordinates q j in an open subset U of Q, when X = X j ∂j and ε = dq 1 ∧ . . . ∧ dq n ∈
Γ(U, Λn (T Q)) we have the familiar expression
n
X ∂X j
divε (X) = .
1
∂q j

A general n-form η ∈ Γ(U, Λn (T Q)) can be written as η = hε with h ∈ E(U ). Then

LX η = (LX h)ε + hLX ε = (LX h + h divε (X))ε , and we deduce

h divη (X) = LX h + h divε (X) .

Using the local flow ΦX t : Qt → Q−t of the vector field X there is the alternative
description of the Lie derivative by the formula
d
ε T ΦX X


LX ε(ξ1 , . . . , ξn ) = t (ξ1 ), . . . , T Φt (ξn ) ,
dt t=0

when (ξ1 , . . . , ξn ) is a basis of Tq Q, q ∈ Q, see (71).

For the Lie derivative of densities, we take the last equation as the definition.
Definition 13.5. For ν ∈ Γr (T QC ) define
d ♯
(LX ν)♯ (ξ1 , . . . , ξn ) := ν T ΦX X



t (ξ1 ), . . . , T Φt (ξn ) .
dt t=0

We obtain the induced divergence divν through LX ν = divν (X)ν.69 .

For instance, locally, for coordinates q j in an open subset U of Q, when X = X j ∂j

and ν = |dq 1 ∧ . . . ∧ dq n |r ∈ Γ(U, δr (T QC )
n
X ∂X j
divν (X) = r .
1
∂q j

We now want to define LXF τ for 1/2-densities τ ∈ Γ1/2 (Z P ) for the special vector
fields X = XF , F ∈ RP , by lifting the Lie derivative LX : Γ1/2 (T QC ) → Γ1/2 (T QC ) just
69
divν is not a one form, in general, since it is not E(Q)-linear; it does not define a connection!
192 13. Half-Density Quantization of the Momentum Phase Space

defined. Let τ1 ∈ Γ1/2 (Z D ) be the lift of a non-vanishing ν1 ∈ Γ1/2 (T QC ), as before. For

a general τ ∈ Γ−1/2 (Z D ) , τ = gτ1 with g ∈ E(M ) we set:
LXF τ := (LXF g + divν1 (T π(XF ))g) τ1 .
It is easy to see that this definition does not depend on the choice of τ1 .
In case of XF ∈ Γ(M, P ), the definition entails LXF τ1 = 0, since T π(XF ) = 0.
Hence, for general τ = gτ1 :
LXF (gτ1 ) = (LXF g)τ1 .

Comparison: Recall that for XF ∈ Γ(M, P ) the partial connection is defined by

∇δXF (gτ1 ) = (LXF g)τ1 , for a non-vanishing lifted τ1 (cf. Definition 13.1). Comparing
with the last equation it follows that the partial connection ∇δX for X = XF ∈ Γ(M, P )
agrees with LX just defined. Hence, LXF for general F ∈ RP can be considered to be an
extension of the partial connection ∇XF , XF ∈ Γ(M, P ) to mor general Hamiltonian
vector fields XF by LXF .
We now can define the quantum operator q δ (F ) on the polarized sections s ⊗ τ ∈
ΓP (M, L ⊗ δ) improving our first attempt in (53):
Definition 13.6 (Quantum Operator). For directly quantizable F ∈ E(M ) the Quan-
tum Operator is

i
q δ (F )(s ⊗ τ ) := (q(F )s) ⊗ τ − s ⊗ LXF τ ,
2π

where q(F ) is the prequantum operator. Hence, the quantum operator also has the
form
i
q δ (F )(s ⊗ τ ) := − (∇XF s ⊗ τ + s ⊗ LXF τ ) + F s ⊗ τ ,
2π
An additional motivation for this definition is explained later in the context of
defining the quantum operator as an infinitesimal generator for the general case in
Definition 14.8.

Main Result
Summarizing the results of this chapter we have the following.
Theorem 13.7. The constructed half-density quantization for the cotangent bundle
M = T ∗ Q with its vertical polarization P yields the representation space HδP as de-
scribed above and is a full geometric quantization in the following sense:

1. The quantization map q δ : RP → S(HδP ) is R-linear and satisfies (D1) and (D2),
now for the new representation space, the well-defined Hilbert space HδP ∼
= L2 (Q)70
70
This means that the Dirac conditions are satisfied for any set o of classical observables contained
in RP .
13.3 Quantum Operators 193

2. If XF is complete, F ∈ RP , then q δ (F ) is self-adjoint.

The description of half-density quantization of the momentum phase space

M = T ∗ Q with its vertical polarization is now completed.

But we want to make the formula for q δ (F ) slightly more explicit in our special case
M = T ∗ Q regarding the fact, that the quantum operators act essentially on functions
ϕ ∈ E(Q). Let s1 be the usual global section of L and ν1 a nowhere vanishing 21 -density
on Q with its unique lift τ1 ∈ Γ1/2 (Z D ). Then every section ψ of L ⊗ δ has the form
ϕs1 ⊗ τ1 . The quantum operator is

i
q δ (F )(ϕs1 ⊗ τ1 ) = (q(F )ϕs1 ) ⊗ τ1 − ϕs1 ⊗ divν1 (T π(XF ))τ1
 2π 
i
= q(F ) − divν1 (T π(XF )) ϕs1 ⊗ τ1 .
2π

This means acting on E(M ), or more precisely on the functions with compact
support, the operator is

i
q δ (f ) = q(F ) − divν1 (T π(XF )) ,
2π

and the additional term induced by the half-density quantizaion in comparison to the
prequantization is apparent in this form. Let us study the additional term in the case
of a simple phase space:

Example 13.8 (Simple phase space). When the configuration space Q is an open
subset of Rn with the standard volume form ε = dq 1 ∧ . . . dq n = dq we get a formula for
the quantum operators q δ (F ) which is more explicit. Here, F is a directly quantizable
observable on the simple phase space M = T ∗ Q.
Let us begin by calculating the prequantum operator q(F ). F has the form F =
A + B j pj with functions A, B j ∈ E(Q).
The Hamiltonian vector field is

∂A ∂B j
 
∂ k ∂
XF = B − + k pj
∂q k ∂q k ∂q ∂pk

in global coordinates q k of Q and the induced pj . Applied to the functions ϕ ∈ E(Q)

only the first term of XF is relevant. Moreover, λ(XF ) = B j pj . Prequantization then
194 13. Half-Density Quantization of the Momentum Phase Space

yields for s = ϕs1 :  

i
q(F )s= − ∇XF + F ϕs1
2π
 
i j
= − LXF ϕ − B pj ϕ + F ϕ s1
2π
 
i j ∂
= − B ϕ + Aϕ s1 .
2π ∂q j
Acting on ϕ ∈ E(Q) this means

i j ∂
q(F ) = A − B .
2π ∂q j

To bring the additional term into a simpler form√we use the density ν := |ε|, also
denoted as |dq|, and the positive half-form ν1 := ν = |dq|1/2 ∈ Γ1/2 (T Q). From
LX ν12 = 2ν1 LX ν1 and LX ν = divε (X)ν we receive
1
LX ν1 = divε (X)ν1
2
for X ∈ V(Q). We have mentioned already that div(X) = divε (X) has the familiar
form
X ∂X k
div(X) = .
∂q k
Furthermore, T π(XF ) = B j ∂j =: B, and so

1 1 X ∂B k
L B ν1 = div(B)ν1 = ν1 .
2 2 ∂q k
As a consequence

i X ∂B k
 
δ j i j ∂
q (A + B pj ) = A − k
− B .
4π ∂q 2π ∂q j

This formula reveals the impact of LXF resp. of 12 div(B) as the essential part
of the additional term. In comparison to the prequantization of T ∗ Rn (cf. Example
7.18) without half-density quantization by simply restricting to those functions which
are independent of the variable pj and using the Lebesgue volume on Q we see that
half-density quantization imposes an additional term in the quantization, namely mul-
tiplication with
i X ∂B k i 1
− = − div(B) . (54)
4π ∂q k 2π 2
13.3 Quantum Operators 195

Restricting to the special cases F = A = q j and F = B j = pj we obtain the known

quantized observables

i ∂
q δ (q j ) = q j , q δ (pj ) = − .
2π ∂q j

The additional term is 0 in these cases.

For a single particle, or a particle system, one wants to quantize the energy H of
the system in order to complete the Schrödinger picture. However, an energy like H =
g ij pi pj − V (q) is not directly quantizable when one uses the vertical polarization. The
rules of Geometric Quantization are too restrictive to treat this important case. They
have to be extended to cover this case and more. We come back to the quantization of
the energy after having introduced the so-called BKS-pairing in the context of half-form
quantization.
Observation 13.9. The additional term resembles operator ordering, a concept which
is not part of Geometric Quantization in a direct way. For instance, in case of M = T ∗ R
and F = qp we see from the above formulas
i ∂
q(qp) = − q ,
2π ∂q
and
i i ∂
q δ (qp) = − − q .
4π 2π ∂q
With the notation
i ∂
Q := q δ (q) = q and P := q δ (p) = − ,
2π ∂q
this amounts to
q δ (qp) = QP ,
i
q δ (qp) = − + QP ,
4π
with −i/4π as the additional term. Inserting the canonical commutaion relation
i
P Q − QP = −
2π
we deduce
1
q δ (qp) = (P Q + QP ) .
2
Similarly,
i
q δ (q 2 p) = − Q + Q2 P = P Q2 − QP Q + Q2 P .
2π
196 13. Half-Density Quantization of the Momentum Phase Space

For the typical component of the angular momentum F = q k pj − q j pk (here M =

∗
T Rn , n > 1 and j ̸= k) the additional terms cancels out and we get the familiar
quantum operator
q δ (F ) = Qk Pj − Qj Pk
agreeing with the prequantum operator.

13.4 The Case of a Real Polarization

Before we study the case of a general polarization in the next chapter, let us consider
the case of a real reducible polarization, i.e. P = P and therefore DC = P ∩P = P = P .
This case is close to the case of a cotangent bundle M = T ∗ Q with vertical polarization
and can be treated in the same way.
Step 1: The quotient bundle Z D = T M C /P is essentially the complexified tangent
bundle T (M/D)C and the r-densities on T (M/D)C can be lifted to r-densities on Z D .
Moreover, a section τ ∈ Γr (Z D ) is a lift of an r-density (i.e. descends to) ν ∈
Γr (T (M/D)C ) if and only if ∇δX τ = 0 for the partial connection ∇ on δr (Z D ), defined
as in Definition 13.1.
Step 2: L will be replaced with

L ⊗ δ , δ := δ1/2 (Z D ) ,

and ∇ will be replaced with ∇ ⊗ ∇δ .

As before, any two sections ψ = s ⊗ τ, ψ ′ = s′ ⊗ τ ′ ∈ Γ(M, L ⊗ δ), which are
polarized, i.e. on which ∇ ⊗ ∇δ vanishes, induce on M/D the 1-density

H(s, s′ )νν ′ ,

where τ (resp. τ ′ ) is the lift of ν (resp. ν ′ ).

This leads to the scalar product
Z
′
⟨ψ, ψ ⟩ := H(s, s′ )ν ♯ ν ′♯
M/D

on the space of polarized sections HP ⊂ Γ(M, L ⊗ δ) with compact support, Finally,

the representation space HδP = Hδ (M, L, P ) is the completion of HP with respect to
the norm given by the scalar product.
Step 3:
i
q δ (F )(s ⊗ τ ) := (q(F )s) ⊗ τ − s ⊗ LXF τ .
2π
We conclude this chapter with the following remark which shows how the 1/2-density
bundle δ = δ1/2 (Z D ) can be replaced by −1/2-densities on P . This remark has no impact
13.4 The Case of a Real Polarization 197

to the half-density quantization of T ∗ Q which we just have carried through, but it

prepares the use of P instead of Z D in the subsequent considerations for the general
case of a complex polarization P given on a symplectic manifold (M, ω).

Proposition 13.10. There is a natural isomorphism τP : δ−r (P ) → δr (Z D ) induced

by the exact sequence

0 −→ DC −→ T M C −→ Z D = T M C /DC −→ 0 .

Recall P = DC . And we are back in the case of the momentum phase space M = T ∗ Q.

Proof. We know this from Proposition 12.7. With the r-density ε = εω on T M C defined
by
(ε♯ω )a (b) := |ωan (b)|r , b ∈ Ra (T M C ) , a ∈ M,
and with the notation [Z] := Z + P ∈ ZaD for Z ∈ Ta M C , an r-density τP (ρ) on Z D
can be defined for ρ ∈ Γ−r (P ) by

(τP (ρ))♯ ([Z1 ], . . . , [Zn ]) := ε♯ (X1 , . . . , Xn , Z1 , . . . , Zn )ρ♯ (X1 , . . . , Xn ) ,

where b = (X1 , . . . , Xn , Z1 , . . . , Zn ) ∈ Ra (T M C ) with x = (X1 , . . . , Xn ) ∈ Ra (P )

and z = (Z1 , . . . , Zn ) such that ([Z1 ], . . . , [Zn ]) ∈ Ra (Z D ). This is well-defined for
fixed (X1 , . . . , Xn ). When Z̃j with [Zj ] = [Z̃j ], we have ε♯ (X1 , . . . , Xn , Z1 , . . . , Zn ) =
ε♯ (X1 , . . . , Xn , Z̃1 , . . . , Z̃n ), since the matrix g with bg = ε♯ (X1 , . . . , Xn , Z̃1 , . . . , Z̃n )
satisfies det g = 1. The definition is independent of the choice of the frame x. Any other
basis y = (Y1 , . . . , Yn ) ∈ R(P ) of Pa has the form y = xg for a suitable g ∈ GL(n, C),
and we conclude

ε♯ (xg, z)ρ♯ (xg) = ε♯ (Y1 , . . . , Yn , Z1 , . . . , Zn )ρ♯ (Y1 , . . . , Yn )

= ε♯ (x, z)| det g|r ρ♯ (x)| det g|−r
= ε♯ (X1 , . . . , Xn , Z1 , . . . , Zn )ρ♯ (X1 , . . . , Xn ) .

Moreover, τ (ρ) is an r-density on Z D since

(τ (ρ))♯ ([z]) := ε♯ (x, zg)ρ♯ (x) = ε♯ (x, g)| det g|r ρ♯ (x) = (τ (ρ))♯ ([z])| det g|r .

(cf. proof of Proposition 12.7, part 4.)

Note, that in the case of M = T ∗ Rn the isomorphism has the following simple
and explicit form: With the notation |dp| = |dp1 ∧ . . . ∧ dpn | ∈ Γ(M, δ1 (P )) and with
the lift τ1 of a suitable non-vanishing r-density ν1 on T QC the above isomorphism
τ : δ−r (P ) → δr (Z D ) is given by

f |dp|−r 7→ f τ1 .
198 13. Half-Density Quantization of the Momentum Phase Space

Moreover, when we choose ν1 := |dq|r , dq = d11 ∧ . . . ∧ dq n and the lift τ1 is denoted

by the same symbol the isomorphism is simply

f |dp|−r 7→ f |dq|r .

As a consequence, the crucial half-densities τ, τ ′ ∈ Γ(M, δ1/2 (Z D )) tensored to the

sections s, s′ ∈ Γ(M, L) of the original prequantum bundle L could also be considered
as to be −1/2-densities ρ, ρ′ on P with τ = τ (ρ) , τ ′ = τ (ρ′ ). The partial connection
on δ1/2 (Z D ) in this interpretation will correspond to a partial connection on δ−1/2 (P )
which has been defined in the preceding chapter (see Definition 12.16).

Summary
 199

14 Half-Density Quantization in General

We now come to the construction of the representation space in case of a general

complex polarization in order to achieve the half-density quantization in general. Let
(M, ω) be a quantizable symplectic manifold with prequantum bundle (L, ∇, H) and let
P be a complex polarization with real part D, i.e. D is the distribution D = P ∩P ∩T M
of rank k = dimR D. It is assumed throughout this section, that M/D exists as
a (2n − k)-dimensional differentiable quotient manifold such that the quotient map
π : M → M/D is a submersion (i.e. the complex polarization P is reducible, according
to Definition 9.15).
In the following, we generalize the three construction steps described in the preced-
ing chapter for the vertical polarization on the momentum phase space M = T ∗ Q to
the case of an arbitrary reducible complex polarization P on a general symplectic man-
ifold (M, ω). This is the content of the next three sections. First, we have to generate
half-densities on the quotient manifold M/D which are induced by the polarization P .
Then the representation space will be determined and finally the quantum operators
are defined.

14.1 Descend of Densities

We use again the quotient bundle Z D := T M C /DC → M which is essentially the

(lifted) tangent bundle T (M/D)C : Since the quotient map π : M → M/D is a sub-
mersion by definition, the tangent map T π : T M → T (M/D) induces a vector bundle
homomorphism T : Z D → T (M/D)C which is fibrewise an isomorphism, i.e. the re-
strictions Ta : ZaD → Tπ(a) (M/D)C are vector space isomorphisms. Notice, that the
action of T is essentially that of T π: When pr : T M C → Z D = T M C /DC is the nat-
ural projection, we have T π = T ◦ pr. Altogether there is the following commutative
diagram:
0 /P / T M C pr / ZD / 0
Tπ
 | T
TQ C

In particular, we obtain a bundle morphism R(Z D ) → R(T (M/D)C ) which we

denote again by T with isomorphisms Ta : Ra (Z D ) → Ra (T (M/D)C ) , a ∈ M . For
vectors Zj ∈ Ta M C , j = 1 . . . , 2n − k, such that ([Z1 ], . . . , [Z2n−k ]) is a basis of ZaD
([Zj ] := Zj + Da ), Ta is given by

Ta ([Z1 ], . . . , [Z2n−k ]) := (Ta π(Z1 ), . . . Ta π(Z2n−k )) = (Ta ([Z1 ]), . . . , Ta ([Z2n−k ])) .

With the aid of T one can lift the r-densities on M/D, i.e. the r-densities on T (M/D)C ,
to r-densities on Z D as we have done in Chapter 13.
200 14. Half-Density Quantization in General

Let us describe this lifting in detail for a given section ν ∈ Γr (M/D) =

Γ(M/D, δr (T (M/D)C )): For Zj as above the lifted r-density τ is defined by
♯
τa♯ ([Z1 ], . . . , [Z2n−k ]) := νπ(a) (Ta π(Z1 ), . . . Ta π(Z2n−k )) .

This yields a unique r-density L(ν) = τ ∈ Γr (Z D ) = Γ(M, δr (Z D )) with the property

L(ν)♯ = ν ♯ ◦ T . The lifting defines an E(M/D)-homomorphism L : Γr (T (M/D)C ) →
Γr (Z D ).
We describe how suitable −r-densities on P descend to r-densities on M/D, and
apply the result to the case r = 1/2. To achieve this goal we first discuss which densities
on Z D are lifted densities and then consider isomorphisms δ−r (P ) → δr (Z D ).
The lifted densities can be characterized by a partial connection (as in the case of
M = T ∗ Q, cf. Chapter 13):

Definition 14.1. Let ν1 ∈ Γ(M/D, T (M/D)C ) be a nowhere vanishing r-density on

M/D (recall that all density line bundles are trivial according to Lemma 12.5, and
consequently admit nowhere vanishing sections), and let τ1 ∈ Γr (Z D ) = Γ(M, δr (Z D ))
its lift: τ1 := L(ν). Any section τ ∈ Γr (Z D ) has the form τ = f τ1 with a function
f ∈ E(M ). Now, for X ∈ Γ(M, DC ) we define

∇X τ := (LX f )τ1 .

This definition is independent of the choice of ν1 and yields a well-defined partial

connection for X ∈ Γ(M, DC ). This can be shown as in the case of M = T ∗ Q in
Section 13.1.

Lemma 14.2. τ ∈ Γr (Z D ) is a lift of an r-density ν on M/D if and only if ∇X τ = 0

for all X ∈ Γ(M, D).

Proof. A general τ ∈ Γr (Z D ) has the form τ = f τ1 with f ∈ E(M ). ∇X τ = LX f τ1 = 0,

i.e. LX f = 0, for all X ∈ Γ(M, DC ) is equivalent to f having the form f = h ◦ π with
h ∈ E(M/D). And f τ1 is the lift of hν1 if and only if f = h ◦ π.

To proceed in the construction of the half-density quantization we want to determine

a line bundle isomorphism δ−r (P ) → δr (Z D ) respecting the partial connections. Recall
from Proposition 12.7 that the exact sequences

0 −→ DC −→ T M C −→ Z D −→ 0

0 −→ DC −→ P −→ P/DC −→ 0
induce natural isomorphisms

δr (Z D ) ∼
= δr (T M C ) ⊗ δ−r (DC ) ,
14.1 Descend of Densities 201

δ−r (DC ) ∼
= δr (P/DC ) ⊗ δ−r (P ) ,
and as a combination

δr (Z D ) ∼
= δr (T M C ) ⊗ δr (P/DC ) ⊗ δ−r (P ) .

As a result, there exist natural isomorphisms (see Remark 12.8)

Γ−r (DC ) → Γr (Z D )
and
Γ−r (P ) → Γr (Z D ) .

In the following, we work with explicit isomorphisms τD : Γ−r (DC ) → Γr (Z D ) , resp.

τP : Γ−r (P ) → Γr (Z D ) (cf. Proposition 14.3), and use them for the construction of the
representation space. Later we show that the construction is essentially independent
of the isomorphism used (see Observation 14.7). And we will see in Remark 14.15,
that using τD or τP in the construction will lead to the same representation space up
to unitary equivalence71 , and thus to the same geometric quantization.
The main technical result is the following proposition. The proof is rather elemen-
tary, but due to the importance of the result in the construction of the representation
space we will present all details of the proof.

Proposition 14.3. There exists a natural line bundle isomorphism

τP : δ−r (P ) → δr (Z D )

with
∇X ◦ τP = τP ◦ ∇X
for all X ∈ Γ(M, DC ). In particular, the following diagram is commutative

τP
Γ−r (P ) / Γr (Z D )
∇X ∇X
 
Γ−r (P ) / Γr (Z D )
τP

Here, the partial connections ∇X are defined previously: ∇X on δr (P ) in Definition

12.16 and ∇X on δr (Z D ) in Definition 14.1.
The isomorphism τP can be understood as an isomorphism of line bundles with
connection as in Definition 8.5, but here the connections are only partial.
71
In the literature τP is preferred.
202 14. Half-Density Quantization in General

Proof. We introduce the following notation: Let

β = (X1 , . . . , Xk , Yk+1 . . . , Yn , Z1 . . . , Zk , Zk+1 . . . , Zn , )
be a basis of Ta M C such that
1. ξ D := (X1 , . . . , Xk ) is a basis of Da ,
2. ξ := (X1 , . . . , Xk , Yk+1 , . . . , Yn ) is a basis of Pa ,
3. ξ ′ := (X1 , . . . , Xk , Zk+1 , . . . , Zn ) is a basis of P a , (55)
4. γ := (Yk+1 , . . . , Yn , Z1 . . . Zn ) determines a basis Ta π(γ)
of Tπ(a) M/DC .

For Z ∈ Ta M C let us denote [Z] := Z + DaC ∈ ZaD , the image of the projection
Ta M C → T MaC /DaC = ZaD . Then the image of γ := (Yk+1 , . . . , Yn , Z1 . . . Zn ) is a
basis [γ] = ([Yk+1 ], . . . , [Yn ], [Z1 ] . . . [Zn ]) of ZaD . Moreover, with η := (Yk+1 , . . . , Yn )
we obtain a basis [η] of Pa /DC and we see that (η, η̄) := (Yk+1 , . . . , Yn , Ȳk+1 , . . . , Ȳn )
induces a basis ([η], [η̄]) of Ea /DaC .
Let ε := |ω n |r be the r-density on T M C induced by the symplectic form ω. And
let θ♯ ([η]) := |ω n−k |r/2 (η, η̄) define a corresponding r-density θ on P/DC . Then for
ρ ∈ Γ−r (DC ) we set:
τP (ρ)♯ ([γ]) := τP (ρ)♯ ([Yk+1 ], . . . , [Yn ], [Z1 ] . . . [Zn ]) := ε♯ (β)θ♯ ([η])ρ♯ (ξ) . (56)

τP (ρ) ∈ Γr (Z D ) is a well-defined r-density as we have shown in similar situations

before:
For an alternative choice β̃ = (X̃1 , . . . , X̃k , Ỹk+1 , . . . , Ỹn , Z̃1 . . . Z̃n ) ∈ Ra (Ta M C ) to
β satisfying the analogue of (55) we denote

ξ˜D := (X̃1 , . . . , X̃k ),

ξ˜ := X̃1 , . . . , X̃k , Ỹk+1 , . . . , Ỹn ),
(57)
η̃ := (Ỹk+1 , . . . , Ỹn ),
γ̃ := (Ỹk+1 , . . . , Ỹn , Z̃1 . . . Z̃n ).

Then there exists a unique G ∈ GL(2n, C) with β̃ = βG such that G can be written
as a block matrix  
d ∗ ∗
G= 0 h ∗
0 0 k
with d ∈ GL(k, C), h ∈ GL(n − k, C) and k ∈ GL(n, C). Moreover, ξ˜D = ξd, η̃ = ηh,
ξ˜ = ξH and γ̃ = γg, where H is the block matrix
 
d ∗
H :=
0 h
14.1 Descend of Densities 203

and g is the block matrix  

h ∗
g := . (58)
0 k

Then τP♯ ([γ̃]) as in (56) satisfies

˜
τP♯ ([γ]g) = τP♯ ([γ̃]) = ε♯ (β̃)θ♯ ([η̃])ρ♯ (ξ)
= ε♯ (βG)θ♯ ([η]h)ρ♯ (ξH)
= ε♯ (β)| det G|r θ♯ ([η])| det h|r ρ♯ (ξ)| det H|−r
= ε♯ (β)| det d|r | det h|r | det k|r θ♯ ([η])| det h|r ρ♯ (ξ)| det d|−r | det h|−r
= τP♯ ([γ])| det g|r ,

since det G = det d det h det k, det H = det d det h and det g = det h det k. Conse-
quently, there is the following transformation rule.

τP♯ ([γ]g) = τP♯ ([γ])| det g|r = τP♯ ([γ])| det G|r | det d|−r , (59)

for the special G, g, d as above.

In order to check whether the definition (56) is independent of the choice of the
representatives Z of the classes [Z] = Ta M C /DaC we assume ξ D to be fixed and [γ̃] = [γ].
In that case all elements of the diagonal of G are 1. Hence det G = 1 = det d = det g,
and the equation (59) yields
τD♯ ([γ̃]) = τD♯ ([γ]) .
Hence the term τP (ρ)♯ ([γ]) is well-defined when the basis ξ D = (X1 , . . . , Xk ) of Da is
fixed.
Next we show that the expression (56) is independent of the choice of the basis
ξ ∈ Ra (DC ). Any other basis ξ˜D can be described as ξ˜D = ξ D d with a unique
D

d ∈ GL(k, C). With h = 1 and k = 1 in our block matrix G we have det d = det G and
we obtain from (59)

τP♯ ([γ̃]) = τP♯ ([γ]g) = ε♯ (β)| det G|r θ♯ ([η])ρ♯ (ξ)| det d|−r = ε♯ (β)θ♯ ([η])ρ♯ (ξ) = τP♯ ([γ]) .

Thus, (56) is independent of the choice of ξ D .

As a result, τP (ρ)♯ ([γ]) is well-defined.
Let us confirm that τP (ρ) determines an r-density on Z D . Let β̃ an alternative
choice of a frame for Ta M C according to (57). We can require ξ˜D = ξ D since the
expression (56) is independent of ξ D as we have just shown. There is a unique g ∈
GL(2n − k, C) such that γ̃ = γg and which can be written as a block matrix as in (58).
With G as the block matrix defined by β̃ = βG we obtain immediately from (59)

τP♯ ([γ]g) = τP♯ ([γ])| det g|r .

204 14. Half-Density Quantization in General

As a consequence, fixing a β = (ξ D , γ) with (55) the assignment

τP♯ ([γ]g) := τP♯ ([γ])| det g|r , g ∈ GL(2n − k, C)
for general g ∈ GL(2n − k, C) coincides with just proven transformation property for
special g with (58), and thus defines a well-defined r-density on Z D .
We have shown that τP (ρ) is an r-density and defines an isomorphism of line bundles
τP : δ−r (P ) → δr (Z D ).
Finally, let ρ ∈ Γ(M, δ−r (DC )). We have to show ∇X τP (ρ) = τP (∇X ρ) for all
X ∈ Γ(M, DC ). It is enough to show this locally. Let a ∈ M and let U be an
open neighbourhood of a for which there exists a frame field β : U → R(T M C ), β =
(ξ1 , . . . , ξk , ηk+1 , . . . , ηn , ζ1 , . . . , ζn ), which satisfies (55) in each point of U . In particular
γ := (ηk+1 , . . . , ηn , ζ1 , . . . , ζn , ) generates Z D , i.e. [γ] := ([ηk+1 ], . . . , [ηn ], ζ1 , . . . , [ζn , ]) is
a frame field for Z D , and T π(γ) is a frame field of T (M/D)C . Finally, by changing
ξ D we can assume ε♯ (β)θ♯ (η) = 1 (replace ξ D with λξ D , where λ ∈ C satisfying λ−k =
ε♯ (β)θ♯ (η)).
From Section 12.3 we know that there is a unique density ρ1 = ρξ ∈ Γ−r (U, P ) with
ρ1 (ξ) = 1 and ∇X (ρ1 ) = 0 for all X ∈ Γ(M, DC ) (even for all X ∈ Γ(M, P )). Moreover,
any ρ ∈ Γ−r (U, P ) is of the form ρ = ρ♯ (ξ)ρ1 with ∇X ρ = LX (ρ♯ (ξ))ρ1 . Hence
(∇X ρ)♯ (ξ) = LX (ρ♯ (ξ)) .

In the same way there exists a unique density τ1 : U → δr (Z D ) with τ1♯ ([γ]) = 1 with
∇X (τ1 ) = 0 for the partial connection ∇X , X ∈ Γ(M, DC ). Any τ = f τ1 ∈ Γ(U, Z D )
has the representation τ = τ ♯ ([γ])τ1 and satisfies ∇X τ = LX (τ ♯ ([γ]))τ1 . In particular,
τP (ρ) = τP (ρ)♯ ([γ])τ1 = ε♯ (β)θ♯ (η)ρ♯ (ξ)τ1 = ρ♯ (ξ)τ1 .
It follows τP (ρ1 ) = τ1 since τP (ρ1 )([γ]) = ρ♯ (ξ)τ( [γ]) = 1. In addition we obtain
∇X τP (ρ) = LX (ρ♯ (ξ))τ1 .
Replacing in the last term the expression LX (ρ♯ (ξ)) by (∇X ρ)♯ (ξ) completes the proof
of the proposition due to the following identities
∇X τP (ρ) = (∇X ρ)♯ (ξ)τ1 = ε♯ (β)θ♯ (η)(∇X ρ)♯ (ξ)τ1 = τP (∇X ρ) .

Observation 14.4. Let τ1 ∈ Γ(M, δr (Z D )) be a nowhere vanishing (now global!)

lift of a nowhere vanishing r-density ν1 ∈ Γr (T (M/D)C ). The result of the preceding
proposition implies that there exists a nowhere vanishing density ρ1 such that ∇X ρ1 = 0
for all X ∈ Γ(M, DC ), namely ρ1 := (τP )−1 (τ1 ). Conversely, the existence of such a ρ1
yields the result of the proposition, by simply defining
τP : δ−r (P ) → δr (Z D ) , f ρ1 7→ f τ1 .
14.2 Representation Space 205

Lemma 14.5. Any other line bundle isomorphism τ : δ−r (P ) → δr (Z D ) has the form
τ = f τP with f ∈ E(M ). Moreover, τ satisfies
∇X ◦ τ = τ ◦ ∇ X
for all X ∈ Γ(M, D) if and only if f = h ◦ π for a suitable h ∈ E(M/D).

Proof. The first statement is clear. Moreover, when τ = f τP satisfies ∇X ◦ τ = τ ◦ ∇X

for all X ∈ Γ(M, D), then ∇X (f τP (ρ1 )) = ∇X (f τ1 ) = LX f τ1 = τ (∇X (ρ1 ) = 0, hence
LX f = 0 and f = h ◦ π , h ∈ E(M/D). And viceversa.

14.2 Representation Space

The construction of the representation space can now be carried through as in Chapter
13 using directly the bundle δ1/2 (Z D ). However, we are interested in describing the
representation space using the density bundle δ−1/2 (P ) (and δ−1/2 (DC ), see below in
Section 14.4).
Construction 14.6 (Representation Space). We replace L with
L ⊗ δ , δ := δ−1/2 (P ) ,
and ∇ with ∇ ⊗ ∇δ , where ∇δ is the partial connection on δ = δ−1/2 (P ).
Let s ⊗ ρ ∈ Γ(M, L ⊗ δ−1/2 (P )) be polarized72 . Then ∇X s = 0 and ∇X ρ = 0 for
all X ∈ Γ(M, DC )73 . According to Proposition 14.3 ∇X (τP (ρ)) = 0. This implies that
there is a unique ν = ν(ρ) ∈ Γ1/2 (T (M/D)C ) (cf. Lemma 14.2), such that τP (ρ) is the lift
of ν. As a consequence, any two polarized sections ψ = s⊗ρ, ψ ′ = s′ ⊗ρ′ ∈ Γ(M, L⊗δ),
determine the 1-density
H(s, s′ )ν̄ν ′
on M/D (with ν ′ = ν(ρ′ )). This 1-density on M/D defines the scalar product
Z
′ ♯
⟨ψ, ψ ⟩ := H(s, t)ν̄ ♯ ν ′
M/D

and the pre Hilbert space

HP = {ψ ∈ Γ(M, L ⊗ δ)ψ polarited and ⟨ψ, ψ⟩ < ∞}
of polarized sections with with finite integral. The completion of HP with respect to
the induced norm p
∥ψ∥ = ⟨ψ, ψ⟩
is the representation space HδP = Hδ (M, L, P ) which we wanted to construct.
72
Every section ψ ∈ Γ(M, L ⊗ δ) can be written in the form ψ = s ⊗ ρ, since there exists a nowhere
vanishing global section ρ of δ.
73
ρ can be chosen to be polarized and nowhere vanishing, hence ∇X (s⊗ρ) = (∇X s)⊗ρ+s⊗∇X ρ = 0
if and only if ∇X s = 0.
206 14. Half-Density Quantization in General

Observation 14.7. Using a different isomorphism

τ = f τP : Γ−1/2 (P ) → Γ−1/2 (Z D )
with the property ∇X ◦ τ = τ ◦ ∇X for all X ∈ Γ(M, DC ) the scalar has the form
f = h ◦ π, cf. Lemma 14.5. The use of τ instead of τP leads to the representation space
Hδ (τ ) which is unitarily equivalent to Hδ by the unitary map
Hδ (τ ) → Hδ , ϕ 7→ hϕ .

14.3 Quantum Operator

Before we state the main result of this section in the theorem below the quantum
operator q δ has to be defined. Let (L, ∇, H) be a prequantum line bundle on the
symplectic manifold (M, ω) with complex polarization P and with the additional line
bundle δ := δ−1/2 (P ). For every directly quantizable observable F ∈ E(M, R)74 the
local flow (ΦFt ) of the Hamiltionian vector field XF on M induces – with the aid of
the naturally lifted vector field ZF on L× – the local one-parameter group (ρFt ) of
transformations, ρFt : Γ(Mt , L) → Γ(M−t , L), where the impact of the connection is
already included as we have deduced in Lemma 7.12. This local one-parameter group
(ρFt ) has as its infinitesimal generator the prequantum operator
i d F
q(F )s = ρ (s) .
2π dt t t=0

See Proposition 7.13. Moreover, the local flow (ΦFt ) of the Hamiltonian vector field XF ,
given by the diffeomorphisms ΦFt : Mt → M−t , induces a pull-back (ΦF )∗t : Γ(Mt , δ) →
Γ(M−t , δ), a local one-parameter group, which we denote again by ΦFt = (ΦF )∗ . We
know
d d
LXF σ = ΦFt σ = (ΦFt )∗ σ
dt t=0 dt t=0
by definition of the Lie derivative LXF .
The two local one-parameter groups (ρFt ) and ΦFt ) on the sections of L resp. δ define
a local one-parameter group
κFt : Γ(Mt , L ⊗ δ) → Γ(M−t , L ⊗ δ).
Locally, we set κFt (φ) = ρFt s ⊗ ΦFt σ for φ = s ⊗ σ.
Definition 14.8. The (half-density) quantum operator q δ (F ) for directly quantizable
F is defined by

i d F
q δ (F )φ := κ (φ) ,
2π dt t t=0

where φ is a section of Γ(M, L ⊗ δ).

74
Only real observables are relevant.
14.3 Quantum Operator 207

Lemma 14.9. For sections φ = s ⊗ σ

i
q δ (F )(s ⊗ σ) = q(F )s ⊗ σ − s ⊗ LXF σ
2π
i
=− ((∇XF s + 2πiF s) ⊗ σ + s ⊗ LXF σ) .
2π

Proof.
i d F
q δ (F )φ = q δ (F )(s ⊗ σ) = ρt s ⊗ Φ∗t σ
2π dt t=0
i d F i d ∗
= ρ s ⊗σ+s⊗ Φσ
2π dt t t=0 2π dt t t=0
i
= q(F )s ⊗ σ − s ⊗ LX σ
2π F

Certainly, the formula

i
q δ (F )(s ⊗ σ) = q(F )s ⊗ σ − s ⊗ LXF σ (60)
2π

can also serve as a definition of the quantum operator.

Observe, that Pt := (ΦF )∗t (P ) = P , when XF is complete, and this implies that the
corresponding representation spaces HPt agree with HP . Therefore, κFt and q δ (F ) is
defined on HP and q δ (F ) is an operator in HP .

Lemma 14.10. q(F )δ is the infinitesimal generator of the one-parameter group (κFt )
of unitary operators.

In particular, when XF is complete, q δ (F ) is self-adjoint. Summerizing:

Main Result

Theorem 14.11 (Half-Density Quantization). Let (L, ∇, H) be a prequantum line

bundle on the symplectic manifold (M, ω) with a reducible complex polarization P . The
half-density quantization for P has as its quantum operators q δ : RP → S(Hδ ) the
maps
i
q δ (F )(s ⊗ ρ) = q(F )s ⊗ ρ − s ⊗ LXF ρ ,
2π
for polarized sections s ⊗ ρ ∈ Γ(M, L ⊗ δ), where q(F ) is the prequantum operator.
Moreover, half-density quantization is a full geometric quantization in the following
sense:
208 14. Half-Density Quantization in General

1. q δ (F ) is R-linear and satisfies (D1) and (D2), now for the new representation
space Hδ = Hδ (M, L, P ) (resp. Hδ (M, L, D), see below).
2. If for F ∈ RP , the vector field XF is complete, then q δ (F ) is self-adjoint.

Let us consider the special case D = 0, i.e. P is a Kähler polarization. Then

M = M/D. In particular, there is no need to search for a natural measure or volume
form, we have the natural volume form ω n on M .
Example 14.12. Let M = T ∗ Rn = Cn be the simple phase space (see Examples
9.16, 10.9, 10.23) in complex coordinates zj = pj + iq j and with symplectic form
ω = 21 i dz̄j ∧ dzj . The holomorphic polarization P is given by
P
 
∂
P := spanE(M ) 1≤j≤n .
∂ z̄j
The prequantum bundle is the trivial bundle L = M × C with connection given by
iX
α= z̄j dzj .
2
The additional line bundle δ = δ−1/2 (P ) is generated by the −1/2-density |dz̄|−1/2 , where
dz̄ is the n-form dz̄ = dz̄1 ∧ . . . ∧ dz̄n .
In order to describe the transformation τP : δ−1/2 (P ) → δ1/2 (M ) (note that Z D = M )
let ρ ∈ Γ−1/2 (P ) be defined by

ρ♯ (X1 , . . . , Xn ) := (|ω n |− /4 )(X1 , . . . , Xn , X1 . . . , Xn ) ,

1

for a basis (X1 , . . . , Xn ) ∈ Pa . Up to a constant ρ coincides with |dz̄|−1/2 . The corre-

sponding τP (ρ) according to (56) is
1
τP (ρ) = |ω n | 2 ∈ Γ 1 (M ) .
2

A general section ψ ∈ Γ(M, L ⊗ δ) has the form ψ = f se ⊗ ρ with f ∈ E(M ) and

n
!!
X π π
se (a) := a, exp − z̄j zj = exp(− z̄z)(a, 1)
j=1
2 2

(with z̄z := ∥z∥2 ), as in Example 10.9.

Since ∇X se := 2πiα(X)se , for X ∈ V(M ), and hence
∇X f se ⊗ ρ = (LX f )se ⊗ ρ
for X ∈ Γ(M, P ), the polarized sections of the line bundle L ⊗ δ are the sections
ψ = f se ⊗ ρ with
∂
f = 0 , j = 1, . . . n .
∂ z̄j
14.4 Other Constructions 209

Consequently, they are in one to one correspondence to the holomorphic functions f

on Cn and the representation space HδP of the half-density quantization is – according
to the construction in Section 14.2 – essentially the Hilbert space
Z
δ
HP = {f ∈ O(C ) |n
f¯f exp(−πz̄z) < ∞} .
Cn

This is the Bargmann-Fock space F, the reduced representation space HP of the

uncorrected quantization as in Example 10.9. As a consequence, the quantum operators
q and q δ are both defined on F with
i
q δ (F )(s ⊗ ρ) = q(F )(s) ⊗ ρ − s ⊗ LXF ρ . (61)
2π
We are interested to know to which extend q(F ) and q δ (F ) differ for directly quantizable
observables F which amounts to determine LXF ρ.
We restrict to real observables, i.e. F has the form F = A+ D̄k zk +Dk z̄k + Cj z̄j zj
P
with real constants A, Cj and complex constants Dk . We concentrat on A = 0 = Dk
and Cj = 12 . Then F is the energy F = H = 21 z̄z of the harmonic oscillator. We have
seen LXH |dz̄| = 0 in the case of n = 1 (cf. Example 12.27) and this result holds true
for arbitrary n. But then we have

LXH ρ = 0 .

As a result, the additional term of the quantum operator q δ (H) in (61) is zero,
and the half-density quantization does not change the Kähler quantization q(H) we
had discussed before in Chapter 10 without using half-densities (see Remark 10.8). In
particular, the half-density quantization does not resolve the problem of the shift in
the spectrum of the quantized energy (cf. Example 10.23).
For complex observables F the partial Lie derivative does not always annihilate |dz̄|
as the Example 12.28 shows. In general, for F = iC(z)z̄z, a non-zero imaginary part
of C(z) will contribute a non-zero factor γ: LXF |dz̄| = γ|dz̄|.

The half-form correction, however, will lead to a correct quantization in the case
of the harmonic oscillator as we show in the next chapter, Chapter 15, on half-form
quantization.

14.4 Other Constructions

Instead of δ−1/2 (P ) one can also use the line bundle δ1/2 (Z D ) as we have done in the
preceding chapter in the case of the momentum space M = T ∗ Q. Moreover, one can
take the line bundle δ = δ−1/2 (DC ) as the additional line bundle, when D ̸= 0, i.e. when
P is not Kähler. The general message is that these other choices lead to the same
half-density quantizations up to unitary equivalence.
210 14. Half-Density Quantization in General

Before we go into details needed for δ = δ−1/2 (DC ) we look at the case of a momen-
tum phase space M = T ∗ Q and compare the half-density quantization of this chapter
with the quantization described in the preceding chapter.

Proposition 14.13. Let P be the vertical polarization on M = T ∗ Q. The construc-

tions using δ = δ1/2 (ZD ) or δ = δ−1/2 (P ) lead to the same half-density quantization.

Proof. We use the line bundle isomorphism τP : δ−1/2 (P ) → δ1/2 (Z D ) introduced in

Proposition 14.3 which induces an isomorphism

Γ(M, L ⊗ δ−1/2 (P )) → Γ(M, L ⊗ δ1/2 (Z D ))

respecting the connections. For polarized sections ψ = s ⊗ ρ, ψ ′ = s′ ⊗ ρ′ of L ⊗ δ−1/2 (P )

the scalar product of HP (with respect to δ−1/2 (P )) is
Z
′
⟨ψ, ψ ⟩ = H(s, s′ )ν(ρ)♯ ν(ρ′ )♯ .
Q

(Recall M/D ∼ = Q.) For the corresponding sections s ⊗ τP (ρ), ψ ′ = s′ ⊗ τP (ρ′ ) of

L ⊗ δ1/2 (Z D ) the scalar product of HP (with respect to δ1/2 (Z D )) is
Z
′ ′
⟨s ⊗ τP ρ, s ⊗ τP ρ ⟩ := H(s, s′ )ν̄ ♯ ν ′♯ ,
Q

according to (52), where ν = ν(τP (ρ)) , ν ′ = ν(τP (ρ′ )). This establishes a unitary
equivalence between the two representation spaces.
For the quantum operators we look at the local situation and assume Q ⊂ Rn . The
quantizable classical variables are the F = A + B j pj with A, B j ∈ E(Q). Then the
Hamiltonian of F is
∂ ∂F ∂
XF = B j j − j .
∂q ∂q ∂pj
The quantum operators agree on the sections s of L. So one only has to show that the
additional terms agree, as well.
In case of δ = δ−1/2 (P ) the partial Lie derivative LXF on δ1 (P ), executes on |dp| ∈
Γ(M, δ1 (P ) yields
 2 
∂ F ∂B j
LXF |dp| = div(XF )|dp| = |dp| = − |dp| .
∂q j ∂pj ∂q j

Hence,

1 ∂B j
LXF |dp|− /2 = |dp|− /2 .
1 1

2 ∂q j
14.4 Other Constructions 211

As a consequence, the additional term is

i 1 ∂B j
− .
2π 2 ∂q j

In Chapter 12 the Hamiltonian vector field X = XF leads to the operator LXF on

sections τ of δr (Z D ) by LXF τ = divν (T π(XF ))τ , where τ is a lift of ν. Note, that

∂
T π(XF ) = B j .
∂q j

The Lie derivative LX of densities on Q is determined by LX |dq| , |dq| ∈ Γ(Q, δ1 (T QC )):

∂X j j ∂
LX |dq| = |dq| , for X = X ∈ V(Q) .
∂q j ∂q j
In particular,
∂B j
LXF |dq| = |dq| .
∂q j
This implies

1/2 1 ∂B j 1
LXF |dq| = j
|dq| /2 .
2 ∂q

Therefore, the additional term is the same as above:

i 1 ∂B j
− .
2π 2 ∂q j

We now want to show that in the case of D ̸= 0 one can use δ = δ−1/2 (DC ) instead
of δ−1/2 (P ). The construction using is similar to the one described in the preceding
sections. One starts with Z D = T M C /DC and considers the lifting of densities from
T (M/D)C to Z D . In Section ?? we recall that densities on Z D are lifts of densities on
T (M/D)C if they are covariantly constant along D (cf. Lemma 14.2). As a further part
of obtaining a suitable descend one proves the analogue of Proposition 14.3, namely

Proposition 14.14. There exists a natural line bundle isomorphism

τD : δ−r (DC ) → δr (Z D )

with
∇X ◦ τD = τP ◦ ∇X
for all X ∈ Γ(M, DC ).
212 14. Half-Density Quantization in General

The definition of τD is simpler than that of τP : With the notation of the proof of
Proposition 14.3 we set for σ ∈ Γ−r (DC ):

τD♯ ([γ]) := ε♯ (β)σ ♯ (ξ D ) .

This result is used in the construction of the representation space on the basis
of δ = δ−1/2 (DC) along the same lines as in the case of δ−1/2 (P ). The outcome is
the representation space Hδ = Hδ (M, L, D) induced by polarized sections of L ⊗ δ.
Moreover, the quantum operator is defined the same way and leads to a corresponding
main theorem, now based on DC and δ = δ−1/2 (DC ).

Remark 14.15. The representation spaces Hδ (M, L, D) and Hδ (M, L, P ) are unitarily
equivalent.

One advantage of using P instead of D – as is done in the literature – is, that the
construction (based on the definition of τP ) has a natural interpretation also for D = 0.
Another advantage of using P instead of D is that a pairing between HP and HP ′
for different complex polarizations can be defined in a natural way as we explain in the
next section.

14.5 Half-Density Pairing

In general, given two different polarizations P, P ′ on a quantizable manifold with pre-

quantum bundle, one should be able to compare the geometric quantization induced
by P with the corresponding geometric quantization induced by P ′ . In particular,
the resulting representation spaces HP and HP ′ should be closely related. In an ideal
case they should be unitarily equivalent by a natural isomorphism and the quantum
operators should be intertwined by this unitary equivalence.
The results of the preceding section can be used to formulate a pairing between
representation spaces HP and HP ′ as a natural relation between these Hilbert paces.

Definition 14.16. Let P, P ′ be complex polarizations on the symplectic manifold

(M, ω). They are called Compatible with each other, when the following three con-
ditions are satisfied.

C1 The intersection D := P ∩ P ′ ∩ T M is an integrable distribution.

C2 D is reducible in the sense that the quotient M/D exists as a manifold such that
the projection π : M → M/D is a submersion.

C3 E := (P + P ′ ) ∩ T M is integrable.

P is called Transversal to P ′ if P ∩ P ′ = {0}, i.e. if D = 0.

14.5 Half-Density Pairing 213

Transversal polarizations are compatible to each other.

It is easy to see, that P is compatible to itself if it is reducible and if E = (P +
P ) ∩ T M ) is integrable, i.e. if P is strongly reducible.
Note, that P ′ is a complex polarization, which we sometimes write Q := P ′ in the
following in order to simplify notation.
In order to construct the pairing we consider the quotient vector bundle Z D :=
T M C /DC , as before. Recall that Z D is essentially the tangent bundle T (M/D)C of the
quotient M/D, and Z D = T M C if D = 0.
The polarization P resp. Q = P ′ has its real distribution DP := P ∩ P ∩ T M
′
resp. DQ = Q ∩ Q ∩ T M = P ∩ P ′ ∩ T M . Moreover, the distribution D as in C1 is
the intersection DP ∩ DQ : By definition D = P ∩ Q ∩ T M contains the intersection
DP ∩ DQ = P ∩ P ∩ Q ∩ Q ∩ T M . Since for X ∈ D = P ∩ Q ∩ T M we have
X = X ∈ P ∩ Q ∩ T M we see X ∈ P ∩ Q ∩ P ∩ Q ∩ T M , hence

D ⊂ DP ∩ DQ = P ∩ P ∩ Q ∩ Q ∩ T M .

Along the same arguments as in the proof of Proposition 14.3 we obtain

Proposition 14.17. Let P be a complex polarization on (M, ω) and let D ⊂ DP

be a distribution such that M/D exists as quotient manifold where the quotient map
M → M/D is a submersion. Then there is a natural line bundle isomorphism

τP : δ−r (P ) → δr (Z D ) with ∇X ◦ τP = τP ◦ ∇X

for all X ∈ Γ(M, DC ) ⊂ Γ(M, P ). Here ∇X are the partial connections on δ−r (P )
resp. on δr (Z D ).

Observation 14.18. Any isomorphism satisfying the compatibility condition ∇X ◦τ =

τ ◦ ∇X for all X ∈ Γ(M, DC ) is of the form f ρ1 → f τ1 , where τ1 is the lift of a nowhere
vanishing ν1 ∈ Γ(M/D, δr (T (M/D)C ) and where ρ1 is a nowhere vanishing section of
δ−r (P ) satisfying ∇X ρ1 = 0 for all X ∈ Γ(M, DC ). We know that ν1 and τ1 exist. And
ρ1 can locally be defined by

ρ♯1 (ξ) := (ε♯ (β)θ♯ (η))−1 τ ♯ ([γ]) ,

where the notation is the one used in the proof of Proposition 14.3.

Now we come to the construction of the pairing on the basis of two compatible
polarizations P and P ′ = Q. Let ψ = s ⊗ ρ ∈ Γ(M, L ⊗ δ)75 resp. ψ = s′ ⊗ ρ′ ∈
Γ(M, L ⊗ δ ′ ) global sections (where δ ′ := δ− 1 (Q)) which are polarized, i.e.
2

′
(∇ ⊗ ∇δ )X ψ = 0 for all X ∈ Γ(M, P ) , (∇ ⊗ ∇δ )X ψ ′ = 0 for all X ∈ Γ(M, Q) .
75
Every global section can be written in this form.
214 14. Half-Density Quantization in General

Then
∇X s = 0 for all X ∈ Γ(M, P ) , ∇X s′ = 0 for all X ∈ Γ(M, Q) ,
and
∇δX ρ = 0 , ∇δX ρ′ = 0 for all X ∈ Γ(M, DC ) .

As a consequence, s and s′ are constant on the leaves of D and the result of Propo-
sition 14.17 implies that τP (ρ) ∈ Γ1/2 (Z D ) resp. τQ (ρ′ ) ∈ Γ1/2 (Z D ) are lifts of suitable
half-densities on M/D which we denote by ν = ν(ρ) resp. ν ′ = ν(ρ′ ). Altogether, the
sections ψ, ψ ′ induce a 1-density
♯
(ψ, ψ ′ ) := H(s, t)ν̄ ♯ ν ′

on M/D, which defines a natural sesquilinear map

Z Z
′ ′ ′ ♯
BP,Q : HP × HQ → C , (ψ, ψ ) 7→ BP,Q (ψ, ψ ) := (ψ, ψ ) = H(s, t)ν ♯ ν ′ .
M/D M/D

′
Recall, that HP = HPδ resp. HQ = HQδ 76 is the space of polarized sections of L⊗δ−1/2 (P )
resp. L ⊗ δ−1/2 (Q) with compact support.
To make the formula for the pairing BP,Q more concrete let us assume, that L is
trivial and has a global nowhere vanishing section s1 ∈ Γ(M, L), a situation which is
locally true. Moreover, let ρ1 resp. ρ′1 be global nowhere vanishing sections of δ =
δ− 1 (P ) resp. δ ′ = δ− 1 (Q) with ∇X ρ1 = 0 for all X ∈ Γ(M, P ) resp. ∇X ρ′1 = 0 for
2 2
all X ∈ Γ(M, Q). Then the polarized sections ψ ∈ Γ(M, L ⊗ δ), ψ ′ ∈ Γ(M, L ⊗ δ ′ )
can be written in the form ψ = ϕs1 ⊗ ρ1 resp. ψ ′ = ϕ′ s1 ⊗ ρ′1 with functions ϕ, ϕ′ ∈
E(M/D). Moreover, τP (ρ1 ) resp. τQ (ρ′1 ) is the lift of a half-density ν1 resp. ν1′ in
Γ(M/D, δ 1 (T (M/D)C )). As a result, the density (ψ, ψ ′ ) has the form (ψ, ψ ′ ) = ϕ̄ϕ′ ν̄1♯ ν1′♯
2
which implies Z
′
B(ψ, ψ ) = ϕ̄ϕ′ ν̄1♯ ν1′♯ . (62)
M/D

Note that ν1 and ν1′

can be chosen to be µ1/2 with an everywhere positive 1-density µ
on M/D77 . With this choice the formula simplifies to
Z
′
B(ψ, ψ ) = ϕ̄ϕ′ µ♯ .
M/D

Moreover, B is non-degenerated.
We have defined a natural pairing B = BP,Q on HP × HQ which is directly induced
by the construction of the half-density representation spaces. To obtain a map T =
TP,Q : HP → HQ from this pairing the continuity of B would be helpful.
76
In the following we omit the upper indices δ, δ ′ for convenience.
77
Such a µ exists according to Proposition 12.5.
14.5 Half-Density Pairing 215

Lemma 14.19. When B is partially continuous, i.e. B is continuous in each variable,

there exists a linear map T : HP → HQ such that for each ψ ∈ HP

B(ψ, ψ ′ ) = ⟨T ψ, ψ ′ ⟩

for all ψ ′ ∈ HQ .

Proof. In fact, fixing ψ ∈ HP defines a continuous and linear map

HQ → C , ψ ′ 7→ B(ψ, ψ ′ ) ,

which can be extended to HQ as a continuous linear functional Bψ : HQ → C. By the

Riesz representation theorem the continuous linear functional Bψ can be represented
by a unique T ψ ∈ HQ in the sense that B(ψ, ψ ′ ) = ⟨T ψ, ψ ′ ⟩ for all ψ ′ ∈ HQ . The map
T : HP → HQ is linear and injective.

When B is continuous, i.e

B(ψ, ψ ′ ) ≤ C ∥ψ∥ ∥ψ ′ ∥ , (ψ, ψ ′ ) ∈ HP × HQ

for a constant C, then the map T = TP,Q in the preceding lemma is continuous as well
and can be continued to all of HP yielding a continuous and bijective T : HP → HQ .
However, T will not be unitary, in general.
There seems to be no general method to show the continuity of B. One of the
problems to implement such a method originates in the fact, that the norms of HP
resp. HQ are, in general, not related to the definition of the form B: The three distri-
butions D = P ∩ Q ∩ T M , DP := P ∩ P ∩ T M , DQ := Q ∩ Q ∩ T M are, in general
different form each other. Therefore, the integration over M/D, which is used to define
B, is not related to the integration over M/DP , which is used to define the norm on
HP and not related to the integration over M/DQ , which is used to define the norm of
HQ , Q = P ′ .
Summarizing, we cannot be sure that the pairing is defined on HP × HQ in a
reasonable way, nor that it is continuous. However, in important cases T turns out to
be well-defined and unitary as we see in the following proposition.

Proposition 14.20 (Fourier Transform). Let M = T ∗ Rn be the momentum phase

space (simple case) with standard symplectic form ω = dq j ∧ dpj and prequantum
bundle (L, ∇, H), where L = M × C, ∇ given by the connection form −pj dq j and H
the constant Hermitian metric on M × C. In the case of the vertical polarization P and
the horizontal polarization P ′ = Q the natural pairing BP,Q is continuous on HP × HQ ,
and the corresponding map TP,Q is unitary. T is the Fourier transform up to a scaling
factor.
216 14. Half-Density Quantization in General

Proof. P is transversal to Q and Z D = T M C , since D = P ∩ Q ∩ T M = {0}. The

connection can be written in the form

∇X f s1 = LX f − 2πi pj dq j (X)f s1

for a general section s = f s1 , f ∈ E(M ), of L. Here, s1 is the special section s1 (a) =

(a, 1) , a ∈ M , as before.
Let dp denote the n-form dp := dp1 ∧ dp2 ∧ . . . ∧ dpn ∈ Γ(M, Λn (P ∨ )) with the
densities |dp| = |dp1 ∧ dp2 ∧ . . . ∧ dpn | ∈ Γ(M, δ1 (P )) and |dp|−1/2 ∈ Γ(M, δ−1/2 (P )).
Any −1/2-density of P is of the form f |dp|−1/2 with f ∈ E(M ). Hence, a general section
ψ = s ⊗ ρ ∈ Γ(M, L ⊗ δ) with δ = δ−1/2 (P ) can be written as

ψ = ϕs1 ⊗ |dp|− /2
1

with ϕ ∈ E(M ). The generating section |dp|−1/2 is polarized with respect to P , since
∂ ∂
|dp|− /2 (
1
,..., ) = 1,
∂p1 ∂p1

so that ∇X (|dp|−1/2 ) = 0 for X ∈ Γ(M, P ). As a consequence, the section ψ = ϕs1 ⊗

|dp|−1/2 ∈ Γ(M, L ⊗ δ) is polarized if and only if ϕs1 is polarized and this is equivalent
to
∂
ϕ = 0, j = 1 ...,n.
∂pj
Hence, we have recovered the result of Construction 13.4 that the space of polarized
sections is

Γ∇⊗∇δ ,P (M, L ⊗ δ) = {f s1 ⊗ |dp|− /2 | f = f (q) ∈ E(Rn )} ,

1

and HδP (M ) can be identified with L2 (Rn , dλ(q).

In the same way, dq := dq 1 ∧ dq 2 ∧ . . . ∧ dq n ∈ Γ(M, Λn (Q∨ )) induces the polarized
section |dq|−1/2 ∈ Γ(M, δ−1/2 (Q)). And a section ψ ′ = ϕ′ s1 ⊗ |dq|−1/2 ∈ Γ(M, L × δ ′ ),
where δ ′ = δ−1/2 (Q) and ϕ′ ∈ E(M ), is polarized if an only if
∂ ′
ϕ − 2πipj ϕ′ = 0 , j = 1 . . . , n .
∂q j

The general solution of this system of differential equations is

j
ϕ′ (q, p) = f ′ (p)e2πipj q , (q, p) ∈ T ∗ Rn = Rn × Rn ,

with an arbitrary function f ′ = f ′ (p) ∈ E(Rn , C).

Hence, the space of polarized sections with respect to the horizontal polarization Q
is
j
Γ∇⊗∇δ′ ,Q (M, L ⊗ δ ′ ) = {f ′ e2πipj q s1 ⊗ |dq|− /2 | f ′ = f ′ (p) ∈ E(Rn )} ,
1
14.5 Half-Density Pairing 217

and HδP (M ) can be identified with {f ′ e2πpj q | f ′ ∈ L2 (Rn , dλ(p))} ∼

j
= L2 (Rn ).
The induced 1-density (ψ, ψ ′ ) on T ∗ M C = Z D – which leads to the form BP,Q :
HP × HQ → C according to the construction explained above (see (62)) – is
(ψ, ψ ′ ) = ϕ̄ϕ′ |dq||dp| = ϕ̄ϕ′ |dq ∧ dp| .
Hence, Z Z
′ ′
BP,Q (ψ, ψ ) = (ψ, ψ ) = ϕ̄ϕ′ dq ∧ dp
ZM M
′ 2πipj q j
= f (q)f (p)e dq ∧ dp
M
Z Z 
= f (q)e −2πipq dq f ′ (p)dp .
Rn Rn

B = BP,Q is continuous, since it is bounded:

Z sZ Z
′ ′
|B(ψ, ψ )| ≤ |f ||f | ≤ |f |2 ) |f ′ |2 = ∥ψ∥∥ψ ′ ∥

for (ψ, ψ ′ ) ∈ HP × HQ . Therefore, B can be extended uniquely to HP × HQ as a

continuous bilinear form. Furthermore, T : HP → HQ defined by T ψ = T f s1 ⊗ |dq|−1/2
with Z
j
T f (p) := f (q)e−2πipj q dq ,
Rn
satisfies Z
′
B(T ψ, ψ ) = T f f ′ dp = ⟨T f, f ′ ⟩ = ⟨T ψ, ψ ′ ⟩
Rn
for all (ψ, ψ ′ ) ∈ HP × HQ .
T is linear, continuous and bijective. Since ⟨T f, T f ⟩ = ⟨f, f ⟩ the map T is even
unitary. Up to a scaling factor c > 0, T is the Fourier transform. Altogether we have
a natural unitary mapping T : HP → HQ between the representation spaces HP and
HQ .
Remark 14.21. T intertwines the quantum operators Qj = q δ (qj ), Pj = q δ (pj ),
′ ′
resp. Q′ j = q δ (qj ), Pj′ = q δ )pj ) in the sense that T ◦ Qj = Q′ j ◦ T resp. T ◦ Pj = Pj′ ◦ T .
In other words, T intertwines the representations
′
q := q δ resp. q ′ := q δ
of the algebra o := {a + bj pj + ck q k | a, bj , ck ∈ R}, a subalgebra of the Poisson algebra
(R2n , R). This means that the following diagram is commutative

HP T / HQ
q q′
 
HP / HQ
T
218 14. Half-Density Quantization in General

In fact, it can be shown that Pj′ = pj and

j i ∂
Q′ = .
2π ∂pj

Hence Z
j
k
T ◦ Q (f )(p) = q k f (q)e−2πipj q dq

and Z
′k i ∂ j
Q T (f )(p) = f (q)e−2πipj q dq
2π ∂pk
Z
i j
= f (q)(−2πiq k )e2πipj q dq
2π
Z
j
= q k f (q)e−2πipj q dq = T ◦ Qk (f )(p) .

As a consequence, T ◦ Qk = Q′ k ◦ T . T ◦ Pj = Pj′ ◦ T can be shown in an analoguous

manner.

Summary
 219

15 Half-Form Quantization

In principle, half-form quantization is similar to half-density quantization. The differ-

ences are:

1. Half-form quantization corrects the shift of the eigenvalues in many cases, which
is not achieved by the half-density approach. Recall, that the geometric quanti-
zation of the harmonic oscillater using the holomorphic polarization (i.e. Kähler
quantization) leads to an incorrect model with a shift of the eigenvalues of the
quantized energy operator, see Example 10.23, and that the half-density quan-
tization does not change this, see Example 14.12. However, in the sketch of
the half-form quantization for the harmonic oscillator in the Example 10.25 we
arrived at the correct model.

2. Half-form quantization is only possible when a certain topological condition of

the phase space (M, ω) (more precisely of the frame bundle R(P ) assigned to
P ) is satisfied, while half-density quantization works without any additional as-
sumption.

3. The topological condition forces one to consider rather involved new structures,
like the concept of a metalinear frame bundle as a special case of a metalinear
structure for a principal fibre bundle with structure group GL(n, C) or the concept
of a metaplectic structure. These structures make the half-form quantization less
accessible, although from an elementary standpoint forms can be considered as
to be more basic than densities.

In this chapter we present the half-form quantization without using the concept of
a metalinear or metaplectic structure. Instead, we only need the existence of a square
root line bundle S of a certain line bundle K, i.e. S ⊗ S ∼ = K, where K = K(P ) is
naturally induced by the polarization P , it is the canonical bundle of P .
In general, such a square root bundle S will not exist. And in case of existence there
might be several inequivalent choices. Existence is guaranteed if a certain cohomology
class induced by P vanishes. We give a detailed explanation of this result in the Section
15.5 below. Note, that a given metalinear structure for the frame bundle R(P ) of P
always induces such a square root.
Interestingly enough, the condition which is necessary for the existence of a square
root bundle of K is exactly the same condition which ensures the existence of a met-
alinear frame bundle associated to the frame bundle R(P ). A metalinear frame bundle
immediately leads to a square root S and the metalinear structure allows one to define
this bundle by transformation properties of the sections similar to the properties of
half-densities. We come back to metalinear structures in the next chapter where also
metaplectic structures are studied. A metaplectic structure on (M, ω) induces met-
alinear frame bundles for several different polarizations at once and thus opens the
220 15. Half-Form Quantization

possibility of comparing polarizations and the corresponding half-density quantization

in an efficient and elegant way, in particular in order to construct a natural pairing.
We study all this in the next chapter.

15.1 Canonical Bundle of a Vector Bundle

In order to define half-forms of a polarization we introduce the notion of the canonical

line bundle of a general complex vector bundle.
Definition 15.1. Let V → X be a complex vector bundle of rank k over an m-
dimensional manifold X: Then K(V ) := Λk (V ∨ ) is called the Canonical Bundle of
V.

Note, that the canonical bundle of the tangent bundle T X C on the m-dimensional
manifold X is the line bundle of forms of top degree m on X: Λm (T ∗ X C ) ∼ =
m C ∨
Λ ((T X ) ) = K(T X ). This line bundle K(T X ) is sometimes denoted by K(X)
C C

and called the canonical bundle of X. For η ∈ RK(X) we have the standard integral of
m-forms with respect to an orientation of X: X η . Without orientation a reasonable
general integration is only possible for densities.
Lemma 15.2. Let V → X be a complex vector bundle of rank k with transition func-
tions gij with respect to an open cover (Uj ) of X. Then det gij−1 are suitable transition
functions of the canonical bundle K(V ) = Λk (V ∨ ). Moreover, K(V ) is (isomorphic to)
the line bundle R(V ) ×GL(k,C) C associated to the frame bundle R(V ) of V with respect
to the representation ρ = det−1 : GL(k, C) → C× . As a result, the global sections
α ∈ Γ(M, K(V )) can be identified with the smooth functions α♯ : R(V ) → C on the
frame bundle R(V ) of V satisfying the equivariance property

α♯ (bg) = (det g) α♯ (b)

for all b ∈ R(V ) and g ∈ GL(k, C) (for the notation bg see Section 12.1).

Proof. The dual vector bundle V ∨ has the transition functions gij−1 which implies that
det gij−1 = (det gij )−1 are transition functions of Λk (V ∨ ). From this result we can read off
that K(V ) can also be defined as the line bundle R(V ) ×ρ C with respect to ρ = det−1 ,
since, in general, the transition functions of such associated bundle are ρ(gij ) (cf. D.9).
Finally, the transformation property for α♯ is the general transformation property in
case of associated bundles (cf. D.7) which we have encountered in a similar form in
the case of r-densities (cf. 12.1).

The correspondence α 7→ α♯ is very simple in the case of the canonical bundle:

Each section α ∈ Γ(M, K(V ) induces a family of maps αx : Vxk → C , x ∈ X, and
therefore, α♯ (b) := αx (b1 , . . . , bk ) is well-defined for each basis b = (b1 , . . . , bk ) of Vx . In
other words, α♯ is the restriction of α to R(P ).
15.1 Canonical Bundle of a Vector Bundle 221

As a result of the lemma, there is a similarity of sections α of K(V ) with 1-densities

on V : The line bundle δ1 (V ) of 1-densities on V is determined by transition functions
| det gij |−1 . And the sections µ ∈ Γ(M, δ1 (V )) are in bijective correspondence to maps
µ♯ : R(V ) → C with the equivariance property

µ♯ (bg) = |det g|µ♯ (b)

for all b ∈ R(V ) and g ∈ GL(k, C), cf. Section 12.1. In particular, any α ∈ Γ(M, K(V ))
induces a 1-density |α| ∈ Γ(M, δ1 (V )).

Notation. The sections α ∈ Γ(X, K(V )) are called 1-forms of V . Of course, this
concept of a 1-form is different from the notion of a differential one form η ∈ A1 (X).

The transformation property of the canonical bundle K(V ) gives rise for introducing
the following definition.

Definition 15.3. Let V → X be a complex vector bundle of rank k over an m-

dimensional manifold X. Then for ℓ ∈ N: Kℓ (V ) := K(V )⊗ℓ 78 , K−ℓ (V ) := K(V ∨ )⊗ℓ ∼
=
∨
Kℓ (V ) .

Observation 15.4. Let gij transition functions of V , then (det gij )−ℓ are suitable
transition functions of Kℓ (V ) for ℓ ∈ Z. And the equivariance property of sections
β ∈ Γ(M, Kℓ (V ) is
β ♯ (bg) = (det g)ℓ β ♯ (b)
for b ∈ R(V ) and g ∈ GL(k, C). Moreover, Kℓ (V ) can also be described as the
associated bundle R(V ) ×GL(k,C) C with respect to the representation ρ = det−ℓ .

Remark 15.5. A concept which is similar to the canonical bundle is the determinant
line bundle which is simply the dual of the canonical bundle (up to isomorphism): For
a vector bundle V of rank r we have the following isomorphisms

det(V ) := Λr (V ) ∼
= (Λr (V ∨ )∨ ∼
= K(V )∨ = K−1 (V ) .

In generalization to the sequence Kℓ (V )(ℓ∈Z) of ”canonical” bundles we intend to

introduce Kℓ (V ) for half-integers ℓ ∈ {ℓ | 2ℓ ∈ Z} or at least for ℓ = 1/2 and −1/2. This
means we need to require the existence of a square root S of K(V ) in order to define
the new sequence by Kℓ (V ) := S 2ℓ for ℓ ∈ {ℓ | 2ℓ ∈ Z}. Note, that S is a square root
of K(V ) = K1 (V ) if and only if S ∨ is a square root of K−1 (V ):

Observation 15.6. One might be tempted to define a square root of K(V ) by simply
requiring the transformation property
1
u(bg) = (det g) /2 u(b) ,
78
W ⊗ℓ is the ℓ-fold tensor product W ⊗ W ⊗ . . . ⊗ W
222 15. Half-Form Quantization

for a function u : R(V ) → C. Or equivalently, to define the square root of K(V )

through the transition functions (det gij )−1/2 , when gij are transition functions for V .
But the square root (det h)−1/2 is not well-defined for general h ∈ E(M, GL(k, C)).
In general, for a manifold M every given function g ∈ E(M, C) has a square root in
E(M, C) if and only if M is simply connected.
Let us have a look at the related problem of describing a possible square root of a
given holomorphic function f : U → C on an open U ⊂ C in Complex Analysis: One
e → U of U on which a holomorphic fe : U
uses a double covering p : U e → C is defined
with fe = f ◦ p and such that fe has a holomorphic square root g : Ũ → C, i.e. g 2 = fe.
On the basis of this idea to obtain a square root of K(V ) one can consider a double
cover of GL(k, C) – namely the metalinear group ML(n, C) – together with a suitable
double cover R(V
e ) of R(V ). This will be carried through in Section 16.1 of the next
chapter.

We postpone the discussion of these problems to the next chapter. In the actual
chapter we want to circumvent these issues and treat the half-form quantization under
the hypothesis that we only have a line bundle S with S ⊗ S ∼= K(P ). Before that, let
us comment the usage of the term ”canonical bundle”:
Remark 15.7. We have introduced the notion of canonical line bundle in Definition
15.1 which works for a general complex vector bundle. The usage in the literature
is not completely uniform. In the context of Geometric Quantization the canonical
bundle of a polarization P ⊂ T M C is often defined as Λn P 0 where P 0 ⊂ T ∗ M C is
the polar or annihilator P 0 := {µ ∈ T ∗ M C | µ|P = 0}. Since P is isomorphic to P 0
when P is a polarization79 this definition amounts essentially to Λn P 0 ∼ = Λn P ∼= KP−1 .
i.e. the canonical line bundle defined with the aid of P 0 is the inverse of the canonical
bundle K = KP used in these Notes, and changing the definitions leads merely to
interchanging Kℓ and K−ℓ , ℓ ∈ Z, as well as K1/2 and K−1/2 in case of existence. We
have introduced the general Definition 15.1 in accordance with the usage in Algebraic
Geometry and Complex Analysis, where the canonical line bundle for a non-singular
variety X is the line bundle of forms of top degree.

15.2 Need for a Square Root of a Line Bundle

This section is inserted into this chapter in order to motivate the search for a square
root of K(V ) resp. K−1 (V ). The results are not needed in the following, so the reader
can proceed immediately with the next two sections to learn how the square root is
applied to achieve the half-form quantization.
Recall that a square root of a line bundle B over M is a line bundle C with
C := C ⊗ C = B. We also speak of a square root when S ⊗ S ∼
2
= B with a fixed
line bundle isomorphism S ⊗ S → B.
79
with respect to the isomorphism X 7→ ω(X, ) , X ∈ Pa
15.2 Need for a Square Root of a Line Bundle 223

Our starting point is a prequantum line bundle (L, ∇, H) on a symplectic manifold

(M, ω) with a reducible complex polarization P . The polarized sections s ∈ Γ(M, L)
are constant on the leaves of the distribution D = P ∩ P ∩ T M and consequently they
define sections on the quotient manifold M/D with values in the induced line bundle
L|M/D . In particular, for two polarized sections s, t ∈ Γ(M, L) the Hermitian structure
H on L yields a function H(s, t) ∈ E(M/D). Such functions we intend to integrate in
order to obtain the representation space (a Hilbert space) of the quantization. However,
as we have stated already several times, there is, in general, no natural measure on
M/D. Therefore, as a first possibility one studies densities on M/D which can be
integrated without depending on an extra structure on M/D, e.g. like a volume. This
approach leads to the half-density quantization considered in the preceding chapter.
If one wants to base the quantization on forms instead on densities, one sees that
sections on K−1 (P ) can be directly composed to yield 2-densities on the quotient Q :=
M/D. We explain this in the case of a reducible real polarization P , i.e. DC = P = P :
Let Z D = T M C /DC ∼ = π ∗ (T (Q)C ) be the pullback of T QC with respect to the quotient
map π : M → M/D = Q.
Lemma 15.8. Let P = P . There is a natural sesquilinear pairing

⟨ , ⟩ : K−1 (P ) × K−1 (P ) → δ2 (Z D )

defined by

⟨α, β⟩♯ ([Z1 ], . . . , [Zn ]) := |ω n (X1 , . . . , Xn , Z1 , . . . , Zn )|2 α♯ (X1 , . . . Xn )β ♯ (X1 , . . . , Xn ) ,

where (α, β) ∈ K−1 (P ) × K−1 (P ) and (X1 , . . . , Xn , Z1 , . . . , Zn ) ∈ Ra (T M C ) such that

(X1 , . . . , Xn ) ∈ Ra (P ) and ([Z1 ], . . . , [Zn ]) ∈ Ra (Z D ) (recall [Z] := Z + Pa ∈ Z D a for
Z ∈ Ta M C ).

Proof. For a fixed Z = (Z1 , . . . , Zn ) the definition is independent of X = (X1 , . . . , Xn ):

For another basis X ′ of Ra (P ) there exists g ∈ GL(n, C) with X ′ = Xg; and

|ω n (Xg, Z)|2 α♯ (Xg)β ♯ (Xg) = | det g|2 |ω n (X, Z)|2 (det g)−1 α♯ (X)(det g)−1 β ♯ (X)

= |ω n (X, Z)|2 α♯ (X)β ♯ (X) ,

Replacing Z by Z̃ = Z + Y, Yj ∈ P the following holds for fixed X

|ω n |2 (X, Z̃)α♯ (X)β ♯ (X) = |ω n |2 (X, Z)α♯ (X)β ♯ (X),

hence ⟨α, β⟩ is well-defined. Moreover, for g ∈ GL(n, C)

⟨α, β⟩♯ ([Zg]) = |ω n (X, Zg)|2 α♯ (X)β ♯ (X)

= | det g|2 |ω n (X, Z)|2 α♯ (X)β ♯ (X)
= | det g|2 ⟨α, β⟩♯ ([Z])
Therefore, ⟨α, β⟩ is a 2-density.
224 15. Half-Form Quantization

Disregarding at the moment the question of how to descend from Z D to Q = M/D

(which has been settled in the preceding chapters) we see that (α, β) would lead to a
suitable 1-density on M/D if one would be able to take square roots of the occurring
objects. This amounts to take square roots of the involved line bundles, in particular
of K−1 (P ).
Let us explain this idea: If S is a square root of K−1 (P ), i.e. S ⊗ S = K−1 (P ), we
obtain for sections α, β ∈ Γ(M, S) a natural 1-density ⟨α, β⟩S ∈ Γ(M, δ1 (Z D )) defined
by

⟨α, β⟩♯S ([Z1 ], . . . , [Zn ]) = |ω n (X1 , . . . , Xn , Z1 , . . . , Zn )|α♯ (X1 , . . . Xn )β ♯ (X1 , . . . , Xn ) ,

as in the preceding lemma.

Furthermore, ⟨α, β⟩S descends to a 1-density on Q = M/D, if the α, β are constant
along D, i.e. if they are polarized. This is the starting point of forming a representation
space from polarized sections of L ⊗ S as the first main step in half-form quantization.

15.3 Descend of Half-Forms

We present in this section a direct approach to the half-form quantization which avoids
the use of metalinear or metaplectic structures on the vector bundles in question. In
this way we obtain a straightforward and rather elementary construction similar to
the half-density quantization. But we cannot obtain stronger results on isomorphisms
of the relevant line bundles or relations between square roots, since we are not yet
in the position to describe half-forms by transformation properties on the frame bun-
dles. The metalinear structure will enable us to formulate the half-form condition via
transformation properties (cf. next chapter).
In the following, we assume that there exists a square root S of K−1 (P ), i.e. a line
bundle S with S ⊗ S = K−1 (P ), or – more generally – a line bundle S with a fixed
isomorphism S ⊗ S → K−1 (P ). S will be denoted also by S = K−1/2 (P ). The existence
and uniqueness of such a line bundle will be discussed in the next section. A section
in K−1 (P ) will be called a −1-P -form and correspondingly a section of S = K−1/2 (P )
will be called a −1/2-P -form. The term ”half-form” will be used, in general, to denote
sections of K1/2 (V ) or K−1/2 (V ) for a vector bundle, in particular for P = V .
As before, P is a reducible complex polarization on a symplectic manifold (M, ω)
and we consider the quotient bundle Z D = T M C /DC which can be described as the
pullback of T QC where Q := M/D is the space of leaves of the distribution D =
T M ∩ P ∩ P with its projection π : M → Q = M/D. We intend to define a natural
sesquilinear map (a pairing) BP : K−1/2 (P ) × K−1/2 (P ) → δ1 (ZD ), as was explained in
the preceding section. We need the following elementary result:

Lemma 15.9. Any two −1/2-P -forms α, α′ ∈ Γ(M, K−1/2 (P )) define a −1-density
ᾱα′ ∈ Γ(M, δ−1 (P )).
15.3 Descend of Half-Forms 225

Proof. The squares α2 = α ⊗ α , α′ 2 = α′ ⊗ α′ are in Γ(M, K−1 (P )), hence, the product
β := ᾱ2 α′ 2 = (ᾱα′ )2 is a −2-density β ∈ Γ(M, β−2 (P )): Indeed, for b ∈ Ra (P ) and
g ∈ GL(n, C) one has
2 2
β ♯ (bg) = (ᾱ2 )♯ (bg)(α′ )♯ (bg) = (det g)−1 (ᾱ2 )♯ (b)(det g)−1 (α′ )♯ (b) = | det g|−2 β ♯ (b) .
√
Hence, ᾱα′ = β = β 1/2 is a −1-density80 on P .
Another way to see that ᾱα′ is a 1-density is to determine the transition functions
for the line bundle
K −1/2 (P ) ⊗ K−1/2 (P ) :
Let zij be transition functions for of K−1/2 (P ) . Then z̄ij zij = |zij |2 are transition func-
tions of K −1/2 (P ) ⊗ K−1/2 (P ). Since |zij |2 = | det gij |, they are also transition functions
of of the density bundle δ−1 (P ) according to Definition 12.2. Hence the two line bundles
are isomorphic. In particular, the sections of K −1/2 (P )⊗K−1/2 (P ) are −1-densities.

In other words, the line bundle K −1/2 (P ) ⊗ K−1/2 (P ) is trivial and isomorphic to
the line bundle δ−1 (P ). The choice of an isomorphism K −1/2 (P ) ⊗ K−1/2 (P ) → δ−1 (P )
determines a non-degenerate sesquilinear pairing K−1/2 (P ) × K−1/2 (P ) → δ−1 (P ) and
vice versa.
We know from Proposition 14.3 that there exists a natural isomorphism

τP : δ−1 (P ) → δ1 (Z D )

preserving the partial connections.

Corollary 15.10. We thus obtain a natural pairing map

⟨ , ⟩P : K−1/2 (P ) × K−1/2 (P ) → δ1 (ZD ) , 7→ ⟨α, α′ ⟩P := τP (ᾱα′ ) ,

where ᾱα′ is the −1-density from the previous lemma.

Because of the central importance of the statement of the corollary, we recall the
resulting definition of the pairing ⟨ , ⟩P in detail: It is enough to define the map locally
around each a ∈ M .
In a suitable open neighbourhood U ⊂ M of a ∈ M there exists a frame field

(ξ, ζ) = (ξ1 , . . . , ξn , ζ1 , . . . , ζn ) : U → R(T M C )

for T M C such that

80
Here we notice a little disadvantage resulting from the decision to avoid the use of metalinear
frames in this chapter: We cannot give a seemingly obvious proof by establishing the transformation
♯
property for (ᾱα′ )♯ through inserting bg in ᾱ♯ and in α′ , since we cannot use a transformation property
like α(bg) = (det g)− /2 α(b). Such a transformation property does not hold. It is, however, satisfied
1

on the appropriate double cover of R(P ) as we explain in the next chapter.

226 15. Half-Form Quantization

1. ξ : U → R(P ) is a frame field for P ,

2. (ξ1 , . . . ξk ) : U → R(DC ) is a frame field for DC ,
3. the [ξj ] := ξj + DC ∈ Z D and [ζj ] := ζj + DC ∈ Z D define a frame field
[γ] := ([ξk+1 ], . . . , [ξn ], [ζ1 ], . . . , [ζn ]) : U → R(Z D ) for Z D , and
4. the [ξj ] ∈ P C /DC and their conjugates [ξ¯j ] := ξ¯j + DC ∈ P /DC yield a
C

frame field η := ([ξk+1 ], . . . , [ξn ], [ξ¯k+1 ], . . . [ξ¯n ]) : U → R(E C /DC ) for E C /DC . Here,
E = T M ∩ (P + P ) and therefore, E C = P + P .
Finally, let θ be the 1-density on E C /DC induced by |ω n−k |1/2 , i.e.

1/2
θ♯ (η) = |ω n−k (ξk+1 , . . . , ξn , ξ¯k+1 , . . . , ξ¯n )| .

Definition 15.11. For every pair (α, α′ ) ∈ Γ(U, K−1/2 (P )) × Γ(U, K−1/2 (P )) we define
⟨α, α′ ⟩P ∈ Γ(U, δ1 (ZD )) by

⟨α, α′ ⟩♯P ([γ]) := |ω n (ξ, ζ)|θ♯ (η)(αα)♯ (ξ).

Proposition 15.12. ⟨α, α′ ⟩P yields a well-defined 1-density in Γ(M, δ1 (Z D )). More-

over, it is a lift of a unique 1-density ν(α, α′ ) ∈ Γ(Q, δ1 (T QC )) on Q = M/D if and
only if it is polarized and this is the case if the −1-density ᾱα′ satisfies ∇X (ᾱα′ ) = 0 for
all X ∈ Γ(M, D), where ∇ is the partial connection on δ−1 (P ) (cf. Definition 12.16).

Proof. To be completed in a similar way as before in the preceding chapter.

Partial Connection
To exploit the last statement about describing the density ⟨α, α′ ⟩P as the lift of a
density on Q := M/D, we need to define a partial connection also on our square root
line bundle S = K−1/2 (P ). We follow the definition of the partial connection on δr (P )
(see Section 12.3), but now for the line bundle Kℓ (P ) , 2ℓ ∈ Z.
We begin with ℓ ∈ Z. Locally, on a suitable (e.g. for example contractible) open
subset U ⊂ M there exists a Hamiltonian frame field ξ : U → R(P ). This frame field
determines a unique σξ ∈ Γ(U, Kℓ (P )) with σξ♯ (ξ) = 1.

Definition 15.13. Let ℓ ∈ Z. Every σ ∈ Γ(U, Kℓ (P )) has the form σ = f σξ with

f = σ ♯ (ξ). For X ∈ Γ(U, P ) the partial connection is defined by

∇X σ := (LX f )σξ = LX σ ♯ (ξ) σξ .

In particular, ∇X σξ = 0, i.e. σξ is polarized. For ℓ ∈ { 21 , − 12 } the definition of ∇X

is given below in Corollary 15.15. See also Section 16.2.
The partial connection is well-defined for all X ∈ Γ(M, P ) and ℓ ∈ Z (i.e. indepen-
dent of the choice of the frame field ξ), and satisfies the properties of a flat connection.
15.3 Descend of Half-Forms 227

This can be shown in the same way as for the partial connection on δr (P ) in Section
12.3.
Moreover, an analogous partial connection is given on the conjugate canonical bun-
dles K ℓ (P ) := {ρ̄ | ρ ∈ Kℓ (P )} which can be defined also by

∇X ρ̄ := ∇X ρ.

Lemma 15.14. All these partial connections are compatible to each other, i.e. for
instance
∇X (σρ) = (∇X σ)ρ + σ∇X ρ
for σ ∈ Γ(M, Km (P )), ρ ∈ γ(M, Kℓ (P )) where σρ = σ ⊗ ρ ∈ Γ(M, Km+ℓ (P )).

Proof. Given the Hamiltonian frame field ξ the sections σξ ∈ Γ(U, Km (P )) , ρξ ∈

Γ(U, Kℓ (P )) and τξ ∈ Γ(U, Km+ℓ (P )) are determined and they satisfy τξ = σξ ρξ . For
general sections σ = f σξ , ρ = gρξ and τ = σρ = f gτξ we have

∇X τ =LX (f g)τξ = ((LX f )g + f LX g)σξ ρξ

=(LX f σξ )gρξ + f σx iLX gρξ
=(∇X σ)ρξ + σx i∇X ρξ .

Corollary 15.15. In particular, ∇X α2 = 2α∇X α for α ∈ Kℓ (P ) and for nowhere

vanishing α
1
∇X α = ∇X α 2 .
2α
This result is used to define ∇X = ∇SX on the root bundle S = K−1/2 or S = K−1/2 :
For α ∈ Γ(U, S) we set
1
∇SX α = ∇X α 2 .
2α

The compatibility extends also for mixing the conjugate with non-conjugate sections
of Kℓ (P ), Km (P ).
For instance, for σ, ρ ∈ Γ(M, K−1/2 (P )) the following holds:

∇X (σ̄ρ) = (∇X σ̄)ρ + σ̄(∇X ρ). (63)

Again there can be proven many compatibility conditions for all these partial con-
nections, including the special connections ∇S . For instance

∇X (ᾱα′ ) = (∇SX ᾱ)α′ + ᾱ∇SX α′ , (64)

for any pair of sections α, α′ ∈ Γ(M, K−1/2 (P )).

228 15. Half-Form Quantization

Lemma 15.16. For polarized sections α, α′ ∈ Γ(M, K−1/2 (P )) the density ᾱα′ ∈
Γ(M, δ−1 (P )) satisfies
∇X (ᾱα′ ) = 0
for X ∈ Γ(M, D). Here, ∇ is the partial connection on δ1 (P ) defined in 12.16.

Proof. This follows immediately from (64).

Collecting these results we obtain:

Corollary 15.17. For any pair of polarized sections α, α′ ∈ Γ(M, K−1/2 (P )) the density
⟨α, α′ ⟩P ∈ Γ(δ1 (Z D )) is the lift of a unique 1-density ν(α, α′ ) ∈ Γ(Q, δ1 (T QC )).

15.4 Representation Space and Quantum Operator

We are now in the position to describe the half-form quantization in full generality:

Construction 15.18 (Representation Space, Using K−1/2 (P )). Let (L, ∇, H) be a

prequantum line bundle on the symplectic manifold (M, ω) and let P be a reducible
complex polarization on (M, ω). In addition, we assume that the line bundle K−1 (P )
has a the square root S = K−1/2 (P ).
As before, we replace L with

L ⊗ S , S = K−1/2 (P ) ,

and ∇ with ∇⊗∇S , where ∇S is the partial connection on S = K−1/2 (P ) (see Definition
15.13).
Let ψ = s ⊗ α, ψ ′ = s′ ⊗ α′ ∈ Γ(M, L ⊗ S) polarized sections. Then s, s′ are
polarized, hence ∇X H(s, s′ ) = 0 for all X ∈ Γ(M, DC ), i.e. H(s, s′ ) is constant on
the leaves of D and therefore descends to a smooth function on Q = M/D which we
denote by H(s, s′ ) again. Moreover, the sections α, α′ of S are also polarized, which
by Corollary 15.17 implies that the result BP (α, α′ ) ∈ Γ(M, δ1 (Z D )) of the pairing is a
lift of a density ν = ν(α, α′ ) ∈ Γ(Q, δ1 (T QC )).
As a consequence, we obtain the scalar product
Z
′ ′ ′
⟨ψ, ψ ⟩ = ⟨s ⊗ α, s ⊗ α ⟩ := H(s, t)ν(α, α′ )♯ .
M/D

The completion of

HP = {ψ ∈ Γ(M, L ⊗ S) | ψ polarized and ⟨ψ, ψ⟩ < ∞}

with respect to the induced norm is the representation space HSP = HS (M, L, P ) which
we wanted to construct.
15.4 Representation Space and Quantum Operator 229

For the definition of the quantum operator we need, as in the case of half-density
quantization, the concept of a partial Lie derivative
LX : Kℓ (P ) → Kℓ (P )
for vector fields X preserving P , in particular for ℓ = − 12 :
Definition 15.19. A vector field X ∈ Γ(U, P ) preserving P induces the flow Φt :
Mt → M−t with T Φt (Pa ) = Pa , a ∈ Mt . Let ℓ ∈ Z. Then for α ∈ Γ(M, Kℓ (P )) the
natural partial Lie derivative
d ♯
(LX α)♯ (ξ) := α (T Φt (ξ))
dt t=0

is well-defined.
The definition can be transferred to S = K−1/2 (P ) to obtain LSX = LX : Γ(M, S) →
Γ(M, S), as well, by setting
1
LSX α := LX (α2 )
2α
on {a ∈ M | αa ̸= 0}81 . Similarly, we can define the partial Lie derivative on general
Kℓ (P ) , 2ℓ ∈ Z.
Definition 15.20. The quantum operator q S (F )82 for F ∈ RP on polarized sections
φ = s ⊗ α of L ⊗ S is defined by

i
q S (F )(s ⊗ α) = (q(F )s) ⊗ α − s ⊗ LXF α.
2π

Since every global section φ can be written locally in the form φ = s⊗α, this determines
a unique polarized section q(F )φ of L ⊗ S.

Main Result
The main result of this section is the following
Theorem 15.21. Let (L, ∇, H) be a prequantum line bundle on the symplectic manifold
(M, ω) with reducible complex polarization P , and let S be a square root of K−1 (P ).
The constructed representation HS has as its quantum operators q S : RP → S(HS ) the
linear maps
i
q S (F )(s ⊗ α) = q(F )s ⊗ α − s ⊗ LXF α ,
2π
81
The partial Lie derivative on half-forms can be defined in a more natural manner by using the
transformation property which we only have on the metalinear frame bundle R(P e ) linked with the
square root bundle.
82
A definition based on the local flow (ϱF F
t ) on Γ(M, L ⊗ S) induced by the local flow (Φt ) of XF
(as in the last chapter, see Definition 14.8) can only be given using the metalinear structure linked to
the square root. This will be explained in the next chapter.
230 15. Half-Form Quantization

for polarized sections s ⊗ α ∈ Γ(M, L ⊗ S), where q(F ) is the prequantum operator.
Moreover, half-form quantization is a full geometric quantization in the following
sense:

1. q S (F ) is R-linear and satisfies (D1) and (D2), now for the new representation
space HS = HS (M, L, P ).

2. If XF is complete, F ∈ RP , then q S (F ) is self-adjoint.

Example 15.22 (Harmonic Oscillator). We want to show the effect of half-form quan-
tization in the case of the harmonic oscillator, thereby continuing Example 10.23 and
see that the quantized operator q S (H) will have the correct eigenvalues. Therefore,
one speaks of ”half-form correction”.
The phase space is M = T ∗ Rn ∼
= Cn with its standard symplectic form ω = dq j ∧dpj .
j i
P
We use complex coordinates zj = pj +iq
i
P , j = 1, . . . , zn , so that ω = 2 dz̄j ∧dzj with
connection form and potential θ = 2 j z̄j dzj . The Hamiltonian H of the harmonic
oscillator is the energy
n
1X
H(z) = zk z̄k .
2 k=1
The corresponding Hamiltonian vector field is
X 
∂ ∂
XH = i zj − z̄j .
∂zj ∂ z̄j

The prequantum line bundle (L, ∇, H) is the trivial bundle L = M with the con-
nection ∇ given by θ and the natural Hermitian structure H on L = M × C. The
polarization P is the holomorphic polarization P generated by the vector fields Xzj .
Observe that XH preserves P , i.e. H is directly quantizable. The representation space
HP can be identified with the Bargmann space F.
For the prequantum operator we know already, according to 10.23:

n
1 X ∂
q(H) = zk
2π k=1 ∂zk
on the space of holomorphic functions on Cn .
In order to determine the additional term in the quantum operator
i
q S (H)(s ⊗ α) = q(H) ⊗ α − s ⊗ LXH α ,
2π
we have to evaluate the partial Lie derivative LXH α for α ∈ Γ(M, K−1/2 (P )).
The standard n-form dz̄ := dz̄1 ∧ . . . ∧ dz̄n ∈ K(P ) generates K(P ) and K(P ) is
trivial. Moreover, K−1 (P ) = K ∨ ∼
= M × C has the square root S = K−1/2 (P ) generated
15.5 Square Root: Existence and Uniqueness 231

by a −1/2-form α, α−2 = dz̄ which we denote by α =: dz̄ −1/2 . f Cartan’s formula yields
LX dz̄ = iX ddz̄ + d(iX dz̄) = d(iX dz̄). From
X
iXH dz̄ = −i z̄j (−1)j dz̄1 ∧ . . . ∧ dz̄
cj ∧ . . . ∧ dz̄n
j

we deduce
LXH dz̄ = −indz̄ ,
and
1
LXH dz̄ − /2 = indz̄ .
1

2
Therefore, the additional term is multiplication by
i n 1 n
− i = .
2π 2 2π 2
As a consequence, the quantum operator for H is

n
!
1 X ∂ n
q S (F ) = zk +
2π k=1
∂zk 2

acting on holomorphic functions ϕ ∈ O(Cn ).

The eigenvalues will be determined by the equation
 
1 X ∂ n
zk + ϕ = Eϕ .
2π ∂zk 2
Using the result of Example 10.23 we conclude that the eigenvalues are the
1  n
EM = M+ , M ∈ N,
2π 2
and the corresponding eigenspaces VM are the spaces of M -homogenous polynomials
in n complex variables. The VM can be understood as V1 ⊙M and F as the Fock space
of V1 .
This is in accordance to the standard quantum model for the harmonic oscillator.

15.5 Square Root: Existence and Uniqueness

The objective of this section is to study the topological condition which is needed for
the existence of a square root of a given complex line bundle, and to determine to
which extent a possible square root is unique. The conditions are formulated in the
context of Čech cohomology with values in the group Z2 , cf. Section E.1.
232 15. Half-Form Quantization

Let K be a complex line bundle on the manifold M with transition functions hij
with respect to an open cover U = (Uj ). We require that U has the property that the
Uj , Uij . . . are all contractible.
From the characterization of vector bundles by transition functions we know that a
square root S of K will have transition functions zij with zij2 = hij , when hij are tran-
sition functions for K. As a consequence, we are looking for functions zij ∈ E(Uij , C× )
satisfying
1. zij2 = hij on Uij ,
(65)
2. zij zjk zki = 1 on Uijk .
As a first try to find suitable zij we pick for each pair (i, j) ∈ I 2 a smooth square
root dij ∈ E(Uij , C× ) of hij , i.e. dij 2 = hij . This is possible since Uij is contractible.
Since hij satisfies the cocycle condition we have d2ij d2jk d2ki = 1. But the condition 2. for
the choice zij = dij , here dij djk dki = 1, will not be satisfied, in general. We define
aijk := dij djk dki on Uijk .

The collection a := (aijk ) is a Čech cocycle in C 2 (U, Z2 ), since aijk ∈ Z2 = {1, −1} 83
because of a2ijk = 1. a induces a cohomology class [a] ∈ Ȟ 2 (U, Z2 ). This cohomology
class is independent of the choice of the square roots dij , it only depends on the tran-
sitions functions hij . In fact, with another choice of d′ij ∈ E(Uij , C× ) with (d′ij )2 = hij
we set cij := d′ij d−1 ′
ij ∈ Z2 and obtain for the new cocycle a := (aijk )
′

a′ijk := d′ij d′jk d′ki = dij djk dki cij cjk cki = aijk cij cjk cki .
Hence
a′ijk
= cij cjk cki ,
aijk
which means that a′ a−1 is the coboundary of c = (cij ) ∈ Č 1 (U, Z2 ). The two cocycles
a, a′ define the same cohomology class [a] = [a′ ] ∈ Ȟ 2 (U, Z2 ). Moreover, different
transition functions for K lead to the same class, as well.
This cohomology class depending on the line bundle K will be denoted by w(K) =
[a] = [(hij )] and it will be called the obstruction class. It is the obstruction for the
existence of a square root, as we will see in the following.
Of course, a = (aijk ) can be regarded as a cocycle in Č 2 (U, C× ) with values in
the group C× , and by definition aijk = dij djk dki it is a coboundary there. In order
that a is a coboundary in Č 2 (U, Z2 ), and hence trivial, it would require that a cocycle
b = (bij ) ∈ Č 1 (U, Z2 ) exists which satisfies84
aijk = bij bjk bki .
We have prepared the proof of the following proposition:
83
We write the abelian group Z2 as the multiplicative group {1, −1} instead of the additive group
{[0], [1]}.
84
When the condition is written additively, it is aijk = bij + bjk + bki = bjk − bik + bij = δ(b)ijk .
15.5 Square Root: Existence and Uniqueness 233

Proposition 15.23. There exists a square root line bundle for the complex line bundle
K over M if and only if the obstruction class w(K) = [a][(hij )] is trivial in Ȟ 2 (U, Z2 ),
i.e. if it is the class [1]. One also says, w(K) vanishes, when one emphasizes the
additive notation Z2 = {[0], [1]}.

Proof. We have stated earlier that the existence of a square root implies the existence of
zij ∈ E(Uij , C× ) satisfying zij2 = hij with zij zjk zki = 1, see (65). Since aijk = zij zjk zki ,
the obstruction class [a] = [(aijk )] is trivial.
Conversely, let the class [a] given by aijk := dij djk dki for a choice of dij and assume
taht it is trivial. Then [a] is a boundary δ([b]), i.e. there is a cocycle b = (bij ) ∈
Č 1 (U, Z2 ) with aijk = δ([b])ijk = bij bjk bki . As a consequence, the functions
dij
zij := ∈ E(Uij , C× )
bij
satisfy 1. and 2. of (65), and thus define a square root S of K with zij as its transition
functions.

Concerning the uniqueness of the square root we prove:

Proposition 15.24. Given a complex line bundle K the isomorphism classes of line
bundles S with S 2 ∼
= K are in one-to-one correspondence to the elements of the Čech
cohomology group Ȟ 1 (U, Z2 ).

Proof. Assume that w(K) = [(hij )] is trivial, so that there is a line bundle S with
S2 ∼
= K with transition functions zij satisfying zij2 = hij .
For another line bundle S ′ isomorphic to S its transition functions satisfy
hi
zij′ = zij
hj

with hj ∈ E(Uj , C× ) according to (I) in Section 3.2. Because of (zij′ )2 = hij = zij2 it
follows that hi 2 = hj 2 = 1. Therefore,
zij′ hi
is the coboundary of ,
zij hj

and (zij ), (zij′ ) determine the same cohomology class in Ȟ 1 (U, Z2 ).

Any cocycle c = (cij ) ∈ Č 1 (U, Z2 ) determines a zij′ := cij zij which defines another
square root S ′ = S([c]) of K. It is not isomorphic to S when its cohomology class
[c] ∈ Ȟ 1 (U, Z2 ) is not trivial. In this way, we obtain a bijection

S : Ȟ 1 (U, Z2 ) −→ {isomorphism classes of square roots of K} , [c] 7→ S([c]) .

234 15. Half-Form Quantization

Corollary 15.25. For the program of half-form quantization it follows that K−1 (P )
has a square root if and only if the corresponding obstruction class w(K−1 (P )) of the
bundle K−1 (P ) vanishes. This is equivalent to the vanishing of the obstruction w(KP )
of the canonical bundle KP = K1 (P )) of P .
Moreover, whenever there exists a square root of K−1 (P ) the Čech cohomology group
1
Ȟ (U, Z2 ) parametrizes the inequivalent square roots of K−1 (P ).

Summary:
 235

16 Metalinear Structure

16.1 Metalinear Frame Bundle

A naive approach to define half-forms on the frame bundle R(P ) of P (where P is a

polarization or more generally a vector bundle) would be to require, that such a half-
form corresponds to a function u on R(P ) with the following transformation property
1
u(bg) = (det g) /2 u(b) (66)
1
for frames b ∈ R(P ) and g ∈ GL(n, C). However, the square root (det g) /2 is not
well-defined, in general. We have discussed this in Observation 15.6 in the preceding
chapter.
To remove the ambiguity in the square root the general linear group GL(n, C) will
be replaced be its connected double covering, the Metalinear Group ML(n, C), and
subsequently the frame bundle R(P ) by a metalinear frame bundle R(Pe ).
The (complex) metalinear group ML(n, C) is the connected central extension of the
Lie group GL(n, C) by Z2 :
ρ
1 → Z2 → ML(n, C) → GL(n, C) → 1 ,
in particular, ρ : ML(n, C) → GL(n, C) is a double covering.
The metalinear group can be defined as the quotient (C × SL(n, C))/2Z as we show
in the following:
We start with the simply connected Lie group C × SL(n, C) with the group law
((u, s), (u′ , s′ )) 7→ (u + u′ , ss′ ), and consider the action
Z × (C × SL(n, C)) → C × SL(n, C)
of Z on C × SL(n, C) given by
 
2πik − 2πik
(k, (u, s)) 7→ u + ,e n s .
n
Lemma 16.1. The homomorphism
p : C × SL(n, C) → GL(n, C) , (u, s) 7→ eu s ,
induces an isomorphism
C × SL(n, C) ∼ = GL(n, C) .
In other words, C ×Z SL(n, C) ∼
= GL(n, C). Moreover, the injection
 
2πik − 2πik
j : Z → C × SL(n, C) , k 7→ , e n In , k ∈ Z ,
n
satisfies Im j = Ker p, i.e. we obtain the following exact sequence
j p
1 −→ Z −→ C × SL(n, C) −→ GL(n, C) −→ 1 .
236 16. Metalinear Structure

Proof. It is easy to see that p is a homomorphism and surjective. The induced homo-
morphism
C × SL(n, C) ∼
= GL(n, C)
is injective, since p is invariant under the action of Z:
 
2πik − 2πik 2πik 2πik
p u+ , e n s = eu+ n e− n s = eu s = p(u, s) .
n
In particular,   
2πik − 2πik
Ker p = , e n In | k ∈ Z = Im j .
n

We conclude that p : C × SL(n, C) → GL(n, C) is a universal covering of

GL(n, C) since SL(n, C) and hence C × SL(n, C) are simply connected. It follows, that
π1 (GL(n, C)) ∼
= Z. This covering ”contains” a 2-fold covering, which is the metalinear
group:
Definition 16.2. The quotient group (C × SL(n, C))/2Z ∼ = C ×2Z SL(n, C) with its Lie
structure is called the (complex) metalinear group and will be denoted by ML(n, C).

Two elements (u, s), (u′ , s′ ) ∈ C × SL(n, C) are equivalent with respect to 2Z if and
only if there is k ∈ Z such that u = u′ + 4πik
n
and s = exp(− 4πik
n
)s′ . As a consequence,
the coset [u, s] of (u, s) is
  
4πik − 4πik
π(u, s) := [u, s] = u+ ,e n s | k ∈ Z , (u, s) ∈ C × SL(n, C) .
n

The definition of ML(n, C) comprises the quotient homomorphism

ρ : ML(n, C) → GL(n, C) , [u, s] 7→ eu s ,

which is surjective such that the following diagram is commutative

C × SL(n, C)
p π
 ρ
'
ML(n, C) / GL(n, C)

The kernel of ρ consists of the elements [u, s] satisfying eu s = In 85 which implies s =

e−u In , hence det s = e−nu = 1, where u = 2πih
n
for suitable h ∈ Z. As a consequence,
     
2πik 4πih − 2πik − 4πih 2πi − 2πi ∼
Ker ρ = + , e n e n In | k, h ∈ Z = [0, In ], , e n In = Z2 .
n n n
85
Here In denotes, as in other occasions, the unit n × n- matrix.
16.1 Metalinear Frame Bundle 237

We obtain the above mentioned exact sequence

ι ρ
1 → Z2 −→ ML(n, C) → GL(n, C) → 1 ,

where (we write again Z2 multiplicatively: Z2 = {1, −1})

 
2πi − 2πi
ι(1) := [0, In ], ι(−1) := , e n In .
n

Observe, that this exact sequence characterizes ML(n, C) as a connected double

covering of GL(n, C).
We also want to note that the fibre of ρ over an element g ∈ GL(n, C) is
  
−1
 −u  2πi −(u+ 2πi )
ρ (g) = u, e g , u + ,e n g ,
n

where u ∈ C is a complex number with det g = enu .

n
The Lie group homomorphism χ : ML(n, C) → C× defined by χ([u, g]) := e 2 u
satisfies χ2 = det ◦ ρ, i.e. the following diagram is commutative

ML(n, C)
χ2

%
ρ ×
9C
det

GL(n, C)

In this sense, χ can be regarded as to be the square root of the determinant.

In order to use this square root not only for a single point of M but globally over the
manifold M we consider an equivariant double (i.e. 2-to-1) covering ρ̃ : R(P e ) → R(P )
of the frame bundle R(P ), where P is a polarization. Over a point a ∈ M where the
ea (P ) → Ra (P ) should represent
fibre Ra (P ) is essentially GL(n, C) the covering ρ̃a : R
the covering ρ : ML(n, C) → GL(n, C) just described. More precisely:

Definition 16.3. A Metalinear Frame Bundle for a polarization P ⊂ T M C is a

e ) → M with structure group ML(n, C) together with a
principal fibre bundle π̃ : R(P
e ) → R(P ), such that
2-to-1 covering ρ̃ : R(P
1. ρ̃ is compatible with the projections, i.e. the diagram
ρ̃
R(P
e ) / R(P )

π̃ π
" |
M
238 16. Metalinear Structure

is commutative: π̃ = ρ̃ ◦ π, and
2. ρ̃ is equivariant, i.e. the following diagram is commutative

e ) × ML(n, C)
R(P / R(P
e )
(ρ̃,ρ) ρ̃
 
R(P ) × GL(n, C) / R(P ) ,

where the horizontal arrows are the right actions of the respective groups: In other
words, ρ̃(b̃g̃) = ρ̃(b̃)ρ(g̃)) for (b̃, g̃) ∈ R(P
e ) × ML(n, C).

Two such metalinear frame bundles R(P e ) and ρ̃′ : Re′ (P ) → R(P ) for P are equiva-
e )→R
lent, if there exists an equivariant diffeomorphism ϕ : R(P e′ (P ), i.e. a diffeomor-
′
phism satisfying ϕ(b̃g̃) = ϕ(b̃)g̃ and ϕ ◦ π̃ = π̃. Finally, a Metalinear Structure
on P is an equivalence class of metalinear frame bundles for P .

Given a metalinear frame bundle R(P e ) and a frame b ∈ Ra (P ) each of the two
−1
objects in ρ̃ (b) ⊂ R(P
e ) is called metaframe. A metaframe is sometimes denoted
by b̃ although this notation is slightly ambiguous.
In general, a metalinear frame bundle does not exist. It exists whenever a certain
obstruction class in Ȟ 2 (M, Z2 ), which is induced by P resp. R(P ), is trivial. In case
of existence the metalinear structures on P are parametrized by the Čech cohomology
group Ȟ 1 (M, Z2 ). This is similar to the existence and uniqueness of square roots of
the canonical bundle K(P ), see Section 15.5 in the preceding chapter, and will be
explained in Section 16.3 below in detail.
e ) → R(P ) a metalinear frame bundle for P and let g̃ij be transition
Let ρ̃ : R(P
functions for this metalinear frame bundle with respect to some open cover U = (Uj ),
again with the property that Ui , Uij , Uijk , ... are contractible. Then there exist functions
uij : Uij → C and sij : Uij → SL(n, C) such that g̃ij = [uij , sij ]. It is easy to show that
gij := ρ ◦ g̃ij = euij sij are transition functions for the frame bundle R(P ) (see the proof
of Proposition 16.8 below). Moreover, zij := χ(g̃ij ) = exp n2 uij is a cocycle. Therefore
(zij ) defines a complex line bundle S → M , see Proposition 3.9.
Now the square of S, the bundle S 2 := S ⊗ S, has transition functions zij2 satisfying
zij2
= χ2 (g̃ij ) = det gij by definition of the character χ. We know by Observation 15.4
that det gij are the transition functions of K−1 (P ) as well. As a result, S ⊗S ∼
= K−1 (P )
and the dual S ∨ is a square root of the canonical bundle K(P ) = K1 (P ).
We conclude

Proposition 16.4. For a polarization P the equivalence classes of square roots of the
bundle K−1 (P ) are in one-to-one correspondence to the metalinear structures on P .
More explicitly, we can describe this bijection with the aid of transition functions: Let
[(g̃ij )] denote the equivalence class of the metalinear frame bundle R(P
e ) determined by
16.1 Metalinear Frame Bundle 239

transition functions g̃ij : Uij → ML(n, C), i.e. [(g̃ij )] is a metalinear structure on P .
Then the bijection is given by

[(g̃ij )] 7→ [(χ(g̃ij )] .

Proof. We have just seen that a metalinear structure on P with transition functions
g̃ij induces a square root of K−1 (P ) determined by transition functions χ(g̃ij ).
Conversely, any square root S of K−1 (P ) is given by transition functions zij satis-
fying zij2 = det gij . We can find uij with
n
zij = exp uij .
2
Since (zij ) is a cocycle, exp n2 (uij + ujk + uki ) = 1, i.e. n2 (uij + ujk + uki ) = 2πim,
thus uij + ujk + uki = 4πim n
, where m is a suitable integer m ∈ Z. We define g̃ij :=
[uij , exp(−uij gij )] : Uij → ML(n, C). We show that (g̃ij ) is a cocycle:
 
−uij −ujk −uki 4πim − 4πim
g̃ij g̃jk g̃ki = [uij + ujk + uki , e e e gij gjk gkj ] = , e n In = [0, In ] = 1
n
by the definition of the metalinear group as (C × SL(n, C))/2Z. This cocycle (g̃ij )
defines a metalinear frame bundle R(P e ) determined by the square root of K−1 (P )
given by the transition function zij . Moreover, zij = χ(˜gij ).
As a result, the assignment [(g̃ij )] 7→ [(χ(g̃ij )] is bijective.

Note, that the set of equivalence classes of square roots of K−1 is also in bijection
with Ȟ 1 (U, Z2 ) ∼
= Ȟ 1 (M, Z2 ), see Proposition 15.24, and thus the set of equivalence
clases of metalinear structure on P are in bijection with Ȟ 1 (M, Z2 ).
A slightly different look at the line bundle S induced by the metalinear frame bundle
is the following: With respect to the charecter χ : ML(n, C) → C× , (z, g) 7→ z, we
obtain the associated complex line bundle to the principal fibre bundle R(P e )

S ′ := R̃(P ) ×ML(n,C) C = R̃(P ) ×χ C .

The transition functions for S ′ turn out to be χ(g̃ij ) = zij according to Proposition
D.9. It follows that S and S ′ are isomorphic.
In the same way one can prove:
Lemma 16.5. In case a square bundle of K(P ) exists, for each half integer r the
bundle Kr (P ) of r-forms is isomorphic to the associated complex line bundle

R̃(P ) ×ML(n,C) C = R̃(P ) ×χ−2r C .

With this result we can formulate the correct form of the transformation property
adjusting the naive ansatz (66):
240 16. Metalinear Structure

e ) → R(P ) be a metalinear frame bundle, where P is a

Corollary 16.6. Let ρ̃ : R(P
polarization of the symplectic manifold (M, ω). For each half integer r the sections α ∈
Γ(M, Kr (P )) are in one-to-one correspondence with the functions α̃♯ = u : R(P
e )→C
satisfying
u(b̃g̃) = χ(g̃)2r u(b̃) ,
1
where b̃ ∈ R(P
e ) and g̃ ∈ M L(n, C). In particular, with r =
2

u(b̃g̃) = χ(g̃)u(b̃) ,

Moreover, for r ∈ Z

u(b̃g̃) = (det ρ(g̃))r u(b̃) = (det g)r u(b̃) ,

when ρ(g̃) = g.

Proof. The transformation rule follows immediately from the construction of the asso-
ciated bundle R̃(P ) ×χ−2r C, cf. Corollary D.8.

Note, that in the case of an integer r ∈ Z a section α ∈ Γ(M, Kr (P )) induces

α♯ : R(P ) → C with
α♯ (bg) = (det g)r α♯ (b)
and α̃♯ : R(P
e ) → C with

α̃♯ (b̃g̃) = (det g)r α̃♯ (b̃) , g = ρ̃(g̃) .

The two functions are simply related by α̃♯ = α♯ ◦ ρ̃.

16.2 Half-Form Quantization Based on the Metalinear Group

In this section a fixed metalinear structure is given by a metalinear frame bundle R̃(P )
where P is a reducible complex polarization on the quantizable symplectic manifold
(M, ω) with a prequantum bundle (L, H, ∇). The metalinear frame bundle R̃(P ) in-
duces a square root S = K−1/2 (P ) of the dual K−1 (P ) of the canonical bundle K(P ),
as we have seen in the preceding section.
With the aid of this square root line bundle S = K−1/2 (P ) the programme of half-
form quantization can be carried through in the same way as is done in Section 15.4.
We mention four occasions where, in comparison to the Section 15.4 and before,
the transformation property of half-forms can be used directly in order to make the
arguments more transparent. Recall, that the transformation property in question is
(cf. Corollary 16.6):
α̃♯ (b̃g̃) = χ(g̃)− /2 α̃♯ (b̃)
1

e ) × ML(n, C) and α̃♯ = α♯ ◦ ρ̃.

for sections α ∈ Γ(M, K−1/2 (P )), where (b̃, g̃) ∈ R(P
16.2 Half-Form Quantization Based on the Metalinear Group 241

1. Proof of the statement of Lemma 15.9: ”Any two sections α, β ∈ Γ(M, K−1/2 (P ))
determine a −1-density ᾱβ ∈ Γ(M, δ−1 (P )).”
Here, ᾱβ =: µ is given by
µ♯ (b) := α̃♯ (b̃)β̃ ♯ (b̃)

for ρ̃(b̃) = b. To prove this statement, we first observe that χ(g̃)χ(g̃) = | det g|
when ρ(g̃) = g. The new transformation property

α̃♯ (b̃g̃) = χ(g̃)−1 α̃♯ (b̃)

(see Corollary 16.6) yields:

µ♯ (bg) = α̃♯ (b̃g̃)β̃ ♯ (b̃g̃) = χ(g̃)−1 α̃♯ (b̃)χ(g̃)−1 β̃ ♯ (b̃) = (χ(g̃)χ(g̃))−1 µ♯ (b) = | det g|−1 µ♯ (b) .

Hence, µ = ᾱβ ∈ Γ(M, δ−1 (P )).

2. Definition of partial connection ∇X σ for σ ∈ Γ(U, Kℓ ), 2ℓ ∈ Z, and X ∈ Γ(U, P ):

Locally, there exists a metaframe field ξ˜ : U → R(P
e ) such that the corresponding
frame field ξ = ρ̃ ◦ ξ˜ is Hamiltonian. We obtain as in the case of R(P ) a section
σ ∈ Γ(U, Kℓ ) such that for the induced σ̃ ♯ : R(P
ξ̃
e )|U → C the following holds
ξ̃
˜
σ̃ξ̃♯ (ξ) ˜ and the definition
= 1. A general section σ ∈ Γ(U, Kℓ ) satisfies σ = σ̃ ♯ (ξ)σξ̃
is
˜
∇X σ := LX (σ̃ ♯ (ξ))σ ξ̃ ) .

This direct definition avoids the twa step definition using the square σ 2 , cf. Corol-
lary 15.15.

3. Definition of partial Lie derivative: Let X ∈ Γ(U, P ) preserve P and denote the
flow of X by Φt : Mt → M−t with T Ψt (Pa ) = Pa , a ∈ Mt . Let ℓ ∈ { 21 , − 21 }.
Then for α ∈ Γ(M, Kℓ (P )) the partial Lie derivative LX α is given by

♯ ˜ d ♯ g ˜
(L
] X α) (ξ) := α (T Φt (ξ))
dt t=0

is well-defined since T Φt can be transferred to R(P e )|U →

e ) as a map TeΦt : R(P
e )|U such that T Φt ◦ ρ̃ = ρ̃ ◦ Tg
R(P Φt .
Thus we avoid the 2 step definition 15.19.

4. Definition of the quantum operator for classical variables F using the local flow
of F in dynamic form, see Proposition 7.13.
242 16. Metalinear Structure

16.3 Metalinear Structure: Existence and Uniqueness

This section discusses under which topological conditions on the manifold M and the
polarization P there exists a metalinear frame bundle for P and to which extent it will
be unique up to isomorphism. These questions will be investigated in the case of a
general principal fibre bundle π : B → M with structure group GL(n, C) instead of
R(P ). A comparison with spin structures on a manifold will be described in the next
chapter.
Definition 16.7. For a principal fibre bundle π : B → M with structure group
GL(n, C) a metalinear bundle over B is a principal fibre bundle π̃ : B e → M with
e → B, such that
structure group ML(n, C) together with a 2-to-1 covering ρ̃ : B
1. ρ̃ is compatible with the projections, i.e. the diagram
ρ̃
B
e / B
π
π̃  
M
is commutative: π̃ = π ◦ ρ̃, and
2. ρ̃ is equivariant, i.e. the following diagram is commutative
e × ML(n, C)
B / B
e (67)
(ρ̃,ρ) ρ̃
 
B × GL(n, C) / B,
where the horizontal arrows are the right actions of the respective groups: We have
ρ̃(b̃g̃) = ρ̃(b̃)ρ(g̃) for (b̃, g̃) ∈ R(P
e ) × ML(n, C).
Two such metalinear bundles ρ̃ : B e → B and ρ̃′ : Be ′ → B over B are called to
be equivalent if there exists an equivariant map ϕ : B e → B e ′ , i.e. ρ̃ = ρ̃′ ◦ ϕ and
ϕ(b̃g̃) = ϕ(b̃)g̃ for (b̃, g̃) ∈ B e × ML(n, C). An equivalence class of metalinear bundles
over B is called a Metalinear Structure.

Evidently, when P is a polarization on a symplectic manifold (M, ω), then a met-

alinear structure on B = R(P ) is represented by a metalinear frame bundle for P as
defined in Section 16.1.
Let ρ̃ : B̃ → B be such a metalinear bundle over B and let g̃ij : Uij → ML(n, C) be
transition functions for B̃ with respect to an open cover U = (Uj ), where the Uj , Uij . . .
are contractible. Then g̃ij has the form g̃ij = [uij , sij ] with suitable uij : Uij → C, sij :
n
Uij → SL(n, C) as before. Moreover, zij := χ(g̃ij ) = e 2 uij : Uij → C× , gij := ρ(g̃ij ) =
euij sij : Uij → GL(n, C) satisfy
1. zij2 = det gij on Uij ,
(68)
2. zij zjk zki = 1 on Uijk .
16.3 Metalinear Structure: Existence and Uniqueness 243

Proposition 16.8. The gij are transition functions for the bundle B. Conversely,
when gij are transition functions for B and uij : Uij → C can be found such that with
zij := exp n2 uij the above conditions 1. and 2. are satisfied, then g̃ij := [uij , e−uij gij ] :
Uij → ML(n, C) are transition functions of a metalinear bundle π̃ : B e → B over B.

Proof. To show the lemma is easy, only the various conditions have to be checked,
and this is not very interesting. Nevertheless, in order to get acquainted with the new
structure, which is important also in the next chapter, we present a detailed proof.
So let ρ̃ : B̃ → B a metalinear structure over B. The compatibility conditions on
ρ̃, ρ and the group actions (see the commutative diagram (67) in the definition) imply
that gij are transition functions for B as we see by investigating the local situation:
Let ψ̃j : BeU → Uj × ML(n, C) be the local trivializations with respect to the cover
j
(Uj ) which define the transition functions g̃ij by

ψ̃i ◦ ψ̃j−1 (a, h̃) = (a, g̃ij (a).h̃) , (a, h̃) ∈ Uij × ML(n, C) .

The double covering ρ̃ induces local trivializations ψj : BUj → Uj × GL(n, C) for the
principal bundle B: Each b ∈ B with π(b) ∈ Uj has two inverse images b̃± ∈ ρ̃−1 (b),
and ψj (b) := (id × ρ)(ψ̃j (b̃± )) is well-defined: ψ̃j (b̃± ) = (a, h̃± ) with two h± ∈ SL(n, C)
with a common image ρ(h± ) = h, hence (id × ρ)(ψ̃j (b̃+ )) = (a, h) = (id × ρ)(ψ̃j (b̃− )).
Since ρ̃ is a local diffeomorphism, ψj is smooth, and, moreover, a diffeomorphism which
respects the action of GL(n, C). Hence, the ψj are local trivializations of the principal
fibre bundle B satisfying ψj ◦ ρ̃ = (id×ρ)◦ ψ̃j , i.e. the following diagram is commutative

ψ̃j
B
eU / Uj × ML(n, C)
j

ρ̃ id×ρ
 
BUj / Uj × GL(n, C) .
ψj

Finally, the diagram helpsto varify

ψi ◦ ψj−1 (a, h) = ψi ◦ ρ̃ ◦ ψ̃j−1 (a, h̃± )

= (id × ρ) ◦ ψ̃i ◦ ψj−1 (a, h̃± )
= (id × ρ)(a, g̃ij (a).h̃± )
= (a, ρ(g̃ij (a).h̃± ))
= (a, (ρ(g̃ij (a)).(ρ(h̃± ))))
= (a, gij (a).h) ,

As a result, gij = ρ(g̃ij ) are transition functions for B.

Conversely, 1. and 2. imply that g̃ij : Uij → ML(n, C), given by g̃ij := [uij , e−uij gij ],
are well-defined and satisfy the cocycle condition (C) which we have shown already
244 16. Metalinear Structure

in the special case of B = R(P ) in Proposition 16.4: In fact, (zij ) = (exp n2 uij ) is a
cocycle, i.e. exp n2 (uij + ujk + uki ) = 1, such that n2 (uij + ujk + uki ) = 2πim for a
suitable m ∈ Z. Hence,
 
−uij −ujk −uki 4πim − 4πim
g̃ij g̃jk g̃ki = [uij + ujk + uki , e e e gij gjk gkj ] = ,e n In = [0, In ] = 1 .
n

e → M having g̃ij as its transition

The corresponding principal fibre bundle π̃ : B
functions can be constructed as in the case of line bundles, see Proposition 3.9. The
fact gij = ρ ◦ g̃ij gives immidiately the 2-1 covering map ρ̃ : Be → B with all its
compatibilities establishing that in the way we obtain a metalinear structure over B.

We now come to the essential part of this section where we determine when a
metalinear structure exists and to which extent it is unique. From the preceding
proposition we know that a metalinear structure on a principal GL(n, C)-bundle B
with transition functions gij exists if and only if one can find uij : Uij → C such that
n
for zij := e 2 ui j and gij the conditions 1. and 2. are satisfied.
In the following, we require again that the open cover U = (Uj ) has the property
that the Uj , Uij . . . are all contractible.
We pick for each i, j ∈ I a smooth square root dij ∈ E(Uij , C× ) of det gij , i.e. d2ij =
det gij . Since gij satisfies the cocycle condition this holds for det gij as well and thus
d2ij d2jk d2ki = 1. In order to discuss condition 2., here dij djk dki = 1, we define

aijk := dij djk dki on Uijk .

The collection a := (aijk ) is a cocycle with aijk ∈ Z2 = {1, −1} since a2ijk = 1. It
induces a cohomology class [a] ∈ Ȟ 2 (U, Z2 ) ∼= Ȟ 2 (M, Z2 ). This cohomology class is
independent of the choice of the square roots dij , it only depends on the transitions
functions gij . For different transition functions with respect to possibly other covers
U′ we obtain the same class depending only on B. This cohomology class w(B) := [a]
will be called the obstruction class.
a = (aijk ) can be regarded as a cocycle in Č 2 (U, C), and by definition it is a
coboundary there. In order that a is a coboundary in Č 2 (U, Z2 ), there should exist a
cocycle b = (bij ) ∈ Č 1 (U, Z2 ) which satisfies

aijk = bij bjk bki .

We have prepared the proof of the following proposition:

Proposition 16.9. There exists a metalinear structure for B if and only if the obstruc-
tion class w(B) = [a] is trivial, i.e. if it is the class [1]. One also says, it vanishes,
when one emphasizes the notation Z2 = {[0], [1]}.
16.3 Metalinear Structure: Existence and Uniqueness 245

Proof. We just have deduced that the existence of a metalinear structure for B implies
the existence of uij ∈ E(Uij , C such that zij := exp n2 uij satisfy zij2 = det gij and
zij zjk zki = 1, see (68). Since a = zij zjk zki = 1, the obstruction class [a] is trivial.
Conversely, let a be given by aijk := dij djk dki for a choice of dij with d2ij = det gij .
When [a] is trivial in Ȟ 2 (U, Z2 ), then there is b = (bij ) ∈ Č 1 (U, Z2 ) with aijk = bij bjk bki .
As a consequence,
dij
zij :=
bij
satisfy 1. and 2. of (68).

Corollary 16.10. For the program of geometric quantization it follows for a complex
polarization P that a metalinear frame bundle R(P e ) → R(P ) exists if and only if the
obstruction class w(R(P )) is trivial.
If w(R(P )) is trivial and R(P
e ) is a metalinear frame bundle we have a natural
1
bijection between Ȟ (M, Z2 ) and the set of metalinear structures in the same way as
for the case of square roots, see the end of the preceding chapter.

Examples 16.11. 1. Momentum Space (T ∗ Q, ω).

2. P2k+1 (C), but not P2k (C), k ∈ Z.
3. Graßmannian Gr(2, 4).

Summary:
246 17. Metaplectic Structure

17 Metaplectic Structure

In case of a quantizable symplectic manifold (M, ω) with a prequantum line bundle

and with two different complex polarizations admitting a square root of the canonical
bundle we want to compare the corresponding representations induced by the half-form
quantization (see Section 15). In order to construct a pairing between the two repre-
sentation spaces the square roots have to fit together. In other words, the metalinear
structures related to the square roots (see 16) have to be compatible. In which way?
It turns out that the different metalinear structures on the polarizations should be
induced by a joint metaplectic structure given on the symplectic manifold (M, ω). The
metaplectic structure is similar to a spin structure on a Riemannian manifold.
As the main result of this section we show, that a metaplectic structure on (M, ω)
determines on each positive complex polarization P of (M, ω) a unique metalinear
structure thereby yielding a square root óf the canonical bundle KP for each P .
This result is applied to constrcut a pairing between representations determined by
differentpositive presentations.
Moreover, this chapter ends with a presentation of the metaplectic representation
and its relation to the CCR representation.

17.1 Metaplectic Frame Bundle

Let (M, ω) be a symplectic manifold. A symplectic frame at a ∈ M is an ordered

basis
(u; v) = (u1 , . . . , un , v1 , . . . , vn ) of Ta M
such that ω(ui , vj ) = δij and ω(ui , uj ) = ω(vk , vl ) = 0, i, j, k, l ≤ n. The collection
Ra (M, ω) of symplectic frames at a ∈ M is in one-to-one correspondence with the
symplectic group Sp(n, R), the group of linear canonical transformations R2n → R2n
with respect to the standard symplectic form.
The symplectic group Sp(n, R) can be defined as the concrete matrix group of
2n × 2n real block matrices  
A B
S= ,
C D
where A, B, C, D are real n × n-matrices,satisfying
 
T 0 1
S JS = J, with J = Jn := , i.e.
−1 0

AT D − C T B = 1, AT C = C T A, DT B = B T D. (69)
There is a natural right action of Sp(n, R) on Ra (M, ω):
 
A B
(u; v) × = (uA + vC; uB + vD),
C D
17.1 Metaplectic Frame Bundle 247

0 0
which provides a bijection Sp(n, R) → Ra (M, ω) by fixing a   ; v ) ∈ Ra (M, ω):
frame (u
A B
For each (u; v) ∈ Ra (M, ω) there exists exactly one S = ∈ Sp(n, R) with
C D
(u0 ; v 0 )S = (u; v).

Lemma 17.1. The subgroup Un of Sp(n) consisting of all block matrices S with A =
D and C = −B is isomorphic to U(n). The S ∈ Sp(n) which are in Un can be
characterized by JS = SJ or S −1 = S T . Un is a maximal compact subgroup of Sp(n, R).

Proof. In fact,    
A B A B
7 A + iB =: γ
→ ,
B A B A
AT A + B T B = 1, AT B = B T A (cf. (69)), is an isomorphism γ : Un → U (n): For
T
U := A + iB we see U U = 1 by
T
(A + iB) (A + iB) = AT A + B T B + i(AT B − B T A) = 1 + i0 = 1 .

Therefore, γ is a well-defined homomorphism and it is easy to see that it is bijective.

Given  
A B
S= ∈ Sp(n),
C D
the condition JS = SJ is
         
0 1 A B C D A B 0 1 −B A
= = =
−1 0 C D −A −B C D −1 0 −D C

which is equivalent to A = D and B = −C.

Note, that U(n) occurs also as a maximal compact subgroup of the general linear
group GL(n, C).
The symplectic group is homeomorphic to the product of the unitary group U (n)
and a contractible space: Let S+ (n) denote the symmetric and positive symplectic
matrices, then
S+ (n) × Un → Sp(n, R) , (S, U ) 7→ S ◦ U ,
is a homeomorphism.
Therefore, since S+ (n) is contractible, the fundamental group of the symplectic
group is π1 (Sp(n, R)) = π1 (U(n)) = Z 86 . As a consequence, there exists a double
covering of Sp(n, R), denoted by Mp(n, R), which is again a Lie group, and we have
the central extension
ρ
1 −→ Z2 −→ Mp(n, R) −→ Sp(n, R) −→ 1 .
86
Here, we use the result π1 (U(n)) = Z.
248 17. Metaplectic Structure

Mp(n, R) is called the metaplectic group.

The collection of all the symplectic frames over a symplectic manifold (M, ω) defines
the symplectic frame bundle on (M, ω)
π
[
R(M, ω) := Ra (M, ω) → M , Ra (M, ω) ∋ (u; v) 7→ a ∈ M .
a∈M

R(M, ω) is a right principal bundle over M with structure group Sp(n, R). This
fibre bundle comes automatically with the structure of a symplectic manifold and
is constructed in an analogous way as the frame bundle R(E) of a complex vector
bundle E → M (see Construction D.4). In particular, the symplectic frame bundle
R(M, ω) is a subbundle of the (tangent) frame bundle R(M ) = R(T M ) of all bases of
Ta M , a ∈ M , with structure group GL(2n, R).
Definition 17.2. A Metaplectic Frame Bundle over a symplectic manifold
(M, ω) is a right principal bundle π̃ : R(M,
e ω) → M with structure group Mp(n, R)
together with a double covering
ρ̃ : R(M,
e ω) → R(M, ω) ,
such that
1. ρ̃ is compatible with the projections, i.e. the diagram
ρ̃
R(M,
e ω) / R(M, ω)

π̃ π
$ z
M
is commutative: π̃ = ρ̃ ◦ π, and
2. ρ̃ is equivariant, i.e. the following diagram is commutative

R(M,
e ω) × Mp(n, R) / R(M,
e ω)
(ρ̃,ρ) ρ̃
 
R(M, ω) × Sp(n, R) / R(M, ω),

where the horizontal arrows are the right actions of the respective groups: In other
words, the condition 2. is
ρ̃(b̃g̃) = ρ̃(b̃)ρ(g̃))
for (b̃, g̃) ∈ R(M,
e ω) × Mp(n, R).
Two such metaplectic frame bundles R(M, e ω) and ρ̃′ : R e′ (M, ω) → R(M, ω) on a
symplectic manifold (M, ω) are equivalent, if there exists an equivariant diffeomorphism
ϕ̃ : R(M,
e ω) → R e′ (M, ω), i.e. a diffeomorphism satisfying ρ̃′ ◦ ϕ̃ = ρ̃ and ϕ̃(b̃g̃) = ϕ̃(b̃)g̃
and ϕ ◦ π ′ = π̃. Finally, a Metaplectic Structure on P is an equivalence class of
metaplectic frame bundles for (M, ω).
17.2 General Metastructures 249

Remark 17.3. This definition is completely analogous to the definition of a metalinear

structure (see Definitions 17.2 or 16.7), and can be generalized to further geometric
situations as we explain in the next section. Moreover, the similarity extends to the
questions of existence and uniqueness.

A symplectic manifold (M, ω) admits a metaplectic structure if and only if the

second Stiefel-Whitney class w2 (M ) ∈ H 2 (M, Z2 ) of M is trivial. Here, the second
Stiefel-Whitney class w2 (M ) is the modulo 2 reduction of the first Chern class c1 (M ) ∈
H 2 (M, Z). Hence, (M, ω) admits metaplectic structures if and only if c1 (M ) is even, i.e.
w2 (M ) is zero. In case of w2 (M ) = 0 the equivalence clases of metaplectic structures
are in 1-to-1 correspondece to H 1 (M, Z2 ).

17.2 General Metastructures

The notion of a metaplectic structure of a symplectic manifold is not only analogous

to the notion of a metalinear structure of a complex polarization but also to the notion
of a spin structure of an oriented Riemannian manifold (M, g) and other geometric
structures like the metaunitary structure of a Hermitian vector bundle (E, H). For the
general definition recall the notion fo central extension of a Lie group G:
Let
ι ρ
1 −→ Z −→ G
e −→ G −→ 1 ,

an exact sequence of Lie groups where Z is abelian with ι(Z) in the center of G e and
where we write the trivial group as 1. In this situation G
e is called a central extension
of G by Z. Sometimes, the exact sequence is called central extension.

Definition 17.4. Consider the central extension above, and let B → X a principal
fibre bundle over a manifold X with structure group G. A ρ-lifting is a principal fibre
bundle Be → X with structure group G̃ together with a ρ-equivariant principal bundle
e → B. That is, the following diagrams are commutative:
morphism ρ̃ : B

ρ̃
B
e /B

π
π̃  
X
e×G
B / B
e
(ρ̃,ρ) ρ̃
 
B×G / B,

e → B and ρ̃′ : B
Two ρ-liftings ρ̃ : B e ′ → B are equivalent if there exists an
e → B
equivariant principal bundle map ϕ : B e ′ with ρ̃ = ρ̃′ ◦ ϕ. In particular the
250 17. Metaplectic Structure

following diagram is commutative:

ϕ
B
e / e′
B

ρ̃   ρ̃′
B

In case of a central extension of the Lie group G by Z2 ,

ι ρ
1 −→ Z2 −→ G̃ −→ G −→ 1 ,
e → B with the usual equivariance properties. An
a ρ-lifting is a double coverimg ρ̃ : B
equivalence class of ρ-liftings could be called a meta-G-structure in this case.
Metalinear and Metaplectic Structure
It is easy to see that the notions of metalinear structure and metaplectic structure fit
into the new definition of a ρ-lifting with respect to the corresponding central extensions
ι ρ
1 −→ Z2 −→ ML(n, C) −→ GL(nC) −→ 1 ,

of GL(n, C) as the structure group of the frame bundle R(E) of a complex vector
bundle E resp.
ι ρ
1 −→ Z2 −→ Mp(n) −→ Sp(n) −→ 1 ,
of Sp(n, R) as the structure group of the symplectic frame bundle by Z2 .
Let us mention another ”metastructure” which we need in the following, namely
the metaunitary structure, before we discuss the spin structure as another example.
Metaunitary Structure
Here, the group G is the unitary group G = U(n) = U(n, C). The unitary group is
the structure group of the frame bundle R(E, H) of orthonormal frames of a hermitian
vector bundle (E, H).
The metaunitary group MU(n) is the (connected) double covering of U (n) deter-
mined by the following central extension:

ι ρ
1 −→ Z2 −→ MU(n) −→ U (n) −→ 1 .

For the unitary frame bundle R(E, H) corresponding to a Hermitien vector bundle
(E, H) a ρ-lifting ρ̃ : R(E,
e H) → R(E, H) is a metaunitary frame bundle and leads to
the notion of a metaunitary structure.
Spin Structure
Let X be an oriented manifold of real dimension k with Riemannian metric g and
let π : R(X, g) → X be the principal fibre bundle of oriented orthonormal frames with
structure group SO(k) called orthonormal frame bundle. R(X, g) can be constructed
17.2 General Metastructures 251

in a way similar to the construction of the frame bundle R(X) or other frame bundles
we have encountered. In particular, the fibre π −1 (x) = Rx (X, g) is the set of ordered
bases b of Tx X, which are oriented and orthonormal with respect to g(x). Rx (X, g)
can be parametrized by SO(k) by the action (b, g) 7→ bg of SO(k): In particular,
for a fixed oriented, orthonormal and ordered basis b ∈ Rx (X, g) , x ∈ X the map
SO(k) → Bx , g 7→ bg is bijective.
Recall, that the spin group Spin(k) is the 2-1 covering group of the special orthog-
onal group SO(k), or in other words, Spin(k) is the central extension of SO(k) by Z2 ,
with respect to the exact sequence of Lie groups:
ρ
1 −→ Z2 −→ Spin(k) −→ SO(k) −→ 1 .

ρ denotes the covering map ρ : Spin(k) → SO(k).

Definition 17.5. A Spin Frame Bundle for the orthonormal frame bundle π :
R(X, g) → M over a k-dimensional oriented Riemannian manifold (X, g) is ρ-lifting

ρ̃ : R(X,
e g) → R(X, g) .

A spin structure is an equivalence class of spin frame bundles.

As in the case of metalinear structures

Remark 17.6. We sketch the cohomological condition for a principal G-bundle B → X
to admit a ρ-lifting. The isomorphism classes of principal G-bundles B can be described
by transition functions (gij ) and have corresponding cohomology classes [gij ] = [B] ∈
Ȟ 2 (U, G)87 , H 2 (U, G) is essentially Ȟ 2 (X, G). The central extension induces a long
exact sequence
δ
−→ Ȟ 1 (X, Z) −→ Ȟ 1 (X, G)
e −→ Ȟ 1 (X, G) −→ Ȟ 2 (X, Z) .

Let us call wρ (B) := δ([B]) the obstruction class of B. One can prove, that a ρ-lifting
of B exists if and only if wρ (B) is trivial. Moreover, the equivalence classes of ρ-liftings
of B are in a 1-to-1 correspndence to cohomology classes of Ȟ 1 (X, Z).
We recognize our results for the existence of a metalinear structure of a GL(n, C)-
bundle B proved in the previous chapter.
Completely in the same way one can prove a similar result for a U(n)-bundle
B → M . For instance, for the bundle of orthonormal frames B = R(E, H) of a
hermitian complex vector bundle (E, H), the following holds: A metaunitary frame
bundle R(E,
e H) exists if an only if the corresponding obstruction class w(E) vanishes.
Moreover, whenever this is the case, there is a bijection between the set of metaunitary
structures on (E, H) and Ȟ 1 (M, Z2 ).
87
Here, U is an open cover of X consisting of contractible open subsets where all the intersections
are contractible.
252 17. Metaplectic Structure

And for the spin case we obtain: A spin structure for the orthonormal frame bundle
R(X, g) exists if and only if w2 (X) ∈ Ȟ 2 (U, Z2 ), the second Stiefel-Whitney class of
T X is trivial. When this is the case, the equivalence classes of spin structures are
parametrized by Ȟ 1 (U, Z2 ).

17.3 Square Roots, Metalinear und Metaplectice Structures

After these preparations on metastructures we show in this section how a given meta-
plectic structure on the symplectic manifold (M, ω) induces on every positive polar-
ization P ⊂ T M C on (M, ω) a metalinear structure and thus a square root of the
canonical bundle KP .88
We begin with the reduction of the symplectic frame bundle R(M, ω) on the sym-
plectic manifold to a U(n)-bundle. We pick a compatible almost complex structure J
which is positive89 :

1. J is a section M → End (T M, T M ) satisfying J ◦ J = 1 = idT M .

2. gaJ (X, Y ) := ωa (X, Ja Y ) defines a symmetric, positive definite bilinear form g J

on Ta M for each a ∈ M . Hence, (M, g J ) is a Riemannian manifold.

3. ω(X, Y ) = ω(JX, JY ) for all X, Y ∈ Ta M , a ∈ M 90 .

Proposition 17.7. Let

R(M, ω, J) := {(u; v) ∈ R(M, ω) | v = (Ju1 , . . . , Jun )} .

Then R(M, ω, J) is a reduction of R(M, ω) to a U(n)-bundle and every U(n)-reduction

arises this way. Moreover, the Sp(n)-bundle R := R(M, ω, J) ×U (n) Sp(n) is a sym-
plectic frame bundle isomorphic to the symplectic frame bundle R(M, ω) .

Proof. The condition (Ju1 , . . . , Jun ) := Ju = v for a symplectic frame (u; v) remains
true for those g ∈ Sp(n), which satisfy Ju′ = v ′ when (u′ ; v ′ ) := (u; v)g, i.e. Jg = gJ.
These group elements g form exactly the subgroup Un of Sp(n) isomorphic to U(n)
according to Lemma 17.1.
The isomorphism R ∼ = R(M, ω) is given by

[(u; v), g] 7→ (u; v)g , ((u; v), g) ∈ R(M, ω, J) × Sp(n) .

88
This section is developed along the lines of a paper of Rawnsley [Raw78].
89
Such an almost complex structure exists on any symplectic manifold (M, ω): One equips M with
a Riemannian metric g and uses the condition 2. – g(X, Y ) = ω(X, JY ) – to determine J.
90
3. follows from 2.
17.3 Square Roots, Metalinear und Metaplectice Structures 253

With the almost complex structure J the real vector bundle T M becomes a complex
vector bundle T M J of dimension n on which J and ω induce the hermitian metric H J :

HaJ (X, Y ) := gaJ (X, Y ) − iωa (X, Y ) , X, Y ∈ Ta M , a ∈ M .

The U(n)-bundle R(T M J , H J ) of unitary frames of (T M J , H J ) is isomorphic to the

reduced bundle R(M, ω, J) by the map

R(M, ω, J) ∋ (u; v) 7→ u ∈ R(T M J , H J ) .

In fact, for (u; v) ∈ R(M, ω, J) with u = (u1 , . . . , un ) the condition H J (ui , uk ) =

ω(ui , Juk ) + iω(ui , uk ) = ω(ui , uk ) = δik is satisfied, so u is an orthonormal frame, and
every orthonormal frame u can be expanded to become a symplectic frame (u; v).
Now, let P be a positive polarization. We are interested in a natural bijection
between the metalinear structures on P and the metaplectic structures on (M, ω).
We have learned in the previous chapter that the metalinear structures on P are
parametrized by Ȟ 1 (U, Z2 ).
The same is true for the metaunitary structures on (T M J , H J ), cf. Remark 17.6.
Regarding the isomorphisms R(M, ω) ∼ = R(T M J , H J ) we conclude that the equivalence
classes of metaplectic frame bundles of (M, ω) is parametrized by Ȟ 1 (U, Z2 ) as well.
As a result, there is a bijection between the set of metalinear structures on P and the
set of metaplectic structures on M given by any bijection of Ȟ 1 (U, Z2 ), in paricular by
the identity id.
However, we are interested to describe and construct a natural bijection using
geometric insight. So we have to go into some details.
As a first step of our way from a metaplectic frame bundle R(M, e ω) to the corre-
sponding metalinear frame bundle R(P ) on a given positive polarization P we have
e
started with the symplectic frame bundle R(M, ω) and arrived at the U(n)-bundle
R(T M J , H J ) of hermitian frames of the tangent bundle of M as a reduction of R(M, ω).
The strategy is to identify this bundle with the U(n)-bundle R(P, H J ) of unitary frames
of P where P is the positive polarization in question.
When this is done, a metalinear structure R(Pe ) on P gives a metaunitary structure
e H ) by reduction. Using the close relations R(M, ω) → R(M, ω, J) ∼
R(P, J
= R(P J , J)
this metastructure is transferred to yield a metaplectic structure on M .
We introduce te notion of Lagrangian subbundle which generalizes the notion of
polarization:

Definition 17.8. A Lagrangian subbundle P ⊂ T M C is an n-dimensional complex

vector bundle in the complexification T M C of the tangent bundle T M such that each
Pa , a ∈ M is a Lagrangian subspace of (Ta M C , ωa ), i.e. ω|P ×P = 0.
P is called positive, if −iωa (X, X) ≥ 0 for all a ∈ M , X ∈ Pa .
254 17. Metaplectic Structure

A polarization is an involutive Lagrangian subbundle as introduced in Chapter 9.

The results we present in the following hold for the more general case of Lagrangian
subbundles.
Proposition 17.9. Let P be a positive Lagrangian subbundle of T M C and let J be a
positive almost complex structure on (M, ω). T M (1,0) denotes the i-eigenspace of J :
T M C → T M C and T M J the tangential bundle equipped with the complex multiplication
by J. Then the complex vector bundles P , T M J and T M (1,0) are isomorphic in a
natural way.
As a consequence, when the hermitian form H J is transformed to P by the isomor-
phim T M J → P we have a natural isomorphism R(T M J , H J ) ∼
= R(P, H J ).
Moreover the Chern classes of the three vector bundles agree with ci (M, ω), the
Chern classes of M .

Proof. The i-eigenspace of J is Im π for the projection

1
π : T M C → T M C , π(X) := (X − iJX) .
2
Moreover, T M (1,0) = π(T M ) and the restriction of π to the complex tangent bundle
T M J is an isomorphism, i.e. R-linear and
1 1
π(JX) = (JX − iJJX) = (JX + iX) = i(π(X)) , X ∈ T M .
2 2
T M (1,0) is a positive Lagrangian and −iω(X, X) > 0 for X ∈ T M (1,0) , , X ̸= 0.
Hence, the orthogonal complement T M (1,0) = Im (1 − π) satisfies −iω(X, X) < 0 for
X ∈ T M (1,0) , X ̸= 0. We conclude P ∩ T M (1,0) = ∅. As a result, the restriction of π
to P is an isomorphism, since π(P ) ⊂ T M (1,0) and π|P is injective: For X, Y ∈ P the
equality π(X) = π(Y ) implies X − Y ∈ Ker π = T M (1,0) , hence X − Y ∈ P ∩ T M (1,0) =
∅, i.e. X = Y .

Construction
After these preparations we now concentrate on a positive Lagrangian P on (M, ω)
and assume that it is equipped with a metalinear structure represented by a metalinear
e ) → R(P ). We use this structure to construct an associated
frame bundle ρ̃ : R(P
metaplectic frame bundle R(M,
e ω) → R(M, ω) and hence a metaplectic structure on
(M, ω).
As before, pick a positive compatible almost complex structure J on M and trans-
form the Hermitian form H J on T M J to P by the isomorphism T M J ∼ = P (cf. Propo-
sition 17.9). The Lagrangian P thus becames a Hermitian vector bundle (P, H J ). The
corresponding unitary frame bundle R(P, H J ) is a subbundle of the GL(n, C)-frame
bundle R(P ). The subbundle
e := (ρ̃)−1 (R(P, H J )) ⊂ R(P
B e )
17.3 Square Roots, Metalinear und Metaplectice Structures 255

is a M U (n)-bundle and turns out to be a metaunitary frame bundle

e H J ) := B
ρ̃ : R(P, e → R(P, H J ) .

Again using the isomorphism T M J ∼= P this construction yields a metaunitary

frame bundle
e M J , H J ) → R(T M J , H J ) .
R(T

Considering the isomorphism R(M, ω) ∼

= R(M, ω, J)×U(n) Sp(n) of Proposition 17.9
we define the Mp(n)-bundle
eJ := R(T
R e M J , H J ) ×MU(n) Sp(n) −→ R(M, ω)

with the 2-to-1 mapping

eJ → R(T M J , H J )×U(n) Sp(n) , [(b, S)] 7→ [(ρ̃(b), S)] , (b, S) ∈ R(T
σ:R e M J , H J )×Sp(n) .

Then it is not difficult to check that

e M J , H J ) ×MU(n) Sp(n) → R(T M J , H J ) ×U(n) Sp(n)
σ : R(T

is a metaplectic frame bundle determining a metaplectic structure on (M, ω).

Proposition 17.10. For each positive Lagrangian P the above construction exhibits a
natural bijection between the metalinear structures on P and the metaplectic structures
on M .

Proof. We have constructed a map [R(P e M J , H J ) ×MU(n) Sp(n)] from the

e )] 7→ [R(T
metalinear structures on P to the metaplectic structures on M . Since both sets are
paramterized bx Ȟ 1 (M ; Z2 ) it is enough to check that the assignment is injective. The
inverse map can be determined by reversing each step: Given a metaplectic frame bun-
dle ρ̃ : R(M,
e ω) → R(M, ω) the reduction to U(n), B e J := (ρ̃)−1 (R(M, ω, J)) , B
eJ →
R(M, ω, J) ∼ = R(P, H J ) yields a metaunitary frame bundle B e J → R(P, H J ) of (P, H J ).
R(P
e ) := B e J ×MU(n) ML(n, C) → R(P ) is the metalinear frame bundle of (P ) yielding
the inverse map [R(M,
e ω)] 7→ [Be J ×MU(n) ML(n, C)].

Remark 17.11. Blattner [Bla77] (see also Sniaticky [Sni80] and Tuynman [Tuy16])
presents an alternative proof of Proposition 17.10 using as an intermediate structure
the bundle of positive Lagrangian frames F+ (M, ω) on (M, ω). A positive Lagrangian
frame at a ∈ M is an n-tuple (u1 , . . . , un ) of vectors uj ∈ Ta M C such that ωa (ui , uk ) = 0
and −iωa (ūi , uk ) is positive semi definite. Note, when P is a positive Lagrangian then
each element b ∈ R(P ) of the frame bundle R(P ) is a positive Lagrangian frame.
F+ (M, ω) consists of the collection of all positive Lagrangian frames at all points of M ,

F+ (M, ω) = {b ∈ P | P ⊂ T M C positive Lagrangian subbundle} .

256 17. Metaplectic Structure

In particular, every positive polarization P is a subbundle of F+ (M, ω). On F+ (M, ω)

there are natural actions of Sp(n) (from the left) and of GL(n, C) from the right. Given
a metaplectic frame bundle R(M,e ω) → R(M, ω) by direct geometric considerations
about F+ (M, ω) a ”metaplectic” bundle of positive Lagrangian frames

ρ̃ : Fe+ (M, ω) → F+ (M, ω)

can be constructed which induces on each positive Lagrangian R(P ) ⊂ F+ (M, ω) a

e ) := (ρ̃)−1 (R(P )).
metalinear frame bundle R(P

Remark 17.12. A different way to introduce and apply the concept of metaplectic
structure has been suggested by Woodhouse [Woo91], p.232. In order to explain this
approach we need the space L+ (M, ω) of positive Lagrangian subspaces over M :
[
L+ (M, ω) := {P ⊂ Ta M C | P positive Lagrangian subspace of (TM , ωa )} .
a∈M

This is a topological fibre bundle with fibres

{P ⊂ Ta M C | P positive Lagrangian subspace of (TM , ωa )}.

A metaplectic structure on (M, ω) is defined to be a square root of the canonical bundle

K = KL+ M of L+ (M, ω). A positive Lagrangian P on (M, ω) is essentially the same
as a section s : M → L+ (M, ω) , a 7→ Pa : P ∼ = s(M ). P obtains a square root of
∗
the canonical bundle KP by pullback: KP = s (K). In this way, a given metaplectic
structure induces directly a square root on every positive polarization of (M, ω).

17.4 Half-Form Pairing

17.5 Mteplectic Representation

Summary:
 257

Appendix: Mathematical Supplements

A Manifolds

There are many different classes of manifolds which are investigated in the various fields
of mathematics and physics. In these lecture notes we are interested only in differen-
tiable91 or complex manifolds and in the differentiable bundles over these manifolds.
In this chapter we summarize the basic notions and results for differentiable man-
ifolds. Complex manifolds will be treated in a later section (in Chapter B) as well as
the principal bundles and the vector bundles over manifolds (in Chapter D).

A.1 Basic Definitions

Definition A.1 (Manifold). Let M be a Hausdorff space and n ∈ N, n ≥ 0.

• A manifold 92 of dimension n is M together with a differentiable structure D.

• A differentiable structure on M is an equivalence class D of differentiable atlases

A on M .

• A differentiable atlas 93 (”Atlas”) on M is a collection A = (qι : Uι → Vι )ι∈J of

(n-dimensional) charts of M which are smoothly compatible to each other and
which cover M : [
M= Uι = M.
ι∈J

• An (n-dimensional) chart (”Karte”) on M is topological map q : U → V (i.e. q

is continuous with a continuous inverse q −1 : V → U ), where U ⊂ M is an open
subset of M and V ⊂ Rn is an open subset of Rn .

• Two such charts q : U → V, q ′ : U ′ → V ′ on M are called smoothly compatible94

if the induced map

q ′ ◦ q −1 |q(U ∩U ′ ) : q(U ∩ U ′ ) → q ′ (U ∩ U ′ )

is smooth (i.e. infinitely often differentiable95 ), or if U ∩ U ′ = ∅.

91
i.e. infinitely differentiable, also called smooth
92
more precisely a differential manifold; but in these notes we will only deal with differentiable
manifolds.
93
In the following mostly called simply atlas, since only differentiable atlases will be considered.
94
in the following we say simply compatible
95
We only consider smooth functions and C ∞ structures in these notes. In general, also C k -
differentiable atlases and C k -differentiable structures are studied.
258 A. Manifolds

• Two atlases A, B are equivalent, if their union is an atlas, i.e. if each chart of A
is compatible with each chart of B.
Observation A.2 (Maximal Atlas). Let M be a manifold with its differentiable struc-
ture D.

• Any atlas A in the equivalence class D determines a maximal atlas M given as

the collection of all charts on M which are compatible to the charts of A. Then
M is maximal in D with respect the inclusion of atlases.

• The maximal atlas M of D is the union of all the atlases in D.

• As a result one could say that a manifold is a Hausdorff space M together with
a maximal atlas M.

• If we refer to a chart on M then it is always a chart of some atlas determining

the differentiable structure of M , in particular, it will be chart of M.

In general, in these notes a manifold will be metrizable, and therefore paracompact.

In most cases the manifold is also assumed to be connected.
Definition A.3 (Smooth Mappings and Functions). A smooth (or differentiable) map-
ping on an open subset D ⊂ M of a manifold into another manifold N is a mapping
f : D → N such that for every a ∈ D there exist charts φ : U → V ⊂ Rn on M and
φ′ : U ′ → V ′ ⊂ Rm on N with U ⊂ D and f (U ) ⊂ U ′ such that

φ′ ◦ f ◦ φ−1 : V → Rm

is smooth. Notation for manifolds M, N :

E(M, N ) := {f : M → N | f , smooth}.

E(M ) = E(M, K),

where K ∈ {R, C}. Smooth mappings with values in K are often called smooth func-
tions or simply functions. A diffeomorphism is a smooth map F : M → N with a
differentiable inverse.

Submanifold:
Definition A.4. A submanifold (”Untermannigfaltigkeit”) N of a manifold M is given
by a subset N ⊂ M such that restrictions of suitable charts on M to N provide an
atlas on N : For each a ∈ N there exists a chart q : U → V on M such that a ∈ U and
q(U ∩ N ) = {y ∈ Rn | yd+1 = . . . = yn = 0} ∼
= V ∩ Rd , where d ∈ N, d ≤ n, providing
the chart
q|U ∩N : U ∩ N → V ⊂ Rd
of N .
A.1 Basic Definitions 259

These charts are automatically compatible to each other and define the structure
of a d-dimensional manifold. In particular, open subsets U ⊂ M are submanifolds.

Proposition A.5. A subset N of an n-dimensional manifold M defines a submanifold

of dimension d ≤ n if and only if either

• d = n and N is open in M , or

• d < n, N is contained in an open subset W ⊂ M , and for every a ∈ W there

are an open neighborhood U and functions f1 , . . . fn−d ∈ E(U ) such that U ∩ N =
{x ∈ U | fj (0) = 0 for all j = 1, . . . , n − d} and rk(f1 , . . . , fn−d ) = n − d.96

Notation. n − d is called the codimension of N .

A closed submanifold N of codimension 1 is called a hypersurface (”Hyperfläche”).

Observation. With the notation of Proposition A.5 a submanifold is closed in the

open subset W of M , but, in general, not in M .

Product Manifold:

Definition A.6. The product (manifold) (”Produktmannigfaltigkeit”) of two mani-

folds M, N is the Hausdorff space M × N with the differentiable structure given by all
the charts
q × q′ : U × U ′ → V × V ′,
where q ′ : U → V is a chart on M and q ′ : U ′ → V ′ is a chart on N .

Proposition A.7. The product M1 × M2 of two manifolds M1 , M2 together with the

projections pj : M1 × M2 → Mj , (x1 , x2 ) 7→ xj , j = 1, 2 satisfies the following universal
property: Every map f : M → M1 × M2 is smooth if and only if the compositions
p1 ◦ f and p2 ◦ f are smooth. And any manifold P with smooth pj : P → Mj satisfying
the above universal property is isomorphic to M1 × M2 (i.e. there is a diffeomorphism
f : P → M × N ).

Slightly more general is the notion of a fibre product:

Definition A.8. The fibre product (”Faserprodukt”) of two mappings f : M → S, g :

N → S over a third manifold S is the submanifold

M ×S N := {(x, y) ∈ M × N | f (x) = g(y)}

of the product M × N , together with its map π : M ×S N → S , (x, y) → f (x) = g(y).

96
here we use the notion of rank of a smooth mapping which is explained later.
260 A. Manifolds

Note, that the two induced maps f ∗ (x, y) := x , g ∗ (x, y) =: y for (x, y) ∈ M ×S N
satisfy: π = f ∗ ◦ g = g ∗ ◦ f , i.e. we have the following commutative diagram

f∗
M ×S N / N
π
g∗ g
 $/ 
M f
S

Proposition A.9. The fibre product M ×S N → S of satisfies the following universal

property: When the (smooth) mappings r : Z → M , t : Z → N satisfy f ◦ r = g ◦ t :
Z → S then there exists a unique r ×S t : Z → M ×S N such that r = g ∗ ◦ (r ×S t) and
t = f ∗ ◦ (r ×S t), in particular π ◦ (r ×S t) = f ◦ r = g ◦ t : Z → S. And any manifold P
with smooth fˆ : P → M , ĝ : P → N with the above universal property is isomorphic
to M ×S N .

Quotient Manifold
Definition A.10. A quotient (manifold) (”Quotientenmannigfaltigkeit”, ”Quotient”)
of a given manifold M with respect to an equivalence relation ∼ on M is any manifold
Q together with a surjective and smooth map π : M → Q such that the following
universal property is satisfied: For every smooth f : M → N such that f is constant
on the equivalence classes of ∼ there exists a unique smooth g : Q → N such that
f = g ◦ π.
f
M /
>N
π
g

Q

A quotient manifold is unique up to diffeomorphism.

Note, that the existence of the quotient as a manifold is guarantied only for special
equivalence relations ∼. In particular, the relation ∼ = {(x, y) ∈ M × M | x ∼ y} has
to be closed as a subset of M × M to ensure that M/ ∼ is at least Hausdorff within
topological spaces. We come back to quotient manifolds in Section A.4.

A.2 Tangent and Cotangent Bundle

Each curve x ∈ E(I, M ) defined on an open interval I ⊂ R with 0 ∈ I through the

point a := x(0) ∈ M determines a Tangent Vector [x]a at a: [x]a is the equivalence
class of (germs of) curves in M through a which is given by the equivalence relation
d d
x ∼a y (i.e. [x]a = [y]a ) ⇐⇒ (q ◦ x)|t=0 = (q ◦ y)|t=0 ,
dt dt
where y ∈ E(I, M ) with y(0) = a.
A.2 Tangent and Cotangent Bundle 261

Definition A.11. The Tangent Space (”Tangentialraum”) at a ∈ M is the space

Ta M := {[x]a | x ∈ E(I, M ), x(0) = a}

of all equivalence classes [x]a .

Observation A.12. The set Ta M carries a natural structure of real n-dimensional
vector space: For [x]a , [y]a ∈ Ta M (with x(0) = y(0) = a) and λ ∈ R one chooses a
chart q : U → V ⊂ Rn with q(a) = 0 and defines

z(t) := q −1 (q ◦ x(t) + λq ◦ y(t)) , t ∈ I0 ,

for a suitable small open interval I0 containing 0. Then [z]a is independent of the choice
of the chart q and we set
[x]a + λ[y]a := [z]a .
Proposition A.13. The Tangent Bundle (”Tangentialbündel”)
[
TM = Ta M,
a∈M

together with the projection

τ = τM : T M → M , [x]a 7→ a,

has a natural structure of a 2n-dimensional manifold where τ is a smooth map and the
fibers τ −1 (a) = Ta M are n-dimensional vector spaces.

Proof. For a chart q : U → V ⊂ Rn the corresponding bundle chart q̃ : τ −1 (U ) →

U × Rn is defined by
d
q̃([x]a ) := (a, q ◦ x(t)|t=0 ) , x(0) = a.
dt
q̃ is bijective, since q̃a := q̃|Ta M : Ta M → Rn is bijective (and linear). For another chart
q ′ : U ′ → V ′ we obtain, with g := q ′ ◦ q −1 , the change of the bundle charts

q˜′ ◦ q̃ −1 |T (U ∩U ′ ) : q̃(U ∩ U ′ ) = q(U ∩ U ′ ) × Rn → q˜′ (U ∩ U ′ ) = q ′ (U ∩ U ′ ) × Rn ,

as the map given by

(q, v) 7→ (g(q), Dg(q).v) ,
where Dg(q).v is the Jacobi matrix (or derivative) Dg(q) of g at q ∈ q(U ∩U ′ ) applied to
v ∈ Rn . In particular, this description shows that q˜′ ◦ q̃ −1 |T (U ∩U ′ ) is a diffeomorphism.
Now, the topology on T U := τ −1 (U ) will be defined by q̃ : T U → U × Rn in such
a way that q̃ is a topological map (i.e. continuous with a continuous inverse). For
another chart q ′ : U ′ → V ′ the topologies on T (U ∩ U ′ ) induced by q̃ and q˜′ agree, since
q˜′ ◦ q̃ −1 |T (U ∩U ′ ) is a diffeomorphism, hence in particular, a topological map. Because
262 A. Manifolds

of the smoothness, the charts q̃ and q˜′ are compatible and thus define a differentiable
structure when the topology on T M turns out to be Hausdorff. But the Hausdorff
property can indeed be proven quite easily.
The proof is complete, but let us describe the differentiable structure using atlasses.
Note, that with respect to an atlas A = (qj : Uj → Vj ) of charts qj defining the
differentiable structure of the manifold M the change of charts is given by qi ◦ qj−1 :
qj (Uij ) → qi (Uij ), where Uij = Ui ∩ Uj , and for the induced bundle charts we obtain
the diffeomorphisms q̃i ◦ q̃j−1 : qj (Uij ) × Rn → qi (Uij ) × Rn . Thus, the bundle charts
(q̃j ) form an atlas of the differentiable structure of the tangent bundle T M .

Observation. For the bundle charts q̃ : T U = τ −1 (U ) → U ×Rn the following diagram

is commutative
q̃
TU /U × Rn
τ
pr1
 z
U
where pr1 : U × Rn → U, (x, v) 7→ x, is the natural projection. Moreover, for a ∈ U the
induced map Ta M → {a} × Rn ∼ = Rn is liner over R.
This property essentially implies that T M → M is a (smooth) vector bundle of
rank n (see Section D.1 for an exposition of vector bundles). Moreover, for an atlas of
charts (qj ) as at the end of the proof the preceding proposition the transition of bundle
charts has the form

q̃i ◦ q̃j−1 : qj (Uij ) × Rn → qi (Uij ) × Rn , (q, v) 7→ (qi ◦ qj−1 (q), D(qi ◦ qj−1 )(q).v) .

Consequently, the transition functions gij : Uij → GL(n, R) of the vector bundle T M
in the way they are used in Section D.1 and elsewhere are given by gij (q) = Dqi ◦
qj−1 (q)GL(n, R).

Observation A.14. For a smooth map f : M → N between manifolds M, N the

Tangent Map (”Tangentialabbildung”), Derivative (”Ableitung”) (also called total
derivative) or Differential of f at a point a ∈ m is given by

Ta f : Ta M → Tf (a) N , Ta F ([x]a ) := [f ◦ x]f (a) .

Ta f is linear. Moreover, the Ta f, a ∈ M, fit together to define the Tangent Map

(Derivative or Differential)

T f : T M → T N , v 7→ Ta (v) , v =∈ τ −1 (a) = Ta M ,

which turns out to be a smooth map. The smoothness can be proven by using the
respective bundle charts. T f is compatible with the projections (i.e. f ◦ τM = τN ◦ T f )
and R-linear in the fibers Ta M, Tf (a) N . Hence, T f is a vector bundle (homo-) morphism
A.2 Tangent and Cotangent Bundle 263

over f . The compatibility condition can also be expressed by stating that the diagram
Tf
TM / TN
τM τN
 
M /N
f

is commutative: f ◦ τM = τN ◦ T f .

Cotangent Bundle:
Similarly, one introduces the cotangent bundle (”Kotangentenbündel”)
[
T ∗ M := Ta∗ M,
a∈M

where
Ta∗ M := Hom(Ta M, R) = (Ta M )∗
with the projection T ∗ M → M denoted by τ or τ ∗ . T ∗ M is a vector bundle of rank n,
as well.
Every manifold is endowed not only with the two vector bundles T M ans T ∗ M ,
but in addition, with further naturally associated vector bundles. Important for the
sequel are the bundles of k-forms for k = 1, 2, . . . , n:
^k
Λka M := (Ta M ) := {α : (Ta M )k → R | α k−multilinear over R and alternating}
[
Λk M := Λka M
a∈M

with smooth projections

τ : Λk M → M.
The Λk M are vector bundles of rank nk .

Exercise A.15. Describe the differentiable structure of the real vector bundles Λk M
by bundle charts.

Vector Fields:
A vector field (”Vektorfeld”) on the manifold M is a smooth map

X : M → T M , with X(a) ∈ Ta M for all a ∈ M,

i.e. X is a smooth Section of the tangent bundle:

X ∈ E(M.T M ) , with τ ◦ X = idM .

264 A. Manifolds

We denote the set of all vector fields on M by V(M ). V(M ) is a vector space
over R and a module over the commutative ring E(M ) with respect to the following
operations for X, Y ∈ V(M ) and f ∈ M:

(X + Y )(a) := X(a) + Y (a) , a ∈ M and

(f X)(a) := f (a)X(a) , a ∈ M.
It turns out that, in addition, V(M ) is a Lie algebra over E(M ). This fact explains
the notation V(M ) with a V in Gothic type.

Differential Forms
Definition A.16. A differential form (”Differentialform”) or simply form of de-
gree k is a section in the bundle Λk M :

Ak (M ) := {α ∈ E(M, Λk M ) | τ ◦ α = idM }

The differential forms of degree k are mostly called k-forms.

Let α ∈ Ak (M ) be k-form and f ∈ E(M ) a smooth function. Then f α : M →

Λk (M ) is defined by (f α)(a) := f (a)α(a), a ∈ M . Certainly, f α is smooth and a
section, hence f α ∈ Ak (M ). Moreover,

E(M ) × Ak (M ) → Ak (M ) , (f, α) 7→ f α ,

is linear over the commutative ring E(M ),

Similarly, for a k-form α ∈ Ak (M ) and (X1 , . . . , Xk ) ∈ (V(M ))k by

α(X1 , . . . , Xk )(a) := α(a)(X1 (a), . . . , Xk (a)), a ∈ M,

one obtains a smooth function α(X1 , . . . , Xk ). Tha map

(V(M ))k → E(M ) , (X1 , . . . , Xk ) 7→ α(X1 , . . . , Xk )

ia k-multilinear over E(M ). It is easy to show:

Proposition A.17. Ak (M ) is an E(M )-module with respect to the multiplication de-
fined above. Moreover the following map is an isomorphism of E(M )-modules
^k
Ak (M ) ∼
= (V(M )) , ω 7→ ((X1 , . . . , Xk ) 7→ ω(X1 , . . . , Xk )) ,

identifying the respective E(M )-modules:

A(M ) := A1 (M ) ∼
= (V(M ))∗ := HomE(M ) (V(M ), E(M )) ,
^k
Ak (M ) ∼
= (V(M )) .
A.2 Tangent and Cotangent Bundle 265

Here, for a module W over a commutative ring R with 1 one defines

^k ^k
W = W := {β : W k → R | β k−multilinear over R and alternating}
R

Remark A.18. The preceding Proposition A.17 suggests an alternative definition of

a k-form by
^k
Ak (M ) := (V(M )) ,
k
VkE(M )-multilinear and alternating η : V(M ) → E(M ) induces a section
since every
η:M → M by η(a)(X1 , . . . , Xk ) = η(X1 , . . . , Xk )(a) for Xj ∈ V(M ).

Local Expressions
In the following we present local expressions for the vector fields and forms which
are used throughout the notes.
Notation A.19. Let M be a manifold of dimension n with its tangent bundle T M
and its cotangent bundle T ∗ M .

1. The Charts on the manifold M defining the differentiable structure of M are

mostly written in the following way

q : U → V , q = (q 1 , q 2 , . . . , q n ) ,

where U ∈ M is an open subset in M , V ∈ Rn is an open subset of Rn and q is

differentiable with differentiable inverse.
2. The smooth functions q j : U → R (the components of q) are called the (local)
Coordinates given by the chart q.
3. A chart q : U → V provides for each a ∈ U a natural vector space basis
 
∂ ∂ ∂
(a), 2 (a) . . . , n (a)
∂q 1 ∂q ∂q
of the Tangent Space Ta M of M at a, where
∂
(a) := q −1 ((q(a) + tej )) a
 
∂q j

is given as the tangent vector of the curve q −1 (q(a)+tej ) , t ∈ ]−ε, ε [ through a ∈

U and where (e1 , . . . , en ) is the standard unit vector basis of Rn : For convenience,
these tangent vectors at a are abbreviated as
∂ ∂
∂j (a) := j
(a) , or resp. ∂j ,
∂q ∂q j
when it is clear from the context which coordinates q are used resp. for which
point a the expressions are employed.
266 A. Manifolds

4. Note that
∂ ∂
j
: U → T U , a 7→ j (a) ,
∂q ∂q
is a Vector Field on U . Moreover, every vector field X : U → T U can be
described uniquely by
∂
X = X j j = X j ∂j ,
∂q
where the coefficients X are smooth, X j can be obtained by
j

d
X j (a) = (q j ◦ x)(t) |t=t0


dt
where the curve x : I → U, x(t0 ) = a, represents X at a: X(a) = [x]a .
5. Every chart q : U → V induces a Bundle Chart (cf. A.2) q̃ : T U → V × Rn
on the Tangent Bundle T M :
 
1 n 1 n

n d
q̃ = q , . . . , q , v , . . . , v : T U → V × R , [x]a 7→ (q(a), (q ◦ x)|t=t0 ,
dt
when x(t0 ) = a. Here, v j acts as
 
j d j
v ([x]a ) = (q ◦ x)|t=t0 , resp.
dt
v j (X(a)) = X j (a) , for a vector field X = X j ∂j ∈ V(U ).
6. A chart q : U → V provides for each a ∈ U also a natural vector space basis
dq 1 (a), dq 2 (a) . . . , dq n (a)

of the Cotangent Space Ta∗ M of M at a, where

 
j d j
dq (a) ([x]a ) := (q ◦ x)|t=t0
dt
Hence, dq j coincides with v j (see above). For convenience, dq j (a) is abbreviated
as dq j when it is clear for which point a the expressions are employed.
7. Note that
dq j : U → T ∗ U , a 7→ dq j (a) ,
is a 1-Form on U . Moreover, every 1-form α : U → T ∗ U over U can be described
uniquely by
α = αj dq j ,
where the coefficients αj are smooth. αj can be obtained by
 
∂
αj (a) = α(a) (a) = α(∂j )(a) .
∂q j
And αj (X) = X j for a vector field X ∈ V(U ).
A.3 Vector Fields and Dynamical Systems 267

8. Every chart q : U → V induces a Bundle Chart (cf. A.2) q̃ : T U → V × Rn

on the Cotangent Bundle T ∗ M :
 X 
q̃ = q 1 , . . . , q n , p1 , . . . , pn : T ∗ U → V × Rn , α 7→ q(a),


αj ej ,

when α = αj dq j (a). As a consequence, pj acts as

pj (α) = αj , if α = αj dq j (a) in Ta∗ U , or

pj (α(a)) = αj (a) , for a 1-field α = αj dq j ∈ A(U ).

A.3 Vector Fields and Dynamical Systems

A Dynamical System is essentially an autonomous differential equation of first order

on a manifold M . Such a dynamical system will be represented by a a pair (M, X)
consisting of a manifold M and a vector field X on M .97 A special example has been
introduced in Section 1.1 with M = U an open subset of Rn and X = XH a Hamiltonian
vector field (see 1.2) determined by a function H on T ∗ U .
For a dynamical system (M, X) the corresponding differential equation is

q̇ = X(q).

Here, q̇ is the same as the tangent vector [q]a at a given by the curve q = q(t). Any
curve q : I → M (I ⊂ R an open interval) is a solution of the dynamical system (also
called integral curve) if it satisfies q̇(t) = X(q(t)) for all t ∈ I.
The elementary theory of differential equations of first order establishes the follow-
ing result

Proposition A.20. Let (M, X) be dynamical system. For each point a ∈ M there
exists a unique maximal solution qa : ]t− (a), t+ (a)[ → M with q̇a = X(qa ) and qa (0) = a
satisfying

1◦ Mt := {a ∈ M | t ∈ ]t− (a), t+ (a)[ } is open in M with

S
t∈R Mt = M .

2◦ Φt : Mt → M−t , Φt (a) := qa (t), is a diffeomorphism with Φ−1

t = Φ−t .

3◦ M∗ = a∈M {a} × ]t− (a), t+ (a)[ is open in M × R and

S

Φ : M∗ → M , (a, t) 7→ Φ(a, t) := Φt (a) = qa (t)

is differentiable.
97
This is the right concept for geometry and elementary analysis. There are more general concepts
in other mathematical domains, e.g. ergodic systems in Stochastics.
268 A. Manifolds

Notation A.21. The map Φ =: ΦX in the above Proposition A.20 is called the (local)
Flow of the vector field X. X is called Complete if M∗ = M ×R i.e. if ]t− (a), t+ (a)[ =
R for all a ∈ M and t ∈ R. In this case, (ΦXt )t∈R is a one parameter group: Φs+t =
X
X X
Φs ◦ Φt of diffeomorphisms.
In the general case, when X is not complete, (ΦX ) is a local one parameter group.

Note, that on a compact manifold M every vector field is complete.

Proposition A.22. A one parameter group (Φt )t∈R of diffeomorphisms

Φt : M → M , t ∈ R

(i.e. Φ0 = idM , Φs+t = Φs ◦ Φt for s, t ∈ R with (a, t) 7→ Φt (a)) differentiable) induces

a unique vector field X such that ΦXt = Φt .

Proof. The induced vector field is

 
d
X : E(M ) → E(M ) , g 7→ a → g(Φt (a))|t=0 .
dt
Or, in another description, the tangent vector X(a) = [Φt (a)]a ∈ Ta M at a ∈ M is the
tangent vector given by the curve t 7→ Φt (a) through a. X is called the Infinitesimal
Generator of (Φt ).

Since only local curves are needed for the definition of X the result extends to the
local one parameter groups, as well. As a consequence, the vector fields (the dynamical
systems) on M can be identified with the local one parameter groups on M .
Lie Derivative of Vector Fields
The concept of the flow of a vector field X on a manifold M enables us to extend
the notion of a Lie derivative of functions to the notion of a Lie derivative of vector
fields (and, moreover, of differential forms, see (71)). Let X ∈ V(U ) a vector field on
an open subset U of a manifold M and let Φ = ΦX the local flow of the vector field.
As before, let Φt (a) = Φ(a, t).
In the case of a function f ∈ E(U ) on an open subset U ⊂ M one can compare f (a)
to a neighbouring f (Φt (a)) (a ∈ U ), and define
d
LX f (a) = f (Φt (a))|t=0 .
dt
In the case of a vector field Y : U → T M such a comparison is not available,
in general: Y (a) and Y (Φt (a)) live in different fibres Ta M and TΦt (a) M . Therefore, a
suitable isomorphism between these fibres could help. And, indeed, the flow Φ provides
such a natural isomorphism: Denote

((Φ−t )∗ Y )(a) = (Ta Φt )−1 Y (Φt (a)) = TΦt (a) Φ−t Y (Φt (a)) ;
A.3 Vector Fields and Dynamical Systems 269

then
(Φ−t )∗ : TΦt (a) M → Ta M
is an isomorphism. Now, Y (a) and (Φ−t )∗ Y (Φt (a)) are both contained in Ta M and can
be compared.

Definition A.23. The Lie Derivative of the vector field Y along the vector field X
is
d
LX Y (a) := ((Φ−t )∗ Y )(a)|t=0 , a ∈ U.
dt
Proposition A.24. For a vector field X ∈ V(M ), for f ∈ E(M ) and for Y ∈ V(M )
one has
1. LX f = df (X).
2. LX Y = [X, Y ].
3. LX (f Y ) = (LX f )Y + f LX Y
Moreover, some natural linearity properties are satisfied.

Proof. 1. is obvious. To show 2. let Ψ be the flow of Y . Then for fixed t:

TΦt (a) Φ−t Y (Φt (a)) = TΦt (a) Φ−t [Ψu ]Φt (a) = [Φ−t ◦ Ψu ◦ Φt )]Φt (a) .

For f ∈ E(M ) it follows that

d
(Φ−t )∗ Y (a)f = TΦt (a) Φ−t Y (f (Φt (a)) = f (Φ−t ◦ Ψu ◦ Φt (a))|u=0 .
du
Therefore,

d
LX Y f (a) = ((Φ−t )∗ Y )f (a)|t=0
dt  
d d
= f (Φ−t ◦ Ψu ◦ Φt (a))|u=0 |t=0
dt du
d d
= (f (Ψu ◦ Φt (a)) − f (Φu ◦ Ψt (a))) |u=0 |t=0
dt du
= L[X,Y ] f (a)

where we use the product rule. This completes the proof of 2., and 3. is again obvious.

Note, that the Lie derivative can be extended to all tensor fields. The case of
differential forms is treated in Section A.5.
270 A. Manifolds

A.4 Quotient Manifold

Let M be a manifold with an equivalence relation R ⊂ M × M on M and let π : M →

M/R be the natural projection onto the quotient, i.e. the set of equivalence classes of R.
In several situations one is interested in the quotient M/R as a manifold, the quotient
manifold or differential quotient. The pragmatic way to achieve this, is to endow M/R
with a differentiable structure and to use this structure for further investigation. In
general, little attention is paid to the question, whether this choice yields the quotient
manifold, or whether or not the quotient manifold exists at all. One reason for this
carelessness might be, that the definition of quotient which is shaped as a universal
property is not well adapted to the situation in the study of differentiable manifolds.
Quite general, it is known that in many categories the notion of subobject, quotient
object or epimorphism is not easy to define in a satisfying matter.
In this section we present some examples in detail, describe results about the exis-
tence of the quotient structure and establish criteria which make sure whether or not
a candidate, i.e. a differential structure on M/R is indeed a quotient structure.
In order that the quotient manifold exists, M/R has to be a Hausdorff space in
the quotient topology. We thus begin with the investigation of topological quotients of
spaces M which are merely topological spaces.

A.4.1 Topological Quotients

Definition A.25. Let M be a topological space and R an equivalence relation. A

Topological Quotient with respect to R is a topological space X together with a
map p : M → X with the following properties:

1◦ The fibres p−1 (x) of p are the equivalence classes of R, and p is surjective,
i.e. (X, p) describes the equivalence relation exactly,

2◦ p is continuous,

3◦ whenever f : M → Y is continuous and constant on the equivalence classes, the

induced map fˆ : X → Y , f = fˆ ◦ p, is continuous.

The topology on X is called the Quotient Topology.

The last property can be formulated as a universal property: For every commutative
diagram
f
M /
>Y
p
g

X
with Y a topological space and f, g maps the following holds:
A.4 Quotient Manifold 271

f is continuous if and only g is continuous.

It is evident, that a topological quotient is unique up to isomorphism, i.e. for another
topological quotient p′ : M → X ′ there exists a unique homeomorphism h : X → X ′
with h ◦ p = p′ .
Lemma A.26. Let M be a topological space with an equivalence relation R and the
natural projection π : M → M/R. The topological quotient exists. The quotient
topology on M/R is the collection of all subsets V ⊂ M/R for which the inverse image
π −1 (V ) is open in M .

Proof. π is continuous: If V ⊂ M/R is open, then π −1 (V ) is open by definition.

Moreover for any continuous f : M → Y into another topological space Y the function
fˆ(p(a)) := f (a) , a ∈ M , is continuous: If W ⊂ Y is open, then U := f −1 (W ) is
open in M . Since π −1 (π(U )) = U the image V := π(U ) is open by definition. Hence,
V = π(U ) = fˆ−1 (W ) is open, and fˆ is continuous.
Examples A.27. In these lecture notes quotients often appear as orbit spaces. The
starting point is a manifold M and a partition of M into orbits of some geometric
origin, for instance solution curves of a vector field, integral manifolds of a foliation
(cf. Section 9) or equivalence classes of a Lie group action (cf. Section C.5. Already
the case of orbits of a Hamiltonian system is interesting.
1. Let M be Rn with the equivalence relation: q ∼ q ′ (i.e. (q, q ′ ) ∈ R) if and only if
∥q∥ = ∥q ′ ∥ with respect to the euclidian norm. The equivalence classes are the spheres
Sn (r) = {q ∈ Rn | ∥q∥2 = r} , r ∈ [0, ∞[. In the case of n = 2 these equivalence classes
are the orbits of the Hamiltonian vector field H(q, p) = 12 (p2 + q 2 ). The quotient M/R
is the collection of spheres {Sn (r) | r ∈ [0, ∞[} with a natural bijection

h : M/R → [0, ∞[ , Sn (r) 7→ r .

The quotient topology on M/R is the topology, which makes h to a homeomorphism,

with respect to the standard topology on the interval [0, ∞[: h−1 (W ) is open for an
open subset W , W ⊂ [0, ∞[ , since
[
{Sn (r) | r ∈ W } = π −1 (h−1 (W ))

is open in Rn . h(U ) is open in the interval for an open subset U in M/R since π −1 (U )
is open in Rn and therefore π −1 (U ) ∩ ([0, ∞[ × {0}) is open in [0, ∞[ × {0} ∼
= [0, ∞[.
In particular, the quotient topology is Hausdorff.
This simple example shows that it is helpful to call p := h◦π : M → [0, ∞[ quotient
as well.
2. Let M = R2 with the differential equation (q̇, ṗ) = (p, 0). The solutions (i.e. the
motions) are q(t) = p0 t + q0 , p(t) = p0 , t ∈ R. The set of all orbits O is

O = {{(t, y) | t ∈ R} | y ∈ R , y ̸= 0} ∪ {{(x, 0)} | x ∈ R} .

272 A. Manifolds

Thus the orbits are the vertical lines through (0, p), p ̸= 0 and the points (x, 0), xR. The
orbits determine an equivalence relation R with the orbits as the equivalence classes.
A natural way to parametrize the orbit space O = M/R seems to be the use of the
axes of coordinates A := {(x, y) ∈ R2 | x = 0 or y = 0} by h : O → A , {(x, 0)} 7→
(x, 0) , {(t, y) | t ∈ R}, (y ̸= 0) 7→ (0, y) . In this way one is tempted to take A with
the topology induced from the inclusion A ⊂ R2 as the topological quotient.
But p =:= h ◦ π : M → A is not continuous. The inverse image −1 (V ) of the
open subset ]r, r + 1[ × {0} ⊂ A is the set ]r, r + 1[ × {0} ⊂ R2 which is not open as a
subset of R2 . The quotient topology on O = M/R is strictly weaker than the topology
induced on A from R2 .
In particular, the quotient topology is not Hausdorff: Any open neighbourhood V
of a point {(x, 0)} ∈ O contains an open neighbourhood of the form Vr (x) := π(Ur (x))
where Ur (x) = ]x − r, x + r[ × ]−r, +r[ for a suitable r > 0. This neighbourhood
of {(x, 0)} is Vr (x) = {{(x′ , 0)} | x′ ∈ ]x − r, x + r[ \ {0}} ∪ {{(t, y) | y ̸= 0} | y ∈
]−r, +r[}. For a different point {(x′ , 0)} ∈ O any neighbourhood V ′ of {(x′ , 0)} contains
an open Vr′ (x′ ). Let r ≤ r′ . Then Vr (x) ∩ Vr (x′ ) ̸= ∅ since this intersection contains
{0} × ]−r, +r[ \ {(0, 0)}. Hence, V ∩ V ′ ̸= ∅, i.e. every pair of neighbourhoods V of
{(x, 0)} resp. V ′ of {(x′ , 0)} has a nonempty intersection V ∩ V ′ ̸= ∅.
3. Projective Space: Let M be Kn+1 \ {0} and consider the equivalence relation
z ∼ w (i.e. (z, w) ∈ R) if and only if there is λ ∈ K with z = λw. The quotient
M/R =: Pn (K) is the set of lines in Kn+1 through 0. Here, K ∈ {C, R}98 . Let
γ : M → Pn (K) be the natural projection. The homogeneous coordinates for z =
(z 0 , z 1 . . . , z n ) ∈ M are [z 0 : z 1 : . . . : z n ]. They describe the equivalence classes and γ
completely: γ(z) = [z 0 : z 1 : . . . : z n ].
The quotient topology on Pn (K) can be described as follows: For each j ∈
{0, 1, . . . , n} let Uj := {[z 0 : z 1 : . . . : z n ] | z j ̸= 0} and Hj := {w ∈ Kn+1 | wj = 1}.
The Uj cover Pn (K). Each Hj is a hypersurface in Kn+1 and also an n-dimensional
affine subspace of Kn+1 . Hj is naturally isomorphic to Kn . Now,

1
φj : Uj → Hj , z 0 : z 1 : . . . : z n 7→ j z 0 , z 1 , . . . , z n
 

z
is a bijection. Each Uj will be endowed with the topology induced by φj . These
topologies determine a unique topology on all of Pn (K) which we call the ”chartwise”
topology. It consists of all unions of subsets of P(K) which are open in one of the Uj .
In particular, φj is a homeomorphism and can be called a topological chart.
We claim that this topology is the quotient topology. First of all, the projection γ
is continuous with respect to this topology.
Define Mj by Mj := γ −1 (Uj ) = Kn+1 \{z j ̸= 0}. γ is continuous, since all restrictions
γ|Mj : Mj → Uj are continuous which is a consequence of the fact that the ”projections”
98
or other topological fields
A.4 Quotient Manifold 273

pj := φj ◦ γ|Mj : Mj → Hj given by

1
pj (z) = z
zj
are obviously continuous maps.

M ⊃ Mj
pj
γ|Mj
 (
Pn (K) ⊃ Uj / Hj ∼
= Kn
φj

Moreover, γ is also open in the chartwise topology. For an open subset U ⊂ M and
j = 0, 1, . . . , n the intersections U ∩ Hj are open in Hj , hence γ(U ) ∩ Uj = φ−1 (U ∩ Hj
is open for all j. It follows that

γ(U ) = ∪{γ(U ) ∩ Uj | j = 0, 1, . . . , n}

is open in the chartwise defined topology.

By the universal property it follows that the quotient topology is finer than the
chartwise defined topology. It is also coarser: Let V ⊂ Pn (K) be open in the quotient
topology. Then γ −1 (V ) is open, and by the openness of the projection γ with respect to
the chartwise topology, V is open in the chartwise topology. Hence the two topologies
coincide.
Finally, it is not difficult to prove that Pn (K) is Hausdorff. Given two different
points [z1 ] ̸= [z2 ] in Pn (K) by homogeneous coordinates we can assume that ∥z1 ∥ =
∥z2 ∥ = 1. In case of K = C the two compact subsets Sk := {exp itzk | t ∈ R} , k = 1, 2 ,
of Cn+1 have no point in common. In case of K = R or K = Q the same is true for
Sk := {zk , −zk }. Hence, there is a positive r > 0 suchSthat ∥w1 − w2 ∥ ≥ 3r for all
(w1 , w2 ) ∈ S1 × S2 . Now, the union of open balls Uk := {B(wk , r) | wk ∈ Sk } is open
(k = 1, 2) and therefore, Vk := γ(Uk ) is an open neighbourhood [zk ]. By construction
V1 ∩ V2 = ∅. Otherwise, there would exist a point [w] ∈ V1 ∩ V2 with w ∈ Uk , k = 1, 2.
Since w ∈ B(wk , r) for suitable wk ∈ Sk the distance between w1 and w2 would be
∥w1 − w2 ∥ ≤ ∥w1 − w∥ + ∥w − w2 ∥ < 2r contradicting ∥w1 − w2 ∥ ≥ 3r.
Finally, Pn (K) is compact, since it is the image of the restriction of γ to the compact
sphere Sn (K = R) resp. (K = C).
4. One can generalize the last result to ℓ2 or KN instead of Kn+1 and to other
(infinite dimensional) sequence spaces by essentially the same arguments. If S is the
sequence space99 there are the natural coordinates z = (z j )j∈N of S and M = S \ {0}
as in t he finite dimensional case. The equivalence relation is the same with the lines
through 0 as the equivalence classes and the natural projection γ : M → P(S). We
99
S is a locally convex space with a Schauder basis
274 A. Manifolds

obtain infinitely many coordinate neighbourhoods Uj , j ∈ N, covering P(S) with the

charts
φj : Uj → Hj = {z ∈ M | z j = 1} ∼
=S.
Moreover, γ −1 (Uj ) = Mj projects to Hj by the continuous maps pj (z) = z
zj
with
pj = φj ◦ γ|Mj .
M ⊃ Mj
pj
γ|Mj
 '
P(S) ⊃ Uj / Hj ∼
=S
φj

Consequently the chartwise topology is the quotient topology. Moreover the quo-
tient topology is Hausdorff. However it is not compact when S is infinite dimensional.

The final arguments in the third example above works in general and leads to a
general criterium for a continuous map being a quotient. We have explained before
that in many cases one is interested to know whether a given continuous and surjective
map g : M → X is a quotient, i.e. X carries the quotient topology.

Proposition A.28. A continuous and surjective map g : M → X is a quotient

1. whenever g is an open map, or

2. whenever g has local continuous sections, i.e. for all x ∈ X there exists an open
neighbourhood V and a continuous s : V → M with g ◦ s = idV .

Proof. Let f : M → Y continuous and constant on the equivalence classes, that means
on the fibres f −1 (y) , y ∈ Y . We have to show that fˆ : X → Y is continuous. For an
open W ⊂ Y , f −1 (W ) is open in M hence , if g is open, g(f −1 (W )) = fˆ−1 (W ) is open,
i.e. fˆ is continuous. If f has local continuous sections s : V → M , the composition
f ◦ s = fˆ|V : V → Y is continuous, and it follows that fˆ is continuous.

A.4.2 Differentiable Quotients

Similarly, for the differentiable case. Note, that the notion of differential quotient is
analogous to that of a topological quotient, see Definition A.10:

Proposition A.29. A differentiable and surjective map g : M → X for differentiable

manifolds M, X is a differentiable quotient

1. whenever g has local differentiable sections, i.e. for all x ∈ X there exists an open
neighbourhood V and a continuous s : V → M with g ◦ s = idV , or

2. whenever f is a submersion, i.e. the derivative Ta f is surjective for all a ∈ M .

A.5 Operations on Differential Forms 275

Proof. 1. has the same proof as in the last proposition. And a submersion always has
local differentiable sections according to the implicit mapping theorem.
Examples A.30. We investigate the 4 examples in A.27:
1. We know that h ◦ π : M = Rn → [0, ∞[ is a topological quotient and the
quotient is Hausdorff. [0, ∞[ is not a manifold in any sense, but it is a manifold
with boundary {0}. Deleting 0 in M and in [0, ∞[ we obtain a differentiable map
g : M \ {0} → ]0, ∞[. It is the differentiable quotient since g has differentiable local
sections and is a submersion.
2. The quotient O in example 2. above is not Hausdorff, so there is no chance
that the differentiable quotient exists. Note, that the defining differential equation
(q̇, ṗ) = (p, 0) is the equation of motion of a Hamiltonian system: The Hamiltonian
vector field is XH (q, p) = (p, 0) where H(q, p) = 21 p2 . XH does not define a distribution,
since X( H)(q, 0) = 0.
3. The natural projection γ : Kn+1 \ {0} → Pn (K) is the topological quotient
according to the third axample above in A.27, and the quotient is Hausdorff. The
projective space Pn (K) obtains a differentiable structure by the charts φ : Uj → Hj
which are (smoothly) compatible to each other. The projection γ is differentiable with
respect to this differentiable structure: Since γ is a submersion, γ : Kn+1 \{0} → Pn (K)
is the differentiable quotient.
In the case of K = C γ : Kn+1 \{0} → Pn (K) is, moreover, the holomorphic quotient.
In particular the charts φj are biholomorphic and Pn (C) is a complex manifold (see
Chapter ”Complex Analysis” B).
4. For sequence spaces like ℓ2 or KN we obtain in the same manner differentiable
resp. holomorphic quotients. However, we need the notion of an infinite dimensional
manifold.
Example A.31. Here is an example of a surjective smooth mapping which is not a
submersion: f : R → R , f (x) = x3 . f is not a submersion, since f ′ (x) = 3x2 is zero
at x = 0. Note, that f is an open mapping. Hence, it is a quotient mapping in the
topological category, but it is not a differentiable quotient.
In a later section on Lie groups actions and explain in Theorem C.25 the following
general result on quotients arising from suitable Lie group action on manifolds:
Proposition A.32. Suppose G is a Lie group acting smoothly, freely and properly
on a manifold M . Then the orbit space M/G is a Hausdorff space, and exists as
differential quotient manifold of dimension equal to dim M − dim G. The quotient map
π : M → M/G is a submersion.

A.5 Operations on Differential Forms

We introduce the main operations on forms in order to describe the interplay between
the Lie derivative, the exterior derivative and the interior derivative on differential
276 A. Manifolds

forms and prove the formula

s
X
dη(X0 , X1 , ..., Xs ) = (−1)j LXj η(X0 , ..., X̂j , ..., Xs )
j=0
X (70)
i+j
+ (−1) η([Xi , Xj ], X0 , ..., X̂i , ..., X̂j , ..., Xs )
0≤i<j≤s

which is used as the definition of the exterior derivative d in the first chapter, cf. (5).100
Pullback

Definition A.33. Given a smooth map F : N → M between manifolds N and M ,

the Pullback
F ∗ : Ak (M ) → Ak (N )
is defined as follows: For η ∈ Ak (M ) and Yj ∈ V(N ) , j = 1, . . . , k, we set:

F ∗ η(Y1 , . . . Yk ) := η (T F (Y1 ), . . . , T F (Yk )) = η(F∗ (X1 ), . . . , F∗ (Xk ) ,

where T F : T N → T M is the derivative of F (see Observation A.14)101 and F∗ (Xj ) :=

T F (Xj ). Pointwise we have for a ∈ N

F ∗ η(Y1 , . . . Yk )(a) = η(F (a)) (Ta F (Y1 ), . . . , Ta F (Yk )) .

Since T F : V(M ) → V(N ) induces a map Λk T F : Ak (M ) → Ak (N ) the pullback F ∗ η

is the same as η ◦ Λk T F .

Exterior Product
The wedge product ∧ : Ak (M ) × Am (M ) → Ak+m (M ) is given as in multilinear
algebra. The wedge product endows
∞
M
◦
A (M ) := Ak (M )
i=0

with the structure of an algebra that is Graded Commutative,

i.e. for α ∈ Ap (M ), β ∈ Aq (M ) we have

α ∧ β = (−1)|α||β| β ∧ α
where |α| = p , |β| = q, denote the respective degrees of α and β.
100
The essential part of this section is taken from an exercise of M. Stankiewicz
101
Exercise: Describe F ∗ η in local coordinates.
A.5 Operations on Differential Forms 277

In physics, especially in the context of supersymmetry, this such a graded commu-

tative algebra is often refereed to as a superalgebra.
Lie Derivative
Given a vector field X with the corresponding flow Φt (c.f. Proposition A.20), the
Lie derivative along X of a differential form is a K-linear map LX : Ap (M ) −→ Ap (M )
defined by
d
LX ω = (Φ∗t ω)|t=0 . (71)
dt
For functions g ∈ E(M ) this reduces to the known directional derivative LX g = Xg =
dg(X). Moreover, since the wedge product is natural with respect to pullbacks, we
obtain

LX (α ∧ β) = LX α ∧ β + α ∧ LX β (72)

Interior Derivative
The interior derivative iX along a vector field X is the E(M )-linear map

iX : Ap (M ) −→ Ap−1 (M )

defined by inserting X into the first argument of the form, i.e.

(iX ω)(Y1 , ..., Yp−1 ) = ω(X, Y1 , ..., Yp−1 )

The interior derivative of a function is defined to be zero.

The interior derivative is sometimes called interior product; but here we are inter-
ested in the property that iX is a derivation with respect to the graded commutative
algebra structure of A◦ (M ).
Since differential forms are alternating it follows that
Observation A.34.
iX iY + iX iY = 0. (73)
Moreover iX satisfies a graded version of the Leibniz rule

iX (α ∧ β) = iX α ∧ β + (−1)|α| α ∧ iX β. (74)

which follows from the definition of the wedge product.

Exterior Derivative
The exterior derivative can be defined axiomatically as a K-linear map

d : Ap (M ) −→ Ap+1 (M )

that satisfies the following properties:

278 A. Manifolds

1◦ For any smooth function f , df is the differential of f .

2◦ d2 = 0.
3◦
d(α ∧ β) = dα ∧ β + (−1)|α| (α ∧ dβ). (75)

Equivalently d can be defined locally as in Equation (8) Chapter 1.

An important property of d is that it respects pullbacks, i.e. for any smooth map
F : M −→ N and α ∈ A◦ (N ) we have d(F ∗ α) = F ∗ (dα).
Superalgebra of Derivations
Although iX , LX , d are all defined in quite a different way, one can actually view
them as three instances of the same kind of object, namely a derivation on superalgebra.
Definition (Derivation). We say that a linear map δ : A◦ (M ) −→ A◦ (M ) is a Deriva-
tion of degree |δ| if for any α, β ∈ A◦ (M ) it satisfies the graded Leibniz rule
δ(α ∧ β) = δα ∧ β + (−1)|α||δ| α ∧ δβ
and δα ∈ A|α|+|δ| (M ).

From equations (72),(74),(75), one sees that LX , iX , d are derivations of degrees

respectively 0, −1 and +1.
Definition. The Commutator of two derivations δ1 , δ2 on A◦ (M ) is defined as
[δ1 , δ2 ] = δ1 δ2 − (−1)|δ1 ||δ2 | δ2 δ1
Proposition. If δ1 , δ2 are derivations on A∗ (M ) then [δ1 , δ2 ] is a derivation of degree
|δ1 | + |δ2 |.

Proof. Clearly, for any α ∈ Ap (M ), we have that [δ1 , δ2 ]α ∈ Ap+|δ1 |+|δ2 | . Therefore it
suffices to check that the graded Leibniz rule holds. This follows from direct computa-
tion.
δ1 δ2 (α ∧ β) = δ1 (δ2 α ∧ β + (−1)|α||δ2 | α ∧ δ2 β)
= δ1 δ2 α ∧ β + (−1)|δ1 |(|α|+|δ2 |) δ2 α ∧ δ1 β
+ (−1)|δ2 ||α| δ1 α ∧ δ2 β + (−1)|α|(|δ1 |+|δ2 |) α ∧ δ1 δ2 β
(−1)|δ1 ||δ2 | δ2 δ1 (α ∧ β) = (−1)|δ1 ||δ2 | δ2 δ1 α ∧ β + (−1)|α||δ2 |+2|δ1 ||δ2 | δ1 α ∧ δ2 β
+ (−1)|δ1 ||α|+|δ1 ||δ2 | δ2 α ∧ δ1 β
+ (−1)|α|(|δ2 |+|δ1 |)+|δ1 ||δ2 | α ∧ δ2 δ1 β
In the commutator the mixed terms cancel leaving
[δ1 , δ2 ](α ∧ β) = (δ1 δ2 − (−1)|δ1 ||δ2 | δ2 δ1 )α ∧ β
+ (−1)|α|(|δ1 |+|δ2 |) α ∧ (δ1 δ2 − (−1)|δ1 ||δ2 | δ2 δ1 )β
as required.
A.5 Operations on Differential Forms 279

By considering the degrees of iX , LX and d, the above proposition and the fact that
[iX , iY ] and [d, d] both vanish one may suspect that the algebra generated by these
derivations should close. And indeed, iX , LX and d satisfy the commutation relations
as stated below, thereby forming, what is sometimes called a Lie superalgebra.
Theorem A.35. The following relations hold on A◦ (M )

[iX , iY ] = [iX , LY ] = [iX , d] =

= iX iY + iY iX = 0 = iX LY − LY iX = i[X,Y ] = iX d + diX = LX

[LX , LY ] = [LX , d] =
= LX LY − LY LX = L[X,Y ] = LX d − dLX = 0

[d, d] =
= d2 + d2 = 0

Proof. The cases of [iX , iY ] and [d, d] have already been considered. Also [LX , d] = 0
clearly holds, because the exterior derivative respects pullback.
To prove the remaining three relations let us notice that it suffices to do so locally.
In particular one only needs to check them for the case of functions and exact 1-forms
because of the coordinate expression (8) and fact that we are comparing derivations of
equal degree.
Let us start with LX = [iX , d] also known as Cartan’s Magic Formula which
has the dorm
LX η = diX η + iX dη (76)
for a differential form η. For a function f ∈ E(M ) we have
[iX , d]f = iX df = X(f ) = LX f
since iX f = 0. Using the above and the fact that [LX , d] = 0 we also get
[iX , d]df = diX df = dLX f = LX df.

In the similar way we prove the remaining two relations:

[iX , LY ]f = iX (LY f ) = 0 = i[X,Y ] f
[iX , LY ]df = iX LY df − LY iX df
= iX d(LY f ) − LY (X(f )) = iX d(Y (f )) − Y (X(f ))
= X(Y (f )) − Y (X(f )) = [X, Y ](f )
= i[X,Y ] df.
280 A. Manifolds

[LX , LY ]f = LX (LY f ) − LY (LX f )

= X(Y (f )) − Y (X(f )) = [X, Y ](f )
= L[X,Y ] f
[LX , LY ]df = LX (LY (df )) − LY (LX (df ))
= d(LX (LY f )) − d(LY (LX f ))
= d(L[X,Y ] )f
= L[X,Y ] (df )
This finishes the proof.

These commutation relations are quite useful, in particular they allow us to prove
the formula used to define the exterior derivative in Chapter 1:
Proof of formula (70): The proof proceeds by induction on the degree p of the
differential form involved.

1) p=0: For a function, the commutator term of (70) is simply zero and we are left
with
df (X) = X(f ) = LX (f )

2) Induction step: Suppose the formula holds for all differential forms of degree at
most p − 1 and let η ∈ Ap (M ). We can then write

dη(X0 , ..., Xp ) = iXp ...iX1 iX0 (dη)

= −iXp ...iX1 d(iX0 η) + iXp ...iX1 (LX0 η)
= −d(iX0 η)(X1 , ..., Xp ) + iXp ...iX1 (LX0 η).

Now, since iX0 η is a (p − 1)-form we can apply our formula which will give us
p
X
=− (−1)j−1 LXj ((iX0 η)(X1 , ..., X̂j , ..., Xp ))
j=1
X
− (−1)i+j−2 (iX0 η)([Xi , Xj ], X1 , ..., X̂i , ..., X̂j , ..., Xp ) + iXp ...iX1 (LX0 η)
1≤i<j≤p
p
X
= (−1)j LXj (η(X0 , X1 , ..., X̂j , ..., Xp ))
j=1
X
+ (−1)i+j η([Xi , Xj ], X0 , X1 , ..., X̂i , ..., X̂j , ..., Xp ) + iXp ...iX1 (LX0 η)
1≤i<j≤p

where the negative sign in the commutator term vanishes because we exchange X0 and
[Xi , Xj ] in the arguments of η. We can now use [iX , LY ] = i[X,Y ] to commute the Lie
A.5 Operations on Differential Forms 281

derivative in the last term all the way to the left yielding
p
X
= (−1)j LXj (η(X0 , X1 , ..., X̂j , ..., Xp ))
j=1
X
+ (−1)i+j η([Xi , Xj ], X0 , X1 , ..., X̂i , ..., X̂j , ..., Xp )
1≤i<j≤p
p
X
+ LX0 (η(X1 , ..., Xp )) + iXp ...i[Xj ,X0 ] ...iX1 η.
j=1

The second last term can be then absorbed into the first sum and in the last term we
move i[Xj ,X0 ] all the way to the right with j − 1 swaps gaining a factor of (−1) for each
of them and finally we get
p
X
= (−1)j LXj (η(X0 , X1 , ..., X̂j , ..., Xp ))
j=0
X
+ (−1)i+j η([Xi , Xj ], X0 , X1 , ..., X̂i , ..., X̂j , ..., Xp )
1≤i<j≤p
p
X
+ (−1)j−1 η([Xj , X0 ], X1 , ..., X̂j , ..., Xp )
j=1
p
X
= (−1)j LXj (η(X0 , X1 , ..., X̂j , ..., Xp ))
j=0
X
+ (−1)i+j η([Xi , Xj ], X0 , X1 , ..., X̂i , ..., X̂j , ..., Xp )
1≤i<j≤p
p
X
+ (−1)j η([X0 , Xj ], X1 , ..., X̂j , ..., Xp )
j=1
p
X
= (−1)j LXj (η(X0 , X1 , ..., X̂j , ..., Xp ))
j=0
X
+ (−1)i+j η([Xi , Xj ], X0 , X1 , ..., X̂i , ..., X̂j , ..., Xp )
0≤i<j≤p

which is the desired formula. □

282 B. Complex Analysis

B Complex Analysis

Elementary properties of holomorphic functions in one variable are assumed to be

known in the following. General reference for the theory of holomorphic functions in
several vasriables, i.e. in Complex Analysis is the book of D. Huybrechts [Huy05].

B.1 Cauchy Integral

A domain is a connected open subset U ⊂ Cn in Cn . A polydisc D = D(r) of polyradius

r = (r1 , . . . , rn ) ∈ Rn+ in Cn is a product of discs, it has the form D(r) := D1 ×. . .×Dn ⊂
Cn , where Dj = {zj | |zj | < rj } = D(rj ) is the usual disc in C.

Definition B.1. A Holomorphic Function on a domain U ⊂ Cn is a function

f : U → C, which is partially holomorphic in the following sense: For each a ∈ U and
for each j = 1, . . . , n the one variable function z 7→ f (a + zej ) is holomorphic in a
neighbourhood of 0 ∈ C.

F. Hartogs102 proved about 100 years ago (published 1906 in Math. Ann.) that
a holomorphic function is already continuous. This result is difficult to prove and it
seems that it has no relevant implications for the theory of holomorphic functions in
several variables. Consequently, one mostly works with the definition that a function is
holomorphic if it is continuous and partially holomorphic. From now on, we assume a
holomorphic function to be continuous. Let O(U ) be the space of holomorphic functions
in this sense. By pointwise operations, O(U ) is in a natural way an algebra over C.
We can apply the 1-dimensional Cauchy Integral to obtain:

Proposition B.2 (Cauchy Integral). Let f ∈ O(U ), and assume 0 ∈ U with D ⊂ U

for a polydisc D = D(r) = D1 × . . . × Dn .

• Then for zj ∈ Dj the Cauchy integral representation

Z Z
1 f (ζ1 , . . . , ζn )
f (z1 , . . . , zn ) = ··· dζ1 · · · dζn (77)
(2πi)n ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )

holds true.

• The partial derivatives have the following description

∂ k1 +···+knf (z1 , . . . , zn ) k1 ! · · · kn !
Z Z
f (ζ1 , . . . , ζn )
k k
= n
··· k1 +1 · · · (ζ − z )kn +1
dζ1 · · · dζn .
∂z1 · · · ∂zn
1 n (2πi) ∂D1 ∂Dn (ζ1 − z1 ) n n
102
F. Hartogs was university professor at the LMU in Munich
B.1 Cauchy Integral 283

Proof. The integrals are interated integrals in one variable, and they do not depend on
the order of evaluating the single integrals since the integrand is continuous.
The first formula can be proven by induction: The result is known for n = 1. Let
n > 1. For fixed zn ∈ Dn , by the induction hypothesis we can assume
Z Z
1 f (ζ1 , . . . , ζn−1 , zn )
f (z1 , . . . , zn ) = n−1
··· dζ1 · · · dζn−1 (78)
(2πi) ∂D1 ∂Dn−1 (ζ1 − z1 ) · · · (ζn−1 − zn−1 )

For fixed (ζ1 , . . . , ζn−1 ) ∈ ∂D1 × · · · × ∂Dn−1 the 1-dimensional Cauchy integral is
Z
1 f (ζ1 , . . . , ζn )
f (ζ1 , . . . , ζn−1 , zn ) = dζn .
2πi ∂Dn (ζn − zn )

Inserting this equality in the formula (78) we receive the result.

Since the integrand is continuous we can exchange integration and differentiation
to obtain the second result.

The Cauchy integral representation of holomorphic functions exhibits a special kind

of mean value property. The value f (0) of f ∈ O(U ) is the average of the values on
the product ∂D1 × . . . × ∂Dn (if D ⊂ U ): For z = 0 and with respect to new variables
ζj = rj eiθj , dζj = ireiθj dθj , we obtain
 n Z 2π Z 2π
1
f (0) = ... f (r1 eiθ1 , . . . , rn eiθn ) dθ1 . . . dθn . (79)
2π 0 0

This result leads to

Proposition B.3 (Mean Value). For a holomorphic function f ∈ O(U ) and w ∈ U

with w + D(r) ⊂ U the value f (w) is the average of the values of f at w + D(r):
Z
1 1
f (w) = 2 . . . 2 f (z)dz ,
πr1 πrn w+D(r)

where dz is Lebesgue integration over Cn in this situation.

R
Proof. We set w = 0 without loss of generality. The integral D(r) f (z)dz is, with
respect to new variables ρ, θ and dzj = ρj dρj dθj ,
Z 2π Z 2π Z r1 Z rn
... ... f (ρ1 eiθ1 , . . . , ρn eiθn ) dθ1 . . . dθn ρ1 dρ1 . . . ρn dρn .
0 0 0 0

Inserting (79) yields

 n Z Z r1 Z rn
1
f (z)dz = ... f (0) ρ1 dρ1 . . . ρn dρn .
2π D(r) 0 0
284 B. Complex Analysis

The result follows from

Z r1 Z rn
1 1
... ρ1 dρ1 . . . ρn dρn = r12 . . . rn2 .
0 0 2 2

The Cauchy integral can be used to define holomorphic functions by integrals:

Proposition B.4. For continuous g : ∂D1 × . . . × ∂Dn → C and (z1 , . . . , zn ) ∈ D1 ×

. . . × Dn the integral
Z Z
1 g(ζ1 , . . . , ζn )
f (z1 , . . . , zn ) = n
··· dζ1 · · · dζn
(2πi) ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )

defines a holomorphic function f : D1 × . . . × Dn → C.

Proof. f is certainly continuous (the integrand is continuous in the variables ζj and zj ).

Again by interchanging integration and differentiation one deduces that the function
is partially holomorphic.

B.2 Power Series

Proposition B.5 (Power Series Development). Let f ∈ O(U ). For each a ∈ U there
exists a sequence of homogeneous polynomials P k = P k f (a) ∈ C(k) [z1 , . . . , zn ] such that
for every polydisc D such that a + D ⊂ U
X
f (a + z) = P k (z)
k∈N

for all z = (z1 , . . . , zn ) ∈ D. The convergence is absolute and uniform on the polydisc
D.

Proof. We may assume a = 0 and use the formula (77): For ζj ∈ ∂Dj and zj ∈ Dj (i.e.
|zj | < |ζj | = rj ) the fraction

1
has a series development
(ζ1 − z1 ) · · · (ζn − zn )

1 X z1j1 · · · znjn
= ,
(ζ1 − z1 ) · · · (ζn − zn ) j ,...,j ∈N ζ1j1 +1 · · · ζnjn +1
1 n

which converges uniformly on compact subsets K contained in the polydisc D = D1 ×

. . . × Dn . Therefore, summation and integration can be interchanged to yield, using
B.2 Power Series 285

the abbreviation f (ζ)dζ := f (ζ1 , . . . , ζn )dζ1 · · · dζn :

z1j1 · · · znjn
Z Z Z Z
f (ζ)dζ X
··· = ··· j1 +1 jn +1 f (ζ)dζ
∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn ) ∂D1 ∂Dn j ,...,j ∈N ζ1 · · · ζn
1 n

X Z z1j1 · · · znjn
Z
= ··· f (ζ) j1 +1 dζ
j1 ,...,jn ∈N ∂D1 ∂Dn ζ1 · · · ζnjn +1
X Z Z
f (ζ)dζ

= ··· j1 +1 jn +1 z1j1 · · · znjn
j1 ,...,jn ∈N ∂D1 ∂Dn ζ1 · · · ζn

Let Z Z
1 f (ζ)
cj1 ,...,jn := ··· dζ
(2πi)n ∂D1 ∂Dn ζ1j1 +1 · · · ζnjn +1
be the coefficients in the last formula. Using (77) we obtain
Z Z
1 f (ζ)dζ
f (z) = n
···
(2πi) ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )
X
= cj1 ,...,jn z1j1 · · · znjn
j1 ,...,jn
∞
X X
= cj1 ,...,jn z1j1 · · · znjn
k=0 j1 +...+jn =k
X∞
k
= P (z) ,
k=0

if P k (z) := j1 +...jn =k cj1 ,...,jn z1j1 · · · znjn . P k is a homogeneous polynomials of degree k,

P
which proves the proposition.

From the proof one can deduce the following results:

Corollary B.6. The coefficients of the homogeneous polynomials P k f (a) are suitable
sums of higher partial derivatives according to Proposition B.2. In fact, the P k f are of
the form
X 1 ∂ j1 +···+jn f (a) j1
P k f (a)(z) = z · · · znjn .
j1 · · · ∂z jn 1
(80)
j +...+j =k
j1 ! · · · jn ! ∂z 1 n
1 n

Hence the power series development is the Taylor expansion.

Furthermore, the mappings P k f : U → C(k) [z1 , . . . , zn ] are holomorphic.

The following result is in sharp contrast to the theory of smooth functions:

286 B. Complex Analysis

Proposition B.7 (Identity Theorem). Let f, g ∈ O(U ) holomorphic functions on the

domain U ⊂ Cn .
1. If P k f (a) = P k g(a) holds for all k ∈ N at one point a ∈ U the two functions
agree in all of U .
2. If f |V = g|V for a nonempty open subset V ⊂ U the two functions agree in all
of U .

Proof. 1. If P k f (b) = P k g(b) holds for all k ∈ N at b ∈ U the two functions agree
in a neighbourhood of b according to Proposition B.5. Hence the set W := {b ∈ U |
P k f (b) = P k g(b) for all k ∈ N} is open and W T ̸= ∅ because of a ∈ W . By the
continuity of all P f, P g) the intersection W = k∈N {b ∈ U | P k f (b) = P k g(b)}
k k

is closed as well. Therefore, W = U because U is connected (we assume U to be a

domain).
2. The assumption implies P k f (a) = P k g(a) for all k ∈ N and for any a ∈ V . The
result follows from 1.

Corollary B.8. O(U ) is an integral domain (i.e. has no zero divisors).

Proposition B.9 (Open Mapping). Every nonconstant f ∈ O(U ) is an open mapping.

Proof. We have to show that f (W ) is open in C for any open subset W ⊂ U . Let
a ∈ W and let D be a polydisc with V := a + D ⊂ W . There exists b ∈ V with
f (a) ̸= f (b) according to Proposition B.7. The function h(ζ) := f ((1 − ζ)a + ζb) is
holomorphic in the unit disc ∆ = {ζ | |ζ| < 1} and not constant. Therefore, h is open
as a holomorphic function in one variable, and h(∆) is an open neighbourhood of f (a).
As a result, f (V ) is a neighbourhood of f (a) contained in f (W ) which proves that
f (W ) is open.

Proposition B.10 (Maximum principle). If a holomorphic f ∈ O(U ) attains its max-

imum in a point a ∈ U , i.e. |f (a)| = max{|f (z)| , z ∈ U } , then f is constant.

Proof. Otherwise, f would be open, in particular f (U ) would be an open subset of C.

But for an open W ⊂ C the maximum of the |z| , z ∈ W , is not attained: In every
neighbourhood V of a point z ∈ C there e exists a point w ∈ V with |z| < |w|.

B.3 Hartogs’ Extension Theorem

The extension theorem considers certain configurations of two open subsets V, W ⊂

Cn , V ⊂ W and V ̸= W , such that every holomorphic function f : V → C has
a holomorphic continuation f˜ : W → C: f˜ is holomorphic and f˜|V = f . As a
consequence, the restriction mapping O(W ) → O(V ) , g 7→ g|V is an isomorphism of
vector spaces. This property of holomorphic functions in more than one variable is in
B.3 Hartogs’ Extension Theorem 287

strong contrast to smooth functions on open subsets of Rn as well as with holomorphic

functions in one variable.
We prove a special case of the extension theorem of Hartogs. Let ∥ ∥ be a norm
on Cn and B(a, r) := {z ∈ Cn | ∥a − z∥ < r} the open ball of radius r around a with
respect to this norm.

Proposition B.11 (Kugelsatz). Let U be domain such that B(a, R) \ B(a, r) ⊂ U

for some a ∈ U and for R, r with 0 < r < R. Then every holomorphic function
f ∈ O(U ) has a unique Holomorphic Continuation103 to U ∪ B(a, R), i.e. there
is a holomorphic f˜ ∈ O(U ∪ B(a, R)) such that f˜|U = f .

Proof. We prove the statement for the supnorm and for n = 2 for simplicity. We can
assume a = 0. Let (z1 , z2 ) ∈ B(0, R) \ B(0, r) with r < |zj | < ρ < R , j = 1, 2. This
configuration is illustrated in the following sketch of the absolute space:

|z2 | B(0, R) \ B(0, r)

0<r<ρ<R
R
ρ
(|z1 |, |z2 |)
r

r ρ R |z1 |

For fixed z1 and varying z2 the Cauchy integral in one variable yields
Z
1 f (z1 , ζ2 )
f (z1 , z2 ) = dζ2 ,
2πi ∂D2 (ρ) ζ2 − z2

where D2 (ρ) is the open disc of radius ρ. For each fixed ζ2 ∈ ∂D2 (ρ) we obtain in the
same way Z
1 f (ζ1 , ζ2 )
f (z1 , ζ2 ) = dζ1 .
2πi ∂D1 (ρ) ζ1 − z1
Inserting one formula in the other yields
Z Z
1 f (ζ1 , ζ2 )
f (z1 , z2 ) = dζ1 dζ2 . (81)
(2πi)2 ∂D1 (ρ) ∂D2 (ρ) (ζ1 − z1 )(ζ2 − z2 )
103
Also called analytic continuation.
288 B. Complex Analysis

This is again the Cauchy integral as presented above. The clou is, that the integral
gives sense not only for the z = (z1 , z2 ) with r < |zj | < ρ but for all z = (z1 , z2 ) , |zj | <
ρ, i.e. for all z ∈ B(0, ρ). Therefore, (81) defines a holomorphic function f˜(z) for
z ∈ B(0, ρ) by Proposition B.4, i.e. f˜ ∈ O(B(0, ρ)) which agrees with f on U ∩ B(0, ρ)
and thus defines a holomorphic continuation to U ∪ B(0, R).

Corollary B.12. A holomorphic function in 2 or more variables can have no isolated

singularity, i.e. if a ∈ U is a point in the domain U and f is holomorphic in U \ {a},
then f can be continued holomorphically to all of U .

B.4 Sequences of Holomorphic Functions

The convergence behaviour of sequences of holomorphic functions in several variables

is in many aspects the same as for holomorphic functions in one variable. The following
is an important result which is easy to prove using the Cauchy integral representation.
Note, that a corresponding result for smooth function does not hold.

Proposition B.13 (Weierstrass). Let (fk ) be a sequence of holomorphic functions

fk ∈ O(U ) which converges uniformly on compact subsets of U to a function f , Then
f is holomorphic.

Proof. Because of the uniform convergence on compacta the limit function is continu-
ous. Hence, for z = (z1 , . . . , zn ) ∈ U and D(r) ⊂ U
Z Z
1 fk (ζ1 , . . . , ζn )
f (z) = lim fk (z) = lim · · · dζ1 · · · dζn
(2πi)n ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )
Z Z
1 lim fk (ζ1 , . . . , ζn )
= n
··· dζ1 · · · dζn ,
(2πi) ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )

since (fk ) converges uniformly on ∂D1 × . . . × ∂Dn and, therefore, integration and limit
can be interchanged. As a consequence,
Z Z
1 f (ζ1 , . . . , ζn )
f (z) = n
··· dζ1 · · · dζn ,
(2πi) ∂D1 ∂Dn (ζ1 − z1 ) · · · (ζn − zn )

and f is holomorphic according to Proposition B.4.

Note, that all partial derivatives of fk converge – uniformly on compact subsets –

to the corresponding partial derivative of f . Likewise, P m fk converges to P m f .
We present an application of this result which is useful in the context of geometric
quantization of the phase space T ∗ Rn = Cn (simple phase space) with respect to
the holomorphic polarization. In this application a standard representation space of
B.4 Sequences of Holomorphic Functions 289

Quantum Mechanics is described, the Bargmann space, also called Fock space: It is
the space Z
n
F := {f ∈ O(C ) | |f (z)|2 exp(−πz z̄)dz < ∞} ,
Cn
Pn
where z z̄ := |z|2 = 1 zj zj and dz is Lebesgue integration over Cn . Let us denote
the density in the above integral by k(z) = exp(−πz z̄), and dµ(z) := k(z)dz the
corresponding measure on Cn .104 The Bargmann space F is a subspace of the Hilbert
space L2 (Cn , dµ) of functions on Cn which are square integrable with respect to dµ. In
this way, F is a prehilbert space and a normed space with the norm
sZ
∥f ∥ = ∥f ∥µ = |f (z)|2 dµ(z) .

Proposition B.14. The Bargmann space F is complete and, hence, a closed subspace
of the Hilbert space L2 (Cn , dµ). In particular, F is a Hilbert space. The Hilbert space F
will be denoted by HP with respect to the holomorphic polarization P in the context of
representation spaces determined by polarizations (e.g. in Example 10.9 and Example
10.13).

Proof. According to Proposition B.3, for every holomorphic function f ∈ O(Cn ) the
value f (w) at a point w ∈ Cn is the average of the values f (z), z ∈ w + D(r), where
D(r) is any polydisc. If we assume all radii rj to be equal, rj = ρ, we obtain
 n Z  n Z
1 1 1
f (w) = f (z)dz = f (z)k(z)dz .
πρ w+D(r) πρ w+D(r) k(z)

Applying the Cauchy-Schwarz inequality to the right hand side leads to

 n
1
|f (w)| ≤ sup k(z)−1 ∥χD ∥µ ∥f ∥µ ,
πρ z∈D

where the abbreviation D = w + D(r) is used and where χX is the indicator function
of X.
In particular, this inequality implies that the evaluation ŵ : F → C , f 7→ f (w), is
continuous.
The inequality can be extended to be uniform over compact subsets: Let K ⊂ Cn
be a compact subset. Then
[
L := w + D(r) = K + D(r)
w∈K

104
The result below is true for more general densities k.
290 B. Complex Analysis

is compact as well and contained in a big polydisc D(R) ⊂ Cn . For each w ∈ K the
inequality now reads.
 n
1
|f (w)| ≤ sup k(z)−1 ξD(R) µ ∥f ∥µ . (82)
πρ z∈D(R)

Now let (fk ) be a Cauchy

 nsequence in the Bargmann space F with respect to the
norm ∥ ∥µ . With C := πρ 1
supz∈D(R) k(z)−1 χD(R) µ we obtain, by (82), for each
w∈K
|fk (w) − fm (w)| ≤ C ∥fk − fm ∥µ
and conclude: (fk (w)) is a Cauchy sequence in C converging to a value f (w) defining
a function f : Cn → C. And the convergence fk → f is uniform on K. Hence, by
Proposition B.13, f is holomorphic. Moreover, the last inequality implies, that f is in
F and that fk → f in the norm ∥ ∥µ , i.e. fk → f in F ⊂ L2 (Cn , dµ).

To get to know more about the Bargmann space, we observe that ̸=O(Cn ). As an
example, the function
π
f (z) = exp( z12 )
2
is not contained in F since f (z)f (z) = exp(π(x21 − y12 )) and hence
Z Z Z
¯
f f dµ(z) = 2
exp(−2πy1 )dz1 exp(−πz ′ z ′ )dz ′ = ∞,
C Cn−1

where x1 = Rez1 , y1 = Imz1 and z ′ = (z2 , . . . , zn ). Of course, F is not empty. The con-
stants are in F as well as all complex polynomials, since the monomials z j := z1j1 . . . znjn
are in F for multiindices j = (j1 , . . . , jn ). In fact, introducing polar coordinates for
each variable zk and using Fubini’s theorem it is easy to show

Lemma B.15.
zj µ
< ∞ , and z j , z k = 0 if j ̸= k .

As a consequence,
P = C[z1 , . . . , zn ] ⊊ F ⊊ O(Cn ).

We conclude this subsection with

Proposition B.16. The space of polynomials P is dense in F. As a consequence the

normed monomials
zj
mj (z) := j , j ∈ Nn ,
∥z ∥µ
form an orthonormal basis of the Hilbert space F.
B.4 Sequences of Holomorphic Functions 291

Proof. Let f ∈ F. As a holomorphic function f has the power series expansion

X X
f (z) = P f k(z) = cj z j ,
k∈N j∈Nn

where cj ∈ C (cf. Proposition B.5). The convergence is uniform with respect to every
polydisc P (r) , r ∈ R+ . We want to show that the series converges in norm, as well.
By the preceding lemma we know P that the terms cj z j (no summation!) are orthogonal
in F. Therefore, in order that cj z j converges to f in the norm of F it is sufficient,
j
that the sequence (|cj | ∥z ∥) is square summable.
Restricting to a polydisc D(r), the cj z j are again orthogonal 2
P jin L (D(r), dµ) (same
proof as for the preceding lemma). The convergence of cj z is uniform on D(r),
which implies, that summation and integration can be interchanged in
Z Z X X Z
∥f ∥2L2 (D(r),dµ) = f¯f dµ = cj zj c jz
j
dµ = 2
|cj | z j z j dµ
D(r) D(r) D(r)

to obtain
2
X
∥f ∥2L2 (D(r),dµ) = |cj |2 z j L2 (D(r),dµ)
.

For r → ∞ we conclude (monotone convergence)

2
X
∥f ∥2L2 (Cn ,dµ) = |cj |2 z j L2 (Cn ,dµ)
,

cj z j → f in the norm of F.
P
which is enough to assure the convergence

This result has the following interpretation.

Remark B.17. Let V := Cn ∨ ⊂ F the space of complex linear functionals on Cn with

the induced Hilbert space structure and let V⊙k its k-fold symmetric tensor product.
Then F can be identified with the symmetric Fock space of V
∞
M
V⊙k .
k=0

Moreover, the operators P k : F → V⊙k are the projections of the above decompositon
of F.

Remark B.18. It is easy to show that for a complex polynomial g the multiplication
operator Mg : P → P ⊂ F , f 7→ gf := Mg (f ) is a closed operator with domain P in
the Hilbert space F (see Definition F.7) and densely defined.
292 B. Complex Analysis

B.5 Complex Manifolds

Most of the basic notions of smooth manifolds (see Section A.1) carry over to the
holomorphic case.
A map F : V → V ′ between open subsets V ⊂ Cn and V ′ ⊂ Cm is holomorphic by
definition, if for every a ∈ V and b ∈ Cn , the map

z 7→ F (a + zb)

is a holomorphic map in one variable in a neighbourhood of 0 (with values in Cm ).

This condition is equivalent to the property that all the components Fk : V → C of
F = (F1 , . . . , Fm ) are holomorphic in the sense Section B.1; but V is no longer assumed
to be connected, in general. As before, by the result of Hartogs, F is continuous.
Definition B.19. M is a Complex Manifold of dimension n, if it is a smooth
manifold of real dimension 2n where an atlas of the differentiable structure is specified
which determines the complex structure of M and which consists of Holomorphic
Charts (φj )j∈I which are holomorphically compatible to each other. This means

φj : Uj → Vj ⊂ Cn , j ∈ I, Vj open in Cn ,

are diffeomorphisms, where (Uj )j∈I is an open cover of M and Vj ⊂ Cn is open, such
that the transition maps:

φk ◦ φ−1
j : φj (Uj ∩ Uk ) → φk (Uj ∩ Uk )

are biholomorphic.
This atlas determines the complex structure by defining general holomorphic charts:
A holomorphic chart on the complex manifold is a continuous map φ : U → V from
an open U ⊂ M to an open V ⊂ Cn such that all

φ ◦ φ−1
j : φj (U ∩ Uj ) → φ(U ∩ Uj ) , j ∈ I ,

are biholomorphic.

A mapping F : M → N between complex manifold is holomorphic if the mappings

φ ◦ F |U ◦ φ−1 : V → U ′ are holomorphic for all holomorphic charts φ : U → V ⊂ Cn
′

of M and all holomorphic charts φ′ : U ′ → V ′ ⊂ Cm with F (U ) ⊂ U ′ .

F |U
UO / U′
φ−1 φ′

V / V′
O(M, N ) denotes the set of holomorphic maps F : M → N .
B.5 Complex Manifolds 293

Submanifolds, Products, Quotients

The notion of submanifolds, product manifolds and quotient manifolds are com-
pletely analogous to the real case (see Section A.1). The universal property of quo-
tients, however, need some care since there exists complex manifolds which have no
global holomorphic functions except for the locally constant functions.
Observation B.20. In fact, the maximum principle (Proposition B.10) implies that
for a complex manifold M , which is connected and compact, the space of holomorphic
functions is O(M ) ∼
= C.
Holomorphic Vector Bundles
The notion of holomorphic vector bundle carries over as well (c.f. Definition D.1):
π : E → M is a holomorphic vector bundle of rank r over a complex manifold M ,
when E is a complex manifold such that π is holomorphic and when each fibre Ea :=
π −1 (a) , a ∈ M , has the structure of a r-dimensional complex vector space. Moreover,
there is an open cover (Uj )j∈I of M with biholomorphic maps (the local trivializations)
ψj : EUj := π −1 (Uj ) → Uj × Cr such that
1. π|EUj = pr1 ◦ ψj , i.e. the following diagram is commutative
ψj
EUj / Uj × Cr
π|EU
j pr1
 z
Uj

2. the restrictions Ea → Cr , v → pr2 (ψj (v)) , are C-linear.

Tangent and Cotangent Bundles

In particular, for a n-dimension complex manifold M the tangent bundle τ : T M →
M is a complex manifold of dimension 2n and a holomorphic vector bundle of rank n,
and the same is true for the cotangent bundle τ ∗ : T ∗ M → M . Some facts about these
bundles are summarized:
Each curve γ ∈ O(D, M ) defined on an open disc D := {z ∈ C | |z| < r} through
the point a := γ(0) ∈ M determines a (complex) Tangent Vector X = [γ]a at a to
M : [γ]a is the equivalence class of (germs of) curves in M through a which is given by
the equivalence relation
d d
γ ∼a β ⇐⇒ (φ ◦ γ)|z=0 = (φ ◦ β)|z=0 ,
dz dz
where β ∈ O(D, M ) with β(0) = a and φ : U → V is a holomorphic chart with a ∈ U .
The equivalence relation is independent of the holomorphic chart φ. The tangent vector
X = [γ]a is denoted also by
d
γ|z=0 or simply γ̇(0) .
dz
294 B. Complex Analysis

Every holomorphic chart φ : U → V induces a Bundle Chart (cf. A.2) φ̃ : T U →

V × Cn on the tangent bundle T M by which the complex structure on the tangent
bundle is determined:
 
1 n 1 n

n d
φ̃ = z , . . . , z , w , . . . , w : T U → V × C , [γ]a 7→ (φ(a), (φ ◦ γ)|z=0 ,
dz

when γ(0) = a. Here, wj acts in the following way

 
j d j
w ([γ]a ) = (z ◦ γ)|z=0 .
dz

The holomorphic sections of the two bundles are the holomorphic vector fields
resp. holomorphic one forms.

Differential Forms
Let X(U ) be the O(U )-module of the holomorphic vector fields on an open subset
U ⊂ M of a complex manifold. As in the smooth case the O(U )-module Ω(U ) of
holomorphic one forms and the O(U )-module of holomorphic vector fields X(U ) are in
duality: The O(U )-bilinear form

⟨ , ⟩ : Ω(U ) × X(U ) −→ O(U ) , (α, X) 7→ α(X) ,

is non degenerate and defines the isomorphisms

X(U ) → Ω(U )∗ , X 7→ (α 7→ α(X) = ⟨α, X⟩ ,

Ω(U ) → X(U )∗ , α 7→ (X 7→ α(X) = ⟨α, X⟩ .

The holomorphic k-forms are the k-multilinear and alternating maps

X(U )k → O(U )

over O(U ), and the O(U )-module of holomorphic k-forms is denoted by Ωk (U ) with
Ω0 (U ) = O(U ), and Ω1 (U ) = Ω(U ).
In local coordinates given by a holomorphic chart φ = (z 1 , . . . , z n ) : U → V ⊂ Cn
a holomorphic vector field X ∈ X(U ) can be represented by X = X j ∂j , where ∂j is the
holomorphic vector field ∂j a = [φ−1 (φ(a) + zej )]a and Xj ∈ O(U ). This vector field
acts on holomorphic functions f as the Lie derivative
∂f
LX f = X j L∂j f = X j .
∂z j
In particular,
∂f d
L ∂j f = j
:= f (φ−1 (φ(a) + zej )) ,
∂z dz z=0
B.5 Complex Manifolds 295

for holomorphic f ∈ O(U ) , and it seems natural to denote ∂j by

∂
,
∂z j
analogous to the case of a differentiable manifold, where a coordinate xj generates the
vector field
∂
.
∂xj
However this notation is reserved for a slightly more general definition and leads to
a vector field which acts not only on holomorphic but also on differentiable functions
(see below (83)).
A typical 1-form is df ∈ Ω defined by df (X) = LX f , X ∈ X(U ), for f ∈ O(U ). In
particular, dz j is dual to ∂j with dz j (∂k ) = δkj .
A k-form η ∈ Ωk (U ) has the representation
X
η= ηj1 ,...,jn dz j1 ∧ . . . dz jn ,
j1 +...+jn =k

with
1
ηj1 ,...,jn = η(∂j1 , . . . , ∂jn ) ∈ O(U ) .
k!
Underlying Smooth Structure
A complex manifold considered as a differentiable manifold has differential forms
which are not holomorphic. They have additional graduations coming from the complex
structure of the manifold and, in addition, from the complexification of the tangent
bundle. This graduation will be described in detail in the following.
The results are best motivated by investigating the local situation first.
So, let U ⊂ Cn be an open subset with standard holomorphic coordinates
(z , . . . , z n ) and related real coordinates xj , y j satisfying z j = xj + iy j . Let us re-
1

gard the real tangent space Ta U , a ∈ U , considered as the tangent space of dimension
2n with respect to the differential structure. Then
 
∂ ∂
, , 1 ≤ j ≤ n,
∂xj ∂y j
is a natural basis of Ta M over R. Recall that
∂ ∂
= [a + tej ]a , and = [a + tiej ]a ,
∂xj a ∂y j a

where ej is the standard basis of Cn defining the complex (and holomorphic) coordinates
z j : z = z j ej for z ∈ Cn . The
∂ ∂
,
∂xj ∂y j
296 B. Complex Analysis

are smooth sections in the tangent bundle and hence they are (smooth) vector fields
which are not holomorphic vector fields!
The multiplication by i giving Cn the structure of a complex vector space carries
over to the tangent space Ta U : Applied to the basic vector fields
∂ ∂
,
∂xj ∂y j
we obtain
∂ ∂ ∂ ∂
= a + ti2 ej a = [a − tej ]a = − j
 
i = [a + tiej ]a = , and i .
∂xj a ∂y j a ∂y j a ∂x a

As a consequence, the multiplication by i in the tangential space Ta U is the real

isomorphism determined by
∂ ∂ ∂ ∂
j
7→ j , j
7→ − j .
∂x ∂y ∂y ∂x
Definition B.21. Let V be a real vector space. An Almost Complex Structure
on a V is a R-linear map J : V → V with J 2 = J ◦ J = −idV .
An Almost Complex Structure on a differentiable manifold M is a section
J ∈ Γ(M, End (T M )) satisfying J ◦ J = −idT M . (M, J) is called an almost complex
manifold.

It is easy to see that an almost complex structure requires V to be even dimensional

when V is finite dimensional. In Section 9.5 on Kähler polarizations the concept of an
almost complex structure is introduced and investigated (cf. Definitions 9.19, 9.23 ff.).

Examples B.22. 1. When V is a complex vector space, then the multiplication

with i induces an almost complex structure v 7→ iv since i2 v = −v. Conversely,
a given almost complex structure J : V → V on a real vector space defines the
structure of a complex vector space on V by (α + iβ)v := αv + βJ(v) , α, η ∈
R, v ∈ V .

2. As we have just seen above, the real tangent space Ta U for U open in Cn , a ∈ U
admits the almost complex structure Ja , a ∈ U given by
∂ ∂ ∂ ∂
Ja : Ta U → Ta U , →
7 , →
7 − ,
∂xj ∂y j ∂y j ∂xj
which is independent of the choice of the holomorphic chart.

3. The corresponding section J ∈ Γ(U, End (T U )) is an almost complex structure

on U . Therefore, every complex manifold is an almost complex manifold. The
converse is only true for integrable almost complex manifolds (see Theorem 9.25).
B.5 Complex Manifolds 297

4. The symplectic involution (c.f. Section 1.1.2) defining the symplectic structure
in Chapter 1 is an almost complex structure.

Now, let V be a real vector space V with an almost complex structure J and let
V be its complexification V ⊗ C. Then J has a linear continuation J C : V C → V C by
C

J(v ⊗ λ) = J(v) ⊗ λ. J C is an almost complex structure of V C , so that on V C we have

the two different (almost) complex structures given by J C and by the multiplication
with i : v ⊗ λ 7→ v ⊗ iλ.
V C decomposes into the eigenspaces of J C for the eigenvalues i, −i (see (43)). This
decomposition carries over to the tangent space T M C of an almost complex manifold
(see also in Section 9.5):
Proposition B.23. Let (M, J) be an almost complex manifold. The continuation
(1,0)
JaC : Ta M ⊗ C → Ta M ⊗ C has the two complementary eigenspaces: Ta M with
(0,1)
eigenvalue i and Ta M with eigenvalue −i.

In case of a complex manifold M the decomposition can be described with the aid of
local holomorphic coordinates z j in an open subset U ⊂ M . The complexified tangent
space Ta U ⊗ C has the basis
 
∂ ∂
⊗ 1, ⊗1
∂xj ∂y j
over C. We define
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
j
:= j
−i j and j
:= j
+i j . (83)
∂z 2 ∂x ∂y ∂ z̄ 2 ∂x ∂y
We obtain  
∂
Ta(1,0) U
= Ker (J − i) = spanC
C
j = 1, . . . , n ,
∂z j
 
(0,1) C ∂
Ta U = Ker (J + i) = spanC j = 1, . . . , n .
∂ z̄ j
(1,0) (0,1)
Remark B.24. The decomposition Ta M ⊗ C = Ta M ⊕ Ta M holds true in all of
M and remains true for the bundles:

T M C = T (1,0) M ⊕ T (0,1) M

with the obvious definition for the holomorphic bundle structure on the direct sum of
the bundles T (1,0) U , T (0,1) U .

The operator J induces J ∗ by J ∗ (µ) := µ ◦ J for µ ∈ T ∗ M . J ∗ ◦ J ∗ = −id. In the

same way as before we obtain:

Ta∗ M C = Ta∗(1,0) M ⊕ Ta∗(0,1) M

298 B. Complex Analysis

Corollary B.25. The analogous decomposition holds for the cotangent bundle with
respect to J ∗ :
T ∗ U ⊗ C = T ∗(1,0) U ⊕ T ∗(0,1) U
with
Ta∗(1,0) U = Ker (J ∗ − i) = spanC {dz j | j = 1, . . . , n} ,
Ta∗(0,1) U = Ker (J ∗ + i) = spanC {dz̄ j | j = 1, . . . , n}.
Here, for a differentiable f ∈ E(U ) the form df ∈ Γ(U, T ∗ U C ) by

∂ ∂ ∂ ∂
df ( j
) = j f , df ( j ) = j f .
∂z ∂z ∂ z̄ ∂ z̄
Definition B.26. We define as bundles and spaces
r,s r s
^ ^ ^
T ∗ U := T ∗(1,0) U ∧ T ∗(0,1) U

for r, s ∈ N , r + s ≤ n
r,s
^
r,s
A (V ) := Γ(V, T ∗ U ) , V ⊂ U , V open.

The k-forms η ∈ Ar,s (V ) , k = r + s , are called differential forms of degree (r, s) or

(r, s)-forms.

Note, that Ak (V ) = r+s=k Ar,s (V ). In local coordinates

L

 
 X 
r,s j1 jn j̄1 j̄n
A (V ) = ηj1 ,...jr ;j̄1 ,...j̄s dz ∧ · · · ∧ dz ∧ dz̄ ∧ · · · ∧ dz̄ } ηj1 ,...jn ;j̄1 ,...j̄r ∈ E(V, C) .
 
j1 ,...jr ;j̄1 ,...j̄s
 299

C Lie Theory

The concept of a Lie group is important in many areas in physics, since essentially
all Symmetries are formulated with the aid of Lie groups and correspondingly the
Infinitesimal Symmetries are formulated by using Lie algebras105 . In mathematics
Lie groups are equally important in particular in Differential and Algebraic Geometry
as well as in Number Theory.
In this chapter we will present the basics of Lie groups and Lie algebras. Our ap-
proach will be introductory, we start from the basic definition of a Lie group, then
specialize to matrix Lie groups (Section C.1), and explain several examples, in partic-
ular by semidirect products or central extensions. Afterwards, we study Lie algebras,
which can be defined over general fields, a priori without emphasizing the connection
to Lie groups (Section C.3). The two concepts are related as described in Section C.4,
namely: to every Lie group, there is a corresponding Lie algebra. Conversely for every
finite dimensional Lie algebra L over R there exists a Lie group whose Lie algebra is
L106 . Note, that this correspondence is not one-to-one. There might be several Lie
groups with the same Lie algebra.
Symmetry often leads to an identification of elements which are related directly by
the symmetry. In this way orbits are created in the space on which the symmetry group
acts, and the orbit space is indeuced. If the symmetry is given by a Lie group G acting
on a differentiable manifold M it is of great interest to have a natural differentiable
structure on the corresponding Orbit Space M/G, where natural means that it is
the quotient structure. Under quite general assumptions on the action such a result is
known, as we explain in the Section C.5 of this chapter.

C.1 Lie Groups

Definition C.1. A Lie group G is a group, which at the same time is a manifold,
such that the multiplication
µ : G × G → G , (f, g) 7→ µ(f, g) := f g,
and the inversion
j : G → G , f 7→ j(f ) = f −1 ,
are smooth maps. A Lie Group Homomorphism between Lie groups G, G′ is a
group homomorphism h : G → G′ which is smooth.

Of course, Rn with the addition as group operation is a Lie group. The same is true
for R× and C× with respect to multiplication. A typical Lie group homomorphism is
exp : C → C× from the additive group C of complex numbers onto the multiplicative
group C× : exp(z + w) = exp z exp w. Moreover:
105
This is the topic of our book [Sch95].
106
Theorem of Ado
300 C. Lie Theory

Examples C.2.

1. Circle Group. The circle group U(1) := {z ∈ C | |z| = 1} with the multi-
plication inherited from C is an abelian group. Evidently, U(1) as a group is
isomorphic to the group of rotation matrices
  
cos t − sin t
SO(2) := t∈R .
sin t cos t

As a subset of C1 ∼ = R2 the group U(1) inherits a natural smooth manifold

structure from R , namely as a submanifold it is the circle S1 . The multiplication
2

µ : S1 × S1 → S1 , µ(eiφ , eiψ ) = ei(φ+ψ) ,

and the inversion

j : S1 → S1 , j(eiφ ) = e−iφ ,
are clearly smooth, hence U(1) is a Lie group. Notice, that C× ∼
= R+ × U(1) as
Lie groups.

2. Special Unitary Group SU(2). Important in Quantum Mechanics is the special

unitary group SU(2):

SU(2) := {A ∈ U(2) | det A = 1}.

U(2) := {A ∈ C(2)107 | ⟨Az, Aw⟩ = ⟨z, w⟩ for all z, w ∈ C},

where ⟨z, w⟩ := z̄ 1 w1 + z̄ 2 w2 is the Hermitean scalar product on C2 .
SU(2) can also be written as:
  
z w 2 2
SU(2) = |z| + |w| = 1. .
−w̄ z̄

The equation |z|2 + |w|2 = 1 shows that SU(2) is a manifold diffeomorphic to the
3-sphere S3 in C2 ∼
= R4 . One can easily see that matrix multiplication is smooth
and – by using Cramer’s rule – the same is true for building the inverse. Hence,
SU(2) is a Lie group.

3. General Linear Group. The group GL(n, R) of all real invertible (n × n)-
matrices is an open subset of the R−vector space of all (n × n)-real matrices
R(n) ∼
2
= Rn . This follows from:

GL(n, R) = {A ∈ R(n) : det A ̸= 0} ,

107
K(n) denotes the algebra of (n × n)-matrices with coeffincients in K.
C.1 Lie Groups 301

since the map det : R(n) → R is continuous. The general linear group
2
GL(n, R) has the structure of a differentiable manifold as open subset of Rn .
The matrix multiplication

µ : GL(n, R) × GL(n, R) −→ GL(n, R) , (A, B) 7−→ AB ,

is polynomial of degree 2 in the coefficients aki , bij of the matrices A = (aki ) and
B = (bij ):
X n
AB = aki bij .
i=1

Therefore, matrix multiplication is smooth (even analytic). Similarly, using

Cramer’s rule, one can see that the inversion is also smooth and analytic. There-
fore, GL(n, R) is a Lie group. In the same way, one can show that GL(n, C), the
group of invertible n×n complex matrices is a Lie group. GL(n, C) is, in addition,
a complex manifold, and the group operations are holomorphic mappings.

4. Special Linear Group:

SL(n, R) := {A ∈ R(n) : det A = 1}

SL(n, C) := {A ∈ C(n) : det A = 1}

To show that the special linear groups are Lie groups one only has to check, that
they are submanifolds of the general linear groups. This is not difficult since
they are the zero sets of the differentiable function det. However, one can use a
general result on closed subgroups of a Lie group to deduce the Lie property as
we explain in the following.

Let us define:
Definition C.3. A matrix Lie group or simply a matrix group is a closed subgroup
G of GL(n, C) , n ∈ N.

Of course, GL(n, C) itself is a matrix group. Moreover, GL(n, R), SL(n, K),
U(1), SU(2), SO(3) are all matrix groups.
With some work, one can show that every matrix group is a Lie group. Since the
inversion and multiplication are smooth, it is enough to show the following result (cf. in
[RuS13], for example).
Proposition C.4. A closed subgroup of a Lie group is a submanifold of the Lie group
and thus a Lie group. In particular, a matrix Lie group G ⊂ GL(n, C) is a Lie group
2
and a differentiable submanifold of R4n .

A large class of geometrically induced matrix Lie groups is given by the following
result:
302 C. Lie Theory

Lemma C.5. Let B ∈ K(n) be a non-degenerate (n × n)-matrix. Then

OB (n, R) = {A ∈ R(n) | A⊤ BA = B}

in case of K = R, resp.
⊤
OB (n, C) = {A ∈ C(n) | A BA = B} ,

is a matrix Lie group.

Proof. For A, A′ ∈ OB (n, K) the equality (AA′ )⊤ BAA′ = A′⊤ A⊤ BAA′ = A′⊤ BA′ = B
holds true. Furthermore, A is invertible because of 0 ̸= det B = det A⊤ det B det A.
−1
Finally, A⊤ BA = B implies (A−1 )⊤ BA−1 = B (resp. (A )⊤ BA−1 = B). As a result
OB (n, K) is a subgroup of GL(nK). OB (n, R) is closed since it is the zero set of the
continuous map A 7→ A⊤ BA−B. Hence, OB (n, K) is a matrix group and Lie group.

We list a couple of matrix groups of the form described in the Lemma:

Examples C.6.

1. Orthogonal Group:

O(n) := A ∈ R(n) | A⊤ A = 1 = O1 (n, R)



is the Orthogonal Group. The orthogonal matrices A ∈ O(n) leave invariant

the Euclidean scalar product ⟨ , ⟩ on Rn . O(n) is compact, since the coefficients
of the matrices A ∈ O(n) are bounded: |Aij | ≤ 1 and since O(n) is closed in C(n).
It is easy to see that | det A| = 1 for all A ∈ O(n). The Special Orthogonal
Groupspecial orthogonal group

SO(n) := {A ∈ O(n) | det A = 1}

is connected, and it is a proper subgroup of O(n). O(n) is not connected, since

{A ∈ O(n, R) | det A = −1} is open and closed. O(n, R) = SO(n, R) ∪ {−A |
A ∈ SO(n, R)} .

2. Unitary Group.
n ⊤
o
U(n) := A ∈ C(n) | A A = 1 = O1 (n, C)

is the Unitary Group, the operators A ∈ U(n) respect the hermitian scalar
product. The special unitary group is

SU(n) := {A ∈ U(n) : det A = 1}.

The groups U(n) and SU(n) are connected and compact.

C.1 Lie Groups 303

3. Generalized Orthogonal Group. For p, q ∈ N, n = p + q, one defines using

the bilinear form p
X n
X
B(x, y) := xj y j − xj y j .
j=1 j=p+1

The matrix Lie groups OB (n, R) are called the Generalized Orthogonal
Groups and they are denoted by O(p, q). The corresponding special orthogonal
groups SO(p, q) are defined as the connected components of the identity of O(p, q).

4. Symplectic Group. Let I ∈ R(2n) be the matrix

 
0 1
I= ,
−1 0

where 0 and 1 are n × n matrices (see Section 1.1). The corresponding matrix
group OI (2n, R) is the symplectic group:

Sp(n) := A ∈ R(2n) : A⊤ IA = I .

(84)

A matrix Lie group of the form OB (n, K) can be understood as a symmetry group,
namely as the group of mappings Kn → Kn preserving the vector space structure
and the structure given by B. Many more Lie groups occur as symmetry groups,
for instance the Euclidean group E(n), which is the group of all differentiable maps
Rn → Rn preserving the euclidean scalar product and the orientation, in short the
isometries. Similar to the Galilean group and the Poincaré group, the Euclidean group
has a description as a semidirect group.

Definition C.7. Let G, H be Lie groups and σ : G → Aut H 108 a group homomor-
phism such that G × H → H , (g, h) 7→ σ(g)(h) , is smooth. The Semidirect Prod-
uct of G over H (denoted by G ⋉ H or G ⋉σ H) is the group with the underlying set
G × H and the group operation

((g, h)(g ′ , h′ )) 7→ (gg ′ , (σ(g)h′ )h) .

It is not difficult to check that a semidirect product G ⋉ H of Lie groups is a Lie

group. The underlying manifold is the product G × H of the two manifolds. The group
operation ((g, h)(g ′ , h′ )) 7→ (gg ′ , (σ(g)h′ )h) is smooth since σ and the group operations
on G and H are smooth. The inversion (g, h) 7→ (g −1 , σ(g −1 )h−1 ) is smooth as well.
Finally, the associativity can directly be checked.

Examples C.8.
1. The product E = G × H of Lie groups is a special case of a semidirect product
where σ(g) = idH .
108
Aut H is the automorphism group of H, i.e. the group of Lie group isomorphisms
304 C. Lie Theory

2. The Euclidean group E(n) is (isomorphic) to the semidirect product SO(n) ⋉ Rn

with respect to the inclusion σ : SO(n) ⊂ GL(n, R) ⊂ Aut (Rn ), where Rn is
considered as the abelian Lie group.

3. The affin linear group Aff(Rn ) can be described as the semidirect product
GL(n, R) ⋉ Rn .

4. The Galilei group Γ is isomorphic to the semidirect product of the Euclidean

group E(3) and the translation group V ∼ = R4 : Γ ∼ = E(3) ⋉ V . The action σ :
E(3) → Aut V is defined by σ(g)(q, t) = (Aq + tv, t) for g = (A, w) ∈ SO(3) × R3
and (q, t) ∈ R4 = R3 × R. Eventually, Γ ∼
= (SO(3) ⋉ R3 ) ⋉ R4 .

5. The Poincaré group P is isomorphic to the semidirect product of the Lorentz

group O(3, 1) and the translation group R3+1 : P ∼
= O(3, 1) ⋉ R4 .

C.2 Extensions of Lie Groups

A Lie group E has a representation as a semidirect product E = G ⋉ H if and only if

there is an exact sequence
ι π
1 −→ H −→ E −→ G −→ 1

of Lie group homomomorphisms such that there exists a splitting s : G → E, i.e. a Lie
group homomorphism with π ◦ s = idG .
Such a sequence is an extension of groups in the following sense:

Definition C.9. An Extension E of G by the group H is given by an exact sequence

of group homomorphisms
ι π
1 −→ H −→ E −→ G −→ 1 .

Exactness of the sequence means that the kernel of every map in the sequence equals
the image of the previous map. Hence the sequence is exact if and only if ι is injective,
π is surjective, and Ker π = Im ι .
The extension is called central if H is abelian and its image im ι is in the center
of E, that is
a ∈ H, b ∈ E ⇒ ι (a) b = b ι (a) .

In case all groups are Lie groups we have (central) extensions of Lie groups.

Note, that H is written multiplicatively and 1 is the neutral element even if H is

supposed to be abelian.
There are extensions with abelian H which are not central, as e.g. in the above
examples of semidirect products.
C.2 Extensions of Lie Groups 305

Examples C.10.
1. A trivial extension has the form
ι π
1 −→ H −→ E = G × H −→ G −→ 1 .

with ι(h) = (1, h) and π = pr1 . More generally, a central extension is called
trivial if it is isomorphic to a trivial central extension, and this is, in turn,
equivalent to the existence of a splitting s : G → E, π ◦ s = idG .

2. Given k ∈ N , k ≥ 2, a nontrivial central extension is the sequence

ι π
1 −→ Z/kZ −→ E = U(1) −→ U(1) −→ 1 ,

with π(z) = z k and ι([n]) = exp 2π nk . A splitting s : U(1) → U(1) would

correspond to a global root of z 7→ z k .

3. A familiar central extension is

ι π
1 −→ U(1) −→ U(n) −→ SU(n) −→ 1 .

4. The universal cover SU(2) → SO(3) is a central extension

ι π
1 −→ Z/2Z −→ SU(2) −→ SO(3) −→ 1 ,

as well as the corresponding cover SL(2, C) → SO(3, 1):

ι π
1 −→ Z/2Z −→ SL(2, C) −→ SO(3, 1) −→ 1 .

5. In general, the universal covering E of a Lie group G is again a Lie group and
the projection π : E → G gives rise to an extension of Lie groups
ι π
1 −→ H = Ker π −→ E −→ G −→ 1 ,

6. The Metalinear Group ML(n, C) is the following 2-1-covering of the general

linear group GL(n, C): ML(n, C) := {(z, A) ∈ C × GL(n, C) | z 2 = det A}. It is
a Lie group as a closed subgroup of a Lie group: ML(n, C) ⊂ GL(n + 1, C). The
map π : ML(n, C) → GL(n, C) , (z, A) 7→ A , is a Lie group homomorphism. For
A ∈ GL(n, C) there are exactly two roots z, −z of det A: z 2 = (−z)2 = det A.
Hence π −1 (A) = {(z, A), (−z, A)}. In case if A = idCn = id we have det A = 1
and 12 = (−1)2 = det A, Hence, The kernel of π is {(1, id), (−1, id)} ∼
= Z/2Z,
and we obtain the exact sequence of Lie groups:

1 −→ Z/2Z −→ ML(n, C) −→ GL(n, C) −→ 1 .

The natural homomorphism χ : ML(n, C) → C× , (z, A) 7→ z, can be viewed as

a root of the determinant on GL(n, C), since it satisfies χ2 = det ◦π(A).
306 C. Lie Theory

7. The Heisenberg Group HSn is defined as the central group extension of the
abelian group R2n by R: HSn can be described by HSn := R × R2n with the group
composition
 
′ ′ ′ ′ 1 ′ ′ ′ ′
(r, p, q) · (r , p , q , ) := r + r + (p · q − q · p ), p + p , q + q ,
2
where p, q, p′ , q ′ ∈ Rn , r, r′ ∈ R. The corresponding exact sequence of Lie groups
is
1 −→ R −→ HSn −→ R2n −→ 1 .
A variant of the Heisenberg group is the polarized version HSpol
n = U(1) × R
n

with exact sequence

1 −→ U(1) −→ HSpol
n −→ R
2n
−→ 1 .

A coordinate free version is the following: Let (V, ω) be a symplectic vector space
of dimension n. The Heisenberg group HS(V, ω) is R × V with the group law
1
(r, v)(r′ , v ′ ) = (r + r′ + ω(v, v ′ ), v + v ′ ) .
2

Notice, that HSn is not abelian although R2n and R are abelian.
8. An important example in the context of quantization of symmetries is the fol-
lowing: Let H be a Hilbert space and let P = P (H) be the projective space of
one-dimensional linear subspaces of H with the natural projection γ : H → P(H).
P is the space of states in quantum physics, that is the quantum mechanical phase
space (see Appendix F). The group U(H) of unitary operators on H turns out to
be e central extension, as we explain in the following.
By definition, the group U(P) = U(P(H)) of projective unitary operators or
quantum symmetries consists of the bijective mappings V : P → P which are
induced by unitary operator U ∈ U(H) by V (γ(ϕ)) := γ(U (ϕ)) , ϕ ∈ H. Let us
denote V =: γ b : U (H) → U(P) is surjective with kernel {λ idH | λ ∈
b(U ). Then γ
∼
U(1)} = U(1) and yields in a natural way a nontrivial central extension of the
group U(P) of (unitary) projective transformations on P by U(1):
ι γ
1 −→ U(1) −→ U(H) −→ U(P) −→ 1 .
b

U(H) is a topological group 109 with respect to the strong topology (see [Sch95]).
The strong topology on U(H) is the topology which is generated by the sub-
sets V (T, ϕ, ε) := {S ∈ U(H) | ∥Sϕ − T ϕ∥ < ε}: The open subsets of
109
i.e. the group multiplication and the inversion are continuous. Note, that in the infinite dimen-
sional case, no Lie group (or differentiable) structure on U(H) is known. However, in finite dimen-
sions, the Hilbert space is isomorphic to Cn , H ∼= Cn , and U(H) ∼= U(n) with a Lie group quotient
∼
U(n) → PU(n) = U(P). In particular, the sequence is an exact sequence of Lie groups.
C.2 Extensions of Lie Groups 307

U(H) are arbitrary unions of finite intersections of sets in the subbase B =

(V (T, ϕ, ε) | T ∈ U(H), ϕ ∈ H, ε > 0). Note, that the strong topology is weaker
than the operator norm topology when H is infinite dimensional. In case of finite
dimension the strong topology on U(H) ∼ = U(n) is simply the usual topology
induces from C(n). U(P) becomes a topological group by the quotient topology
with respect to γ̂ : U(H) → U(P). This topology is called the strong topology as
well.
We have introduced these topologies on U(H), U(P) in order to state, that the
above exact sequence consists of continuous maps, and therefore is an exact
sequence (and a central extension) of topological groups.

Remark C.11. In the light of the last examples it is possible to explain to which
extent a classical symmetry given by a Lie group G can induce a quantum symmetry.
Let G act on a symplectic manifold (M, ω) by symplectomorphisms, i.e. there is an
action Ψ : M × G → M such that q 7→ Ψ(q, g) := Ψg (q) is a symplectomorphism.
If Symp(M, ω) denotes the group of diffeomorphisms which are symplectomorphisms,
the action induces a group homomorphism Ψ : G → Symp(M, ω) , g 7→ Ψg . Assume
now, that a Lie algebra R ⊂ E(M ) of classical observables has been quantized yielding
a subalgebra q(R) of self-adjoint operators on a Hilbert space H. The best one can
hope is that the action Ψ transforms into a projective unitary representation ρ : G →
U(P(H))110 . In general, this property can not be deduced from physics or mathematics.
But of course, it can be formulated as a postulate.
Since in Quantum Mechanics calculations are essentially carried through in the
Hilbert space H associated to the quantum mechanical phase space P one is interested to
lift the representation ρ : G → U(P) to a unitary representation G → U(H) i.e. a group
homomorphism which is continuous with respect to the strong topology on U(H)111 .
Such a lift will not exist in general, but it exists up to a central extension. In fact, there
exist a surjective Lie group homomorphism π : Ĝ → G with ι : U(1) ∼ = Ker π → Ĝ
in the center of G and a unitary representation ρ̂ : Ĝ → G such that the following
diagram commutes (see, for instance [Sch08]):

1 / U(1) / Ĝ
π / G / 1
id ρ̂ ρ
  γ̂

1 / U(1) / U(H) / U(P) / 1

According to a theorem of Bargmann, for a simply connected Lie group G with

trivial cohomology H 2 (Lie G, R) = 0 the upper row can be replaced by a trivial central
extension and therefore, the representation ρ : G → U(P) has a direct lift ρ̃ : G → U(H)
110
ρ is a projective unitary representation if ρ is a group homomorphism and a continuous map with
respect to the strong topology on U(P) defined above.
111
often enough it is simply assumed that such a representations exists.
308 C. Lie Theory

with ρ = ρ̃ ◦ γ̂: ρ̃(g) = ρ̂(1, g) , g ∈ G. The following diagram is commutative:

1 / U(1) / U(1) × G π / G / 1
ρ̃
id ρ̂ ρ
  y γ̂

1 / U(1) / U(H) / U(P) / 1

Note, that G = SU(2) and G = SL(2, C) are simply connected and satisfy
2
H (Lie G, R) = 0.
As a consequence, in the case of G = SO(3) the upper row in the diagram can be
replaced by
ι π
1 −→ Z/2Z −→ SU(2) −→ SO(3) −→ 1 ,
and analogously in the case of G = SO(3, 1) by
ι π
1 −→ Z/2Z −→ SL(2, C) −→ SO(3, 1) −→ 1 .

These properties explain why it is reasonable to call SU(2) the quantum mechanical
rotation group or SL(2, C) the corresponding Lorentz group.

C.3 Lie Algebras

Definition C.12. A Lie Algebra L over a field k is a k-vector space together with
the map [ , ] : L × L → L (the Lie Bracket) with the following properties: For all
X, Y, Z ∈ L and λ ∈ k the Lie bracket satisfies

1. [X + λY, Z] = [X, Z] + λ [Y, Z]

2. [X, Y ] = − [Y, X]

3. [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0

A Lie algebra homomorphism between Lie algebras L, L′ is a linear map h : L → L′

with h([X, Y ]) = [h(X), h(Y )] , X, Y ∈ L.

Examples C.13.

1. Abelian Lie Algebra. Every k−vector space L with [X, Y ] = 0 for all X, Y ∈ L
is a Lie algebra over k, the so called trivial or abelian Lie algebra.

2. Cross product. R3 with [X, Y ] := X × Y (cross product) is a three-dimensional

Lie algebra over R.
C.3 Lie Algebras 309

3. Endomorphism Algebra. Let V be a k-vector space and let L = Hom (V, V ) :=

End V be the k-vector space of k-linear maps (endomorphisms) from V to V .
With the ”commutator” [X, Y ] := X ◦ Y − Y ◦ X for X, Y ∈ EndV as the Lie
bracket, this defines a Lie algebra structure as can easily be shown be direct
calculation. The triple (End V, ◦, [ , ]) is called the endomorphism algebra.
4. Matrix Algebra. In case of k = K ∈ {R, C} and V = Kn the endomorphism
algebra consists of matrices X ∈ Kn×n and the Lie subalgebras of End Kn are
called matrix algebras. End Kn is denoted by gl(n, K). Several matrix algebras
are described further below in Examples C.18.
5. The Lie algebra of vector fields. Let M be a manifold. For smooth vector
fields X on M let LX : E(M ) → E(M ) be the Lie derivative. To vector fields
X, Y there is a unique vector field Z, for which LZ = LX ◦ LY − LY ◦ LX . Let
Z =: [X, Y ]. Therefore the R− vector space of vector fields on M becomes an
infinite-dimensional Lie algebra V(M ). In a local chart q, we have:
∂Y k µ ∂X
k
[X, Y ]k = X µ − Y .
∂q µ ∂q µ

6. The Poisson algebra. Let (M, ω) be the phase space of Hamiltonian Mechanics,
namely a manifold M with symplectic form ω. Then E(M ) with the Poisson
bracket is a Lie algebra. Moreover, the Hamiltonian vector fields Ham(M ) form
a Lie algebra, and the map F 7→ −XF is a surjective Lie algebra homomorphism
(cf. Corollary 1.31 and Observation 1.35).
Example C.14. (Heisenberg Algebra) Let V be a finite-dimensional vector space over
R with a non-degenerate (constant) 2-form ω, i.e. a symplectic vector space. The
Heisenberg Algebra is the central (Lie algebra) extension hs = hs(V, ω) of the
abelian Lie algebra V with respect to the 2-form ω: Its underlying vector space is
R × V and the Lie bracket is
[(r, X), (s, Y )] := (rsω(X, Y ), 0).
The projection p = pr1 : R × V → V is a surjective Lie algebra homomorphism and
the injection j : R → V , t 7→ (t, 0), is an injective Lie algebra homomorphism with
Im j = Ker p. We have the exact sequence
0 −→ R −→ hs(V, ω) = R × V −→ V −→ 0 .
Moreover, the elements (z, 0) are central in hs(V, ω): [(z, 0), (r, X)] = 0 for all (X, r) ∈
hs(V, ω).
There exists a smplectic frame, i.e. a basis (ej ), (fk ), 1 ≤ j, k ≤ n of V such
that ω(ej , fk ) = δjk . The elements Z := (1, 0), Qj = (1, ej ), , Pk = (1, fk ) satisfy the
canonical commutation relations (CCR)
[Pk , Z] = [Qj , Z] = 0 , [Pj , Qk ] = −δjk Z .
310 C. Lie Theory

In case of V = Rn and the standard symplectic form on Rn the Heisenberg algebra

is also denoted by hsn (or hs2n+1 ).
One can introduce a so-called central charge c ∈ R changing the above Lie bracket
to c · (0, rsω(X, Y )) resp. −c · δjk . The result is an isomorphic algebra and the change
can also be achieved by replacing ω by c · ω.

Remark C.15. The Heisenberg algebra yields an abstract form of the canonical com-
mutation relations (CCR) which is an important concept of Quantum Mechanics. In
Section F.3 of the Appendix F on Quantum Mechanics we study the representations
of hsn and the corresponding Heisenberg group HSn in a Hilbert space, thus realiz-
ing the canonical commutation relations. Up to unitary equivalence, the Heisenberg
Lie group has only one irreducible unitary representation and this is the Schrödinger
representation. In particular, hsn is not matrix algebra.

After having seen the abstract definition and concrete examples of both abelian and
non-abelian Lie algebras, let us discuss the connection between Lie algebras and Lie
groups. We will do this in two steps: first we discuss it for the simpler case of matrix
Lie groups, then turn to abstract Lie groups.

C.4 The Lie Algebra of a Lie Group

For any matrix X ∈ C(n) the exponential series

∞
tX
X 1
e = (tX)ν , t ∈ R ,
ν=0
ν!

converges in C(n), and etX is invertible with inverse e−tX . Furthermore, the exponential
map e : R × C(n) → GL(n, C) is smooth112 .
Let us begin with the Lie group G = GL(n, C). The exponential map etX induces
the so-called Fundamental Field, namely the left-invariant vector field X̃ ∈ V(G)
defined by
d
X̃(A) := [AetX ]A = AetX t=0 = AX , A ∈ G .
dt
tX
In fact, t 7→ Ae , t ∈ R, is a curve γ in G through A = γ(0) and determines the
tangent vector [γ]A = X̃(A) ∈ TA G in the tangent space TA G at A ∈ G. Using the
trivialisation T GL(n, C) ∼= GL(n, C) × C(n), the fundamental field can be described
in the simple form X̃ : G → T G , A 7→ (A, AX). Hence, the fundamental field can be
considered to be constant and we conclude

Assertion C.16. The Lie bracket of X̃, Ỹ defined in V(G) ∼= E(G, C(n)) coincides
with the Lie bracket induced by the commutator in End Cn = C(n): [X̃, Ỹ ] = [X,
^ Y ].
112
even analytic!
C.4 The Lie Algebra of a Lie Group 311

The flow of the vector field X̃ (i.e. the solution of the differential equation
Φ̇ = X̃(Φ) , Φ(A, 0) = A, cf. Notation A.21) is Φ(A, t) := AetX , (A, t) ∈ G × R.
In particular, the vector field X̃ is complete.
X̃ is called left-invariant, since for the left multiplication Lg : G → G , A 7→
gA , A, g ∈ G, the following invariance condition holds:

X̃ ◦ Lg = T Lg ◦ X̃ .

This invariance gives rise to an alternative definition of the vector field X̃, namely

X̃(g) = Te Lg (X) , g ∈ G .

The main result of this section is:

Proposition C.17. Let G ⊂ GL(n, C) be a matrix Lie group. Then

g := Lie G := X ∈ C(n) | ∀t ∈ R : etX ∈ G



is a Lie algebra over R with the Lie bracket [X, Y ] = X ◦ Y − Y ◦ X, X, Y ∈ g, i.e. Lie G
is a Lie subalgebra of the endomorphism algebra C(n). Moreover,

Lie G = {γ̇(0) : γ curve in G with γ(0) = idCn = e } .

Therefore, g = Lie G can be identified with the tangent space at the identity element of
G.

Proof. We first show that Lie G = Te G. For each matrix X ∈ Lie G the curve t 7→ etX
in G through e = e0 induces a tangent vector [etX ]0 at identity e ∈ G which can be
identified with dtd etX |t=0 = X. Hence, Lie G ⊂ Te G.
Conversely, each tangent vector X ∈ Te G determines a left-invariant vector field
X̃ on G (even on all of GL(n, C)) by X̃(g) := Te Lg (X). The differential equation
γ̇ = X̃(γ) has a locally unique solution γ : I → G on an open interval I containing 0
such that γ(0) = e. Since the flow Φ(A, t) = AetX is also a solution of γ̇ = X̃(γ) with
γ(0) = e the two curves agree on I, which implies etX = γ(t) ∈ G for t ∈ I. It follows
etX ∈ G for all t ∈ R, which implies X ∈ Lie G. Thus Te G ⊂ Lie G.
Te G obtains a Lie algebra structure through the left-invariant vector fields. Given
X, Y ∈ Te G the fundamental fields X̃ and Ỹ determine the Lie bracket [X̃, Ỹ ] ∈ V(G).
Then Z := [X̃, Ỹ ](e) ∈ Te G is well-defined as the bracket Z = [X, Y ] of X, Y . It is easy
to check that in this way, Te G becomes a Lie algebra. Furthermore, by the assertion
C.16 the Lie bracket [X, Y ] = [X̃, Ỹ ](e) ∈ Te G for X, Y coincides with the commutator
of X, Y in End Cn . In particular, this shows that g, as defined in the proposition, is a
subalgebra of C(n).
312 C. Lie Theory

We now come to the general case: the notion of the Lie algebra Lie G assigned to a
given abstract Lie group G. Once again, we focus on the tangent space at the identity
Te G of G and we define for each tangent vector X ∈ Te G the left-invariant vector field
X̃ to X as:
X̃(g) := Te Lg (X),
where g ∈ G and where Lg : G → G is the left multiplication in G. X̃ is a vector
field on the manifold G, and for two such left-invariant vector fields X̃ and Ỹ the Lie
bracket [X̃, Ỹ ] ∈ V(G) is a vector field on G. As a consequence, the tangent vector
given by [X, Y ] := [X̃, Ỹ ](e) ∈ Te G is well-defined. By construction, the tangent space
Te G with this bracket [ , ] becomes a Lie algebra. This Lie algebra is called the Lie
algebra of the Lie group G and is denoted with g or Lie G.
Examples C.18.
1. The Lie algebra of GL(n, R) is gl(n, R) ∼
= R(n) ∼
= End Rn .
2. The Lie algebra of SL(n, R) is sl(n, R) = {X ∈ R(n) | trX = 0}.
3. The Lie algebra of O(n) is o(n) = {X ∈ R(n) | X ⊤ + X = 0}. Moreover,
Lie SO(n) = Lie O(n) = o(n).
⊤
4. The Lie algebra of U(n) is u(n) = {X ∈ C(n) | X + X = 0}.
⊤
5 The Lie algebra of SU(n) is su(n) = {X ∈ C(n) | X + X = 0 and trX = 0}.
Moreover, Lie SO(3) = Lie SU(2).
6. The Lie algebra of OB (n, R) is oB (n, R) = {X ∈ R(n) | X ⊤ B + BX = 0}.
7. The Lie algebra of the Heisenberg group HSn is the Heisenberg algebra hsn .
Proposition C.19. The assignment G 7→ Lie G for Lie groups G is categorial in the
following sense: Every Lie group homomorphism h : G → G′ induces a natural Lie
algebra homomorphism Lie h : Lie G → Lie G′ given by Lie h := Te h : Te G → Te′ G′ .
Any further Lie group homomorphism h′ : G′ → G′′ satisfies Lie h′ ◦ h = Lie h′ ◦ Lie h.

Note, that – according to result of Cartan – a continuous homomorphism between

Lie groups is already smooth.
Similar to the exponential series eX for matrices X, in the case of an abstract Lie
group G, there is the exponential map

exp : Lie G → G , X 7→ exp X , X ∈ Te G .

To define exp X for X ∈ Te G, we start with a solution of the autonomous differ-

ential equation γ̇ = X̃(γ), through e, γ(0) = e. Such a curve exists because of the
existence and uniqueness theorem for ordinary differential equations. It is, in general,
not assured, that such a solution curve can be defined on all R. However, for left-
invariant vector fields X̃ on a Lie group this can be done because of the invariance, as
we show in the following:
C.4 The Lie Algebra of a Lie Group 313

First, let the curve γ be defined only on ] − ε, ε[ with γ̇ = X̃(γ) and γ(0) = e. For
any g ∈ G the curve φg (t) := gγ(t) , t ∈] − ε, ε[, satisfies φ̇g = X̃(ϕg ). In fact, using
γ̇(t) = [γ(t + s)]γ(t) , we have:


φ̇g (t) = [gγ(t + s)]γ(t) = Tγ(t) Lg [γ(t + s)]γ(t) = Tγ(t) Lg (γ̇(t)) = Tγ(t) Lg (X̃(γ(t))


= Tγ(t) Lg ◦ Te Lγ(t) (X) = Te Lg ◦ Lγ(t) (X) = Te Lgγ(t) (X) = X̃(gγ(t)) = X̃ (φg (t)) .

Thus, we can define for g := γ 12 ε :


γ(t) for −
ε < t < ε
γ1 (t) :=
φg t − 12 ε für − ε + 21 ε < t < ε + 12 ε

γ1 is a well-defined smooth curve since on ] − 12 ε, ε[ the values γ(t) and φg (t) =

γ( 12 ε)γ(t− 12 ε) = γ(t) agree. Hence γ1 is solution of γ̇ = X̃(γ) on the interval ]−ε, ε+ 12 ε[.
By repetition of the argument γ can be extended as a solution to all of R. This solution
corresponds to the exponential series etX and is denoted by exp tX.
In particular, exp X is a well-defined group element exp X ∈ G and determines
the so-called Exponential Map exp : g → G. exp is smooth and locally invertible.
As a consequence, exp provides local charts for the differentiable structure of G. In
particular, there is a neighbourhood U ⊂ G of e ∈ G and a neighbourhood V ⊂ Lie G
such that the restriction exp |V : V → U is diffeomorphism: Near e the Lie group looks
like the flat neighbourhood in the linear space Lie G, in other words, at U the Lie group
appears infinitesimally as V (modulo exp).
We have the following relation between the exponential mapping and the induced
Lie algebra homomorphism:

Proposition C.20. A Lie group homomorphism h : G → G′ satisfies:

h ◦ exp = exp ◦Lie h ,

in other words, the following diagram is commutative:

GO
h / G′
O
exp exp

g / g′
Lie h

We conclude this section with

Proposition C.21. The tangent bundle T G of a Lie group is trivial.

Proof. In fact, the map G × g → T G , (g, X) 7→ X̃(g), is an isomorphism of vector

bundles.
314 C. Lie Theory

C.5 Lie Group Action

Definition C.22. A (left) action of a group G on a manifold M is a map Ψ : G×M →

M such that the maps Ψg : M → M , Ψg (a) := Ψ(g, a) = ga, satisfy

Ψg ◦ Ψh = Ψgh , Ψe = idM , g, h ∈ G .

When G is a Lie group, such a group action Ψ is called a Lie Group Action if Ψ is
differentiable.

Equivalently, a Lie group action of G on M consists of a group homomorphism

Ψ : G → Diff(M ), such that the induced map G × M → M is smooth. A smooth
manifold endowed with a Lie group action of G is also called a G-manifold.
In an analogous way one defines the notion of a right Lie group action, which is
used, for instance, in the context of principal fibre bundles (see Section D.2).
The left action of a Lie group will often be denoted as ga instead of Ψ(g, a) = Ψg (a).

Definition C.23. Let Ψ : G × M → M be a Lie group action. The Isotropy Group

at a ∈ M is the subgroup Ga := {g ∈ G | ga = a}. The Orbit through a ∈ M is
Ma := {ga | g ∈ G}.
The action Ψ is said to be

1. Transitive if for each pair (a, b) ∈ M × M there exists g ∈ G with ga = b.

i.e. all orbits Ma are M = Ma .

2. Effective (or faithful) if Ψg = idM implies g = e, i.e. if Ψ : G → Diff(M ) is

injective.

3. Free if all Ψg , g ∈ G \ {e}, have no fixed points, i.e. Ψg (a) ̸= a for all a ∈ M .
Equivalently, for all a ∈ M the isotropy groups Ga are trivial.

4. Proper if Ψ̃ : G × M → M × M , (g, a) 7→ (a, ga) is a proper mapping, that

is, the inverse images Ψ̃−1 (K) of compact subsets K ⊂ M × M are compact.
Equivalently, if for all sequences (gn ) inG, (pn ) ∈ M such that (pn ) and gn pn )
converge the sequence (gn ) has a convergent subsequence.

Note, that the isotropy groups are closed subgroups of G, hence they are Lie groups.

Examples C.24. Let G be a Lie group and let M be a manifold. The following are
examples of Lie group actions of g on M :
1. The trivial action Ψg = idM , g ∈ G.
2. The action of G on itself by left multiplication, right multiplication or by conju-
gation. The action is transitive and free.
C.5 Lie Group Action 315

3. The action of a given Lie subgroup H ⊂ G on G by left multiplication or by

conjugation. For instance, U(1) acting on SU(2).
4. The action of a matrix Lie group G ⊂ GL(n, K) on Kn .
5. The action of the group R on M given by the flow Φ of a complete vector field
on M .
6. The action of the multiplicative group K× on V \ {0} for a K-vector space V
(K ∈ {R, C}).
7. The adjoint action of G on its Lie algebra g. Similarly, the coadjoint action on
g∗ .
8. The group action of G on a principal fiber bundle π : P → M over a manifold
M (cf. Section D.2). Here, the action is a right action of the Lie group G on P . The
isotropy groups are trivial and the orbits are the fibres Pa = π −1 (a) , a ∈ M .
9. Hamiltonian group action on a symplectic manifold.

The following important theorem completes the discussion on quotient manifolds

in Section A.4. A proof can be found e.g. in [RuS13].

Theorem C.25. The orbit space M/G of a proper and free Lie group action exists as
a quotient manifold and the quotient map π : M → M/G is a submersion.

Examples C.26. 1. P n (R) ∼

= Rn+1 /R× . P n (C) ∼
= Cn+1 /C× . P n (C) ∼
= S n+1 /U(1).
2. S2 ∼
= SU(2)/U(1), the Hopf fibration.
3. Let G be a compact Lie group. For every µ ∈ g∗ the coadjoint orbit satisfies
G/Gµ ∼ = Mµ . The theorem applies since the action of the isotropy group Gµ on G
action is free and proper.
4. In the case of a principal fibre bundle π : P → M with structure group G: the
orbit space P/G is isomorphic to M .
5. For a closed subgroup H of a Lie group G the right multiplication is free and
proper. The quotient G → G/H yields a principal fibre bundle with structure group
H.
6. Another important general example is the associated fibre bundle P ×G F to a
principal fibre bundle P (see Section D.3). In this situation the Lie group G acts from
the right on the principal fibre bundle P and from the left on the fibre F . These actions
induce on P × F the right action ((p, x), g) 7→ (pg, g −1 x), which turns out to be free
and proper, as shown in Section D.3. By the theorem the orbit space (P × F )/G exists
as a manifold. The orbit space is the associated fibre bundle P ×G F ∼ = (P × F )/G.
316 D. Fibre Bundles

D Fibre Bundles

This section adds some concepts generalizing line bundles. The aim is to present line
bundles and connections within the framework of vector bundles and principal fibre
bundles with their associated fibre bundles and to make them available for the later
chapters in these notes beginning with the chapter on half-density quantization. The
line bundles are special vector bundles, namely those with 1-dimensional fibres. Also,
vector bundles are in close connection with principal fibre bundles. To discover these
correlations we describe the relationship between vector bundles, principal fibre bundles
and their associated bundles.

D.1 Vector Bundles

Definition D.1. A vector bundle of rank r ∈ N, r ≥ 1, over a manifold M , is

given by a total space E and a (smooth) map π : E → M such that:

1. For each a ∈ M , Ea = π −1 (a) is an r-dimensional vector space over K (where K

is R or C),
2. E is locally trivial, i.e. there exists an open cover (Uj )j∈I of M and diffeomor-
phisms
ψj : EUj := E|π−1 (Uj ) → Uj × Kr
with
(a) the diagram
ψj |EU
EUj /j Uj × Kr
π|π−1 (U
j) pr1
 x
Uj
is commutative: pr1 ◦ ψj |EUj = π|π−1 (Uj ) 113
(b) For all b ∈ Uj , the following induced map
ψj |E pr2
(ψj )b : Eb −→b {b} × Kr −→ Kr
is a homomorphism (in fact an isomorphism) of vector spaces over K.

As before with line bundles, a vector bundle π : E → M is determined by transition

functions (gjk )j,k∈I with respect to any open cover (Uj )j∈I . The gjk are defined by the
trivializations (ψj ) through the condition
ψj ◦ ψk−1 (a, y) = (a, gjk (a).y), (a, y) ∈ Ujk × Kr ,
113
pr1 , pr2 denote the natural projections pr1 : W × V → W , (x, y) 7→ x resp. pr2 : W × V →
V , (x, y) 7→ y for a product W × V .
D.2 Principal Fibre Bundles and Frame Bundles 317

where gjk (a).y stands for applying the matrix

gjk (a) := (ψj )a ◦ (ψk )−1

a ∈ GL(r, K)

to the vector y ∈ Kr . Note, that in the case of r > 1 the gjk have their values in the
group G = GL(r, K) instead of K× = GL(1, K) = K \ {0} in the case of line bundles:

gjk ∈ E(Ujk , G).

The transition functions satisfy the cocycle condition:

(C) gij (a) · gjk (a) · gki (a) = idKr

for a ∈ Uijk := Ui ∩ Uj ∩ Uk , where ”·” denotes matrix multiplication.

Conversely, every cocycle (gjk ) , gjk ∈ E(Ujk , GL(r, K)), yields a vector bundle of
rank r. This can be proven in the same way as for the case r = 1, see Proposition 3.9.
Observation D.2. The direct sum, the tensor product, taking the dual, the space of
endomorphisms of vector spaces carries over to the vector bundles. Thus, we obtain
from two vector bundles E and F the bundles
k
^
∨
E ⊕ F, E ⊗ F, E , End (E, F ), E, ...

Here E ∨ is the dual bundle of E with fibre the dual (Ea )∗ = Hom K (Ea , K) of the fibre
Ea of E at a ∈ M . In particular, if (gjk ), resp. (hjk ) is a cocycle corresponding to E,
resp. F , then
−1
(gjk ) is the cocycle for E ∨ ,
(gjk ⊕ hjk ) is the cocycle for E ⊕ F ,
(gjk ⊗ hjk ) is the cocycle for E ⊗ F ,
−1


gjk ⊗ hjk is the cocycle for End (E, F ),
(gjk ∧ . . . ∧ gjk ) is the cocycle for k E.
V

In the case of a complex manifold M and K = C one also studies holomorphic

vector bundles, where all occurring maps are holomorphic.

D.2 Principal Fibre Bundles and Frame Bundles

Closely related to vector bundles are the principal fibre bundles with structure group
being a Lie group G. Recall, that a right Lie group action of a group G on a manifold
P is a (smooth) map Ψ : P × G → P , such that with the notation

pg := Ψ(p, g), p ∈ P, g ∈ G,
318 D. Fibre Bundles

the associativity on the right

(pg)h = p(gh) , h ∈ G,

holds. In contrast to the notion of a left action (see Section C.5) the induced maps
Ψg : M → M , p 7→ pg , satisfy the opposite of the homomorphism property, namely

Ψg ◦ Ψh = Ψhg , g, h ∈ G .

On a product P := M × G one has the natural right action, also called trivial
action:
P × G → P , ((a, g), h) 7→ (a, gh) , a ∈ M , g, h ∈ G.

Such a product bundle is the local version of a principal fibre bundle with structure
group G. In general:

Definition D.3. A Principal Fibre Bundle (”Hauptfaserbündel”) with Struc-

ture Group (”Strukturgruppe”) G is a manifold P (the Total Space (”Total-
raum”) together with a smooth projection map π : P → M and a differentiable Right
Action Ψ : P × G → P , with the following properties:

1◦ The action respects the projection, i.e. for all (p, g) ∈ P ×G one has π(pg) = π(p),
and the action G → Pa := π −1 (a), g → pg, is a diffeomorphism for each p ∈ Pa ,
and

2◦ there is an open cover (Uj ) of M with Local Trivializations

ψj : PUj := π −1 (Uj ) → Uj × G

satisfying π|PUj := pr1 ◦ ψj and

ψj (pg) = ψj (p)g, g ∈ G, p ∈ PUj .

In particular, with ψj (p) = (a, h): ψj (pg) = (a, h)g = (a, hg).

Similar to vector bundles, principal fibre bundles are determined by transition func-
tions (gjk ) , gjk ∈ E(Ujk , G), with respect to an open cover (Uj ) of M . They are defined
by the analogous condition as above:

ψj ◦ ψk−1 (a, h) = (a, gjk (a).h)) , (a, h) ∈ Ujk × G.

The condition implies gjk (a) := pr2 ◦ ψj ◦ ψk−1 (a, 1) ∈ G, where 1 is the unit in the
group G and with this definition we confirm

ψj ◦ψk−1 (a, h) = ψj ◦ψk−1 (a, 1)h = (a, pr2 ◦ψj ◦ψk−1 (a, 1))h = (a, gjk (a))h = (a, gjk (a).h).
D.2 Principal Fibre Bundles and Frame Bundles 319

As before, the cocycle condition is satisfied for the transition functions gjk : Ujk →
G:
(C) gij (a) · gjk (a) · gki (a) = 1

for a ∈ Uijk , where 1 is the unit in the group G and the ”·” denotes the group multi-
plication, but · mostly is omitted.
Again, similar to the statement in Proposition 3.9 any cocycle with values in G
yields a principal fibre bundle.

Construction D.4 (Frame Bundle). A vector bundle π : E → M induces a principal

fibre bundle R(E) → M with structure group being the general linear group G =
GL(r, K), the so called Frame Bundle (”Reperbündel”) R(E) of E → M .
In the case of a complex line bundle π : L → M over M the corresponding principal
fibre bundle (which is used in Section 4.2) can be obtained simply by deleting the
zero section: The map, a → 0a , where 0a is the zero element in La defines a section
z : M → L the so called zero section. Let L× := L \ z(M ). Then the restriction of π
to L× defines a principal fibre bundle L× over M with structure group C× = GL(1, C).
L× has the ”same” transition functions as L.
In the general case the frame bundle R(E) of E → M can be constructed as
follows: The total space R(E) is fibrewise S the set Ra (E) of all ordered vector space
bases b = (b1 , . . . , br ) of Ea : R(E) := a∈M Ra (E). For g ∈ G = GL(r, K) one defines
the right action of G on R(E) by

(bg)α := gαβ bβ , (1 ≤ ρ, σ ≤ r).

Then bg = ((bg)1 , . . . , (bg)r ) ∈ Ra (E) with (bg)h = b(gh) for g, h ∈ GL(r, K), and
using elementary linear algebra, we see that the map G → Ra (E), g 7→ bg, is bijective.
R(E) obtains its topological and differential structure from the local trivializations

ψ : E U → U × Kr

of the vector bundle E, as we see in the following:

To each a ∈ U there corresponds a special basis ê(a) of Ea depending on ψ: ê(a) =
êψ := (ψ −1 (a, e1 ), . . . , ψ −1 (a, er )), where (e1 , . . . , er ) ∈ (Kr )r is the standard basis of
Kr , defined by eσ = (δσρ ). Now, for every b ∈ Ra (E) , a ∈ U, there exists a unique
matrixψ̂(b) ∈ GL(r, K) such that

b = eˆψ (a)ψ̂(b), .

This construction leads to the definition of the map

 
ψ R : R(E)U → U × G , b 7→ π(b), ψ̂(b) .
320 D. Fibre Bundles

ψ R is bijective with inverse

−1
ψR (a, g) = ê(a)g , (a, g) ∈ U × G.
In particular,
   
ψ (bh) = π(ah), ψ̂(bh) = π(a), ψ̂(b)h = ψ R (b)h.
R

Finally, the topology and the differentiable structure on R(E)U (and hence on R(E))
will be defined by requiring that all these ψ R are diffeomorphisms.
This construction immediately yields that the transition functions of E are also
transition functions of R(E): Let gjk be transition functions for E coming from local
trivializations ψj : EUj → Uj × Kr . Then
ψjR ◦ (ψkR )−1 (a, 1) = ψjR (êψk (a)) = (a, ψ̂j (êψk (a)) ,
R
which imples that gjk := ψ̂j (êψk ) ∈ E(Ujk are transition functions for R(E). More-
over, because of êψj (a)ψ̂j (êψk (a)) = êψk (a) and êψj (a)gjk (a) = êψk (a) it follows
R
gjk = ψ̂j (êψk ) = gjk . Thus, up to isomorphism, R(E) can as well be defined as the
principal fibre bundle given by (gjk ) directly.
In the case of a holomorphic vector bundle, the transition functions are holomorphic
(note, that GL(r, C) is a complex Lie group). Therefore, R(E) can be endowed with a
complex structure and thus becomes a holomorphic principal fibre bundle.
Definition D.5. It is evident how to define homomorphisms of vector bundles over M
and similarly morphisms (also called homomorphisms) of principal fibre bundles: For
two such bundles π : P → M and π ′ : P ′ → M with structure group G a morphism114
is a smooth map
Θ : P → P′
respecting the projections and right actions on P, P ′ , i.e. π ′ ◦ Θ = π, described by the
commutative diagram
Θ /
P P′
π π′
 
M / M,
idM

and ΘΨ(p, g) = Ψ′ (Θ(p)g), or Θ(pg) = Θ(p)g , (p, g) ∈ P × G, described by the

commutative commutative diagram

P ×G
Ψ / P
Θ×idG Θ
 
Ψ′ / P′ .
P′ × G
The last property is sometimes called equivariance.
114
More general definitions are possible, for instance allowing G to change or M to change
D.2 Principal Fibre Bundles and Frame Bundles 321

Such a morphism of principal bundles determines, as in the case of line bundles

over M , local maps hj ∈ E(Uj , G) satisfying
′
(I) gjk = hj gjk h−1
k .
Vice versa, a family (hj ) with (I) determines a morphism P → P ′ .
Example D.6. In particular, a principal fibre bundle P over M is called to be trivial,
if there exists a morphism
Θ : P → M × G,
which has an inverse Θ−1 as a morphism (which means here that Θ is a diffeomorphism)
and thus establishing an isomorphism of principal fibre bundles. Such a morphism Θ is
given by a map θ : M → G satisfying θ(pg) = θ(p)g such that Θ(p) = (π(p), θ(p)) , p ∈
P.
When transition functions gjk for P are given, then P is trivial if and only if there
exist local functions hj ∈ E(M, G) such that
gjk = hj hk −1 .
In the context of bundles, one studies the sections of the bundles. In case of a
vector bundle π : E → M the set of sections over an open U ⊂ M ,
Γ(U, E) := {s ∈ E(U, E) | π ◦ s = idU }
is an E(U )-module in a natural way. For a principal fibre bundle π : P → M the set
of sections over an open U ⊂ M ,
Γ(U, P ) := {σ ∈ E(U, P ) | π ◦ σ = idU }
comes with a natural right G-action
Γ(U, P ) × G → Γ(U, P ), (σ, g) 7→ σg,
where σg(a) := σ(a)g.
As an example, the
ê : U → R(E)
in the construction of the frame bundle above is a section.
Note, that a principal fibre bundle π : P → M admitting a global section σ ∈
Γ(M, P ) is already trivial:
Θσ : P → M × G , σ(a)g 7→ (a, g)
defines a trivialization with inverse
Θ−1
σ : M × G → P , (a, g) 7→ σ(a)g .

In this way, every local section σ : U → P provides a local trivialization: Θσ :

PU → U × G
Θσ (p) = (π(p), θ(p)), p ∈ PU ,
where θ(p) ∈ G is the unique group element with p = σ(π(p))θ(b) , and where θ ∈
E(PU , G) satisfies θ(pg) = θ(p)g.
322 D. Fibre Bundles

D.3 Associated bundles

How to come back from principal fibre bundles to vector bundles? Consider a principal
fibre bundle π : P → M with structure group G and let F be a manifold (the general
fibre) with a differentiable left action on F by G, denoted
Λ : G × F → F , (g, x) 7→ gx := Λ(x).
(cf. Section C.5 for the notion of a Lie group action). Being a Lie group action includes
that the associativity in the following form holds: h(gx) = (hg)x , g, h ∈ G, x ∈ F .
We define P ×G F := P × F/ ∼, where the equivalence relation is
(p, x) ∼ (p′ , x′ ) ⇐⇒ ∃g ∈ G : (p′ , x′ ) = pg, g −1 x .

Note, that the equivalence classes are the orbits of the following right action on
P × F induced by the left action Λ and the given right action on the principal fibre
bundle P :
(P × F ) × G → P × F , ((p, x), g) 7→ (pg, g −1 x) .

The result is a fibre bundle P ×G F → M

• with the projection πF : P ×G F → M , [(p, x)] → π(p),
• with general fibre F , where for a ∈ M and p0 ∈ Pa the map [(p0 , x)] 7→ x , x ∈ F
is a diffeomorphism Fa := πF−1 (a) ∼
= F,
• with structure group G,
• and with the right action (P ×G F ) × G → P ×G P where ([(p, x)], g) → [(p, g −1 x)]
for [(p, x)] ∈ P ×G F , g ∈ G.
P ×G F is called the Associated Fibre Bundle.
We have to ensure that the quotient exists as a differentiable manifold. This can
be done by applying the result on the existence of the orbits space X/G of a Lie group
action on a manifold X (cf. Theorem C.25): Since the quotient P ×G F := P × F/ ∼ is
the orbit space (P × F )/G with respect to the above mentioned right Lie group action
of G on P × F it is enough, according to the theorem, to check that the action Φ is
free and proper.
Another argumentation to show that the quotient manifold exists is to use the local
trivializations of P → M to verify that they yield local trivializations of πF : P ×G F →
M and that the trivializations glue together to obtain the desired manifold structure
of the quotient. Yet another proof uses the transition functions induced by the original
bundle π : P → M and the Lie group action on F , see Proposition D.9 below.
The sections s : U → P ×G F of the associated bundle are defined as before: s is
a section if s is smooth and π ◦ s = idU . Let Γ(U, P ×G F ) denote the E(U )-module
of sections of P ×G F over U . The sections can be described by functions on PU → F
with an equivariance or invariance property:
D.3 Associated bundles 323

Proposition D.7. Γ(U, P ×G F ) is isomorphic to the E(U )-module

EG (PU , F ) := {f ∈ E(PU , F ) | ∀g ∈ G : f (pg) = g −1 f (p)} .

Proof. For f ∈ EG (PU , F ) we define sf (a) := [(p, f (p))] , a ∈ U , where π(p) = a. sf (a)
is well-defined, since for another p′ ∈ Pa there exists a unique g ∈ G with p′ = pg.
Therefore, (p′ , f (p′ )) = (pg, f (pg)) = (pg, g −1 f (p), by the invariance property of f ,
and consequently (p′ , f (p′ )) ∼ (p, f (p)). It is easy to see that sf is a section, and
EG (PU , F ) → Γ(U, P ×G F ) , f 7→ sf , is linear over E(U ) and injective. Finally the
surjectivity follows from the inverse construction: a section s : U → P ×G F , s(a) =
[(p, x)] determines a map s♯ (p) := x with s♯ (pg) = g −1 x = g −1 s♯ (p). This s♯ is well-
defined and satisfies s♯ ∈ EG (PU , F ) with ss♯ = s.

Associated Vector Bundle

We now concentrate on the special case of a vector space F = Cr as the general
fibre and on the left action on F given by a Lie group representation ρ : G → GL(r, C),
i.e. ρ is a smooth homomorphism. The induced left action on F = Cr is

gx = ρ(g)x , x ∈ Cr , g ∈ G .

ρ(g)x defines indeed a left action: (hg)x = ρ(hg)x = ρ(h)ρ(g)x = ρ(h)(gx) = h(gx).
On the basis of such a representation one obtains a vector bundle P ×G Cr = P ×ρ Cr
of rank r, also denoted by Eρ . The last result leads to the following.

Corollary D.8. Γ(M, Eρ ) is isomorphic to the E(U )-module

Eρ (P, F ) := {f ∈ E(P, F ) | ∀g ∈ G : f (pg) = ρ(g)−1 f (p)} .

Let us compare the transition functions gjk ∈ E(Ujk , G) of the principal fibre bundle
π : P → M with structure group G and the corresponding transition functions of the
associated vector bundle Eρ induced by a representation ρ : G → GL(r, C).
The result is: ρ(gjk ) ∈ E(Ujk , GL(r, C)) can serve as the transition functions of Eρ
and a corresponding result holds for the case of a general associated fibre bundle. We
describe this in some detail.

Proposition D.9. Let Λ : G × F → F be a left action on a manifold F in the form

of the induced map ρ : G → Diff(F ) (where Diff(F ) is the group of diffeomorphisms
F → F ) given by
ρ(g)(x) := Λ(g, x) = gx , (g, x) ∈ G × F.
Then ρ is a group homomorphism. Moreover, the associated bundle P ×G F has as
F
suitable transition functions gjk the functions ρ(gjk ) : Ujk → Diff(F ) where gjk ∈
E(Ujk , G) are transition functions of P .
324 D. Fibre Bundles

Proof. In order to show this result we compare in the situation of a local trivialization
ψ : PU → U × G of P over an open U ⊂ M the quotients
PU × F → PU ×G F and η F : (U × G) × F → (U × G) ×G F.
Because of
η F ((a, h), x) := [((a, h), x)] = {((a, hg), ρ(g −1 )x) : g ∈ G},
for (a, h) ∈ U × G and x ∈ F this equivalence class has a unique representative of the
form ((a, 1), ρ(h)x) ∈ (U × G) × F . Using this, we identify (U × G) ×G F with U × F
and note that the quotient map now is
η F : (U × G) × F → U × F , ((a, h), x) 7→ (a, ρ(h)x).
As a result, a suitable trivialization of PU ×G F is
ψ F : PU ×G F → U ×G F , [(p, x)] 7→ (π(p), ρ(h(p))x),
if ψ(p) = (π(p), h(p)) , h : PU → G, p ∈ PU . Now, we conclude

−1  −1
ρFj ◦ ρFk (a, x) = ρFj = a, ρ ψj ◦ ψk−1 (a, 1)x .





ρk (a, 1), x
This equality yields the desired result
gijF = ρ(gij ) : Uij → Diff(F ),
respectively
ρ(gij ) : Uij → GL(r, C)
in the case of ρ : G → GL(r, C).

We finally present an example of the process of creating vector bundles as associated

fibre bundles induced by a representation – an example which we already know.
Example D.10. M is the projective space Pn (C) of complex dimension n. On M we
consider the dual H = T ∨ of the tautological line bundle T → M , i.e. the hyperplane
bundle. H is determined by the transition functions
zk
gjk (z0 : · · · : zn ) = , (z0 : · · · : zn ) ∈ Ujk = Uj ∩ Uk
zj
with respect to the homogeneous coordinates (z0 : · · · : zn ).
The corresponding principal fibre bundle of H, the frame bundle of H, is H × → M
with structure group C× = GL(1, C). This bundle is determined by the same transition
functions gjk .
Now, to each representation ρm : C× → GL(1, C), z 7→ z m , where m ∈ Z, we obtain
the associated vector bundle Eρm , which is a line bundle with the transition functions
 m
m zk
ρm (gjk ) = gjk =
zj
according to our proposition. Hence, for m ∈ Z the associated line bundle Eρm is
equivalent to our previously defined line bundle H(m) = H ⊗m , see Construction 3.18
in Section 3.3.
D.4 Principal Connection 325

D.4 Principal Connection

In Chapter 4 we present at least 5 different ways to define the concept of a connection

on a line bundle L → M : In form of a collection of covariant derivatives ∇X in 4.1,
through local connection forms αj on the base manifold M in 4.3, through a global
connection form α on the frame bundle L× in 4.7, through a horizontal subbundle
H of T L× in 4.8, and through a vertical projection on the tangent bundle T L× in
4.9. We now introduce connections in principal fibre bundles. In this way, we can
regard connections on line bundles in the framework of general connections. Some
parts become more complicated, but others look simpler in the general case.
Let π : P → M be a principal fibre bundle with structure group G (as in Section
D.2). G is a Lie group and we restrict, for simplicity to matrix groups, i.e. to closed
subgroups of the general linear group GL(m, C) for some m ∈ N. We have a right
action of G on P , that is a differentiable map

Ψ : P × G → P , (p, g) 7−→ Ψ(p, g),

(mostly written in the form pg := Ψ(p, g)) satisfying pe = p (e ∈ G is the unit)

and (pg)h = p(gh) for all g, h ∈ G and p ∈ P . The action is compatible with π,
i.e. π(pg) = π(p) for all g ∈ G and p ∈ P , and it is free, meaning that the map
G ∋ g 7−→ pg ∈ P is a diffeomorphism onto the fibre Pa = π̄ −1 (a) , a = π(p). Most
importantly, the projection π : P → G, is locally trivial: Each a ∈ M has an open
neighbourhood U ⊂ M with a diffeomorphism

ψ : PU = π̄ −1 (U ) → U × G

which respects π and the right action: pr1 ◦ ψ = π|PU and ψ(pg) = ψ(p)g for all p ∈ PU
and g ∈ G - where the action on U × G is the standard right action: (a, h)g = (a, hg).
Finally, let n denote the dimension of M and k the dimension of G.
Let g = Lie G denote the Lie algebra of G and exp : g → G the exponential map.

Definition D.11. The Fundamental Field associated to X ∈ g is the vector field

X̃ ∈ V(P ) given by

d
X(p)
e := (p exp(tX))|t=0 = [p exp(tX)]p ∈ Tp P.
dt

Compare with the case of C× = G and C = Lie G : X = ξ ∈ C, 2πiξ˜ = Yξ (Definition

4.5).

For each point p ∈ P the tangent vector X(p)

e ∈ Tp P points in the direction of the
fibre Pa , a = π(p), which is the same as to say that X(p)
e ∈ Ker Tp π ⊂ Tp P :

Tp π(X(p))
e = [π(p exp(Xt))]a = 0
326 D. Fibre Bundles

since π(p exp(Xt)) = π(p) = a is constant.

For each a ∈ M and p ∈ Pa the tangent space to the fibre Pa at p agrees with the
kernel of Tp π:
Tp (Pa ) = Ker Tp π.
We call V := Ker T π ⊂ T P the Vertical Bundle. The inclusion V ⊂ T P induces
the structure of a real vector bundle on V of (real) dimension k := dim G = dimR Tp Pa ,
where the projection V → P of the vector bundle V is the restriction τ |V of the
projection of the tangent bundle τ : T P → P .
Lemma D.12. The fibres of V are isomorphic to g, where G is considered as a real
vector space. The isomorphism is given by

g → Tp Pa , X 7−→ X(p)
e ,

with a = π(p). In particular,

Vp = Ker Tp π = Tp Pa = {X(p)
e : X ∈ g} ,

Proof. For each p ∈ P the map

g → Tp Pa , X 7−→ X(p)
e

is R-linear and injective, since X(p)

e = 0 means p exp(Xt) is constant, hence X = 0.
Its image is all of Tp Pa because of dimR g = dimR Tp Pa . Hence X 7−→ X(p)
e is an
isomorphism. The other equalities have been shown in the considerations before.

As a result, X 7→ X(p)
e has an inverse σp : Vp → g and the induced map

σ : V → P × g , σ(v) := (τ (v), σp (v)) , v ∈ V,

turns out to be a diffeomorphism with pr1 ◦ σ = τ and σp : Vp → {p} × g linear.

Therefore, σ : V → P × g ∼= P × Rk is a vector bundle isomorphism. Altogether, we
have shown
Proposition D.13. The vertical bundle V ⊂ T P is a trivial vector subbundle of T P
of rank k over R.

Principal Connections:
Now, we introduce the concept of a connection on P . Before that, let us extend the
notion of a differential 1-form on P to the vector valued case: A one form with values
in Kr is an E(P )-linear map α : V(P ) → E(P, Kr ) and one denotes the space of these
1-forms by
A1 (P, Kr ) := Hom E(P ) (V(P ), E(P, Kr )) ∼
= A1 (P ) ⊗ Kr .
In the same way we have A1 (P, g) := Hom E(P ) (V(P ), E(P, g)).
D.4 Principal Connection 327

Definition D.14. A (global) Connection Form on P is a one form α ∈ A1 (P, g)

satisfying

(I1) α(X̃) = X for all X ∈ g

(I2) Ψ∗g α = g −1 αg for all g ∈ G

Here, Ψg : P → P is the diffeomorphism induced by the right action Ψ of P :

Ψg (p) = pg , Ψg : p 7−→ pg. g −1 α(X)g is well-defined since for a matrix group G ⊂
GL(m, C) the Lie algebra g = Lie G is a subalgebra g ⊂ Lie GL(m, C) = Cm×m of
the LIe algebra Cm×m of all m × m-matrices. Hence, g −1 α(X)g is simply defined by
matrix multiplication. And every g ∈ G induces a map g → g, X 7→ g −1 Xg (which is
adg−1 : g → g).
Evidently, the conditions (I1), (I2) above agree with the line bundle case (c.f. Propo-
sition 4.7), since c−1 αc = α in the case of c ∈ C× = G.
Given a principal fibre bundle π : P → M with a connection, i.e. a connection
form α ∈ A1 (P, g) with (I1) and (I2) we obtain an associated Horizontal Bundle
H ⊂ T P in the following way:
H := Ker α ⊂ T P,
with the fibres Hp = Ker αp = {Yp ∈ Tp P | αp (Yp ) = 0}. Since αp |Vp : Vp → g is an
isomorphism (by (I1) and Lemma D.12) the dimension of Hp is n = dim M (note, that
the dimension of P is n+k). Therefore, the induced structure from T P yields on H the
structure of a real vector bundle of dimension n. Since Hp ∩ Vp = {0} for all p ∈ P :,
we obtain the decomposition
TP = H ⊕ V
of TP into the direct sum of two real vector subbundles H, V of TP. The action Ψg
induces an isomorphism Tp Ψg : Hp → Hpg for all (p, g) ∈ P × G. We have shown the
first half of the following.
Proposition D.15. A connection form α ∈ A1 (P ) on a principal fibre bundle P → M
defines the horizontal bundle H := Ker α with

(H1) TP = H ⊕ V
(H2) Tp Ψg (Hp ) = Hpg for all (p, g) ∈ P × G

Conversely, any vector subbundle H ⊂ T P satisfying (H1) and (H2) induces a connec-
tion form α ∈ A1 (P ) with H = Ker α.

Proof. To define α using (H1) and (H2) let v : T P → T P be the projection which
fibrewise is the linear projection vp : Tp P → Tp P with Ker vp = Hp and Im vp = Vp . v
is a vector bundle homomorphism. Now, α := σ ◦ v as the map
p 7−→ αp = σp ◦ vp ∈ EndR (Tp P, g) , p ∈ P ,
328 D. Fibre Bundles

induces the one form α ∈ A1 (P, g) through

α(X)(p) := σp ◦ vp (Xp ) ∈ g , X ∈ V(P ) .

Evidently, we have H = Ker α. It remains to show that α is a connection form.
(I1) is immediate, since α(X̃) = σ ◦ v(X̃) = σ(X̃) = X for X ∈ g. To show (I2) let
Y ∈ Hp , i.e. α(Y ) = σ(v(Y )) = 0. Then Ψ∗g α(Y ) = σ ◦ v ◦ T Ψg (Y ) = 0, by (H2).
Consequently Ψ∗g α(Y ) = 0 = g −1 α(Y )g. For X ∈ g we know

Tp Ψg (X̃(p)) = [Ψg (p exp tX)]pg = [pgg −1 exp tXg]pg = [pg exp(g −1 tXg)]pg = g^
−1 Xg(pg).

−1 Xg(pg)) = g^
Therefore, because of vpg (g^ −1 Xg(pg),

−1 Xg(pg)) = g −1 Xg.
Ψ∗g α(X̃)(p) = σpg ◦ vpg (Tp Ψg (X̃(p)) = σpg ◦ vpg (g^
This proves (I2).

Exploiting the preceding proof we obtain another description of a principal connec-

tion (given by a one form α satisfying the above axioms or by a decomposition H ⊕ P
of T P with (H1) and (H2)):
Proposition D.16. A principal connection is also given by a vector bundle homomor-
phism v : T P → T P with the properties:

(V1) v ◦ v = v and ℑ v = V ,
(V2) T Ψg ◦v = v ◦ T Ψg for each g ∈ G.

Proof. With the complementary vector bundle H := Ker v ⊂ T P we have T P = H ⊕V .

And T Ψg (Y ) ∈ Hpg for Y ∈ Hp by (V2), hence (H2).
Conversely, a decomposition T P = H ⊕ V with (H2) immediately yields the pro-
jection v : T P → T P onto V satisfying
T Ψg ◦ v = v ◦ T Ψg .

ln physics, P is called the space of phase factors, and α is the (global) gauge
potential.

We obtain local gauge potentials by pullback: Let s : U → P be a section,

i.e. smooth and s ◦ π = idU (U ⊂ M open). Denote
As := s∗ α ∈ A1 (U, g).
Then As is called a local gauge potential (given by s ). How do these local gauge
potentials fit together?
D.4 Principal Connection 329

Proposition D.17. Given two sections s, s′ : U → P over U ⊂ P the corresponding

local gauge potentials
A = s∗ α, A′ = s′∗ α
satisfy

A′ = gAg −1 + gdg −1

where g(a) ∈ G is the uniquely defined group element with s(a) = s′ (a)g(a).

Proof. Let Y ∈ Ta M be given by the curve γ(t), i.e. Y = [γ]a . Then

 
′∗ d ′
s α(Y )(a) = αs′ (a) s ◦ γ(t) and
dt t=0
d ′ d d
s · g −1 ◦ γ(t) = (s ◦ γ(t)) g −1 ◦ γ(t)



s ◦ γ(t) =
dt t=0 dt t=0 dt t=0
d d
= s ◦ γ(t)g −1 (a) + s(a) · g −1 ◦ γ(t)
dt t=0 dt t=0

With h := g −1 (a) ∈ G, the first term is

d
Ψh ◦ s ◦ γ(t) = Tp Ψh (Ta s(Y )) , p = s(a) .
dt t=0

To analyze the second term dtd s(a) · g −1 ◦ γ(t) t=0 = dtd s′ (a)g(a)g −1 ◦ γ(t) t=0 we
observe that X := dtd g(a)g −1 (γ(t)) t=0 is a tangential vector X ∈ Te G, since
g(a)g −1 (γ(0)) = e = identity, and can be viewed as to be a Lie algebra element X ∈
g∼= Te G. Since g(a)g −1 (γ(t)) is the matrix multiplication: X = g(a) dtd g −1 (γ(t)) t=0 =
g(a)Ta g −1 (Y ). For every curve η(t) in G with η(0) = e and [η]e = X one has

d ′
X̃ (p′ ) = (p η(t)) , p′ ∈ P
dt t=0
 
−1 ′
in particular for η(t) = g(a)g (γ(t)) : According to (H1), we have α p′ X̃ (p ) = X
and we conclude (p′ = s′ (a)) :

1. αp′ dtd s′ (a)g(a)g −1 (γ(t)) t=0 = X = g(a)Ta g −1 (Y ).

Moreover, for the first term Tp Ψh (Ta s(Y )) the second condition reads

αph Tp Ψh (Ta s(Y )) = h−1 αp (Ta s(Y )) h,

and because of p′ = ph, h−1 = g(a) and Aa (Y ) = s∗ α(Y ) = αp (Ta s(Y )) we obtain

2. αp′ (Tp Ψh (Ta s(Y )) = g(a)Aa (Y )g −1 (a)

330 D. Fibre Bundles

Putting everything together we have

   
′ d ′ d
Aa (Y ) = αp′ s γ(t) = αp′ Tp Ψh (Ta s(Y )) + s(a)g −1 γ(t)
dt t=0 dt t=0
−1 −1
= g(a)Aa (Y )g (a) + g(a)Ta G (Y )

As before, local trivalizations (here given by local sections) over Uj with respect to
an an open cover (Uj ) of M yield local gauge potentials Aj with a transition rule and
vice versa. In detail:
Let sj : uj → p smooth sections with its corresponding trivialization φj : Puj →
Uj × G, p 7→ (πp, ŝj (p)), where ŝj (p) ∈ G is given by p = sj (π(p)) ŝj (p). These data
induce transition functions gjk ∈ E (Ujk , G), given by gjk = ŝk ◦ ŝ−1
j . In particular, (gjk )
satisfies (C), and
sj = sk gjk on Ujk ̸= ∅
Let Aj := s∗j α ∈ A1 (Uj , g).

Proposition D.18. Let α be a connection form on the principal fibre bundle P → M .

Define the local gauge potentials Aj and the transition functions gik as above (depending
on the sections sj ). Then
−1 −1
(Z) Ak = gjk Aj gjk + gjk dgjk .

Conversely, a collection (Aj ) of 1-forms with (Z) defines a connection form α whose
local gauge potentials are the Aj .

Note that (Z) is essentially the same as (Z) in the line bundle case above for the
case G = C× since dgg −1 + gdg −1 = 0 in general and gαg −1 = α for g ∈ C× .

D.5 Connection on a Vector Bundle

In our proposition 4.16 we have already anticipated that there is also a general pro-
cedure in which a given connection on the principal fibre bundle L× induces a con-
nection on the line bundle L : One simply uses the globally given connection form
α ∈ A1 (L× ) and local sections sj : Uj → L (where (Uj )j∈I is an open cover ) to obtain
αj := s∗j α ∈ A1 (Uj ) satisfying (Z) and thus defining a connection on L.
This procedure has its generalization to associated vector bundles of a principal
fibre bundle P → M with connection: Let ρ : G → GL(r, C) be a representation of the
of structure group G of P → M . The associated vector bundle Eρ is

Eρ = P ×ρ Cr −→ M
D.5 Connection on a Vector Bundle 331

the quotient manifold of P × Cr with respect to the equivalence relation

(p, v) ∼ (p′ , v ′ ) ⇐⇒ there exits g ∈ G with (p′ , v ′ ) = pg, ρ( g −1 v

Eρ := P × Cρ / ∼ .

Assume that the connection on P → M is given by a connection one form α ∈

A1 (P ). Let (Uj )j∈I be an open cover of M with sections sj : Uj → P , and define αj by
αj := ρ∗ s∗j α ∈ A1 (Uj , Cr ). Here, the representation Lie ρ : g → g (Cr ) = EndC (Cr )

is induced by ρ :
d
Lie ρ(X) := ρ(exp tX) ∈ g (Cr )
dt t=0
And it leads to the definition ρ∗ β(Y ) := Lie ρ(β(Y )) ∈ Cr for a g-valued form β ∈
A1 (U, g) and Y ∈ V(U ).
The collection (αj ) satisfies the compatibility condition
−1 −1
(Zg ) αk = g̃jk αj g̃jk + g̃jk dg̃jk ,

where g̃jk := ρ (gjk ). These (αj ) induce (as in D.14) a covariant derivative

∇X : Γ (U, Eρ ) → Γ (U, Eρ ) , (85)

U ⊂ M open, compatible with restrictions to V ⊂ U and satisfying (K1) and (K2)

with L replaced by Eρ .
Proposition D.19. On every line bundle over a paracompact manifold there exists a
connection.

Proof. We take an open cover (φj )j∈I over which a given line bundle L −→ M is trivial
with trivializations φj : LUj → Yj × C and local sections sj (a) = φ−1 j (a, 1), a ∈ Uj .ωe
(j)
choose βj ∈ A1 (Uj ) and obtain the connections ∇X f sj := (LX f + 2πiβj (x)f ) sj on
LUj , j ∈ I. Let (kj ) be a partition Pof unity subordinate to (Uj ), i.e kj ∈ E(M ),
Supp kj ⊂ Uj , (kj ) locally finite and j∈I kj (a) = 1 for each a ∈ M . Then
X (i)
∇X s := ki ∇X s|Ui ∩U , s ∈ Γ(U, L)
i∈I

defines a connection on L.

The sum of 2 connections ∇, ∇′ on L → M is is general not a connection. The

difference ∇ − ∇′ is a one form on M . In fact, for X ∈ D(M ) and s ∈ Γ(M, L) the
equation
(∇X − ∇′X ) s = β(X)s
defines a value β(X)(a) ∈ C for s(a) ̸= 0. Because of

(∇X − ∇′X ) f s = LX f s + f ∇X s − LX f s − f ∇′X s = f (∇X − ∇′X ) s

332 D. Fibre Bundles

this value is independent of s. Since for every a ∈ M there exists s ∈ Γ(M, L) with
s(a) ̸= 0 we obtain a uniquely defined β(x) ∈ E(M ) such that X 7→ β(X) is E(M )-
linear, and hence β ∈ A1 (M ).

Proposition D.20. Given a fixed connection ∇ on L, every other connection on L

has the form
∇′ = ∇ + β
for an arbitrary β ∈ A1 (M ). The set of connections is the affine space ∇X + A1 (M ).
This set can therefore be understood as an affine subspace of A1 (L× ).

Proof. It only remains to check that ∇X + β, β ∈ A1 (M ), is a connection. The last

statement follows from 4.3.

Isomorphism classes of line bundles with connections are classified by cohomology,

similar to the description of Pic∞ (M ) ∼
= H 1 (M, ε× ), now using (gij ) and (αi ).
 333

E Cohomology

In this chapter, we give a short introduction to Čech cohomology and compare it with
de Rham cohomology. One major objective is to show that these two cohomology
theories on a manifold are equivalent.
It is helpful but not necessary to base everything in terms of sheaves. To support
this aspect, we proceed in a twofold way: We first present Čech cohomology with values
in an abelian group G. In particular, we compare the cases G = R, G = C with de
Rham cohomology. We then extend the situation to sheaves. As a result, we see that
Čech cohomology with values in sheaves is not much more complicated or difficult than
the case of Čech cohomology with values in groups, - except for the notion of a sheaf,
which is a bit involved.
In the following, M will be a topological space, which we assume to be paracompact.
We are mainly interested in the case of a paracompact manifold.
The quite restricted but already interesting case is the following:

E.1 Čech Cohomology with Values in an Abelian Group

Let G be a fixed abelian group. The elements of G are sometimes called coefficients in
the context of cohomology with values in G.
Definition E.1. For open U ⊂ M we define

F(U ) = F(U, G) := {g : U → G | g locally constant} .

In another formulation one has F(U ) = {g : U → G | g continuous} when G is

endowed with the discrete topology (which is the topology where all subsets H ⊂ G
are open).

F(U ) is an abelian group for each open subset U ⊂ M by pointwise multiplication

or addition depending of whether the composition in G is written multiplicatively or
additively. We choose the additive notation in the following. To every inclusion

V ⊂ U of open subsets U, V ⊂ M

there corresponds the natural restriction map

ρV,U : F(U ) → F(V ) , g 7→ g|V .

Lemma E.2. ρV,U is a group homomorphism and for open subsets W, V, U ⊂ M with
W ⊂ V ⊂ U the identities

ρW,V ◦ ρV,U = ρW,U , ρU,U = idF (U )

are satisfied.
334 E. Cohomology

This property resembles the cocycle condition (C) in Section 3.2 for cocycles
(gij )i,j∈I induced by line bundles. Here, with respect to the index set I := {U ⊂
M | U open }.
The case G = R or G = C is particularly interesting for manifolds. It leads to de
Rham cohomology in case M is a manifold. In Section 15.5 and in Section 16.3 we
need the cases G = Z and G = Z2 .
When the abelian group E is a vector space over R, the groups F(U ) = F(U, E)
are R-vector spaces as well and the restrictions ρU,V are linear maps.

Definition E.3. Let U = (Uj )j∈I be an open cover of M .

1. A basic q-simplex σ (q ∈ N) is an ordered (q + 1) tuple (Uj0 , Uj1 , . . . , Ujq ), ji ∈ I,

such that
Uj0 ...jq := Uj0 ∩ Uj1 ∩ · · · ∩ Ujq ̸= ∅ .
|σ| = Ui0 ...iq is the support of the q-simplex σ.

2. Σb (q) denotes the set of such basic q-simplices. The set Σ(q) is the collection of
finite (formal sums) of q-simplices, i.e.
( )
X
Σ(q) := nµ σµ | σµ ∈ Σb (q) , nµ ∈ Z, only finitely many nµ ̸= 0 .
µ

Σ(q) is an abelian group with respect to the obvious addition, namely the so-
called free abelian group Σb (q)(Z) spanned by Σb (q). The finite sums in Σ(q) are
called q-simplices.

3. Let q ∈ N, q > 0. The k-th partial boundary ∂k σ of a q-simplex σ =

(Uj0 , Uj1 , . . . , Ujq ) ∈ Σb (q) , 0 ≤ k ≤ q, is the (q − 1)-simplex

jk , . . . Ujq ),
∂k σ := (Uj0 , . . . , Uc

where Uc
jk means, that the entry Ujk has to be omitted.

4. The boundary ∂σ of σ is
q
X
∂σ := (−1)k ∂k σ.
k=0

The
Pmap b (q) → Σ(q − 1) induces a homomorphism ∂ : Σ(q) → Σ(q − 1) by
∂ : ΣP
∂( n σµ ) := nµ ∂σµ .
µ

We describe q-simplices for low q in some detail:

E.1 Čech Cohomology with Values in an Abelian Group 335

In the case of q = 0 a 0-simplex in Σb (0) is of the form σ = (Uj ) where j ∈ I is a

fixed index, and a general element τ ∈ Σ(0) is a finite sum of the form τ = nµ Ujµ where
nµ ∈ Z, and only finitely many of the nµ are not zero. σ and τ have no boundary.
In the case of q = 1 a 1-simplex in Σb (1) is of the form σ = (Ui , Uj ) or σ = (Uj0 , Uj1 ),
where i, j, j0 , j1 ∈ I. The support of σ is |σ| = Ui ∩ Uj = Uij . The partial boundaries
are ∂0 σ = (Uj ) , ∂1 σ = (Ui ). And the boundary is
∂σ = (Uj ) − (Ui ) , for σ = (Ui , Uj ) .

In the case of q = 2 a 2-simplex in Σb (2) is of the form σ = (Ui , Uj , Uk ) where

i, j, k ∈ I. The support of σ is |σ| = Ui ∩ Uj ∩ Uk = Uijk . The partial boundaries are
∂0 σ = (Uj , Uk ) , ∂1 σ = (Ui , Uk ) , ∂2 (Ui , Uj ). And the boundary is
∂σ = (Uj , Uk ) − (Ui , Uk ) + (Ui , Uj ) , for σ = (Ui , Uj , Uk ) .

In the following we fix an abelian group G.

Definition E.4. Let G be an abelian group and consider the induced groups F(U ) =
F(U, G) of locally constant functions (cf. Definition E.1) with the restriction homo-
morphisms ρV,U : F(U ) → F(V ), V ⊂ U ⊂ M . Moreover, let U be an open cover of
M.
1. A q-cochain of U with coefficients in F (or in G) is a family
η = (ησ )σ∈Σb (q) , ησ ∈ F(|σ|) ,
or a map on Σb (q)
σ 7→ η(σ) = ησ ∈ F(|σ|), σ a q-simplex of U .
C q (U, F) = C q (U, G) denotes the abelian group of q-cochains with pointwise
addition: (η + η ′ )σ := ησ + ησ′ . For a q-simplex σ = (Uj0 , . . . ..., Ujk ) a q-cochain
η = (ησ ) will often be written in the form η = (ηj0 ,...,jk ) with ηj0 ,...,jk := ησ =
η(Uj0 ,......,Ujk ) .
2. The coboundary operator δ = δ q
δ q : C q (U, F) → C q+1 (U, F)
is the homomorphism given by

q+1
X
q
δ η(σ) := (−1)j ρ|σ|,|∂j σ| (η(∂j σ)).
j=0

Here, σ is a (q + 1)-simplex, hence ∂j σ is a q-simplex and η(∂j σ) ∈ F(|∂j σ|).

This element will be restricted to |σ| as ρ|σ|,|∂j σ| to obtain the summands
(−1)j ρ|σ|,|∂j σ| η(∂j σ) in the above formula defining δ q η ∈ C q+1 (U, F).
A q-cochain η with δ q η = 0 is called a q-cocycle.
336 E. Cohomology

It is easy to show that δ is a homomorphism and that δ 2 = 0 i.e. δ q+1 ◦ δ q = 0.

Hence, the following notations make sense:

Z q (U, G) := Ker δ q : C q (U, G) → C q+1 (U, G) ,

B q (U, G) := Im δ q−1 : C q−1 (U, G) → C q (U, G)

Ȟ q (U, G) := Z q (U, G)/B q (U, G) .

As an example, we consider the case q = 1. A 1-cochain η ∈ C 1 (U, G) is given by

η(σ) = η(Ui ,Uj ) ∈ F(Uij ) written as ηij := η(Ui ,Uj ) ∈ F(Uij ). The coboundary of η is

(δ 1 η)ijk = ηjk − ηik + ηij ,

where the restrictions ρ of the ηjk etc. to Uijk have been suppressed. Therefore, if
ηij + ηjk + ηki = 0 for all i, j, k ∈ I, in particular ηii = 0 and ηij = −ηji the cochain
is a cocycle η ∈ Z 1 (U, G) and determines a cohomology class [η] ∈ Ȟ 1 (U, G). Two
such η, η ′ are equivalent (and hence determine the sam class) if and only if there is a
′
0-cochain h = (hj ) with ηjk = ηjk + hk − hj .
The cohomology groups Ȟ q (U, G) are interesting algebraic invariants of the topo-
logical space M . But one is interested to find invariants which do not depend on special
open coverings U of M . To get rid of this dependence one uses direct limits, also called
inductive limits or, in the language of modern category theory, colimits.

Definition E.5.

1. Let (I, ≤) be a directed set. That is, ≤ is a reflexive and transitive partial
order on I such that every pair of elements i, j ∈ I has an upper bound k ∈ I,
i ≤ k , j ≤ k.

2. Let (Xi )i∈I be a family of structured objects, for instance groups, vector spaces,
topological spaces, rings, etc.115 We stick to the case of groups in the following.
Let fij : Xi → Xj be a homomorphism for all i ≤ j with the following properties:

• fii = idXi , and

• fik = fjk ◦ fij for all i ≤ j ≤ k.

fij
Xi / Xj

fik ~ fjk
Xk
Then the pair (Xi , fij ) is called a direct system over I.
115
or, in general, objects in a category
E.1 Čech Cohomology with Values in an Abelian Group 337

3. A direct limit of the direct system (Xi , fij ) is a group X with homomorphisms
fj : Xj → X satisfying fi = fj ◦ fij for i ≤ j, such that the following universal
property is fulfilled: If gj : Xj → Y are homomorphisms with gi = gj ◦ fij for
i ≤ j, then there exists a unique homomorphism f : X → Y such that
f ◦ fj = gj .
The universal property is expressed by the diagram:
fij
Xi / Xj
fi fj

~
gi X gj

f
  
Y

The direct limit of the direct system (Xi , fij ) is unique up to isomorphism. It is
denoted by lim Xi if it exists. In the case of groups Xj it always F exists and can be
−→
constructed as follows. One starts with the disjoint union Xi ’s i Xi , and defines the
following equivalence relation: For xi ∈ Xi and xj ∈ Xj , xi ∼Fxj if there is k ∈ I with
i ≤ k and j ≤ k for which fik (xi ) = fjk (xj ) . Now, lim Xi := i Xi / ∼ and
−→
fi : Xi → lim Xi , xi 7→ [xi ]
−→
sending each element to its equivalence class. The algebraic operations on lim Xi are
−→
defined such that these maps become homomorphisms.
In order to obtain cohomology groups, which are independent of open coverings
U of M , we use the concept of a direct limit in the following way: The direct set is
Co(M ) := {U | U open cover of M } with the following partial order relation: For
two open covers U = (Ui )i∈I and V = (Vk )k∈K of M we set U ≤ V if for each k ∈ K
there exists a j(k) ∈ I with Vk ⊂ Uj(k) . Then V is called a refinement of U and
j : K → I is a refinement map. The refinement map induces a homomorphism
j̃ : C p (U, G) → C p (V, G)
for each q ∈ N by restriction: To σ = (Vk0 , . . . , Vkq ) there corresponds σ ′ =
(Uj(k0 ) , . . . Uj(kq ) ) and
j̃(η)(σ) := ρ||σ|,|σ′ | η(σ ′ ).
Because of j̃(Z q (U, G)) ⊂ Z q (V, G) and j̃(B q (U, G)) ⊂ B q (V, G) the restriction map
j̃ descends to a homomorphism
j̃ := jUV : H q (U, G) → H q (V, G).


In this way we obtain a direct system of abelian groups Ȟ q (U, G) , jUV indexed by
CoM . The direct limit of this direct system is - by definition - the p-th Čech coho-
mology group:
338 E. Cohomology

Definition E.6.
Ȟ q (M, G) := lim Ȟ q (U, G).
−→
U

is the q-th Čech Cohomology Group on M with values in the abelian group G.

Remark E.7. Fortunately, one can avoid to consider the limit in some concrete cal-
culations. This is the case if there exist open covers U of M such that

Ȟ q (M, G) = Ȟ q (U, G) (independently of U).

More precisely, if the canonical homomorphism Ȟ q (U, F) → Ȟ q (M, F) is an isomor-

phism. Such open covers U are called Leray covers.

E.2 DeRham Cohomology

Let us now come to the comparison of Ȟ q (U, R) (and Ȟ q (M, R)) with the de Rham
q
cohomology groups HdR (M, R) . Recall:
q
HdR (M, R) := {α ∈ Ap (M ) | dα = 0}/{α ∈ Ap (M ) | ∃β : dβ = α}.

Proposition E.8. Let U = (Ui )i∈I be an open cover of a smooth manifold M such that
all intersections Ui0 i1 ...ip := Ui0 ∩ Ui1 ∩ · · · ∩ Uip are empty or diffeomorphic to a convex
open subset of Rn . Then there exists a natural isomorphism
q
HdR (M, R) → Ȟ q (U, R) , q ∈ N.

In particular, U is a Leray cover, and

q
HdR (M, R) ∼
= Ȟ q (M, R) , q ∈ N.

Proof. Note that there always exist such open covers for a manifold. First of all we
prove the isomorphism in the case q = 1 in full detail:
Let α ∈ A1 (M ) with dα = 0. Then α|Uj is exact by the lemma of Poincaré (since
Uj is diffeomorphic to a convex open subset). Let fj ∈ E(Uj ) with dfj = α|Uj . Then
ηij := (fi − fj )|Uij is a constant ηij ∈ R since

d(fi − fj )|Uij = α|Uij − α|Uij = 0, (86)

and the Uij are connected.

The 1-cochain ηα := η := (ηij ) satisfies

(δη)ijk = ηjk − ηik + ηij = fj − fk − fi + fk + fi − fj = 0,

hence ηα = η is a cocycle η ∈ Z 1 (U, R) and defines an element [ηα ] ∈ Ȟ 1 (U, R) .

E.2 DeRham Cohomology 339

To what extent is this cohomology element [η] independent of the various choices
made? Let α∗ ∈ A1 (M ) be in the same de Rham class as α, i.e. α∗ − α = dg,
where g ∈ E(M ). Choose fj∗ ∈ E(Uj ) with dfj∗ = α∗ |Uj and ηij∗ = (fi∗ − fj∗ )|Uij . By
d(fj∗ − fj ) = (α∗ − α)|Uj = dg|Uj we can write

fj∗ − fj = g + cj

with suitable constants cj ∈ R. Hence,

ηij∗ − ηij = ci − cj , i.e. δc = η ∗ − η ,

when c = (ci ). As a result, η ∗ − η ∈ B 1 (U, C), hence [η ∗ ] = [η ] ∈ Ȟ 1 (U, R), and the
map
1
Ψ : HdR (M, R) → Ȟ 1 (U, R) ,
with Ψ(α) := [ ηα ], is well-defined. Evidently Ψ is a homomorphism. Moreover, Ψ
is injective: Ψ(α) = [ηα ] = 0 implies ηij = ci − cj for ci ∈ F(Ui , R), cj ∈ F(Uj , R),
i.e. ci , cj ∈ R. Hence, gj := fj + cj satisfy dgj = α|Uj and gi |Uij = gj |Uij for all i, j ∈ I.
Therefore, g|Uj := gj defines a function g ∈ E(M ) with dg = α which means [α] = 0 in
1
HdR (M, R).
To show surjectivity of Ψ let [η] ∈ Ȟ 1 (U, R) , with η = (ηij ) and ηij ∈ F(Uij , R)
satisfying δη = 0. We find a smooth, locally finite partition of unity (hk )k∈I such that
the support supp hj of hj is compact and supp hj ⊂ Uj . Define
X
αη := ηij hi dhj .
i,j

η = (ηij ) ∈ Z 1 (U, R). We can show

Claim:
αη |Uk = dfk ,
P
where fk = ηkj hj . (All sums are finite, since (hk )k∈I is locally finite.)
Using this result it is clear that the 1-form αη is closed. Furthermore, we get
X
fk − fl = (ηkj − ηlj ) hj
j
X
= ηkl hj , because of δη = 0
j
X
= ηkl hj
j

= ηkl .

We conclude Ψ(αη ) =[η] ∈ Ȟ 1 (U, R) and the surjectivity of Ψ is proven.

340 E. Cohomology

In order to prove the claim

X
αη |Uk = dfk = d( ηkj hj ) ,
j

consider:
XX X
αη = ηij hi dhj + ηkj hk dhj
j i̸=k j
!
XX X X
= ηij hi dhj + ηkj 1− hi dhj
j i̸=k j i̸=k
XX X
= (ηij + ηjk ) hi dhj + ηkj dhj
j i̸=k j
!
X X X 
= ηik hi dhj + d ηkj hj
j i̸=k
! !
X X X  X
= ηik hi dhj + d ηkj hj and d hj =0
i̸=k j j
X 
=d ηkj hj .

1
We have shown that Ψ : HdR (M, R) → Ȟ 1 (U, R) is an isomorphism.
The proof extends directly to cases q > 1. The definition of Ψ will be done as before
by descending from α ∈ Aq (M ), dα = 0, first to βj ∈ Aq−1 (M ) with dβj = α|Uj , then
to γij ∈ Aq−2 (M ), with dγij = βi − βj |Uij etc.
The main part of the proof is again to establish the surjectivity of Ψ with the help
of a smooth partition of unity (hj ). To η ∈ Z q (U, R) we define:
X
αη := ηi0 i1 ...ip hi0 dhi1 ∧ · · · ∧ dhip

and see that Ψ(αη ) =[η].

Remark E.9. With an open cover U of a manifold M as in Proposition E.8 the
same result holds for the corresponding C-valued cohomologies: There exists a natural
isomorphism
q
HdR (M, C) → Ȟ q (U, C), ∀q ∈ N.

We want to explain the definition of Ψ in the case of q = 2 in order to comment

the integrality condition in the form we need it in Section 8.1.
Remark E.10. The natural isomorphism
2
Ψ : HdR (M, C) → Ȟ 2 (M, C)
E.3 Sheaf Cohomology 341

can be constructed as follows. Choose an open cover U = (Uj ) as before in the last
proposition so that all intersections Uj0 j1 ...jn are diffeomorphic to a convex open subset
of Rn or empty.
Given a closed 2-form α ∈ A2 (M ), we find βj ∈ A1 (M ) with dβj = α|Uj and
functions fij ∈ E(Uij ) with dfij = βi − βj |Uij . Hence,

ηijk = fij + fjk + fki ∈ C

is constant and defines η = (ηijk ) ∈ Z 2 (U, C). The class [η]=Ψ(α) is independent of
the choices α, βj , fij and yields the isomorphism
2
Ψ : HdR (M, C) → Ȟ 2 (U, C) = Ȟ 2 (M, C)

The integrality condition can now be reformulated in a complete and satisfying way.

Definition E.11. A closed ω ∈ A2 (M ) is entire (or satisfies condition (E) if there

2
exists an open cover (Uj )j∈I of M such that the class [ω] ∈ HdR (M, C) contains as a
2 ∼ 2
Čech class [c] ∈ Ȟ ((Uj )j∈I , C) = Ȟ (M, C) a cocycle c = (cijk ), with cijk ∈ Z for all
i, j, k ∈ I with Uijk ̸= ∅.

E.3 Sheaf Cohomology

As a general structure we investigate in the following the situation in which for all open
subsets U ⊂ M of a topological space M a collection F(U ) of functions or sections of
a special type (for example locally constant or smooth, or continuous or holomorphic
etc.) is given with compatibility conditions with respect to the inclusion V ⊂ U ,
U, V ⊂ M open. A careful analysis of such data lead to presheaves and sheaves.
For such sheaves we give a short description is the corresponding Čech cohomology.

Definition E.12. For a topological space M we always have the category t(M ) of
open subsets. The objects are the open subsets and the morphisms are the inclusions
U ⊂ V, U, V ∈ M open. A Presheaf of abelian groups on M is a contravariant
functor
F : t(M ) → Ab
from t(M ) into the category of abelian groups Ab.

Hence, F(U ) is an abelian group for each U ⊂ M open and to every inclusion
V ⊂ U there corresponds a homomorphism

ρV,U = F(V ⊂ U ) : F(U ) → F(V ) (87)

such that
ρW,V ◦ ρV,U = ρW,U and ρU,U = idFU (88)
342 E. Cohomology

for open W ⊂ V ⊂ U .
We use the notation ρV,U instead of F(V ⊂ U ) (which corresponds to the use in
category theory) since in many respects, these homomorphisms behave like a restriction.
In many cases of interest they are in fact restrictions but not in all. We also use the
notation g|V instead of ρV,U g for an element g ∈ F(U ).
We do not need more than the properties of a presheaf listed above in (87),(88),
in particular, we can avoid using the language of category theory.

Examples E.13.

1. G an abelian group, F(U ) := {f : U → G | f a map} and ρV,U (f ) = f |V .

2. G as before, F(U ) := {f : U → G | f locally constant}. This is the case we
have studied in the first section of this chapter.
3. F(U ) = C(U ) = C(U, K) and ρ restriction, where K = R, C, as before.
4. Let G be a topological group and abelian. Then F(U ) = C(U, G) and ρ restriction
as before defines a presheaf, generalizing the above.
5. F(U ) = C ∞ (U ) = E(U ).
6. F(U ) = O(U ) if M is a complex manifold and O(U ) is the space of holomorphic
functions on M .
7. F(U ) = Γ(U, L) the E(U )-module of differentiable sections for a line bundle over
M.
8. F(U ) = Γhol (U, L) the O(U )-module of holomorphic sections for a holomorphic
line bundle over a complex mainfold M .
9. F(U ) = Ap (U ), U ⊂ M open subset of a smooth manifold.
10. Similarly, for the holomorphic forms.

The notion of a presheaf of abelian groups can be extended to presheaves with

values in other mathematically structered objects like vector spaces, rings, Banach
spaces, etc. In most of the examples F(U ) is a vector space, in some cases an algebra,
or a module over another presheaf, for instance, Aq is a presheaf of E-modules.
S
A presheaf F is a sheaf if in case of an open cover (Uj ) of an open U ⊂ M , U = Uj ,

• the elements f ∈ F(U ) are determined by its ”restrictions” f |Uj = ρUj ,U f and
• local elements fj ∈ F(Uj ) can be glued together to obtain an f ∈ F(U )
with ρUj ,U f = f |Uj = fj whenever they are compatible in the following sense:
ρUjk ,Uj fj = ρUjk ,Uk fk , j, k ∈ I .
E.3 Sheaf Cohomology 343

More precisely:
Definition E.14. A presheaf F is a sheaf if for all open subsets U ⊂ M and all open
covers U = (Uj )j∈I of U the following property is satisfied:
A collection fi ∈ F(Ui ), i ∈ I, is of the form

fi = ρUi ,U (f ) ∀ i ∈ I,

for a unique element f ∈ F(U ) if and only if for all i, j ∈ I the compatibility property

ρUi ∩Uj ,Ui (fi ) = ρUj ∩Ui ,Uj (fj )

holds.

For instance, if a collection of maps

fi : Ui → X

into a topological space X is continuous and fi |Ui ∩Uj = fj |Ui ∩Uj , then the map

f (a) := fi (a), a ∈ Ui

is a well-defined continuous map with f |Ui = fi , i ∈ I. Moreover, f is unique.

All the examples in E.13 are sheaves.
The presheaf F(U ) = G, U ⊂ M open and ρV,U = idG : G → G is in general not a
sheaf.
For the cohomology of sheaves one needs the simplices introduced in the first section
of this chapter.
Definition E.15. Let F be a sheaf on a topological space M and let U = (Uj )j∈I be
an open cover of M .

1. A q-cochain of U with coefficients in F is a map

σ → η(σ) ∈ F(|σ|), σ a q-simplex of U,

C q (U, F ) is the abelian group of cochains.

2. The coboundary operator

δ = δ q : C q (U, F) → C q+1 (U, F),

is
q+1
X
η→
7 δη, δη(σ) = (−1)j ρ|σ|,|∂j σ| η(∂j σ).
j=0

It is easy to show that δ 2 = 0 and that δ is a homomorphism.

344 E. Cohomology

3. We define:

Z q (U, F) = Ker δ q : C q (U, F) → C q+1 (U, F) ,

B q (U, F) = Im δ q−1 : C q−1 (U, F) → C q (U, F) ,

Ȟ q (U, F) = Z q (U, F)/B q (U, F) .

Example E.16. Let L be a line bundle over the manifold M which is given by transi-
tion functions gij ∈ E × (Uij ) with respect to an open cover U = (Uj ) of M . As usual, we
abbreviate for a q-cochain η the value η(σ), σ = (Uj0 , . . . , Ujq ) , by ηj0 ,...,jq . In particular
g(Ui , Uj ) = gij . Then the transition functions form a 1-cochain g = (gij ) with values
in the sheaf E t imes. Since g satisfies the cocycle condition, g ∈ Z 1 (U, E × ) is a Čech
cocycle. Hence it determines a Čech cohomology class [g] ∈ Ȟ 1 (U, E × ).
Now, let U an open cover, where all intersections Uj are contractible. Then every
line bundle L can be described by suitable transition functions on Uij . Recall, that
Picdiff (M ) is the abelian group of equivalence classes of line bundles on M . Each
such equivalence class [L] is determined by a collection g = (gij ) of suitable transition
functions which define a Čech class [g] ∈ Ȟ 1 (U, E × ). We thus have established an
isomorphism
Picdiff (M ) → Ȟ 1 (U, E × ).

Finally, if V = (Vk )k∈K is a refinement of U with refinement map j : K → I, then

we again have an induced homomorphism

j̃ : Ȟ q (U, F) → Ȟ q (V, F).

The corresponding direct limit gives the cohomology.

Definition E.17. The q-th Čech cohomology group on M with values in the
sheaf F on M is defined by

Ȟ q (M, F) = lim Ȟ q (V, F)

−→
U

Summary: The main objective of the chapter is to prove the equivalence of deR-
ham cohomology and Čech cohomology with values in R resp. in C. The mechanism of
defining Čech cohomology can be transferred to sheaf cohomology without great effort.
The need of sheaves (instead of presheaves) is not apparent so far. In the applications
one considers short exact sequences

0 −→ F −→ G −→ H −→ 0

of sheaves and the corresponding cohomologies which only behave well for sheaves.
 345

F Quantum Mechanics

In this chapter we present the main principles of Quantum Mechanics in form of four
postulates and explain the mathematical framework needed for the formulation of
these postulates. These postulates have been chosen in accordance with the princi-
ples of Quantum Mechanics which are described in the literature on the foundation
of Quantum Mechanics. They comprise much more than is needed for these notes on
Geometric Quantization. But they manifest the geometric nature of Quantum Me-
chanics. And, of course, when Geometric Quantization is presented in these notes on
a mathematical basis it is reasonable and helpful to have a mathematical formulation
of Quantum Mechanics at hand.
We restrict this chapter to the formulation of the mathematical models of Quantum
Mechamics and do not discuss, for example, the measurement process in Quantum
Mechanics or any interpretation.
The mathematics of Quantum Mechanics is advanced and requires a profound un-
derstanding of self-adjoint operators in a Hilbert space. Therefore, after the formulation
of the postulates in the first section of the chapter, Section F.1, we provide in Section
F.2 a short exposition of the theory of self-adjoint operators in a Hilbert space including
examples and the Spectral Theorem.
In many sources about the foundation of Quantum Mechanics more postulates are
required, for instance the representation of the canonical commutation relations (CCR).
Such a postulate can be viewed as to be a special example of a quantum mechanical
system along the postulates 1-4. We investigate the CCR and the closely related Stone-
von Neumann Theorem in Section F.3.

F.1 Four Postulates of Quantum Mechanics

Definition F.1. A Quantum Mechanical System is a pair (H, H), which satisfies
the following four postulates.

Postulate 1. The States of the system are the complex lines through the
origin of H, where H is a Complex Separable Hilbert Space. In other
words, the state space is the projective Hilbert space P(H).

Remarks and Explanations F.2 (to the first postulate).

1◦ Hilbert Space: A complex Hilbert space H is a complex vector space together
with a Hermitian scalar product ⟨·, ·⟩ or Hermitian metric, such that H is complete with
respect to the norm induced by ⟨·, ·⟩. More explicitly, the Hermitian scalar product is
a map:
⟨·, ·⟩ : H × H → C, (89)
with the following properties, which hold for all ϕ, ψ, θ ∈ H, λ ∈ C:
346 F. Quantum Mechanics

• ⟨ϕ + ψ, θ⟩ = ⟨ϕ, θ⟩ + ⟨ψ, θ⟩ and ⟨ϕ, ψ + θ⟩ = ⟨ϕ, ψ⟩ + ⟨ϕ, θ⟩, ⟨λϕ, ψ⟩ = λ⟨ϕ, ψ⟩,
⟨ϕ, λψ⟩ = λ⟨ϕ, ψ⟩. In other words, this means that ⟨·, ·⟩ is R-bilinear, complex
linear in the second and complex antilinear in the first entry.

• ⟨ϕ, ψ⟩ = ⟨ψ, ϕ⟩,

• ⟨ϕ, ϕ⟩ > 0, if ϕ ̸= 0.
p
A scalar product defines a norm on H through ||ϕ|| := ⟨ϕ, ϕ⟩. The space H is
called complete with respect to this norm, if every Cauchy sequence (ϕn ) in H converges
in H.
The homomorphisms between Hilbert spaces are the linear maps T : H → H′ which
respect the norm, i.e. for which ∥T ϕ∥=∥ϕ∥ for all ϕ ∈ H. Equivalently, a linear map
T : H → H is a Hilbert space homomorphism, whenever T leaves the scalar product
invariant, i.e. ⟨T ϕ, T ψ⟩ = ⟨ϕ, ψ⟩ for all ϕ, ψ ∈ H. Such a homomorphism of Hilbert
spaces is injective but in general not surjective.116 The surjective homomorphisms of
Hilbert spaces are called Unitary Operators. Thus, the unitary operators are the
isomorphisms of the theory.
More generally, when Hilbert spaces are viewed as special Banach spaces, one con-
siders the bounded operators T : H → H′ , i.e. the linear maps T from H to H′ with
finite operator norm
∥T ∥ := sup{∥T ϕ∥ | ∥ϕ∥ = 1} < ∞ .
The operator norm equips the complex vector space B(H, H′ ) of bounded operators
from H to H′ with the structure of a Banach space. B(H) denotes the Banach space of
bounded operators from H to H.
Bounded operators are continuous linear maps with respect to the natural norm
topology. The norm topology is the metric topology induced by the metric d(ϕ, ψ) =
∥ϕ − ψ∥ on H: A subset U ⊂ H is open in the metric topology if and only if for every
ϕ ∈ U there exists an r > 0 such that the ball

B(ϕ, r) := {ψ ∈ H | ∥ϕ − ψ∥ < r}

of radius r around ϕ is contained in U . Notice, that continuous linear maps H → H′

are bounded.
2◦ The Examples: The most familiar complex Hilbert spaces are the finite di-
mensional complex Hilbert spaces, which are also called Unitary Spaces. Typical
examples are the number spaces H = Cn with the scalar product:
n
X
⟨z, w⟩ = z i wi .
i=1

116
In the finite dimensional case an injective and linear map is always surjective. Hence, it is an
isomorphism of vector spaces.
F.1 Four Postulates of Quantum Mechanics 347

for z = (z 1 , . . . , z n ), w = (w1 , . . . , wn ) ∈ Cn .
Note, that each n-dimensional Hilbert H is unitarily equivalent to the above unitary
space Cn , where n = dimC H: There exists a unitary operator T : H → Cn . The
existence of T is equivalent to the existence of an orthonormal basis (ej ) of H: ⟨ej , ek ⟩ =
δjk .
A typical infinite dimensional complex Hilbert space is the space ℓ2 of square
summable complex sequences
X
ℓ2 := {(z j )j∈N | z̄ j z j < ∞}
P j j
with the scalar product ⟨z, w⟩ := z̄ w . ℓ2 is complete: Any Cauchy sequence
convergences coordinatewise, and the corresponding limit is the limit of the sequence
in norm. ℓ2 is separable, since - with the basis elements ek := (δkj )j∈N - the countable
set D = { k=n k k 2
P
k=0 q ek | q ∈ Q , n ∈ N} is dense in ℓ .

However, the most important examples of Hilbert spaces for quantum mechanics
are certain function spaces. Namely, the space of square integrable functions on the
configuration space. For example, for an open Q ⊂ Rn
 Z 
2 2
L (Q) := ϕ : Q → C | ϕ measurable and |ϕ(q)| dq < ∞ ,
Q

where in this situation dq stands Rfor the Lebesgue integral117 . The scalar product for
H = L2 (Q) is given by ⟨ϕ, ψ⟩ := Q ϕ(q)ψ(q)dq. In this approach to define L2 (Q) one
needs to identify those functions, which differ only on a set of measure zero.
Without using the Lebesgue integral, one can construct the Hilbert space L2 (Q) in
the following way: One defines
 Z 
2 2
R (Q) := ϕ : Q → C | ϕ is continuous and |ϕ(q)| dq < ∞ ,
Q
R
where now h(q)dq is the Riemann integral for continuous functions h on Q. The
scalar product on R2 (Q) will be obtained in the same way as before,
Z
⟨ϕ, ψ⟩ := ϕ(q)ψ(q)dq ,
Q

with the only difference that the integration is Riemann integration,

p instead of2 Lebesgue
integration. This scalar product determines a norm by ∥ϕ∥ := ⟨ϕ, ϕ⟩ on R (Q). And
the final Hilbert space is the abstract completion R\2 (Q) of the space R2 (Q) with respect

to the norm. Since the scalar product ⟨·, ·⟩ : R2 (Q) × R2 (Q) → C is continuous with
117
More precisely, instead of functions ϕ : Q → C one has to take equivalence classes of measurable
functions ϕ which only differ on a set of measure zero.
348 F. Quantum Mechanics

respect to the topology given by the norm, it can be continued to the completion.
\
Altogether, R 2 (Q) = H ∼
= L2 (Q) is a Hilbert space, unitarily equivalent to L2 (Q).
Note, that it is possible to start with a smaller space than R2 (Q), for example
with E := R2 (Q) ∩ E(Q, C) or with the space E = E0 (Q) of smooth functions on
Q with compact support in Q. Scalar product and norm can be defined on E as
before. The completion process encompasses R2 (Q) and leads to the same completion:
Eb=R \ 2 (Q) ∼
= L2 (Q). Note that E can be viewed as to be a subspace of L2 (Q) where
the scalar product on L( Q) given by Lebesgue integration is, when restricted to E,
the scalar product given by Riemann integration. As a result, the space E b can be
understood as the closure E in L2 (Q), i.e. U
b = E = L2 (Q)
One can show that the examples ℓ2 and L2 (Q) are unitarily equivalent. In fact,
every infinite dimensional and separable Hilbert space is unitarily equivalent to ℓ2
since there exists an orthonormal (Hilbert space) basis H, i.e. ⟨ej , ek ⟩ = δjk and
P (ej ) of118
each ϕ ∈ H has a unique expression as a sum ϕ = ⟨ej , ϕ⟩ej .
Moreover, any Hilbert space is unitarily equivalent to a suitable L2 (X, µ) =
L2 (X)119 where (X, Σ, µ) is a measure space with σ-algebra Σ and measure µ. As
before,
 Z 
2 2
L (X, µ) := ϕ : X → C | ϕ measurable and |ϕ(x)| dµ(x) < ∞ ,
X
R
with the scalar product ⟨ϕ, ψ⟩ := X
ϕ(x)ψ(x)dµ(x) .
n
For instance, for H = C let (X, µ) be the measure space with finite X := {1, . . . , n}
and µ({x}) := 1, x ∈ X . This yields the n-dimensional Hilbert space L2 (X, µ) ∼
= Cn .
The Hilbert space ℓ2 can be defined as L2 (N, µ) with µ({x}) = 1 , x ∈ N.
Moreover, for an uncountable set X and µ({x}) = 1 , x ∈ X one obtains the non-
separable Hilbert space ℓ2 (X) = L2 (X, µ).
3◦ State Space: As required in Postulate 1, the space of states is the projective
Hilbert space P(H) = PH. To obtain this projective space we consider on H \ {0} the
equivalence relation
z ∼ w ⇐⇒ ∃λ ∈ C with z = λw .
The equivalence classes are the complex lines in H through the origin. For an element
z ∈ H , z ̸= 0 , we denote its equivalence class by γ(z) or [z]. In case of H = ℓ2 (or Cn )
we can denote the class [z] also by its homogeneous coordinates [z 0 : z 1 : . . . : zj : . . .].
The quotient space

P(H) = PH := H/ ∼ = {γ(z) | z ∈ H \ {0}},

118
There is no constructive way to define an orthonormal basis, one has to use the axiom of choice.
It is quite remarkable, the the converse is true, as well: When it is possible to find an orthonormal
basis in every separable Hilbert space, then the axiom of choice is valid.
119
The measure µ will often be omitted if it is clear from the context which measure is meant.
F.1 Four Postulates of Quantum Mechanics 349

is the Projective Hilbert Space. PH is the space of states of a quantum mechanical

system according the Postulate 1. PH is determined by the canonical map γ : H\{0} →
PH and inherits its structures by that projection map.
For instance, PH is endowed with a natural topology, the quotient topology
(c.f. Section A.4), for which a subset V ⊂ P(H) is open if and only if γ −1 (V ) is open in
H. By definition, γ is continuous. Moreover, γ is an open map and it is holomorphic
with respect to the holomorphic quotient structure given by the holomorphic charts
(c.f. Examples A.30).
4◦ Example: Particle. An example is the description of a non-relativistic particle
in Quantum Mechanics. Let Q ⊂ R3 be an open subset of R3 . H = L2 (Q, dq) is the
Hilbert space of square integrable functions on Q. The elements ϕ of H are called
Wave Functions and the equivalence classes

a = γ(ϕ) ∈ PH , ϕ ∈ H \ {0} ,
R
are the states of the system. In case of ||ϕ|| = 1 the quantity B |ϕ(q)|2 dq for a
measurable subset B ⊂ Q has the interpretation of the probability that the particle in
state a = γ(ϕ) is contained in B.
5◦ Pseudometric: In contrast to classical mechanics, where we have the phase
space equipped with a symplectic form ω which together with a classical Hamiltonian
H determines the dynamics of the theory, here we have as state space the projective
space PH with the structures induced by the canonical map γ. Part of the dynamics120
of the quantum theory is given by the ”pseudometric” c : PH × PH → R induced by
the scalar product ⟨·, ·⟩: The transition probability between ϕ, ψ ∈ H1 := {ϕ ∈ H |
||ϕ|| = 1} , ϕ ̸= ψ, is
 2
ϕ ψ
t(ϕ, ψ) := , .
||ϕ|| ||ψ||
And for a = γ(ϕ), b = γ(ψ) the pseudometric is
p
c(a, b) := t(ϕ, ψ) .

c is independent of the choice of representatives ϕ, ψ and thus well-defined. The pseu-

dometric is invariant under symmetry transformations of the system, the maps induced
by unitary or antiunitary operators in H.
Note, that the pseudometric c determines the natural topology on the state space
PH. By elementary geometry we see that for two vectors ϕ, ψ ∈ H of unit length the
quantity c([ϕ], [ψ]) is nothing else than cos α, where α can be understood as to be the
angle between ϕ and ψ. Therefore, for each r > 0 there exists h , 0 < h < 1 such that
the ”pseudodisc”
D1 (ϕ, h) := {ψ ∈ H1 | c([ϕ], [ψ]) > h}
120
The dynamics of (H, H) is determined by H, see Postulate 3 below.
350 F. Quantum Mechanics

is contained is the ball

B1 (ϕ, r) = {ψ ∈ H1 | ||ϕ − ψ|| < r} .

Conversely, for each h , 0 < h < 1 there is an r > 1 with B1 (ϕ, r) ⊂ D1 (ϕ, h). Hence,
the collection (D1 (ϕ, h))h<1,ϕ∈H1 of subsets of H1 induces on H1 the natural norm
topology. As a consequence, with the notation D(a, h) := γ(D1 (ϕ, h)) = {b ∈ PH |
c(a, b) > h} the collection
(D(a, h))a∈PH,h<1
generates the natural (quotient) topology on PH as a basis of open subsets.
Moreover, the pseudometric obtains its significance in the fourth postulate: the
quantity t(a, b) = t(ϕ, ψ) ∈ [0, 1] for two states a = [ϕ], b = [ψ] ∈ PH is the transition
probability from a to b. Namely, when the system is initially in the state a, after a
measurement the probability of the system being in the state b is t(a,b).

Postulate 2. Every observable of the system is represented by a self-adjoint

operator acting on the Hilbert space H. As a consequence, the set of all possible
observables is:

SA := {T : D(T ) → H | D(T ) ⊂ H and T : D(T ) → H self-adjoint} .

Remarks and Explanations F.3 (to the second postulate).

1◦ Self-Adjointness: An operator T is a C-linear map T : D(T ) → H from a
linear subspace D(T ) ⊂ H into H which is called the domain of T .
The Adjoint T ∗ of a linear operator T with dense domain D(T ) is defined as
follows: The domain of T ∗ is

D(T ∗ ) := {ψ ∈ H | there exists ξ ∈ H : ⟨ξ, ϕ⟩ = ⟨ψ, T ϕ⟩ for all ϕ ∈ D(T )} .

For ψ ∈ D(T ∗ ) the ξ with ⟨ξ, ϕ⟩ = ⟨ψ, T ϕ⟩ for ϕ ∈ D(T ) is unique (since D(T ) is
dense), and T ∗ (ψ) := ξ.
T is a self-adjoint operator if D(T ) is dense and the adjoint T ∗ of T agrees
with T , i.e. D(T ) = D(T ∗ ) and T ϕ = T ∗ ϕ for all ϕ ∈ D(T ) = D(T ∗ ).
T is Symmetric, if for all ϕ, ψ ∈ D(T ) : ⟨T ϕ, ψ⟩ = ⟨ϕ, T ψ⟩. Hence a self-adjoint
operator is symmetric. The converse does not hold, in general.
Details and results about self-adjoint operators will be explained in the next section,
we only point out the following:
2◦ Finite Dimensional Hilbert Space: In case of H ∼ = Cn with the usual scalar
product it is easy to see: Every dense linear subspace D of H is all of H: D = H,
F.1 Four Postulates of Quantum Mechanics 351

and every C-linear map T : H → H is automatically continuous and therefore also

closed. T is self-adjoint, if the matrix M = MT which represents T with respect to an
⊤
orthonormal basis of H satisfies M = M . One also says T (or M ) is symmetric with
respect to the hermitian scalar product.
3◦ Infinite Dimensional Hilbert Space: In case of an infinite dimensional
Hilbert space H any self-adjoint operator T with D(T ) = H is bounded (equivalently:
continuous) since T is closed (c.f. Definition F.7) and closed operators T with domain
D(T ) = H are bounded (see Proposition F.10). However, there exist self-adjoint op-
erators, which cannot be defined for all points of the Hilbert space H, i.e. D(T ) ̸= H,
the so called unbounded operators. For Quantum Mechanics most of the important
operators are unbounded and self-adjoint.
4◦ Multiplication operator: Let v : Rn → C a continuous function, and let
H = L2 (Rn , dλ)121 be the Hilbert space of square integrable complex functions on Rn .
Consider the multiplication operator

Mv =: D(M ) → H , ϕ 7→ vϕ ,

defined on Z
D(M ) := {ϕ ∈ H | |v(q)ϕ(q)|2 dλ(q) < ∞} .
Rn

D(M ) is dense in H since all bounded ϕ with bounded support are contained in D(M ).
So the adjoint exists.
Now, for all ψ ∈ H and all ϕ ∈ D(M ) the equality
Z Z
⟨vψ, ϕ⟩ = vψdλ = ψ̄vϕdλ = ⟨ψ, vϕ⟩
Rn Rn

holds. As a consequence D(M ∗ ) = D(M ) and M ∗ ϕ = vϕ.

We conclude: M = Mv is self-adjoint if and only if v = v, i.e. v is real-valued. M
is bounded if and only if v is essentially bounded, i.e. there is a subset N ⊂ Rn of
measure 0 such that sup{|v(q)| | q ∈ Rn \ N } < ∞.
The example has a straightforward generalization to general measure spaces
(Ω, Σ, µ) instead of Rn , see Example F.22 below.
5◦ Position: The example of a quantum mechanical system of a non-relativistic
particle on the real axis. Here, we set H := L2 (R). A typical observable is the position
operator Q with  Z 
D(Q) := ϕ∈H | |qϕ(q)|2 dq < ∞ ,

and Qϕ(q) := qϕ(q) for q ∈ R and ϕ ∈ D(Q). Q is self-adjoint according to the

preceding example with v(q) = q. However, Q is not bounded: The sequence ϕn :=
121
λ the Lebesgue measure
352 F. Quantum Mechanics

x[n,n+1] of indicator functions of the interval [n, n + 1] is bounded by 1 : ||ϕn || = 1, but

||Qϕn || ≥ n.
6◦ Momentum Another unbounded observable is the momentum operator P in
H = L2 (R) with

D(P ) := {ϕ ∈ H | ∃Dϕ ∈ H : ⟨Dϕ, ψ⟩ = −⟨ϕ, ψ ′ ⟩ ∀ψ ∈ E(R) , ψ compactly supported}

and P ϕ := −iDϕ for ϕ ∈ D(P ). Here ψ ′ = ψ̇ is the usual derivative of the differentiable
function ψ, while Dϕ is the ”weak derivative” of ϕ ∈ D(P ).

The notion of a self-adjoint operator is fundamental for the mathematical modeling

of quantum mechanical systems. Several deep results about self-adjoint operators are
needed in a mathematical treatment of Quantum Mechanics. Therefore, we include
a separate section on self-adjoint operators (see Section F.2), in which elementary
properties of self-adjoint operators are described and important results as e.g. the
Theorem of Stone and the Spectral Theorem are explained.

Postulate 3. H is an observable and determines the dynamics of the quantum

mechanical system (H, H) in the following sense: Let p0 ∈ P(H) be a state of the
quantum mechanical system with a representative (state vector) ϕ0 ∈ H , ||ϕ0 || =
1. Then the time evolution ϕ(t) of the state vector ϕ0 is given by:

d
ϕ(t) = ϕ̇(t) = −iH(ϕ(t)) , t ∈ R , (90)
dt
with the initial condition ϕ(0) = ϕ0 . This means that the time evolution p(t) of
the state p0 is represented by the unique solution ϕ(t) of the above equation with
ϕ(0) = ϕ0 , i.e. p(t) := [ϕ(t)] with p(0) = p0 .

Remarks and Explanations F.4 (to the third postulate).

1◦ Schrödinger Equation: The above equation (90) is called the Schrödinger
equation and H is called the Hamiltonian (”Hamiltonoperator”) of the quantum
mechanical system (H, H).
2◦ Unitary Group: In order to better understand the time evolution of a quan-
tum mechanical state we want to formulate Stone’s Theorem which describes a strong
relation between self-adjoint operators and unitary operators in a Hilbert space H.
Recall that a unitary operator U is a bijective C-linear map U : H → H which leaves
the scalar product invariant, that is: ⟨U ϕ, U ψ⟩ = ⟨ϕ, ψ⟩ for all ϕ, ψ ∈ H. A unitary
operator U is automatically bounded since ||U ϕ|| = ⟨U ϕ, U ϕ⟩ = ||ϕ||. In particular,
the operator norm of U is ||U || = sup{||U f || : ||f || = 1} = 1 and U is continuous.
F.1 Four Postulates of Quantum Mechanics 353

The inverse U −1 of a unitary operator U exists and is again a unitary operator.

Finally, the composition U ◦ V : H → H of two unitary operators U, V in H is also a
unitary operator. As a consequence, the set of unitary operators in H form a group,
called the Unitary Group, which is denoted by U(H).
A One-Parameter Group of Unitary Operators in H is given by an R-
indexed family of unitary operators (Us )s∈R , which can be described by an action Φ of
R on H, i.e.
Φ : R × H → H, Us f := Φ(s, f ), (s, f ) ∈ R × H
satisfying

1. Us ∈ U(H) for all s ∈ R,

2. Us ◦ Ut = Us+t , for s, t ∈ R

3. for all f ∈ H, the map R → H , s → Us (f ), is continuous.

In another terminology, R → U(H) , s 7→ Us , is a representation, i.e. a continuous

group homomorphism, where U (H) is endowed with the strong topology.

Theorem F.5 (Theorem of Stone). To any one-parameter group of unitary operators

(Us ) there corresponds an Infinitesimal Generator A, which is the operator defined
by:
1
D(A) := {ϕ ∈ H | lim (Us ϕ − ϕ) exists},
s→0 s
1
A(ϕ) := i (lim (Us ϕ − ϕ)), ϕ ∈ D(A) .
s→0 s

Then

1. the infinitesimal generator of a one-parameter group of unitary transformations

is self-adjoint,

2. to each self-adjoint operator A on H there is a corresponding one-parameter group

of unitary operators (Us ), whose infinitesimal generator is A. We denote this as
Us = e−isA (in accordance with the functional calculus of self-adjoint operators,
see below).

Therefore, the observables defined in Postulate 2 are in one-to-one correspondence

with one-parameter groups of unitary operators.
We conclude that the Hamiltonian H induces the one parameter group Ut = e−itH
and the solution of the Schrödinger equation (90) is ϕ(t) = Ut ϕ . for ϕ(0) = ϕ0 . Or, in
another form
ϕ(t) = e−itH ϕ , t ∈ R .
354 F. Quantum Mechanics

Postulate 4. Let [ϕ] ∈ P(H) be a state of the quantum mechanical system with
representative ϕ ∈ H , ||ϕ|| = 1, and let T be an observable with its corresponding
spectral family (Eλ )λ∈R . Then, for an open interval J = ]a, b[, the probability
that an eigenvalue of the observable T in the state [ϕ] is contained in the interval
J is given by the formula
Z b
2
p(ϕ, T, J) := ||E(J)ϕ|| = ⟨ϕ, E(J)ϕ⟩ = d(||Eλ ϕ||2 ) .
a

The concept of a spectral family of projection operators corresponding to a self-

adjoint operator will be explained in the next section.

F.2 Self-Adjoint Operators

This section is a short introduction to self-adjoint operators and the spectral theorem.
Although we do not need details about self-adjoint operators in the Lecture Notes,
the notion and the results of self-adjoint operators are necessary for the mathematical
formulation of quantum mechanics as is apparent from the preceding section on the
postulates of Quantum Mechanics. Moreover, we need a certain familiarity with self-
adjoint operators in the next section.
The material presented in this section can be found in any book on linear operators
in Hilbert spaces, e.g. in [Wei12] or [Hal13].
As before, in this section H denotes a separable Hilbert space. A Linear Opera-
tor in H – often called Operator – is a linear map A defined on a linear subspace
D(A) of H with values in H: A : D(A) → H. D(A) will be called the domain of the
operator A and Im (A) = A(D(A)) will be called its range122 .
It is important to realize, that in the case of an operator B with D(A) ⊂ D(B)
and B|D(A) = A the two operators are regarded as to be different operators when
D(A) ̸= D(B). But B will be called an Extension of A.

Definition F.6. A linear operator A in the Hilbert space H will be called Densely
Defined if the linear subspace D(A) is dense in H.
For a densely defined operator A the Adjoint Operator A∗ of A is defined as
follows: The domain if A∗ is

D(A∗ ) := {ϕ ∈ H | ∃ χ ∈ H : ⟨ϕ, Aψ⟩ = ⟨χ, ψ⟩ for all ψ ∈ D(A) }

122
In the Hilbert space literature the range of an operator A is denoted by R(A). We prefer Im (A)
since we use this notation in the linear algebra context throughout the Lecture Notes.
F.2 Self-Adjoint Operators 355

and the value A∗ ϕ ∈ H is defined by ⟨ϕ, Aψ⟩ = ⟨A∗ ϕ, ψ⟩ for all ψ ∈ D(A). A∗ ϕ is
well-defined since D(A) is dense.
Finally, a densely defined operator A is called

1. Self-adjoint if A and A∗ agree, i.e. D(A) = D(A∗ ) and Aϕ = A∗ ϕ for all

ϕ ∈ D(A) (this is a reformulation of the definition in Remark 1◦ of F.3).

2. Symmetric, if ⟨ϕ, Aψ⟩ = ⟨Aϕ, ψ⟩ for all ϕ, ψ ∈ D(A).

For a symmetric operator A the domain of D(A∗ ) contains D(A) and the two
operators agree on the domain D(A) of A: A∗ |D(A) = A. Thus, A∗ is an extension
of A: A ⊂ A∗ . As a consequence, in order that an operator A is self-adjoint, it is
necessary that A is Symmetric.

Definition F.7. An operator A in H is said to be

1. Closed if the graph Γ(A) of A, defined as Γ(A) := {(ϕ, Aϕ) ∈ H × H | ϕ ∈

D(A)}, is closed in H × H.

2. Closable if the closure Γ(A) ⊂ H × H in H × H is the graph of an operator

B. In that case B is called the closure of A and will be denoted by A := B.

3. Essentially Self-adjoint if A is closable and the closure A satisfies A = A∗ ,

i.e. is self-adjoint.

In particular, for a closable operator A, Γ(A) = Γ(A) and A is a closed operator.

The following assertions are easy to prove.

Proposition F.8. Let A be a densely defined operator. Than the adjoint A∗ of A is

closed. Moreover, the double adjoint A∗∗ = (A∗ )∗ exists if A is closable, and in that
∗
case A = A∗ and A = A∗∗ .
In particular, a self-adjoint operator A is closed and fulfills A = A∗∗ .

By definition, an operator A is closed if for all sequences (ϕn ) in D(A) for which
(ϕn ) and (Aϕn ) converge in H the following holds:

lim ϕn ∈ D(A) , and lim Aϕn = A(lim ϕn ) .

This looks very much like a continuity condition, but a closed operator need not be
continuous as the following example shows.
P∞
Example F.9. In the Hilbert space H = ℓ2 = {(zj ) ∈ CN | 2
1 |zj | < ∞} of
square summable sequences the operator A(zj ) := (jzj ) with domain D(A) = {(zj ) |
j 2 |zj |2 < ∞} is densely defined. Let ϕ = ( 1j ) ∈ H and define ϕn ∈ H by (ϕn )j := 1j
P

for j = 1, 2, . . . , n and (ϕn )j := 0 for j > n. Then ϕn → ϕ but Aϕn does not converge
356 F. Quantum Mechanics

in H, since ∥Aϕn ∥2 = n → ∞. Hence, A is not continuous. However, A is closed. When

(n)
ψn = (zj ) → (vj ) =: ψ and Aψn → (wj ) =: ϕ , for each fixed j ∈ N the coordinates
converge:
(n) (n)
lim zj = vj , lim jzj = wj .
n n
P 2 2
Hence, jvj = wj . Therefore, j |vj | < ∞ which means that ψ ∈ D(A) and Aψ =
(jvj ) = (wj ) = ϕ.
By the way, A is self-adjoint. Notice, that in Remark 4◦ of F.3 a general example
of an unbounded and closed (and self-adjoint) operator, the multiplication operator, is
presented.

A continuous (or equivalently bounded) operator A : H → H (D(A) = H) is always

closed. We cite the following well-known result:

Assertion F.10 (Closed Graph Theorem). An operator A with D(A) = H is contin-

uous if and only if A is closed.

Spectrum of an Operator
An eigenvalue of the operator A in H is a complex number λ ∈ C such that there
exists a ϕ ∈ H , ϕ ̸= 0, satisfying Aϕ = λϕ i.e. such that λI − A = λ − A123 is not
injective. ϕ is called eigenvector and Ker (λ − A) = N(λ − A)124 is the eigenspace of λ.

Proposition F.11. Every eigenvalue λ of a symmetric operator is real, i.e. λ ∈ R,

and eigenvectors of different eigenvalues are orthogonal to each other. Moreover, for
each λ ∈ C\R the operator λ−A is injective and the inverse (λ−A)−1 : Im(λ−A) → H
is a bounded operator with ∥(λ − A)−1 ∥ ≤ |ℑλ|−1 .

Proof. Let λ be an eigenvalue of A with Aϕ = λϕ , ϕ ̸= 0. Then

(λ̄ − λ) ∥ϕ∥ = (λ̄ − λ) ⟨ϕ, ϕ⟩ = ⟨λϕ, ϕ⟩ − ⟨ϕ, λϕ⟩ = ⟨Aϕ, ϕ⟩ − ⟨ϕ, Aϕ⟩ = 0 ,

hence λ = λ̄.
For Aϕj = λj ϕj , j = 1, 2, the following holds:

(λ1 − λ2 ) ⟨ϕ1 , ϕ2 ⟩ = ⟨λ1 ϕ1 , ϕ2 ⟩ − ⟨ϕ1 , λϕ2 ⟩ = ⟨Aϕ1 , ϕ2 ⟩ − ⟨ϕ1 , Aϕ2 ⟩ = 0

As a consequence, ⟨ϕ1 , ϕ2 ⟩ = 0 if λ1 ̸= λ2 .
For λ = ξ + iη and ϕ ∈ D(A) the symmetry of A implies

⟨(ξ − A)ϕ, iηϕ⟩ + ⟨iηϕ, (ξ − A)ϕ⟩ = 0 ,

123
I denotes the identity operator H → H, I(ϕ) = ϕ, and the symbol I will often be omitted in the
notations like λ − A for λI − A.
124
In the Hilbert space literature the kernel of a linear operator is mostly denoted by N (A).
F.2 Self-Adjoint Operators 357

and we obtain
∥(λ − A)ϕ∥2 = ∥(ξ − A)ϕ + iηϕ∥2 = ∥(ξ − A)ϕ∥2 + η 2 ∥ϕ∥2 ≥ η 2 ∥ϕ∥2 ,
When λ ∈ C \ R, i.e. ℑλ = η ̸= 0, this implies that λ − A has to be injective. Moreover,
for ψ = (λ − A)ϕ , ϕ ∈ D(A):
(λ − A)−1 ψ = ∥ϕ∥ ≤ |ℑλ|−1 ∥(λ − A)ϕ∥ = |ℑλ|−1 ∥ψ∥ ,
which is the inequality we intended to prove.
Definition F.12. Let A be an arbitrary operator A in H. complex number λ ∈ C is
in the Resolvent Set ρ(A) if λ − A : D(A) → H is bejective and the Resolvent
Operator R(λ, A) (for A at λ)
R(λ, A) := (λ − A)−1
is a bounded operator R(λ, A) : H → H.
The complement σ(A) := C \ ρ(A) is the Spectrum of A:
In most cases A will be assumed to be closed. Since for an operator A which is not
closed, the operators λ − A and R(λ, A) (in case λ − A is injective) will not be closed.
Therefore, ρ(A) = ∅. For a closed operator A, the resolvent set has the slightly simpler
description ρ(A) = {λ ∈ C | λ − A : D(A) → H is bijective } according to the closed
graph theorem F.10.
Assertion F.13. For a closed operator A the resolvent set ρ(A) is an open subset of
C and the spectrum σ(A) is closed. Moreover, when λ0 ∈ ρ(A) the open disc D = {λ ∈
C | |λ − λ0 | < ∥R(λ0 , A)∥−1 } is contained in ρ(A) and
∞
X
R(λ, A) = (λ0 − λ)n R(λ0 , A)n+1 ,
0

where the series converges in norm and uniformly on compact subsets of the disc. In
particular, R( · , A) : ρ(A) → B(H) is continuous and holomorphic.
Observation F.14. The spectrum σ(A) of an operator is divided into the following
subsets:
σp (A) := {λ ∈ σ(A) | λ − A is not injective}
(Point Spectrum)
σc (A) := {λ ∈ σ(A) | λ − A is injective, Im(λ − A) = H, and Rλ (A)
is not bounded}
(Continuous Spectrum)
σr (A) := {λ ∈ σ(A) | λ − A is injective and Im(λI − A) ̸= H}
(Residual Spectrum)
σp (A), σc (A), σr (A) are pairwise disjoint and σp (A) ∪ σc (A) ∪ σr (A) = σ(A).
Note, that λ ∈ σp (A) if and only if A = λϕ has a nontrivial solution ϕ, i.e. if λ is
an eigenvalue of A.
358 F. Quantum Mechanics

It is easy to see, that a bounded operator A with D(A) = H has bounded spectrum
σ(A) ⊂ {λ ∈ C | |λ| ≤ ∥A∥} and σ(A) ̸= ∅. If the operator A s not bounded it can
happen that σ(A) = C, even if A is densely defined, but also that σ(A) = ∅.
We have deduced that a self-adjoint operator has to be densely defined, symmetric
and closed. What property is missing? The crucial missing property can be expressed
using the spectrum as the following result shows.

Proposition F.15. Let A be an operator on a Hilbert space H which is closed and

symmetric. Then

1. The index dλ (A) := dim(Im(λ − A))⊥

(a) is constant throughout the open upper half-plane.

(b) is constant throughout the open lower half-plane.

2. For the spectrum σ(A) one of the following alternatives holds true

(a) σ(A) is the closed upper half-plane,

(b) σ(A) is the closed lower half-plane,
(c) σ(A) is the entire plane,
(d) σ(A) is a subset of the real axis.

3. A is self-adjoint if and only if the indices dλ (A) are zero for λ ∈ C \ R.

4. A is self-adjoint if and only if σ(A) ⊂ R i.e. the case 2.(d) holds.

In particular: σ(A) ⊂ R if and only if A is essentially self-adjoint. When A is

self-adjoint: σr (A) = ∅.
The indices d+ (A) := di (A) , d− (A) := d−i (A) are called the deficiency indices.
d± (A) can be expressed as follows

d± (A) = dim{ϕ ∈ H | A∗ ϕ = ±iϕ} ,

since Ker (λ̄ − A∗ ) = (Im(λ − A))⊥ .

Corollary F.16. A symmetric operator A is self-adjoint if it satisfies d+ (A) =

d− (A) = 0. Moreover, A is essentially self-adjoint if d+ (A) = d− (A) = 0, and it
has a self-adjont extension if d+ (A) = d− (A).

We present three illustrative examples: Let H = L2 (I) with I ⊂ R a closed interval

in R. I is of the form I = R, I = [a, ∞[, T =]∞, b] or I = [a, b] for a, b ∈ R.
F.2 Self-Adjoint Operators 359

Example F.17. The multiplication operator M = Mv for a measurable function v :

I → R (position operator in the appropriate context and with v(x) = x) is given by
Z
D(M ) := {ϕ ∈ H | |v(x)|2 |ϕ(x)|2 < ∞},
I

ϕ 7→ M ϕ := vϕ , phi ∈ D(M ).
We have d± = 0 since vϕ = ±iϕ is satisfied only for ϕ = 0. Hence M is self-adjoint.
This example is related to the example in Remark 4◦ of F.3.
Example F.18. The differentiation operator P (momentum operator in the appropri-
ate context)
ϕ 7→ P ϕ := −iϕ′ ,
with ϕ ∈ D(P ) , where D(P ) := C∞0 (I) is the space of infinitely often differentiable
functions on I with compact support in the interior of I. Of course, P is symmetric.
In case of I = R the indices are d± = 0 and P is essentially self-adjoint. This is the
example in Remark 5◦ of F.3.
In case of I = [a, ∞[ we have d+ (P ) = 1 and d− (P ) = 01, since P (ϕ) = −iϕ′ = iϕ for
ϕ = e−x and since P (ϕ) = −iϕ, i-e. ϕ′ = ϕ has no non-trivial solution ϕ ∈ L2 ([a, ∞[).
Thus there is no self-adjoint extension. The case ]∞, b] is analogous.
In the case of I = [a, b], d± (P ) = 1. A class of self-adjoint extensions can be
described by boundary conditions. For t ∈ ]0, 1] set

D(Pt ) := {ϕ ∈ H | ϕ absolutely continuous with ϕ′ ∈ H , ϕ(a) = ϕ(b) exp(2πit)} .

Then Pt ϕ = −iϕ , ϕ ∈ D(Pt ), is a self-adjoint extension of P .

Example F.19. The Laplace operator H := −∆ , ϕ 7→ −ϕ′′ , is symmetric on D =
C0∞ (I) ⊂ H.
In case of I = R the indices are d± = 0 and H is essentially self-adjoint.
In case of I = [a, ∞[ we have d± (H) = 1. The self-adjoint extensions of H are
determined by boundary conditions of the form ϕ(a) cos θ + ϕ′ (a) sin θ = 0. The case
]∞, b] is analogous.
In the case of I = [a, b], d± (H) = 2. The self-adjoint extensions are determined by
boundary conditions of the form ϕ(a) = ϕ(b) = 0, or ϕ′ (a) = ϕ(b)′ = 0, or ϕ(a) = ϕ(b)
and ϕ′ (a) = ϕ(b)′ .

Spectral Theory for Self-Adjoint Operators

One of the most beautiful and efficient results for self-adjoint operators is the spec-
tral theorem.
In order to motivate the spectral theorem let us have a look at the finite dimensional
case. A symmetric operator A in a finite dimensional complex Hilbert space H is already
360 F. Quantum Mechanics

self-adjoint. Symmetry can be described by the matrix which represent A: Let (ej ) be
an orthonormal basis of H. An operator A is symmetric if and only if its matrix (ajk )
with respect to the basis (ej ) is Hermitian, i.e. if (ajk ) = (ājk )⊤ . From Linear Algebra
we know:
Theorem F.20 (Spectral Theorem in Finite Dimension). Every symmetric operator
A in an n-dimensional Hilbert space H has diagonal form
n
X
Aϕ = λj ⟨ej , ϕ⟩ ej ,
1

where (ej ) is a suitable orthonornal basis and λj ∈ R.

The λj , j = 1, . . . , n, are the eigenvalues of A and constitute the spectrum: σ(A) =

{λj }. The eigenspaces are the subspaces Ker (λj − A).
Example F.21. As a generalization we consider a sequence (λj )j∈N of complex num-
bers λj ∈ C and define in H = ℓ2 the operator Aϕ := (λj ϕj ) , ϕ = (ϕj ) ∈ D(A), where
D(A) := {ϕ ∈ ℓ2 | |λj ϕj |2 < ∞}. D(A) is dense and the adjoint A∗ is given through
P
the sequence (λj ) with D(A∗ ) = D(A). Hence, A is self-adjoint if and only if λj = λj
for all j ∈ N. The operator A can be expressed as
n
X
Aϕ = λj ⟨ej , ϕ⟩ ej ,
1

where (ek ) is the orthonormal basis ek := (δjk )j∈N .

Now, the spectral theorem for a self-adjoint and compact operator125 A in a Hilbert
space H says that T is essentially of the form just described: There exist a unitary
operator U : H → ℓ2 and a sequence (λj ) of real numbers such that T = U −1 AU . In
that case (λj ) has at most 0 as an accumulation point and the eigenspaces Ker (λj − A)
are finite dimensional for λj ̸= 0.
Note, that the example A is a multiplication operator: ℓ2 is L2 (N) for the measure
µ(ej ) = 1 and with v(j) = λj the operator A has the form Aϕ = vϕ.
In the following we need a generalization of the example in Remark 4◦ of F.3 which
includes the example just described.
Example F.22 (General Multiplication Operator). Let (Ω, Σ, µ) a measure space and
let v : Ω → C be a measurable function. The multiplication operator M = Mv in
H := L2 (Ω, µ) is defined by
M ϕ := vϕ , ϕ ∈ D(M ) ,
Z
D(M ) := {ϕ ∈ H | |v(x)ϕ(x)|2 dµ(x) < ∞} .
Ω
125
T is compact if the image T (B(0, 1)) of the unit ball has a compact closure in H.
F.2 Self-Adjoint Operators 361

It can be proven that D(M ) is dense in H. Hence, the adjoint exists. As in the
example in Remark 4◦ of F.3 one shows that Mv∗ = Mv . Therefore, Mv is self-adjoint
if and only if v = v, i.e. v is real-valued. Mv is bounded if and only if v is essentially
bounded,i.e. there is a subset N ⊂ Ω of measure 0 such that sup{|v(q)| | q ∈ Ω \ N } <
∞..
Theorem F.23 (Spectral Theorem – Multiplication Form). Let (A, D(A)) be a self-
adjoint operator on a separable Hilbert space H. Then there exists a (Σ-finite) measure
space (Ω, Σ, µ), a measurable function v : Ω → R and a unitary operator U : H →
L2 (Ω, µ) with

• ϕ ∈ D(A) ⇐⇒ U ϕ ∈ D(Mv ),

• U AU −1 = Mv on D(Mv ).

H ⊃ D(A) A / H
O
U U −1 U
 
H ⊃ D(M ) M / H

Example F.24 (Hamiltonian of a Free Particle). In classical mechanics the dynamics

of the free particle in Q = Rn is determined by the Hamiltonian H : M = T Q → R,
the energy:
1 X
H(q, p) = (pj )2 .
2m
We set m = 1 in the following. The quantum mechanical counterpart is the Hamilto-
nian n  2
1X ∂
Ĥ = −i ,
2 1 ∂pj

in H = L2 (Rn ). We obtain Ĥ from canonical quantization in elementary Quantum

Mechanics. The Hamiltonian Ĥ is essentially the negative of the Laplacian: Ĥ = − 12 ∆.
In order to describe an equivalent multiplication operator we use the Fourier trans-
form F given as
Z
1
F(ϕ)(x) = n ϕ(y)ei⟨x,y⟩ dy , ϕ ∈ E(Rn ) ∩ L2 (Rn ) ,
(2π) 2 Rn

defining a unitary map F : L2 (Rn ) → L2 (Rn ). For differentiable ϕ ∈ L2 (Rn ) partial

differentiation −i ∂x∂ k and multiplication by xk is interchanged by F in the folloewing
way:
∂ϕ
−i k = F −1 xk Fϕ = F −1 Mxk F ϕ .


∂x
Therefore,
−∆ = F −1 Mv F , Mv = F(−∆)F −1 , (91)
362 F. Quantum Mechanics

with v = ||x||2 and we have a concrete example of a self-adjoint operator being unitarily
equivalent to a multiplication operator. Note, that we have neglected to describe the
domain and check the self-adjointness of −∆. But this can be done now using (91).
We also can determine the spectrum of −∆ as an application of the subsequent
lemma: σ(−∆) = [ 0, ∞ [.
Lemma F.25. The spectrum of the multiplication operator Mv is the essential range
of v defined as

essrg(v) := {λ ∈ C | for all ε > 0 : µ({ω ∈ Ω | |v(ω) − λ| < ε}) > 0} :

σ(Mv ) = essrg(v).
Observation F.26. Note that the spectral theorem allows one to introduce a Func-
tional Calculus for self-adjoint operators: If A is a self-adjoint operator in H with
U AU −1 = Mv (according to Theorem F.23) and f : R → C a measurable function.
Then an operator f (A) can be defined in the following way:

D(f (A)) := {ϕ ∈ H | (f ◦ v)U ϕ ∈ L2 (Ω, Σ, µ)} ,

f (A)ϕ := U −1 Mf ◦v U ϕ , for ϕ ∈ D(f (A)) .

We come back to the functional calculus in Observation F.37 after having introduced
spectral families in order to formulate the spectral theorem in the spectral measure
form.

To motivate the spectral theorem in the spectral measure form let us go back again
to the finite dimensional case. Let A be a self-adjoint (i.e. symmetric) operator in the
n-dimensional Hilbert space H. Then the eigenspaces of A yield a decomposition of
H: Let P1 , P2 , . . . , Pk be the orthogonal projections Pj : H → H onto the pairwise
orthogonal eigenspaces of A corresponding to the eigenvalues λq , . . . , λk . Then
k
X
A= λj Pj .
1

Example F.27. A more general situation is the decomposition of a general separable

L H
into closed subspaces Hj with projections Pj : H → H with Im Pj = Hj and H = Hj .
Given a sequence (λj )j∈N ) of pairwise different real numbers the definition
X X
Aϕ := λj Pj ϕ , ϕ ∈ D(A) := {ϕ | λj Pj ϕ converges}

yields a self-adjoint operator described by projections.

The sum can be given by a suitable integral as well. This approach can be gener-
alized to the concept of a spectral family, or spectral resolution of the identity. For the
formulation of the concept we need some elementary results on orthogonal projections.
F.2 Self-Adjoint Operators 363

Recall that a projection (more precisely an orthogonal projection) in the Hilbert

space H is a bounded operator P : H → H with P ◦ P = P . Then Q := I − P is again
a projection and I = P + Q. As a consequence, Ker P and Im P are closed subspaces
with H = Ker P ⊕ Im P . A projection is self-adjoint. Each closed subspace V ⊂ H has
a unique orthogonal complement W and defines a unique projection P with V = Im P
and W = Ker P .
For a second projection Q we define P ≥ Q if and only if Ker P ⊂ Ker Q which
is equivalent to Im Q ⊂ Im P . In case of P ≥ Q the equality Q = P ◦ Q = Q ◦ P
holds and the difference P − Q is again a projection, namely the projection onto the
subspace Im P ∩ Ker Q. The following result is easy to show.
Proposition F.28. Let (Pk ) be a increasing sequence of projections. Then the projec-
tion P induced by the closed subspace
∞
[
Im P = Im Pk
0

is the pointwise limit of the Pk : P ϕ = lim Pk ϕ. Analogously, for a decreasing sequence

(Pk ) of projections. In that case the limit is the projection onto
∞
\
Im P = Im Pk .
0

Definition F.29 (Spectral Family). A spectral family is a map E : R → B(H), often

written in the form (Eλ ) = (Eλ )λ∈R , with the following properties:

1. Each Eλ is a projection.

2. Eλ ≤ Eµ for λ ≤ µ , λ, µ ∈ R.

3. limλ↘−∞ Eλ = 0 , limλ↗∞ Eλ = 1.

4. limε↘0 Eλ+ε = Eλ for all λ ∈ R.

(Eλ )λ∈R is also called spectral resolution of the identity.

The support Supp(Eλ ) of (Eλ ) is the interval

I = Tr(Eλ ) = {λ ∈ R : Eλ ̸= 0 or Eλ ̸= 1}.

Eλ is called to have bounded support when this interval is bounded.

Note that for increasing (resp. decreasing) sequences of real numbers (λk ) the cor-
responding projections (Eλk ) is increasing (resp. decreasing), so that Proposition F.28
is applicable. The convergence in 3. and 4. is meant is the sense of this proposition, it
is pointwise convergence.
364 F. Quantum Mechanics

Example F.30. Let P be a projection and a, b ∈ R , a < b. Set Eλ = 0 for λ < a,

Eλ = P for a ≤ λ < b and Eλ = 1 for b ≤ λ. The spectral family has [a, b] as its
support. The spectral theorem in finite dimension leads to a similar spectral family
with finitely many projections Pk
Example F.31. The spectral family for the Example F.27 and generalizing Example
F.21 is the following X
Eλ = λj Pj .
λj ≤λ

Example F.32. As in the Example F.22, let (Ω, Σ, µ) a measure space and let v :
Ω → C be a measurable function defining the multiplication operator M = Mv in
H := L2 (Ω, µ), M ϕ = vϕ , ϕ ∈ D(M ). Assume v to be real-valued. Denote S(λ) :=
{ω ∈ Ω | v(ω) ≤ λ} and let χS(λ) : Ω → {0, 1} be the corresponding characteristic
function of S(λ). Then
Eλ ϕ := χS(λ) ϕ , ϕ ∈ H
is a spectral family. (It is the spectral family of M , c.f. Example F.35.)

Recall the Riemann-Stieltjes integral: Let w : R → R be weight function which we

assume
R to be increasing and continuous from the right. The Riemann-Stieltjes integral
f dw for continuous functions f : R → R is defined in essentially the same manner as
the Riemann integral: Z b
f (λ)dw(λ) , a < b ,
a
is the limit of sums n
X
f (tj )(w(tj ) − w(tj−1 )) ,
1

with a = t0 < t1 > . . . < tn = b where the length sup(tj − tj−1 ) tends to zero.
In order to introduce the concept of integrating a spectral family (Eλ ) we use the
weights
wϕ (λ) := ⟨ϕ, Eλ ϕ⟩ ,
wϕ is decreasing, continuous from the right and bounded. wϕ determines a measure
wϕ dλ = dwϕ on R and so we obtain a well-defined integral
Z Z
f (λ)wϕ dλ = f (λdwϕ (λ) .
R R

(Without measure theory this integral can also be expressed by the (improper)
Riemann-Stieltjes integral for continuous f . )
This integral is also denoted by
Z Z Z Z
f (λ)⟨ϕ, Eλ ϕ⟩dλ = f (λ)d⟨ϕ, Eλ ϕ⟩ = f (λ)⟨ϕ, dEλ ϕ⟩ = f (λ)dEλ ϕ .
F.2 Self-Adjoint Operators 365

Proposition F.33 (Integrating a Spectral Family). Let (Eλ )λ∈R be a spectral family.
For each measurable function f : R → R one obtains the operator Ê(f ) in H in the
following way: Z
D(Ê(f )) := {ϕ ∈ H | |f |2 ⟨ϕ, Eλ ϕ⟩dλ < ∞} ,
R
Z
Ê(f )ϕ := f (λ)⟨ϕ, Eλ ϕ⟩dλ , ϕ ∈ D(Ê(f )) .
R

Ê(f ) is self-adjoint and we write

Z Z
Ê(f ) = f (λ)dEλ = f dEλ .

Ê(f ) is bounded whenever f is bounded.

Example F.34. Applied to the Examples F.27, F.21 and F.31 let f : R → R. Then
X
Ê(f ) = f (λj )Pj .
j

Example
Pm F.35. In the Example F.32 let f : R → R a step function, i.e. f (λ) =
c χ
1 j Ij with finitely many pairwise disjoint intervals. The integral is
Z  Xm
Ê(f )ϕ(ω) = f (λ)dEλ ϕ (ω) = cj (v(ω))ϕ(ω) = f ◦ v(ω)ϕ(omega) ,
1

for ϕ ∈ H, so Ê(f )ϕ)(F ◦ v)ϕ. As a consequence, for any measurable f

D(Ê(f )) = {ϕ ∈ H | (f ◦ v)ϕ ∈ H}
Ê(f )ϕ = (f ◦ v)ϕ .
In particular, with f = idR :
Z
Ê(id) = λdEλ = Mv .

As a generalization we have:
Theorem F.36 (Spectral Theorem – Spectral Measure Form). To each self-adjoint
operator A ∈ SA, there corresponds a unique spectral family (Eλ )λ∈R such that A =
Ê(idR ), i.e.
Z
A = λdEλ .

The spectral family is given by

Z b+ε
1
⟨ψ, (Eb − Ea )ϕ⟩ = lim lim ⟨ψ, (R(λ − iεA) − R(λ + iεA)) ϕ⟩ .
δ↘0 ε↘0 2πi a+δ
366 F. Quantum Mechanics

With the aid of the Spectral Theorem – Multiplication Form F.23 this version of
the spectral theorem can be proven using the preceding Example F.35.

Observation F.37. The functional calculus introduced in Observation F.26 can now
be reformulated. If Eλ is the spectral family of a self-adjoint operator A, then for
measurable functions f : R → R the corresponding self-adjoint operator is

Z
f (A) = f (λ)dEλ .

R
For each spectral family a self-adjoint operator is defined as A := λdEλ . The
corresponding one-parameter group of unitary operators is, in accordance with Stone’s
Theorem, but now using functional calculus the family Us = e−isλ dEλ of unitary
R

operators.
A final remark concerning the use of the term ”spectral measure form”: A spectral
family induces an abstract projection valued measure on R. In fact for interval J :=
[a, b[ one can define p(J) := Eb − Ea to obtain a map on the set B(R) of Borel subsets
of R whose values are projections: p : B(R) → B(H). This is the so called projection
valued measure induced by the spectral family. Vice versa, any spectral family can be
induced by a spectral measure Eλ := p(] − ∞, λ]).

F.3 Canonical Commutation Relations

Definition F.38. The Canonical Commutation Relations (CCR) or Heisenberg

commutation relations are

[Pj , Pk ] = 0 , [Qj , Qk ] = 0 , [Pj , Qk ] = −iδjk , for 1 ≤ j, k ≤ n ,

for a set P1 , . . . , Pn and Q1 , . . . , Qn of operators or elements of a Lie algebra.

In Quantum Mechanics one is interested to realize the CCR by linear operators in

a Hilbert space H. In fact, in some treatises on the foundation of Quantum Mechanics
a realization of the CCR is part of the postulates, for example in the following form:
Postulate For a quantum mechanical system in which cartesian coordinates qj
with corresponding momenta pj , j = 1, . . . , n, are represented by self-adjoint
operators Qj and Pj in a Hilbert space H the following realization of the CCR
has to be satisfied

[Pj , Pk ] = 0 , [Qj , Qk ] = 0 , [Pj , Qk ] = −iδjk idH , for 1 ≤ j, k ≤ n .

F.3 Canonical Commutation Relations 367

Observation F.39. We have seen a realization of the CCR at several places in these
lecture notes, for instance in the context of prequantization in (35) or later in the
case of using natural polarizations in the simple case (see Examples 10.12). We recall
a special case, the so-called Schrödinger Representation: The Hilbert space is
H = L2 (Rn , λ) (λ = dq Lebesgue measure), and the Qj , Pj are the unbounded self-
adjoint operators in H with

Qj (ϕ) := qj ϕ ,
∂
Pj (ϕ) : −i j ϕ ,
∂q

j = 1, . . . , n , where the ϕ ∈ H are in the respective domains D(Pj ) , D(Qj ) ⊂ H.

Note, that the Schrödinger representation is a representation of the Heisenberg

algebra hsn (see Example C.14) by self-adjoint operators.

Are there simpler realizations of the CCR?

It is easy to see that for any realization of the CCR the Hilbert space H has to
be infinite dimensional. The identity P Q − QP = −i · idH (CCR with n = 1!) is not
possible for linear maps in a d-dimensional Hilbert space H ̸= {0} , since the trace of
a commutator is zero while the trace of idH is d ̸= 0.

Furthermore, P Q − QP = −i · idH (with self-adjoint P, Q) can only hold for un-

bounded operators. To see this, we need a formula for iterated commutators: We define
P (m) (Q) by recursion, P (0) (Q) := Q and P (m+1) (Q) := [P, P (m) (Q)], and obtain the
following formula by induction.

Lemma F.40.
m m−k
1 (m) X 1
k (−1)
P (Q) = P Q P m−k .
m! k=0
k! (m − k)!

Proof. The formula holds for m = 0. The induction step m → m + 1:

 
1 1 1 1 (m) 1 (m)
P (m+1) (Q) = [P, P (m) (Q)] = P (Q) − P (Q)P
(m + 1)! (m + 1)! m+1 m! m!
368 F. Quantum Mechanics

By the induction hypothesis:

m m
!
1 X 1 k (−1)m−k m−k X 1 k (−1)m−k m−k
= P P Q P − P Q P P
m+1 k=0
k! (m − k)! k=0
k! (m − k)!
m+1 m
!
1 X 1 k (−1)m−k+1 m−k+1 X 1 k (−1)m−k m−k+1
= P Q P − P Q P
m+1 k=1
(k − 1)! (m − k) + 1! k=0
k! (m − k)!
 
1 1 m+1
= P Q
m + 1 m!
m m
!
m−k+1 m−k
1 X 1 (−1) X 1 (−1)
+ P kQ P m−k+1 − P kQ P m−k+1
m + 1 k=1 (k − 1)! (m − k) + 1! k=1
k! (m − k)!
m
 
1 (−1)
− Q P m+1
m+1 (m)!
m
1 m+1 1 X k k (−1)m+1−k m+1−k
= P Q+ P Q P
(m + 1!) m + 1 k=1 k! (m + 1 − k)!
m
!
X 1 k (−1)m+1−k (m + 1 − k) m+1−k (−1)m+1
+ P Q P + QP m+1
k=1
k! (m + 1 − k)! (m + 1)!
m 
m + 1 X 1 k (−1)m+1−k m+1−k (−1)m+1

1 m+1
= P Q+ P Q P + QP m+1
(m + 1!) m + 1 k=1 k! (m + 1 − k)! (m + 1)!
m+1
X1 (−1)m+1−k m+1−k

k
= P Q P
k=0
k! (m + 1 − k)!

Proposition F.41. Let P, Q be self-adjoint operators in the Hilbert space H and define
U (t) := eitP . Then
∞
X (it)m (m)
U (t)QU (−t) = P Q.
m=0
m!
In particular, if P Q − QP = −i · idH , it follows that U (t)QU (−t) = Q + t · idH .

Proof. Formally, we have

X (it)k X (−it)n
U (t)QU (−t) = P kQ Pn ,
k
k! n
n!

and using Lemma F.40

X (it)k X (−it)n X X (it)k (−it)n n X (it)m (m)
P kQ Pn = P kQ P = P (Q)
k! n! m k+n=m
k! n! m
m!
F.3 Canonical Commutation Relations 369

we obtain the first result. (The formal calculation can be justified by applying it to
vectors ϕ in the range of a spectral projection of Q.) If now [P, Q] = −i holds, the
brackets P (m) Q vanish for m > 1 and the sum
X (it)m
P (m) (Q)
m
m!

reduces to Q + (it)P (1) Q = Q + t.

Corollary F.42. Self-adjoint operators P, Q on a Hilbert space H with P Q − QQ = −i

are unbounded with spectrum σ(Q) = σ(P ) = R.

Proof. If Q were bounded then λ − Q would be invertible as a bounded operator for

|λ| > ∥Q∥. Since Q + t is unitarily equivalent to Q for all real t the operator Q would
have empty spectrum contradicting the fact that self-adjoint operators have non-empty
spectrum. For Q as an unbounded operator each t ∈ R turns out to be in the spectrum
of Q, since Q + t is unitarily equivalent to Q + s for all s ∈ R. σ(P ) = R by using the
symmetry (P, Q) 7→ (Q, −P ).

We now introduce the so-called Weyl relations.

Proposition F.43. Let P, Q be self-adjoint operators in the Hilbert space H satisfying

P Q − QP = −i · idH , and define U (t) := eitP , V (s) := eisQ . Then

U (t)V (s) = eist V (s)U (t) .

Proof. Generalizing Lemma F.40 one gets eitP f (Q)e−itP = f (Q+t) for any measurable
function on R using the functional calculus or some induction formulas as above. As a
result
eitP e+isQ e−itP = e+is(Q+t) = eist e+isQ .

This is the ”integrated” version of the CCR in case of n = 1. In order to formulate

the Theorem of Stone-von Neumann we need some definitions:

Definition F.44. A pair of strongly continuous unitary groups (U (t)) and (V (s)) on
a Hilbert space H is called to represent the Weyl relations, if U (t)V (s) = eist V (s)U (t)
holds for all s, t ∈ R.
The representation is called Irreducible, if there is no non-trivial closed linear
subspace H0 ⊂ H such that U (t)H0 ⊂ H0 , V (s)H0 ⊂ H0 for all s, t ∈ R.
Two representations (U (t), V (t)) on H and (U ′ (t), V ′ (s)) on another Hilbert space
H are called unitarily equivalent if there exists a unitary map Φ : H → H′ with
′
370 F. Quantum Mechanics

U (t) = Φ−1 U ′ (t)Φ and V (s) = Φ−1 V ′ (s)Φ, i.e. if the following diagram is commutative:
U (t)
H / HO
V (s)
Φ Φ−1
 U ′ (t)
H′ / H′
V ′ (s)

Remark F.45. A representation of the Weyl relations by U (t), V (s) is essentially a

projective unitary representation of the abelian group R2 :
W : R2 → U(P) , (p, q) 7→ [U (p)V (q)] ,
where [U ] ∈ U(P) denotes the unitary projective operator determined by the unitary
operator U ∈ U(H) , P = P(H). Recall, that a projective unitary representation can
be lifted to a central extension of the group (see Remark C.11). In our situation we
have as central extension of R2 the variant of the Heisenberg group HSpol
1 = U(1) × R
2

with projection π : U(1) × R2 → R2 and exact sequence

π
1 −→ U(1) −→ HSpol 2
1 −→ R −→ 0 .

For the corresponding unitary representation Ŵ : HSpol

1 → R2 the following diagram
commutes:
1 / U(1) / HSpol π / R2 /1
1

id Ŵ W
  γ̂

1 / U(1) / U(H) / U(P) / 1

The definition of Ŵ in case of the Weyl relations is

1
Ŵ (eir , p, q) := eir e− 2 ipq U (p)V (q) , (eir , p, q) ∈ HSpol ∼ 2
1 = U(1) × R .

Proposition F.46. The Schrödinger representation of the Weyl relations on the

Hilbert space L2 (R) given by U (t)ψ(q) = ψ(q + t) and V (s)ψ(q) = eisq ψ(q) is irre-
ducible.

Proof. Let H0 ̸= {0} be an invariant sub-Hilbert space of H and ϕ ∈ H0 ϕ ̸= 0. Let

ψ be an arbitrary element of the orthogonal complement H1 of H0 . Since H1 is also
invariant, we have U (t)ψ, V (s)U (t)ψ ∈ H1 . Hence, ⟨ϕ, V (s)U (t)ψ⟩ = 0 , i.e.
Z
ϕ̄(q)eisq ψ(q + t)dq = 0.
R

Since the Fourier transformation

Z
F : H → H , g 7→ F(g)(s) := c eisq g(q)ds ,
R

(c some constant) is bijective, the functions ϕ̄(q)ψ(q + t) vanish for all t ∈ R. We

conclude ψ = 0, and the orthogonal complement H1 of H0 is zero, hence H = H0 .
. 371

Theorem F.47 (Stone-von Neumann). Any irreducible representation of the Weyl

relations is unitarily equivalent to the Schrödinger representation.

Proofs can be found, for example, in in [Hal13] or [Spe20]. It is interesting that

Hermann Weyl was the first to formulate the result in his book on group theory and
quantum mechanics in 1928 [Wey31], but he did not provide a proof. Marshall Stone
shortly after the appearance of the book pointed out that the result needs a proof
and presented in 1930 the unitary map yielding the unitary equivalence. He was not
providing a proof that his unitary map really does the job. A thorough proof was
given later by John von Neumann in 1931. This and other interesting aspects of the
Stone-von Neumann Theorem can be found in the article of J. Rosenberg, see [Ros04].
The corresponding result for n degrees of freedom is of the same nature, and has
essentially the same proof. Transforming a representation of the Weyl relations

Uk (t)Vj (s) = eδjk ist Vj (s)Uk (t) , j, k = 1, . . . , n ,

besides Uk (t)Uj (t′ ) = Uj (t′ )Uk (t) as well as Vk (s)Vj (s′ ) = Vj (s′ )Vk (s) , into a unitary
representation of the Heisenberg Lie group HSn as in Remark F.45, the result can be
described in the following way.

Theorem F.48 (Stone-von Neumann). Any irreducible unitary representation of the

Heisenberg group is unitarily equivalent to the Schrödinger representation.

This is a remarkable result. In general, when investigating Lie groups as symmetry

groups in physics, geometry or number theory there appear plenty of nonequivalent
unitary and irreducible representations with finite and with infinite dimensional Hilbert
spaces. For applications and also to obtain a good overview, one is quite content to
give a complete list of irreducible unitary representations for a given Lie group (or,
more general, for a topological group). The Heisenberg group is different. There is
only one irreducible unitary representation up to unitary equivalence and it is infinite
dimensional!

Summary:
372 REFERENCES

References

[AE05] S.T. Ali,and M. Englis. Quantization methods: a guide for physicists and analysts.
Reviews in Mathematical Physics 17.4 (2005), 391–490.

[Arn89] V.I. Arnold Mathematical Methods of Classical Mechanics, 2nd edition Springer
(1989)

[AM78] R. Abraham and J.E. Marsden Foundations of Mechanics Benjamin/Cummings,

(1978).

[APW91] S. Axelrod, S. Della Pietra, and E. Witten. Geometric quantization of Chern-

Simons gauge theory. J. Diff. Geom. 33 (1991), 787–902.

[Bla77] R. J. Blattner The metalinear geometry of non-real polarizations. In: Differential

Methods in Mathematical Physics. Lecture Noted in Mathematics 570 Springer (1977),
11–45.

[BW97] S. Bates and A. Weinstein Lectures on the geometry of quantization Berkeley Math-
ematics Lecture Notes 8, AMS (1997).

[Bot91] R. Bott. Stable bundles revisited. Surveys in Differential Geometry (Supplement to

J. Diff. Geom.) 1 (1991), 1–18.

[Bry93] J.-L. Brylinski. Loop Spaces, characteristic classes, and geometric quantization ,
Birkhäuser-Verlag (1993).

[GS77] V. Guillemin and S. Sternberg. Geometric Asymptotics. AMS Math. Surveys 14,
1977. Revised 1990.

[GS84] V. Guillemin and S. Sternberg. Symplectic techniques in physics. Cambridge Univer-

sity Press, 1984.

[GS10] V. Guillemin and S. Sternberg. Semi-classical analysis. Preprint,

http://www-math.mit.edu/ vwg/semiclassGuilleminSternberg.pdf, 2010.

[Hal13] B.C. Hall Quantum Theory for Mathematicians. Springer-Verlag, 2013.

[HJJS08] D. Husemöller, M. Joachim, B. Jurco, and M. Schottenloher. Basic Bundle Theory

and K-Cohomological Invariants. Lect. Notes in Physics 726, Springer, 2008.

[HN91] J. Hilgert and K.-H. Neeb. Lie Gruppen und Lie Algebren. Vieweg, 1991.

[Hur82] N.E. Hurt Geometric Quantization in Action, Reidel Company (1982).

[Huy05] N. Huybrechts Complex Geometry, Springer-Verlag (2005).

[Kos70] B. Kostant Quantization and unitary representations. I. Prequantization. Lecture

Notes in Math. bf 170 (1970), 87–208.

[Kir76] A. A. Kirillov. Theory of Representations. Springer Verlag, 1976.

REFERENCES 373

[Ler12] E. Lerman Geometric quantization: a crash course Contemp. Math583 (2012),

147–174.

[Nas91] C. Nash. Differential Topology and Quantum Field Theory, Academic Press (1983).

[LM87] P. Libermann and C.-M. Marle. Symplectic geometry and analytical mechanics, Rei-
del Company (1987).

[Nei83] B. O’Neill. Semi-Riemannian geometry - With applications to relativity, Academic

Press (1983).

[Poo81] W.A. Poor Differential Geometric Structures. McGraw-Hill Book Company, 198.

[PS86] A. Pressley and G. Segal. Loop Groups. Oxford Univ. Press, 1986.

[Put93] M. Puta Hamiltonian Mechanical Systems and Geometric Quantization, Kluwer

Publications (1993).

[Raw78] J.H. Rawnsley On the pairing of polarizations. Comm. Math. Phys. 58 (1978), 1–8.

[Raw79] J.H. Rawnsley A nonunitary pairing of polarizations for the Kepler problem. Trans.
Amer. Math. Soc. 250 (1979), 167–180.

[Ros04] J. Rosenberg A Selective History of the Stone-von Neumann Theorem. Contemporary

Mathematics 365 (2004), 331–353

[RS80] M. Reed and B. Simon. Methods of modern Mathematical Physics, Vol. 1: Functional
Analysis. Academic Press, 1980.

[RuS13] G. Rudolph and M. Schmidt Differential Geometry and Mathematical Physics I and
II, Springer-Verlag (2013).

[Sch95] M. Schottenloher. Geometrie und Symmetrie in der Physik. Vieweg, 1995.

[Sch08] M. Schottenloher. A Mathematical Introduction to Conformal Field Theory. Springer,

2008.

[Sim68] D. Simms. Lie Groups and Quantum Mechanics. Lecture Notes in Math. 52, Springer
Verlag, 1968.

[SW76] D. Simms and N.N.J. Woodhouse. Lectures on Geometric Quantization. Lecture

Notes in Physics 53, Springer Verlag, 1976.

[Sni80] J. Sniaticky Geometric Quantization and Quantum Mechanics, Springer-Verlag

(1980).

[Spe20] R. Speicher Mathematical Aspects of Quantum Mechanics. Script, Universität des

Saarlandes, 2020.

[Tuy16] G.M. Tuynman The metaplectic correction in geometric quantization. Journal of

Geometry and Physics 106 (2016), 401–426.
374 REFERENCES

[Wei12] J. Weidmann Linear Operators in Hilbert Spaces, Springer-Verlag (2012).

[Wey31] H. Weyl Gruppentheorie und Quantenmechanik, Hirzel-Verlag (1931).

[Wit89] E. Witten. Quantum field theory and the Jones polynomial. Commun. Math. Phys.
121 (1989), 351–399.

[Woo80] N. Woodhouse. Geometric Quantization. Clarendon Press, 1980.

[Woo91] N. Woodhouse. Geometric Quantization. 2nd edition Oxford University Press, 1991.
REFERENCES 375

Sign Conventions

Let (M, ω) be a symplectic manifold with Poisson bracket { , }, and let F, G ∈ E(M )
be classical observables, ı.e. differentiable functions.
In the first table we list the possibilities for the signs and assign the attributes ”A”
and ”B”.
In the second table it is reported how the signs are used in the literature which we
cite in these lecture notes.

Table 1: Possible sign conventions

A B
Hamiltonian vector field XH by iXH ω = dH iXH ω = −dH
Poisson Bracket {F, G} = ω(XF , XG ) −ω(XF , XG )
Representation [XF , XG ] = −Xω(XF ,XG ) Xω(XF ,XG )
Lie Derivative LXF G = XF G = −{F, G} {F, G}

Table 2: Usage

Source citation XH {,} =⇒ ±X{F,G} ±XF G

This course, Abraham-Marsden [AM78] A A =⇒ A A
Sniaticky, Puta [Sni80],[Put93] B B =⇒ A B
Woodhouse, Liberman-Merle [Woo80],[LM87] B A =⇒ B B
Brylinski [Bry93] A B =⇒ B A
Index
r-density, 163 circle bundle, 88
1-form, 211 circle group, 88
closable operator, 334
adjoint operator, 329, 333 closed curve, 84
almost complex structure, 130, 132, 276 closed operator, 334
angular momentum, 96 closed surface, 84
antiholomorphic polarization, 129
coadjoint representation, 29
antiunitary, 328
cochain, 314
associated fibre bundle, 301
cocycle, 46, 296, 298, 314
associated vector bundle, 163
coefficients, 312
axiom of choice, 327
cohomology, vii
Balmer series, 97 colimits, 315
Bargmann, 150 commutator, 259
Bargmann representation, 143 compact operator, 339
Bargmann space, 219, 270 compatible connection, 89, 92
Bargmann-Fock representation, 143, 144 compatible connection, with holomorphic
BKS-pairing, 186 structure, 91
Bohr-Sommerfeld variety, 157 compatible polarizations, 202
boundary, 313 compatible symplectic form, 130
bounded operator, 325 complete vector field, 103, 249
bundle chart, 242, 247, 248, 274 completely integrable system, 22, 122
bundle of r-densities, 163 completeness, 17
complex m-form, 162
called total derivative, 243
complex Lagrangian subspace, 124
canonical bundle, viii, 170, 210
complex manifold, 272
canonical commutation relations, 104, 137,
complex polarization, 126
142, 144, 288, 345
configuration space, 1, 29
canonical coordinates, 13
connecting homomorphism, 59
canonical Hilbert space, 169
connection, 61
canonical line bundle, 151
canonical quantization, 33 connection form, 306
canonical transformation, 11, 29 connection form, global, 65, 71
Cartan’s magic formula, 260 connection form, local, 64, 71
category theory, 315 constant of motion, 4
Cauchy integral, 263 continuous spectrum, 336
CCR, canonical commutation relations, 40, cotangent bundle, 248
142, 345 covariant derivative, 62, 71
central charge, 288 covariantly constant, 172, 179, 180
central extension, 20, 288 curvature, 60, 74
of a group, 283 curvature form, 81
charged particle, 23 curvature operator, 81
Chern class, 113 cyclic summation, 17
Chern connection, 92 cylinder, 116

376
INDEX 377

Darboux’s theorem, 13 entire two form, 86, 320

deficiency index, 337 equations of motion, 1
deformation of the symplectic structure, 23 equations of motion in Poisson form, 16
degree, 257 equivalent line bundles with connection,
degree of freedom, 1 108
densely defined operator, 333 equivariance, 210, 211, 299, 301
density, 163 equivariant, 228, 233
deRham cohomology, 85 essential range, 341
derivation, 9, 259 essentially self-adjoint operator, 334
derivative of a map, 243 Euler operator, 150
descend of densities, 178 Euler-Lagrange equations, 30
determinant bundle, 170 exact sequence, 283
determinant line bundle, 211 examples
diffeomorphism, 239 harmonic oscillator, 149
differentiable quotient, 256 line bundles on projective space, 303
differential equation, 248 momentum phase space, 40, 94, 103,
differential form, 9 127, 129, 139, 143, 184, 199, 205,
differential of a map, 243 219
differential quotient, 251, 255 simple phase space, 184
differentiation operator, 338 twisted momentum space, 95
Dirac condition, vi, 33 exponential map, 291
direct limit, 315 extension, 333
directional derivative, 7 of a group, 283
directly quantizable, 181 exterior derivative, 9
directly quantizable observable, vii, 146,
153 fibre bundle, 301
distribution, 119 fibre derivative, 31
horizontal, 121 fibre manifold, 240
radial, 121 first integral, 4, 5, 16, 20
distribution, generalized function., 158 flat connection, 114
divergence, 182 flat line bundle, 114
divergence of density, 182 flow, 249
domain, 263, 333 foliation, 120
double covering, 211 Fourier transform, 41, 205
dual bundle, 116 frame bundle, 64, 298
dual line bundle, 58 frame field, 171
dynamical system, 248 free, 293
free abelian group, 313
effective, 293 free particle, 340
Ehresmann connection, 71 Fubini’s theorem, 271
eigenspace, 335 functional calculus, 341, 345
eigenvalue, 335 fundamental field, 66, 99, 289, 304
eigenvalues, 150 fundamental group, 112
eigenvector, 335
endomorphism bundle, 81 gauge transformation, 118
energy representation, 155 general linear group, 304
378 INDEX

generalized function, 158 holomorphic vector field, 274

generalized momenta, 13 holomorphic, partially, 263
generalized section, 158 holonomy group, 156
graded commutative, 257 holonomy representation, 118
graph, 334 homogeneous coordinates, 51, 303
homomorphism of line bundles, 44
Hairy Ball Theorem, 53 Hopf fibration, 293
half-density pairing, 202 horizontal bundle, 68, 71, 74, 306
half-density quantization, 69, 101, 161, horizontal distribution, 121, 136
190, 200, 209 horizontal lift, 70, 74
half-form, 214 horizontal polarization, 136, 143, 154
half-form correction, 152 horizontal transport, 74
half-form quantization, 69, 101, 186, 209 hydrogen atom, 38, 95
Hamiltonian, 331 hyperplane bundle, 55, 116
Hamiltonian frame field, 171
Hamiltonian function, 1 identity theorem, 267
Hamiltonian system, 15, 256 implicit mapping theorem, 256
Hamiltonian vector field, 14 inductive limit, 315
harmonic oscillator, 1, 123, 145, 149 infinitesimal generator, 101, 197, 249, 332
Hartogs, 263 infinitesimal symmetry, 278
Hartogs’ extension theorem, 267 inner product, 38
Heisenberg algebra, 288 integrable almost complex structure, 132
Heisenberg commutation relations, 345 integrable distribution, 120
Heisenberg group, 285 integral curve, 248
Heisenberg representation, 137, 143 integral domain, 267
Hermitian connection, 89 integral manifold, 120
Hermitian form, 87 integrality, 42
Hermitian line bundle, 87 integrality condition, 74, 84, 85
Hermitian matrix, 339 interior derivative, 258
Hermitian metric, 87, 324 interior product, 258
Hermitian scalar product, 324 irreducible, vii, 104, 348
Hermitian structure, 60 isolated singularity, 269
Hilbert space, 33, 270, 285, 324 isometry, 282
Hilbert space homomorphism, 325 isomorphism of line bundles, 44
holomorphic k-form, 275 isotropy group, 292
holomorphic chart, 273
Jacobi identity, 17, 19
holomorphic connection, 91
holomorphic continuation, 268 Kähler polarization, 127
holomorphic coordinates, 91 Kepler problem, 95
holomorphic function, 263 kinetic energy, 27
holomorphic line bundle, 56, 91 Koszul connection, 62
holomorphic one form, 274 Kugelsatz, 268
holomorphic polarization, 129, 139, 140,
149, 174, 270 Lagrangian, 29
holomorphic section, 91 Lagrangian distribution, 126
holomorphic vector bundle, 273 Lagrangian system, 29
INDEX 379

Laplace operator, 338 non-relativistic particle, 328

Laplacian, 41, 340 norm topology, 325
leave, 120
Legendre transformation, 32 observable, 33
Leray cover, 317 observables, 1, 8
Levi-Civita connection, 89 obstruction class, 221, 231
Lie algebra, 34, 278, 287 one form, 8, 247
Lie bracket, 287 one parameter group, 249
Lie derivative, 7, 174, 258 one-parameter group of unitary operators,
Lie derivative of a vector field, 146, 250 332
Lie group, 278 open mapping, 267
Lie group action, 292, 301 operator, 333
Lie group homomorphism, 278 operator norm, 325
Lie superalgebra, 260 operator ordering, 151, 186
lift of densities, 178 operator, densely defined, 333
line bundle, 43 orbit, 292
Liouville measure, 102 orbit space, 252, 278
local connection form, 64 orientation bundle, 170
local coordinate, 6, 246 orthogonal group, 281
local one parameter group, 249 orthonormal, 326
local trivialization, 297 othonormal basis in Hlibert space, 327
locally trivial, 304
loop, 82 parallel operator, 77
Lorentz force, 27 parallel transport, 60, 74, 76
lowering operator, 142 parallel transport, along a curve, 76
partial connection, 179, 215
manifold, 238 partial integration, 42
matrix group, 304 partial Lie derivative, 174
matrix Lie group, 278 partially holomorphic, 263
mean value property, 264 phase space, 285
metaframe, 225 Picard group, 58
metaliear structure, 225 piecewise smooth curve, 84
metalinear frame bundle, 209, 224, 225 point spectrum, 336
metalinear group, 212, 224, 284 Poisson algebra, 17, 35
metalinear structure, 228, 231 Poisson bracket, vi, 3, 4, 16
metaplectic structure, 209 Poisson equation, 4
moduli space, 118 Poisson form, 16
momentum operator, 338 polar, 151
momentum phase space, 1, 13, 40, 139, 177, polarization, vii, 38, 104, 118
184, 190 antiholomorphic, 129
morphism of principal fibre bundles, 299 complex, 126
motion, 15, 30 holomorphic, 129, 140, 270
multiplication operator, 338, 339 horizontal, 136, 143, 154
Kähler, 127
natural lift, 99, 197 positive complex, 127
natural system, 30 real, 122, 127
380 INDEX

reducible, 122 real polarization, 122, 127

reducible complex, 127 reducible complex polarization, 127
vertical, 136, 143 reducible foliation, 120
polarized section, 134, 135, 180 reducible polarization, 122
polydisc, 263 regular Lagrangian system, 31
polyradius, 263 relativistic charged particle, 23
position operator, 338 representation space, 99, 101, 134, 138,
positive complex polarization, 127 179, 196
potential, 30 residual spectrum, 336
power series, 265 resolvent operator, 336
prequantization, vi, 35, 42, 86, 135 resolvent set, 336
prequantum Hilbert space, 102 Riemann sphere, 50
prequantum line bundle, vi, 86, 94 Riemannian metric, 30, 89
prequantum operator, 38, 97, 99, 141, 197 Riesz representation theorem, 205
presheaf, 320 right action, 297, 304
principal fibre bundle, 61, 64, 88, 296, 297, right Lie group action, 296
304 Runge-Lenz vector, 96
product bundle, 297
product manifold, 240 Schrödinger equation, 331
product rule, 17 Schrödinger representation, 346
projection, 297, 304, 342 Schrödinger representation, 104, 136
projection valued measure, 345 section, 44, 244, 300
projective manifold, 145 Segal-Bargmann representation, 143
projective space, 253, 285, 303, 328 self-adjoint operator, 33, 103, 324, 329, 334
projective unitary operator, 285 semi-classical limits, 33
proper, 293 semi-Riemannian manifold, 89
pseudo Kähler, 127 semidirect product, 282
pseudometric, 328 sheaf, 321
pseudoscalar, 163 simple curve, 84
pull-back, 174 simple Hamiltonian system, 1
pullback, 65, 257 simple phase space, 11, 38, 40, 94, 103, 127,
129, 136, 139, 141, 143, 161, 184,
quantizable, 181 199, 205, 219, 269
quantizable manifold, vii, 86, 94 simplex, 313
quantization special orthogonal group, 281
of symmetries, 285 spectral family, 342
quantization map, 33 spectral resolution of the identity, 342
quantum mechanical system, 324 spectral theorem, 338
quantum operator, 136 spectral theory, 338
quantum operators, 181 spectrum, 118, 335, 336
quotient manifold, 241, 251 spin structure, 232
quotient topology, 251, 252, 328 square root line bundle, 209
square root of a line bundle, 212
radial distribution, 121, 155 state space, 285
raising operator, 142 state vector, 331
range, 333 Stiefel-Whitney class, 233
INDEX 381

Stokes’ theorem, 84 unitary operators, 325

strong topology, 285, 332 unitary space, 325
structure group, 64, 296, 297 universal property, 240, 241, 251
submanifold, 239
submersion, 255 vector field, 247
superalgebra, 258 velocity phase space, 29
supersymmetry, 258 vertical bundle, 66, 71, 305
support of spectral family, 342 vertical distribution, 121, 136, 143
symmetric, 330 vertical polarization, 136, 143
symmetric operator, 329, 334 vertical vector field, 66
symmetry, 278 wave function, 328
symmetry group, 282 Weierstrass, 269
symplectic form, 3, 10 Weil, André, 86
symplectic group, 130, 132, 282 Weyl relations, 348
symplectic manifold, 11
symplectomorphism, 11, 29 zero section, 64, 298

tangent bundle, 247, 274

tangent map, 243
tangent vector, 241
tangent vector, complex, 274
tautological bundle, 51, 54, 116
Taylor expansion, 266
tensor product, 116
tensor product connection, 114
tensor product of line bundles, 58
the automorphism group, 282
Theorem of Frobenius, 120
Theorem of Stone-von Neumann, 348
topological group, 285
topological quotient, 251, 256
total energy, 20
total space, 43, 297
transition function, 52, 297, 298
transition functions, 46, 295
transition probability, 329
transitive, 292
transversal polarizations, 203
trivial bundle, 62
trivial central extension, 284
trivial line bundle, 44
trivial principal fibre bundle, 300
trivializable, 44
twisted symplectic form, 23

unitary group, 281, 332