Modave Schools and Workshops

Volume 1

First Modave Summer School

in Mathematical Physics

19-25 June, 2005

PROCEEDINGS

Editors:
V. Bouchard
A. Wijns

International Solvay Institute

Modave Schools and Workshops

Volume 1 - First Modave Summer School

in Mathematical Physics
 i

Foreword

While attending the ”Rencontres Mathématiques de Glanon”, a friendly, small scale math confer-
ence in Bourgogne (France), some of us thought about organizing a similar summer school which
would be more “physics oriented” but with the same spirit: a relaxed and informal atmosphere,
a nice location in the countryside, no internet access, mostly Ph.D. students or post-docs as
participants.
The idea was pursued and finally the school was held in Modave (Belgium), in the hostel “Les
Cent Fontaines” situated at the border between the Condroz and the Ardennes. Practically, the
first “Modave Summer School in Mathematical Physics” was aimed at providing mathematical
tools which are useful for researchers in theoretical physics of fundamental interactions and which
are generally supposed to be known but too rarely explained in details. Therefore the courses
had to begin with the basics, to be synthetic and self-contained. Besides that, the lectures were
mostly given by the students themselves, in their own field of interest. Indeed, the Modave
experience was intended to be a school made by and for both post-grad students and post-docs,
which were the positions of all members of the organizing committee. In addition to these specific
features, most of the afternoons were left free, in order to allow spontaneous discussions and/or
meetings for questions and answers in connection with the courses in the morning. Pastoral
walks and cultural visits were also welcome in the afternoon, promoting some inspiration or
informal exchanges.
We would like to thank all lecturers and participants for their contribution to the success of
the school. The International Solvay Institutes are acknowledged for their logistic support and
for printing these proceedings. We hope these lecture notes would be helpful to other young
researchers in mathematics or physics.

The organizers,
X. Bekaert, V. Bouchard, N. Boulanger, S. Cnockaert,
S. de Buyl, S. Detournay, S. Nevens, A. Wijns.
ii

Local Organizing Committee

X. Bekaert, V. Bouchard,
N. Boulanger, S. Cnockaert,
S. de Buyl, S. Detournay,
S. Nevens, A. Wijns
 iii

List of Participants

Joke Adam (KUL)

Glenn Barnich (ULB)
Xavier Bekaert (IHES)
Nazim Bouatta (ULB)
Vincent Bouchard (Oxford)
Nicolas Boulanger (UMH)
Fabien Buisseret (UMH)
Sandrine Cnockaert (ULB)
Laurent Claessens (UCL)
Geoffrey Compère (ULB)
Sophie de Buyl (ULB)
Stphane Detournay (UMH)
Francis Dolan (Cambridge)
Johanna Erdmenger (Humboldt U)
Jarah Evslin (ULB)
Wen Jiang (Oxford)
Serge Leclercq (UMH)
Vincent Mathieu (UMH)
Frank Meyer (Max- Planck Institute for Physics)
Stijn Nevens(VUB)
Jeong-Hyuck Park (IHES)
Jan Rosseel (KUL)
Alexander Wijns (VUB)
Robert Wimmer (Hannover)
Martin Wolf (Hannover)
Fonger Ypma (Oxford)
Contents

Field theory from a bundle point of view

L. Claessens 1
Fiber bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1 Differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Example : Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3 Vector bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4 Principal bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5 Associated bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 Connections on fiber bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7 Frame bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8 Product of principal bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Building gauge invariant theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
9 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
10 Gauge transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
11 A few physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
12 Inclusion of the Lorentz group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
13 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Complex geometry, Calabi-Yau manifolds and toric geometry

V. Bouchard 47
1 Complex geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2 Kähler geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3 Calabi-Yau geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4 Toric geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

Aspects of Two-Dimensional Conformal Field Theory

S. Detournay 93
1 Conformal transformations in two dimensions . . . . . . . . . . . . . . . . . . . . 94
2 Conformal Ward identities in two dimensions . . . . . . . . . . . . . . . . . . . . 99
3 Operator product expansion and conformal transformation of fields . . . . . . . . 107
4 Operator formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5 The free boson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
A Another derivation of the conformal Ward identities . . . . . . . . . . . . . . . . 134
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

iv
CONTENTS v

Semi-Simple Lie Algebras and Representations

(Parts I and II)
N. Boulanger, S. de Buyl 139
1 The structure of Simple Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . 140
2 Highest Weight Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
3 Applications to Hadrons : A1 and A2 . . . . . . . . . . . . . . . . . . . . . . . . . 162
A Notation and Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Semi-simple Lie algebras and representations (Part III)

F. Dolan 182
1 Verma Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
2 The Weyl group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
3 Formal characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4 The Racah-Speiser Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

Lecture Note on Clifford Algebra

J.-H. Park 197
1 Preliminary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2 Gamma Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
3 Spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
4 Majorana Representation and SO(8) . . . . . . . . . . . . . . . . . . . . . . . . . 206
5 Superalgebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
A Proof of the Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

Universal Enveloping Algebras and some applications in physics

X. Bekaert 214
1 Associative versus Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
2 Universal enveloping algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3 Casimir operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
4 Symmetries of the S-matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
A Proof of Berezin’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
B Proof of BCH’s product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

β function and asymptotic freedom in QCD

G. Barnich 235
1 Effective action at one loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
2 Asymptotic behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
3 Quantum gauge fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

Twistor Geometry and Gauge Theory

M. Wolf 246
1 Lecture 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
2 Lecture 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
3 Lecture 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
A Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
vi CONTENTS

B Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

Moyal’s Star Product

S. Cnockaert 269
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
2 Ordering and notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
3 Standard Moyal Star Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
4 Moyal star product as a matrix-like multiplication . . . . . . . . . . . . . . . . . 273
5 Integral representation of the Moyal star product . . . . . . . . . . . . . . . . . . 274
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

Noncommutative Spaces and Gravity

F. Meyer 276
1 Noncommutative Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
2 Symmetries on Deformed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
3 Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
4 Deformed Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
5 Noncommutative Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
6 Star Products and Expanded Einstein-Hilbert Action . . . . . . . . . . . . . . . . 286
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
 Field theory from a bundle
point of view

Laurent Claessens

Université Libre de Bruxelles

E-mail: [email protected]

Based on the lectures presented by L. Claessens at the First Modave Summer School in
Mathematical Physics held at Modave, Belgium on June 19-25, 2005
2 Claessens

Warning
These notes are intended to fit what I said during the Modave Summer School 2005; I just added
some physical features. A much more extended version (containing more proofs, topological
discussions and physical relevance) of this text and the LATEX source files of these notes will be
deposed on http://student.ulb.ac.be/˜lclaesse.
A few after the Modave Summer school, I remarked a sign incoherence between the definition
of the connection 1-form and and of a fundamental vector field. I think it is corrected, but the
readrer should be careful in the computations. If anybody has some advice which can improve
the quality of the text, he is most welcome at [email protected] !

Introduction
There are two reasons to study the fiber bundle formalism in quantum field theory. The first
is this formalism itself : it gives powerfull tools and geometrical interpretations to field theory.
Electromagnetic and gravitational fields are know to be “connections”. Bundle formalism allow
us to undersatd how these two are related to the same idea : to make a global symmetry a local
one. From a geometrical point of view, the way to go from the galillean equation of motion for
2 α
a free particle ddλx2 = 0 to the famous geodesic equation

d2 xα
+ Γα β σ
βσ ẋ ẋ = 0
dλ2
is exactly the Yang-Mills’s trick1 !
The second reason is that the fiber bundle formalism gives an elegant description to the
famous “law of transformations” of quantum fields. If we say that a Weyl spinor is just a map
ψ : R4 → C2 where R4 is the physical space on which the Lorentz transformations acts, there
is a problem : the law of transformation of a map under a change of basis Λ is mathematics
for children : the result is ψ ′ (x) = ψ(Λ−1 x). In particular, there are no reasons to mix the
components. So the space C ∞ (R4 , C2 ) in which ψ lives canot be seen as the space of maps from
physical space to C2 . This space is just a mathematical trick on which we build representations
of the Lorentz group. In particular, we cannot see this R4 as any physical space.
We will see that ψ is not a function from R4 to C2 ; the building of associated vector bundle
will give (I hope) a coherent picture of the quantum fields, their law of transformations and the
physical space R4 .
These notes are destinated to physicists who have yet a good knowledge of field theory, and
Yang-Mills features. The electromagnetic U (1)-gauge is taken as example.

1 with a non-compact group and other more complicated features.

Fiber bundle

1 Differentiable manifolds
1.1 Definition and examples
A systematic exposition of manifolds and such can be found in [5].
A n-dimensional differentiable manifold is a set M and a system of charts {(Uα , ϕα )}α∈I
where each set Uα is open in Rn and the maps ϕα : Uα → M are injective and satisfy the three
following conditions :

• every x ∈ M is contained in at least one set ϕα (Uα ),

• for any two charts ϕα : Uα → M and ϕβ : Uβ → M , the set

ϕ−1
α (ϕα (Uα ) ∩ ϕβ (Uβ ))

is an open subset of Uα ,
• the map
(ϕ−1 −1
β ◦ ϕα ) : ϕα (ϕα (Uα ) ∩ ϕβ (Uβ )) → Uβ

is differentiable2 as map from Rn to Rn .

Each time we say “manifold”, we mean “differentiable manifold”. We will only consider
manifolds with Hausdorff topology (see later for the definition of a topology on a manifold).
Any open set of Rn is a differentiable manifold if we choose the identity map as chart system.
Most of surfaces z = f (x, y) in R3 are manifolds, depending on certain regularity conditions
on f .
If M1 and M2 are two differentiable manifolds, a map f : M1 → M2 is differentiable if
f is continuous and for each two coordinate sytstems ϕ1 : U1 → M1 and ϕ2 : U2 → M2 , the
map ϕ−12 ◦ f ◦ ϕ1 is differentiable on its domain. One can show that if f : M1 → M2 and
g : M2 → M3 are differentiables, then g ◦ f : M1 → M3 is differentiable.

Example : the sphere The sphere S n is the set

S n = {(x1 , . . . , xn+1 ) ∈ Rn+1 | kxk = 1}

for which we consider the following open set in Rn :

U = {(u1 , . . . , un ) ∈ Rn | kuk < 1}

2 In the sequels, by “differentiable” we always mean smooth. If this map is differentiable, C k , analytic,. . . then

the manifold is said to be differentiable, C k , analytic,. . .

3
4 Claessens

and the charts ϕi : U → S, and ϕ̃i : U → S

p
ϕi (u1 , . . . , un ) = (u1 , . . . , ui−1 , 1 − kuk2 , ui , . . . , un ) (1.1a)
p
ϕ̃i (u1 , . . . , un ) = (u1 , . . . , ui−1 , − 1 − kuk2 , ui , . . . , un ). (1.1b)
These map are clearly injective. To see that ϕ(U) ∪ ϕ̃(U) = S, consider (x1 , . . . , xn+1 ) ∈ S.
Then at least one of the xi is non zero. Let us suppose x1 6= 0, thus x21 = 1 − (x22 + . . . + x2n+1 )
and p
x1 = ± 1 − (. . .). (1.2)
If we put ui = xi+1 , we have x = ϕ(u) or x = ϕ̃(u) following the sign in relation (1.2). The fact
that ϕ−1 ◦ ϕ̃ and ϕ̃−1 ◦ ϕ are differentiable is a “first year in analysis exercice”.

Example : projective space On Rn+1 \ {o}, we consider the equivalence relation v ∼ λw

for all non zero λ ∈ R, and we put


RP n = Rn+1 \ {o} / ∼ .
This is the set of all the one dimensional subsapces of Rn+1 . This is the real projective space
of dimension n. We set U = Rn and
ϕi (u1 , . . . , un ) = Span{(u1 , . . . , ui−1 , 1, ui , . . . , un )}.
One can see that this gives a manifold structure to RP n . Moreover, the map
A : S n → RP n
(1.3)
v → Span v
is differentiable.

1.2 Topology on manifold and submanifold

A subset V ⊂ M is open if for every chart ϕ : U → M , the set ϕ−1 (V ∩ ϕ(U)) is open in U.
Theorem 1.1. This definition gives a topology on M which has the following properties :
(i) the charts maps are continuous,
(ii) the sets ϕα (Uα ) are open.
Proof. First we prove that the open system defines a topology. For this, remark that ϕ−1α is
injective (if not, there should be some multivalued points). Then ϕ−1 (A∩B) = ϕ−1 (A)∩ϕ−1 (B).
If V1 and V2 are open in M , then
ϕ−1 (V1 ∩ V2 ∩ ϕ(U)) = ϕ−1 (V1 ∩ ϕ(U)) ∩ ϕ−1 (V2 ∩ ϕ(U))
which is open in Rn . The same property works for the unions.
Now we turn our attention to the continuity of ϕ : U → M ; for an open set V in M , we have
to show that ϕ−1 (V ) is open in U ⊂ Rn . But the definition of the topology on M , is precisely
the fact that ϕ−1 (V ∩ ϕ(U)) is open.
If M is a differentiable manifold and N , a subset of M , we say that N is a submanifold of
dimention k if ∀ p ∈ N , there exists a chart ϕ : U → M around p such that
ϕ−1 (ϕ(U) ∩ N ) = Rk ∩ U := {(x1 , . . . , xk , 0 . . . , 0) ∈ U}.
In this case, N is itself a manifold of dimention k for which one can choose the ϕ of the
definition as charts.
Field theory from a bundle point of view 5

1.3 Tangent vector

A tangent vector to the manifold M is a map X : C ∞ (M ) → R which can be written under
the form
dh i
Xf = (f ◦ X)′ (0) = f (X(t)) (1.4)
dt t=0
for a certain path X : R → M . Notice the abuse of notation between the tangent vector and the
path which defines it.
A more formal way to define a tangent vector is to say that it is an equivalent class of path
in the sense that two path are equivalent if and only if they induced maps by (1.4) are equals.
Using the chain rule d(g ◦ f )(a) = dg(f (a)) ◦ df (a) for the differentiation in Rn , one sees that
this equivalence notion doesn’t depend on the choice of ϕ. In other words, if ϕ and ϕ̃ are two
charts for a neighborhood of x, then (ϕ−1 ◦ γ)′ (0) = (ϕ−1 ◦ σ)′ (0) if and only if (ϕ̃−1 ◦ γ)′ (0) =
(ϕ̃−1 ◦ σ)′ (0). The space of all tangent vectors at x is denoted by Tx M . There exists a bijection
[γ] ↔ (ϕ−1 ◦ γ)′ (0) between Tx M and Rn , so Tx M is endowed with a vector space structure.
If (U, ϕ) is a chart around X(0), we can reexpress Xf with only well know objects by defining
the function f˜ = f ◦ ϕ and X̃ = ϕ−1 ◦ X
 
dh ˜ i ∂ f˜  dX̃ α 
Xf = (f ◦ X̃)(t) =   .
dt t=0 ∂xα  dt 
x=X̃(0) t=0

In this sense, we write

dX̃ α ∂
X= (1.5)
dt ∂xα
and we say that {∂1 , . . . , ∂n } is a basis of Tx M . As far as notations are concerned, from now
a tangent vector is written as X = X α ∂α where X α is related to the path X : R → M by
X α = dX̃ α /dt. We will no more mention the chart ϕ and write
dh i
Xf = f (X(t)) .
dt t=0

Corectness of this short notation is because the equivalence relation is independent of the choice
of chart. When we speak about a tangent vector to a given path X(t) without specification, we
think about X ′ (0).
All this construction gives back the notion of tangent vector when M ⊂ Rm . In order to see
it, think to a surface in R3 . A tangent vector is precisely given by a derivative of a path : if
c : R → Rn is a path in the surface, a tangent vector to this curve is given by
c(t0 ) − c(t0 + t)
lim
t→0 t
which is a well know limit of a difference in Rn .

1.4 Differential of a map

Let f : M1 → M2 be a differentiable map, x ∈ M1 and X ∈ Tx M1 , i.e. X : R → M1 with
X(0) = x and X ′ (0) = X. We can consider the path Y = f ◦ X in M2 . The tangent vector to
this path is written dfx X.
Proposition 1.2. If f : M1 → M2 is a differentiable map between two differentiable manifolds,
the map
dfx : Tx M1 → Tf (x) M2
(1.6)
X ′ (0) → (f ◦ X)′ (0)
6 Claessens

is linear.
Let x : U → M and y : V → M be two charts systems around p ∈ M . Consider the path
c(t) = x(0, . . . , t, . . . 0) where the t is at the position k. Then, with respect to these coordinates,
dh i ∂f dci ∂f
c′ (0)f = f (c(t)) = i
= ,
dt t=0 ∂x dt ∂xk
so c′ (0) = ∂/∂xk . Here, implicitly, we wrote ci = (xi )−1 ◦ c where (xi )−1 is the ith component
of x−1 seen as element of Rn . We can make the same computation with the system y. With
these abuse of notation,
∂ X ∂y j ∂
= (1.7)
∂xi j
∂xi ∂y j

as it can be seen by applying it on any function f : M → R. More precisely if x : U → M and

y : U → M are two charts (let U be the intersection of the domains of x and y), let f : M → R
and f = f ◦ x, f˜ = f ◦ y. The action of the vector ∂xi of the function f is given by

∂f
∂xi f =
∂xi
where the right hand side is a real number that can be computed with usual analysis on Rn .
This real defines the left hand side. Now, f = f˜ ◦ y −1 ◦ x, so that

∂f ∂(f˜ ◦ y −1 ◦ x) ∂ f˜ ∂y j
i
= i
=
∂x ∂x ∂y j ∂xi
∂f˜ ∂y j
where ∂y j is precisely what we write now by ∂y j f and ∂xi must be understood as the derivative

with respect to xi of the function (y −1 ◦ x) : Rn → Rn .

Let f : M → N and g : N → R; the definitions give
dh i
(dfx X)g = (g ◦ f )(X(t))
dt t=0
∂g ∂f i dX α
= .
∂y i ∂xα dt
∂f i dX α
This shows that ∂xα dt is (dfx X)i . But dX α /dt is what we should call X α in the decompositon
∂f i
X = X α ∂α then the matrix of df is given by ∂xα . So we find back the old notion of differential.
Remark 1.3. If X ∈ Tx M and f is a vector valued function on M , then one can define Xf by
exactly the same expression. In this case,
dh i
df X = f (v(t)) = Xf.
dt t=0

A map f : M1 → M2 is an immersion at p ∈ M1 if dfp : Tp M1 → Tf (p) M2 is injective. It

is a submertion if dfp is surjective.

1.5 Tangent bundle

If M is a n dimensional manifold, as set the tangent bundle is
[
TM = Tx M.
x∈M
Field theory from a bundle point of view 7

Theorem 1.4. The tangent bundle admits a 2n dimentional manifold structure for which the
projection
π : TM → M
(1.8)
Tp M → p
is a submertion.

The structure is easy to guess. If ϕα : Uα → M is a coordinate system on M (with Uα ⊂ Rn ),

we define ψα : Uα × Rn → T M by
X 
∂ 
ψ(x1 , . . . xn , a1 , . . . an ) = ai .
| {z } | {z }
i
∂xi ϕ(x1 ,...,xn )
∈Uα ∈Rn

The map ψβ−1 ◦ ψβ is differentiable because

X 
∂yj 
(ψβ−1 ◦ ψβ )(x, a) = (y(x), ai )
i
∂xi y(x)

which is a composition of differentiable maps.

The set T M endowed with this structure is called the tangent bundle.

Commutator of vector fields

If X, Y ∈ X(M ), one can define the commutator [X, Y ] in the following way. First remark
that, if f : M → R, the object X(f ) is also a function from M to R by X(f )(x) = Xx (f ), so
we can apply Y on X(f ). The definition of [X, Y ]x is

[X, Y ]x f = Xx (Y f ) − Yx (Xf ). (1.9)

If X = X i ∂i and Y = Y j ∂j , then

XY (f ) = X i ∂i (Y j ∂j f )
= X i ∂i Y j ∂j f + X i Y j ∂ij
2
f.
2 2
From symmetry ∂ij f = ∂ji f , the difference XY f − Y Xf is only X i ∂i Y j − Y i ∂i X j , so that

[X, Y ]i = XY i − Y X i (1.10)

where X i and Y i are seen as functions from M to R.

1.6 Cotangent bundle

The cotangent space Tp∗ M of M at p is the dual space of Tp M , i.e. the vector space of all the
(real valued) linear3 1-forms on Tp M . In the coordinate system x : U → M , we naturally use,
on Tp∗ M , the dual basis of the basis {∂/∂xi , . . . ∂/∂xi } of Tp M . This dual basis is denoted by
{dx1 , . . . , dxn }, the definition being as usual :

dxi (∂ j ) = δij . (1.11)

3 When we say a form, we will always mean a linear form.
8 Claessens

The notation comes from the fact that equation (1.11) describes the action of the differential of
the projection xi : U → R on the vector ∂ j .
If (Uα , ϕα ) is a chart of M , then the maps

φα : Uα × Rn → T ∗ M
(1.12)
(x, a) → ai dxi |x

give to T ∗ M a 2n dimensional manifold structure such that the canonical projection π : T ∗ M →

M is an immersion.

Pull-back and push-forward

Let ϕ : M → N be a smooth map, α a k-form on N , and Y a vector field on N . Consider the
map dϕ : Tx M → Tϕ(x) M . The aim is to extend it to a map from the tensor algebra of Tx M
to the one of Tϕ(x) M . See [6] for precise definition of the tensor algebra.
The pull-back of ϕ on a k-form α is the map

ϕ∗ : Ωk (N ) → Ωk (M )

defined by
(ϕ∗ α)m (v1 , . . . , vk ) = αϕ(m) (dϕm v1 , . . . , dϕm vk ) (1.13)
for all m ∈ M and vi ∈ X(M ).
Note the particular case k = 0. In this case, we take –instead of α– a function f : N → R
and the definition (1.13) gives ϕ∗ f : M → R by

ϕ∗ f = f ◦ ϕ.

The push-forward of ϕ on a k-form is the map

ϕ∗ : Ωk (M ) → Ωk (N )

defined by ϕ∗ = (ϕ−1 )∗ . For v ∈ Tn N , we explicitly have :

(ϕ∗ α)n (v) = αϕ−1 (n) (dϕ−1

n v).

Let now ϕ : M → N be a diffeomorphism. The pull-back of ϕ on a vector field is the map

ϕ∗ : X(N ) → X(M )

defined by
(ϕ∗ Y )(m) = [(dϕ−1 )m ◦ Y ◦ ϕ](m),
or
(ϕ∗ Y )ϕ−1 (n) = (dϕ−1 )n Yn ,
for all n ∈ N and m ∈ M . Notice that

(dϕ−1 )n : Tn N → Tϕ−1 (n) M,

and that ϕ−1 (n) is well defined because ϕ is an homeomorphism.

The push-forward is, as before, defined by ϕ∗ = (ϕ−1 )∗ . In order to show how to maniulate
these notations, let us prove the following equation :

f∗ξ = (df )ξ .
Field theory from a bundle point of view 9

For ϕ : M → N and Y in X(N ), we just defined ϕ∗ : X(N ) → X(M ), by

(ϕ∗ Y )ϕ−1 (n) = (dϕ−1 )n Yn . (1.14)
Take f : M → N ; we want to compute f∗ = (f −1 )∗ with (f −1 )∗ : X(M ) → X(N ). Replacing
the “−1 ” on the right places, the definition (1.14) gives us :
h i
(f −1 )∗ X = (df )m Xm ,
f (m)

if X ∈ X(M ), and m ∈ M .
We can rewrite it without any indices : the coherence of the spaces automatically impose the
indices : (f −1 )∗ X = (df )X. It can also be rewritten as (f −1 )∗ = df , and thus f∗ = df . From
there to f∗ξ = (df )ξ , it is straightforward.

Hodge operator
Let us take a manifold M indowed with a metric g. We can define a map r : Tx∗ M → Tx M by,
for α ∈ Tx∗ M ,
hr(α), vi = α(v).
for all v ∈ Tx M , where h·, ·i stands for the product given by the metric g. If we have α, β ∈ Tx∗ M ,
we can define
hα, βi = hr(α), r(β)i.
With this, we can define an inner product on Λp (Tx∗ M ) :
hα1 ∧ . . . αp , β1 ∧ . . . βp i = dethαi , βj i.
ij

The Hodge operator is ⋆ : Λ p

(Tx∗ M ) →Λ n−p
(Tx∗ M ) such that for any φ ∈ Λp (Tx∗ M ),
φ ∧ (⋆ψ) = hφ, ψiΩ (1.15)
p
where Ω = | det(g)|dx1 ∧ . . . ∧ dxn .

2 Example : Lie groups

A Lie group is a manifold G endowed with a group structure such that the inversion map
i : G → G, i(x) = x−1 and the multiplication m : G × G → G, m(x, y) = xy are differentiable.
The Lie algebra of the Lie group G is the tangent space of G at the identity : G = Te G. Let
us present two classical Lie group stuff in our new differential geometry formalism.

2.1 The Lie algebra of SU (2)

Let consider G = SU (2); the elements are complexes 2 × 2 matrices U such that U U † = 1 and
det U = 1. An element of the Lie algebra is given by a path u : R → G in the group with
u(0) = 1. Since for all t, u(t)u(t)† = 1,
dh i
0= u(t)u(t)†
dt t=0
dh †
i dh i
(2.1)
= u(0) u(t) + u(t) u(0)†
dt t=0 dt t=0
= [dt u(t)]† + [dt u(t)].
So a general element of the Lie algebra su(2) is an antihermician matrix. The same trick gives
the condition of nul trace.
10 Claessens

2.2 What is g −1 dg ?
The expression g −1 dg is often written in the physical litterature. In our framework, the way
to gives a sense to this expression is to consider it pointwise acting on a tangent vector. More
precisely, the scheme is the data of a manifold M , a Lie group G and a map g : M → G.
Pointwise, we have to apply g(x)−1 dgx to a tangent vector v ∈ Tx M .
Note that dgx : Tx M → Tg(0) G 6= Te G, so dgx ∈ / G. But the product g(x)−1 dgx v is defined
by
dh i
g(x)−1 dgx v = g(x)−1 g(v(t)) ∈ G.
dt t=0

3 Vector bundle
Let M be a smooth manifold. A V -vector bundle of rank r on M is a smooth manifold F and
a smooth projection p : F → M such that

• for any x ∈ M , the fiber Fx := p−1 (x) is a vector space of dimention r on the same field
that V (let’s say K = R or C).

• for any x ∈ M , there exists an open neighborhood U of x and a “chart diffeomorphism”

φ : p−1 (U) → U × V such that for any l ∈ p−1 (y),

– φ(l) = (y, φy (l))

– φy : Ey → V is a vector space isomorphism.

The pair (U, φ) is a local trivialisation; M is the base space; F , the total space, p the projection
and r, the rank of the bundle. The denominations of total and base spaces will also be used in
the same way for principal bundles.
We will sometimes use charts diffeomorphism φ : U × V → p−1 (U) instead of φ : p−1 (U) →
U × V . Since they are diffeomorphism, this difference don’t affects anything.

3.1 Transition functions

The trivialisations will be denoted by greek indices : Uα , φα ,. . . The symbol Uαβ naturally
denotes Uα ∩ Uβ . If we consider two local trivialisations (Uα , φα ) and (Uβ , φβ ), we have to look
at φα ◦φ−1
β : Uαβ ×Kr → Uαβ ×Kr . We define the transition functions gαβ : Uαβ → GL(r, K)
by
φα ◦ φ−1
β (x, v) = (x, gαβ (x)v). (3.1)
These functions take their values in GL(r, K) because φy : Ey → V is a vector space isomor-
phism. Since (φα ◦ φβ )−1 = φβ ◦ φ−1
α , it is clear that gαβ (x) = gαβ (x)
−1
.
−1
If x ∈ Uαβγ = Uα ∩ Uβ ∩ Uγ , we have φα ◦ φγ (x, v) = (x, gαγ (x)v), but also φα ◦ φ−1
γ =
φα ◦ φ−1
β φ β ◦ φ −1
γ , then

(x, gαγ (x)v) = (φα ◦ φ−1

β )(x, gβγ (x)v)
(3.2)
= (x, gαβ (x)βγ (x)v).

Thus gαγ (x) = gαβ (x)gβγ (x). So, as linear maps, we have

gαβ ◦ gαγ ◦ gγα = 1. (3.3)

Field theory from a bundle point of view 11

3.2 Equivalence of vector bundle

p p′
Let E −→ M and F −→ M be two vector bundles on M . They are equivalent if there exists
a smooth diffeomorphism f : E → F such that

• p′ ◦ f = p,

• f |Ex : Ex → Fx is a vector space isomorphism.

Let E and F be two equivalent vector bundles, {Uα | α ∈ I}, an open covering which trivialize
E and F in the same time and φE F
α , φα the corresponding trivializations. A map f : E → F
reads “in the trivialisation” as φα ◦ f |p−1 (µα ) ◦ φE −1
F r r
α : Uα × K → Uα × K and defines a map
λα : Uα → GL(r, K) by
E −1
(φF
α ◦ f |p−1 (µα ) ◦ φ α )(x, v) = (x, λα (x)v). (3.4)

If we denote by g E the transition functions for E (and g F for F ),

F −1 E −1 E −1 −1 −1
φF
α ◦ φβ = (φF
α ◦ f ◦ φα ) ◦ (φE
α ◦ φβ ) ◦ (φE
β ◦f ◦ φE
β ),

so that
F
gαβ E
(x) = λα (x)gαβ (x)λ(x)−1 . (3.5)

Proposition 3.1. A vector bundle over Rn is trivial.

Proof. Let p : F → M be a vector bundle on M = Rn . Let {Uα } be an open covering of Rn as

in the definition. For this, we consider a partition of unity related to the covering Uα : a set of
functions fα : M → R such that

• fα > 0,

• ∀x ∈ M , one can find a neighborhood of x in which only a finite number of fα is non zero,
P
• ∀x ∈ M , α fα (x) = 1.

• fα = 0 outside of Uα .

With this we define the map f : F → Rn × V by

X
f (l) = (x, fα (x)φαx (l)).
α

3.3 Sections of vector bundle

A section of the vector bundle p : E → M is a smooth map s : M → E such that p ◦ s = id |M .
The set of all the sections is denoted by Γ∞ (M, E) or simply Γ(E).
If (Uα , φα ) is a local trivialization, one can describe the section s by a function sα : Uα → V
defined by φα (s(x)) = (x, sα (x)), or equivalently by

s(x) = φ−1
α (x, sα (x)).
12 Claessens

As usual when we define such a local quantity, we have to ask oursel how are related sα and sβ
on Uα ∩ Uβ . The best is sα = sβ , but most of the time it is not. Here, we compute
φβ ◦ φ−1
α ◦ φα (s(s)) = (x, gαβ (x)sα (x)),

which is obviously also equal to (x, sβ (x)). Then

sβ (x) = gαβ (x)sα (x) (3.6)
(without summation).

3.4 Connection on vector bundle

A connection on the vector bundle p : E → M is a bilinear map

∇ : X(M ) × Γ(M, E) → Γ(M, E),

(3.7)
(X, s) → ∇X s
such that
• ∇f X s = f ∇X s,
• ∇X (f s) = (X · f )s + f ∇X s
for all X ∈ X(M ), f ∈ C ∞ (M ) and s ∈ Γ(M, E). The operation ∇ is ofen called a covariant
derivative.
An easy example is given on the trivial bundle E = pr1 : M × C → M . For this bundle,
Γ(M, E) = C ∞ (M, C) and the common derivation is a covariant derivation : ∇X s = (ds)X.

Local description
Proposition 3.2. The value of (∇X s)(x) depends ony on Xx and s on a neighborhood of x ∈ M .
Proof. Let X, Y ∈ X(M ) such that Yz = f (z)Xz with f (x) = 1 and f (z) 6= 1 everywhere else.
Then
(∇Y s)(x) − (∇X s)(x) = (f (x) − 1)(∇X s)(x) = 0.
Since it is true for any function, the linearity makes that it cannot depend on Xz with z 6= x.
If we consider now two sections s and s′ which are equals on a neighborhood of x, we can write
s′ = f s for a certain function f which is 1 on the neighborhood. Then
(∇X s′ )(x) − (∇X s)(x) = (f (x) − 1)(∇X s)(x) + (Xf )s(x)
which zero because on a neighborhood of x, f is the constant 1.
This proposition shows that it makes sense to consider only local descritions of connections.
Let {e1 , . . . , er } be a basis of V and consider the local sections S αi : Uα → E,
S αi (x) = φ−1
α (x, ei ).

A local section sα : Uα → V can be decomposed as sα (x) = siα (x)ei with respect to this basis
(up to an isomorphism bewteen the differents V at each point). Then on Uα ,
siα S αi (x) = siα (x)φ−1
α (x, ei )
= φ−1 i
α (x, sα ei )
(3.8)
= φ−1
α (x, sα (x))
= s(x).
Field theory from a bundle point of view 13

The first line is the definition of the product R × F → F .

So any s ∈ Γ(E) can be (localy !) written under the form4 s = siα S αi ; in particular ∇X (S αi )
can. We define the coefficients θ by

∇X (S αi ) = (θα )ji (X)S αj . (3.9)

where, for each i and j, (θα )ji is a 1-form on Uα . We can consider θα as a matrix-valued 1-form
on Uα .
Proposition 3.3. The formula

(∇X s)α = Xsα + θα (X)sα (3.10)

gives a local description of the connection.

Proof. For any s ∈ Γ(E), we have
X

∇X s = ∇X sjα S αj
j
X 
= (Xsjα )S αj + sjα ∇X S αj (3.11)
j
Xh i
= (Xsiα ) + sjα (θα )ij (X) S αi .
i

Connection and transition functions

A connection determines some local matrix-valued 1-forms θα on the trivialization Uα . Two
natural questions raise. The first is the converse : does a matrix-valued 1-form defines a con-
nection ? The second is to know what is θα in funcion of θβ on Uα ∩ Uβ ? The answer to the
latter is given by the following proposition :
Proposition 3.4. The 1-form θα relative to the trivialization (Uα , φα ) is related to the 1-form
θβ relative to the trivialization (Uβ , φβ ) by
−1 −1
θβ = gαβ dgαβ + gαβ θα gαβ . (3.12)

This equation looks like something you know ? If you think to equation (9.3) or (10.3) or
any physical equation of gauge transformation for the bosons, then you are almost right.
Proof. We can use equation (3.6) pointwise on (∇X s)α :

(∇X s)α = gαβ (∇X s)β

= gαβ Xsβ + θβ (X)sβ (3.13)


= gαβ X(gαβ sα ) + θβ (X)gαβ sα .

We have to compare it with equation (3.10). Note that gαβ and θα (X) are matrices, then one
cannot do
gαβ θβ (X)gαβ = gαβ gαβ θβ (X) = θβ (X)
4 be careful on the fact that the “coefficient” si depends on x : the right way to express this equation is
α
s(x) = siα (x)S αi (x).
14 Claessens

by using gαβ gαβ = 1. By taking explicitly the indices into account, one must write
(gαβ )ij θβ (X)jk (gαβ )kl . Applying Leibnitz formula (X(f g) = f (Xg) + (Xf )g), and making the
simplification gαβ gαβ = 1 in the first term, we find
−1
θα (X)sα = gαβ (Xgαβ )sα + gαβ θβ (X)gαβ sα .

The claim follows from the fact that Xgαβ = dgαβ (X).
The inverse is given in the
Proposition 3.5. If we choose a family of GL(V )-valued 1-forms θα on Uα satisfying (3.12),then
the formula
(∇X s)α = Xsα + θα (X)sα
defines a connection on E.

3.5 Curvature
For X, Y ∈ X(M ), we consider the map

R(X, Y ) : Γ∞ (M, E) → Γ∞ (M, E)

(3.14)
s → ∇X ∇Y s − ∇Y ∇X s − ∇[X,Y ] s.

It satisfies R(f X, Y )s = f R(X, Y )s = R(X, Y )f s for every f ∈ C ∞ (M ). The map R : Tx M ×

Tx M → End Γ∞ (M, E) is the curvature of ∇.

Local description
In a trivialisation (Uα , φα ), we have (∇X s)α = Xsα +θα (X)sα . In the expression of (R(X, Y )s)α ,
the terms coming from the Xsα part of nabla make

XY sα − Y Xsα − [X, Y ]sα = 0.

The other terms are no more than matricial product, then the formula

(R(X, Y )s)α = Ωα (X, Y )sα (3.15)

well defines a 2-form Ωα with values in GL(r, K). We can find an expression for Ω in terms of θ
:
Ωα (X, Y ) = Xθα (Y ) − Y θα (X) − θα ([X, Y ]) + θα (X)θα (X) − θα (Y )θα (Y );
it is written as
1
Ωα = dθα + θα ∧ θα = dθα + [θα , θα ] (3.16)
2
which is a notational shortcut for

Ωα (X, Y ) = dθα (X, Y ) + [θα (X), θα (Y )].

These equations are called structure equations. Pointwise, the second term is a matrix com-
mutator; be careful on the fact that, when we wil speak about principal bundle, the forms θ’s
will take their values in a Lie algebra. On Uα ∩ Uβ , we have
−1
Ωβ (X, Y ) = gαβ Ωα (X, Y )gαβ .

The curavature and the connection fulfill the Bianchi identities :

Field theory from a bundle point of view 15

Lemma 3.6.
dΩα + [θα ∧ Ωα ] = 0.

Proof. For each matricial entry, θα is a 1-form on Uα , then θα (X) is a function which to x ∈ M
assign θα (x)(Xx ) ∈ R. So we can apply d and Leibnitz on the product θα (X)θα (Y ).


d θα (X)θα (Y ) = θα (X)dθα (Y ) + dθα (X)θα (Y ).

Differentiating equation (3.16), dΩα = dθα ∧ θα − θα ∧ dθα .

4 Principal bundle
Let M be a manifold and G, a Lie group whose unit is denoted by e. A G-principal bundle
on M is a smooth manifold P , a smooth map π : P → M and a right action of G on P denoted
by ξ · g with g ∈ G and ξ ∈ P such that

• π(ξ · g) = π(ξ),

• ∀ξ ∈ π −1 (x), π −1 (x) = {ξ · g | g ∈ G} ≃ G,

• for any x ∈ M , there exists a neighborhood Uα of x in M , a diffeomorphism φα :

π −1 (Uα ) → Uα × G and a diffeomorphism φαx : P → G such that

– φα (ξ) = (x, φαx (ξ)),

– φαx (ξ · g) = φαx (ξ) · g.

The group G is often called the structure group. We suppose that the action is effective. We
will sometimes use the notation P (G, M ) to precise that P is a principal bundle over M with
structure group G.
The whole construction is given in figure 1.1. All is not yet defined, but in the following, the
notations will follow this scheme.

φP
α

P Uα × G

E = P ×ρ V
π −1 |Uα π p
Uα × V
φE
α
Uα
M

Figure 1.1: Some bundles

16 Claessens

Lemma 4.1. The map φ−1

α fulfills

φα−1 (x, h) · g = φ−1

α (x, hg).

Proof. From the definition of a principal bundle, any ξ ∈ P can be written under the form
ξ = φ−1
α (x, φαx (ξ)) with φx satisfying φx (ξ · h) = φx (ξ)h for a certain function φx : P → G. We
consider in particular ξ = φ−1
α (x, h)·g. Then ξ·g
−1
= φ−1α (x, h). But ξ·g
−1
= φ−1
α (x, φαx (ξ)g
−1
),
−1
then h = φαx (ξ)g and φαx (ξ) = hg. So we have
ξ = φ−1 −1 −1
α (x, h) · g = φα (x, φαx (ξ)) = φα (x, hg).

Let
R = {(x, y) ∈ P × P | x = y · g for a certain g ∈ G}.
Proposition 4.2. The function u : R → G defined by the condition
p · u(p, q) = q.
is differentiable.

4.1 Transition functions

Let (Uα , φα ) be a local trivialization of P . This induces transition functions gαβ : Uα ∩ Uβ → G
defined by
φα ◦ φ−1
β : Uα ∩ Uβ × G → Uα ∩ Uβ × G
(4.1)
(x, a) → (x, gαβ (x)a).
Clearly, gαα = e and gαβ gαβ = e on Uα ∩ Uβ . Then the triviality
φα ◦ φ−1 −1 −1
β ◦ φβ ◦ φγ ◦ φγ ◦ φα = id

implies the compatibility conditions

gαβ gβγ gγα = e (4.2)
on Uα ∩ Uβ ∩ Uγ .
There is an inverse construction. Let {Uα | α ∈ I} be an open covering of M and gαβ :
Uα ∩ Uβ → G a family of functions such that gαα = e, gαβ gαβ = e on Uα ∩ Uβ and gαβ gβγ gγα = e
on Uα ∩ Uβ ∩ Uγ . Then the following construction gives a G-principal bundle whose transition
functions are the gαβ ’s.
F
• P̃ = α∈I Uα × G (disjoint union),
• if (x, a) ∈ Uα × G and (y, b) ∈ Uβ × G, then (x, a) ∼ (y, b) if and only if x = y and
b = gαβ (x)a,
• π : P̃ → M is defined by π[(x, a)] = x where [(x, a)] is the class of (x, a) for ∼,
• the action is defined by [(x, a)] · g = [(x, ag)].
Theorem 4.3. Let G be a Lie group and M , a differentiable manifold, {Uα }α∈I be an open
covering of M and some functions ϕαβ : Uα ∩ Uβ → G such that ϕαβ (x) = ϕαγ (x)ϕγβ (x). Then
there exists a principal bundle P whose transisitions functions are the ϕα ’s for the covering
{Uα }α∈I .
The trivial bundle is simply P = M × G and π(x, g) = x with the action (x, a) · g = (x, ag).
Field theory from a bundle point of view 17

4.2 Morphisms and such. . .

An homomorphism between P (G, M ) and P ′ (G′ , M ′ ) is a differentiable map h : P → P ′ such
that ∀ξ ∈ P, g ∈ G,
h(ξ · g) = h(ξ) · hG (g) (4.3)
where hG : G → G′ is a Lie group homomorphism. From the definition, h maps a fiber to
only one fiber, but it is not specially surjective on any fiber. So h induces an homomorphism
hM : M → M ′ such that π ′ ◦ h = hM ◦ π.
An isomorphism is an homomorphism g : P (G, M ) → P ′ (G′ , M ′ ) such that

• hP is a diffeomorphism P → P ′ ,
• hG is a Lie group homomorphism G → G′ , and
• hM is a diffeomorphism M → M ′ .

A principal bundle is trivial if one can find an isomorphism h : G × M → P such that

π ◦ h = id ◦ pr2 , i.e. the following diagram commutes :

h /P
G×M (4.4)
pr2 π
 
M /M
id

We say that P is locally trivial if for every x ∈ M , there exists an open neighborhood U in M
such that π −1 (U) endowed with the induced structure of principal bundle is trivial.

4.3 Frame bundle

In the ideas, the building of a vector bundle is just to put a vector space on each point of the
base manifold. A principal bundle is to put something on which a group acts on each point. If
you have a vector bundle on a manifold, you can consider, on each point x ∈ M , the set of all
the basis of the fiber Ex over x. The group GL(r, K) naturally acts on this set which becomes
a candidate to be a GL(r, K)-principal bundle.
p
More formally, we consider a vector bundle F −→ M , and for each x, the set of the basis of
the vector space Fx = p−1 (x). We define
[
P = (basis of Fx ).
x∈M

We naturally consider the projection π : P → M , π(bx ) = x if bx is a basis of Fx .

−1
Let φFα : p (Uα ) → Uα × Kr be a local trivialization of F , and {e1 , . . . , er }, the canonical
r
basis of K . We naturally define
−1
S αi (x) = φF
α (x, ei ).
The set {S α1 (x), . . . , S αr (x)} is a “reference” basis of Fx with respect to the trivialization φα .
If we choose another basis {v 1 , . . . v r } of Fx , we can find a matrix A ∈ GL(r, K) such that
v k = Alk S αl (x). This gives a bijection
−1
φP
α : π (Uα ) → Uα × GL(r, K)
(4.5)
(v 1 , . . . , v r ) → (x, A).
18 Claessens

One can gives to P a GL(r, K)-principal bundle structure such that the φP α are diffeomor-
phism.
Let (Uα , φF F
α ) be a local trivialization of F and gαβ : Uα ∩ Uβ → GL(r, K). In this case,
P P F
(Uα , φα ) is a trivialization of P whose transition function is gαβ = gαβ . Indeed
P −1
φP
α ◦ φβ (x, A) = φP
α ({v 1 , . . . , v r })
−1 −1
where v s = (φFβ) (x, Als el ). In order to see it, recall that v s = Als S αl (x) and that φF
α (x, es ) =
S αs (x). Then
−1
v s = (φF
β) (x, Als el ) = Als S αs (x).
P −1
On the other hand, from the definition of φP
β , the basis (φβ ) (x, A) is the one obtained by
applying A on S. With all this,
P −1 F −1
φP
α ◦ (φβ ) (x, A) = φP
α {(φβ ) (x, Als el )}s=1,...r
F −1 F −1
= φP
α {(φα ) ◦ (φE
α ◦ φβ )(x, Als el )}s=1,...r
E −1
(4.6)
= φP
α {(φα )
F
(x, gαβ (x)si Als el )}i=1,...r
F
= (x, gαβ (x)A).
F
The last product gαβ (x)A is a matricial product.

4.4 Sections of principal bundle

A section of a G-principal bundle is a smooth map s : M → P such that s(x) ∈ π −1 (x) for
any x ∈ M .
If φP
α is a trivialization of P on Uα , it defines a section over Uα by
−1
σα (x) = (φP
α) (x, e)
where e is the neutral of the group. In the inverse sense, we have the following :
Proposition 4.4. If σα : Uα → P is local section of P over Uα ⊂ M , then the definition
φP
α (ξ) = (x, a) if ξ = σα (x) · a is a local trivialization.
−1
Proof. The function φP α is well defined because ξ ∈ π (Uα ) implies the existence of a x ∈ Uα
such that ξ ∈ π −1 (x) = {ξ · g} ≃ G. For this x, there exists a g ∈ G such that ξ = σα (x) · g.
Now we prove that the couple (x, a) is unique in the sense that sα (x) · a = σα (y) · b implies
(x, a) = (y, b). The left hand side belongs to π −1 (x) while the right one belongs to π −1 (y). Then
x = y. The condition π −1 (x) ≃ G imposes the unicity of the g making ξ = η · g for each couple,
ξ, η ∈ π −1 (x).
If σ and σ ′ are two sections of the same principal bundle P , then there exists a differentiable
map f : M → G such that σ ′ (x) = σ(x) · f (x). So all the sections can be deduced from only
one by multiplicating by such a f .
Theorem 4.5. If π : P (G, M ) → M is a principal bundle, then the four following propositions
are equivalents :
(i) P is trivial,
(ii) P has a global section,
(iii) there exists a differentiable map γ : P → G such that γ(ξ · g) = g −1 γ(ξ) for all ξ ∈ P and
g ∈ G,
(iv) there exists a differentiable map ρ : P → G such that ρ(ξ · g) = ρ(ξ)g.
Field theory from a bundle point of view 19

4.5 Equivalence of principal bundle

Two principal bundles π : P → M and π ′ : P ′ → M are equivalent if there exists a diffeo-
morphism ϕ : P → P ′ such that

• π′ ◦ ϕ = π

• ϕ(ξ · g) = ϕ(ξ) · g.

If {Uα }α∈I is an open covering of M on which we have trivialisations φα of P and ψα of P ′ ,

the diffeomorphism ϕ induces some functions λ : Uα → G by setting

(φα ◦ ϕ−1 ◦ ψα−1 )(x, a) = (x, λα (x)a).

This definition works because from the definitions of principal bundle and equivalence, one sees
that (φα ◦ ϕ−1 ◦ ψα−1 )(x, ·) = (x, ·).

Transition functions
We have some transition functions for P and P ′ given by equations

(φα ◦ φ−1
β )(x, g) = (x, gαβ (x)g)

(ψα ◦ ψβ−1 )(x, g) = (x, gαβ

′
(x)g).

′
Now, we want to know what is gαβ in function of gαβ . First remark that (ψα ◦ ϕ ◦ φ−1
α )(x, a) =
−1
(x, λα (x) )a, and next, compute

(x, gαβ (x)a)a = (ψα ◦ ϕ ◦ φ−1

β ◦ φβ ◦ ϕ
−1
◦ ψβ−1 )(x, a)
= (ψα ◦ ϕ ◦ φ−1
β )(x, λβ (x)a)
−1
(4.7)
= (ψα ◦ ϕ ◦ φ−1
α ◦ φα ◦ φβ )(x, λβ (x)a)

= (x, λα (x)−1 gαβ (x)λβ (x)a).

Then
gαβ (x) = λα (x)−1 gαβ (x)λβ . (4.8)
One can show that if two principal bundle have transition functions whose fulfill this condition,
they are equivalent. A G-principal bundle is trivial if it is equivalent to the one given by
π1 : M × G → M .

5 Associated bundle
Let π : P → M be a G-principal bundle and ρ : G → GL(V ), a representation of G on a vector
space V (on K = R or C) of dimension r.
p
The associated bundle E = P ×ρ V −→ M is defined as following. On P × V , we consider
the equivalence relation
(ξ, v) ∼ (ξ · g, ρ(g −1 )v)
for g ∈ G, ξ ∈ P and v ∈ V . Then we define

• E = P ×ρ V := P × V / ∼,
20 Claessens

• p[(ξ, v)] = π(ξ)

where [(ξ, v)] is the class of (ξ, v) in P × V .

If φPα (ξ) = (π(ξ), a(ξ)) is a trivialization of P on Uα , then

φE [(ξ, v)] = (π(ξ), ρ(a)v) (5.1)

is a trivialization of E.
In order to see that it is a good definition, let us consider (η, w) ∼ (ξ, v). It immediately
gives the existence of a g ∈ G such that η = ξ · g and w = ρ(g −1 )v. Then φE [(ξ · g, ρ(g −1 )v)] =
(π(ξ · g), ρ(b)ρ(g −1 )v). From the definition of φE , the vector b is given by φP (ξ · g) = (π(ξ · g), b),
and the definition of a principal bundle gives b = φπ(ξ) (ξ · g) = φπ(ξ) (ξ) · g = ag. The fact that
ρ is an homomorpism makes ρ(ag)ρ(g −1 ) = ρ(a)v and φE is well defined.
Let G be a Lie group, ρ a representation of G on V and M , a manifold. We consider
pr1 p
P = M × G −→ M , the trivial G-principal bundle on M . Then E = P ×ρ V −→ M is trivial,
i.e. we can build a ϕ : P ×ρ V → M × V such that pr1 ◦ϕ = p. It is rather easy : we define


ϕ (x, g), v = (x, ρ(g)v).

It is easy to see that (pr1 ◦ϕ)[(x, g), v] = x and p[(x, g), v] = pr1 (x, g) = x.

5.1 Transition functions

π
Proposition 5.1. Let (Uα , φP α ) be a trivialization of P −→ M whose transition functions are
p
gαβ : Uα ∩ Uβ → G. Then (Uα , φE α ) given by (5.1) is a local trivialization of E −→ M whose
E
transition functions gαβ : Uα ∩ Uβ → GL(dim V, K) are given by

E P
gαβ (x) = ρ(gαβ (x)).

Proof. If we write a := φE βx
(π −1 (x)), we have φP
β (π
−1
(x)) = (x, a) and φE E −1
α ◦ (φβ ) (x, v) =
E −1 −1
φα [(π (x), ρ(a) v)]. So,


−1 
−1
φE
α [(π (x), ρ(a)−1 v)] = x, ρ φαx (π −1 (x)) ρ φβx (π −1 (x)) v

 (5.2)
= x, ρ φαx (π −1 (x))φβx (π −1 (x)) .

Then


E
gαβ = ρ φαx (π −1 (x))φβx π −1 (x)
(5.3)
P
= ρ(gαβ (x)).

5.2 Sections on associated bundle

p π
We consider a bundle E = P ×ρ V −→ M associated to the principal bundle P −→ M and a
section ψ : M → E. We define the function ψ̂ : P → V by

ψ(π(ξ)) = [ξ, ψ̂(ξ)]. (5.4)

Field theory from a bundle point of view 21

Let us see the condition under which this equation well defines ψ̂. First, remark that a ψ defined
by this equation is a section because p[ξ, v] = π(ξ), so that (p ◦ ψ)(π(ξ)) = π(ξ). Now, consider
a η such that π(η) = π(ξ). Then there exists a g ∈ G for which η · g = ξ. For any g and for this
one in particular,
ψ(π(η)) = [η, ψ̂(η)] = [η · g, ρ(g −1 )ψ̂(η)].
Then equation (5.4) defines ψ̂ from ψ if and only if

ψ̂(ξ · g) = ρ(g −1 )ψ̂(ξ). (5.5)

This condition is called the equivariance of ψ̂. Reciprocally, any equivariant function ψ̂ defines
a section of E = P ×ρ V .
If η = ξ · g = χ · k, one can define a sum

[ξ, v] + [χ, w] = [η, ρ(g)v + ρ(k)w]. (5.6)

If ψ, η : M → E are two sections defined by the equivariant functions ψ̂, η̂ : P → V , then the
section ψ + η is defined by the equivariant function ψ̂ + η̂.

Local expressions
−1
We consider a local trivialisation φP
α : π (Uα ) → Uα × G of P on Uα and the corresponding
section σα : Uα → P given by
−1
σα (x) = (φPα) (x, e).
We saw at page 20 that a trivialisation of P gives a trivialization of the associated bundle
E = P ×ρ V ; the definiton is
φE
α [(ξ, v)] = (π(ξ), ρ(a)v) (5.7)
if φP
α (ξ) = (π(ξ), a). With ξ = σα (x), we find

φE
α [(σα (x), v)] = (π(σα (x)), ρ(a)v)
(5.8)
= (x, v).

The section ψ can also be seen with respect ot the “reference” sections σα by means of the
definition
ψ(x) = [σα (x), ψ (α) (x)] (5.9)
for a function ψ (α) : M → V .

Lemma 5.2. Let ψ : M → E be a section and ψ̂ : P → V , the corresponding equivariant

function. Then
ψ (α) (x) = ψ̂(σα (x)).
Proof. By definition,
ψ(x) = ψ(π(ξ)) = [ξ, ψ̂(ξ)].
Thus if we consider in particular ξ = σα (x),

φE E
α (ψ(x)) = φα [ξ, ψ̂(ξ)]

= φE
α [sα (x), ψ̂(σα (x))] (5.10)
= (x, ψ̂(σα (x))).
22 Claessens

Let us anticipate. A spinor is a section of an associated bundle E = P ×ρ V where P is

a Lorente-principal bundle, V = C2 and ρ is the spinor representation of Lorentz on C2 . So a
spinor ψ : M → E is locally described by a function ψ (α) : M → C2 . The latter is the one that
we are used to handle in physics. In this picture, the transformation law of ψ under a Lorentz
transformation comes naturally.
Let {ei } be a basis of V; we consider some “reference” sections γαi of the associated bundle
E = P ×ρ V defined by
γαi (x) = [φ−1
α (x, e), ei ]. (5.11)
A general section ψ : M → E is defined by an equivariant function ψ̂ : P → V which can be
written as ψ̂(ξ) = ai (ξ)ei . If η = φ−1
α (x, e) and ξ = η · g(ξ),

ψ(x) = [ξ, ai ei ]
= ai [η, ρ(g)ei ]
(5.12)
= ai (ξ)ρ(g(ξ))ij [η, ej ]
= cj (ξ)γαj (x).
Since the left hand side of this equation just depends on x, the functions cj must actually not
depends on the choice of ξ ∈ π −1 (x). So we have cj : M → R. Indeed, if we choose χ ∈ π −1 (x),
ψ(x) = cj (ξ)γαj (x)
!
= [ξ, ai (χ)ei ]
= ...
= cj (χ)γαj (x),
so that cj (ξ) = cj (χ). So any section ψ : M → E can be decomposed (over the open set Uα ) as
ψ(x) = siα (x)γαi (x). (5.13)

5.3 Gauge transformation of sections

A gauge transformation of the G-principal bundle π : P → M is a diffeomorphism ϕ : P → P
such that
• π ◦ ϕ = π,
• ϕ(ξ · g) = ϕ(ξ) · g.
When we consider some local sections on σα : Uα → P , we can describe a gauge transfor-
mation with a function ϕ̃α : M → G by requiring
ϕ(σα (x)) = σα (x) · ϕ̃α (x).
This formula defines ϕ from ϕ̃ as well that ϕ̃ from ϕ.
Such a transformation ϕ : P → P naturally acts on a section ψ : M → E by
(ϕ · ψ)(x) = [ϕ(ξ), v]. (5.14a)

if ψ(x) = [ξ, v]. This law can also be seen on the equivariant function ψ̂ which defines ψ. The
rule is
[
ϕ · ψ(ξ) = ψ̂(ϕ−1 (ξ)). (5.14b)
Field theory from a bundle point of view 23

[ !
Indeed if ψ(x) = [ξ, v] = [ξ, ψ̂(ξ)], in the same way (ϕ · ψ)(x) = [ξ, ϕ · ψ(x)] = [ϕ(ξ), v] =
[ϕ(ξ), ψ̂(ξ)]. Taking ξ → ϕ−1 (ξ) as representative, (ϕ · ψ)(x) = [ξ, ψ̂ ◦ ϕ−1 (ξ)].

6 Connections on fiber bundles

Now we begin the most important —and the most difficult I think— part of our fiber bundle
excursion. Much more details and matter can be found in [6].

6.1 Connection on principal bundle

First definition : 1-form
π
We consider a G-principal bundle P −→ M , and G, the Lie algebra of G. A connection on P
is a 1-form ω ∈ Ω(P, G) which fulfills

• ωξ (A∗ξ ) = A,

• (Rg∗ ω)ξ (Σ) = Ad(g −1 )(ωξ (Σ)),

for all A ∈ G, g ∈ G, ξ ∈ P and Σ ∈ Tξ P .

We denote by A∗ the fundamental field associated to A for the action of G on P :
dh i
A∗ξ = ξ · e−tA ,
dt t=0

and Rg : P → P is the action of G on P , i.e. Rg ξ = ξ ·g. For each ξ ∈ P , we have ωξ : Tξ P → G.

Second definition : horizontal space

For each ξ ∈ P , we define the vertical space Vξ P as the subspace of Tξ P whose vectors are
tangent to the fibers : each v ∈ Vξ P fulfills dπv = 0. Any such vector is given by a path
contained i fiber of ξ. So, v ∈ Vξ P if and only if there exists a path g(t) ∈ G such that
h in the
d
v = dt ξ · g(t) .
t=0
A connection Γ is a choice, for each ξ ∈ P , of an horizontal space Hξ P such that

• T ξ P = V ξ P ⊕ Hξ P ,
• Hξ·g = (dRg )ξ Hξ ,
• Hξ P differentiably depends on ξ.

The second condition means that the distribution ξ → Hξ is invariant under G. Thanks to the
first one, for each X ∈ Tξ P , there exists only one choice of Y ∈ Hξ P and Z ∈ Vξ P such that
X = Y + Z. These are denoted by vX and hX and are naturally named horizontal and vertical
components of X. The third condition means that if X is a differentiable vector field on P , then
vX and hX are also differentiable vector fields. We will often write Vξ and Hξ instead of Vξ P
and Vξ P .
The word connection probably comes from the fact that the horizontal space gives a way to
jump from a fiber to the next one. When we consider a connection Γ, we can define a G-valued
connection 1-form by
ω(X)∗ξ = vXξ .
24 Claessens

Theorem 6.1. If Γ is a connection on a G-principal bundle, and ω is its 1-form, then

(i) for any A ∈ G, we have ω(A∗ ) = A,

(ii) (Rg )∗ ω = Ad(g −1 )ω, i.e. for any X ∈ Tξ P , g ∈ G and ξ ∈ M ,

ω((dRg )ξ X) = Ad(g −1 )ωξ (X)

Conversely, if one has a G-valued 1-form on P which fulfills these two requirement, then one has
one and only one connection on P whose associated 1-form is ω.

The projection π : P → M induces a linear map dπ : Tξ P → Tx M . We will see that, when

a connection is given, it is an isomorphism between Hξ and Tx M (if x = π(ξ)). The horizontal
lift of X ∈ X(M ) is the unique horizontal vector field (i.e. it is pointwise horizontal) such that
dπ(X ξ ) = Xπ(ξ) . The proposition which allows this definition is the following.

Proposition 6.2. For a given connection on the G-principal bundle P and a vector field X on
M , there exists an unique horizontal lift of X. Moreover, for any g ∈ G, the horizontal lift is
invariant under dRg .
The inverse implication is also true : any horizontal field on P which is invariant under dRg
for all g is the horizontal lift of a vector field on M .

This proposition comes from [6], chapter II, proposition 1.2.

Proof. We consider the restriction dπ : Hξ → Tπ(ξ) M . It is injective because dπ(X −Y ) vanishes

only when X − Y is vertical or zero. Then it is zero. It is cleat that dπ : Tξ P → Tπ(ξ) M is
surjective. But dπX = 0 if X is vertical, then dπ is surjective from only Hξ .
So we have existence and unicity of an horizontal lift. Now we turn our attention to the
invariance. The vector dRg X ξ is a vector at ξ · g. From the definition of a connection, dRg Hξ =
Hξ·g , then dRg X ξ is the unique horizontal vector at ξ · g which is sent to Xx by dπ. Thus it is
X ξ·g .
For the inverse sense, we consider X, an horizontal invariant vector field on P . If x ∈ M , we
choose ξ ∈ π −1 (x) and we define Xx = dπ(X ξ ). This construction is independent of the choice
of ξ because for ξ ′ = ξ · g, we have

dπ(X ξ′ ) = π(dRg X ξ ) = π(X ξ ).

An other way to see the invariance is the following formula :

X ξ·g = (dRg )ξ X ξ .

By definition, X ξ·g is the unique vector of Tξ·g P which fulfils dπX ξ·g = Xx if ξπ −1 (x), so the
following computation proves the formula :

(dπ)ξ·g ((dRg )ξ X ξ ) = d(π ◦ Rg )ξ X ξ

= dπξ X ξ (6.1)
= Xx .
Field theory from a bundle point of view 25

Curvature
The curvature of a vector or associated bundle satisfies Ωα = dθα + θα ∧ θα . So we naturally
define the curvature of the connection ω on a principal bundle as the G-valued 2-form

Ω = dω + ω ∧ ω. (6.2)

When we consider a local section σα : M → P on Uα , we can express the curvature with a

2-form on M instead of P : F (α) = σα∗ Ω; it is explicitly given by

F (α)x (X, Y ) = Ωσα (x) (dσα X, dσα Y ).

Note that if G is abelian, Ω = dω and dΩ = 0.

6.2 Covariant derivative on associated bundle

Now we consider a general G-principal bundle π : P → M and an associated bundle E = P ×ρ V .
We define a product R × E → E by

λ[ξ, v] = [ξ, λv]. (6.3)

c defines the section λψ. A covariant derivative is
It is clear that the equivariant function λψ
a map
∇ : X(M ) × Γ(M, E) → Γ(M, E)
(6.4)
(X, ψ) → ∇X ψ
such that

∇f X ψ = f ∇X ψ, (6.5a)
∇X (f ψ) = (X · f )ψ + f ∇X ψ (6.5b)

where all the products have to be understood by formula (6.3).

Theorem 6.3. A connection on a principal bundle gives rise to a covariant derivative on any
associated bundle by the formula
[
∇ E ψ(ξ) = X (ψ̂) (6.6)
X ξ

where ψ̂ : P → V is the function associated to the section ψ : M → E.

We have to prove that it is a good definition : the function ∇ [E ψ must define a section
X
∇R
X ψ : M → E and the association ψ → ∇ E
X ψ must be a covariant derivative.
[
Proposition 6.4. The function ∇E ψ defines a section of P .
X

[
Proof. We have to see that ∇E ψ is an equivariant function. The equivariance of ψ̂ gives ψ̂ ◦R =
X g
−1
ρ(g )ψ̂, thus

[
∇E ψ(ξ · g) = X
X ξ·g (ψ̂)


= (dRg )ξ X ξ (ψ̂)
= X ξ (ψ̂ ◦ Rg ) (6.7)
= X ξ (ρ(g −1 )ψ̂)
= ρ(g −1 )X ξ (ψ̂).
26 Claessens

The last equality comes from the fact that the product ρ(g −1 )ψ̂ is a linear product “matrix times
vector” and that X ξ is linear.
Theorem 6.5. The definition
[
∇E ψ(ξ) = X (ψ̂)
X ξ
defines a covariant derivative.
Theorem 6.6. Using the local coordinates related to the sections σα : Uα → P , the covariant
derivatives reads :
(∇X ψ)(α) (x) = Xx ψ (α) − ρ∗ (σα∗ ωx (X))ψ (α) (x) (6.8)
where ρ∗ : G → End(V ) is defined by
d h tA i
ρ∗ (A) = ρ(e ) (6.9)
dt t=0

Proof. The problem reduces to the search of X because

[
(∇X ψ)(α) (x) = ∇X ψ(σα (x)) = X σα (x) (ψ̂).

We claim that X σα (x) = dσα Xx − ω(dσα Xx )∗ . We have to check that dπX = X and ω(X) = 0.
The latter comes easily from the fact that ω(A∗ ) = A. For the first one, remark that sα is a
section, then d(π ◦ sα ) = id, and dπ(dsα Xx ) = Xx , while
dh i
dπ(A∗ξ ) = dπ ξ · e−tA
dt t=0
dh i
(6.10)
= π(ξ)
dt t=0
= 0.
Since the horizontal lift is unique, we deduce


(∇X ψ)(α) (x) = dσα Xx − ω(dσα Xx )∗ ψ̂. (6.11)
From the definition of a fundamental vector field,
dh
i
ω(dσα Xx )∗σα (x) ψ̂ = ψ̂ σα (x) · e−tω(dσα Xx )
dt t=0
d h tω(dσα Xx ) i
= ρ(e )ψ̂(σα (x)) from (5.5) (6.12)
dt t=0
= (dρ)e (ω ◦ dσα )Xx (ψ̂ ◦ σα )(x)


= ρ∗ (σα∗ ω)(Xx ) ψ (α) (x) by (6.9)

We can express the covariant derivative by means of some maps θα : X(M ) × M → End(V )
given by
∇X γαi = θα (X)ij γαj . (6.13)
where the γαi ’s were given in equation (5.11). By the definition (6.5b),
(∇X ψ)(x) = (X · siα )x γαi (x) + siα (x)(∇X γαi )(x)
= (X · siα )x γαi (x) + siα (x)θα (X)ij γαj (x).
On the othre hand with the notations of equation (5.9), γαj = ei and Xx γαj = 0. Then equation
(6.8) gives θα (X) = ρ∗ (σα∗ ωx (X)), or
θα = ρ∗ (σα∗ ωx ). (6.14)
Field theory from a bundle point of view 27

Curvature on associated bundle

From the definition (5.6), it makes sense to define the curvature 2-form by

R(X, Y )ψ = ∇X ∇Y ψ − ∇Y ∇X ψ − ∇[X,Y ] ψ.

It is also clear that ψ (α) defines a section of the trivial vector bundle F = M × V by x →
(x, ψ (α) (x)), so one can define Ωα (X, Y ) : Γ(M, E) → Γ(M, E) by


R(X, Y )ψ (α) = Ωα (X, Y )ψ (α)

and take back all the work around Bianchi because of the relation (6.8) which can be written as
(∇X ψ)(α) (x) = Xx ψ (α) + θα (X)ψ (α) (x) and which is the same that 3.5.

6.3 Connection on frame bundle

p
The frame bundle was defined at page 17. Let F −→ M be a K-vector bundle with some local
trivialization (Uα , φE
α ) and the corresponding transition functions gαβ : Uα ∩ Uβ → GL(r, K).
We consider π : P → M , the frame bundle of F ; it is a GL(r, K)-principal bundle. Let ∇ be a
covariant derivative on F and θα , the associated matrices 1-form. The frame bundle is
[
P = (frame of Fx ).
x∈M

A connection is a G-valued 1-form; in our case it is a map

ωξα : Tξ π −1 (Uα ) → gl(r, K).

We define our connection by, for g ∈ GL(r, K), x ∈ Uα , Xx ∈ Tx M and A ∈ gl(r, K),


ωSαα (x)·g Rg ∗ sα (x)∗ Xx + A∗Sα (x)·g := A + Ad(g −1 )θα (Xx ). (6.15)

where Sα : Uα → P is the section defined by the trivialization φP

α :

−1
Sα (x) = {v α = φE
α (x, ei )}i=1,...,r .

Since θα (Xx ) ∈ End(Kr ) ⊂ gl(r, K), the second term of (6.15) makes sense. This formula is a
good definition of ω because of the following lemma :

Lemma 6.7. If ξ = Sα (x) · g and Σ ∈ Tξ P , there exists a choice of A ∈ G, and Xx ∈ Tx M such

that
Σ = Rg ∗ sα (X)∗ Xx + A∗Sα (x)·g . (6.16)

Proof. If ξ ∈ P is a basis of E at y, there exists only one choice of x ∈ M and g ∈ G such that
ξ = Sα (x) · g.
Let us consider a general path c : R → P under the form c(t) = sα (x(t)) · g(t) where x and g
are path in M and GL(r, K). The frame c(t) is the one of Fx(t) obtained by the transformation
g(t) from sα (x(t)). It is a set of r vectors, and each of them can be written as a combinaison of
the vectors of sα (x(t)), so we write

ci (t) = sjα (x(t))gji (t) (6.17)

28 Claessens

where sjα (x(t)) ∈ Fx(t) and gji (t) ∈ K. We compute Σ = c′ (0) by using the Leibnitz rule and we
denote x′ (0) = Xx , x(0) = x and gji (0) = gji (the matrix of g) :

dh j i dh i i
Σi = sα (x(t)) gji + sjα (x) g (t)
dt t=0 dt j t=0 (6.18)
′
= (dsjα )x Xx gji + gji (0)sjα (x).

Going to more compact matricial form, it gives

Σ = (dsα )x Xx · g + sα (x)g ′ (0).

The second term, sjα (x)gj′i (0), is a general vector tangent to a fiber. So it can be written as a
fundamental field A∗ξ .

Lemma 6.8. On Uα ∩ Uβ , the form fulfills ω α = ω β .

Proposition 6.9. The ω defined by formula (6.15) is a connection 1-form.

6.4 Connection and gauge transformation

Proposition 6.10. If ω is a connection on a G-principal bundle and ϕ, a gauge transformation,
the form β = ϕ∗ ω is a connection 1-form too.
Proof. It is rather easy to see that ϕ∗ A∗ξ = A∗ϕ(x) :

dh i dh i
ϕ∗ A∗ξ = ϕ(ξe−tA ) = ϕ(ξ)e−tA = A∗ϕ(x) .
dt t=0 dt t=0

The same kind of reasoning leads to ϕ∗ Rg ∗ = Rg ∗ ϕ∗ . With this at hand,

(ϕ∗ ω)ξ (A∗ξ ) = ωϕ(ξ) (ϕ∗ A∗ξ ) = A,

and


Rg∗ (ϕ∗ ω)ξ (Σ) = (Rg∗ ω)ϕ(ξ) (ϕ∗ Ω) = Ad(g −1 ) (ϕ∗ ω)ξ (Σ) .

So, the “gauge transformed” of a connection is still a connection. It is hopeful in order to

define gauge invariants objects (lagrangians) from connections (electromagentic fields).

Local description Let π : P → M be a G-principal bundle given with some trivializations

−1
φP
α : π (Uα ) → Uα × G over Uα ⊂ M and sα : Uα → π −1 (Uα ), a section. In front of that, we
consider an associated bundle p : E = P ×ρ V → M with a trivialization φE α : E → Uα ×V . One
can choose a section sα comptatible with the trivialization in the sense that φP α (sα (x)·g) = (x, g);
the same can be done with E by choosing φE α ([sα (x), v]) = (x, v). All this given in figure 1.1.
A section ψ : M → E is described by a function ψα : Uα → V defined by

φE
α (ψ(x)) = (x, ψα (x)). (6.19)

In the inverse sense, ψ is defined (on Uα ) from ψα by

ψ(x) = [sα (x), ψα (x)]. (6.20)

Field theory from a bundle point of view 29

In the same way, a gauge transformation ϕ : P → P is described by functions ϕ̃α : Uα → G,

ϕ(sα (x)) = sα (x) · ϕ̃α (x). (6.21)

The function ϕ̃α also fulfil

−1
(φP P
α ◦ ϕ ◦ φα )(x, g) = (x, ϕ̃(x) · g) (6.22)

because
−1
(φP P
α ◦ ϕ ◦ φα )(x, g) = (φP
α ◦ ϕ)(sα (x) · g)

= φP
α (ϕ(sα (x)) · g)
(6.23)
= φP
α (sα (c) · ϕ̃α (x)g)
= (x, ϕ̃α (x)g).

We know that a connection on P is given by its 1-form ω. Moreover we have the following :
Proposition 6.11. A connection on P is completely determined on π −1 (Uα ) from the data of
the G-valued 1-form σα∗ ω on Uα .
If we have a connection on P , we can define a covariant derivative on the associated bundle
E by
(∇X ψ)(α) (x) = Xx (ψα ) + ρ∗ (s∗α ωx (X))ψ (α) (x),
the matricial 1-form being given by θα = ρ∗ σα∗ ω. The gauge transformation ϕ acts on the
connection ω by defining ω ϕ := ϕ∗ ω.
Proposition 6.12. If β = ϕ∗ ω, then

s∗α (β) = Adϕ̃α (x)−1 s∗α (ω) + ϕ̃α (x)−1 dϕ̃α .

7 Frame bundle
7.1 Construction
We already saw in subsection 4.3 how to build a frame bundle over a manifold. One can
get another expression of the frame bundle when we express a basis of Tx M by means of an
isomorphism between Rn and Tx M . If M is a n-dimensional manifold, a frame at x is an
ordered basis
b = (b1 , . . . , bn )
of Tx M . It is clear that any frame defines an isomoprhism (linear bijective map)

b̃ : Rn → Tx M
(7.1)
ei 7→ ei

where {ei } is the canonical basis of Rn . It is also clear that any isomorphism gives rise to a
frame. Then we see a frame of M at x as an isomorphism b̃ : Rn → Tx M . Let B(M )x be se the
of all the frames of M at x; we define
[
B(M ) = B(M )x .
x∈M
30 Claessens

For all b ∈ B(M )x , we define pB (b) = x and the action B(M ) × GL(n, R) by b · g = (b′1 , . . . , b′n )
where
b′j = bi gi j . (7.2)

It is easy to see that bg

· g = b̃ ◦ g : Rn → Tx M . So we can give to

GL(n, R) /o /o /o / B(M ) (7.3)

pB

M

a structure of principal bundle5 . If (Uα , ϕα ) is a local coordinate chart on M , we define

ϕ̃ : p−1
B (Uα ) → ϕα (Uα ) × GL(n, R)
(7.4)
b → (ϕα (x), A(b))
where A(b) ∈ GL(n, R) is defined by the condition bj = Aj i ∂i |x . The matrix A(b) is the one
which transforms the canonical basis (in the trivialisation ϕα ) into b ∈ B(M )x . That’s for the
principal bundle structure.
The manifold structure of B(M ) is given by Φα : p−1 B (Uα ) → Uα × GL(R),

Φ(b) = (ϕ−1
α × id |GL(n,R) ) ◦ ϕ̃(b)
= (x, A(b)) (7.5)
= (pB (b), A(b)).

It fulfils A(b · g) = A(b) · g. A section s : Uα → B(M ) is sometimes called a moving frame

over Uα .

7.2 Example : tangent bundle

As B(M ) is a principal bundle, one can consider some associated bundles on it. We are now
going to see that the one given by the definition representation ρ : GL(n, R) → GL(n, R) on Rn
is the tangent bundle. So we study B(M ) ×ρ Rn . By choosing a basis on each point of M , we
identify each Tx M to Rn . An element of B(M ) × Rn is a pair (b, v) with b = (b1 , . . . , bn ) and
v = (v 1 , . . . , v n ). We can identify v to the element of Tx M given by v = v i bi .
In order to build the associated bundle, we make the identifications

(b, v) · g ∼ (b · g, g −1 v).

Here, by gv we means the vector whose components are given by (gv)i = v j gj i . The tangent
vector given by (b · g, g −1 v) is

(g −1 v)i (b · g)i = v j (g −1 )j i gi k bk
(7.6)
= v k bk

So the identification map ψ : B(M ) ×ρ Rn → T M given by

ψ([b, v]) = v i bi

is well defined.
5 All this is given with much more details and proofs in [4].
Field theory from a bundle point of view 31

8 Product of principal bundle

In this section, we build a G1 × G2 -principal bundle from the data of a G1 and a G2 -principal
bundle. The physical motivation is clear : as far as electromagnetism is concerned, particles are
sections of U (1)-principal bundle while the reletivistic invariance must be expressed by means
of a SL(2, C)-associated bundle. So the physical fields must be sections of something as the
product of the two bundles. See section 12.

8.1 Putting together principal bundle

Let us consider two principal bundle over the same base space
p1
G1 /o o/ o/ / P1 / M,

p2
G2 /o /o o/ / P2 / M.

First we define the set

P1 ◦ P2 = {(ξ1 , ξ2 ) ∈ P1 × P2 | p1 (ξ1 ) = p2 (ξ2 )} (8.1)
which will be the total space of our new bundle. The projection p : P1 ◦ P2 → M is naturally
defined by
p(ξ1 , ξ2 ) = p1 (ξ1 ) = p2 (ξ2 ),
while the right action of G1 × G2 on P1 ◦ P2 is given by
(ξ1 , ξ2 ) · (g1 , g2 ) = (ξ1 · g1 , ξ2 · g2 )
With all these definitions,
G1 × G2 /o /o o/ / P1 ◦ P2
p

M
is a G1 × G2 -principal bundle over M . We define the natural projections

πi : P 1 × P 2 → P i
(8.2)
(ξ1 , ξ2 ) → ξi ,
and if ei denotes the identity element of Gi , we can indentify G1 to G1 ×{e2 } and G2 to G2 ×{e1 };
in the same way, G1 = G1 × {0} ⊂ G1 × G2 . So we get the following principal bundles :
π1
G2 /o /o o/ / P1 ◦ P2 / P1

π2
G1 /o /o o/ / P1 ◦ P2 / P2 .

It is clear that the following diagram commutes :

π1 π2
P1 Go P1 ◦ P2 / P2
GG ww
GG w
GG p ww
P1 GG ww p2
#  {ww
M
32 Claessens

8.2 Connections
Let ωi be a connection on the bundle pi : Pi → M . Using the identifications, π1∗ ω1 is a
connection on π2 : P1 ◦ P2 → P2 (the same is true for 1 ↔ 2), and π1∗ ω1 ⊕ π2∗ ω2 is a connection
on p : P1 ◦ P2 → M . Let us prove the first claim.
Let A ∈ G1 . We first have to prove that π1∗ ω1 (A∗ ) = A. For this, remark that A = (A, 0) ∈
G1 ⊕ G2 and
dh i
A∗ξ = ξ · e−t(A,0)
dt t=0
dh −tA
i
= (ξ1 , ξ2 ) · (e , e2 ) (8.3)
dt t=0
dh −tA
i
= (ξ1 · e , ξ2 ) ,
dt t=0

so h i h i
d d
dπ1 A∗ = dt π1 (. . .) = ω1 (A∗ ) = A. Let now Σ ∈ T(ξ1 ,ξ2 ) (P1 ◦P2 ). If Σ = dt (ξ1 (t), ξ2 (t)) ,
t=0 t=0
then
∗


R(g,e ∗
2 )π1 ω1 (ξ1 ,ξ2 )
Σ = (π1∗ ω1 )(dR(g,e2 ) Σ)
dh i
= ω1 ( ξ1 (t) · g )
dt t=0
d h i
= ω1 (dRg ξ1 (t) ) (8.4)
dt t=0
dh i
= Ad(g −1 )π1∗ ω1 ( (ξ1 (t), ξ2 (t)) )
dt t=0
= Ad(g −1 )π1∗ ω1 Σ.

8.3 Representations
Let V be a vector space and ρi : Gi → GL(V ) be some representations such that

[ρ1 (g1 ), ρ2 (g2 )] = 0 (8.5)

for all g1 ∈ G1 and g2 ∈ G2 (in the sense of commutators of matrices). In this case, one cas
define the representation ρ1 × ρ2 : G1 × G2 → GL(V ) by

(ρ1 × ρ2 )(g1 , g2 ) = ρ1 (g1 ) ◦ ρ2 (g2 ) = ρ2 (g2 ) ◦ ρ1 (g1 ). (8.6)

The relation (8.5) is needed in order ρ1 × ρ2 to be a representation, as one can check by writing
down explicitly the requirement


(ρ1 × ρ2 ) (g1 , g2 )(g1′ , g2′ ) = (ρ1 × ρ2 )(g1 g1′ , g2 g2′ )
Building gauge invariant theories

9 Connections
This chapter actually don’t deals with quantum field theory in the sense that our wavefunctions
aren’t operators which acting on a Fock space. So this is just relativistic field theory. For
reference, see [2, 4].

9.1 Gauge potentials

Let us consider a section σα of P over Uα . It is a map σα : Uα → P such that π ◦ σα = id. A
connection on P is a 1-form ω : Tp P → G ∈ Ω1 (P ) which satisfies the following two conditions :

ωp (Yp∗ ) = Y, (9.1a)
−1
ω(dRg ξ) = g ω(ξ)g. (9.1b)
The gauge potential of ω with respect of the section σα is the 1-form on Uα given by
Aα (x)(v) = (σα∗ ω)x (v). (9.2)
We will not always explicitly write the dependance of Aα in x. Now we consider another section
σβ : Uβ → P which is related on Uα ∩ Uβ to σα by σβ (x) = σα (x) · gαβ (x) for a well defined
map gαβ : Uα ∩ Uβ → G.
Proposition 9.1. The gauge potentials Aα and Aβ are related by
Aβ = g −1 Aα g − g −1 dg. (9.3)
Proof. By definition, for v ∈ Tx Uα ,


Aβ (v) = (σβ∗ ω)x (v) = ωσα (x)·gαβ (x) (dσβ )x (v) .
We begin by computing dσβ (v). Let us take a path v(t) in Uα such that v(0) = x and v ′ (0) = v.
We have :

d 
(dσβ )x (v) = σβ (v(t))
dt t=0

d 
= σα (v(t)) · gαβ (v(t))
dt t=0
dh i dh i (9.4)
= σα (v(t)) · gαβ (x) + σα (x) · gαβ (v(t))
dt t=0 dt t=0
dh i
= dRgαβ (x) dσα (v) + σα (x) · gαβ (x)e−ts
dt t=0
= dRgαβ (x) dσα (v) + s∗σα (x)·gαβ (x)

33
34 Claessens

where s is defined by the requirement that gαβ (x)−1 gαβ (v(t)) can be replaced in the derivative
by e−ts , so that we can replace gαβ (v(t)) by gαβ (x)e−ts . As far as the derivatives are concerned,
e−ts = gαβ (x)−1 gαβ (v(t)), then

d 
s = − gαβ (x) gαβ (v(t))
−1
= −gαβ (x)−1 dgαβ (v),
dt t=0

this equality being a notation. Now, properties (9.1a) and (9.1b) make that

Aβ (v) = gαβ (x)−1 ωσα (x) (dσα (v))gαβ (x) + s.

The thesis is just the same, with “reduced” notations .

An explicit form for this transformation law is :
d h −1 tAα (v) i d h −1 i
Aβ (v) = g e g − g gαβ (v(t)) , (9.5)
dt t=0 dt t=0

where g := gαβ (x).

9.2 Covariant derivative

When we have a connection on a principal bundle, we can define a covariant derivative on
any associated bundle. Let us quickly review it. An associated bundle is the semi-product
E = P ×ρ V where V is a vector space on which acts the representation ρ of G. We denote
the canonical projection by πp : E → M . The classes are taken with respect to the equivalence
relation (p, v) ∼ (p · g, ρ(g −1 )v).
A section of E is a map ψ : M → E such that π ◦ ψ = id. We denote by Γ(E) the set of all
the sections of E. A section of E defines (and is defined by) an equivariant function ψ̂ : P → V
such that

ψ(π(ξ)) = [ξ, ψ̂(ξ)], (9.6a)

−1
ψ̂(ξ · g) = ρ(g )ψ̂(ξ). (9.6b)

For a section ψ ∈ Γ(E), we define ψ (α) : Uα → V by

ψ (α) (x) = ψ̂(σ(x)).

We saw in (6.8) that a covariant derivative on E is given by

 
(DX ψ)(α) (x) = Xx ψ (α) − ρ∗ (σα∗ ω)x (Xx ) ψ (α) (x). (9.7)

Since (dψ)(X) = X(ψ), we can rewrite this formula in a simpler manner by forgetting the index
α and the mention of X :
Dψ = dψ − (ρ∗ Aα )ψ.
If we note (ρ∗ Aα )ψ by Aα ψ, we have

Dψ = dψ − Aψ. (9.8)

We have to see this equation as a “notational trick” for (9.7). By the way, remark that (ρ∗ Aα )
is the only “reasonable” way for A to acts on ψ.
Field theory from a bundle point of view 35

10 Gauge transformation
A gauge transformation of a G-principal bundle is a diffeomorphism ϕ : P → P which
satisfies
π ◦ ϕ = π, (10.1a)
ϕ(ξ · g) = ϕ(ξ) · g. (10.1b)
It is defined by a function ϕ̃α : Uα → G by the requirement that
ϕ(σα (x)) = σα (x) · ϕ̃α (x). (10.2)
We have shown in proposition 6.10 that, if ω is a connection 1-form on P , the form ϕ·ω := ϕ∗ ω
is still a connection 1-form on P . Of course, with the same section σα than before, we can define
a gauge potential (ϕ · A)α for this new connection. We will see how it is related to Aα . The
reader can guess the result (it will be the same that proposition 9.1). We show it.
Proposition 10.1.
(ϕ · A) = ϕ̃−1 Aϕ̃ − ϕ̃−1 dϕ̃. (10.3)
Proof. Let us consider x ∈ Uα , and v ∈ Tx Uα , the vector which is tangent to the curve v(t) ∈ Uα .
We compute
σα∗ (ϕ∗ ω)x (v) = ω(ϕ◦σα )(x) ((dϕ ◦ dσα )(v)),
but equation (10.2) makes

d 
(dϕ ◦ dσα )(v) = ϕ(σα (v(t)))
dt t=0
 (10.4)
d 
= σα (v(t)) · ϕ̃α (v(t)) .
dt t=0

Now, we are in the same situation than in equation (9.4).

If ψ : M → E is a section of E, the gauge transformation ϕ : P → P acts on ψ by
[
ϕ · ψ(ξ) = ψ̂(ϕ−1 (ξ)). (10.5)
On the other hand, ϕ acts on the covariant derivative (and the potential) : ϕ · D is the covariant
derivative of the connection ϕ · ω. Of course, we define
(ϕ · D)ψ = dψ − (ϕ · A)ψ. (10.6)
Lemma 10.2. If ϕ : P → P is a gauge transformation, then
g
(i) ϕ−1 is also a gauge transformation and (ϕ −1 ) (x) = ϕ̃ (x)−1 ,
α α

(ii) (ϕ · ψ)(α) (x) = ρ(ϕ̃−1

x )ψ (α) (x).

Proof. The first part is clear while the second is a computation :

[
(ϕ · ψ)(α) = ϕ · ψ(σα (x))
= ψ̂(ϕ−1 (σα (x)))
(10.7)
= ψ̂(σα (x) · ϕ̃α (x)−1 )
= ρ(ϕ̃α (x))ψ (α) (x).
36 Claessens

Now, we will proof the main theorem : the one which explains why the covariant derivative
is “covariant”.
Theorem 10.3.
(ϕ · D)(ϕ−1 · ψ) = ϕ−1 (Dψ). (10.8)
Remark 10.4. Let us use more intuitive notations : we write (10.3) under the form A′ = g −1 Ag −
g −1 dg. If we have two sections ψ and ψ ′ , they are necessarily related by a gauge transformation :
ψ ′ = g −1 ψ. Then, the theorem tells us that the equation Dψ = dψ −Aψ becomes D′ ψ ′ = g −1 Dψ
“under a gauge transformation”. This is : Dψ transforms under a gauge transformation as dψ
transforms under a constant linear transformation. This is the reason why D is a covariant
derivative. The physist way to write (10.8) is

D′ ψ ′ = g −1 Dψ (10.9)

Proof of theorem 10.3. First, we look at (ϕ · A)ψα . Using all the notational tricks used to give
a sens to Aψ, we write :

[(ϕ · A)X ψ](α) (x) = (ϕ · A)X ψ (α) (x) = ρ∗ (ϕ · A(X))ψ (α) (x).

But we know that ϕ · A = ϕ̃−1 Aϕ̃ − ϕ̃−1 dϕ̃, then

(ϕ · A)X ψ (α) (x) = ρ∗ (ϕ̃−1 A(X)ϕ̃)ψ (α) (x)

− ρ∗ (ϕ̃−1 dϕ̃(X))ψ (α) (x)
dh i
(10.10)
= ρ(ϕ̃−1 etA(X) ϕ̃)ψ (α) (x)
dt t=0
dh −1
i
− ρ(ϕ̃ ϕ̃(Xt ))ψ (α) (x)
dt t=0

Now, we have to write this equation with ϕ−1 · ψ instead of ψ. Using lemma 10.2, we find :
dh i
(ϕ · A)X (ϕ−1 · ψ)(α) (x) = ρ(ϕ̃−1 etA(X) ϕ̃ϕ̃−1 )ψ (α) (x)
dt t=0
(10.11)
dh −1 −1
i
− ρ(ϕ̃ ϕ̃(Xt )ϕ̃ )ψ (α) (x)
dt t=0

After simplification, the first term is a term of the thesis : ϕ̃(x)−1 (Aψ)α (x) and we let the second
one as it is. Now, we turn our attention to the second term of (10.8); the same argument gives
:
d h −1 i
d(ϕ−1 ψ (α) )x X = (ϕ · ψ)(α) (Xt )
dt t=0
dh −1
i
= ρ(ϕ̃(Xt ) )ψ (α) (Xt ) (10.12)
dt t=0
dh i dh i
= ρ(ϕ̃(Xt )−1 )ψ (α) (x) + ρ(ϕ̃−1 )ψ (α) (Xt ) .
dt t=0 dt t=0

The second term is ϕ̃−1 dψα (X). In definitive, we need to prove that the two excedent terms
cancel each other :
dh i dh i
ρ(ϕ̃−1 ϕ̃(Xt )ϕ̃−1 )ψ (α) (x) + ρ(ϕ̃(Xt )−1 )ψ (α) (x) (10.13)
dt t=0 dt t=0

must be zero.
Field theory from a bundle point of view 37

One can find a g(t) ∈ G such that ϕ̃(Xt ) = ϕ̃g(t), g(0) = e. Then, what we have in the ρ of
these two terms is respectively g(t)ϕ̃−1 and g(t)−1 ϕ̃−1 . As far as the derivative are concerned,
g(t) can be written as etZ for a certain Z ∈ G. So we see that g(t)−1 = e−tZ and the derivative
will come with the right sign to makes the sum zero.
Remark 10.5. If we naively make the computation with the notations of remark 10.4, we replace
ψ ′ = g −1 ψ and A′ = g −1 Ag − g −1 dg in
D′ ψ ′ = dψ ′ − A′ ψ ′ ,
using some intuitive “Leibnits formulas”, we find : D′ ψ ′ = dg −1 ψ+g −1 dψ+g −1 Aψ+g −1 dgg −1 ψ.
It is exactly g −1 dψ + g −1 Aψ with two additionals terms : dg −1 ψ and g −1 dgg −1 ψ. One sees that
these are precisely the two of equation (10.13).

11 A few physics
11.1 Example : electromagnetism
Let us consider the electromagnetism as the simplest example of a gauge invariant physical
theory. We first discuss the theory of free electromagnetic field (this is : without taking into
account the interactions with particles) from the Maxwell’s equations, see [1, 3]. The electric
field E and the magnetic field B are subjet to following relations :
∇ · E = ρ, (11.1a)
∇ · B = 0, (11.1b)
∇ × E + ∂t B = 0, (11.1c)
∇ × B − ∂t E = j. (11.1d)
Comparing (11.1a) and (11.1b), we see that the Maxwell’s theory don’t incorpore magnetic
monopoles. Suppose that we can use the Poincaré lemma. Equation (11.1b) gives a vector field
A such that B = ∇ × A, so that (11.1c) becomes ∇ × (E + ∂t A) = 0 which gives a scalar field
φ such that −∇ · φ = E + ∂t A.
Now the equations (11.1a)–(11.1d) are equations for the potentials A and φ, and we find
back the “physical” field by
B = ∇ × A, (11.2a)
E = −∇φ − ∂t A. (11.2b)
One can easily see that there are several choice of potentials which describe the same elec-
tromagnetic field. Indeed, if
A′ = A + ∇λ, (11.3a)
φ′ = φ − ∂t λ, (11.3b)
the the electromagnetic field given (via (11.2)) by {φ′ , A′ } is the same that the one given by
{φ, A}
The Maxwell’s equations can be written in a more “covariant” way by defining
 
0 −Ex /c −Ey /c −Ez /c
· 0 −Bz · 
F = ·
, (11.4)
· 0 −Bx 
· −By · 0
38 Claessens

F µν = −F νµ and

J = cρ jx jy jz .
We also define ⋆F αβ = 21 eαβλµ Fλµ . With all that, the Maxwell’s equations read :

∂µ F µν = µ0 J ν ,
(11.5)
∂α ⋆ F αβ = 0.

If we define  
A= φ −Ax −Ay −Az , (11.6)
c
the physical fields are given by
Fµν = ∂µ Aν − ∂ν Aµ .
The gauge invariance of this theory is the fact that
′
Fµν = ∂µ A′ν − ∂ν A′µ = Fµν (11.7a)

if
A′µ (x) = Aµ (x) + ∂µ f (x) (11.7b)
for any scalar function f (to be compared with (11.3)).
This is : in the picture of the world in which we see the A as fundamental field of physics,
several (as much as you have functions in C ∞ (R4 )) fields A, A′ ,. . . describe the same physical
situation because the fields E and B which acts on the particle are the same for A and A′ .
Now, we turn our attentions to the interacting field theory of electromagnetism. As far as
we know, the electron makes interactions with the electromagnetic field via a term ψAµ ψ in the
lagrangian. The free lagrangian for an electron is

L = ψ(γ µ ∂µ + m)ψ. (11.8)

The easiest way to include a ψAψ term is to change ∂µ to ∂µ + Aµ . But we want to preserve
the powerfull gauge invariance of classical electrodynamics, then we want the new lagrangian to
keep unchanged if we do
Aµ → A′µ = Aµ − i∂µ φ. (11.9)
In order to achieve it, we remark that the ψ must be transformed simultaneously into

ψ ′ (x) = eiφ(x) ψ(x). (11.10)

The conclusion is that if one want to write down a lagrangian for QED, one must find
a lagrangian which remains unchanged under certain transformation A → A′ and ψ → ψ ′ . In
other words the set {ψ, A} of fields which describe the world of an electron in an electromagnetic
field is not well defined from the alone given of the physical situation : it is defined up to a certain
invariance which is naturaly called a gauge invariance.
Remark 11.1. In the physics books, the matter is presented in a slighty different way. We observe
that the lagrangian (11.8) is invariant under

ψ(x) → ψ ′ (x) = eiα ψ(x) (11.11)

for any constant α. One can see that the associated conserved current (Noether) is closely related
to the electric current. The idea (of Yang-Mills) is to develop this symmetry. Since the symmetry
Field theory from a bundle point of view 39

(11.11) depends only on a constant, we say it a global symmetry; we will simultaneously add a
new field Aµ and upgrade (11.11) to a local symmetry:

ψ(x) → ψ ′ (x) = eiφ(x) ψ(x). (11.12)

Then, we deduce the transformation law of Aµ .

Because of the form of (11.10), we say that the electromagnetism is a U (1)-gauge theory.
The fact that this is an abelian group have a deep physical meaning and many consequences.

11.2 Little more general, little more formal

The aim of this text is to interpret the field A as a gauge potential for a connection. But equation
(11.9) is not exactly the expected one which is (10.3). The point is that equation (11.9) concerns
a theory in which the gauge transformation of the field was a simple multiplication by a scalar
field, so that simplifications as e−iφ(x) Aµ (x)eiφ(x) = Aµ (x) are allowed.
Now, we consider a vector space V , a manifold M and a function ψ : M → V which
“equation of motion” is
Li (∂i + mi )ψ = 0
Where we imply an unit matrix behind ∂ and m; the indices i, j are the (local) coordinates in
M and a, b, the coordinates in V . Let G be a matrix group which acts on V . If ψ is a solution,
Λ−1 ψ is also a solution as far as Λ is a constant –does not depend on x ∈ M – matrix of G. In
other words, Li (∂i + mi )ψa = 0 for all a implies Li (∂i + mi )((Λ−1 )ba ψb ) = 0.
The function, ψ ′ (x) = Λ(x)−1 ψ(x) is no more a solution. If we want it to be solution of the
same equation than ψ, we have to change the equation and consider

Li (∂i + Ai + mi )ψ = 0.

This equation is preserved under the simultaneous change

(
ψa′ = (Λ−1 )ba ψb
(11.13)
(A′i )ab = (Λ−1 )cb (Ai )dc (Λad ) − (∂i Λ−1 )db Λad .

The second line show that the formalism in which A is a connection is the good one to write down
covariant equations. This has to be compared with (9.3). Logically, a theory which inlcudes an
invariance under transformations as (11.13) is called a G-gauge theory.

11.3 A “final” formalism

Now, we work with fields which are sections of some fiber bundle build over M , the physical
space. More precisely, let G be a matrix group. We search for a theory which is “locally invariant
under G”. In order to achieve it, we consider a G-principal bundle P over M and the associated
bundle E = P ×ρ V for a certain vector space V , and a representation ρ of G on V . Typically,
V is C or the vector space of spinors6 on M .
The physical fields are sections ψ : M → E. If we choose some sections σα : M → P , they
can be exprimed by ψ (α) (x) = ψ̂(σα (x)). We translate the idea of a local invariance under G by
requiring an invariance under
ψ ′ (α) (x) = ρ(g(x))ψ (α) (x)
6 Formally, the set of spinors is not a vector space, but we can locally forget this precision.
40 Claessens

for all g : M → G. By (ii) of lemma 10.2, we see that ψ ′ (α) (x) = (ϕ−1 · ψ)(α) (x), where
ϕ : P → P is the gauge transformation given by

ϕ(σα (x)) = σα (x) · g(x).

We want ψ and ψ ′ to “describe the same physics”. From a mathematical point of view, we
want ψ and ψ ′ to satisfy the same equation. It is clear that equation dψ = 0 will not works.
The trick is to consider any connection ω on P and the gauge potential A of ω. In this case
the equation
(d − A)ψ = 0 or Dψ = 0 (11.14)
is preserved under

A → ϕ · A,
ψ → ϕ−1 · ψ.

Theorem 10.3 powa !

In this sense, we say that equation (11.14) is gauge invariant, and is thus taken by physicists
to build some theories when they need a “local G-covariance”. This gives rise to the famous
Yang-Mills theories.
In this picture the matter field ψ and the bosonic field A are both defined from a U (1)-
principal bundle. The sense of “ψ transforms as . . . under a U (1) transformation” is the sense of
the transformation of a section of an associated bunble; the sense of “A transforms as . . . under
a U (1) transformation” is the one of the transformation of the gauge potential of a connection
on a U (1)-principal bundle.
Remark 11.2. The mathematics of equation (11.14) only requires a G-valued connection on
P . There are several physical constraints on the choice of the connection. These give rise to
interaction terms between the gauge bosons. We will not discuss it at all. This a matter of
books about quantum field theories.
The most used Yang-Mills groups in physics are U (1) for the QED, SU (2) for the weak
interactions and SU (3) for the chromodynamics.

11.4 The electromagnetic field F

Now, we are able to interpret the field F introduced in equation (11.4). We follow [2]. From
now, we use the usual Minkowsky metric g = diag(−, +, +, +). From the vector given by (11.6),
we define a (local) potential 1-form

A = Aµ dxµ = −φdt + Ax dx + Ay dy + Az dz.

The field strenght is F = dA. We easily find that

F = (dt ∧ dx)(∂x φ + ∂t Ax ) + . . .
(11.15)
+ (dx ∧ dy)(−∂z Ax + ∂x Ay ) + . . .

But the fields B and E are defined from A and φ by (11.2), then

F = −Ex (dt ∧ dx) − Ey (dt ∧ dy) − Ez (dt ∧ dz)

(11.16)
+ Bx (dy ∧ dz) + By (dz ∧ dx) + Bz (dx ∧ dy).
Field theory from a bundle point of view 41

We naturally have dF = d2 A = 0. But conversely, dF = 0 ensures the existence of a 1-form

A such that F = dA. If we define7 B = ∇ × A and E = −
nablaφ − ∂t A, equations (11.1b) and (11.1c) are obviously satisfied. So in the connection for-
malism, the eqations “without sources” are written by

dF = 0. (11.17)

In order to write the two others, we introduce the current 1-form :

j = jµ dxµ = −ρdt + jx dx + jy dy + jz dz.

One can see that

δF := ⋆d ⋆ F = −dt(∇ · E)
+ dx(−∂t Ex + (∇ × B)x )
(11.18)
+ dy(−∂t Ey + (∇ × B)y )
+ dz(−∂t Ez + (∇ × B)z ),

so that equation δF = j gives equations (11.1a) and (11.1d). Now, the complete set of Maxwell’s
equations is :

dF = 0 (11.19a)
δF = j (11.19b)

with
j = −ρdt + jx dx + jy dy + jz dz, (11.20)

B=∇×A (11.21a)
E = −∇φ − ∂t A (11.21b)

where A is a 1-form such that F = dA whose existence is given by (11.19a).

12 Inclusion of the Lorentz group

Until now we had seen how to express the gauge invariance of a physical theory. In particle
physics, a really funny field theory must be invariant under the Lorentz group; it is rather clear
that, from the bundle point of view, this feature will be implemented by a Lorentz-principal
bundle and some associated bundles. A spinor will be a section of an associated bundle for
spin one half representation of the Lorentez group on C4 . In order to describe non-zero spin
particle interacting with an electomagnetic field (represented by a connection on a U (1)-principal
bubndle), we will have to build a correct SL(2, C) × U (1)-principal bundle.
A space-time is a differentiable 4-dimentional maniflod M endowed with a semi-riemannian
metric. This is a 2-form g ∈ Ω2 (M ) for which we can find at each point x ∈ M a basis
b = (b0 , . . . , b3 ) which fulfils
gx (bi , bj ) = ηij .
When we use an adapted coordinates, the metric reads g = ηij dxi ⊗ dxj .
7 i.e. we consider F as the main physical field while E and B are “derived” fields.
42 Claessens

One says that M is time orientable if one can find a vector field T ∈ X(M ) such that
gx (Tx , Tx ) > 0 for all x ∈ M . A time orientation is a choice of such a vector field. A vector
v ∈ Tx M is future directed if gx (Tx , v) > 0.
The Lorentz group L acts on the orthogonal basis of each Tx M , but you may note that L
don’t acts on M ; it’s just when the metric is flat that one can identify the whole manifold with
a tangent space and consider that L is the space-times isometry group. In the case of a curved
metric, the Lorentz group have to be introduced pointwise and the building of a frame bundle
is natural.
Now, we are mainely interested in the frame related each other by a transformation of L↑+ .
An arising question is to know if one can make a choice of some basis of each Tx M in such a
manner that
(i) pointwise, the chosen frames are related by a transformation of L↑+ ,
(ii) the choice is globally well defined.
The first point is trivial to fulfil from the definition of a space-time. The second one is more
difficult that one can guess. One can see that a good choice can be performed if and only if
there exists a vector field V ∈ X(M ) such that gx (Vx , Vx ) > 0 for all x ∈ M . We suppose that
it is the case8 .
So our first principal bundle attempt to describe the space-time symmetry is the L↑+ -principal
bundle of orthonormal oriented frame on M :
L↑+ /o /o o/ / L(M ) (12.1)
pL

M
The notion of “relativistic invariance” has to be understood in the sense of associated bundle
to this one. The next step is to recall ourself that the physical fields doesn’t transform under
representation of the group L↑+ but under representations of SL(2, C). So we build a SL(2, C)-
principal bundle
SL(2, C) /o /o /o / S(M )
pS

M
In order this bundle to “fit” as close as possible the bundle (12.1), we impose the existence of a
map λ : S(M ) → L(M ) such that
(i) pB (λ(ξ)) = pS (ξ) for all ξ ∈ S(M ) and
(ii) λ(ξ · g) = λ(ξ) · Spin(g) for all g ∈ SL(2, C).
You can recognize the definition of a spin structure. Notice that the existence of a spin
structure on a given manifold is a non trivial issue.
Now a physical field is given by a section of the associated bundle E = S(M ) ×ρ V where ρ
is a representations of SL(2, C) on V . For an electron, it is V = C4 and ρ = D(1/2,0) ⊕ D(0,1/2) .
That describes a free electron is the sense that it doesn’t interacts with a gauge field. So in
order to write down the formalism in which lives a non zero spin particle, we have to build a
U (1) × SL(2, C)-principal bundle. For this, we follow the procedure given in section 8
8 The condition is rather restrictive because we cannot, for example, find an everywhere non zero vector field

on the sphere S n with n even.

Field theory from a bundle point of view 43

13 Interactions
13.1 Spin zero
The general framework is the following :

U (1) /o /o o/ / P `A E = P ×ρ V
AA 9
AAσα ss
π AA sss
s
ss φ
 A ss
M o ? _ Uα

a U (1)-principal bundle over a manifold M (as far as topological subtleties are concerned, we
suppose M = R4 ) and a section φ of an associated bundle for a representation ρ of U (1) on
V . We consider on M the Lorentzian metric, but since we are intended to treat with scalar
(spin zero) fields, we still don’t include the Lorentz (or SL(2, C)) group in the picture. We also
consider local sections σα : Uα → P , a connection ω on P and Ω its curvature. We define
Aα = σα∗ ω.
Now we particularize ourself to the target space V = C on which we put the scalar product
1
hz1 , z2 i = (z1 z 2 + z2 z 1 ),
2
and the representation ρn : U (1) → GL(C),

ρn (g)z = g · z = g n z

where we identify U (1) to the unit circle in C in order to compute the product. A property of
this product is to be an isometry : for all g ∈ U (1), z1 , z2 ∈ C,

hρn (g)z1 , ρn (g)z2 i = hz1 , z2 i.

Our first aim is to write the covariant derivative of φ with respect to the connection ω. For this
we work on the section φ under the form φα : M → V and we use formula (9.7) :


(DX φ)(α) (x) = Xx φ(α) − ρ∗ (σα∗ ω)x Xx φ(α) (x). (13.1)
σ ω
Let us study this formula. We know that (σα∗ ω)x = Aα (x) : Tx Uα → Tσα (x) P → u(1). Thus
Aα (x)Xx is given by a path in U (1); it is this path which is taken by ρ∗ . Then (we forget some
dependances in x ∈ M )

dh
i
ρ∗ Aα (x)Xx φ(α) (x) = ρn (Aα X)(t) φ(α) (x)
dt t=0
dh n
i
= (Aα X)(t) φ(α) (x)
dt t=0 (13.2)
dh i
=n (Aα X)(t) φ(α) (x)
dt t=0
= nAα (X)φ(α) (x).

Thus the covariant derivative is given by

(DX φ)(α) (x) = Xx φ(α) − nAα (x)(Xx )φ(α) (x). (13.3)

44 Claessens

One can guess an electromagnetic coupling for a particle of electric charge n. If this reveals
to be physically relevant, it shows that the “electromagnetic identity card” of a particle is given
by a representation of U (1). It is a remarkable piece of quantum field theory : the properties of
a particle are encoded in representations of some symmetry groups.
Now we are going to prove that kDφk2 is a gauche invariant quantity. The first step is to
give a sense to this norm. We consider Xi (i = 0, 1, 2, 3), an orthonormal basis of Tx M and we
naturally denote Di = DXi , ∂i = Xi and Aαi = Aα (∂i ). Remark that

Aα (x)Xx = (σα∗ )x Xx
= ω(dσα Xx ) (13.4)
dh i
=ω σα (X(t)) ∈ u(1),
dt t=0

then this is given by a path in U (1) which can be taken by ρ. Let c(t) be this path, then
d h ic(t) i
Aα φ(α) (x) = e φ(α) (x) ,
dt t=0

so that under the conjugaison, Aα φ(α) (x) = −Aα φ(α) (x). Now our definition of kDφk2 is a
composition of the norm on V and the one on Tx M :

kDφk2 = η ij hDi φ(α) , Dj φ(α) i (13.5)

Using the notation in which the upper indices are contractions with η ij , we have
  
kDφk2 = (∂i φ(α) )(x) − nAαi φ(α) (x) (∂ i φ(α) )(x) + nAiα φ(α) (x) .

Gauge transformation law

A gauge transformation ϕ is given by an equivariant function ϕ̃α : Uα → U (1) which can be
written under the form
ϕ̃α (x) = eiΛ(x)
for a certain function Λ : Uα → R. Then from the general formula (ii) of lemma 10.2,

(ϕ · φ)(α) (x) = ρn (e−iΛ(x) )φ(α) (x)

(13.6)
= e−niΛ(x) φ(α) (x).

The transformation of the gauche field A is given by equation (10.3). Let us see what means the
term dϕ̃. For v ∈ Tx Uα ,
dh i
(dϕ̃α )x v = ϕ̃α (v(t))
dt t=0
d h iΛ(v(t)) i
= e
dt t=0 (13.7)
d h i
iΛ(v(0))
=i Λ(v(t)) e
dt t=0
= i(dΛ)x veiΛ(x) .

Thus ϕ̃−1
α (x)(dϕ̃α )x = i(dΛ)x . Since U (1) is abelian, ϕ̃
−1
Aϕ̃ = A. Finally,

(ϕ · A)α (x) = Aα (x) + i(dΛ)x . (13.8)

Field theory from a bundle point of view 45

Now we are able to prove the invariance of kDφk2 . First,

(ϕ · A)iα (x) = (ϕ · A)α (∂i )

(13.9)
= Aiα (x) + i(∂i Λ)(x);

second,  
∂i e−niΛ(x) φ(α) (x) = −ni(∂i Λ)(x)φ(α) (x) + e−inΛ(x) (∂i φ(α) )(x). (13.10)

With these two results,

∂i (ϕ · φ)(α) (x) + n(ϕ · A)αi (ϕ · φ)(α) (x) = e−inΛ(x) (nAαi (x) + ∂i φ(α) (x)). (13.11)

The Yang-Mills field strength is given by F (α) = σα∗ Ω (cf. page 25). Since U (1) is abelian,
dF (α) = 0, so that the second pair of Maxwell’s equations is complete without any lagrangian
assumptions.
The full Yang-Mills action is written as
Z  
1 ij 1 2 1
S(ω, φ) = − F (α)ij F (α) + kDφk + mφ(α) φ(α) .
4 2 2
M

The Euler-Lagrange equations are

(∂i − inAαi )(∂ i − inAiα )φα + m2 φα = 0 (13.12a)

ij
∂i F (α) = 0. (13.12b)

So the Yang-Mills lagrangian only gives the first pair of Maxwell’s equations while the second
one is given by the geometric nature of fields.

13.2 Non zero spin formalism

The formalism for a non zero spin particle in an electromagnetic field is described in section 8.
We consider the spinor bundle
SL(2, C) /o /o /o / S(M )
pS

M
with the spinor connection on S(M ), and ρ1 , a representation of SL(2, C) on V . For an electron,
it is V = C4 and ρ1 = D(1/2,0) ⊕ D(0,1/2) , so for g1 ∈ SL(2, C),
    z 
z1 g1 1
   ..  .
ρ1 (g1 )  ...  =  . (13.13)
t −1
z4 (g1 ) z4

On the other hand, we consider the principal bundle

U (1) /o /o o/ / P
pU

M
46 Claessens

with a connection ω2 which describe the electromagnetic field. As representation ρ2 : U (1) →

GL(C4 ) we choose the coordinatewise multiplication :
   
z1 g2 z1
   
ρ2 (g2 )  ...  =  ...  . (13.14)
z4 g2 z4

The physical picture of the electron is now the principal bundle

SL(2, C) × U (1) /o /o o/ / S(M ) ◦ P

p

M,

and the field is a section of the associated bundle (S(M ) ◦ P ) ×ρ C4 .

References
[1] C.Schoblond. Électrodynamique classique. 2003–2004. Notes de cours
d’électromagnétisme de deuxième année en physique à l’université libre de Bruxelles.
http://homepages.ulb.ac.be/˜cschomb/notes.html.

[2] G.Svetlichny. Preparation for gauge theory. 1999. Almost all you need —and wish— to
know about differential geometry (Lie groups, fibre bundles, connections, gauge transfor-
mation, spin bundle and so on) in order to understand the gauge theories of the mathe-
matical physics, math-ph/9902027.

[3] L.Landau and E.Lifchitz. Physique Théorique. 1989. Théorie des champs.

[4] Gregory L. Naber. Topology, geometry, and gauge fields, volume 141 of Applied Mathe-
matical Sciences. Springer-Verlag, New York, 2000. Interactions.

[5] P.W.Michor. Topics in differential geometry. 2003. These notes are from a lecture course
Differentialgeometrie und Lie Gruppen. Corrections and complements to this book will be
posted on the internet at the URL http://www.mat.univie.ac.at/ michor/dgbook.ps.

[6] S.Kobayashi and K.Nomizu. Foundation of differential geometry, volume 1. Interscience

publishers, 1963.
 Complex geometry, Calabi-Yau manifolds
and toric geometry

Vincent Bouchard

Perimeter Institute
31 Caroline St N
Waterloo, Ontario
Canada
N2L 2Y5

E-mail: [email protected]

Abstract. These are lecture notes supplementing a four-hour introductory course on

complex geometry, Calabi-Yau manifolds and toric geometry, given at the First Modave
Summer School in Mathematical Physics. We first define basic concepts of complex
geometry, leading to an analysis of the different definitions of Calabi-Yau manifolds.
The last section provides a short introduction to toric geometry, aimed at constructing
Calabi-Yau manifolds in two different ways in toric geometry; as hypersurfaces in toric
varieties and as local toric Calabi-Yau threefolds.

Based on the lectures presented by V. Bouchard at the First Modave Summer School in
Mathematical Physics held at Modave, Belgium on June 19-25, 2005
48 Bouchard

General remark

In these lectures I assume a basic knowledge of differential geometry and vector bundles on
real manifolds, at the level of Laurent Classens’ lectures in the present school. If needed the
reader may want to consult for instance [1].
The two first sections use what I would call the ‘bundle’ approach to complex geometry,
following closely the treatment of [2] and the first chapters of [3]. For a more traditional approach
(from a physicist’s point of view), I would recommend the lectures on complex geometry by Philip
Candelas [4] and Nakahara’s book [1].
In the third section I define Calabi-Yau manifolds and consider in details two examples of
Calabi-Yau threefolds. A good reference on Calabi-Yau manifolds is [5].
Section 4 is almost independent of the three other sections. It provides a quick introduction
to toric geometry, focussing on constructing Calabi-Yau manifolds in toric geometry. It is clearly
not self-contained and should be seen as a complement to standard references in toric geometry
such as [6, 7, 8]. This section is mainly based on the second chapter of [9].
I do not pretend to add anything new to these standard mathematical topics in these lecture
notes; rather, the aim is to provide an opportunity for students and researchers in mathematical
physics to get a grip on these mathematical concepts without having to go through the standard
lengthy books. I however tried to keep as much mathematical rigour as possible. The drawback
is that many proofs are omitted and various interesting topics are not even discussed.
Indeed, it is obvious that a proper introduction to the subject of these lectures would neces-
sitate much more than four hours. Therefore, I had to make some choices in the topics covered;
my selection was mainly dictated by applications in physics, particularly in string theory.
Finally, if the reader is not sure about the meaning of certain concepts, I recommend doing
a search on http://wikipedia.org/. Wikipedia is a collaborative project; add your own
comments and definitions if needed!

1 Complex geometry
In this section we define complex manifolds, develop calculus on them and define a few important
topological invariants.

1.1 Complex manifolds

Let us first define complex manifolds in two different ways. For a third definition involving
principal bundles, see [2].
Definition 1.1. Let M be a real 2m-dimensional manifold and {Ui } an open covering on M . On
each open subset Ui , we define a coordinate chart to be the pair (Ui , ψi ) where ψi : U → Cm is an
homeomorphism from Ui to an open subset of Cm (that is ψi gives a set of complex coordinates
z1i , . . . , zm
i
on Ui ). We say that the data (M, {Ui , ψi }) is a complex manifold if for every non-
empty intersections Ui ∩ Uj the transition functions ψij = ψj ◦ ψi−1 : ψi (Ui ∩ Uj ) → ψj (Ui ∩ Uj )
are holomorphic as maps from Cm to itself (i.e. they depend only on the zµ but not on their
complex conjugate). m is called the complex dimension of a complex manifold.
See figure 2.2 for a pictorial description of the above definition.
Remark 1.2. In the following, we will always use latin indices for real coordinates and greek
indices for complex coordinates. Moreover, zµ̄ will be a shorthand for z̄µ̄ , where µ̄ is a normal
tensor index.
Complex geometry, Calabi-Yau manifolds and toric geometry 49

M
Ui Uj

ψi ψj
Cm Cm

Figure 2.2: Pictorial representation of a complex manifold.

Thus, roughly speaking a complex manifold is a topological space that locally looks like Cm .
By the definition above we see that a complex manifold is always a real manifold. However, the
converse is not always true. The second definition will answer the following interesting question:
when does a real manifold can be seen as a complex manifold, that is when does a real manifold
admit a complex structure?

Remark 1.3. To fix notation: let M be a real n-manifold, with tangent bundle T M and cotan-
gent bundle T ∗ M . A smooth section S of the tensor product bundle ⊗k T M ⊗l T ∗ M is called a
tensor field of type (k, l) on M , and its components in a local coordinate basis are denoted by
l ∗
Sab11,...,b k
,...,al . The set of tensor fields of type (k, l) on M is denoted by Γ(⊗ T M ⊗ T M ).
k

Let M be a real 2m-dimensional manifold. We define an almost complex structure J on M to

be a smooth tensor field J ∈ Γ(T M ⊗ T ∗ M ) on M satisfying Jab Jbc = −δac . Now let v ∈ Γ(T M )
be a vector field, with components v a in a coordinate basis, and use the fact that J is a map
on tangent spaces to define a new vector field (Jv)b = Jab v a . Since Jab Jbc = −δac , we see that
J(Jv) = −v, that is J 2 = −1. Therefore, roughly speaking J is a generalisation of the usual
multiplication by ±i in complex analysis; it gives to each tangent space Tp M to a point p in M
the structure of a complex vector space. We say that a real 2m-dimensional manifold endowed
with an almost complex structure J is an almost complex manifold.
For two vector fields v, w, define now a vector field NJ (v, w) by

NJ (v, w) = [v, w] + J[v, Jw] + J[Jv, w] − [Jv, Jw], (1.1)

where [, ] denotes the Lie bracket of vector fields.1 N is called the Nijenhuis tensor. In local
coordinates it is given by
a
Nbc = Jbd (∂d Jca − ∂c Jda ) − Jcd (∂d Jba − ∂b Jda ). (1.2)
1 The Lie bracket [v, w] of two vector fields given in a coordinate basis by v a ∂ and w b ∂ is the ‘commutator’
a b
(v a ∂a wb − wa ∂a v b )∂b .
50 Bouchard

Definition 1.4. Let M be a 2m-dimensional real manifold and J an almost complex structure
on M . If N ≡ 0, we call J a complex structure on M . A complex manifold is defined by the
data (M, J) where J is a complex structure on M .
It follows from the well-known theorem by Newlander and Nirenberg (which we will not prove
here) that the two definitions are equivalent. Namely, the theorem states that it is possible to
find local complex coordinates with holomorphic transition functions if and only if the almost
complex structure is integrable, which is equivalent to say that its Nijenhuis tensor vanishes.
Therefore, a real manifold can be considered as a complex manifold only if it admits a complex
structure J. Let us now pause to give a few examples of complex manifolds.
Example 1.5. The simplest example of a complex manifold of complex dimension m is Cm ,
which obviously admits a global coordinate chart.
Example 1.6. A very important family of complex manifolds are the complex projective spaces,
usually denoted by CPm . These are the spaces of complex lines through the origin in Cm+1 . Take
the space Cm+1 \{0} and quotient by the identification

(z0 , . . . , zm ) ∼ λ(z0 , . . . , zm ) (1.3)

where λ is any non-zero complex number. We call the zµ homogeneous coordinates on CPm . One
way to show that it is a complex manifold is to define a set of coordinate charts with holomorphic
transition functions. Since the zµ are not all zero, we can choose an open covering defined by
the open subsets
Uα = {zα 6= 0}. (1.4)
On each Uα we define coordinates ζµα = zµ /zα , which we call inhomogeneous coordinates (or
affine coordinates). On the intersection Uα ∩ Uβ we have that

zµ zµ zβ ζµβ
ζµα = = = α, (1.5)
zα zβ zα ζβ
(µ)
since both zα and zβ are non-zero on the intersection. Therefore, the transition function ψαβ :
ζµα (Uα ∩Uβ ) → ζµβ (Uα ∩Uβ ) is simply a multiplication by (ζβα )−1 , which is of course holomorphic.
The manifolds CPm are also compact, which we will not prove here.
Example 1.7. As our final examples for this section, let us consider submanifolds of the above
complex manifolds. If, for the moment, we restrict our attention to compact complex manifolds,
we see that submanifolds of Cm are not very interesting, since a theorem that we will not prove
here (see [4] for a simple proof ) states that a connected compact analytic submanifold of Cm is a
point. However, many compact complex manifolds can be constructed as submanifolds of CPm .
We saw that CPm is compact; all its closed complex submanifolds are also compact. In fact, there
is a theorem by Chow that states that all such submanifolds of CPm can be realised as the zero
locus of a finite number of homogeneous polynomial equations (which means polynomial equations
homogeneous in the homogeneous coordinates zµ ). One important example is the quintic in CP4 ,
given as the zero locus of the equation
5
X
(zµ )5 = 0. (1.6)
µ=1

This three-dimensional compact complex manifold turns out to be Calabi-Yau, probably the most
studied Calabi-Yau threefold. We will come back to this manifold in section 3.3.
Complex geometry, Calabi-Yau manifolds and toric geometry 51

A generalisation of the construction above consists in constructing complex manifolds as the

zero locus of a finite number of homogeneous polynomial equations in a product of projective
spaces, rather than a single projective space. This generalisation leads to a large number of
manifolds, many of them Calabi-Yau. These are usually called complete intersection Calabi-Yau
(or CI Calabi-Yau manifolds). We will give some examples of this construction in section 3.3.

1.2 Tensors on complex manifolds

In the previous section we defined a complex structure J, which is an endomorphism of real
tangent spaces. At a point p of M , J gives a linear map J : Tp M → Tp M . Let us now
complexify the tangent space Tp M to get Tp M ⊗ C, which is a complex vector space isomorphic
to C2m . The map Jp extends naturally to a map J : Tp M ⊗ C → Tp M ⊗ C.
(1,0) (0,1)
Since J 2 = −1, the eigenvalues of J in Tp M ⊗ C are ±i. Now let Tp M (Tp M )
denote the eigenspace of Jp with eigenvalue i (−i). Both eigenspaces are isomorphic to Cm ,
(1,0) (0,1)
complex conjugate to each other, and we have the decomposition Tp M ⊗ C = Tp M ⊕ Tp M .
Since this works at any point p of M , it can be extended to the full bundle T M , and we
obtain that the complexified tangent bundle (which we will now denote by TC M ) decomposes
as TC M = T (1,0) M ⊕ T (0,1) M . We call T (1,0) M (T (0,1) M ) the holomorphic (anti-holomorphic)
tangent bundle.
Remark 1.8. As noted above to obtain this decomposition we must complexify the tangent
bundle, that is we consider complex-valued tensors on complex manifolds. It is important to note
the difference between having a complex manifold and complexifying the vector bundles, which
are not the same thing. We could consider real vector bundles on complex manifolds, which lead
to real-valued tensors on complex manifolds, or complex vector bundles on real manifolds, which
give complex-valued tensors on real manifolds.
The decomposition into holomorphic and anti-holomorphic pieces carries through to com-
plexified cotangent bundles as well, which gives that TC∗ M = T ∗(1,0) M ⊕ T ∗(0,1) M . Now, since
a tensor field is defined as a section of tensor products of tangent and cotangent bundles, we
expect to find a similar decompositions for complex-valued tensor fields on complex manifolds.
Remark 1.9. We remarked earlier that zµ̄ was a shorthand for z̄µ . Similarly, for general
tensors, the indices ᾱ, β̄, µ̄, . . . are tensor indices, but they indicate a modification to the tensor
itself.
We define the tensors S... α...
= 21 (S...
a... a m...
−iJm ᾱ...
S... ) and S... = 12 (S...
a... a m...
+iJm S... ), and similarly
... 1 ... m ... ... 1 ... m ... a... α... ᾱ...
Tβ... = 2 (Tb... − iJb Tm... ) and Tβ̄... = 2 (Tb... + iJb Tm... ). It is clear that S... = S... + S...
... ... ...
and Tb... = Tβ... + Tβ̄... . Therefore, these operations on tensors are projections: the ‘α’ operation
projects on the holomorphic piece, while the ‘ᾱ’ operation projects on the anti-holomorphic
piece. Moreover, it is easy to show that in this notation Jba = iδβα − iδβ̄ᾱ , which means that J
act on tensor indices α, β, . . . (ᾱ, β̄, . . .) by multiplication by i (−i). Thus, tensors (at a point p)
with indices α, β, . . . lie in a tensor product of the holomorphic tangent and cotangent spaces,
while tensors with indices ᾱ, β̄, . . . lie in a tensor product of the anti-holomorphic tangent and
cotangent bundles. In particular, a vector v α is called a holomorphic vector, while a vector v ᾱ
is called a anti-holomorphic vector, and we have the decomposition v a = v α + v ᾱ .

1.3 Exterior forms on complex manifolds

Definition 1.10. Let M be a real n-dimensional manifold. The k-th exterior (or wedge) power
of the cotangent bundle T ∗ M (defined by the exterior product of its sections below) is written
52 Bouchard

Λk T ∗ M . Smooth section of Λk T ∗ M are called k-forms, and the vector space of k-forms is
denoted by Γ(Λk T ∗ M ) or Ωk (M ).
Thus r-forms are totally antisymmetric tensor fields of type (0, r). We can define the exterior
product (or wedge product) ∧ and the exterior derivative d of k-forms as follows. Let α be a
k-form and β be a l-form, then α ∧ β is a (k + l)-form and dα is a (k + 1)-form defined by (in
component notation)

(α ∧ β)a1 ...ak+l = α[a1 ...ak βak+1 ...ak+l ] , (1.7)

(dα)a1 ...ak = ∂[a1 αa2 ...ak+1 ] , (1.8)

where [. . .] denotes complete antisymmetrisation.

It follows that

d(dα) = 0, α ∧ β = (−1)kl β ∧ α, d(α ∧ β) = (dα) ∧ β + (−1)k α ∧ (dβ). (1.9)

The first property is usually denoted by d2 = 0. We say that a r-form α is closed if it satisfies
dα = 0, and exact if it can be written as α = dβ for some (r − 1)-form. Since d2 = 0, any exact
form is closed.
Now we want to study exterior forms on complex manifolds. Using the decomposition of
the complexified cotangent bundle, it is easy to show that the following complexified bundles
decompose as
Mk
Λk TC∗ M = Λj,k−j M, (1.10)
j=0

where we defined Λp,q M := Λp T ∗(1,0) M ⊗Λq T ∗(0,1) M . A section of Λp,q M is called a (p, q)-form,
which is a complex-valued differential form with p holomorphic pieces and q anti-holomorphic
pieces. We denote the vector space of (p, q)-forms by Γ(Λp,q M ) or Ωp,q (M ), and the vector space
of complexified k-forms by Γ(Λk TC∗ M ) or ΩkC (M ).
On a complex manifold, the exterior derivative also admits a simple decomposition: d = ∂+ ∂, ¯
p,q
where we defined the operators ∂ : Ω (M ) → Ω p+1,q ¯ p,q
(M ) and ∂ : Ω (M ) → Ω p,q+1
(M ). The
identity d2 = 0 implies that ∂ 2 = ∂¯2 = 0 and ∂ ∂¯ + ∂∂
¯ = 0. We can also define a real operator
dc : ΩkC (M ) → Ωk+1 c ¯
C (M ) by d = i(∂ − ∂), which satisfies

1 1
ddc + dc d = 0, (dc )2 = 0, ∂= (d + idc ), ∂¯ = (d − idc ), ¯
ddc = 2i∂ ∂. (1.11)
2 2

1.4 Cohomology
From a mathematical perspective, topological invariants are of prime importance in the study
of manifolds. From a physics point of view, many physical questions can be reformulated as
questions about topological invariants of manifolds, for example in compactifications of string
theory and in supersymmetry.
There are various topological invariants that one can define on a manifold; cohomology
theories and characteristic classes (for instance Chern classes) provide such invariants. We will
start by defining cohomology groups. Cohomology groups have many applications in physics;
for instance they give the massless spectrum in the low energy four-dimensional theory coming
from compactifications of heterotic strings on Calabi-Yau threefolds.
Definition 1.11. Let A0 , A1 , . . . be abelian groups connected by homomorphisms dn : An →
An+1 , such that the composition of two consecutive maps is zero: dn+1 ◦ dn = 0 for all n. We
Complex geometry, Calabi-Yau manifolds and toric geometry 53

can then form the cochain complex

d0 d1 d2 d3
(1.12)
0 → A0 → A1 → A2 → ...

The cohomology groups H k are defined by

Ker(dk : Ak → Ak+1 )
Hk = . (1.13)
Im(dk−1 : Ak−1 → Ak )
With this general definition of cohomology groups, defining a particular cohomology boils
down to finding a collection of abelian groups and homomorphisms such that the above definition
holds.
Definition 1.12. Let M be a real n-dimensional manifold. As d2 = 0, we can form the complex

d d d d d
(1.14)
0 → Ω0 (M ) → Ω1 (M ) → ... → Ωn (M ) → 0.
k
We define the de Rham cohomology groups HdR (M, R) of M by

k Ker(d : Ωk (M ) → Ωk+1 (M ))
HdR (M, R) = . (1.15)
Im(d : Ωk−1 (M ) → Ωk (M ))
k
In other words, HdR (M, R) is the set of closed k-forms where two forms are considered
equivalent if they differ by an exact form, i.e. ω ≃ ω + dα; it is the quotient of the vector space
of closed k-forms on M by the vector space of exact k-forms on M . That is, given a closed
k
k-form ω, its cohomology class [ω] ∈ HdR (M, R) is the set of closed k-form which differ from ω
by an exact form. We call ω a representative of [ω].
We can see in the definition of de Rham cohomology groups that the complex terminates.
This is because there are no antisymmetric (n + 1)-tensors on an n-manifold.
k
Remark 1.13. The R in HdR (M, R) means that the closed k-forms are real, i.e. k ∈ Ωk (M ).
However, we can also define the de Rham cohomology for complexified k-forms, that is k ∈
ΩkC (M ), which we denote by HdR k
(M, C). It is also possible to define the integral cohomology
H k (M, Z), but it will not be needed in these lectures.
Using cohomology groups we can define some important topological invariants. We define
the Betti numbers bk = dimR HdR
k
(M, R). We can also define the Euler class as the alternating
sum of the Betti numbers:
n
X
χ= (−1)k bk . (1.16)
k=0
Now what is the analog of the de Rham cohomology groups for complex manifolds?
Definition 1.14. Let M be a complex manifold of complex dimension m. As ∂¯2 = 0, we can
form the complex
∂¯ ∂¯ ∂¯ ∂¯ ∂¯
p,0 p,1 (1.17)
0 → Ω (M ) → Ω (M ) → ... → Ωp,m (M ) → 0.

We define the Dolbeault cohomology groups H∂p,q

¯ (M ) of M by

Ker(∂¯ : Ωp,q (M ) → Ωp,q+1 (M ))

H∂p,q
¯ (M ) = . (1.18)
Im(∂¯ : Ωp,q−1 (M ) → Ωp,q (M ))
54 Bouchard

The Dolbeault cohomology groups depend on the complex structure of M . Note that we could
¯ this is just a matter of convention
have defined the cohomology groups using ∂ instead of ∂,
since they are complex conjugate.
We now define the Hodge numbers to be hp,q = dim H∂p,q
¯ (M ). The Hodge numbers of a
complex manifold are summarised in what is commonly called the Hodge diamond:
hm,m
..
hm,m−1 . hm−1,m
m,0
h ··· ··· h0,m (1.19)
..
h1,0 . h1,0
0,0
h
The (m + 1)2 Hodge numbers are not independent; there are many relations between them,
depending on the kind of complex manifold you look at. We will investigate some of them in
sections 2 and 3.
There are many other cohomology groups one can define on a manifold, for instance the well-
known Čech cohomology groups; they are all isomorphic on smooth manifolds. For the purpose
of these lectures the de Rham and Dolbeault cohomology groups will suffice.

1.5 Chern classes

Given a fibre F , a structure group G2 and a base space M , we may construct many fibre bundles
over M , depending on the choice of transition functions. It would be interesting to classify these
bundles and to see how much they differ from the trivial bundle M × F . In fact, characteristic
classes are what we are looking for; they are subsets of cohomology classes of the base space
which measure the non-triviality or twisting of the bundle. In other words, they are obstructions
which prevent a bundle from being a trivial bundle.
In this section we focus on a particular kind of characteristic classes, namely Chern classes,
using their differential geometry definition (for a purely topological definition see for instance [3]).
Much more than what we present here could be said about characteristic classes. For instance,
we could have discussed in more details the Chern character and the Todd classes, which have
important applications in physics; in particular while computing the index of some physically
relevant operators using the Atiyah-Singer index theorem. But four hours is not enough to enter
into this subject; more information on these topics can be found for instance in [1, 3, 10, 11].
For the sake of brevity we will not give here the general definition of characteristic classes
in terms of invariants polynomials and cohomology classes; the reader is referred to chapter 11
of [1]. It may be an interesting exercise to show that Chern classes indeed are characteristic
classes according to the general definition.
Definition 1.15. Let E be a complex vector bundle over a manifold M , and let F = dA + A ∧ A
be the curvature two-form of a connection A on E. We define the total Chern class c(E) of E
by
i
c(E) = det(1 + F ). (1.20)
2π
Since F is a two-form, c(E) is a direct sum of forms of even degrees. We define the Chern classes
ck (E) ∈ H 2k (M, R) by the expansion of c(E):
c(E) = 1 + c1 (E) + c2 (E) + . . . . (1.21)
2 The structure group G of a fibre bundle E is a Lie group that act on the left on the fibres F . Roughly

speaking, the transition functions on overlapping coordinate charts take values in the structure group G.
Complex geometry, Calabi-Yau manifolds and toric geometry 55

Remark 1.16. To be more precise, the ck (E) in the expansion are called the Chern forms,
which are closed (2k)-forms, while the Chern classes are defined as the cohomology classes of
the Chern forms. Therefore, Chern classes are cohomology classes, while the Chern forms are
representatives of the Chern classes.
This definition relies on a connection A on the bundle, so one may think that Chern classes
depend on a choice of connection A. Fortunately this is not the case. For different curvatures
F and F ′ , the difference between the two invariant polynomials is an exact form, that is the
two invariant polynomials are in the same cohomology class. Since the cohomology classes
defined by invariant polynomials form what is called characteristic classes, in the present case
Chern classes, it follows that Chern classes are independent of the choice of connection, but
that different connections will lead to different representatives of the cohomology classes ck . A
complete proof of this fact is given in chapter 11 of [1].
Since F is a two-form, on a n-dimensional manifold the Chern class cj (E) with 2j > n
vanishes identically. Also, irrespective of the dimensions of M , the series terminates at ck (E) =
det(iF/2π) and cj (E) = 0 for j > k. Therefore cj (E) = 0 for j > k where k is the rank of the
bundle E.
Remark 1.17. When the complex bundle E is the holomorphic tangent bundle T (1,0) M , we say
that ck (E) is the Chern class of the manifold M and usually denote it simply by ck .
It is useful for computations to have explicit formulae for the Chern classes. One can show
that (see [1]):

c0 (E) = [1],
 
i
c1 (E) = Tr F ,
2π
"   #
2
1 i
c2 (E) = (Tr F ∧ Tr F − Tr(F ∧ F )) ,
2 2π
..
. " #
k
i
ck (E) = det F , (1.22)
2π

where [. . .] denotes the cohomology class.

Chern classes also have a few important properties that are useful in practical calculations.
Let V be the direct sum bundle E ⊕ F . Then the total Chern class of V is c(V ) = c(E) ∧ c(F ),
which follows from properties of the determinant. In fact, c(V ) = c(E) ∧ c(F ) whenever 0 →
E → V → F → 0 is a short exact sequence.3
We now give an alternative description of Chern classes that is useful in understanding their
topological meaning, using cycles.4 Let E be a rank r complex vector bundle on a manifold M
of complex dimension m. Let s1 , . . . , sr be r global sections of E (not necessarily holomorphic).
Define Dk to be the locus of points where the first k sections develop a linear dependence, that
is where s1 ∧ . . . ∧ sk = 0 as a section of Λk E. Then the cycles Dk are Poincaré dual to the Chern
3A α
1 2 α 3 α
sequence of spaces and maps X1 −−→ X2 −−→ X3 −−→ . . . such that Im(αk ) = Ker(αk+1 ) is called exact.
α β
In particular, 0 → A −
→ B means that α is 1 − 1; B −
→ C → 0 means that β is onto. Therefore, the short exact
α β
sequence 0 → A − →B− → C → 0 means that A ⊆ B and that C = B/A. B is called an extension of A by C.
4 This description comes from the duality between homology and cohomology. However, we will not discuss

homology in these lectures; see for instance [3, 4] for more information on that topic.
56 Bouchard

classes cr+1−k (see section 3.3 for a definition of Poincaré duality, and [3] for a more general
discussion). Thus we can use the cycles Dk to understand the topological meaning of the Chern
classes cr+1−k . For instance, the Chern class c1 (E) corresponds to the cycle Dr , which is defined
by s1 ∧ . . . ∧ sr = 0. This represents the zeroes of the sections of the determinant line bundle
Λr E (see section 1.6); therefore c1 (E) = 0 is the same thing as Λr E being trivial. Indeed,
c1 (E) = c1 (Λr E). For k = 1, we find that the top Chern class cr (E) is represented by the zeros
of a single section of E. In particular, if E = T (1,0) M , then cm (M ) represents the zeros of a
generic section of the holomorphic tangent bundle; this is what is called the Euler characteristic
of M , and the Euler class is given by the integral of the Euler characteristic over M , that is by
the integral of the top Chern class of M :
Z
χ = cm (M ). (1.23)
M

Chern character
To end this section we rapidly define the Chern character, which will
Qr be useful in computations
of Chern classes later on. Suppose that we define xi by c(E) = i=1 (1 + xi ), Pwhere r is the
rank of the bundle E. Then the Chern character ch(E) is defined by ch(E) = i exi . The first
few terms in the expansion of the exponential are
1 1
ch(E) = r + c1 (E) + (c1 (E)2 − 2c2 (E)) + (c1 (E)3 − 3c1 (E)c2 (E) + 3c3 (E)) + . . . . (1.24)
2 6
Note that the Chern character satisfies the useful identities ch(E ⊕ F ) = ch(E) + ch(F ) and
ch(E ⊗ F ) = ch(E)ch(F ).

1.6 Holomorphic vector bundles

So far we have used complex vector bundles over complex manifolds. We will now formally define
holomorphic vector bundles over complex manifolds.

Definition 1.18. Let M be a complex manifold. Let {Ep } : p ∈ M be a family of complex

vector spaces of dimension k, parameterised by M . Let E be the total space of this family, and
π : E → M be the natural projection. Suppose also that E has the structure of a complex
manifold. E with its complex structure is called a holomorphic vector bundle with fibre Ck if
the map π is a holomorphic map of complex manifolds5 and for each p there exists an open
neighbourhood U ⊂ M and a biholomorphic map φU : π −1 (U ) → U × Ck such that for each
u ∈ U the map φU takes Eu to {u} × Ck , and this is an isomorphism between Eu and Ck as
complex vector spaces. k is called the rank of the bundle; it is the dimension of its fibers.

See figure 2.3 for a pictorial representation of the above definition of holomorphic vector
bundles.
Basically, the important points in the definition are that the total space E has the structure
of a complex manifold, the projection map π is a holomorphic map of complex manifolds, and
5 Let f : M → N , and M and N be complex manifolds with complex dimensions m and n. Take a point p

in a chart (U, φ) of M . Let (V, ψ) be a chart of N such that f (p) ∈ V . Let {zµ } = φ(p) and {wν } = ψ(f (p))
be their coordinates in Cm and Cn . We thus have a map ψ ◦ f ◦ φ−1 : Cm → Cn . If each function wν is a
holomorphic function of zµ , we say that f is a holomorphic map. A map f is called biholomorphic if an inverse
map f −1 : N → M exists and both f and f −1 are holomorphic maps. See figure 2.4 for a pictorial description
of holomorphic maps of complex manifolds.
Complex geometry, Calabi-Yau manifolds and toric geometry 57

φU Ck
E
×U

π
M p
U

Figure 2.3: Pictorial representation of a holomorphic vector bundle.

M N
f
U V

φ ψ
Cm Cn

Figure 2.4: Pictorial representation of a holomorphic map of complex manifolds.

58 Bouchard

the trivialisation φU is a biholormophism. This is not always obvious, and not all complex vector
bundles are holomorphic vector bundles.
Let us now give a few examples of holomorphic vector bundles. The simplest holomorphic
vector bundle is M × Ck , which is called the trivial vector bundle over M . The complexi-
fied tangent and cotangent bundles admit natural complex structures, which make them into
holomorphic vector bundles.
Now what about the complex vector bundles Λp,q M ? It turns out that, although these
bundles all have complex vector spaces as fibers, only the bundles with q = 0 are holomorphic
vector bundles. A holomorphic section of Λp,0 M – i.e. a section s for which ∂s ¯ = 0 – is called a
holomorphic p-form.
If the fibre of a holomorphic vector bundle is C, i.e. its rank is one, then we say that it is
a holomorphic line bundle. As an example, the bundle KM = Λm,0 M on a complex manifold
M of complex dimension m, sections of which are (m, 0)-forms, is a holomorphic line bundle,
usually called the canonical bundle. Its sections are sometimes called holomorphic volume forms
on M . Actually, given any holomorphic vector bundle E of rank r, we can form the holomorphic
line bundle Λr E, the determinant line bundle, whose transition functions are the determinants
of those for E.
Given two line bundle L and L′ over M , one can construct many other line bundles. First,
there is the dual line bundle L∗ to L. But also, any tensor product of line bundles is also a line
bundle, so L ⊗ L′ forms a new line bundle. This is so because if we look at the fibers, they are
one-dimensional vector spaces; but we have that dim(U ⊗ V ) = dim U dim V for vector spaces
U and V , so we see that the fibers of L ⊗ L′ also have dimension one, therefore L ⊗ L′ is a
line bundle. In fact, the set of isomorphism classes of holomorphic line bundles over M form an
abelian group, where multiplication is given by the tensor product, inverses are dual bundles,
and the identity is the trivial line bundle L ⊗ L∗ . This group is called the Picard group of M .
Now let us consider the special case where M is CPm . First, there is the natural line bundle
whose fiber over a point l in CPm is the line it represents in Cm+1 ; this is the tautological line
bundle (also called the universal line bundle), which we denote L−1 . Its dual, which we denote
L, is called the hyperplane line bundle. In fact, since the tensor product of two holomorphic
line bundles is always a holomorphic line bundle, we can construct holomorphic line bundles Lk
over CPm for any k ∈ Z. Actually, it can even be shown that every holomorphic line bundle
over CPm is isomorphic to Lk for some k ∈ Z. For instance, the canonical bundle KCPm is
isomorphic to L−m−1 . By abuse of notation, we usually denote the line bundle Lk by O(k),
although technically O(k) denotes the sheaf of holomorphic section of Lk . It is interesting to
note that for k ≥ 0 the vector space of holomorphic sections Γ(Lk ) is canonically identified with
the set of homogeneous polynomials of degree k in CPm . Therefore, homogeneous coordinates
of CPm are sections of the hyperplane line bundle L.
Finally, we can use these results to refine our definition of the cohomology groups on complex
manifolds. First, we must generalise our definition of a k-form. We defined k-forms as sections
of the bundle Λk M , that is they take values in M . We can now define a E-valued k-form ω,
that is a form that takes values in the vector bundle E, by the map ω : Γ(E) → Γ(E) ⊗ Ωk (M ).
Thus ω is a section of E ⊗ Λk M . For instance, we could define a Λp,0 M -valued (0, q)-form
ω : Γ(Λp,0 M ) → Γ(Λp,0 M ) ⊗ Γ(Λ0,q M ); hence ω is a section of Λp,q M , that is ω is simply what
we previously defined as a (p, q)-form. But using this generalised definition we can consider
forms that take values in complex vector bundles E different from Λp,0 M .
We defined the (p, q) Dolbeault cohomology group as the quotient of the vector space of ∂- ¯
¯ ¯
closed (p, q)-forms by the vector space of ∂-exact (p, q)-forms. However, the operator ∂ only acts
on the ‘anti-holomorphic’ piece of a (p, q)-form, i.e. on the (0, 1) part of the decomposition of the
Λp,q M bundle. Therefore, we can look at the elements of the Dolbeault cohomology groups in
Complex geometry, Calabi-Yau manifolds and toric geometry 59

¯
the following way; they are ∂-closed (0, q)-forms taking values in the holomorphic vector bundle
Λ M . From that point of view, we can define the cohomology groups H∂q¯(Λp,0 M ), where Λp,0 M
p,0

is a holomorphic vector bundle, sections of which are (p, 0)-forms. These cohomology groups
are indeed isomorphic to the Dolbeault cohomology groups previously defined. However, they
can be generalised; since ∂¯ commutes with holomorphic transition functions, ∂¯ may now act on
forms taking values in any holomorphic vector bundle E, not just the holomorphic vector bundle
Λp,0 M . This leads to the following generalised definition of Dolbeault cohomology.

Definition 1.19. Let M be a complex manifold, and E a holomorphic vector bundle on M .

We have the complex
∂¯ ∂¯ ∂¯ ∂¯
→ Γ(E) ⊗ Ω0,1 (M ) −
0 → Γ(E) − → Γ(E) ⊗ Ω0,m (M ) −
→ ... − → 0. (1.25)

We define the Dolbeault cohomology groups taking values in E, H∂q¯(E), by

Ker(∂¯ : Γ(E) ⊗ Ω0,q (M ) → Γ(E) ⊗ Ω0,q+1 (M ))

H∂q¯(E) = . (1.26)
Im(∂¯ : Γ(E) ⊗ Ω0,q−1 (M ) → Γ(E) ⊗ Ω0,q (M ))

This definition reduces to the former definition of Dolbeault cohomology groups when E =
Λp,0 M .

1.7 Divisors and line bundles

To end this section, we explore the connection between line bundles and divisors, which are
important objects in algebraic geometry and in its physical applications. In practice, it is
convenient to be able to think of divisors in terms of line bundles and vice-versa. For more on
that topic see [11].

Definition 1.20. Let M be a complex manifold. Let Ni be a hypersurface, that is a codimension

1 submanifold that can be written locally as the zeroes of a holomorphic function. Moreover, let
Ni be an irreducible hypersurface, that is it cannot be written as the union of two hypersurfaces.
P
Then we define a divisor to be the formal finite sum of irreducible hypersurfaces D = i ni Ni
with integer coefficients ni . The divisor is called effective if ni ≥ 0 for all i.

Let L be a holomorphic line bundle over M , and s a nonzero holomorphic section of L. Let
N be the hypersurface defined by N = {m ∈ M : s(m) = 0}. Then N may be written in a
unique way as a union N = ∪i Ni , where Ni are irreducible hypersurfaces. For each i, there
is a unique
P positive integer ai which tells us the order to which s vanishes along Ni . Define
D = i ai Ni . Then D is an effective divisor. In fact, since the divisors constructed that way
from line bundles are equal if and only if the line bundles are isomorphic, there is a 1 − 1
correspondence between effective divisors on M and isomorphism classes of holomorphic line
bundles equipped with nonzero holomorphic sections.
If we nowPconsider holomorphic line bundles with a nonzero meromorphic6 section, then the
divisor D = i ai Ni corresponds to a section s with a zero of order ai along Ni if ai > 0, and
a pole of order −ai along Ni if ai < 0. Thus there is a 1 − 1 correspondence between divisors
on M and isomorphism classes of holomorphic line bundles equipped with nonzero meromorphic
sections.
6 A meromorphic function is a function that is holomorphic on an open subset of the complex number plane

C except at points in a set of isolated poles, which are certain well-behaved singularities. Every meromorphic
function can be expressed as the ratio between two holomorphic functions (with the denominator not constant
0): the poles then occur at the zeroes of the denominator.
60 Bouchard

2 Kähler geometry
We will now consider a special type of complex manifolds, namely Kähler manifolds. Roughly
speaking, manifolds with a Kähler metric are those for which the parallel transport of a holomor-
phic vector remains holomorphic. Kähler manifolds are very important in physics, for instance
such manifolds admit the N = 2 supersymmetric sigma models which is crucial in string theory.

2.1 Kähler manifolds

Definition 2.1. Let (M, J) be a complex manifold, and let g be a Riemannian metric on M .
We call g a Hermitian metric if the three following equivalent conditions hold:

(i) g(v, w) = g(Jv, Jw) for all vector fields v, w on M ;

(ii) In component notation, gab = Jac Jbd gcd ;

(iii) Using the greek indices notation, gab = gαβ̄ + gᾱβ , that is gαβ = gᾱβ̄ = 0.

In other words, a Hermitian metric is a positive-definite inner product T (1,0) M ⊗T (0,1) M → C

at every point on a complex manifold M .

Remark 2.2. Note that this is a restriction on the metric, not on the manifold M , since it can
be shown that a complex manifold always admits a Hermitian metric.

Using this Hermitian metric g, we can define a two-form ω on M called the Hermitian form
by ω(v, w) = g(Jv, w) for all vector fields on v, w on M . The equivalent definition in terms of real
components is ωab = Jac gcb , while in terms of complex components it is given by ωab = igαβ̄ −igᾱβ .
Therefore, ω is a (1, 1)-form.
If g was not Hermitian, then ω would not be a form (that is it would not be antisymmetric).
Therefore the Hermitian condition is equivalent to the condition ωab = −ωba .

Definition 2.3. Let (M, J) be a complex manifold, and g a Hermitian metric on M , with
Hermitian form ω. g is a Kähler metric if dω = 0. In this case we call ω a Kähler form, and we
call a complex manifold (M, J) endowed with a Kähler metric a Kähler manifold.

In the remaining of this section we will explore properties of Kähler manifolds. First, it can
be shown that locally, the Kähler condition dω = 0 is equivalent to the condition ∂µ gν ᾱ = ∂ν gµᾱ
and its conjugate equation ∂ρ̄ gµᾱ = ∂ᾱ gµρ̄ . Moreover, for a Kähler metric g the Levi-Civita
connection has no mixed indices, meaning that vectors with holomorphic indices remain with
holomorphic indices after parallel transportation. This implies a restriction on the holonomy of
Kähler manifolds, as we will see in section 2.4.
An important consequence of Kählerity is the existence of a Kähler potential. Let φ be a
real smooth function on M . Clearly, ddc φ is a closed real two-form, as both d an dc are real
operators. But since ddc = 2i∂ ∂, ¯ ddc φ is also a closed (1, 1)-form.
In fact, any closed (1, 1)-form ω can be expressed locally as ω = ddc φ for a real smooth
function φ on M . This is however not true globally; it only holds if ω is also exact.
Therefore, it is always possible to express the Kähler form ω in terms of a smooth function φ
locally, and we call this function the Kähler potential. This also follows from the Kähler condition
in component notation given above. However this is not true globally, for the following reason.
Let M be a compact manifold of real dimension 2m, with Kähler form ω. Since ω is closed,
it defines a Dolbeault cohomology class [ω] ∈ H∂1,1 ¯ (M ), or a de Rham cohomology class [ω] ∈
2
HdR (M, R). The latter is usually called the Kähler class. Further, the wedge product of m copies
Complex geometry, Calabi-Yau manifolds and toric geometry 61

of ω, denoted by ω m , is proportional to the volume form of g (one can sees that by working
out the explicit expression of ω m using the component definition of ωab ).RTherefore it defines a
non-trivial element in both H∂m,m
¯ (M ) and HdR2m
(M, R), and we have that M ω m ∝ vol(M ). But
R
for a compact manifold M , vol(M ) > 0, and M ω m only depends on the cohomology class [ω]
(by Stoke’s theorem). Thus, [ω] must be non-zero. However, ddc φ is exact and therefore zero
as a cohomology class. It follows that on a compact Kähler manifold it is impossible to find a
globally defined Kähler potential.
However, an interesting result is that we can parameterise Kähler metrics with a fixed Kähler
class by smooth functions on the manifold. This goes as follows. Let ω and ω ′ be two different
Kähler forms in the same Kähler class. Therefore, ω − ω ′ is an exact form. But we saw that
exact forms can be expressed globally as ddc φ for a smooth function φ. Therefore, globally we
have that ω = ω ′ + ddc φ. Moreover, φ is unique up to the addition of a constant; suppose φ1 and
φ2 are two different such functions, then ddc (φ1 − φ2 ) = 0 on M , which implies that φ1 − φ2 is
constant, as M is compact. Therefore, smooth functions on M parameterises the set of Kähler
forms in a particular Kähler class.
Example 2.4. In this example we sketch the proof that CPm is aPKähler manifold, that is
m
it admits a Kähler metric. Consider the function u(z0 , . . . , zm ) = µ=0 |zµ |2 where zµ , µ =
0, . . . , m are homogeneous coordinates on Cm+1 \{0}. Define a (1, 1)-form α by α = ddc (log u).
α cannot be the Kähler form of any metric on Cm+1 \{0}, since it is not positive. However, if
we consider the projection π : Cm+1 \{0} → CPm defined by π : (z0 , . . . , zm ) 7→ [z0 , . . . , zm ], one
can show that there exists a unique positive (1, 1)-form ω on CPm such that α = π ∗ (ω). ω is
a Kähler form on CPm ; its associated Kähler metric is called the Fubini-Study metric, and is
given in components by gµν̄ = ∂µ ∂ν̄ log u.
There is a general result that says that any submanifolds of a Kähler manifold is also Kähler
(since the restriction of the Kähler form to a complex submanifold is also a closed, positive
(1, 1)-form). We just saw that CPm is Kähler; therefore all its submanifolds are also Kähler.
This important family of complex manifolds was constructed in section 1.1; we now know that
they all admit a Kähler metric.

2.2 Forms on Kähler manifolds

A complex manifold with a Kähler metric has now enough structure to define operators analog
to the Hodge star and the d† and ∆d operators on a Riemannian manifold, thus leading to a
Hodge theory on Kähler manifolds. We can also relate this theory to the real version – since
complex manifolds are also real manifolds – and find interesting properties of Kähler manifolds.
Before doing so let us summarise quickly some results for real manifolds.
Let M be a compact Riemannian n-manifold, with metric g. Let α and β be k-forms on
M . We define the pointwise inner product of α and β by (in component notation) (α, β) =
αa1 ...ak βb1 ...bk g a1 b1 . . . g ak bk . We define a second inner product, called
R the L2 inner product,
using the volume form dVg on M given by the metric, by hα, βi = M (α, β)dVg .
The Hodge star is an isomorphism of vector bundles ∗ : Λk T ∗ M → Λn−k T ∗ M such that if
β is a k-form on M , then ∗β is the unique (n − k)-form satisfying α ∧ (∗β) = (α, β)dVg for all
k-forms α on M .
We define an operator d† : Ωk (M ) → Ωk−1 (M ) by d† β = (−1)kn+n+1 ∗d(∗β). It is sometimes
called the formal adjoint of d. We have that (d† )2 = 0, so we say that a form satisfying d† α is
coclosed, and if α = d† β then it is coexact. We also define the Laplacian ∆d = dd† + d† d. If
a form satisfies ∆d α = 0, we say that it is harmonic. It can be shown easily that a form on a
compact manifold M is harmonic if and only if it is closed and coclosed.
62 Bouchard

Now, let Hk be the vector space of harmonic k-forms on M . The Hodge decomposition
theorem states that Ωk (M ) = Hk ⊕ Im(dk−1 ) ⊕ Im(d†k+1 ), that is every k-form can be expressed
uniquely as a sum of a harmonic form, an exact form and a coexact form. Moreover, we have
that Ker(dk ) = Hk ⊕ Im(dk−1 ) and Ker(d†k ) = Hk ⊕ Im(d†k+1 ), that is closed (coclosed) forms
can be expressed uniquely as a sum of a harmonic and a exact (coexact) form. Therefore,
k
since HdR (M, R) = Ker(dk )/Im(dk−1 ), we see that there is an isomorphism between Hk and
k
HdR (M, R). In other words, every de Rham cohomology class on M contains a unique harmonic
representative. However, although cohomology classes are topological invariants of M , their
harmonic representatives depend on a particular choice of a metric g.
We are now ready to see what the analogs of these constructions are on Kähler manifolds.
Let M be a complex manifold or real dimension 2m, with a Kähler metric g. Let α and β be
complex k-forms on M . First, define a pointwise inner product (in real component notation) by
(α, β) = αa1 ...ak βb1 ...bk g a1 b1 . . . g ak bk , that is (α, β) is a complex functionRon M , which is bilinear
in α and β̄. For a compact M , define the L2 inner product by hα, βi = M (α, β)dVg . hα, βi is a
complex number, bilinear in α and β̄.
Let the Hodge star on Kähler manifolds to be the isomorphism of complex vector bundles
∗ : Λk TC∗ M → Λ2m−k TC∗ M defined by its action on the sections; α ∧ (∗β) = (α, β)dVg for all
complex k-forms α and β. Define the following operators taking complex k-forms to complex
(k − 1)-forms (there is no (−1) factor as in the real analog since the real dimension of a complex
manifold is always even):

d† = − ∗ d(∗α), ∂ † = − ∗ ∂(∗α), ∂¯† = − ∗ ∂(∗α).

¯ (2.1)

We also define the usual Laplacian ∆d = dd† + d† d, and the two Laplacians ∆∂ = ∂∂ † + ∂ † ∂ and
∆∂¯ = ∂¯∂¯† + ∂¯† ∂,
¯ which satisfy ∆∂ = ∆ ¯ = 1 ∆d . Reformulating in terms of holomorphic and
∂ 2
anti-holomorphic parts, we have defined the following operators on Kähler manifolds:

∗ : Ωp,q (M ) → Ωm−p,m−q (M );
∂ : Ωp,q (M ) → Ωp+1,q (M ), ∂¯ : Ωp,q (M ) → Ωp,q+1 (M );
∂† : Ωp,q (M ) → Ωp−1,q (M ), ∂¯† : Ωp,q (M ) → Ωp,q−1 (M );
∆∂ : Ωp,q (M ) → Ωp,q (M ), ∆∂¯ : Ωp,q (M ) → Ωp,q (M ). (2.2)

¯
Remark 2.5. In the literature the ∂-Laplacian ∆∂¯ on a Kähler manifold is often simply called
the Laplacian and denoted by ∆. Similarly, we call a (p, q)-form satisfying ∆∂¯ a harmonic
¯ = ∂ † α = ∂¯† α = 0.
(p, q)-form. A (p, q)-form α is harmonic if and only if ∂α = ∂α

How can we define the Hodge theory of a Kähler manifold? In fact, we can formulate a
Hodge theory for the ∂¯ operator which is very similar to the Hodge theory for real manifolds.
Let M be a complex manifold of real dimension 2m. Let Hp,q be the vector space of harmonic
(p, q)-forms. The following decompositions hold:
   
Ωp,q (M ) = Hp,q ⊕ ∂¯ Ωp,q−1 (M ) ⊕ ∂¯† Ωp,q+1 (M ) ,
 
Ker∂¯ = Hp,q ⊕ ∂¯ Ωp,q−1 (M ) ,
 
Ker∂¯† = Hp,q ⊕ ∂¯† Ωp,q+1 (M ) . (2.3)

We see that the vector space of harmonic (p, q)-forms is isomorphic to the Dolbeault cohomology
groups, that is Hp,q ∼ p,q
= H∂¯ (M ). This means that there is a unique harmonic representative in
each Dolbeault cohomology classes.
Complex geometry, Calabi-Yau manifolds and toric geometry 63

Let us now define HCk to be the vector space
of complex harmonic k-forms (with respect to
∆d ), that is HCk = Ker ∆d : ΩkC (M ) → ΩkC (M ) . Using (1.10), and the fact that 21 ∆d = ∆∂¯, we
see that there is a further decomposition
k
M
HCk = Hj,k−j , (2.4)
j=0

of complex harmonic k-forms into a sum of harmonic (p, q)-forms with p + q = k. By the above
isomorphisms, we learn that for a Kähler manifold, the complexified de Rham cohomology
decomposes into the Dolbeault cohomology (note that the de Rham cohomology groups are
complex, as we are now considering complexified k-forms)
k
M
k
HdR (M, C) = H∂j,k−j
¯ (M ). (2.5)
j=0

This relation and others lead to various properties of cohomology groups on Kähler manifolds,
which is the topic of the next section.

2.3 Cohomology
In this section we explore the relations between cohomology groups on Kähler manifolds.
First, using (2.5) we see directly that for a Kähler manifold,
k
X
bk = hj,k−j . (2.6)
j=0

k k
Note that this is true because dimC HdR (M, C) = dimR HdR (M, R), since we previously defined
the Betti numbers for real de Rham cohomology groups. Moreover, the Hodge star on Kähler
manifolds and complex conjugation tell us that
hp,q = hq,p ,hp,q = hm−q,m−p . (2.7)
P2k−1 Pk−1
Furthermore, since by the above properties b2k−1 = j=0 hj,2k−1−j = 2 j=0 hj,2k−1−j , we
have that b2k−1 is even for 1 ≤ k ≤ m. In practice, this last condition is useful to deduce that
some complex manifolds do not admit a Kähler metric. For example, the manifold S 3 × S 1 has
a complex structure, but it does not admit a Kähler metric since b1 = 1.
It would be interesting to understand what the Hodge numbers of a Kähler manifold exactly
mean in terms of geometry. A first step towards this end consists in defining the Kähler cone.
Definition 2.6. Let (M, J) be a complex manifold admitting Kähler metrics. If g is a Kähler
metric on M , then the Kähler form ω is a closed (1, 1)-form, that is [ω] ∈ H∂1,1
¯ (M ). We define
1,1
the Kähler cone K of M to be the set of cohomology classes [ω] ∈ H∂¯ (M ) such that ω is the
Kähler form of a Kähler metric on M .
In other words, the Kähler cone defines the set of possible Kähler forms on M . Thus, in
the Kähler cone h1,1 (M ) counts the number of possible Kähler forms. Moreover, as any class
H∂1,1
¯ (M ) inside the Kähler cone can be used to deform the metric slightly while preserving
Kählerity, we can also say that h1,1 (M ) classifies infinitesimal deformations of the metric that
preserve Kählerity. We see that h1,1 (M ) is intimately related to Kähler structure deformations
on M ; we will see in section 3 that other Hodge numbers are related to complex structure
deformations on M .
64 Bouchard

2.4 Holonomy
Before we close this section, we will find what the holonomy of a Kähler manifold is. We have
not discussed holonomy of manifolds so far, but they are crucial in the definition of Calabi-Yau
manifolds. Let us first recall what the holonomy of a manifold is.
Definition 2.7. Let M be an n-dimensional Riemannian manifold with metric g and affine
connection ∇. Let p be a point in M and consider the set of closed loops at p, {c(t)|0 ≤ t ≤
1, c(0) = c(1) = p}. Take a vector X in Tp M and parallel transport along a closed curve c(t); we
end up with a new vector Xc ∈ Tp M . Thus, the loop c(t) and the connection ∇ induce a linear
transformation Pc : Tp M → Tp M . The set of all these transformations is denoted by Holp (M )
and called the holonomy group at p.
The holonomy group measures how vectors are transformed by parallel transport around a
closed curve at a point p of M . In fact, holononmy groups can be defined more generally for a
vector bundle E with a connection, not necessarily the tangent bundle with its affine connection;
the reader is referred to [2] for a detailed discussion of holonomy.
Note that Holp (M ) must be a subgroup of GL(n, R), which is the maximal holonomy group
possible. Holp (M ) is trivial if and only if the Riemann tensor vanishes.
Now, suppose that M is connected (which we always assume in these lectures), and that p
and q are two points of M connected by a curve a. The curve a defines a map τa : Tp M → Tq M
by parallel transporting a vector in Tp M to Tq M along a. Then the holonomy groups are related
by Holp (M ) = τa−1 Holq (M )τa , hence Holq (M ) is isomorphic to Holp (M ). For that reason, the
holonomy group Holp (M ) is independent of the base point p, and we usually omit the subscript
p and denote by Hol(M ) the holonomy group of a manifold M .
In particular, if M is a Riemaniann manifold and ∇ is a metric connection, then parallel
transport preserves the length of a vector, which implies that the holonomy group must be a
subgroup of SO(n) (if M is orientable).
Now what is the holonomy of a complex manifold with a Kähler metric?
Proposition 2.8. Let M be a complex manifold of real dimension 2m, with a Kähler metric g.
The holonomy group of M is contained in U (m).
(1,0)
Take a vector X ∈ Tp M in the holomorphic tangent space at a point p of M . We saw
in section 2.1 that for an affine connection ∇ corresponding to a Kähler metric, vectors with
holomorphic indices remain with holomorphic indices after parallel transport. Therefore, if
Xc is the vector resulting from parallel transporting X around a closed loop c, we must have
(1,0)
Xc ∈ Tp M . Moreover, ∇ preserves the length of a vector. This implies that the holonomy
(1,0) (1,0)
group is the set of all transformations Pc : Tp M → Tp M which preserve the length of a
vector, which is (a subgroup of) U (m).
Note that we can use proposition 2.8 as a definition of Kähler manifolds; all the above
properties of Kähler manifolds follow from it.

3 Calabi-Yau geometry
We are now ready to investigate Calabi-Yau manifolds, which are a particular kind of Kähler
manifolds.
It was in 1954 that Calabi stated his conjecture [12, 13], which was proved by Yau in 1976
[14, 15]. Given a compact Kähler manifold M with c1 = 0, the proof of the conjecture guarantees
the existence of a Ricci-flat Kähler metric on M , that is a Kähler metric with zero Ricci form.
Such a manifold is called a Calabi-Yau manifold.
Complex geometry, Calabi-Yau manifolds and toric geometry 65

However, many different definitions of Calabi-Yau manifolds exist in the literature; we will
review some of the most common definitions and study some relations among them. We will
also investigate properties of Calabi-Yau manifolds and study in details a few examples. We will
end this section by quickly describing ‘local’ Calabi-Yau manifolds (i.e. noncompact Calabi-Yau
manifolds), which have many applications for instance in topological strings.
Calabi-Yau manifolds have been studied extensively in the recent decades, particularly be-
cause of their importance in string theory. While the mathematical study of Calabi-Yau man-
ifolds has helped us understand compactifications of string theory, the study of string theory
has led to fascinating insights in the geometry of Calabi-Yau manifolds, for example the study
of the Calabi-Yau moduli space and the conjectured mirror symmetry. Calabi-Yau manifolds
are thus a very good example of the fruitful interactions between mathematics and physics that
have been taking place in the recent decades.

3.1 Calabi-Yau manifolds

Let us first list some of the most common definitions of Calabi-Yau manifolds. A Calabi-Yau
manifold of real dimension 2m is a compact Kähler manifold (M, J, g):

(i) with zero Ricci form,

(ii) with vanishing first Chern class,

(iii) with Hol(g) = SU (m) (or Hol(g) ⊆ SU (m)),

(iv) with trivial canonical bundle,

(v) that admits a globally defined and nowhere vanishing holomorphic m-form.

We now describe some of the relations between these definitions. There are many ways to
understand these relations. We only describe a few relations, but many others can be found in
the references at the end of these lecture notes.
[1-2]
First, let us show that a compact Ricci-flat Kähler manifold has c1 = 0. We saw in (1.22)
i
that the first Chern class of a complex vector bundle E over M is given by c1 (E) = 2π Tr F ,
where F is the curvature of a connection A. The first Chern class of a manifold was defined in
section 1.5 to be the first Chern class of the holomorphic tangent bundle, and in this case, the
curvature two-form F is simply −iR, where R is the Ricci two-form. Therefore, given a Kähler
manifold M , its first Chern class is given by
 
1
c1 (M ) = Tr R . (3.1)
2π

In other words, the Ricci form defines the first Chern class of a manifold. Hence it is clear
that if a Kähler manifold admits a Ricci-flat metric then it has c1 = 0. However, the converse,
namely, does a Kähler manifold with c1 = 0 admit a Ricci-flat metric, is much more complicated
to prove. It was conjectured by Calabi that the answer is yes and that the Ricci-flat metric is
unique; uniqueness was proved by Calabi, existence by Yau twenty years later. More precisely,
it was proved that given a complex manifold M with a Kähler metric g, a Kähler form ω and
c1 = 0, then there exists a unique Ricci-flat metric g ′ whose Kähler form ω ′ is in the same Kähler
class as ω. In other words, there is a unique Ricci-flat Kähler metric in each Kähler class of M .
This means that the Ricci-flat Kähler metrics on M form a smooth family of dimension h1,1 (M ),
isomorphic to the Kähler cone of M . A proof of this deep theorem is given in chapter 5 of [2].
66 Bouchard

Therefore, h1,1 (M ) counts the number of possible Ricci-flat Kähler forms on a Calabi-Yau
manifold M . These are usually called the Kähler parameters of M .
[4-5]
We saw that the canonical bundle of a complex manifold M of real dimension 2m is the
complex vector bundle KM = Λm,0 M , that is its sections are (m, 0)-forms. Triviality of this
bundles implies that the total space of KM is given by M × C (since it is a line bundle).
Therefore, corresponding to the unit section M × {1}, that is the constant function 1, there is a
globally defined and nowhere vanishing holomorphic (m, 0)-form Ω, which is usually called the
holomorphic volume form. Moreover, it is clear that any globally defined (m, 0)-form can be
written as f Ω for some function f on M . Then, if M is compact and the form is holomorphic,
then f must be holomorphic, and the extension of the maximum modulus principle of complex
analysis tells us that f is constant. Hence hm,0 = 1. On the other hand, the existence of
a globally defined and nowhere vanishing holomorphic (m, 0)-form α directly implies that the
canonical bundle is trivial.
[2-4]
The canonical bundle is the determinant line bundle of the holomorphic cotangent bundle,
i.e. it is the highest antisymmetric tensor product of the holomorphic cotangent bundle. We saw
in section 1.5 that c1 (E) = 0 is the same thing as the determinant line bundle Λk E being trivial,
where k is the rank of E. Therefore KM = Λm T ∗(1,0) M is trivial if and only if c1 (T ∗(1,0) M ) =
−c1 (T (1,0) M ) = −c1 = 0.
[1-3]
We now show that if M is a Ricci-flat Kähler manifold of real dimension 2m, then its holonomy
group is contained in SU (m). Let V = V k ∂k ∈ Tp M be a tangent vector, and parallel transport
it along an infinitesimal parallelogram of area δamn with edges that are parallel to the vectors
∂m and ∂n . It is a standard result that

V ′k = V k + δamn Rmn k l V l . (3.2)

The matrices δlk + δamn Rmn k l are the elements of the holonomy group that are infinitesimally
close to the identity. For a Kähler metric, the matrices δamn Rmn k l are in the Lie algebra of
U (m). In fact, in a neighbourhood of the identity we have that U (m) ∼ = SU (m) × U (1), where
the U (1) factor is generated by the trace δamn Rmn k k . One can easily show that this is equal to
−4δaµν̄ Rµν̄ . Therefore, if the metric is Ricci-flat, then the U (1) part of the holonomy vanishes,
and the holonomy groups must be contained in SU (m). The converse is also true; if the holonomy
of a Kähler manifold is contained in SU (m), then its Kähler metric is Ricci-flat.
In fact, we have only shown that this is true for simply connected manifolds, that is for
closed curves that can be continuously shrunk to a point. The equivalence still holds for multiply
connected manifolds, but the proof is more involved.

3.2 Cohomology
The Hodge numbers of a Calabi-Yau manifold satisfy a few more properties, which drastically
decrease the number of undetermined Hodge numbers. We will now focus on Calabi-Yau three-
folds, that is Calabi-Yau manifolds with complex dimension 3, for the sake of brevity. These
are the most important Calabi-Yau manifolds in physics applications. But most results extend
straighforwardly to higher dimensional Calabi-Yau manifolds.
We have already shown that the Hodge numbers of Kähler manifolds satisfy a Hodge star
duality hp,q = h3−q,3−p and a complex conjugation duality hp,q = hq,p . For Calabi-Yau man-
ifolds, there is a further duality, called holomorphic duality. We have shown in the previous
section that the triviality of the canonical bundle of a Calabi-Yau manifold M of real dimension
Complex geometry, Calabi-Yau manifolds and toric geometry 67

6 implies that h3,0 = 1, i.e. the existence of a unique holomorphic volume form Ω.
R Given a (0, q)
cohomology class [α], there is a unique (0, 3 − q) cohomology class [β] such that M α ∧ β ∧ Ω = 1
(using Stoke’s theorem). Thus h0,q = h0,3−q . Therefore, for a Calabi-Yau manifold we have that
h3,0 = h0,3 = h0,0 = h3,3 = 1.
Moreover, one can show that h1,0 = 0 [2, 5] (from the fact that b1 = 0 – since we assume M
to be simply connected – or otherwise). Thus, h1,0 = h0,1 = h0,2 = h2,0 = h2,3 = h3,2 = h3,1 =
h1,3 = 0. Therefore, the only remaining independent Hodge numbers are h1,1 and h2,1 , and the
Hodge diamond takes the form:

1
0 0
0 h1,1 0
1 h2,1 h2,1 1 (3.3)
0 h1,1 0
0 0
1
P2m
The Euler class of a Calabi-Yau manifold also simplifies. Recall that χ = k=0 (−1)k bk , so
we now have that χ = 2b0 − 2b1 + 2b2 − b3 = 2 − 0 + 2h1,1 − 2 − 2h2,1 , that is

χ = 2(h1,1 − h2,1 ). (3.4)

Therefore, if the Euler class is easily computed, we only have to compute one of the two
independent Hodge numbers. In fact, we saw in section 1.5 that the Euler class is given by the
integral over M of the top Chern class of M , which is c3 (M ) for a Calabi-Yau threefold:
Z
χ = c3 (M ). (3.5)
M

This formula can be used to compute the Euler class of M .

We saw earlier that h1,1 classifies infinitesimal deformations of the Kähler structure. For a
Calabi-Yau threefold, similarly, h2,1 classifies infinitesimal deformations of the complex structure.
We refer the reader to chapter 6 of [3] for a detailed discussion of this interpretation and of the
moduli space of Calabi-Yau manifolds.

Remark 3.1. One of the fascinating property of Calabi-Yau threefolds (and manifolds in gen-
eral) is that they come in mirror pairs, (M, W ), such that H 2,1 (W ) ∼
= H 1,1 (M ) and H 1,1 (W ) ∼
=
2,1
H (M ). Roughly speaking, the complex structure moduli is exchanged with the Kähler structure
moduli. This is the basic idea behind mirror symmetry. See [3, 16] for more information on this
subject.

3.3 Examples
We will now study in some details two particular examples of Calabi-Yau threefolds. The first
one, the quintic in CP4 was the first Calabi-Yau threefold to be studied. Our second example,
the Tian-Yau manifold, was the first three-generation manifold to be discovered. Both examples
have been very important in the history of string theory, and they will help us find our way in
the asbtract jungle of Calabi-Yau threefolds.
There are various ways one can follow to see if a Kähler manifold is Calabi-Yau. The more
‘hands-on’ way is probably to find a globally defined and nowhere vanishing holomorphic volume
68 Bouchard

form (see for instance chapter 9 of [4] for this approach). Another approach, more abstract, is
to compute explicitely the first Chern class of the manifolds and see that it vanishes. In our two
examples, we will first follow the latter, as in the process we will learn how to compute Chern
classes. Then we will quickly review how to construct the holomorphic volume form.
Through these two examples we will study in more generality complete intersection (CI)
manifolds in complex projective spaces and products thereof. But to start with we need to know
the Chern class of the complex projective space CPm .
Obviously we cannot prove here all the results that are needed to carry on the computations.
The reader is referred to [5] for a detailed discussion of various constructions of Calabi-Yau
threefolds.

Chern class of CPm

First, we need to compute the total Chern class of CPm . We recall that homogeneous coordinates
zi , i = 0, . . . , m of Cm+1 are sections of the hyperplane line bundle L. Thus, the holomorphic
∂
tangent bundle of Cm+1 is spanned by tangent vectors si (z) ∂z i
, where the si are any sections of
m (1,0) m ∂
L. Now, on CP , the holomorphic tangent bundle T CP is also spanned by si (z) ∂z i
, with
the si any sections of the hyperplane line bundle – which we now denote by OCPm (1), but we
L
have to take equivalence classes with respect to overall rescaling, since overall rescaling is trivial
in CPm . That is, we have a map from OCPm (1) (m+1) to T (1,0) CPm such that its kernel is the
∂ ∼
trivial line bundle C of multiples of a nowhere-vanishing generator (z0 , . . . , zm ) 7→ zi ∂zi
= 0 in
m
CP . This is summarised in the following exact sequence called the Euler sequence:
L (m+1)
0 → C → OCPm (1) → T (1,0) CPm → 0. (3.6)

L
Trivially c(C) = 1, so by properties of Chern classes we have that c(CPm ) = c(T (1,0) CPm ) =
m+1
c(OCPm (1) (m+1) ) = [c(OCPm (1)] . OCPm (1) is a line bundle, that is its fibers are one-
dimensional, so the expansion of the total Chern class is simply c(OCPm (1)) = 1 + c1 (OCPm (1)).
If we let x = c1 (OCPm (1)), we find that
c(CPm ) = (1 + x)m+1 . (3.7)

Calabi-Yau condition for complete intersection manifolds

We now want to see what the Calabi-Yau condition is for CI manifolds.
Let X be a smooth hypersurface in CPm defined as the zero-locus of a degree d poly-
nomial p. We can see p as a section of the holomorphic line bundle OCPm (d). Consider
T (1,0) X, the holomorphic tangent bundle of X. We define the normal bundle NX on X
to be the quotient NX = T (1,0) CPm |X /T (1,0) X. As a result, we have the exact sequence
0 → T (1,0) X → T (1,0) CPm |X → NX → 0.
Roughly speaking, on X the section p maps points of X to 0 in the fibers of OCPm (d), since
X is defined as the zero-locus of p. Thus, p serves as a coordinate near X, and in fact the normal
bundle NX of X is simply OCPm (d)|X . The above exact sequence becomes (this is also known
as the adjunction formula 1; see [11] for more about this)
0 → T (1,0) X → T (1,0) CPm |X → OCPm (d)|X → 0, (3.8)
which implies that c(X) = c(CPm )/c(OCPm (d)). OCPm (d) is a line bundle, so c(OCPm (d)) =
1 + c1 (OCPm (d)). We know from above that c(OCPm (1)) = 1 + x. Therefore, its Chern character
is ch(OCPm (1)) = ex . It follows that ch(OCPm (d)) = edx = 1 + c1 (OCPm (d)) + . . ., hence
c(OCPm (d)) = 1 + dx. (3.9)
Complex geometry, Calabi-Yau manifolds and toric geometry 69

Using these results we find that the Chern class of X is

(1 + x)m+1
c(X) = . (3.10)
1 + dx
Since x is a closed two-form, we can expand c(X) in such a way that products of x are wedge
products, from which we can extract the Chern classes ck (X) of X. The first Chern class is
c1 (X) = [(m + 1) − d]x. (3.11)
Therefore, we have found an explicit realisation of the Calabi-Yau condition for CI manifolds;
the condition c1 = 0 implies a condition on the degree of the polynomial equation, d = m + 1. If
we want a Calabi-Yau threefold, we have m = 4, and therefore X must be given by the zero-locus
of a degree 5 polynomial, that is a quintic in CP4 .
In fact, it is straightforward to generalise the above computation to CI manifolds Y given
by the zero-locus of a finite number of polynomials in CPm . If there are l such polynomials of
degree di , i = 1, . . . , l, the Chern class is given by
(1 + x)m+1
c(Y ) = Ql . (3.12)
i=1 (1 + di x)
Pl
Therefore, the vanishing of the first Chern class implies the restriction m + 1 = i=1 di on the
degrees of the l polynomial defining Y .
There are only five solutions to the above condition, if we are looking for Calabi-Yau three-
folds, that is l = m − 3 (and we assume di ≥ 2, since if one equation has dj = 1 the manifold
defined by the CI in CPm is equivalent to the manifold defined by the other equations in CPm−1 ):
the quintic in CP4 , the intersection of two cubics in CP5 , the intersection of a quadric and a
quartic in CP5 , the intersection of two quadrics and a cubic in CP6 , and the intersection of four
quadrics in CP7 .
Finally, if we expand completely the total Chern class (using the fact that c1 = 0), we find
" l # " l #
1 X 2 2 1 X 3
c(Y ) = 1 + ( d ) − (n + 1) x − ( d ) − (n + 1) x3 . (3.13)
2 i=1 i 3 i=1 i

This result will be useful for the computation of the Euler class through the integration of the
third Chern class over the manifold Y .

The quintic in CP4

We will now concentrate on the quintic Q in CP4 , which is given by a polynomial equation
of degree 5 in the homogeneous coordinates of CP4 . According to the results of the previous
section, we have that
c(Q) = 1 + 10x2 − 40x3 . (3.14)
To find its Euler class, we must integrate c3 = −40x3 over Q. How do we do that? We invoke
Poincaré duality, which is an intersection pairing of cohomology classes.7 In fact, we want to
‘lift’ the integral to the embedding space where the integration is trivial. Using Poincaré duality
and de Rham’s theorems relating homology and cohomology one can prove the following theorem
about complex integration over submanifolds: [3, 4]
7 Let M be a n-dimensional manifold. Using the operators defined in section 2, one can show that a k-form

ω on M is harmonic if and only if ∗ω is also harmonic. Since the space of harmonic k-forms is isomorphic to
the de Rham cohomology group HdR k (M, R), and that ∗ω is a (n − k)-form, we have an isomorphism between
k ∼ n−k
HdR (M, R) = HdR (M, R). This is Poincaré duality.
70 Bouchard

Theorem 3.2. Let X be a closed k-dimensional submanifold of M , where R M is n-dimensional.

k
For any closed form τ ∈ HdR (M, R), we can define the integration X τ . By Stoke’s theorem,
R
this integral is independent of the choice of representative of the cohomology class. Thus X is a
n−k
linear map H k → R, and Poincaré duality says that there is a (n − k)-form ηX ∈ HdR (M, R)
such that Z Z
τ = τ ∧ ηX . (3.15)
X M

We call ηX the Poincaré dual class to X.

Therefore, the Poincaré dual class restricts the integration to the submanifold X like a delta
function. But how do we find ηX ? In fact in the special case where the normal bundle NX to X
is the restriction to X of some bundle over M , i.e. NX = E|X – which is the case we consider,
with E = OCP4 (5) – then it is easy to find ηX ; ηX = ck (E), where k is the rank of E; that it is
it is the top Chern class of E. R
Thus, for the quintic we have that ηQ = c1 (OCP4 (5)) = 5x. Now, since CPm xm = 1 (this is
so because x is Poincaré dual to a hyperplane and m hyperplanes intersect at a point), we find
Z Z Z
3
χ(Q) = c3 (Q) = (−40x ) = (−40x3 ) ∧ (5x) = −200. (3.16)
Q Q CP4

Now to pursue the study of the quintic further we must determine its Hodge numbers. Let
us first consider h2,1 , which classifies infinitesimal deformations of the complex structure. In
other words, given a polynomial equation of a certain degree in CPm , the complex structure is
determined by the free coefficients (usually called parameters) in the polynomial equation. Fixing
these parameters ‘chooses’ a particular complex structure; but by modifying these coefficients we
move in the moduli space of complex structures of the polynomial equation of a certain degree.
Therefore, these parameters classify infinitesimal deformations of the complex structure, and
h2,1 of the CI manifold is equal to the number of free parameters.

Remark 3.3. Note that this simple method for finding the Hodge numbers of a Calabi-Yau
manifold does not always work; the manifold must satisfy a few extra conditions. However it
works in this example and in the next example we will consider. Another method is to use
the Lefschetz hyperplane theorem to compute h1,1 directly – see [5] for an explanation of this
technique.

For the quintic in CP4 , there are initially 126 parameters.8 The group of holomorphic
automorphisms of CPm being P GL(m + 1, C),9 25 − 1 of them can be removed by a ho-
mogeneous linear change of variables. Moreover, one parameter corresponds to an overall
rescaling. Therefore, there are 101 parameters describing the complex structure of Q. Hence
h2,1 (Q) = 126 − (25 − 1) − 1 = 101.
Since χ = 2(h1,1 − h2,1 ), we have that h1,1 (Q) = 1, i.e. there is only one Ricci-flat Kähler
form on the quintic.
8 Thenumber of independent degree d homogeneous polynomials in n variables is given by the binomial
 
d+n−1
coefficient .
n−1
9 The projective linear group P GL(m + 1, C) is the general linear group GL(m + 1, C) quotiented by the group

Z(m + 1) of all nonzero scalar transformations, that is P GL(m + 1, C) = GL(m + 1, C)/Z(m + 1). This is the
group of holomorphic automorphisms of CPm since the action of GL(m + 1, C) on Cm+1 descends to an action
of P GL(m + 1, C) on CPm .
Complex geometry, Calabi-Yau manifolds and toric geometry 71

To summarise our result; the quintic Q in CP4 has Euler class χ = −200 and Hodge diamond

1
0 0
0 1 0
1 101 101 1 (3.17)
0 1 0
0 0
1

Now we know that Q is a Calabi-Yau manifold, and we studied it using Chern classes. We
will now construct a holomorphic volume form Ω on Q. Alternatively, we could have started our
study of the quintic by attempting a direct construction of an holomorphic volume form, and
show that way that Q is indeed Calabi-Yau.
P4
Define the form τ on C5 by τ = µ=0 dz0 ∧ . . . ∧ zµ ∧ . . . ∧ dz4 (notice that we have replaced
dzµ by zµ ). τ is clearly a holomorphic (4, 0)-form. However, it is not invariant under scaling
zµ → λzµ , so it is not well-defined on CP4 . But the form τ /Q is invariant, where Q is a degree
5 homogeneous polynomial in CP4 . However, it is singular at Q = 0.
Now let γQ be a small loop around Q = 0 in CP4 . Define
Z
τ
Ω= . (3.18)
Q
γQ

Ω is a globally defined and nowhere vanishing

 holomorphic (3, 0)-form on Q = 0. To see this,
∂z0
in a coordinate patch, rewrite dz0 = ∂Q dQ, and integrate along the loop γQ ; by the residue
theorem we find !
P4
µ=1 dz1 ∧ . . . ∧ zµ ∧ . . . ∧ dz4
Ω = (2πi) , (3.19)
(∂Q/∂z0 )
Q=0

which is a nowhere vanishing holomorphic (3, 0)-form on Q = 0.

We have found a holomorphic volume form Ω on Q; therefore as we saw in the previous
section all other holomorphic (3, 0)-forms are constant multiples of Ω.
We can easily extend this construction to CI manifolds constructed as the zero-locus of a
finite number of polynomials in a projective space.
Consider a CI of N polynomials P i ,P i = 1, . . . , m − 3 in a projective space CPm (in order to
m
have a threefold). Define the form τ = µ=0 dz0 ∧ . . . ∧ zµ ∧ . . . ∧ dzm on Cm+1 . Again, this is
Qm−3
not invariant under scaling. However, define the form τ /( i=1 P i ); it is invariant under scaling
if the degrees di of the polynomials P i satisfy

m−3
X
(m + 1) = di , (3.20)
i=1

which is exactly the restriction on the degrees of the polynomial that we found earlier for the
first Chern class to be zero.
Now, consider a (m − 3)-dimensional contour

Γm−3 = γ1 × γ2 × . . . × γm−3 , (3.21)

72 Bouchard

which is the Cartesian product of (m − 3) small loops around the (m − 3) curves defined by
P i = 0, i = 1, . . . , m − 3. Define the form
Z
τ
Ω= Qm−3 i ; (3.22)
i=1 P
Γm−3

this is a globally defined and nowhere vanishing holomorphic (3, 0)-form on the complete inter-
section of the polynomials P i in CPm .

The Tian-Yau manifold

In order to study the Tian-Yau manifold, we must generalise the results of the last section to
manifolds defined by the zero-locus of a finite number of homogeneous polynomial equations in
a product of projective spaces. Let us first introduce some notation.
Such spaces will be denoted by a configuration matrix which gives the degree of each poly-
nomial in the variables of each projective space. Each column corresponds to the degree of one
of the polynomial. We also usually add the Euler class of the manifold at the bottom right of
the configuration matrix. For instance, in this notation the quintic in CP4 is given by

CP4 |5|−200 , (3.23)

while the Tian-Yau manifold is given by

CP3 1 3 0
.
CP3 1 0 3 −18

In other words, the Tian-Yau manifold is given by three polynomial equations in CP3 × CP3 ; one
of degree 1 in both CP3 , one of degree 3 in the first CP3 , and one of degree 3 in the second CP3 .
That is, if xµ and ym are respectively homogeneous coordinates of the two CP3 , it represents
the system of equations

f µνρ xµ xν xρ = 0, g mnr ym yn yr = 0, hµm xµ ym = 0, (3.24)

where f, g, h are coefficients of the equations.

Remark 3.4. In fact, the configuration matrix does not specify a particular manifold, but rather
the family of all complete intersections parameterised by the space of coefficients; we call this
family a configuration. As we noted earlier, two different sets of coefficients correspond to two
complete intersections which in general are different as complex manifolds. The space of these
coefficients is a parameter space for the complete intersection manifolds, and by taking into
account automorphisms of the ambient space and overall rescaling it parameterises the complex
structure moduli space of this family of complete intersections.

Therefore, we should say that the Tian-Yau manifold is an element of the configuration
above, rather than the configuration itself. In fact, it is given by the following equations with
fixed coefficients:
X3 X 3 X3
xi yi = 0, (xi )3 = 0, (yi )3 = 0. (3.25)
i=0 i=0 i=0

Now we want to compute the Euler class and the Hodge numbers of this configuration. It is
straightforward to generalise the results of the previous section.
Complex geometry, Calabi-Yau manifolds and toric geometry 73

Let X be a smooth CI manifold defined by the configuration matrix

CPn1 d11 ··· d1N
.. .. .. ,
. . .
CPnl dl1 ··· dlN
that is it is a CI manifold in a product of l projective spaces of dimensions ni , i = 1, . . . , l, defined
by the zero-locus of N polynomials of degree vectors dj , j = 1, . . . , N in the l projective spaces.
Given such a configuration, it is easy to generalise the previous computation of the Chern class
to obtain Ql
(1 + xr )nr +1
c(X) = QN r=1 Pl . (3.26)
s
a=1 (1 + s=1 da xs )
By expanding, the first Chern class is
l N
!
X X
c1 (X) = nr + 1 − dra xr . (3.27)
r+1 a=1

For c1 (X) to be zero, all the coefficients in the sum must vanish, and we find the condition
N
X
dra = nr + 1, ∀ r = 1, . . . , l. (3.28)
a=1

For the Tian-Yau manifold, l = 2, n1 = n2 = 3, N = 3 and d11 = d21 = 1, d12 = 3, d22 = 0 and
d13= 0, d23 = 3. The condition is satisfied, and therefore the Tian-Yau manifold is a Calabi-Yau
manifold.
To find its Euler class, we must expand the total Chern class to find an expression for the
third Chern class. If c1 (X) = 0, we find that
l
" N
#!
X 1 rst X
r s t
c3 (X) = δ (nr + 1) − da da da xr xs xt . (3.29)
r,s,t=1
3 a=1

To integrate this result, we need a Poincaré dual class. As before, the normal bundle to X is
restriction of a bundle on the covering
 space M , that it NX = E|X for some bundle E over M .
LN Nl r
In fact, E = a=1 O
r=1 r a . Therefore, E is a bundle of rank N , hence ηX = cN (E).
(d )
Since " N !# !
M O l N
^ Ol
r r
c Or (da ) = c Or (da ) , (3.30)
a=1 r=1 a=1 r=1
Nl r
and r=1 Or (da ) is a line bundle for any a, we then have that
" N l
!# N l
! N l
!
M O ^ O ^ X
r r
ηX = cN Or (da ) = c1 Or (da ) = dra xr . (3.31)
a=1 r=1 a=1 r=1 a=1 r=1
R
Thus, since CPni (xi )ni = 1 for i = 1, . . . , l, we find that the result of the integral is the coefficient
of Λlr=1 (xr )nr , the volume form on X – which we denote by the subscript ‘top’ – in the following
expression:
" l " N
#! N l
!#
X 1 rst X ^ X p
r s t
χ(X) = δ (nr + 1) − da da da xr xs xt · db xp . (3.32)
r,s,t=1
3 a=1 p=1 b=1 top
74 Bouchard

Using this general formula one can compute easily that the Euler class of the Tian-Yau manifold
is χ = −18. Furthermore, one can show that the Euler class of any CI Calabi-Yau manifold
must be nonpositive, that is χ ≤ 0.
Now let us try to find the Hodge numbers of the Tian-Yau manifold. We will use the
same method as for the quintic, namely simply counting the free parameters in the polynomial
equations.
Two equations are of degree 3 in 4 variables, so together they have 40 free parameters.
However, in each CP3 (16 − 1) of them can be removed by a homogeneous linear change of
variables, and 1 by overall rescaling. Therefore in these two equations there are in total 8 free
parameters.
Now the third equation has 16 coefficients, and 1 can be removed by oversall rescaling.
Therefore, in total there are 15 + 8 = 23 free parameters. Hence h2,1 (X) = 23. Further, from
the equation χ = 2(h1,1 − h2,1 ), we find that the Tian-Yau manifold has Euler class χ = −18
and Hodge diamond
1
0 0
0 14 0
1 23 23 1 (3.33)
0 14 0
0 0
1
We can construct the holomorphic volume form in exactly the same way as we did before.
Let X be a smooth manifold given by the configuration matrix
CPn1 d11 ··· d1N
.. .. .. ,
. . .
CPnl dl1 ··· dlN

Let zir , i = 0, . . . , nr be coordinates on Cnr +1 . On each complex space we define a form τr =

Pm+1 r r r
Ql
µ=0 dz0 ∧ . . . ∧ zµ ∧ . . . ∧ dzm . The product of all these, τ = r=1 τr , is a form on the space
Ql nr +1
QN
r=1 C . Again, this is not invariant under scaling. However, define the form τ /( a=1 P a );
it is invariant under scaling if the condition (3.28) is satisfied, and thus defined on the space
Ql nr
r=1 CP .
Now, consider a contour
ΓN = γ1 × γ2 × . . . × γN , (3.34)
which is the Cartesian product of N small loops around the N curves defined by P i = 0,
i = 1, . . . , N . Define the form Z
τ
Ω= QN ; (3.35)
a
a=1 PΓN

this is a globally defined and nowhere vanishing holomorphic (3, 0)-form on the complete intere-
Ql
section of the polynomials P N in r=1 CPnr .
The Tian-Yau manifold was important historically as it was the first manifold to give a three-
generation spectrum for the low-energy physics coming out of string theory. The first attempts at
finding the standard model from string theory used what is called the ‘standard’ compactification
of heterotic string theory.10 In these compactifications, the number of generations of the low-
10 ‘Standard’ compactification simply means that the heterotic vacuum is defined by choosing the vector bundle

on the compact Calabi-Yau threefold to be the tangent bundle.

Complex geometry, Calabi-Yau manifolds and toric geometry 75

energy theory is given by half the absolute value of the Euler class of the compact Calabi-Yau
threefold. Therefore, to find a three-generation model we must have χ = ±6.
The Tian-Yau manifold does not satisfy this condition; however, it admits a free Z3 action
which leads to a new Calabi-Yau manifold, X/Z3 , which has Euler class χ = −18/3 = −6. There-
fore, compactification of heterotic strings on the quotiented manifold yields a three-generation
model.
It may seem easy to construct three-generation manifolds, since there is a large number
of complete intersection Calabi-Yau manifolds in products of projective spaces (at least a few
thousands). But in fact, only three phenomelogically interesting constructions (I think) have
been found that way, and they seem to be simply related (see [5]). This is rather surprising,
especially because Tian and Yau constructed their manifold before a list of CI manifolds was
even compiled.
However, at the moment there are many other ways to construct Calabi-Yau manifolds,
although the technique we have explained is still probably the simplest one. For instance, one
can construct Calabi-Yau manifolds as hypersurfaces in weighted projective spaces, as blow-up
of orbifolds, as double fibrations, etc. These constructions yield many other three-generation
manifolds.
Moreover, the standard compactification of heterotic strings was the first attempt at extract-
ing real physics from string theory, but there are now many other ways to obtain phenomeno-
logically interesting physics from string theory. For example, one can consider ‘non-standard’
compactifications of heterotic strings, that is heterotic vacua defined by vector bundles on Calabi-
Yau threefolds which are not the tangent bundle. Recently, such a compactification has been
constructed, giving not only three generations of particles, but a fully realistic massless spectrum
that exactly reproduces the Minimal Supersymmetric Standard Model, with no exotic particles,
and realistic tri-linear couplings in the low-energy superpotential [17, 18].

3.4 ‘Local’ Calabi-Yau manifolds

To end this section, we give a quick definition of ‘local’ Calabi-Yau manifolds. So far, we only
considered compact manifolds, and our definitions of Calabi-Yau manifolds assumed that the
manifolds were compact. However, it is possible to generalise this definition to admit noncom-
pact Calabi-Yau manifolds; by local (or noncompact) Calabi-Yau manifolds, we mean that they
are open neighbourhoods in compact Calabi-Yau manifolds. These are very useful in many ap-
plications in physics, for instance in topological strings [19, 20]. They are also relevant in the
study of geometric transitions [21].
Strictly speaking, if M is noncompact, Yau’s theorem does not apply; we have to supplement
it by some boundary conditions at infinity. Therefore, c1 = 0 does not directly imply that there
exists a Ricci-flat Kähler metric on a noncompact manifold. We will then use the condition on
the first Chern class to generalise our definition of Calabi-Yau manifolds to both compact and
noncompact manifolds.

Definition 3.5. A Calabi-Yau manifold is a Kähler manifold (M, J, g) with c1 (M ) = 0.

This is the same thing as asking for a nowhere vanishing holomorphic (m, 0)-form, or a trivial
canonical bundle, as before. However, the equivalence with the definitions using the holonomy
group and the zero Ricci form hold only when M is compact.
The simplest noncompact Calabi-Yau manifold is obviously Cm .
76 Bouchard

4 Toric geometry
So far we explored various aspects of complex geometry using tools of differential geometry,
sometimes bifurcating in the realm of algebraic geometry. To understand toric geometry, we
must now dive into the abstract world of algebraic geometry.
Toric varieties11 are a special kind of varieties which provide an elementary way to understand
many abstract concepts of algebraic geometry. Owing to its beauty and simplicity, toric geometry
also gives the possibility to compute various non-trivial results in string theory that could not
be calculated otherwise.
In this section we explore various aspects of toric geometry relevant for applications in physics,
mainly in string theory. Our main goal will be to construct Calabi-Yau manifolds in toric
geometry. Therefore we will skip some important concepts and applications of toric geometry.
For good and more complete introductions to toric geometry, the reader is referred to [3, 6, 7, 8].
This section, although based on the concepts of complex geometry developed in the first three
lectures, is almost independent from the rest of these lecture notes.

4.1 Homogeneous coordinates

Toric varieties may be approached from various points of view. They can be described using
fans and homogeneous coordinates, or viewed as symplectic manifolds, or correspondingly as
the Higgs branch of the space of supersymmetric ground states of the gauged linear sigma
model (GLSM), or even associated to convex polytopes in integral lattices. Perhaps the simplest
approach is the homogeneous coordinate description [22]; therefore we will proceed as far as
possible using this approach.
An interesting aspect of Cox’s approach to toric geometry is that by using the homogeneous
coordinate construction, toric varieties look very much like the usual complex (weighted) pro-
jective spaces. In fact, from that point of view we can understand toric varieties as an algebraic
generalisation of complex (weighted) projective spaces.
To start with, let us explain roughly what a toric variety is. Consider Cm and an action by
an algebraic torus (C∗ )p , p < m. We identify and then substract a subset U that is fixed by a
continuous subgroup of (C∗ )p , then safely quotient by this action to form

M = (Cm \ U ) /(C∗ )p . (4.1)

M is called a toric variety, as it still has an algebraic torus action by the group (C∗ )m−p
descending from the natural action of (C∗ )m on Cm .
For instance, CP2 is a toric variety. Indeed, a standard way of describing CP2 is by embedding
it into C3 :
CP2 = (C3 \ {0})/(C∗ ), (4.2)
where the quotient is implemented by modding out by the equivalence relation

(x, y, z) ∼ λ(x, y, z), (4.3)

where λ ∈ C∗ . We see that this description of CP2 satisfies the definition of a toric variety given
above.
11 Roughly speaking, a variety is the algebraic analog of a manifold in differential geometry. More precisely, an

algebraic variety V ⊂ CPm is the zero-locus in CPm of a collection of homogeneous polynomials. In fact, any
analytic subvariety of CPm is an algebraic variety; this is a restatement of Chow’s theorem mentionned in section
1.1. For more information about analytic and algebraic varieties see [11].
Complex geometry, Calabi-Yau manifolds and toric geometry 77

We now describe how to extract toric varieties from a fan using the homogeneous coordinate
approach developed by Cox [22].
Let M and N be a dual pair of lattices, viewed as subsets of vector spaces MR = M ⊗Z R
and NR = N ⊗Z R. Let (u, v) → hu, vi denote the pairings M × N → Z and MR × NR → R.

Definition 4.1. A strongly convex rational polyhedral cone σ ∈ NR is a set

s = {a1 v1 + a2 v2 + . . . + ak vk |ai ≥ 0} (4.4)

generated by a finite number of vectors v1 , . . . , vk in N such that σ ∩ (−σ) = {0}.

Let us put words on this definition. Suppose the lattice N is n-dimensional, that is N =∼ Zn .
A convex rational polyhedral cone is a n or lower dimensional cone in NR , with the origin of the
lattice as its apex, such that it is bounded by finitely many hyperplanes (‘polyhedra’), its edges
are spanned by lattice vectors (‘rational’) and it contains no complete line (‘strongly convex’).
A face of a cone σ is either σ itself or the intersection of σ with one of the bounding hyper-
planes.

Remark 4.2. In the remaining of this section we will refer to convex rational polyhedral cones
simply as cones.

Definition 4.3. A collection Σ of cones in NR is called a fan if each face of a cone in Σ is also
a cone in Σ, and the intersection of two cones in Σ is a face of each.

Now let Σ be a fan in N . Let Σ(1) be the set of one-dimensional cones (or edges) of Σ.
From now on we will focus on three-dimensional toric varieties, or correspondingly on three-
dimensional lattices M, N ≃ Z3 .
Let vi , i = 1, . . . , k be the vectors generating the one-dimensional cones in Σ(1), where
k = |Σ(1)|. To each vi we associate an homogeneous coordinate wi ∈ C. From the resulting Ck
we remove the set [
ZΣ = {(w1 , . . . , wk ) : wi = 0 ∀ i ∈ I}, (4.5)
I

where the union is taken over all sets I ⊆ {1, . . . , k} for which {wi : i ∈ I} does not belong to
a cone in Σ. In other words, several wi are allowed to vanish simultaneously only if there is a
cone such that the corresponding vi all belong to this cone.
Then the toric variety is given by

Ck \ Z(Σ)
MΣ = (4.6)
G
where G is (C∗ )k−3 times a finite abelian group. For all the toric varieties we consider in
these lectures the finite abelian group is trivial, so from now on we will omit it (see [6] for
an explanation of this group). The quotient by (C∗ )k−3 is implemented by taking equivalence
classes with respect to the following equivalence relations among the coordinates wi
1 k
(w1 , . . . , wk ) ∼ (λQa w1 , . . . , λQa wk ) (4.7)
Pk
with λ ∈ C∗ and i=1 Qia vi = 0. Among these relations, k − 3 are independent. We choose the
Qia such that they are integer and the greatest common divisor of the Qia with fixed a is 1.
Using this construction, it is easy to see that the complex dimension of a toric variety is
always equal to the real dimension n of the lattice N ∼
= Zn .
78 Bouchard

v2
11
00
00
11 11
00
00
11 11
00
00
11

00
11 11
00
000000
111111 11 v1
00
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
11
00 11
00 11
00
v3

Figure 2.5: The fan Σ of CP2 .

Example 4.4. Let us come back to the example of CP2 (which is two-dimensional rather than
three-dimensional, but easier to visualise as a first example). The fan is given in figure 2.5.
There are three one-dimensional cones generated by the vectors v1 = (1, 0), v2 = (0, 1) and
v3 = (−1, −1), to which we associate the homogeneous coordinates w1 , w2 and w3 of C3 . The
set ZΣ is simply {0}, and thus the toric variety is given by

MΣ = (C3 \ {0})/(C∗ ) (4.8)

Moreover, we have that 1(1, 0) + 1(0, 1) + 1(−1, −1) = (0, 0), so the C∗ quotient is implemented
by the equivalence relation (w1 , w2 , w3 ) ∼ λ(w1 , w2 , w3 ). This is the usual description of CP2 .
In toric geometry it is straightforward to know whether a toric variety is compact or not:
Proposition 4.5. A toric variety MΣ is compact if and only if its fan Σ fills NR .
The reader is referred to [8] for a proof of this proposition.

Toric divisors
In a toric variety there is a natural set of divisors called toric divisors.
Definition 4.6. Let MΣ be a toric variety described by a fan Σ. Associate a homogeneous
coordinates wi to each vector vi generating the one-dimensional cones of Σ. The toric divisors
Di of MΣ are the hypersurfaces defined by the equations wi = 0.
Since we associated a homogeneous coordinates wi to each one-dimensional cones vi in the
fan Σ of MΣ , we can think of the vectors vi as correponding to the toric divisors defined by
wi = 0.
Similarly, higher-dimensional cones of Σ correspond to lower dimensional algebraic subvari-
eties of MΣ .
In fact, it can be shown (see [8]), by using methods very similar to those used for complex
projective spaces, that the canonical bundle of MΣ is given by
X
KMΣ = O(− Di ). (4.9)
i

This result will be useful to determine whether a toric variety is Calabi-Yau or not.
Complex geometry, Calabi-Yau manifolds and toric geometry 79

4.2 Toric Calabi-Yau threefolds

We will now implement the Calabi-Yau condition on toricPthreefolds.
We have seen in the first section that to a divisor D = i ai Ni we can associate a line bundle
with a meromorphic section such that the meromorphic section has a zero of order ai along Ni
if ai > 0 and a pole of order −ai along Ni if ai < 0. The Ni are irreducible hypersurfaces, that
is hypersurfaces that cannot be written as the union of two hypersurfaces.
In the toric case, the toric divisors Di defined by wi = 0 are irreducible hypersurfaces.
Therefore, using the above correspondence we see that the toric divisor Di is associated to a
line bundle O(Di ) with a section s that has a zero of order one along Di ; thus the section s is
simply wi . Hence we see that each homogeneous coordinate wi is a section of the line bundle
O(Di ) associated to the toric divisor Di .
Now, if we consider a monomial w1a1 · · · wkak ; for ai > 0, it has zeroes of order ai along Di ,
P for aj < 0, it has poles of order −aj along Dj . Therefore it is a section of the line bundle
while
O( i ai Di ).
Let us now consider the case where ai = hvi , mi, i = 1, . . . , k for some m ∈ M . Under the
equivalence relations of the toric variety the monomial becomes
1 k P
k
Qia vi ,mi hv1 ,mi hvk ,mi
(λQa w1 )hv1 ,mi · · · (λQa wk )hvk ,mi = λh i=1 w1 · · · wk . (4.10)
Pk
But since i=1 Qia vi = 0, this monomial is invariant under the equivalence relations and there-
fore it is a true meromorphic function on our toric variety. This means that it must be a section
of the trivial line bundle, i.e.
k
X
hvi , miDi ∼ 0 for any m ∈ M. (4.11)
i=1

Pk
Conversely, if i=1 ai Di ∼ 0, then there exists a m ∈ M such that ai = hvi , mi for all i.
Now, we know that a Kähler manifold is Calabi-Yau if and only if its canonical class is trivial.
We saw in theP previous section that the canonical line bundle of a toric variety MPΣ is given by
KMΣ ∼
k k
= O(− i=1 Di ). Therefore the canonical bundle is trivial if and only if i=1 Di ∼ 0.
Using (4.11), we see that this condition is equivalent to the existence of a m ∈ M such that
hvi , mi = 1 for all i, which leads to the following proposition.
Proposition 4.7. Let MΣ be a toric manifold defined by a fan Σ. MΣ is Calabi-Yau if and
only if the vectors vi generating the one-dimensional cones of MΣ all lie in the same affine
hyperplane.
It is thus very easy to see whether a toric variety is Calabi-Yau or not; in fact, it can be read
off directly from the fan Σ of the toric variety.
A consequence of proposition 4.7 is the following:
Corollary 4.8. A toric Calabi-Yau manifold is noncompact.
Since the vi lie in a hyperplane, Σ does not fill NR . Thus proposition 4.5 tells us that MΣ
is noncompact.
This seems like a serious limitation of toric geometry, since in string theory we are often
interested in compact Calabi-Yau manifolds. However, we will see in section 4.5 how to construct
compact Calabi-Yau manifolds in toric geometry.
The Calabi-Yau condition can be rewritten in yet another equivalent form. In (4.7) we
defined the ‘charges’ (the meaning of this name will become clear in section 4.3) Qia satisfying
80 Bouchard

Pk Pk
Qia vi = 0. Therefore i=1 Qia hvi , mi = 0 for any m ∈ M . In particular, there exists an
i=1 Pk
m ∈ M such that hvi , mi = 1 for all i if and only if i=1 Qia = 0 for all a. But we showed that
a toric manifold is Calabi-Yau if and only if there exists and m ∈ M such that hvi , mi = 1 for
all i. Therefore, the condition can be restated as follows:

Proposition 4.9. A toric manifold is Calabi-Yau if and only if the charges Qia satisfy the
Pk
condition i=1 Qia = 0 for all a.

This condition is also very simple to verify. We only have to check that the charges Qia given
in the toric data describing the manifold add up to zero. Thus, if we are given a fan we simply
check that the vi lie in an affine hyperplane, while if we are given the toric data we simply verify
that the charges add up to zero.
To conclude this section we introduce a nice pictorial way of characterising toric Calabi-Yau
threefolds. We showed that for toric Calabi-Yau threefolds the vi lie in a two-dimensional plane
P . Therefore, we can draw the two-dimensional graph Γ̃ given by the intersection of the plane
P and the fan Σ. Γ̃ determines completely the fan Σ of a toric Calabi-Yau threefold. Given Γ̃,
we can draw a ‘dual’ graph Γ in the sense that the edges of Γ̃ are normals to the edges of Γ and
vice-versa. Γ is called the toric diagram of a toric Calabi-Yau threefolds MΣ . It represents the
degeneration of the fibers of the torus fibration. We will describe in more details toric diagrams
in section 4.3.

(−1, −1)

(1, 0)
(0, 0)

(0, −1)

Figure 2.6: The Γ and Γ̃ graphs for O(−3) → CP2 . The toric diagram Γ is the normal diagram
drawn in thick lines. The points (vi , 1) give the fan Σ, where the vi are the vertices of Γ̃ and are
shown in the figure.

Conversely, given a toric diagram Γ, it is straightforward to recover the fan Σ of the toric
Calabi-Yau threefold. One first draws the dual graph Γ̃, and then define the vectors vi = (νi , 1)
where νi are the vertices of Γ̃. Because of the symmetries of a three-dimensional lattice, the vi
must be the generators of the edges of the fan Σ of the toric Calabi-Yau threefold MΣ . Linear
relations between the vectors vi give the charges Qia . In other words, the fan Σ is a three-
dimensional cone over the two-dimensional graph Γ̃. An example of graphs Γ and Γ̃ is given in
figure 2.6.
Complex geometry, Calabi-Yau manifolds and toric geometry 81

4.3 Toric diagrams and symplectic quotients

In this section we describe the toric diagrams introduced above. To do so, we need to leave
momentarily the homogeneous coordinates approach to toric varieties and see toric manifolds
as symplectic quotients, or correspondingly as the Higgs branch of the space of supersymmetric
vacua of the GLSM.

Toric manifolds as symplectic quotients

Let z1 , . . . , zk be the coordinates of Ck . Let µa : Ck → C, a = 1, . . . , k − 3 be the k − 3 moment

maps defined by
k
X
Qia |zi |2 = Re(ta ), (4.12)
i=1

where the ta are complex numbers. The Qia are the same charges that were introduced in (4.7).
Pk
Therefore, the Calabi-Yau condition imposes that i=1 Qia = 0 for all a. We also consider the
action of the group G = U (1)k−3 on the coordinates defined by

zj → exp(iQja αa )zj , a = 1, . . . , k − 3. (4.13)

It turns out that

Tk−3
a=1 µ−1 (Re(ta ))
M= (4.14)
G
is a toric Calabi-Yau threefold. The k − 3 parameters ta are the complexified Kähler parameters
of the Calabi-Yau threefold.
Furthermore, since the charges Qia are the same as in (4.7), it is easy to recover the fan of
Pk
M. One only has to find distinct vectors vi satisfying i=1 Qia vi = 0; the vi generate the one-
Pk
dimensional cones of Σ. Moreover, since the Calabi-Yau condition tells us that i=1 Qia = 0,
we can choose (because of the symmetries of three-dimensional lattices) vectors vi of the form
vi = (νi , 1). The problem is then reduced to a two-dimensional problem which can easily be
solved by inspection. We see that the charges Qia are the important data defining the Calabi-Yau
toric manifolds. This is usually called the toric data of the manifold.
This description of toric manifolds also arise in gauged linear sigma models. This is a two-
dimensional U (1)k−3 gauge theory with k chiral superfields Φi , whose scalar components are
the zk . The charges of the superfields Φi under the gauge group U (1)k−3 are denoted by Qia ,
a = 1, . . . , k − 3. This is why the Qia are generally called charges. It turns out that – in the Higgs
branch – the supersymmetric ground states of the theory are parameterised by the so-called D-
term equations modulo gauge equivalence, which are nothing but the moment maps µa defined
in (4.12). In other words, the Higgs branch of the space of supersymmetric ground states of the
GLSM is the toric variety M defined above.
Now equipped with the description of toric Calabi-Yau threefolds as symplectic quotients,
let us come back to the toric diagrams introduced in section 4.2. There, we claimed that these
diagrams encode the degenaration of the fibers of the manifold. This can be seen in two different
ways: by looking at the threefold as a T 3 fibration or as a T 2 × R fibration. We will start with
the first approach, which is probably simpler, in the next section. In the section after that, we
will explore the second point of view using the topological vertex approach to toric Calabi-Yau
threefolds.
82 Bouchard

T 3 fibration
We look at the threefolds as T 3 fibrations over three dimensional base manifolds with corners.
Locally, we can introduce complex coordinates on the toric manifold: these are the zi introduced
in (4.12). They are not all independent; for a threefold, there are k − 3 relations between them
given by the moment maps (4.12). Let us rewrite these coordinates as zj = |zj |eiθj , and introduce
a new set of coordinates {(p1 , θ1 ), . . . , (pk , θk )}, with pi ≡ |zi |2 , i = 1, . . . , k. The base of the
threefold is then parameterised by the coordinates pi , while the phases θi describe the fiber T 3 .
Since |zi |2 ≥ 0, the coordinates pi satisfy pi ≥ 0. Therefore the boundaries of the base are
where some of the coordinates pi vanish. But when pj = 0 the circle |zj |eiθj degenerates to a
single point. Hence, the boundaries of the base correspond to degenerations of the corresponding
fiber directions θj . Geometrically, this means that the fiber degenerates in the direction given
by the unit normal to the boundary.
To draw the toric diagram, we first use the moment maps (4.12) to express the coordinates pj ,
j = 4, . . . , k in terms of the three coordinates p1 , p2 , p3 . Consequently, the boundary equations
pj = 0, j = 4, . . . , k become equations in the coordinates p1 , p2 and p3 involving the Kähler
parameters tj of (4.12). In fact, each boundary equation gives a plane in the space generated
by p1 , p2 and p3 . The intersections of these planes are lines; they form the toric diagram of the
toric variety, visualised as a three dimensional graph in the space generated by p1 , p2 and p3 .
Hence, in this approach the toric diagram is simply the boundary of the three dimensional
base parameterised by the pi . There is a T 3 fiber over the generic point, which degenerates at
the boundaries in a way determined by the unit normal. Thus, from this point of view toric
diagrams should be visualised as three dimensional diagrams, encoding the degeneration of the
T 3 fiber. It is perhaps simpler to understand this approach by working out a specific example.
Example 4.10. Let us find the toric diagram of O(−3) → CP2 from this point of view. This
manifold is defined by the moment map p1 + p2 + p3 − 3p4 = t, which we can use to express
p4 = 31 (p1 + p2 + p3 − t). The boundary planes are then given by p1 = 0, p2 = 0, p3 = 0 and
p1 + p2 + p3 = t. The intersections of these planes give the toric diagram of O(−3) → CP2 ,
which is drawn in figure 2.7. We see that it is the same toric diagram as the one shown in figure
2.6, but visualised as a three dimensional graph. Note that from the fourth boundary equation
one can see that the Kähler parameter t controls the size of the CP2 , as it should be.
This is indeed an easy way to visualise the geometry of the manifold from the toric diagram;
another example of this approach will be given in section 4.4. However, it turns out that in
many situations it is more enlightening to consider the manifold as a T 2 × R fibration, especially
from the topological vertex perspective. Let us now describe this alternative viewpoint.

T 2 × R fibration
In this language, a toric diagram Γ is a two-dimensional graph which represents the degeneration
locus of the T 2 × R fibration over the base R3 . Over a line in Γ in the direction (q, p), the cycle
(−q, p) of the T 2 fiber degenerates.
To exhibit this structure, we will now follow the topological vertex approach to toric Calabi-
Yau threefolds developed by Aganagic, Klemm, Mariño and Vafa in [23]. A good reference
is [19].
The fundamental idea behind this approach is that toric Calabi-Yau threefolds are built by
gluing together C3 patches. Therefore, the first step is to describe C3 (which is the simplest
noncompact toric Calabi-Yau threefold) as a T 2 × R fibration and exhibit its degeneration locus
in a two-dimensional graph Γ, which turns out to be a trivalent vertex. Then, more general
geometries are constructed by gluing together C3 patches, which, in the toric diagram language,
Complex geometry, Calabi-Yau manifolds and toric geometry 83

p3

p1

p2

Figure 2.7: Toric diagram Γ of O(−3) → CP2 visualised as a three dimensional graph. It encodes
the degeneration loci of the T 3 fiber.

corresponds to gluing together trivalent vertices in a way specified by the toric data of the
manifold.
Conversely, given a toric Calabi-Yau threefold, we can find a decomposition of the set of all
coordinates into triplets that correspond to the decomposition of the threefold into C3 patches.
The moment maps (4.12) relate the coordinates between the patches, therefore describing how
the trivalent vertices corresponding to the C3 patches are glued together to form the toric diagram
of the manifold.
Let us start by describing C3 from this point of view. Here we will only sketch the description;
the details are given in [19, 23]. Let zi , i = 1, 2, 3 be complex coordinates on C3 . Define the
functions

rα (z) = |z1 |2 − |z3 |2 ,

rβ (z) = |z2 |2 − |z3 |2 ,
rγ (z) = Im(z1 z2 z3 ). (4.15)

It turns out that these functions generate the fiber T 2 × R. More specifically, R is generated by
rγ while the T 2 fiber is generated by the circle actions

exp(iαrα + iβrβ ) : (z1 , z2 , z3 ) → (eiα z1 , eiβ z2 , e−i(α+β) z3 ). (4.16)

The cycles generated by rα and rβ are then respectively referred to as the (0, 1) and (1, 0) cycles.
We now describe the degeration loci of the fibers. We see from (4.15) and (4.16) that the
(0, 1) cycle degenerates when rα = 0 = rγ and rβ ≥ 0, while the (1, 0) cycle degenerates when
rα ≥ 0 = rγ and rβ = 0. There is also a one-cycle parameterised by α + β that degenerates
when rα − rβ = 0 = rγ and rα ≤ 0.
The toric diagram is a planar graph that encodes the degeneration loci of the fibers. We can
set rγ = 0 and draw the graph in the plane rα −rβ . The graph consists in lines prα +qrβ = c where
c is a constant. Over this line the (−q, p) cycle of the T 2 fiber degenerates (up to the equivalence
84 Bouchard

(q, p) ∼ (−q, −p)). For C3 , the degeneration loci can be represented as a toric diagram with
lines defined by the equations rα = 0, rβ ≥ 0; rβ = 0, rα ≥ 0 and rα − rβ = 0, rα ≤ 0. Over
these lines respectively the cycles (0, 1); (−1, 0) ∼ (1, 0) and (1, 1) degenerate. This gives the
trivalent vertex associated to C3 , which is shown in figure 2.8.

(0, 1)

(1, 0)

(−1, −1)

Figure 2.8: Trivalent vertex associated to C3 , drawn in the rα -rβ plan. The vectors represent
the generating cycles over the lines.

For more general geometries, we first find a decomposition of the set of coordinates zi ,
i = 1, . . . , k into triplets of coordinates associated to the C3 patches. We choose a patch and
describe the functions rα and rβ as above. It turns out that we can use these coordinates as
global coordinates for the T 2 fiber in the R3 base. As usual, we refer to the cycles rα and rβ
respectively as the (0, 1) and (1, 0) cycles. Using the moment maps (4.12) defining the toric
Calabi-Yau threefold, we can find the action of the functions rα and rβ on the other patches
and therefore draw the toric diagram giving the degeneration loci of the T 2 fiber. An explicit
example of this approach will be worked out in section 4.4.
This decomposition of toric Calabi-Yau threefolds into C3 patches leads to a similar decom-
position of topological string amplitudes on toric Calabi-Yau threefolds into a basic building
block associated to the trivalent vertex of the C3 patches, which is called the topological vertex.
By gluing together these topological vertices one can build topological string amplitudes on any
toric Calabi-Yau threefold. This beautiful property of topological amplitudes is the essence of
the topological vertex approach developed in [23]. We will not expand further on this subject;
the interested reader is encouraged to go through the details of the construction in [23].
In the next section we illustrate these different approaches to toric Calabi-Yau threefolds in
specific examples.

4.4 Examples
We now describe two examples of toric Calabi-Yau threefolds. The first example is the resolved
conifold, namely O(−1) ⊕ O(−1) → CP1 . In this simple case, we illustrate in details the differ-
ent viewpoints explained in the previous sections. The second example is a more complicated
geometry. It is a noncompact Calabi-Yau threefold whose compact locus consists of two com-
pact divisors each isomorphic to a del Pezzo surface dP2 12 and a rational (−1, −1) curve that
12 A del Pezzo surface dP , n = 0, . . . , 8 is a complex two-dimensional Fano variety, that is CP2 blown up in n
n
points.
Complex geometry, Calabi-Yau manifolds and toric geometry 85

intersects both divisors transversely. We will give the toric data describing the manifold and
draw the corresponding toric diagram.

O(−1) ⊕ O(−1) → CP1

The resolved conifold Y = O(−1) ⊕ O(−1) → CP1 is a noncompact Calabi-Yau threefold which
admits a toric description given by the following toric data:

z1 z2 z3 z4
(4.17)
C∗ 1 1 −1 −1

The lines in this table give the

Pcharges Qia corresponding to the torus actions on the homogeneous
i
coordinates zi . We see that i Q = 1 + 1 − 1 − 1 = 0; therefore Y is Calabi-Yau. Y is defined
as the space obtained from
|z1 |2 + |z2 |2 − |z3 |2 − |z4 |2 = t (4.18)
after quotienting by the U (1) action specified by the charges in (4.17).
P4
We now find the fan Σ describing Y . We have the relation i=1 Qi vi = v1 + v2 − v3 − v4 = 0.
We choose distinct vectors vi = (wi , 1) where wi is two-dimensional. A solution is v1 = (1, 0, 1),
v2 = (−1, 0, 1), v3 = (0, 1, 1) and v4 = (0, −1, 1). These four vectors generate the four one-
dimensional cones of Σ.
The two-dimensional graph Γ̃ is given by the intersection of the plane z = 1 and Σ. The
vertices are (1, 0),(−1, 0),(0, 1) and (0, −1). We can also draw the toric diagram, which is the
dual graph Γ. They are shown in figure 2.9.

(0, 1)

(−1, 0) (1, 0)

(0, −1)

Figure 2.9: The Γ and Γ̃ graphs for O(−1) ⊕ O(−1) → CP1 . The toric diagram Γ is the normal
diagram drawn in thick lines. The points (vi , 1) give the fan Σ, where the vi are the vertices of
Γ̃ and are shown in the figure.

If we look at the resolved conifold as a T 3 fibration, we have to understand the toric diagram
Γ as a three-dimensional graph representing the base, where the T 3 fiber degenerates at the
boundaries. The base is parameterised by the four coordinates pi ≡ |zi |2 subject to the relation
(4.18). We can use (4.18) to eliminate p4 ,

p4 = p1 + p2 − p3 − t. (4.19)
86 Bouchard

Therefore, since |zi |2 ≥ 0, the boundary equations of the toric base are given by

p1 = 0,
p2 = 0,
p3 = 0,
p1 + p2 − p3 = t. (4.20)

The intersections of these planes give the toric diagram of the resolved conifold shown in
figure 2.9, but visualised as a three dimensional graph as in figure 2.10. Note that as in example
4.10, by the fourth boundary equation above one can see that the Kähler parameter t controls
the size of the CP1 , as it should be.

p2

(0, 0, t)
p3

p1

Figure 2.10: Toric diagram Γ of O(−1) ⊕ O(−1) → CP1 visualised as a three-dimensional graph.
It encodes the degeneration loci of the T 3 fiber.

We can also describe the resolved conifold as a T 2 × R fibration, using its decomposition into
3
C patches. We choose the first patch to be defined by z1 6= 0. Using (4.18) we can express z1
in terms of the other coordinates, so the patch is parameterised by (z2 , z3 , z4 ). We define the
functions

rα = |z3 |2 − |z2 |2 ,
rβ = |z4 |2 − |z2 |2 . (4.21)

This gives the usual trivalent graph of C3 .

The other patch is defined by z2 6= 0, therefore parameterised by (z1 , z3 , z4 ). Using (4.18) we
can rewrite the functions (4.21) in terms of the coordinates on this patch:

rα = |z1 |2 − |z4 |2 + t,
rβ = |z1 |2 − |z3 |2 + t. (4.22)

These functions generate the circle action

exp(iαrα + iβrβ ) : (z1 , z3 , z4 ) → (ei(α+β) z1 , e−iβ z3 , e−iα z4 ). (4.23)

In this patch, the (0, 1) cycle degenerates when rα ≤ −t and rβ = −t. The (1, 0) cycle degenerates
when rα = −t and rβ ≤ −t. The (1, 1) cycle degenerates when rα − rβ = 0 and rα ≥ −t.
Therefore, the graph associated to this patch is identical to the first one, although it is shifted
Complex geometry, Calabi-Yau manifolds and toric geometry 87

such that its origin is at the point (−t, −t). The two graphs are joined through the common
edge given by rα − rβ = 0. t gives the ‘length’ of the internal edge, and correspondingly is the
Kähler parameter associated to the CP1 . This gives the toric diagram of the resolved conifold
shown in figure 2.11.

(0, 1)

U1

(1, 0)

(−1, −1)

(1, 1)
(−1, 0)

U2

(0, −1)

Figure 2.11: Toric diagram of O(−1) ⊕ O(−1) → CP1 , drawn in the rα -rβ plan. The vectors
represent the generating cycles over the lines. The origin of the second patch U2 is shifted to
(−t, −t).

Two dP2 ’s connected by a CP1

The next example is a noncompact Calabi-Yau threefold X whose compact locus consists of two
compact divisors each isomorphic to a del Pezzo surface dP2 and a rational (−1, −1) curve that
intersects both divisors transversely. The divisors do not intersect each other. This manifold is
described by the following toric data:
z1 z2 z3 z4 z5 z6 z7 z8 z9 z10
C∗ −1 1 1 −1 0 0 0 0 0 0
C∗ 1 0 −1 −1 0 1 0 0 0 0
C∗ 1 −1 0 −1 1 0 0 0 0 0
(4.24)
C∗ 0 0 0 1 −1 −1 1 0 0 0
C∗ 0 0 0 0 1 0 −1 −1 0 1
C∗ 0 0 0 0 0 1 −1 0 −1 1
C∗ 0 0 0 0 0 0 −1 1 1 −1.
We see that the charges in each line add up to zero, hence X is Calabi-Yau. The toric
diagram Γ of X and its dual Γ̃ are shown in figure 2.12.
This Calabi-Yau threefold, a related one and their orientifolds were studied in details via
topological strings in [24, 25].
88 Bouchard

Figure 2.12: The Γ and Γ̃ graphs for the Calabi-Yau threefold X whose compact locus consists
of two dP2 ’s connected by a CP1 . The toric diagram Γ is the normal diagram drawn in thick
lines.

4.5 Hypersurfaces in toric varieties

In the remaining of this section we explain how compact Calabi-Yau manifolds may be obtained
in toric geometry, namely as hypersurfaces in compact toric manifolds using Batyrev’s well known
reflexive polytopes [26].

Reflexive polytopes
In section 4.1 we described in details toric Calabi-Yau threefolds. In particular, we showed that
toric Calabi-Yau threefolds are noncompact. However, from a string theory perspective, it is
often desirable to consider compact Calabi-Yau manifolds. Hence it seems that toric geometry
is not a good setup for such geometries.
Fortunately, there is a way to construct compact Calabi-Yau manifolds in toric geometry,
namely as compact hypersurfaces in compact toric varieties. Batyrev’s reflexive polytopes [26]
provide a very useful description of such compact Calabi-Yau manifolds. The toric variety itself
is not Calabi-Yau consequently it can be compact. Reflexivity of the polytopes then ensures
that the compact hypersurface, which is not toric itself, is Calabi-Yau.
An elementary introduction to these concepts and their applications to string theory and
dualities can be found in [6]. The following is partly based on the first sections of [27].
As in section 4.1, in the following we focus our attention on three-dimensional toric varieties,
therefore leading to two-dimensional Calabi-Yau hypersufaces, i.e. K3 surfaces. It is however
straightforward to generalise the concepts to higher dimensional toric varieties.
A polytope in MR is the convex hull of a finite number of points in MR , and a polyhedron
in MR is the intersection of finitely many half-spaces (given by inequalities hu, vi ≥ c with some
v ∈ NR and c ∈ R) in MR . It is well known that any polytope is a polyhedron and any bounded
polyhedron is a polytope. If a polyhedron S ⊂ MR contains the origin 0, its dual

S ∗ = {v ∈ NR : hu, vi ≥ −1 for all u ∈ S}. (4.25)

Complex geometry, Calabi-Yau manifolds and toric geometry 89

is also a polyhedron containing 0, and (S ∗ )∗ = S.

A lattice polytope in MR is a polytope with vertices in M .

Definition 4.11. A polytope ∆ ⊂ MR containing 0 is called reflexive if both ∆ and ∆∗ are

lattice polytopes.

This is equivalent to ∆ being a lattice polytope

P whose bounding equations are of the form
hu, vi i ≥ −1 with vi ∈ N (in coordinates, j u j v ij ≥ −1 with integer coefficients vij ). By
convexity it is sufficient to consider only those equations corresponding to vi that are vertices
of ∆∗ . In this way there is a duality between vertices of ∆∗ and facets of ∆; similarly, there
are dualities between p-dimensional faces of ∆ and (n − p − 1)-dimensional faces of ∆∗ (in three
dimensions: between edges and dual edges).
An interior point u of a reflexive polytope must satisfy hu, vi i > −1 for all vi , so an interior
lattice point must satisfy hu, vi i ≥ 0. Thus if u is an interior lattice point, then nu is also an
interior lattice point for any non-negative integer n. For u 6= 0 this would be in conflict with
the boundedness of ∆, implying that 0 is the only interior lattice point.

Toric interpretation
Given a three dimensional pair of reflexive polytopes ∆ ∈ MR , ∆∗ ∈ NR , a smooth K3 surface
can be constructed in the following manner. Any complete triangulation of the surface of ∆∗
defines a fan Σ whose three dimensional cones are just the cones over the regular (i.e., lattice
volume one) triangles. To any lattice point pi = (x̄i , ȳi , z̄i ) on the boundary of ∆∗ one can
assign a homogeneous coordinate wi ∈ C, with the rule that several wi are allowed to vanish
simultaneously only if there is a cone such that the corresponding pi all belong to this cone. The
equivalence relations among the homogeneous coordinates are given by
1 k
(w1 , . . . , wn ) ∼ (λQa w1 , . . . , λQa wk ) for any λ ∈ C∗ (4.26)
P
with any set of integers Qia such that Qia pi = 0; among these relations, k − 3 are independent.
This construction gives rise to a smooth compact three dimensional toric variety MΣ (smooth
because the generators of every cone are also generators of N , compact because the fan fills NR ).
The loci wi = 0 are the toric divisors Di .
Q hq ,p i+1
To any lattice point qj of M we can assign a monomial mj = i wi j i ; the exponents
are non-negative as a consequence
P of reflexivity. The hypersurface defined by the zero-locus of
a generic polynomial P = aj mj transforms homogeneously under (4.26) and can be shown to
define a K3 hypersurface in MΣ (actually it defines a family of hypersurfaces depending on the
coefficients aj ).

Remark 4.12. A good way to remember this construction is by noting that the polytope in
∆∗ ∈ NR gives the faN of the ambient toric variety, while the polytope in ∆ ∈ MR gives the
MonoMials.

Calabi-Yau condition
In fact, it was shown by Batyrev [26] that the hypersurface defined by the vanishing of a generic
polynomial in the class determined by ∆ is a smooth Calabi-Yau manifold for n ≤ 4, where n is
the dimension of the lattice M . For n ≤ 3 the underlying toric variety is smooth; in particular
for n = 3 the hypersurface describes a smooth K3 surface as explained above. For n = 4 it may
have point-like singularities, which are however missed by the generic hypersurface describing
the Calabi-Yau threefold.
90 Bouchard

Let us now explain why the hypersurface is a Calabi-Yau manifold. Let a manifold X be
defined by the equation P = 0 in a toric variety M. As we have seen in the first section, the
polynomial P defines a section of a line bundle (other sections are defined by different coefficients
aj ). The divisor class of the line bundle can be read off from any monomial in P . Since the origin
Qk
is always included in the polytopes, P always includes the monomial i=1 wi , which corresponds
Pk
to the divisor class [ i=1 Di ]. Thus, the polynomial P determines a section of the anticanonical
bundle of the toric variety M.
We have seen in section 3 that in this case P serves as a coordinate near X, and in fact the
∗
normal bundle NX of X is simply KM |X , since P is a section of the anticanonical bundle of M.
Thus, the exact sequence 0 → T X → T 1,0 M|X → NX → 0 becomes (this result is also known
1,0

as the adjunction formula 1 – see [5, 11])

∗
0 → T 1,0 X → T 1,0 M|X → KM |X → 0. (4.27)

Now, given any holomorphic vector bundle B over X of rank k and any holomorphic sub-bundle
A, one can always form the respective determinant bundles det B and det A which satisfy the
identity det B = det A ⊗ det(B/A). Using the above exact sequence, we can then write
∗
det T 1,0 M|X = det T 1,0 X ⊗ det KM |X . (4.28)

Now, using the definition of the anticanonical bundle as the determinant line bundle of the
∗ ∗ ∗
holomorphic tangent bundle and the fact that det KM = KM since KM is a line bundle, we find
∗ ∗ ∗
KM |X = KX ⊗ KM |X , (4.29)
or equivalently
∗
KX = (KM ⊗ KM )|X , (4.30)
that is the canonical bundle KX of X is trivial. If X is smooth, which is guaranteed by reflexivity
for n ≤ 4, then X is a Calabi-Yau manifold.

Fibration structure
Suppose the intersection of ∆∗ with the plane z̄ = 0 gives a reflexive polygon. We may reinterpret
P as a polynomial in the wi for which z̄i = 0, with coefficients depending on the remaining wi ,
i.e. we are dealing with an elliptic curve parameterised by the wi for which z̄i 6= 0. The map
MΣ → P1 , Y
(w1 , . . . , wn ) → W = wiz̄i (4.31)
i:z̄i 6=0

is easily checked to be consistent with (4.26) and thus well defined. At any point of the P1 that
is neither 0 nor ∞ all the wi with z̄i 6= 0 are non-vanishing, and (4.26) can be used to set all
except one of them to 1. This gives the K3 surface the structure of an elliptic fibration.

Hodge numbers
An interesting property of the Batyrev’s construction is that the Hodge numbers can be read off
directly from the lattice data describing the Calabi-Yau manifold. We will not give the explicit
formulae here; the reader is referred to [26]. Owing to this fact, Calabi-Yau hypersurfaces in toric
varieties offer a fantastic playground to learn more about mirror symmetry, which exchanges the
Hodge numbers h2,1 and h1,1 of a Calabi-Yau threefold. In fact, mirror symmetry for hyper-
surfaces of toric varieties has been studied extensively and led to many interesting insights, as
Complex geometry, Calabi-Yau manifolds and toric geometry 91

explained in [3].

Acknowledgments. I would like to thank Wen Jiang, Fonger Ypma and Philip Candelas for
interesting discussions while writing these lecture notes. I owe special thanks to Wen Jiang
for proof reading these lecture notes before the school. My work was supported by a Rhodes
Scholarship and a NSERC PGS Doctoral Scholarship.

References
[1] M. Nakahara, Geometry, Topology and Physics (The Institute of Physics, 2002), 520 p.

[2] D. D. Joyce, Compact Manifolds with Special HolonomyOxford Mathematical Monographs

(Oxford University Press, 2000), 436 p.

[3] K. Hori et al., Mirror Symmetry, Clay Mathematics Monographs Vol. 1 (American Mathe-
matical Society, 2003), 929 p.

[4] P. Candelas, Lectures on complex geometry, in Trieste 1987, Proceedings, Superstrings ’87,
1987.

[5] T. Hübsch, Calabi-Yau Manifolds: A Bestiary for Physicists (World Scientific, 1992), 374
p.

[6] H. Skarke, String dualities and toric geometry: An introduction, (1998), hep-th/9806059.

[7] B. R. Greene, String theory on Calabi-Yau manifolds, (1996), hep-th/9702155.

[8] W. Fulton, Introduction to Toric VarietiesAnnals of Mathematics Studies (Princeton Uni-

versity Press, 1993), 157 p.

[9] V. Bouchard, Toric Geometry and String Theory, PhD thesis, University of Oxford, 2005.

[10] M. B. Green, J. H. Schwarz, and E. Witten, Superstring Theory. Vol. 2: Loop Amplitudes,
Anomalies and PhenomenologyCambridge Monographs On Mathematical Physics (Cam-
bridge University Press, Cambridge, UK, 1987), 596 p.

[11] P. Griffiths and J. Harris, Principles of Algebraic Geometry (John Wiley & Sons, 1978),
813 p.

[12] E. Calabi, The space of kähler metrics, in Proceedings of the International Congress of
Mathematicians, Amsterdam, 1954, vol. 2, pp. 206–207, Amsterdam, 1956, North-Holland.

[13] E. Calabi, On kähler manifolds with vanishing canonical class, in Algebraic geometry and
topology, a symposium in honour of S. Lefschetz, pp. 78–89, Princeton, 1957, Princeton
University Press.

[14] S.-T. Yau, On Calabi’s conjecture and some new results in algebraic geometry, Proceedings
of the National Academy of Sciences of the U.S.A. 74, 1798 (1977).

[15] S.-T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge-
Ampère equations. I., Communications on pure and applied mathematics 31, 339 (1978).
92 Bouchard

[16] D. Cox and S. Katz, Mirror Symmetry and Algebraic Geometry, Mathematical Surveys and
Monographs Vol. 68 (American Mathematical Society, 1999), 469 p.

[17] V. Bouchard and R. Donagi, An SU(5) heterotic standard model, Phys. Lett. B633, 783
(2006), hep-th/0512149.

[18] V. Bouchard, M. Cvetic, and R. Donagi, Tri-linear couplings in an heterotic minimal

supersymmetric standard model, (2006), hep-th/0602096.

[19] M. Mariño, Chern-Simons theory and topological strings, (2004), hep-th/0406005.

[20] A. Neitzke and C. Vafa, Topological strings and their physical applications, (2004), hep-
th/0410178.

[21] M. Rossi, Geometric Transitions, (2004), math.AG/0412514.

[22] D. A. Cox, The Homogeneous Coordinate Ring of a Toric Variety, Revised Version, (1993),
alg-geom/9210008.

[23] M. Aganagic, A. Klemm, M. Mariño, and C. Vafa, The topological vertex, Commun. Math.
Phys. 254, 425 (2005), hep-th/0305132.

[24] V. Bouchard, B. Florea, and M. Mariño, Topological open string amplitudes on orientifolds,
JHEP 2, 002 (2005), hep-th/0411227.

[25] V. Bouchard, B. Florea, and M. Mariño, Counting higher genus curves with crosscaps in
Calabi-Yau orientifolds, JHEP 12, 035 (2004), hep-th/0405083.

[26] V. Batyrev, Dual Polyhedra and Mirror Symmetry for Calabi-Yau Hypersurfaces, J. Alg.
Geom. 3, 493 (1994), alg-geom/9310003.

[27] V. Bouchard and H. Skarke, Affine Kac-Moody algebras, CHL strings and the classification
of tops, Adv. Theor. Math. Phys. 7, 205 (2003), hep-th/0303218.
 Aspects of Two-Dimensional Conformal Field Theory

Stéphane Detournay

Mécanique et Gravitation
Université de Mons-Hainaut, 20 Place du Parc
7000 Mons, Belgium

E-mail: [email protected]

Abstract. These notes constitute the write-up of a four hour lecture given at the
First Modave Summer School on Mathematical Physics, held in Modave, 19-25 june
2005. It is intended to introduce basic tools of conformal field theory in two dimensions
using as guideline the conformal Ward identities. Comments are welcome!

Based on the lectures presented by S. Detournay at the First Modave Summer School in
Mathematical Physics held at Modave, Belgium on June 19-25, 2005
94 Detournay

1 Conformal transformations in two dimensions

Consider flat space-time in d dimensions, that is, Rd endowed with a flat metric gµν (which will
be taken euclidian or lorentzian) and coordinates xµ , µ = 0, · · · , d − 1. First recall that Poincaré
transformations are the set of transformations
′
xµ → x µ (x) , µ = 0, · · · , d − 1 (1.1)

leaving the components of the flat metric (with lorentzian signature) unchanged :
′ !
gµν →gµν (x′ (x)) = gµν (x) . (1.2)

This statement can also be expressed from the ”active” point of view by demanding that the
squared ”Minkowskian norm” ds2 of a vector with components dxµ be preserved under (1.1) :
!
ds2 = gµν dxµ dxν → ds′2 = gµν dx′µ dx′ν = ds2 , (1.3)

that is
∂x′µ ∂x′ν
gµν = gαβ . (1.4)
∂xα ∂xβ
Conformal transformations are defined as the set of transformations (1.1) leaving the com-
ponents of the metric tensor invariant up to a scale :
′
gµν →gµν (x′ (x)) = Λ(x)gµν . (1.5)

Condition (1.5) expresses that the scalar product of basis vectors of the tangent space is conserved
only up to a local scale factor (possibly depending on the point). In particular, angles are
preserved by conformal transformations. Equivalently, conformal transformations are such that
!
ds2 → ds′2 = Λ−1 (x)ds2 . (1.6)

Let us focus on the two-dimensional case. We will be working with an euclidian metric, such
that the line element is (dx1 )2 + (dx2 )2 . We introduce complex coordinates

z = x1 + ix2 and z = x1 − ix2 , (1.7)

in which the metric reads ds2 = dzdz. Notice it is only in 2 dimensions that the metric in
complex coordinates factorizes in dz and dz. Consequently, any change of coordinates
△ △
z→f (z) = z + α(z) , z→f¯(z) = z + ᾱ(z) (1.8)

with f and f¯ depending only upon z and z respectively1 will satisfy (1.6), because
   ¯
df df
ds2 →ds′2 = ds2 . (1.9)
dz dz
Actually, these are the only conformal transformations in two dimensions. To verify this, one
may consider infinitesimal transformations x′µ = xµ + εµ (x) in (1.5), for infinitesimal εµ (x).
′ △
With gµν = gµν + ∆(x)gµν , one gets, to first order in εµ (x) :

∂µ εν + ∂ν εµ = −f (x)gµν . (1.10)
1 It is a common abuse of language to call f (z) (resp. f¯(z)) a holomorphic (resp. anti-holomorphic) function,

because it may in general possess singularities. One usually restricts to meromorphic functions, i.e. functions
admitting a set of isolated poles
Aspects of Two-Dimensional Conformal Field Theory 95

By taking the trace of (1.10), one determines ∆(x), so that we get

2
∂µ εν + ∂ν εµ = (∂α εα )gµν , (1.11)
d
which are called conformal Killing equations. Their solution, in two dimensions, expresses that
△ △
εz (z, z) = εz (z) = ε(z) and εz (z, z) = εz (z) = ε̄(z), see [3] for a detailed computation.
Now, suppose α(z) in (1.8) admits a Laurent expansion around say z = 0 (we only consider
the ”holomorphic” sector) :
+∞
X
α(z) = αn z n+1 . (1.12)
n=−∞

On functions (scalar fields) depending only upon one coordinate (say z), the mappings (1.8) are
generated by differential operators. Indeed, by considering a transformation of the form x→x′ ,
φ(x)→φ′ (x′ ), the variation of a field is defined as :
△
δφ(x) = φ′ (x) − φ(x) = ω a Ga φ(x) , (1.13)

where {ω a } is a set of infinitesimal parameters, and Ga are the symmetry generators2 . By

considering an infinitesimal version of (1.8), the effect on a function F (z) is
X
δF (z) = −α(z)∂z F (z) = αn ln F (z) , (1.14)
n

where we used (1.12) with infinitesimal parameters αn , and defined the generators

ln = −z n+1 ∂z . (1.15)

They satisfy the following commutation relation

[ln , lm ] = (n − m)ln+m , (1.16)

with analogous definition and commutation relations for the ¯ln ’s. The infinite Lie algebra defined
by (1.16) is known as the Witt algebra. Notice that the infinitesimal transformations resulting
from (1.15) are the most general that are analytic (or holomorphic, in strict sense) near the
point z = 0. They may introduce singularities at z = 0, but not branch cuts3 .
An important subset of the transformations (1.8) consists in transformations defining invert-
ible mappings globally defined on C ∪ {∞}, that is, the complex plane plus a point at infinity.
This space is called the Riemann sphere, as it may be compactified (through stereographic pro-
jection) to the two-dimensional sphere S 2 . These transformations are called global conformal
transformations.
To determine them, a first argument is based on the form (1.15) of the generators. Clearly
ln is non-singular at z = 0 if n ≥ −1. To investigate the behavior at ∞, consider the conformal
map z→z ′ = z1 . Under this transformation, the generator ln transforms to
 ′
′−(n+1) ∂z
−z n+1
∂z −→ − z ∂z′ = z ′(1−n) ∂z′ . (1.17)
∂z
2 As an example, for translations, the infinitesimal parameters are ω µ , defined by x′µ = xµ + ω µ and φ′ (x′ ) =

φ(x), so that Gµ = −∂µ

3 to admit a Laurent expansion around z , a function must be analytic in a crown centered at z . If the
0 0
function has a branch cut at z0 (like for instance log z at z = 0), then one cannot ”close a contour” around z0
and the Laurent expansion cannot be defined
96 Detournay

This operator is non-singular at z ′ = 0 if n ≤ 1. The generators (1.15) are thus defined

everywhere on the Riemann sphere if −1 ≤ n ≤ 1.
Let us see to what kind of transformations they correspond. For n = −1, the infinitesimal
transformation is
z→z ′ = z + δA , (1.18)
where the infinitesimal parameter α−1 is supposed to arise from the variation of a finite complex
parameter A. The finite transformation follows from a trivial integration :
′
Zz ZA
dz
=1⇒ dw = dA ⇒ z ′ = z + A . (1.19)
dA
z 0

Recalling the definition of z and going back to the original coordinates x1 and x2 , this corresponds
to
x1 →x1 + Re(A) , x2 →x2 + Im(A) , (1.20)
and hence to translations : xµ →xµ + aµ . For n = 0, one finds, with α0 = δT :
′
Zz ZT
dz dw
=z⇒ = dT ⇒ z ′ = Bz , B = eT . (1.21)
dT w
z 0

Let us consider two cases : B = eiθ and B = λ. With the first one, we find that the effect on
the coordinates (x1 , x2 ) is
xµ →Λµν xν , Λ ∈ SO(2) , (1.22)
thus corresponding to rotations. At this point, we just found the (euclidian) Poincaré group in
two dimensions, which was expected to be a subgroup of the conformal group. With B = λ, we
find that
xµ →λxµ , (1.23)
corresponding to dilations. Finally, for n = 1, we have
dz z
= −D ⇒z ′ = . (1.24)
z2 1 + Dz
This transformation is called special conformal transformation (SCT), and can be seen as the
combination of an inversion, a translation and an inversion (notice that inversions are conformal,
even though they do not have an infinitesimal form).
By combining (1.19),(1.21) and (1.24), one obtains the so-called global conformal transfor-
mations in two dimensions :
Bz + A Bz + A
z′ = = , (1.25)
1 + D(Bz + A) (1 + BD) + BDz
whose general form is
az + b
z′ =
. (1.26)
cz + d
Notice that (1.26) has 4 complex parameters, while (1.25) has only 3. The excessive parameter
is not relevant, since we could divide numerator and denominator of (1.26) by d. It is however
common to use this parametrization for the two-dimensional global (or finite) conformal group,
with additional condition  
a b
det =1 , (1.27)
c d
Aspects of Two-Dimensional Conformal Field Theory 97

which leaves us with 3 complex parameters. The parametrization (1.26) is very convenient
because successive transformations correspond to the product of the corresponding matrices.
Indeed, it can be checked, by looking at successive transformations of the form
a1 z + b1 a2 z1 + b2
z1 = , z2 = , (1.28)
c1 z + d1 c2 z1 + d2
that the resulting transformation is
a′ z + b′
z2 = , (1.29)
c′ z + d′
where the parameters a′ ,b′ ,c′ and d′ are given by
 ′    
a b′ a2 b2 a1 b1
= . (1.30)
c′ d′ c2 d2 c1 d1

This establishes that the global conformal group in two dimensions is equivalent to Sl(2, C)/Z2 ,
the quotient coming from the fact that A and −A ∈ Sl(2, C)/Z2 correspond to the same confor-
mal transformation4 .
At this level, it seems interesting to make a point about the status of the pair of variables
(z, z). It is customary to consider (x1 , x2 ) as a point in C2 , so that (1.7) is a mere change of
coordinates , z being not the complex conjugate of z but rather a distinct complex coordinate.
This constitutes a convenient simplification in some situations, where one could for instance
completely forget about the z (sometimes called anti-holomorphic) sector and focus on the z
(holomorphic) sector, because both sectors are regarded as independent. The global conformal
transformations would then be given by
az + b āz + b̄
z→ , z̄→ , (1.31)
cz + d c̄z + d¯
with 8 complex parameters subject to the constraints ad − bc = 1 and ād¯ − b̄c̄ = 1, thus
corresponding to the group Sl(2, C)/Z2 × Sl(2, C)/Z2 . In the complex plane C2 parameterized
by (z, z), the surface where z = z ∗ will be called the real surface, because it corresponds to
(x1 , x2 ) ∈ R2 , i.e. to the physical space. Furthermore, conformal transformations send points
from the real surface on itself if the condition

f ∗ (z) = f¯(z) (1.32)

is satisfied. Obviously, when restricted to the real surface the global conformal group reduces to
Sl(2, C)/Z2 .
Note that in obtaining (1.26), we assumed that the parameters of the transformations,
α−1 , α0 , α1 , are complex, and we forgot about the z dependance (assuming we are on the real
surface). An equivalent approach used in literature consists in keeping the z dependance, and
then restrict the transformations
X X
z→z + αn z n+1 , z→z + ᾱn z n+1 , (1.33)
n n

with real parameters αn and ᾱn to the real surface z = z ∗ . The transformations preserving the
′
real surface (z = z ∗ ⇒z ′ = z ∗ ) are generated by (ln + ¯ln ) and i(ln − ¯ln ). The global conformal
4 One recognizes easily in (1.26) the different transformations : translations, z ′ = z + b for a = d = 1 and

c = 0, rotations, z ′ = eiθ z for c = b = 0, a = d1 = eiθ/2 , dilations, z ′ = eλ z for c = b = 0, a = d1 = eλ/2 and

z
SCT, z ′ = 1+cz , for a = d = 1, b = 0
98 Detournay

group is thus generated by the 6 generators obtained by taking n = −1, 0, 1, which can be seen
to satisfy the commutation relations of the so(1, 3) real Lie algebra. It can indeed by checked
△
that ∂1 = ∂x1 = −(l1 + ¯l1 ), ∂2 = i(¯l1 − l1 ) (translations), x2 ∂2 − x2 ∂1 = −(¯l0 − l0 ) (rotations),
x1 ∂1 + x2 ∂2 = −(l0 + ¯l0 ) (dilations), the remaining two corresponding to SCT.
Another way of finding the form of the global conformal transformations consists in looking
at the finite transformations (1.8). To describe a bijective map on the Riemann sphere, the
function f (z) should not have any branch point nor any essential singularity. Indeed, around a
branch point the map is not uniquely defined5 , whereas the image of a small neighborhood of
an essential singularity under f is dense in C (Weierstrass’ theorem). Thus f is not invertible
in these cases, and the only acceptable singularities are poles. Then, f can be written as a ratio
of polynomials without common zeros6
P (z)
f (z) = . (1.34)
Q(z)
If P (z) has several distinct zeros, then the inverse image of zero is not unique and f is not
invertible. If P (z) has a multiple zero z0 of order n > 1, then the image of a small neighborhood
of z0 is wrapped n times around zero, and therefore f is not invertible, see Fig.1.

z0 + ε exp iθ
n
z P(z)=(z-z0)
ε
z0

εn exp inθ

εn

Figure 1

Thus P (z) must be of the form P (z) = az + b. The same argument holds for Q(z) when
looking at the behavior of f (z) near z = ∞. Thus
az + b
f (z) = , (1.35)
cz + d
5 Consider log z = ln ρ + iϕ, for z = ρeiϕ . If one adds 2π to ϕ, this doesn’t change z, but the imaginary part

of log z gets modified, showing that log z is a ”multivalued function” having distinct branches
6 Any meromorphic function, i.e. holomorphic except at isolated poles, can be expressed as the ratio of two

holomorphic functions. These could be exponential functions or combination thereof, but we exclude them because
they are periodic in the complex plane and thus not invertible. Furthermore, such functions are even not defined
on the whole Riemann sphere, because (x, y) = (0, ∞) and (x, y) = (∞, 0)are supposed to represent the same
point (the North Pole in the stereographic projection), while ez = ex (cos y + i sin y) is not univocal at this point
Aspects of Two-Dimensional Conformal Field Theory 99

with ad−bc 6= 0 in order for the mapping to be invertible (this is the Jacobian of the transforma-
tion). Since an overall scaling of all coefficients does not change f , the normalization ad − bc = 1
has been adopted , so that we recover the Sl(2, C) global conformal group in two dimensions.

2 Conformal Ward identities in two dimensions

2.1 Preliminaries
The objective of a quantum field theory is to determine the scattering amplitudes between
various asymptotic states (free particles). In practice these amplitudes are computed from the
correlation functions (Green’s functions) via the so-called reduction formulas. The quantum
description of a physical system may be tackled using different methods. One of them, the
operator formalism, consists in replacing classical quantities by operators acting on a vector
space in which the states of the system reside. Another method is called path integration or
functional integration, and can be related to the first one, see e.g. Appendix A of [11]. To begin,
we will work in the second formalism (the operator formalism of two dimensional CFT will be
discussed in Sect. 4), in which a euclidian correlation function of N fields φi (xi ) reads
Z
1
hφ1 (x1) · · · φN (xN )i = DΦ φ1 (x1) · · · φN (xN ) exp(−S[φ]) , (2.1)
Z
R
where Z = DΦ exp(−S[φ]), R Φ denotes the set of all functionally independent fields in the
theory7 , and S[Φ, ∂µ Φ] = d2 x L(Φ(x), ∂µ Φ(x)) is the action.
Classically, the invariance of the action under a continuous symmetry8 implies the existence
of a conserved current via Noether’s theorem. At the quantum level, a continuous symmetry of
the action and of the functional measure leads to constraints on the correlation functions, which
are expressed via the so-called Ward identities.
Before turning to their derivation in the context of 2D CFT, we define the notion of a primary
field. Under a conformal map z→w(z), z→w(z), a primary field transforms as
 −h  −h̄
dw dw
φ(z, z)−→φ′ (w(z), w(z)) = φ(z, z) , (2.2)
dz dz

where h (resp. h̄) is called holomorphic (resp. anti-holomorphic) conformal dimension of the
field. Eq. (2.2) states that a primary field of conformal dimensions (h, h̄) transforms as the
components of a tensor with h covariant indices z and h̄ covariant indices z. The conformal
dimensions are related to the spin s and scaling dimension ∆ of the field by s = h − h̄ and
∆ = h + h̄ 9 .
These fields play a crucial rôle in conformal field theories, namely because correlation func-
tions involving any field of the theory can be reduced to correlations functions involving only
primary fields. This is one of the properties rendering conformal field theories in two dimensions
solvable, where a theory is said to be solved when all correlation function can be written (at least
7 When one speaks of a field in CFT, it does not necessarily mean that this field figures independently in the

functional integration measure. For instance, a single boson φ, its derivative ∂µ φ and a composite quantity such
as the energy-momentum tensor are called fields, since they are local quantities with a coordinate dependence.
However, only some fields are integrated over in the functional integral
8 This means that under a transformation x→x′ , φ(x)→φ′ (x′ ), we have S ′ = S, with S ′ =
R R
d d x′ L(Φ′ (x′ ), ∂ ′ Φ′ (x′ )) = dd x L(Φ′ (x), ∂ Φ′ (x)), where D is the integration domain (often D = R2 )
D′ µ D µ
9 The variation of a field under a transformation can be decomposed in δφ(x) = (φ′ (x)−φ′ (x′ ))+(φ′ (x′ )−φ(x)),

where the second term encodes the spin and scaling properties of the field. For a rotation, z→eiθ z, we find
φ′ (x′ ) = eiθ(h−h̄) φ(x), while under a scaling z→eλ z, we have φ′ (x′ ) = e−λ(h+h̄) φ(x)
100 Detournay

in principle). Fields transforming like (2.2), but only under global conformal transformations
(1.31) are called quasi-primary.

2.2 A derivation of the conformal Ward identities

I first would like to mention that there are quite a lot seemingly different derivations of the
conformal Ward identity. In this section, I will present that of [1], which is to my opinion the
shortest way to get to the goal and gives a good flavor of how things fit all together. Skeptical
readers may wish to consult another derivation, maybe more precise, based on [2] and presented
in Appendix A (see also [2] p123 for an alternative approach to global conformal Ward identities).
Later on, I will return to these identities in the context of the operator formalism (see Sect. 4.4
and 4.7).
To study the consequences of the conformal symmetry on correlation functions, it is usually
more convenient to consider infinitesimal transformations of the form (1.8), for small α(z). With
w(z) = z + α(z), w(z) = z + ᾱ(z), one finds the infinitesimal transformation of a primary field:

δα φ(z, z) = −(hα′ (z) + α(z)∂ + h̄ᾱ′ (z) + ᾱ(z)∂) φ(z, z) . (2.3)

At this point, I would like to draw attention on the infinitesimal character of the transformation
we are to use. As we found, general conformal transformations are given by (1.8), with f (z) and
α(z) depending only upon the variable z (I will sometimes omit the z dependence, being under-
stood it is always present). This makes f and α holomorphic functions, except at the location
of possible singularities. Now, it is clear that close to a singularity, the functions cannot remain
infinitesimal, whatever the parameters αn in (1.12). Furthermore, on the Riemann sphere, even
globally holomorphic (analytic) functions (admitting everywhere a Taylor expansion), like poly-
nomials, are unbounded, except the constant function (for another discussion on this point and
another way round, see [4], p129).
To avoid this problem, we will define transformation (1.8) in a finite region D, say in the
neighborhood of z = 0, with α(z) analytic in D, and put α(z) = 0 outside D. This way of
cutting α(z) will produce boundary terms which we will have to keep track of. The region D
will be chosen so to as to contain the positions of all the fields inside the correlation function,
see Fig. 2.
An analytic function in a region D containing the origin may be Taylor expanded around
z = 0 as
X∞
α(z) = αn z n+1 . (2.4)
n=−1
n+1
Because |z| is bounded in D, the coefficients {αn } can always be chosen sufficiently small to
render α(z) infinitesimal. These coefficients are the parameters of the corresponding conformal
transformation, and their number is infinite expressing the infinite-dimensional character of the
symmetry.
Let us consider a variation affecting both coordinates and fields10 of the form

x → x′ , (2.5)
φ(x) → φ′ (x′ ) . (2.6)

In the functional integral (2.1), let us make the following change of functional integration variable
:
φ(z, z)−→φ′ (z, z) = φ(z, z) + δφ(z, z) . (2.7)
10 We consider the variation of all fields, even those not appearing in the functional integration
Aspects of Two-Dimensional Conformal Field Theory 101

x1 x2

.
.
.
xn C

Figure 2

We will assume that the functional measure is invariant under conformal transformations , that
is, that the Jacobian of this change of variable is equal to 1. This cannot be checked formally,
but only on specific theories, so we will take this as a definition of a conformally invariant filed
theory. With this assumption, Dφ = Dφ′ , and replacing (2.7) in (2.1) does not change the
correlation function. With X = φ(x1 ) · · · φ(xN ) (xi = (zi , z i )), we find that
Z
1 ′
hXi = DΦ′ φ′ (x1 ) · · · φ′ (xN ) e−S[Φ ] (2.8)
Z
Z
1
= DΦ′ (φ1 (x1 ) + δφ1 (x1 )) · · · (φN (xN ) + δφN (xN )) e−S[Φ]−δS[φ] (2.9)
Z
Z Z
1 1
= hXi + DΦ δXe−S[Φ] − DΦ XδS[Φ]e−S[Φ] , (2.10)
Z Z
P
where δX = i (φ(x1 ) · · · δφ(xi ) · · · φ(xN ) ). Thus, we have
△
δhXi = hδXi = hδS Xi . (2.11)
When all fields in the correlation function are primary, we may use (2.3), so that the r.h.s. of
(2.11) becomes
N
X
δα,ᾱ hXi = − [hi α′ (zi ) + α(zi )∂i + h̄i ᾱ′ (z i ) + ᾱ(z i )∂ i ]hXi , (2.12)
i=1

assuming that all points xi in the correlation function are distinct11 . The l.h.s. can be rewritten
11 Correlation functions typically become potentially singular when two or more points come to coincide, so the

operation of ”taking the derivative out of the correlation function” , h· · · ∂i φ(xi ) · · · i = ∂i h· · · φ(xi ) · · · i may pose
problems in this case. We will always assume xi 6= xj for i 6= j.
102 Detournay

by using the relation I

δS = dsµ αν Tµν , (2.13)
C
where the contour C is the boundary of D, and Tµν is the canonical energy momentum tensor12
∂L
Tµν (φ) = δµν L(φ, ∂µ φ) −
∂ν φ , (2.14)
∂φ,µ
with φ,µ = ∂µ φ. This is the conserved current associated with translation invariance of the
action. Eq. (2.13) expresses that, although the action S is supposed to be conformally invariant,
we find a boundary term because α(z) is discontinuous on C.
Let us derive (2.13), in the case where αν = aν = cst. Let S[φ] be an action given by
Z
S[φ] = d2 x L(φ(x), φ,µ (x)) , (2.15)

where the lagrangian L is supposed to be local in φ(x) and φ,µ (x), as well as invariant under
translations. This implies that
Z Z
2 ′ ′
δS[φ] = d xL(φ (x), φ,µ (x)) − d2 xL(φ(x), φ,µ (x)) = 0 , (2.16)

where the integration is over the whole space and φ′ (x) = φ(x) − aµ ∂µ φ(x) = φ(x − a). Now,
when restricted to a finite region D, the transformation of the coordinates reads
xµ → x′µ = xµ + aµ (x) , (2.17)
with
(
µ aµ if x ∈ D ,
a (x) = (2.18)
0 otherwise
and the variation of the field is
(
′ φ(x − a) if x ∈ D ,
φ (x) = (2.19)
φ(x) otherwise
A one-dimensional section of φ′ (x) is given in Fig.3
The variation of the action now becomes
Z Z
2 ′ ′
δS[φ] = d xL(φ (x), φ,µ (x)) − d2 xL(φ(x), φ,µ (x)) , (2.20)
D D

where only the contribution of the part inside D appears. By restricting ourselves to a finite
region, we broke translation invariance, in some sense. There are two contributions to δS[φ] in
(2.20). To see this, we rewrite it as
Z     
∂L ∂L
δS[φ] = d2 x δφ + ∂µ δφ (2.21)
∂φ ∂φ,µ
D
Z     Z  
2 ∂L ∂L 2 ∂L
= d x − ∂µ δφ + d x ∂µ δφ (2.22)
∂φ ∂φ,µ ∂φ,µ
D D
△
= δSreg [φ] + δSjump [φ] . (2.23)
12 From now one, we consider for simplicity the situation where Φ reduces to a single field φ
Aspects of Two-Dimensional Conformal Field Theory 103

φ'(x) = φ(x-a)

φ'(x) = φ(x)

aµ

x
D

Figure 3

The first term corresponds to a ”bulk term”, while the second one can be rewritten using Stokes’
theorem as a boundary term, whose contribution comes from the ”jump” that φ experiences on
∂D = C :   I  
Z
△ 2 ∂L ∂L
δSjump [φ] = d x ∂µ δφ = δφ dsµ . (2.24)
∂φ,µ ∂φ,µ
D C

Thus, with δφ = −aν ∂ν φ ,we get

I  
∂L
δSjump [φ] = − aν ∂ν φ dsµ . (2.25)
∂φ,µ
C

The analysis of the regular term is better performed starting directly from (2.20) and performing
the change of variable x̃ = x − a in the first integral. So, the function to integrate becomes the
same, the difference being the integration domains :
Z Z
δSreg [φ] = d2 x̃L(φ(x̃), φ,µ (x̃)) − d2 xL(φ(x), φ,µ (x)) (2.26)
D̃ D
Z
= d2 xL(φ(x), φ,µ (x)) , (2.27)
δD

where the domain δD is represented in Fig.4 .

Because aµ is infinitesimal, this integral may be transformed in a contour integral (on C or
C̃, to first order), with
Z I I I
2 µ ν µ
d x= ǫµν a dx = ds aµ = dsµ aν δµν . (2.28)
δD C C C
104 Detournay

x2

~
C δD

a
µ C

dx
µ x1

δD µ
ds

Figure 4

Finally, with (2.25) ,(2.26) and (2.28), we find

I    
∂L
δS[φ] = dsµ aν δµν L − ∂ν φ , (2.29)
∂φ,µ
C

and hence Eq. (2.13). The same argument may still be applied if αµ is replaced by a conformal
Killing vector αµ (x). Eq. (2.11) becomes, using (2.13)
I
1
δhXi = − dsµ αν hTµν Xi , (2.30)
2π
C

1
where the factor of − 2π is placed for convenience and corresponds to a normalization choice
for Tµν or the action. Let us now express (2.30) in complex coordinates. With z = x1 + ix2 ,
△
z = x1 −ix2 and (dxµ ) = (dx1 , dx2 ), (dsµ = ǫµν dxν ) = (dx2 , −dx1 ), we find that dsz = ds = −idz
△
and dsz = ds̄ = idz, as well as αz = α(z), αz (z) (with the help of the conformal Killing equations
(1.11)), so that (2.30) becomes

I
1
δα,ᾱ hXi = − dw [α(w)hTzz (w, w)X i + ᾱ(w)hTzz (w, w)Xi]
2πi
C
I
1
+ dw [ᾱ(w)hTzz (w, w)X i + α(w)hTzz (w, w)Xi] . (2.31)
2πi
C

This is a first version of the ward identity, in integral form. It is possible to go a step further
to obtain it in local form. We are first going to simplify (2.31) by analyzing two special cases,
that is (1) α(z) = a = cst. and (2) α(z) = bz, and by applying (2.31) to an arbitrary domain
D∗ which does not contains the points xi (see Fig 5).
Aspects of Two-Dimensional Conformal Field Theory 105

D*
x1 x2

.
.
.
C* xn

Figure 5

In this way, the l.h.s. vanishes, because now δφ(xi ) = 0, as everywhere outside the original
domain D. With the first case, we find
I I
0=− dwhTzz (w, w)X i + dwhTzz (w, w)Xi , (2.32)
C∗ C∗
R R
The two contour integrals may be rewritten using Stokes’s theorem M d2 x∂µ F µ = ∂M dsµ Fµ ,
which implies in particular
Z I Z I
1 1
d2 x ∂f (z, z) = − dzf (z, z) and d2 x ∂f (z, z) = dzf (z, z) . (2.33)
2i 2i
D C D C

Eq. (2.32) can then be rewritten

Z
d2 x [∂w hTzz (w, w)X i + ∂w hTzz (w, w)X i] = 0 . (2.34)
D∗

Because this holds for any domain D∗ (excluding the points xi ), this implies
∂w hTzz (w, w)X i + ∂w hTzz (w, w)X i = 0 , (w, w) 6= xi . (2.35)
With the second case, we find
I I
0=− dw w hTzz (w, w)X i + dw w hTzz (w, w)X i , (2.36)
C∗ C∗

and by using Stokes’ theorem

Z
d2 x [w∂w hTzz (w, w)X i + w∂w hTzz (w, w)X i + hTzz (w, w)X i] , (2.37)
D∗
106 Detournay

where the first two terms of the integrand vanish because of (2.35). Thus (2.37) implies

hTzz (w, w)X i = 0 , (w, w) 6= xi . (2.38)

From (2.34) and (2.38), we find that

∂w hTzz (w, w)X i = 0 , (w, w) 6= xi . (2.39)

Similar relations hold for hTzz (w, w)X i and ∂w hTzz (w, w)X i, so that we end up with

∂w hTzz (w, w)X i = ∂w hTzz (w, w)X i = 0 if (w, w) 6= (wi , wi ) (2.40)

and
hTzz (w, w)X i = hTzz (w, w)X i = 0 if (w, w) 6= (wi , wi ) . (2.41)
Eq. (2.40) shows that inside correlation functions Tzz (resp. Tzz ) does not depend on z (resp.
z), and is said to be an holomorphic operator (resp. anti-holomorphic operator ), except at the
points xi . We set
△ △
Tzz = T (z) and Tzz = T̄ (z) . (2.42)
Furthermore, Tµµ = g µν Tµν = T11 + T22 = 12 (Tzz + Tzz ), so (2.41) expresses the vanishing
of the trace of the energy-momentum tensor inside a correlation function except at coincident
points1314 Finally, eq. (2.31) reduces, with (2.40), (2.41) and (2.42), to15
I I
1 1
δα,ᾱ hXi = − dwα(w)hT (w)X i + dwᾱ(w)hT̄ (w)X i . (2.43)
2πi 2πi
C C

Because of the form of (2.12) and (2.43), we observe that (2.43) splits into two independent
equations (recall we may consider z and z as two independent variables, as well as α(z) and ᾱ(z)
as two independent functions). Let us focus on the z-part (”holomorphic” part), and consider a
correlation function involving primary fields :
I X
1
− dwα(w)hT (w)X i = − (α′ (zi )hi + α(zi )∂i )hXi . (2.44)
2πi i
C

Using the residue theorem16 , the r.h.s. may be rewritten as

N I  
1 X hi 1
− dwα(w) + ∂ i hXi , (2.45)
2πi i=1 (w − zi )2 w − zi
Ci

because α(w) is analytic in D and in particular in any of the contours Ci represented in Fig.6 .
13 Classically, the canonical energy-momentum tensor of a system invariant under rotations can be made sym-

metric on-shell, and put in the so-called Belinfante form (see [2], p47). An alternative definition of the energy-
momentum tensor is given by the functional derivative of the action w.r.t. the metric, evaluated in flat space :
T µν = δgδSν , which makes it identically symmetric. For a system invariant under dilations, it can be shown that
µ
in d > 2 and assuming a technical (”virial”) condition that the canonical energy-momentum tensor can be made
symmetric on-shell ( [2], p102, [4] p122). In d = 2, this is assumed to hold.
14 The tracelessness of the energy-momentum tensor is sufficient to guarantee the invariance of the action under
R
conformal transformations . Indeed, under xµR→xµ + εµ , the action varies as δS = dd x T µν ∂µ εν . If εν is a
d µ ρ µ
conformal Killing vector, then (see 1.11) δS = d x Tµ ∂ ερ , which vanishes if Tµ = 0. The converse is not true,
since ∂ ρ ερ is not an arbitrary function.
15 We may use (2.40) and (2.41) because the contour C does not pass exactly through the points x
i
H φ(z) (n−1)
16
C dz (z−z0 )n
= 2πi φ(n−1)! , if φ(z) is analytic in C
Aspects of Two-Dimensional Conformal Field Theory 107

C2
C1

x1 x2

.
.
Cn
.
xn

Figure 6

For the same reason (and because hXi does not depend on z!), the contours Ci may be
deformed into the contour C, see Fig.2 . Eq. (2.44) then becomes
I " N   #
X hi 1
dwα(w) hT (w)Xi − + ∂i hXi = 0 . (2.46)
i=1
(w − zi )2 w − zi
C
H
This is an equation of the form C dz α(z)f (z) = 0, for all analytical function α(z). Thus f (z)
has to be holomorphic inside C and may not contain singularities. Therefore we write
XN  
hi 1
hT (w)Xi = + ∂ i hXi + reg , (2.47)
i=1
(w − zi )2 w − zi

where ”reg” stands for a holomorphic function of z inside C, regular in particular as w→zi .

3 Operator product expansion and conformal transforma-

tion of fields
The main result of the preceding section is encapsulated in (2.47) (when X contains only primary
fields) and in (2.43), which are called the Ward identities of two-dimensional conformal field
theories. These will be of great importance in what follows. We saw that the z and z parts
of the transformations split in two parts, so in the following we will only consider the z part,
keeping in mind that the z part is always present.
Eq. (2.47) delivers an important information. It yields the singular behavior of the corre-
lator of the field T (z) with the primary fields φi (zi , z i ), as z→zi . For a single primary field of
holomorphic conformal dimension h, this is usually written as
h 1
T (z)φ(w, w) ∼ φ(w, w) + ∂w φ(w, w) . (3.1)
(z − w)2 z−w
108 Detournay

The precise meaning of this expression is the following : when z→w, the (correlation) function
h 1
hT (z)φ(w, w)Y i behaves as the function (z−w) 2 hφ(w, w)Y i + z−w ∂w hφ(w, w)Y i, where

Y = φ1 (w1 , w1 ) · · · φn (wn , wn ), none of the points being coincident in hT (z)φ(w, w)Y i. An

expression like (3.1) is called (the singular part of) the operator product expansion (or OPE ) of
the energy-momentum tensor with a primary field. As we stressed, it only makes sense inside a
correlation function. The symbol ” ∼ ” means equality up to terms which are regular as z→w.
A general OPE of two fields takes the form

XN N
{AB}n (w) X {AB}n (w)
A(z)B(w) = ∼ , (3.2)
n=−∞
(z − w)n n=1
(z − w)n

where the composite fields {AB}n (w) are non-singular at w = z. For instance, {T φ}1 (w) =
∂w φ(w). Let us now return to eq. (2.43), focusing on the z part :
I
1
δα hXi = − dw α(w)hT (w)Xi , (3.3)
2πi
C

where δα hXi stands for h(δα φ1 )φ2 · · · φN i + · · · + hφ1 · · · (δα φN )i as before. This Ward identity
can be interpreted as defining the transformation of the fields within the correlation function.
The variation of the function on the l.h.s. is by definition that produced by the variation of the
operators. Now remember the contour C encircles the positions of all operators. This contour
may be ”distributed” on the operators φ1 , · · · , φN as in Fig.6, because the only singular points
of hT (w)Xi are located at w = zi . Thus
N
X
△
δα hφ1 · · · φN i = hφ1 · · · δα(zk ) φk (zk ) · · · φN i (3.4)
k=1
N
X I
1
= dw α(w)hT (w)φ1 · · · φN i . (3.5)
2πi
k=1 Ck

Each integral in (3.5) corresponds to the variation of a field in (3.4). So, the variation of a single
field is represented by the integral
I
1
δα(zi ) φ(zi ) = dw α(w)T (w)φi (zi ) , (3.6)
2πi
Ci

where Ci encircles only zi , see Fig.6. Consequently, because α(z) is analytic inside Ci , we
conclude that the transformation of a field under z→z + α(z) is completely determined by the
singular terms of the OPE of this field with the holomorphic part T (z) of the energy-momentum
tensor, and vice-versa. This explains the important rôle of OPE’s in conformal field theories.
We now concentrate on the transformation properties of the holomorphic energy-momentum
tensor, T (z). From its definition, it seems natural that it should transform as a rank 2 covariant
tensor,
 −2
dw
T (z)−→T ′ (w) = T (z) (3.7)
dz
under z→w(z), thus as a primary field of conformal weight h = 2, h̄ = 0 (see eq.(2.2)) (similarly,
T̄ (z) should have h = 0, h̄ = 2. Actually, it turns out, in passing to the quantum theory, that
Aspects of Two-Dimensional Conformal Field Theory 109

the situation is complicated by the addition of a term in the transformation law (3.7), so that
the latter becomes in general
 −2 h i
′ dw c
T (z)−→T (w) = T (z) − {w; z} , (3.8)
dz 12

where cis a numerical constant characterizing the theory and where

 2
d3 w/dz 3
△ 3 d2 w/dz 2
{w; z} = − (3.9)
dw/dz 2 dw/dz

is called the Schwartzian derivative. The additional term in (3.8) is called Schwinger term and
is related, as we will see in Sect. 4.2, to the conformal anomaly. It can be shown (see for
instance [4], pp134-141) that, requiring that (3.8) defines a group law and that (3.9) be a local
functional (i.e. containing derivatives of w(z) only up to finite order and degree), the form of
c 17
the functional (3.9) is completely fixed up to a constant, hence the term 12 . Notice that the
Schwartzian derivative of the global conformal map
az + b
w(z) = , ad − bc = 1 (3.10)
cz + d
vanishes, making T (z) a quasi-primary field of conformal weight h = 2. The infinitesimal
variation δα T (z) = T ′ (z) − T (z) can be obtained from (3.9) by considering {z + α(z), z} to first
order in α :  2 2
∂3α 3 ∂ α
{z + α(z), z} = − ∼ ∂3α (3.11)
1 + ∂α 2 1 + ∂α
then (3.8) gives
h c i
T ′ (z + α) = (1 − 2∂α(z)) T (z) − ∂ 3 α
12
= T ′ (z) + α(z)∂T (z) , (3.12)

hence
c 3
δα T (z) = T ′ (z) − T (z) = −
∂ α − 2∂α(z)T (z) − α(z)∂T (z) . (3.13)
12
An alternative way of proceeding is to consider the OPE of T (z) with itself which, according
to the discussion of preceding section and eq. (3.6), should encode (3.13). It again turns out
that, for a general CFT, this OPE has the form
c/2 2T (w) ∂T (w)
T (z)T (w) ∼ 4
+ 2
+ . (3.14)
(z − w) (z − w) z−w

Given a specific CFT, (3.14) is generally easier to obtain by standard techniques than (3.13),
as we will see on some particular examples in the exercises (see also [2] pp128-135, p138). The
last two terms of (3.14) again reflect the fact that T (z) is a quasi-primary field of conformal
17 The group property to be verified is that the result of two successive transformations z→w→u should coincide
 −2  
du c
with what is obtained from the single transformation z→u, that is T ′′ (u) = dw
T ′ (w) − 12
{u; w} =
 −2  −2   −2 
du dw c c ! du c

dw dz
(T (z) − 12
{w; z}) − 12
{u; w} = dz
T (z) − 12
{u; z} . The last equality requires the
 −2
dw
following relation between the Schwartzian derivatives : {u; z} = {w; z} + dz
{u; w}, which is indeed
satisfied.
110 Detournay

c/2
dimension h = 2 (see (3.1)). The term (z−w) 4 is actually the only addition compatible with

the scaling transformation of (3.14). Let us see that (3.13) and (3.14) really expresses the same
thing. The relation between the variation of a field and the OPE is given by (3.6). By plugging
(3.14) in (3.6), we get
I
1
δα T (z) = − dw α(w)T (w)T (z)
2πi
Cz
I  
1 c/2 2T (z) ∂T (z)
= − dw α(w) + + , (3.15)
2πi (w − z)4 (w − z)2 w−z
Cz

so that after performing the integration, we effectively recover (3.13).

4 Operator formalism
Throughout the previous sections, all our manipulations were assumed to hold inside correlation
functions. The consequences of conformal symmetry on two-dimensional field theories were
embodied in constraints imposed on these correlation functions via the Ward identities (2.43)
(we will elaborate on this point in Sect 4.4). These Ward identities were most easily expressed
in the form of an OPE of the energy-momentum tensor with local fields (see (3.1) and (3.14)).
Up to now, we only used the path-integral representation of the theory in which all correlation
functions could in principle be obtained. We would now like to give an operator interpretation
in terms of states in a Hilbert space.

4.1 Radial quantization

The operator formalism distinguishes a time direction from a space direction. This is natural in
Minkowski space-time, but somewhat arbitrary in euclidian space. This allows to choose as time
direction the radial direction form the origin, the space direction being orthogonal to it. This
choice leads to the so-called radial quantization of two-dimensional conformal field theories.
We may start from a two-dimensional Minkowski space with coordinates t and σ. One usually
takes the space direction σ to be periodic, σ ∈ [0, L], defining this way the theory on a cylinder18 .
We continue to euclidian space, t = −iτ , and then perform the conformal transformation
2π 2π
z = e(τ +iσ) L , z = e(τ −iσ) L , (4.1)

which maps the cylinder onto the complex plane C ∪ {∞}, topologically a sphere (the Riemann
sphere). Surfaces of equal euclidian time τ on the cylinder will become circles of equal radii on the
complex plane. This means that the infinite past (τ = −∞) gets mapped onto the origin of the
plane (z = 0) and the infinite future becomes z = ∞, see Fig.7. Time reversal becomes z→1/z ∗
(= e−τ +iσ ) on the complex plane, and parity is z→z ∗ . The operator (l0 + ¯l0 ) (preserving the real
surface, see below eq. (1.33)) generates dilations in the complex plane : z→eλ z,z→eλ z which,
according to (4.1), corresponds to (euclidian) time translations on the cylinder. Consequently,
this operator represents the hamiltonian of our system (since we are working in euclidian space,
a more appropriate name would be transfer operator, which upon Wick rotation becomes the
18 This is natural from the closed-string theory point of view, where the two-dimensional space describes the

worldsheet of the string. Also, this procedure allows to avoid infrared divergences, e.g. as the one present in the
vacuum functional of a free boson (see [2], p142). L can be chosen very large and taken to infinity at the end of
the calculation.
Aspects of Two-Dimensional Conformal Field Theory 111

|z| = cst.
z = exp (τ+iσ)2π/L
τ

τ=−

8
σ

Figure 7

hamiltonian; similarly, the exponential of the transfer operator gives the transfer matrix, which
would become the time evolution operator upon Wick rotation). Finally, an integral over the
space direction σ, at fixed τ , will become a contour integral on the complex plane. This enables
us to use all the powerful techniques of complex analysis. We would now like to translate a
relation like (3.3) : I
1
δα(z) hXi = − dw α(w)hT (w)Xi
2πi
C
in the operator formalism language. Expressions like hφ(z1 , z 1 )φ(z2 , z 2 )i which were understood
in the path integral formalism (see (2.1)) must now be interpreted as
h 0 |T (φ(τ1 , σ1 )φ(τ2 , σ2 )| 0 i , (4.2)
where | 0 i is the vacuum state of the Hilbert space of states H (see Sect. 4.3) and where the
fields become operators acting on H. As is known (see Appendix B for a reminder), in ordinary
QFT insertion of operators in the path integral represents time-ordered matrix elements of the
corresponding operators. The time-ordering, defined by
(
φ(τ1 , σ1 )φ(τ2 , σ2 ) if τ1 > τ2 ,
T [φ(τ1 , σ1 )φ(τ2 , σ2 )] = (4.3)
εφ(τ2 , σ2 )φ(τ1 , σ1 ) if τ2 > τ1

where ε = 1 for bosons and −1 for fermions, appears naturally by the nature of the path
integral, which builds amplitudes from successive time steps. In radial quantization, this notion
obviously generalizes to the radial-ordering
(
φ(z1 , z 1 )φ(z2 , z 2 ) if |z1 | > |z2 | ,
R [φ(z1 , z 1 )φ(z2 , z 2 )] = (4.4)
εφ(z2 , z 2 )φ(z1 , z 1 ) if |z2 | > |z1 |

Since all fields within correlation functions must be radially ordered, so must be the l.h.s. of
any OPE if it is to have an operator meaning. In particular, all OPE’s written previously have
an operator meaning if |z| > |w|.
112 Detournay

Recall that in (3.6), C is a contour encircling all points zi appearing in X = φ1 · · · φN . Let

us consider the variation of a single operator, X = φ(z, z) :
I
1
δα(z) φ(z, z) = − dw α(w)R [T (w)φ(z, z)] (4.5)
2πi
C

where C is represented in Fig.8 . We will forget about the z dependence in what follows. As we

w C

Figure 8

already pointed out, it is expected that a correlation function involving the product of operators
T (w)φ(z) will potentially exhibit a singularity as w→z (see e.g. (3.1)). We will assume that
w = z is the only singularity in the product. Because of the radial ordering, the integrand of
(4.5) is to be changed along the path C. To avoid this, we deform C in C ′ , as shown in Fig. 9.
This can be done safely, because there are no other singularities. The integral can further be
rewritten as I I I I I
= = − − , (4.6)
C C′ C1 C2 Cε

see Fig.10, and where the last integral vanishes because the contour Cε does not contain any
singularity. Thus, (4.5) can be rewritten
 
I I
1 1
δα(z) φ(z) = −  dw α(w)T (w)φ(z) − dw α(w)φ(z)T (w) (4.7)
2πi 2πi
C1 C2
△
= −[Qα , φ(z)] , (4.8)

where I
1
△
Qα = dw α(w)T (w) . (4.9)
2πi
Aspects of Two-Dimensional Conformal Field Theory 113

Cε

C'

C' + Cε = C1 - C2

Figure 9

The integrals over C1 and C2 correspond to integrals over the space direction, at fixed time, and
there is no ordering ambiguity in them. Note that to allow an arbitrary number of fields to lie
beside φ(z) (recall an operator relation involving a field φ is obtained by considering a string of
the form X = φφ1 · · · φN ), the contours C1 and C2 have to be chosen appropriately. Indeed, the
decomposition into two contours is valid if φ(z) is the only field of the string having a singular
OPE with T (w) between the two circles C1 and C2 . Therefore these circles are taken to have
radii respectively equal to |z| + ε and |z| − ε, with ε→0. The commutator (4.8) may then be
seen as an equal time commutator 19 .
The operator (4.9) is called conformal charge and represents the generator of conformal
transformations z−→z + α(z). This situation is familiar from QFT. We know in this context
from Noether’s theorem that to every continuous symmetry of the action one may associate a
current jaµ (where ”a” stands for a collection of indices coming from the set {ωa } of infinitesimal
parameters of the symmetry transformation) that is classically conserved :

∂µ jaµ ≈ 0 (4.10)

where ≈ stands for equality when the equations of motion hold (on-shell). One may define the
conserved charge associated with jaµ , as the integral over the spatial dimensions of the time
component of the current :
Z
Qa = dd−1 x ja0 , (4.11)

whose time derivative vanishes : Q̇a = 0 (provided the spatial components of the current vanish
sufficiently rapidly at infinity). This charge is also seen to generate the infinitesimal symmetry
19 It
seems to me that this construction does not work if there is a point zi in the correlator such that |z| = |zi |
because then, we cannot avoid it yo lie between to two contours
114 Detournay

C1

z
C2

Figure 10

transformations in the operator formalism :

[Qa , φ] = Ga φ , (4.12)

where the generators Ga are defined in (1.13). This justifies the appelation ”conformal charge”
for Qα as well as the usual expression ”the energy-momentum tensor T (z) generates conformal
transformations”.
More generally, if a(w) and b(z) denote two holomorphic fields (i.e. depending only upon
one complex coordinate), then
I I I
dw a(w)b(z) = dw a(w)b(z) − dw b(z)a(w) (4.13)
Cz C1 C2
△
= [A, b(z)] , (4.14)

where the contours

H are as in Fig.10, and the operator A is the integral over space at fixed time
of a(w) : A = a(w)dw (compare to (4.9)). In the l.h.s. of (4.13), the radial-ordering symbol
R is understood, as everywhere where products of fields are involved. The commutator of two
operators, each the integral of a holomorphic field, is obtained by integrating (4.14) over z :
I I
[A, B] = dz dw a(w)b(z) , (4.15)
C0 Cz
H H
where A = a(w)dw and B = b(w)dw. This commutator can also be seen as
 
I I I I
[A, B] =  dw dz − dz dw R[b(z)a(w)] . (4.16)
C1 C2 C1 C2
Aspects of Two-Dimensional Conformal Field Theory 115

Indeed, the difference of contours can be deformed as shown in Fig.11, by fixing for instance
z and then deforming the contour integral on w as in Fig.10 . We then recover the same

w z w
C1 C1
z w z

C2 C2

- = O

w C1

because = C1 - C2
C2

Figure 11

integrations as in (4.15). We will extensively use these relations in the next section. Notice that
we have similar
H relations for operators defined by a contour integration around an arbitrary point
: A(z) = z a(w)dw. This integral does not represent a fixed-time integral, but the commutation
of two such operators can still be defined through (4.15) and (4.16) by replacing the point z = 0
by an arbitrary z.
From now on, whenever a contour integral is written without a specific contour, it is under-
stood that we integrate at fixed time, i.e. along a circle centered at the origin. Otherwise the
relevant points surrounded by the contours are indicated below the integral sign.
As a conclusion, formulas (4.14) and (4.15) are important because they allow to relate OPE’s
to commutation relations, as we will see in the next section.

4.2 Mode expansions and Virasoro algebra

A conformal field (primary or quasi-primary) of dimensions (h, h̄) may be mode expanded as
follows: X
φ(z, z) = z −m−h z −n−h̄ φm,n , (4.17)
m,n∈Z

with I I
1 m+h−1 1
φm,n = dz z φ(z, z) dz z m+h̄−1 φ(z, z) . (4.18)
2πi 2πi
With this definition, the modes transform under scaling z→λz as φn →φ′n = λn φn , thus having
scaling dimension n. Eq. (4.17) corresponds to a Laurent expansion around z = 0. Later on we
116 Detournay

will use a development around an arbitrary point w:

X
φ(z, z) = (z − w)−m−h (z − w)−n−h̄ φm,n (w, w) , (4.19)
m,n∈Z

with
I I
1 m+h−1 1
φm,n (w, w) = dz (z − w) φ(z, z) dz (z − w)m+h̄−1 φ(z, z) . (4.20)
2πi 2πi

△
We set hereafter φm,n (0, 0) = φm,n .
For the energy-momentum tensor, we have

X X
T (z) = z −n−2 Ln , T̄ (z) = z −n−2 L̄n , (4.21)
n∈Z n∈Z

where the modes are given by the contour integrals

I I
1 1
Ln = dz z n+1 T (z), L̄n = dz z m+1 T̄ (z) . (4.22)
2πi 2πi

P
Considering an infinitesimal conformal transformation z→z + α(z), with α(z) = n z n+1 αn ,
the conformal charge (4.9) can be reexpressed as

I
1 X X
Qα = dw αn wn+1 T (w) = αn Ln . (4.23)
2πi n n

The variation of a field then reads with (4.8)

X
δα(z) φ(z) = − αn [Ln , φ(z)] . (4.24)
n

△
In particular , for a transformation z→z + αn z n+1 = z + αn (z), we have

δαn (z) φ(z) = −αn [Ln , φ(z)] . (4.25)

The mode operators Ln (and similarly L̄n , which we will omit) are the generators of the
local conformal transformations on the Hilbert space, exactly like ln (and ¯ln ) (see (1.15)) are the
generators of conformal mappings on the space of functions20 . Using (4.14), the algebra they

20 Recall the Ln ’s generate the conformal transformations of the operators φ

Aspects of Two-Dimensional Conformal Field Theory 117

satisfy may be derived, with the help of the OPE (3.14) :

I I
1
[Ln , Lm ] = dw dz z n+1 wm+1 T (z)T (w)
(2πi)2
0 w
I I  
1 n+1 m+1 c/2 2T (w) ∂T (w)
= dw dz z w + +
(2πi)2 (z − w)4 (z − w)2 z−w
0 w
I  
1 m+1 c/2 3 n+1 n+1 n+1
= dw w ∂ (z ) + 2T (w)∂(z ) + ∂T (w)z
2πi 3! |z=w
0
I h i
1 c m+n−1
= dw w n(n2 − 1) + 2T (w)(n + 1)wm+n+1 + wm+n+2 ∂T (w)
2πi 12
0
2 c
= n(n − 1) δm+n,0 + 2(n + 1)Lm+n
 12 
I I
1 
+ dw∂(wm+n+2 T (w)) − dwT (w)(m + n + 2)wm+n+1  , (4.26)
2πi
0 0

hence
c
[Ln , Lm ] = n(n2 − 1)δm+n,0 + (n − m)Lm+n . (4.27)
12
A similar relation holds for the L̄n ’s and [Ln , L̄m ] = 0. This is the Virasoro algebra. It can
be seen as a central extension of the Witt algebra (1.16). It is clear that the central term
proportional to c is due to the presence of the Schwinger term in (3.8) and (3.14). We note that
the generators L−1 ,L0 and L1 form a Sl(2, C) subalgebra identical to that generated by the ln ’s
for n = −1, 0, 1 :
[L−1 , L0 ] = −L−1 , [L1 , L0 ] = L1 , [L1 , L−1 ] = 2L0 . (4.28)
Thus, only the global subgroup Sl(2, C) is not affected by the conformal anomaly. This will be
reflected when returning to the Ward identities in Sect. 4.4, where we will see that the correlation
functions are invariant only under the global conformal group.

4.3 The Hilbert space and representations of the Virasoro algebra

To describe the Hilbert space H, we will use the standard formalism of ”in” and ”out” states of
quantum field theory. We first have to assume the existence of a vacuum state | 0 i upon which
H is constructed by application of creation operators (or their likes). In free field theories (in
the QFT sense), the vacuum may be defined as the state annihilated by the positive frequency
part of the field. For an interacting field, we assume following the usual procedure of QFT that
the Hilbert space is the same as for a free field, except that the actual energy eigenstates are
different. We suppose that the interaction is attenuated as t→ ± ∞ and that asymptotic fields
are free.
In a two-dimensional CFT on the complex plane, the in-states are defined as
| φin i = lim φ(z, z)| 0 i , (4.29)
z,z→0

for some quasi-primary field φ(z, z). On H, we must also define a bilinear product, which we do
indirectly by defining asymptotic out-states together with the action of Hermitian conjugation
on conformal fields. One sets
h φout | = | φin i+ , (4.30)
118 Detournay

with
△ 1 1
[φ(z, z)]+ = z −2h z −2h̄ φ( , ) , (4.31)
z z
where φ is a quasi-primary field of dimensions (h, h̄). In Minkowski space, Hermitian conjugation
does not affect the space-time coordinates, while in euclidian space, τ = it must be reversed
(τ → − τ ) upon Hermitian conjugation if t is to be left unchanged. So this corresponds to the
mapping z→ z1∗ = z1 on the real surface (see Sect. 4.1 and (4.1)) and explains the interchange
z→ z1 , z→ z1 in (4.31). The additional factors on the r.h.s. of (4.31) may be justified by demanding
that the inner product h φout |φin i is well-defined21 . Let us focus on the holomorphic part of the
stress-energy tensor. By taking the adjoint of the mode expansion (4.21), one finds on the one
hand X
+
[T (z)] = z −n−2 L+n , (4.32)
n

and on the other hand, from (4.31)

+
X X
[T (z)] = z −4 T (1/z) = z −4 z n+2 Ln = z −m−2 L−m . (4.33)
n m

These expressions are compatible if

L+
n = L−n , and similarly L̄+
n = L̄−n . (4.34)

We may also determine the action of the Ln ’s on H. With (4.20) and (4.22), we have
I
△ 1
Ln (z)φ(z, z) = (Ln φ)(z, z) = dw (w − z)n+1 T (w)φ(z, z) . (4.35)
2πi
z

Consider a primary field φ(z) of conformal weight h (leaving the anti-holomorphic part aside)
and the asymptotic state
△
| h i = φ(0)| 0 i . (4.36)
Acting with the Ln ’s on this state yields
I
△ 1
(Ln φ)(0) = Ln | h i = dw wn+1 T (w)φ(0)| 0 i . (4.37)
2πi
0

Using the OPE (3.1), one gets

I  
1 h 1
Ln | h i = dw wn+1 φ(0) + ∂φ(0) + reg |0i (4.38)
2πi w2 w
0
I
1  
= dw hwn−1 φ(0) + wn ∂φ(0) + wn+1 .reg | 0 i . (4.39)
2πi
0

21 h φ
out |φin i = lim h 0 |φ(z, z)φ(w, w)| 0 i = lim z −2h z −2h̄ h 0 |φ(1/z, 1/z)φ(w, w)| 0 i =
z,z,w,w→0
lim h 0 |φ(ξ̄, ξ)φ(w, w)| 0 iξ 2h̄ ξ̄ 2h . We will see in Sect 4.4 that the 2-point functions are completely fixed
ξ,ξ̄→0
C
by conformal invariance, and have the form hφ(ξ̄, ξ)φ(w, w)i = , so that the product is well defined
ξ̄2h ξ2h̄
because the factors cancel out. See also [5] p52, [4] p147.
Aspects of Two-Dimensional Conformal Field Theory 119

For n ≥ −1, none of the regular terms of the OPE will contribute the integral. We thus find
that
Ln | h i = 0 if n > 0 , L0 | h i = h| h i (4.40)
Remark we also have following the same computation that (Ln φ)(w) = 0 for n > 0 and
(L0 φ)(w) = hφ(w).
Eq. (4.40) defines a highest weight representation of the Virasoro algebra (4.27)22 of , just like
the ones one usually encounters in the su(2) case (see the Lectures on Lie algebras).The operator
L0 measures the conformal dimension of the state, while the Ln ’s play the rôle of the raising
and lowering operators, taking one state to another into a representation. The major difference
between the Virasoro (est arrivéééé) representations and those of su(2) is of course that the
Virasoro representations are infinite-dimensional. Although the algebra is infinite-dimensional,
the Cartan subalgebra (i.e. the maximal set of commuting Hermitian generators) contains just
the identity operator and L0 (from (4.27), we see that no pair of generators commute, so we
may choose L0 which, from (4.40), is diagonal in the representation space). The operator L0 is
used to label the states, and from (4.27) we find

[L0 , L−n ] = nL−n , (4.41)

so the state L−n | h i has eigenvalue h + n under L0 (n > 0).

We see that the Virasoro algebra V plays an important rôle in conformal field theory. The
states of the theory span a set of infinite-dimensional representations of V. Eqs. (4.36) to (4.40)
exhibit what is generally referred to as field-state correspondence : the primary fields of the
theory are in one-to-one correspondence with the highest-weight states (h.w.s.) of V. Then,
using (4.41), one obtains excited states by successive applications of the raising operators on
| h i:
L−k1 · · · L−kn | h i . (4.42)
When the L−ki appear in increasing order of the ki ’s, 1 ≤ k1 ≤ · · · ≤ kn , the states (4.42)
provide a complete basis for the representation descended from | h i, because a different ordering
can always be brought into a linear combination of the well-ordered states (4.42) by applying the
commutation rules (4.27) as necessary. The state (4.42) is an eigenstate of L0 with eigenvalue
△
h′ = h + k1 + · · · + kn = h + N . (4.43)

The states (4.42) are called descendants of the asymptotic state | h i and the integer N is called
the level of the descendant. The number of distinct, linearly independent states at level N is
the number p(N ) of partitions of the integer N. The complete set of states generated by the
h.w.s. | h i and its descendants is closed under the action of the Virasoro generators and forms
a representation of V, called a Verma module. A Verma module in conformal field theory is
characterized by the central charge c and the dimension h of the h.w.s., and is denoted by
V (c, h). The Verma modules for the anti-holomorphic part are denoted by V̄ (c̄, h̄). Now recall
that the hamiltonian of the system was identified with the operator L0 + L̄0 , so that the energy
eigenstates belong to the tensor product V ⊗ V̄ . In general ,the total Hilbert space is a direct
sum of such tensor products, over all conformal dimensions of the theory
X
H= V (c, h) ⊗ V̄ (c̄, h̄) , (4.44)
h,h̄

22 Representations of the anti-holomorphic counterpart of (4.27) are constructed by the same method. Since

the two parts of the overall algebra decouples ([Ln , L̄m ] = 0), representations of the latter are obtained simply
by taking tensor products
120 Detournay

where the number of terms in this sum may finite or infinite (we will see that in the simple case
of a free boson, this number is infinite).
To close this section, we notice that there is a very special Verma module formed from the
identity operator I = I(z, z). The h.w.s. is just the vacuum | h i, having h = 0. Clearly
L1 | 0 i = 0 because the vacuum is a h.w.s. The Virasoro algebra tells us that L1 (L−1 | 0 i) = 0, so
L−1 | 0 i is also a h.w.s., with h = 1. It is common to require that L−1 | 0 i = 0, so that the vacuum
is Sl(2, C) invariant (recall L−1 , L0 , L1 generate a Sl(2, C) subalgebra). This originates from the
following. In a theory, the vacuum is generally taken to be invariant under the underlying
symmetry algebra. In principle, we should thus take

Ln | 0 i = 0 , ∀n (4.45)

but this would imply

[Ln , Lm ]| 0 i = 0 , ∀n, m , (4.46)
which is not true because of the central term in (4.27) (take e.g. n = m = −2). So, the best we
can do is
Ln | 0 i = 0 , n ≥ −1 . (4.47)
This condition can be recovered by requiring that T (z)| 0 i be well-defined as z→0 :

lim T (z)| 0 i = lim z −n−2 Ln | 0 i (4.48)

z→0 z→0

is well defined if Ln | 0 i = 0 for n ≥ −1, which includes in particular invariance under the global
subgroup.

4.4 Ward identities strike back : global Ward identities

The operator formalism provides us with a useful framework to investigate how the conformal
symmetry puts constraints on correlation functions. In this formalism, a generic N-point function
of primary fields is written

h 0 |φ1 (z1 ) · · · φN (zN )| 0 i , with |z1| ≥ · · · ≥ |zN | . (4.49)

From (4.45), we know that the vacuum satisfies Ln | 0 i = 0 for ∀n ≥ −1. This also implies
(Ln | 0 i)+ = h 0 |L+
n = 0 ∀n ≥ −1, by using (4.34). Thus, we find that

h 0 |Li = Li | 0 i = 0 for i = −1, 0, +1 , (4.50)

that is for the generators of the global subgroup (that are not affected by the conformal anomaly
in (4.27)). Therefore, we may write

0 = αi h 0 |Li φ1 (z1 ) · · · φN (zN )| 0 i

X
= αi h 0 |φ(z1 ) · · · φ(zj−1 )[Li , φ(zj )]φ(zj+1 ) · · · φ(zN )| 0 i + αi h 0 |φ1 · · · φN Li | 0(4.51)
i
j

where the last term vanishes because of (4.50) and where αi is an infinitesimal parameter. But,
according to (4.25),
δαi (z) φ(z) = −αi [Li , φ(z)] , (4.52)
and (4.51) becomes
X
h 0 |φ1 (z1 ) · · · δαi (zj ) φj (zj ) · · · φN (zN )| 0 i = 0 (4.53)
j
Aspects of Two-Dimensional Conformal Field Theory 121

that is, by definition

δαi hXi = 0 , i = −1, 0, 1 , (4.54)
for X = φ1 · · · φN . Let us analyze the consequences of (4.54). Recall that

δα(z) φ(x) = −(hα′ (z) + α(z)∂z )φ(z) , (4.55)

P n+1 △
P
where α(z) = n αn z = n αn (z). We now develop (4.54) explicitly. For the 1-point
function of a primary field, this yields the following set of equations:

 ∂z hφ(z)i = 0 ,

(h + z∂z )hφ(z)i = 0 , (4.56)


(2zh + z 2 ∂z )hφ(z)i = 0

which imply that hφ(z)i = cst., where the constant may be different from zero only if h = 0.
A 2-point function G12 (z1 , z2 ), with z1 6= z2 has to satisfy

hδαi (z1 ) φ1 (z1 )φ2 (z2 )i + hφ1 (z1 )δαi (z2 ) φ2 (z2 )i = 0 (4.57)

and thus
[h1 αi′ (z1 ) + αi (z1 )∂1 + h2 αi′ (z2 ) + αi (z2 )∂2 ] G12 (z1 , z2 ) = 0 . (4.58)
For i = −1, one gets
(∂1 + ∂2 )G(z1 , z2 ) = 0 (4.59)
and hence
G(z1 , z2 ) = G(x) , x = z1 − z2 , (4.60)
as is natural from translation invariance. The equation for i = 0 then reduces to

(h1 + h2 + x∂x )G(x) = 0 , (4.61)

with solution
G(x) = Cx−(h1 +h2 ) . (4.62)
Finally, by substituting in the equation for i = 1, we find

(h1 − h2 ) x G(x) = 0 , (4.63)

leading to the result

(
C12 (z1 − z2 )−2h if h1 = h2 = h ,
G(z1 , z2 ) = . (4.64)
0 otherwise

Taking the anti-holomorphic part into account would have given rise to an additional factor
of (z 1 − z 2 )−2h̄ .
A similar reasoning also completely fixes the form of the three-point functions (see [3] p43,
[2]p105, and [6] p89-97 for an explicit computation). The result is
h3 −h1 −h2 h1 −h2 −h3 h2 −h3 −h1
G123 (z1 , z2 , z3 ) = C123 z12 z23 z31 , (4.65)

with zij = zi − zj .
As a consequence, the coordinate dependance all n-point functions up to n = 3 are completely
fixed (up to multiplicative constant) by the (global) conformal symmetry. This can be understood
122 Detournay

as follows. We have three complex transformations at our disposal (i = −1, 0, 1). Using them,
we can always bring three variables z1 , z2 , z3 to three arbitrary fixed points α1 , α2 , α3 , by means
of
az1 + b az2 + b az3 + b
= α1 , = α2 , = α3 . (4.66)
cz1 + d cz2 + d cz3 + d
These 3 equations for the four complex variables a, b, c, d, ad − bc = 1 always have a solution if
all the zi ’s are different. One usually chooses α1 = 0 , α2 = 1 and α3 = ∞. Hence, the entire
answer is determined if we know the 3-point function in just three points.
This doesn’t work any longer for n-point functions, n ≥ 4. The 4-point function takes the
following form (see e.g. [6], Exercises 3,4,8) :
 
z12 z34 Y −hi −hj +h/3
G(z1 , z2 , z3 , z4 ) = f z , (4.67)
z13 z24 i<j ij

P4
where h = i=1 hi .

4.5 Descendant fields

We found in Sect. 4.3 that a highest weight state | h i of the Virasoro algebra is obtained by
applying a primary field of conformal dimension h to the vacuum state | 0 i, see (4.36), (4.40).
This state is the source of an infinite tower of descendant states of higher conformal dimension
(see (4.42), (4.43)). Under a conformal transformation, i.e. by application of Ln ’s, the state | h i
and its descendants tranform among themselves.
Each descendant state L−n | h i can be viewed as the result of the application on the vacuum
of a descendant field. Indeed, as seen in (4.37) :
I
△ 1
L−n | h i = L−n φ(0)| 0 i = (L−n φ)(0)| 0 i = dz z −n+1 T (z)φ(0)| 0 i . (4.68)
2πi

The descendant field associated with the state L−n | h i is denoted (see (4.35) )
I
△ 1
φ (−n)
(w) = (L−n φ)(w) = dz (z − w)−n+1 T (z)φ(w) . (4.69)
2πi
w

With (4.69), we may write the complete OPE of T (z) with a primary field (including all regular
terms) as
X
T (z)φ(w) = (z − w)k−2 φ(−k) (w) . (4.70)
k≥0

Because we know the singular part of the OPE (see (3.1)), we have that

φ(0) (w) = hφ(w) and φ(−1) (w) = ∂φ(w) , (4.71)

see also below eq. (4.40). The other descendant fields are defined by the regular part of the
OPE, and can be computed explicitly once the regular part is known (see Sect. 4.6)
By taking φ(w) = I, the identity field, we obtain
I
1
I (−n) (w) = dz z −n+1 T (z) , (4.72)
2πi
w
Aspects of Two-Dimensional Conformal Field Theory 123

so I (−n) (w) = 0 for n ≤ 1, and

1
I (−n) (w) = ∂ (n−2) T (w) for n ≥ −2 . (4.73)
(n − 2)!

In particular, this shows for n = 2 that the energy-momentum tensor can be seen as a descendant
field of the identity operator (field).
A remarkable property of these fields is that their correlation functions may be derived from
those of their corresponding primary field. Consider the correlator

hφ(−n) (w)Xi , (4.74)

with X = φ1 (w1 ) · · · φN (wN ) a string of primary fields of conformal dimensions hi :

I
1
hφ (−n)
(w)Xi = dz (z − w)−n+1 hT (z)φ(w)Xi , (4.75)
2πi
w

where the contour circles w only, excluding the positions of the other fields. This can further be
expressed, by ”reversing” the contour (see Fig.12), deforming it in a sum over contours circling
only one point wi and by considering, in each contour, the OPE (3.1) of T (z) with the relevant
field φi (wi ):
I
1 X
hφ(−n) (w)Xi = − dz (z − w)−n+1 hφ(w)φ1 (w1 ) · · · T (z)φi (wi ) · · · φN (wN )i
2πi i
wi
I  
1 X hi (z − w)−n+1 (z − w)−n+1
= − dz + ∂wi hφ(w)Xi
2πi i (z − wi )2 z − wi
wi
X 
= − hi (1 − n)(wi − w)−n + (wi − w)−n+1 ∂wi hφ(w)Xi
i
≡ L−n hφ(w)Xi , n≥1

with the differential operator

X 
L−n = hi (1 − n)(wi − w)−n + (wi − w)−n+1 ∂wi . (4.76)
i

We have thus reduced the evaluation of a correlation function containing a descendant field to
that of a correlator of primary fields, on which we must apply a differential operator. Because
X
(∂w + ∂wi )hφ(w)Xi = 0 (4.77)
i

P as a consequence of translation invariance (see (4.53) and (4.55) for i = −1)

for any correlator
and L−1 = − i ∂wi , we have

L−1 hφ(w)Xi = ∂w hφ(w)Xi = 0 . (4.78)

A general descendant field, corresponding to the state (4.42), has the form
△
φ(−k1 ,··· ,κN ) (w) = (L−k1 · · · L−kN φ)(w) . (4.79)
124 Detournay

w2 w2
w1 w1
. .

. .

.
w .
w

wN wN

w2 w2
w1 w1
.
.
.
.
. w w
.

wN wN

Figure 12

These field are defined recursively, for instance

φ(−k,−n) (w) = (L−k L−n φ)(w)

I
1
= dz (z − w)1−k T (z)(L−n φ)(w) , (4.80)
2πi
w

and so on. In particular,

φ(0,−n) (w) = (h + n)φ(−n) (w) and φ(−1,−n) (w) = ∂w φ(−n) (w) . (4.81)

The first relation is clear from the Virasoro algebra:

φ(0,−n) (w) = (L0 L−n φ)(w) = (L−n L0 φ)(w) + n(L−n φ)(w) = (h + n)(L−n φ)(w) . (4.82)

The second expresses the transformation of the fields under translations. The variation of
φ(−n) (w) under translations (z−→z + α−1 , Φ(z)−→Φ′ (z ′ ) = Φ(z)) is given by

δα−1 φ(−n) (w) = −α−1 ∂φ(−n) (w) (4.83)

Aspects of Two-Dimensional Conformal Field Theory 125

as for any field. This can also be expressed from (4.25) and (4.22), for n = −1 :
I
(−n) 1
δα−1 φ (w) = − α−1 dw T (w)φ(−n) (w) . (4.84)
2πi
z

But the r.h.s. of (4.84) is just −α−1 φ(−1,−n) (w), see (4.80), and thus with (4.83) we find the
second relation.
Finally, it can be shown that
hφ(−k1 ,··· ,κN ) (w)Xi = L−k1 · · · L−kN hφ(w)Xi , (4.85)
that is, we simply need to apply the differential operators in succession. We may also consider
correlators containing more than one descendant field, but at the end the result is the same
: correlation functions of descendant fields may be reduced to correlation functions of primary
fields. This is why primary fields are of prime interest in CFT’s.
The set comprising a primary field φ and all of its descendants is called a conformal family,
and is denoted by
[φ] = {φ, (L−n φ), · · · , (L−k1 · · · L−kN )} , n > 0, ki > 0 . (4.86)
The members of a family transform amongst themselves under a conformal transformation. This
means that the OPE of T (z) (which generates conformal transformations , see (4.9), (4.23)) with
any member of the family will be composed solely of other members of the same family (conformal
fields have an anti-holomorphic part as well, so there will also be descendants of the field through
the action of the anti-holomorphic generators L̄−n ).

4.6 Operator algebra

We have seen in section 4.4 that conformal invariance takes us a step towards solving a given
conformal field theory (i.e. to be able, at least in principle, to write down all correlation functions
of all the fields present in the theory), by completely fixing the coordinate dependence of the
two- and three-point functions. To fix the numerical coefficients Cijk in (4.65) and to go beyond
three-point functions, some additional information is needed (namely the complete OPE of all
primary fields with each other). Besides conformal invariance, another basic property of two-
dimensional CFT’s is that operator products can be defined for arbitrary pair of fields. That
is, there is a closed operator product algebra among all the fields (called the operator algebra
in short). The operator involved is indeed a product in the sense that it is associative. It is
expected that this structure is a direct consequence of the fundamental properties of quantum
field theory 23 (see e.g. [8] p. 21), but are usually postulates the existence of the operator algebra
as a separate input (this is called the bootstrap approach, where the set of OPE’s is treated as
the fundamental information of the theory, for which the whole theory can be reconstructed ).
In general, the operator algebra is expressed as

X
Oi (x) Oj (y) = Cij k (x − y) Ok (y), (4.87)
k
23 A heuristic argument goes as follows . According to the OPE assumption, in any QFT, the product of local

operators acting at points P

that are sufficiently close to each other may be expanded in terms of the local fields of
the theory: φi (x)φj (y) ∼ k Cij k (x − y)φk (y), where Cij k (x − y) are (possibly singular for x → y) functions.
This is usually an asymptotic statement in QFT. In CFT though it is believed it becomes an exact statement
because scale invariance (dilatations) prevents the appearance of any length l in the theory. So, there is no
parameter to control the expansion, and thus no terms like el(x−y) which would the breakdown of the exactness
of the asymptotic expansion above.
126 Detournay

where the sum runs over all fields present in the theory. (4.87) is to be understood as constraints
on correlations functions. In particular, if φ(y) is some field in the QFT,
X
k
hφ(z)Oi (x) Oj (y)i = Cij (x − y) hφ(z)Ok (y)i , (4.88)
k
and hence (4.87) allows to compute all correlation functions of the theory recursively, reducing
ultimately to the 2-point functions, which are known. Let us make this more precise. We have
seen that the 2-point function vanishes if the conformal dimensions of the two fields are different
(4.64). We are free to choose a basis of primary fields such that
(
1
(w−z)2h (w̄−z̄)2h̄
if h1 = h2 = h and h̄1 = h̄2 = h̄ ,
hφ1 (w, w̄)φ2 (z, z̄)i = (4.89)
0 otherwise .

it is a simple matter of normalization.

Thus, primary fields φ1 and φ2 with different conformal dimensions are orthogonal in the
sense of the two-point functions. This is also the case for all the descendants fields of φ1 and φ2 ,
because as we have seen in section 4.5, correlation functions of descendant fields can be reduced
to correlation functions of primary fields. Thus, if h1 6= h2 , then the conformal families [φ1 ]
(−n ,...,−nh ) −l1 ,...,−ln )
and [φ2 ] are orthogonal (hφ1 1 φ2 i = 0 if h1 6= h2 ). The same is true for the
corresponding Verma modules. Let

|h1 , h̄1 i = φ1 (0, 0)|0i (4.90)

|h2 , h̄2 i = φ2 (0, 0)|0i, (4.91)
see (4.29). Then

hh1 , h̄1 |h2 , h̄2 i = lim h0|φ+

1 (z, z̄)φ2 (w, w̄)|0i
z,z̄,w,w̄→0
11
= lim z̄ −2h1 z −2h̄1 h0|φ1 ( )φ2 (w, w̄)|0i
z,z̄,w,w̄→0 z̄ z
= lim ρ̄2h1 ρ2h̄1 h0|φ1 (ρ, ρ̄)φ2 (0, 0)|0i
ρ,ρ̄→0
(
1 if h1 = h2 = h ,
= (4.92)
0 if h1 = 6 h1 .

using (4.89).
This is also evident from the fact that two eigenspaces of a Hermitian operator (here L0 )
having different eigenvalues are orthogonal. Furthermore, the othogonality of the h.w.s. implies
the othogonality of the Verma modules associated to the two fields (this is also as consequence
of the orthogonality of the conformal families, because descendant states can be seen as created
for the vacuum by the application of the descendant fields, see (4.68)). The operator algebra
takes the general form
X p
φ1 (z1 , z2 )φ2 (0, 0) = C12 z hp −h1 −h2 z̄ h̄p −h̄1 −h̄2 × [φp (0, 0) + zβp(−1) φ(−1)
p (0, 0)
p
¯ ¯ ¯ ¯
+z̄ β̄p(−1) φ(−1)
p (0, 0) + z z̄βp(−1) β̄p(−1) φp(−1,−1) (0, 0)
+z 2 (βp(−1,1) φ(−1,1)
p (0, 0) + βp(−2) φ(−2)
p (0, 0)) + z̄ 2 (...) + ...] (4.93)
Aspects of Two-Dimensional Conformal Field Theory 127

(−1) ¯
(−1)
where the sum runs over all conformal dimensions present in the theory, φp = (L−1 φp ), φp =
¯
(−1)(−1)
(L̄−1 φp ), φp = (L−1 L̄−1 φp ), ... as usual, and the coefficients βp are constants defining how
the descendants of a given family [φp ] contribute to the OPE, which can be determined as func-
tions of the central charge and of the conformal dimensions by requiring that both sides of (4.93)
behave identically upon conformal transformations (for an example of such a computation, see [2]
p. 181 and [1] p. 51).

We now focus on the z-part of (4.93):

X p hp −h1 −h2
φ1 (z)φ2 (0) = C12 z [φp (0) + zβp(−1) φp(−1) (0) + z 2 (βp(−1,1) φ(−1,1)
p (0)
p

+β (−2) φ(−2)
p (0)) + ...] (4.94)

The prefactor z hp −h1 −h2 ensures invariance under scaling transformations. Indeed, consider
z → λz, φ(z) → φ′ (z ′ ) = λ−h φ(z). Applying this to both sides, we have for the l.h.s. :

φ1 (z)φ2 (0) → λ−h1 −h2 φ1 (z)φ2 (0), (4.95)

while for the first term on the r.h.s.:

z hp −h1 −h2 φp (0) → λ−h1 −h2 λhp z hp −h1 −h2 λ−hp φp (0). (4.96)

Thus both sides get multiplied by the same factor λ−h1 −h2 . For the descending fields, one needs
(−1) (0,−1) (−1)
additional factors, because L0 φp = φp = (hp + 1)φp , see (4.81), which translates into
(−1)
the fact that φp (w) transforms under dilations (L0 ) as a primary field of conformal dimension
(hp + 1). So,

z hp −h1 −h2 zφ(−1)

p (0) → λhp +1 λ−h1 −h2 λ−hp −1 φp(−1) (0), (4.97)

and so on for the other descendants at higher levels.

In conclusion, the complete operator algebra may be deduced form the conformal symmetry,
the only necessary ingredients being the central charge, the conformal dimensions of the primary
k
fields and the coefficients Cij . The latter can be obtained from another source, through a
procedure called the conformal bootstrap, consisting essentially in implementing the constraint of
associativity of the operator product algebra (crossing symmetry) (see [2]Sections 6.6.4 and 6.6.5,
[3] p49, [7] p16, [8] p42, [9] p22). Thus, any n-point function can in principle be calculated from
the operator algebra by successive reduction of the products of primary fields. The correlations
of descendant fields obtained can be expressed in terms of primary field correlators, and so on,
up to two-point functions which are known. So the theory is solved, in principle!

4.7 Ward identities: final chapter

We discussed Ward identities at different places in these notes, using two different formalisms
of quantum field theory. Within the path integral representation, we found that the variation
of a correlation function hφ(x1 )...φ(xn )i = hXi under an arbitrary conformal transformation
z → z + α(z), φ(z) → φ′ (z + α(z)) is given by
128 Detournay

I
1
δα hXi = − dξα(ξ)hT (ξ)Xi (4.98)
2πi
C

where C encircles the position of all fields in X. On the other side, in the operator formalism,
we observed in section 4.4 that for α(z) corresponding to a global conformal transformation

αg (z) = α−1 + α0 z + α−1 z 2 , (4.99)

this variation vanishes:

δαg hXi = 0. (4.100)

We would now like to verify that (4.98) effectively vanishes for α(z) of the form (4.99). This
may be done using the operator algebra (4.94). Let us work out explicitly a simple example.
Consider the case X = φ1 (z1 )φ2 (0). This product can be expanded using (4.94). Now, recall
from (4.72) that T (z) belongs to the family of the identity operator I (with h = 0), and that
conformal families corresponding to different primary fields are orthogonal. Thus, when con-
sidering hT (ξ)φ1 (z1 )φ2 (0)i and expanding φ1 (z1 )φ( 0) using (4.94), the only family involved in
the expansion will be that of the identity operator (which is the only (one) primary field with
h = 0). Then,

(−1,−1) (−1,−1) (−2) (−2)

hT (ξ)Xi = hT (ξ)z1−h1 −h2 (I + z1 β0−1 I (−1) (0) + z12 (β0 I (0) + β0 I (0)) + ...)
(4.101)

But, from (6.1), we know that hT (ξ)i = 0, and I (−1) = (L−1 I) = 0, because the vacuum is
1
SL(2, C) invariant (L−1 |0i = 0). We have, because I (−n) (w) = (n−2)! ∂ (n−2) T (w), see (4.68),

hT (ξ)Xi = z1−h1 −h2 +2 hT (ξ)T (0)iβ0−2 + terms involvinghT (ξ)∂ n T (0)i, n ≥ 1 (4.102)

In (4.98), note that we can send the contour to infinity, because all operators are already
inside C. This way, the integrand behaves as

α(ξ)
α(ξ)hT (ξ)Xi ∼ α(ξ)hT (ξ)T (0)i ∝ , (4.103)
ξ→∞ ξ4

where we used (6.2). Thus for α(ξ) given by (4.99), the integral in (4.98) vanishes. We re-
cover the invariance of hXi w.r.t. to transformations of the global conformal group. For local
transformations it is relation (4.98) that holds, instead of the expected invariance. Returning to
the derivation of section 4.4, we may see that this originates from the apparition of the central
charge c, which prevents to apply the same procedure, because we don’t have h0|Ln = Ln |0i ∀n,
but only for i = −1, 0, 1 corresponding to the global subgroup.

4.8 A short story on normal ordering

In this section, I will introduce the notion of conformal normal ordering, and show that it
reduces in some situations to the more familiar operator normal ordering, consisting in moving
annihilator operators to the right. We saw that the OPE of two fields can be written in general
as
Aspects of Two-Dimensional Conformal Field Theory 129

XN
{AB}n (w)
A(z)B(w) = (4.104)
n=−∞
(z − w)n

The conformal normal ordering of the two fields, denoted by :AB : , is by definition

△
:AB : (w) = {AB}0 (w), (4.105)

that is, the term of order (z − w)0 in the OPE. This is a particular prescription, which in general
differs from the definition

(
Am Bn for n > 0 ,
:Am Bn : = (4.106)
Bn Am for n ≤ 0 .

where

I
1
Am := Am (0) = dzz m+hA −1 A(z)
2πi
I
1
Bn := dzz m+hB −1 B(z) (4.107)
2πi
are the modes of the fields A(z) and B(z) (see (4.18)). We will return to this.

The singular terms in an operator product are called contraction and are denoted by

XN
△ {AB}n (w)
A(z)B(w) = . (4.108)
n=1
(z − w)n

Again this definition is in general different from the QFT one, where contraction generally means
propagator (two-point function).
This is well illustrated in the case of the field T (z). Indeed, with (4.108) and (3.14), one has

c/2 2T (w) ∂T (w)

T (z)T (w) = + + , (4.109)
(z − w)4 (z − w)2 (z − w)
while
c/2
hT (z)T (w)i = . (4.110)
(z − w)4
From (4.105), the conformal normal ordering can be represented using contour integration by

Z
1 dz
:AB : (w) = A(z)B(w) (4.111)
2πi (z − w)
where, as usual radial ordering is understood in the r.h.s.
130 Detournay

The OPE (4.104) is then

A(z)B(w) = A(z)B(w) + :AB : (w) + O(z − w) (4.112)

Notice that, when the contraction coincides with the propagator, this formula is exactly the
one we get in usual QFT, where the ordering is the familiar operator normal ordering (4.106).
A field for which the contraction contains only one singular term (necessarily coinciding with
the 2-point function) is called a free field. In this situation, both orderings are equivalent, and
Wick’s theorem may be applied as such. When the field are not free, Wick’s theorem must be
adapted (see [2] p188).

5 The free boson

Consider the following action for a massless scalar field in two dimensions
Z
1
S= d2 x ∂µ φ∂ µ φ , (5.1)
8π

or in the usual complex coordinates z = x1 + ix2 , ∂ = ∂z , · · · :

Z
1
S= d2 z ∂φ∂φ , (5.2)
4π

where d2 z = dzdz = 2dx1 dx2 . This action is invariant under scalings x−→λx, φ(x)−→φ′ (x′ ) =
λ−∆ φ(x) if ∆ = 0.
The equations of motion are
δS
= 0 ⇒ ∂∂φ(z, z) = 0 . (5.3)
δφ
The energy-momentum tensor (see (2.14)) is given by
1 1
Tµν = (∂µ φ∂ν φ − gµν ∂α φ∂ α φ) (5.4)
4π 2
and thus (gzz = 1/2, gzz = gzz = 0)
1 1
Tzz = ∂φ∂φ , Tzz = ∂φ∂φ (5.5)
4π 4π
Later on we will use the following convenient normalization

T (z) = −2πTzz , T̄ (z) = −2πTzz . (5.6)

The primary input we need is the OPE of the fields appearing in the action. This can be
obtained by the following method. Consider the variation

δφ(x) = ε(x) (5.7)

△
and its effect on correlation functions of the form hφ(x1 ) · · · φ(xN )i = hXi. The variation of hXi
is given by (2.11), with Z
1
δS = d2 x∂µ ε∂ µ φ . (5.8)
4π
Aspects of Two-Dimensional Conformal Field Theory 131

Thus, we have
X Z
ˆ i ) · · · φ(xN )i = 1
ε(xi )hφ(x1 ) · · · φ(x d2 x ∂µ ε(x)∂ µ hφ(x)Xi , (5.9)
i
4π

where ˆ stands for omission. Integrating the r.h.s. by parts and discarding the boundary term,
one has Z
1
− d2 x ε(x)∂µ ∂ µ|x hφ(x)Xi . (5.10)
4π
Therefore, (5.9) implies that

1 X
− 2x hφ(x)Xi = ˆ i ) · · · φ(xN )i
δ 2 (x − xi ) hφ(x1 ) · · · φ(x . (5.11)
4π i

This can be solved as

X
hφ(x)Xi = ˆ i ) · · · φ(xN )i + h :φ(x)φ(x1 ) · · · φ(xN ): i
ln |z − zi |−2 hφ(x1 ) · · · φ(x , (5.12)
i

where |z − w|2 = (z − w)(z − w). This is understood as follows. The first term originates from
the fact that 2x = 4∂∂ and that

∂∂ ln |z − w|−2 = −πδ 2 (x − y) , (5.13)

R
x = (z, z), y = (w, w) and d2 x δ 2 (x)f (z) = f (0). Eq. (5.13) follows from the following
representation of the δ function :
1 1 1 1
δ 2 (x) = ∂z = ∂z , (5.14)
π z π z
see e.g. [2] p.119. The second term of the r.h.s. of (5.12) actually defines :φ(x)φ(x1 ) · · · φ(xN ): ,
where the function h :φ(x)φ(x1 ) · · · φ(xN ): i satisfies

∂∂h :φ(x)φ(x1 ) · · · φ(xN ): i = 0 , (5.15)

i.e. it is a harmonic function. It is thus locally the sum of an analytic and anti-analytic function,
and can be Taylor expanded. In particular, it has no pole as x→xi . Eq. (5.12) gives us the
behavior of the function hφ(x)φ(x1 ) · · · φ(xN )i as x→xi , i.e. by definition the OPE of φ(x) with
itself. This is written
φ(z, z)φ(zi , z i ) ∼ ln |z − zi |−2 , (5.16)
where this relation is to be understood inside a correlation function with other fields, and ∼
means equality modulo terms regular as x→xi . If there are no other fields, this gives us the
propagator
hφ(z, z)φ(zi , z i )i = ln |z − zi |−2 , (5.17)
up to a harmonic function (regular as z→zi ). This shows, as expected, that φ(z, z) is a free field,
as defined in the preceding section.
Finally, eq. (5.12) is usually rewritten as

φ(z, z)φ(zi , z i ) = ln |z − zi |−2 + :φ(z, z)φ(zi , z i ):

= hφ(z, z)φ(zi , z i )i + :φ(z, z)φ(zi , z i ): , (5.18)
132 Detournay

which is the familiar relation between time(radial)-ordering and normal ordering in QFT.
This can be brought in the form (4.112) by Taylor expanding the whole series of regular
terms :φ(z, z)φ(zi , z i ): as

:φ(z, z)φ(zi , z i ): = :φ φ: (zi , z i ) + (z − zi ) :∂φ φ: (zi , z i ) + (z − z i ) :∂φ φ: (zi , z i ) + · · · , (5.19)

from which we deduce

φ(z, z)φ(zi , z i ) = ln |z−zi |−2 + :φ φ: (zi , z i )+(z−zi ) :∂φ φ: (zi , z i )+(z−z i ) :∂φ φ: (zi , z i )+· · · .
(5.20)
Focusing on the z sector, we are left with

φ(z)φ(w) = − ln(z − w) + :φ φ: (w) + (z − w) :∂φ φ: (w) + · · · , (5.21)

which coincides exactly with (4.112). We may also write

:φ φ: (w, w) = lim [φ(z, z)φ(w, w) − hφ(z, z)φ(w, w)i] (5.22)

z,z→w,w

6 Exercises
6.1 Operator formalism
1. Show that (4.47) implies that

hT (z)i = h 0 |T (z)| 0 i = 0 (6.1)

2. Compute the two-point function of T (z) and show that

c/2
hT (z)T (w)i = h 0 |T (z)T (w)| 0 i = , (6.2)
(z − w)4

by using (4.47), (4.34) and (4.27).

This can also be inferred from (3.14) and (6.1)

3. Compute the norm of the state L−2 | 0 i and show that a necessary condition for the the-
ory to be unitary is c > 0.

4. Compute the singular part of the OPE of φ(−1) (w) with T (z), and deduce from this the
transformation law of the descendant field φ(−1) (w). For this, use the expansion T (z) =
P −k−2
k (z − w) Lk (w) and the Virasoro algebra (4.27).
2hφ(w) φ(−1) (w) ∂φ(−1) (w)
(R : T (z)φ(−1) (w) ∼ (z−w)3
+(h+1) (z−w)2 + z−w and δα(z) φ(−1) (z) = −hα′′ (z)φ(z)−(h+1)α′ (z)φ(−1) (z)−
α(z)∂φ(−1) (z), where h is the conformal weight of the primary field φ(z) from which the field descends.)

6.2 Free boson

1. From (5.16), write the OPE of ∂z φ(z, z)∂w φ(w, w) and ∂z φ(z, z)∂w φ(w, w).

1 1
(R : ∂z φ(z, z)∂w φ(w, w) ∼ − (z−w) 2 and ∂z φ(z, z)∂w φ(w, w) ∼ − (z−w)2 . Thus everything splits in a holo-
△
morphic and anti-holomorphic part. In the following, we will focus on the holomorphic field ∂φ = ∂z φ.)
Aspects of Two-Dimensional Conformal Field Theory 133

2. The energy-momentum tensor (5.5) is not well defined because ∂φ(z)∂φ(w) have singularities
as z→w. Therefore the quantum version of (5.6) is
1 1
T (z) = − :∂φ ∂φ: , T̄ (z) = − :∂φ ∂φ: . (6.3)
2 2
that is, with (5.22)

T (z) = −2π lim [∂φ(z) ∂φ(w) − h∂φ(z, z) ∂φ(w, w)i] . (6.4)

w→z

Compute the OPE T (z)∂φ(w). What does this show? You could need the following Wick’s
rules:
T (φ1 φ2 φ3 ) = :φ1 φ2 φ3 : + :φ1 φ2 φ3 : + :φ1 φ2 φ3 : + :φ1 φ2 φ3 : , (6.5)

with similar relations holding when some factors are already normal-ordered in the the T-
product, but with omission of contractions inside a same normal ordering. For instance,
T ( :φ1 φ2 : :φ3 φ4 : ) does not contain the term hφ1 φ2 i :φ3 φ4 : .
∂φ(w) ∂ 2 φ(w)
(R : T (z)∂φ(w) ∼ (z−w)2
+ z−w
)

3. Compute the OPE T (z)T (w). Check it has the form (3.14) and deduce the central charge of
the system.

4. Consider the following field, called a vertex operator :

Vα (w) = :eiαφ(w) : . (6.6)

a. Compute its OPE with ∂φ(z)

iα
(R: ∂φ(z)Vα (w) ∼ − z−w Vα (w) )
b. Compute its OPE with T (z) and show this operator is primary with conformal weight α2 /2.

5. Compute the OPE :(∂φ(z))k : :Vα (w): . For this, compute the OPE :Ak : :eB : , for two
general operators A and B.

6. Using the result of Ex.5, show that :eA : :eB : = ehABi :eA eB : . From this, deduce the
correlation function hVα (z)Vβ (w)i (the correlator being given by the most singular part of the
OPE).
(R : See [6], p47)

6.3 Ghost system

The b − c ghost system is defined by
Z
1
d2 x bµν ∂ ν cν , (6.7)
2π
where the field bmuν is a traceless symmetric tensor, and where both cµ and bµν are fermions.
Going to complex coordinates, with cz = c, cz = c̄, bzz = b and bzz = b̄, the action is rewritten
(d2 x = 21 dzdz = 12 d2 z) Z
1
d2 z (b∂c + b̄∂c̄) . (6.8)
2π
134 Detournay

From the equations of motion the fields c and b are holomorphic. They satisfy the following
OPE
1
b(z)c(w) ∼ . (6.9)
z−w
The normal-ordered holomorphic energy-momentum is obtained as (see e.g. [2], p133)

T (z) = :2∂c b + c ∂b: . (6.10)

1. Compute the OPE’s T (z)c(w) and T (z)b(w). Deduce the conformal weights of c and b.

2. Compute the (most singular part of the) OPE T (z)T (w). What is the central charge of
this system?

3. This system appears in the covariant quantization of the bosonic string through the gauge-
fixing of two dimensional gravity (see [10], Chapter 3). In this context, each space-time coordi-
nate is represented by a free boson. How many free bosons do you need in order for the central
charge of the system ”D free bosons + ghosts” to vanish? You just found one way of computing
the critical dimension of bosonic string theory.

Acknowledgments
I would like to heartily thank Ph. Spindel for numerous enjoyable, useful (and sometimes
lively) discussions on the topics discussed here. I am grateful to Vl. Dotsenko for enlightening
correspondences. I thank E. Rodrigues Ferreira for a discussion and references on complex
functions. Finally, big kisses to S. de Buyl for her help at all stages of this work, including
discussions, support (in all senses), typing part of these notes, etc. etc!

A Another derivation of the conformal Ward identities

This section is mainly based on [2]. Therefore I won’t go through all calculations in details, since
they may be found in the reference. R
Let us start with an action S[φ] = dd x L(φ(x), φ,µ (x)), and consider the following infini-
tesimal transformations
δxµ
xµ −→ x′µ = xµ + ω a , (A.1)
δω a
δF
φ(x) −→ φ′ (x′ ) = φ(x) + ω a (x) . (A.2)
δω a
which are supposed to leave the action S invariant when the infinitesimal parameters {ω a } are
constant (the subscript ”a” indicating as before a collection of indices). Noether’s theorem states
that to every continuous symmetry of a field theory corresponds a conserved current, and hence
a conserved charge. To determine this current, one may use the ”Noether’s trick”, which consists
in looking at the variation of the action under the same transformation, but with ω a depending
on the position. This doesn’t leave the action invariant anymore, and the variation may be
written in general as

S ′ = S + terms involving derivatives of ω a (x) + terms without derivatives of ω a (x) . (A.3)

Aspects of Two-Dimensional Conformal Field Theory 135

Because S ′ = S when the ω a ’s are constants, the last contributions must sum up to zero. Then
the variation of the action can be found to be
Z
δS = dd x jaµ ∂µ ωa , (A.4)

with   ν
∂L δx ∂L δF
jaµ = ∂ν φ − δνµ L a
− . (A.5)
∂(∂µ φ) δω ∂(∂µ φ) δω a
If ω a (x)→0 at infinity, this relation can be integrated by parts, to yield
Z
δS = dd x ∂µ jaµ ωa . (A.6)

Now, if the field configuration satisfies the classical equations of motion, the action is stationary
against any variation of the fields, and δS should vanish for any position-dependent parameters
ω a (x). This implies
∂µ jaµ ≈ 0 . (A.7)
This is Noether’s theorem : every continuous symmetry implies the existence of a current,
conserved on-shell.
We may now return to the Ward identity in field theory. We follow the same procedure which
led us to (2.11). In the path integral, we make the following change of variable :

φ(x)−→φ′ (x) = φ(x) − iω a (x)Ga φ(x) , (A.8)

where the parameters depend on x, and hence δS is not 0 and given by (A.6). Notice that the
variation of a field is here defined by δφ(x) = −iω a Ga φ(x) and differs from (1.13) by a factor
−i. We keep this factor to be coherent with [2] but it is unimportant. We may now rewrite
(2.11) using (A.6) and (A.8) as
n
X Z
−i ω a (xi )hφ1 · · · Ga φi · · · φn i = dd x ω a (x) ∂µ hjaµ (x)φ1 · · · φn i , (A.9)
i=1

where ω a (x) are arbitrary functions.R In the l.h.s., only the values of ω a (xi ), i = 1, · · · , n matter.
This equation is thus of the form f (x)h(x)dx = f (y)φ(y) ∀f (x), which necessarily implies
h(x) = φ(x)δ(x − y). From (A.9), we thus conclude that
n
X
∂µ hjaµ (x)φ1 (x1 ) · · · φn (xn )i = −i δ(x − xj )hφ1 · · · Ga φj (xj ) · · · φn i . (A.10)
j=1

This is the Ward identity for the current jaµ . Applying this to translations, we find from (A.5)
that
∂l △
j µν = −η µν L + ∂ ν φ = Tcµν , (A.11)
∂(∂µ φ)
and recover the fact that the conserved current associated to translation invariance is the canon-
ical energy-momentum tensor. This also shows that the energy-momentum tensor is conserved
inside correlation functions except at coincident points.
Notice that the form of the current may be modified from its canonical definition (A.5)
without affecting the Ward identity. We may indeed freely add to it the divergence of an
antisymmetric tensor without affecting its conservation. In the case of a system invariant under
136 Detournay

Lorentz transformations (rotations), it can be shown that there always exist an antisymmetric
tensor B µνρ , with B ρµν = −B µρν such that the tensor
△
TBµν = Tcµν + ∂ρ B ρµν (A.12)
[µν]
is symmetric on-shell, that is TB ≈ 0. This new tensor is called the Belinfante tensor (see [2]p46
for its construction). Furthermore, much in the same way, the energy-momentum tensor of a
theory with scale invariance can be made traceless on-shell, under certain certain conditions
which are generally assumed to be fulfilled (see [2] p.102). By using this suitably modified
energy-momentum tensor, the Ward identities associated with translation, rotation and dilation
invariance can be obtained from (A.10). In two dimensions, the result is (see [2] p106,118,123)
n
X
∂µ hTνµ (x)Xi = − δ(x − xi )∂xνi hXi (A.13)
i=1
Xn
ǫµν hT µν (x)Xi = −i si δ(x − xi )hXi (A.14)
i=1
n
X
hTµµ (x)Xi = − ∆i δ(x − xi )hXi (A.15)
i=1

Here X stands for a string of primary fields. In two dimensions, the spin generators are given
by S µν = sǫµν , and ∆ stands for the scaling dimensions of the fields. These relations show
that, inside correlations functions, the energy-momentum tensor is conserved, symmetric and
traceless, except at coincident points. This is the quantum counterpart of the classical statement
that the energy-momentum tensor is conserved, symmetric and traceless on-shell. A derivation
of these identities which does not require the hypothesis that the energy-momentum can be made
classically traceless can be found in [2], p123.
We may now turn to the conformal Ward identity. Let εµ (x) be (the components of) a con-
formal Killing vector, satisfying (1.11), and defining an infinitesimal conformal transformation.
We may write
1 1
∂µ (εν T µν ) = εν ∂µ T µν + (∂µ εν + ∂ν εµ )T µν + (∂µ εν − ∂ν εµ )T µν (A.16)
2 2
1 1
= εν ∂µ T µν + (∂ρ ερ )Tµµ + ǫαβ ∂α εβ ǫµν T µν , (A.17)
2 2
where we used
1 1
(∂µ εν + ∂ν εµ ) = (∂ρ ερ )ηµν (A.18)
2 2
1 1 αβ
(∂µ εν − ∂ν εµ ) = ǫ ∂α εβ ǫµν . (A.19)
2 2
Remark that ∂µ (εν T µν ) ≈ 0, therefore εν T µν is sometimes called the conformal current. Let us
compute Z
d2 x ∂µ hεν (x)T µν (x)Xi , (A.20)
M

where M is a domain including all insertion points xi . We will also assume that inside this
domain, the functions ε(z) and ε̄(z) have no singularities (this can always be realized if ε and
Aspects of Two-Dimensional Conformal Field Theory 137

ε̄ are regular at x = xi ). By using (A.17) and eqs. (A.13) to (A.15), as well as ǫzz = −2i,
hi = 21 (∆i + si ), h̄i = 21 (∆i − si ), we get, using complex coordinates
Z X
d2 x ∂µ hεν (x)T µν (x)Xi = − (ε(zi )∂zi + ε̄(z i )∂zi + hi ∂zi ε(zi ) + h̄i ∂zi ε̄(z i ))hXi
M i

△
= δε,ε̄ hXi , (A.21)
R
where εz = ε(z), εz = ε̄(z). The l.h.s. can further be expanded with the relation M d2 x ∂µ F µ =
i
R  z z
 µ µν
2 ∂M −dzF + dzF , applied to F = εν (x)T (x). The fact that T µν is conserved, symmet-
ric and traceless except at coincident points implies that
1△
Tzz (z, z) = Tzz (z) = − T (z) (A.22)
2π
△ 1
Tzz (z, z) = Tzz (z) = − T̄ (z) (A.23)
2π
Tzz (z, z) = Tzz (z, z) = 0 , for (z, z) 6= (zi , z i ) , (A.24)

so that (A.21) finally reduces, because the domain M strictly contains all insertion points (and
so the contour C = ∂M does not go through the insertion points), to
I I
1 1
δε,ε̄ hXi = − dwε(w)hT (w)X i + dwε̄(w)hT̄ (w)X i . (A.25)
2πi 2πi
C C

In deriving this relation, we have used the property that the fields in the set X are primary,
through the Ward identities (A.13)-(A.15) (see their derivation p123 of [2]). However, eq. (A.25)
may be taken as a definition of the effect of conformal transformations on an arbitrary local field
within a correlation function, see [2], p122.

References
[1] Vl. Dotsenko, Série de Cours sur la Théorie Conforme, http://cel.ccsd.cnrs.fr/
cours/cel-2/cel-2.html
[2] Ph. Di Francesco, P. Mathieu, D. Sénéchal, Conformal Field Theory, Springer
[3] A.N. Schellekens , Introduction to conformal field theory, Fortsch. Phys, 1996, http:
//staff.science.uva.nl/∼jdeboer/stringtheory/CFT.ps
[4] Y. Saint-Aubin, Phénomènes critiques en deux dimensions et invariance conforme, Lecture
Notes, Université de Montreal (1987)
[5] E. Kiritsis,Introduction to Superstring Theory,hep-th/9709062
[6] C.J. Efthimiou, D.A. Spector, A Collection of Exercises in Two-Dimensional Physics, Part
1, hep-th/0003190
[7] J. Fuchs, Lectures on conformal field theory and Kac-Moody algebras, hep-th/9702194
[8] P.Ginsparg, Applied Conformal Field Theory, in “Fields, Strings and Critical Phenomena,
(Les Houches, Session XLIX, 1988) ed. by E. Brézin and J. Zinn Justin, 1989”, hep-
th/9108028
138 Detournay

[9] J. Bagger, Basic Conformal Field Theory, http://ccdb3fs.kek.jp/cgi-bin/img

index?200031975

[10] M.Green,J.Schwartz, E. Witten, Superstring theory, Cambridge University Press, 1987

[11] J.Polchinsky, String theory, Vol.1 ,Cambridge Univ. Pr. (1998)

 Semi-Simple Lie Algebras and Representations
Parts I and II

Nicolas Boulangera and Sophie de Buylb

a
Service de Mécanique et Gravitation
Université de Mons-Hainaut, Académie Wallonie-Bruxelles,
6, Avenue du Champ de Mars, 7000 Mons, Belgium

E-mail: [email protected]

b
Physique Théorique et Mathématique and International Solvay Institutes
Université Libre de Bruxelles, Académie Wallonie-Bruxelles,
Boulevard du Triomphe, C.P. 231, B-1050 Bruxelles, Belgium

E-mail: [email protected]

Abstract. These lectures consist in a pedagogical introduction to finite dimensional

representations of complex semi-simple Lie algebras. They are fully exemplified and in
particular illustrate the representation theory of su(2) and su(3) by the hadron theory.

Based on the lectures presented by N. Boulanger and S. de Buyl at the First Modave Summer
School in Mathematical Physics held at Modave, Belgium on June 19-25, 2005
140 Boulanger, de Buyl

1 The structure of Simple Lie Algebras

The goal of this section is to explain the construction of an explicit basis for any simple Lie
algebra. This construction exhibits the remarkable structure of these algebras and provides an
elegant way of describing all finite-dimensional irreducible representations of finite-dimensional
complex simple Lie algebras g. In order to achieve this, one needs to introduce the notion
of Cartan subalgebra, a very particular abelian subalgebra of g. The elements of a Cartan
subalgabra possess, in particular, the property that the other generators of the algebra g can be
taken to be the eigenvectors of these elements under the adjoint action. Generators chosen in this
way are called “ladder operators” and the eigenvalues are called “roots”. The next ingredient
is the Killing form. This form enables one to identify the Cartan subalgebra with its dual space
(the roots live in this dual space). Many properties of these roots are stated, that appear to be
extremely useful for the classification of the complex simple Lie algebras — although it is not
the purpose of the present lectures to go through this classification. One of the most important
properties of the basis that we are going to construct is that its generators naturally group
themselves into sl(2) generators. Using this decomposition and the representation theory of
sl(2), the representation theory of every simple Lie algebra can be achieved.

1.1 Basic definitions

A complex Lie algebra g is a vector space equipped with a binary operation [ , ] called a
commutator, mapping g × g into g with the following properties; ∀ x, y, z ∈ g and ∀ α, β ∈ C,

1. Antisymmetry : [x, y] = −[y, x] ;

2. Bilinearity : [αx + βy, z] = α[x, z] + β[y, z] ;
3. Jacobi identity : [[x, y], z] + [z, [x, y]] + [y, [z, x]] = 0 .

In this series of lectures, we will be mainly concerned with finite-dimensional complex Lie al-
gebras and their finite-dimensional representations. Real Lie algebras will be explicitly specified
by g(R).

A Lie algebra g can be specified by a set of generators {J b } and their commutations relations

[J b , J c ] = f bc d J d .

The dimension of the algebra is the dimension of the underlying vector space. The constants
f bc d are the structure constants.

Simple Lie algebras are non-abelian Lie algebras that contain no proper1 ideal. An ideal i of
g is a subspace of g such that [i, g] ⊂ i .

Semi-simple Lie algebras are Lie algebras that contain no abelian ideal and that are not
empty. Every semi-simple Lie algebra can be decomposed into the direct sum of its simple ideals
(see theorem in subsection 5.2 of [1]).
Example 1.1. A1 algebra: A1 ∼ = sl2 can be definedP
by its fundamental representation which consists of all
traceless antihermitian 2 × 2 matrices of the form a = 3p=1 λp ap , where λ1 , λ2 and λ3 are arbitrary complex
numbers and the ap ’s are ap ≡ 2i σp ,
     
1 0 i 1 0 1 1 i 0
a1 = , a2 = , a3 = , (1.1)
2 i 0 2 −1 0 2 0 −i
1A proper subset of a set is a set which is not the set itself neither the empty set.
Semi-simple Lie algebras and representations (Parts I and II) 141

satisfying
[ai , aj ] = −εijk ak . (1.2)

Levi theorem: Every Lie algebra can be decomposed into the direct sum of simple Lie algebras
and solvable algebras (a solvable algebra is a Lie algebra such that the series L(0) = L, L(1) =
[L(0) , L(0) ], L(k) = [L(k−1) , L(k−1) ] stops for some k).

A linear representation Γ of a Lie algebra g in the vector space V is an homomorphism2

Γ : g → gl(V ) : x → Γ(x) , Γ(x) : V → V : v → Γ(x) · v . (1.3)
The linear operator Γ(x) is sometimes simply denoted by x .

1.2 The Cartan-Weyl basis and roots

From now on, it will be understood that we are only treating semi-simple complex Lie algebras,
unless explicitly specified. In the Cartan-Weyl basis, the generators are constructed as follows.
We first find the maximal set of commuting generators H i , i = 1, ..., r , such that adH i is semi-
simple for all i (an endomorphism is called semi-simple if in a suitable basis it can be expressed
by means of a diagonal matrix). This set of operators forms the Cartan subalgebra 3 g◦ . The
rank of g is defined to be the dimension r of its Cartan subalgebra.
Example 1.2. Cartan subalgebra of A1 : Let a be an element of A1 , it is represented in the adjoint repre-
sentation by Γ(a) = ada acting on the algebra itself as in Eq.(1.4).
3
X
[a, aj ] = (ada )ij ai . (1.4)
i=1
Therefore, we find
0 1
0 1 0
ada3 =  −1 0 0 A, (1.5)
0 0 0
0 1 0 1
0 0 0 0 0 −1
ada1 = 0 0 1 A, ada2 = 0 0 0 A. (1.6)
0 −1 0 1 0 0
The subspace of matrices of the form λ3 ada3 may be taken as a Cartan subalgebra g◦ (adH 1 ≡ ada3 ). The
P
subspace of matrices of the form λ3 ada3 is certainly abelian and also maximally abelian because if a = 3p=1 µp ap
is such that [a, λ3 a3 ] = 0 then, by the commutation relations (1.2), µ1 = 0 = µ2 . Notice that ada2 and ada3
are also diagonalizable and they will lead to isomorphic g◦ .

Because the matrices adH i can be simultaneously diagonalized, the remaining generators are
chosen to be those particular combinations E α of the J a ’s that satisfy the following eigenvalue
equations
adH i E α = [H i , E α ] = α(H i )E α
where the α(H i ), also noted αi , are the (not all vanishing) components of elements of g⋆◦ , the
dual of the Cartan subalgebra g◦ . The non-zero elements α of g⋆◦ such that the set
gα = { X ∈ g | [H i , X] = αi X ∀ H i ∈ g◦ }
is not empty are called roots. The set of all roots of g is denoted ∆ and is called the root system
of g.
2 ∀x, y ∈ g and ∀ v ∈ V , [Γ(x), Γ(y)] · v = Γ([x, y]) · v and Γ(αx + βy) · v = αΓ(x) · v + βΓ(x) · v
3 An equivalent definition can be found in subsection 15.3 of [1]. It is also shown that all Cartan subalgebras
of finite-dimensional semi-simple Lie algebras are isomorphic (see the Corollary in subsection 15.3. (subsection
8.1. for definition) and subsection 16.4).
142 Boulanger, de Buyl

Example 1.3. Roots of A1 : From Equ. (1.2),

[H 1 , (a1 + ia2 )] = i(a1 + ia2 ) ,
[H 1 , (a1 − ia2 )] = −i(a1 − ia2 ) , (1.7)
so that there are two non-zero roots α1 and −α1 , with α1 (H 1 ) = i . Thus gα1 and g−α1 are subspaces of matrices
of the form λ(a1 + ia2 ) and µ(a1 − ia2 ) respectively (λ, µ ∈ C).

E α is the corresponding ladder operator. To get the commutation relations between the
ladder operators, one observes that the Jacobi identity implies

[H i , [E α , E β ]] = (αi + β i )[E α , E β ] .

If α + β ∈ ∆, the commutator [E α , E β ] must be proportional to E α+β . It certainly must

vanish if α + β is not a root. Formally,

α+β ∈ ∆ =⇒ [E α , E β ] = Nα,β E α+β , Nα,β ∈ C ,

α β
α+β ∈
/ ∆ =⇒ [E , E ] = 0 .

Note that, at this juncture, the possibility is not excluded that [E α , E β ] = 0 holds together with
α + β ∈ ∆ . This would correspond to Nα,β = 0 .4
When α = −β, the commutator [E α , E −α ] commutes with all H i , which is possible only if
it is a linear combination of the generators of the Cartan subalgebra (see Eq.(1.12) for precise
relation). Therefore, the full set of commutation relations in the Cartan-Weyl basis is

[H i , H j ] = 0
[H i , E α ] = αi E α
[E α , E β ] = Nα,β E α+β
P for some Nα,β ∈ C if α + β ∈ ∆ ,
r i
= i=1 ai H for some ai ∈ C if α = −β ,
= 0 otherwise .

1.3 The Killing form and properties of roots

In order to know the Cartan-Weyl basis more explicitly, more must be said about the root sys-
tem ∆. The Killing form plays an important role in this perspective. This form establishes an
isomorphism between g◦ and g⋆◦ which contains ∆.

A bilinear and symmetric form on g, called Killing form, can be defined by

K(x, y) := tr(adx ady ) ∀x, y ∈ g .

A very important property of the Killing form is its invariance (under the adjoint action)

K(x, [y, z]) = K([x, y], z) ∀ x, y, z ∈ g.

Proposition 1.4 (Cartan’s criterion). A finite-dimensional Lie algebra g is semi-simple if and

only if its Killing form is non-degenerate. [For a proof see e.g. [1], subsection 5.1.]
4 However,
this never occurs (cf. Proposition 1.11, subsection 1.5), so that one actually has Nα,β 6= 0 and
/ ∆ ⇐⇒ [E α , E β ] = 0 .
α+β ∈
Semi-simple Lie algebras and representations (Parts I and II) 143

The fundamental role of the Killing form is to establish an isomorphism between g◦ and
g⋆◦ . Before making it precise, it is necessary to show that the restriction of the Killing form
to g◦ is non-degenerate. We must prove that K(h, h′ ) = 0 ∀ h′ ∈ g◦ implies that h = 0 ,
with h ∈ g◦ . By Cartan’s criterion, we know that K(h, x) = 0 ∀ x ∈ g implies that
h = 0 , so it suffices to show that K(h, x) = 0 ∀ x such that x ∈ g and x ∈ / g◦ . If x ∈ g
and x ∈ / g◦ , then we can find and h′ ∈ g◦ such that [h′ , x] = α x and α 6= 0 . Therefore
K(h, x) = α1 K(h, [h′ , x]) = α1 K([h, h′ ], x) = 0 .

To every γ ∈ g⋆◦ corresponds, up to a normalization factor cγ , an element H γ ∈ g◦ through

γ(h) = cγ K(h, H γ ) ∀ h ∈ g◦ . (1.8)

To avoid the normalization factor, we also define hγ := cγ H γ , giving

γ(h) = K(hγ , h) ∀ h ∈ g◦ . (1.9)

With this isomorphism, the Killing form can be transferred into a (non-degenerate) inner product
on g⋆◦ :

(γ, β) := K(hβ , hγ ) = cβ cγ K(H β , H γ ) . (1.10)

Using the Killing form, the commutation relation between E α and E −α can be established. To
do so, one needs to evaluate the Killing form between two step operators E α and E β such that
α + β 6= 0. The endomorphism adE α ◦ adE β maps gγ into gγ+α+β . Since γ + α + β 6= γ this map
has zero trace and

K(E α , E β ) = 0 with α + β 6= 0 . (1.11)

Since the Killing form is non degenerate, K(E α , E β ) cannot be zero for all root spaces gβ .
Thus there exists a non-zero element E −α ∈ g−α such that K(E α , E −α ) 6= 0. This result
will be helpful for the evaluation of the Killing form between E α and E −α for α 6= 0. One
uses the invariance property with x = h, y = E α and z = E −α and takes into account that
[h, E α ] = α(h)E α , which gives us the relation K(h, [E α , E −α ]) = α(h)K(E α , E −α ). This last
relation tells us that [E α , E −α ] 6= 0 because K(E α , E −α ) 6= 0 and for each α there exists at
least one h ∈ g◦ such that α(h) 6= 0. Using again the invariance property and Eq.(1.8), one
learns that K(h, [E α , E −α ]) = cα K(h, H α )K(E α , E −α ) for all h ∈ g◦ . The Killing form being
non degenerate, one has

[E α , E −α ] = cα K(E α , E −α )H α = K(E α , E −α )hα . (1.12)

Example 1.5. Killing form of A1 : An explicit expression for hα1 will now be found. From α1 (hα1 ) =
K(hα1 , hα1 ) with hα1 = κH 1 , κ2 K(H 1 , H 1 ) = κα1 (H 1 ). By using Eq. (1.5), one finds K(H 1 , H 1 ) =
K(a3 , a3 ) = −2, so κ = − 21 i since α1 (H 1 ) = i . Thus
 
1 1 0
hα1 = (1.13)
4 0 −1

We have (α1 , α1 ) = α1 (hα1 ) = κ2 K(H 1 , H 1 ) = (−i/2)2 (−2) = 1

2
which is real, positive and rational. With
(1.5) and (1.6) we find K(ai , aj ) = −2δij .
144 Boulanger, de Buyl

At this stage and before proceeding, we list some important results concerning the root
system ∆ of (finite-dimensional) semi-simple complex Lie algebras. These results are proved in
e.g. [2] Chapt. 13, Sect. 5.
⋆ The roots span all g⋆◦ : spanC (∆) = g⋆◦ [From this result and by the isomorphism (1.9)
between g◦ and g⋆◦ , a subset of r linearly independent elements {hβi }ri=1 ({βi }ri=1 ∈ ∆)
may be selected from {hα }α∈∆ and thus forms a basis of g◦ .];
⋆ The root spaces gα are one-dimensional (gα = spanC {E α }), i.e. the roots are non-
degenerate;
⋆ The only multiples of α ∈ ∆ which also belong to ∆ are ±α ;
⋆ For every α, β ∈ ∆, (α, β) ∈ Q. Moreover, for every α ∈ ∆, (α, α) > 0 ;
Pr
⋆ Every non-zero α ∈ ∆ can be written α = i=1 κi βi where the κi , 1 6 i 6 r , are rational
numbers;
⋆ On g◦ (R), the real vector space with basis {H i }ri=1 (or {hβi }ri=1 ), we have α(H i ) ∈ R ∀ i
and ∀ α ∈ ∆.
With the Cartan-Weyl basis, for any h ∈ g◦ , adh is a diagonal matrix with zero diagonal
elements corresponding to the basis elements H i and with diagonal element γ(h) corresponding
to each E γ . Since the root spaces are one-dimensional, we have thus
X
K(h, h′ ) = tr(adh adh′ ) = γ(h)γ(h′ ) ∀ h, h′ ∈ g◦ . (1.14)
γ∈∆

In particular, with h = hα and h′ = hβ we have

X
(α, β) = (α, γ)(β, γ) . (1.15)
γ∈∆

In particular, for λ ∈ g⋆◦ , the norm squared of λ is a sum of squares:

X
(λ, λ) = (γ, λ)2 . (1.16)
γ∈∆

From the results listed above together with the relations (1.14)—(1.16), it follows that K(h, h′ )
is real for all h, h′ ∈ g◦ (R). Moreover, for all h ∈ g◦ (R), K(h, h) > 0 and K(h, h) = 0 implies
h = 0.
The Killing form thus provides an euclidean metric on the root space g⋆◦ (R), so that we have
g⋆◦ (R) = Rr .
∼

This highly non-trivial fact is one of the key ingredients in the classification of all finite-
dimensional simple Lie algebras.
Definition 1.1. Suppose that α, β ∈ ∆. Then the α-string of roots containing β is the set of
all roots of the form β + kα, where k is an integer.
Proposition 1.6. Let α, β ∈ ∆. Then there exists two non-negative integers p and q (which
depend on α and β) such that β + kα is in the α-string containing β for every integer k that
satisfies the relation −p 6 k 6 q. Moreover, p and q are such that
2(β, α)
p−q = . (1.17)
(α, α)
Semi-simple Lie algebras and representations (Parts I and II) 145

Also,
2(β, α)
β− α (1.18)
(α, α)
is a non-zero root. For a proof, see e.g. Ref. [2] Theorem IX p. 510.

Proposition 1.7. For all α, β ∈ ∆, 2(β, α)/(α, α) can take only the integral values 0, ±1, ±2
or ±3. [The quantities 2(β, α)/(α, α) are called the ‘Cartan integers’.]
Proof:
That 2(β, α)/(α, α) must be an integer is an immediate consequence of Equation (1.17), as p and q
are integers. Clearly, with β = ±α, 2(β, α)/(α, α) = ±2. Suppose then that β 6= ±α, so that hα and
hβ are linearly independent. As the Killing form K provides an inner product for g◦ (R), applying
the Schwarz inequality gives |K(hα , hβ )|2 < K(hα , hα )K(hβ , hβ ) or |(α, β)|2 < (α, α)(β, β) so that
|2(α, β)/(α, α)| · |2(α, β)/(β, β)| < 4, from which it follows that |2(α, β)/(α, α)| can only take values
0, 1, 2 or 3 .

1.4 Simple roots, Cartan matrix and Dynkin diagram

Let hβ1 , hβ2 , . . . hβr be a set of r linearly independent elements of g◦ , where βi ∈ ∆, 1 6 i 6 r .
As we mentioned in the previous subsection, every non-zero root of ∆ can be written in the form
r
X
α∈∆ ⇒ α= κi βj , κi ∈ Q , i = 1, . . . , r . (1.19)
j=1

Definition 1.2. A non-zero root of ∆ is said to be positive (with respect to the basis {βi }ri=1 )
if the first non-vanishing coefficient of the set {κi }ri=1 appearing in Equation (1.19) is positive.
One then writes α > 0 . The set of all positive (resp. negative) roots is denoted ∆+ (resp. ∆− ).
A simple root is a positive root that cannot be written as the sum of two positive roots. The
set of simple roots will be denoted by {α(i) }pi=1 .
Clearly, if α ∈ ∆+ (which is also denoted by α > 0) then −α is not positive, so that exactly
one half of the set of non-zero roots are positive roots. Also, if α > 0 and β > 0, then α + β > 0
(provided α + β ∈ ∆).
Definition 1.3. Lexicographic ordering of roots
Let α and β be any two roots of ∆ . Then if α − β is positive, (α − β) > 0 , one writes α > β .
Theorem 1.8. If α and β are two simple roots of ∆ and if α 6= β, then
(1) α − β is not a root of ∆ ; and
(2) (α, β) 6 0 .
Proof:
(1) Suppose that α − β is a root of ∆. Then, if α − β is positive, as α = (α − β) + β, α cannot be
simple. On the other hand, if α−β is not positive then β −α is positive and, as β = (β −α)+α,
β cannot be simple. In either case there is a contradiction with the initial assumptions, so
α − β cannot be a root of ∆.
(2) Let β − pα, . . ., β + qα be the α-string of roots containing β. Then, by Equation (1.17),
2(β,α)
p − q = (α,α) . But, by point (1), p = 0 and it was mentioned in the previous subsection that
(α, α) is real (rational, as a matter of fact) and positive for every α ∈ ∆ so (β, α) = (α, β) 6 0 .
146 Boulanger, de Buyl

Theorem 1.9. If g has rank r then g possesses precisely r simple roots {α(i)P }ri=1 . They form a
r
basis of the dual space g◦ . Moreover, if α is any positive root of ∆, then α = i=1 ki α(i) , where
⋆

the coefficients {ki }ri=1 are a set of non-negative integers.

For a proof, see e.g. Ref. [2] p. 522.
Pr (i)
Definition
Pr 1.4. If α = i=1 ki α ∈ ∆+ , the level or height of α is the positive integer
k = i=1 ki .
With the properties of the simple roots established, the next stage is to introduce the Cartan
matrix which plays a crucial role in group theory.
Definition 1.5. The Cartan matrix A of g is an r × r matrix whose elements Aij are defined
in terms of the simple roots {α(i) }ri=1 of g by

2(α(i) , α(j) )
Aij = , i, j = 1, . . . , r . (1.20)
(α(j) , α(j) )

Clearly Aii = 2 ∀ i ∈ {1, . . . , r}, while Proposition 1.7 and part (2) of Theorem 1.8 above
together imply that for i 6= j the only possible values of Aij are 0, −1, −2 or −3. Moreover, Aij =
0 if and only if Aji = 0 . (This follows because (α(i) , α(j) ) = 21 Aij (α(j) , α(j) ) and (α(j) , α(i) )(=
(α(i) , α(j) )) = 12 Aji (α(i) , α(i) ) . As (α(i) , α(i) ) and (α(j) , α(j) ) are both non-zero, Aij = 0 if and
only if (α(i) , α(j) ) = 0, that is, if and only if Aji = 0 .)

The coroot of the simple root α(i) is

2α(i)
α(i)∨ := .
|α(i) |2
P
There Pis a highest root Θ, the unique root for which, in the expansion mi α(i) , the height
(or level) mi is maximized (see e.g. Ref. [1] subsection 10.4. Lemma A).

One introduces a “weight” wi for each simple root α(i) , defined by wi = w(α(i) , α(i) ), where
w is a constant independent of i chosen so that the minimum value in the set w1 , w2 , . . . , wr is
1 . Explicitly, w = min(|α(i) |−2 ), 1 6 i 6 r . By Equation (1.20), wi /wj = Aij /Aji (provided
Aji 6= 0). As Aij 6 0 for i 6= j , it follows that for i 6= j
√ q
Aij = − Aij Aji wi /wj . (1.21)

One depicts a so-called Dynkin diagram for each simple complex Lie-algebra g where each simple
root α(i) of g is assigned a point (a vertex) with its weight wi and Aij × Aji lines are drawn
between the vertices corresponding to α(i) and α(j) . Then, since Aii = 2 ∀ i ∈ {1, . . . , r} , it is
obvious from Equation (1.21) that the Dynkin diagram of g completely determines the Cartan
matrix of g. For Cartan matrices which differ only by renumbering the simple roots (and, as a
consequence, describe the same Lie algebra), one obtains identical Dynkin diagrams.
Example 1.10. Cartan matrices and Dynkin diagrams : The Cartan matrices of A2 , B2 ∼
= C2 and G2 are
 
2 −1
A= (1.22)
−x 2
with x being respectively 1, 2 and 3. The corresponding Dynkin diagrams are depicted as follows.
Semi-simple Lie algebras and representations (Parts I and II) 147

1 1 1 2 1 3
i i i i i i

Actually, the classification of simple finite-dimensional Lie algebras is achieved by the presen-
tation of their respective Dynkin diagrams. A table of Dynkin diagrams for the finite-dimensional
simple Lie algebras can be found in e.g. Ref. [3] p. 114, or in Ref. [2] p. 524. Notice that there
exists different conventions for Dynkin diagrams.

1.5 The remaining commutation relations, conventions

Suppose α, β and α + β belong to the set of roots ∆, with β 6= −α. Let eα , eβ and eα+β be basis
elements of gα , gβ and gα+β respectively. As we have seen in subsubsection 1.2, there exists a
complex number Nα,β such that

[eα , eβ ] = Nα,β eα+β , Nα,β ∈ C . (1.23)

The properties of the Nα,β are embodied in the following three propositions5 . The canonical
basis of Weyl is such that their properties take particularly simple forms.
Proposition 1.11. If α, β and (α + β) ∈ ∆ then Nα,β 6= 0 .
The Killing form does not provide any natural normalization of the eα ∈ gα , α ∈ ∆, since
K(eα , eα ) = 0 (cf. Eq. (1.11)). Several different sets of conventions for the choice of the eα will
be found in the mathematics literature. To avoid confusion, the properties of the Nα,β will be
summarized for an arbitrary choice of the eα ’s, α ∈ ∆ .
Proposition 1.12. Let Nα,β be the structure constants defined by Equation (1.23) and let
Kα := K(eα , e−α ) (α ∈ ∆). Then
(a) Nβ,α = −Nα,β ,
(b) if α, β, γ ∈ ∆ and α + β + γ = 0, then

Nα,β Kγ = Nβ,γ Kα = Nγ,α Kβ ; (1.24)

(c) if α, β, γ, δ ∈ ∆ are such that the sum of no two of them is zero, and if α + β + γ + δ = 0,
then

Nα,β Nγ,δ Kα+β + Nβ,γ Nα,δ Kβ+γ + Nγ,α Nβ,δ Kα+γ = 0 ; (1.25)

(d) for any α, β ∈ ∆

1 Kα Kβ
Nα,β N−α,−β = − (α, α) q(p + 1) , (1.26)
2 Kα+β

where p and q are such that the α-string containing β is β − p α, . . ., β, . . ., β + q α.

5 Thefirst proposition is proved in e.g. Ref. [2] p.512. A proof of the last two propositions can be found in the
same reference: Appendix E, subsection 8 (pp. 829—831).
148 Boulanger, de Buyl

Proposition 1.13. With K(eα , e−α ) taking any assigned value Kα for each pair of roots α,
−α ∈ ∆, the basis elements of g may be chosen so that either Nα,β = N−α,−β ∀ α, β ∈ ∆ or
Nα,β = −N−α,−β ∀ α, β ∈ ∆.

It should be noted that both the choices Nα,β = N−α,−β or Nα,β = −N−α,−β are allowed
for any arbitrarily chosen set of values for the quantities Kα := K(eα , e−α ). Clearly, with
Nα,β = N−α,−β , Eq. (1.26) gives
1 Kα Kβ
(Nα,β )2 = − (α, α) q(p + 1) , (1.27)
2 Kα+β

whereas with Nα,β = −N−α,−β Equation (1.26) gives

1 Kα Kβ
(Nα,β )2 = + (α, α) q(p + 1) . (1.28)
2 Kα+β
There are definite advantages in having the Nα,β all real . This can be achieved either
(i) by taking Kα to be real and negative for all pairs α and −α of ∆ and by taking Nα,β =
+N−α,−β ; or
(ii) by taking Kα to be real and positive for all pairs α and −α of ∆ and by taking Nα,β =
−N−α,−β .

Here are three sets of conventions widely used in the mathematics literature:
(A) For all pairs α and −α of ∆,

Kα ≡ K(eα , e−α ) = −1 , (1.29)

and for all α, β ∈ ∆,

N−α,−β = Nα,β . (1.30)

With this convention the Nα,β are all real and the relation (1.12)

[aα , a−α ] = K(aα , a−α )hα (aα ∈ gα , aα ∈ g−α ) (1.31)

gives

[eα , e−α ] = −hα . (1.32)

Equation (1.24) becomes

Nα,β = Nβ,γ = Nγ,α , (1.33)

Equation (1.25) becomes

Nα,β Nγ,δ = Nβ,γ Nα,δ = Nγ,α Nβ,δ = 0 (1.34)

and Equation (1.27) reduces to

1
(Nα,β )2 = (α, α)q(p + 1) . (1.35)
2
Semi-simple Lie algebras and representations (Parts I and II) 149

(B) Kα = 1 for all pairs α and −α of ∆ and N−α,−β = −Nα,β ∀ α, β ∈ ∆ . With this
convention the Nα,β are again all real and Equations (1.24), (1.25) and (1.28) reduce to
Equations (1.33) (1.34) and (1.35) respectively.
(C) For all pairs α and −α of ∆ ,
2
Kα ≡ K(eα , e−α ) = (1.36)
(α, α)
and for all α, β ∈ ∆ ,
N−α,−β = −Nα,β . (1.37)
Again, with this convention the Nα,β are all real, but Equations (1.24) become
Nα,β Nβ,γ Nγ,α
= = , (1.38)
(γ, γ) (α, α) (β, β)
Equation (1.25) becomes
Nα,β Nγ,δ Nβ,γ Nα,δ Nγ,α Nβ,δ
+ + = 0, (1.39)
(α + β, α + β) (β + γ, β + γ) (α + γ, α + γ)
and Equation (1.28) gives
(α + β, α + β)
(Nα,β )2 = q(p + 1) . (1.40)
(β, β)
Furthermore, case-by-case examination (e.g. Exercise 6.10 of [3]) shows that
(α+β,α+β)
(β,β) q = p + 1 , so that Equation (1.40) reduces to

Nα,β = ±(p + 1) . (1.41)

Thus, with this convention the structure constants Nα,β are all integers.
Now, to make contact with the generators used in subsection 1.3, one defines H α ∈ g◦ by
2 1
H α := hα (= hα ) . (1.42)
(α, α) cα
Then, one chooses E α ∈ gα and E −α g−α such that Equ. (1.36) is satisfied
2
K(E α , E −α ) = . (1.43)
(α, α)
Since by definition E ±α ∈ g±α one has
[h, E ±α ] = ±α(h)E ±α ∀ h ∈ g◦ . (1.44)
α −α
Then, with K(hα , h) = α(h) ∀ h ∈ g◦ , α(hβ ) = β(hα ) = (α, β), [E , E ] =
K(E α , E −α )hα , (1.42) and (1.43), one finds
[H α , E ±α ] = ±2E ±α ,
[E α , E −α ] = Hα . (1.45)
Thus all the results on the representations of sl(2) [and su(2)] apply immediately to this
subalgebra.
Note that the analogue of Equation hα+β = hα + hβ is not true in general for the elements
(α,α) (β,β)
H α . That is, in general H α+β 6= H α +H β . Instead, H α+β = (α+β,α+β) H α + (α+β,α+β) Hβ.
150 Boulanger, de Buyl

Of course, in all of these three cases, with eα ∈ gα and for all h ∈ g◦ , one has

[h, eα ] = α(h)eα (1.46)

and in particular

[hβ , eα ] = α(hβ )eα = (α, β)eα . (1.47)

Each of these sets of conventions has both advantages and disadvantages. To enable all the most
convenient properties to be available, it will be assumed that the basis elements of gα and g−α
denoted by eα and e−α satisfy Equation (1.29), while the alternative basis elements of gα and
g−α denoted by E α and E −α satisfy Equation (1.36)
K(E α , E −α ) = 2/(α, α)
as in Equation (1.43), and that the two sets are related by the equations
q 
Eα = 2 
(α,α) eα ,
q (1.48)
E −α = − (α,α) 2
e−α , 

It can be shown that, in the case of a Lie algebra of matrices, it is possible, after an appropriate
similarity transformation if necessary, to take the elements hα of g◦ to be diagonal Hermitian
matrices, while for each pair α and −α of ∆ the matrices eα and e−α may be chosen so that
e−α = −e†α (1.49)
and correspondingly, by Equations (1.48),
†
E−α = (E−α ) . (1.50)

The Chevalley basis and Serre relations

The Chevalley basis is the particular Cartan-Weyl basis corresponding to the case (C) above.
This basis is the one most useful for studying representations.
(i)
• One denotes hi := H α the Cartan generators associated with the simple roots α(i) by
Eq. (1.8). Remind that the coefficient cα(i) appearing in the latter equation is taken
here, in the Chevalley basis, to be cα(i) = (α(i) , α(i) )/2. Also, do not confuse hi with
(i) (α(i) ,αi ) (i)
hα(i) := cα(i) H α = 2 H α . In general hi 6= hα(i) ;
(i)
• The step operators associated with the simple roots are denoted by ei := E α and those
(i)
associated with minus the simple roots are denoted by f i := E −α . Remind that the
normalization of these generators is such that (1.43) is satisfied.
It is then straightforward to see, from the formulas obtained up to now, that these generators
obey the following commutation relations

[hi , hj ] = 0,
[hi , ej ] = Aji ej ,
[hi , f j ] = − Aji f j ,
[ei , f j ] = δ ij hi .
Semi-simple Lie algebras and representations (Parts I and II) 151

Moreover, their multi-commutators are subject to the so-called Serre relations,

1−Aji j
(adei ) e = 0,
1−Aji j
(adf i ) f = 0.

The Serre relations contain, in a compact way, the information about the full root system. They
determine which multiple brackets of step operators must vanish and hence fix the length of the
α(i) -string through α(j) . The full set of relations listed above is known as the Chevalley–Serre
relations. These relations not only summarize the structure of g in a very compact form, they
also characterize g uniquely. On the particular following example, it is shown how a whole Lie
algebra g can be reconstructed uniquely from the Chevalley–Serre relations. This construction
can be enlarge to include less restricted Cartan matrices (allow Aij with i 6= j to be any negative
integer) and lead the so-called Kac-Moody algebras which are infinite-dimensional.

Example 1.14. The root system of B2 ∼

= C2 , the complexification of so5 ∼
= sp2
The Cartan matrix (deduced from the corresponding Dynkin diagram) is
 
2 −2
A= . (1.51)
−1 2

The positive roots of level 1 are just the simple roots, all other positive roots being of a higher level (cf. Definition
1.4). Since the rank is 2, we have two simple roots α(1) and α(2) . Computing the α(1) -string through α(1) gives
no new positive root, of course. Similarly for the α(2) -string through α(2) . The α(1) -string through α(2) ( i.e.
2(α(2) ,α(1) )
α(2) − pα(1) , . . . , α(2) , . . . , α(2) + qα(1) ) is characterized by p − q = = A21 = −1. One deduces q = 1
(α(1) ,α(1) )
because p = 0 (α(i) − α(j) is never a root). We thus have the positive root α(2) + α(1) at level 1. Computing
the α(2) -string through α(1) , we find again α(2) + α(1) . However, since A12 = −2, we also find α(1) + 2α(2) , at
level 2. There is no other positive root, so that there are 4 of them : {α(1) , α(2) , α(1) + α(2) , α(1) + 2α(2) } and
the dimension of B2 ∼ = C2 is 2 + 4 + 4 = 10 . Then, one can write down all step (or ladder) operators E α of g
(i)
as multiple Lie brackets of the E ±α .

In order to classify finite-dimensional complex simple Lie algebras up to isomorphism, one

only needs to classify the possible Cartan subalgebras and root systems, which is done by using
the Chevalley–Serre relations and the algorithm explained in the previous example. The only free
parameters occurring in the Chevalley–Serre relations are the entries of the Cartan matrix. The
classification of simple Lie algebras therefore amounts to the classification of the Cartan matrices.
This classification has been known for some time, is referred to as the Cartan classification and
is conveniently presented in tables of Dynkin diagrams. We showed in the previous subsection
how the Cartan matrices can be recovered from the corresponding Dynkin diagrams.

Orthonormal basis

In the theoretical physics literature, a different choice of basis of the Cartan subalgebra g◦ is
often encountered. The Killing form of g provides an inner product for the real vector space
g◦ (R). Consequently a basis may be set up in g◦ (R) that is orthonormal with respect to the
Killing form, and this basis is also a basis g◦ (R). Thus there exist r elements Hi , i = 1, . . . , r of
g◦ (R) such that

K(Hi , Hj ) = δij , i, j = 1, 2, . . . , r . (1.52)

152 Boulanger, de Buyl

Then, for any linear functional α(h) defined on g◦ , the element hα defined by Equation (1.9) is
given by6
r
X
hα = α(Hi )Hi . (1.53)
i=1

The advantage of this basis is that (α, β) can always be expressed in a particularly simple form.
Indeed for any two linear functionals α and β belonging to g⋆◦ , Equations (α, β) := K(hα , hβ )
and (1.53) imply that

r
X
(α, β) = α(Hi )β(Hi ) . (1.54)
i=1

Every linear functional ∈ g⋆◦ is completely specified by its values on a basis of g◦ . In particular,
α(h) is completely specified by the sets of values α(Hi ), i = 1, . . . , r which are all real whenever
α is a root of g. This set can be written as an r-component vector α ~,

α
~ = (α(H1 ), . . . , α(Hr )) .

~ of two such vectors α

~ ·β
Consequently, if the scalar product α ~ is defined by
~ and β
r
X
α ~=
~ ·β α(Hi )β(Hi ) ,
i=1

that is, as the inner product of Rr , Equation (1.54) becomes simply

(α, β) = α ~.
~ ·β (1.55)

With this choice of basis, Equation (1.46) gives

adHi eα = [Hi , eα ] = α(Hi )eα

for each α ∈ ∆, and Equations (1.32) and (1.53) give

r
X
[eα , e−α ] = − α(Hi )Hi
i=1

for each α ∈ ∆, both relations involving only components of α~ . It may be noted that adHi is
a diagonal matrix with zero diagonal elements corresponding to the basis elements of g◦ and
α(Hi ) corresponding to the basis (non-degenerate) element eα , tr(adHi ◦
with diagonal elementsP
adHj ) = K(Hi , Hj ) = α∈∆ α(Hi )α(Hj ), so the condition in Equation (1.52) implies that
X
α(Hi )α(Hj ) = δij , i, j = 1, . . . , r .
α∈∆

6 As
P
P
Hi , i = 1, . . . , r form a basis of g◦ , hα = µiα Hi . But, by Equation (1.9), K(hα , Hj ) = α(Hj ), so
i
i implies that µiα = α(Hi ).
i µα K(Hi , Hj ) = α(Hj ), whence Equation (1.52)
Semi-simple Lie algebras and representations (Parts I and II) 153

1.6 Compact semi-simple real Lie algebras

If g is a complex Lie algebra, with the basis elements a1 , . . . , an chosen so that the structure
constants f ijk are all real , then real linear combinations of a1 , . . . , an form a real Lie algebra g
that is one of the “real forms” of g . If g is a finite-dimensional simple Lie algebra over C, one can
ask what are the real Lie algebras g(R) such that g is the complexification of it: g ∼ = g(R) ⊗R C.
One can associate with g several distinct Lie algebras over R, called real forms of g; and there is
always a unique real Lie algebra gc called the compact real form of g such that, in an appropriate
basis, its Killing form is negative definite. For more details, see Lecture 26 of [4], Chapt. 8 of [3]
or Chapt. 14 of [2].

Definition 1.6. Compact and non-compact semi-simple real Lie algebras

A semi-simple real Lie algebra g is said to be “compact” if its Killing form is negative definite
(that is, if K(a, b) < 0 ∀a ∈ g such that a 6= 0), and is said to be “non-compact” otherwise.

This terminology is used because compact semi-simple real Lie algebras correspond to com-
pact Lie groups. The words “compact” and “non-compact” do not describe the topology of g
itself, which, being a vector space of dimension n, is isomorphic to the whole of Rn which is
unbounded and consequently always non-compact in the usual topology on Rn .

Example 1.15. su(2) as a compact semi-simple real Lie algebra

In part (i) of Example 1.5 one reads K(ai , aj ) = −2δij . Thus, for a general element a of su(2), writing
P
a = 3i=1 µi ai with {µi }3i=1 all real it follows that K(a, a) < 0 ∀ a 6= 0 . Thus su(2) is a compact Lie algebra.
su(2) is the real Lie algebra of the Lie group SU (2) which is a compact Lie group. This provides a particular
example of the following general Theorem 1.17.

If g(R) is a compact semi-simple real Lie algebra, then with ha, bi defined for all a, b ∈ g by

ha, bi := −K(a, b) , (1.56)

it follows that h· , ·i is a scalar product. It is then possible to set up an orthonormal basis in

g(R) .

Proposition 1.16. If a1 , . . . , an is an orthonormal basis of a compact semi-simple real Lie

algebra g(R) then the adjoint representation of g(R) defined relatively to this basis consists of
antisymmetric matrices. Put another way, with this basis the structure constants f ijk are com-
pletely antisymmetric under the exchanges of its three indices: f ijk = −f ikj = −f kji = f kij =
−f jik = f jki .

Proof:

For any a ∈ g and i, j = 1, 2, . . . , n, K([ai , a], aj ) = K(ai , [a, aj ]) . Thus, by linearity of the
P Pn
Killing form, − n k=1 [ada ]ki K(ak , aj ) = k=1 [ada ]kj K(ai , ak ) , from which it follows, using
K(ai , aj ) = −δij , that [ada ]ji = −[ada ]ij . The relation f ijk = −f ikj applies for any basis of any
Lie algebra, while the other antisymmetry properties needed for the complete antisymmetry of f ijk
follow from [adaj ]ik = f ijk and [ada ]ji = −[ada ]ij .

Theorem 1.17 (Weyl). A connected semi-simple linear Lie group G is compact if and only if
its corresponding real Lie algebra g is compact.
154 Boulanger, de Buyl

For a proof, see e.g. Ref. [2] Appendix E, subsection 10, with Weyl’s “unitary trick ”.

The following theorem (proved in the same Appendix E, subsection 10 of Ref. [2]) shows that
every complex Lie algebra g has a compact real form, derivable from g in a very straightforward
manner.

Theorem 1.18. For each semi-simple complex Lie algebra g , if hα(i) (j = 1, . . . ,r) and eα , e−α
(α ∈ ∆+ ) are the basis elements of the Weyl canonical form, then the set of elements

ihα(i) , ({α(i) }ri=1 being the simple roots of g̃)

eα + e−α for each α ∈ ∆+ ,

i(eα − e−α ) for each α ∈ ∆+

form the basis of a compact semi-simple real Lie algebra denoted by gc and referred to as the
(unique up to isomorphism) “compact real form” of g .

It is easily verified that the following elements provide a natural orthonormal basis for gc
with respect to the inner product ha, bi := −K(a, b) :

iHi i = 1, . . . , r ,

1
√ (eα + e−α ) for each α ∈ ∆+ ,
2

i
√ (eα − e−α ) for each α ∈ ∆+ ,
2
{Hi }ri=1 being the orthonormal basis of g◦ , see Eq. (1.52).
In the case of Lie algebra of matrices, as every representation of a compact Lie group is
equivalent to a unitary representation, the matrices of the corresponding compact real Lie algebra
gc can be taken to be anti-Hermitian. The previous theorem then implies that

h†α = hα

and
e†α = −eα

for all α ∈ ∆, as mentioned earlier [see Eq. (1.49)].

2 Highest Weight Representations

The key idea in the analysis of finite-dimensional representations of simple Lie algebras is to
reduce the problem to the representation theory of sl(2) algebras. This is possible because each
H α together with E α and E −α span a sl(2) subalgebra in the Chevalley basis. Note that we
will only be interested in finite-dimensional representations of simple Lie algebras.
Semi-simple Lie algebras and representations (Parts I and II) 155

2.1 Representations of sl(2)

The generators h, e and f of sl(2) obey (cf. Equations (1.45))

[h, e] = 2e [h, f ] = −2f [e, f ] = h . (2.1)

Let V be the vector space of a finite dimensional irreducible representation of sl(2). Each operator
on a complex vector space possesses at least one eigenvalue. Let v ∈ V be the eigenvector with
eigenvalue λ of h (we denote the element of gl(V ) associated with h by h),

h v = λ v.

From Eq. (2.1), one gets

h (e v) = e (h v) + 2 e v = (λ + 2) (e v)

which means that, if e v 6= 0, e v is an eigenvector of h with eigenvalue λ + 2. Similarly, if

f v 6= 0, f v is a proper vector of h with proper value λ − 2. Acting repetitively with e and f on
v, one gets a chain of eigenvectors of h with eigenvalues ..., λ − 4, λ − 2, λ + 2, λ + 4, ... As we
are interested in finite dimensional representation, this chain must stop, i.e. there exists k and
l such that ek v = 0 and f l v = 0. Let v0 be the last “upper” vector of the chain, i.e. e v0 = 0
and f v0 6= 0, and µ its eigenvalue with respect to h. One can describe the chain starting from
v0 , all vectors are obtained by applying f . One defines vj = f j v0 . By direct calculation, one
gets

h vj = (µ − 2j) vj e vj = j(µ − j + 1) vj−1 f vj = vj+1 . (2.2)

Again, since we demand a finite-dimensional representation, we also have f p+1 v0 = 0 for a given
p ∈ Z+ . It can then be checked that the relation µ = p must be verified. The vector space W
spanned by v0 , f v0 ,... , f p v0 is invariant under the action of sl(2) and furnish a representation
of this algebra. This representation is irreducible because each vector of W can be mapped to
any other vector of W by the action of es , f r . Thus W = V . The proper values of h are called
weights of the representation, p is the maximal weight and v0 the vector of maximal weight. One
concludes with the
Theorem 2.1. For all integer p > 0 there exists a unique, up to isomorphism, irreducible
representation of dimension p+1 of maximal weight p. One can always choose a basis {v0 , ..., vp }
of the representation space such that e, f and h acts as in Eq.(2.2) with µ = p and j = 0, ..., p.

2.2 Weights of a representation

For any simple Lie algebra g, with the convention corresponding to the Chevalley basis, each
H α together with E α and E −α ≡ F α span an sl2 subalgebra (cf. Eqs. (1.45)). These generators
obey the sl2 commutation relation given by Eq. (2.1) with the following identifications,

Hα → h Eα → e E −α → f . (2.3)

One learns therefore that for an arbitrary representation V of g, a basis {|λi} can always be
found such that the whole Cartan subalgebra acts diagonally on it. Indeed, su2 acts on V and
the representation theory of this algebra tell us that there exists a basis of V such that H α acts
(i)
diagonally on it. The hi ’s (= the H α ’s in the Chevalley basis) spanning the Cartan subalgebra
156 Boulanger, de Buyl

all commute and therefore the whole Cartan subalgebra acts diagonally on a well chosen basis
of V

hi |λi = λi |λi . (2.4)

This notation is a short-cut for Γ(hi )|ψλ i = λ(hi )|ψλ i , in a given representation ΓV of g where
hi is represented by the linear operator Γ(hi ) acting on the vectors |ψλ i spanning V . The
eigenvalues λi can be collected into an r-dimensional vector which is called a weight. This
weight can be seen as an element of g⋆◦ , λ(hi ) = λi .
Moreover, the weight λi are integers. Indeed, the representation ΓV provides a representation
(i) (i)
of the sl2 (α(i) ) subalgebra spanned by the Chevalley generators {H α , E ±α } . The latter
representation of sl2 (α(i) ) can be decomposed into a direct sum of irreducible representations of
(i)
sl2 (α(i) ) discussed in Theorem 2.1. Thus the diagonal elements of Γ(H α ) must all be integers,
(i) (i)
which means that λ(H α ) is an integer. We have λi = λ(hi ) = λ(H α ) . From this, one
concludes that (λ, α∨ ) is an integer for any α ∈ ∆. One just repeats the previous argument for
2
any subalgebra sl2 (α) of g, α ∈ ∆ , and by noting (cf. Eq. (1.42)) that λ(H α ) = (α,α) λ(hα ) =
2(λ,α)
(α,α) = (λ, α∨ ) . It is worth keeping in mind that, upon identification of the Cartan subalgebra
(i)
2
g◦ with its dual g⋆◦ , the Chevalley generator hi := H α = |α(i) |2
hα(i) corresponds to α(i)∨ .

Example 2.2. The adjoint representation of g

Let the basis elements {ai }di=1 of g be chosen to be hα(1) , ... , hα(r) , together with eα for each α ∈ ∆. Then,
by Eq. (1.4) with a = h, it follows that ad(h) is diagonal for each h ∈ g◦ , the diagonal element corresponding
to hα(i) being 0 and the diagonal elements corresponding to eα being α(h). Thus each non-zero root α of g is
a weight of the adjoint representation of g and, as dimgα = 1, each such weight is simple ( i.e. of multiplicity
one). The only other weight λ of the adjoint representation is such that λ(h) = 0 for all h ∈ g◦ , and this weight
has multiplicity r .

Theorem 2.3. If λ is a weight of a representation, then λ + α is also a weight of the same

representation for each α ∈ ∆ such that E α |λi 6≡ 0.

Proposition 2.4. Every weight λ can be written in terms of the simple roots α(i) , 1 6 i 6 r:
r
X
λ= µj α(j) , (2.5)
j=1

where the coefficients µi are all real and rational. Consequently λ(h) is real for each
h ∈ g◦ (R).

Theorem 2.5. For each root α ∈ ∆ and each weight λ of a representation Γ of g, define the
linear functional ωα λ on g◦ by

2(λ, α)
ωα λ (h) = λ (h) − α(h) , ∀ h ∈ g◦ . (2.6)
(α, α)

Then ωα λ is also a weight of the representation Γ with the same multiplicity as λ, i.e.

mult(ωα λ) = mult(λ) .

Proof :
Semi-simple Lie algebras and representations (Parts I and II) 157

By virtue of the equality H α = 2/(α, α)hα , one has λ(H α ) = 2(λ, α)/(α, α), so
ωα λ(h) = λ(h) − λ(H α )α(h) .
Let

α if λ(H α ) ≥ 0 ,
β=
−α if λ(H α ) < 0 .

Then λ(H β ) ≥ 0 and ωβ λ(h) = ωα λ(h) for all h ∈ g◦ . Let R = λ(H β ), so that R is a non-negative
integer and ωβ λ(h) = λ(h) − Rβ(h). If R = 0 the result follows trivially, so it will be assumed
henceforth that R > 0. Applying Theorem 2.3 above R times, one sees that ωβ λ is a weight of
the representation with multiplicity m(λ) if eR −β |λip 6= 0 for each of the m(λ) linearly independent
eigenvectors |λip (p = 1, 2, ..., m(λ)) of h corresponding to the eigenvalue λ(h). Now consider the A1
subalgebra with basis H β , E β and E −β . By Equation (1.48), (E −β )R = ((−2/(α, α))1/2 )R eR −β , so
that eR −β )R |λi 6= 0. But the equation h|λi = λ(h)|λi implies that R = λ(H β ) is an
−β |λip 6= 0 iff (E p
β
eigenvalue of H of multiplicity m(λ) with linearly independent eigenvectors |λip (p = 1, 2, ..., m(λ)).
Therefore (E −β )R |λip is an eigenvector of H β with eigenvalue −R for p = 1, 2, ..., m(λ), so that
(E −β )R |λip 6= 0, from which the required result follows.

The following properties of wα are valid for any α ∈ ∆ :

(a) The element wα acts as a reflexion through the plane whose normal to α
~

wα α = −α ; (2.7)

(b) for any β ∈ g⋆◦ ,

wα wα β = β ;

(c) for any β and γ ∈ g⋆◦ ,

(wα β, wα γ) = (β, γ) ; (2.8)

(d) for any β and γ ∈ g⋆◦ and any two complex numbers µ and λ,

wα (λβ + µγ) = λwα β + µwα γ .

The interpretation of these transformations wα becomes clear if the orthonormal basis {Hi }ri=1
of g◦ is used. Each linear functional β on g◦ has associated with it an r-component vector corre-
sponding β ~ defined by β
~ := (β(H1 ), . . . , β(Hr )) . Attention will be concentrated solely on those
linear functionals β for which the β(Hi ) are all real. This includes all the roots and all the
weights of g.
−−→
Denoting the r-component vector corresponding to wα β by wα β, Equations (1.55) and (2.6)
give

−−→ ~ 2(β ~·α

~)
wα β = β − α
~. (2.9)
α·α
(~ ~)
p
Let θ be the angle between the vectors α ~ so that β
~ and β, ~ ·~ ~ α| cos θ, where |~
α = |β|·|~ α| = (~ α·α ~ ).
~ −−→ ~ ~ ~ ~
Then, if 1α = α α| is the unit vector along α
~ /|~ ~ , wα β = β − 1α (2|β| cos θ) is the reflection of β in
the hyperplane (through the origin) perpendicular to α ~ . As the operators wα were introduced
by Weyl, they are known as “Weyl reflections”.
158 Boulanger, de Buyl

With the identity operator E defined by Eα = α (for any linear functional α on g◦ ) and with
products such as wα wβ defined by wα wβ γ = wα (wβ γ) (for any linear functional γ on g◦ ), the set
consisting of the identity operator, the Weyl reflections and all products of the Weyl reflections,
forms a group called the “Weyl group” denoted by W.
The above theorem states that any Weyl reflection of any weight is a weight with the same
multiplicity. Repeated application gives the result that ωλ is also a weight with the same
multiplicity as λ for every element ω of the Weyl group W .
As every element of W maps the set of non-zero roots of g on itself (the roots are the non-zero
weights of the adjoint representation), W must be a subset of the set of all permutations of the
non-zero roots, the symmetric group Sd−r . The latter group is finite, so is the Weyl group of g .
Note that the composition of reflections leads again to reflections as well as to rotations. Thus
a generic element of W is not of the form (2.9).
It can be shown that every element of the Weyl group W can be expressed as a product of
reflexions w(i) := wα(i) associated with the simple roots. The construction of W is helped by
the observations that if σ and τ ∈ W, then σ = τ iff σα(i) = τ α(i) for every simple root α(i) ,
and that (by (1.20) and (2.6))

w(i) α(j) = α(j) − Aij α(i) , i, j = 1, . . . , r (no summation on j) . (2.10)

Definition 2.1. The α-string of weights containing λ

Suppose that α is a root of g and λ is a weight of some representation of g. Then the “α-string
of weights containing λ” is the set of all weights of that representation of the form λ + kα, where
k is an integer.

Proposition 2.6. Let α be a non-zero root of g and λ a weight of some representation Γ of

g. Then there exist two non-negative integers p and q (which depend on α and λ) such that
λ + kα is in the α-string containing λ for every integer k that satisfies the relations −p 6 k 6 q.
Moreover, p and q are such that

2(λ, α)
p−q = = (λ, α∨ ) . (2.11)
(α, α)

Proof :
With the notation of the proof of the previous Theorem 2.5, it is clear that, for k > 0, λ + kβ is a
weight if and only if (E β )k |λi 6= 0, for some eigenvector |λi corresponding to the eigenvalue λ(h) of
h (in the given representation). However, this is so if and only if λ(H β ) + kβ(H β ) is an eigenvalue
of H β in the reduction of the representation of g on the A1 sublgebra with basis H β , E β and E −β .
As Equation (1.42) implies that β(H β ) = 2, this condition can be restated to be that 12 λ(H β ) + k
be an eigenvalue of 21 H β in this reduction. The same conclusion follows if k < 0. Suppose that in
this reduction the largest irreducible representation of the A1 subalgebra has dimension 2j + 1, so
the eigenvalues of 21 H β range from −j to j in integral steps. Thus k takes all integral values from
−p to q, where
1
λ(H β ) + q = j,
2
1
λ(H β ) − p = −j ,
2
from which Equation (2.11) follows immediately.
Semi-simple Lie algebras and representations (Parts I and II) 159

2.3 Fundamental weights

The dual basis of the coroot basis is denoted {Λ(i) }ri=1 and its elements obey

(Λ(i) , α(j)∨ ) = δij . (2.12)

The Λ(i) are called fundamental weights.

One defines the quadratic form matrix G = (Gij ) with lower indices by Gij := (Λ(i) , Λ(j) ).
(j) (j)
In terms of the Cartan matrix Aij , (G)ij = (A−1 )ij (α 2,α ) . It inverse G−1 is written Gij in
Pr
components ( k=1 Gij Gjk = δik ) and is called the symmetrized Cartan matrix. In terms of the
Cartan matrix, Gij = (α(i)∨ , α(j)∨ ) = (α(i)2,α(i) ) Aij . The symmetrized Cartan matrix coincides
with the restriction of the Killing form to the Cartan subalgebra g◦ .

In order to describe the inner products on the root and weight spaces explicitly, one expresses
roots and weights through their components with respect to the basis of simple coroots and the
Dynkin basis B ⋆ := {Λ(i) | i = 1, 2, . . . , r}, respectively. Thus one writes
r
X r
X
λ= λi α(i)∨ = λj Λ(j) with λi = (λ, Λ(i) ) , λi = (λ, α(i)∨ ) . (2.13)
i=1 j=1

The coefficients λi are called Dynkin labels. As a consequence of (Λ(i) , α(j)∨ ) = δij one has

j (α(j) , α(j) ) j
(α(j)∨ )i = δij = (Λ(i) ) , (α(j) )i = δi . (2.14)
2
The quadratic form matrix and the symmetrized Cartan matrix serve to lower and raise indices,
respectively:
r
X r
X
λi = Gij λj , λi = Gij λj . (2.15)
j=1 j=1

The matrix Gij provides an inner product on the weight space; for simple Lie algebras, this
product is in fact euclidean, i.e. its signature is (r, 0). Combining (2.14) and the expression of
Gij in terms of the Cartan matrix, it follows that
r
X
j ij
(α(i) ) ≡ (α(i) )k Gkj = (A) . (2.16)
k=1

In words: the components of the simple roots in the Dynkin basis coincide with the rows of the
Cartan matrix.

The Dynkin labels are the eigenvalues of the Chevalley generators ofPthe Cartan subal-
gebra in the basis were Eq. (2.4) holds. Indeed hi |λi = λ(hi )|λi = j i
j λ Λ(j) (h )|λi =
P j 2 i i
j λ |α(i) |2 Λ(j) (hα(i) )|λi = λ |λi . As we said before Example 2.2, the eigenvalues λ of h
are integers. They can be computed by using Proposition 2.6.

A weight of special importance is the Weyl vector,

X 1 X
ρ= Λ(i) = α.
i
2
α∈∆+
160 Boulanger, de Buyl

2.4 Weights and their multiplicities

Any finite-dimensional irreducible representation has a unique highest weight state |Λi which
is completely determined by its Dynkin labels. The importance of the highest weight of a
representation stems from the fact that each irreducible representation is uniquely and completely
specified by its highest weight, all of its properties, such as its dimension and the other weights
being easily deducible from it. Moreover, there is a straightforward procedure for constructing
every possible highest weight. The highest weight state is such that E α |Λi = 0, ∀α > 0. The
existence of such a highest weight state is obvious (for finiteness of the representation) but the
uniqueness is not trivial (we will not show this point here)

Definition 2.2. Highest weight Λ of a representation

If Λ is a weight of a representation of g, such that Λ > λ for every other weight λ, then Λ is
said to be the “highest” weight of the representation.

Theorem 2.7. If Λ is the highest weight of an irreducible representation of a semi-simple

complex Lie algebra g then

(a) Λ is a simple weight (i.e. mult(Λ)= 1); and

(b) every other weight λ of the representation has the form

r
X
λ=Λ− qi α(i) , where qi ∈ Z+ , 1 6 i 6 r. (2.17)
i=1

The following theorem is fundamental.

Theorem 2.8. For every irreducible representation Γ of a semi-simple complex Lie algebra g
the highest weight Λ can be written as
r
X
Λ= ni Λ(i) , ni ∈ Z+ , 16i6r (2.18)
i=1

where {Λ(i) }ri=1 is the set of fundamental weights of g. Moreover, to every set of non-negative
integers {ni }ri=1 there exists an irreducible representation of g with highest weight Λ given by
Equation (2.18), and this representation is unique up to equivalence.

Proof:
Only the proof of the first part of the theorem will be given here. For a proof of the second part,
see e.g. [1].
Suppose that Λ is the highest weight of an irreducible representation of g. Then, by Theorem 2.5,
for every α ∈ ∆, ωα Λ is also a weight, so that Λ > ωα Λ. But Λ − ωα Λ = [2(Λ, α)/(α, α)]α which
implies that the right-hand side must be non-negative for every α ∈ ∆. In particular this implies
that [2(Λ, α(i) )/(α(i) , α(i) )] > 0 for every simple root α(i) . We know (see e.g. Proposition 2.6) that
2(Λ, α(i) )/(α(i) , α(i) ) = (Λ, α(i)∨ ) is an integer, then, for each i = 1, 2, . . . , r,

2(Λ, α(i) )/(α(i) , α(i) ) = ni ,

where ni is a non-negative integer.
P r (k) (h). As ni = 2(Λ, α(i) )/(α(i) , α(i) ) = rP ki
Now let Λ(h) = k=1 µk α k=1 µk (A) , one has
Pr i −1
Pr Pr i −1 (k)
Pr i
µl = i=1 n (A )il , so Λ(h) = k=1 i=1 n (A )ik α (h) = i=1 n Λ(i) (h), with Λ(i) :=
Pr −1 ) α(k) the fundamental weights.
k=1 (A ik
Semi-simple Lie algebras and representations (Parts I and II) 161

The irreducible representation of g with highest weight Λ specified by Equation (2.18) will
be denoted by Γ{n1 ,n2 ,...,nr } . Its dimension d is given by the following “Weyl’s dimensionality
formula”
Y (Λ + ρ, α) r
X
d= , ρ= Λ(i) . (2.19)
(ρ, α) i
α∈∆+

For any simple complex Lie algebra g, Weyl’s formula (2.19) can be rewritten as

Y h X
r r
X  i
d= ni kiα wi / kjα wj + 1 , (2.20)
α∈∆+ i=1 j=1

Pr α (i)
Pr i
where α = i=1 ki α , Λ = i=1 n Λ(i) , and wi is the “weight” of the Dynkin diagram
corresponding to α(i) , cf. subsection 1.4.

Although less concise, Equation (2.20) is much easier to apply than Equation (2.19).
Pr
Exercise 2.9. Deduce (2.20) from (2.19). Hint : (i) compute (Λ, α) and find ( i=1 ni kiα wi )/(2w). (ii)
Pr α 1
compute (ρ, α) and find ( i=1 ki wi )/(2w) using (ρ, α(i) ) = 2
(α(i) , α(i) ) ∀i ∈ {1, 2, . . . , r} (which you may
show as well).

The set of eigenvalues of all the states in VΛ is called the weight system and denoted ΩΛ . All
the states in the representation space VΛ are obtained by the action of the lowering operators of
g as
E −β E −γ · · · E −η |Λi , where β, γ, . . . , η are simple roots .
P
Then, as stated in Theorem
P 2.7, each weight λ ∈ ΩΛ is necessarily of the form Λ − ni α(i) with
ni ∈ Z+ . The sum ni is called the level of λ and an analysis level by level can be done. Let
(i)
us focus on the action of a specific simple root α(i) . As E α |Λi = 0 and hi |Λi = Λi |Λi, the
(i)
representation theory of sl(2) tells us that E −α |Λi is non vanishing iff Λi > 1, more precisely
(i)
one can act Λi times with E −α on |Λi. With this criterion, the systematic construction of all
weights in the representation can be done by means of the following algorithm. We start with
the highest weight Λ = (Λ1 , ..., Λr ) in the Dynkin basis. For each positive Dynkin label Λi > 0
we construct the sequence of weights Λ − α(i) , ..., Λ − Λi α(i) which belong to ΩΛ . The process
is repeated with Λ replaced by each of the weights just obtained.

The Freudenthal recursion formula gives the multiplicity of the weight λ in the representation
Λ
∞
XX
[(Λ + ρ, Λ + ρ) − (λ + ρ, λ + ρ)]multΛ (λ) = 2 (λ + kα, α)multΛ (λ + kα) (2.21)
α>0 k=1

Example 2.10. Some representations of sl(3)

The A2 ∼ = sl(3) algebra will be discuss in details in the next section. We give here the roots of this algebra in
order to illustrate the construction explained above; these are α1 , α2 , α1 + α2 and the associated negative roots.
We know that to every dominant integral weight corresponds a representation. Let us begin by illustrate the
simplest case, namely Λ = Λ(1) which is denoted
162 Boulanger, de Buyl

1 0

According to the procedure, we must act once on |Λi with E −α1 . As α1 = 2Λ(1) − Λ(2) , the state obtained has
weight,

-1 1

We must now act on the state E −α1 |Λi once with E −α2 , which gives a state with weight

0 -1

and the procedure stops since no component is non-negative. We can easily obtain the weight system for the
representation associated with every dominant integral weight in this way. We conclude with the weight systems
for Λ = Λ(2) and for Λ = 2Λ(1) :

0 1 2 0
? ?
1 -1 0 1
? @
R
-1 0 -2 2 1 -1
R
@
-1 0
?
0 -2

3 Applications to Hadrons : A1 and A2

This section is based almost verbatim on [2].
Elementary particles possess internal symmetries. These are related to quantum numbers
such as baryon number, lepton number, isotopic spin, hypercharge and charm, which distinguish
the different elementary — and composite — particles. Here we will be concerned with global
internal symmetries based on the groups SU (2) and SU (3).
The starting point of all symmetry schemes for the elementary particles is the observation
that there appear to be four fundamental interactions between these particles. These are, in
decreasing order of strength :

(i) the strong interaction, first discussed in the context of the binding of nucleons in the
nucleus;

(ii) the electromagnetic interaction;

(iii) the weak interaction, responsible for beta decay;

(iv) the gravitational interaction.

In terms of these four interactions it is possible to divide the observed particles into two major
categories, the leptons (and antileptons) which never experience strong interaction, and the
hadrons which, at least in some circumstances, interact through the strong interaction. In
addition there are intermediate particles that are the carriers of the interactions: the photon γ,
the weak gauge bosons W ± and Z 0 , the gluons and the hypothetical graviton. The category
of hadrons can be further divided into two classes, those whose intrinsic spin s is an integer,
Semi-simple Lie algebras and representations (Parts I and II) 163

being called mesons and the others of half-integer spin being referred to as baryons. The “lepton
number” and “baryon number” may be defined for all the presently observed particles by


1 if the particle is a lepton,
L = −1 if the particle is an antilepton, (3.1)


0 for any other type of particle,

and


1 if the particle is a baryon,
B = −1 if the particle is an antibaryon, (3.2)


0 for any other type of particle.

For many years it has been accepted that lepton number and baryon number are each additively
conserved, but developments in grand unified theories together with the observation that in the
Universe more matter exists compared to antimatter have changed this point of view.

3.1 The global symmetry algebra su(2) and isotopic spin

The object of this subsection is to introduce the concept of isotopic spin and present the basic
ideas in such a way that they are easily generalizable to other internal symmetries.
Based on experimental evidences, the proton and the neutron were proposed to be two states
of the same particle, the nucleon. The latter is assigned an isotopic spin I with value 1/2 . This
value is chosen so that the multiplicity of the I = 1/2 irrep. of SU (2) is 2I + 1 = 2 , the number
of states corresponding to the nucleon: proton and neutron. Furthermore, it is suggested that in
the absence of electromagnetic interactions (in a universe with no electromagnetic interactions)
the proton and the neutron would be identical, and each of their interactions, which are all
“strong”, would also be identical.
Developing this further, one can introduce three self-adjoint linear operators I1 , I2 and I3
that satisfy the commutation relations

[Ii , Ij ] = iεijk Ik . (3.3)

As the Lie group SU (2) is compact, each irrep can be taken to be unitary and the corre-
sponding representations of the real Lie algebra su(2) consist of anti-Hermitian matrices. Thus
if {ψ1 , ψ2 , . . . ψd } is an ortho-normal basis of such a d-dimensional irrep Γd of su(2) in an inner
product space V and if the linear operators Φ(a) are defined for all a of su(2) by
d
X
Φ(a)ψp = Γ(a)qp ψq , p = 1, 2, . . . d , (3.4)
q=1

∗
then, as (ψq , Φ(a)ψp ) = Γ(a)qp = −Γ(a)pq = −(Φ(a)ψq , ψp ), it follows that

(φ, Φ(a)ψ) = −(Φ(a)φ, ψ) ∀φ, ψ ∈ V, ∀a ∈ su(2) . (3.5)

It is therefore convenient to define 3 linear operators A1 , A2 , A3 by

Ap = −iΦ(ap ) , p = 1, 2, 3 (3.6)

so that, by Eq. (3.5) ,

(φ, Ap ψ) = (Ap φ, ψ) ∀φ, ψ ∈ V and p = 1, 2, 3. (3.7)

164 Boulanger, de Buyl

That is to say, A1 , A2 and A3 are self-adjoint operators. Also, because Γd is a representation,

implying [Φ(ap ), Φ(aq )] = Φ([ap , aq ]) , we have
[Ap , Aq ] = iεpqr Ar , p, q = 1, 2, 3. (3.8)
This follows from the commutation relations between the generators ap of su(2) , see Example
1.1.

The real Lie algebra su(2) is indeed the set of all 2 × 2 traceless anti-hermitian matrices. This
example also shows that the matrices of a real Lie algebra need not themselves be real . The
reality condition of a real Lie algebra g(R) requires only that the elements of g(R) be real linear
combinations of a1 , a2 , . . . , an .
Now one defines the ladder operators A+ and A− by
A+ = A1 + iA2 , A− = A1 − iA2 , (3.9)
so that A1 = 21 (A+ + A− ), A2 = − 12 i(A+ − A− ) . Then, by Equation (3.8),
[A3 , A+ ] = A+ , (3.10)
[A3 , A− ] = −A− , (3.11)
[A+ , A− ] = 2A3 . (3.12)
Moreover, Equation (3.7) implies
(A− φ, ψ) = (φ, A+ ψ) ∀ φ, ψ ∈ V . (3.13)
Finally, on defines the operator A2 by
A2 = A21 + A22 + A23 . (3.14)
Then , by (3.8),
[A2 , Ap ] = 0 , p = 1, 2, 3 , (3.15)
which in turn implies that
[A2 , A+ ] = 0 = [A2 , A− ] . (3.16)
Also, as A− A+ = A21 + A21 + i[A1 , A2 ], equations (3.8) and (3.14) give
A− A+ = A2 − A23 − A3 , (3.17)
A− A+ = A2 − A23 − A3 . (3.18)
In the quantum mechanical theory of angular momentum, the operators Jx , Jy ,Jz corre-
sponding to the components about Ox, Oy and Oz satisfy the commutation relations
[Jx , Jy ] = i~Jz , [Jy , Jz ] = i~Jx , [Jz , Jx ] = i~Jy . (3.19)
The operators Jx , Jy and Jz are general angular momentum operators, in the sense that they
may correspond to intrinsic spin, orbital angular momentum, or a combination of both. Apart
from a factor ~, Equations (3.8) and (3.19) are identical. Therefore the identifications
Jx = ~A1 , Jy = ~A2 , Jz = ~A3 ,
J+ = ~A+ , J− = ~A− ,
J2 = ~ 2 A2 (3.20)
Semi-simple Lie algebras and representations (Parts I and II) 165

provide the connection between the representation theory of su(2) and the theory of angular
momentum.
Returning to g(R) = su(2) and its complexification g = sl2 , in any representation Γ of g the
P3
matrix representing A2 [defined by Γ(A2 ) = p=1 {Γ(Ap )}2 ] commutes [by Equ. (3.15)] with
Γ(a) for all a ∈ g. Schur’s Lemma then implies that if Γ is irreducible then Γ(A2 ) must be a
multiple of the unit matrix, so all the basis vectors ψ1 , ψ2 , . . . ψd of the irreducible representation
space V are eigenvectors of A2 with the same eigenvalue.
One summarizes the main conclusions concerning the irreducible representations of su(2)
in the form of a theorem. The results have immediate application in the theory of angular
momentum and in the theory of isotopic spin. Theorem 2.1 already contains those results, but
we give them here in a more “physical” way, also providing the explicit expressions for the
representation matrices.
Theorem 3.1. To every non-negative integer or half-integer j there exists an irreducible repre-
sentation of su(2) (and its complexification A1 ∼
= sl2 ) of dimension d = 2j + 1. The orthonormal
j
basis vectors of the irreducible representation specified by j may be denoted by ψm , where m
j
takes all values from j down to −j by unit steps. Each basis vector ψm may be chosen to be a
simultaneous eigenvector of A2 and A3 with eigenvalue j(j + 1) and m respectively, that is

A2 ψ m
j j
= j(j + 1)ψm , (3.21)
j j
A3 ψm = mψm , m ∈ {−j, −j + 1, . . . , j − 1, j} . (3.22)

Moreover, the relative phases of the basis vectors may be chosen so that
j
p j
A+ ψm = (j − m)(j + m + 1) ψm+1 for m ∈ {−j, −j + 1, . . . , j} , (3.23)
j
p j
A− ψm = (j + m)(j − m + 1) ψm−1 for m ∈ {−j, −j + 1, . . . , j} . (3.24)

Up to equivalence, these are the only irreducible representations of su(2) (and its complexification
sl2 ).
The matrices of the irreducible representation may be denoted by Dj (a), with elements
D (a)m′ m where m′ and m take the values −j, −j + 1, . . . , j − 1, j (these all being integers
j

if j is an integer and all being half-integers if j is a half-integer). Then

1 p p
Dj (a1 )m′ m = i[δm′ ,m+1 (j − m)(j + m + 1) + δm′ ,m−1 (j + m)(j − m + 1)] ,
2
1 p p
Dj (a2 )m′ m = [δm′ ,m+1 (j − m)(j + m + 1) − δm′ ,m−1 (j + m)(j − m + 1)] ,
2
Dj (a3 )m′ m = imδm′ m . (3.25)

For a proof, see e.g. [2] p. 442.

One can write, by analogy with Equation (3.6),

Ip = −iΦ(ap ) , p = 1, 2, 3 (3.26)

where a1 , a2 , a3 are the basis elements of su(2). The operators I1 , I2 and I3 may be regarded as
being operators corresponding to the measurement of the “components” of isotopic spin in three
mutually perpendicular directions in an “isotopic spin space”. Introducing the linear operator
I 2 by

I 2 = I12 + I22 + I32 . (3.27)

166 Boulanger, de Buyl

The operator I 2 , which does not belong to su(2) but to its enveloping algebra, has eigenvalues
of the form I(I + 1), where I takes one of the values 0, 12 , 1, . . .. This quantity I is then regarded
as the “isotopic spin”, and the possible values of its component in the third direction in isotopic
space associated with the operator I3 is given by the eigenvalue I3 of I3 , which assume any of
the (2I + 1) values −I, −I + 1, . . ., I − 1, I. The simultaneous eigenvector of I 2 and I3 with
eigenvalues I(I + 1) and I3 may be denoted (by analogy with Equations (3.21) and (3.22)) as
ψII3 , so that

I 2 ψII3 = I(I + 1)ψII3 ,

I3 ψII3 = mψII3 , I3 ∈ {−I, −I + 1, . . . , I − 1, I} . (3.28)
For convenience, the relative phases of the eigenvectors may be chosen as in Equations (3.23)
and (3.24), so that with the operators I+ and I− defined by I± = I1 ± iI2 , the phases are such
that
p
I+ ψII3 = (I − I3 )(I + I3 + 1) ψII3 for I3 ∈ {−I, −I + 1, . . . , I − 1} ,
p
I− ψII3 = (I + I3 )(I − I3 + 1) ψII3 for I3 ∈ {−I + 1, . . . , I − 1, I} .
Indeed, for any element a of the su(2) Lie algebra spanned by the basis elements a1 , a2 and a3
of Equation (3.26),
I
X
Φ(a)ψII3 = DI (a)I ′ I3 ψII3′ , (3.29)
3
I3′ =−I

where DI is the irreducible representation of su(2) introduced above. In terms of I+ and I−

the commutation relations of Equation (3.3) can be rewritten as in Equations (3.10), (3.11) and
(3.12) as
[I3 , I+ ] = I+ ,
[I3 , I− ] = −I− ,
[I+ , I− ] = 2I3 . (3.30)
It may also be assumed that all these isotopic spin operators commute with all the operators
corresponding to space-time transformations, so that the state vector of each hadron is the direct
product of a function of space-time and one of the vectors ψII3 . Each value of I3 corresponds to
a particle, the set of (2I + 1) particles associated with a particular value I being said to form
an “isotopic multiplet”. It is implied from Equation (3.29) that the vectors ψII3 form the basis
of the (2I + 1)-dimensional irreducible representation DI of su(2). In the case of the nucleons,
the proton is assigned the value I3 = 12 and the neutron the value I3 = − 21 of the I = 12 isotopic
“nucleon” multiplet.
It is assumed that all hadrons can be classified within this scheme. Historically, the earliest
particles to be incorporated in this scheme after the nucleons were the three pions π + , π 0 , π − ,
which were assigned to an isotopic I = 1 multiplet, the values of I3 being 1, 0 and −1 respectively.
For both the nucleons and the pions, the electric charge Qe of the hadron is given by
1
Q = I3 + B , (3.31)
2
where B, the baryon number introduced in the previous subsection, has value 1 for the nucleons
and 0 for the pions. Actually, Equation (3.31) holds only for all non-strange and un-charmed
hadrons, the generalizations for strange hadrons being given later in Equation (3.53).
Semi-simple Lie algebras and representations (Parts I and II) 167

3.2 Structure of A2 , the complexification of su3

Roots and Cartan subalgebra of sl2 ≡ A2

In his original paper on the SU (3) symmetry scheme for hadrons, Gell-mann set up a basis for
su3 and its complexification sl3 which has been widely used in the elementary particle literature
ever since. He defined the eight traceless hermitian matrices λ1 , . . ., λ8 by
     
0 1 0 0 −i 0 1 0 0
λ1 =  1 0 0  , λ2 =  i 0 0  , λ3 =  0 −1 0  ,
0 0 0 0 0 0 0 0 0
     
0 0 1 0 0 −i 0 0 0
λ4 =  0 0 0  , λ5 =  0 0 0  , λ6 =  0 0 1 ,
1 0 0 i 0 0 0 1 0
   
0 0 0 1 0 0
1
λ7 =  0 0 −i  , λ8 = √  0 1 0 . (3.32)
0 i 0 3 0 0 −2

These satisfy the commutation relations

8
X
[λp , λq ] = 2ifpqr λr (3.33)
r=1

where the fpqr are antisymmetric in all three indices, the non-zero values being listed in Table
4.1

pqr 123 147 156 246 257 345 367 √458 √678
fpqr 1 1/2 −1/2 1/2 1/2 1/2 −1/2 3/2 3/2

Table 4.1: Non-zero values of the completely antisymmetric constants fpqr of su3 as defined in
Eq. (3.33).

A convenient basis for the real Lie algebra su3 is then provided by the traceless anti -hermitian
matrices a1 , a2 , . . ., a8 defined by ap = iλp , p = 1, . . . , 8. As these are linearly independent
over the field of complex numbers, the complexification A2 ∼ = sl3 of su3 may be defined easily.
We take the same basis ap or λp p = 1, . . . , 8 but allow for linear combinations with complex
P8
coefficients. Direct calculation using [ap , aq ] = −2 r fpqr ar shows that

K(ap , ap ) = −12δpq , p, q = 1, . . . , 8 . (3.34)

For example, ada1 (a2 ) = [a1 , a2 ] = −2a3 implies that (ada1 )i 2 = 0 except for i = 3, where
(ada1 )3 2 = −2. The first column of the matrix (ada1 )i j is trivially zero because [a1 , a1 ] = 0 and
the entries of the other seven columns are zero except for one entry. We find

(ada1 )2 3 = 2, (ada1 )3 2 = −2, (ada1 )4 7 = 1, (ada1 )5 6 = −1,

(ada1 )6 5 = 1, (ada1 )7 4 = −1, (ada1 )8 i = 0 = (ada1 )0 i . (3.35)

Then, one finds tr(ada1 ada1 ) = −12.

168 Boulanger, de Buyl

A convenient choice of Cartan subalgebra is the subspace spanned by λ3 and λ8 , which

implies that the rank l is 2. Then, with h1 = λ3 and h2 = λ8 ,

[h1 , (λ1 + iλ2 )] = 2(λ1 + iλ2 ) , [h2 , (λ1 + iλ2 )] = 0 ,

√
[h1 , (λ6 + iλ7 )] = −(λ6 + iλ7 ) , [h2 , (λ6 + iλ7 )] = 3(λ6 + iλ7 ) ,
√
[h1 , (λ4 + iλ5 )] = (λ4 + iλ5 ) , [h2 , (λ4 + iλ5 )] = 3(λ4 + iλ5 ) ,
[h1 , (λ1 − iλ2 )] = −2(λ1 − iλ2 ) , [h2 , (λ1 − iλ2 )] = 0 ,
√
[h1 , (λ6 − iλ7 )] = (λ6 − iλ7 ) , [h2 , (λ6 − iλ7 )] = − 3(λ6 − iλ7 ) ,
√
[h1 , (λ4 − iλ5 )] = −(λ4 − iλ5 ) , [h2 , (λ4 − iλ5 )] = − 3(λ4 − iλ5 ) .

Thus there are six non-zero roots: α1 , α2 , α3 and −α1 , −α2 , −α3 with

α
~1 = (α1 (h1 ), α1 (h2 )) = (2, 0) ,
√
α
~2 = (α2 (h1 ), α2 (h2 )) = (−1, 3) ,
√
α
~3 = (α3 (h1 ), α3 (h2 )) = (1, 3) .

Clearly, gα1 , gα2 , gα3 , g−α1 , g−α2 , g−α3 are all one-dimensional, with basis elements (λ1 + iλ2 ),
(λ6 + iλ7 ), (λ4 + iλ5 ), (λ1 − iλ2 ), (λ6 − iλ7 ) and (λ4 − iλ5 ) respectively. It should be noted that
α
~3 = α ~1 + α~ 2.
It remains to calculate explicit expressions for hα1 , hα2 and hα3 . Suppose hα1 = κ1 h1 +κ2 h2 .
Then, for j = 1, 2, the Equation

K(hα , h) = α(h) ∀ h ∈ g◦ (3.36)

with h = hj gives κ1 K(h1 , hj ) +κ2 K(h2 , hj ) = α1 (hj ), a pair of simultaneous linear equa-
tions for κ1 and κ2 . As K(ap , aq ) = −12δpq , it follows that K(h1 , h1 ) = 12 = K(h2 , h2 ) and
K(h1 , h2 ) = 0. Thus κ1 = 61 , κ2 = 0, and so
 
1 0 0
1 1
hα1 = h1 = 0 −1 0  . (3.37)
6 6
0 0 0

Similarly,
 
√ 0 0 0
1 3 1
hα2 = − h1 + h2 =  0 1 0 . (3.38)
12 12 6
0 0 −1

Finally, as hα3 = hα1 +α2 = hα1 + hα2 ,

 
√ 1 0 0
1 3 1
hα3 = h1 + h2 = 0 0 0 . (3.39)
12 12 6
0 0 −1

It follows from (α, β) := K(hα , hβ ) = α(hβ ) = β(hα ) and (3.34) that

1 1
(α1 , α1 ) = , (α1 , α2 ) = − ,
3 6
1 1
(α2 , α2 ) = − , (α2 , α2 ) = , (3.40)
6 3
Semi-simple Lie algebras and representations (Parts I and II) 169

all of which are real and rational, and both (α1 , α1 ) and (α2 , α2 ) are also positive. Finally, using
the bilinearity of K(· , ·), one finds
1
(α3 , α3 ) =
3
and the off-diagonal part of the Cartan metric is given by
2(α1 , α2 )
A12 = = −1.
(α2 , α2 )

The complex Lie algebra A2 possesses three pairs of non-zero roots, namely7 {~ α1 , −~
α1 },
{~
α2 , −~
α2 } and {~
α3 , −~
α3 } where α
~3 = α
~1 + α
~ 2 . For the first pair, as (α1 , α1 ) = 1/3 [see (3.40)],
Equations (3.37) and (1.42) imply that
 
1 0 0
Hα1 =  0 −1 0  .
0 0 0
With Eα1 = κ(λ1 + iλ2 ), E−α1 = µ(λ1 − iλ2 ), the third commutation relation of Equations
(1.45), together with Equation (3.33) gives κµ = 1/4. A convenient choice is κ = µ = 1/2,
producing
   
0 1 0 0 0 0
Eα1 =  0 0 0  , E−α1 =  1 0 0 .
0 0 0 0 0 0
In the SU (3) symmetry scheme for hadrons, this algebra is usually called the “I-spin” subalgebra
and a very common notation is defined by
1 α1
I+ = E α1 , I− = E −α1 , I3 = H .
2
Similarly, for the pair {α2 , −α2 }, one has
     
0 0 0 0 0 0 0 0 0
Hα2 =  0 1 0  , Eα2 =  0 0 1 , E−α2 =  0 0 0 .
0 0 −1 0 0 0 0 1 0
This subalgebra is called the “U-spin” or “L-spin” subalgebra, the identification being
1 α2
U+ = E α2 , U− = E −α2 , U3 = H .
2
Finally, for the pair {a3 , −α3 }, one finds
     
1 0 0 0 0 1 0 0 0
Hα3 =  0 0 0  , Eα3 =  0 0 0 , E−α3 =  0 0 0 .
0 0 −1 0 0 0 1 0 0
This subalgebra is called the “V-spin” or “K-spin” subalgebra, the identification being
1
V− = E α3 , V+ = E −α3 , V3 = − H α3 .
2
Note that in the special case of A2 where (α3 , α3 ) = (α1 + α2 , α1 + α2 ) happens to be equal to
(α1 , α1 ) = (α2 , α2 ), we have Hα3 = Hα1 + Hα2 .
7 We will not always explicitly indicate that the roots are vectors.
170 Boulanger, de Buyl

Strings of roots As we have seen, the non-zero roots are ±α1 , ±α2 and ±(α1 + α2 ). Thus the
α1 -string containing α2 consists of α2 = α2 + 0 · α1 and α1 + α2 = α2 + 1 · α1 and the α2 -string
containing (α1 + α2 ) consists of α1 = (α1 + α2 ) − 1 · α2 and α1 + a2 = (α1 + α2 ) + 0 · α2 .

Basis vectors of A2 As noted above, gα1 , gα2 , g(α1 +α2 ) , g−α1 , g−α1 , and g−(α1 +α2 ) have basis
elements (λ1 + iλ2 ), (λ6 + iλ7 ), (λ4 + iλ5 ), (λ1 − iλ2 ), (λ6 − iλ7 ) and (λ4 − iλ5 ) respectively.
Thus let

eα1 = κα1 (λ1 + iλ2 ) , e−α1 = κ−α1 (λ1 − iλ2 ) ,

eα2 = κα2 (λ6 + iλ7 ) , e−α2 = κ−α2 (λ6 − iλ7 ) ,
e(α1 +α2 ) = κ(α1 +α2 ) (λ4 + iλ5 ) , e−(α1 +α2 ) = κ−(α1 +α2 ) (λ4 − iλ5 ) . (3.41)

Then, as K(λp , λq ) = 12δpq for p, q = 1, . . . , 8, Equation (1.29) is satisfied for α = α1 , α2 and

α1 + α2 if

24κα1 κ−α1 = −1 , 24κα2 κ−α2 = −1 , 24κ(α1 +α2 ) κ−(α1 +α2 ) = −1 . (3.42)

Equation [eα , eβ ] = Nα,β eα+β with α = α1 and β = α2 gives 2κα1 κα2 = Nα1 ,α2 κ(α1 +α2 ) while
with α = −α1 , β = −α2 it gives −2κ−α1 κ−α2 = N−α1 ,−α2 κ−(α1 +α2 ) so that, by Equation (3.42),
Equation (1.30) is satisfied if

24κ2α1 κ2α2 = κ2(α1 +α2 ) . (3.43)

This shows that the requirement in Equation (1.30) imposes a constraint that is additional to
those consequent on Equation (1.29). If the further condition in Equation (1.49) is applied,
κ−α = −κ∗α for α = α1 , α2 and α1√+ α2 , so that Equations (3.42) imply |κα1 |2 = |κα2 |2 =
|κα1 +α2 |2 = 1/24 . Thus, if κα = 1/ 24 exp iθα for α = α1 , α2 and α1 + α2 , Equation (3.43)
reduces to exp 2i(θα1 + θα2 ) = exp 2iθα1 +α2 , so that θα1 + θα2 − θ(α1 +α2 ) = pπ, where p is some
integer. A convenient choice is θα1 = θα2 = θ(α1 +α2 ) = 0 (corresponding to p = 0), giving
   
0 1 0 0 0 0
1  1
eα1 = √ 0 0 0  , e−α1 = − √  1 0 0 
6 0 0 0 6 0 0 0
   
0 0 0 0 0 0
1  1
eα2 = √ 0 0 1  , e−α2 = − √  0 0 0 
6 0 0 0 6 0 1 0
   
0 0 1 0 0 0
1  1
eα1 +α2 = √ 0 0 0  , e−(α1 +α2 ) = − √  0 0 0  .
6 0 0 0 6 1 0 0

The matrices hα1 and hα2 complete the basis

   
1 0 0 0 0 0
1 1
hα1 = 0 −1 0  , hα2 =  0 1 0 . (3.44)
6 6
0 0 0 0 0 −1

The structure constants Nα,β The complex algebra A2 has non-zero roots ±α1 , ±α2 and
±(α1 + α2 ), so the only non-zero structure constants Nα,β are

Nα1 ,α2 , Nα2 ,α1 , N−α1 ,−α2 , N−α2 ,−α1 , Nα1 ,−(α1 +α2 ) , N−(α1 +α2 ),α1 , (3.45)
N−α1 ,(α1 +α2 ) , Nα1 +α2 ,−α1 , Nα2 ,−(α1 +α2 ) , N−(α1 +α2 ),α2 , N−α2 ,(α1 +α2 ) , N(α1 +α2 ),−α2 .
Semi-simple Lie algebras and representations (Parts I and II) 171

Of these, only Nα1 ,α2 , N−(α1 +α2 ),α1 and Nα2 ,−(α1 +α2 ) are unrelated by Nα,β = −Nβ,α and by
Equation (1.30). However, from Equation (1.33) with α = α1 , β = α2 and γ = −(α1 + α2 ),

Nα1 ,α2 = Nα2 ,−(α1 +α2 ) = N−(α1 +α2 ),α1 ,

so the sign of the whole set of constants Nα,β depends only on the sign of Nα1 ,α2 . [That
is, √they depend on the choice of the√sign of eα1 +α2 relative to eα1 and eα2 .] Indeed, Nα1 ,α2 =
1/ 6 exp i(θα1 + θα2 − θα1 +α2 ) = 1/ 6 exp ipπ. Thus, with the choice p = 0 that corresponds √ to
the matrices displayed explicitly already before, Nα1 ,α2 = Nα2 ,−(α1 +α2 ) = N−(α1 +α2 ),α1 = 1/ 6 .

Ortho-normal basis of the Cartan subalgebra g◦ of A2 Applying Gram-Schmidt orthog-

onalization process to φi := hαi (i = 1, 2) with (φi , φj ) = K(hαi , hαj ) = (αi , αj ), Equations
(3.40) give
√
H1 = 3 hα1 ,
(3.46)
H2 = hα1 + 2hα2 .

Then, using αi (Hj ) = K(hαi , Hj ) and Equations (3.40),

1
√
α~1 = 6 (2 √3, 0) ,
1 (3.47)
α~2 = 6 (− 3, 3)

and, using the explicit matrix representation given above,

   
√ 1 0 0 √ 1 0 0 √
3 3 1 3
H1 = 0 −1 0  (= λ3 ) , H2 =  0 1 0  (= λ8 ) .
6 6 6 6
0 0 0 0 0 −2

The simple roots

Positive roots of A2 As shown in subsection 3.2, the non-zero roots of A2 are ±α1 , ±α2 and
±(α1 + α2 ). With respect to β1 = α1 , β2 = α2 , the positive roots are α1 , α2 and (α1 + α2 ).
However, with respect to β1 = α1 , β2 = α1 + α2 , the positive roots are α1 , −α2 (= β1 − β2 ) and
α1 + α2 .

Simple roots of A2 With β1 = α1 , β2 = α2 , it is obvious that α1 and α2 are the only simple
roots. However, with β1 = α1 , β2 = α1 + α2 , the simple roots are −α2 and α1 + α2 . This
follows because with this basis the set of positive roots is α1 , −α2 (= β1 − β2 ) and α1 + α2 , but
α1 = (α1 + α2 ) + (−α2 ), so α1 cannot be simple, whereas −α2 and α1 + α2 are simple as they
cannot be expressed as the sum of two other positive roots.
In the following we will choose the basis where α(1) = α1 , α(2) = α2 (for A2 ).

The Cartan matrices of A2 For A2 , Equations (3.40) and (1.20) give

 
2 −1
A= .
−1 2

As this example shows, it is elementary to construct the Cartan matrix A of g from the
knowledge of the root system of g . What is very remarkable is that the process can be reversed,
as we explained in subsection 1.5. Note that the Cartan matrix of A1 is simply A = 2 .
172 Boulanger, de Buyl

Orthonormal basis for the compact real form gc of A2 . Using the results of subsection
3.2 together with the Gell-Mann matrices {λi }8i=1 , one finds
√ √
i 3 i 3
iH1 = λ3 , iH2 = λ8 ,
6
√ 6 √
1 i 3 i i 3
√ (eα1 + e−α1 ) = λ2 , √ (eα1 − e−α1 ) = λ1 ,
2 6 2 6
√ √
1 i 3 i i 3
√ (eα2 + e−α2 ) = λ7 , √ (eα2 − e−α2 ) = λ6 ,
2 6 2 6
√ √
1 i 3 i i 3
√ (eα1 +α2 + e−(α1 +α2 ) ) = λ5 , √ (eα1 +α2 − e−(α1 +α2 ) ) = λ4 .
2 6 2 6
As the basis of g = su3 chosen in subsection 3.2 was aj = iλj , j = 1, . . . , 8, it follows that
the compact real form gc of A2 constructed
√ according to Theorem 1.18 is precisely the real Lie
algebra su3 with orthonormal basis {( 3/6) aj }8j=1 . Consequently,

6
K(ap , aq ) = −( √ )2 δpq = −12δpq
3
as noted previously. Moreover, Proposition 1.16 asserts that with this basis the structure con-
stants f rpq are completely antisymmetric, which is indeed the case, see Table 4.1.

The Weyl group of g = A2

Using the Cartan matrix of A2 and Equation (2.10), we find

w(1) α(1) = −α(1) w(2) α(1) = α(1) + α(2)

w(1) α(2) = α(1) + α(2) w(2) α(2) = −α(2) .

It then follows that

w(2) w(1) α(1) = −(α(1) + α(2) ) w(1) w(2) α(1) = α(2)
w(2) w(1) α(2) = α(1) w(1) w(2) α(2) = −(α(2) + α(1) )

and further that

w(1) w(2) w(1) α(1) = w(2) w(1) w(2) α(1) = −α(2)
(3.48)
w(1) w(2) w(1) α(2) = w(2) w(1) w(2) α(2) = −α(1) .
2 2
As every element of W is a string of products of w(1) and w(2) , and w(1) = w(2) = E, the fact that
w(1) w(2) w(1) = w(2) w(1) w(2) implies that there are no further distinct elements in W . Thus W is
the non-Abelian group of order 6 with elements {E, w(1) , w(2) , w(1) w(2) , w(2) w(1) , w(1) w(2) w(1) } ,
isomorphic to the symmetric group S3 = {e, (1, 2), (2, 3), (3, 2, 1), (1, 2, 3), (3, 1)} .
Incidentally, Equation (2.6) shows that wα(1) +α(2) = w(1) w(2) w(1) . It is instructive to display
these results using
√ the orthonormal basis {Hi }ri=1 of g◦ discussed √above. As noted in Eq.
√
(1) (2)
3.47, α
~ = (1/ 3, 0) , α ~ = (−1/2 3, 1/2) , so α~ (1) + α
~ (2) = (1/2 3,
√ 1/2) , so the reflection
√
(1) (2) (1) (2)
lines normal to α ~ , α ~ and α~ +α ~ have directions (0, 1), 1/2( 3, 1) and 1/2( 3, −1),
~ does not lie on a reflection line, the six distinct elements {σa }6 of W acting
respectively. If β a=1
on β~ produce six distinct vectors {σa β} ~ 6 . However, if β ~ lies on a reflection line and β
~ 6= 0 ,
a=1
~ ~ ~
the set {σ β | σ ∈ W} contains only three members, while if β = 0 then {σ β | σ ∈ W} contains
only the one member ~0 .
Semi-simple Lie algebras and representations (Parts I and II) 173

6
• •
~
ω(1) β ~ = Eβ
β ~

H
H 
HH ~ (2)
α ~ (1)+ α
α ~ (2)

HHAK
 
•
~
HH A
 •
~

 -α~
ω(1) ω(2) β ω(2) β
AH (1)
-
 H
 HH
 HH
 HH
reflection line reflection line
normal to α ~ (2) normal to α~ (1) + α
~ (2)
 H
reflection line
normal to α ~ (1)

• •
~
ω(1) ω(2) ω(1) β ~
ω(2) ω(1) β

~ ∈ Φr .
FIG. 1. The action of the elements of the Weyl group of A2 on a general 2-component vector β

Irreducible representations of A2
We first remind the following important relation established previously in subsection 2.2: hi |λi =
(i) (i)
λi |λi or Γ(H α )|ψλ i = λ(H α )|ψλ i .

Weights of a three-dimensional representation of A2 Consider the explicit three-

dimensional representation of A2 constructed in subsection 3.2. This is the fundamental, or
defining representation of su3 with basis vectors |λ1 i = (1, 0, 0)T , |λ2 i = (0, 1, 0)T and |λ3 i =
(0, 0, 1)T . By inspection (looking at the 3 diagonal elements of the 3 × 3 matrices hα1 , hα2 and
hα3 ), one finds
1
λ1 (hα1 ) = , λ1 (hα2 ) = 0 ,
6
1 1
λ2 (hα1 ) = − , λ1 (hα2 ) = ,
6 6
1
λ3 (hα1 ) = 0 , λ3 (hα2 ) = − ,
6

so the 3 weights λ~1 , λ~2 and λ~3 are all simple. It is interesting to note that, as α(j) (hα(k) ) =
(α(j) , α(k) ), Equations (3.40) imply that for h = hα(1) and h = hα(2)

λ1 (h) = 32 α(1) (h) + 31 α(2) (h) ,

λ2 (h) = λ1 (h) − α(1) (h) , (3.49)
λ3 (h) = λ1 (h) − α(1) (h) − α(2) (h) ,

from which it follows that Equations (3.49) are true for all h ∈ g◦ .

The fundamental weights of A1 and A2 ; Weyl’s dimensionality formula We can invert

the Cartan matrix A = 2 of A1 to obtain A−1 = 1/2, so that the only fundamental weight of
A1 is Λ = 21 α, while for A2 we find
 
2/3 1/3
A−1 =
1/3 2/3
174 Boulanger, de Buyl

so that the fundamental weights are Λ(1) = 23 α(1) + 13 α(2) and Λ(2) = 31 α(1) + 23 α(2) .
Note that one can now identify the irreducible representation Γ{1,0} of A2 with the three-
dimensional representation of A2 studied above. The equations (3.49) show that the complete
weight system of Γ{1,0} is obtained by subtracting successively α(1) and α(2) from Λ(1) .

Since the fundamental weights of g = A2 are Λ(1) = 32 α(1) + 13 α(2) and Λ(2) = 31 α(1) + 23 α(2) ,
the corresponding two-component vectors are then Λ ~ (1) = 2 α (1) 1 (2) ~ (2) = 1 α (1) 2 (2)
3~ +3α~ and Λ 3~ +3α ~ .
It follows that [cf. equations√(3.47)], in the two-dimensional orthonormal basis where α ~ (1) =
√1 (1, 0) and α 1
~ (2) = 2√ (−1, 3), we have
3 3

√
~ (1) = 1 ( 3, 1) ,
Λ ~ (1) = 1 (0, 2) .
Λ
6 6
Weyl’s dimensionality formula (see also Equation (2.20)) implies that the dimension d of the
irreducible representation Γ{λ1 ,λ2 } is given by
1 1
d{λ1 ,λ2 } = (λ + 1)(λ2 + 1)(λ1 + λ2 + 2) . (3.50)
2
It can be shown that for A2 , the knowledge of the weights of Γ{λ1 ,λ2 } yields those of Γ{λ2 ,λ1 } ,
because the weights of Γ{λ2 ,λ1 } are the negatives of those of Γ{λ1 ,λ2 } . Also, if µ1 α(1) + µ2 α(2) is
a weight of Γ{λ1 ,λ2 } , then µ1 α(2) + µ2 α(1) is a weight of Γ{λ2 ,λ1 } with the same multiplicity.

Some irreducible representations of A2 In the elementary particle literature it is common

to label each irrep by its dimension d. When λ1 6= λ2 the irreps Γ{λ1 ,λ2 } and Γ{λ2 ,λ1 } have
the same dimension, one being the “complex conjugate” of the other. In this case one irrep is
denoted by {d} and the other by {d∗ }, the usual convention being

{d} if λ1 > λ2 ,
Γ{λ1 ,λ2 } = ∗ (3.51)
{d } if λ1 < λ2 .

As shown in subsection 3.2 the Weyl group W of A2 consists of six elements {σa }6a=1 ≡
{E, w(1) , w(2) , w(1) w(2) , w(2) w(1) , w(1) w(2) w(1) } . As σa λ is a weight with the same multiplic-
ity as λ for each σa ∈ W , the weights of an irrep may be arranged in sets of six, three or one,
the first occuring when ~λ is not on a reflection line, the second when ~λ is on a reflection line and
~λ 6= 0, and the third when ~λ = 0 .
Some physically important sets of weights of low-dimensional irreps of A2 will now be exam-
ined.

(a) Γ{0,0} = {1}

With λ1 = 0 = λ2 , Equation (3.50) gives d = 1. Consequently this irrep has only one
weight, namely the highest weight Λ = 0 .
(b) Γ{1,0} = {3}
With λ1 = 1, λ2 = 0, Equation (3.50) gives d = 3. The highest weight is Λ = Λ(1) ,
√
~ = Λ~(1) = 1 ( 3, 1), which lies on a reflection line. Consequently this irrep has
so Λ 6
two other simple weights, ω(1) Λ ~ and ω(2) ω(1) Λ,
~ for which , by inspection of Figure 1,
√
~ = 1 (− 3, 1) and ω(2) ω(1) Λ
ω(1) Λ ~ = 1 (0, −2). This implies that ω(1) Λ ~ = Λ−α ~ (1) and
6 6
~ =Λ
ω(2) ω(1) Λ ~ − (~ (1)
α +α (2) 2
~ ) , so the weights of this irrep are 3 α (1)
~ + 3α1 (2) 1
~ , −3α~ + 13 α
(1)
~ (2)
and − 31 α~ (1) − 23 α
~ (2) . The weight diagram is given in Figure 2a.
Semi-simple Lie algebras and representations (Parts I and II) 175

(c) Γ{0,1} = {3∗ }

~ = Λ
The weights of {3∗ } are the negatives of those of {3}, that is, they are Λ ~ (2) =
1 (1) 2 (2) 1 (1) 1 (2) 2 (1) 1 (2)
3α
~ + 3α ~ , 3α ~ − 3α ~ and − 3 α
~ − 3α ~ . The weight diagram is given in Figure
2b.

(d) Γ{1,1} = {8}

The dimension is 8. The highest weight is Λ = Λ(1) + Λ(2) = α(1) + α(2) , so Λ ~ =α ~ (1) + α
~ (2)
1
√
= 6 ( 3, 3). As this does not lie on a reflection line, there are five other simple weights
obtained from√it by Weyl reflections,√ which may be √ found by inspection using Figure
√ 1.
They are 61 (− 3, 3) = α ~ (2) , 61 ( 3, −3) = −~
α(2) , 61 (− 3, −3) = −~ ~ (1) , 61 (2 3, 0) =
α(2) − α
√
~ (1) and 61 (−2 3, 0) = −~
α α(1) . As only six of the eight weights are thereby accounted for,
all that can remain lies at level 2 (the other weights were found at level 1,3 and 4) and is
the weight ~0 of multiplicity 2. This representation is the adjoint representation. Its weight
diagram is an hexagon centered at the origin.

(e) Γ{2,0} = {6}

With λ1 = 2, λ2 = 0, √ Weyl’s dimensionality formula gives d = 6 . The highest weight is

~ = 1 (2 3, 2), which lies on a reflection line. Exactly as for Γ{1,0} = {3},
Λ = 2Λ(1) , so Λ 6 √
Weyl reflections then produce two more simple weights 61 (−2 3, 2) and 16 (0, −4). This
leaves three other weights to be determined. We will use Proposition 2.6. For the α(1) -
string containing Λ, Equation (2.11) gives

h4 2 i
p − q = 2(Λ, α(1) )/(α(1) , α(1) ) = 2 (α(1) , α(1) ) + (α(2) , α(1) ) /(α(1) , α(1) ) = 2 .
3 3

As Λ is the highest weight, it follows that there can be no weight of the form Λ + kα(1)
with k > 0, and since −p 6 k 6 q , we necessarily have q = 0, which gives p = 2 . We thus
have, in the α(1) -string containing Λ , the weights {Λ, Λ − α(1) , Λ − 2α(1) } . But Λ − 2α(1)
(1)
was already obtained previously by Weyl reflection of Λ . However,
1
√ Λ − α is 1new.√ Weyl
reflections applied to this weight produce two other weights: 6 (− 3, −1) and 6 ( 3, −1) .
We have thus d = 6 weights, so we have got all of them for this irrep.

(f) Γ{2,2} = {27}

The highest weight is Λ = 2Λ(1) +2Λ(2) = 2α(1) +2α(2) . By application of Weyl reflections,
we find the weights −Λ , 2α(1) , 2α(2) , −2α(1) and −2α(2) . Application of Freudenthal’s
recursion formula (Equation (2.21)) shows that the weights of level 1 are Λ − α(1) and
Λ − α(2) and that both are simple. They produce, upon Weyl reflection, the weights From
these and by application of Weyl reflections, we find the weights −α(2) −2α(1) , −α(1) −α(2) ,
α(2) − α(1) and −α(1) − 2α(2) .

Now consider the only possible weight of level 2 which has not already been found by Weyl
reflection of Λ and of the weights at level 1. Since we already obtained 2α(1) and 2α(2)
from Weyl reflection of Λ, we are left with only α(1) + α(2) at level 2. As the Weyl vector
ρ = 21 (α(1) + α(2) + (α(1) + α(2) )) = α(1) + α(2) , Freudenthal’s recursion formula Equation
176 Boulanger, de Buyl

(2.21) gives

h   i
3(α(1) + α(2) ), 3(α(1) + α(2) ) − 2(α(1) + α(2) ), 2(α(1) + α(2) ) ×
×mult(α(1) + α(2) ) =
h  
2 mult α(1) + α(2) + 1 · (α(1) + α(2) ) (2α(1) + 2α(2) , α(1) + α(2) ) +
 
mult α(1) + α(2) + 1 · α(2) (α(1) + 2α(2) , α(2) ) +
  i
mult α(1) + α(2) + 1 · α(1) (2α(1) + α(2) , α(1) ) .

Since (α(1) , α(1) ) = 31 = (α(2) , α(2) ), (α(1) , α(2) ) = − 16 and mult(2α(1) +2α(2) ) = mult(α(1) +
2α(2) ) = mult(2α(1) + α(2) ) = 1, this gives mult(α(1) + α(2) ) = 2: the level-2 weight
α(1) + α(2) has a multiplicity 2. Applying Weyl reflections to this double weight generates
2 × 6 = 12 further weights on top of the 6 + 3 + 3 already obtained from Λ, α(1) + 2α(2)
and 2α(1) + α(2) . This makes 24. Proceeding further, there is a weight at level 4 which
was not yet generated : the weight ~0 which happens to have multiplicity 3.

6
√ √
~ = 1 (− 3, 1)
ω(1) Λ ~ = 1 ( 3, 1)
Λ
6
• • 6
= −1α ~ (1)+ 1 α
~ (2) = 2α~ (1)+ 1 α
~ (2)
3 3 3 3

~ = 1 (0, −2)
ω(2) ω(1) Λ
6
= −1α ~ (1)− 2 α
~ (2)
• 3 3

FIG. 2a. Weight diagram of the irrep Γ{1,0} of A2 .

•6
~ = 1 (0, −2)
Λ
6
= 1α~ (1)+ 2 α
~ (2)
3 3

• √
~ = 1 (− 3, −1)
• √
~ = 1 ( 3, −1)
ω(2) Λ
ω(1) ω(2) Λ 6
6
(1) = 1α ~ (1)− 1 α
~ (2)
= − α2 ~ 1
− α~ (2) 3 3
3 3

FIG. 2b. Weight diagram of the irrep Γ{0,1} of A2 .

Semi-simple Lie algebras and representations (Parts I and II) 177

3.3 The global symmetry group SU (3) and strangeness

The concept of strangeness quantum number was developed out of the “associated production”
of Pais to explain the observation that certain hadrons are created by strong interactions, but
decay through the weak interaction. The proposal was that every hadron possesses a strangeness
quantum number S, which is assumed to be an integer, and that production or decay takes place
through the strong interaction if and only if the quantity ∆S , defined by

∆S = {sum of initial values of S} − {sum of final values of S} , (3.52)

is zero, that is, if and only if strangeness is additively conserved. The electric charge of a hadron
is defined to be
1
Q = I3 + Y , (3.53)
2
where the hypercharge Y is defined by

Y =B+S. (3.54)

Assuming that B is conserved, the selection rule for strong interaction is that they act if and
only if

∆Y = 0 . (3.55)

It is natural to assume that the possible values of Y are eigenvalues of a self-adjoint linear
operator Y. As all the particles in an isotopic multiplet are assumed to have the same value of
Y , and as Y is assumed to be simultaneously measurable with I3 , it is necessary that

[Y, Ip ] = 0 , p = 1, 2, 3, (3.56)

implying that

[Y, I 2 ] = 0 (3.57)

as well. Moreover, as Y is unchanged by transformations of the Poincaré group, Y must commute

with all the operators corresponding to the basis elements of the Lie algebra of that group. As
Y is an integer for all observed particles, it is reasonable to assume that iY is the basis element
of a u(1) real Lie algebra, so the set consisting of iY, iI1 , iI2 and iI3 forms the basis of a
u(1) ⊕ su(2) real Lie algebra. However, this alone does not imply any correlation between the
eigenvalues of Y and I3 . To obtain this, it is necessary to make the further assumption that
this u(1) ⊕ su(2) Lie algebra is the proper subalgebra of a larger real Lie algebra. The natural
candidates to consider are the rank-2 compact semi-simple real Lie algebras, because all their
relevant properties are known. Being compact, all the finite-dimensional representations of their
associated Lie groups are equivalent to unitary representations. A rank-2 algebra is appropriate
because it can accommodate two mutually commuting operators (corresponding to Y and I3 )
in its Cartan subalgebra. The non-simple candidates su(2) ⊕ su(2) can be eliminated because it
would leave the values of Y and I3 unrelated, so the choice is narrowed to the rank-2 compact
simple real lie algebras. There are 3 non-isomorphic rank-2 real Lie algebras: su(3) (the compact
real form of A2 ), so(5) (which is the compact real form of B2 ∼ = C2 ), and the compact real form
of G2 . It is now clear that the scheme based on su(3) agrees well with experimental observation,
and that it is not the case for the schemes based on the other algebras.
178 Boulanger, de Buyl

Even with su(3) selected as being the appropriate algebra, there still remains the question of
the precise relationship of Y and I3 to the basis elements of the Cartan subalgebra of A2 . This
is equivalent to the problem of assigning particles to multiplets (which was resolved by Gell-man
and Ne’eman).
The basic idea of the su(3) scheme is that Y and I3 are members of the Cartan subalgebra
of A2 , and their eigenvalues Y and I3 are determined by the weights of the irreducible represen-
tations of A2 . The set of hadrons corresponding to a particular irrep. is said to form a “unitary
multiplet” and the hadrons involved are assumed to be identical apart from their values of Y , I3
and I, so that they all have the same spin, parity and baryon number. Moreover, it is assumed
that in an ideal universe there is only one type of interaction, the strong interaction, such that
the Hamiltonian for this interaction satisfies
[ΦSQ (a), HsSQ ] = 0 , ∀ a ∈ su(3) , (3.58)

where the second-quantized operator ΦSQ (a) obeys

[ΦSQ (a), ΦSQ (a′ )] = ΦSQ ([a, a′ ]) . (3.59)

With this assumption, all the particles in the same unitary multiplet have exactly the same mass.
At this point there is a problem, because in the real world the particles in a unitary multiplet
haves masses that are only very roughly equal. The situation is quantitatively quite different
from that in the isotopic scheme, where the masses within an isotopic multiplet differ by at
most a few per cent, and where the difference can be attributed to the weaker electromagnetic
interaction. It is clear that the considerable mass-splittings between isotopic multiplets in a
unitary multiplet cannot be attributed to the electromagnetic interaction, so that it is necessary
to make the assumption that there are two types of strong interactions. The weaker version,
which will be called the “medium-strong interaction”, is assumed to be responsible for these
mass-splittings, so that, for some a ∈ su(3),
[ΦSQ (a), Hms
SQ
] 6= 0 . (3.60)

The stronger version will still be referred to as “the” strong interaction and is assumed to be
such that Eq. (3.58) still holds.
The first thing to do is to establish the relationship of Y, I1 , I2 and I3 to the basis elements
hα(1) , hα(2) , eα(1) , e−α(1) , eα(2) , e−α(2) , eα(1) +α(2) and e−(α(1) +α(2) ) of the Weyl canonical basis of
A2 . The requirements are that
(1) I1 , I2 and I3 satisfy the commutation relations in Eq. (3.3);
(2) Y satisfies the commutation relation in Eq. (3.56); and
(3) if any hadron in a unitary multiplet has integral electric charge (that is, if Q is an integer),
then all the particles in the multiplet must have integral electric charge.
These requirements lead to the assignments:
1 (1) 2 (2)
Y = Φ(H α ) + Φ(H α ) = 2Φ(hα(1) ) + 4Φ(hα(2) ) = 2Φ(H2 ) , (3.61)
3 3
1 α(1) 1 p
I1 = Φ(E ) + Φ(E −α(1) ) = (3/2)[Φ(eα(1) ) − Φ(e−α(1) )] , (3.62)
2 2
i α(1) i p
I2 = − Φ(E ) + Φ(E −α(1) ) = −i (3/2)[Φ(eα(1) ) + Φ(e−α(1) )] , (3.63)
2 2
1 α(1)
√
I3 = Φ(H ) = 3Φ(hα(1) ) = 3Φ(H1 ) , (3.64)
2
Semi-simple Lie algebras and representations (Parts I and II) 179

where H1 and H2 are the orthonormal basis elements of the Cartan subalgebra of A2 given in
subsection 3.2.
The detailed argument that leads to Equations (3.61)—(3.64) is a follows.
• The requirement (1) is easily dealt with, because the elements 12 H α , E α and E −α defined
by Equations (1.42) and (1.43) 8 satisfy exactly the same commutation relations as the
operators A3 , A+ and A− of su(2), and hence the same commutation relations (Equation
(3.30)) as I3 , I+ and I− . The choice will be made that α = α(1) 9 . The expressions for
I1 , I2 and I3 then follow immediately, as (α(1) , α(1) ) = 31 for A2 (cf. Equation (1.48) and
subsection 3.2). We thus obtain the last three Equations (3.62), (3.63) and (3.64).
√
• Because I3 is the operator corresponding to 3H1 , the requirement (2) that [Y, I3 ] = 0
implies
√ that Y = Φ(hY ) for some hY of g◦ , the Cartan subalgebra of A2 . As I+ =
6Φ(eα(1) ), the requirement [Y, I+ ] = 0 means (by Equation (1.46)) that α(1) (hY ) = 0.
In terms of the ortho-normal basis of g◦ given in subsection 3.2, let hY√= aY1 H1 + aY2 H2 ,
so that α(1) (hY ) √= aY1 α(1) (H1 ) +paY2 α(1) (H2 ) . However, as H1 = 3hα(1) , it follows
that α(1) (H1 ) =p 3(α(1) , α(1) ) = 1/3 and, by Equation (1.10) and (1.52), α(1) (H2 ) =
K(hα(1) , H2 ) = 1/3K(H1 , H2 ) = 0. Consequently the condition α(1) (hY ) = 0 implies
that aY1 = 0. The parameter aY2 is determined by the requirement (3).
• By Equation (3.53), Q = 3Φ(hα(1) ) + aY2 (Φ(hα(2) + 21 hα(1) )). However, as noted in Theorem
2.3, if λ is a weight of a representation then λ + α is also a weight for each root α (provided
that Φ(eα )|λi = 6 0). Thus the requirement (3) implies that, since each particle

in a multiplet
is associated with a weight, the number α 3hα(1) + aY2 (hα(2) + 21 hα(1) ) must be an integer
for every root α of A2 , a condition that will be satisfied if it is true for α = α(1) and α = α(2) .
Direct calculation shows that this is already true for α = α(1) , as α(1) (3hα(1) + aY2 (hα(2) +
1
2 hα(1) )) = 1 , so it remains only to consider α = α
(2)
, for which [using (α(2) , α(2) ) = 31 and
(1) (2) 1
(α , α ) = − 6 ]

1 1
α(2) (3hα(1) + aY2 (hα(2) + hα(1) )) = (aY2 − 2) . (3.65)
2 4
The simplest choice that gives an integer in Equation (3.65) is aY2 = 2, which leads to
Equation (3.61).
The irreducible representations of A2 were investigated in detail in the previous subsections.
For a weight

λ = µ1 α(1) + µ2 α(2) (3.66)

the associated eigenvalues I3 and Y of the operators I3 and Y are given by


I3 = µ1 − 21 µ2 ,
(3.67)
Y = µ2 .

The argument is simply that, by Equations (3.61)—(3.64) above, the eigenvalue I3 possessed by
the particle corresponding to the weight λ in the irrep ΓΛ (the particle belongs to a multiplet
labeled by the highest weight Λ) is I3 = λ(3hα(1) ) = 3[µ1 (α(1) , α(1) )+ µ2 (α(1) , α(2) )] = µ1 + 12 µ2 ,
8 for any positive root α of any complex semi-simple Lie algebra g, the latter definition being equivalent to

Equations (1.29) and (1.48)

9 The other possibilities α = α(2) or α = α(1) + α(2) merely corresponding to automorphisms applied to this

choice, and hence to essentially identical embeddings of su(2) in su(3).

180 Boulanger, de Buyl

and Y = λ(2hα(1) +4hα(2) ) = µ1 [2(α(1) , α(1) )+4(α(1) , α(2) )] +µ2 [2(α(1) , α(2) )+4(α(2) , α(2) )] = µ2 .
Put another way, Equations (3.67) say that the weight λ corresponding to the eigenvalues I3
and Y is (I3 + 21 Y )α(1) + Y α(2) . The resulting pairs of eigenvalues I3 and Y for the irreps {3},
{3∗ }, {8}, {6}, and {27} can thus be obtained from subsection 3.2. By Equation (3.53) the
corresponding values of the electric charge Qe are given by
Q = µ1 . (3.68)
Some very well-established non-trivial unitary multiplets are the baryon and meson octets {8}
and baryon decuplet {10} corresponding to {n, p, Σ± , Σ0 , Λ0 , Ξ− , Ξ0 } (the baryon octet with
j = 1/2 and parity +), {K ± , K 0 , K̄ 0 , π ± , π 0 , η 0 } (the meson octet with j = 0 and parity −),
and {∆± , ∆0 , ∆++ , Σ± , Σ0 , Ξ− , Ξ0 , Ω− } (the baryon decuplet with j = 3/2 and parity +). See
tables in Ref. [2] pp. 748–749 for the corresponding quark contents and the values of (I3 , Y ).

One point that appears immediately is that for the irrep {3} the values of Q are 32 , − 31 and
− 31 ,
i.e., they are not integers. There are two ways of dealing with this situation.
The first is to note that the eigenvalues Q for the unitary multiplet belonging to the irrep
Γ{n1 ,n2 } are integers if and only if (n1 − n2 )/3 is an integer. The argument is that, by Equation
(2.17) and from the definition of the fundamental weights
r
X
Λ(i) = (A−1 )ij α(j) , (3.69)
j=1

every weight λ in Γ{n1 ,n2 } is of the form

2 hX
X 2 i
1 2
λ = n Λ(1) + n Λ(2) − q α 1 (1)
−q α 2 (2)
= nj (A−1 )jk − q k α(k) ,
k=1 j=1

(where q 1 ∈ Z+ and q 2 ∈ Z+ ), so that, from Equations (3.66) and (3.68) ,

2
X 2 1 1 2 1
Q= nj (A−1 )j1 − q 1 = n + n − q 1 = − (n1 − n2 ) + n1 − q 1 .
j=1
3 3 3

As n , n , q , q are all integers, this expression is an integer if and only if (n1 − n2 )/3 is an
1 2 1 2

integer.
Actually, it can be shown (see e.g. [2]) that this is precisely the condition for Γ{n1 ,n2 } to
exponentiate and give a representation of the factor group SU (3)/Z3 . Thus it could be argued
that the non-appearance of non-integral electric charges for the observed particles is the result
of SU (3)/Z3 being the true internal symmetry group, rather than its covering group SU (3).
Much more fruitful (and bold) is the opposite proposal, made by Gell-Mann (1964) and Zweig
(1964), that the particles corresponding to the irreps {3} and {3∗ } do exist and are the basic
constituents of all the observed hadrons. Gell-Mann (1964) called the particle of the {3} quarks,
so that those of the {3∗ } become anti-quarks. The assumption is that the quarks and anti-quarks
have baryon number B = 31 while the anti-quarks correspond to B = − 13 . The three quarks are
usually called the u, d and s quarks (u corresponding to isotopic spin up, d to isotopic spin down
and s to non-zero strangeness) and the associated anti-quarks are denoted by ū, d¯ and s̄.

A Notation and Conventions

The Cartan-subalgebra generators are denoted by
Semi-simple Lie algebras and representations (Parts I and II) 181

Hi for any basis,

Hα when associated with α ∈ g⋆◦ through Eq. (1.8),
hα for cα H α [cf. Eq. (1.8)].
(i)
One denotes by hi := HCh α
the Cartan generators associated with the simple roots α(i) in
the Chevalley basis [the basis where the coefficient cα appearing in Eq. (1.8) is taken to be
cα = (α, α)/2].
In the Chevalley basis, the step operators associated with the simple roots are denoted by
i α(i) −α(i)
e := ECh and those associated with minus the simple roots are denoted by f i := ECh .
Remind that the normalization of these generators in the Chevalley basis is such that the Killing
α(i) −α(i)
form is normalized to K(ECh , ECh ) = (α(i)2,α(i) ) [cf. Eq. (1.43)].

The elements of g⋆◦ are denoted by α. The set of all roots is denoted ∆ while the positive
(resp. negative) roots are denoted ∆+ (resp. ∆− ). Note that αi means α(H i ) but α(i) stands
for a simple root. The coroot of a simple root is denoted α(i)∨ .

The Cartan matrix is chosen to be

2(α(i) , α(j) )
Aij = , i, j = 1, . . . , r
(α(j) , α(j) )

The fundamental weights are denoted by Λ(i) , see Eq.(2.12).

Acknowledgments
We thank X. Bekaert and F. A. Dolan for their reading and comments. SdB warmly thanks S.
Detournay for his advice.

References
[1] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer-Verlag,
New York, Heidelberg and Berlin (1972).

[2] J. F. Cornwell, Group Theory in Physics, Vol II, London, Academic (1984), (Techniques Of
Physics, 7).

[3] J. Fuchs and C. Schweigert, Symmetries, Lie algebras and representations: A graduate course
for physicists, Cambridge, UK University Press (1997).

[4] W. Fulton and J. Harris, Representation Theory: A First Course, Springer-Verlag (1991) 3rd
Edition
 Semi-simple Lie algebras and representations
Part III

Verma modules and the Weyl character formula

Francis Dolan

Department of Applied Mathematics and Theoretical Physics

Wilberforce Road
Cambridge
CB3 0WA, UK.

E-mail: [email protected]

Abstract. In this chapter we discuss more advanced aspects of the representation

theory of complex semi-simple Lie algebras. This discussion overlaps Parts I and II in
places. The discussion therein of finite dimensional representations of Lie algebras is
especially useful for what follows. We include a brief description of the proof of the
Weyl character formula which is often lacking in standard text books. This proof relies
mainly on the properties of standard cyclic modules and the action of the Weyl group
on finite dimensional irreducible representations - described in the next sections. Cer-
tain applications of this formula to irreducible representations are discussed including
the Kostant multiplicity formula, the dimension formula and, for decomposing prod-
ucts, the Racah-Speiser algorithm. While this discussion is by no means detailed nor
self-contained an effort has been made to illustrate the important concepts by simple
examples/exercises.

Based on the lectures presented by F. Dolan at the First Modave Summer School in Mathe-
matical Physics held at Modave, Belgium on June 19-25, 2005
Semi-simple Lie algebras and representations (Part III) 183

1 Verma Modules

We here introduce the notion of a Lie algebra module or g module. This is a vector space V
endowed with an operation g × V → V , (x, v) → xv such that the mapping ( , ) is bilinear and
[x, y]v = xyv − yxv for x, y ∈ g and v ∈ V . For two g modules V, W , considering the linear
mapping ϕ : V → W such that ϕ(xv) = xϕ(v) then ϕ is called a homomorphism of g modules
and the kernel is a g sub-module of V . When the kernel is 0 then ϕ is an isomorphism of g
modules. If a g module has no g sub-modules (other than itself and 0) then it is said to be
irreducible.
The universal enveloping algebraP∞ U (g) for a Lie algebra g is a quotient vector space formed from
(n)
the tensor algebra T (g) = i=0 T (g), T (n) (g) = g ⊗ · · · ⊗ g (n times) with the two-sided
ideal of T (g) generated by all elements of the form x ⊗ y − y ⊗ x − [x, y], x, y ∈ g. Calling
this ideal I then U (g) = T (g)/I. In this way U (g) becomes an associative algebra for which
the commutator of two elements with representatives x, y gives the element with representative
[x, y] i.e. the commutator reproduces the Lie bracket. The basis for U (g) is provided for by the
Poincaré-Birkhoff-Witt theorem: for any x1 , . . . , xk being a basis of g (and assuming that this
is countable) then x1 a1 . . . xk ak where ai is a non-negative integer is a basis for U (g).
A maximal vector in a g module is one which is annihilated by all xα ∈ g, α ∈ ∆+ . In order
to investigate the structure of finite dimensional irreducible g modules it is useful to study first
the larger class of g modules ZΛ generated by a maximal vector vΛ with weight Λ so that vΛ is
cyclic - ZΛ = U (g)vΛ . This is called a standard cyclic g module.
Various important properties of standard cyclic modules are as follows:

• Essentially due to the Poincaré-Birkhoff-Witt theorem, ZΛ is spanned by the vectors

x1 a1 . . . xk ak vΛ with ai non-negative integers. In particular ZΛ is a direct sum of the
weight spaces ZΛ,µ = {vµ ∈ ZΛ |hvµ = h(µ)vµ , h ∈ g0 }.
• All weights µ of ZΛ satisfy µ 6 Λ.
• dim ZΛ,µ < ∞ for all weights µ and dim ZΛ,Λ = 1.
• ZΛ is an indecomposable g module - in other words, ZΛ cannot be written as a direct sum
of two proper sub-modules - rather it has a unique maximal proper sub-module MΛ and a
corresponding unique irreducible quotient IΛ = ZΛ /MΛ .
• Every non-zero homomorphism of ZΛ is standard cyclic with highest weight Λ.
• Any two irreducible standard cyclic modules with the same highest weight are isomorphic.
• For every λ ∈ g0 ∗ there is an irreducible non-zero standard cyclic module of highest weight
λ.

These assertions are proved in [1].

Standard cyclic modules may be realised in terms of the universal enveloping algebra in the
following way. Choosing a basis for the Lie algebra to be hi , E ±α , α ∈ ∆+ , let IΛ ⊂ U (g)
be the left-ideal generated by E α and hi − Λi 1. We define the g module VΛ = U (g)/IΛ - the
184 Dolan

Verma module with highest weight Λ. It is now possible to show that VΛ and ZΛ = U (g)vΛ are
isomorphic as IΛ annihilates vΛ [1].
For our purposes, we can think of the Verma module VΛ as the vector space spanned by the
highest weight vector vΛ and all its descendants. In other words for

h i vΛ = Λ i vΛ , E α vΛ = 0 , α ∈ ∆+ , (1.1)

then the Verma module is the vector space with basis

a aj
(E −αi ) i . . . (E −αj ) . . . vΛ , (1.2)

for some arbitrary (but fixed) ordering of the positive roots αi , 1 6 i 6 |∆+ | and for non-negative
integers ai .
All vectors in the maximal sub-module MΛ of VΛ are called null vectors since they map to zero
in the irreducible quotient IΛ = VΛ /MΛ . Among these are distinguished ones called primitive
null vectors - these are ones for which the action of the raising operators does not yield another
null vector but in fact annihilates them. They satisfy the defining requirement for a highest
weight vector and are themselves highest weights of Verma sub-modules of VΛ .
For a unitary group so that hi† = hi , E αN † = E −αN then any state in proper sub-modules will
have vanishing norm. Supposing that a primitive null state is vΛ′ so that EαN vΛ′ = 0 then as
such a state is a linear combination of states of the form (1.2) for positive nN then clearly the
Hermitian product (vΛ′ |vΛ′ ) ≡ kvΛ′ k2 = 0. Similarly, it is not difficult to show that all other
null states have vanishing norm. It can be shown that for Λ dominant integral then the states
in IΛ will have positive definite norms.

Example 1.1. A simple example provided by A1 ≃ sl2 .

Here the Verma module Vj has basis,1

vj−n ≡ (J− )n vj , n = 0, 1, 2, . . . , (1.3)

so that

J3 vj−n = (j − n)vj−n , J− vj−n = vj−n−1 , J+ vj−n = n(2j − n + 1)vj−n+1 . (1.4)

Thus we see from the last equation that v−j−1 is the primitive null state in the Verma module
Vj for j being a non-negative integer. The irreducible module is given by the quotient

Ij = Vj /V−j−1 . (1.5)

Notice also that we may easily show that Vj for j not being a non-negative integer is irreducible
since then Vj contains no proper sub-modules.
z

1 We denote the Lie group generators by J , J for [J , J ] = ±J , [J , J ] = 2J and take the J eigenvalues
3 ± 3 ± ± + − 3 3
j to be j = n/2, n ∈ N, as is the usual convention in physical applications. We then have that that [J+ , (J− )n ] =
(J− )n−1 n(2J3 − n + 1).
Semi-simple Lie algebras and representations (Part III) 185

Example 1.2. More primitive null vectors

Assume we have a dominant integral weight Λ with Dynkin labels Λi . Recall the sl2 sub-algebra
(i) (i)
formed by hi , E α , E −α of any Lie algebra for some i. For similar reasons as in the previous
example the states of the form
(i) Λi +1
(E −α ) vΛ = 0 , (1.6)
so that these are primitive null states. z

2 The Weyl group

Consider the following mapping (for any roots α, β) defined by,

wα (β) = β − (α∨ , β) α , (2.1)

being a reflection with respect to the hyperplane through the origin and perpendicular to α.
This is a permutation of the root system as (α∨ , β) ≡ 2(α, β)/(α, α) is an integer and so wα (β)
belongs to the root lattice. In fact it is also a root of g.

The mapping (2.1) is called a Weyl reflection and the set of all such reflections forms a group
called the Weyl group. We will denote this by W. This group has generators given by simple
Weyl reflections, w(i) := wα(i) for α(i) being a simple root. Any other element of the Weyl group
w ∈ W may be decomposed in terms of these as

w = w(i1 ) · · · w(in ) , (2.2)

for some n which is generally not unique. However the signature of w defined, in the present
case, by
sgn(w) = (−1)n , (2.3)
is uniquely defined.

Note also that wα 2 = 1 and so the image of the simple roots under the mapping wα is again
another basis for the simple roots. In fact the Weyl group acts transitively and freely on the set
of bases of simple roots so that between any two choices of basis for the simple roots there exists
a unique Weyl group element relating them.2

On simple roots themselves we have from (2.1) that

w(i) (α(j) ) = α(j) − Aij α(i) , (2.4)

for [Aij ] being the Cartan matrix. A number of important points follow from (2.4). First it
implies that w(i) and w(j) commute for Aij = 0. Also, w(i) (α(i) ) = −α(i) and since Aij 6 0 for
i 6= j then w(i) permutes all other positive roots so that,


w(i) ∆+ /{α(i) } = ∆+ /{α(i) } . (2.5)
2 This fact can be used to show that the group of automorphisms of the root system is composed of Weyl group

elements and permutations of simple roots. The latter permutations are related to the ambiguity in ordering a
given basis of simple roots and so correspond to the symmetries of the Dynkin diagram [3].
186 Dolan

The action of the Weyl group thus far defined, extends naturally to any weight system. For
an arbitrary weight λ we define
wα (λ) = λ − (α∨ , λ) α . (2.6)

Thus defined, Weyl reflections leave invariant the inner product on weight space,
(λ, µ) = (wα (λ), wα (µ)) , (2.7)
and so any Weyl group element is also an isometry.
1
P
Using (2.5) and (2.7) we may show that the Weyl vector ρ = 2 α∈∆+ α has Dynkin labels
[1, 1, . . . , 1]. To see this note that (2.5) implies that
wi (ρ) = ρ − α(i) , (2.8)
so that the ith component of ρ is
ρi = (α(i)∨ , ρ) = (wi (α(i)∨ ), wi (ρ)) = (−α(i)∨ , ρ − α(i) ) = −ρi + 2 , (2.9)
so that ρi = 1 for every i ∈ {1, . . . , r}.

The Weyl group divides the weight space into a family of open sets called Weyl chambers. These
are simplicial cones defined by
Hw = {λ : (α(i)∨ , w(λ)) > 0 , 1 6 i 6 r} , (2.10)
for w ∈ W. The number of such equals the order of W. The weights lying on the boundary of the
Weyl chambers are the points on the hyperplanes perpendicular to the roots (α(i)∨ , w(λ)) = 0.
In terms of Dynkin labels these are the weights having at least one vanishing Dynkin label
(w(i) (λ) = λ if λi = 0 ).

The Weyl chamber corresponding to the identity of the Weyl group H1 is the fundamental
or dominant Weyl chamber. In terms of Dynkin labels the weights in this chamber have strictly
positive Dynkin labels.

The Weyl group acts transitively and freely on the Weyl chambers also so that for any λ ∈
/ Hw
there is precisely one element of the Weyl group which maps it into Hw . Another way of saying
this is that the Weyl group orbit of any weight has exactly one member in any Weyl chamber.

An important result is that the weight system of a finite dimensional weight module of a semi-
simple Lie algebra is the union of Weyl group orbits, each of which contains just one dominant
integral weight. Apart from the highest weight these weights have generally multiplicity greater
than one. In particular this applies for irreducible modules with dominant integral highest weight
as these are finite dimensional.

These assertions are proved rigorously in e.g. [1]. In fact from the arguments therein (essentially, that the
Weyl group acts transitively and freely on the bases of simple roots) then the restrictions on the Cartan matrix
derived previously may be easily found.
Semi-simple Lie algebras and representations (Part III) 187

Remark:
Most important for what follows are the shifted or affine Weyl reflections defined for any
weight λ and w ∈ W, in terms of the Weyl vector ρ, by

λw = w(λ + ρ) − ρ . (2.11)

Example 2.1. A simple example is for the Lie algebra of An ≃ sun+1 . In terms of the orthonor-
mal basis of the Lie algebra, a simple Weyl reflection wi acts on an arbitrary weight (ℓ1 , . . . , ℓn+1 )
as (exercise: check!)

wi ((ℓ1 , . . . , ℓi , ℓi+1 , . . . , ℓn+1 )) = (ℓ1 , . . . , ℓi+1 , ℓi , . . . , ℓn+1 ) , (2.12)

that is, it exchanges ℓi and ℓi+1 . Thus the Weyl group in this case is isomorphic to the permu-
tation group Sn+1 . For the other classical Lie algebras of Bn ≃ so2n+1 , Cn ≃ sp2n , Dn ≃ so2n
the action (2.12) is extended to allow sign flips in the labels. z

On the orthonormal basis for slN

slN Lie algebra elements can be represented by N × N traceless matrices whereby the Cartan subalgebra basis consists

1 6 i 6 N , we see that
P
of diagonal traceless matrices. Thus the Cartan subalgebra is N − 1 dimensional. Defining [Hi ]jk := δij δik − δjk /N ,
N
i=1 Hi ≡ 0 and as there is no further linear dependence among the Hi the latter matrices span
the Cartan subalgebra. We can take [Eij ]kl = δik δjl with i 6= j (of which there are N 2 − N such matrices) as a basis
for ladder operators. Then the commutator [x · H, Eij ] = x · (ei − ej )Eij , for ei forming an orthonormal basis of vectors

space as they all lie on the hyperplane normal to P =

P
in RN , implies that the root associated with Eij is ei − ej . Notice that the roots span an N − 1 dimensional vector
N
i=1 ei . The N − 1 simple roots in this basis are α
(i)
= ei − ei+1
where i = 1, . . . , N − 1. Then roots and coroots are identified and the fundamental weights are given by

Λ(j) =
X j
ek −
j X
N
ek . (2.13)
k=1
N k=1

P
P
N −1
Thus in terms of Dynkin labels the orthonormal basis labels read ℓi = λi +· · ·+λN −1 − N
1
j=1 j λj with the convention
N i N
λ ≡ 0. In terms of orthonormal basis labels the Dynkin labels are given by λ = ℓi − ℓi+1 with i=1 ℓi = 0.

Example 2.2. Consider the weight system of the eight-dimensional adjoint representation of
sl3 . In terms of the orthonormal basis the eight weights are given by those for the root system

α1 → (1, −1, 0) , α2 → (0, 1, −1) , α1 + α2 → (1, 0, −1) ,

−α1 → (−1, 1, 0) , −α2 → (0, −1, 1) , −α1 − α2 → (−1, 0, 1) , (2.14)

along with 2 × (0, 0, 0). (For the orthogonal basis weight (a, b, c) the corresponding Dynkin labels
are [a − b, b − c].) The positive roots are indicated in the first line while the negative roots are
on the second. Note that indeed permuting the labels of any such root in the orthonormal basis
generates the whole system of roots. From (2.12) it should be evident that, for α ≡ α1 + α2 , for
example

w1 (α) = α2 , w2 (α) = α1 , w1 w2 (α) = −α1 , w2 w1 (α) = −α2 ,

w1 w2 w1 (α) = −α ,
(2.15)
and that w1 w2 w1 = w2 w1 w2 . Note also that the weight system for the adjoint representation
splits into two Weyl group orbits - that for the roots (2.12) and that for (0, 0, 0). Using that the
188 Dolan

metric in the orthonormal basis is just the normal dot product it should be easy to check that the
Weyl chambers in this case are given by,

H1 = {α}, Hw1 = {α2 }, Hw2 = {α1 } ,

Hw1 w2 = {−α1 } , Hw2 w1 = {−α2 } , Hw1 w2 w1 = {−α} . (2.16)

3 Formal characters
Consider a generic weight module VΛ with a highest weight Λ. It is convenient to keep track
of all the weights λ of VΛ , including their multiplicities MVΛ (λ), by introducing the generating
function,
X
χVΛ ≡ MVΛ (λ) eλ . (3.1)
λ∈VΛ

(3.1) is more properly to be understood as a functional on weight space for which,

eλ eµ = eλ+µ , eλ (µ) = e(λ,µ) , (3.2)

where µ is some arbitrary weight. Under Weyl reflections we define

w(eλ ) = ew(λ) . (3.3)

We will refer to χVΛ (or its value on weight space) as the character of the module VΛ .
For a unitary group and with MVΛ (λ) always finite we may recover a trace formula for the
character by normalising each vector vλ ∈ VΛ corresponding to the weight λ so that kvλ k2 = 1.
Then we may write,
X

χVΛ (µ) = kvλ k2 e(λ,µ) = Tr eR(hµ ) , (3.4)
vλ ∈VΛ

where
P R is the g representation on VΛ and hµ is the Cartan sub-algebra element hµ = (µ, h) =
i
i µi h .

Example 3.1. For A1 ≃ sl2 and the irreducible representation Ij we have that
1 1
xj+ 2 − x−j− 2
χIj (x) ≡ χIj (s) = xj + xj−1 + · · · + x−j = 1 1 , (3.5)
x 2 − x− 2
′ ′
where we have defined ej (s) ≡ xj for any weight s. This is the usual character formula for sl2
irreducible representations. z

Note that all weights in a general Verma module VΛ as defined previously are given by

Λ − n1 α1 − · · · − ni αi − . . . , (3.6)

for any non-negative integers ni , where (recall that) i enumerates the positive roots. For fixed
ni in (3.6) let us call the corresponding weight λ(n1 ,...,ni ... ) then for these values of ni this weight
Semi-simple Lie algebras and representations (Part III) 189

has multiplicity one in the weight system for VΛ . From (3.1) then we may write the character
as
|∆+ | ∞ |∆+ | ∞ |∆+ |
X X X X Y Y
−ni αi
χV Λ = λ(n1 ,...,ni ... )
e =e Λ
e Λ
=e (1 − e−αi )−1 ≡ eΛ (1 − e−α )−1 .
i=1 ni =0 i=1 ni =0 i=1 α∈∆+
(3.7)
This formula is important for what follows.

It is important to realise that a given weight λ of VΛ does not have multiplicity one since there are
of course different choices of ni in (3.6) which give the same weight. We define the multiplicity
for the weight λ of VΛ to be MVΛ (λ) = P (Λ − λ) where

P (µ) = the number of ways µ can be written as a

non-negative linear combination of positive roots . (3.8)

Example 3.2. The partition number P (µ) may be easily found for sl3 , for example. Supposing
the weight µ may be decomposed in terms of the two simple roots as µ = k1 α(1) + k2 α(2) for
k1 , k2 ∈ N. Recall that the positive roots are α(1) , α(2) , α ≡ α(1) + α(2) . The number of solutions
to µ = n1 α(1) + n2 α(2) + n3 α for n1 , n2 , n3 ∈ N is clearly then the number of different ni for
which k1 = n1 + n3 , k2 = n2 + n3 . Thus P (µ) = min(k1 , k2 ) + 1 since n3 can take any integer
value between zero and min(k1 , k2 ) for n1 , n2 to remain positive or zero.
z

Example 3.3. Recall that in the orthonormal basis the metric on weight space is just the usual
dot product. For some arbitrary weight µ with components µ = (µ1 , . . . , µj , . . . ) we define the
variables
xj ≡ eej (µ) = e(ej ,µ) = eµj . (3.9)
Thus for some weight ℓ with components ℓ = (ℓ1 , . . . , ℓj , . . . ) in the aforementioned orthonormal
basis,
eℓ (µ) = x1 ℓ1 . . . xj ℓj . . . . (3.10)
Consider the gl3 adjoint representation having highest weight with Dynkin labels [1, 1] or ortho-
normal basis labels (1, 0, −1). With the variables (3.9) we have that the character is (exercise:
check using the foregoing example in the Weyl groups section!)

1
χI(1,0,−1) (x1 , x2 , x3 ) = (x1 + x2 )(x1 + x3 )(x2 + x3 ) . (3.11)
x1 x2 x3

(Note that this is the same as for the adjoint representation of sl3 considered previously save
that for sl3 then x1 x2 x3 = 1.) z
190 Dolan

Example 3.4. Using (3.7) and the variables (3.9) we may write down the character for some
gln Verma module with highest weight ℓ having components ℓ = (ℓ1 , . . . , ℓn ) in the orthonormal
basis. Recall that the positive roots in the orthonormal basis are given by

ei − ej , 1 6 i < j 6 n. (3.12)

Thus the Verma module character is

n
Y Y n
Y
χVℓ (x1 , . . . , xn ) = xi ℓi (1 − xi −1 xj )−1 = xi ℓi +n−i ∆(x1 , . . . , xn )−1 , (3.13)
i=1 16i<j6n i=1

where ∆(x1 , . . . , xn ) is the Vandermonde determinant,

Y
∆(x1 , . . . , xn ) := (xi − xj ) = det[xi j−1 ] . (3.14)
16i<j6n
Qn
The modification for an sln Verma module would be that i=1 xi = 1 in (3.13). z

3.1 The Weyl character formula

We here give an outline of a proof of the Weyl character formula for an irreducible highest weight
module with dominant integral highest weight Λ. The method is based on an approach by Kac
and recounted in [1]. Recall that an irreducible module IΛ , for highest weight Λ, is identical to
the quotient vector space VΛ /MΛ , where MΛ is the maximal sub-module contained in the Verma
module VΛ . We now wish to argue that any Verma module VΛ may be written as a composition
series in terms of irreducible (sub-)modules IΛ′ , for Λ′ 6 Λ, so that we may write,
X
χVΛ = aΛ′ χIΛ′ , (3.15)
Λ′ 6Λ

where aΛ′ are some integers.

For WΛ being some standard cyclic module, we define BWΛ = {Λ′ 6 Λ} to be the set of highest
weights of the irreducible (sub-)modules IΛ′ in the composition series of WΛ . A restriction on
such weights for BVΛ arises due to the action of the quadratic Casimir on VΛ so that they must
satisfy |Λ′ + ρ|2 = |Λ + ρ|2 . Since the set of weights λ satisfying |λ + ρ|2 = |Λ + ρ|2 is compact
and the weights of VΛ form a discrete set then necessarily BVΛ is a finite set. We may now
proceed by induction on |BVΛ |. Supposing |BVΛ | = 1 and there exists a non-zero sub-module Wλ
of VΛ . Then Wλ has a maximal vector of some weight λ < Λ which must, along with Λ, belong
to BVΛ which is a contradiction. Thus in this case VΛ is irreducible. Unless VΛ is irreducible
it contains at least one sub-module, Wλ , λ < Λ. Then as |BWλ |, |BVΛ /Wλ | < |BVΛ |, induction
implies that the standard cyclic modules Wλ , VΛ /Wλ may be written as a composition series in
terms of respective irreducible sub-modules. The two composition series may be added together
to yield a composition series of the desired type for VΛ . In terms of characters, this composition
series may be expressed as (3.15).
Let us number the weights in BVΛ by integers in such a way that for λp , λq ∈ BVΛ being such
that λq 6 λp then p 6 q. With the weights in BVΛ ordered in this way then the above argument
allows us to write X
χV λ p = bpq χIλq , (3.16)
q
Semi-simple Lie algebras and representations (Part III) 191

where the integers bpq satisfy bpp = 1 (since the highest weight of the Verma module has multi-
plicity one) and bpq is nonzero only for p 6 q (since λq ∈ Vλp obeys λq 6 λp ). Thus the matrix
[bpq ] is upper triangular with unit diagonal elements. Such matrices are generally invertible over
the integers with the inverse having, again, unit diagonal elements.
In particular for the weight Λ we may write,
X
χIΛ = cλ χVλ , (3.17)
λ∈BVΛ

where cλ ∈ Z with cΛ = 1. To compute the coefficients cλ we use a trick. Observe that the
right-hand-side of (3.17) may be rewritten as, using (3.7),
X X
cλ χVλ = P(α) cλ eλ+ρ , (3.18)
λ∈BVΛ λ∈BVΛ

where Y
P(α) = e−ρ (1 − e−α )−1 . (3.19)
α∈Φ+

Under Weyl reflections w(P(α)) = sgn(w)P(α). Thus, we necessarily have that

X X
cµ eµ+ρ = sgn(w) cλ ew(λ+ρ) , (3.20)
µ∈BVΛ λ∈BVΛ

since, for an irreducible module having dominant integral highest weight, then its weight system
is symmetric with respect to Weyl reflections. Thus
cµ = sgn(w)cλ for µ + ρ = w(λ + ρ) . (3.21)
Thus the numbers cµ are effectively determined once we know cλ for µ + ρ lying on the Weyl
group orbit of λ + ρ. Recall that the Weyl group permutes weights so that in a Weyl group orbit
there is exactly one weight in each Weyl chamber. Thus we can choose one namely the dominant
integral weight as a representative for such an orbit. For Λ dominant then Λ + ρ is the unique
dominant weight on the Weyl group orbit of any λ + ρ for λ ∈ BVΛ .3 Thus cλ = sgn(w)cΛ for
λ + ρ = w(Λ + ρ) i.e. for λ = Λw defined in (2.11). However cΛ = 1 so that we arrive at the
Weyl character formula,
Q P
χIΛ = α∈Φ+ (1 − e−α )−1 w∈W sgn(w)ew(Λ+ρ)−ρ . (3.22)

Notice that (3.7) with (3.19) imply that (3.22) may be rewritten as
X X
χIΛ = w(χVΛ ) = sgn(w)χVΛw . (3.23)
w∈W w∈W
P
We will refer to W = w∈W (∗) as the Weyl symmetry operator. (3.8) with (3.23) leads directly
to another multiplicity formula for the weight λ ∈ IΛ called the Kostant multiplicity formula,
given by X X
MIΛ (λ) = sgn(w)MVΛw (λ) = sgn(w)P (Λw − λ) . (3.24)
w∈W w∈W
3 For λ, µ, µ 6 λ, dominant weights then |µ + ρ|2 6 |λ + ρ|2
with equality only if µ = λ. This may be seen by
writing µ = λ−π where π is a sum of positive roots. Then |λ+ρ|2 −|µ+ρ|2 = (λ+ρ, π)+(π, µ+ρ) > (λ+ρ, π) > 0
with equality holding if π = 0 since λ + ρ is (strongly) dominant.
192 Dolan

Example 3.5. An insightful consistency check of (3.23) for sl3

Notice that the highest weights of the Verma module weight systems in (3.23) are shifted Weyl
reflections Λw . These correspond to primitive null vectors in the Verma module VΛ and in the
irreducible module IΛ they are quotiented out. In other words they should have multiplicity zero in
the weight system for the irreducible module IΛ . We may show this explicitly for sl3 . Let [Λ1 , Λ2 ]
denote the Dynkin labels of the highest weight Λ. Defining, for convenience, x = Λ1 +1, y = Λ2 +1
then we may show that the non-trivial shifted Weyl reflections are given by
Λw1 = Λ − xα(1) , Λw2 = Λ − yα(2) , Λw1 w2 = Λ − xα(1) − (x + y)α(2) ,
Λw2 w1 = Λ − (x + y)α(1) − yα(2) , Λw1 w2 w1 = Λ − (x + y) α(1) + α(2) ) . (3.25)
Recall that the multiplicity of the weight λ in the weight system for VΛ is MVΛ (λ) = P (Λ − λ)
where, for sl3 , P (µ) = min(k1 , k2 ) + 1 for µ = k1 α(1) + k2 α(2) , k1 , k2 ∈ N. Thus we have that
MVΛ (Λ) = 1 , MVΛ (Λw1 ) = 1 , MVΛ (Λw2 ) = 1 ,
w1 w2 w2 w1
MVΛ (Λ ) = x + 1 , MVΛ (Λ ) = y + 1 , MVΛ (Λw1 w2 w1 ) = x + y + 1 .(3.26)
Similarly
MVΛw1 (Λw1 ) = 1 , MVΛw1 (Λw1 w2 ) = 1 , MVΛw1 (Λw2 w1 ) = y + 1 ,
MVΛw1 (Λw1 w2 w1 ) = y + 1 , MVΛw2 (Λw2 ) = 1 , MVΛw2 (Λw1 w2 ) = x + 1 ,
MVΛw2 (Λw2 w1 ) = 1 , MVΛw2 (Λw1 w2 w1 ) = x + 1 , MVΛw1 w2 (Λw1 w2 ) = 1 ,
MVΛw1 w2 (Λw1 w2 w1 ) = 1 , MVΛw2 w1 (Λw2 w1 ) = 1 , MVΛw2 w1 (Λw1 w2 w1 ) = 1 , (3.27)
′
and finally, of course, MVΛw1 w2 w1 (Λw1 w2 w1 ) = 1. Otherwise MVΛw (Λw ) = 0. The point of this
is that we may then show explicitly that
′ X ′
MIΛ (Λw ) = sgn(w)MVΛw (Λw ) = 0 , (3.28)
w=1,w1 ,w2 ,w1 w2 ,
w2 w1 ,w1 w2 w1

for any w′ 6= 1.
z

Example 3.6. The denominator identity and dimension formula

The Weyl character formula may be used to write down the dimension of any irreducible
representation
P with dominant integral highest weight. For Λ = 0 in (3.22) and using that χI0 = 1
and ρ = 21 α∈∆+ α then we arrive at the denominator identity,
Y 1 1
X
(e 2 α − e− 2 α ) = sgn(w)ew(ρ) . (3.29)
α∈Φ+ w∈W

Thus the Weyl character formula my be rewritten as

P w(Λ+ρ)
w∈W sgn(w)e
χIΛ = P w(ρ)
. (3.30)
w∈W sgn(w)e

By using the denominator identity with the Weyl character formula we may show that the di-
mension of the representation is
Y (Λ + ρ, α)
dim(IΛ ) ≡ χIΛ (0) = . (3.31)
(ρ, α)
α∈∆+

z
Semi-simple Lie algebras and representations (Part III) 193

Exercise 3.7. By using the denominator identity and l’Hospital’s rule show that
P (w(Λ+ρ),tρ)
w∈W sgn(w)e
χIΛ (0) = lim P (w(ρ),tρ)
, (3.32)
w∈W sgn(w)e
t→0

leads to (3.31). Hence show that a sl3 irreducible representation with Dynkin labels [Λ1 , Λ2 ] has
dimension
dim(I[Λ1 ,Λ2 ] ) = 12 (Λ1 + 1)(Λ2 + 1)(Λ1 + Λ2 + 2) . (3.33)
z

Example 3.8. The character formula for gln

Now that we have written down the Verma module character for gln (3.13) we may without
too much effort use (3.23) to write the character for an arbitrary irreducible representation of
gln . Recall that the Weyl group is the permutation group Sn and that it acts by straightforward
permutation of the labels in the orthonormal basis. Thus we have that for any element of the
Weyl group σ ∈ Sn then, with the variable identification in (3.9) and (3.10),

eσ(ℓ) (µ) = xσ−1 (1) ℓ1 . . . xσ−1 (n) ℓn . (3.34)

Using (3.23) and the fact that we are summing over all such σ ∈ Sn and that in (3.13)
∆(xσ(1) , . . . , xσ(n) ) = sgn(σ)∆(x1 , . . . , xn ), then we arrive at an expression in terms of Schur
functions,

X n
Y
χI(ℓ1 ,...,ℓn ) (x1 , . . . , xn ) = sgn(σ) xσ(i) ℓi +n−i ∆(x1 , . . . , xn )−1
σ∈Sn i=1

= det[xi ℓj +n−j
]∆(x1 , . . . , xn )−1 . (3.35)
Qn
Note that (3.11) is a special case of (3.35). Also, again the modification for sln is that i=1 xi =
1 in (3.35). z

Exercise 3.9. Using the character formula to give the weight system.
Using the usual procedure, construct the weight system for the six-dimensional [2, 0] irreducible
representation of SU3 . Hence show that the the weights in the weight system are of multiplicity
one and are given by

[2, 0] ↔ (2, 0, 0) − ℓ′ , [−2, 2] ↔ (0, 2, 0) − ℓ′ , [0, −2] ↔ (0, 0, 2) − ℓ′ ,

[0, 1] ↔ (1, 1, 0) − ℓ′ , [1, −1] ↔ (1, 0, 1) − ℓ′ , [−1, 0] ↔ (0, 1, 1) − ℓ′ , (3.36)

where (ℓ1 , ℓ2 , ℓ3 ) denote the orthonormal basis labels and where ℓ′ = ( 23 , 32 , 23 ). (How many Weyl
orbits are there?) Show using (3.35) that in this case

χI(2,0,0) (a, b) = a2 + b2 + c2 + ab + bc + cd (3.37)

so that the character formula does indeed encode the weight system of the [2, 0] irreducible rep-
resentation. z
194 Dolan

Example 3.10. An example of where a subgroup of the Weyl group can be used to give a
concise formula for a character.
The character in (3.35) gets successively complicated the higher up in n we go. However it may
simplify more for particular representations. Consider the (p, p, 0, 0) representation of gl4 . (For
the restriction to sl4 the character corresponds to that for the [0, p, 0] representation.) Notice that
under the Klein-4 subgroup realised by K4 = {1, (12), (34), (12)(34)} then the second expression
in (3.35) is

(p)
X (x1 x2 )p+2
H12;34 = ∆(x1 , x2 , x3 , x4 )−1 xσ(1) p+3 xσ(2) p+2 xσ(3) = .
(x1 − x3 )(x1 − x4 )(x2 − x3 )(x2 − x4 )
σ∈K4
(3.38)
By considering the (six) left cosets of S4 quotiented with {1, (12), (34), (12)(34)} we may then
show that
(p) (p) (p) (p) (p) (p)
χI( p,p,0,0) (x1 , x2 , x3 , x4 ) = H12;34 + H13;24 + H14;23 + H23;14 + H24;13 + H34;12 , (3.39)

which is a significant simplification over the 24 possible terms. z

4 The Racah-Speiser Algorithm

The Racah-Speiser algorithm is a simple method of decomposing the products of irreducible rep-
resentations of some Lie algebra into a sum over other irreducible representations, where all such
representations are assumed to have dominant integral highest weights. It can be summarised
by the following rules for decomposing the product of two irreducible representations, IΛ , IΛ′ :

(i) Find the weight system of one of the representations according to the usual rules. Let us
suppose that this is done for IΛ and that the weights are given by λ.

(ii) Add the weights in this weight system to the highest weight of the other representation,
λ + Λ′ .

(iii) For all weights λ′ + Λ′ which have non-negative

L Dynkin labels (and so lie in the dominant
Weyl chamber) then these contribute λ′ (λ′ + Λ′ ) to Λ′ ⊗ Λ.

(iv) It may be the case that at least one of the Dynkin labels of some λ + Λ′ is given by −1.
For such weights it may be shown that Iλ+Λ′ = 0 and so these weights do not contribute
to the decomposition.

(v) All other weights λ′′ + Λ′ with negative Dynkin labels must be related to ones with pos-
itive Dynkin labels by shifted Weyl reflections, (λ′′ + Λ′ )wλ′′ for some unique wλ′′ ∈
W.
L The modification to the decomposition of Λ ⊗ Λ′ for the weights λ′′ + Λ′ is then
′′ ′ wλ′′
λ′′ sgn(wλ′′ )(λ + Λ ) .
Semi-simple Lie algebras and representations (Part III) 195

Example 4.1. The product [j] ⊗ [j ′ ] for sl2

Assume j − j ′ > 0. The weight system for Ij ′ is [j ′ ], [j ′ − 1], . . . , [−j ′ + 1], [−j ′ ] and so by step 2

[j] ⊗ [j ′ ] = [j + j ′ ] ⊕ [j + j ′ − 1] ⊕ · · · ⊕ [j − j ′ + 1] ⊕ [j − j ′ ] , (4.1)

which is the well known formula. If we choose instead the weight system for Ij in the situation
where j − j ′ > 0 we would gain, in addition to [j + j ′ ], . . . , [j − j ′ ] the extra representations
[j ′ − j], [j ′ − j + 1], . . . , [j − j ′ − 1] however these have negative labels and so steps 4 and 5 apply.
The Weyl reflection for the generator σ of S2 is given by

[j]σ = [−j − 1] , (4.2)

so that [j ′ − j] → sgn(σ)[j ′ − j]σ = −[j − j ′ − 1] and [j ′ − j + 1] → −[j − j ′ − 2] etc. and these

extra representations cancel among themselves. Thus we are left with (4.1) again. z

Example 4.2. A simple sl3 example

Consider the well known decomposition 8 ⊗ 8 = 27 ⊕ 10 ⊕ 10 ⊕ 8 ⊕ 8 ⊕ 1 which corresponds to
[1, 1] ⊗ [1, 1] (in terms of Dynkin labelled irreducible representations). The weights in the weight
system of I[1,1] are given by [1, 1], [−1, 2], [2, −1], 2 × [0, 0], [1, −2], [−2, 1], [−1, −1]. Thus adding
these to [1, 1] we obtain [2, 2], [0, 3], [3, 0], 2 × [1, 1], [2, −1], [−1, 2], [0, 0]. Of these, only [2, −1]
and [−1, 2] lie out side of the dominant Weyl chamber. The shifted Weyl reflections for the
generators σ1 , σ2 of S3 are given by

[a, b]σ1 = [−a − 2, a + b + 1] , [a, b]σ2 = [a + b + 1, −b − 2] . (4.3)

Thus [2, −1] → sgn(σ2 )[2, −1]σ2 = −[2, −1] and [−1, 2] → sgn(σ1 )[−1, 2]σ1 = −[−1, 2] so that
these representations in fact vanish. This is an example of step 4. We are left with

[1, 1] ⊗ [1, 1] = [2, 2] ⊕ [3, 0] ⊕ [0, 3] ⊕ 2 × [1, 1] ⊕ [0, 0] , (4.4)

which corresponds to original expression in terms of dimensions, using (3.33). z

Example 4.3. A slightly more trying sl3 example which uses step 5 more!
This time we want to show that 8 ⊗ 6 = 24 ⊕ 15 ⊕ 6 ⊕ 3. In terms of Dynkin labels this is
the decomposition of [1, 1] ⊗ [2, 0]. We have already mentioned what the weight system of I[1,1]
is above so accordingly step 2 gives the weights [3, 1], [1, 2], [4, −1], 2 × [2, 0], [3, −2], [0, 1], [1, −1].
Of these, steps 4 and 5 apply to [4, −1], [3, −2], [1, −1]. Step 4 says that [4, −1], [1, −1] can
be discarded because [4, −1] → 0 and [1, −1] → 0. For [3, −2] then notice from (4.3) that
[3, −2] → sgn(σ2 )[3, −2]σ2 = −[2, 0]. Thus in fact [3, −2] cancels one of the [2, 0] in the product.
We are left with
[1, 1] ⊗ [2, 0] = [3, 1] ⊕ [1, 2] ⊕ [2, 0] ⊕ [0, 1] . (4.5)
Using (3.33) we see that this corresponds to original expression in terms of dimensions. In fact
we have also mentioned what the weight system of I[2,0] is - the weights are given by [2, 0], [0, 1],
[−2, 2], [1, −1], [−1, 0] ,[0, −2]. Adding [1, 1] to these weights we get [3, 1], [1, 2], [−1, 3], [2, 0], [0, 1],
[0, −1]. By step 4 we are left with just [3, 1], [1, 2], [2, 0], [0, 1] which agrees with above. z
196 Dolan

The Racah-Speiser algorithm is a simple consequence of the way in which character formulae for
irreducible representations multiply. Notice that, for WW being the Weyl symmetry operator,


X

χIΛ χIΛ′ = χIΛ WW χVΛ′ = WW χIΛ χVΛ′ = MIΛ (λ)WW eλ χVΛ′ (4.6)
λ∈IΛ

this following from the fact that WW is linear and that χIΛ is invariant under the action of WW .
The last equation can be simplified, using (3.7), to
X
X
MIΛ (λ)WW χVλ+Λ′ = MIΛ (λ)χIλ+Λ′ (4.7)
λ∈IΛ λ∈IΛ

which explains step 2 since {M(λ) × λ; λ ∈ IΛ } corresponds to nothing other than the weight
system of IΛ . The extra rules for dealing with when λ + Λ′ fall outside of the dominant Weyl
chamber are consequences of


χIΛ = w χIΛ = sgn(w)χIΛw . (4.8)

Example 4.4. An example of where characters are so much more efficient for finding a weight
system!
Consider sl3 and the character χI(p,0,0) where (p, 0, 0) ↔ [p, 0] - this may be shown to be given
in general by X
χI(p,0,0) (a, b) = ar bs ct , (4.9)
0≤r,s,t≤p
r+s+t=p

for abc = 1. From (4.9) we may easily read off the weight system for the [p, 0] irreducible
representation. Thus the general rule for multiplying the general [p, 0] and [j, k] irreducible
representations of sl3 is given by
M
[p, 0] ⊗ [j, k] = [j + r − s, k + s − t] . (4.10)
0≤r,s,t≤p
r+s+t=p

(For low values of j, k, p then rules 4 and 5 are applied.)

References
[1] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer-Verlag,
New York, Heidelberg and Berlin (1972).

[2] J. F. Cornwell, Group Theory in Physics, Vol II, London, Academic (1984), (Techniques Of
Physics, 7).

[3] J. Fuchs and C. Schweigert, Symmetries, Lie algebras and representations: A graduate course
for physicists, Cambridge, UK University Press (1997).

[4] W. Fulton and J. Harris, Representation Theory: A First Course, Springer-Verlag (1991) 3rd
Edition
 Lecture Note on Clifford Algebra

Jeong-Hyuck Park

Institut des Hautes Etudes Scientifiques

Bures-sur-Yvette, 91440, France

E-mail: [email protected]

Abstract. This lecture note surveys the gamma matrices in general dimensions with
arbitrary signatures, the study of which is essential to understand the supersymmetry
in the corresponding spacetime. The contents supplement the lecture presented by the
author at the First Modave Summer School in Mathematical Physics, Belgium, June,
2005.

Based on the lectures presented by J.-H. Park at the First Modave Summer School in Math-
ematical Physics held at Modave, Belgium on June 19-25, 2005
198 Park

1 Preliminary
Where do we see Clifford algebra?
• Dirac equation, for sure.
• Supersymmetry algebra.
• Non-anti-commutative superspace.
• Division algebra, R, C, H, O.
• ADHM construction for instantons, F = ± ∗ F .

The gamma matrices in the Euclidean two-dimensions provide the fermionic oscillators,
f2 = 0 , f¯2 = 0 , {f, f¯} = 1 , (1.1)

where f = 21 (γ 1 + iγ 2 ), f¯ = 21 (γ 1 − iγ 2 ). Consequently, the irreducible representation is given

uniquely by 2 × 2 matrices acting on two dimensional spinors, |+i and |−i,
   
0 0 ¯ 0 1
f = |−ih+| = , f = |+ih−| = . (1.2)
1 0 0 0
Higher dimensional gamma matrices are then constructed by the direct products of them.

2 Gamma Matrix
We start with the following well known Lemma.

Lemma 2.1. Any matrix, M , satisfying M 2 = λ2 6= 0, λ ∈ C is diagonalizable, and furthermore

if there is another invertible matrix, N , which anti-commutes with M , {N, M } = 0, then M is
2n × 2n matrix of the form  
λ 0
M =S S −1 . (2.1)
0 −λ
In particular, trM = 0. See Sec.A for our proof.

2.1 In Even Dimensions

In even d = t + s dimensions, with metric
η µν = diag(+ + · · · + − − · · · −) , (2.2)
| {z } | {z }
t s
gamma matrices, γ µ , satisfy the Clifford algebra

γ µ γ ν + γ ν γ µ = 2η µν . (2.3)

With1
γ µ1 µ2 ···µm = γ [µ1 γ µ2 · · · γ µm ] , (2.4)
1 “[ ]” means the standard anti-symmetrization with “strength one”.
Lecture Note on Clifford Algebra 199

we define ΓM , M = 1, 2, · · · 2d by assigning numbers to independent γ µ1 µ2 ···µm , e.g. imposing

µ1 < µ2 < · · · < µm ,

ΓM = (1, γ µ , γ µν , · · · , γ µ1 µ2 ···µm , · · · , γ 12···d ) . (2.5)

Then {ΓM }/Z2 forms a group

ΓM ΓN = ΩM N ΓL , ΩM N = ±1 , (2.6)

where L is a fuction of M, N and ΩM N = ±1 does not depend on the specific choice of repre-
sentation of the gamma matrices.
Lemma (2.1) implies
1
tr(ΓM ΓN ) = ΩM N δ M N , (2.7)
2n
which shows the linear independence of {ΓM } so that any gamma matrix should not be smaller
than 2d/2 × 2d/2 .

In two-dimensions, one can take the Pauli sigma matrices, σ 1 , σ 2 as gamma matrices with a
possible factor, i, depending on the signature. In general, one can construct d + 2 dimensional
gamma matrices from d dimensional gamma matrices by taking tensor products as

(γ µ ⊗ σ 1 , 1 ⊗ σ 2 , 1 ⊗ σ 3 ) : up to a factor i . (2.8)

Thus, the smallest size of irreducible representations is 2d/2 × 2d/2 and {ΓM } forms a basis of
2d/2 × 2d/2 matrices.

By induction on the dimensions, from eq.(2.8), we may require gamma matrices to satisfy
the hermiticity condition
γ µ† = γµ . (2.9)
With this choice of gamma matrices we define γ (d+1) as
q
(d+1) t−s
γ = (−1) 2 γ 1 γ 2 · · · γ d , (2.10)

satisfying
γ (d+1) = (γ (d+1) )−1 = γ (d+1)† ,
(2.11)
{γ µ , γ (d+1) } = 0 .
For two sets of irreducible gamma matrices, γ µ , γ ′µ which are 2n × 2n, 2n′ × 2n′ respectively,
we consider a matrix X
S= Γ′M T (ΓM )−1 , (2.12)
M
′
where T , is an arbitrary 2n × 2n matrix.
This matrix satisfies for any N from eq.(2.6)

Γ′N S = SΓN . (2.13)

By Schur’s Lemmas, it should be either S = 0 or n = n′ , det S 6= 0. Furthermore, S is unique up

to constant, although T is arbitrary. This implies the uniqueness of the irreducible 2d/2 × 2d/2
gamma matrices in even d dimensions, up to the similarity transformations. These similarity
200 Park

transformations are also unique up to constant. Consequently there exist similarity transforma-
tions which relate γ µ to γ µ† , γ µ∗ , γ µT since the latter form also representations of the Clifford
algebra. By combining γ (d+1) with the similarity transformations, from eq.(2.11), we may ac-
quire the opposite sign, −γ µ† , −γ µ∗ , −γ µT as well.
Explicitly we define2 q
t(t−1)
A= (−1) 2 γ1γ2 · · · γt , (2.14)
satisfying

A = A−1 = A† , (2.15)

γ µ† = (−1)t+1 Aγ µ A−1 . (2.16)

If we write
−1
±γ µ∗ = B± γ µ B± , (2.17)
then from
γ µ = (γ µ∗ )∗ = B±
∗ ∗
B± γ µ (B± B± )−1 , (2.18)
one can normalize B± to satisfy [2, 3]
1
∗
B± B± = ε± 1 , ε± = (−1) 8 (s−t)(s−t±2) , (2.19)

†
B± B± = 1 , (2.20)

T
B± = ε± B± , (2.21)

where the unitarity follows from

−1 ∗ † µ∗ † −1 † †
γ µ = 㵆 = (±B± γµ B± )† = ±B± γ (B± ) = B± B± γ µ (B± B± )−1 , (2.22)
†
and the positive definiteness of B± B± . The calculation of ε± is essentially counting the dimen-
sions of symmetric and anti-symmetric matrices [2, 3].

The charge conjugation matrix, C± , given by

T
C± = B± A, (2.23)

satisfies3 from the properties of A and B±

−1
C± γ µ C± = ζγ µT , ζ = ±(−1)t+1 , (2.24)

†
C± C± = 1 , (2.25)

1 1
T
C± = (−1) 8 d(d−ζ2) C± = ε± (±1)t (−1) 2 t(t−1) C± , (2.26)

1
−1 −1
ζ t (−1) 2 t(t−1) AT = B± AB± = C± AC± . (2.27)
2 Alternatively, one can construct C± explicitly out of the gamma matrices in a certain representation [?].
3 Essentially all the properties of the charge conjugation matrix, C± depends only on d and ζ. However it is
useful here to have expression in terms of the signature to dicuss the Majorana supersymmetry later.
Lecture Note on Clifford Algebra 201

ε± is related to ζ as
1 1
ε± = ζ t (−1) 2 t(t−1)+ 8 d(d−ζ2) . (2.28)
Eqs.(2.24, 2.26) imply
1 1
(C± γ µ1 µ2 ···µn )T = ζ n (−1) 8 d(d−ζ2)+ 2 n(n−1) C± γ µ1 µ2 ···µn
(2.29)
1
= ε± (±1)t+n (−1)n+ 2 (t+n)(t+n−1) C± γ µ1 µ2 ···µn .

γ (d+1) satisfies
γ (d+1)† = (−1)t A± γ (d+1) A−1
± =γ
(d+1)
,
t−s −1
γ (d+1)∗ = (−1) 2 B± γ (d+1) B± , (2.30)
t+s −1
γ (d+1)T = (−1) 2 C± γ (d+1) C± ,
where {A+ , A− } = {A, γ (d+1) A}.

In stead of eq.(2.8) one can construct d + 2 dimensional gamma matrices from d dimensional
gamma matrices by taking tensor products as
(γ µ ⊗ σ 1 , γ (d+1) ⊗ σ 1 , 1 ⊗ σ 2 ) : up to a factor i . (2.31)
Therefore the gamma matrices in even dimensions can be chosen to have the “off-block diagonal”
form    
µ 0 σµ (d+1) 1 0
γ = , γ = , (2.32)
σ̃ µ 0 0 −1
d d
where the 2 2 −1 × 2 2 −1 matrices, σ µ , σ̃ µ satisfy
σ µ σ̃ ν + σ ν σ̃ µ = 2η µν , (2.33)

σ µ† = σ̃µ . (2.34)
In this choice of gamma matrices, from eq.(2.30), A± , B± , C± are either “block diagonal” or
“off-block diagonal” depending on whether t, t−s t+s
2 , 2 are even or odd respectively.
In particular, in the case of odd t, we write from eqs.(2.14, 2.15) A as
  q
0 a t(t−1)
A= , a = (−1) 2 σ 1 σ̃ 2 · · · σ t = ㆠ= ã−1 , (2.35)
ã 0
t+s
and in the case of odd 2 we write from eq.(2.26) C± as
 
0 c t(t−1)
C± = , c = ε+ (−1) 2 c̃T = (c† )−1 , (2.36)
±c̃ 0
where a, ã, c, c̃ satisfy from eqs.(2.16, 2.24)
σ µ† = ãσ µ ã , σ̃ µ† = aσ̃ µ a ,
(2.37)
σ µT = (−1)t+1 c̃σ µ c−1 , σ̃ µT = (−1)t+1 cσ̃ µ c̃−1 .
t+s
If both of t and 2 are odd then from eq.(2.27)
t−1 t−1
aT = (−1) 2 c̃ a c−1 , ãT = (−1) 2 c ã c̃−1 . (2.38)
202 Park

2.2 In Odd Dimensions

The gamma matrices in odd d + 1 = t + s dimensions are constructed by combining a set of even
d dimensional gamma matrices with either ±γ (d+1) or ±iγ (d+1) depending on the signature of
even d dimensions. This way of construction is general, since γ (d+1) serves the role of γ d+1

−γ µ = γ d+1 γ µ (γ d+1 )−1 , for µ = 1, 2, · · · , d ,

(2.39)
(γ d+1 )2 = ±1 ,

and such a matrix is unique in irreducible representations up to sign.

However, contrary to the even dimensional Clifford algebra, in odd dimensions two different
choices of the signs in γ d+1 bring two irreducible representations for the Clifford algebra, which
can not be mapped to each other4 by similarity transformations

γ µ = (γ 1 , γ 2 , · · · , γ d+1 ) and γ ′µ = (γ 1 , γ 2 , · · · , γ d , −γ d+1 ) . (2.40)

If there were a similarity transformation between these two, it should have been identity up to
constant because of the uniqueness of the similarity transformation in even dimensions. Clearly
this would be a contradiction due to the presence of the two opposite signs in γ d+1 .

In general one can put5


 ±γ 12···d for t − s ≡ 1 mod 4 ,
γ d+1 = (2.41)

±iγ 12···d for t − s ≡ 3 mod 4 .

2d/2 × 2d/2 gamma matrices in odd d + 1 dimensions, γ µ , µ = 1, 2, · · · , d + 1, induce the

following basis of 2d/2 × 2d/2 matrices, Γ̃M

Γ̃M = (1, γ µ , γ µν , · · · , γ µ1 µ2 ···µd/2 ) , M = 1, 2, · · · 2d . (2.42)

From eq.(2.41)
Γ̃M Γ̃N = Ω̃M N Γ̃L ,

 ±1 for t − s ≡ 1 mod 4 , (2.43)
Ω̃M N =

±1, ±i For t − s ≡ 3 mod 4 .
Here, contrary to the even dimensional case, Ω̃M N depends on each particular choice of the
representations due to the arbitrary sign factor in γ d+1 . This is why eq.(2.13) does not hold
in odd dimensions. Therefore it is not peculiar that not all of ±γ µ† , ±γ µ∗ , ±γ µT are related
to γ µ by similarity transformations. In fact, if it were true, say for ±γ µ∗ , then the similarity
transformation should have been B± (2.17) by the uniqueness of the similarity transformations in
even dimensions, but this would be a contradiction to eq.(2.30), where the sign does not alternate
4 Nevertheless, this can be cured by the following transformation. Under xµ = (x1 , x2 , · · · , xd+1 ) → x′µ =

(x1 , x2 , · · · , −xd+1 ), we transform the Dirac field ψ(x) as ψ(x) → ψ ′ (x′ ) = ψ(x) , to get ψ̄(x)γ ·
∂ψ(x) → ψ̄ ′ (x′ )γ ′ · ∂ ′ ψ ′ (x′ ) = ψ̄(x)γ · ∂ψ(x) . Hence those two representations are equivalent describing
the same physical system.
5 Our results (2.41-2.50) do not depend on the choice of the signature in d dimensions, i.e. they hold for either

increasing the time dimensions, d = (t − 1) + s or the space dimensions, d = t + (s − 1).

Lecture Note on Clifford Algebra 203

under the change of B+ ↔ B− . Thus, in odd dimensions, only the half of ±γ µ† , ±γ µ∗ , ±γ µT are
related to γ µ by similarity transformations and hence from eq.(2.30) there exist three similarity
transformations, A, B, C such that

(−1)t+1 γ µ† = Aγ µ A−1 , (2.44)

t−s−1
(−1) 2 γ µ∗ = Bγ µ B −1 , (2.45)

t+s−1
(−1) 2 γ µT = Cγ µ C −1 . (2.46)

A, B, C are all unitary and satisfy

A = A−1 = A† , C = BT A , (2.47)

1
B ∗ B = ε 1 = (−1) 8 (t−s+1)(t−s−1) 1 , (2.48)

ts 1
B T = εB , C T = ε(−1) 2 C = (−1) 8 (t+s+1)(t+s−1) C , (2.49)

ts
(−1) 2 AT = BAB −1 = CAC −1 . (2.50)

In particular, A is given by eq.(2.14).

2.3 Lorentz Transformations

Lorentz transformations, L can be represented by the following action on gamma matrices in a
standard way
L−1 γ µ L = Lµν γ ν , (2.51)
where L and L are given by
µν 1 µν
L = ewµν M , L = e 2 wµν γ ,
(2.52)
(M µν )λρ = η µλ δ νρ − η νλ δ µρ .

For a even d, if a 2d/2 × 2d/2 matrix, M µ1 µ2 ···µn , is totally anti-symmetric over the n spacetime
indices
M µ1 µ2 ···µn = M [µ1 µ2 ···µn ] , (2.53)
and transforms covariantly under Lorentz transformations in d or d + 1 dimensions as
n
Y
L−1 M µ1 µ2 ···µn L = Lµi νi M ν1 ν2 ···νn , (2.54)
i=1

then for 0 ≤ n ≤ max(d/2, 2), the general forms of M µ1 µ2 ···µn are


 (1 + cγ (d+1) )γ µ1 µ2 ···µn In even d dimensions ,
µ1 µ2 ···µn
M = (2.55)
 µ1 µ2 ···µn
γ In odd d + 1 dimensions ,
204 Park

where c is a constant.

To show this, one may first expand M µ1 µ2 ···µn in terms of γν1 ν2 ···νm , γ (d+1) γν1 ν2 ···νm or
γν1 ν2 ···νm depending on the dimensions, d or d + 1, with 0 ≤ m ≤ d/2. Then eq.(2.54) im-
plies that the coefficients of them, say T µ1 µ2 ···µm+n , are Lorentz invariant tensors satisfying
m+n
Y
Lµνii T ν1 ν2 ···νm+n = T µ1 µ2 ···µm+n (2.56)
i=1

Finally one can recall the well known fact [4] that the general forms of Lorentz invariant tensors
are multi-products of the metric, η µν , and the totally antisymmetric tensor, ǫµ1 µ2 ··· , which verifies
eq.(2.55).

2.4 Crucial Identities for Super Yang-Mills

The following identities are crucial to show the existence of the non-Abelian super Yang-Mills
in THREE, FOUR, SIX and TEN dimensions.

(i) The following identity holds only in THREE or FOUR dimensions with arbitrary signature

0 = (γ µ C −1 )αβ (γµ C −1 )γδ + cyclic permutations of α, β, γ (2.57)

To verify the identity in even dimensions we contract (γ µ C −1 )αβ (γµ )γδ with (Cγ ν1 ν2 ···νn )βα and
take cyclic permutations of α, β, γ to get
1 1
0 = 2d/2 δ1n + (d − 2n)(ζ + ζ n (−1) 2 n(n−1) )(−1)n+ 8 d(d−ζ2) (2.58)

This equation must be satisfied for all 0 ≤ n ≤ d, which is valid only in d = 4, ζ = −1.
Similar analysis can be done for the d+1 odd dimensions by adding (γ (d+1) C −1 )αβ (γ (d+1) C −1 )γδ
term into eq.(2.57). We get
1 1
0 = 2d/2 (δ1n + δdn ) + (d − 2n + 1)(ζ + ζ n (−1) 2 n(n−1) )(−1)n+ 8 d(d−ζ2) ,
ζ = (−1)d/2
(2.59)
Only in d = 2 and hence three dimensions, this equation is satisfied for all 0 ≤ n ≤ d.

(ii) The following identity holds only in TWO, FOUR or SIX dimensions with arbitrary sig-
nature
0 = (σ µ )αβ (σµ )γδ + (σ µ )γβ (σµ )αδ (2.60)
To verify this identity we take d dimensional sigma matrices from f = d − 2 dimensional gamma
matrices as in eq.(2.31)
σ µ = (γ µ , γ (f +1) , i) (2.61)
to get
(σ µ )αβ (σµ )γδ = (γ µ )αβ (γµ )γδ + (γ (f +1) )αβ (γ (f +1) )γδ − δαβ δγδ (2.62)
Again this expression is valid for any signature, (t, s). Now we contract this equation with
−1
(γ ν1 ν2 ···νn C+ )βδ . From eqs.(2.24, 2.30) in the case of odd t we get
 f

−1
(−1)n (f − 2n) + (−1) 2 +n − 1 (γ ν1 ν2 ···νn C+ )αγ (2.63)
Lecture Note on Clifford Algebra 205

To satisfy eq.(2.60) this expression must be anti-symmetric over α ↔ γ for any 0 ≤ n ≤ f .

f
Thus from eq.(2.29) we must require 0 = (−1)n (f − 2n) + (−1) 2 +n − 1 for all n satisfying
1 1
(−1) 8 f (f −2)+ 2 n(n−1) = 1. This condition is satisfied only in f = 0, 2, 4 and hence d = 2, 4, 6 (f =
6 case is excluded by choosing n = 6 and f ≥ 8 cases are excluded by choosing either n = 0 or
n = 3).

(iii) The following identity holds only in TWO or TEN dimensions with arbitrary signature

0 = (σ µ c−1 )αβ (σµ c−1 )γδ + cyclic permutations of α, β, γ (2.64)

2.5 Super Yang-Mills

• Six-dimensions of arbitrary signature
With

T −1
ΓM = ±C± ΓM C± , C± T
= ∓C± , (2.65)
we have
T
C± ΓM = −C± ΓM . (2.66)
We introduce a pair of Weyl spinors of the same chirality,
(ψ1 , ψ2 ) , Γ(7) ψi = sψi , s2 = 1 , (2.67)
and define the charge conjugate spinor by
ψ̄ci := ǫ−1ij ψjT C± . (2.68)

The super Yang-Mills Lagrangian reads

1


L6D = tr 4 FM N F
MN
+ 12 ψ̄ci ΓM DM ψi , (2.69)
and the supersymmetry transformations are given by
δAM = ε̄ic ΓM ψi = −ψ̄ci ΓM εi ,
(2.70)
δψi = − 12 FM N ΓM N εi ,
so that, in particular, δ ψ̄ci = + 12 FM N ε̄ic ΓM N . The Lagrangian transforms as, from (2.60),


δL6D = ∂M tr F M N δAN − 21 ψ̄ci ΓM DM δψi . (2.71)
∗
Only if B± B± = −1, as in the Minkowskian signature, one can impose the pseudo-
Majorana condition,
ψ̄ci = ψ̄D
i
:= (ψi )† A . (2.72)

3 Spinors
3.1 Weyl Spinor
In any even d dimensions, Weyl spinor, ψ, satisfies
γ (d+1) ψ = ψ (3.1)
†
and so ψ̄ = ψ A satisfies from eq.(2.30)
−1 T t−s −1 T
ψ̄γ (d+1) = (−1)t ψ̄ γ (d+1) C± ψ̄ = (−1) 2 C± ψ̄ (3.2)
206 Park

3.2 Majorana Spinor

By definition Majorana spinor satisfies

ψ̄ = ψ T C± or ψ̄ = ψ T C (3.3)

depending on the dimensions, even or odd. This is possible only if ε± , ε = 1 and so from
eqs.(2.19, 2.48)
η = +1 : t − s = 0, 1, 2 mod 8
(3.4)
η = −1 : t − s = 0, 6, 7 mod 8
where η is the sign factor, ±1, occuring in eq.(2.17) or eq.(2.45)6 .

3.3 Majorana-Weyl Spinor

Majorana-Weyl spinor satisfies both of the two conditions above

γ (d+1) ψ = ψ ψ̄ = ψ T C± (3.5)

Majorana-Weyl Spinor exists only if

η = +1 : t − s = 0 mod 8
(3.6)
η = −1 : t − s = 0 mod 8

4 Majorana Representation and SO(8)

Fact 1:
Consider a finite dimensional vector space, V with the unitary and symmetric matrix, B = B T ,
BB † = 1. For every |vi ∈ V if B|vi∗ ∈ V then there exists an orthonormal “semi-real ” basis,
V = {|li, l = 1, 2, · · · } such that B|li∗ = |li.

Proof
Start with an arbitrary orthonormal bais, {|vl i, l = 1, 2, · · · } and let |1i ∝ |v1 i + B|v1 i∗ . After
the normalization, h1|1i = 1, we can take a new orthonormal basis, {|1i, |2′ i, |3′ i, · · · }. Now we
assume that {|1i, |2i, · · · |k − 1i, |k ′ i, |(k + 1)′ i, · · · } is an orhonormal basis such that B|ji∗ = |ji
for 1 ≤ j ≤ k − 1. To construct the k th such a vector, |ki we set |ki ∝ |k ′ i + B|k ′ i∗ with the
normalization. We check this is orthogonal to |ji, 1 ≤ j ≤ k − 1

hj|( |k ′ i + B|k ′ i∗ ) = 0 + hk|ji = 0 . (4.1)

In this way one can construct the desired basis.

In the spacetime which admits Majorana spinor from Eq.(3.4)

η = +1 : t − s = 0, 1, 2 mod 8
(4.2)
η = −1 : t − s = 0, 6, 7 mod 8 ,
6 In [2], η = −1 case is called Majorana and η = +1 case is called pseudo-Majorana.
Lecture Note on Clifford Algebra 207

more explicitly in the even dimensions having ε+ = 1 (or ε− = 1) where B+ (or B− ) is symmetric
and also in the odd dimensions of ε = 1 where B is symmetric, from the fact 1 above we can
choose an “semi-real ” orthonormal basis such that Bη† |li∗ = |li In the basis, we write the gamma
matrices X µ
γµ = Rlm |lihm| . (4.3)
From η γ µ∗ = Bη γ µ Bη−1 and the property of the semi-real basis, Bη |li∗ = |li we get
µ
∗ µ
Rlm = ηRlm . (4.4)

Since Rµ is also a representation of the gamma matrix

Rµ Rν + Rν Rµ = 2η µν , (4.5)

adopting the true real basis, we conclude that there exists a Majorana represention where
the gamma matrices are real, η = + or pure imaginary, η = − in any spacetime
admitting Majorana spinors.

Furthermore from Eq.(2.30), in the even dimension of t − s ≡ 0 mod 8, ε± = 1 and γ (d+1)∗ =

Bγ (d+1)
B −1 (here we omit the subscript index ± or η for simplicity.). The action, |vi → B † |vi∗
preserves the chirality, and from the fact 1 above we can choose an orthonormal semi-real basis
for the chiral and anti-chiral spinor spaces, V = V+ + V− , V± = {|l± i} such that

hl± |m± i = δlm , hl± |m∓ i = 0 , γ (d+1) |l± i = ±|l± i , B † |l± i∗ = |l± i . (4.6)

With the semi-real basis  

(d+1) 1 0
γ = , (4.7)
0 −1
and the gamma matrices are in the Majorana representation
 
µ 0 rµ
γ = , rµ ∈ O(2d/2−1 ) , rµ rν T + rν rµT = 2δ µν . (4.8)
rµT 0

From Eq.(4.6) any two sets of semi-real basis, say {|l± i} and {|˜l± i} are connected by an
O((2d/2−1 )) transformation
X X
|˜l± i = Λ±ml |m± i , Λ±lm Λ±nm = δln . (4.9)
m m

If we define X
Λ± = Λ±lm |l± ihm± | , (4.10)
l,m

then |˜l± i = Λ± |l± i and from the definition of the semi-real basis

Λ± = B † Λ∗± B = Λ± P± = P± Λ± , Λ± Λ†± = P± . (4.11)

We write
∞
X 1
Λ± = eM± , M± ≡ (−1)n+1 (Λ± − P± )n = ln Λ± . (4.12)
n=1
n
Thus for Λ± such that the infinity sum converges we have

M± = −M±† = B † M±∗ B = M± P± = P± M± . (4.13)

208 Park

This gives a strong constraint when we express M± by the gamma matrix products. For the
Eucledean eight dimensions only the SO(8) generators for the spinors survive in the expansion!

M± = 21 wab γ ab P± . (4.14)

Namely we find an isomorphism between the two SO(8)’s, one for the semi-real vectors and the
other for the spinors in the conventional sense. Alternatively this can be seen from
 [a b] T 
ab r r 0
γ = , (4.15)
0 r[a T rb]

where the each block diagonal is a generator of SO(D) while the dimension of the chiral space
is 2d/2−1 . Only in d = 8 both coincide leading to the “so(8) triolity” among sov (8), soc (8) and
soc̄ (8).

Fact 2: Relation to octonions.

In Euclidean eight dimensions, the 16×16 gamma matrices can be taken of the off-block diagonal
form,  
0 ra
γa = , ra rbT + rb raT = 2δab , (4.16)
raT 0
where the 8 × 8 real matrices, ra , 1 ≤ a ≤ 8, give the multiplication of the octonions, oa ,

oa ob = (ra )b c oc . (4.17)

Fact 3:
Consider an arbitrary real self-dual or anti-self-dual four form in D = 8
±
Tabcd = ± 14 ǫabcdef gh T ±ef gh . (4.18)

Using the SO(8) rotations one can transform the four form into the canonical form where the non-
± ± ± ± ± ± ±
vanishing components are T1234 , T1256 , T1278 , T1357 , T1368 , T1458 , T1467 and their dual counter
parts only.

Proof
We start with the seven linearly independent traceless Hermitian matrices

E±1 = γ 2341 P± , E±2 = γ 2561 P± , E±3 = γ 2781 P± , E±4 = γ 1357 P± ,

(4.19)
E±5 = γ 3681 P± , E±6 = γ 4581 P± , E±7 = γ 4671 P± .

As they commute each other, there exists a basis V± = {|l± i} diagonalizing the seven quantities
X
E±r = λrl |l± ihl± | , (λrl )2 = 1 . (4.20)
l
∗
Further, since C|l± i is also an eigenvector of the same eigenvalues, from the fact 1 we can
impose the semi-reality condition without loss of generality, C|l± i∗ = |l± i.
Lecture Note on Clifford Algebra 209

Now for the self-dual four form we let

±
T± = 1
4 Tabcd γ abcd . (4.21)

Since T ± is Hermitian and C(T ± )∗ C † = T ± , one can diagonalize T ± with a semi-real basis
X
T± = λl |˜l± ih˜l± | , C|˜l± i∗ = |˜l± i . (4.22)
l

For the two semi-real basis above we define a transformation matrix

O± = |l± ih˜l± | . (4.23)

Then, since T ± is traceless, O± T ± O±† can be written in terms of E±i ’s. Finally the fact O±
gives a spinorial SO(8) rotation completes our proof.

Some useful formulae are

±P± = E±1 E±2 E±3 = E±1 E±4 E±5 = E±1 E±6 E±7 = E±2 E±4 E±6
(4.24)
= E±2 E±5 E±7 = E±3 E±4 E±7 = E±3 E±5 E±6 .

For an arbitrary self-dual or anti-self-dual four form tensor in D = 8, from

± 1 2 ±
Tacde T ±bcde = ( 4! ) ǫacdef ghi ǫbcdejklm T ±f ghi Tjklm
(4.25)
1 ± ±
= 4 δa b Tcdef T ±cdef − Tacde T ±bcde ,

we obtain an identity
± ±
Tacde T ±bcde = 1
8 δa b Tcdef T ±cdef . (4.26)

5 Superalgebra
5.1 Graded Lie Algebra
Supersymmetry algebra is a Ẑ2 graded Lie algebra, g = {Ta }, which is an algebra with commu-
tation and anti-commutation relations [?, ?]
c
[Ta , Tb } = Cab Tc (5.1)
c
where Cab is the structure constant and

[Ta , Tb } = Ta Tb − (−1)#a#b Tb Ta (5.2)

with #a, the Ẑ2 grading of Ta ,


0 for bosonic a
#a = (5.3)
1 for fermionic a

The generalized Jacobi identity is

[Ta , [Tb , Tc }} − (−1)#a#b [Tb , [Ta , Tc }} = [[Ta , Tb }, Tc } (5.4)

210 Park

which implies

(−1)#a#c Cab
d e
Cdc + (−1)#b#a Cbc
d e
Cda + (−1)#c#b Cca
d e
Cdb =0 (5.5)

For a graded Lie algebra we consider

g(z) = exp(z a Ta ) (5.6)

where z a is a superspace coordinate component which has the same bosonic or fermionic property
as Ta and hence z a Ta is bosonic.
In the general case of non-commuting objects, say A and B, the Baker-Campbell-Haussdorff
formula gives !
X∞
eA eB = exp Cn (A, B) (5.7)
n=0

where Cn (A, B) involves n commutators. The first three of these are

C0 (A, B) = A + B

C1 (A, B) = 21 [A, B] (5.8)

1 1
C2 (A, B) = 12 [[A, B], B] + 12 [A, [A, B]]

Since for the graded algebra

[z a Ta , z b Tb ] = z b z a [Ta , Tb } = z b z a Cab
c
Tc (5.9)

the Baker-Campbell-Haussdorff formula (5.7) implies that g(z) forms a group, the graded Lie
group. Hence we may define a function on superspace, f a (w, z), by

g(w)g(z) = g(f (w, z)) (5.10)

Since g(0) = e, the identity, we have f (0, z) = z, f (w, 0) = w and further we assume that f (w, z)
has a Taylor expansion in the neighbourhood of w = z = 0.
Associativity of the group multiplication requires f (w, z) to satisfy

f (f (u, w), z) = f (u, f (w, z)) (5.11)

5.2 Left & Right Invariant Derivatives

For a graded Lie group, left and right invariant derivatives, La , Ra are defined by

La g(z) = g(z)Ta (5.12)

Ra g(z) = −Ta g(z) (5.13)

Explicitly we have

b b ∂f b (z, u) 
La = La (z)∂b La (z) = (5.14)
∂ua  u=0

∂f b (u, z) 
Ra = Ra b (z)∂b Ra b (z) = − (5.15)
∂ua u=0
Lecture Note on Clifford Algebra 211

∂
where ∂b = ∂z b.

It is easy to see that La is invariant under left action, g(z) → hg(z), and Ra is invariant under
right action, g(z) → g(z)h.
From eqs.(5.12, 5.13) we get
c
[La , Lb } = Cab Lc (5.16)

c
[Ra , Rb } = Cab Rc (5.17)

and from eqs.(5.12, 5.13) we can also easily show

[La , Rb } = 0 (5.18)

Thus, La (z), Ra (z) form representations of the graded Lie algebra separately. For the super-
symmetry algebra, the left invariant derivatives become covariant derivatives, while the right
invariant derivatives become the generators of the supersymmetry algebra acting on superfields.

5.3 Superspace & Supermatrices

In general a superspace may be denoted by Rp|q , where p, q are the number of real commuting
(bosonic) and anti-commuting (fermionic) variables respectively. A supermatrix which takes
Rp|q → Rp|q may be represented by a (p + q) × (p + q) matrix, M , of the form
 
a b
M= (5.19)
c d

where a, d are p × p, q × q matrices of Grassmanian even or bosonic variables and b, c are p × q,

q × p matrices of Grassmanian odd or fermionic variables respectively.
The inverse of M can be expressed as
 
(a − bd−1 c)−1 −a−1 b(d − ca−1 b)−1
M −1 = (5.20)
−d−1 c(a − bd−1 c)−1 (d − ca−1 b)−1

where we may write

∞
X
(a − bd−1 c)−1 = a−1 + (a−1 bd−1 c)n a−1 (5.21)
n=1

Note that due to the fermionic property of b, c, the power series terminates at n ≤ pq + 1.
The supertrace and the superdeterminant of M are defined as

str M = tr a − tr d (5.22)

sdet M = det(a − bd−1 c)/ det d = det a/ det(d − ca−1 b) (5.23)

The last equality comes from

det(1 − a−1 bd−1 c) = det−1 (1 − d−1 ca−1 b) (5.24)

which may be shown using

∞
!
X 1
det(1 − a) = exp − tr an (5.25)
n=1
n
212 Park

and observing
tr (a−1 bd−1 c)n = −tr (d−1 ca−1 b)n (5.26)
−1
From eq.(5.23) we note that sdet M 6= 0 implies the existence of M . Thus the set of super-
matrices for sdet M 6= 0 forms the supergroup, Gl(p|q). If sdet M = 1 then
M ∈ Sl(p|q).
The supertrace and the superdeterminant have the properties

str (M1 M2 ) = str (M2 M1 ) (5.27)

sdet (M1 M2 ) = sdet M1 sdet M2 (5.28)

We may define the transpose of the supermatrix, M , either as

 t 
a ct
Mt = (5.29)
−bt dt
or as  
′ at −ct
Mt = (5.30)
bt dt
where at , bt , ct , dt are the ordinary transposes of a, b, c, d respectively.
We note that
′ ′ ′
(M1 M2 )t = M2t M1t (M1 M2 )t = M2t M1t (5.31)

′ ′
(M t )t = (M t )t = M (5.32)

A Proof of the Lemma

Lemma 1
Any N × N matrix, M , satisfying M 2 = λ2 1N ×N , λ 6= 0, is diagonalizable.

Proof
Suppose for some K, 1 ≤ K ≤ N , we have found a basis,

{ea , vr : 1 ≤ a ≤ K, 1 ≤ r ≤ N − K} (A.1)

such that
M ea = λa ea , for 1 ≤ a ≤ K ,
(A.2)
M vr = P s r vs + ha r ea , for K + 1 ≤ r, s ≤ N .
From M 2 = λ2 1N ×N ,
λ2a = λ2 ,
(A.3)
λ2 vr = (P 2 )s r vs + [(hP )a r + λa ha r ] ea ,
and hence,
P 2 = λ2 1(N −K)×(N −K) ,
(A.4)
(hP )a r + λa ha r = 0 .
Lecture Note on Clifford Algebra 213

The assumption holds for K = 1 surely. In order to construct eK+1 we first consider an eigen-
vector of the (N − K) × (N − K) matrix, P ,

P r s cs = λK+1 cr , λ2K+1 = λ2 , (A.5)

and set
v = cr vr , ha = ha r cr ,
(A.6)
M v = λK+1 v + ha ea .
Consequently
(λK+1 + λa )ha = 0 : not a sum , (A.7)
so that
ha = 0 if λK+1 + λa 6= 0 . (A.8)
a
We construct eK+1 , with K unknown coefficients, d , as

eK+1 = v + da ea . (A.9)

From
M eK+1 = λK+1 eK+1 + [ha + (λa − λK+1 )da ] ea , (A.10)
we determine 

 ha
 if λK+1 6= λa ,
λK+1 − λa
da = (A.11)


 any number if λK+1 = λa .
From (A.8) and λ2K+1 = λ2a = λ2 6= 0, we have

M eK+1 = λK+1 eK+1 . (A.12)

This completes our proof.

If we set a N × N invertible matrix, S, by

(S)b a = (ea )b , M ea = λa ea , 1 ≤ a, b ≤ N , (A.13)

then
S −1 M S = diag(λ1 , λ2 , · · · , λN ) . (A.14)

References
[1] J. A. Strathdee, Int. J. Mod. Phys. A 2 (1987) 273.

[2] T. Kugo and P. K. Townsend, Nucl. Phys. B 221, 357 (1983).

[3] F. Gliozzi, J. Scherk and D. I. Olive, Nucl. Phys. B 122, 253 (1977).

[4] H. Weyl, The Classical Groups, Princeton University Press, 1946, P53.
 Universal Enveloping Algebras and some applications in
physics

Xavier Bekaert

Institut des Hautes Études Scientifiques

Le Bois-Marie, 35 route de Chartres
91440 Bures-sur-Yvette, France

E-mail: [email protected]

Abstract. These notes are intended to provide a self-contained and pedagogical in-
troduction to the universal enveloping algebras and some of their uses in mathematical
physics. After reviewing their abstract definitions and properties, the focus is put
on their relevance in Weyl calculus, in representation theory and their appearance as
higher symmetries of physical systems.

Based on the lectures presented by X. Bekaert at the First Modave Summer School in Math-
ematical Physics held at Modave, Belgium on June 19-25, 2005
Universal Enveloping Algebras 215

These lecture notes are written by a layman in abstract algebra and are aimed for other aliens
to this vast and dry planet, therefore many basic definitions are reviewed. Indeed, physicists
may be unfamiliar with the daily-life terminology of mathematicians and translation rules might
prove to be useful in order to have access to the mathematical literature. Each definition is
particularized to the finite-dimensional case to gain some intuition and make contact between
the abstract definitions and familiar objects.
The lecture notes are divided into four sections. In the first section, several examples of
associative algebras that will be used throughout the text are provided. Associative and Lie
algebras are also compared in order to motivate the introduction of enveloping algebras. The
Baker-Campbell-Haussdorff formula is presented since it is used in the second section where the
definitions and main elementary results on universal enveloping algebras (such as the Poincaré-
Birkhoff-Witt theorem) are reviewed in details. Explicit formulas for the product are provided.
In the third section, the Casimir operators are introduced as convenient generators of the center
of the enveloping algebra. Eventually, in the fourth section the Coleman-Mandula theorem is
reviewed and discussed on Lie algebraic grounds, leading to a rough conjecture on the appearance
of enveloping algebras as physical higher symmetry algebras.1

1 Associative versus Lie algebras

An algebra A over a field K is a vector space over K endowed with a bilinear map ∗ : A×A → A
generally referred to as multiplication or product. The algebra is associative iff

x ∗ (y ∗ z) = (x ∗ y) ∗ z , ∀x, y, z ∈ A .

The algebra is unital if it possesses a unit element 1 such that

x ∗ 1 = x = 1 ∗ x.

Proposition 1.1. Let A be a unital associative algebra, of finite dimension over the field K.
Then A is isomorphic to a subalgebra of the algebra M (n ; K) of n × n matrices for some non-
negative integer n ∈ N.

The center of A is the subalgebra of elements that commute with all elements of A and is
denoted by
Z(A) = {z ∈ A | z ∗ a = a ∗ z , ∀a ∈ A} .
The centralizer of a subset S ⊂ A is the subalgebra of elements that commute with all elements
of S and is referred to as

CA (S) = {z ∈ A | z ∗ s = s ∗ z , ∀s ∈ S} .

If {ei } is a basis of A, then the product ∗ is completely determined by the structure

constants f ijk ∈ K defined by
ej ∗ ek = ei f ijk .

1 Slight additions to standard textbook material on the enveloping algebra topic are presented: An independent

heuristic proof of Berezin’s formula for the enveloping algebra composition law is provided, and a mean to evade
the negative conclusions of the no-go theorems on S-matrix symmetries is discussed, while mentioning higher-spin
algebras as a specific example.
216 Bekaert

For (anti)commutative algebras, the associativity is equivalent to the condition

f i j[k f j l]m = 0 .

1.1 Representations of algebras

An algebra homomorphism from an algebra A to an algebra B is a linear map Φ : A → B
such that Φ(x ∗ y) = Φ(x) ∗ Φ(y) for all x, y ∈ A.
A linear representation of the associative algebra A over the vector space V is an alge-
bra homomorphism from A to the associative algebra End(V ) of endomorphisms (i.e. linear
operators on V )
RV : A → End(V ) .
The vector space V carries the representation and is called a representation space of A
or, in more fancy terms, a (left) A-module. The kernel of RV is called the annihilator of the
A-module V and is denoted by Ann(V ). The image RV A) of the algebra A in End(V ) is called
the realization of the algebra A on the vector space V .
An invariant subspace is a vector subspace W ⊆ V such that RV (x)W ⊆ W for all x ∈ A.
In abstract jargon, it is also christened submodule or left ideal. An irreducible (or simple)
module V has only two distinct submodules, {0} and V itself.

1.2 Tensor algebras

The direct sum of all possible tensorial powers of V is
O ∞
M
(V ) ∼
= ⊗p V ,
p=0

where the first summand (p = 0) is the field K and the second (p = 1) is the space V itself. The
vector space ⊗(V ) endowed with a unital associative algebra structure via the tensor product
⊗, is called the tensor algebra of V .
Let K be a field. The free algebra on n indeterminates X1 , ..., Xn (the construction
works also for any countable set S of “indeterminates”), is the algebra spanned by all linear
combinations X
P (Xi ) = Πi1 ...ik Xi1 . . . Xik ,
k=0

of formal products of the generators Xi , with coefficients Πi1 ...ik ∈ K. This algebra is denoted
by K < Xi > and is said to be freely generated by the X’s.

Proposition 1.2. The free algebra on n indeterminates can be constructed as the tensor algebra
of an n-dimensional vector space. More precisely, if the set {ei } is a basis of a vector space V
over a field K, then we have the following isomorphism of algebras

⊗(V ) ∼
= K < ei > .

1.3 Presentation modulo relations

Because of the wide generality of the tensor algebra (that may be encoded in abstract terms in
its “universality” property), many other interesting algebras are constructed by starting with
Universal Enveloping Algebras 217

the tensor algebra and then imposing certain relations on the generators, i.e. by constructing
certain quotients of the tensor algebra.
A subalgebra I ⊆ A of an algebra A is a left (or right) ideal if A ∗ I ⊆ I (or I ∗ A ⊆ I).
Moreover, if I is both a left and a right ideal, then it is called invariant subalgebra or (two-
sided) ideal. In such case, one may define the quotient algebra A/I that is the algebra of
equivalence classes [a] defined by the equivalence relation a ∼ a + b where b ∈ I.
Exercise 1: Check that the product of classes is well defined.
For any representation RV : A → End(V ), the annihilator Ann(V ) is a (two-sided) ideal
made of all elements of A which are represented by the operator zero in End(V ).

Proposition 1.3. Let RV : A → End(V ) be a linear representation of the algebra A over the
space V .
Then the realization of the algebra A on the space V is isomorphic to the quotient of the
algebra A by the annihilator of the A-module V ,
A
RV (A) ∼
= .
Ann(V )

Let S be a set of generators and R ⊂ K < S > a subset of the free algebra generated by S. A
presentation of an algebra A by generators modulo relations is the definition of A as the
quotient of the free algebra K < S > by its smallest (two-sided) ideal containing all elements of
R. In more pragmatic terms, this procedure puts all elements of R to zero in A and the relations
R = 0 are thereby imposed on the corresponding product of generators in A.

1.4 Symmetric and exterior algebras

The symmetric algebra2 denoted by ⊙(V ) (or also S(V ) or ∨(V )) is the commutative associa-
tive algebra defined as the quotient of the tensor algebra ⊗(V ) by the smallest two-sided ideal
containing all elements of the form x ⊗ y − y ⊗ x. It is graded by the order p of the contravariant
tensors
K ∞
M
(V ) = ⊙p (V ) .
p=0

Another way to construct the symmetric algebra is by using the projector S : ⊗(V ) → ⊙(V ) on
the symmetric part of a contravariant tensor. The quotient of the tensor algebra ⊗(V ) by the
kernel of S is ⊙(V ) ∼
= ⊗(V )/Ker(S).
Let K be a field. The polynomial algebra on n indeterminates X1 , ..., Xn is the algebra
spanned by all linear combinations over K of products of the commuting variables Xi . This
algebra is denoted by K[Xi ].

Proposition 1.4 (Polynomial and symmetric algebra ∼ =). Let {ei } be a basis of an n-dimensional
space V over the field K. Then we have the following isomorphism of associative commutative
algebras
⊙(V ) ∼= K[ei ] .
Moreover, ⊙r (V ) is isomorphic to the space of homogeneous polynomials of order r in the vari-
r
ables ei and its dimension is equal to Cn+r−1 = (n+r−1)!
(n−1)! r ! .

2 The exterior algebra ∧(V ) of antisymmetric forms is constructed analogously to the symmetric algebra
∨(V ) as a quotient of the tensor algebra ⊗(V ) by the relations x ⊗ y + y ⊗ x.
218 Bekaert

Exercise 2: Verify that the correspondence is a bijection (hint: an arbitrary element p of ⊙r (V )

decomposes as p = pi1 ...ir ei1 ⊙ . . . ⊙ eir , where the components pi1 ...ir are completely symmetric)
and a commutative algebra homomorphism (hint: the product in ⊙(V ) reads in components
(p ⊙ q)i1 ...ir+s = p(i1 ...ir q ir+1 ...ir+s ) , where the curved bracket denotes complete symmetrization
with strength one).

1.5 Lie algebras

A Lie algebra is an algebra g whose product [ ] : g × g → g is called (Lie) bracket and obeys
the following properties:

(i) [X, Y ] = − [Y, X] for all X, Y ∈ g (antisymmetry)

(ii) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 for all X, Y, Z ∈ g (Jacobi identity).

A Lie algebra c is said to be Abelian if its Lie bracket vanishes identically, [ c , c ] = 0. The
central extension of an arbitrary Lie algebra g by an Abelian Lie algebra c is the Lie algebra
that is the direct sum g ⊕ c endowed with a Lie bracket such that [ g ⊕ c , c ] = 0 .

Proposition 1.5. A Lie algebra g is associative if and only if

 
X, [Y, Z] = 0 , for any X, Y, Z ∈ g

if and only if g is a central extension of an Abelian Lie algebra.

Exercise 3: Prove Proposition 1.5 by using the Jacobi identity and the definition of associativity
and central extension.
Example: The Heisenberg algebra hn is the Lie algebra of dimension 2n + 1 that is alge-
braically generated by the generators X i , Pj (i, j = 1, 2, . . . , n) and C which are subject to the
following Lie bracket relations

[Pj , X i ] = Cδji , [C, X i ] = 0 , [C, Pj ] = 0 ,

hence C is a central element. The Heisenberg algebra is an associative Lie algebra (it is actually
the elementary building block of any finite-dimensional non-Abelian associative Lie algebra).
From Proposition 1.5, one knows that, generally speaking, a Lie algebra is not associative.
Nevertheless, there is a canonical procedure to construct a Lie algebra out of an associative one.

Proposition 1.6. Any associative algebra A ∼ = (V , ∗ ) may be turned into a Lie algebra g ∼ =
(V , [ ] ) that has the same underlying vector space V , but whose multiplication operation [ , ] is
given by the commutator bracket

[x, y] := x ∗ y − y ∗ x .

It seems there is no name for this widely spread construction but I propose to call the obtained
Lie algebra g “commutator algebra associated to A” and to denote it by [A ].
A natural question arises: Does the converse of Proposition 1.6 hold? More accurately: Is it
possible to canonically construct an associative algebra out of any given Lie algebra? The answer
is positive and “universal”: the enveloping algebra does the job, which explains its usefulness.
Universal Enveloping Algebras 219

The commutator algebra is very useful for defining a linear representation of a Lie
algebra g acting on V (or Lie algebra module) as a Lie algebra homomorphism

RV : g → gl(V ) ,

where V is the representation space and gl(V ) := [End(V )]. Given two g-modules V and W ,
one can represent g also on the tensor product V ⊗ W by
 

(RV ⊗ RW )(X) (v ⊗ w) := RV (X)v) ⊗ w + v ⊗ RW (X)w ,

for all X ∈ g, v ∈ V and w ∈ W . The latter expression may be rewritten as

RV ⊗ RW = RV ⊗ 1W + 1V ⊗ RW .

1.6 Baker-Campbell-Haussdorff formula

Baker-Campbell-Hausdorff (BCH) type relations arise naturally when one describes Lie group
elements by exponentials of Lie algebra elements. The most crucial consequence of the BCH
formula is that it shows that the local structure of the Lie group G (the multiplication law for
elements near the identity) is completely determined by its Lie algebra g (where the Lie bracket
is the multiplication law).
Theorem 1.7 (Baker-Campbell-Haussdorff formula). For any two elements X, Y ∈ g with k X k
and k Y k “sufficiently small”,

Z1 Z1
X Y t adX adY



log(e e ) = Y + dt Ω e e (X) = X + dt Ψ eadX et adY (Y )
0 0

where ad is the adjoint representation defined by adX (Y ) = [X, Y ] and Ω is the function

log z
Ω(z) =
z−1
and Ψ(z) = zΩ(z).
Though entirely explicit, the genuine computation of the group multiplication law from the
Lie brackets is out of reach in most cases. Nevertheless, at first order it is possible to get more
illuminating expressions.
The left and right shift are the respective actions of the group G on itself defined by
λ(g) : h 7→ gh and ρ(g) : h 7→ hg. For all elements X, Y of a Lie algebra g, one may consider the
expansion of log(eX eY ) either in X or in Y :



log(eX eY ) = Y + LY (X) + OY k X k2 = X + RX (Y ) + OX k Y k2



where OY k X k2 and OX k Y k2 denote complicated functions of X and Y obeying



OY k X k2 kXk→ 0 OX k Y k2 kY k→ 0
−→ 0 , −→ 0 .
kXk kY k

The Lie algebra endomorphisms LY and RX are respectively called the left and right shift
operators since they correspond to the infinitesimal action of the Lie group element.
220 Bekaert

Corollary 1.8. The left and right shift operators are respectively given by

LY = B ( adY ) , RX = B (−adX ) ,

where B is the generating function of the Bernoulli numbers

∞
X
z 1
B(z) = = bm z m
exp z − 1 m=0 m!

the coefficients of the Taylor series of which are the Bernoulli numbers bm .
Even more explicitly, for a finite-dimensional Lie algebra with structure constants defined by

[Tj , Tk ] = f i jk Ti

the left shift matrix reads

X∞
bm j ℓm−2 ℓm−1
Lj i (Y ) = fk1 ℓ1 fkℓ21 ℓ2 . . . fkm−1 ℓm−1 fkm i Y
k1 k2
Y . . . Y km−1 Y km .
m=0
m!

Exercise 4: Prove
Corollary 1.8 by

expanding the BCH

formula of Theorem 1.7 at first order
(hint: Ω et adX eadY (X) = Ω eadY (X) + OY′ k X k2 )

2 Universal enveloping algebras

Mathematically speaking, the construction of the universal enveloping algebra of a Lie algebra is
useful in order to pass from a non-associative structure to a more familiar (associative) algebra
over the same field while preserving the representation theory.
The rather abstract definition is as follows (fortunately a more concrete definition is presented
in Theorem 2.1): A universal enveloping algebra U of a Lie algebra g over a field K is a
unital associative algebra U over K, together with a homomorphism i : g → [U] of Lie algebras,
such that if A is another associative algebra over K and φ : g → [A] is another Lie algebra
homomorphism, then there exists a unique homomorphism Ψ : U → A of associative algebras
inducing a Lie algebra homomorphism ψ : [U] → [A] such that the following diagram commutes:
i
g −→ [U]
φց ւψ .
[A]

The universal property above ensures that all universal enveloping algebras of g are isomorphic;
this justifies the standard notation U(g).

Theorem 2.1 (Canonical universal enveloping algebra). Any Lie algebra g has a universal
enveloping algebra:
Let ⊗(g) be the tensor algebra of g and let I be the smallest two-sided ideal of ⊗(g) containing
all elements of the form
x ⊗ y − y ⊗ x − [x, y] ,
for x, y ∈ g. Then the quotient ⊗(g)/I is a universal enveloping algebra of g, i.e.

U(g) ∼
= ⊗(g)/I .
Universal Enveloping Algebras 221

Example: If g is Abelian, then this construction gives the symmetric algebra ⊙(g).
In the finite-dimensional case, combining Theorem 2.1 with Proposition 1.2 leads to a con-
struction more familiar to physicists.
Corollary 2.2. Let {Ti } be a basis of the finite-dimensional Lie algebra g over the field K.
The universal algebra U(g) is isomorphic to a quotient of the free algebra K < Ti > and the
elements of U(g) may be thought as formal polynomials P (Ti ) in the generators of g. More pre-
cisely, the universal algebra U(g) may be presented by the generators Ti modulo the “commutation
relations” of the Lie algebra g
Tj Tk − Tk Tj = [Tj , Tk ] .
Example: The Weyl algebra An is the universal enveloping algebra of the Heinsenberg algebra
hn , that is An ∼
= U(hn ). The Weyl algebra is isomorphic to the algebra of operators polynomial
in the positions and momenta (i.e. textbook quantum mechanics) of which only the associative
algebra structure is retained (in other words, An does not know about Hilbert space, hermiticity
properties, etc).
By definition a representation RV : g → gl(V ) is a Lie algebra homomorphism. The universal
property implies the existence of an associative algebra homomorphism RV : U(g) → End(V )
which is nothing else than a representation of U(g) acting on V . Since the map RV is unique,
the representations of the Lie algebra g and of the associative algebra U(g) are in one-to-one
correspondence.

The realization of the universal enveloping algebra in End(V ), i.e. the image
RV U(g) , will be called here the enveloping algebra of the realization of g on the module V ,
and denoted by Env(g ; V ).

2.1 Poincaré-Birkhoff-Witt theorem

Theorem 2.3 (Poincaré-Birkhoff-Witt). Consider any countable basis {Ti } of the Lie algebra
g with elements Ti indexed by a subset I ⊆ N0 .
The set of all lexicographically ordered monomials of U(g)

Ti1 Ti2 . . . Tik (i1 ≤ i2 ≤ . . . ≤ ik−1 ≤ ik ; k ∈ N)

is a basis of U(g).
Remarks:
By the Poincaré-Birkhoff-Witt (PBW) theorem, the map i (in the abstract definition) is injec-
tive. Usually g is identified with i(g) since i is the canonical embedding.
The PBW theorem is useful in representation theory because it determines a proper basis of
Verma modules that are roughly spaces of the form U(g− )| h.w.i, where g− is the space of lowering
operators and | h.w.i is a highest weight state (see F. Dolan’s lectures for more comments).
Corollary 2.4. Let {Ti } be a countable basis of the Lie algebra g. The set of all distinct Weyl-
ordered homogeneous polynomials

Ti1 Ti2 . . . Tik + all permutations ,

is a basis of U(g). Therefore, the universal enveloping algebra U(g) and the symmetric algebra
⊙(g) are isomorphic as vector spaces.
Exercise 5: Prove Corollary 2.4 by showing that the change of basis passing from lexicographic
to Weyl ordering is triangular.
222 Bekaert

In order to illustrate Corollary 2.4, let us introduce a variable ti for each generator Ti of
g. Using the isomorphism of Proposition 1.4, we identify the symmetric algebra ⊙(g) with the
polynomial algebra K[ti ]. It is convenient to use the set of all distinct completely symmetric
products of the variables ti as a basis of K[ti ]. In such case any polynomial reads
X
P (ti ) = Πi1 ...ik ti1 . . . tik
k=0

where the coefficients Πi1 ...ik ∈ K are symmetric over all contravariant indices. Consequently,
to any polynomial P (ti ) ∈ K[ti ] corresponds a unique Weyl-ordered polynomial PW (Ti ) ∈ K <
Ti >, given by X
PW (Ti ) = Πi1 ...ik Ti1 . . . Tik ,
k=0

and conversely. A Weyl-ordered polynomial in the generators is called a Weyl polynomial.

The symbol of the Weyl polynomial PW (Ti ) ∈ K < Ti > is simply defined as the polynomial
PW (ti ) ∈ K[ti ] obtained by replacing each generator with its counterpart ti .
Remarks
The symbol of a Weyl polynomial is not necessarily written in Weyl-ordered form since it is an
ordinary polynomial in the polynomial algebra K[ti ] where the variables ti commute.
For complex algebras, a useful – though somewhat formal – alternative bijection is given by
the “double Fourier transform” formula
Z
PW (Ti ) = dn p Pe(pj ) exp(i pi Ti ) ,

where Pe is the formal3 Fourier transform of the polynomial P

Z
e j 1
P (p ) = dn t P (ti ) exp(−i pi ti ) .
(2π)n

By construction, for any (not necessarily written in Weyl-ordered form) polynomial P (ti ) the re-
sulting polynomial PW (Ti ) is Weyl-ordered because the power series expansion of the exponential
does the job.

2.2 Weyl map and star product

The bijection between symbols and Weyl polynomials is extremely straightforward at the vector
space level (it is even a mere exchange of arguments in the Weyl-ordered bases). However, the
non-commutativity introduces some complications since in general the product of two Weyl-
ordered polynomials is not properly ordered any more. Indeed, cumbersome reordering of the
generators Ti should be performed before getting a Weyl polynomial. Moreover, the latter does
not correspond to the naive (commutative) product of symbols. In order to take into account
the reordering performed in U(g) one should look for its counterpart in K[ti ], i.e. deform its
commutative product.
The associative (but in general non-commutative) algebra of symbols ⋆(g) is defined as
3 Actually, the Fourier transform of a polynomial should require some care because it is neither a polynomial

nor a power series in x but a distribution. I will not enter in these issues and consider this as a formal definition
of PW which, in any case, is not problematic because one plays with polynomials only (at the beginning and the
end of the procedure).
Universal Enveloping Algebras 223

the space K[ti ] of polynomials endowed with a product ⋆ such that the Weyl map between the
space of symbols onto the universal enveloping algebra

W : ⋆(g) → U(g) : P 7→ W [P ] := PW

is an associative algebra isomorphism:

W [P ⋆ Q] = W [P ] W [Q] , ∀ P, Q ∈ K[ti ] .

The product in U(g) is determined from the Lie algebra bracket, hence the right-hand-side is
known. The crucial point is that the polynomial in the generators W [P ] W [Q] should be Weyl-
ordered in order to determine the left-hand-side and get the symbol P ⋆ Q. This procedure
uniquely defines the product ⋆ . More explicit formulas of the latter are presented in the next
subsection.
Remark Since the universal enveloping algebra U(g) is unique, up to isomorphisms, it is clear
that the algebra ⋆(g) of symbols is simply another realization of the same algebra. Nevertheless,
the ⋆-product is quite useful from conceptual (←֓ deformation quantization) and computational
(←֓ Weyl calculus) points of view. The ⋆ -product is indeed a bidifferential operator that may
be expanded as a formal power series in the structure constants.

2.3 Weyl calculus

The “composition formula” of the universal enveloping algebra has been written explicitly first
by Berezin. It has been subject to many generalizations afterwards by other mathematicians.
The explicit formulae for the star product are the cornerstone of the Weyl calculus and found
direct application in physics for field theories on non-commutative space-times with constant or
linear commutation relations.
On the one hand, the Berezin formula implements the left multiplication by a symbol P (i.e.
the endomorphism P ⋆ ) as a differential operator. On the other hand, the BCH product is an
explicit formula for the ⋆ -product acting as a bidifferential operator. The main ideas of their
proofs is to use the “double Fourier transform” formula of Subsection 2.1 and to interpret the
exponentials of Lie algebra elements (in the Fourier transformations) as Lie group elements.

Theorem 2.5 (Berezin’s formula). Let g be a complex Lie algebra of dimension n ∈ N with
structure constants defined by

[Tj , Tk ] = i f i jk Ti (i, j, k = 1, 2, . . . , n) .

The left multiplication by a given symbol P (ti ) is the differential operator

 ∂

P (ti ) ⋆ = PW tj Lj i ,
∂t
where Lj i (Y ) are the components of the left shift matrix LY ∈ End(g).

In Appendix A, a proof of Theorem 2.5 is presented, which uses only the corollary 1.8.
The Berezin formula is of interest because it shows that it is sufficient to know the BCH
formula at first order in order to determine the left ⋆ -multiplication. Nevertheless, if the BCH
formula is known at all orders, then the following formula is much more convenient (because it
does not involve Weyl symmetrization of one factor and it puts both factors on equal footing).
224 Bekaert

Theorem 2.6 (Baker-Campbell-Haussdorff’s product). The product of two polynomial symbols

of ⋆(g) is a bidifferential operator acting as
 ∂ ∂
 

P (t) ⋆ Q(t) = exp ti mi , P (u) Q(v)
∂u ∂v u=v=t

where the functions mi are the components of

m(X, Y ) = log(eX eY ) − X − Y ,

in some basis {Ti }.

The proof is a straightforward application of the BCH formula together with a couple of Fourier
transformations and integration by parts. It is given in Appendix B.
Example: Theorem 2.6 applied to the Lie brackets of the Heisenberg algebra hn leads to
the well-known Moyal product in the Weyl algebra An (see S. Cnockaert’s lectures for more
details).

3 Casimir operators
3.1 Intertwiners
In representation theory, it is useful to find maps between modules which commute with the
action of the algebra. Let A be an algebra, let RV and RW be two representations of A acting
respectively on the spaces V an W . A linear map φ : V → W is called a intertwiner of
representations if
φ ◦ RV (x) = RW (x) ◦ φ ∀x ∈ A.
If V = W then an intertwiner φ : V → V is called a self-intertwiner.

Theorem 3.1 (Schur’s lemma). Let φ ∈ End(V ) be a self-intertwiner of a finite-dimensional

irreducible complex module V . Then φ = λI, for some λ ∈ C. The space of self-intertwiners of
a finite-dimensional irreducible complex module is isomorphic to C.

Corollary 3.2. Let R be a finite-dimensional irreducible representation of a complex algebra A.

If z is in the center of A, then R(z) = λI, with λ ∈ C.

Proof: The central elements z ∈ Z(A) are represented as self-intertwiners φ := R(z) ∈

End(V ) because R is a Lie algebra homomorphism.

Lie algebra representations are in one-to-one correspondence with their universal enveloping
algebra representations. It is therefore useful to know the basis elements of the center of U(g)
since they act as multiple of the identity on irreducible complex representation spaces and their
corresponding complex values may characterize the irreducible module V .

3.2 Quadratic and higher order Casimir operator

Lemma 3.3. The center Z U(g) of the universal enveloping algebra is equal to the centralizer
CU (g) (g) of the (canonically embedded) Lie algebra g.
Universal Enveloping Algebras 225

Proof: Of course, any element of the center of the universal enveloping algebra must com-
mute with all elements of any subset of U(g), e.g. the canonically embedded Lie algebra g.
Moreover, by the PBW theorem any element that commutes with a basis of g commutes
with any basis of U(g).

Theorem 3.4 (Gel’fand). Let {Ti } be any basis of a finite-dimensional Lie algebra g with Lie
brackets [Tj , Tk ] = f i jk Ti .

A Weyl-ordered polynomial PW (T ) belongs to the center Z U(g) of the universal enveloping
∈ ⊙(g) is invariant under the adjoint
algebra if and only if the corresponding symmetric tensor Π P
action. More explicitly, the Weyl polynomial PW (Ti ) = Πi1 ...in Ti1 . . . Tin (with completely
symmetric contravariant components) commutes with all elements of U(g) iff

0 = f i1 kℓ Πℓi2 ...in + . . . + f in kℓ Πi1 ...in−1 ℓ .

Exercise 6: Prove Gel’fand’s Theorem by only verifying that PW (T ) commutes with any basis
element Tk of g.
The Casimir operators
of a finite-dimensional semisimple Lie algebra g are a distinguished
basis of the center Z U(g) of the universal enveloping algebra made of homogeneous polynomials

Ck = di1 ...ik Ti1 . . . Tik

with di1 ...ik suitable symmetric invariant tensors of the adjoint representation. The degree k of
this homogeneous polynomial is called the order of Ck .
Example: The inverse matrix of the Killing form κij defines the quadratic Casimir operator
equal to
C2 := κij Ti Tj .
The quadratic Casimir C2 operator belongs to the center of U(g) since the Killing form κij is
invariant under the adjoint action.
Theorem 3.5. Let g be
a finite-dimensional semisimple Lie algebra of rank r.
The center Z U(g) of the universal enveloping algebra is isomorphic to the polynomial al-
gebra K[C (i) ] over the base field in r variables C (i) (i = 1, 2, . . . , r). Therefore, the number of
algebraically independent Casimir operators is equal to the rank.
For applications in physics, the Casimir operators are sometimes more useful than the r
Dynkin labels (of which they are non-linear functions) because they more often correspond to
physical quantities (such as the square of the momentum or the Pauli-Lubanski vector).

4 Symmetries of the S-matrix

4.1 Rudiments of scattering theory
A proper beginning to address the symmetries of the S-matrix is a brief review of some funda-
mental definitions of scattering theory.
To start up with first quantization, Wigner showed that the rules of quantum mechanics,
combined with the principle of special relativity, imply that the classification of all possible
wave equations K| ψi = 0 describing the evolution of the states | ψi ∈ H of a free relativistic
particle moving in the Minkowski space Rd−1,1 is equivalent to the classification of all unitary
irreducible representations of the Poincaré algebra iso(d − 1, 1). Moreover, Wigner proved that
226 Bekaert

unitary irreducible iso(d − 1, 1)-modules are labeled by the mass-square m2 (a continuous real
parameter), the “spin” degrees of freedom s (for finite component representations, these are a
finite set of non-negative (half) integers). Therefore, the one-particle Hilbert space H(1) ⊂ H
of solutions decomposes into a direct sum of unitary irreducible iso(d − 1, 1)-modules
M
H(1) = H(m2 , s, i)
m2 , s, i

where the index i labels the particle type. A standard basis is made of plane-wave states | p, s, i i
where pµ is the momentum.
In second quantization, the Hilbert space of scattering theory is the direct sum
∞
M
In := H(n) ,
n=1

of the n-particle Hilbert spaces H(n) that are subspaces of n-fold tensor products ⊗n H(1) ,
determined by the generalized exclusion principle (for instance, in the case without fermions
H(n) = ⊙n H(1) ). The dual of the in-going particle space In is called the out-going particle
space and will be denoted by Out(:= In∗ ). The S-matrix is a unitary operator on In, i.e.
S ∈ U (In). A scattering amplitude is a complex number given by hout| S|ini for some elements
|ini ∈ In and hout| ∈ Out.

4.2 Higher symmetries

Since the kinetic operator K is Hermitian, the linear wave equation K| ψi = 0 comes from an
action principle: S[ψ] = hψ| K| ψi. A local symmetry transformation of the free action is
a unitary operator U on H that commutes with K and acts locally, in the sense that hx, s, i| U| ψi
can be written in the form hx′ , s′ , i′ |ψi where x′ = f (x) is a smooth change of coordinates. The
group of all such transformations is said to be the local symmetry group of the free action.
A symmetry transformation of the S-matrix is a unitary operator on In that commutes
with S (so that it leaves scattering amplitudes invariant), turns one-particle states into one-
particle states and acts on many-particle states as tensor products of one-particle states. The
group G of such transformations is said to be the symmetry group of the S-matrix. The
S-matrix is said to be Poincaré invariant if G ⊇ ISO(d − 1, 1).
It turns out to be convenient to translate the previous definitions into Lie algebraic language:
the (Lie) algebra g corresponds to the symmetry group G and a
• local symmetry generator of the free action is a Hermitian differential operator H
on H that
(a) commutes with the kinetic operator, HK = KH,
(b) is at most linear in the momentum operator Pa .
• symmetry generator of the S-matrix is a Hermitian operator H on In that
(i) commutes with the S-matrix, HS = SH,
(ii) turns one-particle states into one-particle states, HH(1) ⊆ H(1) ,
(iii) acts additively on many-particle states (roughly, “like” the Leibnitz rule). More
precisely, it acts on multiparticle spaces H(n) as the nth tensor product module of
H(1) (see Section 1.5 for the definition).
Universal Enveloping Algebras 227

As shown by J.T. Lopuszanski and others, the property (iii) necessarily follows from the
locality of the symmetry transformations (together with some natural asymptotic condi-
tions).
Contemporary theoretical physics has frequently questioned the ansatz of locality and it
has become common to relax locality (at tiny scales and in a controlled way!) in order to
build new physical models (e.g. extended objects, non-commutative space-time, etc). With
such considerations in mind, it might be interesting to speculate on the relaxation of the last
conditions, which are the less natural. Therefore one may extend the previous definition to
higher(-derivative) symmetry generators of the free action for which the condition (b)
is relaxed, and higher symmetry generators of the S-matrix for which the last condition
(iii) is removed. They are allowed to be general differential operators (i.e. smooth functions of
the momentum operator Pa ). More comments (and contact with enveloping algebras) will be
made in Section 4.4.

4.3 No-go theorem

Most frequently, the Coleman-Mandula theorem is interpreted as the destruction of any reason-
able hope for a fusion between internal and space-time symmetries. (An internal symmetry
of the S-matrix is one that commutes with Poincaré transformations. This implies that it acts
on particle-type indices only.) But another far-reaching consequence of the no-go theorem is the
assertion that the maximal space-time symmetry algebra of relativistic quantum field theory in
flat space-time is the (super)conformal group.
Theorem 4.1 (Coleman-Mandula). Let g be a Lie algebra of symmetries of the matrix S and
let the following four conditions hold:
(i) (Poincaré invariance) g ⊇ iso(d − 1, 1),
(ii) (Particle-finiteness) All unitary iso(d − 1, 1)-modules H(m2 , s, i) are finite-dimensional for
fixed mass-square. Moreover, for any finite mass M , there are only a finite number of
particle types with mass less than M ,
(iii) (Weak elastic analyticity) Elastic-scattering amplitudes are analytic functions of the scat-
tering angle at almost all energies and angles,
(iv) (Occurence of scattering) Let |pi and |p′ i be any two one-particle momentum eigenstates of
H(1) . If |p, p′ i denotes the two-particle state of H(2) made from these, then S|p, p′ i =
6 |p, p′ i
at almost all energies.
Then, the S-matrix symmetry algebra g is isomorphic to the direct sum

g∼
= ginternal ⊕ gspace−time

of an internal symmetry Lie algebra ginternal and a space-time symmetry Lie algebra gspace−time .
Moreover, the space-time symmetry algebra gspace−time can be the conformal algebra so(d, 2) only
if all particles are massless. In all other cases, it is the Poincaré algebra iso(d − 1, 1).
Remarks:
The S-matrix symmetry algebra g is not assumed to be finite-dimensional.
The extension of the theorem to the massless and supersymmetric case is actually due to Haag,
Lopuszanski and Sohnius, who obtained the (super)conformal group as maximal possibility for
gspace−time .
228 Bekaert

For almost each hypothesis of the theorem, there exists a famous counter-example to the
conclusions of Coleman-Mandula theorem which violates the corresponding assumption. For
instance,

1. There are examples of Galilean-invariant models (such as the SU (6)-invariant model of non-
relativistic quarks) which unite multiplets of different spin and thus contradict the conclu-
sions of the theorem.

2. Tensionless strings violate the second assumption since the spectrum contains an infinite set
of massless particles.

3. Symmetry groups are known which are not direct products, but which do allow scattering,
although only in the forward and backward directions.

4. If the S-matrix is the identity, the underlying field theory is free and the symmetry algebra
may be enhanced. Notice that the free case is still of interest if the theory is conformal4 ,
as was argued by Sundborg and Witten in 2000, because in the tensionless limit the
AdS/CFT correspondence should relate an interacting theory in the bulk to a free theory
on the boundary.

0. The most celebrated counter-example to Coleman-Mandula’s theorem is supersymmetry

which violates the initial “hidden” assumption that the algebra is a Lie algebra instead of
the more general possibility of a Lie superalgebra.

The no-go theorem has direct application only for symmetries of the S-matrix, i.e. “physical”
or “visible” symmetries. In other words, gauge symmetries are beyond the scope of the no-go
theorem. Nevertheless, local symmetry groups contain rigid symmetries as a subgroup. As a
conclusion, the theorem restricts the possibilities of reasonable candidates for rigid symmetries
to be gauged. Indeed the standard model corresponds to the gauging of an internal symmetry
group while (super)gravity roughly corresponds to the gauging of the (super)Poincaré group.
As a corollary, the S-matrix no-go theorems impose some constraints on the spin of the gauge
fields associated with the gauged symmetries. Indeed, gauging a symmetry group G requires the
introduction of a connection taking values in the symmetry algebra g. On the one hand, when
the algebra is internal the generators Ti do not carry any space-time index, hence the gauge field
is a vector gauge field Aiµ . On the other hand, for a space-time symmetry algebra the generators
do carry space-time indices and the Poincaré generator Pa is associated to the vielbein eaµ which
in turn leads to a spin-two field: the metric gµν (correspondingly, the supersymmetry generator
leads to a Rarita-Schwinger field of spin three-half). The restriction on the space-time symmetry
algebra therefore rules out gauge fields of higher (i.e. greater than two) spin.

4.4 Yes-go conjecture

To evade the conclusions of the S-matrix no-go theorems one should remove (at least) one of its
assumptions. Symmetries mixing particles with different spin were already obtained by intro-
ducing graded symmetry generators: the supersymmetry generators are somehow “squareroots”
of the translation generators. Thus, it is tempting to try the reverse procedure by introducing
new generators that are “powers” Pa1 . . . Pam of the translation generators. The corresponding
connections eµa1 ... am may lead to a symmetric tensor gauge fields gµ µ1 ... µm of rank (“spin”) equal
to m + 1. Including Lorentz generators in the game naturally leads to the enveloping algebra
4 Strictly speaking, for conformal field theories the S-matrix does not exist (since asymptotic states are not
well defined because of the scale invariance) and has to be replaced by correlation functions.
Universal Enveloping Algebras 229

of the space-time symmetry algebra! The main point is that such powers of the generators are
no more standard symmetries but rather higher symmetries (defined at the end of Section 4.1)
generating space-time symmetry transformations which are non-local.
Exercise 7: Show that Weyl-ordered polynomials of Hermitian operators satisfying the prop-
erties or (i)-(iii) are Hermitian and do obey the properties (i)-(ii) but violate5 the axiom (iii).
Idem with the properties (a)-(b).
Therefore, one may propose the following rather general way to circumvent usual no-go
theorems on maximal symmetry groups:

⋄ Given a symmetry group G ⊂ U (H) of unitary operators acting on some Hilbert space H, one
first builds the enveloping algebra Env(g ; H) ⊂ u(H) of the realization of g as Hermitian
operators acting on H. By Proposition 1.3, this associative algebra Env(g ; H) is isomorphic
to the quotient U(g)/Ann(H) of the universal enveloping algebra of the symmetry algebra
g by the annihilator of the U(g)-module H.

⋄ Secondly, one makes use of Proposition 1.6 to define its commutator algebra6 [Env(g ; H)]
that I propose to call higher-symmetry algebra of the g-module H and to denote by
hs(g ; H).

⋄ Thirdly, since the latter algebra is a Lie algebra one may exponentiate it (at least formally)
to get an higher-symmetry group HS(G ; H) of the group G.

G ֒→ HS(G ; H)
log ↓ ↑ exp
g ֒→ hs(g ; H)
iց ր[]
Env(g ; H)

To conclude this more speculative section, one may summarize the previous considerations
in the following loose conjecture:
If one allows higher symmetries, then there exist counter-examples to the conclusions of symme-
try no-go theorems, and the “gauging” of higher-space-time symmetry algebras hs(gspace−time ; H)
should correspond to some higher-spin gauge theories.

Remarks:
Actually, these speculations are a mere retrospective viewpoint on the higher-spin gauge theories
introduced by M. A. Vasiliev where the gauge fields take values in the higher-symmetry algebra
of the defining representation of the anti de Sitter isometry algebra o(d−1, 2) (i.e. the enveloping
algebra of AdSd Killing tensors). Nevertheless, the present perspective underlines how general
such constructions might be and suggests that the rigid counterpart of the gauge symmetries are
actually realized as genuine space-time symmetries of the corresponding theories.
If these rough speculations are correct, then an interesting issue is whether higher-spin symme-
tries are either realized on quadratic actions and satisfy the Leibnitz rule, or violate the Leibnitz
5 This last property is intimately related to the fact that the universal enveloping algebra has a canonical Hopf

algebra structure (used in the construction of the non-commutative theory of gravity presented by F. Meyer in
his talk). In this context, the property (iii) is christened “primitivity” and it is known that the space of primitive
elements of U (g) is precisely g.
6 Commutator of enveloping algebras were first considered in physics by E. S. Fradkin and V. Y. Linetsky in

1990.
230 Bekaert

rule and be realized on non-linear matter field theories (i.e. be free action or interacting S-matrix
higher-symmetries).

Acknowledgments
I thank N. Boulanger, V.A. Dolgushev, D. Fiorenza, F. Meyer and J.-H. Park for useful discus-
sions.

Appendices

A Proof of Berezin’s formula

We decompose the proof into some lemmas that we glue together at the end.

Lemma A.1. Let X

P (ti ) = Πi1 ...ik ti1 . . . tik
k=0

be a symbol of K[ti ] written with coefficients Πi1 ...in ∈ K which are symmetric over all indices.
The stared symbol defined as
X
P ⋆ (ti ) := Πi1 ...ik ti1 ⋆ . . . ⋆ tik
k=0

is equal to the original symbol: P ⋆ (ti ) = P (ti ) in ⋆(g).

Proof: Indeed, their Weyl polynomials are equal, W [P ∗ ] = W [P ]. More explicitly,

X
W [P ∗ (t)] = Πi1 ...ik W [ti1 ⋆ . . . ⋆ tik ]
k=0
X
= Πi1 ...ik W [ti1 ] . . . W [tik ] = PW (T ) ,
k=0

by linearity, the algebra homomorphism property and the definition of the Weyl map,
since W (ti ) = Ti .

Lemma A.2. The product of two polynomial symbols of ⋆(g) is equal to

P (t) ⋆ Q(t) = PW (t⋆) Q(t) ,

where the right-hand-side should be interpreted as the action of the linear operator PW (t⋆) on the
polynomial Q(t). The linear operator PW (t⋆) is defined as the Weyl-ordered polynomial function
of the elementary operators of left multiplication ti ⋆ that act as ti ⋆ Q(t).
Universal Enveloping Algebras 231

Proof: From Lemma A.1, one gets the first line of the chain of equalities
X 
P (t) ⋆ Q(t) = Πi1 ...ik ti1 ⋆ . . . ⋆ tik ⋆ Q(t)
X
= Πi1 ...ik (ti1 ⋆ . . . ⋆ tik ) ⋆ Q(t)
X     
= Πi1 ...ik ti1 ⋆ . . . ⋆ tik ⋆ Q(t) . . .
X
= Πi1 ...ik (ti1 ⋆) . . . (tik ⋆) Q(t)
= PW (t⋆) Q(t) ,

where we used the linearity and associativity properties of the star product ⋆ of symbols
and of the composition ◦ of operators.

As a consequence of Lemma A.2, in order to compute the product of any symbols, it is

sufficient to know explicitly the action of the elementary operators of left multiplication ti ⋆.
Actually, the latter result was the formula obtained originally by Berezin:

Lemma A.3. The elementary operators of left multiplication in ⋆(g) are explicitly given by
∂
ti ⋆ = tj Lj i
∂t
where Lj i (Y ) are the component of the matrix of the left shift operator L(Y ) = B(adY ).

Proof: The computation is performed in terms of Weyl polynomials, that is one evaluate
the product of the two Weyl polynomials
Z
∂  
Ti = i dn u i δ(u) exp(i uj Tj )
∂u
and Z
QW (T ) = e
dn v Q(v) exp(i v j Tj ) ,

where Qe is the Fourier transform of the polynomial Q. By definition of the left-shift

operator, one knows that


exp(i uj Tj ) exp(i v j Tj ) = exp i v j + ui Lj i ( i v ) + O(u2 ) Tj
 ∂ 
= 1 + ui Lj i ( i v ) j + O(u2 ) exp(i v j Tj )
∂v
Therefore, the product of the Weyl polynomials is equal to

Ti QW (T ) =
Z
∂  
e
 ∂ 
= i dn u dn v i δ(u) Q(v) 1 + uk Lj k ( i v ) j + O(u2 ) exp(i v j Tj )
∂u ∂v
Z
e ∂
= −i dn v Q(v) Lj i ( i v ) j exp(i v j Tj )
∂v
Z  
∂ e
= i dn v Q(v) L j
i ( i v ) exp(i v j Tj ) .
∂v j
232 Bekaert

Hence, the corresponding symbol is by definition

Z
∂ e 
ti ⋆ Q(t) = i dn v j
Q(v) Lj i ( i v ) exp(i v j tj )
∂v
Z
= tj dn v Q(v)e Lj i ( i v ) exp(i v j tj )
 ∂ Z
= tj Lj i e
dn v Q(v) exp(i v j tj )
∂t
 ∂ 
= tj Lj i Q(t) .
∂t

B Proof of BCH’s product

The computation is performed in terms of Weyl polynomials, that is one evaluate the product
of the two Weyl polynomials
Z
PW (T ) = dn r Pe(r) exp(i rj Tj ) ,

and Z
QW (T ) = e
dn s Q(s) exp(i sj Tj ) ,

where Pe and Q
e are the Fourier transform of the respective polynomials P and Q. From Theorem
1.7, one knows that
h
i
exp(i rj Tj ) exp(i sj Tj ) = exp i (rj + sj ) + mj ( ir , is ) Tj ,

where the functions mj are defined in Theorem 2.6. Therefore, the product of the Weyl polyno-
mials is equal to
Z h
i
PW (T ) QW (T ) = dn r dn s Pe(r) Q(s)
e exp i (rj + sj ) + mj ( ir , is ) Tj ,
Universal Enveloping Algebras 233

and the corresponding symbol is by definition the star product

P (t) ⋆ Q(t) =
Z



= dn r dn s Pe(r) Q(s)
e exp i (rj + sj ) tj exp tj mj ( ir , is )
Z
1
= dn r dn s dn u′ dn v ′ P (u′ ) Q(v ′ ) ×
(2π)2n




exp − i rj (u′j − tj ) exp − i sj (vj′ − tj ) exp tj mj ( ir , is )
Z
1
= dn r dn s dn u dn v P (t + u) Q(t + v) ×
(2π)2n



exp tj mj ( ir , is ) exp − i (rj uj + sj vj )
Z
1
= dn r dn s dn u dn v P (t + u) Q(t + v) ×
(2π)2n
 ∂ ∂ 

exp tj mj − ,− ) exp − i (rj uj + sj vj )
∂u ∂v
where we performed the changes of integration variables uj := u′j − tj and vj := vj′ − tj to obtain
the fourth line. Integrating by part, one gets

P (t) ⋆ Q(t) =
Z
1

= 2n
dn r dn s dn u dn v exp − i (rj uj + sj vj ) ×
(2π)
 ∂ ∂

exp tj mj , P (t + u) Q(t + v)
∂u ∂v
which achieves the proof of Theorem 2.6 since the integration over r and s provides Dirac’s
deltas over u and v.
234 Bekaert

Bibliography

Universal enveloping algebras

J. Fuchs and C. Schweigert, Symmetries, Lie algebras and representations, Cambridge Mono-
graphs on Mathematical Physics (Cambridge University Press, 1997) chapter 14.

N. Bourbaki, Elements of mathematics: Lie groups and Lie algebras (Hermann, 1975) part
I, chapter 1.

Weyl calculus

M. V. Karasev and V. P. Maslov, Nonlinear Poisson brackets: geometry and quantization, Math-
ematical Monographs 119 (American Mathematical Society, 1993) appendix I.

S. Waldmann, “Deformation quantization: observable algebras, states and representation the-

ory,” hep-th/0303080, section 1.

F. A. Berezin, “Some remarks about the associated envelope of a Lie algebra,” Funct. Anal.
Appl. 1 (1967) 91.

Intertwiners and Casimir operators

J. Fuchs and C. Schweigert, Symmetries, Lie algebras and representations, Cambridge Mono-
graphs on Mathematical Physics (Cambridge University Press, 1997) chapters 5 & 17.

A. O. Barut and R. Raczka, Theory of group representations and applications (Polish Scien-
tific Publishers, 1977) chapter 9.

S-matrix no-go theorems

S. Weinberg, The quantum theory of fields: Supersymmetry, volume III (Cambridge University
Press, 2000) chapter 24.

S. R. Coleman and J. Mandula, “All possible symmetries of the S matrix,” Phys. Rev. 159
(1967) 1251.

R. Haag, J. T. Lopuszanski and M. Sohnius, “All possible generators of supersymmetries of

the S matrix,” Nucl. Phys. B 88 (1975) 257.

Higher symmetries

M. G. Eastwood, “Higher symmetries of the Laplacian,” hep-th/0206233.

M. A. Vasiliev, “Higher spin gauge theories in any dimension,” Comptes Rendus Physique 5
(2004) 1101 [hep-th/0409260].
 β function and asymptotic freedom in QCD

Glenn Barnich∗

Physique Théorique et Mathématique

Université Libre de Bruxelles
and
International Solvay Institutes
Campus Plaine C.P. 231, B-1050 Bruxelles
Belgique

Abstract. Prerequistes for the lectures were a first course on quantum field theory
including the path integral representation for Green’s functions. The aim of the course
was a self-contained computation of the 1-loop β in QCD and the interpretation of this
result in terms of the high energy behavior of quarks. These lecture notes highlight
the main results and serve as a guide to standard literature where all the intermediate
steps of the computations can be found.

Based on the lectures presented by G. Barnich at the First Modave Summer School in Math-
ematical Physics held at Modave, Belgium on June 19-25, 2005

∗ Maı̂tre de Recherches du Fonds National Belge de la Recherche Scientifique

236 Barnich

1 Effective action at one loop

1.1 Legendre transform

The generating functional for Green’s function Z(J) admits the following path integral repre-
sentation
Z
i
Z(J) = Dφ exp (S[φ] + J · φ), (1.1)
~
R
where J · φ ≡ dn xJA (x)φA (x) with JA (x) external sources and φA (x) the quantum fields. For
simplicity, we consider in the first two sections the case of a single real scalar field φ with action
Z
1 1
S = dd x[− ∂µ φ∂ µ φ − m2 φ2 − V (φ) + iǫ terms]. (1.2)
2 2
R R R
If S = S0 +S1 with S0 = − 21 dd x dd y D(x, y)φ(x)φ(y), and S1 = − dd x V [φ], the iǫ terms im-
plement the boundary conditions and guarantee that the propagator, the inverse of the quadratic
part of the action, coincides up to normalization with the two point function of the free theory,
~ −1 b φ(y)|0
b
D (x, y) =< 0|T φ(x) >. (1.3)
i

After introducing W [J] defined by

~ Z[J]
exp W [J] = , (1.4)
i Z[0]

which can be shown to be the generating functional of connected Green’s functions, the effective
action is defined as the Legendre transform of W [J] with respect to J:
δW
φ= ⇐⇒ J = J[φ], Γ[φ] = (W [J] − J · φ)|J[φ] . (1.5)
δJ

1.2 Semi-classical expansion

Expanding the path integral around the unique classical solution φ0 [J] of the equation
δS
= −J, (1.6)
δφ

one finds in a first step

~  ′′

W [J] = S[φ0 [J]] + J · φ0 [J] − tr ln δ 4 (x, y) + D−1 (x, y)V [φ[J]0 (y)] + O(~2 ). (1.7)
2i
After performing the Legendre transformation, one gets Γ[φ] = S[φ] + ~Γ(1) [φ] + O(~2 ), where
1  ′′

Γ(1) [φ] = − tr ln δ 4 (x, y) + D−1 (x, y)V [φ(y)] . (1.8)
2i
β function and asymptotic freedom in QCD 237

Expanding the logarithm as a power series and using the Fourier representation of the propagator
gives, after a change of variables,
Z Z
1 X (−)n ′′ ′′
Γ(1) [φ] = dd z1 . . . dd zn V [φ(z1 )] . . . V [φ(zn )]
2i n=1 n
Z d Z d
d q2 d qn
. . . exp −iz1 (q2 + · · · + qn ) exp iz2 q2 . . . exp izn qn γ (n) (q2 , . . . , qn ), (1.9)
(2π)d (2π)d
Z Z
1 X (−)n ′′ ′′
= dd z1 . . . dd zn V [φ(z1 )] . . . V [φ(zn )]
2i n=1 n
Z d Z d
d q1 d qn
. . . e(n) (q1 , . . . , qn ), (1.10)
exp iq1 z1 exp iz2 q2 . . . exp izn qn γ
(2π)d (2π)d

where
Z
(n) dd qn 1 1 1
γ (q2 , . . . , qn ) = ...
(2π)d q 2 + m2 − iǫ (q + q2 )2 + m2 − iǫ (q + q2 + · · · + qn )2 + m2 − iǫ
(1.11)

while

e(n) (q1 , . . . , qn ) = (2π)4 δ 4 (q1 + · · · + qn )γ (n) (q2 , . . . , qn ) .

γ (1.12)

1.3 Structure of divergences

If one goes to Euclidean momentum and uses spherical coordinates, one finds for large radius κ,
i.e., when all momenta go to infinity together with a common factor, that γ (n) ∼ κd−1−2n , so
that the integrals are convergent for n > d/2.
In 4 dimensions, only γ (1) and γ (2) (q2 ) diverge. Furthermore, if one differentiates γ (2) with
respect to q2 , the result converges, so that only the q2 independent part of γ (2) (q2 ) diverges.
(1) (2)
One can thus write γ (1) = A + γf inite , γ (2) (q2 ) = B + γf inite (q2 ), with A, B infinite constants.
The one loop divergent part of the effective action thus becomes
Z
′′ A ′′ B

Γ(1)
∞ [φ] = dd x − V [φ(x)] + (V [φ(x)])2 . (1.13)
2i 4i
It follows that the divergences at order 1 in ~ are
• local functionals, i.e., integrals of polynomials in the fields and their derivatives, like the
action itself;
• the effective action, and thus also all other Green’s functions, can be made finite at first
order by adding divergent counterterms to the action;
• if the canonical dimension of the two fields differentiating the potential is strictly bigger
than 1 and if the potential is of canonical dimension less or equal to 4, then the canonical
dimension of the first order divergences is less or equal to 4.
4
Furthermore, for V (φ) = g φ4! , the theory can be made finite at one loop, by requiring mR =
m − ~ gA gB
2i , gR = g + ~ 16i to be finite.
At higher loops in order to get a finite theory, one needs to renormalize the wave function
as well, φR = Z −1/2 φ with Z = 1 + ~(divergent term). In fact, one can show that with
238 Barnich

appropriately chosen mR , gR , φR , the effective action can be made finite to all orders in ~. The
finite theory is then defined through the renormalized effective action

ΓR [φR , mR , gR ] = lim ΓΛ [Z 1/2 (Λ, mR , gR )φR , m(Λ, mR , gR ), g(Λ, mR , gR )]. (1.14)

Λ→∞

4
For V (φ) = g φ4! in 4 dimensions, an explicit computation of the divergent part using a cut-off
Λ in the radial part of spherical coordinates for the Euclidean four-momentum gives

3g 2 Λ 2
2
3gR Λ
gR = g − ~ ln + O(~ ) ⇐⇒ g = gR + ~ ln + O(~2 ). (1.15)
32π 2 m 32π 2 m

1.4 Symmetries
Let δQ φA = QA [φ] be a linear symmetry of the action,
Z
δS
0 = δQ S = dd x QA [φ] . (1.16)
δφA (x)

If the path integral measure can be assumed to be invariant, one can prove, by a shift of
integration variables, the Ward identities
δ δ
J · Q[ ]Z[J] = 0 = J · Q[ ]W [J]. (1.17)
δJ δJ

For the effective action, this becomes

Z
δΓ
δQ Γ = dd x QA [φ] = 0. (1.18)
δφA (x)

Under the above assumptions, one thus finds that the symmetries constrain the effective action,
and also the one loop divergences, in the same way than they constrain the classical action.

1.5 Background fields

Consider additional external sources, the background fields φeA (x) and define
Z
e φ]
Z[J, e = Dφ exp i (S[φ + φ] e + J · φ), (1.19)
~
e e
W e = ~ ln Z[J, φ] .
f [J, φ] (1.20)
i e 0]
Z[0,

The Legendre transform on J is now performed in the presence of the background fields, which
are passive,
e
f [J, φ]
φA
J,φe(x) =
δW
δJA (x)
e
⇔ JA (x) = JA [φ, φ](x), (1.21)
e φ]
Γ[φ, e = W [J[φ, φ],
e φ]e − J[φ, φ]
e · φ. (1.22)

e one can prove that

By shifting integration variables in (1.19) from φ to φ − φ,

e φ]
Γ[φ, e = Γ[φ + φ],
e (1.23)
β function and asymptotic freedom in QCD 239

e = Γ[0, φ],
and in particular, that Γ[φ] e i.e., the effective action can be obtained by computing
only vacuum graphs, but in the presence of background fields.

1.6 Guide to the literature

The computation of the one loop contribution to the effective action can be found in chapter
9-2-2 of [1]. The structure of divergences at one loop order are discussed in chapter 9.1 of [2].
Background fields are briefly discussed in chapter 16 of [3] and with more details in [4]. Copies of
detailed lecture notes (in french) for this part, with prerequisites and intermediate computations,
can be found in [5], chapter 2 and in [6]. In the next section, we follow chapter 13 of [1].

2 Asymptotic behavior
2.1 Dilatation invariance
Consider a change of scale,

x′ = λx, (2.1)
A
φA′ (x′ ) = λ−∆ φA (x). (2.2)

The action
Z
1 1 g 4
S= d4 x L, L = −[ ∂µ φ∂µ φ + m2 φ2 + φ ]. (2.3)
2 2 4!
is invariant under such a transformation, provided ∆φ = 1 and m = 0. The associated infinites-
imal transformation for decreasing scale is
∂
δφ = (1 + x · )φ. (2.4)
∂x
In the presence of a mass term, one finds
∂
δS + m S = 0. (2.5)
∂m
Because the transformation of the fields is linear, one naı̈vely expects, by repeating the reasoning
that leads to the Ward identities, the effective action to satisfy the same relation,
∂
δΓ + m Γ = 0. (2.6)
∂m
We will now see how renormalization, i.e., the process of absorbing divergences, leads to quantum
corrections.

2.2 Callan-Symanzik equation

In order to respect conventional dimensional analysis, one can adjust the finite part of the
subtraction so that
Λ Λ m m Λ
Z(Λ, mR , gR ) = Z( , gR ), g(Λ, mR , gR ) = g( , gR ), (Λ, mR , gR ) = ( , gR ) ,(2.7)
mR mR mR mR mR
240 Barnich

and similarly for the renormalized in terms of the bare quantities. Defining Γ∆,Λ [φ, m, g] =
1 ∂ΓΛ
2 m ∂m [φ, m, g] and varying m with Λ, g, φ fixed in (1.14), one finds
h Z i
dm ∂ ∂ δ
lim 2 Γ∆,Λ [φ, m, g] = dmR + dgR + d4 x δφR (x) ΓR [φR , mR , gR ]. (2.8)
Λ→∞ m ∂mR ∂gR δφR (x)

As ΓR [φR , mR , gR ] is of power counting dimension zero, (there is no longer any Λ which changes
the dimensional analysis), one also has
h Z i
∂ ∂ δ
dmR + d4 x (1 + x · )φR (x) ΓR [φR , mR , gR ] = 0. (2.9)
∂mR ∂x δφR (x)

Using δφ(x) = 0 = 21 Z −1/2 dZφR (x) + Z 1/2 δφR (x) which gives δφR (x) = − 21 d(ln Z)φR (x),
mR
multiplying (2.8) by dm R
and substituting (2.9), one finds

∂m mR
lim 2 Γ∆,Λ [φ, m, g] =
Λ→∞ ∂mR m
h Z i
∂gR ∂ 1 ∂ ln Z ∂ δ
= mR − d4 x (1 + mR +x· )φR (x) ΓR [φR , mR , gR ]. (2.10)
∂mR ∂gR 2 ∂mR ∂x δφR (x)

To simplify the RHS, one defines

∂gR ∂gR 1 ∂ ln Z 1 ∂ ln Z
β(gR ) = mR = −Λ , γ(gR ) = m =− Λ (2.11)
∂mR ∂Λ 2 R ∂mR 2 ∂Λ

where the limit Λ → ∞ is understood.

To understand the LHS, one couples to the starting point action φ2 (x) with an external
source K(x),
Z
1 g 1
SK = − d4 x[ ∂µ φ∂ µ φ + φ4 + m2 (1 + K(x))φ2 (x)] (2.12)
2 4! 2

∂ΓΛ R δΓΛ,K 
which gives Γ∆,Λ = 21 m = d4 x . In the computation of Γ(1) [φ], there is a new
∂m δK(x) K=0
′′
term because VK [φ(x)] = g2 φ2 (x) + m2 K(x). We thus have
Z
(1) ′′ A ′′ B
Γ∞,K = d4 x − VK [φ(x)] + (VK [φ(x)])2 (2.13)
2i 4i
with A, B divergent constants, which gives
(1)
δΓ∞,K  A B
 + gm2 φ2 (x) .
= −m2 (2.14)
δK(x) K=0 2i 4i
R
The frst term can be absorbed by a term of the form d4 xνK(x) in the Lagrangian. We
suppose that one uses normalization conditions that guarantee the absence of this term in the
renormalized effective action, cf. absence of tadpoles. The second term can be absorbed by a
B
redefinition of the source K(x): K(x) = ZK KR (x) with ZK = (1 − ~g 4i ). This implies that the
1/2
part linear in K(x) of ΓΛ,K [ZK KR (x), Z φR (x), m(), g()] = ΓR,K [KR , φR , mR , gR ] is finite as
β function and asymptotic freedom in QCD 241

well at 1-loop for Λ → ∞. One then defines

Z
δΓR,K 
ΓR,∆ [φR , mR , gR ] = d4 x =
δKR (x) KR =0
Z
δΓΛ,K 
= d4 xZK = ZK ΓΛ,∆ [Z 1/2 φR , m(), g()] , (2.15)
δK(x) K=0

∂m mR 1
The LHS of (2.10) becomes 2 ∂m R m ZK
ΓR,∆ [φR , mR , gR ]. If one defines

1 mR ∂m 1 Λ ∂m
1 + δ(gR ) = =− (2.16)
ZK m ∂mR ZK m ∂Λ

equation (2.10) writes as

2(1 + δ(gR ))ΓR,∆ [φR , mR , gR ] =

h Z i
∂ ∂ δ
= β(gR ) − d4 x (1 + γ(gR ) + x · )φR (x) ΓR [φR , mR , gR ]. (2.17)
∂gR ∂x δφR (x)

This is the Callan-Symanzik equation. The naı̈ve relation is thus violated by the terms containing
δ(gR ), γ(gR ), β(gR ) which are of order ~. They are due to the fact that the relation between
renormalized and bare quantities, needed in order to absorb the divergences, involves the cut-off
Λ.

2.3 High energy behavior and massless theory

In terms of Fourier transforms, the Callan-Symanzik reads

n
X
e R,∆ (0; p1 , . . . , pn ) = [β(gR ) ∂ ∂ e R (p1 , . . . , pn )(2.18)
2(1 + δ(gR ))Γ − n(1 + γ(gR )) − pk · ]Γ .
∂gR ∂pk
k=1

At tree level, one can check that the terms involving m2 vanish on their own. Because
e (1) (p1 , . . . , p2n ) ∼ γ
Γ e(n) (p1 + p2 , . . . , p2n−1 + p2n ) , (2.19)
e (1) (0; p1 , . . . , p2n ) ∼ γ
Γ e(n+1) (0; p1 + p2 , . . . , p2n−1 + p2n ) , (2.20)
∆

one can then show that, if all external momenta grow large together, p → λp,

e (1) (λp1 , . . . , λp2n ) = O( 1

Γ ), (2.21)
λ2(n−1)+4
e (1) (0; λp1 , . . . , λp2n ) = O( 1
Γ ∆ ). (2.22)
λ2n+4
Hence, the LHS of (2.18) decreases faster than the RHS. Because this property is not affected
by renormalization, one concludes, by denoting by Γ e as (p1 , . . . , pn ) the dominant behavior of
R
e
ΓR (p1 , . . . , pn ) at high energy, that
n
X
∂ ∂ e as (p1 , . . . , pn ) = 0 .
[β(gR ) − n(1 + γ(gR )) − pk · ]Γ R (2.23)
∂gR ∂pk
k=1
242 Barnich

Because usual dimensional analysis applies to the renormalized theory, scaling p → λp in

−1
Γas
R (pi ; gR , mR ) is equivalent, up to an overall factor, to scaling mR → λ mR . One can then
as
consider the Green’s functions ΓR (pi ; gR ) solutions of (2.23) as defining the Green’s functions
of the massless theory.

2.4 Running coupling constant

Forgetting about the subscript R and introducing

g(λ)
Z
d dg ′
λ g(λ) = β(g(λ)), g(1) = g ⇐⇒ λ = exp , (2.24)
dλ β(g ′ )
g

Zλ g(λ)
Z
d dλ′ γ(g(λ′ )) dg ′ γ(g ′ )
λ z(λ) = γ(g(λ))z(λ), z(1) = 1 ⇐⇒ z(λ) = exp = exp , (2.25)
dλ λ′ β(g ′ )
1 g

equation (2.23), which is valid for all values of the external momenta and of g, can be rewritten
as
d −n e as (λ−1 pi ; g(λ))] = 0,
λ [λ (z(λ))−n Γ (2.26)
dλ
which is equivalent to
e as (λpi ; g) = λ−n (z(λ))−n Γ
Γ e as (pi ; g(λ)) . (2.27)

Starting from x′ = λ−1 x, φ′ (x′ ) = λφ(x), one finds in terms of the Fourier transform, p′ =
λp, φe′ (p′ ) = λ−3 φ(p).
e At the quantum level, one defines φe′ (p′ ) = λ−3 z(λ)φ(p), e justifying the
name “anomalous dimension” for z(λ). In terms of the functional
X 1 Z d 4 p1 Z 4
d pn e
e g] =
e as [φ;
Γ . . . e
φ(−p1 ) . . . φ(−p e as
n )Γ (p1 , . . . , pn ; g) , (2.28)
n! (2π)4 (2π)4
n=2

equation (2.27) reads as

e as [φe′ ; g] = Γ
Γ e as [φe′ ; g(λ)]. (2.29)

We have thus shown that dilatation invariance of the massless theory can be restored at the
quantum level by the introduction of an anomalous dimension for the field and a coupling
“constant” that varies with energy.

The behavior of the coupling constant as a function of the energy is determined by the
β-function:
β function and asymptotic freedom in QCD 243

1 2 3 4
-1

-2

R ∞ dg′
• Singularity at finite energy: β(g) > 0 increases fast enough so that β(g ′ ) < ∞, for
3
instance β(g) = g . In this case, g(λ) increases from 0 to ∞ and g = ∞ is reached for a
R ∞ dg′
finite value of the energy, λ∞ = exp g β(g ′) .

R ∞ dg′ √
• Singularity at infinite energy: β(g) > 0 and β(g ′ ) → ∞, for instance, β(g) = g.
In this case, g(λ) increases from 0 to ∞ and reaches g = ∞ for an infinite value of the
energy, λ∞ = ∞.

• Ultraviolet fixed point: β(g) > 0 for 0 < g < g ∗ , β(g ∗ ) = 0, β(g) < 0 for g ∗ < g,
for instance, β(g) = −g(g − 2). In this case, g(λ) increases for 0 < g < g ∗ and g(λ)
decreases for g ∗ < g. Whatever side of g ∗ one starts off, g(λ) → g ∗ . If the zero of β(g)
d
is simple, β(g) → a(g − g ∗ ) for g → g ∗ with a > 0. The solution of λ dλ g = a(g − g ∗ ) is
g(λ) − g ∗ = (g(1) − g ∗ )λ−a . Assuming that γ(g) → γ(g ∗ ), the anomalous dimension at
∗
high energy becomes z(λ) = λγ(g ) .

• Asymptotic freedom: β(g) = −bg n , b > 0, n > 2 for g small. Integrating the differential
1
equation, one finds g(λ) = g[1+g n−1 (n−1)b ln λ]− n−1 . We thus have g(λ) → 0 for λ → ∞.
In this case, the assumption on which the perturbation theory is based, namely that g is
small, is valid at high energy.

g 4
For the scalar field with V [φ] = 4! φ in 4 dimensions, (1.15) and the definition of the β
function (2.11) imply that

d 3 2
β(gR ) = −Λ gR = ~ g + O(~2 ) . (2.30)
dΛ 16π 2 R
We are thus in case 1. This also applies to quantum electrodynamics where

e3
eR = e − ~ ln(Λ/µ) + O(~2 ) , (2.31)
12
with µ an infrared cut-off and thus

e3R
β(eR ) = ~ + O(~2 ). (2.32)
12π 2
For non abelian Yang-Mills theories based on the gauge group SU (N ), however, one can
244 Barnich

show that
g 3 11 nf
gR = g + ~ 2
( N− ) ln(Λ/µ) + O(~2 ) , (2.33)
4π 12 6
where nf is the number of fermions. One thus finds
3
gR nf 11
β(gR ) = ~ ( − N ) + O(~2 ). (2.34)
4π 2 6 12
For nf 6 5N , β(gR ) < 0 and the case 4 of asymptotic freedom applies. In particular, this holds
for quantum chromodynamics, where nf = 6 and N = 3. Since gR (λ) → 0 for λ → ∞, quarks
behave as free particles at high enough energies.

3 Quantum gauge fields

The rest of the lectures was devoted to the derivation of (2.33).
After shortly discussing the classical action for quantum chromodynamics, it was pointed out
that the quadratic part of the gluon action is not invertible as a consequence of gauge invariance.
The discussion of gauge fixing and BRST invariance was done in the context of the Batalin-
Vilkovisky approach: after introducing ghosts, antifields and an antibracket, one first constructs
the proper minimal solution to the Batalin-Vilkovisky master equation. After introducing a non
minimal sector that contains the antighosts, auxiliary fields and their antifields, gauge fixing
amounts to performing an anticanonical transformation generated by a gauge fixing fermion in
such a way that the classical master equation still holds. Formal gauge independence was then
shown to be a consequence of this master equation. This approach is reviewed for instance
in [7, 8]. Its relation to the more standard BRST approach is reviewed briefly in appendix 2.A
of [9].
The background field gauge was then introduced as in chapter 17 of [3]. The computation
of the renormalized coupling constant was performed following the same reference: in the back-
ground field gauge, the renormalization of the coupling constant and the background field are
related due to background gauge invariance. Hence, in the presence of the background fields, the
computation can be done by computing vacuum amplitudes. Furthermore, it can be reduced to
straightforward matrix algebra by using constant background fields.

References
[1] C. Itzykson and J. B. Zuber, “Quantum field theory,”. New York, USA: McGraw-hill
(1980) 705 P.(International Series In Pure and Applied Physics).

[2] J. Zinn-Justin, Quantum field theory and critical phenomena. International series of
monographs on physics. Oxford, UK: Clarendon, third ed., 1996.

[3] S. Weinberg, “The Quantum Theory of Fields. vol. 2: Modern Applications,”. Cambridge,
UK: Univ. Pr. (1996) 489 p.

[4] L. F. Abbott, “Introduction to the background field method,” Acta Phys. Polon. B13
(1982) 33.
β function and asymptotic freedom in QCD 245

[5] G. Barnich, “Méthodes de la théorie quantique des champs.” Lecture notes Université
Libre de Bruxelles, available online at
http://homepages.ulb.ac.be/%7Egbarnich/methodes.pdf, 2005.

[6] C. Schomblond, “Théorie quantique des champs.” Lecture notes Université Libre de
Bruxelles, available online at
http://homepages.ulb.ac.be/%7Ecschomb/intfonc+QED+QCD.pdf, 2005.

[7] M. Henneaux and C. Teitelboim, Quantization of Gauge Systems. Princeton University

Press, 1992.

[8] J. Gomis, J. Parı́s, and S. Samuel, “Antibracket, antifields and gauge theory
quantization,” Phys. Rept. 259 (1995) 1–145, hep-th/9412228.

[9] G. Barnich, F. Brandt, and M. Henneaux, “Local BRST cohomology in gauge theories,”
Phys. Rept. 338 (2000) 439–569, hep-th/0002245.

[10] S. Weinberg, “The Quantum Theory of Fields. vol. 1: Foundations,”. Cambridge, UK:
Univ. Pr. (1995) 609 p.

[11] G. Sterman, “An introduction to quantum field theory,”. Cambridge, UK: Univ. Pr.
(1993) 572 p.

[12] B. L. Voronov, P. M. Lavrov, and I. V. Tyutin, “Canonical transformations and the gauge
dependence in general gauge theories.,” Sov. J. Nucl. Phys. 36 (1982) 292.
[13] P. Di Francesco, P. Mathieu, and D. Senechal, “Conformal field theory,”. New York, USA:
Springer (1997) 890 p.

[14] M. Henneaux, “Physique mathématique II.” Notes de cours, 2ième licence en sciences
physique, 1993/94.
 Twistor Geometry and Gauge Theory

Martin Wolf

Institut für Theoretische Physik

Universität Hannover
Appelstraße 2, 30167 Hannover, Germany

E-mail: [email protected]

Abstract. These are lecture notes supplementing a three-hour introductory course

on twistor geometry and gauge theory given at the Modave Summer School on Math-
ematical Physics in June 2005. In the first lecture, we discuss the basics of twistor
geometry and a supersymmetric extension thereof. The second lecture introduces the
Penrose transform. Finally, the third lecture deals with the supertwistor correspon-
dence relating a certain theory on supertwistor space to super gauge theory in four
dimensions. Of course, I have tried to track down and to remove all mistakes from
these notes, but nevertheless it is rather unlikely I succeeded in doing so. In case you
find any, please let me know.

General remark: As these notes are intended to be on an introductory level, you may
wish to consult the literature cited in the bibliography for a more thorough discussion
of the subsequent topics.

Based on the lectures presented by M. Wolf at the First Modave Summer School in Mathe-
matical Physics held at Modave, Belgium on June 19-25, 2005
Twistor Geometry and Gauge Theory 247

1 Lecture 1
In this first lecture, we give basic underlying definitions and assertions of (super)twistor geometry
needed for the discussion of (super) gauge theory. We just start with the purely bosonic setup
and afterwards generalize the concepts to the supersymmetric setting. The following discussion
is mainly borrowed from [5, 6, 12].

1.1 Twistor space

Our starting point is to consider a complex vector space V of (complex) dimension n (that is,
= Cn ) and to introduce its flag manifolds
V ∼

Fd1 ···dm (V ) ≡ {(S1 , . . . , Sm ) | Si ⊂ V, dimC Si = di , S1 ⊂ S2 ⊂ · · · ⊂ Sm }, (1.1)

where the Si are subspaces of V . Typical examples of such flag manifolds are the projective space
F1 = CP n−1 and the Graßmannian Fk = Gk,n (C). All of these manifolds are compact complex
manifolds. Note that the flag manifolds (1.1) can equivalently be understood as homogeneous
spaces, i.e., consider the decomposition of Cn according to

Cn ∼
= Cd1 ⊕ Cd2 −d1 ⊕ Cd3 −d2 ⊕ · · · ⊕ Cdm −dm−1 ⊕ Cn−dm .

The subgroup of U (n) which preserves this decomposition is

U (d1 ) × U (d2 − d1 ) × · · · × U (dm − dm−1 ) × U (n − dm ) ⊂ U (n)

and hence
U (n)
Fd1 ···dm (V ) ≡ (1.2)
U (d1 ) × U (d2 − d1 ) × · · · × U (dm − dm−1 ) × U (n − dm )

From this definition, their dimensionality is easily computed to be

dimC Fd1 ···dm (V ) = d1 (n − d1 ) + (d2 − d1 )(n − d2 ) + · · · + (dm − dm−1 )(n − dm ).

As we eventually want to describe field theories in four dimensions, we consider a complex four-
dimensional vector space T4 and call it twistor space.1 Then we have a natural double fibration
in terms of flag manifolds,
F12 (T4 )
π1 @ π2 (1.3)
R
@
F1 (T4 ) F2 (T4 )
together with the canonical projections πi (S1 , S2 ) = Si for i = 1, 2. Next, we follow the lit-
erature and define P3 ≡ F1 (T4 ) = CP 3 , M4 ≡ F2 (T4 ) = G2,4 (C) and F5 ≡ F12 (T4 ) and
call them projective twistor space, compactified complexified Euclidean four-dimensional space
and correspondence space, respectively. The following proposition tells us how geometric data is
translated from M4 to P3 and vice versa:
Proposition 1.1. We have the following geometric twistor correspondence:
(i) point in P3 ←→ CP 2 ⊂ M4
(ii) CP 1 ⊂ P3 ←→ point in M4
1 Later on, we introduce real structures to obtain, e.g., Euclidean signature.
248 Wolf

Proof: Let us start with (i). By definition, a fixed point in P3 is a one-dimensional subspace S10 of T4 .
Thus,
π2 ◦ π1−1 (S10 ) = {S2 ⊂ T4 | dimC S2 = 2, S10 ⊂ S2 }.
Let e0 ∈ S10 be a nonzero vector and choose a basis for T4 of the form {e0 , e1 , e2 , e3 }. Letting [w1 , w2 , w3 ]
be homogeneous coordinates of CP 2 , we define

w = [w1 , w2 , w3 ] 7→ S2w = span{e0 , w1 e1 + w2 e2 + w3 e3 }.

Note that S10 ⊂ S2w . In fact, all subspaces arise in this manner and thus we have established a complex-
analytic isomorphism.

To prove (ii), we consider a fixed two-dimensional subspace of T4 and denote it by S20 . Therefore,
π1 ◦ π2−1 (S20 ) = {S1 ⊂ T4 | dimC S1 = 1, S1 ⊂ S20 }.

∼ C2 and hence π1 ◦ π −1 (S20 ) is isomorphic to the set of one-dimensional subspaces of

But S20 = C2 , that
2
is, to CP 1 . 

The next step is to introduce local coordinates on all the manifolds in the double fibration
(1.3). To do this, let  αα̇ 
x
x = (x ) ∈ C
αα̇ 2×2 ϕ
7→ (1.4)
12
be a coordinate mapping for M4 . The brackets denote, as usual, the equivalence relation induced
by rescalings. Then we define a coordinate chart on M4 by

M4 ≡ ϕ(C2×2 ) ∼
= C4 . (1.5)

We call M4 affine complexified Euclidean four-dimensional space noting that it is simply one
of six possible choices of standard coordinate charts for M4 . Using the projections, we may
naturally define the affine parts of F5 and P3 according to

F 5 ≡ π2−1 (M4 ) and P 3 ≡ π1 ◦ π2−1 (M4 ). (1.6)

Then we have the following little proposition:

= M4 × CP 1 .
Proposition 1.2. F 5 ∼

Proof: Let [λα̇ ] = [λ1̇ , λ2̇ ] be homogeneous coordinates of CP 1 and x as above, then consider the
mapping  αα̇   αα̇   αα̇   αα̇ 
x x x λα̇ x
(xαα̇ , [λα̇ ]) 7→ λα̇ , = ,
12 12 λα̇ 12
 
= S1x,λ , S2x,λ ∈ F5 ,
which in fact proves the assertion. 

The projection π1 : F 5 → P 3 is then readily given by the above considerations. We find

 αα̇ 
x λα̇
∈ P3 .
π
(xαα̇ , [λα̇ ]) 7→1 (1.7)
λα̇
Twistor Geometry and Gauge Theory 249

Therefore, our double fibration (1.3) in terms of these coordinates takes the form

(xαα̇ , [λα̇ ]) ∈ F5 F5 ∼= M4 × CP 1
π1 @ π2 π1 @ π2 (1.8)
R
@ R
@
 αα̇   αα̇  P3 M4
x λα̇ x
∈P
12 ∈ M
3 4
λα̇

Altogether, we may thus use the following local (inhomogeneous) coordinates:

• M4 : xαα̇ ,

• F 5 : xαα̇ and λ± with λ+ = λ−1

− on U+ ∩ U− , where U = {U± } denotes the canonical cover
of CP 1 ,

• P 3 : z±
α 3
and z± ; π1 : (xαα̇ , λ± ) 7→ (xαα̇ λ± α 3 + − +
α̇ , λ± ) = (z± , z± ), with λα̇ = λ+ λα̇ and (λα̇ ) ≡
t
(1, λ+ ).

In particular, the last point shows that P 3 is a holomorphic fibration over the Riemann sphere,
i.e., it is the total space of
O(1) ⊕ O(1) → CP 1 . (1.9)

Remark 1.3. Our projective twistor space P3 is, as already mentioned, the same as CP 3 . On
this space we may introduce the homogeneous coordinates [ω α , λα̇ ]. Then consider CP 3 \ CP 1 ,
where CP 1 ⊂ CP 3 is defined by putting λα̇ = 0 and ω α 6= 0. We can cover CP 3 \ CP 1 by two
coordinate patches, say U+ and U− for which λ1̇ 6= 0 and λ2̇ 6= 0, respectively, and introduce the
coordinates

α ωα 3 λ α ωα 3 λ
z+ ≡ , z+ ≡ 2̇ ≡ λ+ on U+ , z− ≡ , z− ≡ 1̇ ≡ λ− on U− , (1.10)
λ1̇ λ1̇ λ2̇ λ2̇

which are related on U+ ∩ U− by

α 1 α 3 1
z+ = 3 z− and z+ = 3 . (1.11)
z− z−

This shows that CP 3 \ CP 1 is the same as P 3 = O(1) ⊕ O(1). In a given trivialization, global
holomorphic sections of this bundle are locally given by
α
z± = xαα̇ λ±
α̇ , (1.12)

where (xαα̇ ) ∈ M4 and

 
1 −
(λ+
α̇ ) ≡ and λ+
α̇ = λ+ λα̇ . (1.13)
λ+
α
In other words, the equations (1.12) make proposition 1.1 transparent: a fixed point p = (z± , λ±
α̇ ) ∈
P 3 corresponds to a (null) two-plane in M4 and furthermore, a fixed point x = (xαα̇ ) ∈ M4
corresponds to a holomorphic embedding of a rational degree one curve CPx1 ֒→ P 3 .
250 Wolf

1.2 Supertwistor space

Up to now, we have discussed the purely bosonic setup. In the third lecture, we shall discuss,
however, supersymmetric gauge theories in the twistor approach, for which we need an extension
of the twistor space to the supertwistor space, etc.2
Therefore, let us already at this point give the basic underlying concepts.
Let R = R0 ⊕ R1 be a Z2 -graded ring, that is, R0 R0 ⊂ R0 , R1 R0 ⊂ R1 , R0 R1 ⊂ R1 and
R1 R1 ⊂ R0 . We call elements of R0 even and elements of R1 odd. An element of R is said to
be homogeneous if it is either even or odd. The degree (or parity) of a homogeneous element is
defined to be 0 if it is even and 1 if it is odd, respectively. We denote the degree of a homogeneous
element r ∈ R by pr (p for parity). The ring R is called supercommutative if it satisfies

r1 r2 = (−)pr1 pr2 r2 r1 , (1.14)

for all homogeneous elements r1,2 ∈ R. An R-module M is a Z2 -graded bimodule which satisfies
rm = (−)pr pm mr, (1.15)
3
for r ∈ R, m ∈ M , with M = M0 ⊕ M1 . Then there is a natural map Π – called the parity
operator – which is defined by

(ΠM )0 ≡ M1 and (ΠM )1 ≡ M0 . (1.16)

We should stress that R is an R-module itself, and as such ΠR is an R-module, as well. However,
ΠR is no longer a Z2 -graded ring since (ΠR)1 (ΠR)1 ⊂ (ΠR)1 , for instance.
For our purposes, the most important example of such a Z2 -graded ring is the Graßmann
algebra over some (complex) vector space V of dimension n,
M
R = Λ• V ≡ Λp V, (1.17)
p

with the Z2 -grading M M

R = Λ2p V ⊕ Λ2p+1 V. (1.18)
p p

A free module of rank m|n over R is defined by

M m|n = Rm ⊕ (ΠR)n , (1.19)

where Rm = R ⊕ · · · ⊕ R. An example is Cm|n , where we consider the complex numbers as

Z2 -graded ring (just put R1 = 0). We call this module the complexified Euclidean superspace.
Let now U be an open subset in Cm and define

A ≡ OCm ⊗ Λ• V = OCm (Λ• V ), (1.20)

where OCm is the sheaf of holomorphic functions on Cm . Thus, A is a Z2 -graded sheaf of su-
percommutative rings consisting of local Λ• V -valued holomorphic functions. Let (z 1 , . . . , z m ) be
coordinates on Cm and (η1 , . . . , ηn ) be basis sections of V ∼
= Λ1 V . Then (z 1 , . . . , z m , η1 , . . . , ηn )

2 For a detailed introduction to the concepts of supermanifolds, we refer to [1, 5].

3 More precisely, it is a functor from the category of R-modules to the category of R-modules.
Twistor Geometry and Gauge Theory 251

are coordinates for the space (U, OU (Λ• V )). Any f ∈ Γ(U, OU (Λ• V )) can thus be expressed as
X
f (z, η) = fI (z)η I , (1.21)
I

where I is a multiindex. These are the fundamental functions in supergeometry.

These preliminaries allow us to introduce supermanifolds. Let X be some topological space
of real dimension 2m, and let O be a sheaf of Z2 -graded rings on X. Furthermore, let N be the
ideal subsheaf in O of all nilpotent elements in O, and define

Ored ≡ O/N . (1.22)

Then (X, O) is called a complex supermanifold of dimension m|n if:

• (X, Ored ) is a complex manifold.4 We call (X, Ored ) the body of (X, O).

• For each point x ∈ X there is a neighborhood N ∋ x such that

= Ored |N (Λ• Cn ).
O|N ∼

Consider again our complexified Euclidean superspace Cm|n . Then we can define the flag
supermanifolds Fp1 |q1 ···pk |qk (Cm|n ) as the set of all k-tuples (S1 , . . . , Sk ) of free submodules of
Cm|n satisfying S1 ⊂ · · · ⊂ Sk ⊂ Cm|n and rankSi = pi |qi . This naturally generalizes our
definition (1.1). In fact, one can also introduce suitable coordinate systems and thus a suitable
structure sheaf which makes this set into a supermanifold. We shall discuss this for a particular
example momentarily.
Similarly to our discussion in the previous section, consider a free C-module T4|N of rank
4|N . We call this space N -extended supertwistor space. Then we introduce

P3|N ≡ F1|0 (T4|N ),

M4|2N ≡ F2|0 (T4|N ), (1.23)
F5|2N ≡ F1|0,2|0 (T4|N ).
Therefore, the double fibration (1.3) becomes:

F5|2N
π1 @ π2 (1.24)
R
@
P3|N M4|2N
Note that P3|N = CP 3|N . Furthermore, proposition 2.1 generalizes accordingly:
Proposition 1.4. We have the following geometric supertwistor correspondence:

(i) point in P3|N ←→ CP 2|N ⊂ M4|2N

(ii) CP 1|0 ⊂ P3|N ←→ point in M4|2N

Proof: see exercise 2.3 

4 Recall that for a ringed space (X, O ) with the property that for each x ∈ X there is a neighborhood N ∋ x

C
X
such that there is a ringed space isomorphism (N, OX |N ) ∼= (U, OCm ), where U ⊂ m . Then X can be given
the structure of a complex manifold and moreover, any complex manifold arises in this manner.
252 Wolf

In the following, we shall abbreviate X d|0 = X d for any ordinary manifold X.

As before, we introduce local coordinates and it is then rather obvious that one gets

F 5|N ∼= M4|2N × CP 1
π1 @ π2 (1.25)
R
@
P 3|N M4|2N
as well as
• M4|2N : xαα̇ and ηiα̇ ,
• F 5|2N : xαα̇ , ηiα̇ and λ± with λ+ = λ−1
− on U+ ∩ U− , where U = {U± } denotes again the
canonical cover of CP 1 ,
• P 3|N : z±
α 3
, z± and ηi± ; π1 : (xαα̇ , λ± , ηiα̇ ) 7→ (xαα̇ λ± α̇ ± α 3 ±
α̇ , λ± , ηi λα̇ ) = (z± , z± , ηi ), with
+ − + t
λα̇ = λ+ λα̇ and (λα̇ ) ≡ (1, λ+ ).
The last point shows that P 3|N is the total space of the holomorphic fibration over the Riemann
sphere
O(1) ⊗ C2 ⊕ ΠO(1) ⊗ CN → CP 1 . (1.26)
Moreover, the relations
α
z± = xαα̇ λ±
α̇ and ηi± = ηiα̇ λ±
α̇ (1.27)

explicitly say that a fixed point p = (z±α

, λ± ±
α̇ , ηi ) ∈ P
3|N
corresponds to a superplane of dimen-
4|2N
sion 2|N in M while a fixed point (x, η) = (x , ηi ) ∈ M4|2N corresponds to a holomorphic
αα̇ α̇

embedding of a rational curve CPx,η 1

֒→ P 3|N .
In the remainder of this subsection, we compute the first Chern number of the total space
of the fibration (1.26). Remember that O(1) → CP 1 is by definition the dual tautological line
bundle L∗ over the Riemann sphere. Recall also that the natural curvature two-form on the
tautological line bundle takes the form

dλ± ∧ dλ̄±
F|U± = − . (1.28)
(1 + λ± λ̄± )2
i
Hence, integrating the first Chern class c1 (L) = 2π F then gives the first Chern number of L
Z
C1 (L) = c1 (L) = −1. (1.29)
CP 1

Since c1 (L∗ ) = −c1 (L), the first Chern number bundle O(1) → CP 1 is one.5 Upon using
c(E ⊕ F ) = c(E) ∧ c(F ), the above results and the fact that the first Chern number of the sphere
CP 1 is two, we obtain for the total first Chern number of the supertwistor space6 P 3|N
C1 (P 3|N ) = 4 − N . (1.30)

Here, we noticed that the fermions contribute with an opposite sign due to the definition of
the Berezin integral. Note that (1.30) vanishes iff N = 4. In particular, this means that the
5 Generally speaking, the line bundle O(n) ≡ O(1)⊗n (respectively, O(−n) ≡ O(n)∗ ) has first Chern number

n (respectively, −n).
6 For brevity, we shall also call P 3|N supertwistor space.
Twistor Geometry and Gauge Theory 253

Berezinian line bundle of the holomorphic tangent bundle T 1,0 P 3|N (holomorphic Berezinian
line bundle for short),
Ber = KP 3 ⊗ ΛN (O(1) ⊗ CN ), (1.31)
which replaces the canonical bundle in the case of supermanifolds, becomes trivial in the case
of maximal N = 4 supersymmetry, since the canonical bundle KP 3 of P 3 is O(−4). Therefore,
we say that P 3|4 is a Calabi-Yau supermanifold. This observation will be crucial in our later
discussions on super gauge theory (see section 4.4).

Remark 1.5. The following is taken from [4]. In our above definition of a supermanifold (X, O),
we have assumed that the sheaf O is locally isomorphic to Ored (Λ• Cn ). Let now O be globally
of the form O = Ored (Λ• E ∗ ), where E is some holomorphic vector bundle of rank n called the
fermionic tangent bundle. The holomorphic Berezinian line bundle is then defined by

Ber ≡ K ⊗ Λn E,

where K is the canonical bundle of X. Let now X be some complex three-manifold that admits a
spin structure (for instance, the twistor space of any anti-self-dual real four-dimensional manifold
is of this form). Then X can be extended to a Calabi-Yau supermanifold of dimension 3|4 by
setting E = J 1 K −1/2 , which is the bundle of one-jets of holomorphic sections of K −1/2 . How
can one see this? Well, we have a short exact sequence

0 −→ Ω1,0 (X) ⊗ K −1/2 −→ J 1 K −1/2 −→ K −1/2 −→ 0

from which we obtain

Λ4 E ∼
= K −1/2 ⊗ Λ3 (Ω1,0 (X) ⊗ K −1/2 )
∼
= K −1/2 ⊗ Ω3,0 (X) ⊗ K −3/2
∼
= K −1/2 ⊗ K ⊗ K −3/2 ∼
= K −1
and hence Ber = K ⊗ Λ4 E ∼ = K ⊗ K −1 is trivial. Therefore, (X, Ored (Λ• (J 1 K −1/2 )∗ )) is Calabi-
Yau. In particular, in the case of the twistor space P 3 , we have J 1 K −1/2 = O(1) ⊗ C4 , which
reproduces the supertwistor space P 3|4 (and similarly for the compactifications CP 3 and CP 3|4 ).

1.3 Some reality

So far, we have only discussed the complex setting. But in order to discuss field theories on
Euclidean four-dimensional space, we need to introduce certain real structures. On P 3|N , a
good real structure for our purposes is given by the anti-linear transformations [10]
 2 1 
z̄± z̄± 1 1
1 2 3
τ (z± , z± , z± ) = ± 3 , ∓ 3 , − 3 and τ (ηi± ) = ± 3 Ti j η̄j± , (1.32a)
z̄± z̄± z̄± z̄±

where  
0 1 0 0
−1 0 0 0
(Ti ) = 
j
0
. (1.32b)
0 0 1
0 0 −1 0
254 Wolf

Here, we have just given the definition for N = 4 as it will be, by the above reasoning, the most
interesting case for us.7 It is not so difficult to see that τ has no fixed points in P 3|N . However,
it does leave invariant those curves CPx,η 1
֒→ P 3|N defined in (1.27) which are parametrized by
the moduli (x , ηi ) ∈ C
αα̇ α̇ 4|2N
and obey

τ (xαα̇ ) = ǫαβ ǫα̇β̇ x̄β β̇ = xαα̇ and τ (ηiα̇ ) = ǫα̇β̇ Ti j η̄jβ̇ = ηiα̇ . (1.33)

Therefore, we restrict our moduli to the real slice R4|2N ⊂ C4|2N on whose body we can introduce
real coordinates (xµ ) ∈ R4 , with µ = 1, . . . , 4, according to

x22̇ = x̄11̇ ≡ x4 − ix3 and x21̇ = −x̄12̇ ≡ −x2 + ix1 , (1.34)

yielding the Euclidean metric

ds2 = det(dxαα̇ ) = δµν dxµ dxν .

Thus, upon imposing the above reality conditions, we reduce the space M4|2N in (1.25) to
R4|2N and the correspondence space becomes R4|2N × CP 1 , respectively. What about the
supertwistor space P 3|N ? In fact, the formulas (1.27) together with their complex conjugates
define the diffeomorphism
R4|2N × CP 1 ∼ = P 3|N .
Altogether, we see that as real manifolds the supertwistor space and the correspondence space
are diffeomorphic and hence the double fibration (1.25) reduces to the single fibration

P 3|N → R4|2N .
In the sequel, we shall mostly work under the assumption that the reality conditions are imposed.

– Exercises –

Exercise 1.1. Let E → X be a holomorphic vector bundle over a complex manifold X with
structure sheaf Ohol and define O ≡ Ohol (Λ• E). Show that (O, X) is a supermanifold according
to our definition given above.

Exercise 1.2. Show that the dimension of Fp|q (Cm|n ) is p(m − p) + q(n − q)|mq + np.

Exercise 1.3. Prove proposition 2.3.

Exercise 1.4. In the third lecture, we need to consider the anti-holomorphic tangent bundle
T 0,1 P 3|N of the supertwistor space P 3|N . Perform again the change of coordinates (z± α 3
, z± , ηi± ) 7→
(xαα̇ , λ± , ηiα̇ ) given by z±α
= xαα̇ λ± 3 ± α̇ ±
α̇ , z± = λ± and ηi = ηi λα̇ , where (x , ηi ) ∈ R4|2N . Show
αα̇ α̇
0,1 3|N
that a basis of T P is given by
∂ ∂ ∂
V̄α± = λα̇
± , V̄3± = , and V̄±i = λα̇
± .
∂xαα̇ ∂ λ̄± ∂ηiα̇

7 But it works similarly for N = 2. Recall that Euclidean reality can only be imposed if the number of
supersymmetries is even.
Twistor Geometry and Gauge Theory 255

2 Lecture 2
Subject of this second lecture is to show how one can solve zero rest mass field equations in the
twistor approach. For simplicity, we work in the purely bosonic setting.

2.1 Penrose transform

To begin with, let P 3 be again the twistor space and consider the following integral
I
1 ± β β̇ ± ±
φ(x) = dλα̇
± λα̇ f+− (x λβ̇ , λβ̇ ), (2.1)
2πi
C

where f+− is a meromorphic function of z± α

= xαα̇ λ± 3
α̇ and z± = λ± on twistor space, which
= S ⊂ CP . Clearly, only in the case when
is holomorphic in the vicinity of the curve C ∼ 1 1

the integrand has total homogeneity zero, the contour integral gives a nonzero contribution.
Therefore, f+− should be homogeneous of degree −2, or equivalently, a section of the bundle
O(−2).8 Then one readily computes that

φ = ∂ αα̇ ∂αα̇ φ = 0, (2.2)

i.e., the function φ satisfies the Klein-Gordon equation. That means any f+− in twistor space,
which has the required homogeneity, will yield via the above contour integral a solution to the
1 2
Klein-Gordon equation. In particular, in exercise 3.1 you will show that f+− ∼ 1/z± z± indeed
µ
gives the well-kown result φ ∼ 1/x xµ .
What about the other zero rest mass field equations? These can be solved in a similar
fashion by contour integrals over certain functions on twistor space. Consider a zero rest mass
field φα̇1 ···α̇2h of helicity h. Then
I
1 ± ± ± β β̇ ± ±
φα̇1 ···α̇2h (x) = dλα̇
± λα̇ λα̇1 · · · λα̇2h f+− (x λβ̇ , λβ̇ ) (2.3)
2πi
C

solves the equation

ǫα̇α̇1 ∂αα̇ φα̇1 ···α̇2h = 0. (2.4)
Again, in order to have a nonzero integral, the integrand should have total homogeneity zero,
which is equivalent to require that f+− is O(−2h − 2)-valued. We may also consider a zero rest
mass field φα1 ···α2h of helicity −h for which we take
I
1 ± ∂ ∂ β β̇ ± ±
φα1 ···α2h (x) = dλα̇
± λα̇ α1 · · · α2h f+− (x λβ̇ , λβ̇ ), (2.5)
2πi ∂z± ∂z±
C

such that f+− is a section of O(2h − 2). Hence,

ǫαα1 ∂αα̇ φα1 ···α2h = 0. (2.6)

These contour integral formulas are the advertised Penrose transforms.

As we have just learned, any function on twistor space – provided it is a section of an
appropriate O(n) bundle for some n ∈ Z and holomorphic in the twistor coordinates in the
vicinity of the contour C ∼= S 1 ⊂ CP 1 – can be used to construct solutions to certain zero
8 Note that for fixed xαα̇ it is O(−2) over CP 1 otherwise over P 3 .
256 Wolf

rest mass field equations. Certainly, there are a lot of different functions leading to the same
solution. For instance, we could simply change f+− by adding a function which has singularities
on one side of the contour but is holomorphic on the other, since the contour integral does not
feel such functions. So, how can we understand what is happening? The tool which helps us to
clarify this issue is Cech cohomology.

2.2 Cech cohomology groups and Penrose’s theorem

Consider some Abelian9 sheaf S over some topological space X, and an open cover U = {Ua }
of X. A q-cochain of the covering U with values in S is a collection f = {fa0 ···aq } of sections of
the sheaf S over nonempty intersections Ua0 ∩ · · · ∩ Uaq . The fa0 ···aq are assumed to be totally
skew-symmetric in their indices.
The set of all q-cochains has an Abelian group structure (with respect to addition) and is
denoted by C q (U, S). Then we define the coboundary map by

δq : C q (U, S) → C q+1 (U, S),

(δq f )a0 ···aq+1 ≡

q+1
X b
a ···a ···a
(−)i ra00···aq+1
i q+1
b
fa0 ···ai ···aq+1 ,
(2.7)

i=0

where
b
a ···a ···a
i
ra00···aq+1 q+1
b
: S(Ua0 ∩ · · · ∩ Uai ∩ · · · ∩ Uaq+1 ) → S(Ua0 ∩ · · · ∩ Uaq+1 ) (2.8)
is the sheaf restriction homomorphism and b
ai means omitting ai . It is clear that δq is a group
homomorphism, and one may check that δq ◦ δq−1 = 0 (see exercise 3.2). Furthermore, we see
straight away that ker δ0 = S(X). We define

Z q (U, S) ≡ ker δq and B q (U, S) ≡ im δq−1 . (2.9)

We call elements of Z q (U, S) q-cocycles and elements of B q (U, S) q-coboundaries, respectively.

Note that Z q (U, S) and B q (U, S) are both Abelian groups and since the coboundary map is
nilquadratic, B q (U, S) is a normal subgroup of Z q (U, S). The q-th Cech cohomology group is the

quotient
H q (U, S) ≡ Z q (U, S)/B q (U, S). (2.10)
In order to get used to these definitions, let us consider a simple example and take CP 1
together with the holomorphic line bundle O(n) → CP 1 . Then we have the following proposition:
Proposition 2.1. Consider CP 1 with the canonical cover U = {U± }. Then
n ··· −4 −3 −2 −1 0 1 2 ···
H 0 (U, O(n)) 0 0 0 0 C1 C2 C3
H 1 (U, O(n)) C3 C2 C1 0 0 0 0
Proof: Let us just pick two particular examples. You may wish to prove the rest on your own (see
exercise 3.3).
9 In the next lecture, we will also introduce cohomology sets for non-Abelian sheaves.
Twistor Geometry and Gauge Theory 257

+
Consider H 0 (U, O(0)). It is the space of pairs of functions f± analytic on U± satisfying r+− f+ =
− ±
r+− f− , where r+− : O(0)|U± → O(0)|U+ ∩U− is the restriction homomorphism. In other words f+
and f− agree on the overlap U+ ∩ U− and therefore define a global holomorphic function on CP 1 . By
Liouville’s theorem this must be a constant, so H 0 (U, O(0)) ∼= C. Next, let us consider H 1 (U, O(0)).
This is the space of functions f+− on U+ ∩ U− modulo coboundaries. Note that the domain U+ ∩ U− is
an annulus, so these functions f+− can be expanded in Laurent series according to
X
∞ X
∞
f+− = an λn
+ − a−n λ−n
+ .
n=0 n=1

So, let
X
∞ X
∞
f+ = an λn
+ and f− = a−n λ−n
+ .
n=0 n=1
+ −
Then the f± converge on U± and f+− = r+− f+ − r+− f− , that is, f+− is a coboundary. Therefore,
1
H (U, O(0)) = 0. 

The above proposition hints that there is some sort of duality. In fact,

H 1 (U, O(−n − 2)) ∼

= H 0 (U, O(n))∗ , (2.11)

which is a special instance of Serre duality. Here, the star denotes the vector space dual. To
understand this better, consider

g = {g+ , g− } ∈ H 0 (U, O(n)), with g± = aα̇1 ···α̇n λ± ±

α̇1 · · · λα̇n (2.12)

and f = {f+− } ∈ H 1 (U, O(−n − 2)). Then define the pairing

I
1 ±
(f, g) ≡ dλα̇
± λα̇ g± f+− , (2.13)
2πi
C

where the contour is chosen as before. Evidently, this expression is complex linear and non-
degenerate and depends only on the cohomology class of f . Hence, it gives the duality (2.11).
A nice way of writing (2.13) is as (f, g) = bα̇1 ···α̇n aα̇1 ···α̇n , where
I
1 ± ± ±
bα̇1 ···α̇n = dλα̇± λα̇ λα̇1 · · · λα̇n f+− , (2.14)
2πi
C

such that the contour integral formula (2.3) can be recognized as an instance of Serre duality.
Thus, we have reached our goal: our twistor functions f+− from the section 2.1 leading to
solutions of zero rest mass field equations should be thought of as representatives of certain
sheaf cohomology classes. In fact, we can state:

Theorem 2.2. (Penrose) We have the following two isomorphisms

H 1 (P 3 , O(∓2h − 2)) ∼
= Γ(M4 , Z±h ),

where Z±h is the sheaf of solutions to the helicity ±h zero rest mass field equations.
Proof: see [7, 8, 12] 

Finally, it remains to clarify one issue. Apparently, all of our above calculations seem to
depend on the particular cover U of CP 1 . But is this really the case?
258 Wolf

Consider again some topological space X with cover U together with some Abelian sheaf S.
If another cover V is the refinement of U, that is, for U = {Ua }a∈A and V = {Vb }b∈B there is a
map λ : B → A of index sets, such that for any b ∈ B, Vb ⊆ Uλ(b) , then there is a natural group
homomorphism (induced by the restriction mappings of the sheaf S)
q q
hU
V : H (U, S) → H (V, S). (2.15)

We can then define the inductive limit of these cohomology groups with respect to the partially
ordered set of all coverings,

H q (X, S) ≡ lim ind H q (U, S), (2.16)

U

which we call the q-th Cech cohomology group of X with coefficients in S. By the properties of

inductive limits, we have a homomorphism

H q (U, S) → H q (X, S). (2.17)

Now the question is when this becomes an isomorphism. Sometimes it is possible to find a fixed
covering of X such that
H q (U, S) ∼
= H q (X, S). (2.18)
The following proposition tells us, when this is going to happen:
Proposition 2.3. (Leray) Let U = {Ua } be a covering of X with the property that for all tuples
(Ua0 , . . . , Uap ) of the cover H q (Ua0 ∩ · · · ∩ Uap , S) = 0 for all q > 1. Then

H q (U, S) ∼
= H q (X, S).

Proof: see literature cited in the bibliography 

Such covers a called Leray covers and in fact our cover U± of CP 1 is of this form such that
proposition 3.1 becomes:
Proposition 2.4. Consider CP 1 with the canonical cover. Then
n ··· −4 −3 −2 −1 0 1 2 ···
H 0 (CP 1 , O(n)) 0 0 0 0 C1 C2 C3
H 1 (CP 1 , O(n)) C3 C2 C1 0 0 0 0
For more details about Cech cohomology, we refer to the literature cited in the bibliography.

– Exercises –
1 2
Exercise 2.1. Consider the following twistor function f+− ∼ 1/z± z± . Show that the integral
µ
(2.1) gives φ ∼ 1/x xµ , i.e., the Greens function of the Klein-Gordon operator. You may wish
to use (1.34).
Exercise 2.2. Check that δq+1 ◦ δq = 0 for δq defined in (2.7).
Exercise 2.3. Proof the rest of proposition 3.1.
Twistor Geometry and Gauge Theory 259

3 Lecture 3
In the preceding lecture, you learned something about the Penrose transform relating certain
sheaf cohomology groups with solutions to zero rest mass field equations. However, these field
equations were free field equations. In this last lecture, we focus on a particular field theory –
namely on self-dual super gauge theory – and discuss how solutions to its (non-linear) equations
of motion are related to certain sheaf cohomology sets eventually resulting in the supertwistor
correspondence and the Penrose-Ward transform.

3.1 Holomorphic vector bundles in the Cech approach

Let X be some (super)manifold with cover U = {Ua }. Furthermore, we are interested in smooth
maps from open subsets of X into the group GL(n, C) as well as in the non-Abelian sheaf G of
such GL(n, C)-valued functions. A q-cochain of the covering U with values in G is a collection
f = {fa0 ···aq } of sections of the sheaf G over nonempty intersections Ua0 ∩ · · · ∩ Uaq . We denote,
as before, the set of such q-cochains by C q (U, G). We stress that it has a group structure, where
the multiplication is just pointwise multiplication.
We may define the subsets of cocylces Z q (U, G) ⊂ C q (U, G). For example, for q = 0, 1 these
are given by

Z 0 (U, G) ≡ {f ∈ C 0 (U, G) | fa = fb on Ua ∩ Ub 6= ∅},

−1
Z 1 (U, G) ≡ {f ∈ C 1 (U, G) | fab = fba on Ua ∩ Ub 6= ∅ (3.1)
and fab fbc fca = 1 on Ua ∩ Ub ∩ Uc 6= ∅}.

We remark that from the first of these two definitions it follows that Z 0 (U, G) coincides with
the group
H 0 (X, G) ≡ G(X) = Γ(X, G),
which is the group of global sections of the sheaf G. Note that in general the subset Z 1 (U, G) ⊂
C 1 (U, G) is not a subgroup of the group C 1 (U, G).
We say that two cocycles f, f ′ ∈ Z 1 (U, G) are equivalent if fab
′
= ga−1 fab gb for some g ∈
0
C (U, G). The set of equivalence classes induced by this equivalence relation is the first coho-
mology set and denoted by H 1 (U, G). If the Ua are all Stein manifolds10 we have the bijection

H 1 (U, G) ∼
= H 1 (X, G).

Furthermore, we shall also need the sheaf of holomorphic sections of the trivial bundle X ×
GL(n, C), which we denote by OGL , in the sequel.
For the discussion of the supertwistor correspondence, holomorphic vector bundles over
the supertwistor space are needed. Generally speaking, a collection consisting of five objects,
(E, X, pr, U, f ), is called a rank n holomorphic vector bundle, whenever E and X are complex
(super)manifolds, the map
pr : E → X
is a surjective holomorphic projection, U = {Um } is an open covering of X and f = {fab } is
a collection of holomorphic transition functions defined on nonempty intersections Ua ∩ Ub and
10 We call a complex manifold X Stein if X is holomorphically convex (that is, the holomorphically convex hull

of any compact subset of X is again compact in X) and for any x, y ∈ X with x 6= y there is some f ∈ OX such
that f (x) 6= f (y).
260 Wolf

taking values in GL(n, C).

After these generalities, let us now stick to the supertwistor space P 3|N with its two-set
open covering U = {U+ , U− }. We note that the U± are indeed Stein manifolds. Any cocyle
f = {f+− } ∈ Z 1 (U, G) uniquely defines a rank n complex vector bundle E over P 3|N . As
equivalent cocyles define isomorphic complex vector bundles and since the U± are Stein mani-
folds, the isomorphism class of rank n complex vector bundles is parametrized by H 1 (P 3|N , G).
Choosing G to be OGL , we see that holomorphic vector bundles over the supertwistor space are
parametrized by H 1 (P 3|N , OGL ).

3.2 Penrose-Ward transform I

Consider a rank n holomorphic vector bundle E → P 3|N . In exercise 2.4, you have shown that
the anti-holomorphic tangent bundle of P 3|N is spanned by
∂ ∂ ∂
V̄α± = λα̇
± , V̄3± = , and V̄±i = λα̇
± . (3.2)
∂xαα̇ ∂ λ̄± ∂ηiα̇

Thus, by definition of a holomorphic vector bundle, the transition function f = {f+− } ∈

H 1 (P 3|N , OGL ) is annihilated by those vector fields. Recall from the first lecture that by virtue
of the geometric supertwistor correspondence, any (xαα̇ , ηiα̇ ) ∈ R4|2N corresponds to a projective
line CPx,η1
֒→ P 3|N . Now, we additionally want to assume that the bundle E → P 3|N becomes
holomorphically trivial when restricted to any of those lines. Let us denote their moduli space
by
Mhol (P 3|N ) ⊂ H 1 (P 3|N , OGL ).
Then the Birkhoff decomposition theorem tells us that the bundle E becomes topologically trivial
such that the transition function f+− can be split into GL(n, C)-valued functions ψ± satisfying
−1
f+− = ψ+ ψ− , with ∂λ̄± ψ± = 0 (3.3)

Applying the remaining vector fields of (3.2) to f+− , we realize that

−1 −1 −1 −1
ψ+ V̄α+ ψ+ = ψ− V̄α+ ψ− and ψ+ V̄+i ψ+ = ψ− V̄+i ψ− (3.4)

must be at most linear in λ+ . Therefore, we may introduce a (Lie algebra valued) (0, 1)-form
A0,1 such that
± −1
V̄α± yA0,1 ≡ A± α̇
α ≡ λ± Aαα̇ = ψ± V̄α ψ± ,
∂λ̄± yA0,1 ≡ Aλ̄± = 0, (3.5)
−1
V̄±i yA0,1 ≡ Ai± ≡ λα̇ i
± Aα̇ = ψ± V̄±i ψ± ,
and hence,
(V̄α± + A±
α )ψ± = 0,
∂λ̄± ψ± = 0, (3.6)
(V̄±i + Ai± )ψ± = 0.
The compatibility conditions for the linear system (3.6) read as

[∇αα̇ , ∇β β̇ ] + [∇αβ̇ , ∇β α̇ ] = 0, [∇iα̇ , ∇β β̇ ] + [∇iβ̇ , ∇β α̇ ] = 0,

(3.7)
{∇iα̇ , ∇jβ̇ } + {∇iβ̇ , ∇jα̇ } = 0,
Twistor Geometry and Gauge Theory 261

where we have defined the covariant derivatives

∇αα̇ ≡ ∂αα̇ + Aαα̇ and ∇iα̇ ≡ ∂α̇i + Aiα̇ . (3.8)

In particular, the field content of N = 4 self-dual super Yang-Mills (SYM) theory consists of a
◦ ◦ ◦ ◦
(self-dual) gauge potential Aαα̇ , four positive chirality spinors χiα , six scalars W ij = −W ji , four
◦ ◦
eiα̇ and an anti-self-dual two-form Gα̇β̇ , all in the adjoint representation
negative chirality spinors χ
of GL(n, C). The circle refers to the lowest component in the superfield expansions of the
corresponding superfields Aαα̇ , χiα , W ij , χ
eiα̇ and Gα̇β̇ , respectively. Now we follow the literature
and impose the transversal gauge,
ηiα̇ Aiα̇ = 0, (3.9)
in order to remove the superfluous gauge degrees of freedom associated with the fermionic co-
ordinates ηiα̇ . Therefore, we expand Aαα̇ and Aiα̇ with respect to the odd coordinates according
to
◦ ◦
Aαα̇ = Aαα̇ + ǫα̇β̇ χiα ηiβ̇ + · · · ,
◦ ◦
ij β̇
Aiα̇ = ǫα̇β̇ W ηj ekγ̇ ηlγ̇ ηjβ̇ −
+ 43 ǫijkl ǫα̇β̇ χ (3.10)
◦ ◦ ◦
− 65 ǫijkl ǫα̇β̇ (Gγ̇ δ̇ δlm + ǫγ̇ δ̇ [W mn δ̇ β̇
, W nl ])ηkγ̇ ηm ηj + · · · ,
◦ ◦
where W ij ≡ 21 ǫijkl W kl
. Therefore, (3.7) is equivalent to
◦
f α̇β̇ = 0,
◦ ◦
ǫαβ ∇αα̇ χiβ = 0,
◦ ◦ ◦ ◦
W ij
= −ǫαβ {χiα , χjβ }, (3.11)
◦ ◦ ◦ ◦
ǫα̇β̇ ∇αα̇ χ
eiβ̇ = 2[W ij , χjα ],
◦ ◦ ◦ ◦ ◦ ◦ ◦
ǫα̇β̇ ∇αα̇ Gβ̇ γ̇ = {χiα , χ
eiγ̇ } − 21 [W ij , ∇αγ̇ W ij
],

which are the N = 4 self-dual SYM equations. The equations for less supersymmetry are
◦
obtained from those by suitable truncations. In (3.11), f α̇β̇ denotes the anti-self-dual part of
the curvature
◦ ◦ ◦ ◦
[∇αα̇ , ∇β β̇ ] = ǫαβ f α̇β̇ + ǫα̇β̇ f αβ
◦ ◦ ◦
and  ≡ 21 ǫαβ ǫα̇β̇ ∇αα̇ ∇β β̇ . Note that the system (3.11) takes the same form for the superfields
Aαα̇ , χiα , W ij , χ
eiα̇ and Gα̇β̇ . We remark also that the field equations for N < 4 are simply
obtained by suitable truncations of (3.7) and (3.11), respectively.
The equations (3.5) can be integrated to give the formulas
I I
1 A+α 1 Ai+
Aαα̇ = dλ+ and Aiα̇ = dλ+ (3.12)
2πi λ+ λα̇
+ 2πi λ+ λα̇
+
C C

where the contour C = {λ+ ∈ CP 1 | |λ+ | = 1} encircles λ+ = 0. These contour integrals give
the explicit form of the Penrose-Ward transform.
Altogether, we may write down the following theorem:
262 Wolf

Theorem 3.1. There is a one-to-one correspondence between equivalence classes of holomorphic

vector bundles over the supertwistor space P 3|N which are holomorphically trivial on any projec-
tive line CPx,η
1
֒→ P 3|N and gauge equivalence classes of solutions to the N -extended self-dual
SYM equations. In other words, we have the bijection

Mhol (P 3|N ) ∼
= MN
SDYM ,

where MN SDYM denotes the moduli space of solutions to (3.11), which is obtained from the solution
space by factorizing with respect to the group of gauge transformations.11

Remark 3.2. In (1.32a) and (1.33) we have introduced the real structure τ . Imposing the
condition
f+− (x, η, λ) = f+− (τ (x, η, λ))†
on the transition function of E → P 3|N , where dagger means Hermitian conjugation, we even-
tually obtain real SYM fields.

3.3 Penrose-Ward transform II

Above we have chosen a trivialization of E → P 3|N which was convenient for the derivation of
the N -extended self-dual SYM equations. However, there are other convenient trivializations
of E over U± such that the compatibility conditions of the corresponding linear system are
described by holomorphic Chern-Simons (hCS) theory on the supertwistor space. Namely, since
restrictions of the bundle E to 2|N -dimensional leaves of the fibration (1.26) are trivial, there
exist matrix-valued functions ψb± on U± such that
−1
f+− = ψ+ ψ− = ψb+
−1 b
ψ− , with V̄±i ψb± = 0. (3.13)

From (3.13) it then follows that

ϕ ≡ ψ+ ψb+
−1
= ψ− ψb−
−1
(3.14)

is a matrix-valued function generating a gauge transformation

ψ± 7→ ψb± = ϕ−1 ψ± ,
A± b±
α 7→ Aα = ϕ Aα ϕ + ϕ−1 V̄α± ϕ = ψb± V̄α± ψb±
−1 ± −1
,
(3.15)
0 = Aλ̄± 7→ Abλ̄± = ϕ ∂λ̄± ϕ = ψb± ∂λ̄± ψb± ,
−1 −1

Ai± 7→ Abi± = ϕ−1 Ai± ϕ + ϕ−1 V̄±i ϕ = ψb± V̄±i ψb±

−1
= 0.

Thus, we obtain the linear system

(V̄α± + Ab± b
α )ψ± = 0,

(∂λ̄± + Abλ̄± )ψb± = 0, (3.16)

V̄±i ψb± = 0,

which is gauge equivalent to the linear system (3.6).

11 Note that gauge transformations of the gauge potential A are induced by transformations of the form ψ 7→

C
±
g −1 ψ± for some smooth GL(n, )-valued g. Under such transformations the transition function f+− does not
change.
Twistor Geometry and Gauge Theory 263

The compatibility conditions of the linear differential equations (3.16) are the field equations
of hCS theory on the supertwistor space P 3|N . On U± , they read as

V̄α± Ab± ± b± b± b±
β − V̄β Aα + [Aα , Aβ ] = 0,
(3.17)
∂λ̄± Ab± ± b b b±
α − V̄α Aλ̄± + [Aλ̄± , Aα ] = 0.

As in the preceding section, we now have to find the explicit superfield expansions of the compo-
nents Ab± b
α and Aλ̄± , respectively. However, their form is fixed by the geometry of the supertwistor
space. Recall that Ab± b
α and Aλ̄ are sections of the bundles O(1) and O(−2). Together with
±

the fact that the ηi± take values in the bundle ΠO(1), this determines the dependence of Ab±
α
and Abλ̄± on λα̇ bα̇ α̇
± and λ± ≡ τ (λ± ) (of course, only up to gauge transformations generated by
group-valued functions which may depend on λ± and λ̄± ). In the N = 4 case, this dependence
takes the form

Ab± α̇ ± i 1 bα̇ W ij + γ 2
ηi± ηj± λ 1
ηi± ηj± ηk± λ bβ̇ eijk +
bα̇ λ
α = λ± Aαα̇ + ηi χα + γ± 2! ± αα̇ ± 3! ± ±χ αα̇β̇
3
+ γ± 1 bα̇ λ
ηi± ηj± ηk± ηl± λ bβ̇ bγ̇ ijkl (3.18)
4! ± ± λ± Gαα̇β̇ γ̇ ,

Abλ̄± = ±(γ±
2 1
ηi± ηj± W ij + γ±
3 1 bα̇ χ
ηi± ηj± ηk± λ ijk 4 1 bα̇ λ
ηi± ηj± ηk± ηl± λ bβ̇ ijkl
2! 3! ± eα̇ + γ± 4! + ± Gα̇β̇ ),

where γ± ≡ (1 + λ± λ̄± )−1 . Here, Aαα̇ , χiα , W ij , χ

eα̇ i ≡ 1
ejkl
3! ǫijkl χ α̇ and Gα̇β̇ ≡ 1 ijkl
4! ǫijkl Gα̇β̇
is the field content of N = 4 self-dual SYM theory.12 The remaining coefficient fields are not
independent degrees of freedom, they are rather composite expressions,

Wαijα̇ = −∇αα̇ W ij , eijk

χ αα̇β̇
eijk
= − 21 ∇α(α̇ χ β̇)
and Gijkl
αα̇β̇ γ̇
= − 31 ∇α(α̇ Gijkl
β̇ γ̇)
, (3.19)

which follow upon substituting the field expansions (3.18) into the second equation of (3.17).
The field expansions (3.18), together with the first equation of (3.17), eventually reproduce
(3.11).
Therefore, we may summarize:

Theorem 3.3. There is a one-to-one correspondence between gauge equivalence classes of solu-
tions to the N -extended self-dual SYM equations on R4 and equivalence classes of holomorphic
vector bundles over the supertwistor space P 3|N which are holomorphically trivial on any pro-
jective line CPx,η
1
֒→ P 3|N , i.e., the moduli spaces MNhCS of hCS theory on P
3|N
and MNSDYM
are bijective,
MN ∼ N
hCS = MSDYM .

Remark 3.4. In fact, in this and in the preceding sections we have demonstrated the equiva-
lence of the Cech and the Dolbeault descriptions of holomorphic vector bundles (obeying certain
triviality conditions) over the supertwistor space. That is, we have shown the equivalence of the
data
¯ ∼ (Ee → P 3|N , fe+− = 1n , ∂¯A ≡ ∂¯ + A0,1 ),
(E → P 3|N , f+− , ∂)
where ∂¯ is the anti-holomorphic part of the exterior derivative on P 3|N . We should stress that
the bundles E and Ee belong to the same equivalence class in the category of smooth vector bundles
but not in the category of holomorphic vector bundles.

12 Here, by a slight abuse of notation, we have omitted the circles on top of the letters (cf. above).
264 Wolf

3.4 Comment on N = 4
Let us now fix N = 4. From the second lecture we know that in this case the supertwistor space
becomes Calabi-Yau. In particular, this means that the holomorphic Berezinian line bundle is
trivial ensuring the existence of a globally well defined holomorphic volume form Ω.13 On the
patches U± of P 3|4 , it is given by
1
Ω|U± = ±dz± 2
∧ dz± 3
∧ dz± dη1± dη2± dη3± dη4± . (3.20)

Then we may consider a trivial rank n complex vector bundle with connection one-form A and
furthermore introduce the action functional
Z

ShCS = ¯ 0,1 + 2 A0,1 ∧ A0,1 ∧ A0,1 ,
Ω ∧ tr A0,1 ∧ ∂A (3.21)
3
Y

where Y is the submanifold of P 3|4 constrained by η̄i± = 0. Moreover, it is assumed that A0,1 ,
which is the (0, 1)-part of A, does not contain components in the anti-holomorphic fermionic
directions and that its other components depend holomorphically on the ηi± . The equations of
motion of this theory are readily obtained to be
¯ 0,1 + A0,1 ∧ A0,1 = 0,
F 0,2 = ∂A (3.22)

which turn the vector bundle into a holomorphic vector bundle. In components, the equations
(3.22) are given by (3.17).
Finally, substituting the field expansions (3.18) into (3.21), integrating over the fermionic
coordinates and over the Riemann sphere, we end up with
Z n o
S = d4 x tr Gα̇β̇ fα̇β̇ + χiα ∇αα̇ χ
eα̇
i + 1
2 W ij W ij
+ W ij {χ i
α , χjα
} . (3.23)

From this functional, one easily derives the equations of motion (3.11) of the N = 4 self-dual
SYM theory. Therefore, the Calabi-Yau property of P 3|4 allows to define an action functional
of hCS theory which eventually gives an action functional of N = 4 self-dual SYM theory.

– Exercises –
Exercise 3.1. Find the superfield expansions (3.10). You may find it useful to introduce the
recursion operator D = ηiα̇ ∇iα̇ = ηiα̇ ∂α̇i . See also gauge condition (3.9).
Exercise 3.2. Prove (3.19).
Exercise 3.3. Derive (3.23).
Exercise 3.4. Put in (3.11) all fields – besides the gauge potential – to zero. Then we are left
with fα̇β̇ = 0 together with the linear system λα̇
± ∇αα̇ ψ± = 0. We know from the second lecture
that I
1
G0α̇β̇ = dλγ̇± λ± ± ±
γ̇ λα̇ λβ̇ f+− ,
2πi
C
13 We stress that since Ω is a section of the holomorphic Berezinian line bundle, it is an integral form rather
than a differential form.
Twistor Geometry and Gauge Theory 265

where f = {f+− } ∈ H 1 (P 3 , O(−4) ⊗ gl(n, C)), solves ∂α α̇ G0α̇β̇ = 0. Find a similar contour
integral for Gα̇β̇ such that ∇α α̇ Gα̇β̇ = 0.

Acknowledgments
It is a pleasure to thank S. Petersen, A. D. Popov, E. Schrohe and R. Wimmer for com-
menting on the manuscript. I am grateful to the organizers of the Modave Summer School on
Mathematical Physics for the invitation to give these lectures.

Appendices

A Categories
A category C consists of the following data:
(i) a collection Ob(C ) of objects,
(ii) sets Mor(X, Y ) of morphisms for each pair X, Y ∈ Ob(C ), including a distinguished iden-
tity morphism id = idX ∈ Mor(X, X) for each X,
(iii) a composition of morphism function ◦ : Mor(X, Y ) × Mor(Y, Z) → Mor(X, Z) for each
triple X, Y, Z ∈ Ob(C ) satisfying f ◦ id = f = id ◦ f and (f ◦ g) ◦ h = f ◦ (g ◦ h).
There are many examples of categories, such as:
(i) The category of affine spaces over some field F consists of all affine spaces over F (=
objects). The morphisms are linear maps.
(ii) The category of topological spaces consists of all topological spaces (= objects). The
morphisms are continuous maps.
(iii) The category of C k -manifolds consists of all C k -manifolds (= objects). The morphisms
are C k -maps.
(iv) The category of Lie algebras over some field F consists of all Lie algebras over F (= objects)
and the morphisms are those linear maps respecting the Lie bracket.
(v) The category of Lie groups consists of all Lie groups (= objects). The morphisms are the
Lie morphisms, which are group morphisms and C ∞ .
Besides the notion of categories we need so-called functors which relate different categories. A
functor F from a category C to another category D assigns to each object X ∈ Ob(C ) an object
F (X) ∈ Ob(D) and to each morphism f ∈ Mor(X, Y ) a morphism F (f ) ∈ Mor(F (X), F (Y )),
such that F (id) = id and F (f ◦ g) = F (f ) ◦ F (g). Pictorially, we have
f
X −−−−→ Y
 
 
Fy yF
F (X) −−−−→ F (Y )
F (f )

What we have just defined is a covariant functor. A contravariant functor differs from the
covariant functor by assigning to f ∈ Mor(X, Y ) the morphism F (f ) ∈ Mor(F (Y ), F (X)) with
F (id) = id and F (f ◦ g) = F (g) ◦ F (f ). The standard example is the so-called dual vector space
functor. This functor assigns to a vector space V over a field F the dual vector space F (V ) = V ∗
and to each linear map f : V → W the dual map F (f ) = f ∗ : W ∗ → V ∗ , with ω 7→ ω ◦ f and
ω ∈ W ∗ , in the reverse direction. Hence, it is a contravariant functor.
266 Wolf

B Sheaves
Let X be a topological space. A presheaf S of Abelian groups on X consists of the following
data:

(i) for all open subsets U ⊆ X, an Abelian group S(U ) and

(ii) for all inclusions V ⊆ U of open subsets of X, a morphism of Abelian groups rVU : S(U ) →
S(V )

subject to the conditions:

(a) r(∅) = 0,
U
(b) rU = idU : S(U ) → S(U ) and
(c) W ⊆ V ⊆ U , then rVU ◦ rW
V U
= rW .

Put it differently, a presheaf is just a contravariant functor from the category Top(X) (objects
= open sets of X, morphisms = inclusion maps) to the category Ab of Abelian groups. In fact,
we can replace Ab by any other fixed category C . Let S be a presheaf on X, then S(U ) are
the sections of S over U . Sometimes we shall write Γ(U, S) instead of S(U ). The rVU are called
restriction maps.
A presheaf S on a topological space is a sheaf if it satisfies:

(d) U open, {Va } open covering of U , s ∈ S(U ) such that rVUa (s) = 0 for all a, then s = 0,
(e) U open, {Va } open covering of U , sa ∈ S(Va ), with rVUa ∩Vb (sa ) = rVUa ∩Vb (sb ), then there
exists an s ∈ S(U ) such that rVUa (s) = sa for all a.

Point (d) says that any section is determined by its local behavior while (e) means that local
sections can be pieced together to give global sections.
Examples are:

(i) Let X be a real manifold and U ⊆ X. Then C ∞ (U ), Ωp (U ), X (U ), etc. are presheaves,

where the restriction mappings are the usual restriction mappings. In fact, they are sheaves.
(ii) Similar stuff as (i) but on complex manifolds.
(iii) Let R be a ring, X a topological space and U ⊆ X. Then R(U ) is the ring of locally
constant continuous functions on U . This determines a presheaf that is a sheaf. We call
this the constant sheaf R on X, e.g., R = R, C, . . ..
(iv) B(U ) – the bounded holomorphic functions on an open subset U of C. Then U → B(U ) is
a presheaf but not a sheaf.

A morphism of (pre)sheaves φ : A → B consists of a morphism of Abelian groups φU :

A(U ) → B(U ) for all open subsets U such that whenever V ֒→ U is an inclusion, the diagram
φU
A(U ) −−−−→ B(U )
 
U  ′U
rV y yr V
A(V ) −−−−→ B(V )
φV

is commutative. An isomorphism is a morphism that has a two-sided inverse. A typical example

is the de Rham complex on a real manifold, where the sheaf morphism is the usual exterior
derivative.
Twistor Geometry and Gauge Theory 267

Let S be a presheaf and

[
˙
Sex = S(U ).
U ∋x

Then we say that two elements of Sex , f ∈ S(U ), U ∋ x and g ∈ S(V ), V ∋ x, are equivalent if
there exists and open set W ⊂ (U ∩ V ), with x ∈ W such that
U V
rW (f ) = rW (g).

We define the stalk Sx to be the set of equivalence classes induced by this equivalence relation.
Of course, Sx inherits the algebraic structure of the presheaf S, i.e., we can add elements in Sx
by adding representatives of equivalence classes. We shall let

π ≡ rxU : S(U ) → Sx

be the natural restriction mapping to stalks.14

Finally, let φ : A → B a morphism of sheaves on a topological space X. Then φ is an
isomorphism if and only if the induced map on the stalks φx : Ax → Bx is an isomorphism for
all x ∈ X. The proof of this assertion can be found in the literature cited in the bibliography.
Note that this is not true for presheaves.

References
[1] B. DeWitt, Supermanifolds, Cambridge University Press, Cambridge, 1992.
[2] P. Griffith and J. Harris, Principles of algebraic geometry, John Wiley & Sons, 1978.
[3] S. A. Huggett and K. P. Tod, Introduction to twistor theory, Cambridge University Press,
Cambridge, 1994.
[4] C. R. LeBrun, Geometry of twistor spaces, Simons workshop lecture, July 2004.
[5] Yu. I. Manin, Gauge field theory and complex geometry, Springer Verlag, 1988.
[6] L. J. Mason and N. M. J. Woodhouse, Integrability, self-duality, and twistor theory, Claren-
don Press, Oxford, 1996.
[7] R. Penrose, The twistor program, Rept. Math. Phys. 12, 65 (1977).
[8] R. Penrose and W. Rindler, Spinors and space-time, Vol. 1: Two spinor calculus and rel-
ativistic fields, Cambridge University Press, Cambridge, 1984; Spinors and Space-Time,
Vol. 2: Spinor and twistor methods in space-time geometry, Cambridge University Press,
Cambridge, 1985.
[9] A. D. Popov, Self-dual Yang-Mills: Symmetries and moduli space, Rev. Math. Phys. 11,
1091 (1999) [hep-th/9803183].
[10] A. D. Popov and C. Saemann, On supertwistors, the Penrose-Ward transform and N = 4
super Yang-Mills theory, Adv. Theor. Math. Phys. 9, 931 (2005) [hep-th/0405123].
[11] R. S. Ward, On self-dual gauge fields, Phys. Lett. A 61, 81 (1977).
14 Note that a stalk is similar to the fiber of a vector bundle and is the localization of the information of S near
the point x.
268 Wolf

[12] R. S. Ward and R. O. Wells, Twistor geometry and field theory, Cambridge University Press,
Cambridge, 1990.

[13] R. O. Wells, Complex manifolds and mathematical physics, Bul. Amer. Math. Soc. 2, 296
(1979).

[14] R. O. Wells, Differential analysis on complex manifolds, Springer Verlag, 1980.

[15] E. Witten, Perturbative gauge theory as a string theory in twistor space, Commun. Math.
Phys. 252, 189 (2004) [hep-th/0312171].
 Moyal’s Star Product

Sandrine Cnockaert

Physique Théorique et Mathématique, Université Libre de Bruxelles

International Solvay Institutes,
U.L.B. Campus Plaine C.P. 231, B-1050, Bruxelles, Belgium

E-mail: [email protected]

Abstract. The aim of this lecture is to provide a basic introduction to Moyal’s star
product and to some of its representations.

Based on the lectures presented by S. Cnockaert at the First Modave Summer School in
Mathematical Physics held at Modave, Belgium on June 19-25, 2005
270 Cnockaert

1 Introduction
The Moyal star product is an associative deformation of the usual product law. It was first intro-
duced during the early developments of quantum mechanics [1, 2], when people were wondering
about a possible statistical meaning of quantum mechanics and the relation between physical
quantities a and quantum mechanical operators a. The physical quantities are functions of the
commuting phase space variables {q µ , pν } (µ, ν = 1, . . . d), while the quantum operators are
functions of the non-commuting operators {qµ , pν }, that satisfy the commutation relation

[qµ , pν ] = i~δνµ , [qµ , qν ] = 0 = [pµ , pν ] .

The star product appears in this framework as a tool to express quantum laws in terms of
commuting variables. Considering two functions a(q, p), b(q, p) and their operator counterparts
a(q, p), b(q, p), it is obvious that the usual product ab does not reproduce the commutation
relations of the operators: e.g. [a, b] = ab − ba = 0 while [a, b] does not necessarily vanish
(generically it vanishes only in the classical limit ~ → 0). A new product law, the Moyal star
product ⋆ , was introduced, such that a ⋆ b would correspond to the operator product ab . The
⋆-commutator [a, b]⋆ = a ⋆ b − b ⋆ a then corresponds to [a, b] .
The star product is thus a product that allows to rewrite in terms of commuting coordinates
theories depending on non-commuting ones. This property is used in noncommutative field
theories. In these theories, the space-time coordinates xµ no longer commute, they satisfy the
commutation relations [xµ , xν ] = iθµν . See [3] for a review.
A more fundamental use of the star product can be found in string field theory and higher
spin field theories. Indeed, in string field theory, a new producet, called Witten’s star product,
is used to describe the interactions of strings. It can be shown that this product is related to
the Moyal star product [4]. In the domain of higher spins, an interacting theory can be build,
which also uses the Moyal star product (see [5] for a pedagogical (or trying to be pedagogical)
introduction and for more precise references).
Finally, the Moyal star product also appears in the context of deformations of Lie algebras [6]
and probably in many other areas, we won’t try to be exhaustive.
This lecture is organized as follows. Section 2 presents the problem of ordering in quantum
mechanics and introduces some notations. The three following sections provide different repre-
sentations of the Moyal star product: the standard representation, a matrix-like representation
and an integral representation, which are used in the various contexts presented above.

2 Ordering and notations

The correpondance between a function (also called symbol) a(q, p) and an operator a(q, p) that
yields a(q, p) in the classical limit ~ → 0 is not unique. For example, the function a = pq
is the classical limit of the operators a1 = pq , a2 = qp , a3 = 21 (qp + pq) and many others.
Fortunately, the correspondance can be made one-to-one if a specific ordering for the operators
qµ , pν is chosen.
Some orderings are more useful than others, which is the case of the normal ordering and
the Weyl ordering [7].
• normal ordering: the symbol a(q, p) is mapped onto the operator a(q, p) that has all p’s
to the right. e.g. the symbol a = pq corresponds to the operator a2 = qp .
• Weyl ordering: the symbol a(q, p) is mapped onto the Weyl-ordered operator a(q, p), which
is the sum over all possible orders for the operators q, p, with equal weight. Some Weyl-
Moyal’s Star Product 271

ordered operators are a3 = 12 (qp + pq) or b = 13 (qpp + pqp + ppq) . More precisely, the
Weyl map is the bijection between symbols a(q, p) and Weyl-ordered operators aW (q, p)
defined (formally) by
Z
µ µ
W : a(q, p) → aW (q, p) = e a(τ, σ) ei(q τµ +pµ σ ) dd τ dd σ ,

1
R µ
τµ +pµ σ µ )
where e
a(τ, σ) = (2π)2d
a(q, p) e−i(q dd qdd p . The power series expansion of
i(qµ τµ +pµ σ µ )
e automatically symmetrizes the q and p: the products of these operators
appears exactly once in each possible order.

When hermiticity is important, the Weyl ordering is used, as the Weyl-ordered operators are
automatically hermitian. Except when explicitely stated, we use the Weyl ordering in the sequel.
To make the link with noncommutative field theory, the operators of which are the coor-
dinates xµ satisfying [xµ , xν ] = iθµν , we introduce a new notation for {q, p}. The coordinate
Xiµ is defined by Xiµ ≡ (q µ , pµ ), where i = (1, 2) refers to (q, p) respectively. The greek indices
are raised and lowered with the Minkowski metric η µν and its inverse. The invariant Sp(2, R)
metric ǫij such that ǫij = −ǫji and ǫ12 = ǫ12 = 1 is used to raise and lower the latin indices:
ai = ǫij aj , ai = aj ǫji . We furthermore introduce the notation X m ≡ Xiµ , where m denotes
collectively the upper index µ and the lower index i. Xm denotes the corresponding operators.
In terms of this variable, the commutation relations for q, p read

[Xm , Xn ] = iθmn

where θmn = ~η µν ǫij . In the sequel, unless specified otherwise, X m can denote either the phase
space coordinates {q, p}, or the symbols xµ of the coordinates of some noncommutative space-
time. The index m then takes respectively n = 2d or n = d values.
The Weyl map now reads
Z
m
W : a(X) → aW (X) = e a(τ ) eiX τm dn τ ,

1
R m
where e
a(τ ) = (2π)n a(q, p) e−iX τm
dn X .

3 Standard Moyal Star Product

The standard expression for the Moyal star product is
←
− →
−
i mn ∂ ∂
(a ⋆ b)(X) = a(X)e 2 θ ∂X m ∂X n b(X) . (3.1)

It is defined by the property that it reproduces, at the level of their symbols, the usual product
of operators:
W (a ⋆ b) = W (a)W (b) . (3.2)
Before proving this property for the star product (3.1), let us prove a preliminary lemma:
Lemma: (simple case of the Baker-Campbell-Haussdorff formula) If the commutator C =
[A, B] commutes with A and B, then
1 1
eA eB = eA+B+ 2 [A,B] = eA+B e 2 [A,B] .
272 Cnockaert

1
Proof of the lemma: Let us define Y (t) = eA+t(B+ 2 C) . Then

d  d X 1 h in 
∞
1
Y (t) = A + t(B + C))
dt dt n=0 n! 2

1 hX i
X∞ n−1
1 1 1
= (A + t(B + C))n−k−1 (B + C)(A + t(B + C))k
n=0
n! 2 2 2
k=0

1 Xh i
X∞ n−1
1 1 1
= (A + t(B + C))n−1 (B + C) + k(A + t(B + C))n−2 [B, A]
n=0
n! 2 2 2
k=0

1h i
X∞
1 1 n(n − 1) 1
= n(A + t(B + C))n−1 (B + C) − (A + t(B + C))n−2 C
n=0
n! 2 2 2 2
1 1 1 1 1
= eA+t(B+ 2 C) (B + C) − eA+t(B+ 2 C) C = eA+t(B+ 2 C) B
2 2
= Y (t)B .
d
The solution of dt Y (t) = Y (t)B with the initial condition Y (t = 0) = eA is Y (t) = eA etB .
Equating the two expressions for Y (t = 1) proves the lemma. QED.
m n m
(τm +σm ) − 2i θ mn τm σn
In particular, by this lemma, eiX τm
eiX σn
= eiX e .

Proof that (3.1) satisfies (3.2):

Z ←
− →
−
m i mn ∂ ∂ m
a⋆b = a(τ ) eiX τm e 2 θ ∂X m ∂X n eb(σ) eiX σm
dn τ d n σ e
Z hX ←
− − →
∂ k i iX m σm
∞
m 1 i mn ∂
= a(τ ) eb(σ) eiX
dn τ d n σ e τm
( θ ) e
k! 2 ∂X ∂X n
m
k=0
Z hX
∞ i
m 1 i mn m
= a(τ ) eb(σ) eiX τm
dn τ d n σ e ( θ iτm iσn )k eiX σm
k! 2
k=0
Z
iX m (τm +σm ) − 2i θ mn τm σn
= n
a(τ ) eb(σ) e
n
d τd σ e e , (3.3)

which implies that the operator corresponding to a ⋆ b by the Weyl map is

Z
m i mn
W (a ⋆ b) = a(τ ) eb(σ) eiX (τm +σm ) e− 2 θ τm σn
dn τ dn σ e
Z
m m
= a(τ ) eb(σ) eiX τm eiX σm
dn τ dn σ e

= W (a)W (b),

QED.
To conclude this section, we illustrate the fact that the star product depends on the ordering
chosen for the operators, by giving explicitely the expression of the Moyal star product in terms
of the phase space variables, in the Weyl and normal ordering.
In the Weyl ordering, the Moyal star product (3.1) reads
←
− →
− ←
− →
−
i~
( ∂q∂µ ∂ ∂
− ∂p ∂
∂q µ )
(a ⋆ b)(q, p) = a(q, p) e 2 ∂pµ µ b(q, p) .
Moyal’s Star Product 273

On the other hand, the appropriate star product in the normal ordering is
←
− →
−
∂ ∂
−i~ ∂p ∂q µ
(a ⋆ b)(q, p) = a(q, p) e µ b(q, p) . (3.4)

The proof of this statement is left as an exercise, but let us illustrate this star product by two
examples.
e.g.1: if b = b(p), then (3.4) gives a ⋆ b = ab, which is coherent with (ab)N.O. = ab (where
N.O. stands for normal-ordered).
←
− →
−
e.g.2: if a = q µ , b = pν , then [q µ , pν ]⋆ = q µ pν − pν (1 − i~ ∂p∂ ρ ∂q∂ ρ )q µ = q µ pν − pν q µ + i~δνµ =
i~δνµ .

4 Moyal star product as a matrix-like multiplication

In this section, we show that the representation of the Moyal star product in the space of “half-
Fourrier transformed” symbols A e has the structure of matrix multiplication. This is done in the
phase space coordinates. In the sequel, ~ is set equal to one.

For any symbol a(q, p), let us define the “half-Fourrier transform” Ae by
Z Z
−iy µ pµ µ
e µ , rµ )
µ
a(q , pµ ) = d y ed
A(q , y ) = dd y e−iy pµ A(l
µ µ

yµ yµ
where lµ = q µ + 2 , rµ = q µ − 2 .
e µ , rµ ) is related to A(l
If (a ⋆ b)(q, p) = c(q, p), then C(l e µ , rµ ) and B(l
e µ , rµ ) by a matrix-like
multiplication with continuous indices
Z
e µ , rµ ) = dd y A(l
C(l e µ , y µ )B(y
e µ , rµ ) . (4.1)

Proof:
Z ←
− →− ←
− → − ′ ′
a⋆b = e + y , q − y ) e−iy·p e 2i ( ∂q
dd ydd y ′ A(q
∂ ∂ ∂ ∂ ′
e + y ,q − y )
∂p − ∂p ∂q ) e−iy ·p B(q
2 2 2 2
Z ←− →
− ′ ′
= e + y , q − y )e 2 ( ∂q y −y ∂q ) B(q
dd ydd y ′ e−iy·p A(q
1 ∂ ′ ∂
e + y , q − y )e−iy ·p ′

2 2 2 2
Z ′ ′ ′
′
e + y + y y − y e − y − y y + y′
= dd ydd y ′ e−i(y+y )·p A(q ,q − ) B(q ,q − )
2 2 2 2
Z  
= e + y + , y− ) B
dd y+ dd y− e−ip·y+ A(q e y− , q − y+
2 2
Z
y
e + +,q − +). y
= dd y+ e−ip·y+ C(q
2 2
QED

The representation (4.1) of the star product is useful to make the link with Witten’s star
product of string field theory. Indeed, it has the same structure as the latter, expressed in terms
e (lµ [σ] , rµ [σ]) and B
of left and right halfs of strings, lµ [σ] and rµ [σ]. If A e (lµ [σ] , rµ [σ]) are two
1 1 2 2
274 Cnockaert

string fields, then Witten’s string star product is formally given by the functional integral [8]
Z
Ce (lµ [σ] , rµ (σ)) = [dz] A
e (lµ [σ] , z µ [σ]) B
e (z µ [σ] , rµ [σ]) .
1 2 1 2

The exact relation is somewhat more subtle and can be found in [4].

5 Integral representation of the Moyal star product

In some applications such as higher spin field theory (see e.g. [5] and references therein), the
following expression for the Moyal star product is used, in the particular phase space-like case
θmn = η µν ǫij , n = 2d:
Z
1 iµ j
(a ⋆ b)(X) = n dn Sdn T a(X + S) b(X + T ) e2iS Tµ ǫij (5.1)
π

Proof of the equivalence of (5.1) and (3.1): Let us first recall a useful representation
of the Dirac delta function: Z
1
δ(x) = dn p eixp .
(2π)n
We can now prove the equivalence:
Z
1 iµ j
(a ⋆ b)(X) = dn Sdn T a(X + S) b(X + T ) e2iS Tµ ǫij
πn
Z
1 µ µ i µ µ i iµ j
= n
dn Sdn T dn τ dn σ e a(τ ) eb(σ) ei(Xi +Si )τµ ei(Xi +Ti )σµ e2iS Tµ ǫij
π
Z Z
1 n n n e iXiµ (τµi +σµ i
) iTiµ σµ
i µ i i
= d Td τd σ e a(τ ) b(σ) e e dn S eiSi (τµ +2Tµ )
πn
Z
µ i i µ i
= 2n dn T dn τ dn σ e a(τ ) eb(σ) eiXi (τµ +σµ ) eiTi σµ δ(τµi + 2Tµi )
Z
µ i i i µ i
= a(τ ) eb(σ) eiXi (τµ +σµ ) e− 2 τi σµ .
dn τ d n σ e

The last line is the same as (3.3) for the chosen θmn , which proves the equivalence.

References
[1] H. J. Groenewold, Physica 12, 405 (1946).
[2] J. E. Moyal, Proc. Cambridge Phil. Soc. 45, 99 (1949).
[3] M. R. Douglas and N. A. Nekrasov, Rev. Mod. Phys. 73 (2001) 977 [arXiv:hep-th/0106048].
[4] I. Bars, Phys. Lett. B 517, 436 (2001) [arXiv:hep-th/0106157].
[5] X. Bekaert, S. Cnockaert, C. Iazeolla and M. A. Vasiliev, arXiv:hep-th/0503128.
[6] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz and D. Sternheimer, Annals Phys. 111
(1978) 61.
[7] H. Weyl, Z. Phys. 46, 1 (1927).
Moyal’s Star Product 275

[8] E. Witten, Nucl. Phys. B 268 (1986) 253.

 Noncommutative Spaces and Gravity

Frank Meyer

Max–Planck–Institut für Physik

(Werner-Heisenberg Institut)
Föhringer Ring 6, D-80805 München, Germany

Arnold Sommerfeld Center, Department für Physik

Ludwig-Maximilians-Universität München, Theresienstraße 37
D-80333 München, Germany

E-mail: [email protected]

Abstract. This lecture gives an introduction to an algebraic construction of a grav-

ity theory on noncommutative spaces which is based on a deformed algebra of (in-
finitesimal) diffeomorphisms. We start with some fundamental ideas and concepts of
noncommutative spaces. Then the θ-deformation of diffeomorphisms is studied and a
tensor calculus is defined. A deformed Einstein-Hilbert action invariant with respect
to deformed diffeomorphisms is given. Finally, all noncommutative fields are expressed
in terms of their commutative counterparts up to second order of the deformation pa-
rameter using the ⋆-product. This allows to study explicitly deviations to Einstein’s
gravity theory in orders of θ. This lecture is based on joined work with P. Aschieri, C.
Blohmann, M. Dimitrijević, P. Schupp and J. Wess.

Based on the lectures presented by F. Meyer at the First Modave Summer School in Mathe-
matical Physics held at Modave, Belgium on June 19-25, 2005
Noncommutative Spaces and Gravity 277

1 Noncommutative Spaces
In field theories one usually considers differential space-time manifolds. In the noncommutative
realm, the notion of a point is no longer well-defined and we have to give up the concept
of differentiable manifolds. However, the space of functions on a manifold is an algebra. A
generalization of this algebra can be considered in the noncommutative case. We take the
algebra freely generated by the noncommutative coordinates x bi which respects commutation
relations of the type

xµ , x
[b bν ] = C µν (b
x) 6= 0 . (1.1)
Without bothering about convergence, we take the space of formal power series in the coordinates
bi and divide by the ideal generated by the above relations
x

b
Abx = Chhb
x0 , . . . , x
bn ii/([b
xµ , x
bν ] − C µν (b
x)) .

The function C µν (b
x) is unknown. For physical reasons it should be a function that vanishes at
large distances where we experience the commutative world and may be determined by experi-
ments. Nevertheless, one can consider a power-series expansion

C µν (b
x) = iθµν + iC µν ρ x bµν ρσ − δρν δσµ )b
bρ + (q R xρ x
bσ + . . . ,

bµν ρσ are constants, and study cases where the commutation relations
where θµν , C µν ρ and q R
are constant, linear or quadratic in the coordinates. At very short distances those cases provide
a reasonable approximation for C µν (bx) and lead to the following three structures

(i) canonical or θ-deformed case:

xµ , x
[b bν ] = iθµν . (1.2)

(ii) Lie algebra case:

xµ , x
[b bν ] = iC µν ρ x
bρ . (1.3)

(iii) Quantum Spaces:

bµ x
x bµν ρσ x
bν = q R bρ x
bσ . (1.4)

We denote the algebra generated by noncommutative coordinates x bµ which are subject to the re-
b
lations (1.2) by A. We shall often call it the algebra of noncommutative functions. Commutative
functions will be denoted by A. In what follows we will exclusively consider the θ-deformed case
(1.2) but we note that the algebraic construction presented here can be generalized to more com-
plicated noncommutative structures of the above type which possess the Poincaré-Birkhoff-Witt
(PBW) property. The PBW-property states that the space of polynomials in noncommutative
coordinates of a given degree is isomorphic to the space of polynomials in the commutative co-
ordinates. Such an isomorphism between polynomials of a fixed degree is given by an ordering
prescription. One example is the symmetric ordering (or Weyl-ordering) W which assigns to
any monomial the totally symmetric ordered monomial

W : A → Ab
µ
x 7→ bµ
x (1.5)
1 µ ν
xµ xν 7 → (b bν x
b +x
x x bµ )
2
··· .
278 Meyer

To study the dynamics of fields we need a differential calculus on the noncommutative algebra
b Derivatives are maps on the deformed coordinate space [1]
A.

∂bµ : Ab → Ab .

This means that they have to be consistent with the commutation relations of the coordinates.
In the θ-constant case a consistent differential calculus can be defined very easily by1

[∂bµ , x
bν ] = δµν (∂bµ x
bν ) = δµν
[∂bµ , ∂bν ] = 0. (1.6)

It is the fully undeformed differential calculus. The above definitions yield the usual Leibniz-rule
for the derivatives ∂bµ
(∂bµ fbgb) = (∂bµ fb)b
g + fb(∂bµ gb). (1.7)
This is a special feature of the fact that θµν are constants. In the more complicated examples of
noncommutative structures this undeformed Leibniz-rule usually cannot be preserved but one
has to consider deformed Leibniz-rules for the derivatives [2]. Note that (1.6) also implies that

\
(∂bµ fb) = (∂ µ f ). (1.8)

The Weyl ordering (1.5) can be formally implemented by the map (see S. Cnockaert’s Modave
lecture for a detailed discussion)
Z
f 7→ W (f ) =
1
n
µ
dn k eikµ x fe(k) b
(2π) 2

where fe is the Fourier transform of f

Z
1 µ
fe(k) = n dn x e−ikµ x f (x).
(2π) 2

This is due to the fact that the exponential is a fully symmetric function. Using the Baker-
Campbell-Hausdorff formula one finds
b µ
bν
eikµ x eipν x = ei(kµ +pµ )x b
µ
− 2i kµ θ µν pν
. (1.9)

This immediately leads to the following observation

Z
b
f gb = W (f )W (g) =
1 µ ν
dn kdn p eikµ x eipν x fe(k)e g (p) b b
(2π)n
Z
=
1 µ i µν
dn kdn p ei(kµ +pµ )x e− 2 kµ θ pν fe(k)e b
g (p)
(2π)n
i µν
∂µ ⊗∂ν
= W (µ ◦ e 2 θ f ⊗ g), (1.10)

where µ(f ⊗ g) := f g is the multiplication map. With (1.8) we deduce from (1.10) the equation

µ ◦ e− 2 θ
i µν b b fb ⊗ gb = fcg.
∂ µ ⊗∂ ν
(1.11)
1 We use brackets to distinguish the action of a differential operator from the multiplication in the algebra of
differential operators.
Noncommutative Spaces and Gravity 279

The above formula shows us how the commutative and the noncommutative product are related.
It will be important for us later on.

2 Symmetries on Deformed Spaces

In general the commutation relations (1.1) are not covariant with respect to undeformed sym-
metries. For example the canonical commutation relations (1.2) break Lorentz symmetry if we
assume that the noncommutativity parameters θµν do not transform.
The question arises whether we can deform the symmetry in such a way that it acts consis-
tently on the deformed space (i.e. leaves the deformed space invariant) and such that it reduces
to the undeformed symmetry in the commutative limit. The answer is yes: Lie algebras can be
deformed in the category of Hopf algebras (Hopf algebras coming from a Lie algebra are also
called Quantum Groups)2 . Important examples of such deformations are q-deformations: Drin-
feld and Jimbo have shown that there exists a q-deformation of the universal enveloping algebra
of an arbitrary semisimple Lie algebra3 . Module algebras of this q-deformed universal enveloping
algebras are noncommutative spaces with commutation relations of type (1.4). There exists also
a so-called κ-deformation of the Poincaré algebra [3, 4] which leads to a noncommutative space
of the Lie type (1.3). A Hopf algebra symmetry acting on the θ-deformed space was for a long
time unknown. But recently also a θ-deformation of the Poincaré algebra leading to the algebra
(1.2) was constructed [5, 6, 7, 8].
Quantum group symmetries lead to new features of field theories on noncommutative spaces.
Because of its simplicity, θ-deformed spaces are very well-suited to study those. First results on
the consequences of the θ-deformed Poincaré algebra have already been obtained [6, 8]. However,
it remains unknown and subject of future investigations in which precise way this recently
discovered quantum group symmetry restricts the degrees of freedom of the noncommutative
field theory.
In the following we will construct explicitly a θ-deformed version of diffeomorphisms which
consistently act on the noncommutative space (1.2). Then we present a gravity theory which is
invariant with respect to this deformed diffeomorphisms. This construction is based on [8].

3 Diffeomorphisms
Gravity is a theory invariant with respect to diffeomorphisms. However, to generalize the Ein-
stein formalism to noncommutative spaces in order to establish a gravity theory, it is important
to first understand that diffeomorphisms possess more mathematical structure then the algebraic
one: They are naturally equipped with a Hopf algebra structure. In the common formulations
of physical theories this additional Hopf structure is hidden and does not play a crucial role. It
is our aim to deform the algebra of diffeomorphism in such a way that it acts consistently on a
noncommutative space. This can be done by exploiting the full Hopf structure. In this section
we first introduce the concept of diffeomorphisms as Hopf algebra in the undeformed setting.
Diffeomorphisms are generated by vector-fields ξ. Acting on functions, vector-fields are
represented as linear differential operators ξ = ξ µ ∂µ . Vector-fields form a Lie algebra Ξ over the
2 To be more precise the universal enveloping algebra of a semisimple Lie algebra can be deformed. A semisimple

Lie algebra cannot be deformed continuously within the set of semisimple Lie algebras since this is a discrete
set. But the rich Hopf algebra structure of the universal enveloping of a semisimple Lie algebra gives rise to
deformations in the category of Hopf algebras.
3 It is called q-deformation since it is a deformation in terms of a parameter q.
280 Meyer

field C with the Lie bracket given by

[ξ, η] = ξ × η

where ξ × η is defined by its action on functions

(ξ × η)(f ) = (ξ µ (∂µ η ν )∂ν − η µ (∂µ ξ ν )∂ν )(f ).

The Lie algebra of infinitesimal diffeomorphisms Ξ can be embedded into its universal enveloping
algebra which we want to denote by U(Ξ) (see X. Bekaert’s lecture for an introduction to universal
enveloping algebras). The universal enveloping algebra is an associative algebra and possesses a
natural Hopf algebra structure. It is given by the following structure maps4 :

• An algebra homomorphism called coproduct defined by

∆ : U(Ξ) → U(Ξ) ⊗ U(Ξ)

Ξ ∋ ξ 7→ ∆(ξ) := ξ ⊗ 1 + 1 ⊗ ξ. (3.1)

• An algebra homomorphism called counit defined by

ε : U(Ξ) → C
Ξ ∋ ξ 7→ ε(ξ) = 0. (3.2)

• An anti-algebra homomorphism called antipode defined by

S : U(Ξ) → U(Ξ)
Ξ∋ξ 7→ S(ξ) = −ξ. (3.3)

For a precise definition and more details on Hopf algebras we refer the reader to text books
[9, 10, 11]. For our purposes it shall be sufficient to note that the coproduct implements how the
Hopf algebra acts on a product in a representation algebra (Leibniz-rule). Below we will make
this more transparent. It is now possible to study deformations of U(Ξ) in the category of Hopf
algebras. This leads to a deformed version of diffeomorphisms - the fundamental building block
of our approach to a gravity theory on noncommutative spaces. Before studying this in detail,
let us shortly review the Einstein formalism. This way we first understand better the meaning
of the structure maps of a Hopf algebra introduced above.
Scalar fields are defined by their transformation property with respect to infinitesimal coor-
dinate transformations:
δξ φ = −ξφ = −ξ µ (∂µ φ). (3.4)
The product of two scalar fields is transformed using the Leibniz-rule

δξ (φψ) = (δξ φ)ψ + φ(δξ ψ) = −ξ µ (∂µ φψ) (3.5)

such that the product of two scalar fields transforms again as a scalar. The above Leibniz-rule
can be understood in mathematical terms as follows: The Hopf algebra U(Ξ) is represented on
the space of scalar fields by infinitesimal coordinate transformations δξ . On scalar fields the
4 The structure maps are defined on the generators ξ ∈ Ξ and the universal property of the universal enveloping

algebra U (Ξ) assures that they can be uniquely extended as algebra homomorphisms (respectively anti-algebra
homomorphism in case of the antipode S) to the whole algebra U (Ξ).
Noncommutative Spaces and Gravity 281

action of δξ is explicitly given by the differential operator −ξ µ ∂µ . Of course, the space of scalar
fields is not only a vector space - it possesses also an algebra structure - such as U(Ξ) is not
only an algebra but also a Hopf algebra - it possesses in addition the co-structure maps defined
above. We say that a Hopf algebra H acts on an algebra A (or more precisely we say that A is
a left H-module algebra) if A is a module with respect to the algebra H and if in addition for
all h ∈ H and a, b ∈ A

h(ab) = µ ◦ ∆h(a ⊗ b) (3.6)

h(1) = ε(h). (3.7)

Here µ is the multiplication map defined by µ(a ⊗ b) = ab. In our concrete example where
H = U(Ξ) and A is the algebra of scalar fields we indeed have that the algebra of scalar fields is
a U(Ξ)−module algebra. This can be seen easily if we rewrite (3.5) using (3.1) for the generators
ξ ∈ Ξ for U(Ξ):
δξ (φψ) = (δξ φ)ψ + φ(δξ ψ) = µ ◦ ∆ξ(φ ⊗ ψ).
It is also evident that
δξ 1 = 0 = ε(ξ)1.
Now we are in the right mathematical framework: We study a Lie algebra (here infinitesimal
diffeomorphisms Ξ) and embed it in its universal enveloping algebra (here U(Ξ)). This universal
enveloping algebra is a Hopf algebra via a natural Hopf structure induced by (3.1,3.2,3.3).

Physical quantities live in representations of this Hopf algebras. For instance, the algebra of
scalar fields is a U(Ξ)-module algebra. The action of U(Ξ) on scalar fields is given in terms of
infinitesimal coordinate transformations δξ .

Similarly one studies tensor representations of U(Ξ). For example vector fields are introduced
by the transformation property

δξ Vα = −ξ µ (∂µ Vα ) − (∂α ξ µ )Vµ

δξ V α = −ξ µ (∂µ V α ) + (∂µ ξ α )V µ .

The generalization to arbitrary tensor fields is straight forward:

···µn ···µn ···µ
δξ Tνµ11···ν n
= −ξ µ (∂µ Tνµ11···νn
) + (∂µ ξ µ1 )Tνµ···µ n
1 ···νn
+ · · · + (∂µ ξ µn )Tνµ11···ν n
µ1 ···µn µ1 ···µn
−(∂ν1 ξ ν )Tν···ν n
− · · · − (∂ ν n
ξ ν
)T ν 1 ···ν .

As for scalar fields, we also find that the product of two tensors transforms like a tensor.
Summarizing, we have seen that scalar fields, vector fields and tensor fields are representations
of the Hopf algebra U(Ξ), the universal enveloping algebra of infinitesimal diffeomorphisms. The
Hopf algebra U(Ξ) is acts via infinitesimal coordinate transformations δξ which are subject to
the relations:

[δξ , δη ] = δξ×η ε(δξ ) = 0

∆δξ = δξ ⊗ 1 + 1 ⊗ δξ S(δξ ) = −δξ . (3.8)

The transformation operator δξ is explicitly given by differential operators which depend on the
representation under consideration. In case of scalar fields this differential operator is given by
−ξ µ ∂µ .
282 Meyer

4 Deformed Diffeomorphisms

The concepts introduced in the previous subsection can be deformed in order to establish a
consistent tensor calculus on the noncommutative space-time algebra (1.2). In this context it is
necessary to account the full Hopf algebra structure of the universal enveloping algebra U(Ξ).

In our setting the algebra Ab possesses a noncommutative product defined by

xµ , x
[b bν ] = iθµν , (4.1)

We want to deform the structure maps (3.8) of the Hopf algebra U(Ξ) in such a way that
the resulting deformed Hopf algebra which we denote by U(Ξ) b consistently acts on A.
b In the
b b
language introduced in the previous section this means that we want A to be a U(Ξ)-module
algebra. We claim that the following deformation of U(Ξ) does the job. Let U(Ξ) b be generated
as algebra by elements δbξ , ξ ∈ Ξ. We leave the algebra relation undeformed and demand

[δbξ , δbη ] = δbξ×η (4.2)

but we deform the co-sector

∆δbξ = e− 2 hθ
i ρσ b b (δb
∂ ρ ⊗∂ σ
ξ ⊗ 1 + 1 ⊗ δbξ )e 2 hθ
i ρσ b b,
∂ ρ ⊗∂ σ
(4.3)

where
[∂bρ , δbξ ] = δb(∂ρ ξ) .
The deformed coproduct (4.3) reduces to the undeformed one (3.8) in the limit θ → 0. Antipode
and counit remain also undeformed

S(δbξ ) = −δbξ ε(δbξ ) = 0. (4.4)

We have to check whether the above deformation is a good one in the sense that it leads to
a consistent action on A.b First we need a differential operator acting on fields in Ab which
represents the algebra (4.2). Let us consider the differential operator

X∞
1 i
bξ :=
X (− )n θρ1 σ1 · · · θρn σn (∂bρ1 · · · ∂bρn ξbµ )∂bµ ∂bσ1 · · · ∂bσn . (4.5)
n=0
n! 2

This is to be understood like that: A vector-field ξ = ξ µ ∂µ is determined by its coefficient

functions ξ µ . In Section 1 we saw that there is a vectorspace isomorphism W from the space of
commutative to the space of noncommutative functions which is given by the symmetric ordering
prescription. The image of a commutative function f under the isomorphism W is denoted by fb

W : f 7→ W (f ) = fb.

In (4.5) ξbµ is therefore to be interpreted as the image of ξ µ with respect to W . Then indeed we
have
bξ , X
[X bη ] = X
bξ×η . (4.6)
Noncommutative Spaces and Gravity 283

b :
bξ φ)
To see this we use result (1.11) to rewrite (X

X∞
b = 1 i
bξ φ)
(X (− )n θρ1 σ1 · · · θρn σn (∂bρ1 · · · ∂bρn ξbµ )(∂bµ ∂bσ1 · · · ∂bσn φ)
b
n=0
n! 2
X∞
1 i
= (− )n θρ1 σ1 · · · θρn σn (∂bρ1 · · · ∂bρn ξbµ )(∂bσ1 · · · ∂bσn ∂d
µ φ)
n=0
n! 2

= ξ µ\ d
(∂µ φ) = (ξφ). (4.7)

From (4.7) follows

bξ (X
(X b − (X
bη φ)) bη (X
bξ φ)) \
b = ([ξ, b
bξ×η φ),
η]φ) = (X
which amounts to (4.6) and this is what we wanted to show.
It is therefore reasonable to introduce scalar fields φb ∈ Ab by the transformation property

δbξ φb = −(X b
bξ φ).

bξ on the product of two

The next step is to work out the action of the differential operators X
fields. A calculation [8] shows that

b = µ ◦ (e− 2i hθρσ ∂ρ ⊗∂σ (X

bξ (φbψ))
(X
b b bξ )e 2i hθρσ ∂ρ ⊗∂σ φb ⊗ ψ).
bξ ⊗ 1 + 1 ⊗ X b b b
This means that the differential operators Xbξ act via a deformed Leibniz rule on the product
of two fields. Comparing with (4.3) we see that the deformed Leibniz rule of the differential
operator Xbξ is exactly the one induced by the deformed coproduct (4.3):

δbξ (φbψ)
b b b b
b = e− 2i hθρσ ∂ρ ⊗∂σ (δbξ ⊗ 1 + 1 ⊗ δbξ )e 2i hθρσ ∂ρ ⊗∂σ (φbψ)
b = −X
bξ ⊲ (φbψ).
b

Hence, the deformed Hopf algebra U(Ξ) b is indeed represented on scalar fields φb ∈ Ab by the
b b
differential operator Xξ . The scalar fields form a U(Ξ)-module algebra.
In analogy to the previous section we can introduce vector and tensor fields as representations
b The transformation property for an arbitrary tensor reads
of the Hopf algebra U(Ξ).

δbξ Tbνµ11···ν
···µr
s
bξ Tbνµ1···ν
= −(X 1
···µn
n
) + (X b(∂ ξµ1 ) Tbνµ···µ
µ
n
1 ···νn
) + · · · + (Xb(∂ ξµn ) Tbνµ1···ν
µ 1
···µ
n
)
b(∂ ξν ) Tbν···ν
−(X µ1 ···µn
) − · · · − (X b(∂ ξν ) Tbνµ1···ν
···µn
).
ν1 n νn 1

Up to now we have seen the following:

• Diffeomorphisms are generated by vector-fields ξ ∈ Ξ and the universal enveloping algebra

U(Ξ) of the Lie algebra Ξ of vector-fields possesses a natural Hopf algebra structure defined
by (3.8).

• The algebra of scalar fields φ ∈ A is a U(Ξ)-module algebra.

b defined in
• The universal enveloping algebra U(Ξ) can be deformed to a Hopf algebra U(Ξ)
(4.2,4.3,4.4).

b consistently acts on the algebra of noncommutative functions A,

• U(Ξ) b i.e. the algebra of
b
noncommutative functions is a U(Ξ)-module algebra.
284 Meyer

b as the underlying “symmetry” of the gravity theory to be built on the

• Regarding U(Ξ)
b we established a full tensor calculus as representations of the
noncommutative space A,
b
Hopf algebra U(Ξ).

5 Noncommutative Geometry
The deformed algebra of infinitesimal diffeomorphisms and the tensor calculus covariant with
respect to it is the fundamental building-block for the definition of a noncommutative geome-
try on θ−deformed spaces. In this section we sketch the important steps towards a deformed
b µ . Algebraically, it
Einstein-Hilbert action [8]. A first ingredient is the covariant derivative D
can be defined by demanding that acting on a vector-field it produces a tensor-field
!
δbξ D
b µ Vbν = bξ D
−(X b µ Vbν ) − (X
b(∂ ξα ) D
µ
b α Vbν ) − (X
b(∂ ξα ) D
ν
b µ Vbα ) (5.1)

b µν ρ
The covariant derivative is given by a connection Γ

b µ Vbν = ∂bµ Vbν − Γ

D b µν ρ Vbρ .

b µν ρ
From (5.1) it is possible to deduce the transformation property of Γ

δbξ Γ
b µν ρ = (X b µν ρ ) − (X
bξ Γ b(∂ ξα ) Γ
µ
b αν ρ ) − (X
b(∂ ξα ) Γ
ν
b µα ρ ) + (X b µν α ) − (∂bµ ∂bν ξbρ ).
b(∂ ξρ ) Γ
α

The metric G b µν is defined as a symmetric tensor of rank two. It can be obtained for example
by a set of vector-fields E bµ a , a = 0, . . . , 3, where a is to be understood as a mere label. These
vector-fields are called vierbeins. Then the symmetrized product of those vector-fields is indeed
a symmetric tensor of rank two

b µν := 1 (E
G bµ a E
bν b + E
bν b E
bµ a )ηab .
2
Here ηab stands for the usual flat Minkowski space metric with signature (− + ++). Let us
assume that we can choose the vierbeins Ebµ a such that they reduce in the commutative limit to
a
the usual vierbeins eµ . Then also the metric Gb µν reduces to the usual, undeformed metric gµν .
The inverse metric tensor we denote by upper indices
b µν G
G b νρ = δµρ .

We use G b µν respectively G
b µν to raise and lower indices.
The curvature and torsion tensors are obtained by taking the commutator of two covariant
derivatives5

b µ, D
[D b ν ]Vbρ = R
bµνρ α Vbα + Tbµν α D
b α Vbρ

which leads to the expressions

bµνρ σ
R = ∂bν Γ
b µρ σ − ∂bµ Γ
b νρ σ + Γ
b νρ β Γ
b µβ σ − Γ
b µρ β Γ
b νβ σ
Tbµν α b νµ α − Γ
= Γ b νµ α .

5 The generalization of covariant derivatives acting on tensors is straight forward [8].

Noncommutative Spaces and Gravity 285

If we assume the torsion-free case, i.e.

b µν σ = Γ
Γ b νµ σ ,

we find an unique expression for the metric connection (Christoffel symbol) defined by

bαG !
b βγ =
D 0

in terms of the metric and its inverse6

b αβ σ = 1 (∂bα G
Γ b βγ + ∂bβ G
b αγ − ∂bγ G
b αβ )G
b γσ .
2

bµνρ σ we get the curvature scalar by contracting the indices

From the curvature tensor R

b := G
R b µν R
bνµρ ρ .

b indeed transforms as a scalar which may be checked explicitly by taking the deformed coprod-
R
uct (4.3) into account.
To obtain an integral which is invariant with respect to the Hopf algebra of deformed infini-
tesimal diffeomorphisms we need a measure function E. b We demand the transformation property

δbξ E
b = −X
bξ E
b−X
b(∂ ξµ ) E.
µ
b (5.2)

Then it follows with the deformed coproduct (4.3) that for any scalar field Sb

δbξ E
b Sb = −∂bµ (X
bξµ (E
b S)).
b

Hence, transforming the product of an arbitrary scalar field with a measure function Eb we obtain
a total derivative which vanishes under the integral. A suitable measure function with the desired
transformation property (5.2) is for instance given by the determinant of the vierbein E bµ a

E bµ a ) := 1 εµ1 ···µ4 εa ···a E

b = det(E bµ a1 E bµ a4 .
bµ a3 E
bµ a2 E
1 4 1 2 3 4
4!

That E b transforms correctly can be shown by using that the product of four E
bµ ai transforms
i
as a tensor of fourth rank and some combinatorics.
Now we have all ingredients to write down an Einstein-Hilbert action. Note that having
chosen a differential calculus as in (1.6), the integral is uniquely determined up to a normalization
factor by requiring7 [12] Z
∂bµ fb = 0

for all fb ∈ A.
b Then we define the Einstein-Hilbert action on Ab as
Z
SbEH := det(E bµ a )R
b + complex conj..

6 We don’t introduce a new symbol for the metric connection.

7 We consider functions that “vanish at infinity”.
286 Meyer

It is by construction invariant with respect to deformed diffeomorphisms meaning that

δbξ SbEH = 0.

In this section we have presented the fundamentals of a noncommutative geometry on the algebra
Ab and defined an invariant Einstein-Hilbert action. There is however one important step missing
which is subject of the following section: We want to make contact of the noncommutative gravity
theory with Einstein’s gravity theory. This we achieve by introducing the ⋆-product formalism.

6 Star Products and Expanded Einstein-Hilbert Action

To express the noncommutative fields in terms of their commutative counterparts we first ob-
serve that we can map the whole algebraic construction of the previous sections to the algebra
of commutative functions via the vector space isomorphism W introduced in Section 1. By
equipping the algebra of commutative functions with a new product denoted by ⋆ be can render
W an algebra isomorphism. We define

f ⋆ g := W −1 (W (f )W (g)) = W −1 (fbgb) (6.1)

and obtain
(A, ⋆) ∼ b
= A.
The ⋆-product corresponding to the symmetric ordering prescription W is then given explicitly
by the Moyal-Weyl product8 (see also S. Cnockaert’s lecture)

i µν
∂µ ⊗∂ν i
f ⋆ g = µ ◦ e2θ f ⊗ g = f g + θµν (∂µ f )(∂ν g) + O(θ2 ).
2
It is a deformation of the commutative point-wise product to which it reduces in the limit θ → 0.
In virtue of the isomorphism W we can map all noncommutative fields to commutative
functions in A
Fb 7→ W −1 (Fb) ≡ F.
We then expand the image F in orders of the deformation parameter θ

F = F (0) + F (1) + F (2) + O(θ3 ),

where the zeroth order always corresponds to the undeformed quantity. Products of functions
in Ab are simply mapped to ⋆-products of the corresponding functions in A. The same can be
done for the action of the derivative ∂bµ and consequently for an arbitrary differential operator
acting on Ab [8].
The fundamental dynamical field of our gravity theory is the vierbein field E bµ a . All other
quantities such as metric, connection and curvature can be expressed in terms of it. Its image
with respect to W −1 is denoted by Eµ a . In first approximation we study the case

Eµ a = eµ a ,

where eµ a is the usual vierbein field. Then for instance the metric is given up to second order

8 This is an immediate consequence of (1.10).

Noncommutative Spaces and Gravity 287

in θ by
1 1
Gµν = (Eµ a ⋆ Eν b + Eν b ⋆ Eµ a )ηab = (eµ a ⋆ eν b + eν b ⋆ eµ a )ηab
2 2
1 α1 β1 α2 β2
= gµν − θ θ a
(∂α1 ∂α2 eµ )(∂β1 ∂β2 eν b )ηab + . . . ,
8
where gµν is the usual, undeformed metric. For the Christoffel symbol one finds up to second
order: The zeroths order is the undeformed expression
1

Γ(0)ρ
µν = ∂µ gνγ + ∂ν gµγ − ∂γ gµν g γρ , (6.2)
2
the first order reads
i αβ
Γ(1)
µν =
ρ
θ (∂α Γ(0)σ τρ
µν )gστ (∂β g ) (6.3)
2
and the second order

1

Γ(2)
µν =
ρ
= − θα1 β1 θα2 β2 (∂α1 ∂α2 Γ(0) σρ (0) στ
µνσ )(∂β1 ∂β2 g ) − 2(∂α1 Γµνσ )∂β1 (∂α2 g )(∂β2 gτ ξ )g
ξρ
8

(0)
− Γµνσ (∂α1 ∂α2 g στ )(∂β1 ∂β2 gτ ξ ) + g στ (∂α1 ∂α2 eτ a )(∂β1 ∂β2 eξ b )ηab

 1

− 2∂α1 (∂α2 g στ )(∂β2 gτ λ )g λκ (∂β1 gκξ ) g ξρ + ∂µ (∂α1 ∂α2 eνa )(∂β1 ∂β2 eσb )
2 
a b

a b

 σρ
+ ∂ν (∂α1 ∂α2 eσ )(∂β1 ∂β2 eµ ) − ∂σ (∂α1 ∂α2 eµ )(∂β1 ∂β2 eν ) ηab g , (6.4)

where
Γ(0) (0)ρ
µνσ = Γµν gρσ . (6.5)

The expressions for the curvature tensor read

 
(1) σ i (0) τ
Rµνρ = − θκλ (∂κ Rµνρ (0) τ
)(∂λ gτ γ )g γσ − (∂κ Γ(0) β
νρ ) Γµβ (∂λ gτ γ )g
γσ
2


(0)
−Γ(0) σ
µτ (∂λ gβγ )g
γτ
+ ∂µ (∂λ gβγ )g γσ + (∂λ Γµβ σ )

(0)
+(∂κ Γ(0)
µρ
β
) Γνβ τ (∂λ gτ γ )g γσ − Γ(0) σ
ντ (∂λ gβγ )g
γτ


(0)
+∂ν (∂λ gβγ )g γσ + (∂λ Γνβ σ ) (6.6)
(2) σ
Rµνρ = ∂ν Γ(2) σ (2) γ (0) σ
µρ + Γνρ Γµγ + Γνρ Γµγ
(0) γ (2) σ

i  
+ θαβ (∂α Γ(1) γ (0) σ (0) γ (1) σ
νρ )(∂β Γµγ ) + (∂α Γνρ )(∂β Γµγ )
2
1 α1 β1 α2 β2
− θ θ (∂α1 ∂α2 Γ(0) γ (0) σ
νρ )(∂β1 ∂β2 Γµγ ) − (µ ↔ ν), (6.7)
8
where the second order is given implicitly in terms of the Christoffel symbol.
288 Meyer

The deformed Einstein-Hilbert action is given by

Z
1
SEH = d4 x det⋆ eµ a ⋆ R + c.c.
2
Z
1
= d4 x det⋆ eµ a ⋆ (R + R̄)
2
Z
1
= d4 x det⋆ eµ a (R + R̄)
2
Z
(0)
= SEH + d4 x (deteµ a )R(2) + (det⋆ eµ a )(2) R(0) , (6.8)

where we used that the integral together with the Moyal-Weyl product has the property9
Z Z Z
d4 x f ⋆ g = d4 x f g = d4 x g ⋆ f.

In (6.8) det⋆ eµ a is the ⋆-determinant

1 µ1 ···µ4
det⋆ eµ a = ε εa1 ···a4 eµ1 a1 ⋆ eµ2 a2 ⋆ eµ3 a3 ⋆ eµ4 a4
4!
= deteµ a + (det⋆ eµ a )(2) + . . . ,

where
1 1 α1 β1 α2 β2 µ1 ...µ4
(det⋆ )(2) = − θ θ ε εa1 ...a4
 8 4!
(∂α1 ∂α2 eµ1 a1 )(∂β1 ∂β2 eµ2 a2 )eµ3 a3 eµ4 a4
+ ∂α1 ∂α2 (eµ1 a1 eµ2 a2 )(∂β1 ∂β2 eµ3 a3 )eµ4 a4

+ ∂α1 ∂α2 (eµ1 a1 eµ2 a2 eµ3 a3 )(∂β1 ∂β2 eµ4 a4 ) . (6.9)

The odd orders of θ vanish in (6.8) but the even orders of θ give nontrivial contributions.
Equation (6.8) shows explicitly the corrections to Einsteins gravity predicted by the noncom-
mutative theory.

Remarks
For an introduction to field theories on noncommutative spaces, we recommend the review
articles [12, 13]. To learn more about related approaches to noncommutative geometry the reader
is referred to [14, 15]. More about Hopf algebras and Quantum Groups can be found in [9, 10, 11].
A good pedagogical introduction to ⋆-products can be found in [16]. The construction presented
in this lecture is based [8]. The mathematically interested reader is encouraged to read also our
forthcoming work [17] where a generalization to a large class of noncommutative spaces will be
considered.
In connection to X. Bekaert’s Modave lecture it should be mentioned that Hopf algebra sym-
metries (Quantum Groups) fit in the context of generalized symmetries via universal enveloping
algebras which relax the assumptions made for the Coleman-Mandula theorem. One motivation
to study Quantum Groups in physics may therefore be the hope to avoid the conclusions of the
9 This follows by partial integration.
Noncommutative Spaces and Gravity 289

Coleman-Mandula theorem. We hope to gain new insight to the meaning of Quantum Groups in
noncommutative field theories by the above studied θ-deformation of diffeomorphisms and (as a
subset) of the Poincaré algebra.

Acknowledgments
I would like to thank the organizers of the first Modave Summer School for an interesting,
productive and enjoyable meeting. I acknowledge useful discussions with many participants,
among them G. Barnich, X. Bekaert and J.-H. Park. The results presented in this lecture were
obtained together with P. Aschieri, C. Blohmann, M. Dimitrijević, P. Schupp and J. Wess and
I would like to thank them for their collaboration from which I learned so much.

References
[1] J. Wess, Deformed coordinate spaces: Derivatives, Lecture given at the Balkan workshop
BW2003, August 2003, Vrnjacka Banja, Serbia , hep-th/0408080.

[2] J. Wess and B. Zumino, Covariant Differential Calculus on the Quantum Hyperplane, Nucl.
Phys. Proc. Suppl. 18B, 302 (1991).

[3] M. Dimitrijević, L. Möller, E. Tsouchnika, J. Wess, and M. Wohlgenannt, Deformed field

theory on kappa-spacetime, Eur. Phys. J. C31, 129 (2003), hep-th/0307149.

[4] M. Dimitrijević, F. Meyer, L. Möller, and J. Wess, Gauge theories on the kappa-Minkowski
spacetime, Eur. Phys. J. C36, 117 (2004), hep-th/0310116.

[5] R. Oeckl, Untwisting noncommutative Rd and the equivalence of quantum field theories,
Nucl. Phys. B581, 559 (2000), hep-th/0003018.

[6] M. Chaichian, P. Kulish, K. Nishijima, and A. Tureanu, On a Lorentz-invariant interpre-

tation of noncommutative space-time and its implications on noncommutative QFT, Phys.
Lett. B604, 98 (2004), hep-th/0408069.

[7] F. Koch and E. Tsouchnika, Construction of θ-Poincare algebras and their invariants on
Mθ , Nucl. Phys. B717, 387 (2005), hep-th/0409012.

[8] P. Aschieri, C. Blohmann, M. Dimitijević, F. Meyer, P. Schupp, and J. Wess, A gravity

theory on noncommutative spaces, Class. Quant. Grav. 22, 3511 (2005), hep-th/0504183.

[9] V. Chari and A. Pressley, A Guide to Quantum Groups, Cambridge: University Press
(1995) 651 p.

[10] A. Klimyk and K. Schmüdgen, Quantum groups and their representations, Berlin, Germany:
Springer (1997) 552 p.

[11] S. Majid, Foundations of Quantum Group Theory, Cambridge: University Press (200) 640
p.

[12] M. R. Douglas and N. A. Nekrasov, Noncommutative field theory, Rev. Mod. Phys. 73, 977
(2001), hep-th/0106048.
290 Meyer

[13] R. J. Szabo, Quantum field theory on noncommutative spaces, Phys. Rept. 378, 207 (2003),
hep-th/0109162.

[14] A. Connes, Noncommutative Geometry, Academic Press Inc., San Diego, New York, Boston,
(1994).
[15] J. Madore, An introduction to noncommutative differential geometry and physical applica-
tions, Cambridge, UK: Univ. Pr. (2000) 371 p.

[16] S. Waldmann, An Introduction to Deformation Quantization, (2002),

http://idefix.physik.uni-freiburg.de/ stefan/Skripte/intro/index.html, Lecture notes.

[17] P. Aschieri et al., in preparation.