Geometric Quantization: An Introduction To Spin Quantization and Related Topics

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 41

Geometric Quantization

An Introduction to Spin
C
quantization and related topics.
Winter semester 2011/12 at the university of Bonn
Preliminary Version
Fabian Meier
July 2011
Gedruckt auf alterungsbestndigem Papier ISO 9706
Contents
1 Introduction and Background material 2
1.a Idea and physical background . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.b Background material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.b.1 Spin
C
structures and Dirac operators . . . . . . . . . . . . . . . . . 2
1.b.2 Representation theory of abelian Lie groups . . . . . . . . . . . . . 8
1.b.3 Presymplectic structures and Moment maps . . . . . . . . . . . . . 10
1.b.4 Equivariant cohomology . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.b.5 Equivariant Spin
C
structures . . . . . . . . . . . . . . . . . . . . . . 13
1.c Geometric Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.c.1 Spin
C
quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.c.2 Relation to "Physical Quantization" . . . . . . . . . . . . . . . . . . 14
2 Additivity under Cutting 15
2.a Manipulation of Spin
C
structures . . . . . . . . . . . . . . . . . . . . . . . . 15
2.b Spin
C
cutting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.c Kostant Formula for Isolated Fixed Points . . . . . . . . . . . . . . . . . . . 19
2.d Kostant Formula for Non-Isolated Fixed Points . . . . . . . . . . . . . . . 23
2.d.1 The A-roof class andequivariant cohomology for trivial S
1
-manifolds 23
2.d.2 The Atiyah-Segal-Singer xed point formula for Spin
C
Dirac op-
erators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.d.3 Atiyah-Segal-Singer index theorem . . . . . . . . . . . . . . . . . . 29
2.e Additivity under cutting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.f An example: The Two-Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3
4 CONTENTS
3 Spinc Prequantization 32
3.a Cutting of a Spin
C
prequantization . . . . . . . . . . . . . . . . . . . . . . . 32
4 Quantization commutes with Reduction 33
4.a Spin
C
reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.b Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.c Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.d Cut space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.e Proof of Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
A Short introduction to Principal bundles 34
Abstract
This text contains a detailed introduction into the concepts and methods of Spin
C
quantization. After a thorough discussion of the basic material we will describe the
(symplectic) cutting scheme andthencome tothe Spin
C
versionof the famous "Quant-
ization commutes with reduction" theorem.
Chapter 1
Introduction and Background
material
1.a Idea and physical background
In physics, the quantization of a classical system which associates to every classical
oversable a quantum one. Formally, this is a map fullling several properties to be
stated below.
Unfortunately these requirements cannot be fullled simultaneously, so they have to
be weakened. Geometric Quantization nowwants to construct weak kinds of quant-
izations out of the geometry of manifolds.
1.b Background material
In the following M always denotes a compact, oriented Riemannian smooth manifold
of even dimension n =2k.
1.b.1 Spin
C
structures and Dirac operators
For a tangent bundle TM of a Riemannian smooth manifold M, we can reduce the
structure group from Gl
n
to SO(n). From Lie group theory, we know that the fun-
damental group
1

SO(n)

equals Z/2. Now the Spin(n) be the Lie group which is


a double cover of SO(n) (it will be constructed explicitly below). Now the question
arises: Can the structure group SO(n) be "replaced by" Spin(n)?
Such a replacement (where, of course, all diagrams induced by charts and the projec-
tion
SO
: Spin(n) SO(n), should commute) is called a Spin structure.
2
1.B. BACKGROUND MATERIAL 3
A complexied version of this will be called Spin
C
structure.
Both Spin and Spin
C
structures do not always exist; when they exist, they are normally
not unique. The existence question can be formulated in terms of Stiefel-Whitney-
classes and is, therefore, investigated by topological means. We will give some in-
formation about the classication of Spin
C
structures after we have made the relevant
denitions. A detailed account can be found in the appendix of [Lawson89].
A Spin or Spin
C
structure gives by a canonical construction a complex vector bundle
(called Spinor bundle); with the help of a connection and Clifford multiplication, we
get a rst order differential operator on that Spinor bundle: It is called Dirac operator.
In the following paragraphs we will discuss the construction of the Spin and Spin
C
group, the necessary representation theory and, at last, the construction of the Spinor
bundle and the Dirac operator.
Clifford algebras, Spin and Spin
C
The main source for the following material is [Morgan96].
Let
T
n

R
n

=R R
n
R
n
R
n
. . .
be the tensor algebra of R
n
and I
n
the ideal generated by x x x, x, x R
n
(the
term, denotes the standard scalar product of R
n
). We dene
Cl
n
=T
n

I
n
.
This is an 2
n
-dimensional Z/2-graded R-algebra, which is isomorphic to the exterior
algebra as a vector space. If e
1
, . . . , e
n
is an orthonormal basis of R
n
, ordered tensor
products of mutually different e
i
form a basis of Cl
n
. Notice also that in this case, we
have e
i
e
j
=(1)e
j
e
i
for i = j and e
2
i
=1.
The elements x R
n
of norm 1 generate a group lying in Cl
n
. This group is called
Pin(n).
The grading splits Cl
n
into Cl
0
n
Cl
1
n
. If we take the intersection of the subalgebra Cl
0
n
with Pin(n), we get the group Spin(n).
To keep the promise made at the beginning, we have to construct a two-fold covering
map from Spin(n) to SO(n). Consider the map
Spin(n) R
n
Cl
n
(g, x) g xg
1
.
Decomposing x into the part orthogonal and parallel to g, we see that g xg
1
R
n

Cl
n
. Hence above map induces a group homomorphism
SO
: Spin(n) Gl
n
.
4 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
Lemma 1.b(i). The map
SO
has the following properties:
1. For each g Spin(n) we have
SO
(g) is an orthogonal and orientation-preserving
map.
2. Every reection lies in the image of
SO
(which implies that
SO
is surjective)
3. The kernel of
SO
is {1}.
Proof. Can be found e.g. in [Baker06].
Now we consider the complexied Clifford Algebra Cl
n
C. If we again consider the
group generated by the unit elements and intersect it with Cl
0
n
C, we get the group
Spin
C
(n). A direct calculation shows the following canonical isomorphism
Spin
C
(n)

=Spin
1
S
1
,
where S
1
is the circle group in C. If we represent elements of Spin
C
(n) by (g, z) with
g Spin(n) and z S
1
, we get the following well-dened maps:

Spin
: Spin(n) Spin
C
(n)
S
1 : Spin
C
(n) S
1
g (g, 1) (g, z) z
2
These maps will be important in the construction of vector bundles described later.
Representation theory of Clifford algebras
The complexied Clifford algebras have a simple structure as matrix algebras. Gen-
erally, you have to distinguish odd and even n, but since we assumed n = 2k, we will
skip the odd-dimensional case.
Adirect calculationshows Cl
2
C

=C[2], where C[2] means the algebra of 22-matrices


over C. Now we can construct an inductive argument and get:
Lemma 1.b(ii). Cl
2k
C and C[2
k
] are isomorphic as C-algebras.
So Cl
2k
has a unique nite-dimensional irreducible representation (up to isomorph-
ism). This follows from Wedderburns theorem . We call this space S
n
C
(meaning the
Source for appro-
priate Wedderburn
theorem
space C
2
k
with the representation described above).
Now we need one further splitting of the complexied Clifford algebra. For that pur-
pose, we dene a special element. Choose an arbitrary oriented orthonormal basis
e
1
, e
2
, . . . , e
n
of R
n
. Then

C
=i
n+1
e
1
e
2
e
n
1.B. BACKGROUND MATERIAL 5
A direct calculation shows that
2
C
= 1 and that
C
is independent of the choice of
oriented orthonormal basis (a change of the basis is given by multiplication with the
determinant of the matrix used for this change). Now every element of Cl
n
C can be
written as sum of two elements, which are eigenvectors for 1 with respect to
C
:
v =
1
2

v +
C
v

+
1
2

v
C
v

.
This splits Cl
n
C into two spaces called

Cl
n
C

. If we consider the action of


C
on
S
n
C
, this space also splits into S
n

.
The following lemma describes the situation in detail.
Lemma 1.b(iii). The decomposition S
n

is a decomposition of Cl
0
n
C-modules. We
have the following four isomorphisms (as C-algebras):

Cl
0
n
C

=End
C

S
n
+

Cl
0
n
C

=End
C

S
n

Cl
1
n
C

=Hom
C

S
n
+
, S
n


Cl
1
n
C

=Hom
C

S
n

, S
n
+

Proof. Look at the commutator relations of e


i
und
C
.
Since Spin and Spin
C
are "subgroups" of Cl
0
C, we get two representations

n
of
these groups on S
n

. These will be the key tool to construct the Spinor bundles.


Spin
C
bundles
We now dene the concept of a Spin
C
structure; afterwards we look at spinor bundles
and Clifford bundles. All constructions work in a similar fashion for Spin bundles, but
since our main aimis to understand Spin
C
Dirac operators, we omit this extra inform-
ation. For the next sections, the knowledge of principal bundles, their connections
and their induced vector bundles is essential. We advice the non-experienced reader
to have a look at appendix A.
Let P
M
be the prinicipal SO bundle which describes the tangent bundle TM. As de-
scribed in the appendix, every surjective group homomorphismG H can be used to
transforma principal G-bundle into a principal H-bundle. Using this strategy and the
map
SO
: Spin
C
SO, we make the following denition: A principal Spin
C
-bundle

P
M
is called a Spin
C
structure (for P
M
), if the "strategy" applied to

P
M
gives us P
M
.
Anequivalent denitioninterms of vector bundles would be: Take the tangent bundle
TM

=P
M

SO
R
n
and "replace" the structure group SO by Spin
C
. Sometimes we need
Spin
C
structures for other vector bundles with structure group SO. The denitionthen
is verbatim the same.
The existence of a Spin
C
structure is equivalent to the vanshining of the second integ-
ral Stiefel-Whitney-class W
2
(see [Lawson89], appendix). We do not want to go into
6 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
detail, but mention the fact, that for n 4, Spin
C
structures always exist. If Spin
C
structures exist, they are (non-canonically) parametrised by the second integral co-
homology group H
2
(M; Z). On the level of Spinor bundles, one can give a fairly direct
description of this parametrisation.
From now on, let us assume we chose a Spin
C
structure

P
M
for our manifold M. We
dene two bundles, called Clifford bundle and Spinor bundle respectively:
Cl
C
(M) :=P
M

SO
Cl
n
C
S
C

P
M

:=

P
M

Spin
C S
n
C
On Cl
n
we use the action of SO(n) induced by the canonical action of SO(n) on R
n
. For
the denition of the Clifford bundle it is not necessary to have a Spin
C
structure; but
we can also write in a different way:
Cl
C
(M) =

P
M

Spin
C Cl
n
C
The action of Spin
C
on Cl
n
is given by the above action composed with
SO
. If
Cl
n
C and S
n
C
, we dene as the representation. If we show that this action
commutes with the actionof Spin
C
(n), we get a brewise actionof Cl
C
(M) onS
C

P
M

.
Careful thinking shows that we have to prove the equation
(
1
) () =( ) for all Spin
C
(n),
which is obvious.
All splittings dened in the last sections are always well-dened over bundles; the
only non-obvious thing about this is to mention, that
C
is invariant under the action
of SO(n), and thus the +/ splitting of Cl
n
C is preserved. This allows us to dene
the bundles Cl
0
C
(M)
+
, Cl
0
C
(M)

, Cl
1
C
(M)
+
, Cl
1
C
(M)

, S
+
C

P
M

and S

P
M

.
Remember that we also dened a surjective group homomorphism
S
1 : Spin
C
(n)
S
1
, (g, z) z
2
. In this way we get an S
1
-bundle P
S
1
M
, which is called the associated line
bundle (since principal S
1
-bundles are in one-to-one correspondence with complex
line bundles).
At this point we start a short digressionabout characteristic classes: For any n-dimensional
complex vector bundle V we can dene Chern classes c
1
(V ), . . . , c
n
(V ), where c
i
(V )
H
2i
(M; Z). They can be calculated in various ways, e.g. fromthe curvature, by topolo-
gical intersections or by universal properties. Complex line bundles are classied by
their rst chern class. This gives an isomorphism between

complex line bundles over M

= H
2
(M; Z).
Let nowK be an arbitrary complex line bundle. If we tensor S
C

P
M

by K, we also get
a Spinor bundle (of a different Spin
C
structure, if c
1
(K) = 0). The idea behind this is
1.B. BACKGROUND MATERIAL 7
the following: If we look at the chart changing maps of S
C

P
M

K and S
C

P
M

, they
only differ in the S
1
-component of Spin
C
= Spin
C

1
S
1
. Since this is quotiened out
by the map
SO
, we get the same principal SO-bundle.
The following lemma, which we state without proof, is shown by [Lawson89] and
[Morgan96]:
Lemma 1.b(iv). Let S
C

P
M

be the Spinor bundle of one xed Spin


C
structure and let
a H
2
(M; Z) be the rst Chern class of the associated line bundle L. Then the Spin
C
structures are classiedby a+2H
2
(M; Z) H
2
(M; Z). For the Spin
C
structure belonging
to b

a +2H
2
(M; Z)

, we construct the Spinor bundle in the following way:


Take a line bundle K with c
1
(K) =
ba
2
and take S
C

P
M

K. The action of the Clifford


bundle is given by combining the action on S
C

P
M

with a trivial action on K. The new


associated line bundle is L K
2
.
To dene Dirac operators, we rst have to discuss connections.
Connections on the Spinor bundle
We now want to construct a connection on the Spinor bundle which behaves in a
natural way with respect to the Clifford multiplication. From now on take the Spinor
bundle S
C

P
M

with associated line bundle L.


Let
M
be the Levi-Civita connection for TM and
L
be an arbitrary connection for
L. Out of these two ingredients, we can build a connection on S
C

P
M

. This works in
the following way:
We change to the level of principal bundles and look again at the structure groups
SO(n), S
1
and Spin
C
(n)

=Spin(n)
1
S
1
. If we interpret connections as distributions
of horizontal spaces, we see that our two connections induce a connection on the
product principal bundle for the group SO(n) S
1
. Now Spin
C
(n) is a twofold cov-
ering of SO(n) S
1
. The covering map allows us to lift the horizontal spaces to

P
M
.
This induces the connection

on S
C

P
M

, which we wanted to construct. It can be


computed in local coordinates (see [Morgan96], p.39) in a usable way.
We get the following formula for S
C

P
M

and

TM

Cl
C
(M) ((V ) always
denotes the smooth sections of a vector bundle V ):

( ) =
M
() +

().
The Spin
C
Dirac operator
Now we have two kinds of maps:
8 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
1. The connection

can be interpreted as a map from(S
C

P
M

) to

S
C

P
M

TM

.
2. The Clifford multiplication gives us a map from

S
C

P
M

TM

to (S
C

P
M

).
If we also identify TM and TM

with the help of the given metric, we combine those


two maps to get an operator from

S
C

P
M

to itself. This is called a Spin


C
Dirac
operator. If we choose a local basis e
1
, . . . , e
n
for TM, we get the following expression:
D() =

i
e
i

e
i
.
The operator D depends on two choices:
1. The Spin
C
structure.
2. The connection
L
on L.
Here the rst one is the more crucial choice. For the second choice remember that two
connections on L differ by a one-formi with values in the Lie algebra iR. If we take
(TM) to be the dual of , we see that the difference of the two Dirac operators is
just Clifford multiplication by .
ASpin
C
Dirac operator has some nice properties: It is a rst-order, self-adjoint, elliptic
operator.
Remark 1.b(v). The operator Dsplits withrespect tothe splitting

S
+
C

P
M

P
M

into D
+
: (S
+
C

P
M

) (S

P
M

) and D

: (S

P
M

) (S
+
C

P
M

). The com-
ponents from (S

P
M

) to (S

P
M

) vanish: The connection



xes the com-
ponent, but multiplication with e
i
exchanges the components, since e
i
Cl
1
C (see
lemma 1.b(iii)).
1.b.2 Representation theory of abelian Lie groups
A short discussion of general representation theory of compact Lie groups is given in
the appendix.
Here, we will restrict ourselves to the Lie groupS
1
(ofteninterpreted as embedded into
C). Interpreted as "matrix group", the Lie algebra s
1
is canonically given by iR and
the exponential map is the usual complex exp with kernel {2ik | k Z}. For various
reasons, we will often identify the space iR with R by multiplication with i, i.e. 2i
is identied with 2. The respective map belonging to s

1
is called u. Therefore, the
exponential map exp
S
1 from s
1
to S
1
becomes x expxi.
Lemma 1.b(vi). The complex irreducible representations of S
1
are always one-dimen-
sional and in natural one-to-one correpondence with Z (for general Lie groups, this is
called weight lattice). The representation for n Z will be called z
n
.
The real irreducible representations are given by the one-dimensional trivial represent-
ation and two-dimensional representations coming fromthe complex representations.
1.B. BACKGROUND MATERIAL 9
Proof. Schurs lemma says that aG-equivariant mapbetweenirreducibleG-representations
is always an isomorphismor trivial. The proof for this is actually very easy: The kernel
and image are sub-representations, so can only be equal to zero or the whole space.
Now every complex S
1
-representation on a space V is an S
1
-map, since S
1
is an
abelian group. Every C denes S
1
-equivariant maps
g
: V V . Now every
linear map has a complex eigenvalue

g
. Since
g

g
cannot be an isomorphism is
has to be zero by Schurs lemma. So
g
is given by multiplication with

g
. This can
only be irreducible if V is 1-dimensional. Here, has to be a group homorphismfrom
S
1
to itself; those are indexed by Z.
For a real representationonW we do the following: We take the complexiedrepres-
entation
C
on C
R
W. It decomposes into 1-dimensional complex representations
W
i
. Now W
1
has a real basis of the form v +iw, w +iv. We now want to show that
R{v, w} is an S
1
-invariant subspace of W. Let g S
1
. Then we have
g v =

g (v +iw)

span{v +iw, w+iv}

,
which shows the claim. Therefore, real irreducible representations of S
1
are one-
or two-dimensional. One-dimensional represenations are necessarily trivial. Since
SO(2)

=U(1)

=S
1
, the two-dimensional representations are complex representations
after "forgetting the complex structure".
Now assume that we have real (not necessarily irreducible) representation on an
even-dimensional vector space W. We want to choose a complex structure (i.e. a
linear map J with J
2
=1) on W which commutes with ; this would convert W into
a complex representation space W
J
.
For that purpose we split W into two-dimensional representation spaces W
1
W
k
.
Lemma 1.b(vi) tells us that the spaces W
i
fall into two categories: trivial representa-
tions or those coming from a complex representation. Now we can choose a complex
structure J
i
for every W
i
and put them all together to get J . The complex represent-
ations we get on W
i
are independent of the choice of J
i
up to sign: That means that
different choices of J
i
might lead to the representations z
n
and z
n
but the norm of n
is well-dened.
In the general representation theory of Lie groups, the concept of a character is very
important. If is a representation on V , it is dened as g tr

(g)

. For (nite di-


mensional) representations of S
1
a character is of the form

nZ
k
n
z
n
where k
n
are non-negative numbers indicating "how often" the irreducible repres-
entation z
n
appears in (only nitely many k
n
are non-zero). Since we are in the
one-dimensional world, the trace is nothing but an embedding S
1
C. Since exp
S
1
is surjective, the character z
n
is determined by the map v trexp
S
1 (nv) for v s
1
.
10 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
Making the identications above we can view this map as v exp

inu(v)

. We will
often interpret z as a map from s
1
to C.
We will also work with virtual representations. They are formal differences of repres-
entations (in the sense of the Grothendieck construction which is also used in topolo-
gical K-theory). Their characters are also of the form specied above, but then k
n
can
be any integer.
1.b.3 Presymplectic structures and Moment maps
Fromnowonwe consider manifolds withS
1
-action. Let M be equippedwitha smooth
action of S
1
; this action should x the Riemannian metric on M (such metrics can
always be found, look at the equivariant partition of unity etc. in [Guillemin02], ap-
pendix B).
Let now be a closed two-form on M which is preserved by the S
1
-action. It is called
a presymplectic form.
Let be the element 2 s
1

=R. Then exp


S
1 (t ) acts on M, and the curve exp
S
1 (t ) x
denes an element of TM
x
. The induced vector eld shall be called
M
.
Now, a moment map is a map : M s
1

= R such that the following equation for
one-forms holds:
d=

A triple

M, ,

of this kind will be called a moment triple. It will be used in the


concept of prequantization dened later.
1.b.4 Equivariant cohomology
In this section we will describe the concept of a cohomology theory H

G
(M) for mani-
folds M withG-action. We will followappendix Cof [Guillemin02], but restrict ourselves
mainly to the case where G =S
1
and the action is smooth.
There is a topological constructionof this cohomology whichuses the universal bundle
EG over BG. We will consider the Cartan model which is a contruction in the spirit of
de Rhamcohomology (hence we need smoothness everywhere, but this is no problem
in our situation).
Denition 1.b(vii). The equivariant differential forms of degree q are given by

q
G
(M) =

2i +j =q

S
i

j
(M)

G
The terms are dened as follows
1.B. BACKGROUND MATERIAL 11
g

is the dual of the Lie algebra of G.


S
i
denotes the symmetric tensor product; you can see this as a polynomial in
g

.
( )
G
means the forms invariant under the action of G. Here, G acts on g

by the
(dual of the) conjugation, and on
j
(M) by the action induced by the action on
M.
Notice the grading which gives double weight to the rst factor.
For G = S
1
, the structure becomes easier: The action of S
1
on Li e(S
1
) is trivial and
polynomials in R

are just usual polynomials a


n
u
n
+. . . +a
0
. If we split this sum into
its summands and move a
n
to the
j
(M)-part with the help of the tensor product, we
see that elements of
q
S
1
(M) are of the form

2i +j =q

j
u
i
, where
j
is a S
1
-invariant
j -form and u is the identication s
1
.
Denition 1.b(viii). The equivariant exterior derivative d
G
is given by the following
formula:
(d
G
)() =d

()

().
Here we interpret
q
G
(M) is interpreted as polynomial map from g to
j
(M), i.e.
if we plug in an element g, we get a usual one-form. As above,
M
is the vector
eld on M induced by the Lie algebra element . Counting the degrees, we see that d
G
maps from
q
G
(M)
q+1
G
(M).
Lemma 1.b(ix). d
2
G
=0.
Proof. Direct calculation.
The construction above gives us a cohomology theory H

G
(M).
For G =S
1
, the map d
S
1 is given by :
d
S
1

j
u
i

=d(
j
)u
i
+(
M
)
j
u
i +1
,
where
M
is dened as in 1.b.3.
As discussed above, the space

G
(M) can be interpreted as space of G-invariant poly-
nomials on g with values in

(M). Nowwe can replace "polynomials" by a more gen-


eral space of functions; a slightly strange but important space for us is the following:
Let C
#
g
be the space almost everywhere analytic functions on the (real vectorspace) g,
i.e. the space of all analytic functions fromg\P to Cwhere P is a closed set of measure
zero.
12 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
This still allows us to write down an exterior differential and to dene cohomology.
The spaces H
k
G
(M) remain unchanged but we gain elements of "innite order". Since
the usual G-equivariant cohomology is a module over the ring S

(g)
G
, we might hope
to "replace" this ring by C
#
g
, getting the new cohomology H

G
(M)
S
C
#
g
. This is par-
ticularly true if the action of G on M is trivial (this case will be explicitly used for the
Konstant formula).
Add the "almost
everywhere ana-
lytic" coefcient
ring, Asterisque 215,
Cor. 64 and 104
We denote this cohomology by H
#
G
(M). For G =S
1
(and assuming the tensor splitting
above), taking the identication of s
1
with R, this can be interpreted as an analytic
map fromR\ P to H

(M; C), where P is a set of measure zero. So integration over M


gives a map from H
#
G
(M) to C
#
g
.
We now analyse the situation for different q:
1. q =0: The space
0
S
1
(M) consists of the S
1
-invariant functions. Since d
S
1 =d on

0
S
1
(M), we know that H
0
S
1
(M) consists of invariant locally constant functions
on M, i.e. H
0
S
1
(M)

= R
k
, where k is the number of connected components of
M/S
1
.
2. q = 1: Let
1

1
S
1
(M) (there can be no u, since this would already exceed the
degree limit). Now d
S
1 (
1
) = d
1
+
1
(
M
)u, so d
S
1
1
= 0 means d
1
= 0 and

1
(
M
) = 0. So
1
is closed and horizontal (in the sense of principal bundles).
Exact forms are of the formd f , where f is invariant. You can prove that we have
H
1
S
1
(M)

= H
1
(M/S
1
) (where, of course, M/S
1
might be no manifold, so that we
have to apply the topological denition of cohomology).
3. q = 2: Equivariant two-forms are elements of the form u, where is an
invariant two-form and : M R. The derivative d
S
1 (u) is given by d+
(
M
)u du. This is zero iff is closed and
(
M
)=d.
So u is closed if and only if is a presymplectic form for M and an
associated moment map. compare signs
Classication of equivariant line bundles
Let L be a hermitian line bundle with a unitary S
1
-action on it. This means the fol-
lowing: For g S
1
we get a linear map l
g
: L
x
L
g x
, where x M, which preserves
the hermitian metric. We call L an equivariant S
1
-bundle. We are now interested
in the classication of these line bundles (up to S
1
-equivariant isomorphisms of line
bundles).
1.B. BACKGROUND MATERIAL 13
We rst have to dene the equivariant rst Chern class. There is a general denition
of equivariant Chern classes (see p. 212 in [Guillemin02]), but we will just introduce a
formula which can be used for computations "in practise".
Let p : P M be the circle bundle for the hermitian line bundle L. We choose an
S
1
-invariant connection form on P. We construct a moment triple as follows:
The two-formd on P is S
1
-invariant, so it is equal to p

for some two-form on


Here, we have to
check which iden-
tication between
the Lie algebra and
R is chosen
M. Furthermore (
P
) denes an S
1
-invariant function on P, which is given by p

for some : M R. Of course, we want to be a moment map for the just dened.
We have to check the following equation:

P
d +d

() =0
The left side coincides withthe Lie derivative of indirection
M
. Fromthe invariance
we know that this has to vanish .
nd source
Putting this together we know that
1
2

denes an element of H
2
S
1
(M). This
element is called the equivariant rst Chern class of L. The constant factor 1/2 ap-
pears for the same reason as in the ordinary theory of characteristic classes: We want
to dene an element of the integral cohomology (which can be embedded into the
real cohomology).
Theorem1.b(x). For connected S
1
-manifolds M the equivariant rst Chern class c
S
1
1
(L)
denes a bijection between equivariant line bundles and H
2
S
1
(M; Z).
We will not give a proof here, but make some concrete calculations ... .
Fuchs, line bundle
theory
1.b.5 Equivariant Spin
C
structures
After discussing several S
1
-invariant structures, we nowwant to force our Spin
C
struc-
ture and the associated Dirac operators to be S
1
-equivariant. The construction works
as follows:
The S
1
-action on M induces an S
1
-action on the tangent bundle and, therefore, also
on P
M
. An S
1
-equivariant Spin
C
structure is now a Spin
C
-bundle

P
M
with the usual
properties and, additionally, an S
1
-action which commutes with the map

P
M
P
M
.
On the associated bundle S
C

P
M

, we now have an S
1
-equivariant Clifford multiplic-
ation. If we also take an S
1
-equivariant connection, we get an S
1
-equivariant Dirac
operator D.
Since L can be written as

P
M

Spin
C C, where the action on C is given by
S
1 , we see
that L becomes an S
1
-equivariant hermitian line bundle.
14 CHAPTER 1. INTRODUCTION AND BACKGROUND MATERIAL
1.c Geometric Quantization
There are different kinds of geometric quantization. We will restrict ourselves to the
so-called Spin
C
quantization; other related quantization schemes will be discussed
at the end of the next chapter.
Our main sources will be [Fuchs08] and [Da Silva00].
1.c.1 Spin
C
quantization
We assume that we have chosen an S
1
-equivariant Spin
C
structure with respective
Dirac operator D. Our quantizationwill be a virtual, nite-dimensional S
1
-representation,
i.e. the formal difference of two S
1
-representations. We derive this space from the
equivariant index of D
+
.
Every elliptic operator over a closedmanifoldhas nite-dimensional kernel andcoker-
nel. They form a virtual vector space [ker ] [coker ], the index space. Since D and
also D
+
are S
1
-equivariant, [kerD
+
] and [cokerD
+
] are S
1
-representations. Their dif-
ference is dened to be the quantization Q(M, P) of M with Spin
C
structure P.
show that it is in-
dependent of the
chosen S
1
-invariant
connection
Why did we choose D
+
instead of D? As we stated above, the operator D is self-ad-
joint. From spectral decomposition we see that the index of D has to vanish always.
Hence there is no use looking at ind(D) (this is also one of the main reasons why we
only consider quantization in even dimensions: There is no splitting of the Dirac op-
erator in odd dimensions, therefore the quantization space is always zero). ind(D
+
)
and ind(D

) produce the same index with opposite signs; thus it is only a matter of
convention which of them we choose.
What "kindof" quantizations canexist? Remember that all irreducible S
1
-representations
are of the formexp(2in), n Z. So Qis a nite sumof those representations, marked
with positive or negative sign.
1.c.2 Relation to "Physical Quantization"
What is the quant-
ization map from
functions to unit-
ary operators on
Q(M, P)?
Chapter 2
Additivity under Cutting
The aim of this chapter is to prove the Cutting-Additivity of Spin
C
quantization. The
idea is the following: We choose a hypersurface Z in M with some nice properties and
call it splitting hypersurface. With the help of Z we construct two new manifolds M

cut
which represent the two sides of Z. Now we get the formula:
Q(M) =Q

M
+
cut

+Q

cut

.
The key to understand the quantization spaces is to understand the representations
at xed points. They can be expressed with the Atiyah-Segal-Singer index theorem,
which will be discussed later.
We start with discussing the process of cutting and dening the relevant terms. After
discussing the xed point issues we prove the theorem stated above and give an ex-
ample.
2.a Manipulation of Spin
C
structures
At rst we want to discuss the product of two Spin
C
structures, i.e. if we have two
Spin
C
manifolds M
m
and N
n
, how do we construct a Spin
C
structure on MN?
There is natural map Cl
m
Cl
n
Cl
m+n
coming from the inclusions R
m
R
m+n
and
R
n
R
m+n
. Since this map respects gradings and norms, we get a map Spin(m)
Spin(n) Spin(m+n) and by complexication a map
Spin
C
(m) Spin
C
(n) Spin
C
(m+n).
If

P
M
and

P
N
are the respective Spin
C
-bundles, we can dene a product bundle by

P
MN
=

P
M


P
N

Spin
C
(m)Spin
C
(n)
Spin
C
(m+n).
15
16 CHAPTER 2. ADDITIVITY UNDER CUTTING
It is easy to check that this bundle is a S
1
-equivariant Spin
C
-bundle over MN for the
induced S
1
-action on the product. If L
M
and L
N
are the associated line bundles, then
L
MN
is given by the exterior tensor product L
M
L
N
(this can also we written as the
usual tensor product on the pullback bundles).
Next we want to discuss the restrictionto aninvariant hypersurface, meaning a co-ori-
ented submanifold Z of codimension 1 which is invariant under the action of S
1
.
Let v
Z
be the positive unit normal vector eld for Z. By choosing a collar, we can take
a chart, where Z is represented by R
n1
in R
n
and v
Z
is mapped to the unit vector
e
n
. So, locally, we can dene an embedding of P
Z
into P
M
by embedding SO(n 1)
into SO(n) (so that the image are the matrices with e
n
in the last column). Since v
Z
is dened globally on Z, the embedding works globally as well, so that we get a map
P
Z
P
M
. We have the following Pullback diagram:
Spin
C
(n 1)

SO

//
Spin
C
(n)

SO

SO(n 1)

//
Spin
C
(n)
This diagram tells us the following: If we take the product bundle of P
Z
and

P
M
|Z
over Z, we can replace the structure group Spin
C
(n) SO(n 1) by Spin
C
(n 1) by
the surjective map that comes from the pullback diagram. We call this new principal
Spin
C
(n 1)-bundle

P
Z
. It is a Spin
C
structure for P
Z
(follows from the diagram as
well). The associated line bundle is just the restriction of L to Z.
At last we consider the quotient of a manifold by the S
1
-action. Since we want to apply
this method to the hypersurface, we name the manifold in this paragraph Z. We want
to show the following (intuitive) fact: If we have an S
1
-equivariant Spin
C
-structure
over a manifold Z
n1
with a free S
1
action, we get a (non-equivariant) Spin
C
structure
over Z/S
1
.
Like before the action of S
1
induces a vector eld
Z
on Z, which is nonzero every-
where since the action is free. Hence we can dene an (n2)-dimensional subbundle
V of T Z by taking the orthogonal complement in the (n1)-dimensional bundle T Z.
The associated SO(n 2)-bundle P
V
can therefore be embedded into P
Z
(we embed
SO(n 2) into SO(n 1) by xing the direction of
Z
). Again, we have a pullback dia-
gram
Spin
C
(n 2)

SO

//
Spin
C
(n 1)

SO

SO(n 2)

//
Spin
C
(n 1)
which allows us to dene a Spin
C
(n2)-structure

P
V
for P
V
, i.e. for the vector bundle
V over Z. Since V is equal to the pullback of T(Z/S
1
) and everything is S
1
-equivariant,
we see that

P
V
is also the pullback of a Spin
C
structure over Z/S
1
. This will be called

P
Z/S
1 .
2.B. SPIN
C
CUTTING 17
2.b Spin
C
cutting
At rst we give a list of the necessary ingredients:
1. A manifold M
n
as before (compact, oriented, with S
1
-equivariant Riemannian
metric)
2. A co-oriented submanifold Z of dimension n 1 with free (induced) S
1
-action
with the additional property that M \ Z consists of exactly two components M
+
and M

. (Z is also called reducible splitting hypersurface.)


3. An S
1
-equivariant Spin
C
-structure

P
M
over M.
Lemma 2.b(i). There is an invariant smooth function : M R, such that
1
(0) = Z,

(0, )

=M
+
,
1

(, 0)

=M

and 0 is a regular value for .


Remark 2.b(ii). In the prequantization process, which is discussed in the next chapter,
the role of this function will be played by the moment map discussed in the preceding
chapter.
Proof. Choose a collar for Z and cover this collar by charts of the form U

(1, 1).
On each of these charts, we choose the function (x, t ) t . Furthermore, choose the
constant functions 1 on M

. If we combine all these functions by a partition of unity


subordinate to the cover given by the U

(1, 1) and M

, we get a non-invariant
function

, fullling all properties except for the S


1
-invariance.
If
Z
denotes the dual of
Z
, we can integrate

along the curves dened by the


S
1
-action. This gives a map

S
1
on M/S
1
, which can be pulled back to dene .
Why is this still
smooth?
We dene the anti-diagonal action of S
1
on MC by
a (m, z) =(a m, a
1
z).
Now let
M
+
cut
=

(m, z)|(m) =|z|


2

S
1
In the same fashion we use the diagonal action to dene
M

cut
=

(m, z)|(m) =|z|


2

S
1
Lemma 2.b(iii). The spaces M
+
cut
and Mm are oriented manifolds with an induced
S
1
-action and S
1
-invariant Riemannian metric.
18 CHAPTER 2. ADDITIVITY UNDER CUTTING
Proof. If we take the M-like S
1
-actionon MCgivenby a(m, z) =(am, z), we see that
MCbecomes anoriented manifold with S
1
-invariant Riemannianmetric (taking the
usual metric and orientation on C).
In the following paragraphs, we will only discuss the case M
+
cut
since the other case is
similar.
On this manifold, 0 is a regular value of

:=|z|
2
(because this property holds for
). As

is also S
1
-equivariant with respect to the anti-diagonal action, we see that

Z :=

1
(0) is a manifold with S
1
-action. This S
1
-action is even free because it is free
add to widetilde
on the C-component unless z =0; for vanishing z we use the fact that S
1
acts freely on
Z.
If we divide a manifold by a free action of a compact Lie group, the result is again
a manifold. So M
+
cut
=

Z/S
1
is a manifold. Now we want to show that it inherits an
orientation and S
1
-invariant metric.
Since

is also S
1
-equivariant concerning the M-like S
1
-action, the space

Z inherits
this S
1
-action from MC (which is, of course, normally not free).Since the anti-diag-
onal and M-like action of S
1
commute, this S
1
-action is also present on M
+
cut
.
As the metric on MC is S
1
-invariant under both S
1
-actions, it induces a metric on
M
+
cut
which is still S
1
-invariant with respect to the M-like action.
For the orientation, we do the following: Since 0 is a regular value of

, we know that

Z divides MC into the positive and negative part with respect to



. Therefore, we
have a positive unit vector eld, which can be used to impose an orientation on

Z.
In general, a non-vanishing vector eld on an oriented manifold induces an orient-
ation on the orthogonal complement. Here, we use the vector eld

Z
induced by
the anti-diagonal action for this purpose. The orientation on

Z
gives us the required
orientation on M
+
cut
.
Now the manifolds M
+
cut
and M

cut
shall be equipped with S
1
-equivariant Spin
C
struc-
tures coming from the Spin
C
structure on M. We again start on the manifold MC.
We take the trivial Spin
C
structure CSpin
C
over C and and equip it with a S
1
-action
in the following way:
e
i

z, (g, b)

e
i
z, (x
/2
g, e
i/2
b)

, (2.b-1)
where x

is dened as cos +sin e


1
e
2
Spin(2). The value /2 is only well-dened
up to multples of ; but this does not matter since for = 2, the action on (g, b) is
trivial, since we quotiened out 1. This Spin
C
structure shall be given the name it
deserves:

P
+
C
.
Following the steps of 2.a, we rst construct the product Spin
C
structure

P
MC
on
MC from the

P
M
and

P
+
C
. It is equivariant with respect to the anti-diagonal action
on MC.
2.C. KOSTANT FORMULA FOR ISOLATED FIXED POINTS 19
We nowmake a restrictionof

P
MC
to

Z =

(m, z) | (m) =|z|


2

, giving a Spin
C
-structure

Z
.
After that we use the free anti-diagonal action to used the quotient construction, giv-
ing a Spin
C
structure on M
+
cut
. This Spin
C
structure is also equivariant with respect to
the M-like action since this action commutes with the anti-diagonal action. It will be
called

P
+
cut
.
For the construction of

P

cut
over M

cut
, we have to adjust a few steps.
1. We dene

P

C
to be the same bundle as

P
+
C
, but with different action given by
Change notation for
Spin
C
elements to
[, ]
e
i

z, (g, b)

e
i
z, (x
/2
g, e
i/2
b)

, (2.b-2)
2.

Z is now given by

(m, z) | (m) =|z|


2

.
The rest goes through unchanged.
At last we want to discuss the effect on the associated line bundles. We name the line
bundles for

P
M
,

P
+
cut
, Pm,

P
+
C
and

P
C
by L, L
+
cut
, L

cut
, L
+
C
and L

C
respectively. Then we
have the following remark:
Remark 2.b(iv). We have L

=
l
CC, where

=
l
denotes anisomorphismas line bundles,
but not (necessarily) as equivariant line bundles. The action on these bundles is given
by (see 2.b-1 and 2.b-2):
a (z, b) =

a
1
z, a b

for

P

cut
.
Following the construction we get for L

cut
:
L

cut
=

L L

S
1
.
Here induces the product S
1
-action on the line bundles. Notice that

Z and /S
1
mean different things for 1.
2.c Kostant Formula for Isolated Fixed Points
We use a manifold M as before and want to give a description of Q(M) by using
The dimension of
M shall always be
n =2m, not 2k
the representations of S
1
at xed points. In this section, we restrict ourselves to an
S
1
-action with isolated xed points (the xed point set will be denoted M
S
1
. Let
: S
1
C be the virtual character of Q(M).
First, we need the following technical lemma (which does not need isolated xed
points):
20 CHAPTER 2. ADDITIVITY UNDER CUTTING
Lemma 2.c(i). Let x M
S
1
be a xed point. As described in 1.b.2, we choose a complex
structure J on T
x
M which is compatible with the S
1
-action. Let
1
, . . . ,
m
Z be the
weights of the complex representaionof S
1
onT
x
M. Furthermore, let Zbe the weight
of the action of S
1
on L
x
. Then
1
2

m
j =1

is also a weight, i.e. an element of Z.


Proof. We will construct a representation of S
1
with weight
1
2

m
j =1

by combin-
ing a homomorphism fromS
1
to (a maximal torus in) Spin
C
(2m) with a map from the
maximal torus of Spin
C
(2m) to S
1
. The details are as follows:
Decompose T
x
M into V
1
. . . V
m
, so that S
1
acts on V
i
with exp

i
. Fix a point p

P
M

x
.
For each g S
1
, there is a unique element [A
g
, w
g
] Spin
C
(2m) such that g p = p
[A
g
, w
g
] where the second denotes the right action of Spin
C
(2m) on

P
M
. This denes
a homorphism : S
1
Spin
C
(2m).
Now let e
C
1
, . . . , e
C
m
be a complex basis of T
x
M with e
C
i
V
i
. With respect to this basis,
an element g S
1
acts on T
x
M by the matrix:
A

g
=

exp

1
(g) 0
.
.
.
0 exp

m
(g)

U(m) SO(2m)
This enables us todene another homomorphism

: S
1
SO(2m)S
1
by g

g
, exp

(g)

.
Remember that we can dene a map : Spin
C
(2m) SO(2m) S
1
by using the pro-
jection
SO
and the map [A, w] w
2
.
We now claim that

= : On the SO(2m)-factor this is true because



P
M
is an
S
1
-equivariant bundle. On the S
1
-factor this follows directly from the construction
of L.
The two pre-images of A

g
under the map
SO
can described explictly. Let R be
dened by the equation g =expi. Then we have:
A
g

1
SO

A
g

g
=
m
i =1

cos

+sin

e
C
i
J(e
C
i
)

Spin(2m).
Here we interpret Spin(2m) as a subspace of Cl
2m
. If you remember that Spin(2m) acts
on R
2m
by conjugation and if you take e
C
1
, J

e
C
1

, e
C
2
, . . . as an orthogonal basis of Cl
2m
you can check this ad hoc formula by direct calculation.
Now

g
, exp
/2
(g)

form elements of
1

g
, exp

(g)

. Since maps g =1 to [1, 1] in


Spin
C
(2m), we must have A
g
= A

g
.
Putting this all together, we have shown that the image of lies in the (maximal) torus:
T
Spin
C
(2m)
=

m
j =1

cost
j
+sint
j
e
C
j
J

e
C
j

, u

t
j
R, u S
1

Spin
C
(2m).
2.C. KOSTANT FORMULA FOR ISOLATED FIXED POINTS 21
We now dene the map fromT
Spin
C
(2m)
to S
1
by

m
j =1

cost
j
+sint
j
e
C
j
J

e
C
j

, u

u exp(i

j
t
j
),
which we get by mapping e
C
j
J

e
C
j

to i for all j and multiplying the two factors. It


is well-dened since

e
C
j
J

e
C
j

2
= i
2
= 1 and the multiplication of the two factors
eliminates the -ambiguity for elements [A, w] of Spin
C
(2m).
If we now compose and , we get a map : S
1
S
1
given by
exp(i) exp

1
2
i

.
Since this map is well-dened, we know that
1
2

m
j =1

is an integer.
The following proposition will be important for our calculations.
Proposition2.c(ii). Assume that the xedpoints M
S
1
of the S
1
-actionon M are isolated.
For each x M
S
1
, choose a compatible complex structure on T
x
M and denote by
1.
1,x
, . . . ,
m,x
Z the (complex) weights of the action of S
1
on T
x
M.
2.
x
the weight of the action on L
x
.
3. (1)
x
the orientation sign: It is +1 iff the orientation coming from the complex
structure on T
x
M coincides with the orientation of T
x
M coming from (the ori-
ented manifold) M.
Then the character : S
1
C of Q(M) ist given by
() =

xM
S
1

p
()
p
() =(1)
p
exp

x
/2
()
m
j =1
exp

j ,x
/2
() exp

j ,p
/2
()
(1exp

j ,p
())(1exp

j ,p
())
,
where exp
n
is the representation S
1
S
1
for the weight n Z.
Remark 2.c(iii). 1. Note that, although the invidual "weights"
j ,x
/2 may not be
integers, the whole expression is well-dened since it can be written in the form
(1)
p
exp
(
x

j

j ,x
)/2
()
m
j =1
1exp

j ,x
()

1exp

j ,x
()

1exp

j ,x
()
.
2. Since the xed points M
S
1
are isolated, all weights
j ,x
are non-zero (which is
consequence of the slice theorem, see [Guillemin02], appendix B).
22 CHAPTER 2. ADDITIVITY UNDER CUTTING
Proof of Proposition. Here we need some representation theory of Spin
C
.
give a sketch of the
proof, following pa-
per [7]
We nowwant to apply this formula to give an expression for the multiplicity of exp
n
in
Q(M) for every n. In 1.b.2 we said a change of complex structure can transform
j ,x
into
j ,x
. To avoid this problem, we impose a polarization, i.e. we alter the complex
structure J
i
(if necessary) to make all
j ,x
positive (following the remark above, they
are never zero).
Denition 2.c(iv). For n Z, we denote by Q(M)
n
the multiplicity of exp
n
in Q(M).
Furthermore, let
N
x
() :=#

(k
1
, . . . , k
m
)

Z+
1
2

+
m

j =1
k
j

j ,x
=0, k
j
>0

for Z
1
2
,
where the right hand side is always nite since all
j ,x
are positive.
Theorem2.c(v). For any weight n Z, we have
Q(M)
n
=

xM
S
1
(1)
x
N
x

n
1
2

.
Proof. We x a point x M
S
1
and set
j
=
j ,x
, =
x
. If we take the formula fromthe
remark above and divide numerator and denominator by

1exp

j
()

, we get

x
() =(1)
x
exp
(

j

j
)/2
()
m

j =1
1
1exp

j
()
.
Note that
m

j =1
1
1exp

j
()
=

nZ
N
x
(n) exp
n
(),
where N
x
(n) is the number of non-negative integer solutions (k
1
, . . . , k
m
) Z
m
0
for
n +
m

j =1
k
j

j
=0.
This result follows directly from the geometric series expansion of each factor. Plug-
ging that in we get

x
() =(1)
x

nZ
N
x
(n) exp
n+(

j

j
)/2
().
Now we shift by the integer (

J
)/2 and get

x
() =(1)
x

nZ
N
x

n
1
2

1
2

exp
n
().
2.D. KOSTANT FORMULA FOR NON-ISOLATED FIXED POINTS 23
If you compare the denitions of N
x
and N
x
, we see that
N
x

n
1
2

1
2

=N
x

n
1
2

,
and, therefore, we have

x
() =(1)
x

nZ
N
x

n
1
2

exp
n
().
Now we sum this over all xed points and get
() =

nZ


xM
S
1
(1)
x
N
x

n
1
2

exp
n
(),
from which we derive the required formula by equating the coefcients.
2.d Kostant Formula for Non-Isolated Fixed Points
Now we want to give a generalized formula for the case of non-isolated xed points.
First, we need a few more things about equivariant cohomology.
2.d.1 The A-roof class andequivariant cohomology for trivial S
1
-manifolds
In 1.b.4, we dened the equivariant rst Chern class c
1
(K) for an equivariant line
bundle K. For the Atiyah-Segal-Singer index theorem, we need to further construc-
tions which are derived from this. These constructions will be applied to an equivari-
ant complex vector bundle E over a manifold X.
For the following denitions we use the equivariant splitting principle. Following it,
we canassume that E splits as the sumof equivariant complex line bundles K
1
. . .K
k
for the purpose of constructing characteristic classes. This is justied by Appendix I
of [Guillemin02].
justify equivariant
splitting principle
We can then dene expressions depending on the (equivariant) rst Chern classes of
K
1
, . . ., K
k
.
First we discuss the non-equivariant case. Let f (x) be any formal power series in x
(over R). Then, for any H
2
(M; Z), the term f () is a well-denedobject in H

(X; Z)
since
j
=0 for j >m (we use the ring structure of H

(X; Z) givenby the cup product).


Now we can dene an object f (E) by
f (E) := f

c
1
(K
1
)

. . . f

c
1
(K
k
)

.
24 CHAPTER 2. ADDITIVITY UNDER CUTTING
The splitting principle tells us that this term is well-dened (especially independent
of the K
i
). It is common to write f as a (meromorphic) function and mean by that its
Taylor series in the point 0. Let

A be the function

A :=
x
exp(x/2) exp(x/2)

A(0) :=1
which is holomorphic near zero and can therefore be expanded.

A(E) is called the
A-roof-class of E.
Now we want to discuss the equivariant case.
We only want to look at manifolds X = F with trivial S
1
-action, i.e. only consider
vector bundles with brewise S
1
-action. The reason for this is the following: The
Atiyah-Segal-Singer index formula applies to the xed point set of the S
1
-action on
M which consists of components F with trivial S
1
-action.
In 1.b.4, we gave an explicit description of equivarivant rst Chern classes. Let K
j
be an equivariant line bundle over F (coming from the splitting principle), where the
S
1
-action is brewise given by z
n
j
for a xed n
j
, and let
j
be a connection one-form
for the associated S
1
-bundle. It is automatically S
1
-equivariant, since the S
1
-actionon
K
j
is just a multiple of the right S
1
-actionof the principal S
1
-bundle. Hence
1
2
d
j
=:

j
is just a representative of the usual rst Chern class c
1
(K
j
). Furthermore =

j
(
P
) = n
j
= 2n
j
, since
j
applied to the innitesimal generator of z
1
has to be 1
since this action equals the right action of the principal bundle. Putting this together,
we get the following expression for the equivariant rst Chern class
c
1
(K
j
) =

j
n
j
u

.
We want to use this to dene classes in H
#
S
1
(F). First we take
exp

1
2
c
1
(L)

:=exp

1
2

exp

1
2
inu

.
Since exp

1
2
inu

is an analytic function on the whole of s


1
this denes an element in
H
#
S
1
(F).
The second term is more complicated. Let f be the meromorphic function
f (z) :=
1
e
z/2
e
z/2
.
It has poles in {2i, 4i, . . .}. Now f

c
1
(K
j
)

is dened to be
f

c
1
(K
j
)

= f

j
n
j
u

:=

=0
f
(i )

in
j
u

c
1
(K
j
)

.
2.D. KOSTANT FORMULA FOR NON-ISOLATED FIXED POINTS 25
From above we know that f
(i )

in
j
u

is an analytic function on s
1
\Z

2/n
j

. Now,
analogous to the denition of

A, we set

A
e
(E) :=

j
f

c
1
(K
j
)

From the splitting principle we know that this is independent of the choice of the K
j
. Give some argu-
ment for this It is a function from s
1
\Z{/

j
n
j
} to H

(M; C).
2.d.2 The Atiyah-Segal-Singer xed point formula for Spin
C
Dirac operat-
ors
We now want to describe the character of Q(M) in terms of characteristic classes
like above.
Let F be any component of the xed point set of the S
1
-action on M. At every point
y F, we get a real representation of S
1
on the vector space T
y
M. Now T
y
M splits
into the trivial representation Triv
y
and the non-trivial part N
y
. Since the non-trivial
part has to be even-dimensional, this is also true for the trivial part. Using the Local
Linearization theorem (see B.26 in [Guillemin02]), we know that T
y
M and M are loc-
ally S
1
-diffeomorphic which implies that F is a manifold an Triv
y
is its tangent space
at y. The space N
y
then represents the normal bundle NF of F in M. The complex
dimension of NF will be called m(F) (it may be different for different F).
Now equip NF with a complex structure compatible with the S
1
-action . Also assume
check [LM] for
bundle structure
that TF can be equipped with a complex structure (this may not always possible; oth-
erwise the rst Chern classes of the splitted bundle are only dened up to a special
sign ambiguity. Since this ambiguity does not change the A-roof-class, this causes
no real problems. Details are discussed in 2.d.3.). The following proposition will be
proved in the next section by a reduction to the Atiyah-Singer index theorem:
Proposition 2.d(i). There is an >0, so that for v s
1

=R with 0 < v < the following


expressions are equal (as elements of C):

exp
S
1 (v)

FM
S
1
(1)
F
(1)
m(F)

F
exp

1
2
c
1
(L|
F
)

(v)

A(TF)

A
e
(NF)(v).
Here, (1)
F
is giving by comparing the orientationof TFNF coming fromthe complex
check the
m(F)-sign
structure and the orientation of TM.
Remark 2.d(ii).
Since is an analytic map fromS
1
to C, above formula denes it uniquely.
26 CHAPTER 2. ADDITIVITY UNDER CUTTING
M
S
1
consists of a nite number of compact manifolds: First we show that the
xed point set is closed. Take a topological generator g S
1
and consider M as
embedded into Euclidean space. Then g x x is a continuous map an the xed
point set is the preimage of the origin. As a subset of M, the space M
S
1
has to be
compact manifold, i.e. consist of nitely many compact manifolds.
The theorem will be derived from a theorem of Atiyah, Segal and Singer, originally
given in the famous paper [Atiyah68b]. In 2.d.3 we will deduce it from a theorem in
[Lawson89] and give some background information.
Now we want to give a denition of N
F
and generalise theorem 2.c(v).
Assume that NF splits into a direct sum of equivariant complex line bundles:
NF =K
F
1
K
F
m(F)
.
For each xed component F M
S
1
, denote by {n
j ,F
} the weights of the action of S
1
on K
F
j
. From the discussion at the beginning of this section we know that all n
j ,F
are
nonzero. By inverting the complex structure on K
F
j
if necessary we can assume that
all n
j ,F
are positive. Let furthermore
F
denote the weight of the action of S
1
on L|
F
.
For Z
1
2
, dene
S

:=

k
1
, . . . , k
m(F)

Z+
1
2

m(F)

=
m(F)

j =1
k
j
n
j ,F

and for each tuple

k =(k
1
, . . . , k
m(F)
) S

let
p

k,F
:=(1)
m(F)

F
exp

1
2
c
1

L|
F

1
2

j
c
1

K
F
j


A(TF) exp

j
k
j
c
1

K
F
j

.
Now dene
N
F
() :=

kS

k,F
.
Remark 2.d(iii). Let us compare this denition to the one given in the preceding sec-
tion if F is just a point. In this case all bundles are trivial, so that all Chern classes van-
ish and

A(TF) =1. This gives p

k,x
=(1)
m(F)

x
1 =(1)
m(F)
Since N
x
() was dened
Maybe we have to
exchange this sign
by mm(F).
to be the cardinality of S

, we are done.
Now we want to state and prove the generalisation of 2.c(v):
Theorem2.d(iv). For each weight n Z, the multiplicity Q(M)

of n in Q(M) is given
by
Q(M)

FM
S
1
(1)
F
N
F

1
2

.
2.D. KOSTANT FORMULA FOR NON-ISOLATED FIXED POINTS 27
Proof. To get simpler notation, we will leave out the F in the index while looking at
just one individual component. For every F, we dene a meromorphic function.For
that purpose, let f be f (y) = 1

1exp(y)

. For z = exp(iv +w) = 0 this allows us to


dene the following meromorophic function:
g
F
(z) :=z
1
2

1
2

j

j

F
exp

1
2
c
1

L|
F

1
2

j
c
1

K
j


A(TF)
m(F)

j =1
mm(F)

n=0
f
(n)
(n
j
iv +n
j
w)
n!

c
1
(K
j
)

n
Remember that the rst exponent of z is an integer due to lemma 2.c(i). For negative
w, f (n
j
iv +n
j
w) can be written as the power series
1+exp

n
j
(iv +w)

+exp

2n
j
(iv +w)

+. . . =1+z
n
j
+z
2n
j
+. . . (2.d-1)
. So for |z| <1, the function g
F
(z) has a Laurent expansion with only (possible) pole in
zero (so particularly, g
F
is meromorphic on the whole of C).
Now, if we set w =0, we get
g
F

exp(i v)

=exp

1
2

1
2

F
exp

1
2
c
1

L|
F

1
2

j
c
1
(L
j
)


A(TF)
m(F)

j =1
1
1exp

c
1
(K
j
)

following the denition of "applying a meromorphic function to an equivariant rst


Chern class"
=exp

1
2

F
exp

1
2
c
1

L|
F


A(TF)
m(F)

j =1
exp

1
2
c
1
(L
j
)

1exp

c
1
(K
j
)

F
exp

1
2
c
1

L|
F


A(TF)

A
e
(NF)
which is (up to sign) just one summand in the formula in 2.d(i). So for we have

exp(iv)

FM
S
1
(1)
F
(1)
m(F)
g
F

exp(iv)

for 0 <v < with like in theorem 2.d(i).


This means we can extend fromS
1
meromorphically to the whole complex plane in
two ways: Firstly, we take its representation as a Laurent polynomial in z and interpret
it as a meromorphic function with only (possible) pole in zero. Secondly, we take the
formula above andplug arbitrary z into g
F
. Bothextensions have to agree (uniqueness
of meromorphic extensions). This will be useful for our further investigations.
So why do we introduce the variable w if we are only interested in the case w =0? The
reason for that is that we want to use the power series (2.d-1) which is only dened for
negative w.
So let us assume that w is negative. Then we can do the following calculation:
28 CHAPTER 2. ADDITIVITY UNDER CUTTING
f

c
1
(L
j
)

=
mm(F)

n=0
f
(n)
(n
j
iv +n
j
w)
n!

c
1
(K
j
)

n
=

l
l
n
exp

l n
j
(iv +w)

c
1
(K
j
)

n
=

l
exp

l n
j
(iv +w)

l c
1
(K
j
)

n
=

l
z
l n
j
exp

l c
1
(K
j
)

Now we take the product over all j and order the terms with respect to the power of z:
m(F)

j =1
f

c
1
(L
j
)

l =0


l =

j
k
j
n
j
=0
exp

j
k
j
c
1
(L
j
)

z
l
(2.d-2)
Still assuming |z| <1, we can now write the (extended) character as
(z) =

FM
S
1
(1)
F
(1)
m(F)
z
1
2

1
2

j
n
j ,F

F
exp

1
2
c
1

L|
F

1
2

j
c
1

K
j


A(TF)

l =0


l =

j
k
j
n
j
=0
exp

j
k
j
c
1
(L
j
)

z
l
=

l =0

FM
S
1

l =

j
k
j
n
j
(1)
F
p

k,F
z
l +
1
2
(
F

j
n
j ,F
)
Now since
1
2

j
n
j ,F
is an integer, we can shift the exponent of z:
(z) =

kS

1
2

F
(1)
F
p

k,F

F
(1)
F
N
F

1
2

.
Remember that we still assume |z| < 1. The right side forms a Laurent series which
is nite in negative direction and converges at least for 0 < |z| < 1. Since (z) has no
poles outside zero, its Laurent expansion around zero describes the function every-
where. So the equation is true for all z, especially for |z| =1. The coefent of z

there-
fore equals the multiplicity of in the character . This proves the theorem.
The case m(F) =1
For the additivity under cutting, we need above formula explicitly for m(F) =1, i.e. for
(real) codimension two xed point sets.
Here, NF is just a line bundle K
F
and we have only one weight which we call n
F
. Now
S

k Z+
1
2

=k n
F
, k >0

2.D. KOSTANT FORMULA FOR NON-ISOLATED FIXED POINTS 29


is empty or has exactly one element or is empty. The expression for p
k,F
is given by
p
k,F
=

F
exp

1
2
c
1
(L|
F
)
1
2
c
1
(NF)


A(TF) exp

k c
1
(NF)

,
and this implies that
sign problem?
N
F

1
2

= p
k,F
if S

1
2

F
non-empty, otherwise zero.
2.d.3 Atiyah-Segal-Singer index theorem
Here we want to prove proposition 2.d(i) by deducing it from [Atiyah68b] (also using
the discussion of the topic in 14 of [Lawson89]). First we discuss some of the back-
ground material, but cannot go through all denitions in detail since this would imply
to undertake a course in index theory. After this discussion we will deduce 2.d(i) from
the generalised Lefschetz xed point theorem developed by Atiyah, Singer and Segal.
Equivariant K-theory
Topological K-theory builds a ring out of G-equivariant vector bundles over a topo-
logical space X (here, G denotes a compact Lie group; we only need the case G = S
1
later). We loosely follow the exposition in [Atiyah68a].
First assume that X is compact. The isomorphism classes of G-equivariant complex
vector bundles over X form a semi-group with respect to . This can be formaly ex-
tended to a group. So, a standard element of K
G
(X) has the form
[E] [F] K
G
(X)
where E and F are complex, G-equivariant vector bundles and denotes the "formal
difference". The tensor product introduces a ring structure on K
G
(X).
The pullback construction for vector bundles can also be applied to K-theory: A con-
tinuous map f : X Y induces a ring homomorphism f

: K
G
(Y ) K
G
(X). If we
assign a base point to X, we get a map K
G
(X) K
G
(). The space K
G
() is iso-
morphic to R(G), the space of virtual (nite-dimensional) representations of G. For
G =S
1
, R(G) consists of nite sums

n
z
n
,
n
Z.
Now let the reduced K-theory

K
G
(X) be the kernel of the map K
G
(X) K
G
(). The re-
spective sequence has a natural splitting map (assigning every virtual representation
a virtual trivial line bundle), so we get a natural decomposition
K
G
(X)

=

K
G
(X) K()

=K
G
(X) R(G).
30 CHAPTER 2. ADDITIVITY UNDER CUTTING
We also want to dene K-theory for locally compact spaces X
l
. We need this for dis-
cussing elements of K
G
(T X) for T X being the tangent bundle of a compact manifold
later. For this, we take the reduced K-theory of the one-point-compactication, i.e.
K(X
l
) :=

K

X
+
l

.
This denition is consistent with the denition for compact X, since X
+
is just X {}
and the tilde kills the -component.
There is a different way of dening K
G
(X) which will be important for us: Let
0 E
0

E
1



E
n
0
be a complex of vector bundles E
0
, . . ., E
n
over a space X
l
with xing X
l
and being
G-equivariant. The support of the complex shall be subset of X
l
for which is not
exact. We always assume that the support is compact. Two complexes E, F are said to
be homotopic if there is a complex G over X
l
[0, 1] which agrees with E at 0 and with
F at 1.
We denote the set of homotopy classes by C(X
l
). It has a subgroup C

(X
l
) consisting
of the complexes with empty support. Now the quotient C(X
l
)

(X
l
) is isomorphic
to K
G
(X
l
) (dening the addition inC(X
l
) by in every component).
This isomorphism will be discussed in detail in [Segal68]. At this point we only notice
that for compact X
l
, the isomorphism is just given by mapping a complex E to the
alternating sum of its elements.
So why did we introduce this second denition? It allows us to interpret the symbol

D
of an elliptic G-equivariant differential operator
D :

E
0

E
1

over a compact Riemannian manifold X in the language of K-theory. If denotes the


projection T X

=T

X X, the symbol gives us a sequence


0

E
0

E
1

0,
whichis exact for all (x, ) T

X with =0 (andso has compact support). This denes


an element

E
0
, E
1
;
D

K
G
(T X).
It can be shown that every element of K
G
(T X) can be expressed in this form, i.e. the
symbol denes a surjective map from the space Ell
G
X
of elliptic operators over X to
K
G
(T X).
Of course, this construction will later be applied to the Spin
C
Dirac operator.
2.D. KOSTANT FORMULA FOR NON-ISOLATED FIXED POINTS 31
The topological and analytical index
Now [Atiyah68a] denes two maps
a-ind :K
G
(T X) R(G)
t-ind :K
G
(T X) R(G).
We now discuss their denitions:
a-ind: We have two maps on Ell
G
X
: The symbol map to K
G
(T X) and the index map
ind to R(G). You can check that the kernel of consists of operators of index
zero. This allows us to dene a-ind as ind
1
.
t-ind: Let i : X Y be an inclusion of manifolds, with X compact. Applying some con-
structions to the normal bundle, this allows us to dene a (covariant functorial)
map
i
!
: K
G
(T X) K
G
(TY ).
Now let i
E
be a G-equivariant embedding of X into a Euclidean G-space E and
i
0
be the inclusion 0 E. From the Thom isomorphism theorem, we know that
i
0!
is an isomorphism. Hence we can dene the topological index as:
t-ind :=(i
0!
)
1
i
E!
: K
G
(T X) K

T{0}

=R(G)
Of course, it is necessary toshowthat this denitionis independent of the choice
of E. We omit the details here.
The celebrated Atiyah-Singer-index theorem says:
Theorem2.d(v). The maps a-ind and t-ind coincide.
The theorem is usually used to calculate the analytical index by topological formulae.
For extracting the formula 2.d(i), we have to do two things:
1. Reduce the topological calculation to terms involving the xed point set.This is
essentially done in [?]. We will skip most of that.
2. Express the topological index in terms of cohomology and group actions.
Characteristic classes
In general, a characteristic class can be dened as a functor which assigns to every
principal H-bundle P over X a cohomology class of X (we usually take C as coef-
cient ring, but could have taken Z as well). We denote this space by H

G
(Z) or H

G
(C)
respectively. The following things are known:
32 CHAPTER 2. ADDITIVITY UNDER CUTTING
1. If H =T is a torus, H

T
(Z) is naturally isomorphic to the character group R(T) of
T. Particulary, if x
1
, . . . , x
n
is a basis for T (e.g. given by the n projections of T
onto S
1
), we get
H

T
(C)

=C

[x
1
, . . . , x
n
]

,
the ring of formal power series in x
1
, . . . , x
n
.
2.e Additivity under cutting
2.f An example: The Two-Sphere
Chapter 3
Spinc Prequantization
3.a Cutting of a Spin
C
prequantization
33
Chapter 4
Quantization commutes with
Reduction
4.a Spin
C
reduction
4.b Main Theorem
4.c Examples
4.d Cut space
4.e Proof of Main Theorem
34
Appendix A
Short introduction to Principal
bundles
35
Todo list
Source for appropriate Wedderburn theorem . . . . . . . . . . . . . . . . . . . . . 4
Add the "almost everywhere analytic" coefcient ring, Asterisque 215, Cor. 64
and 104 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
compare signs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Here, we have to check which identication between the Lie algebra and R is
chosen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
nd source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Fuchs, line bundle theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
show that it is independent of the chosen S
1
-invariant connection . . . . . . . . . 14
What is the quantization map from functions to unitary operators on Q(M, P)? . 14
Why is this still smooth? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
add to widetilde . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Change notation for Spin
C
elements to [, ] . . . . . . . . . . . . . . . . . . . . . . . 19
The dimension of M shall always be n =2m, not 2k . . . . . . . . . . . . . . . . . 19
give a sketch of the proof, following paper [7] . . . . . . . . . . . . . . . . . . . . . 22
justify equivariant splitting principle . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Give some argument for this . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
check [LM] for bundle structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
check the m(F)-sign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Maybe we have to exchange this sign by mm(F). . . . . . . . . . . . . . . . . . . 26
sign problem? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
36
Bibliography
[Atiyah68a] ATIYAH, M.F. & SINGER, I.M. The index of elliptic operators: I. The
Annals of Mathematics, volume 87(3): pages 484530 (1968).
[Atiyah68b] ATIYAH, M.F. & SINGER, I.M. The index of elliptic operators: III. The
Annals of Mathematics, volume 87(3): pages 546604 (1968).
[Baker06] BAKER, ANDREW. Matrix Groups. Springer, third printing of rst edi-
tion (2006).
[Da Silva00] DA SILVA, A.C., KARSHON, Y. & TOLMAN, S. Quantization of presym-
plectic manifolds and circle actions. TRANSACTIONS-AMERICAN
MATHEMATICAL SOCIETY, volume 352(2): pages 525552 (2000).
[Fuchs08] FUCHS, SHAY. Spin
C
quantization, prequantization and cutting. Ph.D.
thesis, University of Toronto (2008).
[Guillemin02] GUILLEMIN, V., GINZBURG, V.L. & KARSHON, Y. Moment maps, cobord-
isms, and Hamiltonian group actions, volume 98. American Mathem-
atical Society (2002).
[Lawson89] LAWSON, H.B. & MICHELSOHN, M.L. Spin geometry. Princeton Univ
Pr (1989).
[Morgan96] MORGAN, JOHN W. The Seiberg-Witten equations and applications to
the topology of smooth four-manifolds. Princeton University Press, rst
edition (1996).
[Segal68] SEGAL, G. Equivariant K-theory. Publications Mathmatiques de
lIHS, volume 34(1): pages 129151 (1968).
37

You might also like