Raynor - RR Is A Special Case of The As Index Formula

S.C.
Raynor
The Riemann-Roch theorem is a special case of
the Atiyah-Singer index formula
Master thesis defended on 5 March, 2010
Thesis supervisor: dr. M. L ubke
Mathematisch Instituut, Universiteit Leiden
Contents
Introduction 5
Chapter 1. Review of Basic Material 9
1. Vector bundles 9
2. Sheaves 18
Chapter 2. The Analytic Index of an Elliptic Complex 27
1. Elliptic dierential operators 27
2. Elliptic complexes 30
Chapter 3. The Riemann-Roch Theorem 35
1. Divisors 35
2. The Riemann-Roch Theorem and the analytic index of a divisor 40
3. The Euler characteristic and Hirzebruch-Riemann-Roch 42
Chapter 4. The Topological Index of a Divisor 45
1. De Rham Cohomology 45
2. The genus of a Riemann surface 46
3. The degree of a divisor 48
Chapter 5. Some aspects of algebraic topology and the T-characteristic 57
1. Chern classes 57
2. Multiplicative sequences and the Todd polynomials 62
3. The Todd class and the Chern Character 63
4. The T-characteristic 65
Chapter 6. The Topological Index of the Dolbeault operator 67
1. Elements of topological K-theory 67
2. The dierence bundle associated to an elliptic operator 68
3. The Thom Isomorphism 71
4. The Todd genus is a special case of the topological index 76
Appendix: Elliptic complexes and the topological index 81
Bibliography 85
3
Introduction
The Atiyah-Singer index formula equates a purely analytical property of an
elliptic dierential operator P (resp. elliptic complex E) on a compact manifold
called the analytic index ind
a
(P) (resp. ind
a
(E)) with a purely topological prop-
erty, the topological index ind
t
(P)(resp. ind
t
(E)) and has been one of the most
signicant single results in late twentieth century pure mathematics. It was an-
nounced by Michael Atiyah and Isadore Singer in 1963, with a sketch of a proof
using cohomological methods. Between 1968 and 1971, they published a series of
papers
1
in which they proved the formula using topological K-theory, as well as
lling in the details of the original proof.
The history of the Atiyah-Singer index formula reads as aWhos Who in twen-
tieth century topology and analysis. The formula can be seen as the culmination
of a project of generalisation of index theorems that began in the mid 1800s with
the Riemann-Roch theorem (and the Gauss-Bonnet theorem), and which involved
many of the greatest names in topology and analysis of the last 150 years. It is
an achievement for which Atiyah and Singer were awarded the Abel Prize in 2004.
The signicance of their formula reaches beyond the elds of dierential topology
and functional analysis: it is also fundamental in much contemporary theoretical
physics, most notably string theory.
For the purpose of this paper however, the only results which we shall consider
are the classical Riemann-Roch theorem (1864), the Hirzebruch-Riemann-Roch the-
orem (1954), and the Atiyah-Singer index formula (1963). In fact, we will only really
look at the latter two in the context of being direct generalisations of the classical
Riemann-Roch theorem.
The (classical) Riemann-Roch theorem, proved as an equality in 1864, links
analytic properties of certain objects called divisors on compact Riemann surfaces,
with topological properties of holomorphic line bundles dened in terms of the
divisors. Though the terms involved will only be properly dened later in this
paper, it is convenient, nonetheless, to state the theorem here.
Let X be a compact Riemann surface and D a divisor on X, that is, a function
D : X Z with discrete support. Then the Riemann-Roch theorem states that
(0.1) h
0
(X, O
D
) h
1
(X, O
D
) = 1 g + deg(D).
Here h
0
(X, O
D
) is the dimension of the space of meromorphic functions f such
that, for all x X, ord
x
(f) D(x), where ord
x
(f) = n if f has a zero of order
n or a pole of order n at x, and h
1
(X, O
D
) is the dimension of another space of
meromorphic functions also with only certain prescribed poles and zeroes (we will
discuss this in detail in chapter 3). The degree, deg(D), of the divisor D is the sum
1
The index of elliptic operators: I-V. (Paper II from 1968 is authored by Atiyah and Segal,
rather than Atiyah and Singer.) [AS1, AS2, AS3, AS4, AS5].
5
6 INTRODUCTION
of its values over X. Since X is compact, the support of D is nite and so deg(D)
is well-dened. Finally g denotes the genus of the surface X. It is clear that these
are all integral values.
The left hand side of equation (0.1) can be described in terms which depend
on the holomorphic structure of certain line bundles on X, whilst we shall see that
the right hand side depends only on the topology of these bundles.
There is a natural equivalence relation on the space of divisors of a Riemann
surface X and it will be shown that there is a one to one correspondence between
equivalence classes of divisors on X and isomorphism classes of line bundles on X.
(This will be described in chapter 3.)
The Riemann-Roch theorem provides the conditions for the existence of mero-
morphic functions with prescribed zeroes and poles on a compact Riemann surface.
Its signicance did not go unnoticed and its implications were studied by many of
the greatest names in topology and analysis (even including Weierstrass). Interest-
ingly it was initially regarded fundamentally as a theorem of analysis and not of
topology.
It was not until 1954, nearly a century after its original discovery, that Hirze-
bruch found the rst succesful generalisation of the Riemann-Roch theorem to
holomorphic vector bundles of any rank on compact complex manifolds of any di-
mension.
2
This came a few months after J.P Serres 1953 discovery of what is now
known as Serre duality, which provides a powerful tool for calculation with the
Riemann-Roch theorem, but also deep insights into the concepts involved. Serre
had applied sheaf theory to the Riemann-Roch theorem and Hirzebruch also used
these newly emerging methods of topology to nd techniques suitable for the project
of generalisation. The so-called Hirzebruch-Riemann-Roch theorem says that the
Euler characteristic (E) of a holomorphic vector bundle E on a compact complex
manifold X is equal to its T-characteristic T(E). We will dene these terms in
chapters 3 and 4. Of signicance here is that, in the case that the X has dimension
1 and E rank 1, if D is the divisor that corresponds to E, then the Euler charac-
teristic (E) is equal to the left hand side of equation (0.1) and T(E) is equal to
the right hand side of (0.1).
After Hirzebruchs theorem, progress to the Atiyah-Singer index formula was
very swift indeed. Grothendieck discovered the Grothendieck-Riemann-Roch theo-
rem around 1956
3
, and the Atiyah-Singer index formula was published in its com-
plete form in 1964.
The Atiyah-Singer index formula is a direct generalisation of the Hirzebruch-
Riemann-Roch theorem since we can assosciate a certain elliptic complex (E) with
any holomorphic vector bundle E on a compact complex manifold X, and it can
be shown that (E) = ind
a
((E)) and T(E) = ind
t
((E)).
In this paper, we will show how the original Riemann-Roch theorem, formu-
lated for divisors on compact Riemann surfaces, is a special case of the Hirzebruch-
Riemann-Roch Theorem and the Atiyah-Singer index formula. The paper does not
set out to prove any of these theorems. One of the most striking features of the
2
These results can be found in [Hi], originally published as Neue topologische Methoden in
der algebraischen Geometrie in 1956.
3
Grothendieck had originally wished to wait with publishing a proof. With Grothendiecks
permission, a proof was rst published by Borel and Serre [BS] in 1958.
INTRODUCTION 7
Atiyah-Singer index formula, and a good illustration of the depth and signifcance
of the result, is that it admits proofs by many dierent methods, from the initial
cohomology and K-theory proofs, to proofs using the heat equation. We will limit
ourselves here to a cohomological formulation of the formula since this is the most
natural choice when dealing with the Riemann-Roch theorem. However it is per-
haps worth mentioning that the K-theoretic formulation lends itself best to a more
general exposition on the Atiyah-Singer index formula.
The paper begins with two purely expository chapters. Chapter 1 sets out
the basic denitions and notations concerning vector bundles, sheaves and sheaf
cohomology which will be used throughout the paper. Most proofs will not be
given. In chapter 2, elliptic dierential operators, complexes and the analytic index
of an elliptic complex will be dened and a number of examples will be given.
The substantial part of the paper begins in chapter 3. Divisors on a Riemann
surface X are dened and the Riemann-Roch theorem is stated in terms of divisors.
By constructing a holomorphic line bundle L = L
D
on X, associated with the
divisor D, it is then shown that the left hand side of the Riemann-Roch equation
(0.1) can be interpreted as a special case of the analytic index of an elliptic operator.
Finally we show that this also corresponds to the Euler characteristic (L) of L on
a Riemann surface.
In chapter 4, we turn to the right hand side of the Riemann-Roch equation
(0.1) and show that this can be described in terms of purely topological properties
of the surface X and the bundle L = L
D
. To this end we also dene the rst
Chern classes for the line bundles L
D
over X. However, the formulation we obtain
for the right hand side of the equation (0.1) is not yet the formulation for the
topological index, ind
t
, of the Atiyah-Singer index formula or the T-characteristic
of the Hirzebruch-Riemann-Roch theorem.
Chapter 5 provides the rst step in this further path of generalisation. We show
how the Chern classes dened in the previous chapter as topological quantities of
holomorphic line bundles over Riemann surfaces, can be generalised to properties of
rank r holomorphic bundles over compact complex manifolds of higher dimension n.
We then dene a number of topological objects on vector bundles which are needed
in the description of the T-characteristic and the topological index. Most proofs
will be omitted from these expository sections. This information leaves us in a
position to show that the right hand side of the Riemann-Roch equality (0.1) is a
special case of the T-characteristic of a holomorphic bundle over a compact complex
manifold. We will therefore have shown that the classiscal Riemann-Roch theorem
is a special case of the Hirzebruch-Riemann-Roch theorem.
In the nal chapter 6 it remains to show how, in the case of a holomorphic
line bundle L over a compact complex Riemann surface X, the T-characteristic of
L is equal to the topological index of L. In doing so we complete the proof that
the classical Riemann-Roch theorem is a special case of the Atiyah-Singer index
formula.
Unfortunately, there is not space in this paper to show the more general result
that the Hirzebruch-Riemann-Roch theorem for higher dimensions is implied by the
Atiyah-Singer index formula. However, in the appendix we shall briey describe
some steps that are necessary for doing this.
CHAPTER 1
Review of Basic Material
This chapter serves to review the some of the basic concepts and to establish
the notation that we will be using in the rest of the paper. Most proofs of the
results will not be included. The books [We], [Fo], [Hi] are excellent sources for
this material.
Throughout the paper we will assume that the base manifold X is paracompact
and connected.
1. Vector bundles
1.1. Vector bundles, trivialisations, frames and forms. Familiarity with
vector bundles is assumed in this paper. The purpose of this section is not to in-
troduce new material but to establish the notation and conventions for the rest of
the paper.
In the following, the eld K can be R or C. Let U be an open subset of K
n
.
We will use the following notation:
((U) refers to the collection of K-valued continuous functions on U.
c(U) refers to the collection of K-valued dierentiable functions on U.
O(U) refers to the collection of C-valued holomorphic functions on U.
In general we will refer to o- functions and o-structures where o = (, c, O.
In this paper we will be dealing with manifolds with real dierentiable and
complex analytic (holomorphic) structures. That is, manifolds such that the tran-
sition (change of chart) functions are real dierentiable or holomorphic. We will
call these c-, and O- manifolds respectively.
Definition 1.1. Let E, X be Hausdor spaces and : E X be a continuous
surjection. : E X is called a K- vector bundle of rank r over the base space
X with total space E if
(1) There exists an open cover | = U
i
iI
of X and, for all i I, there
exists a homeomorphism
i
:
1
(U
i
) U
i
K
r
such that
i
(E
x
) = x K
r
, for all x U
i
where E
x
:=
1
(x) is the bre of E over x.
For x U
i
, (U
i
,
i
) is called a local trivialisation of : E X at x.
A local trivialisation of E over X is a collection (U
i
,
i
)
iI
.
(2) For all i, j I we dene the transition function g
i,j
:=
i
1
j
[
(UiUj)K
r .
Then, for all x U
i
U
j
, the map
K
r

= x K
r gi,j
x K
r

= K
r
9
10 1. REVIEW OF BASIC MATERIAL
is a linear isomorphism.
We usually simply say that E is a vector bundle over X and rk E = r.
Remark 1.2. For x U
i
, identifying the bre E
x
with K
r
= x K
r
via
i
gives E
x
the structure of an r-dimensional K-vector space. By (2), this is
independent of the choice of i I with x U
i
.
Definition 1.3. For o = c, O, a vector bundle E over X is an o-bundle if E
and X are o-manifolds, : E X is an o-morphism, and the local trivialisations
are o- isomorphisms. It is easily seen that this last condition is equivalent to the
transition functions being o-morphisms.
Remark 1.4. Note that the denitions imply that, if : E X is an o- bundle
over X with local trivialisation (U
i
,
i
)
i
relative to some cover | = U
i
i
, then, if
i
: U
i
GL(n, K)
i
are o-maps on U
i
, (U
i
,
i

i
)
i
is also a local trivialisation
for E.
We calculate the transition functions g
i,j
i,j
for E relative to (U
i
,
i
)
i
, in
terms of the transition functions g
i,j
i,j
for E relative to (U
i
,
i
)
i
:
By denition
g
i,j
=
i

1
j
, on U
i
U
j
so
g
i,j
=
1
j
= (
i

i
) (
1
j

1
j
) =
i
g
i,j
1
j
, on U
i
U
j
.
Example 1.5. The complex projective line CP
1
is a compact Riemann surface.
A point in CP
1
can be specied in homogeneous coordinates [z
0
: z
1
] where z
o
, z
1

C and z
0
and z
1
are not both zero. Then
[z
0
: z
1
] = [z
0
: z
1
] if [z
0
: z
1
] = [z
0
: z
1
], C
.
(C
denotes the non-zero complex numbers.)

We view CP
1
as the space of complex lines l in C
2
which go through the origin
and dene O
CP
1(1) as the submanifold of CP
1
C
2
given by
O
CP
1(1) = (l, p) : p l = ([z
0
: z
1
], (z
0
, z
1
)) : C CP
1
C
2
.
Now, CP
1
= U
0
U
1
where, for i = 0, 1, U
i
is the open set given by
U
i
:= [z
0
: z
1
] CP
1
: z
i
,= 0.
We wish to show that : O
CP
1(1) X (where (l, p) = l) is a holomorphic line
bundle over CP
1
:
Local trivialisations
i
:
1
(U
i
) U
i
C, i = 0, 1 are given by
0
: ([1 : z], (, z)) ([1 : z], )
and
1
: ([w : 1], (w, )) ([w : 1], ) .
So, on U
0
U
1
, [w : 1] = [1 : z] and therefore w =
1
z
.
We calculate the transition functions relative to U
0
and U
1
.
g
0,1
=
0

1
1
[
(U0U1)C
: ([1 : z]), )
_
[1 : z], (
z
, )
_
_
[1 : z],

z
_
.
Since z is non-zero on U
0
U
1
, g
0,1
[
(U0U1)C)
is clearly holomorphic. As a
map, g
0,1
: U
0
U
1
GL(1, C) = C
,
g
0,1
([z
0
: z
1
]) =
z
0
z
1
( so g
0,1
([1 : z]) =
1
z
).
1. VECTOR BUNDLES 11
It is easy to check that g
1,0
= g
1
0,1
: ([w : 1]), )
_
[w : 1],

w
_
, and so
g
1,0
([z
0
: z
1
]) =
z
1
z
0
: U
0
U
1
GL(1, C).
Definition 1.6. Let E and F be K-vector bundles over X. A map
f : E F
is a vector bundle homomorphism if it preserves bres and f
x
= f[
Ex
is a K-linear
map for every x X. Two o- bundles are isomorphic if there is an o-isomorphism
f : E F
which is a K vector space isomorphism on the bres of E.
Proposition 1.7. For o = c, O, given a covering U
i
i
of a manifold X and
non-vanishing o-functions g
i,j
: U
i
U
j
GL(r, K) such that for all i, j, k and for
all x U
i
U
j
U
k
,
g
i,j
(x)g
j,k
(x) = g
i,k
(x)
we can construct an o-bundle : E X which has transtion functions g
i,j
i,j
with respect to the covering U
i
i
. The bundle E is unique up to isomorphism.
Proof. For an outline of this construction see [We, 13-14] .
Definition 1.8. A (local) section of a vector bundle : E X is a map from

X (or an open subset U of X) to E such that s = id
X
(resp. id
U
). We denote
the o-sections of E over X by o(E) := o(X, E). The collection of S-sections of E
over an open subspace U X is denoted by o(U, E). The o-sections of a vector
bundle E, dened by U
i
i
and g
i,j
i,j
are given by o-functions f
i
: U
i
K
r
such
that
f
i
= g
i,j
f
j
, on U
i
U
j
.
When E is the trivial line bundle X C, we write o := o(X C).
Finally, a meromorphic section f of a holomorphic line bundle L over a holo-
morphic manifold X is, relative to a trivialisation U
i
i
, a collection of meromor-
phic functions f
i
: U
i
C such that
f
i
= g
i,j
f
j
, on U
i
U
j
.
The space of meromorphic sections of a line bunle L X is denoted by /(L).
Definition 1.9. A frame at x X for a bundle E X is an ordered basis
for E
x
.
Since there is a locally trivialising neighbourhood U
x
for E, it is clear that we
can extend this and dene a frame for E above U
x
as an ordered set of sections
f = (f
i
)
i
of E over U
x
such that, for each y U
x
, (f
i
(y))
i
is an ordered basis for
E
y
. A frame over U
x
is an o-frame if the sections are o-sections
Remark 1.10. A frame for E on U X denes in a natural way a local
trivialisation of : E X and vice versa.
Namely, let f = (f
i
)
i
be a frame for E over U. We wish to construct a local
trivialisation :
1
(U)

U K
r
. Given e E
x
, x X, e =
r
i=1
i
(x)f
i
(x)
where
i
: U K is an o- function. We dene
(e) = (
i
(x), . . . ,
r
(x)).
It is easily checked that this is an o-isomorphism.
Conversely, given a trivialisation :
1
(U)

U K
r
, we can dene an
o-frame f = (f
i
)
i
over U by
f
i
(x) :=
1
(x, e
i
)
with (e
1
, . . . , e
r
) an ordered basis for K
r
.
Definition 1.11. A vector eld V on X is a continuous section of the tangent
bundle TX of X.
If E is a vector bundle over X, then
p
E denotes the bundle of p-vectors with
coecients in E. That is, for x X, the bre
p
E
x
of
p
E over x consists of
K-linear combinations of elements of the form v
1
v
p
with v
1
, . . . , v
p
E
x
,
where denotes the exterior product in the exterior algebra
_
E
x
of E
x
.
For, o = (, c, O, let o
k
(E) denote the o- k-forms of X with coecients in E.
That is
o
k
(E) := o(E
k
T
X)
where T
X is the (real) cotangent bundle of X.

(When E, X are complex, E
k
T
X := E
C

k
T
X.)
If X is a complex manifold with basis of local coordinates (z
1
, . . . , z
n
), then
(dz
1
, . . . , dz
n
) is a local frame for T, the holomorphic cotangent bundle of X. T is
dened as the bundle for which (dz
1
, . . . , dz
n
) is a local frame.
We denote by c
p,q
(E) the dierentiable (p, q)-forms of X with coecients in
E. That is
(1.1) c
p,q
(E) := c(E
p
T
q
T).
We have
(1.2) c
p
(E) =
q+r=p
c
q,r
(E).
When E = X C, we will often write simply c
q,r
:= c
q,r
(X C).
1.2. Metrics on a vector bundle.
Definition 1.12. Let E be a real dierentiable vector bundle over a real dier-
entiable manifold X. A (bundle) metric on E is an assignment of an inner product
g
x
on every bre E
x
such that such that for any open set U X and sections ,
of E over U, g(, ) is smooth on U.
Using a trivialisation and a partition of unity, it is easy to see that
Proposition 1.13. A vector bundle E over a paracompact dierentiable man-
ifold X admits a metric.
Since all base spaces in this paper are paracompact, all bundles will be metris-
able (admit a metric).
Definition 1.14. If E X is a complex vector bundle over a manifold X
then a Hermitian metric on E is the assignment of a Hermitian inner product h
x
on every bre E
x
such that for any open set U X and sections , of E over U,
h(, ) is smooth on U.
Given a Hermitian bundle (E, h) of rank r over X and a set of local frames
f = f
i
i
where f
i
= (f
i
1
, . . . , f
i
r
) for E, we can dene the function matrix
(1.3) h(f
i
) := (h
i
(f
i
, f
i
))
,
, h(f
i
, f
i
) : U
i
GL(r, C).
Then
h(f
j
) = (h
j
(g
j,i
f
i
, g
j,i
f
i
))
,
= g
i,j
t
h
i
(f
i
)g
j,i
.
Remark 1.15. The above implies that for a line bundle L over a Riemann
Surface X, dened in terms of a covering | = U
i
i
and transition functions
g
i,j
i,j
, a Hermitian metric h on L is therefore entirely dened by a collection
of positive functions =
i
: U
i
R
+
. Namely let f
i
be a holomorphic frame
for L over U
i
. Then h
i
is completely determined by
i
:= h
i
(f
i
, f
i
) > 0 which is a
continuous positive valued function on U
i
.
So a Hermitian metric on L is uniquely determined by a collection of positive
functions
i
on U
i
such that
j
= g
i,j
g
i,j
i
on U
i
U
j
.
Proposition 1.16. If E X is a smooth complex vector bundle over a com-
plex manifold X, E admits a Hermitian metric.
Proof. [We, 68].
Example 1.17. Let : O
CP
1(1) CP
1
be as in example 1.1.5. We wish to
dene a Hermitian metric h on CP
1
. If z is a local coordinate on CP
1
, the standard
Hermitian metric on CP
1
C
2
CP
1
is given by
[ (l, (, )) [
2
= [[
2
+[[
2
, l CP
1
, , C.
Since O
CP
1(1) CP
1
C
2
, we can take the restriction of this metric to O
CP
1(1).
Then, if CP
1
= U
0
U
1
as in example 1.1.5, on U
0
we have
[ ([1 : z], (1, z)) [
2
= 1 +[z[
2
,
and on U
1
we have
[ ([w : 1], (w, 1)) [
2
= 1 +[w[
2
.
On U
0
U
1
, with w =
1
z
,
1 +[w[
2
= 1 +
1
z
2
=
1
[z[
2
1 +[z[
2
= g
0,1
g
0,1
_
1 +[z[
2
_
.
So, the restriction of the standard metric on CP
1
C
2
is indeed a Hermitian metric
for O
CP
1(1).
(In the notation of remark 1.1.15, we have
0
[1 : z] = 1 + [z[
2
on U
0
and
1
[w : 1] = 1 +[w[
2
on U
1
.)
Remark 1.18. A Hermitian metric h on a complex bundle E X induces a
metric g on the underlying real vector bundle. Dene
g := Re h =
1
2
(h +h).
Then g is positive denite, symmetric, bilinear and real valued.
Definition 1.19. If : E X is a bundle, the dual bundle
: E
X
is the bundle with bres E
x
:= (E
x
)
for all x X. A choice of metric g on E

induces an isomorphism E E
: g(, ).
Proposition 1.20. If a vector bundle E on X has transition functions
g
i,j
o(U
i
U
j
, GL(n, K)) with respect to a given covering, the dual bundle E
has transtion functions (g

j,i
)
1
.
Proof. This is a simple exercise in linear algebra.
Proposition 1.21. Every complex vector bundle E over a complex manifold
X can be described by unitary transition functions.
Proof. Let h be a Hermitian metric on E and f
i
i
a collection of frames for
E. We can apply Gram-Schmidt orthonormalisation to each h(f
i
). The transition
maps so obtained are then unitary.
Proposition 1.22. If : E X is a complex bundle with Hermitian metric
h, then
E

= E
.
Proof. By the previous proposition, E can be described by unitary transition
functions g
i,j
with respect to a given covering. We have seen, in proposition 1.1.20,
that E
has transition functions (g

j,i
)
1
, but since g
i,j
(x) U(n) for all x U
i
U
j
,
(g
j,i
)
1
= g
i,j
.
In other words E

= E and we are done.
1.3. Complexication and almost complex structures. We wish to be
able to move from complex vector bundles to the underlying real vector bundle and
conversely to dene (almost) complex structures on even dimensional real bundles.
The map : GL(n, C) GL(2n, R) is the embedding obtained by regarding a
linear map of C
q
with coordinates z
1
, . . . , z
q
as a linear map of R
2q
with coordinates
x
1
, . . . , x
2q
where z
k
= x
2k1
+ix
2k
.
The map : GL(n, R) GL(n, C) is the complexication map, that is the
embedding obtained by regarding a matrix of real coecients as a matrix of complex
coecients.
We have the following commutative diagrams of embeddings:
(1.4)
U(n)

-
O(2n)
GL(n, C)
?
-
GL(2n, R)
?
(1.5)
O(n)

-
U(n)
GL(n, R)
?
-
GL(n, C)
?
where in both diagrams the vertical arrows are simply inclusion.
If X is a compact complex manifold, we can extend the maps and to maps
of vector bundles over X.
Lemma 1.23. There is an automorphism of U(2q) such that, for A U(q),
( (A)) U(2q) has the form
_
A 0
0 A
_
,
up to a permutation of coordinates. Similarly, if B O(q), then, up to a permuta-
tion of coordinates, (B) O(2q) has the form
_
B 0
0 B
_
.
Proof. (N.B. In this proof, we will not consider the permutations of coordi-
nates. However, this does become relevant when considering the orientation of the
spaces.)
(A) =
_
Re A Im A
Im A Re A
_
O(2n).
For M U(2n), let
(M) =
1
2
_
1 i
i 1
_
M
_
1 i
i 1
_
.
This is clearly an automorphism and it is easily checked that it is the desired map
: U(2n) U(2n).
We regard B O(n) as an element B = B of U(n). For M U(n),
(M) =
_
Re M Im M
Im M Re M
_
O(2n).
Since B is real
(B) =
_
B 0
0 B
_
.
Proposition 1.24. If E X is a complex bundle described by unitary tran-

sition functions, ( )(E)

= E E

= E E
.
Similarly, if W X is a real bundle, ( )(W)

= W W.
In this case the orientations of ( )(W) and W W dier by a factor
(1)
q
2
(q1)
.
Proof. By proposition 1.1.22, E is described by unitary transition functions
so E

= E and the isomorphism follows from the above lemma 1.1.23. Similarly
for the second statement. As for the orientations, ( )(W) is represented by
transition matrices g
i,j
GL(2q, R) with coordinates x
1
, y
1
, . . . , x
q
, y
q
and the
transition matrices of W W have coordinates x
1
, . . . x
q
, y
1
, . . . , y
q
.
Definition 1.25. Let V be a real r-dimensional vector space. The complexi-
cation V
C
of V is given by
V
C
:= V
R
C.
This is equivalent to V
C
= V iV and therefore there is a natural isomorphism of
R-vector spaces
V
C

= V V.
V
C
is a complex r-dimensional vector space with complex multiplication given by
(v ) = v , v V, , C.
There is a canonical conjugation map on V
C
dened by
v = v .
If W X is a real vector bundle, the complexication W
C
of W over X is
the bundle with bres (W
x
)
C
. If W is given by transtion functions g
i,j
i,j
with
g
i,j
GL(r, R), then W
C
is given by the same transition functions g
i,j
i,j
but
now with the functions g
i,j
(= (g
i,j
)) regarded as elements of GL(r, C).
Furthermore W
C
= WW, although the orientation diers by a factor (1)
r
2
(r1)
.
Definition 1.26. Given a 2n-dimensional real vector space V , there exists a
linear map J End(V ) such that J
2
= id
V
. Then J is called a complex struc-
ture for V . J gives V the structure of a complex vector space with complex scalar
multiplication dened by
(a +ib)v = av +bJv, a, b R, v V.
J can be extended to V
C
by J(v ) = Jv .
Definition 1.27. An almost complex structure on a smooth 2n-dimensional
real manifold X is a complex structure on each bre T
x
X of the tangent space TX
of X which varies smoothly with x X.
Equivalently, given a trivialisation for the tangent bundle of X with transition
functions g
i,j
i,j
with g
i,j
GL(2n, R), an almost complex structure for X is a
bundle E over X with transition functions t
i,j
i,j
, t
i,j
GL(n, C) relative to the
same trivialisation and such that (t
i,j
) = g
i,j
(for all i, j).
In particular, if X is a complex manifold then the complex tangent bundle
T = T (X) is an almost complex structure for X.
Henceforth we shall use the following notation: If X is a complex manifold then
TX denotes the real tangent bundle of X, and T
X its dual, the real

cotangent bundle of X.
T denotes the complex (holomorphic) tangent bundle of X and T its dual,
the complex cotangent bundle of X.
If T is given by transition functions f = f
i,j
i,j
, we can dene a bundle T
given by f = f
i,j
i,j
. T is the bundle dual to T .
The maps and imply the following:
Proposition 1.28. The following identities hold:
(1.6) TX
C
= T T ,
(1.7) (TX
C
)
= T
X
C
= TT.
And the r-th exterior power of T
X
C
,
r
T
X
C
is given by
(1.8)
r
T
X
C
=
p+q=r
(
p
T) (
q
T).
Proof. This follows directly from the maps and and proposition 1.1.24.
Corollary 1.29. The isomorphism T
X
C
= T
X iT
X, together with the

projection p : T
X
C
= TT T induces an isomorphism T
X T.
1.4. Connections.
Definition 1.30. A connection on a dierentiable K-vector bundle E X
with X paracompact, is a collection of K-linear maps
U
: c(U, E) c
1
(U, E), U X open,
such that, if U
U is open and c(U, E), then

(
U
)[
V
=
V
([
V
)
and which satises the Leibniz formula
U
(s) = ds +s
U
()
for any s c and any c(U, E).
If E X is a complex bundle, a connection on E can be written as
=
1,0
+
0,1
, with
1,0
: c(E) c
1,0
(E),
0,1
: c(E) c
0,1
(E).
Essentially a connection provides a rule for dierentiating a section with re-
spect to a vector eld.
Definition 1.31. If f = (f
)
r
=1
is a frame for E on an open set U, then we
can dene an r r matrix A = A(, f) of dierentials on U, called the connection
matrix of with respect to f such that
(1.9) A
,
(, f) c
1
(U), f
=
r
=1
A
,
(, f) f
.
A dierentiable section of of E over U can be written as
i
=

i
i
f
i
where
i
c(U, K). Let (f) := (
1
, . . . ,
r
). Then, by the dening properties of the
connection ,
(1.10) ([
U
) =
r
=1
_
_
d
+1
A
,
f
)
_
_
= A(, f)((f)) +d(, f)
where d(, f) :=
(d
).
Proposition 1.32. Every dierentiable vector bundle over a paracompact man-
ifold X admits a connection.
Proof. [We, 67].
Example 1.33. Let (X, g) be a Riemannian manifold (g is a metric on the
tangent bundle TX of X) with tangent bundle TX. The Levi-Civita connection
on X is the unique connection on TX which satises:
(1) For vector elds V
1
, V
2
, V
3
on X
V
1
(g(V
2
, V
3
)) = g(
V1
(V
2
), V
3
) +g(V
2
,
V1
(V
3
)).
It is then said that preserves the metric.
(2) For vector elds V
1
, V
2
on X
V1
(V
2
)
V2
(V
1
) = [V
1
, V
2
]
where [V
1
, V
2
] is the Lie bracket of TX. is then said to be torsion free.
The following proposition says that, with respect to a given Hermitian metric h
on a holomorphic bundle E X, a unique special connection with very convenient
properties called the canonical connection exists. If E is taken to be T , the complex
tangent bundle of X, this is a Hermitian analogue of the Levi-Civita connection.
Proposition 1.34. Let X be a complex manifold and E a holomorphic bundle
over X with Hermitian metric h. There exists a unique connection
(E,h)
on E
which satises.
(1)
(E,h)
is compatible with h. I.e.
(1.11) d(h(, )) = h(
(E,h)
, ) +h(,
(E,h)
)
(2) For every holomorphic section of E over any U X open, it holds that
(1.12)
0,1
(E,h)
= 0
In this case
(E,h)
is the so-called canonical connection.
Proof. [We, 78-79].
2. Sheaves
2.1. Some denitions.
Definition 2.1. A presheaf T on a topological space X is an assignment of
an Abelian group T(U) to every non-empty open U X, together with a collection
of restriction homomorphisms
U
V
: T(U) T(V )
V U
for U, V open in X. The
restriction homomorphisms satisfy:
(i) For every U open in X,
U
U
= id
U
the identity on U.
(ii) For W V U open in X,
U
W
=
V
W
W
V
.
If T is a presheaf, an element of T(U) is called a section of the presheaf T
over U.
A subpresheaf ( of T is a presheaf on X such that for all U open in X,
((U) T(U) and, if
U
V
V U
are the restriction functions for (, then
U
V
=
U
V
[
G(U)
.
Definition 2.2. Given two presheaves T and (, a sheaf morphism h : T (
is a collection of maps
h
U
: T(U) ((U)
dened for each open set U X such that the h
U
commute naturally with the
restriction homomorphisms
U
V
on T and
U
V
on (. That is, if V U, with U, V
open in X then
U
V
h
U
= h
V
U
V
.
Definition 2.3. A sheaf is a presheaf T such that for every collection U
i
i
of open sets of X with U =
i
U
i
, the following axioms are satised:
(1) For s, t T(U) such that
U
Ui
(s) =
U
Ui
(t) for all i, it holds that s = t.
(2) Given s
i
T(U
i
) such that
Ui
UiUj
s
i
=
Uj
UiUj
s
j
,
there exists an s T(U) which satises
U
Ui
s = s
i
for all i.
A subsheaf ( of a sheaf T is a subpresheaf of T which satises the
sheaf axioms 1.2.3 (1), and 12.3 (2).
2. SHEAVES 19
Remark 2.4. If T is a sheaf over X then T() is the group consisting of exactly
one element.
Example 2.5. Given a topological space X, we note that, for K = R, C, and
U X open, the space C(U, K) of continuous K-valued functions on U is a K-
algebra. So, we can dene the presheaf (
X
by (
X
(U) = C(U, K). For V U
open subsets of X, the restriction homomorphisms
U
V
are given by
U
V
(f) = f[
V
,
f C(U, K) = (
X
(U). It can easily be checked that this is a sheaf of K-algebras.
Example 2.6. For o = c, O, if X is an o-manifold, then we can dene the
sheaf o
X
by o
X
(U) := o(U, K). Then o
X
(
X
and o
X
is called the structure
sheaf of the manifold X.
Definition 2.7. Let be a sheaf of commutative rings over X. Say T is
a sheaf such that, for every U open in X, we have given T(U) the structure of a
module over (U) in a manner compatible to the sheaf structure, i.e., for (U)
and f T(U),
U
V
(f) =
U
V
()
U
V
(f)
where V U open in X and
U
V
, resp.
U
V
are the corresponding T, resp.
restrictions. Then we call T a sheaf of -modules.
Now, for p 1, we dene the presheaf
p
by
U
p
(U) := (U) (U)
. .
p times
, (
p
)
U
V
:=
U
V

U
V
. .
p times
.
A sheaf ( over X is called a locally free sheaf of -modules of rank p if ( is
sheaf of - modules and, for each x X, there is a neighbourhood U x, such
that, for all open U
U, ((U
)

=
p
(U
) as -modules.
Theorem 2.8. Given a o-manifold X there is a natural equivalence between
the category of o-vector bundles on X of dimension p and the category of locally
free sheaves of o-modules on X of nite rank p. So, given a vector bundle E on
X, we can dene uniquely the locally free sheaf of rank p, o(E)
X
on X where
o(E)
X
(U) := o(U, E).
Proof. For proof that there is a natural one-to-one correspondence, see [We,
40-41]. It is then easy to see that this correspondence induces an equivalence of
categories.
Definition 2.9. Let T be a sheaf over X. For, x X, we dene an equivalence
relation on the disjoint union

Ux
T(U) where U runs over all open neighbour-
hoods U X of x:
If U, V X are open neighbourhoods of x, we say that two elements s T(U)
and t T(V ) are equivalent if there exists and open neighbourhood W of x with
W U V and s[
W
= t[
W
.
The set of equivalence classes is called the stalk of T at x and is denoted by
T
x
. In other words, T
x
is the direct limit of the groups T(U) (x U) with respect
to the restriction homomorphisms
U
V
, x V U, i.e.
T
x
= lim
Ux
T(U) =
_
Ux
T(U)
_
_
.
If T is a sheaf of Abelian groups or commutative rings then T
x
will also inherit
that structure.
An element of the stalk T
x
of T at x X is called a germ.
Definition 2.10. A sheaf T over a paracompact Haussdor space is called
ne if given any locally nite open cover | = U
i
i
of X, there exists a partition
of unity on T subordinate to |. That is, there exists a family of sheaf morphisms
i
: T T
i
such that
(i) supp(
i
) U
i
for all i,
(ii)

i
i
= id
F
.
Example 2.11. If E is a dierentiable vector bundle over a dierential manifold
X and c(E)
X
is the sheaf associated to E via theorem 1.2.8, then c(E)
X
is ne.
Namely for any locally nite open cover | = U
i
of X, there exists a partition
of unity
i
on X subordinate to | where each
i
is a globally dened dierentiable
function and therefore multiplication by
i
of elements of c(E)
X
gives a sheaf
homomorphism which induces a partition of unity on c(E)
X
.
If K X is a closed subspace of X and T is a sheaf over X, we dene T(K)
as the direct limit of T(U) over all open U X such that K U. That is
T(K) := lim
UK
T(U).
Definition 2.12. A sheaf T over a space X is called soft if for any closed
subset K X, the natural restriction map
T(X) T(K)
is surjective. That is, any section over K of a soft sheaf T can be extended to a
global section of T.
Proposition 2.13. Fine sheaves are soft. In particular the sheaf c(E) associ-
ated to a vector bundle E X by theorem 1.2.8 is soft.
Example 2.14. Below are some commonly occurring examples of sheaves:
Constant sheaves
If T is a sheaf such that T(U) = G for some Abelian group G and for
every non-zero connected open set U X, then T is a constant sheaf.
Examples are the sheaves T = Z
X
, R
X
, C
X
given by T(U) = Z, R, C
respectively (so the restriction functions on T(U) are simply the identity
on T(U)). Constant sheaves on a manifold of dimension greater than zero
are not soft and therefore also not ne. See [We, 53].
Sheaves of functions and forms
We have seen in example 1.2.11 that, if E X is a vector bundle, the
sheaf c(E)
X
is ne. Similarly we can show that ((E)
X
is ne for a para-
compact dierential manifold X and c
p,q
(E)
X
is ne for a paracompact
complex manifold X.
The sheaf O(E)
X
of locally holomorphic sections of a complex bundle
E X is, in general, not soft and therefore also not ne. The same
applies to the sheaf O
(E)
X
of nowhere vanishing locally holomorphic
sections of E.
In particular, if E is the trivial bundle 1 := ((C X) X), the
sheaves (
X
(:= ((1)
X
), c
X
and c
p,q
X
are ne and, if X is a manifold of
dimension at least 1, O
X
, O
X
are neither soft nor ne.
2. SHEAVES 21
2.2. Cohomology groups. Most of the proofs of the results in this section
can be found in e.g. [Fo]. However, it is worthwhile to note that in some cases the
results and denitions are given here in a more general form than in [Fo].
Definition 2.15. Let T be a sheaf on a topological space X and let | = U
i
iI
be an open covering of X. For q = 0, 1, 2, . . . , a q- cochain is an element of the
q-th cochain group of T, C
q
(|, T), dened by
C
q
(|, T) :=
(i0,...,iq)I
q+1
T(U
i0
U
iq
)
(where I
q+1
is the direct product of q + 1 copies of I). The group operation on
C
q
(|, T) is componentwise addition.
Definition 2.16. For q = 0, 1, . . . , the coboundary operators
q
: C
q
(|, T) C
q+1
(|, T)
are dened by
q
(f)
i0,...,iq,iq+1
= g
i0,...,iq,iq+1
i0,...,iq,iq+1
where
g
i0,...,iq,iq+1
=
q+1
k=0
(1)
k
f
i0,...,
b
i
k
,...,iq,iq+1
on
k=0,...,q+1
U
i
k
.
(Here f
i0,...,
b
i
k
,...,iq,iq+1
:= f
i0,...,i
k1
,i
k+1
,...,iq,iq+1
.)
So, in particular
0
(f)
i,j
= f
j
f
i
on U
i
U
j
and
1
(g)
i,j,k
= g
j,k
g
i,k
+g
i,j
on U
i
U
j
U
k
.
Where there is no possibility of confusion,
q
will be referred to simply as .
It is easily checked that the coboundary operators are group homomorphisms.
Definition 2.17. Let
Z
q
(|, T) := Ker (
q
)
and
B
q
(|, T) := Im(
q1
).
The elements of Z
q
(|, T) are called q-cocycles and the elements of B
q
(|, T) are
called q-coboundaries.
Lemma 2.18. For q = 0, 1, . . . , B
q
Z
q
.
Proof. This follows immediately from the denitions. [We, 63].
Definition 2.19. For q = 0, 1, . . . the q-th cohomology group H
q
(|, T) of T
with respect to | is dened by
H
q
(|, T) :=
Z
q
(|, T)
_
B
q
(|, T)
.
Definition 2.20. Given two coverings | = U
i
iI
and 1 = V
k
kK
of X,
1 is called ner than |, written 1 < |, if, for every k K, there exists an i I
such that V
k
U
i
. In other words, there exists a rening map : K I such that
V
k
U
(k)
for all k K.
Given a sheaf T on X and covers 1 < | of X, the rening map enables us
to construct a homomorphism t
U
V
: Z
q
(|, T) H
q
(1, T) given by
t
U
V
: f
i0,...,iq
i0,...,iq
f
(k0),...,(kq)
i0,...,kq
.
We note that t
U
V
(B
q
(|, T)) B
q
(1, T) for all q, so t
U
V
denes a homomorphism
of cohomology groups
t
U
V
: H
q
(|, T) H
q
(1, T).
Lemma 2.21. The map t
U
V
: H
q
(|, T) H
q
(1, T) is independent of the choice
of rening map : K I.
Proof. [Fo, 98].
Lemma 2.22. t
U
V
: H
q
(|, T) H
q
(1, T) is injective.
Proof. [Fo, 99].
Given three open coverings J < 1 < |, the above implies that
t
V
W
t
U
V
= t
U
W
.
Therefore, we can dene an equivalence relation () on the disjoint union
H
q
(|, T), where | runs over all open coverings of X, by for H
q
(|, T)
and H
q
(1, T) if there is a covering J < |, J < 1 such that t
U
W
= t
V
W
.
Definition 2.23. The q-th cohomology group of X with coecients in T is
dened as the set of all the equivalence classes of H
q
(|, T) running over all open
coverings | of X. That is, H
q
(|, T) is the direct limit of the cohomology groups
H
q
(|, T) over all open coverings | of X.
H
q
(|, T) := lim
U
H
q
(|, T) =
_
H
q
(|, T)
_
_

.
Proposition 2.24. Let T be a sheaf over X. For any covering | = U
i
i
of
open subsets of X,
H
0
(X, T)

= H
0
(|, T)

= T(X).
Proof. [Fo, 103].
Proposition 2.25. If T ( are sheaves then there is a well-dened natu-
ral homomorphism : H
i
(X, T) H
i
(X, () , i 0 induced by the inclusions
Z
i
(|, T) Z
i
(|, () and C
i1
(|, T) C
i1
(|, () relative to an open cover
| = U
i
i
for X
Proof. Since T (, an element Z
i
(|, T) is in Z
i
(|, () and can therefore
be mapped onto the corresponding cohomology class in H
i
(X, ().
Now, let ,
be representatives of the same class in H

i
(X, T). Then there is
a cover | for X such that
=
i1
() for all i, j, k and some C
i1
(|, T).
But, C
i1
(|, T) C
i1
(|, () so, ,
are mapped onto the same element in

H
i
(X, ().
That this is a homomorphism follows directly from the denition and the alge-
braic structure on T and (.
2. SHEAVES 23
Remark 2.26. Occasionally we use the subscript ()
X
to distinguish a sheaf
F
X
over X with F
X
(U) = F(U) from F = F
X
(X), the global sections F
X
(X)
of F
X
. Examples are the constant sheaves, Z
X
, R
X
and C
X
and the sheaves of
functions (
X
, c
X
and O
X
. We will drop the subscript when referring to the associ-
ated cohomology groups H
k
(X, F) := H
k
(X, F
X
) (and similarly when referring to
C
K
(|, F), Z
k
(|, F) and B
k
(|, F)) since there is no possibility of confusion. For
example, we will write H
k
(X, R) rather than H
k
(X, R
X
).
Definition 2.27. A sequence
T

(

. . .
where , , . . . are sheaf morphisms, is called exact if, for every x X, the corre-
sponding sequence of stalks and restriction maps
T
x
|
Fx
(
x
|
Gx
. . .
is exact.
It is not necessarily the case that
T(U)

((U)

. . .
is exact for every U open in X. However, the following does hold:
Proposition 2.28. If
0 T

(

H
is an exact sequence of sheaves then,
0 T(U)

((U)

H(U)
is exact for every U open in X.
Proof. [Fo, 121].
Example 2.29. The Dolbeault sequence
Let X be a Riemann surface. As usual, c
0,1
X
is the sheaf of local dierentiable (0,1)-
forms on X. If the Dolbeault operator denotes the antiholomorphic component
of the exterior derivative, then the Dolbeault sequence
(2.1) 0 O
X
c
X

c
0,1
X
0
where denotes inclusion, is a short exact sequence of sheaves. This follows from
the Dolbeault lemma [Fo, 105], which says that every dierentiable function g on
X is locally of the form g =
f
z
for some dierentiable function f on X.
Theorem 2.30. If
0 T

(

H 0
is an exact sequence of sheaves over a paracompact Hausdor space X, then, for
q = 1, 2, . . . , there exists a connecting homomorphism
:=
q
: H
q1
(X, H) H
q
(X, T)
so that
(2.2) H
q1
(X, () H
q1
(X, H)

H
q
(X, T) H
q
(X, G) . . .
is an exact sequence.
Proof. [We, 56-58].
Theorem 2.31. If T is a soft sheaf over X, then, for q = 1, 2, . . . , the coho-

mology groups H
q
(X, T) vanish. In particular, if E X is a o-vector bundle with
o = (, c, H
q
(X, E) := H
q
(X, o(E)) vanishes for q 1.
Proof. [We, 56-57], [Hi, 34].
Definition 2.32. Let
(2.3) 0 T
h
T
0
h
0
T
1
h
1
T
2
h
2
. . .
h
p1
T
p
h
p
. . .
be an exact sequence of sheaves over a compact space X. If H
q
(X, T
p
)

= 0 for
p 0 and q 1, then 1.(2.3) is called a resolution of T. In particular, by theorem
1.2.31 , this is the case if T
p
is ne for all p 0. In this case, 1.(2.3) is called a
ne resolution of T.
Example 2.33. The sequence 1.(2.1) in example 1.2.29 is a ne resolution for
O
X
.
Theorem 2.34. Let
(2.4) 0 T
h
T
0
h
0
T
1
h
1
T
2
h
2
. . .
h
p1
T
p
h
p
. . .
be a resolution of a sheaf over a compact manifold X. This denes naturally a
sequence
(2.5) 0 T(X)
h
T
0
(X)
h
0
T
1
(X)
h
1
. . .
h
p1
T
p
(X)
h
p
. . . .
There are natural isomorphisms
H
q
(X, T)

=
Ker (h
q
)
_
Im(h
q1
)
, q 1
and
H
0
(X, T)

= Ker (h
0
).
Proof. By proposition 1.2.28
0 T(X)
h
T
0
(X)
h
0
T
1
(X)
is exact so T(X) = H
0
(X, T) = Ker (h
0
) as required.
Now let /
p
denote the kernel of h
p
: T
p
T
p+1
. Then, for all p,
(2.6) 0 /
p
T
p
/
p+1
0
is a short exact sequence of sheaves on X.
Then, for p 0, q 2,
(2.7) H
q1
(X, T
p
) H
q1
(X, /
p+1
) H
q
(X, /
p
) H
q
(X, T
p
) . . .
is exact by theorem 1.2.30, and since H
q
(X, T
p
) = 0 for q 1, p 0, it follows
that
(2.8) H
q1
(X, /
p+1
)

= H
q
(X, /
p
).
Letting p = q 1, we obtain
H
q
(X, T) = H
1
(X, /
q1
), q 1.
by repeated application of equation (2.8). For q = 1 and letting p = 0, T = Ker h
0
,
soH
1
(X, T) = H
1
(X, /
0
).
2. SHEAVES 25
Since
0 /
q1
T
q1
/
q
0
is a short exact sequence, the sequence
H
0
(X, T
q1
) H
0
(X, /
q
) H
1
(X, /
q1
) (H
1
(X, T
q
) = 0)
is exact for q 1 by theorem 1.2.30. Rewriting gives
T
q1
(X)
h
q1
/
q
(X)
h
q
|
K
q
H
q
(X, T) 0, q 1.
In other words, for q 1,
H
q
(X, T)

=
Ker (h
q
)
_
Im(h
q1
)
as required.
CHAPTER 2
The Analytic Index of an Elliptic Complex
Most of the material in this chapter can be found in [We].
1. Elliptic dierential operators
Let K denote the eld R or C.
Definition 1.1. A linear dierential operator P acting on dierentiable func-
tions dened on an open set U R
n
is an operator of the form
(1.1) P(x, D) =
||m
a
(x)D
where = (
1
, . . . ,
n
),
i
= 0, 1, . . . is a multi-index ([[ :=

i
i
) and D
1
1
. . .
n
n
,
j
=

xj
, j = 1 . . . n. For each , a
is a dierentiable function on U.
Similarly, a linear dierential operator P acting on dierentiable functions de-
ned on an open set U C
n
has the form
(1.2) P(z, D) =
||m
a
(x)D
where = (
1
,
1
, . . . ,
n
,
n
),
i
,
i
= 0, 1, . . . and D
=
1
1

1
1
. . .
n
n

n
n
,
j
=

zj
,
j
=

zj
j = 1 . . . n.
The order k of the operator P in equation 2.(1.2) is the greatest integer such
that there is an with [[ = k and a
, 0.
Let X be a compact dierentiable n-dimensional manifold without boundary
and E X and F X vector bundles of rank p and q respectively. As usual,
c(E) resp. c(F) denote the spaces of dierentiable sections of E resp. F.
Definition 1.2. A K-linear operator P : c(E) c(F) is a dierential op-
erator of order k if it has the following form: Let U X such that (e
j
)
p
j=1
, resp
(f
i
)
q
i=1
are frames for E, resp. F over U, and let =
_
p
j=1
j
e
j
_
c(U, E),
then P can locally be expressed in the form
P(x, D)() = P(x, D)
_
_
p
j=1
j
e
j
_
_
=
q
i=1
_
_
_
p
j=1
||k
a
i,j
(x)D
(
j
)(x)
_
_
_f
i
(x),
and there is at least one i such that

p
j=1
||=k
a
i,j
, 0 on U. Here (a
i,j
)
i,j
is a q p
matrix of smooth functions, so each component of the above is a dierential operator
in the sense of denition 2.1.1.
27
28 2. THE ANALYTIC INDEX OF AN ELLIPTIC COMPLEX
The space of dierential operators P : c(E) c(F) of order less than or
equal to k is denoted by Di
k
(E, F), and the space of all dierential operators
c(E) c(F) is denoted by Di (E, F) :=
k
Di
k
(E, F).
If T
X is the real cotangent bundle of X we dene T
X as the collection of
nonzero cotangent vectors on X:
T
X :=
_
xX
(T
x
X/0) T
X.
Let : T
X X be the canonical projection and
E,
F the pullbacks of E
and F over T
X. Then
E (resp.
F) can be regarded as the subset of E T
X
( resp. F T
X) consisting of the elements (e, ) with T
X and e E
(resp.
e F
).
Definition 1.3. For k Z, the k-symbol
k
(P) :
F of a dierential
operator P : E F is dened as follows:
For x X, let T
X
x
, and s c(E) be a dierentiable section of E with
s(x) = e. If f is a dierentiable function on an open neighbourhood of x with
df(x) = , then we dene
(1.3)
k
(P)(s(x), ) := P
_
i
k
k!
(f f(x))
k
s
_
(x) F
x
.
Proposition 1.4. The k-symbol
k
(P) :
F of P is a well-dened
homomorphism which is homogeneous of degree k, i.e. for > 0,
k
(P)(s(x), ) =
k
k
(P)(s(x), ).
Proof. The symbol
k
(P) is well-dened. That is, it is independent of the
choices of f c(X) and s c(E) = c(X, E):
Let f
be another dierential function on X with df
(x) = df(x) = and

s
another section of E with s
(x) = s(x) E
x
. Locally, P has the form P =
||k
A
where A
are q p matrices of locally smooth functions. So, in

applying P to
_
i
k
k!
(f
(x))
k
s
_
and evaluating at x, derivatives of order < k will
vanish (by the chain rule) since a factor of (f
(x)) (x)s(x) = 0 will remain. For

derivatives of order k we notice that the kth derivative of (f
(x))
k
s depends
only on df
(x) = df(x) = and s
(x) = s(x). So, the k-symbol is independent of

the choices of f and s.
That
k
(P) it is linear (in P) is immediate. If t c(E) denotes another section
of E,
k
(P)((s +t)(x), ) = P
_
i
k
k!
(f f(x))
k
(s +t)
_
(x)
=
k
(P)(s(x), ) +
k
(P)(t(x), ), , C
by the linearity of P, so
k
(P)(s(x), ) is a homomorphism from E
x
to F
x
.
Finally, for > 0
k
(P)(s(x), ) = P
_
i
k
k!
(f f(x))
k
s
_
(x)
= P
_
i
k
k!
k
(f f(x))
k
s
_
(x) =
k
k
(P)(s(x), )
1. ELLIPTIC DIFFERENTIAL OPERATORS 29
by the linearity of P. In other words
k
(P) is homogeneous of degree k.
Clearly
k
(P) 0 if and only if the degree of P is less than k. In fact, for
a dierential operator P of degree k, we will only be interested in the k-symbol
k
(P) and so, henceforth, unless otherwise stated, we shall refer to this simply as
the symbol of P and write (P) :=
k
(P).
Definition 1.5. The operator P is said to be elliptic if for all x X and for
all T
X
x
, (P)() is an isomorphism from E
x
to F
x
.
Of course, if an elliptic operator P : E F exists then rk E = rk F.
Example 1.6. The Laplacian.
As an example, we will show that the Laplace-Beltrami operator for Riemannian
manifolds is elliptic.
(X, g) is a n-dimensional Riemannian manifold with Levi-Civita connection
(see example 1.1.33). c(TX) is the space of dierentiable vector elds on X. For
a smooth real valued function f C
(X, R), the operator

Hf : c(TX) c(TX) C
(X, R) , Hf(V
1
, V
2
) := L
V1
(L
V2
(f)) L
(
V
1
V2)
(f)
(where L
V1
f denotes the Lie derivative of f with respect to V
1
) is the called the
Hessian of f.
The Laplace-Beltrami operator, or Laplacian on X is the trace of this oper-
ator. That is
: C
(X, R) C
(X, R), (f)(x) :=

n
j=1
Hf(v
j
, v
j
)
with (v
1
, . . . , v
n
) a g- orthonormal basis of T
x
X. Clearly the degree of is 2. So,
we wish to show that for any T
X, x = () X and function s C
(X, R),
()(s(x), ) : R R is an isomorphism.
Let f C
(X, R) such that df(x) = and f(x) = 0 and choose an orthonormal

frame v
1
, . . . v
n
of T
x
X such that (v
j
) = ||
1,j
, j = 1, . . . , n. Locally, we can
extend this to an orthonormal frame of vector elds V
1
, , V
n
, with V
1
(x) = v
1
,
on some neighbourhood U of x.
Let s C
(X, R). Then,

()(s(x), ) =
_
1
2
(f
2
s)
_
(x) =
1
2
n
j=1
_
L
Vj
L
Vj
(f
2
s) L
V
j
Vj
(f
2
s)
_
(x).
Using the Leibniz property of the Lie derivative we have, for all j = 1, . . . n,
L
Vj
(f
2
s) = f
2
L
Vj
(s) + 2fsL
Vj
(f)
so
L
Vj
L
Vj
(f
2
s) = L
Vj
_
f
2
L
Vj
(s) + 2fsL
Vj
(f)
_
= 4fL
Vj
(f)L
Vj
(s) +f
2
L
Vj
L
Vj
s + 2s(L
Vj
(f))
2
+ 2fsL
Vj
L
Vj
(f).
By construction, L
Vj
(f) = ||
1,j
at x. So
n
j=1
L
Vj
L
Vj
(f
2
s) = 4fL
V1
fL
V1
s + 2fsL
V1
L
V1
(f) + 2s(L
V1
(f))
2
+
n
j=1
f
2
L
Vj
L
Vj
s.
Using the fact that f(x) = 0, we get
n
j=1
L
Vj
L
Vj
(f
2
s)(x) = 2s(L
V1
(f))
2
= 2s(x) |
x
|
2
.
Again, by the Leibniz rule and the fact that f(x) = 0 we have
L
V
j
Vj
(f
2
s)(x) = f
2
(x)L
V
j
Vj
(s)(x) + 2f(x)s(x)L
V
j
Vj
(f)(x) = 0.
So,
()(s(x), ) =
1
2
n
j=1
_
L
Vj
L
Vj
(f
2
s) L
V
j
Vj
(f
2
s)
_
(x) = |
x
|
2
s(x) = 0
if and only if s(x) = 0.
In other words, the endomorphism ()(, ), T
x
X is injective and hence
an isomorphism and we have proved that : C
(X, R) C
(X, R) is elliptic.
2. Elliptic complexes
Definition 2.1. Given a nite number of dierentiable vector bundles (E
i
)
l
i=1
on X and dierential operators d
i
: c(E
i
) c(E
i+1
), E = (E
i
, d
i
)
l
i=1
is called a
complex if d
i+1
d
i
0.
A complex is said to be elliptic if for any T
x
X, the sequence
E
i,x
(di)()
E
i+1,x
. . .
is exact. (In particular, an elliptic operator is an elliptic complex of the form
0 c(E)
P
c(F) 0.)
Example 2.2. The de Rham complex on a complex manifold
For a dierentiable manifold X of real dimension m, the de Rham complex is
given by
0 c
0
(X)
d
c
1
(X)
d
. . .
d
c
m
(X)
where d denotes exterior dierentiation. We let T
X denote the real cotangent

bundle and T
X
C
= T
X
R
C its complexication. Using the notation above we
write the de Rham complex as
0 c(E
0
)
d
c(E
1
)
d
. . .
d
c(E
m
) 0
where E
k
:=
k
T
X
C
.
d is a dierential operator of degree 1 so we calculate, for each k = 1, . . . , nm,
the associated 1-symbol homomorphisms
(2.1) E
0,x
(d)(s0(x),)
E
1,x
(d)(s1(x),)
. . .
(d)(sm1(x),)
E
m,x
.
Given T
x
X T
X, we choose an f c(X) such that df(x) = and f(x) = 0.

Let s
i
c(E
i
) such that s
i
(x) = e E
i,x
. Then
(d)(s(x), ) = id(fs
i
)(x) = idf(x) e +if(x)ds
i
(x) = i e
by the Leibniz property for exterior dierentiation.
So, the sequence 2.2.1 is exact. Namely e = 0 if and only if e and are
linearly dependent, i.e. e = for some E
i1,x
.
2. ELLIPTIC COMPLEXES 31
Example 2.3. The Dolbeault complex on a complex manifold
We generalise example 1.2.29 to higher dimensional complex manifolds and show
that the complex thus obtained is elliptic.
For a complex n-dimensional dierentiable manifold X the de Dolbeault com-
plex (X) is given by
c
p,0
(X)

c
p,1
(X)

. . .

c
p,n
(X) 0
where the Dolbeault operator : c
p,q
(X) c
p,q+1
(X) is the antiholomorphic
component of the exterior derivative.
Let T denote the complex cotangent bundle and T its conjugate. Then, we
may write the Dolbeault complex as
c(E
0
)

c(E
1
)

. . .

c(E
n
) 0
where E
k
:=
p
T
k
T.
is a dierential operator of degree 1 so we calculate, for each k = 1, . . . , n,
the associated 1-symbol homomorphisms
(2.2) E
0,x
()(s0(x),)
E
1,x
()(s1(x),)
. . .
()(sn1(x),)
E
n,x
.
Given T
X
x
, we choose an f C
(X) such that df(x) = =

1,0
+
0,1
(with
1,0
T and
0,1
T) and f(x) = 0. Let s
i
c(E
i
) such that s
i
(x) = e E
i,x
.
Then
()(s
i
(x), ) = i(fs
i
)(x) = i(f(x) e +f(x)s(x)) = i
0,1
e
by the Leibniz property.
So, the symbol sequence is exact. Namely
0,1
e = 0 if and only if e and
x
are linearly dependent, i.e. e =
0,1
for some E
i1,x
.
Now let E X be a holomorphic bundle over a paracompact manifold X.
Choose a holomorphic frame f = (f
i
)
rk E
i=1
for E on a neighbourhood U X which
is small enough that it is contained in a chart neighbourhood for X. Recall that
c
p,q
(E) = c(E
p
T
q
T), so for an element c
p,q
(E), [
U
=

i
f
i

i
where each
i
is a (p, q)-form on U.
Then dene
E
([
U
) :=
i
(f
i

i
). =
i
f
i
(
i
)
Since f
i
is holomorphic,
E
([
U
) =
i
f
i
(
i
) .
Proposition 2.4. The operator
E
of a complex vector bundle E X is a
well-dened global operator. That is, it is independent of the choice of frame f.
Proof. Let f
= (f
j
)
j
be another holomorphic frame for E on U. Then, there
is a change of frame matrix (t
i,j
)
i,j
i, j = 1, . . . , rk E such that each of the maps
t
i,j
is holomorphic and f
i
=
j
t
i,j
f
j
.
Now
[
U
=
j
f
j

_
i
t
i,j
i
_
,
so, in terms of f
,
([
U
) =
j
f
j

_
i
t
i,j
i
_
=
j
f
j

_
i
t
i,j
(
i
)
_
i,j
f
j
f
i,j
(
i
) =
i
f
i
(
i
).
In other words, () c
p,q+1
(E) is globally dened. So
E
: c
p,q
(E) c
p,q+1
(E)
is a well-dened map.
Definition 2.5. Given a holomorphic vector bundle E X, the operator

E
is called the Dolbeault operator of E.
Example 2.6. The Dolbeault complex for a holomorphic bundle
We wish to show that the Dolbeault complex (E) of E given by
(2.3) c
p,0
(E)

E
c
p,1
(E)

E
. . .

E
c
p,q
(E) 0
is elliptic. In fact the argument of example 2.2.3 carries over to this more general
case.
E
is a dierential operator of degree 1 so we calculate, for each k = 1, . . . , n,
the associated 1-symbol homomorphisms:
Let s
i
c(
p
T
i
T) such that s(x) = e E
p
T
i
T and T
X
x
.
We choose an g C
(X) such that dg(x) = =

1,0
+
0,1
(with
1,0
T and
0,1
T) and g(x) = 0.
Then
(
E
)(s
i
(x), ) = i
E
(gs
i
)(x) = i(g(x) e +g(x)
E
s(x)) = i
0,1
e
by the Leibniz property.
As before
0,1
e = 0 if and only if e and are linearly dependent, i.e. e =
0,1
for some E
p
T
i1
T.
Remark 2.7. If E X is a holomorphic vector bundle, we consider the kernel
of
E
: (c(E) = c
0,0
(E)) c
0,1
(E):
As above, we choose a holomorphic frame f = (f
i
)
rk E
i=1
for E on a neighbourhood
U X which is small enough that it is contained in a chart neighbourhood for X.
Then, given c(E), =

rk E
i=1

i
f
i
with
i
c(U) on U. In this case is
holomorphic if and only if
i
is holomorphic for all i and, by the construction of
E
above, this is precisely when
E
() = (
i
f
i
) =
i
(
i
) f
i
= 0
(where the rst equality follows from the denition of
E
in example 2.2.3).
Definition 2.8. The cohomology groups H
i
(E) of a complex E = (E
i
, d
i
)
i
are dened by
H
i
(E) =
Ker d
i
_
Imd
i1
.
Proposition 2.9. The cohomology groups H
i
(E) of an elliptic complex E =
(E
i
, d
i
)
i
over a compact manifold X are nite dimensional for all i.
Proof. This is an analytic proof which rests on some deep results in functional
analysis. These are described in [Ho]. Given the ndings of Hormander, there is
an elegant proof in [AB, 395-398]. [We, 119-153] also gives a good overview.
2. ELLIPTIC COMPLEXES 33
Definition 2.10. The analytic index of an elliptic complex E of length l is
dened to be ind
a
(E) :=
l
i=0
(1)
i
dimH
i
(E).
In particular, the analytic index of an elliptic operator P : E F where
E X and F X are vector bundles over X, is given by
ind
a
(P) = dimKer (P) dimCoker (P).
CHAPTER 3
The Riemann-Roch Theorem
1. Divisors
1.1. Denitions.
Definition 1.1. A divisor on a Riemann surface X is an integer valued func-
tion D : X Z with discrete support. If D(x) = n
x
, we write formally
D =
xX
n
x
x.
A divisor is called eective if n
x
0 for every x in X.
We can dene a partial ordering on the set of divisors on X by D D
if
and only if D(x) D
(x) for all x X. (D < D
if D D
and there is at least

one x X such that D(x) < D
(x).)
The set of divisors Div (X) on a Riemann surface X denes in a natural way an
Abelian group with operation + such that, for x X, (D+D
)(x) := D(x)+D
(x).
Let X now be a Riemann surface and let /(X) denote the eld of meromorphic
functions on X. If /(X) is not identically zero, then, at every x X, if z is
a local coordiinate for X centered at x, then has the unique form (z) = z
k
g(z)
with k Z and g(z) holomorphic and non-zero at z = 0. k is the order of at x,
ord
x
() and ord
z
() ,= 0 only on a discrete set in X. So, denes a divisor div ()
by
div () :=
xX
ord
x
() x.
In particular, a non-zero holomorphic function on X denes an eective divisor.
Proposition 1.2. Let X be a compact Riemann surface. Given points x
i
N
i=1

X and complex numbers c
i
N
i=1
, there exists a function /(X) such that
(x
i
) = c
i
, i = 1, . . . , N.
Proof. [Fo, 116].
Definition 1.3. A divisor of the form div () for /(X)
(where /(X)
is the group of non-zero elements of /(X)) is called a principal divisor.

Definition 1.4. We say that two divisors D and D
on X are linearly equiv-

alent if there is a meromorphic function on X such that D
= D + div ().
Definition 1.5. A meromorphic dierential on a Riemann surface X is a
holomorphic 1-form on X S where S X is discrete, with the following local
description: Since S is discrete, for all s S, there exists an open neighbourhood
U
s
of s with U
s
S = s and U
s
is contained in a chart neighbourhood of X with
local coordinate z. Then
[
Us
= dz
35
36 3. THE RIEMANN-ROCH THEOREM
where is meromorphic on U
s
.
We denote the set of meromorphic dierentials on X by /
1
(X).
Remark 1.6. It follows from denition 3.1.5 that the meromorphic dieren-
tials on X correspond to the meromorphic sections (see denition 1.1.8) of the
holomorphic cotangent bundle T of X.
Lemma 1.7. If is a non-zero meromorphic dierential on a Riemann surface
X, then every meromorphic dierential on X is of the form with /(X).
Proof. Let (U
i
, z
i
)
i
be a holomorphic atlas for X. On U
i
U
j
, g
i,j
:=
dzj
dzi
is a transition function for the cotangent bundle T. So
[
Uj
=
j
dz
j
= g
j,i
i
dz
i
, and [
Uj
=
j
dz
j
= g
j,i
i
dz
i
on U
i
U
j
.
But
j
dz
j
dz
j
=

j
j
and
j
dz
j
dz
j
=
g
j,i
i
g
j,i
i
dz
i
dz
i
=

i
i
for all i, j. So, , given locally by [
Ui
=
i
i
, is a globally dened meromorphic
function such that = .
Remark 1.8. Proposition 3.1.2 asserts the existence of a non-constant mero-

morphic function on X. Therefore, the set of meromorphic dierentials on a
Riemann Surface X is non-empty. Namely, given locally by :=

z
dz is a
meromorphic dierential on X.
Definition 1.9. Given a mermorphic dierential dz on an open V C, we
dene its order at x V as the order of at x.
If X is a Riemann surface, x X and z a local coordinate at x, the order
ord
x
() of a meromorphic dierential with = dz in a neighbourhood of x is
ord
x
() := ord
x
(). So, we can dene
div () =
xX
ord
x
() x.
A divisor of the form D = div () with a meromorphic dierential on X is a
called a canonical divisor on X.
Definition 1.10. On a compact Riemann Surface X a divisor D has compact
support and therefore the map deg : Div (X) Z dened by deg(D) :=
xX
n
x
is
well dened and clearly a homomorphism. This is called the degree homomorphism.
Proposition 1.11. Every principal divisor has degree 0 and so linearly equiv-
alent divisors have the same degree.
Proof. [Fo, 80-81]. This is a consequence of the residue theorem on Riemann
surfaces.
An immediate consequence of lemma 3.1.7 and proposition 3.1.11 is
Corollary 1.12. All canonical divisors on a Riemann surface X have the
same degree.
1. DIVISORS 37
1.2. The line bundle L
D
. Let D be a divisor on a compact Riemann Surface
X. We associate to D a (holomorphic) line bundle L
D
over X.
Since X is compact, supp(D) = p
j
X[n
pj
,= 0 is nite. We choose a nite
covering U
k
k
of X such that for each U
k
, there is a meromorphic function g
k
on
U
k
with ord
g
k
(p) = n
p
for every p U
k
. For example, we could choose U
k
so that
at most one point p
k
supp(D) is contained in U
k
. (See e.g. [Jo, 228-229] for why
it is possible to construct such functions g
k
k
.)
Then, the zero and polar sets of g
i
are well dened since, by construction,
g
i
and g
j
have the same zero and polar sets on U
i
U
j
. In particular,
gi
gj
is a
nowhere zero holomorphic function on U
i
U
j
and so the covering U
i
i
together
with the transition functions g
i,j
i,j
:=
gi
gj
[U
i
U
j
i,j
specify a one dimensional
holomorphic vector bundle (a complex line bundle) L
D
on X. That is, we can take
each open set U
i
as a trivialising neighbourhood for L
D
so that L
D
[
Ui
U
i
C.
Fibres over points x U
i
U
j
are then identied by the function g
i,j
=
gi
gj
. So L
D
is a holomorphic line bundle over X.
Conversely, we shall see that all holomorphic line bundles on a compact Rie-
mann surface X can be associated to a divisor on X.
Definition 1.13. If s /(L) is a non-zero meromorphic section of a holo-
morphic line bundle L represented by functions s
i
/(U
i
) relative to an open
cover U
i
i
of X, then the order ord
x
(s) of s at x U
i
is given by
ord
x
(s) := ord
x
(s
i
).
This is clearly independent of the choice of cover U
i
i
.
Lemma 1.14. Every holomorphic line bundle L X on a compact Riemann
surface X admits a global meromorphic section s such that s , 0 on X. In partic-
ular, since s has only isolated zeroes and poles, we can associate a divisor (s) on
X to s by (s)(x) = ord
x
(s) for x X.
Proof. [Fo, 225].
Lemma 1.15. Let L be a holomorphic line bundle on a compact Riemann surface
X. If s is a global meromorphic section of L and (s) the divisor associated to s,
then (s) denes a holomorphic line bundle L
(s)
on X.
In this case L
(s)
is isomorphic to L.
Proof. Let L have transition functions g
i,j
i,j
and s = s
i
i
be a global
meromorphic section of L. Then, for all, i, j,
si
sj
is non-vanishing and holomorphic
on U
i
U
j
so we can dene L
(s)
by means of the meromorphic functions s
i
on U
i
.
Since s = s
i
i
is a section, s
i
= g
i,j
s
j
so L
(s)
has transition functions s
i,j
:=
si
sj
=
g
i,j
.
Definition 1.16. The degree deg(L) of a holomorphic line bundle L is dened
as deg(L) := deg(s) where s is a global meromorphic section of L. By the above
lemma 3.1.15, this is well dened.
The following theorem says that there is a one to one correspondence between
the linear equivalence classes of divisors on X and the isomorphism classes of line
bundles on X. In particular, the line bundle L
D
associated to a divisor D is well-
dened up to isomorphism.
Theorem 1.17. If L = L
D
and L
= L
D
, then L and L
are isomorphic if and

only if D and D
are linearly equivalent.

Proof. Let D
= D + div (), /(X). Then, according to the above

method, if L is associated to the data U
i
, g
i
i
, D
can be associated to g
i
and
so we obtain transition functions g
i,j
=
gi
gj
= g
i,j
for L
. By proposition 1.1.7,
L

= L
.
For the converse, let L have transition functions g
i,j
i,j
=
gi
gj
i,j
and L
has
transition functions g
i,j
i,j
=
g
i
g
i,j
with respect to a cover | = U
i
i
, and with
g
i
, g
i
/(U
i
) for all i. The divisors corresponding to g
i
i
and g
i
are denoted
by D and D
respectively.
Say f : L L
is an isomorphism of holomorphic line bundles. We wish to

show that D
= D + div () for some /(X). That is, that

g
i
= g
i
, for all i.
For all i, let s
i
, resp. s
i
denote the holomorphic section of L[
Ui
resp. L
[
Ui
which is mapped to the constant function 1 C by a trivialising map. Then, in
particular, s
i
, resp. s
i
is nowhere vanishing on U
i
, so, since f is linear on bres
f(s
i
) = f
i
s
i
where f
i
is a nowhere vanishing holomorphic function on U
i
.
So, on U
i
U
j
we have
f(s
j
) = f
j
s
j
= f
j
g
j
g
i
s
i
and also
f(s
j
) = f
_
g
j
g
i
s
i
_
=
g
j
g
i
f(s
i
) =
g
j
g
i
f
i
s
i
.
So, since s
i
,= 0 on U
i
U
j
g
j
g
i
f
j
=
g
j
g
i
f
i
on U
i
U
j
and therefore
i
:=
g
i
g
i
f
i
denes a meromorphic function on X. Furthermore, on U
i
, the divisor associated
to is the divisor associated to
g
i
gi
(since f
i
is non-vanishing and holomorphic on
U
i
). I.e. div () = D
D. This is what we wished to prove.
Example 1.18. By proposition 3.1.11, linearly equivalent divisors on a compact

Riemann surface have the same degree. On CP
1
the converse also holds. That is,
if divisors D and D
on CP
1
have the same degree, they are linearly equivalent:
Let D be a divisor on CP
1
such that deg(D) = 0. We cover CP
1
with the open
sets U
0
and U
1
as in example 1.1.5. On U
0
, we have the local coordinate [1 : z] z
and on U
1
we have the local coordinate [w : 1] w. By translating if necessary, we
may assume that D is supported on (U
0
U
1
) U
0
(since the complement of U
0
U
1
in CP
1
is a discrete set of two points) and we may therefore write D =
zC
n
z
z.
1. DIVISORS 39
If supp(D) = z
i
i
, then

i
n
zi
= 0. We wish to associate a meromorphic
function on CP
1
to the divisor D. On U
0
,dene
0
(z) :=
i
(z z
i
)
nz
i
.
Clearly div (
0
) = D[
U0
. Now, on U
0
U
1
, w =
1
z
, w
i
:=
1
zi
, and we have
0
(w) =
i
_
1
w
z
i
_
nz
i
=
_
1
w
_
P
i
nz
i
i
(1 wz
i
)
nz
i
=
i
_
1
w
w
i
_
nz
i
=
i
_
1
w
w
i
_
nz
i
,
=
i
_
1
w
i
_
nz
i
(w
i
w)
nz
i
.
Since

i
_
1
wi
_
nz
i
(w
i
0)
nz
i
is nite, we can extend
0
to all of U
1
. Furthermore,
is non-zero at w = 0 so, we have dened a global meromorphic function on CP
1
with div () = D. We have shown that if D is a divisor on CP
1
such that deg D = 0,
then D = div () for some meromorphic function on CP
1
, and therefore that two
divisors on CP
1
have the same degree if and only if they are linearly equivalent.
This, together with theorem 3.1.17, implies that two line bundles on CP
1
have
the same degree if and only if they are isomorphic.
Let : O
CP
1(1) CP
1
be as in examples 1.1.5 and 1.1.17. We dene a global
meromorphic section s on O
CP
1(1) via the covering CP
1
= U
0
U
1
. Dene
s
0
: [1 : z] (1, z) on U
0
, and s
1
: [w : 1] (
1
w
,
1
w
) on U
1
.
Then s = s
0
, s
1
transforms according to s
1
= g
1,0
s
0
on U
0
U
1
and s is a nowhere
vanishing global meromorphic section of O
CP
1(1) with a single pole of order 1 in
the point given by [0 : 1].
Therefore we have deg(O
CP
1(1)) = 1 and, if D is a divisor on CP
1
with
deg(D) = 1, there is an isomorphism L
D

O
CP
1(1).
Proposition 1.19. Let K be a canonical divisor on a compact Riemann Sur-
face X. Then L
K
is isomorphic to the holomorphic cotangent bundle T of X. We
call L
K

= T the canonical line bundle on X.
Proof. A canonical divisor is the divisor of a meromorphic dierential on X.
The statement follows from denition 3.1.5 (see remark 3.1.6).
Definition 1.20. By theorem 1.2.8, we associate to L
D
a sheaf O
D
:= O(D)
of local holomorphic sections of L
D
.
Proposition 1.21. There is a natural isomorphism between O
D
and the sheaf
T where T(U) = /(U) : D[
U
+ div () 0 and the restriction functions
U
V
, V U are simply the restrictions
U
V
() = [
V
, /(U).
Proof. D is the divisor of a meromorphic section s
0
of L = L
D
. T(U) if
and only if div ( s
0
) = div () +div (s
0
) 0. That is div ( s
0
) is eective on U
and therefore s
0
O
D
(U).
Conversely, given a local holomorphic section s O
D
(U),
div (s) div (s
0
) D[
U
so the formal expression
s
s0
denes an element of T(U).
2. The Riemann-Roch Theorem and the analytic index of a divisor

Definition 2.1. The genus g of a compact Riemann Surface X is dened by
g := dimH
1
(X, O).
Theorem 2.2. The Riemann-Roch theorem
If D is a divisor on a compact Riemann surface X and O
D
is the sheaf of local
holomorphic sections of L
D
(or, by proposition 3.1.21, the sheaf of local meromor-
phic functions on U open in X such that div () + D[
U
0), then H
0
(X, O
D
)
and H
1
(X, O
D
) are nite dimensional vector spaces with dimensions h
0
(X, O
D
)
and h
1
(X, O
D
) respectively and
(2.1) h
0
(X, O
D
) h
1
(X, O
D
) = 1 g + deg D.
It is not the aim of the current thesis to prove this theorem (refer to e.g. [Fo,
129-130]). In this chapter we are interested in showing that the integer quantity
given on the left hand side of the equation can be interpreted as the analytic index
of a dierential operator associated to the divisor D.
2.1. Serre duality. The Serre duality theorem provides a powerful tool for
calculating with the Riemann-Roch theorem.
Theorem 2.3. Serre duality theorem
There is an isomorphism
: H
0
(X, O
KD
)
_
H
1
(X, O
D
)
_
.
Proof. [Fo, 132-138].
Corollary 2.4. If K is a canonical divisor on a compact Riemann surface
X, then
deg K = 2g 2.
Proof. By Serre duality, theorem 3.2.3,
H
1
(X, O
K
) = (H
0
(X, O
KK
))
= (H
0
(X, O))
.
H
0
(X, O) consists of the holomorphic functions on X but since X is compact,
H
0
(X, O) = C so h
1
(X, O
K
) = h
0
(X, O) = 1.
Furthermore, H
0
(X, O
K
) = (H
1
(X, O
KK
))
= (H
1
(X, O))
but
h
1
(X, O) =: g. So the Riemann-Roch theorem gives
h
0
(X, O
K
) h
1
(X, O
K
) = g 1 = 1 g + deg K.
That is deg K = 2g 2.
2. THE RIEMANN-ROCH THEOREM AND THE ANALYTIC INDEX OF A DIVISOR 41

Example 2.5. The projective line CP
1
has genus 0. Therefore a canonical
divisor K on CP
1
has degree -2. In particular, by proposition 3.1.19, the degree
deg(T) of the holomorphic cotangent bundle T of CP
1
is -2.
2.2. The analytic index of a divisor.
Definition 2.6. If D is a divisor on a compact Riemann surface X then the
Dolbeault operator
D
of D is dened as
D
:=
L
D
: c(L
D
) c
0,1
(L
D
) (see
denition 2.2.5).
Remark 2.7. Where the bundle is clear from the context, we shall simply refer
to the operators
L
,
D
as .
We will use the notation
(2.2) c
D
:= c(L
D
) and c
0,1
D
:= c
0,1
(L
D
).
In example 2.2.6, we have seen that the Dolbeault operator c
p,q
(E)

E
c
p,q+1
(E)
for a holomorphic vector bundle E X is elliptic with symbol
(
E
)(s
q
(x), ) = i
E
(gs
q
)(x) = i(g(x) e +g(x)
E
s
q
(x)) = i
0,1
e
for s
q
c
p,q
(E) such that s
q
(x) = e E
p
T
q
T, T
X and g C
(X)
such that dg(x) = =
1,0
+
0,1
(with
1,0
T and
0,1
T) and g(x) = 0.
In particular if D is a divisor on a compact Riemann surface X, the operator
D
is elliptic. Furthermore, by remark 2.2.7, Ker (c
D
D
c
0,1
D
) = O
D
.
Lemma 2.8. The sequence
(2.3) 0 (O
D
)
X
(c
D
)
X
D
(c
0,1
D
)
X

0
(where the arrow denotes inclusion) is an exact sequence of sheaves. Here we
denote the sheaf morphism induced by
D
also by
D
.
Proof. Clearly the map O
D
(U) c
D
(U) is injective for all U X open,
and we have just seen that, by remark 2.2.7, Ker (c
D
D
c
0,1
D
) = O
D
. So, by the
denition of the Dolbeault operator
D
, Ker (c
D
(U)

D
c
0,1
D
(U)) = O
D
(U) for
all U X open.
It therefore remains to show that (c
D
)
X
D
(c
0,1
D
)
X
is surjective: Let x X
and E
0,1
D
(V ) where V x is an open neighbourhood. On a suciently small
neighbourhood U V of x, = s where s c
D
(U) and c
0,1
(U). The
Dolbeault lemma ([Fo, 105]) says that, in a small enough neighbourhood U
U
X, = f where f c. Therefore, by the denitions 3.2.6, and 2.2.5 of
D
=
L
D
,
s = (f s) =:
D
() for some (c
D
)(U).
Lemma 3.2.8 can be generalised in an obvious fashion to holomorphic vector

bundles of rank r over compact complex manifolds of any dimension.
Theorem 2.9. If D is a divisor on a compact Riemann surface X and
D
: c
D
c
0,1
D
is the Dolbeault operator on L
D
, then
Ker (c
D

c
0,1
D
)

= O
D
= H
0
(X, O
D
)
and
Coker (c
D

c
0,1
D
)

= H
1
(X, O
D
).
Proof. The rst statement has been proved in lemma 3.2.8 above. Again by
lemma 3.2.8,
0 (O
D
)
X
(c
D
)
X

(c
0,1
D
)
X

0
is a short exact sequence of sheaves. So, by theorem 1.2.30
0 O
D
c
D

c
0,1
D
H
1
(X, O
D
) H
1
(X, c
D
)
is exact
Furthermore, by theorem 1.2.31, H
1
(X, c
D
) = H
1
(X, c
0,1
D
) = 0. So, it follows
directly from theorem 1.2.34 that
H
1
(X, O
D
)

= Coker (c
D

c
0,1
D
).
Definition 2.10. The analytic index of a divisor D on a compact Riemann

surface X is given as
ind
a
(D) := ind
a
((L
D
)),
the analytic index of the elliptic operator
D
: c
D
c
0,1
D
.
Corollary 2.11. By theorem 3.2.9,
ind
a
(D) = h
0
(X, O
D
) h
1
(X, O
D
),
the left hand side of equation 3.(2.1).
3. The Euler characteristic and Hirzebruch-Riemann-Roch
Definition 3.1. Let E X be a holomorphic vector bundle of rank r over a
compact complex manifold X of dimension n. The Euler characteristic (E) of E
is given by
(E) :=
i0
(1)
i
h
i
(X, E),
where h
i
(X, E) := dimH
i
(X, E) and H
i
(X, E) := H
i
(X, O(E)) as in theorem
1.2.8.
Lemma 3.2. H
i
(X, E) = 0 for i n.
Proof. Let E X be a holomorphic vector bundle with X is a compact
complex manifold of complex dimension n. It follows from remark 2.2.7 that the
sequence of sheaves
0 O(E)
X
c(E)
X
E
c
0,1
(E)
X
E
. . . ,
is a resolution for O(E)
X
. In fact, since the sheaves c
0,k
(E)
X
are ne for all k (by
example 1.2.14), the above sequence is a ne resolution for O(E)
X
. So the sequence
of global sections
0 O(E) c(E)

E
c
0,1
(E)

E
. . .
is a complex. Therefore, by theorem 1.2.34
H
i
(X, E) := H
i
(X, O(E))

=
Ker
_
E
(E
i
T)
_
_
Im
_
E
(E
i1
T)
_
= 0
for i > n since
i
T = 0 in that case.
3. THE EULER CHARACTERISTIC AND HIRZEBRUCH-RIEMANN-ROCH 43

Corollary 3.3. The analytic index of a divisor D on a Riemann surface X
is equal to the Euler characteristic (L
D
) of the bundle L
D
over X.
Proof. This is a restatement of corollary 3.2.11.
Remark 3.4. As a special case of this, we have already seen in 3.2.4 that the
analytic index of a canonical divisor K is equal to
1 g + deg K = 1 g + (2g 2) = g 1,
the Euler number of X.
In fact as a direct corollary of theorem 1.2.34, we have the following general
result:
Theorem 3.5. Let E be a holomorphic vector bundle over a compact complex
manifold X . Then
(E) = ind
a
((E)).
CHAPTER 4
The Topological Index of a Divisor
We now turn to the right hand side of the Riemann-Roch equation:
(0.1) T(D) := 1 g + deg D.
The aim of this chapter will be to show that, if L is a holomorphic line bundle on a
compact Riemann surface X, and D is the divisor of a meromorphic section of L,
T(D) is dependent only on the topological, and not the analytic, structure of L.
1. De Rham Cohomology
Let X be a real n-dimensional dierentiable manifold. As before (page 32), for
p 0, c
p
= c
p
(K) denotes the space of dierential p-forms on X with coecients
in the eld K = R, C. In particular, c
0
is the space of dierentiable K-valued
functions on X.
Recall (equation 1.(1.1)) that if K = C and X is complex, c
p
:=
q+r=p
c
q,r
where c
q,r
:= c(
q
T
r
T).
The exterior derivative d
p
: c
p
c
p+1
is a K-homomorphism.
For p 1, we dene
Z
p
:= Ker (c
p dp
c
p+1
)
and
B
p
:= Im(c
p1 dp1
c
p
).
Then Z
p
is the space of closed p-forms on X, and B
p
the space of exact p-forms on
X.
Definition 1.1. For K = R, C, the p-th de Rham cohomology group
H
p
deRh
(X) := H
p
deRh
(X, K) of X is dened by
H
p
deRh
(X) (= H
p
deRh
(X, K)) :=
Z
p_
B
p.
Theorem 1.2. de Rhams theorem.
For X, a real paracompact n-dimensional dierentiable manifold with
K
X
= R
X
, C
X
the constant real or complex sheaf on X, there is a natural isomor-
phism
H
p
(X, K)

H
p
deRh
(X, K), p 0.
Proof. We consider the resolution
(1.1) 0 K
X
c
0
X
d0
c
1
X
d1
. . .
dn1
c
n
X
dn
0,
where denotes inclusion. The result follows immediately from theorem 1.2.34.
Henceforth we shall often identify H

p
(X, K) and H
p
deRh
(X, K).
45
46 4. THE TOPOLOGICAL INDEX OF A DIVISOR
Remark 1.3. For a Riemann surface X, we describe the map
: H
2
(X, R) H
2
deRh
(X)(= H
2
deRh
(X, R)) explicitly:
Let | = (U
i
)
i
be a covering of X by open sets, and choose an =
i,j,k
i,j,k
in Z
2
(|, R) Z
2
(|, c).
c
X
is a ne sheaf (example 1.2.11) so, by theorem 1.2.31 we have
Z
2
(|, c) = B
2
(|, c) :=
1
(C
1
(|, c)).
In other words, there is a :=
i,j
i,j
C
1
(|, c) such that =
1
().
Now, exterior dierentiation d : c c
1
is well dened on c and, d = d.
Therefore, since is locally constant we have
0 = d = d
1
() =
1
(d)
so d Z
1
(|, c
1
).
Since c
1
X
is also ne, d =
0
() for some C
0
(|, c
1
). But then
1
d =
d
2
= 0 so () := d Z
0
(|, c
2
) = c
2
(X) is a global 2-form on X. Since d
2
= 0,
d is closed and therefore represents an element of H
2
deRh
(X).
To show that this is independent of the choices made, it is sucient to show
that, given a representative =
i,j,k
i,j,k
of 0 H
2
(|, R), the two-form d
constructed in the above manner is exact. Namely, if is a representative of
0 H
2
(|, R), =
1
for some C
1
(|, R) C
1
(|, c). So, d = (d
i,j
)
i,j
= 0.
Therefore, if C
0
(|, c) is such that
0
= d = 0,
i
=
j
for all i, j.
In other words, is a global 1-form and d is exact.
Proposition 1.4. If X is a compact, connected Riemann surface, then a
dierential 2-form c
2
(X) is exact if and only if
_
X
= 0. In particular
H
2
deRh
(X, K)

= K.
Proof. If c
2
(X) is exact,
_
X
= 0 by Stokes theorem since X is without
boundary.
For the converse see [Lo, 35-36].
We will need the following lemma in the nal chapter.
Lemma 1.5. If X, X
are homotopy equivalent topological spaces, then

H
k
deRh
(X, R)

= H
k
deRh
(X
, R)
for all k.
Proof. [BT, 36]
2. The genus of a Riemann surface
In the preceding (denition 3.2.1) we have dened the genus g of a compact
Riemann surface X as the dimension h
1
(X, O
X
) of the rst cohomology group
H
1
(X, O) of the sheaf of local holomorphic functions (the sheaf with locally con-
vergent power series as germs) on X. In what follows, we consider the equivalence
of alternative denitions in order to arrive at a more obviously topological charac-
terisation.
Let O
1
:= O
1
(X) denote the space of holomorphic 1-forms on X and O
1
X
the
sheaf of locally holomorphic 1-forms on X. Then, if K is a canonical divisor on X,
O
1
:= O(T) = O
K
by proposition 3.1.19.
2. THE GENUS OF A RIEMANN SURFACE 47
Proposition 2.1.
h
1
(X, O) = h
0
(X, O
1
).
Proof. Since O = O
0
where 0 is the divisor with empty support, and
H
0
(X, O
1
) = O
1
= O
K
,
this follows from Serre Duality, proposition 3.2.3.
Definition 2.2. Let X be a Riemann surface. A dierentiable function f :
X C is said to be harmonic if
(f) = 0.
Example 2.3. Let f be a non-vanishing holomorphic function on U X.
Then log [f[
2
= log ff is harmonic. Namely,
2
zz
log [f[
2
=

2
zz
(log f + log f) =

z
(log f) = 0.
Definition 2.4. A complex dierential 1-form on a Riemann surface X is
a harmonic 1-form if it is locally of the form df with f a harmonic function on
U X.
Proposition 2.5. A dierential 1-form on a Riemann surface X is harmonic
if and only if it is of the form
=
1
+
2
with
1
,
2
holomorphic dierentials (locally of the form dz, holomorphic). There-
fore, the space H
1
(X) of harmonic dierentials on X is equal to the direct sum
O
1
(X) O
1
(X).
Proof. [Fo, 154].
In particular, it follows from the above proposition 4.2.5 that
dimH
1
(X) = 2 dimO
1
(X) = 2g.
Lemma 2.6. The space H
1
(X) is isomorphic to the rst de Rham cohomology
group H
1
deRh
(X, C) of X.
Proof. Since H
1
(X) consists of closed dierentials, there is an obvious natural
map H
1
(X) H
1
deRh
(X, C).
The proof that this is an isomorphism rests on Hodge theory which we will not
discuss here. See [Lo, 39].
Theorem 2.7. H
1
(X) denotes the abelianised fundamental group of X, that is
the group obtained by taking the quotient of the fundamental group with its commu-
tator subgroup. It holds that
H
1
deRh
(X, C)

= Hom(H
1
(X), C).
In particular, H
1
deRh
(X, C) has dimension 2g.
Proof. This a theorem of de Rham.
The map
_
: H
1
deRh
(X, C) Hom(H
1
(X), C) is given by
_
([])([]) :=
_

where is a representative of [] H
1
deRh
(X, C) and a representative of []
H
1
(X). A proof that this is an isomorphism can be found in [Lo, 29-30].
Therefore the genus g of a Riemann surface X is a topological invariant of X,

independent of its holomorphic structure. We have
Theorem 2.8.
g := dimH
1
(X, O) = dimH
0
(X, O
1
) =
1
2
dimHom(H
1
(X), C).
From the description of g as half the number of generators of the Abelianised
fundamental group, we are able to come to the intuitive idea of genus as the number
of handles or holes in a Riemann surface.
3. The degree of a divisor
We wish to show that, if L is a holomorphic line bundle on a compact Riemann
surface X, and D is the divisor of a meromorphic section of L, the degree deg(D)
of D is only dependent on the topological structure of L.
Let Pic (X) denote the space of isomorphism classes of holomorphic line bun-
dles
1
on X. C
= GL(1, C) denotes the non-zero complex numbers.

Lemma 3.1. Pic (X) is a group with operation .
Proof. Pic (X) has identity id
Pic (X)
= [X C X].
If L, L
are holomorphic line bundles over X then [L] [L
] := [LL
] is well-
dened and an element of Pic (X). Namely, if L is dened by g
i,j
i,j
and L
by
g
i,j
i,j
with respect to | = U
i
i
, L L
is dened by t
i,j
i,j
,
t
i,j
:= g
i,j
g
i,j
: U
i
U
j
C
and so is also a holomorphic line bundle on X.

In the proof of theorem 3.1.17, we have seen that isomorphic line bundles can
be represented by the same transition functions. Therefore the class of [L L
] is
independent of the choices of representatives L and L
for [L] and [L
].
Finally, by proposition 1.1.20, [L]
1
Pic (X) is given by [L
] where L
is the
holomorphic line bundle dual to L.
Let O
X
denote the sheaf of non-vanishing locally holomorphic functions on X.
Proposition 3.2. For a compact, connected Riemann surface X, there is a
natural isomorphism
H
1
(X, O
)

= Pic (X).
Proof. A representative of an isomorphism class

L of line bundles on X is
dened by a covering | = (U
i
)
iI
of X and non-zero holomorphic functions g
i,j
on U
i
U
j
such that g
i,k
= g
i,j
g
j,k
for all i, j, k I. So, L corresponds to an
element (g
i,j
)
i,j
of Z
1
(|, O
). Conversely, an element (g
i,j
)
i,j
of Z
1
(|, O
) denes
the transition functions relative to | of a line bundle L on X.
1
The notation Pic (X) stands for the Picard group of the space X.
3. THE DEGREE OF A DIVISOR 49
Now, L

L if and only if there exist non-zero holomorphic functions f
i
on
each U
i
such that, if L
is dened by (g
i,j
)
i,j
, g
i,j
=
fi
fj
g
i,j
. But this is precisely
when (g
i,j
)
i,j
and (g
i,j
)
i,j
belong to the same class in H
1
(|, O
).
Therefore, there is a well dened natural one to one correspondence between
Pic (X) and H
1
(X, O
).
Since the trivial bundle on X can be dened by the transition functions g
i,j
(x) =
id = 1 for all x U
i
U
j
, and the correspondence is dened in terms of the mul-
tiplicative property of the transition functions, it is straightforward to show that
this is an isomorphism.
In what follows we will therefore often use isomorphism classes of holomorphic

line bundles and elements of H
1
(X, O
) interchangeably. We may also use the same

notation when referring to representatives of the classes of Pic (X)

= H
1
(X, O
),
though, of course, only where we have shown the results to be independent of the
particular choice of representative.
Remark 3.3. In particular deg(L
D
) = deg(D) by lemma 3.1.15.
Lemma 3.4. The degree map deg : Pic (X) Z is a surjective group homo-
morphism. In other words
(1) deg(L M) = deg(L) + deg(M)
(2) deg(L
) = deg(L)
Proof. Given two line bundles L, and L
over X dened via the same covering

| = U
i
by the transition functions g
i,j
= g
i
/g
j
and g
i,j
= g
i
/g
j
respec-
tively, we calculate the degree of the tensor product L L
. A section of L L
has locally the form

i
i
and on U
i
U
j
we have
j
j
= (g
j,i
i
)(g
j,i
i
) = g
j,i
g
j,i
i
.
But g
j,i
g
j,i
=
gjg
j
gig
i
and so deg(L L
) = deg(L) + deg(L
).
We have already seen that L
can be described by the transition functions

g
1
i,j
=
gj
gi
i,j
. So deg(L
) = deg(L) as required
The map is surjective since for all n Z we can dene a divisor D
n
with
deg(D
n
) = deg(D
n
) = n, for example we take the point divisor D
n
= n x for
some x X.
Example 3.5. In example 3.1.18 we have seen that two holomorphic line bun-
dles on CP
1
are isomorphic if and only if they have the same degree. Therefore,
by the above lemma 4.3.4 and examples 1.1.5 and 3.1.18, if T is the holomorphic
cotangent bundle on CP
1
,
T

= O
CP
1(1) O
CP
1(1).
In particular, if CP
1
= U
0
U
1
is the covering of CP
1
given in example 1.1.5, then
relative to this covering, T has transition function g
0,1
[1 : z] = (
1
z
)
2
=
1
z
2
.
Clearly, if T is the holomorphic tangent bundle of CP
1
, deg(T ) = deg(T) = 2
and
T = T
= O
CP
1(1)
O
CP
1(1)
.
Moreover, by 3.1.18, two holomorphic line bundles on CP
1
are isomorphic if and
only if they have the same degree. So, if we dene
O
CP
1(k) :=
_

k
O
CP
1(1)
=:
k
O
CP
1(1), k > 0
k
O
CP
1(1), k 0
(where
0
O
CP
1(1) = C X X is the trivial line bundle), then, for every
holomorphic line bundle L on CP
1
,
deg(L) = k L

= O
CP
1(k).
2
By proposition 4.3.2, a holomorphic line bundle L over X can be viewed as an
element of the cohomology group H
1
(X, O
) represented by the cocycle g

i,j
i,j
(with respect to a covering | = U
i
i
of X).
Let (L, h) be a holomorphic line bundle with Hermitian metric h over a compact
Riemann Surface X and let =
i
i
be a collection of positive functions
i
on U
i
obtained as in remark 1.1.15 relative to a cover | = U
i
i
of X.
Proposition 3.6. There is a global closed 2-form c
1
(L) on X dened by
c
1
(L)[
Ui
=
i
2
2
zz
log
i
dz dz.
Proof. On U
i
U
j
i
2
2
zz
log
j
dz dz =
i
2
2
zz
log(g
i,j
g
i,j
i
)dz dz
=
i
2
2
zz
(log [g
i,j
[
2
+ log
i
)dz dz =
i
2
2
zz
log
i
dz dz.
So we can write c
1
(L) =
i
2
2
zz
log dz dz is a globally dened 2-form on X. By
denition c
1
(L) is closed.
Since c
1
(L) is closed it represents an element of the de Rham cohomology group
H
2
deRh
(X).
Proposition 3.7. The class of c
1
(L) in H
2
deRh
(X) is independent of the choice
of metric h on L.
Proof. Let
1
h and
2
h be Hermitian metrics on L which induce collections
of positive functions
1
,
2
respectively. Then = (
i
)
i
=
_
1i
2i
_
i
is positive.
Furthermore, is globally dened since
j
=
1
j
2
j
=
1
i
g
i,j
g
i,j
2
i
g
i,j
g
i,j
=
1
i
2
i
=
i
on U
i
U
j
for all i, j. We have
i
2
_

2
zz
log
1
i
dz dz

2
zz
log
2
i
dz dz
_
2
In fact, if E CP
1
is holomorphic, then it can be proved that E is a direct sum of
holomorphic line bundles on CP
1
. Therefore, if rk (E) = r, there exist unique k
1
kr such
that
E

= O
CP
1(k
1
) O
CP
1(kr).
This is the Grothendieck splitting theorem ([Gr]).
=
i
2
2
zz
log dz dz = d
_
i
2
z
log dz
_
which is exact.
Definition 3.8. The 2-form c

1
(L) is called the rst Chern form of L.
We denote the class of c
1
(L) in H
2
deRh
(X), also by c
1
(L). This is the rst
Chern class of the line bundle L.
Lemma 3.9. The map c
1
:
_
Pic (X)

= H
1
(X, O
)
_
H
2
deRh
(X) is a homo-
morphism of groups.
Proof. Let 1 denote the trivial line bundle on X. Then we can choose a metric
h on 1 such that 1 (since 1 is trivial we can choose =
i
for all i relative to
any cover U
i
of X, is as in remark 1.1.15). Since log 1 = 0, c
1
(1) = 0 c
2
(X)
so c
1
(1) = 0 in the additive group H
2
deRh
(X).
If L is a vector bundle with metric h and local frames f
i
and L
is a vector
bundle with metric h
and local frames f
i
, we obtain positive functions (as in remark
1.1.15)
i
:= h
i
(f
i
, f
i
) : U
i
R
+
i
=: and
i
:= h
i
(f
i
, f
i
) : U
i
R
+
i
=:
.
Furthermore hh
denes a metric on LL
and f
i
f
i
are local frames for LL
so
hh
is given by
=
i
i
: U
i
R
+
i
. Then
c
1
(L L
) =
i
2
log(
)dz dz =
i
2
log dz dz +
i
2
log
dz dz
= c
1
(L) + c
1
(L
) H
2
deRh
(X)
as required.
Example 3.10. Let O

CP
1(1) CP
1
be the bundle dened in example 1.1.5.
Then, if z is a local coordinate, we have
c
1
(O
CP
1(1)) =
i
2
2
zz
log(1 +[z[
2
)dz dz =
i
2
1
(1 +[z[
2
)
2
dz dz.
We consider the short exact sequence of sheaves
0 Z
X
O
X
exp
O
X
0.
where exp : O
X
O
X
is the map dened by exp(f) := e
2if
for f O(U) and
U X open.
The connecting homomorphism theorem 1.2.30, says that the sequence
(3.1) H
1
(X, O) H
1
(X, O
H
2
(X, Z) . . .
is exact. In proposition 1.2.25 we have seen that the obvious map : H
2
(X, Z)
H
2
(X, R) induced by the inclusions Z
2
(|, Z) Z
2
(|, R) and C
1
(|, Z) C
1
(|, R)
is a well-dened homomorphism.
Theorem 3.11. Identifying H
2
(X, R) and H
2
deRh
(X) via theorem 4.2.7, it holds
that
c
1
=
(: H
1
(X, O
) H
2
deRh
(X))
where
: H
1
(X, O
) H
2
(X, Z) is the connecting homomorphism.
Proof. By the footnote in [We, 104], there is an open covering | = U
i
i
of
X such that each intersection U
i
U
j
is simply connected. Let g
i,j
i,j
denote the
corresponding transition functions.
We denote a map Z
1
(|, O
) Z
2
(|, Z) which induces the connecting homo-
morphism
also by
and, given g = g
i,j
i,j
Z
1
(|, O
), we wish to construct
a
(g) Z
2
(|, Z).
For all x X, exp
x
: (O
X
)
x
(O
X
)
x
is a surjective homomorphism so,
since connected components of U
i
U
j
are simply connected for all i, j, there is a
f = (f
i,j
)
i,j
C
1
(|, O) such that f
i,j
=
i
2
log g
i,j
= exp
1
(g
i,j
).
Then
1
f C
2
(|, O) and, in particular,
1
f Z
1
(|, O) (since
2
=
2
1
= 0).
Using the fact that g
i,k
= g
i,j
g
j,k
for all i, j, k we have
(
1
f)
i,j,k
=
i
2
(log g
j,k
log g
i,k
+ log g
i,j
) Z O
on each connected component of U
i
U
j
U
k
for all i, j, k.
So,
1
f Z
2
(|, Z) and is a representative for an element
1
f H
2
(X, Z) as
well as for the
1
f H
2
(X, R) .
We construct the corresponding element of H
2
deRh
(X): Using the method of
remark 4.1.3, we let :=
1
f and = f. By the inclusion O c, f = f
i,j
i,j
is
in C
1
(|, c).
L X is a holomorphic line bundle dened by g = g
i,j
i,j
and h is a Her-
mitian metric on L. =
i
i
is a collection of positive functions obtained from h
as in remark 1.1.15. We choose =
i
C
0
(|, c
1
),
i
=
i
2
z
log
i
dz. Then
(
0
)
i,j
=
i

j
=
i
2
z
log
_
j
_
dz
=
i
2
z
log (g
i,j
g
i,j
) dz =
i
2
z
log(g
i,j
)dz = df
i,j
.
Therefore
d = d = d
2
f = 0
so
d =
i
2
2
zz
log dz dz Z
0
(|, c
2
) = c
2
is a global 2-form and equal to c
1
(L).
Remark 3.12. Sometimes it will also be convenient to refer to the element of

H
2
(X, R) represented by
1
(f) := (
1
f)
i,j,k
i,j,k
where
(
1
f)
i,j,k
=
i
2
(log g
j,k
log g
i,k
+ log g
i,j
) Z
also as the rst Chern form c
1
(L) of L.
Theorem 3.13.
_
X
c
1
(L) = deg(L).
Proof. By lemmas 4.3.4 and 4.3.9, it suces to prove this for L = (p) a
line bundle associated to a simple point divisor.
For such an L there exists a global holomorphic section s of L which is non-zero
on X p and which vanishes to rst degree at p X.
Since the result is independent of the choice of metric on L by proposition 4.3.7,
we can choose a covering | = U
i
N
i=0
for X, and a metric h for L as follows:
Let | be an atlas for X such that, relative to a local coordinate, U
0
= B
1
is
a disc centered at p with radius 1 relative to the metric on X (with scaling if
necessary), B1
2
B
1
is the concentric disc with radius
1
2
, and B1
2
U
i
= for all
i ,= 0. We also choose s = z on U
0
, and s 1 on U
i
, i ,= 0. Via a partition of unity
we can construct a Hermitian metric h on L and choose a frame f for U
0
= B
1
such that
0
[
B1
2
:= h(f)[
B1
2
1.
g
i,j
i,j
is the set of corresponding transition functions for L.
We let B
r
:= B(p, r) be a disc about p with radius r in a coordinate neighbour-
hood of X.
There is a positive function [s[
2
on X given by
[s[
2
=
i
s
i
s
i
, on U
i
.
Then [s[
2
is globally dened since
j
s
j
s
j
=
i
g
i,j
g
i,j
g
j,i
s
i
g
j,i
s
i
=
i
s
i
s
i
.
s is holomorphic and non-zero on XB
r
. So, on U
i
B
r
,
log [s[
2
= log
i
+ log [s
i
[
2
, and therefore
(3.2)

2
zz
log =

2
zz
log [s[
2
.
Although c
1
(L, ) =
i
2
2
zz
log dzdz is well-dened on X,

z
log dz is not,
in general, globally dened (as can be easily seen by writing out how it transforms
under the transition functions g
i,j
). However, since [s[
2
is globally dened
2
zz
log [s[
2
dzdz = d(

z
log [s[
2
dz).
so by equation 4.(3.2),
_
X\Br
c
1
(L) =
i
2
_
X\Br
2
zz
log dz dz =
i
2
_
X\Br
d
_

z
log [s[
2
dz
_
.
We may apply Stokes theorem to obtain
_
X\Br
c
1
(L) =
i
2
_
X\Br
d
_

z
log [s[
2
dz
_
=
i
2
_
Br
z
log [s[
2
dz.
For r <
1
2
, [s[
2
= [z[
2
so
i
2
_
Br
z
log [s[
2
dz =
i
2
_
Br
z
log [z[
2
dz =
i
2
_
Br
1
z
dz.
Since, by the residue theorem
_
1
z
dz = 2i when the closed curve of integration
contains z = 0,
_
X
c
1
(L) = lim
r0
i
2
_
X\Br
2
zz
log dz dz = lim
r0
i
2
_
Br
1
z
dz = 1
as required.
Example 3.14. In example 4.3.10, we saw that

c
1
(O
CP
1(1))(z) =
1
(1 +[z[
2
)
2
dz dz.
Since lim
z
1
(1+|z|
2
)
2
= 0, theorem 4.3.13 gives
1
2i
_
C
1
(1 +[z[
2
)
2
dz dz = 1.
Let (
X
denote the sheaf of locally continuous functions on X and (
X
denote
the sheaf of locally continuous functions on X which are nowhere vanishing on their
domain. A holomorphic line bundle L H
1
(X, O
) (see proposition 4.3.2) is, in

particular, a continuous line bundle. That is, L H
1
(X, (
).
Since the sequence
0 Z
X
(
X
exp
(
X
0
is exact, so too is the sequence
(3.3) H
1
(X, () H
1
(X, (
)

H
2
(X, Z) H
2
(X, () . . . .
Lemma 3.15. H
1
(X, () = H
2
(X, () = 0.
Proof. (
X
is a ne sheaf since there exists a continuous partition of unity on
X. The result follows from theorem 1.2.31.
Theorem 3.16. The degree deg D of a divisor D on a compact Riemann sur-
face X is a topological invariant of the line bundle L
D
and not dependent on the
holomorphic structure of L
D
.
Proof. The commutative diagram
(3.4)
0
-
Z
X
-
O
X
exp
O
-
O
X
-
0
0
-
Z
X
id
?
-
(
X
?
exp
C
-
(
?
-
0.
induces a commutative diagram
(3.5)
. . .
-
H
1
(X, O)
exp
O
-
H
1
(X, O
O
-
H
2
(X, Z)
-
. . .
(0 = H
1
(X, ())
?
exp
C
-
H
1
(X, (
C
-
H
2
(X, Z)
id
?
-
0.
([We, 56-57]) where denotes inclusion in each case, and
:=
_
: H
1
(X, O) H
1
(X, ()(= 0)
_
and
:=
_
: H
1
(X, O
) H
1
(X, (
)
_
are natural homomorphism induced by the inclusions (proposition 1.2.25). In par-
ticular 0 is the zero-map.
Since the map H
2
(X, Z) H
2
(X, Z) in diagram 4.(3.5) is the identity, so, in
particular a bijection, by theorem 4.3.11, the following diagram commutes
(3.6)
H
2
(X, R)
_
-
R
. . .
-
H
1
(X, O
O
-
c
1
-
H
2
(X, Z)
6
-
. . .
. . .
-
H
1
(X, (
C
-
H
2
(X, Z)
6
-
. . . .
That is, the map obtained by rst ignoring the holomorphic structure of a line
bundle L X and then applying the connecting homomorphism
C
and
_
is
the same as
_
c
1
. By theorem 4.3.13 (
_
c
1
deg) : Pic (X) Z. In other words,
the degree of a holomorphic and therefore continuous line bundle is a topological
property independent of its holomorphic structure.
It has therefore been shown that the right-hand side of the Riemann-Roch equa-
tion 3.(2.1) can be characterised by the topology of the line bundle L
D
associated
to the divisor D on the Riemann surface X.
CHAPTER 5
Some aspects of algebraic topology and the
T-characteristic
1. Chern classes
We generalise the Chern classes dened in chapter 4 to characteristic classes
of complex dierentiable vector bundles of rank r on smooth complex manifolds of
dimension n.
1.1. Curvature. Connections on a vector bundle E X have been dened
in denition 1.1.30.
Definition 1.1. A connection on a vector bundle E X denes in a
natural fashion an element K
c
2
(X, Hom(E, E)) called the curvature tensor of
.
If f is a frame at x, and A is the matrix for with respect to f at x, then K
at x is given by the r r matrix of 2-forms
(f) = dA+A A.
Lemma 1.2. If g is a change of frame then we have
(gf) = g
1
(f)g
and so K
: c(E) c
2
(E) is globally dened,
Proof. [We, 72-73].
Remark 1.3. Given vector elds V
1
, V
2
dened on an open set of X and a
dierentiable section c(E), we have
K(V
1
, V
2
)() = (V
1
)((V
2
)()) (V
2
)((V
1
)()) ([V
1
, V
2
])().
We have already dened (denition 1.1.34) the canonical connection
(E,h)
with respect to a Hermitian metric h on a holomorphic bundle E.
Definition 1.4. The canonical curvature K
h
on a holomorphic vector bundle
E with hermitian metric h is the curvature form K
h
:= K
(E,h)
.
For simplicity of notation, we will usually refer to the canonical connection
(E,h)
simply as and the canonical curvature as K.
Now let f = (f
i
)
r
i=1
be a holomorphic frame for E on U X open and the
canonical connection with respect to h. A is the corresponding connection matrix
on U. Since f is a frame, h(f) is invertible. Then, on U,
Lemma 1.5. (1) A(f) = h(f)
1
h(f),
(2) (f) = A.
57
58 5. SOME ASPECTS OF ALGEBRAIC TOPOLOGY AND THE T-CHARACTERISTIC
Proof. As in (1.3), page 13, we set h
,
:= h(f
, f
), , = 1 . . . r and simi-
larly dh
,
:= dh(f
, f
), and use the notation h := h(f) = (h

,
)
,
,
dh := dh(f) = (dh
,
)
,
= (h
,
)
,
+(h
,
)
,
=: h +h . Then h, resp. dh
(and h, h) are r r matrices of functions, resp. dierentials on U.
If is the canonical connection with respect to h it holds by equation 1.(1.11)
that
dh
,
= h(f
, f
) +h(f
, f
)
= h(
r
=1
A
,
f
, f
) +h(
r
=1
A
,
f
, f
)
= hA+A
t
h.
Now by equation 1.(1.12), we have
= 0 for c(E) holomorphic. Therefore,

in particular A = A(f) is of type (1,0) so we have h = hA and h = A
t
h.
In other words, if f is a holomorphic frame, A = h(f) h(f)
1
so (1) is proved.
Since
0 = (1) =
_
h h
1
_
= h h
1
+hh
1
,
it holds that
(1.1) h
1
= h
1
h h
1
.
Therefore, by part (1) and 5.(1.1)
A = (h
1
h)
= h
1
h
= h
1
h h
1
h
= h
1
h h
1
h
= A A.
It follows that
(f) : dA+A A = A+A+A A = A A+A+A A = A
as required.
1.2. Invariant Polynomials. Let /

r
denote the set of r r complex ma-
trices.
Definition 1.6. A multi-linear form

: /
r
/
r
C is called invariant
if
(gA
1
g
1
, . . . , gA
k
g
1
) =

(A
1
, . . . , A
k
)
for all g GL(r, C) and all A
i
/
r
.
Now, let X be a complex manifold and
i
c
p
= c
p
(X) for i = 1, . . . k. We
can extend the action of

to /
r
c
p
by settting
(1.2)

(A
1

i
, . . . , A
k

k
) := (
1

k
)
(A
1
, . . . , A
k
) c
pk
.
It is simple to check that

so dened is well dened and multi-linear on /
r
c
p
.
Let : E X be a complex vector bundle with rk E = r. Given a choice of
frame f at x X, the restriction of a homomorphism E E to a single bre E
x
1. CHERN CLASSES 59
can be written as an element of /
r
and it is possible to further extend the action
of

to c
p
( Hom(E, E)) := c(( Hom(E, E)
p
T
X) to obtain a map
(1.3)

X
: c
p
( Hom(E, E)) c
p
( Hom(E, E)) c
pk
(X).
To this end, let U X be an open subset over which E is trivial. If
i

c
p
(U, Hom(E, E)), for i = 1, . . . k and f is a frame for E over U ,
i
(f) has the
form A
i
i
, with A
i
/
r
and
i
c
p
(U), so it is natural to dene
(1.4)

U
(
1
, . . . ,
k
) :=

U
(
1
(f), . . . ,
k
(f)).
Lemma 1.7. This denition is independent of the choice of frame so

U
can be
extended to all of X to obtain a map

X
as in 5.(1.5).
Proof. Given a choice of frame f for E on U X open, another frame on
U has the form g f where g has values in GL(r, C). If Hom(E, E) then
(g f) = g
1
(f)g. Therefore, by denition, if

is an invariant multi-linear form
and
i
c
p
(U, Hom(E, E)) as above,
U
(
1
(g f), . . . ,
k
(f f)) =

U
(g
1
1
(f)g, . . . , g
1
k
(f)g)
=

U
(
1
(g f), . . . ,
k
(f f)).
In other words,

U
is independent of the choice of frame.
Definition 1.8. We call a map : /

r
C an invariant polynomial of degree
k if for every g GL(r, C) and for every A /
r
, (gAg
1
) = (A) and (A) is
a homogeneous polynomial of degree k in the entries of A.
Example 1.9. The determinant map det : /
r
C is an invariant polynomial
of degree r. Furthermore
det(I +A) =
r
k=0
k
(A)
where each
k
is an invariant polynomial of degree k in the entries of A.
Remark 1.10. We note that an invariant k-linear form

naturally denes an
invariant polynomial by setting
(A) :=

(A, . . . , A), A /
r
.
The converse is also true (see [We, 85]): Every invariant polynomial of degree k
acting on /
r
denes a linear map
: /
r
/
r
. .
k times
C
such that

is invariant, and

(A, . . . , A) = (A) for all A. In particular there is a
1-1 correspondence between the invariant k-linear forms on /
r
and the invariant
homogeneous polynomials of degree k acting on /
r
. So we can use the same
symbol for both the invariant k-linear form

and its restriction to the diagonal
.
Remark 5.1.10 implies that, given an invariant polynomial of degree k, we
can extend the action of to c
p
( Hom(E, E)) by constructing the map (5.1.3)
corresponding to

and then evaluating this on the diagonal. That is, for
c
p
( Hom(E, E)),
(1.5)
X
() : c
p
( Hom(E, E)) c
pk
(X),
X
() =

X
(, . . . , ).
Given a connection : c(E) E
1
(E) on E, its curvature K is an element of
c
2
( Hom(E, E)) given locally by an rr matrix of two forms. So, if : /
r
C
is an invariant polynomial of degree k,
X
(K) c
2k
(X) is well-dened by lemma
5.1.7.
Proposition 1.11. If is an invariant homogeneous polynomial of degree k
acting on /
r
, then the 2k-form (K) c
2k
is closed.
Proof. [We, 86-87] or [MS, 296-298].
Therefore (K) denes an element of the de Rham cohomology group H
2k
deRh
(X).
Proposition 1.12. If is an invariant homogeneous polynomial of degree k
acting on /
r
, and ,
are connections on a complex vector bundle E X, the

forms (K
) and (K
) represent the same element in the de Rham cohomology

group H
2k
deRh
(X).
Proof. [We, 86-87] or [MS, 298].
1.3. Chern classes. As before let E

X be a complex diferentiable vector
bundle and let : c(E) c
1
(E) be a connection on E with curvature
K
: c(E) c
2
(E).
As in example 5.1.9 above, for k = 0, . . . , r, we dene the invariant polynomials
k
(A) given by det(I +A) =
r
k=0

k
(A).
Definition 1.13. The k-th Chern formc
k
(E, ) of E relative to the connection
is the closed dierential 2k-form given by
c
k
(E, ) := (
k
)
X
(
i
2
K
)
where (
k
)
X
(
i
2
K
) := (
k
)
X
(
i
2
) is well-dened by lemma 5.1.7.

The total Chern form of E relative to is
c(E, ) =
r
k=0
c
k
(E, )
r
k=0
H
2k
deRh
(X)(X).
By proposition 5.1.11, we can dene the k-th Chern class c
k
(E) of E relative to
as the cohomology class of c
k
(E, ) in H
2k
deRh
(X, C). By proposition 5.1.12, this
is independent of the connection . In particular, if E is holomorphic, we can
henceforth assume that the Chern classes are calculated relative to the canonical
connection.
The total Chern class c(E) of E is then given by
c(E) :=
r
k=0
c
k
(E) H
deRh
(X).
In what follows, we will use the same notation to refer to the (total or k-th)
Chern class in H
deRh
(X) and a representative of this class.
1. CHERN CLASSES 61
The Chern classes c
i
(X) of a complex manifold X are dened to be the Chern
classes c
i
(T ) of the holomorphic tangent bundle of X.
Remark 1.14. Note that the denition of the Chern classes implies that, if
E X is a complex bundle with rk E = r, then c
i
(E) = 0 for all i > r.
Proposition 1.15. The Chern classes have the following properties:
(1) Let E be a complex vector bundle over a dierentiable manifold X. For
all i, c
i
(E) is only dependent on the isomorphism class of E.
(2) If Y is also a dierentiable manifold and : Y X is a dierentiable
map then
c(
E) =
c(E)
where
c(E) is the pullback of the cohomology class c(E) H
deRh
(X, C).
(3) Let E, F be complex dierentiable bundles over a dierentiable manifold
X. Then
c(E F) = c(E) c(F)
where the product is dened in terms of the wedge product in the de Rham
cohomologies. That is
c
k
(E F) =
i+j=k
c
i
(E) c
j
(F).
(4) For all i,
(1.6) c
i
(E
) = (1)
i
c
i
(E).
Proof. [We, 92].
Theorem 1.16. For a holomorphic line bundle L over a Riemann surface X,
the rst Chern class c
1
(L) dened in denition 4.3.8 corresponds to c
1
(L) according
to denition 5.1.13 above.
Proof. Let U
i
i
be a trivialising cover for L and f
i
be a holomorphic frame
for L over U
i
. Furthermore, let h be a Hermitian metric on L, and h(f) = :=
i
i
be as in remark 1.1.15. By, lemma 5.1.5 the matrix A for the connection is given
on U
i
by
A(f) =
1
i
z
dz =

z
log
i
dz
and so the curvature K is represented on U
i
by the matrix
i
= A(f
i
) =

z
_

z
log
i
_
dz dz.
Since L is one dimensional, det(I +) = 1+ for /
1
= C. So by denition
5.1.13 the Chern form c
1
(L) is given by
c
1
(L) =
i
2
zz
log
i
dz dz = c
1
(L) H
2
deRh
(X).

2. Multiplicative sequences and the Todd polynomials
[Hi, 9-16] is the authorative reference for this section.
2.1. Denitions and basic properties. Let R be a commutative ring with
identity id
R
=: 1, and let p
0
= 1 R, and p
i
i=1
be variables.
The ring of polynomials in the variables p
i
with coecients in R is denoted by
= R[p
1
, p
2
, . . . ] which can be graded as follows:
The weight of the product p
j1
p
j2
. . . p
jr
is given by

r
i=1
j
i
. We let
0
:= R
and
k
be the group of polynomials consisting only of terms with weight k. That is,
k
consists of linear combinations of products of weight k. So,
k
is the R-module
spanned by all products p
j1
p
j2
. . . p
jr
of weight k.
Clearly =
k=0
k
.
Definition 2.1. Let (K
j
)
j=0
be a sequence of polynomials in p
i
such that
K
0
= 1 and K
i

i
(so K
i
is a polynomial in the variables p
1
, . . . , p
i
). (K
j
)
j=0
is called a multiplicative (or m-) sequence if every identity of the form
(2.1) 1 +p
1
z +p
2
z
2
+ = (1 +q
1
z +q
2
z
2
+. . . )(1 +r
1
z +r
2
z
2
+. . . )
implies that
(2.2)
j=0
K
j
(p
1
, p
2
, . . . )z
j
=
i=0
K
i
(q
1
, q
2
, . . . , q
i
)z
i
k=0
K
k
(r
1
, r
2
, . . . , r
k
)z
k
.
Definition 2.2. We write
K
_

i=0
p
i
z
i
_
:= 1 +
i=1
K
i
(p
1
, . . . , p
i
)z
i
.
The characteristic power series associated to the m-sequence (K
j
)
j
is given by
K(1 +z) =
i=0
b
i
z
i
, (b
0
= 1, b
i
= K
i
(1, 0, . . . , 0) R, i 1).
Proposition 2.3. Every formal power series Q(z) =
i=0
b
i
z
i
is the charac-
teristic power series of a unique m-sequence (K
j
)
j
.
Proof. [Hi, 10].
Example 2.4. The sequence (p
j
)
j=0
is an m-sequence since in this case equa-
tions (2.1) and (2.2) are equivalent. It follows immediately from the denition
(5.2.2) that (p
j
)
j=0
has characteristic power series 1 +z.
2.2. Todd polynomials. In what follows we will need the m-sequence of Todd
polynomials (T
j
)
j
Definition 2.5. The Todd polynomials (T
j
)
j
are the elements of the m-
sequence associated to the characteristic power series
Q(x) =
x
1 e
x
= 1 +
1
2
x +
k=1
(1)
k1
B
k
(2k)!
x
2k
where B
k
is the k
th
Bernoulli number.
1
1
The Bernoulli numbers (B
k
)
k
are the coecients of (1)
k1 z
2k
2k!
in the power series expan-
sion of
z
e
z
1
= 1
z
2
+
B
1
2!
z
2
. . . .
3. THE TODD CLASS AND THE CHERN CHARACTER 63
The rst few Todd polynomials are given by
2
T
1
(p
1
) =
1
2
p
1
,
T
2
(p
1
, p
2
) =
1
12
(p
2
+p
2
1
),
T
3
(p
1
, p
2
, p
3
) =
1
24
p
2
p
1
,
T
4
(p
1
, p
2
, p
3
, p
4
) =
1
720
(p
4
+p
3
p
1
+ 3p
2
2
+ 4p
2
p
2
1
p
4
1
). (2.3)
In what follows, we will only need the rst two Todd polynomials T
1
and T
2
.
However, the polynomials T
3
, T
4
, . . . , are also required for generalising the results
of this paper to higher dimensions.
Remark 2.6. Let E X be a complex dierentiable bundle with Chern
classes c
k
H
2k
deRh
(X)
rk E=r
k=1
. If we take p
k
= c
k
k
and dene the product
c
i
c
j
:= c
i
c
j
, then
T
1
(c
1
) =
1
2
c
1
,
T
2
(c
1
, c
2
) =
1
12
(c
2
c
1
c
1
), etc.
(2.4)
These are elements of H
deRh
(X) =
dim
R
(X)
k=0
H
k
deRh
(X).
3. The Todd class and the Chern Character
Let X be a locally compact complex manifold and E X be a continuous
complex bundle with Chern classes c
i
H
2i
deRh
(X).
Definition 3.1. The (total) Todd class of E is dened by
td(E) =
j=0
T
j
(c
1
, . . . c
j
) H
deRh
(X)
where (T
j
(c
1
, . . . , c
j
))
j
is the m-sequence of Todd polynomials (denition 5.2.5).
The Todd class, td(X), of a compact complex manifold X, is dened as
td(X) := td(T ),
the Todd class of its holomorphic tangent bundle T = T (X).
Lemma 3.2. If E and F are dierentiable complex bundles over X, then
(3.1) td(E F) = td(E)td(F).
Proof. This follows from the dening property of m-sequences 5.(2.2) and
proposition 5.1.15, 3.
Remark 3.3. Since td(E) is a nite series starting with 1 (the constant func-
tion 1 on X), the inverse (td (E))
1
exists.
2
See [Hi, 14].
Example 3.4. Let L be a continuous complex line bundle over X. Then
c
i
(E) = 0 for i > 1 (by remark 5.1.14), so, by the denition 5.2.2 of the characteris-
tic formal series for an m-sequence, we have that td (E) is equal to the characteristic
formal series associated to (T
j
(c
1
, . . . c
j
))
j
. So, formally
td(E) = Q(d) :=
d
1 e
d
,
where d := c
1
(E).
Example 3.5. The Todd class T(L) of a complex line bundle L X over a
Riemann surface X where c
1
= c
1
(L) is the rst Chern class of L, is given by
td(L) = 1 +T
1
(c
1
) =
d
1 e
d
= 1
1
2
c
1
H
deRh
(X).
The formal polynomial P(x) given by P(x) =
q
j=0
c
j
x
j
, has a unique formal
factorisation
(3.2)
q
j=0
c
j
x
j
=
q
i=0
(1 +
i
x).
The c
i
s are symmetric polynomials in the
i
s.
Remark 3.6. Let
q
j=0
c
j
x
j
=
q
i=0
(1+
i
x) be a formal factorisation. Then,
(3.3) td(E) =
q
i=1
i
1 e
i
H
deRh
(X).
See [Hi, 91].
Definition 3.7. If, E is a continuous complex vector bundle of rank q over
X, with Chern classes c
i
, i = 1, . . . q, and such that

q
j=0
c
j
x
j
has the factorisation
5.(3.2), we dene the (total) Chern character ch(E) of E by
ch(E) :=
q
i=1
e
i
H
deRh
(X).
If L
D
is the line bundle associated with a divisor D on a compact Riemann
surface X, then we denote ch(L
D
) by ch(D).
Lemma 3.8. Let E and F be continuous rank q complex vector bundles over
X. Then
ch(E F) = ch(E) + ch(F)
and
ch(E F) = ch(E)ch(F).
Proof. [Hi, 91,(64)].
Proposition 3.9. X is a compact complex manifold and E is a continuous
complex bundle of rank q over X. Then
q
k=0
(1)
k
ch(
k
E
) = (td(E))
1
c
q
(E)
where c
q
(E) is the q-th Chern class of E.
Proof. [Hi, 92].
4. THE T-CHARACTERISTIC 65
4. The T-characteristic
Definition 4.1. If X has complex dimension n and H
deRh
(X), with
=
<
+
n
where
n
is an n-form and
<
is a linear combination of k-forms
with k < n, where dened, the evaluation of the form over the fundamental class
of X is dened by
[X] :=
_
X
n
.
Definition 4.2. The T-characteristic T(E) of a complex vector bundle E X
is given by
(4.1) T(E) := ch(E)td(X)[X].
Theorem 4.3. The Hirzebruch-Riemann-Roch theorem
For a holomorphic vector bundle E over a compact complex manifold X, the
Euler characteristic (E) of E is equal to the T-characteristic T(E) of E.
Proof. The proof of this theorem provides the subject matter of most of
[Hi].
If X is a Riemann surface and L is a holomorphic line bundle over X, then
denition 5.3.7 gives
ch(L) = 1 +c
1
(L).
Theorem 4.4. The topological index of a divisor D on a compact Riemann
surface X of genus g is equal to the T-characteristic of L
D
. That is
1 g + deg D = ch(D)td(X)[X].
Proof. Using the equations above, we have
td(X) = 1 +T
1
(c
1
(X)) = 1 +
1
2
c
1
(X)
since the complex tangent bundle T has complex rank 1. Now, by proposition
5.1.15, (1.6), c
1
(X) := c
1
(T ) = c
1
(T), and since, if K is a canonical divisor,
L
K
= T (as in proposition 3.1.19). So,
_
X
c
1
(X) =
_
X
c
1
(T) = deg K. We
have
ch(D)td(X) = (1 +c
1
(D))
_
1
1
2
c
1
(T)
_
= 1 +c
1
(D)
1
2
c
1
(T)
and so (by an abuse of notation)
T(L) =
_
X
_
c
1
(D) +
c
1
(X)
2
_
= deg D
1
2
deg K.
We have already seen in corollary 3.2.4 that deg K = 2g 2 and so, it follows
immediately that T(L) = deg D g + 1 as required.
CHAPTER 6
The Topological Index of the Dolbeault operator
Theorem 0.5. The Atiyah-Singer Index formula
X is a compact manifold of real dimension m and E, F are dierentiable com-
plex vector bundles over X and P : c(E) c(F) is an elliptic operator. Let be an
almost complex structure (denition 1.1.27) for the total space of the cotangent bun-
dle T
X. The orientation of T
X is given by local coordinates (x

1
,
1
, . . . , x
m
,
m
),
where (x
1
, . . . , x
m
) are chart coordinates for some x X and
i
= dx
i
for all i.
Then the analytic index (P) of P is equal to its topological index ind
t
(P) where
ind
t
(P) := ch(P) td()[T
X].
ch(P) is dened as the Chern character of the dierence bundle associated to
the operator P. This will be constructed later in this chapter.
1
1. Elements of topological K-theory
Definition 1.1. Let X be a compact topological space. It can be shown that the
isomorphism classes of continuous complex vector bundles over X form an Abelian
semi-group with semi-group operation .
The induced Abelian group is called the topological K-group K(X) of X.
2
If E
is a vector bundle over X the element of K(X) associated to E is denoted by [E].
We will sometimes refer to an element of K(X) as a virtual bundle over X.
Example 1.2. For a point space x we have K(x)

= Z since two complex
vector spaces are isomorphic if and only if they have the same dimension.
Remark 1.3. The elements of K(X) are the classes of formal dierences of the
form E F where E, F are complex vector bundles on X.
EF is equivalent to E
if and only if there exists another complex bundle

G on X such that
E F
G

= E
F G.
The bundle G is necessary to ensure transitivity of the relation since the can-
cellation rule as in Z doesnt, in general, apply to vector bundles. That is, it is not,
in general the case that a +c = b +c a = b where a, b, c are isomorphism classes
of complex bundles over a compact manifold X.
It is straightforward to show that K(X) is, in fact, a ring under the operations
of tensor product and sum .
1
For an elliptic complex E, we can associate a dierential operator P
E
to E and then the
Atiyah-Singer index formula says that inda(E) = indt(P
E
). This will be discussed briey in the
appendix, page 81.
2
K-theory was introduced by Grothendieck in 1957. The denitions here are part of the
topological K-theory rened by Hirzebruch and Atiyah in 1959, rather than the more general
algebraic K-theory still associated with Grothendieck and further developed from the 1960s.
67
68 6. THE TOPOLOGICAL INDEX OF THE DOLBEAULT OPERATOR
Remark 1.4. Let X, X
be compact spaces and f : X X
a continuous map.
Then f induces in a natural fashion a ring homomorphism f
!
: K(X
) K(X) as
follows:
Let U
i
i
be an open covering for X
and E
a vector bundle over X
given by transition functions g

i,j
i,j
with respect to U
i
i
. Since f is continuous,
f
1
U
i
i
is a cover for X and we can dene a vector bundle E X by transition
functions g
i,j
i,j
with g
i,j
:= g
i,j
f[
UiUj
. Then f
!
: K(X
) K(X) is the group

homomorphism induced by the semi-group map f
!
([E
]) := [E].
Proposition 1.5. Let X, X
be homotopy equivalent compact spaces with ho-

motopy equivalence f : X X
. Then, f
!
: K(X
) K(X) is an isomorphism.
Proof. [At, 16-18].
Definition 1.6. If we choose x
0
X, then the inclusion : x
0
X in-
duces a homomorphism
!
: K(X) K(x
0
)

= Z. When X is connected, this is
independent of the choice of basepoint x
0
. We dene
K(X) := Ker
!
.
Now, let Y X be a closed non-empty subspace. The space obtained by con-
tracting Y to a basepoint y
0
Y is denoted by
X
_
Y
, and : y
0

X
_
Y
is the
inclusion map. We dene the relative K-group K(X, Y ) by
K(X, Y ) :=

K
_
X
_
Y
_
.
In particular, K(X, Y ) is an ideal of K
_
X
_
Y
_
.
If Y = , we dene
X
_
as the disjoint sum X + p of X and a point p.

Then
K(X, ) =

K(X +p) = Ker (
!
: K(X +p) K(p)) = K(X).
Definition 1.7. For a locally compact space W we dene K(W) :=

K(W
+
)
where W
+
:= W +p is the one-point compactication of W and
K(W
+
) := Ker (
!
: K(W
+
) K(p)).
Using this denition, if X is a compact or locally compact space and f : W X
is a proper map (that is f
1
(K) W is compact for all K X compact), we can
dene f
!
: K(X) K(W) as in remark 6.1.4.
2. The dierence bundle associated to an elliptic operator
Let Y X be a closed subset of a compact manifold X and let E
0
, E
1
be con-
tinuous vector bundles over X such that there is an isomorphism : E
0
[
Y
E
1
[
Y
.
We construct the dierence bundle d(E
0
, E
1
, ). This will be an element of the
relative K-group K(X, Y ).
Let
Z := (X 0) (X 1) (Y I) X I
where I = [0, 1] is the closed unit interval. p : X I X is the projection. We
dene a vector bundle E over Z as follows:
Z is covered by the open sets
Z
0
= (X 0) (Y [0, 1)) and Z
1
= (X 1) (Y (0, 1]).
2. THE DIFFERENCE BUNDLE ASSOCIATED TO AN ELLIPTIC OPERATOR 69
For i = 0, 1, the restriction of p to Z
i
is p
i
:= p[
Zi
.
Then p
i
E
i
is a vector bundle over the open set Z
i
and p
is an isomorphism
on the open set Z
0
Z
1
. Identifying p
E
i
via p
: p
E
0
[
Y (0,1)

p
E
1
[
Y (0,1)
on Z
0
Z
1
, we glue these together to obtained the desired bundle E Z.
The projection : Z
Z
_
(X 0)
induces a homomorphism
!
: K
_
Z
_
(X 0)
_
K(Z)
and the inclusion : X = X 0 Z induces a homomorphism
!
: K(Z) K(X 0)) = K(X).
Lemma 2.1. The sequence
0 K
_
Z
_
X 0
_

!
K(Z)

!
K(X) 0
is exact.
Proof. Given a vector bundle E on X we form a bundle p
(E) on Z XI,
by taking one copy of E over X0 and another over X1 and gluing them together
with the identity map on Y (0, 1). So
!
is surjective.
Given a trivial virtual bundle E over Z, we observe that its restriction E[
U
to
any open U Z is also trivial. In particular E[
Z(X0)
is trivial. So
(
!
)
1
(0
K(Z)
) = 0
K
Z
_
X 0
.
In other words,
!
is injective.
Finally, let V be a vector bundle over Z. Then
!
([V ]) = 0 if and only if V is
isomorphic to a bundle which is trivial over X 0 Z. That is [V ] Im (
!
).
Lemma 2.2. The exact sequence

(2.1) 0 K
_
Z
_
(X 0)
_

!
K(Z)

!
K(X) 0
splits. So, there is a homomorphism g
!
: K(Z) K
_
Z
_
(X 0)
_
such that g
!

!
is the identity on K
_
Z
_
(X 0)
_
.
Proof. Let f : Z X = (X0) be given by f(x, t) = (x, 0) for all (x, t) Z.
Then f is a deformation retraction and f
!
: K(X) K(Z) is a group homomor-
phism such that
!
f
!
is the identity on K(X). Now, given a K(Z),
a (f
!

!
)(a) Ker (
!
) = Im(
!
)
since
!
(a(f
!
!
)(a)) =
!
(a)
!
(a) = 0. Therefore, since
!
is injective, there is a
unique b K
_
Z
_
(X 0)
_
, such that
!
(b) = a(f
!

!
)(a). We dene g
!
(a) := b.
Then
!
(g
1
f
!

!
(a)) = (f
!

!
)(a) (f
!

!
f
!

!
(a)) = (f
!

!
(a)) (f
!

!
)(a) = 0
and therefore, since
!
is injective
(g
!
f
!

!
)(a) = 0.
So, for b K
_
Z
_
(X 0)
_
,
g
!

!
(b) = b
and g
!
: K(Z) K
_
Z
_
(X 0)
_
is the required homomorphism.
Remark 2.3. Since E[

X0
= p
0
E
0
, the virtual bundle E p
E
0
is trivial over
X = X 0, and therefore E p
E
0
Ker
!
.
In particular g
!
(E p
E
0
) K
_
Z
_
X 0
_
is trivial near the base-point x
0
.
I.e. g
!
(E p
E
0
) Ker
_
K
_
Z
_
X 0
_
K(x
0
)
_
.
Lemma 2.4. K(Z, X 0) is isomorphic to K(X, Y ).
Proof. [At, 69]
Definition 2.5. By the above remark 6.2.3, g
!
(E p
E
0
) K(Z, (X 0)).
The dierence bundle d(E
0
, E
1
, ) is dened as the image of the virtual bundle
g
!
(E p
E
0
) K(Z, X 0) K
_
Z
_
(X 0)
_
in K(X, Y ) K
_
X
_
Y
_
.
Proposition 2.6. Let X, X
be compact spaces and Y X, Y
closed
subspaces. Let E, F and E
, F
be vector bundles over X and X
respectively, and
: E[
Y
F[
Y
,
: E
Y
F
Y
be isomorphisms.
(1) If f : (X, Y ) (X
, Y
) is a map, then
d(f
, f
, f
) = f
!
d(E
, F
).
(2) d(E, F, ) is only dependent on the homotopy class of .
(3) If Y is the empty set then d(E, F, ) = E F.
(4) Let j : (X, ) (X, Y ) be inclusion. j induces naturally a homomorphism
j
!
: K(X, Y ) (K(X, ) = K(X)). If E and F are vector bundles over
X and : E[
Y
F[
Y
is an isomorphism, then
j
!
d(E, F, ) = E F.
(5) d(E, F, ) = 0 if and only if there is a bundle G X such that id[
G|
Y
extends to an isomorphism E G F G over the whole of X.
(6) d(E E
, F F
) = d(E, F, ) +d(E
, F
).
(7) d(E, F, ) +d(E, F,
1
) = 0.
(8) If G is a bundle over X and : F[
Y
G[
Y
is an isomorphism, then
d(E, G, ) = d(E, F, ) +d(F, G, ).
Proof. [AH, 33-34].
Definition 2.7. If W is a rank k vector bundle over a compact manifold X
of dimension n and W is equipped with a metric, then we dene the disk bundle
B(W) W as the closed n +k dimensional submanifold with boundary consisting
of the vectors W such that [[ 1. Similarly, the sphere bundle S(W) of W is
the embedded n + k 1 dimensional compact sub-manifold of B(W) consisting of
elements of W with [[ = 1.
We dene B(X) := B(T
X) (where T
X is the real cotangent bundle of X)

and similarly S(X) := S(T
X).
3. THE THOM ISOMORPHISM 71
Let : B(X) X be the canonical projection. By denition, if E and F are
complex bundles over X and P : c(E) c(F) is an elliptic operator, the symbol,
P
:
F is an isomorphism on S(X) := S(T
X). Here, we let

P
denote
the restriction of the sumbol (P) to the sphere bundle S(X). Therefore, we can
construct the dierence bundle d(P) := d(
E,
F,
P
).
In particular we have already seen in example 2.2.6 that the Dolbeault operator
L
: c(L) c(L T), where X is a Riemann surface and L is a holomorphic line
bundle on X, is an elliptic operator. So we can construct the associated dierence
bundle d(
(,
(L T),
L
).
We will return to this shortly.
3. The Thom Isomorphism
3.1. The Thom isomorphism in topological K-theory. Let X be a com-
pact space and p : W X a vector bundle.
Proposition 3.1. There is an isomorphism
!
: K(X) K(W)
called the Thom isomorphism.
Proof. [AS1, 494].
(Note that, since W is only locally compact, K(W) := K(W +p, p) as in
denition 6.1.7 above.)
Remark 3.2. For a vector bundle E X over X, the pullback bundle p
!
E
over W is well dened. Therefore, K(W) can be regarded as a module over K(X).
Corollary 3.3. For X compact, there is a Thom isomorphism
!
: K(X) K(B(X), S(X))
Proof. There is a natural isomorphism (T
(X))
+
=
B(X)
_
S(X)
. The result
then follows from the denitions 6.1.6 and 6.1.7 together with the above proposition.
In particular, since K(T
X) is a module over K(X), K(B(X), S(X)) is a mod-

ule over K(X).
3.2. The Thom isomorphism for cohomology. Let X be a compact space.
By theorem 4.1.2 we can view the Chern character of a vector bundle E X as
an element of H
(X, R) :=
k
H
k
(X, R).
Lemma 3.4. The Chern character is a semi-group homorphism on the semi-
group of isomorphism classes of complex vector bundles over X. It extends naturally
to a ring homomorphism
ch : K(X) H
(X, R).
Proof. [Hi, 177]

Lemma 3.5. Let f : X X
be a continuous map between compact spaces. If

| = U
i
i
is an open cover for X
then f
1
U
i
i
is an open cover for X and we
can dene a homomorphism
f
: H
(X
, R) H
(X, R)
induced by
f
(
i0,...,iq
i0,...,iq
) :=
i0,...,iq
f
i0,...,iq
where
i0,...,iq
f is dened on f
1
(U
i0
U
iq
).
The diagram
(3.1)
K(X
)
f
!
-
K(X)
H
(X
, R)
ch
?
f
-
H
(X, R)
ch
?
commutes
Proof. [Hi, 177].
In particular, if X is a compact space and Y

X is a closed subspace, with
inclusion map , and y
0
X
_
Y
the inclusion of the basepoint y
0
in
X
_
Y
,
then
(3.2)
K(X, Y )
-
K
_
X
_
Y
_
(
)
!
-
K(y
0
)
H
_
X
_
Y
, R
_
ch
?
(
-
H
(y
0
, R).
ch
?
commutes. And since K(X, Y ) is contained in K
_
X
_
Y
_
, the restriction of the
Chern character is a well-dened homomorphism
ch[
K(X,Y )
: K(X, Y ) H
_
X
_
Y
, R
_
such that Im(ch[
K(X,Y )
) Ker ((
).
Definition 3.6. Let X be a compact space and Y

X a closed subspace, with
inclusion map . For k = 0, 1, . . . , the relative cohomology groups H
k
(X, Y, R) are
dened by means of the long exact sequence
H
k
(X, Y, R) H
k
(X, R)

H
k
(Y, R) H
k+1
(X, Y, R) . . . .
Proposition 3.7. The Poincare duality theorem
Let X be an oriented manifold of real dimension n. Then there is an isomor-
phism
H
k
deRh
(X)

= (H
nk
c
(X))
for each k n.
Proof. [BT, 44-46].
Now if X is a real manifold of dimension n and : Y X a closed orientable
submanifold of dimension k ( is the inclusion map), Poincare duality says that there
exists a unique cohomology class [
Y
] H
nk
deRh
(X) called the closed Poincare dual
of Y in X, such that, if
Y
is a representative for [
Y
]
_
Y
=
_
X

Y
for every closed k-form with compact support on X.
If Y is compact in X, then
_
Y
is well-dened for any dierential k-form on X, not just those with compact sup-
port, and so n denes a functional
_
Y
on H
k
deRh
(X). I.e Y denes an element
of (H
k
deRh
(X))
and so by, Poincare duality, there exists a corresponding [
Y
]
H
nk
c
(X) called the compact Poincare dual of Y .
If the dierential (n k)-form with compact support
Y
is a representative of
[
Y
], then we have the unique characterisation
_
Y
=
_
X

Y
for every closed k-form on X.
Now, if
Y
Y
] then clearly it is also a representative for
[
Y
]. So, when Y X is a compact submanifold, it is possible to ensure that the
closed Poincare dual [
Y
] has compact support.
For simplicity of notation, we will henceforth (except for extra emphasis) make
no distinction between a closed form and its class [] in the de Rham cohomology
H
deRh
(X), and denote them both simply by .
Lemma 3.8. Let : Y X be a k-dimensional compact orientable submanifold
of an n-dimensional orientable manifold X, and
Y
its compact Poincare dual in
X. Then the support of
Y
may be shrunk into any open neighbourhood U Y of
Y in X.
Proof. Let
Y,U
be the compact Poincare dual of Y in U. This has compact
support in U so we can extend it by 0 to a form
Y
H
nk
c
(X). Now, for
H
k
(X),
_
Y
=
_
U

Y,U
=
_
X

Y
so
Y
is the compact Poincare dual of Y in X.
Let : W X be a real vector bundle of rank r over a compact manifold
X of dimension n. We view X as a compact submanifold of W, : X W, by
embedding it as the zero-section in W.
By Poincare duality, the map
H
n
deRh
(X) [] [

X
] H
r+n
c
(W)
is well-dened. In fact, the following proposition tells us that this map is an iso-
morphism, and that H
k
deRh
(X)

= H
k+r
c
(W) for all k 0.
Proposition 3.9. There exists a Thom isomorphism
: H
k
deRh
(X) H
k+r
c
(W)
given by
([]) = [

X
]
where is representative for [] H
k
deRh
(X) and
X
X
]
the Poincare dual of X.
3
Proof. [BT, 63-64].
Henceforth we will denote both
X
and [
X
] by and call the Thom class
of W. It should be clear from the context whether we mean the class or its repre-
sentative.
By lemma 6.3.8 we can shrink the support of to any open neighbourhood of
X W. In particular, we can ensure that the support of is contained in B(W).
Proposition 3.10. Let W X be a real vector bundle of rank r. Then
H
k+r
c
(W)

= H
k+r
(B(W), S(W), R)
for all k 0.
Proof. [LM, 239].
By proposition 6.3.10, we may regard the Thom class H
r
c
(W) of W as an
element of H
r
(B(W), S(W), R). We denote this also by . In particular, we can
also write the Thom isomorphism
: H
k
deRh
(X) H
k+r
c
(W)
as an isomorphism
: H
k
(X, R) H
k+r
(B(W), S(W), R).
We will use these two forms of the Thom isomorphism interchangeably in what
follows.
Definition 3.11. As usual, : X W is the embedding of X as the zero
section in W. The Euler class e(W) of W is the pullback
of to X.
Proposition 3.12. For a complex vector bundle E of rank q over a compact
manifold X, the Euler class e(W) of the underlying real manifold W is equal to the
top Chern class c
q
(E) of E.
Proof. [AS3, 550].
Now let E be a complex vector bundle of rank q over X and : W X
the underlying real vector bundle. Further let j : (B(W), ) (B(W), S(W)) be
the natural embedding. The dening exact sequences of H
(B(W), S(W), R) and

H
(B(W), , R) = H
(B(W), R) induce a homomorphism

j
: H
(B(W), S(W), R) H
(B(W), , R) = H
(B(W), R).
Lemma 3.13. H
(B(W), S(W), R) is a module over H
(B(W), R) = H
(W, R)
and j
: H
(B(W), S(W), R) H
(B(W), R) is a module homomorphism.

3
If X is not compact we can use cohomology with compact vertical support instead and the
following arguments can be carried over to apply to this case.
Proof. [Hi, 180].
Remark 3.14. Let X is a compact manifold and E, F complex vector bundles
over X with P : c(E) c(F) an elliptic operator. Furthermore, let d(P)
K(B(X), S(X)) be the dierence bundle associated to P. Then ([Hi, 187])
ch(d(P)) H
(B(X), S(X), R).

In particular, if H
(T
X) then, by the above lemma 6.3.13, ch(d(P))

H
(B(X), S(X), R) and so, by proposition 6.3.10, we may evaluate the form of
ch(d(P)) over the fundamental class of T
X. That is,
ch(d(P))[T
X]
is well-dened.
Lemma 3.15.
: H
k
deRh
(X) H
k
deRh
(W)

= H
(B(W), R) is an isomor-
phism.
Proof. If X is embedded as the zero-section of W, then, F : W [0, 1] W
given by ((x, ), t) (x, (1 t)), x X, W
x
, t [0, 1], is a deformation
retraction (as is F[
B(W)
). So W, B(W) and X are homotopy equivalent and the
conclusion follows by application of lemma 4.1.5 and then theorem 4.1.2.
We assume the support of is contained in the interior of B(W) (that this

is permitted follows from lemma 6.3.8). Denote the restriction of : W X to
B(W) also by .
Corollary 3.16. Using the notation above,
j
e(W) =
c
q
(E).
This gives
j
()) =
( c
q
(E))
for H
deRh
(X).
Proof. The following diagram naturally commutes:
(3.3)
H
(B(W), S(W), R)
j
-
H
(B(W), R)
H
(X, R).
-
In particular, we have already seen in proposition 6.3.10 that we can regard
as an element of H
(B(W), S(W), R). So

(3.4) j
= (
) =
e(W) =
c
q
(E).
Therefore for H
(X, R) = H
deRh
(X),
j
()) = j
) =
) =
( e(W)) =
( c
q
(E))
where the second equality follows from lemma 6.3.13 and the third from equation
6.(3.4) above.
We will need the following lemma in the next section.
Corollary 3.17. Let E and F be complex vector bundles over a compact
complex manifold X and let W be a even-dimensional real oriented vector bundle
over X, with disc and sphere bundles B(W) and S(W) respectively and projection
: B(W) X. Furthermore, let :
E[
S(W)

F[
S(W)
be an isomorphism.
Then,
e(W)
1
(ch d(
E,
F, )) = ch(E) ch(F).
Proof. By proposition 6.2.6, 4 above,
j
(ch (d(
E,
F, )) = ch(j
!
(d(
E,
F, ))) = ch(
E) ch(
F).
Using corollary 6.3.16,
(ch(E) ch(F)) = ch(
E) ch(
F) = j
ch(d(
E,
F, ))
=
(
1
ch(d(
E,
F, )) e(W)).
Since
is an isomorphism,
e(W)
1
ch(d(
E,
F, )) = ch(E) ch(F)
as required.
4. The Todd genus is a special case of the topological index
In the following we are interested in the case W = T
X the real cotangent

bundle over X.
Let X be an n-dimensional complex compact manifold, so an (m = 2n)-
dimensional real compact manifold. We choose a Hermitian metric h and frame e on
the complex tangent bundle T such that T , and therefore T(= T
), are described
by unitary transition functions. Furthermore, the real tangent and cotangent bun-
dles TX and T
X are described by orthogonal transition functions under the real

bundle metric g induced by h (as in remark 1.1.18) and relative to the frame of the
underlying real bundle induced by e.
If TX is the real tangent bundle of X, and T
X the total space of the real

cotangent bundle, with projection : T
X X, then T
X is a 2m-dimensional
manifold with tangent bundle
(TX)
(T
X). The real metric induced by h

gives an isomorphism TX

= T
X.
Recall the maps 1.(1.5) : O(m) U(m) (the complexication of O(m),
obtained by simply expressing a matrix with real coecients as one with complex
coecients) and 1.(1.4) : U(m) O(2m). By proposition 1.1.24, if W is a real
bundle described by orthogonal transition functions,
( )(W)

= W W.
Therefore,
TX
X

=
TX
TX = (
(TX)).
In particular, the GL(m, C)) bundle, =
(TX) is an almost complex structure

for T
X.
4
4
Note that we could equivalently have chosen to dene the almost complex structure for T
X
in terms of T
X rather than TX. The choice of convention here is for simplicity in the last stage
of the paper.
4. THE TODD GENUS IS A SPECIAL CASE OF THE TOPOLOGICAL INDEX 77
Now, by proposition 1.1.28, equation (1.6), the complexcation
(TX) = TX
C
(
= TX TX) of TX is isomorphic to T T , so

=
(T T ).
Remark 4.1. Note that for a given complex manifold M of real dimension m
and complex dimension n the orientation of TM
C
diers by a factor
(1)
m(m1)/2
= (1)
n(2n1)
= (1)
n
from T (M) T (M). Namely, if the orien-
ation of TM
C
is given by the coordinates z
1
,
1
, . . . , z
m
,
m
(with (z
1
, . . . , z
m
)
chart coordinates for some z X and (
1
, ,
m
) T
z
X), the orientation of
T (M) T (M) is given by the coordinates
1
, . . .
m
, z
1
, . . . , z
m
.
Lemma 4.2. Given an elliptic operator P : E F with E, F complex vector
bundles over the n =
m
2
-dimensional complex manifold X, it holds
ind
t
(P) :=
2m
ch(P) td()[T
X]
= (1)
n
(ch(P)) td(T ) td(T )[X] (4.1)

Proof. By Poincare duality, proposition 6.3.7 and the denition of the Thom
isomorphism in proposition 6.3.9
2m
ch(P) td()[T
X] = (1)
n
(ch(P) td())[X].
Now ch (P) H
c
(T
X) so
1
(ch(P)) is well dened and, by denition

ch(P) =
(
1
(ch(P))) =
(
1
(ch (P))) .
Therefore
ch(P)td() =
(
1
(ch(P))) td =
(
1
(ch (P)) (
)
1
(td()))
=
(
1
(ch(P))) (
)
1
(td()))
since
is an isomorphism (lemma 6.3.15) and therefore (
)
1
(td()) = td((T
X))
is well-dened.
Finally, by lemma 5.3.1, td((T
X)) = td(T T ) = td(T ) td(T ). Putting

this together we get
ind
t
(P) :=
2m
ch(P) td()[T
X]
= (1)
n
(ch(P) td())[X]
= (1)
n
(ch(P)) td((T
X))[X]
= (1)
n
(ch(P)) td(T ) td(T )[X]. (4.2)
Now let X be a compact Riemann surface with holomorphic cotangent bundle

T, and L X a holomorphic line bundle. Then c(LT) = c
0,1
(L). In denition
2.2.5, we have dened the operator
L
: c(L) c(L T)
and shown that it is an elliptic dierential operator of order 1. As before we choose
a Hermitian metric and a frame on T so that T is described by unitary transition
functions, and we let T
X be the real cotangent bundle on X, with T
X
C
its
complexication.
By corollary 1.1.29, T
X

= T so we may identify the (real) disc bundle B(X) =
B(T
X) with B(T). We calculated the symbol (

L
) in example 2.2.6. Since
B(T
X)

= B(T), we can express the symbol as
()(s(x), f(x)) = (is(x)f(x), f(x))
for s c(L) and f a non-constant dierentiable function on X such that f(x) = 0
and f is non-zero on a neighbourhood U of x X. In particular, ()[
S(T
is an
isomorphism.
So (
L
) = (i
, id ) in relation to B(T) where

: L T L T is the
natural map (s(x), f(x)) s(x) f(x).
Now, since for a vector bundle E X, E

= iE with the orientation unchanged,
we can henceforth ignore the factor i in the symbol isomorphism.
Lemma 4.3.
(4.3) d(
L, (
L) (
T), ) =
L d(
C,
T, ).
Proof. We simply follow the general construction of the dierence bundle
given above. The map
:
C[
S(T)

T[
S(T)
is given by (z, , x) := (z, , x) for z C, S(T)[
x
over x X. In particular,
since ,= 0, this is an isomorphism.
Let Z = (B(T) 0) (B(T) 1) (S(T) I) and p : (B(T) I) B(T) be
the projection onto B(T). Z is covered by the open sets
Z
0
= (B(T) 0) (S(T) [0, 1))
and
Z
1
= (B(T) 1) (S(T) (0, 1]),
with p
i
:= p[
Zi
, i = 0, 1 the restriction map.
Then p
0
(
L) = p
0
(
(L C)) is a vector bundle over the open set Z

0
, and
p
1
(L T)is a vector bundle over Z

1
. p

= id
L
p
is an isomorphism on
the open set Z
0
Z
1
.
We can therefore glue the bundles p
0
(
(LC)) over Z
0
and p
1
(LT) over
Z
1
together with the isomorphism id
L
p
on Z
0
Z
1
to obtain a vector bundle
E(L, L T,

) over Z.
Since L = L C and the maps
, p
i
and p
are homomorphisms, we have

E(L, L T,

) = p
LE(C, T, ) over Z and E(L, L T,

) p
L) is trivial
over Z
0
. Applying the splitting map g
!
: K(Z) K
_
Z
_
(T0)
_
(see lemma
6.2.1 above) to this bundle, we obtain
g
!
_
E(L, L T,

) p
L
_
= g
!
_
p
L E(C, T, ) p
(L C)
_
= g
!
p
L g
!
_
E(C, TT, ) p
C
_
=
L d(
C,
T, ).
as required.
So, if
: H
i
deRh
(X) H
i+2
deRh
(B(T)/S(T)) is the Thom isomorphism
ch(
L
) =
1
ch d(
L,
(L T), ) =
1
ch(L d(
C,
T, )).
Lemma 4.4.
(ch(d(
L,
(L T), ))) =
1
_
ch(
L) ch(d(
C,
T, ))
_
= ch (L)
_
ch(d(
C,
T, ))
_
. (4.4)
4. THE TODD GENUS IS A SPECIAL CASE OF THE TOPOLOGICAL INDEX 79
Proof. This follows from the denition of the Thom isomorphism and the fact
that ch is a homomorphism.
Lemma 4.5.
(ch(d(
C,
T, ))) = (td(T ))
1
.
Proof. [Hi, 181-182]
Combining all the above we arrive at

Theorem 4.6. If X is a compact Riemann surface and L a holomorphic line
bundle over X, then
ind
t
(
L
) = T(L)
where T(L) is the Todd genus of L.
Proof. Let T be the complex cotangent bundle of X and

= T T be an
almost complex structure for T
X, the real cotangent bundle of X. By the above

ind
t
(
L
) :=
2
ch(
L
) td[T
X]
= (1)
1
ch(
L
) td))[X] (4.5)
= (1)(1)ch(L)(td(T ))
1
) td(T )td(T )[X] (4.6)
= ch(L) td(T )[X] = T(L). (4.7)
Here 6.(4.5) follows from lemma 6.4.2 and 6.(4.6) follows from lemmas 6.4.4
and 6.4.5
We have therefore proved that, in the case that E is a holomorphic line bundle
over a Rieman surface X, the T-characteristic T(E) of E is equal to the topological
index ind
t
(E) of E. In doing so we have shown that the classical Riemann Roch
theorem 3.2.2 is a special case of the Atiyah-Singer index formula 6.0.5.
Appendix: Elliptic complexes and the topological
index
In the introduction (page 5) it was mentioned that the Atiyah-Singer index
formula can be applied to elliptic complexes (denition 2.2.1) dened on compact
complex manifolds. If E is an elliptic complex on a compact complex manifold X
then, analogue to the operator case, the Atiyah-Singer index formula says that
ind
a
(E) = ind
t
(E).
The analytic index ind
a
(E) of an elliptic complex E has been dened in def-
inition 2.2.10. In this appendix we shall give a denition of the topological index
ind
t
(E) of an elliptic complex E over a compact complex manifold X which cor-
responds to the denition already given in chapter 6, page 67 for the operator
case.
For the sake of completion, we shall briey mention the Dolbeault complex
(E) associated to a holomorphic vector bundle E of rank r over a compact complex
manifold X of dimension n, and its corresponding operator. Of course, since the
Hirzebruch-Riemann-Roch theorem is a special case of the Atiyah-Singer index
formula and we have seen that the analytic index of the Dolbeault complex (E) of
E is equal to the Euler characteristic (E) of E (theorem 3.3.5), it then also holds
that T(E) = ind
t
((E)).
Proposition 7.7. Let (E, h), (F, h
) be Hermitian bundles over a compact com-

plex manifold X and P Di
k
(E, F) a dierential operator. Then P has a unique
formal adjoint P
Di
k
(F, E) with respect to the metrics h, h
. That is, we can

dene an inner product , ) on c(E) and an inner product , )
on c(F) by
, ) =
_
X
h((x), (x))dvol, , c(E)
and
=
_
X
h
(x),
(x))dvol,
c(F)
where dvol is a volume form on X induced by the underlying Riemannian structure.
Then, there exists an unique operator P
: c(F) c(E) such that

P(),
= , P
)), for all c(E),
c(F).
Furthermore, it holds that, if
k
(P)
is the adjoint of the linear map

k
(P)(, v
x
) :
E
x
F
x
,
k
(P
) =
k
(P)
.
In particular, P
is elliptic if and only if P is elliptic.

Proof. [We, 117-118]. The last statement follows immediately from the de-
nitions and the rst two statements.
81
82 APPENDIX: ELLIPTIC COMPLEXES AND THE TOPOLOGICAL INDEX
Lemma 7.8. If (E, h), (F, h
) are Hermitian bundles over a complex manifold

X and P Di
k
(E, F) has adjoint P
Di
k
(F, E), then
Ker (P
) = Coker (P)
and
Ker (P) = Coker (P
).
Proof. Given c(F),
Ker (P
) P
() = 0
(, P
()) = 0, for all c(E)

(P(), ())
= 0, for all c(E)

Coker (P).
The proof of the second statement follows exactly the same method.
Given an elliptic complex E = (E
i
, d
i
)
l
i=0
(of length l + 1) over a compact
complex manifold X:
0 c(E
0
)
d0
c(E
1
)
d1
. . .
d
l1
c(E
l
)
d
l
0,
we can dene a unique operator P
E
: c(F) (F
) by
F :=
k=0
2kl
E
2k
F
:=
k=0
2kl
E
2k+1
and
P
E
:= d
0

_
_
_
k=1
2kl
_
d
2k
+d
2k1
_
_
_
_ : c(F) c(F
).
That is
P
E
(
0
,
2
, . . . ) = (d
0
(
0
) +d
1
(
2
), d
2
(
2
) +d
3
(
4
), . . . ).
Since P
E
is a direct sum of elliptic dierential operators, it is itself an elliptic
dierential operator and therefore the analytic index ind
a
(P
E
) and the topological
index ind
t
(P
E
) of P
E
are dened.
Remark 7.9. If we view an elliptic operator P as a complex E
P
of length 1,
then trivially P
E
P
= P.
Example 7.10. If L is a holomorphic line bundle over a compact Riemann
surface X, then L has Dolbeault sequence (L) given by
0 c(L)

L
c(L T) 0
and so P
(L)
=
L
.
Proposition 7.11. The analytic index of the operator P
E
is equal to the an-
alytic index of the complex E. In particular, the analytic index ind
a
(P
E
) is inde-
pendent of the choice of metrics h
i
on E
i
.
APPENDIX: ELLIPTIC COMPLEXES AND THE TOPOLOGICAL INDEX 83
Proof. This follows from the denitions 2.2.10 and lemma A.7.8 above: Say,
k
and
k+2
are such that d
k
(
k
) d
k+1
(
k+2
) . Then, for all x in X,
[d
k
(
k
)(x)[
2
= h
k+1
(d
k
(
k
)(x), d
k
(
k
)(x))
= h
k
(d
k
(
k
)(x), d
k+1
(
k+2
)(x))
= h
k+1
(d
k+1
d
k
(
k
)(x),
k+2
)(x))
= 0.
So Im(d
k+1
) Im(d
k
) 0 and the dimensions of the kernel and cokernel of P
E
are obatained by summing the dimensions of the respective kernels and cokernels
of the constituent maps d
0
, d
1
, d
2
, d
3
, . . . . Therefore,
ind
a
(P
E
) := dimKer (P
E
) dimCoker (P
E
)
=
_
_
_dimKer (d
0
) +
k=1
2kl
_
dimKer (d
2k
) + dimKer (d
2k1
)
_
_
_
_
_
_
_dimCoker (d
0
) +
k=1
2kl
_
dimCoker (d
2k
+ dimCoker (d
2k1
_
_
_
_
=
_
_
_dimKer (d
0
) +
k=1
2kl
(dimKer (d
2k
) + dimCoker (d
2k1
))
_
_
_
_
_
_dimCoker (d
0
) +
k=1
2kl
(dimCoker (d
2k
) + dimKer (d
2k1
))
_
_
_
= (dimKer (d
0
) dimCoker (d
0
)) +
k=1
2kl
(dimKer (d
2k
) dimCoker (d
2k
))
(dimKer (d
2k1
) dimCoker (d
2k1
))
=
l
i=0
(dimKer (d
i
) dimCoker (d
i
))
=: ind
a
(E)
as required.
Definition 7.12. The topological index ind

t
(E) of an elliptic complex E on
a compact complex manifold X is dened as
ind
t
(E) := ind
t
(P
E
).
Proposition 7.13. The topological index ind
t
(E) is independent of the choice
of metrics h
i
on E
i
, i = 0, . . . , l.
Proof. For more details on the construction see [AS1, 489-508], [AS3, 552-
559].
84 APPENDIX: ELLIPTIC COMPLEXES AND THE TOPOLOGICAL INDEX

If (E, h) is a holomorphic Hermitian bundle over a complact complex manifold
X with rk E = r, as before, the Dolbeault complex, (E), of E is given by
0 c
0,0
(E)

E
c
0,1
(E)

E
. . .

E
c
0,q
(E) 0
Proposition 7.14. The topological index of (E) is equal to T-charateristic
T(E).
Proof. [Hi, 189-190].
In particular, the Hirzebruch-Riemann-Roch theorem (theorem 5.4.3) is a spe-
cial case of the Atiyah-Singer index formula (theorem 6.0.5).
Remark 7.15. Here we have dened the topological index of an elliptic complex
by reducing the complex to an operator between vector bundles. In chapter 6, the
topological K-group of a compact manifold X was dened (denition 6.1.1) as the
Abelian group induced by the semi-group of isomorphism classes of vector bundles.
Equivalently, K(X) can be dened as the group of certain equivalence classes of
complexes over X. Using this denition, the topological index of an elliptic complex
is obtained directly without rst reducing to the operator case. This was also the
approach taken by Atiyah and Singer in their proof of the Atiyah-Singer index
formula using topological K-theory. See [AS1, AS3]
Bibliography
[At] M.F. Atiyah. K-Theory. W.A. Benjamin, inc., New York, 1967.
[AB] M.F. Atiyah and R.Bott. A Lefschetz xed point formula for elliptic complexes: I.
Annals of Mathematics, 86: 374-407, 1967.
[AH] M.F. Atiyah and F. Hirzebruch. Analytic cycles on complex manifolds. Topology 1:25-45,
1962.
[AS] M.F. Atiyah and I. M. Singer. The index of elliptic operators on compact manifolds. Bul-
letin of American Mathematical Society, 69: 422-433, 1963.
[AS1] M.F. Atiyah and I. M. Singer. The index of elliptic operators: I. Annals of Mathematics,
87: 484-530, 1968.
[AS2] M.F. Atiyah and G.B. Segal. The index of elliptic operators: II. Annals of Mathematics,
87: 531-545, 1968.
[AS3] M.F. Atiyah and I. M. Singer. The index of elliptic operators: III. Annals of Mathematics,
87: 547-603, 1968.
[AS4] M.F. Atiyah and I. M. Singer. The index of elliptic operators: IV. Annals of Mathematics,
93: 119-138, 1968.
[AS5] M.F. Atiyah and I. M. Singer. The index of elliptic operators: V. Annals of Mathematics,
93: 139-149, 1968.
[BB] B. Booss and D.D. Bleecker. Topology and Analysis: The Atiyah-Singer Index Formula
and Guage-Theoretic Physics. SpringerVerlag, BerlinHeidelbergNew York, 1985.
[BS] A. Borel and J.-P. Serre. Le theor`eme de Riemann-RochBulletin de la Societe
Mathematique de France, 86: 97-136, 1958.
[BT] R. Bott and L.W. Tu. Dierential Forms in Algebraic Topology. SpringerVerlag, Berlin
HeidelbergNew York, 1982.
[Fo] O. Forster. Lectures on Riemann Surfaces. SpringerVerlag, BerlinHeidelbergNew York,
1981.
[Gr] A. Grothendieck. Sur la classication des bres holomorphes sur la sphere de Riemann.
American Journal of Mathematics, 79: 121-138, 1957.
[Hi] F. Hirzebruch. Topological Methods in Algebraic Geometry. SpringerVerlag, Berlin
[Ho] L. Hormander. Pseudo-dierential Operators. Communications on Pure and Applied Math-
ematics, 18: 501 - 517, 1965.
[Jo] J. Jost.Compact Riemann Surfaces. Princeton University Press, 1997.
[LM] H.B. Lawson, Jr. and M-L. Michelsohn. Spin Geometry. Princeton University Press, 1989.
[Lo] E. Looijenga. Riemann Surfaces. Notes of Dutch Mastermath course Riemann Surfaces,
spring 2007. Available online: www.math.uu.nl/people/looijeng/riemannsurfaces.pdf.
[MS] J.W. Milnor and J.D.Stashe. Characteristic Classes. SpringerVerlag, BerlinHeidelberg
New York, 1974.
[Sh] P. Shanahan. The Atiyah-Singer Index Theorem: An intrduction. SpringerVerlag, Berlin
[We] R. O. Wells, Jr. Dierential Analysis on Complex Manifolds. SpringerVerlag, Berlin
85

Raynor - RR Is A Special Case of The As Index Formula

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Raynor - RR Is A Special Case of The As Index Formula

Загружено:

Авторское право:

Доступные форматы

S.C.

denotes the non-zero complex numbers.)

Definition 1.8. A (local) section of a vector bundle : E X is a map from

X is the (real) cotangent bundle of X.

for all x X. A choice of metric g on E

has transtion functions (g

has transition functions (g

Proposition 1.24. If E X is a complex bundle described by unitary tran-

X its dual, the real

Corollary 1.29. The isomorphism T

X, together with the

U is open and c(U, E), then

be representatives of the same class in H

are mapped onto the same element in

Theorem 2.31. If T is a soft sheaf over X, then, for q = 1, 2, . . . , the coho-

X is the real cotangent bundle of X we dene T

X X be the canonical projection and

F) can be regarded as the subset of E T

X) consisting of the elements (e, ) with T

be another dierential function on X with df

(x) = df(x) = and

another section of E with s

are q p matrices of locally smooth functions. So, in

(x)) (x)s(x) = 0 will remain. For

(x) = df(x) = and s

(x) = s(x). So, the k-symbol is independent of

(X, R), the operator

(X, R), (f)(x) :=

(X, R) such that df(x) = and f(x) = 0 and choose an orthonormal

(X, R). Then,

X denote the real cotangent

X, we choose an f c(X) such that df(x) = and f(x) = 0.

(X) such that df(x) = =

Definition 2.5. Given a holomorphic vector bundle E X, the operator

(X) such that dg(x) = =

(x) for all x X. (D < D

and there is at least

Definition 1.3. A divisor of the form div () for /(X)

is the group of non-zero elements of /(X)) is called a principal divisor.

on X are linearly equiv-

Remark 1.8. Proposition 3.1.2 asserts the existence of a non-constant mero-

are isomorphic if and

are linearly equivalent.

= D + div (), /(X). Then, according to the above

is an isomorphism of holomorphic line bundles. We wish to

= D + div () for some /(X). That is, that

D. This is what we wished to prove.

Example 1.18. By proposition 3.1.11, linearly equivalent divisors on a compact

2. The Riemann-Roch Theorem and the analytic index of a divisor

2. THE RIEMANN-ROCH THEOREM AND THE ANALYTIC INDEX OF A DIVISOR 41

Lemma 3.2.8 can be generalised in an obvious fashion to holomorphic vector

Definition 2.10. The analytic index of a divisor D on a compact Riemann

3. THE EULER CHARACTERISTIC AND HIRZEBRUCH-RIEMANN-ROCH 43

Henceforth we shall often identify H

are homotopy equivalent topological spaces, then

48 4. THE TOPOLOGICAL INDEX OF A DIVISOR

Therefore the genus g of a Riemann surface X is a topological invariant of X,

= GL(1, C) denotes the non-zero complex numbers.

are holomorphic line bundles over X then [L] [L

and so is also a holomorphic line bundle on X.

for [L] and [L

In what follows we will therefore often use isomorphism classes of holomorphic