Introduction To Characteristic Classes and Index Theory Jean-Pierre Schneiders

TEXTOS DE MATEM
ATICA
INTRODUCTION
TO
CHARACTERISTIC CLASSES
AND
INDEX THEORY
Jean-Pierre Schneiders
(
X
)
=
#
(
[
]
.
[
]
)
=
_
X
e
T
X
w
(
E
F
)
=
w
(
E
)
w
(
F
)
(
X
;
O
X
(
D
)
)
=
d
e
g
D
+
1
(
X
;
T
)
=
_
X
c
h
T
t
d
T
X
UNIVERSIDADE DE LISBOA
Faculdade de Ciencias
Departamento de Matematica
Textos de Matematica, Volume 13,
Faculdade de Ciencias da Universidade de Lisboa, 2000
Editores: Fernando C. Silva e L. Trabucho
Ttulo: Introduction to Characteristic Classes and Index Theory
Autor: Jean-Pierre Schneiders
ISBN: 972-8394-12-8
INTRODUCTION
TO
CHARACTERISTIC CLASSES
AND
INDEX THEORY
Jean-Pierre Schneiders
UNIVERSIDADE DE LISBOA
Faculdade de Ciencias
2000
Authors address:
Laboratoire Analyse, Geometrie et Applications (UMR 7539)
Institut Galilee, Universite Paris 13
Avenue Jean-Baptiste Clement
F-93430 Villetaneuse, France
E-mail:
jps@math.univ-paris13.fr
URL:
<http://www-math.univ-paris13.fr/~jps>
Mathematics Subject Classication (2000): 55R40, 57R20, 19L10, 14C40.
Preface
This book is based on a course given by the author at the university of
Lisbon during the academic year 19971998. This course was divided in
three parts dealing respectively with characteristic classes of real and com-
plex vector bundles, Hirzebruch-Riemann-Roch formula and Atiyah-Singer
theorem. In the text which follows, we have decided to treat only the rst
two subjects. For an introduction to the last one, we refer the reader to [21]
for the classical point of view or to [22, 23] for recent developments.
The theory of characteristic classes is a very well developed branch of
mathematics and the literature concerning Riemann-Roch theorem is huge.
So, we will not try to give a full view of these subjects. We will rather
present a few basic but fundamental facts which should help the reader to
gain a good idea of the mathematics involved.
Although the reader is assumed to have a good knowledge of homological
algebra and topology, we begin with a chapter surveying the results of sheaf
theory which are needed in the rest of the book. In particular we recall
results concerning acyclic sheaves, taut subspaces, Poincare-Verdier duality
and Borel-Moore homology, illustrating them by means of examples and
exercises.
We refer the reader who would like a more detailed treatment of this part
to standard texts on sheaf theory (e.g. [11, 6, 28, 18]) and algebraic topology
(e.g. [9, 29, 14]). Older works may also be of interest (e.g. [31, 25, 1, 19]).
Chapter 2 is devoted to Euler classes. As a motivation, we begin by prov-
ing the classical Lefschetz xed point formula and applying it to compute
the Euler-Poincare characteristic of a compact oriented topological mani-
fold by means of its Euler class. Next, we study Thom and Euler classes of
oriented real vector bundles. In particular, we consider Thom isomorphism,
Gysin exact sequence and functorial properties of Euler classes. We end
iii
iv Preface
with results on the Euler class of a normal bundle which allow us to link the
Euler class of an oriented dierential manifold with the one of its tangent
bundle.
The rst part of Chapter 3 deals mainly with Stiefel-Whitney classes.
We dene them in the Grothendieck way by means of projective bundles
and Euler classes of associated tautological bundles. Then, we establish the
pull-back and direct sum formulas. Thanks to the splitting principle, we
also obtain a formula for the Stiefel-Whitney classes of a tensor product. In
the second part of the chapter, we study in general the characteristic classes
of real vector bundles. We begin by classifying real vector bundles of rank
r by means of homotopy classes of maps with values in the innite Grass-
mannian G
,r
. This establishes a link between the characteristic classes
of real vector bundles with coecients in a group M and the elements of
H
(G
,r
; M). By computing this last group for M = Z
2
, we show that the
modulo 2 characteristic classes of real vector bundles are polynomials in
the Stiefel-Whitney classes. We end by explaining the usual cohomological
classication of real vector bundles and deducing from it that line bundles
are classied by their rst Stiefel-Whitney class. This allows us to give a
criterion for a real vector bundle to be orientable.
Chapter 4 is centered on Chern classes. We begin by adapting most of
the results concerning Stiefel-Whitney classes to the complex case. Next,
we consider specic results such as the Chern-Weil method of computing
Chern classes using the curvature of a connection. We also treat briey of
the Chern character. The last part of the chapter is little bit technical. It is
devoted to Iversens construction of the local Chern character for complexes
of complex vector bundles (see [17]) and to its application to the denition
of a local Chern character for coherent analytic sheaves.
For more details on the three preceding chapters, the reader may consults
classical books on the theory of ber bundles (e.g. [30, 20, 16]).
The last chapter of this book is about Riemann-Roch theorem. After
a short review of the niteness and duality results for coherent analytic
sheaves, we reach the central question of this part i.e. how to compute
(X; T)
for a coherent analytic sheaf on a compact complex analytic manifold X.
The answer to this question is essentially due to Hirzebruch (see [15]) and
states that
(X; T) =
_
X
chT tdTX
where chT is the Chern character of T and tdTX is the Todd class of the
tangent bundle of X. To better understand the meaning of this formula, we
Preface v
devote Sections 24 to the easy case of line bundles on complex curves. In
this situation, X is a compact Riemann surface and we can link the gener-
alized Riemann-Roch theorem considered above with the original results of
Riemann and Roch. We end the chapter by proving Hirzebruch-Riemann-
Roch theorem for complex projective manifolds. We follow Grothendieck
approach (see [5]) by reducing the result to the case of the complex projec-
tive space by means of a relative Riemann-Roch formula for embeddings.
However, to treat this relative case, we have not followed [5] but used a sim-
pler method based on the ideas of [2, 3, 4] and the proof of the Grothendieck-
Riemann-Roch formula in [10].
For bibliographical informations concerning the subject treated in this
chapter we refer to [15, 10]. Interesting historical comments may also be
found in [7, 8].
It is a pleasure to end this preface by thanking heartily the CMAF for
its hospitality during my stay at Lisbon university. I think in particular to
T. Monteiro Fernandes who invited me to give the course on which this book
is based and suggested to publish it in this collection. I am also grateful
to her for taking a set of lectures notes which served as a rst draft for
this work. All my thanks also to O. Neto and to the various people who
attended the course and whose interest has been a strong motivation for
turning the raw lecture notes into a book.
Let me nally thank F. Prosmans whose help was invaluable at all the
stages of the preparation of the manuscript.
March 2000 Jean-Pierre Schneiders
Contents
1 Survey of sheaf theory 1
1.1 Abelian presheaves and sheaves . . . . . . . . . . . . . . . . . 1
1.2 Sections of an abelian sheaf . . . . . . . . . . . . . . . . . . . 7
1.3 Cohomology with supports . . . . . . . . . . . . . . . . . . . 9
1.4 Flabby and soft abelian sheaves . . . . . . . . . . . . . . . . . 11
1.5 Cohomology of subspaces and tautness . . . . . . . . . . . . . 13
1.6 Excision and Mayer-Vietoris sequences . . . . . . . . . . . . . 14
1.7 Inverse and direct images . . . . . . . . . . . . . . . . . . . . 21
1.8 Homotopy theorem . . . . . . . . . . . . . . . . . . . . . . . . 31
1.9 Cohomology of compact polyhedra . . . . . . . . . . . . . . . 40
1.10 Cohomology of locally compact spaces . . . . . . . . . . . . . 48
1.11 Poincare-Verdier duality . . . . . . . . . . . . . . . . . . . . . 51
1.12 Borel-Moore homology . . . . . . . . . . . . . . . . . . . . . . 58
1.13 Products in cohomology and homology . . . . . . . . . . . . . 60
1.14 Cohomology of topological manifolds . . . . . . . . . . . . . . 65
1.15 Sheaves of rings and modules . . . . . . . . . . . . . . . . . . 72
2 Euler class of manifolds and real vector bundles 73
2.1 Lefschetz xed point formula . . . . . . . . . . . . . . . . . . 73
2.2 Euler classes of manifolds and index theorem . . . . . . . . . 81
2.3 Basic notions on real vector bundles . . . . . . . . . . . . . . 81
2.4 Orientation of real vector bundles . . . . . . . . . . . . . . . . 83
2.5 Thom isomorphism and Gysin exact sequence . . . . . . . . . 87
2.6 Euler classes of inverse images and direct sums . . . . . . . . 90
2.7 Euler classes of normal bundles . . . . . . . . . . . . . . . . . 94
vii
viii Contents
3 Characteristic classes of real vector bundles 99
3.1 Stiefel-Whitney classes . . . . . . . . . . . . . . . . . . . . . . 99
3.2 Splitting principle and consequences . . . . . . . . . . . . . . 106
3.3 Homotopical classication of real vector bundles . . . . . . . 118
3.4 Characteristic classes . . . . . . . . . . . . . . . . . . . . . . . 126
3.5 Cohomological classication of real vector bundles . . . . . . 133
4 Characteristic classes of complex vector bundles 139
4.1 Generalities on complex vector bundles . . . . . . . . . . . . . 139
4.2 Chern classes . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.3 Chern-Weil construction . . . . . . . . . . . . . . . . . . . . . 146
4.4 Chern character . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.5 Local chern character . . . . . . . . . . . . . . . . . . . . . . 161
4.6 Extension to coherent analytic sheaves . . . . . . . . . . . . . 174
5 Riemann-Roch theorem 179
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.2 Cohomology of compact complex curves . . . . . . . . . . . . 186
5.3 Divisors on complex curves . . . . . . . . . . . . . . . . . . . 188
5.4 Classical Riemann and Roch theorems . . . . . . . . . . . . . 190
5.5 Cohomology of coherent analytic sheaves on P
n
(C) . . . . . . 198
5.6 Hirzebruch-Riemann-Roch theorem for P
n
(C) . . . . . . . . . 207
5.7 Riemann-Roch for holomorphic embeddings . . . . . . . . . . 208
5.8 Proof of Hirzebruch-Riemann-Roch theorem . . . . . . . . . . 214
1
Survey of sheaf theory
1.1 Abelian presheaves and sheaves
Let X be a topological space and let Op(X) denote the category of open
subsets of X and inclusion maps.
Denition 1.1.1. An abelian presheaf on X is a functor F : Op(X)
op

/b where /b denotes the category of abelian groups. In other words, an
abelian presheaf is a law which associates an abelian group F(U) to any
open subset U of X and which associates to any open subset V U a
restriction morphism
r
F
V U
: F(U) F(V )
in such a way that
r
F
WV
r
F
V U
= r
F
WU
for any chain of open subsets W V U of X. We often denote r
F
V U
(s)
simply by s
|V
when there is no risk of confusion.
A morphism of abelian presheaves is simply a morphism of the cor-
responding functors. More explicitly, a morphism of abelian presheaves
f : F G is a law which associates to any open subset U of X a morphism
f(U) : F(U) G(U)
1
2 1. Survey of sheaf theory
in such a way that the diagram
F(U)
f(U)
r
F
V U
G(U)
r
G
V U
F(V )
f(V )
G(V )
is commutative for any open subsets V U of X.
With this notion of morphisms, abelian presheaves form an abelian cat-
egory, denoted Tsh(X).
Examples 1.1.2.
(a) On X, we may consider the abelian presheaf (
0,X
dened by setting
(
0,X
(U) = f : U C : f continuous
and dening the restriction morphisms by means of the usual restric-
tions of functions.
(b) If X is endowed with a Borelian measure , we may consider the
abelian presheaves L
p,X
dened by associating to an open U of X
the quotient of the abelian group
f : U C :
_
U
[f[
p
dV +
by the subgroup
f : U C : f = 0 almost everywhere;
the restriction morphisms being the obvious ones.
Denition 1.1.3. An abelian sheaf on X is an abelian presheaf T such
that
(a) we have
T() = 0;
(b) for any open covering | of an open subset U of X, we have the exact
sequence
0 T(U)

U
V U
T(V )

V,WU
T(V W)
where
U
(s) = (r
F
V U
(s))
V U
and
U
((s
V
)
V U
) = (r
F
(V W)V
(s
V
) r
F
(V W)W
(s
W
))
V,WU
.
1.1. Abelian presheaves and sheaves 3
A morphism of abelian sheaves is a morphism of the underlying abelian
presheaves. With this notion of morphisms, abelian sheaves form a full
additive subcategory of Tsh(X). We denote it by ohv(X).
Examples 1.1.4.
(a) Since continuity is a local property, the abelian presheaf (
0,X
is clearly
an abelian sheaf.
(b) The abelian presheaf L
p,X
associated to a Borelian measure on
X is not in general an abelian sheaf. As a matter of fact, the condition
_
U
[f[
p
dV < +
may be satised locally on U without holding globally. Note however
that the presheaf L
p,X
of functions which are locally in L
p,X
is an abelian sheaf.
(c) If X is a dierential manifold, the presheaves (
p
,X
and Tb
p
X
of smooth
and distributional p-forms are clearly abelian sheaves.
(d) Similarly, on a real analytic manifold X, we have the abelian sheaves
/
p
X
and B
p
X
of analytic and hyperfunction p-forms.
(e) On a complex analytic manifold X, we have the abelian sheaves O
X
,
O
X
of holomorphic and antiholomorphic functions and the abelian
sheaves
p
X
and
p
X
of holomorphic and antiholomorphic p-forms. We
have also the abelian sheaves (
(p,q)
,X
and Tb
(p,q)
X
of smooth and distri-
butional (p, q)-forms and the sheaves /
(p,q)
X
and B
(p,q)
X
of analytic and
hyperfunction (p, q)-forms.
Denition 1.1.5. The stalk at x X of an abelian presheaf F is the
abelian group
F
x
= lim
Ux
U open
F(U)
where the inductive limit is taken over the set of open neighborhoods of x
ordered by . We denote
r
F
xU
: F(U) F
x
the canonical morphism and often use the shorthand notation s
x
for
r
F
xU
(s)
when there is no risk of confusion.
Remark 1.1.6. Let x X and let F be an abelian presheaf on X. To deal
with elements of F
x
, we only have to know that:
(a) for any F
x
there is an open neighborhood U of x in X and
s F(U) such that = s
x
;
(b) if U, U
are two open neighborhoods of x in X and s F(U), s

F(U
) then s
x
= s
x
if and only if there is an open neighborhood U
of x such that U
U U
and s
|U
= s
|U
.
Proposition 1.1.7. Let F be an abelian presheaf on X. Dene /(F)(U)
to be the subgroup of
xU
F
x
formed by elements which are locally in
F (i.e. such that for any x
0
U there is a neighborhood U
0
of x
0
in U
and s F(U
0
) with s
x
=
x
for any x U
0
). Turn /(F) into an abelian
presheaf by setting
_
r
A(F)
V U
()
_
x
=
x
for any x V and consider the morphism
a : F /(F)
dened by setting
[a(U)(s)]
x
= s
x
for any x U. Then, /(F) is an abelian sheaf and for any abelian sheaf
( and any morphism g : F ( there is a unique morphism g
making the
diagram
F
a
/(F)
g
(
commutative. Moreover, a induces an isomorphism
a
x
: F
x
/(F)
x
for any x X.
Denition 1.1.8. We call /(F) the abelian sheaf associated to F.
Examples 1.1.9.
(a) The abelian sheaf L
p,X
considered in Examples 1.1.4 is isomorphic
to /( L
p,X
).
1.1. Abelian presheaves and sheaves 5
(b) To any abelian group M, we may associate the constant presheaf
U M.
This presheaf is in general not a sheaf. We denote its associated sheaf
by M
X
and call it the constant sheaf with ber M. For any open
subset U of X, we have
M
X
(U) = : U M : locally constant.
Proposition 1.1.10. The category ohv(X) is abelian. The kernel of a
morphism
f : T (
is the abelian sheaf
U Ker f(U);
its cokernel is the abelian sheaf associated to the abelian presheaf
U Coker f(U).
Proposition 1.1.11. A sequence of abelian sheaves
0 T
f
(
g
H 0
is exact if and only if the sequence of abelian groups
0 T
x
f
x
(
x
g
x
H
x
0
is exact for any x X.
Examples 1.1.12.
(a) Exponential sequence. Let (
0,X
denotes the (multiplicative) abelian
sheaf formed by non vanishing continuous complex valued functions.
Denote
exp : (
0,X
(
0,X
the morphism which sends a continuous complex valued function f to
expf and denote
2i : Z
X
(
0,X
the morphism which sends a locally constant integer valued function
n to a complex valued function 2in. Then, it follows from the local
existence on C
of the complex logarithm that

0 Z
X
2i
(
0,X
exp
(
0,X
0
is an exact sequence of sheaves.
(b) de Rham sequences. Let X be a dierential manifold of dimension n
and let d denotes the exterior dierential. Working by induction on
n, it is relatively easy to show that for any convex open subset U of
R
n
the sequence
0 C
R
n(U) (
0
,R
n(U)
d

d
(
n
,R
n(U) 0
is exact. This result, often referred to as the Poincare lemma, shows
directly that
0 C
X
(
0
,X
d

d
(
n
,X
0
is an exact sequence of abelian sheaves. Similar results hold with (
,X
replaced by Tb
X
(and by /
X
or B
X
if X is a real analytic manifold).
(c) Dolbeault sequences. Let X be a complex analytic manifold. Then,
for any smooth (p, q)-form we have
d = +
with (resp. ) of type (p +1, q) (resp. (p, q +1)). This gives rise
to morphisms
: (
(p,q)
,X
(
(p+1,q)
,X
, : (
(p,q)
,X
(
(p,q+1)
,X
such that
2
= 0,
2
= 0, + = 0. If U is a convex open subset
of C
n
, one checks by induction on n that the sequences
0
p
C
n
(U) (
(p,0)
,C
n
(U)

(
(p,n)
,C
n
(U) 0
and
0
p
C
n(U) (
(0,p)
,C
n
(U)

(
(n,p)
,C
n
(U) 0
are exact. Therefore, we see that
0
p
X
(
(p,0)
,X

(
(p,n)
,X
0
and
0
p
X
(
(0,p)
,X

(
(n,p)
,X
0
are exact sequences of abelian sheaves.
1.2. Sections of an abelian sheaf 7
Proposition 1.1.13.
(a) Let (I
x
)
xX
be a family of injective abelian groups. Then, the abelian
sheaf 1 dened by setting
1(U) =
xU
I
x
for any open subset U of X and
_
r
I
V U
(s)
x
= s
x
for any open subset V of U and any x V is injective.
(b) Let T be an abelian sheaf on X. Then, there is a monomorphism
T 1
where 1 is a sheaf of the type considered in (a). In particular, the
abelian category ohv(X) has enough injective objects.
Remark 1.1.14. As a consequence of the preceding proposition, we get
that any functor F : ohv(X) / where / is an abelian category has a
right derived functor. Note that, in general, ohv(X) does not have enough
projective objects.
1.2 Sections of an abelian sheaf
Denition 1.2.1. Let A be a subset of X and let T be an abelian sheaf
on X.
A section of T on A is an element

xA
T
x
with the property that that for any x
0
A there is an open neighborhood
U
0
of x
0
in X and s T(U
0
) such that
x
= s
x
for any x AU
0
. When A = X, we call sections of T on A global section
of T.
The support of a section of T on A is the set
supp() = x A :
x
,= 0.
It is the relatively closed subset of A.
Sections of T on A form an abelian group that we denote by (A; T).
Let B be a subset of X such that B A and let (A; T). Then,
r
F
BA
() is the element of (B; T) dened by setting
_
r
F
BA
()
x
=
x
for any x B. We will often use
|B
as a shorthand notation for r
F
BA
().
Of course, we have
supp(
|B
) = supp() B.
Remark 1.2.2. Let U be an open subset of X and let T be an abelian
sheaf on X. Then, the canonical morphism
T(U) (U; T)
which sends s T(U) to (s
x
)
xU
is an isomorphism. Hereafter, we will
often use this isomorphism to identify T(U) and (U; T) without further
notice. Note that if V is an open subset of U, the two denitions of r
F
V U
are compatible with this identication.
Denition 1.2.3. Let X be a topological space and let A be a subspace of
X.
We say that A is relatively Haussdorf in X if for any x ,= y in A we can
nd open neighborhoods U and V of x and y in X such that U V = .
By an open covering of A in X, we mean a set | of open subsets of X
such that for any x A there is U | containing x. Such a covering is
locally nite on A if any x A has a neighborhood V in X for which the
set
U | : U V ,=
is nite.
We say that A is relatively paracompact in X if it is relatively Haussdorf
and if for any open covering | of A in X we can nd an open covering 1
of A in X which is locally nite on A and such that for any V 1 there is
U | with V U.
Remark 1.2.4. One checks easily that a subspace A of a topological space
X is relatively paracompact if it has a fundamental system of paracompact
neighborhoods. This will be the case in particular in the following cases:
(a) X is completely paracompact (e.g. metrizable);
(b) A is closed and X is paracompact.
1.3. Cohomology with supports 9
Other examples of relatively paracompact subspaces are given by relatively
Haussdorf compact subspaces.
Proposition 1.2.5. Let A be a relatively paracompact subspace of X and
let T be an abelian sheaf on X. Then,
(a) for any (A; T) there is an open neighborhood U of A in X and
s (U; T) such that s
|A
= ;
(b) if U, U
are two open neighborhoods of A in X and s (U; T), s

(U
, T) then s
|A
= s
|A
if and only if there is an open neighborhood
U
of A such that U
U U
and s
|U
= s
|U
.
In other words, we have
(A; T) = lim
UA
U open
(U; T).
Proof. See [24].
Proposition 1.2.6. Let A be a topological subspace of X and let T be an
abelian sheaf on X. Then, the abelian presheaf T
|A
dened by setting
T
|A
(U) = (U; T)
for any open subset U of A and
r
F
|A
V U
= r
F
V U
for any chain V U of open subsets of A is an abelian sheaf.
Remark 1.2.7. It follows from the preceding results that sections of T on
A may be considered as the global sections of the abelian sheaf T
|A
.
1.3 Cohomology with supports
Denition 1.3.1. A family of supports of X is a set of closed subsets of
X such that
(a) if F
1
is a closed subset of X included in F
2
, then F
1
;
(b) for any F
1
, F
2
, there is F
3
such that F
1
F
2
F
3
.
Let T be an abelian sheaf on X. Then global sections s of T such that
supp(s)
form an abelian group that we denote
(X; T).
Examples 1.3.2.
(a) The set
X
of all the closed subsets of X is clearly a family of supports
and we have
X
(X; T) = (X; T)
for any abelian sheaf T.
(b) Let F be a closed subset of X. Then, the set
F
of all the closed
subsets of F is a family of supports. In this case we set for short
F
(X; T) =
F
(X; T)
for any abelian sheaf T. Note that this special case allows us to recover
the general one. As a matter of fact, we have
(X; T) = lim
F
(X; T).
(b) Let X be a Haussdorf space. Then, the set
c
of all compact subsets
of X is a family of supports. We set for short
c
(X; T) =
c
(X; T)
for any abelian sheaf T.
Let be a family of supports of X.
Proposition 1.3.3. The functor
(X; ) : ohv(X) /b
is left exact and has a right derived functor
R
(X; ) :
T
+
(ohv(X))
T
+
(/b).
Proof. The left exactness follows directly from the structure of kernels
in ohv(X). The existence of the right derived functor follows from Re-
mark 1.1.14.
Denition 1.3.4. Let T be an abelian sheaf on X. We dene the k-th
cohomology group of X with coecients in T and supports in as the
group
H
k
[R
(X; T)].
To lighten notations, we denote it by
H
k
(X; T).
1.4. Flabby and soft abelian sheaves 11
If is the family of all closed subsets of X, we shorten the notation by
dropping the . Similarly, if is the family
F
(resp.
c
) considered in
Examples 1.3.2, we replace it by F (resp. c).
An abelian sheaf T is -acyclic if H
k
(X; T) = 0 for any k > 0.

Remark 1.3.5. Let T be an abelian sheaf on X. By a well-known result
of homological algebra
R
(X; T)
(X;
)
if
is a right resolution of T by -acyclic sheaves. The aim of the next

section is to give basic examples of such sheaves
1.4 Flabby and soft abelian sheaves
Denition 1.4.1. An abelian sheaf T on X is abby if for any chain U
1

U
2
of open subsets of X
r
F
U
1
U
2
: (U
2
; T) (U
1
; T)
is an epimorphism.
Examples 1.4.2.
(a) Any injective abelian sheaf is abby.
(b) Let (M
x
)
xX
be a family of abelian groups. Then,
U
xU
M
x
is a abby sheaf.
(c) The sheaf L
p,X
of Examples 1.1.4 is abby.
(d) The sheaf B
X
of hyperfunctions is abby.
Proposition 1.4.3. A abby abelian sheaf T is -acyclic for any family
of supports . Moreover, if in an exact sequence of abelian sheaves
0 T ( H 0
T and ( are abby, then so is H.
Denition 1.4.4. An abelian sheaf T on X is -soft if for any chain
F
1
F
2
of
r
F
F
1
F
2
: (F
2
; T) (F
1
; T)
is an epimorphism.
Examples 1.4.5.
(a) Assume X is normal. Thanks to Urysohns extension result, it is clear
that the abelian sheaf (
0,X
is -soft for any family of supports .
(b) If X is a dierential manifold, it follows from the existence of partitions
of unity that the abelian sheaves (
p
,X
and Tb
p
X
are -soft for any
family of supports . The same is true of the sheaves (
(p,q)
,X
and Tb
(p,q)
X
if X is a complex analytic manifold.
Denition 1.4.6. A family of supports is paracompactifying if
(a) any F is paracompact;
(b) for any F
1
, there is F
2
with F
1

F
2
.
Examples 1.4.7.
(a) If X is paracompact, then the family formed by the closed subsets of
X is paracompactifying.
(b) If X is a locally compact space, then the family formed by the compact
subsets of X is paracompactifying.
Proposition 1.4.8. Assume the abelian sheaf T is -soft and the family
is paracompactifying. Then, T is -acyclic. Moreover, if in the exact
sequence of abelian sheaves
0 T ( H 0
T and ( are -soft, then so is H.
Corollary 1.4.9 (de Rham and Dolbeault theorems).
(a) For any dierential manifold X, we have the canonical isomorphisms
H
k
(X; C
X
) H
k
((X; (
,X
)) H
k
((X; Tb
X
))
and
H
k
c
(X; C
X
) H
k
(
c
(X; (
,X
)) H
k
(
c
(X; Tb
X
))
for any k N.
(b) For any complex analytic manifold X, we have the canonical isomor-
phisms
H
k
(X;
p
X
) H
k
((X; (
(p,)
,X
)) H
k
((X; Tb
(p,)
X
))
and
H
k
c
(X;
p
X
) H
k
(
c
(X; (
(p,)
,X
)) H
k
(
c
(X; Tb
(p,)
X
))
for any k, p N.
1.5. Cohomology of subspaces and tautness 13
Proof. Thanks to Examples 1.4.5 (b) and Examples 1.4.7, this follows di-
rectly from Remark 1.3.5 and Proposition 1.4.8.
Exercise 1.4.10. Let B
n
denote the open unit ball of R
n
. Show that for
n 1
H
k
(B
n
; C
B
n
)
_
C if k = 0
0 otherwise
Solution. Recall that for any convex open subset U of R
n
, the Poincare
lemma for d shows that the sequence
0 (U; C
R
n) (U; (
0
,R
n) (U; (
n
,R
n) 0
is exact. Therefore, by the de Rham theorem we have
H
k
(U; C
U
)
_
C if k = 0
0 otherwise
Since B
n
is convex, the conclusion follows.
1.5 Cohomology of subspaces and tautness
Denition 1.5.1. Let X be a topological space and let be a family of
supports on X. We say that a subspace A of X is -taut if the canonical
morphism
lim
UA
U open
H
k
U
(U; T
|U
) H
k
A
(A; T
|A
)
is an isomorphism for any k 0 and any abelian sheaf T on X.
Remark 1.5.2. It is easily seen that a subspace A of X is -taut if and
only if for any abby sheaf T on X
(a) the abelian sheaf T
|A
is A-acyclic;
(b) the canonical morphism
(X; T)
A
(A; T
|A
)
is surjective.
In this case, it follows that
T R(A; T
|A
)
is the right derived functor of
T (A; T).
One should however be aware that this result is false in general.
Proposition 1.5.3. Let X be a topological space and let be a family of
supports on X. Assume A is a topological subspace of X. Then A is -taut
in the following cases:
(a) is arbitrary and A is open;
(b) is paracompactifying and A is closed;
(c) is paracompactifying and X is completely paracompact;
(d) is the family of all closed subset X and A is both compact and
relatively Haussdorf.
n
denote the closed unit ball of R
n
. Show that for
n 1
H
k
(B
n
; C
B
n
)
_
C if k = 0
0 otherwise
Solution. Since B
n
has a fundamental system of neighborhoods formed by
open balls of R
n
, this follows directly from Proposition 1.5.3 and Exer-
cise 1.4.10
1.6 Excision and Mayer-Vietoris sequences
Denition 1.6.1. Let A be a subset of X and let be a family of supports
of X. We set
|A
= F : F A
and
A = F A : F .
Remark 1.6.2. Let A be a subset of X and let a family of supports of
X. Clearly, for any abelian sheaf T on X, we have canonical morphisms
|A
(X; T)
(X; T)
and
(X; T)
A
(A; T)
which induce similar morphisms at the level of derived functors.
1.6. Excision and Mayer-Vietoris sequences 15
Proposition 1.6.3 (Excision). Let A be a subset of X and let be a
family of supports of X. Assume either A is open or A is closed and is
paracompactifying. Then, for any object T
of
T
+
(ohv(X)), we have the
canonical distinguished triangle
R
|X\A
(X; T
) R
(X; T
) R
A
(A; T
)
+1
.
In particular, for any abelian sheaf T, we have the excision long exact
sequence:
H
k
|X\A
(X; T)

H
k
(X; T)

H
k
A
(A; T)
-, *+
/.
()

H
k+1
|X\A
(X; T)

H
k+1
(X; T)

H
k+1
A
(A; T)
Proof. Let us recall the proof of this result since a detailed understanding
of its mechanism will be necessary in various parts of this book. We treat
only the case where A is open; the other case being similar.
Assume ( is a abby sheaf and let s (A; () be such that
supp(s) F A
with F in . The zero section on X F and the section s on A coincide on
(X F) A = A F. Therefore, there is a section s
of ( on (X F) A
such that s
|X\F
= 0, s
|A
= s. Since ( is abby and (X F) A is open, we
may extend s
into a section s
of ( on X. For this section, we have

s
|X\F
= 0, s
|A
= s.
Hence, supp s
F and belongs to . These considerations show that
(X; ()
A
(A; ()
is an epimorphism. Moreover, a simple computation shows that
0
|X\A
(X; T)
(X; T)
A
(A; T)
is exact for any abelian sheaf T. It follows that if (
is a abby resolution
of the complex T
, then
0
|X\A
(X; (
(X; (
)
A
(A; (
) 0
is an exact sequence of complexes of abelian groups. Since abby sheaves are
acyclic for the various functors involved, we get the announced distinguished
triangle. The last part of the result follows from the snakes lemma.
Remark 1.6.4. Let us recall that the snakes lemma states that an exact
sequence of complexes of abelian groups
0 A
0
induces a long exact sequence of cohomology
H
k
(A
)
H
k
(u
H
k
(B
)
H
k
(v
H
k
(C
)
-, *+
/.

k
()

H
k+1
(A
)
H
k+1
(u
H
k+1
(B
)
H
k+1
(v
H
k+1
(C
)
where
k
is dened as follows. Let [c
k
] be a cohomology class in H
k
(C
).
Since v
k
is surjective, there is b
k
B
k
such that
v
k
(b
k
) = c
k
.
Using the fact that d
k
(c
k
) = 0, one sees that d
k
(b
k
) is in Ker v
k+1
. Hence,
there is a
k+1
A
k+1
such that u
k+1
(a
k+1
) = d
k
(b
k
). The cohomology class
of a
k+1
is the image of [c
k
] by
k
. A way to remember this denition is to
follow the dotted path in the following diagram:
A
k
u
k
B
k
v
k

C
k
A
k+1
u
k+1
B
k+1
v
k+1
C
k+1
Proposition 1.6.5 (Mayer-Vietoris). Let A, B be two subsets of X and
let be a family of supports on X. Assume that either A and B are open
or A and B are closed and is paracompactifying. Then, for any object T
of
T
+
(ohv(X)), we have the canonical distinguished triangle
R
(AB)
(A B; F
) R
A
(A; F
) R
B
(B; F
) R
(AB)
(A B; F
)
+1
In particular, if T is an abelian sheaf, we have the Mayer-Vietoris long exact

sequence
H
k
(AB)
(A B; F)

H
k
A
(A; F) H
k
B
(B; F)

H
k
(AB)
(A B; F)
-, *+
/.
()

H
k+1
(AB)
(A B; F)

H
k+1
A
(A; F) H
k+1
B
(B; F)

H
k+1
(AB)
(A B; F)
Proof. The proof being similar to that of Proposition 1.6.3, we will not recall
it in details. We only recall that it is based on the fact that the sequence
0
(AB)
(A B; G)

A
(A; G)
B
(B; G)

(AB)
(A B; G) 0
where (s) = (s
|A
, s
|A
), (s, s
) = s
|AB
s
|AB
is exact when A, B are
open and ( is abby or when A, B are closed, is paracompactifying and
( is -soft.
Exercise 1.6.6.
(a) Let S
n
denotes the unit sphere in R
n+1
. By using Mayer-Vietoris
sequence and de Rham theorem, show that for n 1
H
k
(S
n
; C
S
n
)
_
C if k = 0, n
0 otherwise
(b) Show by excision that for n 1
H
k
c
(B
n
; C
B
n
)
_
C if k = n
0 otherwise
Solution. (a) Assume n 1. Set
S
+
n
= x S
n
: x
n+1
0, S
n
= x S
n
: x
n+1
0
and identify S
n1
with S
+
n
S
n
(see gure 1.6.1). Since S
n
= S
+
n
S
n
, we
have the Mayer-Vietoris long exact sequence
H
k
(S
n
; C)

H
k
(S
+
n
; C) H
k
(S
n
; C)

H
k
(S
n1
; C)
-, *+
/.
()

H
k+1
(S
n
; C)

H
k+1
(S
+
n
; C) H
k+1
(S
n
; C)

H
k+1
(S
n1
; C)
Recall that, for any ]0, 1[, x S
n
: x
n+1
> is an open subset of S
n
which is dieomorphic to a ball of R
n
. Therefore, working as in (a), we see
that
H
k
(S
+
n
; C)
_
C if k = 0
0 otherwise
Of course, there is a similar result for S
n
. It follows that
H
k+1
(S
n
; C) H
k
(S
n1
; C) (*)
Figure 1.6.1:
S
n
S
+
n
S
n1
x
n+1
0
if k > 0 and that H
0
(S
n
; C) and H
1
(S
n
; C) are isomorphic to the kernel and
the cokernel of the morphism
H
0
(S
+
n
; C) H
0
(S
n
; C) H
0
(S
n1
; C) (**)
(, )
|S
n1

|S
n1
Note that since S
+
n
and S
n
are clearly connected spaces, the locally constant
complex valued functions and are in fact constant.
Let us assume rst that n = 1. Since S
0
= 0, 1, we have
H
k
(S
0
; C)
_
C
2
if k = 0
0 otherwise.
Therefore, the morphism (**) becomes up to isomorphisms the additive
map
C
2
C
2
(x, y) (x y, x y)
It follows that H
1
(S
1
; C) C and that H
0
(S
1
; C) C. This last isomor-
phism reecting the fact that the circle S
1
is connected.
Assume now that n > 1 and that
H
k
(S
n1
; C)
_
C if k = 0, n 1
0 otherwise
The morphism (**) now becomes equivalent to the additive map
C
2
C
(x, y) x y
Hence, H
1
(S
n
; C) 0 and H
0
(S
n
; C) C. Moreover, thanks to (*) we see
that for k > 1 we have
H
k
(S
n
; C)
_
C if k = n
0 otherwise
The conclusion follows by induction.
(b) Assume n 1. Since B
n
B
n
= S
n1
and since any closed subset
of B
n
is compact, we have the excision distinguished triangle
R
c
(B
n
; C) R(B
n
; C) R(S
n1
; C)
+1
From the associated long exact sequence, we deduce that

H
k
(S
n1
; C) H
k+1
c
(B
n
; C)
for k > 0 since in this case
H
k
(B
n
; C) H
k+1
(B
n
; C) 0.
Moreover, we see also that H
0
c
(B
n
; C) and H
1
c
(B
n
; C) are the kernel and
cokernel of the morphism
H
0
(B
n
; C) H
0
(S
n1
; C)

|S
n1
For n = 1, this morphism is equivalent to
C C
2
x (x, x)
and we get H
0
c
(B
1
; C) 0 and H
1
c
(B
1
; C) C.
For n > 1, it becomes equivalent to
C C
x x
and we get H
0
c
(B
n
; C) H
1
c
(B
n
; C) 0. The conclusion follows by induc-
tion on n.
Exercise 1.6.7. Let I = [a, b] be a compact interval of R. Show by using
tautness and a suitable Mayer-Vietoris sequence that for any abelian sheaf
T on I, we have
H
k
(I; T) = 0
for k > 1. Establish also that this relation holds for k = 1 if
(I; T) T
x
is an epimorphism for every x I. As an application compute
H
(I; M
I
)
for any abelian group M.
Solution. Fix k > 0 and assume there is c H
k
(I; T) which is non-zero.
Set
x
0
= infx I : c
|[a,x]
,= 0.
Since by tautness
lim
x>x
0
H
k
([a, x]; T) H
k
([a, x
0
]; T)
we see that c
|[a,x
0
]
,= 0 and hence that x
0
> a. Since
lim
x<x
0
H
k
([x, x
0
]; T) = H
k
(x
0
; T) = 0,
there is x < x
0
with c
|[x,x
0
]
= 0. For such an x, the decomposition
[a, x
0
] = [a, x] [x, x
0
]
gives the Mayer-Vietoris exact sequence
H
k1
([a, x
0
]; F)

H
k1
([a, x]; F) H
k1
([x, x
0
]; F)

H
k1
({x}; F)
-, *+
/.
()

H
k
([a, x
0
]; F)

H
k
([a, x]; F) H
k
([x, x
0
]; F)

H
k
({x}; F)
If k > 1 or k = 1 and
(I; T) T
x
is an epimorphism, it follows from this sequence that
H
k
([a, x
0
]; T)

H
k
([a, x]; T) H
k
([x, x
0
]; T).
This gives us a contradiction since both c
|[a,x]
and c
|[x,x
0
]
are 0 although
c
|[a,x
0
]
,= 0.
The application to T = M
I
is obvious. We get
H
k
(I; M
I
) =
_
M if k = 0;
0 otherwise.
1.7. Inverse and direct images 21
1.7 Inverse and direct images
Denition 1.7.1. Let f : X Y be a morphism of topological spaces and
let ( be an abelian sheaf on Y .
Let U be an open subset of X. We dene f
1
(()(U) as the abelian
subgroup of
xU
(
f(x)
formed by the elements such that for any x
0
U
there is an open neighborhood U
0
of x
0
in U, an open neighborhood V
0
of
f(U
0
) and s ((V
0
) such that
x
= s
f(x)
for any x U
0
. Clearly,
U f
1
(()(U)
is an abelian sheaf on X. We call it the inverse image of ( by f. Note that
by construction there is a canonical pull-back morphism
f
: (Y ; () (X; f
1
(()).
Remark 1.7.2. It follows at once from the preceding denition that we
may identify f
1
(()
x
and (
f(x)
. In particular, the functor
f
1
: ohv(Y ) ohv(X)
is exact.
Proposition 1.7.3. Let f : X Y and g : Y Z be two morphisms of
topological spaces. Then,
f
1
(g
1
(H)) (g f)
1
(H)
canonically and functorially for H ohv(Z). Moreover,
(id
1
X
)(T) T
canonically and functorially for T ohv(X).
Examples 1.7.4.
(a) Let a
X
: X pt be the morphism which maps the topological space
X to a point. Identifying sheaves on pt with their global sections,
we have
M
X
a
1
X
(M)
(b) Combining the preceding proposition with example (a), we get a
canonical isomorphism
f
1
(M
Y
) M
X
for any abelian group M and any morphism of topological spaces
f : X Y .
(c) Let i : A X be the canonical inclusion of a subspace A of X and
let T be an abelian sheaf on X. One checks easily that
i
1
(T) = T
|A
.
Denition 1.7.5. We say that (X; T) is an abelian sheafed space if X is a
topological space and T is an abelian sheaf on X. We say that
(f; ) : (X; T) (Y ; ()
is a morphism of abelian sheafed spaces if (X; T) and (Y ; () are abelian
sheafed spaces, f : X Y is a morphism of topological spaces and
: f
1
( T
is a morphism of sheaves.
Examples 1.7.6.
(a) Let f : X Y be a morphism of topological spaces. Then the
f
: f
1
M
Y
M
X
gives rise to a morphism of sheafed spaces
(f; f
) : (X; M
X
) (Y ; M
Y
).
(b) Let f : X Y be a morphism of topological spaces. Then,
g g f
induces a canonical morphism of abelian sheaves
f
: f
1
(
0,Y
(
0,X
and hence a morphism of sheafed spaces
(f; f
) : (X; (
0,X
) (Y ; (
0,Y
).
(c) Similarly, a morphism of dierential manifolds f : X Y induces
morphisms of abelian sheaves
f
: f
1
(
p
,Y
(
p
,X
corresponding to the pull-back of dierential p-forms. Hence, we have
a canonical morphism of sheafed spaces
(f; f
) : (X; (
p
,X
) (Y ; (
p
,Y
).
Proposition 1.7.7. A morphism of sheafed spaces
(f; ) : (X; T) (Y ; ()
induces a morphism
R(Y ; () R(X; T)
and, in particular, a canonical morphism
(f; )
: H
(Y ; () H
(X; T).
Moreover, these morphisms are compatible with the composition of mor-
phisms of sheafed spaces.
Proof. We have a functorial morphism
(Y ; () (X; f
1
() (*)
Since the functor f
1
is exact, a standard result of homological algebra
gives us a morphism
R(Y ; () R(X; f
1
().
Composing with the canonical morphism
R(X; f
1
() R(X; T)
induced by , we get the requested morphism. It is possible to visualize
this abstract construction more explicitly as follows. Assume we are given
a quasi-isomorphism
(

where
is a complex of (Y ; )-acyclic sheaves and a commutative diagram

of the form
f
1
(()
f
1
()
f
1
(
where is a quasi-isomorphism and 1
is a complex of (X; )-acyclic

sheaves. Then, in
T
+
(/b), we have the commutative diagram
R(Y ; ()
/o
(1)
(2)
(Y ;
)
(3)
(X; f
1
(
))
(4)
R(X; T)
/o
(5)
(X; 1
)
where (1) and (5) are induced by and , (3) is induced by morphism of
the type (*), (4) is induced by and (2) is the morphism dened abstractly
above.
Examples 1.7.8.
(a) Let f : X Y be a morphism of topological spaces. By applying the
preceding proposition to the canonical morphism of sheafed spaces
(f; f
) : (X; M
X
) (Y ; M
Y
)
we get a canonical morphism
R(Y ; M
Y
) R(X; M
X
)
in
T
+
(/b). The associated morphism
(f; f
: H
(Y ; M
Y
) H
(X; M
X
)
is often simply denoted by f
. Clearly, f
= id if f is the identity of
X. Moreover, if g : Y Z is another morphism of topological spaces,
we have f
= (g f)
. This shows that X H
(X; M
X
) is a
functor on the category of topological spaces. In particular, X Y
implies H
(X; M
X
) H
(Y ; M
Y
).
(b) Let f : X Y be a morphism of dierential manifolds. It is well
known that the pull-back of dierential forms is compatible with the
exterior dierential. In other words, we have
d(f
()) = f
(d)
for any (Y ; (
p
,Y
). Using what has been recalled in the proof
of the preceding proposition, we see that we have the commutative
diagram
H
p
(Y ; C
Y
)
f
H
p
(X; C
X
)
O
H
p
((Y ; (
,Y
))
f
H
p
((X; (
,X
))
where the rst horizontal arrow is the one dened in (a) and the
two vertical isomorphisms come from the de Rham theorem. Thanks
to what has been said in (a), we see also that, up to isomorphism,
de Rham cohomology of X
H
((X; (
,X
))
depends only on the topology of X and not on its dierential structure.
Denition 1.7.9. Let f : X Y be a morphism of topological spaces and
let T be an abelian sheaf on X.
The direct image of the abelian sheaf T by f is the sheaf f(T) on Y
dened by setting
f(T)(V ) = T(f
1
(V ))
for any open subset V of Y ; the restriction morphisms being the obvious
ones.
Example 1.7.10. Let a
X
: X pt be the morphism which maps the
topological space X to a point. Identifying sheaves on pt with their
abelian group of global sections, we have
a
X
(T) (X; T)
for any abelian sheaf T on X.
Proposition 1.7.11 (Adjunction formula). We have a canonical func-
torial isomorphism
Hom
Shv(Y )
((, f(T)) Hom
Shv(X)
(f
1
((), T).
This isomorphism is induced by two canonical functorial morphisms
f
1
(f(T)) T
and
( f(f
1
(()).
Proof. We will only recall the construction of the canonical functorial mor-
phism
Hom
Shv(Y )
((, f(T)) Hom
Shv(X)
(f
1
((), T) (*)
Let h : ( f(T) be a morphism of abelian sheaves. For any open subset
V of Y , we get a morphism
h(V ) : ((V ) T(f
1
(V ))
and using Remark 1.1.6, it is easy to deduce from these morphisms a mor-
phism
h
x,f(x)
: (
f(x)
T
x
for any x X. Now, let U be an open subset of X and let f
1
(()(U).
One checks directly that (h
x,f(x)
(
x
))
xU
T(U). Hence, for any open
subset U of X, we have a morphism
h
(U) : f
1
(()(U) T(U).
These morphisms give rise to the morphism
h
: f
1
(() T
image of h by (*).
Proposition 1.7.12. Let f : X Y and g : Y Z be two morphisms of
topological spaces. Then,
g(f(T)) (g f)(T)
canonically and functorially for T ohv(X). Moreover,
(id
X
)(T) T
canonically and functorially for T ohv(X).
Proposition 1.7.13. The direct image functor
f : ohv(X) ohv(Y )
Rf :
T
+
(ohv(X))
T
+
(ohv(Y )).
Moreover, if g : Y Z is another morphism of topological spaces, then
Rg Rf R(g f).
Example 1.7.14. Denoting a
X
: X pt the canonical map, we deduce
from the functorial isomorphism
a
X
(T) (X; T)
that
Ra
X
() R(X; ).
Therefore, the second part of the preceding proposition contains the fact
that
R(Y ; Rf(T)) R(X; T);
a result which replaces Lerays spectral sequence in the framework of derived
categories.
Proposition 1.7.15. Assume that
(a) the map f is closed (i.e. such that f(F) is a closed subset of Y if F is
a closed subset of X);
(b) the ber f
1
(y) is a taut subspace of X for any y f(Y ).
Then,
[Rf(T)]
y
= R(f
1
(y); T)
for any y Y and any abelian sheaf T on X.
Example 1.7.16. Thanks to Proposition 1.5.3, it is clear that the condi-
tions on f in the preceding proposition are satised if one of the following
conditions holds :
(a) f is closed and X is metrizable;
(b) f is closed, Y is Haussdorf and X is paracompact;
(c) f is proper and X is Haussdorf.
Remark 1.7.17. Note that under the assumptions of Proposition 1.7.15,
we have of course
[f(T)]
y
= (f
1
(y); T)
but that this formula may be false in general.
Corollary 1.7.18 (Vietoris-Begle). Assume that
(a) the map f is closed and surjective,
(b) the ber f
1
(y) is a taut subspace of X for any y Y ,
(c) the ber f
1
(y) is connected and acyclic (i.e.
H
k
(f
1
(y); M
f
1
(y)
) 0
for any k > 0 and any abelian group M) for any y Y .
Then, the canonical morphism
( Rf(f
1
(())
is an isomorphism for any (
T
+
(ohv(Y )). In particular, the canonical
morphism
R(Y ; () R(X; f
1
()
is an isomorphism in
T
+
(/b).
Proof. It is sucient to note that
(f
1
()
|f
1
(y)
((
y
)
f
1
(y)
and that the canonical morphism
M H
0
(f
1
(y); M
f
1
(y)
)
is an isomorphism if f
1
(y) is non-empty and connected.
Figure 1.7.1:
x
0
(x
1
, , x
n
)
x(U X)
x(U)
Remark 1.7.19. Let us recall a few facts about Stokes theorem which
are needed in the following exercise. Let X be an oriented n-dimensional
dierential manifold with boundary. As is well-known, the orientation of X
induces an orientation on X. This orientation is characterized by the fact
that if x : U R
n
is a positively oriented local coordinate system of X on
an open neighborhood of u X such that
x(u) = 0, x(U) = x B
n
: x
0
0, x(U X) = x B
n
: x
0
= 0;
(see Figure 1.7.1) then (x
1
, , x
n
)
|UX
is a positively oriented coordinate
system of X. With this orientation of X, Stokes formula states that
_
X
d =
_
X
for any
c
(X; (
n1
,X
).
Exercise 1.7.20.
(a) Let X be an oriented n-dimensional dierential manifold. Show that
integration of smooth compactly supported n-forms induces a mor-
phism
_
X
: H
n
c
(X; C) C
(b) Let X be an oriented n-dimensional dierential manifold with bound-
ary. Express the various morphisms of the excision long exact sequence
H
k
c
(X X; C)
u
k
H
k
c
(X; C)
v
k
H
k
c
(X; C)
-, *+
/.

k
()

H
k+1
c
(X X; C)
u
k+1
H
k+1
c
(X; C)
v
k+1
H
k+1
c
(X; C)
in terms of de Rham cohomology. Show in particular that
_
X
n1
c
n1
=
_
X
c
n1
for any c
n1
H
n1
c
(X; C).
Solution. (a) Integration gives us a morphism
_
X
:
c
(X; (
n
,X
) C.
By Stokes theorem, we know that
_
X
d =
_
X
= 0
for any
c
(X; (
n1
,X
). Therefore,
_
X
induces a morphism
_
X
:
c
(X; (
n
,X
)/d
c
(X; (
n
,X
) C
and the conclusion follows from the de Rham theorem.
(b) Since (
,X
is a c-soft resolution of the sheaf C
X
, the long exact
sequence of cohomology comes from the application of the snakes lemma
to the exact sequence of complexes
0
c
(X X; (
,X
|X\X
)
c
(X; (
,X
)
c
(X; (
,X
|X
) 0
The canonical restriction morphism
(
,X
|X\X
(
,X\X
is clearly an isomorphism. As for the restriction morphism
(
,X
|X
(
,X
it is a quasi-isomorphism since in the commutative diagram of complexes
C
X
|X
/o

C
X
,X
|X
,X
both vertical arrows are quasi-isomorphisms. It follows that
c
(X X; (
,X
|X\X
)
c
(X X; (
,X\X
)
and that
c
(X; (
,X
|X
)
qis
(X; (
,X
).
Computation of u
k
. Let c
k
H
k
c
(X X; C) be represented by
k

c
(X X; (
k
,X\X
).
By extension by zero,
k
becomes a class
k

c
(X; (
k
,X
)
and u
k
(c
k
) is represented by
k
.
Computation of v
k
. Let c
k
H
k
c
(X; C) be represented by
k

c
(X; (
k
,X
)
and let j : X X denotes the inclusion map. Then, v
k
(
k
) is represented
by
j
(
k
)
c
(X; (
k
,X
).
Computation of
k
. Let c
k
H
k
(X; C) be represented by
k

c
(X; (
k
,X
).
Set K = supp
k
. We know from elementary dierential geometry (collar
neighborhood theorem) that there is a neighborhood U of X in X, a
dierentiable map p : U X such that p j = id
X
and a smooth
function equal to 1 on a neighborhood of X and such that supp is
a p-proper subset U. Denote
k
the k-form on X obtained by extending
p
by zero outside of U. Clearly,
k
has compact support and
j
k
=
k
.
Therefore, it follows from the snakes lemma that
k
(c
k
) is represented by
d
k
|X\X
. As expected, this form has compact support. As a matter of
fact,
d
k
|U
= d p
k
+ p
d
k
= d p
k
and d = 0 in a neighborhood of X. Assuming now that k = n 1, we
see using Stokes theorem that
_
X\X
n1
c
n1
=
_
X
d
n1
=
_
X
n1
=
_
X
j
n1
=
_
X
n1
=
_
X
c
n1
.
1.8. Homotopy theorem 31
1.8 Homotopy theorem
Denition 1.8.1. Let T, ( be abelian sheaves on X and Y respectively
and let
(h; ) : (X I; p
1
X
T) (Y ; ()
be a morphism where I = [0, 1] and p
X
: X I X denotes the rst
projection. Let t [0, 1]. We denote
i
t
: X X I
the morphism dened by setting i
t
(x) = (x, t) and h
t
the morphism h i
t
.
Applying i
1
t
to
: h
1
( p
1
X
T
we get a morphism
i
1
t
: i
1
t
h
1
( i
1
t
p
1
X
T.
Since h i
t
= h
t
and p
X
i
t
= id
X
, this gives us a morphism
t
: h
1
t
( T
and a corresponding morphism
(h
t
;
t
) : (X; T) (Y ; ().
We call (h; ) a homotopy between
(h
0
;
0
) : (X; T) (Y, ()
and
(h
1
;
1
) : (X; T) (Y ; ().
Two morphisms of abelian sheafed spaces connected by a homotopy are said
to be homotopic.
Proposition 1.8.2. Let T, ( be abelian sheaves on X and Y . Assume the
morphisms
(f
0
;
0
) : (X; T) (Y ; ()
and
(f
1
;
1
) : (X; T) (Y ; ()
are homotopic. Then, the morphisms
(f
0
;
0
)
: H
(Y ; () H
(X; T)
and
(f
1
;
1
)
: H
(Y ; () H
(X; T)
are equal.
Proof. Since the application
p
X
: X I X
is proper and surjective and I is connected and acyclic (see Exercise 1.6.7),
Corollary 1.7.18 shows that the canonical morphism
(p
X
;
X
) : (X I; p
1
X
T) (X; T)
induces the isomorphism
(p
X
;
X
)
: H
(X; T)

H
(X I; p
1
X
T).
If
(i
t
;
t
) : (X; T) (X I; p
1
X
T)
denotes the canonical morphism, we have
(p
X
;
X
) (i
t
;
t
) = id.
It follows that (i
t
;
t
)
is the inverse of the isomorphism (p

X
;
X
)
and thus
does not depend on t [0, 1].
Let
(h; ) : (X I; p
1
X
T) (Y ; ()
be a homotopy between (f
0
;
0
) and (f
1
;
1
). Since
(h; ) (i
0
;
0
) = (f
0
;
0
), (h; ) (i
1
;
1
) = (f
1
;
1
),
we see that
(f
0
;
0
)
= (i
0
;
0
)
(h; )
= (i
1
;
1
)
(h; )
= (f
1
;
1
)
.
Corollary 1.8.3. If two morphisms
f
0
: X Y and f
1
: X Y
are homotopic in the topological sense, then, for any abelian group M,
f
0
: H
(Y ; M
Y
) H
(X; M
X
)
and
f
1
: H
(Y ; M
Y
) H
(X; M
X
)
are equal. In particular, if X and Y are homotopically equivalent, then
H
(X; M
X
) H
(Y ; M
Y
)
Examples 1.8.4.
(a) The preceding corollary allows us to show that
H
k
(B
n
; M
B
n
) =
_
M if k = 0
0 otherwise
for any abelian group M. As a matter of fact,
h : B
n
I B
n
(x, t) tx
is a homotopy between the constant map
h
0
: B
n
B
n
x 0
and the identity map
h
1
: B
n
B
n
x x
Therefore, the inclusion map j
0
: 0 B
n
and the projection map
p
0
: B
n
0 are inverse of each other in the category of topological
spaces modulo homotopy. Hence,
H
(B
n
; M
B
n
) H
(0; M
{0}
)
and the conclusion follows.
Note that in contrast with Exercise 1.4.10, we have not made use of
de Rham theorem. A similar reasoning shows that
H
k
(B
n
; M
B
n
)
_
M if k = 0
0 otherwise
(b) Working as in Exercise 1.6.6 (a) and (b), we can deduce from (a) that,
for n 1 and any abelian group M, we have
H
k
(S
n
; M
S
n
)
_
M if k = 0, n
0 otherwise
and
H
k
c
(B
n
; M
B
n
)
_
M if k = n
0 otherwise
Exercise 1.8.5. Assume n 1. Endow B
n+1
with the orientation induced
by the canonical orientation of R
n+1
and orient S
n
as B
n+1
. Denote
_
the
morphism obtained by composing the morphism
H
n
(S
n
; Z) H
n
(S
n
; C)
induced by the inclusion of Z in C with the integral
_
S
n
: H
n
(S
n
; C) C
of Exercise 1.7.20. Show that the group
H
n
(S
n
; Z)
has a unique generator v
S
n
such that
_
v
S
n
= 1 and that
_
induces an
isomorphism between H
n
(S
n
; Z) and Z.
Solution. Since H
n
(S
n
; Z) Z, this group has only two generators g
1
and
g
2
= g
1
. The uniqueness is thus obvious. Let us prove the existence. We
shall use the same notations as in Exercise 1.6.6. It is clear that we have
the following canonical morphism of distinguished triangles
R
c
(S
n
; Z)

R
c
(S
+
n
; Z) R
c
(S
n
; Z)

R
c
(S
n1
; Z)
+1
R
c
(S
+
n
S
n1
; Z)

R
c
(S
+
n
; Z)

(
id
0
)
R
c
(S
n1
; Z)
+1
id
where the rst vertical arrow is induced by zero extension of sections. It

follows that we have the commutative diagram
H
n1
(S
n1
; Z)

n1
H
n
(S
n
; Z)
H
n1
(S
n1
; Z)
n1
H
n
c
(S
+
n
S
n1
; Z)
where
k
and
k
are the Mayer-Vietoris and excision coboundary opera-
tors. Since S
+
n
is an oriented manifold with boundary, we get from Exer-
cise 1.7.20 that
_
S
n
n1
(c
n1
) =
_
S
+
n
\S
n1
n1
(c
n1
) =
_
S
n1
c
n1
where the sign + appears if S
n1
is oriented as the boundary of S
+
n
and
the sign appears in the other case (a simple computation shows that the
sign is in fact (1)
n
).
For n = 1, we have the commutative diagram
H
0
(S
+
1
; Z) H
0
(S
1
; Z)

O
H
0
(S
0
; Z)

0
H
1
(S
1
; Z)
O
Z
2
1 1
1 1

Z
2
( 1 1 )
Z
Therefore, the generators of H
1
(S
1
; Z) are the images by
0
of the functions
1
and
2
dened on S
0
by setting
_
1
(1) = 0
1
(1) = 1
and
_
2
(1) = 1
2
(1) = 0
Since
_
S
1
2
= (
2
(1)
2
(1)) = 1
we may take v
S
1
=
2
.
Assume now that n > 1 and that we have found v
S
n1
H
n1
(S
n1
; Z)
such that
_
S
n1
v
S
n1
= 1.
Since
_
S
n
n1
v
S
n1
= (1)
n
_
S
n1
v
S
n1
= (1)
n
we may choose v
S
n
= (1)
n
n1
v
S
n1
. The conclusion follows easily by
induction.
Exercise 1.8.6. Assume n 1. Show that if C
n
is an open cell of R
n
(i.e.
an open subset of R
n
which is homeomorphic to B
n
) then
H
k
c
(C
n
; Z) 0
for k ,= n and
_
induces an isomorphism
H
n
c
(C
n
; Z) Z.
In particular, there is a unique class v
C
n
H
n
c
(C
n
; Z) such that
_
v
C
n
= 1.
Show also that if C
n
n
included in C
n
, then the canonical
morphism
H
n
c
(C
n
; Z) H
n
c
(C
n
; Z)
is an isomorphism which sends v
C
n
to v
C
n
.
Solution. Let us rst assume that C
n
= B
n
. We start from the excision
distinguished triangle
R
c
(B
n
; Z) R(B
n
; Z) R(S
n1
; Z)
+1
For n = 1, we get the exact sequence

H
0
(B
n
; Z)

O
H
1
(S
0
; Z)

0
H
1
c
(B
1
; Z)

O
0
Z
(
1
1
)
Z
2
( 1 1 )
Z

0
It follows that a generator of H
1
c
(B
1
; Z) is v
B
1
=
0
([]) where
_
(1) = 0
(1) = 1
Since
_
B
1
v
B
1
=
_
B
1
0
([]) = (1) (1) = 1,
we see that
_
B
1
induces the isomorphism
_
B
1
: H
1
c
(B
1
; Z)

Z.
For n 2, we get the isomorphism
H
n1
(S
n1
; Z)
/o
n1
H
n
c
(B
n
; Z)
O
Z
1
Z
Setting v
B
n
=
n1
v
S
n1
, we see that
_
B
n
v
B
n
=
_
B
n
n1
v
S
n1
=
_
S
n1
v
s
n1
= 1
Assume now C
n
is a general open cell of R
n
. By functoriality, it is clear
that
H
k
c
(C
n
; Z) H
k
c
(B
n
; Z)
_
Z if k = n
0 otherwise
Let B be an open ball of R
n
with center x. By inversion, R
n
B B x.
Therefore,
R
c
(R
n
B; Z) R
c
(B x; Z)
From the excision distinguished triangle
R
c
(B x; Z) R(B; Z) R(x; Z)
+1
it follows that R
c
(B x; Z) 0. Hence, R
c
(R
n
B; Z) 0 and
R
c
(B; Z) R
c
(R
n
; Z)
is an isomorphism in the derived category.
It follows that
H
n
c
(B; Z) H
n
c
(R
n
; Z)
is an isomorphism. If we assume moreover that x C
n
and B C
n
, we see
from the commutative diagram
H
n
c
(B; Z)

H
n
c
(C
n
; Z)
.
.
.
.
.
.
H
n
c
(R
n
; Z)
that
H
n
c
(C
n
; Z) H
n
c
(R
n
; Z)
is surjective. Since these groups are both isomorphic to Z, the preceding
morphism is in fact an isomorphism. Using the commutativity of the dia-
gram
H
n
c
(B
n
; Z)
/o

B
n

H
n
c
(R
n
; Z)
R
n
C
we see that
_
R
n
_
R
n
: H
n
c
(R
n
; Z) Z.
A similar argument with B
n
replaced by C
n
allows us to conclude.
Exercise 1.8.7. Let : C
n
C
n
be a dieomorphism between two open
cells of R
n
. Show that
v
C
n
= sgn(J
)v
C
n
where J
denotes the Jacobian of .

Solution. It follows from the preceding exercise that
v
C
n
= mv
C
n
.
On one hand, we have
_
mv
C
n
= m
_
v
C
n
= m.
On the other hand, using de Rham cohomology, we get
_

v
C
n
= sgn(J
)
_
v
C
n
= sgn(J
).
The conclusion follows.
Exercise 1.8.8.
(a) Show that the canonical morphism
R
{0}
(B
n
; Z) R
c
(B
n
; Z)
is an isomorphism.
(b) Deduce from (a) that for any open neighborhood U of u in R
n
we have
H
k
{u}
(U; Z) =
_
Z if k = n;
0 otherwise.
Show, moreover, that H
n
{u}
(U; Z) has a unique generator v
u
such that
_
U
v
u
= 1
(c) Let : U V be a dieomorphism between open subsets of R
n
and
denote J
its Jacobian. Show that for any u U, the map
: H
n
{(u)}
(V ; Z) H
n
{u}
(U; Z)
sends v
(u)
to sgn(J
(u))v
u
.
Solution. (a) Consider the morphism of distinguished triangles
R
{0}
(B
n
; Z)

R(B
n
; Z)

R(B
n
0; Z)
+1
R
c
(B
n
; Z)

R(B
n
; Z)

R(S
n1
; Z)
+1
In this diagram, the second vertical arrow is the identity and by the homo-
topy theorem the third vertical arrow is an isomorphism. It follows that the
rst vertical arrow is also an isomorphism. Since this arrow is the compo-
sition of the isomorphism
R
{0}
(B
n
; Z) R
{0}
(B
n
; Z)
with the canonical morphism
R
{0}
(B
n
; Z) R
c
(B
n
; Z),
we get the conclusion.
(b) The cohomology table follows directly from (a) and Exercise 1.8.6.
Let B
U
U be an open ball with center u. By (a),
H
n
{u}
(U; Z) H
n
{u}
(B
U
; Z) H
n
c
(B
U
; Z).
Denote v
u
the element of H
n
{u}
(U; Z) corresponding to v
B
U
H
n
c
(B
U
; Z).
Since the diagram
H
n
{u}
(U; Z)

H
n
c
(U; Z)
H
n
{u}
(B
U
; Z)

H
n
c
(B
U
; Z)
is commutative, we get
_
U
v
u
=
_
B
U
v
B
U
= 1.
(c) Thanks to (b), the result may be obtained by working as in the
preceding exercise.
n+1
= (x
0
, , x
n
) B
n+1
: x
0
0 and identify
B
n
with
(0, x
1
, , x
n
) : (x
1
, , x
n
) B
n
.
(see Figure 1.8.1). Denote v
B
n+1
\B
n
H
n+1
c
(B
n+1
B
n
; Z) and v
B
n

H
n
c
(B
n
; Z) the classes which have an integral equal to 1. Show that
v
B
n+1
\B
n
=
n
(v
B
n
)
where
n
is the coboundary operator associated to the distinguished triangle
R
c
(B
n+1
B
n
; Z) R(B
n+1
; Z) R
c
(B
n
; Z)
+1
Solution. Thanks to the preceding exercise, this follows directly from Exer-
cise 1.7.20.
Figure 1.8.1:
x
0
(x
1
, , x
n
)
B
n
B
n+1
1.9 Cohomology of compact polyhedra
Denition 1.9.1. A nite simplicial complex is a nite set of non-empty
nite sets called simplexes such that if and
then
. We
call the nite set S = , the set of vertices of .
The dimension of a simplex is the number dim = # 1.
Simplexes of dimension n N are called n-simplexes. A p-face of an n-
simplex is a p-simplex
such that
. The dimension of the nite

simplicial complex is the number
dim = sup
dim.
The n-skeleton of is the simplicial complex
n
= : dim n.
The realization of is the compact subspace [[ of R
#S
dened by
setting
[[ = : S R : [ 0, supp ,
sS
(s) = 1.
If , we set
[[ = [[ : supp = .
Clearly, [[ [
[ ,= if and only if =
and [[ =
[[.
The data of a nite simplicial complex and an isomorphism
h : [[ X
of topological spaces is a nite triangulation of X. A topological space X
which has a nite triangulation is a compact polyhedron.
1.9. Cohomology of compact polyhedra 41
Examples 1.9.2. The set whose elements are
A, B, C, D, A, B, A, D, B, D, B, C, D, C, A, B, D
is a nite simplicial complex of dimension 2. It has A, B, C, D as set of
vertices and contains
- four 0-simplexes (A, B, C, D),
- ve 1-simplexes (A, B, A, D, B, D, B, C, D, C),
- one 2-simplex (A, B, D).
A compact polyhedron homeomorphic to [[ is
A
B
C
D
Compact polyhedra homeomorphic to [
1
[ and [
0
[ are respectively
A
B
C
D
and
A
B
C
D
Denition 1.9.3. Let be a k-simplex of . Two bijections
: 0, , k , : 0, , k
have the same sign if the signature of
1
is positive. Clearly, the
relation to have the same sign is an equivalence relation on the set of
bijections between 0, , k and . An equivalence class for this relation
is called an orientation of . Of course, a k-simplex of has only two
possible orientations, if o is one of them, we denote o the other. An
oriented k-simplex is a k-simplex of endowed with an orientation. If
: 0, , k is a bijection, we denote by [(0), , (k)] the oriented
k-simplex obtained by endowing with the orientation associated to .
A k-cochain of is a map c from the set of oriented k-simplexes of to
Z such that if c
(,o)
denotes the values of c on the oriented simplex (, o),
we have c
(,o)
= c
(,o)
. It is clear that k-cochains form a group. We
denote it by C
k
(). We dene the dierential
d
k
: C
k
() C
k+1
()
by setting
(d
k
c)
[x
0
, ,x
k+1
]
=
k+1
l=0
(1)
l
c
[x
0
, , x
l
, ,x
k+1
]
.
One checks easily that the groups C
k
() (k 0) together with the dier-
entials d
k
(k 0) form a complex C
() canonically associated with . We

call it the simplicial cochain complex of .
Lemma 1.9.4. Let = x
0
, , x
k
be a k-simplex of and let
l
be the
point of [[ corresponding to x
l
. Set
(x
0
, ,x
k
)
(t
1
, , t
k
) =
0
+
k
l=1
t
l
(
l
0
)
and
J
k
= (t
1
, , t
k
) R
k
: t
1
> 0, , t
k
> 0,
k
l=1
t
l
< 1.
Then, J
k
k
and
(x
0
, ,x
k
)
: J
k
[[ is an homeo-
morphism. Moreover, if v
(x
0
, ,x
k
)
is the image of v
J
k
by the isomorphism
1
(x
0
, ,x
k
)
: H
k
c
(J
k
; Z)

H
k
c
([[; Z), then we have
v
(x
0
, ,x
k
)
= (sign)v
(x
0
, ,x
k
)
for any permutation of 0, , k.
Proof. The fact that is an homeomorphism is obvious. Let be a per-
mutation of 0, , k. Set
=
1
(x
0
, ,x
k
)

(x
0
, ,x
k
)
.
Clearly,
is the restriction to J
k
of the anity of R
k
characterized by
(P
l
) = P
l
where P
0
= 0, P
1
= e
1
, , P
k
= e
k
. It follows that
preserves or
reverses the orientation of R
k
according to the fact that is even or odd.
Using Exercise 1.8.7, we see that
(v
J
k
) = (sign)v
J
k
.
Hence,
1
(x
0
, ,x
k
)
(v
J
k
)
=
1
(x
0
, ,x
k
)

(v
J
k
)
= (sign)
1
(x
0
, ,x
k
)
(v
J
k
)
Lemma 1.9.5. Let = x
0
, , x
k
be a k-simplex of and let
=
x
1
, , x
k
. The distinguished triangle
R
c
([[; Z) R
c
([[ [
[; Z) R
c
([
[; Z)
+1
induces a canonical morphism
k1
: H
k1
c
([
[; Z) H
k
c
([[; Z)
and we have
k1
(v
[x
1
, ,x
k
]
) = v
[x
0
, ,x
k
]
.
Proof. The existence of the distinguished triangle follows from the fact that
[[ (resp. [
[) is open (resp. closed) in [[ [
[. Thanks to the morphism
(x
0
, ,x
k
)
of the preceding lemma, we may assume that [[ = J
k
, [
[ =
(t
1
, , t
k
) : t
1
> 0, , t
k
> 0,
k
l=1
t
k
= 1. Then, [[ [
[ appears
as a manifold with boundary and the result follows from Exercises 1.8.9
and 1.7.20 since
1 1 1 1
1 1 0 0
1 0 1 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 0 0 1
> 0.
Proposition 1.9.6. For any k 0, there is a canonical isomorphism be-
tween H
k
([[; Z) and H
k
(C
()).
Proof. Let us consider the excision distinguished triangle
R
c
([
p
[ [
p1
[; Z) R
c
([
p
[; Z) R
c
([
p1
[; Z)
+1
(*)
Since [
p
[ [
p1
[ =

p
\
p1
[[ and [[ is open in [
p
[ for any
p

p1
, we get
R
c
([
p
[ [
p1
[; Z)
p
\
p1
R
c
([[; Z).
Using the fact that [[ is homeomorphic to an open ball of R
p
if
p

p1
, we see that
H
k
c
([
p
[ [
p1
[; Z) =
_
p
\
p1
H
p
c
([[; Z) if k = p
0 otherwise.
Using Lemma 1.9.4, we get a canonical isomorphism
H
p
c
([
p
[ [
p1
[; Z) C
p
().
The long exact sequence of cohomology associated to (*) is
H
k
c
([
p
[ [
p1
[; Z)

H
k
([
p
[; Z)

H
k
([
p1
[; Z)
-, *+
/.
()

H
k+1
c
([
p
[ [
p1
[; Z)

H
k+1
([
p
[; Z)

H
k+1
([
p1
[; Z)
For k > p, we get that
H
k
([
p
[; Z) H
k
([
p1
[; Z).
By decreasing induction on p, we see that H
k
([
p
[; Z) H
k
([
0
[; Z) 0.
For k < p 1, we obtain
H
k
([
p
[; Z) H
k
([
p1
[; Z).
By increasing induction on p, this gives us the isomorphism
H
k
([
p1
[; Z) H
k
([[; Z).
For k = p 1, we get the exact sequence
0

H
p1
([
p
[; Z)

H
p1
([
p1
[; Z)
-, *+
/.
()

H
p
c
([
p
[ [
p1
[; Z)

H
p
([
p
[; Z)

H
p
([
p1
[; Z)
Using the isomorphisms obtained above, we may rewrite this sequence as
0 H
p1
(||; Z)

p1
H
p1
(|
p1
|; Z)

p1
C
p
(||)

p
H
p
(|
p
|; Z) 0.
Set
p
=
p

p
. Clearly, Coker
p1
Coker
p1
H
p
([
p
[; Z) and
through this isomorphism, the canonical map
(
p
)
: Coker
p1
C
p+1
([[)
becomes the map
p
: H
p
([
p
[; Z) C
p+1
([[).
It follows that Ker
p
/ Im
p1
is canonically isomorphic to
Ker
p
H
p
([[; Z).
To conclude, it remains to show that
p
= d
p
. Let [x
0
, , x
p+1
] be an
oriented (p + 1)-simplex of and let c C
p
() be a simplicial p-cochain.
We have to show that
[
p
(c)]
[x
0
, ,x
p+1
]
=
p+1
l=0
(1)
l
c
[x
0
, , x
l
, ,x
p+1
]
.
Denote the simplicial complex formed by the nite non-empty subsets of
x
0
, , x
p+1
.
Using the morphisms of distinguished triangles
R
c
([
k
[ [
k1
[; Z)

R
c
([
k
[; Z)

R
c
([
k1
[; Z)
+1
R
c
([
k
[ [
k1
[; Z)

R
c
([
k
[; Z)

R
c
([
k1
[; Z)
+1
One sees that we may assume = . Moreover, we may assume that

c
[x
0
, , x
l
, ,x
p+1
]
=
_
1 if l = 0
0 otherwise
Set = x
0
, , x
p+1
and
= x
1
, , x
p+1
. Since [[ [
[ is open
in [[ and [
[ is open in [
p1
[, we have the morphism of distinguished
triangles
R
c
([[; Z)

R
c
([[ [
[; Z)

R
c
([
[; Z)
+1
R
c
([[ [
p
[; Z)

R
c
([[; Z)

R
c
([
p
[; Z)
+1
Therefore, the diagram

H
p
c
([
[; Z)

H
p+1
c
([[; Z)
H
p
c
([
p
[ [
p1
[; Z)
H
p
([
p
[; Z)

H
p+1
c
([[ [
p
[; Z)
is commutative. Since c is the image of the orientation class v
[x
1
, ,x
p
]

H
p
c
([
[; Z) by the canonical morphism

H
p
c
([
[; Z) H
p
c
([
p
[ [
p1
[; Z) C
p
(),
p
(c) is the image of (v
[x
1
, ,x
p
]
) by the canonical morphism
H
p+1
c
([[; Z) H
p+1
c
([[ [
p
[; Z) C
p+1
().
But Lemma 1.9.5 shows that
(v
[x
1
, ,x
p
]
) v
[x
0
, ,x
p
]
Corollary 1.9.7 (Eulers theorem). Let be a nite simplicial complex.
Then, the abelian groups
H
k
([[; Z) k 0
are nitely generated and the Euler-Poincare characteristic
([[) =
k
(1)
k
rk H
k
([[; Z)
is equal to
k
(1)
k
#(
k

k1
).
Proof. The rst part follows directly from the fact that
C
k
() Z
#(
k
\
k1
)
is a free abelian group with nite rank.
Denote Z
k
() and B
k+1
() the kernel and the image of the dierential
d
k
: C
k
() C
k+1
()
and set
H
k
() = Z
k
()/B
k
().
From the exact sequences
0 Z
k
() C
k
() B
k+1
() 0
0 B
k
() Z
k
() H
k
() 0
we deduce that
rk C
k
() = rk Z
k
() +rk B
k+1
()
and that
rk Z
k
() = rk B
k
() +rk H
k
().
It follows that
rk C
k
() = rk B
k
() + rk B
k+1
() + rk H
k
()
and, hence, that
k
(1)
k
rk C
k
() =
k
(1)
k
rk H
k
().
The conclusion follows since
rk C
k
() = #(
k

k1
)
and
H
k
() H
k
([[; Z).
Examples 1.9.8.
(a) Consider a simplicial complex with [[ homeomorphic to
Clearly, has four 0-simplexes, ve 1-simplexes and one 2-simplex. Hence,
([[; Z) = 4 5 +1 = 0.
(b) Consider a simplicial complex with [[ homeomorphic to
We have three 0-simplexes and three 1-simplexes. Therefore, ([[; Z) =
3 3 = 0. Note that [[ S
1
and our result is compatible with the fact
that
H
k
(S
1
; Z)
_
Z if k = 0, 1
0 otherwise
since dimH
0
(S
1
; Z) dimH
1
(S
1
; Z) = 1 1 = 0.
(c) Similarly, for
we have = 4 6 +4 = 2. This is compatible with the relation
dimH
0
(S
2
; Z) dimH
1
(S
2
; Z) + dimH
2
(S
2
; Z) = 1 0 + 1 = 2.
1.10 Cohomology of locally compact spaces
Denition 1.10.1. Let X be a locally compact topological space. We call
the cohomological dimension of the functor
T
c
(X; T)
the cohomological dimension of X and denote it by dim
c
X.
Remark 1.10.2. Note that by a well-known result about cohomological
dimensions, dim
c
X n (n N) if and only if
H
k
c
(X; T) 0
for any k > n and any T ohv(X). Note also that contrarily to what
may appear at rst glance, cohomological dimension is a local notion. More
precisely, if | is an open covering of X, we have
dim
c
X = sup
UU
dim
c
U.
In particular, dim
c
U dim
c
X for any open subspace U of X. Note that
although a similar majoration holds for closed subspaces, it may be false for
arbitrary subspaces.
Exercise 1.10.3. Show that the cohomological dimension of an open sub-
space of R
n
is equal to n. Deduce from this fact that if U and V are
homeomorphic open subspaces of R
n
and R
m
then n = m.
Solution. Let us prove that dim
c
R
n
n. The conclusion will follow since
we know that
H
n
c
(B; Z) = Z
for any open ball B of R
n
.
Since
1
2
+
1
arctg x : R ]0, 1[
1.10. Cohomology of locally compact spaces 49
is a homeomorphism, R
n
is homeomorphic to ]0, 1[
n
. Since this last space
is an open subspace of [0, 1]
n
, it is sucient to show that dim
c
[0, 1]
n
n.
We will proceed by induction on n.
For n = 1, this follows from Exercise 1.6.7. Assuming the result is true
for n, we prove it for n + 1 by using the isomorphism
R([0, 1]
n+1
; T) R([0, 1]; p(T))
(where p : R
n+1
R
n
is a canonical projection) and the bers formula for
Rp.
The last part follows from the fact that cohomological dimension is
clearly invariant by homeomorphism.
Remark 1.10.4. It follows at once from the preceding exercise that the
cohomological dimension of a dierential manifold is equal to its usual di-
mension.
Exercise 1.10.5. Let be a nite simplicial complex. Show that the
cohomological dimension of [[ is equal to dim.
Solution. We will proceed by induction on dim. For dim = 0, the result
is obvious. To prove that the result is true for dim = n+1 if it is true for
dim n, it is sucient to use the excision distinguished triangle
R
c
([
n+1
[ [
n
[; T) R
c
([
n+1
[; T) R
c
([
n
[; T)
+1
together with the fact that [

n+1
[ [
n
[ is a nite union of open cells of
dimension n +1.
Denition 1.10.6. The reduced cohomology

H
(X; Z) of X with coecient

in Z is dened by setting
H
k
(X; Z) =
_
H
0
(X; Z)/Z if k = 0,
H
k
(X; Z) otherwise.
A topological space X is cohomologically locally connected (clc for short) if,
for any x X and any neighborhood U of x, there is a neighborhood V of
x included in U such that all the morphisms
H
k
(U; Z)

H
k
(V ; Z)
are zero.
Examples 1.10.7. One checks directly that dierential manifolds are clc
spaces. With a little more work, one sees also that the same is true of
compact polyhedra.
Proposition 1.10.8 (Borel-Wilder). Assume X is a locally compact clc-
space. Then, for any pair K, L of compact subsets of X such that L K
,
all the restriction morphisms
H
k
(K; Z) H
k
(L; Z)
have nitely generated images.
Proof. We will proceed by increasing induction on k. Denote L the family
of compact subsets L of K
which have a compact neighborhood L
for which the image

r
k
L
K
: H
k
(K; Z) H
k
(L
; Z)
is nitely generated. It is clear that any point x K
has neighborhood
in L. Hence, it is sucient to show that if L
1
, L
2
L then L
1
L
2
L.
Choose compact neighborhoods L
1
, L
2
of L
1
, L
2
for which r
k
L
1
K
and r
k
L
2
K
have nitely generated images. Let L
1
, L
2
be compact neighborhoods of
L
1
, L
2
such that L
1
L
, L
2
L
. Consider the diagram

H
k
(K; Z)

H
k
(K; Z) H
k
(K; Z)
H
k1
(L
1
L
2
; Z)

H
k
(L
1
L
2
; Z)
H
k
(L
1
; Z) H
k
(L
2
; Z)
H
k1
(L
1
L
2
; Z)

H
k
(L
1
L
2
; Z)
where the horizontal morphisms come from Mayer-Vietoris sequences and
the vertical ones are restriction maps. We know that Im is nitely gener-
ated. Since
(Im) Im
we see that
(Im) is also nitely generated. Hence, so is Im/(Im

Im
). Using the epimorphism
(Im/(Im Im
)) Im(
)/ Im(
)
we see that Im(
)/ Im(
) is nitely generated. Since the induction

hypothesis shows that Im
is nitely generated, it follows that Im(
)
is nitely generated. This shows that L
1
L
2
L and the conclusion
follows.
Corollary 1.10.9. Assume X is a compact clc space. Then, the abelian
groups
H
k
(X; Z)
1.11. Poincare-Verdier duality 51
are nitely generated. If, moreover, X has nite cohomological dimension,
then the Euler-Poincare characteristic
(X) =
kZ
rk H
k
(X; Z)
is well-dened.
Remark 1.10.10. It follows from the preceding corollary that a compact
dierential manifold has a well-dened Euler-Poincare characteristic. We
will study it with more details in Chapter 2.
1.11 Poincare-Verdier duality
Let f : X Y be a continuous map between locally compact spaces.
Denition 1.11.1. A closed subset F of X is f-proper if the map
f
|F
: F Y
is proper or in other words if F f
1
(K) is compact for every compact
subset K of Y . Clearly, f-proper subsets of X form a family of supports.
Let T be a sheaf on X and let U be an open subset of Y . We set
f
!
(T)(U) =
fproper
(f
1
(U); T).
One checks easily that f
!
(T) is a sheaf on Y . We call it the direct image
with proper supports of T by f.
f
!
: ohv(X) ohv(Y )
Rf
!
:
T
+
(ohv(X))
T
+
(ohv(Y ))
which is computable by means of c-soft resolutions.
Remark 1.11.3. For the canonical map
a
X
: X pt
we see easily that
Ra
X!
(T) R
c
(X; T).
Proposition 1.11.4 (Fibers formula). For any T
T
+
(ohv(X)), we
have a canonical isomorphism
[Rf
!
(T)]
y
R
c
(f
1
(y); T).
Corollary 1.11.5. The cohomological dimension of the functor f
!
is equal
to
sup
yY
[dim
c
f
1
(y)]
Corollary 1.11.6 (Cartesian square formula). Assume
Y
f
X
T
g
Z
g
is a cartesian square of locally compact spaces. Then, we have the canonical

isomorphisms
g
1
f
!
f
!
g
1
and g
1
Rf
!
Rf
!
g
1
Proposition 1.11.7. Let g : Y Z be another continuous map of locally
compact spaces. Then, there are canonical isomorphisms
(g f)
!
g
!
f
!
and R(g f)
!
Rg
!
Rf
!
.
Remark 1.11.8. Combining the preceding result with Remark 1.11.3, we
see that
R
c
(Y ; Rf
!
T) R
c
(X; T).
A result which may be seen as a kind of Leray theorem with compact sup-
ports.
Theorem 1.11.9 (Poincare-Verdier duality). Assume f
!
has nite co-
homological dimension (i.e. assume that there is n 0 such that
H
k
(Rf
!
T) = 0
for k > n and any T ohv(X)). Then,
Rf
!
:
T
+
(ohv(X))
T
+
(ohv(Y ))
has a right adjoint
f
!
:
T
+
(ohv(Y ))
T
+
(ohv(X)).
In other words, there is a canonical functorial isomorphism
Hom
D
+
(Shv(Y ))
(Rf
!
T, () Hom
D
+
(Shv(X))
(T, f
!
().
Moreover, for T
T
(ohv(X)) and (
T
+
(ohv(Y )), there is a functorial
isomorphism
RHom(Rf
!
T, () RHom(T, f
!
().
Example 1.11.10. Let F be a closed subspace of X. Denote
F
( the
abelian sheaf
U
FU
(U; ().
Let i : F X be the canonical inclusion of F in X. Then, one checks easily
that i
!
is exact and that
RHom(i
!
T, () RHom(T, i
1
R
F
().
In particular, there is a canonical functorial isomorphism
i
!
( i
1
R
F
(.
Corollary 1.11.11 (Absolute Poincare duality). Assume X is a nite
dimensional locally compact space. Then, we have the canonical isomor-
phism
RHom(R
c
(X; Z
X
), Z) R(X;
X
)
where
X
= a
!
X
Z is the dualizing complex of X.
Proof. Take f = a
X
, T = Z
X
, ( = Z in Theorem 1.11.9 and use the
isomorphism
RHom(Z
X
,
X
) R(X;
X
).
Proposition 1.11.12. Let g : Y Z be another continuous map of locally
compact spaces. Then, the canonical isomorphism
R(g f)
!
Rg
!
Rf
!
induces by adjunction the canonical isomorphism
(g f)
!
f
!
g
!
.
Corollary 1.11.13 (Alexander duality). Assume X has nite cohomo-
logical dimension and let F be a closed subset of X. Then, there is a
RHom(R
c
(F; Z), Z) R
F
(X;
X
).
Proof. It is clear that F is locally compact and has nite homological di-
mension. Moreover, thanks to Example 1.11.10 and Proposition 1.11.12, we
have
F
= a
!
F
Z i
!
a
!
X
Z i
1
R
F
X
where i : F X is the canonical inclusion. Hence, the conclusion follows
from Corollary 1.11.11 with X replaced by F.
Proposition 1.11.14. For R
n
there is a canonical isomorphism
R
n Z
R
n[n].
Proof. Using Corollary 1.11.11, we see that
(
R
n)
x
lim
Ux
U open ball
RHom(R
c
(U; Z), Z).
Thanks to Examples 1.8.4, we know that
R
c
(U; Z) Z[n].
Hence,
(
R
n)
x
Z[n].
This shows in particular that
H
k
(
R
n) 0
for k ,= n and that the sheaf
or
R
n = H
n
(
R
n)
has all its bers canonically isomorphic to Z. Moreover, it follows from
Corollary 1.11.11 that for any open ball U of R
n
,
R(U; or
R
n) RHom(R
c
(U; Z
U
)[n], Z) Z.
Using the fact (see Exercise 1.8.6) that the canonical diagram
R
c
(U; Z)[n]
/o

Z
R
c
(U
; Z)[n]
/o

Z
is commutative if U U
are open balls of R

n
, we see that or
R
n is canoni-
cally isomorphic to Z
R
n.
Corollary 1.11.15. Let U (resp. F) be an open (resp. a closed) subset of
R
n
. Then, we have the canonical isomorphisms
RHom(R
c
(U; Z), Z) R(U; Z)[n]
and
RHom(R
c
(F; Z), Z) R
F
(R
n
; Z)[n].
In particular, there are exact sequences of the form
0 Ext
1
(H
k+1
c
(U; Z), Z) H
nk
(U; Z) Hom(H
k
c
(U; Z), Z) 0
and
0 Ext
1
(H
k+1
c
(F; Z), Z) H
nk
F
(R
n
; Z) Hom(H
k
c
(F; Z), Z) 0.
Proof. The rst part follows from Corollary 1.11.11 and Corollary 1.11.13
combined with the preceding proposition.
As for the second part, it follows from the fact that for any complex C
of abelian groups we have an exact sequence of the form

0 Ext
1
(H
k+1
(C
), Z) H
k
(RHom(C
, Z)) Hom(H
k
(C
), Z) 0
for any k Z.
Exercise 1.11.16. Let f : S
1
R
2
be a continuous map and set C =
f(S
1
). We dene the multiplicity of x C to be
x
= #f
1
(x). A point
x C is simple if
x
= 1 and multiple otherwise. Assume C has a nite
number of multiple points each of which has nite multiplicity. Compute
the cohomology of C and R
2
C. In particular, show that the number of
bounded connected components of R
2
C is 1 +
xC
(
x
1) generalizing
in this way the well-known theorem on Jordan curves. For example in
x
1
x
2
x
3
x
4
I
II
III
IV
V
VI
we have
x
1
=
x
2
=
x
3
= 2,
x
4
= 3
and
1 +
xC
(
x
1) = 1 +1 + 1 +1 + 2 = 6.
Solution. Let g : S
1
C be the map induced by f. We have
Rg(Z
S
1
)
x
R(g
1
(x); Z) Z
x
(*)
for any x C. In particular, Rg(Z
S
1
) g(Z
S
1
). Consider the canonical
morphism Z
C
g(Z
S
1
). Since g is surjective, it is a monomorphism. More-
over, (*) shows that its cokernel is supported by the set of multiple points
of C. Therefore, we get the exact sequence of sheaves
0 Z
C
g(Z
S
1
)
xC
x
>1
Z
x
1
{x}
0
where Z
{x}
is the direct image of the constant sheaf on x in C. Applying
R(C; ) we get the distinguished triangle
0 R(C; Z
C
) R(C; Rg(Z
S
1
)) R(C;

xC
x
>1
Z
x
1
{x}
)
+1
Moreover,
R(C; Rg(Z
S
1
)) R(S
1
; Z
S
1
) and R(C;

xC
x
>1
Z
x
1
{x}
)
xC
x
>1
Z
x
1
.
Taking cohomology and setting
=

xC
x
>1
(
x
1),
we get the exact sequence
0 H
0
(C; Z
C
) H
0
(S
1
; Z
S
1
) Z
H
1
(C; Z
C
) H
1
(S
1
; Z
S
1
) 0
and the isomorphisms
H
k
(C; Z
C
) H
k
(S
1
; Z
S
1
) 0
for k > 1. Since S
1
is connected, so is C and H
0
(C; Z
C
) H
0
(S
1
; Z
S
1
) Z.
Through these isomorphisms, the rst morphism appears as id : Z Z.
Hence, using the fact that H
1
(S
1
; Z
S
1
) Z, we get the exact sequence
0 Z
H
1
(C; Z
C
) Z 0.
The last group being projective, we get
H
1
(C; Z
C
) Z
+1
.
The cohomology table for C is thus
H
k
(C; Z
C
)
_
_
Z if k = 0,
Z
1+
if k = 1,
0 otherwise.
Since all these groups are free, Corollary 1.11.15 shows that
H
2k
C
(R
2
; Z
R
2) Hom(H
k
(C; Z), Z).
Therefore,
H
k
C
(R
2
; Z
R
2)
_
_
Z
1+
if k = 1,
Z if k = 2,
0 otherwise.
Consider the excision distinguished triangle
R
C
(R
2
; Z) R(R
2
; Z) R(R
2
C; Z)
+1
Taking cohomology and using the fact that H

k
(R
2
; Z) 0 for k > 0, we get
the exact sequence
0 H
0
C
(R
2
; Z) H
0
(R
2
; Z) H
0
(R
2
C; Z) H
1
C
(R
2
; Z) 0
and the isomorphisms
H
k
(R
2
C; Z) H
k+1
C
(R
2
; Z)
for k > 0. Since R
2
is connected, a locally constant function in R
2
is
constant and we have H
0
(R
2
; Z) Z and H
0
C
(R
2
; Z) 0. It follows that
the sequence
0 Z H
0
(R
2
C; Z) Z
1+
0
is exact. From these results we deduce that
H
k
(R
2
C; Z)
_
_
Z
2+
if k = 0,
Z if k = 1,
0 otherwise.
Hence the number of connected components of R
2
C is 2 + . Since C
is compact, R
2
C has exactly one non bounded connected component. It
follows that the number of bounded connected components of R
2
C is 1+.
Moreover, our method of proof shows also that
H
1
(R
2
C; Z) H
1
(R
2
B; Z)
is an isomorphism if B is a closed ball of R
2
containing C. It follows that,
for any bounded connected component U of R
2
C, we have
H
k
(U; Z)
_
Z if k = 0,
0 otherwise
and that
H
k
(U; Z)
_
Z if k = 0, 1,
0 otherwise
if U is the non bounded connected component of R
2
C.
1.12 Borel-Moore homology
From the point of view of sheaf theory, cohomology is more natural than
homology. However, to make the link with classical homology theories, it is
convenient to introduce the following kind of co-cohomology.
Denition 1.12.1. Let be a family of supports of X. We dene the
Borel-Moore homology H
k
(X; Z) of X with integer coecients and supports
in by setting
H
k
(X; Z) = H
k
R(X;
X
).
Remark 1.12.2. Note that thanks to Corollary 1.11.11, we have a canon-
ical epimorphism
H
k
(X; Z) Hom(H
k
c
(X; Z), Z)
which becomes an isomorphism when H
k+1
c
(X; Z) has no torsion.
Proposition 1.12.3. If X is a Homologically Locally Connected space in
the sense of singular homology (HLC for short), then there are canonical
isomorphisms
H
c
k
(X; Z) SH
k
(X; Z) (k N)
where SH
k
(X; Z) denotes the singular homology with integer coecients of
X.
Corollary 1.12.4. Assume X is a HLC space and K is a compact subset
of X. Then, there are canonical isomorphisms
H
k
R(K;
X
) SH
k
(X, X K; Z).
1.12. Borel-Moore homology 59
Proof. This follows from the preceding proposition combined with the dis-
tinguished triangle
R
c
(X K;
X
) R
c
(X;
X
) R(K;
X
)
+1
Proposition 1.12.5. Let f : X Y be a continuous map and assume

X, Y are locally compact spaces of nite cohomological dimension. Then,
there is a canonical morphism
Rf
!
X

Y
.
This morphism induces a morphism
f
: H
c
(X; Z) H
c
(Y ; Z)
which is compatible with the usual push-forward morphism
f
: SH
(X; Z) SH
(Y ; Z).
Proof. We have
f
!
Y
f
!
a
!
Y
Z a
!
X
Z
X
.
Thanks to Poincare-Verdier duality,
Hom
D
(Shv(Y ))
(Rf
!
X
,
Y
) Hom
D
(Shv(X))
(
X
, f
!
Y
).
Hence, to the isomorphism
X
f
!
Y
corresponds a canonical morphism
Rf
!
X

Y
.
Applying R
c
(Y ; ) to this morphism, we get a morphism
R
c
(X;
X
) R
c
(Y ;
Y
)
which induces f
at the level of cohomology. For the link with singular

homology, we refer to standard texts.
Remark 1.12.6. For f = a
X
, we see that
(a
X
)
: H
c
0
(X; Z) Z
corresponds to the classical augmentation
# : SH
0
(X; Z) Z.
which associates to a singular 0-cycle c =

J
j=1
m
j
[x
j
] the number #c =
J
j=1
m
j
. This is why we will often use # as a shorthand notation for
(a
X
)
.
1.13 Products in cohomology and homology
Denition 1.13.1. Let T, ( be two abelian sheaves on the topological
space X. We dene the tensor product of T and ( to be the abelian sheaf
associated to the presheaf
U T(U) ((U).
Let and be families of supports on X. Since F G for any
F and any G , we have a canonical morphism
:
(X; T)
(X; ()
(X; T ().
: ohv(X) ohv(X) ohv(X)
is right exact, left derivable and has nite homological dimension. Moreover,
if and are families of supports on X, we have a canonical morphism
: R
(X; T)
L
R
(X; () R
(X; T
L
()
which induces the generalized cup-products
: H
k
(X; T) H
l
(X; () H
k+l
(X; T
L
()
at the level of cohomology.
Remark 1.13.3. For T = ( = Z
X
, we have T
L
( Z
X
and one can
show that the morphisms
: H
k
(X; Z
X
) H
l
(X; Z
X
) H
k+l
(X; Z
X
)
given by the preceding proposition coincide with the classical cup-products.
On can also recover the usual formulas
(c
k
c
l
) c
m
= c
k
(c
l
c
m
) and c
k
c
l
= (1)
kl
c
l
c
k
.
For T = Z
X
and ( =
X
, we have T
L
( =
X
and we get the
morphisms
: H
k
(X; Z
X
) H
l
(X;
X
) H
kl
(X;
X
).
Using the equality H
m
(X;
X
) = H
m
(X; Z
X
), we recover the classical
cap-products
: H
k
(X; Z
X
) H
l
(X; Z
X
) H
lk
(X; Z
X
).
1.13. Products in cohomology and homology 61
Thanks to the associativity of the generalized cup-products, we also recover
the cup-cap associativity formula
(c
k
c
l
) c
m
= c
k
(c
l
c
m
).
If the elements of are compact, we also get a pairing
, ) : H
k
(X; Z) H
k
(X; Z) Z
by composing
: H
k
(X; Z) H
k
(X; Z) H
0
(X; Z)
with
# : H
c
0
(X; Z) Z.
This pairing is a natural generalization of the classical pairing between ho-
mology and cohomology. Thanks to the cup-cap associativity formula, we
have
c
k
c
l
, c
k+l
_
=
c
k
, c
l
c
k+l
_
.
Proposition 1.13.4 (Projection formula). Let f : X Y be a contin-
uous map between locally compact spaces. Then, for any T
T
+
(ohv(X))
and any (
T
+
(ohv(Y )), there is a canonical isomorphism
(
L
Rf
!
T

Rf
!
(f
1
(
L
T).
Corollary 1.13.5 (Universal coecient formula). Let X be a locally
compact space. Then, for any M
T
+
(/b), we have the canonical isomor-
phism
R
c
(X; M) M
L
R
c
(X; Z).
Remark 1.13.6. Thanks to the preceding proposition, one can prove that
if f : X Y is a morphism of locally compact spaces we have
f
(f
(c
k
) c
l
) = c
k
f
(c
l
)
for any c
k
H
k
(Y ; Z) and any c
l
H
c
l
(X; Z). In particular, we get
(c
k
), c
l
_
=
c
k
, f
(c
l
)
_
if k = l.
Denition 1.13.7. Let X, Y be topological spaces and let T (resp. () be
a sheaf on X (resp. Y ). Denote p
X
, p
Y
the canonical projections of X Y
on X and Y . We dene the exterior tensor product T ( of T and ( by
the formula
T ( = p
1
X
T p
1
Y
(.
Let and be families of supports on X and Y . Denote the family
of supports on X Y formed by the closed subsets of the products of the
form F G with F and G . By denition of T (, we get a
canonical morphism
:
(X; T)
(Y ; ()
(X Y ; T ()
Proposition 1.13.8. Assume X, Y are topological spaces. Then, the func-
tor
: ohv(X) ohv(Y ) ohv(X Y )
is right exact and left derivable. It has nite homological dimension and if
and are families of supports on X and Y , we have a canonical morphism
R
(X; T)
L
R
(Y ; () R
(X Y ; T
L
()
which induces the generalized cross-products
: H
k
(X; T) H
l
(Y ; () H
k+l
(X Y ; T ()
at the level of cohomology.
Remark 1.13.9. If T = Z
X
and ( = Z
Y
, we get T ( Z
XY
and the
generalized cross-products
: H
k
(X; Z) H
l
(Y ; Z) H
k+l
(X Y ; Z)
corresponds to the classical cross-products in cohomology. Note that this is
not really a new operation since
c
k
c
l
= p
X
(c
k
) p
Y
(c
l
).
So, most of the properties of cross-products in cohomology may be deduced
from this formula. In particular, if f : X X
and g : Y Y

are
continuous maps, we have
(f g)
(c
k
c
l
) = f
(c
k
) g
(c
l
).
Proposition 1.13.10 (K unneth theorem for cohomology).
Assume X, Y are locally compact spaces. Then, the canonical morphism
R
c
(X; T)
L
R
c
(Y ; () R
c
(X Y ; T
L
()
is an isomorphism for any T in
T
+
(ohv(X)) and any ( in
T
+
(ohv(Y )).
1.13. Products in cohomology and homology 63
Proof. This follows directly from the projection formula and the cartesian
square formula.
Lemma 1.13.11. Assume X and Y are locally compact spaces. Denote
p
Y
: X Y Y
the second projection. Then, there is a functorial canonical morphism
L
( p
!
Y
(
for (
T
+
(ohv(Y )). This morphism becomes an isomorphism if X is a clc
space.
Proof. We have the chain of morphisms
Rp
Y
!
(
X

L
() Rp
Y
!
(p
1
X

X

L
p
1
Y
() (1)
Rp
Y
!
(p
1
X

X
)
L
( (2)
a
1
Y
Ra
X!
X

L
( (3)
where (1) comes from the denition of , (2) follows from the projection
formula and (3) from the cartesian square formula. Using the canonical
morphism
Ra
X!
X
Z,
we get a canonical morphism
Rp
Y
!
(
X

L
() (.
By adjunction, this gives us the requested morphism
L
( p
!
Y
(. (*)
At the level of sections, these morphisms may be visualized as follows. Let
U, V be open subsets of X and Y . Then, on one hand, we have
R(U V, p
!
Y
() RHom(Z
UV
, p
!
Y
(
|V
)
RHom(Rp
V
!
Z
UV
, (
|V
)
RHom(R
c
(U; Z
U
), R(V ; ()).
On the other hand,
R(U;
X
)
L
R(V ; () RHom(R
c
(U; Z
U
), Z)
L
R(V ; ()
and the canonical morphism
R(U;
X
)
L
R(V ; () R(U V ; p
!
Y
()
induced by (*) corresponds to the canonical morphism
RHom(R
c
(U; Z
U
), Z)
L
R(V ; () RHom(R
c
(U; Z), R(V ; ()).
The last part of the result follows by taking limits and cohomology and
using the fact that
RHom(C
, Z)
L
P
RHom(C
, P
)
if C
is a bounded complex with nitely generated cohomology.

Proposition 1.13.12. Let X and Y be locally compact topological spaces.
Then, there is a canonical morphism
Y

XY
. (1)
If and be families of supports on X and Y , this morphism induces a
canonical morphism
R
(X;
X
)
L
R
(Y ;
Y
) R
(
XY
) (2)
and, hence, cross-products in homology
: H
k
(X; Z) H
l
(Y ; Z) H
k+l
(X Y ; Z).
Moreover, if X or Y is a clc space and (resp. ) is the family of all
compact subsets of X (resp. Y ), then both (1) and (2) are isomorphisms.
Proof. This follows directly from the preceding lemma and the fact that
p
!
Y
Y
=
XY
.
Remark 1.13.13. A link between the cross-products in homology and co-
homology is given by the formula
(c
p
c
q
) (c
r
c
s
) = (1)
p(sq)
(c
p
c
r
) (c
q
c
s
)
which entails the formula
c
k
c
l
, c
k
c
l
_
=
c
k
, c
k
_
c
l
, c
l
_
.
Note also that homology cross-products are compatible with push-forwards.
Namely, if f : X X
and g : Y Y
are morphisms of locally compact

spaces, then
(f g)
(c
k
c
l
) = f
(c
k
) g
(c
l
).
1.14. Cohomology of topological manifolds 65
1.14 Cohomology of topological manifolds
Denition 1.14.1. A topological manifold of dimension n is a Haussdorf
topological space which is locally isomorphic to R
n
.
Proposition 1.14.2. A topological manifold X of dimension n is a clc
locally compact space. Its cohomological dimension is n,
X
is concentrated
in degree n and H
n
(
X
) is a locally constant sheaf with ber Z.
Proof. This follows directly from the denition and Proposition 1.11.14.
Denition 1.14.3. We dene the orientation sheaf or
X
of X by setting
or
X
= H
n
(
X
).
The manifold X is orientable if and only if the sheaf or
X
is constant. In
such a case, an orientation of X is an isomorphism
Z
X
or
X
.
The manifold X endowed with an orientation forms an oriented manifold.
The orientation class of the oriented manifold X is the section
X
(X; or
X
)
image of the section
1
X
(X; Z
X
)
by the orientation of X.
Proposition 1.14.4. Assume X is a topological manifold. Then,
(K; or
X
) SH
n
(X, X K; Z)
canonically for any compact subset K of X. In particular, the notions
related to orientability considered above are compatible with the ones con-
sidered in singular homology.
Proof. As a matter of fact, or
X

X
[n] and
(K; or
X
) H
0
R(K;
X
[n]) H
n
R(K;
X
)
and the announced isomorphism follows from Corollary 1.12.4. Recall that
an orientation of X from the point of view of singular homology corresponds
to the data of a generator
x
SH
n
(X, X x; Z) for any x X in
such a way that for any x
0
X there is a neighborhood K of x
0
and
K
SH
n
(X, XK; Z) such that
x
is the image of
K
in SH
n
(X, Xx; Z)
for any x K. Using the isomorphism
(K; or
X
) SH
n
(X, X K; Z)
one sees easily that the family (
x
)
xX
corresponds to a section of or
X
on X which generates (or
X
)
x
for any x X and the conclusion follows.
Remark 1.14.5.
(a) If X is orientable and connected, we have
Hom(Z
X
, Z
X
) (X; Z
X
) Z.
It follows that the sheaf Z
X
has only two automorphisms (id). Therefore,
X has exactly two orientations. If
X
is the class of one of them,
X
is
the class of the other.
(b) Any open subset U of a topological manifold X is a topological
manifold. If
X
is an orientation class of X,
X
|U
is an orientation class for
U.
(c) We may restate Proposition 1.11.14 by stating that the topological
manifold R
n
is canonically oriented. We will denote
R
n the corresponding
orientation class.
(d) Since
X
= or
X
[n], we have
(U; or
X
) = H
n
(U;
X
).
Thanks to Corollary 1.11.11, we get that
(U; or
X
) = H
n
(RHom(R
c
(U; Z
X
), Z)).
Since H
n+1
c
(U; Z
X
) 0, we obtain a canonical isomorphism
(U; or
X
) Hom(H
n
c
(U; Z
X
), Z).
This provides a more explicit way to view the sheaf or
X
.
Denition 1.14.6. A homeomorphism : X Y of oriented topolog-
ical manifolds is oriented if the orientation class of X corresponds to the
orientation class of Y through the canonical isomorphism
1
or
Y
or
X
.
Lemma 1.14.7. A dieomorphism : U V between open subsets of R
n
is oriented if and only if its Jacobian J
is strictly positive.
Proof. Let x
0
and y
0
be points of U and V . Denote
U
and
V
the canonical
orientation classes of U and V and denote
U
the image of
U
by the
1
or
V
or
U
. Let B
U
(resp. B
V
) be an open ball
of U (resp. V ) containing x
0
(resp. y
0
). Assume that (B
U
) B
V
. Using
part (d) of Remark 1.14.5, we see that
U
|B
U
=
U
|B
U
according to the
fact that the diagram
H
n
c
(B
U
; Z)
Z
H
n
c
((B
U
); Z)
O
H
n
c
(B
V
; Z)
commutes or anticommutes. Using Exercise 1.8.6, we know that if

v
(B
U
)
H
n
c
((B
U
); Z)
has integral 1 then so has i(v
(B
U
)
). Let m be an integer such that
(v
(B
U
)
) = mv
U
where v
U
H
n
c
(B
U
; Z) is a class with integral 1. On one hand, we have
_

(v
(B
U
)
) = m.
On the other hand, representing the classes by means of de Rhamcomplexes,
we get
_

(v
(B
U
)
) =
_
v
(B
U
)
= 1
according to the fact that J
is positive or negative in B
U
. The conclusion
follows.
Proposition 1.14.8. Let (
i
: U
i
V
i
)
iI
be a family of homeomorphisms
such that X =

iI
V
i
each U
i
being an open subset of R
n
. Assume that
the homeomorphism
1
j

i
:
1
i
(V
i
V
j
)
1
j
(V
i
V
j
)
is oriented for any i, j I. Then, there is a unique orientation of X such
that
i
: U
i
V
i
is oriented for any i I. Moreover, any orientation of X
may be obtained in this way.
Proof. The rst part is obtained by gluing the isomorphisms
Z
V
i
or
V
i
induced by the isomorphisms Z
U
i
or
U
i
corresponding to the canonical
orientations of the U
i
s.
As for the second part, it follows directly from the denition if one keeps
in mind that it is always possible to reverse the orientation of a homeomor-
phism
i
: U
i
V
i
(U
i
open subset of R
n
)
by composing it with the reection
(x
1
, , x
n
) (x
1
, , x
n
).
Corollary 1.14.9. On a dierential manifold, the topological and dier-
ential notions of orientation coincide
Exercise 1.14.10. Let T be an abelian sheaf on the topological space X
(a) Assume U and V are two open subsets of X with X = U V and
U V non-empty and connected. Show that if the abelian sheaves
T
|U
and T
|V
are constant, then so is T.
(b) Deduce from (a) that a locally constant sheaf on [0, 1]
n
is constant.
Remark 1.14.11 (Classication of locally constant sheaves).
Let X be a topological space which is path-connected and locally path-
connected, let T be a locally constant sheaf on X and let : [0, 1] X be
a continuous path between x and y. It follows from the preceding exercise
that we have
1
T M
[0,1]
where M is an abelian group. Hence, the canonical morphisms
(
1
T)([0, 1])
_
1
T
_
0
and (
1
T)([0, 1])
_
1
T
_
1
are isomorphisms. This gives us a canonical isomorphism
_
1
T
_
0

_
1
T
_
1
and consequently a canonical isomorphism
m
: T
x
T
y
.
which is called the monodromy along . A simple computation shows that
if
is a path from y to z then m
= m
. If the path
0
and
1
are
connected by a homotopy h : [0, 1]
2
X i.e. if
h(t, 0) =
0
(t) and h(t, 1) =
1
(t);
then h
1
T is constant on [0, 1]
2
and one sees easily that the morphisms
m
1
: T
x
T
y
and m
2
: T
x
T
y
are equal. In particular, m
induces a representation of the Poincare

group
1
(X, x) of X at x on T
x
which is called the monodromy representa-
tion of T at x.
It can be shown that the functor from the category of locally constant
sheaves on X to the category of representations of
1
(X, x) on abelian
groups obtained by associating to a locally constant sheaf T on X its mon-
odromy representation at x is an equivalence of categories.
A trivial consequence is that locally constant sheaves on X are constant
if X is simply path-connected.
Another consequence is that locally constant sheaves with ber Z on X
correspond to representations of
1
(X, x) on Z. Since the only automor-
phisms of Z are id, these representations may be classied by morphisms
1
(X, x) Z
2
;
non trivial representations corresponding to epimorphisms. Therefore, non
constant locally constant sheaves with ber Z on X are classied by invari-
ant subgroup of index 2 of
1
(X, x).
In particular, if X is a topological manifold and
1
(X, x) has no invariant
subgroup of index 2, then X is orientable.
Proposition 1.14.12. Let X be an oriented topological manifold of di-
mension n. Denote
X
its orientation class. Then,
X
may be viewed as
an element of H
n
(X; Z) and

X
: H
k
(X; Z) H
nk
(X; Z)
is an isomorphism.
Proof. Since
X
= or
X
[n], we have
(X; or
X
) = H
n
(R(X;
X
)) = H
n
(X; Z
X
)
and
X
may be viewed as an element of H
n
(X; Z
X
). By the functoriality
of the cup product, we have the following commutative diagram
H
k
(X; Z
X
) H
n
(X;
X
)

H
kn
(X;
X
)
H
k
(X; Z
X
) H
n
(X; Z
X
[n])
id H
n
(X;)
H
kn
(X; Z
X
[n])
H
kn
(X;)
where : Z
X
[n]
X
is the isomorphism associated to
X
. Since
X
is
the image of 1
X
(X; Z
X
) = H
n
(X; Z
X
[n]) by H
n
(X; ), the diagram
H
k
(X; Z
X
)

X
H
kn
(X;
X
)
H
k
(X; Z
X
)
id
1
X
H
kn
(X; Z
X
[n])
H
kn
(X;)
is commutative and the conclusion follows.

Remark 1.14.13. When we work with oriented topological manifolds, the
preceding proposition allows us to transform operations on cohomology into
operations on homology and vice-versa. In particular, if f : X Y is a
continuous map between oriented topological manifolds of dimension n
X
,
n
Y
, then
f
: H
c
k
(X; Z) H
c
k
(Y ; Z)
and
f
: H
k
(Y ; Z) H
k
(X; Z)
induce canonical morphisms
f
!
: H
n
X
k
c
(X; Z) H
n
Y
k
c
(Y ; Z)
and
f
!
: H
n
Y
k
(Y ; Z) H
n
X
k
(X; Z).
Moreover,
: H
k
(X; Z) H
l
(X; Z) H
k+l
(X; Z)
gives rise to the intersection product
: H
n
X
k
(X; Z) H
n
X
l
(X; Z) H
n
X
kl
(X; Z).
Among the many compatibility formulas relating these operations, let us
just recall that
f
!
(f
c
p
c
q
) = c
p
f
!
c
q
or dually that
f
(f
!
c
p
c
q
) = c
p
f
(c
q
)
and that
(c
p
c
q
) (c
r
c
s
) = (1)
(n
X
p)(n
Y
s)
(c
p
c
r
) (c
q
c
s
).
Denition 1.14.14. Let X, Y be oriented topological manifolds of dimen-
sion n and p. Assume Y is a closed subspace of X and denote i : Y X
the inclusion map. Then, the fundamental homology class of Y in X is the
class
[Y ] = i
(
Y
)
of H
Y
p
(X; Z). Dually, the fundamental cohomology class (or Thom class) of
Y in X is the class
Y/X
= i
!
(1
Y
)
of H
np
Y
(X; Z) which is also characterized by the formula
Y/X

X
= [Y ].
Proposition 1.14.15. Let X be an oriented dierential manifold of dimen-
sion n and let Y , Z be closed oriented dierential submanifolds of dimension
p and q of X. Assume Y and Z meet transversally. Then Y Z is a closed
dierential submanifold of dimension p + q n of X which is canonically
oriented. Moreover, for this orientation, we have
[Y Z] = [Y ] [Z]
and
Y Z/X
=
Y/X

Z/X
.
Proof. We will only treat the case where n > p, n > q, p + q > n, leav-
ing the adaptation to the other cases to the reader. Since Y and Z meet
transversally, we have
T
x
X = T
x
Y + T
x
Z and T
x
(Y Z) = T
x
Y T
x
Z
at any point x Y Z. It follows that it is possible to nd an oriented
basis of T
x
X of the form
u
1
, . . . , u
nq
, w
1
, . . . , w
p+qn
, v
1
, . . . , v
np
where
u
1
, . . . , u
nq
, w
1
, . . . , w
p+qn
and w
1
, . . . , w
p+qn
, v
1
, . . . , v
np
are oriented bases of T
x
Y and T
x
Z and
w
1
, . . . , w
p+qn
is a basis of T
x
(Y Z). We dene the canonical orientation of Y Z as the
one corresponding to such a basis.
Let us now prove that
[Y Z] = [Y ] [Z]
in H
Y
p+qn
(X; Z). Since
U H
UY
p
(U; Z), U H
UZ
q
(U; Z), U H
UY Z
p+qn
(U; Z)
are sheaves on X, the problem is of local nature. Therefore, we may assume
that
X = R
nq
R
p+qn
R
np
and that
Y = R
nq
R
p+qn
0, Z = 0 R
p+qn
R
np
the orientations being the products of the canonical orientations of the fac-
tors. In this case,
Y Z = 0 R
p+qn
0
with the orientation given by the canonical orientation of R
p+qn
. There-
fore,
[Y ] =
R
nq
R
p+qn [0], [Z] = [0]
R
p+qn
R
np
and
[Y Z] = [0]
R
p+qn [0].
Since by Remark 1.14.13 we have
(
R
nq
R
p+qn [0]) ([0]
R
p+qn
R
np )
= (
R
nq [0]) (
R
p+qn
R
p+qn) ([0]
R
np )
= [0]
R
p+qn [0]
the conclusion follows.
1.15 Sheaves of rings and modules
In order to focus the survey contained in the preceding sections on the basic
ideas of sheaf theory, we have chosen to deal only with sheaves of abelian
groups. In the rest of this book, we will however need to use sheaves of
rings and modules. The adaptation of the theory reviewed above to this
more general situation being rather mechanical, we will not do it explicitly
here and refer the interested reader to standard texts on the subject.
2
Euler class of manifolds and
real vector bundles
2.1 Lefschetz xed point formula
Let X, Y be two compact oriented topological manifolds of dimension n.
Recall that a correspondence between X and Y is a subset C of X Y .
The image of a subset A X by the correspondence C is the set
C(A) = y Y : x A, (x, y) C
which may also be described as
p
Y
(p
1
X
(A) C).
Following Lefschetz, we will introduce similar notions in the framework of
homology.
Denition 2.1.1. A homological correspondence from X to Y is a class
n
H
n
(X Y ; Z).
For any abelian group M, the image of a class c
p
H
p
(X; M) by
n
is
the class
n
(c
p
) = (p
Y
)
(p
!
X
(c
p
)
n
) H
p
(Y ; M).
Let f : X Y be a continuous map. As usual, set
f
(x) = (x, f(x))
and let
f
= Im
f
= (x, y) : y = f(x)
73
74 2. Euler class of manifolds and real vector bundles
denote the graph of f. Clearly,
f
: X X Y
induces a homeomorphism between X and
f
which turns
f
into a com-
pact oriented manifold of dimension n. We will denote
i
f
:
f
X Y
the canonical inclusion. As is well-known, we may recover f from the cor-
respondence
f
. A similar result is true in homology.
Proposition 2.1.2. The class
f
= (
f
)
(
X
) = (i
f
)
f
)
is a homological correspondence from X to Y for which we have
f
(c
p
) = f
(c
p
).
Proof. Recall that
[
f
] = (i
f
)
f
) = (
f
)
(
X
).
Recall also that c
p

X
= c
p
. As a matter of fact, for c
p
= c
np

X
, we
have
c
p

X
= (c
np

X
)
X
= (c
np
1)
X
= c
p
.
Therefore, keeping in mind that
p
X

f
= id
X
, p
Y

f
= f,
we have successively
f
(c
p
) = (p
Y
)
(p
!
X
c
p
(
f
)
(
X
))
= (p
Y
)
(
f
)
(
!
f
p
!
X
c
p

X
)
= f
(c
p

X
)
For p N, let
p,r
(r = 1, , R
p
) be a basis of the nite dimension
vector space H
p
(X; Q). Thanks to Poincare-Verdier duality, we know that
the pairing
H
p
(X; Q) H
np
(X; Q) Q
2.1. Lefschetz xed point formula 75
which sends (c
p
, c
np
) to #(c
p
c
np
) is perfect. Therefore, there is a unique
basis
p,r
(r = 1, , R
p
) of H
p
(X; Q) such that
#(
p,r

np,r
) =
rr
for any r, r
1, , R
p
. Denote
q,s
,
q,s
(s = 1, , S
q
) similar basis
for the rational homology of Y .
Proposition 2.1.3. Let be a homological correspondence from X to Y .
Then,
=
n
p=1
R
p
r=1
S
p
s=1
(
p
)
sr
np,r
p,s
where
p
is the matrix of
() : H
p
(X; Q) H
p
(Y ; Q)
with respect to the basis
p,r
(r = 1, , R
p
) and
p,s
(s = 1, , S
p
).
Proof. Thanks to K unneth theorem, we know that the classes
p,r
np,s
(p = 1, , n; r = 1, , R
p
; s = 1, , S
p
)
form a basis of H
n
(X Y ; Q). Hence
=
n
p=1
R
p
r=1
S
p
s=1
(
p
)
sr
np,r
p,s
where the (
p
)
sr
are rational numbers. Let us compute (
p
0
,r
0
). Note that
p
!
X
(c
p
) = (1)
n(np)
c
p
Y
.
As a matter of fact, we may assume c
p
= c
np

X
. Hence,
p
!
X
(c
p
) = p
X
(c
np
)
XY
= (c
np
1
Y
) (
X

Y
)
= (1)
n(np)
(c
np

X
) (1
Y

Y
) (1)
= c
p
Y
(2)
where in (1) we have used the formula
(c
p
c
q
) (c
r
c
s
) = (1)
p(sq)
(c
p
c
r
) (c
q
c
s
)
and in (2) the formula
1
Y
c
p
= c
p
.
Therefore, (
p
0
,r
0
) is equal to
n
p=1
R
p
r=1
S
p
s=1
(1)
n(np
0
)
(
p
)
sr
(p
Y
)
_
(
p
0
,r
0

Y
) (
np,r
p,s
)
.
Using the formula
(
p
q
) (
r

s
) = (1)
(np)(ns)
(
p

r
) (
q

s
)
we see that
(p
Y
)
_
(
p
0
,r
0

Y
) (
np,r
p,s
)
= (1)
(np
0
)(np)
(p
Y
)
p
0
,r
0
(
np,r
) (
Y

p,s
)
= (1)
(np
0
)(np)
#(
p
0
,r
0

np,r
)
p,s
(*)
= (1)
(np
0
)(np)
(1)
p(np
0
)
p,p
0
r,r
0
p,s
where (*) follows from the formula (f g)
(
p

q
) = f
(
p
) g
(
q
).
Therefore,
(
p
0
,r
0
) =
n
p=1
R
p
r=1
S
p
s=1
(
p
)
sr
p,p
0
r,r
0
p,s
=
R
p
r=1
S
p
s=1
(
p
0
)
sr
0
p
0
,s
Recall that a xed point of a correspondence C from X to X is a point
x X such that x C(x). Such a point is characterized by the fact that
(x, x) C where is the diagonal of X X.
Denition 2.1.4. Let H
n
(X X; Z) be a homological correspondence
from X to X. We dene the algebraic number of xed points of as the
number
#( [])
and the Lefschetz number of as the number
=
n
p=0
(1)
p
tr [() : H
p
(X; Q) H
p
(X; Q)] .
Theorem 2.1.5 (Lefschetz xed points formula). Assume
H
n
(X X; Z)
is a homological correspondence from X to X. Then,
#( []) =
.
Proof. Using the classes
p,r
,
p,r
introduced above and Proposition 2.1.3,
we see that
=
n
p=1
R
p
r=1
R
p
s=1
(
p
)
sr
np,r
p,s
where
p
is the matrix of
() : H
p
(X; Q) H
p
(X; Q)
with respect to the basis
p,r
(r = 1, , R
p
). Similarly, we have
id
=
n
=1
R
p
=1
R
p
=1
,r

np
,r

p
,s
.
Let : X X X X be the morphism dened by setting
(x, x
) = (x
, x).
We have
(
id
) =
n
=1
R
p
=1
R
p
=1
(1)
p
(np
,r

p
,s

np
,r
and
(
id
) =
X
=
X
=
id
.
Therefore,
[] =
n
=1
R
p
=1
R
p
=1
(1)
p
(np
,r

p
,s

np
,r
.
Since
#
_
(
np,r
p,s
) (
p
,s
np
,r
)
_
= (1)
pp
(1)
p
(np
pp
rs
sr
we see that
#( []) =
n
p=1
R
p
r=1
R
p
s=1
(1)
p
2
(
p
)
sr
sr
=
n
p=1
(1)
p
2
tr
p
and the conclusion follows since p
2
p (mod 2).
Corollary 2.1.6.
(a) Let f : X X be a continuous map. Set
f
=
f
. Then,
#(
f
[]) =
f
=
n
p=0
(1)
p
tr(f
: H
p
(X; Q) H
p
(X; Q)).
In particular,
f
= 0 if f has no xed point.
(b) For f = id
X
, we get
#([] []) =
id
=
n
p=0
(1)
p
dim
Q
H
p
(X; Q) = (X).
Exercise 2.1.7. Assume n N
0
. Let f : S
n
S
n
be a continuous map.
Dene the degree of f as the unique integer deg(f) making the diagram
H
n
(S
n
; Z)
f
H
n
(S
n
; Z)
Z
O
v
S
n
deg(f)
Z
O
v
S
n
commutative.
(a) Show that
f
= 1 + (1)
n
deg(f).
(b) Deduce from (a) that f has a xed point if deg(f) ,= (1)
n+1
.
(c) Apply (b) to show that if there is a homotopy connecting f to id
S
n
and n is even then f has a xed point.
(d) As an application, show that any vector eld of class C
1
on S
n
must
vanish at some x S
n
if n is even.
Solution. (a) follows directly from the denition of deg(f) and the fact that
H
k
(S
n
; Z) =
_
Z if k = 0, n;
0 otherwise.
(b) Since
f
= 0 if and only if deg(f) = (1)
n+1
, the conclusion follows
directly from Corollary 2.1.6.
(c) When f id
S
n
, we have f
= (id
S
n
)
= id. Hence, deg(f) = 1 and

the conclusion follows from (b).
(d) Let
t
(x) be the ow of the vector eld . From (c), we know that
t
has a xed point x
t
S
n
for any t R. Since S
n
is compact, we may
nd a sequence t
k
0
+
such that x
t
k
x
0
S
n
. Let t R and choose
the sequence m
k
of integers such that m
k
t
k
t. Clearly,
t
(x
0
) = lim
k
m
k
t
k
(x
t
k
) = lim
k
x
t
k
= x
0
.
It follows that
t
(x
0
) = x
0
for any t R and hence that (x
0
) = 0.
Exercise 2.1.8.
(a) Let X be a compact oriented manifold of dimension n. Show that
(X) = 0
if n is odd.
(b) Let X be a compact oriented C
1
manifold. Show that if X has a
nowhere vanishing C
1
vector eld then (X) = 0.
Solution. (a) Set
R
p
= dimH
p
(X; Q).
By Poincare duality, we have R
p
= R
np
. Therefore, if n = 2k + 1 (k N)
we have
(X) =
n
p=0
(1)
p
R
p
=
k
p=0
(1)
p
R
p
+
2k+1
p=k+1
(1)
p
R
p
=
k
p=0
(1)
p
R
p
+
k
p=0
(1)
np
R
np
= 0.
(b) Working as in Exercise 2.1.7, we see that (X) ,= 0 entails that any
C
1
-vector eld on X vanish for some x X. The conclusion follows.
Corollary 2.1.9. Let X be a compact dierential manifold of dimension
n.
(a) Assume C is a closed dierential submanifold of dimension n of XX
which meets transversally. Then C is nite and
[C]
=
(x,x)C
i
(x,x)
(C, )
where i
(x,x)
(C, ) is equal to 1 if the canonical orientation of
T
(x,x)
C T
(x,x)
coincides with that of T

(x,x)
(X X) and to 1 otherwise.
(b) Assume f : X X is a dierentiable map with a non-empty set F of
xed points. Assume that for any x F,
f
x
: T
x
X T
x
X
does not x any non-zero tangent vector. Then, the set F is nite and
f
=

xF
sgn(det(idf
x
)).
Proof. (a) Thanks to Theorem 2.1.5, the result follows directly from Propo-
sition 1.14.15.
(b) Clearly, (x, x)
f
if and only if x F. Moreover, for any
x F, we have
T
(x,x)
f
= (, f
x
()) : T
x
X
and
T
(x,x)
= (, ) : T
x
X.
Hence, we have
T
(x,x)
f
T
(x,x)
= (, ) : T
x
X, f
x
() = 0.
It follows that
f
and meet transversally. If
1
, . . . ,
n
is an oriented
basis of T
x
X, then
(
1
, f
x
(
1
)), . . . , (
n
, f
x
(
n
))
and
(
1
,
1
), . . . , (
n
,
n
)
are oriented bases for T
(x,x)
f
and T
(x,x)
. Therefore,
(
1
, f
x
(
1
)), . . . , (
n
, f
x
(
n
)), (
1
,
1
), . . . , (
n
,
n
)
is an oriented basis for T
(x,x)
f
T
(x,x)
. This basis has the same orien-
tation as the basis
(0, f
x
(
1
)
1
), . . . , (0, f
x
(
n
)
n
), (
1
,
1
), . . . , (
n
,
n
).
Using the assumptions, we know that f
x
id is injective and hence bijective.
It follows that
f
x
(
1
)
1
, . . . , f
x
(
n
)
n
is a basis of T
x
X. Hence,
(0, f
x
(
1
)
1
), . . . , (0, f
x
(
n
)
n
), (
1
, 0), . . . , (
n
, 0).
is an oriented basis for T
(x,x)
f
T
(x,x)
and
i
(x,x)
(
f
, ) = (1)
n
sgn(det(f
x
id)) = sgn(det(idf
x
)).
The conclusion follows from (a).
2.2. Euler classes of manifolds and index theorem 81
2.2 Euler classes of manifolds and index theorem
Denition 2.2.1. The Euler class of an oriented topological manifold X
is the class
X
=
/XX
Proposition 2.2.2. For any oriented compact topological manifold, we
have
(X) =
_
X
X
.
Proof. We know that
(X) = #([] []).
Therefore,
(X) = #((
/XX

/XX
)
XX
)
=
_
XX
/XX

/XX
=
_
XX
/XX

!
(1
X
)
=
_
XX
!
(
/XX
) =
_
X
X
2.3 Basic notions on real vector bundles
A (continuous) real vector bundle of rank r is the data of a continuous map
p
E
: E B
E
between topological spaces together with structures of real vector spaces of
dimr on each ber E
b
= p
1
E
(b) (b B) of p. These data being such that
for any b B there is a neighborhood U of b in B and a family (e
1
, , e
r
)
of continuous sections of p
E
|p
1
E
(U)
: p
1
E
(U) U with the property that
(e
1
(b
), , e
r
(b
)) is a basis of p
1
E
(b
) for any b
U. We call p
E
(resp.
E, B
E
) the projection (resp. the total space, the base) of the vector bundle.
A family like (e
1
, , e
r
) is called a (continuous) frame of the real vector
bundle over U. We will often refer to a vector bundle by giving its total
space alone assuming that the projection and the basis are clear from the
context.
Let E be a real vector bundle of rank r and let U be an open subset of its
base B
E
. Then, p
E
|p
1
E
(U)
: p
1
E
(U) U has clearly a canonical structure
of real vector bundle of rank r. We call it the restriction of E to U and
denote it by E
|U
.
A morphism between two real vector bundles E, F with common base B
is a continuous map f : E F such that p
F
f = p
E
, the map f
b
: E
b
F
b
induced by f being R-linear for any b B. Clearly, morphisms of real
vector bundles may be composed and R-linearly combined in a natural way.
Hence, real vector bundles with base B form a category. We will denote it
by 1ect
R
(B). One sees easily that this category is additive, the direct sum
E F of two real vector bundles E and F with base B being characterized
by the fact that
(E F)
b
= E
b
F
b
,
a local frame of E F being given by (e
1
, , e
r
, f
1
, , f
s
) if (e
1
, , e
r
)
and (f
1
, , f
s
) are local frames of E and F.
A trivial vector bundle is a vector bundle of the form
p
B
: V B B
where V is a real vector space. A vector bundle is trivializable if it is
isomorphic to a trivial vector bundle. By denition, any vector bundle is
locally of this type.
A sub-bundle of a vector bundle F with basis B is the data of a vector
bundle E with basis B together with a morphism i : E F for which
i
b
: E
b
F
b
is injective for any b B.
A quotient bundle of a vector bundle E with basis B is the data of a
vector bundle F with basis B together with a morphism q : E F for
which q
b
: E
b
F
b
is surjective for any b B.
Let i : E F be a sub-bundle of F. Then, i has a cokernel in 1ect
R
(B)
characterized by the fact that
(Coker i)
b
= Coker i
b
,
a local frame of Coker i being obtained by considering a local frame of F of
the type (i(e
1
), , i(e
r
), f
r+1
, , f
s
) with (e
1
, , e
r
) a local frame of E
and taking the images of f
r+1
, , f
s
in Coker i. From this construction, it
follows that Coker i together with the canonical morphism F Coker i is
a quotient bundle of F that one often denotes F/E when i is clear from the
context.
Let q : E F be a quotient bundle of E. Then, q has a kernel in
1ect
R
(B) characterized by the fact that
(Ker q)
b
= Ker q
b
,
a local frame of Ker i being obtained by considering a local frame of E of the
type (f
1
, , f
r
, f
r+1
, , f
s
) where (q(f
r+1
), , q(f
s
)) is a local frame of
2.4. Orientation of real vector bundles 83
F and viewing (f
1
, , f
r
) as local sections of Ker q. Clearly, Ker i together
with the canonical morphism j : Ker i E is a sub-bundle of E.
Note that if i : E F (resp. q : E F) is a sub-bundle of F (resp. a
quotient bundle of E) then
E Ker(F Coker i) (resp. F Coker(Ker q E)).
Note also that when the base space B is paracompact, one can show,
using a partition of unity, that for any sub-vector bundle E of F one has
F E F/E.
Finally, let us recall that the inverse image of vector bundle E of base
B by a continuous map f : B
B is the vector bundle f

1
(E) with base
B

f
1
(E)
b
= E
f(b
)
for any b
, a frame of f
1
(E) on f
1
(U) being given by (e
1
f, , e
r
f) where (e
1
, , e
r
) is a frame of E on U.
2.4 Orientation of real vector bundles
Let A be a noetherian ring with nite global homological dimension and let
E be a real vector bundle of rank r and base B.
Denition 2.4.1. The relative dualizing complex of B in E for sheaves of
A-modules is the complex
A
B/E
= (R
B
A
E
)
|B
.
Proposition 2.4.2.
(a) The canonical restriction morphism
Rp(R
B
A
E
) (R
B
A
E
)
|B
is an isomorphism. In particular,
R(U;
A
B/E
) R
U
(p
1
(U); A)
for any open subspace U of X.
(b) The canonical restriction morphism
(
A
B/E
)
b
R
{b}
(E
b
; A)
is an isomorphism. In particular,
A
B/E
is concentrated in degree r.
Proof. (a) One checks easily by working at the level of bers that
p(
B
T) (
B
T)
|B
is an isomorphism for any abby sheaf T on E. The rst part follows. As
for the second part, it is a consequence of Leray theorem.
(b) The problem being local on B, we may assume E = R
r
B. Consider
the morphism of distinguished triangles
[Rp(R
U
A
R
r
U
)]
b

(1)
[Rp(A
R
r
U
)]
b

(2)
[R p(A

E
)]
b
+1
(3)
R
{b}
(R
r
b; A)

R(R
r
b; A)

R(R
r
0 b; A)
+1
We will prove that (2) and (3) are isomorphisms. This will show that (1) is
also an isomorphism and the conclusion will follow.
By simple homotopy arguments, one gets the isomorphisms
[Rp(A
R
r
U
)]
b
A
and
R(R
r
b; A) A.
Since these isomorphisms transform (2) into the identity, (2) is also an
isomorphism.
Denote q : S
r1
U U the second projection. Working as above one
sees that (3) will be an isomorphism if the canonical morphism
[Rq(A
S
r1
U
)]
b
R(S
r1
b; A)
is an isomorphism. Since S
r1
is compact, the ber of q at b is compact
and relatively Haussdorf in S
r1
U. It is thus a taut subspace and the
conclusion follows from the ber formula for Rq
Denition 2.4.3. The relative A-orientation sheaf of B in E is the sheaf
or
A
B/E
= H
r
A
B/E
Proposition 2.4.4.
(a) We have
R(U; or
A
B/E
) R
U
(p
1
(U); A)[r]
and in particular
(U; or
A
B/E
) H
r
U
(p
1
(U); A)
2.4. Orientation of real vector bundles 85
(b) The canonical restriction morphism
[or
A
B/E
]
b
H
r
{b}
(E
b
; A)
is an isomorphism.
(c) Any frame (e
1
, , e
r
) of E on U induces a canonical isomorphism
(e
1
, ,e
r
)
: A
U
(or
A
B/E
)
|U
.
In particular, or
A
B/E
is a locally constant sheaf with ber A.
(d) Let (e
1
, , e
r
) and (e
1
, , e
r
) be two frames of E on U. Set
e
i
(u) =
r
s=1
S
ji
(u)e
j
(u)
for i = 1, , r. Assume det S(u) 0 on U. Then,
(e
1
, ,e
r
)
=
(e
1
, ,e
r
)
.
Proof. (a) It follows from part (b) of the preceding proposition that
A
B/E
or
A
B/E
[r].
Therefore,
R(U; or
A
B/E
) R(U;
A
B/E
)[r]
and the conclusion follows from part (a) of preceding proposition.
(b) This follows directly from part (b) of the preceding proposition.
(c) Let (e
1
, , e
r
) be a frame of E on U. Consider the morphism
: R
r
U p
1
(U)
dened by setting
(x
1
, , x
r
, u) =
r
j=1
x
j
e
j
(u).
Clearly, is an isomorphism of real vector bundles. It follows that the
canonical morphism
: or
A
U/p
1
(U)
or
A
U/R
r
U
(*)
is an isomorphism. By pull-back through the rst projection, the canonical
generator
v
{0}
H
r
{0}
(R
r
; A)
gives the canonical class
p
R
r v
{0}
H
r
U
(R
r
U; A).
Thanks to the second isomorphism of (a), we get a canonical class
U/R
r
U
(U; or
A
U/R
r
U
).
This class induces a morphism
A
U
or
A
U/R
r
U
. (**)
which corresponds at the level of bers to the isomorphism
A H
r
{0}
(R
r
; A)
induced by v
{0}
. It follows that (**) is an isomorphism. By combining it
with (*), we get the conclusion.
(c) Consider the isomorphism
: R
r
U R
r
U
dened by setting (x
, u) = (S(u)x
, u) and assume det S(u) > 0 (resp.

det S(u) < 0) on U. To conclude, it is sucient to prove that the canonical
morphism
: or
A
U/R
r
U
or
A
U/R
r
U
sends
U/R
r
U
to itself (resp. minus itself). At the level of the ber at u,
this amounts to show that the pull-back by the linear bijection
x
S(u)x
of v
{0}
is equal to itself (resp. minus itself). This follows from Exercise 1.8.8.
Denition 2.4.5. The vector bundle E is A-orientable if the sheaf or
A
B/E
is constant. An A-orientation of E is the data of an isomorphism
A
B
or
A
B/E
or of the corresponding A-orientation class
A
B/E
(B; or
A
B/E
).
An orientation of the vector bundle E is the data of an orientation on
every ber E
b
(b B) in such a way that for any point b B there is a
neighborhood U and a frame (e
1
, , e
r
) of E on U with the property that
the basis (e
1
(b
), , e
r
(b
)) is positively oriented in E
b
for any b
U. The
vector bundle E is orientable if it can be given an orientation.
2.5. Thom isomorphism and Gysin exact sequence 87
Proposition 2.4.6.
(a) Any vector bundle E is canonically Z
2
-oriented.
(b) A vector bundle E is Z-orientable if and only if it is orientable. More-
over, any Z-orientation of E corresponds to a canonically determined
orientation and vice-versa.
Proof. We use the notations of Proposition 2.4.4. Denote
A
U/e
(U; or
A
B/E
)
the section corresponding to the isomorphism
e
. Clearly,
A
U/e
=
A
U/e
if det S(u) 0 on U. Therefore, in case (a), we get
Z
2
U/e
=
Z
2
U/e
and there
is a unique class
Z
2
B/E
(U; or
Z
2
B/E
) such that
Z
2
B/E
|U
=
Z
2
U/e
for any
frame e of E on U.
To prove (b), let us proceed as follows. Assume E is Z-orientable and
let
B/E
be a Z-orientation class of E. Let b B and let e be a frame of
E on a neighborhood U of b. The only generators of Z being 1 and 1, we
know that
(
B/E
)
b
= (
U/e
)
b
.
Hence, restricting U if necessary, we may assume that
B/U
|U
=
U/e
.
Changing the sign of one of the sections of e if necessary, we see that for
any b B there is a neighborhood U of b and a frame e of E on U such
that
B/E
|U
=
U/e
.
Choosing the orientations of the bers E
b
in order to make all these frames
positively oriented gives us a canonical orientation of E. The reverse pro-
cedure is similar.
2.5 Thom isomorphism and Gysin exact sequence
Denition 2.5.1. Let E be an A-oriented real vector bundle with base
B. The Thom class of E is the image of the orientation class
A
B/E
by the
(B; or
A
B/E
) H
r
B
(E; A).
We denote it by
A
E
.
Proposition 2.5.2 (Thom isomorphism). Let E be an A-oriented real
vector bundle with base B. Then,
H
k
(B; A
B
) H
k+r
B
(E; A
E
)
c
k
A
E
p
(c
k
)
is an isomorphism.
Proof. This follows directly from part (a) of Proposition 2.4.4.
Denition 2.5.3. The Euler class of an A-oriented vector bundle E with
base B is the class e
A
E
H
r
(B; A
B
) dened by setting e
A
E
= (
A
E
)
|B
where
A
E
is the image of
A
E
in H
r
(E; A
E
).
Proposition 2.5.4. Let E be an A-oriented real vector bundle. Then, a
necessary condition for the existence of a nowhere vanishing section of E is
that
e
A
E
= 0.
Proof. Set

E = EB and denote s
0
: B E the zero section of E. Assume
s : B E is a nowhere vanishing section of E. Thanks to the homotopy
theorem, we know that
s
= s
0
.
It follows that e
E
= s
A
E
= s
A
E
. Using the inclusion s(B)

E, we see
that e
E
= s
((
A
E
)
|

E
). Since
A
E
H
r
B
(E; A), it is clear that (
A
E
)
|

E
= 0.
Hence, e
A
E
= 0 and the conclusion follows.
Proposition 2.5.5 (Gysin exact sequence). For any A-oriented vector
bundle E with rank r and base B, there is a long exact sequence of the form
H
kr
(B; A)
e
A
E

H
k
(B; A)
p
H
k
(

E; A)
-, *+
/.
()

H
k+1r
(B; A)
e
A
E
H
k+1
(B; A)
p
H
k+1
(

E; A)
(*)
Proof. Consider the excision distinguished triangle
R
B
(E; A
E
) R(E; A
E
) R(

E; A

E
)
+1
Taking cohomology, we get the long exact sequence

H
k
B
(E; A
E
)

H
k
(E; A
E
)

H
k
(

E; A
E
)
-, *+
/.
()

H
k+1
B
(E; A
E
)

H
k+1
(E; A
E
)

H
k+1
(

E; A
E
)
(**)
2.5. Thom isomorphism and Gysin exact sequence 89
We know by Proposition 2.5.2 that
A
E
: H
kr
(B; A
B
) H
k
B
(E; A
E
)
is an isomorphism. Moreover, the homotopy theorem shows that
p
: H
k
(B; A
B
) H
k
(E; A
E
)
is an isomorphism, its inverse isomorphism being
s
0
: H
k
(E; A
E
) H
k
(B; A
B
).
Using these isomorphisms, we transform easily (**) into (*).
Exercise 2.5.6. Let n N
0
. Consider the real projective space of dimen-
sion n
P
n
(R) = d : d line of R
n+1
through 0 (R
n+1
0)/R
and the tautological real line bundle on P

n
(R)
U
n
(R) = (v, d) R
n+1
P
n
(R) : v d
the projection p : U
n
(R) P
n
(R) being dened by p(v, d) = d. Show by
using a suitable Gysin exact sequence that the morphism of graded rings
Z
2
[X]/(X
n+1
) H
(P
n
(R); Z
2
)
dened by sending X to e
Z
2
U
n
(R)
H
1
(P
n
(R); Z
2
) is an isomorphism.
Solution. For short, set e = e
Z
2
U
n
(R)
. Consider the Gysin exact sequence
H
k1
(P
n
(R); Z
2
)
e

H
k
(P
n
(R); Z
2
)
p
H
k
(

U
n
(R); Z
2
)
-, *+
/.
()

H
k
(P
n
(R); Z
2
)
e
H
k+1
(P
n
(R); Z
2
)
p
H
k+1
(

U
n
(R); Z
2
)
Clearly, the rst projection

U
n
(R) R
n+1
0 is a homeomorphism.
Hence, H
k
(

U
n
(R); Z
2
) H
k
(R
n+1
0; Z
2
). Moreover, by homotopy, we
know that
H
k
(R
n+1
0; Z
2
) H
k
(S
n
; Z
2
)
_
Z
2
if k = 0, n;
0 otherwise.
Assuming n > 1 and denoting q : S
n
P
n
(R) the canonical map, we get
the exact sequences
0 H
0
(P
n
(R); Z
2
)
q
H
0
(S
n
; Z
2
) H
0
(P
n
(R); Z
2
)
e
H
1
(P
n
(R); Z
2
) 0
0 H
n1
(P
n
(R); Z
2
)
e
H
n
(P
n
(R); Z
2
)
q
H
n
(S
n
; Z
2
)

H
n
(P
n
(R); Z
2
) 0
and isomorphisms
H
k
(P
n
(R); Z
2
)
e
H
k+1
(P
n
(R); Z
2
)
for 1 k n1. We know that P
n
(R) is a compact connected topological
manifold of dimension n. Therefore, H
0
(P
n
(R); Z
2
) Z
2
and
q
: H
0
(P
n
(R); Z
2
) H
0
(S
n
; Z
2
)
is an isomorphism. Hence,
H
0
(P
n
(R); Z
2
)
e
H
1
(P
n
(R); Z
2
)
is an isomorphism. By Poincare duality, we have
H
n
(P
n
(R); Z
2
) H
0
(P
n
(R); Z
2
) Z
2
.
The morphism being surjective is thus an isomorphism. It follows that
H
n1
(P
n
(R); Z
2
)
e
H
n
(P
n
(R); Z
2
)
is also an isomorphism. Summing up, for n > 1, we have established that
H
k
(P
n
(R); Z
2
)
e
H
k+1
(P
n
(R); Z
2
)
is an isomorphism for k = 0, , n1. Since we have also H
0
(P
n
(R); Z
2
)
Z
2
and H
k
(P
n
(R); Z
2
) 0 if k > n, the conclusion follows easily. The case
n = 1 is treated similarly.
2.6 Euler classes of inverse images and direct sums
Proposition 2.6.1. Let f : Y X be a continuous map and let E be a
vector bundle on X. Then, there is a canonical isomorphism
f
1
or
A
X/E
or
A
Y/f
1
(E)
.
In particular, any A-orientation of E on X induces an A-orientation of
f
1
(E) on Y .
Proof. Consider the commutative diagram
f
1
(E)
g
E
p
Y
f
X
2.6. Euler classes of inverse images and direct sums 91
where
g
|f
1
(E)
y
: f
1
(E)
y
E
f(y)
is the canonical isomorphism, p and q being the projections of the vector
bundles E and f
1
(E). Since g is continuous, it induces canonical pull-back
morphisms
R
U
(E
|U
; A
E
) R
V
(f
1
(E)
|V
; A
f
1
(E)
)
for any open subsets U, V of X, Y such that V f
1
(U). Taking coho-
mology, we obtain in the same conditions a canonical morphism
(U; or
A
X/E
) (V ; or
A
Y/f
1
(E)
).
This gives us a morphism of sheaves of A-module
f
1
or
A
X/E
or
A
Y/f
1
(E)
(*)
It remains to prove that this is an isomorphism. At the level of bers, (*)
may be visualized through the commutative diagram
(or
A
X/E
)
f(y)

O
(or
A
Y/f
1
(E)
)
y
O
H
r
{f(y)}
(E
f(y)
; A)
g
|f
1
(E)
y
H
r
{y}
(f
1
(E)
y
; A)
Since g
|f
1
(E)
y
: f
1
(E)
y
E
f(y)
is a homeomorphism, g
|f
1
(E)
y
is an
isomorphism and the conclusion follows.
Remark 2.6.2. In the situation of the preceding proposition, assume E
is oriented. Then, it is canonically Z-oriented and we get a canonical Z-
orientation on f
1
E. One checks easily that one can characterize the cor-
responding orientation by the fact that if e
1
, . . . , e
r
is a positively oriented
frame of E on U then e
1
f, . . . , e
r
f is a positively oriented frame of f
1
E
on f
1
(U).
Proposition 2.6.3. Let f : Y X be a continuous map and let E be an
A-oriented vector bundle of rank r on X. Assume f
1
(E) is endowed with
the A-orientation induced by that of E. Then,
e
A
f
1
(E)
= f
e
A
E
.
Proof. It follows from the proof of the preceding proposition that the image
by
g
: H
r
X
(E; A) H
r
Y
(f
1
(E); A)
of
E
is
f
1
(E)
. Therefore, denoting s
0,X
and s
0,Y
the zero sections of E
and f
1
(E), we have
e
A
f
1
(E)
= s
0,Y
A
f
1
(E)
= s
0,Y
g
A
E
= f
0,X
A
E
= f
e
A
E
.
Exercise 2.6.4. Let E be a real vector bundle of rank r on X with e
Z
2
E
,= 0.
Showthat if E can be represented as an inverse image of a real vector bundle
on S
n
, then n = r.
Solution. This follows directly from the preceding proposition and the fact
that
H
k
(S
n
; Z
2
) = 0
for k , 0, n.
Proposition 2.6.5. Let E (resp. F) be a vector bundle of rank r (resp. s)
on X. Then, there is a canonical isomorphism
or
A
X/E
A
or
A
X/F
or
A
X/EF
.
In particular, given A-orientations for E and F, we can construct canonically
an A-orientation for E F.
Proof. Denote p
E
: EF E, p
F
: E F F the canonical projections.
For any open subset U of X, consider the pull-back morphisms
R
U
(E
|U
; A
E
) R
p
1
E
(U)
((E F)
|U
; A
EF
)
R
U
(F
|U
; A
F
) R
p
1
F
(U)
((E F)
|U
; A
EF
).
Combining them with a cup product, we get a canonical morphism
R
U
(E
|U
; A
E
)
L
A
R
U
(F
|U
; A
F
) R
U
((E F)
|U
; A
EF
)
since p
1
E
(U) p
1
F
(U) is the zero section of (E F)
|U
. This gives us a
canonical morphism
(U; or
A
X/E
)
A
(U; or
A
X/F
) (U; or
A
X/EF
)
and consequently a morphism of sheaves of A-modules
or
A
X/E
A
or
A
X/F
or
A
X/EF
.
2.6. Euler classes of inverse images and direct sums 93
To prove that it is an isomorphism, we will work at the level of bers. Using
the commutative diagram
(or
A
X/E
)
x
A
(or
A
X/F
)
x

O
(or
A
X/EF
)
x
O
H
r
{x}
(E
x
; A)
A
H
s
{x}
(F
x
; A)

H
r+s
{x}
(E
x
F
x
; A)
we see that it is sucient to establish that the second horizontal arrow is
an isomorphism. This follows directly from the commutative diagram
H
r
{x}
(E
x
; A)
A
H
s
{x}
(F
x
; A)

O
H
r+s
{x}
(E
x
F
x
; A)
O
H
r
c
(E
x
; A)
A
H
s
c
(F
x
; A)

H
r+s
c
(E
x
F
x
; A)
and K unneth theorem. (There is no torsion problem since both H
r
c
(E
x
; A)
and H
s
c
(F
x
; A) are isomorphic to A.)
Remark 2.6.6. In the situation of the preceding proposition, assume E
and F are oriented. Then, the canonical Z-orientations of E and F induce a
canonical Z-orientationof EF. One checks easily that one can characterize
the corresponding orientation by the fact that if e
1
, . . . , e
r
(resp. f
1
, . . . , f
r
)
is a positively oriented local frames of E (resp. F) then
e
1
, . . . , e
r
, f
1
, . . . , f
r
is a positively oriented local frame of E F.
Proposition 2.6.7. Let E and F be two A-oriented vector bundles of rank
r and s on X. Assume E F is endowed with the A-orientation induced
by that of E and F. Then,
e
A
EF
= e
A
E
e
A
F
.
Proof. It follows from the proof of the preceding proposition that
A
EF
= p
A
E
p
A
F
.
Therefore, denoting s
0,EF
, s
0,E
and s
0,F
the zero sections of E F, E
and F, we get
s
0,EF
(
A
EF
) = s
0,EF
p
A
E
s
0,EF
p
A
F
= s
0,E
A
E
s
0,F
A
F
= e
A
E
e
A
F
.
Exercise 2.6.8. Show that an oriented real vector bundle E on S
n
with
e
Z
E
,= 0 has no proper sub-bundle.
Solution. Let us proceed by contradiction. Assume F is a proper sub-bundle
of rank s of E. Denote G the quotient bundle E/F and t its rank. The
assumption on E combined with the cohomology table
H
k
(S
n
; Z)
_
Z if k = 0, n
0 otherwise,
shows that the rank of E is n. It follows that 0 < s < n and 0 < t < n.
Since
1
(S
n
) 1, F and G are orientable. Moreover, we may choose their
orientations in such a way that
E F G
as oriented real vector bundles. In this case, we have e
Z
E
= e
Z
F
e
Z
G
.
Since both e
Z
F
and e
Z
G
are 0, we get e
Z
E
= 0 contrarily to what has been
assumed.
2.7 Euler classes of normal bundles
Proposition 2.7.1. Let X and Y be topological manifolds of dimension
d
X
and d
Y
. Assume Y is a closed subspace of X and denote i : Y X the
canonical inclusion. Then,
i
1
R
Y
A
X
i
!
A
X
or
A
Y

A
i
1
(or
A
X
)
[d
Y
d
X
].
Proof. We know that
A
X
= or
A
X
[d
X
] and that
A
Y
= or
A
Y
[d
Y
]. Since
A
Y
=
i
!
A
X
, and since or
A
X
is locally isomorphic to A
X
, the conclusion follows.
Denition 2.7.2. In the situation of the preceding proposition, we call
or
A
Y/X
= or
A
Y

A
i
1
(or
A
X
)
the relative A-orientation sheaf of Y in X. A relative A-orientation of Y

in X is the datum of an isomorphism
A
Y
or
A
Y/X
.
The associated relative A-orientation class is the image
A
Y/X
H
0
(Y ; or
A
Y/X
)
2.7. Euler classes of normal bundles 95
of 1
Y
by this isomorphism. The Thom class of a relative A-orientation of
Y in X is the class
A
Y/X
H
d
X
d
Y
Y
(X; A
X
)
corresponding to
A
Y/X
through the isomorphism of the preceding propo-
sition. The restriction of
A
Y/X
to Y is called the Euler class e
A
Y/X
of the
relative A-orientation.
Remark 2.7.3. If, in the situation considered above, X and Y are both
oriented, then we have isomorphisms
Z
X
or
X
, Z
Y
or
Y
and consequently an isomorphism
Z
Y
or
Y/X
.
We leave it to the reader to check that the Thom class of the corresponding
relative orientation of Y in X is
i
!
(1).
In particular, the preceding denition is compatible with Denition 1.14.14.
Proposition 2.7.4. Let Y be a closed dierential submanifold of dimension
d
Y
of the compact dierential manifold X of dimension d
X
. Then, there is
a canonical isomorphism
or
A
Y/X
or
A
Y/T
Y
X
.
In particular, relative A-orientations of Y in X correspond to A-orientations
of the real vector bundle T
Y
X Y . Moreover, for corresponding orienta-
tions, we have
e
A
Y/X
= e
A
T
Y
X
.
Proof. Endow X with a Riemannian metric and identify T
Y
X with the
orthogonal complement of TY in TX
|Y
. For any > 0, set V
= v
y

(T
Y
X)
y
: [v
y
[ < . As is well-known, for suciently small, the map
: V
(V
)
which sends v
y
V
to exp
y
(v
y
) X is a dieomorphism such that (y) = y
for any y Y . It follows that
: R
Y
(A
(V
)
) R
Y
(A
V
)
is an isomorphism. Since
R
Y
(A
(V
)
) R
Y
(A
X
) or
A
Y/X
[d
Y
d
X
]
and
R
Y
(A
V
) R
Y
(A
T
Y
X
) or
A
Y/T
Y
X
[d
Y
d
X
],
: or
A
Y/X
or
A
Y/T
Y
X
(*)
A priori, this isomorphism depends on the chosen Riemannian metric. How-
ever, since two such metrics are homotopic, it is not dicult to see that the
associated maps
0
: V
0
X ,
1
: V
1
X
become homotopic when restricted to an appropriate neighborhood V of Y
in T
Y
X. One can even request that the homotopy h : V [0, 1] X is
such that h(y, t) = y for any y Y . It is then clear that the isomorphisms
0
: or
A
Y/X
or
A
T
Y
X/Y
and
1
: or
A
Y/X
or
A
T
Y
X/Y
associated to
0
and
1
are equal. Moreover, by construction, for corre-
sponding A-orientations, we have
((
A
Y/X
)
|(V
)
) = (
A
T
Y
X
)
|V
.
Hence,
e
A
T
Y
X
= [
((
A
Y/X
)
|(V
)
]
|Y
= (
A
Y/X
)
|Y
= e
A
Y/X
.
Remark 2.7.5. In the preceding proposition, assume X and Y oriented
and take A = Z. Thanks to Remark 2.7.3, there is a canonical relative
orientationof Y in X. By working locally, one checks easily that this relative
orientation corresponds to the orientation of T
Y
X obtained by quotienting
the orientation of (TX)
|Y
by that of TY .
Corollary 2.7.6. Let X be a compact oriented dierential manifold. Then,
X
= e
TX
.
In particular,
(X) =
_
X
e
TX
.
2.7. Euler classes of normal bundles 97
Proof. The orientation of X induces canonical orientations on and XX.
If we give T
(X X) its usual quotient orientation, Remark 2.7.5 and

Proposition 2.7.4 show that
e
/XX
= e
T
(XX)
.
Since the canonical isomorphism
TX

1
T
(X X)
dened by sending T
x
X to the class of (0, ) T
(x,x)
(X X) modulo
T
(x,x)
is compatible with the orientations, we get
e
TX
=
e
T
(XX)
.
The conclusion follows Denition 2.2.1.
Exercise 2.7.7. Show that the tangent bundle TS
n
to an even dimensional
sphere S
n
has no proper sub-bundle.
Solution. Assume E has a proper sub-bundle. It follows from Exercise 2.6.8
that e
Z
TS
n
= 0 and hence that
(S
n
) =
_
e
Z
TS
n
= 0.
But for n even, we have (S
n
) = 1 +(1)
n
= 2 and we get a contradiction.
Exercise 2.7.8. Let p : E X be an oriented dierential real vector
bundle of rank r on an oriented manifold X of dimension n. Assume s :
X E is a section of E transverse to the zero section which vanishes at
some points of X. Then,
Z
s
= x X : s(x) = 0
is a closed oriented dierential submanifold of X of codimension r and the
image of
Z
s
/X
H
r
Z
s
(X; Z)
in H
r
(X; Z) is the Euler class of E.
Solution. It follows from our assumptions that E has a canonical structure
of oriented dierential manifold. Therefore, the zero section s
0
: X E of
E induces canonical morphisms
s
0!
: H
k
(X; Z) H
k+r
X
(E; Z) (k Z).
Since
s
0
!
(c
k
) = s
0
!
(s
0
p
c
k
) = p
c
k
s
0
!
(1)
and s
0
!
(1) =
X/E
=
E
, these morphisms coincide with Thom isomor-
phisms. Therefore, we have only to show that
s
0!
(e
E
) = s
0!
(
Z
s
/X
) =
Z
s
/E
.
Since
T
s(X)
E (p
1
E)
|s(X)
as oriented vector bundles, we have
e
E
= s
e
T
s(X)
E
= s
s(X)/E
= s
s(X)/E
;
the last equality coming from the fact that s
0
and s are homotopic. There-
fore,
s
0
!
(e
E
) = s
0
!
s
0
(
s(X)/E
)
=
s(X)/E
(s
0!
1)
=
s(X)/E

X/E
and the conclusion follows from Proposition 1.14.15.
3
Characteristic classes of
real vector bundles
For the sake of simplicity, all the topological spaces in this chapter will be
implicitly assumed to be paracompact.
3.1 Stiefel-Whitney classes
Lemma 3.1.1 (Leray-Hirsch). Let f : X Y be a proper map. Assume
the classes
g
1
H
k
1
(X; A), , g
n
H
k
n
(X; A)
are such that
g
1
|f
1
(y)
, , g
n
|f
1
(y)
form a free family of generators of H
(f
1
(y); A) as an A-module for any
y Y . Then,
g
1
, , g
n
(X; A) for the H
(Y ; A)-module struc-
ture given by the left action
(, ) f
() .
99
100 3. Characteristic classes of real vector bundles
Proof. Using the isomorphisms
RHom
A
(A
Y
[k], Rf(A
X
)) RHom
A
(f
1
A
Y
, A
X
)[k] R(X; A
X
)[k]
we associate to g
1
, , g
n
canonical morphisms
h
1
: A
Y
[k
1
] Rf(A
X
), , h
n
: A
Y
[k
n
] Rf(A
X
).
This gives us a morphism
h :
n
j=1
A
Y
[k
j
] Rf(A
X
).
Its ber at y Y is the morphism
h
y
:
n
j=1
A[k
j
] [Rf(A
X
)]
y
R(f
1
(y); A)
associated to the cohomology classes g
1
|f
1
(y)
, , g
n
|f
1
(y)
. Therefore, our
assumption ensures that h is an isomorphism. Applying R(Y ; ), we see
that h induces the isomorphism
n
j=1
R(Y ; A
Y
)[k
j
] R(Y ; Rf(A
X
)) R(X; A
X
).
It follows that the morphism of graded A-modules
n
j=1
H
(Y ; A
Y
)[k
j
] H
(X; A
X
)
(
1
, ,
n
) (p
(
1
) g
1
+ +p
(
n
) g
n
)
is an isomorphism, hence the conclusion.
Let E be a vector bundle of rank r based on X. As usual, we dene the
real projective bundle P(E) associated to E by setting
P(E) = (d, x) : x X, d line of E
x
trough 0
x
,
the projection : P(E) X being dened by setting
(d, x) = x.
Of course, P(E)

E/R
where the action of R
on

E is dened berwise
using the real vector space structure of E
x
(x X). As is well-known, P(E)
3.1. Stiefel-Whitney classes 101
has a canonical topology. For this topology, is continuous and using local
frames of E, we see that for any x X there is a neighborhood U of x and
a commutative diagram of the form
1
(U)
/o

P
r1
(R) U
p
U _
U
(*)
In particular, is proper and its bers are homeomorphic to P
r1
(R). Recall
that
E
X
P(E) = (e, (d, x)) : p(e) = x
and denote U(E) the subset of E
X
P(E) dened by
U(E) = (e, (d, x)) : p(e) = x, e d.
One checks easily that the second projection
p
P(E)
: U(E) P(E)
turns U(E) into a line bundle with base P(E). By construction, a homeo-
morphism
1
(U) P
r1
(R) U
of the type (*) transforms U(E)
|
1
(U)
into a line bundle isomorphic to
p
1
P
r1
(R)
U
r1
(R).
Proposition 3.1.2. Let E be a vector bundle of rank r on X. Set
= e
Z
2
U(E)
H
1
(P(E); Z
2
).
Then,
1, , ,
r1
(P(E); Z
2
) as an H
(X; Z
2
)-module.
In other words, any H
(P(E); Z
2
) may be written in a unique way as
=
0
+
1
+ +
r1

r1
with
0
, ,
r1
H
(X; Z
2
).
Proof. We know from Exercise 2.5.6 that the morphism
Z
2
[X]/(X
r
) H
(P
r1
(R); Z
2
)
which sends X to e = e
Z
2
U
r1
(R)
is an isomorphism. It follows that
1, e, , e
r1
is a free family of generators of H
(P
r1
(R); Z
2
) as a Z
2
-module. Since
|
1
(x)
corresponds to e through the canonical isomorphism
U(E)
|p
1
P(E)
(x)
U
r1
(R),
we see that
1,
|
1
(x)
, ,
r1
|
1
(x)
(
1
(x); Z
2
) for any x X and the
conclusion follows from Lemma 3.1.1.
Denition 3.1.3. Let E be a vector bundle of rank r based on the space
X. Using the preceding proposition, we dene the Stiefel-Whitney classes
w
1
(E), , w
r
(E) of E by the formula
r
=
(w
r
(E)) +
(w
r1
(E)) + +
(w
1
(E))
r1
.
We also set w
0
(E) = 1 and w
k
(E) = 0 for k > r. By construction,
w
k
(E) H
k
(X; Z
2
)
for any k N. We also dene the total Stiefel-Whitney class w(E) of E by
the formula
w(E) = w
0
(E) + +w
r
(E) H
(X; Z
2
).
Proposition 3.1.4. Let f : Y X be a morphism of topological spaces.
Assume E is a vector bundle of rank r on X. Then,
w(f
1
(E)) = f
w(E).
Proof. Consider the canonical diagram
f
1
(E)
g
E
p
Y
f
X
Dene P(g) : P(f
1
(E)) P(E) as the map
(d, y) (g(d), f(y))
and U(g) : U(f
1
(E)) U(E) as the map
(e, (d, y)) (g(e), (g(d), f(y))).
We get the commutative diagram
U(f
1
(E))
U(g)
U(E)
P(f
1
(E))
P(g)
P(E)
Y
f
X
Since g
|f
1
(E)
y
: f
1
(E)
y
E
f(y)
is an isomorphism, so is
U(g)
|U(f
1
(E))
(d,y)
: U(f
1
(E))
(d,y)
U(E)
(g(d),f(y))
.
It follows that U(f
1
(E)) P(g)
1
U(E) and hence that
f
1
(E)
= e
Z
2
U(f
1
(E))
= P(g)
e
Z
2
U(E)
= P(g)
E
.
Applying P(g)
to the relation
r
E
=
(w
r
(E)) +
(w
r1
(E))
E
+ +
(w
1
(E))
r1
E
,
we get
r
f
1
(E)
=
(w
r
(E)) +
(w
r1
(E))
f
1
(E)
+
+
(w
1
(E))
r1
f
1
(E)
and the conclusion follows from the denition of the Stiefel-Whitney classes
of f
1
(E).
Proposition 3.1.5. Let E, F be vector bundles of rank r and s on the
topological space X. Then,
w(E F) = w(E) w(F).
Proof. Consider the map
i : E E F
dened at the level of bers by setting
i(e
x
) = (e
x
, 0)
and denote j : P(E) P(E F) the map induced by i. Consider also the
map
p : E F F
dened at the level of bers by setting
p
x
(e
x
, f
x
) = f
x
and denote
q : P(E F) j(P(E)) P(F)
the map induced by p. Finally, denote
: P(E F) j(P(E)) P(E F)
the inclusion map. One checks easily that
j
1
U(E F) U(E)
and that
1
U(E F) q
1
U(F).
Set
=
r
k=0
EF
(w
k
(E))
rk
EF
and
=
s
l=0
EF
(w
l
(F))
sl
EF
.
Clearly, we have
j
=
r
k=0
j
EF
(w
k
(E))j
rk
EF
=
r
k=0
E
(w
k
(E))
rk
E
= 0
and
=
s
l=0
EF
(w
l
(F))
sl
EF
=
s
l=0
q
F
(w
l
(F))q
sl
F
= q
_
s
l=0
F
(w
l
(F))
sl
F
_
= 0.
From the second relation, it follows that = () where
: H
j(P(E))
(P(E F); Z
2
) H
(P(E F); Z
2
)
is the canonical morphism. Using the fact that for
: H
(j(P(E)); Z
2
) H
j(P(E))
(P(E F); Z
2
) H
j(P(E))
(P(E F); Z
2
)
we have
() = (j
),
we see that
= 0
in H
(P(E F); Z
2
). It follows that
r
k=0
s
l=0
EF
(w
k
(E) w
l
(F))
r+skl
EF
= 0
and hence that
r+s
n=0
EF
_
n
k=0
w
k
(E) w
nk
(F)
_
r+sn
EF
= 0.
This shows that
r+s
EF
=
r+s
n=1
EF
([w(E) w(F)]
n
)
r+sn
EF
Exercise 3.1.6. Let E be a vector bundle of rank r on the topological space
X. Show that a necessary condition for E to have a trivial sub-bundle of
rank s is that
w
r
(E) = = w
rs+1
(E) = 0.
In particular, a necessary condition for E to be trivial is that
w(E) = 1.
Solution. Assume rst that E is trivial. It follows that
E R
r
X a
1
X
(R
r
)
where a
X
: X pt is the canonical map. Hence,
w(E) = a
X
w(R
r
) = a
X
w
0
(R
r
) = 1.
Assume now that E has a trivial sub-bundle T of rank s. From the isomor-
phism
E T E/T
we deduce that
w(E) = w(T) w(E/T) = w(E/T).
Since E/T has rank r s, the conclusion follows.
3.2 Splitting principle and consequences
Proposition 3.2.1 (Splitting principle). Let E be a real vector bundle
of rank r on the space X. Then, there is a proper map
f : Y X
for which the canonical map
f
: H
(X; Z
2
) H
(Y ; Z
2
)
is injective and such that
f
1
(E) L
1
L
r
with L
1
, , L
r
real line bundles on Y .
Proof. Let us proceed by induction on r. For r = 1, the result is obvious.
Assume it is true for r 1 and let us prove it for r. We know that
: P(E) X
is a proper map for which
: H
(X; Z
2
) H
(P(E); Z
2
)
is injective. Using the canonical inclusion
U(E) E
X
P(E)
1
(E)
we see that U(E) is a sub-bundle of
1
(E). Setting F =
1
(E)/U(E),
we get the isomorphism
1
E U(E) F.
By the induction hypothesis, there is a proper map
g : Y P(E)
3.2. Splitting principle and consequences 107
for which
g
: H
(P(E); Z
2
) H
(Y ; Z
2
)
g
1
(F) L
1
L
r1
with L
1
, , L
r
line bundles on Y . Setting f = g and L
r
= g
1
(U(E))
allows us to conclude.
Remark 3.2.2. As shown hereafter, the preceding result allows us to reduce
the proof of formulas concerning Stiefel-Whitney classes of vector bundles
to the case of line bundles.
Proposition 3.2.3. Let E be a vector bundle of rank r on X. Then,
w
r
(E) = e
Z
2
E
in H
r
(X; Z
2
).
Proof. By the preceding proposition, we may assume that there is a proper
map
f : Y X
with
f
: H
(X; Z
2
) H
(Y ; Z
2
)
injective and such that
f
1
(E) = L
1
L
r
.
For such a map, we have
f
(w(E)) = w(f
1
(E)) = w(L
1
) w(L
r
).
Moreover, w(L
k
) = 1 + w
1
(L
k
), hence
f
(w
r
(E)) = w
1
(L
1
) w
1
(L
r
).
Since
f
(e
E
) = e
Z
2
f
1
(E)
= e
Z
2
L
1
e
Z
2
L
r
the proof is reduced to the case where r = 1. When E is a line bundle,
: P(E) X is a homeomorphism and
1
E U(E). It follows that
E
=
e
E
. By the denition of Stiefel-Whitney classes, this shows that
w
1
(E) = e
E
, hence the conclusion.
Remark 3.2.4 (Symmetric Polynomials). Let A be a ring with unit.
Recall that the elementary symmetric polynomials of A[X
1
, , X
n
] are the
polynomials S
n,1
, , S
n,n
(X X
1
) (X X
n
) =
n
k=0
(1)
k
S
n,k
(X
1
, , X
n
)X
nk
in A[X
1
, , X
n
, X]. In particular, S
n,1
= X
1
+ + X
n
and S
n,n
=
X
1
X
n
. Recall also that any symmetric polynomial P of degree r of
A[X
1
, , X
n
] may be written in a unique way as Q(S
n,1
, , S
n,n
) with
Q in A[X
1
, , X
n
]. An algorithm to nd Q by induction on n and on the
degree of P is the following:
(a) Write P(X
1
, , X
n1
, 0) as Q
1
(S
n1,1
, , S
n1,n1
).
(b) Find the polynomial R
1
(X
1
, , X
n
) such that
P(X
1
, , X
n
) Q
1
(S
n,1
, , S
n,n1
) = S
n,n
R
1
(X
1
, , X
n
).
(c) Express R
1
(X
1
, , X
n
) as Q
2
(S
n,1
, , S
n,n
).
(d) Write P as
Q
1
(S
n,1
, , S
n,n1
) + S
n,n
Q
2
(S
n,1
, , S
n,n
).
Exercise 3.2.5. Write
P(X
1
, X
2
, X
3
) = X
1
X
2
3
+ X
1
X
2
2
+ X
2
X
2
1
+ X
2
X
2
3
+ X
3
X
2
1
+ X
3
X
2
2
as a polynomial in S
3,1
= X
1
+X
2
+X
3
, S
3,2
= X
1
X
2
+X
1
X
3
+X
2
X
3
and
S
3,3
= X
1
X
2
X
3
.
Solution. Consider the polynomial
X
1
X
2
2
+ X
2
X
2
1
.
Clearly,
X
1
X
2
2
+ X
2
X
2
1
= X
1
X
2
(X
1
+X
2
) = S
2,1
S
2,2
.
We have
P S
3,1
S
3,2
= P (X
1
+X
2
+ X
3
)(X
1
X
2
+ X
1
X
3
+ X
2
X
3
)
= P X
2
1
X
2
X
2
1
X
3
X
1
X
2
X
3
X
1
X
2
2
X
1
X
2
X
3
X
2
2
X
3
X
1
X
2
X
3
X
1
X
2
3
X
2
X
2
3
= 3X
1
X
2
X
3
= 3S
3,3
.
Therefore,
P = S
3,1
S
3,2
3S
3,3
.
Proposition 3.2.6. Let P be a symmetric polynomial of Z
2
[Z
1
, , Z
r
].
Then, to any vector bundle E on X of rank r is canonically associated a
class
w
P
(E) H
(X; Z
2
).
This class is characterized by the fact that
(a)
w
P
(f
1
(E)) = f
w
P
(E)
for any continuous map f : Y X;
(b)
w
P
(E) = P(e
Z
2
L
1
, , e
Z
2
L
r
)
if E = L
1
L
r
with L
1
, , L
r
line bundles.
Proof. Write
P(Z
1
, , Z
r
) = Q(S
r,1
, , S
r,r
)
where Q Z
2
[Z
1
, , Z
r
] and S
r,1
, , S
r,r
are the elementary symmetric
polynomials in r unknowns. Set
w
P
(E) = Q(w
1
(E), , w
r
(E)).
Clearly, if f : Y X is a continuous map, we have
f
w
P
(E) = f
Q(w
1
(E), , w
r
(E))
= Q(w
1
(f
1
(E)), , w
r
(f
1
(E)))
= w
P
(f
1
(E)).
Moreover, if E = L
1
L
r
with L
1
, , L
r
line bundles, we get
w(E) = (1 +
1
) (1 +
r
)
with
1
= e
Z
2
L
1
, ,
r
= e
Z
2
L
r
. It follows that
w(E) = 1 + S
r,1
(
1
, ,
r
) + + S
r,r
(
1
, ,
r
)
and hence that
w
k
(E) = S
r,k
(
1
, ,
r
).
Therefore,
w
P
(E) = Q(w
1
(E), , w
r
(E))
= Q(S
r,1
(
1
, ,
r
), , S
r,r
(
1
, ,
r
))
= P(
1
, ,
r
).
The fact that these two properties characterize w
P
() uniquely follows from
the splitting principle.
Denition 3.2.7. Let E B, F B be two real vector bundles of rank
r and s. Recall that EF denotes the real vector bundle of rank rs dened
by setting
E F = E
b
F
b
the frame of EF on U associated to a frame e
1
, , e
r
of E
|U
and a frame
f
1
, , f
s
of F
|U
being given by the sections
e
i
f
j
(i 1, . . . r, j 1, . . . s).
Proposition 3.2.8. Let p
1
: L
1
B and p
2
: L
2
B be two real line
bundles. Then,
e
Z
2
L
1
L
2
= e
Z
2
L
1
+e
Z
2
L
2
.
Proof. Denote : L
1
L
2
L
1
L
2
the continuous map dened at the
level of bers by setting
b
(e
1
, e
2
) = e
1
e
2
and denote q
1
: L
1
L
2
L
1
, q
2
: L
1
L
2
L
2
the two projections. By
homotopy, we have
R
L
1
(L
1
L
2
; Z
2
) R
B
(L
2
; Z
2
)
and
R
L
2
(L
1
L
2
; Z
2
) R
B
(L
1
; Z
2
).
Hence, we deduce from the Mayer-Vietoris distinguished triangle
R
L
1
L
2
(L
1
L
2
; Z
2
) R
L
1
(L
1
L
2
; Z
2
) R
L
2
(L
1
L
2
; Z
2
) R
L
1
L
2
(L
1
L
2
; Z
2
)
+1
a canonical morphism
H
1
B
(L
1
; Z
2
) H
1
B
(L
2
; Z
2
) H
1
L
1
L
2
(L
1
L
2
; Z
2
). (*)
Since
1
(B) = L
1
L
2
, we also have an isomorphism
R
L
1
L
2
(L
1
L
2
; Z
2
) R
B
(R(Z
2
))
and hence a morphism
H
1
B
(L
1
L
2
; Z
2
) H
1
L
1
L
2
(L
1
L
2
; Z
2
). (**)
Let us prove that the image of the Thom class of L
1
L
2
by (**) is the
image of the Thom classes of L
1
and L
2
by (*). This will give the result
thanks to the commutative diagrams
H
1
(B; Z
2
)
1
H
1
(B; Z
2
)
H
1
(L
1
L
2
; Z
2
)
|B
H
1
(L
1
L
2
; Z
2
)
|B
H
1
B
(L
1
L
2
; Z
2
)
H
1
L
1
L
2
(L
1
L
2
; Z
2
)
and
H
1
(B; Z
2
) H
1
(B; Z
2
)
( 1 1 )
H
1
(B; Z
2
)
H
1
(L
1
; Z
2
) H
1
(L
2
; Z
2
)
|B
( q
1
q
2
)
H
1
(L
1
L
2
; Z
2
)
|B
H
1
B
(L
1
; Z
2
) H
1
B
(L
2
; Z
2
)
( q
1
q
2
)
H
1
L
1
L
2
(L
1
L
2
; Z
2
)
First, note that it follows from Thom isomorphism that

U H
1
U
(L
1
|U
; Z
2
); U H
1
U
(L
2
|U
; Z
2
)
are sheaves on B. Moreover, note also that since
H
0
L
1|U
L
2|U
(L
1|U
L
2|U
; Z
2
) = 0
for any open subset U of B,
U H
1
L
1|U
L
2|U
(L
1
|U
L
2
|U
; Z
2
)
is also a sheaf on B. From these remarks, it follows that our problem is of
local nature and it is sucient to work at the level of bers and to show
that the image of the appropriate Thom classes by canonical maps
H
1
{b}
((L
1
)
b
; Z
2
) H
1
{b}
((L
2
)
b
; Z
2
) H
1
(L
1
)
b
(L
2
)
b
((L
1
)
b
(L
2
)
b
; Z
2
)
and
H
1
{b}
((L
1
)
b
(L
2
)
b
; Z
2
) H
1
(L
1
)
b
(L
2
)
b
((L
1
)
b
(L
2
)
b
; Z
2
)
corresponding to (*) and (**) are equal. Denote : R
2
R the multipli-
cation, the generator of H
1
{0}
(R; Z
2
) and q
1
, q
2
the two projections from
R
2
to R. Using local frames for L
1
and L
2
, we are reduced to show that
() = q
1
() + q
2
() (***)
in H
1
(R{0})({0}R)
(R
2
; Z
2
). We know that is the image of
]0,+[
by the
canonical map
H
0
R\{0}
(R; Z
2
) H
1
{0}
(R; Z
2
).
It follows that
() (resp. q
1
(), q
2
()) is the image of
Q
1
+
Q
3
(resp.
Q
1
+
Q
4
,
Q
1
+
Q
2
) by the canonical map
H
0
R
2
\(R{0})({0}R)
(R
2
; Z
2
) H
1
(R{0})({0}R)
(R
2
; Z
2
).
Here, Q
1
, , Q
4
are the open quarter planes dened in the following gure
Q
1
Q
2
Q
3
Q
4
To establish (***), it is sucient to note that
Q
1
+
Q
4
+
Q
1
+
Q
2
=
Q
2
+
Q
4
=
Q
1
+
Q
3
+
R
2
modulo 2.
Corollary 3.2.9. Let E be the a real vector bundle of rank r on B. Denote
Det(E) the determinant bundle associated to E. Then,
w
1
(E) = e
Z
2
Det E
.
Proof. Thanks to the splitting principle, we may assume that
E L
1
. . . L
r
where L
1
, . . . , L
r
are real line bundles on B. Denote
1
, . . . ,
r
their Euler
classes modulo 2. Since
w(E) = w(L
1
) . . . w(L
r
) = (1 +
1
) . . . (1 +
n
),
we see that
w
1
(E) =
1
+ +
n
.
Since moreover
Det(E) Det(L
1
) . . . Det(L
r
) L
1
. . . L
r
,
the preceding proposition shows that
e
Z
2
Det(E)
= e
Z
2
L
1
+ + e
Z
2
L
r
Lemma 3.2.10. For any r, s N
0
there is a unique polynomial
T
r,s
Z[U
1
, . . . , U
r
, V
1
, . . . V
s
]
such that
r
i=1
s
j=1
(1 + X
i
+ Y
j
) = T
r,s
(S
r,1
(X), . . . , S
r,r
(X), S
s,1
(Y ), . . . , S
s,s
(Y ).
Proof. This follows directly from the fact that
r
i=1
s
j=1
(1 + X
i
+ Y
j
)
is symmetric in the X and Y variables.
Exercise 3.2.11. Compute T
1,s
and T
2,2
explicitly.
Solution. Since
(1 +X
1
+Y
1
) . . . (1 + X
1
+ Y
s
) =
s
k=0
(1 + X
1
)
k
S
s,k
(Y
1
, . . . , Y
s
)
and
(1 + X
1
)
k
=
k
l=0
C
l
k
X
l
1
,
we see directly that
T
1,s
(U, V ) =
s
k=0
k
l=0
C
l
k
U
l
1
V
k
.
The computation of T
2,2
is also easy but more tedious. The nal answer is
1 + 2U
1
+ 2U
2
+ 2V
1
+ 2V
2
+U
2
1
+ 2U
1
U
2
+U
2
2
+ 3U
1
V
1
+ 2U
2
V
1
+ 2U
1
V
2
2U
2
V
2
+V
2
1
+ 2V
1
V
2
+V
2
2
+U
2
1
V
1
+U
1
U
2
V
1
+U
2
1
V
2
+U
1
V
1
V
2
+U
1
V
2
1
+ U
2
V
2
1
Proposition 3.2.12. Assume E (resp. F) is a real vector bundle of rank
r (resp. s) on B. Then,
w(E F) = T
r,s
(w
1
(E), . . . , w
r
(E), w
1
(F), . . . , w
s
(F)).
Proof. Thanks to the splitting principle, we may assume that
E L
1
. . . L
r
and F N
1
. . . N
s
where L
1
, . . . , L
r
and N
1
, . . . , N
s
are real line bundles on X. Denote
1
,
. . . ,
r
and
1
, . . . ,
s
their Euler classes modulo 2. Since
E F
r
i=1
s
j=1
L
i
N
j
,
we have
w(E F) =
r
i=1
s
j=1
w(L
i
N
j
).
Hence, using Proposition 3.2.8, we get
w(E F) =
r
i=1
s
j=1
(1 +
i
+
j
)
and the conclusion follows from the preceding lemma.
Exercise 3.2.13. Let E and F be real plane bundles on B. Show that
w
1
(E F) = 0
w
2
(E F) = w
1
(E)
2
+w
1
(E)w
1
(F) + w
1
(F)
2
w
3
(E F) = w
1
(E)
2
w
1
(F) +w
1
(E)w
1
(F)
2
w
4
(E F) = w
2
(E)
2
+w
1
(E)w
2
(E)w
1
(F) +w
1
(E)
2
w
2
(F)
+w
1
(E)w
1
(F)w
2
(F) + w
2
(F)
2
+w
2
(E)w
1
(F)
2
Solution. This follows by reducing modulo 2 the polynomial T
2,2
computed
in Exercise 3.2.11.
Denition 3.2.14. Let E B, F B be two real vector bundles of rank
r and s. Recall that Hom(E, F) denotes the real vector bundle of rank rs
dened by setting
Hom(E, F) = Hom(E
b
, F
b
)
the local frame of Hom(E, F) on U associated to a frame e
1
, , e
r
of E
|U
and a frame f
1
, , f
s
of F
|U
being given by the morphisms
h
ij
: E
|U
F
|U
h
ji
(e
k
) =
_
0 if k ,= i
f
j
if k = i
If F is the trivial bundle R B B, then Hom(E, F) is simply denoted
E
and is called the dual of E. In particular, the local frame of E
on U
associated to a frame e
1
, , e
r
of E
|U
is the family of sections e
1
, , e
r
of E
on U characterized by the relations
j
, e
i
_
=
ji
.
Remark 3.2.15. One checks easily that the canonical morphisms
E
F Hom(E, F)
is an isomorphism. So, if we know the Stiefel-Whitney classes of E
and F,
we can compute that of Hom(E, F).
Proposition 3.2.16. For any real vector bundle E with base B and rank
r, there is a (non-canonical) isomorphism
E
E.
In particular,
w(E
) = w(E).
Proof. Using a partition of unity, one can construct easily on each ber E
x
an euclidean scalar product (, )
b
in such a way that
b (e
j
, e
k
)
b
is a continuous function on U if (e
1
, , e
r
) is a local frame of E on U.
Using these scalar products, we obtain the requested isomorphism
: E E
by setting
b
(e) = (e, )
b
for any b B and any e E
b
.
Exercise 3.2.17.
(a) Show that
TP
n
(R) Hom(U
n
(R), U
n
(R)
)
where U
n
(R)
b
is the orthogonal complement of U
n
(R)
b
in R
n+1
for
any b P
n
(R).
(b) Deduce from (a) that
w(TP
n
(R)) =
_
1 + e
Z
2
U
n
(R)
_
n+1
.
(c) Prove that
w(TP
n
(R)) = 1
if and only if n = 2
r
1 (r N
0
). As a consequence, show that if
n + 1 is not a power of 2 then P
n
(R) is not parallelizable.
Solution. (a) Let d P
n
(R). By denition, d is a line of R
n+1
containing
the origin. Denote
q : R
n+1
0 P
n
(R)
the canonical map and let v R
n+1
0 be such that q(v) = d. Clearly,
the linear map
T
v
q : R
n+1
T
d
P
n
(R)
has d as kernel. Hence,
T
v
q
|d
: d
T
d
P
n
(R)
is an isomorphism. Moreover, since
Hom(U
n
(R), U
n
(R)
)
d
= Hom(U
n
(R)
d
, U
n
(R)
d
)
Hom(d, d
)
v induces a canonical isomorphism
v
: Hom(U
n
(R), U
n
(R)
)
d
d
.
This isomorphism sends h : d d
to h(v). By composition, we get the

isomorphism
T
v
q
v
: Hom(U
n
(R), U
n
(R)
)
d
T
d
P
n
(R). (*)
Since
q(v) = q(v)
for ,= 0, the chain rule shows that
T
q
v
( ) = T
q
v
()
for any R
n+1
. Therefore, (*) does not depend on v and we get a
Hom(U
n
(R), U
n
(R)
)
d
T
d
(P
n
(R)).
One checks easily that it extends to the requested isomorphism of vector
bundles.
(b) Since
U
n
(R) U
n
(R)
= R
n+1
P
n
(R)
we see that
Hom(U
n
(R), U
n
(R)) Hom(U
n
(R), U
n
(R)
)
Hom(U
n
(R), R
n+1
P
n
(R))
and therefore that
(R P
n
(R)) TP
n
(R) [U
n
(R)
]
n+1
.
It follows that
1 w(TP
n
(R)) = w(U
n
(R))
n+1
,
hence the conclusion.
(c) Assume n = 2
r
1 with r N. It follows that
w(TP
n
(R)) =
_
1 +e
Z
2
U
n
(R)
_
2
r
= 1 +
_
e
Z
2
U
n
(R)
_
2
r
since we work modulo 2. But 2
r
= n + 1 > dimP
n
(R) and we get
w(TP
n
(R)) = 1.
Assume now n is not of the preceding type. Then, n = 2
r
m1 where m is
an odd number which is strictly greater than 1. In this case 2
r
n and
w(TP
n
(R)) =
_
1 +
_
e
Z
2
U
n
(R)
_
2
r
_
m
= 1 + m
_
e
Z
2
U
n
(R)
_
2
r
+
Since m , 0 (mod 2) and 2
r
dimP
n
(R), it follows that w(TP
n
(R)) ,= 1.
To conclude, it remains to note that if P
n
(R) is parallelizable then TP
n
(R)
is trivializable and we have w(TP
n
(R)) = 1.
Exercise 3.2.18.
(a) Show that if a manifold M of dimension m can be immersed in R
m+s
then the components of degree k > s of w
1
(TM) vanish.
(b) Deduce from (a) that P
2
r (R) cannot be immersed in R
2
r
+s
with s <
2
r
1.
Solution. (a) Let i : M R
m+s
be an immersion. Consider the associated
exact sequence of bundles
0 TM i
1
TR
m+s
T
M
R
m+s
0.
Clearly,
1 = w(i
1
TR
m+1
) = w(TM)w(T
M
R
m+s
).
It follows that w(T
M
R
m+s
) = w
1
(TM) and since T
M
R
m+s
has rank s, the
conclusion follows.
(b) For n = 2
r
, setting e = e
Z
2
U
n
(R)
, we have
w(TP
n
(R)) = (1 +e)
2
r
+1
.
Therefore,
w
1
(TP
n
(R)) = (1 +e +e
2
+ +e
n
)
2
r
+1
.
Using the fact that (a +b)
2
= a
2
+b
2
in H
(P
n
(R); Z
2
), we get
w
1
(TP
n
(R)) = (1 + e + e
2
+ +e
n
)(1 +e
2
r
)
= 1 +e + e
2
+ +e
n1
3.3 Homotopical classication of real vector bundles
Denition 3.3.1. Hereafter, we denote G
n,r
the Grassmannian formed by
real vector subspaces of dimension r of R
n
.
Proposition 3.3.2. The Grassmannian G
n,r
has a canonical structure of
compact topological manifold of dimension r(n r).
Proof. Denote V
n,r
the subset of (R
n
)
r
formed by the sequences (v
1
, , v
r
)
which are linearly independent. Since (v
1
, , v
r
) V
n,r
if and only if there
is i
1
, i
r
1, , n such that
v
1i
1
v
ri
1
.
.
.
.
.
.
.
.
.
v
1i
r
v
ri
r
,= 0
3.3. Homotopical classication of real vector bundles 119
it is clear that V
n,r
is an open subset of (R
n
)
r
. Clearly, GL
r
(R) acts freely
and continuously on the right on V
n,r
. Denote q : V
n,r
G
n,r
the map
which sends a sequence (v
1
, , v
r
) to its linear envelope and endow G
n,r
with the associated quotient topology. By construction,
q(v
1
, , v
r
) = q(v
1
, , v
r
)
if and only if there is M GL
r
(R) such that
(v
1
, , v
r
) = (v
1
, , v
r
) M
or equivalently if and only if
rk(v
1
, , v
r
, v
1
, , v
r
) r. (*)
It follows that q identies V
n,r
/GL
r
(R) with G
n,r
. In particular, q is open
and since condition (*) is clearly closed we see that G
n,r
is separated. Let
L
0
be an element of G
n,r
and let L
1
be a supplement of L
0
in R
n
. Denote
p : R
n
L
0
the projection associated to the decomposition R
n
= L
0
L
1
.
Set U = L G
n,r
: p(L) = L
0
. Clearly, U is an open subset of G
n,r
.
Moreover, since p
|L
: L L
0
is bijective, any L U may be viewed as the
graph of a linear map h : L
0
L
1
. This gives us a bijection
U Hom(L
0
, L
1
)
which is easily checked to be an homeomorphism. Since Hom(L
0
, L
1
)
R
r(nr)
, it follows that G
n,r
is a topological manifold of dimension r(nr).
To prove that it is compact, remark that G
n,r
= q(S
n,r
) where S
n,r
is the
compact subset of V
n,r
formed by orthonormed sequences.
Remark 3.3.3. The reader will easily complete the preceding proposition
to show that G
n,r
has in fact a canonical structure of dierential manifold.
Denition 3.3.4. We dene U
n,r
as the subset of R
n
G
n,r
formed by the
pairs (V, L) with V L.
Proposition 3.3.5. The canonical projection U
n,r
G
n,r
is a real vector
bundle of rank r.
Proof. The only non obvious part is to show that U
n,r
has locally a contin-
uous frame. Using the notations introduced in the proof of the preceding
proposition, we construct such a frame by choosing a basis v
1
, , v
r
of L
0
and associating to any L U the basis p
1
|L
(v
1
), , p
1
|L
(v
r
).
Proposition 3.3.6. Let B be a compact topological space. Assume E is a
vector bundle of rank r on X. Then, there is n N and a continuous map
f : B G
n,r
such that f
1
(U
n,r
) E.
Proof. It is sucient to construct a commutative diagram of continuous
maps of the type
E
g
p
E
U
n,r

p
U
n,r
R
n
G
n,r
B
f
G
n,r
with g
|E
b
injective. Note that in such a diagram f is determined by g.
Moreover, if we denote q
1
, q
2
the two projections of R
n
G
n,r
on R
n
and
G
n,r
, we have (q
2
g)(e) = (q
1
g)(E
p
E
(e)
). Therefore, we have only to
construct the continuous map

f := q
1
g : E R
n
in such a way that

f
|E
b
is injective and linear. To construct this map, let us proceed as follows. We
cover B by a nite number of open subsets U
1
, , U
N
on which E has a
continuous frame. For each k 1, , N we choose a trivialization
U
k
: E
|U
k
R
r
U
k
and set h
U
k
= q
1

U
k
. Denoting (
U
1
, ,
U
N
) a partition of unity
subordinated to the covering U
1
, , U
N
, we set
f(e) = ((
U
1
p)(e)h
U
1
(e), , (
U
N
p)(e)h
U
N
(e)) R
rN
.
It is clear that

f is continuous and linear on the bers of E. It is also
injective on the bers of E. As a matter of fact, if e
1
, e
2
E
b
are such that
f(e
1
) =

f(e
2
), there is k 1 , N such that (
U
k
p)(b) ,= 0. For such
a k, we get
h
U
k
(e
1
) = h
U
k
(e
2
)
and hence e
1
= e
2
.
Denition 3.3.7. We denote R
the real vector space formed by the se-

quences
(x
n
)
nN
of real numbers for which n N : x
n
,= 0 is nite and endow it with the
topology of the inductive limit
lim
nN
R
n
corresponding to the transition morphisms
t
m,n
: R
n
R
m
being dened by setting
t
m,n
(x) = (x, 0).
We denote V
,r
the topological subspace of (R
)
r
formed by the sequences
(v
1
, , v
r
)
which are linearly independent. We denote G
,r
the set of r-dimensional
vector subspace of R
and endow it with the quotient topology associated

with the map
q : V
,r
G
,r
which sends a sequence (v
1
, , v
r
) to its linear envelope. Finally, we denote
U
,r
the subset of R
G
,r
formed by the pairs (v, L) with v L.
Lemma 3.3.8. Let (X
n
, x
m,n
)
nN
be an inductive system of locally com-
pact topological spaces. Assume that Y
n
is a closed subspace of X
n
and
that
x
1
m,n
(Y
m
) = Y
n
and denote y
m,n
the map x
m,n
|Y
n
: Y
n
Y
m
. Then, (Y
n
, y
m,n
)
nN
is an
inductive system and lim
nN
Y
n
is a closed subspace of lim
nN
X
n
.
Proof. Set X = lim
nN
X
n
and Y = lim
nN
Y
n
and identify Y with a subset of X.
Denote x
n
: X
n
X, y
n
: Y
n
Y the canonical maps.
(a) Let v Y and let V be an open neighborhood of v in Y . We have
to show that there is a neighborhood U of v in X such that U Y V .
We know that v = y
n
(v
n
) for some n N and that V
n
:= y
1
n
(V ) is a
neighborhood of v
n
in Y
n
. Let K
n
be a compact neighborhood of v
n
in
X
n
such that K
n
Y
n
V
n
and let us construct by induction a sequence
(K
m
)
mn
such that
(i) K
m
is a compact neighborhood of v
m
:= y
m,n
(v
n
) in X
m
,
(ii) K
m
Y
m
V
m
:= y
1
m
(V ),
(iii) x
m+1,m
(K
m
) K
m+1
.
This is possible since if K
m
is a compact neighborhood of v
m
in X
m
such
that K
m
Y
m
V
m
, we can construct K
m+1
as follows. Using the fact that
x
m+1,m
(K
m
)Y
m+1
is a compact subset of X
m+1
included in V
m+1
together
with the fact that any w
m+1
V
m+1
has a compact neighborhood W
m+1
with W
m+1
Y
m+1
V
m+1
, it is easy to obtain a compact subset L
m+1
of
X
m+1
such that L
m+1
Y
m+1
V
m+1
, L
m+1
x
m+1,m
(K
m
)Y
m+1
. Since
x
m+1,m
(K
m
)L
m+1
is a compact subset of X
m+1
disjoint from Y
m+1
, it has
a compact neighborhood L
m+1
disjoint from Y
m+1
. Taking K
m+1
= L
m+1
m+1
, we see that K
m+1
Y
m+1
V
m+1
and that K
m+1
x
m+1,m
(K
m
) as
requested. Now, set K = lim
mn
K
m
. By construction, KY V . Moreover,
since x
m+1,m
(K
m
) K
m+1
, K is an open neighborhood of v in X.
(b) Let u X Y . We know that u = x
n
(u
n
) for some n N and
that for any m n, u
m
:= x
m,n
(u
n
) , Y
m
. Working as in (a), it is clearly
possible to construct by induction a sequence (K
m
)
mn
such that
(i) K
m
is a compact neighborhood of u
m
in X
m
,
(ii) K
m
Y
m
= ,
(iii) x
m+1,m
(K
m
) K
m+1
.
Set K = lim
mn
K
m
. By construction, KY = and K is an open neighbor-
hood of u; hence the conclusion.
Lemma 3.3.9. Let (X
n
, x
m,n
)
nN
and (Y
n
, y
m,n
)
nN
be inductive systems
of locally compact topological spaces. Then,
lim
nN
(X
n
Y
n
) (lim
nN
X
n
) (lim
nN
Y
n
)
as topological spaces.
Proof. Set Z
n
= X
n
Y
n
and set
X = lim
nN
X
n
, Y = lim
nN
Y
n
, Z = lim
nN
Z
n
.
Denote x
n
: X
n
X, y
n
: Y
n
Y , z
n
: Z
n
Z the canonical maps. Since
the canonical continuous map
i : Z X Y
is clearly bijective, it remains to show that if W is a neighborhood of w in
Z and i(w) = (u, v), then there are neighborhoods U, V of u, v in X and
Y such that i(W) U V . We know that w = z
n
(w
n
) for some n N
and some w
n
Z
n
. Set w
m
= z
m,n
(w
n
) and W
m
= z
1
m
(W) for any m n.
Let u
n
X
n
, v
n
Y
n
be such that i(w
n
) = (u
n
, v
n
). Since i(W
n
) is a
neighborhood of (u
n
, v
n
) in X
n
Y
n
, there are compact neighborhoods K
n
,
L
n
of u
n
, v
n
in X
n
and Y
n
such that i(W
n
) K
n
L
n
. Starting with these
neighborhoods, let us construct by induction sequences (K
m
)
mn
, (L
m
)
mn
such that
(i) K
m
(resp. L
m
) is a compact neighborhood of u
m
(resp. v
m
),
(ii) K
m
L
m
i(W
m
),
(iii) x
m+1,m
(K
m
) K
m+1
, y
m+1,m
(L
m
) L
m+1
.
This is possible since i(W
n+1
) is a neighborhood of
x
m+1,m
(K
m
) y
m+1,m
(L
m
).
Set
K = lim
mn
K
m
, L = lim
mn
L
m
.
By construction, K and L are open subsets of X and Y and i(W) KL.
Proposition 3.3.10. We have
V
,r
lim
nr
V
n,r
,
G
,r
lim
nr
G
n,r
,
U
,r
lim
nr
U
n,r
.
In particular, V
,r
is an open subset of R
and the canonical projection

U
,r
G
,r
is a real vector bundle of rank r.
Proof. Thanks to Lemma 3.3.9, (R
)
r
lim
nN
(R
n
)
r
and it follows from the
fact V
,r
(R
n
)
r
= V
n,r
that V
,r
is open in (R
)
r
and that
V
,r
lim
nr
V
n,r
.
The relation
G
,r
lim
nr
G
n,r
follows easily. Since U
n,r
is a closed subspace of R
n
G
n,r
, the relation
U
,r
= lim
nr
U
n,r
follows directly from Lemma 3.3.8 together with the relation
R
G
,r
= lim
nr
R
n
G
n,r
which follows from Lemma 3.3.9.
Remark 3.3.11. Although we will not prove it here, the space G
,r
is
paracompact (see e.g. [20]).
Lemma 3.3.12. Let B be a topological space and let E be a real vector
bundle of rank r on B. Then, there is a locally nite countable family
(V
k
)
kN
of open subsets of B such that
(a)
kN
V
k
= B;
(b) E
|V
k
R
r
V
k
.
Proof. Let | be a locally nite covering of B by open subsets U on which
E is trivializable and let (
U
)
UU
be continuous partition of unity subor-
dinated to |. For any non-empty nite subset o of |, set
V (o) = b B[ inf
US
U
(b) > sup
UU\S
U
(b).
Set also
V
k
=
_
#S=k
V (o).
One checks easily that V (o) is an open subset of

o. It follows that E is
trivializable on V (o) and that (V
k
)
kN
is a locally nite family. Since
V (o) V (o
) ,=
entails that o o
or o
o, we see that V (o) and V (o
) are disjoint if o
and o
are distinct nite subsets of | with k > 0 elements. It follows that

E is trivializable on V
k
. To conclude, note that we have
_
kN
V
k
= B
since any b B belongs to V (o) with
o = U | :
U
(b) > 0.
Theorem 3.3.13. For any topological space B and any real vector bundle
E of rank r on B, there is a continuous map
f : B G
,r
such that f
1
(U
,r
) E. Moreover, such a map is unique up to homotopy.
Proof. Thanks to the preceding Lemma, the rst part is obtained by work-
ing as in the proof of Proposition 3.3.6.
For the second part, we have to show that given two commutative dia-
grams of the type
E
g
0
U
,r

G
,r
_
E
g
1
U
,r

G
,r
_
B
f
0
G
,r
B
f
1
G
,r
where g
0
and g
1
are injective on the bers of E, there is a homotopy between
f
0
and f
1
. Setting

f
0
= q
1
g
0
and

f
1
= q
1
g
1
, we get two maps
f
0
: E R
,

f
1
: E R
which are linear and injective on the bers of E and whose knowledge makes
it possible to reconstruct the whole diagrams.
(a) Assume

f
0
(e) ,
f
1
(e) : < 0 for e

E. Then, for t [0, 1], the
continuous map
h
t
: E R
dened by setting
f
t
(e) = (1 t)
f
0
(e) + t
f
1
(e)
is linear and injective on the bers of E. Hence it gives rise to a commutative
diagram
E
g
t
U
,r

G
,r
_
B
f
t
G
,r
here q
1
g
t
=
h
t
and g
t
is injective on the bers of E. The conclusion follows
since f
t
(t [0, 1]) is a homotopy between f
0
and f
1
.
(b) In general, consider the maps
: R
, : R
dened by setting
(e
i
) = e
2i
, (e
i
) = e
2i+1
where e
i
are the vectors of the canonical basis of R
. Set

f
=

f
0
and
=

f
1
and consider the associated diagrams
E
g
U
,r

G
,r
_
E
g
U
,r

G
,r
_
B
f
G
,r
B
f
G
,r
here q
1
g
= g
and q
1
g
= g
. By (a), we have f
0
f
, f
,
f
f
1
; hence the conclusion.
Corollary 3.3.14. Assume B is a topological space. Then, the map
[B, G
,r
] Isom(1ect
r
R
(B))
which associates to the homotopy class of f : B G
,r
the isomorphism
class of f
1
(U
,r
) is a bijection.
Proof. Since Lemma 3.3.15 shows that the map considered above is well-
dened, the conclusion follows easily from the preceding result.
Lemma 3.3.15. Assume B is a topological space. Then, for any real vector
bundle F on B [0, 1], there is a real vector vector bundle E on B and a
non-canonical isomorphism
F p
1
B
E.
In particular, if h : B I B
is a continuous homotopy and E
is a real
vector bundle on B
, then the isomorphy class of

h
1
t
E
does not depend on t [0, 1].

Proof. See e.g. [16, p. 28].
3.4 Characteristic classes
Denition 3.4.1. A characteristic class with coecients in the abelian
group M for real vector bundles of rank r is a law which associates to any
real vector bundle E of rank r and base X a class (E) H
(X; A) in such
a way that
(f
1
(E)) = f
(E)
for any continuous map f : Y X.
3.4. Characteristic classes 127
Proposition 3.4.2. Characteristic classes with coecients in M for real
vector bundles of rank r are canonically in bijection with
H
(G
,r
; M).
Proof. Thanks to Theorem 3.3.13, we know that any real vector bundle E
of rank r on B may be written as
f
1
(U
,r
)
where f : B G
,r
is uniquely determined up to homotopy by E. If is
a characteristic class, it follows that
(E) = f
(U
,r
)
and is uniquely determined by (U
,r
) H
(G
,r
; M). Moreover, if
c H
(G
,r
; M)
is given, we can construct a characteristic class such that (U
,r
) = c by
setting
(E) = f
c
since f
depends only on E.
Lemma 3.4.3. Let X be a topological space. Assume (A
i
)
iI
is a directed
family of subspaces of X such that
X lim
iI
A
i
.
Then, for any sheaf T on X, we have
(X; T) lim
iI
(A
i
; T).
Proof. The fact that the canonical map
(X; T) lim
iI
(A
i
; T)
is injective is a direct consequence of the fact that X =

iI
A
i
. Let us
prove that it is also surjective. Let (
i
)
iI
be an element of lim
iI
(A
i
; T).
Dene the family (s
x
)
xX
of

xX
T
x
by setting
s
x
= (
i
)
x
for any x A
i
. We have to prove that (s
x
)
xX
comes locally from a
section of T. Fix x X. There is i I such that x A
i
, an open
neighborhood U of x in X and T(U) such that
|A
i
U
=
i
|A
i
U
. Set
V = x U :
x
= s
x
. Clearly,
V A
j
= x U A
j
:
x
= (
j
)
x
is open in A
j
. Therefore, V is open in X and the conclusion follows.
Lemma 3.4.4. Let X be a topological space and let (F
n
)
nN
be an in-
creasing sequence of closed subspaces of X such that
X lim
nN
F
n
.
Then, for any sheaf T on X, we have
R(X; T) R lim
nN
R(F
n
; T).
In particular, if
H
k
(F
n
; T)
satises Mittag-Leer condition for k l, then
H
l
(X; T) lim
nN
H
l
(F
n
; T).
Proof. It is sucient to apply Lemma 3.4.3 to a soft resolution of T and to
use the fact that Mittag-Leer condition implies lim
acyclicity.
Proposition 3.4.5. For n r, the Z
2
-algebra H
(G
n,r
; Z
2
) is generated by
w
1
(U
n,r
), , w
r
(U
n,r
).
Proof. Recall that the universal bundle of rank r
p : U
n,r
G
n,r
is dened by setting
U
n,r
= (v, L) R
n
G
n,r
: v L.
Dene
U
n,r
= (v, L) R
n
G
n,r
: v L
and denote
q : U
n,r
G
n,r
the canonical projection. One can check easily that q is a real vector bundle
of rank n r. Denote

U
n,r
(resp.

U
n,r
) the space U
n,r
(resp. U
n,r
) with-
out its zero section and denote p (resp. q) the canonical projection of

U
n,r
(resp.

U
n,r
) on G
n,r
. Dene the map
f :

U
n,r

U
n,r1
by setting f(v, L) = (v, L)v
) for v R
n
0. One checks directly that
this application is a homeomorphism; the inverse being given by the map
f
1
:

U
n,r1

U
n,r
dened by setting
f
1
(v, L
) = (v, L
+)v)
for any v L
0. It follows that the Gysin sequences for U

n,r
and
U
n,r1
:
H
kr
(G
n,r
; Z
2
)
e
Z
2
U
n,r
H
k
(G
n,r
; Z
2
)
p
H
k
(

U
n,r
; Z
2
)
H
kn+r1
(G
n,r1
; Z
2
)
e
Z
2
U
n,r1
H
k
(G
n,r1
; Z
2
)
q
H
k
(

U
n,r1
; Z
2
)
are connected by the isomorphism
f
: H
k
(

U
n,r1
; Z
2
) H
k
(

U
n,r
; Z
2
).
Let us prove that
p
(w(U
n,r
)) = f
( q
(w(U
n,r1
))). (*)
To this end, consider the commutative diagram
U
n,r
qf
G
n,r1
p
1
U
n,r
g

U
n,r1
where the map g is dened by setting

g(v
, v, L) = (v
, L )v
)
where L G
n,r
, v L 0, v
L and v
is the orthogonal projection of

v
on )v
. This gives us a morphism

p
1
U
n,r
( q f)
1
U
n,r1
which is surjective. Its kernel is given by
L
n,r
= (v
, v, L) : (v, L)

U
n,r
, v
)v
which is a trivializable line bundle on

U
n,r
. From the exact sequence
0 L
n,r
p
1
U
n,r
( q f)
1
U
n,r1
0
we get that
w( p
1
U
n,r
) = w(L
n,r
) w(( q f)
1
U
n,r1
) = w(( q f)
1
U
n,r1
)
and hence that
p
w(U
n,r
) = f
( q
(w(U
n,r1
))).
We will now prove by increasing induction on k 0 and r 1 that
for n r, any c
k
H
k
(G
n,r
; Z
2
) may be expressed as a polynomial
R(w
1
(U
n,r
), , w
r
(U
n,r
)) in the Stiefel-Whitney classes of U
n,r
. The start-
ing point will be the case k = 0, r = 1 which is obvious. From the two exact
triangles
H
(G
n,r
; Z
2
)
e
Z
2
U
n,r
(G
n,r
; Z
2
)
p

(

U
n,r
; Z
2
)
.
(G
n,r1
; Z
2
)
e
Z
2
U
n,r1
(G
n,r1
; Z
2
)
q
.
.
.
.
.
.
.
H
(

U
n,r1
; Z
2
)
.
and Lemma 3.4.6 below, we deduce that

dimIm p
=
1
2
dimH
(

U
n,r
; Z
2
)
=
1
2
dimH
(

U
n,r1
; Z
2
)
= dimIm q
= dimIm(f
).
Moreover, it follows from (*) and the induction hypothesis that
Im p
Im(f
).
Putting these two facts together, we see that Im p
= Im(f
). Let us
x a class c in H
k
(G
n,r
; Z
2
). From what precedes, we know that there is
c
H
k
(G
n,r1
; Z
2
) such that
p
(c
) = f
( q
(c)).
By the induction hypothesis, there is a polynomial relation
c
= R
(w
1
(U
n,r1
), , w
r1
(U
n,r1
)).
Using (*), it follows that
p
(c) = p
(w
1
(U
n,r
), , w
r1
(U
n,r
))
and hence that
c R
(w
1
(U
n,r
), , w
r1
(U
n,r
)) = c
e
Z
2
U
n,r
= c
w
r
(U
n,r
)
with c
H
k2
(G
n,r
; Z
2
). By the induction hypothesis,
c
= R
(w
1
(U
n,r
), , w
r
(U
n,r
))
Lemma 3.4.6. Let
E
u
E
v ,
F
w
be an exact triangle of vector spaces (i.e. Ker w = Imv, Ker v = Imu,

Ker u = Imw). Assume E has nite dimension. Then, F has nite dimen-
sion and
dimImv =
1
2
dimF.
Proof. We know that
Imv Coker u, Coker v Imw Ker u.
Since
dimImu + dimKer u = dimE,
we have
dimCoker u = dimKer u.
It follows that
dimImv = dimCoker v < +.
Hence,
dimF = dimImv + dimCoker v = 2 dimImv.
Proposition 3.4.7. The canonical morphism
Z
2
[W
1
, , W
r
] H
(G
,r
; Z
2
)
which sends W
k
w
k
(U
,r
) is an isomorphism.
Proof. Using the notations introduced in the proof of Proposition 3.4.5,
we get from the Gysin sequences that, for k < n r, we have the exact
sequences
0 H
k
(G
n,r1
; Z
2
)
q
H
k
(

U
n,r1
; Z
2
) 0
and
H
kr
(G
n,r
; Z
2
)
e
Z
2
U
n,r
H
k
(G
n,r
; Z
2
)
p
H
k
(

U
n,r
; Z
2
)
It follows from the equality Im(f
) = Im p
that p
is surjective in
degree k < n r. Hence, we get the exact sequence
0 H
kr
(G
n,r
; Z
2
)
e
Z
2
U
n,r
H
k
(G
n,r
; Z
2
)

H
k
(G
n,r1
; Z
2
) 0 (*)
where = ( q
)
1
f
1
p
. Thanks to the formula (*) of the proof of

Proposition 3.4.5 we have (w(U
n,r
)) = w(U
n,r1
). Since U
n,r
|G
n1,r
=
U
n1,r
, it follows from Proposition 3.4.5 that the restriction map
H
k
(G
m,r
; Z
2
) H
k
(G
n,r
; Z
2
)
is surjective for m n. Hence, Lemma 3.4.4 shows that
H
k
(G
,r
; Z
2
) lim
nr
H
k
(G
n,r
; Z
2
).
Taking the projective limit of the sequences (*) and using the fact that the
Mittag-Leer condition implies lim
-acyclicity, we get the exact sequence

0 H
kr
(G
,r
; Z
2
)
e
Z
2
U
,r
H
k
(G
,r
; Z
2
) H
k
(G
,r1
; Z
2
) 0.
Working by induction, as in the proof of Proposition 3.4.5, we see easily
that any c H
k
(G
,r
; Z
2
) may be written polynomially as
c = R(w
1
(U
,r
), , w
r
(U
,r
)).
To show the uniqueness of such a writing, let us proceed by induction on
k 0 and r 1 as follows. Assume
R(w
1
(U
,r
), , w
r
(U
,r
)) = S(w
1
(U
,r
), , w
r
(U
,r
)). (*)
3.5. Cohomological classication of real vector bundles 133
Write
R(w
1
, , w
r
) = R
(w
1
, , w
r1
) + w
r
R
(w
1
, , w
r
)
S(w
1
, , w
r
) = S
(w
1
, , w
r1
) + w
r
S
(w
1
, , w
r
).
Applying to (*), we see that
R
(w
1
(U
,r1
), , w
r1
(U
,r1
)) = S
(w
1
(U
,r1
), , w
r1
(U
,r1
)).
By the induction hypothesis, we get R
= S
and, hence,
w
r
(U
,r
) R
(w
1
(U
,r
), , w
r
(U
,r
))
= w
r
(U
,r
) S
(w
1
(U
,r
), , w
r
(U
,r
)).
Since w
r
(U
,r
) = e
Z
2
U
,r
and e
Z
2
U
,r
is injective, the conclusion follows
by the induction hypothesis.
3.5 Cohomological classication of real vector bundles
Denition 3.5.1. Let G be a topological group and let | be an open
covering of the topological space X. A continuous

Cech 1-cochain on |
with values in G is the datum for any U, V | of a continuous map
V U
: U V G.
Such a cochain is a cocycle if
WV

V U
=
WU
on U V W.
Two continuous 1-cocycles (
UV
)
V,UU
, (
V U
)
V,UU
are equivalent if we
can nd for any U | a continuous map
U
: U G
in such a way that
V U
=
V

V U

1
U
on U V.
The set of equivalence classes of continuous 1-cocycles on | is denoted
H
1
cont
(|; G).
If 1 is an open covering of X such that 1 |, there is a canonical restriction
map
H
1
cont
(|; G)

H
1
cont
(1; G).
These restriction maps turn the family

H
1
cont
(|; G) (| open covering of X)
into an inductive system and we set
H
1
cont
(X; G) = lim
H
1
cont
(|; G).
Proposition 3.5.2. For any topological space B, there is a canonical bi-
jection
Isom(1ect
r
R
(B))

H
1
cont
(B; GL
r
(R)).
Proof. Let E be a real vector bundle with base B and rank r. By denition,
there is an open covering | of B such that E
|U
has a continuous frame for
any U |. This gives us a family of trivializations
U
: E
|U
R
r
U (U |).
For any U, V |,
V

1
U
: R
r
(U V ) R
r
(U V )
is an isomorphism of real vector bundles. Therefore, the map
v q
2
_
(
V

1
U
)(v, x)
denes an element
V U
(x) of GL
r
(R). Moreover, it is clear that
V U
: U V GL
r
(R)
is continuous and that
WV
(x)
V U
(x) =
WU
(x)
for any x U V W. It follows that (
V U
)
U,V U
is a continuous 1-
cocycle on | with values in GL
r
(R). We leave it to the reader to check that
its class in

H
1
cont
(X; GL
r
(R)) depends only on the isomorphy class of E. As
a consequence, we get a well-dened map
Isom(1ect
r
R
(B))

H
1
cont
(X; GL
r
(R)).
Its injectivity is almost obvious. To prove its surjectivity, it is sucient to
consider a continuous 1-cocycle (
V U
)
U,V U
on | with values in GL
r
(R)
and to show that its equivalence class is the class associated to the real
vector bundle obtained by gluing together the family of trivial bundles
R
r
U
through the transition isomorphisms
R
r
(U V ) R
r
(U V )
(v, x) (
V U
(x)v, x).
Details are left to the reader.
Lemma 3.5.3. For any x R
, set
s(x) =
_
0 if x > 0
1 if x < 0.
Then,
s : R
Z
2
is a morphism of groups and the sequence of abelian groups
0 R
exp
R
s
Z
2
0
is exact. Moreover, for any topological space X, this sequence induces the
exact sequence of sheaves
0 (
R
X
(
R
X
(Z
2
)
X
0
where (
R
X
denotes the sheaf of real valued continuous functions.
Proof. Direct.
Proposition 3.5.4. Let B be a topological space. The exact sequence
0 (
R
B
(
R
B
(Z
2
)
B
0
H
1
(B; (
R
B
) H
1
(B; Z
2
).
There is a bijection
H
1
cont
(B; GL
1
(R)) H
1
(B; (
R
B
).
The associated bijection
Isom(1ect
1
R
(B)) H
1
(B; Z
2
)
may be realized by the map
L e
Z
2
L
.
Proof. Taking the long exact sequence of cohomology associated to the se-
quence of sheaves
0 (
R
B
(
R
B
(Z
2
)
B
0,
H
1
(B; (
R
B
) H
1
(B; (
R
B
) H
1
(B; Z
2
) H
2
(B; (
R
B
).
Since (
R
B
is soft, the rst and last terms vanish. Hence the isomorphism
H
1
(B; (
R
B
) H
1
(B; Z
2
).
Clearly, there is a bijection
H
1
cont
(B; GL
1
(R))

H
1
(B; (
R
B
)
and using the isomorphism between

Cech cohomology and ordinary coho-
mology, we get a bijection
H
1
cont
(B; GL
1
(R)) H
1
(B; (
R
B
).
Combining this with the bijection of Proposition 3.5.2, we get a bijection
Isom(1ect
1
R
(B)) H
1
(B; Z
2
). (*)
What remains to prove is that this bijection may be realized by the Euler
class. Since this bijection is clearly compatible with the pull-back of bundles
and cohomology classes, Theorem 3.3.13 shows that the result will be true
if the image of U
in H
1
(P
(R), Z
2
) by (*) is the Euler class of U
. We
know that the restriction map
H
1
(P
m
(R); Z
2
) H
1
(P
n
(R); Z
2
)
is an isomorphism for m n 1. It follows that the restriction map
H
1
(P
(R); Z
2
) H
1
(P
1
(R); Z
2
)
is also an isomorphism. Hence, we are reduced to prove that the image
of U
1
(R) by (*) in H
1
(P
1
(R); Z
2
) is e
Z
2
U
1
(R)
. Set V
0
= [x
0
, x
1
] : x
0
,= 0
and V
1
= [x
0
, x
1
] : x
1
,= 0. Clearly, V
0
, V
1
are open subsets of P
1
(R)
which are homeomorphic to R. Moreover, P
1
(R) = V
0
V
1
and V
0
V
1
is homeomorphic to R
. The line bundle U

1
(R)
|V
0
has a continuous frame
given by
[x
0
, x
1
]
_
(1,
x
1
x
0
), [x
0
, x
1
]
_
.
Similarly, the line bundle U
1
(R)
|V
1
has a continuous frame given by
[x
0
, x
1
]
_
(
x
0
x
1
, 1), [x
0
, x
1
]
_
.
The continuous 1-cocycle on 1 = V
0
, V
1
with values in GL
1
(R) = R
associated to U
1
is thus given by
V
0
V
0
= 1,
V
1
V
1
= 1,
V
1
V
0
=
x
1
x
0
,
V
0
V
1
=
x
0
x
1
.
Its image c in

H
1
(1; Z
2
) is thus the class of the 1-cocycle
V
0
V
0
= 0,
V
1
V
1
= 0,
V
1
V
0
= s(
x
1
x
0
),
V
0
V
1
= s(
x
0
x
1
).
Let (
V
0
,
V
1
) be an element of

C
0
(1; Z
2
). Then,
V
0
(resp.
V
1
) is constant
on V
0
(resp. V
1
) and
d(
V
0
,
V
1
)
V
1
V
0
=
V
0|V
1
V
0
V
1|V
1
V
0
is also constant on V
1
V
0
. Since
V
1
V
0
is not constant on V
1
V
0
, it follows
that c ,= 0. Using the fact that

H
1
(1; Z
2
) H
1
(P
1
(R); Z
2
) Z
2
, it follows
that c = e
Z
2
U
1
(R)
.
Corollary 3.5.5. Assume B is a topological space. Then, a real line bundle
L on a B is trivial if and only if
e
Z
2
L
= 0.
In particular, a real vector bundle E on B is orientable if and only if
w
1
(E) = 0
Proof. Since a real vector bundle E is orientable if and only if the line
bundle Det(E) is trivializable, the result follows directly from the preceding
proposition combined with Corollary 3.2.9.
4
Characteristic classes of
complex vector bundles
In this chapter as in the previous one, all topological spaces are implicitly
assumed to be paracompact.
4.1 Generalities on complex vector bundles
Denition 4.1.1. A complex vector bundle of rank r is the data of a con-
tinuous map
p
E
: E B
E
between topological spaces together with structures of complex vector spaces
on each ber E
b
= p
1
E
(b) (b B
E
) of P
E
. These data being such that for
any b B there is a neighborhood U of b in B and a family (e
1
, , e
r
)
of continuous sections of p
E
|p
1
E
(U)
: p
1
E
(U) U with the property that
(e
1
(b
), , e
r
(b
)) is a basis of p
1
E
(b
) for any b
U.
As seen from the preceding denition, the notion of complex vector
bundle is completely similar to that of real vector bundle. We just have
to replace R-linearity with C-linearity when appropriate. This is why we
will not dene in details the vocabulary concerning complex vector bundles,
assuming the reader can adapt easily what has been done for real vector
bundles.
139
140 4. Characteristic classes of complex vector bundles
Lemma 4.1.2. Let u : E E be a morphism of complex vector bundles.
Denote E
R
the real vector bundle associated to E and
u
R
: E
R
E
R
the morphism of real vector bundles associated to u. Then,
det
R
u
R
= [det
C
u[
2
.
Proof. Let e be an eigenvector of u and denote the associated eigenvalue.
Choose F such that E = )e F and denote p : E F the associated
projection. Denote : )e )e the multiplication by and v : F F the
map p u
|F
. Then,
det
C
u = det
C
v
and
det
R
u
R
= det
R
R
det
R
v
R
.
Therefore, an induction argument reduces the problem to the case where
dim
C
E = 1. In this case, if e is a non-zero vector of E and u(e) = e, we
have
det
C
u = .
Since (e, ie) is a basis of E
R
and the matrix of u
R
in this basis is
_
'
'
_
we have det
R
u
R
= '
2
+
2
= [[
2
. The conclusion follows.
Corollary 4.1.3. Let E be a complex vector bundle of rank r on B. Then,
the underlying real vector bundle E
R
is canonically oriented. If (e
1
, , e
r
)
is a continuous local frame of E, then (e
1
, ie
1
, , e
r
, ie
r
) is a continuous
oriented local frame of E
R
. In particular, the Euler class e
Z
E
R
H
2r
(B; Z)
is well-dened.
Exercise 4.1.4. Denote P
n
(C) the complex projective space and U
n
(C)
the complex universal bundle. Set = e
Z
U
n
(C)
H
2
(P
n
(C); Z). Then,
H
(P
n
(C); Z) Z Z Z
n
.
In particular,
(P
n
(C)) = n + 1.
Solution. Work as in Exercise 2.5.6.
4.2. Chern classes 141
Proposition 4.1.5. Let E be a complex vector bundle of rank r on B. De-
note : P(E) B the associated complex projective bundle. Let U(E) be
the universal complex line bundle on P(E) and let = e
Z
U(E)
H
2
(P(E); Z)
be its Euler class. Then
H
(P(E); Z)
is a free H
(B; Z)-module of rank r with

1, ,
2
, ,
r1
as basis.
Proof. Work as in the proof of Proposition 3.1.2.
Corollary 4.1.6. Let E be a complex vector bundle of rank r on the
paracompact base X. Then, there is a proper map
f : Y X
for which the canonical map
f
: H
(X; Z) H
(Y ; Z)
f
1
(E) L
1
L
r
with L
1
, , L
r
complex line bundles on Y .
Proof. Work as in the proof of Proposition 3.2.1.
4.2 Chern classes
Denition 4.2.1. Using the notations of Proposition 4.1.5, we dene the
Chern classes of E as the classes c
1
(E) H
2
(B; Z), , c
r
(E) H
2r
(B; Z)
characterized by the relation
r
=
(c
1
(E))
r1
(c
2
(E))
r2
+ + (1)
r
(c
r
(E)).
By convention, we extend the preceding denition by setting
c
0
(E) = 1, c
k
(E) = 0 (k > r).
As for Stiefel-Whitney classes, we also dene the total Chern class c(E) as
the sum
k
c
k
(E) H
(B; Z).
Remark 4.2.2. Note that c(E) is in fact an element of
H
ev
(B; Z) :=
kN
H
2k
(B; Z)
which is a commutative subalgebra of H
(B; Z).
Proposition 4.2.3. Assume E, F are complex vector bundles on B of rank
r, s and let f : B
B be a continuous map. Then,

(a) c(f
1
(E)) = f
(c(E));
(b) c(E F) = c(E) c(F);
(c) c
r
(E) = e
Z
E
R
.
Proof. Work as in Propositions 3.1.4, 3.1.5 and 3.2.3.
Denition 4.2.4. Let E be complex vector bundle with base B. We denote
E the complex vector bundle obtained from E by changing the C-vector
space structure of each ber into its conjugate one.
Proposition 4.2.5. Let E be a complex vector bundle with rank r and
base B. Then,
c(E
) = c(E) = (1)
r
c(E).
Proof. Endowing E with a Hermitian structure, one sees easily that
E
E
and the rst equality follows. To get the second one, we may use the splitting
principle and treat only the case where E is a complex line bundle. In this
case, we have only to show that
e
Z
E
= e
Z
E
.
In other words, we have to show that that the canonical orientations o and
o of E
R
induced by the complex structure of E and E are opposite. This
follows from the fact that if e is a complex local frame of E, then e, ie is a
positively oriented local frame for o and e, ie is a positively oriented local
frame for o.
Exercise 4.2.6. With the notations of Exercise 4.1.4, show that
c(TP
n
(C)) = (1 + )
n+1
and deduce from this formula that
_
P
n
(C)
n
= 1.
Solution. The rst relation is obtained by working as in Exercise 3.2.17.
From this relation we deduce that
P
n
(C)
= c
n
(TP
n
(C)) = (n + 1)
n
.
Therefore, using the index theorem for compact manifolds, we get
_
P
n
(C)
(n + 1)
n
=
_
P
n
(C)
P
n
(C)
= (P
n
(C)) = n + 1.
Proposition 4.2.7. Denote G
,r
(C) the complex innite Grassmannian
of rank r and U
,r
(C) the associated universal complex vector bundle of
rank r. Assume E is a complex vector bundle of rank r on B. Then, there
is a continuous map f : B G
,r
(C) such that
E f
1
U
,r
(C).
Moreover, such a map is unique up to homotopy.
Proof. Work as for Corollary 3.3.14.
Corollary 4.2.8. Characteristic classes with values in the abelian group
M of complex vector bundles of rank r are in bijection with
H
(G
,r
(C); M).
Proposition 4.2.9. The morphism of rings
Z[C
1
, , C
r
] H
(G
,r
(C); Z)
dened by sending C
k
to c
k
(U
,r
(C)) is an isomorphism.
Proof. Work as for Proposition 3.4.7.
Proposition 4.2.10. The exact sequence of sheaves
0 Z
B
2i
(
B
(
B
0
H
1
(B; (
B
) H
2
(B; Z
B
).
By this isomorphism, the class of a complex line bundle L is sent to c
1
(L).
Proof. Working as in the proof of Proposition 3.5.4, we may assume B =
P
1
(C), L = U
1
(C). Set V
0
= [z
0
, z
1
] : z
0
,= 0, V
1
= [z
0
, z
1
] : z
1
,= 0.
Dene v
0
: V
0
C and v
1
: V
1
C by setting
v
0
=
z
1
z
0
, v
1
=
z
0
z
1
.
Clearly, U
1
(C)
|V
0
has a continuous frame given by s
0
dened by
[z
0
, z
1
]
_
(1,
z
1
z
0
), [z
0
, z
1
]
_
.
Similarly, U
1
(C)
|V
1
has a continuous frame given by s
1
dened by
[z
0
, z
1
]
_
(
z
0
z
1
, 1), [z
0
, z
1
]
_
.
Since
s
0
|V
0
V
1
= v
0
s
1
|V
0
V
1
,
the class of L in H
1
(P
1
(C), (
P
1
(C)
) is the image of v
0
by the coboundary
operator
H
0
(V
0
V
1
; (
P
1
(C)
) H
1
(P
1
(C); (
P
1
(C)
)
of the Mayer-Vietoris sequence associated to the decomposition P
1
(C) =
V
0
V
1
. Dene Z
1
as the kernel of the de Rham dierential

d
1
: (
1
(
2
and
dlog : (
Z
1
by setting
dlog(f) =
df
f
.
The morphism of exact sequences
0

Z
2i
2i
exp
id

dlog
0
0

C

(
Z
1
0
induces the commutative diagram of coboundary operators
H
1
(P
1
(C); (
)
/o

dlog
H
2
(P
1
(C); Z)
2i
H
1
(P
1
(C); Z
1
)
/o

H
2
(P
1
(C); C)
We have to show that the image of [L] in H
2
(P
1
(C); Z) is a generator. Hence
it is sucient to show that
_
P
1
(C)
dlog([L]) = 2i.
To compute dlog([L]), let us proceed as follows. First, the commutative
diagram
H
0
(V
0
V
1
; (
)

dlog
H
1
(P
1
(C); (
)
dlog
H
0
(V
0
V
1
; Z
1
)

H
1
(P
1
(C); Z
1
)
for the coboundary operators of the Mayer-Vietoris sequences of (
and
Z
1
, shows that dlog[L] is the image of

dv
0
v
0
H
0
(V
0
V
1
; Z
1
)
in H
1
(P
1
(C); Z
1
). To compute this image, we will use the soft resolution

0 Z
1
(
1
d
(
2
0
of Z
1
. We have the commutative diagram with exact rows

0

(P
1
(C); (
1
(V
0
; (
1
) (V
1
; (
1
(V
0
V
1
; (
1
0
0

(P
1
(C); (
2
)

(V
0
; (
2
) (V
1
; (
2
)

(V
0
V
1
; (
2
)

0
To obtain the image of
dv
0
v
0
in (P
1
(C); (
2
)/d(P
1
(C); (
1
), we have rst
to nd (,
) (V
0
; (
1
) (V
1
; (
1
) such that
|V
0
V
1

|V
0
V
1
=
dv
0
v
0
and then to use the relations
|V
0
= d,
|V
1
= d
. Let be a C
-function
on C which is zero for [z[ 1/4 and 1 for [z[ 3/4. Clearly, we may take
= (v
0
)
dv
0
v
0
,
= (1 (v
0
))
dv
0
v
0
.
Then,
d
|V
0
V
1
= d
|V
0
V
1
and these forms have compact supports. Moreover, is equal to d on
V
0
V
1
and 0 on P
1
(C) (V
0
V
1
). Therefore, denoting D(0, 1) the unit
disk in C, we have
_
P
1
(C)
=
_
V
0
d =
_
C
d((z)
dz
z
)
=
_
D(0,1)
d((z)
dz
z
) =
_
D(0,1)
(z)
dz
z
=
_
D(0,1)
dz
z
= 2i
where the last formula follows from Cauchys theorem.
Proposition 4.2.11. Assume E (resp. F) is a complex vector bundle of
rank r (resp. s) on B. Then,
c(E F) = T
r,s
(c
1
(E), . . . , c
r
(E), c
1
(F), . . . , c
s
(F)).
Proof. Working as in the proof of Proposition 3.2.12, we see that it is suf-
cient to treat the case r = s = 1. In this case, we have only to prove
that
c
1
(E F) = c
1
(E) +c
1
(F).
This follows easily from the preceding proposition.
4.3 Chern-Weil construction
In this section, p : E B will denote a dierentiable complex vector bundle
of rank r on the paracompact dierential manifold B. As usual, for such a
bundle, C
p
(B; E) will denote the space of dierentiable p-forms with values

in E (i.e. the dierentiable sections of
_
p
T
X
C
C
E).
A C-linear connection on E is the data of a C-linear operator
_ : C
0
(B; E) C
1
(B; E)
satisfying Leibnitz rule
_(fs) = (df)s + f(_s)
for any f C
(B), s C
(B; E). Such an operator gives rise to a family

of operators
_
p
: C
p
(B; E) C
p+1
(B; E)
which is uniquely characterized by the fact that
_
p
(s) = (d)s + (1)
p
(_s)
4.3. Chern-Weil construction 147
for any C
p
(B; E) and any s C

0
(B; E). Let e

1
, , e
r
be a dieren-
tiable frame of E on the open subset U of B. Then,
_e
j
=
r
k=1
kj
e
k
where is a r r matrix of dierential 1-forms on U. We call the matrix
of the connection _ in the local frame e
1
, , e
r
. If s is a dierentiable
section of E on U, then s may be written in a unique way as
s =
r
j=1
s
j
e
j
.
For such an s, we have
_s =
r
j=1
(ds
j
)e
j
+
r
j=1
r
k=1
s
j
kj
e
k
.
Hence,
(_s)
j
= ds
j
+
r
k=1
jk
s
k
.
Moreover,
_
1
_
0
s
= _
1
(
r
j=1
(ds
j
+
r
k=1
jk
s
k
)e
j
)
=
r
j=1
r
k=1
((d
jk
)s
k

jk
ds
k
)e
j
(ds
j
+
r
k=1
jk
s
k
)
r
l=1
lj
e
l
.
Therefore,
(_
1
_
0
s)
j
=
r
k=1
(d
jk
)s
k

jk
ds
k
l=1
(ds
l
+
r
k=1
lk
s
k
)
jl
=
r
k=1
(d
jk
)s
k

r
l=1
r
k=1
lk

jl
s
k
=
r
k=1
_
d
jk
+
r
l=1
jl

lk
_
s
k
.
In particular, _
1
_
0
: C
0
(B; E) C
2
(B; E) is C
(E)-linear and hence

comes from a morphism of vector bundles
K : E
2
X
C
C
E
whose matrix in the local frame e
1
, , e
r
is
= d + . (*)
The morphism K is called the curvature of the connection _. Note that
more generally,
_
p+1
_
p
( s) = _
p+1
((d)s + (1)
p
(_s))
= (1)
p+1
(d) (_s) +(1)
p
(d) (_s)
+(1)
2p
(_
1
_
0
(s))
= K(s).
Let f : B
B be a dierentiable map and let _ be a C-linear connec-

tion on E. Clearly, f
1
(E) is a dierentiable complex bundle on B
. We
denote f
1
_ the C-linear connection on f
1
E characterized by the fact
that its matrix with respect to the dierentiable frame e
1
f, , e
r
f of
f
1
E on f
1
(U) associated to a dierentiable frame e
1
, , e
r
of E on U
is the pull-back by f of the matrix of _ in e
1
, , e
r
(i.e. we set
f
1
= f
).
We leave it to the reader to check that this denition is meaningful. From
this denition and formula (*), it follows easily that, if
f
1
(resp.
)
denotes the matrix of the curvature of f
1
_ (resp. _) in the local frame
e
1
f, , e
r
f (resp. e
1
, , e
r
) of f
1
(E) (resp. E), then
f
1
= f
.
Denote M
r
(C) the algebra of r r matrices of complex numbers. An
invariant homogeneous polynomial of degree k on M
r
(C) is a homogeneous
polynomial map
P : M
r
(C) C
of degree k such that
P(T
1
AT) = P(A)
for any T GL
r
(C) and any A M
r
(C). (Examples of such invariant
polynomials are given by the determinant (degree r) or the trace (degree
1)). Let P be an invariant homogeneous polynomial of degree k on M
r
(C)
and let _ be a C-linear connection on E. Denote
,e
the matrix of the
curvature of _ in a frame e = (e
1
, , e
r
) of E on the open subset U of B.
Since C
ev
(U; E) is a commutative C-algebra, the expression

P(
,e
)
is a well-dened dierential form of degree 2k on U. Moreover, since P is
invariant, P(
,e
) does not depend on the frame e. It follows that there is
a unique dierential form P(K
) characterized by the fact that

P(K
)
|U
= P(
,e
)
for any local frame e = (e
1
, , e
r
) of E. Moreover, by construction, we
have
P(K
f
1
) = f
P(K
).
Proposition 4.3.1. For any C-linear connection _ on E and any invariant
homogeneous polynomial P on M
r
(C),
P(K
)
is a closed dierential form of degree 2k.
Proof. Let us dene the matrix P
(A) by setting
P
(A)
jk
=
P(A)
A
kj
.
Then, for any H M
r
(C), we have
L
H
P(A) =
r
j=1
r
k=1
P(A)
A
kj
H
kj
= tr(P
(A)H)
where L
H
denotes the derivative in the direction H. The fact that P is
invariant entails that P
(A) commutes with A. As a matter of fact, let

H M
r
(C). From the relation
P(A(I + tH)) = P((I + tH)A)
which holds for any t R, we deduce by derivation that
tr(P
(A)AH) = tr(P
(A)HA) = tr(AP
(A)H).
Since H is arbitrary, we see that
P
(A)A = AP
(A).
Let (resp. ) be the matrix (resp. the curvature matrix) of the connection
_ with respect to a dierentiable frame e
1
, , e
r
of E on U. A simple
computation shows that
dP() =
r
j=1
r
k=1
P
()
jk
d
kj
= tr(P
() d).
Since
= d + ,
we have
d = d d
= (Bianchy identity).
Therefore,
dP() = tr(P
() ) tr(P
() )
= tr( P
() ) tr(P
() )
= 0.
Proposition 4.3.2. Let P be an invariant homogeneous polynomial on
M
r
(C) and let _
0
and _
1
be two C-linear connections on E. Then, the
closed dierential forms P(K
0
) and P(K
1
) determine the same de Rham
cohomology class.
Proof. Consider the projection
p
B
: R B B
and the two connections p
1
B
_
0
, p
1
B
_
1
on p
1
B
E. Denote t : R B R
the rst projection and set
_ = (1 t)(p
1
B
_
0
) + t(p
1
B
_
1
).
Clearly, _ is a connection on p
1
B
E. Denote
i
0
: B R B, i
1
: B R B
the two embeddings dened by setting
i
0
(x) = (0, x), i
1
(x) = (1, x).
Clearly, i
1
0
p
1
B
(E) E and i
1
1
p
1
B
(E) E. Moreover, by construction
i
1
0
_ = _
0
and i
1
1
_ = _
1
. Therefore,
P(K
0
) = P(K
i
1
0

) = i
0
P(K
),
P(K
1
) = P(K
i
1
1

) = i
1
P(K
).
Using the isomorphism between de Rham cohomology and the usual coho-
mology together with the homotopy theorem, we see that the maps
i
0
= H
dR
(R B; C) H
dR
(B; C)
i
1
= H
dR
(R B; C) H
dR
(B; C)
are equal. The conclusion follows directly.
Proposition 4.3.3. There is at least one C-linear connection on E.
Proof. This is clear if E is trivializable. In general, we may nd a locally
nite covering | of B by open subsets U such that E
|U
is trivializable. For
each U |, x a connection _
U
on E
|U
. Let (
U
)
UU
be a partition of
unity subordinated to |. Then,
_ =

uU
U
_
U
is a well-dened C-linear connection on E.
Corollary 4.3.4. There is a canonical way to associate to any invariant
polynomial on M
r
(C) a characteristic class for dierentiable complex vector
bundles of rank r with values in C.
Proof. Write P = P
0
+ P
1
+ + P
k
where each P
l
is a homogeneous
invariant polynomial of degree l. Let E be a dierentiable complex vector
bundle. By Proposition 4.3.3, there is at least one C-linear connection _ on
E. By Proposition 4.3.1, P
l
(K
) is a well-dened closed dierential form

of degree 2l. Moreover, Proposition 4.3.2 shows that the cohomology class
P
l
(E) H
2l
(B; C)
of P
l
(K
) depends only on E. Set
P
(E) =
P
0
(E) + +
P
k
(E) H
ev
(B; C).
Since we have clearly
P
(f
1
E) = f
P
(E)
for any dierentiable map f : B
B, the conclusion follows.

Proposition 4.3.5. For any dierentiable complex vector bundle E of rank
r on B, we have
c(E) =
C
r
(E)
where C
r
is the invariant polynomial of A M
r
(C) dened by setting
C
r
(A) = det(I +
A
2i
).
Proof. (a) Assume rst r = 1. Since B is paracompact and has nite
dimension, it is possible to nd a dierentiable map
f : B P
n
(C)
such that E f
1
U
n
(C). It follows that we have only to treat the case of
U
n
(C) on P
n
(C). Moreover, since the restriction map
H
1
(P
n
(C); C) H
1
(P
1
(C); C)
is injective, we can even assume n = 1. In this case, we have to show that
if _ is a connection on U
1
(C), then c
1
(U
n
(C)) is represented by
[K
]
2i
or in
other words that
_
P
1
(C)
K
= 2i.
As usual, set
V
0
= [z
0
, z
1
] : z
0
,= 0, V
1
= [z
0
, z
1
] : z
1
,= 0
and dene the coordinates v
0
: V
0
C and v
1
: V
1
C by setting
v
0
([z
0
, z
1
]) =
z
1
z
0
, v
1
([z
0
, z
1
]) =
z
0
z
1
.
We know that U
1
|V
0
has a frame s
0
dened by setting
s
0
([z
0
, z
1
]) =
_
(1,
z
1
z
0
), [z
0
, z
1
]
_
and that similarly, U
1|V
1
has a frame s
1
dened by setting
s
1
([z
0
, z
1
]) =
_
(
z
0
z
1
, 1), [z
0
, z
1
]
_
.
A connection _ on U
1
(C) is thus characterized by the dierential forms
0
C
1
(V
0
; U
1
(C)),
1
C
1
(V
1
; U
1
(C)) dened by the relations
_s
0
=
0
s
0
, _s
1
=
1
s
1
.
Moreover, since
s
0
= v
0
s
1
on V
0
V
1
, we have
_s
0
= (dv
0
)s
1
+v
0
_s
1
and hence
0
=
(dv
0
)
v
0
+
1
.
Conversely, the two 1-forms
0
,
1
are given and satisfy the preceding gluing
condition, we can use them to construct a unique connection _ on U
1
(C).
Let be a dierentiable function on C such that
= 0 on D(0,
1
4
), = 1 on D(0,
3
4
).
Set
0
= (v
0
)
dv
0
v
0
on V
0
and
1
=
_
1 (
1
v
1
)
_
dv
1
v
1
on V
1
. On V
0
V
1
, we have
v
1
=
1
v
0
,
hence,
dv
1
=
1
v
2
0
dv
0
and
dv
1
v
1
=
dv
0
v
0
.
It follows that
0
=
dv
0
v
0
+
1
on V
0
V
1
and that the forms
0
,
1
dene a connection _ on U
1
(C). For
this connection, we have
0
= d
0
+
0

0
= d
_
(v
0
)
dv
0
v
0
_
on V
0
. Since
0
has compact support in V
0
, K
corresponds to the form

obtained by extending
0
by 0 outside V
0
. Therefore,
_
P
1
(C)
K
=
_
V
0
0
=
_
C
d
_
(z)
dz
z
_
=
_
D(0,1)
d
_
(z)
dz
z
_
=
_
D(0,1)
(z)
dz
z
=
_
D(0,1)
dz
z
= 2i
(b) In the general case, it is easy to nd a dierentiable map
f : B
B
such that f
1
E L
1
L
r
with L
1
, , L
r
of rank 1 and for which
f
: H
(B; C) H
(B
; C)
is injective. Let _
1
, , _
r
be connections on L
1
, , L
r
and denote
_ = _
1
_
r
the associated direct sum connection on f
1
E. In the
local frame e
1
, , e
r
of f
1
(E) corresponding to local frames e
1
, , e
r
of
L
1
, , L
r
, the curvature matrix has the diagonal form
_
_
_
1
.
.
.
r
_
_
_
where
k
is the curvature form of _
k
with respect to e
k
. Therefore,
det(I +

2i
) = (I +

1
2i
) (I +

r
2i
)
and
C
r
(f
1
(E)) =
C
1
(L
1
)
C
1
(L
r
).
It follows from (a) that
C
r
(f
1
(E)) = c(L
1
) c(L
r
) = c(f
1
(E)).
Hence
f
C
r
(E) = f
c(E)
Corollary 4.3.6. For any A M
r
(C), denote
1
(A), ,
r
(A) the com-
plex numbers dened by setting
det(I +tA) = 1 +t
1
(A) + +t
r
r
(A).
Then,
1
(A), ,
r
(A) are invariant homogeneous polynomials of degree
1, , r and we have
1
(E) = 2ic
1
(E)
.
.
.
r
(E) = (2i)
r
c
r
(E)
for any dierentiable complex vector bundle E of rank r on B.
Proof. This follows directly from the equality
C
r
(A) = 1 +

1
(A)
2i
+ +

r
(A)
(2i)
r
.
Corollary 4.3.7. There is a one to one correspondence between invariant
polynomials on M
r
(C) and characteristic classes for complex vector bundles
of rank r with coecients in C.
Proof. We know that any characteristic class for complex vector bundles
of rank r may be written in a unique way as a polynomial Q(c
1
, , c
r
) in
Chern classes. It follows that
=
Q

1
2i
, ,
r
(2i)
r
.
Conversely, let P be an invariant polynomial on M
r
(C). We know that for
any matrix A M
r
(C) there is T GL
r
(C) such that
B = T
1
AT =
_
_
_
_
_
_
1
0 0

2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0

r
_
_
_
_
_
_
For
S() =
_
_
_
_
_
2
.
.
.
r
_
_
_
_
_
the matrix
C() = S()BS()
1
is such that
C
jk
() =
jk
B
jk
.
It follows that
lim
0
C
jk
() =
k
jk
.
But since P is invariant, we have
P(A) = P(B) = P(C()) = lim
0
P(C()) = P(diag(
1
, ,
r
)).
It follows that P(A) is a symmetric polynomial in the eigenvalues
1
, ,
r
of A. Therefore, there is a unique polynomial Q(S
1
, , S
r
) such that
P(A) = Q(
1
(A), ,
r
(A)).
It follows that
P
= Q(2ic
1
, , (2i)
r
c
r
).
Hence, the conclusion.
4.4 Chern character
Denition 4.4.1. Let X be a topological space and let T be an abelian
sheaf. Hereafter, we denote
(X; T)
the completion of the topological abelian group obtained by endowing the
graded abelian group
H
(X; T)
with the topology for which
kl
H
k
(X; T) (l N)
is a basis of neighborhoods of 0. Of course, forgetting the topologies, we
have
(X; T)
kN
H
k
(X; T).
Remark 4.4.2. Consider a formal series
P(x
1
, , x
r
) =
k
1
=0

k
r
=0
a
k
1
k
r
x
k
1
1
x
k
r
r
with coecients in a ring with unit A. Assume P(x
1
, , x
r
) is symmetric
(i.e. such that
P(x
1
, , x
r
) = P(x
1
, , x
r
)
for any permutation of 1, , r). Then, the associated homogeneous
polynomials
P
m
(x
1
, , x
r
) =

|k|=m
a
k
x
k
(m 0).
are also symmetric. As was explained in Remark 3.2.4 all these polynomials
may be written in a unique way as
Q
m
(S
r,1
(x
1
, , x
r
), , S
r,r
(x
1
, , x
r
)).
4.4. Chern character 157
This allows us to associate to P and to any complex vector bundle E of
rank r on X a class c
P
(E)

H
(X; A) by setting
c
P
(E) =
m=0
Q
m
(c
1
(E), , c
r
(E)).
Note that if f : Y X is a continuous map, then it follows from the
preceding construction that
c
P
(f
1
E) = f
c
P
(E).
Denition 4.4.3. The class ch(E)

H
(X; Q) given by the construction

above for A = Q and
P(x
1
, , x
r
) = e
x
1
+ +e
x
r
,
is called the Chern character of E.
Proposition 4.4.4. Let X be a topological space. Assume E and F are
complex vector bundles on X. Then,
(a) ch(E F) = ch(E) + ch(F);
(b) ch(E F) = ch(E) ch(F).
Proof. This follows directly from the splitting principle and the denition
of the Chern character.
Proposition 4.4.5. There is a unique extension
ch :
/
b
(1ect
C
(B))

H
(B; Z)
of the usual Chern character
ch : 1ect
C
(B)

H
(B; Z)
which is invariant by isomorphism and such that
(i) for any E
in
/
b
(1ect
C
(B)) we have
ch(E
[1]) = ch(E
)
(ii) for any distinguished triangle
E
+1
of
/
b
(1ect
C
(B)), we have
ch(F
) = ch(E
) +ch(G
).
This extension is given by the formula
ch(E
) =
kZ
(1)
k
ch(E
k
).
Proof. Uniqueness. Let E
be an object of
/
b
(1ect
C
(B)) and let us prove
that
ch(E
) =
kZ
(1)
k
ch(E
k
).
We know that there are integers a, b such that E
k
= 0 if k , [a, b]. Let us
proceed by increasing induction on b a. If b a = 0, the result follows
directly from (i) and the fact that ch extends the usual Chern character.
Assume now that b a > 0. Remark that E
is isomorphic to the mapping

cone of
E
a
[a 1]
d
a
>a
E
where
>a
E
denotes the complex

0 E
a+1
E
b
0
with E
a+1
in degree a +1. It follows from (i) and (ii) that
ch(
>a
E
) = (1)
a1
ch(E
a
) + ch(E
).
Hence, by the induction hypothesis
ch(E
) = (1)
a
ch(E
a
) +

ka+1
(1)
k
ch(E
k
) =
kZ
(1)
k
ch(E
k
).
Existence. Let us dene ch(E
) as

kZ
(1)
k
ch(E
k
). Assume rst
that E
0 in
/
b
(1ect
C
(B)). As is well-known, this means that E
is split
exact. Hence, if Z
k
denotes the kernel of d
k
: E
k
E
k+1
, we have
E
k
Z
k
Z
k+1
and
ch(E
) =
kZ
(1)
k
ch(E
k
) =
kZ
(1)
k
ch(Z
k
)
kZ
(1)
k+1
ch(Z
k+1
) = 0.
Assume now that M
(u
) is the mapping cone of

u
: E
.
By construction,
M
k
(u
) = E
k+1
F
k
.
4.4. Chern character 159
Hence,
ch(M
(u
)) =
kZ
(1)
k
_
ch(E
k+1
) + ch(F
k
)
_
= ch(F
) ch(E
).
Combining the two preceding results, we see that if
u
: E
/
b
(1ect
C
(B)), then ch(E
) = ch(F
). It follows that
ch is invariant by isomorphism and that (ii) is satised.
Corollary 4.4.6. The Chern character for complexes has the following
properties:
(i) if E
and F
are objects of
/
b
(1ect
C
(B)), then
ch(E
) = ch(E
) ch(F
);
(ii) if b : B
B is a continuous map and E
is an object of
/
b
(1ect
C
(B)),
then
ch(f
1
E
) = f
ch(E
).
Proof. This follows at once from the denition of ch for complexes and the
similar properties of the usual Chern character. As a matter of fact, for (i),
we have
(E
)
k
=
lZ
E
l
F
kl
and
ch(E
) =
kZ
(1)
k
lZ
ch(E
l
) ch(F
kl
)
=
kZ
lZ
(1)
l
ch(E
l
) (1)
kl
ch(F
kl
)
= ch(E
) ch(F
).
As for (ii), we have
(f
1
E
)
k
= f
1
(E
k
)
and hence
ch(f
1
E
) =
kZ
(1)
k
ch(f
1
E
k
)
=
kZ
(1)
k
f
ch(E
k
)
= f
ch(E
).
Denition 4.4.7. The Todd class of a complex vector bundle E with base
X is the class td(E)

H
(X; Q) given by the construction of Remark 4.4.2

for A = Q and
P(x
1
, , x
r
) =
x
1
1 e
x
1

x
r
1 e
x
r
.
Remark 4.4.8. It follows directly from the preceding denition that
td(E F) = td(E) td(F)
if E and F are complex vector bundles on the same base.
Proposition 4.4.9. Let E be a complex vector bundle of rank r on the
topological space X. Then,
ch(
E) = c
r
(E
)/ td(E
).
Proof. By the splitting principle, we may assume that E = L
1
L
r
where L
1
, , L
r
are line bundles. Set x
1
= c
1
(L
1
), , x
r
= c
1
(L
r
). We
have
E = (
L
1
) (
L
r
).
Hence,
ch(
E) = ch(
L
1
) ch(
L
r
) = (1 e
x
1
) (1 e
x
r
).
On the other hand,
c(E
) = c(L
1
) c(L
r
) = (1 x
1
) (1 x
r
)
and
td(E
) = td(L
1
) td(L
r
) =
x
1
1 e
x
1

x
r
1 e
x
r
.
It follows that
c
r
(E
) = (x
1
) (x
r
).
Hence,
c
r
(E
)/ td(E
) = (1 e
x
1
) (1 e
x
r
)
4.5. Local chern character 161
4.5 Local chern character
Denition 4.5.1. Let E
be an object of
/
b
(1ect
C
(B)) and let S be a
closed subset of B. We say that E
is supported by S if E
B\S
0 in
/
b
(1ect
C
(B S)) and denote
/
b
S
(1ect
C
(B))
the subcategory of
/
b
(1ect
C
(B)) formed by the complexes which are sup-
ported by S.
Assume the object E
of
/
b
(1ect
C
(B)) is supported by S. Then,
ch(E
)
|B\S
= ch(E
|B\S
) = 0
in

H
(B S; Z). From the exact sequence
S
(B; Z)
i

H
(B; Z)
r
(B S; Z)
it follows that ch(E
) is the image of a class in

H
S
(B; Z). Although such a
class is in general not unique, we shall prove the following result (cf [17]).
Proposition 4.5.2. There is a unique way to dene
ch
S
:
/
b
S
(1ect
C
(B))

H
S
(B; Z)
for any topological space B and any closed subset S of B in such a way that
(i) if E
and F
are isomorphic objects of

/
b
S
(1ect
C
(B)), then
ch
S
(E
) = ch
S
(F
);
(ii) for any E
in
/
b
S
(1ect
C
(B)) we have
i(ch
S
(E
)) = chE
;
(iii) if b : B
B is a continuous map and E
is an object of
/
b
S
(1ect
C
(B))
then
ch
S
(f
1
E
) = f
ch
S
(E
)
for any closed subset S
of B
such that f
1
(S) S
.
Moreover,
(iv) for any E

/
b
S
(1ect
C
(B)) we have
ch
S
(E
[1]) = ch
S
(E
);
(v) for any distinguished triangle
E
+1
of
/
b
S
(1ect
C
(B)) we have
ch
S
(F
) = ch
S
(E
) +ch
S
(G
).
Proof. Uniqueness. By Lemma 4.5.3 below, we know that to any
E

/
b
S
(1ect
C
(B))
is canonically associated a continuous map
s : B

B,
a closed subset

S and an object

E
of
/
b
S
(1ect
C
(

B)) such that
(i) s
1
(
S) S;
(ii) s
1

E
;
(iii) the map
i :

H
S
(

B; Z)

H
(

B; Z)
is injective.
It follows that
ch
S
(E
) = ch
S
(s
1

E
) = s
ch
S
(

E
)
and since ch
S
(

E
) is the unique cohomology class such that
i(ch
S
(

E
)) = ch(

E
),
we get the conclusion.
Existence. Using the notations introduced above, we dene ch
S
(E
) by
setting
ch
S
(E
) = s
ch
S
(

E
)
where ch
S
(

E
) is characterized by the relation
i(ch
S
(

E
)) = ch(

E
).
Note that from this denition it follows easily that ch
S
(E
) = 0 if E
is exact.
As a matter of fact, by construction of

S we have in this case s(B)

S = .
Let E
1
, E
2
be two objects of
/
b
S
1
(1ect
C
(B
1
)) and
/
b
S
2
(1ect
C
(B
2
)). De-
note

B
1
,

S
1
,

E
1
and

B
2
,

S
2
,

E
2
the objects associated to E
1
and E
2
by
Lemma 4.5.3 and let s
1
: B
1

B
1
, s
2
: B
2

B
2
be the canonical maps.
(a) Assume rst that b : B
2
B
1
is a continuous map, that E
2
b
1
E
1
in
(
b
(1ect
C
(B
2
)) and that b
1
(S
1
) S
2
and let us prove that
b
ch
S
1
(E
1
) = ch
S
2
(E
2
).
By construction of

B
2
and

B
1
, we get a continuous map
b :

B
2

B
1
such that

b s
2
= s
1
b and an isomorphism
b
1

E
1
in
(
b
(1ect
C
(

B
2
)). From the denitions of

S
1
and

S
2
, it follows also that
b
1
S
1

S
2
.
Using Corollary 4.4.6, we see that
ch(

E
2
) =

b
ch(

E
1
).
This entails that
i
2
ch
S
2
(

E
2
) =

b
i
1
ch
S
1
(

E
1
)
=
i
2
(
ch
S
1
(

E
1
))
and hence that
ch
S
2
(

E
2
) =

b
ch
S
1
(

E
1
).
Therefore,
ch
S
2
(E
2
) = s
2
ch
S
2
(

E
2
) = s
ch
S
1
(

E
1
)
= b
1
ch
S
1
(

E
1
) = b
ch
S
1
(E
1
)
Let E

/
b
S
(1ect
C
(B)). Applying the result obtained above for B
1
=
B
2
, b = id, E
1
= E
, we see that ch
S
(E
) depends only on the isomorphy

class of E
in
(
b
(1ect
C
(B)). By a similar reasoning, we see also that
ch
S
(E
[1]) = ch
S
(E
).
(b) Assume now that B
1
= B
2
and S
1
= S
2
. Denoting B and S these
spaces, let us prove that
ch
S
(E
1
E
2
) = ch
S
(E
1
) + ch
S
(E
2
).
Set

C =

B
1

B

B
2
. Denote t : B

C the map deduced from from s
1
and
s
2
and p
1
:

C

B
1
, p
2
:

C

B
2
the two projections. Set

F
1
= p
1
1
1
,
2
= p
1
2
2
and T = p
1
1
(
S
1
) p
1
2
(
S
2
). Thanks to Lemma 4.5.5, we know
that
i :

H
T
(

C; Z) H
(

C; Z)
is injective. Since
i(ch
T
(

F
2
)) = ch(

F
2
)
= ch(

F
1
) +ch(

F
2
)
= i(ch
T
(

F
1
) + ch
T
(

F
2
))
it follows that
ch
T
(

F
2
) = ch
T
(

F
1
) + ch
T
(

F
2
).
Using the fact that t
1
(T) = s
1
1
(
S
1
) s
1
2
(
S
2
) S and the isomorphisms
t
1

F
1
s
1
1
1
E
1
, t
1

F
2
s
1
2
2
E
2
,
we get that
ch
S
(

E
2
) = t
ch
T
(

F
2
)
= t
ch
T
(

F
1
) +t
ch
T
(

F
2
)
= ch
S
(

E
1
) +ch
S
(

E
2
).
(c) With the same notations as in (b), let us now prove that if M
(u
)
is the mapping cone of the morphism
u
: E
1
E
2
of
(
b
S
(1ect
C
(B)) then ch
S
(M
(u
)) = ch
S
(E
2
) ch
S
(E
1
). To this end,
consider the complex vector bundle p : H B whose ber at b B is
Hom
C
b
(Vect
C
)
(E
b
, F
b
),
a continuous local frame being obtained by xing continuous local frames
for each E
k
and each F
k
and associating to a morphism v
the components
of the matrices of the v
k
s with respect to the xed frames. Set F
1
= p
1
E
1
and F
2
= p
1
E
2
. By construction, there is a canonical morphism
v
: F
1
F
2
and we can associate to u
: E
1
E
2
a section
u
: B H
making the diagram
1
u
(F
1
)

1
u
(v
1
u
(F
2
)
O
1
u
2
commutative. It follows that
1
u
(v
) M
(u
)
in
(
b
(1ect
C
(B)) and hence that
ch
S
(M
(u
)) =
u
ch
S
(M
(v
)).
Since
u
and
0
are clearly homotopic, we see that
ch
S
(M
(u
)) = ch
S
(M
(0
))
= ch
S
(E
1
[1] E
2
)
= ch
S
(E
2
) ch
S
(E
1
).
(d) When u
/
b
(1ect
C
(B)), M
(u
) is exact and
we get
ch
S
(E
2
) = ch
S
(E
1
).
It follows that for any distinguished triangle
E
1
E
2
E
3
+1
of
/
b
(1ect
C
(B)), we have
ch
S
(E
3
) = ch
S
(E
2
) ch
S
(E
1
)
and this completes the proof.
Lemma 4.5.3. Let S be a closed subset of the topological space B. Then,
to any object E
of
(
b
S
(1ect
C
(B)) one can associate canonically
(i) a continuous projection p :

B B with a canonical section s : B

B;
(ii) a closed subset

S of

B such that s
1
(
S) S and for which the canon-

ical map
i :

H
S
(

B; Q) H
(

B; Q)
is injective;
(iii) an object

E
of
(
b
S
(1ect
C
(

B)) such that
s
1
(

E
) E
.
Proof. The case S = B being obvious, we shall assume S ,= B.
(a) Construction of

B. Let
= (
l
)
lZ
be a family of natural integers
for which l Z :
l
,= 0 is nite and let V be a nite dimensional complex
vector space. By a ag of nationality
of V , we mean an increasing family

V
= (V
l
)
lZ
of complex vector subspaces of V such that
dimV
l+1
= dimV
l
+
l+1
.
Generalizing what has been done for complex Grassmannians, it is easy
to see that the set Fl
(V ) of ags of nationality
of V has a canonical
structure of dierential manifold. Working as in the construction of the
projective bundle associated to a complex vector bundle, we see more gen-
erally that any complex vector bundle E B gives rise to the ag bundle
Fl
(E) B.
In the special case where
E =
kZ
E
k
and
= (rk E
l
)
lZ
the ag bundle Fl
(E) B has a canonical section given by

b E
l,b
where
E
l
=
kl
E
k
for any l Z. Let us dene

B as the closed subset of Fl
(E) formed by
ags F
such that
E
l1,b
F
l
E
l+1,b
for any l Z if F
Fl
(E
)
b
. We denote p :

B B the map induced by
the canonical projection Fl
(E) B.
(b) Construction of

E
and

S. Denote

F
l
the complex vector bundle on
B whose ber at F

B is F
l
. By construction,
F
l

F
l+1
, p
1
E
l
p
1
E
l+1
, p
1
E
l1

F
l
p
1
E
l+1
for any l Z. Set
E
k
=

F
k
/p
1
E
k1
.
Clearly,

E
k
is a complex vector bundle of rank
k
on

B. Denote
d
k
:

E
k

E
k+1
the morphism of complex vector bundles deduced from the inclusions

F
k

F
k+1
, p
1
E
k1
p
1
E
k
. Since
F
k
p
1
E
k+1
we have

d
k+1

d
k
= 0 and

E
= (

E
k
,

d
k
)
kZ
is a complex of complex vector
bundles on

B. We denote

S the support of

E
in

B.
(c) Construction of s. Let us dene s : B

B as the map which
associates to any b B the ag F
Fl
(E)
b
dened by setting
F
l
= x
kl+1
E
k
b
: x
l+1
= d
l
x
l
.
One checks easily that s is continuous and that
s
1

E
in
(
b
(1ect
C
(B)). In particular, we see that the support of E
is s
1
(
S) and
hence that s
1
(
S) S.
(d) Construction of the sequence (
k
)
kZ
if

S ,=

B. Since

S ,=

B, the
complex

E is exact at some point

b of

B. It follows that there is a unique
sequence
= (
k
)
kZ
of natural numbers such that
k
=
k
+
k+1
.
As a matter of fact, if

Z
k
b
is the kernel of

d
k
b
:

E
k
b

E
k+1
b
, we have
E
k
b

Z
k
b

Z
k+1
b
and we may take
k
= dim

Z
k
b
. The existence follows. As for the uniqueness,
it is sucient to note that
l
=
k<l
(1)
kl+1
k
.
(e) Characterization of

B

S if

S ,=

B. Let us prove that a ag F

B
above b B is not in

S if and only if
dim(F
k
E
k,b
/E
k1,b
) =
k
(*)
for all k Z. To this end, recall that the ber of

E
at F
is isomorphic to
the complex
F
k1
/E
k2,b
F
k
/E
k1,b
F
k+1
/E
k,b

the k-cocycles of which are given by
Z
k
F
= F
k
E
k,b
/E
k1,b
.
Hence, if

E
is exact at F
, (*) follows from the construction of
in (a). Con-
versely, assume (*) holds for all k Z and denote

B
k
F
the k-coboundaries
of

E
. The exact sequence

0

Z
k1
F

E
k1
F

B
k
F
0
shows that
dim

B
k
F
= dim

E
k1
F
dim

Z
k1
F
=
k1
k1
=
k
.
It follows that dim

B
k
F
= dim

Z
k
F
and hence that

B
k
F
=

Z
k
F
.
(f) Cohomology of

B

S if

S ,=

B. Recall that if E is an n-dimensional
complex vector bundle on B and l is an integer such that 0 l n, Then
G
l
(E) denotes the Grassmannian bundle whose ber at b B is the set
of complex vector subspaces of dimension l of E
b
. Denote G the bered
product of the Grassmannian bundles G
k
(E
k
) (k Z). For any ag F
of
B

S above b and any k Z, part (e) shows that
(F
k
E
k,b
)/E
k1,b
is
k
-dimensional. Since
E
k,b
/E
k1,b
E
k
this gives us a continuous map
f :

B

S G.
We shall prove that this map is a ber bundle whose bers are isomorphic
to C
d
with d =
kZ

2
k
. This will entail that
f
: H
(G; Q) H
(

B

S; Q)
is an isomorphism.
For any k Z, let L
k
be an element of G
k
(E
k
) above b B. Denote
L
the element of G associated to the family (L

k
)
kZ
. A ag F
of

B above
b belongs to f
1
(L
) if and only if the image of

F
k
E
k,b
/E
k1,b
in E
k
is equal to L
k
for any k Z. Assume F
l
is known for l k +1, then
F
k
may be chosen arbitrarily between the subspaces of dimension

lk

l
of F
k+1
E
k+1,b
such that
F
k
E
k,b
= E
k1,b
L
k
.
These subspaces are in bijection with the subspaces G
k
of dimension
k

k
=
k+1
of
F
k+1
E
k+1,b
/E
k1,b
L
k
such that G
k
E
k,b
/E
k1,b
L
k
= 0. Since
dim(F
k+1
E
k+1,b
/E
k+1,b
) =
k+1
and
dim(E
k,b
/E
k1,b
L
k
) =
k
k
=
k+1
,
we have
dim(F
k+1
E
k+1,b
/E
k1,b
L
k
) =
k+1
+
k
k
= 2
k+1
.
It follows that G
k
may be chosen arbitrarily among the supplementary sub-
spaces of a xed
k+1
dimensional subspace in a space of dimension 2
k+1
.
Thanks to Lemma 4.5.4, these subspaces are in bijection with C
2
k+1
. The
conclusion follows by decreasing induction on k.
Now, consider the continuous map
g : G

B

S
dened by sending an element L
of G above b B to the ag
(E
l1,b
L
l
L
l+1
)
lZ
.
By construction,
f g = id
G
.
Therefore,
g
: H
(

B

S; Q) H
(G; Q)
is the inverse of
f
: H
(G; Q) H
(

B

S; Q).
(g) Injectivity of

i. A long exact sequence of cohomology shows that the
canonical morphism
i : H
S
(

B; Q) H
(

B; Q)
is injective, if the canonical morphism
: H
(

B; Q) H
(

B

S; Q)
associated to the inclusion

j :

B

S

B is surjective. Set
G
kZ
G
k
(E
k
E
k+1
)
and denote
f
:

B G
the continuous map which sends a ag F
of

B
b
to the element
(F
k
/E
k1,b
)
kZ
of G
b
. Clearly, f
j g is the map
h : G G
which sends an element L
G
b
to the element
(L
k
L
k+1
)
kZ
of G
b
. Since
h
= g
j)
(f
and since g
is an isomorphism, it is sucient to show that h
is surjective.
Denote
p
k
: G G
k
(E
k
)
the canonical projection and T
k
the tautological bundle on G
k
(E
k
). Re-
peated applications of the Leray-Hirsch theorem show that
H
(G; Q)
is a free H
(B; Q)-module generated by the p
k
c
l
(T
k
). Denote
p
k
: G
k
(E
k
E
k+1
)
the canonical projection and T
k
the tautological bundle on G
k
(E
k
E
k+1
).
Clearly, we have
(p
k
h)
1
(T
k
) p
1
k
T
k
p
1
k+1
T
k+1
.
Therefore,
(h
(p
k
)
)(c
(T
k
)) = p
k
c
(T
k
) p
k+1
c
(T
k+1
).
It follows that
p
k
c
(T
k
) p
k+1
c
(T
k+1
)
and its inverse both belong to the image of h
. Hence, proceeding by de-

creasing induction on k, we see that p
k
c
(T
k
) Imh
for all k Z. This

entails that Imh
= H
(G; Q) and the conclusion follows.

Lemma 4.5.4. Let E, F be two nite dimensional complex vector spaces
and set G = EF. Then, the vector subspace H of G such that EH = G
are canonically in bijection with the complex vector space
Hom(F, E).
whose dimension is dimE dimF.
Proof. Let p
E
: G E and p
F
: G F be the two projections associated
to the decomposition G = E F. For any vector subspace H of G such
that E H = G, the linear map
p
F
|H
: H F
is bijective. Denote q its inverse and set h
H
= p
E
q. By construction,
q(x) = (h
H
(x), x) and H = (h
H
(x), x) : x F.
Now, let h Hom(F, E) and set
H
h
= (h(x), x) : x F.
Clearly, E H
h
= G and one checks easily that the maps H h
H
and
h H
h
are reciprocal bijections.
Lemma 4.5.5. Let S
1
, S
2
be two closed subsets of B and let E
1
, E
2
be
two bounded complexes supported respectively by S
1
and S
2
. Denote
q
1
:

B
1
B, s
1
: B

B
1
,

S
1
,

E
1
and
q
2
:

B
2
B, s
2
: B

B
2
,

S
2
,

E
2
the objects associated to E
1
and E
2
by Lemma 4.5.3. Set

C =

B
1

B

B
2
and denote p
1
:

C

B
1
, p
2
:

C

B
2
the two projections. Then the
canonical morphism
H
T
(

C; Q) H
(

C; Q)
is injective if T is equal to one of the closed subsets
p
1
1
(
S
1
), p
1
2
(
S
2
), p
1
1
(
S
1
) p
1
2
(
S
2
), p
1
1
(
S
1
) p
1
2
(
S
2
).
Proof. Thanks to Lemma 4.5.6, we have the following (horizontal) exact
sequence of (vertical) complexes
0 0 0
0

H
S
1
(

B
2
\
S
2
)
(

B
1

B
2
\

S
2
; Q)

(

B
1

B
2
\

S
2
; Q)

(

B
1
\

S
1

B
2
\

S
2
; Q)

0
0

H
S
1
B
2
(

B
1

B
2
; Q)
(

B
1

B
2
; Q)

(

B
1
\

S
1

B
2
; Q)

0
0

H
S
1
(

B
1
; Q) H
S
2
(

B
2
; Q)

(

B
1
; Q) H
S
2
(

B
2
; Q)

(

B
1
\

S
1
; Q) H
S
2
(

B
2
; Q)

0
0
where cartesian products (resp. tensor products) are to be understood over

B (resp. H
(B; Q)). The third (vertical) complex being exact, the snakes
Lemma combined with the fact that is surjective shows that is also
surjective. Combining this result with the long exact sequence associated
with the distinguished triangle
R
S
1
S
2
(

B
1

B
2
; Q) R
S
1
B
2
(

B
1

B
2
; Q) R
S
1
B
2
\
S
2
(

B
1

B
2
\

S
2
; Q)
+1
we get that
H
S
1
S
2
(

B
1

B
2
; Q) H
S
1
B
2
(

B
1

B
2
; Q)
is injective. Since is also injective, we have established the result for
T =

S
1

S
2
and T =

S
1

B
2
. Since and are both surjective we also
get the result for T =

B
1

S
2
and T = (
S
1

B
2
) (

B
1

S
2
).
Lemma 4.5.6. Let E
be a bounded complex supported by the closed

subset S of B. Let q :

B B and

S be as in Lemma 4.5.3. Then,
(i) H
(

B

S; Q) is a nite free H
(B; Q)-module;
(ii) for any continuous map b : C B, we have the canonical morphism
H
(

B

S; Q)
H
(B;Q)
H
(C; Q)

H
(

B

S
B
C; Q).
Proof. (i) Using the notions introduced in Lemma 4.5.3, we have
H
(

B

S; Q) H
(G; Q)
and the conclusion follows from Leray-Hirsch theorem.
(ii) Working as in part (c) of this same lemma, we see that
f
B
id
C
:

B

S
B
C G
B
C
H
(GC; Q) H
(

B

S
B
C; Q)
and the conclusion follows by applying once more Leray-Hirsch theorem.
Proposition 4.5.7. If E
1
and E
2
are complexes of complex vector bundles
on B with support respectively in S
1
and S
2
then E
1
E
2
has a support in
S
1
S
2
and
ch
S
1
S
2
(E
1
E
2
) = ch
S
1
(E
1
) ch
S
2
(E
2
).
Proof. Let us use the notations of Lemma 4.5.5. Denote t : B

C the map
associated to the canonical maps s
1
: B

B
1
, s
2
: B

B
2
and set
1
= p
1
1

E
1
,

F
2
= p
1
2

E
2
,

T
1
= p
1
1

S
1
,

T
2
= p
1
2

S
2
.
Clearly,

F
1
and

F
2
have support respectively in

T
1
and

T
2
and

F
2
has
support in

T
1

T
2
. Since
E
1
= t
1

F
1
, E
2
= t
1

F
2
and
S
1
S
2
t
1
(
T
1

T
2
)
it follows that
ch
S
1
S
2
(E
1
E
2
) = t
ch

T
1
T
2
(

F
2
).
By Lemma 4.5.5, we know that the canonical map
i :

H
T
1
T
2
(

C; Q)

H
(

C; Q)
is injective. Since
i(ch
T
1
T
2
(

F
2
)) = ch(

F
2
)
= ch

F
1
ch

F
2
=
i(ch
T
1
(

F
1
) ch
T
2
(

F
2
))
we get that
ch
T
1
T
2
(

F
2
) = ch
T
1
(

F
1
) ch
T
2
(

F
2
)
and the conclusion follows since
ch
S
1
(E
1
) = t
ch
T
1
(

F
1
)
and
ch
S
2
(E
2
) = t
ch
T
2
(

F
2
).
4.6 Extension to coherent analytic sheaves
Let X be a topological space and let be a sheaf of ring on X. Recall that
is coherent if the kernel of any morphism of the type
p
|U

|U
(U open subset of X)
is locally of nite type. In this case, a -module T is coherent if and only
if it admits locally presentations of the form
p
1

p
0
T 0.
Moreover, these sheaves form an abelian category denoted (oh(). When
X (resp. M) is a complex (resp. real) analytic manifold, O
X
(resp. /
M
) is
a coherent sheaf of rings and one sets for short
(oh(X) = (oh(O
X
) (resp. (oh(M) = (oh(/
M
) )
and calls coherent analytic sheaves the objects of this category. Note that
among coherent analytic sheaves on X (resp. M), one nds the locally free
O
X
-modules (resp. /
X
-modules). They correspond to the sheaves of com-
plex (resp. real) analytic sections of complex (resp. real) analytic complex
vector bundles on X and form an additive subcategory of (oh(X) (resp.
(oh(M)) which will be denoted Bnd(X) (resp. Bnd(M)).
Although we will not review in details the theory of coherent analytic
sheaves here, we will feel free to use its main results without proof since the
interested reader can nd them in standard texts (see e.g. [12, 13]).
Denition 4.6.1. Let M be a real analytic manifold and let S be a closed
subset of M. We denote
/
b
S
(Bnd(M)) (resp.
T
b
S
((oh(M))) the full tri-
angulated subcategory of
/
b
(Bnd(M)) (resp.
T
b
((oh(M))) formed by the
complexes which are exact on M S.
Lemma 4.6.2. Assume M is a compact real analytic manifold. Then,
(a) For any T

(
b
((oh(M)) there is L

(
b
(Bnd(M)) and a quasi-
isomorphism
u
: c
.
(b) Any exact sequence
0 c
p
. . . c
q
0
with c
p
, . . . , c
q
in Bnd(M) splits.
4.6. Extension to coherent analytic sheaves 175
(c) For any closed subset S of M, the canonical inclusion
Bnd(M) (oh(M)
induces the equivalence of triangulated categories
/
b
S
(Bnd(M))
T
b
S
((oh(M)).
Proof. (a) As is well known (see e.g. [18, Corollary 1.7.8]), it is sucient to
prove that:
(i) For any coherent analytic sheaf T on M, there is an epimorphism
c T
with c Bnd(M).
(ii) There is an integer n such that for any exact sequence
0 c
n
c
n1
. . . c
0
of (oh(M) with c
n1
, . . . , c
0
in Bnd(M) we have c
n
Bnd(M).
Property (i) may be established as follows. Let T be a coherent analytic
sheaf on M. Since M has a complex neighborhood which is a Stein manifold,
Cartans theorem A shows that T is generated by its global sections. Using
the compactness of M and the fact that T is locally of nite type, one nds
global sections s
1
, . . . , s
N
of T inducing an epimorphism
/
N
M
T.
It follows that any object of (oh(M) is a quotient of an object of Bnd(M).
Let us now show that property (ii) holds with n equal to the dimension
of M. Let
0 c
n
c
n1
. . . c
0
be an exact sequence of (oh(M) with c
n1
, . . . , c
0
in Bnd(M). Denote
T the cokernel of the last morphism. The global homological dimension of
/
M
being equal to n, the at dimension of T is not greater than n. Since
c
n1
, . . . , c
0
are at /
M
-modules, it follows that c
n
is also at. Applying
Nakayamas lemma, we see that c
n
is locally free and hence belongs to
Bnd(M) as expected.
(b) Let
0 c
c c
0 (*)
be an exact sequence with c and c
in Bnd(M). Since c
is locally free,
we have
RHom
A
M
(c
, c
) Hom
A
M
(c
, c
).
Moreover, Hom
A
M
(c
, c
) being a coherent analytic sheaf, Cartans theo-

rem B shows that
R(M; Hom
A
M
(c
, c
)) (M; Hom
A
M
(c
, c
)).
Therefore
RHom
A
M
(c
, c
) Hom
A
M
(c
, c
)
and c
is a projective /
M
-module. In particular, the sequence (*) splits.
Since a direct summand of a at module is at, c
is at and hence locally

free. With these results at hand, we can conclude by a simple iteration
procedure.
(c) This follows directly from (a) and (b) thanks to well-known results
of homological algebra.
Proposition 4.6.3. Let M be a compact real analytic manifold and let
S be a closed subset of M. There is a unique way to dene a local Chern
character
ch
S
:
T
b
S
((oh(M)) H
S
(M; Z)
which is invariant by isomorphism and such that
ch
S
(c
) = ch
S
(E
)
if c
is an object of
/
b
S
(Bnd(M)) and E
is the associated complex of com-

plex vector bundles. Moreover, such a local chern character is additive and
multiplicative.
Proof. Thanks to Proposition 4.5.2 and Proposition 4.5.7, the result follows
directly from the preceding lemma.
Remark 4.6.4. It follows easily from the additivity of the local Chern
character introduced in the preceding proposition that
ch
S
(T
) =
kZ
(1)
k
ch
S
(H
k
(T)).
So, the local Chern character for complexes of coherent analytic sheaves may
be reduced to the local Chern character for coherent analytic sheaves. A
similar reduction has however no meaning for complexes of complex vector
bundles since the corresponding cohomology sheaves are not locally free in
general.
4.6. Extension to coherent analytic sheaves 177
Denition 4.6.5. Let M be a compact real analytic manifold, let S be a
closed subset of M and let T
be an object of
T
b
coh
(/
M
) (i.e. a complex of
/
M
-modules with bounded coherent cohomology). Assume T
is supported
by S. We extend the denition of ch
S
by setting
ch
S
(T
) =
kZ
(1)
k
ch
S
(H
k
(T))
where the classes
ch
S
(H
k
(T)) H
S
(M; Z)
are dened according to the preceding proposition.
Proposition 4.6.6. Let M be a compact real analytic manifold and let S
and T be closed subsets of M.
(a) Assume
c
+1
is a distinguished triangle of
T
b
coh
(/
M
) which is supported by S, then
ch
S
(T
) = ch
S
(c
) +ch
S
((
).
(b) Assume c
and T
are objects of
T
b
coh
(/
M
) supported respectively by
S and T. Then c

L
A
M
T
is an object of
T
b
coh
(/
M
) supported by
S T and
ch
ST
(c
L
A
M
T
) = ch
S
(c
) ch
T
(T
)
Proof. Part (a) follows from the long exact sequence of cohomology and
the additivity of the local Chern character for coherent analytic sheaves.
Part (b) follows from the multiplicativity of the local Chern character of
Proposition 4.6.3 combined with the fact that any object of
T
b
coh
(/
M
) is
isomorphic in
T
b
coh
(/
M
) with an object of
T
b
((oh(M)).
Denition 4.6.7. Let X be a compact complex manifold and let S be a
closed subset of X. Assume T
is an object of
T
b
coh
(O
X
) supported by S.
Using the real analytic structure of X, we set
ch
S
(T) = ch
S
(/
X

O
X
T).
Remark 4.6.8. Since /
X
is at over O
X
, the local Chern character in-
troduced in the preceding proposition is clearly invariant by isomorphism,
additive and multiplicative.
5
Riemann-Roch theorem
5.1 Introduction
Let X be a compact complex analytic manifold of complex dimension n. To
x the notations, let us recall that
O
X
= sheaf of holomorphic functions on X
p
X
= sheaf of holomorphic p-forms on X
/
X
= sheaf of meromorphic functions on X
/
(p,q)
X
= sheaf of real analytic forms of bitype (p, q)
B
(p,q)
X
= sheaf of hyperfunction forms of bitype (p, q)
(
(p,q)
,X
= sheaf of smooth forms of bitype (p, q)
Tb
(p,q)
X
= sheaf of distribution forms of bitype (p, q)
/
n
X
= sheaf of real analytic n-forms
B
n
X
= sheaf of hyperfunction n-forms
(
n
,X
= sheaf of smooth n-forms
Tb
n
X
= sheaf of distribution n-forms
These various sheaves enter in the following well-known resolutions.
179
180 5. Riemann-Roch theorem
Proposition 5.1.1.
Real de Rham resolutions. The sequence
0 C
X
(
0
,X
d

d
(
2n
,X
0
is exact and we get analogous exact sequences if we replace (
,X
by
/
X
, B
X
or Tb
X
.
Dolbeault resolutions. The sequence
0
p
X
(
(p,0)
,X

(
(p,n)
,X
0
is exact and we get analogous exact sequences if we replace (
,X
by
/
X
, B
X
or Tb
X
.
Complex de Rham resolution. The sequence
0 C
X
O
X
d

d
n
X
0
is exact.
Since X is a canonically oriented compact topological manifold of di-
mension 2n, we have the two following propositions.
Proposition 5.1.2 (Topological niteness). We have
H
k
(X; C
X
) = 0
if k , [0, 2n] and
dimH
k
(X; C
X
) < +
for any k [0, 2n].
Remark 5.1.3. It follows from the preceding proposition that the complex
Betti numbers
b
k
(X) = dimH
k
(X; C
X
) (k Z)
are well-dened topological invariants associated to X. Another topological
invariant of X is its Euler-Poincare characteristic. Note that it follows from
the universal coecient formula that
(X) =
kZ
(1)
k
b
k
(X).
5.1. Introduction 181
Proposition 5.1.4 (Topological duality). By Poincare duality, we have
an isomorphism
H
k
(X; C
X
)
H
2nk
(X; C
X
)
which can be made more explicit by means of the pairing
, ) : (X; Tb
2nk
X
)
c
(X; (
k
,X
) C
dened by the formula
u, ) =
_
u .
Remark 5.1.5. It follows directly from the preceding proposition that
b
2nk
(X) = b
k
(X)
for any k Z. In particular,
(X) = (1)
n
b
n
(X) +2
n1
k=0
(1)
k
b
k
(X)
so that
(X) b
n
(X) (mod 2).
Finiteness and duality results also holds for coherent analytic sheaves.
Proposition 5.1.6. For any coherent analytic sheaf T on X, we have
H
k
(X; T) = 0
if k , [0, n] and
dimH
k
(X; T) < +
if k [0, n].
Proof. We will only give a sketch of the proof for the case where T is locally
free. For the general case, we refer to standard texts on complex analytic
geometry.
By tensorization on O
X
with the Dolbeault resolution
0 O
X
(
(0,0)
,X
(
(0,n)
,X
0
we obtain the resolution
0 T T
O
X
(
(0,0)
,X
T
O
X
(
(0,n)
,X
0.
Since T
O
X
(
(0,k)
,X
is locally isomorphic to a direct sum of a nite number
of copies of (
(0,k)
,X
, it is a soft sheaf. Therefore,
R(X; T) (X; T
O
X
(
(0,)
,X
).
The vanishing part of the result follows easily. To obtain the niteness part,
we consider the resolution
0 T T
O
X
/
(0,0)
X
T
O
X
/
(0,n)
X
0
obtained by applying the exact functor T
O
X
to the Dolbeault resolution
0 O
X
/
(0,0)
X
/
(0,n)
X
0.
Since T
O
X
/
(0,k)
X
is a coherent real analytic sheaf on X, it is acyclic.
Hence,
R(X; T) (X; T
O
X
/
(0,)
X
)
and the canonical morphism
(X; T
O
X
/
(0,)
X
) (X; T
O
X
(
(0,)
,X
)
is a quasi-isomorphism. One checks easily that the components of the rst
complex are (DFS) spaces and that the components of the second complex
are (F) spaces. Since the morphism from the left complex to the right
complex is clearly continuous, the conclusion follows from the next lemma.
Lemma 5.1.7. Let
u
: E
be a morphism of complexes of locally convex topological vector spaces.

Assume that for any k Z,
(i) E
k
is a (DFS) space;
(ii) F
k
is a (F) space;
(iii) H
k
(u
) is surjective.
Then, H
k
(F
) is nite dimensional.
Sketch of proof. The basic idea is to write E
k
= lim
iN
E
k
i
where E
k
i
is a Ba-
nach space; the transitions being compact and then to use Baires theorem
to reduce the result to Schwartz compact perturbation Lemma.
Proposition 5.1.8. For any coherent analytic sheaf T on X, we have the
canonical isomorphisms
H
k
(X; T)
H
nk
(X; RHom
O
X
(T,
X
))
where
X
denotes as usual the sheaf of holomorphic n-forms.
Proof. We will only sketch the proof in the case where T is locally free.
Working as in the proof of Proposition 5.1.6, we get the two isomor-
phisms
R(X; T) (X; T
O
X
(
(0,)
,X
)
and
R(X; RHom
O
X
(T,
X
)) (X; RHom
O
X
(T, Tb
(n,)
X
)).
Consider the canonical pairing
[T
O
X
(
(0,k)
,X
]
C
[RHom
O
X
(T, Tb
(n,nk)
X
)] Tb
(n,n)
X
which sends (f w, h) to w h(f). Combining it with the integration of
distributions, we get the pairing
(X; T
O
X
(
(0,k)
,X
)
C
(X; RHom
O
X
(T, Tb
(n,nk)
X
)) C
and hence the morphism of complexes
(X; T
O
X
(
(0,)
,X
)
(X; RHom
O
X
(T, Tb
(n,n)
X
)).
Using the denition of distributions, one checks easily that this morphism
is in fact an isomorphism. The cohomology of the complexes involved being
nite dimensional, the dierentials are strict and we get
H
k
((X; T
O
X
(
(0,)
,X
))
H
nk
((X; RHom
O
X
(T, Tb
(n,)
X
))).
Example 5.1.9. Since
p
X
is a coherent analytic sheaf, Proposition 5.1.6
shows that the Hodge numbers
h
p,q
(X) = dimH
q
(X;
p
) (q Z)
are well-dened holomorphic invariants of X which vanish for q , [0, n]. Of
course, we have
p
(X) := (X;
p
X
) =
qZ
(1)
q
h
p,q
.
Moreover, it follows from the complex de Rham resolution that
(X) =
pZ
(1)
p
p
(X).
Note that the canonical morphism
:
p
X

np
X

n
X
np
X
Hom
O
X
(
p
X
,
n
X
).
Hence, applying the duality result for coherent analytic sheaves, we see that
h
np,q
= h
p,nq
.
The h
p,q
have many other interesting properties. However, since they are
related to Hodge theory we will not review them here.
Thanks to the results in this section, one sees that the Euler-Poincare
characteristic
(X; T) =
k=0
(1)
k
dimH
k
(X; T)
makes sense for any coherent analytic sheaf on a compact complex ana-
lytic manifold. The computation of this number forms what one may call
the generalized Riemann-Roch problem (its link with the original results
of Riemann and Roch will be explained in Sections 24 below). In [15],
Hirzebruch solved this problem when X is a projective manifold and T is
the sheaf of complex analytic sections of a complex analytic vector bundle
F of rank r by expressing the number (X; T) by means of a formula of
the type
(X; T) =
_
X
P(c
1
(F), , c
r
(F), c
1
(TX), , c
n
(TX))
where P is a polynomial with rational coecients which depends only on r
and n. In Sections 58 we will establish the following slightly more general
result.
Theorem 5.1.10. Let X be a projective complex analytic manifold of
dimension n and let T be a coherent analytic sheaf on X. Then,
(X; T) =
_
X
ch(T) td(TX).
To better understand the link between this formula and the preceding
one let us write it explicitly in terms of Chern classes in a two simple cases.
Case of line bundles on curves.
We need to compute ch(E) and td(TX) in degrees 0 and 2. We have
x
1 e
x
=
x
1 (1 + (x) +
(x)
2
2
+o(x
2
))
=
1
1
x
2
+ o(x)
= 1 +
x
2
+ o(x)
and
e
x
= 1 + x +o(x).
Therefore, using the fact that E and TX are line bundles, we get
ch(E) = 1 + c
1
(E)
td(TX) = 1 +
c
1
(TX)
2
.
Hence,
[ch(E) td(TX)]
2
= c
1
(E) +
c
1
(TX)
2
so that
(X; c) =
_
X
c
1
(E) +
c
1
(TX)
2
.
Case of line bundles on surfaces.
We need to compute ch(E) and td(TX) in degrees 0, 2 and 4. We have
x
1 e
x
=
x
1 (1 + (x) +
(x)
2
2
+
(x)
3
6
+o(x
3
))
=
1
1
x
2
+
x
2
6
+o(x
2
)
= 1 +
x
2
+
x
2
12
+ o(x
2
)
and
e
x
= 1 + x +
x
2
2
+o(x
2
).
It follows that we have at order 2
x
1
1 e
x
1
x
2
1 e
x
2
=
_
1 +
x
1
2
+
x
2
1
12
__
1 +
x
2
2
+
x
2
2
12
_
= 1 +
x
2
2
+
x
2
2
12
+
x
1
2
+
x
1
x
2
4
+
x
2
1
12
= 1 +
x
1
+x
2
2
+
x
2
1
+3x
1
x
2
+ x
2
2
12
= 1 +
(x
1
+x
2
)
2
+
(x
1
+ x
2
)
2
+ x
1
x
2
12
= 1 +
S
2,1
(x
1
, x
2
)
2
+
S
2
2,1
(x
1
, x
2
) +S
2,2
(x
1
, x
2
)
12
.
Therefore,
td(TX) = 1 +
c
1
(TX)
2
+
c
1
(TX)
2
+ c
2
(TX)
12
and
ch(E) = 1 +c
1
(E) +
c
1
(E)
2
2
.
Finally,
[ch(E) td(TX)]
4
=
c
1
(TX)
2
+c
2
(TX)
12
+
c
1
(E)c
1
(TX)
2
+
c
1
(E)
2
2
and
(X; c) =
_
X
c
1
(TX)
2
+ c
2
(TX)
12
+
c
1
(E)c
1
(TX)
2
+
c
1
(E)
2
2
.
5.2 Cohomology of compact complex curves
Let X be a connected compact (smooth) complex curve.
Proposition 5.2.1. There is a natural integer g such that
H
0
(X; O
X
) C H
0
(X;
X
) C
g
H
1
(X; O
X
) C
g
H
1
(X;
X
) C
H
k
(X; O
X
) 0 (k 2) H
k
(X;
X
) 0 (k 2)
H
0
(X; C
X
) C
H
1
(X; C
X
) C
2g
H
2
(X; C
X
) C
H
k
(X; C
X
) 0 (k 3)
5.2. Cohomology of compact complex curves 187
Proof. Writing the long exact sequence of cohomology associated to the
short exact sequence
0 C
X
O
X

X
0,
0

H
0
(X; C
X
)

H
0
(X; O
X
)

H
0
(X;
X
)
-, *+
/.
()

H
1
(X; C
X
)

H
1
(X; O
X
)

H
1
(X;
X
)
-, *+
/.
()

H
2
(X; C
X
)

0.
Since X is connected, H
0
(X; C
X
) C and since X is compact, the maxi-
mum principle shows that H
0
(X; O
X
) C. Therefore, the morphism
H
0
(X; C
X
) H
0
(X; O
X
) (*)
is an isomorphism. By duality, we see that
H
2
(X; C
X
) C, H
1
(X;
X
) C
and that the morphism
H
1
(X;
X
) H
2
(X; C
X
)
is the dual of the isomorphism (*). It follows that the sequence
0 H
0
(X;
X
) H
1
(X; C
X
) H
1
(X; O
X
) 0
is exact. Since
H
1
(X; O
X
) H
0
(X;
X
)
,
the spaces H
1
(X; O
X
) and H
0
(X;
X
) have the same dimension g. It follows
that H
1
(X; O
X
) C
g
, H
0
(X;
X
) C
g
and that H
1
(X; C
X
) C
2g
.
Denition 5.2.2. The integer g which appears in the preceding table is
called the genus of the Riemann surface X. One denes it classically as the
maximal number of linearly independent holomorphic 1-forms.
Remark 5.2.3. Note that since 2g is the rst Betti number of X, g is a
topological invariant of X. One can show that this is the only invariant of
this kind. Note also that it follows from the cohomology table of X that
(X) = 2 2g.
5.3 Divisors on complex curves
Let A be a commutative integral ring and let K be its eld of quotients.
Denote A
the multiplicative monoid formed by the non-zero elements of

A. Dene the divisibility relation [ on A
by
a[b c A
with b = a c.
This is clearly a preorder relation compatible with the multiplication of
A
. Since A is integral, any element of A
is cancelable and A
has
an associated group. This group may be identied with the group K
of
invertible (i.e. non-zero) elements of K. The relation [ induces a preorder
relation on K
dened by
f[g c A
with g = f c g/f A
.
Remark that if f, g K
are such that

f[g and g[f
then g/f A
and conversely. If follows that D = K
/A
has a natural
structure of ordered commutative group. We denote + the group operation
of D and we denote its canonical order relation. We denote (f) the
element of D associated to f K
. The ordered group (D, +, ) is called

the group of principal divisors of the ring A. This group enters in the exact
sequence
1 A
D 0.
In the case of a complex curve X, we may consider by analogy the sheaves
of commutative groups O
X
(resp. /
X
) of invertible elements of O
X
(resp.
/
X
) and we may dene the sheaf of abelian groups Tiv
X
as the cokernel of
the canonical inclusion
O
X
/
X
.
It follows from the exact sequence
1 O
X
/
X
Tiv
X
0
that (Tiv
X
)
x
is the group of principal divisors of the integral ring (O
X
)
x
.
So, it is canonically endowed with an order relation ()
x
. Hence, we get an
order relation on the sections of Tiv
X
. A divisor on X is the data of a
global section D of Tiv
X
. A divisor coming from a global section f of /
X
is
called principal, we denote it (f). It follows from the construction of Tiv
X
5.3. Divisors on complex curves 189
that a divisor on X is locally principal. It is however not globally principal
in general. To study this phenomenon, one introduces the quotient group
(X; Tiv
X
)/(X; /
X
).
This group is called the Picard group of X and denoted Pic(X). Two divisor
which have the same image in Pic(X) are said to be equivalent.
The structure of (Tiv
X
)
x
is very simple. As a matter of fact, if t is a
local coordinate such that t(x) = 0, any meromorphic function f may be
uniquely written in a neighborhood of x as
f = t
n
h
where n Z and h O
X,x
(consequence of Taylor formula). It follows that
(f) = n(t). Moreover, we see easily that if t
is another local coordinate such

that t
(x) = 0, we have (t) = (t
). Therefore, (t) is a canonically dened

divisor supported by x. We denote it by [x] and we denote ord
x
(f) the
unique integer n such that
(f) = n[x].
We may sum up what precedes by saying that
ord
x
: (Tiv
X
)
x
Z
is an isomorphism of ordered abelian groups. Any divisor D of X being
locally principal, x : ord
x
(D) ,= 0 is locally nite; hence nite (X is
compact). It follows that
D =
xsupp D
ord
x
(D)[x].
Thus, there is a bijection between the ordered group of divisors of X and
the ordered group
Z
(X)
.
We dene the degree of the divisor D by the formula
deg D =

xsupp D
ord
x
(D).
To a divisor D on X, one associates the sheaf O
X
(D) dened by setting
O
X
(D)(U) = f /
X
(U) : (f) D[
U
0
for any connected open subset U of X. One checks easily that O
X
(D) is an
O
X
-submodule of /
X
. Consider a point x X. This point has a connected
neighborhood U on which D is principal. So, we nd a meromorphic func-
tion g on U such that D[
U
= (g). It follows that f O
X
(D)(U) if and only
if f = 0 or (f) (g). Hence,
O
X
(D)(U) =
h
g
: h O
X
(U)
and O
X
(D)[
U
is the free O
X
-submodule of rank 1 of /
X
generated by 1/g.
Therefore, O
X
(D) is a locally free O
X
-module of rank 1. Remark that if
D, D
are two divisors on X such that

D
= D + (g)
with g (X; /
X
), then
O
X
(D
) O
X
(D)
f fg
is an isomorphism of O
X
-modules. Therefore, the isomorphy class of O
X
(D)
depends only on the class of D in Pic(X).
5.4 Classical Riemann and Roch theorems
Denition 5.4.1. Let L be a locally free O
X
-module of rank 1 on X.
Consider a non-zero meromorphic section s of L, i.e. assume that
s (X; /
X

O
X
L) 0.
Consider a neighborhood U of x X on which L is isomorphic to O
X
and
denote l a generator of L[
U
. Clearly, we have
s[
U
= f l
where f is a meromorphic function on U uniquely determined by this rela-
tion. If l
is another generator of L[
U
, we have l = hl
where h O
X
(U)
and
s[
U
= f
where f
/
X
(U) is given by f
= fh. It follows that (f) = (f
) and
that this divisor depends only on s[
U
. We denote it (s[
U
). Using a gluing
process, we obtain a well-dened divisor (s) on X.
From this denition it follows easily that
5.4. Classical Riemann and Roch theorems 191
Proposition 5.4.2. We have an O
X
-linear isomorphism
L
O
X
O
X
(D) L(D)
where L(D) is the subsheaf of /
O
X
L dened by setting
L(D)(U) = s (U; /
X
O
X
L) 0 : (s) D 0
for any connected open subset U of X.
Proposition 5.4.3. For any locally free O
X
-module L of rank 1 and any
divisor D on X, we have
(X; L(D)) = (X; L) + deg D.
In particular,
dimH
0
(X; L(D)) (X; L) + deg D.
Proof. Let D and D
be two divisors on X such that D D
. By denition,
it is clear that O
X
(D) is a subsheaf of O
X
(D
). Denote Qthe quotient sheaf

O
X
(D
)/O
X
(D). Write D, D
as
D =
m
j=1
n
j
[x
j
], D
=
m
j=1
n
j
[x
j
]
where x
1
, , x
m
are points of X and n
1
, . . . , n
m
; n
1
, . . . , n
m
are elements
of Z (which may be equal to 0). One sees easily that Q
x
= 0 if x /
x
1
, . . . , x
m
. Now, consider a coordinate neighborhood U of a point x
j

x
1
, , x
m
such that U x
1
, , x
m
= x
j
. Let t be a coordinate on
U such that t(x
j
) = 0. Since D[
U
= n
j
(t) and D
[
U
= n
j
(t), we see that
O
X
(D)
x
j
=
h
t
n
j
: h (O
X
)
x
j
, O
X
(D
)
x
j
=
h
t
n
j
: h
(O
X
)
x
j
.
Since n
j
n
j
, we may write
h
= a
0
+a
1
t + + a
n
j
n
j
1
t
n
j
n
j
1
+t
n
j
n
j
h
where a
0
, , a
n
j
n
j
1
C and h
(O
X
)
j
are uniquely determined by
this condition. It follows that
Q
x
j
= (O
X
(D
)/O
X
(D))
x
j
C
n
j
n
j
.
The sheaf Q is thus supported by x
1
, , x
m
and we have
dimQ
x
j
= n
j
n
j
(1 j m).
Applying L
O
X
to the exact sequence
0 O
X
(D) O
X
(D
) Q 0
and taking Euler-Poincare characteristics, we see that
(X; L(D)) (X; L(D
)) +(X; L
O
X
Q) = 0
and hence that
(X; L(D)) (X; L(D
)) +deg D
deg D = 0.
Therefore,
(X; L(D
)) deg D
= (X; L(D)) deg D.

If D, D
are divisors on X, there is a divisor D
D, D
. It follows
from what precedes that
(X; L(D)) deg D
does not depend on D. The conclusion follows since
L(0) L.
Corollary 5.4.4. For any divisor D on X, we have
(X; O
X
(D)) = deg D +1 g.
In particular, deg D depends only on the class of D in Pic(X) and
dimH
0
(X; O
X
(D)) deg D +1 g.
Proof. This follows directly from the preceding proposition since
(X; O
X
) = dimH
0
(X; O
X
) dimH
1
(X; O
X
) = 1 g.
Remark 5.4.5. Let x
1
, , x
m
be distinct points of X. The problem which
led to the classical Riemann and Roch theorems was the computation of the
dimension of the space of meromorphic functions which are holomorphic on
X x
1
, , x
m
and have poles of order lower or equal to 1 at x
1
, , x
m
.
One checks easily that this dimension is nothing but
dimH
0
(X; O
X
(D))
for D = [x
1
] + + [x
m
]. Therefore, the preceding corollary contains Rie-
manns inequality in its original form.
Proposition 5.4.6. Let L be a locally free O
X
-module of rank 1. Fix
x X. Then, L has meromorphic sections whose only pole is at x. In
particular, L has non-zero meromorphic sections.
Proof. In the preceding proposition, set D = m[x] and choose m large
enough in order that (X; L) + deg D > 0.
Corollary 5.4.7. Let L be a locally free O
X
-module of rank 1. Then, there
is a divisor D on X and an isomorphism
O
X
(D) L.
Moreover, although D is not unique its class in Pic(X) depends only on L.
Proof. Let s be a non-zero meromorphic section of L. Set D = (s). Working
on the open subsets where L is trivial, we see easily that
O
X
(D) L
f fs
is a well-dened isomorphism. To see that the class of D in Pic(X) depends
only on L, it is sucient to note that non-zero meromorphic sections s, s
of L are always linked by a relation of the type

s
= gs
with g (X; /
X
) and that such a relation entails that
(s
) = (s) +(g).
Denition 5.4.8. Let L be a locally free O
X
module of rank 1 on X. We
denote [L] the class in Pic(X) of any divisor D such that
L O
X
(D).
Example 5.4.9. Since
X
is a locally free O
X
-module of rank 1, we may
consider
[
X
].
This class is traditionally called the canonical class of divisors of X and
denoted by K.
Remark 5.4.10. It follows from what precedes that
D O
X
(D)
induces a bijection between Pic(X) and the set of isomorphy classes of
locally free O
X
-module of rank 1 on X whose inverse is the map induced
by
L [L].
Since
O
X
(D)
O
X
O
X
(D
) O
X
(D +D
),
the preceding bijection is in fact an isomorphism of abelian groups.
As a consequence of this isomorphism we get that
(X; L) = deg[L] + 1 g
for any locally free O
X
module of rank 1 on X
Let us now look at Rochs original contribution.
Denition 5.4.11. We denote
: Pic(X) N
the map induced by
D dimH
0
(X; O
X
(D)).
Example 5.4.12. We have
(0) = dimH
0
(X; O
X
) = 1
and
(K) = dimH
0
(X;
X
) = g.
Proposition 5.4.13. For any divisor D on X, we have
([D]) (K [D]) = deg D + 1 g
Proof. From the duality theorem for coherent analytic sheaf we deduce that
H
1
(X; O
X
(D)) H
0
(X; Hom
O
X
(O
X
(D),
X
)).
Using the isomorphism
Hom
O
X
(O
X
(D),
X
) O
X
(D)

X
,
and the properties of [], we see that
[Hom
O
X
(O
X
(D),
X
)] = K [D].
It follows that
dimH
1
(X; O
X
(D)) = (K D).
Since
(X; O
X
(D)) = deg D +1 g,
the proof is complete.
Corollary 5.4.14. We have
deg K = 2g 2.
Proof. Applying the preceding proposition for D = K, we obtain
(K) (O) = deg K + 1 g.
Therefore,
g 1 = deg K +1 g
Proposition 5.4.15.
(a) If deg D < 0, then
dimH
0
(X; O
X
(D)) = 0 and dimH
1
(X; O
X
(D)) = g 1 deg D.
(b) If deg D > 2g 2, then
dimH
0
(X; O
X
(D)) = deg D +1 g and dimH
1
(X; O
X
(D)) = 0.
Proof. (a) Thanks to Riemann-Roch formula, it is sucient to show that
H
0
(X; O
X
(D)) = 0.
Let us proceed by contradiction. Assume f is a non-zero section of O
X
(D).
Then,
(f) D
and
deg(f) +deg D 0.
Since [(f)] = 0, we have deg(f) = 0 and we get deg D 0. The conclusion
follows.
(b) As in (a), we have only to prove the rst equality. Using Proposi-
tion 5.4.13, we see that this equality is true if
(K [D]) = 0.
Since by the preceding corollary we have deg(K [D]) = 2g 2 deg(D),
this follows from directly from part (a).
Remark 5.4.16. Note that the second part of the preceding proposition
contains a complete answer to the original Riemann-Roch problem for high
degree divisors.
Proposition 5.4.17. For any divisor D on X, we have
deg D =
_
X
c
1
(O
X
(D)).
Proof. Assume D =
m
i=1
n
i
[x
i
] where x
1
, . . . , x
m
are distinct points of X.
For any i 1, . . . , m, let t
i
be local coordinate dened on an open neigh-
borhood U
i
of x
i
such that t
i
(x
i
) = 0, t
i
(U
i
) D(0, 1) and assume that
U
1
, . . . , U
m
are disjoint. Set U
0
= Xx
1
, . . . , x
m
and denote | the cover-
ing U
0
, . . . , U
m
. We know that O
X
(D) is generated on U
0
, U
1
, . . . , U
m
by
the functions 1, t
n
1
1
, . . . , t
n
m
m
. So, we may represent the class of O
X
(D)
in H
1
(X; O
X
)

H
1
(|, O
X
) by the

Cech cocycle g dened by setting
g
U
0
U
1
= t
n
1
1
, g
U
0
U
2
= t
n
2
2
, . . . , g
U
0
U
m
= t
n
m
m
.
Thanks to the proof of Proposition 4.2.10, we also know that the diagram
H
1
(X, O
X
)
dlog
c
1

H
1
(X,
1
)
H
2
(X, C)
is commutative. It follows that c
1
(O
X
(D)) is the image in H
2
(X; C) of the
Cech cocycle of

H
1
(|,
1
) dened by
c
1
U
0
U
1
= dlogg
U
0
U
1
= n
1
dt
1
t
1
.
.
.
c
1
U
0
U
m
= dlogg
U
0
U
m
= n
m
dt
m
t
m
.
The Dolbeault resolution
0
1
(
(1,0)
(
(1,1)
0
H
1
(X,
1
) (X, (
(1,1)
)/(X; (
(1,0)
).
We may follow it in terms of

Cech cohomology thanks to the Weil Lemma.
We have to follow the dotted path in the diagram
(X;
1
)

(X; (
(1,0)
(X; (
(1,1)
C
0
(|,
1
)

C
0
(|, (
(1,0)
C
0
(|, (
(1,1)
C
1
(|,
1
)

C
1
(|, (
(1,0)
C
1
(|, (
(1,1)
.
.
.
.
.
.
.
.
.
To write c
1
as the

Cech coboundary of c
0

C
0
(|, (
(1,0)
), we will use a
partition unity (
U
)
UU
relative to | and set
c
0
U
0
=
U
1
c
1
U
0
U
1

U
m
c
1
U
0
U
m
c
0
U
1
=
U
0
c
1
U
0
U
1
.
.
.
c
0
U
m
=
U
0
c
1
U
0
U
m
Then, we may represent the image of c in H
2
(X, C) by the

Cech 0-cocycle
of

C
0
(|, (
(1,1)
) given by
U
0
dc
0
U
0
.
.
.
U
m
dc
0
U
m
which corresponds to the dierential form with compact support
= d
U
1
n
1
dt
1
t
1
d
U
m
n
m
dt
m
t
m
.
It follows that
_
X
c
1
(O
X
(D)) =
_
X
=
m
i=1
n
i
_
U
i
d
U
i

dt
i
t
i
.
Denote V
i
the image in U
i
of a disk centered on 0 whose radius is suciently
small in order that
U
i
= 1 on V
i
U
i
. Stokes theorem shows that
_
X
c
1
(O
X
(D)) =
m
i=1
n
i
_
U
i
\V
i
d(
U
i
dt
i
t
i
)
=
m
i=1
n
i
_
V
i
U
i
dt
i
t
i
=
m
i=1
n
i
_
V
i
dt
i
t
i
= deg D.
Corollary 5.4.18. For any locally free O
X
-module L of rank 1 on X, we
have
(X; L) =
_
X
c
1
(L) +
c
1
(TX)
2
.
Proof. Since TX is a holomorphic line bundle, we know that c
1
(TX) is the
Euler class of TX and hence of X. Therefore,
_
X
c
1
(TX) = (X, C) = 1 2g + 1 = 2(1 g)
and
1 g =
_
X
c
1
(TX)
2
.
Since by the preceding proposition, we have
deg([L]) =
_
X
c
1
(L),
Remark 5.4.10 allows us to conclude.
Remark 5.4.19. Thanks to the computations at the end of Section 1, the
preceding corollary shows that Hirzebruch-Riemann-Roch formula is true
for compact complex curves.
5.5 Cohomology of coherent analytic sheaves on P
n
(C)
Denition 5.5.1. Let
q : C
n+1
0 P
n
(C)
be the canonical projection. For any open subset U of P
n
(C) and any l Z,
set
O
P
n
(C)
(l)(U) = f O
C
n+1 (q
1
(U)) : f(z) =
l
f(z) ( C
).
5.5. Cohomology of coherent analytic sheaves on P
n
(C) 199
From this denition it follows immediately that :
Proposition 5.5.2. For any l Z,
U O
P
n
(C)
(l)(U)
is a locally free O
P
n
(C)
-module of rank 1. Moreover, there are canonical
isomorphisms
O
P
n
(C)
(l)
O
P
n
(C)
O
P
n
(C)
(l
) O
P
n
(C)
(l +l
),
Hom
O
P
n
(C)
(O
P
n
(C)
(l), O
P
n
(C)
(l
)) O
P
n
(C)
(l
l).
Exercise 5.5.3. There is a canonical isomorphism
P
n
(C)
O
P
n
(C)
(n 1).
Solution. As usual, set
U
k
= [z
0
, , z
n
] : z
k
,= 0
and dene the map u
kl
: U
k
C by setting
u
kl
=
z
l
z
k
.
Then, (u
k0
, , u
kk
, , u
kn
) gives a biholomorphic bijection between U
k
and C
n
. It follows that any (U;
P
n
(C)
) may be written in a unique
way on U U
k
as
= f
k
du
k0

du
kk
du
kn
with f
k
O
P
n
(C)
(U U
k
). For l < k, we have
u
km
=
u
lm
u
lk
.
Therefore,
du
km
=
u
lk
du
lm
u
lm
du
lk
u
2
lk
and
du
k0

du
kk
du
kn
=
k1
m=0
du
l0
u
lk

u
lm
du
lk
u
2
lk

du
lk
u
lk

du
ln
u
lk
+
n
m=k+1
du
l0
u
lk

du
lk
u
lk

u
lm
du
lk
u
2
lk

du
ln
u
lk
.
Since du
ll
= 0, we get
du
k
0

du
kk
du
kn
=
du
l0
u
lk

u
ll
du
lk
u
2
lk

du
lk
u
lk

du
ln
u
lk
=
(1)
kl
u
n+1
lk
du
l
0

du
ll
du
ln
.
It follows that
f
l
= f
k
(1)
kl
u
n+1
lk
.
Hence,
(1)
l
z
n+1
l
f
l
([z
0
, , z
n
]) =
(1)
k
z
n+1
k
f
k
([z
0
, , z
n
])
on q
1
(UU
k
U
l
). This shows that there is a unique h (U; O
P
n
(C)
(n
1)) such that
h(z
0
, z
n
) =
(1)
k
z
n+1
k
f
k
([z
0
, , z
n
])
on q
1
(U U
k
) for k 0, , n. We have thus constructed a canonical
morphism
P
n
(C)
O
P
n
(C)
(n 1).
Since it is clearly an isomorphism, the proof is complete.
Exercise 5.5.4. Denote |
n
(C) the sheaf of holomorphic sections of the
universal bundle U
n
(C) on P
n
(C). Then, there is a canonical isomorphism
|
n
(C) O
P
n
(C)
(1).
Solution. Using the same notations as in the proof of the preceding propo-
sition, set
s
k
([z
0
, , z
n
]) =
_
(
z
0
z
k
, ,
z
k
z
k
, ,
z
n
z
k
), [z
0
, , z
n
]
_
.
Clearly, s
k
is a holomorphic frame of U
n
(C)
|U
k
. Therefore, for any
(U; |
n
(C)) there is a unique
k
(U U
k
; O
P
n
(C)
) such that
=
k
s
k
on U U
k
. Since
s
k
= u
kl
s
l
n
(C) 201
on U
k
U
l
, we have
l
=
k
u
kl
on U U
k
U
l
. It follows that
l
([z
0
, , z
n
])
z
l
=

k
([z
0
, , z
n
])
z
k
on q
1
(U U
k
U
l
) and hence that there is a unique h (U; O
P
n
(C)
(1))
such that
h(z
0
, , z
n
) =

k
([z
0
, , z
n
])
z
k
on q
1
(U U
k
). The conclusion follows as in the preceding proposition.
Theorem 5.5.5. The cohomology of the sheaf O
P
n
(C)
(l) is given by the
table :
H
0
(P
n
(C); O
P
n
(C)
(l)) =
_
C
(
n+l
l
)
if l 0
0 otherwise
H
k
(P
n
(C); O
P
n
(C)
(l)) = 0 if 0 < k < n
H
n
(P
n
(C); O
P
n
(C)
(l)) =
_
C
(
l1
n
)
if l n 1
0 otherwise
In particular,
(P
n
(C); O
P
n
(C)
(l)) =
(l +1) (l +n)
n!
for any l Z.
Proof. We will only do the easy part and prove the rst equality. We refer
to [26] and [27] for the other results. Note that it follows from denitions
that
H
0
(P
n
(C); O
P
n
(C)
(l)) = f O
C
n+1 (C
n+1
0) : f(z) =
l
f(z) ( C
).
Since the codimension of 0 in C
n+1
is at least 2, any
f H
0
(P
n
(C); O
P
n
(C)
(l))
is in fact holomorphic on C
n+1
and may thus be written as
f(z) =
0
=0

n
=0
a
0
, ,
n
z
0
0
z
n
n
with a
0
, ,
n
C. From the relation
f(z) =
l
f(z) ( C
),
we deduce that a
0
, ,
n
||
= a
0
, ,
n
l
and hence that a
0
, ,
n
= 0 if
[[ ,= l. It follows that H
0
(P
n
(C); O
P
n
(C)
(l)) is 0 if l < 0 and equal to the
space of homogeneous polynomials of degree l for l 0. In this case, the
dimension is given by
d
n,l
= #(
0
, ,
n
) :
0
0, ,
n
0,
0
+ +
n
= l.
To compute this number, we note that, since
_

0
=0
z
0
_

_

n
=0
z
n
_
=
0
=0

n
=0
z
0
++
n
,
we have
1
(1 z)
n+1
=
l=0
d
n,l
z
l
.
Hence,
d
n,l
=
1
l!
_
1
(1 z)
n+1
_
(l)
z=0
=
(n +1)(n + 2) (n + l)
l!
=
_
n + l
l
_
.
To conclude, note that it follows from the cohomology table that
(P
n
(C); O
P
n
(C)
(l)) =
_
_
_
n+l
l
_
if l 0
0 if n l 1
(1)
n
_
l1
n
_
if l n 1
and the announced formula holds since for l n 1, we have
(1)
n
(l n) (l 1)
n!
=
(l + 1) (l + n)
n!
.
Remark 5.5.6. Note that, using the duality theorem for coherent analytic
sheaves, we have
H
k
(P
n
(C); O
P
n
(C)
(l)) H
nk
(P
n
(C); Hom
O
P
n
(C)
(O
P
n
(C)
(l),
P
n
(C)
))
H
nk
(P
n
(C); O
P
n
(C)
(l 1 n)).
Therefore, the last isomorphism in the preceding theorem follows from the
rst since
_
n l 1 n
l 1 n
_
=
_
l 1
n
_
if l n 1.
n
(C) 203
Corollary 5.5.7. The graded C-algebra
S =
lZ
(P
n
(C); O
P
n
(C)
(l))
is canonically isomorphic to C[z
0
, , z
n
].
Denition 5.5.8. Recall that the homogeneous part of degree l of a graded
S-module M is denoted M
l
and that M(l) is the graded S-module charac-
terized by the fact that M(l)
l
= M
l+l
(l
Z).
We denote o the sheaf of graded C-algebras dened by setting
o =
lZ
O
P
n
(C)
(l).
For any graded S-module M, consider the graded tensor product
o
S
M.
This is clearly a graded o-module. We denote

M its homogeneous part
of degree 0. By construction,

M is an O
P
n
(C)
-module. Finally, we denote
T^ (resp. TT) the thick subcategory of /od(S) formed by the graded
S-modules M such that there is l
0
Z with

ll
0
M
l
isomorphic to zero
(resp. of nite type).
M

M
from the category of graded S-modules to that of O
P
n
(C)
-modules is exact.
Moreover,
(a)

S(l) = O
P
n
(C)
(l),
(b)

M = 0 if M is an object of T^,
(c)

M is coherent if M is an object of TT.
Proof. The dicult part is the exactness, for it we refer the reader to [26]
and [27]; the other parts are easier. As a matter of fact, (b) follows at once
from the fact that for x U
k
, any section of (

M)
x
of the form
l
0
m
l
0
with
l
0
(o
l
0
)
x
, m
l
0
M
l
0
is equal to
l
0
z
l
0
l
k
z
ll
0
k
m
l
0
for any l Z. As for (c), we know that there is a morphism
R
0
r
0
=1
S(l
r
0
) M
with a cokernel in T^. Since S is noetherian, we get an exact sequence of
the type
R
1
r
1
=1
S(l
r
1
)
R
0
r
0
=1
S(l
r
0
) M N 0
with N T^. This gives us the exact sequence of O
P
n
(C)
-modules
R
1
r
1
=1
O
P
n
(C)
(l
r
1
)
R
0
r
0
=1
O
P
n
(C)
(l
r
0
)

M 0
Denition 5.5.10. Let T be a coherent analytic sheaf on P
n
(C). We set
T(l) = O
P
n
(C)
(l)
O
P
n
(C)
T
and denote S(T) the graded S-module
lZ
(P
n
(C); T(l)).
Proposition 5.5.11. For any coherent analytic sheaf T on P
n
(C), the
graded S-module S(T) is an object of TT. Moreover, the functor
T S(T)
induces an equivalence between the category of coherent analytic sheaves
on P
n
(C) and the abelian category TT/T^. The quasi-inverse of this
equivalence is induced by the functor
M

M.
Proof. We refer the reader to [26] and [27].
Corollary 5.5.12. Let T be a coherent analytic sheaf on P
n
(C). Then,
(a) there is an exact sequence of the form
0
R
p
r
p
=0
O
P
n
(C)
(l
r
p
)
R
0
r
0
=0
O
P
n
(C)
(l
r
0
) T 0;
n
(C) 205
(b) the O
P
n
(C)
-module T(l) is generated by its global sections for l 0;
(c) we have
H
k
(P
n
(C); T(l)) = 0 (k > 0)
for l 0.
Proof. (a) We know that S(T) is an object of TT. It follows that there is
a morphism
R
0
r
0
=0
S(l
r
0
) S(T)
with a cokernel N in T^. Since its kernel K is a graded submodule of a
graded S-module of nite type, Hilbert syzygies theorem shows that K has
a presentation of the form
0
R
p
r
p
=0
S(l
r
p
)
R
1
r
1
=0
S(l
r
1
) K 0.
It follows that we have an exact sequence of the form
0
R
p
r
p
=0
S(l
r
p
)
R
0
r
0
=0
S(l
r
0
) S(T) N 0.
The conclusion follows by applying the exact functor to this sequence.
(b) follows from (a) since the result is true for T = O
P
n
(C)
(l) with l 0.
(c) follows directly from (a) thanks to Theorem 5.5.5.
Remark 5.5.13. Parts (b) and (c) of the preceding proposition may be
viewed as stating that O
P
n
(C)
(1) is an ample locally free O
P
n
(C)
-module of
rank 1.
Proposition 5.5.14. Let M be a graded S-module which is in TT. Then,
there is a unique polynomial P
M
Q[z] such that
P
M
(l) = dimM
l
for l 0.
Proof. We know that for M in TT there is an exact sequence of the type
0
R
p
r
p
=0
S(l
r
p
)
R
0
r
0
=0
S(l
r
0
) M N 0
with N in T^. Moreover,
dimS(l
0
)
l
= dimS
l
0
+l
=
(l +l
0
+1) (l +l
0
+n)
n!
.
Hence, the conclusion follows from the additivity of M dimM
l
and the
fact that the zeros of a polynomial are in nite number.
Denition 5.5.15. The polynomial P
M
of Proposition 5.5.14 is called the
Hilbert-Samuel polynomial of M.
Corollary 5.5.16. For any coherent analytic sheaf T on P
n
(C) the map
l (P
n
(C); T(l))
is the Hilbert-Samuel polynomial P
S(F)
of the graded S-module S(T) (in
other words
(P
n
(C); T(l)) = dim(P
n
(C); T(l))
for l 0).
Proof. It follows from Theorem 5.5.5 that
(P
n
(C); O
P
n
(C)
(l
0
)(l)) =
(l +l
0
+1) (l +l
0
+n)
n!
.
Therefore,
(P
n
(C); T(l))
is a polynomial in l for T = O
P
n
(C)
(l
0
). Part (a) of Corollary 5.5.12 and
the additivity of (P
n
(C); ) allow us to extend the result to any coherent
analytic sheaf T on P
n
(C). Moreover, using part (c) of Corollary 5.5.12, we
see that
(P
n
(C); T(l)) = dim(P
n
(C); T(l))
if l 0 and the conclusion follows from the denition of S(T) and of its
Hilbert-Samuel polynomial.
Exercise 5.5.17. Let Z be a closed hypersurface of P
n
(C) corresponding
to the zeros of a homogeneous polynomial Q of degree d. Denote O
Z
the
coherent analytic sheaf on P
n
(C) associated to Z. Show that
P
S/(Q)
(l)
= (P
n
(C); O
Z
(l))
=
(l + 1) (l +n)
n!

(l d +1) (l d + n)
n!
5.6. Hirzebruch-Riemann-Roch theorem for P
n
(C) 207
and that in particular
(P
n
(C); O
Z
) =
_
1 if d n
1 (1)
n
_
d1
n
_
if d > n.
Solution. From the exact sequence
0 O
P
n
(C)
(d)
Q
O
P
n
(C)
O
Z
0
we deduce that
(P
n
(C); O
Z
) = (P
n
(C); O
P
n
(C)
) (P
n
(C); O
P
n
(C)
(d))
and the rst formula follows immediately. For the second one, we note that
(d + 1) (d + n)
n!
=
_
_
_
0 if d n
(1)
n
(d 1) (d n)
n!
if d > n.
Hence, the conclusion.
5.6 Hirzebruch-Riemann-Roch theorem for P
n
(C)
Proposition 5.6.1. Assume T is a coherent analytic sheaf on X = P
n
(C).
Then,
(X; T) =
_
X
ch(T) td(TX).
Proof. By Corollary 5.5.12, we know that T has a resolution of the form
0
R
p
r
p
=0
O
X
(l
r
p
)
R
0
r
0
=0
O
X
(l
r
0
) T 0.
Using the additivity of both sides of Hirzebruch-Riemann-Roch formula, it is
thus sucient to treat the case T = O
X
(l) (l Z). Thanks to Exercise 4.2.6
we know that c
(TX) = (1 + )
n+1
where H
2
(X; Z) is the rst Chern
class of U
n
(C). Hence, TX and U
n
(C)
n+1
have same total Chern class. It
follows that
td(TX) = td(U
n
(C)
n+1
) = (tdU
n
(C))
n+1
=
_

1 e
_
n+1
.
Recall that the sheaf of holomorphic sections of U
n
(C) is O
X
(1). Hence,
for l > 0, we have
ch(O
X
(l)) = ch(O
X
(1) O
X
(1))
= ch(O
X
(1))
l
= e
l
(*)
Using the fact that
ch(O
X
(l)) ch(O
X
(l)) = ch(O
X
) = 1
we see that formula (*) holds for any l Z. Combining the results, we get
that
ch(O
X
(l)) td(TX) =
_

1 e
_
n+1
e
l
.
Therefore,
(ch(O
X
(l)) td(TX))
2n
=
n
where is the coecient of z
n
in the Taylor expression at 0 of
_
z
1 e
z
_
n+1
e
lz
.
Using Cauchys formula, we have
= Res
_
e
lz
(1 e
z
)
n+1
; 0
_
=
1
2i
_
C
e
lz
(1 e
z
)
n+1
dz
where C is a path of C
such that Ind(, 0) = 1. Using the change of

variables
w = 1 e
z
we get
=
1
2i
_
C
(1 w)
l
w
n+1
dw
1 w
=
1
2i
_
C
dw
w
n+1
(1 w)
l+1
=
1
n!
_
1
(1 w)
l+1
_
(n)
w=0
=
(l +1) (l +n)
n!
=
_
l +n
n
_
.
By Exercise 4.2.6, we know that
_
X

n
= 1. Therefore,
_
X
ch(O
X
(l)) td(TX) =
_
l + n
n
_
and the conclusion follows from Theorem 5.5.5.
5.7 Riemann-Roch for holomorphic embeddings
Our aim in this section is to prove the following result.
5.7. Riemann-Roch for holomorphic embeddings 209
Theorem 5.7.1. Let i : X Y be a closed embedding of complex analytic
manifolds. Then, for any T

T
b
Coh
(O
X
), we have
ch
X
(i
!
T) = i
!
(chT
/ tdT
X
Y )
in H
X
(Y ; Q).
We will start by recalling a few facts about Koszul complexes which will
be needed in the proof.
Let be a commutative ring on the topological space X, let c be locally
free -module of rank r and let s be a section of c
. Recall that the Koszul

complex K
(c; s) is the complex

0
r
c
L
s

k
c
L
s
c

1
0
where L
s
is the interior product with s. As is well-known, the -linear
morphism L
s
is characterized by the formula
L
s
(
1

k
) =
k
l=1
(1)
l1
s,
l
)
1

l

k
.
As a graded -module, K
(c; s) is isomorphic to the exterior algebra
c
and hence has a canonical structure of anticommutative graded -algebra.
Note that the dierential L
s
is compatible with this structure in the sense
that
L
s
(
k

l
) = L
s
(
k
)
l
+(1)
k
k
L
s
(
l
)
for any sections
k

_
k
c,
l

_
l
c.
In particular, if
1
, . . . ,
p
are global sections of , we have
K
(
p
; (
1
, ,
p
)) K
(;
1
)
R

R
K
(;
p
).
From this formula, it follows easily that if
l
: /
1
+ +
l1
/
1
+ +
l1
is injective for l 1, , p then K
(
p
; (
1
, ,
p
)) is a projective reso-
lution of
/
1
+ +
p
.
This allows us to prove the following lemma.
Lemma 5.7.2. Let X be a complex analytic manifold and let c be a locally
free O
X
-module of rank p. Denote E the associated holomorphic complex
vector bundle. Assume s is a section of E which is transverse to the zero
section. Then,
Z
s
= x X : s(x) = 0
is a complex analytic submanifold of X. Moreover, if i : Z
s
X is the
inclusion map, then the Koszul complex
0
p
L
s
p1
L
s
O
x
0
is a resolution of the O
X
-module
i
!
O
Z
s
.
Proof. The problem being clearly local on X, we may assume X is a coor-
dinated neighborhood U of x
0
and c = O
p
X
. Then, E = C
p
U and we
have
s(x) = (f(x), x)
where f : U C
p
is a holomorphic map. The fact that s is transverse to the
zero section entails that f is a submersion. So, restricting U to a smaller
neighborhood of x
0
if necessary, we may nd a holomorphic coordinate
system (z
1
, , z
p
) on U with f
1
= z
1
, , f
p
= z
p
. For such a coordinate
system, the Koszul complex
K
(c, s) K
(O
p
X
, (z
1
, , z
p
))
and the conclusion follows easily.
With this lemma at hand, we can now prove the following special case
of Theorem 5.7.1.
Proposition 5.7.3. Let p : E X be a holomorphic complex vector
bundle of rank r. Denote i : X E its zero section. Then, for any
T

T
b
Coh
(O
X
), we have
ch
X
(i
!
T
) = i
!
(chT
/ tdE)
in H
X
(E; Q).
Proof. First, note that since p i = id
X
, we have
ch
X
(i
!
T
) = ch
X
(i
!
i
)
= ch
X
(i
!
O
X

L
O
E
p
)
= ch
X
(i
!
O
X
)p
chT

and the result will be true for any T

T
b
Coh
(O
X
) if it is true for O
X
.
To prove the result in this case, let us consider the relative projective
compactication p : E X of E. Recall that
E = P(E (C X))
and that we have an open embedding
j : E E
and a complementary closed embedding
k : P(E) E
which we can use to identify E and P(E) with subspaces of E. Set i = j i.
Let us show that
(a) the canonical morphism
: H
X
(E; Q) H
(E; Q)
is injective;
(b)
ch(i
!
O
X
) = i
!
(1/ tdE)
in H
(E; Q).
The result will follow since
(ch
X
(i
!
O
X
)) = ch(i
!
O
X
)
and
j
ch
X
(i
!
O
X
) = ch
X
(i
!
O
X
).
(a) Since
H
X
(E; Q) H
X
(E : Q) H
pproper
(E; Q)
the injectivity of will follow from the surjectivity of
H
(E; Q) H
(P(E); Q).
Denote U (resp. U) the tautological line bundle on E (resp. P(E)). We
know that H
(E; Q) is a free H
(X; Q)-module with basis

1, c
1
(U), , c
1
(U)
r
and that H
(P(E); Q) is a free H
(X; Q)-module with basis

1, c
1
(U), , c
1
(U)
r1
.
Since U
|P(E)
U, the conclusion follows.
(b) Since
U p
1
(E (C X))
we get a morphism
U p
1
E
and hence a section t of
F = Hom(U, p
1
E) U
p
1
E.
One checks easily that t is transverse to the zero section of F and that
Z
t
= X. Therefore, Exercise 2.7.8 shows that
c
r
(F) =
E/X
= i
!
1.
Moreover, Lemma 5.7.2 shows that the Koszul complex
K
(F, t)
is a resolution of i
!
O
X
by locally nite free O
E
-modules. It follows that
ch(i
!
O
X
) = chK
(F, t) = ch
.
Using Proposition 4.4.9, we get
ch(i
!
O
X
) = c
r
(F)/ tdF.
Note that
c
r
(F)c
1
(U
) = c
r+1
(F U
) = 0
since
F U
Hom(U, p
1
(E (C X)))
has a nowhere vanishing section. But
td(F)
1
= td(U
)
1
p
td(E)
1
p
td(E)
1
(mod c
1
(U
)).
Therefore,
c
r
(F)/ tdF = c
r
(F)/p
td E
= (i
!
1)/p
tdE
= i
!
(1/ tdE).
Proof of Theorem 5.7.1.
The idea is to use the deformation of X to the normal bundle of Y in
order to reduce the general case to the case treated in Proposition 5.7.3.
Recall that with the complex deformation

X of X to the normal bundle
p : T
Y
X Y is given a closed embedding

i : Y C

X, a submersion
:

X C and a projection p :

X X such that
(a) the maps
p
i : Y C Y,
i : Y C C
coincide with the canonical projections;
(b) there is a commutative diagram of the form
Y 1

i

1
(1)
Y
O
i
1
X
O
j
1
(c) there is a commutative diagram of the form

Y 0

i

1
(0)
Y
O
i
0
T
Y
X
O j
0
with p j
0
= i
0
p.
Since both and
i are submersions, it follows that j

0
and j
1
are transverse
to

i. Hence,
Lj
i
!
(T O
C
) = i
0
!
(T)
and
Lj
i
!
(T O
C
) = i
1!
(T).
Setting ( =
i
!
(T O
C
), it follows that
ch
Y
(i
0!
T) = j
0
ch
Y C
(()
and that
ch
Y
(i
1
!
T) = j
1
ch
Y C
(().
Hence
j
0
!
ch
Y
(i
0!
T) = j
0
!
(1) ch
Y C
(()
in H
Y{0}
(

X; Q) and
j
1
!
ch
Y
(i
1!
T) = j
1
!
(1) ch
Y C
(()
in H
Y{1}
(

X; Q). We have
j
0
!
(1) =
X/
1
(0)
=
C/{0}
in H

1
(0)
(

X; Z) and a similar formula for j
1
!
(1). Since
C/{0}
=
C/{1}
in H
2
c
(C; Z), they are equal in H
2
D
(C; Z) for D a sucient large compact
disk. It follows that
j
0
!
(1) = j
1
!
(1)
in H

1
(D)
(

X; Z) and we get
j
0
ch
Y
(i
0
!
T) =

j
1
ch
Y
(i
1
!
T)
in H
YD
(

X; Q). Since Y D is a p-proper closed subset of

X,
p
!
: H
Y D
(

X; Q) H
Y
(X; Q)
is well-dened and we get
ch
Y
(i
1!
T) = p
!
j
1
!
ch
Y
(i
0!
T)
= p
!
j
0
!
ch
Y
(i
0!
T).
Using Proposition 5.7.3, we get nally
ch
Y
(i
1!
T) = p
!
j
0
!
i
0!
(chT/ tdT
Y
X)
= i
1!
ch(T/ tdT
Y
X)
as requested.
5.8 Proof of Hirzebruch-Riemann-Roch theorem
Using the results in the three previous sections, we can now prove Theo-
rem 5.1.10.
Proposition 5.8.1. Assume i : X Y is a closed embedding of com-
pact complex analytic manifolds. Then Hirzebruch-Riemann-Roch theorem
holds for X if it holds for Y .
Proof. Let T be a coherent analytic sheaf on X. By Theorem 5.7.1, we
have
ch
X
(i
!
T) = i
!
(chT/ tdT
X
Y ).
5.8. Proof of Hirzebruch-Riemann-Roch theorem 215
Therefore,
ch(i
!
T) tdTY = i
!
(chT i
tdTY/ tdT
X
Y ).
Thanks to the exact sequence
0 TX i
1
TY T
X
Y 0
we have
i
tdTY = tdTX tdT

X
Y.
So
ch
Y
(i
!
T) tdTY = i
!
(chT tdTX)
Corollary 5.8.2. The Hirzebruch-Riemann-Roch theorem holds for any
complex projective manifold.
Proof. This follows directly from the preceding proposition if one takes
Y = P
n
(C) and uses Proposition 5.6.1.
Bibliography
[1] P. Alexandro and H. Hopf, Topologie I, Grundlehren der mathema-
tischen Wissenschaften, no. 45, Springer, Berlin, 1935.
[2] M. F. Atiyah and F. Hirzebruch, Riemann-Roch theorems for dieren-
tiable manifolds, Bull. Amer. Math. Soc. 65 (1959), 276281.
[3] M. F. Atiyah and F. Hirzebruch, Analytic cycles on complex manifolds,
Topology 1 (1962), 2545.
[4] M. F. Atiyah and F. Hirzebruch, The Riemann-Roch theorem for ana-
lytic embeddings, Topology 1 (1962), 151166.
[5] A. Borel and J.-P. Serre, Le theor`eme de Riemann-Roch, Bull. Soc.
Math. France 86 (1958), 97136.
[6] G. E. Bredon, Sheaf theory, McGraw-Hill Series in Higher Mathematics,
McGraw-Hill, New York, 1967.
[7] J. Dieudonne, Cours de geometrie algebrique I. Apercu historique sur le
developpement de la geometrie algebrique., Le Mathematicien, no. 10,
Presses Universitaires de France, Paris, 1974.
[8] , A history of algebraic and dierential topology. 19001960,
Birkhauser, Boston, 1989.
[9] A. Dold, Lectures on algebraic topology, Grundlehren der mathematis-
chen Wissenschaften, no. 200, Springer, Berlin, 1972.
[10] W.Fulton, Intersection theory, Ergebnisse der Mathematik und ihrer
Grenzgebiete, 3. Folge, no. 2, Springer, Berlin, 1984.
217
218 Bibliography
[11] R. Godement, Topologie algebrique et theorie des faisceaux, Hermann,
Paris, 1964.
[12] H. Grauert and R. Remmert, Theory of Stein spaces, Grundlehren der
mathematischen Wissenschaften, no. 236, Springer, Berlin, 1979.
[13] H. Grauert and R. Remmert, Coherent analytic sheaves, Grundlehren
der mathematischen Wissenschaften, no. 265, Springer, Berlin, 1984.
[14] M. Greenberg, Lectures on algebraic topology, Mathematics Lecture
Notes, Benjamin, New York, 1967.
[15] F. Hirzebruch, Topological methods in algebraic geometry, 3 ed.,
Grundlehren der mathematischen Wissenschaften, no. 131, Springer,
Berlin, 1966.
[16] D. Husemoller, Fibre bundles, 2 ed., Graduate Texts in Mathematics,
no. 20, Springer, New York, 1975.
[17] B. Iversen, Local Chern classes, Ann. scient.

Ec. Norm. Sup. 4
e
serie 9
(1976), 155169.
[18] M. Kashiwara and P. Schapira, Sheaves on manifolds, Grundlehren der
mathematischen Wissenschaften, no. 292, Springer, Berlin, 1990.
[19] S. Lefschetz, Algebraic topology, A. M. S. Colloqium Publications, no.
XXVII, American Mathematical Society, New York, 1942.
[20] J. W. Milnor and J. D. Stashe, Characteritic classes, Annals of Math-
ematics Studies, no. 76, Princeton University Press, Princeton, New
Jersey, 1974.
[21] R. S. Palais and al, Seminar on the Atiyah-Singer index theorem,
Annals of Mathematics Studies, no. 57, Princeton University Press,
Princeton, 1965.
[22] P. Schapira and J.-P. Schneiders, Elliptic pairs I: Relative niteness
and duality, Asterisque 224 (1994), 560.
[23] , Elliptic pairs II: Euler class and relative index theorem,
Asterisque 224 (1994), 6198.
[24] J.-P. Schneiders, Relative paracompactness as tautness condition in
sheaf theory, Bull. Soc. Roy. Sci. Li`ege 53 (1984), 179186.
[25] H. Seifert and W. Threlfall, Lehrbuch der Topologie, B. G. Teubner,
Leipzig, 1934.
Bibliography 219
[26] J.-P. Serre, Faisceaux algebriques coherents, Annals of Mathematics 61
(1955), 197278.
[27] , Geometrie algebrique et geometrie analytique, Ann. Inst.
Fourier (Grenoble) 6 (1956), 142.
[28] Seminaire Heidelberg-Strasbourg 1966/67. Dualite de Poincare, Publi-
cations I.R.M.A. Strasbourg, no. 3, 1969.
[29] E. H. Spanier, Algebraic topology, McGaw-Hill Series in Higher Math-
ematics, McGraw-Hill, New York, 1966.
[30] N. Steenrod, The topology of ber bundles, Princeton Mathematical
Series, no. 14, Princeton University Press, Princeton, 1951.
[31] O. Veblen, Analysis situs, 2 ed., A. M. S. Colloqium Publications, no.
V/II, American Mathematical Society, New York, 1931.
Montagem e impressao em :
SILVAS, crl R. D. Pedro V, 126
Telef. 342 31 21 1250-094 LISBOA
Deposito Legal # 112 106/2000
TEXTOS DE MATEM
ATICA
Volumes ja publicados
1 Lus Sanchez,
Metodos da Teoria de Pontos Crticos, 1993.
2 Miguel Ramos,
Teoremas de Enlace na Teoria dos Pontos Criticos, 1993.
3 Orlando Neto,
Aequac ones Diferenciais em Superfcies de Riemann, 1994.
4 A. Bivar Weinholtz,
Integral de Riemann e de Lebesgue em R
N
, (3
a
edicao), 1999.
5 Mario S. R. Figueira,
Fundamentos de Analise Innitesimal, (2
a
edicao), 1997.
6 Owen J. Brison,
Teoria de Galois, (3
a
edicao), 1999.
Aequac ones Diferenciais - Uma Introduc ao, (2
a
edicao), 2000.
8 Armendo Machado,
Topicos de Analise e Topologia em Variedades, 1997.
9 Armendo Machado,
Geometria Diferencial - Uma Introduc ao Fundamental, 1997.
Teoria dos Operadores, 1998.
11 Teresa Monteiro Fernandes,
Topologia Algebrica e Teoria elementar dos Feixes, 1998.
12 Owen J. Brison,
Grupos e Representac oes, 1999.
13 Jean-Pierre Schneiders,
Introduction to Characteristic Classes and Index Theory, 2000
Abstract
This book is based on a course given by the author at the university of
Lisbon during the academic year 19971998. Its aim is to give the reader
an idea of how the theory of characteristic classes can be applied to solve
index problems. Starting from the Lefschetz xed point theorem and its
application to the computation of the Euler-Poincare characteristic of a
compact orientable manifold, we rst develop the theory of Euler classes
of orientable manifolds and real vector bundles. Then, we study Stiefel-
Whitney classes and the general modulo 2 characteristic classes of real vector
bundles. Similar considerations for complex vector bundles lead us to the
Chern classes. We conclude the part devoted to characteristic classes by a
study of global and local Chern characters. The rest of the book is then
centered around the Riemann-Roch theorem. We present rst a very simple
proof which works for compact complex curves and allows us to make links
with the original results of Riemann and Roch. Then, we treat in details the
case of compact complex projective manifolds by more advanced methods.
Faculdade de Ciencias Departamento de Matematica
Rua Ernesto de Vasconcelos Bloco C1, 3
Piso
1749-016 LISBOA PORTUGAL

Introduction To Characteristic Classes and Index Theory Jean-Pierre Schneiders

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Introduction To Characteristic Classes and Index Theory Jean-Pierre Schneiders

Transféré par

Droits d'auteur :

Formats disponibles

TEXTOS DE MATEM

are two open neighborhoods of x in X and s F(U), s

of the complex logarithm that

are two open neighborhoods of A in X and s (U; T), s

(X; T) = 0 for any k > 0.

is a right resolution of T by -acyclic sheaves. The aim of the next

of ( on X. For this section, we have

F and belongs to . These considerations show that

In particular, if T is an abelian sheaf, we have the Mayer-Vietoris long exact

From the associated long exact sequence, we deduce that

is a complex of (Y ; )-acyclic sheaves and a commutative diagram

where is a quasi-isomorphism and 1

is a complex of (X; )-acyclic

. This shows that X H

by zero outside of U. Clearly,

is the inverse of the isomorphism (p

where the rst vertical arrow is induced by zero extension of sections. It

For n = 1, we get the exact sequence

denotes the Jacobian of .

its Jacobian. Show that for any u U, the map

. The dimension of the nite

() canonically associated with . We

induces a canonical morphism

[) is open (resp. closed) in [[ [

[. Thanks to the morphism

One sees that we may assume = . Moreover, we may assume that

Therefore, the diagram

[; Z) by the canonical morphism

together with the fact that [

(X; Z) of X with coecient

which have a compact neighborhood L

for which the image

. Consider the diagram

(Im) is also nitely generated. Hence, so is Im/(Im

). Using the epimorphism

) is nitely generated. Since the induction

is nitely generated, it follows that Im(

is a cartesian square of locally compact spaces. Then, we have the canonical

are open balls of R

of abelian groups we have an exact sequence of the form

Taking cohomology and using the fact that H

Proposition 1.12.5. Let f : X Y be a continuous map and assume

at the level of cohomology. For the link with singular

is a bounded complex with nitely generated cohomology.

are morphisms of locally compact

commutes or anticommutes. Using Exercise 1.8.6, we know that if

is a path from y to z then m

induces a representation of the Poincare

is commutative and the conclusion follows.

= id. Hence, deg(f) = 1 and

coincides with that of T

B is the vector bundle f

characterized by the fact that

, u) and assume det S(u) > 0 (resp.

Taking cohomology, we get the long exact sequence

and the tautological real line bundle on P

the relative A-orientation sheaf of Y in X. A relative A-orientation of Y

(X X) its usual quotient orientation, Remark 2.7.5 and

(X; A) for the H

where the action of R

First, note that it follows from Thom isomorphism that

and is called the dual of E. In particular, the local frame of E