Lectures On Sheaf Theory: C.H. Dowker

Lectures on Sheaf Theory
by
C.H. Dowker
Tata Institute of Fundamental Research

Bombay
1957
Lectures on Sheaf Theory

by
C.H. Dowker
Notes by
S.V. Adavi and N. Ramabhadran
Tata Institute of Fundamental Research

Bombay
1956
Contents
1 Lecture 1
2 Lecture 2
3 Lecture 3
4 Lecture 4
15
5 Lecture 5
21
6 Lecture 6
27
7 Lecture 7
31
8 Lecture 8
35
9 Lecture 9
41
10 Lecture 10
47
11 Lecture 11
55
12 Lecture 12
59
13 Lecture 13
65
14 Lecture 14
73
iii
iv
Contents
15 Lecture 15
81
16 Lecture 16
87
17 Lecture 17
93
18 Lecture 18
101
19 Lecture 19
107
20 Lecture 20
113
21 Lecture 21
123
22 Lecture 22
129
23 Lecture 23
135
24 Lecture 24
139
25 Lecture 25
143
26 Lecture 26
147
27 Lecture 27
155
28 Lecture 28
161
29 Lecture 29
167
30 Lecture 30
171
31 Lecture 31
177
32 Lecture 32
183
33 Lecture 33
189
Lecture 1
Sheaves.
1
onto
Definition. A sheaf S = (S , , X) of abelian groups is a map : S

X, where S and X are topological spaces, such that
1. is a local homeomorphism,
2. for each x X, 1 (x) is an abelian group,
3. addition is continuous.
That is a local homeomorphism means that for each point p S ,
there is an open set G with p G such that |G maps G homeomorphically onto some open set (G).
Sheaves were originally introduced by Leray in Comptes Rendus
222(1946)p. 1366 and the modified definition of sheaves now used was
given by Lazard, and appeared first in the Cartan Sem. 1950-51 Expose
14.
In the definition of a sheaf, X is not assumed to satisfy any separation
axioms.
S is called the sheaf space, the projection map, and X the base
space.
The open sets of S which project homeomorphically onto open sets
of X form a base for the open sets of S .
Proof. If p is in an open set H, there exists an open G, p G such that
|G maps G homeomorphically onto an open set (G). Then H G is
open, p H G H, and |H G maps H G homeomorphically onto
(H G) open in (G), hence open in X.

1
Lecture 1
2
2
is a continuous open mapping

Proof. Continuity of follows from the fact that it is a local homeomorphism, and openness follows from the result proved above.

The set S x = 1 (x) is called the stalk of S at x. S x is an abelian
group. If (p) , (q), p + q is not defined.
S x has the discrete topology.
Proof. This is a consequence of the fact that is a local homeomorphism.

Let S S be the cartesian product of the space S with itself and let
S + S be the subspace consisting of those pairs (p, q) for which (p) =
(q). Addition is continuous means that f : S + S S defined by
f (p, q) = p+q is continuous. In other words, if p, q S and (p) = (q),
then given an open G containing p + q, there exist open sets H, K with
p H, q K such that if r H, s K and (r) = (s), then r + s G.
We may write this as H + K G.
Proposition 1. Zero and inverse are continuous.
(i) Writing O x for the zero element of the group S x , zero is continuous
means that f : X S , where f (x) = O x , is continuous.
Proof. Let x X and let G be an open set containing f (x) = O x .

Then there is an open set G1 such that O x G1 G and |G1 is a
homeomorphism of G1 onto open (G1 ). Since O x + O x = O x , and
addition is continuous, there exist open sets H, K with O x H,
O x K such that H + K G1 . Let L = G1 H K, then L is open,
O x L and |L is a homeomorphism of L onto open (L). Then
x = (O x ) (L) and if y (L) there exists q L with (q) = y.
Then q H, q K and hence q + q G1 . But q G1 , and |G1 is
11; hence q = S y G1 . Therefore, since q+q S y G1 , q+q = q,
hence q = Oy . Thus if y (L), f (y) = Oy L. Thus x is in open
(L), with f ((L)) = L G. Hence f is continuous. (Incidentally
we have proved that each O x is contained in an open set which
Lecture 1
consists of zeros only and which projects homeomorphically onto

an open set of X.)

(ii) Writing-p for the inverse of p in the group S (p) inverse is continuous means that g : S S where g(p) = p is continuous.
Proof. Let p S , and let G be an open set containing-p. Then
there exists open L containing O(p) and consisting of zeros only.
Since p+ (p) = O(p) and addition is continuous, there exist open
H, K, with p H, p K and H + K L. Hence if q H, r K,
(r) = (q), then q + r = O(q) , i.e. r = q. We may assume that
|H is a homeomorphism. Let
H1 = (|H)1 ((H) (K G)),
then p H1 , and since is open, continuous, H1 is open. Then if
q H1 , there exists r KG with (r) = (q); then r = q = g(q). 4
Thus H1 is open and g(H1 ) G. Hence g is continuous.

Corollary 1. Subtraction is continuous.
i.i. f : S + S S , where f (p, q) = p q is continuous.
Corollary 2. The set of all zeros is an open set.
Example 1. If X is a topological space, and G is an abelian group furnished with the discrete topology, let S = (X G, , X) where (x, g) =
x and (x, g1 ) + (x, g2 ) = (x, g1 + g2 ). Each stalk S x is isomorphic to G.
Axioms a) and c) are easily verified. This sheaf is called the constant
sheaf associated with G.
Example 2. Form the constant sheaf (A Z, , A) where A is the square
{(x, y) : 0 x 1, 0 0 y 1}, and Z is the group of integers.
Then identify (x, 0) with (1 x, 1) in A to get a Mobius band X, and
identify (x, 0, n) with (1 x, 1, n) in A Z to get S . The resulting sheaf
S = (S , , X) is the sheaf of twisted integers over the Mobius band.
Each S x is isomorphic to the group of integers.
Lecture 1
Example 3. Let X be the sphere of complex numbers. Let S x be the

additive group of function elements at x, each function element being
a power series converging in some neighbourhood of x. Let S = x S x
and define : S X by (S x ) = x. If p is a function element, a
neighbourhood of p in S is defined by analytic continuation. Then S =
(S , , X) is the sheaf of analytic function elements. Each component
(maximal connected subset) of S is a Riemann surface without branch
points. The sheaf space S is Hansdorff.
Lecture 2
Example 4. Let S : {(x, y) : x2 + y2 = 1, x < 1} together with the group 6
of integers Z}. The topology on the former set is the usual induced
topology, and the neighbourhoods for an integer n Z are given by
Ga (n) = {n, (x, y) : x2 + y2 = 1, a < x < 1}; X : {(x, y) : x2 + y2 = 1} and
let : S X be defined by (n) = (1, 0), (x, y) = (x, y). 1 (1, 0) is the
group Z, for other pints (x, y) X, 1 (x, y) = (x, y) is regarded as the
zero group. It is easily verified that S = (s, , X) is a sheaf. Here S is
locally euclidean, and has a countable base. The set of all zeros is open
(Corollary 2) and compact but not closed; its closure is not compact, S
is T 1 but not Hausdorff though X is Hausdorff.
Exercise. If X is a T o or a T 1 space, space, so is S
So far we have only defined sheaves of abelian groups. It is now
quite clear how we can extend the definition to the case where the stalks
are any algebraic systems.
A sheaf of rings is a local homeomorphism : S X such that
each 1 (x) is a ring and addition and multiplication are continuous, i.e
f : S + S S where f (p, q) = p + q,
g : S + S S where g(p, q) = p q,
are continuous.
The sheaf of function elements (Example 3) where multiplication 7
of two function elements in the same stalk is defined to be the usual
multiplication of power series is a sheaf of rings.
In the sheaf of twisted integers (Example 2) each S x is isomorphic
to the ring Z, but this sheaf is not a sheaf of rings.
5
Lecture 2
A sheaf of rings with unit is a local homeomorphism : S X

such that each 1 (x) is a ring with unit element 1x and addition, multiplication and unit are continuous; i,e.,
f : S + S S , f (p, q) = p + q,
g : S + S S , f (p, q) = p q,
h : X S ,h(x) = 1x are continuous.
Example 5. Let A be the ring with elements 0, 1, b, c; where the rules
of addition and multiplication are given by
1 + 1 = b + b = c + c = 0;
1 + b = c, 1 + c = b, b + c = 1;
b2 = b, c2 = c, bc = cb = 0
[The ring A may be identified with the ring of functions defined on
a set of two elements with values in the field Z2 .]
Let X : {x, k x 1}, S : subspace of X A obtained by omitting
the points (k, 1), (k, b); and let : S X, be defined by (x, a) = x.
Addition and multiplication in a stalk are defined by
(x, a1 ) + (x, a2 ) = (x, a1 + a2 ),
(x, a1 ) (x, a2 ) = (x, a1 a2 ).
8
Then S = (S , , X) is a sheaf of rings, each S x is a ring with unit, but

S is not sheaf of rings with unit.
A sheaf of (unitary left) a-modules, where a = (A, , X) is a sheaf of
rings with unit, is a local homeomorphism : S X such that each
1 (x) is a unitary left A x -module and addition, and multiplication by
elements of A x (for each x) are continuous; i.e.,
f : S + S S,
f (p, q) = p + q,
g : A + S S,
g(a, p) = ap
are continuous.
A + S is the subspace of A S consisting of all pairs (a, p) for which
(a) = (p).
Lecture 2
[If R is a ring with unit element, we say that M is a unitary, left

R-module if it is a left R-module, and if 1 m = m for each m M.]
Example 6. Let a denote the sheaf of function elements on the complex
sphere X. Let S x consist of all (p, q) of function elements at x. Let S =
S
S x . A neighbourhood of (p, q) is defined by analytic continuation of
xX
p and q. In each S x addition is defined as (p1 , q1 ) + (p2 , q2 ) = (p1 +

q2 , q1 + q2 ); and if a is a function element at x, define multiplication
as a(p, q) = (a.p, a.q). Define : S X, by (p, q) = x if p, q are
function elements at x. Then (S , , X) is a sheaf of a-modules.
Any sheaf a of rings with unit can be regarded as a sheaf of a- 9
modules; the product ap for a A x , p A x being defined as the product
ap in A x .
A sheaf of B-modules where B is a ring with unit element is a local
homeomorphism : S X such that 1 (x) is a unitary left B-module
and addition, and multiplication by elements of B are continuous; i.e.,
f : S + S S,
f (p, q) = p + q,
gb : S S , gb (p) = b.p for each b B

are continuous.
Let S = (S , , X) be a sheaf of B-modules. This is equivalent to
saying that S is a sheaf of B-modules, where B is the constant sheaf
(X B, , X).
Proof:
B is a ring with unit element.
B(X B, , X) is a constant sheaf.
S = (S , , X) is a sheaf of abelian groups.
1) S is a sheaf of B-modules. 1) S is a sheaf of B-modules.
This means that b p is deThis means that ((p), b) p is
fined such that S x is a unidefined such that S x is a unitary
left B-module.
left x B module.
2) Addition is continuous.
2) Addition is continuous.
3) Multiplication is continuous 3) Multiplication is continuous
means that h : B S S
means that g : (X B) + S S
h(b, p) = b p is continuous.
g((p), b, p) = b.p is continuous.
Lecture 2
8
10
To prove the assertion, it is enough to show that the continuity of

h is equivalent to the continuity of g. To do this, define : B S
(X B) + S as (b, p) = ((p), b, p), then g = h. is clearly 1-1. We
show that is a homeomorphism and the result will follow from this. A
base for (X B) + S is formed by the sets (U b) + G where G projects
homeomorphically onto (G).
(U b) + G = (V b) + G where V = U (G)
= (V b) + H where H = (|G)1 V
= ((H) b) + H
= (b H).
Since the sets b H form a base for B S , it follows that is a homeomorphism.
Thus we may identify sheaves of (X B, , X)-modules with sheaves
if B-modules. By abuse of language, we write B for the ring as well as
for the constant sheaf (X B, , X).
11
Example 7. Let C be the ring of complex numbers, S = (S , , X) be

the sheaf of function elements on the complex sphere X and for c C,
p S define c p to be the usual product of a complex number with a
power series. Then S becomes a sheaf of C- modules.
Example 8. Let C be the ring of complex numbers, S = (S , , X) be
the sheaf of function elements on the complex sphere X. For c C and
(p, q) S define c(p, q) = (c p, c q).
Then S becomes a sheaf of C-modules.
Example 9. Let S = (S , , X) be any sheaf of abelian groups and let Z
be the ring of integers. For n Z and p S define n p = p + + p (n
times) if n > 0, n p = (n)p if n < 0, and 0 p = 0(p) . Thus S may
be considered as a sheaf of Z-modules.
Thus sheaves of rings with unit, sheaves of B-modules and sheaves
of abelian groups can be considered as special cases of sheaves of amodules.
Lecture 3
Sections.
Definition . A section of a sheaf (S , , X) over an open set U X is a 12
map f : U S such that f = 1|U where 1|U denotes the identity
function on U. (A map is a continuous function).
By abuse of language, the image f (U) is also called a section.
For each open set U X the function f : U S , where f (x) = 0x
is a section.
Proof. Zero is continuous, and (0x ) = x.
This section will be called the 0-section (zero section).

Proposition 2. If S = (S , , X) is a sheaf of abelian groups, the set
of all sections of S over a non-empty open set U forms an abelian
group (U, S ). If S is a sheaf of sings with unit, (U, S ) is ring with
unit element. If S is a sheaf of a-modules, (U, S ) is a unitary left
(U, a)-module. If S is a sheaf of B-modules, (U, S ) is a unitary left
B-module.
Note. The operations are the usual ones for functions. If f, g (U, S ),
a (U, a), b B, define
( f + g)(x) = f (x) + g(x), ( f g)(x) = f (x) g(x),
(a f )(x) = a(x) f (x), (b f )(x) = b f (x).
9
Lecture 3
10
The zero element of (U, S ) is the 0-section over U. If S is a sheaf of

rings with unit, the unit element of (U, S ) is the unit section 1 : U
S where 1(x) = 1x .
13
Proof. The proposition follows from the fact that addition zero, inverse,
unit and multiplication (ring multiplication as well as scalar multiplication) are continuous.

Remark. If a = (X B, , X) is a constant sheaf of rings with units, for
each open set U X we can identify B with the ring of constant sections
fb , b B where fb (x) = (x, b) over U. Then B (U, a) is a subring,
and by restricting the ring of scalars, each (U, a)-module becomes a
B-module. (B need not be the whole of (U, a)).
Abuse of . If S is a sheaf of a-modules, we agree that the unique
section over the empty set is the 0-section, and the set (S ) = 0.
Example 6. If S = (S , , X) is the sheaf of function elements over the
complex sphere X, (X, S ) can be identified with the ring of functions,
analytic in U. Then (X, S ) is the ring of functions, analytic everywhere, hence is isomorphic to the ring of complex numbers C.
Note . Usually a sheaf S = (S , , X) may be interpreted as describing
some local property of the space X; then (X, S ) gives the corresponding global property.
A section f : U S is an open mapping
Proof. This is proved using the fact that f is continuous and that is a
local homeomorphism.
14
We can now characterise the sections of S .

The necessary and sufficient condition that a set G S is a section f (U) over some open set U X is that G is open and |G is a
homeomorphism.
Proof. The sufficiency is easy to prove. To prove the necessity let f be a
section over U, then since f is open f (U) is open. Since f : U f (U)
is 1-1, open, continuous, | f (U) : f (U) U is a homeomorphism
Lecture 3
11
We have shown that if f is a section over U then f is a homeomorphism of U onto f (U).

The sections f (U) form a base for the open sets of S .
Proof. We have already proved that the open sets of S which project
homeomorphically onto open sets of X form a base for the open sets of
S

The intersection f (U) g(V) of two sections is a section.
Proof. f (U)g(V) is open and projects homeomorphically onto an open
set of X since each of f (U) and g(V) has this property.

If f : U S is a section, the set {x : f (x) = 0x } is open in U.
Proof. {x : f (x) = 0x } = ( f (U) 0(U)) (0 denotes the 0-section over
U), hence is open in U.

Definition. If f : U S is a section, the support of f , denoted as supp
f , is the set {x : f (x) , 0x }. This set is closed in U. If f is a section over 15
X, supp f is a closedsubset of X
Note . Since the sections of S form a base for the open sets of S , the
topology of S = S x can be described by specifying the sections. (See
appendix at the end of the lecture).
Let a = (A, , X) be a sheaf of rings with unit, and let S = (S , , X)
and R = (R, , X) be sheaves of a-modules (all over the same base space
X).
Definition. A homomorphism h : S R is a map h : S R such that
h = and its restriction h|S x = hx : S x R x is an A x homomorphism
for each x X
This definition includes as a special case the definition of homomorphisms of sheaves of B-modules and sheaves of abelian groups.
If h : S R is a homomorphism, the image of each section is a
section.
12
Lecture 3
Proof. If f : U S is a section, then h f : U R, the image of

the section f defined by (h f )(x) = h( f (x)), is continuous and (h f ) =
f = 1|U.
(1) h is an open mapping.
(2) h is a local homeomorphism.
16
Proof. (1) If G S is open, then G is a union of sections, hence h(G) is

a union of sections of R, hence is open
(2) Each p S is contained in some section f (U). So h f (U) is a
section in R and h| f (U) = h f | f (U), and each of | f (U) : f (U) U
and h f : U h f (U) is a homeomorphism.

Appendix.
A sheaf may be described by specifying its sections as follows: Suppose that we are given a space X and mutually disjoint abelian groups
S x , one for each point x X. Also suppose that we are given a famP
ily
= {s} of functions with domain open in X and values in x S x ,
s : dom(s) x S x , such that, if x dom(s), s(x) S x . Suppose
further that
P
(i) the images for all s cover S x ,
(ii) if s1 (x), s2 (x) are defined then, for some open U with x
P
dom(s1 ) dom(s2 ), (s1 + s2 )|U ,
(iii) if s(x) = 0x then, for some open U with x U dom(s)s(U)

consists entirely of zeros.
P
If S = S x with {s(U)}, for all s and open U dom(s), as base
for open sets and if (p) = x for p S x then (S , , X) is a sheaf.
Proof. Let p s(U) s1 (U1 ) and let x = (p). By (i) there exists
s2 with s2 (x) = p, by (ii) there exists a neighbourhood V of x with
P
(s + s2 )|V
and by (iii), since (s + s2 )x = p p = 0x there is a
smaller neighbourhood V of x with (s + s2 )(V ) consisting of zeros.
Lecture 3
13
Similarly there is a neighbourhood V1 of x with (s1 + s2 )(V1 ) consisting

of zeros.

Let W = V V1 , then (s + s1 + s2 )|W = s|W = s1 |W. Then s(W) is in 17
the proposed base and p s(W) s(U) s1 (U1 ). Therefore the axioms
for a base are satisfied.
Then s : U S is continuous. For if x dom(s) and p = s(x),
any neighbourhood G of p contains a neighbourhood s1 (U1 ) and again
there is an open W with p s(W) = s1 (W) s(U1 ) G. Hence
s : U s(U) is a homeomorphism, since s is clearly 1-1 and open.
Then |s(U) is the inverse homeomorphism and its image U is open.
Thus is a local homeomorphism.
Addition is continuous. For if p, q S x , with p+q s2 (U2 ) suppose
p s(U) and q s1 (U1 ). By (ii) there is a neighbourhood V of x with
P
(s1 + s2 )|V . Then p+q s2 (U2 )(s+ s1 )(V) and hence for some W,
with x W U2 V, s2 |W = (s + s1 )|W. Thus, if r s(W), t s1 (W)
and (r) = (t), then s(r) + s1 (t) = s2 (t) s2 (U2 ).
Lecture 4
If (S , , X) and (S , , X) are two sheaves of a-modules with S S and 18
if the inclusion map i : S S is a homomorphism, then S is an open
subset of S since i is an open map; further the topology on S is the one
induced from S . Then = i = |S and since i|S x : S x S x is a
homomorphism, S x is a sub - A x - module of S x .
This suggests the following definition of a subsheaf.
Definition. (S , |S , X) is called a subsheaf of (S , , X) if S is open in
S and, for each x, S x = S S x is a sub - A x - module of S x .
A subsheaf is a sheaf and the inclusion map i is a homomorphism.
Proof. For each p S there exists an open set G, p G S , with
|G a homeomorphism. Then G S is open in S and (|S )|G S =
|G S is a homeomorphism. S x is an A x -module and the operations
which are continuous in S are continuous in the subspace S . Therefore
(S , |S , X) is a sheaf.

Since i : S S is a map, and i = |S and i|S x : S x S x is
the inclusion homomorphism of the submodule S x , it follows that i is a
homomorphism.
The set of all zeros in S is a subsheaf of S .
Proof. The set of zeros is open in S , and 0x is a sub-A x -module of S x .

This sheaf is called the 0-sheaf (zero sheaf) and is usually identified 19
with the constant sheaf 0 = (X 0, , X).
15
Lecture 4
16
If h : S R is a homomorphism of sheaves, the set S of p S

such that h(p) = 0(p) forms a subsheaf S of S called the kernel of
h(S = ker h), and the image set S = h(S ) forms a subsheaf of R
called the image of h(S = imh) .
Proof. (1) Since h is continuous and 0 (0 denotes the set of zeros in R)
is open, therefore S = h1 (0) is open. Each S x = S S x is the
kernel of h|S x : S x R x , hence S x is a sub-A x -module of S x .
(2) Since h is an open map, S is open. Each S x = S x R is the image
of the homomorphism h|S x : S x R x , hence S x is a sub - A x module of R x .

Definition. A homomorphism h : S R is called a monomorphism if
ker h = 0, an epimorphism if im h = R, and and isomorphism if ker h = 0
and im h = R.
Definition. A sequence
hj
h j+1
S j1 S j S j+1
of homomorphisms of sheaves is called exact at S j if ker h j+1 = im h j ;
it is called exact if it is exact at each S j .
If h : S R is a homomorphism, the sequence
i
0 ker h
S
im h 0
20
is exact.
Here i : ker h S is the inclusion homomorphism, and h : S
im h is defined by h (p) = h(p). It is a homomorphism. The other two
homomorphisms are, of course, uniquely determined.
Proof. The statement is the composite of the three trivial statements:
(i) i : ker h S is a monomorphism,
(ii) ker h = ker h,
Lecture 4
17
(iii) h : S imh is an epimorphism.

Definition. A directed set (, <) is a set and a relation <, such that
1) < ( ),
2) if < and < then < (, , ),
3) if , , there exists a such that < and < .
That is, < is reflexive and transitive and each finite subset of has
an upper bound. (We also write > for < ).
Example. Let be the family of all compact subsets C of the plane let
C < D mean C D. is then a directed set.
Definition . A direct system {G , } of abelian groups, indexed by a
directed set , is a system {G } of abelian groups and a system { :
G G }< of homomorphisms such that
(i) : G G is identity,
(ii) if < < , = : G G .
Thus if < < and < k < , then = k k .
The definition of a direct system will be the same even when the G
s are any algebraic systems.
Definition . Two elements a G and b G of G are said to be
equivalent (ab) if for some , a = b.
This relation is easily verified to be an equivalence relation. The
equivalence class determined by a will be denoted by (a).
We will now define addition of equivalence classes. If (a) and (b)
are equivalence classes, a G , b G , choose a > and > and
21
Lecture 4
18
define (a) + (b) = ( a + b).
G
G TTTT
TTTT jjjjjjj
jjTT
tjjjjjjj TTTTT*
G 1
G I
II
II
I$
G 2
tt
tt
ztt
To show that this does not depend on the choice of choose 1 >
and > , let 2 > and > 1 . Then
2 ( a + b) = 2 a + 2 b
= 2 1 (1 a + 1 b),
hence a+ b 1 b. Clearly the class ( a+ b) is independent
of the choice of a and b.
If {G , } is a direct system of abelian groups, the equivalence
classes form an abelian group G called the direct limit of the system.
Proof. That G is an abelian group follows easily from the fact that each
G is an abelian group.

22
The zero element of G is the class containing all the zeros of all the
groups G .
Clearly, if each G is a ring, then G is a ring, and similarly for any
other algebraic system.
The function : G G, where a = (a) is a homomorphism
and if < , = .
Proof.
(a + b) = (a + b) = (a) + (b) = a + b,
( a) = ( a) = (a) = a.
Lecture 4
19
Example . Let (N, ) be the directed set of natural numbers. For each
natural number n let Gn = Z and if n m, let mn : Gn Gm be defined
m!a
. The direct limit is isomorphic to the group of rational
by mn a =
n!
numbers.
Example. Let (N, ) be as before. For each natural number n, let Gn be
the group of rational numbers modulo 1 and if n m let mn : Gn Gm
m!a
be defined by mn a =
. The direct limit is zero.
n!
If {G , } is a direct system of abelian groups and if { f : G H}
are homomorphisms into an abelian group H with f = f , there is
a unique homomorphism f : G H of the limit group G such that
f = f .
Proof. Since f = f , all elements of an equivalence class have the
same image in H. Then f is uniquely determined by f ( a) = f a.
For any two equivalence classes, choose representatives a1 , b1 in 23
some G . Then
f ( a1 + b1 ) = f (a1 + b1 )
= f a1 + f b1
since f = f is a homomorphism. Thus f is a homomorphism.
Lecture 5
Definition . Let S be a unitary left A-module and R a unitary left B- 24
module; a homomorphism : (A, S ) (B, R) is a pair ( : A
B, : S R) where
(a + b) = (a) + (b), (ab) = (a) (b), (1) = 1, a, b A
(s + t) = (s) + (t), (as) = (a) (s), s, t S , a A.
[Remark. For a homomorphism : (A, S ) (B, R), we sometimes
write only for both and .]
Direct systems and direct limits are defined for arbitrary algebraic
systems. Thus if is a directed set and {A , S , }, , where
= (:A A , :S S ), is a direct system of unitary modules, the
direct limit consists of a ring A with unit element, and a unitary left Amodule S , and there are homomorphisms : (A , S ) (A, S ) such
that, if < , = .
The unit element of A is the equivalence class containing all the
unit elements of all A , and the zero of A is the class containing all the
zeros. Thus, if 1 = 0 in some A , 1 = 0 in A and hence for all a A,
a = 1 a = 0 a = 0, and for all s S , s = 0, and the direct limit consists
of the pair (0,0).
If h : (A , S ) (B, R) are homomorphisms with h mu = h ,
there is a unique homomorphism h : (A, S ) (B, R) such that h = h
for each .
Proof. This is proved exactly as in the last lecture.
21
25
Lecture 5
22
Definition. If is a directed set, a subset of is said to be a subdirected set of , if, with the induced order relation, it is a directed set.
Definition . A subset of a directed set is said to be cofinal in if,
for any element , there exists a with < .
If is cofinal in , is a subdirected set.
Proof. This is simple.

P
If = {A , S , }, is a direct system and if is a subdirected
P
set of , then = {A , S , }, , is also a direct system. Let
(A , S ) be its direct limit and : (A , S ) (A , S ) for . Since,
for < = ; there is a unique induced homomorphism
i : (A , S ) (A, S ) with i = .
/ (A , S ) / (A , S )
KKK
UUUU
UUUU
K
UUUU KKKK
i
U
UUUUKK%
U*
(A , S ) UU
(A, S )
If is cofinal in , then i : (A , S ) (A, S ) is an isomorphism.

S
S
Proof. Each class a of A has a representative in
A and if
S
P
P
a,
A and a o in then a o in . Thus i : A A is an
isomorphism. Similarly i : S S is an an isomorphism.
26
If {A , S , } and {B , R , } are direct systems indexed by the

same directed set , and if { f : (A , S ) (B , R ) are homomorphisms such that f = f : (A , S ) (B , R ), there is a
unique homomorphism f : (A, S ) (B, R) of the limit modules such
that f = f .
(A , S )
/ (A , S )
f
(B , R )
/ (B , R )
/ (A, S )
JJ
JJh
JJ
f
f
JJ
%

/ (B, R)
(B , R )
(A , S J)
Lecture 5
23
Proof. Let h = f : (A , S ) (B, R). Then h = f f =

f = f = h . Therefore there is a unique homomorphism f :
(A, S ) (B, R) with f = h , i.e. with f = f .

If {A , S , }, {A , R , }, {A , Q , } are direct systems of unitary left modules (with direct limits S , R, Q respectively) indexed by the
= : A A , and if, for each
same directed set , with =
, S R is an exact sequence of homomorphisms of A

- modules, and if commutativity holds in
(A , S )
/ (A , R )
(A , S )
/ (A , Q )
/ (A , R )

/ (A , Q )
then the sequence of induced homomorphisms of A-modules

g
S
R
Q
is exact.
27
Proof. Consider the following commutative diagram:

(A , S )
/ (A , R )
(A , S )
/ (A , R )
(A, S )

/ (A , Q )
/ (A , Q )
/ (A, R)

/ (A, Q)
(i) im g ker f . For if s1 S , for some , s = S 1 , s S . Then

g( s) im g, and f (g s) = ( f g s) = (0) = (0). Hence
g s ker f .
Lecture 5
24
(ii) ker f im g. For if r1 R, then for some , r = r1 , r R . Let

r ker f . Then f ( )r = 0 = f r, this means that f r 0,
hence there exists a > such that
( f r) = 0 = f ( r), i.e. r ker f
and ker f is equal to im g by assumption, hence there must exist
an s S such that r = g s, i.e.
(Q.e.d.)
r = r = g s = g s im g.
Definition. A bihomomorphism f : (A, R, S ) (B, T ), where A, B are

commutative rings with unit element, R, S , are unitary A-modules and
T is a unitary B-module, is a pair ( f : A B, f : R S T ) such
that f is bilinear. More precisely,
f (a1 + a2 ) = f (a1 ) + f (a2 ), f (a1 a2 )
= f (a1 ) f (a2 ), f (1) = 1, a1 , a2 A,
f (r, a1 s1 + a2 s2 ) = f (a1 ) f (r, s1 ) + f (a2 ) f (r, s2 ), r R, s1 , s2 S ,
f (a1 r1 + a2 r2 , s) = f (a1 ) f (r1 , s) + f (a2 ) f (r2 , s), r1 , r2 R, s S .
28
If A = B and f is the identity, we write f : (R, S ) T .

Given (A, R, S ) there exists a unitary A-module R A S and a bihomomorphism : (R, S ) R A S where im generates R A S such
that, for any bihomomorphism f : (A, R, S ) (B, T ) there is a unique
homomorphism f : (A, R A S ) (B, T ) with f = f .
The module R A S together with the bihomomorphism is called
a tensor product of R and S over the ring A; it is uniquely determined
upto isomorphism. Hence we can say the tensor product of R and S .
Proof. For the proof of the existence of the tensor product, see Bourbaki, Algebre multilineaire. We give only the proof of uniqueness.
Lecture 5
25
(A, R A O S )
llll5
l
lll
(A, R, S )R
RRR
RRR
R(
(A, R A S )
Let R A S and R A S be two tensor products of R and S over

A. Then by definition of the tensor product the bihomomorphism
induces a homomorphism : (A, R A S ) (A, R A S ) such that
= . Similarly induces a homomorphism such that
= .
Then is the identity of R A S , similarly is the identity of
R A S . Hence is an isomorphism.
If we identity R A S with R A S under this isomorphism, will 29
coincide with
Example . Let A be a commutative ring with unit element, S a unitary A-module, and consider A itself as a unitary A-module. Define
: (A, A, S ) (A, S ) as (a, s) = as. Then is verified to be a bihomomorphism. If f : (A, A, S ) (B, T ) is any bihomomorphism, define
the homomorphism f : (A, S ) (B, T ) as f a = f a, f s = f (1, s).
Then
f (a, s) f (as) = f (1, as) = f (a) f (1, s) = f (a, s).
Thus f = f , and S with the bihomomorphism : (A, S ) S is the
tensor product A A S .
Similarly S A A = S with (s, a) = as.
Lecture 6
Definition . A homomorphism : (A, R, S ) (B, P, Q) consists of ho- 30
momorphisms (A, R) (B, P) and (A, S ) (B, Q) (in the sense already
defined), where the homomorphism A B is the same in both cases.
A homomorphism : (A, R, S ) (B, P, Q) induces a map : R
S P Q. We now consider the following diagram:
/ (B, P Q)
OOO
OO
OOO
OOO

/ (B, P B Q)
(A, R A S )
(A, R SO)
/ (C, T U)

/ (C, T C U)
Here , are the induced maps and , , the bihomomorphisms included in the definition of tensor products. Further, the homomorphism
= ,
induces a unique homomorphism as indicated, such that
= . From the uniqueness, we have = bar. If
and similarly
is the identity then also is the identity.
The operator of taking the tensor product commutes with the operation of taking the direct limit.
Proof. Let {A , R , S , }, be a direct system, where each A is a
commutative ring with unit element, R and S are unitary A -modules
and : (A , R , S ) (A , R , S ) are homomorphisms. Then, since
is the identity and = = ( < < ), the system
{A , R A S , } is a direct system. Let its direct limit be denoted by
(A, Q).

27
Lecture 6
28
(A , R , S )
/ (A , R , S )
(A , R A S )
/ (A, R, S )
JJJ
JJJf
JJJ
J%

/ (A , R A S )
(B, T )
t9
t
t
tt
tt
ttt f

/ (A, Q)
31
We define a bihomomorphism : (A, R, S ) (A, Q) as

(r, s) = (r , s ),
where r and s are respresentatives of r R, s S , for the same .
Since
( r , s ) = (r , s ) = (r , s ),
: (R, S ) Q is independent of the choice of representatives. For a
suitable choice of representatives,
(r, bs + cs ) = (r , b s + c s )
= ((b (r , s ) + c (r , s ))
= (b ) (r , s ) + (c ) (r , s )
= b (r, s) + c (r, s )
and similarly
(br + cr , s) = b (r, s) + c (r , s)
Thus : (R, S ) Q is verified to be a bihomomorphism. Clearly im
generates Q, for, each q Q has a representative q in some R A S
P
and, since im generates R A S , q = ki=1 ai . (ri , si ), ai A ,
ri R , si S . Then
q=
k
X
i=1
(ai ) (ri , si ) =
k
X
i=1
(ai )( ri , si )
Lecture 6
32
29
which proves that im generates Q.

We now show that Q together with the bihomomorphism : (R, S )
Q is the tensor product R A S . To do this, let f : (A, R, S ) (B, T )
be any bihomomorphism, then for each , f : (A , R , S ) (B, T ) is
also a bihomomorphism, hence it induces a unique homomorphism f :
(A , R A S ) (B, T ) where for each < , f = f =
f = f = f . Hence, since im generates R A S , f =
f . Therefore there is a unique homomorphism f : (A, Q) (B, T ) with
f = f . Then
f(r, s) = f (r , s ) = f (r , s ) = f (r , s ) = f (r, s);
and the proof of the statement is complete.
Presheaves.
Let be the set of all open sets of X, with the order relation , i.e.
U V is equivalent to saying that U < V. then U U and if U V
and V W then U W and given U, V there exists W = U V with
U W, V W; hence is a directed set.
Definition. A presheaf of modules over a base space X is a direct system
{AU , S U , VU } indexed by , such that (A , S ) = (0, 0), where is the
empty set.
For a presheaf over X, the index set is always the family of all
open sets of X.
The definition of a presheaf includes, as a special case, the definition 33
of a presheaf of B-modules, presheaf of rings with unit element and a
presheaf of abelian groups.
Example 10. Let X be the complex sphere, AU the ring of all functions
analytic in U if U is a non-empty open set and A = O; and, if f AU
and U V, let VU f = f |V, i.e. VU is the restriction homomorphism.
Then {AU , VU } is a presheaf of rings with unit element.
Presheaf of sections. Let a be a sheaf of rings with unit and S a sheaf
of a-modules. For each open U, the ring (U, S ) is a unitary (U, a)module. If V U, let
VU : ((U, a), (U, S )) ((V, a), (V, S ))
Lecture 6
30
34
denote the restriction homomorphism. By convention (, a) = o,

(, S ) = o where denotes the empty set. Thus {(U, a), (U, S ), VU }
is a presheaf denoted by (a, S), and is called the presheaf of sections of
(a, S ).
For each x X let x denote the family of all open subsets of X
containing x. Then x is a subdirected set of . If {AU , S U , VU } is a
presheaf, let (A x , S x ) denote the direct limit of the subsystem
{AU , S U , VU } indexed by x , and let xU : (AU , S U ) (A x , S x ) be
the homomorphism which sends each element into its equivalence class.
If a AU , its image xU a = ax is called the germ of a at x; similarly
S
for s S U . We will denote by a : U A x the function for which
S x
a (x) = ax , and similarly for s : U S x . For each W U we write
x
aW = a (W) = {ax : x W}, and similarly for sW .

[For instance, in Example 10, if f is analytic in U, x U the germ
f x is the class of those functions each of which coincides with f in some
neighbourhood of x.]
S
S
Let A = x A x , S = x S x . Define : A X, : S X by
(A x ) = x, (S x ) = x. Then a : U U, s : U U are the identity
maps on U.
We can take {aU }U , a AU as a base for open sets in A. For, {aU }
covers A and if p aU bV , x = (p), we have p = ax = bx with
a AU , b AV and x U V. Then for some W with x W U V,
WU a = WV b = c say. Since
xU a = xW WU a = xW c = xW WV b = xV b for each x W,
we have cW = aW = bW . Then p cW = aW = bW aU bV . Similarly
the sets {sU } form a base for open sets in S .
Lecture 7
With the notations introduced in the last lecture, we prove
35
Proposition 3. If {AU , S U , VU } is a presheaf of modules over the space

X, then a = (A, , X) is a sheaf of rings with unit and S = (S , , X) is a
sheaf of a-modules.
Proof. If a AU , a : U A is continuous. For, if ax bV with b AV ,
there exists c AW with x W U V such that a (W) = aW = cW =
bW bV . Also a : U A is an open mapping since, for open V U,
aV = (VU a)V is open by definition.

Hence a : U aU being 1-1 is a homeomorphism and the inverse
|aU : aU U is a homeomorphism of aU onto the open set U.
Similarly s : U S is continuous and |sU : sU U is a homeomorphism. Thus and are local homeomorphisms.
For each x X, 1 (x) = A x is a ring with unit element, and 1 (x) =
S x is a unitary left A x -module.
Addition is continuous, for if ax , bx A x (with a AU , b AU1 ,
x U U1 ) and ax + bx cU2 with c AU2 , then for some W with
x W U U2 U2 , WU a + WU1 b = WU2 c. Thus ax aW ,
bx bW and for any p aW , q bW with (p) = (q) = y say, we have
p + q = ay + by = cy cU2 .
The unit is continuous, for if 1x A x is the unit element of A x and
1x bU , then for some V with x V U, VU b is the unit element of
AV . Then bV bU and consists of the unit elements 1y for y V.
Similarly the other operations of a and S are continuous.
36
31
Lecture 7
32
We remark that aU , sU are sections and that such sections form

a base for open sets of A, S .
Remark . The sheaves introduced originally by Leray were actually
presheaves with the indexing set consisting of the family of all closed
sets instead of the family of all open sets.
Example 11. Let X be the circle |z| = 1; for each open set U of X let
S U be the abelian group of all integer valued functions in U and let VU
be the restriction homomorphism. This system is a presheaf and the
induced sheaf has Example 4 as a subsheaf.
Example 12. Let X be the real line, and S U the R module (R denotes
the ring of real numbers) of all real indefinitely differentiable functions
in U and let VU denote the restriction homomorphism. This system is
a presheaf and the sheaf space S of the induced sheaf is not Hausdorff.
Let (a, S ) be a sheaf, (a, S) its presheaf of sections and (a , S )
the sheaf determined by (a, S). We show that (a , S ) and (a, S ) are
canonically isomorphic.
z
(a, ) o
(a,) F
z<
F
h
F"
(a , )
If x U and f (U, a), let hxU f = f (x) A x . Similarly, if

s (U, S ), let hxU s = s(x) S x . Then
hxU : ((U, a), (U, S )) (A x , S x )
37
is a homomorphism and, if x V U, hxV VU = hxU . Then there is an

induced homomorphism hx : ( x , S x ) (A x , S x ) with hx xU = hxU .
In fact hx is an isomorphism. For if p A x , then there is some
section f : U A with f (x) = p, then
p = f (x) = hxU f = hx xU f imhx ,
Lecture 7
33
and if p Ax with hx p = O, choose a representative f (U, a) for p .

Then f (x) = hxU f = hx p = O. Hence, for some V, with x V U,
f |V = o. Therefore
p = xU f = xV VU f = xV OV = (O).
Thus hx : Ax A x is an isomorphism and similarly hx : S x S x
is an isomorphism.
Let h : (A , S ) (A, S ) be given by h|(Ax , S x ) = hx . If f (U, a)
and fU is the induced section in a , given by f(x) = xU f , then
h f(x) = hx f(x) = hx xU f = hxU f = f (x)
and thus h( f(U)) = f (U). The same holds if s (U, S ).
Thus h is an isomorphism of stalks for each x and, since it maps
section fU onto sections f (U), h is a local homeomorphism and hence
is continuous. Thus h : (a .S ) (a, S ) is a sheaf isomorphism.
We identify (a , S ) with (a, S ) under this isomorphism.
38
1
If a A, h a is the class of all sections f : U A where f (U)
contains a, and similarly for h1 s.
P
, } and P = {A , S ,
} are presheaDefinition. If = {AU , S U
VU
P U U VU
P
ves over X, a homomorphism f := is a system { fU } of homo ) (A , S ) such that f
morphisms fU : (AU , S U
U
U
V VU = VU fU , that
is, the following diagram is commutative.
)
(AU , S U
VU
/ (A , S )
V
V
fV
fU
(AU , S U )
VU

/ (AV , S V ).
P P
Let (a , S ), (a, S ) be the sheaves determined by , . Then
P
P
the homomorphism f :
induces a sheaf homomorphism f :
(a , S ) (a, S ).
Proof. For each x, { fU } xU induces a homomorphism f x : (Ax , S x )
(A x , S x ), with f x xU = xU fU and these homomorphism f x define a
function f : (A , S ) (A, S ).
Lecture 7
34
If for a AU , fU a = a, then
ax = xU a = xU fU a = f x xU a = f x (ax ).
39
Thus f (aU ) = aU and f is a local homeomorphism, hence is continuous.

Hence f is a sheaf homomorphism, and this completes the proof.
P
Let be a presheaf which determines the sheaf (a, S ), (a, S ) the
presheaf of sections and (a , S ) the sheaf determined by it. The func-
tions fU : (AU , S U ) (U, a), (U, S ) (where fU a is the section
a : U A determined by a, and similarly for fU s), determine a homoP

morphism, f = { fU } :
(a, S). In general, the homomorphism f
is neither an epimorphism nor a monomorphism, hence obviously not a
isomorphism.
The induced homomorphism f : (a, S ) (a , S ) is the identifying
isomorphism h1 .
Proof. Let a A and suppose that a = xU b, b AU . f (a) is the class
at x containing fU b which is a section with ( fU b)(x) = xU b = a. Thus
f (a) is the class h1 a of all sections g : U A with g(x) = a.
/ (a, s)
u:
u

u

u u
f
/
(a, s)
(a , s ).

h1
Lecture 8
P g P f P
Proposition 4. If
is an exact sequence of homomor- 40

phisms of presheaves, i.e. if each sequence
gU
fU
(AU , S U
) (AU , S U ) (AU , S U
)
g
S
S of sheaves is also
is exact, then the induced sequence S
exact.
Proof. The sequences (A x , S x ) (A x , S x ) (A x , S x ) are exact by a
property of the direct limit.

Induced homomorphism of presheaves of sections. If S , S are
sheaves of a-modules and g : S S is a homomorphism, there
is a homomorphism g : S S of the presheaves of sections with
g U : (U, S ) (U, S ) defined by g U ( f ) = g f . This homomorphism
takes all sections through s S into sections through gs . Thus, with
the usual identification, g inducts the sheaf homomorphism g : s s.
Quotient Sheaves.
Proposition 5. If S is a sheaf of a-modules and S is a subsheaf of
S , there is a unique sheaf S , whose stalks are the quotient modules
S x = S x /S x , such that j : S S where j|S x = jx : S x S x is the
natural homomorphism, is a sheaf homomorphism.
S is the quotient sheaf S = S /S .
S
Proof. If S = x S x /S x is to have a topology such that j : S 41
S is a sheaf homomorphism, S must be covered by sections which
35
Lecture 8
36
are image j f (U) of sections of S , and this uniquely determines the

topology of S . This topology has the property that a set V of S is
open if and only if j1 V is open. Thus S if it exists, is unique.

(U, S ) is a sub-(U, a)-module of (U, S ) and let VU |(U, S )
= (U, S )/(U, S ) and let j : (U, S ) S be
= VU . Let S U
U
U
S denote the homomorthe natural homomorphism. Let
:
S
VU
U
V
P
, } is a presheaf and the
phism induced by VU . Then = {AU , S U
VU
f
j P
sequence 0 (a, S )
(a, S )
0 is exact, where 0 is the

presheaf {AU , 0U , . . .}. Then the induced sequence of sheaves
i
0 S
S
S 0,
where S is the sheaf determined by
the sequence
ix
, is exact. That is, for each x,
jx
0 S x
S x S x 0
is exact. Thus j x induces an isomorphism S x /S x S x , and if we
identify S x with S x /S x , jx : S x S x /S x is the natural homomorphism.
Thus a sheaf S having the required properties exists.
42
Definition . A homomorphism f : (a, S ) (B, R) consists of maps

f : A B and and f : S R, commuting with the projections, such
that the restrictions f x = f |A x B x and fx = f |S x : S x R x give a
homomorphism fx = ( fx , f x ) : (A x , S x ) (B x , R x ).
If f : (a, S ) (B, R) is a sheaf homomorphism and if S , R
are subsheaves of S , R respectively such that f (S ) R , there are
induced homomorphisms
f : (a, S ) (B, R ), f : (a, S /S ) (B, R/R )
with f i = i f , f j = j f , where i denotes the inclusion homomorphism
and j the natural homomorphism of a sheaf onto a quotient sheaf.
Proof. The result is clear for stalks, and the fact that f , f are homomorphisms follows from the fact they are continuous.
Lecture 8
37
Example 13. Let X be the circle |z| = 1, let S be the constant sheaf
(X Z, , X) of integers over X and let S be the subsheaf obtained by
omitting the points (1, n) for n , o. Then S /S is isomorphic to the
sheaf of Example 4.
Example 14. Let X be the complex plane. Let S U be the additive abelian
be the multiplicative abelian
group of functions analytic in U, let S U
group of non-vanishing analytic functions in U and let jU : S U S U

2i
f
be the homomorphism defined by jU f = e . The system { jU } gives
a homomorphism of presheaves and there is an induced exact sequence
of sheaves
0 Z S S 0
where Z is the constant sheaf of integers. An element of (X, S ) is a 43
function analytic in the whole plane.
More generally, in this example X can be replaced by a complex
analytic manifold.
Tensor products of sheaves.
Let a, B be sheaves of commutative rings with unit element let R,
S be sheaves of a-modules and let J be a sheaf of B-modules.
Definition . A bihomomorphism f : (a, R, S ) (B, J) consists of
maps f : A B, f : R + S T , which commute with the projections,
such that, for each x X, the restriction fx : (A x , R x , S x ) (B x , T x ) is
a bihomomorphism.
If r (U, R), s (U, S ) there is a section t : U T defined by
t(x) = f (r(x), s(x)). That t is a map follows from the fact that it is the
composite of the two maps
r,s
U = U + U R + S T ;
where U + U is the set of points (x, x), x U. We write t = fU (r, s).
Then fU : (AU , RU , S U ) (BU , T U ) where AU = (U, a), etc., is a
bihomomorphism, as follows from the property at each x, e.g.
fU (ar, s)(x) = f ((ar)(x), s(x)) = f (a(x)r(x), s(x)
Lecture 8
38
= f (a(x)). f (r(x), s(x)) = ( fU (a))(x). fU (r, s)(x)

= ( fU (a). fU (r, s))(x),
44
and similarly the other properties can be proved. Clearly fU commutes

with the restriction of functions; fV VU = VU fU .
(AU , RU , S U )
fU
VU
/ (BU , T U )
VU
(AV , RV , S V )
fV
xV

/ (BV , T V )
xV
(A x , R x , S x )
fx

/ (B x , T x )
Also f x xU (r, s) = f (r(x), s(x)) = fU (r, s)(x) = xU fu (r, s); i.e.,

f x xU = xU fU . Thus the bihomomorphism f is determined by the
system of bihomomorphisms { fU }.
45
Proposition 6. If a is a sheaf of commutative rings with unit, and R,

S are sheaves of a-modules, there exists a sheaf Q of a-modules and
a bihomomorphism : (R, S ) Q with im x generating Q x for
each x, such that if f : (a, R, S ) (B, J ) is any bihomomorphism
there is a (unique) homomorphism f : (a, Q) (B, J ) with f. = f .
The sheaf Q together with the bihomomorphism is called the tensor
product R a S and is unique upto isomorphism. Each Q x together with
x : (R x , S x ) Q x is the tensor product R x Ax S x . The sections q(U)
P
where q(x) = ki=1 x (ri (x), si (x)) with ri (U, a), si (U, S ) and
S
S
o < k < , form a base for the open sets of = x Q x = x R x Ax S x .
S ) i.e., AU =
Proof. Let {AU , RU , S U , VU } be the presheaf (a, R,
(U, a), etc. For each VU , VU : (AU , RU , S U ) (AV , RV , S V ) there is
an induced homomorphism
VU : (AU , RU AU S U ) (AV , RV AV S V )

and the system AU , RU AU S U , VU is a presheaf determining some
sheaf (a, Q).
Since tensor products and direct limits commute, for each x, there is
a unique induced bihomomorphism x : (R x , S x ) Q x with x xU =
Lecture 8
39
xU U and Q x together with x is the tensor product R x Ax S x . An

P
arbitrary element q = ki=1 U (ri , si ) RU AU S U determines a section
qU : U where
qU (x) = xU q =
=
k
X
i=1
k
X
xU U (ri , si ) =
k
X
x xU (ri , si )
i=1
x (ri (x), si (x))
i=1
and such sections from a base for Q.

If f : (a, R, S ) (B, J ) is any bihomomorphism, there is an
S) (B,
J ) of presheaves.
induced homomorphism { fU } : (a, R,
Then, if fU : (AU , RU AU S U ) (BU , T U ) is the homomorphism induced

fU,
by

J is a homomorphism of
fU : AU , RU A S U , VU B,
U
presheaves which induces a homomorphism f : (a, Q) (B, J ) of

sheaves. Then
(AU , RU , S U )
rrr
rr U
r
r
rr

rLrr (A , R S )
U
U AU U
LLL
LLL
LLL fU
LL
%
(BU , T U )
VU
VU
VU
/ (AV , RV , S V ) xV / (A x , R x , S x )
HH
PPP
HH fx
PPPfV
HH
PPP
x
V
HH
PPP
HH

/ (A x , Q x )
/ (AV , RV AV S V ) xV
ww
ww
nnn
n
w
n
n
ww
fx
fV
nnn
ww
{ww

vnnn xV
/ (B x , T x )
/ (BV , T V )
46
fx x xU (r, s) = xU fU U (r, s) = xU fU (r, s) = f x xU (r, s);

thus fx x = f x and hence f = f . Since im x generates R x Ax S x = Q x ,
fx is uniquely determined by fx ; hence f is unique. That Q and are
unique upto isomorphism is proved in the usual manner.
Corollary. If : (a, R, S) (B, J, U) is a homomorphism, there is a
unique induced homomorphism : (a, R a S ) (B, J B U) with
Lecture 8
40
=
(a, R, )
/ (B, J, U)
(a, R a S
/ (B, J U),
Lecture 9
1 Cohomology groups of a space with coefficients in
a presheaf
47
Definition . A covering (an indexed covering) {Ui }iI of a space X is a

system of open sets whose union is X.
P
Definition . If
= {S U , VU } is a presheaf of A-module where A is a
fixed ring with unit element, a q-cochain f (q = 0, 1, . . .) of a covering
P
U = {Ui }iI with values in is an alternating function of q + 1 indices
with
f (io , i1 , . . . , iq ) S Uio Uiq
or more briefly f () S U where is the simplex io , . . . , iq . In particular f (i0 , i1 , . . . , iq ) = 0 if Ui0 Uiq = . ( A function f is called
alternating if
(i) f (i0 , i1 , . . . , iq ) = 0 if any two of i0 , . . . , iq are the same,
(ii) f ( j0 , j1 , . . . , jq ) = f (i0 , i1 , . . . , iq ) according as the permutation
j0 , . . . , jq of i0 , . . . , iq is even or odd).
We will often write (V, U) for VU .
P
P
The q-cochains of U with values in form an A-module C q (U , ).
P
For q < 0, we define C q (U , ) = 0.
41
Lecture 9
42
P
Definition . The coboundary q+1 f (or simply f ) of f C q (U , ) is
the function on (q + 1)-simplexes defined by
(q+1 f )() =
q+1
X
(1) j (U , U j ) f ( j ),
j=0
48
where j = i0 , . . ., ij , . . ., iq+1 = i0 , . . ., i j1 , i j+1 , . . . , iq+1 is the j-th

face of = io , . . . , iq+1 .
P
P
If f C q (U , ), then f C q+1 (U , ).
Proof. It is sufficient to verify that f is an alternating function, e.g.,

f (i1 , io , . . . , iq+1 ) = (U , U1 ) f (1 ) (U , Uo ) f (o )
+
q+1
X
(1) j (U , U j ) f (i1 , io , . . . , ij , . . . , iq+1 )
j=2
= (U , Uo j ) f (o ) + (U , U1 ) f (1 )
q+1
X
(1) j (U , U j ) f (io , i1 , . . . , ij , . . . , iq+1 )

j=2
= f (io , i1 , . . . , iq+1 )
and, if io = i1 ,
f (io , i1 , . . . , iq+1 ) = (U , Uo ) f (o ) (U , U1 ) f (1 )
= 0,
where = io , . . . , iq+1 .
It follows, since (U , U j ) is a homomorphism, that

X
X
q+1 : C q (U , ) C q+1 (U , )
is a homomorphism. One verifies by computation that q+1 q f = 0 for
P
f C q1 (U , ), using the fact that for j k
, . . . , iq+1
k j = k (io , . . . , ij , . . . , ik+1 , . . . , iq+1 ) = io , . . . , ij , . . . , ik+1
1. COHOMOLOGY GROUPS OF.....
43
= j k+1 .
49
(The computation is carried out at the end of the lecture).

Thus in q ker q+1 in the sequence
0 C o (U ,
X
X
) C (U , )
C q1 (U ,
X q
X q+1
) C q (U , )
P
The quotient module H q (U , ) = ker q+1 / im q is called the q-th
P
cohomology module of U with coefficients in the presheaf .
P
The elements of the module Z q (U , ) = ker q+1 are called qP
cocycles and the elements of the module Bq (U , ) = im q are called
P
P
P
q-coboundaries. Since Bo (U , ) = 0, we have H o (U , ) Z o (U , ).
Definition . A covering W = {V j } jJ is said to be a refinement of the

covering U = {Ui }iI if for each j J there is some i I with V j Ui .
If W is a refinement of U , choose a function : J I with V j

U( j) . Then there is a homomorphism
X
X
+ : C q (U , ) C q (W , )
defined by
+ f () = (V , U() ) f ()
where = j0 , . . . , jq ; and = ( j0 ), . . . , ( jq ).
+ commutes with since
q+1 + f () =
=
=
q+1
X
(1)k (V , Vk )+ f (k )
k=0
q+1
X
(1)k (V , Vk )(Vk , U(k ) ) f ((k )),
k=0
q+1
X
(1)k (V , U(k ) ) f ((k ),
k=0
+ q+1 f () = (V , U ) f ()
Lecture 9
44
=
=
q+1
X
(1)k (V , U )(U , Uk ) f (k ),
k=0
q+1
X
(1)k (V , Uk ) f (k )
k=0
50
and k = k . Hence there is an induced homomorphism,

X
X
W U : H q (U , ) H q (W , ).
P
P
The homomorphism W U : H q (U , ) H q (W , ) is independent
of the choice of .
Proof. Let : J I, : J I be two such choices. Let the set J be
linearly ordered and define the function
X
X
kq1 : C q (U , ) C q1 (W , )
by
(k
q1
q1
X
f )() =
(1)h (V , Uh ) f (h )
h=0
for = j0 , . . . , jq1 with j0 < j1 < < jq1 , where

h = ( j0 ), . . . , ( jh ), ( jh ), . . . , ( jq1 ),
and let kq1 f be alternating. Then kq1 is a homomorphism, since
(V , Uh ) : S Uh S V is a homomorphism.

51
Using the facts that, for = j0 , j1 , . . . , jq ,

h i = i h+1
(0 i h q 1),
h i = i+1 h
(0 h < i q),
h h1 = h h
0 0 = ,
(1 h q),
q+1 q = ,
one finds that

q kq1 f + kq q+1 f = + f + f.
1. COHOMOLOGY GROUPS OF.....
45
(The computation is given at the end of the lecture.)

P
This holds for q = 0 with the obvious meaning of k1 : C 0 (U , )
P
0. Thus if r H q (U , ) is represented by a cocycle f , + f + f is a
coboundary and W U r is uniquely determined.
If the covering W is a refinement of W and W is a refinement of U ,
then W W W U = W U and U U is the identity.

Proof. If W = Wk kK is a refinement of W , choose 1 : K J so that
Wk V1 k . Then Wk V1 k U1 k and 2 : K I can be chosen to
be 1 . Then
(+1 + f )() = (W , V1 )(+ f )(1 )
= (W , V1 )(V1 , U1 ) f (1 )
= (W , U2 ) f (2 )
= (+2 f )().
Thus +1 + = +2 and so for the induced homomorphisms,
X
X
W W W U = W U : H q (U , ) H q (W , ).
Similarly, for the refinement U of U , : I I can be chosen to be

P
P
the identity, hence U U : H q (U , ) H q (U , ) is the identity.

(1)
q+1 q = 0.
(2)q kq1 + kq q+1 = + + .
(1)
(q+1 q f )() =
q+1
X
(1) j (U , U j )(q f )( j )
j=0
q+1
q
X
X
(1) j
(1)k (U , U j )(U j , Uk j ) f (k j )
j=0
k=0
q
k X
X
(1) j+k (U , U j k+1 ) f ( j k+1 )+
j=0 k=0
q
q+1 X
X
(1) j+k (U , Uk j ) f (k j )
j=k+1 k=0
52
53
Lecture 9
46
=0
(2)
(q kq1 f )() =
q
X
(1)i (V , Vi )(kq1 f )(i )
i=0
q1
q X
X
=
(1)i+h (V , Uh i ) f (h i )
i=0 h=0
q1
h X
X
(1)i+h (V , Ui h+1 ) f (i h+1 )
i=0 h=0
q1
q X
X
(1)i+h (V , Ui+1 h ) f (i+1 h )
i=h+1 h=0
q
h1
XX
i+h1
(1)
i=0 h=0
q1
q+1 X
X
(V , Ui h ) f (i h )
(1)i+h1 (V , Ui h ) f (i h )(kq q+1 f )()
i=h+2 h=0
q
X
(1)h (V , Uh )(q+1 f )(h )
h=0
q+1
q X
X
=
(1)i+h (V , Ui h ) f (i h ), (q kq1 f + kq q+1 f )()
h=0 i=0
=
=
q
X
h=0
q
X
(V , Uh h ) f (h h )
(V , Uh h ) f (h h )
h=0
q
X
h=0
q+1
X
(V , Uh+1 h ) f (h+1 h )
(V , Uh h1 ) f (h h1 )
h=1
= (V , U0 0 ) f (0 0 ) (V , Uq+1 q ) f (q+1 q )
= (V , U ) f ( ) (V , U ) f ()
= (+ f )() (+ f )().
54
Lecture 10
P
Let h = {hU } :
be a homomorphism of presheaves, i.e., each 55
S is a homomorphism and, if V U, h
hU : S U
U
V VU = VU hU .
We define, for each q 0, the mapping
X
h+ : C q (U , ) C q (U , )
by (h+ f )() = hU f (). Then, h+ is a homomorphism since each hU
is a homomorphism.
h+ commutes with
Proof.
(h+ q+1 f )() = hU (q+1 f )()
q+1
X
= hU
(1) j (U , U j ) f ( j ),
j=0
and
h f )() =
q+1
X
(1) j (U , U j )(h+ f )( j )
q+1
X
(1) j (U , U j )hU j f ( j )
q+1 +
j=0
j=0
q+1
X
= hU
(1) j (U , U j ) f ( j )
j=0
47
Lecture 10
48
Q. e. d.
P
Hence h+ induces a homomorphism hU : H q (U , ) H q (U , ). h+
commutes with +

Proof.
(h+ + f )() = hV (+ f )()
= hV (V , U ) f ()
= (V , U )hU f ()
= (+ h+ f )().
56
P
Hence hW W U = W U hU : H q (U , ) H q (W , ), i.e., the
following diagram is commutative.

H q (U ,
W U
P
H q (W ,
hU
/ H q (U , P)
W U
hW
P
/ H q (W ,
).
P h P g P
is a sequence of homomorphisms of presheaves,
If
then gh induces a homomorphism
X
X
q
(gh) : C (U , ) C (U , )
+
such that (gh)+ = g+ h+ and, if h is the identity, h+ is the identity.

Proof.
((gh)+ f )() = (gh)U f ()
= gU hU f ()
= (g+ h+ f )().
If h is the identity, i.e., if each hU is the identity, then (h+ f )() =
hU f () = f (); so h+ is the identity, q.e.d.
Lecture 10
49
If one has the commutative diagram of homomorphisms of presheaves, i.e., gh = h1 g1 , then

P
/ P
g1
P
h1
P
g+ h+ = (gh)+ = (h1 g1 )+ = h+1 g+1 ,

and hence gU hU = h1U g1U .
P h P g P
If U = {Ui }iI is a covering and the sequence
of 57
homomorphisms of presheaves is exact, then the sequence
C q (U ,
is also exact.
X
X g+
X
h+
) C q (U , ) C q (U , )
P
Proof. (i) If f C q (U , ) is an element of im h+ , clearly f ker g+ ,
hence im h+ ker g+ .
P
(ii) Linearly order the index set I, and let f C q (U , ) be an element of ker g+ . Then f () ker gU = imhU for each q-simplex
, hence there is an element r in the module corresponding to
the open set U , of the presheaf , such that hU (r) = f (). If
= (i0 , . . . , iq ) with i0 < < iq , define the function t on by
t() = r. If = ( j0 , . . . , jq ) is a permutation of = (i0 , . . . , iq )
define t on by t( ) = t() according as is an even or odd
permutation of . If is a q-simplex in which two indices are
repeated, define t() to be zero. It then follows that t C q (U , )
and it is easily verified that h+ (t) = f , hence ker g+ imh+ .

P i P j P
If the sequence 0
0 of homomorphisms of
presheaves is exact, there is an induced homomorphism
U : H q (U ,
) H q+1 (U ,
).
Lecture 10
50
Proof. Since the homomorphisms i+ , j+ commute with the homomorphism , there is commutativity in the following diagram:
()
/ C q (U ,
P
/ C q+1 (U ,
i+
P
/ C q (U ,
i+
/ C q+1 (U , P)
j+

/ C q (U , P )
/0

/ C q+1 (U , P )
/0
j+

58
P
P
P
Since the sequence 0
0 is exact, each
row of the diagram is an exact sequence of homomorphisms. We will
P
P
construct a homomorphism : Z q (U , ) H q+1 (U , ) which is
P
zero on Bq (U , ), and hence will induce a homomorphism from
P
P
H q (U , ) H q+1 (U , ).
P
To do this, let r Z q (U , ), and choose s C q (U , ) with j+ s =
r. Since j+ s = j+ s = r = 0, s ker j+ and by exactness, there is
a unique t C q+1 (U , ) with i+ t = s. Then i+ t = i+ t = s = 0,
hence t = 0. Let H q+1 (U , ) be the element represented by t. To
show that is unique. let s1 , t1 be the result of a second such choice,
then j+ (s s1 ) = r r = 0 and s s1 = i+ u for a unique u C q (U , ).
Then since i+ is a monomorphism and
i+ (t t1 ) = (s s1 ) = i+ u = i+ u,
59
hence t t1 = u. Thus t and t1 represent the same element

H q+1 (U , ).
Let = (r). If r = ar1 + br2 Z q (U , ), suppose that r1 = j+ s1 ,
s1 = i+ t1 and that r2 = j+ s2 , s2 = i+ t2 , and let 1 = (r1 ), 2 = (r2 )
be the elements represented by t1 and t2 . Then since j+ , and i+ are
homomorphisms,
r = j+ (as1 + bs2 ), (as1 + bs2 ) = i+ (at1 + bt2 )
Lecture 10
51
and, since at1 + bt2 represents a1 + b2 , we have

(r) = a(r1 ) + b(r2 ).
Thus : Z q (U , ) H q+1 (U , ) is a homomorphism.
P
If r Bq (U , ), let r = v . For some w C q1 (U , )v = j+ w.
Then j+ (w) = j+ w = r and there exists a unique t C q+1 (U , )
with i+ t = (w) = 0, hence t = 0; i.e., (r) = 0. Thus induces a
homomorphism
U : H q (U , ) H q+1 (U , ).
Q.e.d.
W ,U commutes with U , i.e., the following diagram is commutative.
U
P
/ H q+1 (U , P )
H q (U , )
W U
W U
P
H q (W ,
/ H q+1 (W , P )
Proof. + commutes with j+ , , i+ .

If 0
0 is exact, then the sequence
X
iU
q
0 H U , H U ,

X jU
X
U
X
q+1
q
q
H U , H U ,
H U,
is exact.
Proof. The exactness of this sequence is the result of six properties of 60

the form ker im and im ker. Each can be easily verified in (). (See
Eilenberg-Steenrod, Foundations of Algebraic Topology, p. 128).
Lecture 10
52
P i P j P
P i1 P j1 P
If 0
0 and 0 1
1 0
P P P
P P1 P
are exact sequence, and if h : ( , ,
) ( 1 , 1 ,
1 ) is a homomorphism commuting with i, j, i1 , and j1 , then hU commutes with
U .
Proof. The homomorphism h+ commutes with the homomorphisms j+ ,
and i+ , q.e.d.

With the same assumptions as in the above statement, we then have
the following commutative diagram, in which each row is exact.
/
H 0 (U ,
...
H 0 (U ,
1)
iU
h
U
hU
P
H q (U , )
...
P
H q (U , )
jU
H q (U ,
1)
i1U
H q (U ,
1)
j1U
H q+1 (U ,
h
U
h
U
hU
H q (U ,
H q (U ,
1)
1U
H q+1 (U ,
1)
...
...
Definition. A proper covering of X is a set of open sets whose union is

X.
61
A proper covering U = {U} of X may be regarded as an indexed

covering UU UU if each open set of the covering is indexed by itself.

Every covering Ui iI has a refinement which is a proper covering, e.g.,
the set of all open sets U such that U = Ui for some i I.
Let be the set of all proper coverings of X and let U < W mean
that W is a refinement of U . Then is a directed set, for 1) U < U 2)
if U < W and W < W then trivially U < W and 3) given U , W there
exists W with U < W , W < W , e.g, W may be chosen to consists of
all open sets W with W = V for some U U , V W .
(There is no set of all indexed coverings).
P

The system H q (U , ), W U U ,W is then a direct system. Its
P
direct limit H q (X, ) is called the q-th cohomology module (over A) of
P
P
P
X with coefficients in . Let U : H q (U , ) H q (X, ) denote the
usual homomorphism in to the direct limit.
P
P
If h : is a homomorphism of presheaves, there is a induced
homomorphism
h : H q (X,
) H q (X,
X
)with h U = U hU .
Lecture 10
53
Proof. This follows from the fact that hU W U = W U hU .

Q.e.d.
P i P j P
O is an exact sequence of presheaves,
If O
there is an induced exact sequence
0
H 0 (X,
H q (X,
P
H q (X, )
H q (X,
H q+1 (X,
Proof. This is a consequence of the fact that the direct limits of exact
sequences is again an exact sequence.

P P P
P P P
If h : ( , , ) ( 1 , 1 ,
1 ) is a homomorphism of exact 62
sequence of presheaves
i
/ P
P
/P
/ P
i1
P
/0
j1
P
/0
and h commutes with i, j, i1 , j1 then the following diagram, where h is

the homomorphism induced from h, is a commutative diagram.
...
/ H q (X, P )
...
P
/ H q (X,
1)
/ H q (x, P)
i1
P
/ H q (X,
1)
/ H q (x, P )
j1
P
/ H q (X,
1)
/ H q (x, P )
/ ...

/ H q+1 (X, P1 )
/ ...
Proof. The result is a consequences of the fact that hU commutes with

iU , jU and U .
Lecture 11
If the coefficient presheaf is the presheaf of sections of some sheaf S 63
of A-modules, we write C q (U , S ) instead of C q (U , S ) etc. Then, if
U = Uio Uiq is called the support of the simplex = io , . . . , iq ,
a q-cochain f C q (U , S ) is an alternating function which assigns to
each q-simplex a section over the support of .
If U = {Ui }i I is any covering of X, H o (U , S ) is isomorphism to
(X, S ).
Proof. A 0-cochain belonging to C 0 (U , S ) is a system ( fi )iI , each fi
being a section of S over Ui . In order that this cochain be a cocycle,
it is necessary and sufficient that fi f j = O over Ui U j ; in other
words, that there exist a section f (X, S ) which coincides with fi
on Ui for each i I. Thus there is an isomorphism U : (X, S )
Z 0 (U , S ) H 0 (U , S ).

Proposition 7. H 0 (X, S ) can be identified with (X, S ).
Proof. Since W U U = W , there is an induced isomorphism :
(X, S ) H 0 (X, S ) with U U = . A homomorphism h : S S1
of sheaves induces a homomorphism {hU } of the presheaves of section
and hence induced homomorphism hU , h with commutativity in

(X, S )
/ H 0 (U , S )
/ H 0 (X, S )
h
hU
hX
(X, S1 )
/ H 0 (U , S )
1
55
/ H 0 (X, S )
1
Lecture 11
56
64
Thus we can identify (X, S ) with H o (X, S ) under if we also

identify hx : (X, S ) (X, S1 ) with h : H o (X, S ) H o (X, S1 ).

Definition. A system Ai iI of subset of a space X is called finite if I is
finite, countable if I is countable. The system is said to be locally finite
if each point x X has a neighbourhood V such that V Ai = expect
for a finite number of i. (This finite number may also be zero).

We notice that a locally finite system Ai iI is always point finite.

(A system Ai iI of subsets of X is said to be point finite if each point
x X belongs to Ai for only a finite number of i).

If Ai iI is locally finite, so is B j jJ if J I and each B j A j . If

Ai iI is locally finite, so is A i iI , where A i denotes the closure of Ai ,
S
S
S
and Ai = Ai . In particular, if each Ai is closed, so is Ai
iI
iI
iI
Definition . The order of a system {Ai }iI of subsets of X is 1 if Ai is

the empty set for each i I. Otherwise the order is the largest integer n
such that for n + 1 values of i I, the Ai s have a non-empty intersection,
and it is infinity if there exists no such largest integer.
65
Definition. The dimension of X, denoted as dim X, is the least integer n

such that every finite covering of X has a refinement of order n, and
the dimension is infinity if there is no such integer.
Definition . A space X is called normal, if for each pair E, F of closed
sets of X with E F = , there are open sets G, H with E G, F H
and G H = .
Definition . A covering U = {Ui }iI of the space X is called shrinkable
if there is a refinement W = {Vi }iI of U with V i Ui for each i I.
X is normal if and only if every locally finite covering of X is shrinkable.
Proof. See S. Lefschetz, Algebraic Topology, p.26.
If X is normal, dim X n if and only if every locally finite covering

of X has a refinement of order n.
Lecture 11
Proof. See C.H.Dowker, Amer.Jour of Math. (1947), p.211.
57
Definition. A space X is called paracompact if every covering of X has

a refinement which is locally finite.
If X is paracompact and normal, dim X n if and only if every
covering of X has a refinement of order n.
Proof. (i) Since X is paracompact, every covering of X has a locally
finite refinement and since X is normal and dim X n, using the
above result, every locally finite covering has a refinement of order 66
n, thus every covering has a refinement of order n.
(ii) Since every covering of X has a refinement of order n, in particular, every locally finite of X has a refinement of order n, hence,
since X is normal, using the above result, we obtained dim n.

Remark. Since a paracompact Hausdorff space is normal, (see J. Dieudonne, Jour. de Math. 23, (1944), p.(66), this result holds, in particular,
when X is a paracompact Hausdorff space.
If a covering U of X has a refinement W of order n, then it has a
proper refinement W of order n.
Proof. If W = {V j } jJ has order n, let W be the proper covering
formed by all open sets W such that, for some j j, W = V j . Then W
has order n and is a refinement of U .

P
If X is paracompact and normal and dim X n, then H q (X, ) = 0
P
for q > n and an arbitrary presheaf .
Proof. Replace the directed set of all proper coverings of X by the cofinal sub-directed set of all proper covering of order n. If U ,
P
q > n and f C q (U , ), then f (U0 , U1 , . . . , Uq ) S Uo Uq = S = 0
for any q + 1 distinct open sets of U . If the open sets U0 , U1 , . . . , Uq
are not all distinct, then f (U0 , U1 , . . . , Uq ) = 0 since f is alternatP
P
ing. Hence C q (U , ) = 0 and hence also H q (U , ) = 0. Therefore
P
H q (X, ) = 0, q > n.
Lecture 11
58
If
67
0.
P
is a presheaf which determines the zero sheaf, then H o (X, ) =
P
Proof. For any element H 0 (X, ) choose a representative
P
P
f z0 (U , )SH 0 (U , ), where U is some proper covering of X, so
that U f = . For each x X choose an open set U = (x) such
P
that x (x) U . Since determines the 0-sheaf, one can choose
an open set V x such that x V x (x) and (V x , (x)) f ((x)) = 0.

Then W = V x xX is a refinement of U , and, for each x, (+ f )(x) =
(V x , (x)) f ((x)) = 0, hence + f = 0. If W is a proper refinement
of W , choose 1 : W X so that each W V1 (W) and (1 )+ f =
+1 + f = 0. Thus W U f = 0 and hence = U f = W W U f = 0.
P
Hence H o (X, ) = 0.

This result is not true in general for the higher dimensional cohomology groups. However, if the space X is assumed to be paracompact and
normal, we will prove the result to be true for the higher dimensional
cohomology groups.
Lecture 12
P
Proposition 8. If X is paracompact and normal and if is a presheaf 68
P
which determines the zero sheaf, the H q (X, ) = 0 for all q 0.
P
Proof. Let f C q (U , ) where U = {Ui }iI is any locally finite covering. Since X is normal, we can shrink U to W , W = {Wi }iI with
i Ui . For each x X choose a neighbourhood V x of x such that the
W
following conditions are satisfied:

a) If x Ui , V x Ui ,
b) If x Wi , V x Wi ,
i , V x Wi = ,
c) if x < W
d) if x Uio Uiq = U , Vx U f () = 0
Conditions a) and b) can be satisfied, for the coverings U and W
being locally finite, each x is contained only in a finite number of sets of
i
the coverings. To see that condition c) can be satisfied, consider all W
i . The union of these sets is the closed since W is locally
for which x < W
finite, and x is in the open complement of this union. Next, by condition
P
a), V x U , and since
determines the 0-sheaf, we can choose V x
small enough so that d) is satisfied. We can thus always choose V x small
enough so that the above conditions are fulfilled.

If the V x are chosen as above, the covering V x xX is a refinement 69
of U . Choose the function : X I so that x W(x) , then by b),
V x W(x) U(x) . Then
+ f () = + f (xo , . . . , xq ) = (V , U ) f ((xo ), . . . , (xq )).
59
Lecture 12
60
If V = , + f () = 0. If V , then V xo meets each V x j , hence

(x j ) . Then since xo W
(x j )
meets each W(x j ) and hence by c), xo W
U(xi ) , by a) V xo U(xi ) for each j and hence V xo U() . Hence
+ f () = V U() f () = V Vx0 Vx0 U() f ()
= 0 by d).
Thus + f () = 0 for all , hence + f = 0.
P
If U1 is a proper covering and f C q (U1 , ), there is a locally finite
refinement U of U1 (since X is paracompact). Then there is a refinement W of U (found as above), a proper refinement W1 of W (existence of W1 is trivial) and a function 1 : W1 U1 with V 1 (V) for
S
P
V W1 such that +1 f = 0. Hence every element of
H q (U1 , )
P (U1 proper)
is equivalent to zero, i.e., the direct limit H q (X, ) consists only of zero.
70
Example 15. Let X be the space with four points a, b, c, d and let a
P
base for the open sets be the sets (a, c, d), (b, c, d), (c), (d). Let be the
presheaf for which S U = z, the group of integers, if U = (c, d); and S U =
P
0 otherwise. The homomorphisms VU are the obvious ones. Then
P
determines the 0-sheaf, but H 1 (X, ) = Z. The space X paracompact
but not normal.
i
S
S is an exact sequence of sheaves, then
If 0 S
iU
jU
0 (U, S ) (U, S ) (U, S )

is exact and hence
0 S S S
is exact
Proof. We will show that for ker jU imiU (the rest is trivial). Let
f ker jU . Then, x U, j f (x) = ( jU f )(x) = 0x and hence by exactness,
f (x) = ip for some p S x . Thus f (U) i(S ). But i : S i(S ) is
homeomorphism. Then g : U S , where g(x) = 11 f (x), is a section
of S over U, and f = iU g.
Lecture 12
61
One cannot in general complete the sequence

0 (U, S ) (U, S ) (U, S )
by a zero on the right as the following example shows.
Example 16. Let X be the segment {x : 0 x 1}. Let G be the 4group with elements 0, a, b, c. Let S be the subsheaf of the constant
sheaf G = (X G, , X) formed by omitting the point (0, a), (0, c), (1, b),
(1, c). Let S be the subsheaf of S formed by omitting all the points
(x, a), (x, b). Let S = (X Z2 , X) and let j : S S be the 71
homomorphism induced by j : G Z2 where j(a) = j(b) = 1, j(c) =
j(0) = 0. The the sequence
j
S
S 0
0 S
is exact, but the sequence
0 (X, S ) (X, S ) (X, S ) O,
i.e.,
0 O O Z2 O
is not exact.
P i P j P
If 0
is an exact sequence of presheaves, there is

P
P
an image presheaf
and a quotient presheaf Q such that the

0
sequences
X
i X jo X
O
0,
0
Q0
are exact. These sequences are natural in the sense that if h is a homomorphism of exact sequences, commuting with i and j:
0
/ P
P
/
/P
/ P
P
/

/P
Lecture 12
62
then there are induced homomorphisms h of the exact cohomology sequences

...
/ H q (X, P )
h
72
j0
/ H q (X, P)
h
P
...
/ H q (X,
...
/ H q (X, P0 )
1)
P
/ H q (X,
/ H q (X, P0 )
h
j0
1)
/ ...
P
/ H q (X,
/ H q+1 (X, P )
1 0)

/ H q+1 (X, P1 )
/ ...
and
i
...
P
/ H q (X,
10 )
/ H q (X, P )
P
/ H q (X,
1)
/ H q (X, Q)
h
/ H q+1 (X, P0 )
/ ...
/ H q (X, Q1 )

/ H q+1 (X, P10 )
/ ...
commuting with i , j0 , 0 and i , j , respectively.

P
, }, let S = im j . Then since j : P P is a
Proof. If = {S U
U
VU
oU
homomorphism,
maps
im
j
into
im
j
U
V . Hence, writing S oU = im jU
VU
and QU = S U /S oU , there are induced homomorphisms oVU and VU

with comutativity in
SU
joU
iU
/ S
oU
oVU
VU
SV
joV
/ S
U
jU
VU
/ S
oV
VU
iV
/ QU
/ S
V
jV
/ QV
Clearly the systems

VU } are
o = {S oU , oVU } and Q = {QU ,
presheaves and the sequences
i
O

o O
and
i
O
QO

o
Lecture 12
63
are exact. Since h commutes with i, j,

/ P
/P
i1
P
P
/ P
j1
P
jU
SU
/ S
U
h
U
hU
j1U
S 1U
/ S
1U
h
U maps S oU = im jU into S 1oU = im j1U . Hence there are induced 73
homomorphisms hoU , hU with comutativity in

joU
SU
/ S
oU
iU
/ S
U
iU
h
oU
hU
S 1U
joU
/ S
1oU
jU
/ QU
h
U
h U
jU
/ S
1U
/ Q1u
Since h is a homomorphism of presheaves, hU commutes with VU
and h
U with VU . Hence, since joU and jU are epimorphisms and joU ,
jU commute with and h, h and commutes with and h U with
oU
oVU
VU , i.e., the diagrams given below are commutative:
S oU
oVU
^==
== joU
==
=
SU
h
oU
VU
/ SV
hU
S 1U
S 1oU
VU
VU
QU ^<
<<
<<jU
<<
VU
h
V
/ S 1V
==
== jo
==
=
oVU
/ S 1oV
/ S V
h
U
h U
/ QV
A

jV
S U
h
oV
hV

jo

/ S oV
@

joV

S 1U
VU

j
VU
Q1U
h V
/ S 1V
<<
<< j
<<
<
/ Q1V

Thus h, h
o , h , h are homomorphisms of presheaves commuting
with jo , i, j.
jo
/ P
1o
P
h
o
P
/ P
jo
P
/Q
h
j
/ Q1
Lecture 12
64
74
Thus {h} is a homomorphism of the exact sequences

/ P
P
jo
/P
/ P

/ P1
/0
P
/0
/Q
/0
h
o
jo
1o
commuting with i and jo , and a homomorphism of the exact sequences

/ P
h
o
/ P
P
1o
h
i
P
h
j

/ Q1
/0
commuting with i and j. Therefore the induced homomorphisms of

the exact cohomology sequences commute with i , jo , o and i , j ,
respectively.
Lecture 13
Example 17. Let X consist of the natural numbers together with two 75
special points p and q. Each natural number forms an open set. A
neighbourhood of p (resp. q) consists of p (resp; q) together with all but
a finite number of the natural numbers. Let S U = Z if U consists of all
but a finite number of the natural numbers and if V U is another such
set, let VU : Z Z be the identity. If U is an open set not containing
all but a finite number of the natural numbers or if it contains either p
or q, let S U = O and let VU , UW be the zero homomorphisms. Then
P
= {S U , VU } is a presheaf determining the 0-sheaf, but H 1 (X, Z) = Z.
The space X is T 1 and paracompact but not normal.
Example 18. Let R be a set with cardinal number N1 let S = 2R be the
set of all subsets of R and let T = 2S be the set of all subsets of S . If
r R, let r T be the largest subset of S , which is such that, each
of its elements considered as a subset of R contains the elements r. Let
R T consists of all r, for all r R and let T 1 = T R .
Let X be a space consisting of (1) all elements r R and (2) all
triples (t, r1 , r2 ) with t T 1 , r1 , r2 R and r1 , r2 . Each point (t, r1 , r2 )
is to form an open set. Neighborhoods of points r of the first kind are
sets N(r; s1 , . . . , sk ), where o k < and s1 , . . . , sk S , consisting
of r together with all points (t, r1 , r2 ) with r (r1 , r2 ) and, for each 76
i = 1, . . . , k, either r si t or r < si < t. [cf. Bings Example G,
Canadian Jour, of Math. 3 (1951) p.184].
For sets U X of cardinal number 2 and consisting of points
of the second kind, let S U = Z and, if V U is another such set, let
65
Lecture 13
66
VU : Z Z be the identity. If U is an open set containing any point

of the first kind
or consisting of at most one point, let S U = 0.P Then
P
= S U , VU is a presheaf determining the 0-sheaf, but H (X, ) , 0
(although dim X = 0). X is a completely normal, Hausdorff space but
is not paracompact. (A space X is said to be completely normal if each
subspace of X is normal).
i
S
S 0 is an exact sequence of sheaves, let
If 0 S
j
i
be the image and quotient presheaves in 0 S
S
S
S o , Q,
for which the sequences
jo
0 S
S So 0,
j
i
0 S0
Q 0,
S
are exact. Then Q determines the zero sheaf.

Proof. An arbitrary element of a stalk Qk of the induced sheaf has the
form xU jU f where x U and f (U, S ). Since j maps S onto
S , there is an open set V, x V U, for which f |V im jV . Then
f = f |V im jV = im iV and by exactness jV
f = 0. Hence
VU
VU
xU jU f = xV VU jU f = xV jV
VU f = 0.
77
Therefore the sheaf determined by Q is the 0-sheaf.
Note . In example 16, if o (U, S ) = im jU , we have o (U, S ) =

(U, S ) for all U expect X, but o (X, S ) = 0, o (X, S ) = Z2 .
Thus Q X = Z2 , Q U = 0 for all smaller U, and thus Q determine the
0-sheaf.
i
Proposition 9. If X is paracompact and normal and if 0 S

S 0 is an exact sequence of sheaves, there is an exact cohomology
sequence
i
0 H o (X, S ) H q (X, S )
H q (X, S ) H q (X, S ) H q+1 (X, S ) .
Lecture 13
67
If h : (S , S , S ) (S 1 , S1 , S 1 ) is a homomorphism of exact
sequences, commuting with i and j,
/ S
/ S
/ S
1
/S
/0
/ S1

/ S
1
/0
the induced homomorphisms S of the cohomology sequences commute

with i , j and , i.e. the following diagram is commutative:
...
/ H q (x, S )
...

/ H q (X, S1 )
/ H q (X, S )
/ H q (x, S )
h
i
/ H q+1 (X, S )

/ H q (X, S 1 )

/ H q (X, S1 )
/ ...

/ H q+1 (X, S )
1
/ ...
Proof. As before, if So , Q denote the image and quotient presheaves

in the exact sequence of presheaves of sections
j
i
0 S
S
S ,
we obtain the exact sequence of presheaves
78
j0
0 S
S S0 0,
and
j
i
0 S
S
Q 0,
From these exact sequences of presheaves we obtain the following
Lecture 13
68
exact cohomology sequences:
=0
H q1 (X, Q)

...
/ H q (X, S )
/ H q (X, S )
/ H q (X, So ) o / H q+1 (X, S )
LLL
q8
LLL
qqq
q
i
L
q
LLL
q
j
% qqq
jo
/ ...
H q (X, S )
j
=0
H q (X, Q)
= 0 for
Since Q determine the 0-sheaf, by Proposition 8, H q (X, Q)
all q 0 and hence, by exactness, i is an isomorphism. Hence if

= o (i )1 : H q (X, S ) H q+1 (X, S ), the cohomology sequence
0 H o (X, S ) H o (X, S ) H o (X, S )
H q (X, S ) H q (X, S ) H q (X, S ) H q+1 (X, S )
79
is exact.
Next, since the homomorphism h commutes with i and j, the induced homomorphism h of presheaves also commutes with i and j:
0
/ S
/ S
1
/ S
h
/ S

/ S
1

/ S
1
Hence, the induced homomorphism h of the cohomology modules

commutes with i , jo , i and o . Thus in the exact cohomology sequences, h commutes with i , j and .
Lecture 13
69
Note . If X is not paracompact and normal, in general, i is not an isomorphism, to be precise, the cohomology sequence is not defined. One
does, however, have the exact sequence
0 H o (X, S ) H o (X, S ) H o (X, S )
H 1 (X, S ) H 1 (X, S ) H 1 (X, S )
as one sees from the exact sequences
70
0 = H o (X, Q)
/ H o (X, S )

j
/ H o (X, S ) o / H o (X, S ) o / H 1 (X, S )

o
MMM
q8
MMM
qqq
q
M
q
i
M
qq
j MMM
& qqq
H o (X, S )

j
/ H 1 (X, S ) o / H 1 (X, S )
o
LLL
LLL
i
L
j LLL&

H 1 (X, S )
= 0.
H o (X, Q)
Lecture 13
Lecture 13
71
The following examples show that Proposition 9 is not true, in gen- 80

eral,unless the space is both paracompact and normal.
Example 19. Let X consist of the unit segment I with the usual topology
and of two points p and q. A neighbourhood of p (resp. q) consists
of p (resp. q) together with the whole of I. Let S , S , S be the
sheaves S , S , S of Example 16 over I together with zeros at p
and q. A neighbourhood of 0 p (resp. 0q ) consists of the zeros over a
neighbourhood of p (resp. q). Then there is an exact sequence
i
0 S
S
S 0
where i, j correspond to those in Example 16. Since H 1 (X, S ) =
H 2 (X, S ) = 0 and H 1 (X, S ) = Z2 , there is no exact cohomology
sequence. The space X is paracompact but not normal.
Example 20. Let X consists of a sequence of copies In of the unit segment together with two special points p and q. A neighbourhood of p
(resp q) consists of p (resp. q) together with all but a finite number of
the segments In . Let G be the 4-group and let S be the subsheaf of
(X G, , X) consisting of zero at p and q and on each In a copy of
the sheaf S of Example 16. Let S be the subsheaf of (X Z2 , , X)
formed by omitting the points (p, 1), (q, 1) and let the homomorphism
j : S S be induces by j : G Z2 as defined in Example 16. Then
there is exact sequence
i
S
S 0
0 S
but H 1 (X, S ) = H 2 (X, S ) = 0 while H 1 (X, S ) , 0. Thus there is 81
no exact cohomology sequence. The space X is paracompact and T 1 but
not normal.
Example 21. Let R, S , T 1 be as in Example 18. Let X be the space
consisting of (1) the elements r R and (2) segments Intr1 r2 where n is a
natural number, t T 1 , r1 and r2 are in R, and r1 , r2 . Neighbourhoods
of the points r are sets N(r; n, s1 , ., sk ) where n is a natural number and
s1 , . . . , sk S , consisting of r together with all segments Imtr1 r2 with
Lecture 13
72
m > n, r (r1 , r2 ) and, for each i = 1, . . . , k, either r si t or

r < si < t.
Let G be the 4-group and let S be the subsheaf of (X G, , X)
consisting of zero at each r and a copy of the sheaf S of Example 16 on
each Intr1 r2 . Let S be the subsheaf of (X Z2 , , X) formed by omitting
(r, 1) for all r, and let the homomorphism j : S S be induced by
j : G Z2 as mentioned before. Then there is an exact sequence
0 S S S 0
but H 1 (X, S ) = H 2 (X, S ) = 0 while H 1 (X, S ) , 0. Thus again,
there is no exact cohomology sequence. X is a perfectly normal Hausdorff space but is not paracompact. (A space X is said to be perfectly
normal if, for each closed set C of X there is a continuous real valued
function defined on X and vanishing on C but not at point x X C.
Perfectly normal spaces are completely normal.)
Lecture 14
Definition. A resolution of a sheaf G of A-modules is an exact sequence 82
of sheaves A-modules
e
dq
d1
0G
S o S 1 S q1 S q
such that H q (X, S q ) = 0, p 1, q 0.
There are than induced homomorphisms
do
dq
d1
d q+1
0 (X, S o ) (X, S q1 ) (X, S q ) (X, S q+1 )
for which im dq ker dq+1 , i.e., dq+1 dq = 0. The A-modules (X, S k )

(k 0) together with the homomorphisms dq form a formal cochain
complex denoted by (X, S ). Let the q th cohomology module of the
complex (X, S ) be denoted by H q (X, S ) = ker dq+1 / im dq .

Example 22. Let X be the unit segment x : o x 1 , and let G be
the subsheaf of the constant sheaf Z2 formed by omitting the points (0,1)
and (1,1). A resolution
(1)
d1
0G
S o S 1 0
of G is obtained by identifying G with the sheaf S of Example 16 and

taking S , S , i, j for S o , S 1 , e, d1 . (That H p (X, S q ) = 0, p 1,
q 0 can be verified.) The induced sequence
0 (X, S o ) (X, S ) 0
73
Lecture 14
74
is
83
0 0 Z2 0.
Another resolution
(2)
d1
d2
0G
S o S 1 S 2 0
of G is obtained by taking e, S o as before, S 1 = Z2 +Z2 , S 2 = Z2 , d1 =

k j where k(x, 1) = (x, (1, 0)), and d2 with d2 (x, (1, 1)) = d2 (x, (0, 1)) =
(x, 1).
Another resolution
(3)
d1
0G
S o S 1 0
of G is obtained by taking S o to be subsheaf of Z2 formed by omitted

(0,1), with e : G S o as the inclusion homomorphism and with d1 as
the natural homomorphism onto the quotient sheaf S 1 = S o /G .
Yet another resolution
(4)
d1
0G
S o S 1 0
of G is obtained by taking S o = Z2 and S 1 = S o /G .

In each case H 1 (X, S ) = Z2 , H p (X, S ) = 0, p > 1.
84
Example 23. Let X be the sphere x2 + y2 + z2 = 1, and let G be the

constant sheaf Z2 . Let R denote the constant sheaf Z2 + Z2 with i : G
R defined by i(1) = (1, 1). Let R R consist of all zeros together
with ((x, y, z), (0, 1)) for z < 0; let S o = R/R and let j : R S o be
the natural homomorphism. Let e = ji : G S o .
Let Io be the quotient sheaf S o /e(G ) and let h : S o Io be the
natural homomorphism. The stalks of Io are Z2 on the equator and 0
elsewhere.
Let I be the quotient sheaf of R consisting of Z2 + Z2 on the equator and 0 elsewhere. Identify Io with the subsheaf of I consisting of
all zeros and all ((x, y, 0), (1, 1)), and let k : Io I be the inclusion
homomorphism. Let I be for y > 0 and ((x, y, 0), (1, 0)) for y < 0. Let
Lecture 14
75
S = I/I and let l : I S 1 be the natural homomorphism. Let

d1 = 1kh : S o S 1 . Let S 2 = S 1 /d1 S o and let d2 be the natural
homomorphism. Then from the diagram:
0
R

= Io
>R
|
}}
||
}}
j ||
}
}
|
}} e
|| h
o
/
G
S
i

/I

/ S
d2
/ S2
we see that
e
d1
d2
0G
S o S 1 S 2 0
is a resolution of G and the induced sequence
0 (X, S o ) (X, S 1 ) (X, S 2 ) 0
is
85
0 Z2 + Z2 Z2 + Z2 Z2 + Z2 0
and
H o (X, S ) = H 2 (X, S ) = Z2 , H 1 (X, S ) = 0
Proposition 10. If X is paracompact normal and if
dq
0G
S o S q1 S q
is a resolution of G , there is a uniquely determined isomorphism :
H q (X, S ) H q (X, G ).
If
e
d1
q
q1 d
0 G1
S1o S11 S1
S1
Lecture 14
76
is a resolution of another sheaf G1 and if
h : (G , S o , S 1 , . . .) (G1 , S1o , S11 , . . .)
is a homomorphism commuting with e, d1 , d2 , . . ., then the induced homomorphism H commutes with .

H q (X, S )
/ H q (X, G )
H q (X, S1 )
86
/ H q (X, G1 )
Proof. The homomorphism h : H q (X, G ) H q (X, G1 ) is the usual

induced homomorphism. Now, since h commutes with dq , q 1, h also
commutes with the homomorphisms
dq : (X, S q1 ) (X, S q ) (q 1),
and hence there is an induced homomorphism
h : H q (X, S ) H q (X, S1 ).

Let zq = im dq = ker dq+1 S q ; then there are exact sequences
do1
e
0G
S o z1 0
(1)
q+1
q
q do
g+1
q i
S z
0 (q 1)
0z
q
where iq is the inclusion homomorphism and do is the homomorphism

q
q
induced by dq : iq do = dq . Since h commutes with d, h maps zq in z1 ,
and commutes with i, do .
0
/G
/ So
do1
/ G1
/ z1
/ 0,
/ So
1
do1

/ z1
1
/ 0,
Lecture 14
77
/ zq
iq
/ Sq
/ zq+1
/ zq
1
doq+1
iq
/ 0,
/ Sq
1
q+1
do

/ zq+1
1
/ 0.
Hence the induced homomorphism h of the corresponding exact

cohomology sequences also commutes also commutes with e , do ,
and i , do , respectively.
d1
S o S 1 is exact, so is the se- 87

Case 1. q = 0. Since 0 G
d1
quence 0 (X, G )
(X, S o ) (X, S 1 ). Then H o (X, S ) =
1
ker d = ime, but e is a monomorphism and (X, G ) = H o (X, G ), hence
e : H o (X, G ) H o (X, S ) is an isomorphism commuting with h . Let
= e1 .
Case 2. q > 0. The exact cohomology sequence corresponding to that
exact sequences (1) for q 1 (where zo = G ) is
doq
iq1
0 (X, zq1 ) (X, S q1 ) (X, zq ) H 1 (X, zq1 ) 0
since H 1 (X, S q1 ) = 0. Thus induces an isomorphism

q
: (X, zq )/imdo H 1 (X, zq1 ) (q 1).

q+1
Since im iq = ker do
isomorphism
= ker dq+1 , the monomorphism iq induces an
iq : (X, zq )/ im do im iq / im dq
= ker dq+1 / im dq
= H q (X, S ).
Thus we have an isomorphism
(iq )1 : H q (X, S ) H 1 (X, zq1 ) (q 1)
Lecture 14
78
commuting with h , since h commutes with and (i )1 .
0

/ (X, zq )
(X, S q1N)
do
NNN
NNN
N
dq NNN'
/ H 1 (X, zq1 )
/0
iq
(X, S qN)
NNN q+1
NNdN
NNN
N'

/ (X, zq+1 )
/ (X, S q+1 ).
q+1
do
iq+1
88
Also, the exact cohomology sequences corresponding to (1) contain

qp+1
)
(iqp )
(do
0 H p1 (X, zqp+1 ) H p (X, zqp ) 0

for 1 < p < q and
do1
0 H q1 (X, z1 ) H q (X, G ) 0 for
p = q.
Thus we have isomorphisms (q 1),
(iq )1
H q (X, S ) H 1 (X, zq1 ) H q1 (X, z1 ) H q (X, G )

commuting with h . Let be the composite of these isomorphisms,
: H q (X, S ) H q (X, G ).
Theorem 1 (Uniqueness theorem). If X is paracompact normal, and if
e
d1
d1
dq
0G
S o S 1 S q1 S q ,
0G
S1o S11
89
q
q1 d
S1
S1 ,
are two resolutions of the same sheaf G of A-modules, there is a canon-
Lecture 14
79
ical isomorphism
: H q (X, S ) H q (X, S1 ).
Moreover, if h : (S o , S 1 , S 2 , . . .) (S1o , S11 , S12 , . . .) is a homomorphism commuting with e, d1 , d2 , . . .,
S
|>
e ||
|
||
||
/G
@@
@@
@@
e @@

d1
/ S1
d2
/ ...
/ S q1

/ S1
1
d1
S1o
dq
d2

/ S q1
1
/ ...
/ Sq
/ ...
dq

/ S1q
/ ...
then the induced homomorphism

h : H q (X, S ) H q (X, S1 )
is the isomorphism .
Proof. We have the canonical isomorphisms , 1 ,
H Q (X, S )
H q (X, G ) H q (X, S1 );
let = 1
1 .
There is commutativity in the diagram:

H q (X, S )
/ H q (X, G )
h
H q (X, S1 )

/ H q (X, G ),
where the homomorphism h on the right is the identity. Hence the

homomorphism h on the left is equal to 1
1 = .
Lecture 15
We now given example to show that the uniqueness theorem fails in 90
more general spaces.
Example 24. Let X consist of the unit segment I = {x : o x 1}
together with two points p, q. A neighbourhood of p (resp. q) consists
of p (resp. q) together with all of I. Let G be the subsheaf of the constant
sheaf Z2 formed by omitting the points (p, 1), (q, 1), (0, 1), (1, 1). Let
S o be the subsheaf of the constant sheaf Z2 + Z2 formed by omitting
(p, a), (q, a) for all a , 0 and (0, (1, 0)), (0, (1, 1)), (1, (0, 1)), (1, (1, 1)).
Let S 1 be the subsheaf of Z2 formed by omitting (q, 1). Let S 2 have
the stalk Z2 at p and 0 elsewhere; a neighbourhood of (p, 1) consists of
(p, 1) together with all the zeros over I. Then there is a resolution
d1
d2
d3
0G
S o S 1 S 2 0
and the corresponding sequence
d2
d1
d3
0 (X, S o ) (X, S 1 ) (X, S 2 ) 0

is
0 0 0 Z2 0;
so H 2 (X, S ) = Z2 .
There is also a resolution
91
e
d1
0G
R o 0
81
Lecture 15
82
where e is an isomorphism; then H p (X, R) = 0 for all p. There is even

a homomorphism h, commuting with e, d1 , . . .,
o
R
{=
e {{{
{{
{{
/G
h
BB
BB
B
e BBB

So
d1
d1
/0
/0
/0

/ S1

/ S2
/ 0.
[The space X is not normal.]

Definition . A sheaf S of A-modules is called fine if for every closed
set E in X and open set G in X with E G, there is a homomorphism
h : S S such that
i) h(s) = s
ii) h(s) = 0(s)
if
(s) E,
if
(s) < G
example of a fine sheaf.

Example 25. For each open subset U of X, let S U be the A-module
of all functions f : U A. If V U, define VU to be restriction
homomorphism. Let S be the sheaf of germs of functions determined
P
by the presheaf = {S U , VU }. If E G with E closed and G open, let
hU : S U S U be defined by
(hU f )(x) = f (x) G (x)
92
where x U and G is the characteristic function of G.

(G (x) = 1 A
if
x G, G (x) = 0 A
if
x < G).
P
P
Then {hU } :
is a homomorphism. If h : S S is the
induced homomorphism, h(s) = s if (s) E and h(s) = 0 if (s) < G,

hence the sheaf is fine.
Lecture 15
83
Exercise. If M is any non-zero A-module and the space X is normal, the

constant sheaf (X M, , X) is fine if and only if dim X 0.
Note. The set of all endomorphisms h : S S forms an A-algebra, in
general, non commutative, where h1 h2 is the composite endomorphism.
The identity 1 : S S is the unit element of the algebra.
If S = (S , , X) is a sheaf and X1 is a subset of X, let X1 and S 1 =
1 (X1 ) have the induced topology. Then (S 1 , |S 1 , X1 ) is a sheaf called
the restriction of S to X1 .
If X is normal, the restriction of a fine sheaf S to any closed set C
is fine.
Proof. Let E be any closed subset of C and G any open subset of C with
E G. Extend G to an open set H of X, G = H C. Then, since X is
normal, E closed in X, H open in X with E H, there is an open subset
V of X with E V V H. Since S is fine, there is a homomorphism
h : S S with
h(s) = s
= 0(s)
if (s) E,
if (s) X V.
93
Then if S1 is the restriction of S to C, h|S 1 : S 1 S 1 is a homomorphism h1 : S1 S1 and we have

h1 (s) = h(s) = s
= 0(s)
if (s) E
if (s) C G C G X H X V.
Proposition 11. If X is normal, U = {Ui }iI a locally finite covering of

X, and if the restriction of S to each U i is fine (in particular, if S is
fine), there is a system {li }iI of homomorphisms li : S S such that
i) for each i I there is a closed set Ei Ui such that 1i (S x ) =
0x if x < Ei ,
Lecture 15
84
ii)
iI
11 = 1.
(1 denotes the identity endomorphism S S ).

Proof. Using the normality of X, we shrink the locally finite covering
U = {Ui }iI to the covering U {Vi }iI with V i Ui and we further shrink
i Vi .
the locally finite covering U to the covering U = {Wi }iI with W
Since the restriction Si of S to U i is fine, there is a homomorphism
gi : Si Si with
gi (s) = s
= o(s)
94
i,
if (s) W
if (s) U i V i .
Let the homomorphism hi : S S be defined by

hi (s) = gi (s) if
= 0(s)
(s) U i ,
if (s) X Ui .
(This definition is consistent, since gi (s) = 0(s) on U i Ui ). This

hi : S S is continuous and is a homomorphism with
i,
if (s) W
hi (s) = s
= 0(s)
if (s) X V i .
Let the set I of indices be well ordered and define the homomorphisms li : S S by
Y
1i (1 h j ) hi ,
j<i
where the product is taken in the same order as that of the indices.
Each point x X has a neighbourhood N x meeting Ui for only a

finite number of i, say i1 , i2 , . . . , iq with i1 < i2 < < iq . If (s) N x ,
li (s) = (1 hi1 ) (1 hik1 )hik (s), i = ik , k = 1, . . . , q,
= 0(s) for all other i.
Lecture 15
85
Clearly li (S x ) S x and li |S x : S x S x is a homomorphism. The

function li is continuous on each 1 (N x ) and coincides on the overlaps
of two such neighbourhoods, hence li : S S is continuous. Thus
li : S S is a homomorphism, and
hi (S x ) = 0x , x < V i ,
hence
li (S x ) = 0x , x < V i .
Take Ei = V i Ui . Let (s) N x ; then for some ik , 1 k q,
(s) Wik and hence hik (s) = s. Hence
95
(1 hi1 ) (1 hiq )(s) = 0.
Therefore
X
1i (s) = hi1 (s) + (1 h11 )hi2 (s) + + (1 hi1 ) (1 hiq1 )hiq (s)
iI
= s (1 hi1 ) (1 hiq )(s)

= s.
Note . The homomorphisms li are usually not uniquely determined and
they cannot therefore be expected to commute with other given homomorphisms.
Lecture 16
Let {li }iI be a system of endomorphisms of a fine sheaf S correspond- 96
ing to a locally finite covering {Ui }iI of a normal space X. Each li gives
a homomorphism li (U) : (U, S ) (U, S ) for each open U and
P
li (U) has a meaning and is the identity endomorphism of (US ).
iI
Also li determines a homomorphism
li (U) : (Ui U, S ) (U, S )

defined by
(li (U)g)(x) = li (g(x))
if x Ui U,
if x (X Ui ) U.
=0
One verifies that the following diagrams are commutative.

li (U)
/ (U, S )
nn7
li (U) nnn
n
(Ui U,S )
(Ui U,S )
nnn

nnn
/ (Ui U, S )
(Ui U, S )
(U, S )
li (Ui U)
(Ui U, S )
li (U)
/ (U, S )
VU
(Ui V,Ui U)
(Ui V, S )
li (V)
/ (V, S )
If X is normal, S is fine and U = {Ui }iI is a locally finite covering

of X, then H q (U , S ) 0 for q 1.
Proof. Let kq1 : C q (U , S ) C q1 (U , S ) for q 1 be the homomorphism defined by
X
(kq1 f )() =
li (U )( f (i)),
iI
87
Lecture 16
88
97
where i = i, io , . . . , iq1 if = io , . . . , iq1 . (This infinite sum of sections is finite neglecting zeros, in some neighbourhood of each point.)
Using the fact that o (i) = and j (i) = i j1 () for j > 0, one
verifies that
q kq1 f + kq q+1 f = f.
(The computation is given at the end of the lecture.)
Hence each cocycle f is a coboundary q.e.d.
Proposition 12. For a fine sheaf S over a paracompact, normal space

X, H q (X, S ) = 0 for q 1
Proof. H q (X, S ) = 0 q 1 for each locally finite covering U of the
space X. Since the space X is paracompact, this means that H q (X, S ) =
0, q 1.

Corollary . If X is paracompact and normal, any exact sequence of
sheaves
e
d1
dq
0G
S o S q1 S q . . . ,
where each S q (q 0) is fine, is a resolution of G .
Definition. A sheaf S is called locally fine, if for each open U and each
x U, there is an open V with x V U such that the restriction of S
to V is fine.
If X normal, a fine sheaf S is locally fine.
Proof. The restriction of S to an arbitrary closed set is fine.
98
Proposition 13. If X is paracompact normal, a locally fine sheaf S is

fine.
Proof. Let E G1 , with E closed and G1 open. If G2 = X E then
{Gi }i=1,2 is a covering of X. Since S is locally fine, for each x Gi ,
there is an open V x with x V x Gi such that the restriction of S to
V x is fine. Since X is paracompact, there is a locally finite refinement
U {U j } jJ of {V x } xX , hence U is also a refinement of {Gi }. If U j V x ,
Lecture 16
89
then U j V x , and V x , being closed in X, is normal. Since the restriction

of S to V x is fine, the restriction of S to U j is also fine. Now by
P
l j = 1 and l j
proposition 11, there exist endomorphisms l j such that
jJ
is zero outside a closed set E j U j . Choose the function : J (1, 2)

so that U j G( j) and let
li =
l j,
i = 1, 2.
( j)=i
Then I1 + I2 = l and
li (s) = 0 if (s) X
U j X Gi .
( j)=i
Hence
l1 (s) = 0 if
(s) X G1
and
l1 (s) = s if (s)X G2 = E.
l1 thus gives the required function, and this completes the proof.
Corollary . If X is paracompact and normal, any exact sequence of

sheaves
e
d1
dq
0G
S o . . . S q1 S q ,
where each S q (q 0) is locally fine, is a resolution of G .
The following examples shows that, in more general spaces, fineness
need not coincide with local fineness.
Example 26. Let X have points a, b, . . . , h with base for open sets consisting of ( f ), (g), (h), (d, f, h), (e, g, h), (c, f, g), (b, e, g, h), (a, d, e, f,
g, h). Let S be the subsheaf of the constant sheaf Z2 formed by omitting
(c, 1), ( f, 1). Then S is fine but not locally fine. In fact, V = (c, f, g) is
the least open set containing c and the restriction of S to V = X (h)
is not fine. (X is not normal.)
99
Lecture 16
90
Example 27. Let T be the space of ordinal numbers 1 with the usual
topology induced by the order. Let A be the space or ordinal numbers
o and let X be the subspace of T XA formed by omitting the point
(1 , o ). Let S be the constant sheaf Z2 over X. Then S is locally
fine, for every point has a closed neighbourhood which is normal and
zero dimensional. But S is not fine. If B is the set of even numbers,
then B A. Let E = 1 B and G = T B. Then E G X
with E closed and G open. There is no endomorphism of S which is
(X in neither paracompact nor
the identity on E and is zero outside G.
normal.)
100
101
Example 28. The space M of Quart. Jour. Math. 6 (1955), p. 101 is normal and locally zero dimensional but not zero dimensional. Therefore
the constant sheaf Z2 is locally fine but not fine. (M is not paracompact).
q kq1 f + kq q+1 f = f.
q
X
q q1
k f () =
(1) j (U , U j )(kq1 f )( j )
j=0
q
X
X
= (1) j (U , U j )
Ii (U j ) f (i j )
j=o
q
X
(1) j
l1 (U )(Ui , U j j ) f (i j ).
l1 (U )(q+1 f )(i)
j=0
q q+1
i
q+1
X
li (U )
li (U)(Ui , U ) f ()
(1) j (Ui , U j i) f ( j i)
j=0
X
i
li (U)
q+1
X
(1) j (U j , Ui j1 ) f (i j1 )
j=1
Lecture 16
91
li (U ) f ()
q q1
li (U )
q q+1
f () + k
q
X
(1) j+1 (Ui , Ui j ) f (i j ).
j=0
= i li (U ) f () = f ()
Lecture 17
In this and the next lecture, we shall give a proof of de Rhams theorem. 102
Let X be an indefinitely differentiable (C ) manifold of dimension
n, which is countable at infinity (i.e) a countable union of compact sets);
we assume that X is a Hausdorff space. Then X is paracompact and
normal. (Dieudonne, Jour. de Math. 23 (1944)). The set E p (U) of all
C (alternating) differential p-forms on an open set U forms a vector
space over the field of real numbers. Exterior differentiation gives a
homomorphism d p ,
d p : E p1 (U) E p (U)
with d p+1 d p = 0. In particular, there is a sequence
do
d1
dp
dn
dn+1
0 E o (X) E 1 (X) E p (X) E n (X) 0

with im d p ker d p+1 . Let
H p (E (X)) = ker d p+1 / im d p .
This vector space of the closed p-forms modulo the derived p-forms
is called the p-th de Rham cohomology vector space of the manifold X.
If V U, the inclusion map i : V U induces a homomorphism
VU = i1 : E p (U) E p (V)
which commutes with d. Thus the system E p = {E p (U), VU } is a 103
93
Lecture 17
94
presheaf which determines a sheaf p, called the sheaf of germs of

p-forms, and
d p : E p1 E p
is a homomorphism of presheaves which induces a homomorphism
d p : p1 p .
There is a constant presheaf {R, VU } where R is the filed of real
numbers and VU : R R is the identity. This presheaf determines the
constant sheaf R. There is a homomorphism e : R E o (U), (E o (U) is
the space of C functions on U) where e(r) is the function on U with
the constant value r, and further e commutes with VU . Thus, there is an
induced sheaf homomorphism e : R o . Hence, we have a sequence
of homomorphisms of sheaves
dp
d1
0 R
o p1 p . . . ,
with d p+1 d p = 0.
There is a homomorphism
{ fU } : {E p (U), VU } p ,
104
p denotes the presheaf of sections of p ), where the image of an el(

ement of E p (U) is the section over U which it determines in p . Then
d commutes with fU and, in particular, with fx . Thus we have the commutative diagram:
E p1 (X)
dp
/ E p (X)
dp
fX
fX
(X, p1 )
/ (X, p ).
If a p-form E p (X) is not zero, there is some point x X at

which it does not vanish, and hence xX , 0. Thus fX , 0, i.e., fX
is a monomorphism. (In the same way, fU is a monomorphism for each
open set U.) fX is also onto, hence an isomorphism. For, if g (X, p ),
Lecture 17
95
then since p is the sheaf of germs of p-forms, for each x X there is a

neighbourhood U x of x and a p-form x defined on U x such that section
g and the section determined by x coincide on U x . Then {U x } forms a
covering for X, and since the section determined by x and y coincide
on U x Uy , using the fact that fU is a monomorphism for each open set
U, we see that the forms x and y themselves coincide on U x Uy ;
hence they define a global from , such that fX () = g. Thus fX gives
an isomorphism of the sequences:
/ E o (X)
d1
/ ...
fX
/ E p (X)
fX
/ (X, o )
dp
/ E p1 (X)
d1
fX
/ ...
/ ...
dp
/ (X, p1 )

/ (X, p )
/ ...
Hence there is an induced isomorphism of the cohomology vector 105

spaces:
fX : H q (E (X)) H q (X, )
(q 0).
Poincares lemma. The sequence
d1
dp
0 R o p1 p
is exact.
Proof. We have to prove that for each point a X, the sequence
e
d1
p
p1 d
0 Ra
oa a
is exact, where the subring Ra of oa , consisting of the germs of constant

functions at a, is identified with the filed R of real number. Choose
a coordinate neighbourhood W of a with coordinates (x1 , . . . , xn ) and
p
suppose that a = (0, . . . , 0). Then a is the direct limit of the system
p

E (U), VU aU where U belongs to the cofinal set of those spherical
neighbourhoods x21 + + x2n < r2 which are contained in W.

For each such U, let
h : E o (U) R
Lecture 17
96
k p1 : E p (U) E p1 (U)
and
(p 1)
be the homomorphisms defined by

h( f ) = f (0, . . . , 0),
and
k p1 ( f (x1 , . . . , xn )dxi dxi p )
Z 1
P
X
i j . . . dxi p
=(
f (tx1 , . . . , txn )t p1 dt)
(1) j1 xi j dxi1 . . . dx
0
106
j=1
respectively. (The formula on the right is an alternating function of

i1 , . . . , i p ; h and k p1 are then extended by linearity to E o (U) and E p (U)
respectively.) One now verifies that
eh + k0 d1 = 1,
d p k p1 + k p d p+1 = 1 (p 1),
where 1 denotes the identity map.
(The computation is carried out at the end of the lecture.)
Thus f ker d implies that f im e and ker d p+1 implies that
im d p . Hence ker d1 = im e and ker d p+1 = imd p , since already
im e ker d1 and im d p ker d p+1 . Hence the sequence
d1
dp
0 RU
E o (U) . . . E p1 (U) E p (U) . . .
is exact, and since exactness is preserved under direct limits, therefore
the limit sequence
e
d1
dp
0 Ra
oa ap1 ap . . .
is exact, q.e.d.
The sheaf p is fine.
Proof. Since the space X is paracompact and normal, by Proposition 13
(Lecture 16), it is enough to prove that the sheaf p is locally fine. Let
U be an open set of X, and let a U.
Lecture 17
107
97
We may assume that U is compact and that it is contained in some

coordinate neighbourhood N of a. Let V be an open subset with a V
and V U. We will now prove that the restriction Vp of p to V is fine.
Extend G to an open
Let E G with E closed and G open in V.
set H U, so that G = V H. Then U is covered by a finite number

of spherical neighbourhoods S i contained in N, such that S i either does
not meet E or is contained in H.
For each i, choose an indefinitely differentiable function fi which
is positive inside S i and vanishes out side S i . We construct one such
n
P
function as follows: For the spherical neighbourhood (x j bi j )2 < ri2 ,
j=1
let
gi (r) = 0
Z
=
(r ri ),
)
1
ri
exp
dt ( r ri ),
ri
2
(t 2 )(t ri )
r
(
)
Z ri
1
ri
=
exp
dt (0 r )
ri
2
(t 2 )(t ri )
ri/2
ri
and define fi by
v
u
tX
n
fi (x1 , . . . , xn ) = gi (
(x j bi j )2 .
j=1
Let 1 (x) = fi (x), summed for all i for which S i meets E and let
P
2 (x) = fi (x), summed for all the remaining i. Then 1 +2 is positive
in U and, if
(x) = 1 (x)/(1 (x) + 2 (x)),
P
is indefinitely differentiable in U, is zero outside H and is constant, 108

equal to 1, in a neighbourhood of E.
Let h : E p (W) E p (W), for open W U be defined by h() =
. Then h is a homomorphism commuting with YW , Y open in U,
Y W. Hence h induces a homomorphism h : p (U) p (U) for
which
h(b ) = b if b E,
Lecture 17
98
h(b ) = 0b if b V G U H.
(b denotes the germ determined by at b U.)
Proposition 14. There is an isomorphism
fX : H p (E (X)) H P (X, R).
Proof. By the corollary to Proposition 12, the exact sequence

d1
dp
0 R
o p1 p . . .
is a resolution of the constant sheaf R and hence, by Proposition 10,
there is an isomorphism
: H p (x, ) H p (X, R);
but we already have (as proved in the earlier part of this lecture) an
isomorphism
fX : H p (E (X)) H p (X, ).

109
(1)eh + k0 d1 = 1.
(2)d p k p1 + k p d p+1 = 1 (p 1).
(1) If f (x) = f (x1 , . . . , xn ) E 0 (U),

1
eh f (x) = f (0, . . . , 0) and d f (x) =
n
X
Di f (x)dxi , (),
i=1
hence ko d1 f (x) =
Pn R 1
f (0), thus eh f (x) +
i=1 0 Di f (tx)dt
o
k d1 f (x) = f (x).
xi =
R1 d
f (tx)dt = f (x)
0 dt
(2) If = f (x1 , . . . , xn )dxi1 dxi p ,
p p1
d k
Z
=d (
p
f (tx)t
p1
p
X
dt
(1) j1 xi j dxi1 d xi j dxi p
j=1
Lecture 17
99
n Z
X
= ( (
i=1
j=1
Z
+(
f (tx)t p1 dt) pdxi1 dxi p .
p p+1
k d
p

X
j1
Di f (tx)t dt))
(1) xi j dxi1 d xi j . . . dxi p
n
X
=k (
Di f (x)dxi dxi1 dxi p )
p
i=1
n Z i

X
p
Di f (tx)t dt) xi dxi1 dxi p
(
i=1
p

X
(1) j1 xi j dxi dxi1 d xi j dxi p .

j=1
() Di denotes partial derivation with respect to the i - th variable concerned.

Thus d p k p1 + k p d p+1
110
Z 1
=(
f (tx)p t p1 dt)dxi1 . . . dxi p
0
n Z
X
i=1
Di f (tx)t p dt xi dxi1 . . . dxi p
= f (tx)t p
+
1
n Z 1
X
i=1
n
1X
Di f (tx) t p dt xi dxi1 . . . dxi p
= f (x)dxi1 . . . dxi p
= .
i=1

Di f (tx) xi t p dt dxi1 . . . dxi p
Lecture 18
Let s p be a fixed p-simplex in Euclidean p-space R p , with vertices 111
a0 , a1 , . . . , a p , i.e. s p is the convex set spanned by points a0 , . . . , a p
which are in general position. We may assume that ao is the origin
and a1 , . . . , a p are unit points of a coordinate axes in R p , and that s p1 is
the face opposite a p in s p .
Definition . A differentiable singular p-simplex in a C manifold X is
a C map t : s p X. The image, im t, is called the support of the
singular simplex t. The j-th face j t is the composite map td j : s p1 X
where d j : s p1 s p is the linear map which maps ao , . . . , a p1 into
ao , . . . , a j , . . . , a p .
The support of j t is contained in the support of t.
Definition . A differentiable singular p-cochain in an open set U X
is a real valued function of differentiable p-simplexes with supports in
U; f (t) R if suppt U.
p
The set S U of of all differentiable P cochains in U forms a real

p
vector space. There is a restriction homomorphism VU : S UP S V for
p1
p
V U and a coboundary homomorphism d p : S U S U defined by
p
X
(d f )(t) =
(1) j f ( j t).
p
j=0
The homomorphisms VU and d p commute, and imdo ker d p+1 .

101
Lecture 18
102
In particular, there is a sequence
p
p1 d
d1
0 S oX S 1X S X
112
S X . . .
with d p+1 d p = 0, i.e., im d p ker d p+1 . Let

H p (S X ) = ker d p+1 /imd p .
This vector space is called the pth cohomology vector space of X
based on differentiable singular cochains.
Since a singular 0-simplex may be identified with the point which is
o can be identified with the vector space of all functions
its support, S U
f : U R. The vector space E o (U) of C functions on U is a
o , and the space R of constant functions on U is a subspace
subspace of S U
o
o.
of E (U), i.e., R E o (U) S U

p
The presheaf S U , VU determines a sheaf S p and since VU com0 , there
mutes with d p and with the inclusion homomorphism e : R S U
are induced homomorphisms
dp
d1
0 R
S 0 S p1 S p . . .
Here the constant sheaf R is identified with the sheaf of germs of
constant functions.
There is a homomorphism

p
gU : S U , VU S p ,
where the image of an element of S Up is the section which it determines.
Then d commutes with gU and, in particular, with gX . Then we have the
commutative diagram:
0
/ So
X
/ ...
gX
/ S p1
X
dp
gX
/ (X, S o )
/ ...
/Sp
X
/ ...
gX
/ (X, S p1 )

/ (X, S p )
/ ...
113
The induced homomorphism gX : H p (S X ) (X, S )is an isomorphism for each p.
Lecture 18
103
p
Proof. For each covering U of X, let S U denote the vector space of

all real valued functions of differentiable psimplexes, which are defined for each simplex t with supp t contained in some open set of U .
p1
p
d p mapsS U
into S U
and, if W is a refinement of U , there is the obp
p
vious restriction homomorphism gW U : S U S W commuting with

p
d p . Then S U , gW U is a direct system and it can be proved, using a
method similar to the one used in the proof of Proposition 8 (Lecture

12), that its direct limit is (X, S p ).

The induced homomorphisms gW U : H p (S U U ) H p (S W ) are
isomorphisms. (See Cartan Seminar, 1948-49, Expose 8, 3.). Hence
gU : H p (S U ) H p (X, S ) is an isomorphism and, in particular, taking U as the covering by one open set X,
gX : H p (S X ) H p (X, S )
is an isomorphism, q.e.d.
The sequence
dp
0 R
S S p1 S p . . .
is exact.
Proof. It is sufficient to show that
e
114
p1
0
0 R
SU
SU
dp
S U . . .
is exact for a cofinal system of neighbourhoods U of each point a X.

For this system, take the spherical neighbourhoods U contained in a
coordinate neighbourhood N of a. The result is proved using the conical
homotopy operator. (See Cartan Seminar, 1948-49, Expose 7, 6 ; his
formula should be replaced by
y(0 , . . . . . . p+1 ) = (0 )x (1 /(1 0 ), . . .)
=0
0 , 1,
0 = 1;
where the indefinitely differentiable function is chosen so that 0

()0 ) 1 for 0 0 1, (0) = 1 and (0 ) = 0 for o in some
neighbourhood of 1.)
Lecture 18
104
The sheaf S p is fine.
Proof. Let E G with E closed and G open. Define hU : S U S U by

if supp t G
(hU f )(t) = f (t)

=0
otherwise.
Then hU is a homomorphism commuting with VU and induces a

p
p
homomorphism h : S p S p such that hx : S x S x is the identity if
Thus S p is fine, q.e.d.
x G, and is zero if x X G.

115
Now let hU : E p (U) S U be the homomorphism defined by

Z
t1 ,
(hU )(t) =
sp
where t1 is the inverse image of the form by t. Clearly hU commutes with VU , hence induces a homomorphism h : p S p with
commutativity in
E p (U)
xU
/ p
x
hU
hx
SU
xU
/ S p.
x
Hence there is an induced homomorphism h : (X, p ) (X, S p )

with commutativity in
E p (X)
fX
/ (X, p )
gX
hX
S Xp
hU commutes with d p .
Proof.
(hU d p )t =
Z
sp
t1 (d p )
/ (X, S p ).
Lecture 18
105
=
d p (t1 ) ( sinced p commutes with t1 )
sp
p
X
(1)
j=0
t1 ( by Stokes theorem for s p ),
d j sp
p
X
(1) j (hU ) j t
j=0
p
= (d hU )t.
Thus hX induces a homomorphism hX : H p (E (x)) H p (S X ). Also 116
the homomorphisms h : p S P and hence the induced homomorphisms h : (X, p ) (X, S P ) commute with d p , and thus there
are induced homomorphisms h : H p (X, ) H p (X, S ). There is
commutativity in
H p (E (X))
fX
/ H p (X, )
hX
gX
H p (S X )
/ H p (X, S ).
The homomorphism h is an isomorphism.
Proof. We have the two resolutions

0
>
}
e }}
}}
}
}}
/R
AA
AA
AA
e AA

/ ...
S0
/ p1
dp
/ ...

/ S p1
/ p
/ ...
dp

/Sp
/ ...
of R. The homomorphism h commutes with d p , and commutativity in

the triangle follows from the fact that R o S o and e, h and
e are inclusion homomorphisms. Hence h is the isomorphism of the
uniqueness theorem, q.e.d.
Lecture 18
106
Theorem 2 (de Rham). The homomorphism
hX : H p (E (X)) H p (S X )
is an isomorphism.
Proof. The following diagram is commutative :
H p (E (X))
fX
/ H p (X, )
hX
H p (S X )
gX
/ H p (X, S ).
117
Since fX , gX , and h are isomorphisms, and the above diagram is

commutative, we have hX = g1
X h fX . Therefore, hX is an isomorphism.
Lecture 19
Definition. A double complex K is a system
of A modules (A is a com- 118

mutative ring with unit element K p,q , indexed by pairs (p, q) of integers, together with homomorphisms d1 and d2 with
p,q
p,q
d1 : K p1,q K p,q ,
p+1,q
d1
d2 : K p,q1 K p,q ,
p,q
d1 = 0,
p+1,q p+1,q1
d1
d2
p,q+1 p,q
d2 = 0,
p+1,q p,q
d1
d2 = 0,
d2
(i.e., d1 and d2 are differential operators of bi degree (1, 0) and (0,

1) respectively, which anticommute. Usually we omit the superscripts
attached to d1 and d2 .) We have then the anticommutative diagram:
...

/ K p1, q1
d2
d2
d1
d1
...

/ K p1, q
/ K p,q1

d2

/ K p1, q+1
/ ...
d1
/ K p,q

d2

/ K p,p+1
/ ...
(Each row and column of a double complex forms a (single) complex

with the homomorphisms d2 and d1 respectively.)
Definition. A subcomplex L of K is a system of submodules L p,q K p,q
stable under d1 and d2 ; thus d1 (L p1,q ) L p,q and d2 (L p,q1 ) L p,q .
119
107
Lecture 19
108
If L is a subcomplex of a double complex K, then clearly K/L =

with the homomorphisms induced by d1 and d2 is again a
K p,q /L p,q
double complex.
p,q
p,q
Let Z1 (K) be the kernel of d1 : K p,q K p+1,q and let B1 (K) be
p,q
p,q
the image of d1 : K p1,q K p,q . Since d12 = 0, B1 Z1 K p,q .
p,q
p,q1
Now d1 (Z P1,Q)1 = 0 Z1 and, since d1 d2 (Z1
)

p,q1
p,q1
p,q
p,q
is a
= d2 d1 (Z1
) = 0, d2 (Z1
) Z1 . Thus Z1 (K) = Z1
subcomplex of K.
P1,q
p,q
p,q1
Also d1 (B1
) = 0 B1 and d2 (B1 ) = d2 d1 (K p1,q1 ) =

p,q
p,q
d1 d2 (K p1,q1 ) B1 . Thus B1 (K) = B1 is a subcomplex of

p,q
p,q
p,q
Z1 (K). Let H1 (K) = Z1 (K)/B1 (K) with H1 (K) = H1 (K) = Z1 /B1 .
In the double complex H1 (K), the homomorphism induced by d1 is the
trivial (zero) homomorphism.
...
/ H p1,q1
1
d2
/ H p1,q
1
d2
/ H p1,q+1
1
/ ...
...
/ H p,q1
1
d2
/ H p,q
1
d2
/ H p,q+1
1
/ ...
p,q
p,q
Similarly if Z2 = ker d2 and B2 = im d2 there is a double complex

p,q
p,q p,q
H2 (K) with H2 (K) = H2 (K) = Z2 |B2 . In H2 (K), the homomor-
phism induced by d2 is the trivial homomorphism.
p1,q1
H2
d1
p1,q
H2
d1
d1
p,q1
H2
H p1,q+1
p,q
H2

p,q+1
H2
Lecture 19
109
120
In particular,
there is a double complex
H2 (H1(K)), which we write
p,q
p,q
p,q
p,q p,q
as H12 (K) = H12 (K) , where H12 = Z12 B12 and Z12 = ker d2 :
p,q
p,q+1
p,q1
p,q
H1 H1
; B12 = im d2 : H1
H1 . In the double complex
H12 (K), the induced homomorphisms d1 and d2 are the trivial homomorphisms.
Similarly there is a double complex

p,q
H21 (K) = H21 (K) = H1 (H2 (K)).
p,q
Notations. In terms of the more usual notation, H12 (K) = HII (HI (K))
p,q
q
and H21 (K)) = HIp (K)(HII (K)).

p,q
To the double complex K = K , d1 , d2 we can now associate the

P
K p,q (each
(single) complex K n , d K n being the direct sum K n =
p+q=n
Kn
is an A module) with the differential operator d = d1 + d2 : K n1

(d is a homomorphism and d2 = d12 + d1 d2 + d2 d1 + d22 = 0).
K n.
dn
dn+1
K n1 K n K n+1
Thus im dn ker dn+1 and there are cohomology modules H n (K) =
ker dn+1 /imdn .
Definition . A homomorphism f : K L (of bidegree (r, s)) of double 121
complexes is a system of homomorphisms f : K p,q L p+r,q+s .
Definition. A map f : K L of double complexes is a homomorphism
of bidegree (0, 0), which commutes with d1 and d2 .
Clearly a map f : K L induces homomorphisms
p,q
p,q
p,q
p,q
p,q
p,q
f + : H1 (K) H1 (L), f : H12 (K) H12 (L)

and
f : H21 (K) H21 (L)
Also, f determines obvious homomorphisms f : K n Ln which

commute with d = d1 + d2 and there are induced homomorphisms f :
H n (K) H n (L)
Lecture 19
110
Definition. A sequence
hr
hr+1
Kr1 Kr Kr+1
of homomorphisms of bidegree (0, 0) of double complexes is called exact
if, each pair (p, q), the sequence
p,q
p,q
Kr1 Kr
p,q
Kr1
is exact.
Given an exact sequence of maps of double complexes
j
K
K 0,
0 K
there is an exact cohomology sequence
j
H n (K )
H n (K) H n (K ) H n+1 (K )
122
Proof. The sequences

i
0 K n
Kn
K n 0
are clearly exact for each each n, and d commutes with i and j. Then,
using the standard arguments of Lecture 10, we obtain the result.

Definition. Two maps of double complexes, f : K L and g : K L
are called homotopic ( f g) if there exist homomorphisms h1 : K p+1,q
L p,q and h2 : K p,q+1 L p,q (i.e., h1 and h2 are homomorphisms K L
of bidegree (1, 0) and (0, 1) respectively) such that
d1 h1 + h1 d1 + d2 h2 + h2 d2 = g f,
d1 h2 = h2 d1 ,
d2 h1 = h1 d2 .
(Homotopy of maps is obviously an equivalence relation.)

Homotopic maps f : L and g : K L induce the same homomorphisms
Lecture 19
111
p,q
p,q
p,q
p,q
(i) f = g : H12 (K) H12 (L),

(ii) f = g : H21 (K) H21 (L),
(iii) f = g : H n (K) H n (L).
Proof. Let (h1 , h2 ) be the pair of homomorphisms K L which express
the homotopy between f and g. Since h2 , like d2 , anticommutes with
d1 , there are induced homomorphisms
p,q+1
h+2 : H1
p,q
(K) H1 (L).
Further, since
123
d1 h1 + h1 d1 = g f d2 h2 h2 d2 ,
h1 expresses the homotopy of g and f + d2 h2 + h2 d2 from a column
complex of K to the corresponding column complex of L.

Hence
p,q
p,q
f + + d2 h+2 + h+2 d2 = g+ : H1 (K) H1 (L),

d2 h+2 + h+2 d2 = g+ f + .
i.e.,
Thus h+2 expresses the homotopy of g+ and f + from a row complex

p,q
p,q
of H1 (K) to the corresponding row complex of H1 (L) .

Hence
p,q
p,q
f = g : H12 (K) H12 (L). This proves (i).

The proof of (ii) is carried out in a similar manner, using the other
anti-commutativity d2 h1 = h1 d2 .
To prove (iii), let h = h1 + h2 : K n+1 Ln . Then
dh + hd = (d1 + d2 )(h1 + h2 ) + (h1 + h2 )(d1 + d2 )
= (d1 h1 + h1 d1 + d2 h2 + h2 d2 ) + (d1 h2 + h2 d1 ) + (d2 h1 + h1 d2 )
= g f.

Thus h is a homotopy of the complexes K n and Ln , and we obtain
f = g : H n (K) H n (L).
112
Lecture 19
Note . In the double complexes which occur in the usual applications,

one has commutativity d1 d2 = d2 d1 , d1 h2 = h2 d1 and d2 h1 = h1 d2 124
rather than anti-commutativity. The commutative case can be transformed into the anti-commutative case and vice versa by replacing d2
by (1)P d2 : K p,q1 K p,q and h2 by (1)P h2 : K p,q1 K p,q . These
substitutions do not change ker d2 , im d2 , etc., and so the cohomology
p,q
p,q
modules H1 , H12 , etc., remain unchanged. But if K is a commutative
double complex, K n , d, H n (K) are understood to refer to the associated
anticommutative double complex.
Lecture 20
. If is a directed set, a direct system of double complexes 125
Definition
K ,
is a system of double complexes K and maps
,
: K K
( < ),
such that
(i) is the identity,
(ii) is homotopic to for < < .

If K , is a direct system of double complexes, there are unique-
ly determined direct limits:

p,q
p,q
(i) H12 (K) = direct limit {H12 (K ), },

p,q
p,q
(ii) H21 (K) = direct limit{H21 (K ), },

(iii) H n (K) = direct limit{H n (K ), }.

p,q
Proof. (i) The system H12 (K ), is a direct system, as

(a) is the identity, since is the identity,
(b) = ( ) = , since homotopic maps induce the same
homomorphism on the cohomology groups.

113
Lecture 20
114
The proofs of (ii) and (iii) are carried out in a similar manner.
126
Definition . We say that a system {K } of double complexes is bounded

p,q
on the right if there is some integer m, such that K = 0 whenever
q > m (for all p and ). The system is said to be bounded on the left
p,q
(resp. above, below) if there is an integer m such that K = 0 whenever
q < m (resp. p < m, p > m).

Proposition 15. If K , is a direct system of double complexes
p,q
which is bounded above or on the right and if H12 (K) = 0 for all p
and q, then H n (K) = 0 for all n.
Proof. Let H n (K), let be its representative in some H n (K ) and
p,q
p1,q+1
+ . . .,
let a Z n (K ) represent the class . Let a = a + a
m,nm
nm,m
where p + q = n and the sum terminates with a
(resp. a
).
Since a Z n (K ), da = (d1 + d2 )a = 0, i.e.,
p,q
p,q
p1,q+1
da = d1 a + (d2 a + d1 a
) + ,
p,q
p,q
p1,q+1
and the sum being direct, we have d1 a = 0 and d2 a +d1 a

= 0.
p,q
p,q
p,q
p,q+1
p,q
Thus a Z1 (K ) and d2 a B1 (K ). Therefore, a represents
p,q
p,q
an element of Z12
(K ). Since H12
= 0, there is some > such
p,q
p,q
p,q
that a = a represents an element of B12 (K ). Thus, for some
p,q1
p,q
p,q
p,q
b Z1
(K ), a d2 b B1 (K ), and hence a = d2 b + d1 c for
p1,q
.
some c K
Let a = a and let
e = a d(b + c)
= a d1 b d2 b d1 c d2 c
p,q
= a a d2 c
p1,q+1
= (a
=
127
p1,q+1
where e
ep1,q+1
p1,q+1
= a
( since d1 b = 0)
p2,q+1
d2 c) + a
ap2,q+2
d2 c. Then since e and a represent the same
Lecture 20
115
class H n (K ), they represent the same element H n (K). The

reason for choosing this representative e is the fact that its (p, q) th
component is zero. Continuing thus, after a finite number k of steps for
a suitable , is represented by e = d2 c , consisting of a single term
pk,q+k
in K
. Continuing the construction still further, since the system of
double complexes is bounded above or to the right, after a finite number
of steps, we obtain a representative e = d2 c = 0, i.e., is represented
by 0 Z n (K ) for a suitable . Hence = 0.

Proposition 15-a. If K , is a direct system of double complexes
p,q
which is bounded below or on the left, and if H21 (K) = 0 for all p and
q then H n (K) = 0 for all n.
Proof. This is carried out exactly as in Proposition 15, except that we

eliminate the component of a with highest second degree q instead of
the one with highest first degree, and d1 c plays the role of d2 c.

Proposition 16. If K , is a direct system of double complexes
p,q
which is bounded above or on the right and if H12 (K) = 0 except (at
o,q
most) for p = 0, then there exist isomorphisms : H q (K) H12 (K) for
all q.

If another direct system Kl ambda, is bounded above or on the

p,q
right with H12 (K ) = 0 except for p = 0 and if h : K K are maps

with h = h , then there are induced homomorphisms
128
o,q
o,q
h : H q (K ) H q (K) and h : H12 (K ) H12 (K)

commuting with the isomorphisms .
H q (K )
/ H 0,q (K )
12
H q (K)
/ H 0,q (K).
12
Lecture 20
116
Proof. Let L be the subcomplex
L :
...

/ K 1,q1

/ K 1,q
/ ...
...

/ Z 0,q1
1

/ Z 0,q
1
/ ...
0
p,q
p,q
p,q
o,q
of K , with L = K for p < 0; L = 0 for p > 0 and L =

o,q
Z1 (K ). Since L is stable under d1 and d2 , it is a subcomplex of K .
Let M be the subcomplex
M :
...

/ K 1,q1

/ K 1,q
/ ...
...

/ B0,q1
1

/ B0,q
1
/ ...
0
129
p,q
p,q
p,q
0
p,q
p,q
of L , with M = L = K for p < 0; M = L = 0 for p > 0

o,q
o,q
and M = B (K ). Since M is stable d1 and d2 , it is a subcomplex
of L .

Since h : K K commutes with d1 and d2 , we have h : L
L and h : M M . Thus there are induced maps h : K /L
K /L and h : L /M L /M which commute with i and j in the
Lecture 20
117
exact sequences
i
/ L
/ K
/ K /L

/0
/ L
/ K

/ K /L
/ M
/ L
/ L /M
/ 0,
and
0
/0
j
/ L /M
/ 0.
/ M i / M
0
Hence h commutes with d , i and j in the exact cohomology sequences:
...
/ H n1 (K /L )
h
...
/ H n (L )
h

/ H n1 (K /L )
/ H n (K /L )

/ H n (L )
/ H n (K )

/ H n (K )

/ H n (K /L )
and
...
/ H n (M )
h
...
/ H n (L )
/ H n (L /M )
/ H n+1 (M )
/ H n (M )
/ H n (L )
/ ...
/ H n (L /M )

/ H n+1 (M )
/ ...
In the direct limit, we have the following commutative diagram 130

where each row is exact.
(A1 )
H n1 (K /L )
(A2 )
H n1 (K/L)
H n (L)
H n (K /L )
H n (K )
H n (L )
...
H n (K)
H n (K/L)
...
and
(B1 )
H n (M )
H n (L )
(B2 )
H n (L /M )
H n (M)
H n (L)
H n+1 (M )
...
H n (L/M)
H n+1 (M)
...
Lecture 20
118
The quotient double complex K /L is
K /L :
...

/ K 0,q1 /Z 0,q1

/ K 0,q /Z 0,q
1
/ ...
...

/ K 1,q1

/ K 1,q
/ ...

o,q
o,q
1,q
and since the sequence 0 K /Z1

H1 (K /L ):
K is exact, we have
...

/ H 1,q1 (K )

/ H 1,q (K )
/ ...
...

/ H 2,q1 (K )

/ H 2,q (K )
/ ...
131
p,q
p,q
Thus H12 (K /L ) = 0 for p 0, and is equal to H12 (K ) for p > 0,

p,q
p,q
hence H12 (K/L) = direct limit H12 (K /L ) = 0 for p 0, and
by hypothesis is also zero for p > 0, hence is zero for all pairs (p, q).
Since K /L is bounded above or to the right, by Proposition 15, we
have H n (K/L) = 0 for all n. Thus, in the sequence (A2 ), we see that
i : H n (L) H n (K) is an isomorphism.
Lecture 20
119
1,q
Again, since the sequence K
o,q
B1 0 is exact, we have for
H1 (M ) :

/ H 1,q1 (K )
...

/ H 1,q (K )
/ ...
p,q
p,q
Thus, H12 (M ) = 0 for p > 0 and is equal to H12 (K ) for p < 0,

p,q
p,q
hence H12 (M) = direct limit H12 (M ) is equal to zero for p 0,
and by hypothesis, is also zero for p < 0, hence is zero for all pairs
(p, q). As before, the conditions of Proposition 15 being satisfied, we 132
have H n (M) = 0 for all n. Thus, in the sequence (B2 ), we see that
j : H n (L) H n (L/M) is an isomorphism.
The quotient double complex L /M is given by
L /M :
...

/ H 0,q1 (K )

/ H 0,q (K )
o,q
/ ...
o,q1
o,q
Thus (L /M )q = H1 (K ) and d = d2 : H1 (K ) H1 (K ).
o,q
o,q
Hence H q (L /M ) = H12 (K ); similarly H q (L /M ) = H1 (K ). Furo,q
thermore, in the limit we have H q (L/M) = H12 (K) and H q (L /M ) =
o,q
H1 (K ).
From the sequences (A1 ), (A2 ), (B1 ), (B2 ), we have the commutative
Lecture 20
120
diagram :
H q (K ) o
H q (L )
/ H q (L /M )
H q (K) o
0,q
H12 (K )
H q (L)

/ H q (L/M)
0,q
H12 (K).
Then there is an isomorphism

o,q
: H q (K) H12 (K),

o,q
133
where = j (i )1 : H q (K) H q (L) H q (L/M) = H12 (K),

being an isomorphism since we have proved that each of i and j is an
isomorphism.
Further, form (I), we have obviously commutativity in the following
diagram :
H q (K )
/ H 0,q (K )
12
h
H q (K)
/ H 0,q (K).
12

Proposition 16-a. If K , is a direct system of double complexes
p,q
which is bounded below or on the left and if H21 (K) = 0 except for
p,o
q = 0, then there exist isomorphisms : H p (K) H21 (K) for all p.

If another direct system K , is bounded below or on the left
p,q
with H21 (K ) = 0 except for q = 0, and if h : K K are maps with

h = h , then there are induced homomorphisms
p,o
p,o
h : H p (K ) H p (K) and h : H21 (K ) H21 (K)

which commute with .
Remark . In particular, all the propositions
proved in this lecture are

p,q
satisfying the conditions stated
true for a double complex K = K
in the propositions. We have only to replace the by the identity map

K K.
Lecture 20
121
Example 29. Let
134
K p,q = Z (ring of integers) if q 0 and p = q or q 1,

= 0 otherwise.
For q 0, let d1 : K q1,q K q,q and
d2 : K q1,q K q1,q+1
be the identity isomorphisms of Z onto itself. The other homomorphisms are all the trivial homomorphisms. Then K = K p,q is a double
complex with
p,q
p,q
H21 (K) = H2 (K) = Z if (p, q) = (0, 0),

= 0 otherwise ;
p,q
H12 (K)
p,q
H1 (K)
= 0 for all (p, q),
and
H n (K) = Z if n = 0,
= o otherwise .
This double complex is bounded below and on the left, but is unbounded above and on the right.
Lecture 21
Introduction of the family
135
Let be a family of paracompact normal closed subsets of a topological

space X such that
(1) if F , then every closed subset of F is in ,
(2) if F1 , F2 , then F1 F2 ,
(3) if F , there is an open U with F U and U .
For example, if X is paracompact and normal, can be taken to
the family of all closed subsets of X, and, if X is locally compact and
Hausdorff, then can be taken to be the family of all compact sets of X.
Sections with supports in the family . If S is a sheaf of
A mod-
ules, the set of all sections f (X, S ) such that supp f = x : f (x) ,

0x is in , forms an A module (if supp f1 and supp f2 then
supp ( f1 f2 ) (supp f1 supp f2 ) is in ), a submodule of (X, S ),

which we denote by (X, S ).
Any homomorphism h : S 1 S of two sheaves of A modules
induces a homomorphism
h : (X,
S ) (X,
S ),
since a homomorphism of sheaves decreases supports (i.e., supp h f

supp f ).
123
Lecture 21
124
Definition . A - covering of X is a locally finite proper covering U 136

S
such that, if X < , there is a special open set U U with
U
UU (U )
Remark. If X < , U is unique, and is not the empty set, for otherwise,
in each case, X would belong to . If X , a - covering is just a
locally finite proper covering of X.
The - coverings of X form a subdirected set of the directed set
of all locally finite proper coverings of the space X.
Proof. (i) If X , is the directed set of all locally finite proper
coverings of X.

(ii) If X < and U , W are any two - coverings of X, let W = W :

W = U V for some U U and some V W with W = U V .
Then W is a locally finite proper covering of X and
[
[
[
(
=(
V)
U)
W
WW (W )
UU (U )
VW (V )
is in , since each set contained in brackets is in . Thus W is a

-covering which is a common refinement of U and W .

Remark. If U and W are two coverings with special sets U and V
then U V is not empty, and if W is refinement of U , then V U
and V is not contained in any other U U . In particular, if W is
equivalent to U , then V = U .
137
Cohomology groups with supports in the family . If U is a P

covering and a presheaf of A- modules, we define
X
X
p
C U ,
= Cp U ,
for p > 0,
X
X

p
for p = 0,
and
C U ,
Co U ,
Lecture 21
125
P
P
p
whereC (U , ) is the submodule of C o (U , ) consisting of those zero
cochains which assign to U the zero of S U . Then we have a mapping
X
X
p : Cp1 U ,
Cp U ,
.
Let
138
X
p
H U ,
= ker p+1 / im p .
P
p
Then H (U , ) is called the p th cohomology module of the covP
ering U with coefficients in the presheaf and supports in the family
.
If a - covering W is a refinement of U , for each choice of the
function : W U , (V ) = U . We then have the mapping (Lecture
9)
X
X
p
p
(p 0).
C W ,
+ : C U ,
+ induces the homomorphism

X
X
W U : Hp U ,
Cp W ,
with U U = identity, and W W W U = W U if U < W < W . Thus

P
{Hp (U , ) , W U } is a direct system of A-modules.
Let

P
P
P
p
p
p
H (X, ) = direct limit H (U , ) , W U
. H (X, ) is
U ,W
called the p-th cohomology module of the space X with coefficients in

P
the presheaf and supports in the family .
The result analogous to Proposition 7 (Lecture 11) is true in this
case.
Proposition 7-a If S is a sheaf of A-modules, Ho (X, S ) = (X, S ).
Proof. We can identify Ho (U , S ) = Zo (U , S )(see Lecture 11) with
the submodule of (X, S ) consisting of the sections f with supp f
S
X U
U ; then supp f . If W is a refinement of U ,
US (U )
since V U , we have X U X V , hence
W U : Ho (U , S ) Ho (W , S )
Lecture 21
126
is the inclusion homomorphism. Thus Ho (X, S ) can be identified with

a submodule of (X, S ) i.e., Ho (X, S ) (X, S ).
If f (X, S ), let U = X supp f , and let U be an open set

containing supp f with U . Clearly U = U, U is a -covering
(with special set U ) and the cochain g defined by
g(U) = f |U; g(U ) = 0
is the cocycle in Zo (U , S ) = Ho (U , S ) which is identified with f
(X, S ). Thus (X, S ) Ho (X, S ), hence Ho (X, S ) = (X, S ).

139
Given a sequence of homomorphisms of presheaves

q1
X
dq
q
X
dq+1
q+1
X
with im dq ker dq+1 (i.e., d2 = 0) for each -covering U the system

P
p,q
p
{kU = C U , q with the homomorphisms
q
q
X
X
p
p
: C (U , ) C U ,
and
p
:C U
q
q1
X
p
, C U ,
forms a double complex denoted by KU = C U ,
Pq
P d P
of two presheaves induces a hoProof. Any homomorphism

P
P d p
p
momorphism of C U ,
C (U , ) commuting with the coboundary operator . We have 2 = 0, and by hypothesis d2 = 0. Further d = d; so we have the commutative case of a double complex,
q.e.d.

For each pair U , W of - coverings for which W is a refinement of
U choose : W U with V (V); if W = U , let : U U be
the identity and let
q
q
X
X
p
p
W U = + : C (U , ) C W , .
Lecture 21
127

X
X
, W U
C ( ) = C U ,
U ,W
is a direct system of double complexes.

(If U < W , W U = + , for an arbitrary but fixed choice of :
W U if W , U , and : U U is the identity.)
140
P
Proof. Since and d commute with W U , W U : C (W , )
P
C (U , ) is a map of double complexes, and by construction U U
is the identity.

If W is a - refinement of W , W U W U corresponds to a possible
choice of : S U . Since for all possible choices of , (W ) = U ,
the homotopy operator k (see Lecture 9)
q
q
X
p
p1
W ,
k : C U , C
1 U , Pq = C 1 U , Pq into C o W , Pq . Thus we have two

maps C
P
P
homomorphisms k : Cp U , q Cp1 W , q and the trivial ho

P
P
p
p
momorphism C U , q C W , q1 such that
k + k = W W W U W U ,
and further, d commutes withk. Hence W U W U is homotopic to

P
W U , hence C (U , ), W U is a direct system.
Lecture 22
P
Proposition 8-a. If
is a presheaf which determines the zero sheaf, 141
P
p
then H (X, ) = 0 for all p 0.
S
P
p
Proof. Let f C (U , ). Then
U , and hence has
UU (U )
H . Shrink the covering

neighbourhoods
G, H with G H; G,

U = U (U ), U H of H to a covering W = WU UU with
U U. For each x H choose a neighbourhood V x of x such that
W
a) if x U, V x U H,
b) if x WU , V x WU H,
U , V x WU = ,
c) if x < W
d) if x Uo . . . U p = U , Vx U f () = 0,
S
and let V = X
U.
UU (U )

Then V , V x
is a refinement of U . Choose : H () U
xH
W(x) and
+
()
= U . Then it can be verified that f = 0.
The covering V x G
of G has a locally finite refinement {Yi }iI .
such that x
xH
Let U1 be the proper covering consisting of V together with all V such

that V = Yi G for some
i I. Then W1 is a - covering which is a

. Hence there is a function 1 : W1 U with

P
p
V 1 (V) and such that +1 f = 0. Therefore H (X, ) = 0, q.e.d.
refinement of V , V x
xH
129
Lecture 22
130
142
Let
dq
dq+1
S q1 S q S q+1
be a sequence of homomorphisms of sheaves of A-modules with im dq
ker dq+1 . Let Bq = im dq , zq = ker dq+1 and H q = zq /Bq .
There is an induced sequence of homomorphisms of presheaves
dq+1
dq
S q1 S q S q+1
q
with im dq = Bo Bq and ker dq+1 = zq . Also there is an induced
sequence of homomorphisms

dq p

dd+1 p
p
C U , S q1 C U , S q C U , S q+1

p
p
p
q and
where C (U , S q ) = C U , S q , with im dq = C U , B
p,q
p
p
p
ker dq+1 = C (U , zq ). Then H2 C (U , S ) = C (U , zq ) /C

qo .
U ,B
Let : zq H q be the natural homomorphism. There is an in qo ) = 0.
duced homomorphism : zq H q with (B
Hence there is an induced homomorphism
p
: C (U , zq ) C (U , H q ),
p
q )) = 0.
which commutes with and W U such that (C (U , B
Hence there are induced homomorphisms :
p,q
: H2 C (U , S ) C (U , H q ),
p,q
: H21 C (U , S ) Hp (U , H q ),
p,q
: H21 C (S ) H (X, H q ).
143
p,q
The homomorphism : H21 C (S ) C (X, H q ) is an isomorphism.

Proof. The exact sequence
q
o 0
0 o zq zq /B
Lecture 22
131
gives rise to an exact sequence

p
q
p
p
qo ) 0.
0 C (U , B o ) C (U , zq ) C (U , zq /B
Hence the induced homomorphism

p,q
p
p
qo ) C p (U , zq /B
qo )
H21 C (U , S ) = C (U , zq )/C (U , B
is an isomorphism. Therefore
p
p,q
H21 C (U , S ) H (U , zq /B o )
and hence the homomorphisms
p,q
p
qo )
H21 C (S ) H (X, zq /B
(1)
are isomorphisms.
The exact sequence

q
q B
q /B
o 0
o B
0B
j
i
qo )
q ) C p (U , B
q /B
qo ) 0
0 Cp (U , B
Cp (U , B
and i, j commute with . Hence there is an exact cohomology sequence 144
i
p1
q /B
qo ) H p (U , B
qo )
H (U , B
i
p
p
q /B
qo ) .
H (U , Bq )
H (U , B
Since i, j and commute with W U , there is an exact cohomology

sequence of the direct limits
q
q /B
o ) H p1 (X, B o )
Hp1 (X, B
p
q ) H p (X, B
q /B
qo )
H (X, B
Lecture 22
132
q /B
qo determines the 0-sheaf and hence H p (X,
The presheaf B
q /B
qo ) = o for all p. Hence, by exactness,
B
qo ) H p (X, B
qo )
i : Hp (X, B
is an isomorphism.
From the exact sequences of homomorphisms
0
/B
q
0
/ zq
/ zq /B
q
0
/0

/B
q

/ zq

/ zq /B
q
/ 0,
one obtains exact sequences of homomorphisms

q)
Hp (X, B
0
Hp (X, zq )
Hp (X, zq /B q0 )
p+1
Hp+1 (X, zq )
Hp+1 (X, zq )
)
H (X, B
0
q)
Hp+1 (X, B

q)
H (X, B
H (X, zq )

H (X, zq /B q )
where four of the vertical homomorphisms are isomorphisms. Hence,

by the five lemma (see Eilenberg-Steenrod, Foundations of Algebraic
Topology, p. 16), the homomorphism
p
qo ) H p (X, zq B
q)
H (X, zB
(2)
145
is an isomorphism.
Next, the exact sequence
0 Bq zq H q 0
q zq H q .
0B
q
Let Ho be the image of zq in H q . Then there is an exact sequence

q
0 zq /B q H q H q /Ho 0
Lecture 22
133
q
p
and the presheaf H q /Ho determines the 0-sheaf. Hence H (X,
q
H q /Ho ) = 0, and in the exact cohomology sequence, the homomorphism
p
p
H (X, H q /B q ) H (X, H q )
(3)
is an isomorphism.
Then is the composite isomorphism
p,q
o ) H (X, zq /B
q ) H (X, H q ).
H21 C (S ) H (X, zq /B
Definition. We say that the degrees of {S q } are bounded below if there

is an integer m such that S q = 0 for q < m.
Lecture 23
Definition. The - dimension of a space X, dim X, is sup dim F.
146
dim X n if and only if every - covering has a - refinement

of order n.
Proof. Necessity. Let dim X n and let U be a -covering of
S
X. Let G be a neighbourhood of
U with G . Then U =
UU (U )

U G forms a locally finite covering of G.
Since G is
(U (U )) G,
normal and dim G n, the covering U (see Lecture 11) has a (locally
finite) proper refinement W of order n. Let V be the union of X G
and those elements W which are contained in U G together with V
form a - covering W of order n which is a - refinement of U .
Sufficiency. Let F and let U be a finite proper covering of F. Let
G be an open set with F G and G . Extend each U U to an open
set V of G with V F = U. These sets together with V = X F form a
- covering W of X. Then W has a refinement W of order n and
{W F}WW is a refinement of U of order n. Thus dim F n, and
hence dim X n.
Note. The paracompactness of the sets of was not used in this proof.
Example . In Example M (see C. H. Dowker, Quart. Jour. Math 6 147
(1955), p. 115) let be the family of all paracompactsets of M (the
space is also denoted by M). Then dim M = 0 and dim M = 1.
Remark. It is always true that dim X dim x.
135
Lecture 23
136
Proposition 17. Let
dq+1
dq
S q1 S q S q+1
be a sequence of homomorphisms of sheaves of A modules with im dq =
ker dq+1 for q , 0 and im d0 ker d1 and let G = ker d1 / im d0 . If
dim X is finite or if the degrees of {S q } are bounded below there is
an isomorphism : H pC (S ) Hp (X, G ).
If
S
q1
dq
dq+1
q+1
is another such sequence (with G isomorphic to G and identified with

G ) and if h : S q S q are homomorphisms commuting with dq such
that the induced homomorphism h : G G is the identity, then there
are induced homomorphisms h : H PC (S ) H PC (S ) with commutativity in
H p C (S )
/ H p (X, G )
h = identity
H pC (S )

/ H p (X, G ).
Proof. If the degrees of {S q } are bounded below, then Cp (U , S q ) = 0

for q < n for some n, and so the system C (U , S ), W U
is
U ,W
148
bounded on the left. If dim X m, then there is a cofinal directed set

, consisting of coverings of order m. If U , CP (U , S q ) =

0 for p > m; thus the system C (U , S ), W U
is bounded
U ,W
below.
If q , 0, we have H q = 0, and by a result in Lecture 22, we have

p,q
Hp (, H ) = 0, hence H21 C (S ) = 0. Therefore by
Proposition 16-a, there is an isomorphism
p,q
H21 C (S )
p,o
: H pC (S ) H21 C (S ).
Lecture 23
137
Also for q = 0, there is an isomorphism (see Lecture 22)

p,o
: H21 C (S ) H (x, G ).
Let be the composite isomorphism
= : H PC (S ) HP (X, G ).
Next, the homomorphisms h : S q S q induce homomorphisms
h : CP (U , S q ) CP (U , S q ) which commute with d, and W U ,
and hence give rise to maps h : C (U , S ) C (U , S ) which
commute with W U . Therefore, there are induced homomorphisms h
which commute with ,
H pC (S )
/ H p,0 C (S )
21
H p C (S )
/ H p,0C (S ).
21
Since h commutes with dq , h maps zq into zq and Bq into Bq , hence 149

induces a homomorphism h : H q H q , and there is commutativity
in
p
C (U , zq )
/ C p (U , H
q )
C (U , zq )
/ C p (U , H q ),
where is the homomorphism induced by the natural homomorphisms

zq H q and zq H q . Therefore there is commutativity in
p,q
H21 C (S )
/ H p (X, H
h
p,q
q )
H21 C (S )
/ H p (X, H q ).
Lecture 23
138
Therefore, taking q = 0, we see that h commutes with = .

H p C (S )
/ H p,0C (S )
21
/ H p (X, G )
h = identity
H pC (S )
/ H p,0C (S )
21
/ H p (X, G ).
Lecture 24
Every covering is shrinkable as is shown by the following result.
150
Let {Ui }iI be a locally finite covering of space X with some i I
such that U i is normal for i I (i ). Then there is a refinement {Vi }iI
with V i Ui .
Proof. The union of the locally finite system {U i }iI(i ) of normal closed
S
S
Ui with X Ui
Ui
sets is normal and closed and X Ui
i,i
i,i
S
Ui open. Hence there are open sets G, H with XUi G,
closed and
i,i
S
then V i X G Ui .
Ui . Let Vi = X G,

G H, H
i,i
Since {H Ui }iI(i ) is a covering of H and the closed subset H of

S
Ui is normal, there is a covering {Pi }iI(i ) of H with P i H Ui .
i,i
S
Vi = H.
Let Vi = H Pi for i I (i ). Then Vi is open, V i Ui and
i,i
Then {Vi }iI is a covering of X and V i Ui for all i I q.e.d.

If S is a fine sheaf and if C U X with C closed, U open and U
normal, then the restriction of S to C is fine.
Proof. This result is proved in the same way as in the case (see Lecture
15) that X is normal except that the open set H is to be replaced by its
intersection with U if necessary, so that H U.

Proposition 11-a If {Ui }iI is a locally finite covering of a space X with 151
some i I such that U i is normal for i I (i ), and if S is a sheaf
whose restriction to each closed subset C of each Ui , i I (i ) is fine
139
Lecture 24
140
(in particular this is true if S is fine), then there is a system {li }iI of
homomorphisms li : S S such that
(i) for each i I there is a closed set Ei Ui such that 1i (S x ) = 0x if
x < Ei ,
P
li = 1.
(ii)
iI
i Vi , V i Gi , G i Ui ,
Proof. Shrink to a covering {Wi }iI with W
where Wi , Vi and Gi , are open. Using the fineness of the restriction of
S to G i . one constructs homomorphisms hi : S S i , i , (actually
the homomorphisms are hi : SG i SG i , and we extend these by zero
outside G i ; SG i denotes the restriction of S to G i ) with
hi (s) = s
i,
if (s) W
= 0(s) if (s) X V i .
Let the set I (i ) be well-ordered and define
Y

li =
(1 h j ) hi (i , i ),
j<i
li =
(1 h j ).
{ j I(i )}
152
(In a neighbourhood of each point of X, li , i I, is only a finite

product.) Then li : S S is a homomorphism. Let Ei = V i for
i , i ; then if (s) X V i , we have li (s) = 0(s) since hi (s) = 0(s) . Let
S
Ei = X
Wi ; then Ei Wi Ui . If (s) X Ei , then, for
iI(i )
some i I (i ), (s) Wi and hence hi (s) = s; so li (s) = 0(s) .
If (s) = x, choose a neighbourhood N x of x meeting at most a finite

number of the sets Ui , i I (i ), say, for i = i1 , . . . , iq with i1 < < iq .
Then
X
li (s) = hi1 (s) + (1 hi1 )hi2 (s)
iI
+ (1 hi1 ) . . . (1 hiq1 )hi1 (s)
Lecture 24
141
+ (1 hi1 ) . . . (1 hiq ) (s)
= s,
and this completes the proof.

Let
dq+1
dq
S q1 S q S q+1
be a sequence of homomorphisms of sheaves with dq+1 dq = 0. Such a
sequence of sheaves is called a complex of sheaves.
Definition. A complex of sheaves {S q } is called homotopically fine, if,
for each locally finite covering {Ui }iI with some i I such that U i is
normal for i I (i ), there exist homomorphisms hq1 : S q S q1 153
q
and a family {li }iI of endomorphisms of S q such that
q
(i) for each i I there is a closed set Ei Ui such that li (S x ) = 0x if

q
x < Ei ,
P q
li = 1 + dq hq1 + hq dq+1 .
(ii)
iI
If each S q is fine, then the sequence {S q } is homotopically fine.
Proof. Taking hq = 0, this result follows immediately from Proposition

11-a.

If the sequence {S q } is homotopically fine, and U is a locally finite covering satisfying the conditions of the previous definition, then
p,q
H12 C (U , S ) = 0 for all p > 0. (This result is trivially true for p < 0)
Proof. As in the proof of Proposition 12, there are induced homomorq
q
phisms li (U) : (U, S q ) (U, S q ) induced by li , and homomorphisms
p
p1
k p1 : C (U , S q ) C (U , S q ) (p > 0)
such that
p k p1 f () + k p p+1 f () =
li (U ) f ()
iI
= f () + dq hq1 f () + hq dq+1 f ()
Lecture 24
142
Thus k f + k f = f + dh f + hd f . and hence
p
154
p
k + k = 1 + dh + hd : C (U , S q ) C (U , S q ).
Since d and h commute with , there are induced homomorphisms
dq : Hp (U , S q1 ) Hp (U , S q ),
q
h+ : H (U , S q+1 ) H (U , S q ).
p,q
Now, H1 C (U , S ) = Hp (U , S q ) and, from the homotopy k we

have
p,q
p,q
dh+ + h+ d = 0 1 : H1 C (U , S ) H1 C (U , S )
p,q
is homotopic to zero and hence H12 C (U , S ) = 0.
Lecture 25
Proposition 18. If the complex of sheaves {S q } is homotopically fine, 155
there is an isomorphism : H q C (S ) H q (X, S ).
If the complex of sheaves {S q } is also homotopically fine, and h :
q
S S q are homomorphisms commuting with dq , then h commutes
with .
H q C (S )
/ H q (X, S )
H qC (S )
/ H q (X, S ).
p
Proof. The system C (U , S ) = {(C (U , S q ))} of double complexes

p,q
is bounded above by p = 0. Since (see Lecture 24) H12 C (S ) = 0 for
p,q
p > 0, and H12 C (S ) = 0 trivially for p < 0, by Proposition 16, there
is an isomorphism
o,q
: H q C (S ) H12 C (S ).
Since h : Cp (U , S q ) Cp (U , S q ) commutes with d, and
W U , h : C (U , S ) C (U , S ) is a map of double complexes
which commutes with W U . Therefore h commutes with .

Since (X, S q ) = Ho (X, S q ) = dir lim Ho (U , S q ) and the homomorphism U : Ho (U , S q ) (X, S q ) commutes with dq , there
are induced homomorphisms
U : H q Ho (U , S ) H q ( (X, S ),
: dir lim H q Ho (U , S ) H q (X, S ).
143
Lecture 25
144
156
Since the operation of forming cohomology groups commutes with

the operation of forming direct limits, (see Cartan - Eilenberg, Homological Algebra, Proposition 9.3 , p. 100), is an isomorphism, and
since h commutes with U , h commutes with .
Now,
o,q
H1 C (U , S ) = Ho (U , S q ),
o,q
H12 C (U , S ) = H q Ho (U , S ),
o,q
H12 C (S ) = dir lim H q Ho (U , S ).

Thus we have an isomorphism
o,q
: H12 C (S ) H q (X, S )
which commutes with h . Let = be the composite isomorphism
then commutes with h .
H qC (S )
/ H o,qC (S )
12
H qC (S )
/ H q (X, S )
/ H 0,q C (S )
12

/ H q (X, S ).
Theorem 3 (Uniqueness Theorem). Let

dq
dq+1
S q1 S q S q+1
157
be a homotopically fine complex of sheaves of A modules with im dq =

ker dq+1 for q , 0 and im do ker d1 and let H o = ker d1 / im do . Let
{S q } be another such complex with H o isomorphic to H o .
If dim X is finite or the degrees of {S q } and {S q } are bounded
below, then any isomorphism : H o H o induces an isomorphism
: H q (X, S ) H q (X, S )
and if {S q } is another such complex and : H o H o is an
isomorphism, then
= : H q (X, S ) H q (X, S ).
Lecture 25
145
If h : S q S q are homomorphisms (for each q) commuting with

and if the induced homomorphism h : H o H o is an isomorphism, then the homomorphism h : H q (X, S ) H q (X, S is the
isomorphism h .
dq ,
Proof. Since the hypotheses of Propositions 17 and 18 are satisfied,

there are isomorphisms
: H q C (S ) Hq (X, H o ),
: H qC (S ) H q (X, S ).
o
Since : H
H o is an isomorphism, so is
q
: H (X, H o ) H (X, H o ).
Let be the isomorphism 1 1 . Since () = , = 158
.
H q (X, S ) o
H qC (S )
/ H q (X, H
H q (X, S ) o
0 )
H qC (S )
/ H q (X, H 0 )
H q (X, S ) o
H qC (S )
/ H q (X, H
0 ).
If h : S q S q are homomorphisms commuting with dq and

inducing an isomorphism h : H o H o , then it follows from the
commutative diagram:
H q (X, S ) o
H q C (S )
h
h h
H q (X, S ) o
/ H q (X, H
H qC (S )
that h = h : H q (X, S ) H q (X, S ).
0 )
/ H q (X, H 0 ),
Lecture 26
Singular chains
159
Definition. Let T p be the set of all singular p- simplexes in X. A singular

p-chain with integer coefficients is a function c : T p Z such that
ct = c(t) is zero for all but a finite number of t T p . It is usually written
P
ct t. The support of c is the union of the supports
as a formal sum
tT
t(s p ) of those simplexes t for which ct , 0. The boundary of the p - chain

c is the (p 1)-chain
p1 c =
tT p
ct
p
X
(1) j j t.
j=o
The singular p-chains of X form a free abelian group C p (X, Z) and

p1 : C p (X, Z) C p1 (X, Z) is a homomorphism with p1 p = 0;
p1 decreases support, i.e., supp p1 c supp c .
If W = {Vi }iI is a covering of X, let C p (X, Z, W ) be the subgroup
of C p (X, Z) consisting of chains c such that ct = 0 unless supp t Vi
for some i I. Let j : C p (X, Z, W ) C p (X, Z) be the inclusion homomorphism. Since p1 decreases supports, there is an induced homomorphism p1 : C p (X, Z, W ) C p1 (X, Z, W ) which commutes with
j.
It is known (Cartan Seminar, 1948-49, Expose 8, 3) that there is a
subdivision consisting of homomorphisms r : C p (X, Z) C p (X, Z, W )
such that
(i) rc = c if c C p (X, Z, W ),
160
147
Lecture 26
148
(ii) supp rc = supp c.
Further, there is a homotopy h p+1 : C p (X, Z) C p+1 (X, Z) such

that
(iii) p h p+1 c + h p p1 c = jrc c,
(iv) h p+1 jc = 0,
(v) supp h p+1 c supp c.
Let I be well-ordered, let ji (t) = t if supp t Vi but supp t 1 Vk for
all k < i and let ji (t) = 0 otherwise. This defines a homomorphism
ji : C p (X, Z, W ) C p (X, Z),
with j =
ji and supp ui c Vi supp c.
iI
li =
Let
j r : C (X, Z) C p (X, Z). Then supp li c Vi supp c.
P iP p
Let 1 = li = ji r = jr. Since r j = 1 : C p (X, Z, W ) C p (X, Z, W ),
iI
l2 = j(r j)r = jr = l. Further, since jr = l, we have

p h p+1 + h p p1 = l 1,
and supp lc supp c.

If U is open, let C p (X, Z)U be the set of chains c C p (X, Z) such that
U does not meet supp c. Then C p (X, Z)U is a subgroup of C p (X, Z); let
S pU = C p (X, Z)/C p (X, Z)U . Since p1 , h p+1 , li and l decrease supports,
there are induced homomorphisms
p1 : S pU S p1,U
li : S pU ,
h p+1 : S pU S p+1,U
and l : S pU S pU .
161
If V U then C p (X, Z)U C p (X, Z)V and there is an induced epimorphism VU : S pU S pV which commutes with p1 , h p+1 , li and
l. Then {S pU , VU } is a presheaf which determines a sheaf S p called the
sheaf of singular p-chains. There are induced sheaf homomorphisms
p1 : S p S p1 , h p+1 : S p S p+1 ,
Lecture 26
149
li : S p S p , and l : S p S p ,
with
1 + p h p+1 + h p p1 = l =
li .
iI
If c C p (X, Z) and U = X V i then, since supp li (c) Vi , li (c)

C p (X, Z)U . Therefore li : S pW S pW is the zero homomorphism for
each open W U. Hence li (S px ) = 0x for all x X V i .
Definition. If G is a sheaf of A modules, the sheaf S p Z G is a sheaf
of A-modules called the sheaf of singular p - chains with coefficients in
G.
Let S p = Skp Z G for some fixed integer k and let d p+1 : S p
S p+1 , h p1 : S p S p1 , li : S p S p and l : S p S p be the
homomorphisms induced by kp1 , hkp+1 , li and l. Then
162
X
1 + d p h p1 + h p d p+1 = l =
li ,
p
and li (S x )0x for all x X V i .
dp
d p+1
The sequence S p1 S p S p+1 is homotopically fine.

Proof. Let U = {Ui }iI be a locally finite covering with i I such
that U i is normal for i I (i ). Shrink U to a covering W = {Vi }iI
with V i Ui . Construct the homomorphisms h p1 : S p S p1 ,
li : S p S p (as above) such that
X
li ,
1 + d p h p1 + h p d p+1 =
iI
and li (S xp ) = 0x if x < V i . Thus we can take Ei = V i .
Definition . Let C
p (X, G ) = (X, S p Z G ); this A module is called
the module of singular p - chains of x with coefficients in G . Let H
p (X,
G ) = ker p1 /im p in the sequence
p
p1
C
C
p (X, G ) C p1 (X, G )
p+1 (X, G )
Lecture 26
150
163
where the homomorphism p : C

p+1 (X, G ) C p (X, G ) is the one
induced by the homomorphism p : S p+1 S p . The A module
H
p (X, G ) is called the p-th singular homology module of the space X
with coefficients in the sheaf G , and supports in the family .
P
ct t
An element of a stalk S px can be written uniquely in the form
tT p
where ct Z, ct = 0 if x is not in the closure of the support of t, supp t,

and ct = 0 except for a finite number of t.
An element of a stalk S qx Z G x can be written uniquely in the form
P
gt t where gt G x , gt = 0 if x < supp t and gt = 0 except for a finite
xT p
number of t.
P
t
An element of C
p (X, G ) can be written uniquely in the form
tT p
t where t is a section of G over supp t, t = 0 except for a set of

simplexes whose supports form a locally finite system and the set of
points x such that, for some t, t (x) is defined and , 0x is contained in
a set of . (A section over a closed set E is a map : E G such that
: E E is the identity.)
Definition . An n - manifold is a Hausdorff space X which is locally
euclidean, i.e., each point x X has a neighbourhood which is homeomorphic to an open set in Rn .
If X is an n-manifold, then dim X = n.
164
Proof. Since X can be covered by open sets whose closures are homeomorphic to subsets of Rn , X is locally n-dimensional. Then each closed
set E is locally of dimension n and E is paracompact and normal, hence dim E n. Further, any non-empty set E has a closed
neighbourhood V and V contains a closed set homeomorphic to the
closure of an open set in Rn . Hence dim V n; thus dim X = n, and
this completes the proof.

p
p1
In the sequence S p+1 S p S p1

Let H p = ker p1 / im p ; H p is called the p - the singular homology sheaf in X.
Lecture 26
151
If X is an n-manifold, the p-th singular homology sheaf in X is locally isomorphic with the p-th singular homology sheaf in Rn .
Proof. Let x0 U1 X where U1 is open in X and let f : U1 U1
be a homeomorphism onto an open set U1 Rn . Choose an open set
U = Rn f (V).
Then
V with xo V, V U1 and let U2 = X V,
2
{U1 , U2 } is a covering of X, and there is the homotopy defined above,
h + h = l1 + l2 t,
with l2 (S px ) = 0 for x V. Hence H p (S ) is isomorphic with H p (l1 S )
in V. But f : U1 U1 takes l1 (S p ) into l1 (S p ) where S p is the sheaf
of singular p chains in Rn and l1 is the corresponding homomorphism
for the covering {U1 , U2 } of Rn . In f (V), H p (l1 S ) is with H p (S ).
Using triangulations of Rn , one can verify, for Rn , that H p = o for
p , n and Hn is isomorphic with the constant sheaf (Rn Z, , Rn).
One uses a homotopy which does not decrease supports and which does 165
not induce a sheaf homotopy. The isomorphism is not a natural one but
depends on the choice of an orientation for Rn .
In an n-manifold X, H p = 0 for p , n and Hn is locally isomorphic
with Z. Let J = Hn ; if J is isomorphic with Z the manifold is called
orientable, otherwise the manifold is said to be non-orientable and J is
called the sheaf of twisted integers over X. Example 2 is the restriction
ot the Mobius band of the sheaf of twisted integers over the projective
plane.
If S p = Snp Z G on an n-manifold X, then H p (S ) for p , 0
and H o (S ) = J Z G .
Proof. Since S px is a free abelian group, so are the subgroups Z px and
B px . Also, H px being either 0 or Z, is free. It is known (Cartan Seminar,
1948-49, Expose 11) that if
0 F F F 0
is an exact sequence of abelian groups and F is without torsion and if
G is an abelian group, then the induced sequence
0 F G F G F G 0
Lecture 26
152
is exact. From the following commutative diagram of exact sequences,

one can see that
H p (S G ) = ker p1 / im p z p G /B p G H p G .
166
Thus H p (S ) = Hnp G = O for p , O and H p (S ) = J G

for p = O.
0o
Bp G o
S p+1 G o
z p+1 G o
|
/ Sp G
zp G

/ B p1 G
z p1 G
Hp G

q
qqq
q
q
q
x qq
q
B p2 G
qq
qqq
q
q
q
xqqq
S p1 G
0p
v
vv
vv
v
v
v
{v
/0
0
vv
vv
v
v
vv
vz v
p1
H p1 G

0
Q.e.d.
Lecture 26
153
Proposition 19. If X is an n- manifold, there is an isomorphism:

1
: Hnp
(X, G ) H (X, J Z G ).
(X, G ) = (X, S
Proof. Since Cnp
(X, S ), Hnp (X,
np z G ) =
p
o
p
G ) = H (X, S ). And, since H (S ) = J z G , H (X, J z G ) =
p
H (X, H o ). By proposition 17 and 18, there are isomorphisms
p
: H pC (S ) H (X, H o ),
: H pC (S ) H p (X, S ).
Thus 1 is the required isomorphism.
This proposition is part of the Poincare duality theorem.
Lecture 27
Given any sheaf G of A-modules, there exists an exact sequences of 167
sheaves
dq
d1
OG
S o S 1 S q1 S q
where each S q (q O) is finite.
q
Proof. For each open set U of X, let S U (q = O, 1, . . .) be the abelian

group of integer valued functions f (x0 , . . . , xq ) of q + 1 variables x0 , . . .
xq U. If V is an open set with V U, the restriction of the functions
q
q
f gives a homomorphism VU : S U S V . There is a homomorphism
q+1
q
q+1
dU : S U S U defined by
d
q+1
f (xo , . . . , xq+1 ) =
q+1
X
(1) j f (xo , . . . , x j , . . . , xq+1 ).
j=o
o is the inclusion homomorphism of the constant funcIf e : Z S U

tions on U, the sequence
e
q
q1 d
d1
o
S U
OZ
SU
S U
is exact (Cartan Seminar 1948-49, Expose 7, 8). Clearly VU dq+1 f =

q
dq+1 VU f and VU commutes with e. The presheaves {S U , VU }(q =
0, 1, . . .) determine sheaves S q and there is an induced exact sequence
e
q
q1 d
d1
o
S U
OZ
SU
155
S q
Lecture 27
156
p
It is easily verified that each abelian group S U is without torsion,

q
and this property is preserved in the direct limit. Hence each stalk S x = 168
q
q
dir lim{S U , VU } xU is without torsion, i.e., the sheaves S are without
torsion. Therefore the sequences of sheaves of A-modules
e
d1
dq
O Z Z G
S o Z G S q1 Z G S q Z G
is exact, (this follows from the fact that if O F F F O is
an exact sequence of abelian group, F is without torsion, and G is an
abelian group, then the sequence
O F G F G F G O
is exact), that is, the sequence
e
d1
dq
(1) 0 G
S o Z G S q1 Z G S q Z G
is exact.
We now show that each of the sheaves S q Z G is a fine sheaf. To
q
do this, let E G with E closed and G open and let h : S U S Up be
the homomorphism defined by
h f (xo , . . . , xq ) = f (xo , . . . , xq ) if xo , . . . , xq U G,
= 0 otherwise.
Then h commutes with VU and induces a homomorphism h : S q
and h(S qx ) = 0x if
for which hx : S qx S qx is the identity if x G,
There is then an induced homomorphism h : S q Z G
x X G.
q
q
S Z which is the identity on the stalks S x Z G x for x G and zero
q
q
Thus S Z G is fine, and the sequence (1)
on S x Z G x for x X G.
is a fine resolution of G .
We now give a definition of the cochains of a covering of a space
X coefficients in a sheaf G and support in a -family and also give an
alternative definition of the cohomology groups of X with coefficients
in G and supports in the family . We then prove Lerays theorem on
acyclic coverings.
Sq
169
Lecture 27
157
p
Definition. Let C (U , G ), for an arbitrary covering U = {Ui }iI be the

C p (U , G ) consisting of those cochains f such that the closure of the set
supp f = {x : f (io , . . . , i p )(x) is defined for some = (io , . . . , i p ) and
, 0x }, belongs to the family . (If p = 0, this definition does not agree
with the previous one even when U is a -covering.)

Then, since the homomorphisms
p+1 : C p (U , G ) C p+1 (U , G )
+ : C p (U , G ) C p (W , G )
decrease supports, they induce homomorphisms
p
p+1
p+1 : C (U , G ) C (U , G )
and
+ : Cp (U , G ) Cp (W , G )
p
respectively. Hence there are cohomology module H (U , G ) and hop

p
momorphisms W U : H (U , G ) H (W , G ). For p = 0, Ho (U , 170
G ) = (X, G ) for every covering G , and
W U : (X, G ) (X, G )
is the identity. Using the directed set of all proper covering of
p
X, {H (U , G ), W U }U ,W is a direct system. The direct limit of this
p
system will be denoted by H (X, G ). This module also is called the p- th
cohomology module of the space X with coefficients in the sheaf G and
supports in the family . (This cohomology module is isomorphic with
that previously defined by means of -coverings, Lecture 21.) There are
p
p
homomorphisms into the direct limits, U : H (U , G ) H (X, G ).
Let X be a paracompact normal space, G a sheaf of A-modules over
X, and let U = {U} be locally finite proper covering of X where each
U U is an F set. (U is indexed by itself.) Let be the set of all
intersection E U with E . Since U is an open F set in X, U
is paracompact and normal. Hence each E U is paracompact
and normal. One easily verifies that is a -family in U . We now as
sume the following conditions on the family and the covering U .
Lecture 27
158
(i) For some infinite cardinal number m, the union of fewer than m
elements of is contained in a set belonging to . (If X ,
choose m greater than the number of closed sets of X; if is family
of compact sets, let m = No .)
171
(ii) Each set in meets fewer than m sets of the covering U .

(iii) Each U U is an F sets, i.e., is a countable union of closed
subsets of X.
(iv) For each U = Uo U p , Hp (U , G ) = 0 for q > 0; here G
denotes the restriction of G to U
(A covering U is called acyclic is conditions (iv) is satisfied.)
Under the conditions (i), (ii), (iii), (iv) stated above, the homomorphism U : Hp (U , G ) Hp (X, G ) is an isomorphism.
Proof. Let
d1
OG
S o S q1 dq S q
p
be any fine resolution of G . Then there is a system {(C (U , S q ))}U

of double complexes, where is the cofinal directed set of all locally
finite proper coverings of X. This system is bounded above by p = 0
and on the left by q = 0.

Since X is normal, S q is fine and U is locally finite, there is a
homotopy (see Lecture 16)
k p1 : C p (U , S q ) C p1 (U , S q )
(p > 0).
Since k p1 decreases supports, it induces a homomorphism

p
k p1 : C (U , S q ) C p1 (U , S q ) (p > 0),
172
with p k p1 + k p p+1 = 1. Hence H (U , S q ) = 0 for p > 0, and for

p = 0, Ho (U , S q ) = (X, S q ). Hence
O,q
H12 C (U , S ) = H q (X, S ).
Lecture 27
159
Thus we have the isomorphisms indicated below (see Lecture 20):

H q (X, S ) o
0,q
H12
C (U , S ) o
0,q
H12 C (S )
H qC (U , S )
H qC (S ).
Hq
o
(X, S )
Since S q is fine and X is normal, S q is locally fine, hence its restriction to U is locally fine. But U is paracompact and normal, so
that the restriction of S q to U is fine. Hence there is an isomorphism
(see Proposition 17 and 18, lectures 23 and 25 respectively),
q
1 : H q (U , S ) H (U , G ).
Hence by condition (iv), H q (U , S ) = 0 for q > 0, H o (U , S )
Ho (U , G ) = (U , G ).
p
If f C (U , S q ) (q > O) and dq+1 f = O, then (dq+1 f )(Uo , . . . ,
U p ) = O in each U = Uo U p . Since H q (U , S ) = 0(q >
0), there is a section g(Uo , . . . , U p ) (U , S q1 ) with dg(Uo , . . . ,
U p ) = f (Uo , . . . , U p ) (choose g(Uo , . . . , U p ) = 0 if f (Uo , . . . , U p ) = 0.
There is then a cochain g C p (U , S q1 )) with dg = f , (see p.57).
o (U , S q ), supp f is contained in a set belonging to and 173
Since f C
hence f () is different from zero on fewer than m sets U . Then g()
is different from zero on fewer than m sets U and hence supp g is the
union of fewer than m set {x U : g()(x) , 0, each of which is in
and hence has its closure in . Hence supp g is contained in a set
p
p,q
belonging to and g C (U , S q1 ). Hence H2 C (U , S ) = 0(q >
0).
Since the sequences
d1
0 C p (U , G ) C p (U , S o ) C p (U , S 1 )
is exact, if f Cp (U , S O ) and d1 f = 0, then f = e(g) for some
g C p (U , G ) and clearly g Cp (U , G ). Thus
p,o
p,0
H2 C (U , S ) C (U , G ) and H21 C (U , S ) H (U , G ).
Lecture 27
160
Thus we have the isomorphism indicated below:
H pC (U , S )
/ H p,0C (U , S )
21
/ H p (U , G )
H pC (S )
/ H p,0C (S )
21
/ H p (X, G ).
Combining this diagram with the previous one, we see that the homomorphism : Hp (U , G ) Hp (X, G ) is an isomorphism.
Q.e.d
In particular, we have proved the following proposition (Cartan Seminar, 1953-54, Expose 17, p.7).
174
Proposition 20. If U is a locally finite proper covering of a paracompact normal space X by open F sets, and if G is a sheaf of A-modules
such that H q (U , G ) = O(q > O) for every U = Uo Uk (k =
O, 1, . . .), then
U : H p (U , G ) H p (X, G )
is an isomorphism.
p
In the case that is the family of all compact sets of X, we write H

instead of Hp .
Proposition 20 (-a). If U is a locally finite proper covering of a locally

compact and paracompact Hausdorff space by open F sets with compact closures, and if G is a sheaf of A-modules such that H p (U , G ) =
O(q > O) for every
U = Uo Uk (k = 0, 1, . . .)
then
p
: H (U , G ) H (X, G )
is an isomorphism.
Note . It is no restriction to assume that U is a proper covering. If W
is any covering, there is an equivalent proper covering U with the open
sets. Then W U and U ,W are isomorphisms.
Lecture 28
Direct sum of modules
175
{M i}
Definition . The direct sum of a system

iI of A-modules is an Amodule whose elements are systems {mi }iI , usually written as formal
P
sums i mi where mi M i and mi = Q for all but a finite number of i.
The operations in M are defined by
X
X
X
X
X
a
mi =
ami ,
(mi1 + mi2 ),
mi2 =
mi1 +
where
i
i m1
M,
i
i m2
M,
im
M and a A.
P
Clearly there is a homomorphism p j : i M i M j defined by
P
P
p j ( i mi ) = m j and a homomorphism h j : M j i M i defined by
pi h j (m j ) = 0 for i , j, p j h j (m j ) = m j , m j M j .
A system of homomorphism (g , gi ) : (A, M i ) (B, N i ), i I, inP

P
duces a homomorphism (g , g) : (A, i M i ) (B, i N i ) where
P i
P i i
g( i m ) = i g (m ). There is commutativity in
(A, M j)
hj
/ (A, P M i )
i
pj
(g ,g)
(g ,g j )
(B, N j )
hj
/ (B,
P
i
iN )
161
/ (A, M j )
(g ,g j )

/ (B, N j ).
Lecture 28
162
P
P
P
P
P
P
Proof. Let i mi1 i M i, i mi2 i M i, i mi i M i and let a A.
Then
X
X
X
X
gi (mi1 + mi2 )
mi2 ) = g( (mi1 + mi2 ) =
g( mi1 +
i
176
X
X
X
(gi (mi1 ) + gi (mi2 )) =
gi (mi1 ) +
gi (mi2 )
i
X
X
= g( mi1 ) + g( mi2 )
g(a
X
i
i
X
Xi
i
m ) = g( am ) =
gi (ami )
i
g (a)g (m ) = g (a)
i
gi (mi )
X
= g (a)g(
mi ).
i
X
X
i
j j
g p ( m ) = g m = p j ( gi mi )
i
X
= p g( mi )
j
i
k k j
p gh (m ) = g p h (m j ) = gk (0) = 0 for k , j
and
p j gh j (m j ) = g j p j h j (m j ) = g j (m j ),
hence
gh j (m j ) = h j g j (m j ),
m j M j.
The operation of forming the direct limit commute with the operation of
forming the direct sum.

Proof. Let {A , Mi , , i }, ( a directed set) be a direct system
for each i I. Let the direct limits be (A, M i) with homomorphisms
( , i ) : (A , Mi ) (A, M i). There are induced homomorphisms
X
X
( , ) : (A ,
Mi ) (A ,
Mi )
i
Lecture 28
177
163
for which = , < < v and is the identity. Thus

P
{A , i Mi , , } is direct system. Let its direct limit be (A, M)
with homomorphisms
X
( , ) : (A ,
Mi ) (A, M).
i
(A , Mi )
/ (A, M i )
hi
hi
(A ,
/ (A , M i )
P
P
/ (A, M)
/ (A , i M i )
VVVV
NNN
VVVV
NNN
VVVV
N
V
V
V
V
V
VVVVNNN&
+
P }
Mi ) V
(A,
hi
Mi)
The system of homomorphisms (i , i ) : (A , Mi ) (A, M i ) inP i

P i
duces a homomorphism ( ,
) : (A , i M ) (A, i M ). Since for
< ,
X
X
X
i
i
i
(
m
)
=
m
)
=
i i mi
i
i mi
Xi
mi ),
therefore
= and there is an induced homomorphism : M
P i

iM.
P
P
If i mi i M i then mi = 0 except for a finite number of i, say
i
i
i1 , . . . , ik . Then, for some , each mi j has a representative mj Mj ; let
P
P
P
P
i
mi = 0 for i < (i1 , . . . , ik ). Then i mi i Mi and i mi =
i m =
P i
i m . Thus is an epimorphism.
178
To show that is a monomorphism, let m M and (m) = 0.
P
Choose a representative i mi of m. Then mi = 0 except for a finite
number of i, say i = i1 , . . . , ik . Since
X
X
0 = (m) =
mi ) =
i mi ,
(
i
Lecture 28
164
ij
j
therefore each mj = 0. Now choose so that each
mj = 0.
P i
Then m is represented by 0 in i M and hence m = 0. Thus is an
P
isomorphism. We identity M with i M i under the isomorphism .
Direct sum of sheaves

Let a be a sheaf of rings with unit and let {S i }iI be a system of
sheaves of a-modules. Then there is a unique sheaf S whose stalks
P
j
j
are the direct sums S ix such that the homomorphisms hx : S x S x
iI
determine a sheaf homomorphism h j : S j S .

This S is called the direct sum of the sheaves S i and is denoted by
P
S = S i.
iI
P
P
Proof. Uniqueness. If s = i si i S ki with si = 0 except for i1 , . . . , ik ,
choose a neighbourhood U of x and sections f j : U S i j , j = 1, . . . , k
SP i
S x ) be defined by f (y) =
such that f j (x) = S i j . Let f : U S (=
x iI
P ij
ij
j h f j (y). Since h is a sheaf homomorphism, the composite function
hi j
fj
U S i j S
179
is a section and hence f is a section. Since

f (x) =
X
j
hi j f j (x) =
X
j
hi j si j =
si = s,
the section goes through s. Thus, since such section cover S , they
uniquely determine the topology of S .
i , , i }
Existence From the presheaves of sections (a, S i ) = {AU , S U
VU
VU
P
i = (U, S i ). Then {A ,
i
where AU = (U, a) and S U

U
i S U , VU , VU }
(where U runs through the directed set of all the open sets of X) is a
presheaf. It determines a sheaf (a, S ) and there is a homomorphism
Lecture 28
j
hU : S U
165
P
i
i SU
with commutativity in
j
j
(AU , S U )
j
)
(VU , VU
hU
/ (AU , Pi S i )
U
(VU , VU )
(AV , S V )
j
hV

/ (AV , Pi S i ).
V
180
Hence there is an induced sheaf homomorphism h j : S j S .

The stalk (A x , S x ) of (a, S ) is the direct limit of the direct system
P i
{AU , i S U
, VU , VU } xU which is identified with the direct sum of the
P
i , , i }
. Thus S x = i S ix
direct limits (A x , S ix ) = dir lim{AU , S U
VU xU
VU
and the required sheaf exists, q.e.d.
P i
j
S U (as defined in the
There is also a homomorphism piU : i S U
beginning of the lecture) with commutativity in
(AU ,
i
i S U)
pUj
(VU , VU )
(AV ,
P
i
i S V)
/ (A , S j )
U
U
j
(VU , VU )
pVj

/ (A , S j )
V
V
Hence there is a sheaf homomorphism p j : S S j which is

P
clearly onto. If s = i si S , it is easily verified that p j (s) = s j , hence
p j h j (s j ) = s j and p j hi (s j ) = 0 if i , j.
Lecture 29
Given homomorphisms gi : S i T (i I), there is an induced homo- 181
P
morphism g : S i T with ghi = gi .
iI
P
P
Proof. If s = i si S x , let g(s) = igi (si ) T x .Then g|S x : S x T x is
clearly a homomorphism. Choose an open set U and sections f j : U
P
S i j so that the section defined by f (y) = hi j f (y) goes through s. Then
P
g f (y) = gi j f j (y) and g f being the sum of a finite number of sections,
is a section. Thus g is continuous and is a sheaf homomorphism.

Note. Since a itself is a sheaf of a-modules, there is, for any I, a direct
P
sum a where each direct summand is a. This is again a sheaf of a
iI
-modules.
If S is a sheaf of a-modules over a space X, and aY is the restriction

of a to a subset Y, then clearly the restriction SY of S to Y is a sheaf of
aY -modules.
Definition. The following properties of sheaves S of a- modules over a
space X are called property (a1 ) and property (a) :
Property (a1 ). There is a covering {U j } jJ of X, and for each j J,
P
aU j SU j .
there is an index set I j and an epimorphism j :
iI j
Property (a). There is a covering {U j } jJ of X, and for each j J,

Pk j
there is a natural number k j and an epimorphism j : i=1
aU j SU j . 182
167
Lecture 29
168
S has property (a1 ) if and only if each point x X has a neighbourhood U such that the sections in (U, S ) generate SU . (That is,
P
for each y U and s S y , s = kj=1 ayj f j (y) for some finite number k of
j
elements a j Ay and sections f j (U, S ).)
Proof. Necessity. Let S have property (a1 ). Since {U j } is a covering,

x U j for some j; let U = U j . Then there is an index set I and an
P

epimorphism : aU SU .
iI
Let f i = hi 1 : U S where 1 is the unit section in aU ,
1
hi X
U
aU
aU
SU .
iI
P
Then if s SU , s = ( i aiy ) for some aiy Ay with aiy = Oy except
for a finite number of i, say i = i1 , i2 , . . . , ik . Then
k
X
i
ayj fi j (y)
j=1
k
X
i
ayj hi j 1y
j=1
hi j ayj 1y
j=1
k
X
= (
j=1
183
k
X
X
i
hi j ayj ) = ( aiy ) = s.
iI
Sufficiency. For each x X, there is a neighbourhood U x of x such

that the sections over U x generate SU x . Then {U x } xX is a covering of
X. Let Ix be the set of sections (U x , S ). For each section i I x ,
there is a sheaf homomorphism ix : aU x SU x given, for a Ay , by
ix (a) = a i(a).
Then there is an induced homomorphism
X
x :
aU x SU x .
iIx
Then for s S y and y U x ,

s=
k
X
j=1
ay i j (y) =
k
X
j=1
X
i
j
xj (ay ) x ( aU x ).
iIx
Lecture 29
169
Thus x is an epimorphism.
S has property (a) if and only if each point x has a neighbourhood
U such that some finite number of sections fi (U, S ) (i = 1, . . . , k)
generate SU .
Proof. Similar to the proof given above.
It is clear that (a) implies (a1 ), i.e., each sheaf with property (a) has
P
P
property (a1 ). The sheaf a has property (a1 ) and ki=1 a has property
iI
(a). In particular, the sheaf a of a-modules has property (a).

If Si , i = 1, . . . , k are sheaves of a- modules with property (a1 )
P
(resp. (a)), then the direct sum ki=1 ai has property (a1 ) (resp (a)).
Proof. Clear.
Statements (1), (2), (3), (4), (5), (6), (Lectures 29, 30, 31,) are required to prove Serres theorem on coherent sheaves, (see the next lec- 184
ture for the definition of coherent sheaves), i.e., if 0 S S
S 0 is an exact sequence of sheaves, and if two of them are coherent, then the third is also coherent.
(1)
If f : S S is an epimorphism and S has property (a1 )
(resp (a)), then S has property (a1 ) (resp (a)).

Proof. Clear.
Example . If M is a finitely generated A- module, the constant sheaf

M has property (a) with respect to the constant sheaf A. If a constant
sheaf is a sheaf of a- modules, then it has property (a1 ). If X is the unit
segment 0 x 1, the subsheaf S of the constant sheaf Z2 obtained
by omitting (1, 1) does not have property (a1 ) either as a sheaf of Zmodules or as a sheaf of Z2 -modules. With the same X, the sheaf S of
germs of functions f : X Z2 , considered as a sheaf of Z2 -modules,
has property (a1 ) but not property (a). But, considering S as a sheaf
of rings with unit, it has property (a) with respect to itself, The sheaf
a of germs of analytic functions in the complex plane has property (a)
as a sheaf of a- modules, but as a sheaf of C- modules (where C is the
field of complex numbers) it does not even have property (a1 ). For, there
170
Lecture 29
are natural boundaries for analytic functions, e.g., let f be an analytic

function in |z| < 1, with |z| = 1 as natural boundary. If a has property (a1 )
as a sheaf of C-modules, then by considering a point on the boundary,
we see that f can be continued to a neighbourhood of this boundary
point and this is a contradiction.
Lecture 30
Definition. The following properties of sheaves S of a- modules over a 185
space X are called property (b1 ) and property (b).
Property (b1 ). For each open set U of X and each homomorphism
P
f : aU SU , ker f has property (a1 ) as a sheaf of aU - modules.
iI
Property (b). For each open set U of X and each homomorphism f :

Pk
i=1 aU SU , ker f has property (a) as a sheaf of aU - modules.
Note. Since (a1 ) and (a) are local properties, properties (b1 ) and (b) are
equivalent to the following properties (b1 ) and (b ) respectively.
Property (b1 ). For each neighbourhood V of each point x X, there
exists an open set U, x U V, such that for each homomorphism
P
f : aU SU , ker f has property (a1 ) as a sheaf of aU - modules.
iI
Property (b ). For each neighbourhood V of each point x X, there

exists an open set U, x U V such that for each homomorphism
k
P
f : aU SU , ker f has property (a) as a sheaf of aU -modules.
i=1
Thus (b1 ) and (b) are also local properties. The sheaf ker f is called 186
the sheaf of relations between the sections fi : U SU , where fi =
f hi 1,
f
1
hi X
aU
SU .
U
aU
iI
The sheaf of relations between the sections fi is described by the

following result.
171
Lecture 30
172
P
The element of (ker f )x for x U are the elements i ai of which
i ai fi (x) = 0.
P
Proof. For each element ai , only a finite number of the ai being different from zero, we have
P
ai =
q
X
hi j ai j and
j=1
q
q
X
X
X
ij
f ( ai ) = f ( h ai j ) =
f hi j (ai j )
j=1
q
X
j=1
q
X
ai j f hi j (lx ) =
j=1
ai j fi j (x)
j=1
ai fi (x),
and this completes the proof.
If we start with a system { fi }iI of sections of S over an open set U,

then each fi defines a homomorphism (again denoted by fi ) fi : aU
SU where fi (a) = a fi (x), a A x . Then the system { fi } of homomorP
phisms defines a homomorphism f : aU SU , and the sheaf ker f
iI
187
is called the sheaf of relations between the sections fi : U SU .

(2)
If S has property (b1 ) (resp (b)), then every subsheaf of S
has property (b1 ) (resp (b)).
Proof. Clear.
Definition. A sheaf S of a- modules is called coherent if it has properties (a) and (b).
Note . If S is a coherent sheaf, then SU is coherent for each open set
U. Coherence is a local property, i.e., if each point has a neighbourhood
U such that SU is coherent, then S is coherent. If we define a sheaf S
to be of finite type if, for each x X, there is an open set U, x U, such
that each stalk of SU is generated by the same finite number of sections
Lecture 30
173
f1 , . . . , fk over U; it is then easily verified that conditions (a) and (b) for
coherence are equivalent to the following conditions:
(i) The sheaf S is of finite type.
(ii) If f1 , . . . , fk are any finite number of sections of S over any open
set U, then the sheaf of relations between these finite number of
sections is of finite type.
Example 30. Let a be the constant sheaf Z2 on 0 x 1, let R be
the subsheaf obtained by omitting (1, 1) and let S be the quotient sheaf
with stalks Z2 at 1 and zero elsewhere. Then the natural homomorphism
a S has kernel R and R does not have property (a1 ). Hence S has
neither (b1 ) nor (b).
Example 31. Let A be the ring of Example 5, with elements 0, 1, b, c, 188
such that b2 = b, c2 = c, bc = cb = 0. Let a be the subsheaf of the
constant sheaf A on 0 x 1 obtained by omitting (1, b) and (1, c). Let
S be the constant sheaf Z2 on which 1 and c of A operate as the identity
and b operates as zero. The sheaf R of relations for the section 1 of Z2
consists of 0 and (x, b) for x < 1. If U is a connected neighbourhood of
1, the only homomorphism of aU into RU is the zero homomorphism.
Thus R does not have (a1 ) and S has neither (b1 ) nor (b).
Example 32. Let A be the ring Z[y1 , y2 , . . .] of polynomials in infinitely
many variables with integer coefficients. Let a be the constant sheaf A
on 0 x 1 and let S be the constant sheaf Z on which all the indeterminates y1 , y2 , . . ., operate as zero. The sheaf of relations for the
section 1 of Z is the constant sheaf formed by the ideal of all polynomials without constant terms. This ideal is not finitely generated, hence S
P
does not have (b). However, for every homomorphism f : aU SU ,
iI
ker f is constant on each component of U and hence has property (a1 ).

Thus S has (b1 ) but not (b).
f
S
S 0 is an exact sequence of sheaves
(3)
If 0 S
of a - modules such that S has (a1 ) (resp (a)) and S has (b1 ) (resp
(b)), then S has (a1 ) (resp (a)).
Lecture 30
174
189
Proof. If S has (a1 ), each x X has a neighbourhood U for which

P
P
there is an epimorphism : aU SU . Since g : aU SU is a
iI
iI
homomorphism and S has (b1 ), ker g has (a1 ). Hence for some open
set V with x V U, there is a homomorphism
X
X
:
aV
aV
jJ
iI
such that im = (ker g)V . Hence, since is an epimorphism im =

(ker g)V , and therefore im = (im f )V . Then, since f is a monomorphism, there is an epimorphism
X
aV SV .
f 1 :
jJ
Thus S has (a1 ). Similarly if S has (a) and S has (b), then S
has (a).
aV
jJ
aV
iI
CC
CCg
CC
CC

!
g
/S
/ S
/ S
/0
Q.e.d.
S
S
0 is an exact sequence of
(4)
If 0
sheaves of a- modules such that S and S have (b1 ) (resp (b)) then
S also has (b1 ) (resp (b)).
Proof. Let S and S have (b1 ) and let :
iI
190
aU SU be a given
homomorphism. Since S has (b1 ), ker g has (a1 ) and hence for each
x U, there is an open set V with x V U and a homomorphism,
X
X
:
aV
aV
jJ
iI
Lecture 30
175
such that im = (ker g)V . Then im (ker g)V = (im f )V and hence
there is a homomorphism
X
:
aV SV
jJ
with f = . Since S ahs (b1 ), ker has (a1 ) and there is an open set
W with x W V, and a homomorphism
X
X
:
aW
aW
kK
jJ
such that im = (ker )W .
aW
kK
mmm
mmmmm

mm

m
m

m
P
mmm

aV vm

jJ

BB

BB

BB

B

B!

P

aU

iI

GG
GGg
xxx

GG
x

x
GG

x
x
G#
x

|
f
g
/S
/ S
/ S
/0
To show that S has (b1 ) it is enough to show that im = (ker )W .

P
For any element r
aW ,
kK
(r) = f (r) = f (0) = 0.

191
Thus im (ker )W . Next, for any element p (ker )W , we

P
aW such that (q) = p;
have p (ker g)W = (im )W . choose q
jJ
then
(q) = f 1 (q) = f 1 (p) = f 1 (0) = o.
176
Lecture 30
Thus q (ker )W = im , and hence p = (q) im . Thus im =

(ker )W and S has (b1 ). Similarly if S and S have (b), it can be
proved that S also has (b).
Lecture 31
P
If f : S S is an epimorphism, : ki=1 aU SU a homomor- 192
phism, and if x is a point of the open set U, then there exists an open set
V with x V U and a homomorphism
:
k
X
aV SV such that f = |
i=1
k
X
aV .
i=1
Proof. Since f : S S is an epimorphism, each of the sections i =

hi 1 : U S , i = 1, . . . , k, is locally the image of a section in S .
Hence there is an open set Vi with x Vi U and a section i : Vi S
such that f i i |Vi . Let i : aVi SVi be the homomorphism defined
by i (a) = a i (a) These homomorphisms i induce a homomorphism
:
k
X
aV SV
i=1
where V =
Tk
i=1 Vi .
Then, if
Pk
i=1 ai
Pk
i=1 aV ,
we have
k
k
k
X
X
X
f ( ai ) = f ( ai i (ai )) =
ai i (ai )
i=1
i=1
k
X
i=1
ai hi 1(ai ) =
i=1
k
X
= (
k
X
hi ai
i=1
ai ),
i=1
i.e., f = |
Pk
i=1 aV .

177
Lecture 31
178
f
193
S
(5)
If 0 S
of a- modules such that S has property (a) and S has property (b),
then S has property (b).
P
Proof. Let U be an open set and let : ki=1 aU SU be a homomorphism. Since g : S S is an epimorphism, by the result proved
above, if x U, there is an open set V with x V U and a homoP
P
morphism : ki=1 aV SV such that g = | ki=1 aV . Since S has
property (a), there is an open set W with x W V and an epimorP
phism : 1i=k+1 aW SW . Then and f induce a homomorphism
P
: 1i=1 aW SW .

P
P
We also have the projection homomorphisms p : li=1 aW ki=1
P
P
aW and p : 1i=1 aW li=k+1 aW such that = p + f p .
Pl
p
Pl
o p
/ Pk aW
i=1 aW
i=1
KK f
t
ttt
KK
KK
tt
KK
tt
t
K

t
y
%
g
f
/ S
/ SW
i=k+1 aW
K
194

/ S W
/0
The second square forms a commutative diagram since g = gp +

g f p = gp = p and hence p maps ker into ker . Actually, p maps
ker onto ker , for, if a ker , then ga = a = 0 and by exactness
there exists b SW such that f b = a. Since is an epimorphism, there
P
exists c li=k+1 aW such that c = b. Then (a + c) = a + f c = 0
and p(a + c) = a. Thus p maps ker . Since S has (b) ker has (a), and
since p| ker : ker ker is an epimorphism, ker has (a). Hence
S has (b).
The corresponding statement, with (a1 ) and (b1 ) in place of (a) and
(b), is not true as the following example shows.

Example 33. Let X be the union of the sequence of circles Cn = (x, y) :

x2 + y2 = x/n , n = 1, 2, . . .. let each stalk of a be the ring Z[x1 , x2 , . . .]
of polynomials in infinitely many variables, with coefficients in Z, and
Lecture 31
179
let a be constant except that, on going around the circle Cn , xn and -xn
interchange.
More precisely, let T n be the automorphism of the ring Z[x1 , x2 ,. which
interchanges xn and xn . If U is open in X and U Cn , let AU be the
ring of functions f defined on U and with values in Z[x1 , x2 , . !. .] such
1
that f is constant on each component of U not containing , 0 and, if
!
! n
1
1
a component W of U contains , 0 , f (x, y) = f , 0 for (x, y) W
n !
n
1
and y < 0 and f (x, y) = T n f , 0 for (x, y) W and y > 0. If U is
n
not contained in any Cn let AU be the ring of functions, constant on each
component of U, with values in
Z[xn1 , xn2 , . . .] Z[x1 , x2 , . . .]
!
1
where n1 , n2 , . . ., are those values of n for which , 0 < U.
n
If V U let VU : AU AV be given by VU f = f |V. Let a be the
sheaf of rings determined by the presheaf {AU , VU }.
Let I be the sheaf of ideals formed by polynomials with even coef- 195
ficients, then I is generated by the section given by the polynomial 2.
Let
S = a/I = Z2 [x1 , x2 , . . .],
then
0 I a S 0
is exact. Then, as sheaves of a-modules, I has properties (a) and (a1 ),
a has (b) and (b1 ) and S has (b) but not (b1 ).
f
S
(6)
If 0 S
of a-modules such that S and S have property (a), then S has property (a).
Proof. Since S has property (a), for each point x there is a neighbourP
hood U of x and an epimorphism : ki=1 aU SU . There is an open
Lecture 31
180
196
P
set W with x W U, a homomorphism : ki=1 aW SW , an epiPk
morphism : i=k+1 aW SW and homomorphisms , p, p as in the
P
previous proof. If s SW , there is some a ki=1 aW such that a = gs.
Then g(s a) = 0, and by exactness, for some b SW , s a = f b.
P
Then for some c ki=k+1 aW , b = c and s = a + f c = (a + c).
Pl
Hence , i=1 aW SW is an epimorphism, and hence S has property
(a).

The corresponding statement, with (a1 ) in place of (a), is not true as
the following example shows.
S
Example 34. Let X =
n=1 C n as in Example 33. Let a = Z and let Sn
be a sheaf which is locally Z4 , but on going around the circle Cn , 1 and
3 interchange. Let Sn be the subsheaf with stalks consisting of 0 and 2;
it is the constant sheaf Z2 . Let S n = Sn /S n ; this is also Z2 . Then
the sequence
0 S n Sn S n 0
is exact. Let
S =
n=1 S n , S = n=1 Sn , S = n=1 S n .
Then the sequence

0 S S S 0
is exact. Since S and S are constant sheaves, they have property
(a1 ) but S does not have property (a1 ).
Statements (1), (2), (3), (4), (5), (6) (Lectures 29, 30, 31) prove the
following proposition due to Serre.
f
S
S 0 is an exact sequence
Proposition 21. If 0 S
of sheaves of a-modules and if two of them are coherent (i.e., possess
property (a) and (b)), then the third is also coherent.
197
Corollary. If Si , i = 1, . . . , k are coherent sheaves of a- modules, then

Pk
i1 Si is coherent
Lecture 31
181
Proof. Since the sequence of sheaves

k1
Si 0
0 Sk ki1 Si i=1
is exact, the result follows by induction.
Lecture 32
Let S and S be sheaves of a- modules and let S have property (a). 198
If f and g are two homomorphisms of S S , the set of points x for
which f |S x = g|S x is an open set.
Proof. Let W be the set of points x for which f |S x = g|S x and let xo
W. Since S has property (a), there is an epimorphism
:
k
X
aU S U
i=1
for some neighbourhood U of xo . Then since xo W, f h j l xo = gh j l xo

( j = 1, . . . , k), and hence for some open set V j , with x0 V j U,
f h j 1 = gh j 1 : V j S .
f
1
hj
/ aV
/ Pk
i=1 aV
/ S
V
9 SV
g

Let V =
Tk
j=1 V j .
If
Pk
i1 ai
Pk
i1
Ai with x V, then
k
k
k
X
X
X
f ( ai ) = f ( hi ai ) =
ai f hi 1x
i=1
i=1
k
X
i=1
i=1
k
X
ai ghi 1x = g( ai ).
i=1
183
Lecture 32
184
199
P
P
Hence f | ki=1 aV = g| ki=1 aV and, since is an epimorphism,
f |S V = g|S V . Thus xo V W with V open. Hence W is an open
set.
The following result deals with the extension of stalk homomorphisms.
Let S , S be sheaves of a- modules such that S is coherent and
S has property (a). At a point x X, let f : S x S x be a homomorphism of A x modules. Then there is a neighbourhood U of x and a
homomorphism f : S U SU whose restriction to S x is f x .
Proof. Since S and S have property (a), there is a neighbourhood W
of x and epimorphisms
:
and
p
X
j=1
q
X
aW S W ,
aW SW .
j=1
For each j = 1, . . . , p, choose an element

that
q
X j
a = fx h j lx
Pq
j
i=1 ai
Pq
i=1
A x such
i=1
and choose sections
j : Vj
q
X
aV j
i=1
with x V j W, V j open, such that j (x) =

the homomorphism
q
X
g j : av j
av j
Pq
j
i=1 ai ,
i=1
be defined by g j (a) = a j ((a)). Then, for a A x ,

q
X
j
g j (a) = (a j (x)) = a ( ai )
i=1
j = 1, . . . , p. Let
Lecture 32
185
= a fx h j 1x = f x h j (a).
200
Then the homomorphisms {g j } j = 1, . . . , p, induce a homomorphism

p
q
X
X
g:
av
av
i=1
where V =
pj=1 V j ,
Pr
such that g |
k=1 aU
i=1
Pp
i=1
A x = f x .
Pp
j=1 aU
Pq
i=1 aU
/ S
U
fx f

/0
/ SU
Since S has property (b), there is an open set Y with x Y V,

and homomorphism
p
r
X
X
:
aY
aY
k=1
j=1
such that im = ker. Then

r
r
X
X
Ax = 0
A x = f x
g
k=1
k=1
by exactness. Hence by the previous result, there

P is an open set U with
x U Y, such that the homomorphism g rk=1 aU coincides with
the zero homomorphism. Therefore g induces a homomorphism
p
X
f : aU / im SU .
j=1
Pp
We can identify this quotient ( j=1 aU )/ im with S U so that 201
becomes the natural homomorphism; then f is a homomorphism f :
Lecture 32
186
P
p
S U SU with f = g j=1 au . If s S x since is an epimorphism
Pp
Pp
Pp
there is some element j=1 a j j=1 A x such that s = ( j=1 a j ). Then
p
p
X
X
f (s) = f ( a j ) = g ( a j )
j=1
p
X
= f x (
j=1
a j ) = f x (s);
j=1
i.e., the restriction of f to S x is f x , q.e.d.
Let Y X and either, let x be paracompact and normal and Y

closed, or, let X be hereditarily paracompact and normal. Let S , S
be sheaves of a- modules over X such that S is coherent and S has
property (a). If f : S y Sy is a homomorphism of sheaves of ay
modules, there exists an open set U with Y U, and a homomorphism
g : SU SU such that g|S y = f .
Proof. By the previous result, for each point y Y there is an open set
Vy in X with
x Vy and a homomorphism y : S vy SVy such that

y S = f S . Then by the first result of this lecture, the set of points of
y
202
Vy Y at which y = f is open
in Y. Hence there is a set Wy open in X
with y Wy Vy such that y S Wy Y = f S Wy Y .

We now show that there are systems {Gi }iI and {Hi }iI of open sets
S
of X such that, if G = Gi ,
iI
(i) H i G Gi ,
(ii) the system {Gi } is locally finite in G,
(iii) Y
Hi ,
iI
(iv) each Gi is contained in some Wy .
Lecture 32
187
(1) If X is paracompact and normal and Y is a closed subset of X,

then the covering {X Y, Wy }yY of X has a locally finite refinement {G j } j j , and the covering {G j } jJ can be shrunk to a covering
{H j } jJ with H j G j . Let I J be the set of indices for which
G j Y is not empty. Then, for i I, Gi is not contained in X Y and
S
hence is contained in some Wy . Clearly Y Hi and conditions
iI
(i), (ii), (iii), (iv) are satisfied.
(2) If X is hereditarily paracompact and normal, then G =
Wy is
yY
paracompact and normal. Then there is a locally finite refinement

{Gi }iI of the covering {Wy }yY of G. Since G is normal, the covering
{Gi }iI of G can be shrunk to a covering {Hi }iI with H i G Gi .
The sets Gi , Hi being open in G are open in X. Conditions (i), (ii),
(iii), (iv) are thus satisfied.
Since Gi is contained in
are homomorphisms i :
some Wy , there

G i SG i such that i S G i Y = f S G i Y . Let E i j be the set of 203
points x H i H j G at which i |S x , j |S x ; then Ei j is closed in G.
S
Let E = Ei j ; it is the union of a locally finite system of closed sets in
S
i, j
G, hence is closed in G. Let U = G E, then U is open in G, hence

open in X, and Y U.
Let g : S U SU be defined as follows: For x H i U let
g|S x = i |S x . This gives a consistent definition of g, and g is continuous

(These closed
on each closed set S H U of a locally finite system in S U
i
sets cover S U ). Thus g is a sheaf homomorphism and g|S Y = f .

The above result for the case when X paracompact and normal is
more useful in applications. In particular, the above results are when
both S and S are coherent sheaves of amodules.
Example 35. Let T be the space of ordinal numbers 1 with the topology induced by the order, let Q be the space of ordinal numbers o
and let X = T Q. Then X is compact Hausdorff and hence paracompact
normal. Let Y1 = (T (1 )) o , Y2 = 1 (Q (o )) and Y = Y1 Y2 .
Let a = S = S = Z2 , then S is coherent. Let f : SY SY be the
Lecture 32
188
homomorphism which is the identity on Y1 and is zero on Y2 , There is

no extension of f over an open set containing Y.
Let Y X, and let S be a coherent sheaf of a modules over X. Then
the restriction SY is a coherent sheaf of ay modules.
204
Proof. Since S has property (a), for each y Y there is an open set W
Pp
of X with y W and an epimorphism : i=1 aW SW . Then the
Pp
Pp
restriction
aWY :
aWY SWY is an epimorphism. Thus
i=1
i=1
SY has property (a).
P
To prove that SY has property (b), let ; ki=1 aWY SWY be a
homomorphism where W is an open set in X, and let y W Y. Choose
a section fi : Vi S , (i = 1, . . . , k)y Vi , through hi 1y S y and let
P
T
V = ki=1 Vi . There is a homomorphism : ki=1 aV SV defined by
P
P
P
P
ki=1 ai = ki=1 ai fi (ai ). If ki=1 ai ki=1 Ay , then
P
P
P
P
P
ki=1 ai = ki=1 ai fi (y) = ki=1 ai hi 1y ki=1 hi ai = ki=1 ai .
Since the set of points of Y where = is open in Y, there is
P
an open set G of X with y G V W, such that ki=1 aGY =
P
k aGY . Since S has property (b), there is an open set U with
i=1
y U G, and a homomorphism
:
1
X
aU
i=1
k
X
aU
i=1
such that the sequence

1
X
aU
i=1
k
X
aU SU
i=1
P
is exact. Then if = 1i=1 aUY , the sequence
1
X
i=1
aUY
k
X
i=1
is exact. Thus SY has property (b).
aUY
SUY
Lecture 33
Definition . A sheaf a of rings with unit is called a coherent sheaf of 205
rings if it is coherent as a sheaf of -modules, i.e., it has property (b)
(Property (a) is trivially satisfied.)
If a is coherent and S is a sheaf of a- modules, then S is coherent
if and only if for each point x there is an open set U with x U, and an
exact sequence
l
k
X
X
aU
aU
SU 0.
i=1
i=1
Proof. Necessity. If S is coherent, property (a) implies the existence

of and property (b) implies the existence of .

Sufficiency. Since a is coherent, aU is coherent in U for each open
P
P
P
set U and so are li=1 aU and kj=1 aU . As the image of li=1 aU , im
P
has property (a), and as a subsheaf of kj=1 aU , im has property (b).
Thus im is coherent, and there is an exact sequence
0 im
k
X
aU SU 0.
i=1
Hence, since two of the sheaves are coherent, the third, S )U, is
coherent for a neighbourhood of each point x and hence S is coherent.
Example 36. In the ring B = Z[y, x1 , x2 , . . .] of polynomials in infinitely
many variables with integer coefficients, let I be the ideal generated 206
by yx1 , yx2 , . . . and let A = B/I. Then multiplication by y gives a
189
190
Lecture 33
homomorphism f : A A whose kernel C consists of polynomials

in Z[x1 , x2 , . . .] without constant terms. Then the A- module C is not
finitely generated. Hence if X is a consisting of one point, the constant
sheaf A is not a coherent sheaf of rings.
Example 37. Let X = {x : 0 x 1} and let F be the ring of functions
f : X Z4 for which f (x) = f (1) for x > 0 and f (0) = f (1) mod 2.
There are eight functions in F. Let a be the sheaf of germs of function
in F. Let g be the (constant) global section defined by the function
g : X Z4 where g(0) = 2 and g(x) = 0 for x > 0. The sheaf R
of relations for this section is obtained by omitting from a the germs at
0 of functions f with f (0) = 1 or 3. Then the sections of R over any
connected neighbourhood U of 0 contain only the germs of even valued
functions, hence do not generate RU . Thus R does not have (a1 ) and
hence a has neither (b1 ) nor (b).
The sheaf of germs of analytic functions in the complex plane is a
coherent sheaf of rings.
207
Proof. Let a be the sheaf of germs of analytic functions in the complex

place, and let f1 , . . . , fk be sections of a over a neighbourhood U of
a point zo , i.e., fi is an analytic function in U. We can write fi (z) =
(z zo )n gi (z) where gi does not vanish at zo and hence does not vanish in
TK
Vi . Let R be the
a neighbourhood Vi of zo , zo Vi U. Let V = i=1
sheaf of relations between the sections fi |V, i = 1, . . . , k. We will show
that R is finitely generated in V.
Let P = C[z] be the ring of polynomials in z with complex coefP

ficient and let M be the submodule (over P) of the direct sum ki=1 P
Pk
consisting of elements (p1 , . . . , pk ) for which i=1 pi (z)(z zo )ni = 0.
P
Since P is a euclidean ring and ki=1 P is finitely generated over P, the
submodule M is finitely generated over p. (See van der Waerden , Modj
j
ern Algebra, Vol, I, p. 106). Let (p1 , . . . , pk ), j = 1, . . . , l, be a system of
j
j
j
generators for M. Let ri (z) = pi (z)/gi (z); then each ri (z), i = 1, . . . , k,
Lecture 33
191
is an analytic function in V, and

k
X
i=1
j
ri (z) fi (z)
k
X
pi (z)(z zo )ni = 0
( j = 1, . . . , l).
i=1
Thus for each j, the section determined by (ri , . . . , rk ) is in R. We

will show that these sections generate R.
P
Let ki=1 tiz1 , fiz1 = 0 be a relation between germs at z1 V, i.e., let
P
ti (z), i = 1, . . . , k, be analytic functions at z1 such that ki=1 ti (z) fi (z) = 0.
P
Then ki=1 ti (z)gi (z)(z zo )ni 0. Let n = max(n1 , . . . , nk ) and suppose
that nk = n. We can write
ti (z)gi (z) = (z zo )n qi (z) + si (z),
(i = 1, . . . , k 1)
where qi (z) is analytic at z1 and si (z) 0 of z1 , zo and is a polynomial 208

of degree less than n if z1 = zo . Then
(t1 (z)g1 (z), . . . , tk (z)gk (z)) =
q1 (z)((z zo )n , 0, . . . , 0, (z zo )n1 +
+
qk1 (z)(0, . . . , (z zo )n , (z zo )nk1 + (s1 (z), . . . , sk (z))
where sk (z) is the analytic function defined by

sk (z) = tk (z)gk (z) +
k1
X
qi (z)(z z0 )ni .
i=1
Now, ((z z0 )n , 0, . . . , 0, (z z0 )n1 , etc. are in M and by direct

P
ni
verification we have k1
i=1 si (z)(z z0 ) 0. Since s1 (z), . . . , sk1 (z), are
polynomials, it follows that sk (s)(z zo )n , is a polynomial.
(i) If z1 , z0 , then s1 (z), . . . , sk1 (z) are all zero and
sk (z) = tk (z)gk (z) +
k1
X
qi (z)(z z0 )ni ,
i=1
(z z0 )n sk (z) = tk (z)gk (z)(z z0 )n +
k1
X
i=1
qi (z)(z z0 )n (z z0 )ni
Lecture 33
192
= tk (z)gk (z)(z z0 )n +
k1
X
ti (z)gi (z)(z z0 )ni
i=1
0,
hence sk (z) 0,
P
r
(ii) If z1 = z0 , then sk (z) has a series expansion
r=0 ar (z z0 ) and on
n
multiplication by (z zo ) this series has a finite number of terms.
Hence sk (z) is already a polynomial.
209
In either case, (t1 (z)g1 (z), . . . , tk (z)gk (z)) is a linear combination of

elements of M with coefficients analytic at z1 . Hence
(t1 (z)g1 (z), . . . , tk (z)gk (z)) =
l
X
h j (z)(p1 (z), . . . , pk (z)),
j=1
where h j (z) is analytic at z1 . Then

(t1 (z), . . . , tk (z)) =
l
X
h j (z)(r1 (z), . . . , rk (z)).
j=1
Thus the sheaf R of relations is generated by a finite number of

sections, hence the sheaf a is coherent.
This result is a special case of Okas theorem, Cartan Seminar, 195152, Expose 15, 5. The following proposition on the extension of coherent sheaves is based on Expose 19, 1, of the same seminar.
Proposition 22. Let Y X and either, let X be paracompact and normal
with Y closed, or, letX be hereditarily paracompact and normal. Let a
be a coherent sheaf of rings with unit over X and let S be a coherent
sheaf of aY - modules over Y. Then there is an open set U with Y U,
and a coherent sheaf J of aU - modules over U whose restriction to Y is
isomorphic to S .
Proof. Since S is coherent, there is a covering {V j Y} jJ of Y, where
V j is open in X, and for each j J an exact sequence
lj
X
i=1
aV jY
kj
X
i=1
aV j Y SV j Y 0.
Lecture 33
193

210
From the properties of X there exist system s{Gi }iI and {Hi }iI of
open sets of X such that, if G = iGi ,
(i) H i G Gi ,
(ii) the system {Gi }iI is locally finite in G,
(iii) Y i Hi ,
(iv) each Gi is contained in some V j ,
For the first case, we can assume that each Gi is an F set in X, hence
G and all intersections kr=1 Gir are F -sets and hence are paracompact
and normal. In the second case, all subsets of X are paracompact and
normal.
Since each Gi is contained in some V j , there are exact sequences
lj
X
aGi Y
ki
X
aGi Y SG j Y 0.
r=1
i=1
Either Gi is paracompact and normal with Gi Y closed in Gi or Gi

Pi
is hereditarily paracompact and normal, and the sheaves lr=1
aGi and
Pki
r=1 aG i are coherent. Hence (see Lecture 32) there is an open set G i
with Gi Y Gi Gi and an extension
li
X
G i
r=1
ki
X
aGi
r=1
of i . For the first case we may assume that Gi is also an F - set. Let
Pi
S i = ( kr=1
aGi )/ im . Then, if is the natural homomorphism, the
sequence
ki
li
X
X
aGi S i 0
aGi
r=1
is exact, i.e.,
r=1
211
Lecture 33
194
ki
X
0 im
aGi S i 0
r=1
is exact and hence S i is coherent. There is an isomorphism

gi : SGi Y SGi Y .
i
There are open sets Hi with H i Y Hi , H i G Gi where

S
q
= Gi . For the first case, G and all intersections r=1 Gi are F
iI
212
sets, hence are paracompact and normal.

For i, j I, there is an open set Gi j , Gi Gj Y Gi j Gi Gj , and a

j
j
j
homomorphism fi j : SGi j SGi i j such that fi j SGi j Y = g1
g
j SG i j Y .
i
For the first case, we may assume that Gi j is an F set. Then there
is an
j
open set Gi j with Gi G j Y Gi j Gi j G ji such that fi j f ji SG Y

ij

j
is the identity and f ji fi j SGi j Y . is the identity. Let Ei j = H i H j

(G Gi j ), then Ei j is closed in G .
For i, j, k I there is an open set Gi jk , Gi Gj Gk Y Gi jk

Gi Gj Gk , such that fi j f jk |Gi jk = fik |Gi jk . Let Ei jk = H i H i H k
(G Gi jk ), then Ei jk is closed in G .
S
SS
S
Let E = ( i, j Ei j ) ( i, j,k Ei jk ) and let U = (G E) ( i Hi ).
Then E is closed in G and U is open with Y U. Over each Hi U
there is a sheaf SHi U ; over each Hi H j U there is an isomorphism
i
fij

j
= fi j SHi H U : SH H U SHi H U
i j
i
j
i
j
and fij = ( fij )1 . Further, over each Hi H j Hk U these isomorphisms are consistent, i.e., fij f jk = fik . Then by identification there is
determined a sheaf J over U such that JHi U is identified with SHi U .
j
Then the isomorphisms gi : SHi Y SHi Y induce an isomorphism

i
g : JY S . Since each SHi U is coherent. J is coherent.

i
Bibliography
[1] H.Cartan: Seminaire de L Ecole Normale Superiure, 1948-49,
1950-51, 1951-52, 1953-54.
213
[2] H. Cartan: Varietes analytiques complexes et cohomologie, Colloque de Bruxelles, (1953), pp. 41-55.
[3] H.Cartan and S. Eilenberg: Homological Algebra, Princeton University Press, 1956.
[4] H. Cartan and J.P. Serre: C. R. Acad. Sci. (Paris), 237(1953), pp.
128-130.
[5] S.S.Chern: Complex Manifolds, Scientific Report on the second
summer institute. Bull. Amer. Soc. 62(1956). pp. 101-117.
[6] R. Deheuvels: C. R. Acad. Sci. (Paris), 238 (1954) pp. 1089; 240
(1955), p. 1183.
[7] F. Dolbeault: Formes differentielles et cohomologie sur une varieta
anlytique complex I, Annalso of Math., 64(1956), pp. 83-130.
[8] S. Eilenberg and N. Steenrod: Foundations of algebraic topology,
Princetion University Press, 1952.
[9] I. Fary:Notion axiomatique de lalgebre de cochaines dans la theorie de J. Leray, Bull. Soc. Math. France, 82 (1954), pp. 97-135.
[10] J. Frenkel: C. R. Acad. Sci. (Paris), 241(1955), p.17.
195
214
196
BIBLIOGRAPHY
[11] A. Grothendieck: A general theory of fibre spaces with structue

sheaf, National Science Foundation Research Project, University
of Kansas, 1955.
[12] A. Grothendieck: Theoremes de finitude pour la cohomologie des
faisceaux, Bull. Soc. Math., France 84(1956). pp 1-7.
[13] F. Hirzebruch: Neue Topologische methoden in der algebraischen
geometrie, Springer, 1956.
[14] W.V.D.Hodge and M.F.Atiyah: Integrals of the second kind on an
algebraic variety, Annals of Math., 62(1955), pp. 56-91.
[15] K. Kodaira: On cohomology groups of compact analytic varieties
with coefficients in some analytic faisceaux, Proc. Nat. Acad. Sci.
U.S.A., 39(1953). pp. 865-868.
[16] K. Kodaira: On a differential geometric method in the theory of
analytic stacks, Ibid, pp. 1268-73.
[17] K. Kodaira and D.C.Spencer: On Arithmetic Genera of algebraic
varieties, Proc. Nat. Acad. Sci. U.S.A., 39(1953), PP. 641-649.
[18] K. Kodaira and D.C.Spencer: Groups of complex line bundles over
complex Kahler varieties, Ibid, pp. 868-872.
[19] K. Kodaira and D.C.Spencer: Divisor calss groups on algebraic
varieties, Ibid,pp. 872-877.
215
[20] K. Kodaira and D.C.Spencer: On a theorem of Lefschetz and the

lemma of Enriques-Severi-Zariski, Proc. Nat. Acad. Sci. U.S.A,
39(1953), pp. 1273-1278.
[21] J. Leray: C.R. Acad. Sci. (Paris), 232(1946), p. 1366.
[22] J. Leray: Lanneau spectral etlanneau filtre dhomologie dun espace localement compat etdune application continue, Jour. De
Math.Pures et Appliquees, 29(1950), pp. 1-139.
BIBLIOGRAPHY
197
[23] G. de Rham: Varietes differentiables (Formes, Courants, formes

harmoniques), Paris, Hermann, 1955.
[24] L. Schwartz: Lectures on Complex Analytic Manifolds, Tata Institute of Fundamental Research, Bombay, 1955.
[25] J. P. Serre: Queleques problems globaux relatifs aux varietes de
Stein, Colloque de Bruxells, (1953), pp 57-68.
[26] J. P. Serre: Un theorem de daulite, Com. Math. Helv. 29(1955), pp.
9-26.
[27] J. p. Serre: Faisceaux algebriques coherents, Annals of Math., 61
(1955), pp. 197-278.
[28] N. E. Steenrod: Homology with local coefficients, Annals of
Math., 44(1943), pp. 610-627.
[29] A.Weil: Sur les theorems de de Rham, Comm. Math. Helv., 216
26(1952), pp. 119-145.
[30] O. Zariski: Algebraic Sheaf theory, Scientific Report on the second
summer institute, Bull. Amer. Math. Soc., 62(1956). pp. 177-141.

Lectures On Sheaf Theory: C.H. Dowker

Uploaded by

Copyright:

Available Formats

Lectures On Sheaf Theory: C.H. Dowker

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lectures On Sheaf Theory: C.H. Dowker

Uploaded by

Copyright:

Available Formats

What is a sheaf?

What is a sheaf?

What are the properties of the projection map π in a sheaf?

What are the properties of the projection map π in a sheaf?

Lectures on Sheaf Theory

Tata Institute of Fundamental Research

Lectures on Sheaf Theory

S.V. Adavi and N. Ramabhadran

Tata Institute of Fundamental Research

Definition. A sheaf S = (S , , X) of abelian groups is a map : S

is a continuous open mapping

Proof. Let x X and let G be an open set containing f (x) = O x .

consists of zeros only and which projects homeomorphically onto

Example 3. Let X be the sphere of complex numbers. Let S x be the

A sheaf of rings with unit is a local homeomorphism : S X

Then S = (S , , X) is a sheaf of rings, each S x is a ring with unit, but

[If R is a ring with unit element, we say that M is a unitary, left

p and q. In each S x addition is defined as (p1 , q1 ) + (p2 , q2 ) = (p1 +

gb : S S , gb (p) = b.p for each b B

To prove the assertion, it is enough to show that the continuity of

Example 7. Let C be the ring of complex numbers, S = (S , , X) be

This section will be called the 0-section (zero section).

The zero element of (U, S ) is the 0-section over U. If S is a sheaf of

We can now characterise the sections of S .

We have shown that if f is a section over U then f is a homeomorphism of U onto f (U).

Proof. If f : U S is a section, then h f : U R, the image of

Proof. (1) If G S is open, then G is a union of sections, hence h(G) is

(iii) if s(x) = 0x then, for some open U with x U dom(s)s(U)

Similarly there is a neighbourhood V1 of x with (s1 + s2 )(V1 ) consisting

If h : S R is a homomorphism of sheaves, the set S of p S

(iii) h : S imh is an epimorphism.

If is cofinal in , then i : (A , S ) (A, S ) is an isomorphism.

isomorphism. Similarly i : S S is an an isomorphism.

If {A , S , } and {B , R , } are direct systems indexed by the

Proof. Let h = f : (A , S ) (B, R). Then h = f f =

, S R is an exact sequence of homomorphisms of A

then the sequence of induced homomorphisms of A-modules

Proof. Consider the following commutative diagram:

(i) im g ker f . For if s1 S , for some , s = S 1 , s S . Then

(ii) ker f im g. For if r1 R, then for some , r = r1 , r R . Let

Definition. A bihomomorphism f : (A, R, S ) (B, T ), where A, B are

If A = B and f is the identity, we write f : (R, S ) T .

Let R A S and R A S be two tensor products of R and S over

We define a bihomomorphism : (A, R, S ) (A, Q) as

which proves that im generates Q.

denote the restriction homomorphism. By convention (, a) = o,

aW = a (W) = {ax : x W}, and similarly for sW .

Proposition 3. If {AU , S U , VU } is a presheaf of modules over the space

We remark that aU , sU are sections and that such sections form

If x U and f (U, a), let hxU f = f (x) A x . Similarly, if

is a homomorphism and, if x V U, hxV VU = hxU . Then there is an

and if p Ax with hx p = O, choose a representative f (U, a) for p .

Thus f (aU ) = aU and f is a local homeomorphism, hence is continuous.

tions fU : (AU , S U ) (U, a), (U, S ) (where fU a is the section

a : U A determined by a, and similarly for fU s), determine a homoP

is an exact sequence of homomor- 40

are image j f (U) of sections of S , and this uniquely determines the

0 is exact, where 0 is the

, is exact. That is, for each x,