Algebraic Topology Notes, Part I: Homology
Algebraic Topology Notes, Part I: Homology
Algebraic Topology Notes, Part I: Homology
JONATHAN A. HILLMAN
Abstract. The teaching material that forms this web site is copyright. Other
than for the purposes of and subject to the conditions prescribed under the
Copyright Act, no part of it may in any form or by any means (electronic,
mechanical, microcopying, photocopying, recording or otherwise) be altered,
reproduced, stored in a retrieval system or transmitted without prior written
permission from the University of Sydney.
COPYRIGHT. The University of Sydney 2003 (revised in May 2010).
JONATHAN A. HILLMAN
2. introduction
The most interesting spaces for geometrically minded mathematicians are manifolds, cell-complexes and polyhedra.
Notation.
X
= Y means X is homeomorphic to Y , if X, Y are spaces.
Rn+ = {(x1 , . . . xn ) Rn | xn 0}. Rn+ = {(x1 , . . . xn ) Rn | xn = 0}.
Dn = {(x1 , . . . xn ) Rn | x2i 1}.
S n1 = Dn = {(x1 , . . . xn ) Rn | x2i = 1}.
intDn = Dn \ Dn = Dn \ S n1 . O = (0, . . . , 0).
Exercise 1. (a) Show that Dn \ {O}
= S n1 (0, 1] and Rn \ {O}
= S n1 R.
n
n
(b) Show that D = [0, 1] .
Definition. An n-manifold is a space M whose topology arises from a metric and
such that for all m M there is an open neighbourhood U and a homeomorphism
h : U h(U ) onto an open subset of Rn+ . The boundary M is the set of points m
for which there is such a homeomorphism h with h(m) Rn+ .
We shall show later that the dimension n is well-defined, and that (Rn+ ) = Rn+ ,
so M is an (n 1)-manifold and M = . (See Exercise 17 below.)
FACT. The metric condition is equivalent to requiring that M be Hausdorff (T2 )
and that each connected component of M have a countable base of open sets. Open
subspaces of Rn+ clearly satisfy these conditions, but there are bizarre examples
which demonstrate that these conditions are not locally determined.
Examples. (n = 2): disc, sphere, torus (T ), annulus, M
obius band (M b).
The projective plane P 2 (R) = S 2 /(x x)
= M b D2 . Let S 2 = D+ E D ,
where D+ = {(x, y, z) S 2 | z 12 }, D = {(x, y, z) S 2 | z 12 } and E =
{(x, y, z) S 2 | |z| 21 }. Since the antipodal map interchanges D+ and D and
identifying antipodal points of E gives a M
obius band we see that P 2 (R)
= M bD2
2
is the union of a M
obius band with D .
All surfaces without boundary are locally homeomorphic to each other. We need
a global invariant to distinguish them. Homology provides such invariants. (It is
not so successful in higher dimensions.)
3. projective spaces
Let F = R, C or H (the quaternion algebra). Then F is a skew field and has
finite dimension as a real vector space. Let d = dimR F (= 1, 2 or 4).
Given a point X = (x0 , . . . xn ) F n+1 \ {O}, let [x0 : : xn ] be the line
through O and X in F n+1 . Let P n (F ) (or F P n ) be the set of all such lines
through O in F n+1 . Two nonzero points determine the same line if and only if they
are proportional, i.e., [x0 : : xn ] = [y0 : : yn ] if and only if there is a
F = F {0} such that yi = xi for 0 i n. Hence P n (F ) = (F n+1 \ {0})/F .
Each line through O in F n+1
= R(n+1)d passes through the unit sphere S (n+1)d1 ,
and two points on the unit sphere determine the same line if and only if one is a
multiple of the other by an element of S d1 (= {1}, S 1 or S 3 ), the subgroup
of elements of F of absolute value 1. Thus there is a canonical surjection from
S (n+1)d1 to P n (F ), and P n (F ) = S (n+1)d1 /S d1 is the orbit space of a group
action. In particular, P n (R) is obtained from the n-sphere by identifying antipodal
points.
JONATHAN A. HILLMAN
union of the spheres, with the equatorial identifications given by s n+1 (s) for
all s S n . We give S the topology for which a subset A S is closed if and
only if A S n is closed for all n. Then S is a CW complex, with two cells in each
dimension, represented by Dn and S n \ n (Dn ), for n 1.
5. polyhedra and simplicial homology
Definition. An affine q-simplex in RN is the closed convex set determined by
q + 1 affinely independent points. The standard q-simplex q is determined by the
standard basis vectors in Rq+1 . Thus
q = {(x0 , . . . xq ) Rq+1 | xi = 1, xj 0 j}.
Clearly any affine q-simplex is homeomorphic to q and so is a compact metric
space.
Definition. The faces of the affine simplex determined by {P0 , . . . Pq } are the
affine simplices determined by subsets of this set.
Definition. A polyhedron in RN is a subset P which is the union of finitely many
affine simplices (of varying dimensions).
(We may assume that any two simplices meet along a common face, possibly
empty.)
Definition. A triangulation of a space X is a homeomorphism h : K X from
some polyhedron K (in some RN ).
In other words, a triangulation is a representation of X as a finite union of closed
subsets, each homeomorphic to a simplex.
A polyhedron is a special case of a cell complex in which all the defining maps
are one-to-one. Conversely, any such cell complex admits a triangulation as a
polyhedron, but in general more simplices are needed than cells.
Example. 3 = {(x0 , . . . x3 ) R4 | xi = 1, xi = 0, xj 0 j} is a union
of four 2-simplices, and is homeomorphic to the 2-sphere S 2 , which has a cellular
structure with just two cells. The torus S 1 S 1 has a cellular structure with just
four cells. Its simplest triangulation requires seven vertices.
As each affine simplex is determined by its vertices, a polyhedron is determined
by the set of all vertices together with the set of finite subsets corresponding to
the simplices (and their faces). This is essentially combinatorial data, and there
is a purely combinatorial notion of simplicial complex, which makes no mention of
topology. (See [Spanier].) How can we use this to extract invariants of spaces which
are insensitive to the triangulation used?
Euler observed that given any triangulation of the 2-sphere S 2 into V vertices, E
edges and F faces (i.e., 0-, 1- and 2-simplices, respectively) then V E +F = 2.
This is easily proven by induction on V , once you show that any two triangulations
have a common refinement. Riemann and Betti were lead to consider the number
of q-dimensional holes in a space. It was only later that it was realized that these
Betti numbers could be interpreted as the dimensions of certain vector spaces
(or ranks of certain abelian groups), whose alternating sum gives an extension of
Eulers invariant.
6. an example of simplicial homology
A q-dimensional hole in a polyhedron may perhaps be defined informally (in
terms of what surrounds it) as a union of q-simplices which has no boundary -
JONATHAN A. HILLMAN
whose (q 1)-dimensional faces match up in pairs. We want to ignore such qcycles which bound (q + 1)-simplices. From unions of q-simplices to formal
sums is not a great step. With the proper linear analogue of boundary - - we
have = 0. We are lead to look at formal sums of q-simplices with = 0 and to
factor out ((q + 1)-simplex). Thus we define Hq (P ) = Ker()/Im() as a measure
of the q-dimensional holes.
Example. S 2 = 3 may be triangulated as a tetrahedron, with vertices A, B, C, D,
edges AB, AC, AD, BC, BD, CD and faces ABC, ABD, ACD, BCD. We take these
as bases of vector spaces over a field R, of dimensions 4, 6 and 4, respectively. The
boundary maps are given by 0 V = 0 for vertices V , 1 V W = W V for edges
V W and 2 V W X = W X V X + V W for faces V W X. These may be extended
to linear maps betwen the vector spaces. We find that Ker(2 ) is 1-dimensional,
generated by BCD ACD + ABD ABC, Im(2 ) = Ker(1 ) is 3-dimensional and
Im(1 ) is 3-dimensional, generated by the differences B A, C A and D A.
Thus H2 (S 2 )
= R, H1 (S 2 ) = 0 and H0 (S 2 )
= R. (In fact, we could allow R to be
any ring, provided we modify the terminology of vector spaces and dimension.)
Exercise 2. The triangulation of S 2 as an icosahedron (with 12 vertices, 30 edges
and 20 faces) is is invariant under the antipodal map. (See figure.) Identifying
opposite sides gives a triangulation of P 2 (R) with 6 vertices, 15 edges and 10 faces.
Let R be a commutative ring and form the simplicial chain complex with coefficients
R of this polyhedron. Compute the homology. Note in particular what happens if R
is (a) a field of characteristic 6= 2, e.g., the real numbers R; (b) the 2-element field
F2 = Z/2Z; (c) the integers Z.
E
B
D
A
A
E
(The vertices A , B and C are behind the faces ABC, ADE and AEF , respectively, and the edges A B , A C , A D , A E , A F , B C , B D, B E, B F , C D ,
C E , C F, D E and E F are not shown.)
Many interesting spaces cannot be triangulated, and there are others which admit
several essentially distinct triangulations with no common refinement. Thus it is
not clear that simplicial homology provides a useful topological invariant. We shall
have to modify our definition.
7. an algebraic interlude
The main algebraic prerequisite for Homology is linear algebra. We shall ultimately want to consider modules over a ring R, but you may assume that R is a
field and the modules are vector spaces, for much of what follows.
The key notions that you should understand are homomorphism, kernel, image,
submodule (or sub-vectorspace), quotient module (or quotient vectorspace) and
cokernel.
y
y
y
y
y
B1 B2 B3 B4 B5
Let pi : Ai Ai+1 and qj : Bj Bj+1 be the horizontal maps. Show that (i) if
is onto and and are 1-1 then is 1-1; (ii) if and are onto and is 1-1 then
is onto. Identify clearly where each hypothesis is used.
[Hint for (i): suppose that a A3 and (a) = 0. Show p3 (a) = 0 hence a = p2 (a )
for some a A2 . And so on ... Part (ii) is similar.] The argument applies without
change to modules over a ring.
8. exact sequences of complexes
A sequence of chain complexes C D E is exact if each of the corresponding sequences of modules Cn Dn En is exact at Dn , for all n. If
0 C D E 0 is a short exact sequence of chain complexes then there
are connecting homomorphisms from Hn (E ) to Hn1 (C ), for all n, giving rise to
a long exact sequence of homology:
. . . Hn+1 (E ) Hn (C ) Hn (D ) Hn (E ) Hn1 (C ) . . .
Moreover the connecting homomorphisms are natural transformations: a morphism of short exact sequences of complexes gives rise to a commuting diagram
with two parallel long exact sequences of homology.
In particular, if C is a subcomplex of D (i.e., if each homomorphism : Cq
Dq is a monomorphism) then the differentials induce differentials on the quotient
modules Dq /Cq , and we obtain a short exact sequence
0 C D (D/C) 0.
JONATHAN A. HILLMAN
The special case in which the complexes C , D and E are each trivial except
in degrees 1 and 0 is known as the Snake Lemma, and then asserts that there is a
six-term exact sequence
0 Ker(1 ) Ker(1 ) Ker(1 ) Cok(1 ) Cok(1 ) Cok(1 ) 0.
Conversely, the Snake Lemma can be used to establish the long exact sequence in
general.
Exercise 4. Show that the long exact sequence of homology determined by a short
exact sequence of chain complexes is indeed exact.
We shall later define the notion of chain homotopy between chain homomorphisms, and take the linear duals to obtain cochain complexes and cohomology.
9. euler characteristic
Suppose now that R is a field and that all the vector spaces Cn are finite dimensional, and are 0 for all but finitely many values of n. The alternating sum of the
dimensions (C ) = (1)n dimR Cn is then a well-defined integer, called the Euler
characteristic of C .
Exercise 5. Show that
(C ) = (1)n dimR Hn (C ) = (1)n (dimF Ker(n ) dimF Im(n+1 )).
(This is an elaboration of the formula dimR N = dimR M + dimR P , if M is a
subspace of N with quotient P .)
Exercise 6. Show that any triangulation of the torus T requires at least 7 vertices.
[Hint: Use the Euler characteristic, and note each edge is determined by its
vertices, each edge is common to two faces and each face is triangular.]
Exercise 7. Generalize (5) to the other closed surfaces.
(For this you need to know that every such surface may be obtained from S 2
by replacing g 0 disjoint closed discs with copies of the punctured torus T0 =
T intD2 in the orientable case, or by replacing c 1 disjoint closed discs with
copies of the M
obius band in the nonorientable case.)
Exercise. Let G be a finite graph in which no pair of vertices is connected by more
than one edge. Suppose that G is embedded in a surface S and let {Fi | i I} be
the set of components of S G. Show that (S) = V E + iI (Fi ), where G
has V vertices and E edges.
10. singular homology
Let S be a set and R a ring. A function f : S R is 0 a.e. if f (s) = 0 for all
but finitely many s S. For each s S let es : S R be the function defined by
es (s) = 1 and es (t) = 0 if t 6= s. Then es is 0 a.e. The free R-module with basis S
is the set R(S) = {f : S R | f is 0 a.e.}, with the obvious R-module structure.
Every element in this module is uniquely expressible as a linear combination of
the basis elements es . (In fact f = sS f (s)es .) It has the following universal
property: homomorphisms from R(S) to an R-module M correspond bijectively
to functions from S to (the underlying set of) M .
10
JONATHAN A. HILLMAN
particular, this number is independent of the coefficient field, and moreover does
not depend on how X is represented as a cell complex.
11. functoriality, relative homology, long exact sequence
Let f : X Y be a map. If : q X is a singular q-simplex in X then f
is a singular q-simplex in Y . Thus f determines homomorphisms
Cq (f ) : Cq (X) Cq (Y ),
which are compatible with the differentials (qY Cq (f ) = Cq1 (f )qX ) and so together
give a chain homomorphism C (f ). It is easily verified that C (idX ) = idC (X) and
C (gf ) = C (g)C (f ), i.e., that the chain complex construction is functorial. Such
chain homomorphisms induce homomorphisms
Hq (f ) : Hq (X) Hq (Y ),
which again are functorial.
Exercise 12. Verify that Hq is a functor, i.e., that Hq (1X ) = 1Hq (X) and Hq (f g) =
Hq (f )Hq (g) for all q.
We shall need to consider also relative homology, for pairs (X, A), where A is a
subspace of X. The inclusion of A into X induces natural monomorphisms from
Cq (A) to Cq (X), for all q. The quotient module Cq (X, A) = Cq (X)/Cq (A) is the
module of relative q-chains. Thus we have a short exact sequence of chain complexes
0 C (A) C (X) C (X, A) 0,
and a corresponding long exact sequence of homology for the pair (X, A).
We may describe the relative homology modules more explicitly as follows. Let
Zq (X, A) be the module of q-chains c on X such that c is in the image of Cq1 (A),
and let Bq (X, A) be the submodule generated by Cq+1 (X) and the image of Cq (A).
Then it follows easily from the standard isomorphism theorems of linear algebra
that Hq (X, A)
= Zq (X, A)/Bq (X, A).
We shall see that if A is a well-behaved subset of X then the relative homology
depends largely on the difference X \ A; in this way we can hope to analyze the
homology of a space by decomposing the space into simpler pieces.
We have shown that homology is functorial, the homology of a 1-point space is
concentrated in degree 0, and that there is a long exact sequence of homology corresponding to any pair of spaces (X, Y ). There remain two fundamental properties
of homology that we must develop: homotopy invariance and excision. Establishing these very important properties is rather delicate, and so we shall state them
first, and defer the proofs, so that we can get on with computing the homology of a
variety of spaces. (It can be shown that singular homology for cell-complexes may
be characterized axiomatically by the properties listed in this paragraph.)
12. homotopy
If f : X Y is a homeomorphism then (for all n) Hn (f ) is an isomorphism,
since it has inverse Hn (f 1 ). [As we have seen, Hn (f ) Hn (f 1 ) = Hn (f f 1 ) =
Hn (idY ) = idHn (Y ) , and similarly Hn (f 1 ) Hn (f ) = idHn (X) .)
In fact much weaker conditions on f imply that the induced homomorphisms
are isomorphisms. We shall see that homology is a rather coarse invariant, in that
it cannot distinguish Rn from a point.
11
12
JONATHAN A. HILLMAN
n
n
Example. Let D+
= {(x0 , . . . xn ) S n | xn 0} and D
= {(x0 , . . . xn ) S n |
n
n
n
n
n
n1
xn 0}. Then D+ D = S and D+ D = S
. Let N = (0, . . . 0, 1) S n
be the north pole.
n
n
Claim (using excision; proven later): the inclusion of the pair (D+
, D+
\ {N }) into
n
n
(S , S \ {N }) induces isomorphisms on all homology groups.
n
n
Now D+
and S n \ {N } are both contractible, while D+
\ {N } S n1 . On
applying the long exact sequences of homology for these spaces and making use of
the isomorphisms coming from excision and homotopy equivalences we find that
Hq (S n )
= Hq1 (S n1 ), provided q 2. If q = 1 we get an exact sequence
n
0 H1 (S n ) H0 (S n1 ) H0 (D+
).
One might suppose that S is the unit sphere in 2 , the (countably) infinite
dimensional analogue of Cn . However the obvious map is not a homeomorphism.
Let U = {x 2 | ||x|| = 1}, and let U0 be the subset of points with almost all
coordinates 0. The natural map from S to U0 is a continuous bijection. However
the subset C = { nn2 +1 (e1 + n1 en ) | n 2} is closed in S , but is not closed in U0 ,
and so this bijection is not a homeomorphism. On the other hand U0 is a proper
subset of U , which is dense in U . Are U0 or U contractible?
13. homotopic maps induce the same homomorphism
Theorem 1. Let f and g be homotopic maps from X to Y . Then Hq (f ) = Hq (g)
for all q.
Proof. We shall show that this follows by functoriality from a special case. Let
F : X [0, 1] Y be a homotopy from F0 = f to F1 = g, and let it : X X [0, 1]
be the map defined by it (x) = (x, t) for all x X and 0 t 1. Then Ft = F it ,
so Hq (f ) = Hq (F )Hq (i0 ) and Hq (g) = Hq (F )Hq (i1 ), for all q. Thus it shall suffice
to show that Hq (i0 ) = Hq (i1 ), for all q. In other words, we may assume that
Y = X [0, 1] and F = idY (so that Ft = it ).
13
We shall define prism operators Pq : Cq (X) Cq+1 (X [0, 1]) such that
X[0,1]
Cq (i1 ) Cq (i0 ) = q+1
Pq + Pq1 qX . Thus if is a singular q-cycle on X
X[0,1]
14
JONATHAN A. HILLMAN
1
1
Hq ( Dn , Dn \ {O}) Hq (Dn , Dn \ {O})
2
2
are isomorphisms, by excision, while Hq (Dn , S n1 )
= Hq (Dn , Dn \ {O}), by homotopy invariance.
O
A
X = A f eq
1 q
2D
U = X \ 21 Dq
Im(f )
15
on (I, I) and its homology class generates H1 (I, I) = Z. (Consider the long
cI ] = [1] [0]
exact sequence of homology for the pair (I, I), and observe that [id
=
d
.
b
3. Let
(t) = (1 t) and let : 2 X be the singular 2-simplex defined
by (x, y, z) = (y). For any q 0 let q be the constant singular q-simplex with
b 1 +
b +
value (0). Then ( + 2 ) =
b + 1 1 + 1 =
b.
4. Let P be a polygon with n sides j , for 1 j n (numbered consecutively).
Then = c
j is a 1-cycle on P
= Z. For let
= S 1 , which generates H1 (P )
16
JONATHAN A. HILLMAN
U = P \ int1 . Then H1 (P )
= H1 (P, U )
= H1 (1 , 1 ), and the images
of and
c1 in H1 (P, U ) agree. (Here
c1 is considered as a relative 1-cycle on
(1 , 1 )
= (I, I).)
5. Similarly, if (t) = e2it for 0 t 1 then b
generates H1 (S 1 ). Moreover, if
2i(j1+t)/n
j (t) = e
, for 1 j n then [b
] = [c
j ] in H1 (S 1 ).
1
1
th
6. Let let fn : S S be the n power map: fn (z) = z n for all z S 1 . Since
f0 is the constant map H1 (f0 ) = 0. If n 1 then fn j = for 1 j n. Hence
H1 (fn )([b
]) = [f[
] in H1 (S 1 ).
n j ] = n[b
If n < 0 then fn j = for 1 j |n|. Hence H1 (fn )([b
]) = |n|[b] = n[b
]. Thus
fn induces multiplication by n on H1 (S n ), for all n.
7. Generalization (not proven). Let P be a convex polyhedron in Rn , whose
faces are triangulated as (n 1)-simplices. Then if these faces are consistently
oriented their formal sum represents a generator of Hn1 (P )
= Hn1 (S n1 )
= Z.
Exercise 22. The fundamental theorem of algebra. Let P (z) = z n + an1 z n1 +
... + a0 be a polynomial of degree n 1 with complex coefficients. Show that if r is
large enough Pt (z) = (1 t)P (z) + tz n has no zeroes on the circle |z| = r for any
0 t 1. Hence the maps : z z n and : z P (rz)/|P (rz)| are homotopic as
maps from S 1 to S 1 . If P has no zeroes the latter map extends to a map from the
unit disc D2 to S 1 . CONTRADICTION. Why?
17. wedge, bouquets, the torus
Let X and Y be spaces with given basepoints x0 , y0 . The one-point union of
X and Y is X Y = (X Y )/(x0 = y0 ). (Thus S 1 S 1 is the figure eight.) The
common image of x0 = y0 represents a natural basepoint for X Y .
Assume that the base-points have contractible neighbourhoods in X and Y ,
respectively. If X and Y are path-connected then so is X Y , and Hq (X Y )
=
Hq (X) Hq (Y ) for all q > 0.
Example. r S 1 is defined inductively. It may also be obtained by adjoining r
1-cells to a point.
The torus T = S 1 S 1 may be constructed by adjoining a 2-cell to S 1 S 1 . More
precisely, we may construct T by identifying opposite sides of a rectangle R. Let
U Y R be proper subrectangles. Let X = T \ U . Then T = X Y , X Y =
Y \ U S 1 , X S 1 S 1 and Y is contractible. Since Hq (T, X)
= Hq (Y, X Y )
(by excision of T \ Y ), we get Hq (T ) = 0 if q > 2. We also obtain an exact sequence
0 H2 (T ) H1 (X Y ) H1 (X) H1 (T ) 0.
Let A1 , A2 , A3 , A4 be the vertices of R and K, L, M, N be the vertices of Y .
Then H1 (X Y ) is generated by the image of KL + LM + M N + N K, which
is homologous in R U to A1 A2 + A2 A3 + A3 A4 + A4 A1 , and hence represents
a + b a b = 0 in H1 (X). Hence the homomorphism from H1 (X Y ) to H1 (X)
induced by the inclusion is trivial, and so H1 (T )
= Z2 and H2 (T )
= Z.
The other surfaces may also be obtained by adjoining a single 2-cell to a wedge
of circles. In particular, T T = 4 S 1 e2 may be obtained from a regular octagon
in the hyperbolic plane by identifying sides in pairs. This construction generalizes
further: g T 2g S 1 e2 . (See the figure on the next page.)
For simple higher dimensional examples we may look at products of spheres:
q
p
p
q
since S p S q = (S p D
)(D
S q )(D+
D+
), we have S p S q S p S q ep+q .
a1
b2
a1
a2
b1
a2
a1
b2
b1
17
b1
b2
a2
is
=
=
18
JONATHAN A. HILLMAN
the degenerate cases when the vertices {B, A0 , . . . Aq } lie in some q-dimensional
hyperplane. (In particular, we allow B C.)
Joining extends to linear combinations of affine simplices in an obvious way.
Lemma 4. If C is a q-chain with q 1 then (B C) = C B C. If C = ri Pi
is a 0-chain then (B C) = C (ri )B.
Proof. It suffices to assume that C is a q-simplex. The result is geometrically clear
(see figure); the only point to check is that the signs are correct.
B
BC =
B C =
C
1
1
Definition. The barycentre of q is Bq = ( q+1
, . . . , q+1
).
We shall define the subdivision of q inductively, extending the subdivision of
its boundary by joining with Bq .
B2
19
X=
A=
20
JONATHAN A. HILLMAN
Theorem 10. Let X be a space with a subspace A, and let U A. If the closure of
U is contained in the interior of A then the inclusion of (X \ U, A \ U ) into (X, A)
induces isomorphisms on relative homology, Hq (X \ U, A \ U ; R)
= Hq (X, A; R),
for all q and any coefficients R.
Proof. 1-1: Suppose z is a singular q-chain on X \ U such that z is a (q 1)-chain
on A \ U and that z is a relative boundary in (X, A), i.e., that there is a singular
q-chain z on A and a singular q + 1-chain w on X such that z = z + w. Then
(Sd)n (z) = (Sd)n (z ) + (Sd)n (w). If (Sd)n (w) is small of order {X \ U , intA},
say (Sd)n (w) = wX\U + wA then (Sd)n (z) wX\U = (Sd)n (z ) + wA . Since the
LHS is a chain on X \ U and the RHS is a chain on A both sides are chains on
A U . Hence (Sd)n (z) = ((Sd)n (z) wX\U ) + wX\U is a relative boundary in
(X \ U, A \ U ).
onto: similar.
A knot is an embedding K of one sphere inside another, and a link is an embedding L of several disjoint spheres. These are usually studied through invariants
of their complements. The purpose of the next exercise is to show that homology
alone gives little information!
Let K : S n S n+k be an embedding (homeomorphism onto its image). To
avoid pathologies, we asume that the image of K has a product neighbourhood
S n Dk . Let X = S n+k \S n intDk . Then X is a compact manifold with boundary
X = S n Dk = S n S k1 . Moreover the inclusion of X into S n+k \S n 21 Dk is
a homotopy equivalence. (This is not immediately obvious, but should be plausible.)
Exercise. Assume that n, k > 0. Use excision (and homotopy invariance) to show
that Hi (X) = 0 unless i = 0 or k 1, in which case Hi (X; R)
= R.
[Hints. (1) S n+k = Dn+k+1 = Dn+1 Dk , so S n+k = S n Dk Dn+1 S k1 ;
(2) X S n+k \ {P } S n+k , where P is any point in S n+k X.]
What happens if n = 0 or k = 0? Can you see what happens for an embedding
L : S n S n+k ? (The result holds for embeddings of S n (without such product
neighbourhoods) but this extension is usually done via Alexander duality.)
20. spectral theory for orthogonal matrices
Let hx, yi = xi yi be the standard inner product on Rn and hv, wi = vi w
i
be the standard inner product on Cn . If U is a subspace of Rn its orthogonal
complement is U = {x Rn | hx, ui = 0 u U }. The inner product on Rn
induces inner products on subspaces, and Rn
= U U is the orthogonal direct
n
sum. (Similarly for subspaces of C .)
A matrix A GL(n, R) is orthogonal if hAx, Ayi = hx, yi for all x, y Rn . A
matrix B GL(n, C) is unitary if hBv, Bwi = hv, wi for all v, w Cn . Any real
matrix A GL(n, R) acts on Cn in a natural way, and so we may view GL(n, R)
as a subgroup of GL(n, C). Under this identification, the orthogonal matrices are
exactly the real unitary matrices.
The advantage of working over the complex numbers is that we can always find
eigenvalues and eigenvectors. Let u be an eigenvector of the unitary matrix B
corresponding to an eigenvalue and let U = Cu. We may assume that hu, ui =
1. Since B is unitary it also maps U to itself, and so preserves the direct sum
decomposition Cn
= U U . Hence we may prove by induction on n that Cn
21
22
JONATHAN A. HILLMAN
23
24
JONATHAN A. HILLMAN
25
26
JONATHAN A. HILLMAN
26. cohomology
If we consider the linear duals of our chain complexes, we obtain a more powerful
invariant, the cohomology ring of a space. As we do not have time to do justice to
this, I shall just sketch the easy bits here, and indicate in the final section below an
alternative construction of the real cohomology ring for open subsets of Rn , using
differential forms.
Given a chain complex C , we define the associated cochain complex C by
C q = HomR (Cq , R), with codifferential q (f ) = f q+1 . The cohomology modules
H q (C ) are the quotients Ker( q )/Im( q1 ). On applying this construction to the
singular chain complexes of (pairs of) spaces, we obtain contravariant functors:
if f : (X, A) (Y, B) then H q (f ) : H q (Y, B) H q (X, A). A pair of spaces
determines a long exact sequence of cohomology, and cohomology satisfies homotopy
and excision. The cohomology of a point is H ({P }; R) = R, 0, 0, . . . .
In degree 0 the cohomology module H 0 (X; R) may be identified with the Rvalued functions on X which are constant on components. (This is the first indication that cohomology may be related to functions and differential forms, since a
function f on an open subset of Rn is locally constant if and only if df = 0.)
Let ev : H q (C ; R) HomR (Hq (C ; R), R) be the evaluation homomorphism,
defined by ev([f ])([c]) = f (c). The quotient Cq /Zq
= Bq1 is a submodule of the
free module Cq1 . Therefore Cq /Zq is a free module if R = Z or is a PID. (This
is a standard algebraic fact, but is not trivial.) Hence any homomorphism from Zq
to R extends to a homomorphism from Cq to R. It follows that ev is always an
epimorphism. It is an isomorphism if R is a field.
The ring structure arises ultimately from the diagonal map X : X X X.
The ring H (X; R) is a graded R-algebra, and is graded-commutative: =
(1)pq . (Here denotes the product.) In particular, H (S n ; Z)
= Z[n ]/(n2 ),
n
n
where n generates H (S ; Z).
Example. Let h : S 3 S 2 = P 1 (C) be the Hopf fibration, given by h(u, v) =
[u : v] for all (u, v) S 3 , and let c : S 3 S 2 be the constant map. Then
S 2 S 4 = S 2 c e4 and P 2 (C) = S 2 h e4 . It is easy to see that S 2 S 4 and
P 2 (C) have isomorphic homology and cohohomology groups. However their cohomology rings are not isomorphic. In fact H (S 2 S 4 ; Z)
= Z[2 , 4 ]/(2 , 4 )2 and
3
H (P (C); Z) = Z[2 ]/(2 ). Hence these spaces are not homotopy equivalent. In
particular, h is not homotopic to a constant map, although H (h) = H (c).
27
by the submodule generated by the elements r[m, n] [rm, n], r[m, n] [m, rn],
[m + m , n] [m, n] [m , n] and [m, n + n ] [m, n] [m, n ], for all r R,
m, m M and n, n N . Let m n denote the image of [m, n] in M R N .
The obvious function from M N to M R N which sends (m, n) to m n is
bilinear. This is in fact the universal bilinear function; if P is another R-module
and b : M N P is bilinear then there is an unique linear map b : M R N P
such that the obvious diagram commutes.
This construction and those that follow are suitably functorial.
Examples: R a field. Rm R Rn
= Rmn .
28
JONATHAN A. HILLMAN
.
This
pairing
is
nondegenerate
and
so
H
(X)
= H 1 (X; R).
= H1 (X; R)
DR
q
q
More generally, HDR (X) = H (X; R) (for reasonable open subsets X).
Any finite cell complex is homotopy equivalent to an open subset of some Rn .
Thus the De Rham approach applies in considerable generality. De Rham cohomology may also be defined directly for a smooth n-manifold, without first choosing
some embedding in an euclidean space. We illustrate this for the circle S 1 = R/Z.
(Note that as S 1 is compact it is not diffeomorphic to an open subset of any Rn .)
Example. Let S = {f : R R | f is C , f (x + 1) = f (x)x R} be the set of
periodic C functions. Let = RSdx and let d : S be given by df = f dx.
Then dx is closed but not exact. S 1 dx = 1 6= 0.
Sometimes we can get away with much smaller complexes. The circle is given
by polynomial equations: S 1 = {(x, y) R2 | x2 + y 2 = 1}. Let S = R[x, y]/(x2 +
y 2 1) = R[x][y]/(y 2 = 1 x2 ). Since x2 + y 2 = 1 we should have 2xdx + 2ydy = 0,
and so we set = (Sdx Sdy)/S(xdx + ydy)
= Sdx R[x]dy. Then dx dy =
(x2 + y 2 )dx dy = 0 (on using xdx = ydy, etc.). Hence = 0. We find
q
that Ker(d) = R.1 and /dS
= R.[xdy]. Thus HDR (S 1 )
= R for q = 0, 1 and is 0
otherwise.
When n = 3 we may relate exterior derivation of forms to the familiar operations
of vector calculus. Let i, j and k denote the standard unit vectors in R3 . A C
vector field on X is a vector-valued function f i+gj+hk with coefficients in S. Let
V denote the S-module of all such vector
fields. Then
V is a free S-module of
rank 3, with basis {i, j, k}. Since 32 = 3 and 33 = 1 the spaces of 1-forms
and 2-forms on R3 are each free of rank 3, while the space of 3-forms is free of
rank 1. There are obvious isomorphisms from 1 (X) and 2 (X) to V, sending
dx, dy, dz and dy dz, dz dx, dx dy to i, j, k (respectively), and 3 (X)
= S via
f dx dy dz 7 f . Under these identifications wedge product corresponds to cross
product of vector fields and the exterior derivative in degrees 0, 1 and 2 corresponds
to grad, curl and div.
1
If X is a simply connected region of R3 then HDR
(X) = 0, so curl(V ) = 0 if
and only if V = grad(f ) for some function f .
2
If HDR
(X) = 0 then div(W ) = 0 if and only if W = curl(V ) for some V .
Example. X = R3 {O}. V (x) = (m/|x|3 )x. V = d(m/|x|).
(Remark. Vector fields and 1-forms are dual objects: a 1-form may be evaluated on a vector field v to get a function (v). In a proper development of differential
geometry a Riemannian metric is used to determine an isomorphism between these
modules. Changing the metric changes the isomorphism. The identification of vector fields with 1-forms for open subsets of R3 given above is the one determined by
the standard euclidean inner product on R3 .)
29
(8)
30
JONATHAN A. HILLMAN
(3) The abelianization functor ab from ((Grp)) to ((M odZ )) which sends a
group G the the abelian group Gab = G/G and a homomorphism f : G
H to the induced homomorphism f ab : G/G H/H .
(4) If A is an object of A there is a covariant functor from A to ((Set)) sending an object B to the set HomA (A, B) and a morphism f : B C to
HomA (A, f ), or f for short, such that f (g) = f g for all g HomA (A, B).
Similarly there is a contravariant functor sending B to HomA (B, A) and f
to f such that f (g) = gf .
(5) Every contravariant functor from C to D can be thought of as a covariant
functor from the dual category C op to D.
Definition. A natural transformation between functors F and G from A to Z is
a rule which associates to each object A of A a morphism (A) : F (A) G(A) in
Z such that for all morphisms f : A B in A we have
(B)F (f ) = G(f )(A).
Example. Let V = ((M odR )) be the category of finite dimensional real vector
spaces and D be the contravariant functor from V to itself which sends a vector
space V to its dual D(V ) = HomR (V, R) (considered as a vector space in the usual
way). Then DD is naturally isomorphic to the identity functor, i.e., there is a
natural transformation from IV to DD such that (V ) is an isomorphism for all
objects V of V. (It is defined by (V )(v)(f ) = f (v) for all v V and f D(V ).)
There is no such natural transformation from IV to D, although each vector
space has the same dimension as its dual and so is isomorphic to it.
School of Mathematics and Statistics, University of Sydney, NSW 2006, Australia
E-mail address: [email protected]