Professional Documents
Culture Documents
2010
c
M.
A. Aguilar and C. Prieto
ii
Table of Contents
Contents
Preface
ix
1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2
General Denitions . . . . . . . . . . . . . . . . . . . . . . . .
1.3
Further Examples . . . . . . . . . . . . . . . . . . . . . . . . .
1.4
Homotopy Lifting . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5
1.6
1.7
1.8
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.8.1
Covering Maps . . . . . . . . . . . . . . . . . . . . . . 44
1.8.2
Spherical Fibrations . . . . . . . . . . . . . . . . . . . 48
1.8.3
2 Fiber Bundles
51
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2
Topological Groups . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3
Fiber Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.1
2.4
Tangent Bundles . . . . . . . . . . . . . . . . . . . . . 69
Coordinate Transformations . . . . . . . . . . . . . . . . . . . 72
iv
Table of Contents
2.4.1
2.5
Vector Bundles . . . . . . . . . . . . . . . . . . . . . . 79
Principal Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5.1
Stiefel Manifolds . . . . . . . . . . . . . . . . . . . . . 88
2.6
2.7
Induced Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.7.1
2.8
2.9
Functional Bundles . . . . . . . . . . . . . . . . . . . . 96
Universal Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 99
2.8.1
2.8.2
2.9.2
117
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.2
3.3
3.4
3.2.1
3.2.2
3.3.2
3.3.3
3.3.4
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.4.1
3.4.2
3.4.3
Table of Contents
4.1
4.2
vii
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.1.1
4.1.2
4.1.3
4.1.4
. . . . . . . . . . . . . . 156
4.2.2
References
173
Alphabetical Index
175
viii
Table of Contents
Preface
The purpose of this book is the study of ber bundles. The concept of a ber
bundle is one of those ubiquitous concepts in mathematics. Its rst appearance was probably in the thirties of the last century within the study of the
topology and geometry of manifolds. However it was not until the publication of Norman Steenrods book [15] in 1951 that a systematic treatment of
the concept was given. In the meantime some other books not many on
the subject have appeared. Worthy to be mentioned is Sir Michael Atiyahs
book [?] on K-Theory, where special ber bundles are studied, namely the
vector bundles, which constitute the basis for dening K-theory.
We start this book in a very general setup, where we dene as bration
just a continuous map p : E B. Thereon we begin to put some requirements to p and according to those, we put an adjective like Serre bration
for those maps p which have the homotopy lifting property for cubes, or
Hurewicz bration if the maps p have the homotopy lifting property for all
spaces. We also have the locally trivial brations, which are always Serre
brations. They are even Hurewicz brations whenever the base space B is
paracompact. A special case are the covering maps, which are locally trivial
brations whose bers are discrete spaces.
This book was inspired in the notes of a course given by Dieter Puppe in
Heidelberg some time in the seventies to whom we are deeply grateful. The
inuence of Albrecht Dold is also present.
The authors1
Preface
Chapter 1
1.1
Introduction
If one wishes to study topological spaces, one way of doing it is the following. One may take a cell decomposition (or using cells one constructs a new
space) and one tries to reduce its topological properties to algebraic or combinatorial relationships between the boundaries of the cells, for instance, one
can construct simplicial complexes or apply a homology theory.
A second possibility can be illustrated by the following algebraic situation.
One may take an exact sequence
O F E B O
(of groups, say) and ask what possible values of E one can take for given F
and B (for example, E = F B is always possible).
It is a useful idea to compare this question with the following topological
situations. The general setup will be as follows. Let p : E B be any
continuous map. The inverse images p1 (b) of points b in B constitute a
decomposition of E into bers p1 (b). We get closer to the algebraic
situation described above if all bers were homeomorphic to each other as
it will be the case in the following examples. The maps p : E B, that
we shall be dealing with will be generically called brations, without any
conditions. Later on, according to their particular (lifting) properties, they
will be qualied with a special name, such as trivial bration, Serre bration,
Hurewicz bration, locally trivial bration, and so on.
1.1.1 Examples. The following should be brations.
(a) The topological product dened as follows. Let B and F be topological
Figure 1.1
The space E is not homeomorphic to S1 I since the boundary of S1 I
consists of two circles, i.e., it is not connected, but the boundary of E
is a circle, i.e., it is connected. By means of
Bd(M ) = {x M | H2 (M, M x) = 0}
one can dene the boundary of M = S1 I, resp. M = E in a topologically invariant way.
(c) The Klein bottle dened as follows. Let E be obtained from I I by
identifying for every t I the pair (0, t) with (1, 1 t) and for every
s I the pair (s, 0) with (s, 1). Let B = S1 be obtained again as in
(b) and p : E B be induced again by (s, t) 7 s, then p1 (s) S1
for every s B (see Figure 1.2).
The space E is not homeomorphic to the torus S1 S1 . As a proof of
this fact we compute the homology of E using the cell decomposition
shown in Figure 1.3.
1.1 Introduction
Figure 1.2
e0
e1
e0
e0
ee1
e2
e1
ee1
ee1
e0
e1
e0
Figure 1.3
It consists of one 0-cell e0 , two 1-cells e1 and ee1 and one 2-cell e2 . In
the cellular chain complex one has the following:
e2 = 2e
e1
e1 = e
e1 = 0
e0 = 0 ,
from which we obtain
H2 (E) = 0 ,
H1 (E)
= Z Z2 .
H1 (S1 S1 )
= Z Z,
(see [1, 7.3.12]). Since the rst and second homology groups of both
spaces are dierent, they cannot be homeomorphic.
(d) The covering maps, of which a particularly important example will be
the following. Take
p : R S1 C ,
x 7 e2ix ,
Figure 1.4
(e) The tangent bundle of a smooth manifold, of which a concrete example
is the tangent bundle T (Sn ) of the sphere Sn Rn+1 (cf. 1.6.6 (e)). Let
{
}
T (Sn ) = (x, v) Sn Rn+1 | x v
be furnished with the relative topology, and take
p : T (Sn ) Sn ,
(x, v) 7 x .
Consider the restriction of p to
T (Sn ) = {(x, v) T (Sn ) | v = 0} .
The following is an interesting question: Does there exist a continuous
map s : Sn T (Sn ) such that p s = idSn ? One such s is called
a section of p. Geometrically, s can be described as a nonvanishing
continuous vector eld on Sn .
1.1.2 Exercise. Prove that the brations of (b) and (c) have a section and
that, on the contrary, that of (d) does not.
All examples introduced in 1.1.1 are going to be brations in a sense that
we still have to state precisely. On the contrary, the following will not be
one, even though all of its bers are homeomorphic.
(0, 1)
(1, 1)
(0, 0)
(1, 0)
p
Figure 1.5
The map p has the following property: Not for every path : I B and
for every point x0 E, such that (0) = p(x0 ), there exists another path
e : I E such that
e (0) = x0 and p
e = ; i.e., not for every path in B,
there exists a lifting to E with a given origin. For instance, if x0 = (0, 2),
there does not exist
e unless is constant in a neighborhood of 0. (It is an
exercise to prove this fact.) See Section 1.4 for a general treatment of this
question.
1.2
General Definitions
In this section we present the general set up on which the rest of this book
is supported.
1.2.1 Definition. (For the time being) we shall call bration any continuous map p : E B. E will be called the total space and B the base space
of the bration. Moreover, p1 (b) will be called the ber over b, (b B).
1.2.2 Definition. Let p and p be brations. A pair of maps (f, f ) is called
a ber map from p to p if the diagram
E
/ E
/ B
/ B F
projB
/ B
E=
= E
B=
0
=B
0
Figure 1.6
1.2.5 Definition. Let p : B E be a bration and A B. Then
pA = p|p1 (A) : EA = p1 (A) A
is called the restriction of the bration p to A.
1.2.6 Exercise. Prove that if p : E B is trivial, then also pA : EA
A is trivial.
1.2.7 Definition. A bration p is locally trivial if every point b B has a
neighborhood U such that pU es trivial.
1.2.8 Theorem. Let p : E B be a locally trivial bration. If B is
connected, then all bers of p are homeomorphic.
Proof: In a trivial bration, clarly all bers are homeomorphic. Let b0 B
be any point. Then the set
{
}
B0 = b B | p1 (b) p1 (b0 )
is open in E, namely let b B0 and U be a neighborhood of b in B such
that p is trivial over U . Then all bers over U are homeomorphic and so
U B0 . Similarly one proves that B B0 is open in B. Since B0 = and
B is connected, then B = B0 .
??
??
p ???
S1
/ S1 F
ww
ww
w
ww
{ww projS1
1
2
1
2
Figure 1.7
(b) In a similarly simple way one can see that the Klein bottle bration and
the exponential bration R S1 are locally trivial but not trivial.
(c) On the contrary, example 1.1.3 is not locally trivial (exercise).
1.3
Further Examples
Fn+1 {0} o
MMM
MMM
M
p MMM
&
FPn ,
Dene
??
?
hi
Ui
/ Ui (F {0}) ,
nn
nnn
vnnnprojU
Vi Sd(n+1)1
O
OOO
OOO
O'
pU
i
ki
Ui
/ Ui Sd1
r
rrr
r
r
yr projUi
)
(
hi (x0 , x1 , . . . , xn ) = p(x), |xi |1 xxi ,
10
It is easy to verify that gi is well dened for each i and that hi and gi
are continuous and inverse to each other. Thus hi is a homeomorphism.
ki = hi |Vi Sd(n+1)1 and gi |Ui Sd1 are inverse of each other (and have the
desired images), and the diagrams obviously commute with hi and ki .
C2 {0} o ? _ S3
CP1
z0
z1
C {}
S2
with
(s0 , s1 , 0) ,
with
(s0 , s1 , 1) .
11
(s, t) (s, t) S1 S1 .
Figure 1.8
Now we can describe p : S3 S2 by mapping
(
)
(
) (s )
0
2 12
2 12
rs0 , (1 r ) s1 7 r(1 r )
C {} .
s1
The inverse images of a point in S2 correspond to a xed value of r.
They constitute a circle that lies on the torus given by the equation
r = constant, if r = 0, 1. If r = 0 or 1, then they determine full circles.
Each of these circles intersects each parallel and each meridian of the
torus exactly once. Every two circles that are inverse images of a point
are knotted. For this, one might analyze the case p1 (0) and p1 (z)
(z = 0, ), or for two of those circles that lie on the same torus r =
constant.
One might try to study the general map
S1 S1 S2
) ( sm )
(
0
2 12
[s0 , s1 , r] 7 r(1 r )
sn1
where m and n are natural numbers. In general one does not obtain
a locally trivial bration, since the local triviality fails on the points
12
Figure 1.9
The relative position of two of these inverse image tori can be visualized
as follows.
One stretches a (self-intersecting) surface along the rst trefoil knot
and chooses one side of it to be the front (i.e., one takes an orientation
of the surface). After traveling along the second trefoil knot in the
adequate sense, then one crosses the surface 2 3 = 6 times from the
front to the back.
1.4
Homotopy Lifting
X I
e
h0
e
h
v
h
/E
v;
p
/B
13
1.4.3 Theorem. A trivial bration has the HLP for every space.
Proof: A trivial bration is equivalent to the product bration. Therefore,
we can restrict ourselves to the problem
e
h0
X {0}
_
X I
e
h
/BF
r9
projB
/B.
Dene e
h by e
h(x, t) = (h(x, t), projB e
h0 (x, 0)).
1.4.4 Examples.
(a) The bration of example 1.1.3 does not have the HLP for any nonempty
space X, since for instance the homotopy h(x, t) = t cannot be lifted
starting with e
h0 (x, 0) = (0, 2).
(b) There are brations that have the HLP for a one-point space X = {}
but not for X = {} I I. An example of this is the double covering
of the plane branched at the origin given, say, by
p : C C ,
z2
|z|
0
7 0,
z 7
e : I C such that p
e = and such that
e (0) p1 ((0)) is
preassigned. Now, since I 1 (0) is an open set, it is an at most
countable union of intervals In open in I. Since p|C{0} is a covering
map (see Section 1.8 below), |In can be lifted. Let
en be a lifting. If
0 In (and
e (0) = 0) let
en be such
en (0) =
e (0). If we dene
{
en (t) if t In ,
e (t) =
0
if t 1 (0),
then
e is such that
e (0) is as we wanted, and p
e = . Moreover,
e
1
is continuous, since for t In , this is clear, and if t0 (0), then the
continuity of
e at t0 follows from the fact that |e
(t)| = |(t)|, that is,
|e
(t)
e (t0 )| = |(t) (t0 )| and from the continuity of .
Now, if X = I and h : X I C is given by h(s, t) = (s 21 , t 21 )
there does not exist e
h for any e
h0 , since p restricted to p1 (h(I I))
(that is, the inverse image of the boundary of h(I I) (see Figure 1.10)
is a twofold-covering map, and e
h would induce a section of it, fact that
is not true (cf. Section 1.8).
14
Figure 1.10
1.4.5 Exercise. Prove that if the group Z2 acts on C antipodally, then one
has aan isomorphism C/Z2 C such that there is a commutative diagram
zz
zz
z
z
z| z
q
C/Z2
CA
AA
AAp
AA
A
/ C.
/7 E
pp
p
p
p pe
p
h
p
/B
X I
X {0} A I
_
admits a map e
h that makes both triangles commutative.
Even a trivial bration might not always have the relative HLP as one
can easily see in the case B = {}, since in this case, the existence of e
h such
that the upper triangle commutes implies an extension problem, and this
problem is usually nontrivial (note, however, that the commutativity of the
lower triangle is in this case always trivial).
1.4.7 Theorem. The following statements are equivalent:
(a) p has the HLP for the closed unit ball Bn , n = 0, 1, 2, . . . ( Bn = {x
Rn | x 1} ).
15
(b) p has the relative HLP for the pair (Bn , Sn1 ), n = 0, 1, 2, . . . .
(c) p has the relative HLP for a CW-pair (X, A).
(d) p has the HLP for every CW-complex X.
Proof:
be given by
(
)
1+t
x, t
if |x| 12 (2 t),
2t
)
k(x, t) = (
1 (1 + t) x , 2(1 |x|)
if |x| 12 (2 t).
2
|x|
k is a homeomorphism of pairs that converts the relative homotopy lifting
problem for the pair (Bn , Sn1 ) into a homotopy lifting problem for Bn . See
Figure 1.11.
11111111
00000000
00000000
11111111
00000000
11111111
00000000
11111111
11111
00000
00000
11111
Figure 1.11
(c) (d) Just take A = .
(d) (a) Just observe that Bn is a CW-complex.
(b) (c) Let X n be the n-skeleton of X and let Xn = X n A. We shall
inductively construct maps
e
hn+1 : (X {0} Xn I) E
such that e
hn+1 |X{0}Xn1 I = e
hn and such that the composite p e
hn = h,
wherever it is dened.
e
h0 : (X {0} A I) E
is already given. Assume that e
hn has already been constructed. Recall that
X I is a CW-complex with cells of the form
ek (0, 1) ,
ek {0} ,
ek {1}
16
Bn {0} Sn1 I
j id
17
Proof of the theorem: We shall use condition (b) of 1.4.7 for each pU and
prove (a) for p. For technical reasons, we substitute the ball Bn with the
homeomorphic cube I n .
Subdivide I n I by successively halving the sides until each subcube is
mapped by h into some U U . Thus we obtain a decomposition of I n , whose
k-dimensional subcubes (faces if k < n) will be denoted by Vik , as well as a
decomposition of I
0 < t1 < t2 < t3 < < 1 .
We shall extend e
h0 step by step along the layers I n [tj , tj+1 ] to nally
I {0} V k
n
e
hk1
/6 E
m m
m
m
p
m m ehk
m
m
/B,
[0, t1 ]
I n {0} V k1
[0, t1 ]
_
h|
k = 0, 1, . . . , n, where e
h0 = e
h0 .
Assume that e
hk1 has already been constructed. Then we can solve the
problem
k
Vik {0} V
i [0, t1 ]
e
hk1 | 1
/
6p U
l l
l
pu
l le k
hi
l l
/
Vik [0, t1 ]
h|
/E
p
/ B,
18
We omit the proof, since it is quite intrincate and would pull us apart
from the topics we are dealing with. See [3, Thm. 48] for a proof.
Since every CW-complex is paracompact (see [12] or [8]), we have the
following.
1.4.15 Corollary. Let p : E B be a bration and U = {Uj }jJ be an
open cover of B such that pUj is a Hurewicz bration. Then
(a) if B is a CW-complex then p is a Hurewicz bration;
(b) p has the HLP for every CW-complex.
Consider the path space X I = { : I X} furnished with the compactopen topology. Given a bration p : E B, take the bered product
E B B I = {(e, ) E B I | p(e) = (0)} .
19
E B _ B I
i0
E B B I I
/
r8 E
/ B,
where i0 is the inclusion into the bottom of the cylinder (i0 (e, ) = (e, , 0))
and (e, , t) = (t). Since the square is obviously commutative, and the
b : E B B I I I,
bration has the HLP for every space, there exists
such that both triangloes commute. Dening : E B B I E I by
b , t) ,
(e, )(t) = (e,
we have the desired PLM.
Conversely, assume that there is a PLM : E B B I E I for p : E
B and assume a general homotopy lifting problem
f
X _
i0
e
h
X I
/E
v:
p
/ B.
/ E B B I
/ (f (x), h(x))
/ EI
/ (f (x), h(x)) .
Then e
h : X I E given by e
h(x, t) = h (x)(t) is the desired lifting.
20
1.4.18 Exercise. Proving the existence of PLMs show that the following
are Hurewicz brations:
(a) The map B I B, given by 7 (1).
(b) The map P (B) = { B I | (0) = b0 } B, given by 7 (1).
This is the so-called path bration of B (see 3.4.7).
(c) Given f : X B, the map Ef = {(x, ) XB I | f (x) = (1)}
B, given by (x, ) 7 (0). The space Ef is the so-called mapping path
space, the bration is the mapping path bration, and its ber over (a
base point) b0 B, Pf = {(x, ) X B I | (0) = b0 , f (x) = (1)},
is the so-called homotopy ber of f .
The following result states that every map factors as a homotopy equivalence followed by a Hurewicz bration (i.e., every map can by replaced by
a Hurewicz bration, up to a homotopy equivalence). It is an easy exercise
to prove it.
1.4.19 Proposition. Given any continuous map f : X B, the map
: X Ef given by x 7 (x, ex ), where ex : I B is the constant path
with value f (x), is a homotopy equivalence. Moreover, there is a commutative
triangle
Ef
}>
}}
}
fe
}
}}
/ B,
X
21
Proof of 1.4.20: We shall apply Lemma 1.4.21. Assume that u and h are as
in that lemma. Since p : E B is a Hurewicz bration, the lifting problem
/7 E
oo
o
p
i0
oo
o
o
/B
EI
id
h(pid)
22
/ B.
e (namely, a
p
e = ) if and only if (p) admits a section s : A E
map such that (p) s = idA ).
The following is an important result.
1.4.25 Proposition. Let : A B be continuous. If a bration p :
E B has the HLP for a space X, then so does (p).
Proof: We have to show that the homotopy lifting problem
e
h0
X _
i0
X I
e
hy y
e
/E
y<
(p)
/A
23
X _
i0
X I
e
kv
/E
v;
p
/ B,
AA
AA
p0 AA
/ E1
}
}
}}p
}
}
~} 1
and
E1 A
AA
AA
p1 AA
/ E0
}
}
}}p
}
}~ } 0
commute, and these maps are such that B idE0 and B idE1 , that
is, these composites are ber homotopic to the identities in the sense that
they are homotopic through homotopies H and K such that the triangles
E0 GI
GG
GG
GG
p0 proj1 GG
#
are commutative.
/ E0
}
}}
}}p0
}
}
~
and
E1 GI
GG
GG
GG
p1 proj1 GG
#
/ E1
}
}}
}}p1
}
}~
24
Figure 1.12
It is an exercise for the reader to gure out explicitly.
Dene
f : X (I I {0} I) E
by
f (x, s, 0) = G0 (x, s) ,
f (x, 0, t) = K(x, t) ,
f (x, s, 1) = G1 (x, s) .
Then the diagram
e
h0
/ X (I I {0} I) ef e26/( E
_
e e em m m
e
he e e e e
m
p
e
m m H
e e e
e
e
m
e
e
m
e e
/6 B
/X I I
X I I
X I
_ {0}
idX |
idX
25
commutes and both the exterior square as well as the right square pose
lifting problems. Since p is a Hurewicz bration, there exists e
h (that solves
1
e
e
the exterior problem). Then H = h (idX ) solves the problem on the
right. This is the desired homotopy.
The following result combines the concept of homotopic maps with that
of equivalent Hurewicz brations.
1.4.29 Theorem. Let p : E B be a Hurewicz bration and let 0 , 1 :
e0 A and (p) :
A B be homotopic. Then the brations 0 (p) : E
1
e1 A induced by p through 0 and 1 , respectively, are ber homotopy
E
equivalent.
Proof: Let p0 = 0 (p) and p1 = 0 (p) be the induced brations, and let
e0 E and 1 : E
e1 E be the corresponding projection maps such
0 : E
that p 0 = 0 p0 and p 1 = 1 p1 . Given a homotopy F : A I B
e0 I E and G1 : E
e1 I E that
from 0 to 1 , there are maps G0 : E
solve the lifting problems
e0
E
_
i/4 E
i i
i
G0 i i
p
i0
i i
i
i
i i
e0 I
/AI
/B
E
p0 idI
and
e1
E
_
i/4 E
i i
i
G1 i i
i
p
i1
i i
i
i
i i
e1 I
/ B,
/AI
E
p1 idI
where i0 and i1 are the inclusions into the bottom and into the top of the
e0 E
e1 be given by
corresponding cylinders, respectively. Let e0 : E
e0 (a, e) = (a, G0 (a, e, 1)) ,
e1 E
e0 be given by
and e1 : E
e1 (a, e) = (a, G1 (a, e, 0)) .
Then
p (G0 (e1 idI )) = F (p0 idI ) (e1 idI ) = F (p1 idI ) = p G1
and
G0 (e1 idI ) i0 = G1 i0 .
26
e1 e0 A idEe0 .
and
1.4.30 Corollary. If p : E B is a Hurewicz bration and B is contractible, then p is ber homotopy equivalent to the trivial bration B
p1 (b) B for any b B.
Proof: If B is contractible, then idB cb , where cb : B B is the constant map with value b. Obviously idB (p) is equivalent to p and by 1.4.23
the induced bration cb (p) is trivial. Hence, by 1.4.29, p is ber homotopy
equivalent to a trivial bration.
1.5
X I
where
e
h0 (x, 0) = f0 (x)
e
h0
e
h
u
h
and
/
u: E
p
/ B,
h(x, t) = (t) .
We assume further that p has the HLP for X and X I (this is not always
the case, as seen in 1.4.4 (b)), then we can solve the problem and there exists
such a map e
h. Since e
h(x, 1) F1 = p1 (b1 ) we may dene f1 : X F1 by
f1 (x) = e
h(x, 1) and say that f1 is obtained from f0 by translation along .
27
resp. e
h (x, 0) = f0 (x), where h and h are given by h(x, t) = (t) and
e
and f1 (x) = h (x, 1).
Let now g : X I F0 be a homotopy such that g(x, 0) = f0 (x)
and g(x, 1) = f0 (x), and let : I I B be a homotopy such that
(s, 0) = (s) and (s, 1) = (s), (0, t) = b0 and (1, t) = b1 for all s, t.
Consider the problem
e0
H
/6 E
l l
l
l
p
l lH
e
l
l
/ B,
X I I
X (I I {0} I)
_
where
H(x, s, t) = (s, t)
e
H0 (x, s, 0) = e
h(x, s)
e
e
H0 (x, s, 1) = h (x, s)
e 0 (x, 0, t) = g(x, t) .
H
Since the pair (X I I, X (I I {0} I)) is homeomorphic to the
pair (X I I, X {0} I) (see Figure 1.13, and compare with the proof
of 1.4.7), and p has the HLP for X I, the solution of the problem exists.
1111111111
0000000000
0000000000
1111111111
00000000000
11111111111
0000000000
1111111111
00000000000
11111111111
0000000000
1111111111
00000000000
11111111111
0000000000
1111111111
000000000
111111111
00000000000
11111111111
0000000000
1111111111
000000000
111111111
000000000
111111111
00000000000
11111111111
000000000
111111111
00000000000
11111111111
I
11111111111
00000000000
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
I
Figure 1.13
e 1, t) = H(x, 1, t) = (1, t) = b1 , F (x, t) = H(x,
e 1, t) denes a
Since pH(x,
homotopy F : X I F1 from f1 to f1 .
28
1.5.4 Definition. Let [X, Y ] denote the set of homotopy classes of maps
X Y . For each path : b0 b1 , there is a function : [X, F0 ]
[X, F1 ] that sends the homotopy class of any map f0 : X F0 to that of
the map f1 : X F1 obtained from f0 by translation along .
Theorem 1.5.2 guarantees that the function is well dened, and Theorem 1.5.3 shows that
2 1 = 1 2 .
Let eb : I B be the constant path with value eb (t) = b , = 0, 1.
Then
eb = id[X,F ] ,
as one can easily verify. Moreover, if is the inverse path of , then by
Theorem 1.5.3 and the previous remark,
= id[X,F0 ]
and
= id[X,F1 ] .
29
The fundamental groupoid 1 (B) is a small category, that is, its objects
constitute a set (the underlying set of the space B). Given a map f : B
B , there is a covariant functor fb : 1 (B) 1 (B ) that coincides with f
in the objects and is such that for a path : b0 b1 , one has fb([]) = [f ].
Obviously, the functor fb depends only on the homotopy class of f . We have
the following.
1.5.7 Proposition. The assignment B 7 1 (B) is a functor from the homotopy category Toph of topological spaces and homotopy classes of maps, to
the category Cat of small categories and functors between them.
/ [Y, F0 ]
[X, F1 ]
/ [Y, F1 ] ,
/ X {0} f0 i4/ E
_
i iu u:
e
h i i i
p
u
i i
u eh
i i
i
u
i
/X I
/ B,
Y I
Y {0}
_
(y, t)
gid
/ (g(y), t)
/ (t) .
The map y 7 e
h (y, 1) provides a representative of (). On the other
hand, f1 g represents (), and since e
h (y, 1) = f1 g(y), one gets the
assertion of the theorem.
30
There are some consequences of the previous theorem. Since every morphism in the fundamental groupoid is an isomorphism we have the following.
31
1.6
7 [g a f ]
that does not depend on the choice of the representative a [a]. The following rules are easily veried.
(f , g ) (f, g) = (f f , g g) ,
(idX , idY ) = id(X,Y ) ,
(f, g)(0) = 0 .
32
f = (f, id) .
idZmmmm
m
mmm
vmmm
(X Y ) Z
(X Y ) Z _ _ _ _ _ _f _ _ _ _ _ _/ X (Y Z) ,
where , as well as the two vertical maps are identications. f denes a
bijection such that f ((x y) z) = x (y z). f will be a homeomorphism
when the maps idZ and idX are identications. This is the case
under the given hypotheses.
33
1.6.6 Examples.
(a) Y = .
(b) S0 Y Y .
(c) I Y = CX with 0 I as the base point is the (reduced) cone of Y .
(See Figure 1.14 (c), where the thick line represents the base point.)
(d) S1 Y = Y is the (reduced) suspension of Y . (See Figure 1.14 (d),
where the thick line represents the base point.)
Y
(c) CY
(d) Y
Figure 1.14
(e) Let Sn = {x Rn+1 | x = 1} be the unit n-sphere with =
(1, 0, 0, ..., 0) as the base point.
There is a pointed homeomorphism
: Sn = S1 Sn Sn+1
given as follows. If we describe the points of S1 by
(cos 2t, sin 2t) ,
t [0, ] ,
then is given by
((cos 2t, sin 2t) (x0 , . . . , xn )) =
)
1x0
2
2
2
2
= cos t + x0 sin t, x1 sin t, . . . , xn sin t,
sin 2t Sn+1 .
2
(
34
f g
/ X Y
(X Y )
f g
/ X Y ,
q is an identication.
1.6.8 Theorem.
(1) is a two-variable covariant functor.
(2) is compatible with the homotopy relation, i.e., if f0 f1 and g0 g1 ,
then f0 g0 f1 g1 .
Proof: (1) follows immediately.
(2) is obtained as follows: Let h : I X X be a homotopy between
f0 and f1 , and let q : X Y X Y , q : X Y X Y be the
respective identications. Then in the diagram
I X Y
idq
I (X Y )
hg
/ X Y
q
/ X Y
the map idq is again an identication and therefore the arrow at the bottom
describes a homotopy f0 g f1 g. To prove f g0 f g1 one proceeds
similarly; the general case follows combining the two previous cases.
1.6.9 Definition. Let S1 = I/{0, 1}, where we denote its points simply
by their inverse images in I. Let 0 S1 be the base point. Let moreover
f, g : X Y be pointed maps. We dene f + g : X Y by
{
f (2t x)
if 0 t 21 ,
(f + g)(t x) =
g((2t 1) x) if 12 t 1.
f + g is well dened and is continuous. If ft and gt are homotopies, then
also ft + gt is one, so that [f ] + [g] = [f + g] denes an operation + in
(X, Y ).
1.6.10 Theorem. ((X, Y ); +) is a group with the selected element 0 as
neutral element.
35
Proof: Observe that, as we did above, one can write a homotopy as a family
ht : X Y ,
ht (x) = h(x, t) ,
1
1
2 s(2 t) if 1 s 2 ,
t (s) = s 14 t
if 21 s 34 ,
s(1 + t)
if 34 s 1,
describes a pointed homotopy t : S1 S1 . By 1.6.8
((f + g) + h) (t x) : S1 X Y
is a homotopy. From the fact that 0 = idS1 and that ((f +g)+h)(1 x) =
f + (g + h) the associativity is obtained.
Neutral element: The map
{
s(1 + t)
if 0 s 12 ,
t (s) =
t + (1 t)s if 12 s 1,
gives a homotopy t : S1 S1 . If k is the constant map, we have that
gt = f (t idX ) : S1 X Y is a homotopy between g0 = f and
g1 = f + k.
Existence of the inverse: By
f (t x) = f ((1 t) x)
a continuous map f : X Y is dened. The homotopy t : S1 S1
given by
{
2st
if 0 s 12 ,
t (s) =
2t(1 s) if 12 s 1,
is such that f (0 idX ) = k and f (1 idX ) = f + f .
n1
36
1.6.12 Exercise. Prove that the fundamental group 1 (X) is the group of
isomorphisms of the base point to itself in the fundamental grupoid 1 (X)
dened above in 1.5.5.
1.6.13 Theorem. Let f : Y Y , g : X X be pointed maps. Then
f : (X, Y ) (X, Y )
and
(g) : (X, Y ) (X , Y )
are homomorphisms, where g = idS1 g.
Proof: Let a, b : X Y represent two elements in (X, Y ). One has
{
f a (2t x)
if 0 t 12 ,
(f (a + b)) (t x) =
g b((2t 1) x) if 21 t 1,
= ((f a) + (f b)) (t x) ;
hence f ([a] + [b]) = f [a] + f [b], and so f is a homomorphism.
On the other hand, the equalities
((a + b) (id g)) (t x) = (a + b)(t g(x))
{
a(2t g(x))
if 0 t 12 ,
=
b((2t 1) g(x)) if 12 t 1,
{
a (id g) (2t x)
if 0 t 21 ,
=
b (id g) ((2t 1) x) if 12 t 1,
= (a g + b g)(t x) ,
imply that (g) ([a] + [b]) = (g) [a] + (g) [b]; therefore, (g) is a homomorphism.
37
f (2s 2t x)
g((2s 1) 2t x)
=
f (2s (2t 1) x)
g ((2s 1) (2t 1) x)
{
(f + f )(2s t x)
=
(g + g )((2s 1) t x)
if 0 t 12 ,
if 21 t 1,
if 0 s 12 , 0 t 12 ,
if 21 s 1, 0 t 12 ,
if 0 s 12 , 21 t 1,
if 12 s 1, 12 t 1,
if 0 s 12 ,
if 12 s 1,
= ((f + f ) + (g + g ))(s t x) .
from there one obtains by taking special values for the maps
f + g (f + k) + (k + g) = (f + k) + (k + g) f + g ;
f + g (k + f ) + (g + k) = (k + g) + (f + k) g + f g + f
g+f.
The rst of these equation shows that [f ] + [g] = [f ] + [g], and the second,
that [f ] + [g] = [g] + [f ].
38
1.7
One of the most useful algebraic tools is that of an exact sequence. In this
section we show how the homotopy sets and groups introduced in the previous
section t together to yield a long exact sequence.
39
We shall work here under the following assumptions: All spaces, maps and
homotopies, as well as all constructions made, will be pointed. It will usually
be easy to distinguish in the new constructed spaces, which is the base point.
The base point will be generically denoted by , as we do for the one-point
space. A map f : X Y will be called nullhomotopic if it is homotopic
to the constant map; this fact will be denoted by f 0. The bration
p : E B will always be a Serre bration, and i : F = p1 () , E will
denote the inclusion of the ber (F and E have the same base point). X
will be a CW-complex and X will be a 0-cell of some adequate CWdecomposition.
1.7.1 Lemma. The sequence
p
(X, F )
(X, E) (X, B)
/
o7 E
o
o
p
o oe
o
h
o
/B
X I
X {0} _ {} I
to obtain a lifting e
h of h, if we dene e
h0 by e
h0 (x, 0) = f (x) and e
h0 (, t) = .
Since pe
h(x, 1) = h(x, 1) = , h(x, 1) F and it determines a map g :
X F by setting g(x) = h(x, 1). Then e
h : f i g, so that one has
i [g] = [i g] = [f ]. Thus [f ] im(i ).
40
e
hi
i i
i i
i i
I X
i4/ E
i i
i i
const
/ X
/B
0
/3
{0} (I X) I (I
X I {})
g g gE
_
g
g
g
g
I I X
g
e
g g g H
g
g
g
H
/B
e 0 (s, 0, x) = e
H
h0 (s, x) ,
e 0 (s, 1, x) = e
H
h1 (s, x) .
The HLP for the pair (IX, IXI{}) provides us with the existence of
e Since pH(1,
e t, x) = H(1, t, x) = , H
e determines a homotopy gt : X
H.
e t, x), where g0 = g0 and g1 = g1 ; gt is thus the
F through gt (x) = H(1,
desired homotopy.
41
is exact.
Proof: im() ker(i ): If g : X F corresponds to f : X B, then
let e
h : I X E be the homotopy through which g is dened (cf. 1.7.3).
e
h is a homotopy that starts with k (the constant map) and ends with i g.
Thus i [f ] = i [g] = [i g] = [k] = 0.
im() ker(i ): Take g : X F and let e
h : I X E be a
(null)homotopy such that e
h(0, x) = and e
h(1, x) = ig(x) = g(x) F .
Let moreover h = p e
h. Since h(I X I {}) = {}, the map h is
compatible with the identication q and so it determines a continuous map
f : X B such that f q = h. In a diagram
I XI
II
II
II
h II$
B.
|x
/ X
x
{0} X _ I {}
const
pfe
where e
h(t, x) = fe(t x). Thus g : x 7 e
h(1, x) = fe(1 x) = corresponds
to f = p fe. In other words, p [fe] = [f ] = [g] = [k] = 0.
im(p ) ker(): If g : X F corresponds to f : X Y through
e
h : I X E, and is such that [g] = , then let gt : X F be a
nullhomotopy such that g0 = g and g1 = k, and dene e
h : I X E by
{
e
h(2t, x) if 0 t 12 ,
e
h (t, x) =
g2t1 (x) if 12 t 1.
The map e
h is compatible with the identication q : I X X and
therefore it denes a map fe : X E such that q fe = e
h . Now,
p fe = f + k f . Thus p [fe] = [f ], i.e., [f ] im(p ).
42
q
const
{0} X
I {}
_
f1 + f2
{
g1 (2s x)
if 0 s 12 ,
g(s x) = e
h(1, s x) =
g2 ((2s 1) x) if 12 s 1,
= (g1 + g2 )(s x) ,
corresponds to f1 + f2 through e
h, i.e., ([f1 ] + [f2 ]) = [f1 + f2 ] = [g] =
[g1 + g2 ] = [g1 ] + [g2 ] = [f1 ] + [f2 ].
n (X, F )
n (X, E) n (X, B) n1 (X, F )
p
i
0 (X, F )
0 (X, E) 0 (X, B)
is exact, and all arrows (maybe excepting the last three) represent group homomorphisms.
(n f )
n (Y, F )
commutes.
/ n (X, E)
(n f )
/ n (Y, E)
/ n (X, B)
(n f )
/ n (Y, B)
/ n1 (X, F )
(n1 f )
/ n1 (Y, F )
43
In general, (X, F ) does not have a group structure; is not a homomorphism in those cases. However, it is true that sends the right cosets
of im(p ) exactly onto one element. We have the following.
1.7.8 Theorem. 0 = 1 if and only if 0 1 im(p ).
Proof: Assume that 0 = 1 . Let f represent , and let g correspond
to f through a homotopy e
h , = 0, 1. By assumption, g0 g1 , say via the
homotopy gt . If we dene a map l : X E by
if 0 s 41 ,
h0 (4s, x)
l(s x) = g4s1 (x)
if 41 s 12 ,
e
h1 (2 2s, x) if 12 s 1,
then we have that p l = (f0 + k) + f 1 (f 1 is the inverse of f1 ), since by
denition of e
h , one has that pe
h (s, x) = f (s x). Passing to homotopy
classes we have
p [l] = [p l] = [f0 ] + [k] + [f 1 ] = [f0 ] [f1 ] = 0 1 im(p ) .
Conversely, let us suppose that 0 1 im(p ). More specically, 0 =
p () + 1 . Choose representatives f1 of 1 and l of and take f0 = p l + f1
as a representative of 0 .
If g1 corresponds to f1 through e
h1 , then dene e
h0 by
{
l(2s x)
if 0 s 12 ,
e
h0 (s, x) =
e
h1 (2s 1, x) if 12 s 1,
to obtain that p e
h0 = (p l + f1 ) q = f0 q, so that e
h0 lifts f0 q. Since
e
e
h0 (1, x) = h1 (1, x) = g1 (x), then g1 also corresponds to f0 , that is,
1 = [f1 ] = [g0 ] = [f0 ] = 0 .
44
I O F
/E
p
idf
I X
/ 2 X _ _ _/ B
Gf
/3
g g gE
g
g
g g
p
const
e
hg
g
g g g
g
g
g
/ 2 X
I X
q
Gf
/ B.
But setting e
h = g (id f ) (cf. 1.7.9) one gets e
h(1, z) = g(1, f (z)) = f (z).
e
Thus f corresponds to Gf through h, and so [f ] = [Gf ] = [f ].
For any n > 2, just replace X in the previous case with n1 X.
1.8
Applications
In this section, we explain some particular instances of (locally trivial) brations that have special interest in algebraic topology.
1.8.1
Covering Maps
One of the most useful tools of algebraic topology for computing the fundamental group of a space is the concept of a covering map, that we analyze
succintly in what follows. See [13] or [1] for a thorough treatment.
45
1.8 Applications
/
o7 E
o
o
p
o oe
o
h
o
/B
I I
{0} I (I I)
_
a map e
h, where h(t, s) = (t), e
h0 (0, s) = x0 , e
h0 (t, 0) =
e0 (t), e
h0 (t, 1) =
e
e1 (t). For xed t, the mapping s 7 h(t, s) denes a continuous map into
the ber over (t), and is thus constant, since the ber is discrete. Hence
e0 (t) = e
h(t, 0) = e
h(t, 1) =
e1 (t).
Of course, the previous theorem and its proof are still valid if p is a Serre
bration and each ber admits only constant paths.
1.8.3 Corollary.
(a) For a covering map, the homotopy lifting is unique. (This follows since
a homotopy is nothing else but a family of paths.)
(b) For a covering map, the translation of the ber along a path in B is
unique. (This follows from the fact that in order to translate a ber
one has to lift a particular homotopy.)
46
Suppose that in the covering map p : E B, the total space E is 0connected (i.e., 0 (E) = 0), and that B is connected. Then B es 0-connected.
The exact sequence
p
1 (E) 1 (B) 0 (F )
0 (E) = 0 ,
0 (F ) (as sets, since im(p ) does not have to be a normal subgroup of 1 (B)
(cf. for instance [7, III.17.1]).
47
1.8 Applications
Proof: p is bijective, since 0 (F ) = 0, and p1 is continuous, since the projection p of a locally trivial bration is an open map.
In particular, B = Sn does not admit nontrivial covering maps with pathconnected total space if n 2.
If E is simply connected, we have an isomorphism of sets
1 (B)
= 0 (F )
=F.
Considering the special case of the covering map
p : R S1
(cf. 1.1.1(d)), we obtain the following.
1.8.10 Theorem. There is a group isomorphism 1 (S1 )
= Z.
Proof: Let n,m : I R be the path t 7 m+nt from m to m+n, m, n Z.
Any other path from m to m + n is homotopic to m,n , since hs : I R,
given by
t 7 (1 s)m,n (t) + s (t)
is a homotopy from m,n to relative to the end points.
In particular,
0,m + m,n 0,m+n .
Each path : I S1 with (0) = (1) can be lifted to
e : I R, so
that
e (0) = 0 and
e (1) = k (for some k Z). One has
=p
e p 0,k = p n,n+k .
Take k = p 0,k . Since
[k ] + [l ] = [p 0,k ] + [p k,k+l ]
= [p 0,k+l ]
= [k+l ] ,
we have that [1 ] generates 1 (S1 ). Therefore, 1 (S1 ) is cyclic and innite as
a set. Hence it is free.
48
1.8.2
Spherical Fibrations
There are cases in which special bers, base spaces, or even total spaces of a
given bration make the long homotopy exact sequence collapse. One obtains
short exact sequences or even isomorphisms that provide us with valuable
information. In what follows, we shall analyze cases in which one or more of
those spaces are spheres.
We assume well known that
i (Sn ) = 0 if i < n, and n (Sn ) = Z if n 1
(see [1, 5.1.22] or [7, IV.2]; see also 1.6.19).
Take n 1 and consider the brations 1.3.1
p : Sd(n+1)1 FPn
with ber embedding
i : Sd1 , Sd(n+1)1 .
The map i is nullhomotopic. Thus, from the homotopy exact sequence of p,
we obtain the short exact sequences
p
0 Z Z Z2 0 .
2. d = j = 1, n > 1. Then 1 (S1(n+1)1 ) = 1 (Sn ) = 0, thus
1 (RPn )
= 0 (S0 )
= Z2 ,
= rst as sets, but also as groups, since there is only one group with
two elements.
49
1.8 Applications
3. d = 1, j 2, n 2.
Since j (S0 ) = 0, we obtain
j (Sn )
= j (RPn ) .
4. d = 4, j 2.
From Theorem 1.7.9 one has
j (HPn )
= j (S4n+3 ) j1 (S3 ) .
For n = 1, in particular, one has
j (S4 )
= j (S7 ) ji (S3 ) .
1.8.12 Note. The homeomorphism FP1 Sd can be given similarly to the
case F = C (cf. 1.3.1 (b)). Namely, via
F2 {0} Sd(n+1)1 (w0 , w1 ) 7 w0 w11 F {}
= Sd .
(Exercise).
5. d = 2, n 1.
One has
{
Z
if j = 2,
j (CPn )
=
2n+1
j (S
) if n = 1,
since j1 (S1 ) = 0 if j =
2 (cf. 1.8.6). In particular, one has (for n = 1)
that
j (S2 )
= j (S3 ) if j > 2 .
1.8.13 Remark. With the help of the Cayley numbers (octonians), one can
construct an analogous bration to the previous ones
S7 , S15 S3
(cf. [15, 20.6]) and conclude from it that
j (S8 )
= j (S15 ) j1 (S7 ) if j 1 ,
(it is nontrivial if j 8).
50
1.8.3
Sections play an important role in many aspects of the theory and applications of the brations. We analyze here some implications of the existence
of a section for a given bration.
p
/ j (E)
o
p
s
j (B)
Chapter 2
Fiber Bundles
2.1
Introduction
2.2
Topological Groups
52
2 Fiber Bundles
1. Let (Rn , +), resp. (Cn , +), be the real, resp. complex, n-dimensional
vector space with the usual topology and the usual sum of vectors.
They both are topological groups for every n.
2. Let GLn (R) be the set of real invertible n n matrices with the group
structure given by matrix multiplication and the topology given as
follows. Fix an ordering of the entries of each matrix, so that it can be
2
considered as an n2 -tuple of real numbers, i.e., as an element of Rn .
2
This way, GLn (R) can be seen as an (open) subspace of Rn with the
relative topology. In this case, is continuous, since the entries of the
product matrix AB 1 are rational functions of the entries of A and B.
Thus GLn (R) is a topological group. In particular, the group GL1 (R)
is the multiplicative group of the nonzero real numbers, also written as
R . The group GLn (R) is called the general linear group of real n n
matrices.
3. Let GLn (C) be the set of complex invertible n n matrices with the
group structure and topology analogous to the previous example. Similarly, GLn (C) is a topological group. In particular, the group GL1 (C) is
the multiplicative group of the nonzero complex numbers, also written
as C . The group GLn (C) is called the general linear group of complex
n n matrices.
2.2.4 Theorem. Every subgroup H of a topological group G with the relative
topology is a topological group.
Proof: Let be the induced multiplication in H. Let i : H , G be the
inclusion. i = |HH is continuous, and since H has the relative topology,
is continuous. Similarly, one proves that the map sending an element in
H to its inverse is continuous.
2.2.5 Examples.
1. The following are important subgroups of GLn (R):
SLn (R) = {A GLn (R) | det(A) = 1}
is the special linear group of real n n matrices.
On = {A GLn (R) | AA = 1} ,
where A is the transposed matrix of A and 1 is the unit matrix, is the
orthogonal group of n n matrices.
SOn = On SLn (R)
53
the matrix AA has entries nk=1 aki akj . Therefore, On is the inverse
image of the closed set {1} GLn (R) under the continuous mapping
A = (aij ) 7
aki akj ,
k=1
where 1 denotes the unit matrix, with ones in the diagonal and zeroes
elsewhere. The subgroup SOn is closed, since it is the intersection of
2
two closed subgroups. Since On Rn is clearly bounded, the groups
On and SOn are even compact.
2. The following are special cases:
O1 = {1, 1} = Z2 = S0 ,
SL1 (R) = {1} ,
SO2
= S1 ,
On SOn Z2
SO3 RP
For the last of the previous statements, we sketch a proof. Each element
in SO3 is a rotation around some axis. Let B3 R3 be the unit ball
and let f : B3 SO3 be the map that sends an element x B3 to
the rotation around the axis determined by x by an angle |x|. f is
clearly surjective; that is, it is an identication (see Figure 2.1).
|x|
x
O
Figure 2.1
From f (x) = f (y) it follows that either x = y or x = y and |x| =
|y| = 1. That is, f identies antipodal points of S2 B3 , and thus f
induces a homeomorphism
RP3 = B3 / SO3 ,
54
2 Fiber Bundles
A B
B A
)
.
55
p1 pA = AH =
Ax .
xH
/G
p
/G
G/H _ _ _/ G/H ,
2.2.9 Exercise. Prove the previous theorem using the maps and instead
of the maps , , , and .
2.2.10 Theorem. The homogeneous space G/H is Hausdor if and only if
H is closed in G.
56
2 Fiber Bundles
57
/G
p
G G/H _ _ _/ G/H
a map is uniquely dened. The action is continuous, since by 2.2.7,
the product of maps idp is an identication. The map is then given
by
(g1 , g2 H) 7 (g1 g2 )H
58
2 Fiber Bundles
With this, it is routine to verify (a) and (b) in 2.2.12. The action is
always transitive, but not necessarily eective. (For instance, if G is
abelian and H = {1}, it is not eective. It is never free.).
2.2.19 Exercise. Prove the following:
(a) If G acts freely on X, then it also acts eectively.
(b) The orthogonal group On acts eectively on Rn , but it does not act
freely.
2.2.20 Remark. Many transitive actions can be reduced to the one of Example 2.2.18, 2.
Let x0 X be a xed element. As we already noted, by g 7 gx0 one
denes a map f : G X. This map f is surjective when G acts transitively
on X. Take H = {g G | gx0 = x0 } = f 1 (x0 ). Then H is a subgroup of G.
It is called the isotropy subgroup of x0 and is usually denoted by Gx0 . This
subgroup is closed whenever the point x0 is closed in X. Let us consider the
problem
GB
BB
zz
BBf
zz
z
BB
z
B!
}zz
G/H _ _ _ _ _ _ _/ X .
p
GX
is commutative.
/ G/H
/X
59
60
2 Fiber Bundles
2.2.25 Example. The group On acts on the sphere Sn1 through (A, x) 7
Ax (cf. Example 2.2.18, 1). Take x0 Sn1 to be the vector such that x0 =
(0, . . . , 0, 1). The equation Ax0 = x0 is equivalent to the matrix equation
(
)
B 0
A=
, B On1 .
0 1
By means of the embedding given by
(
)
B 0
B
7
0 1
we may consider the group On1 as a subgroup of On and by 2.2.20 we have
a homeomorphism
f : On /On1 Sn1 ,
(since On is compact and Sn1 is Hausdor).
2.2.26 Exercise. Similarly to the previous example, give a transitive action of the group Un on the sphere S2n1 Cn . Conclude that there is a
homeomorphism
Un /Un1 S2n1 .
2.2.27 Note. See Subsection 2.5.1 for further examples similar to 2.2.25
and 2.2.26.
2.3
Fiber Bundles
The denition of a ber bundle that we shall give below was proposed by A. Dold.
61
U = B.
FF
FF
FF
F
proj1 FF
#
Uj
/ p1 Uj
yy
yy
y
yp
y| y Uj
62
2 Fiber Bundles
y1 fx x : F F ,
where y = f (x), is an element g(x) G; in particular, it is a homeomorphism.
(C2) The map g : U f
63
g (x) = 1
z hx x = z fy fx x
1
= (1
z fy y ) (y fx x )
= g (y)g(x) G ,
since by assumption both g (y) and g(x) lie in G. It still remains to prove
1
that the mapping x 7 g (x) is a continuous map g : U h U G.
This is true because
{U f
U h U | A }
2.3.5 Theorem. Let F be a set bundle over B with two given atlases A and
A . Then the set bundle map idF = (e, idB ) is compatible with with A and
A if and only if A A is an atlas.
Proof: This follows immediately from Denition 2.3.3.
64
2 Fiber Bundles
65
(e,idB )
(e ,idB )
(F, A2 )
(f,f )
/ (F , A ) .
2
By assumption, (e, idB ) and (e , idB ) are compatible with the atlases (see
2.3.5). If the bundle map (f, f ) on the bottom is compatible with the atlases,
then by 2.3.4 so is also the bundle map on the top.
and
= (F, G, B ; F , A ) ,
66
2 Fiber Bundles
1
We do it as follows. Take f = f and f = {fy | y B } such that fy = fx1
1
1
g (y) = 1
= g(x)1 ,
x fy y = (y fx x )
since f (y) = x U f U = U f
U .
continuous, since f , g, and the map (which sends a group element to its
inverse) are continuous.
E=
Fx and p(Fx ) = {x} .
xB
We now endow E with a topology. Using the map j given by (x, y) 7 j,x (y)
we have the next commutative diagram.
j
Uj F
(2.3.15)
GG
GG
GG
G
proj1 GG
#
Uj .
/ p1 Uj =
ll
lll
lll
l
l
ll pUj
lu ll
xUj
Fx
Uj F {j}
NNN
NNNproj1
NNN
NNN
'
/B,
p
where (x, y, j) = j (x, y), we endow E with a topology and with it p turns
out to be continuous.
We call p : E B the bration determined by the set bundle F.
67
i
1
Gij : 1
j (Ei Ej ) Ei Ej i (Ei Ej )
/ (x, 1 j,x (y), i)
(x, y, j)
i,x
is continuous, since 1
i,x j,x G and G acts continuously on F . The map
Gji is inverse to Gij , and therefore, Gij is a homeomorphism. Let A be open
in Xi . We have to prove that i A is open in Ei , that is, by the very denition
= Gij (A 1
i (Ei Ej ))
is open in Xj . But this is open in 1
j (Ei Ej ), since Gji is a homeomorphism
1
and j (Ei Ej ) = (Ui Uj )F {j} is open in Xj , we have that Xi 1 i A
is open in Xi .
From Diagram (2.3.15) and the previous lemma, we obtain the following
two consequences.
2.3.17 Proposition. p is locally trivial.
In what follows, we see that not only a ber bundle gives rise to a locally
trivial bration, but also that a bundle map induces a ber map.
2.3.19 Construction. Let F and F be set bundles with atlases A and
A . Let (f, f ) : F F be a bundle map that is compatible with the
atlases. We now want to construct a ber map (fb, f ) between the locally
trivial brations determined by the given set bundles (2.3.14), namely,
b
f
E _ _ _/ E
p
/ B ,
68
2 Fiber Bundles
2.3.20 Theorem. The map fb is continuous, and thus (fb, f ) is a ber map.
Proof: The topologies in E and E are given through identications and
. One has
A U F {}
A U F {}
fb
/E.
It is then enough to prove that fb is continuous. For that, since the sets
1
(U f U ) F {} build an open cover of U F {}, we only
check that fb |(U f 1 U )F {} is continuous for all A and all A .
One has
fb(x, v, ) = fbx (v) = fx x (v)
= y (y1 fx x )(v)
= y g(x)(v)
= (y, g(x)v, ) ,
where y = f (x). The last term clearly depends continuously on (x, v), thus
we obtain the desired continuity.
Let F be a set bundle over B with two atlases A and A . The set map
p : E B (as in 2.3.14) depends only on F. However, there are two
topologies T and T in E.
2.3.21 Theorem. If A and A are equivalent atlases, then the topologies T
and T , generated by A and A on E are the same.
Proof: By 2.3.5, the bundle map (e, idB ) : F F is compatible with the
atlases. By 2.3.19, we have that the identity map
idE = eb : (E, T ) (E, T )
is continuous. Similarly, one may prove that the inverse map is also continuous.
69
2.3.1
Tangent Bundles
with
x Def(f ) .
70
2 Fiber Bundles
f h(x) (N ) ,
71
f
|x ,
xj
j = 1, 2, . . . , n.
2.3.27 Lemma. The vectors D1 (x), . . . , Dn (x) build a basis of the tangent
space Tx (Rn ).
For the proof see [11, 2.3].
72
2 Fiber Bundles
If h is a local chart for the manifold M , one can give a local chart =
{x (M ) | x Im(h)} for T by
x = dhy : Rn Tx (M ) ,
y = h1 (x) .
2.4
Coordinate Transformations
73
gii (x) = 1 G, x Ui , i J.
2.
Fx = F ,
j = {j,x | x Uj } ,
x Uj .
To be more precise, one should have to say, a cocycle with coecients in the sheaf
of germs of continuous maps B G, (cf. Hirzebruch [4, 2.6]). However, no confusion
should arise by our short form of stating it.
74
2 Fiber Bundles
i, j J.
2.4.5 Theorem. Let , e be ber bundles over B, with ber F and structure
group G. Let A and Ae be the corresponding atlases with the same cover U and
coordinate transformations {gij }, {e
gij }. and e are equivalent over B (see
2.3.12) if and only if the cocycles g = {gij } and ge = {e
gij } are cohomologous
in U.
In particular, a ber bundle is characterized, up to equivalence over B,
by its coordinate transformations.
Proof: Let (f, idB ) : e be an equivalence. By means of the mapping
ej (x) =
x 7
e1
j,x fx j,x a continuous map j : Uj G is determined
(by 2.3.3). One has
geij (x)j (x) = (
e1
ej,x ) (
ej,x fx i,x )
i,x
1
=
e1
i,x fx (i,x i,x ) j,x
1
= (
e1
i,x fx i,x ) (i,x j,x )
= i (x)gij (x) .
Thus the cocycles g = {gij } and ge = {e
gij } are cohomologous in U.
Conversely, let g = {gij } and ge = {e
gij } be cohomologous in U. The map
f
fx =
ej,x j (x) 1
j,x : Fx Fx
is independent of j; namely, the right hand side is equal to
ej,x geji (x) i (x) 1
x Ui Uj .
The pair ({fx }, idB ) is a bundle map, since conditions 2.3.3 (C1) and (C2)
are obtained from
75
Not every ber bundle has an atlas for a given cover. For this reason, we
wish to compare dierent covers.
Let U = {Uj | j J} and mathcalV = {Vk | k K} be open covers of B.
Let mathcalV be a renement of U, i.e., there exists a function : K J
with Vk U(k) for every k K. Let g = {gij | i, j J} be a cocycle for U
with coecients in G. By
hkl = g(k)(l) |Vk Vl ,
k, l K ,
we dene a new cocycle # (g) = {hkl | k, l K} for mathcalV with coecients in G. This is the cocycle induced by the renement.
2.4.6 Definition. Let g and ge be cocycles for the covers U = {Uj | j J}
ei | i J}
e with coecients in G. We say that g and ge are
and Ue = {U
cohomologous in B if there exists a common renement mathcalV = {Vk |
k K} of U and Ue and rening functions : K J and
e : K Je
#
#
such that (g) and
e (e
g ) are cohomologous in mathcalV .
Cohomology in B is an equivalence relation. Reexivity and symmetry
are clear. Transitivity will be proved inside the proof of Theorem 2.4.7,
although it is an easy exercise to prove it directly.
We denote by [g] the corresponding equivalence class and call it cohomology class of g.
Let H 1 (B; G) be the set of cohomology classes of cocycles for covers of B
with coecients in G. Let kG (F, B) be the set of equivalence classes (over
B) of ber bundles over B with ber F and structure group G.
2.4.7 Theorem. If to each ber bundle, the cohomology class of the cocycle
consisting of its coordinate transformations is assigned, there is a bijection
: kG (F, B) H 1 (B; G) ,
induced by mapping each ber bundle to the cohomology class of the cocycle
determined by its coordinate transformations.
For the proof, we need some previous considerations.
Let F be a set bundle over B with atlas A = {j | j J} for the cover
U = {Uj | j J}. Let mathcalV ={Vk | k A} be an open renement of U
and : K J the rening function, (i.e. Vk U(k) ).
76
2 Fiber Bundles
Dene
k = {(k),x | x Vk }
# A = {k | k K} .
2.4.8 Lemma. The following statements hold:
(a) # A is an atlas equivalent to A.
(b) If the cocycle g consists of the coordinate transformations of A, then
# g consists of those of # A.
Since the proof is simple, we leave it to the reader.
Proof of 2.4.7:
is well dened:
Let F and Fe be set bundles over B with ber F . Let A and Ae be atlases
for F and Fe with respect to the group G and with covers U = {Uj | j J}
ei | i J}.
e Then
and Ue = {U
ei | (j, i) J J}
e
mathcalV = {Uj U
e As rening functions we have the projecis an open renement of U and U.
tions
e
e(j, i) = i
e kk5 J ,
kk
kkk
J Je SS
SSS
SS)
J,
(j, i) := j .
77
e A)
e are equivalent
2.4.8 (b) and 2.4.5) and so, by 2.4.8 (a), (F, A) and (F,
over B too.
is surjective:
This is exactly the statement of Theorem 2.4.3.
= (F, G, B; F, A) ,
are called associated if the cocycles consisting of their coordinate transformations are cohomologous in B; that is, if the images of their equivalence
classes under
: kG (F, B) H 1 (B; G)
and
e : kG (Fe, B) H 1 (B; G)
coincide.
2.4.10 Definition. Let : G H be a continuous homomorphism of
topological groups. If {gij } is a G-cocycle, then { gij } is an H-cocycle,
as one deduces from (CT1). The assignment {gij } 7 { gij } is compatible
with the cohomology relation and determines a function
: H 1 (B; G) H 1 (B; H) .
A geometric interpretation of is the following. Let : G , H be the
inclusion of a subgroup. If H acts eectively on F and G acts by restricting
the action of H, then one may clearly consider a bundle = (F, G, B; F, A)
as a bundle with structure group H. However, in this case, does not have
to be injective. By passing to the larger group H two nonequivalent bundles
may become equivalent, as we show below in the case of the twisted torus
(cf. 2.4.11, 3).
2.4.11 Examples.
1. Using the method shown in 2.3.2 and the local trivializations of 1.2.9(a)
one may assign to the Moebius strip a set bundle and two local charts.
78
2 Fiber Bundles
1
0
1
0
1
0
1
0
U0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
U1
U0 U1
1
0
1
0
1
0
1
0
Figure 2.2
The only nontrivial coordinate transformation is
{
idS1 if b < x < 1,
g0 1 (x) = 1
0,x 1,x =
g
if 0 < x < b.
The group G = {id, g}
= Z2 is again the structure group. The twisted
torus is associated to the Klein bottle and therefore, it is nontrivial,
since one of two associated bundles is trivial if and only if the other is
also trivial.
If we take 1,x to be also the identity, then instead of the twisted torus
we obtain the trivial bundle. Dening
fx : Fx Fx
s 7 dx s ,
79
2.4.1
Vector Bundles
80
2 Fiber Bundles
81
2.5
Principal Bundles
xB
EG
proj1
/E
p
/B
82
2 Fiber Bundles
is commutative; that is, for every x E and g G, one has that p(xg) =
p(x).
A ber map (f, f ) : p p between principal G-brations is called a
principal map if
f (zu) = f (z)u , z E , u G ,
in other words, if the map f : E E is equivariant.
Again, principal G-brations and principal maps build a category.
The trivial bration proj1 : B G B, together with the action :
(B G) G B G given by ((x, v), u) = (x, vu) is a principal bration
called the trivial principal G-bration. If A B, then : E G E
induces a map A : (p1 A) G p1 A that equips pA : p1 A A with
the structure of a principal bration. We shall denote it again by pA .
A principal bration (p, ) is called locally trivial, if for every z B, there
is a neighborhood U of z and a principal equivalence (that is, a principal map,
that is an equivalence) between (pU , U ) and the trivial principal G-bration
over U .
2.5.3 Theorem. The assignments
7 (p , )
(f, f ) 7 (fb, f )
determine a functor from the category of principal G-bundles to the category
of locally trivial principal G-brations. In particular, they assign to the trivial
principal bundle, the trivial principal bration (cf. 2.3.22).
Proof: First we show that is continuous. For that we recall the denition
of the topology of E, 2.3.14. Let A be an atlas for . In the diagram
A {}
U G G
idG
EG
A {}
U G
/ E,
83
II
II
II
I
proj1 II$
/ p1 U
ww
ww
w
ww p
w{ w U
U .
If is trivial, that is, if its atlas consists of just one chart , then the previous
considerations imply that p is trivial, through the trivialization
= : U G = B G p1 U = E .
By 2.3.22, we still have to prove that the determined ber maps are principal
maps, that is, that they are equivariant. Namely, fb(zu) = fb(z)u, z E , u
G. Take pz = p(zu) = x U and f x = y U . Then fx = y g(x) 1
x ,
and all three maps on the right-hand side are equivariant, i.e., compatible
with the right action of G, (cf. 2.5.1). From 2.3.19 one has
fb(zu) = fx (zu)
= y g(x) 1
x (zu)
= (y g(x) 1
x (z))u
= fx (z)u
= fb(z)u .
for
z Gx .
If y = f (x), then one has fx (z) Gy , since (h, h) is a ber map. We have to
prove that fx is bijective and that f = {fx } is compatible with the atlases.
Take A, A , and v G. Then
(y1 fx x )(v) = (y1 hx )(ev)
= (y1 hx (e))v ;
the last equality holds, since by assumption, y1 , h, and x are compatible
with the right action of G, (see 2.5.1, 2.5.2). Now g(x) = y1 hx (1) G
84
2 Fiber Bundles
denes a bijective map from G into itself (left translation by g(x)); since y
and x are bijective, so is also fx . We still have to prove (cf. 2.3.3 (C2)) that
1
the mapping x 7 g(x) is continuous on U f U . This follows, since we
may write g as the following composite of continuous maps:
U f
/ (U f 1 U ) G
/
x
(x, 1)
/ p1 (U )
/
f x (1)
(| )1
/ U G
/ (y, 1 f x (1))
y
/ p1 (U f 1 U )
/ x (1)
proj2
/G
/ 1 f x (1) .
y
Finally, fb = h is clear.
85
Figure 2.3
2.5.8 Theorem. For each locally trivial principal bration p : E B,
with group action : E G E, there is a unique principal bundle such
that p = p , = .
Proof: Let {Uj | j J} be an open cover of B and let pUj be trivial. Assume
the principal maps
j : Uj G p1 Uj
describe the local triviality. We have to dene = (G, G, B; G, A). Obviously, we have to set Gx = p1 (x). A = {j }, j = {j,x | x Uj }, and
j,x : G Gx will be given by j,x (u) = j (x, u), u G. j,x is bijective,
since j is a homeomorphism. Since j is compatible with the right action
of G, j,x is also bijective, and one has
1
i,x
j,x (v) = 1
i,x j,x (e)v,
x Ui Uj ,
space E = xB Gx recovers its original topology. This is, in fact, the case,
since the map
:
Uj G {j} E ,
which according to 2.3.14 has to be constructed, must be an identication,
which is even an open map if one takes in E the original topology. On the
other hand, = , since j,x is compatible with the right action (see 2.5.1).
All this shows the existence of a bundle with the desired properties. Let
86
2 Fiber Bundles
87
The following special case is easy to grasp. Take E to be a Lie group and
G a closed subgroup, (cf. Chevalley [2, p. 110, Prop. 1]).
2.5.11 Definition. Let = (G, G, B; G, A) be a principal bundle and H a
closed subgroup of G. We dene a bundle
e B; F, A)
e
/H = (G/H, G,
as follows. H acts on every Gx on the right. Consider the equivalence relation
in Gx given by z1 z2 if there exists h H such that z1 = z2 h. Let Gx /H
be the set of equivalence classes. Setting Fx = Gx /H we may dene
Ae = {
e | A} ,
e = {
ex | x U }
such that
ex : G/H Fx
is the bijection canonically induced by x . Thus ex1
ex : G/H G/H
1
is induced by x x = g(x) G.
If the natural action (2.2.18) G G/H G/H is eective, then /H
e = G is a ber bundle.
with this action and structure group G
If this action is not eective, then from uvH = vH (v G, u G), one
/
U G
U G/H
e.
/ E
88
2 Fiber Bundles
2.5.14 Definition. Let = (F, G, B; F, A) be a ber bundle. The princie associated to is described as follows.
pal bundle e = (G, G, B; G, A)
We say that a map f : F Fx is admissible if 1
x f G for A
and x U .
This denition is independent of the choice of such that x U .
Namely, take
Gx = {f | f : F Fx is admissible} ,
ex : G Gx
given by
v 7 (F F Fx ) ;
that is,
ex (v) = x v. Since 1
x (x v) = v G, x v is admissible
and thus it lies in Gx .
ex is bijective. From
ex1
ex (v) = x1 x v = g(x)v ,
it follows that
Ae = {
e | A}
is an atlas and e is associated to (2.4.9).
2.5.1
Stiefel Manifolds
89
Via
On /Onk VS n,k ,
(v1 , . . . , vn ) 7 (v1 , . . . , vk )
since On is compact and VS n,k is Hausdor. We identify both spaces through
this homeomorphism.
Mapping (x1 , . . . , xl ) to
/ VS n,k
(2.5.17)
On /Onl
On /Onk .
90
2 Fiber Bundles
H0 =
BOnl B 1 .
BOnk
2.6
In this section we show how the principal G-bration determined by a principal G-bundle relates to the bration determined by an associated G-bundle
with an arbitrary ber F .
2.6.1 Definition. Let G be a topological group, E a right G-space, and F
a left G-space. There is a left action of G on E F given by
E F G E F ,
(x, y, g) 7 (xg 1 , gy) .
91
NNN
NNN
NNN
NN&
B,
/BF
ww
ww
w
ww
{ww proj1
92
2 Fiber Bundles
e = ( G idF ) 1 : U F q 1 U = p1 U G F .
Since both 1 and G idF are homeomorphisms,
e is one too. Using 2.6.3,
one easily veries that q
e = proj1 : U F U ; in other words, the
diagram
U F
GG
GG
GG
G
proj1 GG
#
/ p1 U G F
s
sss
s
s
ss
sy ss
commutes.
:
U G {} E ,
A
93
U F {} E .
G idF :
U G {} G F E G F ,
A
:
U G {} G F
U F {}
A
HH
HH
H
q HHH
$
/ E
~
~
~~
~~ q
~
~~
2.7
Induced Bundles
Given a ber bundle over a space B and a map A B, we study here how
this map induces a ber bundle over A.
2.7.1 Definition. Let = (F, G, B; F, A) be a ber bundle and : A
B a continuous map. We dene a new ber bundle
e A)
e
() = (F, G, A; F,
by the following:
fa = F(a)
F
fa | a 1 (U )}
e = {
ea : F F
ea = (a)
Ae = {
e | A}
94
2 Fiber Bundles
It is easy to check that 2.3.1 (B1)(B3) hold. The ber bundle () is called
the ber bundle induced by through the map .
In case that : A , B is an inclusion, the induced bundle () is called
the restriction of to A and is usually denoted by |A.
2.7.2 Exercise. Prove that through a constant map, a trivial bundle is
induced.
In what follows, we analyze the relationship between theprevious denition and that of an induced bration (see 1.4.22).
e A be the bration induced by p = p through : A B.
Let E
e = {a} Fa is equal to E () . (Observe that for the
Then, as sets, E
aA
construction of the bration corresponding to , one has to provide that the
bers are mutually disjoint; see 2.3.14). However, it is also true that both
e and E () have the same topology. For seeing this, one has to prove that
E
e : Ue F = 1 U F E
e is a homeomorphism over some
the map
by
fa F(a) .
a = id : F
95
and
= (F, G, B; F, A) .
Then there exists a unique bundle map (h, idA ) : b () such that the
diagram of bundles
b PP
(h,idA )
()
PP(f,)
PPP
PP(
m6
mmm
mmm( ,)
is commutative.
Proof: The commutativity of the diagram requires to dene ha = fa , from
where the uniqueness of h follows. The so-dened map h determines a bundle
map, since
1
ea1 ha
ba = (a)
fa
ba
lies in G and depends continuously on a A, because (f, ), by assumption,
is a bundle map.
96
2 Fiber Bundles
2.7.1
Functional Bundles
b A)
b be ber bundles. We call a
Let = (F, G, B; F, A) and b = (F, G, A; F,
bundle map h : Fba Fb , a A, b B admissible if the homeomorphism
1
ba lies in G. This denition is independent of the choice of the
b h
charts and
b such that a Ub, b U , as follows from
b
b1 h
ba = (b1 b ) 1
ba (
b1
a a ) ,
b h
since each of the compositions in parentheses lies in G, by denition of an
atlas.
b ) is a bundle e = (G, G,
e A
2.7.10 Definition. The functional bundle Apl(,
e A),
e where
B; F,
Fe(a,b) = {h : Fba Fb | h is admissible}
with atlas
such that
Ae = {(,
b ) Ab A}
(,
b )(a,b) : G Fe(a,b) ,
(a, b) Ub U
is given by
v 7 b v
b1
a .
b ) is well dened, we have to
In order to check that the bundle Apl(,
prove that b v
b1
Fe(a,b) ; that is, that it is admissible, and that
a
(,
b )(a,b) is bijective.
The former follows from
1
b1
a ) a = v G ,
b (b v
and the latter from the fact that one has an inverse of (,
b )(a,b) given by
ba .
v 7 1
b v
e We have
Now the question is if A is an atlas for some adequate group G.
1
b
b )1 (,
b1
(,
a a
(a,b) b )(a,b) (v) = b b v
1
= g (b) v g
b
b(a)
(
)
= g (b), g
(a),
v
b
b
if we dene
: (G G) G G
97
b
is obviously continuous.
b ), the bundle map
Let (f, ) : b be a bundle map. If e = Apl(,
determines a map s : A Ee given by
(
)
A a 7 fa : Fba F(a) Fe(a,(a))
that, by denition of a bundle map, is admissible. One has that pes(a) =
(a, (a)).
2.7.11 Lemma. Let : A B be a continuous map. The assignment
(f, ) 7 s given above yields a one-to-one relation between bundle maps
(f, ) : b and continuous maps s : A Ee such that pes(a) =
(a, (a)).
Proof: We prove that f is compatible with the atlases Ab and A if and only
if s is continuous. That the map is bijective is obvious. The diagram
Ub 1 U
QQQ
QQQ
QQ
s QQQQ
(
/ Ub U G
mmm
mmm
m
m
m e
vmmm
pe1 (Ub U )
commutes if
e b, v) = (,
g(a) = (a, (a), 1
ba ) and (a,
b )(a,b) (v) = b v
b1
a .
(a) fa
98
2 Fiber Bundles
assertion follows.
2.7.12 Theorem. Let 0 , 1 : A B be homotopic maps. If A is a CWcomplex and is a bundle over B, then the induced bundles 0 () and 1 ()
are equivalent over idA .
Proof: Let b = 0 (). We shall prove that there is a bundle map (f, 1 ) :
b . By 2.7.7, the bundle 1 () is equivalent to b over idA . Let t be a
homotopy between 0 and 1 . Consider
s0
w
ht
Ee
w;
pe
/ AB,
with the homotopy ht (a) = (a, t (a)). For t = 0 the diagram commutes if s0
corresponds to the bundle map (0 , 0 ) : b as in 2.7.11 (see 2.7.4). If
A is a CW-complex and pe is locally trivial, then by 1.4.8 and 1.4.9, we can
lift ht , with the initial condition s0 , say to a map st . s1 gives us by 2.7.11 a
bundle map (f, 1 ) : b Ee.
2.7.15 Note. The proof of Theorem 2.7.12 required the lifting of a certain
homotopy. The theorems of Dold [3], recalled in 1.4.14, allow us to weaken
the assumptions. One may either assume the space A to be paracompact,4
or the ber bundle to be numerable, i.e., such that it has an atlas whose
corresponding cover is numerable. In this latter case, the induced bundle
0 () and the functional bundle are numerable.
4
99
2.8
Universal Bundles
In 2.7 we saw that homotopic maps induce equivalent ber bundles. We now
ask the opposite question. Namely, if there is a bundle over an adequate
space B such that every bundle over A is induced through a map : A B.
Moreover, we ask if it is possible to choose and B in such a way, that the
equivalence of the induced bundles implies that the maps through which they
are induced are homotopic.
2.8.1
/5
k kE
k
k
r
p
k k
k
k
k
k
/ B.
Bk
Sk1
_
100
2 Fiber Bundles
The composition sk1 (|Sk1 ) is well dened because the section sk1 is
dened on (Sk1 ), since (Sk1 ) B k1 . In case that we can nd the
lifting r, we can extend sk1 to a section s over A B k1 ek by giving it
by
{
sk1 (x) if x A B k1 ,
s (x) =
r1 (x) if x ek .
One can easily check that s is well dened and continuous, since is an
identication, and that it is a section. We need conditions in order for r to
exist.
2.8.4 Lemma. In the diagram
e
E
pe
/E
p
/ B,
let pe = (p) be the bration induced by p through and (, ) the corresponding ber map (1.4.22). The assignment se 7 se denes a bijective
function
{sections of pe} {liftings of }.
e be
Proof: If r : A E is a lifting of , namely, if p r = , let ser : A E
e since (a) = pr(a)). The mapping
the section a 7 (a, r(a)), ((a, r(a)) E,
r 7 ser is the inverse of se 7 se.
where pe = (p)
According to 2.8.4 the lifting sk1 (|Sk ) of i corresponds to a section
e of pe over Sk1 (more precisely, to a section in the bration
t : Sk1 E
induced by i, which can be interpreted as restriction of pe).
() is a bundle over a contractible CW-complex; therefore, it is trivial
(2.7.14). Thus, also pe is trivial. Consequently, there is a homeomorphism f
101
proj2
/
o7 F
OOO
o
o
OOO
proj1
O
o go
o
pe OOOO
O' o o
BO k
/ Bk F
?
Sk1
2.8.5 Lemma. The section t can be extended to a section over Bk if and
only if proj2 f t : Sk1 F can be extended to Bk .
e be an extension of t; then proj2 f t is an extension
Proof: Let t : Bk E
of proj2 f t. Conversely, let g : Bk F be an extension of proj2 f t.
e be given by t (z) = f 1 (z, g(z)). Clearly, t is a section
Let t : Bk E
that extends t.
Proof of 2.8.2: We proceed by induction over (the dimension of) the skeletons.
There exists always an extension s0 of s to A B 0 , since the 0-cells of
B A constitute a discrete subspace. Let sk1 be a section over A B k1 ,
(k 1), that extends s.
0
k1
102
2 Fiber Bundles
proj
E A B 1 A
be the bration determined by .
The bundle is dened as follows:
B = A
F(a,b) = {h : Fa Fb | h is admissible and b is arbitrary} ,
Fa =
bB
G = G ,
a : F = E Fa given by
(v : F Fb ) 7 (v1
a : Fa Fb ) ,
= {a | a U } .
We have to check again that all these elements give us a ber bundle. On
the way, we shall describe the action of G on E = F .
a1 a (v) = v 1
a a
= v (g
(a))1
= (g
(a), v)
103
2.8.9 Definition. The ber bundle is called the partial functional bundle
of (, ). We denote it by Apl1 (, ).
The equation
(2.8.10)
E =
Fa =
aA
F(a,b)
= E
aA bB
A B proj
/A
1
commutes.
2.8.11 Lemma. The spaces E and E have the same topology.
Proof: Let and be local charts corresponding to the open sets U A
and V B. Consider the diagram
U V G
(2.8.12)
id
/ U p1 (v)
/ U E
p1
(U V )
/ p1 (U B)
p1
(U ) .
Diagram (2.8.12). If varies along the atlas A , we obtain that the sets
1
U p1
(V ) build an open cover of U E , and that the sets p (U V )
build an open cover of p1
(U ). Hence, we obtain the assertion of the lemma.
104
2 Fiber Bundles
2.8.2
n-Universal Bundles
We characterize here bundles that are universal for ber bundles over spaces
B of bounded dimension.
2.8.14 Definition. Let be a ber bundle and the associated principal
bundle. will be called n-universal (n ) if i (E ) = 0 for i < n. The
determined bration p is also called n-universal.
2.8.15 Theorem. Let be an n-universal ber bundle, A a CW-complex of
dimension n, A0 A a subcomplex and a ber bundle over A. Then any
bundle map |A0 can be extended to a bundle map .
Proof: Note that Apl1 (|A0 , ) = Apl1 (, )|A0 . With this remark and 2.8.13,
using the bundle map |A0 , we obtain a section s0 : A0 E
( = Apl1 (, )) of p over A0 . By assumption, i (F ) = 0 for i < n, since
F = E . Theorem 2.8.1 guarantees that we can extend s0 to a section
s : A E . From s, we obtain a bundle map (2.8.13) that extends the
given bundle map |A0 .
2.8.16 Definition. Let be a ber bundle. We may assign to each homotopy class [] [A, B ] an equivalence class of ber bundles over A; namely
[] () = [ ()] ,
where [] denotes the homotopy class of the map and [] denotes the
equivalence class of the bundle .
By 2.7.12, this assignment is well dened (if A is a CW-complex). Denote
by kG (A) the set of equivalence classes of ber bundles over A with ber F
and structure group G, and let () : [A, B ] kG (A) be the function just
dened.
2.8.17 Theorem. Let be an (n + 1)-universal ber bundle and A a CWcomplex such that dim(A) n. Then the function () : [A, B ] kG (A)
is bijective.
105
(i ,i )
|A {} o
(f , )
3/ ,
()
[A , B ]
/ kG (A)
()
/ kG (A )
106
2 Fiber Bundles
2.9
2.9.1
Grassmann Manifolds
G/H
= Ok
G/H0
= Ok .
The action G/H0 G/H G/H corresponds, via the canonical identications (2.9.2)(2.9.3), to the group multiplication. Thus, is a principal
107
/ VS n,k
On /Ok Onk
pn,k
/ Grn,k ,
mn,k
VS n,k Ok VS n,k
108
2 Fiber Bundles
by
((x1 , . . . , xk ), A) 7 (x1 , . . . , xk )A ,
where the k-tuple (x1 , . . . , xk ) should be seen as a matrix with k columns
xi ; the columns of the product form an orthonormal k-frame that generates the same plane as (x1 , . . . , xk ), and this is the k-frame we refer to by
(x1 , . . . , xk )A.
Since the pair (f, f ) is a principal map, by denition of the actions one
has
(ABe1 , . . . , ABek ) = (Ae1 , . . . , Aek )B ,
and, obviously, both sides are the rst k columns of the product matrix AB.
The (-)universal Ok -bundles are obtained by passing to the colimit. Let
us consider the diagram
VS n,k
pn,k
/ VS n+1,k
pn+1,k
Grn,k
/ Grn+1,k
/ ,
n VS n,k is such that the canonical map from the topological sum
n VS n,k
109
/ VS ,k
VS n,k
pn,k
Grn,k
p,k
/ Gr,k
VS n,k
/ VS n+1,k Ok
mn+1,k
/ VS n+1,k
Besides, both spaces have the same topology, as one sees in the commutative
diagram
a
/ VS ,k Ok
( VS n,k ) Ok
O
(VS n,k Ok )
/ colim(VS n,k Ok ) ,
110
2 Fiber Bundles
Rnk
E
Rk
Figure 2.4
as follows (cf. 2.5.5). First we need a section
s : Un,k p1
n,k Un,k ,
and we dene
n,k ([x1 , . . . , xk ], B) = (s[x1 , . . . , xk ])B ;
the section is obtained as follows. Each plane E Un,k is generated by
exactly a k-tuple x1 (E), . . . , xk (E) of vectors of the form
x1 (E) = (1, 0, . . . , 0, xk+1,1 , . . . , xn,1 )
x2 (E) = (0, 1, . . . , 0, xk+1,2 , . . . , xn,2 )
..
..
.
..
Observe that the n-tuples (0, . . . , 1, . . . , 0, xk+1,i , . . . , xn,i ) are the solutions
of a system of n k linear equations with n unknowns. They are clearly
linearly independent.
The assignment E 7 (x1 (E), . . . , xk (E)) Rnk is continuous (cf. Milnor,
op. cit). Moreover, the orthonormalization (x1 (E), . . . , xk (E)) 7 (e
x1 (E), . . . , x
ek (E))
given by the Gram-Schmidt process is also continuous (namely, one can give
explicit formulas for the orthonormalized basis; cf. for instance, the formulas
given by Langwitz [?, p.74]). We thus may dene
s(E) = (e
x1 (E), . . . , x
ek (E)) VS n,k .
It is an easy matter to convince oneself that all maps n,k for dierent values
of n are compatible and glue together to yield a map
,k : U,k Ok p1
,k U,k
and the map ,k , as a colimit of homeomorphisms, is also a homeomorphism. Of course, it is also a principal map over the identity map of U,k .
Thus, p,k is trivial over U,k .
111
(d) We prove that i (VS ,k ) = 0 for every i. This follows from the next
lemma.
2.9.10 Lemma. Any compact set K VS ,k lies inside some VS n,k .
Before proving this lemma, we come back to statement (d) in the proof
of 2.9.9.
Let f : Si VS ,k represent any element of i (VS ,k ). Its image f (Si )
is a compact set, and thus, by 2.9.10, it lies inside VS n,k for some n. Let n
be large enough that n k > i. Hence, f is nullhomotopic as a map into
VS n,k , and thus also as a map into VS ,k . This proves the statement.
Proof of 2.9.10: If the statement of the lemma were false, then there would
be a sequence p1 , p2 , . . . of points of K such that pn VS n,k for all n. But
since K is compact, the sequence {pi | i N} has an accumulation point p0 .
2.9.2
Let G be an arbitrary topological group. We want to construct a universal principal G-bration pG : EG BG. First we shall give a formal
description of pG and then we shall explain the geometrical meaning of the
construction.
2.9.11 Construction. First we describe EG as a set.
Consider sequences
(t1 , v1 , t2 , v2 , . . . , ti , vi , . . .)
such that
ti I = [0, 1] ,
vi G ,
i = 1, 2, 3, . . . ,
112
2 Fiber Bundles
EG
/ EG
ti
/I
ti
113
t1
i (0, 1] G
/ t1 (0, 1]
i
GG
vi
/ G,
a b u G with (a, u) = b ,
(see 2.2.21).
Let pG : EG BG be the quotient map. Then pG is a principal
bration.
2.9.13 Theorem. pG is a locally trivial principal bration. We denote by
G the corresponding ber bundle.
Proof: Let us consider the sets Wi = t1
i (0, 1] and Vi = pG Wi . Wi is open
in EG by the denition of the topology in EG. Vi is open in BG, since
1
pG
(Vi ) = Wi . {Vi | i = 1, 2, . . .} is an open cover of BG. We shall prove
that pG is trivial over Vi .
We dene maps i : Vi G Wi by
i (pG (t1 v1 + t2 v2 + ), u) = t1 (v1 vi1 u) + t2 (v2 vi1 u) + .
We show that they are homeomorphisms.
i is well dened.
Namely, if pG (t1 v1 + ) = pG (t1 v1 + ) Vi , then one has tj = tj , and
there exists w G such that vj = vj w for every j such that tj = 0. Thus
vi = vi w and vj vi1 u = vj ww1 vi1 u = vj vi1 u for such values of j.
i is bijective.
Namely,
(pG , vi ) : Wi = t1
i (0, 1] Vi G
is an inverse of i and is continuous.
i is compatible with the action of G.
114
2 Fiber Bundles
Namely,
i (pG (t1 v1 + ), u1 u2 ) = t1 (v1 vi1 u1 u2 ) +
= t1 (v1 vi1 u1 + )u2
= i (pG (t1 v1 + ), u1 )u2 .
i is continuous.
Namely, |Vi {e} is continuous, since
i (pG (a), e) = (a, vi (a)1 ), a Wi ,
and since
i (x, u) = (i (x, e), u),
i is also continuous.
We still have to prove that i (EG) = 0 for every i. Before doing it, we
explain the construction of EG.
Consider inside EG the subset
E k G = {t1 v1 + + tk vk + + tj vj + | tj = 0 if j > k} .
For example, a point t1 v1 +t2 v2 E 2 G can be described by a triple (t, v1 , v2 ),
t = t1 (1 t = t2 ), since t1 + t2 = 1, where the triples
(0, v1 , v2 ) and (0, v1 , v2 )
are identied, as well as are the triples
(1, v1 , v2 ) and (0, v1 , v2 ) .
In other words, E 2 G, as a set, is the join G G (see page 10), up to the fact
that the topology of G G might be ner (larger). Analogously, one may see
that, up to topology, E 3 G can be considered as (G G) G, and so on.
2.9.14 Theorem. i (EG) = 0 for every i 0.
Proof: Dening
sk (t1 v1 + t2 v2 + ) =
j=1
tj ,
115
Take Uk = s1
Uk+1 for k 1, and
k (0, 1]. Then, Uk
k=1 Uk = EG, since
t+(1t)sk (a)
tj (a)
sk (a)
if j k,
(1 t)tj (a)
if j > k,
vj h(a, t) 7 vj (a)
if tj h(a, t) > 0 .
j=1
tj (h(a, t)) =
t + (1 t)sk (a)
sk (a) + (1 t)(1 sk (a))
sk (a)
= 1,
so that, indeed, h(a, t) EG.
By 2.9.12, h is continuous. One has that h(a, 0) = a and h(a, 1) E k G (=
k
k+1
s1
G given
k (1)). We now attach to h another homotopy d : E GI E
by
(1 t)tj (a)
tj d : (a, t) 7 t
0
{
vj (a)
vj d : (a, t) 7
e
if j k,
if j = k + 1,
if j k + 2;
if j k,
if j k + 1.
116
2 Fiber Bundles
[A, BG]
#
[A, BG ]
/ kG (A)
/ kG (A) .
(G )
Let Vi BG be the open set dened in the proof of 2.9.13, and corre1
spondingly, Vi BG . One has that (Vi1 ) = Vi . Let [f ] = [A, BG].
(G )[f ] is represented by a bundle with open cover {f 1 (Vi )} and coordinate transformations
f
gij
f 1 (Vi ) f 1 (Vj ) Vi Vj G G ,
gij
where gij and gij are the coordinate transformations of G and G , respectively.
1
Chapter 3
3.1
Introduction
3.2
Spectral Sequences
3.2.1
Additive Relations
We shall introduce here the concept of the subtitle, since it is a very convenient formalism for studying spectral sequences.
3.2.1 Definition. Let A and B be abelian groups (which we write additively). A relation f : A B is a triple f = (A, B, F ) such that F A B
(cf. 2.3.23). We say that the relation f is additive if f is a subgroup of A B.
If f is on the one hand a function of sets, and on the other, an additive relation, then f is a group homomorphism.
In what follows, we shall only consider additive relations. Let f =
(A, B, F ) and g = (B, C, H) be relations. We dene the composition g f as
the triple (A, C, H), where
H = {(a, c) A C | b B with (a, b) F, (b, c) G} .
The relation g f is additive again. Abelian groups, together with additive
relations, constitute a category. Given a relation f = (A, B, F ), we dene its
inverse relation by
f 1 = (B, A, F 1 ) , where F 1 = {(b, a) B A | (a, b) F } .
This relation f 1 is also additive. One has the following formulas:
(g f )1 = f 1 g 1 ,
(f 1 )1 = f .
118
3 Homology of Fibrations
(b) f ( jJ Uj ) = jJ f (Uj ).
119
AO
/B
_
_
_
/
Def(f )
B/Ind(f ) ,
?
3.2.2
There are several approaches to spectral sequences. We chose here the classical one through exact couples invented by Massey [10]. Before stating the
denition, we need some previous concepts.
3.2.8 Definition. Let A = {Ap,q | (p, q) Z Z} and C = {Cp,q | (p, q)
Z Z} be families of abelian groups. We dene a homomorphism
h : A C
of bidegree (r, s) as a family of homomorphisms
hp,q : Ap,q Cp+r,q+s ;
we denote this fact by bideg(h) = (r, s). A and C are called bigraded groups.
The elements of Ap,q are said to have bidegree (p, q). Let k : C D be
another homomorphism of bidegree (u, v). We say that a sequence
h
A C D
is exact at C if for every (p, q), the sequence
hp,q
kp+r,q+s
120
3 Homology of Fibrations
A _@
@@
@@
@
k @@
(3.2.10)
/A
~~
~
~
~~
~~ j
Cp,q+1
+3 Ap1,q+1
/ Cp1,q+1
+3 Cp,q
/ Ap2,q+1
(3.2.11)
Ap,q
+3 Ap1,q
Ap+1,q1
Ap+q
The double arrows show the intertwined exact sequences. Let Ap,q be the
colimit of the sequence
i
Ap1,q+1 Ap+1,q1 .
Thus we have a system of homomorphisms
ip,q = i : Ap,q An ,
p, q Z ,
n = p+q,
CC
CC
CC
C
i C!
An
is commutative.
/ Ap+1,q1
v
vv
vv
v
v
{vv i
121
: Ap,q An ) = An .
(b)
p+q=n Im(i
Recall that for each sequence of (abelian) groups, there exists a colimit, and
it is unique up to isomorphism (see [1]).
3.2.12 Definition. Let G be a group and {Fp G | p Z} a sequence of
subgroups such that Fp G Fp+1 G; we refer to {Fp G} as a ltration of the
group G.
3.2.13 Construction. In what follows, for simplicity, we shall assume that
Ap,q = 0
if
p < 0.
the groups
Im(f ), Ind(f ), Ker(f ), and Def(f ) ,
(as dened in 3.2.6).
Im(f ) = ij 1 (Cp,q ) = i(Def(j)) = i(Ap,q )
Ind(f ) = ij 1 (0) = i(Ker(j)) = i(i(Ap1,q+1 ))
= i(Ap1,q+1 ) .
By
i(Ap,q ) = Fp Ap+q ,
and since Ind(f ) Im(f ), we obtain a ltration of Ap+q , because we have
proved
Ind(f ) = Fp1 Ap+q ,
(3.2.14)
Im(f ) = Fp Ap+q .
We also have
(
Ker(f ) = Ind f
(3.2.15)
= ji
=j
(
( ))
(0) = j Ker i
)
(ir )1 (0)
r=0
r=0
j (ir )1 k (0) =
j (ir )1 (0)
r=0
r=0
(
)
Ind j (ir )1 k
122
3 Homology of Fibrations
(3.2.16)
ir j 1 (0)
r=0
1 r 1
k ij
(0) =
(
)
Ker j (ir )1 k
r=0
r=0
and
r=0
ir j 1 (0) =
r=0
= 0,
since by assumption Ap,q = 0 if p < 0. These computations can be gured
out in a diagram similar to (3.2.11).
1
In the expressions for Ker(f ) and Def(f ) the relations j(ir1 ) k play a
role.
3.2.17 Definition. Take r 1, r Z, i0 = idA .
drp,q = j(ir1 )1 k : Cp,q Cpr,q+r1 ,
r
Zp,q
= Def(drp,q ) ,
r
Bp,q
= Ind(drp+r,qr+1 ) ,
r
Zp,q =
Zp,q
,
Bp,q
=
r
Ep,q
r=1
r
Bp,q
,
r=1
r
r
Zp,q
/Bp,q
1 r .
123
r 1
= j(i ) (0)
= j(ir )1 k(0)
= Ind(dr+1
p+1,q+1 ) ,
Ker(dr )p,q = (j(ir1 )1 k)1 (0)
= k 1 (ir1 )j 1 (0)
= k 1 (ir1 )i(Apr1,q+r )
1
= k (i )j
=
=
(Cpr1,q+r )
r+1 1
(dp,q ) (Cpr1,q+r )
Def(dr+1
p,q ) ,
Applying again the complete diagram as in (3.2.11), one can rewrite what
we just proved in the following chain of inclusions (since Ind(f ) Im(f ) and
Def(f ) Ker(f )).
0
Ind(d1 )
/ Im(d1 )
/ Im(d2 )
/
/ Ker(d2 )
/ Ker(d1 )
/C
/ Ind(d2 )
/ Ind(d3 )
/
/ Def(d3 )
/ Def(d2 )
/ Def(d1 ) .
In other words,
0 = B1 B2 B3 Z 3 Z 2 Z 1 = C .
In particular, one has
B r+1 = Im(dr ) Def(dr ) = Z r ,
B r = Ind(dr ) Ker(dr ) = Z r+1 ,
124
3 Homology of Fibrations
Def(dr )
E = Def(dr )/Ind(dr )
r
/ C/Ind(dr )
O
?
/ Def(dr )/Ind(dr ) = E r
O
?
/ Im(dr )/Ind(dr )
Def(d )/Ker(dr )
r
r
r
is a homomorphism.
Epr,q+r1
Hence dp,q : Ep,q
r
= Ker(drp,q )/Im(drp+r,qr+1 )
r+1
= Def(dr+1
p,q )/Ind(dp+r+1,qr )
r+1
r+1
r+1
= Zp,q
/Bp,q
= Ep,q
.
Ep,q
= Fp Ap+q /Fp1 Ap+q ,
125
r=0
Ker(dr+1 ) = Zp,q
,
Ind(dr+1 ) = Bp,q
,
r=0
3.3
3.3.1 Construction. Let : E B be a Serre bration over a CWcomplex B. Denote by B p the p-skeleton and by E p its inverse image under
, 1 (B p ), p 0. In particular, set E p = if p < 0. We have an exact
couple (see 3.2.9).
i
A `A
AA
AA
A
k AA
(3.3.2)
C,
/A
}
}
}}
}} j
}
}~
and let
i : Hn (E p ) Hn (E)
126
3 Homology of Fibrations
Hn (E p ) Hn (E p+1 ) Hn (E p+2 ) ,
so that we have to prove the following.
Each element x Hn (E) lies inside the image of some i, and if some
element in Hn (E p ) lies inside the kernel of i, then it also lies inside the kernel
of
ir : Hn (E p ) Hn (E p+r )
for r large enough.
Let x = [z], where z Zn (E) is a cycle. The support |z| of the cycle z,
1
with Hp,q (B p , B p1 ), where d correThus, we can identify the group Ep,q
sponds to d1 .
On the other hand, in this case one has (see [1, 7.3.1]) that
{
iJ p Z if q = 0,
Hp+q (B p , B p1 ) =
0
if q =
0,
127
Since d has bidegree (r, r 1), one has that d = 0 for r 2 and
r
r
2
for all r 2. One has also, from d and
hence we have that Ep,q
= Ep,q
3.2.18, that Def(dr+1 ) = Ker(dr ) = Def(dr ); hence Z 2 = Z , analogously,
from Ind(dr ) = Im(dr ) = Ind(dr+1 ), we obtain B 2 = B . It is now easy to
verify the formula (3.2.22) that ralates E with the ltration of H (B).
We want to compute now E 1 , E 2 , and d1 for the exact couple (4.2.2). For
that, we need the following result.
3.3.5 Lemma. Let A0 , A1 , B0 , and B1 be subspaces of B such that
A0 B 0
A1 B 1
and assume that A0 , resp. A1 , is a strong deformation retract of B0 , resp.
B1 . Then the inclusion induces isomorphisms
Hn ( 1 (A1 ), 1 (A0 ))
= Hn ( 1 (B1 ), 1 (B0 )) ;
besides,
Hn ( 1 (B ), 1 (A )) = 0,
= 0, 1 .
if b B,
(b, 1) A if b B,
(a, t) = a
if a A.
128
3 Homology of Fibrations
b : n I E ,
(b
(x, t) =
bt (x)) ,
/6 E
l l
l
bl
l l
l
l
/B
n I
n {0} i (n1 ) I
_
(3.3.7)
where
h(x, t) = ((x), t) (guaranteed by (iv)),
h0 (x, 0) = (x)
h0 (i (y), t) = (c
i )(y, t)
(guaranteed by (ii)),
(guaranteed by (i)).
(c
i ) has already been constructed, by the induction hypothesis. h0 is well
dened:
h0 (i (y), 0) = (c
i )(y, 0) ,
= i (y)
= (i (y))
= h0 (i (y), 0) .
129
h0 (x, 0) = (x)
= ((x), 0)
= (i (y), t)
= h(x, 0) ;
= ((i (y), t)
= h(i (y), t) .
Sq (n )
/ Sq+1 (n1 I)
(i id)#
/ Sq+1 (n I)
is commutative.
(d) We dene homomorphisms
r : Sn (B) Sn ( 1 (A)) ,
h : Sn (B) Sn+1 (B) ,
by
r() = b1 , h() =
b# F (n ) (n = idn ) ,
if is an n-simplex, and then by extending linearly.
Statement: r is a chain transformation, r() = if Sn ( 1 (A)), and
h is a chain homotopy such that
(h + h) = r ,
Sn (B) .
130
3 Homology of Fibrations
=
(1)i i (b
1 )
=
i=0
n
[
(1)i (
i )1
i=0
(1)i r(i )
i=0
= r() .
By (a) (iii), r() =
b1 = if Sn ( 1 (A)). Thus
h() = b
# F (n )
=
b# F (in ) ,
h() = h( (1)i i )
[
=
(1)i (
i )# F (n1 )
=
(1)i
b# (i id)# F (in1 ) (since by (a) (i) c
b(i id))
i =
=
(1)i c
(by (c))
# F i# (n1 )
= c
# F (n )
(since i (n ) = i# n1 ).
Summarizing:
h() + h() = c
# (F + F ) (n )
1
0
= c
# (j# j# )(n )
= b1 b0
= r() .
Hn ( 1 (A )) Hn ( 1 (B )) ,
= 0, 1 .
t together as follows.
/ Hn+1 ( 1 (A0 ))
/ Hn ( 1 (A1 ), 1 (A0 ))
/ Hn ( 1 (A1 ))
=
/ Hn+1 ( 1 (B0 ))
=
/ Hn ( 1 (B1 ), 1 (A0 ))
=
/ Hn ( 1 (B1 ))
/
/ .
131
Hn+1 ( 1 (A ))
Hn+1 ( 1 (B ))
=0 /
0
Hn ( 1 (B ), 1 (A )) / Hn ( 1 (A )) / Hn ( 1 (B ))
0,
3.3.1
(1)
(2)
Hn ( 1 (B p ), 1 (B p {mpj })) o
(4)
(5)
(3)
(6)
(7)
132
3 Homology of Fibrations
3
1 3 s4
2
3 s4
0 1 3 s4
Figure 3.1
We adopt the abbreviation
hn (A, A ) = Hn ( 1 (A), 1 (A ))
and consider the diagram
hn (sp , s p )
(3.3.9)
pi
/ hn1 (s p , i sp )
O
= (1)
hn1 (i sp , (i sp ) )
=
(2)
/ hn1 (s p i sp , i sp i sp )
=
ep = 01 0p1 0p : hn (sp , s p ) hnp (p sp ) .
3.3.2
133
sp
Figure 3.2
i sp
s p
1 sp
sp i sp
Figure 3.3
Let : I B be a path. Consider the bration
e induced by through
.
E
/E
/ B.
The map
e induces a homeomorphism
e1 (t) 1 ((t)) with whose help
we identify Hn ( 1 ((t))) and Hn (e
1 (t)).
3.3.11 Definition. Let : I B be a path. The translation of the
homology of the ber along is the homomorphism
(1)
: Hn ( 1 ((0)))
= Hn ( 1 (0)) Hn (E )
(2)
Hn (e
1 (1))
= Hn ( 1 ((1))) .
The homomorphisms (1) and (2) induced by the inclusion are isomorphisms by 3.3.10 and 3.3.5.
134
3 Homology of Fibrations
3.3.12 Theorem. The translation of the homology of the ber has the following properties:
(a) = id if is constant.
(b) (1 2 ) = 2 1 .
0 = 1 .
(c) 0 1 rel(I)
In other words, the translation of the homology of the ber is a functor from
the fundamental grupoid of B into the category of abelian groups (and isomorphisms).
Proof: (a) In this case, (1) = (2).
(b) Let 1 2 : E1 2 I be the bration induced through 1 2 and take
the paths
h1 : I I
( )
1
t 7
t,
2
h2 : I I
( )
1
1
t 7 +
t,
2
2
hi
hi
/ E1 2
/ I,
Hn ( 1 (1 2 (0)))
/ Hn (E1 )
OOO
O
OOO
OOO
1
OO'
Hn ( 1 (1 (0)))
/ Hn (E1 2 ) o
dII
u:
II h2
h1 uuu
II
u
II
u
u
I
u
u
Hn ( 1 (1 (1)))
Hn ( 1 (1 2 (1)))
Hn (E2 ) o
O
Hn ( 1 (2 (1)))
o7
ooo
o
o
oo
ooo 2
Hn ( 1 (2 (0)))
The homomorphisms h1 y h2 are bijective by 3.3.5, thus the diagram consists of nothing else but isomorphisms. Hence, we obtain the assertion.
(c) Let h : I I B be such that h(s, 0) = 0 (s), h(s, 1) = 1 (s),
h(0, t) = 0 (0) = 1 (0), h(1, t) = 0 (1) = 1 (1). Let
e : G I I
135
Hn (
/ Hn (E0 ) o
3.3.6
k0
Hn ( 1 (
(1)))
0
1
_
_
_
/
_
o
_
_
/
({(0, 0)}))
Hn (G)
Hn ( ({(1, 0)}))
that consists only of isomorphisms, one has that 0 can be dened through
the dotted arrows. Analogously we may conclude for 1 , and thus the homotopy between
1 (0 (0)) 1 (0, 0) G
and
1 (1 (0)) 1 (0, 1) G
p = 01 . . . 0p : Hn ( 1 (sp ), 1 (s p )) Hnp (F ) .
p
(
aj spj ) =
(fj )(p )1 (aj )
j
we obtain an isomorphism
: Cp (B; Hnp (F )) Hn ( 1 (B p ), 1 (B p1 )) .
(3.3.16)
p,q
136
3.3.3
3 Homology of Fibrations
hn (B p , B p1 )
p
gj
(1)
/ hn (sp , sp mp )
j
j
NNNj
II hp
II j
NNN(2)
II
NNN
II
NN&
I$
/ hn (B p , B p ep )
/ hn (B p , B p mp )
j
j
hn (spj , s pj )
(3)
all homomorphisms are induced by inclusions. (1) and (3) are isomorphisms
by 3.3.5, (2) is one by excision, and thus hpj is also one, and we obtain
{
0
if k = j,
p p
gk
fj =
p
hj if k = j,
p
(the second is an isomorphism). Thus the homomorphisms gk
determine
p
p1
hn (B , B ) as a direct sum; that is, we can give (uniquely) an element in
p
hn (B p , B p1 ) by its images under the homomorphisms gj
.
p1
p1
p1
hp1
(
)
(a
)
=
g
(
fk ( p1 )1 (ak ))
l
l
l
p1
= gl
(aspj )
(3.3.17)
p1 1 p
= gl
d fj ( p )1 (a).
hn (spj , s pj )
/ hn1 (sp1 , s p1 )
l
l
kk
k
k
k
kk
ukkk (2)
KK
KK
KK
%
/ hn1 (s p , i sp )
j
GG j
p
GG
fj
GG
GG(1)
p1
GG
hn1 (B )
GG
9
O
OOO
GG
rrr
GG
O
r
OO'
#
rr
/ hn1 (B p1 , B p1 ep1 ) ,
/ hn1 (B p1 , B p2 )
hn (B p , B p1 )
l
p1
1
hn1 (s pj )
gl
137
where ip is as in 3.3.9, and this, as well as (2), are considered only when
sp1
= i spj . All other homomorphisms are induced by inclusions or are
l
connecting homomorphisms of the corresponding homology sequence.
From this we have that, if sp1
is not a face of spj (see gure 3.4), then s pj
l
p1 1 p
lies in B p1 ep1
; hence (1) is the trivial homomorphism and so gl
d fj =
l
p1
as well as
0, and by the computation 3.3.17, we have al = 0, because hl
p1 are isomorphisms.
spj
slp1
Figure 3.4
Let sp1
= i spj , be the i-face of spj . Thus
l
p
p1 1 p
p 1
hp1
(a) = gl
d fj ( p )1 (a)
l i ( )
p1 1
= hp1
) (al );
l (
(as ) =
ip (a)(i sp ) ,
i=0
138
3 Homology of Fibrations
10
hq (s 1 , 0 s1 ) o
o7
= ooo
ooo
ooo
hq+1 (s1 , s 1 )
hq (0 s1 )
KKK
KKKf
KKK
K%
s1
OOO
OOO
OOO
OO'
hq (s 1 , 1 s1 ) o
#
/ hq (b0 )
h (s1 )
9 q
ss
s
s
sssg
sss
h ( s1 ) / hq (b0 ) .
4 q 1
;
11
139
0 = (as ) =
p p1
jp1 ip (a)(j i sp )
i=0 j=0
The fact that 0p = id has been already proved. For i 1 one has
0 1 sp = i1 0 sp = j k sp
if
and by the induction hypothesis, we obtain from this the assertion. Thus we
have proved that
p
(asp ) =
(1)i a(i sp ) ;
i=0
that is, that is the ordinary boundary homomorphism. In other words, this
states that (3.3.16) is a chain isomorphism. Hence we have the following
theorem, known as the LeraySerre theorem.
3.3.18 Theorem. Let : E B be a homologically simple (orientable)
1
1
, d ) is, through (3.3.16),
Serre bration over a CW-complex B. Then (E,q
isomorphic as a chain complex to (C (B; Hq (F )), 0). Therefore, induces
an isomorphism
2
Epq
= Hp (B; Hq (F )) .
140
3 Homology of Fibrations
3.3.4
1
By the last theorem, we have that Epq
= Cp (B; Hq (F )) = 0 for p < 0 or q < 0.
We say that the spectral sequence is concentrated in the rst quadrant, and
hence,
r
Ep,q
= 0 if p < 0 or q < 0 , r 1 .
r
r
d : Ep,q
Epr,q+r1
the group on the right-hand side is zero for r > p and thus the dierential
r
r
d is also zero. From the diagram that denes d in 3.2.17, we obtain that
r
d = 0 is equivalent to
Ker(dr )p,q = Def(dr )p,q .
Correspondingly, we have that for the dierential
r
r
r
d : Ep+r,qr+1
Ep,q
the group on the left-hand side is zero for r > q + 1 and thus
Ind(dr )p,q = Im(dr )p,q .
From these two equalities, together the chains of inclusions in 3.2.17 and
Denition 3.2.12, we have
1
2
p+1
p+2
Cp,q = Zp,q
Zp,q
Zp,q
= Zp,q
=
1
2
q+2
q+3
0 = Bp,q
Bp,q
Bp,q
= Bp,q
=
r
p+1
r
Zp,q
=
Zp,q
= Zp,q
= Zp,q
if r > p
r
Bp,q
r
q+2
r
Bp,q
= Bp,q
= Bp,q
if r > q + 1
r
Ep,q
= Zp,q
/Bp,q
= Ep,q
if r > max{p, q + 1} .
= Ep,np
Fp Hn (E)/Fp1 Hn (E) = Ep,np
r
= 0 for p > n, so that,
for r large enough. In particular, we have Ep,np
indeed, one has the mentioned ltration as indicated.
141
3.4 Applications
3.4
Applications
3.4.1
Spherical Fibrations
2
Ep,q
= Hp (B; Hq (F )) = 0
if
q = 0, m 1 .
r
Hence, for these values of q, Ep,q
= 0, r 2, and moreover,
2
2
Ep,0
= Ep,m1
= Hp (B; G)
if G(
= H0 (Sm1 ) = Hm1 (Sm1 )) is the coecient group of the homology.
We have that
r
r
r
d : Ep,q
Epr,q+r1
is nonzero, at most in case that
m
m
m
d : Ep,0
Epm,m1
;
/ Hp,0 (E m , dm )
/ Em
p,0
m+1
Ep,0
2
Ep,0
2
Epm,m1
m+1
Epm,m1
Ep,0
Hp (B; G)
Hpm (B; G)
.
Epm,m1
/ Em
pm,m1
/ Hpm,m1 (E m , dm )
/0
142
3 Homology of Fibrations
If we now consider the ltered homology of the total space E (see Subsection
3.3.4)
0 F0 Hn (E) F1 Hn (E) Fn Hn (E) = Hn (E) ,
and since
=0
Fp Hn (E)/Fp1 Hn (E) = Ep,np
if
n p = 0, m 1 ,
/ Hn (E)
/ Hn (E)/Fn1 Hn (E)
/0
En,0
Enm+1,m1
.
Epm+1,m1
Ep,0
@ @
Z
Z
Z
Z
Z
Epm,m1
=} =
}
=
}= }=
}= }=
0.
Hp (E)
Hp (B) Hpm (E) Hp1 (E)
,
Hp (E) Ep,0
Hp (B)
143
3.4 Applications
is indeed induced by .
Let us consider the commutative square
E
/B
id
/B
id
as a ber map from to idB . If we denote with a tilde the spectral sequence
associated to the bration idB (cf. 3.3.4), by the naturality of the spectral
sequence, we have the following commutative diagram:
Hp (E)
Hp (B)
/ E
p,0
/ E2
p,0
e
/E
e2
E
p,0
p,0
/ Hp (B; H0 (F ))
0
/ Hp (B; G) ,
144
3 Homology of Fibrations
{
G
Hp (CPn ) =
0
if p = 0, 2, 4, . . . , 2n,
otherwise.
{
G
Hp (HPn ) =
0
if p = 0, 4, 8, . . . , 4n,
otherwise.
3.4.5 Remark. For RPn the problem is that the corresponding bration
S0 , Sn RPn
has disconnected ber. See [1] for the corresponding computation.
3.4.2
{
Hq (F ) if p = 0, m,
= Hp (Sm ; Hq (F ))
=
0
otherwise.
r
= 0 for p = 0, m and r 2 and the
Therefore, again all terms Ep,q
dierentials can only be nonzero in the case
m
m
m
d : Em,q
E0,q+m1
,
145
3.4 Applications
Figure 3.5
as shown in Figure 3.5.
So again we have
E2 = E3 = = Em
E m+1 = E m+2 = = E .
Moreover, the following sequence is exact:
0
/ Hp,q (E m , dm )
/ Em
m,q
m+1
Em,q
Em,q
/ Em
0,q+m1
/ H0,q+m1 (E m , dm )
2
Em,q
2
E0,q+m1
m+1
E0,q+m1
Hq (F )
Hq+m1 (F )
E0,q+m1
.
/0
/ Hn (E)/Fm1 Hn (E)
/ Hn (E)
/0
Em,nm
E0,n
]
]
]
]
Em,q
[
=} =
}
=
}= }=
}= }=
[
E0,q+m1
[
[
[
<|
<| <|
<
|
|< |<
<
"
0.
146
3 Homology of Fibrations
Hq+m1 (F ) E0,q+m1
Hq+m1 (E)
{b0 }
/E
/B
Hr (F )
Hr (F )
/ H0 ({b0 }; Hr (F ))
/ H0 (Y ; Hr (F ))
e2
/E
0,r
e0,r
E
/ Hr (F )
/ E2
0,r
/ E
0,r
/ Hr (E)
/ Em
0,r
/ E m+1
0,r
/ Hr (E) .
From it, it is easy to convince oneself that the top row yields the identity.
147
3.4 Applications
Figure 3.6
3.4.8 Example. Let : Sm Sk be a homologically trivial Serre bration
with ber Sn , k 1, n 1. The Gysin sequence with coecients in Z and
p = k + n + 1 looks as follows
/ Hk+n+1 (Sk )
/ Hk (Sk )
/ Hk+n (Sm )
/ ,
0
thus Hk+n (Sm ) = 0; therefore,
(3.4.9)
k + n = m.
/ Hnk+1 (Sn )
/ Hn (Sn )
/ Hn (Sm )
/ .
0
Thus Hnk+1 (Sn ) = 0; therefore,
(3.4.10)
n k + 1 = 0;
148
3.4.3
3 Homology of Fibrations
if 0 < q < s.
=0
if
0<p<r
or
0 < q < s.
= 0
0
s
n1
d
Figure 3.7
Thus the nonzero terms of the spectral sequence are distributed according
to Figure 3.7. Again, we omit writing the coecients. For the elements of the
r
, we call p + q their total degree. In what follows, we shall consider
term Ep,q
k
n
n
d : En,0
E0,n1
.
149
3.4 Applications
We have
2
3
n
En,0
= En,0
= = En,0
n+1
n+2
En,0
= En,0
= = En,0
2
n
E0,n1
= = E0,n1
n+1
E0,n1
= = E0,n1
,
/ E n+1
n,0
En,0
n,0
/ En
0,n1
/ E n+1
0,n1
2
En,0
2
E0,n1
E0,n1
/ En
/ 0.
Hn (B)
Hn1 (F )
0 E0,n
Hn (E) En,0
0 .
Hn (F )
Hn (E)
Hn (B) Hn1 (F ) .
i
Hopf bration S1 S3 S2 , the sequence
H3 (S1 ) H3 (S3 ) H3 (S2 )
is not exact. This shows that the inequality n < r + s cannot be improved
in general.
150
3 Homology of Fibrations
Chapter 4
Generalized Cohomology of
Fibrations
In this chapter, we present dierent spectral sequences, according to the
type of bration we are dealing with (Leray-Serre-Whitehead; RothenbergSteenrod).
4.1
Introduction
4.1.1
and
q : hq R hq+1 ,
152
for all q Z.
Excision. For every pair of spaces (X, Y ) and a subset U Y satisfying
q1
and
q : hq hq1 R ,
153
4.1 Introduction
(X , Y )
hq (X , Y ) .
AA
AA
AA
A
/X
~
~
~~
~~
~
~~
fb
(X , Y )
/ (EX , EY )
/ (X, Y )
154
fb : h (EX , EY ) h (EX , EY ) .
EX I
EX I
n6 EX NN
H
NNN
NN&
n n
/E
X
6
NNN
Hmmm
NNN
mmm
'
/ B,
X I
155
4.1 Introduction
for all q.
Namely, EU EY and clearly
EU EU E EY .
Y
.
4.1.6 Remark. If h satises the additivity axiom, then also h
(EX , EY ) =
(X , Y ) ,
and
Then (x,
e e) = ((x), e). If : Y Y X X is the inverse homeomorphism of |XX , then the map
e : EY EY EX EX
e e) = ((y), e) is well dened, since y Y Y , and is the
given by (y,
inverse of |
e EX EX .
Using 1.4.20 and 4.1.7 we have the following.
4.1.8 Theorem. Let : E B be a Hurewicz bration and f : X
B, g : Y B be spaces over B. Let moreover X X, Y Y be
cobrations, and : (X, X ) (Y, Y ) a map over B that is also a relative
homeomorphism. Then
(Y, Y ) h
(X, X )
: h
is an isomorphism.
156
(Y, Y ) = h (EY , EY )
h
h (EY /EY ) ,
=e
h (EY , EY )
/ h (EX /EX )
/ h (EX , EX ) ,
4.1.2
4.1.9 Definition. Let B be a topological space. A system of local coecients on B is a contravariant functor
G : 1 (B) Ab ,
where 1 (B) denotes the fundamental groupoid of B (1.5.5) and Ab is the
category of abelian groups (and isomorphisms). In other words, a system of
local coecients maps every point b B to an abelian group G(b), and every
path : b b to a group isomorphism G() : G(b ) G(b), in such a way
that if 0 1 , then G(0 ) = G(1 ).
4.1.10 Example. Let : E B be a Serre bration. Dene
G : 1 (B) Ab
by
G(b) = H n ( 1 (b)) ,
where
e : E I is the bration induced by over : I B. As in
Denition 3.3.11, the homomorphisms (1) and (2) induced by the inclusions
are isomorphisms.
157
4.1 Introduction
We call this the ordinary system of local coecients induced by the bration : E B on B.
4.1.11 Definition. Let h be a generalized cohomology theory and :
E B a Hurewicz bration, with a path lifting map
: E B B I = {(e, ) E B I | (e) = (0)} E I .
Given any path : b b , dene a map
() : 1 (b) 1 (b )
by ()(e) = (e, )(1).
4.1.12 Exercise. Prove the following facts:
(i) If 0 1 : b b , then (0 ) (1 ) : 1 (b) 1 (b ).
(ii) If : b b and : b b , then ( ) ( ) () : 1 (b)
1 (b ).
4.1.13 Exercise. Prove that there is a category of systems of local coecients on a space B.
4.1.14 Exercise. Prove that a map f : B B induces a covariant functor from the category of systems of local coecients on B to the category
of systems of local coecients on B. Prove that this correspondence is (contravariantly) functorial.
From Exercise 4.1.12 we conclude that there is a system of local coecients as follows.
4.1.15 Theorem. Let h be a generalized cohomology theory and : E
B be a Hurewicz bration. Then the mapping
[] 7 () : hp ( 1 (b )) hp ( 1 (b))
determines a system of local coecients. We call this the hp -system of local
coecients induced by the bration : E B on B, and denote it by
hp (F).
158
4.1.3
s=
gi i , where gi G(i (e0 )) .
This explains the second equality.
Dually we dene
S p (X; G)
(4.1.17) s =
gi i =
G(i )(gi )0 i +
(1)j gi j i .
i
j=1
159
4.1 Introduction
G()
GG
GG
GG
G
(1) GG
#
/ G((0))
w
ww
ww
w
w
{ww (1)
p+1
i=1
160
= H (X; G).
If A X and G is a system of local coecients on X, then we may
consider the restriction G|A of G to A by taking the composition of the
functor G with the morphism of fundamental groupoids 1 (A) 1 (X).
The inclusion S (A; G|A ) S (X; G) has a cokernel that we denote by
S (X, A; G).
4.1.25 Definition. The (singular) homology of the pair (X, A) with local
coecients in G is given by
H (X, A; G) = H (S (X, A; G)) ,
and the short exact sequence of chain complexes
0 S (A; G|A ) S (X; G) S (X, A; G) 0
provides the long exact sequence in homology of a pair
161
4.1 Introduction
G(x0 )
x0
/ G(x1 )
H(x0 )
H()
x1
/ H(x1 )
H (Z; g f G)
QQQ
QQQ
QQ
g QQQQ
(
/ H (X; G)
oo7
ooo
o
o
o
ooo f
H (Y ; f G)
commutes. This is the Functoriality axiom for homology with local coecients.
4.1.27 Exercise. Let G be a system of local coecients on X, and let
f0 , f1 : Y X be homotopic maps. Prove that the induced systems of
local coecients f0 G and f1 G are isomorphic, say by an isomorphism of
systems of local coecients : f0 G f1 G. Prove, moreover, that the
homomorphisms induced by f0 and f1 in homology with local coecients in
G coincide up to the isomorphism, namely, that the diagram
H (Y ; f0 G)
b
TTTfT0
TTT*
jjj4
jjjfj1
H (X; G)
H (Y ; f1 G)
commutes. This is the Homotopy axiom for homology with local coecients.
4.1.28 Exercise. Let G be a system of local coecients on X and let A
162
(Hint: Compare with the proof of [14, 4.6.5].) This is the Excision axiom
for homology with local coecients.
4.1.29 Exercise. Observe that a system of local coecients G on a singular
space is nothing but an abelian group G = G(). Prove that
{
G if p = 0,
Hp (; G) =
0 if p = 0.
This is the Dimension axiom for homology with local coecients.
The previous exercises show that homology with local coecients satises
axioms similar to the EilenbergSteenrod axioms (cf. Subsection 4.1.1 or see
[1]). There is one more axiom that also plays an important role; namely, we
have the following.
4.1.30 Exercise. Take pairs of spaces (X , A ), with the indexes varying
in any set . Let i : (X , A ) (X , A ), , be the canonical inclusion of each of the pairs into their topological sum. If G is a system of local
coecients on the the topological sum and G = i G is the induced system
on each summand, then prove that the inclusions provide an isomorphism
(
)
=
(X , A ) ; G .
Hp (X , A ; G ) Hp
163
4.1 Introduction
=
Hp
(X , A ) ; G
H p (X , A ; G ) .
4.1.4
X p1
p X p
p
is an identication.
4.1.34 Definition. Suppose that (X, A) is a relative CW-complex and G
a system of local coecients on X. If we denote by X p the p-skeleton of
(X, A), p 0, and X 1 = A, we dene the cellular complex of (X, A) with
local coecients in G by
Cp (X, A; G) = Hp (X p , X p1 ; G) ,
and
: Cp (X, A; G) Cp1 (X, A; G)
is the boundary operator of the triple (X p , X p1 , X p2 ).
164
p ) (X p , X p1 )} is a
4.1.35 Proposition. Assume that { : (p ,
p ; G ) Hp (X p , X p1 ; G) = Cp (X, A; G)
( ) :
Hp (p ,
p (X,A)
p (X,A)
)
(
)
(
p ; G
p ; G .
Hp p ,
p ,
(4.1.37)
= Hp
p (X,A)
p (X,A)
i
i
i
i
induce isomorphisms
)
(
(
(
))
(
)
j
p1
p1
p1 p1
p
p p
Hp , Hp1 , i i
Hp1 i , i
,
where the connecting homomorphism on the left-hand side is an isomor p , (p1
p1 )), since both
phism by the exact sequence of the triple (p ,
i
the rst and the third spaces of it are contractible, while j on the right-hand
side is an isomorphism by excision. We dene
[p , p1
] = j1 .
i
165
4.1 Introduction
p1 ) G
p ) G
:
Hp (p ,
Hp (p1 ,
by
(u g) =
p1
p
[p , p1
, (g) .
](u) , i
p1
<p
p ; G )
H p (p ,
( ) : C p (X, A; G) = H p (X p , X p1 ; G)
p (X,A)
166
p1 p1
p1
p1
p1
p1
p p
p
p p
H
i , i
H
, i i
H ,
and we dene
[p , p1
] = j 1 .
i
If we take p , p1
, : G((e0 )) G((ei0 )) as before, we have
i
4.1.42 Theorem. The coboundary homomorphism of the cellular cocomplex
of a pair of spaces (X, A), : C p1 (X, A; G) C p (X, A; G), can be expressed in terms of the direct product decompositions given in Theorem 4.1.41
p1 ) G
p ) G
H p1 (p1
,
H p (p ,
:
by
(1)p1 ((u g ) ) =
[p , p1 ] (u ) p , p1 , 1 (g ) .
p1
<p
i
i
i
i
We proceed inductively on p. Consider p = 1 and the diagram
1) F )
h1+q ((1 ,
g3
ggggg
g
g
g
g
ggg
hq (F ) WWWWW
WWWWW
WW
(j 1 )1 +
1
jF
1 ) hq (F ) ,
H 1 (1 ,
167
p1 ) F )
hp1+q ((p1
,
i
i
p1
p) F )
/ hp+q ((p ,
p
p1 ) hq (F )
H p1 (p1
,
i
i
j 1 1
p ) hq (F ) ,
/ H p (p ,
p , (p1
p1 )) F ,
where on the top corresponds to the triple (p ,
i
p , (p1
p1 )). Since the horizontal
while the one on the bottom to (p ,
i
arrows are isomorphisms, so as also is the left arrow, we may dene p to be
an isomorphism such that the diagram commutes.
4.2
GG
GG
G
j GG
(4.2.2)
/ A
w
w
w
ww
w
w
w{ w k
C ,
168
Zrp,q = Def(dp,q
r ),
Brp,q = Ind(dp+r,qr+1
),
r
p,q
Z
=
Zrp,q ,
p,q
B
=
Erp,q
r=1
Brp,q ,
r=1
Zrp,q /Brp,q
1 r .
p,q
Similarly to Section 3.2, we have that dp,q
C p+r,qr+1 induces
r : C
a homomorphism
p,q
dr : Erp,q Erp+r,qr+1 ,
and one has the following result dual to 3.2.19.
4.2.3 Theorem. The pair (Er , dr ) is a cochain complex and its cohomology
satises
p,q
H p,q (Er , dr )
= Er+1 ;
4.2.1
169
/ Ep
/ Bp .
e
p ) F(e ) ))
: hp+q (E p , E p1 ) hp+q (T , T ) hp+q ((p ,
0
p
p ) hq (F(e ) )) ,
H p (p ,
0
p of p , and p is the
where T is the restriction of T to the boundary
isomorphism given in Theorem 4.1.43. Since by 1.4.26 the map of pairs
p ) F(e ) ) (T , T ) is a (ber) homotopy equivalence, the
: (p ,
0
homomorphism is also an isomorphism.
In what follows, we prove that the homomorphism
= (
e ) : hp+q (E p , E p1 )
hp+q (T , T ) .
p
is an isomorphism.
hp+q (T , T )
170
p , B p1 )
= ( ) : h(B
p,
p)
h(
p ) (B p , B p1 )
(p ,
p
p
p1
.
B B
p
p ,
Given that both, the inclusion p
p , and the inclusion
B p1 , B p are cobrations, by 4.1.8 we have an isomorphism
p (B p , B p1 ) h
p
()
:h
p ) .
(p ,
(see
But since the cohomology theory h is additive, then so is also h
p p
4.1.6); hence the homomorphisms induced by the inclusions i : ( , ) ,
p ) yield an isomorphism
(p ,
p
p
h
p )
(p ,
=
p,
p) .
h(
171
4.2.2
/ C p (B, A; hq (F))
hp+q (E p , E p1 )
hp+q+1 (E p+1 , E p )
(1)p
where on the left-hand side is the connecting homomorphism for the triple
(E p+1 , E p , E p1 ) and on the right-hand side represents the coboundary operator of the cellular cochain complex of the pair (B, A) with local coecients
in hq (F) (see 4.1.40).
Proof: We have to prove the commutativity of the diagram
/
hp+q (E p , E p1 )
(4.2.7)
p ) hq (F )
H p (p ,
(e )
hp+q+1 (E p+1 , E p )
(1)p
p+1
q
p+1
(p+1
),
, ) h (F (e
H
0)
hp+q (E p , E p1 )
/ hp+q (
p p
( , )
hp+q+1 (E p+1 , E p )
/ hp+q+1 (
F (e ) )
0
p+1 p+1
( , )
F (e0 ) ) .
172
Since both vertical arrows are given by connecting homomorphisms, that are
natural, and inclusion maps, the commutativity of this diagram is quite clear.
p) F )
hp+q ( (p ,
(e )
p+1
hp+q+1 ( (p+1
)
, ) F (e
0)
p) F )
hp+q ((p ,
(e )
p+1
hp+q+1 ((p+1
).
, ) F (e
0)
p) F )
hp+q ((p ,
(e )
0
p+q+1
p+1
((p+1
, )
F (e0 ) )
p+q
p+q+1
p ) hq (F )
H p (p ,
(e )
0
p+1
(1)p
p+1
(p+1
, )
hq (F (e0 ) ) ,
where the isomorphisms are given in 4.1.43. This last diagram commutes
because the deniton of the coboundary homomorphism on the right-hand
side is given using the coincidence isomorphisms dened in page 166 previous
to Theorem 4.1.42, that correspond precisely to the way that the coboundary
homomorphism on the left-hand side is dened, and the sign comes from
Theorem 4.1.42.
Putting these three diagrams together, we obtain the commutativity of
Diagram (4.2.7), as desired.
From Lemma 4.2.6, we obtain immediately the main result of this paragraph.
4.2.8 Theorem. Let : E B be a Hurewicz bration. If (B, A) is
a relative CW-complex and Ep,q is the spectral sequence associated to the
ltration of E induced by the skeletal ltration of B, then one has for the
E2 -term an isomorphism
: E2p,q H p (B, A; hq (F)) ,
where H p (B; hq (F)) is the cellular cohomology of (B, A) with local coecients
determined by hq ( 1 (b)), b B.
References
174
References
Index
action
, diagonal, 91
action of a group on a space, 56
, eective, 57
, free, 57
, transitive, 57
additive relation, 117
, denition domain, 118
, image, 118
, indeterminacy, 118
, kernel, 118
additivity axiom
for cohomology with local coefcients, 163
for homology with local coecients, 162
associated
ber bundles, 77
bration, 92
open cover for a partition of unity,
18
atlas, 61
, maximal, 64
, trivial, 61
atlases
, equivalent, 63
axiom
, additivity, for cohomology with
local coecients, 163
, additivity, for homology with
local coecients, 162
, dimension, for homology with
local coecients, 162
, exactness, for cohomology with
176
of ber bundles, 65
of brations, 6
of set bundles, 62
cell decomposition
, Klein bottle, 2
cellular cocomplex with local coefcients, 165
cellular complex with local coecients, 163
chain rule, 70
chart
, local, 60
classifying space, 105
cocycle, 73
cocycles
, cohomologous in a cover, 74
, cohomologous in a space, 75
cohomologous cocycles, 74, 75
cohomology class of a cocycle, 75
cohomology theory, 151
, exactness axiom, 152
, excision axiom, 152
, homotopy axiom, 151
, long exact sequence, 152
cohomology with local coecients,
160
, additivity axiom, 163
, exactness axiom, 162
cohomology with local coecients
of a pair, 162
compatible set bundle map, 62
complex
projective space, 9
composition of relations, 117
cone
, reduced, 33
construction
, Borel, 91
, Milnor, 111
coordinate transformation, 72
covering
Alphabetical Index
map, 45
denition domain
of a one-one relation, 69
of an additive relation, 118
derivation, 70
derivative, 70
diagonal action, 91
dieomorphism
, local, 70
dierentiable relation, 69
dimension axiom
for homology with local coecients, 162
dual vector bundle, 79
eective action of a group on a space,
57
equivalence
, bundle, 65
, ber, 6
over B, 65
equivalent
, ber homotopy, 23
atlases, 63
brations, 6
equivariant map, 58, 82, 91
exact
couple, 120
sequence of bigraded groups, 119
exactness axiom
for cohomology with local coefcients, 162
for homology with local coecients, 160
excision axiom
for homology with local coecients, 162
excision for a cohomology theory,
152
exponential
bration, 4, 8
Alphabetical Index
ber
, translation, 26
bundle, 64
, trivial, 64
map, 65
equivalence, 6
map, 5
determined by a ber bundle,
68
over B, 6
of a bration, 5
ber bundle
, induced, 94
, restricted, 94
, restriction, 94
ber bundles
, associated, 77
, category, 65
ber homotopy, 23
ber homotopy equivalent, 23
ber-preserving map, 23
bration, 5
, n-universal, 104
, Hopf, 10, 147
, Hurewicz, 16
, Klein, 2, 8
, Moebius, 2, 8
, Serre, 16
, associated, 92
, base space, 5
, exponential, 4, 8
, ber, 5
, induced, 22
, locally trivial, 7
, orientable, 135
, path, 20, 146
, principal, 81
, product, 2
, restriction, 7
, spherical, 48
, total space, 5
177
, trivial, 2, 6
determined by a ber bundle,
68
brations
, equivalent, 6
G-brations
, category, 82
brations, category, 6
ltration of a group, 121
k-frame in Rn , 88
free action of a group on a space,
57
Freudenthal suspension theorem,
38
functional bundle, 96
, partial, 103
functoriality axiom
for homology with local coecients, 161
fundamental
group, 35
groupoid, 28
fundamental group
of RPn , 48
general linear group of
complex n n matrices, 52
real n n matrices, 52
germ of a function, 70
Grassmann manifold, 107
group
, bigraded, 119
, fundamental, 35
, general linear, 52
, isotropy, 58
, orthogonal, 52
, special linear, 52, 54
, special orthogonal, 53
, special unitary, 54
, structure, 64
, topological, 51
178
, unitary, 54
action, 56
, orbit, 59
eective action, 57
free action, 57
transitive action, 57
groupoid, fundamental, 28
Gysin sequence, 142
HLP, 12
, relative, 14
homogeneous space, 55
homology
, Klein bottle, 3
, torus, 3
homology theory, 152
homology with local coecients, 159
, additivity axiom, 162
, dimension axiom, 162
, exactness axiom, 160
, excision axiom, 162
, functoriality axiom, 161
, homotopy axiom, 161
homology with local coecients of
a pair, 160
homomorphism
, suspension, 38
homomorphism of bigraded groups,
119
homotopy
, pointed, 31
ber, 20
groups
of the Stiefel manifolds, 90
lifting property, 12
, relative, 14
nth homotopy group, 35
homotopy axiom
for homology with local coecients, 161
homotopyaxiom for a cohomology
Alphabetical Index
theory, 151
Hopf bration, 10, 147
Hurewicz bration, 16
image of an additive relation, 118
incidence isomorphism, 164
indeterminacy of an additive relation, 118
induced
bundle, 94
cocycle, 75
ber bundle, 94
bration, 22
inverse relation, 117
isomorphism
, incidence, 164
isotropy subgroup, 58
join of two spaces, 10
kernel of an additive relation, 118
Klein
bottle, 2, 8
, cell decomposition, 2
, homology, 3
bration, 2, 8
LeraySerre
spectral sequence, 139
theorem, 139
lifting, 12
local chart, 60
of a manifold, 71
local coecients
, singular cohomology, 160
, singular homology, 159
local coecients, system, 156
local dieomorphism, 70
locally
nite partition of unity, 18
trivial
bration, 7
principal bration, 82
179
Alphabetical Index
paracompact space, 18
partial functional bundle, 103
partition of unity, 17
, associated open cover, 18
, locally nite, 18
, subordinate to a cover, 18
path bration, 20, 146
path space, 18
path-lifting map, 19
PLM, 19
pointed
homotopy, 31
map, 31
space, 31
pointed homotopy classes, set, 31
principal
bundle, 81
bration, 81
, locally trivial, 82
G-bration
, trivial, 82
map, 82
product
, smash, 32
, twisted, 91
bration, 2
projective space
, complex, 9
, quaternionic, 9
, real, 9
property
, homotopy lifting, 12
, relative homotopy lifting, 14
quaternionic
projective space, 9
real
projective space, 9
, fundamental group, 48
reduced
cone, 33
180
suspension, 33
renement of an open cover, 75
relation, 117
, additive, 117
, inverse, 117
relations
, composition, 117
relative
HLP, 14
homotopy lifting property, 14
restricted
bundle, 94
ber bundle, 94
restriction
of a ber bundle, 94
of a bration, 7
of a system of local coecients,
160
section, 4, 50
section over a subspace, 99
sequence
, Gysin, 142
, Wang, 146
Serre bration, 16
, spectral sequence, 139
set bundle, 60
map, 62
compatible, 62
set bundles
, category, 62
set of pointed homotopy classes, 31
singular
cohomology groups with local
coecients, 160
homology groups with local coecients, 159
small category, 29
smash product, 32
space
, classifying, 105
Alphabetical Index
, homogeneous, 55
, loop, 146
, paracompact, 18
, path, 18
, pointed, 31
, tangent, 70
of orbits, 59
G-space, 56
special
linear group, 52, 54
orthogonal group, 53
unitary group, 54
spectral sequence, 124
, LeraySerre, 139
sphere
, tangent bundle, 4
spherical bration, 48
Stiefel manifold, 88
, homotopy groups, 90
structure group, 64
suspension
, reduced, 33
homomorphism, 38
system of local coecients, 156
, constant, 159
, restriction, 160
, trivial, 159
induced by hp , 157
induced by ordinary cohomology, 157
tangent
bundle, 4
, unitary, 50
of the sphere, 4
space, 70
vector, 70
tensor product of vector bundles,
80
theorem
, Freudenthal, 38
Alphabetical Index
, LeraySerre, 139
topological
group, 51
topological product, 1
torus, 2
, homology, 3
total space of a bration, 5
transformation, coordinate, 72
transgression, 149
transitive action of a group on a
space, 57
translation
along a path, 26
of the ber, 26
translation of the homology of the
ber, 133
trivial
atlas, 61
ber bundle, 64
bration, 2, 6
principal G-bration, 82
twisted product, 91
unitary
group, 54
tangent bundle, 50
n-universal
bundle, 104
bration, 104
universal bundle, 105
vector
, tangent, 70
vector bundle, 79
, dual, 79
vector bundles
, Whitney sum, 80
, tensor product, 80
Wang sequence, 146
wedge sum, 32
Whitney sum, 80
181