Top Rigidity of Torus
Top Rigidity of Torus
Top Rigidity of Torus
Alexandre Martin
Preface
One of the main goals in topology is the classication of manifolds up to some
equivalence relation (homotopy equivalence, homeomorphism, PL-homeomorphism,
dieomorphism. . .). A very natural question arising from this is to decide whether
two manifolds identical up to some equivalence relation are still the same under
a stronger equivalence relation. In this thesis, we will be interested in topogical
rigidity. A manifold M is said to be topologically rigid if any topological manifold homotopy equivalent to M is actually homeomorphic to M . Maybe the most
famous example of a topological rigidity phenomenon is the Poincar conjecture,
now proven in all dimensions, which asserts that Sn is topologically rigid. An example more related to the subject of this thesis is the following result of Mostow.
If f : M N is a homotopy equivalence between two complete hyperbolic n-manifolds of nite volume (n 3), then f is
homotopic to an isometry.
Motivated by this result, Borel formulated the following conjecture, which can
be thought as a topological analogue of the Mostow Rigidity Theorem.
Borel Conjecture.
The rst example was developed during the sixties. In 1964, Bass, Heller and
Swan [. H. 64] proved the vanishing of W h(Zn ). At the same period, Farrell and
Hsiang developped the theory of codimension one splitting obstructions [FH73],
using the ideas introduced in the doctoral dissertation of Farrell (1967). This in
turn was used by Hsiang and Shaneson [HS70] to classify the PL-structures on a
PL-manifold homotopy equivalent to a high-dimensional torus. This result is a
cornerstone in high-dimensional topology. It was of crucial importance to Kirby
and Siebenmann [KS77] who used it to develop the theory of topological manifolds.
1
This in turn was used by Hsiang and Wall to prove the topogical rigidity of the
torus in 1969 in [HW69].
Since then, this circle of idea has been extensively studied, especially by Farrell and Jones, who were able to prove the Borel Conjecture for a large class of
manifolds.
The aim of this thesis is to present the surgical proof of the original result, namely
the topologicl rigidity of the torus. It is organised as follows. The rst chapter
presents a proof of the Bass-Heller-Swan theorem for Zn , which will be used in
Chapter 2 to split homotopy equivalences along codimension one submanifolds.
This splitting theorem will be a key ingredient for the computation of the PLstructure set SP L (Tk Dn ) presented in Chapter 4. Chapter 5 deals with the
various P L-structure a topological manifold might carry. Finally, we prove the
topological rigidity of the torus in Chapter 6.
While trying to be as self-contained as possible, I was forced to outsource some
results and some of the most technical lemmas to keep this thesis reasonably long.
I shall give references anytime I do that.
I would like to thank here my advisor, Andrew Ranicki. During these few
months I spent in Edinburgh, he guided me through this beautiful world of surgery
theory, showing me its beautiful landscapes, while encouraging me to venture on
some (algebraic) roads I would not have taken alone.
I would like also to thank my advisor at the Ecole Normale Suprieure, Frdric
Paulin. Since my very rst year at the ENS he helped me and encouraged me to
work on a eld that is not so fashionable in France. I could not thank him enough
for that!
Contents
1 The Bass-Heller-Swan Theorem.
1.1
1.2
1.3
1.4
K0 , K1 . . . . . . . . . . . . . . .
The Resolution theorem for K1 .
The Grothendieck's Theorem .
The vanishing of W h(Zn ). . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.2
3.3
. . . . . . .
. . . . . . .
. . . . . . .
Theorem...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 5
. 9
. 14
. 17
.
.
.
.
Surgery obstructions. . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Degree one normal maps, surgery obstructions, and L-groups.
3.1.2 Simply-connected obstructions. . . . . . . . . . . . . . . . .
3.1.3 Codimension-one splitting. . . . . . . . . . . . . . . . . . . .
3.1.4 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The surgery exact sequence. . . . . . . . . . . . . . . . . . . . . . .
3.2.1 The sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2 The maps . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.3 An exact sequence of what ? . . . . . . . . . . . . . . . . . .
Classifying spaces and their homotopy groups. . . . . . . . . . . . .
3.3.1 i (P L/O) and smoothing theory. . . . . . . . . . . . . . . .
3.3.2 BO, BG and the J -homomorphism. . . . . . . . . . . . . . .
3.3.3 i (G/P L) and the surgery obstruction map of spheres. . . .
21
22
29
32
34
36
36
36
37
38
39
40
40
42
42
43
43
43
44
46
59
71
4.1
4.2
5.1
5.2
5.3
6.1
6.2
Chapter 1
The Bass-Heller-Swan Theorem.
The Whitehead torsion (f ) W h(1 (N )) of a homotopy equivalence f : M N
vanishes if f is homotopic to a homeomorphism. Thus the vanishing of W h(1 (M ))
is generally a rst step in proving the topological rigidity of a high-dimensional
manifold M . This chapter is devoted to prove the vanishing of W h(Zn ).
In Section 1.1, we dene lower algebraic K -groups for suitable categories. The
interest of such a general approach will become apparent in Section 1.2 where
we prove the Resolution Theorem for K1 , which will allow us to work with more
tractable categories. We prove in Section 1.3 a theorem of Grothendieck which
will be needed to prove the vanishing of W h(Zn ) in Section 1.4. This chapter is
greatly inspired by [Ros94] and [. H. 64].
Throughout this chapter R will be a
1.1
K0, K1.
Denition 1.1.1. (i) Let Proj R be the category of netaly generated projective
R-modules.
(ii) Let R-modfg be the category of nitely generated R-modules. It is an abelian
category if R is noetherian.
Denition 1.1.2.
0 P 1 P P2 0
is a short exact sequence in A and P1 , P2 Obj P , then P Obj P .
- P has a small skeleton, i.e, P has a full subcategory P0 which is small and
for which the inclusion P0 , P is an equivalence.
Proposition 1.1.3. Let R be a Noetherian ring. Then R-modfg and Proj R are
categories with exact sequences.
Proof. These categories are clearly closed under extensions. Proj R has for (small)
skeleton the set of direct summands in Rn , n 0. R-modfg has for (small) skeleton
the set of quotient modules of Rn , n 0.
fg
Denition 1.1.4.
Denition 1.1.5.
P1
P1
P
/
0
/
P2
P2
Denition 1.1.6.
M 7
M 0
0 1
,
En (R) embeds in En+1 (R). We denote GL(R) (resp. E(R)) the direct limit of the
GLn (R) (resp. En (R)).
(ii) If is a group, let W h() be the quotient of K1 (Z) by the image of GL1 (Z)
in GL(Z), called the Whitehead group of .
Proposition 1.1.9. K1 (R) is an abelian group and K1 (R) ' K1 (Proj R).
For a proof of these results, we refer to [Ros94].
Denition 1.1.10.
Denition 1.1.12.
0 Pn . . . P 1 M
with each Pi a nitely generated projective R-module.
We recall a famous theorem of Hilbert.
Theorem 1.1.13
R[t].
1.2
K1 .
In this section we will prove that for a regular ring R, K1 (R), or equivalently
K1 (Proj R), is naturally isomorphic to K1 (R-modfg ). The main advantage of
this theorem is that it allows us some useful constructions as quotients, since we
now work in an abelian category. This considerations will be of crucial importance
in the proof of the vanishing of W h(Zn ).
The notation [ ]f g (resp. [ ]proj ) will denote an element of K (R-modfg ) (resp.
K (Proj R).
Our starting point, while trying to build a map K1 (R-modfg ) K1 (Proj R),
is the fact that every nitely generated R-module (with R-regular) admits a nite
resolution. We need the following result, which allows to lift an automorphism of
a nitely-generated R-module to an automorphism of some nite resolution.
...
/
...
P1
/ Pr
Pr
0
/
P1
=
0
0 1
=
1
0 1
1
0
1
1
0 1
0 1
1 0
.
1
0 1
1 0
0 1
1
0 1
0 1
1 0
P P
1
P P
M
0
0.
...
dr+1
...
dr+1
dr
Pr
...
dr
Pr
...
P1
0
/
P1
...
dr+1
Pr
dr
...
/M
P1
0,
this implies that ker dr is projective. Since r+1 induces an isomorphism on ker dr ,
we have the following nite resolution:
/ ker d dr+1 /
r+1
Pr
/ ker d
dr+1
r
dr
...
Pr
dr
...
P1
0
/
P1
Pr
Pr
/ ...
/P
1
/ ...
/ P1
10
/ 0.
Set
([M, ]f g ) =
X
(1)i [Pi , i ]proj K1 (Proj R).
i1
...
0
Pr+1
r+1
...
/0
...
Pr0
P10
Pr
...
P1
M0
0.
P0
d0
to a diagram
M
/
P00
0
/0
M0
P0
/M
d0
0.
0
Let B = ker P0 M 0
. Since d0 is surjective, the projection B M 0 is
surjective. Now consider an epimorphism P00 B , with P00 projective. Composing
with the projections B P0 and B M 0 yields a diagram
P00
0
P0
d00
/0
M0
d0
/M
11
0.
Pi0
i
Pi+1
/ ...
/ P0
M0
di
Pi
/M
/ ...
0,
ker di
Pi+1
ker di
0r
/
/
/ P0
/ ...
Pr0
P1
01
P1
/
...
/ ...
Pr0
...
/P
/ Pr
/ P0
/ 0.
Pr
dr d0
Pr0 r
...
P1 P10
d1 d01
M M
. . . P100 M
12
0,
Id
P1
f10
Id
P10
0,
f1 and f10 induce isomorphism on the rst homology groups. It follows that f and
f 0 are homology equivalences, hence their mapping cones are acyclic. But an easy
induction shows that for any exact sequence in Proj R
0 Qn . . . Q1 Q0 0,
P
i
the Euler characteristic (Q ) =
i (1) [Qi ]proj vanishes. Using the fact that
00
(Cf ) = (P ) (P ) and (Cf 0 ) = (P0 ) (P00 ), the result follows.
R)
0 Mn . . . M1 M0 0,
P
the Euler characteristic (M ) = i (1)i [Mi ]f g vanishes. Hence for any nitely
generated R-module and any projective resolution
0 Pn . . . P1 M 0,
we have
!
([M ]f g ) =
X
(1)i+1 [Pi ]proj
i1
X
i1
(1)i+1 [Pi ]f g = [M ]f g .
1.3
In this section we prove that every nitely generated projective R-module is stably
trivial, where R = Z[Zn ]. This amounts to proving that the inclusion Z , R
Denition 1.3.1.
have clearly (R[t] R Q) ' Q for every projective R-module which is bounded
below, it suces to show that R[t] R (P ) ' P for every projective R[t]-modules
which is bounded below. Now consider the quotient map f : P (P ) given
by the denition of (P ), which can be seen as an epimorphism of graded Rmodules. Since (P ) is projective, there exists a right inverse g : (P ) P ,
which yields a map of graded R[t]-modules h : R[t] R (P ) P . Now clearly
(h) : (R[t] R (P ))
(P ). Hence, since one can easily check that is right
exact, we have (coker h) = 0. But since coker h is bounded below, it follows from
1.3.2 that coker h=0, hence h is isomorphic. Now since P is projective, h splits
14
the left exactness. Note that R R[t] M may also be written as M/(t 1)M ,
so this amounts to proving that for any graded R[t]-modules M and any graded
submodule M 0 , then (1 t)M M 0 = (1 t)M 0 . But this follows by an easy
induction, since if an element x = x0 + x1 + . . . M is such that (1 t)x =
x0 + (x1 tx0 ) + . . . + (xn txn1 ) + . . . M 0 , then every xn is actually in M 0 .
Theorem 1.3.6
morphism.
(Grothendieck).
Proof. It is easy to see that the map K0 (R[t]) K0 (R) dened above is a left
inverse to the map we are considering. Hence it suces to prove the surjectivity.
Let P be a projective R[t]-module. We are going to prove that it is of the form
R[t] R[t,s] N for some graded R[t, s]-module. To see that, rst observe thet P =
R[t]n /Q for some n 0 and some module of relations Q R[t]n . Since M is
nitely generated and R is noetherian, it follows from the Hilbert Basis Theorem
that Q is nitely generated. Now choose a nite set of generators of Q
Since R is regular, R[t, s] is regular by the Syzygy theorem, hence we can choose
a nite resolution
0 Pm . . . P1 N 0.
Using 1.3.5, the following sequence
Lemma 1.3.7. If
0 Pm0 . . . P00 0
by the same reasoning as above. Now since the map K0 (R) K0 (R[t, t1 ])
factors through K0 (R[t]) by means of the ring homomorphisms R R[t]
R[t, t1 ], it suce to show that K0 (R[t]) K0 (R[t, t1 ]) is surjective, by ??. Let
P be a nitely generated projective R[t, t1 ]-module. Then P = R[t, t1 ]n /Q for
some module of relations Q R[t, t1 ]n . Since P is nitely generated and R is
noetherian, it follows from the Hilbert Basis Theorem that Q is nitely generated.
Thus we can choose d large enough so that td Q R[t]n . Hence
P ' td R[t, t1 ]n / td Q ' R[t, t1 ] R[t] M
for some nitely generated R[t]-module M . Using the Syzygy theorem, choose a
nite resolution
0 Pm . . . P 1 M 0
16
by nitely generated projective R[t]-modules. Since R[t, t1 ] is a at over R[t], the
following sequence is exact
Corollary 1.3.9. The natural map K0 (Z) K0 (Z[Zn ]) is an isomorphism for all
n 0.
1.4
The vanishing of
W h(Zn).
Theorem 1.4.1
all n 0.
(Bass-Heller-Swan).
tm 0
0 1
(1 + A(t 1)),
0
1
m
t B td Bd td1 Bd
,
t
1
and the last matrix has entries of degree d1, so we conclude by induction. Thus
we have only to deal with the case tm B B0 + tB1 = (B0 + B1 ) + (t 1)B1 . Since
tm B must be invertible as a matrix over R[t, t1 ], B0 +N1 is invertible. So factoring
out by B0 + B1 GL(R), we have a matrix of the form 1 + A(t 1) = (1 A) + tA.
Let Cr , . . . , Cs M (R) with s, r > 0, such that
Lemma 1.4.3. Let P be a projective R[t, t1 ]-module, and an unipotent automorphism of P . Then [P, ]proj = 0 K1 (R[t, t1 ]).
Proof. By the Resolution theorem 1.2.5, it is enough to prove the result in R[t, t1 ]mod . Suppose s = 0, and let Mi = Im (1)si . Each Mi is stable under , and
fg
induces identity on the quotient Mi+1 /Mi , yielding the following commutative
diagram in R[t, t1 ]-modfg
0
Mi+1
Mi
Mi
Mi+1
Mi+1 /Mi
/
Id
Mi+1 /Mi
0.
19
(i)
(i)
(i)
0 P1 P (i) P2 0.
Now one can deduce the following commutative diagrams
/
(i)
P1
/ P (i)
P (i)
P1
tId
(i)
tId
tId
P (i)
/ P (i)
2
0.
So,
X (i)
(i)
(i)
(Q, tId) = R[t, t1 ]s , tId +
(P , tId) (P1 , tId) (P2 , tId) 1 (Proj R[t, t1 ]),
i
20
Chapter 2
From Algebra to Topology: Splitting
obstructions
In this chapter, we study a geometric phenomenon which will be of crucial importance in calculating the PL-structure set of the torus: codimension one splitting.
Let f : M M 0 be a homotopy equivalence, and N 0 a two-sided codimension one
submanifold of M 0 . By making f transverse to N 0 , it induces a degree one map
g : N N 0 , with N = f 1 (N 0 ). The problem is to decide if we can homotop f
to make g a homotopy equivalence. The idea will be to make g highly connected
by performing succesive surgeries on N . Note that, instead of classical surgery,
everything is done inside M , so we will need dierent assumptions to make sure
we can perform surgery on a class.
In the rst section, we prove that this program can be carried out until middle
dimension, where a obstruction to perform surgery will appear. In Section 2.2, we
prove that the vanishing of this obstruction gives algebraic moves, whose geometric counterparts will be developed in Section 2.3 to achieve the surgery program,
yielding the following
Splitting Theorem.
Let f : M Tn be a homotopy equivalence between PL-manifolds of dimension
n 6, and N 0 a two-sided codimension submanifold of Tn . Then we can homotop
f such that:
- f is transverse to N 0 .
- the restriction f| : N N 0 is a homotopy equivalence, where N = f 1 (N 0 ).
Finally, we relate the algebraic obstruction to the algebraic machinery developed
21
in the previous chapter in Section 2.4. This last section is not necessary for the
rest of this thesis, but presents the link between the Whitehead group of a group
and the Nil group of its groups ring given by the general Bass-Heller-Swan theorem.
can rst assume that N is connected. Now g : N N 0 is a degree one map between
connected compact manifolds, thus induces an epimorphism g : 1 (N ) 1 (N 0 ).
Indeed, consider the covering map Z N 0 associated to g 1 (N ). Then by
denition g lifts to g : N Z , and g : Hn1 (N ) Hn1 (N 0 ) factors through
g
Hn1 (N )
Hn1 (Z) Hn1 (N 0 ). Since g is a degree one map, Z is compact
(otherwise Hn1 (Z) = 0), so g 1 (N ) is a subgroup of 1 (N 0 ) of nite index d, and
Hn1 (Z) Hn1 (N 0 ) is multiplication by d. Hence d = 1 and g 1 (N ) = 1 (N 0 ).
Now consider a loop representing an element of ker g . Then, since f : M M 0 ,
the commutativity of the following diagram
/ 1 (M 0 )
1 (N )
g
/ 1 (M )
1 (N 0 )
22
Lemma 2.1.2. A degree one normal map between compact connected orientable
manifolds induces an epimorphism on Hi , i 1.
Proof. Let h : X Y be a degree one normal map between compact connected
Hi (X)
D
Hi (Y )
D
mi
(X)
23
mi
(Y )
Denition 2.1.3.
the inclusion Zn1
f
YM
/ YM
p0
M 0.
.
By denition of p0 , N 0 lifts to a two-sided codimension one submanifold that
we still denote N 0 . N 0 divides YM 0 into AN 0 and BN 0 . The group of covering transformations is isomorphic to Z, and we choose a generator t such that
tAN 0 AN 0 .Furthermore, since 1 (N ) 1 (N 0 ) is an isomorphism, N lifts to a
two-sided codimension one submanifold that will still denote by N . We can choose
a lift such that f (N ) N 0 . N divides YN into AN and BN , and considering the
generator t of the innite cyclic group of covering transformations of p induced
by the isomorphism 1 (N ) 1 (N 0 ), we have f (AN ) AN 0 , f (BN ) BN 0 and
tAN AN .
24
Denition 2.1.4.
We now describe how an ambient surgery (analog to the one done in 2.1.1 )
will be performed. Suppose we have an embedding (Dk , Sk1 ) , (N, AN tAN )
(AN tAN is a fundamental domain of the innite cyclic covering). Then the
normal bundle of the embedded Dk is trivial, thus admits a tubular neighboorhood of the form Dk Dnk . By homotoping f we can suppose that f maps
this tube in N , and by choosing a suciently small tubular neighboorhood we
can assume that the image of this tube is not dense in N . Consider the codimension one submanifold N1 obtained by ambient surgery on this tube (namely,
N1 = N (Sk1 Dnk )Dk Snk1 ). This submanifold projects to a submanifold
N1 in M . By considering a tubular neighboorhood around N , we can assume that
there is a cobordism W between N and N1 , as described in the following picture
Hi (YM , BN )
Hi (YM , tBN , )
(t1 )
Hi (YM , BN )
excision
(t1 )
Hi (AN , N )
excision
Hi (AN , N ),
modules are nitely generated by 2.1.6, it is enough to prove that any element
x is killed by a suciently high power of (t1 ) . Let c be a cycle representing
x. Since c has compact support, there exists a l such that the support of c lies
in AN tl AN . But then x is killed under the map Hi (YM , BN ) Hi (YM , tl BN )
induced by inclusion, and so (t1 )l x = 0.
The following lemma exhibits homology classes on which surgery will be possible. For a proof, we refer to [FH73].
Proof. We keep the same notations as above. Using 2.1.1, we can suppose that
there exists a codimension one connected submanifold N M such that g is a 1 isomorphism. We prove the proposition by induction on k . Since we just proved
it for k = 0, 1, let assume g is k -connected for k < n/2. If k + 1 > n/2, the
result follows by Poincar duality. Otherwise Kk+1 (AN , N ) is a nitely generated
R-module endowed with a nilpotent R-endomorphism (t1 ) , whose nilpotence is
denoted l. Since (t1 ) Kk+1 (AN , N ) is also nitely generated, let x1 , . . . , xs be a
set of generators. By 2.1.8 there are elements in k+1 (AN tAN , N ) whose images
under the composite map are x1 , . . . , xs . Since k + 1 < n/2, we can represent
them by disjoint embeddings (Dk+1 , Sk ) , (AN tAN , N ). Using the procedure
described above, we perform surgery on these embeddings to obtain a map homotopic to f (for simplicity, we still denote it f ) and a codimension one submanifold
N1 N . Let AN1 , BN1 be the corresponding sets for N1 . We have the following
commutative diagram
/
Kk+1 (AN , N )
excision
Kk+1 (AN1 , N1 )
excision
Kk+1 (YM , BN )
Kk (BN1 , BN ).
Kk+1 (AN , N )
(t1 )
Kk+1 (AN1 , N1 )
0,
it follows that (t1 ) Kk+1 (AN1 , N1 ) = 0. We can repeat the argument. Thus after
nitely many times, Kk+1 (AN , N ) can be killed. Note that during that procedure,
Kk+1 (BN , N ) has not been aected. Indeed, since W is obtained from N1 by
adding n k 1 handles and n k 1 > n/2 since k + 1 < n/2, it follows
that Kk+1 (YM , AN1 ) Kk+1 (YM , AN ) is an isomorphism, hence Kk+1 (BN , N ) '
Kk+1 (BN1 , N1 ). So we can apply a similar program to kill Kk+1 (BN , N ). This
completes the induction step.
28
2.2
Lemma 2.2.1. (i) Suppose n = 2k is even. Then under the above assumptions,
we can homotop f so there exists a codimension one submanifold N such that the
induced map g : N N 0 satises: Ki (AN , N ) = 0, Ki (BN , N ) = 0 for i < k,
Kk (AN , N ) = 0 and Kk (BN , N ) is nitely generated.
(ii) Suppose n = 2k + 1 is odd. Then under the above assumptions, we can
homotop f so there exists a codimension one submanifold N such that the induced
map g : N N 0 satises: Ki (AN , N ) = 0, Ki (BN , N ) = 0 for i k, Kk+1 (AN , N )
and Kk+1 (BN , N ) are nitely generated.
Proof. (i) Using 2.1.9, we can assume that Ki (AN , N ) = 0, Ki (BN , N ) = 0 for i <
k . If we try to carry out the same procedure as in 2.1.9, the only problem arising is
the utilisation of the Whitney trick to represent x1 , . . . , xs by disjoint embeddings.
However, since 1 (N ) 1 (AN tAN ) is an isomorphism by denition of the
innite cyclic covering, we can use the Wall's piping out argument to represent
them by disjoint embeddings (we refer the reader to the Chapter 4 of [Wal70]
for the proof of this lemma). Then the same reasoning aplies and one can kill
Kk (AN , N ). By the same argument as in 2.1.5, we have
K i (AN , N, R) K i (BN , N, R) = K i1 (N, R)
for any R-module R. Now K i1 (N, R) = 0 for i > k by Poincar duality, so
K i (BN , N, R) = 0 for i > k . Hence Kk (BN , N ) is a nitely generated projective
R-module.
(ii) By 2.1.9, we can assume that Ki (AN , N ) = 0, Ki (BN , N ) = 0 for i k . Now
since K i1 (N, R) = 0 for i > k + 1 by Poincar duality, we have K i (AN , N, R)
K i (BN , N, R) = 0 for i > k + 1, thus K i (AN , N, R) = 0 and K i (BN , N, R) = 0
for i > k + 1. Hence Kk+1 (AN , N, R) and Kk+1 (BN , N, R) are nitely generated
projective R-module.
Denition 2.2.2.
almost splitting.
Recall that the Grothendieck's theorem implies that K0 (R) ' Z, hence every
nitely generated projective R-module is stably free.
Denition 2.2.3.
Recall that, since K0 (R) ' Z by the Grothendieck theorem 1.3.8, every nitely
generated projective R-module is stably free. Hence, using the usual notations,
f R if n = 2k +1
(Kk+1 (BN , N ), t ) (resp. (Kk (BN , N ), t )) denes an element in Nil
(resp. n = 2k ). We have the following
Denition 2.2.4.
g
We say that an object (P, ) in N
il R is triangular if there
exists a ltration 0 = E0 E1 . . . Rr = P such that Ei+1 /Ei is free and
(Ei+1 ) Ei for all i.
An elementary example of a triangular object is given by the following
P
Lemma
2.2.5.
If
P is a free R-module, then P R R[x]/(xr ) with ( i pi xi ) =
P
i+1
is a triangular object.
i pi x
An important property of triangular object is the following
g
Proposition 2.2.6. Any triangular object represents the zero element in N
il R.
Proof. We proceed by induction on the length of the ltration by which the object
0 (P 0 , 0 )
(P 00 , 00 )
(P, ) 0
30
Fm1
vm1
Em1 .
v
f R = 0.
Proposition 2.2.8. Nil
g
Proof. Let [P, ] N
il R, with m = 0,
since each Si+1 /Si is stably free by the Grothendieck theorem 1.3.8.
31
Thus the element (Kk (BN , N ), t ) or (Kk+1 (BN , N ), t ) is the zero element in
g
N il R = K0 (Nil R). This in turn gives information on the structure of the pair,
which can be thought as an analog of the fact that an element representing 0 in
g
K0 (R) is stably trivial. Namely, an element representing the zero element in N
il
R = K0 (Nil R) is stably trivial in the following sense.
g
Proposition 2.2.9. An object (P, ) Nil R represents 0 N
il R if and only if
there exists triangular objects (T1 , t1 ), (T2 , t2 ) such that
(P, ) (T1 , t1 ) ' (T2 , t2 ).
For a proof, we refer to [Ko9]. We will see in the next section the geometric
operations corresponding to adding or removing a triangular object, which will
allow us to kill the remaining homology kernel.
2.3
We are now going to prove the splitting theorem. So far, we have proved that one
can homotop f so as to have an almost splitting. Furthermore, we saw in the last
section that the remaining homology kernel is stably triangular. In this section,
we describe the geometric operations which allow us to add or remove a triangular
object, thus proving the Splitting Theorem.
with (P, v) ' (Kk (BN , N ), t ) and (F, f ) a triangular object, there exists an almost
splitting (N1 , g1 ) such that (Kk (BN1 , N1 ), t ) ' (P1 , 1 ).
(ii) Suppose that n = 2k + 1 7, and (N, g) is an almost splitting. If we have
an exact sequence in Nil R
v
0 (P, ) (P1 , 1 )
(F, f ) 0
with (P, v) ' (Kk+1 (BN , N ), t ) and (F, f ) a triangular object, there exists an
almost splitting (N1 , g1 ) such that (Kk+1 (BN1 , N1 ), t ) ' (P1 , 1 ).
32
Proof. The proofs are essentially the same, except for some number changes. We
with (P, v) ' (Kk (BN , N ), t ) and (F, f ) a triangular object, there exists an almost
splitting (N1 , g1 ) such that (Kk (BN1 , N1 ), t ) ' (P1 , 1 ).
33
Proof. Once again we can suppose that (F, f ) ' (R, 0). Let a be an element of
Kk (BN , N ) ' P generating F . Let l be the least integer such that tl Kk (BN , N ) =
0. Since t a = 0, a ker t Im tl1
. Thus, using 2.1.8 and Wall's piping out argument, we can represent a by an embedding under the Hurewicz homomorphism.
Carrying out again our surgery program, we obtain a new almost splitting (N1 , g1 )
and a cobordism W between N and N1 . Using the exact sequence of the triple
(BN , W, N )
0 Kk (W, N ) Kk (BN , N ) Kk (BN , W ) ' Kk (BN1 , N1 ) 0,
and using the fact that it preserves the action of t , it follows that this sequence
is isomorphic to
0 (R, 0) (P, ) (P1 , 1 ) 0.
Furthermore, by looking at the exact sequence of the triple (AN1 , W, N1 ), one can
check that Kk (AN , N ) is not aected during the procedure.
with (P, v) ' (Kk+1 (BN , N ), t ) and (F, f ) a triangular object, there exists an
almost splitting (N1 , g1 ) such that (Kk+1 (BN1 , N1 ), t ) ' (P1 , 1 ).
This lemma is quite technical and uses ideas from the Chapter 4 of [Wal70].
We refer the reader to [FH73] for the details.
This last section presents without proof the links between the algebraic obstruction
constructed in Section 2.2 and the Whitehead torsion of the associated homotopy
equivalence. It is not needed for the rest of the thesis.
34
1 The
copy of
g
N
il(Z)
35
W h( Z)
Chapter 3
A crash course in surgery theory.
This chapter is intended to present in a very concise way the necessary background
on surgery theory and homotopy theory, which will be used to prove the topological
rigidity of the torus in high dimensions. For a (incredibly) more detailed exposition
of this material, we refer to [Ran02].
For sake of simplicity, and since this the only case we will encounter in this
thesis, we will assume that all the spaces we study have a fundamental group
with vanishing Whitehead group. That will allow us to avoid complications
in the exposition of surgery theory due to so called decorations of algebraic Lgroups. Furthermore, since we will have to consider possibly TOP-, PL-, or DIFFmanifolds, when a manifold comes with a given structure and that the discussion
applies to equally to all the above structures, we will just call it a CAT-manifold
(CAT = TOP, PL, DIFF ).
3.1
3.1.1
Surgery obstructions.
Degree one normal maps, surgery obstructions, and
L-groups.
36
is an isomorphism for all n. Since for any oriented Poincar space X there exists
a degree one map f : M X with M a compact oriented manifold, the idea is to
modify f in such a way that it becomes a homotopy equivalence (or equivalently
that it induces isomorphisms on homotopy groups). Since a degree one map induces an epimorphism on 1 by 2.1.1, one can try to perform surgery on a set of
generators of ker(f : 1 (M ) 1 (X)) to obtain a new map f1 : M1 X which
is a 1 -isomorphism. The situation is not that simple in higher dimensions since a
degree one map does not necessarily induce epimorphisms on i , i 2. However,
it induces epimorphisms on Hi , i 2 by 2.1.2. Thus, it suces to modify f1 so
that it induces an isomorphism on 1 and Hi , i 2 (or equivalently until middle
dimension by Poincar duality). It appears that to carry out such a program,
one needs some additional bundle data. Hence we are considering instead degree
one normal maps (see the denition below). Below middle dimension, an element
f
Simply-connected obstructions.
A fundamental case is the case of degree one normal maps over simply-connected
manifolds.
Z if n 0 (mod 4),
Ln (e) =
Z if n 2 (mod 4)
2
0 if n 1, 3 (mod 4)
37
0
if n 1, 3 (mod 4).
1
8 ((N ) (M ))
3.1.3
Codimension-one splitting.
such that
- 0 F is a PL-homotopy equivalence,
- 1 F is a homotopy equivalence,
- = S(F, b).
Thus, using the Realization Theorem, let
Proposition 3.1.4. (L) is a well-dened homoprhism, and the following sequence is exact and splits:
Ln+1 (i )
(L)
(, b) : N L [0, 1] [0, 1]
be a degree one normal map with surgery obstruction and such that, with N =
N + N , | : N (L [0, 1] {0}) (L {0} [0, 1]) (L {1} [0, 1])
a PL-homeomorphism. We can use it to identify the copy L {0} [0, 1] and
L {1} [0, 1] in N . By glueing them together, we obtain a normal map
Products
(, b) : M N
and a PL-manifold X . We can construct the normal map
( IdX , b Id) : M X N X.
We would like a formula for S( IdN ). The answer has been given by Morgan
[Mor78] for X simply-connected, and by Ranicki [Ran80] in the general case. There
exist symmetric L-groups Ln ( 0 ), and a pairing
Li () Lj ( 0 ) Li+j ( 0 )
such that S( IdX ) = S() (X), where
: B 0 L ( 0 )
is the Mischenko-Ranicki symmetric signature. The symmetric L-groups are hard
to compute. However, we have the following particular case:
39
Proposition 3.1.7. (i) The symmetric L-groups L (e) are 4-periodic, and given
by
Z if n 0 (mod 4),
n
L (e) =
Z if n 1 (mod 4)
2
0 if n 1, 2 (mod 4).
(ii) If N is simply connected, then
0
if dim N 2, 3 (mod 4).
(N )
Ln+m ( 0 )
/
Ln (e)
Ln+m (pr0 )
Ln+m (e),
The sets
a CAT-isomorphism, with (N1 , f1 ) and (N2 , f2 ) equivalent if there exists a CATisomorphism h : N1 N2 such that the following diagram commutes up to homotopy:
f1
/M
{=
{
{{
h
{{ f2
{
{
N1
N2 .
M,
BG
41
3.2.2
The maps
/ M.
N
.
1
1
3.2.3
One has to be careful with the surgery exact sequence. Although [M, G/CAT ]
and Lm (Z [1 (M )]) both carry a natural group structure, the surgery obstruction
map S is generally NOT a group morphism. However, S is in some special cases a
group morphism. This is the case for example if M is a suspension, or for M Dk
(k 1) appearing in the surgery obstruction map.
42
3.3
We give here some homotopy groups of the classifying spaces BG, BP L, BO, and
their associated homotopy bers G/P L and P L/O, needed for the rest of this
thesis.
3.3.1
i (P L/O)
BO, BG
and the
J -homomorphism.
i
1
2
i (BO)
Z2
Z2
i (BG)
Z2
Z2
i (BO)
i (BG)
Z2
Z2
Z2
Z2
3
4
0
Z
Z2
Z24
0
Z2
pr
Z
Z24
0
0
0
0
0
0
43
3.3.3
i (G/P L)
Theorem 3.3.4
(Stallings).
For i 6, SP L (Si ) = 0.
Corollary 3.3.5. The surgery obstruction map i (G/P L) Li (e) is an isomorphism for i 6.
Proof. Immediate from the surgery exact sequence.
In low dimensions, the same reasoning does not apply. Instead, we start by
showing that i (G/PL) and Li (e) are abstractly isomorphic, then by proving that
the surgery obstruction map is an isomorphism. We already know Li (e), i 5 by
??. Thus we compute the homotopy groups of G/PL.
Recall that P L/O is 7-connected. Thus the exact sequence
. . . n+1 (BO)
n+1 (BG) n (G/O) n (BO)
n (BG) . . .
to calucate the homotopy groups of G/P L in low dimensions.
0 1 (G/O) 1 (BO)
1 (BG) 0
hence 1 (G/O) ' 0.
sequence
0 Z2 2 (G/O)
Z2
Z2 0
hence 2 (G/O) ' Z2 .
44
Z
Z24
3 (G/O) 0
hence 3 (G/O) ' 0.
0 3 (G/O) Z
Z24 0
hence 4 (G/O) ' Z.
Proposition 3.3.11. There exists an almost parallelizable 4-manifold with signature 16.
Corollary 3.3.12. 4 (G/P L) L4 (e) is injective.
Proposition 3.3.13. There exists a degree one normal map (T2 , 2 ) (S2 , 2 )
with Arf invariant 1.
Corollary 3.3.14. 2 (G/P L) L2 (e) in an isomorphism.
Finally, we have proved the following
Theorem 3.3.15. The surgery obstruction map n (G/P L) Ln (e) is an isomorphism in dimension n 6= 4, and a monorphism in all dimensions.
45
Chapter 4
The classication of PL-homotopy
tori.
We now start our program to prove the topological rigidity of the torus in high
dimensions. According to the denition given in the previous chapter, this amounts
to proving that ST OP (Tn ) consists of a single element for n 5.
We start by computing SP L (Tn ). This might seem quite surprising since the
question of the topological rigidity of the torus is merely formulated in terms of
topological manifolds. However, this will appear to be more tractable, mainly because of the following fundamental theorem:
Our aim is to compute the structure set SP L (Tk ), k 5, and more generally
SP L (Tk Dn ), n + k 5. Our main tool is the surgery exact sequence
Tk Dn
Ln+k+1 Z 1 (Tk Dn ) SP L (Tk Dn )
46
Tk Dn
STk Dn
Tk Dn , , (G/P L, ) Ln+k Z 1 (Tk Dn ) .
: Sl1 ([0, 1]l ) , [0, 1]l Tl the attaching map of the top l-cell. For every
1 i l, i,0 = |([0,1]...{0}[0,1]) and i,1 = |([0,1]...{1}[0,1]) are attaching
maps of some (l 1)-cells, and i,0 = i,1 l2 (Tl ) because of the identication.
Now by the induction hypothesis, we can suppose that S factors through
_
C
S : S(([0, 1]l )) Sl
Sl Tl ,
W l
where the collapsing map C : Sl
S is obtained by collapsing the subsets
([0, 1] ... {1} [0, 1]), ([0, 1] ... {1} [0, 1]), 1 i l. But by denition
P
of the addition in l , l 2, this is exactly i (Si,0 + Si,1 ) = 0.
In particular, S n+1 Tk has the homotopy type of a wedge of spheres, so
_
M
M
[S n+1 Tk , Y ] = [ Si+n+1 , Y ] =
i+n+1 (Y ) =
i+n (G/P L)
with ki summands i+n for each i. But Ln+k (Zn ) = ki Ln+i (e) by (a result
which will appear in the crash course in surgery theory), and the following diagram is commutative
[S n Tk , G/P L]
STk Dn
Ln+k (Zk )
k
i n+i (G/P L)
SSn+i
47
/ k L
n+i (e)
i
Corollary 4.1.3. Every PL-manifold of dimension 5 which is homotopy equivalent to the torus is stably parallelizable.
Proof. Tk = 0, so given any homotopy equivalence h : M Tk , we have h N =
Tk = 0, hence N = 0.
Computation of
SP L(Tk Dn), n + k 5.
By the results of the previous section, it is now necessary to describe precisely the
action of Ln+k+1 (Zk ) on SP L (Tk Dn ). We start by expliciting a set of generators.
For every J {1, . . . , k}, let |J| denote its cardinal, J c its complementary,
and set T (J) = {(x1 , . . . , xk ) Tk = (S1 )k |xi = if i
/ J}. For each J with
|J| + n 1(mod 2), we associate an element (J) Ln+k+1 (Zk ) in the following
way (for simplicity, set m = |J|):
- If m + n 5, by the Realization Theorem 3.1.3, choose a degree one normal
map (M, h, F ) over ([0, 1]m+n+1 , m+n+1 ) such that S[0,1]m+n+1 (M, h, F ) is a
generator of Lm+n+1 (e). Let K be obtained from T (J) [0, 1]n [0, 1] by
taking the connected sum with M along T (J) [0, 1]n {1}. This yields the
degree one normal
(T (J) [0, 1]n [0, 1]) ] [0, 1]m+n+1 = T (J) [0, 1]n [0, 1]
with framing the connected sum of the framings. Denote (K, f, E) this degree
one normal map. We write (K, f, E) = (T (J) [0, 1]n [0, 1])](M, h, F ). By
denition of the surgery obstructions,
Proposition 4.2.1. Every element of Ln+k+1 (Zk ) has a unique expression a(J)(J),
where J is a non empty subset of {1, . . . , k} such that |J| + n 1 (mod 2), and
a(J) Z if |J| + n 3 (mod 4) (resp. a(J) Z2 if |J| + n 1 (mod 4).
P
Moreover, we have
50
Proof. Let`j0 JK . Recall that we dened (J) as S(W, , E), where (W, , E) =
(M, h, F ) (T (J)[0, 1]n+1 CP2 ), (M, h, F ) representing an element of B|J|+n+5 ([0, 1]|J|+n+5 , )
with S(M, h, F ) the chosen generator of L|J|+n+5 (e). We can take the boundary
connected sum along a disk that misses T (J {j0 }) [0, 1]n+1 CP2 ,
We are now able to prove 4.2.1. Let A the abelian group with genetors the
subsets J {1, . . . , n} with |J| + n odd, and relations 2J = 0 for |J| + n
P
1 (mod 4). Dene a map : Ln+k+1 (Zk ) A, 7 (w(J)(J))J . is surjective
51
P
P
since ( a(J)(J)) =
a(J)J . But by 3.1.5, these two groups are isomorphic.
Hence is an isomorphism, which conludes 4.2.1.
We now have to understand which elements of Ln+k+1 (Zk ) stabilize [IdTk Dn ].
First of all, note that if |J|+n is an odd integer 5 the construction of (J) directly
shows that (J) stabilizes [IdTk Dn ]. Recall that we constructed two degree one
normal maps
h : M [0, 1]m+n+1
Id ] h : (T (J) [0, 1]n [0, 1]) ] M (T (J) [0, 1]n [0, 1]) ] [0, 1]m+n+1
and constructed (J) by crossing with T (J c ) and taking the associated surgery
obstruction. Now, by denition of the action of Ln+k+1 (Zk ) on SP L (Tk Dn ), we
have
(J).[IdTk Dn ] =
(IdT (J)[0,1]n {1} ] h) IdT (J c )
n
m+n+1
c
(T (J) [0, 1] {1} ] [0, 1]
) T (J )
But (IdT (J)[0,1]n {1} ] h)IdT (J c ) is clearly a PL-homeomorphism, hence (J).[IdTk Dn ] =
[IdTk Dn ].
The situation for |J| + n = 1 is even easier. Since the connected sum of
T (J) [0, 1]n [0, 1] with S1 S1 was taken in the interior, it has no eect on the
boundary, so the restriction of the degree one normal map obtained after taking
the connected sum is a PL-homeomorphism when restricted to the boundary, and
the degree one normal map obtained after crossing with T (J c ) restricts to a PLhomeomorphism on the boundary, hence (J).[IdTk Dn ] = [IdTk Dn ].
The only remaining case is |J| + n = 3. Recall that since there exists no 4dimensional PL-manifold with vanishing w1 and w2 and signature 8 by Rokhlin's
theorem, we were forced to cross with CP2 , so the previous argument does not
apply. However, there exists a 4-dimensional PL-manifold with vanishing w1 and
w2 and signature 16 (references ?), which yields a degree one normal map (W, h, F )
over (D4 , 4 ) with obstruction twice the chosen generator of L4 (e). Once again the
periodicity of surgery obstructions shows that we could have dened 2(J) starting
with this normal map, taking the boundary connected sum with T (J) Dn [0, 1],
52
crossing with T (J c ) and evaluating the surgery obstruction of the resulting degree one normal map. Hence the same reasoning as for |J| + n 5 implies
2(J).[IdTk Dn ] = [IdTk Dn ].
We are now going to use Rokhlin's theorem to prove that there are essentially
the only elements of Ln+k+1 (Zk ) acting trivially on SP L (Tk Dn ). Namely we have
P
Proposition 4.2.5. Let = a(J)(J) Ln+k+1 (Zk ). Then acts trivially on
[IdTk Dn ] if and only if a(J) is even whenever |J| + n = 3.
P
Proof. Suppose we have an element = a(J)(J) Ln+k+1 (Zk ) acting trivially
on [IdTk Dn ] and a subset H with |H| + n = 3 and a(H) odd. The idea here
will be, given a degree one normal map representing , to lower the dimension
using splitting ideas and derive a contradiction to Rokhlin's theorem. Since even
multiples of (H) act trivially, we can assume a(H) = 1. By the Realization
Theorem, let (W, , F ) be a degree one normal map over (Tk Dn [0, 1], n+k+1 )
53
f . Denote by W the PL-manifold obtained, and let Q denote the image of the two
copies of T2 D2 in W . We then have a degree one normal map
g : W (T2 D2 ) (T3 [0, 1]) (T2 D2 ) = T2 S2 .
H2 (P Q)
H2 (P 0 Q0 )
H2 (P ) H2 (Q)
H2 (P ) H2 (Q0 )
54
i j
i0 j0
H2 (W )
H2 (W 0 )
H3 (P Q)
H3 (P 0 Q0 )
where the vertical isomorphisms come form the fact that f restricts to a
PL-homeomorphism on Q and P Q. Let x K2 (W ). The commutativity
of the right square and the fact that the map H3 (P Q) H3 (P 0 Q0 ) is an
isomorphism imply that x is sent to 0 under the map H2 (W ) H3 (P Q).
By exactness, let u H2 (P ), v H2 (Q) such that i (u) j (v) = x. The
commutativity of the middle square and the fact that x K2 (W ) imply
(i0 j0 )(f (u), f (v)) = 0. Hence by exactness there exists s H2 (P Q)
such that 0 (s) = (f (u), f (v)). Let t denote the antecedent of s under the
left vertical isomorphism. By commutativity of the left square and the fact
that H2 (Q) H2 (Q0 ) is an isomorphism, (s) = (u + y, v), with y K2 (P ).
But by exactness, (i0 j0 )(u + y, v) = 0, and x = i (y).
- : K2 (P ) K2 (W ) is injective: By the Mayer-Vietoris exact sequence, we
have
H1 (P Q)
(1)
H1 (P 0 Q0 )
/ H1 (P ) H1 (Q)
i j
(2)
/ H1 (P ) H2 (Q0 )
/ H1 (W )
/ H1 (W 0 )
(4)
(3) f
i0 j0
/ H2 (P Q)
/ H2 (P ) H2 (Q)
(5)
/ H2 (P 0 Q0 )
/ H2 (P ) H2 (Q)
where the vertical isomorphisms come form the fact that f restricts to a PLhomeomorphism on Q and P Q, and the assumption that f induces a 1
isomorphism. Let x K2 (P ) such that (x) = 0. By exactness, there exists
y H2 (P Q) such that (y) = (x, 0). By commutativity of the right square,
the image z of y under the right vertical isomorphism is sent to 0, so there
exists v H1 (W 0 ) such that 0 (v) = z . Now we use the following classical
fact on exact sequences: In the previous situation, if the vertical maps (1),
(2) and (4) are surjective, the map (3) is injective. Thus the map H1 (W )
H1 (W 0 ) is injective. But this argument works equally for homology with
f Id
55
Lemma 4.2.7.
(i) H 1 (W, P ; Z2 ) = 0;
Proof.
(i) We have
(by excision)
H 1 (W, P ; Z2 ) = H 1 (Q, P Q; Z2 )
= H3 (Q; Z2 )
=0
(ii) We have the following commutative diagram
H 1 (P ; Z2 )
f
H 1 (P 0 ; Z2 )
H 2 (W, P ; Z2 )
H 2 (W 0 , P 0 ; Z2 ).
H 1 (W 0 , P 0 ; Z2 )
H 1 (W 0 ; Z2 )
/ H 1 (P 0 ; Z
2)
H 2 (W 0 , P 0 ; Z2 )
Furthermore,
H 2 (W 0 , P 0 ; Z2 ) = H 2 (Q0 , P 0 Q; Z2 )
0
= H2 (Q ; Z2 )
(by excision)
(by Lefschetz duality)
= H2 (T2 D2 ; Z2 ) H2 (T2 D2 ; Z2 )
= Z2 Z2
56
H 1 (P 0 ; Z2 )
H 2 (W 0 , P 0 ; Z2 ) to prove the surjectivity of ,
- (W 0 ) = 0.
First remark that the Splitting Theorem applies in the other cases, even if we are
considering manifolds with boundaries. We then start with a degree one normal
map f : P T3n Dn , n = 0, 1, 2.
- n = 1:
Glue P 0 = T2 D2 and Q0 = T2 D2 along their common boundary T2 D2 ,
to obtain W 0 = T2 S2 . Nothing is changed, except in proving that is onto.
it is now immediate that is either surjective or zero since H2 (Q0 ; Z2 ) = Z2 .
If it was zero, we would have
=0
H 2 (W 0 , P 0 ; Z2 )
H 2 (P 0 ; Z2 )
H 2 (W 0 ; Z2 ) , H 3 (W 0 , P 0 ; Z2 )
but the last injection is impossible by cardinality.
- n = 2:
Glue T1 D3 and Q0 = D2 S2 along their common boundary S1 S2 to obtain
W 0 = S4 . (S4 ) = 0, and H3 (Q0 ; Z2 ) = 0. Furthermore H2 (Q0 ; Z2 ) = Z2 , so
is again either surjective or zero. If it was zero, we would have a surjection
H 1 (W 0 ; Z2 ) 0 Z2 H 1 (P 0 ; Z2 ).
- n = 3:
Glue P 0 = D4 and Q0 = P 0 along their common boundary. This time all
groups involved are zero and the result follows.
57
Finally for n 4 the previous construction of (J) shows that every element of
Ln+k+1 (Zk ) acts trivially on SP L (Tk Dn ), which conludes the proof of 4.2.5.
We are now able to give a simple description of SP L (Tk Dn ).
Denition 4.2.8.
J () = w(J)(J)(mod2)
called a geometric coordinate of .
Consider a basis t1 , . . . , tk of H 1 (Tk ; Z2 ). For a subset J = {i1 , . . . , i|J| |i1 <
. . . < i|J| } {1, . . . , k} with |J| + n = 3, let tJ = t1 . . . t|J| . Then the (tJ ) form
a basis of H 3n (Tk Dn ; Z2 ). We have:
x 7
J ()tJ
|J|+n=3
58
Chapter 5
PL-structures on topological
manifolds.
Considering a homotopy equivalence f : M Tk , we want to prove that M is actually homeomorphic to Tk . Such a homeomorphism naturally endowes M with a
PL-structure. Our rst task is then to prove that M indeed admits a PL-structure.
Note that a PL-structure on M gives rise to a PL-structure on its (topological)
tangent bundle M : M B TOP. In other words, there exists a lift
B
: PL
M
B TOP.
59
5.1
We are now going to study the eect of a nite covering on the PL-structure set.
Namely, we want to prove that a PL-manifold homotopy equivalent to Tk Dn
(n + k 5, n 6= 3) has a nite cover PL-homeomorphic to Tk Dn . We rst dene
what coverings we will be interested in.
f p
Tk Dn
p
Tk Dn
f
M
Tk Dn [0, 1]
h (pId)
pId
Tk Dn [0, 1]
f
one normal map (h, h F ) : (M , M ) (Tk Dn [0, 1], n+k+1 ). One easily
checks that this denes maps
p]
/S
Ln+k+1 (Zk )
p!
k
P L (T
Ln+k+1 (Zk )
Dn )
pt
SP L (Tk Dn )
following sequence
(H)
0 Ln+k+1 (1 (T (H)Dn ))
Ln+k+1 (1 (Tk Dn )) Ln+k (1 (T (H)Dn )) 0
there exists Ln+k+1 (1 (T (H) Dn )) such that (J) = j (). Let us describe
geometrically the eect of p! on such an element.
Let (h, F ) : M (T (H) [0, 1] [0, 1], ) a degree one normal map with
surgery obstruction and such that, with M = M + M , h| : M
(T (H) [0, 1] {0}) (T (H) {0} [0, 1]) (T (H) {1} [0, 1]) is a PLhomeomorphism. We can use it to identify the copy T (H) {0} [0, 1] and
T (H) {1} [0, 1] in M . By glueing them together, we obtain a normal map
(f, E) : (N, N ) (T (H)S1 [0, 1], ) whose surgery obstruction is j () = (J).
Now it is clear that p] (N, f, E) can be obtained as follows: Consider d copies
(Mi , hi , Fi ) of (M, h, F ). Glue together the copy T (H) {1} [0, 1] in Mi and
T (H) {0} [0, 1] in Mi+1 , 1 i d 1, and denote P the space obtained.
Let ti : T (H) [0, 1] [0, 1] T (H) [0, 1] [0, 1], (x, y, t) 7 (x, y, (t + i 1)/d).
Let : (t1 f1 ) . . . (td fd ) and D = F1 . . . Fd .
62
It is now not hard to see that p] (N, f, E) is obtained by glueing together the the
copy T (H) {0} [0, 1] in M1 and T (H) {1} [0, 1] in Md , i.e p] (N, f, E) =
j (S(P, , D)). But by additivity of surgery obstructions, S(P, , D) = dS(M, h, F ),
hence p! (J) = p! j S(M, h, F ) = dj S(M, h, F ) = d(J).
We thus have proved the following
/J
1 if H = J and i
t
H (p (J)) =
d if H = J and i J
0 otherwise
Recall that p : H 3n (Tk ; Z2 ) H 3n (Tk ; Z2 ) is given by p (tJ ) = dtJ if i J ,
p (tJ ) = tJ sinon. Furthermore, every nice covering map p(x1 , . . . , xk , y1 , . . . , yn ) =
(xd11 , . . . , xdkk , y1 , . . . , yn ) can be written as a composite of coverings we have been
considering previously. Thus we have proved the following
63
SP L (Tk Dn )
'
'
H 3n (Tk ; Z2 )
H 3n (Tk ; Z2 ).
5.2
T OP/P L
Denition 5.2.1. Let M be a topological manifold whose boundary is a PLmanifold. A PL-structuralization is a homeomorphism h : N M with N a
PL-manifold and such that h|N : N M is a PL-homeomorphism. Two PLstructuralizations hi : Ni M, i = 0, 1 are called concordant if there exists a PLhomeomorphism : N0 N1 and a homeomorphism H : N0 [0, 1] M [0, 1]
such that:
- HN0 0 = h0
- HN0 1 = h1
- H : N0 [0, 1] M [0, 1] coincides with h0 Id0,1 .
A PL-structure on M is an equivalence class of PL-structuralizations. Let TP L (M )
denote the set of all PL-structures on M .
As we mentionned it earlier, a PL-structure on M yields a PL-structure on the
topological stable tangent bundle. The converse is true, but requires the dicult
theorem
64
The proof can be found in [KS77]. Here is at least the construction of the maps
in both senses.
Consider from now M as a PL-manifold. A PL-structuralization h : N M
yields an isomorphism of topological stable tangent bundles, since h is a homeomorphism. It also yields a new PL-structure on the stable tangent bundle on M .
The situation is resumed in the following diagram, in which the upper triangle and
both lower triangles are commutative.
TOP/PL
/ B PL
6 w;
w
ww
P L
ww
w
ww
h
ww P L
w
ww
ww
w
w
/ B TOP
M
T OP
65
M RHp
HH
HH
HH
HH
H$
E()
yy
yy
y
y
y| y
Since E() is a PL-manifold, the homeomorphism H endowes M Rp with a PLstructure. But now by the Product Structure Theorem, this yields a PL-struture
on M . More precisely, there is a homeomorphism g : N M with N a PLmanifold such that H and g IdRp are concordant. We then dene the map
: 7 g .
Note that a PL-structuralization of M can be seen as representing an element of
the structure set. In other words, there is a forgetful map : TP L (M ) SP L (M ).
We have the following
Tk Dn
Tk Dn
66
Tk Dn
Proof. Identify Tk with (S1 )n and denote e its basepoint, and endow it with its
invariant metric. For an integer > 0, consider the nice covering map p :
Tk Dn Tk Dn , (x, y) 7 (x , y) . Let ht : Tk Dn Tk Dn be a homotopy
rel (Tk Dn ) between h = h0 and Id = h1 (for the special case n = 0, note that
since 1 (Tk ) acts trivially on [Tk , Tk ], we can assume that the ht are basepoint
preserving). The pullback covering h p is isomorphic to p . Thus we have the
following commutative diagram
Tk DIn
/ E(h p )
II
II
II
II
II
p III
II
II
I$
Tk Dn
h p
T D
n
h
T Dn
t : Tk Dn , t [0, 1]
Similarly, we can construct a continuous family of maps h
0 = h
and such that p h
t = ht p .
with h
1 is a
Consider rst the case n = 0. We can assume that h(e)
= e. Thus h
t (e) is a continuous path in the discret set
deck transformation. But since t 7 h
1
1 (e) = h
0 (e) = e, hence h
1 = Id. Now for all x p1 (e), t 7 h
t (x) is a
p (e), h
0 ), h(x))
d(x, h(x))
d(x, x0 ) + d(x0 , h(x))
= d(x, x0 ) + d(h(x
2
IdTk ) , where
So for every > 0 there exists an integer 0 such that d(h,
IdTk ) = supxTk d(h(x),
d(h,
x). Now, by local contractibility of the space of homeomorphisms of a compact topological manifold, this implies that there exists a nice
isotopic to IdTk .
nite covering such that h lifts to a homeomorphism h
67
For n > 0, the previous argument is not sucient since the isotopy has no
reason to be rel (Tk Dn ). To overcome that, we need to do the previous procedure
"far from the boundary. More precisely, for 0 < < 1, let D Dn the disk of
stabilizes every element of p1 ({e}
radius . This time, we can assume that h
(D )). Now choose > 0. Choose 0 < < such that d(h(x), h(y)) < whenever
d(x, y) < , choose close enough to 1 so that given an element x Tk D , one
n
can choose an element x0 p1
({e} (D )) such that d(x, x0 ) , and choose
large enough so that any closed (isometric) foundamental domain has diameter less
|Tk D , IdTk D ) 2. Thus
than . The same reasoning as above shows that d(h
there exists a nice covering such that there exists an isotopy : Tk D [0, 1]
|Tk D . One then construct the desired
Tk D [0, 1] between IdTk D and h
k
n
isotopy rel (T D ) by setting:
(
(a, t)
if |a|
(a, t) =
1|a|
(a, 1 ) if |a|
[(Tk Dn , ), (TOP/PL, )]
TP L (Tk Dn )
SP L (Tk Dn )
Corollary 5.2.7.
5.3
We can now turn back to our problem of determining the PL-structures a topological manifold might carry. We have just proved that we have the following
69
bration
K(Z2 , 3) B PL B TOP
Since
H 4 (BTOP/PL; 4 (BT OP/P L)) ' Hom(H4 (BTOP/PL; Z), 4 (BT OP/P L)).
Denition 5.3.1. (i) Under the natural map BTOP BTOP/PL, pullbacks
to a class
H 4 (BTOP; Z2 ),
Proof. Let f : M BTOP the classifying map of the topological tangent bundle
isomorphism given by
g 7 g .
(f ) = f ( ) = f = (M ) = 0.
The second assertion follows immediately from the properties of the bration
TOP/PL BPL BTOP.
70
Chapter 6
Topological rigidity of the torus
We are now going to prove the main theorem of this thesis. Recall that we found
in Chapter 5 a necessary and sucient condition for a manifold of dimension 5
to admit a PL-structure. In Chapter 4, we were able to compute the PL-structure
set SP L (Tn ), n 5. We now combine these results to deduce the rigidity theorem.
There is still one obstacle left. In order to prove that a high-dimensional
homotopy torus admits a PL-structure, we need to prove the vanishing of its
Kirby-Siebenmann obstruction. But since lies in the cohomology of BTOP, this
obstruction is generally not preserved under homotopy equivalence. Section 5.1 is
thus devoted to constructing an intermediary charactristic class G (M ) carrying
enough data on (M ) to derive the existence of a PL-structure on a homotopy
torus from its vanishing, and such that G (M ) = G (Tn ) for a manifold homotopy
equivalent to Tn . In section 5.2, we nally carry out our program and prove the
rigidity theorem. This chapter follows the strategy of [HW69].
6.1
Now, since P L/O is 4-connected, 3 (B PL) ' 3 (BO) ' 0, and 2 (TOP/PL) ' 0
by the theorem on the homotopy type of TOP/PL.
As for 4 (B Spin-Top) ' 4 (B Spin-Top), the exact sequence of homotopy
groups of the bration TOP/PLBPL BTOP yields
(Hurewicz)
' Hom(Z Z2 ; Z)
' Z.
Now, by the universal coecient theorem,
h : 4 (B Spin-Top) H4 (B Spin-Top, Z)
(Id)Id
(Id)0
Proposition 6.1.4.
(ii) The natural map H 4 (BST op; Z)f ree H 4 (BSpin-Top ; Z) is injective.
(iii) The natural map H 4 (B S Top; Z)f ree H 4 (BSO; Z) is injective 1 .
1 Actually,
isomoprhism.
72
H 4 (B S
Top; Z) H 4 (BSO; Z)
is an
Proof. Since Spin-Top is the universal double cover of ST op, the natural map
Spin-Top S Top yields an isomorphism (Spin-Top) Q (S Top) Q ,
and thus an isomorphism n (B Spin-Top) Q n (B S Top) Q for n 1, by
commutativity of the following diagram
n (B Spin-Top) Q
/ n (B S
Top) Q
n1 (Spin-Top) Q
/ n1 (S
Top) Q.
H 4 (B S Top; Q)
/ H 4 (BSO; Q)
O
H 4 (B Top; Q)
H 4 (BSO; Q),
n (SO) Q n (O) Q,
n (S Top) Q n (Top) Q
n (B SO) Q
/ n (B O) Q
n (B S Top) Q
n1 (SO) Q
/ n (B TOP) Q
/ n1 (O) Q
n1 (S Top) Q
73
/ n1 (TOP) Q.
It also yields an isomorphism on 0 Q ' 0, so (iii) then follows form the generalized Whitehead theorem, and (i) follows from the fact that H 4 (B SO; Z) ' Z.
Denition 6.1.5. Let ptop be a generator of H 4 (BS Top, Z)f ree , called a topological
universal Pontryagin class.
Corollary 6.1.6. There exists an integer d 6= 0 such that df ree is induced from
ptop H 4 (B(ST op); Z).
H 4 (B TOP/PL; Z2 )
'
H 4 (B Spin-Top; Z2 )
'
Hom(4 (B TOP/PL), Z2 )
/
Hom(4 (B Spin-Top), Z2 )
(Z Z2 Z2 ) H 4 (B Spin-Top; Z2 ).
Now the fundamental class H 4 (B Spin-Top; 4 (B Spin-Top)) corresponds, under the canonic identication, to the inverse of the Hurewicz map h1
B Spin-Top : Z
0Id
(Id)Id
0Id
TTTT
TTTT
TTTT
TTT)
H 4 (B5 Spin-Top; Z2 )
j
jjjj
j
j
j
jjj
jjjj
H 4 (B TOP; Z2 ).
74
X2
BSpin-G
/
BSG
/
K(3 (BSG), 3)
K(2 (BSG), 2).
BSpin-Top
BS Top
X2
BSpin-G
/ K(3 (BSG), 3)
/ K(2 (BSG), 2)
BSG
and the map BSpin-Top BSpin-G lifts to X2 if and only if the composite
/
Lemma 6.1.10. The map H 4 (X2 ; 4 (BG)) H 4 (BSpin-Top; 4 (BG)) maps the
fundamental class X2 of X2 to G .
Proof. Since X2 is 3-connected, we have
H 4 (X2 ; 4 (BG)) ' Hom(4 (X2 ), 4 (BG)) ' Hom(4 (BG), 4 (BG))
and, under this isomorphism, X2 represents Z24
Z24 . Furthemore, since BSpin-Top
is 3-connected,
Id
Hom(4 (BTOP), 4 (BG)) is represented by ZZ2 Z24 , and the result follows.
pri
Lemma 6.1.11. Let M a topological oriented manifold whose Spivak normal bration admits a spin structure M BSpin-G. Then there exists a topological
spin structure on M
BSpin-Top
9
M
ss
ss
ss
s
s
ss
/ BTOP
such that the induced spin structure on the Spivak normal bration M BSpin-Top
BSpin-G agrees with the given structure M BSpin-G.
Proof. First consider the case of an oriented manifold M BS Top with a lift
of its oriented Spivak bration M BS G to BSpin-G. Since Spin-Top (resp.
Spin-G) is the universal double cover of S Top (resp. SG), we have brations
/
K(Z2 , 1) ' F
BSpin-Top
F 0 ' K(Z2 , 1)
/
BS Top
BSpin-G
/
76
BS G.
BSpin-Top
BSpin-G
/
BS Top
/
/ BF 0
K(Z2 , 2) ' BF
BS G
BSpin-Top
M
/ BS
Top
BSpin-G
/
K(Z2 , 2) ' BF
BS G
(gf ) 0 = f = 0
hence M BS Top lifts to BSpin-Top. Furthermore, the various lifts are in bi
jective correspondence with [M, F ]. But F
F 0 yields a bijection [M, F ]
[M, F ], and [M, F ] classies the lifts of M BS G. Thus there exists a lift
M BSpin-Top compatible with both the orientation of the tangent bundle
M BS Top and the spin structure on the Spivak bration M BSpin-G.
The same reasoning applies to the brations
K(Z2 , 0) ' F
BS Top
F 0 ' K(Z2 , 0)
/
BS G
/ BG,
BTOP
77
o
ooo
o
o
oo
ooo
/ BSpin-G
BSpin-Top
9
ss
ss
s
s
ss
ss
/ BS Top
/ BS G.
H 4 (X2 ; Z24 )
FF
FF
FF
FF
FF
F
0 FFF
FF
FF
"
H 4 (B Spin-Top; Z24 )
tt
tt
t
tt
tt
tt
t
t
tt
tt
t
t
ty t
H 4 (M ; Z24 ).
Lemma 6.2.1. The topological Pontryagin class ptop is additive modulo torsion
with respect to direct sums of vector bundles.
78
f f
BS Top O BS Top
BSO BSO
BS Top
O
BSO
Let m 6= 0 such that ptop is mapped to mp1 H 4 (BSO, Z) under the map BSO
BS Top. As before, we can prove that BSO BSO BS Top BS Top induces
an isomorphism on H 4 ( ; Q), so that H 4 (BS TopBS Top; Z) ' Z Z (torsion).
Now, by additivity of the rst Pontryagin class,
Top; Z)
Tn
p
Tn
f) = 0 and M
f admits a PL-structure by the
H 4(M, Z2 ) is the zero map, so (M
Obstruction Theorem.
f being a PL-homotopy torus, it is stably parallelizable by 4.1.3. Since M
f
Now M
f) = 0 = ptop (M ).
has torsion-free cohomology, it follows from 6.2.1 that ptop (M
f, Z) is injective, which
But H 4 (M, Z) has no torsion, so : H 4 (M, Z) H 4 (M
implies ptop (M ) = 0. Thus, by 6.1.6, there exists d 6= 0 such that df ree (M ) = 0,
hence f ree (M ) = 0 since M has torsion-free cohomology. It follows that G (M ) =
i (M ), where i : Z2 , Z24 is the canonical injection. Recall that, M being a
homotopy torus, G (M ) = 0 by 6.1.9, hence i (M ) = 0. But i : H 4 (M, Z2 )
f, Z24 ) is clearly injective, so (M ) = 0, and M admits a PL-structure by
H 4 (M
the Obstruction Theorem.
jT OP
H 3 (Tn , Z2 )
= [Tn , TOP/PL] TP L (Tn ) , SP L (Tn )
= H 3 (Tn , Z2 ) .
TP L (Tn ) SP L (Tn ) .
Thus, to the homotopy equivalence f : M Tn representing an element of
SP L (Tn ) there corresponds a homeomorphism g : N Tn and a homeomorphism H : N M such that the following triangle
MO
H
/ Tn
}>
}
}}
}}
}
}
}} g
}
}
}}
}}
80
Bibliography
[BV68] J.M. Boardman and R.M. Vogt. Homotopy-everything H-spaces.
A.M.S 74, (1968).
[FH73] F.T. Farrell and W.C. Hsiang. Manifolds with 1 = G T .
Journal of Mathematics, (1973).
Bull.
American
[H. 64] H. Bass, A. Heller and R.G. Swan. The Whitehead group of a polynomial
extension. Publications mathmatiques de l'IHES 22, (1964).
[HS70] W.C Hsiang and J.L. Shaneson. Fake Tori, in Proc.
Conference . Markham Press, (1970).
Memoirs of
81
82
avail-