POINCARE IMMERSIONS
JOHN R. KLEIN
Abstract. We show that the fundamental theorem of immer-
sion theory admits a Poincare duality space analogue. Along the
way, we obtain new homotopy theoretic proofs of the existence
and uniqueness of the Spivak normal fibration of a closed Poincare
space.
1. Introduction
Assume that V n and Nn are smooth n-manifolds, possibly with
boundary and with V compact. Let g :V -! N be a map. Consider
the statements:
1. The map gx idDj :V x Dj -! Nx Dj is homotopic to a smooth
embedding for some j 0.
2. There exists a stable vector bundle isomorphism V ~= g*N , where
N and V are the stable normal bundles.
The implication 1 ) 2 is shown to hold by taking the differential of an
embedding in the homotopy class of gx idDj. The other direction, 2 )
1, can be proved as follows: a choice of bundle isomorphism gives rise
to a tangent bundle monomorphism T (V x Dj) -! T (Nx Dj) covering
gx idDj, whenever j is sufficiently large. The Hirsch immersion theorem
[Hi ] produces a smooth immersion in the homotopy class of gx idDj.
Transversality enables one to perturb this immersion to an embedding
if j is large.
Poincare immersions. It is natural to ask whether a result of this
kind holds in the category of Poincare duality spaces. Let P nand Mn
be connected Poincare spaces of dimension n (with or without bound-
ary). In analogy with the above, and by a slight misuse of terminology,
let us say that a map f :P -! M immerses if fx idDj :P xDj -! Mx Dj
is the underlying map of a Poincare embedding for some j 0 (for the
definition of Poincare embedding see x2). We call a representative em-
bedding an immersion of f. In this context we prove (see 3.1)
____________
Date: November 12, 1998.
1
2 JOHN R. KLEIN
Theorem A. (1). An immersion of f :P n -! Mn induces a stable
fiber homotopy equivalence P 'f*M where P and M are the Spivak
normal fibrations of P and M respectively.
(2). Conversely, a map f :P ! M and a stable fiber homotopy equiv-
alence P 'f*M determine an immersion of f.
For certain applications, it is desirable to know the extent to which
the fiber homotopy equivalence given Theorem A(1) is compatible with
the immersion given in Theorem A(2).
Addendum. Let OE :P 'f*M be a fiber homotopy equivalence. Then
the immersion in Theorem A(2) can be chosen so that its associated
fiber homotopy equivalence P 'f*M given by Theorem A(1) coincides
with OE up to fiber homotopy.
Spivak thickenings. The main idea in proving Theorem A(2) is the
notion of Spivak thickening, a homotopy theoretic version of the bound-
ary of a regular neighborhood of a finite polyhedron embedded in eu-
clidean space.
Let K be a connected homotopy finite space. An n-thickening of K is
a homotopy finite space A and a map i :A -! K such that the mapping
cylinder pair (K [iAx I; Ax 1) has the structure of an n-dimensional
Poincare space, and moreover, i is (n- k- 1)-connected whenever K
is homotopy equivalent to a CW complex of dimension k. An n-
thickening is a Spivak n-thickening if it comes equipped with a based
map ff :Sn -! K [iCA, such that ff*([Sn]) is a fundamental class, and
K [i CA is the mapping cone of i. The decompression of a Spivak
n-thickening (A; ff) is the Spivak (n+ 1)-thickening (K A; ff) where
K A = Kx 0 [ Ax I [ Kx 1 is the fiberwise suspension of A -! K
(note: the mapping cone of K A -! K is identified with the reduced
suspension of the mapping cone of A -! K). Two Spivak n-thickenings
(A; ff) and (A0; ff0) are elementary concordant if there exists a weak
homotopy equivalence h :A -~!A0covering the identity of K such that
the induced map of mapping cones K [ CA -! K [ CA0 transfers ff to
ff0. Stable concordance of Spivak thickenings is the equivalence relation
generated by decompression and elementary concordance.
Theorem B. A connected homotopy finite space possesses a Spivak
thickening which is unique up to stable concordance.
(In fact, stable concordances between any two Spivak thickenings
can be chosen in some sense canonically, in a way which we won't try
to make precise.)
POINCARE IMMERSIONS 3
We will give two proofs of Theorem B. The first uses manifold tech-
niques: regular neighborhood theory and the Browder-Casson-Sullivan-
Wall theorem. The second proof uses equivariant duality with respect
to a topological group model for the loop space of K. The second proof
is manifold-free.
An important special case occurs when K is a closed (= without
boundary) Poincare space of dimension k. If k n- 3 it can be
shown that the homotopy fiber of the map A -! K underlying an
n-thickening is homotopy equivalent to an (n- k- 1)-sphere (see [Br2 ,
I.4.4] or [Kl1 , Th. B]). Theorem B immediately gives the classical ex-
istence and uniqueness results for the Spivak normal fibration:
Corollary C (Spivak [Sp ], Wall [Wa2 ]).Assume that Kk is a closed
Poincare space. Then there exist an integer n k, an (n- k- 1)-spherical
fibration :S() ! K and a map ff :Sn -! K such that the cap prod-
uct U \ ff*([Sn]) is a fundamental class for K, where K denotes the
Thom space of , and U denotes a Thom class for .
When n k is large, any two such pairs (E; ff) and (E0; ff0) are re-
lated by a stable fiber homotopy equivalence E '-!E0 such that the
induced map of Thom spaces transfers ff to ff0 up to homotopy.
The classical proof of Corollary C invokes regular neighborhood the-
ory for existence and Spanier-Whitehead duality for uniqueness. Our
two proofs of Theorem B yield two proofs of Corollary C. The first
one more-or-less amounts to the classical proof, but the second one
seems to be new, and is based on the existence of Equivariant Spanier-
Whitehead duality maps. Bill Dwyer has informed me that he was also
aware of an argument of this kind. A rough sketch of the second argu-
ment yielding the existence part of Corollary C in the simply connected
case appears in [Kl2 ].
Outline. The material in x2 is for the most part language. In x3 we
prove the first part of Theorem A and provide a quick proof of the sec-
ond part when the source also admits the structure of a closed Poincare
space of dimension p n. In x4 we give the manifold proof of Theo-
rem B. In x5 we prove the second part Theorem A as a consequence
of Theorem B. In x6 we develop equivariant duality. In x7 we provide
a second proof of Theorem B using the material in x6. In x8 we prove
the addendum to Theorem A.
Acknowledgements. I would like to thank Andrew Ranicki for bringing
these issues to my attention.
4 JOHN R. KLEIN
2. Preliminaries
Spaces. Our ground category is Top, the category of compactly gen-
erated Hausdorff spaces, equipped with its usual Quillen model cate-
gory structure (weak homotopy equivalences, Serre fibrations and Serre
cofibrations; see [Qu , II.3]).
A non-empty space is always (-1)-connected. A connected space is
0-connected, and is r-connected for some r > 0 if its homotopy groups
vanish up through degree r, for any choice of basepoint. A map of non-
empty spaces X -! Y is called r-connected if its homotopy fiber with
respect to any choice of basepoint in Y is an (r- 1)-connected space.
An 1-connected map is a weak equivalence.
A space is homotopy finite if it is homotopy equivalent to a finite
CW complex.
If Y - X -! Z is a diagram of cofibrant spaces, its homotopy
pushout is the double mapping cylinder Y x 0 [ Xx [0; 1] [ Zx 1. If
Y -! X is a map of cofibrant spaces, then the homotopy pushout of
X- Y -! X is the (unreduced) fiberwise suspension X Y of Y -! X;
it comes equipped with an evident map X Y -! X.
If Y -! X is a map of cofibrant spaces, we will often write (X ; Y )
for the cofibration pair (X [Y x0Y x [0; 1]; Y x 1) associated with the
mapping cylinder.
A commutative diagram of spaces
A --- ! C
? ?
? ?
y y
B --- ! D
is cocartesian (nee homotopy cocartesian) if the induced map from the
homotopy pushout of B- A -! C to D is a weak equivalence.
Finally, a note about usage. In the majority of instances below, the
term `space' will refer to a cofibrant object of Top, i.e., the retracts of
objects built up from the empty space by attaching a finite number of
cells. However, there is one notable exception: in dealing with spherical
fibrations :S() -! X, we will not always require the total space S()
to be cofibrant. The reason for this is that we will be forming base
changes S()|Y -! Y along maps Y -! X (here S()|Y = Y x XS() is
one of the notations we will be using for the fiber product); cofibrancy
of total spaces is not usually invariant under base changes.
Fibrations. Assume B is a cofibrant space. We let Top =B denote
the category of spaces over B: an object consists of a map Y ! B,
where Y is an object of Top. A morphism Y ! Z is a map covering
POINCARE IMMERSIONS 5
the identity map of B. Quillen [Qu , II.2.8] proved that Top =B has
the structure of a model category in which the weak equivalences and
fibrations are defined by applying the forgetful functor to Top. A
fibration E ! B of spaces is the same thing as fibrant and cofibrant
object of Top =B. Given fibrant and cofibrant objects E; E0 2 Top =B,
there exists a fiberwise homotopy equivalence E ' E0 if and only if
E and E0 descend to isomorphic objects of the homotopy category of
Top =B (here we are applying a well-known fact about model categories
which says that a pair of fibrant and cofibrant objects are homotopy
equivalent if and only are related by a chain of weak equivalences).
Incidentally, we do not assert (since we do not know) that the
fiberwise suspension B E 2 Top =B of a fibrant and cofibrant ob-
ject is again fibrant. It is however cofibrant. Given fibrant objects
E; E0 2 Top =B, their fiberwise join E *B E0 2 Top =B is the (cofi-
brant) object given by the double mapping cylinder of the diagram
Ec (Ex B E0)c ! (E0)c (again, we do not assert that E *B E0 is
fibrant).
If B is a connected space, define Sph B to be the set generated by
objects E -! B whose homotopy fiber is spherical, modulo the equiva-
lence relation generated by weak equivalence in Top =B and fiberwise
suspension. Fiberwise join equips Sph B with the structure of an abelian
monoid. If additionally B is homotopy finite, then a result of Stasheff
[St] implies Sph B is an abelian group.
Thom spaces. Let X be a space equipped with a spherical fibra-
tion :S() ! X. The Thom space X is the mapping cone of the
composite S()ci~ S() -! X, where S()c denotes the cofibrant ap-
proximation of S(). Then X is cofibrant in the sense of based spaces.
Poincare spaces. A Poincare space X of dimension n is a pair (X; @X)
such that X and @X are homotopy finite, @X ! X is a cofibration,
and X satisfies Poincare duality:
o there exist a local system of abelian groups L of rank one defined
on X and a fundamental class [X] 2 Hn(X; @X; L) such that the
cap product homomorphisms
\[X] :H*(X; M) -! Hn- *(X; @X; L M)
and
\[@X] :H*(@X; N) -! Hn- *- 1(@X; L|@X N)
are isomorphisms, where [@X] 2 Hn- 1(@X; L|@X) is the image of
[X] under the connecting homomorphism in the homology exact
6 JOHN R. KLEIN
sequence of the pair (X; @X), and M (N) is any local system on
X (resp. on @X) (compare [Wa2 ]).
A Spivak fibration :S() -! X is an oriented spherical fibration
which comes equipped with a based map ff :Sn+ j -! X =(@X)|@X
(where, say, the fiber of is (j- 1)-spherical) such that the cap product
U \ ff*([Sn+ j])
is a fundamental class for X. Here U 2 Hj() denotes a Thom class
of , where Hj() is the j-th singular cohomology group of the pair
(X ; S()c) with coefficients taken in the local system L. The stable
fiber homotopy type of is unique, more precisely, two such pairs (; ff)
and (0; ff0) can be related by a stable fiber homotopy equivalence ' 0
whose induced map on Thom spaces transfers ff onto ff0up to homotopy
(see [Wa2 , 3.4-5] and [Br2 , I.4.19]; alternatively we will recover the
existence and uniqueness of the Spivak fibration in Corollary C).
Embeddings. Let P n and Mn be connected Poincare spaces of di-
mension n. An embedding of P in M is a commutative cocartesian
square of homotopy finite spaces
@P --- ! C
? ?
incl.?y ?yg
P --- ! M
e
together with a factorization of the inclusion @M -! C -! M, such that
the composite
Hn(M; @M) -! Hn(M ; C) ~=Hn(P; @P )
is an isomorphism and the image of a fundamental class [M] under the
composite
Hn(M; @M) -! Hn(M ; P q @M) ~=Hn(C ; @P q @M)
equips (C ; @P q @M) with the structure of a Poincare space
The space C is called the complement, and e :P -! M is the un-
derlying map of the embedding. In this case, we say that the map
e :P -! M embeds. Sometimes we will have occasion to refer to the
embedding by means of its underlying map.
POINCARE IMMERSIONS 7
The decompression of an embedding e :P -! M with complement C
is the embedding P xI -! Mx I defined by the diagram
@(P xI) --- ! W
? ?
? ?
y y
P xI --- ! Mx I
ex idI
where W := M C, and the factorization @(P xI) -! W -! Mx I is
evident (since P @P ~= @(P xI)).
Let K be a connected homotopy finite space, Xn be a Poincare space
and f :K -! X a map. One says that f embedded thickens (nee PD
embeds [Kl3 ]) if there exists a map A -! K (`normal data') such that
(K ; A) is an n-dimensional Poincare pair, and such that with respect
to this choice, the composite
K -~! K -f!X
embeds (in the above sense). It was shown in [Kl3 , 3.1] that the com-
posite
f j
K -! X -! Xx D
embedded thickens if j is sufficiently large.
3. Immersions
As in the introduction, a map f :P n! Mn of connected n-dimensional
Poincare spaces immerses if fx idDj :P xDj ! Mx Dj embeds for
some integer j 0; such an embedding is called an immersion of f. We
now restate Theorem A.
Theorem 3.1. Let f :P n-! Mn be a map. Then
1. An immersion of f gives rise to a stable fiber homotopy equiva-
lence OE :P ' f*M .
2. Conversely, given a stable fiber homotopy equivalence OE :P 'f*M ,
there is an immersion of f which induces it up to stable fiber ho-
motopy.
Proof of 3.1(1).Let (; ff) be a Spivak fibration for M as guaranteed by
[Sp , Th. A] or by Corollary C, where say, :S() ! M and ff :Sn+j !
M =(@M)|@M ).
8 JOHN R. KLEIN
Suppose first that f embeds. Let
@P --- ! C
? ?
i?y ?yj
P --- ! M
f
be an embedding diagram. Consider the composite fi, defined by
* f* (fOi)*
Sn+j -ff!M =(@M)|@M -!M =Cj ' P =(@P )
*
(to avoid notational clutter we are*fudging slightly in writing M =Cj
for the mapping cone of the map Cj ! M ).
If Uf* denotes a Thom class of f*, then by the naturality of the
Thom isomorphism the cap product Uf* \ fi*([Sn+ j]) is a fundamental
class for P . We infer that f* is a Spivak normal fibration for P .
By the uniqueness of the Spivak fibration ([Wa2 , 3.5], [Br2 , I.4.19] or
Corollary C), there is a stable fiber homotopy equivalence P 'f*.
The establishes 3.1(1) for embeddings of f.
The general case of 3.1(1) follows by replacing the above embedding_
of f by an embedding of fx idDj :P xDj -! Mx Dj. |__|
We next give a quick proof of 3.1(2) in the special case when the
underlying space of P also has the structure of a closed Poincare space.
The argument for the general case is in x5.
Proof of 3.1(2) when P is a closed Poincare space. Assume P nhas the
structure of a p-dimensional closed Poincare space (p n). To distin-
guish the two Poincare duality structures, we let V pdenote the effect of
forgetting the boundary of P . Let V be a Spivak fibration of V . Then
the map @P ! P = V has an (n- p- 1)-spherical homotopy fiber. Let
@ be the result of making @P ! P into a fibration. Then P + @ = V
in Sph P. By hypothesis f*X ' P , so
V = f*X + @ :
in Sph P.
f
Now choose an embedded thickening of the composite V -! X
Xx Dj; this can be achieved when j is chosen sufficiently large (see [Kl3 ,
3.1]). Then V = f*X + in Sph P, where :A -! V is the normal data
of the embedded thickening. We infer that f*X +@ = f*X +, whence
@ = in Sph P. Consequently, the embedded thickening amounts to an__
embedding of f :P xDj -! Xx Dj. |__|
POINCARE IMMERSIONS 9
4. First Proof of Theorem B
The argument which proves the general case of 3.1(2) employs Poincare
thickenings. Recall the statement of Theorem B:
Theorem 4.1. Given a connected homotopy finite space K, there is
one and only one Spivak thickening of K up to stable concordance.
As stated in the introduction, we shall provide two proofs of 4.1.
The first proof uses manifolds and appears in this section. The second
proof is homotopy theoretic and will appear in x7.
First proof of 4.1.The existence of a Spivak thickening of K is given
as follows: K may replaced up to homotopy by a finite k-dimensional
polyhedron embedded in some Euclidean space. The boundary of the
regular neighborhood defines a Poincare boundary A, the retraction of
the regular neighborhood defines an (n- k- 1)-connected map A -! K,
and the Thom-Pontryagin collapse defines a map Sn -! K =A repre-
senting a fundamental class.
To prove uniqueness we introduce an auxiliary notion: an embedded
smooth thickening of K in Sn is a compact codimension zero smooth
submanifold Nn Sn and a homotopy equivalence h :K -'!N.1 Two
embedded smooth thickenings (N0; h0) and (N1; h1) of K are concordant
if there exist an embedded h-cobordism W Snx I from N0 Snx 0
to N1 Snx 1 and a homotopy equivalence H :Kx I -'!W extending
h0 and h1. Regular neighborhood theory and transversality show that
there is one and only one embedded smooth thickening of K in Sn
up to concordance when n is large with respect to k := the minimum
dimension of those CW complexes in the homotopy type of K.
Let (A; ff) and (A0; ff0) be Spivak n-thickenings of K. By a con-
struction due to Browder [Br1 ], the pair (A; ff) defines an embedded
(Poincare) thickening in Sn+1 :
K A --- ! C
? ?
? ?
y y
K --- ! Sn+1
where the complement C is the mapping cone of ff :Sn -! K =A and
the map K A -! C is given by collapsing the inclusion of Kx 0 in
K A = Kx 0 [ Ax [0; 1] [ Kx 1 to a point. In what follows, we let
____________
1This notion is sometimes called an embedding up to homotopy type and is sim*
*ilar
to the definition of Wall [Wa1 ], the essential difference being that Wall requ*
*ires h
to be a simple homotopy equivalence.
10 JOHN R. KLEIN
n+1 denote the double mapping cylinder of K- K A -! C. This
gives a triad of spaces (n+1; K; C; K A).
If n 5, the Browder-Casson-Sullivan-Wall theorem [Wa3 , 12.1]
implies that there exist an embedded smooth thickening (Nn ; h) of K
in Sn+1 , with complement W and a homotopy equivalence of triads
(Sn+1 ; N; W ; @N) -'!(; K; C; K A) :
This in turn defines a concordance between the Spivak (n+ 1)-thickenings
(K A; ff) and (@N; a), where a :Sn -! N=@N denotes the Thom-
Pontryagin collapse map of N Sn.
Repeating the above procedure with (A; ff) replaced by (A0; ff0), we
obtain another embedded smooth thickening (N0; h0) of K in Sn+1 and
a homotopy equivalence of triads
(Sn+1 ; N0; W 0; @N0) -'!(0; K; C0; K A0) :
Therefore we have a concordance between the Spivak (n+ 1)-thickenings
(K A0; ff0) and (@N0; a0), where a0 is the Thom-Pontryagin collapse of
N0 Sn+ 1.
Assuming n is large, and using the stable uniqueness up to concor-
dance of embedded smooth thickenings, we see that (N; h) and (N0; h0)
are concordant. This concordance induces, in an evident way, a concor-
dance between (@N; a) and (@N0; a0). It follows that (K A; ff) and __
(K A0; ff0) are concordant. |__|
5. Completion of the proof of Theorem A
Using Theorem B, we give the proof of 3.1(2).
Construction 5.1. Let K be a connected space, :S() -! K a spher-
ical fibration and E -! K a map of spaces. Let K E be the double
mapping cylinder of
S()c- (S()|E)c -! E
(this is a kind of twisted fiberwise suspension of E -! K). There is an
evident map K E -! K. Furthermore, there is a cocartesian square
K E --- ! E|E
? ?
? ?
y y
K --- ! K
which shows that the mapping cone of K E -! K coincides up to
homotopy with the mapping cone of E|E -!K . If fflj: Kx Sj-1 -! K
POINCARE IMMERSIONS 11
denotes the trivial (j- 1)-spherical fibration, then ffljKE is identified
with the j-fold iterated fiberwise suspension jKE.
If E -! K is an n-thickening of K and if is (j- 1)-spherical, then
K E -! K is an (n+ j)-thickening of K.
Proof of 3.1(2).Set = M . Given a map of spaces g :Y -! M,*to
avoid notational clutter we write Y for the Thom space Y g . By as-
sumption, f*M ' P , so we have an identification P P = P . Using the
construction 5.1, we have a thickening P @P -! P which has P =(@P )
as its mapping cone (up to homotopy). Hence, P @P -! P is a Spivak
thickening.
f
On the other hand, we also know that the composite P -! M
Mx Dj embedded thickens for large j (by [Kl3 , 3.1]). Choose such
an embedded thickening, and let A -! P be its normal data. Then
P A -! P is also a Spivak thickening of P . Reason: let W denote the
complement of the embedded thickening. Then there is a homotopy
equivalence P =A ' M =W since A; P; W and M sit in a cocarte-
sian square (here and in what follows, in order to avoid notational
clutter, we are fudging slightly in writing P =A for the mapping cone
of A ! P ). Let S` ! M =(@M) represent a fundamental class of
(M ; M @M). Follow this up with the map M =(@M) ! M =W '
P =A to obtain a map S` ! P =A representing a fundamental class
of (P ; P A). This shows that P A ! P is a Spivak thickening (again,
we are using the identification P=P A ' P =A ). Consequently, by
4.1, P A ! P and P @P ! P are stably concordant.
Let o represent an inverse for P in Sph P (cf. x2). Then, stabiliz-
ing further with respect to o , it follows that +Po A and +Po @P are
isomorphic in the stable homotopy category of spaces over P . Equiv-
alently, there exist non-negative integers j and ` such that jPA and
`P@P are isomorphic in the homotopy category of spaces over P . But
jPA ! P amounts to the normal data for the j-fold decompression
of the given embedded thickening of f :P ! M with normal data
f `
A ! P . We infer that P -! M Mx D embedded thickens with
normal data `P@P ! P . But the pair (P ; `P@P ) is identified with
(P xD`; @(P xD`)), so what we have is an embedding whose underly-
ing map is fx idD` :P xD` -! Mx D`. We conclude that f :P -! M __
immerses. |__|
12 JOHN R. KLEIN
6. Equivariant duality
We develop duality for spaces equipped with the action of a suitable
kind of topological group G with coefficients in the so-called naive G-
spectra. The reader should not confuse our theory with the equivariant
duality appearing in [B-M-S-M , Chap. 3], rather, our theory reduces
to Ranicki duality [Ra , x3] in the case when G is discrete.
Let G. be a simplicial group. Then G = |G.| (i.e., the realiza-
tion of its underlying simplicial set) is a topological group object in
Top. Let RG (*) denote the category of based (left) G-spaces (where
`space' in this context means an object of Top ) and equivariant based
maps. Declare a morphism of RG (*) to be a weak equivalence (resp.
fibration) if and only if it is one after applying the forgetful functor
RG (*) ! Top . A morphism is a cofibration if and only if it satisfies
the left lifting property with respect to the fibrations which are also
weak equivalences. Then
Proposition 6.1. With respect to the above, RG (*) is a Quillen model
category.
For a proof, see [S-V ].
Since we will be for the most part be using cofibrant objects, we
indicate what these are. The equivariant j-cell is the unbased G-space
Djx G where G-acts by left translation. The equivariant (j- 1)-sphere
is similarly Sj- 1xG. If Z 2 RG (*) is an object and g :Sj-1 xG -! Z is
a G-map, then we may form the amalgamation Z [g(Djx G), which is
again an object of RG (*). Call an object of RG (*) cofibrant if it is built
up from a point by iterated equivariant cell-attachments, or if it is a
retract thereof (in particular, a cofibrant object is free away from the
basepoint). Any object Z comes equipped with a functorial cofibrant
approximation Zc ~iZ.
An object of RG (*) is finite if it is isomorphic to an object built
up from a point by a finite number of equivariant cell attachments. It
is homotopy finite if it admits a chain of weak equivalences to a finite
object (i.e., they are isomorphic in the homotopy category).
The reduced suspension Y of a cofibrant object Y (formed as in the
category of based spaces) is again a cofibrant object; the action of G on
Y is defined by letting G act trivially on the suspension coordinate.
For cofibrant objects Y; Z we define [Y; Z]G to be equivariant homotopy
classes of equivariant maps.
Given cofibrant objects Y; Z 2 RG (*), the smash product Y ^ Z
(formed as in the category of based spaces) inherits the diagonal action
of G. Let Y ^G Z denote the effect of taking orbits with respect to the
action of G. Then Y ^G Z is an (unequivariant) based space.
POINCARE IMMERSIONS 13
A spectrum with G-action E consists of cofibrant objects Ei 2 R*(G)
for integers i 0 and morphisms Ei -!Ei+ 1. (We remark that these
are equivariant spectra in the naive sense, and are not to be confused
with the more elaborate notion used by May et. al. [B-M-S-M ].)
Given a cofibrant object Y 2 RG (*), the equivariant cohomology of
Y with coefficients in E is the graded abelian group
E*G(Y ) := colim [jY; Ej+ *]G :
j!1
Similarly, equivariant homology of Y with coefficients in E is the graded
abelian group
EG*(Y ) := colim [Sj+ *; Ej ^G Y ] :
j!1
Definition 6.2. Assume that Y; Z 2 RG (*) are cofibrant and homo-
topy finite objects. Let d :Sn -! Y ^G Z be a map. Then d is an
equivariant duality with respect to E if the correspondence
f n+ j (f^GidZ)Od
jY -! Ej+ * 7- ! S -- - - - - !Ej+ *^G Z
induces an isomorphism of abelian groups
~= G
E*G(Y ) -! E*- n(Y ) :
If equivariant duality holds with respect to all spectra with G-action
E, then d is said to be an equivariant duality map.
If d :Sn -! Y ^G Z is an equivariant duality map, then so is its
suspension d :Sn+1 = Sn -! (Y ^G Z) = Y ^G Z.
Example 6.3. Let i n be a nonnegative integer. Set
SnG := Sn ^ (G+ )
where G+ means G with the disjoint union of a basepoint, and where
G acts on SnG by left translation. The diagonal map G+ -! (Gx G)+
smashed with Sn = Si ^ Sn- idefines an equivariant map SnG-! SiG^
Sn-Gi. Taking G-orbits gives a map Sn -! SiG^G Sn-Gi. It is straightfor-
ward to check that the latter is an equivariant duality map.
To determine whether a map is an equivariant duality, it is suffi-
cient to verify duality with respect to a suitable version of equivariant
singular homology. We now explain what this means.
Let ss := ss0(G) be the group of path components of G. Then G -! ss
is a homomorphism. Let G0 denote the kernel. The diagonal inclusion
G0 G0x G0 defines a map
Y ^G0 Z -! YG0 ^ ZG0 ;
14 JOHN R. KLEIN
where YG0 denotes the based ss-space given by taking the orbits of G0
acting on Y . This map is ss-equivariant, so we may take ss-orbits to
obtain a canonical map Y ^G Z -! YG0 ^ssZG0.
Proposition 6.4. A map d :Sn -! Y ^G Z is an equivariant duality
if and only if taking the slant product with the fundamental class of Sn
with respect to the composite
d0: Sn -d!Y ^G Z -! YG0 ^ssZG0
yields an isomorphism
~=
=d0*([Sn]) :H*(YG0) -! Hn-*(ZG0) ;
where H* denotes reduced singular homology.
In other words, the map d0: Sn -! YG0 ^ssZG0 is an ss-equivariant
duality map with respect to h(Z[ss]), the Eilenberg-MacLane spectrum
with ss-action associated to integral group ring of ss (cf. Ranicki [Ra ,
x3]). In the special case when ss is trivial, this just means that d0 is a
Spanier-Whitehead duality map.
Proof of 6.4.We shall only provide a sketch of the argument (details
will appear elsewhere). First we introduce some auxiliary definitions.
Define the homotopy groups of a spectrum with G-action E to be the
graded Z[ss]-module ss*(E) := colimj[Sj+ *; Ej]. A morphism of spectra
with G-action D -! E is a collection of morphisms Di -!Ei compatible
with the structure maps of D and E. A morphism is a weak equivalence
if it induces an isomorphism on homotopy groups. If D -! E is a weak
equivalence, it induces isomorphisms DG*(Y ) ~= EG*(Y ) and D*G(Y ) ~=
E*G(Y ) for all cofibrant objects Y 2 RG (*).
For any Z[ss]-module M, let h(M; j) be its associated Eilenberg-
Mac Lane spectrum in degree j. Then h(M; j) is a spectrum with
ss-action. The homomorphism G ! ss provides an action of G on
the spaces h(M; j)k, however we have to free up the action to en-
sure that h(M; j) is a spectrum with G-action. This can be accom-
plished by replacing the k-th ss-space h(M; j)k by the cofibrant G-space
EGx h(M; j)k (with G acting diagonally). Assume this has been done.
With respect to the hypotheses of the proposition, the ss-cellular
chain slant product map
=d0*([Sn]) :C*(YG0) -! Cn-*(ZG0)
is a ss-equivariant chain equivalence. Equivalently, duality is satisfied
with respect to the spectrum with G-action h(Z[ss]; j) for all j.
POINCARE IMMERSIONS 15
Tensoring the above map with any Z[ss]-module M, and using the
finiteness of Y and Z, it follows that the slant product map
=d0*([Sn]) :C*(YG0; M) -! Cn-*(ZG0; M)
(where chains now have coefficients in M) is also an equivariant chain
homotopy equivalence (compare [Wa2 , 1.1]). Equivalently, G-equivariant
duality holds with respect to h(M; j) for all j and all Z[ss]-modules M.
A spectrum with G-action E is said to be Eilenberg-Mac Lane of
type (M; j) for a Z[ss]-module M and an integer j if ss*(E) is trivial
if * 6= j and there is an isomorphism of Z[ss]-modules ssj(E) ~= M.
By an obstruction theory argument which we omit, one shows that
an Eilenberg-MacLane spectrum of type (M; j) admits a weak equiv-
alence E -! h(M; j). This gives equivariant duality for spectra with
G-action which are Eilenberg-MacLane of type (M; j). By the proce-
dure of killing homotopy groups, any spectrum with G-action E ad-
mits a Postnikov system, i.e., a sequence . . .! PkE ! Pk- 1E ! . . .
for k 2 Z and compatible maps E -! PkE such that the homotopy
groups of E vanish in degrees > k and E -! PkE induces a isomor-
phism on homotopy groups in degrees k. Then the homotopy fiber
of PkE -! Pk- 1E is Eilenberg-Mac Lane of type (ssk(E); k). Call E
bounded below if ss*(E) = 0 whenever * is sufficiently small. By induc-
tion up the stages of the Postnikov system using the five lemma, we
obtain equivariant duality for bounded below spectra with G-action.
Any spectrum with G-action E admits a chain of weak equivalences
to a spectrum with G-action of the form colimjE(j)in which E(j)is
bounded below (we can define E(j)as the (-j)-connected cover of E,
i.e., the homotopy fiber of the map E -! P-jE). Equivariant duality
__
for E then follows from equivariant duality for the E(j). |__|
The fundamental result about the existence and uniqueness of equi-
variant duality maps is
Theorem 6.5. (Existence). If Y 2 RG (*) is cofibrant and homotopy
finite, then there exist an integer n 0, a cofibrant and homotopy finite
object Z 2 RG (*) and an equivariant duality map d :Sn -! Y ^G Z.
(Uniqueness). Assume Y comes equipped with two equivariant duality
maps d :Sn -! Y ^G Z and d0: Sn -! Y ^G Z0. Then there exist an
integer j 0 and an equivariant weak equivalence h :jZ -! jZ0 such
that (idY ^G h) O jd is homotopic to jd0. Moreover, if j is large then
h is unique up to equivariant homotopy.
Proof. (Existence). Ranicki [Ra , 3.5] provides a proof of existence with
G = ss is discrete and Y is a finite object (he uses induction on the
16 JOHN R. KLEIN
equivariant cell decomposition; the inductive step uses the duality map
provided in example 6.3). An inspection his proof adapts to the case
of a general G when Y is finite. The case of a general homotopy finite
Y follows by noting that a cofibrant homotopy finite object Y admits
a weak equivalence Y 0-~!Y where Y 0is built up from a finite number
of equivariant cells.
(Uniqueness). The proof will be based on equivariant obstruction the-
ory. A morphism A -! B of RG (*) will be called r-connected if it is so
as a map of spaces. For a cofibrant object A, we write dim A k if A is
built up from a point by attaching equivariant cells of dimension k.
We will require the following fact, whose proof is given by induction on
the number of equivariant cells of A: given an r-connected morphism
f :A -! B and a map g :Z -! B with Z cofibrant and dim Z r, there
exists a map "g:Z -! A such that f O "gis equivariantly homotopic to
g. Moreover, "gis unique up to homotopy if dim Z r- 1.
Assume that d :Sn -! Y ^G Z and d0:Sn -! Y ^G Z0 are equivariant
duality maps. We may assume without loss in generality that Y; Z and
Z0 are finite objects.
Let F (Y; SnG) denote the function space of based (unequivariant)
maps from Y to SnG (recall SnG := Sn ^ (G+ )), topologized using the
compactly generated compact open topology. Then F (Y; SnG) is a based
G-space where G acts on functions by conjugation. The fixed point
space F (Y; SnG)G is the space of equivariant functions. This eqivariant
function space admits a left G-action using the right translation action
on SnG. The correspondence given by f 7! (f ^G id) O d defines a map
ffiZ :F (Y; SnG)G -! F (Sn; Sn ^ Z)
and if we give F (Sn; Sn ^ Z) the action induced by the action of G
on Z and the trivial action on Sn, it follows that ffiZ is G-equivariant.
Equivariant duality implies that ffiZ induces an isomorphism on homo-
topy groups in a certain range (the range where the function spaces
homotopically coincide with their associated function spaces of stable
maps). Specifically, if dim Y k, and Z is r-connected, then ffiZ is
ae := min (2n- k- 2; 2r+ n)-connected. Replacing Z be jZ and using
the duality map jd :Sn+ j -! Y ^G jZ, shows similarly that the
corresponding map
ffijZ:F (Y; Sn+jG)G -! F (Sn+j ; Sn+j ^ jZ)
is (ae+ 2j)-connected.
On the other hand, there is an evident map
jZ:jZ -! F (Sn+j ; Sn+j ^ jZ) :
POINCARE IMMERSIONS 17
If dim Z `, it follows that dim jZ `+ j, and eventually, ae+ 2j ex-
ceeds `+ j as j-increases. By equivariant obstruction theory we obtain
a morphism q :jZ -! F (Y; Sn+jG)G such that ffijZO q is equivariantly
homotopic to jZ. Furthermore, q is unique up to equivariant homotopy.
We obtain therefore a composite map
q n+j G ffijZ0 n+j n+j j 0
jZ -! F (Y; SG ) - ! F (S ; S ^ Z ) :
If s denotes the connectvity of Z0 then the map
jZ0:jZ0 -!F (Sn+j ; Sn+j ^ jZ0)
is (2s+ 2j+ 1)-connected. By requiring j to be large, we can guaran-
tee the inequality `+ j < 2s+ 2j+ 1. Again by equivariant obstruction
theory, we see that there exists a morphism h :jZ -! jZ0 such that
jZ0O h coincides with ffijZ0O q up to equivariant homotopy, and h is
unique up to equivariant homotopy. It is automatic that h is a weak __
equivalence, and that (idY ^G h) O d is homotopic to d0. |__|
7. Second proof of 4.1
Let K be a connected space. Choose a basepoint for K. Let S.K
be the simplicial total singular complex. Let G. denote the Kan loop
group of S.K, and set G = |G.| (with compactly generated topology).
Then there is a functorial identification BG ' K, where BG is the
classifying space of G (i.e., the bar construction of G.). Using this
identification we can assume K = BG without any loss in generality.
If we set K" := EG, then have a fibration K" -! K. For any map of
spaces A -! K, let A" -!K" be the associated pullback along K" -! K.
Then A" Ax K" inherits an (unbased) G-action (here G acts on the
product by acting trivially on the first coordinate and by its usual
action on EG on the second coordinate). In the homotopy category
of spaces over K, the Borel construction K"x GA" -! K is canonically
isomorphic to A -! K
For technical reasons, we replace A" by its cofibrant approximation
(A")ci~ A" in RG (*). By abuse of notation, let K"=A" be the mapping
cone of the composite (A")ci~ A" -!K". Then K"=A" is a cofibrant object
of RG (*). Then the reduced diagonal map K"=A" ! K"+ ^ K"=A" is
equivariant and therefore induces a map of G-orbits
:(K"=A")G -! K"+ ^G K"=A"
which is easily seen to be a weak equivalence of based spaces. Up to
homotopy, (K"=A")G is just K =A, the mapping cone of A -! K.
18 JOHN R. KLEIN
Second proof of 4.1.(Existence). Let K be a connected homotopy fi-
nite space. Then K"+ is a homotopy finite object of RG (*). By the
existence part of 6.5, there exist an integer n 0, a homotopy finite
object Z and an equivariant duality map
d :Sn -! K"+ ^G Z :
Let A = "Kx GZ be the Borel construction. The we have a map A -! K,
and the mapping cone of the latter is identified with K"+ ^G Z up to
homotopy. If ff denotes the map d :Sn+ 1! "K+^G Z ' K =A, then
(A; ff) is a Spivak (n+ 1)-thickening of K.
(Uniqueness). If (A; ff) and (A; ff0) denote a Spivak thickenings of K,
by decompressing a sufficient number of times, we can assume that
they are both Spivak n-thickenings.
Poincare duality for (K ; A) implies that the the composite
d :Sn -ff!K=A ' "K=A" -! K"+ ^G K"=A"
satisfies equivariant duality with respect to h(Z[ss1(K)]). Application
of 6.4 shows d to be an equivariant duality map. Similarly, so is the
composite
0 0 0 0
d0:Sn -ff!K =A ' "K=A" -! "K+^G K"=A" :
By the uniqueness part of 6.5, there exist an integer j 0 and an equi-
variant weak equivalence h :jK"=A" -~!jK"=A"0such that (idK"+^G
h) O jd is homotopic to d0.
The Borel construction "Kx GjK"=A" -!K coincides with the (j+ 1)-
fold fiberwise suspension j+K1A -! K in the homotopy category of
spaces over K. Consequently, h induces an isomorphism in the homo-
topy category of spaces over K between j+K1A -! K and j+K1A0-! K.
Moreover, the fact that (idK"+^G h) O jd is homotopic to d0 implies
that this isomorphism transfers j+ 1ff to j+1ff0. Hence (A; ff) and __
(A; ff0) are stably concordant. |__|
8. Proof of the Addendum to Theorem A
Let (P ; ff) and (M ; fi) represent*the Spivak fibrations for P*and M,
where say, ff :Sn+ k -! P P =(@P )i P and fi :Sn+ k -! MM =(@M)j M
(here Let i :@P ! P and j :@M ! M denote the inclusions). Let
OE :P ' f*M be a fiber homotopy equivalence.
As in the proof of Theorem A(1) (i.e., 3.1(1)), an immersion of f
induces a stable collapse map
fi j* f* (fOi)*
fi0: Sn+ k-! MM =(@M) M -! P M =(@P ) M
POINCARE IMMERSIONS 19
The fiber homotopy equivalence OE0:P ' f*M provided in Theorem
A(1) is given by applying the uniqueness result for the Spivak fibration
with respect to the pairs (f*M ; fi0) and (P ; ff). Using OE, we obtain an
identification of (f*M ; fi0) with (P ; ff0), where ff0 is obtained compos-
* (fOi)* i*
ing the identification P f M =(@P ) M ' P P =(@P ) P (defined by
OE) with fi0.
It it were true that ff and ff0 coincide up to homotopy, then the
uniqueness of the Spivak fibration would imply that OE and OE0 coincide
up to fiber homotopy. Hence, it will suffice to modify the immersion in
such a way that ff0 coincides with ff up to homotopy.
Let Aut (P ) denote the group of (stable)*fiber homotopy auto- *
morphisms of P , and let ssn+ k(P P =(@P )i P )" ssn+ k(P P =(@P )i P )
denote the subset of those elements fl satisfying the condition that
U \ fl*([Sn+ k]) is a fundamental class. Then, by [Wa2 , 3.5], there is an
isomorphism of sets
~= i* "
Aut (P ) -! ssn+ k(P P =(@P ) P )
defined by the*assignment 7! 0 O ff, where 0 is the self-map of
P P =(@P )i P which induces. It follows that there exists a unique
up to fiber homotopy such that ff0' 0O ff.
Let o represent an inverse for P in Sph P. Then the fiber homo-
topy automorphism h := + ido of the trivial fibration induces an
automorphism of the object @(P xDj) ~= P+Po @P in the homotopy
*
category of spaces over P . The self-map of jP P =(@P )i P induced
by Thomifying h coincides with j 0 up to homotopy.
Let
@(P xDj) --- ! W
? ?
? ?
y y
P xDj -- - - ! Xx Dj
fx idDj
be the embedding diagram for the given immersion. Change the map
@(P xDj) ! W by prefixing it with h. The resulting diagram will
only homotopy commute, but this can be repaired by replacing W by
a suitable mapping cylinder. The result of this procedure yields a new
immersion whose stable collapse map coincides with ff up to homotopy.
References
[B-M-S-M] Bruner, R. R., May, J. P., Steinberger, M., McClure, J. E.: Equivaria*
*nt
stable homotopy theory. (LNM, Vol. 1213). Springer 1986
[Br1] Browder, W.: Embedding smooth manifolds. Proc. I.C.M. , 712-719
(1968). Moscow 1966
20 JOHN R. KLEIN
[Br2] Browder, W.: Surgery on Simply Connected Manifolds. (Ergebnisse
Math., Vol. 65). Springer 1972
[Hi] Hirsch, M.: Immersions of manifolds. Trans. AMS 93, 242-276 (1959)
[Kl1] Klein, J. R.: On two results about fibrations. Manuscripta Math. 92,
77-86 (1997)
[Kl2] Klein, J. R.: Poincare duality spaces. To appear in the Wall Birthday
Volume on Surgery Theory
[Kl3] Klein, J. R.: Poincare embeddings and fiberwise homotopy theory. To
appear in Topology
[Qu] Quillen, D.: Homotopical Algebra. (LNM, Vol. 43). Springer 1967
[Ra] Ranicki, A.: The algebraic theory of surgery. II Applications to topo*
*l-
ogy. Proc. London Math. Soc. 40, 193-283 (1980)
[S-V] Schw"anzl, R., Vogt, R.: The categories of A1 - and E1 -monoids and
ring spaces as closed simplicial and topological model categories. Ar*
*ch.
Math. (Basel) 56, 405-411 (1991)
[Sp] Spivak, M.: Spaces satisfying Poincare duality. Topology 6, 77-101
(1967)
[St] Stasheff, J. D.: A classification theorem for fiber spaces. Topology *
*2,
239-246 (1963)
[Wa1] Wall, C. T. C.: Classification problems in differential topology_IV.
Thickenings. Topology 5, 73-94 (1966)
[Wa2] Wall, C. T. C.: Poincare complexes I. Ann. of Math. 86, 213-245 (1967)
[Wa3] Wall, C. T. C.: Surgery on Compact Manifolds. Academic Press 1970
Wayne State University, Detroit, MI
E-mail address: klein@math.wayne.edu