Higher Simple Homotopy Theory

Annals of Mathematics, 102 (1975), 101-137
Higher simple homotopy theory

By A. E. HATCHER*
1. Introduction
In this paper we globalize J. H. C. Whitehead's simple homotopy theory
[17] by constructing a homotopy functor Wh from polyhedra to simplicial
H-spaces, such that Whitehead's theory amounts to the calculation of
IT o Wh (K),the arc-components of Wh (K). "Higher simple homotopy theory"
is then concerned with the full homotopy type of Wh (K), for example, its
higher homotopy groups.
Recall from Whitehead's simple homotopy theory the basic geometric
operation of an elementary collapse, written L o ~ L 1' where L o and L 1 are
finite cell complexes such that L o is obtained from L 1 by attaching a ball
along a face in its boundary. The equivalence relation generated by elemen-
tary collapses is called simple homotopy equivalence, and the main theorem
is that a homotopy equivalence of finite complexes is simple if and only if a
single algebraically defined obstruction (the torsion), lying in an abelian group
which depends only on the fundamental group of the spaces involved,
vanishes.
Simple homotopy equivalences are not hard to find in nature. A useful
recognition criterion in the PL category, due to M. M. Cohen [7], is the
following: A PL map j: L o ~ L 1 is a simple homotopy equivalence if all the
point inversesj-1(*) are non-empty and contractible. Cohen called such maps
contractible mappings. For example, an elementary collapse L o ~ L 1 can be
realized by an evident contractible mapping. More recently, T. A. Chapman
[3] has vastly generalized Cohen's theorem to the CW category (with "con-
tractible" replaced by "cell-like"), thereby proving a conjecture of Whitehead
that homeomorphisms are simple.
One nice property of PL contractible mappings not shared by elementary
collapses is that they are closed under composition. Thus we can form the
category e whose objects are finite polyhedra (say, finite subpolyhedra of ROO
for definiteness) and whose morphisms are PL contractible mappings. We
have also the full subcategory eK of e whose objects are polyhedra homotopy
equivalent to the fixed polyhedron K, and the subcategory e(K) c e K whose
objects contain K as a deformation retract and whose morphisms restrict to
* Supported in part by NSF grant GP 34324X.
102 A.E.HATCHER
the identity on K. One of several equivalent definitions of Wh (K) is the

classifying space Be(K). This is (the geometric realization of) the simplicial
space whose k-simplices are the compositions L o ~ L 1 ~ • • • ~ L k in
e(K). The various (k - 1)-faces of such a k-simplex are obtained by deleting
an L i and, if 0 < i < k, composing fi with fi+l- The classifying spaces Be K
and Be are defined similarly. The arc-components of Be are, by Cohen's
theorem, exactly the simple homotopy types of finite polyhedra. Be K is a
union of components of Be, those containing polyhedra homotopy equivalent
to K. Also, 1C o Wh (K) is just the group called "Wh (K)" in [8], where White-
head's theorem is reformulated to say that 1C o Wh (K) is naturally isomorphic
to Wh 1 (1C 1K), the algebraic torsion group, quotient of K 1Z[1C 1K]. (See also
[10], [13], [15] for similar geometric definitions of Whitehead torsion.)
To breathe a little life into this categorical nonsense, we start by show-
ing that Be actually classifies something: fibrations in the PL category, that
is, PL maps which satisfy the covering homotopy property for polyhedra
(Serre fibrations). Thus e is the "structure group" for PL fibrations. Intui-
tively, the idea is that PL contractible mappings are "local" homotopy
equivalences, and the covering homotopy property is essentially a local
condition. An immediate corollary is that, over a connected base, the fibers
of a PL fibration all have the same simple homotopy type, not just the same
homotopy type, as one might expect.
More usually in topology one works with homotopy fibrations, meaning
maps of arbitrary spaces which are Serre fibrations, or equivalently, PL
maps which are only quasi-fibrations (satisfying the weak covering homotopy
property [9]). Homotopy fibrations with fibers homotopy equivalent to K are
classified by BG(K), where G(K) is the H-space of self-homotopy equivalences
of K. Passing from PL fibrations to homotopy fibrations induces a map of
classifying spaces, whose homotopy fiber turns out to be Wh (K) == Be(K):
Wh(K) ~ Be K ~ BG(K) .
Thus Wh (K) measures the global difference between PL contractible map-
pings and general homotopy equivalences. Curiously, Wh (K) has much more
structure than Be K or BG(K): it is a homotopy functor of K and, functorial-
ly, an infinite loopspace.
A deeper justification for our definition of Wh (K) is that we can use it
to prove a parametrized version of the PL h-cobordism theorem. Recall that
the h-cobordism theorem says in effect that for a given compact connected
PL manifold Mn, n ~ 5, the various h-cobordisms (W, M) are in one-to-one
correspondence with the components of Wh (M), via their torsions. In
HIGHER SIMPLE HOMOTOPY THEORY 103
particular, there exists a product structure (W, M) ~ (M x I, M) if and only
if W lies in the identity component of Wh (M). The parametrized version
deals with the comparison between different product structures on (Mx I, M).
Let g>(M) denote the (simplicial) space of PL homeomorphisms of Mx [fixed
on M (i.e., pseudo-isotopies). Then we construct a natural map g>(M) ~
Q Wh (M) which is k-connected whenever n == dim M is large with respect to
k (n ~ 3k + 8 will do). This is the main theorem of the paper.
An immediate consequence is that there is a "stable" dimension range
n ~ i where 1C i g>(Mn) depends only on the homotopy type of M. In particular,
the inclusion g>(M) ~ g>(M x I), f~ f X id I , induces an isomorphism on 1C j
if n ~ i. With results of Chapman, this leads to a neat reformation of higher

simple homotopy theory in terms of compact Hilbert cube manifolds. Such
manifolds have the form K x Q for K a finite polyhedron and Q the Hilbert
cube. K is determined only up to simple homotopy type, so we may as well
take it to be a PL manifold M. Then g>Top(Mx Q) ~ Uq g>(Mx Iq) ~ Q Wh (M),
the first equivalence, by [5]. For the composite equivalence g>Top(M x Q) ~
Q Wh(M) we can replace M by K, and we have a diagram
g>Top(K x Q) ~ Homeo (K x Q) ~ G(K x Q) ~ G(K x Q)/Homeo (K x Q)
1=
'lWh(K) - ~ 'lBe K -~
1= 1=
G(K) - - ~ Wh(K)
1
where the upper row is a fibration sequence by [6] and the lower row continues
(*). The map G(K x Q)/Homeo(K x Q) ~ Wh(K) is a homotopy equivalence
on identity components, and the precise situation with 1C o is covered by
Chapman's original proof of the topological invariance of Whitehead torsion:
A homotopy equivalence f: K ~ K' is simple if and only if f x id: K x Q ~
K' x Q is homotopic to a homeomorphism [4]. The equivalence of G(K x
Q)/Homeo(Kx Q) with a union of components of Wh(K) therefore globalizes
this result and gives the "topological invariance of higher torsions," viz.,
that the composition Homeo (K) ~ G(K) ~ Wh (K) is null-homotopic. I
should add that the equivalence of G(K x Q)/Homeo(K x Q) with a then-
hypothetical higher simple homotopy theory was predicted to me a couple of
years ago by F. S. Quinn.
A further application of the stable equivalence 9'(M) ~ Q Wh (M) is the
calculation 1
1 Assuming the first Postnikov invariant k 1 e H3(7r 1 M; 7r 2 M) of M vanishes-see the foot-
note in § 10.
104 A. E. HATCHER
the second and third equivalences by [2] and [11], respectively. (One could
also prove this directly.) Here Wh 2(1!1) is a certain quotient of K 2Z[1!1];
Wht(1!l; Z2 x 1!2) is described in Section 10.
The calculation of 1!1 Wh(M) shows that higher simple homotopy theory
is not a functor of fundamental groups alone, as is the classical theory.
Probably the best general statement about the dependence of Wh(K) on
K is that if K ~ K' is k-connected, k > 1, then the induced map Wh(K) ~
Wh (K') is (k - l)-connected. This is proved in Section 7 by a homotopy
excision argument. In fact, we show that Wh satisfies an excision property
formally analogous to excision in ordinary homotopy theory. Consequently
there is a stable simple homotopy theory s*(K) which is a generalized
homology theory. R. K. Lashof had previously constructed this theory in
terms of pseudo-isotopy spaces (using our stability result on g>(M)~g>(Mx I»).
Also, using results of Morlet and Chenciner he calculated the coefficient
groups: Si(SO) F::d 1!i_29'Diff(Dn) for n large. Since 9'PL(Dn) is contractible by
the Alexander trick, the effect of this is that s* measures the difference
between 9'Diff and 9'PL, in the stable range. It is known (see [16], [18]) that
the first non-vanishing Si(SO) occurs for i == 3.
By way of example we give in the last section of the paper an easy con-
struction of some non-trivial elements of 1!1 Wh(K) whenever 1!lK"* 0,
together with a way of injecting these into 1!n+1Wh(K X Tn), Tn the n-torus,
for any n ~ 1.
In a later paper we intend to clarify the relationship between Wh (K)
and higher algebraic K-theory by defining higher Whitehead groups
Whi(1!lK) and natural maps 1!i-l Wh(K) ~ Whi(1!lK) and KiZ[1!lK] ~
Wh i (1!lK). The best one could hope would be for these two maps to be
surjective (they are for i == 1, 2), but even this seems unlikely in general.
Similar remarks apply to a second family of functors Wht (1!lK; Z2 x 1!2K)
which extend the summand Wht(1!lK; Z2 x 7r 2K) of 1!lWh(K). And these two
invariants are just the beginning.
I am indebted to T. A. Chapman and F. S. Quinn for some stimulating
conversations about the material of this paper.
2. PL fibratiollS
We will be "Working in the PL category. All polyhedra will be sub-
polyhedra of Roo though we usually neglect to mention the specific embeddings
in Roo. For simplicity we will consider only finite polyhedra. The extension
to the locally finite case (with proper maps) is straightforward; the result
would be a "higher infinite simple homotopy theory," generalizing [13].
This section contains preliminary material on fibrations in the PL

category, the ma:.l results being the local characterization given in 2.1 and
its global form in 2.5.
We begin by describing a completely general way of decomposing any
PL map n: E ~ B into elemental blocks over the simplices of some triangula-
tion of B. Given a chain of PL maps
11 Ik
Lo~L1~···~Lk'
the iterated mapping cylinder M(fh ... , fk) is defined inductively to be the
ordinary mapping cylinder of the composition M(fh ... , !k-1) --> L k- 1 ~
L k , where the unmarked arrow is the obvious projection. Thus for k == 1 we
have the usual mapping cylinder, for k == 2 the mapping cylinder of M(f1)~
L 1 ~ L 2 , etc. (Note that mapping cylinders are well-defined PL objects by
9.5 of [7]; see also [1].) By an iterated mapping cylinder decomposition of
:r: E ~ B we mean: Over each simplex a of some triangulation of B, 7r- 1(a)
is given as an iterated mapping cylinder M(ft, ... , ff:), where k == dim a,
such that n: n- 1 (a) ~ a is identified with the standard projection M(ft, ... ,
ff:) ~ Ll k • Moreover, these structures are to be compatible when we pass
from a to simplices of aa.
To obtain an iterated mapping cylinder decomposition of an arbitrary
PL map n: E ~ B, choose triangulations of E and B (which we still call E
and B) and barycentric subdivisions E' and B' such that n: E ~ Band
:r: E' ~ B' are simplicial. Let bo, ... , bk be barycenters of simplices (30 >
... > (3k of B and let L i == n- 1(b i). Define PL mapsfi+l: L i ~ L i+1 by sending
a barycenter ei E L i of a simplex Ci of n- 1«(3i) to the barycenter ei+1 of Ci+l ==
Ci n n- «(3i+l)' and extending linearly. Then n- (b o • • • bk) is identified naturally
1 1
with M(fh ... , fk), and n: n- (b o • • • bk) ~ bo • • • bk is the projection M(fl' ... ,
1
fk) ~ Ll k. (To see this it suffices to consider the case that E and Bare
simplices and n is simplicial.)
PROPOSITION 2.1. Suppose n: E ~ B is a PL map which is a Sel~re fibra-
tion (briefly, a PL fibration). Then the maps fI in any iterated mapping
cylinder decomposition of n are contractible mappings, Le., all point-
inverses (fJ)-I( *) are non-empty and contractible. Conversely, if in some
iterated mapping cylinder decomposition of 7r all the maps fI are contract-
ible mappings, then n is a PL fibration.
Proof. We first show that the fI's must be contractible mappings if n
is a PL fibration. For this it suffices to choose E == M(f) for f: L o ~ L 1 • If
x E L u then M(f I f- 1(x») is a cone C on f-l(X), and we apply the covering
106 A. E. HATCHER
homotopy property in the diagram

M(f)
/ in
c~ [0, 1]
to the homotopy C ~ [0, 1] ~ [0, 1], h,(t) == min (1 - 8, t), where n- 1(t) ==
L o for 0 ~ t < 1 and n- 1(1) == L 1 • The effect is to produce a continuous family
of contractions of f-1(X) in each slice L o == n- 1(t), 0 ~ t < 1. By continuity,
the contraction must take place in a neighborhood of f-1(X) c L o for (s, t)
near (0, 1), so f-1(X) must actually be contractible in itself.
For the converse we start with an iterated mapping cylinder decomposi-
tion of n, with respect to some triangulation of B. Since being a PL fibration
is a local property with respect to B, it will suffice to show that n is a PL
fibration over the star, in the barycentric subdivision of B, of each vertex
v E B. (These stars can be enlarged slightly by isotopy so that their interiors
still cover B.) Thus we may assume B == star (v) == C(A), the cone on A ==
link (v), and, by induction on dim B, that on L == n- 1(A), n is a PL fibration.
Moreover, E == n- 1(B) has the structure of a mapping cylinder M(f), where
f: L ~ K == n- 1(v) is such that its restriction to each fiber n- 1(a) in L is a
contractible mapping (being one of the ft's in the given iterated mapping
cylinder decomposition of re or else the identity). Thus the proposition is
reduced to:
LEMMA 2.2. Given a PL fibration n: L ~ A and a fiber-preserving PL
contractible mapping (f, re): L--+Kx A, then the natural projection?C: M(f)--+
C(A) to the cone on A is a PL fibration.
Proof. Let F: M(f) --+ K x C(A) be the obvious map. We claim:
F has a homotopy inverse G for which there is a homotopy H u:
M(f) --+ M(f) from the identity to GF such that ?CHu == ?C and
(H)
such that H u is fixed on K == ?C- 1(v), where v is the cone point
of C(A).
Assuming this, we can proceed as follows. Given a lifting problem
M(f)
, / lif
X~C(A)
g,
let f): X --+ I be such that f)-1(0) == gol(V) and g,(x) *- v if 8 ~ f)(x) =I=- O.
Xxl
-[
XxO
Define g:: X - go-l(V) ~ C(A) - {v} to be gs if s ~ 0 and go if s ~ o. This lifts

to g;: X - go-l(V) ~ ~f(f) - n=-l(V) since n= I M(f) - n=-l(V) is a PL fibration.
Then define Ys: X ~ M(f) for s ~ 0 by Ys == Hs/oY;, where if necessary 0 is
replaced by a smaller function so that each arc gs(x), 0 ~ 8 ~ O(x), approaches
the constant arc go(x o) as x approaches X oE gol(V). (This assures that Ys is a
continuous extension of Yo.) We are left with the problem of lifting gs for
o ~ s ~ 1 with an initial position Yo which factors through the trivial fibra-
tion K x C(A) ~ C(A). This can certainly be done. D
Towards proving (H) we first reprove a result of M. M. Cohen [7]:
PROPOSITION 2.3. A contractible mapping f: L~K is a simple homotopy
equivalence.
Note that this then holds also for the restrictions! I!-l(K'), K' a sub-
polyhedron of K. Hence contractible mappings are closed under composition.
Proof of 2.3. Choose triangulations of K and L such that f is simplicial.
Since f is surjective, we can regard it as a kind of collapsing map: for each
simplex a of K, f collapses the subcomplex f-l(a) of L onto a. Thinking of
f in this way, we can factor it as L ~ L o ~ L 1 ~ • • • ~ K where fi
collapses the inverse image of the i-skeleton K i of K to Ki. We will show
that each fi is a simple homotopy equivalence, which implies that their com-
position f is also.
Let a be an i-simplex of K i == L~. Since f is simplicial we can choose an
i-cell of L'i-l mapped isomorphically onto a by fie Identify a with this i-cell.
Then since!i is a contractible mapping, there is a deformation retraction of
!i-l(a) onto a, rei !i-l(Ja) , preserving the "fibers" fi-l(X). Using these deforma-
tions for the various a in Ki we can pass continuously from L i - 1 to L~ ==
L i U fi-l(K i ) by a homotopy of the attaching maps of the cells inf-l(K - K i ),
thinking of L i - l as a CW complex (with PL attaching maps). The resulting
homotopy equivalence!:: Li-l~L~ is therefore simple. The collapsef:': L~-+
L i is clearly a homotopy equivalence; it is simple since the collapsing splits
into the disjoint collapses fi-l(a) -+ a each of which takes place over a con-
tractible, hence simply-connected, part of L i • The composition f:'f: is
108 A.E.HATCHER
homotopic to fi (in fact, by a homotopy which is arbitrarily small with respect

to projection on K), so fi is a simple homotopy equivalence. D
]{ ]{ ]{
Returning to assertion (H), consider first the unfibered case A == point.

Retracing the steps in the proof that f: L ~ K is a homotopy equivalence, we
can construct g: K ~ L and a homotopy h u : L ~ L from the identity to gf
such that fh u is arbitrarily close to f. Letting fh u approach f as we slide
down M(f) to KcM(f) we obtain G: Kx I~M(f) and a homotopy H u : M(f)~
M(f) from the identity to GF (preserving the projection 7f to I) such that
H u is the identity on Kc M(f).
In the fibered case when f: L~ K is replaced by (f, 7r): L~ K x A, begin
with g: K x A ~ Land h u : L ~ L as above. These two maps commute with
projection to A up to homotopy. Applying the covering homotopy property
to this homotopy, we can deform g and h u so that they commute with projec-
tion to A exactly. Then construct G and H u as in the unfibered case. D
As a consequence of 2.1, being a PL fibration is a purely local property;
i.e., given 7r: E ~ B, if every point of E has a neighborhood U such that
7r I U: U ~ 7r( U) is a PL fibration, then 7r is a PL fibration. Also, 7r is a PL
fibration if over one-dimensional subpolyhedra of B it is a PL fibration.

The next lemma, which follows from 2.1, will be useful later.
LEMMA 2.4. Let K ~ Band L ~ B be PL fibrations, K' c K a subfibra-
tion, and 1>: K' ~ L a fiber map. Then L U <p K ~ B is a PL fibration.
Proof. A mapping cylinder in an iterated mapping cylinder decomposi-
tion of L U <pK can be chosen of the form M(g U <pf), coming from a commuta-
tive diagram
K;~K:
"" 10 "11
'\. ~
Lo~L1
with !, f', and g contractible mappings. To check that g U </Jf is a contractible

mapping, consider first a point x E L 1' so that
(g U </J f)-1(X) g-1(X) U </Jof- 1(91 1(X») •
=:
Now f- 1(q)1 1(x») deforms into (f')-1(91 1(X») since both are homotopy equivalent
to ~11(x) under f and f', respectively. So (g U </Jf)-1(X) deforms into g-1(X)
which is contractible by hypothesis. Hence (g U </Jf)-l(X) is contractible if
x E L 1• In the opposite case x Et L 1' (g U </Jf)-1(X) =:f-1(X) is also contractible. D
We define now a classifying space for PL fibrations.
Definition. The simplicial space S has as a typical k-simplex a finite
subpolyhedron Le ROO X a k such that the projection L ~ a k is a PL fibra-
tion. The face and degeneracy maps are the obvious ones induced by
restriction and projection of ak to its (k - l)-faces.
Using the local characterization in 2.1 it is clear that S is a Kan complex.
Any PL fibration E ~ B (with compact fibers) is induced from a map B ~ S.
Homotopic maps correspond to "homotopic" fibrations, i.e., fibrations which
are restrictions to B x {O} and B x {1} of a fibration over B x I. In this
sense S classifies PL fibrations.
Heuristically, S can be thought of as "the space of all finite polyhedra",
or more precisely, as the (PL) singular complex of this "space". For if 1r: E-+
B is a PL fibration, then the covering homotopy property says somehow that
the fibers 1r- 1 (x) are polyhedra which vary "continuously" with x. For
example, one might ask, when can one finite polyhedron be deformed "con-
tinuously" into another, or in other words, what are the arc-components of
S? By 2.1 this amounts to asking when two polyhedra L o and L 1 can be
joined by a chain of contractible mappings. By 2.3, L o and L 1 must have the
same simple homotopy type. Conversely, since elementary collapses are
contractible mappings, we see that the arc-components of S are exactly the
simple homotopy types of finite polyhedra.
Recall the definition of the category e and its classifying space Be from
Section 1. We can define a map Be ~ S by sending the k-simplex L o ~
L 1->- • •• ~ L k of Be to its iterated mapping cylinder M(f1' .•• , fk) em-
bedded in Roo x a k by general position, preserving the projection to a k •
PROPOSITION 2.5. Be ~ S is a homotopy equivalence.
Proof. We will be a little sketchy, since the result will not be used
essentially in the rest of the paper. Let L be a k-simplex of S with a chosen
lifting L O to Be over aa k , representing an element of 1r k (S, Be). As in the
paragraph preceding 2.1 we can decompose L into iterated mapping cylinders
110 A.E.HATCHER
via triangulations of Land a k such that L -+- a k is simplicial, thereby obtain-

ing a lift L1 to Be over a k • The problem is to arrange things so that L1 is
homotopic in Be to LO over aa k •
LO is given as a union of iterated mapping cylinders M(ft, ... , fj"), one
for each j-simplex 0' of a triangulation of aa k • Triangulate the vertices Lf
of M(ft, ... , fJ) so that each ft is simplicial. This triangulation depends on
0', but we can suppose that if 7: is a face of a, then the 7:-triangulations sub-
divide the a-triangulations. Since the ft's are simplicial, each M(f1", ... , fj")
decomposes into iterated mapping cylinders of the restrictions of the ft's to
simplices of the L~'s. Now triangulate L so that all these iterated mapping
subcylinders are subcomplexes and form L1 from this triangulation.
To begin constructing a homotopy in Be from LO to L1 we first subdivide
aa k so LO and L 1have the same vertices L~ and differ only in the maps L~ ~
L~+u say ft for LO and g~ for L1. The change in LO resulting from subdivid-
ing aa k can clearly be realized by a homotopy of LO in Be.
Next, we construct homotopies from the g~'s to the ft's, inductively over
the skeletons of our triangulation of Lg. On the restriction M(ft, ... ,
fJ) I a l to a simplex a l of Lg, assuming ft's and gi's already agree on M(ft,
• • ., ft) I aa l , we can perform the well-known Alexander trick of radially
coning off g~ to ft. That is, for successively smaller concentric simplices
a~ c a l , we use the maps gi on M(ft, •.. , fJ) I a~ and the maps ft on M(ft,
••• , fJ) I a l - a~. This deformation of the gi's on M(ft, •.. , fJ) I al extends
naturally via a regular neighborhood of al in Lg to a deformation on all of
M(f~, ••• , fJ), and then we continue with (l + 1)-simplices of Lg. In the end
we get a deformation of M(ft, •.• , fJ) to M(gf, ••• , gj) as simplices in Be
(the underlying space M(ft, ••• , fJ) is unchanged during the deformation).
It remains to piece together these deformations over the various simplices
0' of aa , the trouble being that the deformation we have constructed over
k
0' depends on the triangulation of Lg. But for faces 7: of 0' we will have
chosen the triangulation of L~ to be a subdivision of the triangulation of the
appropriate L~, so further applications of the Alexander trick will provide
a way to glue everything together. Details are left to the diligent reader. D
Remark. The results of this section have analogues for PL quasi-fibra-
tions, in which "contractible mappings" are replaced by "homotopy equiva-
lences" throughout.
3. The simple homotopy functor S(K)
Recall that a k-simplex of S is a PL fibration L ~ Roo x a k -+- a k • It
will be convenient to label such a simplex by its fibers Lt == L n Roo x {t},
t E ~k, and to think of Lt as a k-parameter family of polyhedra.
Definition. For a fixed polyhedron Kc Roo, let SK be the subcomplex

of S consisting of those Lt's of the homotopy type of K, and let S(K) c SK
be the subcomplex consisting of Lt's which contain K as a deformation
retract.
SK is just a union of components of S. Restricting 2.5 to the subcategories
e(K) c e K c e (see Section 1) we obtain:
PROPOSITION 3.1. The maps Be(K) ~ S(K) and BeK ~ SK are homotopy
equivalences. D
Thus S(K) is, up to homotopy, the space called Wh (K) in Section 1.
PL fibrations and homotopyfibrations with fibers of the homotopy type
of K are classified by SK and BG(K) , respectively. Here G(K) is the H-space
of homotopy equivalences K ~ K. Since PL fibrations are homotopy fibra-
tions, there is a forgetful map SK ~ BG(K).
PROPOSITION 3.2. The homotopy fiber of SK ~ BG(K) is S(K).
Proof. The fiber ~(K) of SK ~ BG(K) consists of pairs (Lt, ft), where
Lt is a simplex in SK and ft: K ~ Lt is a family of homotopy equivalences.
In fact, this datum is exactly a fiber-homotopy trivialization of the fibration
Lt~t. If (Lt,ft) E ~(K), the mapping cylinder M(ft) lies in S(K), since by 2.4,
M(ft) ~ t is a PL fibration. The correspondence (Lt, ft) ~ M(ft) gives a map
~(K) ~ S(K) which has as a homotopy inverse the map Lt ~ (Lt, K ~ Lt),
as one can easily check. D
Remark. At the 1C o level the fibration sequence
G(K) -----+ S(K) -----+ SK -----+ BG(K)
is just the well-known representation of the set 1COS K of simple homotopy

types within the homotopy type of K as the orbit space of 1CoS(K)~Whl(1C 1K)
under the action of 1CoG(K). (See [8, § 24].)
PROPOSITION 3.3. S(K) is a covariant functor of K, from the homotopy

category of finite polyhedra to the homotopy category of (simplicial) infinite
loopspaces.
COROLLARY 3.4. If K is contractible then so is S(K).
This is because if K is a point we can just cone off all Lt in S(K) uni-
formly.
112 A. E.HATCHER
Proof of 3.3. For a map f: K -+ K' and Lt E S(K), set f*(L t ) == K' U f Lt.
By 2.4 this lies in S(K'). Also by 2.4 a homotopy of f induces a homotopy of
f*(L t ) in S(K').
The composition operation" +" in S(K) is "disjoint union with the two
copies of K identified." To achieve the disjunction we will make essential
use of the given embeddings in Roo. Write Roo as Rr' x R x R~. For each
rectangle R == (a 1, b1) x (a 2 , b2 ) x·· · in R~, with 0 ~ a i < bi ~ 1 and (ai' bi ) ==
(0, 1) for sufficiently large i, let the cone C(R) be the union of rays in R x R~
from the origin 0 x 0 through points in 1 x R. We can assume that all
Lt E S(K) lie in Rr' x C((O, 1) x (0,1) x ... ) and that Lt n R~ x 0 x 0 == K.
Now to form Lt + L~, first compress Lt linearly into Rr' x C((O, 1/2) x (0, 1) x
... ) and L~ linearly into Rr' x C((1/2,1) x (0,1) x ... ). Then set Lt + L;
equal to the union of the shifted Lt and L~, which now intersect only in K.
This sum operation clearly makes $(K) a homotopy associative H-space. In
fact, S(K) now has an obvious "little cubes" structure, making it into an
infinite loopspace [12]. For example, homotopy commutativity follows by
the familiar argument using the first two coordinates of R~ to slide around
in. D
4. Families of PL cell complexes
For a more detailed study of S(K) we will need to replace it by a
homotopy equivalent space, whose k-simplices are k-parameter families of
polyhedra Lt with chosen decompositions into PL cells. Each Lt will be con-
structed from K by successively attaching cells of various dimensions by PL
attaching maps. The idea in defining a k-parameter family is to allow two
kinds of operations: homotopies of attaching maps, and the "collapsing" of
certain collections of cells by coning them off to a point. For example, an
elementary c~llapse L o ~ L 1 can be realized by a one-parameter family L tJ
o ~ t ~ 1, as in the following picture.
Another one-parameter family, demonstrating a homotopy of attaching

maps, is the following.
As this example illustrates, in parametrized settings it is unreasonable to

require attaching maps to be skeletal in each parameter slice. Indeed, it is
impossible to go from L o to L 1 by a homotopy through skeletal attaching
maps. Thus the PL cell complexes we permit may not be decomposed as CW
complexes, though the underlying spaces are nice polyhedra.
The full definition of a k-parameter family of PL cell complexes is some-
what complicated:
(i) Set LiO) == K, tEak.
(ii) Inductively, build Li i ) from Li i - 1 ) by
(a) attaching a PL cell e~i via a PL k-parameter family of maps fJJ~:
Sni- 1 ~ Li i - 1 ) and then
(b) collapsing e~i to a point p~ E Li i - 1 ) over some subcomplex of {t E
a k I fJJ:(Sn i -1) == p~}.
Globally, Lt == Ui Li i ) is assumed to satisfy:
(iii) For each collapse point p, the cells of Lt collapsing to p are attached
consecutively, as a block, without intervening non-collapsing cells.
(iv) The underlying polyhedra of the family Lt form a k-simplex of
S(K). That is,
(a) Lt ~ t is a PL fibration.
(b) K ~ Lt is a homotopy equivalence.
A family of PL cell complexes constructed according to these rules we call
a basic k-parameter family. A general family consists of basic families over
the simplices of some subdivision of a k • Thus in a general k-parameter
family the cells need not be attachable in the same order all over a k • (The
actual order in which cells are attached is not part of the data of Lt' only
the decomposition of Lt - K into cells.)
Definition. For 0 ~ i ~ J ~ 00, S{(K) is the simplicial space whose k-
simplices are general k-parameter families of PL cell complexes Lt such that
all cells of Lt - K have dimensions in the range ri, J]. The face and degen-
eracy maps in S{(K) are the obvious ones.
S~(K) is clearly a Kan complex, since we include general k-parameter
families.
114 A.E.HATCHER
Remarks. (1) If the normalization condition (iii) were not assumed, it

could easily be achieved by a small homotopy of attaching maps.
( 2) In the presence of (iii), condition (iv. a) is equivalent to:
(iv. a') Near each collapse point P, the block Ct(p) of cells which collapse
to P attaches to Li i ) (for appropriate i) inside a small contractible neighbor-
hood N t c Li i ), such that N t is a deformation retract of N t U Ct(p).
Thus S1(K) can be defined without mention of PL fibrations.
( 3) Any family Lt E S~(K) can be homotoped in S~(K) to a family in
which all collapsing is sptit, Le., of the form Lt V Ct ~ L o where Ct is con-
tractible. To achieve this, first use the deformation retractions N t U Ct(p) --+
N t of (iv. a') to push all non-collapsing cells off Ct(p), then use a contraction
of N t to make Ct(p) attach to N t at a point.
( 4) One can always tell by inspection when a non-zero-dimensional cell
of Lt - K collapses: it shrinks to a point. But when two or more O-cells
merge into one, there is no intrinsic way to tell which O-cells collapse and
which is the survivor. So let us agree as a convention that collapsing O-cells
are distinguished. It is easy to achieve this by a homotopy of the family
Lt: Let e~ ELt, tEal c a k, be a family of O-cells. Extend e~ to a point Pt E Lt
for tEak, and attach a line segment [0, l]t (with the usual cell decomposition)
to Lt by identifying 0t with Pt. Then collapse [0, l]t to It over al. The effect
is to replace e~ by It over a l (and It collapses over any part of a l where e~
collapsed). Moreover, any O-cells of the original Lt which abut the new It
over a l are now distinguished as collapsing O-cells, since It survives near al.
PROPOSITION4.1. The natural map SC;(K) ~ S(K) obtained by ignori1~g
cell decompositions is a homotopy equivalence.
Proof. Represent an element of 7r k(S(K), Sc;(K») by a family Lt E S(K),
tEak, having a given lift L~ E S~(K) over oak. Choose a triangulation T of
L == Ut Lt such that the families of cells of L~ are subcomplexes and such
that the projection 7r: L --+ a k is linear on each simplex of T. Intersecting T
with Lt gives each individual Lt a PL cell complex structure, but the result-
ing family Li of PL cell complexes may not lie in S~(K) because cells may
not collapse to points but to cells of positive dimension. However, if we
perturb 7r slightly so that for each simplex a l in T, 7r(a l ) has maximal dimen-
sion, namely min (k, l), then the collapsing will be of the sort in S~(K). By
repeated application of 4.3 below, this perturbation can be done in S(K), and
over oak it is clear that there is induced a deformation of L~ in Sc;(K). The
final step is then given by:
LEMMA 4.2. Let the family Li E Sc;(K) be a subdivision, of the family

Li E S~(K). Then L~ and L~ are homotopic in Sc;(K).
Proof. Proceeding inductively over simplices of the parameter domain,
we need only to do a relative construction for t E ~ k. Consider the following
homotopy within a cell e~ of Lt (n > 0). In the first half of the homotopy
we split e~ into two n-cells and an (n -I)-cell by intersecting it with a family
of parallel hyperplanes:
.-----new n - c e l l - - - - - .
In the second half of the homotopy we reverse the process, but with e~
replaced by its subdivision in L~. Doing this simultaneously for all cells of
L~ gives a homotopy from L~ to L~ in Sc;(K). In order to make this homotopy
fixed over a~k (where by assumption Li == LD we can first deform L~ to be
constant on the segments t x [0, e] of a collar neighborhood a~k x [0, cl,
then damp the homotopy down to zero along these segments. D
LEMMA 4.3. Let 7T:: L ~ ~ k be a PL fibration, and let n' be obtained from
1r by perturbing the image of one vertex of some triangulation of L in which
1r is linear on simplices, then extending li1~early. If the vertex lies over the
interior of ~k and the perturbati01~ is sufficiently small, then 7T:' is also a
PL fibration.
Proof. Along a line in ~k the given triangulation of L gives a decom-
position of 7T:, near a given 7T:-slice L o, into a mapping cylinder projection
M(f-1) U M(f1) ~ [-1, 1] for contractible mappings L_ 1 ~ L o J.!- Lp If
the line in ~k is parallel to the direction of the perturbation, the fibers L: of
7T:' near L o can be obtained, up to isotopy along the rays of M(f-1) or M(f1) ,
as follows. For some function lfit: Lo~[-l, 1], L: intersects M(f_1If~1(x») U
M(f1 I f1- 1(x») in the slices n- 1(ifJt(x»), x E L o• The projection gt: L: ~ L o is a
contractible mapping, being the restriction of M(f-1) U M(f1) ~ L o. So at
least the fibers of n' are homotopy equivalent to the fibers of '!r.
We now check that n' is a PL fibration along this line in ~k parallel to
the direction of the perturbation. A given x E L o lies in a minimal simplex of
the triangulation of L, which intersects L o in a convex cell a. We distinguish
two cases:
( 1) a does not lie in a fiber of 7T:'. Then near x the fibers L~ are inde-
116 A. E. HATCHER
pendent of t, up to isotopy, so n:' is a fibration near x.

( 2) (J lies in a fiber L~. Writing n:' as M(f~) U M(f~) ~ L~ for maps
f' f'
L'- ~ L~ ~ L~, then we can describe (f~)-l(X) as follows. Let N be a
transverse section of (J in L o at x. Thus N is a join aN * x, where aN is the
link of (J in a neighborhood of (J in L o. Let aN+ be spanned by the vertices
of aNfor which 90> 0, and set N+ == aN+*x. Then (f~)-l(X) is gol(N+), and
so is contractible since N+ is a cone. The same arguments apply to points of
L~ near x, hence n:' I M(f~) is a PL fibration near x. And similarly with f!-.
Thus n:' is a PL fibration along lines in the direction of the perturbation.
Consider now a mapping cylinder M(f') for 1C' in a direction other than
that of the perturbation, f': L~ ~ L~. If y E L~ and x E (f')-l(y), let M(f) ,
f: L o~ L H be a mapping cylinder for 1C such that x E L o and YELl. Then
(f')-l(y) == (gof)-l(y), where go: L~ ~ L o is as above. Since go and f are con-
tractible mappings, (f')-l(y) is contractible. D
Probably a more general statement than 4.3 is true, that any perturba-
tion of 1C sufficiently small with respect to the given triangulation of L is
still a PL fibration. (The proof just given applies in fact to the case that all
vertices are perturbed in one direction.)
5. Suspension in S(K)
In this section we define a suspension operation ~: S(K)~S(K) and prove
two facts about it: that it is a homotopy inverse for the H-space structure
"+" on S(K), and that it satisfies a nice stability property. An "external"
suspension S(K) ~ QS(SK) , apparently unrelated to ~, will be defined in
Section 7.
Let r: L ~ K be a retraction. Its suspension ~r: ~L ~ K is defined as
follows: ~L is L x Iwith L x aI collapsed to K x aIvia r x id ar , then with
K x I collapsed to K via projection. And ~r is given by ~r(x, s) == r(x). For
example, if K is a point, ~L is just the usual reduced suspension.
Now let Lt be a family in S(K). ,Then there is a family of deformation
retractions rt: Lt~K (in the strong sense that Kis fixed during the homotopy
r t ~ id), unique up to canonical homotopy. Consider the suspensions ~rt: ~Lt--+
K. By 2.4, the two collapses by means of which ~Lt is obtained from Lt x I
preserve PL fibrations. So ~Lt is again a family in S(K). Thus we obtain
~: S(K) ~ S(K), determined up to homotopy (the choice of rt).
PROPOSITION 5.1. The map ~ + id: S(K) ~ S(K) is null-homotopic.

Proof. Consider the cone operator C on S(K) defined by CL t == Lt x
Ij{(L t x 0) U (K x I) K} as in the definition of~. A deformation r t ~ id
I"J
gives a homotopy C ~ L + id in S(K), by 2.4. But also C ~ 0 from the

projection Lt x I --+ Lt x 0 which is a contractible mapping. D
As a simple application we globalize the well-known product formula for
Whitehead torsion. The sum formula can be treated in a similar fashion
(see e.g., § 23 of [8] for the classical case).
PROPOSITION 5.2. Let C be a finite connected polyhedron. Then the
product map S(K) --+ S(K x C), Lt ~ Lt x C, is homotopic to x(C) ·i, ~vhere
x(C) is the Euler characteristic of C and i: S(K) --+ S(K x C) is induced by
inclusion of a factor.
Proof. Choose a triangulation of C. Then if Lt E S(K), deform Lt x C
as follows. For a~ E C, push Lt x ~~ down into K over oa'n via the deforma-
tion r t ~ 1. Then compose with a contraction of K x an to K x O. This
deforms Lt x ~~ to the n-fold suspension L n Lt. Doing this for all simplices
of C by downward induction on n, Lt x C is deformed to EL\nec (LnL t) ~
x(C) ·i(Lt ). D
It will be useful to have L operate in S~(K) as well as in S(K). A priori
the suspension of a family Lt E S~(K) need not lie in S~(K), since the suspen-
sion of a block Ct(p) of cells which collapse to a point pELt - K collapses to
a line p x I, not a point. But there is an easy way to correct this, by
contracting LCt(p) to the center of p x I as it collapses.
PROPOSITION 5.3. The map L: S{(K) --+ S{ti(K) is (2i - j)-connected,
provided i > 1.
LEMMA 5.4. Let the retraction r: L ~ K be i-connected. Then the sus-
pension L: 7l: k (r) --+ 7l: k +1(Lr) is an isomorphism for k < 2i and an epimor-
phism for k == 2i.
Proof. Let C be the mapping cylinder of r. In C x [-1, 1] we have the
subspaces C == C x 0, A == C x -1 U L x [-1,0], and B == C x 1 U L x [0,1].
Consider the following diagram:
7l: k (C, A n B)
aI"" a
Jr k +1 (C x [0, 1]; C, B) ----~ 7l: k (B, A n B)
1 "" a 1
7l: k+l( C x [-1, 1]; C x [-1, 0], A U B) ~ 7l: k(A U B, A)
I"" ""
7l: k + 1 (C x [-1,1], A U B) .
118 A.E.HATCHER
The four maps labelled isomorphisms come from the various long exact
sequences of the triads (C x [0,1]; C, B) and (C x [-1,1]; C x [-1,0],
A U B), while the two other vertical arrows are induced by inclusion. The
composite 1r k(C, A n B) --+ 1r k+1(C x [-1,1], A U B) can be identified with
~: 1rk(r)--+1rk+l(~r). Since the pairs (A, A n B) and (B, A n B) are i-connected
by hypothesis, (B, A n B) --+ (A U B, A) is 2i-connected by homotopy excision,
and the result follows. D
Proof of 5.3. Consider first a single L~ E Siti(K) which we wish to
desuspend. Suppose inductively that we have homotoped the attaching maps
in a subcomplex of L~ to a suspension ~Lio), together with the deformation
retraction r:: L~ --+ K restricted to ~Lio), say ~rt. Then the attaching map
Sl --+ ~Lio) of a cell e~+l EL:, plus r: I e~+t, give an element of 1r l +1 (~rt). Since
r t: Li o) --+ K is i-connected, the lemma says that e~+l and r; I e~+l can be desus-
pended provided l ~ j ~ 2i. This is the inductive step in showing that ~ is
O-connected if 2i - J- ~ O.
Desuspending a family L; representing an element of 1r*(~) is done
inductively over basic k-simplices, and cell by cell within each basic k-simplex.
Desuspending cells e~+l over a k, with the assumption of a desuspension over
Ja k , is possible if l + k ~ j + k ~ 2i if one uses a straightforward fibered
version of homotopy excision, which we leave to the reader. Collapses are
desuspended to collapses by considering them as the case that K == N t , in
the notation of Remark (2) of Section 4.
The assumption i > 1 is needed to assure that the desuspension Lt of L~
actually lies in S{(K). For if i > 1 then 1r 1K --+ 1r 1L t is an isomorphism and
H*(L t, K; Z[1r1Kl) ~ H*(~Lt, K; Z[1r 1K]) == 0 ,
so K is a deformation retract of Lt. D
6. Trading up cells into two dimensions
One of the main geometric steps in the proof of Whitehead's theorem on
simple homotopy types is the assertion that any homotopy equivalence is,
modulo elementary expansions and collapses, an inclusion, K~L such that
the cells of L - K all have dimension either n or n + 1, for a fixed n. We
prove now a parametrized version of this which will be the basis for all the
deeper results about S(K) in the remainder of the paper.
THEOREM 6.1. The inclusion S~+l(K) ~ Sc;(K) is (n - l)-connected if
n > 1.
Proof. Let the family Lt represent an element of 1r kSC;(K), say Lt E Sg(K)
for some j ~ n + 1. By 6.2 below, Lt can be deformed to a family L~ E

S~-l(K). Then by 5.2 L~ is homotopic to an iterated suspension "Lj-n-1L~' of a
family L~' E S~+l(K), provided k ~ n - 1. If j is chosen so that j - n - 1 is
even, then by 5.1, L~' is homotopic to L~. This gives surjectivity of
1rkS~+l(K) ~ 1rkS~(K) .
Injectivity is similar. If Lt' t E a k+1, is a contraction in Sg(K) of an
element of 1rkS~+l(K) for some j with j - n - 1 even, then adjoin to a k+1 an
external collar supporting a homotopy of Lt I oa k+1 to "L j- n- 1 L t I oa k+1 • The
resulting family L~ E St(K) can be deformed to L? E S~-l(K), reI boundary,
by 6.2 below. If k < n - 1, L~' desuspends to a contraction in S~+l(K) of the
original Lt loa k + 1 • D
It remains to prove:
PROPOSITION 6.2. S{~l(K)~S{(K) is a homotopy equivale1~ce (i+1<j).
Proof. Let Lt represent a class in 1r n(S{(K), S{+l(K»), and let e~ be one
of the i-cells of Lt' attached by 9t: 8 i - 1 ~ Li·). We will show how to cancel
e~ over a k-simplex a k, assuming a cancellation given over oak, introducing
only i + 1- and i + 2-cells in the process. Induction on k will then allow e~
to be cancelled everywhere, and the result will follow by iterating for other
i-cells.
The actual elimination of e~ over a k proceeds just as in the unparame-
trized case:
LEMMA 6.3. If <Pt exte1~ds to a map (fit: D i ~L~·) over a k, which also
collapses to a point over aa k, then e~ can replaced by an i + 2-cell over a k.
Proof. We may use ;Pt to deform <Pt to the constant map. Then Ute/lk e~
is an (i + k)-sphere in L == Ute/lk Lt. By rechoosing the extension (fit by an
el~ment of 1r i +k(L) == 1r i +k(K) if necessary, we may assume the i-sphere e;
(Le., trivially attached i-cell) bounds a disc D ti +1 in Lt which also collapses
over oak. Now introduce a trivial pair of cells e~+l and e~+2 which together
form a disc attached to Lt at a point of D/+ 1 and collapsing over oak. The
discs D/+ 1 give a homotopy of the attaching map of e~+l so that oe~+l == e~.
Next, push all higher cells off the disc e~ U e~+l. Then cancel e~ and e~+l. D
It remains to find the extension (fit. When k == 0 this is trivial-cells of
lowest dimension must be attached by null-homotopic maps. But for k > 0,
Ute/lk <Pt: 8 i +k- 1 ~ Ute/l k Li may well be non-trivial. Our aim will be to kill
o
9t simultaneously for all tEak by attaching new i + 1-cells to Li o
).
First we specify more closely the subcomplexes Li in which of course

o
),
the index (.) varies with t, since the order of attaching e~ may vary with t.
120 A.E.HATCHER
We want to choose a subdivision Tof a k such that each cell of Lt is contained

in L~') over an open subset of a k which is a union of open simplices of T. To
achieve this we begin with a subdivision T' of a k such that Lt consists of
basic simplices in S1(K) over simplices of T'. We can assume each cell et of
Lt lies in L~o) over a subcomplex S(e t) of T'. Then we can make et E Li o) attach
only to cells of L~o) over a regular neighborhood N(e t) of S(e t) by a small
homotopy of attaching maps. For if et is disjoint from a cell e: over S(e t),
then it is disjoint from the center of e: over a neighborhood of S(e t) and so
it can be pushed off e: near S(e t). Finally, we enlarge L~o) by including et
over the interior of N(e t). After doing this for all cells et we subdivide T'
to a triangulation T in which the N(e t ) are subcomplexes. (In this construc-
tion we regard collapse points as extensions of the cells which collapse to
them. Thus if a collapse point p belongs to Li then in nearby t-slices all
o
),
the cells collapsing to p also lie in L~o).)

Now we show how to construct ;Pt inductively over skeletons of the
triangulation T of a k. We have already remarked that ;Pt exists over vertices
of T; in fact, we can first deform ifJt into K by general position and then
choose for (fit the image of e~ under the retraction r t: Lt ~ K. Assume induc-
tively that (fit has been constructed over the (l - I)-skeleton of T such that
ifit :::: e~ (reI CPt) in L o and consider the problem of extending ifit over an l-
simplex al. By the construction of T in the first place, and by induction
thereafter, we can assume:
ifit extends from Oal up to a neighborhood Ne a l of the center of a l
( *) such that Li-) f-+ t is a fibration over N.
The obstruction to finishing the extension is the homotopy class of a map
a: Si-l+l ~ Li o), tEN. In Lt, the cells e~ themselves provide a contraction, a
of a. Since Li ~ t is a fibration over N, we can assume a is spread as an
o
(l-1)-parameter family (D/+\ OD/+l)~(Lh Li o»), t E Dl-l, where N == D l- 1 X

D 1 , such that Dti+l reduces to a point for t E C, a collar neighborhood of
oD l- 1 in Dl-l. Extend (D/+\ oDl+l)~(Lh Li o») over Nx Dk-l, a neighborhood
in a k• Now adjoin a trivial pair of cells (e~+l, e~+2) which together form a
ball attached to L~o) at a point in D/+l. Using D/+ 1 , deform e~+l so that it
attaches to L~o) along OD/+l. If we let Li o) include the cell e~+l over N x Dk-l
and the cell e~+2 over C x D 1 X Dk-l, then we have killed the obstruction to
extending ifit over al.
To guarantee that (*) is preserved we need onlych oose N x D k - l to include
the centers of all simplices of T which it meets and to be radially "starlike"
in these simplices. This completes the induction step in the construction of
;Po and hence the proof of 6.2. D

7. Excision in S(K)
Suppose our base space "K" is the union of two subpolyhedra A and B.
Then we have inclusions
S(A n B) c S(B)
n n
S(A) c S(A U B)
THEOREM 7.1. If (A, A n B) is m-connected and (B, A n B) is n-con-
1~ected, with m > 1, then the map
7r k (S(B), S(A n B)) ~ 7r (S(A U B), S(A))
k
induced by inclusions is surjective if k ~ m +n- 2 and injective if

k < m + n - 2.
As in ordinary homotopy theory, excision implies a suspension theorem.
Consider the (ordinary) suspension SK of a polyhedron K as the union of two
cones A and B on K, with A n B == K. Since S(A) and S(B) are contractible,
the homotopy fibers of S(A n B) ~ S(B) and S(A) ~ S(A U B) are S(K)
and QS(SK) , respectively. The map of homotopy fibers S(K) ~ QS(SK)
induced by the inclusion (S(B), S(A n B)) c (S(A U B), S(A)) sends Lt E S(K)
to the loop L~ in S(SK) defined by
L~ == SK U (Lt x s) c SL t , 0~ s ~ 1 ,
where SL t is, as usual, Lt x I with Lt x 0 and Lt x 1 collapsed to points.
COROLLARY 7.2. If K is m-connected, the suspension map S(K) ~
QS(SK) is (2m - 1)-connected. D
Thus the iteration
S(K) ~ QS(SK) ~ Q2S(S2K) ~ ···
eventually becomes highly connected, and we can speak of the stable simple
homotopy functor s(K) == lim Q""S(S"" K): The stable simple homotopy groups
si(K) == 7r i s(K) == lim 7r i +""S(S""K) satisfy excision by 7.1, hence form a
generalized homology theory s*(K).
In the course of the proof of 7.1 we will obtain also the following:
7.3. If K ~ K' is m-connected, m
PROPOSITION > 1, then the induced
map S(K) ~ S(K') is (m - 1)-connected.
For example, taking K' to be an Eilenberg-MacLane space K(7r 1K, 1),
this is the well-known fact that 7r oS(K) depends only on 1r 1K. (If K' ==
K(7r 1K, 1) is not a finite complex, take "S(K')" here to be Ua S(K;), the union
122 A.E.HATCHER
over finite subpolyhedra K; c K'.) The proposition does not extend to m == 1

since a surjection TC1K -+ TC1K' may not induce a surjection Wh1(TC1K)-+
Wh1(TC1K'), e.g., if TC1K is a free group, in which case Wh1(TC1K) is zero.
LEMMA 7.4. Given K' == KU em +1and Lt E S1+ 1(K'), tEak, we can deform
the attaching maps of cells in Lt - K' so that they attach to em +1 over a
neighborhood of a (k - m)-dimensional subcomplex of a k, staying in S1+ 1(K)
over any part of a k where Lt E S}+l(K).
Proof. Consider for example a cell e~+l E Lt - K' and suppose inductively
that lower cells have already been fixed up. Let <Pt: Si -+ L~·) be the attach-
ing map of e~+l. In general position <Pt will be transverse to the center Xt of
the top cell of L~·) (as a k-parameter family), and the pullback Ut <Pt1(Xt) will
be a manifold 11c Si x a k of dimension ~ k. Outside a small neighborhood
NI of 11 we can push <Pt off this top cell of L~·). Then repeat the process for
the next highest cell of L~·) to get a manifold 12 c Si x a k - NI with neigh-
borhood N 2 , etc. So we can assume that outside N == Ui Ni the family 9t
maps into K'. Extend <Pt to (fit: D i +l -+ L~·) U e~+l by identifying the interior
of D i +l with e~+l. We can choose a family of (i + I)-balls D t c D i +1 x {t}
such that Ut Dt is a small neighborhood of the fiberwise join of N with x °
a k in D i + x a k, as follows. Write D i +l as the union of a small concentric
1
disc D oi +l with an annulus Si x [0,1], Si X 1 == aD i + 1 • Letf: Si x a k -+ [0,1]

°
be 1 on Nand away from N. Then set
Dt == D oi +l x {t} U {(x, s, t) E Si X [0, 1] x {t} Is ~ F(x, t)} ,
Le., the "shadow" of the graph of f. Note that Ut D t has a (k + I)-dimen-
sional spine, since N has a k-dimensional spine.
Now let r t: Lt -+ K' be a family of retractions. Since <Pt == rt<pt outside
N, r t9t I D i +1 - Dt defines a homotopy of <Pt to a new attaching map <p~ which
equals 9t on Nand rtlfit I aDt outside N. In general position the image of the
spine of Ut D t under rtlfit will meet the center of em +1 in a (k - m)-dimensional
complex, so outside a neighborhood of .this we may assume <P: attaches e~+l
to K' - em +1 == K. This is the inductive step. Note that if for certain tEak
the original Lt lies in S1+ 1(K) c S1+ 1(K'), then choosing r t to retract Lt - K
onto K guarantees that the modified Lt is still in S~+l(K). D
Proof of 7.3. Since S is a homotopy functor, we can take K -+ K' to be
an inclusion such that K' - K consists of cells of dimension ~ m + 1. For
each of these cells in turn apply 7.4 to a family Lt representing an element
of TC k(S1+ 1(K'), S1+ 1(K»), k < m. The hypothesis m > 1 is only necessary to
assure that TC1K ~ TC1K', so that the new Lt contains K as well as K' as a
deformation retract. D
Proof of 7.1. By induction it suffices to take A == K Uem+t, B == K U en+1 •
Let Lt E S~+l(A U B), t E Dk, be such that, writing aDk == D~-l U D~-l, we have
Lt E S~+l(A) for t E D~-l, Lt E S~+l(B) for t E D~-l, and Lt E S~+l(A n B) for t E
D~-l n D~-l. Applying 7.4, we obtain a set SA c Dk - D~-l over which cells
of Lt-(A U B) attach to A, and SA has a spine of dimension k-m. Similarly
we obtain SB c Dk - D~-l with a (k - n)-dimensional spine. In general
position we can assume that SB is disjoint from sh(SA)' the "shadow" of SA
under projection to D~-t, provided (k - m) + (k - n) < k - 1, or k ~ m +
n - 2. Then there is an evident homotopy of the family Lt, fixed over D~-l,
which excises sh(SA) from its parameter domain. Again, the hypothesis
m > 1 implies that Lt actually is in S~+l(B) over the new Dk and similarly
that Lt is in S1+ l(A n B) over the new D~-l. This shows that the inclusion
(Sl+l(B), S~+l(A n B») c (S~+l(A U B), S~+l(A») is surjective on TC k and injective
on TCk-l- D
8. Families of PL handlebodies
We continue to work entirely within the PL category. Let Wn+l be a
compact connected manifold and let Mn be a codimension zero submanifold
of aW. By a handlebody structure h on (W, M) we mean:
( 1) There exists a filtration of W by codimension zero submanifolds
W == W(N) => W<N-l) => ••• => W(O) where W(O) ~ M x I is a collar on M x 0 ==
MeW (with aM x I c aW) and W(i) is obtained from W(i-l) by attaching
a handle Dni x Dn+l-ni via an embedding epi: Sni- l X Dn+l-ni ~ aW(i-l).
(2) A product structure on the collar W(O) ~ M x I and on each handle
Dni x Dn+l-ni is given. That is, the homeomorphisms W(O) ~ M x I and
W(i) - W(i-l) ~ iJni X Dn+l-ni are specified only up to product homeomor-
phism fl x f2: M x I ~ or fl x f2: Dni x Dn+l-ni~. (Thus in a product struc-
ture on Xl x X 2 the collections of slices {Xl x {x 2 } I X 2 E X 2 } and {{Xl} x X 2 I Xl E
Xl} are well-defined.)
Now we define the notion of a k-parameter family of handlebody struc-
tures h t on (W, M), t E ~k. We first allow the submanifolds W(i) to vary
through a k-isotopy Wt(i) , as follows. For i == 0 the collar Wt(O) moves by
k-isotopy fixed on M == M x o. For i > 0 the handle Dni x Dn+l-ni moves
by a k-isotopy Dr i x Dtn+l-n i . In particular, the attaching maps epi can vary
by a k-isotopy ep~.
But also we want to allow certain collections of handles to be coned off
to a point. The prescription for this goes inductively on k, as follows. For
124 A.E.HATCHER
to E ~k, suppose that over the boundary of a small neighborhood of to in ~k

the (k - I)-parameter family h t has a filtration with layers Wt(i) C Wt(j) such
that the handles of Wt(j) - Wt:) attach to aWt(i) inside a disc Dt' in such a way
that Wt(i) is homeomorphic to Wt(j) by a family of homeomorphisms fixed out-
side a small neighborhood of Dt' in Wti ). In this situation we can obtain a k-
parameter family Wt(j) by simply shrinking Wt(j) - Wti ) radially to a point
Pt E Dt as t goes radially from the boundary of the neighborhood of to to to
(the Alexander trick). Thus the submanifolds Wt(j) change to Wt(g) == Wt(~) by
isotopy. More generally, we allow simultaneously a finite number of such
collapsing operations to be going on independently at distinct points Pt E W.
A k-parameter family of handlebody structures h t on (W, M) obtained
by the above process we call a basic family. A general family h t is one which,
over the simplices of some subdivision of ~k, consists of basic families. The
distinction between basic and general is that in a basic family all the handles
can be attached in one order independent of t (though the choice of such an
order is not part of the data of h t ). The collection of all k-parameter families,
k == 0, 1, 2, ... , forms a simplicial space which we denote §( W, M). The face
and degeneracy maps are the obvious ones, and §(W, M) is clearly a Kan
complex.
Example. We can consider as one-parameter families according to the
above definition all of the handle operations used in the proof of the PL
h-cobordism theorem, namely isotopies of attaching maps of various sorts or
cancellations of complementary pairs of handles. As a very special case, let
(W, M) == (M x I, M x 0) be given first the handlebody structure h o with no
handles and just the collar Wo(O) == M x I (in the given product structure).
A second handlebody structure h 1 on (M x I, M x 0) can be obtained from
a handlebody structure on M as follows. The collar W1(O) of hI is M x [0, 1/2],
and for each i-handle Di X Dn-i of Mwe can consider Dit, x (Dn-i x [1 - c, 1])
as an i-handle in M x I and (D i x [1/2, 1 - cl) x Dn-i as an (i + I)-handle.
See figure (a). We can isotope these t:wo handles in M x I so they attach
only to M x 1/2, as in figure (b). The resulting complementary pair can then
be coned off, producing figure (c). Doing this successively for all handles of
h u we obtain a one-parameter family h t connecting h 1 with hoe
~-r-------r-r-7 ••••.Jfx 1 .....
~--t------+---+-<. .. .il;f x 1- s
.>-&- J..-.A..-< .•• .iv!xl/2
c.- -J' •••• MxO

(a) (b) (c)
The reason for permitting collapses in §( W, M) more general than the

collapse of a complementary pair of handles is that with the more general
collapses we can prove the following key result.
THEOREM 8.1. §( W, M) is contractible.
Proof. Let ht, t E Sk, represent an element of n:k§(W, M). We will show
how to homotope the family h t to a constant family. To begin, we deform
h t so that the collars Wt(O) ~ M x I all agree with a chosen standard collar.
The collars Wt(O) are all standard at M ~ M x 0, so by an isotopy within Wt(O)
we can assume they are standard near M x 0, say on M x [0, 1/2]. Then
using the one-parameter family in the preceding example, we can decompose
the remaining non-standard half of Wt(O) into handles by a homotopy of the
family ht, leaving us with only standard collars.
The idea for the rest of the proof is to show that ht == h~ and a constant
family h~ have a common subdivision h~/2, in the obvious sense that h~/2
intersects each family of handles Dl x Dt-+ 1 - ' of h~ or h~ in a union of handles
of h~/2, and in fact, in a family in §(D' x Dn+l-i, aD' x Dn+l-'). Then by the
obvious handlebody version of 4.2, a family is homotopic to a subdivision of
itself, so h~ ~ h~!2 ~ h~.
Let T be a triangulation of W x Sk such that the k-parameter families
of handles of h~ and h~ are subcomplexes. We can assume T is transverse
to the slices W t == W x {t}, so that a simplex a l of Tintersects each slice W t
in a ball Btl-k, a point Pt, or not at all. For a slice W t this ball decomposition
leads to a handlebody structure h~/2 in the usual way: small neighborhoods
of the O-balls are O-handles, small neighborhoods of the 1-balls, minus the
previously constructed O-handles, are 1-handles, etc. We shall call such
handles handle neighborhoods Nt(Bl- k) or Nt(pt) (though they are not really
neighborhoods). Unfortunately, this construction fails to give a k-parameter
family in §(W, M). The trouble comes in a neighborhood N of the k - 1
skeleton Tk-l of T, where a ball Bl- k can shrink to a point Pt E T k-l, although
its handle neighborhood Nt(B/- k) cannot change to the O-handle Nt(pt)
"continuously," i.e., by any natural path in §(W, M). So we will have to
splice things together somehow near T k - 1 •
We can choose the neighborhood N of T k - 1 to have the following prop-
erties:
(1) N == Nk-l::) · · · ::) N°, where Ni is a neighborhood of T i obtained
from Ni-l by adding handle neighborhoods N(a') of the i-simplices of T',
such that Nt(a') == N(a') n W t is a k-parameter family of O-handles over the
i-handle 1r(N(a'») in Sk, where n:: W x Sk ---+- Sk is projection.
126 A.E.HATCHER
(2) If the i-handle n(N(a'») is written as D i x Dk-" then over

D' x aDk-' the O-handle Nt(a i ) is the union of O-handles Nt(a i ) for d > i
and handles Nt(Btl-k) of W t - N.
Now we can extend the family of handlebody structures h~/2 on W x Sk - N
to all of W x Sk, one simplex ai of Tk-l at a time by downward induction
on i, as follows. Over D i x aDk-i we have inductively a family of handle-
body structures on the (n + l)-disc Nt(a i ). Since ~(D'n+t, rp) ~ * by 8.2
below, this extends to a family over D' x Dk-i. See figures (d), (e), and (f)
for an example when k == 1. Note also that any two such extensions of h~/2
over N are homotopic, by the relative form of the same argument.
,--- J.V
A. ____..
1
o
o
o 1
t
~
1
(d) Triangulation T (f) Splicing over N
LEMMA 8.2. ~(Dn+l,~) and ~(Dn+l, Dn) are contractible.

Proof. We can uniformly cone off any family h t in ~(Dn+t, rjJ) to a single
O-handle, as in the following sequence:
For ~(Dn+t, Dn) we can take all collars to be standard and then deform h t E
~(Dn+t, Dn) to a fixed (1, O)-handle pair: D
~---- Collar - - - - - - - -
Returning to the main argument, the family h~/2 as constructed is not

yet a subdivision of h~ (or hD. Even in the unparametrized case a handle
neighborhood Nt(Bl-k) of a ball Bl-k in the intersection of two handles of h~
will not lie inside either handle. But such a ball Bl-k lies in D/ x aDt+ 1 - i for
only one handle Dl x Dt+ 1 - , of h~, so let us enlarge this handle by a small
collar on Dl x aDt+ 1-;' consisting of all these Nt(Bl-k) for which Bl-k c Dl x
aDt+ 1- ' . In other words we are moving h~ to h~/4 by a small isotopy so that
h~/2 becomes a subdivision of h~/4. A similar modification is required for our
fixed collar Wt(O) ~ M x I: we expand it slightly to include all Nt(Bl-k) with
Bl-k C Wt(O). Of course we do not want to subdivide Wt(O), so we just let h~/2 ==
h~/4 on Wt(O). See figures (g) and (h).
(g) h~ (h) h~ '4 (heavy lines)
In the parametrized case we also have to take extra care in extending

h~/2 to N so that it restricts to a family of handlebody structures on the
handles of h~/4. This can be done by building the extension of h~/2 to N, one
handle of h~/4 at a time, using the second half of 8.2. Also, where a handle
of h~/4 collapses to a point, choose the extension of h~/2 to N to collapse
simultaneously to a point. The result of these modifications of h~/2 is a family
h~/3, homotopic to h~/2 since it differs only on N, and homotopic to h~/4 since it
is a subdivision of h~/4. Thus h~ ::::: h~/2, and similarly h~/2 ::::: h~. D
It would be interesting to know if the analogue of 8.1 is true in the
topological category (for manifolds admitting a topological handlebody
structure).
9. The parametrized PL h-cobordism theorem
A well-known representation for the classifying space of the simplicial
group PL (W) of (PL) homeomorphisms of a manifold W is the space of all
submanifolds of ROO homeomorphic to W. To make this more precise and to
give a relative form for PL(W, M), the homeomorphisms restricting to the
identity on a codimension zero submanifold M of aW, first let &( W, M) denote
the simplicial space of embeddings of Win Roo agreeing with a given fixed
embedding on M. Then PL(W, M) operates freely on &(W, M) by composi-
tion: fE PL(W, M) sends g E &(W, M) to go /-1 E &(W, M). The principal
simplicial fibration
128 A. E. HATCHER
PL(W, M) ~ &(W, M) ~ &(W, M) == &(W, M)/PL(W, M)

is universal for PL(W, M) since &(W, M) is clearly contractible. The space
&(W, M) is the representation of BPL(W, M) described at the beginning of
this paragraph.
Recall that a k-simplex of the simple homotopy space S(M) is a sub-
polyhedron Le Roo x a k such that the projection 1r: L -+ a k is a PL fibration,
and such that a fixed copy of M == M x {t} is a deformation retract of each
fiber 1r- 1(t). In particular, if (W, M) is an h-cobordism, simplices of &(W, M)
are simplices of S(M). Thus we have an inclusion of &(W, M) in S(M), in
fact, in the component of S(M) having torsion equal to that of the h-cobordism
(W, M). (Note that &(W, M) is always connected.)
THEOREM 9.1. The inclusion &(M x I, M) ~ S(M) is k-connected (onto
the identity component of S(M») provided dim M == n ~ 3k + 5. Consequently
the looping PL (M x I, M) -+ QS(M) is k-connected if n ~ 3k + ·8.
Of course PL (M x I, M) is just the PL pseudo-isotopy space, denoted by
9'(M) in the introduction.
COROLLARY9.2. The stabilization a: 9'(M) ~ 9'(M x I), a(f) == f x id r ,
induces an isomorphism on 1r k if n ~ 3k + 11 and an epimorphism if n ==
3k + 10.
Proof. Consider the diagram
&(M x I, M) c-_-+ S(M)
ul ul
&(M x I x I, M x I) ~ S(M x I)
where a takes Xc Roo to X x le Roo x le ROO x R == ROO. On S, a is clearly
a homotopy equivalence since the projection X x I -+ X is a contractible
mapping. D
We approach the proof of the theorem by giving another version of
BPL(W, M) in terms of handlebodies. PL(W, M) acts by composition on
the space §(W, M) of handlebody structures on (W, M), fE PL(W, M) carry-
ing hE §(W, M) to its image f(h) under f. Then using the diagonal action
we have a principal fibration
PL(W, M) ~ &(W, M) x §(W, M) ~ & x §(W, M) .
By 8.1, this is also universal for PL(W, M). For each (g, h) E &(W, M) x
§(W, M) there is the induced handlebody structure g(h) on (g(W), g(M»).
This is invariant under the diagonal action of PL(W, M), so we can interpret
& x §(W, M) as the space of handlebodies in Roo homeomorphic to W, col-

lared on the fixed M == g(M). Now let W == M x I and denote by X(M) the
subspace of IfJ x §(M x I, M) consisting of handlebodies whose collars agree
with a fixed collar. Note that the inclusion X(M) ~ & x ~(M x I, M) is
a homotopy equivalence. (A collar on M in Wc Roo is essentially a path of
embeddings M ~ Roo having a fixed initial point, and the space of such paths
is clearly, contractible.)
Given a handlebody structure h on (W, M) there is a well-known way
of associating to it a PL cell complex, in the sense of Section 4, by collapsing
each handle to its core and collapsing the collar M x I to M. More precisely,
a handle D i X Dn+l-i collapses to D i x {*} u aD i x Dn+l-i, so there are some
choices involved: first the point * E iJ n + 1 - i and then the actual collapse D i X
Dn+l-i --+ D i X {*} u aD i x Dn+l-i, which depends on choosing a small collar
C on aD i x aDn+l-i in D i x aDn+l-i to collapse to aD i x Dn+l-i (see the figure
below). However, the space of such choices is contractible.
Moreover, collapsing to their cores the handles of a k-parameter family

of handlebodies gives rise to a k-parameter family of cell complexes Lt as in
Section 4. Thus we have a map c: X(M) --+ S~(M), determined up to homo-
topy-the aforementioned choices made in collapsing a handle to its core.
Letting X{(M) c X(M) denote the subspace of handlebodies whose handles
O
have indices in the range [i, J we have a diagram

],
&(M x I, M) ~ S(M)
=1 . =1
X(M) _c~ S~(M)
J
X~+l(M) --~ S~+l(M) .
J
Since collapsing handles to their cores is a contractible mapping, the diagram
is homotopy commutative.
PROPOSITION 9.3. X~+l(M) ~ X(M) is k-connected if k +2~i ~ n -
k - 2.
130 A. E.HATCHER
PROPOSITION9.4. X~+l(M) ~ S~+l(M) is k-connected (onto the identity

component of S~+l(M») if 2i < n - k and n ~ 5.
These two propositions prove the theorem: choose i = k + 2 and use the
fact that S1+ 1 (M) ~ S~(M) is (i - l)-connected by 6.1.
The analogue of 9.3 in the smooth category, with handlebodies replaced
by eoo functions and gradient-like vector fields, is known only for small
values of k (see § 5.3 of [11] for the cases k = 1, 2). This is because the
behavior of singularities of eoo functions of higher codimension is not well
understood. In the PL category there is no such problem since, roughly
speaking, one can always just cone off by the Alexander trick--PL singular-
ities are very "flabby". Perhaps there is an alternate approach in the smooth
category which avoids the mysteries of eoo singularities.
To prove 9.3 it will suffice to show:
(*) X{+l(M) ~ X{(M) is k-connected if i ~ n - k - 3
since by working with dual handlebodies the same argument shows that
X{(M) C-.-.+ X{+l(M) is highly connected. The proof of (*) consists of trans-
lating the proof of 6.2 into handlebody terms. The main difficulty is to
deform attaching maps of cells into embeddings, which are needed in order
to attach handles. The technique for doing this is contained in the following
handlebody analogue of 6.3.
LEMMA 9.5. Suppose the handle D/ x Dt+ 1 - i in the family W t E X{(M),
tEak, collapses over aa k. If its attaching map rpt: Si-l X Dn+l-i --+ aWt(*) is
null-homotopic in W t(*) by a homotopy which also collapses to a point over
aa k and if i ~ n - k - 3, then the given i-handle can be 1·eplaced by an
(i + 2)-handle over a k.
Proof. The hypothesis on rpt gives a family Vt: D i+1 --+ W t such that, if
we write aD i +1 == D~_ U D~,
(i) Vt I D~ is a slice Dl x {x} for some x E aDt+ 1 - i ;
(ii) lJrt(D!:) C W t(*);
(iii) Vt(Di+l) is a point over aak.
We claim that lfrt can be improved to satisfy also
(iv) lfrt is an embedding into a level aWr) in each t-slice, t E ~k (with
the superscript ( .), like (*), depending on t);
(v) lfrt(aD i+l) n (D/ x Dt+ 1 - i) == D/ x {x}.
If we assume this, the given i-handle can be traded for an (i + l)-handle over
a k just as in the unparametrized case: Introduce a trivial (i + 1, i + 2)-
handle pair at the level aWr), use lfrt to isotope the attaching map of the
(i + l)-handle to be 'tt I aD i +\ then cancel the given i-handle with the new
(i + l)-handle, using 'tt I D!.. to isotope the i-handle to a trivially attached
handle.
Now we show how to achieve (iv) and (v). In general position 'tt(D-i,+l)
will be disjoint from the center points of all handles provided i < n - k.
Then we can push 'tt(D-i,+l) into Uz aWt(l). Next, working one handle D! x
Dt,+l- i at a time, in the reverse of the order of attaching, put 'tt in general
position with respect to the sphere {*} x aDt+ 1- i so that V't is an embedding
there (the singularity set of 'to ~t, is of dimension 2(i + k + 1) - (n + k) <
i ~ } == codim ({*} x aD/t+l- i ) assuming i ~ n - k - 3). Then ~t can be
pushed off D/ x Dt+ 1- i , and eventually down into M x {1}. Let It be the
union of the transverse pullbacks 'tt 1({*} x aDt+1-i) for the various handles
D/ x Dt+ 1- i . This has dimension ~ k + 1. We can assume that 'tt maps the
complement of a neighborhood N t of the cone clt C D;,+l into M x {1}. Now
replace 'tt by its restriction to the (i + l)-disc N t. (Even though clt may not
vary continuously in t, the neighborhoods N t can be taken to be a nice PL
family of discs in D i +1 .) This restriction to N t will be an embedding if we
make 2: t nclt == 0, which requires codim2: t == (n+k)-(i+k+1) > dim (clt) ==
k + 2, or i < n - k - 3. We can improve this to i ~ n - k - 3 since condi-
tion (iv) can be weakened to require only that 'tt be an embedding on each
concentric Si in D i +1 •
To push 'tt into a single level a Wt) we must avoid situations where V't
maps D;,+l across the top of one handle, which attaches across a second
handle, which attaches across a third handle, etc., the last handle in this
chain attaching to the image of V't again. The coincidences of 'tt(D i +l) across
a }l-handle across ... a }z-handle are, in general position, of dimension ~ i +
k - }z + 2 - l in the parameter domain ak • Over this part of a k the inter-
sections of 'tt(D i +l) with the attaching sphere of the }z-handle are of dimen-
sion ;£ 2i + k - n + 2 - l. So we can make clt disjoint from these chains of
coincidences (and hence excise them) provided (k + 2) + (2i + k - n + 2 - l) <
i + k + 1, or i < n - k - 3 + l. This finishes 9.5. D
To complete the proof of 9.3 we must kill the obstructions to the
attaching map of an i-handle being (slicewise) homotopically trivial. It was
shown in the proof of 6.2 how this can be done homotopically by adding
(i + 1)- and (i + 2)-cells. The technique in the proof of 9.5 for producing
embeddings suitable for attaching handles works just as well here, to show
that the obstructions can also be killed by introducing (i + 1)- and (i + 2)-
handles. D
132 A. E. HATCHER
Proof of 9.4. Let Lt E S1+ 1 (M) for tEak have a lifting W t E JC1+ 1 (M) for
t E aa k , which we wish to extend over a k • We can assume to start that
collapsing handles of W t are attached after all other handles, since in general
position the collapse points of Wo of dimension ~ k - 2 over aa k , will be
disjoint from (the core spheres of) attaching maps of non-collapsing handles
if (k - 2) + (i + k - 1) < n + k - 1, or i < n - k + 2. Also, over a k we
can deform collapsing cells of Lt to attach last. Now suppose inductively
that we have thickened into handles all non-collapsing cells of Lt up to a
level Li o), to form Wt) over a k, and let CPt: Si ~ Li o) ::::: Wt(o) be the attaching
map of the next (i + I)-cell (the case of an i-cell is similar). In general
position cpt(Si) will be disjoint from the cores of handles in Wt) if (i + k) +
(i + 1 + k) < n + k + 1, or 2i < n - k. Then ept can be deformed into aWt),
where handles are supposed to attach. Approximate CPt by an embedding
(if 2(i + k) < n + k, or 2i < n - k again). In order to use CPt to attach an
(i + I)-handle we need a trivialization of the normal bundle V(cpt(Si), aWt»).
(Note that we are in the stable range.) Let r t: Lt~Mbe a family of deforma-
tion retractions and consider ri(r(M)) where r(M) is the stable tangent
bundle of M. Assume inductively that we have an isomorphism ri(r(M)) F::d
r(Wt») on Wt). The cell e~+l attached via CPt gives a trivialization of ri(r(M))
on cpt(Si) and of r(cpt(Si)), hence also of V(cpt(Si), aWt»). So e~+l can be
thickened to a handle, at least away from where it collapses.
To turn collapsing cells into collapsing handles, choose a triangulation
of the set of collapse points and proceed by downward induction on the
simplices of this collapse set as follows. For an l-simplex a l (which we can
identify with its projection to the parameter domain a k ) one has given over
the boundary of its dual cell Dk-l in a k layers Wt(P) C Wt(q) , with the handles
of Wt(q) - Wt(P) to be coned off at a l n Dk-l. Excising all but a neighborhood
of these handles from Wt(P) leaves us with an h-cobordism (Vt, Dt) in each
t-slice of aDk-l since the corresponding cells of Lt do collapse at a l n Dk-l.
By the (simply-connected) h-cobordism theorem (Vt , Dt) F::d (Dt+ 1 , Dt) if n~ 5.
Moreover, since PL(Dn+l, Dn) is contractible by the Alexander trick, we
can choose trivializations of (Vh Dt) consistently for all t E aDk-l. This is
just what is required for the coning off of the handles of Wt(q) - Wt(P) over
Dk-l. This completes the induction step and the proof of 9.4, hence of 9.1. D
Compared with the proof of 9.3, the proof of 9.4 is rather crude, using
only general position to embed below the middle dimension. It seems that
the more delicate excision technique of 9.3 for deforming attaching maps
into embeddings in levels aWt> does not work in the relative case when
embeddings are given over Oak. Using the absolute case one can only show
that
7r kJCl+1(M) ~ 7rkS~+l(M)
is onto (and presumably split) if i ~ n - k - 2, hence that 7r k - l 9'(M) r:-:::3
7r k X(M) maps onto 7r k S(M) if n ~ 2k + 4.

10. A few examples
From the pseudo-isotopy theorem ([11], [16]) we have 2
7r 1S(K) r:-:::3 Wh 2(7r 1 K) EB Wht(7r K; Z2
1 x 7r 2 K) .
Let us consider the second summand Wht(7r 1 ; Z2 x 7r 2). Denote by (Z2 x 7r 2)[7r 1]
the additive abelian group of finite linear combinations E aiai for a i E Z2 x 7r 2
and a i E 7r 1. Then Wht(7r 1; Z2 x 7r 2) is just the quotient group of (Z2 x 7r 2)[7r 1]
modulo the subgroup generated by elements of the form aa - a'rar:- 1 and
(3 ·1, where a' denotes the usual action of rE 7r 1 on the 7r 2 component of a
and the trivial action on the Z2 component.
A one-parameter family Lt E S(K), t E [0, 1], representing a generator aa
of Whi (7r 1K; Z2 x 7r 2 K) has the form Lt == (K V Si) U <Pt e1+ 1, i > 2, the cell
e~+l being attached by the level slices rpt of a map "proj + aa": Si x I ~
K V Si which is the result of first pinching Si x I as in the following picture,
/-
"..--- .......
- ........ /----- , ........
J-----{
/----- ............... /..----- ........ ".}---------'
/------, /
"..---- ........
""
)----~
I x I Si+1
then, on Si x I projecting to Si c K V Si, on I x I wrapping each level

segment around a loop representing a'E 7r 1K, and on Si+1 taking the sum
a 1 + [a 2 , idsd, where a 1E 7r i + 1(Si) c 7r i + 1(K V Si) is the Z2 component of a and
[a 2, idsd denotes Whitehead product with a 2, the 7r 2K component of a. Thus
L o == L 1 and the family Lt is a loop in S(K). If a loop based at the natural
basepoint K E S(K) is desired, just connect L o to K by first pulling in the
2 It appears now (May, 1975) the proof in Part 11 of [11] (Part 11 is not joint with Wagonp:,-'
that the 7r2 part of the second obstruction is well-defined makes implicit use of an additional
hypothesis on K, namely, that the first Postnikov invariant k 1e H3(7r1K; 7r2K) of K is zero.
When k 1 is non-zero all that the arguments of [11] prove is that 7r1$(K) maps onto Wh 2(1r 1 K) EB
Whi(7r 1K; Z2) with kernel a quotient of Wht(7r1K; 7r 2 K).
134 A.E.HATCHER
"tail" where 90 wraps around a, then collapsing the resulting K V D i +l to K.

It is not too hard to figure out why loops Lt constructed according to
the above scheme for elements aa - a~!'a!,-l and ,e.1 in (Z2 x Jr 2K)[7r 1K]
should be homotopically trivial in S(K). But to prove that this is exactly
the kernel of the resulting map (Z2 x Jr 2K)[Jr 1K] -+ Jr1S(K) is a real task; this
is essentially the content of Part 11 of [11]. Computing the kernel of the k-
parameter analogue Jrk(QK) -+JrkS(K) looks like a very tough problem.
Now let us compute the terms in the exact sequence of 3.2:
(1) ~ Jr 1 G(K) ~ Jr1S(K) ~ Jr1S k ~ JroG(K) -----+ JroS(K) ~
in the case K == SI. Since SI is a K(Z, 1) it is easy to check that G(S1) ~

0(2), the orthogonal group. Thus SSl -+ BG(SI) has a section. Also, Wh 1 (Z)
and Wh 2 (Z) are known to vanish. So (1) becomes, for K == SI,
( 2) ~ Z ~ Z2[t, t- 1] ~ Z2[t, t-
1
] x Z +------- Z ~0
Z2[1] Z2[1] T 2 ~ 2
where T denotes the twisted product in which conjugation by the generator

of Z2 interchanges t and t- 1 •
It is instructive to compare this with the case K == S1 V Si with i > 2.
Clearly Jr OG(SI V Si) f:::::J Z2 X Z2' generated by the degree -1 maps on S1 and
Si. Also, one can check that 7r 1 G(Sl V Si) ~ Z2[t, t- l ], generated by the loops
V't == id s l V 9 tJ where 9t is the family of maps Si -+ S1 V Si described above.
Lemma 10.1 below says that 7r 1G(SI V Si) -+ 7r l S(SI V Si) is surjective, so
the sequence (1) becomes now
( 3) -----+
proj
Z2[t, t- 1 ] ~
Z [t t- 1]
2, ~
0 =
Z2 X Z2 ~ Z2 X Z2 ~ 0 •
Z2[1]
Thus, although 7r*S(K) depends only on the low dimensional skeletons of K,
the maps in (1) depend on all of K.
The interpretation of (2) and (3) in terms of homeomorphisms of Hilbert
cube manifolds (see the introduction) is somewhat curious. On SI X Q there
are many, many homeomorphisms which are homotopic but not isotopic, but
on (S1 V Si) X Q for i > 2 there are none.
LEMMA 10.1. The image of Jr 1G(K V Si) -+ JrlS(K V Si) contains the
> 2.
summand Wht(Jr l ; Z2 x Jr 2 ) i f i
Proof. The map G(K) -+ S(K) sends f: K -+ K to its mapping cylinder
M(f) == (K x [0, 1] U K)/(x x 1 f(x») , considered as lying in S(K) by identi-
t'J
fying K with K x 0 c M(f). In the case at hand V't: K V Si -+ K V Si is the

identity on K, so we may collapse K x [0, 1] c M(V't) to K x 0 by a con-
tractible mapping. The resulting Lt has the form Lt == (K V Soi V S/) U rte~+1,
where S~ and Sli correspond to Si X 0 and S' x 1 in M(Yt) and 'Y t is a level
slice of the map
projo + proj1 + (aa)l: S' x I ~ K V Soi V Sli •
Now we consider Sf as a trivially attached i-cell e~, and, simultaneously for

each t E [0, 1], we deform its attaching map across S;, producing again a
trivially attached e:.
But on examining what has happened to 'Y t we find
that projo + proj1 + (aa)l has changed to (aa)o + proj1 + (aa)l. The follow-
ing diagrams depict this change at t = 0 (or 1), when L o changes to K V
Soi V D'+l.
We can assume that (aa)o is added to proj1 in the t-interval [0,1/2] and that
(aa)l is added in [1/2, 1]. Then fort E [0, 1/2] the pair (ef+t, eD can be collapsed
by an elementary collapse. The remaining Lt for t E [1/2, 1] is just the family
called Lt at the beginning of this section, with Kreplaced now by K V Si. D
Finally, we show that 1I: i S(K) is a direct summand of 1I: i +1S(K x Sl),
giving a cheap way to mass produce elements of 1I:*S( -) from the known
models in 11: 0 and 11: 1 • (The technique is also known, independently, to Hsiang-
Sharpe and Burghelea-Lashof-Rothenberg.)
PROPOSITION 10.2. For a compact manifold M, fY(M x I) is a homotopy
retract of QfY(M x Sl). Hence S(K) is a homotopy retract of QS(K x Sl).
Proof. Let us change notation slightly and let fY(M) denote pseudo-
isotopies M x I ~ M x I fixed on M x 0 U aM x I, rather than just on M x o.
This new fY(M) is homeomorphic to the old one since (M x I, M x 0) r::d
(M x I, M x 0 U aM x I). Also, for M a proper submanifold of some V
(Le., M n aV == aM) denote by 9'(M, V) the space of proper embeddings
M x I ~ V x I agreeing with the given M ~ V on M x 0 U aM x I, and
such that M x 1 ~ V x 1. We will be interested in the following diagram:
Oll .
Q9'(M X Si) - - - 4 Q9'(M x R)
0P1
~T
'"
9'(M
/~
x I)
/
~ QP2
/
/ Ql
"'~
(J~
2
Q9'(M, M x Si) ~ Q9'(M, M x R) .
The maps ~1 and ~2 are obtained by lifting to the cover M x R ~ M X Sl,
136 A.E.HATCHER
while the fibrations P1 and P2 come from restriction to M x 1 c M X SI and

°
M x c M x R. The fiber of P2 is clearly contractible, so Qp2 is a homotopy
equivalence. The fiber of P1 can be identified with g>(M x I), and via this
identification, 1: is defined by taking lE g>(M x I) c 9'(M x SI), forming the
commutator 8f8- 1f-l where 8: M x 8 1 X I ~ is rotation through the angle
8 E 8 1 , then considering OfO-1f-1 as a loop by letting () run around SI. The
two unlabelled maps and a are defined to make the diagram commute.
For example, a is restriction of a pseudo-isotopy (M x I) x I ~ to the
family of slices (M x 8) x I c (M x I) x I c (M x R) x I for ° ~ 8 ~ 1,
followed by the translation x ~ x - 8 in the R direction. With this descrip-
tion we recognize a as a section for the map a in the following fibration
sequence (identifying (R, 0) with (0, 00), 1)):
Q9l(M, M x (0, 00») ~ 9l(M x I) -----+ E ~ 9l(M, M x (0, 00») .

Here E is the space of proper embeddings M x [0, 1] x I ~ M x [0, 00) x I
° °
which are the identity on M x x I U M x [0, 1] x U aM x [0, 1] x I and
carry M x [0, 1] x 1 to M x [0, 00) x 1. The map p is restriction to M x
1 x I; that p is a fibration depends crucially on the compactness of M.
Since E is evidently contractible (essentially by restriction to the seg-
ments M x [0, 8] x I), a and a are homotopy inverses, and we have shown
that 9'(M x I) ~ Qg>(M x R) is a homotopy retract of Qg>(M x Sl). These
constructions respect the stabilization M ~ M x I ~ M X 12~ · · · , so passing
to the limit we get QS(M) a homotopy retract of Q2S(M x SI). To deloop
this we appeal to [14] (which inspired the present proposition) for the asser-
tion that 7r oS(M) F::::J Wh 1 (1C 1M) is a direct summand of 1C 1S(M x SI) ~
1C o9'(M X SI). D
Remarks. 1. The proposition holds also in the smooth and topological
categories, with the same proof.
2. The pseudo-isotopy spaces g>(M) and 9'(M, V) can be replaced by the
spaces of homeomorphisms M ~ M and embeddings M -+ V, both reI aM.
3. The map ~2 has a section induced by embedding R in SI. Burghelea-
Lashof-Rothenberg have shown that P1 also splits. This is clear if M == N x
I for some N-an inclusion I x Sl C I x I gives N x I X Sl C N x I x I,
hence a projection of 9'(N x I x SI) onto the fiber of Pu 9'(N x I x I). Thus
we get a formula
QS(M X SI) ~ Q,S(M) x S(M) x Qm(M)
where m(M) is the fiber of the stabilized ~2. Presumably the relationship of
this to a well-known formula in algebraic K-theory for K*A[t, t- 1] is more
than coincidental.
PRINCETON UNIVERSITY
REFERENCES
[ 1] E. AKIN, Transverse cellular mappings of polyhedra, Trans. A.M.S. 169 (1972), 401-438.
L2] D. BURGHELEA and R. LASHOF, The homotopy type of the space of diffeomorphisms,
Trans. A.M.S. 196 (1974), 1-50.
[~3] T. A. CHAPMAN, Cell-like mappings of Hilbert cube manifolds: applications to simple
homotopy theory, Bull. A.M.S. 79 (1973), 1286-1291.
H4] - - - , Topological invariance of Whitehead torsion, Am. J. of Math. 96 (1974), 488-
497.
[5] - - - , Concordances of Hilbert cube manifolds, preprint.
[6] - - - , Homotopic homeomorphisms of Hilbert cube manifolds, Proc. Park City (Utah)
Topology Conference 1974, to appear.
[7] M. M. COHEN, Simplicial structures and transverse cellularity, Ann. of Math. 85 (1967),
218-245.
[8] - - - , A course in simple-homotopy theory, Springer-Verlag, 1973.
[ 9] A. DOLD, Partitions of unity in the theory of fibrations, Ann. of Math. 78 (1963), 223-
255.
[10] B. ECKMANN and S. MAUMARY, Le groupe des types simples d'homotopie sur un polyedre,
in: Essays on Topology and Related Topics, ed. A. Haefliger and R. Narasimhan,
Springer-Verlag, 1970, 173-187.
[11] A. HATCHER and J. WAGONER, Pseudo-isotopies of compact manifolds, Asterisque 6 (1973).
[12] J. P . MAY, The geometry of iterated loopspaces, Springer Lecture Notes 271, 1972.
[13] L. C. SIEBENMANN, Infinite simple homotopy types, Indag. Math. 32 (1970), 479-495.
[14] - - - , Torsion invariants for pseudo-isotopies of closed manifolds, Notices A.M.S. 14
(1967), 942.
[15] R. STOCKER, Whiteheadgruppe topologischer Raume, Invent. Math. 9 (1970), 271-278.
[16] 1. A. VOLODIN, Generalized Whitehead groups and pseudo-isotopies, Uspekhi Math. Nauk
27 # 5 (1972), 229-230.
[17] J. H. C. WHITEHEAD, Simple homotopy types, Amer. J. of Math. 72 (1950), 1-57.
[18] K. IGUSA, Princeton Ph. D. Thesis, 1975.
(Received March 19, 1975)

Higher Simple Homotopy Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Higher Simple Homotopy Theory

Uploaded by

Copyright:

Available Formats

Annals of Mathematics, 102 (1975), 101-137

Higher simple homotopy theory

the identity on K. One of several equivalent definitions of Wh (K) is the

if n ~ i. With results of Chapman, this leads to a neat reformation of higher

g>Top(K x Q) ~ Homeo (K x Q) ~ G(K x Q) ~ G(K x Q)/Homeo (K x Q)

1 Assuming the first Postnikov invariant k 1 e H3(7r 1 M; 7r 2 M) of M vanishes-see the foot-

This section contains preliminary material on fibrations in the PL

homotopy property in the diagram

Define g:: X - go-l(V) ~ C(A) - {v} to be gs if s ~ 0 and go if s ~ o. This lifts

homotopic to fi (in fact, by a homotopy which is arbitrarily small with respect

Returning to assertion (H), consider first the unfibered case A == point.

fibration if over one-dimensional subpolyhedra of B it is a PL fibration.

with !, f', and g contractible mappings. To check that g U </Jf is a contractible

via triangulations of Land a k such that L -+- a k is simplicial, thereby obtain-

t E ~k, and to think of Lt as a k-parameter family of polyhedra.

Definition. For a fixed polyhedron Kc Roo, let SK be the subcomplex

PROPOSITION 3.2. The homotopy fiber of SK ~ BG(K) is S(K).

G(K) -----+ S(K) -----+ SK -----+ BG(K)

is just the well-known representation of the set 1COS K of simple homotopy

PROPOSITION 3.3. S(K) is a covariant functor of K, from the homotopy

COROLLARY 3.4. If K is contractible then so is S(K).

Another one-parameter family, demonstrating a homotopy of attaching

As this example illustrates, in parametrized settings it is unreasonable to

Remarks. (1) If the normalization condition (iii) were not assumed, it

LEMMA 4.2. Let the family Li E Sc;(K) be a subdivision, of the family

pendent of t, up to isotopy, so n:' is a fibration near x.

PROPOSITION 5.1. The map ~ + id: S(K) ~ S(K) is null-homotopic.

gives a homotopy C ~ L + id in S(K), by 2.4. But also C ~ 0 from the

for some j ~ n + 1. By 6.2 below, Lt can be deformed to a family L~ E

9t simultaneously for all tEak by attaching new i + 1-cells to Li o

First we specify more closely the subcomplexes Li in which of course

We want to choose a subdivision Tof a k such that each cell of Lt is contained

the cells collapsing to p also lie in L~o).)

(l-1)-parameter family (D/+\ OD/+l)~(Lh Li o»), t E Dl-l, where N == D l- 1 X

;Po and hence the proof of 6.2. D

induced by inclusions is surjective if k ~ m +n- 2 and injective if

over finite subpolyhedra K; c K'.) The proposition does not extend to m == 1

disc D oi +l with an annulus Si x [0,1], Si X 1 == aD i + 1 • Letf: Si x a k -+ [0,1]

to E ~k, suppose that over the boundary of a small neighborhood of to in ~k

.>-&- J..-.A..-< .•• .iv!xl/2

c.- -J' •••• MxO

The reason for permitting collapses in §( W, M) more general than the

(2) If the i-handle n(N(a'») is written as D i x Dk-" then over

LEMMA 8.2. ~(Dn+l,~) and ~(Dn+l, Dn) are contractible.

Returning to the main argument, the family h~/2 as constructed is not

(g) h~ (h) h~ '4 (heavy lines)

In the parametrized case we also have to take extra care in extending

PL(W, M) ~ &(W, M) ~ &(W, M) == &(W, M)/PL(W, M)

& x §(W, M) as the space of handlebodies in Roo homeomorphic to W, col-

Moreover, collapsing to their cores the handles of a k-parameter family

have indices in the range [i, J we have a diagram

PROPOSITION9.4. X~+l(M) ~ S~+l(M) is k-connected (onto the identity

is onto (and presumably split) if i ~ n - k - 2, hence that 7r k - l 9'(M) r:-:::3

7r k X(M) maps onto 7r k S(M) if n ~ 2k + 4.

/----- ............... /..----- ........ ".}---------'

then, on Si x I projecting to Si c K V Si, on I x I wrapping each level

"tail" where 90 wraps around a, then collapsing the resulting K V D i +l to K.

in the case K == SI. Since SI is a K(Z, 1) it is easy to check that G(S1) ~

where T denotes the twisted product in which conjugation by the generator

fying K with K x 0 c M(f). In the case at hand V't: K V Si -+ K V Si is the

Now we consider Sf as a trivially attached i-cell e~, and, simultaneously for

while the fibrations P1 and P2 come from restriction to M x 1 c M X SI and

Q9l(M, M x (0, 00») ~ 9l(M x I) -----+ E ~ 9l(M, M x (0, 00») .

You might also like