TOPOLOGY

AND ITS

APPLICATIONS
ELSEVIER Topology and it< Applications 77 (1997) I-18

Comultiplication
J. Klein,
Fakultiit Fuchbereich fiir Mathematik, MathematWInformatik,

and suspension
R.M. Vogt *
D-33615 Bielefeld, Germany German) D-49069 Osnabriick,

R. Schw&xl,
Universittit Bielefeld, Universitiit

Osnabriick.

Received 8 August 1995: revised 26 February 1996. 24 July 1996

Abstract
We prove the conjecture of Berstein, Hilton, and Ganea that an (n - I)-connected CW-complex X with dim X < k(n - 2) + 3 which has a co-Ak_,-space structure desuspends. Moreover, we show the following dual of a result of Segal: given a cosimplicial space Y, which is special in the sense that the canonical maps Vz, Yr + k;, are homotopy equivalences and Yl is 2-connected then there is a functorial desuspension of Y,. 0 1997 Elsevier Science B.V.
Keywords: AMS

Co-Ak-space;

Desuspension;
55P40

Cosimplicial

space; Homotopy limit

clussijication:

551345;

1. Introduction It has been known for a long time that an (n - 1)-connected CW-complex of di-

mension 6 2n - 1 has the homotopy type of a suspension. In 1963 Berstein and Hilton proved that an (n - 1)-connected based CW-complex X of dimension 6 3n - 3 has the homotopy type of a suspension provided X admits a comultiplication X + X
V

X with

homotopy counit [l] (Berstein and Hilton made the additional technical requirement that the homology groups of X are finitely generated). In 1970 Ganea extended this result: an (n - I)-connected co-H-space X of dimension < 4n - 5, n > 2, has the homotopy type of a suspension CY if and only if it is homotopy coassociative. Moreover, the homotopy equivalence CY + X is a homomorphism up to homotopy [5]. Berstein-Hilton and Ganea conjectured that a suitable theory of co-A,-spaces, dual to Stasheffs theory of An-H-spaces [13], would lead to extensions of their results. Saito has developed such a theory in part and managed to extend the results to co-Ad-spaces of dimension < 5n - 7 [ 1I]. In Stasheffs An-space situation the following result holds [14,17].
* Corresponding author. E-mail: rainer@chryseis.mathematik.uni-osnabrueck.de. 0166.8641/97/$17.00 0 1997 Elsevier Science B.V. All rights reserved.

PIISOl66-8641(96)00103-4

J.

KIein er al. / Topologyand its Applications77 (1997) 1-18 well-pointed Ak-l-space, k 3 3, adis the

Proposition 1.1. Let X be an (n - I)-connected

= 0 for i > k(n + 1) - 4. Then there is an Here QMY

f : X + AMY

which is a weak equivalence.

Moore loop space on Y with its natural monoid structure. Recall that an Ak-homomorphism is a homomorphism homotopies. up to homotopy satisfying

coherence conditions up to (k - 1)-dimensional is to prove the foilowing dual result.

The purpose of this paper

Theorem 1.2. Let X be an (n - 1)-connected based CW-complex with dim X < k(n 2) + 3, n > 2. Then X is of the homotopy type of a suspension if and only if X is a co-ilk-l-space. Moreovel; the homotopy equivalence f : CY + X can be chosen to be a co-Ak- 1-homomorphism. Corollary 1.3. A 2-connected space X is of the weak homotopy type of a suspension if
and only if it is a co-Am-space. We then use our approach to prove a dual version of Segals delooping result [ 121

Theorem 1.4. Let X : A + Top* be a cosimplicial

based space such that X0 is con-

tractible, XI is 2-connected and (~1, . , 7riTn) : VT=, Xl + X, is a homotopy equivalence where ri : [l] + [n] maps 0 to i - 1 and 1 to i. Then there is a weak homotopy equivalence ERX + X1, where RX is a&nctorial, thickened version of the topological corealization of X. that

A result of this type has been formulated by Hopkins [7] with the weaker condition

XI is only l-connected. He claimed that the result follows from the homology spectral sequence of the cosimplicial space. In his analysis of this spectral sequence Bousfield addresses Hopkins result [3, (4.9)] but he requires that the corealization of X is nilpotent. This is certainly the case if X1 is 2-connected, and we cannot see that 1-connectedness suffices. Instead of the maps rk Hopkins uses the equivalence

(11,

L,):

\lix, 7
k=I

x,
space in his sense induces

where Lk : [l] +

[n] maps 0 to 0 and 1 to k. A cosimplicial

one in our sense by a base change. The approach to desuspension which we develop here is motivated by potential applications. A fibrant variant of the main result of this paper (see [9]) will be used by the first-named author in a subsequent paper to prove a Haefliger-type embedding theorem in the category of Poincar6 duality spaces [8]. Such embedding and desuspension results also play a fundamental role in the original PoincarC surgery program as formulated by Browder et al. It is hoped that the results of this paper can be adapted so as to complete the Browder program.

.I. Klein et al. / Topology and its Applications 77 (1997) l-18

While Ganea and Saito make use of the comonad more in the spirit of Berstein-Hilton our methods differ. They carefully requires the assumption set topological difficulties who construct

En

in their proofs our proof is directly, although (which of the spaces involved

the desuspension

analyze the homology by Goodwillie

that these groups are finitely generated)

while we use the homogenerated spaces in

topy theory of cubical diagrams developed

[6]. To stay away from point unless stated otherwise.

we work in the category 7%~ of compactly

the sense of [15]. Based spaces are assumed to be well-pointed

2. Al,-spaces

and co-ilk-spaces of an Ak-space using the terminology of all formal expressions of [2]. X, : n + 1

We recall the definition

Let G be the category with ob 6 = IV = (0, 1, . . .}, exactly one morphism for n 3 1, and G(n, Ic) consisting xi, @ . 13 xi, ) i,+...+il,=n

G(O,O) = {ido} and G(O, n) = 0 = G(n, 0) f or n > 0. Composition requirement that @ is a bifunctor. n * G(n)

is defined by the

The category G describes semigroups: a semigroup G determines a functor 4 + 7op, Gn, transforming $ into x. Conversely, if G : 6 + Top is a functor such that = G( 1) and G(f@g) X,, = G(f)xG(g) then G( 1) 1sa semigroup with multiplication XI,. . , .c, and one output, the are obtained by wiring boxes together to a tree

G(,&). We symbolize product x1 2,.

n 3 2, by a box with n inputs

Composite shaped circuit, e.g., 2.1.

operations

Example

represent all possible nontrivial composite operations 3 + 1. In 6 they coincide but for homotopy associative multiplications they only coincide up to homotopy. To account for this we give each connection (between boxes) a length t E I = [0, 11.Given a tree T the various lengths of its connections make up the points of a cube C(T) whose dimension is the number of connections of T. We thus obtain a new category 78 with ob 74 = ob G and 76(n, 7G(n, 1) = u C(T), k) = fl7E(ii, T running through all trees with n inputs, withil+...+ik=n.

1) x...xTG(ik,l)

J. Klein et al. / Topology and its Applications

77 (1997) I-18

Let T be a tree with lengths in 7G(Ic, 1) and let (T,, . . , Tk) E TG(n, k) be a Ic-tuple of trees; then T o (T,, . . , Tk) is the tree obtained from T by sticking Ti onto its ith input and giving the newly created connections idr . We have a continuous (Tr,. . . ,Tk) CE(Tk+l,. Composition the lengths 1. The trivial tree serves as

pairing 63 : 7-G x 7-G + 7-G defined by . . ,Tk+l) = (TI,. . ,Ttc+~). by the requirement that 69 is a bifunctor. It is continuous.

then is determined

Let WG be the quotient category of 7-G obtained by imposing the relation that a connection of length 0 may be shrunk by amalgamating the two boxes at its ends to a single one. Observe that the trees in Example WG(3, 1) which are identified codify an associating Definition homotopy. of WG of objects 0, 1,. . , k. An Ak@ to x (in particular X(n) = 2.1(l) and (3) represent unit intervals in 2.1 at 0 with the tree (2). Hence the trees of Example

2.2. Let M&G be the full subcategory

space is a continuous

functor X : V&G + Top mapping

X( l)n) such that the multiplication X(X,) :X(l)* + X( 1) admits a homotopy unit. (Note that we do not require any coherence for the homotopy unit.) By abuse of notation we often denote X(1) by X. Remark 2.3. The map E : WG + G which maps a tree with n inputs to X, is a continuous functor admitting a nonfunctorial section r] : G + WG mapping X, to the tree with a single vertex and 7~inputs if n > 1 and to the trivial tree for 72= 1. As maps of morphism spaces n o E N id (the homotopy shrinks the lengths of connections by the factor t at time t). It can be shown that PVG(n, 1) is a subdivision of the Stasheff cell Kn of [13] into

cubes. So our definition coincides with the one of Stasheff with the exception that he makes the stronger requirement that X(X,) admits a strict unit e. But if {e} C X is a closed cofibration Co-Ah-spaces Definition both definitions are defined dually: functor X: (V&G)! + I$* into the are equivalent [2, Chapter I].

2.4. A co-Al,-space is a continuous

category of well-based topological spaces mapping @ to the wedge sum V such that the comultiplication X(&) :X( 1) + X( 1) V X( 1) admits a homotopy counit (i.e., pi o X(X2) Y id, where pi :X(l) V X(1) + X( 1) is the ith projection, i = 1,2). Again we often write X for X( 1). Example 2.5. Let C be an operad without permutations to C a category C with ob C = IV, C(lz, k) = u C(ii)
x . . x

in the sense of [IO]. We associate

C(ik)

with ir +. . . + ik = n and all ij > 0.

C(O,O) = {ida} and C(0, n) = C(n, 0) = 0 if n > 0. Taking products defines a pairing @ : C x C + C. Composition C(k, 1) x C(n, k) + C(n, 1)

.I. Klein et al. / Topology and its Applications 77 (1997) I-18

is defined by the structure map y of the operad. General composition by the requirement B-preserving that $ is a bifunctor. Call C an Ak-operud if C(j) there is a homotopy A co-C-space

then is determined

2 * for j < k. If C is an Arc-operad, there is a continuous $-preserving functors from Fc: to F&.

functor FC : Wr;G + C by [2, 3.171. If F& is another such functor then through continuous of maps is a collection i

bLIL : x A c(n)+ (n-fold

x v . . v x = XV = C(n) u {*}, such that ~1 (x: 1) = z and

wedge) where C(n)+

. A C(ik)+

pLi;Aid

A c(i,)+

A . . A c(i,)+ pr4 1 . . . v (X A C(ib)f)

Y 1 x A (C(k) x C(i,) x .
idAy

x c(i,))+

(X A c(i,)+)

1
x A c(n)+

ILn

1 * xvn

commutes.

It determines

a co-C-space,

i.e., a continuous

functor CP _j 7op* mapping ~f3to V.

@ to V, hence a continuous If C is an Ak-operad, a co-ilk-space.

functor (Wk@P + fip*

mapping

k 3 2, then any co-C-space

has a homotopy counit. Hence it is

Example 2.6. Let Q be the little structures. 0 6 Xl < yl 6 22 < embedding of n. intervals
y2 < .'. <

l-cubes
2, <

category
y, < ,

[2, 2.531 describing

loop space as an

A point in Q(n, 1) is a tuple (XI, ~1,. . , cc,, gn) of points in 1 such that 1, and should be considered [zz, yi] into I such that the images have disjoint interiors. The in Q define the little cube operad C, of [lo]. and hence a co-Am-space with

spaces Q(n, 1) and the composition

Let Z be a based space. Then CZ is a co-&-space CL, : cz A Q(n, I)+ + cz v . . v cz

E kth summand
if t E [Q~Y~],

otherwise.

Remark 2.7. In their treatment of multiplications on topological spaces Boardman and Vogt [2] and May [lo] have homotopy units incorporated into the structure as 0-ary operations or points in the 0th space C(0) of an operad C respectively. We here merely require homotopy units outside the structure, i.e., their existence without specifying a particular one. This weaker condition suffices for our purposes and simplifies our argument.

J. Klein et al. / Topology and its Applications 77 (1997) I-18

3. Cubical diagrams For the readers convenience and 7, the full subcategories we recall Goodwillies results about cubical diagrams. and Jn 0. For n = 0 we

Let &, denote the category of all subsets of 24 = { 1, . . . , n} and inclusions, of all objects except for 24 respectively (~1, . . . , E,) where

take n = 0. A diagram D : Q, -+ 7%~ is a cubical diagram of based spaces with D(T) sitting at the comer with coordinates Ei = Dl& 1 0 if i ET. ifi$!T, of all proper initial and terminal faces respectively.

and 017, are the subdiagrams

Definition 3.1. D is called k-Cartesian if the canonical map D(0) + holim 017, is k-connected, and k-co-Cartesian if the canonical map hocolim Df& + D(n) is kconnected. For T c n let aTD be the restriction vertex 0 and terminal of D to the ITI-dimensional the restriction face with initial

vertex T and ~,_TD

of D to the IT/-dimensional

face with initial vertex 14- T and terminal vertex 21 ((T( denotes the number of elements of T). Proposition 3.2 (Goodwillie [6]). Let D: &, + Top*, n 3 1, be a diagram such that

(i) aTD is c(T)-co-Cartesian for all T c Q, T # 0,

(ii) c(U) 6 c(T)for U c T. Then D is k-Cartesian with

Proposition 3.3 (Goodwillie [6]). Let D : &, -+ Top*, n 3 1, be a diagram such that (i) ~,_TD is k(T)-Cartesian for all T c Q, T # 0, (ii) k(U) 6 k(T)for U c T. Then D is c-co-Cartesian with c = Min n - 1 + c k(T,); a {Tcy} is a partition of 21into T, # 0 .

Remark 3.4. We will have to apply these results to coherently homotopy commutative diagrams in the sense of Section 5 rather than commutative ones. Fortunately they hold in this case too (e.g., a stronger version of [6, Theorem 2.41 for homotopy commutative cube diagrams can be found in [18, Proposition 5.51). Indeed, by [16, Proposition 5.41 each coherently homotopy commutative cube is equivalent to a strictly commutative one, and the equivalence induces an equivalence of their homotopy limits and colimits.

J. Klein et al. / Topology and its Applications

77 (1997)

l-18

4. The Berstein-Hilton

result 1.2 we illustrate co-H-space, our strategy by reproving n 3 2, with comultiplication the Berstein-

Before we prove Theorem Hilton result.

Let X be an (n - I)-connected /IX Consider -+xvx. the diagram

of based spaces (translating

into cube coordinates to (1.1. 1)):

P3 = II(@)

corresponds

to (0; 0,O) and X V X = D(J) corresponds

Diagram 4.1.

where ik is the inclusion

of the kth wedge summand.

The maps it, i2, ~1 and j define a

diagram 7; + %p*, which we complete to a &-diagram by taking the homotopy limit 9. Note that P3 can be obtained by taking iterated homotopy pullbacks: let RZ be the homotopy pullback of

xAxvx&x. and P2 the one of * + X t * (hence P2 is the loop space QX and R2 the cojoin pullback:

X c X); then Ps is the homotopy Diagram 4 h

4.2.
p2 I

The bottom face of Diagram 4.1 is a diagram from Q2 to 7%~ which is co-Cartesian (meaning cc-co-Cartesian) and j is (n - I)-connected. By Proposition 3.2 the map

8 j : * + pullback connected.

.I. Klein et al. / Topology and its Applications 77 (1997) I-18

R2 and hence

R2 is (2n - 3)-connected.

Since Diagram

4.2 is a homotopy In particular, Pa is

diagram

the map P3 +

P2 is also (272 - 3)-connected.

Let M3 be the homotopy of * t P3 + * and X t

colimit of Diagram 4.1 restricted to 33. We can obtain A43 since CP, and X V X are the homotopy pushouts we have a homotopy pushout diagram: * + X respectively,

by taking iterated homotopy pushouts:

Diagram 4.3.
CP3AX 41 I XVX-=-+M3 I

where r2 is the map induced by the top face of Diagram 4.1. By construction, the cube of Diagram 4.1 is Cartesian. Each terminal 2-dimensional face is co-Cartesian (here we need that X is a l-connected co-H-space) and hence (2n - 3)-Cartesian by Proposition 3.2 since j is (n- I)-connected. Since CL,ii and i2 are (n- I)-connected the cube is (3n-2)co-Cartesian by Proposition 3.3. In particular, the induced map rg : A43 + X V X is (3n - 2)-connected. Since r3 o ga N id, the map g3 is (3n - 3)-connected. Hence r2 is homology (3n - 3)-connected because Diagram 4.3 is a homotopy pushout. Since CP3 and X are simply connected r2 is homotopy (3n - 3)-connected. We summarize Proposition 4.4. Let X be an (n - 1)-connected
CK N X. CW-complex Hence with a co-H-space strucK

ture, n 3 2. Then CP3 -+ X is (3n - 3)-connected. and a homotopy equivalence

there is a CW-complex

For the second part we refer the reader to the proof of Theorem Section 6.

1.2 at the end of

5. Cubical diagrams

of co-Ak-spaces

In this section we construct Diagram 4.1 in the general case. The category WG of Example 2.1 only codifies the coherence of comultiplications. To incorporate inclusions such as ii, i2 : X -+ X V X or j : * + X we have to enlarge it to a category W*G. We define ob: W*G = ob WG and
W*G(n, k) = fl WG(l, k) x Inj(l, rz)

where Inj(l, n) is the set of ordered injections

1 + n. In particular,

W*G(n,

0) consists

of a single element (ida, 0 + 22). Composition of (f, a) E WG(l, k) x Inj(l, n) with (91 @ . . . B sn, P) E WG(m, n> x Wm, P>

J. Klein et al. / Topology and its Applicarions 77 (1997) I-18

is defined by (f: a) O (9 @ . . @ gnr a) = (f 0 (LL(l) cg . . 83 SC?(l)),0 0 N(ml Here gi E WG(mi, l), m = ml +. . . + m,, and .
>

mn)).

a(ml,
is the injection 1 < j < m,(q,

. . m,) : m,(l) sending m,(,) 1 < i < I.

+ . . . + 7r~l,(~)+ m
+

. . . + m,(i_,i

+ j to ml + m2 + . . . + m,(i)_1

+j

for

The bifunctor

@ extends to W*G by

(f, a) @ (9, P) = (f @ 9, QS u P) where a U /3 is the ordered disjoint union. Let PVlG denote the full subcategory of W*G consisting of all objects j < Ic. A coAl,-space X : (WkG)P + Ti~p* extends to a continuous functor X : (w;4)p mapping @ to V by xv1 -+ yn * Tilp*

Xc(f, o,) : XV x3

for (.f,a) E WG(l,k) x Inj(l, n ), w h ere (Y* maps the ith summand identically onto the a(i)th one (checking of functoriality is left to the reader). In an analogous way we can adjoin the injections to 4 to obtain a category G*. We have the following fact: functor E : WG + G and its section r] : G + WG

Proposition 5.1. The augmentation

to afinctor

E*

is a homotopy

equivalence

: W*G -+ 4* and a nonfinctorial section q* : 4* + W*G. Again on morphism spaces with homotopy inverse q*.

For a co-H-space we started with the commutative x-diagram 4.1. For a co-Ak_ I space with k - 1 3 3 associating homotopies enter the picture, and we only can hope for a coherently homotopy commutative 7k-diagram to start with. To describe this we need are codified by the topoa variant of Example 2.1: Let 27 be a small category. Homotopy logical category WD WD(A, R) =

coherent D-diagrams

[2, Chapter VII] with ob WD = ob 2) and u Dn+, (A, B) x I /> n . . , fo) of morphisms in V such that

where D,+, (A, B) is the set of (n + 1)-tuples (fn, fn o . . . o fo : A + B is defined. The relations are

if fn = id,

(.fn,tn,..

~,f,,t,,fo)=

(fn, . . . . fi+l,max(ti+l,ti),fi-l,...,fo) (fn: tn,. t2. fl) (fm fL+I 0 ft. ti, f >fo)
> . ,

if.fi=id,
if fo = id, if ti+, = 0.

10

.I. Klein et al. / Topology and its Applications

77 (1997) 1-18

Composition

is defined by = (fn,t7L,...,fO,l,gp,Up,...,go)

(~~,tn,...,fo)O(gp,Up,...,go)

WD can be viewed as obtained from YDby taking the free category over the diagram 2) and putting back the relations up to coherent homotopies. Lemma 5.2. The augmentation E : WV + D, (fi7, t,, . . , fo) ti fn, 0. . of0 is a homosection r~: 2) + W2), f t+

topy equivalence of morphism spaces with the nonfinctorial (f) as homotopy inverse [2, Proposition 3.151. Definition rap* . Example 5.4. A homotopy X(,(f))
X0 421

5.3 [2]. A homotopy coherent D-diagram

is a continuous

functor D : WD +

&-diagram

X : &2 --+ %p* is given by

and two based homotopies F:X(q(h)) N X(11(3)) o X(7(f)) and G:X(q(h)) N X(77(7)) o X(q(g)) which are part of the structure. Here 7 : QZ + WQ2 is the canonical section and the morphisms f, 7, g, ?j, and h in QZ are the inclusions of the indexing sets of the spaces at the vertices. Obviously, homotopy induces a homotopy a homotopy commutative and vice versa. square with a specified &-diagram

Before we can construct our homotopy Tk-diagram we need another functor which will turn out to be also useful for the proof of Theorem 1.4. Construction 5.5. Let nJ denote the category of ordered sets [n] = (0, 1, . . . , n} and order preserving injections and ny the full subcategory of objects [j], j < lc. Define a functor 61, :rj ++ AT_,

by &c({il> . ..,i,})=Ip-1].Identifyil<...<i,inorderwithO<1<...<p-1. Then an inclusion cy : {il, . . , ip} c {jl, . . . , j,} defines a unique order preserving map tik((~) : lp - I] + [q - 11. (We always list the elements of a subset of n in increasing order.) Construction 5.6. Cpk : A? + (Gi)P is defined by sending [j] to j and Q : [p] + [q] to

(6
i=l

b(i)-a(i-I) > oa

)
>

.I. Klein et al. / Topology and its Applications

77 (1997) l-18

II

where 0 a : a(p) a(p) + 1,. Construction . , 4.

- a(O) +

g is the ordered injection -

missing

the elements

1, . . ~a(O),

5.7. We now turn to the actual construction

of a continuous

functor

Let X : (wk_ I G)Op+ ?$I* be a co-Ak_ 1-space. Then $k is a composite +k:: WTk, % (w,*_, g) 5 ?-@*

and hence functorial in X. We need all terminal faces of ?,!&highly co-Cartesian. To achieve this we construct ok in such a way that all 2-dimensional faces of X o & are co-Cartesian. It then follows that all higher dimensional faces of $k: are co-Cartesian a diagram of this kind is called strongly co-Cartesian in [6]. As explained too;

in Example 5.4 it suffices to specify ok on all objects and on morphisms

77(f) for f: {it,. . . , ip} C (11, . , I,+ I} in 7;; and to define homotopies ~(9) o n(f) = q(f) 0 q(g) for S o f = f o g : {il, . ,ip} c {jl:. ,&+?}. To simplify notation we write f for n(f) if there is no chance for confusion. On objects and on morphisms E : w7k + Tk is the canonical ical section. In detail:
~k({b. :i,}) =p-

f as above 81, is given by n* o p&t o 81,o E where and n* : ($$,

augmentation

1. 1 if It $ Image f: ifl,$Imagef,
if 1,+1 $

Ifp=

1 wetakeBk(f)=(id,j,8+1),andifp> LI @ idn--:!
ok(f) =

idr_:! @3 x2 $ id,_, i idp_-2 @ ~2

1 <r<p+l,

Image f. the rth element. in 7k into c {j I,..., jp+2} We now define the in 7k and j, < .jq the

where

LT : 1

denotes

the injection

missing

homotopies: elements

Let h: {i,, . . . , ip} c {j,, . . ;jp+2} be a morphism

not in the image of h. Then h decomposes ,,..., ip} c {j ,,..., 3, ,... jr ).... ,jn+?}

h=gof:{i =fog:{i,,

. . . . in} c {j l....,

&+2} c {j,: . . . . &+z}

where ji means that this element is deleted. Case 5.7.1. If 1 < r < q < p + 2 we have a diagram
P-

id,

2%X2CBid,_, *P

12

J. Klein et al. / Topology und its Applications

77 (1997) 1-18

which commutes

if q > T + 1.

If q = T + 1 this diagram reduces to the form

The square commutes v*(id $ A,) o I

up to homotopy N 71*(Xx) =

by the associating
11*(X2 83 id) 071*(x2).

homotopy

Case 5.7.2. If T = 1, q = 2 we have a commutative

square

and similarly

for T = p + 1, q = p + 2.

Case 5.7.3. If T = 1 and q = p + 2 we have commutative

squares

for p > 1, respectively

p = 1.

The diagrams of Cases 5.7.2 and 5.7.3 also take care of the case p = 1

Case 5.7.4. If T = 1 and 2 < q < p + 2 we have a commutative

square

*P

and similarly

for 1 < T < p + 1 and Q = p + 2.

J. Klein et al. / Topology and its Applications 77 (1997) l-18

13

Let V c W7r~ be the subcategory generated by all WQ2 c VW,+ where Q2 runs through all 2-dimensional faces of 7Jk.Then our construction defines Ok on V. We now apply the lifting theorem [2, 3.171 to

wrk
E

(w;_ , G)P
Pk-lOflk

to obtain an extension

of &IV to WTk. co-Ak_ 1-space, k > 3. Then ,for each (k - l)-

Lemma 5.8. Let X be a l-connected

dimensional ,far:e S of Tk the homotopy coherent S-diagram ?JklwS: WS + 7op* co-Cartesian in the sense of [6, 2.11.

is strongly (homotopy)

Proof. We have to show that the specified faces of the functor Ok are homotopy pushouts after composition with X*. This holds for the first diagram of Case 5.7.1, and for Case 5.7.4 if T = 1 and 3 < q or T < p and q = p + 2. For given maps f : A + B and 9 : C + D in Top*, the square AvCfVidBvC I
idVg

I
idvg fvid

+ AVD-BVD

$
pushout. Case 5.7.3 holds by direct investigation. homotopy For the second associative co-H-

is always a homotopy

diagram of Case 5.7.1 we use the fact that a l-connected space which is a CW-complex

has a 2-sided inverse [5, Proposition we appeal to [ 19, Lemma 3.31. 0

3.61. For Case 5.7.2

6. Proof of Theorem

1.2

Recall from Diagrams 4.2 and 4.3 that the homotopy limit Pj of the W7s-diagram we started with can be obtained as an iterated homotopy pullback and that the homotopy colimit Mj of the induced WJs-diagram can be obtained as an pushout. We mimic this in the general case. So let $1 : W7[ + Top*, 2 < diagrams determined by the co-Al-I-structures of the co-Ak_t-space be the face defined by all subsets of Ic containing the element 1 iterated homotopy 1 < k be the cube X. Let Sk-1 c Tk (it corresponds to

14

J. Klein et al. / Topology and its Applications

77 (1997) I-18

the bottom face of Diagram vertex. TSk_1 the homotopy &_r

4.1), and Tsk_

1 obtained

from it by deleting faces of Sk_,. homotopy

the initial x-

is the subdiagram

of all proper terminal WTSk_1

Let Pl denote

limit (in the sense of [16]) of the coherently by [16, Proposition

commutative

diagram $l and Rk_1 the one of the subdiagram are still well-pointed limit of the induced diagram * = &{l} + Rk_, t 4-1 5.41). In particular, = fix

+ %p*. Observe that Pl and

6.91. We can obtain P, as the homotopy

(compare [18, Proposition

we have a sequence of fibrations

plc_p

Pk-1 k ~ ,-...--%pz = flRk_1.

with fiber&)

$&(k\{i} + k) : Xv(k-2) is(l-(k-l)+(k-l)(n-l))-

Since $!&jWSk-~ is strongly Co-Cartesian and each map + X(-1 is (n - I)-connected, the induced map * + Rk-1 connected by Proposition 3.2, and so is pk. Since n > 2

and P2 = 0X each Pk, k 3 2, is connected. Pk together with the universal transformation Pk + $k extends $k to a Mi&&.-diagram @k : WQk + fip*, i.e., to a homotopy coherent k-dimensional cube diagram. Let 7 C
&k be an l-dimensional terminal face, 1 < k. By Lemma 5.8 it is strongly co-Cartesian, and hence by Proposition 3.2 (1 + I(n - 2))-Cartesian. Since @I, is infinitely Cartesian

it is (k - 1 + 2 + k(n - 2))-co-Cartesian Recall that 3~ c &,+ is the subcategory

by Proposition

3.3.

of all subsets of l C b, 1 < k, except for 1

itself. Let Zl-1 c 3~ denote the subcategory of all subsets containing I and JZl-I the full subcategory of Zl-t containing all but the terminal element, and let Ml denote the homotopy colimit of @k 1 WJt . By Lemma 5.8 the homotopy colimit of @I,Iw~~~_I to the terminal have homotopy vertex of Zl-1 is a homotopy pushout diagrams: is X(-1 and the induced map 5.41 we

equivalence.

By [18, Proposition

Diagram 6.1.

for 3 6 1 < k, where rl : A41 -+ X (l-l) is the induced map into of @klw&. Then, as noted above, ~1 o gl N id. Since Ok is (k(n Cartesian, rk is (k(n - 2) + k + I)-connected, hence gk is (k(n Since Diagram 6.1 is a homotopy pushout, r&t is homology (k(n By downwards induction we obtain r2 : CP, = M2 + x

the terminal

vertex

- 2) + k + I)-co2) + k)-connected. 2) + k)-connected.

.I. Klein et al. / Topology and its Applications

77 (1997) l-18

I5

is homology implies of

(k(n - 2) + 3)- connected.

Since Cl,

and X are both l-connected,

this part

that r2 is (k(n - 2) + 3)- connected.

We have proved the connectivity

Proposition 6.2. Let X be an (n - I)-connected be the homotopy limit of its associated coherently Then the induced map

co-ilk-l-space, homotopy

n 3 2, k >, 3. Let Pk Tk-diagram.

commutative

is a (k . (n - 2) + 3)- connected CO-Ak_ 1-homomorphism. It remains to show that r2 : cpk = M* --f X is a CO-..& I -homomorphism. The first

of Diagrams 6.1 is obtained from (see also Diagram 4.3)

r),

with p = X(v(X2)). the standard commutative

The induced

map q3 : Al, = CP,

X V X factors through

pinch map and a wedge of two maps so that we arrive at a homotopy diagram

cP,Ax

CP,

1
CP,

pinch

Since rs o gs r id and the pinch map and 1-1 have counits we deduce f E r2 rv g. Hence rz is a homomorphism of co-H-spaces up to homotopy (to obtain higher coherence one has to include the higher dimensional cubes into the argument; we leave this to the reader).

16

J. Klein et al. / Topology and its Applications 77 (1997) l-18

Proof of Theorem is a CW-complex r2 induces

1.2. Let K be the CW-approximation with dimX + X which is a homology epimorphism CW-complex

of 9,

and suppose that X is free abelian and in dimensions less as

< Ic(n - 2) + 3. Since Hk(n_2~+3X isomorphism in dimension

a map 4: CK

than Ic(n - 2) + 3 and a homology (k(n - 2) +2)-d imensional

k(n - 2) + 3, there is a (cf. [ 1,

Y having the same (Ic(n - 2) + I)-skeleton Since EY and X are l-connected

K and a map f : Y + K such that q o Cf : CY + X is a homology isomorphism Theorem 2.11). Moreover, q o Cf it is a homotopy equivalence. is a co-ilk-l-map.

7. Proof of Theorem

1.4 space such that X0 is contractible, Xi is

Let X : A -+ %p* be a based cosimplicial 2-connected, (W 1.. and n

is a homotopy

equivalence.

Define

p:x%x*+xvx where the second map is a homotopy inverse of (~1, ~2) = (cl*, do). (By abuse of notation we write X for Xi .) Then (X, p) is a co-H-space because

commutes

up to homotopy

since Xa is contractible.

In a similar way one verifies that

(X, p) is homotopy

associative. 1 > 2, of X 0191~ : Tk -+ A?:, -+ Top is strongly

Lemma 7.1. Each l-dimensionalface, homotopy co-Cartesian.

Proof. We evaluate X o flk on 2-dimensional faces. Let a: : { il , . . . , iP} C be an inclusion. If a: misses j, < j, the associated square is Diagram 7.2.

{jl , . . . , jP+2}

J. Klein et al. / Topology and its Applications

77 (1997) I-18

17

Each

such square

is homotopy

equivalent

to a square

considered

in the proof of

Lemma 5.8 and hence a homotopy Diagram 7.3.

pushout: for 0 < i < p + 1 consider

XV(i-)

p(p-9

-=5-x?,

f
XV(i-I)

j,dvdid IdA

(X

X)

Xv(p-i)idv(dido)vd > xv(-l)

x7

_p(P-c

-xp+1

The map id V (d*, do) V id = id V ( rl, q) V id is an equivalence by assumption on X, and so are the other horizontal equivalences. The square commutes while the triangle commutes up to homotopy.
XV(P-1)

Hence, if 1 < T < 4 < p + 2, Diagram 7.2 is equivalent

to

rXVP

+ XVP

id,_zVpVid,+l_,

+jyv(P+l)

which corresponds to Case 5.7.1, and hence is a homotopy pushout, because (X, p) is a homotopy associative co-H-space. For do, dP+ : X, + X,+, the squares corresponding to Diagram 7.3 arise from xvxp~xp + X p+l --++x,vx

d t-----p

II
-x,+1

dP+ I

As before, the remaining hence are homotopy Proof of Corollary . * PJ 3

cases of Diagram 7.2 give diagrams of Cases 5.7.2-5.7.4

and

pushouts.

1.3 and Theorem 1.4. Let Pk denote the homotopy limit of X o v.Y~. 1.2 we obtain a sequence of fibrations P3 3 4-1 P2 N RX is k-connected because X is 2-connected, limit of the 4. and a (Ic + 3)Since

As shown in the proof of Theorem

such that pk : Pk + connected

map rl~ : CP, + X. Let P be the homotopy = 0

hm xi(&) we have nk(P)

limKk(PZ)

rk(Pk).

In particular, the natural projection q,+: P + Pk is k-connected. The maps rk : CPk + X are compatible with the pk up to homotopy and induce a map CP+X. Since CP + CPk -+ X is (Ic + 1)-connected, this map is a weak equivalence.

18

J. Klein et al. / Topology and its Applications

77 (1997) I-18

Acknowledgment We thank the Deutsche Forschungsgemeinschaft ing Saitos paper to our attention, presentation of the paper. for support, M. Golasinski for bringthe

and the referee for helpful comments

concerning

References [I] I. Bernstein and PJ. Hilton, On suspension and comultiplications, Topology 2 (1963) 73-82. [2] J.M. Boardman and R.M. Vogt, Homotopy Invariant Structures on Topological Spaces, Lecture Notes in Math. 347 (1973). [3] A.K. Bousfield, On the homology spectral sequence of a cosimplicial space, Amer. J. Math. 109 (1987) 361-394. [4] G. Ellis and R. Steiner, Higher-dimensional crossed modules and the homotopy groups of (n + I)-ads, J. Pure Appl. Algebra 46 (1987) 117-136. [5] T. Ganea, Cogroups and suspensions, Invent. Math. 9 (1970) 185-197. [6] T. Goodwillie, Calculus II: Analytic functors, K-Theory 5 (1992) 295-332. [7] M.J. Hopkins, Formulations of cocategory and the iterated suspension, Asterisque 113-l 14 (1984) 212-226. [8] J.R. Klein, Embedding theory in the Poincare category, in preparation. [9] J.R. Klein, R. Schwanzl and R.M. Vogt, Parameterized suspension theory, in preparation. [lo] J.P. May, The Geometry of Iterated Loop Spaces, Lecture Notes in Math. 271 (1972). [1 l] S. Saito, On higher coassociativity, Hiroshima Math. J. 6 (1976) 589617. [12] G. Segal. Categories and cohomology theories, Topology 13 (1974) 293-312. [13] J.D. Stasheff, Homotopy associativity of H-spaces I, Trans. Amer. Math. Sot. 108 (1963) 293-312. [ 141 J.D. Stasheff, H-spaces from the Homotopy Point of View, Lecture Notes in Math. 161 (1970). [ 151 R.M. Vogt, Convenient categories of topological spaces for homotopy theory, Arch. Math. 22 (1971) 545-555. [16] R.M. Vogt, Homotopy limits and colimits, Math. Z. 134 (1973) 1 l-52. [17] R.M. Vogt, A remark on An-spaces and loop spaces, J. London Math. Sot. (2) 14 (1976) 321-325. [18] R.M. Vogt, Commuting homotopy limits, Math. Z. 153 (1977) 59-82. [19] M. Walker, Homotopy pull-backs and applications to duality, Canad. J. Math. 29 (1977) 45-64.