We give a proof of the generalised Eilenberg-Zilber theorem of Dold,PuppeandCartier[9]. The arguments used here are adapted from the articles [12], [13] of EilenbergandMac Lane.
Throughout this section, we suppose given a bisimplicial objectAin an abelian category A.
(3.26) Definition (shuffle). We letn∈N0 andp∈ bn,0c. A(p, n−p)-shuffle is a permutationµ∈S[0,n−1]
such that µ|[0,p−1] and µ|[p,n−1] are strictly monotonically increasing maps, where we write S[0,n−1] for the symmetric group on[0, n−1]. The set of all(p, n−p)-shuffles is denoted byShp,n−p.
(3.27) Definition. We let n∈N0, p∈ bn,0c.
(a) The morphism An,n AW
A p,n−p
−−−−−→Ap,n−p
is defined by
AWp,n−p :=AWAp,n−p:= dhbn,p+1cdvbp−1,0c. (b) Further, we let
Ap,n−p−−−−−→∇p,n−p An,n
be given by
∇p,n−p:=∇Ap,n−p:= X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµ.
Our first aim is to show that these morphisms yield complex morphisms betweenC DiagAandTot C(2)A.
(3.28) Remark. Defining CnDiagA−−→AWn TotnC(2)A
byAWnprp,n−p:=AWp,n−pfor allp∈ bn,0c,n∈N0, we obtain a complex morphism C DiagA−→AW Tot C(2)A.
Proof. We have
AWn∂prp,n−1−p =AWp+1,n∂h+ (−1)pAWp,n−p∂v
= dhbn,p+2cdvbp,0c( X
i∈[0,p+1]
(−1)idhi) + (−1)pdhbn,p+1cdvbp−1,0c( X
j∈[0,n−p]
(−1)jdvj)
= X
i∈[0,p+1]
(−1)idhbn,p+2cdhidvbp,0c+ X
j∈[0,n−p]
(−1)j+pdhbn,p+1cdvj+pdvbp−1,0c
= X
i∈[0,p+1]
(−1)idhbn,p+2cdhidvbp,0c+ X
j∈[p,n]
(−1)jdhbn,p+1cdvjdvbp−1,0c
= X
(3.29) Definition (Alexander-Whitney morphism). The complex morphism C DiagA−→AW Tot C(2)A
(3.30) Proposition (recursive characterisation of the shuffle morphism via path simplicial objects). We have
∇A0,0= idA0,0 and
we can conclude
∇Ap,n−p= svn−p X
µ∈Shp,n−p
(p−1)µ=n−1
(sgnµ)svd0,p−2eµshdp,n−1eµ+ shp X
µ∈Shp,n−p
(n−1)µ=n−1
(sgnµ)svd0,p−1eµshdp,n−2eµ
= (−1)n−psvn−p X
µ∈Shp−1,n−p+1
(n−1)µ=n−1
(sgnµ)svd0,p−2eµshdp−1,n−2eµ+ shp X
µ∈Shp,n−p
(n−1)µ=n−1
(sgnµ)svd0,p−1eµshdp,n−2eµ
= (−1)n−psvn−p X
µ∈Shp−1,n−p
(sgnµ)svd0,p−2eµshdp−1,n−2eµ+ shp X
µ∈Shp,n−1−p
(sgnµ)svd0,p−1eµshdp,n−2eµ
= (−1)n−psvn−p∇Pp−1,n−p(2)A + shp∇Pp,n−1−p(2)A
The proof forp=norp= 0is easier since in this case the sum in the shuffle morphism does not split into two sums and since the only(n,0)-shuffle resp.(0, n)-shuffle is the identity: We have
∇An,0= svd0,n−1e= sv0svd0,n−2e= sv0∇Pn−1,0(2)A and
∇A0,n= shd0,n−1e= sh0shd0,n−2e= sh0∇P0,n−1(2)A. (3.31) Lemma. We have
∇Ap,n−psDiagn A= svn−pshp∇Pp,n−p(2)A and
∇Pp,n−p(2)AdDiagn+1A= dvn+1−pdhp+1∇Ap,n−p for allp∈ bn,0c,n∈N0.
Proof. We compute
∇Ap,n−psDiagn A= X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµsvnshn = svn−pshp X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµ
= svn−pshp∇Pp,n−p(2)A and
∇Pp,n−p(2)AdDiagn+1A= X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµdvn+1dhn+1
= dvn+1−pdhp+1 X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµ= dvn+1−pdhp+1∇Ap,n−p for allp∈ bn,0c,n∈N0.
(3.32) Remark. We have
∇p,n−pdDiagn A=
dhn∇n−1,0 ifp=n,
(−1)n−pdhp∇p−1,n−p+ dvn−p∇p,n−1−p ifp∈ bn−1,1c,
dvn∇0,n−1 ifp= 0
for allp∈ bn,0c,n∈N.
Proof. According to proposition (3.30) and lemma (3.31), we have
∇Ap,n−pdDiagn A= ((−1)n−psvn−p∇Pp−1,n−p(2)A + shp∇Pp,n−1−p(2)A )dDiagn A
= (−1)n−psvn−p∇Pp−1,n−p(2)A dDiagn A+ shp∇Pp,n−1−p(2)A dDiagn A
= (−1)n−psvn−pdvn+1−pdhp∇Ap−1,n−p+ shpdvn−pdhp+1∇Ap,n−1−p
= (−1)n−pdhp∇Ap−1,n−p+ dvn−p∇Ap,n−1−p
forp∈ bn−1,1c. The computation forp=nandp= 0 is analogous: We have
∇An,0dDiagn A= (sv0∇Pn−1,0(2)A)dDiagn A= sv0dv1dhn∇An−1,0= dhn∇An−1,0 and
∇A0,ndDiagn A= (sh0∇P0,n−1(2)A)dDiagn A= sh0dvndh1∇A0,n−1= dvn∇A0,n−1. (3.33) Remark. Defining
TotnC(2)A−−→∇n CnDiagA
byembp,n−p∇n:=∇p,n−p for allp∈ bn,0c,n∈N0, we obtain a complex morphism Tot C(2)A−→∇ C DiagA.
Proof. By the definition of the differential morphisms inTot C(2)A, we have to show that
∇Ap,n−p∂C DiagA=
∂C(2)A,h∇An−1,0 ifp=n,
∂C(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)A,v∇Ap,n−1−p ifp∈ bn−1,1c,
∂C(2)A,v∇A0,n−1 ifp= 0
for allp∈ bn,0c,n∈N0. First, we consider the boundary cases: We have
∇An,0∂C DiagA= svd0,n−1e X
k∈[0,n]
(−1)kdDiagk A= svd0,n−1e X
k∈[0,n]
(−1)kdvkdhk
= X
k∈[0,n]
(−1)ksvd0,n−1edvkdhk = X
k∈[0,n]
(−1)kdhksvd0,n−2e=∂C(2)A,h∇An−1,0
and analogously
∇A0,n∂C DiagA= shd0,n−1e X
k∈[0,n]
(−1)kdDiagk A= shd0,n−1e X
k∈[0,n]
(−1)kdvkdhk
= X
k∈[0,n]
(−1)kdvkshd0,n−1edhk = X
k∈[0,n]
(−1)kdvkshd0,n−2e=∂C(2)A,v∇A0,n−1
for alln∈N. It remains to prove
∇Ap,n−p∂C DiagA=∂C(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)A,v∇Ap,n−1−p forp∈ bn−1,1c,n∈N,n≥2. Thereto, it suffices to show that
∇Ap,n−p∂C Diag P(2)A=∂C(2)P(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)P(2)A,v∇Ap,n−1−p because according to remark (3.32) this implies
∇Ap,n−p∂C DiagA=∇Ap,n−p(∂C Diag P(2)A+ (−1)ndDiagn A) =∇Ap,n−p∂C Diag P(2)A+ (−1)n∇Ap,n−pdDiagn A
=∂C(2)P(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)P(2)A,v∇Ap,n−1−p+ (−1)n((−1)n−pdhp∇Ap−1,n−p+ dvn−p∇Ap,n−1−p)
=∂C(2)P(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)P(2)A,v∇Ap,n−1−p+ (−1)pdhp∇Ap−1,n−p+ (−1)ndvn−p∇Ap,n−1−p
=∂C(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)A,v∇Ap,n−1−p.
We proceed by induction on n∈N, n≥2, to show the second identity involving the path bisimplicial object P(2)A and use the recursive characterisation of proposition (3.30). However, by the induction hypothesis, we may also use the first identity involvingAduring our calculations since this is implied by the second as already shown.
First, forn= 2andp= 1we compute
∇A1,1∂C Diag P(2)A= (sh1∇P1,0(2)A−sv1∇P0,1(2)A)∂C Diag P(2)A= sh1∇P1,0(2)A∂C Diag P(2)A−sv1∇P0,1(2)A∂C Diag P(2)A
= sh1∂C(2)P(2)A,h∇P0,0(2)A−sv1∂C(2)P(2)A,v∇P0,0(2)A= sh1∂C(2)P(2)A,h−sv1∂C(2)P(2)A,v
= sh1(dh0−dh1)−sv1(dv0−dv1) = dh0sh0−idA1,1−dv0sv0+ idA1,1 = dh0sh0−dv0sv0
=∂C(2)P(2)A,h∇A0,1−∂C(2)P(2)A,v∇A1,0. Next, we show the asserted formula forp=n−1,n∈N,n≥3:
∇An−1,1∂C Diag P(2)A= (−sv1∇Pn−2,1(2)A+ shn−1∇Pn−1,0(2)A)∂C Diag P(2)A
=−sv1∇Pn−2,1(2)A∂C Diag P(2)A+ shn−1∇Pn−1,0(2)A∂C Diag P(2)A
=−sv1(∂C(2)P(2)A,h∇Pn−3,1(2)A + (−1)n−2∂C(2)P(2)A,v∇Pn−2,0(2)A) + shn−1∂C(2)P(2)A,h∇Pn−2,0(2)A
=−sv1∂C(2)P(2)A,h∇Pn−3,1(2)A + (−1)n−1sv1∂C(2)P(2)A,v∇Pn−2,0(2)A + shn−1∂C(2)P(2)A,h∇Pn−2,0(2)A
=−∂C(2)P(2)A,hsv1∇Pn−3,1(2)A + (−1)n−1(∂C(2)P(2)A,vsv0−idAn−1,1)∇Pn−2,0(2)A + (∂C(2)P(2)A,hshn−2+ (−1)n−1idAn−1,1)∇Pn−2,0(2)A
=−∂C(2)P(2)A,hsv1∇Pn−3,1(2)A + (−1)n−1∂C(2)P(2)A,vsv0∇Pn−2,0(2)A +∂C(2)P(2)A,hshn−2∇Pn−2,0(2)A
=∂C(2)P(2)A,h(shn−2∇Pn−2,0(2)A−sv1∇Pn−3,1(2)A) + (−1)n−1∂C(2)P(2)A,vsv0∇Pn−2,0(2)A
=∂C(2)P(2)A,h∇An−2,1+ (−1)n−1∂C(2)P(2)A,v∇An−1,0. Analogously, forp= 1, n∈N,n≥3, we have
∇A1,n−1∂C Diag P(2)A= ((−1)n−1svn−1∇P0,n−1(2)A+ sh1∇P1,n−2(2)A)∂C Diag P(2)A
= (−1)n−1svn−1∇P0,n−1(2)A∂C Diag P(2)A+ sh1∇P1,n−2(2)A∂C Diag P(2)A
= (−1)n−1svn−1∂C(2)P(2)A,v∇P0,n−2(2)A+ sh1(∂C(2)P(2)A,h∇P0,n−2(2)A−∂C(2)P(2)A,v∇P1,n−3(2)A)
= (−1)n−1svn−1∂C(2)P(2)A,v∇P0,n−2(2)A+ sh1∂C(2)P(2)A,h∇P0,n−2(2)A−sh1∂C(2)P(2)A,v∇P1,n−3(2)A
= (−1)n−1(∂C(2)P(2)A,vsvn−2+ (−1)n−1idA1,n−1)∇P0,n−2(2)A
+ (∂C(2)P(2)A,hsh0−idA1,n−1)∇P0,n−2(2)A −∂C(2)P(2)A,vsh1∇P1,n−3(2)A
= (−1)n−1∂C(2)P(2)A,vsvn−2∇P0,n−2(2)A+∂C(2)P(2)A,hsh0∇P0,n−2(2)A−∂C(2)P(2)A,vsh1∇P1,n−3(2)A
=∂C(2)P(2)A,hsh0∇P0,n−2(2)A−((−1)n−2∂C(2)P(2)A,vsvn−2∇P0,n−2(2)A+∂C(2)P(2)A,vsh1∇P1,n−3(2)A)
=∂C(2)P(2)A,hsh0∇P0,n−2(2)A−∂C(2)P(2)A,v((−1)n−2svn−2∇P0,n−2(2)A+ sh1∇P1,n−3(2)A)
=∂C(2)P(2)A,h∇A0,n−1−∂C(2)P(2)A,v∇A1,n−2. Finally, we letp∈[n−2,2],n∈N, n≥4. Then we get
∇Ap,n−p∂C Diag P(2)A= ((−1)n−psvn−p∇Pp−1,n−p(2)A + shp∇Pp,n−1−p(2)A )∂C Diag P(2)A
= (−1)n−psvn−p∇Pp−1,n−p(2)A ∂C Diag P(2)A+ shp∇Pp,n−1−p(2)A ∂C Diag P(2)A
= (−1)n−psvn−p(∂C(2)P(2)A,h∇Pp−2,n−p(2)A + (−1)p−1∂C(2)P(2)A,v∇Pp−1,n−1−p(2)A ) + shp(∂C(2)P(2)A,h∇Pp−1,n−1−p(2)A + (−1)p∂C(2)P(2)A,v∇Pp,n−2−p(2)A )
= (−1)n−psvn−p∂C(2)P(2)A,h∇Pp−2,n−p(2)A + (−1)n−1svn−p∂C(2)P(2)A,v∇Pp−1,n−1−p(2)A + shp∂C(2)P(2)A,h∇Pp−1,n−1−p(2)A + (−1)pshp∂C(2)P(2)A,v∇Pp,n−2−p(2)A
= (−1)n−p∂C(2)P(2)A,hsvn−p∇Pp−2,n−p(2)A
+ (−1)n−1(∂C(2)P(2)A,vsvn−1−p+ (−1)n−pidAp,n−p)∇Pp−1,n−1−p(2)A
+ (∂C(2)P(2)A,hshp−1+ (−1)pidAp,n−p)∇Pp−1,n−1−p(2)A + (−1)p∂C(2)P(2)A,vshp∇Pp,n−2−p(2)A
= (−1)n−p∂C(2)P(2)A,hsvn−p∇Pp−2,n−p(2)A + (−1)n−1∂C(2)P(2)A,vsvn−1−p∇Pp−1,n−1−p(2)A +∂C(2)P(2)A,hshp−1∇Pp−1,n−1−p(2)A + (−1)p∂C(2)P(2)A,vshp∇Pp,n−2−p(2)A
=∂C(2)P(2)A,h((−1)n−psvn−p∇Pp−2,n−p(2)A + shp−1∇Pp−1,n−1−p(2)A )
+ (−1)p∂C(2)P(2)A,v((−1)n−p−1svn−1−p∇Pp−1,n−1−p(2)A + shp∇Pp,n−2−p(2)A )
=∂C(2)P(2)A,h∇Ap−1,n−p+ (−1)p∂C(2)P(2)A,v∇Ap,n−1−p
By induction, we have shown that the morphisms∇n forn∈N0 yield a complex morphism Tot C(2)A−→∇ C DiagA.
(3.34) Definition (shuffle morphism). The complex morphism Tot C(2)A−→∇ C DiagA
given as in remark (3.33) byembp,n−p∇n=∇p,n−pfor allp∈ bn,0c, that is,
∇n=
∇n,0
...
∇0,n
as a morphism fromTotnC(2)A=L
p∈bn,0cAp,n−ptoCnDiagA=An,nfor alln∈N0, is called (Eilenberg-Mac Lane)shuffle morphism.
At next, we will show that the Alexander-Whitney morphism and the Eilenberg-Mac Lane shuffle morphism restrict to well-defined morphisms onM DiagAresp.Tot M(2)A.
(3.35) Proposition.
(a) We have a morphism of split short exact sequences D DiagA //
C DiagA //
AW
M DiagA
Tot D(2)A //Tot C(2)A //Tot M(2)A
By abuse of notation, the induced morphism M DiagA−→Tot M(2)Ais also denoted byAW:=AWA. (b) We have a morphism of split short exact sequences
Tot D(2)A //
Tot C(2)A //
∇
Tot M(2)A D DiagA //C DiagA //M DiagA
By abuse of notation, the induced morphism Tot M(2)A−→M DiagAis also denoted by∇:=∇A. Proof.
(a) We have
sDiagk AAWp,n−p= shksvkdhbn,p+1cdvbp−1,0c= shkdhbn,p+1csvkdvbp−1,0c=
((shkdhbn,p+1cdvbp−1,0c)svk−p ifp≤k, (dhbn−1,pcsvkdvbp−1,0c)shk ifp > k,
for every k∈[0, n−1], that is,(Im sDiagk A)AWp,n−p= Im(sDiagk AAWp,n−p)D(2)p,n−pAfor allk∈[0, n−1]
and therefore(DnDiagA)AWp,n−pD(2)p,n−pAfor allp∈ bn,0c. Hence we have an induced morphism D DiagA−→Tot D(2)A.
Moreover, M DiagA ∼= C DiagA/D DiagA and Tot M(2)A ∼= Tot(C(2)A/D(2)A) ∼= Tot C(2)A/Tot D(2)A by the normalisation theorem (2.28). Hence we have an induced morphism on the cokernels
M DiagA−→AW Tot M(2)A.
(b) We will show that
(D(2)p,n−pA)∇Ap,n−pDnDiagA
for allp∈ bn,0c, n∈N0. Thereto, we proceed by induction onn, where for n= 0the assertion is trivial since D(2)0,0A ∼= 0and D0DiagA ∼= 0. So we let a natural number n∈ N with n ≥1 and p ∈ bn,0c be given and we assume that the asserted inclusion holds for all bisimplicial sets up to dimension n−1. By proposition (3.30), we compute
shi∇Ap,n−p=
(shisv0∇Pn−1,0(2)A ifp=n, shi((−1)n−psvn−p∇Pp−1,n−p(2)A + shp∇Pp,n−1−p(2)A ) ifp∈ bn−1,1c
)
=
(shisv0∇Pn−1,0(2)A ifp=n, (−1)n−pshisvn−p∇Pp−1,n−p(2)A + shishp∇Pp,n−1−p(2)A ifp∈ bn−1,1c
)
=
(sv0shi∇Pn−1,0(2)A ifp=n,
(−1)n−psvn−pshi∇Pp−1,n−p(2)A + shp−1shi∇Pp,n−1−p(2)A ifp∈ bn−1,1c fori∈[0, p−1],p∈ bn,1c, and therefore
Im(shi∇An,0)Im(sv0shi∇Pn−1,0(2)A) fori∈[0, n−1]and
Im(shi∇Ap,n−p)Im((−1)n−psvn−pshi∇Pp−1,n−p(2)A + shp−1shi∇Pp,n−1−p(2)A ) Im((−1)n−psvn−pshi∇Pp−1,n−p(2)A ) + Im(shp−1shi∇Pp,n−1−p(2)A ) Im(svn−pshi∇Pp−1,n−p(2)A ) + Im(shi∇Pp,n−1−p(2)A )
fori∈[0, p−1],p∈ bn−1,1c. Now by the induction hypothesis, we have
Im(shi∇Pp,n−1−p(2)A )(D(2)p,n−1−pP(2)A)∇Pp,n−1−p(2)A Dn−1Diag P(2)ADnDiagA fori∈[0, p−1],p∈ bn−1,1c, and
Im(svn−pshi∇Pp−1,n−p(2)A )Im(shi∇Pp−1,n−p(2)A )(D(2)p−1,n−pP(2)A)∇Pp−1,n−p(2)A Dn−1Diag P(2)A DnDiagA
fori∈[0, p−2],p∈ bn,1c. Since additionally, by lemma (3.31),
Im(svn−pshp−1∇Pp−1,n−p(2)A ) = Im(∇Ap−1,n−psDiagn−1A)Im(sDiagn−1A)DnDiagA
forp∈ bn,1c, we can conclude that Im(shi∇Ap,n−p)DnDiagA fori∈[0, p−1],p∈ bn,1c.
Analogously, we show Im(svj∇Ap,n−p) DnDiagA for all j ∈ [0, n−1−p], p ∈ bn−1,0c. Indeed, by proposition (3.30), we have
svj∇Ap,n−p=
(svj((−1)n−psvn−p∇Pp−1,n−p(2)A + shp∇Pp,n−1−p(2)A ) ifp∈ bn−1,1c, svjsh0∇P0,n−1(2)A ifp= 0
)
=
((−1)n−psvjsvn−p∇Pp−1,n−p(2)A + svjshp∇Pp,n−1−p(2)A ifp∈ bn−1,1c, svjsh0∇P0,n−1(2)A ifp= 0
)
=
((−1)n−psvn−1−psvj∇Pp−1,n−p(2)A + shpsvj∇Pp,n−1−p(2)A ifp∈ bn−1,1c, sh0svj∇P0,n−1(2)A ifp= 0
forj∈[0, n−1−p], p∈ bn−1,0c, and therefore
Im(svj∇Ap,n−p)Im((−1)n−psvn−1−psvj∇Pp−1,n−p(2)A + shpsvj∇Pp,n−1−p(2)A ) Im((−1)n−psvn−1−psvj∇Pp−1,n−p(2)A ) + Im(shpsvj∇Pp,n−1−p(2)A ) Im(svj∇Pp−1,n−p(2)A ) + Im(shpsvj∇Pp,n−1−p(2)A )
forj∈[0, n−1−p], p∈ bn−1,1c, and Im(svj∇A0,n)Im(sh0svj∇P0,n−1(2)A)
forj∈[0, n−1]. With the induction hypothesis, it follows that
Im(svj∇Pp−1,n−p(2)A )(D(2)p−1,n−pP(2)A)∇Pp−1,n−p(2)A Dn−1Diag P(2)ADnDiagA forj∈[0, n−1−p], p∈ bn−1,1cand
Im(shpsvj∇Pp,n−1−p(2)A )Im(svj∇Pp,n−1−p(2)A )(D(2)p,n−1−pP(2)A)∇Pp,n−1−p(2)A Dn−1Diag P(2)ADnDiagA forj∈[0, n−2−p], p∈ bn−1,0c. Since additionally, by lemma (3.31),
Im(shpsvn−1−p∇Pp,n−1−p(2)A ) = Im(∇Ap,n−1−psDiagn−1A)Im(sDiagn−1A)DnDiagA
forp∈ bn−1,0c. HenceIm(svj∇Ap,n−p)DnDiagAforj∈[0, n−1−p],p∈ bn−1,0c.
Therefore
(D(2)p,n−pA)∇Ap,n−pDnDiagA.
So we have induced morphisms Tot D(2)A−→D DiagA and
Tot M(2)A ∇
A
−−→M DiagA.
(3.36) Theorem (generalised Eilenberg-Zilber theorem, normalised case). The Alexander-Whitney morphism CMA−→AW Tot M(2)A
and the Eilenberg-Mac Lane shuffle morphism Tot M(2)A−→∇ M DiagA
are mutually inverse homotopy equivalences. In particular, M DiagA'Tot M(2)A.
Proof. First, we want to show that ∇AAWA = idTot M(2)A. We let n∈N0 be given. For each p, q ∈ bn,0cwe have
∇Ap,n−pAWAq,n−q= ( X
µ∈Shp,n−p
(sgnµ)svd0,p−1eµshdp,n−1eµ)dhbn,q+1cdvbq−1,0c
= X
µ∈Shp,n−p
(sgnµ)(svd0,p−1eµdvbq−1,0c)(shdp,n−1eµdhbn,q+1c).
By applying the simplicial identities, we recognise that each summand ends with a vertical degeneracy if q < p resp. with a horizontal degeneracy if q > p. Since we are in the normalised case, this means that
∇Ap,n−pAWAq,n−q = 0forp6=q. It remains to consider the caseq=p. Then we have
since the only summand that is not trivial because it ends with a degeneracy, is the one whereµ = id[0,n−1]. Thus
We have to show that these morphisms induce morphisms on the entries of the normalised complexM DiagA.
Thereto, we prove that they restrict to morphisms DnDiagA h
For the first summand, we get by induction that Im(sDiagk AhPn−1(2)A)DnDiag P(2)ADn+1DiagA
ifk ≤n−2. We consider the case k =n−1. By definition and proposition (2.33)(c), hAn is a certain linear combination of morphisms, each one being a composite of a horizonal backal and a vertical backal morphism.
Thus, by proposition (2.33)(a) applied vertically and horizontally, we have sDiagn−1AhPn−1(2)A = hAn−1sDiagn A and hence
Im(sDiagn−1AhPn−1(2)A)Dn+1DiagA.
Altogether,
Im(sDiagk AhAn)Dn+1DiagA.
It remains to show that(hn∈A(MnDiagA,Mn+1DiagA)|n∈N0)is a complex homotopy from idM DiagA to fA, that is,
hAn∂C DiagA+∂C DiagAhAn−1= idCnDiagA−fnA for alln∈N0
up to sums of morphisms whose images are in the degenerate complex. We proceed by induction on n∈N0. Forn= 0, we have
hA0∂C DiagA= sDiag0 Af0A(dDiag0 A−dDiag1 A) = sDiag0 AdDiag0 A−sDiag0 AdDiag1 A= idCnDiagA−idCnDiagA
= idCnDiagA−fnA.
Now we assume thatn≥1and that the assumpted relation holds in all lower dimensions. Then we have hAn∂C Diag P(2)A= (hPn−1(2)A+ (−1)nsDiagn AfnP(2)A)∂C Diag P(2)A
=hPn−1(2)A∂C Diag P(2)A+ (−1)nsDiagn AfnP(2)A∂C Diag P(2)A
=hPn−1(2)A∂C Diag P(2)A+ (−1)nsDiagn A∂C Diag P(2)Afn−1P(2)A
= (idCnDiagA−fn−1P(2)A−∂C Diag P(2)AhPn−2(2)A) + (−1)n(∂C Diag P(2)AsDiagn−1A+ (−1)nid)fn−1P(2)A
= idCnDiagA−fn−1P(2)A−∂C Diag P(2)AhPn−2(2)A+ (−1)n∂C Diag P(2)AsDiagn−1Afn−1P(2)A+fn−1P(2)A
= idCnDiagA−∂C Diag P(2)AhPn−2(2)A+ (−1)n∂C Diag P(2)AsDiagn−1Afn−1P(2)A
= idCnDiagA−∂C Diag P(2)A(hPn−2(2)A+ (−1)n−1sDiagn−1Afn−1P(2)A) = idCnDiagA−∂C Diag P(2)AhAn−1 as well as, by proposition (2.31)(c),
hAndDiagn+1A= (hPn−1(2)A+ (−1)nsDiagn AfnP(2)A)dDiagn+1A=hPn−1(2)AdDiagn+1A+ (−1)nsDiagn AfnP(2)AdDiagn+1A
= dDiagn AhAn−1+ (−1)nsDiagn AdDiagn+1AfnA= dDiagn AhAn−1+ (−1)nfnA. Hence we can conclude
hAn∂C DiagA=hAn(∂C Diag P(2)A+ (−1)n+1dDiagn+1A) =hAn∂C Diag P(2)A+ (−1)n+1hAndDiagn+1A
= (idCnDiagA−∂C Diag P(2)AhAn−1) + (−1)n+1(dDiagn AhAn−1+ (−1)nfnA)
= idCnDiagA−∂C Diag P(2)AhAn−1+ (−1)n+1dDiagn AhAn−1−fnA
=−∂C Diag P(2)AhAn−1−(−1)ndDiagn AhAn−1+ idCnDiagA−fnA
=−(∂C Diag P(2)A+ (−1)ndDiagn A)hAn−1+ idCnDiagA−fnA
=−∂C DiagAhAn−1+ idCnDiagA−fnA, that is,hAn∂C DiagA+∂C DiagAhAn−1= idCnDiagA−fnA.
(3.37) Theorem (generalised Eilenberg-Zilber theorem ofDold,PuppeandCartier, cf. [9, Satz 2.9]). We have
C DiagA'Tot C(2)A.
Proof. By theorem (3.36), we have M DiagA'Tot M(2)A.
Since the normalisation theorem states a homotopy equivalence between the associated (double) complexes and the Moore (double) complexes, cf. theorem (2.28) and theorem (3.24), and since the total complex functor preserves homotopy equivalences due to proposition (3.19), this implies by theorem (3.36) that
C DiagA'M DiagA'Tot M(2)A'Tot C(2)A.
Quillenmentions the following corollary in [28] as well-known.
(3.38) Corollary. There exists a spectral sequence E with Ep,n−p1 ∼= Hn−p(CAp,−) that converges to the homology groupHn(C DiagA), wherep∈[0, n],n∈N0.
Proof. By the generalised Eilenberg-Zilber theorem (3.37), we haveC DiagA'Tot C(2)A and hence Hn(C DiagA)∼= Hn(Tot C(2)A)
for alln∈N0. The spectral sequenceE of the “columnwise” filtered double complexC(2)Ahas the entries Ep,n−p1 ∼= Hn−p(C(2)p,−A) = Hn−p(CAp,−)
forp∈[0, n],n∈N0.
(3.39) Corollary. We suppose given a bisimplicial setX, a commutative ringRand anR-moduleM. (a) There exists a spectral sequence E with Ep,n−p1 ∼= Hn−p(Xp,−, M;R) that converges to the homology
group Hn(DiagX, M;R), wherep∈[0, n],n∈N0.
(b) There exists a spectral sequence E with E1p,n−p ∼= Hn−p(Xp,−, M;R)that converges to the cohomology group Hn(DiagX, M;R), wherep∈[0, n],n∈N0.
Proof.
(a) We apply corollary (3.38) toRX⊗RM. Then we obtain
Hn(C Diag(RX⊗RM)) = Hn(C((DiagRX)⊗RM)) = Hn((C DiagRX)⊗RM)
= Hn((CRDiagX)⊗RM) = Hn(C(DiagX;R)⊗RM)
= Hn(DiagX, M;R) forn∈N0, and
Hn−p(C(RX⊗RM)p,−) = Hn−p(C(RXp,−⊗RM) = Hn−p((CRXp,−)⊗RM)
= Hn−p(C(Xp,−;R)⊗RM) = Hn−p(Xp,−, M;R) forp∈[0, n], n∈N0.
(b) By the generalised Eilenberg-Zilber theorem (3.37), we have
C(DiagX;R) = CRDiagX = C DiagRX'Tot C(2)RX= Tot C(2)(X;R) and hence
Hn(DiagX, M;R) = Hn(R(C(DiagX;R), M))∼= Hn(R(Tot C(2)(X;R), M))
= Hn(TotR(C(2)(X;R), M))
forn∈N0. The spectral sequence of the “columnwise” filtered double complexR(C(2)(X;R), M)has E1p,n−p∼= Hn−p(R(C(2)p,−(X;R), M)) = Hn−p(R(C(Xp,−;R), M)) = Hn−p(Xp,−, M;R)
forp∈[0, n], n∈N0. (1)
1The seeming non-duality in the proofs of (a) and (b) is due to the fact that cohomology ofcosimplicial objects has not been defined.