Topological Models for Arithmetic
William G. Dwyer and Eric M. Friedlander
University of Notre Dame
Northwestern University
Abstract. In this paper, we use topological models to compute the
`-adic topological K-theory of certain arithmetic rings A. Our tech-
nique is to exploit class field theory to show that the etale topological
type of A is equivalent in an appropriate sense to something relatively
simple. Calculating the K-theory of this simple "topological model"
provides an explicit determination of the `-adic topological K-theory
of A and, by means of a comparison map, gives information about the
algebraic K-theory of A. For example, we are sometimes able to com-
pute the mod ` cohomology of certain "unstable" topological K-theory
spaces and verify that it injects into the cohomology of the correspond-
ing unstable algebraic K-theory spaces. This gives an explicit lower
bound for H* (GL (n; A); Z=`):
x0. Introduction
In [4] we introduced the "etale K-groups" of a ring A as the homotopy
groups of a twisted topological K-theory space associated to the etale
topological type Aet of A. The construction of these groups involves
choosing a prime ` invertible in A and working with K-theory completed
at `. Building upon work of C. Soule [19], we described a natural map
from algebraic K-groups (completed at `) to etale K-groups, and showed
that this map is in general surjective for arithmetic rings. The well-
known Lichtenbaum-Quillen Conjectures essentially assert that this map
is an isomorphism.
Many of the basic results of class field theory and arithmetic duality
can be interpreted as statements about the etale cohomology of various
arithmetic rings A [10] or equivalently as statements about the (ordi-
nary) cohomology of Aet: In [5] we exploited this fact to find, for some
specific arithmetic rings A, an elementary topological space XA mapping
to Aet by a cohomology isomorphism. This construction allowed us not
only to compute the etale K-groups of A, but also to identify the under-
lying twisted topological K-theory space and compute its cohomology
(which, by the Lichtenbaum-Quillen conjecture, should be isomorphic
to the cohomology of GL (A)).
______________________________________
Partially supported by the National Science Foundation and by NSA Grant # MDA
904-90-H-4006.
Typeset by AM S-TEX
2 Dwyer and Friedlander
In the present paper, we extend this program to cover other rings A.
In each case, for a specific prime ` invertible in A, we find a "good mod `
model" XA for A; this is an explicit space or pro-space XA which in an
appropriate sense captures the mod ` cohomology of Aet. For example,
there is a very simple mod ` model for A if ` is a regular prime and
A = Z[1=`]. We also find that in fortunate situations (such as the one
just mentioned), the cohomology of the resulting (unstable) K-theory
space of the model XA is not only computable but also injects into the
cohomology of the (unstable) algebraic K-theory space of A. In some
cases this sheds light on the Lichtenbaum-Quillen conjecture (2.4, 6.4,
6.6).
In brief, the contents of our paper are as follows. Section 1 motivates
and presents the definition of a good mod ` model of an arithmetic ring
A. Such a model captures the etale cohomology of A with coefficients
in "Tate twists" of Z=`. In x2, we provide a particularly simple model
for A = Z[1=`] whenever ` is an odd regular prime. Verifying this model
is more delicate than verifying the models for affine curves and local
fields described in x3. The remainder of the paper is concerned with an-
alyzing the cohomology of (unstable) K-theory spaces associated with
our good mod ` models and studying the relationship between this co-
homology and that of the general linear groups of A. In x4, we discuss
the Eilenberg-Moore spectral sequence and apply it to our topological
K-theory spaces. Our comparison of cohomology of algebraic and topo-
logical K-theory spaces is achieved by restricting to maximal tori: the
actual comparison is carried out in x6, following a check in x5 that corre-
sponding algebraic and topological tori have the same cohomology. We
conclude the paper by mentioning a few open problems in x7.
0.1 Notation: We fix a prime number ` and let R denote the ring n
Z[1=`], F the finite field Z=`, i`n the `n 'th root of unity e2ssi=` and `n
the multiplicative group of `n 'th roots of unity. The symbol i without
a subscript denotes i`.
Our fundamental reference for homotopy theoretic completions is [2].
If X is a space or pro-space the `-adic completion tower of X is a pro-
space denoted F! (X); if X is a space then F! (X) is a tower in the
usual sense whose inverse limit is the `-adic completion F1 (X) of X.
If E ! B is a fibration the fibrewise `-adic completion tower of E over
B [4, p. 250] is denoted Fo!(E); if E and B are spaces, the inverse limit
of this tower is the fibrewise `-adic completion Fo1(E) of E over B.
x 1. Good mod ` models
We recall that Xet is a "space" which reflects the etale cohomology
of a noetherian simplicial scheme X. The reader can consult [11] for a
Topological Models 3
coherent discussion of the etale cohomology H*et(X; F ) of X with coef-
ficients in a sheaf F on the "etale site"; endurance now assured, he or
she can then peruse [1] and [7] for the construction and properties of
the pro-space (i.e., inverse system of simplicial sets) Xet.
For any finitely generated R-algebra A, let Aet denote the etale topo-
logical type (Spec A)et. For any connected, noetherian simplicial scheme
X (e.g., Spec A for a noetherian domain A), Xet is an inverse system of
connected simplicial sets [7, 4.4]. If X is normal in each simplicial de-
gree, as is always the case in our arithmetic examples, Xet is homotopy
equivalent to an inverse system in the homotopy category of simplicial
sets each of which has finite homotopy groups [7, 7.3].
Two key properties of Aet are given in the following
1.1 Proposition. Let a : Spec k ! Spec A be a geometric point of the
finitely generated normal R-algebra A, corresponding to a homomor-
phism from A to a separably closed field k and determining a base point
of Aet. Then
(1) ss1(Aet; a) = sset1(A; a) , the (profinite) Grothendieck fundamental
group of A pointed by a (this group classifies finite etale covering
spaces of Spec A).
(2) H* (Aet; C) = H*et(Spec A; C); the etale cohomology of Spec A
with coefficients in the local coefficient system C (such a local co-
efficient system is an abelian group C provided with a continuous
action of sset1(A; a)).
The etale K-theory space associated to Aet is constructed using func-
tion complexes. Namely, for each integer n, we consider the group
scheme GL n;R whose coordinate algebra is the Hopf algebra
R[xi;j; t]=det (xi;j)t = 1 :
The usual bar construction [7, 1.2, x2] (over R) determines a simplicial
scheme BGL n;R with etale topological type (BGL n;R)et.
1.2 Definition: The space BGL n(Aet) is defined to be (see [4, 2.4,
4.5])
BGL n(Aet) Hom 0`(Aet; (BGL n;R)et)Ret
the connected component of the function complex of maps over Ret from
Aet to the pro-space fibrewise mod ` completion over Ret of (BGL n;R)et.
More generally, if GR is a group scheme over R, we define B G(Aet) to
be
B G(Aet) Hom 0`(Aet; (B GR )et)Ret :
4 Dwyer and Friedlander
Remark: The curious reader might wonder how in general terms to
obtain a single space as a function complex of pro-spaces. The answer
is relatively simple: take the homotopy colimit indexed by the indexing
category of the domain and then the homotopy limit indexed by the
indexing category of the range.
1.3 Definition: Let A be a finitely generated R-algebra of finite mod `
etale cohomological dimension. The (connected) etale K-theory space
BGL (Aet) is defined to be
BGL (Aet) colim n{(BGL n(Aet)}
and the i-th mod ` etale K-group of A is defined to be
Keti(A; Z=` ) ssi(BGL (Aet); Z=` ) i > 1:
Remark: The homotopy fibre of the map (BGL n;R)et ! Ret (see 1.2)
is `-equivalent to the completion F! (BU n) [7, x8]; this leads to an in-
terpretation of ss* BGL (Aet) as the "twisted topological K-theory" of
the pro-space Aet. There is an Atiyah-Hirzebruch spectral sequence to
support this interpretation [4, 5.1]. Using the fact that both algebraic
K-theory and etale K-theory are the homotopy groups of infinite loop
spaces and that the map between these spaces is an infinite loop map [4,
4.4], one can extend the above definition to include K0 and K1. More-
over, one can extend this definition to certain rings of virtually finite
etale cohomological dimension (e.g., Z[1=2] at the prime 2) by a "de-
scent procedure" that involves taking homotopy fixed point spaces [5,
p. 140].
1.4 Remark: For any group scheme GR over R there is a natural map
B G(A) ! B G(Aet)
constructed as follows. One views a t-simplex of B G(A) as a morphism
of simplicial schemes Spec A [t] ! B GR over Spec R; by functorality
this determines a map of pro-spaces (Spec A)et x [t] ! (B GR )et over
Ret, and thus a t-simplex of B G(Aet). In particular, there are maps
BGL n(A) ! BGL n (Aet); these pass to a map
BGL (A)+ ! BGL (Aet)
which induces homomorphisms Ki(A; Z=` ) ! Keti(A; Z=` ).
Before proceeding to new results concerning the maps of 1.4, we briefly
summarize some earlier results from [4].
Topological Models 5
1.5 Theorem. Denote the map Ki(A; Z=` ) ! Keti(A; Z=` ) by i (A).
(1) If A is a finite field of order q where ` does not divide q, then
i (A) is an isomorphism for all > 0.
(2) If A is the ring of S-integers in a global field and contai*
*ns a
primitive 4'th root of unity if ` = 2, then for i > 0 and > 0 the
map i (A) is surjective.
(3) The well-known "Lichtenbaum-Quillen Conjecture" is equivalent
to the conjecture that the maps in (2) are isomorphisms.
(4) More generally, if the mod ` etale cohomological dimension of A
is 2 (e.g., if A is a local field or the coordinate algebra of a
smooth affine curve over a finite field), then for i > 0 and > 0
the map i (A) is surjective.
The preceding discussion should provide motivation for our efforts to
obtain computable models for BGL n (Aet). Such models will be function
spaces with domain XA , a "good mod ` model for A", which cover a
structure map reflecting the action of sset1(A) on the `-primary roots of
unity.
Let R1 denote the ring obtained from R by adjoining all `-primary
roots of unity,
R1 R[`1 ] Z[1=`; `1 ] ;
let denote the pro-group
Gal (R1 ; R) {Aut (` ); > 0} ~= {(Z=` )*; 1} ;
and let denote the quotient Aut (`) ~= (Z=`)* of . Observe that Ret
is provided with the natural structure map
Ret ! K(; 1)
which "classifies" the finite etale extensions R ! R[` ]. More generally,
if A is any R-algebra, then Aet is provided with a natural structure map
to Aet ! Ret ! K(; 1).
We denote by Z=`n (i) the local system on K(; 1) given by the i-fold
tensor power of the natural action of on `n. For any pro-space X
provided with a structure map X ! K(; 1), we also use Z=`n (i) to
denote the induced local system on X.
The next definition formalizes a modelling technique implicit in [5].
1.6 Definition: Let A be a finitely generated R-algebra. Then a map
of pro-spaces
f A : XA ! K(; 1)
6 Dwyer and Friedlander
is said to be a good mod ` model for A (or for Aet) if there exist pro-
spaces Xi over K(; 1); each of finite mod ` cohomological dimension,
and a chain of maps
XA = Xm Xn-1 ! . . .! X0 = Aet
over K(; 1) satisfying the condition that Xj ! Xj1 induces isomor-
phisms
H* (Xj1 ; Z=`(i)) ~= H* (Xj; Z=`(i)); i = 1; . . .; ` - 1 :
Remark: For the purposes of this paper it would be enough to work
over K(; 1) instead of over K(; 1) in defining a good mod ` model
(cf. proof of 1.8). There are other situations in which this would not be
sufficient.
1.7 Remark: The condition of 1.6 is equivalent to the condition that
the maps Xj ! Xj1 induce mod ` homology equivalences
"Xj ! X"j1
on covering spaces determined by Xj1 ! K(; 1) ! K(; 1); this, in
turn, is equivalent to the condition that X"j ! X"j1 induces isomor-
phisms
H* (X"j1 ; Z=`n (i)) ~= H* (X"j; Z=`n (i)); all n; i > 0;
or, in fact, to the condition that Xj ! Xj1 induces isomorphisms
H* (Xj1 ; Z=`n (i)) ~= H* (Xj; Z=`n (i)); all n; i > 0:
The significance for us of a good mod ` model is given by the following
proposition.
1.8 Proposition. Let f A : XA ! K(; 1) be a good mod ` model for
Aet. Then the chain of maps XA = Xm . . .! X0 = Aet extends to
a chain of pairs (of pro-spaces)
BGL n;XA - - - - BGL n;Xm-1 --- - ! . . .---- ! Fo!(BGL n;A)et
?? ? ?
y ?y ?y
XA - - - - Xm-1 --- - ! . . .---- ! Aet
Topological Models 7
with each map between pairs inducing a homotopy equivalence on homo-
topy fibres (i.e., with each map between pairs determining a homotopy
cartesian square). As a consequence, the space BGL n (Aet) is homotopy
equivalent to the space of sections (XA ; BGL n;XA ) of BGL n;XA ! XA .
Proof: Given Xi ! Xi-1 and BGL n;Xi-1 ! Xi-1 ; we define
BGL n;Xi ! Xi
as the evident pullback.
Given Xi ! Xi+1 and BGL n;Xi ! Xi, we proceed as follows to define
BGLn;Xi+1 ! Xi+1 : Suppose first that the map ss1(Xi) ! ! is
trivial. The homotopy fibre F of the map BGL n;Xi ! Xi is `-equivalent
to F! (BU n) and ss1Xi acts trivially on the mod ` (co)-homology of this
pro-space (cf. [4, p. 260]). It follows from the fibre lemma [2] that F is
equivalent to the homotopy fibre of the completed map F! (BGL n;Xi) !
F! (Xi), or to the homotopy fibre of the composite map F! (BGL n;Xi) !
F! (Xi) ! F! (Xi+1 ). (Observe that Xi ! Xi+1 gives an isomorphism
on homology and so induces an equivalence F! (Xi) ! F! (Xi+1 )). The
pro-space BGL n;Xi+1 is then the homotopy pullback of F! (BGL n;Xi) !
F! (Xi+1 ) over the completion map Xi+1 ! F! (Xi+1 ).
In the general case, let X"j , j = i, i + 1 be the connected covering
space determined by the composition
Xj ! K(; 1) ! K(; 1)
and let oj denote the corresponding group of covering transforma-
tions. The conditions of 1.7 imply that oi = oi+1 ; let o denote this
common group. Let BGL n;X"idenote the corresponding covering space
of BGL n;Xi. Construct a pro-space "BGL n;X"i+1" as above. By natural-
ity this pro-space has an action of the group o , and we define BGL n;Xi+1
as Eo xo BGL n;X"i+1; the desired map BGL n;Xi ! BGL n;Xi+1 can be
obtained by combining the o -equivariant map BGL n;X"i! BGL n;X"i+1
with some chosen o -equivariant map BGL n;X"i! Eo and then passing
to o orbits.
Finally, the homotopy equivalence of spaces of sections is proved by
comparing homotopy spectral sequences [5, 3.2]. The space BGL n(Aet)
is homotopy equivalent to (A; Fo!(BGL n;A)et) because
Fo!(BGL n;A)et--- - ! Fo!(BGL n;R)et
?? ?
y ?y
Aet --- - ! Ret
8 Dwyer and Friedlander
is homotopy cartesian. ||
We conclude this section with the following example from [5, x4]. Its
verification entails interpreting certain cohomological calculations aris-
ing from class field theory.
1.9 Example: Let A denote the ring R[i], and assume that ` is odd
and regular in the sense of number theory; this last means that ` does
not divide the order of the ideal class group of A or equivalently that
` does not divide the order of the ideal class group of Z[i]. Choose a
prime p which is congruent to 1 mod ` but is not congruent to 1 mod `2.
Let _S1 denote a bouquet of (` + 1)=2 circles, and construct a map
f A : (_S1 ) ! K(; 1)
by sending a generator of the fundamental group of the first circle to
p 2 (Z=` )* and mapping the other circles trivially. Then there exists a
map
_S1 ! Aet
over K(; 1) which satisfies the conditions of 1.7. Consequently, f A is a
good mod ` model for A.
x2. Z[1=`] with ` an odd regular prime
In this section we will assume that ` is an odd regular prime (see 1.9);
recall that R denotes the ring Z[1=`]. Fix a prime p which generates the
multiplicative group of units in Z=`2, let XR denote the space RP 1 _S1
(the wedge of real infinite projective space and a circle), and let
f R : XR ! K(; 1)
be a map with sends the generator of ss1(RP 1 ) ~= Z=2 to (-1) 2 (Z=` )*
and the generator of ss1(S1 ) ~= Z to p 2 (Z=` )*.
2.1 Theorem. The above map f R : XR ! K(; 1) is a good mod `
model for R.
Remark: In [5] there is a description of a parallel good mod 2 model
for the ring Z[1=2].
Let eR : RP 1 ! Ret be a homotopy equivalence and eZp : S1 !
(Zp)et a mod ` equivalence which sends the generator of ss1(S1 ) to the
Frobenius element in sset1Zp (3.2, 3.4). Given an embedding fl : Zp ! C,
we construct a commutative diagram of rings
R --- - ! R
?? ?
y ?y ;
fl
Zp --- - ! C
Topological Models 9
choose a basepoint (in an essentially unique way) in the contractible
pro-space Cet, and use this basepoint together with the commutative
diagram to obtain a map fflR (fl) : Ret _ (Zp)et ! Ret. Let eR (fl) denote
the composite map
eR _ eZp fflR (fl)
eR (fl) : RP 1 _ S1--- - - ! Ret _ (Zp)et--- ! Ret :
Theorem 2.1 is an immediate consequence of the following result.
2.2 Proposition. There exists an embedding fl : Zp ! C such that
the above maps fflR (fl) and eR (fl) each satisfy the conditions of 1.6.
Remark: The reader might be surprised to learn that one can always
choose an embedding fl0 : Zp ! C such that the conditions of 1.6 are not
satisfied by either fflR (fl0) or eR (fl0). Namely, the proof of Proposition
2.2 shows that choices of elements oe, o 2 G as in Remark 2.10 will
determine such a "bad" embedding.
Before proving 2.2, we point out the following corollary. In the state-
ment, F p refers to the space studied by Quillen [14].
2.3 Corollary. The etale K-theory space BGL (Ret) of R is homotopy
equivalent to
F1 (F p x U =O ) :
Proof of 2.3: By Proposition 1.8, we conclude that BGL (Ret) is ho-
motopy equivalent to
colim n{((Zp)et _ Ret; BGL n;(Zp)et_Ret )} :
We interpret each space in the colimit as a fibre product of
((Zp)et; Fo!BGL n;(Zp)et) and (Ret; Fo!BGL n;Ret)
over
(pt ; Fo!BGL n;pt) :
In any given dimension the homotopy groups of the spaces involved in
these fibre products stabilize as n gets large [4, 4.5], and we identify the
colimit of these cartesian squares as
BGL (Ret) --- - ! F1 BO
?? ?
y ?y
F1 (F p) --- - ! F1 BU
10 Dwyer and Friedlander
(see [5, 4.1 ff.]) so that BGL (Ret) fits into a fibration sequence
F1 (U =O )!- BGL (Ret)!- F1 (F p) :
The fibration BGL (Ret) ! F1 (F p) has a section given by the map
F1 (F p) = F1 (im Jp) ! F1 (B+1) ! BGL (Ret) (see for instance
[12, x4]) and, since this is a fibration of infinite loop spaces, the section
gives the desired product decomposition for BGL(Ret). (A more delicate
argument would show that the product decomposition holds even in the
category of infinite loop spaces). ||
2.4 Remark: Let C be the infinite cyclic subgroup of R* generated by
`. Consider the maps
f g
BGL (Z) x S1!- BGL (R)!- F1 (F p x U=O)
obtained on the one hand by passing to the limit with the evident direct
sum maps GL n(Z) x C ! GL n+1 (R) and on the other by combining
1.4 with 2.3. It follows from the localization theorem and the triviality
of K*(Z=`) at ` that the map f is an isomorphism on mod ` homology.
The Lichtenbaum-Quillen conjecture for the ring Z at the prime ` is
thus [5, 3.1] equivalent to the conjecture that the composite g . f is an
isomorphism on mod ` homology.
The rest of this section is concerned with proving 2.2. For simplicity
we will denote the space XR = RP 1 _ S1 by X, and we will let X
be the covering space of X corresponding to the kernel of the composite
homomorphism
ss1(fR )
ss1(XR )--- - ! !- :
We will denote the ring R[i] by R .
Let denote the one dimensional representation of F[] given by the
action of on ` R[i]*. The ring F[] is semisimple (by Maschke's
theorem) and
; 2 ; . . .; (`-2) ; (`-1) = Z=`
is a complete list of the (isomorphism classes) of irreducible F[] mod-
ules. If M is an F[] module, we denote by M [odd ] (resp., M [even ]) the
submodule of M generated by irreducible summands isomorphic to odd
(resp., even) powers of . We shall have frequent occasion to consider
the F[] module L given by
L = (F[])[odd ] = 3 . . .(`-4) (`-2) :
The module L is isomorphic as an F[] module to its F dual.
Topological Models 11
2.5 Lemma. There are isomorphisms of F[] modules
H1 (X ; Z=`) ~= Z=` L
H1(X ; Z=`) ~= Z=` L :
The groups Hi(X ; Z=`) vanish for i 2.
Proof: We construct X geometrically as follows. Above the circle of
X = RP 1 _ S1 , we place a circle (constituting part of X ) mapping to
X as an (` - 1) fold covering. We mark on this upper circle the (` - 1)
points given by the inverse image of the base point of X. The space
X is then defined to be the union of this circle with (` - 1)=2 copies of
the infinite sphere S1 , where each S1 maps via a 2-fold covering map
to RP 1 and where the two points of a given S1 mapping to the base
point of RP 1 are identifed with antipodal marked points on the circle
of X in such a way that each antipodal marked pair belongs to exactly
one S1 . The resulting space X is a cyclic covering space of X of degree
` - 1 and has the homotopy type of a wedge of (` + 1)=2 circles. The
action of the generator
oe 2 ss1S1 ss1X
on X rotates the circle of X through an angle 2ss=(` - 1) and operates
as continuity requires on the copies of S1 . The action of the generator
o 2 ss1(RP 1 ) ss1X
on X rotates the circle through an angle ss and operates as the antipodal
map on each S1 . Thus, the action of o is that of the oe(`-1)=2. Since
is a cyclic group of order ` - 1 generated by p there is a unique action
of on X under which p acts as oe and p(`-1)=2 = -1 as o .
The group acts trivially on the homology class of the circle in X ,
and so this circle determines a copy of the trivial F[] module Z=` in
H1(X ; Z=`). Collapsing this circle to a point and replacing each S1
by an arc yields a bouquet of (` - 1)=2 circles and it is is clear that the
action of on the dimension one Z=` homology of this bouquet gives
the representation L. The desired identification of H1(X ; Z=`) follows
from the long exact homology sequence of a pair and the fact that exact
sequences of F[] modules split; the identification of H1 (X ; Z=`) is by
duality. The fact that the higher cohomology groups of X vanish is
clear. ||
Recall that R denotes R[i].
12 Dwyer and Friedlander
2.6 Lemma. There are isomorphisms of F[] modules
H1et(R ; Z=`) ~=Z=` L
Het1(R ; Z=`) ~=Z=` L :
The groups Hiet(R ; Z=`) vanish for i 2.
Proof: Since R contains the `'th roots of unity `, there is an iso-
morphism of F[] modules
H1et(R ; Z=`) ~= H1et(R ; `) (-1)
where (-1) ~= (`-2) . Let E be the group of units in R and E`
the quotient group E=E`. There is for general reasons a short exact
sequence of F[] modules (cf. proof of 5.2)
0!- E`!- H1et(R ; `)!- `Pic (R )!- 0
and in this case the group `Pic (R ) (i.e., the group of elements of expo-
nent ` in the ideal class group of R ) is trivial because ` is regular. Let
E0 be the group of units in Z[i] and E0`the quotient E0=(E0)`. According
to [20, x8.3] there is an isomorphism of F[] modules
(`-3)=2M
E0`~= 2i
i=1
and so, since E` is the direct product of E0`and a cyclic group generated
by the image of ` 2 R*, there is an isomorphism
(`-3)=2M
E` ~= 2i ~= (Z=` L) :
i=0
The desired isomorphisms follow directly. The fact that the higher
(co)homology groups vanish is proved in [5, 4.4] (cf. 1.9). ||
We shall now let G denote the semidirect product of and the F[]
module L above. Up to isomorphism the group G is the only extension
of by L. Let XG ! X denote the covering space corresponding to
the kernel of the composition
ss1(X ) ! H1(X ; Z=`) ! H1(X ; Z=`)[odd ] ~= L :
Similarly, let R ! RG denote the finite etale map corresponding to the
kernel of the composition
ss1(R ) ! Het1(R ; Z=`) ! Het1(R ; Z=`)[odd ] ~= L :
In what follows we will assume that we have chosen an embedding RG
C which extends the inclusion R C. In view of 2.5 and 2.6, the
following lemma is elementary.
Topological Models 13
2.7 Lemma. The composition XG ! X ! X is a normal covering map
with group G. The composition R ! R ! RG is a Galois extension
with Galois group G.
Let p be the unique prime of R which lies above the rational prime p.
2.8 Lemma. The prime p splits completely in RG .
Proof: By Kummer theory, RG can be obtained from R by adjoining
an element u1=` for each u 2 R* such that u u-1 is an `'th power. (Here
u is the complex conjugate of u). For such a u, adjoining u1=` is the
same as adjoining (uu )1=`, so we can obtain RG from R by adjoining
`'th roots of real units. Let R+ = R \ R, let p+ be the prime of R+
below p , and choose u 2 (R+ )*. The residue class field R+ =p+ has no
`'th roots of unity, so raising to the `'th power is an automorphism of its
multiplicative group and the image of u in R+ =p+ has at least one `'th
root. By Hensel's lemma u has at least one `'th root in the completion of
R+ at p+ . Since ` R* , all of the `'th roots of u lie in the completion
of R at p . The lemma follows. ||
We will have to use some properties of the group G .
2.9 Lemma. The group G can be generated by two elements oe, o 2 G
with oe having order ` - 1 and o having order 2. Moreover, any non-
trivial element of G of order 2 is conjugate to o and any subgroup of G
of order ` - 1 is conjugate to the subgroup generated by oe.
2.10 Remark: There certainly exist elements oe, o 2 G of order (` - 1)
and 2 respectively which do not generate G; for example, choose any
oe 2 G of order (` - 1) and let o = oe(`-1)=2 .
Proof of 2.9: We employ the description of G as the group of deck
transformations of XG over X. Since the group
ss1X = ss1(RP 1 _ S1 ) = Z=2 * Z
is generated by two elements "oeand "o with "oeof infinite order and "o of
order 2, it suffices to prove that the image oe 2 G of "oehas order ` - 1.
This follows from the fact that the element
"oe`-12 ker{ss1(X) ! } = ss1(X ):
corresponds to the circle of X (see proof of 2.5); this circle represents
a homology class invariant under the action of and so by construction
lifts to a closed curve in XG . The assertions involving conjugacy follow
from combining the fact that H1 (0; L) = 0 for any subgroup 0
14 Dwyer and Friedlander
with the fact that H1 (0; L) can be interpreted as the set of L-conjugacy
classes of splittings of the extension
1 ! L ! G0 ! 0 ! 1
obtained by restricting the extension G of to 0. ||
Proof of 2.2: We will fix two elements oe, o 2 G as in 2.9; by the
uniqueness provision of 2.9, we can assume after perhaps changing oe up
to conjugacy that o is the restriction to RG C of complex conjugation.
Denote by Roe the ring fixed by the action of oe on RG . According to
2.8 the subgroup L G acts simply transitively on the set of all primes
of RG which like above p. By 2.9, the isotropy (=inertia) subgroups of
these primes correspond exactly to the subgroups of G of order ` - 1.
Let pG be such a prime which is fixed by oe, and poethe prime below pG
in Roe. The completion of Roeat poeis isomorphic to Zp, so we can view
the completion map as a map fl0 : Roe! Zp.
Let fl : Zp ! C be an embedding such that the composite fl . fl0 is
the inclusion Roe C. We will show that eR (fl) and fflR (fl) satisfy the
conditions of 1.7. All of the rings in question come with embeddings in
C and we will use the chosen basepoint of Cet to give basepoints for all
etale fundamental groups. The pullback of the diagram
fl0et
(Zp)et--! (Roe)et- (RG )et
is (Zp RoeRG )et, which is connected because (poebeing inert in RG ) the
ring Zp Roe RG is the completion of RG at pG and hence a domain. An
elementary covering space argument now shows that the image of the
composite map
sset1Zp!- sset1Roe-! sset1R!- G
contains (and is in fact equal to) the subgroup generated by oe; since
the image under eZp of ss1(S1 ) is dense in the profinite group sset1Zp, the
image of the map
ss1S1!- sset1Zp!- G
also contains oe. In the same way the image of the map ss1RP 1 ! sset1R
is equal to the subgroup generated by the element o 2 G representing
complex conjugation. By 2.9 the composite map
ss1(eR (fl))
ss1X = ss1(RP 1 _ S1 )--- - - - ! sset1R ! G
is surjective, so that the map ss1(X ) ! (R )et induced by eR (fl) induces
a surjection H1(X ; Z=`)[odd ]!- Het1(R ; Z=`)[odd ]. By 2.5. 2.6 and
Topological Models 15
a descent spectral sequence argument, the map H1(X ; Z=`)[even ]!-
Het1(R ; Z=`)[even ] induced by eR (fl) is the same as the induced map
H1(X; Z=`)!- Het1(R; Z=`) ~= Z=` :
This map is surjective because p is inert in the (unique) unramified
Galois Z=` extension of R; in fact, this extension is the maximal real
subring of the ring obtained from R by adjoining a primitive `2 root
of unity. (Here is the one place in which we use the fact that p is a
multiplicative generator mod `2 and not just mod `.) Consequently, the
map H1(X ; Z=`)!- Het1(R ; Z=`) induced by eR (fl) is an epimorphism
and hence by counting (2.5, 2.6) an isomorphism. This shows that eR (fl)
satisfies the conditions of 1.7. Since the map eR _ eZp also satisfies the
conditions of 1.7, the desired result for fflR (fl) follows at once. ||
x3. Affine curves and local fields
Let p be a prime number different from `. In this section, we construct
good mod ` models for
(1) affine curves over the algebraic closure Fp of Fp,
(2) affine curves over a finite field Fq where q = pd,
(3) higher p-adic local fields, and
(4) certain `-adic local fields.
Our first example, that of a smooth affine curve over Fp , is implicit
in [9].
3.1 Proposition. Let Y denote a smooth complete curve over Fp of
genus g and let A denote the coordinate algebra of the complement
U Y of s 1 points. Let XA F! (_S1 ) denote the `-adic completion
of the wedge of 2g + s - 1 circles. Then there exists a mod ` equivalence
Aet ! XA :
Furthermore, the structure map Aet ! K(; 1) is (homotopically) triv-
ial, so that XA (with the trivial structure map to K(; 1)) is a good
mod ` model for A .
Proof: Let T Witt (Fp ) denote the Witt vectors of Fp . Then the
"liftability of smooth curves to characteristic 0" asserts that there is a
proper, smooth map YT ! Spec T with (geometric) fibre Y over the
residue field of T . Moreover, we may find a closed subscheme ZT YT
with the properties that ZT is finite, etale over Spec T and that the fibre
over the residue field of UT YT \ZT is U . So constructed, UT ! Spec T
16 Dwyer and Friedlander
is a "geometric fibration", so that the `-adic completions of its geometric
and homotopy theoretic fibres are equivalent [8], [7, 11.5].
Let fl : Spec C ! Spec T correspond to an embedding T C. The
geometric fibre Yfl is a Riemann surface of genus g with s punctures,
thereby having the homotopy type of a wedge of 2g + s - 1 circles.
Since Tet is contractible, we conclude that the `-adic completion of the
homotopy theoretic fibre of UT ! Spec T is equivalent to XA . On
the other hand, Spec A is the geometric fibre of UT ! Spec T above
the residue field Spec Fp ! Spec T . Thus, Aet has `-adic completion
equivalent to XA . The triviality of the structure map Aet !- K(; 1)
follows from the fact that A contains all of the `-primary roots of unity. ||
3.2 Remark: The fact that the finite field Fq (q = pd) has a unique
extension inside Fp of degree s for any s > 0 and that each such extension
is Galois and cyclic with a canonical generator (the Frobenius) for its
Galois group implies that
(Fq)et ~= {K(Z=s; 1)}s>0 :
In fact, there is a natural mod ` equivalence S1 ! (Fq)et which sends
the generator of ss1S1 to the Frobenius map. The composite of this
mod ` equivalence with the structure map (Fq)et ! K(; 1) is the map
q : S1 ! K(; 1) which sends the generator of ss1S1 to q 2 Aut (` ) ~=
(Z=` )*.
3.3 Proposition. Let Y denote a smooth, complete curve over Fq of
genus g, let A denote the coordinate algebra of the complement U Y
of s 1 Fq-rational points, and let A denote A Fq Fp . Let XA be as
in 3.1. Then there exists an equivalence
OE : XA F! (_S1 ) ! F! (_S1 ) XA
such that the homotopy quotient XA XA xZ EZ of the action of
OE on XA is a good mod ` model for A; the structure map f A is the
composition
f A = q O pr : XA ! BZ = S1 ! K(; 1):
Moreover, if g = 0 then OE is a map which induces multiplication by q on
H1(XA ; Z=` ) for each > 0.
Proof: We view Spec A ! Spec Fq as a geometric fibration with geo-
metric fibre Spec A (cf. proof of Proposition 3.1). Thus, the homotopy
Topological Models 17
theoretic fibre of Aet ! (Fq)et is mod ` equivalent to XA [8]. Equiva-
lently, the mod ` fibrewise completion Fo!(Aet) of Aet fits in a fibration
sequence
XA ! Fo!(Aet) ! (Fq)et
Let S1 ! (Fq)et be given by sending a generator of ss1(S1 ) = Z to the
Frobenius map (as in 3.2). Then the pullback of the above fibration
sequence by S1 ! (Fq)et is a fibration sequence of the form
XA ! E ! S1
This implies that E is the homotopy quotient of OE , where OE : XA ! XA
is the "deck transformation" associated to a generator of ss1(S1 ), the
group of the covering space XA ! XA . Let XA = E.
By construction we have a commutative diagram of prospaces
Aet --- - ! Fo!(Aet) - - - - XA
?? ? ?
y ?y ?y
=
(Fq)et --- - ! (Fq)et - - - - S1
in which the horizontal arrows are (fibrewise) mod ` equivalences. Since
the structure map Aet ! K(; 1) factors through the structure map
(Fq)et ! K(; 1) and since composition of this latter map with S1 !
(Fq)et was checked above to be q we conclude that the indicated map
XA ! K(; 1) is a good mod ` model for A.
Finally, if g = 0 and s = 2 (so that A ~= Fq[t; t-1 ]), then the fact
that Pic A = Z in this case implies that there is a natural isomorphism
H1et(A ; ` ) ~= A* Z=` (cf. the exact sequence in the proof of 2.6). Ob-
serve that A* Z=` is invariant under Gal (Fp ; Fq) so that Gal (Fp ; Fq)
acts on H1et(A ; Z=` ) H1et(A ; ` ) -1` by multiplication by q-1 and
dually on Het1(A ; Z=` ) by multiplication by q. ||
Remark: The "Riemann Hypothesis for Curves" provides further infor-
mation about the self-equivalence OE of XA in Proposition 3.3. Namely,
we conclude that the induced map on `-adic homology is the sum of two
maps: the first is multiplication by q on (Z`)s-1 (as in Proposition 3.3)
and the second has eigenvalues of absolute value q1=2 on (Z`)2g.
3.4 Remark: If A is a hensel local ring (e.g., a complete discrete val-
uation ring) with residue field k and quotient field K, then the map
ket ! Aet is an equivalence, but the relationship between Ket and Aet
is considerably more subtle. The special case r = 1 of the following
proposition treats this question in a special case, as does 3.6.
18 Dwyer and Friedlander
As considered for example in [6] a generalized local field K of "tran-
scendental degree" r over a field k is a field for which there exists a finite
chain of homomorphisms
k = K0 O1 ! K1 . . . Or ! Kr = K
such that each Oj is a hensel discrete valuation ring with residue field
Kj-1 and quotient field Kj.
3.5 Proposition. Let K be a generalized local field of transcenden-
tal degree r over the finite field Fq and let S = S1 denote the circle.
Consider the space
XK F! (Sxr ) xZ EZ
which is the homotopy quotient of the map given by multiplication by q
on the completion of the r-fold cartesian power of S. If XK is provided
with the structure map
f K = q O pr : XK ! BZ ! K(; 1) ;
then it is a good mod ` model for K.
Proof: The proof of [6, 3.4] establishes a chain of ring homomorphisms,
each inducing an isomorphism in etale cohomology with Z=`(i) coeffi-
cients (i = 1; . . .; ` - 1), relating A Fq[t1; :::; tr; t-11; :::; t-1r] and *
*K.
Thus, a good mod ` model for A will also be a good mod ` model for K
(ring homomorphisms are always compatible with the structure maps to
K(; 1)).
We view Spec A as the complement in (PF1q)xr of a divisor with normal
crossings rational over Fq. Replacing Fq by T Witt (Fq), the Witt
vectors of Fq, we obtain U" = Spec A" (PT1)xr (i.e., the r-fold fibre
product of PT1 with itself over T ) with the property that U" ! Spec T
is a geometric fibration. Letting fl : Spec C ! Spec T be given by a
complex embedding of T , we observe that the geometric fibre of Ufl is
mod ` equivalent to Sxr . As in the proof of Proposition 3.3, this implies
the mod ` equivalences over K(; 1)
Aet ! Fo!(Aet) F! (Sxr ) xZ EZ :
Since the action of Gal (Fp ; Fp) on H1et(A; ` ) ~= A* Z=` is trivial,
we conclude as in the proof of Proposition 3.3 that the action of Z on
F! (Sxr ) is given by multiplication by q. ||
In the next proposition, we present a good mod ` model for a p-adic
field which looks a bit different from the one provided by the special
case r = 1 in Proposition 3.5.
Topological Models 19
3.6 Proposition. Let K be a p-adic field with residue field Fq. Con-
sider the 1-relator group ssK given by
ssK = < x; y : yxy-1 = xq> :
Then the map
f"K : K(ssK ; 1) ! K(; 1)
induced by the homomorphism ssK ! sending x to the identity and y
to q 2 Aut (` ) = (Z=` )* is a good mod ` model for K. Furthermore,
K(ssK ; 1) is homotopy equivalent to the (homotopy) pushout K of the
diagram
(S1 _ S1 ) S1 ! pt
in which the left map sends a generator of ss1(S1 ) to x-q yx-1 y-1 , where
x and y are the generators of ss1(S1 _ S1 ) associated to the wedge sum-
mands.
Proof: Using the relationship yxy-1 = xq, we readily verify that the
normal closure of the cyclic subgroup generated by x consists of frac-
tional powers of x with denominators a multiple of q. Hence, ssK lies
in an extension Z[1=p] ! ssK ! Z and K(ssK ; 1) fits into a fibration
sequence
(S1 )1=p ! K(ssK ; 1) ! S1 :
Let XK be the good mod ` model for K resulting from the case r = 1
of 3.5. There is evidently a commutative square
K(ssK ; 1)--- - ! XK
?? ?
y ?y
=
S1 --- - ! S1
in which the induced map on fibres is a mod ` equivalence. We conclude
that f"K : K(ssK ; 1) ! K(; 1) is a good mod ` model for K.
The equivalence K ' K(ssK ; 1) follows from a theorem of Lyndon
(cf. [3, p. 37]) in view of the fact that the relator of ssK is not a power. *
*||
Our last example is that of an `-adic field containing a primitive
`'th root of unity. The proposition below is based upon a theorem of
Demuskin [17].
3.7 Proposition. Let L be an `-adic local field. i.e., the completion
of a number field F at a prime with residue field of characteristic `.
Assume that `s L, where `s > 2, that `s+1 6 L, and that 2g is the
20 Dwyer and Friedlander
degree of L over Q`. Let GL denote the 1-relator group generated by
2g + 2 elements x; y; x1; . . .; x2g subject to the defining relation
s+1 -1 -1
x` yx y = [x1; x2] . .[.x2g-1 ; x2g] :
Then there is a homomorphism GL ! such that the classifying map
f L : K(GL ; 1) ! K(; 1) is a good mod ` model for L. Moreover,
K(GL ; 1) is homotopy equivalent to the (homotopy) pushout L of the
diagram
(S1 _ S1 ) S1 ! (_2gS1 )
in which the left map sends a generator t of ss1(S1 ) to x`s+1 yx-1 y and the
right map sends t to the product of commutators [x1; x2]::::[x2g-1 ; x2g]
(here x; y; x1; :::; x2g are generators of corresponding fundamental groups
associated to wedge summands).
Proof: The second assertion follows from [3, p. 50]; it implies that GL
has cohomological dimension 2.
Let L denote the Galois group Gal (L ; L), where L is the algebraic
closure of L, and let "b " denote profinite `-completion. The completion
map Let ~= K(L ; 1) ! K(^L ; 1) induces an isomorphism on H*(- ; Z=`)
by [16, Ch. II, Prop. 20] and hence in particular [16, Ch. II, Prop. 15]
both L and ^L have cohomological dimension 2.
By [17] (cf. [16, Ch. II, x5.6]) there is an isomorphism G^L ~= ^L . The
map GL ! G^L induces a map
Hi(K(GL ; 1); Z=`) ! Hi(K(G^L ; 1); Z=`)
which for general reasons is an isomorphism when i 1 and an epi-
morphism for i = 2. Since H2(K(GL ; 1); Z=`) is isomorphic to Z=` by
the above construction of K(GL ; 1), H2(K(G^L ; 1); Z=`) is isomorphic to
Z=` by [16, Ch. II, Th. 4], and the higher Z=` homology groups of these
pro-spaces vanish, we conclude that in the diagram
ff fi
Let ~= K(L ; 1)!- K(^L ; 1) ~= K(G^L ; 1)- K(GL ; 1)
both of the maps ff and fi induce isomorphisms on Z=` homology. Since
L contains ` the local systems Z=`(i) are all isomorphic to the trivial
system Z=`. Thus, if we define f L : K(GL ; 1) ! K(; 1) to be given
by the composite
GL ! G^L ~= ^L ! ;
the map f L is a good mod ` model for L. ||
Topological Models 21
x4. The Eilenberg-Moore spectral sequence
In this section, we describe the machinery from algebraic topology
which will be used to prove the cohomological injectivity statements
in x 6. Our main tool will be the Eilenberg-Moore spectral sequence,
denoted EMSS; we will use it to study the cohomology of function com-
plexes and in x6 to investigate the maps in cohomology induced by con-
tinuous maps between function complexes. This resembles Quillen's use
of the EMSS in [14, x1-5].
Throughout this section we will consider graded algebras over the
field F, in particular, we shall use H* X to denote the mod ` cohomology
H* (X; Z=`) of a space X. We begin with a brief discussion of the EMSS;
further details may be found in [18]. We consider a cartesian square of
spaces with f (and thus f 0) a fibration:
E0 --- - ! E
? ?
f0?y ?yf
B0 --- - ! B
where B is assumed to be simply connected and the cohomology (i.e.,
mod ` cohomology) of each space is assumed finite in each dimension.
Then there is a strongly convergent second quadrant spectral sequnce
E-s;t2= Tor -s;tH*B(H* (B0); H* E) ) H* (E0)
whose r'th differential takes the form
dr : E-s;tr! E-s+r;t-r+1r :
This is a spectral sequence of algebras; each differential is a derivation.
Recall that if S is a (positively) graded commutative algebra and
M , N are graded modules over S, then Tor -s;tS(M; N ), where s is the
homological degree and t is the internal degree, is explained as follows.
If
. . .! Fn ! Fn-1 ! . . .! F0 ! M
is a projective resolution of M (as a graded right S-module) and Fn0=
Fn S N , then Tor -s;tS(M; N ) denotes the degree t component of the
graded module
ker{Fs0! Fs0-1 }= im {Fs0+1 ! Fs0} :
22 Dwyer and Friedlander
Let S be a graded commutative algebra (e.g., S = H* B). One of
the most convenient contexts in which to compute Tor S (M; N ) is that
in which M is a quotient of S by an ideal I generated by a regular
sequence x1; . . .; xn of homogeneous elements. Then one obtains a very
efficient free resolution of M , the Koszul resolution, which takes the
following form:
. . .! S n V ! S n-1 V ! . . .! S V ! S ! M
where V is a vector space spanned by elements y1; . . .; yn and where the
tensor products are taken over F.
The following proposition, an almost immediate consequence of the
existence of the Koszul resolution, is essentially in [18, 2.10].
4.1 Proposition. Let S be the polynomial algebra F[x1; :::; xn ] graded
so that each xi is homogeneous of degree di. Let S(r) denote the r-fold
tensor power of S over F, S(r) Sr , and give S the structure of an
S(r) module by means of the multiplication map. Then Tor *S(r)(S; S) is
naturally isomorphic as an S algebra to the (r - 1)-fold tensor power of
the DeRham complex of S:
Tor *S(r)(S; S) ~= *S=F S . . .S *S=F :
Remark: The DeRham complex *S=F is an exterior algebra over S
generated by elements dxi (i = 1; . . .; n) of homological degree 1 and
internal degree di.
Proof of 4.1: The kernel of the multiplication map S(r) ! S is gen-
erated by the regular sequence
{xi;j- xi;r; 1 i n; 1 j < r}
in
S(r) ~= F[x1; :::; xn ]r ~=F[xi;j; 1 i n; 1 j r]:
We obtain the asserted computation by taking the Koszul resolution of S
as an S(r) module using this regular sequence, tensoring this resolution
over S(r) with S, and observing that the resulting complex has trivial
differentials. ||
With the aid of Proposition 4.1, the EMSS easily provides the follow-
ing computation.
Topological Models 23
4.2 Proposition. Let X be a simply connected space with H* X iso-
morphic to the finitely generated polynomial algebra S = F[x1; :::; xn ].
Let Y denote the space
Y Map (_S1 ; X)
of unpointed maps from a bouquet of r circles to X. Then H* Y admits
a filtration associated to the EMSS whose associated graded module is
naturally isomorphic to the r-fold tensor power of the DeRham complex
of S:
gr H* Y ~=*S=F S . . .S *S=F:
Proof: Let _D1 be the one-point union of r copies of the unit interval
D1 formed by glueing these copies together at 0 2 D1. The space
_S1 can be obtained from _D1 by further identifying the r images 1i
(i = 1; . . .; r) of 1 2 D1 with the common image of 0. This implies that
the space Y fits into a cartesian square
Y --- - ! Map (_D1; X) ' X
?? ?
y ?ye
X --- - ! Xr+1
where is the diagonal and e is the fibration given by evaluation at
various interval endpoints. Under the basepoint evaluation equivalence
Map (_D1; X) ' X the map e is homotopic to the diagonal. Using
Proposition 4.1, we identify the E2 term of the EMSS for this square as
M
E-s;*2= s1S=FS . . .S srS=F:
s=s1+...+sr
The shape of this spectral sequence implies that E0;*2and E-1;*2 consist
of permanent cycles; since these cycles generate E*;*2multiplicatively,
we conclude that E*;*2consists of permanent cycles, or, in other words,
that E2 = E1 . ||
The following generalization of Proposition 4.2 will be useful; it is
proved in the same way as 4.2. Let _S1 be a bouquet of r circles, and
f : E ! _S1 a fibration with homotopy fibre X = f -1(*). Let _D1 be
the space in the proof of 4.2 and f 0: E0 ! _D1 the (trivial) fibration
obtained by pulling E back over the quotient map _D1 ! _S1 . As in
24 Dwyer and Friedlander
the proof of 4.2, there is a homotopy fibre square for the space of sections
(f ) of f
(f ) --- - ! (f 0) ' X
?? ?
y ?ye
X --- - ! Xr+1
where again is the diagonal and e is given by evaluation at the points
0 and 1i (i = 1; . . .; r) in _D1. Under the basepoint evaluation equiv-
alence (f 0) ' X the map e is homotopic to a product (OE0; . . .; OEr),
where OE0 is the identity map of X and each OEi, i = 1; . . .; r is a self-
equivalence of X. We will call the self-equivalences OEi the monodromy
maps of the fibration f . The induced cohomology maps OE*igenerate the
usual monodromy action of the free group ss1(_S1 ) on H* X.
4.3 Proposition. Let f : E ! _S1 be a fibration with fibre X over
a bouquet _S1 of r circles. Assume that X is simply connected. Let
OEi, i = 1; . . .; r be the monodromy maps of f . Let e : X ! Xr+1
be the product (id; OE1; . . .; OEr), and eH* X the module over H* (Xr+1 )
obtained using e*. Then there is a strongly convergent EMSS
Tor -s;tH*(Xr+1()H* X; eH* X) ) H* ((f )) :
Finally, if H* X is isomorphic to a finitely generated polynomial algebra
S over F and ss1(_S1 ) acts trivially on H* X then this EMSS collapses
to an isomorphism
gr H* ((f )) ~= *S=F S . . .S *S=F:
x5. Algebraic and topological tori
In this section, we verify the injectivity of the map on mod ` coho-
mology induced by the natural map
BTn (A) ! BTn (Aet) ;
where A is an R algebra and Tn;R is the rank n, R-split torus given by
Tn;R = Spec R[t1; t-11; . . .; tn ; t-1n]:
The map in question is the n-fold cartesian power of the map BT1(A) !
BT1(Aet) associated to the special case n = 1, and so it is this special
case that we will concentrate on. This discussion prepares the way for the
consideration in x6 of the more delicate map BGL n(A) ! BGL n(Aet) :
We begin with the construction of non-connected versions of BT1(A)
and BT1(Aet).
Topological Models 25
5.1 Lemma. Let A be a noetherian R algebra. There exists infinite loop
spaces
BT"1(A) Hom g(A; BT1;R ) and BT"1(Aet) Hom `(Aet; (BT1;R )et)Ret
with the following properties:
(1) BT1(A) and BT1(Aet) are respectively the identity components
of BT"1(A) and BT"1(Aet).
(2) ss0(BT"1(A)) = Pic (A).
(3) The map OE0A : BT1(A) ! BT1(Aet) is the restriction to identity
components of a map of infinite loop spaces OEA : BT"1(A) !
BT"1(Aet).
Proof: The construction of Hom g (A; BT1;R ) and the natural map OEA
of (3) are provided in [4, x2]. One verifies that this map is the map on
zero spaces of a map of spectra as in [4, x3, 4.4] using the multiplicative
structure on BT1;R given by the commutative group scheme T1;R . Now,
(2) follows from [4, A.6] and (1) is implicit in the constructions just
described. ||
The following proposition is implicit in [4, p. 274].
5.2 Proposition. Let A be a noetherian R algebra such that Pic (A)
contains no infinitely `-divisible elements. Then the natural map (1.4)
BT1(A)!- BT1(Aet) induces an isomorphism
H* (BT1(Aet); Z=`)!- H* (BT1(A); Z=`) :
Proof: Let F denote the homotopy fibre of OEA (see 5.1); observe that
F has a natural basepoint. It is enough to show that ssi(F ) is uniquely
`-divisible for i 1 and that the cokernel of the map ss1(OEA ) is also
uniquely `-divisible.
As in [4, p. 294], there is map of fibration sequences
`
Hom g(A; B`)R --- - ! BT"1(A) --- - ! BT"1(A)
? ? ?
~=?y OEA?y OEA?y
`
Hom `(A; (B`)et)Ret --- - ! BT"1(Aet) --- - ! BT"1(Aet)
The statement for ssi(F ), i 1, follows from completing this map of
fibration sequences to a 3 x 3 square by taking vertical fibres.
The map ss1(OEA ) can be identified as a natural map in etale cohomol-
ogy
H0 (Spec A; Gm ) ! H1 (Spec A; Z`(1))
26 Dwyer and Friedlander
where Gm = T1;R denotes the multiplicative group. This map appears
as one of the rungs in a long exact etale cohomology ladder in which the
top rail is the long exact sequence associated to
`
0 ! ` ! Gm !- Gm ! 0
and the bottom rail is the long exact sequence assoicated to
`
0 ! Z`(1)!- Z`(1) ! ` ! 0 :
In this ladder, the rung maps involving H* (Spec A; `) are the identity.
Let C denote the cokernel of ss1(OEA ). We first show that C has no
`-torsion with the following straightforward diagram chase. Pick x0 2 C
such that `x0 = 0. Choose x1 2 H1 (Spec A; Z`(1)) projecting to x0. By
assumption, `x1 is the image of an element x2 2 H0 (Spec A; Gm ). The
image of x2 in H1 (Spec A; `) is the same as the image of x1, namely 0,
so that x2 = `x3 for some x3 2 H0 (Spec A; Gm ). Let x4 be the image of
x3 in H1 (Spec A; Z`(1)). Clearly, `(x1 - x4) = 0, so that x1 - x4 is the
image of an element x5 2 H0 (Spec A; `). Let x6 = x3 + x5. Then the
image of x6 in H1 (Spec A; Z`(1)) is x1 and hence x0 = 0.
We finally verify that C is `-divisible. Consider x0 2 C. Pick
x2 2 H1 (Spec A; Z`(1)) projecting to x0, and let x3 be the image of
x2 in H1 (Spec A; `). Consider first the case that x3 is the image of
an element x4 from H0 (Spec A; Gm ). Let x5 be the image of x4 in
H1 (Spec A; Z`(1)): It is clear that x2 - x5 has image 0 in H1 (Spec A; `)
and so x2 - x5 = `x6 for some x6 2 H1 (Spec A; Z`(1)). Then x0 = `x1,
where x1 is the image of x6 in C: Assume on the contrary that x3 in not
the image of any such x4. Let x7 be the image of x3 in H1 (Spec A; Gm );
it follows that x7 6= 0. Carrying out an inductive argument along similar
lines with the long exact ladders associated to the sequences
0 ! Z`(1) ! Z`(1) ! `i ! 0
and
0 ! `i ! Gm ! Gm ! 0
shows that there are elements x7(i) 2 H1 (Spec A; Gm ) such that x7(1) =
x7 and `x7(i) = x7(i - 1). In view of the hypothesis on Pic (A), this is
impossible. ||
Proposition 5.2 has the following immediate corollary.
Topological Models 27
5.3 Corollary. Let A be a noetherian R algebra such that Pic (A)
contains no infinitely `-divisible elements. Then the natural map (1.4)
BTn (A) ! BTn (Aet) induces an isomorphism
H* (BTn (Aet); Z=`) ! H* (BTn (A); Z=`) :
x6. Cohomological injectivity
In this section, we show that if A = R[i] = Z[1=`; i] and ` is a regular
prime (1.9), then the natural cohomology map (1.4)
H* (BGL n(Aet); Z=`) ! H* (BGL n(A); Z=`)
is a monomorphism. In this case, then, our good mod ` model for A
gives an explicit lower bound for H* (BGL n(A); Z=`): For the ring R
itself with ` regular, we obtain an analogous, but stable, result. Our
basic technique is to show that the natural map
BTn (Aet) ! BGL n (Aet)
induces an injection in mod ` cohomology and then apply Corollary 5.3.
We shall let S denote the graded polynomial algebra
S F[c1; . . .; cn ], deg (ci) = 2i :
The reader should recall that S is isomorphic to the mod ` cohomol-
ogy ring of the classifying space BGL n(Ctop) of the complex Lie group
GLn (Ctop) or to the mod ` cohomology ring of BU n. Let S" denote the
graded polynomial algebra
"S F[t1; :::; tn ], deg (ti) = 2 :
Then S" is isomorphic to the mod ` cohomology ring of the classifying
space BTn (Ctop) of a maximal torus of GL n (Ctop) or to the mod `
cohomology ring of the classifying space BTn of the group Tn U n of
diagonal matrices. The restriction map
S ~= H* (BU n; Z=`) ! H* (BTn ; Z=`) ~= "S
includes S "Sas the ring of symmetric polynomials in the variables ti.
6.1 Remark: If XA is a good mod ` model for the R algebra A, then
for each n > 0 there is a fibration BGL n;XA ! XA with homotopy fibre
F! (BU n) (see 1.8) and the technique of 1.8 produces a parallel fibra-
tion BTn;XA ! XA with homotopy fibre F! (BTn ). There is a map
28 Dwyer and Friedlander
BTn;XA ! BGL n;XA which on fibres is the completion of the natural
map BTn ! BU n . We will denote the spaces of sections of these fi-
brations by, respectively, BGL n(XA ) and BTn (XA ). If XA is a space
(as opposed to a pro-space), then BGL n(XA ) is the space of sections
of the ordinary fibration lim {BGL n;XA } ! XA with homotopy fibre
F1 (BU n); a similar remark holds for BTn (XA ). According to the final
statement in 1.8 (as extended to cover Tn;A as well as GL n;A) there is
up to homotopy a commutative diagram
'
BTn (A) --- - ! BTn (Aet) --- - ! BTn (XA )
?? ? ?
y ?y ?y :
'
BGL n(A) --- - ! BGL n(Aet) --- - ! BGL n (XA )
6.2 Lemma. Let A be a finitely generated R algebra and let f A : XA !
K(; 1) be a good mod ` model for A. Assume that XA has the mod `
homology of a bouquet of r > 0 circles, and further assume that the
composite XA ! K(; 1) ! K(; 1) is homotopically trivial. Then
there are gradings on H* (BGL n(XA ); Z=`) and on H* (BTn (XA ); Z=`)
compatible with the natural map (6.1) between these two cohomology
rings, and a commutative diagram
gr H* (BGL n(XA ); Z=`) --- - ! gr H* (BTn (XA ); Z=`)
? ?
~=?y ?y~=
*SS . . .S *S --- - ! *"SS". . .S"*"S
in which the lower horizontal arrow is the r-fold tensor power of the map
*S=F ! *"S=Fof DeRham complexes induced by S ! S". In particular
the map H* (BGL n(XA ); Z=`) ! H* (BTn (XA ); Z=`) is injective.
Proof: By replacing XA if necessary (see 1.8) we can assume that XA
is a bouquet _S1 of r circles and that BGL n(XA ) (resp. BTn (XA ))
is the space of sections of a fibration over _S1 with fibre F1 (BU n)
(resp. F1 (BTn )) (see 6.1). The triviality of _S1 ! K(; 1) ! K(; 1)
implies that ss1(_S1 ) acts trivially on the mod ` cohomology groups of
the fibres in these fibrations. The existence of the given commutative
diagram is a direct consequence of 4.3; the gradings on the cohomology
rings in question are the gradings associated to appropriate collapsing
Eilenberg-Moore spectral sequences (4.1). Let S(r + 1) and S"(r + 1)
denote the (r + 1)-fold tensor powers of S and S" respectively. The
indicated map on DeRham complexes is the natural map
Tor *S(r+1)(S; S) ! Tor *"S(r+1)(S"; "S)
Topological Models 29
induced by S ! S". One can view this map explicitly by first mapping
the Koszul resolution of S over S(r + 1) to the Koszul resolution of S"
over S"(r + 1), then tensoring the corresponding resolutions by S or S".
Inspection verifies that this map is injective. For another proof of this
injectivity see [14, Lemma 9]. ||
Lemma 6.2 leads quickly to the result we want.
6.3 Proposition. Assume that ` is an odd regular prime and let A
denote the ring R[i] = Z[1=`; i]. Let r = (` + 1)=2. Then the natural
map (1.4) BGL n(A) ! BGL n (Aet) induces a monomorphism
H* (BGL n(Aet); Z=`) ! H* (BGL n(A); Z=`):
Moreover, H* (BGL n(Aet); Z=`) admits a filtration for which the associ-
ated graded module is isomorphic to *SS . . .S *S, the r-fold tensor
power of the DeRham complex of S.
Remark: The proof of 6.3 goes through without change if A is the co-
ordinate algebra of a multiply punctured plane over the algebraic closure
of a field of characteristic different from `, i.e., in the genus 0 case of
3.1. The proof fails for affine curves of higher genus, since Pic (A) then
contains divisible subgroups (cf. 5.3).
Proof of 6.3: It follows from 1.9 that A has a good mod ` model XA
which is a bouquet of circles; this is also a consequence of 2.1, since the
fact that RP 1 _ S1 is a model for R implies easily that the cover of this
space described in 2.5 is a model for R[i]. Since A contains `, the com-
posite XA ! K(; 1) ! K(; 1) is trivial. By 6.2 and the discussion
in 6.1, then, the map BTn (Aet) ! BGL n(Aet) induces a monomor-
phism on mod ` cohomology. By 5.3, the map BTn (A) ! BTn (Aet)
induces an isomorphism on mod ` cohomology. The commutative di-
agram in 6.1 gives the desired injectivity statement. The formula for
gr {H* (BGL n(Aet; Z=`)} is from 6.2. ||
There is a stable version of Proposition 6.3 that holds for R as well
as for R[i].
6.4 Proposition. Assume that ` is an odd regular prime and let A
denote either the ring R = Z[1=`] or the ring R[i]. Then the natural
map (1.4) BGL (A) ! BGL (Aet) induces a monomorphism
H* (BGL (Aet); Z=`)!- H* (BGL (A); Z=`) :
6.5 Remark: The mod ` cohomology of BGL (Aet) can be determined
easily from 2.3 if A = R and by an analogous calculation or from [5, 4.6]
30 Dwyer and Friedlander
if A = R[i]. With almost no extra effort (cf. [5, p. 143]) results similar
to 6.4 can be obtained for any number ring R0 between R and R[i], or
for any normal etale extension R00 of such an R0 of degree a power of
`. (Observe for instance that by 1.9 or 2.1 there are many normal etale
extensions of R[i] of degree a power of `: these correspond to normal
subgroups of finite index in a free pro-` group on (` + 1)=2 generators.)
Proof of 6.4: For A = R[i] the result follows from passing to an
inverse limit over n with 6.3; on both sides the cohomology groups in
any given dimension stablilize with n to a fixed value. Recall from [4,
6.4] that the map BGL (A)+ ! BGL (Aet) is a map of infinite loop spaces
commuting with transfer. Consider the following commutative diagram
whose horizontal arrows are induced on the one hand by the transfer
and on the other hand by the ring homomorphism i : R ! R[i] (in this
diagram H* denotes H* (- ; Z=`)):
tr* i*
H* BGL (R) --- - ! H* BGL (R[i]) --- - ! H* BGL (R)
x? x x
? ?? ?? :
tr* i*
H* BGL (Ret) --- - ! H* BGL (R[i]et) --- - ! H* BGL (Ret)
i tr
Recall that the composite BGL (R)+ !- BGL (R[i])+ -! BGL (R)+ is
multiplication by (` - 1), in view of the fact that R[i] is a free R mod-
ule of rank (` - 1); for the same reason the composite BGL (Ret)!-
BGL (R[i]et)!- BGL (Ret) is multiplication by (` - 1). Since (` - 1) is
prime to `, this implies that both horizontal composites in the above di-
agram are isomorphisms. The middle vertical arrow is a monomorphism
by the considerations above. The fact that the left vertical arrow is a
monomorphism follows at once. ||
We will state our final proposition only for the ring R, although there
is a version for R[i] as well as for the other rings mentioned in 6.5. Let
Dn GL n(`-1)(R) be the image of the diagonal subgroup (R[i]*)n
GL n(R[i]) under the transfer map GL n(R[i])!- GL n(`-1) (R) obtained
by choosing a basis for R[i] as an R module.
6.6 Proposition. Suppose that ` is an odd regular prime and that
the Lichtenbaum-Quillen conjecture holds at ` for Z; equivalently (2.4),
assume that the map BGL (R) ! BGL (Ret) induces an isomorphism on
mod ` cohomology. Then H* (BGL (R); Z=`) is detected on the groups
Dn , in the sense that if x 2 H* (BGL (R); Z=`) is non-zero, then there
exists n such that x has nonzero image under the composite
H* (BGL (R); Z=`)!- H* (BGL n(`-1)(R); Z=`)!- H* (BDn ; Z=`) :
Topological Models 31
Proof: There are commutative diagrams
ff fi
BTn (R[i]) --- - ! BGL n(R[i]et) --- - ! BGL (R[i]et)
? ?
~=?y ?ytr
BDn --- - ! BGL n(`-1)(Ret) --- - ! BGL (Ret)
in which the right hand vertical map is the transfer and induces a
monomorphism on mod ` cohomology (cf. proof of 6.4). The map fi
is an equivalence in a stable range [4, 4.5] which tends to infinity as n
gets large. By 6.2 (cf. proof of 6.3) the cohomology map H* (ff; Z=`)
is a monomorphism. It follows that H* (BGL (Ret); Z=`) is detected on
the groups Dn ; if the Lichtenbaum-Quillen conjecture holds the same is
true of H* (BGL (R); Z=`). ||
Remark: The groups Dn which appear in 6.6 are closely analogous to
the subgroups used by Quillen [14, Lemmas 12 and 13] to detect the
cohomology of general linear groups over a finite field. Propositions 6.3,
6.4 and 6.6 are also true for ` = 2 (in this case R = R[i] = Z[1=2]). This
can be proved along the above lines using [5, 4.2]; it also follows directly
from the recent work of Mitchell [13].
x7. Further questions and problems
It seems reasonable to conclude this paper by mentioning a few of
the many open questions related to what is discussed here. In some
cases, the authors have obtained partial answers whereas in other cases
the answers remain completely elusive. Probably the most compelling
problem is to settle the Lichtenbaum-Quillen conjecture:
7.1 Problem: Determine whether or not the map
BGL (Z[1=`])!- BGL (Z[1=`]et)
induces an isomorphism on mod ` cohomology.
Somewhat less ambitious is
7.2 Question: Are there some general circumstances under which etale
K-theory splits off of algebraic K-theory? More specifically, if A is
a regular noetherian ring of etale cohomological dimension 2 with
1=` 2 A, does the map (1.4)
F1 (BGL (A)+ )!- F1 (BGL (Aet)) ' BGL (Aet)
32 Dwyer and Friedlander
have a right inverse? This would give a strong geometric explanation
for the cohomological injectivity results in x6 and the homotopical sur-
jectivity results in [4].
7.3 Problem: Develop a tractable good mod ` model for R = Z[1=`] if
` is a prime which is not regular. It seems possible to do this if ` is an
irregular prime which is well-behaved [20, p. 201] in a certain number
theoretic sense.
7.4 Question: To what extent can one modify the construction of
BGL n(Aet) to take into consideration the unstable phenomena Quillen
describes in [15, p. 591]? Namely, Quillen for various A produces classes
in H*(BGL n(A); Z=`) which vanish in H*(BGL (A); Z=`); his key as-
sumption is that Pic (A) has nontrivial elements of finite order prime
to `. In particular, if ` is regular his method produces such classes for
A = R[i] whenever R[i] is not a principal ideal domain. These classes
are not accounted for in H*(BGL n(R[i]et); Z=`).
7.5 Question: How can one incorporate transfer more systematically
into the modelling process? For example, transfer plays an important
role in known surjectivity results (see, for example, [4]). Can one incor-
porate transfer into the model itself?
7.6 Question: To what extent can the natural map
BG(A) ! BG(Aet)
of 1.4 be used to study the homology of G(A) for group schemes GR
over R other than GLn;R and Tn;R ?
References
[1] M. Artin and B. Mazur, "Etale Homotopy," Lecture Notes in Math. 100,
Springer, Berlin, 1969.
[2] A.K. Bousfield and D.M. Kan, "Homotopy Limits, Completions and Localiza-
tions," Lecture Notes in Math. 304, Springer, Berlin, 1972.
[3] K. S. Brown, "Cohomology of Groups," Graduate Texts in Math. 87, Springe*
*r,
New York, 1982.
[4] W. G. Dwyer and E. M. Friedlander, Algebraic and etale K-theory, Trans.
Amer. Math. Soc. 272 (1985), 247-280.
[5] W. G. Dwyer and E. M. Friedlander, Conjectural calculations of general l*
*inear
group homology, in "Applications of Algebraic K-theory to Algebraic Geomet*
*ry
and Number Theory," Contemporary Mathematics Volume 55, Part I, Amer.
Math. Soc., Providence, 1986, pp. 135-147.
[6] W. G. Dwyer and E. M. Friedlander, Some remarks on the K-theory of field*
*s,
in "Applications of Algebraic K-theory to Algebraic Geometry and Number
Theory," Contemporary Mathematics Volume 55, Part I, Amer. Math. Soc.,
Providence, 1986, pp. 149-157.
Topological Models 33
[7] E. M. Friedlander, "Etale Homotopy of Simplicial Schemes," Ann. of Math.
Studies 104, Princeton Univ. Press, Princeton, 1982.
[8] E. M. Friedlander, The etale homotopy type of a geometric fibration, Man*
*uscripta
Math. 10 (1973), 109-144.
[9] E. M. Friedlander, K(ss; 1)'s in characteristic p > 0, Topology 12 (1973*
*), 9-18.
[10] B. Mazur, Notes on etale cohomology of number fields, Scient. Ecole Nor*
*m.
Sup., 4e ser. 6 (1973), 521-553.
[11] J. S. Milne, "Etale Cohomology," Princeton Univ. Press, Princeton, 1980.
[12] S. A. Mitchell, The Morava K-theory of algebraic K-theory spectra, K-th*
*eory
3 (1990), 607-626.
[13] S. A. Mitchell, On the plus construction for BGL (Z[1=2]) at the prime*
* 2,
preprint (University of Washington), 1989.
[14] D. G. Quillen, On the cohomology and K-theory of the general linear gro*
*up
over finite fields, Annals of Math. 96 (1972), 552-586.
[15] D. G. Quillen, The spectrum of an equivariant cohomology ring: II, Anna*
*ls
of Math. 94 (1971), 573-602.
[16] J. P. Serre, "Cohomologie Galoisienne," Springer, New York, 1964.
[17] J. P. Serre, Structure de certains pro-p-groupes, Sem. Bourbaki no. 252
(1963).
[18] L. Smith, Homological algebra and the Eilenberg-Moore spectral sequence,
Trans. A. M. S. 129 (1967), 58-93.
[19] C. Soule, K-theorie des anneaux d'entiers de corps de nombres et cohomo*
*logie
etale, Invent. Math. 55 (1979), 251-295.
[20] L. C. Washington, "Introduction to Cyclotomic Fields," Graduate Texts in
Math. 83, Springer, New York, 1982.
University of Notre Dame, Notre Dame, Indiana 46556
Northwestern University, Evanston, Illinois 60208
Processed May 8, 1992