Skip to content
BY 4.0 license Open Access Published by De Gruyter July 16, 2020

Homological epimorphisms, homotopy epimorphisms and acyclic maps

  • Joseph Chuang and Andrey Lazarev ORCID logo EMAIL logo
From the journal Forum Mathematicum

Abstract

We show that the notions of homotopy epimorphism and homological epimorphism in the category of differential graded algebras are equivalent. As an application we obtain a characterization of acyclic maps of topological spaces in terms of induced maps of their chain algebras of based loop spaces. In the case of a universal acyclic map we obtain, for a wide class of spaces, an explicit algebraic description for these induced maps in terms of derived localization.

MSC 2010: 16E45; 55R35

1 Introduction

The notion of epimorphism exists in any category 𝒞: a morphism XY is an epimorphism if for any object Z of 𝒞 the induced map of sets Hom𝒞(Y,Z)Hom𝒞(X,Z) is injective. Assuming the existence of pushouts in 𝒞, this is equivalent to requiring that the codiagonal map Y*XYY is an isomorphism; cf. for example [19, Proposition 2.1] where this and other easy equivalent reformulations are given. Epimorphisms of sets or groups are just surjections; however already for rings the situation is more interesting, e.g., the inclusion (or any localization of rings) is an epimorphism. The tensor product BAB realizes the pushout B*AB for commutative rings. On the other hand, in the category of (not necessarily commutative) rings, BAB is not a pushout, as it is not even a ring in general. However it is still true that AB is an epimorphism if and only if the multiplication map BABB is an isomorphism [11, Proposition 4.1.1].

If 𝒞 is a category with an additional structure allowing one to do homotopy theory in it (such as the category of topological spaces or simplicial sets or, more generally, a closed model category), there is a similar notion of a homotopy epimorphism: it is a map XY such that Y*X𝕃YY is an isomorphism in the homotopy category of 𝒞 where Y*X𝕃Y stands for a homotopy pushout. It is known [22] that homotopy epimorphisms of connected topological spaces are precisely acyclic maps, i.e. maps whose homotopy fibers have zero integral homology groups in positive degrees.

We investigate the notion of homotopy epimorphism in the category of differential graded (dg) algebras, possibly noncommutative. It is known [5, Proposition 3.17] that derived localizations of dg algebras are homotopy epimorphisms. On the other hand, derived localizations are also homological epimorphisms, i.e. maps AB such that the multiplication map BA𝕃BB is a quasi-isomorphism. The property of being a homological epimorphism has many nice implications for the induced functors on derived categories and it is natural to ask (especially having in mind the corresponding non-homotopy result) whether homotopy and homological epimorphisms are the same thing. Our first main result is that this is, indeed, true.

Next, we consider topological applications. Given a connected space X we denote its based loop space by GX; then its chain complex C*(GX,𝐤) (simplicial or singular) with coefficients in a commutative ring 𝐤 is a dg 𝐤-algebra. Our second main result is a characterization of 𝐤-acyclic maps f:XY as those maps for which the map of dg algebras C*(GX,𝐤),C*(GY,𝐤) is a homological epimorphism. Equivalently, f is acyclic if an only if it induces a Verdier localization of the triangulated categories of cohomologically locally constant sheaves (also known as infinity local systems) on X and Y .

Finally, we consider, for a given connected space X, its p-plus-construction Xp+; then the canonical map XX+ is a universal 𝔽p-acyclic map out of X where 𝔽p is the field with p elements. We show that if X is such that π1(X) has a perfect subgroup of finite index, then the map C*(GX,𝔽p)C*(GXp+,𝔽p) admits a purely algebraic description as a certain Bousfield localization of the category of dg C*(GX,𝔽p)-modules. Note that H0(GX,𝔽p)𝔽p[π1(X)], the 𝔽p–group ring of π1(X). In the case when π1(X) is finite we show that the dg algebra C*(GXp+,𝔽p) is (quasi-isomorphic to) the derived localization of C*(GX,𝐤) at a certain idempotent of 𝔽p[π1(X)]. In the case when X has no higher homotopy groups, i.e. is the classifying space of a finite group, this is essentially equivalent to the main result [2] (which, however, was formulated without invoking derived localization).

1.1 Notation and conventions

We work in the category of -graded (dg) modules over a fixed commutative ground ring 𝐤; an object in this category is a pair (V,dV) where V is a graded 𝐤-module and dV is a differential on it; it will always be assumed to be of homological type (so it lowers the degree of a homogeneous element). Unmarked tensor products and homomorphisms will be understood to be taken over 𝐤; we will abbreviate “differential graded” to “dg”. The suspension of a graded vector space V is the graded vector space ΣV so that (ΣV)i=Vi+1. Quasi-isomorphisms and isomorphisms will be denoted by and , respectively.

A dg algebra is an associative monoid in the dg category of dg vector spaces with respect to the standard monoidal structure given by the tensor product. Given a map AB of dg algebras (not necessarily central) we will refer to B as a dg A-algebra. A dg vector space V is a (left) dg module over a dg algebra A if it is supplied with a dg map AVV satisfying the usual conditions of associativity and unitality; a right dg module is defined similarly.

The categories of dg algebras and dg modules over a dg algebra admit structures of closed model categories; cf. [5] for an overview. We will denote by DGA𝐤 the category of dg 𝐤-algebras, and for a dg algebra A we write D(A) for its derived category. Recall that objects of D(A) are cofibrant left dg A-modules and morphisms are chain homotopy classes of dg module maps. For a dg algebra A, a left dg A-module M and a right dg A-module N their tensor product NAM and their derived tensor product NA𝕃M are defined. The latter is defined as either NAM or NAM where M and N are cofibrant replacements of M and N, respectively. Similarly for two left (or right) dg A-modules M,N their derived module of homomorphisms RHomA(M,N) is defined as Hom(M,N) where M is a cofibrant replacement of M.

The category of reduced simplicial sets will be denoted by SSet* and the category of simplicial groups by SGp. By “a space” we mean “a simplicial set”; nevertheless all the results in the paper are of homotopy invariant nature and so they make sense and are valid for topological spaces by the well known correspondence between topological spaces and simplicial sets.

2 Derived free products of dg algebras

Let A,B and C be graded algebras with A being flat over 𝐤. In this case we can form the free product B*AC, this is again a graded algebra satisfying an appropriate universal property; if A,B and C are dg algebras, then so is B*AC. The derived version B*A𝕃C is described in [5, Section 2]. This is a homotopy pushout in the closed model category of dg algebras and can be defined concretely as B*AC where B and C are A-cofibrant replacements of A-algebras B and C, respectively.

For an A-algebra B we denote by B¯ the cokernel of the unit map AB; it is clearly a dg A-bimodule. Then we have the following important technical result.

Lemma 2.1.

Let A be a dg algebra and let B,C be dg A-algebras. Assume that the unit maps AB and AC are cofibrations of left A-modules. Then B*AC has a natural filtration by dg A-bimodules

0=F0B=F1F2

with nFn=B*AC and FnFkFn+k and such that

F2n+1/F2nBA(C¯AB¯)An,
F2n+2/F2n+1BA(C¯AB¯)AnAC¯

for n0.

Proof.

Let F2n,F2n+1B*AC be spanned by the monomials b1c1bncn and b1c1bncnbn+1 with biB and ciC, respectively. It is clear that this filtration consists of dg A-bimodules, is exhaustive and multiplicative. Since AB and AC are cofibrations of left dg A-modules, it follows that B¯ and C¯ are cofibrant as left A-modules, in particular they are A-flat, after forgetting the differential. Now the required conclusion on the associated graded quotients follows from [10, Theorem 4.6]; cf. also [11, p. 206] for a simpler argument in the case when C and B are free as left A-modules. ∎

Remark 2.2.

In fact, the conclusion of Lemma 2.1 holds under the weaker assumptions that AB and AC are injections and B¯,C¯ are flat left A-modules since this is what is required for the application of [10, Theorem 4.6]. Moreover, modifying the filtration so that its components are spanned by monomials ending with an element in B rather than beginning with one, one obtains a similar conclusion under the assumption that B¯,C¯ are flat rightA-modules.

Corollary 2.3.

Let A,B,C be as in Lemma 2.1. Then there is a quasi-isomorphism of dg algebras

B*ACB*A𝕃C.

Proof.

Let B,C be A-cofibrant replacements of B and C, respectively. Then we have a map

f:B*A𝕃C:=B*ACB*AC.

Consider the filtrations on B*AC and B*AC described in Lemma 2.1. Then the map f induces a map on the associated graded to these filtrations and, since B¯,B¯,C¯,C¯ are cofibrant as left A-modules, we conclude that the E1-terms of the corresponding spectral sequences are isomorphic, and the desired conclusion follows. ∎

Remark 2.4.

It is natural to ask whether the conclusion of Corollary 2.3 holds under weaker conditions than A-cofibrancy of B and C; cf. Remark 2.2. Assuming that the unit maps AB and AC are injections and that B¯ and C¯ are flat as left A modules (disregarding the differential) allows to identify the associated graded of the appropriate filtration of B*AC. However, in order to ensure that the iterated tensor product of B¯ and C¯ and B computes the derived tensor product, one has to assume, in addition, that B¯ and C¯ are homotopically flat as left dg A-modules. Recall that a left dg A-module M is called homotopically flat if for any right A-module N it holds that NA𝕃MNAM, e.g., any cofibrant module is homotopically flat. Thus, B*AC computes the derived free product if the unit maps AB,AC are injections and B¯,C¯ are flat left A-modules as well as homotopically flat left A-modules (such dg modules are called semi-flat, cf. for example [7] regarding this nomenclature).

Of course, “left” can be replaced with “right” in the above discussion; cf. Remark 2.2. In particular, the conclusion of Corollary 2.3 holds under the assumption that the unit maps AB and AC are cofibrations of right dg A-modules.

3 Modules of relative differentials for dg algebras

In this section we recall the construction of the modules of relative differentials for dg algebras. The treatment of [11] extends to the dg case in an obvious manner.

Definition 3.1.

For a dg A-algebra B the module of relative differentials ΩA(B) is the kernel of the multiplication map:

(3.1)ΩA(B)BABB.

Thus, ΩA(B) is a dg B-bimodule. If AB is an A-cofibrant replacement of the A-algebra B, we define the derived module of relative differentials ΩA𝕃(B):=ΩA(B). Thus, ΩA𝕃(B) is well defined as an object in the homotopy category of dg B-bimodules.

Suppose that f:AB is a cofibration of left dg A-modules. The short exact sequence (3.1) is split as in the category of left dg B-modules by a map BBAB, bb1. The cokernel of the latter map is isomorphic as a left dg B-module to BAB¯ (recall that B¯ is the cokernel of f). It follows that BAB¯ can be identified with ΩA(B) as a left dg B-module.

The formation of the module of relative differentials behaves well with respect to free products.

Lemma 3.2.

Let A,B and C be dg algebras.

  1. There is an isomorphism of dg B*AC-bimodules:

    ΩA(B*AC)((B*AC)BΩA(B)B(B*AC))((B*AC)CΩA(C)C(B*AC)).
  2. The maps ΩA(B)ΩA(B*AC) and ΩA(C)ΩA(B*AC) induced by the canonical maps BB*AC and CB*AC are the compositions

    ΩA(B)1id1(B*AC)BΩA(B)B(B*AC)ΩA(B*AC)

    and

    ΩA(C)1id1(B*AC)BΩA(C)C(B*AC)ΩA(B*AC).

Proof.

Part (i) of the statement above is [11, Theorem 5.8.8]. Unraveling the proof in op. cit. yields (ii). ∎

4 Equivalence of homotopy and homology epimorphisms

We will now introduce the notions of homotopy and homological epimorphisms for dg algebras and show that they are equivalent.

Definition 4.1.

Let f:AB be a map of dg algebras making B into a dg A-bimodule.

  1. f is said to be a homological epimorphism if the multiplication map BA𝕃BB is a quasi-isomorphism.

  2. f is said to be a homotopy epimorphism if the codiagonal map B*A𝕃BB is a quasi-isomorphism.

Remark 4.2.

Homotopy epimorphism can be defined in any closed model category. The notion of a homological epimorphism is more restrictive as it requires the structure of a closed model category (or something similar) on modules over monoids in a given closed model category. It would be interesting to investigate whether the equivalence between homological and homotopy epimorphism is a general categorical phenomenon (rather than special to dg algebras).

Homological epimorphisms of dg algebras were studied in [21]. Their exceptionally good property is that they induce smashing localization functors on the level of derived categories (indeed they are characterized by this property).

We have the following characterization of homological epimorphisms, whose proof is a straightforward consequence of definitions.

Lemma 4.3.

A map AB is a homological epimorphism if an only if ΩAL(B) is acyclic.∎

Theorem 4.4.

A dg algebra map AB is a homological epimorphism if and only if it is a homotopy epimorphism.

Proof.

Without loss of generality we can assume that AB is a cofibration of dg algebras; then we have B*ABB*A𝕃B. Suppose that AB is a homological epimorphism. It suffices to show that the map id*1:BB*AAB*AB is a quasi-isomorphism. Considering the filtration {Fn} on B*AB constructed in Lemma 2.1 and taking into account that BAB¯0 since AB is a homological epimorphism, we conclude that the associated graded quotients Fn/Fn-1 are acyclic for n>0 and so id*1:BB*AB is indeed a quasi-isomorphism.

Conversely, suppose that AB is a homotopy epimorphism; we will show that ΩA𝕃(B)ΩA(B) is acyclic. We have the following quasi-isomorphisms of B-bimodules:

ΩA(B)ΩA(B*AB)(B*ABBΩA(B)BB*AB)(B*ABBΩA(B)BB*AB)
ΩA(B)ΩA(B).

Here the second quasi-isomorphism follows from Lemma 3.2 (i) and the third follows since B*ABB and taking into account that B*AB is cofibrant both as a left and right B-module. By Lemma 3.2 (ii) the map

Bid*1B*AB

(which is a quasi-isomorphism) induces a quasi-isomorphism ΩA(B)ΩA(B)ΩA(B) that is also the inclusion into the left direct summand. This is only possible if ΩA(B)0 as required.∎

5 𝐤-acyclic maps

Recall that there is a left Quillen functor G:SSet*SGp that is part of a Quillen equivalence between the categories of reduced simplicial sets and simplicial groups; cf. [14]. We will also consider the left Quillen functor C*:SSet*DGA𝐤 associating to a simplicial group H its normalized simplicial chain complex supplied with the Pontryagin product C*(H,𝐤)C*(H,𝐤)C*(H,𝐤) induced by the simplicial group operation H×HH.

Interpreted topologically, the composite functor C*G:SSet*DGA𝐤 associates to a reduced simplicial set X a dg algebra that is quasi-isomorphic to the chain algebra on the loop space of |X|, the geometric realization of X. An obvious modification allows one to consider it as a functor from the homotopy category of connected (not necessarily reduced) spaces.

Definition 5.1.

A map between connected spaces is 𝐤-acyclic if its homotopy fiber F is 𝐤-acyclic, that is Hn(F,𝐤)=0, n>0.

Remark 5.2.

It is customary to call -acyclic maps simply acyclic.

Lemma 5.3.

Let f:XY be a map between connected spaces. Then f is k-acyclic and induces an isomorphism π1(X)π1(Y) if and only if the induced map of dg algebras C*(GX,k)C*(GY,k) is a quasi-isomorphism.

Proof.

Suppose first that X and Y are both simply-connected. Let f be 𝐤-acyclic. It follows by considering the Serre spectral sequence of f that the dg coalgebras C*(X,𝐤) and C*(Y,𝐤) are quasi-isomorphic. Then the spectral sequences associated with cobar-constructions of the chain coalgebras C*(X,𝐤) and C*(Y,𝐤) converge strongly to H*(GX,𝐤) and H*(GY,𝐤), respectively (this is where simple connectivity is needed) and it follows that the dg algebras C(GX,𝐤) and C*(GY,𝐤) are indeed quasi-isomorphic.

Conversely, if the map of dg algebras C*(GX,𝐤)C*(GY,𝐤) is a quasi-isomorphism, then the bar-construction spectral sequence implies that f:XY induces an isomorphism on 𝐤-homology and thus, again by simple connectivity and the Serre spectral sequence of f, it is a 𝐤-acyclic map. The desired statement is therefore proved in the simply-connected case.

If X and Y are not simply-connected, denote by X¯ and Y¯ their universal covers and note that the condition that f:XY induces an isomorphism on fundamental groups implies that there is a homotopy pullback diagram of spaces:

(5.1)

where the map f¯ is induced by f. Then the homotopy fiber of f¯ is 𝐤-acyclic and since X¯,Y¯ are simply-connected, we obtain by the argument above that the dg algebras C*(GX¯,𝐤) and C*(GY¯,𝐤) are quasi-isomorphic. But clearly there is a homotopy fiber sequence of simplicial groups

(5.2)GX¯GXGBπ1(X)π1(X),

and similarly

(5.3)GY¯GXGBπ1(Y)π1(Y),

where Bπ1(X),Bπ1(Y) are classifying spaces of π1(X) and π1(Y), respectively. It follows that the maps of dg algebras C*(GX¯,𝐤)C*(GX,𝐤) and C*(GY¯,𝐤)C*(GY,𝐤) induce homology isomorphisms in positive degrees, and therefore the map C*(GX,𝐤)C*(GY,𝐤) also has this property. Finally,

H0(GX,𝐤)𝐤[π1(X)]𝐤[π1(Y)]H0(GY,𝐤)

is also an isomorphism and so the dg algebras C*(GX,𝐤) and C*(GY,𝐤) are indeed quasi-isomorphic as claimed.

Conversely, suppose that the induced map C*(GX,𝐤)C*(GY,𝐤) is a quasi-isomorphism. In particular, it gives an isomorphism H0(GX,𝐤)𝐤[π1(X)]𝐤[π1(Y)]H0(GY,𝐤) which implies that f induces an isomorphism π1(X)π1(Y). Again, we conclude that diagram (5.1) is a homotopy pullback. Similarly, we conclude from (5.2) and (5.3) that the dg algebras C*(GX¯,𝐤) and C*(GY¯,𝐤) are quasi-isomorphic, and therefore (because of simple-connectivity of X¯ and Y¯) the map f¯ is a 𝐤-acyclic map. Therefore, by 5.1f is also a 𝐤-acyclic map. ∎

Remark 5.4.

If 𝐤=, then Lemma 5.3 implies that a map XY between connected spaces is a weak equivalence if and only if the induced map C*(GX,)C*(GY,) is a quasi-isomorphism. Surprisingly, this simple and fundamental fact appears to have been noticed only recently; cf. [8, 23].

Theorem 5.5.

Let f:XY be a map between two connected spaces. Then the following are equivalent:

  1. f is a 𝐤-acyclic map.

  2. The induced map of dg algebras C*(GX,𝐤)C*(GY,𝐤) is a homotopy epimorphism.

  3. The induced map of dg algebras C*(GX,𝐤)C*(GY,𝐤) is a homological epimorphism.

Proof.

In light of Theorem 4.4 it suffices to prove the equivalence of (i) and (ii). Without loss of generality we assume that X and Y are reduced (as opposed to merely connected) simplicial sets. The functor XC*(GX,𝐤) is a composition of two left Quillen functors and thus is itself left Quillen, and so it preserves homotopy pushouts. It follows that there is a quasi-isomorphism of dg algebras

C*(G(Y*X𝕃Y),𝐤)C*(GY,𝐤)*C*(GX,𝐤)𝕃C*(GY,𝐤).

Let f:XY be a 𝐤-acyclic map. Then the map YY*X𝕃Y mapping Y to the first (or second) wedge component of Y*X𝕃Y is likewise 𝐤-acyclic and, moreover, by the van Kampen theorem, induces an isomorphism on the fundamental groups. By Lemma 5.3, the corresponding map

C*(GY,𝐤)C*(G(Y*X𝕃Y),𝐤)C*(GY,𝐤)*C*(GX,𝐤)𝕃C*(GY,𝐤)

is a quasi-isomorphism, so C*(GX,𝐤)C*(GY,𝐤) is a homotopy epimorphism. This chain of implications is clearly reversible, and so we obtain the desired if and only if statement. ∎

Let 𝐤 be a field. It is known [15] that for a connected simplicial set X the derived category D𝐤(|X|) of cohomologically locally constant sheaves of 𝐤-modules on |X| the topological realization of X (also known as infinity local systems on |X|) is equivalent to D(C*(GX,𝐤)), the derived category of the chain algebra on the based loop space of X. This leads to the following result.

Corollary 5.6.

A map f:XY of connected spaces is k-acyclic if an only if the inverse image functor f-1:Dk(|Y|)Dk(|X|) is fully faithful.

Proof.

By the correspondence between cohomologically locally constant sheaves and modules on the chain algebra of based loop spaces mentioned above, the functor f-1:D𝐤(|Y|)D𝐤(|X|) is fully faithful if an only if the restriction functor D(C*(GY,𝐤))D(C*(GX,𝐤)) is fully faithful. The latter is equivalent by [21, Theorem 3.9 (6)] to the map C*(GY,𝐤))C*(GX,𝐤) being a homological epimorphism, which, by Theorem 5.5, is equivalent to f:XY being a 𝐤-acyclic map. ∎

6 Plus-construction and derived localization

6.1 Recollection on plus-construction, localization and completion of spaces

A 𝐤-plus-construction (called the partial 𝐤-completion in [4]) of a connected space X is a space X𝐤+ supplied with a map XX𝐤+ that is 𝐤-acyclic and is terminal among homotopy classes of 𝐤-acyclic maps out of X. The space X𝐤+ is local with respect to 𝐤-acyclic spaces, i.e. if Y is 𝐤-acyclic, then any basepointed map YX𝐤+ is homotopic to the constant map; and the map XX𝐤+ is initial among homotopy classes of maps from X into local spaces.

A closely related notion is that of a localization with respect to the homology theory H𝐤; for a space X its H𝐤 localization is a space LH𝐤X supplied with a map XLH𝐤X inducing an isomorphism in homology with coefficients in 𝐤 and terminal among homotopy classes of such maps (note that H𝐤 localization is called semi-𝐤-completion in [4]). The space LH𝐤X is local with respect to H𝐤-homology equivalences, i.e. if f:YZ is an H𝐤-homology equivalence, then the induced map on homotopy classes [Z,LH𝐤X][Y,LH𝐤X] is an isomorphism. Furthermore, the map XLH𝐤X is initial among homotopy classes of maps from X into LH𝐤-local spaces. The space LH𝐤X is “farther away” from X than X𝐤+ in the sense that the map XLH𝐤X always factors through X𝐤+; in favorable cases X𝐤+LH𝐤X.

Finally, there is a notion of a 𝐤-completion XX𝐤; the latter map always factors through LH𝐤. Often we have LH𝐤XX𝐤; in this case X is called 𝐤-good.

From now on, we will assume that 𝐤=𝔽p, although all arguments and statements below are valid for an arbitrary field of characteristic p. We will write Xp+,Xp and Xp for X𝔽p+,LH𝔽pX and X𝔽p, respectively.

There is a class of spaces X for which Xp+ is particularly well-behaved. Recall that a group G is p-perfect if H1(G,𝔽p)=0. The maximal p-perfect subgroup of G will be denoted by P(G); it is normal in G and the quotient G/P(G) is p-hypoabelian, i.e. it does not have nontrivial p-perfect subgroups.

Definition 6.1.

A group G is p-reasonable if G contains a p-perfect subgroup of finite index (equivalently, G/P(G) is a finite p-group). A connected space X is p-reasonable if π1(X) is.

Remark 6.2.

Clearly, if G is p-reasonable, then the augmentation ideal in the group ring 𝔽p[G/P(G)] is nilpotent. Conversely, if 𝔽p[G/P(G)] has this property then G/P(G) is a finite p-group by [12], and therefore G is p-reasonable.

Proposition 6.3.

Let X be p-reasonable. Then the following assertions hold:

  1. X is p-good (so that XpXp).

  2. π1(Xp)π1(X)/P(π1(X)).

  3. There is a weak equivalence XpXp+.

Proof.

Parts (i) and (ii) of the required statement are proved [1, Part III, Proposition 1.11] and similar arguments also prove (iii). Namely, consider the canonical map f:XB[π1(X)/P(π1(X))] induced by the quotient map π1(X)π1(X)/P(π1(X)) and recall from [4, Chapter VII, Section 6.2] that Xp+ is constructed as a fibrewise completion of the map f (converted into a fibration). Denoting the homotopy fiber of f by F, we have therefore a homotopy fiber sequence of spaces

(6.1)FpXp+B[π1(X)/P(π1(X))].

Now apply the functor of p-completion to the homotopy fiber sequence FXB[π1(X)/P(π1(X))]. We obtain

(6.2)FpXpB[π1(X)/P(π1(X))]p

and this is also a homotopy fiber sequence by the p-mod Fibre Lemma of [4, Chapter II, Section 5.1] and taking into account that the p-group π1(X)/P(π1(X)) acts nilpotently on H~*(F,𝔽p). There is a map from (6.1) to (6.2) that is a weak equivalence on end terms (since B[π1(X)/P(π1(X))] is already p-complete) and it follows that it is a weak equivalence on the middle terms as required. ∎

6.2 Derived localization of dg algebras

Let A be a dg algebra and let sH0(A) be a zero-dimensional cycle in A. In this situation one can construct [5] another dg algebra LsA together with a dg algebra map f:ALs(A) such that f(s) is invertible in H0(LsA) and initial among the homotopy classes of maps out of A with this property. Moreover, the map ALsA can be interpreted as the Bousfield localization of A as a (left) dg module over itself with respect to the map rs:AA, rs(a)=as, aA. It is also the nullification of A with respect to A/s, the cofiber of rs; cf. [13, Section 4.10 and Proposition 4.11] regarding this result and terminology. There is a homotopy fiber sequence of left dg A-modules

(6.3)LsAALsA

where LsA is the s-colocalization of A (also known as A/s-cellularization and denoted by CellA/s(A) in [13]). This is a kind of a dualizing complex for left dg A modules relative to A/s and it has a nice interpretation in terms of classical homological algebra (cf. [13, Section 4]):

Ls(A)RHomA(A/s,A)REndA(A/s,A/s)𝕃A/s.

The following example taken from [13, Subsection 4.1] is instructive.

Example 6.4.

Let A=, and let s=p so A/s/p, the cyclic group of prime order p. Then Lp[1p], and from (6.3) we obtain

Lp()Σ-1([1p]/)=:Σ-1/p.

Assume that s=e is an idempotent in H0(A); then LeA is quasi-isomorphic as a left A-module to the Bousfield localization of A with respect to the localizing subcategory generated by the left dg A-module A(1-e) (since A(1-e) and A/eA(1-e)ΣA(1-e) generate the same localizing subcategory). Therefore, by [13] there is a quasi-isomorphism of left dg A-modules:

Le(A)RHomA(A(1-e),A)REndA(A(1-e),A(1-e))𝕃A(1-e)
(6.4)A(1-e)(1-e)A(1-e)𝕃(1-e)A.

The A-bimodule

I:=A(1-e)(1-e)A(1-e)𝕃(1-e)A

can be viewed as a “derived two-sided ideal” generated by the idempotent 1-e in A in the sense that the homotopy cofiber A/I (“derived quotient”) of A by I is quasi-isomorphic to the derived localization LeA. This resembles quotienting out by a two-sided ideal generated by an idempotent in a nonderived context.

6.3 Algebraic description of the loop space of a p-plus-construction

Let X be a connected space; below we will write GXp+ and GXp for G(Xp+) and G(Xp), respectively. It follows from Theorem 5.5 that the map of dg algebras C*(GX,𝔽p)C*(GXp+,𝔽p) is a homology epimorphism, and so it is natural to ask for an algebraic description of this map. We will give such a description for a p-reasonable space; the result is particularly pleasant when π1(X) is a finite group.

Set A:=C*(GX,𝔽p). There is a canonical map C*(GX,𝔽p)H0(GX,𝔽p)𝔽p[π1(X)] in the homotopy category of dg 𝔽p-algebras and, since 𝔽p[π1(X)] is augmented, so is C*(GX,𝔽p). Consider the homology functor H𝔽p:M𝔽pA𝕃M on the category of dg A-modules. The notation H𝔽p is designed to invoke an analogy with the homology functor in the stable homotopy category given by smashing with the mod-p Eilenberg–MacLane spectrum (in fact, this is not merely an analogy since the category of dg algebras is Quillen equivalent to the category of algebras over the integral Eilenberg–MacLane spectrum H, cf. [24]). We will denote by M the Bousfield localization of an A-module M with respect to this homology functor. Thus, we have a canonical map MM that is an H𝔽p-equivalence (i.e. induces a quasi-isomorphism upon applying H𝔽p) and such that M is H𝔽p-local (i.e. it does not admit homotopy nontrivial maps from H𝔽p-acyclic modules).

Theorem 6.5.

Let X be a connected space. Then the following assertions hold:

  1. If X is p-reasonable, then C*(GXp+,𝔽p) is quasi-isomorphic to C*(GX,𝔽p) and the canonical map

    C*(GX,𝔽p)C(GXp+,𝔽p)

    is the H𝔽p-localization map on the category of left dg C*(GX,𝔽p)-modules.

  2. If π1(X) is a finite group then there exists an idempotent e in 𝔽p[π1(X)] such that the dg algebra C*(GXp+,𝔽p) is quasi-isomorphic to the derived localization LeC*(GX,𝔽p) and C*(GX,𝔽p)C*(GXp+,𝔽p) is the e-localization map.

Proof.

Let X be p-reasonable and let A:=C*(GX,𝔽p) and LA:=C*(GXp+,𝔽p). To show that LAA we need to show the following:

  1. The map ALA is an H𝔽p-equivalence.

  2. LA is H𝔽p-local.

Denote by M the homotopy fiber of ALA; then MA𝕃LA0 and so

𝔽pA𝕃M𝔽pA𝕃LALA𝕃M0,

which implies (a). Next, note that any dg A-module whose A-action is through the augmentation map A𝔽p is H𝔽p-local. Indeed, if Y is such an A-module and N is H𝔽p-acyclic (i.e. 𝔽pA𝕃N0), then

RHomA(N,Y)RHom𝔽p(𝔽pA𝕃N,Y)0.

Since X is p-reasonable, H0(LA)𝔽p[π1(Xp+)] is the 𝔽p-group ring of a finite p-group and is, thus, a local 𝔽p-algebra with a finite-dimensional nilpotent maximal ideal. Any dg H0(LA)-module has a finite filtration induced by the powers of the maximal ideal in H0(LA) with associated graded quotients being 𝔽p-modules, which implies that any such module is H𝔽p-local. The dg A-module LA can be represented as a homotopy inverse limit of its Postnikov tower {LA[n]:n=0,1,} (so that LA[n] has the same homology as LA up to and including degree n and zero in higher degrees). The homotopy cofibers LA[n]/LA[n+1] are isomorphic as objects in D(A) to dg H0(LA)-modules, and thus are H𝔽p-local. This implies that LA is H𝔽p-local and part (i) is therefore proved.

For part (ii) let e be a primitive idempotent of 𝔽p[π1(X)] which acts as the identity on the trivial module of 𝔽p[π1(X)] and by 0 on other simple modules; it is unique up to conjugation by a unit of 𝔽p[π1(X)]. Then the localization map f:ALeA is easily seen to be the H𝔽p-localization, so the statement follows by part (i). Indeed, the homotopy fiber Ae of f has the property that LeAe0, but then

𝔽pA𝕃Ae𝔽pA𝕃LeAAAe𝔽pA𝕃LeAe0,

so Ae is H𝔽p-acyclic and f is an H𝔽p-local equivalence. Also H0(LeA) is a local ring with residue field 𝔽p so LeA is H𝔽p-local. ∎

The following is a straightforward consequence of Theorem 6.5 (i).

Corollary 6.6.

Let X be a p-reasonable space. The homotopy epimorphism of dg algebras

C*(GX,𝔽p)C*(GXp,𝔽p)

induces a smashing localization functor on the derived categories

D(C*(GX,𝔽p))D(C*(GXp,𝔽p))

that coincides with HFp-localization on the perfect subcategory of D(C*(GXp,Fp)).∎

Remark 6.7.

Note that the H𝔽p-localization functor on the full derived category of C*(GX,𝔽p) is not necessarily smashing, even if it is so on the perfect subcategory.

Corollary 6.8.

Let X be a connected space with π1(X) finite and denote by e an idempotent in Fp[π1(X)] acting as the identity on the trivial π1(X)-module and zero on other simple π1(X)-modules. Then there is a homotopy cofiber sequence of dg C*(GX,Fp)-modules:

C*(GX,𝔽p)(1-e)(1-e)C*(GX,𝔽p)(1-e)𝕃(1-e)C*(GX,𝔽p)C*(GX,𝔽p)C*(GXp+,𝔽p).

Proof.

This follows directly from Theorem 6.5 (ii), taking into account (6.3) and (6.4). ∎

Remark 6.9.

Let us consider the case X=BH where H is a finite group; the homotopy theory of the loop space on BHpBHp+ has been studied extensively, see the review paper [9] and the more recent [2]. From Corollary 6.8 we obtain a homotopy fiber sequence

(6.5)𝔽p[H](1-e)(1-e)𝔽p[H](1-e)𝕃(1-e)𝔽p[H]𝔽p[H]C*(G(BHp+),𝔽p).

Next, taking the homology long exact sequence of (6.5), we obtain, for n>1,

Hn(G(BHp+),𝔽p)Torn-1(1-e)𝔽p[H](1-e)(𝔽p[H](1-e),(1-e)𝔽p[H]),

while for n=1 we have an exact sequence

0H1(G(BHp+),𝔽p)𝔽p[H](1-e)(1-e)𝔽p[H](1-e)(1-e)𝔽p[H]𝔽p[H]𝔽p[H/P(H)]0,

which is the main result of [2] obtained by different methods.

Since the localization functor L:D(C*(GX,𝔽p))D(C*(GXp+,𝔽p)) is always smashing, it makes sense to ask whether it is finite, i.e. whether the localizing subcategory Ker(L)D(C*(GX,𝔽p)) is compactly generated. Theorem 6.5 (ii) tells us that it is the case when π1(X) is finite (indeed, in this case it is even a derived localization at a single element). It turns out that KerL may not be compactly generated, even when X is a classifying space of an abelian group. To see that, let us note first the following useful derived analogue of Nakayama’s lemma; here Perf(A) stands for the perfect derived category of a ring A.

Lemma 6.10.

Let A be an ordinary ring and let S=A/J be the quotient by its Jacobson radical J. Then the functor SAL-:Perf(A)Perf(S) has the zero kernel.

Proof.

This is [17, Lemma 5.3]. ∎

Example 6.11.

Let G be a locally finite p-group; then 𝔽p[G] is local by [20], and so its augmentation ideal coincides with the Jacobson radical. Assume additionally that G is divisible, e.g., G=/pn=1/pn. It follows from the divisibility of G that H1(G,𝔽p)=0, and so G is p-perfect. So, by Theorem 6.5 (i) we have the smashing localization functor L:D(𝔽p[G])D(LH𝔽p𝔽p[G]); moreover, L coincides with the H𝔽p-localization when restricted to perfect dg 𝔽p[G]-modules by Corollary 6.6. Thus, a perfect dg 𝔽p[G]-module M is in KerL if and only if

H*(𝔽p𝔽p[G]𝕃M)=0,

and by Lemma 6.10, we have M0.

Remark 6.12.

The above example is a modification of Keller’s counterexample to the Telescope Conjecture [16].

Example 6.13.

Let p,q be prime numbers such that q divides p-1. Then /q acts faithfully on /p and we can form the semidirect product /p/q, clearly it is p-perfect. Let X:=B(/p/q) An easy calculation with the Serre–Hochschild spectral sequence associated with the normal subgroup /p of /p/q shows that there is an isomorphism of graded algebras

H*(Xp+,𝔽p)H*(/p/q,𝔽p)𝔽p[x,y]

with x,y situated in cohomological degrees 2q-1 and 2q, respectively. Since Xp+ is simply-connected, this implies that its lowest homotopy group is π2q-1Xp+/p. It follows that there exists a homotopy fiber sequence

(S2q-1)p𝑗(S2q-1)pXp+

where j is the self-map of the p-completed 2q-1-sphere (S2q-1)p of degree 2q-1 (this argument is presented in [4, Chapter VII, Proposition 4.4] for p=3,q=2). Thus, GXp+ is the homotopy fiber of j; in particular, the latter is a loop space. This is a well-known fact and it is also known that this loop space structure is unique [6, Corollary 1.2].

Now let p>3 or q>2. The homology of GXp+ can be computed with the help of the cobar spectral sequence whose E2 term is

ExtH*(Xp+,𝔽p)(𝔽p,𝔽p)𝔽p[x,y]

with |x|=2q-2 and |y|=2q-1. Taking into account the Hopf algebra structure on this spectral sequence with x,y primitive, we conclude that it collapses and there is an isomorphism of graded algebras

H*(GXp+,𝔽p)𝔽p[x,y].

Note that the dg algebra C*(GXp+,𝔽p) is not formal, i.e. it is not quasi-isomorphic to its homology. Indeed, if it were, then C*(Xp+,𝔽p) would likewise be formal and quasi-isomorphic to H*(Xp+,𝔽p)𝔽p[x,y]. However, such a result is incompatible with the minimal model of C*(/p,𝔽p) computed in [18].

The case p=3,q=3 requires a separate treatment since the Hopf algebra structure on the cobar spectral sequence for GXp+ is insufficient for deducing the multiplicative structure on H*(GXp+,𝔽p). Note also that in this case XBS3, the classifying space of the symmetric group on three letters. An idempotent in 𝔽3[S3] figuring in the formulation of Theorem 6.5 can be found explicitly; some possible choices are -(12)-1,-(13)-1 and -(23)-1. Let e be any one of these. Then the dg algebra Le𝔽3[S3]C*(GX3+,𝔽3) is formal and quasi-isomorphic to the graded algebra (with zero differential) generated over 𝔽3 by two indeterminates x,y with |x|=2 and |y|=3 subject to the relation xy=yx and x3=y2 (note that this is not a graded commutative algebra, although its associated graded with respect to the powers of the maximal ideal (x,y) is isomorphic to the graded commutative polynomial algebra 𝔽3[x,y]).

To obtain this result, we will realize 𝔽3[S3] as a “squeezed resolution”, following Benson [2]. Let α:AeA(1-e) and β:A(1-e)Ae be nonzero homomorphisms; both are unique up to multiplication by a nonzero scalar. Then the dg A-module P=(Pi,di) given by

Pi={A(1-e)if i>0,Aeif i=0,0if i<0,  di={αβif i>1,βif i=1,0if i<1,

is a left 𝔽3-squeezed resolution for S3, in the sense of [2, Definition 3.2]. The chain maps z(n):PΣ-nP given by

z(n)i={Σ-nif i>0,Σ-nαif i=0,0if i<n,

for n2, together with the identity map on P, descend to a basis of

HomD(A)(P,P)HomD(A)(A,P)H(P).

Since z(m)z(n)=z(m+n) for all m,n2, as is easily verified, it follows that Le𝔽3[S3]RHomA(P,P) is quasi-isomorphic to the subalgebra generated by z(2) and z(3), which is a graded algebra with the claimed presentation.

Remark 6.14.

Some examples of computation of H*(GXp+,𝔽p) for X being a classifying space of a finite group are presented in [3, Section 13C], in particular when this finite group is /p/q. The special case p=3,q=2 was not noticed in op. cit.


Communicated by Frederick R. Cohen


Award Identifier / Grant number: EP/N015452/1

Award Identifier / Grant number: EP/N016505/1

Funding statement: This work was partially supported by EPSRC grants EP/N015452/1 and EP/N016505/1.

References

[1] M. Aschbacher, R. Kessar and B. Oliver, Fusion Systems in Algebra and Topology, London Math. Soc. Lecture Note Ser. 391, Cambridge University, Cambridge, 2011. 10.1017/CBO9781139003841Search in Google Scholar

[2] D. Benson, An algebraic model for chains on ΩBGp, Trans. Amer. Math. Soc. 361 (2009), no. 4, 2225–2242. 10.1090/S0002-9947-08-04728-4Search in Google Scholar

[3] D. J. Benson, J. P. C. Greenlees and S. Shamir, Complete intersections and mod p cochains, Algebr. Geom. Topol. 13 (2013), no. 1, 61–114. 10.2140/agt.2013.13.61Search in Google Scholar

[4] A. K. Bousfield and D. M. Kan, Homotopy Limits, Completions and Localizations, Lecture Notes in Math. 304, Springer, Berlin, 1972. 10.1007/978-3-540-38117-4Search in Google Scholar

[5] C. Braun, J. Chuang and A. Lazarev, Derived localisation of algebras and modules, Adv. Math. 328 (2018), 555–622. 10.1016/j.aim.2018.02.004Search in Google Scholar

[6] C. Broto and R. Levi, Loop structures on homotopy fibers of self maps of spheres, Amer. J. Math. 122 (2000), no. 3, 547–580. 10.1353/ajm.2000.0017Search in Google Scholar

[7] L. W. Christensen and H. Holm, The direct limit closure of perfect complexes, J. Pure Appl. Algebra 219 (2015), no. 3, 449–463. 10.1016/j.jpaa.2014.05.004Search in Google Scholar

[8] J. Chuang, J. Holstein and A. Lazarev, Homotopy theory of monoids and derived localization, preprint (2018), https://arxiv.org/abs/1810.00373. 10.1007/s40062-021-00276-6Search in Google Scholar

[9] F. R. Cohen and R. Levi, On the homotopy theory of p-completed classifying spaces, Group Representations: Cohomology, Group Actions and Topology (Seattle 1996), Proc. Sympos. Pure Math. 63, American Mathematical Society, Providence (1998), 157–182. 10.1090/pspum/063/1603147Search in Google Scholar

[10] P. M. Cohn, On the free product of associative rings, Math. Z. 71 (1959), 380–398. 10.1007/BF01181410Search in Google Scholar

[11] P. M. Cohn, Skew Fields. Theory of General Division Rings, Encyclopedia Math. Appl. 57, Cambridge University, Cambridge, 1995. 10.1017/CBO9781139087193Search in Google Scholar

[12] I. G. Connell, On the group ring, Canadian J. Math. 15 (1963), 650–685. 10.4153/CJM-1963-067-0Search in Google Scholar

[13] W. G. Dwyer and J. P. C. Greenlees, Complete modules and torsion modules, Amer. J. Math. 124 (2002), no. 1, 199–220. 10.1353/ajm.2002.0001Search in Google Scholar

[14] P. G. Goerss and J. F. Jardine, Simplicial Homotopy Theory, Progr. Math. 174, Birkhäuser, Basel, 1999. 10.1007/978-3-0348-8707-6Search in Google Scholar

[15] J. V. S. Holstein, Morita cohomology, Math. Proc. Cambridge Philos. Soc. 158 (2015), no. 1, 1–26. 10.1017/S0305004114000516Search in Google Scholar

[16] B. Keller, A remark on the generalized smashing conjecture, Manuscripta Math. 84 (1994), no. 2, 193–198. 10.1007/BF02567453Search in Google Scholar

[17] H. Krause, Cohomological quotients and smashing localizations, Amer. J. Math. 127 (2005), no. 6, 1191–1246. 10.1353/ajm.2005.0041Search in Google Scholar

[18] D. Madsen, Homological aspects in representation theory, Norwegian University, Trondheim, 2002. Search in Google Scholar

[19] F. Muro, Homotopy units in A-infinity algebras, Trans. Amer. Math. Soc. 368 (2016), no. 3, 2145–2184. 10.1090/tran/6545Search in Google Scholar

[20] W. K. Nicholson, Local group rings, Canad. Math. Bull. 15 (1972), 137–138. 10.4153/CMB-1972-025-1Search in Google Scholar

[21] D. Pauksztello, Homological epimorphisms of differential graded algebras, Comm. Algebra 37 (2009), no. 7, 2337–2350. 10.1080/00927870802623344Search in Google Scholar

[22] G. Raptis, Some characterizations of acyclic maps, J. Homotopy Relat. Struct. 14 (2019), no. 3, 773–785. 10.1007/s40062-019-00231-6Search in Google Scholar

[23] M. Rivera, F. Wierstra and M. Zeinalian, The functor of singular chains detects weak homotopy equivalences, Proc. Amer. Math. Soc. 147 (2019), no. 11, 4987–4998. 10.1090/proc/14555Search in Google Scholar

[24] B. Shipley, H-algebra spectra are differential graded algebras, Amer. J. Math. 129 (2007), no. 2, 351–379. 10.1353/ajm.2007.0014Search in Google Scholar

Received: 2019-09-06
Revised: 2020-02-02
Published Online: 2020-07-16
Published in Print: 2020-11-01

© 2021 Walter de Gruyter GmbH, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 19.4.2024 from https://www.degruyter.com/document/doi/10.1515/forum-2019-0249/html
Scroll to top button