Accessibility and Gorenstein injective envelopes

Statement (1) follows from [adamek-rosicky, Them. 2.11(i)/Def. 2.12] and  [adamek-rosicky, Remarks 2.18(2)].

We prove (2). We first observe that if B\in\mathcal{B} is \lambda-presentable relative to \mathcal{K}, then B is also \lambda-presentable relative to the accessible category \mathcal{B}. Indeed the former means that the functor \operatorname{Hom}_{\mathcal{K}}(B, -) \colon \mathcal{K} \xrightarrow{} \textbf{Set} preserves \lambda-direct limits. Since the restriction of this functor to \mathcal{B} is exactly \operatorname{Hom}_{\mathcal{B}}(B, -) \colon \mathcal{B} \xrightarrow{} \textbf{Set}, it follows from Lemma 2.3(2) that B is \lambda-presentable relative to \mathcal{B}. So we have shown \lambda-presentablility in \mathcal{K} implies \lambda-presentability in \mathcal{B}. So by combining this with statement (1), we have that for any regular cardinal \lambda' \vartriangleright \lambda: If B\in\mathcal{B} is \lambda'-presentable relative to \mathcal{K}, then B is also \lambda'-presentable relative to \mathcal{B}. On the other hand, it follows from [adamek-rosicky, Uniformization Theorem 2.19] that there exist (arbitrarily large) regular cardinals \mu \vartriangleright \lambda such that, conversely, the inclusion functor takes \mu-presentable objects of \mathcal{B} to \mu-presentable objects of \mathcal{K}. (The proof constructs a particular regular \lambda' \vartriangleright \lambda such that this is the case for any other regular \mu \vartriangleright \lambda'.)

By Baer’s criterion, the set of all \{\,U_i/C\,\} where C ranges through all subobjects of each U_i, cogenerates the categorical injective cotorsion pair, (\mathcal{G},\mathcal{I}), in our ground category \mathcal{G}. We define a set of complexes \mathcal{S} as follows where all of the n’s range through \mathbb{Z}:

One checks that \mathcal{S}^{\perp} = \mathcal{I}_A. Indeed by [gillespie-degreewise-model-strucs, Prop. 4.4], the right Ext-orthogonal to the set

is precisely the class of all complexes of injectives. So then \mathcal{S}^{\perp} is precisely the class of all complexes of injectives X such that

for all n and all A_j\in A. But since X is a complex of injectives, one can show (see [gillespie-book, Cor. 10.22 and Exer. 10.4.1]) that for a \mathcal{G}-object M we have an isomorphism

From this it follows that \mathcal{S}^{\perp} = \mathcal{I}_A.

Now defining \mathcal{W} := {}^\perp \mathcal{I}_A, we have that (\mathcal{W}, \mathcal{I}_A) is a cotorsion pair that is cogenerated by \mathcal{S}. We note that both \mathcal{W} and \mathcal{I}_A are closed under suspensions (shifts). The isomorphism in (Proof) is a special case of the well-known isomorphism

between degreewise split extensions of complexes and the homology of the total hom complex. (For example, see again [gillespie-book, Cor. 10.22].) The characterization of \mathcal{W} stated in (3) follows from this isomorphism. What we have asserted in (2) follows from the special case isomorphism shown above in (Proof).

It follows easily from property (3) that \mathcal{W} is closed under direct summands and satisfies the 2 out of 3 property on short exact sequences. It is also clear that \mathcal{W} contains all contractible complexes. This includes the categorically injective complexes, which are precisely the contractible complexes of injectives, as well as the generating set \{D^n(U_i) \,|\, U_i\in\mathcal{U}\} for the Grothendieck category \textnormal{Ch}(\mathcal{G}). From these two facts it follows that (\mathcal{W}, \mathcal{I}_A) is a (cofibrantly generated and functorially complete) injective cotorsion pair; see [gillespie-book, Prop. 2.21(4) and Them. 9.34]. It is then automatic from [gillespie-ding-modules, Prop. 3.2] that \mathcal{W} is closed under direct limits. (This can also be found in [gillespie-book, Cor. 9.57].)

At this point we have constructed a cofibrantly generated abelian model structure

on \textnormal{Ch}(\mathcal{G}) having all of the properties listed in (1), (2), and (3). The properties listed in (4) follow automatically from general statements; see [gillespie-book, Them. 10.27 and 10.29 and Cor. 12.24].

To finish, let us make clear that \mathcal{I}_{\mathcal{B}} = \mathcal{I}_A. Because of the containments

we clearly have that for any complex X of injectives,

But it follows immediately from the properties stated in (1) and (2) that

So the three types of acyclicity coincide. In particular, \mathcal{I}_A = \mathcal{I}_{\mathcal{B}}.

Of course by taking the set A to be \widehat{A} in Theorem 3.3 we do get a model structure, \mathfrak{M}_{\widehat{A}} = (All, \mathcal{W}, \mathcal{I}_{\widehat{A}}), with all of the properties listed there. We only need to show that \mathcal{I}_{\widehat{A}} = \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, the class of all exact \mathcal{B}-acyclic complexes of injectives.

(\subseteq) Assume X \in \mathcal{I}_{\widehat{A}}. Then, by definition, X is a complex of injectives that is both A-acyclic and \mathcal{U}-acyclic. By Theorem 3.3, \mathcal{I}_{\mathcal{B}} = \mathcal{I}_A, so X is \mathcal{B}-acyclic. Being also \mathcal{U}-acyclic implies that X itself is exact, as \mathcal{U} is a set of generators.

(\supseteq) On the other hand, if X\in\textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, then certainly it is A-acyclic, as A\subseteq \mathcal{B}. By hypothesis, X is also \mathcal{U}-acyclic. This means X is in \mathcal{I}_{\widehat{A}}.
This shows \mathcal{I}_{\widehat{A}} = \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, and that is all that was needed to be shown.

Precisely, let T be a set as guaranteed by Corollary 2.6. Then as in Proposition 4.1, set \widehat{T} = T\cup\mathcal{U}, and apply the theorem.

Precisely, let T be a set as guaranteed by Corollary 2.7. Then as in Proposition 4.1, set \widehat{T} = T\cup\mathcal{U}, and apply the theorem.

If X is \mathcal{U}-acyclic then it is certainly exact, as \mathcal{U} is a set of generators. (This holds for any set of generators.) For the converse we use that our generators are of finite projective dimension. Indeed if X is an exact complex of injectives and U_i \in \mathcal{U}, then by dimension shifting we get \operatorname{Ext}^1_{\mathcal{G}}(U_i, Z_nX) \cong \operatorname{Ext}^{1+j}_{\mathcal{G}}(U_i, Z_{n+j}X) and this vanishes for large enough j. It follows that X is \mathcal{U}-acyclic.

We have M\in {}^\perp \mathcal{GI}_{\mathcal{B}}, if and only if, \operatorname{Ext}^1_{\mathcal{G}}(M, N) = 0 for all N \in \mathcal{GI}_{\mathcal{B}}, if and only if, \operatorname{Ext}^1_{\mathcal{G}}(M, Z_0X) = 0 for all X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}. But there is a well-known isomorphism

which holds whenever X is an exact chain complex. (For example, the isomorphism is proved in [gillespie-degreewise-model-strucs, Lemma 4.2].) So the previous conditions happen if and only if \operatorname{Ext}^1_{\textnormal{Ch}(\mathcal{G})}(S^0(M), X) = 0 for all X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, or, if and only if S^0(M)\in {}^\perp \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}} =\mathcal{W}. This is the case, by definition, if and only if, M\in\mathcal{W}_{\mathcal{G}}. This proves \mathcal{W}_{\mathcal{G}} = {}^\perp \mathcal{GI}_{\mathcal{B}}.

Next, note that M\in{}^\perp \mathcal{GI}_{\mathcal{B}} if and only if \operatorname{Ext}^1_{\mathcal{G}}(M, Z_nX) = 0 for all X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, and all n. Since X is an exact complex of injectives it is easy to see that this is equivalent to saying M is \mathcal{B}-Tate trivial in the sense of Definition 5.1.

For X\in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}}, we have \operatorname{Ext}^{1}_{\textnormal{Ch}(\mathcal{G})}(\Sigma^{n}(-),X) \cong H_{n-1}[\operatorname{Hom}_{\mathcal{G}}(-,X)] (as in the proof of Theorem 3.3). It follows that \mathcal{W} and hence \mathcal{W}_{\mathcal{G}} = \mathcal{W}\cap\mathcal{G} are both thick.

The equivalence of the conditions on \mathcal{U} follows immediately from Lemma 5.2. Next, assume ({}^\perp \mathcal{GI}_{\mathcal{B}}, \mathcal{GI}_{\mathcal{B}}) is a complete cotorsion pair. Being Grothendieck, there does exist some set, say \{G_i\}, of generators for \mathcal{G}. For each i, using that ({}^\perp \mathcal{GI}_{\mathcal{B}}, \mathcal{GI}_{\mathcal{B}}) has enough projectives, write an epimorphism W_i \twoheadrightarrow G_i with W_i \in {}^\perp \mathcal{GI}_{\mathcal{B}}. Then \{W_i\} must also be a generating set for \mathcal{G}. In particular, {}^\perp \mathcal{GI}_{\mathcal{B}} contains the generating set \mathcal{U}:=\{W_i\}.

The final statement was shown in Lemma 4.4.

(\implies) This is Lemma 5.4.

(\impliedby) Assume we have a set \mathcal{U} of \mathcal{B}-Tate trivial generators, or equivalently, that each X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}} is \mathcal{U}-acyclic. Then we have the exact \mathcal{B}-acyclic model structure of Corollary 4.2,

Its construction shows the cotorsion pair to be cogenerated by a set of chain complexes \mathcal{S}. (Explicitly, we have taken T to be a set as guaranteed by Corollary 2.6 and then set \widehat{T} := T\cup\mathcal{U}. Then our explicit cogenerating set goes back to the proof of Theorem 3.3, and in this case becomes

It follows from [stovicek-deconstructible, Remark 1.8 and Prop. 2.9(1)] that there exists a set \mathcal{S}' \supseteq \mathcal{S} such that \textnormal{Filt}(\mathcal{S}') = \mathcal{W}. Here, \textnormal{Filt}(\mathcal{S}') denotes the class of all transfinite extensions of objects in \mathcal{S}'. Now let \mathcal{S}'_{\mathcal{G}} := \mathcal{S}'\cap\mathcal{G}, or more formally

Then \mathcal{S}'_{\mathcal{G}} \subseteq \mathcal{W}_{\mathcal{G}}. If M\in \mathcal{W}_{\mathcal{G}}, then S^{0}(M) \in \mathcal{W}, and so the complex S^0(M) must be a transfinite extension of objects of \mathcal{S}'. However, each complex X_{\alpha} in this transfinite extension is necessarily a subobject of S^{0}(M), and as such must again be of the form S^{0}(M_{\alpha}) for some object M_{\alpha}. It follows that M is a transfinite extension of objects in \mathcal{S}'_{\mathcal{G}}, and hence \mathcal{W}_{\mathcal{G}} = \textnormal{Filt}(\mathcal{S}'_{\mathcal{G}}). So by  [gillespie-book, Cor. 9.4] (Eklof’s Lemma) we have \mathcal{W}_{\mathcal{G}} = \textnormal{Filt}(\mathcal{S}'_{\mathcal{G}}) \subseteq {}^\perp (\mathcal{S'_{\mathcal{G}}}^{\perp}). Since our generating set \mathcal{U} is contained in \mathcal{W}_{\mathcal{G}}, we have that ({}^\perp (\mathcal{S'_{\mathcal{G}}}^{\perp}), \mathcal{S'_{\mathcal{G}}}^{\perp}) is a cofibrantly generated cotorsion pair; see [gillespie-book, Def. 9.30 and Remark 9.31]. Since we already know that \mathcal{W}_{\mathcal{G}} = \textnormal{Filt}(\mathcal{S}'_{\mathcal{G}}) is closed under direct summands, we see from [gillespie-book, Them. 9.34(2)] that

and that this is a (functorially) complete cotorsion pair. It now follows from Lemmas 5.6 and 5.7 that \mathcal{W}_{\mathcal{G}}^{\perp} = \mathcal{GI}_{\mathcal{B}} and that this is an injective cotorsion pair.

Finally, we comment on the enumerated properties. First, that \mathcal{G} is well-generated follows from a general result ([gillespie-book, Cor. 12.24]) and the notion of a triangulated localization is discussed in [gillespie-book, §6.7]. Property (2) follows from Lemma 5.7. Finally, the Quillen equivalence stated in (3) will be proved in Proposition 6.2.

We have \mathcal{W}_{\mathcal{G}} = {}^\perp \mathcal{GI}_{\mathcal{B}} by Lemma 5.2. So \mathcal{GI}_{\mathcal{B}} \subseteq ({}^\perp \mathcal{GI}_{\mathcal{B}})^{\perp} = \mathcal{W}_{\mathcal{G}}^{\perp}.

Assume we have a complete cotorsion pair (\mathcal{V}, \mathcal{V}^{\perp}) where \mathcal{V} is some thick class containing \mathcal{B}. We note that the cotorsion pair is necessarily hereditary; (this follows from the completeness and the thickness of \mathcal{V}. See [gillespie-recollement, Lemma 2.3].) Since \mathcal{B} contains all of the injective objects, so does \mathcal{V}. Then by a standard argument we have that (\mathcal{V}, \mathcal{V}^{\perp}) is in fact an injective cotorsion pair. That is, \mathcal{V}\cap\mathcal{V}^{\perp} is precisely the class of injective objects; see [gillespie-recollement, Prop. 3.6(3)]. Now the argument given in [gillespie-recollement, Them. 5.2] generalizes in a straightforward way to show that each N \in \mathcal{V}^{\perp} is equal to the 0-cycle of some exact \mathcal{B}-acyclic complex of injectives. The argument goes like this: Take a standard injective coresolution of N. By the hereditary condition, each cokernel in the coresolution is again in \mathcal{V}^{\perp}. To build the resolution the other way, start by taking a special \mathcal{V}-precover of N. Then the precover must be an injective object and the kernel again is in \mathcal{V}^{\perp}. So we may iterate this process to get an injective resolution of N, again with each kernel in \mathcal{V}^{\perp}. Pasting together the coresolution and the resolution we obtain an exact complex of injectives X, with N = Z_0X, and for which \operatorname{Hom}_{\mathcal{G}}(M,X) is exact whenever M \in\mathcal{V}. We are done since \mathcal{B}\subseteq \mathcal{V}.

This is the analog of [bravo-gillespie-hovey, Lemma 5.1], but we give an alternate proof. By properties previously shown, we have: (i) any disk D^n(M) on any object M\in\mathcal{G} is in the class \mathcal{W}, and (ii) any sphere S^n(B), for any object B \in \mathcal{B}, is also in \mathcal{W}. From (i) we infer that any bounded above exact complex is trivial, as any such complex is a transfinite extension of disks. From (ii) we infer that any bounded below complex of objects from \mathcal{B} is trivial, as any such complex is a transfinite extension of spheres S^n(B_n) with each B_n\in\mathcal{B}. Now the given chain complex W has a subcomplex Y \subseteq W, where Y is the bounded below complex \cdots \xrightarrow{} W_2 \xrightarrow{} W_1 \xrightarrow{} Z_0W \xrightarrow{} 0. Then note that W/Y is (isomorphic to) the complex 0 \xrightarrow{} Z_{-1}W \xrightarrow{} W_{-1} \xrightarrow{} W_{-2} \xrightarrow{} \cdots, which is bounded above and exact and so is trivial in the exact \mathcal{B}-acyclic model structure. Therefore, since the class of trivial complexes is thick, W is trivial if and only if Y is trivial. But we have another subcomplex, S^0(Z_0W) \subseteq Y, whose quotient is a bounded below complex of objects from \mathcal{B}. Thus Y (and hence W) is trivial if and only if S^0(Z_0W) is trivial. That is, if and only if Z_0W \in \mathcal{W}_{\mathcal{G}}.

That (S^0, Z_0) is an adjoint pair is standard; see [gillespie-book, Lemma 10.1(3)]. The left adjoint S^0 is clearly exact and preserves trivial objects. It follows that S^0 preserves cofibrations and trivial cofibrations, so (S^0, Z_0) is indeed a Quillen adjunction [gillespie-book, Prop. 7.7]. To verify that this is a Quillen equivalence we will check the conditions in [gillespie-book, Them 7.17(2)]. These conditions involve the unit and counit. We note that the unit of the adjunction (S^0, Z_0) is merely the identity map \eta_M \colon M \xrightarrow{} Z_0S^0(M). The counit is the evident inclusion \epsilon_X \colon S^0(Z_0X) \xrightarrow{} X. Now for an arbitrary (cofibrant) object M\in\mathcal{G}, we find a short exact sequence

with X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}} and W\in\mathcal{W}. By applying the snake lemma (in degrees 0 and -1) we get a short exact sequence

It follows from Lemma 6.1 that Z_{0}W\in\mathcal{W}, because the W_{i} \in \mathcal{B} are injective for all i\neq 0 and H_{i}W=0 for all i\neq 1. It follows that M \rightarrowtail Z_0X is a trivial cofibration. This verifies the first condition (involving the morphism j^{\#}) in [gillespie-book, Them 7.17(2)]. We now turn to the second condition (the one involving the morphism p^{\flat}). So let X \in \textnormal{\footnotesize{ex}}{\mathcal{I}}_{\mathcal{B}} be an arbitrary fibrant object. Since Z_0X is already cofibrant, the map p^{\flat} is just the evident inclusion \epsilon_X \colon S^0(Z_0X) \xrightarrow{} X that is the counit of the adjunction. We wish to show that it too is a weak equivalence (trivial cofibration) and for this it suffices to show that its cokernel is in \mathcal{W}. But this again follows from Lemma 6.1. Indeed its cokernel is the complex

where \bar{d}_0 is the unique map satisfying d_0 = \bar{d}_0\pi. Since X is an exact complex of injectives, this complex satisfies the hypotheses of Lemma 6.1, and so it is in \mathcal{W} if and only if its 0-cycle is in \mathcal{W}_{\mathcal{G}} (trivial in the Gorenstein injective model structure on \mathcal{G}). But the 0-cycle object of the above complex is clearly \operatorname{Ker}{\bar{d}_0} = 0, which is certainly in \mathcal{W}_{\mathcal{G}}. This completes the proof that that p^{\flat} is a trivial cofibration, and that (S^0, Z_0) is a Quillen equivalence.

The final statement follows from general results; see [gillespie-book, Thems. 7.12 and 7.17].

The class \mathcal{W}_{\mathcal{G}} is closed under direct limits. So the corollary follows from a classical argument due to Enochs that can be found in [xu_flat_covers, §2.2]. Although the arguments there are given for module categories, the results remain valid for Grothendieck categories (cf. [enochs2001, §2]). In particular, note that any special \mathcal{GI}_{\mathcal{B}}-preenvelope of an object M is, in the terminology of [xu_flat_covers, Def. 2.2.1], a generator for \operatorname{Ext}(\mathcal{W}_{\mathcal{G}}, M). Then the existence of Gorenstein injective envelopes follows from [xu_flat_covers, Prop. 2.2.1/Them. 2.2.2]. Likewise, the existence of \mathcal{W}_{\mathcal{G}}-covers follows from [xu_flat_covers, Them. 2.2.12].

This follows from [gillespie-book, Prop. 3.10]. In particular, take \lambda (see [gillespie-book, Setup 3.2.2 on page 63] to be \lambda = \omega := \mathcal{W}_{\mathcal{G}} \cap \mathcal{GI}_{\mathcal{B}}, which is exactly the class of injective objects.