Elsevier

Journal of Algebra

Volume 606, 15 September 2022, Pages 30-103
Journal of Algebra

Closure properties of limC

https://doi.org/10.1016/j.jalgebra.2022.04.029Get rights and content

Abstract

Let C be a class of modules and L=limC the class of all direct limits of modules from C. The class L is well understood when C consists of finitely presented modules: L then enjoys various closure properties. Our first goal here is to study the closure properties of L in the general case when CMod–R is arbitrary. Then we concentrate on two important particular cases, when C=addM and C=AddM, for an arbitrary module M.

In the first case, we prove that limaddM={NMod–R|FFS:NFSM} where S=EndM, and FS is the class of all flat right S-modules. In the second case, limAddM={FSM|FFS} where S is the endomorphism ring of M endowed with the finite topology, FS is the class of all right S-contramodules that are direct limits of direct systems of projective right S-contramodules, and FSM is the contratensor product of the right S-contramodule F with the discrete left S-module M.

For various classes of modules D, we show that if MD then limaddM=limAddM (e.g., when D consists of pure projective modules), but the equality for an arbitrary module M remains open. Finally, we deal with the question of whether limAddM=AddM˜ where AddM˜ is the class of all pure epimorphic images of direct sums of copies of a module M. We show that the answer is positive in several particular cases (e.g., when M is a tilting module over a Dedekind domain), but it is negative in general.

Introduction

Direct limits provide one of the key constructions for forming large modules from families of small ones. In the case when the small modules are taken from a class of finitely presented modules, classic theorems of Lenzing et al. make it possible to describe completely the resulting class of large modules. However, if we start with a class, C, consisting of arbitrary modules, then the structure of the class L=limC is much less clear: for example, L need not be closed under direct limits.

Our first goal here is to investigate which closure properties of the class C carry over to L. Then we will characterize the class L for two particular instances: when C is the class of all, and all finite, direct sums of copies of a single (infinitely generated) module M. The first characterization relies on the well-known equivalence between the category addM of all direct summands of finite direct sums of copies of M and (mod–S)proj, the category of all finitely generated projective right S-modules, where S is the endomorphism ring of M. The second characterization is based on a recently discovered equivalence [39] between the category AddM of all direct summands of arbitrary direct sums of copies of M and (ContraS)proj, the category of projective right contramodules over S, the endomorphism ring of M endowed with the finite topology.

We will prove that in many cases, e.g., when C consists of small or pure projective modules, particular injective or Prüfer modules, the classes limaddC and limAddC coincide. However, whether this is true in general, remains an open problem.

We will also characterize the class limAddP when P is a projective module in terms of its trace ideal. Another problem addressed here is the question of whether limAddM=AddM˜ where AddM˜ denotes the class of all pure epimorphic images of direct sums of copies of a module M. We will give a positive answer in several particular cases, e.g., when M is an (infinitely generated) tilting module over a Dedekind domain. However, we will show that the answer is negative in general, even if the class AddM is closed under direct limits: we will construct an example of a countably generated flat module M such that AddM=limAddMAddM˜.

Let us say a few more words about the applications of contramodules to the study of the class limAddM and to the limaddM versus limAddM question. The notions of a flat module and a flat contramodule [38], [37], [7] play a key role in the descriptions of the classes limaddM and limAddM, respectively. The classical Govorov–Lazard theorem [22], [27] describes the flat modules as the direct limits of projective modules, or even more precisely, as the direct limits of finitely generated free modules. The analogous assertion is not true for contramodules, generally speaking, and we present a counterexample.

Still it is not known whether every direct limit of projective contramodules is a direct limit of finitely generated projective (or finitely generated free) contramodules. When this holds for the topological endomorphism ring S of a module M, it follows that limaddM=limAddM. In particular, this observation is applicable to some Prüfer-type modules M, or more generally, to modules M whose topological endomorphism ring S admits a dense left noetherian subring S such that the induced topology on S is a left Gabriel topology with a countable base of ideals generated by central elements. It is important here that ideals generated by central elements in noetherian rings have the Artin–Rees property, which allows to prove that the underlying S-modules of flat S-contramodules are flat.

We also prove that, for any module M, both the classes limaddM and limAddM are deconstructible (i.e. every module from the respective class is filtered by modules of bounded size from the same class). The assumption that all flat S-contramodules are direct limits of projective ones, for S=EndM, allows to improve the cardinality estimate for deconstructibility of limAddM. In order to obtain the improved cardinality estimate, we study homological properties of the class of all flat contramodules and its natural subclass of so-called 1-strictly flat contramodules. Under a mild assumption (that all flat contramodules are 1-strictly flat), we show that the class of all flat S-contramodules is closed under (transfinite) extensions and kernels of epimorphisms, and that it is quasi-deconstructible modulo the class of all so-called contratensor-negligible contramodules.

Section snippets

Preliminaries

Let R be a ring and let Mod–R (mod–R) denote the class of all (all finitely presented) right R-modules. Let C be any class of modules closed under finite direct sums.

The key subject of this paper is the class L=limC of all modules MMod–R for which there exists a direct system D=(Ci,fji|ijI) in Mod–R with CiC for all iI, such that M is the direct limit of D. That is, (M,fi(iI)) is the colimit of the diagram D in Mod–R. We will use the notation of M=limCi or M=limD.

That M=limCi can

Closure under direct sums and extensions

Let C be a class of modules closed under finite direct sums. The class L=limC is well-understood in the case when C consists of finitely presented modules:

Lemma 2.1

Let R be a ring, Cmod–R, and L=limC.

  • (i)

    The class L is closed under arbitrary direct sums and direct limits, pure submodules and pure epimorphic images, and Lmod–R=addC. Moreover, L is closed under pure extensions.

  • (ii)

    Assume moreover that C is closed under direct summands, extensions, RC, and C consists of FP2-modules. Then L=(C), whence L is

Closure under direct limits and the class limaddM

For a class of modules A, we will denote by A˜ the class of all pure epimorphic images of the modules from A (cf. [18, 8.37]). This class comes up naturally in our context in the case when M is ∑-pure split, i.e., each pure embedding NM with MAddM splits. Note that each ∑-pure injective module is ∑-pure split, cf. [18, 2.32], and the converse is true e.g. when R is left hereditary and M is a tilting module by [4, 5.6].

First, we have the following observations:

Lemma 3.1

Assume that AMod–R is closed

Local splitting and a characterization of lim for classes closed under direct sums

We start by recalling the definition of a locally split monomorphism going back to Azumaya [5] (see also [50]):

Definition 4.1

A monomorphism νHomR(X,Y) is called locally split in case for each finite subset FX there exists ρFHomR(Y,X) such that ρFνF=idF. A short exact sequence 0XνYZ0 is said to be locally split provided that ν is a locally split monomorphism.

It is easy to see that each locally split monomorphism is pure. The converse fails in general:

Example 4.2

Let R be a von Neumann regular ring which is not

limaddD versus limAddD

In this section, we consider the question of when limaddD equals limAddD (or, equivalently, limsumD equals limSumD by Lemma 1.1) for a class of modules D.

The equality is trivial when D is closed under arbitrary direct sums, and easy to prove in the case when D={M} for a self-small module M: one can refine the original direct system D consisting of modules from SumM into a direct system consisting of modules from sumM making use of appropriate restrictions of the maps from D. A similar

The case of projective modules

In this section, we consider the particular case of projective modules. By Corollary 5.3, limaddM=limAddM for a projective module M. In this particular case we can describe limaddM and show that this class is closed under direct summands and direct limits.

Let IR be an ideal, F1,F2 finitely generated free right R-modules, and uHomR(F1,F2). We say that u is supported in I if u(F1)F2I. If F1=Rn,F2=Rm for some n,mN, then u is supported in I if and only if u is given by left multiplication of

The tilting case

We will now consider the particular case of (infinitely generated) tilting modules T in more detail. If T is 0-tilting, i.e., T is a projective generator, then R is isomorphic to a direct summand of Tn for some n>0, whence limaddT=limAddT=AddT˜=F0. However, the situation is much less clear already for infinitely generated 1-tilting modules. In order to cover the case of arbitrary n-tilting modules, it will be convenient to deal with a slightly more general setting.

Recall [18, 13.20] that

A covering class of modules not closed under pure quotients

If A is a class of modules closed under direct sums, but limA is not closed under direct limits, then limAA˜ (see Example 3.5, Example 3.8 above). The aim of this section is to construct a ring R and an R-module F such that the class A=AddF is closed under direct limits, i.e., AddF=limAddF, but it is still not closed under pure epimorphic images in Mod–R. Then it is clear that AddF is a precovering class closed under direct limits, hence a covering class in Mod–R.

In fact, for any given

Contramodule methods

The aim of this section is to formulate and prove Theorem 9.2, which is a version of Theorem 3.3 for limAddM. Various applications of this theorem will be presented in the subsequent sections.

The exposition in this section, as well as in Sections 10–13, is based on the theory of contramodules over topological rings. We recall the basic concepts of this theory and the main results which we need, and then proceed to the desired applications. Our main reference for contramodules is [39, Sections

Contramodules in the limaddM versus limAddM problem

In this section we deduce a corollary of Theorem 9.2 which will be used in the proofs of Proposition 5.7, Proposition 5.8 given in the next section. We also prove a couple of other (related) corollaries, one of which will be used in order to obtain a more generally formulated application to the limaddM versus limAddM problem in Section 12.

We start with a discussion of finitely presented contramodules. Let S be a complete, separated, left linear topological ring. If X is a finite set, then we

Generalized Prüfer modules

In this section we prove Proposition 5.7, Proposition 5.8. The arguments are based on the theory of contramodules over topological rings, and more specifically on Corollary 10.7.

We start with formulating the Artin–Rees lemma for centrally generated ideals in left noetherian rings in the form suitable for our purposes.

Lemma 11.1

Let R be a left noetherian ring and IR be an ideal generated by central elements. Let M be a finitely generated left R-module with a submodule NM. Then there exists an integer m0

Gabriel topologies

The aim of this section is to formulate and prove a generalization of Proposition 5.7, Proposition 5.8 to modules MR whose endomorphism ring S is left noetherian and the finite topology on S satisfies a certain list of conditions. In fact, we will consider a more general setting in which the ring S itself is not necessarily noetherian, but it has a dense noetherian subring on which the additional conditions are imposed. In addition to the application to the limaddM versus limAddM problem, we

Quasi-deconstructibility of flat contramodules

The aim of this section is to improve the cardinality estimate for deconstructibility of the class limAddM in Corollary 5.5 under an additional assumption of condition (C-GL1) from Remark 11. In fact, the cardinality estimate for deconstructibility of limAddM which we obtain under the assumption of (C-GL1) is even better that the one for the class limaddM in Corollary 3.4. However, there is a caveat that the estimates in this section are only for the cardinalities of the sets of generators

Open problems

Problem 1: Does the equality limaddD=limAddD hold for any class of modules D? In particular, does limaddM=limAddM for any module M?

See Corollary 5.3, Lemma 5.6, Proposition 5.4, Proposition 5.7, Proposition 5.8, and Corollary 6.5, Corollary 12.4 for some positive answers.

Does limaddM=limAddM hold for any tilting module M? See Corollary 7.3 and Theorem 7.9 for partial positive answers.

Problem 2: Assume that C is a deconstructible class of modules. Is L=limC also deconstructible?

Lemma 2.5

Acknowledgement

We are grateful to Michal Hrbek and Jan Šťovíček for very helpful consultations, and to Gene Abrams for providing us with a copy of the paper [32]. We also thank the referee for careful reading of the text and many helpful comments.

References (50)

  • L. Angeleri Hügel et al.

    Direct Limits of Modules of Finite Projective Dimension, Rings, Modules, Algebras, and Abelian Groups

    (2004)
  • L. Angeleri Hügel et al.

    Approximations and Mittag-Leffler conditions - the applications

    Isr. J. Math.

    (2018)
  • G. Azumaya

    Locally split submodules and modules with perfect endomorphism rings

  • S. Bazzoni et al.

    Covers and direct limits: a contramodule-based approach

    Math. Z.

    (2021)
  • S. Bazzoni et al.

    Projective covers of flat contramodules

    Int. Math. Res. Not.

    (2021)
  • G.M. Bergman

    Every module is an inverse limit of injectives

    Proc. Am. Math. Soc.

    (2013)
  • M. Boratyński

    A change of rings theorem and the Artin-Rees property

    Proc. Am. Math. Soc.

    (1975)
  • L. Claborn

    Dedekind domains: overrings and semiprime elements

    Pac. J. Math.

    (1965)
  • G. De Marco

    Projectivity of pure ideals

    Rend. Semin. Mat. Univ. Padova

    (1983)
  • R. El Bashir

    Covers and directed colimits

    Algebr. Represent. Theory

    (2006)
  • S.E. Dickson

    A torsion theory for Abelian categories

    Trans. Am. Math. Soc.

    (1966)
  • A. Dress

    On the decomposition of modules

    Bull. Am. Math. Soc.

    (1969)
  • L. Fuchs et al.

    Modules over Non-Noetherian Domains

    (2001)
  • P. Gabriel et al.

    Calculus of Fractions and Homotopy Theory

    (1967)
  • R. Göbel et al.

    Indecomposable almost free modules – the local case

    Can. J. Math.

    (1998)
  • Research supported by GAČR 20-13778S. The first-named author's research is also supported by RVO: 67985840.

    View full text