1. Introduction
The concept of torsion pairs is a useful tool to study the representation theory of finite-dimensional algebras (or more general Artin algebras). Let A be an Artin algebra and
$\mod A$
the category of finitely generated (left) A-modules. Recall that a torsion pair
$(\mathcal {T}, \mathcal {F})$
in
$\mod A$
is a pair of full subcategories of
$\mathrm{mod}\,A$
such that
$\mathrm{Hom}_A(\mathcal {T}, \mathcal {F}) = 0$
and every
$M\in \mathrm{mod}\,A$
admits a short exact sequence
with
$L \in \mathcal {T}$
and
$N \in \mathcal {F}$
. Of particular interest are the functorially finite torsion pairs, which are those fulfilling some nice approximation conditions. They arise in the context of tilting theory [Reference Brenner and Butler9] and
$\tau $
-tilting theory [Reference Adachi, Iyama and Reiten1]. The aim of this work is threefold:
-
• We generalize the notion of torsion pairs. Instead of full subcategories of
$\mathrm{mod}\,A$
, ideals of morphisms in
$\mathrm{mod}\,A$
will be considered. This leads to the concept of ideal torsion pairs (Section 2). -
• We focus on functorially finite ideal torsion pairs, which are those fulfilling some nice approximation conditions similar to functorially finite torsion pairs. We classify them through corresponding subfunctors of the forgetful functor
$\mathrm{mod}\,A \rightarrow \mathrm{Ab}$
(Section 2) and the notion of ideals determined by objects introduced in this work (Section 3), following Auslander’s work on morphisms determined by objects [Reference Auslander3]. -
• We apply the theory developed, generalizing the notion of preprojective modules (Section 5) and establishing a connection between ideal torsion pairs and the Krull–Gabriel dimension (Section 6).
A pair
$(\mathcal {I}, \mathcal {J})$
of ideals of morphisms in
$\mathrm{mod}\,A$
is an ideal torsion pair if
$\psi \varphi = 0$
for all
$\varphi \in \mathcal {I}$
,
$\psi \in \mathcal {J}$
and every
$M\in \mathrm{mod}\,A$
admits a short exact sequence
with
$\varphi \in \mathcal {I}, \psi \in \mathcal {J}$
. Further,
$\mathcal {I}$
is called a torsion ideal and
$\mathcal {J}$
a torsion-free ideal. The notion of ideal torsion pairs was developed parallelly by the first three authors in [Reference Zhu, Fu, Herzog and Schlegel29]. To see that ideal torsion pairs generalize torsion pairs, let
$(\mathcal {T}, \mathcal {F})$
be a torsion pair in
$\mathrm{mod}\,A$
. Then
$(\langle \mathcal {T} \rangle , \langle \mathcal {F} \rangle )$
is an ideal torsion pair, where
$\langle \mathcal {T} \rangle $
and
$\langle \mathcal {F} \rangle $
denote the collection of all morphisms factoring through a module in
$\mathcal {T}$
and
$\mathcal {F,}$
respectively (Remark 5.1). This assignment from torsion pairs to ideal torsion pairs is injective.
The existence of the short exact sequences
$(\ast )$
for ideal torsion pairs is functorial in M, which leads to a one-to-one correspondence between ideal torsion pairs
$(\mathcal {I}, \mathcal {J})$
and subfunctors t of the forgetful functor
$\mathrm{mod}\,A \rightarrow \mathrm{Ab}$
such that
$tM$
is a submodule of M for all
$M\in \mathrm{mod}\,A$
(Proposition 3.4). This correspondence is a crucial tool for the analysis of ideal torsion pairs, as we can now consider finitely presented functors to study them. The following theorem is a generalization of a result on torsion pairs [Reference Smalø28].
Theorem A (Theorem 3.8)
Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair and t the corresponding functor. Then
$\mathcal {I}$
is functorially finite if and only if t is finitely presented if and only if
$\mathcal {J}$
is functorially finite.
The definition of functorially finite ideals is given in Section 1. An ideal torsion pair is functorially finite if it satisfies the equivalent properties in Theorem A. Inspired by Auslander’s notion of morphisms determined by objects [Reference Auslander3], an ideal
$\mathcal {I}$
is right C-determined for
$C\in \mathrm{mod}\,A$
if
$\varphi f \in \mathcal {I}$
for all f starting in C (such that the composition is defined) already implies
$\varphi \in \mathcal {I}$
. Dually,
$\mathcal {I}$
is left C-determined if
$f \varphi \in \mathcal {I}$
for all f ending in C already implies
$\varphi \in \mathcal {I}$
. As it turns out, the right A-determined ideals of
$\mathrm{mod}\,A$
are precisely the torsion ideals (Proposition 4.1). One might ask, when a torsion ideal is also left C-determined for some
$C\in \mathrm{mod}\,A$
. The following result answers this question and gives a second viewpoint on functorially finite ideal torsion pairs.
Theorem B (Theorem 4.5)
-
(a) In
$\mathrm{mod}\,A,$
we have an equality
$$ \begin{align*} \left\{\begin{matrix}\text{functorially finite}\\ \text{torsion ideals}\end{matrix}\right\} = \bigcup_{C \in \mathrm{mod}\,A } \left\{\begin{matrix}\text{left } C\text{-determined} \\ \text{torsion ideals}\end{matrix}\right\}. \end{align*} $$
-
(b) For
$C\in \mathrm{mod}\,A,$
there exists a one-to-one correspondence
$$ \begin{align*} \left\{\begin{matrix}\text{left } C\text{-determined} \\ \text{torsion ideals}\end{matrix}\right\} \longleftrightarrow \left\{\begin{matrix} \text{bi-submodules of} \\ { {}_AC_{\mathrm{End}_A(C)^{\mathrm{op}}} } \end{matrix} \right\}. \end{align*} $$
The main application of ideal torsion pairs in this work is their relation with the Krull–Gabriel dimension of A (see [Reference Geigle15]), denoted by
$\mathrm{KG}(A)$
. In a sense,
$\mathrm{KG}(A)$
measures a complexity related to the representation type of
$\mathrm{mod}\,A$
: We have
$\mathrm{KG} (A) = 0$
if and only if A is of finite representation type by a classical result of Auslander [Reference Auslander2], and
$\mathrm{KG} (A) \neq 1$
by a result of Herzog [Reference Herzog17] and Krause [Reference Krause19, Corollary 11.4]. Geigle proved for hereditary A that
$\mathrm{KG}(A) = 2$
if A is tame and
$\mathrm{KG}(A) = \infty $
if A is wild [Reference Geigle15]. If A is a finite-dimensional k-algebra over a field k, then A is conjectured to be tame domestic if and only if
$\mathrm{KG}(A) < \infty $
[Reference Schröer, Krause and Ringel27, Conjecture 3]. For example, this conjecture is proven for string algebras over an algebraically closed field [Reference Laking, Prest and Puninski20, Corollary 1.2].
We introduce a new dimension, the torsion dimension of A, denoted by
$\mathrm{TD}(A)$
. It is defined to be the m-dimension (see [Reference Prest21]) of the lattice of functorially finite ideal torsion pairs
$(\mathcal {I}, \mathcal {J})$
, where
$(\mathcal {I}, \mathcal {J}) \leq (\mathcal {I}', \mathcal {J}')$
if
$\mathcal {I} \subseteq \mathcal {I}'$
. We always have
$\mathrm{TD}(A) \leq \mathrm{KG}(A)$
and equality if A is commutative (Proposition 7.1 and Remark 7.2). A central link between the torsion dimension and the Krull–Gabriel dimension of A is given by the radical ideal
$\mathrm{rad}_A$
, which is the smallest ideal containing all non-isomorphisms between indecomposable modules in
$\mathrm{mod}\,A$
, and its powers
$\mathrm{rad}_A^\alpha $
for ordinal numbers
$\alpha $
(see [Reference Prest22]). The following conjecture is an “ordinal version” of Schröer’s conjecture [Reference Schröer, Krause and Ringel27, Conjecture 5].
Conjecture C (Conjecture 7.3)
Let
$\alpha $
be a non-zero ordinal number. Then
$\mathrm{KG}(A) = \alpha +1$
if and only if
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
and
$\mathrm{rad}_A^{\omega ({\alpha +1})} = 0$
, where
$\omega $
denotes the first non-finite ordinal.
One step toward this conjecture is a result by Krause [Reference Krause18, Corollary 8.14]: If
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
, then
$\mathrm{KG}(A) \geq \alpha $
. We show a similar result for the torsion dimension: If
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
, then
$\mathrm{TD}(A) \geq \alpha $
(Proposition 7.7). Under the assumption of Conjecture C it then follows that
$\mathrm{KG}(A) = \mathrm{TD}(A)$
or
$\mathrm{KG}(A) = \mathrm{TD}(A)+1$
(Corollary 7.8). There are no known examples, where the second equality holds. Conjecture C is proven if A is a string algebra over an algebraically closed field [Reference Laking, Prest and Puninski20, Corollary 1.3]. Lastly, if there exists
$M\in \mathrm{mod}\,A$
such that the smallest torsion class containing M is not functorially finite, then
$\mathrm{TD}(A)> 1$
(Proposition 7.9). Putting everything together, we can calculate the values
$\mathrm{TD}(A)$
for all hereditary Artin algebras and verify
$\mathrm{TD}(A) = \mathrm{KG}(A)$
in this case.
Theorem D (Theorem 7.10)
Let A be a hereditary Artin algebra.
-
(a) If A is representation finite, then
$\mathrm{TD}(A) = 0$
. -
(b) If A is tame, then
$\mathrm{TD}(A) = 2$
. -
(c) If A is wild, then
$\mathrm{TD}(A) = \infty $
.
A second application of ideal torsion pairs is a generalization of preprojective modules [Reference Auslander and Smalø7]. Again, the radical ideal and its ordinal powers appear in this context. For an ordinal number
$\alpha ,$
let
$\mathrm{I}(\mathrm{rad}_A^\alpha )$
be the smallest torsion ideal containing
$\mathrm{rad}_A^\alpha $
. The projective rank of
$M\in \mathrm{mod}\,A$
is the smallest
$\alpha $
such that the identity
$1_M$
is contained in
$\mathrm{I}(\mathrm{rad}_A^\alpha )$
(or
$\infty $
if no such
$\alpha $
exists). The modules of projective rank
$0$
are precisely the projective modules and those of finite projective rank are precisely the preprojective modules (Proposition 6.5). If
$\lambda \neq 0$
is a limit ordinal (or
$\infty $
), then there exists a module of projective rank greater than or equal to
$\lambda $
if and only if
$\mathrm{rad}_A^\lambda \neq 0$
(Corollary 6.7). As it turns out, the behavior of modules of non-finite projective rank is opposite to the finite case.
Corollary E (Corollary 6.10)
Let
$\lambda $
be a non-zero limit ordinal and
. If there exists a module of projective rank between
$\lambda $
and
$\lambda +n$
, then there exist infinitely many indecomposable modules of projective rank between
$\lambda $
and
$\lambda +n$
and their length is unbounded.
2. Preliminaries
Let A be an Artin k-algebra (k is a commutative artinian ring and A is finitely generated over k),
$\mathrm{mod}\,A$
the category of finitely generated (left) A-modules, and
$\mathrm{Ab}$
the category of abelian groups. Further, let
$\mod k$
be the category of finitely generated k-modules and
$D = \mathrm{Hom}_k (-,I)$
, where I is the injective hull of
$k/ J_k$
with
$J_k$
the Jacobson radical of k. Then D induces a duality between
$\mod k$
and
$\mod k$
as well as between
$\mod A$
and
$\mod A^{\mathrm{op}}$
.
The main objects of concern in this work are the (functorially finite) ideal torsion pairs, a generalization of (functorially finite) torsion pairs. In what follows, we present the classical setup.
2.1. Torsion pairs
A pair
$(\mathcal {T}, \mathcal {F})$
of full subcategories of
$\mathrm{mod}\,A$
is a torsion pair if:
-
(i)
$\mathrm{Hom}_A(M,N) = 0$
for all
$M\in \mathcal {T}$
and
$N\in \mathcal {F}$
, and -
(ii) every
$M\in \mathrm{mod}\,A$
admits a short exact sequence with
$$ \begin{align*} 0 \longrightarrow L \longrightarrow M \longrightarrow N \longrightarrow 0 \end{align*} $$
$L\in \mathcal {T}$
and
$N\in \mathcal {F}$
.
Further,
$\mathcal {T}$
is called a torsion class and
$\mathcal {F}$
a torsion-free class. Instead of (ii), one can define torsion pairs by demanding a maximality condition on
$\mathcal {T}$
and
$\mathcal {F}$
with respect to the orthogonality property (i), that is, if
$M\notin \mathcal {T}$
, then there is
$N\in \mathcal {F}$
with
$\mathrm{Hom}_A(M,N) \neq 0$
and if
$N\notin \mathcal {F}$
, then there is
$M\in \mathcal {T}$
with
$\mathrm{Hom}_A(M,N) \neq 0$
. This turns out to be an equivalent definition of torsion pairs. The following result is a well-known characterization of torsion classes and torsion-free classes in
$\mathrm{mod}\,A$
.
Lemma 2.1. A full subcategory
$\mathcal {C}$
of
$\mathrm{mod}\,A$
is a torsion(-free) class if and only if
$\mathcal {C}$
is closed under extensions and factor modules (submodules).
2.2. Subcategories and approximations
We discuss approximations of modules with respect to full additive subcategories of
$\mathrm{mod}\,A$
as in [Reference Auslander and Smalø7]. A morphism f in
$\mathrm{mod}\,A$
is left (right) minimal if
$\alpha f = f$
(resp.
$f\alpha = f$
) implies that
$\alpha $
is an isomorphism for all
$\alpha $
such that the composition is defined. Let
$\mathcal {C}$
be a full additive subcategory of
$\mathrm{mod}\,A$
. A morphism
$\varphi \colon M \rightarrow C_M$
is a left
$\mathcal {C}$
-approximation of M if
$C_M \in \mathcal {C}$
and every morphism
$M\rightarrow C$
with
${C} \in \mathcal {C}$
factors through
$\varphi $
, that is, there exists a morphism
$C_M\rightarrow C$
such that the diagram

commutes. If
$\varphi $
is also left minimal, then
$\varphi $
is a left minimal
$\mathcal {C}$
-approximation. If every
$M\in \mathrm{mod}\,A$
admits a left
$\mathcal {C}$
-approximation, then
$\mathcal {C}$
is covariantly finite. Dually, a morphism
$\psi \colon C_N \rightarrow N$
is a right
$\mathcal {C}$
-approximation of N if
$C_N\in \mathcal {C}$
and every morphism
$C\rightarrow N$
with
${C} \in \mathcal {C}$
factors through
$\psi $
, that is, there exists a morphism
$C\rightarrow C_N$
such that the diagram

commutes. If
$\psi $
is also right minimal, then
$\psi $
is a right minimal
$\mathcal {C}$
-approximation. If every
$N\in \mathrm{mod}\,A$
admits a right
$\mathcal {C}$
-approximation, then
$\mathcal {C}$
is contravariantly finite. If
$\mathcal {C}$
is both co- and contravariantly finite, then
$\mathcal {C}$
is functorially finite. Further, it is well-known that if M admits a left (right)
$\mathcal {C}$
-approximation, then M also admits a left (right) minimal
$\mathcal {C}$
-approximation.
Let
$(\mathcal {T}, \mathcal {F})$
be a torsion pair and
a short exact sequence with
$M\in \mathrm{mod}\,A$
arbitrary,
$L \in \mathcal {T}$
and
$N \in \mathcal {F}$
. Then the monomorphism
$L \rightarrow M$
is always a right minimal
$\mathcal {T}$
-approximation and the epimorphism
$M \rightarrow N$
a left minimal
$\mathcal {F}$
-approximation. In particular,
$\mathcal {T}$
is contravariantly finite and
$\mathcal {F}$
covariantly finite. For
$M\in \mathrm{mod}\,A,$
we denote by
$\mathrm{gen}\, M$
(
$\mathrm{cogen}\, M$
) the collection of all modules that are isomorphic to a factor module (submodule) of
$M^n$
for some
. One might ask, when
$\mathcal {T}$
and
$\mathcal {F}$
are functorially finite. This is answered by the following result.
Theorem 2.2 [Reference Smalø28]
Let
$(\mathcal {T}, \mathcal {F})$
be a torsion pair in
$\mathrm{mod}\,A$
. The following are equivalent:
-
(i) The torsion class
$\mathcal {T}$
is functorially finite. -
(ii) There exists
$M \in \mathrm{mod}\,A$
such that
$\mathcal {T} = \mathrm{gen}\, M$
. -
(iii) The torsion-free class
$\mathcal {F}$
is functorially finite. -
(iv) There exists
$N \in \mathrm{mod}\,A$
such that
$\mathcal {F} = \mathrm{cogen}\, N$
.
A torsion pair fulfilling the equivalent properties in the above theorem is a functorially finite torsion pair.
2.3. Ideals and approximations
Passing from torsion pairs to ideal torsion pairs, we switch from full additive subcategories of
$\mathrm{mod}\,A$
to ideals of morphisms in
$\mathrm{mod}\,A$
. In the later context, approximations with respect to ideals of morphisms in
$\mathrm{mod}\,A$
are relevant. For this, we follow [Reference Fu, Guil Asensio, Herzog and Torrecillas12] but switch the notation. Recall that a class of morphisms
$\mathcal {I}$
in
$\mathrm{mod}\,A$
is an ideal of
$\mathrm{mod}\,A$
if:
-
(i) for all
$\varphi , \psi \in \mathcal {I,}$
we have
$\varphi + \psi \in \mathcal {I}$
(if the addition is defined), and -
(ii) for all
$\varphi \in \mathcal {I}$
and all
$f,g,$
we have
$g \varphi f \in \mathcal {I}$
(if the composition is defined).
For
$M,N\in \mathrm{mod}\,A,$
we denote by
$\mathcal {I}(M,N)$
the collection of all morphisms in
$\mathcal {I}$
starting in M and ending in N. Now
$\mathcal {I}$
induces the additive functors
$\mathcal {I}(M,-) \colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
and
$\mathcal {I}(-,N) \colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}^{\mathrm{op}}$
in the canonical way.
A morphism
$\varphi \colon M \rightarrow C_M$
is a left
$\mathcal {I}$
-approximation of M if
$\varphi \in \mathcal {I}$
and every morphism
$M\rightarrow C$
in
$\mathcal {I}$
factors through
$\varphi $
, that is, there exists a morphism
$C_M\rightarrow C$
such that the diagram

commutes. If
$\varphi $
is also left minimal, then
$\varphi $
is a left minimal
$\mathcal {I}$
-approximation. If every
$M\in \mathrm{mod}\,A$
admits a left
$\mathcal {I}$
-approximation, then
$\mathcal {I}$
is covariantly finite. Dually, a morphism
$\psi \colon C_N \rightarrow N$
is a right
$\mathcal {I}$
-approximation of N if
$\psi \in \mathcal {I}$
and every morphism
$C\rightarrow N$
in
$\mathcal {I}$
factors through
$\psi $
, that is, there exists a morphism
$C\rightarrow C_N$
such that the diagram

commutes. If
$\psi $
is also right minimal, then
$\psi $
is a right minimal
$\mathcal {I}$
-approximation. If every
$N\in \mathrm{mod}\,A$
admits a right
$\mathcal {I}$
-approximation, then
$\mathcal {I}$
is contravariantly finite. If
$\mathcal {I}$
is both co- and contravariantly finite, then
$\mathcal {I}$
is functorially finite. Further, if M admits a left (right)
$\mathcal {C}$
-approximation, then M also admits a left (right) minimal
$\mathcal {C}$
-approximation.
Given a full additive subcategory
$\mathcal {C}$
of
$\mathrm{mod}\,A$
, we can associate an ideal
$\langle \mathcal {C} \rangle $
of all morphisms factoring through modules in
$\mathcal {C}$
. In this way, the notion of approximations with respect to ideals of
$\mathrm{mod}\,A$
generalizes the notion of approximations with respect to full additive subcategories of
$\mathrm{mod}\,A$
by the following result.
Lemma 2.3. Let
$\mathcal {C}$
be a full additive subcategory of
$\mathrm{mod}\,A$
.
-
(a) Every left (right)
$\mathcal {C}$
-approximation is also a left (right)
$\langle \mathcal {C} \rangle $
-approximation. -
(b) Every left (right) minimal
$\langle \mathcal {C} \rangle $
-approximation is also a left (right) minimal
$\mathcal {C}$
-approximation.
Proof. We only show the “left case.” Let
$\varphi \colon M \rightarrow C_M$
be a left
$\mathcal {C}$
-approximation. Then
$\varphi $
factors through
$C_M \in \mathcal {C}$
, so
$\varphi \in \langle \mathcal {C} \rangle $
. Further, if
$\psi \colon M \rightarrow C$
is in
$\langle \mathcal {C} \rangle $
then it factors through a module in
$\mathcal {C}$
. Hence,
$\psi $
must factor through the left
$\mathcal {C}$
-approximation
$\varphi $
. It follows that
$\varphi $
is a left
$\langle \mathcal {C} \rangle $
-approximation.
Let
$\varphi \colon M \rightarrow C_M$
be a left minimal
$\langle \mathcal {C} \rangle $
-approximation. Given
$\psi \colon M \rightarrow C$
with
$C\in \mathcal {C}$
, clearly,
$\psi \in \langle \mathcal {C} \rangle $
so
$\psi $
factors through
$\varphi $
. It is left to show
$C_M\in \mathcal {C}$
for
$\varphi $
to be a left minimal
$\mathcal {C}$
-approximation. Since
$\varphi \in \langle \mathcal {C} \rangle $
, there exists
$f \colon M \rightarrow C$
and
$g \colon C \rightarrow C_M$
with
$C\in \mathcal {C}$
and
$\varphi = gf$
. Now f must factor through
$\varphi $
, that is, there exists
$h\colon C_M\rightarrow C$
with
$f = h \varphi $
. Thus,
$\varphi = gh\varphi $
. Because
$\varphi $
is left minimal, it follows that
$gh$
is an isomorphism. Hence,
$C_M$
is a direct summand of C and
$C_M \in \mathcal {C}$
. It follows that
$\varphi $
is a left minimal
$\mathcal {C}$
-approximation.
2.4. The radical ideal
A morphism
$\varphi \colon M \rightarrow N$
in
$\mathrm{mod}\,A$
is radical if for all indecomposable modules
$X\in \mathrm{mod}\,A$
and morphisms
$f\colon X \rightarrow M, g\colon N \rightarrow X,$
the composition
$g \varphi f$
is a non-isomorphism. The radical ideal of
$\mathrm{mod}\,A$
is the ideal consisting of all radical morphisms in
$\mathrm{mod}\,A$
, denoted by
$\mathrm{rad}_A$
. It plays an important role in the application of ideal torsion pairs to the Krull–Gabriel dimension and the generalization of preprojective modules (Sections 5 and 6).
A morphism
$\varphi \colon X \rightarrow M$
in
$\mathrm{mod}\,A$
is left almost split if
$\varphi $
is not a split monomorphism and every morphism
$X \rightarrow M'$
that is not a split monomorphism factors through
$\varphi $
. In that case, X must be indecomposable and morphisms starting in X that are not split monomorphisms are precisely the radical morphisms. It follows that left almost split morphisms are left
$\mathrm{rad}_A$
-approximations starting in indecomposable modules. Arbitrary left
$\mathrm{rad}_A$
-approximations can be constructed out of those. Hence, the existence of left almost split morphisms (see [Reference Auslander and Reiten6]) implies that
$\mathrm{rad}_A$
is covariantly finite. Dually, the existence of right almost split morphisms implies that
$\mathrm{rad}_A$
is contravariantly finite.
Proposition 2.4. The ideal
$\mathrm{rad}_A$
is functorially finite.
2.5. Morphisms determined by objects
In [3] and [4], Auslander introduced the concept of morphisms determined by objects. In Section 3, we will introduce the notion of ideals determined by objects to give a second viewpoint on functorially finite ideal torsion pairs. Further, we establish a connection with morphisms determined by objects.
A morphism
$\varphi \colon M \rightarrow N$
in
$\mathrm{mod}\,A$
is left C-determined for
$C\in \mathrm{mod}\,A$
provided the following condition is satisfied: For every morphism
$\varphi ' \colon M \rightarrow N'$
in
$\mathrm{mod}\,A$
, if
$f \varphi '$
factors through
$\varphi $
for all
$f \colon N' \rightarrow C$
, then
$\varphi '$
already factors through
$\varphi $
. As an example, if
$\varphi $
is left almost split, then
$\varphi $
is left M-determined. Dually, a morphism
$\varphi \colon M \rightarrow N$
in
$\mathrm{mod}\,A$
is right C-determined provided the following condition is satisfied: For every morphism
$\varphi ' \colon M' \rightarrow N$
in
$\mathrm{mod}\,A$
, if
$\varphi ' f$
factors through
$\varphi $
for all
$f \colon C \rightarrow M'$
, then
$\varphi '$
already factors through
$\varphi $
. The following result is due to Auslander, with a slight correction by Ringel [Reference Ringel25].
Theorem 2.5. Let
$\varphi $
be a morphism in
$\mathrm{mod}\,A$
, K its kernel and Q its cokernel.
-
(a) Then
$\varphi $
is left
$\tau Q \oplus I$
-determined, where
$\tau $
denotes the Auslander–Reiten translation and I the injective hull of the top of K. -
(b) Then
$\varphi $
is right
$\tau ^{-} K \oplus P$
-determined, where
$\tau ^{-}$
denotes the inverse of the Auslander–Reiten translation and P the projective cover of the socle of Q.
2.6. The functor category
Let
$(\mod A, \mathrm{Ab})$
be the abelian category of additive functors
$\mathrm{mod}\,A \rightarrow \mathrm{Ab}$
and
the full subcategory of finitely presented functors in
$(\mod A, \mathrm{Ab})$
. Recall that
$F\in (\mod A, \mathrm{Ab})$
is finitely presented if there exists a short exact sequence
with
$M,N \in \mathrm{mod}\,A$
. A functor
$F \in (\mod A, \mathrm{Ab})$
is finitely generated if it is a factor of
$\mathrm{Hom}_A(M,-)$
for some
$M\in \mathrm{mod}\,A$
. The functor categories are well-studied (see, e.g., [Reference Prest23]). The category
is closed under extensions, kernels, and cokernels in
$(\mod A, \mathrm{Ab})$
. As a consequence, it inherits the abelian structure of
$(\mod A, \mathrm{Ab})$
. The functor categories will help us to study (functorially finite) ideal torsion pairs (Section 2). Further, the definition of the Krull–Gabriel dimension takes place in
.
For
$F\in (\mod A, \mathrm{Ab})$
and
$M\in \mathrm{mod}\,A$
, since multiplication with an element in k is a morphism in
$\mathrm{mod}\,A$
, we can consider
$F(M)$
as a k-module. If F is finitely presented, then
$F(M) \in \mod k$
. Now the duality D induces a duality between
and
.
Lemma 2.6. [Reference Auslander5, Proposition 3.3]
There is a duality d between
and
given by
$F \mapsto dF$
with
$dF(M) = DF(DM)$
.
2.7. The Krull–Gabriel dimension
Let
$\mathcal {A}$
be an abelian category. A Serre subcategory
$\mathcal {S}$
of
$\mathcal {A}$
is a full subcategory closed under extensions, subobjects, and factor objects. The quotient category
$\mathcal {A}/\mathcal {S}$
is again an abelian category (for more details, see [Reference Gabriel14]).
Following Geigle [Reference Geigle15], the Krull–Gabriel dimension,
$\mathrm{KG}(A)$
, of A is defined as follows: Let
$\mathcal {S}_{-1} = 0$
be the trivial Serre subcategory of
. If
$\alpha $
is an ordinal of the form
$\alpha = \beta +1$
, let
$\mathcal {S}_\alpha $
be the Serre subcategory of all objects in
which become of finite length in
. If
$\lambda $
is a limit ordinal, then let
$\mathcal {S}_{\lambda } = \bigcup _{\alpha < \lambda } \mathcal {S}_{\alpha }$
. Now the Krull–Gabriel dimension of A equals the smallest ordinal
$\alpha $
with
. If no such
$\alpha $
exists, then
$\mathrm{KG}(A) = \infty $
.
In Section 6, we connect the theory of ideal torsion pairs with the Krull–Gabriel dimension. The Krull–Gabriel dimension is an important homological dimension, as its value is connected with the representation type of
$\mathrm{mod}\,A$
. We have
$\mathrm{KG} (A) = 0$
if and only if A is of finite representation type by a classical result of Auslander [Reference Auslander2]. If A is hereditary, then Geigle showed the following theorem.
Theorem 2.7. [Reference Geigle15]
Let A be a hereditary Artin algebra. If A is tame, then
$\mathrm{KG}(A) = 2$
and if A is wild, then
$\mathrm{KG}(A) = \infty $
.
2.8. The m-dimension of a modular lattice
Let
$(L, \lor , \land )$
be a lattice that is modular, which means
$a \lor (x \land b) = (a \lor x) \land b$
for all
$a,b,x \in L$
with
$a \leq b$
. For example, the collection of all subobjects of a fixed object in an abelian category is a modular lattice. For
$x,y\in L,$
we define
$x\sim y$
if the interval
$[x\land y, x\lor y]$
has finite length. Then
$\sim $
defines an equivalence relation on L such that
$L/{\sim }$
is again a modular lattice and the canonical map
$L \rightarrow L/{\sim }$
is a lattice homomorphism. Following Prest [Reference Prest21], the m-dimension,
$\dim L$
, of L is defined as follows: Let
$L_{-1} = L$
. If
$\alpha $
is an ordinal number of the form
$\alpha = \beta +1$
, let
$L_{\alpha } = L_{\beta }/{\sim }.$
If
$\lambda $
is a limit ordinal, then let
$L_{\lambda } = \varinjlim {}_{\alpha < \lambda } L_{\alpha }$
. Now the m-dimension of L equals the smallest ordinal
$\alpha $
such that
$L_\alpha $
consists of exactly one element. If no such
$\alpha $
exists, then
$\dim L = \infty $
. The following result yields a different way to compute
$\mathrm{KG}(A)$
that will be important for the connection between ideal torsion pairs and the Krull–Gabriel dimension.
Proposition 2.8. [Reference Krause18, Proposition 7.2]
Let L be the modular lattice of finitely presented subfunctors of
$\mathrm{Hom}_A(A,-)$
. Then
$\mathrm{KG}(A) = \dim L$
.
3. Ideal torsion pairs
A pair of ideals
$(\mathcal {I}, \mathcal {J})$
in
$\mathrm{mod}\,A$
is an ideal torsion pair if:
-
(i)
$\psi \varphi = 0$
for all
$\varphi \in \mathcal {I}$
and
$ \psi \in \mathcal {J}$
(if the composition is defined), and -
(ii) every
$M\in \mathrm{mod}\,A$
admits a short exact sequence with
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
.
Further,
$\mathcal {I}$
is called a torsion ideal and
$\mathcal {J}$
a torsion-free ideal. The concept of ideal torsion pairs was parallelly developed by the first three authors in [Reference Zhu, Fu, Herzog and Schlegel29] in a more general context.
Remark 3.1. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair,
$M\in \mathrm{mod}\,A$
and
a short exact sequence with
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
. Then for
$f\colon X \rightarrow M$
in
$\mathcal {I,}$
we have
$\psi f = 0$
, so f must factor through
$\varphi $
. Hence,
$\varphi $
is a right
$\mathcal {I}$
-approximation. Similarly,
$\psi $
is a left
$\mathcal {J}$
-approximation. In particular,
$\mathcal {I}$
is contravariantly finite and
$\mathcal {J}$
covariantly finite.
For an ideal
$\mathcal {I,}$
we denote by
$\mathcal {I}^\perp $
the collection of all morphisms
$\psi $
with
$\psi \varphi = 0$
for all
$\varphi \in \mathcal {I}$
and dually
${}^\perp \mathcal {I}$
. One can easily verify that
$\mathcal {I}^\perp $
and
${}^\perp \mathcal {I}$
are ideals. Similar to torsion pairs, ideal torsion pairs can also be defined by replacing (ii) with a maximality condition on
$\mathcal {I}$
and
$\mathcal {J}$
with respect to the orthogonality property (i). This is shown by the following result.
Proposition 3.2. Let
$\mathcal {I}$
and
$\mathcal {J}$
be ideals in
$\mathrm{mod}\,A$
.
-
(a) The pair
$({}^\perp (\mathcal {I}^\perp ), \mathcal {I}^\perp )$
is an ideal torsion pair. -
(b) The pair
$({}^\perp \mathcal {J}, ({}^\perp \mathcal {J})^\perp )$
is an ideal torsion pair. -
(c) The pair
$(\mathcal {I}, \mathcal {J})$
is an ideal torsion pair if and only if
$\mathcal {I}^\perp = \mathcal {J}$
and
${}^\perp \mathcal {J} = \mathcal {I}$
.
Proof. (a) Clearly,
$\psi \varphi = 0$
for all
$\varphi \in {}^\perp (\mathcal {I}^\perp )$
and
$\psi \in \mathcal {I}^\perp $
. For
$M\in \mathrm{mod}\,A, $
let
$L \subseteq M$
be the sum of all images of morphisms in
$\mathcal {I}$
ending in M. Then
$M \rightarrow M/L$
is contained in
$\mathcal {I}^\perp $
. It is left to show that the inclusion
$\varphi \colon L \rightarrow M$
is contained in
${}^\perp (\mathcal {I}^\perp )$
. Suppose there exists
$\psi \in \mathcal {I}^\perp $
with
$\psi \varphi \neq 0$
. Then by the definition of
$L,$
there exists
$\varphi '\colon X \rightarrow M$
in
$\mathcal {I}$
with
$\psi \varphi ' \neq 0$
, which is a contradiction. Thus, always
$\psi \varphi = 0$
and
$\varphi \in {}^\perp (\mathcal {I}^\perp )$
.
(b) Similar to (a).
(c) If
$\mathcal {I}^\perp = \mathcal {J}$
and
${}^\perp \mathcal {J} = \mathcal {I}$
, then
$({}^\perp (\mathcal {I}^\perp ), \mathcal {I}^\perp )$
equals
$(\mathcal {I}, \mathcal {J})$
and is an ideal torsion pair by (a). If
$(\mathcal {I}, \mathcal {J})$
is an ideal torsion pair, then
$ \psi \varphi = 0$
for all
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
. Thus,
$\mathcal {I}^\perp \supseteq \mathcal {J}$
. Now let
$f \colon M \rightarrow N$
be a morphism in
$\mathcal {I}^\perp $
and consider a short exact sequence
with
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
. Then
$f \varphi = 0$
implies that f factors through
$\psi $
. Hence,
$f \in \mathcal {J}$
. It follows that
$\mathcal {I}^\perp = \mathcal {J}$
and similarly
${}^\perp \mathcal {J} = \mathcal {I}$
.
As for torsion classes and torsion-free classes, we have an internal characterization for torsion ideals and torsion-free ideals. However, in each case, it only requires one closure property instead of two (compare Lemma 2.1). This already hints at the fact that, in general, there are much more ideal torsion pairs than torsion pairs.
Lemma 3.3.
-
(a) An ideal
$\mathcal {I}$
of
$\mathrm{mod}\,A$
is a torsion ideal if and only if
$\varphi f \in \mathcal {I}$
implies
$\varphi \in \mathcal {I}$
for all morphisms
$\varphi $
and epimorphisms f. -
(b) An ideal
$\mathcal {J}$
of
$\mathrm{mod}\,A$
is a torsion-free ideal if and only if
$g \psi \in \mathcal {J}$
implies
$\psi \in \mathcal {J}$
for all morphisms
$\psi $
and monomorphisms g.
Proof. (a) Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair and
$\varphi f \in \mathcal {I}$
for a morphism
$\varphi $
and an epimorphism f. Then
$\psi \varphi f = 0$
for all
$\psi \in \mathcal {J}$
. Because f is epic, also
$\psi \varphi = 0$
for all
$\psi \in \mathcal {J}$
. Now Proposition 3.2(c) implies
$\varphi \in \mathcal {I}$
.
For the other implication, let
$\mathcal {I}$
be an ideal such that
$\varphi f\in \mathcal {I}$
implies
$\varphi \in \mathcal {I}$
for all morphisms
$\varphi $
and epimorphisms f. We show that
$(\mathcal {I}, \mathcal {I}^\perp )$
is an ideal torsion pair. Similar to the proof of Proposition 3.2, for
$M\in \mathrm{mod}\,A,$
let
$L\subseteq M$
be the sum of all images of morphisms in
$\mathcal {I}$
ending in M. Then
$M \rightarrow M/L$
is contained in
$\mathcal {I}^\perp $
. It is left to show that the inclusion
$\varphi \colon L \rightarrow M$
is contained in
$\mathcal {I}$
. Because M is of finite length, it follows that L is a finite sum and there exist
$\varphi _i' \colon X_i \rightarrow M$
in
$\mathcal {I}$
such that the image of the induced morphism
$\varphi '\colon \bigoplus _{i=1}^n X_i \rightarrow M$
equals L. Now
$\varphi _i' \in \mathcal {I}$
implies
$\varphi ' \in \mathcal {I}$
and if
$f\colon \bigoplus _{i=1}^n X_i \rightarrow L$
denotes the projection onto the image of
$\varphi '$
, then
$\varphi f = \varphi ' \in \mathcal {I}$
. Since f is an epimorphism, the morphism
$\varphi $
is contained in
$\mathcal {I}$
.
(b) Similar to (a).
Next, we want to show that for an ideal torsion pair
$(\mathcal {I}, \mathcal {J}),$
the ideal
$\mathcal {I}$
is functorially finite if and only if so is the ideal
$\mathcal {J}$
. The same result holds for torsion pairs [Reference Smalø28], however, its proof heavily relies on tilting theory, which is not available in our context. Instead, we relate ideal torsion pairs to certain functors and apply the theory of the functor category
to deduce the desired result.
We denote by
the identity functor
$\mathrm{mod}\,A \rightarrow \mathrm{mod}\,A$
and view
as a functor in the abelian category of additive functors from
$\mathrm{mod}\,A$
to
$\mathrm{mod}\,A$
. Hence, a subfunctor t of
assigns to each module X a submodule
$tX$
of X.
Proposition 3.4. There exists a one-to-one correspondence

defined as follows:
-
(i) For
$M\in \mathrm{mod}\,A,$
consider a short exact sequence with
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
. Then
$M\mapsto \mathrm{Im}\, \varphi $
defines a subfunctor of
.
-
(ii) A subfunctor t of
defines an ideal torsion pair
$(\mathcal {I}, \mathcal {J})$
in
$\mathrm{mod}\,A$
by
$$ \begin{align*} \mathcal{I} &= \{\varphi\colon L \rightarrow M \mid \mathrm{Im}\, \varphi \subseteq tM\},\\ \mathcal{J} &= \{\psi\colon M \rightarrow N \mid tM \subseteq \ker \psi\}. \end{align*} $$
Proof. (i) Let
$f\colon M \rightarrow M'$
be an arbitrary morphism in
$\mathrm{mod}\,A$
. The morphism
$\varphi \colon L \rightarrow M$
is a monic left
$\mathcal {I}$
-approximation by Remark 3.1. Similarly, let
$\varphi ' \colon L' \rightarrow M'$
be a monic left
$\mathcal {I}$
-approximation. Then
$f \varphi \in \mathcal {I}$
must factor through
$\varphi '$
. Thus,
$f(\mathrm{Im}\, \varphi ) \subseteq \mathrm{Im}\, \varphi '$
and the described assignment defines a subfunctor of
.
(ii) First, we show that
$\mathcal {I}$
and
$\mathcal {J}$
are ideals. Let
$\varphi \colon L \rightarrow M$
and
$\varphi '\colon L' \rightarrow M$
be in
$\mathcal {I}$
. Then
$\mathrm{Im}\, \varphi , \mathrm{Im}\, \varphi ' \subseteq tM$
. Hence,
$\mathrm{Im}\, (\varphi + \varphi ') \subseteq tM$
and
$\varphi + \varphi '$
is contained in
$\mathcal {I}$
. For a morphism
$f\colon X\rightarrow L, $
clearly,
$\mathrm{Im}\, \varphi f \subseteq \mathrm{Im}\, \varphi \subseteq tM$
so
$\varphi f \in \mathcal {I}$
. For a morphism
$g\colon M \rightarrow Y$
, because t is a subfunctor of
, the inclusion
$g(tM) \subseteq tY$
holds. Thus,
$\mathrm{Im}\, g \varphi = g(\mathrm{Im}\, \varphi ) \subseteq g(tM) \subseteq tY$
. It follows that
$g \varphi \in \mathcal {I}$
and
$\mathcal {I}$
is an ideal. Similarly,
$\mathcal {J}$
is an ideal. By definition of
$\mathcal {J}$
and
$\mathcal {I}$
, we have
$\psi \varphi = 0$
for all
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {I} = 0$
. Further, every
$M\in \mathrm{mod}\,A$
admits the short exact sequence
with
$\varphi \in \mathcal {I}$
and
$\psi \in \mathcal {J}$
. It follows that
$(\mathcal {I}, \mathcal {J})$
is an ideal torsion pair. Since the short exact sequences are also given as above in (i), it follows that the assignments are mutually inverse.
Remark 3.5. We can consider the collection of ideal torsion pairs as a lattice by
$(\mathcal {I}, \mathcal {J}) \leq (\mathcal {I}', \mathcal {J}')$
if
$\mathcal {I} \subseteq \mathcal {I}'$
(or equivalently
$\mathcal {J} \supseteq \mathcal {J}'$
). The intersection of torsion(-free) ideals is again a torsion(-free) ideal since they are characterized by fulfilling a closure property (Lemma 3.3). Hence, the meet of the lattice is given by
$(\mathcal {I}\cap \mathcal {I}', (\mathcal {I}\cap \mathcal {I}')^\perp )$
and the join by
$({}^\perp (\mathcal {J} \cap \mathcal {J}'), \mathcal {J}\cap \mathcal {J}')$
for ideal torsion pairs
$(\mathcal {I}, \mathcal {J})$
and
$(\mathcal {I}', \mathcal {J}')$
. Because the assignment in Proposition 3.4 is order-preserving, it follows that the collection of ideal torsion pairs is a complete modular lattice (as so is the lattice of subfunctors of
). Note that the lattice of torsion pairs is not necessarily modular.
By the assignment in Proposition 3.4, we will analyze ideal torsion pairs through the corresponding subfunctors of
.
Lemma 3.6. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair in
$\mathrm{mod}\,A$
and t the corresponding subfunctor of
. The natural isomorphism
restricts to an isomorphism
$\mathcal {I}(A,-) \cong t$
.
Proof. The natural isomorphism is given by
$\mathrm{Hom}_A(A,M) \ni \varphi \mapsto \varphi (1)$
for
$M\in \mathrm{mod}\,A$
. Now if
$\varphi \in \mathcal {I}(A,M)$
, then
$\mathrm{Im}\, \varphi \subseteq tM$
and in particular
$\varphi (1) \in t M $
. Conversely, for
$x\in t M, $
let
$\varphi \colon A\rightarrow M$
be defined by
$\varphi (1) = x$
. Then
$\mathrm{Im}\, \varphi \subseteq t M $
and
$\varphi \in \mathcal {I}(A,M)$
. Hence, the natural isomorphism restricts to
$\mathcal {I}(A,-) \cong t$
.
Recall that D denotes the duality between
$\mathrm{mod}\,A$
and
$\mathrm{mod}\,A^{\mathrm{op}}$
. For an ideal
$\mathcal {I}$
in
$\mathrm{mod}\,A,$
we denote by
$D \mathcal {I}$
the collection of all morphisms
$\varphi $
in
$\mathrm{mod}\,A^{\mathrm{op}}$
isomorphic to
$D\psi $
for some
$\psi \in \mathcal {J}$
. Clearly,
$D\mathcal {I}$
is an ideal of
$\mathrm{mod}\,A^{\mathrm{op}}$
.
Lemma 3.7. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair in
$\mathrm{mod}\,A$
and t the corresponding subfunctor of
. Then
$(D\mathcal {J}, D\mathcal {I})$
is an ideal torsion pair and the corresponding subfunctor of
is isomorphic to
.
Proof. For
$D \varphi \in D \mathcal {I}$
and
$D \psi \in D \mathcal {J,}$
the equality
$D \varphi D \psi = D (\psi \varphi ) = 0$
holds. For
$M\in \mathrm{mod}\,A^{\mathrm{op}}$
, there exists a short exact sequence
with the first morphism contained in
$D \mathcal {J}$
and the second one contained in
$D \mathcal {I}$
. Thus,
$(D \mathcal {J}, D\mathcal {I})$
is an ideal torsion pair. Further, by the above short exact sequence, the corresponding subfunctor of
is equivalent to
.
Given an additive functor
$F \colon \mathrm{mod}\,A \rightarrow \mathrm{mod}\,A$
, we can also view F as a functor from
$\mathrm{mod}\,A$
to
$\mathrm{Ab}$
by composing it with the forgetful functor
$U \colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
. We say that F is finitely presented if
.
Theorem 3.8. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair and t the corresponding subfunctor of
. The following are equivalent:
-
(a) The module
$ A$
admits a left
$\mathcal {I}$
-approximation. -
(b) The ideal
$\mathcal {I}$
is functorially finite. -
(c) The functor t is finitely presented.
-
(a)’ The module
$D A^{\mathrm{op}}$
admits a right
$\mathcal {J}$
-approximation. -
(b)’ The ideal
$\mathcal {J}$
is functorially finite. -
(c)’ The functor
is finitely presented.
Proof. (a)
$\Rightarrow $
(b): By Remark 3.1, the ideal
$\mathcal {I}$
is always contravariantly finite. It is left to show that
$\mathcal {I}$
is covariantly finite. Let
$M\in \mathrm{mod}\,A$
and
$\pi \colon A^n \rightarrow M$
an epimorphism. Given a left
$\mathcal {I}$
-approximation
$\varphi \colon A\rightarrow C_A$
consider the following pushout diagram:

We will show that
$\psi $
is a left
$\mathcal {I}$
-approximation. By commutativity of the diagram,
$\psi \pi \in \mathcal {I}$
so
$\mathrm{Im}\, \psi = \mathrm{Im}\, \psi \pi \subseteq tP$
. Hence,
$\psi \in \mathcal {I}$
. Now let
$f\colon M \rightarrow X$
be a morphism in
$\mathcal {I}$
. Then
$f \pi \in \mathcal {I}$
must factor through
$\varphi ^n$
. By the universal property of the pushout diagram, there exists a morphism
$g \colon P \rightarrow X$
with
$g \psi = f$
. It follows that
$\psi $
is a left
$\mathcal {I}$
-approximation.
(b)
$\Rightarrow $
(c): By Lemma 3.6, we can show that
$\mathcal {I}(A,-) \subseteq \mathrm{Hom}_A(A,-)$
is finitely presented. Because
$\mathrm{Hom}_A(A,-)$
is finitely presented, it is enough to show that
$\mathcal {I}(A,-)$
is finitely generated. Let
${\varphi \colon A \rightarrow C_A}$
be a left
$\mathcal {I}$
-approximation. We show that
$\mathrm{Im}\, \mathrm{Hom}_A(\varphi ,-) = \mathcal {I}(A,-)$
. Because
$\varphi \in \mathcal {I}$
, the inclusion “
$\subseteq $
” holds. Now let
$f\colon A\rightarrow X$
be a morphism in
$\mathcal {I}$
. Then f factors through
$\varphi $
and so
$f\in \mathrm{Im}\, \mathrm{Hom}_A(\varphi , X)$
. Hence, also “
$\supseteq $
” holds.
(c)
$\Rightarrow $
(a): By Lemma 3.6, the functor
$\mathcal {I}(A,-) \subseteq \mathrm{Hom}_A(A,-)$
is finitely presented. Hence, there exists an epimorphism
$\mathrm{Hom}_A(M,-) \rightarrow \mathcal {I}(A,-)$
with
$M\in \mathrm{mod}\,A$
. Now, by the Yoneda lemma, there exists a morphism
$\varphi \colon A\rightarrow M$
with
$\mathrm{Im}\, \mathrm{Hom}_A(\varphi ,-) = \mathcal {I}(A,-)$
. We show that
$\varphi $
is a left
$\mathcal {I}$
-approximation. First,
$\varphi = \mathrm{Hom}_A(\varphi , M) (1_M)$
so
$\varphi \in \mathcal {I}$
. Now let
$f\colon A\rightarrow X$
be a morphism in
$\mathcal {I}$
. Then
$f\in \mathcal {I}(A,X) = \mathrm{Im}\, \mathrm{Hom}_A(\varphi , X)$
so f factors through
$\varphi $
. It follows that
$\varphi $
is a left
$\mathcal {I}$
-approximation.
(a)’
$\Leftrightarrow $
(b)’
$\Leftrightarrow $
(c)’: If
$(\mathcal {I}, \mathcal {J})$
is an ideal torsion pair in
$\mathrm{mod}\,A$
, then
$(D\mathcal {J}, D\mathcal {I})$
is an ideal torsion pair by Lemma 3.7 with the corresponding subfunctor of
equivalent to
. By duality,
$ DA^{\mathrm{op}}$
admits a right
$\mathcal {J}$
-approximation iff
$A^{\mathrm{op}}$
admits a left
$D\mathcal {J}$
-approximation and the ideal
$\mathcal {J}$
is functorially finite iff
$D\mathcal {J}$
is functorially finite. Hence, the equivalences follow from (a)
$\Leftrightarrow $
(b)
$\Leftrightarrow $
(c).
(c)
$\Leftrightarrow $
(c)’: Because
, the functor t is finitely presented if and only if
is finitely presented. By Lemma 2.6, the functor
is finitely presented if and only if
is finitely presented.
An ideal torsion pair
$(\mathcal {I}, \mathcal {J})$
in
$\mathrm{mod}\,A$
is called functorially finite if one of the equivalent conditions in Theorem 3.8 is fulfilled.
Corollary 3.9. The one-to-one correspondence

restricts to a one-to-one correspondence

Corollary 3.10. If
$(\mathcal {I}, \mathcal {J})$
and
$(\mathcal {I}', \mathcal {J}')$
are functorially finite ideal torsion pairs, then so are
${(\mathcal {I}\cap \mathcal {I}', (\mathcal {I}\cap \mathcal {I}')^\perp )}$
and
$({}^\perp (\mathcal {J} \cap \mathcal {J}'), \mathcal {J}\cap \mathcal {J}')$
.
Proof. Since
is an abelian subcategory of
$(\mod A, \mathrm{Ab})$
, it follows that the collection of finitely presented subfunctors of
is a sublattice of the lattice of subfunctors of
. Hence, by Corollary 3.9, the lattice of functorially finite ideal torsion pairs is a sublattice of the lattice of ideal torsion pairs. Now the claim follows by Remark 3.5.
We introduce a monoidal structure on pairs
$s \leq t$
of subfunctor of
as a method for producing new ideal torsion pairs via the correspondence in Proposition 3.4. Let
$s\leq t$
and
$s' \leq t'$
be subfunctors of
. Then the composition
$t (t'/s') $
is a subfunctor of
$t'/s'$
and thus isomorphic to
$t"/s'$
for some
$t"\leq t'$
. Similarly, the composition
$s(t'/s')$
is a subfunctor of
$t'/s'$
and thus isomorphic to
$s"/s'$
for some
$s" \leq t'$
. Hence,
In this way, composing factors
$t/s$
of subfunctors
$s\leq t$
of
yields a monoidal structure with the neutral element
. The following lemma shows that the monoidal structure restricts to finitely presented functors.
Lemma 3.11. Let
$F\colon \mathrm{mod}\,A \rightarrow \mathrm{mod}\,A$
be an additive finitely presented functor and
. Then
.
Proof. We can uniquely extend F and G to functors
$\overline {F}\colon \mathrm{Mod}\,A \rightarrow \mathrm{Mod}\,A$
and
$\overline {G}\colon \mathrm{Mod}\,A \rightarrow \mathrm{Ab}$
such that
$\overline {F}$
and
$\overline {G}$
commute with direct limits and
$\overline {F}, \overline {G}$
coincide with
$F,G$
on
$\mathrm{mod}\,A,$
respectively, as follows. For
$M\in \mathrm{Mod}\,A,$
we have
$M = \varinjlim M_i$
for
$M_i \in \mathrm{mod}\,A$
and we set
Then
$\overline {F}$
and
$\overline {G}$
have the desired properties (see, e.g., [Reference Prest23]). Because F and G are finitely presented, it follows from [Reference Krause18, Theorem 9,1] that
$\overline {F}$
and
$\overline {G}$
commute with products. Now
$\overline {G}\,\overline {F}$
is the unique extension, in the above sense, of
$GF$
as the functors coincide on
$\mathrm{mod}\,A$
and
$\overline {G}\,\overline {F}$
commutes with direct limits, since
$\overline {F}$
and
$\overline {G}$
commute with direct limits. Further,
$\overline {G}\,\overline {F}$
commutes with products, since so do
$\overline {G}$
and
$\overline {F}$
. By [Reference Krause18, Theorem 9,1], it follows that
$GF$
is finitely presented.
We continue by investigating special cases of the monoidal structure on pairs
$s\leq t$
of subfunctors of
, namely,
$s = 0$
and arbitrary t, as well as
and arbitrary s. In particular, we are interested in the corresponding ideal torsion pairs. To describe them, we introduce two operations on ideals
$\mathcal {I}$
and
$\mathcal {I}'$
. First, we denote by
$\mathcal {I}' \mathcal {I}$
the collection all morphisms
$\varphi '\varphi $
with
$\varphi \in \mathcal {I}$
and
$\varphi \in \mathcal {I}'$
(if the composition is defined). It is easy to check that
$\mathcal {I}' \mathcal {I}$
is again an ideal. For the other operation, we say that a morphism
$f \colon X \rightarrow Y$
is an extension of a morphism
$\varphi '$
by a morphism
$\varphi $
if there exists a commutative diagram of morphisms

with
$f = g h$
, where the horizontal sequence is exact. Now we denote by
$\mathcal {I} \diamond \mathcal {I}'$
the collection of all morphisms f that are an extension of a morphism
$\varphi ' \in \mathcal {I}'$
by a morphism
$\varphi \in \mathcal {I}$
.
The above definition of extensions of morphisms was introduced in [Reference Fu and Herzog13]. The following result was shown by the first three authors in [Reference Zhu, Fu, Herzog and Schlegel29].
Proposition 3.12. Let
$(\mathcal {I}, \mathcal {J})$
,
$(\mathcal {I}', \mathcal {J}')$
be ideal torsion pairs in
$\mathrm{mod}\,A$
and
$t, t'$
the corresponding subfunctors of
.
-
(a) The functor
$t t'$
corresponds to the ideal torsion pair
$(\mathcal {I}'\mathcal {I}, \mathcal {J} \diamond \mathcal {J}' )$
. -
(b) The functor
$t"$
, defined by
, corresponds to the ideal torsion pair
${(\mathcal {I} \diamond \mathcal {I}', \mathcal {J}' \mathcal {J})}$
.
Proof. (a) For
$M\in \mathrm{mod}\,A,$
the inclusion
$t'M \rightarrow M$
is in
$\mathcal {I}'$
and the inclusion
$tt'M \rightarrow t' M$
is in
$\mathcal {I}$
. Thus, the inclusion
$tt'M \rightarrow M$
is in
$\mathcal {I}' \mathcal {I}$
. Now an arbitrary morphism in
$\mathcal {I}' \mathcal {I}$
factors as
$\varphi ' \varphi $
for
$\varphi \colon L \rightarrow N$
in
$\mathcal {I}$
and
$\varphi ' \colon N \rightarrow M$
in
$\mathcal {I}'$
. Further,
$\varphi '$
factors as
$N\rightarrow t'M \rightarrow M$
and the composition
$L \rightarrow N \rightarrow t'M$
factors as
$L \rightarrow t t'M \rightarrow t'M$
. We conclude that
$\mathrm{Im}\, \varphi ' \varphi \subseteq t t'M$
and the torsion ideal corresponding to
$tt'$
must be equal to
$\mathcal {I}' \mathcal {I}$
. It is left to show that
$(\mathcal {I}' \mathcal {I})^\perp = \mathcal {J} \diamond \mathcal {J}'$
.
For
$f\colon X \rightarrow Y$
in
$\mathcal {J} \diamond \mathcal {J}'$
, there exists a commutative diagram of morphisms

with
$f = gh$
and
$\psi \in \mathcal {J}, \psi ' \in \mathcal {J}'$
, where the horizontal sequence is exact. Now for
$\varphi \in \mathcal {I},\varphi ' \in \mathcal {I}'$
, we have
$\pi h \varphi ' = \psi ' \varphi ' = 0$
. Hence,
$h \varphi '$
must factor through
$\iota $
, that is,
$h \varphi ' = \iota \alpha $
. Thus,
$f \varphi ' \varphi = gh \varphi ' \varphi = g \iota \alpha \varphi = \psi \alpha \varphi = 0$
and so
$f \in (\mathcal {I}' \mathcal {I})^\perp $
.
Let
$f\colon X \rightarrow Y$
in
$(\mathcal {I}' \mathcal {I})^\perp $
and consider the commutative diagram

Since
$t' X \rightarrow t'X/tt'X$
is contained in
$\mathcal {J}$
and
$X\rightarrow X/t'X$
in
$\mathcal {J}'$
, we conclude that
$X \rightarrow X/tt'X$
is contained in
$\mathcal {J}\diamond \mathcal {J}'$
. Because
$tt'X \rightarrow X$
is contained in
$\mathcal {I}' \mathcal {I}$
, the morphism f factors through
$X \rightarrow X/tt'X$
. Thus, also
$f\in \mathcal {J} \diamond \mathcal {J}'$
.
(b) Similar to (a).
Corollary 3.13. If
$(\mathcal {I}, \mathcal {J})$
and
$(\mathcal {I}', \mathcal {J}')$
are functorially finite ideal torsion pairs, then so are
${(\mathcal {I}'\mathcal {I}, \mathcal {J} \diamond \mathcal {J}' )}$
and
$(\mathcal {I} \diamond \mathcal {I}', \mathcal {J}' \mathcal {J})$
.
Proof. Let t and
$t'$
be the subfunctors of
corresponding to the ideal torsion pairs
$(\mathcal {I}, \mathcal {J})$
and
$(\mathcal {I}', \mathcal {J}')$
. By Corollary 3.9, the functors are finitely presented. Now
$tt'$
and
$t"$
, defined by
, correspond to
$(\mathcal {I}'\mathcal {I}, \mathcal {J} \diamond \mathcal {J}' )$
and
$(\mathcal {I} \diamond \mathcal {I}', \mathcal {J}' \mathcal {J}),$
respectively, by Proposition 3.12. By Lemma 3.11, the functors
$tt'$
and
$t"$
are finitely presented. Hence, the corresponding ideal torsion pairs are functorially finite.
For later purposes, the above result will be important for producing new functorially finite ideal torsion pairs. In particular, the involved ideals will be related to subcategories of
$\mathrm{mod}\,A$
. We check that the notion of extensions of morphisms behaves well with the notion of extensions of modules. For full additive subcategories
$\mathcal {C}, \mathcal {C}'$
of
$\mathrm{mod}\,A$
, we denote by
$\mathcal {C} \diamond \mathcal {C}'$
the collection of all modules
$M\in \mathrm{mod}\,A$
such that there exists a short exact sequence
with
$L \in \mathcal {C}$
and
$N \in \mathcal {C}'$
.
Lemma 3.14. Let
$\mathcal {C}$
and
$\mathcal {C}'$
be full additive subcategories of
$\mathrm{mod}\,A$
. Then
Proof. Let
$f \colon X \rightarrow Y$
be in
$\langle \mathcal {C} \diamond \mathcal {C}' \rangle $
. Then f factors as
$gh$
with g starting in
$M \in \mathcal {C} \diamond \mathcal {C}'$
. Hence, there exists a short exact sequence
with
$L\in \mathcal {C}$
and
$N\in \mathcal {C}'$
. We extend this to a commutative diagram

with
$\varphi \in \langle \mathcal {C} \rangle $
and
$\varphi ' \in \langle \mathcal {C}' \rangle $
. Thus,
$f \in \langle \mathcal {C} \rangle \diamond \langle \mathcal {C}' \rangle $
.
Let
$f\colon X\rightarrow Y$
be in
$\langle \mathcal {C} \rangle \diamond \langle \mathcal {C}' \rangle $
. Then there exists a commutative diagram

with exact middle row,
$f = gh$
and
$\varphi \in \langle \mathcal {C} \rangle $
,
$\varphi ' \in \langle \mathcal {C}' \rangle $
. In particular,
$\pi h = \varphi '$
factors through some
$C'\in \mathcal {C}'$
and taking a suitable pullback P, we obtain a commutative diagram

with
$h = h' \alpha $
. Now
$gh' \beta = g\iota = \varphi $
factors through some
$C\in \mathcal {C}$
. Taking a suitable pushout Q, we obtain a commutative diagram

with
$\gamma g' = gh'$
. Hence,
$f = gh = g h' \alpha = \gamma g' h' \alpha $
factors through
$Q \in \mathcal {C} \diamond \mathcal {C}'$
. We conclude that
$\langle \mathcal {C} \rangle \diamond \langle \mathcal {C}' \rangle = \langle \mathcal {C} \diamond \mathcal {C}' \rangle .$
4. Ideals determined by objects
Motivated by Auslander’s concept of morphisms determined by objects (see [Reference Auslander3], [Reference Auslander4]), we introduce the notion of ideals determined by objects. This will offer a different approach to torsion ideals and torsion-free ideals.
Let
$\mathcal {I}$
be an ideal of
$\mathrm{mod}\,A$
and
$C\in \mathrm{mod}\,A$
. Then
$\mathcal {I}$
is right C-determined if
$\varphi f \in \mathcal {I}$
for all f starting in C (such that the composition is defined) already implies
$\varphi \in \mathcal {I}$
. Dually,
$\mathcal {I}$
is left C-determined if
$f \varphi \in \mathcal {I}$
for all f ending in C already implies
$\varphi \in \mathcal {I}$
.
Proposition 4.1.
-
(a) An ideal
$\mathcal {I}$
is a torsion ideal if and only if
$\mathcal {I}$
is right A-determined. -
(b) An ideal
$\mathcal {J}$
is a torsion-free ideal if and only if
$\mathcal {J}$
is left
$D A^{\mathrm{op}}$
-determined.
Proof. (a) Let
$\mathcal {I}$
be a torsion ideal and
$\varphi $
a morphism such that
$\varphi f \in \mathcal {I}$
for all f starting in A. We can choose f as an epimorphism starting in
$A^n$
such that
$\varphi f\in \mathcal {I}$
. By Lemma 3.3, it follows that
$\varphi \in \mathcal {I}$
. Hence,
$\mathcal {I}$
is right A-determined.
Let
$\mathcal {I}$
be right A-determined and
$\varphi $
a morphism such that
$\varphi f \in \mathcal {I}$
for an epimorphism f. Then for an arbitrary morphism
$f'$
starting in A (such that
$\varphi f'$
is defined),
$f'$
factors through the epimorphism f. Thus,
$ \varphi f' \in \mathcal {I}$
and because
$\mathcal {I}$
is right A-determined, we conclude that
$\varphi \in \mathcal {I}$
. By Lemma 3.3, it follows that
$\mathcal {I}$
is a torsion ideal.
(b) Similar to (a).
Remark 4.2. For an ideal
$\mathcal {I}$
of
$\mathrm{mod}\,A,$
let
$\mathrm{I}(\mathcal {I})$
denote the smallest torsion ideal containing
$\mathcal {I}$
. In Section 2, we have indirectly seen two ways to describe
$\mathrm{I}(\mathcal {I})$
. Namely,
$\mathrm{I}(\mathcal {I}) = {}^\perp (\mathcal {I}^\perp )$
by Proposition 3.2 and
$\mathrm{I}(\mathcal {I})$
equals the collection of all morphisms
$\varphi $
such that
$\varphi f \in \mathcal {I}$
for all epimorphisms f by Lemma 3.3. Now Proposition 4.1 offers the most useful description for our purposes: The ideal
$\mathrm{I}(\mathcal {I})$
equals the collection of all morphisms
$\varphi $
such that
$\varphi f \in \mathcal {I}$
for all morphisms f starting in A. In particular,
$\mathrm{I}(\mathcal {I})$
is uniquely determined by
$\mathcal {I}(A,-)$
.
A similar description holds for the smallest torsion-free ideal containing
$\mathcal {I}$
, denoted by
$\mathrm{{J}}(\mathcal {I})$
: The ideal
$\mathrm{{J}}(\mathcal {I})$
equals the collection of all morphisms
$\psi $
such that
$f \psi \in \mathcal {I}$
for all morphisms f ending in
$DA^{\mathrm{op}}$
.
Let
$\mathcal {I}$
be an ideal. We would like to know when
$\mathrm{I}(\mathcal {I})$
or
$\mathrm{J}(\mathcal {I})$
is functorially finite depending on
$\mathcal {I}$
. This is fully answered by the following result.
Corollary 4.3. Let
$\mathcal {I}$
be an ideal of
$\mathrm{mod}\,A$
.
-
(a) The torsion ideal
$\mathrm {I}(\mathcal {I})$
is functorially finite if and only if A admits a left
$\mathcal {I}$
-approximation. -
(b) The torsion-free ideal
$\mathrm {J}(\mathcal {I})$
is functorially finite if and only if
$DA^{\mathrm{op}}$
admits a right
$\mathcal {I}$
-approximation.
Proof. (a) By Remark 4.2, we have
$\mathcal {I}(A,-) = \mathrm {I}(\mathcal {I})(A,-)$
. Thus, A admits a left
$\mathcal {I}$
-approximation if and only if A admits a left
$\mathrm {I}(\mathcal {I})$
-approximation. Now the claim follows by Theorem 3.8.
(b) Similar to (a).
Next, we investigate when a torsion ideal is also left C-determined and when a torsion-free ideal is also right C-determined.
Lemma 4.4. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair.
-
(a) The ideal
$\mathcal {I}$
is left C-determined for some
$C\in \mathrm{mod}\,A$
if and only if
$\mathcal {J}$
is functorially finite. -
(b) The ideal
$\mathcal {J}$
is right C-determined for some
$C\in \mathrm{mod}\,A$
if and only if
$\mathcal {I}$
is functorially finite.
Proof. (a) Let
$\mathcal {I}$
be left C-determined. We denote by
$\mathcal {J}_C$
the ideal consisting of all morphisms
$\psi f$
with
$\psi \in \mathcal {J}$
starting in
$C^n$
for some
$n>0$
and f arbitrary. Clearly,
$\mathcal {J}_C \subseteq \mathcal {J}$
and thus
$\mathrm {J}(\mathcal {J}_C) \subseteq \mathcal {J}$
. To deduce equality, we show
${}^\perp \mathcal {J}_C \subseteq \mathcal {I}$
. Let
$\varphi $
such that
$\psi f \varphi = 0$
for all
$\psi \in \mathcal {J}$
starting in
$C^n$
and arbitrary f. Then
$ f \varphi \in {}^\perp \mathcal {J} = \mathcal {I}$
and because
$\mathcal {I}$
is left C-determined, also
$\varphi \in \mathcal {I}$
. We conclude that
${}^\perp \mathcal {J}_C \subseteq \mathcal {I}$
. Hence,
$\mathrm {J}(\mathcal {J}_C) = \mathcal {J}$
. By Corollary 4.3, it suffices to show that
$DA^{\mathrm{op}}$
admits a right
$\mathcal {J}_C$
-approximation for
$\mathcal {J}$
to be functorially finite. Let
$\psi _i, 1\leq i \leq m$
generate
$\mathcal {J}(C, DA^{\mathrm{op}})$
as a k-module. Then all
$\psi _i$
induce a morphism
$\psi \colon C^m \rightarrow DA^{\mathrm{op}}$
. Clearly,
$\psi \in \mathcal {J}_C$
. Now let
$\alpha \colon M \rightarrow DA^{\mathrm{op}}$
be in
$\mathcal {J}_C$
, so
$\alpha = \beta f$
for some f and for some
$\beta \in \mathcal {J}$
starting in
$C^n$
with
. Then by construction of
$\psi $
, the morphism
$\beta $
factors through
$\psi $
. Hence,
$\alpha $
factors through
$\psi $
and
$\psi $
is a right
$\mathcal {J}_C$
-approximation.
Let
$\mathcal {J}$
be functorially finite and
$\psi \colon C\rightarrow DA^{\mathrm{op}}$
a right
$\mathcal {J}$
-approximation. Further, let
$\varphi \colon M \rightarrow N$
be a morphism such that
$f \varphi \in \mathcal {I}$
for all morphisms f ending in C. We show
$\varphi \in {}^\perp \mathcal {J} = \mathcal {I}$
. Let
$\alpha \colon N \rightarrow L $
in
$\mathcal {J}$
and
$\iota \colon L \rightarrow (DA^{\mathrm{op}})^n$
a monomorphism. Then
$\iota \alpha $
factors through
$\psi ^n$
, that is,
$\iota \alpha = \psi ^k f$
. Now
$f \varphi \in \mathcal {I}$
implies
$\iota \alpha \varphi = \psi ^k f \varphi = 0$
. Thus,
$\alpha \varphi = 0$
and we conclude that
$\varphi \in {}^\perp \mathcal {J} = \mathcal {I}$
. Hence,
$\mathcal {I}$
is left C-determined.
(b) Similar to (a).
It is interesting to note that, for an ideal torsion pair
$(\mathcal {I}, \mathcal {J})$
, the property of one of the ideals to be determined by an object on its non-trivial side is equivalent to the other ideal being functorially finite. Now the property of being functorially finite is equivalent for both of the ideals by Theorem 3.8. Hence, we can describe functorially finite torsion ideals by the objects that they are determined by, plus some additional information. This is done by the following result.
Theorem 4.5.
-
(a) In
$\mathrm{mod}\,A,$
we have an equality
$$ \begin{align*} \left\{\begin{matrix}\text{functorially finite}\\ \text{torsion ideals}\end{matrix}\right\} = \bigcup_{C \in \mathrm{mod}\,A } \left\{\begin{matrix}\text{left } C\text{-determined} \\ \text{torsion ideals}\end{matrix}\right\}. \end{align*} $$
-
(b) For
$C\in \mathrm{mod}\,A,$
there exists a one-to-one correspondence via
$$ \begin{align*} \left\{\begin{matrix}\text{left } C\text{-determined} \\ \text{torsion ideals}\end{matrix}\right\} \longleftrightarrow \left\{\begin{matrix} \text{bi-submodules of} \\ { {}_AC_{\mathrm{End}_A(C)^{\mathrm{op}}} } \end{matrix} \right\} \end{align*} $$
$\mathcal {I} \mapsto tC$
, where t denotes the subfunctor of
corresponding to
$\mathcal {I}$
.
Proof. (a) This follows by Lemma 4.4 and Theorem 3.8.
(b) By the functoriality of t, the submodule
$tC$
of C is invariant under
$\mathrm{End}_A(C)$
. Hence,
$tC$
is a submodule of the bimodule
${}_AC_{\mathrm{End}_A(C)^{\mathrm{op}}}$
. To show that the assignment
$C \mapsto tC$
is one to one, we construct an inverse assignment: For a submodule X of
${}_AC_{\mathrm{End}_A(C)^{\mathrm{op}}}$
, let
$\mathcal {I}_X$
be the collection of all morphisms
$\varphi \colon M \rightarrow N$
such that for all
$f\colon A\rightarrow M$
and
$g\colon N \rightarrow C,$
we have
$g \varphi f (1) \in X$
. Clearly,
$\mathcal {I}_X$
is an ideal and right A-determined as well as left C-determined by construction. By Proposition 4.1, the ideal
$\mathcal {I}_X$
is a left C-determined torsion ideal. Further,
$X\rightarrow C$
is contained in
$\mathcal {I}_X$
and
$C \rightarrow C/X$
in
$\mathcal {I}_X ^\perp $
. Hence, if t denotes the subfunctor of
corresponding to
$\mathcal {I}_X$
, then
$tC = X$
. It is left to show
$\mathcal {I}_{tC} = \mathcal {I}$
, where t denotes the subfunctor of
corresponding to
$\mathcal {I}$
. If
$\varphi \colon M \rightarrow N$
is in
$\mathcal {I}$
, then for all
$f \colon A \rightarrow M$
and
$g\colon N \rightarrow C,$
we have
$\mathrm{Im}\, g \varphi f \subseteq tC$
. Thus,
$g\varphi f (1) \in tC$
and
$\varphi \in \mathcal {I}_{tC}$
. On the contrary, if
$\varphi \colon M \rightarrow N$
is in
$\mathcal {I}_{tC}$
, then for all
$f\colon A \rightarrow M$
and
$g\colon N \rightarrow C,$
we have
$ g \varphi f(1) \in tC $
. Thus, always
$\mathrm{Im}\, g \varphi \subseteq tC$
and
$g \varphi \in \mathcal {I}$
. Because
$\mathcal {I}$
is left C-determined, we conclude that
$\varphi \in \mathcal {I}$
, so
$\mathcal {I}_{tC} = \mathcal {I}$
.
We proceed by giving a connection with the classical notion of morphisms determined by objects (see Section 1).
Lemma 4.6.
-
(a) Let
$\mathcal {I}$
be a functorially finite torsion ideal and
$\varphi \colon A \rightarrow M$
a left
$\mathcal {I}$
-approximation. Then
$\mathcal {I}$
is left C-determined if and only if
$\varphi $
is left C-determined. -
(b) Let
$\mathcal {J}$
be a functorially finite torsion-free ideal and
$\psi \colon N \rightarrow DA^{\mathrm{op}}$
a right
$\mathcal {J}$
-approximation. Then
$\mathcal {J}$
is right C-determined if and only if
$\psi $
is right C-determined.
Proof. (a) Let
$\mathcal {I}$
be left C-determined and
$\varphi ' \colon A \rightarrow M'$
a morphism such that
$f \varphi '$
factors through
$\varphi $
for all
$f \colon M' \rightarrow C$
. Then always
$f \varphi ' \in \mathcal {I}$
and because
$\mathcal {I}$
is left C-determined, we conclude that
$\varphi '\in \mathcal {I}$
. Hence,
$\varphi '$
factors through the left
$\mathcal {I}$
-approximation
$\varphi $
. It follows that
$\varphi $
is left C-determined.
Let
$\varphi $
be left C-determined and
$\varphi ' \colon M' \rightarrow N$
a morphism such that
$f \varphi ' \in \mathcal {I}$
for all
$f \colon N \rightarrow C$
. Then for all
$g \colon A \rightarrow M'$
, the morphism
$f \varphi ' g \in \mathcal {I}$
factors through
$\varphi $
. Because f is arbitrary and
$\varphi $
left C-determined, it follows that
$\varphi ' g$
factors through
$\varphi $
. Thus,
$\varphi ' g\in \mathcal {I}$
and because
$\mathcal {I}$
is right A-determined by Proposition 4.1, we conclude that
$\varphi \in \mathcal {I}$
. Hence,
$\mathcal {I}$
is left C-determined.
(b) Similar to (a).
For a functorially finite ideal torsion pair
$(\mathcal {I}, \mathcal {J}),$
we have already seen a way to find
$C \in \mathrm{mod}\,A$
such that
$\mathcal {I}$
is left C-determined. Namely, we can choose C by a right
$\mathcal {J}$
-approximation
$C \rightarrow DA^{\mathrm{op}}$
(see the proof of Lemma 4.4). The following result shows a way to find
$C \in \mathrm{mod}\,A$
, only considering the ideal
$\mathcal {I}$
.
Corollary 4.7.
-
(a) Let
$\mathcal {I}$
be a functorially finite torsion ideal,
$\varphi \colon A \rightarrow M$
a left
$\mathcal {I}$
-approximation, K the kernel of
$\varphi ,$
and Q its cokernel. Then
$\mathcal {I}$
is left C-determined for
$C = \tau Q \oplus I$
, where
$\tau $
denotes the Auslander–Reiten translation and I the injective hull of the top of K. -
(b) Let
$\mathcal {J}$
be a functorially finite torsion-free ideal,
$\psi \colon N \rightarrow DA^{\mathrm{op}} $
a right
$\mathcal {J}$
-approximation, K the kernel of
$\psi ,$
and Q its cokernel. Then
$\mathcal {J}$
is right C-determined for
$C = \tau ^{-} K \oplus P$
, where
$\tau ^{-}$
denotes the inverse of the Auslander–Reiten translation and P the projective cover of the socle of K.
5. Subcategories related to ideal torsion pairs
We start by discussing the obvious subcategories of
$\mathrm{mod}\,A$
which can relate to ideal torsion pairs: torsion classes and torsion-free classes.
Remark 5.1. Let
$(\mathcal {T}, \mathcal {F})$
be a torsion pair. Then
$\mathrm{Hom}_A(M,N) = 0$
for all
$M\in \mathcal {T}$
and
$N\in \mathcal {F}$
implies
$\psi \varphi = 0$
for all
$\varphi \in \langle \mathcal {T} \rangle $
and
$\psi \in \langle \mathcal {F} \rangle $
. Further, the short exact sequence
with
$L\in \mathcal {T}$
and
$N\in \mathcal {F}$
also fulfills
$\varphi \in \langle \mathcal {T} \rangle $
and
$\psi \in \langle \mathcal {F} \rangle $
. Hence,
$(\langle \mathcal {T} \rangle , \langle \mathcal {F} \rangle )$
is an ideal torsion pair. Thus,
$\langle \mathcal {T} \rangle $
is a torsion ideal and
$\langle \mathcal {F} \rangle $
a torsion-free ideal.
Applying the theory of ideal torsion pairs to torsion pairs, we obtain the following well-known result [Reference Smalø28].
Corollary 5.2. Let
$(\mathcal {T}, \mathcal {F})$
be a torsion pair. Then
$\mathcal {T}$
is functorially finite if and only if
$\mathcal {F}$
is functorially finite.
Proof. By Lemma 2.3, we can check that
$\langle \mathcal {T} \rangle $
is functorially finite if and only if
$\langle \mathcal {F} \rangle $
is functorially finite. This is the case by Remark 5.1 and Theorem 3.8.
For an ideal
$\mathcal {I,}$
let
$\mathrm{ob}\, \mathcal {I}$
be the collection of all
$M\in \mathrm{mod}\,A$
with
$1_M \in \mathcal {I}$
or equivalently
$tM = M$
. A full additive subcategory
$\mathcal {C}$
of
$\mathrm{mod}\,A$
is a mono-closed (epi-closed) class, if
$M\in \mathcal {C}$
and
$N\leq M$
implies
$N \in \mathcal {C}$
(resp.
$M/N \in \mathcal {C}$
). As it turns out, those classes are precisely the ones that arise from ideal torsion pairs.
Lemma 5.3.
-
(a) If
$\mathcal {C}$
is an epi-closed class, then
$\langle \mathcal {C}\rangle $
is a torsion ideal. If
$\mathcal {I}$
is a torsion ideal, then
$\mathrm{ob}\, \mathcal {I}$
is an epi-closed class. -
(b) If
$\mathcal {C}$
is a mono-closed class, then
$\langle \mathcal {C}\rangle $
is a torsion-free ideal. If
$\mathcal {J}$
is a torsion-free ideal, then
$\mathrm{ob}\, \mathcal {J}$
is a mono-closed class.
Proof. (a) Let
$\varphi $
be a morphism and f an epimorphism with
$\varphi f \in \langle \mathcal {C} \rangle $
. Then
$\varphi f$
factors as
$\varphi _1 \varphi _2$
with
$\varphi _1$
starting in
$C\in \mathcal {C}$
. Because
$\mathcal {C}$
is an epi-closed class, the image of
$\varphi _1$
is contained in
$\mathcal {C}$
. Now
$\mathrm{Im}\, \varphi = \mathrm{Im}\, \varphi f \subseteq \mathrm{Im}\, \varphi _1 $
implies
$\varphi \in \langle \mathcal {C}\rangle $
. By Lemma 3.3, it follows that
$\langle \mathcal {C} \rangle $
is a torsion ideal.
Let
$M\in \mathrm{ob}\, \mathcal {I}$
and
$N \leq M$
. Consider the projection
$\pi \colon M \rightarrow M/N$
. Then
$\pi = \pi 1_M \in \mathcal {I}$
. Further,
$\pi = 1_{M/N} \pi $
. By Lemma 3.3, it follows that
$1_{M/N} \in \mathcal {I}$
. Hence,
$M/N \in \mathrm{ob}\, \mathcal {I}$
and
$\mathrm{ob}\, \mathcal {I}$
is an epi-closed class.
(b) Similar to (a).
Remark 5.4. By Lemma 2.1, every torsion(-free) class in
$\mathrm{mod}\,A$
is an epi-closed (mono-closed) class. On the contrary, an epi-closed (mono-closed) class in
$\mathrm{mod}\,A$
is a torsion(-free) class if and only if it is closed under extensions. In particular, this shows with Lemma 5.3 that, in general, there are much more ideal torsion pairs than torsion pairs.
Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair of
$\mathrm{mod}\,A$
. Then
$\langle \mathrm{ob}\, \mathcal {I} \rangle \subseteq \mathcal {I}$
is a torsion ideal and
$\langle \mathrm{ob}\, \mathcal {J} \rangle \subseteq \mathcal {J}$
a torsion-free ideal by Lemma 5.3. We show a criterion for when
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
or
$\mathcal {J} = \langle \mathrm{ob}\, \mathcal {J} \rangle $
.
Proposition 5.5. Let
$(\mathcal {I}, \mathcal {J})$
be an ideal torsion pair and t the corresponding subfunctor of
.
-
(a) The equality
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
holds if and only if
$t^2 = t$
. In that case,
$tM \in \mathrm{ob}\, \mathcal {I}$
for all
$M\in \mathrm{mod}\,A$
. -
(b) The equality
$\mathcal {J} = \langle \mathrm{ob}\, \mathcal {J}\rangle $
holds if and only if
$(1/t)^2 = 1/t$
. In that case,
$M/tM \in \mathrm{ob}\, \mathcal {J}$
for all
$M\in \mathrm{mod}\,A$
. -
(c) The equalities
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
and
$\mathcal {J} = \langle \mathrm{ob}\, \mathcal {J} \rangle $
hold if and only if
$(\mathrm{ob}\, \mathcal {I}, \mathrm{ob}\, \mathcal {J})$
is a torsion pair.
Proof. (a) By Proposition 3.12, the torsion ideal
$\mathcal {I}^2$
corresponds to
$t^2$
. Hence, if
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
, then
$\mathcal {I}^2 = \mathcal {I}$
and
$t^2 = t$
. Let
$M\in \mathrm{mod}\,A$
and consider the canonical inclusion
$\iota \colon t^2 M \rightarrow tM$
. Then
$\iota \in \mathcal {I}$
. If
$t^2 = t$
, then
$\iota = 1_{tM}$
and
$tM\in \mathrm{ob}\, \mathcal {I}$
. In particular
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
, as every morphism
$L\rightarrow M$
in
$\mathcal {I}$
factors through
$tM$
.
(b) Similar to (a).
(c) If
$(\mathrm{ob}\, \mathcal {I}, \mathrm{ob}\, \mathcal {J})$
is a torsion pair, then
$(\langle \mathrm{ob}\, \mathcal {I} \rangle , \langle \mathrm{ob}\, \mathcal {J} \rangle )$
is an ideal torsion pair by Remark 5.1. Since
$\langle \mathrm{ob}\, \mathcal {I} \rangle \subseteq \mathcal {I}$
and
$\langle \mathrm{ob}\, \mathcal {J} \rangle \subseteq \mathcal {J}$
, it follows that
$(\mathcal {I}, \mathcal {J})$
equals
$(\langle \mathrm{ob}\, \mathcal {I} \rangle , \langle \mathrm{ob}\, \mathcal {J} \rangle )$
by Proposition 3.2. On the other hand, if
$\mathcal {I} = \langle \mathrm{ob}\, \mathcal {I} \rangle $
and
$\mathcal {J} = \langle \mathrm{ob}\, \mathcal {J} \rangle $
, then
$\mathrm{Hom}_A(\mathrm{ob}\, \mathcal {I} , \mathrm{ob}\, \mathcal {J} ) = 0$
. Further, for all
$M\in \mathrm{mod}\,A,$
there exists a short exact sequence
$ 0 \rightarrow tM \rightarrow M \rightarrow M/tM \rightarrow 0$
with
$tM\in \mathrm{ob}\, \mathcal {I}$
by (a) and
$M/tM \in \mathrm{ob}\, \mathcal {J}$
by (b). Hence,
$(\mathrm{ob}\, \mathcal {I}, \mathrm{ob}\, \mathcal {J})$
is a torsion pair.
We continue by discussing when epi-closed classes and mono-closed classes are functorially finite. For
$M\in \mathrm{mod}\,A,$
let
$\mathrm{add}\, M$
be the smallest full additive subcategory of
$\mathrm{mod}\,A$
containing M, let
$\mathrm{gen}\, M$
be the closure of
$\mathrm{add}\, M$
under factor modules, and let
$\mathrm{cogen}\, M$
be the closure of
$\mathrm{add}\, M $
under submodules. Clearly,
$\mathrm{gen}\, M$
is the smallest epi-closed class and
$\mathrm{cogen}\, M$
the smallest mono-closed class containing M. We reprove the classification of functorially finite epi-closed (mono-closed) classes in [Reference Auslander and Smalø7, Propositions 4.6 and 4.7], using the theory of ideal torsion pairs.
Proposition 5.6.
-
(a) A full additive subcategory
$\mathcal {C}$
of
$\mathrm{mod}\,A$
is a functorially finite epi-closed class if and only if
$\mathcal {C} = \mathrm{gen}\, M$
for some
$M\in \mathrm{mod}\,A$
. -
(b) A full additive subcategory
$\mathcal {C}$
of
$\mathrm{mod}\,A$
is a functorially finite mono-closed class if and only if
$\mathcal {C} = \mathrm{cogen}\, M$
for some
$M\in \mathrm{mod}\,A$
.
Proof. (a) Let
$\mathcal {C}$
be a functorially finite epi-closed class and
$\varphi \colon A \rightarrow C_A$
a left
$\mathcal {C}$
-approximation. Then
$C_A \in \mathcal {C}$
and for
$M\in \mathcal {C}$
an epimorphism
$A^n \rightarrow M$
must factor through
$\varphi ^n$
for some
. We conclude that
$\mathrm{gen}\, C_A = \mathcal {C}$
.
For the other implication, consider the ideal
$\mathcal {I} = \langle \mathrm{add}\, M \rangle $
. If
$\varphi _1, \dots , \varphi _n$
denotes a basis of the k-module
$\mathrm{Hom}_A(A,M)$
, then the induced morphism
$\varphi \colon A \rightarrow M^n$
is a left
$\mathcal {I}$
-approximation. Thus, by Corollary 4.3, the ideal
$\mathrm {I}(\mathcal {I})$
is functorially finite. By Lemma 4.3, the ideal
$\mathrm {I}(\mathcal {I})$
equals
$\langle \mathrm{gen}\, M \rangle $
, as
$\mathrm{gen}\, M$
is the smallest epi-closed class containing M. Hence,
$\mathrm{gen}\, M$
is functorially finite.
(b) Similar to (a).
It would be nice to have a similar result as above, classifying all functorially finite torsion ideals and torsion-free ideals (not only those generated by epi-closed classes and mono-closed classes). To do so, we consider the abelian category
$\mathrm{mor}\,A$
, where the objects are morphisms in
$\mathrm{mod}\,A$
and the morphisms are commutative squares of morphisms in
$\mathrm{mod}\,A$
. Notice that
$\mathrm{mor}\,A$
is equivalent to the category of finitely generated (left) B-modules over the Artin algebra
$$ \begin{align*} B= \begin{bmatrix} A &0 \\ A & A \end{bmatrix}. \end{align*} $$
Lemma 5.7.
-
(a) If
$\mathcal {I}$
is a torsion ideal of
$\mathrm{mod}\,A$
, then
$\mathcal {I}$
is an epi-closed class in
$\mathrm{mor}\,A$
. If
$\mathcal {C}$
is an epi-closed class in
$\mathrm{mor}\,A$
, then the collection of all morphisms factoring through a morphism in
$\mathcal {C}$
is a torsion ideal. -
(b) If
$\mathcal {J}$
is a torsion-free ideal of
$\mathrm{mod}\,A$
, then
$\mathcal {J}$
is a mono-closed class in
$\mathrm{mor}\,A$
. If
$\mathcal {C}$
is a mono-closed class in
$\mathrm{mor}\,A$
, then the collection of all morphisms factoring through a morphism in
$\mathcal {C}$
is a torsion-free ideal.
Proof. (a) Let
$\varphi \colon M \rightarrow N$
be in
$\mathcal {I}$
. An epimorphism in
$\mathrm{mor}\,A$
starting in
$\varphi $
is given by a commutative square of morphisms

with f and g surjective. Then
$\varphi ' f = g \varphi \in \mathcal {I}$
. Hence,
$\varphi ' \in \mathcal {I}$
by Lemma 3.3. It follows that
$\mathcal {I}$
is an epi-closed class in
$\mathrm{mor}\,A$
.
Let
$\mathcal {C}$
be an epi-closed class in
$\mathrm{mor}\,A$
and
$\mathcal {I}$
the collection of all morphisms factoring through a morphism in
$\mathcal {C}$
. Clearly,
$\mathcal {I}$
is an ideal. To see that
$\mathcal {I}$
is a torsion ideal, we make use of Lemma 3.3. Let
$\varphi \colon L \rightarrow N$
be a morphism and
$f\colon M \rightarrow L$
an epimorphism with
$\varphi f\in \mathcal {I}$
. Then
$\varphi f $
factors as
$\varphi _1 \alpha \varphi _2$
for some
$\alpha \colon M' \rightarrow N'$
in
$\mathcal {C}$
,
$\varphi _1\colon N' \rightarrow N$
and
$\varphi _2 \colon M \rightarrow M'$
. Now consider the commutative square

where
$\pi _1 \colon M'\rightarrow \mathrm{Im}\, \varphi _1\alpha , \pi _2 \colon N'\rightarrow \mathrm{Im}\, \varphi _1$
are the canonical projections and
$\iota \colon \mathrm{Im}\, \varphi _1\alpha \rightarrow \mathrm{Im}\, \varphi _1$
the canonical inclusion. Because
$\mathcal {C}$
is epi-closed and
$\alpha \in \mathcal {C}$
, we conclude that
$\iota \in \mathcal {C}$
. Now
$\mathrm{Im}\, \varphi = \mathrm{Im}\, \varphi f = \mathrm{Im}\, \varphi _1 \alpha \varphi _2 \subseteq \mathrm{Im}\, \varphi _1 \alpha $
. Hence,
$\varphi $
factors through
$\iota \in \mathcal {C}$
and so
$\varphi \in \mathcal {I}$
. Thus,
$\mathcal {I}$
is a torsion ideal.
(b) Similar to (a).
Proposition 5.8.
-
(a) An ideal
$\mathcal {I}$
is a functorially finite torsion ideal if and only if there exists
$\varphi \in \mathrm{mor}\,A$
such that
$\mathcal {I}$
equals the collection of all morphisms factoring through a morphism in
$\mathrm{gen}\, \varphi $
. -
(b) An ideal
$\mathcal {J}$
is a functorially finite torsion-free ideal if and only if there exists
$\psi \in \mathrm{mor}\,A$
such that
$\mathcal {J}$
equals the collection of all morphisms factoring through a morphism in
$\mathrm{cogen}\, \psi $
.
Proof. (a) Let
$\mathcal {I}$
be a functorially finite torsion ideal and
$\varphi \colon A \rightarrow C_A$
a left
$\mathcal {I}$
-approximation. We show that
$\mathcal {I}$
equals the collection of all morphisms factoring through a morphism in
$\mathrm{gen}\, \varphi $
. First,
$\mathrm{gen}\, \varphi \subseteq \mathcal {I}$
by Lemma 5.7. Let
$f\colon M \rightarrow N$
be in
$\mathcal {I}$
and
$\pi \colon A^n \rightarrow M$
an epimorphism for some
. Then
$f \pi $
factors through
$\varphi ^n$
. Hence, there exists
$g \colon C_A^n \rightarrow N$
with
$f \pi = g \varphi ^n$
. It follows that
$\mathrm{Im}\, f = \mathrm{Im}\, f\pi = \mathrm{Im}\, g \varphi ^n \subseteq \mathrm{Im}\, g$
. Thus, there exists a commutative diagram of morphisms

where
$\pi '\colon C_A^n \rightarrow \mathrm{Im}\, g$
denotes the canonical projection and
$f = f_2 f_1$
. Clearly,
$f_1 \in \mathrm{gen}\, \varphi $
. It follows that f factors through a morphism in
$\mathrm{gen}\, \varphi $
and
$\mathcal {I} = \mathrm{gen}\, \varphi $
.
Let
$\mathcal {I}$
equal the collection of all morphisms factoring through a morphism in
$\mathcal {C} = \mathrm{gen}\, \varphi $
for some
$\varphi \in \mathrm{mor}\,A$
. Then
$\mathcal {I}$
is a torsion ideal by Lemma 5.7. Further,
$\mathcal {C}$
is functorially finite in
$\mathrm{mor}\,A$
by Proposition 5.6. Hence, for
$M\in \mathrm{mod}\,A,$
there exists a left
$\mathcal {C}$
-approximation of
$1_M$
, given by a commutative square of morphisms

We show that f is a left
$\mathcal {I}$
-approximation. First,
$f\in \mathcal {I}$
since
$f_2 \in \mathcal {C}$
and
$f = f_2f_1$
. Now let
$g\colon M \rightarrow N$
be a morphism factoring through a morphism
$\alpha \colon M' \rightarrow N'$
in
$\mathcal {C}$
. Then
$g = g_2 \alpha g_1$
for
$g_1 \colon M \rightarrow M'$
and
$g_2 \colon N' \rightarrow N$
. We consider the commutative square

It must factor through the left
$\mathcal {C}$
-approximation of
$1_M$
from above, so there exists a commutative diagram of morphisms

with
$g_1 = h_1 f_1$
. In total,
$g = g_2 \alpha g_1 = g_2 \alpha h_1 f_1 = g_2 h_2 f$
. It follows that f is a left
$\mathcal {I}$
-approximation, so
$\mathcal {I}$
is functorially finite.
(b) Similar to (a).
6. Transfinite powers of the radical and related ideal torsion pairs
Preprojective modules were first introduced by Dlab and Ringel [Reference Dlab, Ringel and Gordon11] for finite-dimensional tensor algebras and later generalized by Auslander and Smalø for arbitrary Artin algebras [Reference Auslander and Smalø7]. Our aim is to extend the class of preprojective modules using the theory of ideal torsion pairs. Let us mention that Krause also extended the class of preprojective modules in [Reference Krause18], using a different approach. In what follows, all results can be dualized leading to an extension of the class of preinjective modules.
Following Prest [Reference Prest22], we define the notion of transfinite powers of an ideal
$\mathcal {I}$
of
$\mathrm{mod}\,A$
: For
let
$\mathcal {I}^n$
denote the collection of all n-fold compositions of morphisms in
$\mathcal {I}$
(in particular,
$\mathcal {I}^0 = \mathrm{Hom}_A$
). If
$\lambda $
is a non-zero limit ordinal, let
$\mathcal {I}^\lambda = \bigcap _{\alpha < \lambda } \mathcal {I}^\alpha $
. If
$\alpha $
is an arbitrary infinite ordinal, then
$\alpha = \lambda + n$
for a limit ordinal
$\lambda $
and
$n\in \mathbb {N}$
, and we let
$\mathcal {I}^\alpha = (\mathcal {I}^\lambda )^{n+1}$
. Lastly, we define
$\mathcal {I}^\infty = \bigcap _{\alpha } \mathcal {I}^\alpha $
. The case
$\mathcal {I} = \mathrm{rad}_A$
will be most important to us.
Remark 6.1. From the definition of
$\mathcal {I}^\alpha $
, it is not hard to see that
$\mathcal {I}^\alpha $
is an ideal for all ordinal numbers
$\alpha $
. This yields a descending chain of ideals
For
$M,N\in \mathrm{mod}\,A,$
the expression
$\mathcal {I}^\alpha (M,N)$
is a submodule of the finitely generated k-module
$\mathrm{Hom}_A (M,N)$
. Thus, if
$\lambda $
is a limit ordinal (or
$\infty $
), then
$\mathcal {I}^\lambda (M,N) = \bigcap _{\alpha < \lambda } \mathcal {I}^\alpha (M,N) = \mathcal {I}^\alpha (M,N)$
for some
$\alpha < \lambda $
.
The projective rank of a module
$M \in \mathrm{mod}\,A$
, denoted by
$\mathrm{prk}\, M$
, is the smallest ordinal number
$\alpha $
(or
$\infty $
) with
$M\in \mathrm{ob}\, \mathrm{I}(\mathrm{rad}_A^\alpha )$
. Now
$M \in \mathrm{ob}\, \mathrm{I}(\mathrm{rad}_A^\alpha )$
is equivalent to
$\varphi \in \mathrm{rad}_A^\alpha $
for all
$\varphi \colon A\rightarrow M$
by Remark 4.2. Hence,
$\mathrm{prk}\, M$
equals the smallest
$\alpha $
such that there exists
$\varphi \colon A \rightarrow M$
in
$\mathrm{rad}_A^\alpha \backslash \mathrm{rad}_A^{\alpha +1}$
(or
$\infty $
if no such
$\alpha $
exists). In particular, if M is indecomposable, then
$\mathrm{prk}\, M = 0$
is equivalent to M being projective, since exactly in that case there exists a split epimorphism
$A\rightarrow M$
.
Remark 6.2. Let
$\alpha $
be an ordinal number (or
$\infty $
). Then
$\mathrm{ob}\, \mathrm{I}(\mathrm{rad}_A^\alpha )$
is an epi-closed class by Lemma 5.3. This has the following immediate consequences:
-
(i) For every epimorphism
$M \rightarrow N$
in
$\mathrm{mod}\,A,$
we have
$\mathrm{prk}\, M \leq \mathrm{prk}\, N$
. -
(ii) For
$M, N \in \mathrm{mod}\,A,$
we have
$\mathrm{prk}\, M \oplus N = \min \{\mathrm{prk}\, M, \mathrm{prk}\, N\}$
.
Recall that an indecomposable module
$X\in \mathrm{mod}\,A$
is preprojective if there exists a finite collection
$\mathcal {C}$
of indecomposable modules with the following property: If there exists an epimorphism
$M\rightarrow X$
, then M must have a direct summand isomorphic to a module in
$\mathcal {C}$
(see [Reference Auslander and Smalø7]). In what follows, we show that the indecomposable preprojective modules are precisely those of finite projective rank.
Lemma 6.3. The ideal
$\mathrm{rad}_A^n$
is functorially finite for all
. Moreover, if
$\varphi \colon A\rightarrow C_A$
is a left minimal
$\mathrm{rad}_A^n$
-approximation, then an indecomposable module
$X\in \mathrm{mod}\,A$
is isomorphic to a direct summand of
$C_A$
if and only if there exists
$\psi \colon A \rightarrow X$
in
$\mathrm{rad}_A^{n}\backslash \mathrm{rad}_A^{n+1}$
.
Proof. By Proposition 2.4, the ideal
$\mathrm{rad}_A$
is functorially finite. For
$M\in \mathrm{mod}\,A,$
consider a sequence
of left
$\mathrm{rad}_A$
-approximations
$M_i \rightarrow M_{i+1}$
. We show that
$f = f_n f_{n-1} \dots f_1$
is a left
$\mathrm{rad}_A^n$
-approximation. Clearly,
$f\in \mathrm{rad}_A^n$
. Let
$g\in \mathrm{rad}_A^n$
starting in M. Then g factors as
$g_n g_{n-1} \dots g_1$
with
$g_i\colon N_{i-1} \rightarrow N_i$
in
$\mathrm{rad}_A$
and
$N_0 = M$
. Because
$f_i$
it a left
$\mathrm{rad}_A$
-approximation for all i, we obtain a commutative diagram

It follows that g factors through f. Hence, f is a left
$\mathrm{rad}_A^n$
-approximation and the ideal
$\mathrm{rad}_A^n$
is covariantly finite. Similarly,
$\mathrm{rad}_A^n$
is contravariantly finite and thus functorially finite.
Let
$X\in \mathrm{mod}\,A$
be indecomposable such that there exists
$\psi \colon A\rightarrow X$
in
$\mathrm{rad}_A^n \backslash \mathrm{rad}_A^{n+1}$
. Then
$\psi $
factors through the left
$\mathrm{rad}_A^n$
-approximation
$\varphi \colon A \rightarrow C_A$
, so
$\psi = g \varphi $
for
$g \colon C_A \rightarrow X$
. Further, the morphism g cannot be radical, as otherwise
$\psi = g \varphi \in \mathrm{rad}_A^{n+1}$
. It follows that g is a split epimorphism, since X is indecomposable. Hence, X is a direct summand of
$C_A$
.
Let
$C_A = X\oplus Y$
with X indecomposable, let
$\pi _X \colon C_A \rightarrow X, \pi _Y \colon C_A \rightarrow Y$
be the canonical projections and
$\iota _X \colon X \rightarrow C_A, \iota _Y \colon Y \rightarrow C_A$
the canonical inclusions. For a contradiction, suppose that
$\mathrm{rad}_A^n(A,X) = \mathrm{rad}_A^{n+1}(A,X)$
. Then
$\pi _X \varphi \in \mathrm{rad}_A^{n+1}$
factors as
$\pi _X \varphi = g f$
for
$f \in \mathrm{rad}_A^n$
and
$g \in \mathrm{rad}_A$
. Now f factors through the left minimal
$\mathrm{rad}_A^n$
-approximation
$\varphi $
, so
$f = h \varphi $
. Hence,
Because
$\varphi $
is left minimal, the morphism
$\iota _X gh+\iota _Y \pi _Y$
is an isomorphism. Thus, there exists
$\psi \colon C_A \rightarrow C_A$
with
$\psi (\iota _X gh+\iota _Y \pi Y) = 1_{C_A}$
. We conclude that
Now
$g \in \mathrm{rad}_A$
implies
$1_X \in \mathrm{rad}_A$
, which is a contradiction.
Corollary 6.4. Let
$X\in \mathrm{mod}\,A$
be indecomposable and
$A \rightarrow C_{n}$
a left minimal
$\mathrm{rad}_A^n$
-approximation for all
. Then X can only be isomorphic to a direct summand of
$C_{n}$
for finitely many n.
Proof. Suppose that X is a direct summand of infinitely many
$C_{n}$
. Then the sequence
$\mathrm{rad}_A(A,X) \supseteq \mathrm{rad}_A^2(A,X) \supseteq \dots $
of submodules of the k-module
$\mathrm{Hom}_A(A,X)$
would have infinitely many proper inclusions by Lemma 6.3, which contradicts that
$\mathrm{Hom}_A(A,X)$
is of finite length as a k-module.
Proposition 6.5. Let
$X\in \mathrm{mod}\,A$
be indecomposable. The projective rank of X is finite if and only if X is preprojective.
Proof. Let
and
$\mathcal {C}$
be the collection of all indecomposable modules
$Y \in \mathrm{mod}\,A$
with
$\mathrm{prk}\, Y \leq n$
. Then for all
$Y\in \mathcal {C,}$
there exists
$A \rightarrow Y$
in
$\mathrm{rad}_A^{m}\backslash \mathrm{rad}_A^{m+1}$
for some
$m\leq n$
. Thus, by Lemma 6.3, the number of isomorphism types of modules in
$\mathcal {C}$
is finite. Now if there exists an epimorphism
$M\rightarrow X$
such that M does not have a direct summand isomorphic to a module in
$\mathcal {C}$
, then
by Remark 6.2. This is a contradiction. It follows that X is preprojective.
Let X be preprojective and
$\mathcal {C}$
a finite collection of indecomposable modules such that for every epimorphism
$M\rightarrow X,$
the module M has a direct summand isomorphic to a module in
$\mathcal {C}$
. By Corollary 6.4, there exists
such that for a left
$\mathrm{rad}_A^{n}$
-approximation
$A\rightarrow C_A$
, the module
$C_A$
does not contain a direct summand isomorphic to a module in
$\mathcal {C}$
. For a contradiction, suppose that X has non-finite projective rank. Then an epimorphism
$A^m\rightarrow X$
is contained in
$\mathrm{rad}_A^n$
so it must factor through
$A^m \rightarrow C_A^m$
. This yields an epimorphism
$C_A^m \rightarrow X$
, which is a contradiction since
$C_A^m$
does not have a direct summand isomorphic to a module in
$\mathcal {C}$
.
We continue by investigating the modules of non-finite projective rank. It is not clear for which ordinal number
$\alpha $
(or
$\infty $
) there exists a non-zero module
$X\in \mathrm{mod}\,A$
of projective rank
$\alpha $
. In what follows, we show a criterion for this.
Lemma 6.6. Let
$\mathcal {I}$
be an ideal of
$\mathrm{mod}\,A$
and
$\lambda \neq 0$
a limit ordinal (or
$\infty $
). Then
$\mathrm { I}(\mathcal {I}^\lambda ) = \langle \mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda ) \rangle $
. In particular,
$\mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda ) \neq 0$
if and only if
$\mathcal {I}^\lambda \neq 0$
.
Proof. Clearly,
$\langle \mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda )\rangle \subseteq \mathrm {I}(\mathcal {I}^\lambda )$
. Let
$M \in \mathrm{mod}\,A$
, t the subfunctor of
corresponding to
$\mathrm {I}(\mathcal {I}^\lambda )$
,
$\iota \colon tM \rightarrow M$
the canonical inclusion, and
$\pi \colon A^n \rightarrow tM$
an epimorphism. Then
$\iota \pi \in \mathcal {I}^\lambda $
by Remark 4.2. By Remark 6.1, we can choose
$\alpha < \lambda $
with
$\mathcal {I}^\lambda (A, M) = \mathcal {I}^\alpha (A,M)$
and
$\mathcal {I}^\lambda (A,tM) = \mathcal {I}^\alpha (A,tM)$
. Further,
$\iota \pi $
factors as
$g f$
for
$f, g \in \mathcal {I}^\alpha $
since
$(\mathcal {I}^\alpha )^2 \subseteq \mathcal {I}^\lambda $
. Now for all
$\varphi $
starting in
$A,$
the composition
$g \varphi $
is in
$\mathcal {I}^\alpha (A,M) = \mathcal {I}^\lambda (A,M)$
. Hence,
$g \in \mathrm {I}(\mathcal {I}^\lambda )$
factors through
$\iota $
and
$g = \iota \psi $
. Since
$ \psi f \in \mathcal {I}^\alpha (A^n, tM) = \mathcal {I}^\lambda (A^n, tM)$
, we deduce that
$\psi f$
factors through the canonical inclusion
$\iota ' \colon t(tM) \hookrightarrow tM$
. Thus,
$\iota = gf = \iota \psi f$
factors through
$\iota '$
and we conclude that
$t(tM) = tM$
. It follows that
$tM \in \mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda )$
and hence
$\mathrm {I}(\mathcal {I}) = \langle \mathrm{ob}\, \mathrm { I}(\mathcal {I}^\lambda ) \rangle $
.
Further, if
$\mathcal {I}^\lambda \neq 0,$
then
$\mathrm {I}(\mathcal {I}^\lambda ) \neq 0$
and thus
$\mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda ) \neq 0$
. On the contrary, if
$\mathrm{ob}\, \mathrm {I}(\mathcal {I}^\lambda ) \neq 0,$
then
$\mathrm {I}(\mathcal {I}^\lambda ) \neq 0$
. In that case,
$\mathcal {I}^\lambda (A,-) \neq 0$
by Remark 4.2.
Corollary 6.7. Let
$\lambda \neq 0$
be a limit ordinal (or
$\infty $
). There exists a non-zero module of projective rank greater than or equal to
$\lambda $
if and only if
$\mathrm{rad}_A^\lambda \neq 0$
.
For the finite case, we have already seen that for each
there can only be finitely many isomorphism types of indecomposable modules of projective rank n. The infinite case is completely opposite. We will show that if there exists an indecomposable module of projective rank
$\alpha $
for an infinite ordinal number
$\alpha $
(or
$\alpha = \infty $
), then there exist infinitely many isomorphism types of indecomposable modules of projective rank close to
$\alpha $
.
Lemma 6.8. Let
$\alpha $
be a non-finite ordinal number (or
$\infty )$
and write
$\alpha = \lambda + n$
for a limit ordinal
$\lambda $
(or
$\infty $
) and
. For all
$M\in \mathrm{mod}\,A$
with
$\mathrm{prk}\, M = \alpha ,$
there exists a radical epimorphism
$M' \rightarrow M$
with
$\lambda \leq \mathrm{prk}\, M' \leq \alpha $
.
Proof. Let t be the subfunctor of
corresponding to
$\mathrm{I}(\mathrm{rad}_A^\lambda )$
, let
$\varphi \colon C_M \rightarrow M$
be a right
$\mathrm{rad}_A$
-approximation, and
$\pi \colon A^n \rightarrow M$
an epimorphism. The morphism
$\pi $
is contained in
$\mathrm{rad}_A^\alpha \subseteq \mathrm{rad}_A^\lambda $
since
$\mathrm{prk}\, M = \alpha $
. We can choose
$\beta < \lambda $
with
$\mathcal {I}^\lambda (A, C_M) = \mathcal {I}^\beta (A,C_M)$
by Remark 6.1. Further, the morphism
$\pi $
factors as
$g f$
for
$f, g \in \mathrm{rad}_A^\beta $
since
$(\mathrm{rad}_A ^\beta )^2 \subseteq \mathrm{rad}_A^\lambda $
. In particular, g must factor through the right
$\mathrm{rad}_A$
-approximation
$\varphi $
, so
$g = \varphi g'$
. Now
$g'f \in \mathcal {I}^\beta (A,C_M) = \mathcal {I}^\lambda (A,C_M)$
. Hence,
$g'f$
factors through the canonical inclusion
$\iota \colon t C_M \rightarrow C_M$
, so
$g'f = \iota f'$
. This yields
$\pi = gf = \varphi g' f = \varphi \iota f'$
. Because
$\pi $
is an epimorphism, it follows that
$\varphi \iota $
is an epimorphism, which is also radical since
$\varphi \in \mathrm{rad}_A$
. The morphism
$\varphi \iota $
starts in
$tC_M$
, which is contained in
$\mathrm{ob}\, \mathrm{I}(\mathcal {I}^\lambda )$
by Lemma 6.6. We conclude that
$\mathrm{prk}\, tC_M \geq \lambda $
. By Remark 6.2, the inequality
$\mathrm{prk}\, tC_M \leq \mathrm{prk}\, M = \alpha $
holds. This proves the claim with
$M' = tC_M$
.
Proposition 6.9. Let
$X\in \mathrm{mod}\,A$
be indecomposable of non-finite projective rank and write
$\mathrm{prk}\, X = \lambda + n$
for a limit ordinal
$\lambda $
(or
$\infty $
) and
. There exists a chain
of radical morphisms
$f_i$
between indecomposable modules
$X_i$
with
$f_1 f_2 \dots f_i \neq 0$
and
$\lambda \leq \mathrm{prk}\, X_i \leq \alpha $
for all i.
Proof. We construct modules
$M_i, N_i$
recursively such that every indecomposable direct summand of
$M_i$
has projective rank in the interval
$[\lambda , \alpha ]$
and
$\mathrm{prk}\, N_i> \alpha $
. Further, we construct epimorphisms
$$ \begin{align*} \varphi_i \colon M_i \oplus N_i \xrightarrow{\begin{pmatrix} \alpha_i & \beta_i\\0& \gamma_i \end{pmatrix}} M_{i-1}\oplus N_{i-1} \end{align*} $$
with
$\alpha _i\in \mathrm{rad}_A$
. Let
$M_0 = X$
and
$N_0 = 0$
. Suppose given
$M_{i-1}, N_{i-1}$
. By Lemma 6.8, there exists an epimorphism
$\varphi \colon M' \rightarrow M_{i-1}$
with
$\mathrm{prk}\, M' \in [\lambda , \alpha ]$
and
$\varphi \in \mathrm{rad}_A$
. By Remark 6.2(ii), we can decompose
$M' = M \oplus N$
such that every indecomposable direct summand of M has projective rank in
$[\lambda ,\alpha ]$
and
$\mathrm{prk}\, N> \alpha $
. We define
$M_{i} = M$
and
$N_{i} = N_{i-1} \oplus N$
. The morphism
$\varphi \oplus 1_{N_{i-1}}$
induces the desired epimorphism
$\varphi _{i}$
(via
$M_i \oplus N_i = M' \oplus N_{i-1}$
).
The composition
$\varphi _1 \varphi _2 \dots \varphi _i$
is an epimorphism for all i. Further,
$\varphi _1 \varphi _2 \dots \varphi _i$
equals
$$ \begin{align*} \begin{pmatrix} \alpha_1 & \beta_1\\0& \gamma_1 \end{pmatrix} \begin{pmatrix} \alpha_2 & \beta_2\\0& \gamma_2 \end{pmatrix} \dots \begin{pmatrix} \alpha_i & \beta_i\\ 0& \gamma_i \end{pmatrix} = \begin{pmatrix} \alpha_1 \alpha_2 \dots \alpha_i & \beta\\0 & \gamma_1 \gamma_2 \dots \gamma_i \end{pmatrix} \end{align*} $$
for some
$\beta $
. If
$\alpha _1 \alpha _2 \dots \alpha _i = 0$
for some i, then
$\beta $
and
$\gamma _1\gamma _2 \dots \gamma _i$
would induce an epimorphism
${N_i \rightarrow M_0 \oplus N_0 = X}$
. By Remark 6.2(i), this yields
which is a contradiction. Hence,
$\alpha _1 \alpha _2 \dots \alpha _i \neq 0$
for all i. Let
$M_i = \bigoplus _{j=1}^{n_i} X_{ij}$
with
$X_{ij}$
indecomposable. Then, by construction,
$\mathrm{prk}\, X_{ij} \in [\lambda , \alpha ]$
and
$\alpha _i$
induces radical morphisms
$\alpha _{ij}^{j'} \colon X_{ij} \rightarrow X_{(i-1)j'}$
. We have
Thus, for all
there exist
$j_0, j_1, \dots , j_i$
with
$\alpha _{1 j_1}^{j_0} \alpha _{2 j_2}^{j_1} \dots \alpha _{i j_i}^{j_{i-1}} \neq 0$
. As in [Reference Bass8, p. 474], the König graph theorem now implies the existence of an infinite sequence
$j_0, j_1, \dots $
such that
${\alpha _{1 j_1}^{j_0} \alpha _{2 j_2}^{j_1} \dots \alpha _{i j_i}^{j_{i-1}} \neq 0}$
for all i.
Corollary 6.10. Let
$\lambda $
be a non-zero limit ordinal. For all
either no indecomposable modules have projective rank in
$[\lambda , \lambda +n]$
or the length of indecomposable modules having projective rank in
$[\lambda , \lambda +n]$
is unbounded.
Proof. Combine Proposition 6.9 and the Harada–Sai lemma [Reference Harada and Sai16, Lemma 1.2].
Example 6.11. Let k be an algebraically closed field,
the Kronecker quiver, and
$A = kQ$
. We can divide the indecomposable modules of
$\mathrm{mod}\,A$
into three parts: The preprojective modules
$\mathcal {P}$
, the regular modules
$\mathcal {R}$
, and the preinjective modules
$\mathcal {I}$
. The Auslander–Reiten quiver of
$\mathrm{mod}\,A$
can be visualized as follows:

By Proposition 6.5, it follows that the modules in
$\mathcal {P}$
are exactly those of finite projective rank. Every morphism from
$\mathcal {P}$
to
$\mathcal {R}$
is in
$\mathrm{rad}_A^\omega $
, where
$\omega $
denotes the smallest non-finite ordinal number. Further, every morphism from
$\mathcal {P}$
to
$\mathcal {I}$
is in
$(\mathrm{rad}_A^\omega )^2 = \mathrm{rad}_A^{\omega +1}$
. Since
$A\in \mathcal {P}$
, it follows that
$\mathcal {R}$
contains exactly the modules of projective rank
$\omega $
and
$\mathcal {I}$
those of projective rank
$\omega +1$
. By Corollary 6.10, we would expect the length of modules in
$\mathcal {R}$
to be unbounded, which is the case.
7. The torsion dimension and the Krull–Gabriel dimension
Through the concept of ideal torsion pairs, we introduce a new homological dimension for Artin algebras and relate it to the Krull–Gabriel dimension. The torsion-dimension of A, denoted by
$\mathrm{TD}(A)$
, is the m-dimension of the lattice of functorially finite ideal torsion pairs of A.
Proposition 7.1. The inequality
$\mathrm{TD}(A) \leq \mathrm{KG}(A)$
holds.
Proof. Let
$U\colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
be the forgetful functor, L the lattice of finitely presented subfunctors of
$f,$
and
$L'$
the lattice of finitely presented subfunctors of
. Note that, in the first case, we consider additive functors from
$\mathrm{mod}\,A$
to
$\mathrm{Ab}$
, while in the second case, we consider them to be from
$\mathrm{mod}\,A$
to
$\mathrm{mod}\,A$
. Clearly,
$L' \rightarrow L, F \mapsto U F$
defines an injective lattice homomorphism. Hence,
$\dim L' \leq \dim L$
. Since
$U \cong \mathrm{Hom}_A(A,-)$
, the equality
$\dim L = \mathrm{KG}(A)$
holds by Proposition 2.8. Further, we have
$\dim L' = \mathrm{TD}(A)$
by Corollary 3.9. It follows that
$\mathrm{TD}(A) \leq \mathrm{KG}(A)$
.
Remark 7.2. Suppose that A is commutative and let
$U\colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
be the forgetful functor. If t is a subfunctor of U, then
$tX$
is an abelian subgroup of X. Because multiplication with an element in A is a morphism in
$\mathrm{mod}\,A$
, it follows from the functoriality of t that
$tX$
is a submodule of X. Thus, subfunctors of
coincide with subfunctors of U. Hence,
$\mathrm{KG}(A) = \mathrm{TD}(A)$
in that case.
Another link between the torsion dimension and the Krull–Gabriel dimension of A is given by the radical ideal
$\mathrm{rad}_A$
of
$\mathrm{mod}\,A$
and its ordinal powers
$\mathrm{rad}_A^\alpha $
. First, we describe the relation between
$\mathrm{rad}_A^\alpha $
and the Krull–Gabriel dimension. Schröer conjectured that
$\mathrm{KG}(A) = n$
if and only if
$\mathrm{rad}_A^{\omega (n-1)} \neq 0$
and
$\mathrm{rad}_A^{\omega n} = 0$
for
, where
$\omega $
denotes the first non-finite ordinal [Reference Schröer, Krause and Ringel27]. He defined the Krull–Gabriel dimension in such a way that it only takes values in
. We state a version of his conjecture that also accounts for ordinal numbers. Note that there is no known example of an Artin algebra A with
$\omega \leq \mathrm{KG}(A) < \infty $
.
Conjecture 7.3. Let
$\alpha $
be a non-zero ordinal number.
-
(i) If
$\mathrm{KG}(A) \geq \alpha +1$
, then
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
. -
(ii) If
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
, then
$\mathrm{KG}(A) \geq \alpha +1$
.
Clearly, the above conjecture is equivalent to:
$\mathrm{KG}(A) = \alpha +1$
if and only if
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
and
$\mathrm{rad}_A^{\omega (\alpha +1)} = 0$
. This is exactly Schröer’s conjecture for
. Further, (ii) is proven for
$\alpha = 1$
[Reference Herzog17] and Krause showed that if
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
, then
$\mathrm{KG}(A) \geq \alpha $
[Reference Krause18, Corollary 8.14]. The whole conjecture is proven if A is a string algebra over an algebraically closed field k [Reference Laking, Prest and Puninski20, Corollary 1.3].
The next aim will be to connect the torsion dimension and the ordinal powers of the radical ideal. To do so, we give a different view on the lattice of finitely presented subfunctors of the forgetful functor
$U \colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
and the lattice of finitely presented subfunctors of
.
Remark 7.4. Let
$U\colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
be the forgetful functor. By the isomorphism
$U\cong \mathrm{Hom}_A(A,-)$
, we can identify subfunctors of U with subfunctors of
$\mathrm{Hom}_A(A,-)$
. A finitely presented subfunctor of
$\mathrm{Hom}_A(A,-)$
equals
$\mathrm{Im}\, \mathrm{Hom}_A(\varphi ,-)$
for some
$\varphi \colon A \rightarrow M$
. Further,
$\mathrm{Im}\, \mathrm{Hom}_A(\varphi ,-)$
is a subfunctor of
$\mathrm{Im}\, \mathrm{Hom}_A(\psi ,-)$
if and only if
$\varphi $
factors through
$\psi $
. Let L be the collection of equivalence classes of morphisms starting in A with the equivalence relation
$\varphi \sim \psi $
if
$\varphi $
factors through
$\psi $
and
$\psi $
factors through
$\varphi $
. Then L is partially ordered by
$\varphi \leq \psi $
if
$\varphi $
factors through
$\psi $
. By the above observations, we can identify the lattice of finitely presented subfunctors of U with L.
Lemma 7.5. Let
$U\colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
be the forgetful functor. The identification of finitely presented subfunctors of U and equivalence classes of morphisms starting in A restricts to an identification between finitely presented subfunctors of
and equivalence classes of morphisms
$\varphi \colon A\rightarrow M$
such that for all
$a\in A,$
there exists
$\alpha \colon M\rightarrow M$
with
$\varphi (a) = \alpha \varphi (1)$
.
Proof. The isomorphism
$\mathrm{Hom}_A(A,-) \cong U$
is given by
$\psi \mapsto \psi (1)$
for
$N\in \mathrm{mod}\,A$
and
$\psi \in \mathrm{Hom}_A(A,N)$
. Hence, for
$\varphi \colon A \rightarrow M,$
the functor
$\mathrm{Im}\, \mathrm{Hom}_A(\varphi ,-)$
becomes a subfunctor of
under the isomorphism if and only if for
$N\in \mathrm{mod}\,A,$
the abelian group
is a submodule of N. In particular, in that case, for all
$a\in A,$
there exists
$\alpha \colon M \rightarrow M$
with
$\varphi (a) = \alpha \varphi (1)$
. Now this property already suffices, since for arbitrary
$\beta \colon M \rightarrow N,$
we have
$a \beta \varphi (1) = \beta \varphi (a) = (\beta \alpha )\varphi (1)$
and so
$N'$
is a submodule of N.
Lemma 7.6. Let A be generated by
$a_1, a_2, \dots , a_n$
as a k-module. For a morphism
$\varphi \colon A \rightarrow M,$
let
$\overline {\varphi }\colon A \rightarrow M^n$
be defined by
$\overline {\varphi }(1) = (\varphi (a_1)\, \varphi (a_2)\,\dots \varphi (a_n))^\top $
. Then for all
$a\in A,$
there exists
${\alpha \colon M^n \rightarrow M^n}$
with
$\overline {\varphi }(a) = \alpha \overline {\varphi }(1)$
.
Proof. For
$a\in A,$
let
$a a_i = \sum _{i=1}^n c_{ij} a_j$
with
$c_{ij} \in k$
. Then the matrix
$(c_{ij})_{i,j}$
defines a morphism
$\alpha \colon M^n \rightarrow M^n$
such that the i’th component of
$\alpha \overline {\varphi }(1)$
equals
$$ \begin{align*} \sum_{j=1}^n c_{ij} \varphi(a_j) = \varphi\left(\sum_{j=1}^n c_{ij} a_j\right) = \varphi(a a_i), \end{align*} $$
which is exactly the i’th component of
$\overline {\varphi }(a)$
. Hence,
$\alpha \overline {\varphi }(1) = \overline {\varphi }(a)$
.
We are now ready to prove a connection between the torsion dimension of A and ordinal powers of the radical ideal of
$\mathrm{mod}\,A$
. It is similar to the result of Krause that
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
implies
$\mathrm{KG}(A)\geq \alpha $
.
Proposition 7.7. Let
$\alpha $
be a non-zero ordinal. If
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
, then
$\mathrm{TD} (A)\geq \alpha $
.
Proof. Let
$\psi \colon M \rightarrow N$
be a non-zero morphism in
$\mathrm{rad}_A^{\omega \alpha }$
and
$\varphi \colon A \rightarrow M$
with
$\psi \varphi \neq 0$
. Consider
and the morphisms
$\overline {\varphi }$
,
$\overline {\psi \varphi }$
as in Lemma 7.6. By construction
$\overline {\psi \varphi } = \psi ^n \overline {\varphi }$
. By Lemma 7.5, the morphisms correspond to finitely presented subfunctors
$t'\leq t$
of
. We show that the m-dimension of the interval
$[t',t]$
is greater than or equal to
$\alpha $
by transfinite induction.
For
$\alpha = 1,$
we have
$\psi \in \mathrm{rad}_A^\omega $
so for all
$m,$
there exists
$\psi _i \colon M_i \rightarrow M_{i+1}$
in
$\mathrm{rad}_A$
for
$i= 1,2,\dots ,m$
with
$\psi = \psi _m \dots \psi _1$
. For
$i=0,1,\dots ,m,$
consider the finitely presented subfunctor
$t_i$
of
corresponding to
$f_i = \overline {\psi _i\dots \psi _1 \varphi }$
. Then
$f_i = \psi _i^n f_{i-1}$
and thus
$t' = t_m \leq t_{m-1} \leq \dots \leq t_0 = t$
. Suppose
$t_i = t_{i-1}$
for some i. Then
$f_{i-1}$
would factor through
$f_i$
and there exists
$g\colon M_{i} \rightarrow M_{i-1}$
with
$f_{i-1}= g f_i = g\psi _i^n f_{i-1} = \alpha f_{i-1}$
for
$\alpha = g \psi _i^n \in \mathrm{rad}_A$
. Hence, there exists
$l>0$
with
$\alpha ^l = 0$
and
$f_{i-1} = \alpha ^l f_{i-1} = 0$
. But then
$\overline {\psi \varphi } = f_m = 0$
and thus
$\psi \varphi = 0$
, which is a contradiction. Hence,
$t' = t_m \lneq t_{m-1} \lneq \dots \lneq t_0 = t$
. Because m was arbitrary, it follows that the interval
$[t',t]$
has m-dimension greater than or equal to
$1$
.
Assume the claim holds for
$\alpha>0$
and let
$\psi \in \mathrm{rad}_A^{\omega (\alpha +1)} = \bigcap _{i=1}^\infty (\mathrm{rad}_A^{\omega \alpha })^i$
. Then for all
$m,$
there exists
$\psi _i \colon M_i \rightarrow M_{i+1}$
in
$\mathrm{rad}_A^{\omega \alpha }$
for
$i= 1,2,\dots ,m$
with
$\psi = \psi _m \dots \psi _1$
. For
$i=0,1,\dots ,m,$
consider the finitely presented subfunctor
$t_i$
of
corresponding to
$f_i = \overline {\psi _i\dots \psi _1 \varphi }$
. Then
$f_i = \psi _i^n f_{i-1}$
and thus
$t' = t_m \leq t_{m-1} \leq \dots \leq t_0 = t$
. By induction, the interval
$[t_i, t_{i+1}]$
has m-dimension greater than or equal to
$\alpha $
. Because m was arbitrary, it follows that the interval
$[t',t]$
has m-dimension greater than or equal to
$\alpha + 1$
.
Assume the claim holds for all
$\alpha < \lambda $
with
$\lambda $
a limit ordinal and let
$\psi \in \mathrm{rad}_A^\lambda $
. Then
$\psi \in \mathrm{rad}_A^\alpha $
for all
$\alpha < \lambda $
and by induction, the m-dimension of
$[t,t']$
is greater than or equal to
$\alpha $
for all
$\alpha < \lambda $
. Hence, it is also greater than or equal to
$\lambda $
.
Corollary 7.8. If Conjecture 7.3(i) is true, then either
$\mathrm{KG}(A) = \mathrm{TD}(A)$
or
$\mathrm{KG}(A) = \mathrm{TD}(A)+1$
.
Proof. If
$\mathrm{KG}(A) = \infty $
, then Conjecture 7.3(i) implies
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
for all ordinal numbers
$\alpha $
. Hence,
$\mathrm{TD}(A) \geq \alpha $
for all
$\alpha $
by Proposition 7.7. We conclude that
$\mathrm{TD}(A) = \infty $
. If
$\mathrm{KG}(A) = \alpha +1$
for some ordinal number
$\alpha $
, then
$\mathrm{rad}_A^{\omega \alpha } \neq 0$
by Conjecture 7.3(i). Hence,
$\mathrm{TD}(A) \geq \alpha $
. Further, Proposition 7.1 implies the inequality
$\mathrm{TD}(A) \leq \alpha +1$
. Thus, either
$\mathrm{TD}(A) = \alpha $
or
$\mathrm{TD}(A) = \alpha +1$
.
The following is a similar result for the torsion dimension as the result on the Krull–Gabriel dimension that
$\mathrm{KG}(A) \neq 1$
for all Artin algebras A. However, we need an additional assumption to conclude that
$\mathrm{TD}(A) \neq 1$
.
Proposition 7.9. If there exists
$M\in \mathrm{mod}\,A$
such that the smallest torsion class
$\mathcal {T}$
containing M is not functorially finite, then
$\mathrm{TD}(A)> 1$
.
Proof. Let
$\mathcal {C} = \mathrm{gen}\, M$
. By Proposition 5.6, it follows that
$\mathcal {C}$
is a functorially finite epi-closed class. Hence, by Lemmas 2.3 and 5.3, the ideal
$\langle \mathcal {C} \rangle $
is a functorially finite torsion ideal. Consider the ideal
$$ \begin{align*} \mathcal{I}_n = \langle \mathcal{C}\rangle\underbrace{\diamond \dots \diamond}_{n\text{-}\mathrm{times}} \langle \mathcal{C} \rangle \end{align*} $$
for
. Then
$\mathcal {I}_n$
is a functorially finite torsion ideal by Corollary 3.13. Further,
$\mathcal {I}_n = \langle \mathcal {C}_n \rangle $
for
by Lemma 3.14. The functorially finite torsion ideal
$\mathcal {I}_n$
corresponds to a finitely presented subfunctor
$t_n$
of
by Corollary 3.9. Since
$\mathcal {I}_{i} \subseteq \mathcal {I}_{i+1}$
, this yields a chain
$t_1 \leq t_2 \leq \dots $
of finitely presented subfunctors of
. We show that
$t_n X \neq t_{n+1} X$
for infinitely many isomorphism types of indecomposable modules
$X\in \mathrm{mod}\,A$
. Suppose the contrary holds. Then there exists
$N>n$
such that
$t_n X = t_{n+1} X$
for all indecomposable modules X in
$\mathcal {C}_{N+1}\backslash \mathcal {C}_N$
. It follows that
$t_{n+1} X = t_n X \in \mathcal {C}_n$
by Proposition 5.5. Further, since
$X\in \mathcal {C}_{N+1}$
, there exists a submodule
$X'\in \mathcal {C}_{n+1}$
of X such that
$X/X' \in \mathcal {C}_{N-n}$
. The canonical inclusion
$X' \rightarrow X$
factors through the left
$\langle \mathcal {C}_{n+1} \rangle $
-approximation
$t_{n+1} X \rightarrow X$
, so
$X' \subseteq t_{n+1} X$
. Because
$C_{N-n}$
is epi-closed, it follows that
$X/t_{n+1} X \in \mathcal {C}_{N-n}$
. Now
$t_{n+1} X \in \mathcal {C}_n$
and
$X/t_{n+1} X \in \mathcal {C}_{N-n}$
imply
$X\in \mathcal {C}_{N}$
. Hence,
$\mathcal {C}_N = \mathcal {C}_{N+1}$
. Thus,
$\mathcal {C}_N$
is closed under extensions. By Remark 5.4, it follows that
$\mathcal {T} = \mathcal {C}_N$
, so
$\mathcal {T}$
is functorially finite, which is a contradiction. Hence, the functors
$t_n$
and
$t_{n+1}$
always disagree on infinitely many isomorphism types of indecomposable modules. Next, we refine the chain of finitely presented functors
Let
$U\colon \mathrm{mod}\,A \rightarrow \mathrm{Ab}$
be the forgetful functor. Then
$U t_n$
is finitely presented, so there exists a maximal subfunctor
$r_{n,1}$
of
$U t_n$
containing
$U t_{n-1}$
. Now the quotient
$U t_n/r_{n,1}$
is simple. Thus, the functors
$U t_n$
and
$r_{n,1}$
agree on all but one isomorphism type of an indecomposable module Y [Reference Auslander5]. Notice that
$r_{n,1} Y$
may not be a submodule of Y. To fix this, we define
$t_{n,1}$
by
$t_{n,1} X = r_{n,1} X = t_n X$
for
$X\not \cong Y$
indecomposable and
where we sum over all morphisms
$\varphi \colon X\rightarrow Y$
with
$X\not \cong Y$
indecomposable. By construction,
$t_{n-1} \subseteq t_{n,1} \subseteq r_{n,1}$
and
$t_{n,1}$
is functorial. Further,
$t_{n,1} X$
is a submodule of X for all indecomposable modules X and
$t_{n,1}, t_n$
agree on all but one isomorphism type of an indecomposable module. It follows that
$U t_n/U t_{n,1}$
is finitely presented and thus
$U t_{n,1}$
is finitely presented. Further, the inequalities
${t_{n-1} < t_{n,1} < t_n}$
hold. Since
$t_{n-1}$
and
$t_n$
disagree on infinitely many different isomorphism types of indecomposable modules, we can proceed inductively to obtain a chain of finitely presented functors
It follows that the m-dimension of the interval
$[t_{n-1},t_n]$
equals at least
$1$
for all n. Thus, the m-dimension of the interval
equals at least
$2$
. We conclude that
$\mathrm{TD}(A)> 1$
.
We are now ready to show that the torsion dimension coincides with the Krull–Gabriel dimension of a hereditary Artin algebra. For the details of the representation theory of hereditary Artin algebras, see [10] and [24].
Theorem 7.10. Let A be a hereditary Artin algebra. Then
$\mathrm{TD}(A) = \mathrm{KG}(A)$
.
Proof. We have
$\mathrm{TD}(A) \leq \mathrm{KG}(A) $
by Proposition 7.1. Thus, if A is of finite representation type, then
$\mathrm{TD}(A) = \mathrm{KG}(A) = 0$
.
If A is of tame representation type, then
$\mathrm{TD}(A) \leq \mathrm{KG}(A) = 2$
. Let M be the direct sum of all simple regular modules (up to isomorphism) of a tube of finite rank. Then the smallest torsion class containing M is not functorially finite by [Reference Ringel26, Proposition 4]. Hence,
$\mathrm{TD}(A)> 1$
by Proposition 7.9 and thus
$\mathrm{TD}(A) = 2.$
If A is of wild representation type, then
$\mathrm{rad}_A^\infty \neq 0$
by [Reference Krause18, Proposition 8.15]. Hence,
$\mathrm{TD}(A) = \infty $
by Proposition 7.7 and thus
$\mathrm{TD}(A) = \infty = \mathrm{KG}(A)$
.
In the following example, we show an application of the monoidal structure on pairs
$s\leq t$
of subfunctor of
(see Section 2) to the torsion dimension and the Krull Gabriel dimension.
Example 7.11. Let k be an algebraically closed field,
the Kronecker quiver, and
$A = kQ$
(as in Example 6.11). Then A is a tame hereditary Artin algebra. Hence,
$\mathrm{TD}(A) = \mathrm{KG}(A) = 2$
by Theorem 7.10. In this example, we show that two finitely presented subfunctors of
are sufficient to produce enough finitely presented subfunctors of
by the monoidal structure to deduce
$\mathrm{TD}(A) \geq 2$
.
Let
$S \in \mathrm{mod}\,A$
be simple injective and
$T\in \mathrm{mod}\,A$
simple projective. We can divide the indecomposable modules of
$\mathrm{mod}\,A$
into three parts: the preprojective modules
$\mathcal {P}$
, the regular modules
$\mathcal {R,}$
and the preinjective modules
$\mathcal {I}$
. Further, the regular modules
$\mathcal {R}$
can be divided into tubes
${\mathcal {R}^\lambda = \{R^\lambda _1, R^\lambda _2, \dots \}}$
for
$\lambda \in k\cup \{\infty \}$
. For all
$\lambda ,$
the full subcategory of modules isomorphic to a direct sum of modules in
$\mathcal {R}^\lambda $
is abelian and has one simple object
$R_1^\lambda $
. Further, it is uniserial, so the lattice of subobjects of
$R_j^\lambda $
is linearly ordered for all j. The linear order is determined by the short exact sequences
Consider
$\mathcal {C} = \mathrm{gen}\, R_1^\lambda $
and
$\mathcal {D} = \mathrm{cogen}\, R_1^\lambda $
. The indecomposable modules in
$\mathcal {C}$
equal
$\{R_1^\lambda , S\}$
and the ones in
$\mathcal {D}$
equal
$\{R_1^\lambda , T\}$
(up to isomorphism). Further,
$\mathcal {C}$
is a functorially finite epi-closed class and
$\mathcal {D}$
a functorially finite mono-closed class by Proposition 5.6. Let
$\mathcal {I} = \langle \mathcal {C} \rangle $
and
$\mathcal {J} = \langle \mathcal {C} \rangle $
. Then, by Lemmas 2.3 and 5.3, the ideal
$\mathcal {I}$
is a functorially finite torsion ideal and
$\mathcal {J}$
a functorially finite torsion-free ideal. Now
$\mathcal {I}$
corresponds to a finitely presented subfunctor t of
and
$\mathcal {J}$
to a finitely presented subfunctor r of
by Corollary 3.9. For
$M\in \mathrm{mod}\,A,$
the module
$t M$
is the largest submodule of M in
$\mathcal {C}$
and
$M/r M$
is the largest factor module of M in
$\mathcal {D}$
by Proposition 5.5. By the above short exact sequences, we conclude that
Next, we employ the monoidal structure on pairs of subfunctors of
. Let
$t_i$
be defined by
. Then
$t_1 \leq t_2 \leq \dots $
are finitely presented by Lemma 3.11. By the values of t, it follows that
$t_i R_k^\lambda = R_k^\lambda $
if
$k < i$
and otherwise
$t_i R_k^\lambda = R_i^\lambda $
. Further, let
$r_{i,j}$
be defined by
$r^j (t_{i+j+1}/t_i) = r_{i,j}/t_i$
. Then
$r_{i,j}$
is finitely presented by Lemma 3.11. By the values of
$t_i$
and r, it follows that
$$\begin{align*} k < i \colon \quad &r_{i,j} R_k^\lambda = R_k^\lambda = t_i R_k^\lambda,\\i \leq k \leq i+j \colon \quad &r_{i,j} R_k^\lambda = R_{j}^\lambda = t_i R_k^\lambda, \\i+j < k \colon \quad &r_{i,j} R_k^\lambda = R_{i+1}^\lambda = t_{i+1} R_k^\lambda. \end{align*}$$
Hence, if we restrict
$r_{i,j}$
on
$\mathcal {R}^\lambda $
, then
and thus
It follows that the m-dimension of the interval
$[t_i, t_{i+1}]$
equals
$1$
for all i. We conclude that the m-dimension of
equals
$2$
and so
$\mathrm{TD} (A) \geq 2$
.
Acknowledgements
I would like to express my gratitude to Frederik Marks for weekly discussions of the mathematical content in this work. I am also thankful to him for guidance while writing this article.