
TL;DR
This paper introduces residual finiteness for categories, demonstrating that certain categories like free categories and subcategories of vector spaces possess this property, which implies several algebraic and computational advantages.
Contribution
It extends the concept of residual finiteness from groups to categories and establishes key properties and examples of residually finite categories.
Findings
Free categories are residually finite.
Finitely generated subcategories of vector spaces are residually finite.
Residually finite categories are Hopfian and have solvable word problem.
Abstract
We introduce the notion of residual finiteness for categories. In analogy with the group-theoretic setting, we prove that free categories and finitely generated subcategories of finite-dimensional vector spaces are residually finite. Moreover, finitely generated residually finite categories are Hopfian and finitely presented residually finite categories have solvable word problem.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsLogic, programming, and type systems · Homotopy and Cohomology in Algebraic Topology · semigroups and automata theory
Residually finite categories
Clara Löh
(Date: . © C. Löh 2019. This work was supported by the CRC 1085 Higher Invariants (Universität Regensburg, funded by the DFG).)
Abstract.
We introduce the notion of residual finiteness for categories. In analogy with the group-theoretic setting, we prove that free categories and finitely generated subcategories of finite-dimensional vector spaces are residually finite. Moreover, finitely generated residually finite categories are Hopfian and finitely presented residually finite categories have solvable word problem.
Key words and phrases:
residually finite category, residually finite group
2010 Mathematics Subject Classification:
20E18, 20E26, 18A99
1. Introduction
Classical mathematics is often concerned with infinite, potentially huge, structures. From a more computational point of view, it is therefore essential to ask which properties can be tested through transformations to finite structures. For example, in group theory, this is captured by the notion of residual finiteness [6]: A group is residually finite, if equality of group elements can be tested via group homomorphisms to finite groups [1, Definition 2.1.1, Proposition 2.1.2].
Definition 1.1** (residually finite group).**
A group is residually finite, if for all , with , there exists a finite group and a group homomorphism with
[TABLE]
In this note, in analogy with the group-theoretic setting, we introduce the following notion of residual finiteness for categories, based on testing via functors to finite categories:
Definition 1.2** (residually finite category).**
A category is residually finite, if for all morphisms and in with , there exists a finite category and a functor with
[TABLE]
Here, a category is finite if is finite and for all also is finite. In Section 2.2, we will say more about the notions of equality of morphisms and finiteness in categories.
For groups (and the canonical interpretations of groups as categories with a single object), the notions of residual finiteness from Definition 1.1 and Definition 1.2 coincide (Proposition 4.1). Moreover, Definition 1.2 also subsumes a notion of residual finiteness for monoids (via categories that only contain a single object) [2] and groupoids (categories all of whose morphisms are isomorphisms).
(Non-)Examples
Important examples of residually finite groups are free groups [1, Theorem 2.3.1] and finitely generated linear groups [7]. Analogously, we show that the following categories are residually finite:
- •
free categories (Proposition 4.7)
- •
finitely generated subcategories of the category of finite-dimensional vector spaces over a field (Corollary 4.18)
Groups are residually finite if and only if they embed into a product of finite groups [1, Corollary 2.2.6]. Similarly, we prove that a small category is residually finite if and only if it is equivalent to a subcategory of a product of finite categories (Corollary 3.7).
In contrast, the following categories are not residually finite (and thus are not amenable to systematic testing via functors to finite categories):
- •
the category of finite sets (Proposition 4.8)
- •
the category of finite-dimensional vector spaces over a field (Proposition 4.11)
- •
the simplex category (Proposition 4.9)
Using residual finiteness
Two classical applications of residual finiteness in group theory are:
- •
All finitely generated residually finite groups are Hopfian [1, Theorem 2.4.3] (i.e., every surjective endomorphism is already an isomorphism); this shows, for instance, that every self-map of an aspherical oriented closed connected manifold with residually finite group induces an isomorphism on fundamental groups and thus is a homotopy equivalence.
Similarly, this holds also in other algebraic situations, e.g., for rings and modules [4, 12].
- •
All finitely presented residually finite groups have solvable word problem [8].
We establish the corresponding versions for residually finite categories: All finitely generated residually finite categories are Hopfian (Theorem 5.1). All finitely presented residually finite categories have solvable word problem (Theorem 5.2); this might be of interest when modelling calculi, deductional systems, or rewriting systems in categories [3].
Organisation of this article
In Section 2, we clarify our setup of category theory. Section 3 contains basic inheritance properties of residual finiteness of categories. Examples of residually finite categories are discussed in Section 4. Finally, in Section 5, we show that finitely generated residually finite categories are Hopfian and that finitely presented residually finite categories have solvable word problem.
2. Setup
2.1. Categories
For simplicity and concreteness, we will use classical class-set theory (such as NBG [11]) as ambient theory for category theory; all categories will be locally small (i.e., while the objects of a category form a class, the morphisms between any two objects form a set). A category is small if the class of objects is a set. Reference to the axiom of choice will be made explicit. Most of this note can be adapted in a straightforward manner to more synthetic settings or settings with more stages of “sizes” of sets.
2.2. Equality and finiteness
Equality of morphisms and finiteness of categories play a central role in the definition of residual finitenes (Definition 1.2). As equality of objects in categories is a delicate subject, we briefly comment on two popular choices:
- •
Using equality of objects: Let and let , . Then the morphisms and in are considered equal if and only if and and (in the set ). In particular, morphisms are supposed to know (at least implicitly) about their domain and target objects.
In this setting, we can use the naive notion of finiteness of categories: A category is finite if is finite and if for all , the set is finite.
- •
Without using equality of objects: If we want to avoid to speak of equality of objects (in order to obtain equivalence-robust notions), we will only define (in)equality for morphisms in the same morphism set. In this version, it does not make sense to talk about (in)equality of morphisms that have different domains/targets.
Moreover, in this setting, a category should be considered to be finite if it is weakly finite in the following sense: A category is weakly finite if contains only finitely many isomorphism classes of objects and if for all , the set is finite.
We will adopt the first, naive, semantics (using equality of objects); in particular, we also will talk about finite generation of categories, etc. in the naive sense. We can then compare the definition using the first semantics and the second semantics. In this case, both interpretations result in the same notion of residual finiteness:
Proposition 2.1**.**
Let be a category, let , , , , and let , with or . Then there exists a finite category and a functor with .
Proof.
We consider the complete directed graph on the set (with at most four elements) and its associated category , which is finite (Section 2.3). We then define the following functor :
- •
on objects: For , we set
[TABLE]
- •
on morphisms: For and , we define as the unique morphism in from to .
By construction, (because the target or the domain objects in are not equal). ∎
In particular, this also shows that different objects can always be separated by functors to finite categories.
Proposition 2.2**.**
Let be a category, let and be morphisms in with , let be a weakly finite category, and let be a functor with . Then there exists a finite category and a functor with .
Proof.
Every weakly finite category is equivalent to a finite category (one can use the axiom of choice or use a more constructive notion of weak finiteness that includes such an equivalence). Hence, there is a finite category and a faithful functor . We can then take . ∎
2.3. Graphs and quivers
A directed graph is a pair consisting of a set (the vertices) and a set (the edges). If is a directed graph, then the category associated with consists of
- •
objects: We set .
- •
morphisms: If , are objects in , then we set
[TABLE]
- •
composition of morphisms: The composition of morphism is uniquely determined by the definition of the morphism sets and the fact that concatenation of directed paths in witnesses that the composition of composable morphisms exists. If , then the unique element of is the identity morphism of .
More generally, a quiver is a quadruple consisting of a set (the vertices), a set (the edges), and two maps (the source and target map, respectively).
3. Basic properties
In order to work efficiently with residually finite categories, we first establish some basic inheritance results.
3.1. Isomorphisms, equivalences, subcategories
Proposition 3.1**.**
Let , be isomorphic categories. If is residually finite, then also is residually finite.
Proof.
Composing separating functors with an isomorphism proves the claim. ∎
More generally, the same argument shows that residual finiteness is inherited under equivalences of categories:
Proposition 3.2**.**
Let and be equivalent categories. If is residually finite, then also is residually finite.
Proof.
Let be an equivalence of categories and let and be morphisms in with . As an equivalence of categories, is faithful; hence, in . Because is residually finite, there exists a finite category and a functor with . Thus, the functor separates and . ∎
Corollary 3.3**.**
Let be a category and let be a skeleton of . Then is residually finite if and only if is residually finite.
Proof.
As a skeleton of , the category is equivalent to (depending on the setting, we can either use the axiom of choice or a constructive notion of skeleton that requires the existence of such an equivalence). We then only need to use the fact that residual finiteness is inherited under equivalences of categories (Proposition 3.2). ∎
Proposition 3.4**.**
Subcategories of residually finite categories are residually finite.
Proof.
This is immediate from the definition (we only need to restrict the corresponding separating functors). ∎
As in the case of groups, in a residually finite category, we can separate any finite number of morphisms:
Proposition 3.5**.**
Let be a residually finite category, let , and let be different morphisms in . Then there exists a finite category and a functor such that the morphisms are all different.
Proof.
Because is residually finite, for all with , there exists a finite category and a functor with
[TABLE]
Then also the product category
[TABLE]
is finite and the product functor has the desired property. ∎
3.2. Products
Proposition 3.6**.**
- (1)
If and are residually finite categories, then also is a residually finite category. 2. (2)
If is a set and is a family of residually finite small categories, then also is residually finite.
Proof.
We only prove the second part (the first part can be proved in the same way). Let and be morphisms in with . By definition of , there exist families and , where and are morphisms in with and . As , there is an with . Because is residually finite, there exists a finite category and a functor with . Let denote the projection functor. Then the composition satisfies
[TABLE]
as desired. ∎
Corollary 3.7**.**
Let be a small category. Then the following are equivalent:
- (1)
*The category is residually finite. * 2. (2)
The category is equivalent to a subcategory of a product (over a set) of finite categories.
Proof.
Ad . Let be residually finite. We consider the index set
[TABLE]
( is small, so this is a set). Because is residually finite, for each , there exists a finite category and a functor with
[TABLE]
Then, the family defines a functor , which is faithful (by construction). More precisely, the existence of such a functor is guaranteed by the axiom of choice.
Hence, is equivalent to a subcategory (namely the image category of ) of the product of finite categories.
Ad . Products of finite categories are residually finite (Proposition 3.6), subcategories of residualy finite categories are residually finite (Proposition 3.4), and residual finiteness is preserved under equivalences (Proposition 3.2). ∎
4. Basic examples
4.1. Groups and groupoids
If is a group, then we can consider the associated category , which consists of a single object and whose morphisms are defined by (with the composition given by the composition in ). For groups, the residual finiteness notions in Definition 1.1 and Definition 1.2 coincide:
Proposition 4.1**.**
Let be a group. Then is residually finite if and only if the associated category is residually finite.
Proof.
Let be residually finite and let , be morphisms in with ; in particular, . Because is residually finite, there is a finite group and a group homomorphism with . The homomorphism induces a functor mapping the only object of to the one of and using on the morphisms:
[TABLE]
As is finite, also the category is finite. Moreover, by construction,
[TABLE]
Hence, the category is residually finite.
Conversely, let the category be residually finite and let with . Because is residually finite and , there exists a finite category and a functor with . We then consider the (finite) group
[TABLE]
where ; the functor induces a group homomorphism
[TABLE]
Because the category is finite, also the group is finite. Moreover, by construction . Hence, the group is residually finite. ∎
Example 4.2** (finitary symmetric group).**
Let be the group of permutations of with finite support. Then is not residually finite (the subgroup of even permutations in is simple and infinite). A similar consideration will show that the category of finite sets is not residually finite (Proposition 4.8).
Corollary 4.3**.**
Let be a category and let . If is residually finite, then is a residually finite group.
Proof.
The subcategory of consisting of the object and the -automorphisms of is isomorphic to . If is residually finite, then also this subcategory is residually finite (Proposition 3.4); thus, also is residually finite (Proposition 3.1). Therefore, the group is residually finie (Proposition 4.1). ∎
In general, the converse of Corollary 4.3 does not hold (Proposition 4.8, Proposition 4.9). However, for groupoids (i.e., small categories all of whose morphisms are isomorphisms), we obtain:
Corollary 4.4**.**
Let be a groupoid.
- (1)
If is connected and , then is residually finite if and only if the group is residually finite. 2. (2)
The following are equivalent:
- (a)
The category is residually finite. 2. (b)
For each , the group is residually finite.
Proof.
For the first part, let be the full subcategory of generated by . Because is a groupoid, is a skeleton of . Moreover, is isomorphic to the category . Applying Corollary 3.3 and Proposition 4.1 finishes the proof of the first part.
For the second part, Corollary 4.3 proves the implication (a) (b). For the converse implication, we can argue as follows: Choosing (via the axiom of choice) one object in each connected component of leads to a skeleton of . Assuming (b), this skeleton is easily seen to be residually finite (Proposition 4.1 and collapsing all but one components to the one-object category ). Hence, applying Proposition 3.3 and Proposition 4.1 shows that is residually finite as well. ∎
4.2. Graphs, posets, and free categories
Proposition 4.5**.**
Let be a directed graph. Then the associated category (Section 2.3) is residually finite.
Proof.
Let be the directed graph
[TABLE]
obtained from by adding for each edge also the inverse edge. Then the category is a subcategory of , which is a groupoid. Moreover, for each vertex of , the automorphism group is trivial, whence residually finite. Therefore, is residually finite (Corollary 4.4) and so also is residually finite (Proposition 3.4).
Of course, alternatively, we can also invoke the more general statement on free categories (Proposition 4.7). ∎
Corollary 4.6**.**
Let be a poset. Then the poset category of is residually finite.
Proof.
If is a poset, then the poset category of is the same as the category associated to the directed graph
[TABLE]
Hence, by Proposition 4.5, the poset category of is residually finite. ∎
Free groups are residually finite [1, Theorem 2.3.1]; we will now establish the corresponding result for categories.
Proposition 4.7**.**
Let be a quiver. Then the free category , freely generated by , is residually finite.
Proof.
Let and be morphisms in with . We can view and as finite (directed) paths in ; because of , they differ in at least one edge.
Let be the quiver obtained from by identifying all vertices to a single vertex (and keeping distinct edges distinct) and let be the corresponding quiver projection. Because is the free category, freely generated by , there exists a functor that induces the quiver morphism on the underlying quivers. Because and are different, also the associated paths in are different.
Because is a quiver with a single vertex, the category is the category associated with the free monoid, freely generated by the edges of . Hence, we can view as subcategory of the category associated with the free group , freely generated by the edges of . Because is residually finite [1, Theorem 2.3.1], also the category is residually finite (Proposition 4.1). Hence, there exists a finite category and a functor with . Then the composition
[TABLE]
separates and . ∎
4.3. Sets and simplices
Proposition 4.8**.**
The category of finite sets is not residually finite.
Proof.
We consider
[TABLE]
(but in fact any two different maps would work). Let be a finite category and let be a functor. We will now show that , using a detour via bigger sets.
If , then the alternating group is a subset of . Because is finite, there exists an with \bigl{|}F(A_{N})\bigr{|}<|A_{N}|. Because restricted to is a group homomorphism and because is simple, it follows that
[TABLE]
for all . Let be the inclusion and let be the projection that sends all to . Then
[TABLE]
Because defines an element of , we obtain
[TABLE]
Therefore, and cannot be separated by a functor to a finite category. Hence, is not residually finite.
Alternatively, one could also argue via the simplex category as subcategory of (Proposition 4.9). ∎
It should be noted that each object in has a (residually) finite automorphism group. Hence, in the case of the non-residual finiteness originates in the overal interaction of “small” objects with “big” objects.
More drastically, the simplex category is not residually finite (even though all objects have trivial automorphism group). We will use the following version of : Objects are all sets of the form with and morphisms are all monotonically increasing functions.
Proposition 4.9**.**
The simplex category is not residually finite.
Proof.
We consider
[TABLE]
Let be a finite category and let be a functor. We will now show that , using a detour via bigger sets.
Because is finite, there exists an such that holds for all . For each , we set
[TABLE]
which is a morphism in . By the choice of , there exist with
[TABLE]
Moreover, we look at the following morphisms in :
[TABLE]
Then
[TABLE]
and we conclude
[TABLE]
Therefore, and cannot be separated by a functor to a finite category, which shows that is not residually finite. ∎
4.4. Module categories
A key example of residually finite groups are finitely generated linear groups [7, 9]. It is therefore natural to wonder about the residual finiteness of module categories. In the following, a ring with unit is a not necessarily commutative ring that has a multiplicative unit with .
Definition 4.10** (module categories).**
Let be a ring with unit and let . Then we introduce the following categories:
- •
: The category of all finitely generated free left -modules and -linear maps.
- •
: The category of all finitely generated free left -modules freely generated by at most elements and -linear maps.
Proposition 4.11**.**
Let be a ring with unit. Then the category is not residually finite.
Proof.
The category contains a subcategory that is isomorphic to the category of finite sets (e.g., taking the objects with and all -linear maps given by right multiplication by matrices that only have entries in and all of whose rows contain exactly one ). Because is not residually finite (Proposition 4.8), we obtain that is not residually finite (Proposition 3.4). ∎
However, we will see that adding appropriate finiteness conditions on the category, will imply residual finiteness (Corollary 4.18).
Proposition 4.12**.**
Let be a ring with unit whose underlying additive Abelian group is not residually finite. Then the category is not residually finite.
Proof.
In view of Corollary 4.3, we only need to show that the automorphism group is not residually finite. Because is isomorphic to the group , it suffices to show that is not residually finite. As
[TABLE]
is an injective group homomorphism from the additive group to and because the additive Abelian group is not residually finite, also is not residually finite. ∎
Example 4.13**.**
The category is not residually finite because the additive group is not residually finite [1, Example 2.1.9].
Proposition 4.14**.**
Let be a residually finite ring with unit and let . Then the category is residually finite.
Proof.
The full subcategory of generated by is a skeleton of . In view of Corollary 3.3, we can therefore restrict attention to morphisms between these modules.
Let and be morphisms in with . In view of Proposition 2.1, we may assume that and have the same domain and the same target .
Let and be the matrices representing and , respectively, with respect to the standard bases (via right multiplication). Because , there exist , with . Because the ring is residually finite, there exists a finite ring and a ring epimorphism such that . In particular,
[TABLE]
in (where the tensor product is taken with respect to ).
Because the ring is finite, also the category is finite and the reduction functor separates and . ∎
Corollary 4.15**.**
Let be a residually finite ring with unit and let be a finitely generated subcategory of . Then is residually finite.
Proof.
Because is finitely generated, it contains only finitely many objects of . Hence, there exists an such that is a subcategory of . We can now apply the previous Proposition 4.14 and Proposition 3.4. ∎
Example 4.16**.**
For each , the category is residually finite: The family of all reductions modulo prime numbers shows that is a residually finite ring. We can therefore apply Proposition 4.14.
Corollary 4.17**.**
Let be a finitely generated commutative ring with unit and let be a finitely generated subcategory of . Then is residually finite.
Proof.
Every finitely generated commutative ring is residually finite [10] and thus we can apply Corollary 4.15. ∎
Finally, we obtain the category-version of Malcev’s theorem on linear groups:
Corollary 4.18**.**
Let be a field and let be a finitely generated subcategory of . Then is residually finite.
Proof.
Because is finitely generated, there exists a finitely generated commutative ring such that all morphisms in can be represented by matrices over . Therefore, we can view as a finitely generated subcategory of . Therefore, applying Corollary 4.17 proves the claim. ∎
5. Using residual finiteness
We will now show how residual finiteness of categories can exploited in the presence of finite generation/finite presentation.
5.1. Hopficity
Every finitely generated residually finite category is Hopfian in the following sense:
Theorem 5.1**.**
Let be a finitely generated residually finite category and let be a full functor that is essentially surjective (i.e., for each there exists a with ). Then is faithful. In particular, in the presence of the axiom of choice, is an equivalence.
Proof.
We proceed as in the case of the corresponding result for groups: Let and be morphisms in with ; in view of Proposition 2.1, we assume that there are with . Because is residually finite, there exists a finite category and a functor with .
As is finitely generated and is a finite category, there exist only finitely many different functors . Hence, there are with and
[TABLE]
Inductively, we find sequences , in , sequences , of morphisms in , and sequences , of isomorphisms in satisfying
[TABLE]
and
[TABLE]
for all . We now set
[TABLE]
and show the following:
- (1)
There exist isomorphisms and in with
[TABLE] 2. (2)
We have . 3. (3)
We have . 4. (4)
We have . 5. (5)
We have (which proves that is faithful).
Ad (1). We can take
[TABLE]
which clearly are -isomorphisms with the claimed property.
Ad (2). This follows from (1) and the fact that .
Ad (3). By construction,
[TABLE]
and, analogously,
[TABLE]
Therefore, we can use (2) to prove (3).
Ad (4). This is an immediate consequence of (3).
Ad (5). By (1), we have
[TABLE]
Moreover, (4) shows that . Because and are isomorphisms in , we conclude that . ∎
5.2. Solving the word problem
Theorem 5.2**.**
Let be a finite presentation of a residually finite category. Then the word problem is solvable for (via an explicit algorithm, specified in the proof).
Let us first recall the corresponding notions: As in the case of groups, presentations of categories are defined via quotient categories of free categories [5, Chapter II.8].
Definition 5.3** (finite presentation).**
A finite presentation of a category is a pair consisting of
- •
a finite quiver ,
- •
a finite set of finite (directed) paths in .
The category presented by such a finite presentation is the quotient category of the free category , freely generated by , modulo the smallest congruence relation on morphisms of containing .
Definition 5.4** (solvability of the word problem).**
Let be a finite presentation of a category. Then the word problem for is solvable if the following holds: There exists an algorithm that given as input two finite (directed) paths in (specified as finite lists of directed edges) decides whether the morphisms in the category represented by these paths are equal or not.
Proof of Theorem 5.2.
Again, we proceed as in the corresponding result for groups. Let , let and be finite directed paths in , and let and be the morphisms in represented by and , respectively.
We simultaneously perform the following tasks (e.g., by interleaving):
- •
We enumerate the congruence relation on the morphisms of the free category generated by and check whether belongs to (these initials of) . If , then the answer is yes (i.e., ).
- •
We diagonally enumerate all natural numbers and all categories with object set and morphism set . For each such finite category , we construct the finite set of functors (using the universal property of ) and its composition with the canonical projection functor . For each such functor , we check whether . If , then the answer is no (i.e., ).
We briefly explain why this algorithm is correct and terminates after a finite number of steps. To this end, we distinguish the following cases:
- •
If , then ; hence, will be found after a finite number of enumeration steps of .
Moreover, because , the morphisms and cannot be separated by a functor in the second branch of the algorithm. Hence, the algorithm correctly terminates with yes.
- •
If , then we can invoke residual finiteness of : There exists a finite category and a functor with . Composing with the canonical projection functor, leads to a functor such that
[TABLE]
Because every finite category is isomorphic to one of the categories enumerated in the second branch of the algorithm, such a separating functor will be found in a finite number of steps.
Moreover, because , we have and thus the algorithm will not stop in the first branch. Hence, the algorithm correctly terminates with no. ∎
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] T. Ceccherini-Silberstein, M. Coornaert. Cellular Automata and Groups , Springer Monographs in Mathematics, Springer, 2010.
- 2[2] È. A. Golubov. Finite separability in semigroups, Dokl. Akad. Nauk SSSR , 189, 20–22, 1969.
- 3[3] J. Lambek, P. J. Scott. Introduction to higher order categorical logic , Cambridge Studies in Advanced Mathematics, 7, Cambridge University Press, 1986.
- 4[4] J. Lewin. Subrings of finite index in finitely generated rings, J. Algebra , 5, 84–88, 1967.
- 5[5] S. Mac Lane. Categories for the Working Mathematician , Graduate Texts in Mathematics, 5, second edition, Springer, 1998.
- 6[6] W. Magnus. Residually finite groups. Bull. Amer. Math. Soc. , 75, 305–316, 1969.
- 7[7] A. Malcev. On isomorphic matrix representations of infinite groups, Rec. Math. [Mat. Sbornik] N.S. , 8(50), 405–422, 1940.
- 8[8] A. W. Mostowski. On the decidability of some problems in special classes of groups, Fund. Math. , 59, 123–135, 1966.
