Universal Categories

Hans Schoutens · 2025

Read Universal Categories by Hans Schoutens (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-18896.txt.

Introduction

Hans Schoutens works in model theory and category theory, where a theory's models form a category of their own. He points out a surprising property: any small category elementarily equivalent to that model category already embeds inside it. The known proof runs through ultrapowers and stays abstract, leaving open which concrete axioms, depending on the theory and language, force the embedding. His paper, posted to arXiv in December 2025, also proposes a first-order logic in which equivalent categories are always elementarily equivalent, tightening the two-way street between the notions.

Catalog record

CREATORHans Schoutens
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-18896

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Universal Categories Hans Schoutens Abstract . The category of models of any theory T T in any first-order language L L has the surprising property that any small category that is elementarily equivalent with it, already embeds in it. The proof uses an abstract argument via ultrapowers, leaving one wonder which concrete categorical axioms, depending on T T and L L , are responsible for this embedding result. We also propose a first-order logic for which equivalent categories are always elementarily equivalent. 1. Introduction The key notion of this paper is that of universality: let us call an L -structure universal , if any elementarily equivalent L -structure of lesser cardinality embeds in it. Uncountably categorical structures are examples of universal structures, but beyond that, there seem to be no immediate examples. For instance, the real field is not universal, for their exist countable non-Archimedean real closed fields, which therefore are elementary equivalent with the reals but can never be embedded in them. In this paper, we will encounter nonetheless an abundance of examples, actually, one for each first-order theory T , to wit, the category of models of T , viewed as a first-order structure in the language of categories. Since categories make a distinction between objects and morphisms, a two-sorted language L_{\text{cat}} is therefore best suited for the model-theoretic study of categories. Apart from that, one needs a (partial) binary operation for composing morphisms, a unary function assigning to an object its identity morphism, and two unary functions assigning to a morphism its domain and range. The axioms in this language are self-evident and the resulting L_{\text{cat}} -theory \mathbf{CAT} will be called the theory of categories . To see how natural all this is: any L_{\text{cat}} -structure is in an obvious way a category, any substructure is a subcategory and the L_{\text{cat}} -homomorphisms between categories are just the functors. Moreover, this language is very expressive: the existence of any finite limit or co-limit is now first-order expressible. But the main observation is universality (see Theorem 6.6 ): Main Theorem . Any category of the form \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T}) , for some theory T in some first-order language L , is universal. Here we write \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T}) for the subcategory of models of T which belong to a once-and-for-all fixed Grothendieck universe \mathfrak{U} (see § 4 for more details). We need to make such an assumption to ensure that the category is small (i.e., is not a class), and therefore a model of \mathbf{CAT} . In this introductory section, let me sketch a proof in one particular case, that of the category of Abelian groups \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T_{\text{Ab}}}) , where T_{\text{Ab}} is the theory of commutative groups in the language L=(+,-,0) . This argument is how one would like to prove the main theorem, by explicitly giving the axioms. Unfortunately, so far, this is basically the only case where I succeeded in doing so. The general case will be proven using ultrapowers: it turns out that any ultrapower \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T})_{\natural} (with respect to an ultrafilter belonging to the universe \mathfrak{U} ) of any model category \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T}) is already a subcategory of the original \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T}) . Together with a characterization of elementarity using ultrapowers, (Proposition 3.1 , a corollary of the famous Keisler-Shelah criterion, but which can easily be proved directly), this will yield the main theorem. But let me explain what I consider to be a better proof at the hand of the category of Abelian groups. 1.1. Abelian groups Let \mathbbm{C} be an arbitrary category in our Grothendieck universe \mathfrak{U} . Assume it admits a null object 0 , meaning that for any object G , there are unique morphisms 0\to G and G\to 0 ; their composition G\to 0\to G will be denoted again by 0 . We will also assume that \mathbb{C} admits (finite) products G_{1}\times G_{2} with projection maps \pi_{1} and \pi_{2} . There are therefore unique morphisms \Delta_{G}\colon G\to G\times G and \tau_{G}\colon G\times G\to G\times G , the diagonal and the twist respectively, such that \pi_{k}\circ\Delta_{G}=1_{G} and p_{k}\circ\tau_{G}\circ\Delta_{G}=p_{l} , with \{k,l\}=\{1,2\} . 1 1 1 If the object G happens to be a set, then \Delta_{G}(a)=(a,a) and \tau_{G}(a,b)=(b,a) . Let us call a morphism \mu\colon G\times G\to G a group arrow , if it satisfies the following four conditions (1) (Associativity) \mu\circ(1_{G}\times\mu)=\mu\circ(\mu\times 1_{G}) ; (2) (Unit) \mu\circ(0\times 1_{G})=p_{2} ; (3) (Inverse) there exists a morphism \lambda\colon G\to G , such that \mu\circ({\tt Id}\times\lambda)\circ\Delta_{G}=0 ; (4) (Commutativity) \mu\circ\tau_{G}=\mu ; Tacitly assumed in these axioms are the various isomorphisms between products, such as (G\times G)\times G\cong G\times(G\times G) , which we simply denote by G\times G\times G , etc. If G is an actual (additive) Abelian group, then we can take \mu(a,b):=a+b and \lambda(a):=-a . Note that commutativity is needed for \mu and \lambda to be actual group homomorphisms. To define the L_{\text{cat}} -theory \mathbf{AB} , we now add to the existence of finite products and a null-object, the two axioms stating (Ab1) that any object G admits a unique group arrow \mu_{G} , and (Ab2), if f\colon G\to H is an arbitrary arrow, with \mu_{G} and \mu_{H} the respective unique group arrows on G and H , then (5) (Linearity) \mu_{H}\circ(f\times f)=f\circ\mu_{G} . To see that \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T_{\text{Ab}}}) satisfies these axioms, taking the group arrow to define the group law on a structure, everything trivially holds except the uniqueness requirement in (Ab1). But this follows from the following simple observation: 1.2 Lemma . Let (G,+,0) be an Abelian group and suppose \mu\colon G\times G\to G defines a monoidal operation on G with the same neutral element 0 . If \mu is additive (for the induced addition on G\times G ), then \mu is just the group law. Proof. By assumption, \mu(a+a^{\prime},b+b^{\prime})=\mu(a,a^{\prime})+\mu(b,b^{\prime}) , for all a,a^{\prime},b,b^{\prime}\in G . Take a^{\prime}=0=b , so that \mu(a,b^{\prime})=\mu(a,0)+\mu(0,b^{\prime}) and since 0 is also the neutral element for the \mu -operation, the latter is just a+b^{\prime} . ∎ Objects in a category \mathbbm{C}\models\mathbf{AB} are actually not yet groups, as the group arrow is only formal. But any functor (preserving products and null objects) into the category of sets will turn them into actual groups and morphisms into group homomorphisms. This is best achieved by positing the existence of a generator Z and taking as actualization the (faithful) functor \operatorname{Hom}_{\mathbbm{C}}(Z,-) (for details, see § 4 ). In this way, \mathbbm{C} becomes a subcategory of \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T_{\text{Ab}}}) . We will therefore also add the axiom asserting the existence of a generator to \mathbf{AB} , and so we proved: 1.3 Corollary . The category \mathbb{M}\!\operatorname{od}_{\mathfrak{U}}({T_{\text{Ab}}}) is universal. ∎ The use of category theory to analyze model-theory is of course not new: Makkai’s work led to what is now called FOCL (=first-order categorical logic; see [ 6 , 7 ] ); or Lawvere’s perspective—from whom I also borrow the use of the language L_{\text{cat}} —leading to ETCC (=elementary theory of the category of categories; see [ 4 , 5 ] ) and its variants, just to mention a few. We use these perspectives in the present paper only in a very simple-minded way, by combining them in order to study some properties of the models of the theory of a model-category. Often, these works have focused on complete theories, so that morphisms

Read the complete text