nLab ultracategory

Contents

Contents

Idea

Ultracategories are categories endowed with extra structure meant to abstract the possibility of forming ultraproducts. They were introduced by Makkai (1987) in order to prove strong conceptual completeness for coherent logic: while the category of models of a coherent theory, on its own, is not sufficient to determine the theory up to any useful notion of equivalence, it becomes sufficient to reconstruct the theory up to Morita equivalence once it is endowed with the structure of an ultracategory.

Theorem

(Makkai) Let PP be a small pretopos and let Mod(P)=Pretop(P,Set)\operatorname{Mod}(P) = \mathbf{Pretop}(P, Set) be its category of models. Then, the evaluation functor

ev:PUlt(Mod(P),Set) ev \colon P \to \mathbf{Ult}\big( \operatorname{Mod}(P) , Set \big)

is an equivalence of categories, where Ult\mathbf{Ult} is the 2-category of ultracategories, ultrafunctors, and transformations thereof.

In other words, ultracategories determine a 2-fully-faithful embedding

Pretop opUlt. \mathbf{Pretop}^{op} \hookrightarrow \mathbf{Ult} .

One of the key motivations to introduce ultracategories lies in the fact that, given two coherent theories T 1T_1 and T 2T_2, the existence of an equivalence of categories Mod(T 1)Mod(T 2)\operatorname{Mod} (T_1) \simeq \operatorname{Mod} (T_2) does not necessarily imply that the two theories are Morita equivalent, i.e. that the syntactic pretoposes P 1P_1 and P 2P_2 of T 1T_1 and T 2T_2, obtained by pretopos completion of their coherent syntactic categories, are equivalent. Indeed, it is only when an equivalence Mod(T 2)Mod(T 1)\operatorname{Mod} (T_2) \to \operatorname{Mod} (T_1) comes from a pretopos morphism I:P 1P 2I : P_1 \to P_2 – corresponding to a suitable kind of interpretation of T 1T_1 in T 2T_2 – that we can apply the usual conceptual completeness theorem, deducing that II is an equivalence itself. The strong conceptual completeness theorem allowed by ultracategories, instead, characterizes the functors Mod(T 2)Mod(T 1)\operatorname{Mod} (T_2) \to \operatorname{Mod} (T_1) coming from a pretopos morphism P 1P 2P_1 \to P_2 in terms of preservation of ultraproducts, in a suitable sense: that is, as the ultrafunctors Mod(T 2)Mod(T 1)\operatorname{Mod} (T_2) \to \operatorname{Mod} (T_1). Thus, the existence of an equivalence Mod(T 1)Mod(T 2)\operatorname{Mod} (T_1) \simeq \operatorname{Mod} (T_2) in Ult\mathbf{Ult} corresponds to the existence of an equivalence of categories P 1P 2P_1 \simeq P_2; in particular, this directly implies the usual conceptual completeness theorem, simply by noting that an ultrafunctor is an equivalence in Ult\mathbf{Ult} if (and only if) it is an equivalence of categories.

To provide intuition on ultracategories, let us start by recalling that a compact Hausdorff space KK can be seen as a relational β \beta -module where each ultrafilter μ\mu on KK converges to a unique limit, namely the unique point xKx \in K such that every open neighborhood of xx in KK lies in μ\mu, so as to determine a function βKK\beta K \to K. More generally, we can say that a function f:IKf \colon I \to K for some set II converges with respect to some ultrafilter μ\mu on II to the limit of the ultrafilter f *(μ)={SK|f 1(K)μ}f_*(\mu) = \{ S \subseteq K | f^{-1}(K) \in \mu\}, i.e. the unique point xKx \in K such that every open neighborhood of xx in KK contains a μ\mu-large subset of values of ff. This way, the convergence relation on KK can be seen as a family of functions

K I×βIK K^I \times \beta I \to K

for each set II: in the same spirit, an ultrastructure (called pre-ultracategory by Makkai ‘87) on a category \mathcal{M} is given at its core by a family of functors

Id: I×βI \int_I - d - : \mathcal{M}^I \times \beta I \to \mathcal{M}

for each set II. Intuitively, for an II-indexed family (m i) iI(m_i)_{i\in I} of objects in \mathcal{M} and an ultrafilter μ\mu on II, the object Im idμ\int_I m_i d \mu is the “abstract ultraproduct” of the family (m i) iI(m_i)_{i\in I} with respect to μ\mu. Different, inequivalent, axiomatizations for the concept of ultracategory exist in the literature: most notably, Lurie‘s definition of ultracategories (together with the introduction of left ultrafunctors) allowed to extend Makkai’s result to a reconstruction theorem for coherent toposes, and it formalizes the idea that ultracategories categorify compact Hausdorff spaces (see Lurie, Thm 3.1.5).

Theorem

(Lurie) Let PP be a small pretopos and let =Sh(P)\mathcal{E} = Sh(P) be its coherent topos of sheaves. Then, the evaluation functor ev:PUlt L(Mod(P),Set)ev : P \to \mathbf{Ult}^L (\operatorname{Mod}(P) , Set ) induces an equivalence

Ult L(Mod(P),Set) \mathcal{E} \simeq \mathbf{Ult}^L ( \operatorname{Mod}(P) , Set )

where Ult L\mathbf{Ult}^L is the 22-category of ultracategories, left ultrafunctors, and transformations thereof. In particular, evev is an equivalence PUlt(Mod(P),Set)P \simeq \mathbf{Ult} (\operatorname{Mod}(P) , Set ), where Ult\mathbf{Ult} is the locally full sub-2-category of Ult L\mathbf{Ult}^L determined by ultrafunctors.

Lurie’s axiomatization was proved by Hamad to coincide with that of normal colax algebras for a pseudomonad on the 2-category of locally small categories. Tarantino and Wrigley weakened Lurie’s definition to obtain an axiomatization of ultracategories sharing the same properties, but emerging universally from the ultrafilter monad. Instead, Saadia and Hamad independently introduced virtual ultracategories as a multicategorical generalization of ultracategories which allows to extend Lurie’s result to toposes with enough points; the former, in particular, introduces yet another axiomatization of ultracategories, satisfying the same reconstruction theorem above, as a particular case of virtual ultracategories. If ultracategories categorify compact Hausdorff spaces, virtual ultracategories categorify all topological spaces, allowing for maps into an abstract ultraproduct which may not exist in the underlying category.

References

On a 2-monadic treatment of ultracategories:

As colax algebras for a pseudomonad:

  • Ali Hamad, Ultracategories as colax algebras for a pseudo-monad on CAT [arXiv:2502.20597]

In terms of Kan extensions of relative monads:

For a generalization to virtual ultracategories, a categorification of relational β \beta -modules designed to allow reconstruction of any topos with enough points:

  • Gabriel Saadia, Extending conceptual completeness via virtual ultracategories [arXiv:2506.23935]

  • Ali Hamad, Generalised ultracategories and conceptual completeness of geometric logic [arXiv:2507.07922]

For a different notion of ultracategory:

Last revised on July 19, 2025 at 16:20:03. See the history of this page for a list of all contributions to it.