Assuming the Grothendieck axiom of universes (see also here or here), let $U_0$ denote the universe of "ZFC sets", i.e. of sets that can be constructed in ZFC alone without assuming axioms equivalent to the existence of inaccessable cardinals. Then the "standard" category of sets "$Set$" corresponds to all sets in $U_0$. (This paper calls $U_0$ "small-definable" sets.)

The "standard" Yoneda embedding is a functor $Y: \mathcal{C} \to Set^{\mathcal{C}^{op}}$ that takes a locally-$U_0$-small category $\mathcal{C}$ to its category of ($U_0$-)presheaves ($Set^{\mathcal{C}^{op}}$) by sending each object of $\mathcal{C}$ to its representable presheaf (and acting correspondingly on morphisms).

Given a Grothendieck universe $U$, let $U$-$0Cat$ denote the ($1$-)category of $U$-small $0$-categories. Then

Question: is it always possible to define a Yoneda embedding functor $Y: \mathcal{C} \to U\text{-}0Cat^{\mathcal{C}^{op}}$ for any locally-$U$-small category $\mathcal{C}$? In other words
(1) can we replace "sets" with "$0$-categories" in the statement of the Yoneda lemma? and
(2) does the Yoneda lemma still work for arbitrary Grothendieck universes $U$, rather than just $U_0$?

I assume from here that the answer to (2) is yes, so the answer (1) is more interesting to me (whence the question title).


  • Size issues are apparently inherent to the notions of presheaf categories and the Yoneda embedding, see e.g. these answers (1)(2)(3) to questions from MathOverflow. In particular, although the "standard" statement gives the impression that the Yoneda Lemma can only possibly apply to categories $\mathcal{C}$ that are locally-$U_0$-small, it seems that in fact there are important applications to categories larger than this. (Cf. example 8.1 here.)
  • The appearance of sets (which can be axiomatized/defined in many different ways) in one of the foundational/most important results/constructions in category theory always seemed bizarre/"artificial" to me. But a theorem relating the structure of ($1$-)categories to those of $0$-categories seems much more "natural" and much less "artificial".

Possibly misguided assumptions motivating question:

  • Due to the "philosophy" of "negative thinking"/"decategorification"/"n-categories", an $(n-1)$-category should be equivalent (strictly? weakly?) to an $n$-category in which all $n$-morphisms are the identity. Thus a $0$-category should be "the same thing" as a $1$-category that is both discrete and skeletal.
  • The category of all $U_0$-small, discrete, and skeletal ($1-$)categories is isomorphic to the category of all $U_0$-small ("ZFC-small") sets "$Set$", while the category of all essentially-$U_0$-small, discrete ($1-$)categories is equivalent to "$Set$". Cf. these two related questions (1)(2). (Although perhaps the latter category cannot exist due to size reasons; I don't know.)
  • If the ($1$-)category $\mathcal{C}$ is $U_n$-small (thus in particular $U_n$-locally-small), the category of presheaves $U$-$0Cat^{\mathcal{C}^{op}}$ into which the ($U$-)Yoneda embedding embeds $\mathcal{C}$ is always $U_n$-large for any universe $U$, even for the $U_0$ ("ZFC") presheaf category $U_0$-$0Cat^{\mathcal{C}^{op}}$ (corresponding to the "standard" Yoneda embedding). At least that's what I inferred from these related questions (1)(2)(3) where one of the main conclusions is that the ("standard") Yoneda embedding is never essentially surjective due to size issues. (This answer seems to be related to the "uncheatable lemma" from here or here.) nLab also seems to imply something like this.
  • 2
    $\begingroup$ It is a little unclear what precisely you think a 0-category is. The category of 0-categories, no matter what, is equivalent to the category of sets, so the Yoneda lemma would be unaffected. $\endgroup$ Apr 6, 2022 at 22:38
  • $\begingroup$ @KevinArlin If this is obviously true, then I consider that great! I just learned that there is even a notion of "$0$-category" within the past two weeks. John Baez's arxiv.org/pdf/q-alg/9705009.pdf article says that the $k$-morphisms for $k>n$ of an $n$-category should all be identity morphisms. So to me that sounds like a $0$-category has to in some sense be "equivalent" to a discrete and skeletal $1$-category. But that's really all I understand. $\endgroup$ Apr 6, 2022 at 23:13
  • $\begingroup$ In particular, that such a thing should be equivalent to a "set" does not seem to me to follow from the (1st order logic/elementary) axioms of a category. (That would just involve "variables" corresponding to objects without specifying a specific "thing" the "variables" need to belong to.) It seems to follow from whatever foundations are implicitly being used when we choose to talk about a "collection" of objects. We could choose ZFC/NBG, in which case a "collection" of objects, and thus a $0$-category, must be a $U_0$ set or $U_1$ set ("proper class"). But are we really free to assume $\endgroup$ Apr 6, 2022 at 23:17
  • 3
    $\begingroup$ It is unnecessary to restrict to $U_0$-sets for the Yoneda embedding. You may as well prove it for all universes. It takes no additional effort. $\endgroup$
    – Zhen Lin
    Apr 7, 2022 at 0:31
  • 3
    $\begingroup$ I don’t see the relevance of any size questions at all here. A (weak) 0-category is any kind of category in which any two objects connected by any morphism are connected by an isomorphism which is itself unique up to an isomorphism, which is…and so on. No reason to be skeletal by the way. The map from such a thing to its set of equivalence classes, viewed as some kind of $n$-category with no non-identity morphisms, is visibly an equivalence. $\endgroup$ Apr 7, 2022 at 0:34


You must log in to answer this question.