# “Largish” cardinals

In what follows, $\mathsf{ZCKP}$ refers to the subset of $\mathsf{ZFC}$ consisting of the axioms of Zermelo set theory with choice and foundation ($\mathsf{ZC}$) plus those of Kripke-Platek set theory ($\mathsf{KP}$): equivalently, as these two theories have much overlap, $\mathsf{ZCKP}$ consists of $\mathsf{ZC}$ plus the axiom of replacement for $\Sigma_1$-formulæ (or equivalently, $\mathsf{KP}$ plus choice, foundation, powerset, infinity, and full separation). A "largish" cardinal property means, informally, one whose existence is provable in $\mathsf{ZFC}$ but not in $\mathsf{ZCKP}$.

Here is the simplest example of a largish cardinal notion. A theorem of Azriel Lévy (see, e.g., Barwise, Admissible Sets and Structures (1975), theorem II.3.5 on page 53 and theorem II.9.1 on page 76) states that for every uncountable cardinal $\kappa$, if $H(\kappa)$ is the set of sets hereditarily of cardinality $<\kappa$, then $H(\kappa)$ is a $1$-elementary submodel of the universe (meaning that every $\Sigma_1$ formula with parameters in $H(\kappa)$ is true [in $V$] iff it is true in $H(\kappa)$). This implies that $H(\kappa)$ satisfies $\Delta_0$-collection (eqvt. $\Sigma_1$-replacement), and if $\kappa$ is a strong limit, then $H(\kappa) \models \mathsf{ZCKP}$ (and the converse is clear). In particular, $\mathsf{ZCKP}$ does not prove the existence of strong limit cardinals.

Now I am interested in strengthenings of this condition on $\kappa$ such that the existence of these cardinals is still provable in $\mathsf{ZFC}$. Two obvious candidates are:

• $H(\kappa)$ satisfies $\mathsf{ZC}$ plus replacement for $\Sigma_n$ formulæ,

• $H(\kappa)$ is an $n$-elementary submodel of the universe (i.e., every $\Sigma_n$ formula with parameters in $H(\kappa)$ is true iff it is true in $H(\kappa)$).

These should at least imply that $\kappa$ is a fixed point of the beth function, so that in fact $H(\kappa) = V_\kappa$. (Perhaps this should be added as a precondition to be worthy of the term "largish cardinal".)

Edit (on 2015-12-11, following the answer by Joel David Hamkins): I didn't realize how very different the two notions above are: the first (call them "$\Sigma_n$-replacing" cardinals) is "local" in that it involves only sets from $V_\kappa$ and can thus be expressed as a $\Delta_1$ property of $V_\kappa$, whereas the latter ("$\Sigma_n$-correct cardinals") is "global" and involves the entire universe. This has a consequence of size: the smallest $\Sigma_2$-correct cardinal, as explained in Joel's response, is larger than the first $\Sigma_n$-replacing cardinal for all $n$, or even the first inaccessible, etc., and there is no hope of "computing" it. There may be some hope for the first $\Sigma_2$-replacing cardinal, however. I really should have asked two different questions.

My question is this: Can these conditions, at least for $n=2$, or perhaps some related ones, be rephrased in purely cardinal-theoretic terms (without appealing to model theory and if possible avoiding the Lévy hierarchy)? Even better, can the smallest cardinal satisfying such a condition be "described" or "computed" in some way? (In the same way that $\beth_\omega$, or "the limit of the sequence defined by $\kappa_0 = \omega$ and $\kappa_{n+1} = \beth_{\kappa_n}$" are descriptions/computations of the smallest strong limit cardinal and the smallest fixed point of the beth function.)

More generally, any comments on these or related properties would be welcome (including a better term than "largish cardinal"). There is probably some connection with powerset-admissible ordinals, although the exact relation escapes me.

One reason why one might be interested in such cardinals is that the corresponding $H(\kappa)$ might serve as a drop-in replacement for Grothendieck universes in a $\mathsf{ZFC}$ formulation of category theory (they are not fully Grothendieck universes, but the point is that the use of a construction that escapes from such a "universish" set is likely to be so rare as to be very conspicuous; and unlike Grothendieck universes, their existence follows from $\mathsf{ZFC}$).

Your cardinals are known as the $\Sigma_n$-correct cardinals, and they arise in diverse set-theoretic contexts. For example, we use them extensively in our paper:

It is a ZFC theorem that the $\Sigma_n$-correct cardinals form a closed unbounded proper class often denoted $C^{(n)}$.

One subtle point about the $\Sigma_n$-correct cardinals is that although we have a concept of $\Sigma_2$-correct and $\Sigma_3$-correct and so on, $\Sigma_n$-correct for any particular $n$, there is no uniform-in-$n$ way to express the concept of $\Sigma_n$-correctness in first-order set theory. The concept is uniformly expressible in some second-order set theories, such as Kelley-Morse set theory, which prove that there is a truth-predicate for first-order truth.

Concerning your question, when $n\geq 2$ there can be no way to define what it means for $\kappa$ to be $\Sigma_n$-correct by looking only below $\kappa$, say, as a limit process, since such a property would be too simple, as it could be verified inside $V_\kappa$ itself, but such verifiable-in-$V_\kappa$ properties have complexity at worst $\Delta_2$. For example, the property of being $\Sigma_n$-correct cannot be $\Sigma_n$-expressible, for then the assertion "There is a $\Sigma_n$-correct cardinal" would reflect from $V$ to $V_\kappa$, even when $\kappa$ is the least $\Sigma_n$-correct cardinal, which gives a contradiction since there are none below the least one. Meanwhile, the property of being $\Sigma_n$-correct is $\Pi_n$-expressible, since one need only say that all the instances of $\Pi_n$ truth in $V_\kappa$ are actually true.

The case of $\Sigma_2$-correct cardinals is particularly attractive, and perhaps this is an example that interests you. The $\Delta_2$ properties are precisely the properties that are local, in the sense that they can be determined in any sufficiently large $H(\theta)$. You can read more on my blog post:

It follows that a cardinal $\kappa$ is $\Sigma_2$-correct, if whenever there is an object having a certain properties inside some possibly very large $H(\theta)$, then there is such an object inside such an $H(\theta)$ with $\theta<\kappa$. In other words, $\kappa$ is $\Sigma_2$-correct, if whenever anything verifiable happens anywhere, then it happens inside $V_\kappa$. Alternatively, everything verifiable has already happened by the time you get to $H_\kappa$. Such a way of understanding $\Sigma_2$-correctness is extremely useful, since it aligns with how set theorists often think about verifying set-theoretic facts.

(A small matter: the distinction between $V_\kappa$ and $H_\kappa$ disappears once $n\geq 2$, since in this case the cardinals are $\beth$-fixed points and so $V_\kappa=H_\kappa$.)

Lastly, your idea of using the $\Sigma_n$-correct cardinals as a universe replacement idea is well known. This is known as the Feferman theory, and I also discussed it here on MathOverflow in my answer to the question What interesting/nontrivial results in Algebraic geometry require the existence of universes?.

• Thanks! Just to clarify, the $\Sigma_n$-correct cardinals are those which satisfy the second property I mentioned ($V_\kappa \mathrel{\prec_n} V$). The first (viz., $V_\kappa \models \mathsf{ZC} + \Sigma_n\textrm{-replacement}$), is of a different nature since it can be checked by looking at $V_\kappa$ alone. (It's also implied by $\Sigma_n$-correctness, although I'm not sure I didn't miss a $\pm1$ on the $n$ here.) So I'll wait a bit before approving your answer to see if someone (or you yourself) has something to say about that other property. – Gro-Tsen Dec 9 '15 at 22:34
• Yes, that's right. Officially, it is defined with $H_\kappa\prec_n V$, but this difference only matters for $n=1$, since once $n\geq2$ then we have $H_\kappa=V_\kappa$ as you noted. – Joel David Hamkins Dec 9 '15 at 22:37