I'm reading through Joel David Hamkins' set theory lecture notes. On page 14, on the subject of inaccessible cardinals and submodels of ZFC in $V$, he defines a universe cardinal to be a cardinal $\gamma$ such that $V_\gamma$ models ZFC. For example, if an inaccessible cardinal $\kappa$ exists, then $\kappa$ is a universe cardinal. But furthermore, by the Löwenheim-Skolem theorem, there are then lots of models of ZFC inside $V_\kappa$, and Hamkins says that this proves that $\kappa$ is the $\kappa$th universe cardinal. Specifically, $\{\gamma < \kappa: V_\gamma \prec V_\kappa\}$ is evidently closed in $\kappa$, and is "unbounded by Löwenheim-Skolem".

  1. This is confusing to me. The Löwenheim-Skolem theorem says that $V_\kappa$ contains submodels of arbitrary cardinality. But in order to have any universe cardinals at all, you need models of ZFC of the form $V_\gamma$, which don't seem to come from Löwenheim-Skolem. Why should these exist? Is there something you can do to a model to put it in the form $V_\gamma$?

  2. Relatedly, it's also true (this is in Kanamori's Higher Infinite) that the only models for second-order ZFC are $V_\kappa$ for $\kappa$ inaccessible. Can someone explain the difference between these two concepts? I'm finding the distinction between first-order and second-order Replacement a little hair-splitting.

I hope I've phrased this the right way -- I'm a set theory rookie. Thanks in advance!

  • 1
    $\begingroup$ First-order replacement schema is handle the definable class function, while second-order replacement axiom can handle arbitrary class function (even the given class function is not definable.) $\endgroup$
    – Hanul Jeon
    May 22, 2014 at 15:13
  • $\begingroup$ @tetori: thanks. So the idea is that if $V_\kappa$ models second-order ZFC, then every chain of ordinals of length less than $\kappa$ is the image of a function, so $\kappa$ must be regular. Whereas if $V_\kappa$ just models first-order ZFC, you could evade regularity of $\kappa$ by somehow blocking all these functions from being definable in $V_\kappa$. Correct? $\endgroup$ May 22, 2014 at 16:25
  • $\begingroup$ @PaulVonKaoghnett I don't think about this part so I don't assure that your arguement is correct. But by Asaf's answer says that the front part of your argument is right. $\endgroup$
    – Hanul Jeon
    May 23, 2014 at 6:29
  • $\begingroup$ It is known that minimal $\kappa$ which $V_\kappa$ makes the model of ZFC is singular; see related post in Cantor's attic $\endgroup$
    – Hanul Jeon
    May 23, 2014 at 6:30

2 Answers 2


Yes, you do need a tiny additional argument to get universe cardinals below an inaccessible $\kappa$. The additional step is an elementary chain argument, which tends to occur relatively frequently.

Starting from an inaccessible $\kappa$, Löwenheim-Skolem gives a small elementary submodel $M_0\prec V_\kappa$. We can then put $M_0$ inside $V_{\alpha_0}$ for some cardinal $\alpha_0<\kappa$. Applying Löwenheim-Skolem again gives a new elementary submodel $V_{\alpha_0}\subseteq M_1\prec V_\kappa$. We then put $M_1$ into some $V_{\alpha_1}$ again. After repeating this $\omega$ many times we get intertwined chains of models $M_n$ of ZFC and initial segments $V_{\alpha_n}$ of the universe which both union up to $V_{\sup \alpha_n}$, which is elementary in $V_{\kappa}$. This implies that $\sup\alpha_n<\kappa$ is a universe cardinal.

By the way, the terminology has shifted slightly since Joel wrote those notes and the term now used for these cardinals is worldly cardinals.

  • $\begingroup$ Thanks, this answers my question! By the way, can you take arbitrary (longer than $\omega$) unions of chains of submodels? $\endgroup$ May 22, 2014 at 16:18
  • 1
    $\begingroup$ @PaulVanKoughnett Indeed. The argument is flexible enough to give worldly cardinals of arbitrary cofinality (below $\kappa$). $\endgroup$ May 22, 2014 at 16:32
  • 1
    $\begingroup$ What's blocking the application of this argument to show (the false statement) that if $\kappa$ is a $\beth$-fixed point of uncountable cofinality, then there's some $\gamma$ such that $V_\gamma\prec V_\kappa$? $\endgroup$ Sep 8, 2020 at 22:45
  • $\begingroup$ @JasonZeshengChen Why is that statement false? $\endgroup$ Sep 9, 2020 at 0:10
  • $\begingroup$ @MihaHabič, oops, I meant unprovable. What I'm thinking is that ZFC proves that if $\alpha,\beta$ are ordinals and $V_\alpha\prec V_\beta$, then $V_\alpha$ is a model of ZFC. ZFC also proves that there are $\beth$-fixed points of uncountable cofinality. So if ZFC were to prove the statement, then ZFC would prove that some $V_\gamma$ is a model of ZFC. $\endgroup$ Sep 9, 2020 at 0:26

Miha gave a very good answer to the first question. Let me add a bit on the second question.

The difference between first-order and second-order replacement is subtle. First-order replacement is a schema, and it says that if $f$ is an internally definable function with a set domain, then its range is a set; whereas the second-order replacement axiom is a single axiom, and it states that whenever $f$ is a subset of $V_\kappa$ which is a function, and its domain is a set of $V_\kappa$, then its range is an element of $V_\kappa$ as well.

If $V_\kappa$ satisfies the second-order axiom, then it $\kappa$ to be regular. To see this note that:

  1. If $A\subseteq\kappa$ is an unbounded set, then $A\notin V_\kappa$.
  2. Every unbounded set of ordinals is the range of a function which enumerates it.

Now if $\kappa$ is singular, there is some $\gamma<\kappa$, and a function $f\colon\gamma\to\kappa$ which enumerates an unbounded set. This function $f$ is a subset of $V_\kappa$, and its domain is an element of $V_\kappa$, but its range is not.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.