In their paper “On Löwenheim–Skolem–Tarski numbers for extensions of first order logic”, Magidor and Väänänen make the following statement:

“For second order logic, $\mathrm{LS}(L^{2})$ [the Löwenheim–Skolem number for second order logic—my comment] is the supremum of $\Pi_{2}$-definable ordinals..., which means that it exceeds the first measurable, the first $\kappa^{+}$-supercompact $\kappa$, and the first huge cardinal if they exist”.

[“The Löwenheim–Skolem number $\mathrm{LS}(L^{2})$ of second order logic $L^{2}$ is the smallest cardinal $\kappa$ such that if a theory $T\subset L^{2}$ has a model, it has a model of cardinality $\le\max(\kappa,|T|)$”, and “$L^{2}$ extends first order logic with quantifiers of the form $\exists R\,\phi(R,x_0,\dots,x_{n-1})$, where the second order variable $R$ ranges over $n$-ary relations on the universe for some fixed $n$”—my comment also but substantially quoting the authors.]

Assume that the first measurable, the first $\kappa^{+}$-supercompact $\kappa$, and the first huge cardinals exist. What type of large cardinal, then, is $\mathrm{LS}(L^{2})$? If the answer is known, please provide the reference.

  • $\begingroup$ Magidor has results on this. $\endgroup$ Jul 19, 2015 at 3:49
  • 3
    $\begingroup$ Why did this get a -1? Seems a fine question to me. $\endgroup$ Jul 19, 2015 at 4:23
  • 5
    $\begingroup$ Thomas, in your definition of the Löwenheim-Skolem number, shouldn't it be $\kappa\leq\text{max}(\kappa,|T|)$, rather than $<$? $\endgroup$ Jul 19, 2015 at 12:20
  • 1
    $\begingroup$ Strange. Wikipedia defines it as I suggest, and that definition makes sense to me: en.wikipedia.org/wiki/L%C3%B6wenheim_number#Extensions. $\endgroup$ Jul 20, 2015 at 1:10
  • 2
    $\begingroup$ @JoelDavidHamkins is right. You can't reasonably expect to get models of $T$ that are smaller than $|T|$. $\endgroup$ Jul 20, 2015 at 15:10

2 Answers 2


The following is due to Magidor:

Theorem 1. Is $\kappa$ is a strong cardinal, then $LS(L^2) < \kappa.$

The proof if easy. Let $T \subseteq L^2$ be a theory and let $A$ be a model for $T$. e may assume the universe is some cardinal $\delta.$ Take some cardinal $\beta > \beth_{\omega}(\delta)$, and let $j: V \to M$ witness $\kappa$ is $\beta$-strong. It is easily seen $M\models$``$A \models T$'', so $M \models \exists B( B \models T, |B| < j(\kappa))$. By elementarity in $V$, $T$ has a model of size $< \kappa.$

Also note that for any theory $T \subseteq L^2,$ there is a least $\delta_T$ such that if $T$ has a model, then it has a model of size $< \delta_T.$ Then $LS(L^2)=\sup \{\delta_T: T$ as above $\}$, so $LS(T^2)$ can be singular, even though it can be above some very large cardinals.


The Löwenheim number (compared to LS, this uses sentences rather than theories) for the second order logic $L^2$ is the least $κ$ such that $V_κ$ satisfies all true $Σ_2$ sentences. This $κ$ has cofinality $ω$, and $V_κ$ does not satisfy ZFC, but $κ$ is a limit of large cardinals. For example, if there is a proper class of huge cardinals, then $κ$ is a limit of huge cardinals, and same with other $Σ_2$ properties.

Assuming countable vocabulary, the Löwenheim-Skolem (LS) number for $L^2$ is the least $κ$ such that $V_κ$ satisfies all true $Σ_2$ sentences with real parameters. $ω_1≤\operatorname{cf}(κ)≤c$, where $c$ is the cardinality of the continuum. As before, $V_κ$ does not satisfy ZFC, but $κ$ is a limit of large cardinals.

If we allowed a proper class of constant symbols, then $L^2$ would not have a Löwenheim-Skolem number. Still, even then, if $V_κ ≺_{Σ_2} V$, then every theory of cardinality $<κ$ with an $L^2$ model has such a model of cardinality $<κ$. For reference, among $κ$ with $V_κ ≺_{Σ_2} V$, the following are in strictly increasing order of the least example (if any, if not assuming large cardinals): $\operatorname{cf}(κ)=ω$, $\operatorname{cf}(κ)=ω_1$, $V_κ⊨\text{ZFC}$, $\operatorname{cf}(κ)=ω_1 ∧ V_κ⊨\text{ZFC}$, $κ$ is regular, $κ$ is Mahlo, $κ$ is weakly compact, $κ$ is measurable, $κ$ is strong. Note that for every strong $κ$, $V_κ ≺_{Σ_2} V$.

Also, the Löwenheim-Skolem-Tarski (LST) number of a logic $L$ is the smallest cardinal $κ$ such that every structure for $L$ has an elementary substructure of size $≤κ$. As noted in the paper in the question, the LST number for $L^2$ (with countable vocabulary) is the least supercompact cardinal, or none if none exists; the Vopěnka's principle holds iff every logic with set-sized vocabulary has an LST number.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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