The Virtual Large Cardinal Hierarchy

Firstly note that (iii)\Rightarrow(ii)\Rightarrow(i) and (v)\Rightarrow(iv) are simply by definition.

(i)\Rightarrow(iv) Assume \delta is virtually Woodin, and fix a function f\colon\delta\to\delta. Let \kappa<\delta be virtually ({<}\delta, f)-strong and let \theta<\delta be a regular cardinal such that \sup_{\alpha\leq\kappa}f(\alpha)<\theta. Then there is a generic elementary embedding \pi\colon (H_\theta, \in, f\cap H_\theta) \to (\operatorname{\mathcal M}, \in, f^+) such that H_\theta\subseteq \operatorname{\mathcal M}, f\cap H_\theta=f^+\cap H_\theta, \operatorname{\mathcal M}\subseteq V, and \pi(\kappa)>\theta. Note that, by our choice of \theta, f\operatorname{\upharpoonright}\kappa\in H_\theta and \pi(f\operatorname{\upharpoonright}\kappa)(\kappa)=f^+(\kappa)=f(\kappa)<\theta.

So it suffices to show that \kappa is a closure point for f. Let \alpha<\kappa. Then

so \pi fixes f(\alpha) for every \alpha<\kappa. Now, if \kappa was not a closure point of f then, letting \alpha<\kappa be the least such that f(\alpha)\geq\kappa, we have

a contradiction. Note that we used that \pi(\kappa)>\theta here, so this argument would not work if we had only assumed \delta to be virtually pre-Woodin.

(iv)\Rightarrow(vi) Assume (iv) holds, let f\colon\delta\to\delta be given and define g\colon\delta\to\delta as g(\alpha):=(2^{<\gamma_\alpha})^+, where \gamma_\alpha is the least regular cardinal above |f(\alpha)|. By (iv) there is a \kappa<\delta which is a closure point of g (and so also a closure point of f), and there is a regular \lambda\in(\kappa, \delta) for which there is a generic elementary embedding \pi\colon H_\lambda\to\operatorname{\mathcal M} with \operatorname{crit}\pi=\kappa, H_\lambda\subseteq\operatorname{\mathcal M}, \operatorname{\mathcal M}\subseteq V, and \pi(g\operatorname{\upharpoonright}\kappa)(\kappa)<\lambda.

Let \theta be the least regular cardinal above |\pi(f\operatorname{\upharpoonright}\kappa)(\kappa)|, and note that H_\theta\in H_\lambda by our definition of g. Thus, both H_\theta and H_{\pi(\theta)}^{\operatorname{\mathcal M}} are elements of \operatorname{\mathcal M}. An application of Countable Embedding Absoluteness ? then yields that \operatorname{\mathcal M} has a generic elementary embedding \pi^*\colon H_\theta^{\operatorname{\mathcal M}}\to H_{\pi(\theta)}^{\operatorname{\mathcal M}} such that \operatorname{crit}\pi^*=\kappa, \pi^*(\kappa)=\pi(\kappa), \pi(f\operatorname{\upharpoonright}\kappa)\in\operatorname{ran}\pi^*, and \pi(f\operatorname{\upharpoonright}\kappa)(\kappa)<\theta. By elementarity of \pi, H_\lambda has an ordinal \bar\theta<\kappa and a generic elementary embedding \sigma\colon H_{\bar\theta}\to H_\theta with \sigma(\operatorname{crit}\sigma)=\kappa, f\operatorname{\upharpoonright}\kappa\in\operatorname{ran}\sigma and f(\operatorname{crit}\sigma)<\bar\theta, which is what we wanted to show.

(vi)\Rightarrow(v) Assume (vi) holds and let f\colon\delta\to\delta be given. Define g\colon\delta\to\delta as g(\alpha):=\langle(2^{<\gamma_\alpha})^+,f(\alpha)\rangle, where \gamma_\alpha is the least regular cardinal above |f(\alpha)|. In particular, g codes f. By (vi) there exist regular \bar\kappa<\bar\lambda<\kappa<\lambda such that \kappa is a closure point of g (so also a closure point of f) and there exists a generic elementary embedding \pi\colon H_{\bar\lambda}\to H_\lambda with \operatorname{crit}\pi = \bar\kappa, \pi(\bar\kappa)=\kappa, g(\bar\kappa)<\bar\lambda, and g\operatorname{\upharpoonright}\kappa\in\operatorname{ran}\pi. Since f is definable from g and g\operatorname{\upharpoonright}\kappa\in\operatorname{ran}\pi, it follows that f\operatorname{\upharpoonright}\kappa\in\operatorname{ran} \pi. So let \pi(\bar f)=f\operatorname{\upharpoonright}\kappa with \bar f\colon\bar\kappa\to\bar\kappa. Now observe that \bar f=f\operatorname{\upharpoonright}\bar\kappa since for \alpha<\bar\kappa, we have f(\alpha)=\pi(\bar f)(\alpha)=\pi(\bar f(\alpha))=\bar f(\alpha).

Let \bar\theta be the least regular cardinal above |f(\bar\kappa)|. By the definition of g, we have H_{\bar\theta}\in H_{\bar\lambda}. Now, following the (iii)\Rightarrow(ii) direction in the proof of Theorem ? we get that H_{\bar\lambda} has a generic elementary embedding \sigma\colon H_{\bar\theta}\to\operatorname{\mathcal M} with \operatorname{\mathcal M} closed under {<}\bar\theta-sequences from V, \operatorname{crit}\sigma=\bar\kappa, \sigma(\bar\kappa)> \bar\theta, and \sigma(\bar f)(\bar\kappa)<\bar\theta. Let \pi(\bar\theta)=\theta and \pi(\operatorname{\mathcal M})=\operatorname{\mathcal N}. Now by elementarity of \pi, we get that there is a generic elementary embedding \sigma^*:H_\theta\to \operatorname{\mathcal N} with \operatorname{crit}\sigma^*=\kappa, \sigma^*(\kappa)>\theta, and \sigma^*(\pi(\bar f))(\kappa)=\sigma^*(f\operatorname{\upharpoonright}\kappa)(\kappa)<\theta.

(vi)\Rightarrow(iii) Let C be the club of all cardinals \alpha such that

Let f\colon\delta\to\delta be given as f(\alpha):=\langle \gamma_0^\alpha,\gamma_1^\alpha\rangle, where \gamma_0^\alpha is the first limit point of C above \alpha and the \gamma_1^\alpha are chosen such that \{\gamma_1^\alpha\mid\alpha<\beta\} encodes A\cap H_\beta for inaccessible cardinals \beta<\delta.

Let \kappa<\delta be a closure point of f such that there are regular cardinals \bar\theta<\kappa<\theta and a generic elementary embedding \pi\colon H_{\bar\theta}\to H_\theta such that \pi(\operatorname{crit}\pi)=\kappa, f(\operatorname{crit}\pi)<\bar\theta, and f\operatorname{\upharpoonright}\kappa\in\operatorname{ran}\pi. Let \bar\kappa=\operatorname{crit} \pi. The same argument as above gives that \pi(f\operatorname{\upharpoonright} \bar\kappa)=f\operatorname{\upharpoonright}\kappa. In particular, it follows that \bar\kappa is a closure point of f, and hence \bar\kappa\in C. We claim that \bar\kappa is virtually ({<}\delta, A)-extendible. Since \kappa\in C because it is a closure point of f, it suffices by the definition of C to show that

Let \beta be the least element of C above \bar\kappa, and note that \beta is below \bar\theta since f(\bar\kappa)<\bar\theta, and the definition of f says that the first coordinate of f(\bar\kappa) is a limit point of C above \bar\kappa. It then holds that

as both \bar\kappa and \beta are elements of C. Since f encodes A in the manner previously described and \pi(f\operatorname{\upharpoonright}\bar\kappa)=f\operatorname{\upharpoonright}\kappa, we get that \pi(A\cap H_{\bar\kappa}) = A\cap H_\kappa, and thus

for A^* := \pi(A\cap H_\beta). Now, as (H_\gamma,\in, A\cap H_\gamma) and (H_{\pi(\gamma)},\in,A^*\cap H_{\pi(\gamma)}) are elements of H_{\pi(\beta)} for every \gamma<\beta, Countable Embedding Absoluteness ? implies that H_{\pi(\beta)} sees that \bar\kappa is virtually ({<}\beta, A^*)-extendible. Since

it follows that (H_\beta,\in A\cap H_\beta) satisfies that \bar\kappa is virtually ({<}\beta, A\cap H_\beta)-extendible. Finally, since

it follows that (H_\kappa,\in, A\cap H_\kappa) satisfies that \bar\kappa is virtually ({<}\kappa,A\cap H_\kappa)-extendible.

As a corollary of the proof, we now have that virtually Woodin cardinals and faintly Woodin cardinals are equivalent.

[prop.weakerwoodin] A cardinal \delta is virtually Woodin if and only if it is faintly Woodin.

Indeed, the argument would work as well if we only assumed existence of generic embeddings \pi:(H_\theta,\in,A\cap H_\theta)\to (\operatorname{\mathcal M},\in,B) such that \operatorname{crit} \pi=\kappa, \pi(\kappa)>\theta, H_\theta\subseteq \operatorname{\mathcal M} and A\cap H_\theta=B\cap H_\theta. In other words, we do not need the a priori stronger assumption that H_\theta=H_\theta^{\operatorname{\mathcal M}}. Using Proposition ?, it suffices to observe that if \pi:(H_\theta,\in,A)\to (\operatorname{\mathcal M},\in,B) is a faintly (\theta,A)-strong embedding such that A codes the sequence of H_\lambda for \lambda<\theta, then H_\theta=H_\theta^{\operatorname{\mathcal M}}.

We should also observe that if \delta is virtually Woodin, then indeed for every A\subseteq H_\delta, we have stationary many virtually ({<}\delta,A)-extendible cardinals by an argument very similar to the proof that virtually Woodin cardinals are Mahlo.

Recall that the virtual Vopěnka Principle states that for every proper class \mathcal C consisting of structures in a common language, there are distinct structures \operatorname{\mathcal M},\operatorname{\mathcal N}\in \mathcal C for which there is a generic elementary embedding \pi\colon\operatorname{\mathcal M}\to\operatorname{\mathcal N}. The second author and Hamkins showed in [GitmanHamkins] that the virtual Vopěnka principle holds if and only if for every class A there is a proper class of weakly virtually A-extendible cardinals (in our terminology, these are ({<}\textsf{On},A)-pre-extendible cardinals). It follows from Theorem ? that if \textsf{On} is faintly Woodin, then the virtual VopěnkaPrinciple holds. However, since it is consistent that the virtual Vopěnka Principle holds and \textsf{On} is not Mahlo [GitmanHamkins], the two assertions are not equivalent.

It turns out, however, that a weakening of the virtual Vopěnka Principle is equivalent to the assertion that \textsf{On} is faintly pre-Woodin. Our formal setting for working with classes throughout this article will be the second-order Godël-Bernays set theory \textsf{GBC} whose axioms consist of \textsf{ZFC} together with class axioms consisting of extensionality for classes, class replacement asserting that every class function when restricted to a set is a set, global choice asserting that there is a class well-order of sets, and a weak comprehension scheme asserting that every first-order formula defines a class.

[defi.VopenkaFinite] The virtual Vopěnka Principle for finite languages states that for every proper class \mathcal C consisting of structures in a common finite language, there are distinct structures \operatorname{\mathcal M},\operatorname{\mathcal N}\in \mathcal C for which there is a generic elementary embedding \pi\colon\operatorname{\mathcal M}\to\operatorname{\mathcal N}.

The arguments in [GitmanHamkins] show:

[theo.virtVopFinAext] The virtual Vopěnka Principle for finite languages holds if and only if for every class A, there is a weakly virtually A-extendible cardinal. The Vopěnka Principle is equivalent to the Vopěnka Principle for finite languages. Indeed, the Vopěnka Principle can be restated in terms of the existence of elementary embeddings between elements of natural sequences [Kanamori]. But as we will see in the next section, this equivalence relies once again relies on Kunen’s Inconsistency.

Say that a class function f\colon\textsf{On}\to\textsf{On} is an indexing function if it satisfies that f(\alpha)>\alpha and f(\alpha)\leq f(\beta) for all \alpha<\beta.

Say that an \textsf{On}-sequence \vec\operatorname{\mathcal M}=\langle \operatorname{\mathcal M}_\alpha\mid\alpha<\textsf{On}\rangle is natural if there exists an indexing function f^{\vec\operatorname{\mathcal M}}\colon\textsf{On}\to\textsf{On} and unary relations R^{\vec\operatorname{\mathcal M}}_\alpha\subseteq V_{f^{\vec\operatorname{\mathcal M}}(\alpha)} such that \operatorname{\mathcal M}_\alpha = (V_{f^{\vec\operatorname{\mathcal M}}(\alpha)}, \in, \{\alpha\}, R_\alpha^{\vec\operatorname{\mathcal M}}) for every \alpha.

The following theorem shows that the virtual Vopěnka Principle for finite languages holds if and only if \textsf{On} is virtually pre-Woodin if and only if \textsf{On} is faintly pre-Woodin if and only if for every class A there is an weakly virtually A-extendible cardinal. In particular, we get that virtually pre-Woodin and faintly pre-Woodin cardinals \delta are equivalent, and both are equivalent to the assertion that for every A\subseteq H_\delta, there is a ({<}\delta,A)-pre-extendible cardinal. In the next section, in Theorem ?, we will separate the virtual Vopěnka Principle for finite languages from the virtual Vopěnka principle.

[theo.vopwood] The following are equivalent.

1. The virtual Vopěnka Principle for finite languages holds.

2. For every class A, there is a ({<}\textsf{On},A)-pre-extendible cardinal.

3. For any natural \textsf{On}-sequence \vec\operatorname{\mathcal M} there exists a generic elementary embedding \pi\colon\operatorname{\mathcal M}_\alpha\to\operatorname{\mathcal M}_\beta for some \alpha<\beta.

4. \textsf{On} is virtually pre-Woodin.

5. \textsf{On} is faintly pre-Woodin.

Proof. By Theorem ?, (i) and (ii) are equivalent. (i)\Rightarrow(iii) and (iv)\Rightarrow(v) are trivial.

(v)\Rightarrow(i): Assume that \textsf{On} is faintly pre-Woodin and fix some class \mathcal C of structures in a common language. Let \kappa be ({<}\textsf{On},\mathcal C)-pre-strong. Fix some regular \theta>\kappa such that H_\theta has a structure M from \operatorname{\mathcal C} of the \kappa-th rank among elements of \mathcal C, and fix a generic elementary embedding

with \operatorname{crit} \pi=\kappa, H_\theta=H_\theta^{\operatorname{\mathcal N}} (note that actually H_\theta\subseteq \operatorname{\mathcal N} suffices here), and \mathcal C\cap H_\theta=\mathcal C^*\cap H_\theta. Now we have that \pi:M\to \pi(M) is an elementary embedding (note that the language is finite and therefore can be assumed to be fixed by \pi), and \pi(M)\neq M because \pi(M) is a structure in \mathcal C^* of rank \pi(\kappa) among elements of \mathcal C^*. The structure (\operatorname{\mathcal N},\in,\mathcal C^*) believes that both M and \pi(M) are elements of \mathcal C^*, and, by Countable Embedding Absoluteness ?, \operatorname{\mathcal N} has a generic elementary embedding between M and \pi(M). Therefore, (\operatorname{\mathcal N},\in,\mathcal C^*) satisfies that there is a generic elementary embedding between two distinct elements of \operatorname{\mathcal C}^*, and hence (H_\theta,\in,\mathcal C\cap H_\theta) satisfies that there is a generic elementary embedding between two distinct elements of \mathcal C, and it must be correct about this.

(iii)\Rightarrow(iv): Assume (iii) holds and assume that \textsf{On} is not virtually pre-Woodin, which means that there exists some class A for which there are no virtually ({<}\textsf{On},A)-pre-strong cardinals. Define a function f\colon\textsf{On}\to\textsf{On} by setting f(\alpha) to be the least regular \eta>\alpha such that \alpha is not virtually (\eta,A)-pre-strong. Also define g\colon\textsf{On}\to\textsf{On} by setting g(\alpha) to be the least strong limit cardinal above \alpha which is a closure point of f. Note that g is an indexing function, so we can let \vec\operatorname{\mathcal M} be the natural sequence induced by g and R_\alpha := A\cap H_{g(\alpha)}. (ii) supplies us with \alpha<\beta and a generic elementary embedding

Let us argue that \pi is not the identity map, so that it must have a critical point. Since g(\alpha) is a closure point of f, the structure (H_{g(\alpha)},\in,A\cap H_{g(\alpha)}) can define f correctly. So it must satisfy that f does not have a strong limit closure point above \alpha, but the structure (H_{g(\beta)}, \in, A\cap H_{g(\beta)}) does have such a closure point, namely g(\alpha). Thus, \pi must have a critical point. Now since g(\alpha) is a closure point of f, it holds that f(\operatorname{crit}\pi)<g(\alpha), so fixing a regular \theta\in(f(\operatorname{crit}\pi), g(\alpha)) we get that \operatorname{crit}\pi is virtually (\theta, A)-pre-strong, contradicting the definition of f. Hence \textsf{On} is virtually pre-Woodin. \hfill\blacksquare