Cofinal families of finite VC-dimension


Abstract

Given infinite cardinals \(\theta\leq \kappa\), we ask for the minimal VC-dimension of a cofinal family \(\mathcal{F}\subseteq[\kappa]^{<\theta}\). We show that for \(\theta=\omega\) and \(\kappa=\aleph_n\) it is consistent with ZFC that there exists such a family of VC-dimension \(n+1\), which is known to be the lower bound. For \(\theta>\omega\) we answer this question completely, demonstrating a strong dichotomy between the case of singular and regular \(\theta\). We furthermore answer some relative and generalized versions of the above question for singular \(\theta\), and answer a related question which appears in BBNKS?.

1 Introduction↩︎

Suppose that \(X\) is an infinite set of size \(\kappa\) and that \(\mathcal{F}\) is a family of subsets, each of cardinality at most \(\theta\) for \(\theta\leq \kappa\). Consider two possible properties:

  1. The family \(\mathcal{F}\) is cofinal: every set of cardinality \(<\theta\) is contained in a member of \(\mathcal{F}\).

  2. The family \(\mathcal{F}\) has finite VC-dimension, meaning that there is some \(n<\omega\) such that given any subset \(s\) of \(X\) of size \(n\), there is some \(t\subseteq s\) such that \(t\) is not of the form \(s \cap A\) for some \(A\in \mathcal{F}\), see 10.

Observe that ([enu:cofinal]) pulls \(\mathcal{F}\) towards having many sets while ([enu:small32VC]) pulls \(\mathcal{F}\) towards having few sets. This precise tension is the center of this paper. Generalizing ([enu:cofinal]) we may introduce a third cardinal \(\lambda\leq \theta\) and ask that \(\mathcal{F}\) is \(\lambda\)-cofinal, meaning that every set of size \(<\lambda\) is contained in a member of \(\mathcal{F}\). This leads us to:

Question 1. Given cardinals \(\lambda\leq\theta\leq\kappa\), what is the minimal VC-dimension of a \(\lambda\)-cofinal family \(\mathcal{F}\subseteq[\kappa]^{<\theta}\) (where \([\kappa]^{<\theta}\) is the set of all subsets of \(\kappa\) of cardinality \(<\theta\))?

Consider the family of proper initial segments of \(\omega\). Every finite subset of the natural numbers is contained in such an initial segment, and it can be seen to have VC-dimension 1. Thus, we have that for \(\lambda=\kappa=\theta=\omega\), the answer to 1 is 1. In BBNKS?, Bays, Simon and the first two authors, motivated by questions in model theory, considered a version of 1. Generalizing the example of initial segments of \(\omega\), they proved:

Fact 2. BBNKS? There is an \(\omega\)-cofinal family \(\mathcal{F}\subseteq[\omega_1]^{<\omega}\) with \(\mathop{\mathrm{VC}}(\mathcal{F})=2\).

The following result in their paper demonstrates that the VC-dimension for such families cannot be smaller, and more generally gives a lower bound for a possible answer to 1.

Fact 3. BBNKS? For \(n<\omega\), a cardinal \(\kappa\) and a family \(\mathcal{F}\subseteq[\kappa^{+n}]^{<\kappa}\) which is \((n+2)\)-cofinal we have \(\mathop{\mathrm{VC}}(\mathcal{F})\geq n+1\).

Remark 4. 3 does not appear exactly in this form in BBNKS?: there, the family is required to cover every finite subset of \(\kappa^{+n}\), not just those of size \(n+1\). However, the proof there gives this stronger formulation.

In this paper, we attempt to continue this project. In particular, we ask when the bound of 3 is optimal. The following is the first result proved.

5 (62([enu:regular],[enu:singular])). Fix infinite cardinals \(\lambda\leq \theta\leq \kappa\).

  1. If \(\theta\) is regular and uncountable, then there is a \(\lambda\)-cofinal \(\mathcal{F}\subseteq[\kappa]^{<\theta}\) with \(\mathop{\mathrm{VC}}(\mathcal{F})\leq n\) if and only if \(\kappa<\theta^{+n}\).

  2. If \(\theta\) is singular and \(\mathop{\mathrm{cof}}(\theta)<\lambda\), then every \(\lambda\)-cofinal \(\mathcal{F}\subseteq[\kappa]^{<\theta}\) has \(\mathop{\mathrm{VC}}(\mathcal{F})=\infty\).

In light of 2 3 5, the following two questions are therefore natural:

Question 6. \({}\)

  1. What can we say for \(\theta=\omega\) and \(\kappa=\omega_n\) for \(n>1\)?

  2. What can we say for singular \(\theta\), \(\kappa=\theta^{+n}\) and \(\lambda\leq \mathop{\mathrm{cof}}(\theta)\)?

Regarding those two questions, we get the following consistency results.

7 (62([enu:omega],[enu:measurable])). \({}\)

  1. For every \(n<\omega\), it is consistent with ZFC that there is an \(\omega\)-cofinal \(\mathcal{F}\subseteq[\omega_n]^{<\omega}\) with \(\mathop{\mathrm{VC}}(\mathcal{F})=n+1\).

  2. Relative to the existence of a measurable cardinal, it is consistent that there is a singular cardinal \(\theta\) of countable cofinality such that for every \(n<\omega\) there is some \(\omega\)-cofinal \(\mathcal{F}\subseteq[\theta^{+n}]^{<\theta}\) such that \(\mathop{\mathrm{VC}}(\mathcal{F})=n+1\).

The notion of VC-dimension is important in machine learning, where it corresponds to PAC learnability (see MLShaiShai?), and also has a model theoretic counterpart, namely the NIP/IP dichotomy, which we will not go into here, but see Simon? for more. An interesting application is the existence of a plethora of examples of 2-sorted structures with IP, whose language each has a single binary NIP relation.

In their paper, the following was also proved:

Fact 8. BBNKS? Suppose \(\theta\) is an infinite cardinal, \(X\) a set with \(|X|\geq \theta^+\) and \(\mathcal{F}\subseteq[X]^{<\theta}\) is \(\omega\)-cofinal, then the structure \(\mathcal{M}=(X,\mathcal{F},\in)\) has \(IP\), i.e., there is some definable family in \(\mathcal{M}\) with IP (see 10).

Since the structure \((X,\mathcal{F},\in)\) is interpretable in every structure in which \(\mathcal{F}\) is definable, 5 7 together with 8 give us a family of examples (consistently) of NIP families of sets which are not definable in any NIP structure, which would hopefully give us more understanding of local NIP phenomena, and perhaps serve as a source for counterexamples to conjectures regarding such formulas.

This work is part of the third author’s master’s thesis. The complementary work towards this thesis was written in Fin? together with Johanna Steinmeyer, in it a finitary analogue of the above question is considered.

The third author would like to thank the first two authors for guiding him towards the completion of this thesis, and to Uri Abraham for refereeing the thesis with great care, leading to improvements in its presentation, and therefore in the presentation of this paper.

A bit on the proofs.↩︎

The main construction throughout this paper is that of an \(n\)-ordering system, see 16. The idea is to construct systems of compatible well-orders on a cardinal and its initial segments (and each of their initial segments, etc.) and considering sets which are closed relative to these systems, in a certain precise sense (see 21). This idea originated in BBNKS? where this was done for \(n=1\). Having this general definition forces us to prove many properties by induction on \(n\). Accordingly, we made an effort to find the correct definitions so that the proofs are as simple as possible.

2 Preliminaries and ordering systems↩︎

2.1 VC dimension and cofinal families of sets↩︎

9. Given a set \(X\) and a cardinal \(\theta\), write \([X]^{<\theta}\) for the collection of subsets of \(X\) of size less than \(\theta\). Say that \(\mathcal{F}\subseteq[X]^{<\theta}\) is \(\lambda\)-cofinal if every \(A\in[X]^{<\lambda}\) is contained in a member of \(\mathcal{F}\).

10. Fix a family \(\mathcal{F}\subseteq\mathcal{P}(X)\) and a cardinal \(\kappa\leq |X|\).

  1. Say that \(A\subseteq X\) is shattered by \(\mathcal{F}\) if for every \(A_0\subseteq A\) there is some \(S_0\in\mathcal{F}\) such that \(A_0=A\cap S_0\).

  2. The VC-dimension of \(\mathcal{F}\) is the maximal size of a finite subset of \(X\) which is shattered by \(\mathcal{F}\), or \(\infty\) if this is unbounded. Say that \(\mathcal{F}\) is a VC-class if \(\mathop{\mathrm{VC}}(\mathcal{F})<\infty\).

11. Given cardinals \(\lambda\leq\theta\leq\kappa\), let \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)\) be the least number \(n\in\omega\cup\{\infty\}\) such that there is a \(\lambda\)-cofinal family \(\mathcal{F}\subseteq[\kappa]^{<\theta}\) of VC-dimension \(n\).

We start by investigating basic properties of this function.

Proposition 12. Given \(\mathcal{F}\subseteq \mathcal{P}(X)\) and \(X_0\subseteq X\) we have that any subfamily \(\mathcal{F}_0\) of the family \(\mathcal{F}\restriction X_0=\{X_0\cap S: S\in\mathcal{F}\}\) satisfies \(\mathop{\mathrm{VC}}(\mathcal{F}_0)\leq \mathop{\mathrm{VC}}(\mathcal{F})\). In particular, for every choice of cardinals \(\lambda\leq\theta\leq\kappa\) and \(\lambda_0\leq\theta_0\leq\kappa_0\) with \(\kappa_0\leq\kappa\), \(\lambda_0\leq\lambda\) and \(\theta\leq\theta_0\), we have that \(\mathop{\mathrm{VCcof}}(\lambda_0,\theta_0,\kappa_0)\leq \mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)\).

Proof. The first part is trivial. For the second part, let \(\mathcal{F}\subseteq[\theta]^{<\kappa}\) be \(\lambda\)-cofinal, and apply the first part with \(X=\theta\) and \(X_0=\theta_0\). It is clearly still \(\lambda\)-cofinal, and consists of sets of size less than \(\kappa \leq \kappa_0\), so in particular \(\lambda_0\)-cofinal, giving us the desired inequality. ◻

Restating 3 in these terms, we have:

Fact 13. BBNKS? For every infinite cardinal \(\kappa\) and natural number \(n\), we have \(\mathop{\mathrm{VCcof}}(n+2,\theta,\theta^{+n})\geq n+1\).

From 13 we get the following interesting corollary, which is not used elsewhere in this work. It also follows more directly from Fin?.

Corollary 14. Suppose \(\mathcal{F}\subseteq[\omega]^{<\omega}\) has the following “uniform cofinality” property: there is a function \(f:\omega\rightarrow \omega\) such that given \(A\in[\omega]^n\) there is \(B\in[\omega]^{<f(n)}\) with \(A\subseteq B\) and \(B\in\mathcal{F}\). Then \(\mathop{\mathrm{VC}}(\mathcal{F})=\infty\).

Proof. Fix such a family \(\mathcal{F}\) with a corresponding function \(f:\omega\rightarrow \omega\). consider the two-sorted structure \(\mathcal{M}=(\omega,\mathcal{F},\in)\), where \(\mathord{\in} \subseteq \omega\times\mathcal{F}\) is the incidence relation. Then for every \(n<\omega\), we have that \(\mathcal{M}\models \psi_n\), where: \[\begin{gather} \psi_n:= \;\forall x_0,\ldots,x_{n-1}\in \omega\;\exists y_0,\ldots,y_{f(n)-1}\in\omega\;\exists S\in\mathcal{F}\\ \big\{x_0,\ldots,x_{n-1}\big\}\subseteq S\land \big\{y_0,\ldots,y_{f(n)-1}\big\}=S \end{gather}\]

Assume towards a contradiction that \(\mathop{\mathrm{VC}}(\mathcal{F})=m<\omega\). Let \(\mathcal{N}=(N,\hat{\mathcal{F}})\) be an elementary extension of cardinality \(\aleph_m\). Without loss of generality we have \(N=\aleph_m\). Now, write \(\hat{\mathcal{F}}_{\mathop{\mathrm{fin}}}=\hat{\mathcal{F}}\cap[\aleph_m]^{<\omega}\). Since having VC-dimension \(m\) is a first-order property, we have \(\mathop{\mathrm{VC}}(\mathcal{\hat{F}})=m\), hence \(\mathop{\mathrm{VC}}(\mathcal{\hat{F}}_{\mathop{\mathrm{fin}}})\leq m\). Because \(\mathcal{N}\models \psi_n\) for every \(n<\omega\), we have that \(\hat{\mathcal{F}}_{\mathop{\mathrm{fin}}}\) is cofinal as well, a contradiction to the bound of 13. ◻

The following example is a strong hint towards the ideas to follow.

Examples 15.

  1. We claim that \(\mathop{\mathrm{VCcof}}(\aleph_0,\aleph_0,\aleph_0)=1\). First, the family \(\mathcal{F}=\big\{\{0,\ldots,n-1\}: n<\omega\big\}\) witnesses \(\mathop{\mathrm{VCcof}}(\aleph_0,\aleph_0,\aleph_0)\leq 1\). Indeed, given any finite nonempty \(A\subseteq \omega\) we have \(A\subseteq\sup A+1\), and given any two distinct \(n_1<n_2<\omega\), the set \(\{n_1,n_2\}\) is not shattered by \(\mathcal{F}\), as any \(I\in\mathcal{F}\) with \(n_2\in I\) has \(n_1\in I\) as well, so \(\mathop{\mathrm{VC}}(\mathcal{F})<2\). Since any family with at least two distinct sets has VC-dimension at least 1, we have \(\mathop{\mathrm{VC}}(\mathcal{F})=1\). 12 13 imply that \(1\leq \mathop{\mathrm{VCcof}}(\aleph_0,\aleph_0,\aleph_0)\) as well.

  2. In fact, given any infinite cardinal \(\kappa\), this generalizes to demonstrate that \(\mathop{\mathrm{VCcof}}(\mathop{\mathrm{cof}}(\kappa),\kappa,\kappa)=1\), with the upper bound witnessed by the family of proper initial segments of \(\kappa\).

  3. Note that for every natural number \(n\) and infinite cardinals \(\theta\leq \kappa\), the family \([\kappa]^{n+1}\) witnesses \(\mathop{\mathrm{VCcof}}(n+2,\theta,\kappa)\leq n+1\). In fact, applying 13 and monotonicity gives us an equality whenever \(\kappa\geq\theta^{+n}\).

Our main method for constructing cofinal families will use the following, which was implicitly constructed in BDHSY? and explicitly for the case \(n=2\) in BBNKS?

16. By induction on \(0 < n < \omega\) we define when a function \(\mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}}:[X]^{<n}\rightarrow\mathcal{P}(X^2)\) is an \(n\)-ordering system on a set \(X\). In the following and throughout, we write \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_s\) for \(\mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}}(s)\) and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) for the irreflexive variant, i.e., \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s := \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_s \setminus \{(x,x) : x\in X\}\).

  • For \(n=1\), \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}\) is a 1-ordering system if \(\mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}}_{\emptyset}\) is a (non-strict) well-order on \(X\).

  • For \(n>1\) and any \(X\), \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}\) is an \(n\)-ordering system if it satisfies the following conditions:

    • \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_\emptyset\) is a (non-strict) well-order on \(X\).

    • For any \(x_0 \in X\), the function \(\mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}}^{x_0}:[X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x_0}]^{<n-1}\rightarrow\mathcal{P}(X^2)\) given by \(\mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}}^{x_0}_t = \mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_{t\cup \{x_0\}}}\) is an \((n-1)\)-ordering system on the \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_\emptyset\)-initial segment \(X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x_0} := \{x\in X : x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x_0\}\).

We also define the domain of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\) (or \(\mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_s)\), denoted by \(\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)=\mathord{\operatorname{dom}}( \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_s)\), as usual: \(\{x\in X : (x,x) \in \mathord{ \mathrel{\stackunder[0.1ex]{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}{\rule{1.2ex}{0.1ex}}}_s}\}\).

As an illustration of the idea, consider the following instance of the above definition.

Example 17. A 3-ordering system on a set \(X\) consists of the following:

  • A well-order \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset\) on \(X\).

  • For every \(x\in X\), a well-order \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x\}}\) on the set \(X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x}\) defined above.

  • For every \(x,y\in X\) such that \(y\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x\), a well-order \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x,y\}}\) on the set \(\{z\in X: z\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x\land z\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x\}}y\}\).

Remark 18. By induction on \(n\), if \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\) and \(k<n\), then \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\restriction [X]^{<k}\) is a \(k\)-ordering system.

19. Given an \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on a set \(X\) and a finite sequence of ordinals \(\bar{v}=(v_0,\ldots,v_{n-1})\in \operatorname{\mathbf{Ord}}^n\), say that \(\bar{v}\) bounds the order type of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) if for every \(s\in[X]^{<n}\) we have \(\mathop{\mathrm{otp}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\leq v_{|s|}\).

Our goal is to use \(n\)-ordering systems for various \(n < \omega\) to produce cofinal families of (relatively) small VC dimension.

Remark 20. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\) for \(1<n\). Then, for every set \(s \in [X]^{<n-1}\) and every \(a\in X\) such that \(a\) is in the domain of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\), the domain of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s\cup \{a\}}\) is \(X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s a} = \{x \in X : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s a\}\). The proof is by induction on \(1<n\). If \(s\) is empty this follows directly from 16. This is the only case when \(n=2\) so this covers the base of the induction. Otherwise, let \(x_0 = \max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset} s\) and let \(t=s \setminus \{x_0\}\). Then the domain of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s \cup \{a\}}\) is the same as the domain of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{\{a\} \cup t}\). Since \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) is an \((n-1)\)-ordering system and \(1<n-1\) (since \(s\) is nonempty), by induction said domain is \(\{x \in X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\emptyset} x_0} : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_t a\}\). But clearly, if \(x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_t a\) then \(x \in X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\emptyset} x_0}\) (since \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) is an \((n-1)\)-ordering system on \(X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\emptyset} x_0}\)), so said domain is \(\{x \in X : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_t a\}\) which is by definition \(\{x \in X : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s a\}\) as required.

21. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\). A set \(S \subseteq X\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed if for every \(s\subseteq S\) of size \(|s|=n-1\) and every \(a,b\in \mathord{\operatorname{dom}}(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s})\), if \(b\in S\) and \(a \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s b\) then \(a\in S\).

Example 22. If \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is a 1-ordering system, then a set is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed if and only if it is an initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset\).

23. Let \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) be an \(n\)-ordering system on a set \(X\).

  • Given \(\lambda\leq\theta\), say that \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) has \((\lambda,\theta)\)-closures if every \(A\in[X]^{<\lambda}\) is contained in an \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed set \(B\in[X]^{<\theta}\).

  • Say that \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) has \(\theta\)-closures whenever it has \((\theta,\theta)\)-closures. Say that it has finite closures if it has \(\omega\)-closures.

24. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\) and \(s\in[X]^{<\omega}\). Then for every \(k\leq \min\{n-1,|s|\}\) there is a unique \(s_{k}\in[s]^k\) such that \(s\setminus s_{k}\subseteq \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s_{k}})\).

Proof. We prove this lemma by induction, where the case \(n=1\) is trivial, as is the case of \(s=\emptyset\). For \(n>1\) and \(s\neq\emptyset\) we write \(x_0=\max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}s\). By the induction hypothesis there exists \(t_{k-1}\in[(s\setminus\{x_0\})]^{k-1}\) such that \(s\setminus(\{x_0\}\cup t_{k-1})\subseteq \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{t_{k-1}})=\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{t_{k-1}\cup\{x_0\}})\), which shows that setting \(s_k=t_{k-1}\cup\{x_0\}\) finishes the existence proof. For uniqueness, suppose \(s_k,t_k\in[s]^k\) both satisfy the conclusion of the lemma. We will first claim that \(s_k,t_k\) must contain \(x_0\). Indeed, for any nonempty \(t\subseteq s\) we have \(x_0\notin\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_t)\subseteq X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset \max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}t}\subseteq X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x_0}\), so the desired inclusion cannot hold otherwise. Write \(s_k'=s_k\setminus\{x_0\}\) and \(t_k'=t_k\setminus \{x_0\}\). Note that we have \(s\setminus\{x_0\}\subseteq \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{x_0})\), so we have \(s\setminus s_k\subseteq \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{s_k'})\subseteq X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset x_0}\) and \(s\setminus t_k\subseteq\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{t_k'})\). Applying the induction hypothesis proves \(s_k'=t_k'\), which completes the proof of uniqueness. ◻

Proposition 25. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\). Then the collection \(\mathcal{F}\) of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed sets has VC-dimension at most \(n\).

Proof. Suppose that \(s\subseteq X\) has size \(n+1\). Apply 24 to get \(s'\in[s]^{n-1}\) such that \(s\setminus s'\subseteq \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'})\), and write \(\{a,b\}=s\setminus s'\) As \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'}\) is a linear order, either \(a \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'} b\) or \(b \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'} a\). Assume without loss of generality that the former holds. Then every closed set containing \(s',b\) also contains \(a\) by 21, so \(s\) cannot be shattered. ◻

The following is a main ingredient in all that follows.

Corollary 26. Suppose \(\lambda\leq\theta \leq \kappa\) are infinite cardinals and \(n<\omega\) satisfy that there exists an \((n+1)\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(\kappa^{+n}\) with \((\lambda,\theta)\)-closures, then the family \(\mathcal{F}\) of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed sets of size less than \(\theta\) has VC-dimension at most \(n+1\). In particular, it witnesses \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa^{+n})\leq n+1\). From 13 and 12 it follows that \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa^{+n})= n+1\).

Proof. Fix such an ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\). By 25, the collection of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed sets has VC-dimension at most \(n+1\). By 12 so does \(\mathcal{F}\), and by having \((\lambda,\theta)\)-closures this family is \(\lambda\)-cofinal. ◻

In BBNKS?, Bays, Simon and the first two authors use this idea in order to prove the following.

Fact 27 (BBNKS?,Theorem 3.8). There is a 2-ordering system on \(\omega_1\) of order type bounded by \((\omega_1, \omega)\) (see 19) with finite closures, so in particular, \(\mathop{\mathrm{VCcof}}(\omega,\omega,\omega_1)=2\).

3 Closed initial segments↩︎

Recall from 21: if \(0<n\) and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on \(X\) then a set \(S \subseteq X\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed if for every \(s\subseteq S\) of size \(|s|=n-1\) and every \(a,b\in \mathord{\operatorname{dom}}(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s})\), if \(b\in S\) and \(a \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s b\) then \(a\in S\). Given \(s \subseteq X\) of size \(n-1\) and \(b \in X\), call the set \(s \cup \{b\} \cup \{a : a \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s b\}\) the closed \(n\)-initial segment for \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) determined by \(s \cup \{b\}\). Note that when \(n=1\), the closed 1-initial segment determined by \(b\) is precisely the closed initial segment \(X_{{\mathrel{\mathpalette\pr@ceqd@t\relax}_\emptyset} b} = \{x\in X : x\mathrel{\mathpalette\pr@ceqd@t\relax}_\emptyset b\}\) (see 22 as well).

If \(S\) is finite of size at least \(n\), then \(S\) is contained in a closed \(n\)-initial segment: let \(\beta_n = \max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset} S\), \(\beta_{n-1}= \max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\beta_n\}}} S \setminus \{\beta_n\}\), …, \(\beta_1 = \max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\beta_i : 1<i\leq n\}}} S \setminus \{\beta_i : 1<i\leq n\}\), then \(S\) is contained in the closed \(n\)-initial segment determined by \(\beta_n,\ldots,\beta_1\). Indeed, this follows from 20. Moreover, if \(S\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed, then the procedure above shows that \(S\) itself is a closed \(n\)-initial segment.

Consider 27: the 2-ordering system there on \(\omega_1\) has order type bounded by \((\omega_1,\omega)\). This means in particular that closed 2-initial segments are finite. On the other hand, by the previous paragraph all \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed finite sets of size at least 2 are closed 2-initial segments. This naturally raises the question: can we improve 27 so that closed 2-initial segments are \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed implying that the two notions coincide (and in particular the 2-ordering system will have finite closures)? This section is dedicated to answering this question by finding a 2-ordering system on \(\omega_1\) for which closed 2-initial segments are \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed. Together with 25 this answers positively the first part of BBNKS?.

28. There is a 2-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(\omega_1\) of order type bounded by \((\omega_1,\omega)\) in which all closed 2-initial segments are \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed. In particular, it has finite closures.

Proof. Given an ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on a set \(X\) and \(A\subseteq X\), \(x\in X\), write \(A\vdash x\) if any \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed set \(S\) which contains \(A\) must contain \(x\).

By induction on \(\alpha<\omega_1\), we will recursively construct a sequence of 2-ordering systems \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\) on \(\alpha +1\) for \(\alpha < \omega_1\) such that the following hold:

  1. \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha_\emptyset\) is the usual ordering on \(\alpha+1\).

  2. \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha_{\{\beta\}} = \prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha'}_{\{\beta\}}\) whenever \(\beta \leq \alpha < \alpha'\).

  3. \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha_{\{\beta\}}\) is of order type \(|\beta|\) for all \(\beta\leq \alpha\).

  4. All closed 2-initial segments of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\) are \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed.

  5. Given a finite and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed set \(S\subseteq \alpha\) and \(\beta,\gamma\in \alpha\setminus S\) distinct, either \(S \cup \{\gamma\}\not\vdash\beta\) or \(S\cup \{\beta\}\not\vdash\gamma\).

If we succeed, then define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(\omega_1\) by setting \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset\) as the usual ordering on \(\omega_1\) and \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{\{\alpha\}} = \mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{\{\alpha\}}^\alpha\) for \(\alpha<\omega_1\) to conclude: ([IC]) tells us that all closed 2-initial segments are \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed.

In the construction, we only explain how to define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha_{\{\alpha\}}\) since ([itm:usual32ordering]) and ([itm:increasing]) determine the rest. Thus, to simplify notation, we will write \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}}\) instead of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha_{\{\alpha\}}\).

Define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{0\}}\) to be the empty ordering, then clearly it satisfies this.

We start with the successor step. Assuming we have defined \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\beta\) for all \(\beta \leq \alpha<\omega_1\) such that the above hold, define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha+1\}}=\{(\alpha,\beta):\beta<\alpha\}\cup \mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{\{\alpha\}}\) (i.e., leave the ordering \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}}\) on \(\alpha\) unchanged and add \(\alpha\) as the first element). We claim that our assumptions still hold.

([itm:usual32ordering]) is clear.

([itm:order32type]): We have to show that the order type of \(\prec_{\{\alpha+1\}}\) is \(|\alpha+1|\). This follows directly by the construction and the induction hypothesis.

([IC]): Suppose that \(D\) is a closed 2-initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha+1}\), i.e., has the form \(\{x<\alpha+2 : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\beta\}} \gamma\} \cup \{\beta,\gamma\}\) for some \(\gamma<\beta<\alpha+2\). If \(\beta<\alpha+1\), then \(D\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha}\)-closed by induction, so also \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha+1}\)-closed. If \(\gamma = \alpha\) (i.e., the minimal element in \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha+1\}}\)) then \(D = \{\beta, \gamma\}\) so it is trivially \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha+1}\)-closed, so we may assume that \(\beta=\alpha+1,\gamma \neq \alpha\). From this and the definition of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha+1\}}\), it follows that \(D \cap (\alpha+1)\) is the closed 2-initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\) determined by \(\alpha,\gamma\), so it is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed by induction.

We are given \(\gamma_2,\delta <\gamma_1\leq \alpha+1\) such that \(\gamma_2,\gamma_1\in D\) and \(\delta\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\gamma_1\}} \gamma_2\) and we have to show that \(\delta \in D\). If \(\gamma_1 = \alpha+1\), this follows directly from the definition of \(D\), so we may assume \(\gamma_1\leq \alpha\). In this case \(\gamma_2,\gamma_1 \in D \cap (\alpha+1)\), so \(\delta \in D \cap (\alpha+1)\) since it is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed, as required.

([dash]): Suppose \(S\subseteq\alpha+1\) is finite and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha+1}\)-closed, and \(\beta,\gamma\in\alpha+1\setminus S\) are distinct and \(\gamma<\beta\).

If \(S=\emptyset\) then ([dash]) holds since singletons are closed. If \(S\cup\{\beta\}\subseteq\alpha\) then ([dash]) follows from the induction hypothesis. Otherwise, we either have \(\alpha\in S\) or \(\beta=\alpha\).

  • If \(\alpha\in S\) and \(\beta,\gamma\in\alpha+1\setminus S\), we must have \(\beta,\gamma<\alpha\). Suppose without loss of generality that \(\beta\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}}\gamma\). Note that as \(S\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha+1}\)-closed and \(\beta \notin S\), \(S \subseteq \{x<\alpha : x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}} \beta\} \cup \{\alpha\}\). Thus, the set \(\{x<\alpha : x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}} \beta\} \cup \{\alpha,\beta\}\) is a closed 2-initial segment of \(\prec^{\alpha}\) not containing \(\gamma\). By ([IC]) applied to \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha}\), this set witnesses that \(S\cup\{\beta\}\not\vdash \gamma\).

  • Otherwise we have \(\beta=\alpha\). Take some initial segment \(D\) of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}}\) which contains \(\gamma\) and \(S\). Note that \(D \cup \{\alpha\}\) is a closed 2-initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\) so \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed by induction. It follows that \(D\) is also \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed (clearly if \(\gamma_1,\gamma_2 \in D\) and \(\delta \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\gamma_1\}} \gamma_2\) then \(\delta <\alpha\)). Thus, \(D\) witnesses that \(S\cup \{\gamma\}\not\vdash \beta\).

We are left with the limit stage of the construction. Suppose \(\delta<\omega_1\) is a limit ordinal and we constructed \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\) for all \(\alpha<\delta\). As above, we only need to define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\) on \(\delta+1\).

Fix a bijection \(f:\omega\rightarrow\delta\). Let \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*}\) be a 2-ordering system on \(\delta\) determined by \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*_\emptyset\) being the usual order on \(\delta\) and for each \(\alpha < \delta\), \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*_{\{\alpha\}}} = \mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{\alpha}_{\{\alpha\}}}\). Note that ([itm:usual32ordering])–([dash]) hold for \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\). Define an increasing union of finite \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed sets \((S_n)_{n<\omega}\) recursively as follows. Take \(S_0=\emptyset\). Suppose we have defined the \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed sets \((S_k)_{k\leq n}\) such that each \(S_k\) contains \(f(0),\ldots,f(k-1)\) and \(S_j\) for \(j<k\). Let \(\alpha < \delta\) be such that \(\alpha+1\) contains \(S_n\). By ([IC]) (and ([itm:order32type]), i.e., the fact that closed 2-initial segments are finite), there is some finite \(S_{n+1} \subseteq \alpha+1\) which is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\alpha\)-closed (so also \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed) and contains \(S_n\) and \(f(n)\). Note that we have \(\bigcup_{n<\omega}S_n=\delta\).

Fix some \(n<\omega\). Suppose that \(m = |S_{n+1} \setminus S_n|\). By induction on \(i\leq m\) define a strictly increasing sequence of injective functions (enumerations) \(f_i:i\to S_{n+1} \setminus S_n\) such that for all \(i<m\), \(S_n \cup \mathop{\mathrm{Im}}(f_i)\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed (where \(\mathop{\mathrm{Im}}(f_i)\) is the image of \(f_i\)).

For \(i=0\), this is possible since \(S_n\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed. Given \(f_i\) for \(i<m\), define the following binary relation on \(S_{n+1}\setminus \mathop{\mathrm{Im}}(f_i)\): \(x \preceq y\) if and only if \(S_n \cup \mathop{\mathrm{Im}}(f_i) \cup \{y\} \vdash x\). Note that \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is antisymmetric: if \(x\preceq y\) and \(y \preceq x\) then \(x=y\) by ([dash]) (applied to the \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed set \(S_n \cup \mathop{\mathrm{Im}}(f_i)\)). It is also clearly transitive, so it is a partial order on a finite set and thus has a minimal element \(x_0\), and let \(f_{i+1}(i)=x_0\). Let us show that \(S_n \cup \mathop{\mathrm{Im}}(f_{i+1}) = S_n \cup \mathop{\mathrm{Im}}(f_i) \cup {\{x_0\}}\) is \(\prec^*\)-closed. Indeed, define the closure \(\mathop{\mathrm{cl}}(S_n \cup \mathop{\mathrm{Im}}(f_{i+1}))\) as the intersection of all \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed sets containing it (note that \(S_{n+1}\) is closed, so this closure is finite and contained in \(S_{n+1}\)). Clearly \(\mathop{\mathrm{cl}}(S_n \cup \mathop{\mathrm{Im}}(f_{i+1}))\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed and contains \(S_n \cup \mathop{\mathrm{Im}}(f_{i+1})\). If \(y \in \mathop{\mathrm{cl}}(S_n \cup \mathop{\mathrm{Im}}(f_{i+1})) \setminus (S_n \cup \mathop{\mathrm{Im}}(f_{i+1}))\) then \(y \preceq x_0\) and \(y \neq x_0\), contradicting the minimality of \(x_0\). Thus, \(\mathop{\mathrm{cl}}(S_n \cup \mathop{\mathrm{Im}}(f_{i+1})) = S_n \cup \mathop{\mathrm{Im}}(f_{i+1})\) and the latter is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed as required.

Being done with the inductive construction, let \(f^n = \bigcup_{i\leq m} f_i\); it is an enumeration of \(S_{n+1} \setminus S_n\). For \(x <\delta\), let \(n_x = \max \{n<\omega : x\notin S_n\}\).

Finally define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\) by \(x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}} y\) if and only if \(n_x < n_y\) or \(n=n_x = n_y\) and \((f^{n})^{-1}(x) < (f^{n})^{-1}(y)\) (i.e., \(x\) appears before \(y\) in the enumeration given by \(f^n\)). Note that any initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed.

Let us check that ([itm:usual32ordering])–([dash]) hold. ([itm:usual32ordering]), ([itm:increasing]) are clear and ([itm:order32type]) holds since every initial segment of the linear order \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\) is finite (so that the order type of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\) is \(\omega\)).

We show ([IC]). Suppose that \(D = \{x<\delta : x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\alpha\}} \beta\} \cup \{\alpha,\beta\}\) for some \(\beta<\alpha\leq \delta\) is a closed 2-initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\delta\). If \(\alpha<\delta\), \(D\) is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed (so also \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\delta\)-closed) by the induction hypothesis, so suppose that \(\alpha = \delta\). We are given \(\gamma_2,\varepsilon<\gamma_1 \leq \delta\) such that \(\gamma_1,\gamma_2 \in D\) and \(\varepsilon\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\gamma_1\}} \gamma_2\) and we have to show that \(\varepsilon \in D\). If \(\gamma_1 = \delta\), this follows directly from the definition of \(D\), so we may assume that \(\gamma_1 < \delta\). But since \(D \cap \delta\) is an initial segment of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{\delta\}}\), it is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^*\)-closed, so \(\varepsilon \in D\) as required.

Finally we show ([dash]). Suppose that \(S \subseteq \delta\) is finite and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\delta\)-closed, and \(\alpha\neq \beta \in \delta \setminus S\). Then, \(S \cup \{\alpha,\beta\} \subseteq \gamma < \delta\), and hence this follows from the induction hypothesis applied to \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^\gamma\). ◻

4 Large cofinality and uncountable cardinals↩︎

Recall 11: Given cardinals \(\lambda\leq\theta\leq\kappa\), \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)\) is the least number \(n\in\omega\cup\{\infty\}\) such that there is a \(\lambda\)-cofinal family \(\mathcal{F}\subseteq[\kappa]^{<\theta}\) of VC-dimension \(n\). In this section we study the case of an uncountable bound \(\theta\) on the size of the members (i.e. the second parameter of \(\mathop{\mathrm{VCcof}}\)).

We prove that when \(\lambda\leq\theta\leq\kappa\) are infinite cardinals, \(n<\omega\) and \(\theta\) uncountable then the behavior of \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)\) depends on whether \(\theta\) is regular or singular.

In the regular case we get an optimal result: \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)=n+1\) if and only if \(\kappa=\theta^{+n}\). In particular, \(\lambda\) is not restricted.

In the singular case, we show that in VC-classes the cofinality parameter \(\lambda\) (the first parameter of \(\mathop{\mathrm{VCcof}}\)) in a VC-class can be at most \(\mathop{\mathrm{cof}}(\theta)\). It is not clear what can be said about \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)\) when \(\lambda \leq \mathop{\mathrm{cof}}(\theta)\), and see 36 for a precise formulation.

However, we do show in 35 that relative to the existence of a measurable1 cardinal it is consistent that there is an uncountable cardinal \(\theta\) of cofinality \(\omega\) such that for every \(n<\omega\), \(\mathop{\mathrm{VCcof}}(\omega,\theta,\theta^{+n})=n+1\).

Notation 29. Given a \(k\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on a set \(X\) and \(s\in[X]^{<n}\), write \(\mathord{\operatorname{dom}}(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s})\) for the domain of the ordering \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\) (so a subset of \(X\)).

We now have a useful lemma, which will be used at its full generality in the next section.

30. Suppose \(\theta\) is an infinite cardinal, \(n<\omega\) and \(X\) is a set satisfying \(|X|\leq \theta^{+n}\). Then there exists an \((n+1)\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(X\) such that \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_\emptyset=\mathord{<}_X\) and \((\theta^{+n},\ldots,\theta)\) bounds the order type of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) (see 19).

Proof. By induction on \(n\). We first set \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_\emptyset\) to be some linear order on \(X\) of order type \(|X|\). If \(n=0\) then we are done. Otherwise, since the set \(X_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\emptyset}x_0}\) has cardinality at most \(\theta^{+(n-1)}\) for every \(x_0\in X\), we can apply the induction hypothesis to get an \(n\)-ordering systems \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) on each of these sets of order type bounded by \((\theta^{+(n-1)},...,\theta)\), and so for \(s\in [X]^{<n}\setminus\{\emptyset\}\) we define \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{s}=\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{x_0}_{s\setminus \{x_0\}}\), where \(x_0=\max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}s\). ◻

Proposition 31. Fix an uncountable regular cardinal \(\theta\) and a natural number \(n<\omega\), then any \((n+1)\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(\theta^{+n}\) with \((\theta^{+n},\ldots,\theta)\) bounding its order type has \(\theta\)-closures (see 23). In particular, we have \(\mathop{\mathrm{VCcof}}(\theta,\theta,\theta^{+n})=n+1\).

Proof. Fix such an ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) and fix \(A\in[\theta^{+n}]^{<\theta}\). Define \(A=A_0\), and given that we have defined \(A_k\) for some \(k<\omega\), define:

\[A_{k+1}=A_k\cup\bigcup\left\{\{x\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s \alpha\}:\alpha\in \mathord{\operatorname{dom}}\left(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\right)\cap A_k,s\in [A_k]^n\right\}.\]

By definition we have \(|A_0|<\theta\), and by induction, the order type bound assumption and the regularity of \(\theta\), it follows that \(|A_{k+1}|<\theta\). Take \(A_\omega=\bigcup_{k<\omega}A_k\) hence \(|A_\omega|<\theta\) by the regularity and uncountability of \(\theta\). Clearly it is \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\)-closed. Hence \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) has \(\theta\)-closures. For the second part, apply 26 [exists] to the first part. ◻

We move on to the case of singular cardinals.

32. Suppose \(\theta\) is singular of cofinality \(\lambda\) and \(\mathcal{F}\subseteq[\theta]^{<\theta}\) is \(\lambda^+\)-cofinal. Then there is some infinite cardinal \(\kappa<\theta\) and \(\mathcal{F}_0\subseteq\mathcal{F}\) with \(\mathcal{F}_0\subseteq[\theta]^{<\kappa}\) that is \(\lambda^+\)-cofinal.

Proof. Assume otherwise. Let \((\kappa_\alpha:\alpha<\lambda)\) be a strictly increasing sequence of cardinals with \(\bigcup_{\alpha<\lambda}\kappa_\alpha=\theta\). Then there are \(A_\alpha\in[\theta]^{<\lambda^+}\) for \(\alpha<\lambda\) such that the minimal cardinality of a set \(B_\alpha\in\mathcal{F}\) with \(A_\alpha\subseteq B_\alpha\) is greater than \(\kappa_\alpha\). Let \(A=\bigcup_{\alpha<\lambda}A_\alpha\), so it is of size \(\leq \lambda\) as well, hence there is some \(B\in\mathcal{F}\) with \(A\subseteq B\). Since \((\kappa_\alpha:\alpha<\lambda)\) is cofinal in \(\theta\) and \(|B|<\theta\), there is some \(\alpha<\lambda\) with \(|B|<\kappa_\alpha\). But \(A_\alpha\subseteq A\subseteq B_\alpha\), a contradiction to our choice of \(A_\alpha\). ◻

Corollary 33. Suppose \(\theta\) is singular of cofinality \(\lambda\). Any \(\lambda^+\)-cofinal \(\mathcal{F}\subseteq[\theta]^{<\theta}\) has infinite VC-dimension.

Proof. Let \(\mathcal{F}\) be as above. Apply 32 to get some \(\kappa\) and \(\mathcal{F}_0\subseteq\mathcal{F}\) as there. Since \(\theta\) is singular, it is a limit cardinal, and because \(\kappa<\theta\) we get \(\kappa^{+n}<\theta\) for every \(n<\omega\). Let \(\mathcal{F}_0' = \{s \cap \kappa^{+n} : s\in \mathcal{F}_0\}\). Assuming towards a contradiction that \(\mathop{\mathrm{VC}}(\mathcal{F})=n<\omega\), and in particular \(\mathop{\mathrm{VC}}(\mathcal{F}_0')\leq n\) by 12. This gives us \(\mathop{\mathrm{VCcof}}(\omega,\kappa,\kappa^{+n})<n+1\), a contradiction to 13. ◻

Therefore, for a singular \(\theta\), we have a critical point \(\lambda=\mathop{\mathrm{cof}}(\theta)\) which would be of interest.

We now demonstrate that consistently (assuming the consistency of a measurable cardinal), at least for some cardinal this is indeed the critical value.

Recall the following celebrated result of Prikry, originating in Prikry?.

Fact 34. Gitik? Assume the existence of a measurable cardinal \(\theta\). Then there exists a forcing notion \(\mathbb{P}\) such that the following hold:

  1. \(\mathbb{P}\) preserves cardinals.

  2. \(\Vdash_{\mathbb{P}}\theta\text{ is uncountable of cofinality \omega}\).

Proposition 35. Assume the consistency of a measurable cardinal. Then it is consistent that there is an uncountable cardinal \(\theta\) of cofinality \(\omega\) such that for every \(n<\omega\) we have \(\mathop{\mathrm{VCcof}}(\omega,\theta,\theta^{+n})=n+1\).

Proof. Let \(\theta\) be measurable and let \(\mathbb{P}\) be the forcing notion of 34. Since \(\theta\) is measurable, it is regular, so by 31 for every \(n<\omega\) there is some \(\mathcal{F}_n\) witnessing \(\mathop{\mathrm{VCcof}}(\theta,\theta,\theta^{+n})=n+1\). Let \(G\subseteq\mathbb{P}\) be a generic filter and fix \(n<\omega\). By 34 the cardinal \(\theta\) is singular and of cofinality \(\omega\) in \(V[G]\), and the \(n\)’th successor of \(\theta\) in \(V[G]\) is its \(n\)’th successor in \(V\), so it remains to show that \(\mathcal{F}_n\) is \(\omega\)-cofinal and \(\mathop{\mathrm{VC}}(\mathcal{F}_n)=n+1\). For the first of these, forcing cannot add a new finite subset to an existing set, so by the \(\theta\)-cofinality of \(\mathcal{F}_n\) in \(V\) we get \(\omega\)-cofinality in \(V[G]\). The latter follows from absoluteness, hence we are done. ◻

Despite of this consistency result, the general case of \(\lambda\) in the range \(\omega\leq \lambda\leq \mathop{\mathrm{cof}}(\theta)\) remains to be considered.

Question 36. Is it true that for every singular cardinal \(\theta\) and \(n<\omega\) we have \(\mathop{\mathrm{VCcof}}(\mathop{\mathrm{cof}}(\theta),\theta,\theta^{+n})=n+1\)? Is it consistent?

5 Cohen-generic ordering systems on \(\omega_n\)↩︎

Given the second result mentioned in [unctbl], we would like to see whether there exists a cofinal family \(\mathcal{F}\subseteq[\omega_n]^{<\omega}\) of VC-dimension \(n+1\). The goal of this section is to establish its consistency.

So far, to construct \((n+1)\)-ordering systems, we constructed it inductively, first choosing \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset\) and then defining \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\) for nonempty \(s\) by an inductive construction. Here we start with an \(n\)-ordering system and use forcing to construct \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s\) for \(s\) of size \(n\). The domain of these new orders can be already defined without forcing, and we want to define these in a uniform way that allows us to argue about these new orders as well as old orders (where the domain is already defined). This motivates the definition of a “pre-domain” which we now explain.

37. Given an \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on a set \(X\) and \(s\in[X]^k\) for \(0<k\leq n\), write \(\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}} s\) for the unique element of the set \(s\setminus s_{k-1}\), where \(s_{k-1}\) is the subset given by 24. Namely, it is the unique \(y\in s\) such that \(y\in\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s\setminus\{y\}})\).

Notation 38. Given an \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on a set \(X\) and given \(s\in[X]^{\leq n}\), if \(s\neq\emptyset\), we write \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s):=\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'})_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'}\mathord{\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}} s} = \{x\in \mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'}) : x \prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'} \mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}} s\}\), and if \(s=\emptyset\), we write \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset)=X\).

Remark 39. By applying [rem:domain of <<s union a,Lemma: inductive maximum], if \(s\in[X]^k\) for \(k<n\) then we have \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)=\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\).

We have the following lemma.

40. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\) and \(x_0\in X\). Then for any \(t\in [\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})]^{<n}\) we have \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_t)=\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{t\cup\{x_0\}})\).

Proof. We may assume \(t\neq \emptyset\). By the definition of \(\mathord{\operatorname{predom}}\), the lemma would follow once we prove that \(\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(t\cup\{x_0\})=\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}}t\). Writing \(m\) for the latter, we get: \[m\in\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{t\setminus\{m\}})=\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{(t\cup\{x_0\})\setminus\{m\}}),\] which is the definition of \(\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}}t\). ◻

We immediately get the following.

Corollary 41. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is an \(n\)-ordering system on a set \(X\), and for every \(s\in[X]^n\) we are given a well-order \(<_s\) on the set \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\). Then the map \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}':[X]^{\leq n}\rightarrow \mathcal{P}(X^2)\) which extends \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) and assigns \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}'(s)=\mathord{<}_s\) to any \(s\in[X]^n\) is an \((n+1)\)-ordering system on \(X\).

Proof. By induction on \(n\), with the case of \(n=1\) being the definition of a 2-ordering system. Suppose \(n>1\) and fix \(x_0\in X\). In order to show that \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}'\) is an \((n+1)\)-ordering system on \(X\), we must show that \((\prec\mathrel{\mkern-5mu}\mathrel{\cdot}')^{x_0}\) is an \(n\)-ordering system on \(\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{x_0})\).

Now, the assignment \((\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0})':[\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})]^{<n-1}\rightarrow \mathcal{P}([\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0})]^2)\) extending \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) and assigning to each \(t\in[\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})]^{n-1}\) the orderings \(\mathord{<'}_t:=\mathord{<}_{t\cup\{x_0\}}\) on each \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{t\cup \{x_0\}})\) (which is equal to \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_t)\) by 40) is an \(n\)-ordering system on \(\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})\) by the induction hypothesis, which is exactly the assignment \((\prec\mathrel{\mkern-5mu}\mathrel{\cdot}')^{x_0}\). Thus the proof is complete. ◻

We note that if there is a uniform bound \(\alpha\) on the order type of the orderings \(<_s\), and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) has order type bounded by \((\alpha_{n-1},...,\alpha_0)\) (see 19), then \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}'\) as constructed in 41 has order type bounded by \((\alpha_{n-1},...,\alpha_0,\alpha)\).

42. Let \((X,<^X)\) be a well-ordered set. For an ordinal \(\alpha\leq \omega\), write \(\alpha^{<^X}\) for the first \(\alpha\) elements of \(<^X\) (or all of \(X\) if \(\mathop{\mathrm{otp}}(<^X)<\alpha\)). An \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(X\) is nice if it satisfies the following:

  1. We have \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}=\mathord{<^X}\).

  2. Writing \(\alpha=\min\{|\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)|,\omega\}\) we have \((\alpha^{<^X},<^X)=(\alpha^{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s},\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) (or equivalently, \(\alpha^{<^X}\) is contained in \(\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) as an initial segment and \(<^X\restriction \alpha^{<^X}=\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s}\restriction\alpha^{<^X}\)).

Remark 43. Note that for any nice \(n\)-ordering system on \((X,<^X)\) and \(x_0\in X\), writing \(\alpha=\min\{|\mathord{\operatorname{dom}}(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{\{x_0\}})|,\omega\}\), since \(\alpha^{<^X}=\alpha^{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}}}\) and the two orderings \(<^X,\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}}\) agree on this set, we have that \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) is a nice \((n-1)\)-ordering system on \((X_{<^X x_0},\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})\).

The following proof is an elaboration of [exists].

44. Suppose \((X,<^X)\) is an infinite well-ordered set satisfying \(\mathop{\mathrm{otp}}(\mathord{<^X})\leq \omega_n\) for some \(n>0\). Then there exists a nice \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \(X\) with \((\omega_n,\ldots,\omega_1)\) bounding the order type of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) (see 19).

Proof. By induction on \(n\). We set \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}=\mathord{<^X}\). If \(n=1\) we are done. Otherwise, since the initial segment \(X_{<^X x_0}\) has cardinality at most \(\aleph_{n-1}\geq \aleph_1\) for any \(x_0\in X\), we fix a well-ordering \(<_{x_0}\) of order type at most \(\omega_{n-1}\) on each \(X_{<^X x_0}\), such that \(\omega^{<^X}\cap X_{<^{X} x_0}\) is an initial segment of \((X_{<^{X} x_0},<_{x_0})\) with both \(<^X,<_{x_0}\) agreeing on \(\omega^{<^X}\cap X_{<^{X} x_0}\). Thus, we have that \(\omega^{<_{x_0}}\) is an initial segment of \(\omega^{<^X}\). Applying the induction hypothesis to \((X_{<^X x_0},<_{x_0})\) for every \(x_0\) gives us nice \((n-1)\)-ordering systems \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}\) on each of these sets of order type bounded by \((\omega_{n-1},...,\omega_1)\), and so for \(s\in [X]^{<n}\setminus\{\emptyset\}\) we define \(\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s}=\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{s\setminus\{x_0\}}}\), where \(x_0=\max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset}s\). ◻

We will start with a lemma.

45. Suppose \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is a nice \(n\)-ordering system on a well-ordered set \((X,<^X)\). For \(s\in[X]^{\leq n}\setminus\{\emptyset\}\) we have that \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is finite if and only if \(s\cap\omega^{<^X}\neq\emptyset\). In this case we have: \[\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)=\{x\in X:x<^X\min(s\cap\omega^{<^X})\}.\]

Proof. Fix \(s\in[X]^{\leq n}\setminus\{\emptyset\}\). The case of \(|s|=1\) is immediate, so assume \(|s|>1\) and write \(x_0=\max_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_\emptyset} s\). We also write \(k=\min\{|\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{\{x_0\}})|,\omega\}\).

We will prove the lemma by induction on \(n\), where the case \(n=1\) follows from the case \(|s|=1\).

For \(n>1\) we have \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}_{s\setminus\{x_0\}})=\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) by 40. By the induction hypothesis and since \(k^{<^X}=k^{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{x_0}}\) by niceness, we first of all get that if the former is finite then \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)=\left\{x\in X: x<^X \min((s\setminus\{x_0\})\cap \omega^{<^X})\right\}\) (where here we use that niceness gives \(\mathord{<}^X=\mathord{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}_{x_0}\) on \(\omega^{<^X}\)). We also get that \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is finite if and only if \((s\setminus\{x_0\})\cap \omega^{<^X}\neq\emptyset\). We now finish both parts by noting that if \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is finite then since \(|s|\geq 2\), any \(y\in s\setminus\{x_0\}\) must satisfy \(y\prec\mathrel{\mkern-5mu}\mathrel{\cdot}x_0\) by maximality, implying both that \(\min((s\setminus\{x_0\}\cap\omega^{<^X})=\min(s\cap \omega^{<^X})\) and that if \(s\cap\omega^{<^X}\neq\emptyset\) then \(s\setminus\{x_0\}\cap\omega^{<^X}\neq\emptyset\). This finishes the proofs of both parts of the lemma. ◻

46. In the context of 45, suppose \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is infinite. Then it contains \(\omega^{<^X}\).

Proof. Write \(y=\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}s\) and \(s'=s\setminus\{y\}\). By definition we have that \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is an initial segment of the ordering \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'}\) on the set \(\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'})\), which contains \(\omega^{<^X}\) as an initial segment of order type \(\omega\) by niceness. Since any infinite initial segment of a well-order contains its first \(\omega\) elements, the result follows. ◻

Throughout the rest of this subsection, fix some \(2\leq n<\omega\), and a nice \(n\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) on \((\omega_n,<)\) (see 42) whose order type is bounded by \((\omega_n,...,\omega_1)\) (19), where \(<\) is the usual ordering on \(\omega_n\). Note that in particular we have \(\omega=\omega^{<}\).

Notation 47. Define the following (where the second equality follows from 45): \[\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}=\{s\in[\omega_n]^{n}:|\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)|\geq \aleph_0\}=\{s\in[\omega_n]^n:s\cap\omega=\emptyset \}.\]

We will now prove another lemma.

48. For any \(s\in[\omega_n]^n\) we have \(|\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)|\leq \aleph _0\). In particular, if \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) then \(|\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)|=\aleph_0\).

Proof. Given \(s\in[\omega_n]^{n}\) we write \(y=\mathop{\mathrm{MIN}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}s\) and \(s'=s\setminus\{y\}\). By the assumption on the bound of the order type of \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) we have \(\mathop{\mathrm{otp}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_{s'})\leq \omega_1\) hence every initial segment is at most countable. ◻

Notation 49. Given sets \(X,Y,Z\), define a partial function \(f:X\times Y\rightharpoonup Z\) and some \(y\in Y\), write \(f_y:X\rightharpoonup Z\) for the function \(f_y(x)=f(x,y)\). We will often not distinguish between a partial function and its graph. Because we often deal with partial functions from a product of sets, an element of the domain will be a tuple. In order to simplify notation, when writing an element of the graph of a partial function we will separate the elements of the domain from those of the range by a semicolon. That is, if \(f:\prod_{i=1}^n X_i\rightharpoonup Y\), \(x_i\in X_i\) and \(y\in Y\), we will write \((x_1,...,x_n;y)\in f\) to mean \((x_1,...,x_n)\in\mathord{\operatorname{dom}}(f)\) and \(f(x_1,...,x_n)=y\).

Remark 50. This section of the paper uses forcing. We use the Jerusalem forcing convention, under which \(p\leq q\) means that \(q\) strengthens \(p\).

51. Write \(\mathcal{Q}\) for the set consisting of the finite partial functions \(p:\bigcup_{s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}}\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\times\{s\}\rightharpoonup\omega\) such that for every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\), the partial function \(p\restriction \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\times\{s\}\) is injective. We consider it as a forcing notion by equipping it with the partial ordering of inclusion of partial functions.

By 48 we have that \(Q\) is isomorphic to the poset finite partial functions \(p:\omega\times\omega_n\rightharpoonup\omega\) which are injective in one coordinate, equipped with the extension ordering. Thus, the following Lemma is standard and extends Kunen?.

52. The forcing \(\mathcal{Q}\) satisfies the countable chain condition.

The following is standard as well, and essentially follows from the same argument as Kunen?.

Proposition 53. Suppose \(G\subseteq\mathcal{Q}\) is a generic filter. Write \(g=\bigcup G\). Then the following hold in \(V[G]\):

  1. \(g\) is a function on its domain.

  2. \(\mathord{\operatorname{dom}}(g)=\bigcup_{s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}}\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\times\{s\}\).

  3. For every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\), the function \(g\restriction \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\times\{s\}\) is injective.

Remark 54. Note that in \(V[G]\), \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}\) is still a nice \(n\)-ordering system on \(\omega_n=\omega_n^{V[G]}\).

55. Suppose \(G\subseteq\mathcal{Q}\) is a generic filter. We define \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{G}\) to be the \((n+1)\)-ordering system on \(\omega_n\) which is given by applying 41 to the orderings \(<_s\), \(s\in[\omega_n]^n\) which are defined as follows (in \(V[G]\)):

  • If \(s\notin\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) then \(\mathord{<}_s=\mathord{<^{\operatorname{\mathbf{Ord}}}}\restriction \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) (in this case \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) is an initial segment of \(\omega\) by 45).

  • If \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) then for every \(x,y\in \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\) write \(x<_s y\) if and only if \(g_s(x)<g_s(y)\).

Our goal is to show the following. Its proof will be provided after [red,const]

56. In \(V[G]\), the \((n+1)\)-ordering system \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{G}\) has finite closures.

Notation 57. For \(B\subseteq\omega_n\) we write \(\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B)=\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\cap [B]^n\).

The key observation is the following.

Proposition 58. Suppose \(B\in[\omega_n]^{<\omega}\) and \(p\in\mathcal{Q}\) satisfy the following:

  1. The set \(B\cap \omega\) is an initial segment of \(\omega\).

  2. For all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B)\) we have \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\cap B=\mathord{\operatorname{dom}}(p_s)\).

  3. For all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B)\) we have that \(\mathop{\mathrm{Im}}(p_s)\subseteq\omega\) is an initial segment.

Then \(p\Vdash B\text{ is \dot{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{G}}-closed}\).

Proof. Let \(B,p\) be as above. Let \(G\subseteq\mathcal{Q}\) be a generic filter containing \(p\) and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^{G}\) the corresponding \((n+1)\)-ordering system as in 55. Suppose \(t\in[B]^n\), \(\alpha,\beta\in\mathord{\operatorname{dom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G_t)=\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_t)\) and \(\alpha\in B\) satisfy \(\beta\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G_t\alpha\). Assume first that \(t\notin\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\), so \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_t)\in\omega\) by 45 and the first bullet of 55, \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G_t\) is the usual ordering on this natural number. Thus, by ([jt1]) we have \(\beta\in B\). Consider now the case \(t\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\), meaning \(g_t(\beta)<g_t(\alpha)\). Since \(\alpha\in \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_t)\cap B\) we have \(\alpha\in\mathord{\operatorname{dom}}(p_t)\) by ([jt2]), and so ([jt3]) and the injectivity part of 53 gives us \(\beta\in \mathord{\operatorname{dom}}(p_t)\) and hence \(\beta\in B\) by ([jt2]). More explicitly, by 53 there is some \(q\in G\) such that \((\beta,t) \in \mathord{\operatorname{dom}}(q)\). Additionally, by ([jt2]) there is some \(\beta'\in \mathord{\operatorname{dom}}(p_t)\) such that \(p_t(\beta')=g_t(\beta)\). Since \(G\) is a filter, \(p\) and \(q\) are compatible, i.e., there is some \(r\in G\) containing both. So \(p_t(\beta') = g_t(\beta) = r_t(\beta)\) so \(\beta = \beta'\) because \(r_t\) is injective. ◻

Proposition 59 (Small edit at end of proof of 3 and change proof of 4). Let \(A\in[\omega_n]^{<\omega}\) and \(p\in\mathcal{Q}\). Then there is some \(B\in[\omega_n]^{<\omega}\) and \(q\in\mathcal{Q}\) such that the following all hold.

  1. \(p\leq q\) and \(A\subseteq B\).

  2. The set \(B\cap\omega\) is an initial segment of \(\omega\).

  3. For every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) we have that \(\mathop{\mathrm{Im}}(q_s)\) is an initial segment of \(\omega\).

  4. For every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) we have \(\mathord{\operatorname{dom}}(q_s)\subseteq \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\cap B\).

  5. For every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B)\) we have \(\mathord{\operatorname{dom}}(q_s)\supseteq \mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\cap B\).

Proof. Given a condition \(r\in \mathcal{Q}\), write \(\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{r}\) for the set of all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) such that \(\mathop{\mathrm{Im}}(r_s)\neq\emptyset\). Our proof will, for each \(1\leq i\leq 5\), construct \(q_i\in \mathcal{Q}\) and \(B_i\in[\omega_n]^{<\omega}\) such that the items \((1),...,(i)\) hold, and then \(q=q_5\), \(B=B_5\) will satisfy the conclusion. Assume without loss of generality that \(p\) is not the empty condition. We start by setting \(q_1=p\) and \(B_1=A\). We now set \(q_2=q_1\) and \(B_2=B_1\cup \bigcup(B_1\cap \omega)\), so that ([J1]) and ([J2]) both hold.

To each \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_2}\) we associate the set \(X_s=\max\{\mathop{\mathrm{Im}}(q_2)_s\}\setminus\mathop{\mathrm{Im}}(q_2)_s\) and write \(j_s=|X_s|\). We now write \(X_s=\{m^s_0,...,m^s_{j_{s-1}}\}\). Let \(\{n^s_0,...n^s_{j_{s-1}}\}\) denote the first \(j_s\) elements of \(\omega\setminus\mathord{\operatorname{dom}}(q_2)_s\), and define \(q_3=q_2\cup\{(n^s_i,s;m^s_i):s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_2},i< j_s\}\) (note that by 46 the elements \(n^s_i\) are elements of \(\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\)) and \(B_3=B_2\). By our choice of \(m_i\) the functions \(q_s\) are still injective, meaning \(q_3\in \mathcal{Q}\) and ([J1]) clearly holds. Trivially ([J2]) still holds. By definition of \(X_s\) and \(q_3\) we have that \(\mathop{\mathrm{Im}}(q_3)_s=\mathop{\mathrm{Im}}(q_2)_s\cup X_s\) is an initial segment of \(\omega\), so ([J3]) holds for all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_2}=\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_3}\), and so the same holds for all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) (for any \(s\notin\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_3}\) we trivially have that \(\mathop{\mathrm{Im}}(q_3)_s\) is an initial segment of \(\omega\)).

To every \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_3}\) we set \(\{a^s_0,...,a^s_{l_s-1}\}=\mathord{\operatorname{dom}}(q_3)_s\setminus B_3\) for \(l_s=|\mathord{\operatorname{dom}}(q_3)_s\setminus B_3|\). We now define \(q_4=q_3\), \(B_4'=B_4\cup\{a^s_i:s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_4},i<l_s\}\) and \(B_4=B_4'\cup \bigcup(B_4'\cap\omega)\). We have that ([J2]) holds by definition of \(B_4\), and since \(q_4=q_3\) we have that ([J1]) and ([J3]) both hold. By choice of \(a^s_i\) we have that ([J4]) holds for all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_3}\), and so the same holds for all \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\) (for any \(s\notin\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}^{q_3}\) we clearly have \(\mathord{\operatorname{dom}}(q_3)\subseteq\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\cap B\).

Finally, for any \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B_4)\), write \(\{x^s_0,...,x^s_{m_s -1}\}\) for the elements of the set \((\mathord{\operatorname{predom}}(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}_s)\cap B_4)\setminus\mathord{\operatorname{dom}}(q_4)_s\) and \(\{y^s_0,...,y^s_{m_s -1}\}\) for the first \(m_s-1\) elements of \(\omega\setminus \mathop{\mathrm{Im}}(q_4)_s\), and set \(B_5=B_4\) and \(q_5=q_4\cup\{(x^s_i,s;y^s_i):s\in\inf_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B),i<m_s-1\}\). We have that ([J1]), ([J2]) hold by induction. For \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}\setminus\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B_5)\) we have that ([J3]) and ([J4]) hold because \((q_4)_s=(q_5)_s\). and for \(s\in\mathop{\mathrm{Inf}}_{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}}(B_5)\) by choice of \(x^s_i,y^s_i\). To finish, ([J5]) also holds by choice of \(x^s_i\). ◻

Proof of 56. Fix \(A\in[\omega_n]^{<\omega}\). We will show that the set \(D_A\) of conditions which force that \(A\) is contained in a finite \(\dot{\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G}\)-closed set \(B\in[\omega_n]^{<\omega}\) is dense in \(\mathcal{Q}\). Fix \(p\in\mathcal{Q}\). Let \((B,q)\) be a pair given by applying 59 to \((A,p)\), so in particular it satisfies the conditions of 58, hence \(q\) forces that \(B\) is closed. Since \(B\) is finite and contains \(A\), the result follows. ◻

Corollary 60. Given \(n<\omega\), it is consistent with ZFC that \(\mathop{\mathrm{VCcof}}(\omega,\omega,\omega_n)=n+1\).

Proof. Let \(G\subseteq\mathcal{Q}\) be generic, then \(V[G]\) and \(V\) have the same cardinals by 52 and \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G\) is an \((n+1)\)-ordering system on \(\omega_n\) with finite closures by 56. The collection of finite \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G\)-closed sets has VC-dimension \(n+1\) by 25, so we finish by 26. ◻

Question 61. Can the above construction be modified so that the closed \((n+1)\)-initial segments of the orderings \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G\) are themselves \(\prec\mathrel{\mkern-5mu}\mathrel{\cdot}^G\)-closed (as in 28)?

6 Conclusions and questions↩︎

Combining [monotonicity,lower1,prop:measurable,singular,regular,Corollary: omega_n] gives us the following.

62. Suppose \(\lambda\leq\theta\leq\kappa\) are infinite cardinals.

  1. For \(\theta^{+\omega}\leq \kappa\) we have \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)=\infty\).

  2. For \(\theta\) regular uncountable, if \(\kappa=\theta^{+n}\) for some \(n<\omega\) then \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)=n+1\).

  3. For \(\theta=\omega\), if \(\kappa=\omega_n\) for some \(n<\omega\) then it is consistent that \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)=n+1\) (note that in this case \(\lambda=\omega\)).

  4. For \(\theta\) singular, arbitrary \(\kappa\), and any \(\lambda\) satisfying \(\mathop{\mathrm{cof}}(\theta)<\lambda\) we have \(\mathop{\mathrm{VCcof}}(\lambda,\theta,\kappa)=\infty\).

  5. Relative to the consistency of a measurable cardinal, it is consistent that there exists an uncountable cardinal \(\theta\) of countable cofinality satisfying
    \(\mathop{\mathrm{VCcof}}(\omega,\theta,\theta^{+n})=n+1\) for every \(n<\omega\).

Question 63. Is there a model of ZFC and a natural number \(n\) for which there is no \(n+1\)-ordering system on \(\omega_n\) with finite closures? More generally, is it consistent that \(\mathop{\mathrm{VCcof}}(\omega,\omega,\omega_n)>n+1\) for some \(n<\omega\)?

64. Can one construct cofinal VC-classes of finite subsets of an infinite set in a way which does not require an ordering system?


  1. See Shoenfield? for a comprehensive review of measurable cardinals.↩︎