---
title: Approachability Property in Set Theory
url: https://www.emergentmind.com/topics/approachability-property
type: topic
---

# Approachability Property in Set Theory

In modern set theory, the approachability property is the family of principles attached to Shelah’s approachability ideal, an ideal designed to formalize when ordinals below a successor cardinal can be approximated by shorter initial segments drawn from a fixed sequence. It is one of the central combinatorial interfaces between square principles, stationary reflection, scales, tree properties, and forcing constructions at successors of regular and singular cardinals. In the literature represented here, the underlying object is uniform—Shelah’s ideal \(I[\lambda]\)—but the phrase “approachability property” is used with two closely related conventions: one says that a regular \(\lambda\) “has the approachability property” when \(I[\lambda]\) is proper, while another writes \(AP_\mu\) for the assertion \(\mu^+\in I[\mu^+]\) [1802.10125] [2404.18571].

## 1. Shelah’s ideal and the two standard conventions

Let \(\lambda\) be a regular uncountable cardinal. A \(\lambda\)-approaching sequence is a sequence
\[
\bar a=(a_\xi:\xi<\lambda)
\]
of bounded subsets \(a_\xi\subset\lambda\). Given such a sequence, \(B(\bar a)\) is the set of all \(\delta<\lambda\) which are singular and for which there is a cofinal club \(c\subset\delta\) of order-type \(<\delta\) so that every initial segment \(c\cap\gamma\) reappears in the list \(\{a_\eta:\eta<\delta\}\). Shelah’s approachability ideal \(I[\lambda]\) is the normal ideal on \(\lambda\) generated by \(NS_\lambda\), the non-stationary ideal, together with all sets \(B(\bar a)\) arising from \(\lambda\)-approaching sequences [1802.10125].

At successors, an equivalent presentation used repeatedly in later work fixes \(\mu\) regular and defines \(I[\mu^+]\) by requiring a sequence \((a_\alpha)_{\alpha<\mu^+}\subseteq[\mu^+]^{<\mu}\) and a club \(C\subseteq\mu^+\) such that whenever \(\gamma\in A\cap C\), there is \(E\subseteq\gamma\) with \(\operatorname{otp}(E)=\operatorname{cf}(\gamma)\), \(E\) unbounded in \(\gamma\), and \(\{E\cap\alpha:\alpha<\gamma\}\subseteq\{a_\alpha:\alpha<\gamma\}\). This is the formulation used in work on \(AP_\mu\) and on internal approachability variants [2404.18571].

The terminological point matters. In the exposition attached to “Guessing models and the approachability ideal,” one says that \(\lambda\) has the approachability property if \(I[\lambda]\) is a proper ideal, i.e. if there is at least one stationary subset of \(\lambda\) outside \(I[\lambda]\) [1802.10125]. By contrast, later papers standardly write
\[
AP_\mu:\iff \mu^+\in I[\mu^+]
\]
or, equivalently in that notation, that the ideal is improper at \(\mu^+\) [2404.18571] [2508.04374]. The shared substance is the same ideal; what changes is which global assertion is singled out.

## 2. Combinatorial content of approachability

The ideal \(I[\kappa^+]\) is closely tied to square, scales, and Aronszajn-tree phenomena. One basic relationship is that \(AP(\kappa)\) is a weakening of Jensen’s weak-square principle: \(\square_\kappa\) implies \(AP(\kappa^+)\) [1702.05062]. At successors of singulars, the relation is especially sharp. If \(\sigma\) is singular of cofinality \(\omega\) and \(\sigma_n\uparrow\sigma\) are regular, then over ZFC,
\[
AP(\sigma^+) \iff \text{every scale of length }\sigma^+\text{ in }\prod_n \sigma_n\text{ is a good scale},
\]
and failure of \(AP(\sigma^+)\) is equivalent to the existence of a bad scale of length \(\sigma^+\) [1702.05062].

Failure of approachability also has direct consequences for trees and square. For a regular \(\kappa\), failure of \(AP(\kappa)\) implies failure of \(\square_{\kappa^-}\), and hence there is no special \(\kappa\)-Aronszajn tree [1702.05062]. From the singular-cardinal side, \(\square^*_\kappa\Rightarrow AP_\kappa\), so negating \(AP_\kappa\) gives strong compactness-type consequences [2508.04374].

The ideal also stratifies by cofinality. If \(\kappa\) is regular then \(I[\kappa^+]\) contains the entire set \(\kappa^+\cap\operatorname{cf}(<\kappa)\); in particular \(AP_\kappa\) holds for all regular \(\kappa\). If \(\kappa\) is singular, ZFC proves only that \(\kappa^+\cap\operatorname{cf}(\le \operatorname{cf}(\kappa))\in I[\kappa^+]\), and whether higher cofinality layers must lie in the ideal was precisely the question left open by Shelah [2508.04374].

A useful interpretive summary, stated explicitly in the guessing-models paper, is that \(I[\kappa^+]\) measures “how far” \(\kappa^+\) fails stationary reflection on cofinality \(\kappa\) points, while principles such as \(TP\), \(ITP\), and \(ISP\) tend to produce clubs of guessing models which in turn witness approachability failures or collapses [1802.10125].

## 3. Guessing models, \(GM^+\), and collapse of \(I[\omega_2]\)

A major refinement in the study of approachability is the connection with guessing models. For a powerful structure \(R\) and regular \(\gamma\le\kappa\), a set \(M\prec R\) of size \(\kappa\) is \(\gamma\)-guessing if whenever \(x\subset M\) is bounded in \(M\) and for every \(a\in M\cap P_\gamma(M)\) one has \(x\cap a\in M\), then there exists \(g\in M\) with \(x=g\cap M\). The principle \(GM(\kappa,\gamma,R)\) says that the class of \(\gamma\)-guessing \(M\prec R\) of size \(\kappa\) is stationary in \(P_\kappa(R)\), and \(GM(\kappa,\gamma)\) means this holds for all large \(H_\theta\). The strengthening \(GM^+(\kappa^{++},\gamma)\) uses strongly \(\gamma\)-guessing models of size \(\kappa^{++}\), defined as increasing unions of \(\kappa^{++}\) many \(\gamma\)-guessing submodels of size \(\kappa\), continuous at points of cofinality \(\kappa\) [1802.10125].

Mohammadpour and Veličković show that \(GM^+(\omega_3,\omega_1)\) has a broad combinatorial footprint. In any model of \(GM^+(\omega_3,\omega_1)\), one gets \(ISP(\omega_2)\) and \(ISP(\omega_3)\), the tree property at \(\omega_2\) and \(\omega_3\), the Singular Cardinal Hypothesis above \(\omega_2\), and
\[
\neg \square(\omega_2,\lambda)\qquad\text{for all regular }\lambda\ge\omega_2
\]
[1802.10125].

For the approachability ideal, the decisive statement is
\[
I[\omega_2]\restriction S_{\omega_2}^{\omega_1}=NS\restriction S_{\omega_2}^{\omega_1}.
\]
Here \(S_{\omega_2}^{\omega_1}\) is the set of ordinals below \(\omega_2\) of cofinality \(\omega_1\). Thus, on cofinality-\(\omega_1\) points, the approachability ideal collapses to the non-stationary ideal [1802.10125]. In the same exposition, a local principle \(FS(\kappa^+,\gamma)\) is isolated: for every \(X\in H_{\kappa^+}\) there is a \(\kappa\)-closed unbounded set of \(\gamma\)-guessing models \(M\prec H_{\kappa^+}\) of size \(\kappa\) containing \(X\). The proposition proved there states that if \(FS(\kappa^+,\kappa)\) holds, then
\[
I[\kappa^+]\restriction S_{\kappa^+}^{\kappa}=NS\restriction S_{\kappa^+}^{\kappa}.
\]
The proof sketch argues that if \(\delta=M\cap\kappa^+\) were in \(B(\bar a)\), then from the cofinal sequence \(c\subset\delta\) one could build a bounded set approximated by \(M\) but not guessed by \(M\), a contradiction [1802.10125].

This collapse result at \(\omega_2\) was previously shown consistent by Mitchell, and the significance of the guessing-model approach is that it places the collapse alongside tree properties, \(ISP\), \(SCH\), and square failures in a single framework [1802.10125].

## 4. Forcing constructions and successive failures

The forcing used to obtain the simultaneous phenomena at \(\omega_2\) and \(\omega_3\) starts from two supercompact cardinals \(\kappa<\lambda\). Conditions are finite side conditions built from two kinds of models: countable elementary submodels of \(V_\lambda\) (“C-models”) and transitive \(<\kappa\)-closed approximations of initial segments of \(V_\lambda\) of size \(<\kappa\) (“Magidor models”). The forcing \(P^\kappa_\lambda\) is \(\kappa\)-c.c., preserves \(\omega_1\) by strong properness with respect to countable models, and preserves \(\kappa\) by strong properness with respect to Magidor models. In the extension \(V[G]\), one gets \(\kappa\mapsto\omega_2\), \(\lambda\mapsto\omega_3\), as well as \(GM(\omega_2,\omega_1)\), \(FS(\omega_2,\omega_1)\), and, if \(\lambda\) is supercompact, \(GM^+(\omega_3,\omega_1)\) [1802.10125].

A complementary direction is to force long intervals of failure of approachability. Unger proves that, assuming a supercompact cardinal, there is a forcing extension \(V[G*H][R]\) in which \(\aleph_{\omega^2}\) is strong limit and, for every regular cardinal \(\lambda\) with
\[
\aleph_2\le \lambda\le \aleph_{\omega^2+3},
\]
\(AP(\lambda)\) fails [1702.05062]. The construction combines an Easton-support preparation by Mitchell-style collapses with a main diagonal Prikry forcing and a full-support side forcing. The Prikry forcing has the Prikry property and adds no new bounded subsets of \(\kappa\), so the target singular remains strong limit [1702.05062].

The analysis of why \(AP\) fails across the interval is not uniform. At double successors, approximation-preservation arguments show that any candidate witness to approachability would already appear before the relevant collapse. At successors of singulars \(\aleph_{\omega\cdot n+1}\), PCF theory is used to preserve bad scales, and bad scale implies failure of \(AP\) at a successor of singular. At \(\aleph_2\), the argument is the classical Mitchell-collapse analysis together with the Gitik–Krueger preservation theorem for failure of approachability under \(\omega\)-centered forcing [1702.05062].

These two forcing lines exhibit complementary modes of interaction: side-condition forcing and guessing models can collapse \(I[\omega_2]\) on a cofinality layer, while Prikry-style constructions can force extended blocks of regular cardinals to fail approachability altogether.

## 5. Internal variants and the fine structure of \(I[\mu^+]\)

The global statement \(AP_\mu\) does not eliminate distinctions among internal approximation properties of elementary submodels. Foreman–Todorcevic style variants include internally unbounded \(IU_\mu\), internally stationary \(IS_\mu\), internally club \(IC_\mu\), and internally approachable \(IA_\mu\), with
\[
IA_\mu \Rightarrow IC_\mu \Rightarrow IS_\mu \Rightarrow IU_\mu
\]
[2404.18571].

Recent work shows that these implications can fail stationarily, and even at successive levels. Under Martin’s Maximum there are stationarily many
\[
N\in[H(\omega_3)]^{\omega_1}
\]
such that \(N\cap H(\omega_2)\) is \(IA_{\omega_1}\) while \(N\cap H(\omega_3)\) is not \(IS_{\omega_1}\), answering a question of Foreman negatively [2404.18571]. Under stronger large-cardinal assumptions and a new Mitchell-style forcing \(M(\mu,\kappa)\), one can force \(AP_\mu\) together with stationarily many
\[
N\in[H(\mu^{+++})]^\mu
\]
for which \(N\cap H(\mu^+)\in IA_\mu\), \(N\cap H(\mu^{++})\in IC_\mu\setminus IA_\mu\), and \(N\cap H(\mu^{+++})\in IS_\mu\setminus IC_\mu\). A two-step extension by
\[
Q=M(\tau,\mu,\kappa)\times Add(\mu,\kappa^+)
\]
also yields \(AP_\mu\) together with stationarily many \(N\in[H(\mu^+)]^\mu\) which are \(IC_\mu\) but fail \(IA_\mu\) [2404.18571].

A separate structural issue concerns maximal generators of the ideal. Krueger develops forcing with finite conditions that adds partial square sequences on stationary sets using adequate sets of models as side conditions, and a side-condition product forcing for simultaneously adding partial square sequences on multiple stationary sets. Assuming the consistency of a greatly Mahlo cardinal, it is consistent that the approachability ideal \(I[\omega_2]\) does not have a maximal set modulo clubs [1607.04772]. This shows that the internal structure of \(I[\omega_2]\) can be substantially more complicated than a single canonical generator would suggest.

## 6. Shelah’s problem and global failure at singular cardinals

Shelah asked whether, for singular \(\kappa\), the ideal \(I[\kappa^+]\) must contain a club subset of \(\kappa^+\cap \operatorname{cf}(\kappa)^{++}\); equivalently, whether ZFC proves
\[
\kappa^+\cap\operatorname{cf}(\kappa)^{++}\in I[\kappa^+].
\]
This remained open from the 1980s [2508.04374].

Jakob and Poveda give a negative consistency answer. Assuming appropriate large cardinal hypotheses, for a singular cardinal \(\aleph_\gamma\) and a regular \(\mu\in(\operatorname{cf}(\gamma),\aleph_\gamma)\), they construct a model of ZFC in which
\[
\aleph_{\gamma+1}\cap\operatorname{cof}(\mu)\notin I[\aleph_{\gamma+1}].
\]
Their forcing is a Prikry-type Magidor-product followed by a collapse, with a Strong Prikry Property used to show that no ground-model family of size \(<\kappa\) can cover an initial segment of the added cofinal map; consequently the stationary set
\[
S=\kappa^+\cap\operatorname{cf}(\mu)
\]
is forced to consist entirely of non-approachable points [2508.04374].

They also obtain a global theorem: assuming GCH and a proper class of supercompact cardinals, there is a class-forcing extension in which, for every singular \(\kappa\), the approachability property \(AP_\kappa\) fails, and moreover for each singular \(\kappa\) there are unboundedly many regular \(\delta<\kappa\) such that
\[
\kappa^+\cap\operatorname{cf}(\delta)\notin I[\kappa^+].
\]
As a corollary, in that model there is no special \(\kappa^+\)-Aronszajn tree for any singular \(\kappa\) [2508.04374].

These results reposition the approachability property from a local combinatorial feature to a global organizing principle for singular-cardinal combinatorics. They also sharpen a common misconception: failure of approachability is not confined to isolated successors of singulars or to the classical \(\omega\)-cofinality case. The recent consistency results show that non-approachability can be forced simultaneously at every singular cardinal, while other constructions show that even when \(AP_\mu\) holds, the surrounding landscape of internal approachability and of the ideal \(I[\mu^+]\) can remain highly nontrivial [2404.18571] [2508.04374].

Source: https://www.emergentmind.com/topics/approachability-property