Limit models — unions of increasing chains in which every step is universal over its predecessor — are the standard surrogate for saturated models in abstract elementary classes (AECs), the semantic generalization of first-order model theory that lacks compactness. Grossberg, VanDieren and Villaveces proved in 2016 that in an AEC with amalgamation that is Galois-stable in a cardinal μ and satisfies superstability-like conditions on μ-splitting (locality and no long splitting chains), any two limit models over a common base are isomorphic over that base — provided the class also has no μ-Vaughtian pairs, a unidimensionality hypothesis. That hypothesis is used at exactly one point of the proof, to show that 'reduced' towers are continuous, and it was added in 2014 to repair an error found in 2012 in the original claim that the superstability-like conditions alone suffice. Subsequent work recast the question — under μ-superstability, continuity of reduced towers is equivalent to symmetry of non-μ-splitting over limit models — and settled the tame case: with tameness, μ-superstability implies uniqueness of limit models, and uniqueness is asymptotically equivalent to every other notion of superstability. What remains open is the general case: whether μ-superstability alone, without tameness and without Vaughtian-pairs or categoricity assumptions, implies symmetry, continuity of reduced towers, and hence uniqueness of limit models of cardinality μ; and whether, without tameness, uniqueness conversely implies superstability.

Background

An abstract elementary class (AEC) is a class K of structures in a fixed similarity type, together with a strong substructure relation ≺_K, closed under isomorphisms and under unions of ≺K-increasing chains, and satisfying a downward Löwenheim–Skolem property with minimal size bound LS(K). AECs were introduced by Shelah in 1977; they generalize the class of models of a complete first-order theory with elementary substructure, and include classes axiomatizable in the infinitary logic L{ω1,ω} and many classes of modules. Unlike the elementary setting there is no compactness, so model-theoretic arguments must be semantic. If K has the amalgamation property (two extensions of a common submodel can be overlaid) and the joint embedding property (any two models embed into a common one), one may work inside a large homogeneous 'monster' model C, and types become orbits: the Galois type of an element a over M is its orbit under automorphisms of C fixing M. K is Galois-stable in μ if every model of size μ realizes at most μ Galois types.

Because relative saturation is unwieldy in AECs, the working surrogate for a saturated model is a limit model: M is a (μ,σ)-limit model over N if M = ⋃{i<σ} M_i with M_0 = N, all M_i of size μ, and each M{i+1} universal over M_i (every extension of M_i of size μ embeds into M_{i+1} over M_i). Galois-stability in μ guarantees that limit models exist. In a first-order superstable theory all limit models of a given cardinality are isomorphic — they are exactly the saturated models — while in stable non-superstable theories they split into several isomorphism types governed by the cardinal κ_r(T). Uniqueness of limit models in an AEC is therefore read as an analogue of superstability.

The relevant independence notion is μ-splitting (Definition 2.1 of Grossberg–VanDieren–Villaveces): a Galois type p ∈ ga-S(M) μ-splits over N ∈ K_{≤μ} with N ≺_K M if there exist N_1, N_2 ∈ K_μ and a K-embedding h: N_1 → N_2 such that N ≺_K N_l ≺K M for l = 1,2, h ↾ N = id_N, and p ↾ N_2 ≠ h(p ↾ N_1). The superstability-like package isolated by Grossberg–VanDieren–Villaveces (their Assumption 2.8(4); following later authors, μ-superstability) asks that, for every infinite ordinal α, every sequence ⟨M_i : i < α⟩ of limit models of cardinality μ with M_α = ⋃{i<α} M_i, and every p ∈ ga-S(M_α): (a) if p ↾ M_i does not μ-split over M_0 for every i < α, then p does not μ-split over M_0 (locality, also called continuity); and (b) there exists i < α such that p does not μ-split over M_i (no long splitting chains). Their proof runs through towers: increasing sequences of models with named realizations and base models coding non-splitting, partially ordered by an extension order; a tower is reduced if it admits no proper diagonal improvement, and reduced towers are dense. The scheme originates in Shelah and Villaveces's 1999 attempt to prove uniqueness under categoricity without amalgamation; gaps in that proof were found and the program was repaired by VanDieren under additional closure hypotheses. The scheme is: reduced towers are dense and continuous, and from this one obtains that any two limit models over a common base are isomorphic over that base.

A minimal type is a non-algebraic Galois type with a unique non-algebraic extension to every model of size μ, and a μ-Vaughtian pair is a pair M' ≺_K N' of models of size μ over a limit model M in which some minimal type over M acquires no new realization in N'. 'No μ-Vaughtian pairs' is a unidimensionality-like hypothesis, much weaker than categoricity (real closed fields, for instance, have no Vaughtian pairs without being categorical). Two further notions frame what is now known. Tameness is a locality property for Galois types: a type over a model of size μ is determined by its restrictions to submodels of some smaller fixed size (χ-tameness). Symmetry for non-μ-splitting over limit models is the two-type interchange property of VanDieren ('Superstability and symmetry', Definition 3): K has symmetry for non-μ-splitting over limit models if whenever M, M_0, N ∈ K_μ and elements a, b satisfy (1) M is universal over M_0 and M_0 is a limit model over N, (2) a ∈ M \ M_0, (3) ga-tp(a/M_0) is non-algebraic and does not μ-split over N, and (4) ga-tp(b/M) is non-algebraic and does not μ-split over M_0, then there exists M^b a limit model over M_0 containing b such that ga-tp(a/M^b) does not μ-split over N.

Grossberg, VanDieren and Villaveces proved: in an AEC without maximal models, with amalgamation and joint embedding, for μ > LS(K), that is Galois-stable in μ, has no μ-Vaughtian pairs, and satisfies the locality and no-long-splitting-chains conditions for μ-splitting, any two (μ,σ_1)- and (μ,σ_2)-limit models over a common base M are isomorphic over M. (The published formulation additionally assumes μ-disjoint amalgamation over limit models; Fred Drueck observed, recorded as Remark 1.12 of the paper, that this assumption is not needed for the tower arguments, so it is not part of the elimination question.) The no-Vaughtian-pairs hypothesis enters at exactly one point: the theorem that reduced towers are continuous, where the existence of minimal types (due to Shelah) forces any element at a discontinuity to be matched inside a proper extension, contradicting reducedness. An early version of the paper claimed that the superstability-like conditions alone suffice; an error was found in 2012 in precisely this step, and the unidimensionality assumption was added in 2014 to repair it. Whether the repair is necessary — whether superstability-like hypotheses alone imply uniqueness of limit models — is the present problem.

Problem Statement

Let K be an abstract elementary class with the amalgamation and joint embedding properties and no maximal models, and let μ > LS(K). Assume K is Galois-stable in μ. No tameness, categoricity, or Vaughtian-pairs assumption is made.

Notation. K_μ is the class of models of K of cardinality μ. A (μ,σ)-limit model over N ∈ K_μ is a model of the form M = ⋃{i<σ} M_i, where ⟨M_i : i < σ⟩ is ≺K-increasing, M_0 = N, each M_i has cardinality μ, and M{i+1} is universal over M_i (every model in K_μ extending M_i embeds into M{i+1} over M_i). A tower is a ≺_K-increasing sequence of models of cardinality μ indexed by a linear order, together with distinguished elements and base models recording that each distinguished element's Galois type does not μ-split over its base model; the tower is reduced if for every extension of it by a tower with the same distinguished elements and base models, M_i′ ∩ ⋃j M_j = M_i for all i (Definition 5.3 of the source paper), and continuous if at every limit index β, M_β = ⋃{i<β} M_i. Symmetry for non-μ-splitting over limit models is the property defined in the Background above (Definition 3 of VanDieren, 'Superstability and symmetry').

(1) Sufficiency — the Vaughtian-pairs elimination. Assume in addition that μ-splitting satisfies the locality/continuity and no-long-splitting-chains conditions of Assumption 2.8(4) (the μ-superstability package of Grossberg–VanDieren–Villaveces). Prove, or refute, that under these hypotheses alone: (a) K has symmetry for non-μ-splitting over limit models; equivalently — by Theorem 3 of VanDieren, 'Superstability and symmetry', whose standing hypothesis that K has no maximal models of cardinality μ⁺ is supplied by the 'no maximal models' assumption above — every reduced tower of models of cardinality μ is continuous; and consequently (b) for any limit ordinals σ₁, σ₂ < μ⁺ and any two (μ,σ₁)- and (μ,σ₂)-limit models over a common base M ∈ K_μ, the two limit models are isomorphic over M. A refutation is an explicit abstract elementary class satisfying all the hypotheses above in which some reduced tower is discontinuous, symmetry for non-μ-splitting over limit models fails, or two limit models over a common base are not isomorphic over that base.

(2) Necessity. In this direction the locality/continuity and no-long-splitting-chains conditions are not assumed: K is an AEC with the amalgamation and joint embedding properties and no maximal models, Galois-stable in μ, and no tameness is assumed. Does uniqueness of limit models of cardinality μ (isomorphism over the base, for all lengths below μ⁺) imply the μ-superstability package — in particular the absence of long μ-splitting chains?

The two directions are parts of a single objective: settling, for abstract elementary classes with amalgamation and without tameness, the conjecture circulating in this literature since the failed early draft of the 2016 theorem — that uniqueness of limit models is equivalent to superstability. In the general (possibly non-tame) setting they are independent questions, and each may be settled by a proof or by an explicit counterexample.

References

  1. Rami Grossberg, Monica VanDieren, Andrés Villaveces, “Uniqueness of limit models in classes with amalgamation”, Mathematical Logic Quarterly 62 (2016), 367–382. DOI: 10.1002/malq.201500033, https://doi.org/10.1002/malq.201500033 (author version: arXiv math/0509338)
  2. Monica M. VanDieren, “Superstability and symmetry”, Annals of Pure and Applied Logic 167 (2016), 1171–1183. DOI: 10.1016/j.apal.2016.05.004, https://doi.org/10.1016/j.apal.2016.05.004 (arXiv: 1507.01990)
  3. Monica M. VanDieren, Sebastien Vasey, “Symmetry in abstract elementary classes with amalgamation”, Archive for Mathematical Logic 56 (2017), 423–452. DOI: 10.1007/s00153-017-0533-z, https://doi.org/10.1007/s00153-017-0533-z (arXiv: 1508.03252)
  4. Rami Grossberg, Sebastien Vasey, “Equivalent definitions of superstability in tame abstract elementary classes”, The Journal of Symbolic Logic 82 (2017), 1387–1408. DOI: 10.1017/jsl.2017.21, https://doi.org/10.1017/jsl.2017.21 (arXiv: 1507.04223)
  5. Will Boney, Monica M. VanDieren, “Limit models in strictly stable abstract elementary classes”, Mathematical Logic Quarterly 70 (2024), 438–453. DOI: 10.1002/malq.202200075, https://doi.org/10.1002/malq.202200075 (arXiv: 1508.04717)
  6. Jeremy Beard, Marcos Mazari-Armida, “On the spectrum of limit models”, Annals of Pure and Applied Logic 176 (2025), Article 103647. DOI: 10.1016/j.apal.2025.103647, https://doi.org/10.1016/j.apal.2025.103647 (arXiv: 2503.11605)
  7. Jeremy Beard, “Long limit models are isomorphic assuming a splitting-like relation”, preprint, 2025. arXiv: 2511.18665, https://arxiv.org/abs/2511.18665
  8. Jeremy Beard, “The spectrum of limit models in a first order setting”, preprint, 2025. arXiv: 2510.03459, https://arxiv.org/abs/2510.03459
  9. Monica M. VanDieren, “Superstability in Tame Abstract Elementary Classes”, preprint, 2015, withdrawn by the author due to a crucial error. arXiv: 1502.04144, https://arxiv.org/abs/1502.04144
  10. Monica M. VanDieren, “A Characterization of Uniqueness of Limit Models in Categorical Abstract Elementary Classes”, preprint, 2015. arXiv: 1511.09112, https://arxiv.org/abs/1511.09112
  11. Will Boney, Monica M. VanDieren, Sebastien Vasey, “Shelah–Villaveces revisited”, preprint, 2016, arXiv: 1611.05292, https://arxiv.org/abs/1611.05292 (withdrawn by the authors in March 2017: version 2 consists of the notice “Paper withdrawn due to a fatal mistake”, so the derivation announced in version 1 is not an established result)
  12. Will Boney, Rami Grossberg, Monica M. VanDieren, Sebastien Vasey, “Superstability from categoricity in abstract elementary classes”, Annals of Pure and Applied Logic 168 (2017), 1383–1395. DOI: 10.1016/j.apal.2017.01.005, https://doi.org/10.1016/j.apal.2017.01.005 (arXiv: 1609.07101)
  13. Saharon Shelah, Andrés Villaveces, “Toward categoricity for classes with no maximal models”, Annals of Pure and Applied Logic 97 (1999), 1–25. DOI: 10.1016/S0168-0072(98)00015-3, https://doi.org/10.1016/S0168-0072(98)00015-3
  14. Monica VanDieren, “Categoricity in abstract elementary classes with no maximal models”, Annals of Pure and Applied Logic 141 (2006), 108–147. DOI: 10.1016/j.apal.2005.10.006, https://doi.org/10.1016/j.apal.2005.10.006