University of Illinois, Chicago
제1장에서는 non-classifiable 이론 T가 있다고 가정하고, 그 이론의 모델들 사이의 비이형을 유지하기 위한 condition에 대하여 논의합니다. 특히, DOP 또는 OTOP를 가진 이론은 classifiable 이론으로 간주되며, 이들은 forcing 확장에서 비이형이 유지됩니다.
제2장에서는 non-classifiable 이론 T가 있다고 가정하고, 그 이론의 모델들 사이의 비이형을 유지하기 위한 condition에 대하여 논의합니다. 특히, ccc-forcing은 비이형을 유지할 수 없으며, DOP 또는 OTOP를 가진 이론도 비이형을 유지하지 못한다는 것을 보여줍니다.
제3장에서는 classifiable 이론 T가 있다고 가정하고, 그 이론의 모델들 사이의 비이형을 유지하기 위한 condition에 대하여 논의합니다. 특히, Skolemized 이론은 classifiable 이론으로 간주되며, 이들은 forcing 확장에서 비이형이 유지됩니다.
논문에서는 forcing 이론과 모델 이론의 접합에 관한 연구를 진행하였으며, non-classifiable 이론, DOP 또는 OTOP를 가진 이론, classifiable 이론에 대한 분석을 수행하였습니다.
University of Illinois, Chicago
arXiv:math/9301208v1 [math.LO] 15 Jan 1993Forcing IsomorphismJ. Baldwin ∗Department of MathematicsUniversity of Illinois, ChicagoM.
C. Laskowski†Department of MathematicsUniversity of Illinois, ChicagoS. ShelahDepartment of MathematicsHebrew University of Jerusalem ‡July 4, 2018If two models of a first order theory are isomorphic then they remainisomorphic in any forcing extension of the universe of sets.
In general, how-ever, such a forcing extension may create new isomorphisms. For example,any forcing that collapses cardinals may easily make formerly non-isomorphicmodels isomorphic.
Certain model theoretic constraints on the theory andother constraints on the forcing can prevent this pathology.A countable first order theory is said to be classifiable if it is superstableand does not have either the dimensional order property (DOP) or the omit-ting types order property (OTOP). Shelah has shown [7] that if a theory Tis classifiable then each model of cardinality λ is described by a sentence of∗Partially supported by N.S.F.
grant 90000139.†Visiting U.I.C. from the University of Maryland thanks to the NSF Postdoctoralprogram.‡The authors thank the U.S. Israel Binational Science Foundation for its support ofthis project.
This is item 464 in Shelah’s bibliography.1
L∞,λ. In fact this sentence can be chosen in the L∗λ.
(L∗λ is the result ofenriching the language L∞,ℶ+ by adding for each µ < λ a quantifier sayingthe dimension of a dependence structure is greater than µ.) Further work([3], [2]) shows that ℶ+ can be replaced by ℵ1.
The truth of such sentenceswill be preserved by any forcing that does not collapse cardinals ≤λ andthat adds no new countable subsets of λ, e.g., a λ-complete forcing. Thatis, if two models of a classifiable theory of power λ are non-isomorphic, theyremain non-isomorphic after a λ-complete forcing.In this paper we show that the hypothesis of the forcing adding no newcountable subsets of λ cannot be eliminated.
In particular, we show thatnon-isomorphism of models of a classifiable theory need not be preserved byccc forcings. The following definition isolates the key issue of this paper.0.1 Definition.
Two structures M and N are potentially isomorphic if thereis a ccc-notion of forcing P such that if G is P-generic then V [G] |= M ≈N.In the first section we will show that any theory that is not classifiablehas models that are not isomorphic but are potentially isomorphic. In thesecond, we show that this phenomenon can also occur for classifiable theo-ries.
The reader may find it useful to examine first the example discussed inTheorem 2.3.1Non-classifiable TheoriesWe begin by describing a class (which we call amenable) of subtrees of Q≤ωthat are pairwise potentially isomorphic. Then we use this fact to show thatevery nonclassifiable theory has a pair of models that are not isomorphic butare potentially isomorphic.1.1 Notation.i) We adopt the following notation for relations on subsets of Q≤ω.
⊏denotes the subsequence relation; < denotes lexicographic ordering; forα ≤ω, levα is a unary predicate that holds of sequences of length (level)α; ∧is the operation on two sequences that produces their largestcommon initial segment. We denote the ordering of the rationals by ii) For η ∈Qω, let Dη = {σ ∈Qω : σ(2n) = η(n)} and Sη = {σ ∈Dη :σ(2n + 1) is 0 for all but finitely many n}. Let C = ∪η∈QωSη.iii) The language Lt (for tree) contains the symbols ⊏, <, levα and unarypredicates Pη for η ∈Qω.iv) For any A ⊆C, A∗denotes the Lt-structure with universe A ∪Q<ωunder the natural interpretation of ⊏, <, levα and with Pη(A∗) = Sη∩A.Note that ⟨C, <⟩is isomorphic to a subordering of the reals. Since C isdense we may assume Q is embedded in C but not necessarily in a naturalway.1.2 Definition. A substructure A∗of C∗is amenable if for all η ∈Qω, alln ∈ω and all s ∈Qn, if Pη(C∗) contains an element extending s then Pη(A∗)does also.1.3 Remark. It is easy to see that a substructure A∗of C∗is amenable justif for all even n and all s ∈Qn, if η(i) = s(2i) for all i < n2 then for everyr ∈Q there is a ν ∈A ∩Sη with ν|n + 1 = s⌢r.1.4 Main Lemma. If A∗and B∗are amenable substructures of C∗thenthey are potentially isomorphic.Proof. Let P denote the set of finite partial Lt-isomorphisms between Aand B under the natural partial order of extension. We can naturally extendany Lt elementary bijection between A and B to an isomorphism of A∗andB∗as Lt-structures.1.5 Claim 1. P satisfies the countable chain condition.In fact, we will show P = ∪n∈ωFn where if p, q ∈Fn then p∪q ∈P. Givenp ∈P, let ⟨a1 . . . an⟩be the lexicographic enumeration of dom p. Let n(p) bethe cardinality of dom p and let k = k(p) be the least integer satisfying thefollowing conditions.i) If i ̸= j, ai|k ̸= aj|k.ii) If i ̸= j, p(ai)|k ̸= p(aj)|k.iii) For all n with 2n + 1 ≥k, ai(2n + 1) = 0.3 iv) For all n with 2n + 1 ≥k, p(ai)(2n + 1) = 0.Now define an equivalence relation on P by p ≃q if n(p) = n(q),k(p) = k(q) and letting ⟨a1 . . . an⟩enumerate (in lexicographic order) dom p,⟨a′1 . . . a′n⟩enumerate (in lexicographic order) dom q, for each i, ai|k(p) =a′i|k(p) and p(ai)|k(p) = q(a′i)|k(p). Then since Q is countable and the do-mains of elements of P are finite, ≃has only countably many equivalenceclasses; we designate these classes as the Fn. We must show that if p ≃qthen p ∪q ∈P. It suffices to show that for all i, j if C |= ai < a′j thenC |= p(ai) < q(a′j).• Case 1: i ̸= j. By the definitions of k = k(p) and ≃, we must havej > i, ai|k < a′j|k and p(ai)|k < q(a′j)|k. This suffices.• Case 2: i = j.Choose the least t such that ai(t) ̸= a′i(t).Thenai(t) < a′i(t). Note that t > k and t must be even since for any oddt > k, ai(t) = a′i(t) = 0.Suppose ai ∈Sν and a′i ∈Sη.We now claim p(ai)|t = q(a′i)|t. Fix any ℓ< t. If ℓ< k, p(ai)(ℓ) =q(a′i)(ℓ) by the definition of ≃. If ℓ≥k is odd then the fourth condi-tion in the definition of k(p) guarantees that p(ai)(ℓ) = q(a′i)(ℓ) = 0.Finally, if ℓ≥k and ℓ= 2u, since p and q preserve the Pη, p(ai)(ℓ) =ν(u) = ai(ℓ) and q(a′i)(ℓ) = η(u) = a′i(ℓ) but these are equal by theminimality of t.It remains to show p(ai)(t) But since ai(t) Let ⟨a1 . . . an⟩enumerate dom p in lexicographic order.Fix s < n with as < a < as+1 (the other cases are similar). Let m be leastsuch that as|(m + 1), a|(m + 1), as+1|(m + 1) are distinct and let c denotea|m. Suppose Pρ(as), Pσ(a), and Pτ(as+1). Note that since as < a < as+1,it is impossible for as and as+1 to agree on a larger initial segment than aand as do. Thus without loss of generality we may assume that as|m = a|m.Two cases remain.4 • Case 1: as|m = a|m = as+1|m = c. Suppose m is odd. Let bs = p(as)and bs+1 = p(as+1). Then bs|m = bs+1|m and bs(m) < bs+1(m). By thedefinition of amenability for any r with bs(m) < r < bs+1(m), there isan η ∈B ∩Sσ with η|(m+1) = bs⌢r so p∪{< a, η >} ∈P as required.If m is even choose as the image of a, p(c)⌢σ(m/2).• Case 2: as|m = a|m = c but as+1|m ̸= c.Again, let bs = p(as)and bs+1 = p(as+1) and denote bs|m by b′. By amenability there is anη ∈B ∩Sσ with η|m = b′. Any such η is less than bs+1. If m is evenη > bs is guaranteed by η(m) > ρ(m); if m is odd by Remark 1.3 whichrecasts the definition of amenability we can choose η(m) > bs(m). Ineither case η is required image of a.We deduce three results from this lemma. First we note that there arenonisomorphic but potentially isomorphic suborderings of the reals. Thenwe will show in two stages that any countable theory that is not classifiablehas a pair of models of power 2ℵ0 that are not isomorphic but are potentiallyisomorphic.1.6 Theorem. Any two suborderings of ⟨C, <⟩that induce amenable Lt-structures are potentially isomorphic.Proof. Since the isomorphism we constructed in proving Lemma 1.4 pre-serves levels, restricting it to the infinite sequences and reducting to < yieldsthe required isomorphism.1.7 Definition. Let M be an L-structure. We say that ⟨aη ∈M : η ∈Q≤ω⟩is a set of L-tree indiscernibles if for any two sequences η, ν from Q≤ω:If η and ν realize the same atomic type in ⟨Q≤ω; ⊏, <, lev, V⟩then ⟨aη1, . . . aηn⟩and ⟨aν1, . . . aνn⟩satisfy the same L-type.Note that the isomorphism given in Theorem 1.6 preserves ∧since ∧isdefinable from ⊏. We have introduced ∧to the language so that atomictypes suffice in Definition 1.7.1.8 Theorem. Let T be a complete unsuperstable theory in a language L.Suppose L ⊆L1 and T ⊆T1 with |T1| ≤2ω. Then there are L1-structuresM1, M2 |= T1 such that each Mi|L is a model of T of cardinality 2ω, M1and M2 are not L-isomorphic but in a ccc-forcing extension of the universeM1 ≈L1 M2.5 Proof. We may assume that T1 is Skolemized. Note there is no assumptionthat T1 is stable. Let M be a reasonably saturated model of T1. By VII.3.5(2)of [4] there are L-formulas φi(x, y) for i ∈ω and a tree of elements ⟨aη ∈M :η ∈Q≤ω⟩such that for any n ∈ω, η ∈Qω, and ν ∈Qn+1 if ν|n = η|n thenφn+1(aη, aν) if and only if ν ⊏η. By VII.3.6(3) of [4] (applied in L1!) wemay assume that the index set is a collection of L1-tree indiscernibles.Let Y = ⟨aν ∈M : ν ∈Q<ω⟩. For η ∈Qω, let pη be the type over Ycontaining φn+1(x; aη|n⌢η(n)) ∧¬φn+1(x; aη|n⌢η(n)+1) for all n ∈ω.Now a direct calculation from the definition of tree indiscernibility (whichwas implicit in the proof of Theorem VIII.2.6 of [4]) shows:Claim. For any η ∈Qω and any Skolem term f, if f(aη1, . . . , aηn) realizespν then some ηi = ν.Let M2 be the Skolem hull of C′ = Y ∪{aη : η ∈C} where C is chosenas in 1.1. Since Y is countable there are at most 2ℵ0 embeddings of Y intoM2; let fη for η ∈Qω enumerate them. For η ∈Qω, define bη ∈Sη bybη(2n) = η(n) and bη(2n + 1) = 0 for all n ∈ω.Let A = ∪η∈QωS′η where S′η = Sη −{bη} if M2 realizes fη(pbη) and S′η =Sη if M2 omits fη(pbη).It is easy to check that A∗⊆C∗is amenable. Let M1 be the Skolem Hullof A′ = Y ∪{aη : η ∈A}.Since A∗and C∗are amenable there is a ccc-forcing notion P such thatV [G] |= A∗≈C∗. Since A′ and C′ are sets of L1-tree indiscernibles, theinduced map is an L1-isomorphism. Thus, V [G] |= M1 ≈L1 M2. Thus weneed only show that M1 and M2 are not isomorphic in the ground universe.Suppose h were such an isomorphism. Choose η ∈Qω such that h|Y = fη.Now if bη ∈A the construction of A guarantees that M2 omits fη(pbη) =h(pbη) but bη realizes pbη ∈M1. On the other hand, if bη ̸∈A then by theClaim, M1 omits pbη but M2 realizes f(pbη).We now want to show the same result for theories with DOP or OTOP.We introduce some specialized notation to clarify the functioning of DOP.1.9 Notation. For a structure M elementarily embedded in a sufficientlysaturated structure M∗, b from M, and a from M∗, dim(a, b, M) is theminimal cardinality of a maximal, independent over b, set of realizationsof stp(a/b) in M. For models M of superstable theories, if dim(a, b, M) is in-finite then it is equal to the cardinality of any such maximal set. For p(x, y) ∈S(∅) and b from M let d(p(x; b); M) = sup{dim(a′, b, M) : tp(a′/b) = p}.6 1.10 Lemma. If a complete, first order, superstable theory T of cardinalityλ has DOP then there is a type p(v, u, x, y) such that for any cardinal κ thereis a model M and a sequence {aα : α ∈κ} from M such that for all α, β ∈κand all c from M, d(p(v; c, aα, aβ); M) ≤λ+ and(∃u ∈M) [d(p(v; u, aα, aβ); M) = λ+] if and only if α < β.(1)Proof. This is the content of condition (st 1) on page 517 of [7]. (As for anyinfinite indiscernible I there is a finite J ⊆I such that if d ∈I \ J thentp(d, ∪J) is a stationary type and Av(I, ∪I) is a non-forking extension of it).1.11 Proposition. Suppose |L| = λ and T is a superstable L-theory witheither DOP or OTOP. There is an expansion T1 ⊇T, |T1| = λ+ such that T1is Skolemized, and an L-type p (p = p(v, u, x, y) if T has DOP, p = p(v, x, y)if T has OTOP) such that v, u, x, y are finite, lg x = lg y and for any ordertype (I, <) there is a model MI of T1 and a sequence {ai : i ∈I} from MI ofL1-order indiscernibles such that:(a) MI is the Skolem Hull of {ai : i ∈I};(b) If T has DOP then for all i, j ∈I,(∃u ∈MI) [d(p(v; u, ai, aj); MI) ≥λ+] if and only if i
Let κ be the Hanf number for omitting types for first-order languagesof cardinality λ+. If T has OTOP then by its definition (see [7] XII §4) thereis a model M of T and sequence {aα : α ∈κ} of finite tuples from M andtype p(v, x, y) such that M |= (∃v)p(v, aα, aβ) iffα < β.By Lemma 1.10 when T has DOP we can find a model M of T, a sequence{aα : α ∈κ} and a type p(v, u, x, y) so that (∃u ∈M) [d(p(v; u, aα, aβ); M) ≥λ+] if and only if α < β. We may also assume that d(p(v; c, aα, aβ); M) ≤λ+for all α, β ∈κ and c.Let L0 be a minimal Skolem expansion of L. That is, L0 is a minimalexpansion of L such that there is a function symbol Fφ(y) ∈L0 for eachformula φ(x, y) ∈L0. Let M0 be any expansion of M satisfying the Skolemaxioms ∀y[(∃x)φ(x, y) →φ(Fφ(y), y)] and let T0 = Th(M0). Without loss ofgenerality λ+ + 1 ⊆M0.7 ¿From now on, assume we are in the DOP case as the OTOP case issimilar and does not require a further expansion of the language (i.e., takeL1 = L0 and T1 = T0.) Expand L0 to L′0 by adding relation symbols <, ∈, P,constants for all ordinals less than or equal to λ+ and a new function symbolf(w, u, x, y). Let M′0 be an expansion of M0 so that < linearly orders the aαand the set of aα is the denotation of P. Interpret the constants and ∈inthe natural way. For all α, β ∈κ and all realizations dbc of p(v, u, aα, aβ) inM0, let (λw)f(w, c, aα, aβ) be a 1–1 map from an initial segment of λ+ to amaximal, independent over c ∪aα ∪aβ, set of realizations of stp(d/c aαaβ).Let L1 be a minimal Skolem expansion of L′0, let M1 be a Skolem ex-pansion of M′0 to an L1-structure and let T1 denote the theory of M1. So|T1| = λ+.Note that if, for some c, the domain of (λw)f(c, aα, aβ) is λ+ then α <β. Also, for all α, β ∈κ and c from M1 the independence of the range of(λw)f(w, c, aα, aβ) is expressed by an L1-type. Thus M1 omits the typesq(v; u, x, y) = p(v, u, x, y)∪{v ⌣uxy {f(γ, u, x, y) : γ < λ+}}∪{P(x)}∪{P(y)}∪{x ̸< y}and r(v) = {v ∈λ+} ∪{v ̸= γ : γ ∈λ+}.To complete the proof of the proposition construct an Ehrenfeucht-Mostowskimodel MI of T1 built from a set of L1-order indiscernibles {ai : i ∈I} omit-ting both q(v, u, x, y) and r(v). The existence of such a model follows as inthe proof of Morley’s omitting types theorem (see e.g., [4, VII.5.4]).Note that in the DOP case of the proposition above the argument showsd(p(v; c, ai, aj); MI) ≤λ+ for all i, j ∈I and c.We have included a sketch of the proof of Lemma 1.11 which is essentiallyFact X.2.5B+6209 of [7] and Theorem 0.2 of [5] to clarify two points. Wewould not include this had not experience showed that some readers missthese points.Note that the parameter c is needed in the DOP case notonly to fix the strong type, but because in general we cannot ensure theexistence of a large, independent set of realizations over aα ∪aβ. Also, it isessential that we pass to a Skolemized expansion to carry out the omittingtypes argument and that the final set of indiscernibles are indiscernible in theSkolem language. We can then reduct to L for the many models argument(if we use III 3.10 of [6]) not just [7] VIII,§3) but for the purposes of thispaper we cannot afford to take reducts as the proof of Theorem 1.14 requires8 that an isomorphism between linear orders I1, I2 induces an isomorphism ofthe corresponding models.Let us expand on why we quote [6] above. In [6], Theorem III 3.10 it isproved that for all uncountable cardinals λ and all vocabularies τ, if there isa formula Φ(x, y) such that for every linear order (J, <) of cardinality λ thereis a τ-structure MJ of cardinality λ and a subset of elements {as : s ∈J}satisfyingi) MJ |= Φ(as, at) if and only if s It istrue that the natural example satisfying these conditions is an Ehrenfeucht-Mostowski model built from ⟨as : s ∈J⟩in some expanded language, but thisis not required. In particular, our generality allows taking reducts, so long asthe formula Φ remains in the vocabulary. Further, there is no requirementthat Φ be first-order.However, in our context we want to introduce an isomorphism betweentwo previously non-isomorphic models. The natural way of doing this is toproduce two non-isomorphic but potentially isomorphic orderings J1 and J2and then conclude that MJ1 and MJ2 become isomorphic. Consequently, itis important for us to know that the models are E.M. models.We can simplify the statement of the conclusion of Lemma 1.11 if wedefine the logic with ‘dimension quantifiers’. In this logic we demand that inaddition to the requirement that ‘equality’ is a special predicate to be inter-preted as identity that another family of predicates also be given a canonicalinterpretation.1.12 Notation. Expand the vocabulary L to ˆL by adding new predicatesymbols Qµ(x, y) of each finite arity for all cardinals µ ≤λ+. Now definethe logic ˆLλ+,ω by first demanding that each predicate Qµ is interpreted inan L-structure M byM |= Qµ(a, b) if and only if dim(a, b, M) = µ.9 Then define the quantifiers and connectives as usual. We will only be con-cerned with the satisfaction of sentences of this logic for models of superstabletheories.1.13 Remarks.i) The property coded in Condition (1) of Lemma 1.11is expressible by a formula Φ(x, y) in the logic ˆLλ+,ω. Each formulain ˆLλ+,ω and in particular this formula Φ is absolute relative to anyextension of the universe that preserves cardinals. More precisely Φ isabsolute relative to any extension of the universe that preserves λ+.ii) If T has OTOP the formula Φ can be taken in the logic Lλ+,ω. So inthis case Φ is preserved in any forcing extension.iii) Alternatively, the property coded in Condition (1) of Lemma 1.11 isalso expressible in Lλ+,λ+. That is, there is a formula Ψ(x, y) ∈Lλ+,λ+(in the original vocabulary L) so thatMI |= Ψ(ai, aj) if and only if i
However,the particular statements MI |= Ψ(ai, aj) and MI |= ¬Ψ(ai, aj) will bepreserved under any cardinal-preserving forcing by the first remark.iv) Note that we could have chosen the type p (in the DOP case) suchthat p(v; c, aα, aβ) is a stationary regular type. Note also that had wefollowed [7],X2.5B more closely, we could have insisted that |T1| = λ.In fact we could have arranged that in MI, every dimension wouldbe ≤ℵ0 or ∥MI∥(over a countable set). However, neither of theseobservations improve the statement of 1.14.1.14 Theorem. If T is a complete theory in a vocabulary L with |L| ≤2ωand T has either OTOP or DOP then there are models M1 and M2 of Twith cardinality the continuum that are not isomorphic but are potentiallyisomorphic.Proof. By Theorem 1.8 we may assume that T is superstable. By Propo-sition 1.11 and Remark 1.13(i) there is a model M of a theory T1 ⊇T ina Skolemized language L1 ⊇L containing a set of L1-order indiscernibles10 {aη : η ∈Q≤ω} and an ˆLλ+,ω-formula Φ(x, y) so that Φ(aη, aν) holds inM if and only if η is lexicographically less than ν.Further, the state-ments “M |= Φ(aη, aν)” and “M |= ¬Φ(aη, aν)” are preserved under anyccc forcing.Note that this L1-order indiscernibility certainly implies L1-tree-indiscernibility in the sense of Definition 1.7.Thus, the construction of potentially isomorphic but not isomorphic mod-els proceeds as in the last few paragraphs of the proof of Theorem 1.8 oncewe establish the following claim.Claim. For any ν ∈Qω there is a collection pν(x) of boolean combina-tions of Φ(x, a) as a ranges over Y such that for any η ∈Qω and any L1-termf, if f(aη1, . . . aηn) realizes pν in M then some ηi = ν.Proof. The conjunction of the Φ(x; aν|n⌢<ν(n)+1>) and ¬Φ(x; aν|n⌢<ν(n)+1>)that define the ‘cut’ of aν will constitute pν. Now if ν is not among the ηichoose any n such that η1|n, η2|n, . . . ηk|n, ν|n are distinct.Then the se-quences ⟨η1, . . . ηk, ν|n⌢⟨ν(n) + 1⟩⟩and ⟨η1, . . . ηk, ν|n⌢⟨ν(n) −1⟩⟩have thesame type in the lexicographic order soM |= Φ(f(aη1, . . . aηk); aν|n⌢<ν(n)+1>) ↔Φ(f(aη1, . . . aηk); aν|n⌢<ν(n)−1>).Thus, f(aη1, . . . aηk) cannot realize pν.1.15 Remarks.i) Note that in Theorem 1.8 we were able to use anyexpansion of T as T1 so the result is actually for PC∆-classes.InTheorem 1.14 our choice of T1 was constrained, so the result is truefor only elementary as opposed to pseudoelementary classes. The caseof unstable elementary classes could be handled by the second methodthus simplifying the combinatorics at the cost of weakening the result.ii) While we have dealt only with models and theories of cardinality 2ω,the result extends immediately to models of any larger cardinality andstraightforwardly to theories of cardinality κ with κω = κ.2Classifiable examplesWe begin by giving an example of a classifiable theory having a pair of non-isomorphic, potentially isomorphic models. We then extend this result to aclass of weakly minimal theories.11 Let the language L0 consist of a countable family Ei of binary relationsymbols and let the language L1 contain an additional uncountable set ofunary predicates Pη. We first construct an L0-structure that is rigid butcan be forced by a ccc-forcing to be nonrigid. Our example will be in thelanguage L0 but we will use expansions of the L0-structures to L1-structuresin the argument.We now revise the definitions leading up to the notion of an amenablestructure in Section 1 by replacing the underlying structure on Q≤ω by onewith universe 2ω. In particular, Dη, Sη, and C are now being redefined.2.1 Notation.i) For η ∈2ω, let Dη = {σ ∈2ω : σ(2n) = η(n)} and Sη = {σ ∈Dη : σ(2n + 1) is 0 for all but finitely many n} ∪{bη}, where bη is anyelement of Dη satisfying bη(2n + 1) = 1 for infinitely many n.LetC = ∪η∈2ωSη.ii) Let M∗be the L1-structure with universe 2ω where Ei(σ, τ) holds ifσ|i = τ|i, and the unary relation symbol Pη holds of the set Sη. LetM1 be the L1-substructure of M∗with universe C.iii) Any subset A of C inherits a natural L1 structure from M1 with Pηinterpreted as Sη ∩A.2.2 Definition. An L1-substructure M0 of M1 is amenable if for all η ∈2ω,all n ∈ω and all s ∈2n, if there is a ν ∈Pη(M1) with ν|n = s then there isa ν′ ∈Pη(M0) with ν′|n = s.Note that any L1-elementary substructure of M1 is amenable. Moreover,it easy to see that i) each Dη is a perfect tree, ii) 2ω is a disjoint union ofthe Dη and iii) for each s ∈2<ω there are 2ω sequences η such that s has anextension b ∈Dη.2.3 Theorem. The theory FERω of countably many refining equivalencerelations with binary splitting has a pair of models of size the continuumwhich are not isomorphic but are potentially isomorphic.This result follows from the next two propositions and the fact that M1is not rigid.12 2.4 Proposition. There is an L1-elementary substructure M0 of M1 suchthati) |Pη(M1) −Pη(M0)| ≤1.ii) M0|L0 is rigid.Proof. Note that each automorphism of M1|L is determined by its restric-tion to the eventually constant sequences so there are only 2ω such. Thus wemay let ⟨fi : i < 2ω⟩enumerate the nontrivial automorphisms of M1|L. Wedefine by induction disjoint subsets Ai, Bi of M1 each with cardinality lessthan the continuum. We denote ∪i At stage i, choose α ∈M1such that fi moves α. Then, by continuity, there is a finite sequence s suchthat every element of Ws = {τ : s ⊆τ} is moved by fi. Since |Ai|, |Bi| < 2ωand by condition iii) of the remark after the definition of amenable there arean η ∈2ω and a β ∈Pη∩(fi(Ws)−Ai). Then let Bi = {β} and Ai = Sη−{β}.Finally, let M0 = M1 −B2ω.Since no element is ever removed from an Ai, condition i) is satisfied.It is easy to see that M0 is rigid, as any nontrivial automorphism h of M0would extend in a unique way to an automorphism fi of M1 but at step i weensured that the restriction of fi to M0 is not an automorphism.2.5 Proposition. If M0 is an amenable substructure of M1, M0 and M1 arepotentially isomorphic.Proof. Let P be the collection of all finite partial L1-isomorphisms be-tween M0 and M1.We first claim that P is a ccc set of forcing conditions. In fact, P =∪n∈ωFn where if p, q ∈Fn then p ∪q ∈P. Given p ∈P, fix an (arbitrary)enumeration ⟨a1 . . . an⟩of dom p. Let n(p) be the cardinality of dom p andlet k(p) be the least integer satisfying the following conditions.i) If i ̸= j, M0 |= ¬Ek(ai, aj).ii) If i ̸= j, M1 |= ¬Ek(p(ai), p(aj)).iii) For all n with 2n + 1 ≥k, ai(2n + 1) = 0.iv) For all n with 2n + 1 ≥k, p(ai)(2n + 1) = 0.13 Now define an equivalence relation on P by p ≃q if n(p) = n(q), k(p) =k(q) and letting ⟨a1 . . . an⟩enumerate dom p, ⟨a′1 . . . a′n⟩enumerate dom q,for each i, ai|k(p) = a′i|k(p) and p(ai)|k(p) = q(a′i)|k(p). Then ≃has onlycountably many equivalence classes; these classes are the Fn. We must showthat if p ≃q then p ∪q ∈P. It suffices to show that for all i, j if M0 |=En(ai, a′j) then M1 |= En(p(ai), q(a′j)).• Case 1: i ̸= j. By the definition of k = k(p), M0 |= ¬Ek(ai, a′j). Let ℓbe maximal so that M0 |= Eℓ(ai, a′j). Then ℓ≥n since M0 |= En(ai, a′j).Since aj|k = a′j|k, ℓis also maximal so that M0 |= Eℓ(aj, ai). As p isan isomorphism, M1 |= Eℓ(p(ai), p(a′j)). But p(aj)|k = q(a′j)|k, so ℓis also maximal with M1 |= Eℓ(q(a′j), p(ai)). Since n ≤ℓwe concludeM1 |= En(q(a′j), p(ai)) as required.• Case 2: i = j. Suppose ai ∈Sν and a′j ∈Sη. We have M0 |= Ek(ai, a′j)by the definition of k and similarly, M1 |= Ek(p(ai), q(a′i)). Now weshow by induction that for each m ≥k, M0 |= Em(ai, a′i) if and onlyif M1 |= Em(p(ai), q(a′i)). Assuming this condition for m we show itfor m + 1. If m + 1 is odd, the result is immediate by parts iii) andiv) of the conditions defining ≃. If m = 2u then ai(m) = ν(u) anda′i(m) = η(u). Since p and q are L1-isomorphisms p(ai)(m) = ν(u) andq(a′i)(m) = η(u). But M0 |= Em(ai, a′i) implies ν(u) = η(u) so we haveM1 |= Em(p(ai), q(a′i)).To show that the generic object is a map with domain M0, it suffices toshow that for any p ∈P and any a ∈M0 −dom p there is a q ∈P with q ≤pand dom q = dom p ∪{a}. Choose n so that the members of dom p ∪{a} arepairwise En-inequivalent. Fix any L1-automorphism g of M∗that extends p.Let s = g(a)|n and Ws = {γ ∈2ω : s ⊆γ}. Since there is a ν ∈C ∩Ws andB is amenable, there is a ν′ ∈B ∩Ws. Choosing ν′ for b, p ∪⟨a, b⟩is therequired extension of p.Since M0 and M1 are isomorphic in a generic extension for this forcing,we complete the proof.2.6 Remark. The notion of a classifiable theory having two non-isomorphic,potentially isomorphic models is not very robust, and in particular can be lostby adding constants. As an example, let FER∗ω be an expansion of FERω14 formed by adding constants for the elements of a given countable model ofFERω. Then every type in this expanded language is stationary and theisomorphism type of any model of FER∗ω is determined by the number ofrealizations of each of the 2ω non-algebraic 1-types.Thus, if two modelsof FER∗ω are non-isomorphic then they remain non-isomorphic under anycardinal-preserving forcing.Similarly, non-isomorphism of models of the theory CEFω of countablymany crosscutting equivalence relations (i.e., Th(2ω, Ei)i∈ω, where Ei(σ, τ)iffσ(i) = τ(i)) is preserved under ccc forcings.We next want to extend the result from Theorem 2.3 to a larger classof theories. Suppose T is superstable and there is a type q, possibly over afinite set e of parameters, and an e-definable family {En : n ∈ω} of properlyrefining equivalence relations, each with finitely many classes that determinethe strong types extending q. Let T be such a theory in a language L andlet M be a model of T. Let L0 be a reduct of L containing the En’s.We say ⟨aη ∈M : η ∈X ⊆2ω⟩is a set of unordered tree L-indiscerniblesif the following holds for any two sequences η, ν from X:If η and ν realize the same L0-type then ⟨aη1, . . . aηn⟩and ⟨aν1, . . . aνn⟩satisfy the same L-type.We say that a superstable theory T with a type of infinite multiplicityas above embeds an unordered tree if there is a model M of T containing aset of unordered tree L-indiscernibles indexed by 2ω. We deduce below theexistence of potentially isomorphic nonisomorphic models of weakly minimaltheories which embed an unordered tree. Every small superstable, non-ω-stable theory has a type of infinite multiplicity with an associated family of{En : n < ω} of refining equivalence relations and a set of tree indiscerniblesin the sense of [1].The existence of such a tree of indiscernibles sufficesfor the many model arguments but does not in itself suffice for this result.Marker has constructed an example of such a theory which does not embedan unordered tree.However, an apparently ad hoc argument shows thisexample does have potentially isomorphic but not isomorphic models.2.7 Notation. Given A = {aη : η ∈2ω} a set of unordered tree indis-cernibles let D = {aη ∈A : η(n) = 0 for all but finitely many n}. For η ∈2ωlet pη(x) ∈S1(D) be q(x) ∪{En(x, aν) : aν ∈D and ν|n = η|n}. Note thatD is a dense subset of A, each aη realizes pη and each pη is stationary.15 2.8 Lemma. Let T be a weakly minimal theory that embeds an unorderedtree. Fix A and D as described in Notation 2.7. There is a set X satisfyingthe following conditions:i) X ∪A is independent over the empty set;ii) for any Y with D ⊆Y ⊆A, and any η ∈2ω, pη is realized in acl(XY )if and only if pη is realized in Y ;iii) for any Y with D ⊆Y ⊆A, acl(XY ) is a model of T.Proof. It is easy to see from the definition of unordered tree indiscernibilitythat if X = ∅, then conditions i) and ii) of the Lemma are satisfied for anyY ⊆A. We will show that for any X and Y with D ⊂Y ⊆A with XYsatisfying conditions i) and ii) and any consistent formula φ(v) over acl(XY )that is not satisfied in acl(XY ) it is possible to adjoin a solution of φ toX while preserving the conditions. By iterating this procedure we obtain amodel of T.Now suppose there is a Y with D ⊆Y ⊆A, such that acl(XY ) is not anelementary submodel of the monster. Choose a formula φ(x, c, a) with c ∈Xand a ∈Y such that φ(x, c, a) has a solution d in M but not in acl(XY ). Ifwe adjoin d to X we must check that conditions i) and ii) are not violated.Since T is weakly minimal and d ̸∈acl(XY ), XAd is independent. Supposefor contradiction that for some a′ ∈Y , pν is not realized in a′ but pν isrealized in acl(Xda′) by say e. Since condition ii) holds for XY , e ̸∈acl(Xa′).Therefore by the exchange lemma d ∈acl(Xea′). Let θ(v, c′, a,′ e) with c′ ∈Xand a′ ∈Y witness this algebraicity. Thenχ(c, c′, a, a′, z) = (∃x)[φ(x, c, a) ∧θ(x, c′, a′, z)] ∧(∃=mx)θ(x, c′, a′, z)is a formula over Xaa′ satisfied by e.Moreover, e ̸∈acl(Xaa′).For, ifso, transitivity would give d ∈acl(Xaa′) ⊆acl(XY ).Now tp(e/Xaa′)and in particular χ(c, c′, a, a′, z) is implied by pν and the assertion thatz ̸∈acl(Xaa′). Since XA is independent, it follows by compactness thatthere is b ∈D such that χ(c, c′, a, a′, b) holds.So there is a solution ofφ(x, c, a) in the algebraic closure of XY . This contradicts the original choiceof φ so we conclude that condition ii) cannot be violated.16 2.9 Theorem. If T is a weakly minimal theory in a language of cardinalityat most 2ℵ0 that embeds an unordered tree then T has two models that arenot isomorphic but are potentially isomorphic (by a ccc-forcing).Proof. Let L be the language of T. Assume that the type q is based ona finite set e. Let T ′ be the expansion of T formed by adding constants fore. Let M be a large saturated model of the theory T ′ and let the sets A andD be chosen as in Lemma 2.8 applied in L′ to T ′.Recall the definition of C from Notation 2.1. For any W ⊆C, let M′W bethe L′-structure with universe acl(X ∪{aη : η ∈W}) and denote M′W|L byMW. We will construct an amenable set W such that MW ̸≈MC. Since bothare amenable, there is a forcing extension where W ≈C as L1-structures.Since {aη : η ∈C} is a set of unordered tree L′-indiscernibles, the inducedmapping of {aη : η ∈W} into {aη : η ∈C} is L′-elementary. Thus, M′W ≈L′M′C and a fortiori MW ≈L MC.To construct W, let {fη : η ∈2ω} enumerate all L-embeddings of De intoMC. Note that each pη can be considered as a complete L-type over De.Let W = ∪η∈2ωS′η where S′η = Sη −{bη} if MC realizes fη(pbη) and S′η =Sη if MC omits fη(pbη). (see Notation 2.1. )Suppose for contradiction that g is an L-isomorphism between MW andMC.Then for some η, g|D = fη.Now if cη ∈W, the definition of Wyields fη(pη) is not realized in MC.This contradicts the choice of g asan isomorphism.But if cη is not in W then by the construction of W,fη(cη) = g(cη) does not realize g(pη). But this is impossible since g is ahomomorphism.References[1] J.T. Baldwin. Diverse classes. Journal of Symbolic Logic, 54:875–893,1989. [2] Steve Buechler and Saharon Shelah. On the existence of regular types.Annals of Pure and Applied Logic, 45:207–308, 1989. [3] Bradd Hart. Some results in classification theory. PhD thesis, McGillUniversity, 1986.17 [4] S. Shelah. Classification Theory and the Number of Nonisomorphic Mod-els. North-Holland, 1978. [5] S. Shelah. Existence of many L∞,λ-equivalent non-isomorphic models ofT of power λ. Annals of Pure and Applied Logic, 34, 1987. [6] S. Shelah. Universal classes: Part 1. In J. Baldwin, editor, ClassificationTheory, Chicago 1985, pages 264–419. Springer-Verlag, 1987. SpringerLecture Notes 1292. [7] S. Shelah. Classification Theory and the Number of Nonisomorphic Mod-els. North-Holland, 1991. second edition.18 출처: arXiv:9301.208 • 원문 보기