## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

by Anand Pillay

Ratings:

267 pages4 hours

This introductory treatment covers the basic concepts and machinery of stability theory. Lemmas, corollaries, proofs, and notes assist readers in working through and understanding the material and applications. Full of examples, theorems, propositions, and problems, it is suitable for graduate students in logic and mathematics, professional mathematicians, and computer scientists. Chapter 1 introduces the notions of definable type, heir, and coheir. A discussion of stability and order follows, along with definitions of forking that follow the approach of Lascar and Poizat, plus a consideration of forking and the definability of types. Subsequent chapters examine superstability, dividing and ranks, the relation between types and sets of indiscernibles, and further properties of stable theories. The text concludes with proofs of the theorems of Morley and Baldwin-Lachlan and an extension of dimension theory that incorporates orthogonality of types in addition to regular types.

Publisher: Inscribe DigitalReleased: May 17, 2013ISBN: 9780486150437Format: book

**INDEX **

*α, β, γ *. . . will denote ordinals, *κ, λ, µ *cardinals, and *m, n, k *natural numbers. *i, j *may denote natural numbers, ordinals, or members of some particular set, it being always clear from the context what is meant. β also has a technical meaning when used in the expression *β*(*p*) which is introduced in Chapter 3. We work in ZFC set theory.

Our model-theoretic notation is fairly standard, as in Chang and Keisler (1973) for example, with one or two differences, which we point out below.

*L, L′ . . . *is always a first-order language (with equality). *M, N, *etc. will *denote *structures (i.e. *L*-structures for some *L*) and *A, B, *etc. subsets of structures. If *M *is an *L*-structure, then we also denote the universe (or domain) of *M *by *M. **M *is the cardinality of (the universe of) *M. *We will allow our languages to contain an unlimited supply of variables. Clearly the cardinality of the set of *L*-sentences modulo logical equivalence is unaffected by this. (In fact, as in pointed out later, we shall work essentially with just countable languages, the cardinality of a language being the cardinality of its set of predicates, function symbols, and constant symbols.) *L*(*A*) denotes the language obtained by adjoining to *L *names for the elements of *A. *We do not normally distinguish between elements of *A *and their names. Variables are denoted by *x, y, z, xi*, etc.

is the length of *x*come from *A *∈ *A. *

, this means that all the free variables in the formula are among the members of *x*may also contain ‘parameters’ (i.e. names of elements from some *A**, *.

I of course assume the *compactness theorem *a set of *formulae *of *L*is *consistent *(i.e. there is a *L*-structure *M *and an assignment of elements of *M *is true in *M*is consistent.

Another basic fact is the *Lowenheim-Skolem *theorem. This states that, if *M *is an infinite *L*-structure and *A *⊂ *M, *then

(*i*) For any *λ **M **, **L *), there is *N *>*M *such that *IN *= *λ; *

(ii) If *λ *≤ |*M *and ≥ max(|*A **L *) then there is *N< M *such that *A *⊂ *N *and |*N*| *λ. *

I now discuss notation for the important notion of a type. Let *T *be a complete theory in *L. *A (complete *n*-type of *T *(*n *< *ω*of *L*-formulae in *n *free variables, say *x0, ... , xn*—1, which is maximal consistent with *T. *This is equivalent to saying:

(i) *T *is consistent;

(ii) For any *φ *= *φ*(*x*0, ... , x*n*−1) ∈ *L*, either *φ **φ *.

A type of *T *is an *n*-type for some *n *< *ω*. Note that a type of *T *is closed under conjunctions. The set of *n*-types of *T *is denoted *Sn*(*T*) and we put *S*(*T*. Types are denoted by *p, q, r. *∈ *Sn*(*T*).

If *M *is an *L*-structure, *A *⊂ *M, *is an *n*-tuple from *M, *the *type *over *A *in *M. *∈ *S*(*T*) if there is a model *M *of *T *∈ *M *.

We also use what are in effect types in infinitely many variables, although we have no developed notation for this. Let *M *be an *L*-structure, *A *⊂ *M, *and *B *⊂ *M. *Let *B *be listed as 〈b*i : i *∈ *I*〉 and assume that we have available variables *xi *for *i *∈ *I. *Then by *tpM*(*B/A*) we mean

Of course, this depends (up to permutation or change of, variables) on the particular indexing of *B *that we use. Clearly tp*M*(*B/A*) is essentially the same as *Th*(*M, a*)*a*∈*A *∪ *B *where we replace the names for *B *in the latter by variables.

is the result of replacing *xi *by *yi *for relevant *i. *.

∈ *Sn*(*Th*(*M*, *a*)*a*∈*A*) (*M *an *L*be an *L-*formula. Then by *p **φ *I mean

If *S *is some statement, and *φ *a formula then *φ*if *S *denotes *φ *if *S **φ *if *S *is false.

If *M *is an *L**∈ L*(*M*, then *øM *∈ *Mn M **φ *)}.

A formula with *n *free variables is often called an *n*-formula.

*Sn*(*T*) is sometimes viewed as a topological space, the topology being as follows: for an *n*Then the sets [*φ*] are taken as basic open sets, and the topology is generated accordingly.

An isolated point of *Sn*(*T*is said to isolate *p *(relative to *T *of course).

Let me review a few more of the results which I shall be assuming.

**Proposition 0.1. (The Beth definability theorem.) ***Let *L *be a language, *P *an *n-*ary predicate not in *L *and *L′ = L ∪ {P}. *Let *T′ *be an *L′-*theory. Suppose that any *L-*structure *M *has at most one expansion to an *L′*-structure *M′ *for which *T′. *Then there is an n*-*formula **of *L *such that *.

**Proposition 0.2. (Omitting types.) ***Let *T *be a countable theory, in a countable language *L. *For each *n < *ω let *n *be a set of L-formulae with free variables, **say. Suppose that, for each *n *is nonprincipal over *T, *i.e. that there is no **consistent with *T *such that **for all ψ *n. *Then *T *has a countable model which omits each *n.

*I*(*λ*, *T*) denotes the number of models of *T *of cardinality *λ, *up to isomorphism. If *I*(*λ*, *T*) = 1, we say that *T *is *λ*-categorical.

**Proposition 0.3. ***Let *T *be a complete theory in a countable language *L. *Then *T*is *0*-categorical if and only if, for each *n < *ω*, *all *p ∈ Sn(T) *are isolated. *

A prime model of the theory *T *is a model *M *of *T *such that, for all *N **T, *there is an elementary embedding of *M *into *N*.

**Proposition 0.4. ***Let *T *be a complete theory in the countable language *L.

T, *then *M *is a prime model of *T *if and only if *

M *is countable; *

*For every *∈ M, *tp*) *is isolated *(*as a type of *S(T)).

(ii) *Any two prime models of *T are *isomorphic. *

(iii) *If *S(T) *is countable then, for any model *M *of *T *and *)) *is also countable. *

(iv) *If *S(T) *is countable, then *T *has a prime model. *

If *M **T*∈ *M*, and (*N, *) is a prime model of *Th*((*M, *)), we also say that *N **. *(*N *is clearly a model of *T.*)

*M *is said to be *κ-saturated *if, for any *A *⊂ *M *with |*A*| < *κ*, for all *p*(*x*) ∈ *S*1(*Th*(*M, a*)*a*∈*A *), *p *is realized in *M. *

**Proposition 0.5***. Let *M *be κ-saturated *(*κ *0). *Then *

(*i*) |M| ≥ *κ, if *M *is infinite; *

(*ii*) *If *A ⊂ M, lAl< *κ, and **is a set of formulae of *L(A) *in less than κ free variables which is consistent with *Th(M, a)a∈A, *then **is realized in *M. *In particular every *∈ S(Th(M, a)a∈A) *is realized in *M;

(*iii*) *Let *N ≡ M *and *|N| *κ*; *then there is an elementary embedding of *N *into *M;

(*iv*) *Suppose that *|M| = *κ and that *= *κ, and *N *is also κ-saturated.Then *N ≅ M;

(*v*) *Suppose that *|M| = *κ. Let *A ⊂ M, |A| < *κ, and let f be an elementarymap of *A *into *M (*i. e. *(M, a)a∈A ≡ (M, f(a))a∈A, *or, less accurately, tp*M(A) = *tp*M(f(A))), *then there is an automorphism *g *of *M *which extends *f.

**Note 0.6. **The property *M *is said to have in (v) above is called *κ-homogeneity *(iv) and (v) above are proved by a back-and-forth argument. The rest are easy.

Now any complete theory has *κ*-saturated models for arbitrarily large *κ*be a *κ*-saturated model of *T*. If *N **T*, |*N*| < *κ*, then, by (iii) of Proposition 0.4, *N *. Thus, without loss of generality, all models of *T *of cardinality less than *κ *. If *φ *then *N **φ *whenever *N *< *M *and the parameters from *φ *are in *N. **φ *and so, if *N *is a model of *T, *|*N*| < *κ, *and *φ *is an *L*(*N*)-sentence, then *N **φ **φ.*Now let *A, B *, with *A, B *of cardinality < *κ*. Suppose that *A *has a property *P *taking *A *to *B *whereby *B *also has *P. *If |*M*| ≠ *κ, *be a *κ*taking *A *to *B. *So *again B *has *P *. Let me note that the existence *in general *of a *κ*-saturated model of cardinality *κ *(*κ *0) cannot be shown without some set-theoretic hypothesis on *κ *(e.g. *κ *is weakly inaccessible) although, as we shall see, if *T *is stable, then *T *has such models for arbitrarily large *κ*.

From now on *L *will be a countable language and *T *a *complete *theory in *L *be a *κ*-saturated model of *T *and is thus *κ*taking *A *to *B*can be bypassed *using an argument as in the previous paragraph, or using directly the saturation of M. *

*By a set we will mean a subset of **and by a model an elementary substructure of *. By what we have said above we lose nothing in generality. If *φ **φ **φ *and, as we pointed out above, if *N *is a model and *φ *an *L *(*N*)-sequence, then *N **φ **φ*.

, and carry on as before.

is sometimes referred to as *the big model. *Note that by our conventions, if *M *and *N *are models and *M *⊂ *N, *then this means that *M *< *N *(as *M *and *N< **, A **/A*. Similarly for a set *A*, we write *Sn*(*A*), *S*(*A*.

**Definition 0.7. **

(*i*) *Let λ *0. *T *is *λ-stable *if, for all *A*, |*A *| ≤ *λ *implies |S1(*A*) | ≤ *λ*.

(*ii*) *T *is *stable *if *T *is *λ*-stable for *some λ. *

**Note 0.8**. Clearly *T *is *λ*-stable iff, for all *M*, |*M*| ≤ *λ *implies |*S*1 (*M*) | ≤ *λ*. Note also that, for any *A*, |S1 (*A*) | ≥ |*A*|, because, for *a, b *∈ *A, a ≠ b *implies tp(*a/A*) ≠ tp(*b/A*).

**Lemma 0.9. ***T *is *λ-stable iff, for *n < *ω and *A *with *|*A*|≤ *λ*, |Sn(A) | ≤ *λ*.

**Proof. ⇐ **is immediate.

⇒. By induction on *n. *Suppose, for all *A *with |*A *| ≤ *λ, *|*Sn*(*A*)| ≤ *λ. *Now let |*A*| ≤ *λ *and we consider *Sn*+1 (*A*). Suppose by way of contradiction that |S*n*+1(*A*)| > *λ. *For each *p*(*x*0, ... , *xn*) ∈ S*n*+1(*A*), let *p*′(*x*0, ... , *xn*−1) be the restriction of *p *to the variables *x*0, ... ,*xn*−1. So, clearly, *p′ *∈ *Sn*(*A*). Thus, by hypothesis, there are *pi*(*x*0, ... , *xn) *∈ *Sn*+1(*A*), for *i *< *λ*+, such that *i *< *j *< *λ*+ implies *pi *≠ *pj*(for some *q *∈ *Sn*(*A*)) for all *i *< *j *< *λ*realize *q*. Then it is clear that, for each *i *< *λ*+*, pi*(*c*0, ... , *cn*−1 *xn*and moreover, that *i *< *j *and so we have a contradiction to the *λ*-stability of *T*.

**Proposition 0.10**. *Let *T *be *0*-stable (or as we shall say, ω-stable). Then *T *is λ-stable for all λ. *

**Proof. **We will use the following simple lemma whose proof is left as an exercise.

**Lemma 0.11. ***Let *|A| ≤ *λ*, (*λ *0) *and, for each φ*(x) ∈ L(A), *let us denote by *[*φ*]A *the set *{p(x) ∈ S1(A) : *φ*(x) ∈ p}. *Suppose that *|[*φ*] A |> λ, *where φ *∈ L(A). *Then there is ψ*(x) ∈ L(A) *such that *|[*φ **ψ*]Al > *λ *and l[*φ**ψ*> *λ. *

Now we prove the proposition. Suppose that *T *were not *λ*-stable (*λ *arbitrary). I show that *T *is not *ω**A *≤ *λ **S*1(*A*> *λ*[*x *= *x*> *λ*. (Notation as in Lemma 0.11.) Thus using repeatedly Lemma 0.11 we find *L*(*A*)-formulae *φn*(*x*) for each *n *∈ *ω*>2, such that

(*i*) *φ*<> = ‘*x *= *x*’;

(*ii*) For each *n *[*φn*(*x*> *λ*;

(*iii*) For each *n*and *i **φn*^<*i*>(*x*) →*φn*(*x*):

(*iv*) For each *n *.

Let *B *be the set of elements of *A *that occur in the formulae *φn*(*x*). So clearly *B *is countable. For each *τ *∈ *ω**τ*(*x*) = {*φτ **n*(*x*) : *n *< *ω**τ*(*x*) is a consistent set of *L*(*B**τ*(*x*) to some *pτ*(*x*) ∈ *S*1(*B*) for each *τ *∈ *ω*2 (by Zorn’s lemma).

Then by (iv) *τ*1 ≠ *τ*and *T *is not *ω*-stable.

**Note 0.12. **To explain the notation above, for *X *a set and a an ordinal, *αX *is the set of sequences of length *α *of elements from *X. *

Now I define indiscernible sequences and sets.

**Definition 0.13. **be a total ordering on a set *I*. Let *n *< *ω*, and, for each *i *∈ *I**i *be an *n*of course). Let *A **i *: *i *∈ *I*〉 is said to be *indiscernible over A relative to *, if for any *m *< *ω*, and *i*1 , ... , *im*, *j*1 , ... , *jm *in *I *such that *i**i**im *and *j**j**jm*is an *L*(*A*for *k *= 1 , ... , *m*, then

If, for some *i *≠ *j *in *I**i **j**i *: *i *∈ *I*〉 nontrivial.

(ii) Let *I **i *an *n*-tuple for each *i *∈ *I**i *: *i *∈ *I*〉 is said to be an *indiscernible set *over *A*, if, for any *m *< *ω*, if *i*1 , ... , *im *are distinct elements of *I*, *j*1 , ... , *jm *are distinct elements of *I*is an *L*(*A*)-formula, *then *

**Note 0.14. **Suppose that *α *is an ordinal and < is the usual ordering on *α**i *: *i *< *α*〉 is indiscernible over *A **i *: *i *< *α*〉 is an indiscernible *sequence *over *A*.

Ramsey’s theorem says that, if *m*, *n *< *ω*, *X *is an infinite set and if the set of m element subsets of *X*, (called [*X*] *m*) is partitioned into *n *sets, i.e. [*X*] *m *= *P*1 ∪ *P*2 ∪ ... ∪ *Pn*, then there is *Y *⊂ *X *such that *Y *is infinite and, for some *i*≤ *n*, [*Y*] *m *⊂ *Pi*.

Ramsey’s theorem (for the case *n *= 2) can be used to produce infinite sequences of indiscernibles satisfying certain requirements. See Chang and Keisler (1973) for the details.

is consistent (with *φ*), where *A *. In a few cases we use compactness in this latter sense.

We also engage commonly in a certain kind of arugment which goes as follows; suppose we have sets *A *and *B *and we seek *C *such that tp(*C *∪ *B*/*A*. Suppose that we have found *C′ *and *B′ *such that tp(*C′ *∪ *B′*/*A*and tp(*B′*/*A*) = tp(*B*/*A*). Then we say that we can assume that *B′ *= *B *and that thus *C′ *is the required *C*. This is because there is an automorphism of the big model taking *B′ *to *B *and fixing pointwise *A *whereby, if *C *is the image of *C′ *under this automorphism, then *C *′ be the result of replacing the variables for *B *by the corresponding names for *B*. Then as tp(*C′ *∪ *B′*/*A*and tp(*B′*/*A*) = tp(*B*/*A*′ will be consistent and thus, by the saturation of the big model, will be realized by *C*, say. Then tp(*C *U *B*/*A*and so *C *is as required.

See Chang and Keisler (1973) for attributions of classical results and Propositions 0.1 – 0.5.

The notion of a theory being *κ*-stable is due to Rowbottom (1964). What are in effect *ω*-stable theories were studied by Morley as totally transcendental theories. The definition of a theory being stable is due to Shelah (1971). Lemma 0.10 is due to Morley (1965). The notion of an indiscernible sequence is due to Ehrenfeucht and Mostowski (1956) and the notion of an indiscernible set is due to Morley (1965).

**Definition 1.1. **(i) Let *p*(*x*) ∈ *S*(*A*) and *B *⊂ *A*. *p *is *definable over B *∈ *L*∈ *L*(*B*∈ *A *with *l*.

(ii) *p*(*x*) ∈ *S*(*A*) is said to be *definable *if *p *is definable over *A*.

**Remark 1.2. **∈ *S*(*A*) be definable over *B *⊂ *A*, and let, for each *L*be an *L*(*B*in Definition 1.1 (i). *d *into the set of *L*(*B*) formulae and we call *d *a *defining schema over B *for *p*. If *B *= *A*, *d *is simply a defining schema for *p*.

**Example 1.3. **Let *p*(*x*) ∈ *S*(*A*. Then *p *.*d *is clearly a defining

You've reached the end of this preview. Sign up to read more!

Page 1 of 1

Close Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Loading