Sei sulla pagina 1di 9

CONSTRUCTING STRONGLY EQUIVALENT NONISOMORPHIC MODELS FOR UNSUPERSTABLE THEORIES, PART C

Tapani Hyttinen and Saharon Shelah

Abstract In this paper we prove a strong nonstructure theorem for (T ) -saturated models of a stable theory T with dop. This paper continues the work started in [HT].

1. Introduction and basic denitions By a strong nonstructure theorem we mean a theorem, which claims that in a given class of structures, there are very equivalent nonisomorphic models. The equivalence is usually measured by the length of Ehrenfeucht-Fraisse games in which has a winning strategy. The idea behind this is, that if models are very equivalent but still nonisomorphic, they must be very complicated, i.e. there is a lot nonstructure in the class. For more background for the theorems of this kind, see [HT]. In this paper we prove the following strong nonstructure theorem (see Denitions 1.2 and 1.3). 1.1 Theorem. Let T be a stable theory with dop and = cf () = (T ) + a <(T ) 1 , = < > + and for all < , < . Then there is F -saturated model M0 |= T of power such that the following is true: for all + , -trees t there a is a F -saturated model M1 of power such that M0 M1 and Mo M1 . = t
a In [HT] Theorem 1.1 was proved for F -saturated models of a countable superstable theory with dop. There we used Ehrenfeucht-Mostowski models to construct
modified:1997-09-15

revision:1997-09-15

Research supported by the United States-Israel Binational Science Foundation. Publ. 602.

602

the required models. To prove that the models are not isomorphic, it was essential that the sequences in the skeletons of the models were of nite length. In the case of unsuperstable theories we cannot quarantee this. Another problem was, of course, a that with Ehrenfeucht-Fraisse models we cannot construct more than F -saturated models. a In this paper we overcome these problems by using F -prime models instead of Ehrenfeucht-Mostowski models. 1.2 Denition. (i) Let be a cardinal and an ordinal. Let t be a tree (i.e. for all x t , the set {y t| y < x} is well-ordered by the ordering of t ). If x, y t and {z t| z < x} = {z t| z < y} , then we denote x y , and the equivalence class of x for we denote [x] . By a , -tree t we mean a tree which satises: (a) |[x]| < for every x t ; (b) there are no branches of length in t ; (c) t has a unique root; (d) if x, y t , x and y have no immediate predecessors and x y , then x = y . (ii) If is a tree and is an ordinal then we dene the tree = ( , <) so that (x, y) < (v, w) i y < w or y = w and x < v . 1.3 Denition. Let t be a tree and a cardinal. The Ehrenfeucht-Fraisse game of length t between models A and B , G (A, B) , is the following. At each t move : (i) player chooses x t , < and either a A , < or b B , < , we will denote this sequence by X ; (ii) if chose from A then chooses b B , < , else chooses a A , < , we will denote this sequence by Y . must move so that (x ) form a strictly increasing sequence in t . must move so that {(a , b )| , < } is a partial isomorphism from A to B . The player who rst has to break the rules loses. We write A B if has a winning strategy for G (A, B) . t t The following theorem is frequently used in this paper. 1.4 Theorem. ([Sh]) Let T be a stable theory. Assume I is an innite indiscernible sequence over A , I B and J I is countable. (i) Av(I, B) does not fork over J and Av(I, J ) is stationary. (ii) I {a} is indiscernible over A i t(a, A I) = Av(I, A I) . Proof. See [Sh] Lemma III 4.17. 1.5 Corolary. Let T be a stable theory. Assume I is an innite indiscernible sequence over A and J I is innite. Then I J is independent over A J . Proof. Follows immediately from Theorem 1.4. 2. Construction Through out this paper we assume that T is a stable theory with dop, = cf () = (T ) + <(T ) 1 , = < > + and for all < , < . 2

602

revision:1997-09-15

modified:1997-09-15

2.1 Theorem. ([Sh]) There are models Ai , i < 3 , of cardinality < and innite indiscernible sequence I over A1 A2 such that (i) A0 A1 A2 , A1 A0 A2 , (ii) Av(I, I A1 A2 ) A1 , Av(I, I A1 A2 ) A2 , (iii) t(I, A1 A2 ) is almost orthogonal to A1 and to A2 , (iv) if Bi , i < 3 are such that B0 A0 A1 A2 , B1 A1 B0 A2 B2 and B2 A3 B0 A1 B1 then t(I, A1 A2 ) t(I, A1 A2
i<3

B3 ).

Proof. This is [Sh] X Lemma 2.4, except that in (iv), only () stp(I, A1 A2 ) t(I, A1 A2
i<3

B3 )

is proved. But since r (T ) , by [Sh] XI Lemma 3.1 A1 A2 is a good set. It is easy to see that this together with (*) implies t(I, A1 A2 ) t(I, A1 A2
i<3

B3 ).

In [HT] the following theorem is proved. 2.2 Theorem. ([HT] Theorem 3.4) There is a + , + 1 -tree such that it has a branch of length and for every + , -tree t there is a + , -tree such that . t
modified:1997-09-15

revision:1997-09-15

Let be a tree. We dene a model M () . Let A, B, C and I be as A0 , A1 , A2 and I in Theorem 2.1. We may assume that |I| = . For all t we choose At , Bt and Ct so that (i) there is an automorphism ft (of the monster model) such that ft (Bt ) = B , ft (Ct ) = C and ft1 A = idA , (ii) Bt Ct A {Bs Cs | s , s = t} . For all s, t , s < t , we choose Ist so that (i) there is an automorphism gst such that gst Bs = fs Bs , gst Ct = ft Ct and gst (Ist ) = I , (ii) Ist Bs Ct {Bp Cp | p } {Ipr | p, r , p < r, p = s or r = t} . a We dene M () to be the F -primary model over S() = {Bt Ct | t } {Ist | s, t , s < t} . By Theorem 2.2, Theorem 1.1 follows immediately from the theorem below. 2.3 Theorem. Let be as in Theorem 2.2 and M0 = M () . Assume t is a + , -tree. Let be a + , -tree such that . If M1 = M () , then t M0 M1 , Mo M1 and the cardinality of the models is . = t The claim on the cardinality of the models follows immediately from the assumptions on . The other two claims are proved in the next two chapters. Notice that in there are no brances of length . Since in there is such a branch, this enables us to prove the nonisomorphism of the models. 3

602

3. Equivalence In this chapter we prove the rst part of Theorem 2.3. We want to remind the reader of the assumptions made in the beginning of Chapter 2. Let (S(), {di | i < }, (Di | i < )) and (S(), {ei | i < }, (Ei | i < )) be a F -constructions of M () and M () , respectively, see [Sh] IV Denition 1.2. If we choose the constructions carefully we can assume = = . We enumerate and : = {t | i < } and = {t | i < } . Furthermore we i i do this so that if t < t then i < j , {, } . If , we write () = {t | i < } i j i and similarly for () . We also enumerate all Ist : Ist = {ai | i < } . st We write S(, ) for {Bt | t ()} {Ct | t ()}

{ai | s < t, s, t (), i < } st and similarly for S(, ) . If < and g : () () is a partial isomorphism then by g we mean the function from S(, ) onto S(, ) which satises: (i) if g(t) = t then for all a Bt and b Ct , g (a) = ft1 (ft (a)) and g (b) = ft1 (ft (b)) , 1 (ii) if g(t) = t , g(s) = s , t < s and a Its then g (a) = gt s (gts (a)) . 3.1 Lemma. If < and g : () () is a partial isomorphism then g is a partial isomorphism.
modified:1997-09-15

Proof. Immediate by the denitions. We write M (, ) = S(, ) {di | i < } and similarly for M (, ) . We say that < is good if for all i < , D i M (, ) and Ei M (, ) . Notice that the set of all good ordinals is cub in . Notice also a that the set of those ordinals < for which M (, ) is F -saturated, is -cub, i.e. it is unbounded in and closed under increasing sequences of conality . 3.2 Lemma. Assume A B , ai and Ci , i < , are such that (i) Ci A {aj | j < i} is of power < , (ii) t(ai , Ci ) t(ai , B {aj | j < i}) . Then for all sequences d {ai | i < } , there is D A of power < such that t(d, D) t(d, B) . Especially, d A B . Proof. See the proof of [Sh] Theorem IV 3.2. 3.3 Lemma. Let < be good, < < , g : () () is a partial isomorphism, f : M (, ) M (, ) is a partial isomorphism and g S(, ) f . Then f g is a partial isomorphism from M (, ) S(, ) onto M (, ) S(, ) . Proof. Follows immediately from Lemmas 3.1, 3.2 and the denition of a good ordinal. 4

602

revision:1997-09-15

3.4 Lemma. Assume < is good, g : () () and f : M (, ) M (, ) are partial isomorphism, g f and (, a)a() (, f (a))a(). If A M0 is of power < then there are good < , partial isomorphisms g : ( ) ( ) and f : M (, ) M (, ) such that (g ) f , f f , g g and A M (, ) . Proof. By playing the Ehrenfeucht-Fraisse game we can nd a good < such that (i) there is a partial isomorphism g : ( ) ( ) such that g g , a a (ii) M (, ) is F -primary over S(, ) and M (, ) is F -primary over S(, ) , (iii) A M (, ) . By (i) above and Lemma 3.3, f (g ) is a partial isomorphism from M (, ) a S(, ) onto M (, ) S(, ) . From (ii) it follows that M (, ) is F -primary a over M (, ) S(, ) and M (, ) is F -primary over M (, ) S(, ) . So the a existence of the required f follows from the uniqueness of the F -primary models ([Sh] Conclusion IV 3.9). 3.5 Theorem. M0 M1 . t

Proof. By Lemma 3.4, it is easy to translate the winning strategy of in G (, ) to her winning strategy in G (M0 , M1 ) . t t
modified:1997-09-15

4. Nonisomorphism In this chapter we prove the second part of Theorem 2.3, i.e. M0 M1 . Again = we want to remind the reader of the assumptions made in the beginning of Chapter 2. For a contradiction we assume that f : M0 M1 is an isomorphism. If a M0 then we write a for the least such that a M (, ) and similarly for a M1 . By A we mean {a | a A} . Let X be such that |X| = and for all x, y X if x = y then either i i x < y or y < x . For every x X we choose ui , Sx and Nx , i {0, 1} , so that x 0 1 (i) x ux and ux , i x (ii) Sx = {Bt | t ui } {Ct | t ui } {Ist | s, t ui , s < t} , where x x x x Ist Ist is of innite power at most , i a i 0 (iii) Nx Mi is F -primary over Sx and furthermore if a Nx S() and 0 1 a = di in the construction of M0 then Di Nx and similarly for Nx , 0 1 (iv) f Nx is onto Nx , i (v) |Nx | , a 0 (vi) if M (, ) is F -saturated, then so is M (, ) Nx . It is easy to see that these sets exist. 4.1 Lemma. Assume Ai , i < , are sets of power . Then there are X and B such that |X| = and for all i, j X , Ai Aj = B . 5

602

revision:1997-09-15

Proof. Without loss of generality we may assume that for all i < , Ai . We dene f () = sup(Ai (j<i Aj )) . Since > + is regular, this function is regressive on a stationary set. So by Fodors lemma, it is constant on some set X of power . Since for all < , < , the claim follows by the pigeon hole principle. By Lemma 4.1 and the pigeon hole principle we may assume that X is chosen so that it satises the following: (i) There are ui , S i and N i , i {1, 2} , such that for all x, y X , if x = y i i i i then ui ui = ui , Sx Sy = S i and Nx Ny = N i . x y o o 0 (ii) For all x X , M (, N 0 ) Nx = N 0 and if x < y then M (, Nx ) Ny = 0 N and similarly for 1 instead of 0 . i i i (iii) For all x, y X , there are elementary maps fxy : Nx Ny and an order i isomorphisms gxy : ui ui such that y x i i 0 (a) fxy N i = idN i , gxy ui = idui and gxy (x) = y , 1 i (b) for all t ui and a Bt Ct , fxy (a) = fg i (t) (ft (a)) , x
i (c) for all s, t ui , s < t , fxy x y x Ist is onto Ig i 1 fxy (f (a)) .
xy i xy (s)gxy (t)

(d) for all a

0 Nx ,

0 f (fxy (a))

4.2 Lemma. Let x, y X , x < y . a (i) N i is F -primary over S i . i a i (ii) Nx is F -primary over N i Sx . i i (iii) N i S i Sx Sy . i i (iv) Nx N i Ny . o o (v) Ixy Bx Cy Nx Ny . Proof. Immediate by (ii) in the choice of X and Lemma 3.2.
modified:1997-09-15

4.3 Corollary. Let x, y X , x < y . 0 0 0 0 (i) If A, B and C are such that A N 0 Nx Ny , and B Nx N 0 A Ny C then t(Ixy , Bx Cy )
o 0 t(Ixy , Ixy Nx Ny A B C) 0 0 (ii) t(Ixy Nx Ny , ) does not depend on x and y .

Proof. (i) By the rst assumption on A and Lemma 4.2 (iii) A N 0 S 0 Bx C y . By the construction of M0 , this implies

revision:1997-09-15

(A) (B) and (C)

A N 0 A Bx C y .
0 0 B Nx Bx N 0 A Ny C 0 0 C Ny Cy N 0 A Nx B.

From the second assumption it follows easily that

By Theorem 2.1 (iv), (A),(B) and (C) imply the claim. 0 0 (ii) By (iii) in the choice of X and Lemma 4.2 (iv), for all x < y , fxx fyy is an elementary map. So the claim follows from (A), (B) and (C) above and Theorem 2.1 (iv). c For x, y X , x < y , let Ixy be some countable subset of Ixy . 6

602

4.4 Lemma. Assume x, y X , x < y . Then there are s u1 u1 and x t u1 u1 such that either y c c c c (i) s < t and Av(f (Ixy ), f (Ixy Bx Cy )) is not orthogonal to Av(Ist , Ist Bs Ct ) , or c c c c (ii) t < s and Av(f (Ixy ), f (Ixy Bx Cy )) is not orthogonal to Av(Its , Its Bt C s ) . Proof. For a contradiction, we assume that such s and t do not exist. Let 0 (x, y) = {(s, t)| s < t and s u1 u1 , t u1 u1 or t u1 u1 , s u1 u1 } x y x y 1 (x, y) = {(s, t)| s < t and s u1 u1 , t u1 u1 or t u1 u1 , s u1 u1 } x y x y and 2 (x, y) = {(s, t)| s < t and s u1 u1 , t u1 u1 or t u1 u1 , s u1 u1 } . y x y x For i {0, 1, 2} , let 1 1 S i (x, y) = S() (Sx Sy ji {Ist | (s, t) j (x, y)}) and Ri (x, y) = {Ist | (s, t) i (x, y)} . 1 1 Now it is easy to see that S 0 (x, y) S 1 Sx Sy . By Lemma 3.2 N 1 S 1 S 0 (x, y) 1 1 Sx Sy . So 1 1 S 0 (x, y) N 1 Sx Sy . By Lemma 4.2 this implies
1 1 (A) S 0 (x, y) N 1 Nx Ny .

modified:1997-09-15

By the construction
1 1 (B) R0 (x, y) Sx S 1 S 0 (x,y) R1 (x, y) Sy .

By Lemma 3.2
1 1 1 Nx Sx S 0 (x, y) R0 (x, y) R1 (x, y) Sy

and so
1 1 1 R0 (x, y) Nx Sx S 0 (x,y)R0 (x,y) R1 (x, y) Sy .

By (B) this implies

revision:1997-09-15

(C)

1 1 R0 (x, y) Nx N 1 S 0 (x,y) R1 (x, y) Sy .

By Lemma 3.2 and (ii) in the choice of X ,


1 1 1 Ny Sy S 0 (x, y) R0 (x, y) R1 (x, y) Nx

and so
1 1 1 R1 (x, y) Ny Sy S 0 (x,y)R1 (x,y) R0 (x, y) Nx .

By (C) this implies


1 1 (D) R1 (x, y) Ny N 1 S 0 (x,y) R0 (x, y) Nx .

602

1 1 Then by (A), (D) and Corollary 4.3 (i), f (Ixy ) is indiscernible over Nx Ny S 2 (x, y) . By Lemma 3.2 and (ii) in the choice of X , we see that for all (s, t) 2 (x, y) , 1 1 Ist is indiscernible over Nx Ny S 2 (x, y) and (Ist )(s,t)2 (x,y) is independent over 1 1 Nx Ny S 2 (x, y) . For all (u, v) 2 (x, y) {(x, y)} we choose innite Iuv Iuv of power < such that 1 1 (i) for all (u, v) 2 (x, y) , if we write B(u, v) = Nx Ny S 2 (x, y) Iuv , then Iuv Iuv B(u,v) f (Ixy ) {Ist | (s, t) 2 (x, y), (s, t) = (u, v)}.

c 1 1 (ii) Ixy Ixy and if we write B(x, y) = Nx Ny S 2 (x, y) f (Ixy ) , then f (Ixy Ixy ) B(x,y) {Ist | (s, t) 2 (x, y)}.

Because | 2 (x, y)| < , it is easy to see that such Iuv exist. c c c c Since Av(f (Ixy ), f (Ixy Bx Cy )) is orthogonal to Av(Ist , Ist Bs Cs ) for all 2 (s, t) (x, y) we see that Ixy Ixy is indiscernible over S() . Because |Ixy Ixy | = , this contradicts [Sh] Theorem IV 4.9 (2). If s, t , then we write st for the set of all innite J such that for some J , J J and there is an automorphism g for which g Bs = fs Bs , g Ct = ft Ct and g(J ) = I .

modified:1997-09-15

4.5 Lemma. Assume x, y X , x < y , s u1 u1 , t u1 u1 and s and t x y c c are incomparable in . If J st , then Av(f (Ixy ), f (Ixy Bx Cy )) is orthogonal to c c Av(J, J Bs Ct ) . Also if J ts , then Av(f (Ixy ), f (Ixy Bx Cy )) is orthogonal to Av(J, J Bt Cs ) .
c c Proof. For a contradiction assume that Av(f (Ixy ), f (Ixy Bx Cy )) is not orthogonal to Av(J, J Bs Ct ) , the other case is similar. Then we can choose J so that in addition, |J | = and J M1 . By Theorem 2.1 (iv), J is indiscernible over S() . By [Sh] Theorem IV 4.14, a Av(J, M1 ) is F+ -isolated. Then we can nd a model D M1 of power such that c (a) f (Ixy Bx Cy ) J Bs Ct D , c (b) Av(f (Ixy ), D) is not almost orthogonal to Av(J, D) , (c) Av(J, D) Av(J, M1) . (For (c), notice that because D is a model, t(a, D) stp(a, D) .) But since |D| < c and |f (Ixy )| = , it is easy to see that Av(f (Ixy ), D) is satised in M1 , a contradiction. Let x, y X be such that x < y . By Lemma 4.4 we can nd sxy and txy such c c that there is J sxy txy txy sxy for which Av(f (Ixy ), f (Ixy Bx Cy )) is not orthogonal to Av(J, J Bsxy Ctxy ) or to Av(J, J Btxy Csxy ) . By Lemma 4.3 (ii) we can choose these so that for all y and y from X , if x < y and x < y then sxy = sxy . We call this element just sx . Similarly we can choose txy so that it does not depend on x ( x < y ). We call this element ty .

revision:1997-09-15

602

4.6 Lemma.

For all x and x from X , sx and sx are comparable in .

Proof. By Lemma 4.5, for all y X , if y > x and y > x then ty is comparable to sx and to sx . Since |{z | z sx z sx }| < and if y = y then ty = ty , we can nd y X such that sx < ty and sx < ty , which implies the claim. 4.7 Theorem. M 0 M1 . = Proof. If M0 M1 then by Lemma 4.6 we can nd Y of power such = that for all s, t Y if s = t then either s < t or t < s . Clearly this contradicts the fact that is a + , -tree. Together with Theorem 3.5, Theorem 4.7 implies Theorem 2.3, and so Theorem 1.1 is proved. 4.8 Remark. As in [HT], we can see that Theorem 1.1 implies the following: Under the assumptions of Theorem 1.1, for every + , -tree t there are models Mi |= T , i < + , such that for all i < j < + , Mi Mj and Mi Mj . = t References. [HT] T. Hyttinen and H. Tuuri, Constructing strongly equivalent nonisomorphic models for unstable theories, APAL 52, 1991, 203-248. [Sh] S. Shelah, Classication Theory, Stud. Logic Found. Math. 92 (North-Holland, Amsterdam, 2nd rev. ed., 1990). Tapani Hyttinen Department of Mathematics P.O. Box 4 00014 University of Helsinki Finland Saharon Shelah Institute of Mathematics The Hebrew University Jerusalem Israel Rutgers University Hill Ctr-Bush New Brunswick New Jersey 08903 U.S.A.

revision:1997-09-15

modified:1997-09-15

602

Potrebbero piacerti anche