Sei sulla pagina 1di 33

The Universality Spectrum : Consistency for more Classes

no. 457

Saharon Shelah Institute of Mathematics, The Hebrew University, Jerusalem, Israel Department of Mathematics, Rutgers University, New Brunswick, N.J., U.S.A.

Last revised, 1 August, 1993

Abstract. We deal with consistency results for the existence of universal models in natural classes of models (more exactlya somewhat weaker version). We apply a result on quite general family to Tfeq and to the class of triangle-free graphs
modified:1994-12-11

0 Introduction: The existence of universal structures, for a class of structures in a given cardinality is quite natural as witnessed by having arisen in many contexts. We had wanted here to peruse it in the general context of model theory but almost all will interest a combinatorialist

revision:1994-12-11

who is just interested in the existence of universal linear order or a triangle free graph. For a rst order theory (complete for simplicity) we look at the universality spectrum USPT = { : T has a universal model in cardinal } (and variants). Classically we know Partially supported by the United States Israel binational science foundation, Publication No. 457, 2, 3 are some years old, 5.2,5.3 done in summer 92, 4 +5.1, rst version done with 2, 3 but written in sp. 92, 11.92 resp.. Revised 12/93. 1

(457)

that under GCH, every > |T | is in USPT , moreover 2< = > |T | USPT (i.e.the existence of a saturated or special model, see e.g. [CK]). Otherwise in general it is hard for a theory T to have a universal model (at least when T is unstable). For consistency see [Sh100], [Sh175], [Sh 175a], Mekler [M] and parallel to this work Kojman-Shelah [KjSh 456] ; on ZFC nonexistence results see Kojman-Shelah [KjSh409], [KjSh447], [KjSh455]. We
get ZFC non existence result (for Tfeq under more restriction , essentially cases of failure

of SCH ) in 2, more on linear orders (in 3), consistency of (somewhat weaker versions of) existence results abstractly (in 4) derived consistency results and apply them to the class of models of Tfeq (an indexed family of independent equivalence relations) and to the class of triangle free graphs (in 5 ). The general theorem in 4 was intended for treating all simple theories (in the sense of [Sh 93] , but this is not included as it is probably too much model theory for the expected reader here (and for technical reasons).
modified:1994-12-11

1 1.1 Denition: For a class K = (K, K ) of models 1) K = {M K : M = } 2) univ(, K) = Min {|P| : P a set of models from K such that for every N K for some N P, M can be K -embedded into N }.

revision:1994-12-11

3) Univ(, K) = Min { N : N K , and every M K can be K -embedded into N }. 4) If K is the class of models of T , T a complete theory, we write T instead (mod T, ) (i.e. the class of model of T with elementary embeddings). If K is the class of models of T , T a universal theory, we write T instead (mod (T ), ).

1.2 Claim: 1) univ(, K) = 1 i K has a universal member of cardinality . 2

(457)

2) Let T be rst order complete, |T | . Then we have univ(, T ) implies univ(, K) = 1 and Univ(, T ) univ(, T ) cf S (Univ(, T ), ) = cov (Univ(, T ), + , + , 2) (see [Sh-g] ; we can replace T with K with suitable properties).

2 The universality Spectrum of Tfeq For Tfeq , a prime example for a theory with the tree order property (but not the strict order property), we prove there are limitations on the universality spectrum; it is meaningful when SCH fails.
2.1 Denition: Tfeq is the model completion of the following theory, Tfeq . Tfeq is dened

as follows: (a) it has predicates P, Q (unary) E (three place, written as yEx z} (b) the universe (of any model of T ) is the disjoint union of P and Q , each innite
modified:1994-12-11

(c) yEx z P (x) & Q(y) & Q(z) (d) for any xed x P , Ex is an equivalence relation on Q with innitely many equivalence classes (e) if n < , x1 , . . . , xn P with no repetition and y1 , . . . , yn Q then for some y Q,
n =1

yEx y .

revision:1994-12-11

(Note: Tfeq has elimination of quantiers).

2.2 Claim: Assume: (a) < < (b) cf = , = cf = cf , + < (c) =: pp() () > + |i | 3

(457)

(d) there is {(ai , bi ) : i < i }, ai []< , bi [] and |{bi : i < i }| such that: for every f : for some i, f (bi ) ai then (1) Tfeq has no universal model in . (2) Moreover, univ(, Tfeq ) = pp() () . Proof: Let D be a -complete lter on , i = cf i < = tcf(
i<

i<

i , tlim D i = , =:

i /D) > i (and for (2), tcf


i<

i<

i /D > univ(, Tfeq )). Also let f : <

be <D -increasing conal in

i /D. Let S = { < : cf = , divisible by +1 }.

Let C = C : S be such that: C a club of , otp(C ) = and [ C > 0 divisible by ] and id a (C) (i.e. for every club E of for stationary many S E, / C E) (exists-see [Sh 365,2]). For (1), let M be a candidate for being a universal model of Tfeq of cardinality ,
modified:1994-12-11

for (2) let M : < exemplify =: univ(, Tfeq ); for (1) let = 1, M0 = M0 . Without

loss of generality |P M | = |QM | = , P M is the set of even ordinals < , QM is the set of odd ordinals < . For each i < i and S and z QM let ai = {2 : ai } and d[z, , i, ] = { :

revision:1994-12-11

nacc C and for some x ai there is y < , such that M |= yEx z but there is no y < sup( C ) such that M |= yEx z}. Clearly d[z, , i, ] is a subset of C of cardinality

|ai | < . Dene gz,,i,


j<

j by: if |ai | < j , C , otp ( C ) = j then gz,,i, (j) =

otp ( C ) where C is Min { : C , > sup(d[z, , i, ] )} and let gz,,i, (j) = 0 if |ai | j . By the choice of f : < for some we have gz,,i, <D f , 4

(457)

let = [z, , i, ] be the rst such . As = tlim D i clearly [z, , i, ] is the rst < such that for the D-majority of i < ,
d[z,,i,] otp (

C ) [f (i), i ); clearly /

it is well dened. Wlog {bi : i < i } = {bi : i < i } As > + + |i |, there is () < such that: z QM , S, i < i , < [z, , i, ] < (). Now we can dene by induction on < , N , such that:
(i) N is a model of Tfeq with universe = (1 + ),

(ii) all x P N are even, all y QN are odd (iii) N increasing continuous, P N = P N+1 (iv) for any x P N there is a y = yx, QN+1 \ QN such that (z QN )[zEx y],
i (v) if S, i < i and bi Min (C ) then there is a z QN+1 \ QN such that i Rang f() = {otp (y C ) : for some x bi , y is minimal such that yEx z } where

bi = {2 : bi }.
modified:1994-12-11

def

[For carrying out this let d,i =df { C : otp (C ) = (f() (j) + 1) for some j < }, so d,i nacc (C ), now choose distinct x,i, bi for d,i . Next choose
i y,i, \ sup(C ) such that it is as in clause (iv) for x,i, and z Ex,i, y,i, .]

If < and f is an embedding of N =

<

N into M , for some i we have

revision:1994-12-11

f (bi ) ai as we can dene f : by f (2) = 2f (), well dened as f maps P N into


P M . Let i1 < be such that bi1 = bi . Let E = { < : (M , N , f )

(M , N, f ) and

> i1 }, clearly it is a club of hence, by the choice of C, for some S we have C E.


i Let z = f (z1 ), so d[z, , i1 , ] is well dened. For each j < there are 0 < 1 in C such def

that otp (C 0 ) = f() (j), otp (C 1 ) = f() (j) + 1 and there is y in (1 \ 0 ) QN


i and x bi1 (= bi ) such that yEx z1 , y minimal for those zi , x. So x = f (x) ai M ,

(457)

i f (zi ) M , and letting y = f (y) we have y < 1 , and y Ex f (z ). Is there y1 < 0 with those properties? if so f (y)Ef (x) y1 , (M , N, f ) |= (t)[t QM & f (t)Ex y1 ] so as i x < 0 , y1 < 0 E1 there is such t < 0 , as E is an equivalence relation f (t)Ex f (z ). N i Now as f is an embedding tEx z , contradicting the choice of y. So y = f (y) witness

1 d[z, , i, ) hence otp (1 C ) gf (z i1 ),,i1 , (j)

We easily get a contradiction.

2.2

2.3 Claim: 1) In 2.2 we can replace clauses (c), (d) by (c)+ , (d) below and the conclusions still hold. (c)+ = ppD () > |i | + , D a lter on , or at least for some i : i < , i = cf i < = tlimJ i : i < and
modified:1994-12-11

i<

i /D is -directed.

(d) {(ai , bi ) : i < i } , ai []< , i or at least {bi : i < i } has cardinality , bi = {i, : < } and for every f : for some i we have { < : f ( i, ) ai } = mod D. 2) Above we can weaken in (c)+ the demand cf (
i< i<

i /D is -directed by

i /D) if in clause (d) we strengthen |= mod D to D.

revision:1994-12-11

Also similarly we can prove 2.4 Claim: Assume (a) < < (b) = cf () = cf () = cf , + < (c) ppD () > |i | + cov (, , + , ), D is -complete. 6

(457)

(d) {(ai , bi ) : i < i } , ai []< , i or at least {bi : i < i } has cardinality , bi = {i, : < } and for every f : for some i we have { < : f ( i, ) ai } = mod D. Then (1) Tfeq has no universal model in (2) moreover univ(, Tfeq ) ppD ()

2.5 Remark: 1) When does (d) of 2.2 hold?; it is a condition on > > , assuming for simplicity > 0 , i = ) e.g. it holds (even with bi = b0 ) if:

()1 for some cardinal we have , = cf , cov (, + , + , ) .

modified:1994-12-11

2) As for condition (d) from claim 2.3, if D is the lter of co-bounded subsets of , it suces to have ()2 for some cardinal we have cov (, , + , ) , or equivalently, [, ) and cf () = imply pp() () .

3) So if = cf () <

()

< + < = cf () < pp() () then by [Sh 460]


()

revision:1994-12-11

condition ()1 holds for some <

4) Why have we require > 0 ? as then by [ Sh-g, Ch. II , 5.4] we can describe the instances of cov by instances of pp ; now even without this restriction this usually holds (see there) and possibly it always hold ; alternatively , we can repeat the proof of 2.2 using cov 5) The parallel of 2.3(2) for 2.4 can be easily stated. 7

(457)

pp() (). Proof: The next step is:

2.7 Question: Let T be f.o. with the tree property without the strict order property; (see [Sh-c]) does 2.2 hold for it?

3 A consequence of the existence of a universal linear order. This section continues, most directly, [KjSh 409].

3.1 Claim: Assume (a) < 2 and 2< + < 2 , is regular. (b) in = + there is a universal linear order
modified:1994-12-11

then , there are f : (for < ) such that: (), for no f : do we have
<

Proof: Assume , fails. We use -tuples of elements to compute invariants. Note that

revision:1994-12-11

2 2< + hence 2 {, + } hence (+ ) = + . Let x : < + list (+ ). Let : < list distinct members of 2 (not necessarily all of them ). Note that as 2< + there is a stationary S I[], S { < + : cf () = } (see [Sh 365 , 2] for the denition of I[S]). As S I[] by [Sh365, 2] there is C = C : S an S-club system such that id p (C), otp C = and / 8

(457)

2.6 Conclusion: If = cf ;

()

, + < = cf < pp+ () then univ(, Tfeq ) ()

f =J bd f .

for each < we have |{C : nacc C }| . Let M be a candidate for being a universal model of Tord of cardinality + , wlog with universe + . For every linear order M with universe + , for every x M (a -tuple of members
x of M ) and S, we dene a (possibly partial) function g = gM, : nacc C as follows:

()0 for nacc C , g() = i for every < we have : () = 1 ( < )( < sup( C )) [ <M x <M <M x ].
x Clearly gM, () can have at most one value . We call (, x) good in M if for every

nacc C there is < such that : x , x realize the same <M -Dedekind cut over {i : i < sup( C )} (necessary if 2< = + ). (The meaning is that for every < , x , x realize the same <M -Dedekind cut over {i : i < sup( C )} ). Let h : nacc C be: h(i) is the (i + 1)-th member of C . We are assuming ,
modified:1994-12-11

x fails, so {gM , h : x 2, S} cannot exemplify it. So we can nd h : M

such that:
x if x (M ), S is (, x) good in M then (gM , h ) satises h =J bd

x (gM , h ).

revision:1994-12-11

Let h = h ; let g : nacc C be h (h1 ) : nacc C . We now as in [KjSh M 409, x.x?????] (using S I[] i.e. ) construct a linear order N = M h with universe + , N =
<

N , N increasing continuous in with universe an ordinal < + and for

each S, there is a sequence y = y : < of members of N+1 such that

()1 if nacc C , g () = , < then


() = 1 ( N )( Nsup(C ) )[ <N y <N <N y ].

(457)

Suppose f : + + is an embedding of N into M , let E = { < + : N universe is and is closed under f , f 1 }. Clearly E is a club of + , hence for some S the set A = (acc E) (nacc C ) is unbounded in (so acc acc E). Let x = x : <
=: f (y ) : < , so we know (similarly to [KjSh 409 3]????) that for A and x x < we have gM , ()() = 1 g () () = 1. Hence A gM , () = g ()

x (gM , h )(otp ( C ) 1) = h (otp ( C ) 1) contradicting the choice of h .

3.1

3.1A Claim: 1) In 3.1 if is a successor cardinal then we can get 0 there are f : for < + such that () for every f for some < + we have f =D f (where D is the club lter on ). 2) If we allow > + , clause (a) of 3.1 holds and (b) below then , of 3.1 holds ;
modified:1994-12-11

similarly in 3.1A(1), where (b) univ(+ , Tord ) Proof: 1) Use [Sh 413, 3.4]. 2) The same proofs.

revision:1994-12-11

So from the existence of a universal linear order of cardinality + , where is as in

3.1+3.1A(1), we get , from this we get below a stronger guessing of clubs.

3.1A

3.2 Claim: Assume is regular uncountable , and 1 there are f : for < + such that: for every f : for some , { < : f () = f ()} is stationary. 10

(457)

1) Let S1 { < + : cf () = }, S2 be stationary, and S1 = sup( S2 ).


We can nd C = C : S1 , < + , such that : (a) C is a club of of order type . (b) nacc C S2 .

(c) for every club E of + , for stationarily many S1 , for some < + ,
= sup : nacc C and sup( C ) nacc C , otp ( C ) is even and {, sup( C )} E . 2) Let = < and S { < + : cf = } stationary. We can nd C = C : S, <

+ such that
(a) C is a club of of order type .

(b) for every club E of + for stationary many S, for some < + , for every < we have E contains arbitrarily large (below ) intervals of C of length
modified:1994-12-11

3) If is a successor cardinal then we can get (2) even if we omit = < and weaken in 1 , f () = f () to f () f (). 4) In part(2), if S2 = + we can omit = < if we restrict ourselves in (b) to a regular cardinal. 3.2A Remark 1) We can in 3.2(3) get the conclusion of 3.2(2) too if we x

revision:1994-12-11

2) We can replace in the assumptions and conclusions , + by is in 3.1A(2). Proof: 1) Let C : S1 be such that: C a club of , otp C = and nacc (C ) S2 .
If < < + , S2 (, ) has at least two elements then let (, , , ) : <

list all increasing pairs from (S2 \ ) (maybe with repetitions). Let f : < + exemplify 1 . Let C = {, : < } (increasing). Let e = e be a club of 11

(457)

such that: if i < j are from e then , < ,j . Now for S1 , < + , we let: ,i
C = {, , , , , : e }. , , Clearly C is a club of of order type . Now if E is a club of + , then E S2 is f () f ()

f (i)

a stationary subset of + so for some S1 , = sup(E S2 ) and dene g : by: , , , , +1, are the rst and second members of (E S2 ) \ (, , )}. By the choice of
f : < + for some < + , (stat )(g() = f ()). So C is as required . g() g()

2) Similar proof (and we shall not use it). 3) In the proof of (1) for < let h(, ) : . We do the construction for each < . The demand on e = e is changed to: if for i < j are from e, then ,i,
h(f (), ) onto

< ,j ,

and C is changed accordingly. For some < we succeed (really this version of 1 implies

the original version .)


4) By the proof above we can get C such that: for every regular < and club E of
modified:1994-12-11

+ for stationarily many S1 , for unboundedly many nacc C , we have: E,

cf () = . Then we correct as usual (see[Sh365 2]).

3.3 Claim: Assume: (a) regular, S stationary, = .

revision:1994-12-11

(b) C = C : S , C a club of . (c) P = P : S , P P(nacc (C )) is closed upward. (d) for every club E of for some , E nacc C P (e) < , T =
nacc C

{T,, : , {, } nacc C } , for < nacc C , T,, {T,, : nacc C ,

( ), |T,, | , and even for each the set

nacc C } has cardinality . 12

(457)

3.2 .

( ) and < are from A f

T,, . Then for some f nacc C ( ) we have [ < from A f T,, ] and for every < + , { A : f () = f ()} P . / Then there is no universal linear order of cardinality + . Proof: Similar to the previous one. 3.5 Conclusion: If 2 > + , = cf > 0 , C = C : S , S { < + : cf = } stationary, + id a (C) and for each we have |{C : nacc C }| then / (a) there is no universal linear order in + (b) moreover , univ(+ , Tord ) 2 ).

3.6 Discussion: (1) The condition from 3.1 holds in the models (of ZFC) constructed
modified:1994-12-11

in [Sh 100, 4] where = 0 , 20 = 2 and there is a non meager subset of 2 of cardinality 1 . (2) It is clear from 3.5 that the existence of a universal graph in does not imply the existence of a universal linear order in every for = + , = < : as by [Sh 175], [Sh 175a], if V |= GCH, = < , C = C : < + , cf = guesses clubs, for some + -c.c.

revision:1994-12-11

forcing notion P we have V P |=P there is a universal graph in + . But in V P the property of C, guessing clubs, is preserved and it shows that there is no universal linear order. (3) We can look at this from another point of view: (a) Considering the following three proofs of consistency results on the existence of universal structures: [Sh 100, 4] (universal linear order in 1 ), [Sh 175, 1] (universal 13

(457)

(f) If A P , for < + we have f

nacc C

graphs in + , = < and [Sh 175a] (universal graphs in other cardinals), the rst result cannot be gotten by the other two proofs. (b) For theories with the strict order property it is harder to have universal models than for simple theories (see [Sh93]) as the results of [Sh500, 1] on simple theories fail for the theory of linear order (by 3.5) and even all (f.o.) theories with the strict order property (as in [KjSh 409, x.x] ) (4) Concerning 3.5(b) , note that (for any complete rst order T ) we have Univ(, T ) 2< hence cf S (2< ), equality (i.e., = + ). univ(, T ) so under reasonable hypotheses we get in 3.5(b)

4 Toward the consistency for simple theories The aim of this proof was originally to deal with the universality spectrum of simple
modified:1994-12-11

countable theories and as a rst approximation to characterize { : univ(+ , T ) ++ < 2 }, but we shall do it more generally and have more consequences. On simple theories see [Sh 93]. The reader may well read the smooth version, i.e. add in Denition 4.1, the (< )-smoothness from 4.2(4), (5), and so we can omit clauses (e)(), (), () + (1) from Denition 4.1. He can also assume in 4.1 that i = 0 .

revision:1994-12-11

4.0 Notation: (1) For a set u S< (+ ) =: {u + : |u| < } let sup (u) = { + :
u} also let S = { < + : cf = }
+

(2) If u1 , u2 S< (+ ), h : u1 u2 is legal if it is one to one, onto, and there is a unique h+ such that: hV is one to one order preserving from sup (u1 ) onto sup (u2 ) and for u1 , h+ ( + ) = h() + . 14

(457)

(3) We say that h is lawful if in addition h+ is the identity . We sometimes use legal and lawful for functions h : u1 u2 when ui + , |ui | . (4) Wide + -trees T = (T , <) are here-just subsets of
+ >

(+ ) of cardinality + closed

under initial segments with the order being initial segment. A branch is a maximal linearly ordered subset, a + -branch is one of order type + . (So the trees are automatically normal).

4.1 Denition: Kap = (Kap , Kap ) is a -approximation family, if for some sequence (= i : i < + of vocabularies , |i | , i increasing with i, M i means (M i ) i; i can have relations and functions with innite arity but < (you may concentrate on the case i = for all i < ) the following hold : (a) Kap is a set of -model with a partial order =Kap (or is a sup(M ) -model).
modified:1994-12-11

(b) if M Kap then |M | is a subset of + of cardinality < and M Kap N M N . then M Kap and M Kap M ; also M 0 Kap (this is

(c) if M Kap , S

just to say we have the joint embedding property). (d) any Kap -increasing chain in Kap of length < has an upper bound.
(e) () if S , M0 = M2 , M0 Kap M1 , |M1 | then M1 , M2 has a common
+

revision:1994-12-11

Kap -upper bound M3 , such that M3 = M1


() if we have M1,i (i < i < ), M1,i Kap increasing with i , |M1,i | i S and
+

M2 i Kap M1,i , then there is a common upper bound M3 to {M2 } {M1,i : i < i }
() if we have M1 Kap , M2,i Kap for i < i < increasing with i , S ,

M2,i M1 then there is a common Kap -upper bound to {M1 } {M2,i : i < i } such that M3 = M1 . 15

(457)

() if (i) i : i i is a strictly increasing sequence of members of S ,

(ii) we have M1,i (i < i < ), M1,i Kap increasing with i , (iii) [i(1) < i(2) Mi(1) = Mi(2) i(1) ] (iv) |M1,i | i (v) M2,j Kap for j < j has universe i , and is <Kap -increasing in j (vi) M2,j i Kap M1,i , then there is a common upper bound M3 to {M2,j : j < j } {M1,i : i < i } such that for every i < i we have M3 i = M1,i (f) For < + , {M Kap : |M | } has cardinality . (g) We call h : M1 M2 a lawful (legal) Kap -isomorphism if h is an isomorphism from M1 onto M2 and h is lawful (legal). We demand: () if M1 Kap , u1 = |M1 |, u2 + and h a lawful mapping from u1 onto u2
modified:1994-12-11

then for some M Kap , |M | = u2 and h is a lawful Kap -isomorphism from M onto M . () lawful Kap -isomorphisms preserve Kap . (h) If M Kap and < + then for some M Kap we have M Kap M and |M | (i) [Amalgamation] Assume M Kap for < 3 and M0 Kap M for = 1, 2 . Then

revision:1994-12-11

for some M Kap and lawful function f we have: M1 Kap M , the domain of f is M2 , f |M0 | is the identity and f is a Kap -embedding of M2 into M , i.e. f d,d ????(M2 ) Kap M (j) If Mi AP for i < i < is <Kap -increasing,
i<i <2

Mi Kap N Kap then

there is N + , N 2 Kap N + Kap and a Kap -embedding f of N into N + over 16

(457)

i<i

Mi .

4.1A Remark: 1) This is similar to + -uniform forcing, see [Sh107], [ShHL 162] see also [Sh326, AP], [Sh405, AP]. 2) From (g)(), () we can deduce () if h is a lawful Kap - isomorphism from M1 Kap onto M2 Kap , and M1 Kap M1 and h can be extended to some lawful h+ with domain |M1 | then for some h , M2 we have M2 Kap M2 , h h and h a lawful Kap -isomorphism from M1 onto M2 . 3) We can use a linear order < of + is < [, + ) is a saturated model of Th( , <) and demand legal (and lawful) maps to preserve it. No real change.

4.1B Denition We call Kap homogeneous if in clause (g) of denition 4.1 we can replace lawful by legal.
modified:1994-12-11

4.2 Denition: 1) For Kap is a -approximation family, we let: Kmd = { : (i) is a Kap -directed subset of Kap (ii) is maximal in the sense that : for every < + for some M we have

revision:1994-12-11

|M | (iii) if M , M Kap M , then for some M , there is a lawful Kap -isomorphism h from M onto M over M }. 2) Kap is a simple -approximation if: (it is a -approximation family and ) for every Kmd and {(Mi , Ni ) : i < + } satisfying Mi , Mi Kap Ni Kap there is a club C of + and pressing down h : C + such that: 17

(457)

() if 1 < 2 are in C S , h(1 ) = h(2 ) and M1 Kap M , M2 Kap M then

we can nd N Kap , M Kap N , and a lawful Kap -embeddings f1 , f2 of N1 , N2 into N over M1 , M2 respectively such that f1 (N1 1 ) = f2 (N2 2 ). Of course by rening h we can demand on 1 , 2 also that () M1 1 = M2 2 , N1 1 = N2 2 , |M1 | 2 , (|N1 | 2 and some f is a lawful isomorphism from N1 onto N2 mapping M1 onto M2 .
3) We dene Kmd as before but M |M | .

4) Kap is -closed if = cf < and: if Mi : i < is Kap -increasing in Kap then


i<

Mi Kap is an Kap - upper bound; moreover (i < )[Mi Kap N ] implies

i< Mi Kap N . 5) Kap is (< )-closed if it is -closed for every < 6) Kap is smooth if
modified:1994-12-11

() it is (< )-closed; () all vocabularies i are nitary;


() in clauses (c),(e)(), and (e)() we can replace S to > 0 is divisible
+

by .

revision:1994-12-11

7) Kap is a -approximationx family if from Denition 4.1 it satises clauses (a), (b), (c), is smooth??, (g), (h), (i), and (i) if M2 M1 1 , then M1 , M2 have an upper bound (j) if Mi Kap is Kap -increasing then
i<

Mi Kap .
+

8) Kap is nice if whenever M0 Kap M1 , S , |M1 | , M0 = M2 and M M

Kap for

< 3, then we can nd M3 Kap M such that M Kap M , M Kap M3 for 18

(457)

< 3 and M3 = M1 . 9) Kap is weakly nice if whenever for


= 1, 2, M0 = M , S , |M1 | 2 , = 1

and M as above, we can nd M as above.

4.2A Observation: 1) If M, N Kmd , < + , then some lawful f is an isomorphism

from M onto N . 2)???

4.3 Lemma: Suppose that (A) = < ; (B) Kap is a -approximation family; (C) Kmd for < ; (D) T is a wide + -tree, A a + -branch of T for < and for = (< ) we
modified:1994-12-11

have A = A , and we let (, ) = the level of the <T -last member of A A , (, ) = ((, ) + 1). Then there is a forcing notion Q such that: (a) Q is -complete of cardinality | |< (b) Q satises the version of + -c.c. from [Sh 288 1 ] (for simplicity - here always for

revision:1994-12-11

= but by smoothness we actually have lub).


(c) For some Q-names h and (for < ) we have: Q for < we have Kmd , h is lawful , maps + onto + , and maps onto such that for < < , (, ) = (, ), so for every M we have h (|M |) is lawful and is an isomorphism from M onto some M .

19

(457)

4.3A Remark: 1) Our freedom is in permuting (, + ); up to such permutation (i) = {M : |M | i} is unique. 2) If we demand that Kap be smooth the proof is somewhat simplied. 3) We can replace assumption (B) by (B) Kap is a -approximationx family. Proof: We dene Q as follows:
p p Q i p = (M , hp ) : w p where

(a) wp [ ]< ;
p (b) M ; p (c) hp a lawful mapping, Dom hp = |M |; p p (d) if = are in w p , then : h (M (, )) and h (M (, )) are Kap

comparable;
modified:1994-12-11

(e) for every w p , for some n < , 0 = i0 < i1 < . . . < in = + , we have: for
[1, n), i S and for every
+

<n
+1 )

() for every w for which (, ) [i , i w such that: j (, ) [i , i


+1 )

and j [i , i

+1 )

S there is

p p and M (, ) Kap M (, )

revision:1994-12-11

p q The order is p q i: w p wq and for w p : M Kap M , hp hq and wp


p q M = M

p q h (M (, )) Kap h (M (, )).

The lemma will follow from the facts 4.4-4.7 below. 4.4 Fact: Any increasing chain in Q of length < has an upper bound. Proof: Let pi : i < be an increasing sequence in Q, < a limit ordinal. Let 20

(457)

w=

{w pi : i < }, and list w as {j : j < j }. We now choose by induction on j < j , a

member Mj of Kap and a lawful mapping hj with domain |Mj | such that :
p (a) if (Mi , hpii ) : i < but j w pi is eventually constant, then this value is i

(Mj , hj ).
p (b) Otherwise let hj (Mj ) j be a Kap -upper bound of {hpii (Mi ) : i < but i

j w pi } {hj1 (Mj1 ) (j , j1 ) : j1 < j}.

If we succeed q =df (Mj , hj ) : j w is a member of Q as required. Why? First we check that q Q . Clauses (a),(b),(c) are obvious; for clause (d) let = be in w , so let {, } = {j1 , j2 }, j1 < j2 ; now if ()(b) holds for j2 just note that hj1 (Mj1 ) (j1 , j2 ) hj2 (Mj2 ) by the choice of the later; and if ()(a) holds for j2 ,
p i then for some i < , (Mj2 , hj2 ) = (Mj2i , hp2 ) and now check the choice of (Mj1 , hj1 ). j

modified:1994-12-11

If for it too clause (b) holds for some i(1) < , (Mj1 , hj1 ) = (Mj1i(1) , hj2i

P (1)

) and use

P i pmax{i(1),i} Q. If for j1 clause (b) holds then by its choice hp2 (Mj2i ) (j1 , j2 ) hj2 (Mj1 ) j p i hence hj1 (Mj2 ) (j1 , j2 ) = hp1 (Mj1i ) (j1 , j2 ) hj1 (Mj2 ) (j1 , j2 ) as required. So we are j

left with the case clause (b) of () apply to j2 , which is even easier. For clause (e) , clearly it is enough to prove :

revision:1994-12-11

() for every i1 (S {+ }) there is i0 i1 (S {0}) such that () of clause (e)

of the denition of Q holds with i0 , i1 taking the role of i , i

+1

Let i1 S {+ } be given ; for each i < i1 let f (i) =df sup{(, )+1 : w, (, )

[i, i1 )} ( if the supremum is on an empty set - we are in a trivial case ). Clearly [j1 < j2 < 21

(457)

i1 f (j1 ) f (j2 )], so for some i0 i1 (S {0}) for all i [i0 , i1 ) (S {0}) we

have f (i) = f (i0 ) . Now for each i < i1 let g(i) =df sup{j + 1 : j < j , (j , ) [i, i1 ) and in case (b) occurs for j}, note: if the suremum is on the empty set then the value is zero; again it is clear that g decrease with i hence wlog for all i [i0 , i1 ) we have g(i) = g(i0 ) case 1 g(i0 ) > 0; this means that for every i [i0 , i1 ) there is w such that : (, ) [i0 , i1 ) and letting = j and in (*) above case (b) occurs. Check case 2 not case 1
p For every w let be the rst ordinal such that (M i , hpi ) : i < , i is

constant, and again wlog for some for every i [i0 , i1 ), < , < f (i0 ) and j < g(i0 ) there is w such that (, ) [i, i1 ), {j : j j < ga (i0 )} and j , the rest should be clear.
modified:1994-12-11

So we have proved that q Q ; now pi Kap q is straightforward. So now we have only to prove that we can carry the inductive denition from (*). In the choice of Mj , hj we rst have chosen hj (Mj ). We do it by choosing h(Mj ) for {(j , ) : w}; there we use clause (e)() of Denition 4.1. Having chosen hj (Mj ) we can nd Mj , hj by clauses (g)() + () of Denition 4.1.

4.4

revision:1994-12-11

4.5 Fact: 1) if p Q, w p and N then for some q: p q, w q = w p and


p p wp \{} (M , h ) q q = (M , hq ) and N M .

2) If p Q, < then for some q, p q Q and w q . Proof: 1) Easier than the previous one (or let = 1 , p0 = p and {j : j < j } list w p with 22

(457)

= 0 , repeat the proof of 4.4 , just use q to choose (M0 , h0 ). 2) Easier. Note the following

4.6 Fact: If Kap is -closed, then the following set is Q dense in Q: {p Q: if , w p ,


p p then hp (M ) (, ) = h (M ) (, )}.

Proof: Follows easily from the previous Facts.

4.7 Fact: The chain condition ()+ , from [Sh 288 1] holds. Proof: For simplicity assume Kap is 0 -closed so we can use 4.6. Suppose p() Q for
S . For some pressing down function h : S + and : < + we have:
+ +

() if h( 1 ) = h( 2 ), 1 < 2 then:
modified:1994-12-11

(a) otp (w p( ) ) = otp (w p( ) ) and w p(

w p(
1

= wh(1 )
2

(b1 ) OPwp(1 ) ,wp(2 ) is the identity on w p( (b2 ) for , w p(


1

w p(

the following are equivalent:

(i) (, ) < 1 ; (ii) ( , ) < 2 where =df OP wp(1 ) ,wp(2 ) (), =df OP wp(1 ) ,wp(2 ) ();

revision:1994-12-11

(iii) ( , ) = (, ) where , are as in (ii). (c) |M


p( 1 )

| is bounded in 2 and also sup{(, ) : = are in w p( ) } < 2

(d) if 2 = OP wp(1 ) ,wp(2 ) (1 ) then

OPu1 ,u2 is the unique order preserving function f such that (Dom f ) an initial segment

of u1 Rang (f ) an initial segment of u2 and Dom (f ) = u1 Rang (f ) = u2 . 23

(457)

4.5

() OP |M p(1 ) |,|M p(2 ) | is an isomorphism from M1 () M1


p( 1 )

p( )

1 1

onto M2

p( 2 )

which is lawful.

( 1 ) = M2

p( 2 )

( 2 ).

Now we have to prove : h( 1 ) = h( 2 ) p( 1 ), p( 2 ) are compatible. In the list {j : j < j } put w p(


1

w p(

an initial segment. Say {j : j < i }. First we restrict ourselves

further by assuming Kap is nice (see Denition 4.2(7)). We dene a common upper bound p; we let w p = w p( For w p(

w p( ) . For w p( ) \ w p(
)

p let (M , hp ) = (M p( )

p( )

, h

p( )

).

p(
p( )

p rst choose M such that M

p Kap M for

= 1, 2,

p and M 1 = M

2 [Why? -by Denition 4.1 clause (e)(), now we can nd such

p M Kap , now we can nd one in by Kap is nice (see Denition 4.2(7))]. + Second, we deal with the case Kap is not nice. Without loss of generality there is 0 S ,

M j
modified:1994-12-11

condition q j , increasing with j, w q = {i : i < j}, q j ,

bookmkeeping is as in the proof of 4.4, the successor case as in the proof above (for nice Kap ) but using amalgamation (=clause (i) of Denition 4.1 in the end). This nishes the proof of 4.3.

The simplicity of Kap is referred to only in 4.8 below, but it is needed to get the

revision:1994-12-11

universality results later.

4.8 Claim: Assume Kap is a simple + -approximation system. If 0 Kap is directed and < + = |[, + ) \ c.c. of [Sh 288 1], |Q| = + ,
M P Q

there is a and a lawful f such that f (0 ) Kmd . 24

(457)

p(

1 0 and Aj 0 = Ai 0 for j < i < i . We choose by induction on j i a


=1,2 i<j

M i

p( )

Kap Mj . The i

4.7

4.3

|M |, then for some forcing notion Q satisfying the + -

Proof: Natural. By renaming, without loss of generality A = {|M | : M 0 } = {2 :

def

< + }. Q = {M : M Kap and M A Kap and M A Kap M } order by Kap .


4.8

4.9 Conclusion: Assume = < < 2 = , and a + -tree T with branches is given For simplicity we assume that + is the set of members of T , 0 is the root and <T < for t T and let ut = {[, + ) : T t} . Then there is a forcing notion P such that: (a) P is -complete, satises the + -c.c. and has cardinality (so the cardinals in V P are the same and cardinal arithmetic should be clear). (b) for any -approximation system Kap there are , Mt : t T for < ++ such t that: () Kmd t
modified:1994-12-11

( g(t)+1)

() t <T s t s () for every Kmd for some < ++ and + -branch B = {t : < + } of T and lawful function from + onto + mapping onto
<+

t .

(c) Is R V P is (< )-complete, satises the version of the + -c.c.()+ , from [Sh2881] and Di (i < + ) is a dense subset of R and |R| + , then for some directed G R,

revision:1994-12-11

Di G = .

Proof: We use iterated forcing of length ++ , (< )-support , each iterand satisfying

if = < , 2 = + < = , and we add Cohen subsets to + (i.e. force

by {f : f a partial function from to {0, 1} of cardinality < + }, then in V P those assumptions hold. 25

(457)

the + -c.c.()+ , from [Sh 288 1], Pi , Qj : i +++ , j < ++ such that: for every Kap (from V or from some intermediate universe) for unboundedly many i < ++ , we use the forcings from 4.3 or 4.8.

4.9

5 Applications 5.1 Lemma: Suppose (A) T is rst order , complete, for simplicity with elimination of quantiers (or just inductive theory with the amalgamation and disjoint embedding property). (B) Kap is a simple -approximation system such that every M Kap is a model of T hence every M , where for Kmd we let M = {M : M }.

(C) every model M of T of cardinality + can be embedded into M for some Kmd with
modified:1994-12-11

M P

|M | = {2 : < + }.

Then: (a) in 4.9 in V P , there is a model of T of cardinality ++ universal for models of T of cardinality + .

revision:1994-12-11

(b) So in V P , univ(+ , T ) ++ but there is a club guessing sequence C : Sx

Proof: Straightforward.

Though for theories with the strict order property, the conclusion of 4 (and 5.1) fails, for some non simple theories we can succeed. Note that in 5.1 we have some freedom in 26

(457)

5.1

choosing Kap even after T is xed .

5.2 Lemma: Let T = Tfeq ; it satises the assumption of 5.1 (hence its conclusions).

In fact we can nd a smooth nice simple -approximation system Kap such that every
md model M of T of cardinality + is embeddable into some M Kap .

5.2A Remark 1) Note that there univ(, Tfeq ) = univ(, Tfeq ) . Actually the -approximation

family we get is also homogeneous. 2) The situation is similar for T3 in 5.3. Proof: By 5.2A(1) we deal with models of Tfeq . Condition (A) of 5.1 clearly holds. The main point is to dene Kap . () M Kap i:
modified:1994-12-11

(i) M is a model of T (ii) |M | [+ ]< () M1 Kap M2 i (i) M1 M2


/ (ii) if S , a P M1 , b QM1 \ and (c M1 )[M1 |= bEa c c ] then
+

revision:1994-12-11

(c M2 ) [M2 |= bEa c c ]. /

Also condition (C) of 5.1 is easy and we turn to condition (B). The checking of (K ap , Kap ) is a -approximation family (see Denition 4.10) as well as smoothness is staightforward. E.g. let us check the amalgamation (Denition 4.1 clause(i)). So assume M Kap for < 3, M0 Kap M1 , M0 Kap M2 ; by Denition 4.2 clause (g)() without loss of generality 27

(457)

|M1 | |M2 | = |M0 |. Now we shall dene a model M with universe |M1 | |M2 |, as follows: P M =df P M1 P M2 , QM =df QM1 QM2 , and for each x P M , we let Ex be the closure to an equivalence relation of the set of cases occurring in M1 and/or M2 , now check. The checking is straightforward.

Now we are left with the main point: the simplicity of Kap (see Denition 4.2(2)). Choose h as implicit in () of Denition 4.2(2); so let 1 < 2 , M1 , M2 , M, N1 , N2 , f be as there. Let f1 , f2 be lawful mappings such that f1 (N1 1 ) = f2 (N2 2 ) and N0 = M and for = 1, 2 the f (N \ M ) is disjoint to M for = 1, 2; let N = f (N ). Now we
def

dene N Kap ; it is a model with universe |N0 | |N1 | |N2 |, P N = pN0 P N1 P N2 , QN = QN0 QN1 QN2 . Lastly for x P N , we let Ex be the nest equivalence relation on QN which extend each Ex , (if x N ). Why is N a model of Tfeq ? Clearly P N , QN
modified:1994-12-11

is a partition of N (as this holds for N ( < 3) and as any two of those models agree
N on their intersection) and each Ex (x P N ) is an equivalence relation on QN (by its

choice). Why N is a submodel of N ? concerning P and Q there are no problems. So


N assume x P N , and we shall prove Ex |N | = Ex , the inclusion is by the choice of N Ex . For the other inclusion, if x belongs (and the proof of amalgamation) to only one

revision:1994-12-11

Nm it is totally trivial. If it belongs to exactly two of them, say Nm1 , Nm2 just note Nm1 Nm2 Nm1 , Nm2 . So assume x
2 =0

P N and here we shall use clause ()(ii)


()

of the denition of Kap . So suppose y0 , . . . , ym() are such that N but N


(m())

|= ym Ex ym+1 ,

|= y0 Ex ym() ; wlog m() is minimal. Of course without loss of generality

(m) = (m + 1) (as then we can omit ym+1 ) and (m) = (m + 2) (otherwise ym+1 N
(m)

(m+1)

= N

(m+1)

(m+2)

ym+2 and we can omit ym+1 . So necessarily 28

(457)

(m) = (m + 3) and { (m), (m + 1), (m + 2)} = {0, 1, 2}, hence enough to deal with the case m() = 3. As x
m<

(m)

=???

2 =0

N , clearly x N1 N2 1 , and for

some m, ym N1 N2 ( 1 ), and for some m1 {m 1, m + 1} and k {1, 2}, ym1 N0 Nk = M Nk = M Nk , so by the choice of Kap there is ym1 N0 N1 k QN , ym1 Ex 0 ym , but so ym1
<3

N (as N1 1 = N2 2 ), and we are done.

5.3 Lemma: Ttrf , the theory of triangle free graphs satises the assumption of 5.1 (hence its conclusions). Proof: Let xRy mean {x, y} is an edge. The main point is to dene Kap () M Kap i (i) M is a model of T (ii) |M | [+ ]<
modified:1994-12-11

() M1 Kap M2 i (i) M1 M2
(ii) if S , a, b M1 and there is no c M1 , M1 |= cRa & cRb then for no
+

c M2 , M2 |= cRa & cRb.

revision:1994-12-11

Let us check Denition 4.1, i.e. that (Kap , Kap ) is a -approximation system Clause (a), (b), (c) are immediate. Clause (d) holds in a strong form: the natural union is a lub; and even Kap is smooth. Clause (e) follows from (d)+ and (i) (amalgamation) Clauses (f ) and (g) are immediate (as in g) the demand is on lawful h only). Clause (h) is trivial. 29

(457)

5.2

Clause (i): Using a lawful f without loss of generality |M1 | |M2 | = |M0 |. Dene M3 : |M3 | = |M1 | |M2 |, RM3 = RM1 RM2 . Clearly M3 Kap as for M Kap M3 , by transitivity and symmetry it is enough to prove M1 Kap M3 , clearly M1 M3 , (i.e.
clause (i) of () above). For proving clause (ii) let S , a, b M1 and c M3 be
+

such that aRM3 c&bRM3 c. If c M2 \ M1 , necessarily a, b M2 hence a, b M0 and use M0 Kap M2 , but if c M1 there is nothing to prove. Clause y follows from smoothness.
Next let us show that Kap is simple. Let 1 < 2 (from S ), M1 , M2 , N1 , N2 , M, f
+

be as in Denition 4.2(2) (). Without loss of generality M N1 = M1 , M N2 = M2 . Dene a model N . |N | = |N1 | |N2 | |M |

modified:1994-12-11

R N = R N 1 R N 2 R N 3 Clearly N extends each of the models N1 , N2 , M (hence M1 , M2 too). Clearly for proving N Kap it suces to show ()1 if N there is no triangle.

revision:1994-12-11

Why? Clearly the only case we should consider is a N1 N2 \ M , b N1 M \ N2 , c N2 M \ N1 . (hence b M1 \ 1 , c M2 \ 2 ). So for some c M1 \ 1 , f (c ) = c but also f (a) = a hence aRN1 c1 , also aRM1 b, so as M1 Kap N1 clearly for some a M1 1 we have a RM1 c Ra RM1 b. Applying again f we get a RM2 c. So (by the last two sentences) in M we have a RM c&a RM b. But by the choice of abc (and as M N ) we get a , b, c is a triangle in M which belongs to Kap , contradiction. 30

(457)

So N Kap ; also M Kap N . [Why? being submodels sould be clear. So suppose , a, b, c contradicts clause (ii) of the denition of Kap , so c N \ M , so c N \ M for some {1, 2}.

If c N \ M2 \ N1 then necessarily (as aRN c, bRN c) we have a, b N hence a, b N M = M using M Kap N we have c M such that M |= aRc &bRc , c is as required). So necessarily c N N1 \ M hence a, b M1 M2 ; if a, b M1 do as above, also if a, b, M2 do as above, so by symmetry without loss of generality a M1 , b M2 . Now use f as in the proof of N is triangle free.) The case N N is similar. Having proved Kap is a simple -approximation family, it is easy to check the assumptions (A) and (C) of 5.1 hold.
modified:1994-12-11

5.3

5.4 Discussion: The similarity between the proofs of 5.2, 5.3 is not incidental. For a complete rst order T , let e.g. M B() be a + -saturated model of T , choose by induction on < , an elementary submodel B of (B(), , < ) of cardinality + such that {M , B : < } (+ + 1) B , B< B , f B+1 a mapping from B + onto { + : < + divisible by }, extending

revision:1994-12-11

<

f . In the

end let N be the model with universe + such that


<+

<+

f as an isomorphism for N

B M onto N . Let E be a thin enough club of + . Let Kap = {N : N


(N , )N nacc E }. (only nacc replaces S ).
+

and (N, )N nacc E

References 31

(457)

[CK] C. C. Chang, J. H. Keisler: Model Theory, North-Holland, 1973. [M] A. Mekler: Universal structures in power 1 , Journal of Symbolic Logic 55 (1990), 466477. [Sh-c] S. Shelah :Classication Theory and the number of non isomorphic models, North Holland Publ. Co.,Studies in Logic and the Foundation of Math vol. 92, 1990, 705 + xxxiv. [Sh-g] Cardinal Arithmetic, , Oxford University Press , to appear. [Sh 80] A weak generalization of MA to higher cardinals, Israel J.of Math., 30 (1978), 297 306. [Sh 92] Remarks on Boolean algebras , Algebra Universalis 11 (1980), 77-89. [Sh 93] Simple Unstable Theories, Annals of Math. Logic., 19 (1980) 177204. [Sh 100] Independence results, J. of Symb. Logic, 45 (1980) 563573.
modified:1994-12-11

[Sh 107] Models with second order properties IV, A general method and eliminating diamonds, Annals. of Math. Logic, 25 (1983) 183212. [KfSh 150] M. Kaufman and S. Shelah: The Hanf number of stationary logic, Notre Dame J. of Formal Logic, 27 (1986) 111 123. [ShLH 162] S. Shelah, C. Laamne and B. Hart: Models with second order properties V: A General principle, Annals of Pure and Applied Logic, to appear. [GrSh 174] R. Grossberg and S. Shelah: On universal locally nite groups, Israel J. of Math., 44 (1983), 289302. [Sh 175] On Universal graphs without instances of CH. Annals of Pure and Applied Logic ,26 (1984) 7587. 32

(457)

revision:1994-12-11

[Sh 175a] Universal graphs without instances of CH, Israel J. of Math., 70 (1990), 6981. [Sh 288] Was Sierpenski right II, Proceedings of the 1/91 conference in Budapest in honour of A. Hajnal and V. T. Ss, Colloquia Mathemtica Societatis Janos Bolyai 60. Sets o Graphs and Numbers , Budapest (Hungary ) ,1991, 637-668. [Sh 326] Vive la Dierence I, Proceedings of the Conference in Set Theory, eds. H. Judah, W. Just, and H. Woodin, Springer Verlag MSRI publications 26 (1992) , 357-405 [Sh 365] There are Jonsson algebras in many inaccessible cardinals, Ch. IV of [Sh g]. [Sh 405] Vive la dierence II - refuting Kims conjecture , Israel J. of Mathematics , accepted [KjSh 409] M. Kojman and S. Shelah: Nonexistence of universal orders in many cardinals, Journal of Symbolic Logic 57 (1992), 875891. [Sh 413] More Jonsson algebras and colorings, Archive f. Math. Logic. , accepted [KjSh 447] M. Kojman and S. Shelah: The universality spectrum of stable unsuperstable theories,
modified:1994-12-11

accepted to Annals of Pure and Applied Logic. 58 (1992) , 57-92 [KjSh 455] M.Kojman and S. Shelah: Universal Abelian Groups. Israel J. of Mathematics , submitted [KjSh 456] in preparation [Sh 460] The generalized continuum hypothesis revisited, preprint

revision:1994-12-11

[Sh500] Classifying Unstable Theories, preprint.

33

(457)

Potrebbero piacerti anche