Professional Documents
Culture Documents
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS
SAHARON SHELAH
Abstract. We prove that some natural outside property is equivalent (for
a rst order class) to being stable.
For a model, being resplendent is a strengthening of being -saturated.
Restricting ourselves to the case > |T| for transparency, a model M is -
resplendent means:
when we expand M by < individual constants c
i
: i < , if
(M, c
i
)< has an elementary extension expandable to be a model of
T
, |T
.
Trivially any saturated model of cardinality is -resplendent. We ask: how
many -resplendent models of a (rst order complete) theory T of cardinality
are there? Naturally we restrict ourselves to cardinals =
+ 2
|T|
. Then
we get a complete and satisfying answer: this depend just on T being stable or
unstable. In this case proving that for stable T we get few, is not hard; in fact
every resplendent model of T is saturated hence determined by its cardinality
up to isomorphism. The inverse is more problematic because naturally we
have to use Skolem functions with any < places. Normally we use relevant
partition theorems (Ramsey theorem or Erd os-Rado theorem), but in our case
the relevant partitions theorems fails so we have to be careful.
0. Introduction
Our main conclusion speaks on stability of rst order theories, but the major (and
the interesting) part of the proof has little to do with it and can be read without
knowledge of classication theory (only the short proof of 1.8 uses it), except the
meaning of < (T) which we can take as the property we use, see inside 2.1(1)
here (or see [Sh:E59, 1.5(2)] or [Sh:c]). The point is to construct a model in which
for some innite sequences of elements we have appropriate automorphism, so we
need to use Skolem functions with innitely many places. Now having functions
with innite arity make obtaining models generated by indiscernibles harder. More
specically, the theory of the Skolemizing functions witnessing resplendence for
(M,
+ 2
|T|
(B): an external denition of stability which happens to be the dividing
line.
Baldwin had told me he was writing a paper on resplendent models: for
0
stable
one there are few ( 2
0
) such models in any cardinality; and for T not superstable
there are 2
2
|T|
, T has > non-isomorphic resplendent
models,
(iv): for every =
2
|T|
, T has 2
T1
below, there is an expansion (M
1
, c
i
)
i<
of (M, c
i
)
i<
to a
model of T
1
, when:
()
T1
: Case 0: = 0: [T
1
[ < ,
Case 1: = 1: for some
(N
1
), [
[ < and T
1
L(
c
i
: i < ),
Case 2: = 2: T
1
is recursive (see 1.4(4) below),
Case 3: = 3: T
1
= Th (N
1
, c
i
)
i<
(but remember that N
1
has
only < relations and functions not of M).
(2): -resplendent means (, 3)-resplendent.
(3): Assume M is a model of T, c
>
[M[ and M
c
is an expansion of (M, c).
We say that M
c
witnesses (, )resplendence for c in M, when:
for every rst order T
1
such that
Th(M, c) T
1
& [(T
1
) (T)[ <
and ()
T1
holds, we have:
M
c
is a model of T
1
up to renaming the symbols in (T
1
) (M, c).
(4): For M, N
1
, c
i
: i < ) and T
1
Th(N
1
, c
i
)
i<
as in part (1), T
1
is
recursive when:
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
4 SAHARON SHELAH
(a): =
0
and T
1
is recursive (assuming the vocabulary of T is repre-
sented in a recursive way or
(b): >
0
and for some
(N
1
), [
(x
0
, . . . , x
n1
) L(
(N
1
) such that is the
identity on
and (
1
) =
2
and
0
<
1
< . . . < then
1
(c
0
, c
1
, . . .) T
1
i
2
(c
0
, c
1
, . . .) T
1
.
(5): We say f is an (M, N)-elementary mapping when f is a partial one-to-
one function from M to N, (M) = (N) and for every (x
0
, . . . , x
n1
)
L((M)) and a
0
, . . . , a
n1
M we have:
M [= [a
0
, . . . , a
n1
] i N [= [f(a
0
), . . . , f(a
n1
)].
(6): f is an Melementary mapping if it is an (M, M)elementary mapping.
(7): M is homogeneous if :
for any Melementary mapping f with [ Dom(f)[ < and a M
there is an Melementary mapping g such that:
f g, Dom(g) = a Dom(f).
(8): M is strongly homogeneous if for any M-elementary mapping f with
[ Dom(f)[ < there is an automorphism g of M, such that f g.
(9): Let
1
2
be vocabularies. We say that is an automorphism of
2
over
1
when: is a permutation of
2
, maps any predicate P
2
to a
predicate of
2
with the same arity, maps any function symbol of F
2
to a function symbol of
2
of the same arity and
1
is the identity.
(10): For ,
2
as in part (9) let be the permutation of the set of formulas
in the vocabulary
2
which induce.
Example 1.5. There is, for each regular , a theory T
such that:
(a): T
is superstable of cardinality ,
(b): for , T
has 2
& [
[ < M
is saturated ]
then M is (, 1)resplendent.
(2) If M is saturated of cardinality then M is -resplendent.
Proof: Easy, e.g., see [Sh:a] and not used here elsewhere.
Proof of 1.5: Let A
0
= (i + 1) : i < and A
1
= A
0
. For every linear
order I of cardinality we dene a model M
I
:
its universe is
I s, t, i, x) : s I, t I, i < , x A
1
and [I [= s < t x A
0
] ,
(and of course, without lost of generality, no quadruple s, t, i, x) as above belongs
to I), its relations are:
P = I,
R =
_
s, t, s, t, i, x)
_
: s I, t I, s, t, i, x) [M
I
[ P
_
,
Q
= s, t, i, x) : s, t, i, x) [M
I
[ P, x for < .
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 5
In order to have the elimination of quantiers we also have two unary functions F
1
,
F
2
dened by:
s I F
1
(s) = F
2
(s) = s,
s, t, i, x) [M
I
[ I F
1
(s, t, i, x)) = s&F
2
(s, t, i, x)) = t.
It is easy to see that:
(a): In M
I
, the formula
P(x) & P(y) & (z)(R(x, y, z) &
<
Q
(z))
linearly orders P
MI
, in fact denes <
I
;
(b): Th(M
I
) has elimination of quantiers;
(c): if (M
I
), [[ < then M
I
is saturated;
(d): Th(M
I
) does not depend on I (as long as it is innite) and we call it T
;
(e): T
is superstable.
Hence: T
= Th(M
I
) is superstable, does not depend on I, and
M
I
= M
J
if and only if I
= J,
and by 1.6 M
I
is (, 1)resplendent.
1.6
Fact 1.7. (1) M is (, 3)resplendent implies M is (, )resplendent implies
M is (, 0)resplendent.
(2) M is (, 0)resplendent implies M is -compact.
(3) M is (, 2)resplendent implies M is -homogeneous, even strongly -homogeneous
(see Denition 1.4(7),(8)).
(4) If M is (, 2)resplendent >
0
and a
n
: n < is an indiscernible set
in [M[, then it can be extended to an indiscernible set of cardinality |M|
(similarly for sequences).
(5) M is (, 3)resplendent implies M is -saturated.
(6) If > [T[ then the notions of 1.4 (, )resplendent for = 0, 1, 2, 3, are
equivalent.
Proof: Straightforward, for example
(3) For given a
i
, b
i
M (for i < , where < ) let
T
1
= g(a
i
) = b
i
: i <
(x, y)(g(x) = g(y) x = y), (x)(y)(g(y) = x)
(x
0
, . . . , x
n1
)[R(x
0
, . . . , x
n1
) R(g(x
0
), . . . , g(x
n1
))] :
R an n-place predicate of (M)
(x
0
, . . . , x
n1
)[F(g(x
0
), . . .) = g(F(x
0
, . . .))] :
F an n-place function symbol of (M).
(4) For notational simplicity let a
n
= a
n
. Let T
1
be, with P a unary predicate, g a
unary function symbol,
g is a one-to-one function into P P(a
n
) : n <
_
(x
0
, . . . , x
n1
)
_
_
<n
P(x
) &
_
<m<n
x
,= x
m
& [a
0
, . . . , a
n1
]
(x
0
, . . . , x
n1
)
_
:
(x
0
, . . . , x
n1
) L((M))
_
1.7
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
6 SAHARON SHELAH
Conclusion 1.8. If M is resplendent, (T) +
1
then M is saturated.
Proof: By 1.7(5) M is saturated, so without loss of generality |M| > . Hence,
by [Sh:a] or [Sh:c, III,3.10(1),p.107], it is enough to prove: for
I an innite indis-
cernible M, dim(
+2
|T|
. Then T has > pairwise non-isomorphic resplendent
models of cardinality .
Before embarking on the proof, we give some explanations.
Remark 1.10. (1) We conjecture that we can weaken in 1.9 the hypothesis
=
+ 2
|T|
to =
<
+2
|T|
. This holds for many s, see [Sh:309,
2]; but we have not looked at this. See 3.
(2) We naturally try to imitate [Sh:a], [Sh:c, VII,2, VIII,2] or [Sh:E59, 3],[Sh:331].
In the proof of the theorem, the diculty is that while expanding to take
care of resplendency, we naturally will use Skolem functions with innite
arity, and so we cannot use compactness so easily.
If the indiscernibility is not clear, the reader may look again at [Sh:a] or
[Sh:c, VII,2], (tree indiscernibility). We get below rst a weaker version
of indiscernibility, as it is simpler to get it, and is totally harmless if we
would like just to get > non-isomorphic models by the old version [Sh:300,
III,4.2(2)] or the new [Sh:309, 2]
Explanation 1.11. Note that the problem is having to deal with sequences of <
elements
b = b
i
: i < ), innite. The need to deal with such
b with all theories
of small vocabulary is not serious there is a universal one though possibly of
larger cardinality, i.e., if M [= T, b
i
M for i < , < , we can nd a f.o. theory
T
2
= T
2
(
b) satisfying Th(M, b
i
)
i<
T
1
, [T
1
[ (2
|T|+||
)
<
such that:
if Th(M, b
i
)
i<
T
and [(T
) (T) b
i
: i < [ <
then renaming the predicates and function symbols outside T, we
get T
T
2
(
b)
this is possible by Robinson consistency lemma. Let us give more details.
Claim 1.12. (1) Let M
0
be a model,
0
= (M
0
),
b = b
i
: i < ) where
b
i
M
0
for i < and
0
be a cardinal. Let
1
=
0
b
i
: i <
so M
1
= (M
0
, b
i
)
i<
is a
1
model. Then there is a theory T
2
= T
2
[
b] =
T
2
[
, the model M
2
is expandable to a model of T
,
when:
(): M
2
is a
1
model,
(): M
2
can be expanded to a model of T
2
,
(): Th(M
2
) T
,
(): T
) (M
2
)[ ,
(a)
+
: if > [T[ +[[ then [
2
[ 2
<
is enough.
(2) If in part (1), clause () of (b) is weakened to
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 7
()
2
: T
) (M
2
)[ < ,
then we can strengthen (a) to
(a)
2
:
2
= (T
2
) extends
1
and has cardinality
<
2
|1|++0+||
,
(a)
+
2
: if > ([T[ +[[)
+
then [
2
[
<
2
<
is enough.
Proof: 1) We ignore function symbols and individual constants as we can replace
them by predicates. Let
T = T
: T
and
(T
) (M
1
) has cardinality .
This is a class; we say that T
, T
),
(c): Rang(h) = (T
),
(d): h preserves arity (i.e., the number of places, and of course being predi-
cate/function symbols),
(e): h ((M
1
)) = identity,
(f ): for a f.o. sentence = (R
1
, . . . , R
k
) L[(T
)], where R
1
, . . . , R
k
are
the non-logical symbols occurring in , we have
(R
1
, . . . , R
k
) T
(h(R
1
), . . . , h(R
k
)) T
.
Now note that
1
T/
= has cardinality 2
|0|+||+
.
Now let T
: < 2
|0|+||+
list members of T such that every equivalence
class of being isomorphic over Th(M
1
) is represented. (T
)
1
: < 2
|0|+||+
)
are pairwise disjoint.
Note that Th(M
1
) T
. Let T
2
=
: < 2
|0|+||+
and
2
T
2
is consistent.
Why? By Robinson consistency theorem.
Let T
2
be any completion of T
2
. So condition (a) holds; proving (b) should be
easy.
Let us prove (a)
+
; this is really the proof that a theory T, [T[ < , has a model
in 2
<
universal for models of T of cardinality . We shall dene by induction
on < , a theory T
2
such that:
(A): T
2
0
= Th(M
1
),
(B): T
2
a f.o. theory,
(C):
2
= (T
2
) has cardinality 2
|0|+||+||+0
,
(D): T
2
,
2
, [
[ [
1
[ +[[,
, T
2
L[
], T
complete and [
[ = R,
then we can nd R
2
+1
2
[replacing R by R
] T
2
+1
There is no problem to carry out the induction, and
<
T
2
is as required.
2) Similar.
1.12
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
8 SAHARON SHELAH
Explanation 1.13. So for M [= T,
b
>
M, we can choose T
1
[
b] Th(M,
b)
depending on Th(M,
b) is expandable to
a model of T
2
[
b].
W.l.o.g. (T
2
[
b]) depends on g(
b) and
0
only, so it is (g(
b),
M
).
The things look quite nitary but T
2
[
b). I.e.,
() , (), where
():
b
>
M, for , ( a limit ordinal) g(
i1
, . . . , b
in
] [b
i1
, . . . , b
in
],
(): for any L(
2
) for some < :
_
T
2
[
] T
2
[
.
[You can make T
1
[
b) is expandable
to a model of T
2
[
b], so if
2
= (T
2
[
b]) (M,
2
). We write here
b as a sequence of
parameters but from another prospective the predicate/function symbol R
()
has + arity(R)places.
Explaning the rst construction 1.14. (i.e., 2.19 below)
Eventually we build a generalization of EM(
(
) witnessing < (T), but the functions have any < places but not ,
and the indiscernibility demand is weak. We start as in [Sh:E59, 2], so for some
formulas
= a
) for
):
&
+1
(a
, a
)
if()
.
Without loss of generality, for any < for some sequence
G
= G
,
: <
g( y
=
G
(a
) := G
,
(a
) :
< g( y
:
as generators. For W [
]
<
, let
N
W
= N[W] be the submodel which a
b] for each
b
>
(N
W
),
(): monotonicity: W
1
W
2
N
W1
N
W2
.
So if U
, then N[U ] = N
W
: W
>
[U ] is a resplendent model of
cardinality .
(): Indiscernibility: (We use here very minimal requirement (see below)
but still enough for the omitting type in (1) below):
(1):
U N[U ] omits p
=:
(x, a
) : < ; (satis-
faction dened in N[
]),
(2):
U N[U ] realizes p
.
Now (2) was already guaranteed: a
realizes p
.
For (1) it is enough
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 9
(1)
: if W
>
[
],
W then p
is omitted by N
W
(satisfaction dened
in N[
]).
Fix W, for (1)
. A sucient condition is
(1)
i
: i < ) is indiscernible over N
W
in
N[
.
[if
r
(T) < , immediately suces; in the general case, and avoiding classication
theory, use
p
(x, a
(+1)
) & (x, a
()+1
) : <
so we use
(x, y
) =
(x, y
g( y
))
(x, y
(g( y
), 2 g( y
)))
in the end].
Note: as [W[ < , for some () < , for every
W : ()
is a singleton
and W W
()
(see below), this will be enough to omit the type. The actual
indiscernibility is somewhat stronger.
Further Explanation: On the one hand, we would like to deal with arbitrary se-
quences of length < , on the other hand, we would like to retain enough freedom
to have the weak indiscernibility. What do we do? We dene our (not as nice
as in [Sh:E59, 2], i.e., [Sh:a, Ch.VII 3]) by approximations indexed by .
For , we essentially have N
W
for
W W
=: W : W
, [W[ < and
the function ( W) is one-to-one .
Now, W
and W
<
W
is equal to [
]
<
.
So if we succeed to carry out the induction for < , arriving to = the
direct limit works and no new sequence of length < arises.
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
10 SAHARON SHELAH
2. Proof of the Main Lemma
In this section we get many models using a weak version of indiscernibility.
Context 2.1. (1) T is a x complete rst order theory, < (T), =
:
) with
a
M we have: if < ,
, < then M [=
[a
, a
]
if(=())
.
(2) Let be innite large enough cardinal; =
([T[) is O.K.
Remark: Why are we allowed in 2.1(1) to use
= I
= I
,
be the model
( (
) : = , E
i
<
i
)
i<
,
where
E
i
= (, ) :
,
, i = i ,
<
i
= (, ) : E
i
and (i) < (i) .
(2) Let W
= W
= W
: W has cardinality < and for any ,=
from W we have ,= , and W
<
=
<
W
.
(3) We say that W is invariant, or (, )invariant, when W W
and:
if W
1
, W
2
W
is hereditary if it W
W W W
W
Denition 2.3. (1) Let =
T,
be the minimal cardinal satisfying:
(a): =
<
[T[,
(b): if M is a model of T,
b
>
M, then there is a complete (rst order)
theory T
b) such that:
if T
Th(M,
b) and (T
)
(M,
b)
has cardinality <
then there is a one-to-one mapping from (T
) into (T
) over
(M,
b)
preserving arity and being a predicate / function symbol, and mapping
T
into T
.
(2) For < , let [T, ] be a vocabulary consisting of
T
, the individual con-
stants b
in the vocabulary
T
b
: <
extending T, let T
[T
b) is a model of T
, then T
[T
] is as in clause (b)
of part (1).
(3) For M [= T and < and
b
M, let T
b, M] = T
[Th(M,
b)].
Remark 2.4. Note that is well dened by 1.12. In fact, = 2
|T|+
:
+
<
is OK.
Main Denition 2.5. We say that m is an approximation (or an approximation,
or (, )approximation) if
()
1
: (so =
m
= (m)),
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 11
()
2
: m consists of the following (so we may give them subscript or superscript
m):
(a): a model M = M
m
;
(b): a set F = F
m
of symbols of functions, each f F has an interpre-
tation, a function f
m
with range M, but when no confusion arises
we may write f instead of f
m
, (or f
m
, note that the role of those f-s
is close to that of function symbols in vocabularies, but not equal to);
(c): each f F has
f
< places, to each place (i.e., an ordinal
<
f
) a unique
=
f
,=
],
and the -th variable of f varies on I
, i.e., f
m
(. . . , x
, . . .)
<
f
is
well dened i
_
<
f
x
= I
,
;
we may write f
m
(. . . ,
, . . .)
w[f]
instead f
m
(. . . ,
(f,)
, . . .)
<
f
, where
w[f] = (f, ) : <
f
; and f F (W W)[w[f] = :
W], see clause (e) below;
(d): for each
b
>
[M[, an expansion M
b
of (M,
b) to a model of
T
, . . .)
<
f
: f F, f(. . . ,
, . . .)
<
f
well dened,
and
W for every ,
(g): a function f = f
m
,
such that m satises the following:
(A): M is a model of T,
(B): [witness for < (T):] for our xed sequence of rst order formulas
,
F for ,
(we also call them f
m
,
) such that:
(i): f
,
is a one place function , with
f
,
,
f
,
0
from clause (c) being
1, respectively.
(ii): f
1,
(
1
) = f
2,
(
2
) if
1
=
2
and they are well dened, i.e.
,
(iii): if
_
f
1,
(
1
), f
2,+1
(
2
)
i [
1
(+1) =
2
(+1)],
(C): N
W
M, for W W,
(D): [f = f
m
witness an amount of resplendency]
(): the domain of f is a subset of
F
m
=:
f :
f = f
: <
f
),
f
< , f
F, and
f
does not depend on , call it
f
and
for <
f
the sequence (f
, ) does not
depend on , call it (
f, ),
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
12 SAHARON SHELAH
(): for
f Dom(f ), f (
f) is a function with domain
( x) : ( x) is a [T,
f
]term, and x = x
: u)
for some nite subset u = u
of
f
and if ( x) Dom(f (
f)) then
f (
f)(( x)) F[
f] := f F :
f
=
f
& ( <
f
)((f, ) = (
f, )),
(): if
f Dom(f ) and
b = f
(. . . ,
(
f,)
, . . .)
<
f
thenthe uni-
verse of M
b
is f((. . . ,
(
f,)
, . . .)
<
f
: f F[
f]
if (f (
f))(( x)) = f
F, W W,
W and
= (
f, )
for <
f
, and x = x
: u), and
b =
f( ) = f
( ) : <
f
), then
M
b
(f
(. . . ,
, . . .)
<
f
: u)) = f
(. . . ,
, . . .)
<
f
.
[explaining (): we may consider
b N
W1
N
W2
, and we better have that
the witnesses for resplendency demands, specialized to
b, in N
W1
and in N
W2
are
compatible so that in the end resplendency holds].
However, we shall not get far without at least more closure and coherence of the
parts of m.
Denition 2.6. (1) An approximation m is called full if W
m
= W
(m)
, and
is called semifull if W
<(m)
W
m
W
(m)
and is called almost full if it
is semi full when is limit ordinal and full when is a non-limit ordinal.
(2) An approximation m is resplendent if
m
and
if W W
W
m
and
f F
m
, and
(
f, ) : <
f
: W,
then
f Dom(f
m
).
(3) In part (2), if we omit , we mean =
m
, and <
.
(4) An approximation m is called term closed if:
(E): Closure under terms of :
Assume that u
, [u[ < , and for some W W
m
, u :
W, and
: <
F
m
,
< n, satises (f
, ) : <
f
u, is an n-place (T)-term
so = (x
0
, . . . , x
n1
). Then for some f F
m
satisfying
f
=
,
(f, ) =
: u) such that
for
u, and
: u W
W for some W
we have
f
m
(. . . ,
, . . .)
w[f]
=
_
. . . , f
m
(. . . ,
, . . .)
w[f
]
, . . .
_
<n
(this clause may be empty, but it helps to understand clause (F); note
that it is not covered by 2.5(D) as the functions are not necessarily with
the same domain, hence this says something even for the identity:
so this implies that in clause (f) of Denition 2.5 we can demand
m
(f, ) : <
f
= W).
(F): Closure under terms of (M
b
):
Assume that u
, [u[ < , and
: <
) lists u with no
repetitions, and for some W W
m
, u : W. If n <
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 13
and f
F
m
for < n,
f = f
, ) u for <
f
, and
(f
, ) u for <
f
, and
b
= f
(. . . ,
(f,)
, . . .)
b = b
for u and
: u W
m
, then
f
m
(. . . ,
, . . .)
u
=
M
b
_
. . . , f
m
(. . . ,
(f
,)
, . . .)
<(f
)
, . . .
_
<n
.
Observation 2.7. In Denition 2.6(4) in clauses (E),(F) it suce to restrict
ourselves to the case n = 1 and is the identity.
Proof: By 2.5(D)().
Of course some form of indiscernibility will be needed.
Denition 2.8. (1) Let E be the family of equivalence relations E on
>
(
) : without repetitions ,
or a subset of it, such that
1
E
2
g(
1
) = g(
2
).
(2) Let E
2
i for some < we have
(i)
1
,
2
) ,
(ii)
1
=
2
for < ,
(iii)
1
,=
2
is nite.
(3A) We say that (
1
,
2
) are immediate neighbours if g(
1
) = g(
2
), and for
some < g(
1
) we have ( < )( ,=
1
=
2
); so the dierence
with (3) is that nite is replaced by a singleton.
(4) Let E
0
<
be dened like E
0
let
seq
(W) =
_
:
>
(
) is with no repetitions,
and for some W W we have
as we dene E
0
,
(): g(f(
1
h()
: <
f
))) = f(
2
h()
: <
f
)) for f, h as above;
(c): Assume
1
,
2
seq
(m)
(W
m
),
f
1
,
f
2
Dom(f
m
),
=
f
1 =
f
2, and for some one-to-one function h from
to g(
) we have
(
f
, ) =
m
h()
for , m = 1, 2, and
1
E
2
. Let
= f
h()
: <
)) : < g(
f
)).
Then there is g such that
(): g is an (M
m
b1
, M
m
b2
)elementary mapping,
(): g = g
m
1
,
2
from clause (b) above.
(2) An approximation m is strongly indiscernible if it is E
1
(m)
indiscernible.
(3) (a) An approximation m is weakly indiscernible when it is E
0
(m)
-indiscernibility.
(b) An approximation m is weakly/strongly nice if it is term closed and
weakly/strongly indiscernible.
(c) An approximation m weakly/strongly good if it is weakly/strongly nice
and is almost full.
(d) An approximation m is weakly/strongly excellent if it is weakly/strongly
good, and is resplendent, see Denition 2.6(2),(3).
Discussion 2.10. Why do we have the weak and strong version?
In the proof of the main subclaim 2.19 below the proof for the weak version is
easier but we get from it a weaker conclusion:
+
non-isomorphic -resplendent
of cardinality =
. But see 3.
Claim 2.11. Let m be an approximation.
(1) In the denition of m is E
0
, . . .)
<
f
1
, and
: <
m1
f
W
(so a N
m1
W
), then (g
m2
m1
[W, h])(a) = f
m2
2
(. . . ,
, . . .)
<
f
2
,
(g): if
f
1
= f
1
: < ) Dom(f
m1
) and
m1
(
f
1
, )
(m2)
for <
m
f
1
,
f
2
= f
2
: < )
(F
m2
), and
m2
f
2
=
m1
f
1
, and
< & <
f
1 (f
2
, ) =
, & h(f
2
) = f
1
, then
():
f
2
Dom(f
m2
),
(): h
_
(f
m2
(
f
2
))((x
: u)))
_
=
_
f
m1
(
f
1
)
_
((x
: u))),
when u is a nite subset of
(): assume
for <
m1
f
1
, and W =
: <
m1
f
1
,
= f
(. . . ,
, . . .)
b
1
[N
m1
W
[ onto
M
m2
b
2
[N
m2
W
[.
(2) We say that m
, h
for ,
(c): h
is the identity,
(d): if
0
<
1
<
2
< then h
2
0
= h
1
0
h
2
1
.
(3) We say that an inverse system of approximations m
, h
: < , )
is continuous at if:
(a): < is a limit ordinal,
(b): W
m
W
m
: < ,
(c): F
m
Dom(h
) : < ,
(d): Dom(f
m
) =
f
2
: for some < and
f
1
Dom(f
m1
) of length
g(
f
2
) we have h
(f
2
) = f
1
for < g(
f
2
) .
Discussion: Having chosen above our order, when can we get the appropriate indis-
cernibility? As we are using nitary partition theorem (with nitely many colours),
we cannot make the type of candidates for
b xed. However we may have a priory
enough indiscernibility to x the type of enough
b
, i.e.,
, with skeleton a
:
) such that for
and
+1
we have
EM(
, ) [=
(a
, a
) i .
Without loss of generality, for some unary function symbols F
(), we have
EM(
, ) [= F
(a
) = a
for
. Now, by induction on < we choose
such that
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
16 SAHARON SHELAH
(a):
) (
) and EM
1
(
) EM
1
(
),
(e): the sequence
) =
<
(
),
(f ): if =
i
(x) : i < i
), and M
+1
= EM
1
(
,
+1
), and for
we dene
b =
b
,
as
M
+1
i
(a
) : i < i
)
i
(EM
1
(,
+1
)),
then we can interpret a model M
+1
b
of T
b, M
+1
T
] in M
+1
, which
means
(): if R
T
[
b, M
+1
(T)]
T
is a kplace predicate, then there is a
(k + 1)place predicate R
(
+1
) (
) such that
M
+1
b
[= R[c
0
, . . . c
k1
] i M
+1
[= R
[c
0
, . . . , c
k1
, a
],
(): if F
T[
b,M
+1
(T)]
T
is a kplace function symbol, then there
is a (k + 1)place function symbol F
(
+1
) (
) such that
M
+1
b
[= F[c
0
, . . . c
k1
] = c i M
+1
[= F
[c
0
, . . . , c
k1
, a
] = c.
Let us carry out the induction; note that there is a redundancy in our contraction:
each relevant
b is taken care of in the -th stage for every < large enough,
independently, for the dierent -s.
For = 0:
Let
0
= .
For a limit :
Let
: < ).
For = + 1:
Let the family of sequences of the form =
i
(x) : i < i
), where
i
(x) is a unary
term in (
), i
< , be listed as
(x) =
i
(x) : i < i
).
Let M
be a
+
resplendent (hence strongly
+
-homogeneous and -resplendent)
elementary extension of EM
1
(
), and let M
= M
T
, and choose
.
For each < let
b
=:
i
(a
) : i < i
). Now, (M
) can be expanded to a
model M
of T
, M
], and let
(T
b, M
])
T
= R
,
j,n
: j < , n < F
,
j,n
: j < , n < ,
where R
,
j,n
is an nplace predicate and F
,
j,n
is an nplace function symbol. Next
we shall dene an expansion M
+
of M
. Its vocabulary is
(
) R
,,j,n
, F
,,j,n
: j < , n < ,
where R
,,j,n
is an (n + 1)place predicate, F
,,j,n
is an (n + 1)place function
symbol, and no one of them is in (
be an automorphismof M
mapping EM
1
(
)
onto EM
1
(,
((a
)) =
(a
) (hence < g
(a
) = a
using (x) = F
(x)).
Now we actually dene M
+
expanding M
+
:
R
M
+
,,j,n
=
_
(g
(c
0
), g
(c
1
), . . . , g
(c
n1
), g
(a
)) :
M
[= R
,
j,n
(c
0
, . . . , c
n1
)
_
,
F
M
+
,,j,n
is an (n + 1)place function such that
M
[= F
,
j,n
(c
0
, . . . , c
n1
) = c implies
F
M
+
,,j,n
(g
(c
0
), . . . , g
(c
n1
), a
) = g
(c).
We further expand M
+
to M
++
=
+1
, (
) = (M
++
1
, . . . ,
n
we have
(): M
++
[= [a
1
, . . . , a
n
] i EM
1
(
) [= [a
1
, . . . , a
n
],
():
1
, . . . ,
n
),
1
, . . . ,
n
) are similar in
.
It is easy to check that
=
+1
is as required.
So we have dened the sequence
, h
: < < ) is an
inverse system of approximations.
(1) There are m
, h
, h
: < < + 1) is an
inverse system of approximations continuous at .
(2) For the following properties, if each m
+1
(for < ) satises the prop-
erty, then so does m
<
W
m
, and let M
= M
m
= F
m
, M
= M
m
and N
W
= N
m
W
and M
b
= M
m
b
below.
First let F
, h
: < ), i.e.,
(): h
onto F
= h
<
Dom(h
),
(): If
< , f
, for [
, ), satisfy h
(f
) = f
when
<
< , then for one and only one f F
we have:
f
=
f
for [
, ) and
f,
=
_
_
,
:
<
_
,
(): every f F
,
are as in (B) of Denition 2.5, i.e., for any
and < we have
< h
(f
, ) = f
,m
,
.
Second, we similarly choose f
m
.
Thirdly, we choose M
) and M
+
b
when
b Rang(f) : f F
& ( <
f
)( W)(
f
)
for some W W
<
. Though we can use the compactness theorem, it seems to me
more transparent to use ultraproduct . So let D be an ultralter on containing
all co-bounded subsets of . Let M
<
M
/D. If f F
, let
f
< and
f
: [
f
, )) be such that
f
< h
(f) = f
, so
f
f
: <
f
)
has no repetitions. Now, when
f
, let
f
m
(. . . ,
, . . .) = c
: < )/D,
where
(
f
, ) c
=
_
h
(f)
m
_
(. . . ,
, . . .) M
,
<
f
c
is any member of M
.
So M
W
is well dened for W W
m()
.
Fourth, if
b = b
: < ())
>
(M
W
), b
= f
m
(. . . ,
, . . .)
<
f
, and
<
and for [
, ): f
,
F
m
, f
,
: < ()) Dom(f
m
), and h
(f
,
) = f
,
then we let
b
= b
is f
m
(. . . ,
, . . .)
<
f
if [
, )
and b
is any member of M
if <
b
=
[,)
M
/D.
We still have to check that if for the same
b we get two such denitions, then they
agree, but this is straightforward.
Fifth, we choose M
b
for other
b
>
(M
) for which M
b
is not yet dened to
satisfy clause (d) of Denition 2.5; note that by the choice of W
m
those choices
do not inuence the preservation of weakly/strongly indiscernible. So m
is well
dened and one can easily check that it is as required.
2.14
Claim 2.15. Assume = + 1 < , and m
1
is a approximation.
(1) There are h
and an approximation m
2
such that m
1
h
m
2
, M
m2
=
M
m1
, M
m2
b
= M
m1
b
, and Dom(h
) = F
m2
.
(2) If m
1
is weakly/strongly nice, then m
2
is weakly/strongly nice.
(3) If m
1
is weakly/strongly indiscernible , then m
2
is weakly/strongly indis-
cernible; simply for E-indiscernible, E E
.
Proof: (1) Should be clear.
Let (m
2
) = , W
m2
= W
m1
, M
m2
= M
m1
and M
m2
b
= M
m1
b
for
b
>
(M
m1
).
Then let
F
m2
= g
f,h
: f F
satisfying h(
f,
) Suc(
f,
) =
f,
) : < ,
where for g = g
f,h
we let
g
=
f
and
g,
= h(
f,
), and if
g,
for <
g
(=
f
), then
g
m2
f,h
(. . . ,
, . . .) = f
m1
(. . . ,
, . . .) M
m1
M
m2
.
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 19
We dene h
by:
Dom(h
) = F
m2
and h
(g
f,h
) = f.
Lastly let
Dom(f
m2
) =
_
g
f,
: <
f
i.e., does not depend on
we have < () g
= g
f,h
_
,
and if h,
f, g = g
f,h
: <
f
) Dom(f
m2
) are as above, ( x) is a [T, ()]term,
x = x
m
2
mean that
(m
1
) = (m
2
) and m
1
h
m
2
with h being the identity on F
m1
F
m2
,
and W
m1
W
m2
, and f
m1
f
m2
, the last mean that if
f Dom(f
m1
) then
f Dom(f
m2
) and the function f
m2
(
f) is equal to the function f
m1
(
f).
(2) Let m
1
<
m
2
mean that
(a): m
1
m
2
,
(b): if
f F
m1
then
f Dom(f
m2
).
Observation 2.17. (1)
is a partial order, m
1
m
1
, and
m
1
<
m
2
m
1
m
2
, and
m
1
m
2
<
m
3
m
1
<
m
3
, and
m
1
<
m
2
m
3
m
1
<
m
3
.
(2) Each
: < ) is <
<
m. So if each m
m
1
.
(2) If m
0
is an approximation, then there is a approximation m
1
such that
m
0
<
m
1
and Dom(f
m1
) = F
m0
.
Proof: 1) Let M
m1
= M
m0
, and M
m1
b
= M
m0
b
for
b
>
(M
m0
). Let W
m1
= W
,
and let
: <
) such that
: < g( ))
is without repetitions and
: < g( ) / W
m0
. Let
=
,
: <
) and
dene
=:
,
: < g(
)), and W
=:
,
: <
for <
. Let
= otp
1
< : (
2
<
1
)(
2
,=
1
).
For each W W
W
m0
, let M
m1
W
be an elementary submodel of M
m1
of
cardinality such that
W
1
W W
1
W
m0
M
m1
W
1
M
m1
W
and
b
>
_
M
m1
W
_
M
m0
b
[M
m1
W
[ M
m0
b
.
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
20 SAHARON SHELAH
Let a
W,i
: i < ) list the elements of M
m1
W
. For <
&
= then
f
,i
= lg(
) = lg(
) = lg(
)-and (f
,i
, ) =
,
, and we dene f
m1
,i
by: if
,
for <
f
,i
, and
: <
f
,i
) =
then
f
m1
,i
(. . . ,
,
, . . .) = a
W,i
.
Next, F
m1
almost is F
m0
f
,i
: <
be a |M
m0
|
+
resplendent elementary extension of M
m0
. We dene an
approximation m
1
as follows:
(a):
m1
=
m0
, W
m1
= W
m0
, M
m1
= M
,
(b): if
b
>
(M
m0
), then M
m1
b
is an elementary extension of M
m0
b
,
(c): f
m1
f
m0
and Dom(f
m1
) = F
m0
,
(d): F
m1
= F
m0
(e): if (f
m1
(
f))(
( x
)) = f, (
f, )
, and
b = f
m1
(. . . ,
, . . .)
<
f
: <
f
) and x
= x
i
: i u),
then
f
m1
(. . . ,
, . . .)
<
f
=
M
m
1
b
(f
i
(. . . ,
, . . .)
<
f
: i u)).
2.18
Main Claim 2.19. Assume m
0
is a weakly nice approximation. Then there is a
weakly good approximation m
1
such that m
0
<
m
1
with W
m1
= W
m0
.
Proof: By 2.18(1)+(2) there is a full term closed m
1
such that m
0
<
m
1
and
Dom(f
m1
) = F
m0
. We would like to correctm
1
so that it is weakly indiscernible.
Let m
2
be an
m1
approximation as guaranteed in the Claim 2.20 below, so it is
good and reecting we clearly see that m
0
m
2
and even m
0
<
m
2
.
Main SubClaim 2.20. (1) Assume m
0
is a weakly nice approximation and
m
0
<
m
1
and Dom(f
m1
) = F
m0
and W
m1
is an ideal (that is closed under
nite union). Then there is a good approximation m
2
such that:
(a):
m2
=
m1
, F
m2
= F
m1
, f
m2
= f
m1
, and W
m2
= W
m1
.
(b): m
0
<
m
2
;
(2) We may add
(c) Assume
(): n < and f
F
m1
,
I
(f
,)
for <
f
, < n, and is a
nite set of formulas in L(
T
)
(): m < and for k < m we have
f
k
= f
k
: <
k
) Dom(f
m1
) and
n
k
< and g
k,
F
m2
(for < n
k
) satisfying
(g
k,
, ) : <
g
k,
) = (
f
k
, ) : <
f
k ),
and
k,
I
(
f
k
,)
for < n
k
, <
f
k , and
k
is a nite set of
formulas in L([
f
, (T)]).
Then we can nd
for < n
k
, <
f
and
k,
for < n
k
, <
f
k for
< n
k
, k < m such that
(i):
I
(f
,)
for <
f
and
k,
I
(
f
k
,)
for <
f
k for < n, k <
m,
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 21
(ii): the sequences
: < n, <
f
k,
: < n
k
, k < m, <
f
k )
and
: < n, <
f
k,
: < n
k
, k < m, <
f
k ) are similar
(see Denition ),
(iii): the type realized by the sequence
f
m2
(. . . ,
, . . .)
<
f
: < n)
in M
m2
is equal to the type which the sequence
f
m1
(. . . ,
, . . .)
<
f
: < n)
realizes in M
m1
,
(iv): for k < m
1
, the
k
type realized by the sequence
g
m2
k,
(. . . ,
k,
, . . .)
<
f
k
: < n
k
)
in the model M
m2
f
k,m
2
:<
k
(. . .
k,
. . .)
<:<
k
is equal to the
k
type
realized by the sequence
g
m1
k,
(. . . ,
k,
, . . .)
<
f
k
: < n
k
)
in the model M
m1
f
k,m
2
:<
k
(. . .
k,
. . .)
<:<
k
(v): if k
1
, k
2
< m then
f
k1,m2
(. . . ,
k1
, . . .)
<
f
k
1
: <
k1
) =
f
k2,m2
(. . . ,
k2
, . . .)
<
f
k
2
: <
k2
)
if and only if
f
k1,m1
(. . . ,
k1
, . . .)
<
f
k
1
: <
k1
) =
f
k2,m1
(. . . ,
k2
, . . .)
<
f
k
2
: <
k2
),
(vi): if < n
k
, k < m,
< n, then
f
m2
(. . . ,
, . . .)
<
f
= f
m2
k,
(. . . ,
k,
, . . .)
<
f
k
if and only if
f
m1
(. . . ,
, . . .)
<
f
= f
m1
k,
(. . . ,
k,
, . . .)
<
f
k
.
Discussion 2.21. Now we have to apply the Ramsey theorem to recapture weak
indiscernibility. Why we only promise m
0
<
m
1
&Dom(f
m1
) = F
m0
, not that m
1
is excellent? As T
b
in order to know T
b, M
m
].
Usually a partition theorem on what we already have is used at this moment,
but partition of innitary functions tend to contradict ZFC. However, in the set
expressing what we need, the formulas are nitary. So using compactness we will
reduce our problem to the consistency of the set of rst order formulas in the
variables
f(. . . ,
, . . .)
<(f)
: f F
m1
and <
f
(f, )
.
This can be easily reduced to the consistency of a set of formulas in L(
T
) (rst
order).
We can get because for all relevant
b we know T
b, M].
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
22 SAHARON SHELAH
Proof: Let Y = y
f(...,,...)
w[f]
: f F
m1
and
: w[
f]) such that
, let
f,
be
[T, lg
f] where w[
f] = w[f
f,
:
f, as above
T
.
Let g
f,
be a one to one function from [T, lg(
f)] onto
f,
which is the identity
on
T
preserve the arity and being a predicate function symbol, individual constant.
Let g
f,
be the mapping from L([T, lg(
f)]) onto L(
f,
) which g
f,
induce.
We now dene a set (the explanations are for the use in the proof of
1
below).
0
=
0
1
2
3
4
5
6
7
8
where
(a):
0
=
(y
f
1
,
(1)
, y
f
2
,+1
(2)
)
t
: where t = truth if and only if
[
1
( + 1) =
2
( + 1)] and < ,
, and
for = 1, 2
[explanation: to satisfy (iii) in clause (B) of Denition 2.5].
(b):
1
= y
f
1
,
(1)
= y
f
2
,
(2)
: < ,
for = 1, 2 and
1
=
2
[explanation:to satisfy (ii) in clause (B) of Denition 2.5].
(c):
2
= y
f(...,...)
w[f]
= (. . . , y
f
(...,,...)
w[f]
. . .) : f, f
: < n) and
(. . . ,
(f
, ), . . .)
<(f
)
, , . . .)
<n
:
f, f
: < n) and f
(...,
,...)
<
, . . .)
<n
: (x
0
, . . . , x
n1
) L(
T
) and f
F
m0
and
=
f
I
(f
,)
for <
and M
m0
[=
[. . . , f
m0
(. . . ,
, . . .)
<
, . . .)
<n
5
= (. . . , y
f
(...,
,...)
<
,...)
<n
: for some
f, and f
: < n) we
have (x
0
, . . . , x
n1
) L(
f,
) and
f Dom(f
m0
), (
f) =
=
f
, f
F
m0
, (
f, ) = (f
, ) for <
, < n and M
m0
[= [. . . f
m
(. . . ,
, . . .)
<
. . .)
<n}
[explanation: this is for being above m
0
, the formulas from the M
m0
s].
(g):
6
= (. . . , y
f
(...
1
,...)
<
, . . .)
<n
(. . . , y
f
(...,
2
,...)
<
, . . .)
<n
:
(x
0
, . . . , x
n1
) L))
T
) and
= (f
), f
F
m1
,
k
I
(f
,)
if k = 1, 2
and <
we have
1
,=
2
F
m
, lg(
f
1
) = lg(
f
2
), (
f
, )
for < lg(
f
) and (
f
1
,
1
) =
(
f
2
,
2
)
1
1
=
2
2
;
L(
f
,
) and g
1
f
1
,
1
(
1
) = g
1
f
2
,
2
(
2
).
[Explanation: this has to show the existence of the M
b
: we can avoid
this if we change the main denition such that instead M
b
we have M
f,
]
(i):
8
= (. . . , y
f
(...,
1
,...)
<
, . . .)
<n
(. . . , y
f
(...,
2
,...)
<
, . . .)
<n
: for
some
f Dom(f
m1
),
f, ,
1
,
2
and f
: < n) we have (x
0
, . . . , x
n1
)
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 23
L(
f,
), f
F
m1
,
f
, (f
, ) =
for <
, and w(
f)
: <
for <
, and
(
f,
1
) =
2
1
=
k
2
, . . .)
w[f]
and the
predicates and relation symbols in each
f,
naturally, M
m
is a model
of or more exactly not M
m
but the common expansion of the M
m
b
-s
for
b f
m
(. . . ,
, . . .)
<
: <
) :
f = f
: <
) Dom(f
m0
),
and
I
(
f,)
.
So it is enough to prove
(
2
): has a model.
We use the compactness theorem, so let
a
be nite. We say that
appears in
a
, if for some variable y
f(...,,...)
w[f]
appearing as a free variable in
some
a
we have
, . . .)
w[f]
appears in
a
(or in ).
Let n
0
= [
a
[. Now, for each
the set of I
appearing in
a
, which we
call J
, is nite but on
1
=: max[J
[ :
is well dened <
0
as well as m
0
= [
a
0
[.
For each
we can nd a nite set u
such that:
(): (i): if
1
,=
2
J
, then min :
1
() ,=
2
() u
(ii): if
[f
1,
(
1
), f
2,
(
2
)]
t
from clause (B) appears in
a
0
, then
, + 1 u
(iii): u
.
(iv): [u
[ (n
1
)
2
+ 2m
0
+ 1.
Clearly n
2
= max[u
[ :
is well dened (<
0
), so without loss of generality,
[u
[ = n
2
.
Let v
be nite, in fact of size [
a
[ = n
0
such that:
(I): if
[y
f
1
,
(1)
, y
f
2
,
(2)
]
t
appears in
a
0
, so 1, 2
,
then
v for 1, 2,
Now, for all
v we replace in
a
all members of J
by one
and
we call what we get
b
, i.e., we identify some variables. It suces to prove
b
is
consistent. Now, by the choice of the set v also
b
is of the right kind, i.e., .
[Why? We should check the formulas , in
a
i
for each i 8; let it be
replaced by
b
. If in
0
a
by clause (ii) of this substitution has
no aect on . If
1
, either
= or
is trivially true. If
3
, clearly
3
. If
4
then
4
as m
0
is nice hence weakly indiscernible, i.e.
clause (b) of Denition 2.9(1) (and the demand f
F
m0
). If
5
, similarly
using clause (c) of Denition 2.9(1). Lastly if
6
we just note that similarly is
preserved and similarly for
7
8
].
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
24 SAHARON SHELAH
We then transform
b
to
c
by replacing each by
by
[]
where
[]
() = () if u
and
[]
() = 0 otherwise. It suces to prove the consistency of
c
. Now, the eect is
renaming variables and again
c
. Let
k
: k < k
) list the
which
appear in
c
such that v. Let
k
=
k
so
k
v, and let = : =
k
: k < k
),
k
k
, ()( < & / u
k
k
() = 0) and is similar
to
i.e., for k
1
, k
2
< k
k1
() <
k2
().
For each we can try the following model as a candidate to be a model of
c
. It expand M
m1
, and if symbols from
f,
T
appear they are interpreted as
their g
1
f,
-images are interpreted in M
m1
f( ):<lg(
f)
. Lastly we assign to the variable
y
f
(. . . ,
, . . .)
<
f
appearing in
c
the element f
m1
(. . . ,
, . . .)
<
f
of M
m1
. Call
this the -interpretation. Considering the formulas in
c
i
for i 0, . . . , 5, 7
they always holds. For the formulas in
c
6
,
8
we can use a partition theorem
on trees with [n
2
[ <
0
levels (use [Sh:E59, 1.16](4), which is an overkill, but has
the same spirit (or [Sh:c, AP2.6, p.662])).
2.20
Claim 2.22. There is an increasing continuous inverse system of approximations
m
, h
: , )
such that each m
is weakly excellent.
Proof: By induction on we choose m
and h
:
1
, < ) is an inverse system of approxima-
tions,
(b): m
:
1
< , < < ) is an inverse system of good weakly excellent
approximations with (m
) = . So by 2.14 we can nd m
, h
( < ) as required.
For = + 1:
By 2.15(1+2) there is m
,0
a weakly nice approximation such that m
,0
.
By 2.19 there is a full term closed approximation m
,1
such that m
,0
,1
and m
,1
is good. We can choose by induction on [1, ] good approximations
m
,
,
increasing continuously, m
,
<
m
,+1
.
For = 1, m
,
is dened; for limit use 2.17(2), for successor use 2.19, and
m
=: m
,
is good by 2.17(3).
2.22
Claim 2.23. Assume m
, h
(e.g., =
2
|T|
). Then there are > pairwise non-isomorphic resplendent
models of T of cardinality .
Proof: Let m = m
and I
, [I[ = and for simplicity
: () = 0
for every large enough <
>
I. Let M
I
be the submodel of M
m
with
universe
_
f
_
. . . ,
(f,)
, . . .
_
<
f
: f F
m
and (f, ) I
:
) is with no repetitions for each <
hence by the indiscernibility the sequence a
: I) is with no repetition, so
|M
I
| [I[ , so |M
I
| = .
Now, M
I
is a -resplendent model of T as m being weakly excellent is full and
resplendent.
For < ,
let a
= f
,m
,
() ( M
I
) for any I
I.
The point is:
(): For
,
+1
,
= ,
,= ( + 1), we have:
the type
_
(x, a
(+1)
) (x, a
) : <
_
is realized in M
I
if and only if I.
[Why? The implication holds by clause (B)(iii) of Denition 2.5. For the other
direction, if c M
I
, then for some W W
, satisfying W I, we have c N
m
W
,
and as / I and [W[ < clearly for some < we have
:
W = .
Let c = f
m
(. . . ,
, . . .)
<
f
, where f F
, so
] .
By the denition of a system, m is full. Choose I
; recalling m
it
suces to prove that
M
m
[= [(h
(f))(. . . ,
, . . .)
<
f
, f
,
(a
)]
[(h
(f))(. . . ,
, . . .)
<
f
, f
,
(a
)] .
But m
2
|T|
,
then T has 2
= f
,
(), and
for
>
of length + 1 let
let a
= f
,+1
(), f
,+1
(
) is immaterial. Let
( x
: < )) = (y)
_
<
((y, x
,0
) (y, x
,1
)
_
.
Now we can choose f
I
: M
I
M
,
such that
(a): if f
I
(b) = (t
i
: i < i
)) such that t
i
I
,
and (f, ) = t
for <
f
we have b = f
m
(. . . , t
, . . .)
<
f
and
(b): f
1
(b) = I if b = a
(see above).
The new point is that we have to prove the statement () in [Sh:331, 2.3](c)().
So assume that for = 1, 2 and < :
b
2
(M
I
), f
I
) =
), and
t
1
= t
2
,
: <
=
2
,
t
1
=
t
2
) though possibly I
1
,= I
2
), and the truth value of each statement
( I
)
_
i<
i
= t
i,i
i
_
does not depend on 1, 2. Assume further that M
I1
[= (. . . ,
b
1
, . . .)
<
, and
we shall prove that M
I2
[= (. . . ,
b
2
, . . .)
<
; this suces.
First note that, as f
I1
, f
I2
f
(
)
, necessarily
b
1
=
b
2
(so call it
b
). Now,
M
I1
[= (. . . ,
b
1
, . . .)
<
means that for some c
1
M
I1
we have
M
I1
[=
<
[c
1
, b
,0
] [c
1
, b
,1
],
and let c
1
= f
1
(. . . , , . . .)
w[f1]
. Let
J = : t
,j
for some < , j < g(
) and
J
+
= : I
).
Now we can choose
I
(f1,)
from I
2
such that (f
1
, ) J
+
=
(f
1
, ). Let f
2
F
m
be such that h
(f
2
) = h
(f
1
) and (f
2
, ) =
for <
f2
=
f2
. Easily, c
2
= f
m
(. . . ,
, . . .)
<
f
1
M
I2
witness that
M
I2
[= (y)[
<
(x, b
,0
) (x, b
,1
)]
(recalling M
I1
, M
I2
M
(
)
).
3.1
Recall and add
(
3
6
3
)
r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
5
-
0
9
m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
5
-
2
4
ON SPECTRUM OF RESPLENDENT MODELS 27
Denition 3.2. (1) E
1
(see De-
nition 2.8(3)) except that we omit clause (iv) there.
(2) For < dene E
2
), with no repetition
1
E
2
2
if and only if
(i)
1
,
2
>
(
.
(iii)
1
=
2
for <
.
(iv) for every
, the sets u
= <
: u
1
),
2
: u
2
) are similar.
Claim 3.3. (1) In 2.22 we can demand that every m
is E
1
indiscernible i.e.
get the strong version.
(2) Moreover we can get even E
2
-indiscernibility.
Proof:
(1) Very similar to the proof of 2.22. In fact, we need to repeat 2 with minor
changes. One point is that dening good we use E
1