4

0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL
IN POWER AND -GENERATED MODELS
Steven Givant and Saharon Shelah
Abstract. We investigate a notion called uniqueness in power that is akin to
categoricity in power , but is based on the cardinality of the generating sets of
models instead of on the cardinality of their universes. The notion is quite useful
for formulating categoricity-like questions regarding powers below the cardinality of
a theory. We prove, for (uncountable) universal theories T, that if T is -unique for
one uncountable , then it is -unique for every uncountable ; in particular, it is
categorical in powers greater than the cardinality of T.
It is well known that the notion of categoricity in power exhibits certain irregu-
larities in small cardinals, even when applied to such simple theories as universal
Horn theories. For example, a countable universal Horn theory categorical in one
uncountable power is necessarily categorical in all uncountable powers, by Morleys
theorem, but it need not be countably categorical.
Tarski suggested that, for universal Horn theories T, this irregularity might be
overcome by replacing the notion of categoricity in power by that of freeness in
power. T is free in power , or -free, if it has a model of power and if all such
models are free, in the general algebraic sense of the word, over the class of all
models of T. T is a free theory if each of its models is free over the class of all
models of T. It is trivial to check that, for > [T[, categoricity and freeness in
power are the same thing. For [T[ they are not the same thing. For example,
the (equationally axiomatizable) theory of vector spaces over the rationals is an
example of a free theory, categorical in every uncountable power, that is -free, but
not -categorical. Tarski formulated the following problem: Is a universal Horn
theory that is free in one innite power necessarily free in all innite powers? Is it
a free theory?
Baldwin, Lachlan, and McKenzie, in Baldwin-Lachlan [1973], and Palyutin, in
Abakumov-Palyutin-Shishmarev-Taitslin [1973], proved that a countable -categorical
universal Horn theory is
1
-categorical, and hence categorical in all innite powers.
Thus, it is free in all innite powers. Givant [1979] showed that a countable -free,
but not -categorical, universal Horn theory is also
1
-categorical, and in fact it is
a free theory. Further, he proved that a universal Horn theory, of any cardinality
, that is -free, but not -categorical, is necessarily a free theory.
Shelahs research was partially supported by the National Science Foundation and the United
States-Israel Binational Science Foundation. This article is item number 404 in Shelahs bibliog-
raphy. The authors would like to thank Garvin Melles for reading a preliminary draft of the paper
and making several very helpful suggestions.
Typeset by A
M
S-T
E
X
1
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


2 STEVEN GIVANT AND SAHARON SHELAH
Independently, Baldwin-Lachlan, Givant, and Palyutin all found examples of
countable
1
-categorical universal Horn theories that are not -free, and of count-
able
1
- and -categorical universal Horn theories that are not free theories.
Thus, Tarskis implicit problem remains: nd a notion akin to categoricity in
power that is regular for universal Horn theories, i.e., if it holds in one innite
power, then it holds in every innite power.
One of the diculties with the notions of categoricity in power and freeness in
power is that they are dened in terms of the cardinality of the universes of models
instead of the cardinality of the generating sets. This causes diculties when trying
to work with powers < [T[.
Lets call a model A strictly -generated if is the minimum of the cardinalities
of generating sets of A. We dene a theory T to be -unique if it has, up to
isomorphisms, exactly one strictly -generated model. For cardinals > [T[, the
notions of -categoricity, -freeness, and -uniqueness coincide (in the case when
T is universal Horn). When = [T[, we have
-categoricity -freeness -uniquenes ,
but none of the reverse implications hold.
Givant [1979], p. 24, asked, for universal Horn theories T, whether -uniqueness
is the regular notion that Tarski was looking for, i.e., (1) Does -uniqueness for one
innite imply it for all innite ? For countable T, he answered this question ar-
matively by showing that -uniqueness is equivalent to categoricity in uncountable
powers. For uncountable T, he provided a partial armative answer by showing
that categoricity in power > [T[ implies -uniqueness for all innite , and is, in
turn, implied by -uniqueness when = [T[ and is regular. However, the problem
whether -uniqueness implies categoricity in powers > [T[ when < [T[, or
when < = [T[ and is singular, was left open.
In this paper we shall prove the following:
Theorem. A universal theory T that is -unique for some > is -unique for
every > . In particular, it is categorical in powers > [T[.
It follows from the previous remarks that a universal Horn theory T which is
-unique for some > is -unique for every . Thus, the only part of (1)
that still remains open is the case when T is uncountable and -unique. A more
general formulation of this open problem is the following:
Problem. Is an -unique universal theory T necessarily categorical in powers >
[T[ ? In particular, is a countable -unique (or -categorical ) universal theory

1
-categorical ?
An example due to Palyutin, in Abakumov-Palyutin-Shishmarev-Taitslin [1973],
shows that a countable universal theory categorical in uncountable powers need
not be -unique. In fact, in Palyutins example the nitely generated models are
all nite, and there are countably many non-isomorphic, strictly -generated mod-
els. Thus, for universal theories, -uniqueness for some > does not imply
-uniqueness.
To prove our theorem, we shall show that, under the given hypotheses, the
theory of the innite models of T is complete, superstable, and unidimensional,
and that all suciently large models are a-saturated. Thus, we shall make use of
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS 3
some of the notions and results of stability theory that are developed in Shelah
[1990] (see also Shelah [1978]). We will assume that the reader is acquainted with
the elements of model theory and with such basic notions from stability theory
as superstability, a-saturatedness, strong type, regular type, and Morley sequence.
We begin by reviewing some notation and terminology, and then proving a few
elementary lemmas.
The letters m and n shall denote nite cardinals, and and innite cardinals.
The cardinality of a set U is denoted by [U[. The set-theoretic dierence of A and
B is denoted by A B. If is a function, and a = a
0
, . . . , a
n1
) is a sequence of
elements in the domain of , then ( a) denotes the sequence (a
0
), . . . , (a
n1
)).
We denote the restriction of to a subset X of its domain by X, and a similar
notation is employed for the restriction of a relation. A sequence X

: < ) of
sets is increasing if X

for < < , and continuous if X

=

<
X

for
limit ordinals < .
We use German letters A, B, C, ... to denote models, and the corresponding Ro-
man letters A, B, C, ... to denote their respective universes. If (x
0
, ..., x
n1
) is a
term in a (xed) language L for A, and if and a an n-termed sequence of elements in
A, symbolically a
n
A, then the value of at a in A is denoted by
A
[ a], or simply
by [ a]. A similar notation is used for formulas. Suppose a
n
A and X A.
The type of a over X (in A), i.e. the set of formulas in the language of A, x )
xX
that are satised by a in the latter model, is denoted by tp
A
( a, X), or simply by
tp( a, X), when no confusion can arise. The strong type of a over X (in A), i.e., the
set of formulas in the language of A, b )
bA
that are almost over X and that are
satised by a, is denoted by stp
A
( a, X), or simply by stp( a, X). If p( x) is a strong
type and E A is a base for p in A, then p

E denotes the set of formulas in p

that are almost over E.
We write A B to express that A is a submodel of B. The submodel of A
generated by a set X A is denoted by Sg
A
(X), or simply by Sg(X), and its
universe by Sg(X). A model is -generated if it is generated by a set of cardinality
, and strictly -generated if it is -generated, but not -generated for any < .
Every model A has a generating set of minimal cardinality, and hence is strictly
-generated for some (nite or innite) . If X is a generating set of A of minimal
cardinality, and Y is any other generating set of A, then there must be a subset Z
of Y of power at most [X[ + such that Z generates X, and hence also A. A set
X A is irredundant (in A) if, for every Y X we have Sg(Y ) ,= Sg(X).
A model A is an n-submodel of B, and B an n-extension of A, in symbols
A
n
B, if for every
n
-formula (x
0
, ..., x
k1
) in the language of A, and every
a
k
A, we have A [= [ a] i B [= [ a]. A 0-submodel of B is just a submodel in
the usual sense of the word, and an elementary submodelin symbols, A Bis
just an n-submodel for each n. We write A B to express the fact that A is a
proper elementary submodel of B, i.e., A B and A ,= B. A theory is model
complete if, for any two models A and B, we have A B i A B. For any
theory T, we denote by T

the theory of the innite models of T.

We shall always use the phrase dense ordering to mean a non-empty dense linear
ordering without endpoints. It is well-known that the theory of such orderings
admits elimination of quantiers. Hence, in any such ordering U = U , <), if a
and

b are two sequences in
n
U that are atomically equivalent, i.e., that satisfy the
same atomic formulas, then they are elementarily equivalent, i.e., they satisfy the
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


4 STEVEN GIVANT AND SAHARON SHELAH
same elementary formulas.
Fix a linear ordering V = V , <), and set W =

V =

V . Let be
the (proper) initial segment relation on W, and, for each , let P

be the set
of elements in W with domain , i.e., P

=

V . There is a natural lexicographic
ordering, < , on W induced by the ordering of V : f < g i either f g or else
there is a natural number n in the domain of both f and g such that fn = gn
and f(n) < g(n) in V. Take h to be the binary function on W such that, for any
f, g in W, h(f, g) is the greatest common initial segment of f and g. We shall call
the structure W= W , <, , P

, h)

the full tree structure over V with +1

levels, or, for short, the full tree over V. Any substructure of the full tree over
V that is downward closed, i.e., closed under initial segments, is called a tree over
V. A tree is any structure isomorphic to a tree structure over some ordering. A
tree U over a dense order V is itself called dense if: (i) for every f in U with nite
domain, say n, the set of immediate successors of f in U, i.e., the set of extensions
of f in U with domain n +1, is densely ordered by < in U, or, put a dierent way,
g(n) : f g is dense under the ordering inherited from V; (ii) every element in U
with a nite domain is an initial segment of an element in U with domain . Just
as with dense orderings, the theory of the class of dense trees admits elimination
of quantiers.
A model A is -homogeneous if, for every cardinal < and every pair a,

b

A
of elementarily equivalent sequences, there is an automorphism of A taking a to

b. It is well-known that, for regular cardinals , any model has -homogenous

elementary extensions (usually of large cardinality). It follows from our remarks
above that, if U is a -homogeneous dense ordering or tree, and if a,

b

U are
atomically equivalent (where < ), then there is an automorphism of U taking a
to

b. A model A is weakly -homogenous provided that, for every n, every pair of
sequences a, a

n
A that are elementarily equivalent, and every b A, there is a
b

A such that ab) and a

) are elementarily equivalent. It is well known that

a countable, weakly -homogenous model is -homogeneous. Dense orderings and
dense trees are always weakly -homogeneous.
We turn, now, to some notions from stability theory. A model A is a-saturated
if, for any strong type p (consistent with the theory of A, a )
aA
), if E A is a
nite base for p, then p

E is realized in A. We say that A is a-saturated in B, in

symbols A
a
B, provided that A B and that, for any strong type p of B based
on a nite subset E of A, if p

E is realized in B, then it is realized in A. A model

A is a-prime over a set X if (i) X A and A is a-saturated; (ii) whenever B is an
a-saturated model elementarily equivalent to A and such that X B, then there is
an elementary embedding of A into B that leaves the elements of X xed. (Here,
we assume that A and B are elementary substructures of some monster model).
As is shown in Shelah [1990], Chapter IV, Theorem 4.18, for complete, superstable
theories, the a-prime model over X exists and is unique, up to isomorphic copies
over X. We shall denote it by Pr
a
(X). Let A be a model and B, C subsets of A
with C B. A type p( x) over B splits over C if there are

b, c from B such that
tp(

b, C) = tp( c, C), and there is a formula ( x, y) over C such that ( x,

b) and
( x, c) are both in p.
We now x a universal theory T in a language L of arbitrary cardinality. In
what follows let L

be an expansion of L with built-in Skolem functions, and T

any
Skolem theory in L

that extends T. Without loss of generality, we may assume

that, for every term (x
0
, ..., x
n1
) of L

there is a function symbol f of L

of rank
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS 5
n such that the equation f(x
0
, ..., x
n1
) = holds in T

.
For every innite ordering U = U , ) and (complete) Ehrenfeucht-Mostowski
set of formulas of L

compatible with T

, there is a model M

= EM(U, )
of T

called a (standard) Ehrenfeucht-Mostowski model of T

such that M

is
generated by U (in particular, U M

), and U is a set of -indiscernibles in M

with respect to the ordering of U, i.e., if (x

0
, ..., x
n1
) , and if a
n
U satises
a
i
< a
j
for i < j < n, then M

[= [ a].
Shelah [1990], Chapter VII, Theorem 3.6, establishes the existence of certain gen-
eralizations of Ehrenfeucht-Mostowski models. Suppose that T is, e.g., nonsuper-
stable, and that the sequence
n
( x, y) : n < ) of formulas witnesses this non-
superstability (see, e.g., Shelah [1990], Chapter II, Theorems 3.9 and 3.14). Then
there is a generalized Ehrenfeucht-Mostowski set from which we can construct,
for every tree U, a generalized Ehrenfeucht-Mostowski model M

= EM(U, ) of
T

. In other words, given a tree U, there a model M

of T

generated by a sequence
a
u
: u U) of -indiscernibles with respect to the atomic formulas of U; in par-
ticular, if w and v in
n
U are atomically equivalent in U, then a
w
0
, . . . , a
w
n1
) and
a
v
0
, . . . , a
v
n1
) are elementarily equivalent in M

. Moreover, for w in P
M

n
and v
in P
M

, we have that M

[=
n
[ a
w
, a
v
] i w v. In general, the sequences a
u
may
be of length greater than 1. However, to simplify notation we shall act as if they
all have length 1, and in fact, we shall identify a
u
with u. Thus, we shall assume
that M

is generated by U and that two sequences from U which are atomically

equivalent in U are elementarily equivalent in M

.
Here are some well-known facts about Ehrenfeucht-Mostowski models. Let U, V
be innite orderings or trees, and M

= EM(U, ) , N

= EM(V, ). (In the case

of trees, we must assume that M

and N

exist.)
Fact 1. If is any embedding of V into U, then the canonical extension of to
N

is an elementary embedding of N

into M

.
In particular,
Fact 2. Any automorphism of U extends to an automorphism of M

.
Fact 3. If V U, then Sg
M

(V ) M

, and Sg
M

(V ) is isomorphic to N

via a
canonical isomorphism that is the identity on V .
Fact 4. If U is a linear order, then U is irredundant in M

. If U is a tree, then
for any element f and subset X of U, if f is not an initial segment of any element
of X, then f is not generated by X in M

.
We shall always denote the reduct of M

to L by M, in symbols, M = M

L.
Facts 1, 2, and 4 transfer automatically from M

to M. However, M is usually not

generated by U. We shall therefore formulate versions of the above facts that apply
to Sg
M
(U), and, more generally, to a collection of models that lie between Sg
M
(U)
and Sg
M

(U) = M

.
Denition. Suppose M

= EM(U, ) , and K L

L.
(i) For each f K, let R
f
be the relation corresponding to f
M

, i.e., R
f
[a
0
, . . . , a
n
]
i f
M
[a
0
, . . . , a
n1
] = a
n
. Set M
K
= M, R
f
)
fK
.
(ii) For each V U set
KV = V f
M

[ a] : f K, a from V .
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


6 STEVEN GIVANT AND SAHARON SHELAH
Thus, the elements of KV are just the elements of V , together with the elements
of M

that can be obtained from sequences of V by single applications of f in K.

Lemma 1. Let U, V be dense orderings or dense trees, M

= EM(U, ) , N

=
EM(V, ), and K L

L. (Thus, in the case of dense trees, we postulate that

generalized Ehrenfeucht-Mostowski models for exist.)
(i) U is a set of indiscernibles in Sg
M
(K U) for the atomic formulas of
(with respect to the atomic formulas of U); in particular, if a and

b are
two atomically equivalent sequences in U, then they satisfy the same atomic
formulas in Sg
M
(KU).
(ii) U is a set of indiscernibles in Sg
M
(KU) for some complete Ehrenfeucht-
Mostowski set compatible with T, i.e., if a and

b are two atomically equiva-
lent sequences in U, then they are elementarily equivalent in Sg
M
(KU).
(iii) If embeds V into U, then the canonical extension of to Sg
N
(K V )
elementarily embeds Sg
N
(KV ) into Sg
M
(KU).
In particular:
(iv) Every automorphism of U extends canonically to an automorphism of Sg
M
(K
U).
(v) If V U, then Sg
M
(KV ) Sg
M
(KU).
The lemma continues to hold if we replace M and N everywhere by M
K

and N
K
respectively.
Proof. Since satisfaction of atomic formulas is preserved under submodels, part (i)
is trivial. Part (i) directly implies (iv), and part (v) implies (iii). Thus, we only
have to prove (ii) and (v). We begin with (v).
First, assume that U is -homogeneous. We easily check that the Tarski-Vaught
criterion holds. Here are the details. Suppose a is an n-termed sequence from
Sg(K V ) satisfying x(x, y) in Sg(K U); say, b is an element of Sg(K U)
such that b) a satises (x, y). Let c and

d be sequences of elements from V and
U that generate a and b respectively. Thus, there are f
0
, . . . , f
k1
and g
0
, . . . , g
l1
in K, and a term ( z, w) and terms
0
( u, v), . . . ,
n1
( u, v) in L such that (adding
dummy variables to simplify notation)
a
i
=
i
_
f
0
[ c], . . . , f
k1
[ c], c

for i < n, and b =

_
g
0
[

d], . . . , g
l1
[

d],

d

.
Hence, the sequence

_
g
0
[

d ], . . . , g
l1
[

d ],

d

,
0
_
f
0
[ c], . . . , f
k1
[ c], c

, . . . ,
n1
_
f
0
[ c], . . . , f
k1
[ c], c
_
satises in Sg(KU). Since V is dense, there is a sequence e from V such that c

d
is atomically equivalent to c e in U. By -homogeneity there is an automorphism of
U taking c

d to c e. In view of (iv), this automorphism extends to an automorphism

of Sg(KU). Thus,

_
g
0
[ e], . . . , g
l1
[ e], e

,
0
_
f
0
[ c], . . . , f
k1
[ c], c

, . . . ,
n1
_
f
0
[ c], . . . , f
k1
[ c], c
_
satises in Sg(K U). Since [g
0
[ e], . . . , g
l1
[ e], e] is in Sg(K V ), this shows
that the Tarski-Vaught criterion is satised. Hence, Sg(KV ) Sg(KU).
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS 7
Now let U be an arbitrary dense ordering or dense tree. Take W to be an -
homogeneous extension of U, and set P

= EM(W, ). Thus, M

, and hence
M P, by Fact 3. By the case just treated we have
Sg
M
(KU) = Sg
P
(KU) Sg
P
(KW)
and
Sg
M
(KV ) = Sg
N
(KV ) Sg
N
(KW) .
Hence, Sg
M
(KV ) Sg
M
(KU), as was to be shown. This proves (v).
For (ii), suppose a,

b
n
U are atomically equivalent in U. Let W be an -
homogeneous elementary extension of U that includes the elements of a and

b, and
set P

= EM(W, ). By Fact 3 we may assume that M

, and hence that

M P. Since W is -homogeneous, there is an automorphism of W taking a to

b. By Fact 2, this extends to an automorphism of P

. Now maps W one-one

onto itself and preserves all the operations of P

. In particular, an appropriate
restriction is an automorphism of Sg
P
(K W). Thus, a and

b are elementarily
equivalent in Sg
P
(K W). By part (v) they remain elementarily equivalent in
Sg
P
(KU) = Sg
M
(KU).
The above proof obviously remains valid if we replace (implicit and explicit oc-
currences of) M and N everywhere by M
K
and N
K
respectively.
Lemma 2. Suppose U has power , M

= EM(U, ), and K L

has power at
most . Then Sg
M
(KU) is strictly -generated.
Proof. Since [U[ = [K[, we have [KU[ = . Thus, Sg(KU) is -generated.
Suppose now, for contradiction, that Sg(K U) is -generated for some (nite or
innite) < . A standard argument then gives us a set V U, of power when
, and nite when < , such that KV generates Sg(KU). In particular,
K V generates U in M. But then V generates U in M

, which is impossible, by
Fact 4. Indeed, either U is irredundant in M

or else a cardinality argument gives

us an f in U that is not an initial segment of any element of V .
Main Hypothesis. Throughout the remainder of the paper we x an uncountable
cardinal and assume that T is a -unique universal theory in a language L of
arbitrary cardinality.
As before, L

will be an arbitrary expansion of L with built-in Skolem functions,

T

a Skolem theory in L

extending T, and a standard or generalized Ehrenfeucht-

Mostowski set compatible with T

. Again, recall that if M

= EM(U, ), then we
set M= M

L. Our rst goal is to show that T

the theory of the innite models

of Tis complete.
Lemma 3. Let U be a dense ordering of power , and M

= EM(U, ). Then
for every n and every countable set H L

L, there is countable set K, with

H K L

L, such that
Sg
M
(KU)
n
M .
The same is true if U is a dense tree, provided that M

exists.
Proof. The proof is by induction on n, for all H at once. The case n = 0 is trivial:
take K = H. Assume, now, that the lemma holds for a given n 0 and for all
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


8 STEVEN GIVANT AND SAHARON SHELAH
countable sets H L

L. Suppose, for contradiction, that H

0
is a countable
subset of L

L such that,
(1) For every countable K with H
0
K L

L we have
Sg
M
(KU) ,
n+1
M .
We construct an increasing sequence H

: <
1
) of countable subsets of L

L
such that, setting
A

= Sg
M
(H

U) ,
we have
(2) A

n
M and A

,
n+1
A
+1
for all <
1
.
The set H
0
is given. Suppose H

has been constructed. Since A

,
n+1
M, by
(1), there is a
n
-formula

(x
0
, . . . , x
p1
, y
0
, . . . , y
q1
) in L, appropriate terms

0
, . . . ,
q1
in L and function symbols f
0
, . . . , f
r1
in H

, and a sequence a =
a
0
, . . . , a
s1
) from U such that (adding dummy variables to simplify notation)
(3) f
0
[ a], . . . , f
r1
[ a], a
0
, . . . , a
s1
) satises x
0
. . . x
p1

(x
0
, . . . , x
p1
,
0
, . . . ,
q1
)
in M, but not in A

.
(Here we are using the fact that H

U generates A

.) Thus, there are Skolem

functions g
0
, . . . , g
p1
in L

such that
(4) g
0
[ a], . . . , g
p1
[ a], f
0
[ a], . . . , f
r1
[ a], a
0
, . . . , a
s1
) satises the
n
-formula

(x
0
, . . . , x
p1
,
0
, . . . ,
q1
) in M .
By the induction hypothesis there is a countable H
+1
with
H

g
0
, . . . , g
p1
H
+1
L

L
and such that, setting A
+1
= Sg(H
+1
U), we have A
+1

n
M. However,
we put witnesses in A
+1
to insure that A

,
n+1
A
+1
. At limit stages take
H

=

<
H

. This completes the construction.

Set
G =

<
1
H

and B = Sg
M
(GU) .
Clearly, B =

<
1
A

. Therefore, from (2) and our construction we obtain

(5) A

n
B
n
M and A

,
n+1
B for every <
1
.
Next, we prove:
(6) For every <
1
and every countable, dense X U we have
Sg(H

X) ,
n+1
B .
Indeed, x <
1
and select a sequence a so that (3) and (4) hold. Given X U,
choose a sequence

b from X that is atomically equivalent to a in U. This is possible
because X is dense. By indiscernibility in M

, and (4),
g
0
[

b], . . . , g
p1
[

b], f
0
[

b], . . . , f
r1
[

b], b
0
, . . . , b
s1
)
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS 9
satises

(x
0
, . . . , x
p1
,
0
, . . . ,
q1
) in M, and hence also in B, by (5). Suppose,
for contradiction, that Sg(H

X)
n+1
B. Then
(7) f
0
[

b], . . . , f
r1
[

b], b
0
, . . . , b
s1
) satises x
0
. . . x
p1

(x
0
, . . . , x
p1
,
0
, . . . ,
q1
)
in Sg(H

X) .
Now we have Sg(H

X) A

, by Lemma 1(v). Therefore, (7) holds with Sg(H

X) replaced by A

. Since a and

b are elementarily equivalent in the expansion
Sg
M
H

(H

U) of A

, by Lemma 1(ii), we conclude that f

0
[ a], . . . , f
r1
[ a], a
0
, . . . , a
s1
)
satises
x
0
. . . x
p1

(x
0
, . . . , x
p1
,
0
, . . . ,
q1
)
in A

. But this contradicts (3); so the proof of (6) is completed.

We now work towards a contradiction of (6) by producing a <
1
and a
countable, dense set X U such that
(8) Sg(H

X) B .
Since B = Sg
M
(G U) and Sg
M
(U) are both strictly -generated, by Lemma 2,
they are isomorphic, by -uniqueness. Thus, there is a set V B and a dense
ordering _ on V such that V = V , _) and U are isomorphic, V generates B,
and V is a set of indiscernibles in B with respect to atomic formulas (under the
ordering _).
We dene increasing sequences, X
n
: n ) and Y
n
: n ), of countable,
dense subsets of U and V respectively, and an increasing sequence
n
: n ) of
countable ordinals, such that
(9) Sg
B
(H

n
X
n
) Sg
B
(Y
n
) Sg
B
(H

n+1
X
n+1
) .
Indeed, set
0
= 0 and take X
0
to be an arbitrary countable, dense subset of U.
Since H
0
is countable, so is H
0
X
0
. Therefore, there is a countable Y
0
V that
generates H
0
X
0
in B. By throwing in extra elements, we may assume that Y
0
is
dense.
Now suppose that
n
, X
n
and Y
n
have been dened. Since Y
n
is countable and
B =

<
1
A

, there is a
n+1
<
1
such that Y
n
A

n+1
. Without loss of
generality we may take
n+1
>
n
. Since H

n+1
U generates A

n+1
, there is a
countable subset X
n+1
U such that H

n+1
X
n+1
generates Y
n
in B. Of course we
may choose X
n+1
so that it is dense and includes X
n
. As before, we now choose
a countable, dense Y
n+1
V that generates X
n+1
in B and includes Y
n
. This
completes the construction.
Set
= sup
n
: n , X =

n
X
n
, Y =

n
Y
n
.
Then <
1
and H

=

n
H

n
, by denition of the sequence H

: <
1
).
Also, X , ) and Y , _) are countable, dense submodels of U and V respectively.
By (5), (9), and Lemma 1(v) we have
Sg
M
(H

X) = Sg
B
(H

X) =

n<
Sg
B
(H

n
X
n
) =

n<
Sg
B
(Y
n
) = Sg
B
(Y )
and
Sg
B
(Y ) = Sg
M
(Y ) Sg
M
(V ) = Sg
B
(V ) = B .
Thus, we have constructed a countable <
1
and a countable, dense X U such
that (8) holds. This contradicts (6).
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


10 STEVEN GIVANT AND SAHARON SHELAH
Lemma 4. For every countable H L

L, there is a countable K, with H

K L

L, such that, whenever V is a dense ordering and N

= EM(V, ), we
have
Sg
N
(KV ) N .
The same is true when V is a dense tree, provided that EM(U, ) exists for dense
trees U.
Proof. Let U be a dense ordering of power , and set M

= EM(U, ). We use
Lemma 3 to dene an increasing sequence H
n
: n < ) of countable subsets of
L

L such that
(1) Sg
M
(H
n
U)
n
M .
Indeed, we set H
0
= H, and given H
n
satisfying (1), we apply Lemma 3 to obtain
H
n+1
. Setting K =

n
H
n
, we easily check that
(2) Sg
M
(KU) M .
Now suppose that V is any dense ordering or tree, and that N

= EM(V, ).
We rst treat the case when V is a submodel of U. By Fact 3 we may assume,
without loss of generality, that N

. Hence,
(3) N M .
In view of (3) and Lemma 1(v), we get
(4) Sg
N
(KV ) = Sg
M
(KV ) Sg
M
(KU) .
Combining (2)(4) gives
(5) Sg
N
(KV ) N .
For the case when V is not a submodel of U, take a countable, dense submodel
W of V. Set P

= EM(W, ). Again, we may assume that P

, so
(6) P N ,
(7) Sg
P
(KW) = Sg
N
(KW) Sg
N
(KV ) .
Furthermore, there is an embedding of W into U, and this embedding extends to
an elementary embedding of P

into M

. Hence, as usual,
(8) induces an elementary embedding of P into M ,
(9) induces an elementary embedding of Sg
P
(KW) into Sg
M
(KU) .
Using (2) and (6)(9), we readily verify that (5) holds. Here are the details. Let
( x) be a formula of L, and a a sequence of appropriate length from Sg
N
(KV ).
Then there is a nite sequence v from V that generates a (with the help of K).
Choose a v

in W that is atomically equivalent to v, and let a

be the sequence
obtained from v

in the same way that a is obtained from v. Then a and a

are
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS11
elementarily equivalent in N

and therefore also in Nand in Sg

N
(KV ) by Fact
3 and Lemma 1(ii). Hence,
N [= [ a] i N [= [ a

]
i P [= [ a

] by (6),
i M[= [( a

)] by (8),
i Sg
M
(KU) [= [( a

)] by (2),
i Sg
P
(KW) [= [ a

] by (9),
i Sg
N
(KV ) [= [ a

] by (7),
i Sg
N
(KV ) [= [ a] .

Theorem 5. T

is complete.
Proof. Let B
1
and B
2
be two innite models of T. For i = 1, 2, let T
i
be the
theory of B
i
, let L

be an expansion of the language of T

i
with built-in Skolem
functions, T

i
a Skolem theory in L

extending T
i
, and
i
a standard Ehrenfeucht-
Mostowski set in L

compatible with T

i
. Let U be a dense ordering of power ,
and set M

i
= EM(U,
i
) and M
i
= M

i
L. By Lemma 4 there is a countable
K
i
L

L such that
(1) Sg
M
i
(K
i
U) M
i
.
Now Sg
M
1
(K
1
U) and Sg
M
2
(K
2
U) are strictly -generated models of T, by
Lemma 2. Therefore, they are isomorphic, by the -uniqueness of T. From this
and (1) we conclude that M
1
and M
2
are elementarily equivalent. But M

i
, and
hence also M
i
, is a model of the theory of B
i
. In consequence, B
1
and B
2
are
elementarily equivalent, as was to be shown.
Our next goal is to prove:
Theorem 6. T

is superstable.
Proof. Set U = (+1) Q and U = U , <), where < is the lexicographic ordering
on U. We dene a substructure V of the full tree over U by specifying its universe.
It is the smallest set V satisfying the following conditions: all nite levels of the full
tree are included in V , i.e.,

n
n
U V ; all eventually constant functions from

U are in V ; for each limit ordinal <

1
we choose a strictly increasing function
f

in

1
such that supf

(n) : n = , and we put into V a copy of f

from

V called g

and determined by: g

(n) = f

(n) , 0 ) for each n.

(1) For every countable set W U, the set V

W is also countable.
Indeed, the nite levels,
n
W, and the set of eventually constant functions of

W
are all countable. Moreover, the domain of W, i.e., : , q ) W for some q
Q, is countable, so the set of g

in

W, with <
1
, is also countable. This proves
(1).
Let be a standard Ehrenfeucht-Mostowski set compatible with T

(our Skolem
extension of T), and set M

= EM(U, ). Assume, for contradiction, that T

4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


12 STEVEN GIVANT AND SAHARON SHELAH
is not superstable, and let
n
(x, y) : n < ) be a sequence of formulas wit-
nessing this nonsuperstability. Then, as mentioned in the preliminaries, there is
an Ehrenfeucht-Mostowski set compatible with T

such that the generalized

Ehrenfeucht-Mostowski model N

= EM(V, ) exists, and, for u from P

V
n
and
v from P
V

, we have N

[=
n
[u, v] i u v. Set M= M

L and N = N

L.
By Lemma 4, there are countable sets J, K L

L such that
(2) Sg(JU) M and Sg(KV ) N .
By Lemma 2, Sg(J U) and Sg(K V ) are both strictly -generated. Hence,
-uniqueness implies that they are isomorphic. Let be such an isomorphism.
We now dene two strictly increasing, continuous sequences, X

: <
1
) and
Y

: <
1
), of countable, dense subsets of U and V respectively, each Y

being
downward closed, i.e., closed under initial segments. (i) Let X
0
= Q and
take Y
0
to be any countable, dense, downward closed subset of V . Suppose, now,
that X

and Y

have been dened. Since both of these sets are countable, by

assumption, so are JX

and KY

. For any a in JU there is obviously a nite

subset C
a
V such that K C
a
generates (a) (in Sg(K V )). (ii) For each a in
JX

, put the elements of C

a
into Y
+1
. Since JX

is countable, this adds only

countably many elements to Y
+1
. (iii) Similarly, for every b in K Y

, choose a
nite set D
b
U such that J D
b
generates
1
(b) (in Sg(J U)), and put each
element of D
b
into X
+1
. (iv) For each , q ) in X

, put all of Q into

X
+1
. (v) Put all of

( Q) V into Y
+1
. The latter set remains countable,
by (1). (vi) If necessary, add countably many more elements to X
+1
and Y
+1
to
insure that these sets are dense, that X

X
+1
and Y

Y
+1
, and that Y
+1
is
downward closed. (vii) For limit ordinals , set X

=

<
X

and Y

=

<
Y

.
Setting
(3) A

= Sg(JX

) and B

= Sg(KY

) for every <

1
,
we see from conditions (ii) and (iii) that, for each <
1
, we have
(A

) B
+1
and
1
(B

) A
+1
.
From this it easily follows that
(4) For each limit ordinal <
1
, the (appropriate restriction of the) mapping is an
isomorphism of A

onto B

.
Set
E =
_
<
1
: is a limit ordinal and

(
1
Q) Y

=
_
<

( Q)

V
_
and
E

=
_
E : = sup( E) = sup E : <
_
.
(5) E and E

are closed, unbounded sets.

To see that E is unbounded, observe, rst of all, that, by conditions (v) and
(vii), we have
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS13
(6)

(
1
Q) Y

_
<

( Q)

V for every limit ordinal <

1
.
Therefore, we need only establish the reverse inclusion for unboundedly many limit
ordinals. Since a v in

(
1
Q) V must have a countable range, there is an
ordinal
v
<
1
such that v is in

(
v
Q). Let
0
be an arbitrary countable
ordinal. Since Y

0
is countable, the supremum of the
v
, over all v in Y

0
, is
countable. Hence, we can nd a countable
1
>
0
such that

(
1
Q) Y

0


(
1
Q) .
Continuing in this fashion, we obtain a strictly increasing sequence
n
: n < ) of
countable ordinals such that

(
1
Q) Y

n


(
n+1
Q) .
Set = sup
n
: n < . Then

(
1
Q) Y

_
<

( Q)

V .
In view of (6), we see that equality actually holds in the preceding line. This shows
that E is unbounded. The rest of the proof of (5) is easy, so we leave it to the
reader.
(7) For every in E

and every in E , the type tp(g

, B

) splits over B

.
To prove (7), recall that g

is not in

( Q) for any < . We thus see, from
the denition of E, that g

is not in Y

. Hence, by Fact 4, g

is not in B

.
The function g

is dened in terms of a strictly increasing, ordinal-valued function

f

whose supremum is . Since < , there must be an n < such that f

(n) > .
Let h
1
and h
2
be the extensions of g

n to n + 1 determined by
h
1
(n) = g

(n) = f

(n) , 0 ) and h
2
(n) = f

(n) , 1 ) .
Notice that h
1
= g

(n + 1). Clearly, h
1
and h
2
are in Y

, by conditions (v) and

(vii). Because is in E, and f

(n) > , we can apply the denition of E to conclude

that h
1
and h
2
are not in Y

, and hence are not initial segments of any elements in

Y

. Therefore, they are not in B

, by Fact 4. They realize the same type over B

,
by tree indiscernibility. Moreover, N [=
n+1
[h
1
, g

] and N [=
n+1
[h
2
, g

], since

n+1
codes the initial segment relation between elements in P
V
n+1
and P
V

. Thus,
both
n+1
(h
1
, x) and
n+1
(h
2
, x) are in tp(g

, B

). This completes the proof of

(7).
We now work towards a contradiction to (7).
(8) For every in E

and every a in Sg(JU) A

, there is an in E such that

tp(a, A

) does not split over A

.
To see that (8) contradicts (7), take any in E

, and set a =
1
(g

). Since g

is in Sg(KV ) B

, we get that a is in Sg(JU) A

, by (4). Hence, there is an

in E such that tp(a, A

) does not split over A

, by (8). But then, applying

to a, tp(a, A

), and A

, we get that tp(g

, B

) does not split over B

, by (4). This
contradicts (7).
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


14 STEVEN GIVANT AND SAHARON SHELAH
To prove (8), x a in E

and an a in Sg(JU) A

. Then there is a sequence

u = u
0
, . . . , u
r1
) from U, function symbols f
0
, . . . f
n1
in J, and a term from
L, such that (adding dummy variables)
(9) a =
_
f
0
[ u], . . . , f
n1
[ u], u

.
Each u
i
has the form u
i
=
i
, q
i
) for a unique
i
and q
i
in Q. The set

i
=
_
: Q X

and
i

_
is nonempty, since it contains , by condition (i). We shall denote the smallest
element of this set by

i
. Notice that u
i
X

i
=
i
, by condition (iv). Now,
by denition,

i
Q X

for each i < r. Since is in E

, there is, for each

i < r, a in E such that

i
Q X

. Take to be the maximum of these

. Then is in E and

i
Q X

for each i < r.

To verify that tp(a, A

) does not split over A

, let (x, y) be a formula of L, and

let

b and c be sequences of equal length from A

, such that (x,

b) and (x, c) are

in tp(a, A

), i.e.,
(10) M[= [a,

b] [a, c] .
We must prove that
(11) tp(

b, A

) ,= tp( c, A

) .
Since

b and c come from A

, there is a nite sequence v from X

that generates

b
and c with the help of some elements of J. Using the ordinals

i
, we shall construct
a sequence u

= u

0
, . . . , u

r1
) in X

such that, for each i, we have u

i
= u
i
i u
i
is
in X

, and
(12) u

i
< u

k
i u
i
< u
k
and u

i
= u

k
i u
i
= u
k
,
u

i
< v
j
i u
i
< v
j
and u

i
= v
j
i u
i
= v
j
.
for all appropriate j, k.
The construction of u

is not dicult: we shall set u

i
=

i
, q

i
), where q

i
is
chosen from Q. Since

i
Q X

, this assures that u

i
will be in X

. If
i
is
in
i
, then

i
=
i
. In this case, take q

i
= q
i
, so that u

i
= u
i
. Suppose, now, that
v
j
=
j
, r
j
). If u
i
< v
j
, then we have either
i
<
j
, or else
i
=
j
and q
i
< r
j
.
In the rst case, observe that
j

i
, since v
j
is in X

(here we are using again

condition (iv)). Therefore,

i

j
. If

i
<
j
, then we may choose q

i
however we
wish. If

i
=
j
, then we must choose q

i
< r
j
. This is possible since

i
Q X

.
In the second case, when
i
=
j
, we have
i
in X

, so u

i
= u
i
. In both cases we
obtain u

i
< v
j
. The other cases in (12) are treated similarly. Notice, however, that
we must choose q

i
so that (12) holds for all appropriate j and k at once. To do
this, we use the density of Q.
From (12) we conclude that u

v and u v are atomically equivalent in U. Because

U is a set of order indiscernibles in M

, it follows that u

v and u v are elementarily

equivalent in M

. Recalling (9), set a

=
_
f
0
[ u

], . . . , f
n1
[ u

], u

, and notice that

a

is in A

, since u

is from X

.
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS15
Since u

generates a

in the same way that u generates a, and since v generates

b and c, we conclude that a

b c and a

b c are elementarily equivalent in M

. In
particular, from (10) we get that
M[= [a

b] [a

, c] .
Thus, (a

, y) is in tp(

b, A

), while (a

, y) is in tp( c, A

). This proves (11), and

hence (8).
As is well-known, the fact that T

is superstable (and hence stable) implies that

order indiscernibles are totally indiscernible. Thus, if M

= EM(U, ), then any

two one-one sequences in
n
U satisfy the same formulas in M(but not necessarily in
M

). It follows that Lemma 1 (with K = ) goes through without any references

to order. Thus, if M

and N

are as in Lemma 1, then any injection of V into

U extends canonically to an elementary embedding of Sg
N
(V ) into Sg
M
(U), and
this extension is an isomorphism i is onto. In particular, the isomorphism
type of a subalgebra (of some very large model) generated by an innite set of
indiscernibles with respect to is uniquely determined by the cardinality of the
set of indiscernibles. To give a precise denition of such algebras, we x a very
large dense ordering Z (as large as we will need for any argument in this paper),
set C = EM(Z, ), and let Z

: an innite cardinal [Z[) be an increasing

sequence of subsets of Z such that [Z

[ = .
Denition 7. Let F

= Sg
C
(Z

). The set Z

is referred to as a -basis of F

The algebras F

, for , have many properties in common with free algebras.

For example, if U and V are -bases of F

and F

respectively, and if maps V one-

one into U then the canonical extension of maps F

elementarily into F

. This is
just a reformulation of what was said above. For another example, notice that F

is
strictly
-generated, by Lemma 2. It follows that any innite -basis of F

must have
cardinality ; otherwise we would have F

= F

for some ,= , forcing the strictly

-generated model F

to be -generated and visa-versa.

For the case when T

is superstable, we readily extend Lemma 1(v) to cover

a
:
Lemma 8. Let U, V be dense orderings, M

= EM(U, ), and K L

L. If
V U, then
Sg
M
(KV )
a
Sg
M
(KU) .
The lemma continues to hold if we replace M by M
K
.
Proof. We begin with the case when U is
1
-homogeneous. Let p be a strong type
of M based on a nite subset E of Sg(KV ), and suppose that p

E is realized in
Sg(KU) by

b. Let v = v
0
, . . . , v
m1
) be a nite sequence in V that generates E
in M, with the help of nitely many f K, and let u = u
0
, . . . , u
n1
) be a nite
sequence from U that generates

b. Because of the density of V, we can certainly nd
a sequence u

= u

0
, . . . , u

n1
) in Vsuch that v u and v u

are atomically equivalent

in U. Using again the density of V, is not dicult to construct an -sequence w in
V such that v w u and v w u

are atomically equivalent and the range, X, of v w is

dense. Assume, for the moment, that this done. Since U is
1
-homogeneous, there
is an automorphism of U taking v w u to v w u

. Extend it to an automorphism
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


16 STEVEN GIVANT AND SAHARON SHELAH
of M

. Then also induces an automorphism of Sg(K U), and it is the identity

mapping on Sg(KX). Because X is dense, Sg(KX) is an elementary submodel
of Sg(K V ), and therefore p is stationary over this submodel. Since xes an
elementary submodel over which p is stationary, we see that must map p

E to
itself. Thus, (

b) realizes p

E in Sg(K U). But Sg(K V ) is an elementary

substructure of Sg(K U), by Lemma 1(v), and all the formulas of p

E are over
Sg(KV ). Thus, (

b) must realize p

E in Sg(KV ).
Now suppose that U is arbitrary. Let W be an
1
-homogeneous extension of U,
and set N

= EM(W, ). Then by the previous case,

Sg
M
(KV ) = Sg
N
(KV )
a
Sg
N
(KW)
and
Sg
M
(KU) = Sg
N
(KU)
a
Sg
N
(KW) .
Therefore, Sg
M
(KV )
a
Sg
M
(KU).
Returning to w, a simple example should suce to indicate how it is to be
constructed. Suppose, for example, that
v
i
< u
0
< u
1
< u
2
< u
3
< v
j
,
and that no other elements of the sequences v and u are in the interval between v
i
and v
j
. Of course,
v
i
< u

0
< u

1
< u

2
< u

3
< v
j
.
If u

0
u
0
, or else if u

3
u
3
, then include in the range of w a dense set between
v
i
and u

0
, or else between u

3
and v
j
, respectively. Now consider the case when
u
0
< u

0
< u

3
< u
3
. If u

1
u
1
, or else if u

2
u
2
then include in the range of
w a dense set between u

0
and u

1
, or else between u

2
and u

3
, respectively. There
remains the case when u
1
< u

1
< u

2
< u
2
. In this case, include in the range of w
a dense set between u

1
and u

2
.
The next lemma is the analogue of Lemma 3 for the notion
a
.
Lemma 9. Let U be a dense ordering of power , and M

= EM(U, ). Then for

every countable H L

L there is a countable K with H K L

L such that
Sg
M
(KU)
a
M .
Proof. The proof is quite similar to the proof of Lemma 3. Suppose, for contradic-
tion, that H is a counterexample to the assertion of the lemma. Analogously to the
proof of Lemma 3, we construct an increasing, continuous sequence H

: <
1
)
of countable subsets of L

L such that
(1) Sg(H

U) Sg(H
+1
U) M, but Sg(H

U) ,
a
Sg(H
+1
U) .
Indeed, by Lemma 4 we can take H
0
to be a countable extension of H such that
Sg(H
0
U) M. Suppose, now, that H

has been dened, and that Sg(H

U)
M. By the assumption on H, Sg(H

U) is not a-saturated in M. Thus, there is

a strong type p

based on a nite subset E

of Sg(H

U) , such that p

is
realized in Msay by a

but p

is not realized in Sg(H

U). Since a

is
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS17
generated by U in M

, there is a nite set F L

L such that F U generates

a
x
i in M. Take H
+1
to be a countable extension of H

F in L

L such that
Sg(H
+1
U) M. Then Sg(H

U) Sg(H
+1
U) . By construction, a

realizes
p

in Sg(H
+1
U), so Sg(H

U) ,
a
Sg(H
+1
U). This completes the
verication of (1).
Set G =

<
1
H

and B =

<
1
Sg(H

U) = Sg(GU).
(2) For every <
1
and every countably innite X U we have Sg(H

X) ,
a
B .
Indeed, let W be a countable, dense subset of U containing X and containing
nite subsets U
0
and U
1
of U such that H

U
0
generates E

and GU
1
generates
a

in M. Then, by Lemma 1(v),

(3) Sg(GW) Sg(GU) = B .
Therefore, to prove (2) it suces to show that Sg(H

X) ,
a
Sg(GW).
Since p

is realized by a

in Sg(G U), by construction, we see from (3)

and the denition of W that it must also be realized by a

in Sg(GW). Also,
(4) Sg(H

W) Sg(H

U) ,
by Lemma 1(v). Since p

is not realized in Sg(H

U), by construction, we
see from (4) that it is not realized in Sg(H

W).
We now transfer this situation to a strong type over Sg(H

X). Because W is
countable and dense, and X is innite, there is an automorphism of W taking U
0
into X (this is the point of introducing W to replace U). Extend it to an automor-
phism of N

= Sg
M

(W). Then appropriate restrictions of are automorphisms

of Sg(GW) and Sg(H

W). Therefore,
(5) (p

(E

) is realized in Sg(GW) (by ( a

)), but not in Sg(H

W) .
Suppose, now, for contradiction, that Sg(H

X)
a
Sg(GW). Then
(6) (p

(E

) is realized in Sg(H

X) .
On the other hand, this supposition also gives
(7) Sg(H

X) Sg(GW) .
Since Sg(H

U) Sg(GU), by construction, we get from (3) and (4) that

(8) Sg(H

W) Sg(GW) .
Combining (7) and (8), we arrive at Sg(H

X) Sg(H

W). In view of (6),

this forces (p

(E

) to be realized in Sg(H

W), which contradicts (5). This

proves (2)
We now construct a <
1
and a countable X U that contradict (2). Since B
is strictly -generated, by Lemma 2, it is isomorphic to F

. Thus, there is a set V of

generators of B, of size , that is totally indiscernible in B. Exactly as in the proof
of Lemma 3, we construct increasing sequences, X
n
: n ) and Y
n
: n ), of
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


18 STEVEN GIVANT AND SAHARON SHELAH
countable subsets of U and V respectively, and an increasing sequence
n
: n < )
of countable ordinals, such that
Sg(H

n
X
n
) Sg(Y
n
) Sg(H

n+1
X
n+1
)
for every n < . Set
= sup
n
: n < , X =
_
n
X
n
, and Y =
_
n<
Y
n
.
Then
Sg(H

X) =

n<
Sg(H

n
X
n
) =

n<
Sg(Y
n
) = Sg(Y ) .
Since Sg(Y )
a
B by Lemma 8 (with K = ) and the total indiscernibility of Y
and V , we conclude that Sg
M
(H

X)
a
B. This is just the desired contradiction
to (2).
Lemma 10. For every countable H L

L, there is a countable K, with H

K L

L, such that, whenever V is a dense ordering and N

= EM(V, ), we
have
Sg
N
(KV )
a
N .
Proof. The proof is almost identical to the proof of the second part of Lemma 4,
with replaced everywhere by
a
. One uses Lemma 9 in place of the rst
part of the proof of Lemma 4, and Lemma 8 in place of Lemma 1(v) and Facts 1
and 3. Notice, for example, that, taking K = L

L, we get essentially Fact 3 with

replaced by
a
. We leave the details to the reader.
Lemma 11. F

is a-saturated for every .

Proof. By Shelah [1987], the proof of Theorem 2.1 on pp. 285287, T has an
Ehrenfeucht-Mostowski model N

= EM(W, ) of power > max(2

|T|
)
+
, such
that N = N

L is a-saturated. By Lemma 10 there is a countable K L

L
such that Sg
N
(K W)
a
N. Let U be a dense submodel of W of power .
Then Sg
N
(K U)
a
Sg
N
(K W), by Lemma 8. Thus, we immediately see that
Sg
N
(KU), too, is a-saturated. Since it is strictly -generated, by Lemma 2, it is
isomorphic to F

, by -uniqueness. Thus, F

is a-saturated.
Next, we turn to F

. Recall that F

has a -basis Z

that extends a -basis Z

of F

. Hence, F

and even F

a
F

, by Lemmas 1(v) and 8 (with K = )

and the total indiscernibility of Z

and Z

. Since F

is a-saturated, by the previous

lemma, so is F

.
Now consider any > . To show that F

is a-saturated, consider an arbitrary

strong typed p based on a nite subset E of F

. Let V a denumerably innite

subset of the -basis Z

of F

that contains a set of generators for E. Since Sg(V )

is isomorphic to F

, and the latter is a-saturated, p

E is realized in Sg(V ). But

Sg(V ) F

, so p

E is also realized in F

.
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS19
Lemma 12. There is a cardinal such that every model of T of power > is
a-saturated.
Proof. Let be the Hanf number for omitting types in languages of cardinality
at most [T[. Suppose, for contradiction, that there is a model A of T of power
> max, that is not a-saturated. Then there is a strong type p = p

E (over
A) based on a nite subset E of A such that p is omitted by A. Since there are at
most [T[ many inequivalent formulas in p, by Shelah [1990], Chapter III, Lemma
2.2(2), we may assume that p is a (possibly incomplete) typeas opposed to a
strong typeover some subset B of A of cardinality at most [T[.
Let L

be a language with built-in Skolem functions that extends L and that

includes constants for the elements of B. Let T

be a Skolem theory in L

that
extends the theory of A, b )
bB
. By Morleys Omitting Types Theorem and the
choice of , there is an Ehrenfeucht-Mostowski model N

of T

over a dense ordering

U of indiscernibles of cardinality such that N

omits the type p (see Morley

[1965], Theorem 3.1, or Chang-Keisler [1973], Exercise 7.2.4). Let V be a dense
submodel of U of power , and set M

= Sg
N

(V ). Then M

is the corresponding
Ehrenfeucht-Mostowski model over V, and M

, by Fact 3. Obviously, M

,
and hence also M, omits p.
Let H L

L be a nite set containing all the individual constant symbols for

elements of E. By Lemma 4, there is a countable K with H K L

L such
that Sg(K V ) M. Thus, E is a subset of the universe of Sg(K V ). Since p,
as a strong type, is based on a subset of the model Sg(KV ), every formula of
p is equivalent to a formula

with parameters from Sg(K V ), by Shelah ibid.,

Lemma 2.15(1). Let p

be the set of these formulas

. Then p

is a strong type
based on E that is equivalent to p. Because M omits p, it also omits p

. Therefore,
the elementary substructure Sg(K V ) must omit p

. This shows that Sg(K V )

is not a-saturated. But that is impossible: since Sg(K V ) is strictly -generated
(by Lemma 2), it is isomorphic to F

, and we have seen that F

is a-saturated.
Lemma 13. F

is a-prime over any -basis, for . Consequently, F

is also
a-prime over .
Proof. Let U be a -basis of F

. Then U generates any given sequence a from F

,
so tp( a, U) is clearly atomic. In particular, F

is a-atomic over U. Moreover, F

is also a-saturated, by Lemma 11, and there are obviously no Morley sequences in
F

over U. By Shelahs second characterization theorem for a-prime models, F

is
a-prime over U (see Shelah [1990], Chapter IV, Denition 4.3 and Theorem 4.18).
Now an a-prime model over a countable set is also a-prime over (see ibid.).
Hence, F

is also a-prime over .

We now work inside of some monster model. Let p
0
, . . . , p
n1
be pairwise or-
thogonal regular types, all based, say, on a nite set E. By realizing the strong type
of E over in Pr
a
(), and then passing to automorphic copies of p
0
, . . . , p
n1
,
we may assume that E is a subset of the universe of Pr
a
(). For each i < n, let I
i
be an innite Morley sequence built from p
i

E.
Lemma 14. Pr
a
(E

i<n
I
i
) is strictly -generated, where = [

i<n
I
i
[.
Proof. The proof is by induction on . If = , then, as mentioned in the preceding
lemma, Pr
a
(E

i<n
I
i
) is a-prime over . But, we just saw that F

is a-prime
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


20 STEVEN GIVANT AND SAHARON SHELAH
over . Hence, these two models are isomorphic, by the uniqueness of a-prime
models. Since F

is strictly -generated, by Lemma 2, so is Pr

a
(E

i<n
I
i
) .
Now suppose that the lemma is true for all innite < . Represent

i<n
I
i
as a
strictly increasing sequence a

: < ) with the property that, for each i < n, the

sequence contains innitely many members of I
i
. For each innite < , set B

=
Pr
a
(E a

: < ). We may arrange this so that B

whenever ,
and, at limit stages , that B

=

<
B

. By the induction hypothesis, each B

is [[-generated, since [[ < . Therefore, Pr

a
(E

i<n
I
i
) =

<
B

is
generated by a set of cardinality at most = .
To see that no set of smaller cardinality can generate it, we proceed by contra-
diction. Let X be an innite set of power < that generates Pr
a
(E

i<n
I
i
) .
Since Pr
a
(E

i<n
I
i
) is a-atomic over E

i<n
I
i
, for each nite sequence

b from
X there is a nite subset I
b


i<n
I
i
such that
stp(

b, E I
b
) stp(

b, E

i<n
I
i
) .
Set J =

I
b
:

b is a nite sequence from X. Then J has power at most and
(1) stp(

b, E J) stp(

b, E

i<n
I
i
) for every (nite) sequence

b from X .
Since < , there is a sequence a = a
0
, . . . , a
k1
) that is in

i<n
I
i
, but not in
J. Thus, a is independent over E J. This is readily seen to contradict (1). For
example, because X generates Pr
a
(E

i<n
I
i
) , there is a sequence

b from X and
terms
i
( x) of L, for i < k, such that
i
[

b] = a
i
. Now
stp(

b, E J)

i<k

i
( x) = a
i
,
by (1), so there is a formula ( x) in L

(E J) (the set of formulas almost over

E J) such that
( x)

i<k

i
( x) = a
i
.
Thus, a is denable, over L

(E J), since, e.g.,

y = a
i
x(( x)
i
( x) = y) for each i < k .
But then a is in the algebraic closure of E J, so it cannot be independent over
this set. We have reached our contradiction.
Theorem 15. T is categorical in every cardinality > [T[.
Proof. We begin by proving that T must be unidimensional. Indeed, suppose for
contradiction that there are two orthogonal regular types, p
1
and p
2
, over a nite
subset E of Pr
a
(). Working inside a monster model, for i = 1, 2, let I
i
be a Morley
sequence of cardinality built from p
i

E. Then Pr
a
(E I
1
) and Pr
a
(E I
1
I
2
)
are both strictly -generated models, by the previous lemma, and hence isomorphic,
by -uniqueness. But this is impossible: the rst model has a single dimension of
cardinality , the p
1
-dimension, and all the other dimensions are ; the second
model has exactly two dimensions of cardinality , the p
1
- and the p
2
-dimensions.
Thus, T can only have one regular type, up to equivalence. Lets denote this type
by p.
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


UNIVERSAL THEORIES CATEGORICAL IN POWER AND -GENERATED MODELS21
Now let ( [T[) be so large that all the models of T of power > are a-
saturated (see Lemma 12). Then T is
+
-categorical. Indeed, if M and N are
models of power
+
, then they both are a-saturated, by choice of , and have p-
dimension
+
, by unidimensionality. Hence, they must be isomorphic, by Shelah
[1990], Chapter V, Theorem 2.10. This shows that T is categorical in some power
> [T[. By the results of Shelah [1974], it is categorical in every power > [T[.
Theorem 16. T is -unique for every uncountable .
Proof. If T is denitionally equivalent to a countable theory, and is categorical in
every innite power, i.e., if T is a totally categorical theory, then the theorem is
obvious. Suppose, now, that it is not totally categorical. We shall use Theorems
13, including the proof of Theorem 2, from Laskowski [1988]. According to these,
T

must have a minimal prime model M

0
. Moreover, there is a type p based on
M
0
with the following properties:
(1) If N is any elementary extension of M
0
, then any two maximal Morley sequences in
N built from pM
0
must have the same cardinality. This is called the p-dimension
of N.
(2) If N and N

are elementary extensions of M

0
, with the same p-dimension, then
there is an isomorphism of N onto N

that xes M
0
.
(3) If M
0
N N

, then pN is realized in N

.
To simplify notation, we shall assume that p is a 1-type.
Since T

is a universal-existential theory categorical in power, it is model com-

plete, by Lindstr oms theorem (see, e.g., Chang-Keisler [1973], Theorem 3.1.12).
Fix > , and let Mbe a strictly -generated model of T, say X is a generating
set of power . Our goal is to prove that M has p-dimension . Let W be a count-
ably innite subset of X. Then Sg(W) M, by model completeness. Without
loss of generality, we may suppose that M
0
is an elementary submodel of Sg(W).
We now dene a strictly increasing, continuous sequence, Y

: < ), of subsets
of X of size < , such that
(4) M
0
Sg(Y

) Sg(Y
+1
) M for < .
Indeed, take Y
0
= W. If Y

has been dened, then Sg(Y

) ,= M, since M is strictly
-generated and Y

has power < . Hence, there is an element u

in X that is not
generated by Y

. We set Y
+1
= Y

. Property (4) follows by our choice of

Y
0
and u

, for each , and by model completeness.

For each < , the type p Sg(Y

) is realized in Sg(Y
+1
), by (3) and (4).
Therefore, using our strictly increasing chain Sg(Y

) : < ) of elementary sub-

structures of M, we can build from pM
0
a Morley sequence in Mof length at least
. Thus, M has p-dimension at least .
Suppose now that I is a maximal Morley sequence in M built from pM
0
. Then
Sg(M
0
I) = M. For otherwise we would have Sg(M
0
I) M, by model
completeness. Hence, we could realize pSg(M
0
I) in M, by (3), and thus extend
I to a larger Morley sequence in M built from pM
0
, contradicting its maximality.
Since X also generates M, and has cardinality > , there is a subset J of I of
cardinality such that M
0
J generates X, and hence also M. In particular,
M
0
J generates I. But then J = I, since any element in IJ would be independent
4
0
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
9


22 STEVEN GIVANT AND SAHARON SHELAH
over M
0
J, and hence could not be generated by this set. We conclude that [I[ .
In other words, M has p-dimension at most , and hence exactly .
We have shown:
(6) Any strictly -generated model of T extending M
0
has p-dimension .
Now let M and N be any two strictly -generated models of T. By passing to
isomorphic copies, we may assume that M
0
is an elementary substructure of each.
Hence, by (2) and (6), M and N are isomorphic over M
0
. This shows that T is
-unique.
By the preceding theorem, the noncountably generated models of T are, up to
isomorphisms, precisely the structures F

, for > .
Bibliography
[1973] Baldwin, J.T. and Lachlan, A.H., On universal Horn classes categorical in some innite
power, Algebra Universals 3 (1973), 98111.
[1973] Abakumov, A.I., Palyutin, E.A., Shishmarev, Yu. E., and Taitslin, M.A., Categorical
quasivarieties, Algebra and Logic 11 (1973), 121.
[1973] Chang, C.C. and Keisler, H.J., Model theory, Studies in Logic and the Foundations of
Mathematics 73, North-Holland Publishing Company, Amsterdam, 1973, xii + 550 pp.
[1979] Givant, S.R., Universal Horn classes categorical or free in power, Annals of Mathematical
Logic 15 (1979), 153.
[1988] Laskowski, M.C., Uncountable theories that are categorical in a higher power, Journal of
Symbolic Logic 53 (1988), 512530.
[1965] Morley, M., Omitting classes of elements, in The theory of models, Proceedings of the
1963 International Symposium at Berkeley (J.W. Addison et al., eds.), Studies in Logic,
North-Holland Publishing Company, Amsterdam, 1965, pp. 265274.
[1974] Shelah, S., Categoricity of uncountable theories, in Proceedings of the Tarski Symposium
(L. Henkin et al., eds.), Proceedings of Symposia in Pure Mathematics XXV, American
Mathematical Society, Providence, R.I., 1974, pp. 187203.
[1978] , Classication Theory and the number of non-isomorphic models, Studies in Logic
and the Foundations of Mathematics 92, North-Holland Publishing Company, Amster-
dam, 1978, xvi + 544 pp.
[1987] , Existence of many L
,
-equivalent non-isomorphic models of power . Proceed-
ings of (Trento) Model Theory Conference, June, 1986, Annals of Pure and Applied Logic
34 (1987), 291310.
[1987a] , On the number of strongly

-saturated models of power ., Annals of Pure and

Applied Logic 36 (1987), 279287.
[1990] , Classication Theory and the number of non-isomorphic models, Studies in Logic
and the Foundations of Mathematics 92, second edition, North-Holland Publishing Com-
pany, Amsterdam, 1990, xxxiv + 705 pp.
Mills College
Hebrew University, Rutgers University, and MSRI