Академический Документы
Профессиональный Документы
Культура Документы
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Categoricity over P for rst order T or
categoricity for L
can stop at
k
while
holding for
0
, . . . ,
k1
Saharon Shelah
The Hebrew University
Department of Mathematics
Jerusalem, Israel
Bradd Hart
University of California at Berkeley
Department of Mathematics
Berkeley, California
October 6, 2003
In the 1950s, Los conjectured that if T was countable rst order theory
in a language L then if it was categorical in some uncountable power then it
was categorical in all uncountable powers. In [7], Morley proved this. Buoyed
by this success, more general forms of the Los conjecture were considered.
In [10], Shelah showed that if T was any rst order theory categorical
in some power greater than [T[ then T was categorical in all powers greater
than [T[. Keisler took up the investigation of the L
sentence which is
n
-
categorical for all n then is excellent.
The question which naturally arises is, under suitable set theoretic as-
sumptions, is categoricity in
n
for n < k sucient to prove full categoricity
for a sentence in L
.
The answer to this question must wait while we introduce another variant
of the Los conjecture.
Suppose L is a relational language and P L is a unary predicate. If M
is an L-structure then P(M) is the L-structure formed as the substructure
of M with domain a : M [= P(a). Now suppose T is a complete rst order
theory in L with innite models. Following Hodges, we dene
Denition 0.1 T is relatively -categorical if whenever M,N [= T, P(M) =
P(N), [P(M)[ = then there is an isomorphism i : M N which is the
identity on P(M).
T is relatively categorical if it is relatively -categorical for every .
The notion of relative categoricity has been investigated by Gaifman ([1]),
Hodges ([3] and [4]), Pillay ([8]) and Pillay and Shelah ([9]). In ([13]), Shelah
gave a classication under some set theory.
Again the question arises whether the relative -categoricity of T for some
> [T[ implies that T is relatively categorical.
In this paper, we provide an example, for every k > 0, of a theory T
k
and
an L
sentence
k
so that T
k
is relatively
n
-categorical for n < k and
k
is
n
-categorical for n < k but T
k
is not relatively
k
-categorical and
k
is
not
k
-categorical.
The examples are due to Shelah. Harrington asked about the possibility
of such examples in Chicago in December, 1985 as he was not happy with the
complexity of the classication. The examples provided L
sentences which
were categorical but not excellent and so a proof of this fact was written up
in [2].
The notation used is standard. [A]
k
will stand for all the k-element subsets
of the set A. T
is used
2
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
to represent the direct sum of groups and
Z
2
(written as 2
).
1 The Example
We rst describe the example informally. Fix a natural number k greater
than one. There will be an innite set I with K = [I]
k
. There are constants
c
n
for n and a predicate R containing all of them. R will be thought
of as levels and we will refer to constants in R as standard levels. We x
Z
2
, the abelian group, G, the direct sum of K-many copies of Z
2
and H, the
direct sum of R-many copies of Z
2
. In addition, all relevant projections onto
Z
2
are available to us. All of this constitutes the P-part of the model.
Outside of this we have two types of objects. First, for every level r R
and every u K, we have a distinct copy of G. Via some connection between
our xed copy of G and this one we will be able to determine the sum of
any three elements of G but we will have lost the zero. Second, for every
u K there will be a distinct copy of H in which we again have lost the
zero.
We will be interested in the possibility of choosing elements from these
copies of G and H to act as the zero in their respective groups. We wont
put any more restraints on Gs from non-standard levels so any element will
do. However, for each n , on the level corresponding to c
n
, and for every
u K, there will be a predicate connecting the copy of H corresponding to
u and k of the copies of G on the n
th
level. It will be these predicates which
make or break the categoricity by putting restraints on choices for the zeroes
of the copies of G and H.
We now wish to x k for the rest of the paper.
Convention 1.1 k will be a xed natural number greater than one.
Now, more formally, we dene the language for the example.
Denition 1.2 L will be the language that consists of
1. unary predicates I, K, R, P, G
a
, H
a
3
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
2. binary predicates , H
b
3. ternary predicates ,, + and G
b
4. a 4-ary predicate h
5. a 5-ary predicate g
6. a k + 1-ary predicate Q
l
for every l < and
7. constants c
a
for every a Z
2
We now describe the standard model on I.
Denition 1.3 If I is an innite set then the standard model on I denoted
by M
I
is the L-structure with universe
I [I]
k
Z
2
[I]
k
Z
2
Z
2
[I]
k
[I]
k
Z
2
[I]
k
Z
2
and where the symbols of L are interpreted as follows:
1. I is interpreted as I, K as [I]
k
, R as , G
a
as
[I]
k
Z
2
and H
a
as
Z
2
2. the constants c
a
are interpreted as a. That is, for example,R(c
a
) holds
for every a .
3. P(x) holds i x is a constant or one of I(x),K(x),G
a
(x) or H
a
(x) holds.
4. G
b
(l, u, x) holds i R(l), K(u) and x = (l, u, y) for some y
[I]
k
Z
2
5. H
b
(u, x) holds i K(u) and x = (u, y) for some y
Z
2
6. (x, y) holds i I(x), K(y) and x y
7. +(x, y, z) holds i x, y and z are all in one of Z
2
,
[I]
k
Z
2
or
Z
2
and
x +y = z.
8. (u, x, a) holds i K(u), G
a
(x) and x(u) = a, an element of Z
2
.
4
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
9. (l, x, a) holds i R(l), H
a
(x) and x(l) = a, an element of Z
2
.
10. g(l, u, x, y, z) holds i R(l), K(u), G
a
(x), y = (l, u, a), z = (l, u, b) (so
G
b
(l, u, y) and G
b
(l, u, z)) and b = a +x.
11. h(u, x, y, z) holds i K(u), H
a
(x), y = (u, a), z = (u, b) (so H
b
(u, y)
and H
b
(u, z)) and b = a +x.
12. Q
l
(x
0
, . . . , x
k
) holds i x
i
= (c
l
, u
i
, y
i
) with G
b
(c
l
, u
i
, x
i
) for i < k and
x
k
= (u
k
, z) with H
b
(u
k
, x
k
) where u
0
, . . . , u
k
are all the k-element sub-
sets of some (k + 1)-element subset of I and
i<k
y
i
(u
k
) = z(c
l
)
Remarks:
1. In the previous denition, all of the direct sums used in the denition
of the universe represent abelian groups. Hence on the right hand side
of items 7, 10, 11 and 12, the addition mentioned is addition in the
appropriate group.
2. In item 12, each y
i
is in
[I]
k
Z
2
and u
k
is in [I]
k
so y
i
(u
k
) is in Z
2
. z is
in
Z
2
and c
l
so z(c
l
) is in Z
2
. Hence, the displayed equality is
comparing elements of Z
2
.
Lets consider some of the sentences in L that the standard model satises.
For a xed innite set I, M
I
satises:
1. I is an innite set, K is the collection of k-element subsets of I and
is the membership relation between elements of I and elements of K.
2. I, K, R, G
a
, H
a
are disjoint and their union together with the constants
c
a
for a Z
2
form P.
3. R(c
a
) for every a .
4. G
b
(l, u, x) implies R(l) and K(u) and H
b
(u, x) implies K(u).
5
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
5. If x is not in P then either for some l and u, G
b
(l, u, x) or for some u,
H
b
(u, x) and for every l R and u, v K, P, H
b
(u, ) and G
b
(l, v, )
are pairwise disjoint.
6. If (u, a, z) then K(u), G
a
(a) and z is one of the constants indexed by
Z
2
.
7. If (l, b, z) then R(l), H
a
(b) and z is one of the constants indexed by
Z
2
.
8. If g(l, u, a, v, w) then R(l), K(u), G
a
(a), G
b
(l, u, v) and G
b
(l, u, w).
9. If h(u, b, x, y) then K(u), H
a
(b), H
b
(u, x) and H
b
(u, y).
10. The constants c
a
for a Z
2
together with + have the group structure
of Z
2
.
11. + restricted to G
a
gives a subgroup of
K
Z
2
which contains
K
Z
2
where
the projections are given by .
12. + restricted to H
a
gives a subgroup of
R
Z
2
which contains
R
Z
2
where
the projections are given by .
13. For every l in R and u in K, G
b
(l, u, ) is non-empty and for every l
in R, u in K and x so that G
b
(l, u, x) g(l, u, , x, ) is a bijection from
G
a
onto G
b
(l, u, ). Moreover, g(l, u, x, y, z) implies g(l, u, x, z, y) and
if g(l, u, a, x, y) and g(l, u, b, y, z) then g(u, l, a + b, x, z) where a + b is
the unique c so that +(a, b, c),.
14. For every u in K, H
b
(u, ) is non-empty and for every u in K and x so
that H
b
(u, x), h(u, , x, ) is a bijection from H
a
onto H
b
(u, ). More-
over, h(u, x, y, z) implies h(u, x, z, y) and if h(u, a, x, y) and h(u, b, y, z)
then h(u, a +b, x, z) where a +b is the unique c so that +(a, b, c).
15. If Q
l
(x
0
, . . . , x
k
) then for i < k, for some u
i
in K, G
b
(c
l
, u
i
, x
i
) and for
some u
k
in K, H
b
(u
k
, x
k
). Additionally, u
0
, . . . , u
k
are all the k-element
subsets of some (k + 1)-element subset of I. If is a permutation of k
then Q
l
(x
(0)
, . . . , x
(k1)
, x
k
).
6
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
16. If Q
l
(x
0
, . . . , x
k
), G
b
(c
l
, u, x
0
), H
b
(v, x
k
), G
b
(c
l
, u, x
0
) and H
b
(v, x
k
)
then Q
l
(x
0
, . . . , x
k
) i the v-projection of the unique element a so that
g(c
l
, u, a, x
0
, x
0
) via is 0 and Q
l
(x
0
, . . . , x
k
) i the c
l
-projection of the
unique element a so that h(v, a, x
k
, x
k
) via is 0.
17. Suppose l , u is in K and i
0
, . . . , i
n1
are distinct elements of I not
in u. For each j < n, let v
j
i
for 1 i k be a list of the k-element
subsets of u i
j
besides u. If G
b
(c
l
, v
j
i
, x
j
i
) for each j < n and i < k
and H
b
(v
j
k
, y
j
) for every j < n then
x
j<n
Q
l
(x, x
j
1
, . . . , x
j
k1
, y
j
).
17 actually follows from the previous axioms but it is in the form that we
will use it in section 2. We make the following denition for the rest of the
paper.
Convention 1.4 Let T be the theory in L made up of the sentences enu-
merated 1 17 above.
The standard model satises some additional sentences in L
. For any
innite set I, M
I
satises:
1. R contains only the constants indexed by .
2. G
a
is canonically isomorphic to
K
Z
2
.
3. H
a
is canonically isomorphic to
Z
2
.
Convention 1.5 Let be the L
= R(M) c
l
: l .
Extend f
M
and f
N
to include R
K(M) (= R
.
If x M so that M [= H
b
(u, x) then there is a unique a so that
M [= h(u, a, f
M
(u), x).
There is a unique y N so that
N [= g(u, a, f
N
(u), y).
Extend j so that j(x) = y.
Using the fact that M and N satisfy T, it is not hard to show that j denes
a function from M onto N. We want to show that it is an isomorphism. Well
check the hardest predicate, Q
l
.
Suppose M [= Q
l
(x
0
, . . . , x
k
) where
M [= G
b
(c
l
, u
i
, x
i
) for i < k and M [= H
b
(u
k
, x
k
).
Choose a
i
for i k so that M [= g(c
l
, u
i
, a
i
, f
M
(l, u
i
), x
i
) for i < k and
M [= h(u
k
, a
k
, f
M
(u
k
), x
k
).
We know
M [= Q
l
(f
M
(l, u
0
), . . . , f
M
(u
k
))
since f
M
is a solution. Suppose
M [= (u
i
, a
i
, z
i
) for i < k and M [= (c
l
, a
k
, z
k
).
Then by using axioms 15 and 16 of T, we conclude that
i<k
z
i
= z
k
9
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
where the sum takes place in Z
2
and we identify the constants indexed by
Z
2
with the elements they represent.
Since P(M) = P(N), this happens in N as well and since N [= T, we
unravel the fact that f
N
is a solution so N [= Q
l
(f
N
(l, u
0
), . . . , f
N
(u
k
)) to
conclude that N [= Q
l
(y
0
, . . . , y
k
) where y
i
= j(x
i
) for i k.
A completely symmetric argument shows that if N [= Q
l
(j(x
0
), . . . , j(x
k
))
then M [= Q
l
(x
0
, . . . , x
k
) so j is an isomorphism. 2
Lemma 2.3 Suppose M [= T.
1. If M is countable then M has a solution.
2. If A B I(M), B is countable and f is an A-solution then f can
be extended to a B-solution.
Proof: The rst follows from second so we will prove the second.
Choose f
so that f f
and dom(f
) = dom(f) [B]
k
where if u , [A]
k
then M [= H
b
(u, f
(u) is arbitrary.
f
,
2. dom(f
n
) = dom(f
) l
i
, u
i
) : i < n and
3. f
n
is a solution on its domain.
If we accomplish this then
f
n
will provide a B-solution extending f.
Suppose we have dened f
n
. We need to choose an a so that M [=
G
b
(c
ln
, u
n
, a) and which will be compatible with the demands of being a
solution.
Say that a (k +1)-element subset v of B puts a constraint on u
n
if u
n
v
and k 1 of the k-element subsets of v, say w
1
, . . . , w
k1
, are such that
l
n
, w
i
) dom(f
n
) for i < k. Note that since u
n
, A, at least one of these
w
i
s must also not be a subset of A.
10
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Now since only nitely many elements are enumerated before l
n
, u
n
),
there are only nitely many (k + 1)-element subsets of B which put a con-
straint on u
n
. This is exactly the situation that axiom 17 of T was designed
for so we can nd an a so that f
n+1
= f
n
l
n
, u
n
), a) is a solution on its
domain. 2
Corollary 2.4 is a complete L
sentence.
Proof: To see this, it suces to see that if M and N are countable models
of then M
= N. But since M and N are models of , P(M) and P(N)
are uniquely determined by so we may assume that P(M) = P(N). By
lemma 2.3, M and N have solutions and hence by lemma 2.2, M
= N. 2
Denition 2.5 Suppose M [= T, A
I(M) and a
0
, . . . , a
m1
are distinct
elements of I(M) A
. A
s
, f
s
: s T
(m)) is a compatible
n
T
(m)-
system of solutions if
1.
sP
(m)
A
s
= A
a
0
, . . . , a
m1
, [A
[
n
and A
s
= A
a
t
: t s
for every s T
(m).
2. f
s
is a A
s
-solution for every s T
(m)
3. for every s, t T
(m) if s t then f
s
f
t
Using the notation from the denition, suppose A
s
, f
s
: s T
(m)) is a
compatible
0
T
sP
(m)
A
s
]
k
sP
(m)
[A
s
]
k
then a
0
, . . . , a
m1
u. Since m < k, there is b u a
0
, . . . , a
m1
u.
If c
sP
(m)
A
s
u then
(u b) c ,
sP
(m)
[A
s
]
k
.
Hence, if u v where v is any (k+1)-element subset of
(m)
A
s
then there
is a k-element subset u
v, u ,= u
so that u
(m)
A
s
as well. Using
this observation and a proof similar to the proof of lemma 2.3, we obtain
11
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Lemma 2.6 If A
s
, f
s
: s T
(m)) is a compatible
0
T
(m)-system
with m < k then there is
sP
(m)
A
s
-solution f so that f
s
f for every
s T
(m)
We use this as the base step in the following lemma
Lemma 2.7 If A
s
, f
s
: s T
(m)) is a compatible
n
T
(m)-system
with m+n < k then there is
sP
(m)
A
s
-solution f so that f
s
f for every
s T
(m)
Proof: We prove this by induction on n. If n = 0 then this is just lemma
2.6. Suppose n > 0 and A
s
= A
b
t
: t s. Enumerate A
, a
: <
n
)
and let A
= a
s
= A
b
t
: t s for every
s T
s
be the restriction of f
s
to an A
s
-solution. We wish to
dene g
is a
sP
(m)
A
s
-solution extending f
s
for every s T
(m)
2. g
<n
g
: <
n
) and let
A
= a
: < Let f
be the restriction of f to an A
-solution. By
induction, we dene A
b-solutions g
extending f
. If we have dened
g
f
+1
to a A
+1
b-
solution. At limits we take unions and
<n
g
is a B-solution extending f.
2
Theorem 2.9 If M [= T and [M[ <
k
then M has a solution.
12
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Proof: By induction on the cardinality of M. If M is countable then this
is lemma 2.3. If [M[ =
n
with n > 0 then we can choose N, N M with
[N[ <
n
. By induction, N has a solution and by using lemma 2.8 repeatedly,
we can extend it to a solution for M. 2
Corollary 2.10 1. T is relatively
n
-categorical for all n < k.
2. is
n
-categorical for all n < k.
Proof: 1. Suppose M and N are models of T, P(M) = P(N) and [P(M)[ =
n
for some n < k. It follows that [M[ = [N[ =
n
. By theorem 2.9, M and
N have solutions and so by lemma 2.2, M
= N.
2. Suppose M and N are models of and [M[ = [N[ =
n
for some
n < k. P(M) is uniquely determined by I(M) and P(N) is determined by
I(N). [M[ = [I(M)[ so we may assume that P(M) = P(N) and it follows
then that M
= N by theorem 2.9 and lemma 2.2. 2
3 The Failure of Full Categoricity
In this section, we show that is not fully categorical.
Suppose M [= and I = I(M). Without loss of generality, we may
assume that K(M) = [I]
k
, R(M) = , G
a
(M) =
K
Z
2
and H
a
=
Z
2
.
Further, we may assume that the constants c
l
= l for l and c
a
= a for
a Z
2
. , and + can also be assumed to be as in the standard model M
I
.
Lemma 3.1 If M, N [= , M N and N has a solution then M has a
solution.
Proof: Suppose that f is a solution for N. Fix some g : K(M) M so
that
M [= G
b
(l, u, g(l, u)) for every l and u K(M).
For u K(M), let c
l,u
be such that
N [= g(l, u, c
l,u
, g(l, u), f(l, u)).
Choose d
l,u
so that for every v K(M) and y Z
2
M [= (v, d
l,u
, y) i M [= (v, c
l,u
, y).
13
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Dene f
: K(M)K(M) M so that f
(l, u
i
)) for i < k
and by the choice of d
l,u
,
(c
l,u
i
+d
l,u
i
)(u
k
) = 0 for all i < k
hence M [= Q
l
(f(l, u
0
), . . . , f(u
k
)). 2
Lemma 3.2 If M [= and > [M[ then there is N [= so that [N[ =
and M N.
Proof: Let I(N) be the disjoint union of I(M) and . From our discussion
at the beginning of the section, this denes the P-part of N. P(M) will be
subset of P(N) except for G
a
(M). The small technical point here is that
we have identied G
a
(N) with
K(N)
Z
2
. We will identify x G
a
(M) with
x
G
a
(N) where x
K(N)\K(M)
Z
2
. If u K(M) and l then let G
b
(l, u, N) =
G
b
(l, u, M)J and identify x G
b
(l, u, M) with (x, 0) where 0 is the identity
in J. If u K(N)K(M), let G
b
(l, u, N) =
K(N)
Z
2
. We leave it to the reader
to dene a reasonable g.
It remains to dene Q
l
on N for each l . Fix an arbitrary function
f : K(M) M so that
M [= H
b
(u, f(u)) for all u K(M).
14
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
f is needed only in case 3 below. Suppose v is a k + 1-element subset of
I(N) and u
0
, . . . , u
k
are all the k-element subsets of v. Note that either
v I(M) or at most one of the u
i
s is a subset of I(M). Further suppose
x
i
G
b
(l, u
i
, N) for i < k and x
k
H
b
(u
k
, N). There are a number of cases:
1. u
i
K(M) for all i. Then x
i
= (x
i
, a
i
) for some x
i
G
b
(l, u
i
, M) and
a
i
J for i < k. Since u
k
K(M), let
Q
l
(x
0
, . . . , x
k
) hold in N i M [= Q
l
(x
0
, . . . , x
k1
, x
k
).
2. For only one j < k, u
j
K(M). x
j
= (x
j
, a
j
) for some a
j
J. Let
Q
l
(x
0
, . . . , x
k
) hold in N i
i<k
x
i
(u
k
) = x
k
(l)
where x
j
(u
k
) means a
j
(u
k
).
3. Only u
k
K(M). Choose c so that M [= h(u
k
, c, x
k
, f(u
k
)). Let
Q
l
(x
0
, . . . , x
k
) hold in N i M [=
i<k
x
i
(u
k
) = c(l).
4. If none of the u
i
s are in K(M) then
Q
l
(x
0
, . . . , x
k
) hold in N i
i<k
x
i
(u
k
) = x
k
(l).
It is not hard to see that N dened in this way is a model of and with
the appropriate identications, M N. 2
Corollary 3.3 If is not -categorical then it is not -categorical for any
> .
Proof: Any two models of of cardinality have isomorphic P-parts. Hence
if is not -categorical there must be M [= , [M[ = so that M does not
have a solution.
By lemma 3.2, we can nd N [= and M N so that [N[ = . If is
-categorical then N has a solution since there is a model of of cardinality
with a solution. But then by lemma 3.1, M has a solution which is a
contradiction. Hence is not -categorical. 2
15
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Denition 3.4 Suppose M [= and i
0
, . . . , i
k
are distinct elements of I(M).
Let A = ([i
0
, . . . , i
k
]
k
i
1
, . . . , i
k
) and f be a function with domain
containing A so that
M [= G
b
(l, u, f(l, u)) for all (l, u) A.
Let
x
j
l
= f(l, i
0
, . . . , i
j1
, i
j+1
, . . . , i
k
) for j ,= 0 and l <
and choose y H
b
(i
1
, . . . , i
k
, M). Dene a function g as follows:
g(l) =
0 if M [= Q
l
(x
0
l
, . . . , x
k1
l
, y)
1 otherwise
The invariant for i
0
, . . . , i
k
via f is g + 2
<
, a coset of 2
<
in the abelian
group 2
.
Lemma 3.5 The denition of invariant given above is independent of the
choice of y.
Proof: Use the notation of the denition. Choose any y
so that
M [= H
b
(i
1
, . . . , i
k
, y
).
Let c H
a
(M) be such that
M [= h(i
1
, . . . , i
k
, c, y, y
).
Let
g
(l) =
0 if M [= Q
l
(x
0
l
, . . . , x
k1
l
, y
)
1 otherwise
Now g
+ 2
<
= g + 2
<
. 2
If m and f, g are functions with the same domain dene the relation
m
by
f
m
g i [x : f(x) ,= g(x)[ <
m
.
Denition 3.6 Suppose M [= , I I(M) and i
1
, . . . , i
k
are distinct ele-
ments of I(M) I. Let f be a function with domain that contains
([I i
1
, . . . , i
k
]
k
i
1
, . . . , i
k
)
16
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
so that
M [= G
b
(l, u, f(l, u)) for all (l, u) A.
The 0-invariant for I, i
1
, . . . , i
k
via f is the function g with domain I so that
g(a) = the invariant for a, i
1
, . . . , i
k
via f.
Suppose 0 < m < k, I I(M) and i
1
, . . . , i
km
are distinct elements of
I(M) I and f is a function whose domain contains
A = ([I i
1
, . . . , i
km
]
k
u : i
1
, . . . , i
km
u)
so that
M [= G
b
(l, u, f(l, u)) for all (l, u) A.
Let I
0
. . . I
m1
I where [I
i
[ =
i
. Choose a function f
so that the
domain of f
contains
B = ([I
m1
i
1
, . . . , i
km
]
k
),
f
(l, u) G
b
(l, u, M) and f
f.
Lemma 3.7 The denition of m-invariant above is independent of the choice
of f
.
Proof: Note that by lemma 3.5, the denition of 0-invariant is well-dened.
Use the notation of the denition for m-invariant for m > 0. Choose any
other applicable f
. Let
C =
v : u K(M), l < , c G
a
(M)so that (l, u) B,
c(v) ,= 0 andM [= g(l, u, c, f
(l, u), f
(l, u)).
[C[
m1
since [B[ =
m1
and if a I (I
m1
C) then the value of h(a)
is not eected by the choice of f
instead of f
. Hence the
m
-class of h is
well-dened. 2
Suppose that I is an innite set and g : [I]
k
2
/2
<
. We will dene
the canonical structure M
g
on I via g.
17
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
The P-part of M
g
is the same as M
I
. Moreover, so are the predicates G
b
and g. However, H
b
(u, M
g
) = u g(u) for all u [I]
k
. We modify h so
that
h(u, x, (u, y), (u, z)) holds in M
g
i x +y = z
where the addition takes place in 2
. (Note 2
<
2
.)
The denition of Q
l
is identical to the one for M
I
. It is not hard to show
that M
g
satises .
Theorem 3.8 Let be the least cardinal such that
k1
< 2
. is not
categorical in . In fact, there are 2
k
.
Proof: Let B
0
= f
a
: a 2
/2
<
where f
a
:
0
2
/2
<
so that
f
a
(i) = a for all i
0
. Dene B
m
inductively for 0 < m < k 1. Suppose
we have dened B
m1
. Let C = h : h :
m
m1
B
m1
. Let B
m
be a
maximal collection of ,
m
-equivalent elements in C. It is not hard to show
that [B
m
[ = 2
m
.
Fix A B
k1
k2
of size . We wish to dene a structure M
A
in such a
way as to be able to recover A. Let I
A
=
k2
k1
k1
A. Choose
g
A
: [I
A
]
k
2
/2
<
so that if i
m
m
m1
for 0 < m < k 1, , <
k1
and a A then g(a, (, ), i
k2
, . . . , i
1
) = a()(i
k2
) . . . (i
1
) and otherwise
g(u) is arbitrary. Let M
A
be the canonical structure on I
A
via g
A
.
We try to recover A by looking at (k 2)-invariants. We need to x
certain functions for the rest of the argument. Let
f : K(M
A
) M
A
be dened so that
f(l, u) = (l, u, 0) where 0 is the identity element in
K(M
A
)
Z
2
. Remember that (l, u, 0) is a member of G
b
(l, u, M
A
). Let f be
the restriction of
f to [
k2
k1
k1
]
k
and let h be the restriction
of
f to [
k2
A]
k
.
Claim 3.9 Suppose m < k 1 and i
j
j
j1
for m < j < k 1.
The m-invariant for
m
, i
m
, . . . , i
k2
, (, ), a via
0
, . . . ,
m1
and
f is the
m
-class of a()(i
k2
) (i
m+1
). (If m = 0 then a()(i
k2
) (i
1
) is the
0-invariant.)
18
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
Proof: Notice that
f contains all possible domains required for calculating
invariants.
f essentially chooses the zero in all the G
b
(l, u, M
A
)s.
We prove this claim by induction on m. Suppose the notation is as it is
in the claim. Choose
y a()(i
k2
) (i
1
) = H
b
(u, M
A
)
where u = i
1
, . . . , i
k2
, (, ), a.
Since
f chooses the zero in all G
b
(l, u, M
A
)s, the value y(l) determines
the truth value of the appropriate instance of Q
l
. This is independent of the
choice of i
0
0
so the 0-invariant is a()(i
k2
) (i
1
).
The induction step is similar. 2
A consequence of the claim is that if a A and , <
k1
then the
(k 2)-invariant for
k2
, (, ), a via
0
, . . . ,
k3
(if k > 3) and f h is
the
k2
-class of a(). The domain of h is too large however to allow us to
say we have captured a.
So suppose we use some h
for any h
k1
k1
and f we are able to recover A. We have
xed
k1
elements then and there are 2
many of
the M
A
s are non-isomorphic since
k1
< 2
. 2
Corollary 3.10 1. is not 2
k1
-categorical.
2. T is not relatively categorical.
Proof: The rst is obvious from theorem 3.8, the remark after it and corollary
3.3. To see the second, notice that all the models built in the proof of theorem
3.8 have isomorphic P-parts and are models of T. Hence T is not relatively
categorical. 2
References
[1] H. Gaifman. Characterization of uniqueness and rigidity properties.
preprint.
19
3
2
3
r
e
v
i
s
i
o
n
:
1
9
9
5
-
1
2
-
0
1
m
o
d
i
f
i
e
d
:
1
9
9
5
-
1
2
-
0
2
[2] Bradd Hart. Some results in classication theory. PhD thesis, McGill
University, 1986.
[3] W. Hodges. Relative categoricity in abelian groups. preprint.
[4] W. Hodges, I. M. Hodkinson, and H. D. Macpherson. -categoricity,
relative categoricity and coordinatisation. preprint, 1987.
[5] J. Keisler. Model Theory for Innitary Logic. North Holland, 1971.
[6] L. Marcus. A prime minimal model with an innite set of indiscernibles.
Israel Journal of Mathematics, 11 (1972), pp. 180183.
[7] M. Morley. Categoricity in power. Trans. Am. Math. Soc., 114 (1965),
pp. 514538.
[8] A. Pillay.
0
-categoricity over a predicate. Notre Dame J. of Formal
Logic, 24, 1983.
[9] A. Pillay and S. Shelah. Classication theory over a predicate 1. Notre
Dame J. of Formal Logic, 26, 1985.
[10] S. Shelah. Categoricity of uncountable theories. In L. Henkin et al,
editor, Proceedings of the Tarski symposium, pp. 187203. Amer. Math.
Soc., 1974.
[11] S. Shelah. Categoricity in
1
of sentences in L
. Israel Journal of
Mathematics, 20 (1975), pp. 127148.
[12] S. Shelah. Classication of non-elementary classes 1: the number of
models of L