(

8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES
SH849
SAHARON SHELAH
Abstract. We consider stability theory for Polish spaces and more generally
for denable structures (say with elements of a set of reals). We clarify by
proving some equivalent conditions for
0
-stability. We succeed to prove exis-
tence of indiscernibles under reasonable conditions; this gives strong evidence
that such a theory exists.
0. Introduction
0(A). General Aims.
{0.4}
Question 0.1. Is there a stability theory/classication theory of Polish spaces/algebras
(more generally denable structures say on the continuum)?
Naturally we would like to develop a parallel to classication theory and in
particular stability theory (see [Sh:c]). A natural test problem is to generalize
Morley theorem = Los conjecture. But we only have one model so does it mean
anything?
Well, we may change the universe. If we deal with abelian groups (or any variety)
it is probably more natural to ask when is such (denable) algebra free.
{0.5}
Example 0.2. If P is adding (2
0
)
+
-Cohen subsets of then
(C)
V
and (C)
V[P]
are both algebraically closed elds of characteristic 0 which are not isomorphic (as
they have dierent cardinalities).
So we restrict ourselves to forcing notions P
1
P
2
such that
(2
0
)
V[P1]
= (2
0
)
V[P2]
and compare the Polish models in V
P1
, V
P2
. We may restrict our forcing notions
to c.c.c. or whatever.
{0.6}
Example 0.3. Under any such interpretation
(a) C = the eld of complex numbers is categorical
(b) R = the eld of the reals is not (by adding 2
0
many Cohen reals).
Date: March 14, 2011.
The author would like to thank the Israel Science Foundation for partial support of this research
(Grant No. 242/03). This was part of [Sh:771] which was rst done in June 2002; but separated
in Feb. 2009. First version - 09/Feb/13.
1
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


2 SAHARON SHELAH
(Why? We say assume P
1
P
2
, (2
0
)
V[P1]
= (2
0
)
V[P2]
but R
V[P1]
,= R
V[P2]
. Triv-
ially: R
V[P2]
is complete in V[P
2
] while R
V[P1]
in V
[P2]
is not complete, but there
are less trivial reasons).
{0.7}
Conjecture 0.4. We have a dychotomy, i.e. either the model is similar to cate-
gorical theories, or there are many complicated models under the present inter-
pretation.
So in particular we expect the natural variants of central notions dened below
(like categoricity) will be equivalent; in particular we expect that it will be enough
to consider the forcing notions of adding Cohen reals. Naturally those questions
call for the use of descriptive set theory on the one hand and model theory on the
other hand; in particular to using denability in both senses and using L
1,0
(Q).
Presently, i.e. here there is no serious use of either; the questions are naturally
inspired by model theory. It would be natural to consider questions inspired by the
investigation of such specic structures; to some extent considering the freeness of
a denable Abelian group fall under this.
A priori, trying to connect dierent directions in mathematics is tempting, but
it may well lack non-trivial results. We suggest that the result on the existence of
indiscernibility, 3.8, give serious evidence that this is not the case; note that even
the weaker 1.7 gives more than set theory, i.e. Erd os-Rado theorem. That is, the
question is: is there a non-trivial theory in this direction? While the present work
does not achieve a real theory of this kind, we believe that it gives an existence
proof.
Let us elaborate suggestions for the denition of categorical and for den-
able. Recall that it is well known that: a Borel (set or structure) is a
1
1
-one, a

1
1
-set/structure is a
1
2
-one and also an
1
-Suslin one, and a
1
2
-set/structure is a

1
-Suslin one, see e.g. [Jec03].
{z7f}
Denition 0.5. 1) Let A denote a denition of a -structure, a countable vocab-
ulary, the set of elements of the structure is the reals or a denable set of reals such
that it is absolute enough, i.e. for any forcing notion PQ we have A[V
P
] A[V
Q
]
but we may say the structure/model A.
2) We say A is F

/Borel/
1
1
/-Suslin, etc., if the denition mentioned above is
F

/Borel/
1
1
/-Suslin, etc.
3) We say A is categorical (or categorical
1
, similarly below) when for any forcing
notions P Q such that (2
0
)
V[P]
= (2
0
)
V[Q]
the structures A[V
P
], A[V
Q
] are
isomorphic in V
Q
.
3A) We say A is categorical under (e.g. = (2
0
=
1
) means that for any
forcing notions P Q satisfying
P
,
Q
and
Q
(2
0
)
V[P]
= (2
0
)
V[Q]

the structures A[V

P
], A[V
Q
] are isomorphic in V
Q
.
4) We say A is K-categorical when above P, Q K (or the pair (P, Q) K).
5) If is a denition of a cardinal then A is categorical in , is dened as in (3A)
but
P
,
Q
is replaced by [V
P
] = [V
Q
] = (2
0
)
V[P]
= (2
0
)
V[Q]
and this
is a non-empty condition; similarly in (3A),(4).
6) Let T be a set of (rst order) equations in the countable vocabulary . Let A
be a model of T. We say A is free
1
for K when for every P K the model A[V
P
] is
a free algebra. Similarly, parallely to (3A),(4),(5).
{z7k}
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 3
Conjecture 0.6. 1) If, e.g. a
1
1
-structure A is categorical
1
in some

for K then it is categorical in every

of conality >
0
for K where
K = (P, Q) : PQ and (2
0
)[V
P
] = (Z
0
)[V
Q
]

[V
P
], Card[V
P
] = Card[V
Q
].
2) Or at least for every

1
.
3) Similarly for freeness.
{0.8}
Thesis 0.7. 1) Classication theory for such models resemble more the case of
L
1,
than the rst order.
2) As there (see [Sh:87a], [Sh:87b], [Sh:600]) if the continuum is too small we may
get categoricity for incidental reasons.
See [Sh:h]; as support for this thesis, in [Sh:771, 5] we prove:
{0.9}
Theorem 0.8. There is an F

-abelian group (i.e. an F

-denition, in fact a very

explicit denition) such that V [= G is a free abelian group i V [= 2
0
<
736
.
Comments: In the context of the previous theorem we cannot do better than F

,
but we may hope for some other examples which is not a group or categoricity is
not because of freeness.
The proof gives
{0.10}
Conclusion 0.9. For any n < for some F

-abelian group A, A is categorical in

i n.
A connection with the model theory is that by Hart-Shelah [HaSh:323] such
things can also occur in L
1,
whereas (by [Sh:87a], [Sh:87b]) if

n
(2
n
< 2
n+1
)
and L
1,
is categorical in every
n
, then is categorical in every . See more
in Shelah-Villaveces [ShVi:648].
The parallels here are still open.
Those questions may cast some light on the thesis that non-rst order logics are
more distant from the so-called mainstream mathematics. This work originally
was a section in [Sh:771]; in it we try to look at stability theory in this context,
proving the modest (in 3.8):
for
0
-stable -Suslin models the theorem on the existence of indis-
cernibles can be generalized.
We may consider another interpretation of categoricity. Of course, we can use
more liberal than L[A
2
, r] or restrict the A

s further (as in the forcing version), see

(0C). A natural question is whether in the existence of indiscernibles (in Theorem
3.8) we can start with a set of cardinality rather than
+
+
. Another natural
question is whether categoricity implies that the L
,0
-theory of A is simple
clearly by (0A) we are sure this holds.
Both questions (and maybe in the right parallel of stable) are addressed in a
work in progress, [Sh:F1134].
0(B). The Content of the Present Work.
Our context is a -candidate (A, ), so A is structure with a set of elements/a
set of reals, which is reasonably denable: usually -Suslin; you can x as the
set of quantier free formulas. In the present work forcing does not appear, so the
universe is xed.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


4 SAHARON SHELAH
In 1 we give some basic denitions in Denition 1.4 but note that the
0
-stable
and
0
-unstable are proved to be complimentary only later. The main result is to
get the end-extension indiscernibility existence lemma. Using stability, we have
that for a

: < ) we get a subsequence a

: S), S stationary such

that the -type of a
1
, . . . , a

k
) over a

: <
0
does not depend on a
k
. We
then improve it to does not depend on
k
,
k1
, . . . ,
kn
for a x n, but this
does not give full indiscernibility. Note that the denition of stability and its use
in 1.7 speak of denability of types but the type are model theoretic ones speaking
only on -formulas whereas the denable means in set theoretic sense, so the
denition is arbitrary, just has to be in the submodel. This seems inherent in our
framework: for a predicate P (A), the set theoretic denition of P
A
may involve,
as approximations, relations on the reals which are very complicated.
In 2 we generalize the order property implies unstability, in 2.1 giving a criterion
for unstability in 2.3. Now the unstable in Denition 1.4(3) really speaks on
having a perfect set of types; we here dene apparently weaker version (, , )-
unstable.
Lastly, in 3, we dene ranks, but here the ranks are for subsets I of
m
A, not
just denable ones; as explained above this seems inherent in our framework.
We then prove (in 3.3,3.7) that A is (
0
, )-stable is really dened, i.e. that
we have several equivalent denitions some from the structure side, some from the
non-structure side (generally on this see [Sh:E53, (1A),(2A),(2B)]).
Lastly, we prove a theorem on the existence of (fully) indiscernible sequence:
from a set of cardinality
+
+
to one of cardinality ; a parallel situation occurs for
strongly dependent T (by [Sh:863]). We end noting that, of course, being (
0
, )-
unstable implies a failure of categoricity (strong one) as expected in our frame.
0(C). Further Comments.
{0.12}
Denition 0.10. 1) For a denition A of a -model (usually with a set of elements
a denable set of reals) we say that A is categorical
2
in 2
0
when : for some
real r: for every A
1
, A
2
the models A
L[A1,r]
, A
L[A2,r]
are isomorphic (in V).
2) For a class K of forcing notions and cardinal we say A is categorical
2
in (, K)
when for every P K satisfying
P
2
0
, we have in V
P
: the structure A is
categorical
2
in , i.e. in the sense of part (1).
Comparing Denition 0.10(1) with the forcing version we lose when V = L, as
it says nothing, we gain as (when 2
0
>
1
) we do not have to go outside the
universe. Maybe best is categorical in in V
P
for every c.c.c. forcing notion P
making 2
0
.
Note also that it may be advisable in 0.10(1) to restrict ourselves to the case
is regular as we certainly like to avoid the possibility (2
0
)
L[A1,r]
= < (2
0
)
L[A2,r]
(see on this and for history in [Sh:g, Ch.VII]).
Of course, any reasonably absolute denition of unstability implies non-categoricity:
if we have many types we should have a perfect set of them, hence adding Cohen
subsets of adds more types realized. If we add
i
: i < 2
0
) Cohen reals for every
A 2
0
, A
V[i:iA]
: A 2
0
) are non-isomorphic over the countable set of pa-
rameters, if we get 2
2

0
non-isomorphic models, we can forget the parameters and
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 5
retain our richness in models. The present work is to some extent a continuation
of [Sh:202], [Sh:522], see history there.
{0.15}
Remark 0.11. For () = 1, 2 letting =
()1
, we may replace everywhere -
Suslin, ,
+
by
1
()
,
()1
,
()
respectively. Presently though, the rst (i.e.

1
1
) case is more restrictive than the second (
1
-Suslin) case, what we can prove is
the same. For () = 1 we have real equivalence.
Note that below the indiscernible sequence are indiscernible sets, justied by 2.1
because of (see, e.g. [Sh:c]).
{z20}
Claim 0.12. 1) Assume I is an innite linear order a
s

m
A for s I, a
s
: r
I) is (, n)-indiscernible sequence (see ?) but not a (, n)-indiscernible set (see
1.9(6)). Then we can nd = ( x
0
, . . . , x
n
, y) and

b
g( y)
A and < n such
that: if s
0
<
I
. . . <
I
s
n
then
M [= [ a
s0
, . . . , a
sn
,

b]
M [= [ a
t0
, . . . , a
tn
,

b] when t
k
is s
k
if k ,= + 1, t
k+1
if k = , t
k1
if
k = + 1.
2) So if I = , this implies the (, )-order property.
3) Similarly for end-(, n)-indiscernible sequence (but we may have to demand
s
0
s

I for a x s

.
Recall
{z17}
Denition 0.13. 1) Let T

be the following tree:

(a) members have the form (
t
,
t
) such that for some n,
t

n
,
t

n

(b) it is ordered by s
T
t i
s

t

s

t
.
2) For a subtree T of T

let proj(T) =

: for some

we have
(n, n) : n < T.
3) A set of this form is called -Suslin.
4) For notational convenience we may replace

by

(H (
0
)) and consider a
m
(A) as a

(n) : < g( a)) : n < ).

(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


6 SAHARON SHELAH
1. Generalizing stability in
0
We may consider the dividing line for abelian groups from [Sh:402] and try to
generalize it for any simply dened (e.g. Polish algebra till -Suslin) model. We
deal with having two possibilities, in the high, complicated side we get a parallel
of non
0
-stability (hence strong non-categoricity, see 3.10; in the low side we have
a rank. But even for minimal formulas, the example in [Sh:771, 5] shows that we
are far from being done, still we may be able to say something on the structure.
We may consider also ranks parallel to the ones for superstable theories. Note
that there are two kinds of denability we are considering: the model theoretic one
and the set theoretic one.
{7.0}
Context 1.1. 1) If not said otherwise, A will be a structure with countable vocab-
ulary and its set of elements is a set of reals; usually a denition - see 0.5(1).
2) L is a logic. We did not specify the logic; we may assume it is L
1,0
or just
L
1,0
(Q) where Q is the quantier there are uncountably many.
3) is a x cardinality, let a -model mean a -Suslin model.
{6a.del}
Denition 1.2. 1) For a structure A, an A-formula or (A, L)-formula is a
formula in the language L(
A
) so in the logic L, and the formula in the vocabulary
of A with nitely many free variables, writing = ( x) means that x is a nite
sequence of variables with no repetitions including the free variables of .
2) denotes a set of such formulas and denote a pair (
0
( x),
1
( x)) of formulas
so is a -pair when
0
,
1
.
3) We say (or or ) is -Suslin (or
1
1
or
1
2
or
1
0
(= Borel)) i they are so
as set theoretic formulas.
{7.0k}
Discussion 1.3. 1) So if the relation A, (
0
, . . . ,
n1
) P
A
where P is
a predicate
1
of
A
, are -Suslin, then they are upward and downward absolute (as
long as the relevant subtrees of T

are in the universe).

2) When we consider a formula ( x) this is more complicated by we are assuming
it, too, is -Suslin, meaning that this (set theoretic) denition denes ( x)
A
also
in the other universes we consider.
{7.1}
Denition 1.4. 1) We say (A, ) is a -candidate (or -Suslin candidate; as is
constant we may omit it) when :
(a) A is a -model
(b) is a countable set of (A, L)-formulas
2
which, are in the set theory sense,
-Suslin (we identify and )
(c) we consider changes of the universe (say by forcing) only which this is
preserved, anyhow we can use each ( x) is quantier free.
1A) We can replace being -Suslin by being
1
1
by
1
2
, etc., (naturally we need
enough absoluteness); if we replace it by we write -candidate. If does not
appear we mean it is
1
1
or understood from the context.
2) If (A, ) is a candidate we say A is (
0
, )-stable (or (A, ) is
0
-stable), when
is a countable set of (A, L)-formulas and for large enough and x H (),
1
pedantically we should change a little in 0.13, or we can translate P
A

n
(

) by a subset
of

2
We may use is a set of pairs = (
0
( x), , ( x) of such formulas, and is later demand
. So far it does not matter.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 7
for every N (H (), , <

) of cardinality such that + 1 N and to which

x belongs and a
m
A where m < the following weak denability condition on
tp

( a, N A, A) holds:
letting
m
A,
be dened in (), for some function c N with domain
m
(A,)
to 0, 1 the statement () holds where
() letting
m
A,
=
m
(A,)
= ( x,

b) : ( x,

b) = (
0
( x,

b),
1
( x,

b))
and x = x

: < m),

b
>
A and
0
,
1
and
A [= ( x)(
0
( x,

b)
1
( x,

b)),
() if = (
0
( x,

b),
1
( x,

b))
m
(A,)
N and < 2 and
A [=

[ a,

b] then = c( ).
3) We say that (A, ) is
0
-unstable (or A is (
0
, )-unstable) when : there are
a

m
A for

2 and
,0
( x, y

) and
,1
( x, y

) and

b

g( y)
A for

>
2 such that:
(a) A [= ( x)(
,0
( x,

)
,1
( x,

))
(b) if
0
,
1
and
0
,
1


2, n = g() and
0
(n) = 0,
1
(n) = 1 then
A [=
,0
[ a
0
,

]
,1
[ a
1
,

].
4) Let ( x, y)
m
(A,)
means that ( x, y) = (
0
( x, y),
1
( x, y)) and
0
( x, y),
1
( x, y)
belongs to and A [= ( x, y)(
0
( x, y)
1
( x, y)).
5) Let ( x, y) means that ( x, y) = (
0
( x, y),
1
( x, y)) and
0
( x, y),
1
( x, y)
belongs to .
Remark 1.5. 1) There are obvious absoluteness results (for
m
(A,)
, (A, ) is

0
-unstable and
0
-stable).
2) On those notions being complimentary see Theorem 3.3.
{7.2}
Observation 1.6. 1) If is closed under negation then in Denition 1.4(2) we
equivalently can replace by:

for some c N we have: if ( x, y) and

b
g( y)
A and

b N then
A [= ( a,

b) i c(( x,

b)) = 1.
2) In Denition 1.4(2) we can x (A, ) and omit <

, at the expense of having to

use a somewhat larger .
Proof. Straight.
{7.4}
Claim 1.7. The End-Extention Indiscernibility existence lemma
Assume (A, ) is a
0
-stable candidate.
1) In Denition 1.4(2), the demand N is countable can be omitted.
2) Assume is closed under negation and permuting the (free) variables, m <
, a

m
A for < and
0
< = cf() and S is stationary and A A has
cardinality < . Then for some stationary S

S the sequence a

: S

) is a
-end extension indiscernible sequence over A in A (see Denition 1.9(3),(4),(5)
below).
3) Moreover for any pregiven n < we can nd stationary S

S such that
a

: S

) is (, n)-end extension indiscernible over A in A.

4) In part (2) we can nd a club E of and regressive function f
n
on S E for
n < such that:
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


8 SAHARON SHELAH
(i) if , S E then f
n+1
() = f
n+1
() f
n
() = f
n
()
(ii) if n < and < , then the sequence a

: S E, f
n
() = ) is
(, n)-end extension indiscernible over A
(ii)
+
moreover, if n < and , < then a

: S E and f
n
() = ) is
(, n)-end extension indiscernible over A a

: < .
{7.4y}
Remark 1.8. 1) This is a rst round on indiscernibility.
2) The assumption is closed under negation is quite strong.
3) Really we get indiscernible sets by 0.12 and 2.1.
4) The claim and proof are similar to [Sh:c, Ch.III,4.23,pg.120-1], but before proving
we dene:
{7.4a}
Denition 1.9. 1) Let (A, ) be a candidate. We say A has (, )-order when :
()

for some m() < and ( x, y)

m()
A,
with g( x) = g( y), this formula
linear orders some I
m()
A of cardinality , see part (2) for denition.
2) We say that the formula ( x, y) linear orders I A i for some a
t
: t I) we
have:
(a) I = a
t
: t I
(b) I is a linear order
(c) = (
0
( x, y),
1
( x, y))
m
(A,)
(d) if s <
I
t then A [=
0
[ a
s
, a
t
]
1
[ a
t
, a
s
].
3) For a linear order J (e.g. a set of ordinals), we say a
t
: t J) is a -end-
extension indiscernible sequence (over A) i for any n < and t
0
<
J
< . . . <
J
t
n1
<
J
t
n
, the sequences a
t0
. . . a
tn2
a
tn1
and a
t0
. . . a
tn2
a
tn
realize the
same
3
-type (over A) in A.
4) We say that a
t
: t J) is (, n
0
, n
1
)-end-extension indiscernible over A in A
when :
(a) J a linear order for some m, a
t

m
A, A A
(b) if r

: < n
0
), s

: < n
1
), t

: < n
1
) are <
J
-increasing sequences,
r
n01
<
J
s
0
, r
n01
<
J
t
0
then a
r0
. . . a
rn
0
1
a
s0
. . . a
sn
1
1
and
a
r0
. . . a
rn
0
1
a
t0
. . . a
tn
1
1
realizes the same -type over A in A
(c) if J has a last element we allow to decrease n
0
and/or n
1
.
5) If we omit n
0
this means for every n
0
, (so -end extension... means (, 1)-end
extension.
6) We say a
t
: t I) is (, n)-indiscernible sequence over A when it is an (, 0, n)-
indiscernible sequence over A.
7) We say a
t
: t I) is a -indiscernible sequence over A when it is an (, n)-
indiscernible sequence over A for every n.
Proof. Proof of 1.7
1) Let N

(H (), , <

) be such that A, N

. Now for every countable

N N

to which (A, ) belongs there is c

N
N as mentioned in the denition
1.4 or really 2.1(2). Hence by normality of the club lter on [N

]
0
, the family of
3
We may consider not containing a contradictory pair.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 9
countable subsets of N

, for some c

the set N = N : N N

is countable and
c
N
= c

is a stationary subset of [N

]
0
, so c

can serve for N

.
2) Let N

: < ) be an increasing continuous sequence of elementary submodels

of (H (), , <

) to which A belongs, such that |N

| < , N

and N

and a

: < ) N
0
(hence a

N
+1
). For each S, applying 1.7(1)
to N

, a

we get c

as in Denition 1.4(2). So for some c

and some
stationary subsets of S

S of we have S

= c

. Now -end extension

indiscernibility follows.
3) We prove this by induction on n:

for all m < a stationary S , a

m
A for < there is a stationary
S

S such that: if < ,

for < n and

0
<

1
<
. . . and

0
<

1
< . . . then a

0
. . . a

n1
, a

0
. . . a

n1
realizes the
same -type over A a

: < .
For n = 0 the demand is empty so S

= S is as required. For n = 1 apply the

proof of part (2). For n+1 > 1 by the induction hypothesis we can nd stationary
S
1
S as required in
n

. For each < we can choose

,
= (, ) for
n such that =
,0
<
,1
< . . . <
,n
and 0 < n
,
S
1
. Let
a

= a
,0
. . . a
,n
so a

m(n+1)
A and apply the induction hypothesis or just
the case n = 1, i.e. part (2) to m (n + 1), S
1
, a

: < ) getting a stationary

S
2
S
1
as required in
n

or just in
1

.
We claim that S
2
is as required. So assume

0
< . . . <

n
< and

0
< . . . <

n
< and

S
2
for n.
Now, letting (, ) =
,
for n we have:
(i) a

0
a

1
. . . a

n
and a

0
a
(

0
,1)
. . . a
(

0
,n)
realizes the same -type
over A a

: < in A.
[Why? As 1, . . . , n (

1
, ) S
1
,

S
2
S
1
and the choice of S
1
).]
(ii) a

0
a
(

0
,1)
. . . a
(

0
,n)
is equal to a

0
.
[Why? By the choice of a

0
.]
(iii) a

0
, a

0
realizes the same -type over A a

: < hence over A a

:
< .
[Why? By the choice of S
2
].
Similarly
(iv) a

0
is equal to a

0
a
(

0
,1)
. . . a
(

0
,n)
(v) a

0
a
(

0
,1)
. . . a
(

0
,n)
and a

0
a

1
. . . a

n
realizes the same -type
over A a

: < .
By (i)-(v) the set S
2
is as required in
n+1

.
4) The proofs of parts (2), (3) actually give this.
1.7
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


10 SAHARON SHELAH
2. Order and unstability
{7.7}
Claim 2.1. The order/unstability lemma:
Assume that

1
(a) (A, ) is a -candidate, (e.g. () 1, 2, =
()1
the relation
are
1
()
hence a -Suslin)
(b)
0
( x, y),
1
( x, y) are contradictory in A
(c) J is a linear order of cardinality
(d) a
t

m
A for t J
(e) A [=
0
[ a
s
, a
t
] and
1
[ a
t
, a
s
]
whenever s <
J
t

2
(a)

+ or
(b) J is with density < [J[ and < [J[.
Then (A, ) is
0
-unstable; even more specically the demand in Denition 1.4(3)
holds with
,0
=
0
,
,1
=
1
.
{7.7q}
Question 2.2. What can : A has a (, )-order, that is
1
of 2.1 holds be?
We rst prove a claim from which we can derive the lemma.
{7.7A}
Claim 2.3. (A, ) is
0
-unstable when we have:
(a) (A, ) is a -candidate and m < , =
m
(A,)
or just
m
(A,)
(b)

P = P

: <
+
)
(c) P

is a non-empty family of subsets of

m
A
(d) if < <
+
and I P

then for some I

0
, I
1
P

which are subsets of

I and some pair (
0
( x,

b),
1
( x,

b)) we have < 2 and a I

A [=

( a,

b)
(e) if I P

and < <

+
and F is a function with domain I and range
of cardinality , then there is I

such that I

I and F I

is
constant.
Proof. Proof of 2.1 from 2.3
Let = (
0
( x,

b),
1
( x,

b)) :

b
g( y)
A.
Case 1:

+
For <
+
we let
P

= I : I
m
A is linearly ordered by and has cardinality
+
.
Case 2: J has density , [J[ > +
For <
+
we let
P

= I : I
m
A is linearly ordered by getting an order of cardinality
> + and density .
This should be clear.
By
2
of the assumption at least one of the cases holds.
2.1
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 11
Similarly
{b7}
Remark 2.4. Like 2.1, replacing clause (e) by
(e)

for some < n and = ( x

0
, . . . , x
n
, y), g( x

) = m;

b
g( y)
A and a
s

m
A for s J and the conclusion of 0.12.
Proof. Proof of 2.3
For each ( x) as a
g( x)
A : A [= [ a] is a -Suslin set and
4
()
( x)
we can nd C
,
: and < ) such that
(a) a
g( x)
A : A [= [ a] = a: for some < and

we have
( a, ) C
,

(b) if < then C

,
is closed subset of
(g( x)+1)
(

).
We can nd functions F
0

, F
1

such that if ( x) and A [= ( a) then F

0

( a)

and F
1

( a)

witnessing this and code a continuously. For notational simplicity
and without loss of generality m = 1. Let W = w : w
>
2 is a front
5
hence
nite.
For w W and n < let Q
n,w
be the family of objects x = (n, u, , , ) =
(n
x
, . . .) such that:
()
n,w,x
for unboundedly many <
+
we can nd a witness (or an -witness)
y = ( a

: < n), B

: w)) which means:

(a) u = (u
0

, u
1

) : w) and w u
0

, u
1

n and =
i
: i <
n),
i
= (
i
0
(x, y

),
i
1
(x, y

))
m
(A,)
(b) a

g( y

)
A
(c) B

for w
(d) if w,

b B

and < n then (

0
( x, a

),

1
( x, a

))
m
(A,)
, g( x) =
m, g( y

) arbitrary (but nite) and

() u
0

A [=

0
[

b, a

]
() u
1

A [=

1
[

b, a

]
(e) if ,= are from w then (u
0

u
1

,= ) (u
0

u
1

,= )
(f) () =
i
,
: w, i 0, 1 and u
i

)
() =
i
,
: w, i 0, 1 and u
i

)
()
i
,

>

() <
>

(g) if w,

b B

, i 0, 1, u
i

then
i
,
F
0

i
(

b, a

) and
i
,

(F
1

i
(

b, a

)).
Clearly

1
Q
0,{<>}
,= .
4
Pedantically note that we sometimes consider a
g( x)
A as a member of

or better

(H (
0
)) rather than
g( x)
(

) or see 0.13(4).
5
i.e. for every

2 there is one and only one n < such that n w
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


12 SAHARON SHELAH
[Why? Let x = (0, <>, (, ), <>, <>, (x)(x = x)) and if <
+
choose I P

we let B
<>
= I; recall that by clause (c) of Claim 2.3 the family P

is non-empty.]

2
if x Q
n,w
and for w then there is y such that:
y Q
n,w
u
y
= u
x

i
x,,

i
y,,

i
y,,
=
i
x,,

y
=
x
.
[Why? As x Q
n,w
we know that for some unbounded Y
+
for each Y
there is an -witness y

= a

: < n)B

: w) as required in ()
n,w
. Let
<
+
and () = Min(Y ( +1)). Now for each w we have B
()

P
()
.
Let k > supg(
i
x,,
: < 2, w and < n + 1 and, of course, the set
F

i
(

b, a
()

)k :

b B
()

has cardinality for each i < 2, < 2, < n, w.

Hence by clause (e) of Claim 2.3 as < () we can nd a subset B
,
of B
()

from P

and
,i
,

w>
w,
,i
,

>
such that

b B
,
F
1

i
x,
(

b, a
()

)k =
i
,

b B
,
F
0

i
x,
(

b, a
()

k =
,i
,
.
Similarly for some
,i
,
,
,i
,
: < 2, w, < n) we have
Y

= Y :
,i
,
=
,i
,
and
,i
,
=
,i
,
is unbounded in
+
.
Now it is easy to choose y

3
if x Q
n,w
and w and u = (w) < 0 >, < 1 > so u W,
then there is y Q
n+1,u
such that:
() n
y
= n
x
+ 1
() u

y,
= u

x,
for w, = 0, 1
u

y,<j>
0, . . . , n 1 = u
x,
for = 0, 1 and j = 0, 1
()
i
y,,
=
i
x,,
for w

i
y,<>,
=
i
x,,
()
i
y,,
=
i
x,,
for w

i
y,<>,
=
i
x,,
()
y
n =
x
.
[Why? Similar to the proof of
2
using clause (d) of the assumption of Claim 2.3
this time.]
Together it is not hard to prove the non
0
-unstability.
2.3
{7.8}
Remark 2.5. 1) This claim can be generalized replacing
0
by , strong limit sin-
gular of conality
0
.

(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 13
{7.9}
Denition 2.6. 1) tp

( a, A, A) = ( x,

b) : ( x, y) and

b
g( y)
(A) and
A [= [ a,

b].
2)
pr,m
A,,A
= (
0
( x,

b),
1
( x,

b)) :
0
( x, y),
1
( x, y) belong to and

b
g( y)
A and
x = x

: < m) and A [= ( x)[

0
( x,

b)
1
( x,

b)] where A A, a set of A-

formulas, and so
pr,m
A,
= (
0
( x, y),
1
( x, y)) :
0
,
1
, A [= y x[
0
( x, y)

1
( x, y)].
3) S
m

(A, A) = tp

( a, A, A) : a
m
A where A A and a set of L(
A
)-formulas.
{7.10}
Denition 2.7. 1) We say (A, ) is (, , )-unstable i there are M A, m <
and a

: < ) such that:

(a) a

m
A
(b) if ,= are < then for some (
0
( x,

b),
1
( x,

b))
pr,m
A,,M
(see Denition
2.6(2)) we have
0
( x,

b) tp

( a

, M, A) and
1
( x,

b) tp

( a

, M, A)
(c) |M| .
1A) Let A be (
0
, , per)-unstable mean that (A, ) is
0
-unstable; here per stands
for perfect.
2) In part (1) and (1A) we add weakly if we weaken clause (b) to
(b)

tp

( a

, M, A) ,= tp

( a

, M, A) for ,= from X
(so if is closed under negation there is no dierence); in part (1),
X = and in part (1A), X =

2.
3) We use (
0
, , x, Q) where Q is a forcing notion i the example is found in V
Q
such that usually M is in V and we add an additional possibility if x = per then
M V and X = (

2)
V
(here per stands for perfect).
4) We may replace a forcing notion Q by a family K of forcing notions (e.g. the
family of c.c.c. ones) meaning: for at least one of them.
5) We replace unstable by stable for the negation.
{7.11}
Observation 2.8. If is closed under negation, then A is weakly (
0
, , )-
unstable i A is (
0
, , )-unstable.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


14 SAHARON SHELAH
3. Rank and Indiscernibility
{7.12}
Denition 3.1. Let (A, ) be a -candidate. Let 1 m N and assume D is
a
+
-complete lter on
m
A; (or a lter on some I
m
A then we interpret it as
J
m
A : J I D; writing rk
m

we meand rk
m
D
, D = I
m
A :
m
AI has
cardinality similarly rk
n
<
).
For I
m
A we dene rk
m
D
(I) = rk
m
D
(I, , A), an ordinal or innity or 1 by
dening for any ordinal when rk
m
D
(I) by induction on .
Case 1: = 0.
rk
m
D
(I) i I ,= .
Case 2: limit.
rk
m
D
(I) i rk
m
D
(I) for every < .
Case 3: = + 1.
rk
m
D
(I) i (a) + (b) holds where
(a) if I = I
i
: i < then for some i < we have rk
m
D
(I
i
)
(b) we can nd ( x,

b)
m
A,
and I
0
, I
1
I such that rk

(I

) and
a I

A [=

( a,

b) for = 0, 1.
{7.13}
Observation 3.2. Assume (A, ) is a -candidate.
1) If are ordinals and rk
m
D
(I) then rk
m
D
(I) .
2) rk
m
D
(I) Ord 1, is well dened (for I
m
A).
3) If I
1
I
2
A then rk
m
D
(I
1
) rk
m
D
(I
2
).
Proof. Trivial.
{7.14}
Theorem 3.3. The following are equivalent if 2
0

+
, (A, ) is a -candidate:
(a) rk
m

(
m
A)
+
for some m
(b) A is (
0
, )-unstable, see Denition 1.4(3)
(c) A is (
0
, ,
+
)-unstable
(d) rk
m
D
(
m
A) = for some m
(e) A is not (
0
, )-stable, see Denition 1.4(2).
Proof. (a) (b).
Let P

= I
m
A : rk
m

(I) for <

+
and apply 2.3.
(b) (c): Trivial as we are assuming 2
0

+
.
(c) (d):
Let A A be countable and a

: <
+

m
A exemplies that A is
(
0
, ,
+
)-unstable.
Without loss of generality
() if

b A, ( x, y) and <
+
: A [= ( a

b) is bounded then it is
empty.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 15
Now let P = a

: S : S
+
is unbounded. Now we can prove by
induction on that I P rk
m
D
(I) .
(d) (a): Trivial.
(e) (c):
Let A A, [A[ and let S
m

(A, A) = tp

( a, A, A) : a
m
A for some m.
Let be large enough, N (H (), ) be of cardinality such that A, A N
and + 1 N. By Denition 3.7(2) and the present assumption (e), for every
m and p S
m

(A, A) there is c
p
N such that p is disjoint to q
no
cp
and if is
closed under negation is equal to p
yes
cp
where for t yes, no we let q
t
c
:= ( x,

b) :
( x, y) ,

b
g( y)
(A N) and c(( x,

b)) = t. Hence if is closed under

negation, then S
m
(A, A) has cardinality |N| = ; as this holds from every
A A of cardinality we get (c), as promised.
If is not closed under negation but (c) holds, let S = tp

( a

, A, A) : <
+

be as in Denition 2.7(1), rest should be clear.

(d) (e):
We have to prove (e), i.e. A is (,
0
)-stable, see Denition 1.4(2).
So assume M N, N (H (), ), A N, +1 N and |N| = and a
m
A
and we should nd a function c as there. Now let Z := rk
m
k
(I) : I B where
B = I : I N is a non-empty subset of
m
A to which a belongs. Now this family
B is not empty (as
m
A B) hence Z ,= .
Of course, B, Z N and / Z by our present assumption, i.e. (d). Hence
= min(Z) is a well dened ordinal, so by the denition of Z there is I B such
that rk
m

(I) = . Now why rk

m

(I) := + 1? By Denition 3.1, case 3, clause

(a) or clause (b) there fail.
If clause (a) fails, then there is a sequence

I = I
i
: i ) of subsets of I
with union I such that i < rk
m

(I
i
) < , clearly

I H () hence without
loss of generality

I N. As N necessarily i < I
i
N and obviously
i < I
i
I
m
A.
Lastly, as a I necessarily there is j < such that a I
j
. So I
j
witness
rk
m

(I
j
) Z and as said above rk
m

(I
j
) < , so contradicting the choice of as
min(Z).
So clause (b) of case 3 of Denition 3.1 fail (for our , , I) so

1
for every ( x,

b)
m
A,
we can choose < 2 such that rk
m

( a

I : A [=

[ a

b]) < .
So there is a function t N with domain
m
A,
and range 0, 1 such that

2
if ( x,

b)
m
A,
then t = t( ( x,

b)) satises rk
m

(I
( x,

b),t
) < where

3
I
( x,

b),t
= a

I : A [=
t
[ a

b].
However

4
if ( x,

b)
m
A,
N and
t
( x,

b) is satised by a, t < 2 and A [=

t
[ a,

b]
then
(a) I
( x,

b),t
B
(b) rk
m

(I
( x,

b),t
) Z
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


16 SAHARON SHELAH
(c) rk
m

(I
( x,

b),t
) =
(d) t ,= t( ( x,

b)).
[Why? For clause (a): as ( x,

b) N clearly I
( x,

b),t
N and it is I
m
A. Also,
a I
( x,

b),t
by the assumption of
4
. Hence by the denition of Z also clause
(b) holds. But by 3.2(3) we have rk
m

(I
( x,

b),t
) rk
m

(I) = so by the choice of

as min(Z ) we get equality, i.e. clause (c) holds. Recalling
2
we get clause (d)
contradicting the present assumption.]
Now by
4
(as t is a function from N), we are done proving (e).
3.3
{7.14d}
Conclusion 3.4. 1) The property rk
m

(A) = is preserved by forcing.

2) If rk
m

(
m
A) = then for some U
1
coding the denitions of A and of the
-Suslin denitions of every ( x) , and U
1
Ord every forcing extension of
L[U
1
, U
2
] satises rk
m

(
m
A) =
{7.15}
Denition 3.5. If p is a (
1
, m)-type over A in A (i.e. a set of formulas ( x, a)
with ( x, y)
1
, a A), we let (may write instead of <
+
)
rk
m
<
(p,
1
, A) = Minrk
m
<
(

<n

(
m
A,

),
1
, A) : n <
and

( x,

) p for < n.
{7.16}
Observation 3.6. 1) If p q (or just q p) are (, m)-types in A then rk
m
<
(q, , A)
rk
m
<
(p, , A).
2) If q is a (, m)-type in A then for some nite p q we have
rk
m
<
(q, , A) = rk
m
<
(p, , A)
hence
p r q rk
m
<
(r, , A) = rk
m
<
(p, , A).
{7.17}
Claim 3.7. Assume
+
2
0
. In 3.3 we can add
(f) (A, ) is not
0
-stable
(g) for some < cf() and, of course, < cf(), 2
0
the pair (A, ) is
(, , )-unstable
(h) like (f ) for every such , .
Proof. (f) (c).
Let M (H (), , <

) be countable such that x M for suitable x and

m < . For every a
m
A there is a function c
a
M from
m
(A,)
to 0, 1
as in Denition 1.4. So if a
i

m
A for i <
+
then for some i < j <
+
we
have c
ai
= c
aj
because M is countable hence for no ( x,

b)
m
(M,)
do we have

0
( x,

b) tp

( a
i
, M, A),
1
( x,

b) tp

( a
j
, M, A). So clearly (c) fails.
(f) (c).
Fix (H (
0
), , <

) and let
S
0
= M (H (
0
), , <

) : A M and |M| =
and + 1 M.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 17
For m < and I
m
A let J
I
= J[I] be the family of S S
0
such that: we can
nd F
x
, c
x
: x H ()) (a witness) such that:
() c
x
:
m
A,
0, 1
() F
x
:
>
(H ()) H ()
() if M S
0
is closed under F
x
for x M then for every a I for some
y M, c
y
is a witness for tp( a
M
, M A, A), see Denition 1.4(2).
Clearly J
I
is a normal ideal on S
0
. Also if m < S
0
J[
m
A] then
increasing we get the desired result. Toward contradiction assume that m <
and S
0
/ J[
m
A] and let P (i.e. P

= P for <
+
) be the family of I
m
A
such that S
0
/ J
I
.
We now nish by 2.3 once we prove
if I P then for some ( x,

b)
m
A,
for each < 2 the set I

(x,

b)
is
a I : A [=

( a,

b) belong to P.
If not, for every ( x,

b)
m
A,
there is = c[ ( x,

b)] < 2 and (F

( x,

b)
x
, c
( x,

b)
x
) :
x H ()) witnessing S
0
J[I

].
Dene (F
y
, c
y
) for y H () by: if y = x, ( x,

b)) then F
y
= F
( x,

b)
x
, c
y
=
c
( x,b)
x
, otherwise c.
Clearly we can nd M S
0
such that

1
if ( x,

b)
m
A,
M and x M then M is closed under F
( x,

b)
x

2
for some a
m
A, no c
y
, y M denes tp

( a, M A, A), in the sense of

1.4(2).
But c does it! So we are done.
(h) (g).
Obvious.
(g) (d).
Like (c) (d).
(c) (h).
Repeat the proof of (c) (d) in 3.3.
3.7
{7.18}
Theorem 3.8. Assume that (A, ) is a -candidate and is
0
-stable or just -
stable.
For some <
1
we have: if , m < , A A, [A[ and a

m
A for
<
+
then for some S
+
of cardinality the sequence a

: S) is
-indiscernible over A in A.
Proof. Assume not. For <
1
let
P

= a

: <
+
: for no S
+
of cardinality
is a

: S) is -indiscernible over A in A.
The central point is to note the obvious:
if
+
is regular, > 0, A A, [A[ , a

m
A for <
+
and S
+
is stationary then (a) or (b) where
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


18 SAHARON SHELAH
(a) for some club E of , a

: S E) is -indiscernible over A in A
(b) for some m < and club E

m
of
+
we have
(b)
m
(i) a

: S E

m
) is (, m)-end extension indiscernible
(ii) for no club E

m
of
+
is a

: S E
1
) a sequence
which is (, m + 1)-end extension indiscernible.
Clearly clause (a) is impossible by our present assumptions so let E

, m be as in
clause (b). By claim 1.7(4) there is a club E of and f
n
: n < ) as there and let
S

= S : f
m+1
() = , so > , P
m+1
= : S

is stationary. Without
loss of generality E

E and / P
m+1
S

= . Without loss of generality

f
m+1
is as in claim 3.9 below.
So by (b)
m
(ii) clearly P
m+1
is not a singleton (and it cannot be empty), so we
clearly have nished.
3.8
{7.19}
Claim 3.9. Let A, a

: < ), E, f
n
: n < ) be as in 1.7(4). Then without loss
of generality (possibly shrinking E and changing the f
n
s) we can add
(iii) if n < and
1
,=
2
are in Rang(f
n+1
) but f
n+1
(
1
) =
1
f
n+1
(
2
) =

2
f
n
(
1
) = f
n
(
2
) letting S = : f
n
() = f
n
(
1
) = f
n
(
2
) and
= Min(S E(
1
+ 1)(
2
+ 1),
then for some formula ( x
0
, . . . , x
n
) with parameters from A a

: < such
that:
() if i < 2,

0
< . . . <

n+1
are from S E( n)()(f
n
() = f
n
(

)
f
n+1
() =
i
) and f(

0
) =
i
then A [= [ a

0
, . . . , a

n
] i = 0.
Proof. Easy.
3.9
{c19}
Claim 3.10. Assume (A, ) is a -candidate and it is (
0
, )-unstable, see Claim
3.3 and Denition 1.4(3). Then A is not categorical, even under 2
0
=
1
, see
0.5(3A).
Remark 3.11. Of course, we can get stronger versions: many models.
Proof. Let x = x
m
,

= (
,0
( x, y

),
,1
( x, y))
m
(A,)
for
>
2 and

:

>
2), a

:

2) be as in Denition 1.4(3).
Without loss of generality this is absolutely, i.e. if P is a forcing extension, and
(

2)
V[P]
then we can choose a

.
Let Q be the forcing of adding
2
Cohens,

: <
1
), so P is trivial and
easily A = A
V
, A
V[Q]
are not isomorphic: if F is such that G Q be generic over
V,

[G

] and toward contradiction p F is an isomorphism from A

V[Q]
onto A
V
where p G. So for some () <
1
, F

(b

) :
>
2) depend just
on

: < ()) so in V[ ()] we can compute it, so F

()
) can have no
possible value contradiction.
3.10
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


BEGINNING OF STABILITY THEORY FOR POLISH SPACES SH849 19
4. Private Appendix
Discussion 4.1. (2011.2.23)
1) Can we characterize categoricity in
1
. So assume
0
-stability. We hope: the
elmination in L(aa), has
0
-homogeneity.
2) Can we get elimination of quantiers.
3) Should we give details in 0.9 instead as in [Sh:771, 5].
Maybe see [Sh:F562].
Moved from pg.5:
Metric Models
Concerning [Sh:771, 1] we may think of a more general context.
{1.6}
Denition 4.2. 1) We say a is a metric -model ( =
a
is a vocabulary, that is a
set of function symbols and predicates; in the main case we say -algebra when
has no predicates only functions) i
(a) a is a metric space with metric d
a
(b) M
a
= M(a) is a model or an algebra with universe [M
a
[, (of course with
a set of elements the same as the set of points of the metric space), with
vocabulary =
a
(c) if F
a
is (an n-place) function symbol, then F
M(a)
is (an n-place)
continuous function from M
a
to M
a
(for d
a
, i.e. by the topology which the
metric d
G
induces, of course)
(d) if R
a
is an n-place predicate, then R
a
= R
M(a)
is a closed subset of
M
n
a
=
n
([M
a
[).
2) We say a is unitary if some e
a
is a unit of a which means that e an individual
constant and e
a
is closed under F
M(a)
for F
G
and R
G
e, . . .) R
a
.
3) We say a is complete if ([M
a
[, d
a
) is a complete metric space.
4) We replace unitary by specially

unitary above if we add:

() for every R
+
, the set a M
a
: d
a
(a, e
a
) < is a subalgebra of A.
5) We replace unitary by specially-unitary if some

witness it which means:
(a)

is a decreasing sequence of positive reals with limit zero
(b) for every F
a
for some n = n(F, a) we have for every m [n, ) the set
a M
a
: d
a
(a, e
a
) <
m
is closed under F
M(a)
.
6) We add the adjective partial if we allow F
M(a)
to be a partial function, so in
clause (c) of part (1) means now:
(c) if F
a
= (M
a
) is an n-place function symbol then the set (a
1
, . . . , a
n
, F
M(a)
(a
1
, . . . , a
n
) :
a
1
, . . . , a
n
M
a
and F
M(a)
(a
1
, . . . , a
n
) is well dened is a closed subset
of
n+1
(M
a
).
7) We say a is specially-unitary i some

witnesses it which means (a),(b) from
part (5) and
(d) for any F
a
a k-place function for every m n(F, a) we also have:
if d
a
(x

, y

) <
m+1
for = 1, . . . , k and d
a
(x

, e
a
) <
m+1
, d
a
(y

, e
a
) <

m+1
then d
a
(F
M(a)
(x
1
, . . . , x
k
), F
M(a)
(y
1
, . . . , y

)) is <
m
.
(
8
4
9
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
3
-
1
4







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
3
-
1
5


20 SAHARON SHELAH
References
[Jec03] Thomas Jech, Set theory, Springer Monographs in Mathematics, Springer-Verlag, Berlin,
2003, The third millennium edition, revised and expanded.
[Sh:c] Saharon Shelah, Classication theory and the number of nonisomorphic models, Studies in
Logic and the Foundations of Mathematics, vol. 92, North-Holland Publishing Co., Amsterdam,
xxxiv+705 pp, 1990.
[Sh:g] , Cardinal Arithmetic, Oxford Logic Guides, vol. 29, Oxford University Press, 1994.
[Sh:h] , Classication Theory for Abstract Elementary Classes, Studies in Logic: Mathe-
matical logic and foundations, vol. 18, College Publications, 2009.
[Sh:E53] , Introduction and Annotated Contents, 0903.3598.
[Sh:87a] , Classication theory for nonelementary classes, I. The number of uncountable
models of L
1
,. Part A, Israel Journal of Mathematics 46 (1983), 212240.
[Sh:87b] , Classication theory for nonelementary classes, I. The number of uncountable
models of L
1
,. Part B, Israel Journal of Mathematics 46 (1983), 241273.
[Sh:202] , On co--Souslin relations, Israel Journal of Mathematics 47 (1984), 139153.
[HaSh:323] Bradd Hart and Saharon Shelah, Categoricity over P for rst order T or categoricity
for L
1
can stop at
k
while holding for
0
, ,
k1
, Israel Journal of Mathematics 70
(1990), 219235, math.LO/9201240.
[Sh:402] Saharon Shelah, Borel Whitehead groups, Mathematica Japonica 50 (1999), 121130,
math.LO/9809198.
[Sh:522] , Borel sets with large squares, Fundamenta Mathematicae 159 (1999), 150,
math.LO/9802134.
[Sh:F562] , Continuing 771.
[Sh:600] , Categoricity in abstract elementary classes: going up inductively,
math.LO/0011215.
[ShVi:648] Saharon Shelah and Andres Villaveces, Categoricity may fail late, Journal of Symbolic
Logic submitted, math.LO/0404258.
[Sh:771] Saharon Shelah, Polish Algebras, shy from freedom, Israel Journal of Mathematics 181
(2011), 477507, math.LO/0212250.
[Sh:863] , Strongly dependent theories, Israel Journal of Mathematics accepted,
math.LO/0504197.
[Sh:F1134] , Stability Theory on -Souslin Structures.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at