(

9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS
SH977
SAHARON SHELAH
Abstract. We prove that the theory of abelian groups and R-modules even
in innitary logic is stable and understood to some extent.
0. Introduction
0(A). Aims.
We like to know how complicated is a class of R-modules which are models of a
sentence in an innitary logic (for rst order logic we know much):
(A) is it stable? (say no formula ( x, y) L
,
(
R
) linearly ordering ar-
bitrarily long sequence of tuples in some models of )? Can we dene
non-forking?
(B) do we have a parallel of the main gap, i.e. proving that either every M
Mod

can be characterized by some suitable cardinal invariants or there

are many complicated M Mod

?
Here we rst show that for any R-module, in L
,
(
R
) or better L
,,
(
R
) we have
a version of eliminating quantiers (however we add parameters). By this we can
prove some versions and consequences of stability. More specically
after expanding by enough individual constants, every formula in L
,,
(
R
)
is equivalent to a Boolean combination of positive existential such formulas
stability, i.e. no long sequences of linearly ordered (< )-tuples
(
ep
,
, 2)-indiscernible implies
ep
,
-indiscernible
convergence follows.
0(B). Preliminaries.
{z7}
Notation 0.1. 1) Let

be if =
+
and if is a limit cardinal.
Date: September 9, 2010.
The author thanks Alice Leonhardt for the beautiful typing. First version done April 2010.
The author would like to thank the Israel Science Foundation for partial support. Paper Number
977.
1
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


2 SAHARON SHELAH
{z0}
Denition 0.2. 1) We say is a -additive (or a -Abelian) vocabulary when
has the two-place functions x + y, x y, the individual constant 0 and the other
predicates and function symbols has arity < .
2) M is a -additive structure when :
(a)
M
is a -additive vocabulary
(b) M is a -structure
(c) G
M
:= ([M[, +
M
,
M
, 0
M
) is an Abelian group
(d) if P T is an -place predicate then P
M
is a sub-group of (G
M
)

(e) if F +, 0, 0 is an -place function symbol then F

M
is a partial -place
function from M to N and graph(F
M
) = aF
M
( a)) : a Dom(F
M
) is
a subgroup of (G
M
)
+1
.
{z12}
Observation 0.3. 1) For a ring R, an R-module can be considered a
0
-additive
structure in the vocabulary

R
.
2) For a -additive model M, for every -term ( x) we have M [= ( a

b) =
( a) (

b) and M [= P( a

b) when M [= P( a) P(

b).
Remark 0.4. Fisher [Fis77] denes and deals with Abelian structure in other
directions, those notions are related.
{a4}
Denition 0.5. 1) Let
R
= (R) be the vocabulary of R-modules, i.e. have binary
functions x + y, x y, individual constant 0 and F
a
, multiplication by a from the
left for every a R.
2) If x, y has length then x + y = x

+y

: < ), x y = x

: < ) and
similarly a x for a R, and when we replace x and/or y by a member of

M.
{z15}
Remark 0.6. 1) We may use +, , 0, 1 P
i
: i < i(), P
i
unary and
instead modules use -models M such that [M[ = P
M
i
: i < i(), +
M
is a
partial two-place function, +
M
= +
M
P
M
i
: i < i(), (P
M
i
, +
M
) an abelian
group, all relations and functions commute with + or at least every relation is
ane, i.e. let F

(x, y, z) = x y + z, and demand G(. . . , F

(x
i
, y
i
, z
i
), . . .)
i<i()
=
F

(G( x), G( y), G( z)) and a,

b, c P
M
F

( a,

b, c) = F

(a
i
, b
i
, c
i
) : i < arity(P))
P
M
.
2) However, as we use innitary logics, if M is the disjoint sum of Abelian groups
G
M
i
:= (P
M
i
, +
M
i
) for i < i() we dene G
M
as the direct sum having predicate for
those subgroups we have bi-interpretability. Concerning having ane structure,
we can expand by choosing an element in each to serve as zero.
3) It is natural to extend our logic by cardinality quantiers saying the denable
subgroup G divided by the denable subgroup H has cardinality . This causes
no serious changes in the proof.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 3
1. Eliminating quantifiers
{a2}
Convention 1.1. 1) R is a xed ring =
R
, see below or is an additive
vocabulary.
2) K is the class of R-modules or of -additive models.
3) M, N will denote R-modules or are -additive models.
4) = cf().
{a6}
Denition 1.2. We dene
pe
,
=
pe,
,
=
pe,
,
(), a set of formulas ( x) in
L
,
() in fact in L
,,
() with g( x) = < , by induction on the ordinal
which is -increasing with and is of cardinality

([[ +

) as follows; we
write
pe
,,
for the set of = ( x, y), g( x) = , g( y) = with
pe
,+
and

pe

=
pe
,
: < ,
pe
,,<
=
pe
,,
: < .
Case 1: = 0
For R-modules:
It is the set of = ( x) of the form:

<n
a

x
()
= 0 with () < g( x) or better,

<
a

= 0 where a

R is 0
R
for all but nitely many s.
For the -additive case:
It is the set of ( x) has the form P( ( x)), a sequence of length arity(P) of
terms (in the variables x), P may be equality.
Case 2: a limit ordinal
It is
pe
,
(R) : < .
Case 3: = + 1 for some < and
pe
,+
we have ( x) = y(( x y) :
( x, y) ).
{a8}
Claim 1.3. For M K and ( x)
pe
,
(), the set (

M) =

b

M : M [= [

b]
is a sub-abelian group of

M and the set

b

M : M [= [

b a] is ane (=
closed under x y + z) for any a

M.
Proof. Easy.
{a10}
Theorem 1.4. For every for every M K there is a subset I = I

of
>
M
of cardinality

([[ +

) such that: in M every formula ( x,

b) from
L
,,
(), so

b
>
M and g( x) < , is equivalent in M to a Boolean combination
of formulas of the form ( x a) with ( x)
pe
,g( x)
() and a I
g( x)
M.
{a12}
Conclusion 1.5. 1) For every M K, < and a
n
M, for some and

( x)
pe
,,
, we have a

M : tp

pe
,,
( a

, , M) = tp

pc
,,
( a, , M)
is equal to a

M : M [=

[ a

]

<

( a

a).
{a15}
Remark 1.6. 1) We shall use Claim 3.1 in the proof; this is better but not manda-
tory.
2) Note that instead of an R-module M we can use (M, c

)
<
, i.e. expand M by
individual constants; the only dierence is using

([R[ +

+) instead.
3) The result has an arbitrary choice: the I

, so e.g. not every formula ( x)

L
,,
and a I

is ( x, a

) equivalent to a formula without parameters. Instead

(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


4 SAHARON SHELAH
of using extra individual constants, in the proof (see

) for any ( x), ( x)

i
( x)
for i < i() <

, I, G, G
i
(i < i()) and the ideal I on

can expand M by:

(a) P
M
= a : M [= [ a] and i <

: a / G
i
I is a subgroup
(b) predicates for the set a +P
M
: a (M).
So the proof shows that we can in M eliminate quantier to quantier-free formulas
in this expansion.
4) Also this may give too much information. Still the result gives elimination of
quantiers: not as low as in the rst order case.
{a17}
Denition 1.7. 1) We say

b
1
,

b
2


M are -equivalent over I
>
M when
( x

)
pe
,
(R), a I M [= [

b
1
a] [

b
2
a].
2) Replacing I by A means I =

A : < .
We shall use freely
{a19}
Observation 1.8. The sequence

b
1
,

b
2


M are -equivalent over I

M i for
any ( x)
e
,
we have (a) (b) where:
(a) for some a I

M we have M [= [

b
1
a] [

b
2
a]
(b) for every a I

M we have M [= [

b
1
a] [

b
2
a]
Proof. Straight.
??
Proof. Proof of 1.4
By induction on we choose I

and prove the statement. For = 0 and a

limit ordinal this is obvious so assume = + 1 and we shall choose I

.
Choose I

such that

(a) I

is a subset of
>
M
(b) [I

[ 2

where

([[ +

)
(c) I

(d) If < and

i
( x)
ep
,
and a
i
I

M for i < i()

and
there is

d

M such that M [=
i
[

d a
i
] for i < i()
then there is such

d I

(e) Assume < , g( x) = , ( x) is a conjunction of formulas from

ep
,
and
i
( x)
ep
,
for i <

and apply 3.1 with

= (2

)
+
,

, (

M), (

M)
i
(

M) for i < here

standing for S, G, G
s
(s S) there; (i.e. the subgroups of ([M[, +
M
)
with universe as above) getting the ideal I on

such that

/ I. For any u I there are

d

i
(

M) for
< () 2

such that for every a (M) there is < ()

satisfying i < : a

d

/
i
(

M) I
(f) if

d
1
,

d
2
I

M then

d
1
+

d
2
I

,

d
1

d
2
I

and
<

0

d
1
I

.
To prove the induction statement for clearly it suces to prove:
assume , < ; if

b
1
,

b
2


M are -equivalent over I

and c
1


M then
for some c
2


M the sequences

b
1
c
1
,

b
2
c
2

(+)
M are -equivalent over
I

.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 5
Why holds? Let x be of length and y of length . Let
1
= ( x, y)
ep
,+
:
for some a I

+
M we have M [= [

b
1
c
1
a] and for ( x, y)
1
choose
a
( x, y)
I

M such that M [= [

b
1
c
1
a
( x, y)
]. Let
2
= ( x, y)
ep
,+
:
( x, y) /
1
.
So by

(d) there is a

b

I
,+
be such that g(

) = g(

b
1
), ( c

) = g( c
1
)
and ( x, y)
1
M [= [

a
( x, y)
]. For transparency note that if
2
=
then by the assumption of there is c
2


M such that ( x, y)
1
M [=
(

b
2
c
2
a
( x, y)
), so c
2
is as required, hence we are done so without loss of generality

2
,= . Clearly [
2
[

and let

= (

, y) : ( x, y)

for = 1, 2.
Let
i
( x y a
i
) : i <

list the set of formulas ( x y a) satised by

c
1

b
1
with ( x, y)
ep
,+
, a I

and let

i
( y) =
i
(0

, y).
Let the ideal I on

be dened as in 3.1 with G =

M) :

( y)

1
and
G
i
= G

i
(

M) for i S :=

, = (2

)
+
.
Case 1:

I.
So clearly M [= [

b
1
b

, c
1
c

] for every ( x, y)
1
.
Let

( x, y) = ( x, y) : ( x, y)

1
, so clearly it
ep
,+
and M [=

b
1

, c
1
c

] hence M [= ( y)

b
1
b

, y]. But ( y)( x, y)

,
so by the
assumption on

b
1
,

b
2
we have M [= ( y)

b
2

, y] hence for some c

2
we have
M [=

b
2

, c

2
] and let c

2
= c

2
+ c

, so M [=

b
2

, c

2
c

]. As we are in case
1, there is a sequence e

: < ) of members of G, i.e. of a

M : M [= (

, a)
such that i <

((1) < (2) < ) e

(2)
e
(1)
/ G
i
.
So for every < , the sequence (

b
2

)( c

2
c

+ e

) belongs to

(
+
M)
and <

: (

b
2

)( c

2
c

+ e

) belongs to ( a
i

) + G

has at most
one member. As

< for some < , (

b
2

b)( c

2
c

+ e

) / a
i

+G
i
:
i <

.
So c

2
:= c
2
+ e

is as required.
Case 2:

/ I
So there is a sequence

: < ()) as in

(e) for , G, G
i
(i <

) as above
so < ()

d

. As clearly c
1
c

G necessarily for some < () the set

u = i <

: ( c
1
c

) / G
i
belongs to I and, of course,

( c

) I

+
M
and we have:
M [=

b
1

, c
1
c

]
if i

u then M [=
i
[

b
1

, c
1
c

].
As in Case 1 there is c

2


M such that
M [=

b
2

, c

2
c

]
if i

u then M [=
i
[

b
2

, c

2
c

].
As u I by 3.1 there is a sequence e
j
: j <
+

) and u

, u u

I such that:
e
j
G
i
for i

e
j2
e
j1
/ G
i
for j
1
< j
2
<
+

, i u

.
So
(

b
2

)( c

2
c

e
j
) belongs to

(M)
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


6 SAHARON SHELAH
if i

u then (

b
2

)( c

2
c

e
j
) belongs to
i
( x y a
i
) in
M hence

b
2
( c

2
e
j
) satises the formula
i
( x y a
i
) in M.
Lastly,
for i u, there is j
i

+

such that for every j

+

j
i
the sequence
(

b
2

)( c

2
c

e
j
) satises
i
( x y a
i
), so for some j, ( c

e
j
)
is as required.

1.4
{a21}
Denition 1.9. Let = cf(), an ordinal,

=

: < ).
1) For an R-module M we say

I is a (

, , )-witness for M when I = I

: )
is dened by induction on as in the proof of 1.4, with (

([R[ +

))
+
(instead equality).
{a23}
Remark 1.10. If

([[+) is the rst beautiful cardinal > [[+, then for =

<
the proof of 1.4 works for L

+
,
(). The point is then having to choose

we try
by induction on <

choose

( x, y)
2
L

+
,

p

+
,g( x y)
such that

1
<
2
H

H
1
where H

= (
+
M) : ( x, y)
1

(
+
M).
As is beautiful (see [EM02] or [GbSh:880]) we are stuck in some () <

.
{a26}
Claim 1.11. For every and =
<(||+)
and M K there is I

M of
cardinality 2

such that: if ( x, y) L

+
,
() and

b
g( y)
M, in M, to a Boolean
combination of formulas of the form ( x a) with ( x)
pe

,g( x)
L

+
,
()
and a I
g( x)
M.
Proof. Similarly.
1.11
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 7
2. Stability
{b2}
Context 2.1. 1)
(a) R a xed ring, =
R
or
(b) is a -additive vocabulary; K the class of -additive models.
2) M K a xed R-module.
3) = cf() and an ordinal () limit for simplicity.
4)

=

: ()) as in 1.4,

>

:=

([R[ +

).
5)

I

is a (

, , ())-witness.
6) A

= a : a I
()
.
7)

=
pe
(),
for < and =

: < .
8) M

= M
A
:= (M, a)
aA
.
{b4}
Remark 2.2. 1) The demand that () is a limit ordinal is not really necessary, but
otherwise well have to be more careful.
2) We may add cardinality and even dimension quantiers for

>

.
{b6}
Denition 2.3. Assume < ,
pe
()
and A

A M K and a

M.
1) S

(A, M) = tp

( a, A, M) : a

M, see below.
2) For a

M let tp

( a, A, M) = ( x

b c) :

b

A and c
+
M and
M [= [ a
1

b c] and ( x, y)
pe
,+
.
{b7}
Claim 2.4. Assume
pe
()
and A M K.
1) The set S

(A, M) has cardinality [A[

()
.
2) There are no a

M,

M for <
()
and ( x, y)
pe
(),,
such that
for < < we have M [= [ a

b
p
] [ a

].
3) If A

A
L
,,()
M and p S

(A, M) and ( x, y)
,,
, A closed under
x y and p( x,

b) :

b

A ,= then for some a

N we have ( x a

)
p.
Proof. 1) Consider the statement
() if ( x, y)
pe
(),,
and p

( x) = tp
{( x, y)}
( a

, A, M) S

(A, M)
for = 1, 2 and

b

M, c
+
M and ( x

b c) p
1
( x) p
2
( x) then
p
1
( x) = p
2
( x).
Why () is true? Assume ( x

) p
1
( x), so a
1

(

M). But we are
assuming ( x

b c) p

( x) = tp
{( x, y)}
( a

, A, M) hence a

b c (M) for
= 1, 2. Together a
2

= ( a
2

b c) ( a
1

b c) + ( a
1

) belongs to
(M), hence (x

) p
2
(x). So ( x

) p
1
( xb

) p
2
and by
symmetry we have hence p
1
( x) = p
2
( x), i.e. we have proved ().
Why () is sucient? For every < , ( x, y)
pe
(),,
and p( x)
S

(A, M) choose (

b
p( x),( x, y)
, c
p( x),( x, y)
) such that

1


b
p( x),( x, y)


A and c
p( x),( x, y)

+
A
if possible ( x

b
p( x),( x, y)
c
p( x),( x, y)
) p( x).
For p( x) S

(A, M) let
p( x)
= ( x, y)
pe
,,
: in
1
we have possible and
let q
p( x)
= ( x

b
p( x),( x, y)
c
p( x),( x, y)
) : ( x, y)
p( x)
.
Now
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


8 SAHARON SHELAH

2
if p
1
( x), p
2
( x) S

(A, M) and
p1( x)
=
p2( x)
and q
p1( x)
= q
p2( x)
then
p
1
( x) = p
2
( x).
[Why? Just think.]

3
the set (
p( x)
, q
p( x)
) : p( x) S

(A, M) has cardinality p

<
2
||
+
([A[
<
)
||
.
[Why? Straightforward.]
2),3) Should be clear.
2.3
Recall ([Sh:300a])
{b8}
Denition 2.5. For we say I

M is (, )-convergent when [I[ and
for every < and ( x)
+
and

b

M, c
+
M for all but < of the a I
the truth value of a

b c (M) is constant.
{b10}
Claim 2.6. 1) A sucient condition for I = a
i
: i <

M is (, )-convergent
is: for some , I

M and i < j < ( x)

a
j
a
i
(M).
2) If < , = cf() >
()
and (i < )([i[

()
< ) and a
i


M for
i < with no repetition then for some stationary S , a
i
: i S is (
+
, )-
convergent.
Remark 2.7. 1) So 2.6(1) says that 2-indiscernible implies (< )-indiscernible.
2) Also 2.6(2) says there are indiscernibles.
Proof. Should be clear.
2.6
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 9
3. How much does the subgroup exhaust a group
{p2}
Claim 3.1. Assume the groups G
s
(for s S) are subgroups of the Abelian group
G and > [S[
+
. There is an ideal I on S (possibly I = P(S)) such that:
(a) for every u I there is a sequence g
u
= g
u,
: < ) of members of G
such that s S < < g
u,
G
s
,= g
u,
G
s
(b) for u P(S)I, clause (a) fails
(c) if S / I, cf() > 2

and < [[

< , e.g. ()( = (

)
+
) then
there is A G of cardinality < such that for every g G for some a A
we have s S : gG
s
,= aG
s
I
(d) moreover, (i.e. under the assumptions of clause (c)) for every u I for
some g and v we have
u v I
g = g

: < )
g

G
s
= g
0
G
s
for s Sv
if s v, < < then g

G
s
,= g

G
s
.
{p3}
Denition 3.2. For G and

G = G
s
: s S) as in 3.1 and
0
let I = I
G,

G,
be as dened in clauses (a),(b) of 3.1, it is an ideal (but may be P(S)).
Proof. Let I be the set of u S such that clause (a) holds.
Now
() () I P()
() I is -downward closed
() I is an ideal.
[Why? Clauses (), () are obvious. For clause (), let u
1
, u
2
I be disjoint and
we shall prove that u := u
1
u
2
T. Let g
,
: < ) witness u

I for = 1, 2.
We try to choose (

) by induction on < such that for every <

we have s u g
1,
g
2,
G
s
,= g
1,
g
2,
G
s
.
Arriving to , if we fail then there are f : and g : u such that
for (i, j) we have g
1,i
g
2,j
G
g(i,j)
= g
1,f(i,j)
g
2,f(i,j)
G
g(i,j)
.
For each i < , < and s u S let U
2
i,,s
= j < : f(i, j) = , g(i, j) = s.
Now j U
2
i,,s
g
1,i
g
2,j
G
s
= g
1,
g
2,
G
s
g
2,j
G
s
= g
1
1,i
g
1,
g
2,
G
s
hence for
s u
2
we have j(1) ,= j(2) U
2
i,,s
g
2,j(1)
G
s
= (g
1
1,i
g
1,
g
2,
)G
s
= g
2,j(2)
G
s
contradiction. Hence U
2
i,,s
has cardinality 1 for i < , < , s u
2
.
For j < , < and s u let
U
1
j,,s
= i < : f(i, j) = and g(i, j) = s.
As G is abelian, as above we have < j < s u
j
[U
1
j,,s
[ 1. For
1, 2 and i < let U

i
= U

i,,s
: < and s u

, so as [u

[ [S[ clearly
[U

i
[ [S[. As > [S[
+
there are i, j < such that i U
1
j
j / U
2
i
; hence the
pair (i, j) satises the demand on the pair (i

, j

).
So we can carry the induction on < , so we are done proving ()() hence
proving ().
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


10 SAHARON SHELAH
Now for 3.1, we have chosen I such that clauses (a),(b) holds and prove that it
is an ideal.
Toward proving clause (c) of 3.1 for each u I
+
= P(S)I, let g
u
= g
u,
: <
(u)) be a maximal sequence of members of u such that < < (u) s u
g
u,
G
s
,= g
u,
G
s
. By the denition of I as u / I, necessarily (u) < , and as we
are assuming cf() > 2
|S|
, clearly () = sup(u) : u I
+
< . So B := g
u,
:
u I
+
and < (u) is a subset of G of cardinality < . For every u I and
h : Su B choose g
h
G such that, if possible, (s Su)(g
h
G
s
= h(s)G
s
), so
A = g
h
: h is a function from Su into B and u I is a subset of G of cardinality
[B[
|S|
< .
We shall show that A is as required, then we are done. Let g

G. Let
u = s S: for no u I
+
and < (u) do we have gG
s
= g
u,
G
s
. Now if u I
+
then g
u
= g
u,
: < (u)) is well dened and g

satises the demand on g

u,(u)
contradicting the maximality of g
u
. So u I and we can nd h : (Su) B such
that s Su g

G
s
= h(s)G
s
. So g
h
is well dened and A and is as required,
so we are done.
Clause (d) is proved similarly.
3.1
{p5}
Claim 3.3. In 3.1 there is a W S such that
(a) there is s = s
i
: i < i()) listing W such that
() (

i<j
G
si
,

G
si
) is nite for j < i() stipulating

i<0
G
si
= G
(b) if W

S satises (A) then W

W.
Proof. Immediate.
Example 3.4. An example of additive structure is a ring satisfying xy = yx, i.e.
if (R
+
) is Zx
s
: s I, f is a function from I I into R
+
is such that f(x, y) =
f(y, x) and f(x, x) = 0 and (

<()
a

x
s

) (

m<n()
b
m
x
tn
) = a

b
m
f(s

, t
m
) : <
(), m < m().
Remark 3.5. Concluding Remark
Questions:
1) Can we build other modules L
,,
(
R
)-equivalent to a given R-module?
2) For =
0
we get upward L ow.Sk.Tar. theorem
3) Can we get many non-isomorphic models from un-superstability?
4) For >
0
we have ways to build

eq

-submodels, but can we go up?

5) Use [Sh:F1050] context, with equality and completeness, characterize the case
of few dimension characterize a module condition for having many pairwise non-
isomorphic models.
6) Probably, as in [Sh:i, V], i.e.[Sh:300g] we can use squares which does not reect
in small conality to get non-structure results, i.e. constructing modules.
Discussion 3.6. 1) (2010.5.25) We may like to characterize when T = Th
L
,,
(M)
has not too many models. So probably either we have some sort of unsuperability
and have many models or have few dimensions.
2) In ZFC, for >
0
seems hopeless.
3) If we allow to change the universe by -complete forcing for large enough
we may hope to do this for any , but have not really looked at it.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 11
4) So for trying the main gap we may restrict ourselves to the case =
0
. Seems
O.K. - DO.
But we like to axiomatize at least completeness (under relevant norms). So we
may consider [Sh:F1050] to avoid coding -sequences - DO. (so the case when the
formulas of a structure determine them in the completion)
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


12 SAHARON SHELAH
4. Private Appendix
Generalizing
{d8}
Denition 4.1. Let be a cardinal, if =
0
we may omit it.
1) We say is an abelian -vocabulary when
(a) the two-place binary operation +, and individual constant 0 belongs to
it
(b) each predicate P has arity arity

(P), number of places, <

(c) similarly for function symbols.
2) For an abelian -vocabulary let AS

be the class of M such that

(a) M is a -structure M
(b) ab(M) = M+, , 0 is an abelian group
(c) if P is an n-place predicate then P
M
is a subgroup of
n
(ab(M)), i.e.
if a,

b P
M
then a

b P
M
hence a +

b P
M
(d) if F is an -place function symbol then < and F
M
is a parti-
tion function with domain a subgroup of

M and F
M
commutes with +
M
(equivalently aF
M
( a) : a
n
M satises (c).
{d12}
Denition 4.2. Let
0
, if =
0
we may omit it.
1) We say is an ane -vocabulary when as above but
(e) for predicate P , Q := gr

(P) is a predicate in of the same arity and

we try -place function symbol as ( + 1)-places predicate.
2) For a sentence L
,
(+) let AS
,
is the class of M AS
,
satisfying ;
similarly for a set of sentences.
3) For as above AfS

, the class of ane -structures is the class of M such that

(a),(b) as above
(c)

for P and n-place predicate and a,

b, c P
M
then a b + c P
M
(d)

for F an n-place function symbol and a,

b, c
n
M we have F( a

b+ c) =
F( a) F(

b) +F( c)
(e) for every predicate P letting Q := gr

(P) we have Q
M
= a

b : a,

b
P
M

4) Af S
,
as above.
Remark 4.3. The use of gr

(P) is just an notational assumption as we can add

them.
{d20}
Theorem 4.4. As in 1.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 13
5. Toward main gap
{e2}
Discussion 5.1. 1) Below maybe instead L we have L
1
L
2
.
2) It seems reasonable
() for L = L
,
() or if M
0

L
M

C, tp(M
2
, M
1
) does not fork over M
0
then M
1
+M
2
C.
3) Unsuperstability means M

n

L
M

n+1
M

c, f
n
: M
1
n

iso
onto
M
2
n
, f
n

f
n+1
, by

n<
f
n
is not L-elementary.
4) So it seems hopeful that as in [Sh:300g] assume existence of non-reecting sta-
tionary S S

0
, f we can construct.
5) But for M

: < independent over M

, but is

<
M

)
C
< C? Probably we
have to complete. We may hope for innitary Skolem functions.
6) Maybe we should restrict ourselves to L
,0
.
{e4}
Discussion 5.2. 1) Restricting ourselves to =
1
, i.e. L

+
,0
() help us to
circumvent this rst obstacle. But if we like to answer the question which classes
of Abelian groups can be characterized by a cardinal invariant it seems to me
reasonable to include the class which Abelian group theorist considered so. But
this inclusion torsion complete p-groups which has completeness demand not
expressible by L
,0
().
So we shall use topological logic. But then it is natural that the topologies are
topological Abelian groups.
2) What about elds?
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


14 SAHARON SHELAH
6. Continuing [Sh:F956, 2]
This is complimentary to [Sh:F956, 2.1] and [Sh:F957, 1].
{f2}
Theorem 6.1. Assume R is a ring, M is an R-module and
0
.
1) If
1
,
2
:= |M|
<
then M
1
, M
2
are L
,
-equivalent.
2) If is the rst beautiful cardinal > [R[+ and
1
,
2

and M is an R-module
then M
1
, M
2
are L
,
-equivalent.
Proof. Similar to [Sh:F957, Theorem 1.1].
Stage A: For < let

1

+

= ( x
[]
) : a positive existential formula from L
,
(
R
) where
x
[]
= x

: < ), x
[,]
= x

: [, ])

1
note
(A) if H K and 1 and f
0
, . . . , f
1
H

then H

[=
[f
0
, . . . , f
1
] i < H [= [f
0
(), . . .]
(B) for every abelian group (or R-module) H the subset [H] :=
a

H : H [= [ a] is an abelian subgroup of H, equivalently satis-
es: if a
1
, a
2
are in it and s

Z, not R! for = 1, 2 then s

1
a
1
+s
2
a
2

[H] (the subset is not necessarily a submodule); note that it suces
to consider the case s
1
= 1 as then a

1
= a
1
+ (s
2
1) a
2
(H) and
a

2
:= (s
1
1) a
1
+ a
2
= a
2
+ (s
1
1) a
1
(H) hence a

1
+ a

2
(H)
but a

1
+ a

2
= s
1
a
1
+s
2
a
2
.
Notation: We stipulate the
+
n
to be pairwise disjoint

2
(a) every equation

<
d

= 0 (where d

R for < )
belongs to
+

(b)

: < is closed under adding dummy variables

(c)
+

is closed under permuting the variables

(d)
+

is closed under conjunction

(e) if ( x
[+1]
)
+
+1
then

( x
[]
) = x

( x
[+1]
)
belong to
+

.
Stage B: As there.
Stage C: Use ( x
[]
, x
[,]
). For part (1) as there. For part (2). Let

( x
[]
, x
[,]
)
be as there; H = c
[,]
M : M [=

( o

, c] for c H let
c
( x
[,]
) =
( o

, x
[,]
) ), M [= [ o

, c). We let c

: <

) be a maximal (under
<) sequence of members of H such that letting H

= c
[,]
M : M [= [0, c] we
have
1
<
2
<

H
2
H
1

<

.
[Why? As in [?].]
()

9
(?) it suces to nd b H
2
such that <

b / (H

)
2
.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 15
[Why? Assume that ( x
[]
, x
[,]
)

tp

( a
1

b
1
, , M
1
) and without loss of
generality

( x
[]
, x
[,]
) is a conjunct of . There is c is - FILL.]

7. Private Appendix
OLD - introduction moved 2010.5.24, pg.1:
For rst order logic, we know much of the theories of modules [here?], e.g. any
formula in L(
modulo
(R)) is equivalent to a Boolean combination of existential pos-
itive formulas
any existential positive formulas ( x) in L(
modulo
(R)) dene, in any R
module M, a subgroup of
g( x)
M
for any R-module M, Th
L(
modulo
(R)
(M) is a stable rst order theory
for any and for many and I
n
M of cardinality there is a -
convergent J [I]

.
Moved 2010.4.30 from Denition 1.2,pg.2:
Recall L
,,
() is the set of formulas of L
,
of quantier depth (where
( x), ( y) contribute 1 to the depth but , do not).
Moved 2010.5.2 from pg.3:
Variant of Denition ??, Case 2: For some < disjoint subsets
1
,
2
of
pe
,+
(R)
and non-zero (

([R[ +

)))
+
, it has the form
( y
0
, . . . , y
i
, . . .)
i<
(

i<

( x)1
( x y)

i<j<

( x)2
( y
i
y
j
)).
Moved 2010.4.30 from Proof of Theorem 1.4, choice of I

, pg.3:
For any < < and disjoint
1
,
2

ep
,
let

d
,,
: [
1
,
2
] : i < i(, )) be a maximal subset of

d
+
M : M [= [

d]
for ( x)
1
and

d = 0, 0, . . .) such that i < j < i(, ) ( x)
2

M [= [

d
,,
j
[
1
,
2
]

d
,,
i
[
1
,
2
]].
Let
I

= d
,,
[
1
,
2
] :
1
,
2

ep
,
are disjoint, < , i < i(, ) and
[i(, )[

([R[ +

).
This is easy.
Moved 2010.4.30 from Proof of Theorem 1.4, getting c
2
, pg.3:
Let
2
= ( x, y)
eq
,+
: ( x, y) . So

( x) = ( y)
1

ep
,
hence by the denition of
1
we have M [=

b
1
] hence by the assumption of
we have M [=

b
2
] hence by the choice of there is c
2


M such that
M [= [

b
2
c
2
a
( x, y)
] for every ( x, y)
1
. The proof now splits according to
the case in () above.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


16 SAHARON SHELAH
For each i < i( + , ) we may by c
2,i
=

b
2
c
2


d
,
i
[
1
,
2
] clearly it is well
dened, belongs to
+
M, also as (

d
,,
i
[
1
,
2
]) is constantly 0

, clearly c
2,i
=

b
2
.
By the rst demand on

d
,,
i
[
1
,
2
] it (
+
M) :
1
, but also

b
2
c
2
is in it so c
2,i
belongs to it. By the second demand, for each ( x, y)
2
, for at
most one i < (

([R[ +

))
+
we have M [= [ c
2,i
].
Case 1: i( +, ) (

([R[ +

))
+
.
As [
2
[ [
ep
+,
[

([R[ +
+
), so for some i, c
2,i
really c
2,i
[, + ) is as
required.
Case 2: i( +, ) <

([R[ +

))
+
.
So

d
(+,)
(

([R[ +

))
+
, , [
1
,
2
) : i < i(+, ) (

([R[ +

))
+
+
, ) I

and we cannot choose

d
,,
i
[
1
,
2
] =

b
2
c
2
as

b
2
c
2
(M) : =
( x, y)
+
necessary.
If for some i < i( +, ) and ( x, y)
2
we have M [= [

b
2
c
2

d
,,
i
[
1
,
2
]
but then ( x, y)
1
by the denition of
1
, contradiction.
For every = ( x, y)
pe
, x+
choose, if possible,

b
1,
c
1,

+
N such that
b
1
c
1

b
1,
c
1,
(M) and let
1
be the set of the ( x, y)
pe
,+
for which
this is possible. So there is

b
1,
c
1,

+
N such that

b
1,8
c
1,

b
1,
c
1,
(M)
for every
1
.
So

b
1
c
1

b
1,
c
1,
(M) : ( x, y)
1
and choose c
2


M be such that

b
2
c
2

b
1,
c
1,
(M) : ( x, y)
1
.
If c
2
is as required we are done. Otherwise, let
2
=
pe
,+

1
and

d I

such
that

b
2
c
2

d (M).
Moved 2010.4.300 from ??, pg.4:
2) We may say I

is a (

, , )-witness.
3) If

is constantly we may write ?
{a15}
Denition 7.1. 1) For
2

pe
,
let E

, i.e. E

( x, y) = xE

y with g( x) = g( y)
be dened by: for an R-module M and a
1
, a
2


M let M [= a
1
E

a
2
i (b

M)(( a
1

b) ( a
2

b)).
2) For
1
,
2

pe
,
let E
1,2
( x, y) = xE
,1,2
y - FILL.
{a17}
Claim 7.2. 1) For
pe
,
and R-module M, the relation E
M

is an equivalence
relation on

M, o

/E
M

is a subgroup and a + o

/E
M

: a

M is the set of
E
M

-equivalence classes.
2) [Check] So instead expanding M by individual constants for I

, we may add
unary predicates naming the E
M
,1,2
-equivalence classes; will then we have to use
the ane version, see ??.
References
[EM02] Paul C. Eklof and Alan Mekler, Almost free modules: Set theoretic methods, North
Holland Mathematical Library, vol. 65, NorthHolland Publishing Co., Amsterdam, 2002, Re-
vised Edition.
[Fis77] Edward R. Fisher, Abelian structures. I., Abelian group theory (Proc. Second New Mexico
State Univ. Conf., Las Cruces, N.M., 1976) (Berlin), Lecture Notes in Math., vol. 616, Springer,
1977, pp. 270322.
(
9
7
7
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
1
8







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


MODULES AND INFINITARY LOGICS SH977 17
[Sh:i] Saharon Shelah, Classication Theory for Abstract Elementary Classes 2, Studies in Logic:
Mathematical logic and foundations, vol. 20, College Publications, 2009.
[Sh:300a] , Stability theory for a model, Chapter V (A), in series Studies in Logic, vol. 20,
College Publications.
[Sh:300g] , Universal classes: Changing the framework, Chapter V (G).
[GbSh:880] Ruediger Goebel and Saharon Shelah, Absolutely Indecomposable Modules, Proceed-
ings of the American Mathematical Society 135 (2007), 16411649, 0711.3011.
[Sh:F956] Saharon Shelah, On non-L
inf,
0
-equivalent innite powers of a.
[Sh:F957] , Equivalence of power of abelian groups.
[Sh:F1050] , Model Theory with continuity.
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