2

1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II:
COMPARISON WITH OTHER LOGICS
Saharon Shelah
The Hebrew University, Jerusalem Israel
Simon Fraser University, Burnaby, B.C., Canada
Rutgers University, New Brunswick, New Jersy, U.S.A.
Aug 15, 1991
Abstract. We show the ordering of the Hanf number of L
,
(wo), (well order-
ing) L
c
,
(quantication on countable sets), L
,
(aa) (stationary logic) and second
order logic, has no more restraints provable in ZFC than previously known (those
independence proofs assume CON(ZFC only). We also get results on corresponding
logics for L
,
.
The author would like to thank the BSF and NSREC for partially supporting this work. Publ
number 211
Typeset by A
M
S-T
E
X
1
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


2 SAHARON SHELAH
0 Introduction.
The stationary logic, denoted by L(aa) was introduced by Shelah [Sh
43]. Barwise, Kaufman and Makkai [BKM] make a comprehensive research
on it, proving for it the parallel of the good properties of L(Q). There has
been much interest in this logic, being both manageable and strong, see [K]
and [Sh 199].
Later some properties indicating its anity to second order logic were
discovered. It is easy to see that coutable conality logic is a sublogic of
L(aa). By [Sh 199], for pairs , of formulas in L
,
(Q
cf

0
), satisfying
there is an interpolant in L(a, a). By Kaufman and Shelah [KfSh
150], for models of power >
1
, we can express in L
,
(aa) quantication
on countable sets. Our main conclusion is (on the logics see Def 1.1 or the
abstract, on h, The Hanf Numbers, see 1.2)
0.1 Theorem. The only restiction on the Hanf numbers of L
,
(wo), L
c
,
, L
,
(aa), L
II
,
are:
h

L
,
(wo)

L
c
,

L
,
(aa)

L
II
,
, h)
h(L
c
,
) < h(L
II
,
.
Proof See 2.1 (neccessity), 2.2, 2.4, 2.5 and 3.3 (all six possibilities are
consistent).
The independence results are proved assuming CON(ZFC) only and the
results are generalized to L

+
,
. We do not always remember to write down
the inequalities of the form L
,
(Q
1
) < L
,
(Q
2
). For some of the results
when we generalize them to L

+
,
or L
,
we need a stronger hypothesis.
The proofs of the results on h(L
1
) h(L
2
) give really stronger information:
we can interpret L
1
in L
2
, usually here by using extra predicates, i.e.,every
formula in L
1
is equivalent to a formula in (L
2
); remember (L
2
) is
dened by: (L
2
)() is represented by (
1
,
2
),
e
L
2
(
e
),
1

2
=
, M i M can be expanded to a model of
1
i M cannot be expanded
to a model of
2
( so the requirement on (
1
,
2
) is strong). Note that
this has two interpretation: one in which we allow
1
,
2
to have new sorts
hence new elements, the other in which we do not allow it. We use an
211 2 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS3
intermediate course, we allow this but the number of new elements are the
power set of the old. But for L
c
,
L
,
(aa), for models of power =

0
we do not need new elements.
We thank Matt Kaufman for discussions on this subject.
Notation. Let cardinals be denoted by , , ,
Ordinals are denoted by , , , , , i, . is a limit ordinal.
Let H() be the family of sets whose transitive closure has cardinality
< (so for regular it is a model of ZFC , i.e., ZFC

except the power

set axiom: and for a strong limit a model of ZC.
Let Levy(, ) = f : f a function from some < into
Levy (, <) = f : f a partial function from to , [Domf[ <
, f(, ) < 1 +.
Notation on Logics. : L will be a logic, a vocabulary (i.e., set of
predicates and fuction symbols, always with a xed arity, usually nite).
We assume that L() is a set of formulas, each with < Oc
1
(L) free variables
and < Oc(L) predicates and function symbols; L() is closed under rst
order operations, substitutions and relativizations and L() is a set (with
and the the family of variables sets)
Two formulas are isomorphic if some mapping from the set of predicates,
function symbols and free variables of one onto those of another is one-to-
one and map one formula to the other.
We are assuming that up to isomorphism there is a set of L-formulas,
this number is denoted by [L[.
Let L
1
L
2
mean L
1
() L
2
() for every vocabulary .
211 3 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


4 SAHARON SHELAH
1 Preliminaries.
1.1 Denition.
(1) L
,
is the logic in which
iI
([I[ < ) and (x
0
, . . . , x
i
. . . )
iJ
([J[ <
) are allowed, with Oc
1
(L
,
) = (so L
,
is rst order logic)
(2) For a logic L, L(wo) extends L by allowing the quantier (wo x, y)(x, y)
saying x : (x, y), (x, y)) is well ordering
(3) For a logic L, L
c
= L(
c
) extends L by allowing a monadic predicate
as free variable and the quantier (
c
X)(X) saying there is a
countable set X such that (X)
(4) For a logic L, L(aa) extends L by allowing monadic predicates as
free variables and the quantiers (aaX)(X) saying that the collec-
tion of countable X satisfying contains a closed unbounded family
of countable subsets of the model
(5) For a Logic L, L
II
= L(
II
) extends L by allowing binary predicates
as free variables and the quantiers R(R)saying there is a two-
place relation R on the model satisfying R
(6) For Q
c
, aa,
II
, L

(Q) is dened similarly allowing a string

(Qx
1
. . . Qx
i
. . . )
i<
, [[ < 0c
1
(L)
(7) Let L
c
= L(
c
), L
wo
= L(wo), L
II
= L(
II
), L
aa
= L(aa)
1.2 Denition.
(1) For a sentence Let h() = sup[M[
+
: M [=
(so it is a cardinal (or innity)) and it is the rst such that
has no model )
(2) For a theory T, h(T) = h(
T
)
(3) For a logic L let h(L) = suph() : h() < , L() for some
vocabulary
(4) For a logic L and cardinal let h(L, ) = suph() : for some
vocabulary of power < , L(), h() <
(5) For a logic L and cardinal let hth(L, ) = suph(T) : for some
vocabulary of power < , T L(), h() <
hth(L) = H(L, )
211 4 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS5
1.3 Claim.
(1) for every L for some L, [h() < h() < h() < ]
(2) h(L) is strong limit
(3) If L is closed under
<
0
for
0
< then cf[h(L)]
(4) If the number of sentences in L (up to isomorphism) is then
cf[h(L)]
1.4 Lemma. assume L is a logic L
II
,
and there is a function f from
Card to Card such that:
(a) f is denable in L
II
,
, i.e., the class of two sorted models , f())
is denable by some sentence of L
II
,
or even just
(a)

For some

< h(L
II
,
) and

L
II
,
for , ,

, < h(L
II
,
)
we have , )

i = f()
(b) If L has a model of power then has a model M,
[[M[[ f()
(c) L is denable in L
II
,
i.e., the class (, , M) : L(), Ma -
model, M is denable by a sentence in L
II
,
(d) For < h(L), f() < h(L)
Then h(L) < h(L
II
,
)
Proof Easy. Let
0
L
II
,
be such that

< h(
0
) < , where

are as in (a)

. We can assume h(
0
) < h(L) (otherwise the conclusion is
trivial). Let L
II
,
say that for some ,
0
:
(i) the model M is isomorphic to some (H(), ), strong limit,
(ii) for every < , M [= ( )[
0
has a model of cardinality
] ( )[, ) [=

]
(iii)
0
< ,
0
has a model of power whose cardinality is in the interval
(
0
, )
(iv) for every < ,
0
, there is L which has a model of
cardinality in the interval (, ), but for some

(, ) has no
model of cardinality in the interval (

, )
Now (H(h(L)), ) is a model of and it has no models of larger cardinality.
211 5 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


6 SAHARON SHELAH

1.4
We can prove similarly:
1.5 Lemma. Suppose L
1
, L
2
are logics and there is f : Card Card such
that
(a) for some

< h(L
2
) and

L for ,

we have: , ) [=

i = f()
(b) if L
1
has a model of cardinality then has a model M,
[[M[[ f()
(c) L
1
is denable in L
2
just in the following weaker sense: for K
1
=
(, ) : L
1
(), K
2
= M, , ) : M [= , L
1
() there
are
e
L
2
.
(x)[x K
e
for some , some expansion of (H(), , x) satis-
es
e
] and for every x : some expansion of (H(), , x) satises

e
is a bounded family of cardinals
(d) For < h(L
1
), f() < h(L
2
)
(e) [[L[[ < h(L
2
), L
II
,
L
2
Then h(L
1
) < h(L
2
)
Remark. Of course if 1.5 is hypothesis holds for L
1
(and L
2
) then the
conclusion holds for L

1
, L

2
whenever L

1
L
1
and L
2
L

2
.
1.7 Lemma.
(1) If M , L
wo
then this is preserved by any forcing, this holds
even for L
wo
,
(2) If M , L
aa
,
then this is preserved by any
1
-complete forcing
this holds even for L
aa
,
(3) If M , L
c
,
this is preseved by forcing not adding new
countabale subsets of [M[ (this holds even for L
,
1
)
(4) If M , L
,
, regular, then this is preserved by forcing by
P where P does not add sequences of ordinals of lenght < . If P
is
1
-complete this holds for L
aa
,
.
(5) Suppose V
1
, V
2
are models of set theory (with the same ordinals),
V
1
V
2
, and letting = h(L)
V
1
where L is L
wo
,
or L
c
,
or L
c
,
,
211 6 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS7
(just a suitable downward Lowenheim Skolem theorem is needed).
If A : A bounded, A V
1
= A : A bounded, A V
2

then h((L)
V
1
= h(L)
V
2
.
Proof Left to the reader.
2 Independence for L
c
,
, L
II

.
In this section we shall deal with the indepedence of the cases where
h(L
wo
,
) = h(L
c
,
).
2.1 Lemma.
(1) For any logic L : h(L(wo)) h(L
c
) h(L(aa)) h(L
II
)
(2) For any logic L we have h(L
c

) < h(L
II
,
)
(3) For any logic L we have h(L
c

+
,
) < h(L
c

+
,
), moreover:
if < h(L
II
,
) then h(L
c

+
,
) < h(L
II
,
)
Proof
(1) By Kaufman and Shelah [KfSh 150, Theorem 4.1]; only L = L
,
is discussed there, but it makes no dierence, the non trivial part
is h(L
c
) h(L
aa
);
(2) See [KfSh 150];
(3) Use 1.5 for the function f : f() = (

0
)
+
2.2 Lemma.
(1) If V = L then h(L
wo
,
) = h(L
c
,
) < h(L
aa
,
) = h(L
II
,
)
(2) If V = L, then for any logic L, h(L
wo
) = h(L
c
) h(L
aa
) =
h(L
II
).
Proof
(1) See [KfSh 150]
(2) Same proof.
211 7 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


8 SAHARON SHELAH
2.3 Fact. For a regular cardinal and L
aa
,
the following are equiv-
alent:
(i) for every large enough
Levy(,)
has a model of power
(ii) for some -complete forcing Q we have:
Q
has a model of power
.
Proof Easy; (i) (ii): as Levy(, ) is a -complete forcing notion, (i)
is a particular case of (ii). (ii)(i) let Q be a -complete forcing notion
such that
Q
has a model of cardinality . Let be such that
> [Q[,
Q
has a model of cardinality but and =

. In
(V
Q
)
Levy(,)
has a model of cardinality by 1.7(4).
But (V
Q
)
Levy(,)
is V
Levy(,)
. (see e.g. [Kun]).
2.3A Notation. Let
0
[, ] be the rst cardinal satisfying 2.3(i), if
one exists, and otherwise.
2.4 Lemma.
(1) In some forcing extension of L, h(L
wo
,
) = h(L
c
,
) < h(L
aa
,
) <
h(L
II
)
(2) Moreover for < h(L
II
,
), we have h(L
aa
,
) < h(L
II
,
)
2.4A Remark. If we want to have: < h(L
aa
,
) h(L
c
,
) < h(L
aa
,
), we
should dene
i+1
= h(L
c

+
i
,
)
+
.
Proof Start with V = L. Let

L
aa
,
a sentence such that h(L
c
,
) <
h(

) < be chosen later. Let

0
> h(

) be regular,

< h(L
II
,
). We
dene an iterated forcing P
i
, Q

j
: i , < ) and cardinals
i
such
that:
(a) the iteration is with set support (so P

is a class forcing)
(b)
i
is regular cardinal
(c)
i


<i

, and
i
is the rst regular cardinal

j<i
(
j
+
j
)
+
(when i > 0)
211 8 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS9
(d) Q
i
( V
P
i
) is
i
-complete
(e) Let
i

: <
i
be the set of all L
aa

i
,
i
sentences (up to isomor-
phism) in V
P
i
.
We dene in V
P
i
, Q
i
to be Levy(
i
,
i
) where
i
is the successor of sup
0
[,
i
]
V
P
i
:
L

i
,
i
and so
i+1
=
+
i
.
Our model is V
P

. Clearly the
i
are not collapsed (as well as limits
of
i
and <
0
) and other successor cardinals
0
are collapsed. So in
V
P

, for regular
0
, if L
aa
,
has a model of cardinality then
it has a model of cardinality . As clearly h(L
II
,
) >
0
, we get by 1.4
h(L
aa
) < h(L
II
) (as well as (2)).
By the Lowenheim Skolem theorem, using 1.7(5) for L
wo
,
or
L
c
,
, h() does not change (being or <
0
) hence (in V
P
) h(L
wo
,
) =
h(L
wo
,
)
V
; h(L
c
,
) = h(L
c
,
)
V
. Hence (in V
P

) h(L
wo
,
) = h(L
c
,
) as this
holds in L.
We still have to choose

L
aa
,
and prove that in V
P

we have
h(L
c
,
) < h(L
a,a
,
). There is

L
aa
,
, L h(L
c
,
) < h(

) < (by
2.2).
Clearly for any such

, V
P

h(L
c
,
) < h(

) (as no new subset

of h(

) is added), but we need also V

P

h(

) < ; but checking

the sentneces produced in [KfSh 150] proof of Theorem 4.3 (for proving
L h(L
aa
) = h(L
II
)), they are like that. So V
P

h(L
c
,
) < h(L
aa
,
).
2.5 Lemma.
(1) In some forcing expresion of L we have h(L
wo
,
) = h(L
c
,
) =
h(L
aa
,
) < h(L
II
,
)
(2) In fact for any logic L we have h(L
wo
) = h(L
c
) = h(L
aa
)
(3) For < h(L
II
,
) then, h(L
aa
,
) = h(L
II
,
).
Proof We start with V = L. We dene a (full set support) iteration,

Q = P
i
, Q

i
: i an ordinal )(Q

i
a P
i
name) and cardinals
i
such that
(a)
i
is regular
1
+[P
i
[ for i limit
i
= (

j<i

i
)
+
(b) Q

i
is
i
-complete
211 9 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


10 SAHARON SHELAH
(c) if i is even, G
i
P
i
the generic set (remember Q

i
V
P
i
) then let
the set of elements of P
i
be listed as p
i

: <
i
, and Q

i
will be
the product of the Levy collapses of

i
+4+2+m
to

i
+4+1+m
for <
i
such that: [p
i

G
i
m = 0] and [p
i

, G
i
m = 1].
Let
i+1
=

i
+
i
+1
(d) if i is odd, let
i

: <
i
list all sentneces of L
aa

i
,
i
in a rich
enough vocabulary of cardinality
i
). For each if there is a
i
-
complete forcing notion Q (which is a set) and (in V
P
i
)
Q
there
is a model of
i

of cardinality
i
then let
i

be such that

Levy(
i
,
i

)
i

has a model of cardinality

i
; otherwise
i

=
i
.
Note that
i

exists by 2.3.
Let Q
i
= Levy(
i
, <
i+1
) where
i+1
= (
i
+

<
i

)
++
.
Let G

be generic over V and V [G

] be our model. Note in V [G

],
(*) [i odd
i+1
=
+
i
]
[i even
i+1
=
+(
i
+1)
i
]
[i limit
i
= (

<i

)
+
].
For =
2+1
, if L
aa
,
has a model of cardinality then it
has a model of cardinality (by 2.3 + 1.7(4)). By (*) we deduce that
V
P

if L
aa
,
has a model of cardinality > then it has a model
M, < [[M[[ <

+.
So 1.5 is applicable to show h(L
aa
,
) < h(L
II
,
) (and by 1.6 and 1.7)
also 2.5(3) holds.
Why h(L
wo
,
) = L
a,a
,
?. Let

describe (L

, G

i<
P
i
).
If M

, then for some and G, M

= (L

, , G), so without loss of

generality equality holds. Now if < [[, M is a [regular] cardinal of
L i is a [regular] cardinal of L. Also we know that for every ordinal ,
if in L,
2i

<
2i+1
, divisible by four then forcing by P

collapses
at most one of the cardinals
+1
,
+2
,
+3
,
+4
of L; if
2i
<

2i
+
2i
then exactly one.
We assume

say so, and so when

L
+4
[[ the answer in M to the
question which of
+1
,
+2
,
+3
,
+4
is collapsed is the right one. So
when
2i+1
< [[, we can in M reconstruct G

P
2i
(see choice of Q
2i
).
211 10 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS 11
But V
P

2i+1

2i
(+1)+1
and
2i+2
= (
2i+1
)
+
and for limit
we have

= (

i<

i
)
+

The rest is as in [KfSh 150] proof of 4.3

211 11 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


12 SAHARON SHELAH
3 h(L
wo
,
) is O.K. but for h(L

3
,
) large cardinals are needed and
sucient.
In section 2 we deal with the three cases for which h(L
wo
,
) = h(L
c
,
).
Here we deal with the three cases where h(L
wo
,
) < h(L
c
,
). The new part
is Lemma 3.2, and then, in 3.3 we get the desired conclusion. For dealing
with L

+
,
we do not assume CON(ZFC) alone, we assume the existence
of a class of large cardinals (weaker than measurability). By 3.4 at least if

3
+ (2

0
)
+
, something of this sort is necessary.
3.1 Fact. : The following are equivalent for L
wo
,
or even L
wo
,
:
(i) for every large enough
Levy(
0
,<)
h(() =
(ii) for some (set) forcing notion P we have
P
h() = .
Proof similar to the proof of 2.3
3.1A Notation. Let the rst satisfying (i) be
1
() (and
0
if there is no
such ).
3.2 Lemma. (V = L).
(1) For some (set) forcing notion P

P
h(L
wo
,
) < h(L
c
,
)
and this is preserved by h(L
wo
,
)
+
-complete forcing.
(2) In (1) we can use Levy(
0
< ) for some > cf =
0
(3) We can use instead Cohen() = f : f a nite function from to
(0, 1). So cardinals are not collapsed
Proof 1) Let

= sup
1
() : L
wo
,

We now dene a nite support iteration P

i
, Q

n
: n < ) and
n
as
follows:

0
=

Q
0
= Levy(
0
,
0
)
for n 0,
n+1
is h(L
wo
,
)
V
P
n
211 12 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS 13
Q
n
= Levy(
0
,
n
).
Let = (

n
). Note that P

satises the
+
c.c.
Now V
P

is our model. Note

(*) V
P

G.C.H. +
1
=
+
, andV = L[R, <] for any well < ordering
of R.
Note that in B = ( T())
V
P

; o, +, , ) we can dene by rst order

formulas (representing ordinals by well ordering of ):
(a)
n

n
(maximal countable ordinal which is a cardinal in L

+
(b) L

+ hence
n
: n < ) (by induction remembering the Lowenheim
Sholem theorem) hence the iteration (really we can omit this as P

is just Levy (
0
, < ))
(c) the set R

=
def
r R : for some n, and G P
n
generic over
V, r V [G].
And for r R

(d) H
r
= L
wo
,
: L[r] h() < as it is equal to
L
wo
,
: L[r] h() <
n

n
.
[Note that P

n
s are homogeneous, hence h() does not depend on G P
n
]
So by 3.1 and the choice of
0
, we can dene in that model B
H

= L
wo
,
: h()
V
P

<
[How ? it is H
r
: r R

,remembering 3.1]
Let = h(L
wo
,
) (in V
P

).
Now we dene a sentence L
c
,
: it just describes (H(), ): it says
(i) enough axioms of ZFC holds
(ii) every countable bounded set of ordinals is represented
(iii) on every innite cardinal there is a model M

with universe
satisfying some H

(which we have shown is denable in any

model M of )
So we have proved the rst assertion from 3.2. Now -complete forcing,
preserve trivially h() as it preserves satisfaction for L
wo
,
. It pre-
serves h() < as this is equivalent to h() < , the forcing adds
no new model power < , and Lowenheim Skolem Theorem nishes the
argument.
211 13 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


14 SAHARON SHELAH
2) We have proved it in the proof of (1)
3) A similar proof, replacing
1
() by

1
= rst such that
Cohen()
h[[ = if there is one
0
otherwise.

3.2
3.3 Conclusion. for some forcing extensions of L:
(1) h(L
wo
,
) < h(L
c
,
) < h(L
aa
,
) = h(L
II
,
)
(2) h(L
wo
,
) < h(L
c
,
) = h(L
aa
,
) < h(L
II
,
)
(3) h(L
wo
,
) < h(L
c
,
) < h(L
aa
,
) < h(L
II
,
).
Proof: Combine 3.2 with 2.
3.4 Claim. (0
#
): For
3
+ (2

0
)
+
we have h(L
wo
,
) = h(L
c
,
).
Remark. : The logics are essentially equivalent.
Proof If L
wo
,
says M is, for some , (L

[A], ) (up to isomorphism),

> 2

0
, A 2

0
, every subset of is in L
(2

0)
[A], and
2
, and
<
2
: cf =
0
in L

2
[A] = <
2
: cf =
0
then by Jensens
covering lemma [ < [[ every countable subset of is represented in
the model
3.4
211 14 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS 15
3.5 Claim. Suppose that:
(*) for every for some , (
1
)
<

or even just
(**) for every for some ,
BG
(c)
<

, which means: for every

f : []
<
for some
n
: n < ) for every <
1
, for some
Y , Y has order type and
n
(w [Y ]
n
)[
n
= f(w)] .
Then for every , h(L
wo

+
,
).
3.5A Remark.
(1) The property (**) was discoverd by Baumgartner and Galvin [BG]
such that:

BG
(c)
<

i h(L
wo

+
,
).
(2) See [KfSh 150, 4.2] (for = )
Proof There is a sentence L
c
,
such that for : there is a model
M, [[M[[ = , [P
M
[ = , i ( < )
BG
(c)
<

.
3.5
On K = K
V
(the core model of V) see Dodd and Jensen [DJ].
3.6 Claim. Suppose V = K, and (**) ( from 3.5 ),then
(1) for every we have h(L
wo

+
,
) < h(L
c

+
,
) < h(L
aa

+
,
) = h(L
II

+
,
)
(2) for every L, h(L
aa
) = h(L
II
).
Proof 1) First inequality by the observation above, the second inequlity
follows from last equality Th 2.1, last equality see ( 2)
(note: if cf >
0
in L
aa
,
we can say for A whether < : cf =

0
, A is a stationary subset of ).
2) As in [KfSh 150]
3.7 Observation. There is L
c
,
such that M i M is isomorphic
to K

for some .
It is known see (see [BG], [DJ])
211 15 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


16 SAHARON SHELAH
3.8 Fact. If in V there are, e.g., measurable cardinals in Card, then K
(**).
3.9 Claim. : Suppose V = K and (**) holds.
For some forcing extension V [G

] of V, V [G

] () and for every

, h(L
wo

+
,
) < h(L
c

+
,
) < h(L
aa

+
,
) < h(L
II

+
,
).
Proof Similar ot 2.4(1) except that we want to preserve (**). We dene
by induction on an iterated forcing, P
i
, Q
j
, < ) with set support
and cardinals
i
increasing such that:
(i)
0
=
2
(ii)

= (

i<

i
+[P

[)
+
(iii) if
i
, P
i
are dened, let
i
be
+
i
+
0
[,
i
] : L
aa

i
,
i
.
Q
i
= Levy (
+
i
,
+
i
)(inV
P
i
) and
i+1
is minimal such that

i+1

BG
(c)
<

++
i
and
i+1
h(L
c

+
i
, ).
We leave the rest to the reader.
3.10 Claim. Suppose V = K and (**) holds.
For some forcing extension V [G

] of V, V [G

] () (hence the con-

clusion of 3.7) and for every
h(L
wo

+
,
) < h(L
c

+
,
) = h(L
aa

+
,
) < h(L
II

+
,
)
Proof Combine the proofs of 3.9 and 2.5.
4 Lowering consistency strength.
We present here some alternative proofs with lower consistency strength
than in 3. Specically 4.1, 4.3 and 3.2(3) justfy the restriction

3
+ (2

0
)
+
in 3.4].
211 16 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS 17
4.1 Lemma. Let V = L. Then there is a forcing notion P L, not adding
reals, such that for G P generic over V , in V [G]:
a) h(L
wo

1
,
) < h(L
c

1
,
)
b) No
1
-complete forcing notion (or even forcing notion satisfying the
I-condition I a set of
V [G]
2
-complete ideals from L changes the truth
value of h() < for L
wo

1
,
c) There is a sentence L
c
,
whose class of models of power
2
is just L

[G] :
2
( and note P L

2
[V [G]])
d) h(L
c

+
,
) < h(L
aa

+
,
) = h(L
II

+
,
)
4.1A Remark. In the proof below, coding generic sets by the decision which
L-cardinals are collapsed is replaced here by which L-regular cardinal have
in V cardinality
0
and which cardinality
1
Proof Let I(, ) be, e.g. the calss of lters D which are -complete over
some (this in V ), where < , [ D[ <
We dene by induction on n,
n
,
n
,
i,j
,
i,j
, P
i
, Q

: i
n
, <
n
)
and f
n
such that
(A)
0
= 0,
n+1
>
n
(B) P
i
, Q

j
,

: i
n
, <
n
) is an RCS iteration suitable for x

n
=
I
i,j
,
i,j
,
i
i,j
, i < j
n
, i not strongly inaccessible ).
See [Sh-b Ch.XI] or [Sh-f (Ch XI)] particularly Def. 6.1
(C) f
n
is a one-to-one function from P
i
onto some ordinal
n
, extending

e<n
f
e
.
G

will denote a generic subset of P

.
For n = 0 there is nothing to do.
For n + 1, note that forcing by P

n
does not add new reals. So
(L
wo

1
,
)
V
= (L
wo

1
,
)
V [G

n
]
and let
i
: i <
1
be a lsit of the
sentences (up to isomorphism).
We now (i.e for dening
n+1
etc.) dene by induction on <
1
, Q

n
+
, x

n
++1
as follows:
(a) P
i
, Q

: i
n
+) is x

n
++1
-suitable RCS iteration
211 17 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


18 SAHARON SHELAH
(b) If there is Q

n
+
, a P

n
+
-name of a forcing notion sattisfying the
I(([P

n
+
[ + sup
i,
: i < j
n
+ )
+
, -condition for some
then
P

n
++1
Q

has arbitrarily large models then Q

n
+
is like
that, otherwise it is, e.g., Levy (
1
, 2

1
).
Next let

= h(L
wo

1
,
)
V [G

n
+
1
]
, Q

n
+
1
= Levy (
1
,
+

). Now (where
<, > is Godels pairing function on ordinals), let in V [G

n
+
1
] : A
n
=
[f
n
(p), f
n
(q)) : p, q P

n
p q and p ,= q f
n
(p), f
n
(p)) : q
G

n
+
and let
n
= supf
n
(p), f
n
(q)) : p, q P

n
+
1
. Now we dene
Q

n
+
1
+i
by induction on i
n
:
Q

n
+
1
is Levy (
1
,
2
)
V [G

n
+
1
]
,
Q

n
+
1
+1+2i+1
is Levy (
1
,
2
[V [G

n
+
1
+1+2i+1
]],
Q

n
+
1
+1+2i
is Namba forcing (of V [G

n
+
1
+1+i
]) if i A
n
and Levy
(
1
,

)
V (n,i)
where V (n, i) = V [G

n
+
1
+1+2i
] if i , A
n
.
Now let
n+1
=
n
+
1
+ 2
n
,
n+1
= [P

n
+
1
+2
n
[, and dene f
n+1
.
We leave the rest to the reader
4.1
4.2 Conclusion.
(1) We can do the forcing from 2.4, 2.5 to the universe we got in 4.1
getting corresponding results (for L

1
,
(Q)s, with CH and G.C.H):
so we need CON(ZFC) only.
(2) the same holds for 4.3 for the L

2
,
(Q)s (so we use CON(ZFC+
the class of ordinals in Mahlo) only).
4.3 Lemma. Suppose V = L, (for simplicity) and is a Mahlo cardinal
(i.e., every closed unbounded class of cardinals has a regular member). then
there is an inaccessible cardinal and a forcing notion P H(), such that
:
(a) P satises the -c.c., does not add reals and collapse every
(
1
, ) : and
P
G.C.H. is
2
and [P[ =
(b) h(L
wo

2
,
) < h(L
c
,
)
(c) there is a sentence L
c
,
whose class of models of power
2
is just suitable expansions of L

[G] :
2
.
211 18 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


THE HANF NUMBERS OF STATIONARY LOGIC II: COMPARISON WITH OTHER LOGICS 19
Proof Like 4.1, but instead of induction on n < we do induction on
< , and in the induction only we rst do the coding (Q

n
+
1
+i
, i < )
(so that for c), we say that for some club of C of
2
, for C we are
coding the set of sentence in L
||
+[G P

].
Do we really need the large cardinal hypothesis in 4.3 (and so in 4.2(2))?
4.4 Claim. Suppose 0
#
, V and
V
2
is a successor cardinal in L and
2

0
=
1
then for some sentence L
wo

2
,
, its models are exactly suitable
expansions of (L

, T
<
1
()), where is the last L-cardinal <
V
2
.
Hence h(L
wo

2
,
) = h(L
c

2
,
).
Proof Should be clear
4.4 Concluding Remarks. : Still we do not settle the exact consistency
strength. In fact e.g. if
V
2
is the rst L-inaccessible, we can still prove the
last sentence of 4.4.
For h(L
wo

2
,
) < h(L
c

1
,
) with 2

0
=
1
we can generalize Lemma 4.3
to this case (using [Sh-f, XV]).
Also there is a gap in consistency strength in 3 for >
3
+ (2

0
)
+
.
It is not hard to show that if
2
+2

0
, cf >
0
and for some A
does not exists, then h(L
wo

+
,
) = h(L
c

+
,
)
211 19 6.10.2003
2
1
1


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
3
-
1
1







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
1


20 SAHARON SHELAH
References
[[BKM] J.Barwise, M.Kaufman and M.Makkai,] Satationary Logic, Annals of Math Logic 13, 171224.
[[BKM]] A correction to Stationary Logic, Annals of Math Logic 20, 231232.
[[BG] J. Baumgartner and F. Galvin] Annals of Math Logic.
[[DJ] Dodd and Jenssen] The core model.
[[K] M.Kaufman] Model Theoretic Logics, J. Barwise and S. Feferman 166 (6.12).
[[Kun] K. Kunen] Set theory, An Introduction to Independence Proofs, Studies in Logic and the Foun-
dation of Math 102.
[[KfSh 150] M. Kaufman and S. Shelah] The Hanf Number of Stationary Logic, Part I, Notre Dame J. of Formal Logic 27,
111123.
[[Sh 43] S. Shelah] Generalized Quantiers and Compact Logics, Trans. Amer. Math Sci. 204, 342364.
[Sh b] S. Shelah, Proper Forcing, Springer Verlag, Lectures notes.
[Sh 199] S. Shelah, Remarks in Abstract Model Theory, Annals of Pure and Applied
Logic, no. 1985.
[Sh f], Proper and improper forcing, Springer Verlog, in prepint.
[V] J. Vaananen, On the Hanf numbers of unbounded logic.
211 20 6.10.2003