You are on page 1of 45

(

7
7
5
)


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







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


MIDDLE DIAMOND
SH775
Saharon Shelah
The Hebrew University of Jerusalem
Einstein Institute of Mathematics
Edmond J. Safra Campus, Givat Ram
Jerusalem 91904, Israel
Department of Mathematics
Hill Center-Busch Campus
Rutgers, The State University of New Jersey
110 Frelinghuysen Road
Piscataway, NJ 08854-8019 USA
Abstract. Under certain cardinal arithmetic assumptions, we prove that for every
large enough regular cardinal, for many regular < , many stationary subsets of
concentrating on conality has the middle diamond. In particular, we have the
middle diamond on { < : cf() = }. This is a strong negation of uniformization.
Key words and phrases. Set theory, Normal ideals, Middle Diamond, Weak Diamond.
I would like to thank Alice Leonhardt for the beautiful typing.
This research was partially supported by the Israel Science Foundation.
Pub. 775
First Typed at Rutgers - 6/13/01
Latest Revision - 04/Oct/11
Typeset by A
M
S-T
E
X
1
(
7
7
5
)


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







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


2 SAHARON SHELAH
Annotated Content
0 Introduction
[We state concise versions of the main results: provably in ZFC there are
many cases of the middle diamond, for many triples (, , ) where <
are regular, 2 < we can give for any S

a function f

from
some subset C

of of order type unbounded in into such that any


f : extends f

for stationarily many s. We also review some


history.]
1 Middle Diamond: sucient conditions
[We dene families of relevant sequences

C = C

: S) and properties
needed for phrasing and proving our results (see Denition 1.5) and give
easy implications; (later in 1.22, 1.25 we dene other versions of middle dia-
mond). Then (in 1.10) prove a version of the middle diamond for (, , , )
guessing only the colour of f C

when = cf(2

), the number of possible


colours, , is not too large and belongs to a not so small set of regular
cardinals < (e.g., every large enough regular <

if

). The
demand on says 2

is not the number of -branches of a tree with < 2

nodes. We then prove that a requirement on (, ) is reasonable (see 1.11,


1.9, 1.13). In 1.20 we get the version of middle diamond giving f C

from
the apparently weaker one guessing only the colour of f C

. We quote
various needed results and denitions in 1.3, 1.24.]
2 Upgrading to larger cardinals
[Here we try to get such results for many large enough regular . We
succeed to get it for every large enough not strongly inaccessible cardinal.
We prove upgrading: i.e., give sucient conditions for deducing middle
diamond on
2
from ones on
1
<
2
(see 2.1) and 2.3 (many disjoint ones),
2.10, 2.11). We deduce (in 2.4) results of the form if is large enough
satisfying a weak requirement, then it has middle diamond for quite many
, , using upgrading claims on the results of 1 and generalization of [EK]
(see 2.5, 2.6, 2.7). We deal with such theorem more than necessary just for
the upgrading.]
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 3
0 Introduction
Using the diamond (discovered by Jensen [J]) we can construct many objects;
wonderful but there is a price - we must assume V = L or at least instances
of G.C.H. (see Gregory [Gr], Shelah [Sh 108]). In fact =
+
= 2

>

{<:cf()=}
for every regular large enough depending on (see [Sh 460] and
citations there).
Naturally we would like to get the benet without paying the price. A way to do
this is to use the weak diamond (see [DvSh 65], [Sh:f, AP,1], [Sh 208], [Sh 638] and
lately [Sh 755], see on the denable version in [Sh:f, AP,3], [Sh 576]).
In the diamond principle for a stationary S , for S we guess for

2.
In the weak diamond, for stationary S for S we do not guess for


2, we just guess whether it satises a property - is black or white, i.e., the value
of F( ) 0, 1 for a pre-given F. Many times it is just enough in constructions
using diamond to avoid undesirable company: whatever you will do I will do
otherwise. This is, of course, a considerably weaker principle, but for
1
it follows
from CH (which does not imply

1
) and even WCH (i.e. 2

0
< 2

1
). In fact, it
holds for many cardinals: 2

> 2
<
= 2

suces. Having F with two colours (=


values) seemed essential. In [Sh 638], we get more colours for
1
if the club lter
on
1
is
2
-saturated but this is a heavy assumption. In [Sh 755], if =
<
= 2

,
it is proved that for many regular < , if S < : cf() = is stationary
then we have the weak diamond for S. Now what does many mean? For example,
if

, many include every large enough regular <

; note that it is for


every stationary S unlike the situation, e.g. for
1
under CH (see [Sh:f, V]).
We get there more cases if the colouring F on

2 depends only on C

, where
C

= sup(C

) and C

is small.
Here we get more than two colours and moreover: we guess C

, C

=
sup(C

) and so can choose M

: S) such that M

is a model with universe


C

and relations only (and e.g. vocabulary H (


0
)) such that if M is a model
with universe and relations only (and vocabulary H (
0
)) then for stationarily
many S, M C

is M

. Of course, if ( < )[2

> ] then we cannot have


C

= so this is a very reasonable restriction. (It would be even better to have


M

M below but we do not know). Also for almost all large enough the
result holds for many regular .
The motivation has been trying to get better principles provably in ZFC.
We use partial squares and I[], see [Sh 108], [AShS 221], [Sh 237e], [Sh 420, 1]
and [DjSh 562]. We prove (in fact this holds for quite many stationary S S

).
0.1 Theorem. For every strong limit of uncountable conality, for every regular
> which is not strongly inaccessible, for every large enough regular < , for
every < , S

has the -middle diamond (called later has the (,


+
)-MD
2
-
(
7
7
5
)


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







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


4 SAHARON SHELAH
property), which means:

,
for some stationary S < : cf() = for every relational vocabulary
of cardinality < (so has no function symbols) we can nd

M = M

:
S) which is a -MD-sequence
(a) M

is a -model
(b) the universe of M

is an unbounded subset of of order type


including
(c) if M is a -model with universe then for stationarily many S
we have M

M.
It seems that this is stronger in sucient measure from the weak diamond to justify
this name; there are many properties between the weak diamond and the property
from 0.1 and we have to put the line somewhere. We shall use the adjective middle
diamond to describe even cases with colouring with more than two values (justied
by 1.20).
I thank Todd Eisworth and the referee and Adi Yarden for their helpful comments.
Some improvements and elaborations are delayed to [Sh:F611]. This work is con-
tinued in [Sh 829].
0.2 Theorem. In 0.1 we can have also

f = f

: S) where f

: u

, u


,
|u

|
= such that:
(d) if M is a -model with universe and f : and u ,
|u|
= , then
for stationarily many S we have M

M & (f

= f u f) & u

=
u.
Proof of 0.1 and 0.2. By 2.9(1) if > is regular not strongly inaccessible, then
for every large enough regular < the following holds (see Denition 1.19(4)):
()

if < then some (, ,


+
)-parameter

C has the -Md property; let

C = C

: S).
Now without loss of generality S ( )

[, and let C

= : < or
( C

)( < + ).
Lastly, let = 2
++||
and by 1.20 we get the conclusion of 0.1.
As for 0.2, Clause (d) is easy by having a pairwise disjoint sequence of suitable
S
i
: i < ) which holds by 2.9(2).
0.1

0.2
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 5
Note that having M

s is deduced in 1.20 from



C has the -Md-property with
large enough. Note that 0.1 has obvious applications on various constructions,
e.g., construction of abelian groups G with Ext(G, K) ,= .
Note that in 0.1, 2.9 instead of using large enough < . . . , strong limit
and quoting [Sh 460] we can specically write what we use on (and on ). Also
in 0.1, if we omit cf() >
0
(but still is uncountable) the case successor is not
aected, and in the case =

when
2
divides , the statement instead every
large enough can be replaced by for arbitrarily large regular < , or even:
for arbitrarily large strong limit

< , for every large enough regular <

.
A consequence is (see more on this principle in [Sh 667]).
0.3 Conclusion. 1) If , , , S, M

: S) are as in 0.1 and < < , then


we can nd =

: S),

is an increasing sequence of length of ordinals


< with limit such that:
if F : , then for stationarily many S we have i < F(

(2i)) =
F(

(2i + 1)).
2) In fact it is enough to nd a (, , )-Md-parameter C

: S) which has the


(, 2
||
+
)-MD-property (see Denition 1.6(1)).
Proof. 1) By (2).
2) Let = P

: < [[
+
, p
i
a monadic predicate. Let M

: S) be as in
0.1 for . Let F : and we let M be the following -model with universe
: P
M

= : = pr(
0
,
1
),
0
<
1
< [[
+
and F( +
0
) = F( +
1
),
M

+[[
+
< Min(M

). Without loss of generality otp([M

[) = , S
[[
+
/. Let (

(2, i),

(2i +1)) = (

i
+

2
,

i
+

2
) when

i
is the i-th member of
M

i
P.
Note that if a natural quite weak conjecture on pcf holds, then we can replace

by

(and similar changes) in most places but still need >


1
>

, 2

1
> 2

0
2

in the stepping-up lemma; e.g. for every regular large enough , for some n, we
have middle diamond with
0
colours on S

n
- is helpful for abelian group theory.
(Under weak conditions we get this for
1
, see [Sh:F611].
Even without this conjecture it is quite hard to have the negation of this state-
ment as it restricts 2

and cf(2

) for every !
(
7
7
5
)


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







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


6 SAHARON SHELAH
0.4 Denition. 1) Let S has square mean that is regular uncountable, S
a stationary subset of and: for some club E of and S

, S E S

there is
a witness

C = C

: S

) which means:
()

C = C

: S

) has the square property which means


(a) C

is closed,
(b) S

limit = sup(C

),
(c) C

= C

and
(d) S otp(C

) < .
1A) We say S has a strict square if we add
(e) S E otp(C

) = cf().
2) We say S has a (

) square if we replace (d) above by


(d)

S otp(C

.
If we replace square by (

)-square this means we add otp(C

.
3) We say ( , )-square if above the demand C

= C

is restricted
to the case cf() .
4) S

= < : cf() = .
On I[] see [Sh 420, 1].
0.5 Denition. 1) For regular uncountable let I[] be the family of sets S
which have a witness (E,

P) for S I[], which means
() E is a club of , P = P

: < ), P

P(), [P

[ < , and for every


E S there is an unbounded subset C of of order type < such that
C C P

.
2) Equivalently there is a pair (E, a), E a club of , a = a

: < ), a

,
a

= a

and E S = sup(a

) > otp(a

) (or even = sup(a

),
otp(a

) = cf() < .
0.6 Notation. 1) Let , , , , , denote cardinals, a set of cardinals, let
, , , , , , i, j, denote ordinals but 2 and 2 and is a limit ordi-
nal if not said otherwise.
Clearly 1.10 and some results later continue Engelking Karlowic [EK] which
continue works on density of some topological spaces, see more history in [Sh 420].
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 7
1 Middle Diamond: Sufficient conditions for cf(2

)
1.1 Denition. For a regular uncountable , an ordinal 2 and cardinal 2
(1) we say S is (, )-small if it is F-(, )-small for some (, , )-colouring
F, also called a -colouring of
>
, that is: a function F from
>
to
where
(2) we say S is F (, )-small if (F is a -colouring of
>
and)
()

F,S
for every c
S
for some

the set S : F( ) = c

is not
stationary.
(3) Let D
md
,,
= A : A is (, )-small, we call it the (, )-middle
diamond lter on . If = 2 we call this weak diamond lter and may write
D
wd
,
.
(4) We omit if = , writing instead of (, ). We may omit from S is
F -small, and ()

F,S
when clear from the context.
1.2 Claim. 1) If = cf() >
0
, 2 and then D
md
,,
is a normal
lter on .
2) Assume S , 2

, 2 for = 1, 2. Then S is (
1
, )-small i
S is (
2
, )-small
1
.
3) In fact

[2, 2
<
] is enough.
Proof. Straightforward.
1.2
Recall
1.3 Denition. 1) Let
<>
= Min[P[ : P []

and every A []

is
included in the union of < members of P.
2) Let
[]
= Min[P[ : P []

and every A []

is equal to the union of <


members of P.
3)
<>
tr
= sup[lim

(t)[ : t is a tree with nodes and levels.


4) For J an ideal on let U
J
() = Min[P[ : P []

and for every f :


Dom(J) for some u P we have i Dom(J) : f(i) u , = mod J; we
say an ideal J is an ideal on a set Dom(J) of cardinality . If J = []
<
we may
write instead of J.
1
so seems immaterial, but below when using

C this is not so (see Denition 1.5)
(
7
7
5
)


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







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


8 SAHARON SHELAH
1.4 Remark. 1) On
<>
tr
see [Sh 589], on
<>
see [Sh 430], on
<>
,
[]
see
there and in [Sh 460] but no real knowledge of these references is assumed here.
2) Note that when = only F

: S is relevant in Denition 1.1 because


for any

for some club E of we have E

. So we can restrict
the colourings to this set.
Below we may usually restrict ourselves to = . But if < = cf(), 2

= 2

even for = = 2, the set is small (as noted by Uri Abraham, see [DvSh 65]).
1.5 Denition. 1) We say

C is a -Md-parameter if:
(a) is regular uncountable,
(b) S is a stationary subset of ,
(c)

C = C

: S) and
2
C

= sup(C

).
1A) We say

C is a (, , )-Md-parameter if in addition ( S)( C

)[cf() =
[C

[ < ].
2) We say that F is a (

C, , )-colouring if: 2, 2,

C = C

: S) is a
-Md-parameter and F is a function from
>
to such that:
if S and
0
,
1


and
0
C

=
1
C

then F(
0
) = F(
1
).
2A) If = we may omit it and write (

C, )-colouring.
2B) In part (2) we can replace F by

F = F

: S) where F

:
(C

)
such
that

S F() = F

( C

). If = we may use F

:
(C

)

(as we can ignore a non-stationary S

S); so abusing notation we may write


F( C

). Similarly if cf() = ,

: < ) is increasing continuous with limit


we may restrict ourselves to
(C

)
(

).
3) Assume F is a (

C, , )-colouring, where

C is a -Md-parameter.
We say c
S
(or c

) is an F-middle diamond sequence, or an F-Md-sequence
if:
() for every

, the set S : F( ) = c

is a stationary subset of .
We also may say that c is an (F, S)-Md-sequence or an (F,

C)-Md-sequence; note
that can be reconstructed from F.
3A) We say c
S
is a D F-Md-sequence or D (F,

C)-Md-sequence if D is a
lter on (no harm to assume that S belongs to D, or at least S D
+
) and
() for every

we have
S : F( ) = c

, = mod D.
2
we can omit the requirement = sup(C

) but then we trivially have middle diamond if


=
<
when = {|C

|
+
: S}
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 9
4) We say that

C is a good (,
1
)-Md-parameter, if

C is a -Md-parameter and
for every < we have
1
> [C

: S & C

[. If
1
= we may write
a good -Md-parameter. Similarly for a (,
1
, , )-good-Md-parameter and a
(, , )-good-Md-parameter (i.e. adding that

C is a (, , )-Md-parameter.
4A) We say that

C is a weakly good -Md-parameter if it is a -Md-parameter and
for every < we have > [C

: S & nacc(C

)[. Similarly we
dene a weakly good (, , )-Md-parameter and a weakly good (,
1
, , )-Md-
parameter.
4B) We say

C is a good
+
-Md-parameter if

C = C

: S) is a -Md-parameter
and C

1
C

2
C

1
= C

2
. Similarly we dene a good
+
(, , )-
Md-parameter.
4C) We dene a weakly good
+
-Md-parameter or (, , )-Md-parameter anal-
ogously, i.e., nacc(C

1
) nacc(C

2
) C

1
= C

2
.
4D) For a set of regular cardinals we say that

C = C

: S) is -closed if
each C

is; where a set C of ordinals is -closed if for every ordinal , cf() &
= sup(C ) < sup(C) C.
4E) We say (C

, C

) : S) is a good (, , )-Md

parameter if C

: S)
is a (, , )-Md-parameter, C

, sup(C

) = and for every < the set


[C

: C

, S[ has cardinality < .


1.6 Denition. 1) We say that

C (or (,

C)) has the (D, , )-Md-property if
(a)

C is a -Md-parameter
(b) D is a lter on
(c) if F is a (

C, , )-colouring, then there is a DF-Md-sequence, see Deni-
tion 1.5, part (3A).
We may omit D

if it is the club lter on .
2) We say that

C has the (D, )-Md-property as in part (1) only that F is a (

C, , )-
colouring. We may omit D if D is the club lter. We may omit if = 2.
3) We say that S has the (, )-Md-property when some

C = C

: S) has,
and then we say the -Md-property if = .
In Denition 1.5(2),(2A) we replace colouring by colouring
2
if we have all
:
>
(or :
>
(C

) ); and let colouring


1
be the original notion (so
in Denition 1.5(3),(3A) we have new meaning when we use such Fs).
4) We write Md
2
-property if we use colouring
2
F and we write Md
1
-property for
the Md-property.
1.7 Remark. We may consider replacing C

by (C

, D

), D

a lter on C

, that is
F

() would depend on ( C

)/D

only. For the time being it does not make a


real dierence.
(
7
7
5
)


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







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


10 SAHARON SHELAH
1.8 Claim. 1) The restrictions in 1.5(2B) are O.K. That is in 1.5(2B), the rst
sentence is obvious; for the second sentence (if = ) recall that for every

,
for some club E of , E

; and similarly in the third sentence.
2) In Denition 1.5(4), if
1
> then goodness holds trivially.
3) If

C = C

: S) is a weakly good (, , )-parameter, then for some club E


of there is

C

= C

: S E) which is a good
+
(, , )-parameter.
4) If

C = C

: S) has the (D, , )-Md-property, and



C

= C

: S), C

, then also

C

has the (D, , )-Md-property.


Proof. 1),2) Easy.
3) Note that the C

s are not required to be a club of , see [Sh 420, 1] but


we elaborate. Let A

: < ) list without repetition C

: S and
nacc(C

) and be such that A

= A

: . Let
E = < : if < and S and nacc(C

) then C

: < ,
clearly E is a club of .
Lastly, for S E let C

= < : for some nacc(C

) we have
C

= A

. Now C

: S E) is as required.
4) Should be clear.
1.8
From the following denition the case Sep(, ) is used in the main claim below.
1.9 Denition. 1) Sep(, ,
0
, , ) means that for some

f:
(a)

f = f

: < )
(b) f

is a function from

(
0
) to
(c) for every

the set

(
0
): for every < we have f

() ,= ()
has cardinality < .
2) We may omit
0
if
0
= . We write Sep(, ) if for some = cf() 2

we
have Sep(, , , , ) and Sep(< , ) if for some = cf() 2

and < we
have Sep(, , , , ) .
1.10 The Main Claim. Assume
(a) = cf(2

), D is a
+
-complete lter on extending the club lter
(b) = cf() <
(c)

C = C

: S) is a (, , )-good-Md-parameter, or just weakly good or


just (C

, C

) : S) is a good (, , )-Md

-parameter and S D
(d) 2
<
2

and
(e) < 2

[[
<>
tr
< 2

(f) Sep(, ).
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 11
Then

C has the (D, 2

, )-Md-property (recall that this means that the number of


colours is not just 2).
We may wonder if clause (f) of the assumption is reasonable; the following Claim
gives some sucient conditions for clause (f) of 1.10 to hold.
1.11 Claim. Clause (f) of 1.10 holds, i.e., Sep(, ) holds, if at least one of the
following holds:
(a) =

(b) U

() = + 2

,
(c) U
J
() = where for some we have J = []
<
, ,


(d) is a strong limit of conality ,= = cf() <
(e)

().
We shall prove this later.
1.12 Remark. 1) Note that =
<
= , = cf() < , S = S

, C

= is a
particular case (of interest) of the assumptions (a)-(b) of 1.10 and that C

: S

)
is a (, , )-good
+
- Md-parameter.
2) We can replace the requirement S S

by (implicit in (c) of 1.10)


( ) [ S is a regular cardinal].
Then clause (e) of 1.10 is replaced by S
<>
tr
< 2

.
3) We may consider nding a guessing sequence c simultaneously for many colour-
ings. In the notation of [Sh:f, AP,1] this is the same as having large (0). We
may reconsider the problem of
+
-entangled linear ordering where =

0
, see [Sh
462].
4) For = 2, (i.e. weak diamond) the proof below can be simplied: we can forget
F and choose

: P

such that for some i < the sequence

(i) : S) is a weak diamond sequence for F.


5) Note that the minimal natural choices for getting -MD property are =

0
, = 2

0
, = 2

, = cf(2

).
6) Note that in the main claim (1.10) if 2

is regular, = 2

is allowed.
Proof of 1.10. In clause (c) we can assume (C

, C

) : S) is a good (, , )-
Md

-parameter (in the other cases we choose C

= nacc(C

) and use 1.8(3)).


Let F be a given (

C, 2

, )-colouring.
(
7
7
5
)


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







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


12 SAHARON SHELAH
By assumption (f) we have Sep(, ) which means (see Denition 1.9(2)) that
for some = cf() 2

we have Sep(, , , , ).
Let F = f

: < where F exemplify Sep(, , , , ), see Denition 1.9(1)


and for

let Sol

=:

: for every < we have () ,= f

() where
Sol stands for solutions, so by clause (c) of the Denition 1.9(1) of Sep it follows
that
()

[Sol

[ < .
Let h be an increasing continuous function from into 2

with unbounded range; if


is regular, h = id

is O.K. Recall that = cf(2

) 2
<
. Let cd be a one-to-one
function from

(2

) onto 2

such that
= cd(

: < )) sup

: <
and let the function cd
i
: 2

for i < be such that cd


i
(cd(

: < ))) =
i
.
Let C = C

: S and C

, (recall that trivially [C[ 2

) and let
T = : for some C C (hence C is of cardinality < ), we have
C
(2

). By
assumptions (c) + (d) clearly [T[ =

CC
2
|C|
[C[ (2

)
<
2

but [T[ 2

hence [T[ = 2

. So let us list T possibly with repetitions as

: < 2

such
that [ < h() Rang(

) h()] and let T


<
=

: < h() for < .


For S let P

=: : is a function from C

into h() such that for every


C

we have (C

) T
<
, clearly P


(C

)
h(). Clearly
(C

); P

, C

is naturally a tree with P

the set of otp(C

)-
branches and cf(otp(C

)) = and set of nodes T


<
hence with < 2

nodes hence
by assumption (e)
() [P

[ < 2

.
For each P

and < we dene


,

C

(h()) by
,
() = cd

(()) for
C

. Now for P

, clearly

=: F(
,
) : < ) belongs to

. Clearly

: P

is a subset of

of cardinality [P

[ which as said above is < 2

.
Recall the denition of Sol

from the beginning of the proof and remember ()


above, so P

Sol

has cardinality < = cf() 2

. Hence we can nd



Sol

: P

.
[Why? if < 2

then [ Sol

: P

[ [Sol

[ : P

[P

[ <
2

= [

[ and if = 2

then 2

is regular, [P

[ < 2

, P

[Sol

[ < 2

and again [ Sol

: P

[ < 2

= [

[.]
Let < . Recall that



for S and f

from the list of F is a function from

to so f

) < . Hence we can consider the sequence c

= f

) : S)
S

as a candidate for being an F-Md-sequence. If one of them is, we are done. So


(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 13
assume towards a contradiction that for each < there is a sequence



(2

)
that exemplies the failure of c

to be an F-Md-sequence, so there is a E

D such
that

1
S E

F(

) ,= f

).
Dene


(2

) by

() = cd(

() : < )). Now as is regular uncountable


it follows by choice of h that E =: < : for every < we have

() < h()
and if

S, C

and C

= C

then

T
<
is a club of (see
the choice of T and T
<
, and recall as well that by assumption (c) the sequence
(C

, C

) : S) is good, see Denition 1.5(4) + (4D)).


By clause (a) in the assumption of our claim 1.10, the lter D includes the clubs
of so clearly E D; also D is
+
-complete hence E

=: E

: < E belongs
to D. Now choose E

S, xed for the rest of the proof; clearly

;
just check the denitions of P

and E, E

. Let < . Now recall that

,
is
the function from C

to h() dened by

,
() = cd

()).
But by our choice of

clearly C

cd

()) =

(), so
C

,
() =

() so

,
=

.
Hence F(

,
) = F(

). As E

clearly F(

) ,= f

) and as

clearly



is well dened. Now easily

() = F(

,
)
by the denition of

, so we have

() ,= f

).
As this holds for every < it follows by the denition of Sol

that


Sol

. But

was proved above so

: P

hence

Sol

Sol

: P

whereas

has been chosen outside this set,


contradiction.
1.10
We return to looking at Sep (see Denition 1.9) which appears in assumption (f)
of 1.10. There are obvious monotonicity properties:
1.13 Claim. If Sep(, ,
0
,
1
, ) holds and

, (

0
)

(
0
)

1

1
and

, then Sep(

0
,

1
,

) holds.
Proof. Let h be a one-to-one function from

0
) into

(
0
). Let

f = f

: < )
exemplify Sep(, ,
0
,
1
, ). For <

we dene f

:
(

)
(

0
)

1
by: f

() =
(
7
7
5
)


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







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


14 SAHARON SHELAH
f

(h()) if < & f

(h()) <

1
and f

() = 0 otherwise. Let

)
(

1
) and
let A

=
(

)
(

0
): for every <

we have f

() ,=

(). Let =:


so

(

1
)

(
1
), hence by the choice of f

: < ) the set A

=

(
0
):
for every < we have f

() ,= () has cardinality < . If A

then

0
) and h()

(
0
), and for <

we have f

() ,=

() hence for
< (

) we have either f

(h()) <

1
& () =

() ,= f

() = f

(h()) or
f

(h())

1
& () =

() <

1
so f

(h()) ,= (). Hence A

h() A

and as h is a one-to-one function we have [A

[ [A

[ <

and we are done.

1.13
To show that 1.10 has reasonable assumptions we should still prove 1.11.
Proof of 1.11. Our claimgives sucient conditions for Sep(, ), i.e. Sep(, , , , )
for some = cf() 2

. Clearly if < 2

we can waive regular as


+
can
serve as well. Let
1
= .
Case 1: =

, = ,
0
[, ] and we shall prove Sep(, ,
0
,
1
, ). Let
F =

f :f is a function with domain



(
0
) and
for some u []

and sequence =
i
: i < )
with no repetition,
i

u
(
0
), we have
(

(
0
)[
i
f() = i] and
(

(
0
)[(

i<
(
i
)) f() = 0]

.
We write f = f

u,
, if u, witness that f F as above.
Recalling =

, clearly [F[ = . Let F = f

: < and we let



f = f

:
< ). Clearly clauses (a),(b) of Denition 1.9 (with , ,
0
,
1
= here standing
for , , ,
1
there) holds and let us check clause (c). So suppose

and let
R = R

=:

(
0
): for every < we have f

() ,= (). We have to prove


[R[ < (as we have chosen = ).
Towards contradiction, assume that R

(
0
) has cardinality and choose
R

R of cardinality . Hence we can nd u []

such that u : R

) is
without repetitions.
Let
i
: i < list R

without repetitions and let


i
=:
i
u. Now let
=
i
: i < ), so f

u,
is well dened and belong to F. Hence for some <
we have f

u,
= f

. Now for each i < ,


i
R

R hence by the denition of


R, ( < )(f

(
i
) ,= ()) in particular for = we get f

(
i
) ,= (). But by the
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 15
choice of , f

(
i
) = f

u,
(
i
) and by the denition of f

u,
recalling
i
u =
i
we
have f

u,
(
i
) = i, so i = f

(
i
) ,= (). This holds for every i < whereas

,
contradiction.
Case 2: 2

and U

[] = .
Let = (2

)
+
or just (2
<
)
+
hence = cf()
+
2

.
Let u
i
: i < []

exemplify U

[] = . Dene F as in case 1 except that


for notational simplicity
0
= and we restrict ourselves in the denition of F to
u

P(u
i
) : i < . Clearly [F[ .
Assume that

and R = R



is dened as in Case 1, and toward
contradiction assume that [R[ (2
<
)
+
, hence we can nd

, (

) : < )
such that:

: < =

: <

) ,=

).
Clearly

: < ) is with no repetitions.


So by the choice of u
i
: i < as examplifying U

[] = , i.e., the denition


of U

[], for some i < the set u


i

: < has cardinality , call it u. So


u : R has cardinality and we can continue as in Case 1. If we wish to
use
0
[, ] instead of
0
= above let pr be a pairing function on . Without
loss of generality each u
i
is closed under pr and its inverses, and f

u,
is dened i
for some < we have u u

and i < Rang(


i
) u

. In the end choose i


such that u
i
pr(

, pr(

),

)) : < has cardinality .


Case 3: U
J
[] = where J = []
<
for such that

, .
Let = [2
<
]
+
.
The proof is similar to Case 2, only nd now

and (

) : < ) and choose


i < such that u
i
pr(

)) : < has cardinality .


Case 4: > ,= cf() and is strong limit.
Follows by case 2.
Case 5:

().
By [Sh 460] we can nd a regular <

() which is > and such that U

[] =
, so Case 2 applies.
(
7
7
5
)


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







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


16 SAHARON SHELAH
Note that in case 2, we can replace 2

by: there is F

of cardinality
such that for every A []

for some f F we have = Rang(f A). The


proof is obvious. Similarly in case 3.
1.11
There are some obvious monotonicity properties regarding the Md-property.
1.14 Claim. 1) If (,

C) has the -Md-property (see Denition 1.6(1)) and

C =
C

: S) and C
1

is an unbounded subset of C

for S then (, C
1

: S))
has a -Md property. We can also increase S and decrease ; we can deal with the
(, )-Md-property.
2) Assume = cf() >
0
, S : < , cf() = is stationary and < .
Then some C

: S) is a (, , )-Md-parameter.
Proof. 1) Obvious.
2) Easy (note: goodness is not required).
1.14
We may consider the following generalization.
1.15 Denition. 1) We say

C is a -Md
3
-parameter if:
(a) is regular uncountable
(b) S is a stationary subset of (consisting of limit ordinals)
(c)

C = C

, C

, C
,
: S, < cf())
(d) C

= sup(C

) and C

= sup(C

), otp(C

) = and C

= C
,
:
< cf()
(e) < cf() sup(

<
C
,
) < and C
,
increase with
2) We say that a -Md
3
-parameter

C is
3
good if < > [C
,
: S, <
cf() and the -th member of C

is [.
3) The (D, , )-Md
3
-property for

C means the (D, , )-Md-property for C

:
S), similarly for the other variants of this property (so the (D, , )-Md
3
-property
for for a stationary S S

is the same as the (D, , )-Md-property).


4) If C
,
: < cf() = C

: nacc(C

) we may write just C

: S)
(so Md
3
is like Md in this case).
We may consider the following variant.
3
if C

: S is a good (, , )-Md-parameter C

= sup(C

), otp(C

) = and we let
C
,
= { C

: otp(C

) +1} for < then C

, C
,
: S, < is a -Md
1
-parameter
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 17
1.16 Claim. Assume
(a) is regular uncountable, S < : cf() = is stationary and <
(b) B is a model with universe such that A []
<
c
B
(A) []
<
; we
may allow innitary functions provided that
()
B,
if A
i
: i < ) is increasing, A
i
[]
<
then c
B
(

i<
A
i
) =

i<
c
B
(A
i
)
(c)

C = C

: S) is a good (, ,
+
)-Md-parameter
(d) S

= S: for every < , c


B
()
(e) C

,
=
,
c
B
(C


,
) for S

and < where


,
C

is
increasing with with limit and C

= C

,
: < cf(), C

=
,
:
< cf().
Then C

, C

, C

,
: S

, < cf()) is a good (S, , )-Md


3
-parameter (and SS

is non-stationary); if then

<
C

,
= c
B
(C

).
Proof. Easy.
1.17 Claim. 1) Assume

C = C

, C

, C
,
: S, < cf()) is a good (, , )-
Md
3
-parameter. Then claim 1.10 holds (i.e. we get there the (D, )-Md
3
-property).
2) In part (1) hence in 1.10, we can weaken clause (e) to
(e)

(if = 2

we can use h = id

): there is h : 2

increasing continuous
with limit 2

and there is a set T =

: < 2

and function g :
such that if

(2

) then for stationary many S we have


(a) C

: < h(g())
(b) < cf() (

<
C
,
)

: < .
Proof. Same proof as 1.10.
1.18 Claim. In the proof of 1.10 instead of f : and f

: C

we can use
f : []
<
0
and f

: [C

]
<
0
(so get the -MD

-property for = 0, 1, 2).


Proof. Let cd

be a one-to-one function from


>
into and without loss of
generality

cd

(
0
,
1
, . . . ,
n1
)) < Max(w
(0)
+. . .
(n1)
: a
(
7
7
5
)


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







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


18 SAHARON SHELAH
permutation of n). We are given

C = C

: S), without loss of generality every


S is innite and divisible by ().
Let C
1,
= cd

() :
>
(C

) and C

1,
= nacc(C

).
1.18
1.19 Denition. 1) For a set A of ordinals and an ordinal let A = : <
or for some A we have [, + ).
2) The pair (,

C) or just

C has the (, )-MD
0
-property (or use MD instead of
MD
0
) if we can nd f

: S) such that:
(a) f

is a function from C

= C

to for some

<
(b) if f is a function from to and < is non-zero then for stationary
many S we have f

f and

= .
If we omit we mean = 2 so

= 1 for every S.
3) (D, , )-MD
2
-property means: if is a relational vocabulary of cardinality < ,
then we can nd M

: S) such that:
(a) M

is a -model with universe C

= C

for some

<
(b) if M is a -model with universe and < then the set A
M,
= S :
M C

= M

and

= is stationary and even belongs to D


+
.
If we omit we mean = 2 so

= 1 for every S. Similarly with D.


4) We say (, S) has the (, )-MD

-property if for some



C = C

: S) the
pair (,

C) has the (, )-MD

-property. We say that has the (, , )-property


if (, S

) has the (, )-MD

-property.
The following claim shows the close anity of the properties stated in theorem
0.1 in the introduction and the -Md-property with which we deal in this section
here. This claim is needed to deduce Theorem 0.1 from 1.10.
1.20 Claim. 1) Assume (,

C) has the -Md-property where

C = C

: S) is a
(, , )-Md-parameter. If < , =
<
then (,

C) has the (, )-MD
0
-property.
2) Assume in addition that

C is a good
+
(, , )-Md-parameter then (,

C) has the
(D, , )-MD
2
-property.
3) In part (2) we can allow to have function symbols but have:
(a)

is a -model with universe C

= C

for some

<
(b)

if M is a -model with universe and < then the following set belongs
to D
+
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 19
A

M,
= S :

= and (M c
M
(C

), c)
cC

= (M

, c)
cC

.
4) In part (2) we can replace good
+
by good if we add
each C

is closed under pr, where pr is a xed pairing function on .


Proof. The proof consists of a translation of one version of middle diamond to
another (note: is a relational is a strong restriction).
1) First x

< and deal with C

: S), C

= C

[0,

) recalling C

= C

. Let cd be a one-to-one function from


>
onto . Dene F

: S) as follows:
if f
(C

)
then F

(f) = cd(f(
,
) : < otp(C

))) where
,
: < otp(C

))
list C

in increasing order so F

is a function from
(C

)
into . As we assume (,

C)
has the -Md-property, we can apply its denition, hence for

F = F

: S)
there is a diamond sequence c

: S).
We now dene f

: C

as follows: if
0
is a member of C

and
[
0
,
0
+

) and < Min(C

(
0
+ 1)) and c

= cd(
,
: < otp(C

)) and

,
= cd(
,,i
: i <

) then f

() =
,otp(C

0
),
0
; otherwise f

() = 0. So
f

: C

. Why is f

: S) as required for C

: S) (and

)? Assume
that f : , so we can dene f

: by f

() = cd(f( + ) : <

)),
hence for stationarily many S we have c

= F

(f

). Now we check that


f C

is equal to f

dened above.
Let
,,i
= f

(
,
) and
,+i
= f(
,
+ i) for i <

, < otp(C

) so
,
=
f

(
,
) = cd(
,,i
: i <

)), c

= F

(f

) = cd(
,
: < otp(C

)).
Hence by the denition of f

we have: if =
,
C

, [, + ) and
< Min(C

(+1)) then f

() =
,,
= f(
,
+()) = f(+()) =
as required.
To have every

< (and guess also f (C

) f [0,

)) we just have
to partition S to S

: < ) such that each



C S

has the -Md-property which


we can do by 1.23(3) below (or work a little more above, i.e., f

() = cd(f(+) :
<

)f() : <

))).
2), 3), 4) Left to the reader.
1.20
1.21 Claim. 1) Assume = cf() > = cf(), S S

and
(a)

C

is a -Md-parameter for = 1, 2
(b)

C
1
= C
1

: S) has the (, )-Md-property


(c) A

[]
<
&
|A

|
= for < and S C
2

= A

: C
1

.
(
7
7
5
)


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







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


20 SAHARON SHELAH
Then

C
2
has the (, )-Md-property.
2) If

C is a (, ,
+
)-Md-parameter with the (, )-Md-property,

= and
C

= + i : C

= 0 and i < , then



C

= C

: Dom(

C)) is a
(, ,
+
)-Md-parameter with the -Md-property.
Proof. Easy by coding (as in the proof of 1.20).
1.22 Denition. Assume > are regular and

C = C

: S) is a (, , )-Md-
parameter with the (, )-Md-property. Let ID
,,,,
(

C) = A :

C (S A)
does not have the (, )-Md-property. If = we may omit it.
1.23 Claim. Let

C, , , , , be as above in Denition 1.22 and J = ID
,,,,
(

C).
1) J is an ideal on such that S J, / J if
0
.
2) J is -complete and even normal if (if
0
< it is still
+
-complete).
3) If S and S ,= mod J, then for any

F
i
= F
i

: S), F
i

:
(C

)
,
can be partitioned to subsets each having a

F
i
-Md-sequence.
Proof. 1) Let cd: be one-to-one onto and cd

: be such
that cd

(cd(
0
,

))) =

. Clearly J is a subfamily of P() and it is closed


under subsets, so we need just to prove that it is closed under union of two.
So suppose A = A
1
A
2
, A

J for = 1, 2. As J is closed under subsets,


without loss of generality A A
2
= . Let the sequence

F

= F

: S A

)
exemplify A

J. Now we dene

F = F

: S A) as follows:
if S A and f
(C

)
then:
A

(f) = F

(cd

f), i.e., F

(f) = F

(f

) where C

() = cd

(f()).
2) Straight as the sequence F

: S) is not required to be nicely denable.


3) Guessing is the same as guessing .
1.23

Well, are there good (, ,
+
)-parameters? Some version of this is crucial for
having interesting consequences of 1.10, but we shall have to quote (on I[] see
Denition 0.5; see on it [Sh 420, 1]).
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 21
1.24 Claim. 1) If S is a stationary subset of a regular cardinal and S I[],
see Denition 0.5, and ( S)[cf() = ], then
() for some club E of , there is a good
+
(S E, ,
+
)-Md-parameter. (Of
course, we can replace the
+
by any ) (recall that we do not required
the C

s to be a closed subset of )
() for some club E of there is a weakly good
+
(S E, ,
+
)-Md-parameter
consisting
4
of clubs (this is of interest when >
0
)
() for some club E of there is a good (SE, ,
+
)-Md-parameter

C = C

:
S E) such that
(i)

C is weakly good
+
, i.e., nacc(C

1
)nacc(C

2
) C

1
= C

(ii)

C is good, i.e., for each < we have > [ C

: S, C

[
(iii)

C is

-closed that is each C

is

-closed which means = sup(C

) & cf()

where

= : = cf() and
< [[
<>
tr
< .
2) If = cf(),
+
< = cf(), then there is a stationary S I[] with (
S)[cf() = ].
3) In part (1), moreover, we can nd a good (S E, ,
+
)-Md-parameter and a
-complete lter D on such that: for every club E

of , the set S E :
C

E belongs to D.
Proof. 1) By the denition of I[] see Denition 0.4 and 1.5 above (see more in [Sh
420, 1]) and for clause () the denition of [[
<>
tr
.
2) By [Sh 420, 1].
3) By [Sh:g, III].
1.24
1.25 Denition. Let

C be a (, , )-Md-parameter
(1) For a family F of

C-colourings and P

2, let id
C,F,P
be

W : for some F F for every c P


for some

the set
W S : F( C

) = c

is a non stationary subset of

4
this means: the parameter

C = C

: S E, C

a club of
(
7
7
5
)


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







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


22 SAHARON SHELAH
(2) If F is the family of all

C-colouring we may omit it. If we write Def instead
of F this mean as in [Sh 576, 1] (not used so the reader can ignore this).
We can strengthen 1.5(2) as follows.
1.26 Denition. We say the -colouring F is (S, , )-Md-good if:
(a) S < : cf() = is stationary
(b) we can nd E and C

: S E) such that
() E is a club of
() C

is an unbounded subset of , [C

[ <
() if ,


, S E, and

= C

then F(

) = F()
() for every < we have
> [C

: S E and C

[
() S [[
<cf()>
tr
< or just S > [C : C is
unbounded in and for every C for some S we have C =
C

and C

[.
We can sum up (relying again on some quotations).
1.27 Conclusion. If = cf(2

) and we let =: : = cf() and ( <


2

)([[
<>
tr
< 2

) then
(a) is large e.g. contains every large enough Reg

if

<
(b) if


+
< then there is a stationary S I[] such that S
cf() =
(c) if , S are as in clause (b) above then for some club E of
(i) there is a weakly good (, ,
+
)-Md-parameter

C = C

: S E)
(ii) there is a good (, ,
+
)-Md-parameter C

: E S)
(iii) there is a good
+
(, ,
+
)-Md-parameter

C = C

: E S)
(d) for , S,

C as in clause (b), (c) above and 2

, if F is a (

C, , )-colouring
and D = D

(or D is a
+
-complete lter on extending D

), and lastly
S D
+
then there is a F-middle diamond sequence for

C and there is an
DF-Md-sequence; see Denition 1.5(3),(3A)
(e) if , S,

C as in clauses (b), (c): 2

, then

C has the (, )-property
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 23
(f) for , S as in clause (b) above 2

, if >
0

C

= C

: S), C

a
stationary subset of for each then we can nd

C

= C

: E S)
such that C

is a club of for each and C

: S) has the
(, )-Md-property (really

C

does not depend on



C

)
(g) in clause (f) if =

, , 2

we can conclude also that C

:
S) has the (, )-MD
2
-property.
Proof. Clause (a) holds by [Sh 460], clause (b), (c) holds by 1.24, clause (d) by the
major claim 1.10 and clause (e) is a rephrasing of clause (d), and lastly for clause
(f) use above case (iii) of clause (c) and use the obvious monotonicity (and C

is unbounded in being stationary).


1.27
Note some open questions.
1.28 Question: 1) Assume is a weakly inaccessible which is not strong limit (i.e.
a limit regular uncountable cardinal but not strong limit) and 2

: < ) is not
eventually constant.
Does have the weak diamond? Other consequences? In this case cf(2
<
) = as
2

: < ) witness it hence 2

> 2
<
=: 2

: < . If = 2
<
then =
<
and such s are investigated in [Sh 755] (for weak diamond on many stationary
sets).
2) Consider generalizing [Sh:f, AP,3] to our present context. Also we may consider
c

:
+n

+(n3)
. . . ) where c

.
1.29 Claim. Assume

,
(i)
(ii) if F

has cardinality < 2

then there is

such that (f
F)( < )(() = f()).
1) Sep(, , , 2

) hence if 2

is a regular cardinal then Sep(, ) holds so clause


(f ) of 1.10 is satised.
2) Even if 2

is singular still in Claim 1.10 we can replace clause (f ) in the as-


sumption by
,
.
Remark. On relatives of
,
see e

, see [MPSh 713].


Proof. 1) Clearly

has cardinality 2

, so let

: < 2

) be a list without
repetitions of

. For 2

, let

: < and we choose



f

by
induction on 2

such that
(
7
7
5
)


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







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


24 SAHARON SHELAH
(a)

f

= f

: < )
(b) f

is a function from

to
(c) f

is increasing continuous in , i.e., < f

= f

Dom(f

)
(d) for every < < the set < : f

) =

() is not empty.
If we succeed, then we let f

= f
2

that is f

: < 2

for < and


f

: < ) is clearly as required.


For clause (d) it is enough to consider the case = +1, as the non trivial case
in the induction is = + 1. We can choose f
+1

) : < )

such that it
holds by the assumption ().
2) In the proof of 1.10, now

[Sol

[ < 2

does not suce. However, before


choosing f

: < ) we are given F and we can choose cd, cd

: < ) and dene


T,

: < 2

), T
<
and P

: S). Note that < [h()[


<>
tr
< 2

hence
< [ P

: S [ < 2

. We can also choose


,
: P

, < ) and
dene



for P

, S.
Now we can phrase our demands on f

: < ), which is in fact weaker than the


full generality of Sep(, ); it is


,=

: for some P

we have ( < )(f

() ,=

()).
For this it is enough that for the list

: < 2

) chosen in the proof of part (1),


for each S, for some < 2

we have

: P

: < which is
easy to obtain.
1.29
1.30 Observation. If
0
, = 2, then
,
holds.
Proof. If F

, [F[ < 2

, let

F and let = 1 () : < ).
1.30
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 25
2 Upgrading to larger cardinals
In 1 we get the -middle diamond property for regular cardinals of the form
cf(2

) on most S S

for many = cf() (and ). We would like to prove


Theorem 0.1. For this we rst give an upgrading, which is a stepping up claim
(for MD

) given as 2.1 below. Then we prove that it applies to


2
(for quite a few

1
), when
2
is a successor of regular, then when
2
is successor of singular and
lastly when
2
is weakly inaccessible (not strong limit). We can upgrade each of
the variants from 1, but we shall concentrate on the one from 1.5(5A) because of
Claim 1.20.
The main result of this section is claim 2.9.
Question: What occurs to
2
strongly inaccessible, does it have to have some Md-
property?
Possibly combining 1 and 2 we may eliminate the
2
> 2

1
++
in 2.1, but
does not seem urgent now.
2.1 Upgrading Claim. 1) Assume
(a)
1
<
2
are regular cardinals such that
2
> 2

1
+
<
+
(this last condition
can be waived for nice colourings, see Denition 2.2 below)
(b) for some stationary S
1
S

, some sequence

C = C
1

: S
1
) is a
(
1
, , )-Md-parameter which has a (D

1
, , )-Md-property
(c)
2
>
+
1
,
2
is a successor of regular.
Then for some stationary S
2
S

, there is

C = C

: S
2
), a good
5
(
2
, , )-
Md-parameters which has the (D

2
, , )-Md-property.
2) We can in part (1) replace clause (c) by any of the following clauses
(c)

there is a
1
square on some stationary S S

1
, (see Denition 0.4(1A);
this holds if
2
=
+
,

1
= )
(c)

2
has a (
1
, S
1
)-square (where S
1
is from clause (b)) which means:

1
,S
1

1
<
2
are regular, S
1

1
is stationary and there are S
2
S

1
stationary and

C
2
= C
2

: S
2
) satisfying C
2

= sup(C
2

), C
2

is closed, otp(C
2

) =
1
and if C
2

1
C
2

2
,
1
,
2
S
2
and =
sup(C
2

1
) and otp( C
2

2
) S
1
, then C
2

2
= C
2

1
.
5
the goodness is a bonus
(
7
7
5
)


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







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


26 SAHARON SHELAH
(c)

2
has a (
1
,

C
1
)-square

C
2
(where

C
1
is from clause (b)) which means:

1
,

C
1
,

C
2
()

C
2
= C
2

: S
2
)
() S
2
S

1
is stationary
() C
2

is a closed unbounded subset of of order type


1
() if C
2

1
C
2

2
(so
1
,
2
S
2
) and otp ( C
2

1
) S
1
,
and = sup(C
2

1
) then otp(C
2

1
) = otp(C
2

2
), call it
and
C
2

1
: otp(C
2

1
) C
1

= C
2

2
: otp(C
2

2
) C
1

.
3) Assume 0, 2
(a)
1
<
2
are regular uncountable
(b) some (
1
, , )-Md-parameter has the -MD

-property
(c) at least one of the variants of clause (c) above holds.
Then some (
2
, , )-Md

-parameter has the -MD

-property.
2.2 Denition. We say that a (, )-colouring F = F

: S) for a -Md-
parameter

C = C

: S) is nice when:
if
1
,
2
S and otp(C

1
) = otp(C

2
) and
1

(C

1
)
,
2

(C

2
)
and

1
C

1

2
C

2
otp(C

1

1
) = otp(C

2

2
)
1
(
1
) =
2
(
2
),
then F

1
(
1
) = F

2
(
2
).
Proof. 1) If clause (c) holds then clause (c)

of part (2) holds by [Sh 351, 4], so it


is enough to prove part (2).
2) Trivially clause (c)

implies clause (c)

and clause (c)

implies clause (c)

; so
assume (c)

; and let

C
2
= C
2

: S
2
) be as there.
Let

h = h

: S
2
) be such that h

is an increasing continuous function from

1
onto C
2

recalling clause (), i.e., C


2

is a closed unbounded subset of of order


type
1
. Let S = h

(i) : S
2
and i S
1
and if S, = h

(i), S
2
and
i S
1
then we let C

= h

(j) : j C
1
i
; and let

C

= C

: S). Why is C

(hence

C

) well dened? We shall check that if S, = h

(i

),

S
2
, i

S
for = 1, 2 then h

1
(j) : j C
1
i
1
= h

2
(j) : j C
1
i
2
; this holds by clause
() of the assumption (c)

. We should check the conditions for



C

is a good
(
2
, , )-Md-parameter with the (D

2
, , )-Md-property.
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 27
The main point is: assume that F is a (

C

, , )-colouring, nd a c as required.
For each S
2
we dene a (

C
1
, , )-colouring F

by F

() = F( h
1

) that
is for i S
1
and
C
1
i
we dene F

() as F() where
C

(i)
, where is
dened by C

(i)
, we let () = (h
1

()). As
6

2
> 2

1
+||
<
+
because

{||
|C

|
:S
1
}
2

1
{||
||
:}
= 2

1
+||
<
+ , for some stationary S

2
S
2
we have S

2
F

= F

, so F

is a (

C
1
, , )-colouring, hence it has a good
(D

1
, , )-Md-sequence c
1
= c
1
i
: i S
1
).
We dene c

= c

: S) by S
2
i S
1
c

(i)
= c
1
i
. By clause () of
(c)

it is well dened. Now for any


(
2
)
and club E
2
of
2
, we can nd
S
2
acc(E
2
), hence E
1
= <
1
: h

() E
2
is a club of
1
and let
1

(
1
)

be
1
() = (h

()), so by the choice of c


1
for some
1
S
1
E
1
, c
1

1
= F

(
1
C
1

1
)
and let
2
= h

(
1
), so
2
E
2
and c

2
= c
1
i
= F

(
1
C
1

1
) = F

(
1
C
1

1
) = F(
C

2
), so we are done.
3) Similar only easier (or see the proof of 2.10).
2.1
2.3 Claim. We can strengthen the conclusion of 2.1(1) and 2.1(2) to: we can nd
a sequence S
2,i
: i <
2
) of pairwise disjoint stationary subsets of
2
and for each
i <
2
a sequence C
i

: S
2,i
) which is a (
2
, , )-Md-parameter having the
(D

2
, , )-Md-property.
Proof. By the proof of 2.1, without loss of generality clause (c)

of 2.1(2) and
clauses (a), (b) of 2.1(1) hold. For some <
1
for unboundedly many i <
2
the set S
2,i
=: S
2
: i is the -th member of C
2

is stationary.
2.3
2.4 Conclusion: 1) If is a strong limit, is a successor cardinal, = cf() > 2

,
moreover > 2
2

(superuous for nice colourings, e.g., for MD

, 0, 2), then
for every large enough regular < , some good
+
(, ,
+
)-Md-parameter

C has
the (2

, )-Md-property for every < .


2) We can add:

C has the -MD

-property for = 0, 2.
3) We can nd a sequence S
i
: i < ) of pairwise disjoint stationary subsets of S

such that for each i < there is



C
i
= C
i

: S) as in parts (1) and (2).


Proof. 1) Let
0
< be large enough such that [ , 2

& < &


0

= cf() < ] [[
[]
< , (see Denition 1.3, exists by [Sh 460]).
By 1.10, for any regular [
0
, ) and < we have the result for cf(2

) in
place of . If is a successor of regular we can apply 2.1(1), i.e., for clause (c)
6
only place this is used, if F is nicely dened this is not necessary
(
7
7
5
)


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







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


28 SAHARON SHELAH
there is a square as required by [Sh 351, 4]. In the case that is a successor of a
singular cardinal, by 2.7 below, clause (c)

of 2.1(2) holds for


2
= ,
1
= cf(2

)
so we can apply 2.1.
2) Follows by 1.20.
3) In the proof of parts (1) + (2) above, instead of 2.1 we use 2.3.
2.4
Below we generalize Engelking Karlowic theorem [EK] (which says that if =

, = 2

then there are f

: for < such that every partial function


from to with domain of cardinality , is extended by f

for some < ); we


do more than is necessary for the middle diamond.
2.5 Claim. 1) Assume
(a) = cov(,
+
,
+
,
+
)
(b) 2


(c) < 2

Then there is

F such that:
()

F = F

: < )
() each F

is a function from to
()

if u []

then we can nd

E such that
(i)

E = E
i
: i < ) is a sequence of equivalence relations on u
(ii) u = /E
i
: i <
(iii) if f : u respects some E
i
(i.e. E
i
f() = f())
then we can nd a partition u

: < ) of u such that < f


u

for some < .


2) In part (1) we can strengthen clause ()

to () below if (d)

for some
1, . . . , 5 below holds
() if u []

and f : u , then we can nd u = u


i
: i < ) such that
(i) u = u
i
: i < and
(ii) for every i < for some < we have f u
i
F

(d)
1
there is a sequence

f = f

: < ) of pairwise distinct functions from


to such that: if u []

then we can nd sequences u = u


i
: i < ) and
B
i
: i < ) such that u = u
i
: i < , B
i
[]
<
0
for i < and for each
i < the sequence f

B
i
: u
i
) is with no repetitions
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 29
(d)
2
there is a sequence

f = f

: < ) of functions from to and an ideal


J on such that

f is
+
-free modulo J, i.e., (
+
, J)-free (which means that
for u []

, we can nd a sequence a = a

: u) of subsets of such
that a

= mod J and i a

& ,= f

(i) ,= f

(i))
(d)
3
like (d)
2
but a

J
+
that is a

/ J
(d)
4
( < )([[

< ) and < pp


J
() for some ideal J on
(d)
5
=

0
< <

.
3) If we weaken the conclusion of part (2), replacing () by
()

,
for every club E of for some closed u E of order type there is u
i
:
i < ) such that u
i
u and u & cf() > u
i
: i < and
(i < )( < )[f u
i
F

)
then we can weaken the assumption in a parallel way, e.g.
(d)

1,,
there is

f = f

: < ), f



such that for any club E of we can nd a
closed e E of order type +1, and u = u
i
: i < ),

B = B
i
: i < ) such
that B
i
[
1
] <
0
, u
i
and u & cf() u
i
: i < and
for each i, f

B
i
: u
i
) is with no repetitions (and in (d)
3
it becomes
easy).
Remark. We can replace (as a domain) by []
2
, etc.
Proof. 1) Let A be a family of subsets of each of cardinality such that
any A []

is included in the union of many of them; it exists by assumption


(a). By assumption (b) without loss of generality A A & B A B A
so we replace included by equal and also without loss of generality A
1
A &
A
2
A A
1
A
2
A. Let pr be a pairing function on .
As < 2

there is a sequence f

: < ) of pairwise distinct functions


from to 0, 1.
Let
Y = (A, B, g,

) :B A, g = g
i
: i < )
g
i

B
2 and i < j < g
i
,= g
j
and

=
i
: i < ) satises
i
: i < A A.
(
7
7
5
)


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







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


30 SAHARON SHELAH
Clearly [Y [ 2

= . For each x = (A, g,



) Y we dene a function
F
x
: as follows:
F
x
() is
i
if g
i
f

and is 0 if there is no such i (clearly we cannot have


i
0
,= i
1
< & g
i
0
f

& g
i
1
f

).
Let

F = F

: < ) list F
x
: x Y , clearly

F satised demand () +(). As
for demand ()

let u []

be given. For any


1
,=
2
u let = (
1
,
2
) < be
minimal such that f

1
() ,= f

2
(), exists as f

2
,= f


2 and let B = (
1
,
2
) :

1
<
2
are from u. By the choice of A we can nd pairwise disjoint B

A for
< such that B =

: < . Lastly, we dene



E = E

: < ) by
for
1
,
2
u,
1
E

2
i f

1
B

= f

2
B

.
Assume that f : u respects E

, then we can nd non empty A

A for <
such that Rang(f) =

<
A

. Now for any pair (


1
,
2
) , let u

1
= Min
u : f() A

1
, let
i
: i < ) list u

2
, let x

1
,
2
be (A

2
, B

1
B

2
, f

i
B

1
: i <
), f(
i
) : i < )). Now check.
2) Naturally the proof splits to ve cases, but rst let A be as in the proof of part
(1), pr(..., ...) be a pairing function on , let f

: < ) be a sequence of functions


from to (chosen separately in each case). Let
X = x :x = (A, g,

) satises
(i) A A
(ii) g = g
i
: i < ),
(iii) each g
i
is a partial function from A to A
(iv)

=
i
: i ) and i
i
A.
Clearly [X[ and for each x = (A, g,

) X we dene F
x
: by
F
x
() =
i
i i = i
x
() = Min : if < then g

.
So let u []

and f : u be given and let v


i
: i < ) be such that u =

v
i
:
i < and A
i
: i < ) be such that i < A
i
A and Rang(f
v
i
) A
i
.
Case 1: Clause (d)
1
holds.
We choose f

= f

for every < where f

is from clause (d)


1
. Let (u
i
, B
i
) :
i < ) be as guaranteed to exist for u in clause (d)
1
. For each i, j < such that
u
i
v
j
,= and let
i,j

: < ) list u
i
v
j
(possibly with repetition) and let
x
i,j
= (A

i,j
, g
i,j

: < ),
i,j

: < )) where
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 31
(a) A

i,j
= B
i
A
j
(b) g
i,j

= f

i,j

B
i
(c)
i,j

= f(
i,j

).
We can check that
() x
i,j
X and
()

u
i
v
j
: i, j < = u
() F
x
i,j
f (u
i
v
j
).
[Why? Let () u
i
v
j
so () =
i,j
()
for some () < so g
i,j
()
= f

i,j
()
B
i
by
clause (b) above hence () : if < then g

()
. But f

B
i
:
u
i
) is a sequence without repetitions, so
< ()
i,j

u
i
g
i,j

= f

i,j

B
i
,= f

i,j
()
= g
i,j
()
(the middle inequality holds by the choice of (u
i
, B
i
) : i < ) above, i.e., by clause
(d)
1
).
So i
x
i,j
(()) = () so F
x
i,j
(()) =
i,j
()
= f(
i,j
()
), the last equality holds by
clause (c). So we are done.
Case 2: Clause (d)
2
holds.
Follows from Case 3.
Case 3: (d)
3
holds.
Let f

be any function from to extending f

. It suces to show that


f

: < ) satises clause (d)


1
. So let u []

, so by (d)
3
we can nd a sequence
a = a

: u) of subsets of such that a

/ J and

1
,=
2
u & i a

1
a

2
f

(i) ,= f

(i).
Let B
i
= i []
<
0
and u
i
= u : i a

, clearly they are as required.


Case 4: Clause (d)
4
holds.
This implies Clause (d)
3
by [Sh:g, II,1.3,p.46+1.4](3),p.50 (or see [Sh 506]).
Case 5: (d)
5
.
(
7
7
5
)


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







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


32 SAHARON SHELAH
Let
1
= Min :

> , so clearly
1
. As , 2

we have
2

<
1
, hence
1
= cf(
1
) is and ( <
1
)([[

<
1
). By (d)
5
we have

1
>
0
. So by [Sh:g, VIII], pp
J
bd

(
1
) (
1
)

> so Clause (d)


4
holds.

2.5
2.6 Claim. 1) Assume
(a) = cov(,
+
,
+
,
+
) so >
(b) 2


(c) < = cf()

(d) = cf(), S S

is stationary and S I[]


(e) T and [ < & T [[
<cf()>
tr
< ].
Then there is

F such that
()

F = F

: < )
() F

is a function from to
() for a club of S, there is an increasing continuous sequence

: < )
with limit such that:
() for every function f from

: T to we can nd a sequence
u
i
: i < ) of subsets of T with union T such that (i < )( <
)(f
j
: j u
i
F

).
Proof. 1) Let a = a

: < ) and E
0
as in 0.5(2) witness S I[]. Let

M = M

: < ) be an increasing continuous sequence of elementary submodels


of (H(), , <

), each of cardinality < satisfying



M ( + 1) M
+1
, |M
0
| =
, M
1+
and , , T, S, a, E
0
(+1) M
0
. Let E = < : M

= ,
clearly a club of and E E
0
as E
0
M
0
.
Now we choose by induction on < a function f

from to such that:


()
1
if b []

& b M
+1
then
Rang(f

) Rang(f

) : b
()
2
f

is the <

-rst object satisfying ()


1
.
This is clearly possible as 2

<

.
For every i < and u M
0
[]

and g : u from M
0
, let F
g,i
be the unique
F
g,i
: and it satises: if f

(i) Dom(g) then F


g,i
() = g(f

(i)). [Why
F
g,i
M
0
? As + 1 M
0
and u, g M
0
.]
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 33
Let (g

, i

) : < ) list the set of all such pairs (g, i) and let F

= F
g

,i

. We
shall show that F

: < ) is as required; so clauses () + () hold trivially.


Now if acc(E)S SE
0
so a

= sup(a

), otp(a

) = cf() = and let

: < ) list the closure in of a

in the increasing order. So for some club C of


consisting of limit ordinals we have C

E. Let = cf() : T, now


suppose C is a limit ordinal with conality . So for every < ,
+1
a

so, as a M
0
M

&
+1
M

clearly a

+1
M

but (see Denition


0.5(2)) c(a

+1
) =

: hence

: M

. Now the tree


( : M

is a sequence of ordinals of length < , ) is a tree which belongs


to M

+1
, has levels and |M

| < nodes so by assumption (e) it has <


-branches. As this set (of -branches) belongs to M

+1
, every -branch of this
tree belongs to M

+1
but

: < ) is (essentially) such a branch; so together


C cf()

: < ) M

+1
.
Let
v
i
= C :cf() and
f

(i) / Rang(f

) :

: < .
So v
i
: i < ) is a sequence of subsets of C with union T. For each i < by
assumption (a) there is v
i,j
: j < ) be a sequence of subsets of v
i
with union v
i
such that for every j < f

(i) : v
i,j
v
+
i,j
M
0
.
Let h be a function from

: T to . So by assumption (a) we can nd a


sequence w
j
: j < ) of subsets of each of cardinality such that Rang(h)

j<
w
j
and w
j
: j < M
0
. Let u
i,j
1
,j
2
= : v
i
, f

(i) v
+
i,j
1
and
h(

) w
j
2
. Hence for i, j
1
, j
2
< there is g = g
i,j
1
,j
2
: f

(i) : u
i,j
1
,j
2

satisfying u
i,j
1
,j
2
g(f

(i)) = h(

), see the choice of v


i
, v
i,j
1
, w
j
2
, u
i,j
1
,j
2
.
Now as v
i
, 2

, +1 M
0
and v
+
i,j
1
, w
j
2
M
0
are subsets of of cardinality
clearly the set of partial functions from v
+
i,j
1
to w
j
2
belongs to M
0
and has
cardinality
|u
i,j
|
[w
j
[

= 2

hence is included in M
0
. So g
i,j
belongs
to M
0
for every i, j < . Also easily F
g
i,j,i
extends h (

: u
i,j
). So

: u
i,j
: i, j < ) and

: C) are as required (modulo some


renaming). But Dom(g
i,j
) = f

(i) : u
i,j
w
j
.
2.6
Compare the following with [Sh 237e].
2.7 Claim. 1) Assume
(a) =
+
and = cf() < and
0
< = cf() <
(
7
7
5
)


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







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


34 SAHARON SHELAH
(b) (i) = :
0
= cf() < and
<>
tr
=
(ii) T

= < : cf() ,
(iii) T
+

= T

0 + 1 : < ,
(iv) T

= T
+

(c) P

A :

A a sequence of length of subsets of each included in T
+

and including T


(d) P []

has cardinality
(e) if

< for < and w

i
: i < ) a sequence of subsets of T
+

and
including T

then for some



A P we have:
(i) for some < g(

A) and j < , A

j
(ii) < g(

A) & j < & A

: A

P.
Then we can nd (

C
i
, A

i
, S
i
) for i < such that
() 0 + 1 : < A

i

()

C
i
= C
i

: S
i
); S
i
and [otp( C
i

) A

i
& C
i

& C
i


S
i
& C
i

= C
i

]
() S

i
= S
i
: cf() = is stationary
() for every club E of there are

E and a sequence w

i
: i < ) of subsets
of T
+

including T

with union T
+

such that for each j < there is



A P
satisfying: ( < g(

A))(A

j
(i < )[

i
& A

i
= A

])
() if B I[] is stationary and B S

then we can demand BS

i
: i <
is not stationary and in clause () any B satises the requirement on

.
2) Assume, in addition < [[

< , T

= < : cf() ,= and we let


P =

A :

A = A
i
: i < ) is a sequence of sets with union T
+

. Then T

= T
+

and for some P all the assumptions hence the conclusion of part (1) holds.
3) Assume

is strong limit < , = cf() < , =


+
, then for some
0
<

,
for any regular (
0
,

) the set T

include < : cf()


1
or nonlimit
and for P =

A :

A is a sequence of <
0
subsets of with union T

such that
A

: < g(

A) is closed under nite unions.
Then for some P the assumptions hence the conclusion of part (1) holds.
4) In (1) (and (2), (3), if

A = A

: < ()) P we can replace it by



A

= A

:
< ()), A

= A

< : cf() >


0
and A

is stationary in and still


obtain an element of P.
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 35
2.8 Remark. 1) If is (weakly) inaccessible we have weaker results: letting = :
= cf() and < [[
<>
tr
< , there is P

: < ), P

[]
<
, [P

[ <
such that for regular < for every stationarily many S

for some increasing


continuous sequence

: < ) with limit , [ < & cf()

: <
P

] (see [Sh 420, 1]).


2) Can we restrict ourselves to: w

j
: j < ) is increasing continuous in part (1)
and similarly in (2).
It seems very reasonable (if =
+
, = cf() < ) that we can nd
i
:
i < ), an increasing sequence of regular cardinals < with limit such that

+
= tcf(

i<

i
,
J
bd

) and j <
j
> max pcf
i
: i < j. Why? This is
trivial if =
0
and if >
0
, it folows by the weak hypothesis (see [Sh 410]), whose
negation is not known to be consistent and which has large consistency strength.
3) If there is
i
: i < ) is as above then in part (1) of 2.7 we can strengthen
clauses (), () in the conclusion. By demanding that w

j
: j < ) is increasing
continuous. Why? Choosing f

: < ) we add
if f

1
(i
1
) = f

2
(i
2
) then i
1
= i
2
& f

1
i
1
= f

2
t
2
.
so there is a sequence g
i
: i < ), g
i
: and g
i
1
(f

1
(i
1
)) = f

1
(i
0
)
when i
0
< i
1
< , < and a function i : such that i(f

(i)) = i.
Now in the end of the proof we replace

= cd(j, f

(j), (

)) (so xing j) by

= cd().
Proof. There is B as in clause (), that is B I[] and B S

is a stationary
subset of , see Claim 1.24(2). So let B be such a set; clearly omitting from B
a nonstationary subset of is O.K. as if B

B, BB

is non-stationary and the


demands in clause () for B

are exemplied by the sequence

C
i
, A

i
, S

i
) : i < )
then the same sequence exemplies it for B.
We can nd a = a

: < ) witnessing B I[] see Denition 0.5(2), so


(possibly omitting from B a nonstationary set) we have: B a

=
sup(a

) & = otp(a

) and [ a

= a

] and [ B a

&
otp(a

) < ].
Let =
+
and

M = M

: < ) be an increasing continuous sequence of


elementary submodels of (H (), , <

) to each of which a, , P, P belongs, and


satisfying

M ( + 1) M
+1
, M

and |M

| < .
As =
+
clearly < |M

| = . For any < let b


,
: < ) list
M
+1
[]
<
.
Let
i
: i < ) be an increasing sequence of regular cardinals with limit ,
i
>
with = tcf(

i<

i
, <
J
bd

) and let

f = f

: < ) be an increasing sequence in


(
7
7
5
)


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







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


36 SAHARON SHELAH
(

i<

i
, <
J
bd

) conal in it and obeying a (i.e. i < & a

(i) < f

(i)) and
<
i
i < j < b
,
f

(j) < f

(j) (there is such


i
: i < ) by [Sh:g, II]);
there is such

f by [Sh:g, I]. For notational help later without loss of generality each
f

(i) is divisible by .
Now for any x = (h
1
, h
2
, A, i) = (h
x
1
, h
x
2
, A
x
, i
x
) where h
1
is a partial function
from to , h
2
a function from A to where T

A T
+

and i < , we dene


S
x
=

< : we have cf() ,= , f

(i) Dom(h
1
) and =: h
1
(f

(i)) A satises
(i) b
,h
2
()
is a subset of of order type ,
(ii) b
,h
2
()
is unbounded in if is a limit ordinal
(iii) if A and b
,h
2
()
and = otp(b
,h
2
()
)
then = h
1
(f

(i)) & b
,h
2
()
= b
,h
2
()

and for S
x
let C
x

= b
,h

2
(h

1
(f

(i))
.
For x as above let S

x
= < : cf() = and there is an increasing continuous
sequence

: < ) with limit such that A

S
x
& C
x

:
< . Now for any such x, for S

x
choose

: < ) witnessing it and let


C
x

C
x

: A; note that S
x
S

x
= because [ S
x
cf() ,= ], [
S

x
cf() = ].
Let x

: <

) list the set of x M


0
which are as above and S

x
is
stationary, and for each <

, let S

=: S
x

,

C

=: C
x

: S

), A

=: A
x

.
We have to check the ve clauses of 2.7(1):
Clause (): See the demand on x above.
Clause (): Holds by clause (iii) in the denition of S
x
and of C
x

for S
x
and similarly for S

x
.
Clause (): Holds by the choice of x

: <

).
Clause () + (): Let B, and we shall show that S

i
: i < , this clearly
suces for () hence for (). Let

: ) list the closure of a

in the increasing
order by the choice of a, a

is a set of non-limit ordinals hence cf(

) = =
and

= .
Now for every T
+

, for some () < we have

: < = b
,()
(()
exists as
<cf()>
tr
= by the denition of T

) and let j() = Minj < : f

(j) >
f

(i) for every < j.


(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 37
For each j < let w

j
= : T

or T

& j() = j and apply clause (e)


of the assumptions to the sequence

: < ) where

= cd(j(), f

(j

), (

)),
cd a coding function on so (by (e)) we can nd

A P satisfying clauses (i) and
(ii) of assumption (e). So by (i) of (e) there are i() < g(

A) and j() < such
that A
()
w

j()
.
For any pair (i, j) such that i < g(

A), j < and A
i
w

j
we shall dene x = x
i,j
.
By (ii) of (e) we have

: A
i
P.
We now dene x = (h
x
1
, h
x
2
, A
x
, i
x
) as follows: we let i
x
= j, A
x
= A
i
,Dom(h
x
1
) =
f

(j) : A
x
, h
x
1
(f

(j)) = , Dom(h
2
) = A
x
, h
2
() = (

). Easily x = x
i,j

M
0
(as P M
0
, [P[ , + 1 M
0
) and

x
, so we are done.
2) Let P = A : A is a bounded subset of of cardinality . Clearly [P[
[[

: < [[ : < = , so clause (d) holds.


Now if = cf() < and ,= then
<>
tr
= . [Why? For completeness if
T is a tree with set of nodes and levels, then very -branch of T includes a
<
T
-unbounded subset which is bounded as a subset of hence belongs to P, and
clearly the branch is determined by any <
T
-unbounded subset hence the number
of -branches of T is [P[ .] So T

= T

and the non-obvious clause to check


is (e). So assume

< for < and w

i
: i < ) is a sequence of subsets of T
+

.
Let =
i
: i < ,
i
< and let

A = A
i
: i < ) list the family

: w

i
and

<
j
: i, j < .
3), 4) Left to the reader.
2.7
2.9 Claim. 1) If is regular and not strongly inaccessible and < is a strong
limit singular of uncountable conality, then for every large enough regular < ,
for every , < , some (, ,
+
)-parameter

C has the (, )-Md-property (see
Denition 1.19(4)); in fact,

C is a sequence of clubs.
2) Moreover, we can nd such

C

for < satisfying: Dom(



C

) : < ) are
pairwise disjoint.
3) Moreover for every large enough regular < for every , < some (, ,
+
)-
parameter has the (, )-Md

-parameter for = 0, 1, 2.
Proof. 1) Let
1
< be such that 2

1
and let
0
< be large enough such
that
()
1
(i) if is not a successor of regular, then < cov([[, , ,
0
) <
(ii) if is inaccessible, then cov(
1
, , ,
0
) =
1
.
(Exist by [Sh 460].)
Note that
1
is needed only if is (weakly) inaccessible.
(
7
7
5
)


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







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


38 SAHARON SHELAH
Let

(
0
, ) be strong limit; this is possible as is a strong limit of un-
countable conality. Let
1
=
1
[

] <

be >
0
and such that < 2

cov([[, , ,
1
) < 2

; (such
1
exists by [Sh 460]).
Let Reg

1
and we shall prove the result for and every , <

, so
without loss of generality = . By Fodors Lemma on this is enough.
Let

= cf(2

). By 1.24(1) for every stationary S <

: cf() = from
I[

] (and there are such, see 1.24(2)) there are



C
S
= C
S

: S), D such that

C
S
is a weakly good
+
(

, ,
+
)-Md-parameter, D is a

-complete lter on

containing the clubs of

and by 1.10 + 1.11 S D


+
, (e.g., D = D

), the pair
(

C
S
, D) has the -Md-property.
For the case is (weakly) inaccessible we add each C

is a club of . This is
permissible as we can use clause () of 1.24(1). Fix such S
1
S

and

C = C

:
S
1
) =

C
S
1
.
Note that we actually have
(a) if S

S
1
is stationary, C

= sup(C

) for S

then

C

=
C

: S

) is a weakly good
+
(

, ,
+
)-Md-parameter and

C

has the
(D, )-Md-property
(b) if

= A
i
: i < i() < then for some i, j < i(),

C

is a good
(

, ,
+
)-Md-parameter and

C

has the (D, )-Md-property where S

=:
S
1
: A
j
& = sup(nacc(C

) A
i
) and for S

we let
C

=: nacc(C

) A
j
.
[Why? Let S

i,j
= S
1
: A
j
and = sup(nacc(C

) A
i
). Clearly
S
i,j
: i, j < i()) is a sequence of subsets of S
1
with union S
1
. By Claim
1.23 for some i, j < i() the sequence

C S
i,j
has the (D, )-Md-property.
By monotonicity, i.e., Claim 1.14(1), the sequence nacc(C

) A
i
: S
i,j
)
has the (D, )-Md-property.]
Now
Case 1: is the successor of regulars.
We apply 2.1(1) with (,

) here standing for (


2
,
1
) there; there is suitable
partial square (in ZFC by [Sh 351, 4]). Note that we do not lose any <
because of the lifting to .
Case 2: is the successor of a singular.
The proof is similar only now we use 2.7(3) to get the appropriate partial square
using to derive the

C

used as input for it (recalling that each sequence



A P
is closed under the union of two!)
Case 3: is (weakly) inaccessible.
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 39
Really the proof in this case covers the other cases (but
1
in this case contributes
to increasing
0
). This time we use 2.6 + 2.10 which is proved below.
Let S
2
be a stationary subset of S

from I[], it exists, see 1.24(2). Let F

:
<
1
) be as in 2.6 with (,
1
,

, ) here standing for (, , , ) there. Let


g

be a function from
1
onto H

). Possibly replacing S
2
by its intersection
with some club of , there is a good

-Md-parameter

C
2
= C
2

: S
2
). Let
e = e

: S

), e

a club of of order type . Clearly


() for some <
1
for any club E of for some increasing continuous sequence

i
: i

) of member of E the set B

(
i
: i

)) = i S
1
:
g

(F

(
i
)) = (i, , otp(

C
2

), otp(

i
)) : < i and

is a stationary subset of

[Why? For every club E we can choose


i
E (i

) increasing continuous so
for some the set B

(
i
: i

)) is a stationary subset of

. As the number of
possible is
1
< = cf(), there is as above.]
For this let
(i) S

3
= < : cf() = and g

(F

()) has the right form


(ii) S

3
let C
3

= e

: (i, , j, otp( e

)) g

(F

())
(iii) let F

be dened by F

() = if:
=
1
for some (
1
,
2
,
3
,
4
) g

(F

()).
We apply 2.10.
2),3) Similarly using Claim 2.3 when using Claim 2.4.
2.9
We now state another version of the upgrading of middle diamonds.
2.10 Claim. 1) Assume
(a)
1
<
2
are regular cardinals
(b) S
1
S

is stationary,

C
1
= C
1

: S
1
) is a (
1
, , )-Md-parameter
which has the (D

1
, , )-Md

-property, D

1
is a lter on
1
to which S
1
belongs
(c) S
2
S

is stationary and

C
2
= C
2

: S
2
) is a (
2
, , )-Md-parameter
(d) F :
2

1
satises: for every club E of
2
there is an increasing con-
tinuous function h from
1
into
2
with sup(Rang(h)) acc(E) such that
S
1
: h() S
2
and C
2
h()
h() : C
1

and F(h()) = belongs


to D

1
(e) 2
2

1
<
2
.
(
7
7
5
)


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







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


40 SAHARON SHELAH
Then

C
2
has the (, )-Md

-property.
2) Assume that in part (1) we omit assumption (e) and replace the Md-property by
MD

-property where 0, 2.
Then

C
2
has the (, )-MD

-property.
3) In part (1) if we omit assumption (e) and strengthen clause (d) by demanding
S
1
: h() S
2
and C
2
h(0)
= h() : C
1

and F(h()) = belongs to D

1
,
then we can still conclude that: for every nice (

C
2
, , )-colouring F (see below)
there is an F-middle diamond sequence.
Remark. 1) Is clause (d) of the assumption reasonable? Later in [Sh:F611, 4.5] we
prove that it holds under reasonable conditions.
2) This + 2.11 are formalized elsewhere.
Proof of 2.10. Similar to 2.1.
First, assume = 2 (so we are in part (2)) and let be a relational vocabulary
of cardinality < and by clause (b) of the assumption there is a sequence

M
1
=
M
1

: S
1
), M
1

a -model with universe C


1

such that

1
if M is a -model with universe
1
, then the set of S
1
for which M
1

=
M C
1

is ,= mod D

1
.
Now for every S
2
if F() S
1
and F() : C
2

C
1
f()
is an unbounded
subset of f(), then let M
2

be the unique -model with universe C


2

such that
F C
2

is an isomorphism from M
2

onto M
1
F()
F() : C
2

. If S
1
does
not satisfy the requirement above let M
2

be any -model with universe C


2

.
Now it is easy to check that

M
2
= M
2

: S
2
) is as required.
Second, assume = 1 (and we are in part (1)) and let F
2
be a (

C
2
, , ) colouring.
For each (

C
1
, , )-colouring F
1
we ask:
Question on F
1
: Do we have, for every club E of
2
an increasing continuous
function h :
1

2
such that
(i) sup(Rang(h)) E
(ii) the following set is a stationary subset of
1
S
1
: h() S
2
and C
2
h()
h() : C
1

and F(h()) = and


(
1

(C
1

)
)(
2

(C
2
h()
)
)[( C
1

)(
1
() =
2
(h())) F
1
(
1
) =
F
2
(
2
).
If for some F
1
the answer is yes, we proceed as in the proof for = 1. Otherwise,
for each such F
1
there is a club E[F
1
] exemplifying the negative answer. Let
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 41
E =

E[F
1
] : F
1
is a (

C
1
, , )-colouring. As there are 2
2

1
such functions
F
1
clearly E is a club of
2
and using (d) of the assumption, there is h :
1

2
as there. Now we can dene F
1
by h, F
2
naturally and get a contradiction.
The case = 0 and part (3) is left to the reader.
2.10
Trying to apply 2.10, clause (d) of the assumption looks the most serious. By the
following if each C
1

is a stationary subset of (so >


0
), it helps.
2.11 Claim. In 2.10, if 2

1
<
2
, and we replace (d) by (d)

+ (f) below, then


we can conclude that for some

C
3
= C
3

: S) has the -MD

-property where
C
3

(f)

D = D

: S
2
), D

is a lter on C
2

containing the co-bounded subsets


of C
2

(or just D

P(C
2

)A C
2

: sup(A) <
(d)

F :
2

1
satises: for every club E of for some increasing continuous
function h from
1
to
2
with sup(Rang(h)) acc(E) the set S
1
:
h() S
2
and h() : C
1

C
2

belong to D

1
.
Proof. Let C = e : e = e

: S
1
), e

is a subset of C
2

.
So [C[ 2

1
< and for each e E we dene

C
e
= C
e

: S
e
2
) by
S
e

= S
2
: C
e

where C
e

= C
2

: otp(C
2

) e
F()
. If for some
e C,

C
e
is as required.
2.11
(
7
7
5
)


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







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


42 SAHARON SHELAH
Glossary :
0 Introduction
Theorem 0.1: Many cases of MD-diamond.
Theorem 0.2: With division to .
Conclusion 0.3: Connection to the principle from [Sh 667].
Denition 0.4: Having (partial) squares.
1 Middle Diamond: sucient conditions
Denition 1.1: Denition of S being F -small and the normal lter.
Claim 1.2: The lter is normal and does not matter.
Denition 1.3: Recalling pcf denitions.
Remark 1.4:
Denition 1.5: The main notions are dened:

C is -Md-parameter, (, , )-Md-parameter,
F is a (

C, , )-coloring,

C is a (D, F)-middle diamond sequence,

C is good and

C has the (D, , )-Md-property or the
(D, , )-Md-property, also (good
+
-Md-property and Md

.
Denition 1.6:
Claim 1.8: Easy implications.
Major Claim 1.10: Sucient conditions for

C having the (D, )-Md-properties
(D a lter on , = cf(2

)).
Claim 1.11: E.g., =

implies Sep(, ), one of the assumptions of 1.10.


Remark 1.12: On variants.
Denition 1.9: Sep(, , ,
1
, ), Sep(, ).
Claim 1.13: Monotonicity properties of Sep.
Claim 1.14: On some implications.
Denition 1.15: -MD-parameter.
Claim 1.16: Existence of good MD-parameters.
Claim 1.17: Like 1.10 for MD-parameters (1.14 - 1.17 are not central).
Claim 1.20: Getting, e.g., (D, , )-MD-property, i.e., M

: S), [M

[ = C

which guess (as in Theorem 0.1) for having



C with the
(D, )-property; also md-properties MD-properties are dened.
Denition 1.22: -MD

-property and the ideal ID


,,,,
(

C)
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 43
Claim 1.23: Non saturation ideal above (needed in 1.20)
Claim 1.24: Quotations on the existence of good Md-parameters (by quoting).
Denition 1.25: A related ideal.
Denition 1.26: F is (S, , )-good (1.25, 1.26 not central).
Conclusion 1.27: Phrasing some conclusions.
Question 1.28: weakly inaccessible, 2

: < ) is eventually constant.


2 Upgrading to larger cardinals
Claim 2.1: For suitable
2
> 2

1
, we construct (
2
, , )-Md-parameters

C which
has the (D

, , )-property from (
1
, , )-Md-parameters which
has the (D

, , )-parameters.
Claim 2.3: Getting

C
i
= C
i

: S
i
) sets S
i
pairwise disjoint for i < in 2.1
Conclusion 2.4: If is strong limit, = cf() > 2
2

is successor of singulars
then for every large enough < , some good (, ,
+
)-Md-parameter
has the (D

, 2

, )-Md-property for every < .


Claim 2.5: A generalization of Engelking Karlowic. [no pf]
Claim 2.7: Dealing with successor of singulars. [ pf]
Claim 2.9: We advance the picture in 2.4 dealing with weakly inaccessible
(i.e. regular limit) not strong limit cardinals; at a minor price -
is of uncountable conality.
Claim 2.10: This is a stepping-up lemma used in 2.9 above.
Claim 2.11: Sucient conditions for the lifting used in 2.10.
(
7
7
5
)


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







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


44 SAHARON SHELAH
REFERENCES.
[AShS 221] Uri Abraham, Saharon Shelah, and R. M. Solovay. Squares with di-
amonds and Souslin trees with special squares. Fundamenta Mathe-
maticae, 127:133162, 1987.
[DvSh 65] Keith J. Devlin and Saharon Shelah. A weak version of which follows
from 2

0
< 2

1
. Israel Journal of Mathematics, 29:239247, 1978.
[DjSh 562] Mirna Dzamonja and Saharon Shelah. On squares, outside guessing
of clubs and I
<f
[]. Fundamenta Mathematicae, 148:165198, 1995.
math.LO/9510216.
[EK] Ryszard Engelking and Monika Karlowicz. Some theorems of set the-
ory and their topological consequences. Fundamenta Math., 57:275
285, 1965.
[Gr] C. Gray. Iterated Forcing from the Strategic Point of View. PhD thesis,
UC Berkeley, 1982.
[J] Thomas Jech. Set theory. Springer Monographs in Mathematics.
Springer-Verlag, Berlin, 2003. The third millennium edition, revised
and expanded.
[MPSh 713] Pierre Matet, Cedric Pean, and Saharon Shelah. Conality of normal
ideals on P

(), I. Archive for Mathematical Logic. math.LO/0404318.


[Sh:F611] Saharon Shelah. Around Very Weak Diamonds.
[Sh 638] Saharon Shelah. More on Weak Diamond. The East-West Journal of
Mathematics, accepted. math.LO/9807180.
[Sh 108] Saharon Shelah. On successors of singular cardinals. In Logic Collo-
quium 78 (Mons, 1978), volume 97 of Stud. Logic Foundations Math,
pages 357380. North-Holland, Amsterdam-New York, 1979.
[Sh 208] Saharon Shelah. More on the weak diamond. Annals of Pure and
Applied Logic, 28:315318, 1985.
[Sh 237e] Saharon Shelah. Remarks on squares. In Around classication theory
of models, volume 1182 of Lecture Notes in Mathematics, pages 276
279. Springer, Berlin, 1986.
[Sh 351] Saharon Shelah. Reecting stationary sets and successors of singular
cardinals. Archive for Mathematical Logic, 31:2553, 1991.
[Sh 420] Saharon Shelah. Advances in Cardinal Arithmetic. In Finite and
Innite Combinatorics in Sets and Logic, pages 355383. Kluwer Aca-
demic Publishers, 1993. N.W. Sauer et al (eds.). 0708.1979.
(
7
7
5
)


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







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


MIDDLE DIAMOND SH775 45
[Sh 410] Saharon Shelah. More on Cardinal Arithmetic. Archive for Mathe-
matical Logic, 32:399428, 1993. math.LO/0406550.
[Sh:g] Saharon Shelah. Cardinal Arithmetic, volume 29 of Oxford Logic
Guides. Oxford University Press, 1994.
[Sh 430] Saharon Shelah. Further cardinal arithmetic. Israel Journal of Math-
ematics, 95:61114, 1996. math.LO/9610226.
[Sh 462] Saharon Shelah. -entangled linear orders and narrowness of products
of Boolean algebras. Fundamenta Mathematicae, 153:199275, 1997.
math.LO/9609216.
[Sh 506] Saharon Shelah. The pcf-theorem revisited. In The Mathematics of
Paul Erd os, II, volume 14 of Algorithms and Combinatorics, pages
420459. Springer, 1997. Graham, Nesetril, eds.. math.LO/9502233.
[Sh:f] Saharon Shelah. Proper and improper forcing. Perspectives in Math-
ematical Logic. Springer, 1998.
[Sh 589] Saharon Shelah. Applications of PCF theory. Journal of Symbolic
Logic, 65:16241674, 2000.
[Sh 460] Saharon Shelah. The Generalized Continuum Hypothesis revisited.
Israel Journal of Mathematics, 116:285321, 2000. math.LO/9809200.
[Sh 576] Saharon Shelah. Categoricity of an abstract elementary class in two
successive cardinals. Israel Journal of Mathematics, 126:29128, 2001.
math.LO/9805146.
[Sh 755] Saharon Shelah. Weak Diamond. Mathematica Japonica, 55:531538,
2002. math.LO/0107207.
[Sh 667] Saharon Shelah. Successor of singulars: combinatorics and not col-
lapsing cardinals in (< )-support iterations. Israel Journal of
Mathematics, 134:127155, 2003. math.LO/9808140.
[Sh 829] Saharon Shelah. More on the Revised GCH and the Black Box. Annals
of Pure and Applied Logic, 140:133160, 2006. math.LO/0406482.

You might also like