Documente Academic
Documente Profesional
Documente Cultură
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED
SH506
DEDICATED TO PAUL ERD
OS
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem, Israel
Rutgers University
Mathematics Department
New Brunswick, NJ USA
Abstract. The pcf theorem (of the possible conability theory) was proved for
reduced products
Q
i<
i
/I, where < min
i<
i
. Here we prove this theorem
under weaker assumptions such as wsat(I) < min
i<
i
, where wsat(I) is the minimal
such that cannot be delivered to sets / I (or even slightly weaker condition).
We also look at the existence of exact upper bounds relative to <
I
(<
I
-eub) as
well as cardinalities of reduced products and the cardinals T
D
(). Finally we apply
this to the problem of the depth of ultraproducts (and reduced products) of Boolean
algebras.
Partially supported by the Deutsche Forschungsgemeinschaft, grant Ko 490/7-1. Publication no.
506.
I would like to thank Alice Leonhardt for the beautiful typing.
First Typed - File saharon.top, May 10, 1993; converted to TeX April 9, 2002
Latest revision - 08/Feb/10
Typeset by A
M
S-T
E
X
1
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
2 SAHARON SHELAH
0 Introduction
An aim of the pcf theory is to answer the question, what are the possible co-
nalities (pcf) of the partial orders
i<
i
/I, where cf(
i
) =
i>
, for dierent ideals
I on . For a quick introduction to the pcf theory see [Sh 400a], and for a detailed
exposition, see [Sh:g] and more history. In 1 and 2 we generalize the basic theo-
rem of this theory by weakening the assumption < min
i<
i
to the assumption
that I extends a xed ideal I
with wsat(I
) < min
i<
i
, where wsat(I
) is the
minimal such that cannot be divided to sets / I
) [Dom(I
)[
+
(and
reg
(I
) [Dom(I
)[
+
so [Sh:g, I,1,2,II,1,VII,2.1,2.2,2.6] are really a special
case of the proofs here.
During the sixties the cardinalities of ultraproducts and reduced products were
much investigated (see Chang and Keisler [\CK ]). For this the notion regular lter
(and (, )-regular lter) were introduced, as: if
i
0
, D a regular ultralter
(or lter) on then
i<
i
/D = (liminf
D
i
)
) and J
<
[
] where usually
=
i
: i < ) is
a sequence of regular cardinals, I
-positive set (so they are pairwise disjoint, not just disjoint
modulo I
). In 1.5, 1.8 we give the basic properties. In lemma 1.9 we phrase the
basic property enabling us to do anything: (1.9 ()): essentially if liminf
I
(
)
wsat(I
) and
/I
is
+
-directed then we prove that
/J
<
[
] is -directed. In
1.11, 1.13 we deduce more properties of J
<
[
] : pcf(
]
-increasing sequence f
]
-bound in
. In 1.14 we
prove pcf(
) has a last element. In 1.13 we deal with the connection between the
true conality of
i<
i
/D
and
i<
i
/E when
i
=: tcf(
i<
i
/D
i
) and D
is the
E-limit of the D
i
s.
In 2.1 we dene normality of for
: J
] = J
<
[
] +B
OS 3
normality: J
] = J
<
[
] + B
: < where B
/J
<
[
] is increasing. We
then (in 2.2) characterize semi-normality (there is a <
J
<
[
]
-increasing
f = f
:
< ) conal in
/D) = ) and when semi normality implies normality (if some such
f has a
<
J
<
[
]
eub).
We then deal with continuity system a and <
J
<
[
]
-increasing sequence obeying
a, in a way adapted to the basic assumption () of 1.9.
Here as elsewhere if min(
)
+
our life is easier than when we just assume
limsup
I
(
) ,
/I
is
+
-directed (where wsat(I
), of course). In 2.3 we
give the denitions, in 2.5 we quote existence theorem, show existence of obedient
sequences (in 2.7), essential uniqueness (in 2.10) and better consequence to 1.12
(in the direction to normality). We dene (2.12) generating sequence and draw a
conclusion (2.13(1)). Now we get some desirable properties: in 2.11 we prove semi
normality, in 2.13(2) we compute cf(
/I
) as max pcf(
) >
+
, we get the strongest version (including normality
using 2.9, i.e. obedience). Lastly, we try to map the implications between the
various properties when we do not use the basic assumption 1.9 () (in fact there
are considerable dependence, see 2.18, 2.19).
In 3.1, 3.3 we present measures of regularity of lters, in 3.2 we present measures
of hereditary conality of
i<
i
/D and
in 3.10 we give a reasonable upper bound by hereditary conality ( (
/D +
hcf
D,
(
i<
I
))
<
when reg
(D)).
In 3.13 - 3.14 we return to existence of eubs and obedience (Saharon, new point
over 2.9) and in 3.15 draw conclusion on downward closure.
In 3.16 - 3.17 we estimate T
D
(
) > =
0
). We also mention
1
-complete lters; (3.19, 3.20) and see what
can be done without relaying on pcf (3.23)).
Now we deal with depth: dene it (3.21, see 3.22), give lower bound (3.25), com-
pute it for ultraproducts of interval Boolean algebras of ordinals (3.27). Lastly we
translate the problem does
i
< Depth
+
(B
i
) for i < implies < Depth
+
(
i<
B
i
/D)
at least when > 2
0
< ], to a pcf problem (in 3.29).
In the last section we phrase a reason 1.9() works (see 4.1), analyze the case we
weaken to 1.9() to lim inf
I
(
) wsat(I
i
is regular, Min(
) = Min
i
: i < (of course also in
we can replace
by another set), and let
i<
i
; usually we are assuming
is regular. Let
] = A
: < ) = A
,
[
= i < :
i
> .
But we can replace by any set (in the denitions and claims). Let I
denote a
xed ideal on .
3) For I a lter on let I
+
= P() I (similarly D
+
= A : A / D), let
lim inf
I
= min : i < :
i
I
+
and
lim sup
I
= Min : i < :
i
> I and
atom
I
= : i :
i
= I
+
.
4) For a set A of ordinals with no last element, J
bd
A
= B A : sup(B) < sup(A),
i.e. the ideal of bounded subsets.
5) Generally, if inv(X) = sup[y[ : (X, y) then inv
+
(X) = sup[y[
+
: [X, y],
and any y such that [X, y] is a wittness for [y[ inv(X) (and [y[ < inv
+
(X)),
and it exemplies this.
6) Let Ord be the class of ordinals.
7) Considering
i<
f(i), considering
i<
f(i)/I formally if (i)f(i) = 0 then
i<
f(i) =
; but we usually ignore this, particularly when i : f(i) = 0 I.
1.2 Denition. 1) For a partial order
1
P:
(a) P is -directed if: for every A P, [A[ < there is q P such that
pA
p q, and we say: q is an upper bound of A;
1
actually we do not require p q p p = q so we should say quasi order
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 5
(b) P has true conality if there is a sequence p = p
: < ) increasing
and conal in P, i.e.:
<
p
< p
and q P[
<
q p
]. We write
tcf(P) = for the minimal such , in fact it is unique and we say that p
witness = tcf(P). (Note: if P is linearly ordered it always has a true
conality but, e.g., (, <) (
1
, <) does not)
(c) P is called endless if p Pq P[q > p] (so if P is endless, in clauses (a),
(b), (d) above we can replace by <)
(d) A P is a cover (of P) if: p Pq A[p q]; we also say A is conal
in P
(e) cf(P) = min[A[ : A P is a cover
(f) We say that, in P, p is a lub (least upper bound) of A P if:
() p is an upper bound of A (see (a))
() if p
2) If D is a lter on S,
s
(for s S) are ordinals, f, g
sS
s
, then: f/D < g/D,
f <
D
g and f < g mod D all mean s S : f(s) < g(s) D. Also if f, g are
partial functions from S to ordinals, D a lter on S then f < g mod D means
i Dom(D) : i / Dom(f) or f(i) < g(i) (so both are dened) belongs to D.
We write X = A mod D if Dom(D) [(XA) (AX)] belongs to D. Similarly for
, and we do not distinguish between a lter and the dual ideal in such notions. So
if J is an ideal on and f, g
sS
f(s) < g(s); similarly f g. So
(
; similarly f or
i<
f(i).
4) If I is an ideal on , F
Ord, we call g
Ord an
I
-eub (exact upper bound)
of F if:
() g is an
I
-upper bound of F (in
Ord)
() if h
Ord, h <
I
Maxg, 1 then for some f F, h < maxf, 1 mod I
() if A , A ,= mod I and [f F f A =
I
0
A
, i.e., i A : f(i) ,= 0
I] then g A =
I
0
A
.
5a) We say the ideal I (on ) is -weakly saturated if cannot be divided to
pairwise disjoint sets from I
+
(which is P() I).
5b) wsat(I) = Min : I is -weakly saturated.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
6 SAHARON SHELAH
1.3 Observation. 1) Concerning 1.2(4), note: g
= Maxg, 1 means g
(i) =
Maxg(i), 1 for each i < ; if for every f F, i < : f(i) = 0 I we
can replace Maxg, 1, Maxf, 1 by g, f respectively in clause () and omit clause
().
2) The ideal I on is -weakly saturated i in the topological space of the ultral-
ters on the subspace D : D an ultralter on disjoint to I has spread < , or
is a limit ordinal, it has spread but the spread is not obtained (hence 2
cf()
but it is consistently singular, see [Sh 233], [JuSh 231]).
1.4 Denition. Below if is the ultralters disjoint to I, we write I instead of
. Recall that we can replace by any set.
1) For a property of ultralters (if is the empty condition, we omit it):
pcf
) = pcf(
, ) = tcf(
) = tcf(
/J) : J
is an ideal on satisfying such that
/D) .
3) J
, ] = J
<
+[
, ].
4) pcf
, I) = tcf(
B) = pcf
I+(\B)
(
) (so if B I it is ), also J
<
(
B, I)
P(B) is dened similarly.
6) If I = I
), J
<
[
, <
J
) and (
,
J
) are endless (even when each
i
is just a
limit ordinal).
1) min(pcf
I
(
)) liminf
I
(
) for
regular.
2)
(i) If B
1
B
2
are from I
+
then pcf
I
(
B
1
) pcf
I
(
B
2
);
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 7
(ii) if I J then pcf
J
(
) pcf
I
(
); and
(iii) for B
1
, B
2
we have pcf
I
(
(B
1
B
2
)) = pcf
I
(
B
1
)
pcf
I
(
B
2
).
Also
(iv) A J
<
[
(B
1
B
2
)] A B
1
J
<
[
B
1
] & A B
2
J
<
[
B
2
]
(v) if A
1
, A
2
I
+
, A
1
A
2
= , A
1
A
2
= , and tcf(
, <
I
) = for
= 1, 2 then tcf(
, <
I
) = ; and if the sequence
f = f
: < ) witness
both assumptions then it witness the conclusion.
3)
(i) if B
1
B
2
, B
1
nite and
regular then
pcf
I
(
B
2
) Rang(
B
1
) pcf
I
(
(B
2
B
1
)) pcf
I
(
B
2
)
(ii) if in addition i B
1
i
< Min(Rang[
(B
2
B
1
)]),
then pcf
I
(
B
2
) Rang(
B
1
) = pcf
I
(
(B
2
B
1
)).
4) Let
be regular (i.e., each
i
is regular);
(i) if = liminf
I
then
/I is -directed
(ii) if = liminf
I
is singular then
/I is
+
-directed
(iii) if = liminf
I
is a regular uncountable cardinal, for some club E of ,i <
:
i
E or
i
= I then
/I is
+
-directed. We can weaken the
assumption to I is not lowly normal for (,
, <
I
) has true
conality
(v) If
)
(vi) pcf
I
(
=: limsup
I
(
). Then pcf
I
(
) (limsup
I
(
))
+
; in fact for some ideal J extending
I,
/J is (limsup
I
(
))
+
-directed.
6) If D is a lter on a set S and for s S,
s
is a limit ordinal then:
(i) cf(
sS
s
, <
D
) = cf(
sS
cf(
s
), <
D
) = cf(
sS
(
s
, <)/D), and
(ii) tcf(
sS
s
, <
D
) = tcf(
sS
(cf(
s
), <
D
)) = tcf(
sS
(
s
, <)/D).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
8 SAHARON SHELAH
In particular, if one of them is well dened, then so are the others. This is true
even if we replace
s
by linear orders or even partial orders with true connality.
7) If D is an ultralter on a set S,
s
a regular cardinal, then =: tcf(
sS
s
, <
D
)
is well dened and pcf(
s
: s S).
8) If D is a lter on a set S, for s S,
s
is a regular cardinal, S
=
s
: s S
and
E =: B : B S
and s :
s
B D
and
s
> [S[ or at least
s
> [t :
t
=
s
[ for any s S then:
(i) E is a lter on S
(ii) S
) > [S
[,
(iii) F = f
sS
s
:
s
=
t
f(s) = f(t) is a cover of
sS
s
,
(iv) cf(
sS
s
/D) = cf(S
sS
s
/D) = tcf(S
/E).
9) Assume I is an ideal on , F
Ord and g
Ord. If g is a
I
-eub of F then
g is a
I
-lub of F.
10) sup pcf
I
(
) [
/I[.
11) If I is an ideal on S and (
sS
s
, <
I
) has true conality as exemplied by
f = f
S
Ord for <
= cf(), f
: < ) is <
I
-increasing with <
I
-eub f then tcf(
i
f(i), <
I
) =
tcf(cf[f(i)], <
I
) = .
13) If I J are ideals on then
(a) wsat(I) wsat(J)
(b) liminf
I
(
) liminf
J
(
)
(c) if = tcf(
i<
i
, <
I
) then = tcf(
i<
i
, <
J
).
14) If f
1
, f
2
are <
I
-lub of F then f
1
=
I
f
2
.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 9
1.6 Denition. 1) Let I is not almost normal for (,
,
for no j < is i < :
i
h(i) < j = mod I.
2) Let I is not lowly normal for (,
, for no < , is
i < :
i
h(i) < I
+
.
Remark. Note that weakly normals implies lowly normal.
Proof. They are all very easy, e.g.
0) We shall show (
, <
J
) is endless (assuming, of course, that J is a proper ideal
on ). Let f
too as
each
), and let h
witness
this. Without loss of generality
i
> h(i) . So assume that f
for
< . We now dene a function f with domain by
f(i) = f
(i) <
i
. So f
.
Second, for any < we have
i < : (f
I
f. Together
we are done. To nish we need
if there is a club E of and i :
i
E or
i
= I then I is not medium
normal for (,
).
[Why ? Without loss of generality
i
, we dene a function h with
domain , h(i) = sup(E
i
) if
i
/ E and h(i) = 0 if
i
E .
So i < h(i) <
i
hence h
), so we are done.
5) Let
=: lim sup
I
(
) and dene
J =: A : for some <
) = lim inf
J
(
) =
.
Case 1:
is
0
.
We do not use the J above. Now the desired conclusion fails then every ultralter
on disjoint to I is
1
-complete. Now if i < :
i
>
0
J
+
the construction
is immediate so without loss of generality i <
i
=
0
. But not weakly
normal for (,
) then j < A
j
=: i < : h(i) < j , = mod I but
A
j
: j < = . There is an ultralter D on disjoint to J A
j
: j < so
A
j
: j < ) exemplies D is not
1
-complete.
Case 2:
is singular.
By part (4), clause (ii),
/I is (
)
+
-directed and by part (4) clause (v) we get
the desired conclusion.
Case 3:
is regular >
0
.
Let h
) and let
J
= A : for every h
the set A
=: i A :
i
> I
hence for every h
. So J J
. Also
J
, A = A
0
A
1
; for every h
, choose j
0
, j
1
<
such that A
=: i A
: h(i) < j
= A
mod I, so j =: maxj
0
, j
1
<
and
A
]. Also / J
[why? as h
witness
that I is not weakly normal for (
)]. So together J
is an ideal on extending
I. Now J
), as witnessed by h
.
[Why? Let us check Denition 1.6(2), so let <
=
i < :
i
h(i) < / J
+
; now A
1
= i < :
i
> J J
and
A
2
hence A
1
A
2
J but it includes A
, so we are
done.]
Lastly,
/J
is (
)
+
-directed (by part (4) clause (iii)), and so pcf
J
(
) is disjoint
to (
)
+
.
9) Let us prove g is a
I
-lub of F in (
Ord,
I
). As we can deal separately with
I + A, I + ( A) where A =: i : g(i) = 0, and the later case is trivial we can
assume A = . So assume g is not a
I
-lub, so there is an upper bound g
of F,
but not g
I
g
. Dene g
Ord:
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 11
g
(i) =
_
0 & if g(i) g
(i)
g
(i) & if g
<
I
g. So, as g in an
I
-eub of F for I, there is f F such that
g
<
I
maxf, 1, but B =: i : g
) so
g
B = g
B <
I
maxf, 1 B. But we know that f
I
g
(as g
is an
upper bound of F) hence f B
I
g
B =
I
0
B
; as g
is a
I
-upper bound of F we know
[f
F f
B =
I
0
B
], hence by () of Denition 1.2(4) we have g B =
I
0
B
, a
contradiction to B / I (see above).
1.5
1.7 Remark. In 1.5 we can also have the straight monotonicity properties of
pcf
I
(
, ).
1.8 Claim. 1) J
<
[
] J
<
[
].
3) If is singular, J
<
[
] = J
<
+[
] = J
].
4) If / pcf(
), then J
<
[
] = J
].
5) If A , A / J
<
[
], and f
A, f
: < ) is <
J
<
[
]
-increasing
conal in (
A)/J
<
[
] then A J
].
Also this holds when A , f
: < ) is <
J
-increasing conal in (
A)/J
for any ideal J on such that I
J J
], A / J.
6) The earlier parts hold for J
<
[
, ], too.
Proof. Straight.
1.9 Lemma. Assume
()
is regular and
() Min(
) > wsat(I
) wsat(I
), and
/I
is
+
-directed.
2
2
note, if cf() < then
+
-directed follows from -directed which follows from lim
inf
I
(
) , i.e. rst part of clause (). Note also that if clause () holds then
/I
is
+
-directed (even (
, <) is
+
-directed), so clause () implies clause ().
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
12 SAHARON SHELAH
If is a cardinal , and / J
<
[
] then (
, <
J
<
[
]
) is -directed (remember:
J
<
[
] = J
<
[
, I
]).
1.10 Remark. Note that above () () so in any case also (
,
I
) is
+
-
directed.
Proof. Note: if f
then f < f + 1
, (i.e., (
, <
J
]
) is endless) where
f + 1 is dened by (f + 1)(i) = f(i) + 1). Let F
such that:
(a) f
0
f
1
mod J
<
[
]
(b) for < we have f
1
f
1
mod J
<
[
].
If is singular, there is C unbounded, [C[ = cf() < , and by the induction
hypothesis there is g
g mod J
<
[
]. Now g is as
required: f
0
f
1
f
1
min(C\)
g mod J
<
[
=: i < :
i
> [[ for < , so < < A
and < A
= mod I
= () < and B
=: i < : f
1
(i) > g
(i)
(iv) for each < , for every [
+1
, ), B
,= B
+1
mod J
<
[
].
We cannot carry this denition: as letting () = sup
()
A
+1
,= B
+1
()
A
+1
mod J
<
[
] for
< (by (iv) and as A
+1
= mod I
and I
J
<
[
]) and B
()
(by (iii))
and [ < B
()
A
()
] (by (ii)), together A
+1
(B
()
B
+1
()
) : < )
is a sequence of pairwise disjoint members of (I
)
+
, a contradiction
3
Now for = 0 let g
be f
1
0
and
= 0.
3
i.e we have noted that for no B
( < ) do we have: B
= B
+1
mod I
where A
= mod I
(e.g., A
= A
).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 13
For limit let g
(i) =
_
<
g
(i) for i A
as
< ,
i
> for i A
and
is a sequence of regular cardinals) and let
= 0.
For = + 1, suppose that g
hence B
J
<
[
] for
unboundedly many < (hence for every < ) then g
] and the proof is complete. So assume this fails, then there is a minimal
() < such that B
()
, J
<
[
]. As B
()
/ J
<
[
] we have B
()
D and cf(
/D) .
But < . Hence f
1
/D where h
. Let
us dene g
:
g
(i) = Maxg
(i), h
(i).
Now (i), (ii) hold trivially and B
mod J
<
[
]
clearly B
()
B
mod J
<
[
]. Moreover J
<
[
()
D implies B
/ J
<
[
].
On the other hand B
is i < : f
1
(i) > g
(i) >
g
(i), h
mod D).
We can conclude B
/ D, whereas B
] as
required in clause (iv).
Now we have said that we cannot carry the denition for all < , so we are stuck
at some ; by the above is successor, say = + 1, and g
is as required: an
upper bound for F modulo J
<
[
].
1.9
1.11 Claim. If () of 1.9, D is an ultralter on disjoint to I
and = tcf(
, <
D
), then for some B D,(
B, <
J
<
[
]
) has true conality . (So B J
]
J
<
[
] by 1.8(5).)
Proof. As (
,
I
) is
+
-directed (by 1.9) clearly
+
. By the denition of
J
<
[
] clearly D J
<
[
] = .
Let f
/D (so f
). By 1.9
without loss of generality ( < )(f
< f
mod J
<
[
]).
Now 1.11 follows from 1.12 below: its hypothesis clearly holds. If
_
<
B
=
mod D, (see (A) of 1.12) then (see (D) of 1.12) J D = hence (see (D) of 1.12)
g/D contradicts the choice of f
D; by (C)
of 1.12 we get the desired conclusion.
1.11
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
14 SAHARON SHELAH
1.12 Lemma. Suppose () of 1.9, cf() > , f
, f
< f
mod J
<
[
] for
< < , and there is no g
< g mod J
<
[
].
Then there are B
, J
<
[
]
(B) < B
mod J
<
[
] (i.e. B
J
<
[
])
(C) for each , f
: < ) is conal in (
, <
J
<
[
]
) (better restrict
yourselves to () (see (A)) so that necessarily B
/ J
<
[
]);
(D) for some g
,
_
<
f
g mod J where
4
J = J
<
[
] +B
: < ;
in fact
(D)
+
for some g
g mod (J
<
[
] + B
), in
fact B
= i < : f
(i) g
(i) f
(i)] = mod J
<
[
]
(hence for every large enough < this holds)
(F) if is a limit ordinal < , f
is a
J
<
[
]
-lub of f
: < then B
is a
lub of B
: < in P()/J
<
[
].
Proof. Remember that for < , A
= i < :
i
> [[ so A
= mod I
and
< A
, () < , B
: < )
such that:
(i) g
(iii) B
=: i : f
(i) > g
(i)
(iv) if () < then B
,= B
+1
mod J
<
[
].
For = 0 let g
= f
0
, and () = 0.
For limit let g
(i) =
_
<
g
(i) if i A
]
holds trivially and g
as each
i
is regular and [i A
i
> ]), and let
() = 0.
4
Of course, if B
= mod J
<
[
OS 15
For = + 1, if < : B
J
<
[
] is unbounded in , then g
is a bound for
f
: < ) mod J
<
[
mod J
<
[
]
hence < : B
J
<
[
/ J
<
[
] (of course,
we may increase () later). If B
and of
(), clause (B) holds by B
s denition as < f
< f
mod J
<
[
], (D)
+
holds with g = g
by the choice of B
such that
<
B
B mod J
<
[
] and
B
B / J
<
[
=: (g
(B))(f
B) contradicts f
is a
J
<
[
]
-lub of f
: < , because: g
(obvious), (f
mod J
<
[
])
[why? as B
B / J
<
[
] and g
(B
B) = g
(B
B) < f
(B
B) by the
choice of B
B
J
<
[
]
f
B = g
B and
f
( B)
J
<
[
]
g
( B) = g
( B)
(the
J
<
[
]
holds as ( B) B
J
<
[
). So only
clause (C) (of 1.12) may fail, without loss of generality for = (). I.e. f
()
: < ) is not conal in (
()
, <
J
<
[
]
). As this sequence of functions
is increasing w.r.t. <
J
<
[
]
, there is h
()
) such that for no < do we
have h
B
j
()
mod J
<
[
]. Let h
= h
0
(\B
()
)
and g
be dened
by g
(i) = Maxg
(i), h
()
A
= mod I
and B
+1
()
,= B
()
mod J
<
[
] so B
()
A
+1
B
+1
()
: < ) is a sequence of
pairwise disjoint members of (J
<
[
])
+
hence of (I
)
+
which give the contradiction
to () of 1.9; so the lemma cannot fail.
1.12
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
16 SAHARON SHELAH
1.13 Lemma. Suppose () of 1.9 and < .
1) For every B J
] J
<
[
], we have:
(
B, <
J
<
[
]
); has true conality ; (hence is regular).
2) If D is an ultralter on , disjoint to I
, then tcf(
/D) is min : DJ
] ,=
.
3)(i) For a limit cardinal J
<
[
] =
_
<
J
<
[
], hence
(ii) For every , J
<
[
] =
<
J
].
4) J
] ,= J
<
[
] i J
] J
<
[
] ,= i pcf(
).
5) J
]/J
<
[
] is -directed (i.e. if B
] for <
] we have B
B mod J
<
[
.)
Proof. 1) Let
J = B :B J
<
[
] or B J
] J
<
[
] and
(
B, <
J
<
[
]
) has true conality .
By its denition clearly J J
/D)
+
, then B / J
]);
contradiction.
On the other hand if F
such that (f
F)(f < g mod J
<
[
] J,
D J = ), and this implies cf(
C), <
J
<
[
]
_
has true
conality , of course C B C and C B D hence C B / J
<
[
]. Clearly
if C
C, C
/ J
<
[
] then also (
, <
J
<
[
]
) has true conality , hence by
the last sentence without loss of generality C B; hence by 1.8(5) we know that
C J
)
+[
]. So [ < B , J
]] (by
the choice of ) hence by 1.13(3)(ii) below, we have B / J
<
[
]. It similarly follows
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 17
that D J
<
[
] = . Now (
B, <
J
<
[
]
) has true conality by 1.11. As we
know that B DJ
], and J
<
[
<
J
<
[
, <
J
) is -directed;
i.e., if
< and f
: <
such that
( <
)(f
<
+
, as () of 1.9 holds, this is obvious by 1.9. So without loss of generality
+
and
= cf(
such that
<
[
+
<
. By 1.9, there is f
such
that ( <
)(f
< f mod J
<
[
]). Since J
<
[
] J, it is immediate that
( <
)(f
] J
<
[
]
_
<
J
<
[
, <
J
) is -directed and DJ = , one has tcf(
/D) ,
but B D J
<
[
: < ) be <
J
<
[
]+(\B
)
-increasing and conal in
mod J
<
[
] +
(
) (for <
, as a
<
J
<
[
]
-bound to f
: < f
: <
, such f
: < ), getting B
: < ), now B
1.13
1.14 Conclusion. If () of 1.9, then pcf(
]. [ exists, since
=: [
[
: J
] , = ] and by 1.5(10).
1.14
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
18 SAHARON SHELAH
1.15 Claim. Suppose () of 1.9 holds. Assume for j < , D
j
is a lter on
extending A : A I
, E a lter on and D
= B : j < : B D
j
E (a lter on ). Let
j
=: tcf(
, <
D
j
) be well dened for j < , and assume
further
j
> + .
Let
= tcf(
, <
D
), = tcf(
j<
j
, <
E
).
Then = (in particular, if one is well dened, then so is the other).
Proof. Without loss of generality . (Why? Otherwise we can add
j
=:
0
,
D
j
=: D
0
for j , and replace by and E by E
= A : A E).
Let f
j
: <
j
) be an <
D
j
-increasing conal sequence in (
, <
D
j
).
Now = 0, 1, for each f
, dene G
(f)
j<
j
by G
mod D
j
and if = 0 then: not f
j
f mod D
j
(it is well
dened for f
by the choice of f
j
: <
j
)).
Note that for f
1
, f
2
B(f
1
, f
2
) =: i < : f
1
(i) f
2
(i) D
A(f
1
, f
2
) =: j < : B(f
1
, f
2
) D
j
E
for some A E, for every i A we have f
1
D
i
f
2
for some A E for every i A we have
G
(f
1
)(i) G
(f
2
)(i)
G
(f
1
) G
(f
2
) mod E.
So
1
G
is a mapping from (
,
D
) into (
j<
j
,
E
) preserving order.
Next we prove that
2
for every g
j<
j
for some f
, we have g G
0
(f) mod E.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 19
[Why? Note that min
j
: j <
+
+
and J
] J
]. By 1.9 we
know (
, <
J
]
) is
+
-directed, hence for some f
:
()
1
for j < we have f
j
g(j)
< f mod J
].
We here assumed <
j
, hence J
] J
<
j
[
] is disjoint
to D
j
by the denition of J
<
j
[
3
for f
we have G
0
(f) G
1
(f).
[Why? Read the denitions].
4
if f
1
, f
2
and G
1
(f
1
) <
E
G
0
(f
2
) then f
1
<
D
f
2
.
[Why? As G
1
(f
1
) <
E
G
0
(f
2
) there is B E such that: j B
G
1
(f
1
)(j) < G
0
(f
2
)(j). For each j we have f
1
D
j
f
j
G
1
(f
1
)(j)
by
the denition of G
1
(f
1
)) and f
j
G
1
(f
1
)(j)
<
D
j
f
2
(as G
1
(f
1
)(j) < G
0
(f
2
)(j)
and the denition of G
0
(f
2
)(j)) so together f
1
<
D
j
f
2
. So A(f
1
, f
2
) =
i < : f
1
(i) < f
2
(i) satises: A(f
1
, f
2
) D
j
for every j B, hence
A(f
1
, f
2
) D
) hence f
1
<
D
f
2
as required.]
Now rst assume = tcf(
, <
D
) is well dened, so there is a sequence
f =
f
: < ) of members of
, <
D
-increasing and conal. So G
0
(f
) : < )
is
E
-increasing in
j<
j
(by
1
), for every g
j<
j
for some f
we
have g
E
G
0
(f) (why? by
2
), but by the choice of
f for some < we have
f <
D
f
hence by
1
we have g
E
G
0
(f)
E
G
0
(f
), so G
0
(f
) : < ) is
conal in (
j<
j
, <
E
). Also for every < , applying the previous sentence to
G
1
(f
) + 1 (
j<
j
) we can nd < such that G
1
(f
) + 1
E
G
0
(f
), so
G
1
(f
) <
E
G
0
(f
) : C) is <
E
-increasing
conal in (
j<
j
, <
E
). So if is well dened then = tcf(
j<
j
, <
E
) is well
dened and equall to .
Lastly, assume that is well dened i.e.
j<
j
/E has true conality , let
g = g
and
ordinals
such that
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
20 SAHARON SHELAH
(i) f
(ii) g
<
E
G
0
(f
)
E
G
1
(f
) <
E
g
(so
<
)
(iii)
1
<
2
<
1
<
2
(so
).
In stage , rst choose
1
+ 1 :
1
< , then choose f
such
that g
+1 <
E
G
0
(f
) (possible by
2
) then choose
such that G
1
(f
) <
E
g
.
Now G
0
(f
)
E
G
1
(f
) by
3
. By
4
we have
1
<
2
f
1
<
D
f
2
. Also if
f
then G
1
(f)
j<
j
hence by the choice of g, for some < we have
G
1
(f) <
E
g
but
so G
1
(f) <
E
g
E
G
0
(f
) hence by
4
, f <
D
f
.
Altogether, f
, <
D
) has true conality , so is
well dened and equal to .
1.15
1.16 Conclusion. If () of 1.9 holds, and , =
j
: j < ), D
j
: j < ) are
as in 1.15 and + < min( ), and J is an ideal on and I an ideal on such
that I
j<
( A) D
j
(e.g. I = I
) then pcf
J
(
j
: j < ) pcf
I
(
).
Proof. Assume pcf
J
(
j
: j < ). Let E be an ultralter on disjoint to J
such that = tcf(
j<
j
/E) then we can dene an ultralter D
on as in 1.15,
so clearly D
) as required.
1.16
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 21
2 Normality of pcf(
) for
] is
increasing continuous in , the question is how J
<
[
], J
<
+[
] are related.
The simplest relation is J
<
+[
] = J
<
[
].
We give a sucient condition for exsitence of such B, using this in 2.11; giving the
necessary denition in 2.3 and needed information in 2.5, 2.7, 2.9; lastly 2.10 is the
essential uniqueness of conal sequences in appropriate
/I.
2.1 Denition. 1) We say pcf(
) is normal (for
) if for some B ,
J
] = J
<
[
] + B.
2) We say pcf(
) is semi-normal (for
) if there are B
mod J
<
[
] and
(ii) J
] = J
<
[
] +B
: < .
3) We say
is normal if every pcf(
) is normal for
. Similarly for semi
normal.
4) In (1), (2), (3) instead
we can say (
, I) or
/I or (
, <
I
) if we replace I
)).
2.2 Fact. Suppose () of 1.9 and pcf(
).
Now:
1) is semi-normal for
i for some F = f
: <
we have: [ <
f
< f
mod J
<
[
], F is
unbounded in (
, <
D
) whenever tcf(
, <
D
) = .
2) In 2.1(2), without loss of generality, we may assume that
either: B
= B
0
mod J
<
[
] (so is normal)
or: B
,= B
mod J
]
-exact upper bound g
i<
(
i
+ 1) and then
B =: i < : g(i) =
i
generates J
] over J
<
[
].
4) If is semi normal for
then for some
f = f
: < ),
B = B
: < )
we have:
B is increasing modulo J
<
[
], J
] = J
<
[
] + B
: < , and
f
: < ) is <
J
<
[
]
-increasing and
f,
B as in 1.12.
Proof. 1) For the direction , given B
, <
J
<
[
]
) has true conality , and let it be
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
22 SAHARON SHELAH
exemplied by f
such that: , f
J
<
[
]
f
and < f
<
J
<
[
]
f
.
Now F =: f
< f
mod J
<
[
and (
, <
D
) has
true conality , then by 1.11 for some B J
] J
<
[
] we have B D, so by
the choice of B
mod J
<
[
] hence B
D.
As f
J
<
[
]
f
, <
D
).]
The other direction, follows from 1.12 applied to F = f
: < . [Why?
By 1.12 there is a sequence B
]
increasing modulo J
<
[
] so J =: J
<
[
] +B
: < J
].
If equality does not hold then for some ultralter D over , D J = but
D J
, <
D
) = contradicting the assumption on F.]
2) Because we can replace B
: < ) by B
i
: i < ) whenever
i
: i < ) is
non decreasing, non eventually constant.
3) If is normal for
, let B be such that J
] = J
<
[
] + B. By 1.13(1)
we know that (
B), <
J
<
[
]
) has true conality , so let it be exemplied by
f
0
: < ). Let f
= f
0
0
(\B)
for < . Now f
: < ) is as required by
1.5(11).
Now suppose f
]
-eub of F,
g
i<
(
i
+ 1) and B = i : g(i) =
i
. Let D be an ultralter on disjoint to
J
<
[
, let f
= (f B) 0
(\B)
, now necessarily
f
(i) <
i
= g(i)] and [i B f
(i) = 0 g(i)]),
hence (see Denition 1.4(4)) for some < we have f
< maxf
, 1 mod J
<
[
]
hence for some < , f
mod J
<
[
] hence f f
mod D; also
< f
< f
, <
D
) = . If B / D then B D so g
= g ( B) 0
B
= g mod D and
< f
<
D
f
+1
D
g =
D
g
, so g
exemplies F is bounded in
(
, <
D
) so as F is as in 2.2(1), tcf(
, <
D
) = is impossible. As D is disjoint
to J
<
[
], necessarily tcf(
, <
D
) > . The last two arguments together give, by
1.13(2) that J
] = J
<
[
= A
OS 23
2.3 Denition. Let there be given regular , < < , possibly an ordinal,
S , sup(S) = and for simplicity S is a set of limit ordinals or at least have no
two successive members.
1) We call a = a
, otp(a
) <
, and [ a
= a
: otp(a
is (, ) E = . We say a
is continuous in S
if S
= sup(a
).
2) Assume f
Ord for < and
A
= A
. We say
f = f
: < ) obeys a = a
: < )
for
A
if:
(i) for a
, if =: otp(a
(note:
A
determine ).
2A) Let ,
, I
be as usual. We say
f obeys a for
A
continuously on S
if: a
is continuous in S
and
f obeys a for
A
and in addition S
S and for S
= f
a
(i) = supf
(i) : a
when [a
[ <
i
.
2B) For given
=
i
: i < ),
f = f
: < ) where f
and a , and
let f
a
be dened by: f
a
(i) is 0 if [a[
i
and f
= a
is replaced by S & a
( < )[a
[ < [a
+
cf() <
and assume = cf() >
+
r
. Then for
some stationary S < : cf() =
r
, there is a continuity condition a for
(S,
r
); moreover, it is continuous in S and S otp(a
) =
r
; so for every
5
Note: if otp(a
) = and = sup(a
& a
(, ) E ,= ].
2) Assume =
++
, then for some stationary S < : cf() = cf() there is
a continuity condition for (S, + 1, ).
3) If a is a (, ,
1
)-continuity condition and
1
then there is a (, + 1, )-
continuity condition.
Proof. 1) By [Sh 420, 1].
2) By [Sh 351, 4.4](2) and
6
.
3) Check.
2.5
2.6 Remark. Of course also if =
+
the conclusion of 2.5(2) may well hold. We
suspect but do not know that the negation is consistent with ZFC.
2.7 Fact. Suppose () of 1.9, f
)) and
A
=
A
= i < :
i
> ).
Then
1) Assume a is a -weak continuity condition for (S, ), = sup(S), then we can
nd
f
= f
,
(ii) for < we have f
<
J
<
[
]
f
(iv)
f
obeys a for
A
.
2) If in addition min(
) > , S
= f
(ii) for S
we have f
and = +1 S
& S
<
J
<
[
]
f
(iv)
f
obeys a for
A
continuously on S
.
3) Suppose f
and g
and [ S
] then
.
6
the denition of B
i
in the proof of [Sh:g, III,2.14](2) should be changed as in [Sh 351, 4.4](2),
[Sh:g, III,2.14](2),clause(c),p.135-7
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 25
4) If < , for < we have
f
= f
: < ), where f
, then in 2.7(1)
(and 2.7(2)) we can nd f
by f
(and f
by f
.
Proof. Easy (using 1.9 of course).
2.8 Claim. In 2.7 we can replace () from 1.9 by
/J
<
[
] is -directed.
2.9 Claim. Assume () of 1.9 and let
A
be as there.
1) In 1.12, if f
/J
<
[
]/J
<
[
] is
+
-directed (hence if is semi normal for
then it is normal to
).
Proof. 1) Assume not, so for some club E of we have
() < < & E B
,= B
mod J
<
[
].
As a is a -weak (S, )-continuity condition, there is S such that b =:
a
: otp(a
,
+1
) E, and let
<
+1
be such that a
and otp(a
) < ;
by shrinking and renaming without loss of generality
<
and
. Let
() =: otp(a
).
Lastly, let B
0
=: i < : f
(i) < f
(i) < f
(i) < f
+1
(i), clearly it is
= mod I
=: A
()+1
(B
) B
0
, now
B
mod J
<
[
,= B
by () above
hence B
,= mod J
<
[
]. Now B
0
, A
()+1
= mod I
by the previous
sentence and by 1.9() which we are assuming respectively and I
J
<
[
] by
the laters denition; so we have gotten B
,= mod J
<
[
= , for suppose i B
, so i A
()+1
and also f
(i) <
f
+1
(i) f
(i) (as i B
0
and as
+1
a
& i A
()+1
respectively); now
i B
hence i B
hence i , B
, contradiction. So B
: < )
is a sequence of pairwise disjoint members of (J
<
[
])
+
, contradiction.
2) The proof is similar to the proof of 1.13(4), using 2.9(1) instead 1.12 (and a from
2.5(1) if >
+
r
or 2.5(2) if =
++
).
2.9
We note also (but shall not use):
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
26 SAHARON SHELAH
2.10 Claim. Suppose () of 1.9 and
(a) f
) and
f = f
: < ) is <
J
<
[
]
-increasing
(b)
f obeys a continuously on S
], and f
/J : < ) is conal in
(
, <
J
) (e.g., J = J
<
[
] + ( B), B J
] J
<
[
]).
(d) f
for S
(alternatively: f
: < cf()) is a
-increasing sequence of members, of J then
_
< cf()
A
J).
Then:
(A) the set
< : if S
and otp(a
) = then f
= f
mod J
contains a club of
(B) the set
< : if S and = sup( a
) and otp( a
) =
then f
= f
a
mod J
contains a club of .
Proof. We concentrate on proving (A).
Suppose S
, and f
,= f
mod J. Let
A
1,
= i < : f
(i) < f
(i)
A
2,
= i < : f
(i) > f
(i).
So A
1,
A
2,
J
+
, suppose rst A
1,
J
+
. By Denition 2.3(2A), for every
i A
1,
for every large enough a
, f
(i) < f
i
. As
J is cf()-indecomposable for some < we have i < :
i
< J
+
so
f
A
1,
< f
A
1,
(and < ). Now by clause (c), E =: < : for every <
we have f
< f
OS 27
E A
1,
J.
If
<
f
E
do we have f
,= f
hence [ E A
2,
J] and we are done.
2.10
We now return to investigating the J
<
[
, <
I
) is
+
-directed hence min pcf
I
(
)
+
(by 1.5(4)(v)) hence let us dene by induction on ,
f
= f
: < ), B
and
D
such that:
(I)(i) f
mod J
<
[
]
(iii) < & < f
mod J
<
[
]
(iv) for < < and < : f
(II)(i) D
/D
) =
(ii) f
/D
/D
(iii) f
+1
/D
/D
; moreover
(iii)
+
B
and f
+1
/(J
<
[
] + (
B
))
(iv) f
+1
0
/D
is above f
/D
: < .
For = 0: No problem. [Use 1.13(1)+(4)].
For limit < : Let g
be dened by g
(i) = supf
and f
such that g
and < f
< f
mod J
<
[
]. This is
possible by 1.9 and clearly the requirements (I)(i),(ii),(iv) are satised.
Use 2.2(1) to nd an appropriate D
: < )
and D
are as required.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
28 SAHARON SHELAH
For = : Choose f
] J
<
[
]. Let g
: < ) be
conal in (
, <
D
) and even in (
, <
J
<
[
]+(\B
)
) and without loss of generality
_
<
f
/D
< g
0
/D
and
<
f
. We get f
: < ) increasing
and conal mod (J
<
[
] + ( B
)) such that g
by 1.9 from g
: < ).
Then get D
s and D
: < ), f
of such that:
() < C
< f
+1
mod J
<
[
]
So C =:
<
C
such that
()
1
g f
mod J
<
[
] for <
by 1.9 without loss of generality
()
2
f
0
< g mod J
<
[
] for <
For each < , by II (iii), (iii)
+
for some
< we have
()
3
< g B
< f
+1
mod J
<
[
]
Let () = sup
<
, so () < and so
()
4
< g B
< f
+1
()
B
mod J
<
[
]
For < , let B
= i A
: g(i) < f
()
(i). By ()
4
, clearly
+1
mod
J
<
[
], but
by (II)(iii)
+
hence B
+1
D
; by (II)(iv)+()
2
we know
B
/ D
, hence B
,= B
+1
mod D
hence B
,= B
+1
mod J
<
[
].
On the other hand by (I)(iv) for each < we have B
: ) is -
increasing and (as A
= mod J
<
[
/I
: < ) is
-increasing, and by the previous sentence B
,= B
+1
mod J
<
[
] hence B
/I
:
< ) is strictly -increasing. Together clearly B
+1
A
+1
B
: < ) is a
sequence of pairwise disjoint members of (J
<
[
])
+
, hence of (I
)
+
; contradiction
to wsat(I
).
2.11
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 29
2.12 Denition. 1) We say
B = B
and c pcf(
)
(ii) J
] = J
<
[
] + B
for each c.
2) We call
B = B
: c) smooth if:
i B
&
i
c B
i
B
.
3) We call
B = B
: Rang(
i < :
i
pcf(
).
4) We call
B = B
).
2.13 Fact. Assume () of 1.9.
1) Suppose c pcf(
),
B = B
mod I
.
2.14 Remark. For another proof of 2.13(2) see 2.17(2) + 2.17(4) and for another
use of the proof of 2.13(2) see 2.19(1).
Proof. 1) If not, then I = I
+B
_
d
B
i<
i
/D); necessarily pcf(
B), hence by
the last assumption of 2.13(1) we have c. By 1.13(2) we know B
D hence
B B
/I
) = max pcf(
).
2) The case =
0
is trivial (as wsat(I
)
0
implies P()/I
is a Boolean
algebra satisfying the
0
-c.c. (as here we can substract) hence this Boolean algebra
is nite hence also pcf(
)
+
let (B) = max pcf
I
B
(
B).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
30 SAHARON SHELAH
We prove by induction on that for every B (I
)
+
, cf(
, <
I
+(\B)
) = (B)
when (B) ; this will suce (use B = and = [
i<
i
[
+
). Given B let
= (B), by renaming without loss of generality B = . By 1.14, pcf(
) has a
last element, necessarily it is =: (B). Let f
: < ) be <
J
<
[
]
increasing
conal in
/J
<
[
) cf(
/I
). Let us prove
the other inequality. For A J
<
[
] I
choose F
A
which is conal in
/(I
+(A)), [F
A
[ = (A) < (exists by the induction hypothesis). Let be a
large enough regular, and we now choose by induction on < , N
, g
such that:
(A)(i) N
(H (), , <
) sn
(ii) |N
| =
(iii) N
: ) N
+1
(iv) N
, I
N
0
, f
: < ) N
0
and the function
A F
A
belongs to N
0
(B)(i) g
and g
N
+1
(ii) for no f N
do we have g
<
I
f
(iii) < &
i
> [[ g
(i) < g
(i).
There is no problem to dene N
exemplies cf(
, g
< f
()
mod J
<
[
] hence () <
g
<
J
<
[
]
f
= i < : g
(i) f
: < )
is increasing with , (by clause (B)(iii)), hence as usual as wsat(I
) (and
>
0
) we can nd () < such that
_
n
B
()+n
= B
()
mod I
[why do we
not demand ((), ) B
= B
()
mod I
N
0
N
()+1
clearly, by its denition, B
()
N
()+1
hence F
B
()
N
()+1
. Now:
g
()+1
( B
()
) =
I
g
()+1
( B
()+1
) < f
( B
()+1
)
=
I
f
( B
()
).
[Why rst equality and last equality? As B
()+1
= B
()
mod I
OS 31
But g
()+1
B
()
iB
()
i
, and B
()
J
<
[
] as g
< f
()
f
mod J
<
[
] so for some f F
B
()
we have g
()+1
B
()
< f B
()
mod I
mod I
Now f
N
()+1
and f N
()+1
(as f F
B
()
, [F
B
()
[ , + 1 N
()+1
the function B F
B
belongs to N
0
N
()+1
and B
()
N
()+1
as g
()
, f
N
()+1
) so together
() maxf, f
N
()+1
;
But (), () together contradict the choice of g
()+1
(i.e., clause (B)(ii)).
2.15
> scite2.10A ambiguous
2.16 Denition. 1) We say that I
, ) if:
(a)
/I
is -directed and
(b) for every pcf
I
(
), (
, <
J
<
[
]
) is -directed and
(c) we can nd B
: pcf
I
(
B
() B
,
() J
<
[
, I
] = I
+B
: pcf
I
(
),
() B
/ J
<
[
, I
] and
() (
)/J
<
[
, I
]
J
<
[
] and J
] = J
<
[
] + B
).
1A) We say that I
, ) if:
(a) (
, <
I
) is -directed
(b) (
, <
J
<
[
]
) is -directed for each pcf
I
(
)
(c) there are B
,
for < pcf
I
(
) such that
() < < pcf
I
(
) B
,
B
,
mod J
<
[
, I
]
() J
<
[
] = I
+B
,
: < < , pcf
I
(
) and
() (
, <
J
<
[
]
) is -directed and
() ((
B
,
), <
J
<
[
]
) has true conality .
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
32 SAHARON SHELAH
1B) We say that I
, ) if:
(a) (
, <
I
) is -directed
(b) each (
, <
J
<
[
) is -directed
(c) for any ultralter D on disjoint to J
<
[
] letting = tcf(
, <
D
) we
have: and for some B D J
] J
<
[
B), <
J
<
[
]
) has true conality .
1C) We say that I
, ) if:
(a) (
, <
I
) is -directed and
(b) (
, <
J
<
[
]
) is -directed for any .
1D) Above we write
instead (
, ) when we mean
= max : (
, <
I
) is -directed.
2) We say that I
) , we have: I
satises
the pcf-th above if it satises the pcf-th for
with lim inf
I
(
) > . Similarly
(in both cases) for the weak pcf-th and the weaker pcf-th.
3) Given I
, let
J
pcf
= A : A I
or A / I
and I
=: A : wsat(I
A) or A I
similarly J
wpcf
; we may write J
x
[I
].
4) We say that I
, for some A I
+
we have ((
A), <
I
) has a true conality.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 33
2.17 Claim. 1) If () of 1.9 then I
,
+
).
2) If () of 1.9 holds, and
/I
is
++
-directed (e.g.,
+
< min
) or just there
is a continuity condition for (
+
, )) then I
,
+
).
3) If I
, ) then I
, ) which
implies that I
satises
the weakest pcf-th for (
, ).
Proof. 1) Let appropriate
be given. By 1.9, 1.13 most demands holds, but we
are left with normality. By 2.11, if pcf(
), then
is semi normal for . This
nishing the proof of (1).
2) Let pcf(
) and let
f,
B be as in 2.2(4). By 2.5(1)+(2) there is a, a (, )-
continuity condition; by 2.7(1) without loss of generality
f obeys a, by 2.9(1) the
relevant B
/I
, <
I
) is given (but possibly () of 1.9 fails).
1) If I
,
satises (the conclusions of ) 1.11, then I
,
satisfy (the conclusion of )
1.13(1), 1.13(2), 1.13(3), 1.13(4), 1.14.
1A) If I
,
satises (the conclusion of ) 1.9 then I
,
satises (the conclusion of )
1.15.
2A) If I
,
satisfy the conclusion of 1.9.
3) If I
,
satises 1.9, 1.11 then I
,
satises 2.2(1) (for 2.2(2) - no assumptions).
4) If I
,
satises 1.13(1), 1.13(2) then I
,
satises 2.2(3).
5) If I
,
satises 1.13(2) then I
,
satises 2.13(1).
6) If I
satisfy 1.13(1) + 1.13(3)(i) then I
,
satises 1.13(2).
7) If I
,
satises 1.13(1) + 1.13(2) and is semi normal then 2.13(2) holds, i.e.,
cf(
, <
I
) sup pcf
I
().
Proof. 1) We prove by parts.
Proof of 1.13(2). Let = tcf(
] = . Also
by 1.11 for some B D we have = tcf((
B), <
J
<
[
]
), so by the previous
sentence B / J
<
[
], together we nish.
Proof of 1.13(1). Repeat the proof of 1.13(1) replacing the use of 1.9 by 1.13(2).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
34 SAHARON SHELAH
Proof of 1.8(3)(i). Let J =:
<
J
<
[
], so J J
<
[
] is an ideal because J
<
[
] :
< ) is -increasing (by 1.8(2)), if equality fail choose B J
<
[
] J and choose
D an ultralter on disjoint to J to which B belongs. Now if = cf() < then
+
< (as is a limit cardinal) and = cf() &
+
< D J
] =
D J
<
+[
] D J
<
[
/D).
Together contradiction by 1.5(7).
Proof of 1.13(3)(ii). Follows.
Proof of 1.13(4). Follows.
Proof of 1.14. As in 1.14.
1) Check.
2) Read the proof of 1.15.
2A) Check.
3) The direction is proved directly as in the proof of 2.2(1) (where the use of
1.13(1) is justied by 2.18(1)).
So let us deal with the direction . So assume
f = f
: < ) is a sequence
of members of
which is <
J
<
[
]
-increasing such that for every ultralter D
on disjoint to J
<
[
] we have: = tcf(
, <
D
) i
f is unbounded (equiva-
lently conal) in (
, <
D
). By (the conclusion of) 1.9 without loss of generality
f
is <
J
<
[
]
-increasing, and let
J =: A : A J
<
[
] or
f is conal in (
, <
J
<
[
]+(\A)
.
Clearly J is an ideal on (by 1.5(2)(v)), and J
<
[
] J J
]. If J ,= J
<
[
]
choose A J
]. By 1.9 there is g
such that f
< g mod J
=: i < : g(i) f
(i). Hence B
/J
<
[
: < ) is <
J
<
[
]
-increasing).
Lastly if B J
], but B B
/ J
<
[
] +B
, g
D
) =
(by 1.13(3) which holds by 2.17(1)) but f
: < )
is as required.
4) 6) Left to the reader.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 35
7) For pcf(
) let B
i
: i < ) be such that J
] = J
<
[
] + B
i
: i <
(exists by semi-normality; we use only this equality). Let f
,i
: < ) be conal
in ((
i
), <
J
,
[
]
), it exists by 1.13(1). Let F be the closure of f
,i
: <
, i < , pcf(
), so
it suce to prove that F is a cover of (
, <
I
). Let g
, if (f F)(g f)
we are done, if not
I = A i < : f(i) > g(i) : f F, A I
is
0
-directed, / I, so there is an ultralter D on disjoint to I, (so f F
g <
D
f) and let = tcf(
] J
<
[
] ,= ,
hence for some i < , B
i
D, and we get contradiction to the choice of the
f
,
: < ( F).
2.18
2.19 Claim. If I
, <
I
) = sup pcf
I
(
)
(2) We can nd (J
) : <
),
,
J
+1
= A : if A / J
then tcf((
A), <
J
) =
and for no
A (J
)
+
does ((
A), <
J
(3) If I
.
Proof. 1) Similar to the proof of 2.13(2).
2) Check (we can also present those ideals in other ways).
3) Check.
2.19
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
36 SAHARON SHELAH
3 Reduced products of cardinals
We characterize here the cardinalities
i<
i
/D and T
D
(
i
: i < )) using pcfs
and the amount of regularity of D (in 3.1 - 3.4). Later we give sucient conditions
for the existence of <
D
-lub or <
D
-eub. Remember the old result of Kanamori
[Kn] and Ketonen [Kt]: for D an ultralter the sequence /D : < ) (i.e., the
constant functions) has a <
D
-lub if reg(D) < ; and see [Sh:g, III,3.3] (for lters).
Then we turn to depth of ultraproducts of Boolean algebras.
The questions we would like to answer are (restricting ourselves to
i
2
or
i
2
2
i<
i
/D :
i
a cardinal for i < , i.e.,
characterize it by properties of D; (or at least Card
D
2
i<
i
) :
i
a cardinal for i < is natural).
Question B: What can be DEPTH
+
D
= Depth
+
(
i<
i
/D) :
i
a regular cardinal
(at least DEPTH
+
D
2
0
< ),
which can be removed if the cov = pp problem is completed (see [Sh:g, AG]). So
the problem is for the other ultralters D, on which we give a reasonable amount on
information translating to a pcf problem, sometimes depending on the pcf theorem.
3.1 Denition. 1) For a lter D let reg(D) = Min : D is not -regular (see
below).
2) A lter D is -regular if there are A
-s is empty.
3) For a lter D let
reg
D
+
for < such that
no i < belongs to innitely many A
s
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 37
and
reg
D
+
for < such that :
< A
s.
4) reg
(D), reg
D
+
. Of course, reg(I),
etc., means reg(D) where D is the dual lter.
3.2 Denition. 1) Let
htcf
D,
(
i
) = suptcf(
i<
i
/D) :
i
= cf
i
i
for i < and
tcf(
i
/D) is well dened
and
hcf
D,
(
i<
i
) = supcf(
i
/D) :
i
= cf
i
i
;
if =
0
we may omit it.
2) For E a family of lters on let hcf
E,
(
i<
i
) be
suptcf(
i<
i
/D) :D E and
i
= cf
i
i
for i < and
tcf(
i<
i
/D) is well dened.
Similarly for hcf
E,
(using cf instead of tcf).
3) hcf
D,
(
i<
i
) is hcf
E,
(
i<
i
) for E = D
: D
a lter on extending D.
Similarly for htcf
D,
.
4) When we write I, e.g., in hcf
I,
we mean hcf
D,
where D is the dual lter.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
38 SAHARON SHELAH
3.3 Claim. 1) reg(D) is always regular.
2) If < reg
(D).
4) reg(D) reg
(D) reg
(D).
5) reg
(D) reg
(E).
Proof. Should be clear. E.g.
2) Let u
: < D
+
exemplify
< reg
: u
. Now D
is a
lter on extending D and for < we have A
D.
Finally, the intersection of A
0
A
1
. . . for distinct
n
< is empty, because
for any memeber j of it we can nd
n
< such that j A
n
and
n
u
n
. Now if
n
: n < is innite then there is no such j by the choice of A
: < ), and if
n
: n < is nite then without loss of generality
n<
n
=
0
contradicting u
0
is nite as
n<
n
u
n
.
3.3
3.4 Observation. [
i<
i
/I[ [
0
/I[ holds when
i<
i
0
.
3.5 Observation. 1) [
i<
i
/I[ htcf
I
(
i<
i
).
2) If I
/I
) = max pcf
I
(
).
3) If I
) then
hcf
D,
(
) = hcf
D,
(
) = htcf
D,
(
)
whenever D is disjoint to I
.
4) hcf
E,
(
i<
i
) = hcf
E,
(
i<
i
).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 39
5)
i<
i
/I hcf
I,
(
i<
i
) = hcf
I,
(
i<
i
) htcf
I,
(
i<
i
) and hcf
I,
(
i<
i
)
htcf
I,
(
i<
i
).
3.6 Remark. In 3.5(3) concerning htcf
D,
see 3.13.
Proof. 1) By the denition of htcf
I
it suces to show [
i<
i
/I[ tcf(
i
/I
),
when I
is an ideal on extending I,
i
= cf
i
i
for i < and tcf(
i<
i
/I
)
is well dened. Now [
i<
i
/I[ [
i<
i
/I[ [
i<
i
/I
[ cf(
i
/I
), so we have
nished.
2) By 2.18(1) and 1.14 and 2.19.
3) Left to the reader (see Denition 2.16(2)).
4), 5) Check.
3.5
3.7 Claim. If = [
i<
i
/I[ (and
i
0
and, of course, I an ideal on ) and
< reg(I) then =
.
Proof. For each i < , let
i
: <
i
) list the nite sequences from
i
. Let
M
i
= (
i
, F
i
, G
i
) where F
i
() = lg(
i
), G
i
(, ) is
i
() if < g(
i
) (= F
i
()),
and F(, ) = 0 otherwise; let M =
i<
M
i
/I so |M| = [
i
/I[ and let M =
(
i
/I, F, G). Let A
i
: i < ) exemplies I is -regular. Now
()
1
We can nd f
and f
i<
f(i) for < such that: < < f
<
I
f
(i) = [ w
i
[ < f(i), and note < &i A
(i) < f
(i)].
()
2
For every sequence g = g
: < ) of members of
i<
i
, there is h
i<
i
such that < M F(h/I, f
/I) = g
/I.
[Why? Let, in the notation of ()
1
, h(i) be such that
i
h(i)
= g
(i) : w
i
)
(in the natural order).]
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
40 SAHARON SHELAH
So in M, every -sequence of members is coded by at least one member so |M|
=
|M|, but |M| = [
i<
i
/I[ hence we have proved 3.7.
3.7
3.8 Fact. 1) For D a lter on , A
1
, A
2
) a partition of and (non zero) cardinals
i
for i < we have
[
i<
i
/D[ = [
i<
i
/(D + A
1
)[ [
i<
i
/(D +A
2
)[
(note: [
i<
i
/P()[ = 1).
2) D
[]
=: A : [
i<
i
/(D+(A))[ < is a lter on ( an innite cardinal
of course) and if
0
i<
i
/D then D
[]
is a proper lter.
3) If [
i<
i
/I[, (
i
innite, of course, I an ideal on ) and A I
+
iA
i
/I[ and < reg
(I) then [
i
/I[
.
Proof. Check (part (3) is like 3.7).
3.9 Claim. If D E are lters on then
[
i<
i
/D[ [
i<
i
/E[ + sup
AE\D
[
i<
i
/(D + ( A))[ + (2
/D) +
0
.
We can replace 2
OS 41
3.10 Lemma. [
i<
i
/D[ (
/D+hcf
D,
(
i<
i
))
<
(see Denition 3.2(1)) pro-
vided that:
() reg
(D)
3.11 Remark. 1) If =
+
1
, we can replace
/D by
1
/D. In general we can
replace
/D by sup
i<
f(i)/D : f
.
2) If D satises the pcf-th above (see 2.16(1A), 2.17(2)) then by 3.5(3) we can
use htcf
(sometime even htcf, see 3.13). But by 3.8(1) we can ignore the
i
,
and when i < 2
i
> we know that 1.9(*)() holds by 3.3(3).
Proof. Let =
/D + hcf
D,
(
i<
i
). Let for < ,
=:
, i.e.,
=:
(
/D + hcf
D,
i<
i
)
||
, clearly
=
||
. Let =
8
(sup
i<
i
)
+
and N
(H (), , <
) be such that |N
| =
, N
||
N
, + 1 N
and D,
i
: i <
) N
and [ < N
]. Let N = N
: < . Let g
i<
i
and
we shall nd f N such that g
(e < 3) and
A
such that:
(a) f
e
i<
(
i
+ 1)
(b) f
1
and f
2
(c)
A
= A
i
: i < ) N
(d)
i
A
i
i
+ 1, [A
i
[ [[ + 1, and A
i
: < ) is increasing continuous
(in )
(e) f
0
(i) = Min(A
i
g
(i) <
i
A
i
(f) f
1
= f
0
mod D
(g) g
< f
2
< f
1
mod (D +i < : g
(i) ,= f
1
(i))
(h) if g
(i) ,= f
1
(i) then f
2
(i) A
+1
i
.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
42 SAHARON SHELAH
So assume everything is dened for every < . If = 0, let A
i
=
i
, if limit
A
i
=
<
A
i
, for = + 1, A
i
will be dened in stage . So arriving to ,
A
is
well dened and it belongs to N
as we have N
||
and < N
/D. As A
i
: i < ) N
, [A
i
[ < and
/D < + 1 N
,
clearly [
i<
[A
i
[/D[ hence f/D : f
i<
A
i
N
hence f
0
/D N
hence
there is f
1
such that f
1
f
0
f
0
clearly
g
f
1
mod D, let y
0
=: i < : g
(i) f
1
(i), y
1
=: i < : i / y
0
and
cf(f
1
2
=: y
0
y
1
. So y
e
: e < 3) is a partition of and g
< f
1
mod (D + y
e
) for e = 1, 2.
Let y
4
= i < : cf(f
1
(i)) so f
1
, and N
hence y
4
N
, so
(
i<
f
1
(i), <
D+y
4
) N
. Now
cf(
i<
f
1
(i), <
D+y
4
) hcf
D+y
4
,
(
i<
i
) hcf
D,
(
i<
i
) < + 1 N
hence there is F N
, [F[ , F
iy
4
f
1
iy
4
f
1
4
)))].
As + 1 N necessarily F N
2
) 0
(\y
2
)
and get f
4
F N such that g
< f
4
mod (D + y
2
). Now use similarly
i<
cf(f
1
(i))/(D + y
1
) [
i<
f
1
(y
1
+ y
2
) < f
2
mod D. Let A
+1
i
be: A
i
if
i y
0
and A
i
f
2
(i) if i y
1
y
2
.
It is easy to check clauses (g), (h). So we have carried the denition.
Let
X
=: i < : f
0
+1
(i) < f
0
(i).
Note that by the choice of f
1
, f
1
+1
we know X
= y
1
y
2
mod D, if this last set is
not D-positive then g
f
1
mod D, hence g
/D = f
1
/D N
, contradiction, so
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 43
y
1
y
2
,= mod D hence X
D
+
. Also y
1
y
2
: < ) is -decreasing hence
X
/D : < ) is -decreasing.
Also if i X
1
X
2
and
1
<
2
then f
0
2
(i) f
0
1
+1
(i) < f
0
1
(i) (rst inequality:
as A
1
+1
i
A
2
i
and clause (e) above, second inequality by the denition of X
1
),
hence for each ordinal i the set < : i X
(D),
contradiction to the assumption ().
3.10
Note we can conclude
3.12 Claim.
i<
i
/D = sup(
i<
f(i))
<reg
(D
1
)
+ hcf
D
1
(
i<
i
)
<reg
(D
1
)
: D
1
is a lter on
extending D such that
A D
+
1
i<
i
/(D
1
+ A) =
i<
i
/D
1
and f
, f(i)
i
Proof. The inequality should be clear by 3.8(3). For the other direction let
be the right side cardinality and let D
1
= A : if A D
+
then
i<
i
/D ,
so we know by 3.8(2) that D
1
is a lter on extending D. Now
0
/D (by
the term (
i
f(i)/D
1
)
<reg
(D
1
)
) so by 3.9 we have
i<
i
/D
1
> . By 3.10 (see
3.11(1)) we get a contradiction.
3.12
Next we deal with existence of <
D
-eub.
3.13 Claim. 1) Assume D a lter on , g
Ord for < , g
= g
: < )
is
D
-increasing, and
() cf() reg
(D).
Then at least one of the following holds:
(A) g
(D)
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
44 SAHARON SHELAH
(C) for some club C of and some
1
< and
i
<
+
1
and w
i
Ord of order
type
i
for i < , there are f
i<
w
i
(for C) such that f
(i) =
min(w
i
g
D
f
& f
=
D
f
& f
D
g
& g
.
2) In (C) above if for simplicity D is an ultralter we can nd w
i
Ord, otp(w
i
) =
i
,
i<
w
i
such that
f
<
D
h
<
D
f
+1
, moreover,
i<
i
< .
Proof. 1) Let = reg
, f
,
(for
< ),
A
such that:
(a)
A
= A
i
: i < )
(b) A
i
= f
,
(i), g
(i)] + 1
(c) f
,
(i) = Min(A
i
g
(i)) (and f
,
Ord, of course)
(d)
is the rst ,
_
<
,
; moreover, g
< maxf
,
, 1
< maxg
, 1 mod D
Let
).
Note
() < <
&
< f
,
=
D
f
,
& f
,
f
,
& f
,
,=
D
f
,
.
[Why last phrase? applying clause (e) above, second phrase with , here standing
for , there we get A
0
=: i < : maxg
(i), 1 g
(i) D
+
and applying
clause (e) above rst phrase with here standing for there we get A
1
= i < :
g
(i) < f
,
(i) or g
(i) = 0 = f
,
(i) D, hence A
0
A
1
D
+
, and g
(i) g
(i) f
,
(i)
g
(i) g
(i) < f
,
(i)
f
,
(i) g
(i) < f
,
(i), together f
,
,=
D
f
,
as required.]
Case A:
= and
_
<
< .
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 45
Let () =
<
= i < : f
(),
(i) ,= f
(),+1
(i) , =
mod D. Now for i < , f
(),
(i) : < ) is non increasing so i belongs to nitely
many y
s only, so y
(D).
Case B:
= and
_
<
= .
So possibility (B) of Claim 3.13 holds.
Case C:
< .
Still A
i
(i < ), f
,
( < ) are well dened.
Subcase C1:
cannot be dened.
Then possibility C of 3.13 holds (use w
i
=: A
i
, f
= f
+,
).
Subcase C2:
can be dened.
Then f
,
is a <
D
-eub of g
as required in
clause (e). Now f
,
is almost as required in possibility (A) of Claim 3.13 only
the second phrase is missing. If for no
1
< , i < : cf[f
,
(i)]
1
D
+
,
then possibility (A) holds.
So assume
1
< and B =: i < :
0
cf[f
,
(i)]
1
belongs to D
+
,
we shall try to prove that possibility (C) holds, thus nishing. Now we choose w
i
for i < : for i we let w
0
i
=: f
,
(i), [sup
<
g
,
(i) of order type cf[f
,
(i)] and for i B let
w
1
i
= , lastly let w
i
= w
0
i
w
1
i
, so [w
i
[
1
as required in possibility (C). Dene
f
Ord by f
(i) = min(w
i
g
: < ) is
D
-increasing; if for some
, ) we have
f
/D = f
Ord by:
g
B = f
(which is < f
,
),
g
( B) = 0
\B
.
Now g
. So
there is no
,=
D
f
,
so we have actually proved possibility (C).
2) Easy (for
i
< , without loss of generality = reg
(D) =
reg(D) so
1
< reg(D)).
3.13
3.14 Claim. 1) In 3.13(1), if = = cf(), g
there are
1
< reg
(D) and A
D
+
for < such that the intersection of
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
46 SAHARON SHELAH
any
+
1
of the sets A
1
)[i A
] (reminds
(,
+
1
)-regularity of ultralters).
2) We can in 3.13(1) weaken the assumption () to ()
where
()
cf() reg(D)
(A)
there is a
D
-upper bound f of g
<
D
f (of course f
Ord) is a
D
-upper bound of g
: <
and lim inf
D
cf[f(i)] : i < ).
3) If g
Ord, g
: < ) is <
D
-increasing and f
Ord satises (A)
above
and
()
cf() wsat(D) and for some A D for every i < , cf(f(i)) wsat(D)
then for some B D
+
we have
i<
cf[f(i)]/(D + B) has true conality cf().
Remark. Compare with 2.9.
Proof. 1) By the choice of a = a
: < ) list a
: otp(a
) <
(or just a subset of it) we have (
,
+1
) C ,= .
Let
,
+1
) C, and
,
+1
) be such that
: a
,
and as we can use
2
: < ), without loss of generality
<
= i < : f
(i) < f
(i) < f
(i) < f
+1
(i) and supf
(i) + 1 : < + 1.
2) In the proof of 3.13 we replace clause (e) by
(e
) g
,
and for < we have f
mod D.
3) By 1.13(1).
3.14
3.15 Claim. 1) Assume = tcf(
<
D
,
(D), min(
) > reg
(D)
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 47
2) Let I
be the ideal dual to D, and assume () above. If ()() of 1.9 holds and
is semi-normal (for (
, I
)) then it is normal.
Proof.
Case 1 < lim inf
D
(
).
We let
=
_
if <
i
1 if
i
and we are done.
Case 2: lim inf
D
(
) reg
(D))[reg
(D) < ].
Let =: reg
for
< such that g
= g
)).
Now if in claim 3.13(1) for g
(D), reg
(I) , contradiction.
Case 3: lim inf
D
(
) reg
(D))[reg
(D) < ].
Like the proof of [Sh:g, Ch.II,1.5B] using the silly square.
We turn to other measures of
/D.
3.16 Denition.
(a) T
0
D
(
) = sup[F[ : F
and f
1
,= f
2
F f
1
,=
D
f
2
(b)
T
1
D
(
) = Min[F[ :(i) F
(ii) f
1
,= f
2
F f
1
,=
D
f
2
(iii) F maximal under (i) + (ii)
(c) T
2
D
(
) = Min[F[ : F
, for some f
2
F we
have f
1
,=
D
f
2
(d) If T
0
D
(
) = T
1
D
(
) = T
2
D
(
) then let T
D
(
) = T
l
D
(
) for l < 3
(e) for f
Ord and < 3 let T
l
D
(f) means T
l
D
(f() : < )).
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
48 SAHARON SHELAH
3.17 Theorem. 0) If D
0
D
1
are lters on then T
D
0
(
) T
D
1
(
) for = 0, 2.
Also if = A
0
A
1
, A
0
D
+
, and A
1
D
+
then T
D
(
) = minT
D+A
0
(
), T
D+A
1
(
)
for = 0, 2.
1) htcf
D
(
) T
2
D
(
) T
1
D
(
) T
0
D
(
).
2) If T
0
D
(
) = T
1
D
(
) = T
2
D
(
) > 2
suce).
3) T
0
D
(
)
<reg(D)
= T
0
D
(
) (each
i
innite of course).
4) [htcf
D
i<
f(i)] T
2
D
(f) [htcf
D
i<
f(i)]
<reg(D)
+ wsat(D)
/D.
5) If D is an ultralter [
/D[ = T
e
D
(
) for e 2.
6) In (4), if
i<
f(i) 2
/D min
i<
f(i)), the second and
third terms are equal.
7) If the sup in the denition of T
0
D
(
) < htcf
D
(
); then we can nd
= cf(
)
(, htcf
D
(
)] and =
i
: i < ), a sequence of regular cardinals,
i<
i
i
such that
: <
i<
i
be : g
(i) =
_
g(i) if g(i) <
i
0 otherwise.
So there is (g) <
such that g
<
D
f
(g)
. Let
= sup(g) : g F, now
<
(as
= cf(
) > = [F[). So g F g ,=
D
f
, contradiction. So
really T
2
D
(
) htcf
D
(
) as required.
If F exemplies the value of T
1
D
(
), it also exemplies T
2
D
(
) [F[ hence
T
2
D
(
) T
1
D
(
).
Lastly if F exemplies the value of T
1
D
(f) it also exemplies T
0
D
(
) [F[, so
T
1
D
(
) T
0
D
(
).
2) Let be [P()/D[ or at least is such that the Boolean algebra P()/D
satises the
+
-c.c. Assume that the desired conclusion fails so T
2
D
(
) < T
0
D
(
), so
there is F
0
, such that [f
1
,= f
2
F
0
f
1
,=
D
f
2
], and [F
0
[ > T
2
D
(
) + (by
the denition of T
0
D
(
).
For every f F
0
there is g
f
F
2
such that f ,=
D
g
f
(by the choice of F
2
).
As [F
0
[ > T
2
D
(
) + for some g F
2
, F
=: f F
0
: g
f
= g has cardinality
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 49
> T
2
D
(f) +. Now for each f F
let A
f
= i < : f(i) = g(i) clearly A
f
D
+
.
Now f A
f
/D is a function from F
,= f
from F
(hence
form F
0
) we have A
f
/D = A
f
/D; but so
i < : f
(i) = f
(i) i < : f
(i) = g(i) = A
f
/D
hence is ,= mod D, so f
,=
D
f
F
0
by the choice of F
0
) we have:
f
1
,= f
2
F
A
f
1
A
f
2
= mod D
so A
f
: f F
). As
+
T
0
D
(
) we can nd
f
,=
D
f
is
nite. Now for every function h : we dene g
h
, a function with domain :
g
h
(i) = (, f
h()
(i)) : w
i
.
So [g
h
(i) : h
[ (
i
)
|w
i
|
=
i
, and if h
1
,= h
2
are from
then for some < ,
h
1
() ,= h
2
() so B
h
1
,h
2
= i : f
h
1
()
(i) ,= f
h
2
()
(i) D that is B
h
1
,h
2
A
D so
1
if i B
h
1
,h
2
A
then w
i
, so g
h
1
(i) ,= g
h
2
(i)
2
B
h
1
,h
2
A
D.
So g
h
: h
) exemplies T
0
D
(
< T
0
D
(
)].
So if T
0
D
(
<
with
< T
0
D
(
=
T
0
D
(
)
<reg(D)
=
<
, and let f
: <
,=
D
f
D
] and repeat the previous proof with f
h()
replacing f
h()
.
7
+
means the left side is a supremum, right bigger than the left or equal but the supremum
is obtained
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
50 SAHARON SHELAH
4) For the rst inequality assume it fails so =: T
2
D
(f) < htcf
D
(
i<
f(i)) hence for
some g
i<f(i)
(f(i)+1), tcf(
i<
g(i), <
D
) is with = cf() > . Let f
: < )
exemplies this. Let F be as in the denition of T
2
D
(f), now for each h F, there
is (h) < such that
i < : if h(i) < g(i) then h(i) < f
(g)
(i) D.
Let
= sup(h) + 1 : h F, now f
i<
f(i) and h F h ,=
D
f
i<
f(i)) exemplies T
2
D
(f) ; we replace clause (g) in the proof by
(g)
< f
2
+1
< f
1
mod D
the construction is for < reg(D) and if we nd satisfy f
1
,=
D
g
we are done.
5) Straightforward.
6) Note that all those cardinals are 2
and 2
wsat(D)
i<
f(i)]
<reg(D)
[T
0
D
(f)]
<reg(D)
= T
0
D
(f).
7) See proof of part (3). Moreover, if =
<
, < T
0
D
(
),
< T
0
D
(
) as
exemplied by f
: < , f
: <
respectively. Let g
be: if
<
< <
then g
(i) = (f
(i), f
(i)). So g
: < show: if T
0
D
(
) is singular then
the supremum is obtained.
3.17
3.18 Claim. Assume D is a lter on , f
Ord,
0
= and 2
< , T
D
(f),
(see Denition 3.16(d) and Theorem 3.17(2)). If < T
D
(f) then for some sequence
f of regulars,
+
= tcf(
/D), or at least
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 51
() there are
i,n
: n < n
i
) : i < ),
i,n
= cf(
i,n
) < f(i) and a lter
D
on
_
i<
i n
i
such that:
+
= tcf(
(i,n)
i,n
/D
) and D = A :
_
iA
i n
i
D
.
Also the inverse is true.
Remark 3.15A. 1) It is not clear whether the rst possibility may fail. We have
explained earlier the doubtful role of
0
= .
2) We can replace
+
by any regular such that
<
[[
0
< and then we use
3.17(4) to get
+
T
D
(f).
(3) The assumption 2
/D 2
, so by 3.17(4)
T
D
(f) [htcf
D
(
i<
f(i)]
<reg(D)
.
If < htcf
D
(
i<
f(i)) we are done (by 3.15(1)), so assume htcf
D
(
i<
f(i)) ,
but we have assumed < T
D
(f) so we can conclude
<reg(D)
+
. Let
be minimal such that
< reg(D)
we know
cf
=
<reg(D)
=
<reg(D)
+
, > 2
,
_
<
[[
<reg(D)
< . By the assumption
=
0
we know >
0
(of course is regular). By [Sh:g, Ch.VIII,1.6](2),IX,3.5
and [Sh 513, 6.12] there is a strictly increasing sequence
: < ) of regular
cardinals with limit such that
+
= tcf(
<
J
bd
).
As clearly htcf
D
(
i<
f(i)), we can nd for each < , a sequence
i
:
i < ) such that
i
= cf(
i
) f(i), and tcf(
i<
i
/D) =
i
> 2
. Let A
[ and
i,n
: n < enumerate
i
: satises i A
3.18
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
52 SAHARON SHELAH
3.19 Conclusion Suppose D is an
1
-complete lter on . If
i
2
) >
0
then for some
i
= cf(
i
)
i
we have
sup
AD
+
tcf
D+A
(
i<
i
) > .
3.20 Conclusion Let D be an
1
-complete lter on . If for i < , B
i
is a Boolean
algebra and
i
< Depth
+
(B
i
) (see below) and
2
<
0
< sup
AD
+
T
D+A
(
)
then
+
< Depth
+
(
i<
B
i
/D).
Proof. Use 3.28 below and 3.19 above.
3.21 Denition. For a partial order P (e.g., a Boolean algebra) let Depth
+
(P) =
Min:we cannot nd a
<
P
a
.
3.22 Discussion 1) We conjecture that in 3.19 (and 3.20) the assumption D is
1
-
complete can be omitted.
2) Note that our results are for =
0
only; to remove this we need to improve
the theorem on pp = cov (i.e., to prove cf() =
0
< pp() = cov(, ,
1
, 2)
(or suppp() : cf() =
0
< < = cf(S
0
(), ) (see [Sh:g], [Sh 430, 1]),
which seems to me a very serious open problem (see [Sh:g, Analytic guide,14]).
3) In 3.20, if we can nd f
i<
i
for < : [ < < f
mod D]
and f
=
D
f
+1
then < Depth
+
(
i<
B
i
/D). But this does not help for
regular > 2
.
4) We can approach 3.18 dierently, by 3.23 - 3.26 below.
3.23 Claim. If 2
2
< T
D
(
), (or at least 2
|D|+
< T
D
(
)) and
<
= ,
then for some -complete lter E D we have T
E
(
) > .
Proof. Without loss of generality is regular (as
<
= & cf() <
<
+
=
). Let f
: <
+
,=
D
f
]. We choose by
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 53
induction on ,
<
+
as follows:
(i) ,= f
(i) : <
_
(note: each generator of E
,
is in D but not necessarly E
,
D!).
Let
, clearly <
<
. Now if
<
+
, then
clearly
=
_
<
<
+
and for every (
,
+
), E
,
D, so for every
such there are A
D
+
and a
]
<
such that A
i < : f
(i) =
f
]
<
we have
: (
,
+
), A
= A, a
= a : f
i<
f
(i) : a
,
hence has cardinality
< . Also [[
]
<
[
<
<
+
, |D
+
| 2
<
=
+
, but the number of possible Es is 2
2
= E[ =
+
. Necessarily E D and E is -complete, and
f
: <
+
, and E
= E exemplies T
E
(
) > , so E is as required.
3.23
3.24 Fact 1) In 3.23 we can replace
+
by
if 2
2
< cf(
T
0
D
(
) and
<
[[
<
<
.
2) We can, in 3.23, [and 3.24(1)] replace T
D
(
) > by
/D has an increasing
sequence of lengths > [ ], we can deduce this also otherwise.
3.25 Claim. 1) If 2
< [
/D[,
i<
[i[
0
< , and D is regular then < Depth
+
(
i<
i
/D).
2) Similarly for D just a lter.
Proof. Without loss of generality = lim
D
= sup(
), so [
/D[ =
(by
[\CK ]). If we are done; otherwise let = Min :
, so
cf()
=
hence
0
< hence cf() >
0
, also by
s
minimality
i<
[i[
cf()
[i[
< cf()
/J
bd
cf()
has true connality . Let
= sup
: < +2
,
let i : cf() be i(i) = sup + 1 :
i
. If there is a function h
i<
i(i)
such that
j<cf()
i < : h(i) < j = mod D then
i<
h(i)
/D has true conality
as required; if not (D, i) is weakly normal (i.e. there is no such h - see [Sh 420]).
But for D regular, D is cf()-regular, some A
3.25
3.26 Discussion 1) In 3.23 (or 3.24) we can apply [Sh 410, 6] so = tcf(
_
i<
a
i
/D
,
where D = A :
_
iA
a
i
D
and each a
i
is nite.
In 3.18 we have gotten this also for (2
, 2
2
).
3.27 Claim. If D is a lter on , B
i
is the interval Boolean algebra on the ordinal
i
, and [
i<
i
/D[ > 2
i<
B
i
/D) i for
some
i
i
(for i < ) and A D
+
, the true conality of
i<
i
/(D+A)) is well
dened and equal to .
Proof. The (i.e., only if direction) is clear. For the direction assume
is regular < Depth
+
(
i<
B
i
/D) so there are f
i<
B
i
such that
i<
B
i
/D
f
/D < f
/D for < .
Without loss of generality > 2
. Let f
(i) =
_
<n(,i)
[j
,i,2
, j
,i.2+1
) where
j
,i,
< j
,i,+1
<
i
for < 2n(, i). As = cf() > 2
=: (i, ) : i <
, < 2n
and
i,
: i < , < 2n
i
) such that (i, ) A
i,
is a limit ordinal
and
() for every f
(i,)A
i,
and < there is (, ) such that
(i, ) A
A j
,i,
=
i,
(i, ) A f(i, ) < j
,i,
<
i,
(i, ) A cf(
i,
) > 2
.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 55
Let (i) = max < 2n(i) : (i, ) A and let B = i : (i) well dened. Clearly
B D
+
(otherwise we can nd < < such that f
/D = f
/D, contradiction).
For (i, ) A dene
i,
by
i,
= sup
j,m
+1 : (j, m) A
and
j,m
<
i,
. Now
i,
<
i,
as cf(
i,
) > 2
. Let
Y = < : if(i, ) A
A then j
,i,
=
i,
and if (i, ) A then
i,
< j
,,i
<
,i
.
Let B
1
= i B : (i) is odd. Clearly B
1
B and B B
1
= mod D (otherwise
as in ()
1
, ()
2
below get contradiction) hence B
1
D
+
. Now
()
1
for < from Y we have
j
,i,(i)
: i B
1
) j
,i,(i)
: i B
1
) mod (D B
1
)
[Why? as f
i<
B
i
/D]
()
2
for every Y for some , < Y we have
j
,i,(i)
: i B
1
) < j
,i,(i)
: i B
1
) mod (D B
1
)
[Why? by () above.]
Together for some unbounded Z Y ,
j
,,(i)
: i B
1
)/(D B
1
) : Z
_
is
<
DB
1
-increasing, so it has a <
(DB
1
)
-eub (as > 2
), say j
i
: i B
1
) hence
iB
1
j
i
/(D B
1
) has true conality , and clearly j
i,(i)
i
, so we have
nished.
3.27
3.28 Claim. If D is a lter on , B
i
a Boolean algebra,
i
< Depth
+
(B
i
) then
(a) Depth(
i<
B
i
/D) sup
AD
+
tcf(
i<
i
/(D + A)) (i.e., on the cases tcf is well
dened)
(b) Depth
+
(
i<
B
i
/D) is Depth
+
(P()/D) and is at least
sup[tcf(
i<
i
/(D + A))]
+
:
i
< Depth
+
(B
i
), A D
+
.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
56 SAHARON SHELAH
Proof. Check.
3.29 Claim. Let D be a lter on ,
i
: i < ) a sequence of cardinals and
2
0
< ) we also
have () () where
() if B
i
is a Boolean algebra,
i
< Depth
+
(B
i
) then < Depth
+
(
i<
B
i
/D)
() there are
i
= cf(
i
)
i
for i < and A D
+
such that = tcf(
i
/(D+
A))
() there are
i,n
: n < n
i
) : i < ),
i,n
= cf(
i,n
) <
i
and a lter D
on
_
i<
i n
i
such that:
= tcf(
(i,n)
i,n
/D
() for some A D
+
, T
D+A
(
i
: i < )).
Remark. So the question whether () () assuming ( < )(
0
< ) is
equivalent to () () which is a pure pcf problem.
Proof. Note () () is easy (as in 3.18, i.e., as in the proof of 3.7, only easier).
Now () () is trivial and () () by 3.28. Next () () holds as we can
use () for B
i
=: the interval Boolean algebra of the order
i
and use 3.27. Lastly
assume ( < )(
0
< ), now () () by 3.18.
3.29
Discussion: We would like to have (letting B
i
denote Boolean algebra)
Depth
(+)
(
i<
B
i
/D)
i<
Depth
(+)
(B
i
)/D
if D is just lter we should use T
D
and so by the problem of attainment (serious
by Magidor Shelah [MgSh 433]), we ask
for D an ultralter on , does
i
< Depth
+
(B
i
) for i < implies
i<
i
/D < Depth
+
(
i<
B
i
/D)
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 57
at least when
i
> 2
i
> 2
for simplicity,
T
D
(
i
: i < )) < Depth
+
(
i<
B
i
/D).
As explained in 3.29 this is a pcf problem.
However changing the invariant (closing under homomorphisms, see [M]) we get
a nice result; this will be presented in [Sh 580].
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
58 SAHARON SHELAH
4 Remarks on the conditions for the pcf analysis
We consider a generalization whose interest is not so clear.
4.1 Claim. Suppose
=
i
: i < ) is a sequence of regular cardinals, and is a
cardinal and I
) wsat(I
()
+
similarly but
(b)
+
for i < we have otp(H(i)) <
i
.
1) In 1.9 we can replace the assumption () by ()
H
above.
2) Also in 1.11, 1.12, 1.13, 1.14, 1.15, ? we can replace 1.9() by ()
H
.
> scite1.11 undened
3) Suppose in Denition 2.3(2) we say
f obeys a for H (instead of for
A
) if
(i) for a
) < we have
otp(a
), otp(a
) H(i) f
(i) f
(i)
and in 2.3(2A), f
(i) = supf
(i) : a
and otp(a
), otp(a
) H(i).
Then we can replace 1.9() by ()
H
in 2.7, 2.8, ?; and replace 1.9() by ()
+
H
in
> scite2,6 undened
2.10 (with the natural changes).
Proof. 1) Like the proof of 1.9, but dening the g
s by induction on we change
requirement (ii) to
(ii)
(i) < g
(i).
We can not succeded as
(B
()
B
+1
()
) i < : , + 1 H(i) : < )
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 59
is a sequence of pairwise disjoint member of (I
)
+
.
In the induction, for limit let g
(i) < g
.
Also the proof of the rest is the same.
3) Left to the reader.
4.1
We want to see how much weakening () of 1.9 to lim inf
I
(
) wsat(I
)
suces. If singular or lim inf
I
(
) > or just (
, <
I
) is
+
-directed then case
() of 1.9 applies. This explains () of 4.2 below.
4.2 Claim. Suppose
=
i
: i < ),
i
= cf(
i
), I
an ideal on , and
() lim inf
I
(
) = wsat(I
), regular.
Then we can dene a sequence
J = J
, J
+1
=: A : A and: A J
or we can nd h : A such
that
i
> h(i) and < i : h(i) < J
, the pair (
, J
+ ( A)) (equivalently
A, J
then (
,
<()
J
for
+
by clause (b) for = 0, successor
and as
_
<
J
for limit. Clause (c) holds by claim 4.4 below. It should be clear
that J
+
+1
= J
iA
i
, choose by induction on < , () <
increasing with such that i < : h(i) ((), ( + 1)) J
+
()
. If we succeed
we contradict wsat(I
(i) =
_
if <
i
0 if
i
exemplies tcf(
/J
()
) = 0.
4.2
Now:
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
60 SAHARON SHELAH
4.3 Conclusion. Under the assumptions of 4.2, I
, <
I
) = sup pcf
I
(
) (see 2.19).
4.4 Claim. Under the assumption of 4.2, if J is an ideal on extending I
the
following conditions are equivalent
(a) for some h
, <
J+(\A)
) is
+
-directed.
Proof. (a) (b)
Let f
by
f
(i) = supf
(i) <
i
as h(i) <
i
= cf(
i
) and f
A <
J
f
A as i A : h(i) <
J.
(b) (a):
Let f
(i) =
_
if <
i
0 if
i
As lim inf
I
, clearly < f
<
I
f
and of course f
. By our
assumption (b) there is h
OS 61
regular cardinals and cardinal invariants of Boolean Algebra,
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
62 SAHARON SHELAH
REFERENCES.
[JuSh 231] Istvan Juhasz and Saharon Shelah. How large can a hereditarily sepa-
rable or hereditarily Lindelof space be? Israel Journal of Mathematics,
53:355364, 1986.
[Kn] Akihiro Kanamori. Weakly normal lters and irregular ultra-lters.
Trans. of A.M.S., 220:393396, 1976.
[Kt] Jussi Ketonen. Some combinatorial properties of ultra-lters. Fund.
Math., CVII:225235, 1980.
[Ko] Peter Komjath. On second-category sets. Proc. Amer. Math. Soc.,
107:653654, 1989.
[MgSh 433] Menachem Magidor and Saharon Shelah. Length of Boolean alge-
bras and ultraproducts. Mathematica Japonica, 48(2):301307, 1998.
math.LO/9805145.
[M] J. Donald Monk. Cardinal functions of Boolean algebras. circulated
notes.
[Sh 7] Saharon Shelah. On the cardinality of ultraproduct of nite sets. Jour-
nal of Symbolic Logic, 35:8384, 1970.
[Sh:a] Saharon Shelah. Classication theory and the number of nonisomor-
phic models, volume 92 of Studies in Logic and the Foundations of
Mathematics. North-Holland Publishing Co., Amsterdam-New York,
xvi+544 pp, $62.25, 1978.
[Sh 233] Saharon Shelah. Remarks on the numbers of ideals of Boolean algebra
and open sets of a topology. In Around classication theory of models,
volume 1182 of Lecture Notes in Mathematics, pages 151187. Springer,
Berlin, 1986.
[Sh:c] Saharon Shelah. Classication theory and the number of nonisomor-
phic models, volume 92 of Studies in Logic and the Foundations of
Mathematics. North-Holland Publishing Co., Amsterdam, xxxiv+705
pp, 1990.
[Sh 351] Saharon Shelah. Reecting stationary sets and successors of singular
cardinals. Archive for Mathematical Logic, 31:2553, 1991.
[Sh 400a] Saharon Shelah. Cardinal arithmetic for skeptics. Ameri-
can Mathematical Society. Bulletin. New Series, 26:197210, 1992.
math.LO/9201251.
(
5
0
6
)
r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
1
1
THE PCF THEOREM REVISITED SH506 DEDICATED TO PAUL ERD
OS 63
[Sh 420] Saharon Shelah. Advances in Cardinal Arithmetic. In Finite and In-
nite Combinatorics in Sets and Logic, pages 355383. Kluwer Academic
Publishers, 1993. N.W. Sauer et al (eds.). 0708.1979.
[Sh 410] Saharon Shelah. More on Cardinal Arithmetic. Archive for Mathemat-
ical Logic, 32:399428, 1993. math.LO/0406550.
[Sh 345a] Saharon Shelah. Basic: Conalities of small reduced products. In Car-
dinal Arithmetic, volume 29 of Oxford Logic Guides, chapter I. Oxford
University Press, 1994.
[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 580] Saharon Shelah. Strong covering without squares. Fundamenta Math-
ematicae, 166:87107, 2000. math.LO/9604243.
[Sh 513] Saharon Shelah. PCF and innite free subsets in an algebra. Archive
for Mathematical Logic, 41:321359, 2002. math.LO/9807177.