(

9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE
SH938
SAHARON SHELAH
Abstract. We deal with relatives of GCH which are provable. In particular
we deal with rank version of the revised GCH. Our motivation was to nd
such results when only weak versions of the axiom of choice are assumed but
some of the results gives us additional information even in ZFC. We also start
to deal with pcf for pseudo-conality (in ZFC with little choice).
Annotated Content
0 Introduction, pg.2
[We present introductory remarks mainly to 3,4.]
1 Preliminaries, pg.3
[We present some basic denitions and claims, mostly used later.]
2 Commuting ranks, pg.8
[If we have ideals D
1
, D
2
on sets Y
1
, Y
2
and a Y
1
Y
2
-rectangle of ordinals,
we can compute rank in two ways: one is rst apply rk
D1
on each row and
then rk
D2
() on the resulting column. In the other we rst apply rk
D2
()
on each column and then rk
D1
() on the resulting row. We give sucient
conditions for an inequality. We use (ZFC + DC and) weak forms of choice
like AC
Y

or AC
P(Y

)
.]
3 Rank systems and a Relative of GCH, pg.13
[We give a framework to prove a relative of the main theorem of [Sh:460]
dealing with ranks. We do it with weak form of choice (DC + AC
<
), a
limit cardinal, this give new information also in ZFC.]
4 Finding systems, pg.21
[The main result in 3 deals with an abstract setting. Here we nd an
example, a singular limit of measurables. Note that even under ZFC this
gives information on ranks.]
5 Pseudo true conality, pseudo tcf, pg.23
[We look again at the pcf( ), but only for
1
-complete lters using pseudo-
conality and the conalities not too small. Under such restrictions we get
parallel to pcf basic results.]
Date: February 24, 2010.
The author would like to thank the Israel Science Foundation for partial support of this re-
search. (Grant No. 242/03). Publ. 938. The author thanks Alice Leonhardt for the beautiful
typing. First Typed - 08/July/28.
1
(
9
3
8
)


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







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


2 SAHARON SHELAH
0. Introduction
In [Sh:460] and [Sh:513], [Sh:829] we prove in ZFC = ZF + AC relatives of G.C.H.
Here mainly we are interested in relatives assuming only weak forms of choice, but
some results add information even working in ZFC, in particular a generalization
of [Sh:460] for ranks. Always we can assume ZF + DC.
Our original motivation was
{y0.3}
Conjecture 0.1. Assume ZF + DC and a limit cardinal such that AC
<
and
is strong limit. For every ordinal , for some < , for any < and -complete
lter D on we have rk
D
() = .
Here we get an approximation to it, i.e. for a limit of measurables restricting
ourselves to ultralters; this is conclusion 4.4 deduced by applying Theorem 3.10
to Claim 4.3. Can we do it with =

?
Also we would like to weaken AC
<
; this is interesting per se and as then we will
be able to combine [Sh:835] + [Sh:513] - see below. We intend to try in [Sh:F955];
starting with

J = J
n
: n < ) such that IND(

J) or something similar.
It may be illuminating to compare the present result with (see [Sh:g, V]).
{y0.7}
Claim 0.2. If >
0
, 2
2

then the following conditions are equivalent:

()
1
for every -complete lter D on , we have rk
D
(
+
) =
+
()
2
<
+
rk
D
() <
+
for every -complete lter D on
()
3
there is no T

of cardinality
+
and -complete lter D in such
that f
1
,= f
2
T f
1
,=
D
f
2
.
Also we can in 0.2 replace
+
by a cardinal of conality > 2
2

. So the result
in [Sh:460] implies a weak version of the conjecture above, say on [rk
D
()[, but
the present one gives more precise information. On the other hand, the present
conjecture is not proved for =

, also it seems less accommodating to the

possible results with

instead of

in [Sh:513] below 2
2

.
{y0.9}
Question 0.3. In [Sh:908] can we prove that the rank is small?
Discussion 0.4. In 0.5 below we present examples showing some limitations.
Below part (1) of the example shows that Claim 2.3 cannot be improved too
much and part (2) shows that Conclusion 4.4 cannot be improved too much. In
fact, in conjecture 0.1 if we demand only is a limit cardinal then it consistently
fails. This implies that we cannot improve too much other results in 3,4.
We may wonder how to compare the result in [Sh:460] and Conjecture 0.1 even
in ZFC.
{y0.4}
Example 0.5. 1) If D

= dual([

]
<

) for = 1, 2 (so if

is regular then
D

= dual(J
bd

)) and
2
<
1
then D
2
does not 2-commute with D
1
, i.e.
2
D1,D2
from Denition 2.1 fail.
2) Consistently with ZFC, for every n, rk
J
bd
n
(

) >

.
Proof. 1) Let A =
1
and let f
2

2
Ord be constantly 1 hence by Denition 1.10
and Claim 1.11(3) the ideal J
2
= J[f
2
, D
2
] is [
2
]
<2
. Choose a function h :
1

2
and ( <
2
)(
1
<
1
)(h() = ) and let B

: A) be such that we have

B

:=
2
h().
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 3
So if A

D
1
, B

J
+
2
then for some

<
1
we have A

and
B

2
, [B

[ =
2
and choose t B

and then choose s A

such that h(s) = t+1,

such s exists by the choice of h so (s, t) A

but (s, t) / sB
s
, contradiction.
2) Assume that the sequence 2
n
: n < ) is increasing with supremum >

and in cf(
(n)
(
n
), <
J
bd
n
) there is an increasing sequence of length
+1
for each
n [1, ) hence it follows that rk
J
bd
n
(

) > rk
J
bd
n
(
n
)
n
for n [1, ).
0.5
We may hope to prove interesting things in ZF + DC by division to cases: if
[Sh:835] apply ne, if not then we have a strict p. But we need AC
<
to prove
even clause (f) in 3.1, see [Sh:F955]. We may consider that even in ZFC, probably
[Sh:908] indicate that we can use weaker assumptions.
Let us say something on our program on set theory with little choice of which this
work is a part. We always know that the axiom of choice is true. In addition we
had thought that there is no interesting general combinatorial set theory without
AC (though equivalence of version of choice, inner model theory and some other
exist). Concerning the second point, since [Sh:497] our opinion changed and have
thought that there is an interesting such set theory, with bounded choice related
to pcf. More specically [Sh:497] seems to prove that such theory is not empty.
Then [Sh:835] suggest to look at axioms of choice orthogonal to V = L[R],
e.g. demand then

can be well ordered (and weaker relatives). The results say
that the universe is somewhat similar to universes gotten by Easton like forcing,
blowing up 2

for every regular without well ordering the new T(). Continuing
this Larson-Shelah [LrSh:925] generalize classical theorem on splitting a stationary
subset of a regular consisting of ordinals of conality .
In [Sh:F955] we shall continue this work. In particular, we continue 5 to get
a parallel of the pcf theorem and more. Recall that in [Sh:513] in ZFC we get
connections between the existence of independent sets and a strong form of [Sh:460].
We prove related theorems on rank.
We thank the referee for many corrections and remarks.
1. Preliminaries
{z0.0}
Context 1.1. 1) We work in ZF in all this paper.
2) We try to say when we use DC but assuming it always makes no great harm.
3) We shall certainly mention the use of any additional form of choice, mainly AC
A
.
4) In 1.2 - 1.11 we quote denitions and claims to be used, see [Sh:835]. The rest
of 1 is used only in 5.
{z0.11}
Denition 1.2. 1) A lter D on Y is ( B)-complete when: if A
t
: t B)
B
D
then A := A
t
: t B D. We can instead say [B[
+
-complete even if [B[
+
is
not well dened.
1A) A lter D on Y is pseudo ( B)-complete when if A
t
: t B)
B
D then
A
t
: t B is not empty (so adopt the conventions of part (1)).
2) For an ideal J on a set Y let dual(J) = Y X : X J, the dual ideal and
Dom(J) = Y , abusing notation we assume J determines Y .
3) For a lter D on a set Y let dual(D) = Y X : X D, Dom(D) = Y . We may
use properties dened for lter D for the dual ideal (and vice versa).
4) For a lter D on Y let D
+
= A Y : Y A / D and for an ideal J on Y let
J
+
= (dual(J))
+
.
(
9
3
8
)


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







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


4 SAHARON SHELAH
Remark 1.3. It may be interesting to try to assume that relevant lters are just
pseudo ( B)-complete instead of ( B)-complete. Now 1.14 clarify the connection
to some extent, but presently we do not pursue this direction.
{z0.13}
Denition 1.4. C is the class of sets A such that AC
A
, the axiom of choice for A
non-empty sets, holds.
{z0.15}
Denition 1.5. 1) (A) = Min: there is no function from A onto .
2) (A) = Min: there is no one-to-one function from into A so (A) (A).
{z0.21}
Denition 1.6. 1) For D a lter on Y and f, g
Y
Ord let f <
D
g or f < g mod
D means that s Y : f(s) < g(s) D; similarly for , =, ,=.
2) For D a lter on Y and f
Y
Ord and Ord we dene when
rk
D
(f) = by induction on :
For < , rk
D
(f) = i < rk
D
(f) ,= and for every g
Y
Ord
satisfying g <
D
f there is < such that rk
D
(g) = .
3) We can replace D by the dual ideal.
{z0.24}
Observation 1.7. 1) Let D be a pseudo
1
-complete lter on Y . If f, g
Y
Ord
and f
D
g then rk
D
(f) rk
D
(g) and so if f =
D
g then rk
D
(f) = rk
D
(g).
2) If D

is a pseudo
1
-complete lter on Y for = 1, 2 then D
1
D
2
f
Y
Ord
rk
D1
(f) rk
D2
(f).
Proof. Easy.
{z0.23}
Claim 1.8. Assume D is a lter on Y such that D is
1
-complete or just pseudo

1
-complete (see Denition 1.2(1A)).
1) [DC] For f
Y
Ord, in 1.6, rk
D
(f) is always an ordinal, i.e. < .
2) [DC] If rk
D
(f) then for some g

tY
(f(t) + 1) we have = rk
D
(g). If
< rk
D
(f) we can add g <
D
f and we can demand (y Y )(g(y) < f(g) g(y) =
0 = f(y)).
2A) If rk
D
(f) < then part (2) holds for f (without assuming DC).
3) If f, g
Y
Ord and f <
D
g and rk
D
(f) < then rk
D
(f) < rk
D
(g).
4) For f
Y
Ord we have rk
D
(f) > 0 i t Y : f(t) > 0 D.
5) If f, g
Y
Ord and f = g + 1 then rk
D
(f) = rk
D
(g) + 1.
Proof. Straight, e.g.
2A) We prove this by induction on = rk
D
(f). If there is nothing to prove.
If = + 1 by the denition, there is g <
D
f such that rk
D
(g) , now by part
(3) we have rk
D
(g) < rk
D
(f) which means rk
D
(g) < +1, so together rk
D
(g) =
and let g

Y
Ord be dened by g

(s) is g(s) if g(s) < f(s) and is 0 if g(s) f(s)

so g

<
D
f and g

D
g
D
g

hence rk
D
(g

) = rk
D
(g) = is as required.
Lastly, if > +1 by the denition there is f

<
D
f such that rk
D
(f

) +1
and by 1.7(1) without loss of generality t Y f

(t) f(t) and by part (3)

rk
D
(f

) < rk
D
(f) so we can apply the induction hypothesis to f

.
1.8
{z0.25}
Claim 1.9. 1) [AC
0
] If D is an
1
-complete lter on Y and f
Y
Ord and
Y = Y
n
: n < then rk
D
(f) = minrk
D+Yn
(f) : n < and Y
n
D
+
.
2) [AC
W
] If D is a [J[
+
-complete lter on Y, J innite and f
Y
Ord and
Y
t
: t J D then rk
D
(f) = minrk
D+Yt
(f) : t J and Y
t
D
+
.
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 5
Proof. Like [Sh:71].
1) By part (2).
2) Note that by AC
W
necessarily t : Y
t
D
+
is non-empty. The inequality
is obvious (i.e. by 1.7(2)). We prove by induction on the ordinal that (v
J)[Y
v
D
+
rk
D+Yv
(f) ] rk
D
(f) .
For = 0 and is limit this is trivial.
For = + 1, we assume (v J)[Y
v
D
+
rk
D+Yv
(f) > ] so by
Denition 1.6 it follows that [v J Y
v
D
+
(g)(g
Y
Ord g <
D+Yv
f rk
D+Yv
(g) ] hence, if v J Y
v
D
+
then t Y : f(t) = 0 = mod
(D + Y
v
), i.e. v : f(v) = 0 Y
v
= mod D. As this holds for every v J and
D is [J[
+
-complete clearly we have t Y : f(t) = 0 = mod D. We can by
1.7(1) replace f by f

Y
Ord when v Y : f(v) = f

(v) D so without loss of

generality t Y f(t) > 0.
But J C, hence by 1.8(2A) there is a sequence g
v
: v J

) such that
J

:= v J : Y
v
D
+
and g
v

Y
Ord, g
v
<
D+Yv
f, rk
D+Yv
(g
v
) and
t Y g
v
(t) < f(t) so g
v
< f.
As D is [J[
+
-complete necessarily Y

:= Y
v
: v JJ

= mod D, but
Y
v
: v J D hence Y

= Y
v
: v J

belongs to D. Dene g
Y
Ord by
g(s) = ming
u
(s) : u J

satises s Y
u
if s Y

and 0 if s Y Y

.
Hence (Y
v
: v J

) D and g
Y
Ord and g <
D
f (and even g < f) so by
the induction hypothesis
it suces to prove v J

rk
D+Yv
(g) .
Fix v J

, and for each u J

let Y
v,u
:= t Y
u
Y
v
: g(t) = g
u
(t) so by the
choice of g(t) we have

1
if v }

, t Y
y
then for some u J

we have t Y
y,x
Y
y
and
g(t) = g
u
(t).
Hence

2
Y
v,u
: u J

) exists and Y
v,u
: u J

= Y
v
(D + Y
v
).
Now

3
if u J

Y
v,u
(D + Y
v
)
+
then rk
D+Yv,u
(g) .
[Why? By the choice of Y
v,u
we have g = g
u
mod(D + Y
v,u
) hence rk
D+Yv,u
(g) =
rk
D+Yv,u
(g
u
), also Y
v,u
Y
u
hence D + Y
v,u
D + Y
u
which by 1.7(2) implies
rk
D+Yv,u
(g
u
) rk
D+Yu
(g
u
) which is . Together we are done.]
By
2
+
3
and the induction hypothesis it follows that v J

rk
D+yv
(g)
so by we are done.
1.9
{z0.27}
Denition 1.10. For Y, D, f in 1.6 let J[f, D] =: Z Y : Y Z D or (Y Z)
D
+
rk
D+Z
(f) > rk
D
(f).
{z0.29}
Claim 1.11. [DC+AC
Y
] Assume D is an
1
-complete [}[
+
-complete lter on Y .
1) If f
Y
Ord then J[f, D] is an
1
-complete and [}[
+
-complete ideal on Y .
2) If f
1
, f
2

Y
Ord and J = J[f
1
, D] = J[f
2
, D] then rk
D
(f
1
) < rk
D
(f
2
) f
1
< f
2
mod J and rk
D
(f
1
) = rk
D
(f
2
) f
1
= f
2
mod J.
3) If f
Y
Ord is e.g. constantly 1 then J[f, D] = dual(D).
4) If f
Y
Ord and A (J[f, D])
+
then (A D
+
and) rk
D+A
(f) = rk
D
(f).
(
9
3
8
)


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







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


6 SAHARON SHELAH
Proof. 1) By 1.9.
2) As J is an ideal on Y (by part (1)) this should be clear by the denitions;
that is, let A
0
:= t Y : f
1
(t) < f
2
(t), A
1
:= t Y : f
1
(t) = f
2
(t) and
A
2
:= t Y : f
1
(t) > f
2
(t). Now A
0
, A
1
, A
0
) is a partition of Y .
First, assume A
0
J
+
, then by the denition of J[f
1
, D] we have (rk
D
(f
1
) <
rk
D+A0
(f
1
)); i.e. rk
D+A0
(f
1
) rk
D
(f
1
) and so by 1.7(2) we have rk
D
(f
1
) =
rk
D+A0
(f
1
). Now as A
0
J
+
, by the choice of A
0
, f
1
<
D+A0
f
2
hence rk
D
(f
1
) =
rk
D+A0
(f
1
) < rk
D+A0
(f
2
) = rk
D
(f
2
).
[Why? By the previous sentence, by 1.8(3), by the previous sentence respectively.]
Second, similarly if A
2
J
+
then f
2
< f
1
mod (D+A
2
) and rk
D
(f
1
) > rk
D
(f
2
).
Lastly, if A
1
J
+
then by 1.7(1) f
1
= f
2
mod (D + A
1
) hence rk
D+A1
(f
1
) =
rk
D+A1
(f
2
) and rk
D
(f
1
) = rk
D+A1
(f
2
) = rk
D+A1
(f
2
) = rk
D
(f
2
).
By the last three paragraphs at most one of A
0
, A
1
, A
2
belongs to J
+
and as
A
0
A
1
A
2
= Y at least one of A
0
, A
1
, A
2
belongs to J
+
, so easily we are done.
3) Obvious.
4) Proved inside the proof of part (2).
1.11
{z0.32}
Denition 1.12. 1) Let FIL
cc
S
(Y ) or FIL
pcc
S
(Y ) be the set of D such that:
D is a lter on the set Y which is [S[
+
-complete and is
1
-complete or is psuedo
[S[
+
-complete and psuedo
1
-complete.
2) Let FIL
cc
(Y ) or FIL
pcc
(Y ) be FIL
cc

or FIL
pcc

.
3) Omitting Y means for some Y and then we let Y = Dom(D). Without enough
choice, the minimal ( S)-complete lter extending a lter D is gotten in stages.
{z0.51}
Denition 1.13. 1) For a lter D on Y and set S we dene comp
S,
(D) by
induction on Ord .
= 0: comp
S,
(D) = D
= limit: comp
S,
(D) = comp
S,
(D) : <
= + 1: comp
S,
(D) = A Y : A belongs to comp
S,
(D) or include the
intersection of some S-sequence of members of comp
S,
(D), i.e. A
s
: s S,
where A
s
: s S) is a sequence of members of comp
S,
(D).
2) Similarly for a family S of sets replacing S by some member of S, e.g. we
dene com
S,
(D) by induction on using ( S)-sequences, i.e. S-sequence for
some S S.
3) If = we may omit it. We say that D is a pseudo ( S, )-complete when
/ comp
S,
(D).
{z0.55}
Observation 1.14. 1) If D is a lter on Y and S is a set, then:
(a) comp
S,
(D) : Ord ) is an -increasing sequence of lters of Y
(starting with D)
(b) if comp
S,+1
(D) = comp
S,
(D) then for every we have comp
S,
(D) =
comp
S,
(D)
(c) there is an ordinal =
S
(D) < (T(Y )) such that comp
S,
(D) = comp
S,+1
(D)
and comp
S,
(D) : ) is strictly -increasing.
2) Assume AC
S
. Then for any lter D on Y we have
S
(D) when := min :
a cardinal such that cf() (S).
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 7
3) Assume DC + AC
S
+ [S S[ = [S[. Then for any lter D on Y we have

S
(D) 1 and comp
S,1
(D) is an ( S)-complete lter or is T(Y ); the latter holds
i D is not pseudo ( S)-complete.
4) Similarly to part (2) for S but AC
S
is replaced by S S AC
S
and
= min : regular and S S (S).
Remark 1.15. Note that in part (2) of 1.14, is regular and (
>
S) but the
inverse is not true, if (S) =
0
but holds if (S) >
0
.
Proof. We prove the versions with S, i.e. for (4). Let D

= comp
S,
(D) for
Ord.
1) Clause (a) is by the denition; clause (b) is proved by induction on , for =
this is trivial, for = +1 use the assumption and for > +1 use the denition
and the induction hypothesis. As for clause (c) let

= min Ord ; if
< then D

= D
+1
, so D

) is -increasing continuous by clause

(a), and by clause (b), D

) is constant. Now dene h : T(Y )

by:
A D
+1
D

h(A) = and h(A) = 0 when there is no such . So h is onto

hence

< (T(A)) so

is as required on
S
(D).
2) We prove also the relevant statement in part (4), so S S AC
S
cf()
(S). Let be an ordinal.
Let
T
1
n
= : is a set of sequences of length n,
closed under initial segments such that for every non-maximal
for some S S we have
s) s S.
T
2
,n
= x : (a) x has the form Y

: )
(b) T
1
n
and Y

Y for
(c) Y

= Y
<s>
: s satises s) if
but is not -maximal in
(d)

<

< 1 +
(e) Y

D if is -maximal in
but g() < n
T
2
n
= T
2
,n
: is an ordinal.
Let n(x) = n for the minimal possible n such that x T
2
n
and let x = Y
x

,
x

:

x
).
Let T
3

= T
2
,n
: n < and let <

be the natural order on T

3

: x <

y i
n(x) < n(y),
x
=
y

n(x)
(S : S S) and (Y
x

,
x

) = (Y
y

,
y

) for
x
.
Now
A D

i there is an -branch x
n
: n < ) of (T
3

, <

) such that Y
xn
<>
= A.
[Why? We prove it by induction on the ordinal . For = 0 and limit this is
obvious so assume we have it for and we shall prove it for + 1.
First assume A D
+1
and we shall nd such -branch; if A D

this is
obvious, otherwise there are S S and a sequence A
s
: s S) of members of D

such that A = A
s
: s S. So X
s
:= x : x witness A
s
D

is well dened and

(
9
3
8
)


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







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


8 SAHARON SHELAH
non-empty by the induction hypothesis, clearly the sequence X
s
: s S) exists,
hence we can use AC
S
to choose x
s
: s S) satisfying x
s
X
s
.
Now dene x = x
n
: n < ) as follows:
xn
= ) s) :
xs,n1
and
s S,
xn
<>
=
xs,n
<>
+ 1 : s S and Y
x
<>
= A and Y
xn
<s>
= Y
xs,n1

. Now
check.
Second, assume that there is such -branch x
n
: n < ) of (T
3

, <

) such that
Y
xn
<>
= A. Let S = (0) :
x1
so necessarily S S. For each n < and
s S we dene y
n,s
as follows:
yn,s
s
= : s)
xn+1
and for
n
s
let
yn,s

=
xn+1
<s>
and Y
yn,s

= Y
xn+1
<s>
. Now clearly y
n,s
: n < ) is an -
branch of (T
3

) so by the induction hypothesis A

x1
<s>
D, comp
S,
(D) and
Y
x0
<>
= A = Y
x
<s>
:<>
x1
comp
S,+1
(D). So we are done.]
Now toward a contradiction assume that
S
(D) > , so there is A D
+1
D

hence here is an -branch x

n
: n < ) of T
3

witnessing that A D
+1
, let
=
xn
: n < and

=
xn

for every n < large enough. So is well

founded (recalling

>

) and we can choose

: ) such
that

= sup

+ 1 : and g() = g() + 1. If

<>
< we are
done otherwise let be -maximal such that

hence

< ,
so necessarily

= =

+ 1 : , g() = g() + 1. Let S S be

such that s) s S, so

<s>
: s S is an unbounded subset of so
cf() (S) < . This takes care of the rst possibility for so the second case is
easier.
3) It suces to show that we can replace x T
2
2
by x T
2
1
.
1.14
{z0.59}
Denition 1.16. 1) For a lter D on a set Y and a set S let
S
(D) be as in clause
(c) of the Observation 1.14(1).
1A) Similarly with S instead S.
2) D is pseudo (S, )-complete if / comp
S,
(D).
2A) Similarly with S instead S.
{z0.79}
Observation 1.17. 1) If h is a function from S
1
onto S
2
then (S
1
) (S
2
) and
every [pseudo] ( S
1
)-complete lter is a [pseudo] ( S
2
)-complete lter.
2. Commuting ranks
The aim of this section is to sort out when two rank rk
D1
, rk
D2
do so called
commute.
{c3.1}
Denition 2.1. Assume that D

is an
1
-complete lter on Y

for = 1, 2. For
1, 2, 3, 4, 5 we say D
2
does -commute with D
1
when:

D1,D2
holds
where:

1
if A D
1
and

B = B
s
: s A)
A
(D
2
) then we can nd A

, B

such
that: A

D
1
, B

D
2
and A

s B
s
: s A so A

2
if A D
1
and

B = B
s
: s A)
A
(D
2
) and J
2
= J[f
2
, D
2
] for some
f
2

Y2
Ord then we can nd A

, B

such that A

D
1
, B

J
+
2
and
A

s B
s
: s A so A

3
if A D
1
and

B = B
s
: s A)
A
(D
2
) and J
1
= J[f
1
, D
1
] for some
f
1

Y1
Ord then we can nd A

, B

such that A

J
+
1
, A

A, B

D
2
and s A

B
s
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 9

4
if A D
1
and

B = B
s
: s A)
A
(D
2
) and

J
1
= J
1
t
: t Y
2
)
satises J
1
t
J[f, D
1
] : f
Y1
Ord and J
2
J[f, D
2
] : f
Y2
Ord then
we can nd A

, B

such that B

J
+
2
and t B

(J
1
t
)
+
and
(s, t) A

s A t B
s
hence A

A, A

D
+
1

5
like
4
but we omit the sequence

J
1
and the demand on A

is A

D
+
1
.
{c3.2}
Remark 2.2. 1) These are seemingly not commutative relations.
2) We shall rst give a consequence and then give sucient conditions.
3) We intend to generalize to systems (see 3.1 and 3.8).
4) Can we below use D

FIL
pcc
(Y
1
), see Denition 1.12? Yes, but only when
we do not use D + A, A D
+
.
{c3.3}
Claim 2.3. rk
D1
(f) rk
D2
(g) when:
(a) D

FIL
cc
(Y

) for = 1, 2
(b) g = g
t
: t Y
2
)
(c) g
t

Y1
Ord
(d) g
Y2
Ord is dened by g(t) = rk
D1
(g
t
)
(e)

f = f
s
: s Y
1
)
(f) f
s

Y2
Ord is dened by f
s
(t) = g
t
(s)
(g) f
Y1
Ord is dened by f(s) = rk
D2
(f
s
)
(a) D
2
does 2-commute with D
1
(b) AC
Y1
holds.
Remark 2.4. In order not to use DC in the proof we should consider as a member
of Ord in clauses (d),(g) of .
Proof. We prove by induction on the ordinal that

if + above hold for D

1
, D
2
, f, g,

f, g and rk
D1
(f) then rk
D2
(g) .
The case = 0 is trivial and the case a limit ordinal follows by the induction
hypothesis. So assume that = + 1.
Let
()
1
A := s Y
1
: f(s) > 0.
As we are assuming rk
D1
(f) > 0 by 1.8(4) necessarily
()
2
A D
1
.
For each s A, f(s) > 0 so applying clause (g) of we get
()
3
rk
D2
(f
s
) > 0 when s A
hence
()
4
B
s
:= t Y
2
: f
s
(t) > 0 belongs to D
2
when s A.
So B
s
: s A)
A
(D
2
). Recall (see (a) of the assumption) that D
2
does 2-
commute with D
1
, apply it to A, B
s
: s A), J
2
:= J[g, D
2
]; so we can nd A

, B

such that
()
5
(a) A

D
1
(and A

A)
(
9
3
8
)


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







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


10 SAHARON SHELAH
(b) B

J
+
2
recalling J
2
= J[g, D
2
] so B

D
+
2
and (by Denition 1.10)
rk
D2+B
(g) = rk
D2
(g)
(c) (s, t) A

s A t B
s
.
Now by the present assumption of

we have
()
6
rk
D1
(f) = + 1.
Hence by the denition of rk and 1.8(2) we can nd f

such that
()
7
(a) f

Y1
Ord and rk
D1
(f

)
(b) f

<
D1
f
(c) by ()
1
without loss of generality s A f

(s) < f(s).

For each s A, clearly f

(s) < f(s) = rk

D2
(f
s
) rk
D2+B
(f
s
), by 1.7(2), clause
(g) of and D
2
D
2
+B

respectively, hence by 1.8(2) for each s Y

1
there is a
function f

s
such that
()
8
(a) f

s

(Y2)
Ord,
(b) f

s
< f
s
mod D
2
if s A and t Y
2
f

s
(t) < f
s
(t)f

s
(t) = 0 = f
s
(t)
(c) rk
D2+B
(f

s
) = f

(s); may require this only for s A.

As Y
1
C by (b) of the assumption, clearly
()
+
8
there is such a sequence

f

= f

s
: s Y
1
).
As s A

t B

f
s
(t) > 0, see ()
4
+ ()
5
, clearly
()
9
if s A

and t B

then f

s
(t) < f
s
(t).
We now dene g

= g

t
: t Y
2
) by
()
10
g

t
(s) = f

s
(t) for s Y
1
, t Y
2
so g

t

Y1
Ord.
So
()
11
s A

t B

t
(s) = f

s
(t) < f
s
(t) = g
t
(s)
hence (recalling A

D
1
by ()
5
(a) and 1.8(3))
()
12
if t B

then g

t
<
D1
g
t
hence rk
D1
(g

t
) < rk
D1
(g
t
).
Dene g

(Y2)
Ord by g

(t) := rk
D1
(g

t
) hence (recalling rk
D1
(g
t
) = g(t))
()
13
g

< g mod D
2
+ B

.
Note that here D
1
+ A

= D
1
, (though not so when we shall prove 2.9).
Now we apply the induction hypothesis to g

, f

, g

:= g

t
: t Y
2
),

f

:= f

s
:
s Y
1
), D
1
+ A

, D
2
+ B

and and get

()
14
rk
D2+B
(g

).
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 11
[Why is this legitimate? First, obviously clauses (a),(b) of holds, second, we have
to check that clauses (a)-(g) of hold in this instance.
Clause (a): First D
1
+ A

FIL
cc
(Y
1
) as we assume D
1
FIL
cc
(Y
1
) and
A

D
1
, see ()
5
(a), actually A

D
+
1
suce (used in proving 2.9).
Second, D
2
+ B

FIL
cc
(Y
2
) as D
2
FIL
cc
(Y
2
) and B

D
+
2
by ()
5
(b).
Clause (b): g

= g

t
: t Y
2
) by our choice.
Clause (c): g

t

Y1
Ord by ()
10
.
Clause (d): g

Y2
Ord is dened by g

(t) = rk
D1
(g

t
) by its choice after ()
12
.
Clause (e):

= f

s
: s Y ) by our choice in ()
+
8
.
Clause (f): f

s

Y2
Ord is dened by f

s
(t) = g

t
(s) holds by ()
10
.
Clause (g): f

Y1
Ord is dened by f

(s) = rk
D2+B
(f

s
) holds by ()
7
(a) +
()
8
(c).
Now

, the induction hypothesis, assumes rk

D1+A
(f

) which holds
by ()
7
(a) + ()
5
(a), actually A

D
+
1
suce here and its conclusion is
rk
D2+B
(g

) as promised in ()
14
.]
Next
()
15
< rk
D2
(g).
[Why?

1
rk
D2+B
(g

) by ()
14

2
rk
D2+B
(g

) < rk
D2+B
(g) by ()
13
and 1.8(3)

3
rk
D2+B
(g) = rk
D2
(g) by ()
5
(b).
Together ()
15
holds.]
So
()
16
= + 1 rk
D2
(g)
as promised. Together we are done.
2.3
{c3.11}
Claim 2.5. Assume D

FIL
cc
(Y

) for = 1, 2.
If D
2
does
1
-commute with D
1
then D
2
does
2
-commute with D
1
when (
1
,
2
) =
(1, 2), (1, 3), (2, 4), (1, 4), (1, 5), (4, 5).
Proof. Obvious for (4,5) use 1.11(3).
2.5
{c3.8}
Claim 2.6. Assume D

FIL
cc
(Y

) for = 1, 2. If at least one of the following

cases occurs, then D
2
does 1-commute (hence 2-commute) with D
1
.
Case 1: D
2
is [Y
1
[
+
-complete.
Case 2: D
1
is an ultralter which is [Y
2
[
+
-complete
Case 3: D
1
, D
2
are ultralters and if

A = A
t
: t Y
2
)
Y2
(D
1
) then for some
A

D
1
we have t : A
t
A

D
2
.
(
9
3
8
)


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







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


12 SAHARON SHELAH
Proof. So let A D
1
and B
s
: s A)
A
(D
2
) be given.
Case 1: Let A

= A and B

= B
s
: s A, so A

D
1
by an assumption
and B

D
2
as we assume B
s
: s A D
2
and D
2
is [Y
1
[
+
-complete (and
necessarily [A[ [Y
1
[).
Case 2: For each t Y
2
let A

t
:= s Y
1
: s A and t B
s
and let A

t
be the unique member of A

t
, Y
1
A

t
D
1
, recalling D
1
is an ultralter on Y
1
.
Clearly the functions t A

t
and t A

t
are well dened hence the sequences
A

t
: t Y
2
), A

t
: t Y
2
) exist and A

t
: t Y
2
D
1
.
As D
1
is [Y
2
[
+
-complete necessarily A

:= A

t
: t Y
2
A belongs to D
1
,
and clearly A

A. Let B

= t Y
2
: A

t
= A

t
.
So now choose any s

(possible as A

D
1
implies A

,= ) so B
s
D
2
and t B
s
s

t
s

t
A

t
A

,= A

t
= A

t
t B

so
B
s
B

but B
s
D
2
hence B

D
2
. So A

, B

are as required.
Case 3:
Like Case 2.
2.6
{c3.10}
Claim 2.7. Assume AC
P(Y2)
.
1) Assume D
1
FIL
cc
(Y
1
) and D
2
FIL
cc
(Y
2
).
Then D
2
does 3-commute with D
1
when D
1
is ( T(Y
2
))-complete.
2) In part (1) if E D
2
is (D
2
, )-conal, it suces to assume D
1
is ( E)-
complete.
Remark 2.8. For part (1) in the denition of ( T(Y
2
))-complete we can use just
partitions, but not so in part (2).
Proof. 1) So let A D
1
and

B = B
s
: s A)
A
(D
2
) and J
1
= J[f
1
, D
1
] for some
f
1

Y
Ord be given. So s B
s
is a function from A D
1
to D
2
T(Y
2
) hence
as AC
P(Y2)
is assumed recalling that by 1.11(1) the ideal J
1
on Y
1
is ( T(Y
2
))-
complete, there is B

D
2
such that A

:= s A : B
s
= B

J
+
1
. Clearly
A

, B

are as required.
2) For B E let A

B
= s A : B B
s
, so clearly A

B
: B E) is a sequence
of subsets of A D
1
with union A, so again by 1.11(1) for some B

E the set
A

:= s A : B

B
s
belongs to J
+
1
, so we are done.
2.7
{c3.14}
Claim 2.9. rk
D1
(f) rk
D2
(g) when:
as in 2.3
but we replace clause () there by

(a) D
2
does 4-commute with D
1
(b) AC
Y1
holds.
Proof. We repeat the proof of 2.3 but:
First change: we replace ()
5
and the paragraph before it by the following:
So

B = B
s
: s A)
A
(D
2
).
Recall that D
2
does 4-commute with D
1
, apply this to A, B
s
: s A),

J
1
=
J
1
t
: t Y
2
) where J
1
t
:= J[g
t
, D
1
], J
2
:= J[g, D
2
] and we get A

, B

such that:
()

5
(a) A

D
+
1
and A

A
(b) B

J
+
2
hence B

D
+
2
and rk
D2+B
(g) = rk
D2
(g)
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 13
(c) (s, t) A

s A t B
s
(d) if t B

then A

(J
1
t
)
+
hence
t B

rk
D1+A
(g
t
) = rk
D1
(g
t
) = g(t).
Second change: we replace ()
12
and the line before, the line after it and ()
13
by:
Dene g

Y2
Ord by g

(t) = rk
D1+A
(g

t
).
Now
()

12
if t B

then
(a) g

t
<
D1+A
g
t
, by ()
11
(b) rk
D1+A
(g

t
) < rk
D1+A
(g
t
) by (a) and 1.8(3),
(c) rk
D1+A
(g
t
) = rk
D1
(g
t
) recalling ()

5
(d) and J
1
t
= J[g
t
, D
1
] hence
(d) rk
D1
(g
t
) = g(t) by clause (d) of
(e) rk
D1+A
(g
t
) = g(t) by (c), (d) above hence
(f) g

(t) < g(t) by the choice of g

, clause (b) and clause (e).

Hence by ()

12
(f) we have
()

13
g

< g mod D
2
+ B

.
Concerning the rest, we quote ()
5
(b) twice but ()

5
(b) = ()
5
(b), and quote ()
5
(a)
twice but noted there that ()

5
(a) suce and g

is dened before ()

12
rather than
apply ()
12
.
2.9
3. Rank systems and A Relative of GCH
To phrase our theorem we need to dene the framework.
{m4.3}
Denition 3.1. Main Denition: We say that p = (D, rk, , j, ) = (D
p
, rk
p
,
p
, j
p
,
p
)
is a weak (rank) 1-system when:
(a) is singular
(b) each d D is (or just we can compute from it) a pair (Y, D) = (Y
d
, D
d
) =
(Y [d], D
d
) = (Y
p,d
, D
p,d
) such that:
() (Y
d
) < , on () see Denition 1.5
() D
d
is a lter on Y
d
(c) for each d D, a denition of a function rk
d
() with domain
Y [d]
Ord and
range Ord, that is rk
p,d
() or rk
p
d
()
(d) () is a function with domain D such that (d) D
() if d D and e (d) then Y
e
= Y
d
[natural to add D
d
D
e
,
this is not demanded but see 3.8(2)]
(e) () j is a function from D onto cf()
() let D
i
= d D : j(d) i and D
i
= D
i
D
i+1
() e (d) j(e) j(d)
(f) for every < for some i < cf(), if d D
i
, then d is (p, )-complete
where:
(
9
3
8
)


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







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


14 SAHARON SHELAH
() we say that d is (p, X)-complete (or ( X)-complete for p) when: if
f
Y [d]
Ord and = rk
d
(f) and A
j
: j X) a partition
1
of Y
d
, then
for some e (d) and j < we have A
j
D
e
and = rk
e
(f); so
this is not the same as D
d
is ( X)-complete; we dene (p, [X[
+
)-
complete, i.e. (p, < [X[
+
)-complete similarly
(g) no hole
2
: if rk
d
(f) > then for some pair (e, g) we have: e (d) and
g <
D[e]
f and rk
e
(g) =
(h) if f = g + 1 mod D
d
then rk
d
(f) = rk
d
(g) + 1
(i) if f g mod D
d
then rk
d
(f) rk
d
(g).
{m4.4}
Denition 3.2. 1) We say p = (D, rk, , j, ) is a weak (rank) 2-system, (if we
write system we mean 2-system) when in 3.1 we replace clauses (d),(f),(g) by:
(d)

() is a function with domain D

() for d D we have (d) (e, h) : e D
j(d)
and h : Y
e
Y
d
;
writing e (d) means then (e, h) (d) for some function h
(f)

for every < for some i < cf(), if d D

i
, then d is (p, )-complete
where:
() we say that d is (p, X)-complete (for p) when: if f
Y [d]
Ord
and = rk
d
(f) and A
j
: j X) a partition
3
of Y
d
, then for some
(e, h) (d) and j < we have h
1
(A
j
) D
e
and = rk
e
(f h);
we dene (p, [X[
+
)-complete similarly
(g)

no hole: if rk
d
(f) > then for some (e, h) (d) and g
Y [e]
Ord we have
g < f h mod D
e
and rk
e
(g) = .
{m4.4d}
Denition/Claim 3.3. Let p be a weak rank 1-system; we can dene q and
prove it is a weak rank 2-system by D
q
= D
p
, rk
q
= rk
p
,
q
(d) = (e, id
Y [d]
) :
e
p
(d), j
q
= j
p
,
q
=
p
.
{m4.5}
Convention 3.4. 1) We use p only for systems as in Denition 3.1 or 3.2.
2) We may not distinguish p and q in 3.3 so deal only with 2-systems.
{m4.5d}
Remark 3.5. The following is an alternative to Denition 3.2. As in 3.1 we can
demand e (d) Y
e
= Y
d
but for every d we have a family c
d
, i.e. the function
d c
d
is part of p and make the following additions and changes:
() c
d
is a family of equivalence relations on Y
d
() we replace
Y [d]
Ord by f
Y [d]
Ord: eq(f) := (s, t) : s, t Y
d
and
f(s) = f(t) c
d

() if E
1
, E
2
are equivalence relations on Y
d
such that E
2
renes E
1
then E
2

c
d
E
1
c
d
() if e (d) then Y
e
= Y
d
and c
d
c
e
.
{m4.6}
Denition 3.6. For = 1, 2 we say that p = (D, rk, , j, ) is a strict -system
when it satises clauses (a)-(i) from 3.1 or from 3.2 and
1
as long as is a well ordered set it does not matter whether we use a partition or just a
covering, i.e. {A
j
: j } = Y
d
2
we may use another function here, as in natural examples here we use (d) = {d} and not
so in clause (f)
3
as long as is a well ordered set it does not matter whether we use a partition or just a
covering, i.e. {A
j
: j } = Y
d
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 15
(j) for every d D and , , f, j
0
satisfying below, there
4
is j < cf() such
that: there are no e, g satisfying below, where:

1
e D
j

2
g
Y [e]

3
g(t) : t Y
e
[,

) for some

<

4
j j
0

5
rk
e
(g) ,

1
f
Y [d]

2
rk
d
(f) =

3
<

4
cf() = cf()

5
j
0
< cf()

6
s Y
d
e

D
j0
rk
e
(f(s)) = f(s).
{m4.9d}
Observation 3.7. 1) If p is a strict -system then p is a weak -system.
2) In Denition 3.6, from (j)
6
recalling (j)
1
+
4
we can deduce: rk
e
(f(s)) =
f(s) for s Y
d
.
3) In
5
of (j) of 3.6 without loss of generality rk
e
(y) > +7 as we can use g +7.
{m4.9}
Denition 3.8. 1) We say that a weak -system p is weakly normal when:

1
in (d)() of Denition 3.1 we add e (d) D
d
D
e

2
in (d)

() of Denition 3.2 we add: if (e, h) (d) then (A D

d
)(h
1
(A)
D
e
).
2) We say p is normal when it is weakly normal and

3
in Denition 3.1, if A D
+
d
, d D, f
Y [d]
Ord and = rk
d
(f) then for
some e (d) we have A D
e
and rk
e
(f) =

4
in Denition 3.2, if d D, f
Y [d]
Ord, rk
d
(f) = and A D
+
d
, then for
some (e, h) (d) we have s Y
e
: h(s) A D
e
and rk
e
(f h) = .
3) We say p is semi normal when it is weakly normal and we have

3
,

4
holds
where they are as above just ending with .
{m4.10}
Claim 3.9. Assume p is a weak -system and d D
p
.
0) If p, q are as in 3.3, then p is [weakly] normal i q is.
1) If f, g
Y [d]
Ord and f <
D
d
g then rk
d
(f) < rk
d
(g).
2) If f
Y [d]
Ord and rk
d
(f) > 0 then s Y
d
: f(s) > 0 D
+
d
.
2A) If in addition p is semi-normal then s Y
d
: f(s) = 0 = mod D
d
.
3) rk
d
(f) depends just on f/D
d
(and d and, of course, p).
Proof. 0) Easy; note that by this part, below without loss of generality
p
= 2.
1) Let f
1

Y [d]
Ord be dened by f
1
(s) = f(s) + 1. So clearly f
1

D
d
g hence by
clause (i) of 3.1 (equivalently 3.2) we have rk
d
(f
1
) rk
d
(g). Also f
1
= f +1 mod
D
d
hence by clause (h) of 3.1 (equivalently 3.2) we have rk
d
(f
1
) = rk
d
(f)+1. The
last two sentences together give the desired conclusion.
4
can we make j depend on f (and a partition of) d? Anyhow later we use d

(d), if
{d} = (d)? Also so = 1, 2 may make a dierence.
(
9
3
8
)


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







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


16 SAHARON SHELAH
2) Toward contradiction assume the conclusion fails. Let f

Y
d
Ord be constantly
zero, so f = f

mod D
d
hence by part (3) we have rk
d
(f

) = rk
d
(f) > 0. By clause
(g)

of Denition 3.2, the no hole applied to (f

, d), there is a triple (e, h, g) as

there, so B := s : s Y
e
and g(s) < f(h(s)) D
e
, i.e. s Y
e
: g(s) < 0 D
e
,
contradiction.
Hence by the weak normality of p we have h(s) : s B ,= mod D
d
but this
set includes s Y
d
: f(s) > 0.
2A) Let A = s Y
a
: f(s) = 0, so toward contradiction assume A D
+
d
. As p is
semi-normal we can nd a pair (e, h) (d) as in

4
of 3.8(3) so A
1
= t Y
e
:
h
1
(t) A D
e
and rk
e1
(f h) rk
d
(f) > 0, but clearly t Y
e
: (f h)(t) =
0 D
e
, contradiction by part (2).
3) Use clause (i) of Denition 3.1 twice.
3.9
{m4.12}
Theorem 3.10. [ZF] Assume that p = (D,rk,, j, ) is a strict rank 1-system (see
Main Denition 3.1) or just a strict 2-system. Then for every ordinal there is
i < cf() such that: if d D
i
then rk
d
() = , i.e. rk
d
( : s Y
d
)) = .
Proof. We shall use the notation:

0
If there is an i as required in the theorem for the ordinal then let i() be
the minimal such i (otherwise, i() is not well dened).
Without loss of generality,

1
every d D
p
is (p, (cf()))-complete, i.e. clause (f) of 3.1 for

:=
cf()
+
holds for every d D.
[Why? Let i

be the i < cf() which exists by clause (f) of Denition 3.1, 3.2 for

. Now we just replace D by D

i
(and j by jD
i
, etc).]
Clearly we have

2
rk
d
() for an ordinal and d D.
[Why? We can prove this by induction on for all d D, by clauses (h) + (i) of
Denition 3.1.]
As a warmup we shall note that:

3
if d D and <

or just d D and is hereditarily (p, )-complete

which means that every e in the -closure of d is (p, )-complete then:
() rk
d
() =
() f
Y [d]
rk
d
(f) < .
[Why? Note that as <

clearly d is (p, )-complete by

1
and clause (f)
of 3.1, so we can assume that d is hereditarily (p, )-complete. We prove the
statement inside
3
by induction on the ordinal (for all hereditarily (p, )-
complete d D). Note that for < , d is (p, )-complete implies d is
(p, )-complete, we shall use this freely.
Arriving to , to prove clause () let f
Y [d]
and for < we dene A

:=
t Y
d
: f(t) = , so A

: < ) is a well dened partition of Y

d
so the sequence
exists, hence as d is hereditarily (p, )-complete recalling () from clause (f)

of 3.2 for some triple (e, h, ) we have (e, h) (d) and < and h
1
(A

) D
e
and rk
e
(f h) = rk
d
(f).
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 17
Now f h = : t Y
e
) mod D
e
hence by Claim 3.9(3) we have rk
e
(f h) =
rk
e
(). But the assumptions on d holds for e hence by the induction hypothesis
on we know that rk
e
() = and < so together rk
d
(f h) < , so clause ()
of
3
holds.
To prove clause () rst consider = 0; if rk
d
() > 0 by clause (g) of Denition
3.1, 3.2 there are (e, h) (d) and g
Y [e]
Ord such that g < : t Y
e
) mod D
e
,
so for some t Y
e
we have g(t) < but = 0, contradiction; this is close to 3.9(2).
Second, consider > 0, so by
2
we have rk
d
() and assume toward
contradiction that rk
d
() > , so by clause (g) of Denition 3.1, 3.2 there is a
triple (e, h, g) as there. Now apply clause () of
3
for (which we have already
proved) recalling (e, h) (d) so also e is (p, )-complete. We get rk
e
(g) < , a
contradiction. So
3
indeed holds.]
Now as in the desired equality we have already proved one inequality in
2
, we
need to prove only the other inequality. We do it by induction on .
Case 1: < .
By clause (f) for some i < cf() we have d D j(d) i d is (p, )-
complete, hence by
3
() we have rk
d
() = , as required.
Case 2: = + 1.
By clause (h) of Denition 3.1 we have d D rk
d
() = rk
d
() + 1. Hence
d D
i()
rk
d
() = rk
d
() + 1 = + 1 = , so i() exemplies that i() exists
and is i() so we are done.
Case 3: is a limit ordinal of conality ,= cf().
So for each < by the induction hypothesis i() < cf() is well dened. For
i < cf() let u
i
:= < : i() i, so is well dened; moreover the sequence
u
i
: i < cf()) exists and is -increasing. If i < cf() sup(u
i
) < then
sup(u
i
) : i < cf()) is a -increasing sequence of ordinals < with limit . So as
cf() ,= cf() necessarily for some i

< cf() the set S := : < and i() < i

is an unbounded subset of . We shall prove that i() is well dened and i

.
Subcase 3A: cf() .
Let d D
i
and g
Y [d]
be given. Clearly Rang(g) is a subset of of
cardinality < (Y
d
) which by clause (b)() of 3.1 is < cf() hence we can
x S such that Rang(g) , hence by clause (i) of 3.1, rk
d
(g) rk
d
() but
i() = i

and d D
i
hence rk
d
() = < so together rk
d
(g) < . As this holds
for every d D
i
by the no-hole clause (g)

and clause (e)() of 3.2 it follows that

d D
i
rk
d
() as required.
Subcase 3B: cf() < (but still cf() ,= cf()).
Let

: < cf()) be an increasing sequence of ordinals from S with limit .

Now let j

< cf() be such that d D

j
d is (p, cf()
+
)-complete, see clause
(f) of Denition 3.1.
Now assume d D
max{i,j}
and g
Y [d]
. For < cf() let A

= t Y
d
:
g(t) <

but j < g(t)

j
so A

: < cf()) is well dened and is a partition

of Y
d
. Hence by clause (f) of Denition 3.2 for some < cf() and (e, h) (d)
we have h
1
(A

) D
e
and rk
d
(g) = rk
e
(g h); but j(e) j(d) i

, j

and by
the choice of A

and clause (i) of 3.1 the latter is rk

e
(

) hence as i(

) i

the latter is =

< . As this holds for every d D

max{i,j}
and g
Y [d]
, by
(
9
3
8
)


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







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


18 SAHARON SHELAH
the no-hole clause (g)

of 3.2 necessarily rk
d
() . So maxi

, j

< cf() is as
required, so we are done.
Case 4: is a limit ordinal such that cf() = cf().
Let
i
: i < cf()) be increasing with limit . Assume toward contradiction that
for every i < cf() there is d
i
D
i
such that rk
di
() > but we do not assume
that such a sequence d
i
: i < cf()) exists. Choose such d
0
; as rk
d0
() > , clearly
there are f
0

Y [d

0
]
and a member d

0
of (d
0
), though not necessarily Y
d

0
= Y
d
,
such that

4
rk
d

0
(f
0
) =
[Why? By using clause (g)

of 3.2.]
Note

5
i(f
0
(t)) is well dened for every t Y
d

0
.
[Why holds? Because f
0
(t) < and the induction hypothesis.]
For j
1
< cf(), j
0
< cf() let A
j1,j0
= t Y
d

0
: f
0
(t) <
j1
and (j <
j
1
)(f
0
(t)
j
) and i(f
0
(t)) = j
0
. By clause (f)

of 3.2 applied to the pair (d

0
, f
0
)
and the partition A
j1,j0
: j
1
< cf(), j
0
< cf()), for some (d

, h

) (d

0
) we
have rk
d
(f
0
h

) = and for some j

1
, j
2
we have h
1

(A
j1,j0
) D
d
. By 3.9(3)
for some f = f
0
h

mod D
d
and letting d := d

we have

6
(a) d D
(b) f
Y [d]
Ord
(c) rk
d
(f) =
(d) t Y
d
i(f(t)) = j
2
f(t) <
j1
(j < j
1
)(f(t)
j
).
Next

7
letting :=
j1
, clause from 3.6 for (d, , , f, j
0
).
[Why? We check the six demands

1
f
Y [d]
which holds by
6
(b) + (d)

2
rk
d
(f) = which holds by
6
(c)

3
< which holds as (i < cf())(
i
< )

4
cf() = cf() which holds by the case assumption

5
j
2
< cf() obvious

6
s Y
d
e

D
j0
rk
e
(f(s)) = f(s) holds by
6
(d).
So
7
indeed holds.]
Now by
7
, clause (j) of Denition 3.6(1) applied with d, , =
j1
, f, j
0
here
standing for d, , , f, j
0
there, we can nd j as there. Let i
2
= maxj, j
1
, j
0
, i(
j1
)
so i
2
< cf() and choose e
0
D
i2
such that rk
e0
() > as in the beginning of
the case. As rk
e0
() > by clause (g)

of 3.2 there are e

1
(e
0
) and g
Y [e1]

such that rk
e1
(g) so g < : t Y
e1
). Now without loss of generality

8
(a) rk
e1
(g) = + 1
(b)

= supg(t) : t Y
e1
<
(c) j(e
1
) i
2
.
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 19
[Why? Because we can use g

Y [e1]
dened by g

(t) = g(t) + 2 for t Y

e1
, by
clause (h) of 3.1, rk
e1
(g

) = rk
e1
(g) + 2 > . By clause (e)() we have j(e
1
)
j(e
0
) i
2
. Now we nd (d

2
, h

) (e
1
) and g
2
as in the proof of
6
and rename.]
Also without loss of generality

9
t Y
e1
g(t)
j1
.
[Why? Let A
0
= t Y
e1
: g(t) <
j1
, A
1
= t Y
e1
: g(t)
j1
so by clause
(f)

of 3.2 for some pair (e

2
, h) (e
1
) we have rk
e2
(g h) = rk
e1
(g) = +1 and
(h
1
(A
0
) D
e2
) (h
1
(A
1
) D
e2
). So if h
1
(A
0
) D
e2
then by clause (i) of 3.1,
rk
e1
(g h) rk
e2
() but i() is well dened i
2
j(e
1
) j(e
2
) so rk
e2
() =
together rk
e2
(g h) contradicting the previous sentence. Hence h
1
(A
0
) / D
e2
so h
1
(A
1
) D
e2
. Let g

Y [e2]
Ord be dened by g

(t) is (g h)(t) if t h
1
(A
1
)
and is
j1
+ 1 if t h
1
(A
0
). By Claim 3.9(3) we have rk
e2
(g

) = rk
e2
(g h) so
(e
2
, g

) satises all requirements on the pair (e

1
, g) and t Y
e2
g

(t)
j1
> 0,
so we have justied the non-loss of generality.]
Recall :=
j1
and let e = e
1
. By the choice of j after
6
, i.e. as in
clause (j) of 3.6, recalling e D
j
we shall get a contradiction to the choice
of (d, , , f, j
0
, e, g, j). To justify it we have to recall by
7
that the quintuple
(d, , , f, j
0
) satises of 3.6(j) and then we prove that the triple (e, g, j) satises
of 3.6(j).
Now of 3.6 says:

1
e D
j
as
as j i
2
, e
0
D
i2
and e = e
1
(e
0
)

2
g
Y [e]

which holds as g
Y [e]

3
g(t) [,

)
holds as g(t) < by
2
+
8
(b) and g(t)
j1
= by
9

4
j j
0

holds as j i
2
j
0

5
rk
e
(g) >
holds by
8
(a).
So we really get a contradiction.
3.10
{m4.14}
Denition 3.11. 1) We say that the pair (d, e) commute (or 6-commute) for p
when d, e D
p
and rk
d
(f) rk
e
(g) whenever (f, g,

f, g) is a (p, d, e)-rectangle,
see below; xing f, g we may say (d, e) commute for f, g.
2) We say that (d, e, f, g,

f, g) is p-rectangle or (f, g,

f, g) is a (p, d, e)-rectangle
when:
(a) d D
p
(b) e D
p
(c) g = g
t
: t Y
e
) and g
t

Y [d]
Ord for t Y
e
(d) g
Y [e]
Ord is dened by g(t) = rk
d
(g
t
)
(e) f
s

Y [e]
Ord is dened by f
s
(t) = g
t
(s)
(f)

f = f
s
: s Y [d])
(g) f
Y [d]
Ord is dened by f(s) = rk
e
(f
s
).
(
9
3
8
)


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







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


20 SAHARON SHELAH
{m4.16}
Claim 3.12. [Assume ZF + AC
<
] If p = (D, rk, , i, ) be a weak rank 1-system
then p is a strict rank 1-system when there is a function
1
such that (and we may
say
1
witness it):
()
0

1
a function with domain D
()
1

1
(d) (d) for d D
()
2
for every d, , , f, j
0
satisfying of 3.6, for some j < cf() for every
e D
j
we have
(a) e is (p,
1
(d))-complete
(b) if d

1
(d), e

1
(e) then (d

, e

) commute (for p) see 3.11, at

least for f
Y [d]
and any g
Y [e]
[, )
()
3
we strengthen clause (g) of Denition 3.1 to
(g)
+
add: rk
e
(f) = rk
d
(f) and e
1
(d)
()
4
AC
Y [d]
and AC
1(d)
whenever d D.
Remark 3.13. 1) In ()
1
, can we make j depend on f and a partition of Y
d
? Will
be somewhat better.
2) We can similarly prove this for a weak rank 2-system. It is natural though not
necessary to add (e, h)
1
(d) Y
e
= Y
d
h = id
Y
d
.
Proof. Let d, , , f, j
0
satisfying of 3.6(j) be given and we should nd j < cf()
such that for no pair (e, g) clause there holds. Without loss of generality s
Y
d
f(s) > 0.
Let j < cf() be as in ()
2
in the claim and without loss of generality j > j
0
and we shall prove that j is as required in clause (j) of Denition 3.6, this is enough.
So assume e D
j
, g
Y [e]
[, ] and toward contradiction, (j, , , e, g) satisfy
there. For each t Y
e
clearly g(t) < = rk
d
(f) hence by clause (g)
+
of ()
3
,
see (g) of Denition 3.1, no hole, there are g
t

Y [d]
and d
t

1
(d) such that
g
t
<
D
d
t
f and rk
dt
(g
t
) = g(t), without loss of generality g
t
< max(f, 1
Y [d]
) = f
and by the ()
3
, we add also rk
dt
(f) = rk
d
(f).
As AC
Ye
by ()
4
, we can choose such sequence (g
t
, d
t
) : t Y
e
). Now e is
(p,
1
(d))-complete and (d, e) commute for p, by clauses (a),(b) respectively of
()
2
(i.e. by the choice of j and as e D
j
), hence we can nd e

1
(e) and
d

1
(d) such that rk
e
(g) = rk
e
(g) = and t Y
e
: d
t
= d

belongs to D
e
.
For s Y
d
= Y
d
let f
s

Y [e]
Ord be dened by f
s
(t) = g
t
(s) so f
s
(t) = g
t
(s) <
and let f

Y [d]
Ord be dened by f

(s) = rk
e
(f
s
) and let

f = f
s
: s Y
d
).
Fixing s Y
d
we have t Y
e
f
s
(t) = g
t
(s) < Maxf(s), 1 = f(s),
i.e. f
s
< f(s) : t Y
e
) hence rk
e
(f
s
) < rk
e
(f(s)). Now by
6
from 3.6, as
j
0
j j(e

) we have s Y
e
rk
e
(f(s)) = f(s) so s Y
e
rk
e
(f
s
) < f(s),
i.e. f

< f.
Clearly (f

, g,

f, g) is a (p, d

, e

)-rectangle hence by clause (b) of ()

2
of the
assumptions, i.e. the choice of (e, g) and Denition 3.11(2) we know that rk
d
(f

)
rk
e
(g).
But recall that rk
e
(g) = rk
e
(g) by the choice of e

. We get a contradiction by
() = rk
d
(f) = rk
d
(f) > rk
d
(f

) rk
e
(g) = rk
e
(g) .
[Why those inequalities? By
2
of from 3.6 we are assuming; as d

d
t
: t Y
e

and the choice of the d

t
s; as f

<
D
d
f and 3.9(3); by an inequality above; by the
choice of e

; by
5
of of 3.6.]
3.12
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 21
4. Finding Systems
(4A) Building weak rank systems and measurable
{e5.3}
Claim 4.1. [ZF + DC]
If
1
holds and p
,
= p

= p = (D, rk, , j, ) is dened in
2
then p is a weak
rank 1-system, even semi normal (and (g)
+
of 3.12 holds) where:

1
(a) =
i
: i < ) is an increasing sequence of regular cardinals
> = cf() with limit such that if i < is a limit ordinal then

i
= (
j
: j < i)
+
(b)

is a cardinal or

2
(a) D
i
= J : J is a
i
-complete ideal on some =
J
< including
[]
<
and satisfying cf(J, ) <

(and if

= we stipulate this as the empty

demand) such that < J and let D = D
0
(b) if d = J D
i
, J an ideal on
J
:= A : A J
then we let Y
d
=
J
and D
d
be the lter dual to the ideal J
(c) j(J) = mini : J is not
+
i+1
-complete
(d) (J) = J + B : B A and
J
B is not in J
(e) rk
J
(f) is as in Denition 1.6.
Proof. So we have to check all the clauses in Denition 3.1.
Clause (a): As =
i
: i < , the sequence
i
: i < ) is increasing and
0
>
(all by
1
) clearly is a singular cardinal (and = cf()).
Clause (b): Let d D, so d = J.
Subclause (): So Y
d
=
J
< hence (Y
d
) = (
J
) =
+
J
< recalling is a
limit cardinal and the denition of D = D
0
in clause (a) of
2
.
Subclause (): Also obvious.
Clause (c): For f
(
d
)
Ord, rk
d
(f) as dened in
2
(e), is an ordinal recalling
Claim 1.8(1).
Clause (d)():
Trivial.
Clause (d)():
Trivially e (d) Y
e
= Y
d
D
e
D
d
; so p is weakly normal, see Denition
3.8, moreover p is semi-normal as rk
D
(f) rk
D+A
(f) for A D
+
.
Clause (e):
Obvious from the denitions.
Clause (f):
Let < be given and choose i < such that <
i
. Let d D be such
that j = j(d) i hence D = D
d
is a lter on some
J
, so assume A

: <

=
J
and

<
i
. Now D is
i
-complete and (see 1.9(2)) we have rk
D
d
(f) =
minrk
D+A
(f) : <

and A

D
+
d
which is what is needed as A

D
+
d

d + (
j
A

) (d).
(
9
3
8
)


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







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


22 SAHARON SHELAH
Clause (g): By 1.8(2).
Moreover, the stronger version with e = d holds so in particular (g)
+
of 3.12
holds.
Clause (h):
Easy. On the one hand, as g < f, the denition of rk
d
(f), we have rk
d
(f)
rk
d
(g) + 1. On the other hand, if g

< f mod D
d
then g

g mod D hence by
clause (i) below we have rk
d
(g

) rk
d
(g) < rk
d
(g) + 1, as this holds for every
g

< f mod D
d
we have rk
d
(f) rk
d
(g) + 1. Together we are done.
Clause (i):
Obvious.
4.1
Discussion 4.2. Assume is a singular cardinal, =

i<

i
, = cf() <
i
<
and
i
is increasing with i. Assume that for each i there is a pair (D, Y ), D is
a
i
-complete ultra-lter on Y, (Y ) < . This seems to be a good case, but
either we have D is a ( (Y ))-complete so
Y
Ord/D is dull or (Y ) > =
completeness(D) and so there is a -complete non-principal ultralter on and on
< so = sup(measurables ).
{e5.13}
Claim 4.3. [ZF + DC + AC
<
]
Assume is singular and = sup( the class of measurable cardinals), (equiv-
alently for every < there is a -complete non-principal ultralter on some

< ). Let =
i
: i < cf()) be increasing with limit ,
i
> cf() such that for
i limit
i
= (
j
: j < i)
+
and
i
is measurable for i non-limit.
Then p = p
uf

is a strict rank 1-system where p is dened by
(a) D
i
= J: dual (J) is a non-principal ultra-lter which is

i
-complete on some =
J
<
so naturally Y
J
=
J
and D
J
= dual(J)
(b) j(J) = mini : J is not
+
i+1
-complete, well dened
(c) (J) = J
(d) rk
J
(f) = rk
dual(J)
(f) as in 1.6.
Proof. We can check clauses (a)-(i) of 3.1 as in the proof of 4.1.
We still have to prove the strict, i.e. we should prove clause (j) from Denition
3.6. We prove this using Claim 3.12, we choose
1
(d) := d (d) for d D
p
so it suces to prove ()
0
()
4
of 3.12.
So in Claim 3.12, we have ()
0
, ()
1
hold by the choice of
1
, and concerning
()
3
in 4.1 we prove (g)
+
, and ()
4
holds as for each < we have AC

as <
by an assumption and for d D we have AC
1(d)
, as
1
(d) is a singleton.
Note that

1
if < < (T()) then is not measurable.
Now we are left with proving ()
2
, so let d D
p
, , , f
Y [d]
be given as in (j)
of in 3.6, and we should nd j as there.
Let j < = cf() be such that (T()) <
j
, and let e D
j
. Now clause (a)
is trivial as [
1
(d)[ = 1, and clause (b) says that the pair (d, e) commute, see
Denition 3.11 recalling
1
(d) = d. So let (f, g,

f, g) be a (p, d, e)-rectangle, see
Denition 3.11(2), and we should prove that rk
e
(g) rk
d
(f); let Y
1
= Y
e
, Y
2
= Y
d
.
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 23
To prove this we apply 2.3 or 2.9, but the f,

f are interchanged with g, g; we
check (a) (g) from 2.3. They hold by (a) (f) of Denition 3.11.
Concerning (a), (b) from 2.3, AC
Y1
holds as AC
<
holds and the denition
of p. Lastly, we should prove (a) there which says D
d
does 2-commute with D
e

which holds by Case 2 of Claim 2.6.

4.3
{5e.17}
Conclusion 4.4. [AC
<
, a singular cardinal] Assume = sup < : is a
measurable cardinal. Then for every ordinal for some < we have rk
D
() =
for every -complete ultralter on some cardinality < .
Proof. t suces to prove this for the case has conality
0
. Now we can apply
Claim 4.3 and Theorem 3.10.
I
5. pseudo true cofinality
We repeat here [Sh:938, 5].
Pseudo PCF
We try to develop pcf theory with little choice. We deal only with
1
-complete
lters, and replace conality and other basic notions by pseudo ones, see below.
This is quite reasonable as with choice there is no dierence.
This section main result are 5.9, existence of lters with pseudo-true-conality;
5.19, giving a parallel of J
<
[].
In the main case we may (in addition to ZF) assume DC + AC
P(P(Y ))
; this will
be continued in [Sh:955].
{r1}
Hypothesis 5.1. ZF
{r2}
Denition 5.2. 1) We say that a partial order P is (< )-directed when every
subset A of P of power < has a common upper bound.
1A) Similarly P is ( S)-directed.
2) We say that a partial order P is pseudo (< )-directed when it is (< )-directed
and moreover every subset P

: < has a common upper bound when:

(a) if < is a limit ordinal
(b)

P = P

: < ) is a sequence of non-empty subsets of P

(c) if
1
<
2
, p
1
P
1
and p
2
P
2
then p
1
<
P
p
2
.
2A) For a set S we say that the partial order P is pseudo ( S)-directed when
P
s
: s S has a common upper bound whenever
(a) P
s
: s S) is a sequence
(b) P
s
P
(c) if s S then P
s
has a common upper bound.
{r3}
Denition 5.3. We say that a partial (or quasi) order P has pseudo true conality
when: is a limit ordinal and there is a sequence P

: < ) such that

(a) P

P and = sup < : P

non-empty
(b) if
1
<
2
< , p
1
P
1
, p
2
P
2
then p
1
<
P
p
2
(c) if p P then for some < and q P

we have p
P
q.
(
9
3
8
)


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







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


24 SAHARON SHELAH
{r4}
Remark 5.4. 0) See 5.2(2) and 5.8(1).
1) We could replace by a partial order Q.
2) The most interesting case is in Denition 5.6.
3) We may in Denition 5.3 demand is a regular cardinal.
4) Usually in clause (a) of Denition 5.3 without loss of generality

,= , as
without loss of generality = cf() using P

= P
f()
where f() = the -th
member of C where C is an unbound subset of < : P

,= of order type
cf(). Why do we allow P

= ? as it is more natural in 5.17(1), but can usually

ignore it.
{r5}
Example 5.5. Suppose we have a limit ordinal and a sequence A

: < ) of
sets with

<
A

= ; moreover u = sup(u)

u
A

= . Dene a partial
order P by:
(a) its set of elements is (, a) : a A

and <
(b) the order is (
1
, a
1
) <
P
(
2
, a
2
) i
1
<
2
(and a

for = 1, 2).
It seems very reasonable to say that P has true conality but there is no increasing
conal sequence.
{r6}
Denition 5.6. 1) For a set Y and sequence =
t
: t Y ) of ordinals and
cardinal we dene
ps-tcf-l

( ) = D : D a -complete lter on Y such that ( /D)

has a pseudo true conality;
see below.
2) We say that /D or ( , D) or ( , <
D
) has pseudo true conality when D
is a lter on Y = Dom( ) and is a limit ordinal and the partial order ( , <
D
)
essentially does
5
, i.e., there is a sequence

T = T

: < ) satisfying:

F
(a) T

f
Y
Ord : f <
D

(b) T

,= 0
(c) if
1
<
2
, f
1
T
1
and f
2
T
2
then f
1
< f
2
mod D
(d) if f
Y
Ord and f < mod D then for some < we have g T

f < g mod D (by clause (c) this is equivalent to: for some <
and some g T

we have f g mod D).

3) ps-pcf

( ) = ps-pcf
-comp
( ) := : there is a -complete lter D on Y such
that /D has pseudo true conality and is minimal for D.
4) pcf-l
,
( ) = D : D a -complete lter on Y such that /D has true co-
nality .
5) In part (2) if is minimal we call it ps-tcf( , D) or simply ps-tcf( , <
D
);
note that it is a well dened (regular cardinal).
5
so necessarily {s Y : s > 0} belongs to D but is not necessarily empty; if it is = Y then
= , so pedantically this is wrong, ( , <
D
) does not have any pseudo true conality hence
we say essentially but usually we shall ignore this or assume
V
t
t = 0 when not said otherwise.
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 25
{r7}
Claim 5.7. 1) If = ps-tcf( , <
D
), then ( , <
D
) is pseudo (< )-directed.
1A) If (S) < = ps-tcf( , <
D
) then ( , <
D
) is pseudo ( S)-directed.
2) Similarly for any quasi order.
3) If cf(
t
) = cf() for t Y then ( , <
D
) is -directed.
4) Assume AC

for < . If cf(

s
) for s Y then ( , <
D
) is pseudo
-directed.
Proof. 1), 1A), 2) As in 5.8(1) below.
3) So assume T satises [T[ < , so there is a sequence f

: < ) listing
T for some < . Let f be dened by f(s) = supf

(s) : < , now

f(s) < (s) as cf(
s
) > .
4) So assume

P = P

: < ), a limit ordinal < and P

non-empty and
< < f P

g P

f <
D
g. As AC

holds we can nd a sequence

f = f

: )

<
P

and apply part (3).

5.7
{r8}
Claim 5.8. Let =
s
: s Y ) and D is a lter on Y .
0) If /D has pseudo true conality then ps-tcf( , <
D
) is a regular cardinal;
similarly for any partial order.
1) If /D has pseudo true conality
1
and true conality
2
then cf(
1
) =
cf(
2
) = ps-tcf( , <
D
), similarly for any partial order.
2) ps-pcf

( ) is a set of regular cardinals so if /D has pseudo true conality

then ps-tcf( , <
D
) is where = cf() and /D has pseudo conality .
3) Always ps-pcf

( ) has cardinality < (D : D a -complete lter on Y ).

4) If

=
s
: s Y )
Y
Ord and s :
s
=
s
D then ps-tcf( /D) =
ps-tcf(

/D) so one is well dened i the other is.

Proof. 0) By the denitions.
1) Let T

: <

) exemplify /D has pseudo true conality

for = 1, 2.
Now
() if 1, 2 and

<

then for some

3
<
3
we have g
1
T

g
2

T
3

3
g
1
<
D
g
2
.
[Why? Choose g

+1
, choose
3
<
3
and g
3
T
3

3
such that
g

< g
3
mod D. Clearly f T

f <
D
g

<
D
g
3
so g
3
is as required.]
Hence
() h
1
:
1

2
is well dened when
h
1
(
1
) = Min
2
<
2
: (g
1
T
1
1
)(g
2
T
2
2
)(g
1
< g
2
mod D).
Clearly h is non-decreasing and it is not eventually constant (as T
1

: <
1
is
conal in /D) and has range unbounded in
2
(similarly).
The rest should be clear.
2) Follows.
3),4) Easy.
5.8
Concerning [Sh:835]
(
9
3
8
)


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







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


26 SAHARON SHELAH
{r9}
Claim 5.9. The Existence of true conality lter [ >
0
+ DC + AC
<
] If
(a) D is a -complete lter on Y
(b)
Y
Ord
(c) := rk
D
( ) satises cf() (Fil
1

(Y )), see below.

Then for some D

we have
() D

is a -complete lter on Y
() D

D
() /D

has pseudo true conality, in fact, ps-tcf( , <

D
) = cf(rk
D
( )).
Recall from [Sh:835]
{r9a}
Denition 5.10. 0) Fil
1

(Y ) = D : D a -complete lter on Y and if D

Fil
1

(Y ) then Fil
1

(D) = D

Fil
1

(Y ) : D D

.
1) Fil
4

(Y ) = (D
1
, D
2
) : D
1
D
2
are -complete lters on Y .
2) J[f, D] where D is a lter on Y and f
Y
Ord is A Y : A = mod D or
rk
D+A
(f) > rk
D
(f).
{r9b}
Remark 5.11. 1) On the Denition of pseudo (< , 1 +)-complete D see 1.13; we
may consider changing the denition of Fil
1

(Y ) to D is
1
-complete and pseudo(<
, 1 + ))-complete lter on Y .
Proof. Proof of the Claim 5.9
Recall y Y :
y
= 0 = mod D as rk
D
(
y
: y Y )) = > 0 but
f
1
, f
2

Y
Ord (f
1
= f
2
mod D) rk
D
(f
1
) = rk
D
(f
2
) hence without loss of
generality y Y
y
> 0.
Let D = D

: D

is a lter on Y extending D which is -complete. So (D)

(Fil
1
1
(Y )) cf(). For any < rk
D
( ) and D

D let
()
2
(a) T
,D
= f : rk
D
(f) = and D

is dual(J[f, D])
(b) T
D
= T
,D
: < rk
D
( )
(c)
,D
= < rk
D
( ) : T
,D
,=
(d) T

= T
,D
: D

D.
Now
()
3
if < rk
D
( ) then T

,= .
[Why? By 1.8(2) there is g
Y
Ord such that g < mod D and rk
D
(g) =
and without loss of generality g . Now let D

= dual(J[g, D]), so (D, D

)
Fil
4

(Y ) by 1.11(1) (using AC
<
) the tler D

is -complete so D

D and clearly
g T
,D
, see 1.8(2), but T
,D
T

so T

,= 0; here we use AC
<
.]
()
4
sup(
,D
) : D

D and
,D
is bounded in rk
D
( ) is a subset of
rk
D
( ) which has cardinality < (D) (Fil
1

(Y )) cf().
[Why? The function D

sup(
,D
) witness this.]
()
5
the set in ()
4
is bounded below rk
D
( ) so let () < rk
D
( ) be its supre-
mum.
[Why? By ()
4
.]
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 27
()
6
there is D

D such that
,D
is unbounded in ( , <
D
).
[Why? Choose < rk
D
( ) such that > (). By ()
3
there is f T
()
and by
()
2
(d) for some D

D we have f T
(),D
so by the choice of () the set
,D

cannot be bounded in rk
D
( ).]
()
7
if
1
<
2
are from
,D
and f
1
T
1,D
, f
2
T
2,D
then f
1
<
D
f
2
.
[Why? By 1.8.]
Together we are done: by ()
6
there is D

D such that
,D
is unbounded in
rk
D
( ). Hence

T = T
,D
:
,D
) witness that ( , <
D
) has pseudo true
conality by ()
7
, and so ps-tcf( , <
D
) = cf(otp(
,D
)) = cf(rk
D
( )), so we
are done.
5.9
So we have
{r10}
Denition/Claim 5.12. 1) We say that = ps-tcf
D
( ), where is a limit ordinal
when, for some set Y :
(a)
Y
Ord
(b)

D = (D
1
, D
2
)
(c) D
1
D
2
are
1
-complete lters on Y
(d) rk
D1
( ) = = sup(
D,
) where
D,
= < rk
D1
( ): for some f <
mod D
1
, we have rk
D1
(f) = and D
2
= dual(J[f, D
1
].
2) If D
1
is
1
-complete lter on Y, =
t
: t Y ) and cf(
t
) (Fil
1
1
(Y ))
for t Y then for some
1
-complete lter D
2
on Y extending D
1
we have ps-
tcf
(D1,D2)
( ) is well dened.
3) Moreover in part (2) there is a denition giving for any (Y, D
1
, D
2
, ) as there,
a sequence T

: < ) exemplifying the value of ps-tcf

D
( ).
Proof. 2), 3) Let := rk
D1
(f), so by Claim 5.7(3) we have cf() (Fil
1
1
(Y ))
hence by Claim 5.9 above and its proof the conclusion holds: the proof is needed
for = sup(
D,
), noting observation 5.13 below.
5.12
{r10d}
Observation 5.13. 1) [DC] or just [AC
0
].
Assume D is an
1
-complete lter on Y and f, f
n

Y
Ord for n < and
f(t) = supf
n
(t) : n < . Then rk
D
(f) = suprk
D
(f
n
) : n < .
Remark 5.14. Similarly for other amounts of completeness, see 5.18.
Proof. As rk
D
(f) = minrk
D+An
(f) : n < if A
n
: n < D, A
n
D
+
by
1.9 or see [Sh:71].
5.13
Remark 5.15. Also in 1.9(2) can use AC
Y
only, i.e. omit the assumption DC, a
marginal point here.
{r11}
Claim 5.16. [AC
<
] The ordinal has conality when :
(a) = rk
D
( )
(b) =
y
: y Y )
Y
Ord
(c) D is an
1
-complete lter on Y
(d) y Y cf(
y
) .
(
9
3
8
)


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







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


28 SAHARON SHELAH
Proof. Note that y Y
y
> 0. Toward contradiction assume cf() < so
has a conal subset C of cardinality < . For each < for some f
Y
Ord we
have rk
D
(f) = and f <
D
and without loss of generality f

yY

y
. By AC
<
there is a sequence f

: C) such that f

yY

y
, f

<
D
and rk
D
(f

) = .
Dene g

yY

y
by g(y) = f

(y) : C and f

(y) <
t
. By clause (d) we
have [y Y g(y) <
y
], so g <
D
, hence rk
D
( g) < rk
D
() but by the choice
of g we have C f

D
g hence C = rk
D
(f

) rk
D
(g) hence
= sup(C) rk
D
(g), contradiction.
5.16
{r12}
Observation 5.17. 1) Assume ( , D) satises
(a) D a lter on Y and =
t
: t Y ) and each
t
is a limit ordinal
(b)

T = T

: < ) exemplify = ps-tcf( , <

D
) so we demand just
= sup < : T

,=
(c) T

= f

tY

t
: for some g T

we have f = g mod D.
Then: T

: < ) exemplify = ps-tcf( , <

D
) that is
()

<
T

is conal in ( , <
D
)
() for every
1
<
2
< and f
1
T

1
and f
2
T

2
we have f
1
f
2
.
2) Similarly, if D,

T satises clauses (a),(b) above and D is
1
-complete and =
cf() >
0
then we can correct

T to make it
0
-continuous that is T

: < )
dened in (c)
1
+ (c)
2
below satises () + () above and () below and so is
0
-
continuous, (see below) where
(c)
1
if < and cf() ,=
0
then T

= T

(c)
2
if < and cf() =
0
then T

= supf
n
: n < ): for some increasing
sequence
n
: n < ) with limit we have n < f
n
T

n
, see below
() if < and cf() =
0
and f
1
, f
2
T

then f
1
= f
2
mod D.
3) This applies to any increasing sequence T

: < ), T

Y
Ord, a limit
ordinal.
Proof. Straightforward.
5.17
{r13}
Denition 5.18. 0) If f
n

Y
Ord for n < , then supf
n
: n < ) is dened as
the function f with domain Y such that f(t) = f
n
(t) : n < .
1) We say

T = T

: < ) exemplifying = ps-tcf( , <

D
) is weakly
0
-
continuous when:
if < , cf() =
0
and f T
n
then for some sequence (
n
, f
n
) : n < ) we
have =
n
: n < ,
n
<
n+1
< , f
n
T
n
and f = supf
n
: n < ); so if
D is
1
-complete then f/D : f T

is a singleton.
2) We say it is
0
-continuous if we can replace the last then by i.
{r14}
Theorem 5.19. The Canonical Filter Theorem Assume DC and AC
P(Y )
.
Assume =
t
: t Y )
Y
Ord and t Y cf(
t
) (T(Y )) and
ps-pcf
1-comp
( ) hence is a regular cardinal. Then there is D = D

, an
1
-complete
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 29
lter on Y such that = ps-tcf( /D) and D D

for any other such D

Fil
1
1
(D).
{r14b}
Remark 5.20. 1) By 5.9 there are some such .
2) We work to use just AC
P(Y )
and not more.
3) If >
0
we can replace
1
-complete by -complete.
Proof. Let

1
(a) D = D : D is an
1
-complete lters on Y such that ( /D) has
pseudo true conality ,
(b) D

= D : D D.
Now obviously
(c) D

is an
1
-complete lter on Y .
For A Y let D
A
= D D : A / D and let T

= A Y : D
A
,= .
As AC
P(Y )
we can nd D
A
: A T

) such that D
A
D
A
for A T

. Let
D

= D
A
: A T

, clearly

2
D

= D : D D

and D

D is non-empty.
As AC
P
holds clearly
()
0
we can choose

T
A
: A T

) such that

T
A
exemplies D
A
D as in
5.17(1),(2), so in particular is
0
-continuous.
For each < let T

= T
A

: A T

, now
()
1
T

.
[Why? As by 5.17(1)(c) we have T
A

for each A T

.]
()
2
if
1
<
2
< , f
1
T

1
and f
2
T

2
then f
1
< f
2
mod D

.
[Why? Note that A T

f
1
<
DA
f
2
by the choice of T

: < ), hence the

set t Y : f
1
(t) < f
2
(t) belongs to D
A
for every A T

hence by
2
it belongs
to D

which means that f

1
<
D
f
2
as required.]
()
3
if f then for some
f
< we have f

: [
f
, ) f < f

mod D

.
[Why? For each A T

there are , g such that < , g T

A

and f < g mod

D hence

[ + 1, ) f

T
A

f < g < f

mod D
A
. Let
A
be the minimal
such ordinal < . As cf() (T(Y )) (T

), clearly

= sup
A
+1 : A T

is < . So A T

g T

: [

, )) f <
DA
g. By
2
the ordinal

is
as required on
f
.]
Moreover
()
4
there is a function f
f
in ()
3
.
[Why? As we can (and will) choose
f
as the minimal such that ...]
()
5
for every

< there is (

, ) such that T

,= .
[Why? We choose by induction on n, a sequence

n
=
n,A
: A T

) and a
sequence

f
n
= f
n,A
: A T

) and a function f
n
such that
()
n
< and m < n
m
<
n
(
9
3
8
)


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







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


30 SAHARON SHELAH
()
0
=

and for n > 0 we let

n
= sup
m,A
: m < n, A T

()
n,A
(
n
, ) is minimal such that there is f
n,A
T
A
n,A
satisfying n =
m + 1 f
m
< f
n,A
mod D
A
() f
n,A
: A T

) is a sequence such that each f

n,A
are as in clause ()
() f
n
is dened by f
n
(t) = supf
m,A
(t) + 1 : A T

and m < n.
[Why can we carry the induction? Arriving to n rst, f
n
is well dened
by clause () as cf(
t
) (T

) for t Y . Second by clause () and the choice of

F
A

: < ) : A T

in ()
0
the sequence
n,A
: A T

) is well dened. Third

by clause () we can choose f
m,A
: A T

) because we have AC
P
. Fourth,
n
is
well dened by clause () as cf() (T

).
Lastly, the inductive construction is possibly by DC.]
Let

=
n
: n < and f = supf
n
: n < ). Easily f T
A

: A T

as each T
A

: < ) is
0
-continuous.]
()
6
if f then for some < and f

we have f < f

mod D

.
[Why? By ()
3
+ ()
5
.]
So we are done.
5.19
(
9
3
8
)


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







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


PCF ARITHMETIC WITHOUT AND WITH CHOICE SH938 31
6. Private Appendix
Moved 10.1.05 from after 0.5, pg.2:
Remark 6.1. We cannot weaken the assumption (a) in 2.3 (which says D
2
does
2-commute with D
2
) too much because even assuming ZFC, if (a) below(?) for
some f, g,

f, g we have of 2.3 but rk
D1
(f) > rk
D2
(g).
Moved 10.1.07 from after 3.12, pgs.20,21:
{m4.17}
Claim 6.2. Like 3.12 assuming
6
AC
cf()
we can weaken ()
2
(b) to
()
2
(b)

if (d, , , f, j
0
) satises of 3.6 and i(f(t)) : t Y
d
) is constant, and
supcf(f(t)) : t Y
d
< , then :
() cf(

tY
d
f(t)/D
d
) < or
() for any j < cf(), if e D
j
and g
Y [e]
[, ] then for every p-
rectangle (d, e, f, g,

f, g) as we have rk
d
(f) rk
e
(g).
Proof. Let (d, , , f, j
0
) satises of 3.6 be given and we should ne j < cf()
such that for on pair (e, g) does clause there hold.
Without loss of generality the set A := s Y
d
: cf(f(s)) cf() is or
is Y
d
. If the former holds, by AC
cf()
as cf() = , we can nd

h = h
i
: i <
cf()) satisfying h
i


sY
d
f(s) such that suprk
d
(h
i
) : i < cf() = and let
h

sY
d
f(s) be dened by h

(s) = f
i
(s) : i < cf(), well dened as A = ,
contradiction. So Y
d
= and without loss of generality s Y
d
f(s) is a
limit ordinal (as without loss of generalityA

= s Y
d
: f(s) is a limit ordinal
is or is Y
d
, and the former is impossible). Similarly without loss of generality
i(f(t)) : t Y
d
) is constant.
As AC
Y
d
holds there is a sequence C
s
: s Y
d
) such that C
s
is an unbounded
subset of f(s) of order type cf(f(s)). We continue as in the rst paragraph of
3.12s proof but demand g
t


sY
d
C
s
. 20.10.07 - not clear why possible and why
suce.
6.2
Moved 10.1.07 from after 5.19,pg.30:
{r16}
Denition 6.3. For
Y
Ord let J
1-comp
<
( ) = X Y : ps-pcf
1-com
( X)
and J
1-comp

is J
2-comp
<
+
.
{r17}
Remark 6.4. In 6.3, see Denition 5.6(3).
References
[Sh:g] Saharon Shelah, Cardinal Arithmetic, Oxford Logic Guides, vol. 29, Oxford University
Press, 1994.
[Sh:71] , A note on cardinal exponentiation, The Journal of Symbolic Logic 45 (1980),
5666.
[Sh:460] , The Generalized Continuum Hypothesis revisited, Israel Journal of Mathematics
116 (2000), 285321, math.LO/9809200.
[Sh:497] , Set Theory without choice: not everything on conality is possible, Archive
for Mathematical Logic 36 (1997), 81125, A special volume dedicated to Prof. Azriel Levy.
math.LO/9512227.
6
hard not to have it by ()
4
of 3.12
(
9
3
8
)


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







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


32 SAHARON SHELAH
[Sh:513] , PCF and innite free subsets in an algebra, Archive for Mathematical Logic 41
(2002), 321359, math.LO/9807177.
[Sh:829] , More on the Revised GCH and the Black Box, Annals of Pure and Applied Logic
140 (2006), 133160, math.LO/0406482.
[Sh:835] , PCF without choice, Archive for Mathematical Logic submitted,
math.LO/0510229.
[Sh:908] , On long increasing chains modulo at ideals, Mathematical Logic Quarterly 56
(2010), 397399, 0705.4130.
[LrSh:925] Paul Larson and Saharon Shelah, Splitting stationary sets from weak forms of Choice,
Mathematical Logic Quarterly 55 (2009), 299306.
[Sh:938] Saharon Shelah, PCF arithmetic without and with choice, Israel Journal of Mathematics
accepted, 0905.3021.
[Sh:F955] , PCF with little choice.
[Sh:955] , Pseudo pcf, preprint.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at