(

8
3
5
)


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







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


PCF WITHOUT CHOICE
SH835
SAHARON SHELAH
Abstract. We mainly investigate models of set theory with restricted choice,
e.g., ZF + DC + the family of countable subsets of is well ordered for every
(really local version for a given ). We think that in this frame much of pcf
theory, (and combinatorial set theory in general) can be generalized. We prove
here that there is a class of regular cardinals, every large enough successor of
singular is not measurable and we can prove cardinal inequalities.
Anotated Content
0 Introduction
(0A) Background, aims and results
(0B) Preliminaries
[Include quoting [Sh:497], (0.1,0.2); hrtg(Y ), wlor(Y ), (0.6); dening rk
D
(f),
(0.7 + 0.8) on J[f, D], (0.9, 0.7, 0.11); H
<,
(Y ), (0.13 and observation
0.14); and on closure operations (0.15).]
1 Representing

[We dene Fil

and prove a representation of

. Essentially under rea-
sonable choice the set

is the union of few well ordered set, i.e., their
number depends on only. We end by a claim on a.]
2 No decreasing sequence of subalgebras
[As suggested in the title we weaken the axioms. We deal with

with
+
not measurable, existence of ladder

C witnessing conality and prove that
many
+
are regular (2.13).]
3 Concluding remarks
Date: November 8, 2010.
This research was supported by the United States-Israel Binational Science Foundation. I
would like to thank Alice Leonhardt for the beautiful typing. First Typed - 2004/Jan/20.
1
(
8
3
5
)


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







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


2 SAHARON SHELAH
0. Introduction
0(A). Background, aims and results.
The thesis of [Sh:497] was that pcf theory without full choice exists. Two theo-
rems supporting this thesis were proved. The rst ([Sh:497, 4.6,pg.117]).
{0.1}
Theorem 0.1. [DC] If H () is well ordered, strong limit singular of uncountable
conality then
+
is regular not measurable (and 2

is an , i.e. P(M) can be

well ordered and no (, 2

] is measurable).
Note that before this Apter and Magidor [AM95] had proved the consistency
of H() well ordered, =

, ( < )DC

and
+
is measurable so 0.1 says
that this consistency result cannot be fully lifted to uncountable conalities. Gen-
erally without full choice, a successor cardinal being not measurable is worthwhile
information.
A second theorem ([Sh:497, 5]) was
{0.2}
Theorem 0.2. Assume
(a) DC + AC

+ regular uncountable
(b)
i
: i < ) is increasing continuous with limit , > , H () well ordered,
strong limit, (we need just a somewhat weaker version, the so-called i <
Tw
D
(
i
) < ).
Then , we cannot have two regular cardinals such that for some stationary S ,
the sequence cf(
+
i
) : i S) is constantly .
A dream was to prove that there is a class of regular cardinals from a restricted
version of choice (see more [Sh:497] and a little more in [Sh:E37]).
Our original aim here is to improve those theorems. As for 0.1 we replace H ()
well ordered by []
0
is well ordered and then by weaker statements.
We know that if, e.g., 0
#
or there is no inner model with a measurable then
though 2

: regular) is quite arbitrary, the size of []

, >> is strictly con-

trolled (by Easton forcing [Eas70], and Jensen and Dodd [DJ82]). It seemed that
the situation here is parallel in some sense; under the restricted choice we assume,
we cannot say much on the cardinality of P() but can say something on the
cardinality of []

for >> .
In the proofs we fulll a promise from [Sh:589, 5] about using J[f, D] from
Denition 0.10 instead of the nice lters used in [Sh:497] and, to some extent, in
early versions of this work. This work is continued in Larson-Shelah [LrSh:925]
and will be continued in [Sh:F1078]. On a dierent line with weak choice (say
DC
0
+ AC

, xed): see [Sh:938], [Sh:955] and [Sh:F1039]. The present work

ts the thesies of [Sh:g] which in particular says: it is better to look at
0
: a
cardinality) then at 2

: a cardinal). Here instead well ordering P() we well

order []
0
, this is enough for much.
A simply stated conclusion is
{0n.2}
Conclusion 0.3. [DC] Assume every []
0
is well ordered for every .
1) If 2
2

is well ordered then for every , []

is well ordered.
2) For any set Y , there is a derived set Y

so called Fil
4
1
(Y ) of power near
P(P(Y )) such that
Levy(0,Y)
for every ,
Y
is well ordered.
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 3
{0n.9}
Thesis 0.4. 1) If V [= ZF + DC and every []
0
is well orderable then V
looks like the result of starting with a model of ZFC and using
1
-complete forcing
notions like Easton forcing, Levy collapsed, and more generally, iterating of -
complete forcing for >
0
.
2) This approach is dual to investigating L[R] - here we assume -sequences are
understood (or weaker versions) and we try to understand V (over this), there over
the reals everything is understood.
We thank for attention and comments the audience in the advanced seminar in
Rutgers 10/2004 (particularly Arthur Apter) and advanced course in logic in the
Hebrew University 4,5/2005 and to Paul Larson for many corrections.
0(B). Preliminaries.
{0n.2.7}
Convention 0.5. We assume just V [= ZF if not said otherwise.
{0.3}
Denition 0.6. 1) hrtg(A) = Min: there is no function from A onto .
2) wlor(A) = Min: there is no one-to-one function from into A or = 0A ,=
so wlor(A) hrtg(A).
{0.A}
Denition 0.7. 1) For D an
1
-complete 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) = .
2) We can replace D by the dual ideal. If f
Z
Ord and Z D then we let
rk
D
(f) = rk(f 0
Y \Z
).
Galvin-Hajnal [GH75] use the rank for the club lter on
1
. This was continued
in [Sh:71] where varying D was extensively used.
{0.B}
Claim 0.8. [DC] In 0.7, rk
D
(f) is always an ordinal and if rk
D
(f) then for
some g

yY
(f(y) + 1) we have = rk
D
(g), (if < rk
D
(f) we can add g <
D
f;
if rk
D
(f) < then DC is not necessary; if rk
D
(f) = this is trivial, as we can
choose g = f).
{0.B1}
Claim 0.9. 1) [DC] 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
+
, ([Sh:71]).
2) [DC + AC

] If D is a -complete lter on Y, a cardinal >

0
and f
Y
Ord
and Y = Y

: <

< then rk
D
(f) = Minrk
D+Y
(f) : <

and
Y

D
+
.
{0.C}
Denition 0.10. For Y, D, f in 0.7 let J[f, D] =: Z Y : Y Z D or
rk(f)
D+(Y \Z)
> rk
D
(f).
{0.D}
Claim 0.11. [DC+AC
<
] Assume D is a -complete lter on Y, >
0
.
1) If f
Y
Ord then J[f, D] is a -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.
Proof. Straight or see [Sh:589, 5] and the reference there to [Sh:497] (and [Sh:71]).

0.11
(
8
3
5
)


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







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


4 SAHARON SHELAH
{0n.D}
Denition 0.12. 1) [Y [ qu(Z) or Y
qu
Z or [Y [
qu
[Z[ or [Y [
qu
Z or
Y
qu
[Z[ means that Y = or there is a function from Z (equivalently a subset
of Z) onto Y .
2) reg() = Min : is a regular cardinal.
{0.E}
Denition 0.13. For a set Y , cardinal and ordinal we dene H
<,
(Y ) by
induction on : if = 0, H
<,
(Y ) = Y , if = +1 then H
<,
(Y ) = H
<,
(Y )
u : u H
<,
(Y ) and [u[ < and if is a limit ordinal then H
<,
(Y ) =
H
<,
(Y ) : < .
{x.5}
Observation 0.14. 1) If is the disjoint union of W
z
: z Z) and z Z
[W
z
[ < and wlor(Z) then = supotp(W
z
) : z Z hence cf() < hrtg(Z).
2) If = W
z
: z Z and wlor(P(Z)) then supotp(W
z
) : z Z = .
3) If = W
z
: z Z and [Z[ < then = supotp(W
z
) : z Z.
Proof. 1) Let Z
1
= z Z : W
z
,= , so the mapping z Min(W
z
) exemplies
that Z
1
is well ordered hence by the denition of wlor(Z
1
) the power [Z
1
[ is an
aleph < wlor(Z
1
) wlor(Z) and by assumption wlor(Z) . Now if the desirable
conclusion fails then

= sup(otp(W
z
) : z Z
1
[Z
1
[) is an ordinal < ,
so we can nd a sequence u

: <

) such that otp(u

, u

and
= u

: <

, so

< [

[, easy contradiction.
2) For x Z let W

x
= < : (z Z)( W
z
z x) hence is the disjoint
union of W

x
: x P(Z). So the result follows by (1).
3) So let <

be a well ordering of Z and let W

z
= W
z
: if y <

z then / W
y
,
so W

z
: z Z) is a well dened sequence of pairwise disjoint sets with union equal
to W
z
: z Z = and otp(W

z
) otp(W
z
). Hence if [W
z
[ = for some z Z
the desirable conclusion is obvious, otherwise the result follows by part (1).
0.14
{0.F}
Denition 0.15. 1) We say that c is a very weak closure operation on of
character (, ) when :
(a) c is a function from P() to P()
(b) u []

[c(u)[
(c) u u 0 c(u), the 0 for technical reasons.
1A) We say that c is a weak closure
1
operation on of character (, ) when
(a),(b),(c) above and
(d) u v u c(u) c(v)
(e) c(u) = c(v) : v u, [v[ .
1B) Let ... character (< , ) or (, < ), or (< , < ) has the obvious meaning
but if is an ordinal not a cardinal, then < means of order type < ; similarly
for < . Let ... character (, Y ) means character (<
+
, < hrtg(Y ))
1C) We omit the weak if in addition
(f) c(u) = c(c(u)) for u .
1
so by actually only c[]

count
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 5
2) is f-inaccessible when Dom(f) f() < .
3) c : P() P() is well founded when for no sequence U
n
: n < ) of subsets
of do we have c(U
n+1
) U
n
for n < .
4) For c a partial function from P() to P() (for simplicity assume = u :
u Dom(c)) let c
1
,<
a function from P() to P() be dened by induction
on the ordinal as follows:
(a) c
1
0,<
(u) = u
(b) c
1
+1,<
(u) = 0c
1
,<
(u)

c(v) : v c
1
,<
(u) and v Dom(c), [v[ <

(c) for limit let c

1
,<
(u) = c
1
,<
(u) : < .
4A) Instead < we may write .
5) For any function F : []
0
and countable u we dene c
2

(u, F) by
induction on
1
(a) c
2
0
(u, F) = u 0
(b) c
2
+1
(u, F) = c
2

(u, F) F(c
2

(u, F))
(c) c
2

(u, F) = c
2

(u, F) : < when

1
is a limit ordinal.
6) For countable u and F as in part (5) let c
3
F
(u) = c
3
(u, F) := c
2
1
(u, F) and
for any u let c
4
F
(u) := u

c
3
F
(v) : v Dom(F) is countable.
7) For a cardinal we say that c : P() P() is -well founded when for
no -decreasing sequence U

: < ) of subsets of do we have < <

c(U

) U

.
8) If F : []

and u then we let c

F
(u) = c
1
F
(u) be the minimal subset
v of such that w [v]

F(w) v and u v (exists).

{0n.F}
Observation 0.16. For F : []
0
, the operation u c
3
F
(u) is a very weak
closure operation of character (
0
,
1
).
{0.G}
Remark 0.17. So for any very weak closure operation,
0
-well founded is a stronger
property than well founded, but if u c(c(u)) = c(u) which is reasonable,
they are equivalent.
{0.H}
Observation 0.18. []

is well ordered i

is well ordered when .
Proof. Use a pairing function on for showing [

[ []

, so holds. If

is well
ordered by <

map u []

to the <

-rst f

satisfying Rang(f) = u.
0.18
(
8
3
5
)


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







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


6 SAHARON SHELAH
1. Representing

Here we give a simple case to illustrate what we do (see latter on improvements

in the hypothesis and conclusion). Specically, if Y is uncountable and []
0
is well
ordered, then the set
Y
can be an analysis modulo countable union over few (i.e.,
their number depends on Y but not on ) well ordered sets.
{r.1}
Denition 1.1. 1)
(a) Fil
1
(Y ) = Fil
1
1
(Y ) = D : D an
1
-complete lter on Y , so Y is dened
from D as X : X D
(b) Fil
2
1
(Y ) = (D
1
, D
2
) : D
1
D
2
are
1
-complete lters on Y , ( / D
2
, of
course); in this context Z

D means Z D
2
(c) Fil
3
1
(Y, ) = (D
1
, D
2
, h) : (D
1
, D
2
) Fil
2
1
(Y ) and h : Y for some
< , if we omit we mean = hrtg(Y )
(d) Fil
4
1
(Y, ) = (D
1
, D
2
, h, Z) : (D
1
, D
2
, h) Fil
3
1
(Y, ), Z D
2
; omit-
ting means as above.
2) For y Fil
4
1
(Y, ) let Y = Y
y
, y = (D
y
1
, D
y
2
, h
y
, Z
y
) = (D
1
[y], D
2
[y], h[y], Z[y]);
similarly for the others and let D
y
= D[y] be D
y
1
+Z
y
.
3) We can replace
1
by any >
1
(the results can be generalized easily assuming
DC + AC
<
, used in 2).
{r.2}
Theorem 1.2. [DC
0
] Assume
(a) []
0
is well ordered.
Then we can nd a sequence F
y
: y Fil
4
1
(Y )) satisfying
() F
y

Z[y]

() F
y
is a well ordered set by f
1
<
y
f
2
rk
D[y]
(f
1
) < rk
D[y]
(f
2
) so f
rk
D[y]
(f) is a one-to-one mapping from F
y
into the ordinals
() if f
Y
then we can nd a sequence y
n
: n < ) with y
n
Fil
4
1
(Y )
such that n < f Z
yn
F
yn
and Z
yn
: n < = Y .
An immediate consequence of 1.2 is
{r.1x}
Conclusion 1.3. 1) [DC +

is well-orderable for every ordinal ].
For any set Y and cardinal there is a sequence F
x
: x

(Fil
4
1
(Y ))) such
that
(a)
Y
= F
x
: x

(Fil
4
1
(Y ))
(b) F
x
is well orderable for each x

(Fil
4
1
(Y ))
(b)
+
moreover, uniformly, i.e., there is a sequence <
x
: x

(Fil
4
1
(Y )) such
that <
x
is a well order of F
x
(c) there is a function F with domain P(
Y
) such that: if S
Y
then
F(S) is a non-empty subset of S of cardinality
qu

(Fil
4
1
(Y ))) recalling
Denition 0.12. In fact, some sequence U

: < ()) is a partition of

Y
, we have F(S) = U
f(S)
where f(S) = Min : U

S ,= (and [U

[ <
hrtg(

(Fil
4
1
(Y )))).
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 7
2) [DC] If [()]
0
is well ordered where () = rk
D
(f) + 1 : f
Y
and
D Fil
1
1
(Y ) then
Y
satises the conclusion of part (1).
Remark 1.4. 1) So clause (c) of 1.3(1) is a weak form of choice.
Proof. Proof of 1.3 1) Let F
y
: y Fil
4
1
(Y )) be as in 1.2.
For each x

(Fil
4
1
(Y )) (so x = x
n
: n < )) let
F

x
= f : f is a function from Y to such that
n < f Z
xn
F
xn
and Y = Z
xn
: n < .
Now
()
1
Y
= F

x
: x

(Fil
4
1
(Y )).
[Why? By clause () of 1.2.]
Let () = rk
D
(f) + 1 : f
Y
and D Fil
1
1
(Y ). For x

(Fil
4
1
(Y )) we
dene the function G
x
: F

x


() by G
x
(f) = rk
D1[xn]
(f) : n < ).
Clearly
()
2
()

G = G
x
: x

(Fil
4
1
(Y ))) exists
() G
x
is a function from F

x
to

()
() G
x
is one to one.
Let <

be a well ordering of

() and for x

(Fil
4
1
(Y )) let <
x
be the following
two place relation on F

x
:
()
3
f
1
<
x
f
2
i G
x
(f
1
) <

G
x
(f
2
).
Obviously
()
4
() <
x
: x

(Fil
4
1
(Y ))) exists
() <
x
is a well ordering of F

x
.
By ()
1
+ ()
4
we have proved clauses (a),(b),(b)
+
of the conclusion. Now clause
(c) follows: for non-empty S
Y
, let f(S) be minotp(g : g <
y
f, <
y
) : y

(Fil
4
1
(Y )) and f F

y
S. Also for any ordinal let U
1

:= f: for some y

(Fil
4
1
(Y )) we have = otp(g : g <
y
f, <
y
) and U

= U
1

U
1

: < .
Lastly, we let F(S) = U
f(S)
S. Now check.
2) Similarly.
1.3
Proof. Proof of 1.2 First

1
there are a cardinal and a sequence u = u

: < ) listing []
0
.
[Why? by assumption (a).]
Second, we can deduce

2
there are
1
and a sequence u = u

: <
1
) such that:
(a) u

[]
0
(b) if u []
0
then for some nite w
1
, u u

: w
(c) u

is not included in u
0
. . . u
n1
when n < ,
0
, . . . ,
n1
< .
(
8
3
5
)


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







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


8 SAHARON SHELAH
[Why? Let u
0
be of the form u

: <

) such that (a) + (b) holds and g( u

0
)
is minimal; it is well dened and g( u
0
) by
1
. Let W = < g( u
0
) : u
0

u
0

: w when w is nite. Let

1
= [W[ and let f :
1
W be a one-to-
one onto, let u

= u
0
f()
so u

: <
1
) satises (a) +(b) and
1
= [W[ g( u
0
).
So by the choice of u
0
we have g( u
0
) =
1
. So we can choose f such that it is
increasing hence u is as required.]

3
we can dene n : []
0
and partial functions F

: []
0

1
for
< (so F

: < ) exists) as follows:

(a) u innite F
0
(u) = Min: for some nite w , u u

:
w mod nite
(b) u nite F
0
(u) undened
(c) F
+1
(u) := F
0
(u(u
F0(u)
. . . u
F

(u)
)) for < when F

(u) is dened
(d) n(u) := Min : F

(u) undened.
Then

4
(a) F
+1
(u) < F

(u) <
1
when they are well dened
(b) n(u) is a well dened natural number and u u
F

(u)
: < n(u)
is nite and k < n(u) (u u
F

(u)
: < k) u
F
k
(u)
is innite
(c) if u
1
, u
2
[]
0
, u
1
u
2
and u
2
u
1
is nite then F

(u
1
) = F

(u
2
)
for < n(u
1
) and n(u
1
) = n(u
2
)

5
dene F

: []
0
by F

(u) = Min(u
F

(u)
: < n(u) 0u) if well
dened, zero otherwise

6
if u []
0
then
() c
3
(u, F

) = c
3
F
(U) is F

(u) := u

u
F

(u)
: < n(u) 0
() c
3
F
(u) = c
2
(u)
(F) for some (u) <
1
() there is

F = F

: <
1
) such that: for every u []
0
, c
3
F
(u) =
F

(u) : < (u) and F

(u) = 0 if [(u),
1
)
() in fact F

(u) is the -th member of c

3
F
(u) if < (u).
[Why? Dene w

u
by induction on by w
0
u
= u, w
+1
u
= w

u
F

(w

u
) and for limit
, w

u
= w

u
: < . We can prove by induction on that w

u
F

(u) which is
countable. The partial function g with domain F

(u)u to Ord, g() = Min :

w
+1
u
is one to one onto an ordinal call it (), so w
()
u
F

(u) and if they

are not equal that F

(w
()
u
) F

(u)w
()
u
hence w
()
u
w
()+1
u
contradicting the
choice of (). So clause () holds. In fact, c
3
(u, F

) = w
()
u
and clause () holds.]

7
there is no sequence U
n
: n < ) such that
(a) U
n+1
U
n

(b) U
n
is closed under F

, i.e. u [U
n
]
0
F

(u) U
n
(c) U
n+1
,= U
n
.
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 9
[Why? Assume toward contradiction that U
n
: n < ) satises clauses (a),(b),(c).
Let
n
= Min(U
n
U
n+1
) for n < hence the sequence =
n
: n < )
is with no repetitions and let
m,
:= F

(
n
: n m) for m < and <
n
m
:= n(
n
: n [m, )). As is with no repetition, n
m
> 0 and by
4
(c)
clearly n
m
= n
0
for m < and
m,
=
m,0
for m < , < n
0
. So letting
v
m
= u
F

({n:n[m,)})
: < n
m
, it does not depend on m so v
m
= v
0
, and by
the choice of F

, as
n
: n [m, ) U
m
and U
m
is closed under F

clearly
v
m
U
m
. Together v
0
= v
m
U
m
so v
0
U
m
: m < . Also, by the denition
of the F

s,
n
: n < v
0
is nite so for some k < ,
m
: n [k, ) v
0
but
v
0
U
k+1
contradicting the choice of
k
.]
Moreover, recalling Denition 0.15(6):

7
there is no sequence U
n
: n < ) such that
(a) U
n+1
U
n

(b) c
4
F
(U
n+1
)U
n
,= .
[Why? As above but letting
n
= Min(U
n
c
3
F
(U
n
)), recalling Denition 0.16(6).]
Now we dene for (D
1
, D
2
, h, Z) Fil
4
1
(Y ) and ordinal the following, recall-
ing Denition 0.15(6) for clauses (e),(f):

8
F
(D1D2,h,Z),
=: f : (a) f is a function from Z to
(b) rk
D1+Z
(f 0
(Y \Z)
) =
(c) D
2
= Y X : X Y satises X = mod D
1
or X D
+
1
and rk
D1+X
(f 0
(Y \Z)
) >
that is rk
D1+X
(f) >
(d) Z D
2
, really follows
(e) if Z

Z Z

D
2
then
c
3
F
(Rang(f Z

)) = c
3
F
(Rang(f))
(f) y Z f(y) = the h(y)-th member of c
3
F
(Rang(f)).
So we have:

9
F
(D1,D2,h,Z),
has at most one member; called it f
(D1,D2,h,Z),
(when de-
ned; pedantically we should write f
(D1,D2,h,Z),c,
)

10
F
(D1,D2,h,Z)
=: F
(D1,D2,h,Z),
: an ordinal is a well ordered set.
[Why? Dene <
(D1,D2,h,Z)
by the s.]

11
if f : Y and Z Y then c
4
F
(Rang(f Z)) has cardinality < hrtg(Z).
[Why? By the denition of hrtg() we are done.]

12
if f : Z and Z Y then c
4
F
(Rang(f)) has cardinality <
hrtg([Z]
0
) or is nite.
[Why? If Rang(f) is countable more hold by 0.16. Otherwise, by
6
() we have
c
4
F
(Rang(f)) = Rang(f) F

(u) : u [Rang(f)]
0
and <
1
.
Let () be minimal such that Rang(f) () has order type
1
. Let h
1
, h
2
:

1

1
be such that h

() < max, 1 and for every

1
,
2
<
1
there is
[
1
+
2
+ 1,
1
) such that h

() =

for = 1, 2. Dene F : [Z]

0
as follows:
if u [Rang(f)]
0
, let

(u) = h

(otp(u ()) for = 1, 2 and F(u) = F

( u:
if < () then otp(u ) <
1
(u)).]
(
8
3
5
)


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







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


10 SAHARON SHELAH
Now

1
if u [Rang(f)]
0
then F(u) is F

(v) for some v [Z]

0
and <
1
.
[Why? As F(u) Rang(F

2(u)
[Rang(f)]
0
).]

2
F(u) : u [Rang(f)]
0
c
4
F
(Rang(f)).
[Why? By
1
recalling
6
.]

3
if u [Rang(f)]
0
and <
1
then F

(u) is F(u) for some v [Rang(f)]

0
.
[Why? Let
1
= otp(u ()),
2
= and <
1
such that h

() =

for
= 1, 2. Let v = u : Rang(f) () and sup(u ()) + 1 and
otp(Rang(f) (sup(u () + 1)) < (
1
)),
so F(u) = F

(u).]

4
in
2
we have equality.
Together c
4
F
(Rang(f)) = F(u) : u [Rang(f)]
0
Rang(f) so it is the union
of two sets; the rst of cardinality < hrtg([Z]
0
) and the second of cardinality
< hrtg[Z], so we are easily done.

13
if f : Y then for some y
n
Fil
4
1
(Y ) and
n
Ord for n < we
have f = f
yn,n
: n < .
[Why? Let
I
0
f
= Z Y : for some y Fil
4
1
(Y ) satisfying Z
y
= Z
and ordinal , f
y,
is well dened and equal to f Z
I
f
= Z Y : Z is included in a countable union of members of I
0
f
.
So it is enough to show that Y I
f
.
Toward contradiction assume not. Let D
1
= Y Z : Z I
f
, clearly it belongs
to Fil
1
(Y ), e.g. noting that I
f
. So () = rk
D1
(f) is well dened (by 0.8)
recalling that only DC
0
is needed.
Let
D
2
= X Y : X D
1
or rk(f)
D1+(Y \X)
(f) > ().
By 0.10 + 0.11 clearly D
2
is an
1
-complete lter on Y extending D
1
.
Now we try to choose Z
n
D
2
for n < such that Z
n+1
Z
n
and c
4
F
(Rang(f
Z
n+1
)) does not include Rang(f Z
n
).
For n = 0, Z
0
= Y is O.K.
By

7
we cannot have such -sequence Z
n
: n < ); so by DC for some (unique)
n = n(), Z
n
is dened but not Z
n+1
.
Let h : Z
n
hrtg([Y ]
0
)
1
be:
h(y) = otp(f(y) c
4
F
(Rang(f Z
n
))).
Now h is well dened by
12
. Easily
f Z
n
F
(D1+Zn,D2,h,Zn),()
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 11
hence Z
n
I
0
f
I
f
, contradiction to Z
n
D
2
, D
1
D
2
.
So we are done proving
13
.]
Now clause () of the conclusion holds by the denition of F
y
, clause () holds
by
10
recalling
8
,
9
and clause () holds by
12
.
1.2
{r.2y}
Remark 1.5. We can improve 1.2 in some way the choice of u.
Assume (a) + (b) where
(a) let u

: <

) be a sequence of members of []
0
such that (u
[]
0
)()(u u

innite), and <

otp(u

) = .
[Why? We dene F
0
(v) := Min <

: vu

innite. Let F

(u) = min(u
F0()
u).
Now we use only F

for < such that F

(u) is the -th member of u

F
or of
u
F0()
u; so (u) = .]
{1p.4}
Observation 1.6. 1) The power of Fil
4
1
(Y, ) is smaller or equal to the power
of the set (P(P(Y )))
2
P(Y )
|Y |
; if
0
[Y [ this is equal to the power of
P(P(Y ))
Y
.
2) The power of Fil
4
1
(Y ) is smaller or equal to the power of the set (P(P(Y )))
2

P(Y )
Y
: < hrtg([Y ]
0
).
3) In part (2), if
0
[Y [ this is equal to [P(P(Y ))
Y
: < hrtg([Y ]
0
);
also < hrtg([Y ]
0
) [P(P(Y ))
Y
[ = [P(P(Y ))[.
Remark 1.7. 1) As we are assuming DC, the case
0
[Y [ is Y nite, so degener-
ated; even [Y [ =
0
is [Fil
4
0
(Y )[
qu
[P(P(Y Y ))[.
Proof. 1) Reading the denition of Fil
4
1
(Y, ) clearly its power is the power of
P(P(Y )) P(P(Y )) P(Y )

. If
0
[Y [ then [P(P(Y )) P(Y )[
[P(P(Y ))P(P(Y ))[ = 2
|P(Y ))
2
|P(Y )|
2
|P(Y )|+|P(Y )|
= 2
|P(Y )|
= [P(P(Y ))[
[P(P(Y )) P(Y )

[ as P(Y ) +P(Y ) = 2
|Y |
2 = 2
|Y |+1
= 2
|Y |
.
2) Read the denitions.
3) If < hrtg([Y ]
0
) then let f be a function from [Y ]
0
onto and for < let
A
f,
= u [Y ]
0
: f(u) < . So A
f,
is a one-to-one function from onto
A
f,
: < P(P(Y )) so [
Y
[ P(P(Y )), P(P(Y ))[
Y
[ P(P(Y ))
P(P(Y )) 2
|P(Y )|
= 2
|P(Y )|
= 2
|P(Y )|+|P)Y )|
= 2
|P(Y )|
. Better, for f a function
from [Y ]
0
onto < (Y ) let A
f
= (y
1
, y
2
) f(y
1
) < f(y
2
) Y
1
Y . Dene
F : P(Y Y ) hrtg(Y ) by F(A) = if A = A
f
and f, are as above, and
F(A) = 0 otherwise.
So [P(P(u))

Y
: < hrtg([Y ]
0
)[
qu
[P(P(Y )) P(P(Y Y ))) =
[P(P(Y ) Y )[. By part (2) we easily get [Fil
4
1
(Y )[
qu
P(P(Y Y )).
1.6
{6.1}
Claim 1.8. [DC
0
] Assume
(a) a is a countable set of limit ordinals
(b) <

is a well ordering of a
(c) a cf() hrtg(P()) or just a/[a]
<0
is hrtg(P())-directed.
Then we can dene (

J,

b,

f ) such that
() (i)

J = J
i
: i i())
(ii) J
i
is an ideal on a
(
8
3
5
)


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







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


12 SAHARON SHELAH
(iii) J
i
is increasing continuous with i, J
0
= , J
i()
= P(a)
(iv)

b = b
i
: i < i()), b
i
a and J
i+1
= J
i
+ b
i
(v) so J
i
is the ideal on a generated by b
j
: j < i
() (i)

f =

f
i
: i < i())
(ii)

f
i
= f
i

: <
i
)
(iii) f
i

a is <
Ji
-increasing with <
i
(iv) f
i

: <
i
is conal in (

a, <
Ji+(a\bi)
)
() (i) cf(

a)

i<i()

i
(ii) for every f a for some n and nite set (i

) : < n such
that i

< i(),

<
i

we have
f < max
<n
f
i

, i.e., ( a)( < n)[f() < f

i

()].
Remark 1.9. Note that no harm in having more than one occurance of a. See
more in [Sh:F1078], e.g. on uncountable a.
Proof. Note that clause () follows from () + ().
[Why? Easily ()(ii) ()(i). Now let g a and let I
g
= b: we can nd n <
and i

< i() and

<
i

for < n such that b ( < n)(g() < f

i

()).
Easily I
g
is an ideal on a; note that if a I
g
we are done. So assume a / I
g
.
Note that J
g
J
i()
hence j
g
= mini i(): some c P(a)I
g
belongs to J
i

is well dened (as a P(a)I

g
a J
i()
). As J
0
= and J
i
: i i()) is
-increasing continuous, necessairly j
g
is a successor ordinal say j
g
= i
g
+ 1 and
let c P(a)I
g
belongs to J
i
. By clause ()(iv) we can nish easily.]
By induction on i we try to choose (

J
i
,

b
i
,

f
i
) where

J
i
= J
j
: j i),

b
i
= b
i
j
:
j < i),

f
i
=

f
j
: j < i) which satises the relevant parts of the conclusion and do
it uniformly from (a, <

). Once we arrive to i such that J

i
= P(a) we are done.
For i = 0 recalling J
0
= there is no problem.
For i limit recalling that J
i
= J
j
: j < i there is no problem and note that if
j < i a / J
j
then a / J
i
.
So assume that (

J
i
,

b
i
,

f
i
) is well dened and we shall dene for i + 1.
We try to choose g
i,
= g
i,

: <
i,
) and b
i,
by induction on <
1
and for
each we try to choose g
i,

a by induction on (in fact < hrtg(a) suce,

we shall get stuck earlier) such that:

i,
(a) if < then g
i,

<
Ji
g
i,

(b) if < and <

i,
then g
i,

g
i,

(c) if cf() =
1
then g
i,

is dened by
a g
i,

() = Min

C
g
i,

() : C is a club of
(d) if cf() ,=
1
, ,= 0 then g
i,

is the <

-rst g a satisfying
clauses (a) + (b) + (c)
(e) if we have g
i,

: < ), cf() >

1
, moreover
cf() mincf() : a and there is no g as required in
clause (d) then
i,
=
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 13
(f) if = 0, then g
i,

is the <

-rst g a such that

< <
i,
g
i,

g
and = + 1 ( <
i,
)[(g
Ji
g
i,

)]
(g) J
i
is the ideal on P(a) generated by b
j
: j < i
(h) b
i,
(J
i
)
+
so b
i,
a
(i) g
i,
is increasing and conal in ((a), <
Ji+(a\bi,)
)
(j) b
i,
is such that under clauses (h) + (i) the set otp(a ) : b
i,

is <

-minimal
(k) if < i then g
i,
0
is a <
Ji+b
i,
-upper bound of g
i,
,
hence b
i,
b
i,
mod J
i
.
The rest should be clear.
We now give details on some points
()
1
Clause (c) is O.K.
[Why? We already have g
i,

: < ) and g
i,

: <
i,
, < ), and we dene
g
i,

as there. Now g
i,

() is well dened as the Min is taken on a non-empty set

as we are assuming cf() =
1
. The value is < because for some club C of ,
otp(C) =
1
, so g
i,

() g
i,

() : C but this set is while cf() >

1
by clause (c) of the assumption. By AC
0
we can nd a sequence C

: a) such
that: C

is a club of of order type

1
such that g
i,

() = g
i,

() : C

hence for every club C of included in C

we have g
i,

() = g
i,

() : C

.
Now a g
1,

() =

C
g
1,

() when C := C

: a, because C too
is a club of recalling a is countable. So if < then for some we have
< C hence the set c := a : g
i,

() g
i,

() belongs to J
i
and
ab g
i,

() < g
i,

() g
i,

(), so indeed g
i,

<
Ji
g
i,

.
Lastly, why < g
i,

g
i,

? As we can nd a club C of which is as

above for both g
i,

and g
i,

and recall that clause (b) of

i,
holds for every C.
Together g
i,

is as required.]
()
2
cf(
i,
) >
1
and even cf(
,t
) mincf() : a.
[Why? We have to prove that arriving to , if cf() < mincf() : a then
we can choose g
i,

as required. The cases cf() =

1
, = 0 are covered, otherwise
let u be unbounded of order type cf(), and dene a function g from a to the
ordinals by g() = supg
i,

() : u g
i,

() : < ). This is a subset of of

cardinality < [a[ + cf() which is < = cf() hence g a, easily is as required,
i.e. satises clauses (a) + (b) holds, and the <

-rst such g is g
i,

.]
Note that clause (e) of
1,
follows.
()
3
if < then
i,

i,
.
[Why? Otherwise g
i,

i,
contradict clause (e) of
i,
.]
Case 1: We succeed to carry the induction, i.e. choose g
i,
for every <
1
.
Let () <
1
be minimal such that [(),
1
)
i,
=
i,()
, exists by
()
3
and let

:=
i,()
. For each a the sequence g
i,
0
() : <
1
) is
(
8
3
5
)


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







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


14 SAHARON SHELAH
non-decreasing hence C

:= <
1
: ()( <
1
g
i,
0
() = g
i,
0
()) or
( < )(g
i,
0
() < g
i,
0
() is a club of
1
(and even an end-segment), hence also
C = C

: a is a club of
1
.
Let c = : for every < from C, g
i,
0
() < g
i,
0
() we get contradiction to the
demand on b
i,
, b
i,
.
Case 2: We are stuck in <
1
.
For = 0 there is no problem to dene g
i,

by induction on till we are stuck,

say in , necessarily is of large enough conality, and so g
i,
is well dened. We
then prove b
i,
exists again using <

.
For limit we can also choose g

.
For = + 1, if g
i,
0
as required cannot be chosen, we can prove that g
i,
, b
i,
exists as above.
1.8
{6.2p}
Remark 1.10. From 1.8 we can deduce bounds on hrtg(
Y
(

)) when <
1
and
more like the one on
0

(better the bound on pp(

)).
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 15
2. No decreasing sequence of subalgebras
We concentrate on weaker axioms. We consider Theorem 1.2 under weaker
assumptions than []
0
is well orderable. We are also interested in replacing
by in no decreasing -sequence of c-closed sets, but the reader may consider
=
0
only. Note that for the full version, Ax
4

, i.e., []

is well orderable, case

=
0
is implied by the >
0
version and suces for the results. But for other
versions the axioms for dierent s seems incomparable.
Note that if we add many Cohens (not well ordering them) then Ax
4

fails below
even for =
0
, whereas the other axioms are not aected. But forcing by
1
-
complete forcing notions preserve Ax
4
.
{22.0}
Hypothesis 2.1. DC

and let () = +
1
. We x a regular cardinal (below
we can use DC
0
+ AC

only).
{22.1}
Denition 2.2. Below we should, e.g. write Ax
,
instead of Ax

and assume
> > . If = we may omit it.
1) Ax
1
,,
means that there is a weak closure operation on of character (, ),
see Denition 0.15, such that there is no -decreasing -sequence U

: < ) of
subsets of with < c(U
+1
) U

. We may here and below replace by

< ; similarly for ; let < [Y [
+
means [Y [.
2) Let Ax
0
,,
mean there is c : []

[]
<
of members of []

such that
u 0 c(u) and there is no -decreasing sequence U

: < ) of members of
[]

such that < c(U

+1
) U

.
Writing Y instead of means c : []
<hrtg(Y )
[]

. Let c
[]
: P() P()
be c
1
,<reg(
+
)
as dened in 0.15(4) recalling reg() = Min : a regular cardinal
.
3) Ax
2

means that there is A []

which is well orderable and for every u []

for some v A, u v has power .

4) Ax
3

means that cf([]

, ) is below some cardinal, i.e., some conal A []

(under ) is well orderable.

5) Ax
4

means that []

is well orderable.
6) Above omitting (or writing ) means for every , omitting we mean
< hrtg(P()).
7) Lastly, let Ax

= Ax

for = 1, 2, 3.
So easily (or we have shown in the proof of 1.2):
{r.2b}
Claim 2.3. 1) Ax
4

implies Ax
3

, Ax
3

implies Ax
2

, Ax
2

implies Ax
1

and Ax
1

implies Ax
0

. Similarly for Ax

,<,
.
2) In Denition 2.2(2), the last demand, if c has monotonicity, then only c []

is relevant, in fact, an equivalent demand is that if

: < )

then for some
,

: (, ).
3) If Ax
0
,<1,<
and hrtg(Y ) and
2

2
= suphrtg(
1
[]

) : < hrtg(Y )
then Ax
0
,<2,<hrtg(Y )
.
Proof. 1) Easy; least obvious is Ax
2
,<,
Ax
1
,<,
which holds by 1.5.
2) First assume that we have a -decreasing sequence u

: < ) such that

< c(U
+1
) U

. Let

= min(U

c(U
+1
)) for < so clearly
2
can do somewhat better
(
8
3
5
)


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







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


16 SAHARON SHELAH

: < ) exist so by monotonicity c(

: [ + 1, ) c(U
+1
) hence

/ c(

: [ + 1, ).
Second, assume that

=

: < )

satises

/ c(

: [ + 1, )
for < . Now letting U

: < satises for < clearly

U

: < ) exists, is -decreasing and <

/ c(U

+1
)

. So we
have shown the equivalence.
3) Let c() witness Ax
0
,<1,<
. We dene c

: []
<hrtg(Y )
[]
<2
by c

(u) =
c(v) : v u has cardinality < . It is enough to note:
()
1
if u []
<hrtg(Y )
then c

(u) has cardinality <

2
:= hrtg(
1
[
1
]
<
).
[Why? Let C
u
= (v, ) : v u has cardinality < and < otp(c(v)) which
is <
1
. Clearly [c

(u)[ < hrtg(C

u
) and [C
u
[ = [
1
[otp(u)]
<
[, so ()
1
holds.
Note that if

<
+
1
we can replace the demand v [u]
<
[c(v)[ <
1
by
v [u]
<
otp[c(v)[ <

.]
()
2
If u

: < ) is -decreasing then u

(u
+1
) for some < .
[Why? If not by AC

we can choose a sequence

: < ) such that <

= min(u

(u
+1
)). Let u

: [, ), and note that as u

is -decreasing also c

(u

) is -decreasing. So by the choice of c() for some

,

: < [, ), but this set is c

(u
+1
) by the denition of c

(), so
we are done.]
2.3
{r.B.3}
Claim 2.4. Assume c witness Ax
0
,<,
so < and c : []

[]
<
and
recall c
1
,
: P() P() is from 2.2(2), 0.15(4).
1) c
1
1,
is a weak closure operation, it has character (

, ) whenever
and

= hrtg( P()), see Denition 0.15.

2) c
1
reg(
+
),
is a closure operation, has characteristic and it has character
(<

, ) when and

= hrtg(H
<
+( )).
Proof. 1) By its denition c
1
1,
is a weak closure operation.
Assume u , [u[ ; non-empty for simplicity. Clearly [[u[]

has the
same power as [u]
<
. Dene
3
the function G with domain [u]
<
as follows:
if < and v [u]

then G((, v)) is the -th member of c(v) if < otp(c(v))

and G((, v)) = min(u) otherwise.
So G is a function from [u]

onto c
1
1,
(u). This proves that c
1
1,
has
character (<

, ) as

= hrtg( P()).
2) If u

: reg(
+
)) is an increasing continuous sequence of sets then [u

+]

=
[u

: < reg(
+
) as reg(
+
) is regular (even of conality > suce) by its
denition, note reg(
+
) =
+
because as AC

holds as DC

holds.
Second, let u , [u[ and let u

= c
1
,
(u) for
+
; it is enough to show
that [u

+[ <

. The proof is similar to earlier one.

2.4
{r.4}
Denition/Claim 2.5. Let c exemplify Ax
0
,<,Y
and Y be an uncountable set
of cardinality ().
1) Let F
y
, F
y,
be as in the proof of Theorem 1.2 for y Fil
4
()
(Y, ) and ordinal
(they depend on and c but note that c determine ; so if we derive c by
3
clearly we can replace < by < for (,
+
)
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 17
Ax
4

then they depend indirectly on the well ordering of []

) so we may write
F
y,
= F
y
(, c), etc.
That is, fully
()
1
for y Fil
4
()
(Y, ) and ordinal let F
y,
be the set of f such that:
(a) f is a function from Z
y
to
(b) rk
D[y]
(f) = recalling that this means rk
D
y
1
+Z
y(f 0
Y \Z
y) by De-
nition 0.7(2)
(c) D
y
2
= D
y
1
Y A : A J[f, D
y
1
], see Denition 0.10
(d) Z
y
D
y
2
(e) if Z D
y
2
and Z Z
y
then c(f(y) : y Z) f(y) : y Z
y

(f) h
y
is a function with domain Z
y
such that y Z
d
h
h
(y) =
otp(f(y) c(f(z) : z Z
y
)
()
2
F
y
= F
y,
: an ordinal.
2) Let
y
=
y
(c) =
y
(, c) = : F
y,
,= and f
y,
is the function f F
y,
when
y
; it is well dened.
3) If D Fil
()
(Y ), rk
D
(f) = and f
Y
then
D
(, c) and f Z
y
= f
y,
for some y Fil
4
1
(Y ) where
D
(, c) :=
y
: y Fil
4
()
(Y ) and D
y
1
= D +Z
for some Z Y mod J[f, D].
4) If D Fil
()
(Y ), f
Y
, Z D
+
and rk
D+Z
(f) then for some g

yY
(f(y) + 1)
Y
( + 1) we have rk
D
(g) = hence
D
(, c).
5) So we should write F
y
[c],
y
[, c], f
y,
[c].
Proof. As in the proof of 1.2 recalling c exemplies Ax
0
,<,hrtg(Y )
holds, this
replaces the use of F

there; and see the proof of 2.11 below.

2.5
{y.21}
Claim 2.6. We have
2
is an ordinal and Ax
0
2,<2,Y
holds when,(note that
2
is
not much larger than
1
):
(a) Ax
0
1,<1,Y
so < hrtg(Y )
(b) c witnesses clause (a)
(c) D Fil
()
(Y )
(d)
2
= : f
y,
[c] is well dened for some y Fil
4
()
(Y,
1
) which satises
D
y
1
= D and necessarily Rang(f
y,
[c])
1

(e)
2
is dened as
2,3
where
() let
2,0
= hrtg(Y )
()
2,1
= sup
<2,0
hrtg( Fil
4
()
(Y,
1
))
()
2,2
= sup
<2,1
hrtg(
1
[]

)
()
2,3
= suphrtg(
Y
Fil
()
(Y )) : <
2,

(this is an overkill).
Proof.

1

2
is an ordinal.
(
8
3
5
)


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







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


18 SAHARON SHELAH
[Why? To prove that
2
is an ordinal we have to assume <
2
and prove

2
. As
2
clearly
y
[c] for some y Fil
4
()
(Y,
1
) for which D
y
1
= D
so there is f
Y
(
1
) such that fZ
y
F
y,
. So rk
D+Z[y]
(f) = hence by 0.7
there is g
Y
such that g f, i.e., (y Y )(g(y) f(y)) and rk
D+Z[y]
(g) = .
By 2.5(4) there is z Fil
4
()
(Y, ) such that D
z
1
= D and gZ
z
F
z,
so we are
done proving
2
is an ordinal.]
We dene the function c

with domain [
2
]
<hrtg(Y )
as follows:

2
c

(u) = u 0 : there is y Fil

4
()
(Y,
1
) such that f
y,
[c] is well
dened
4
and Rang(f
y,
[c]) c(v[u]).
where

3
v[u] := c(v) : v
1
is of cardinality and is w(v) .
where

4
for v
1
we let w(v) = Rang(f
z,
[c]) : z Fil
4
()
(Y,
1
) and u
and f
z,
[c] is well dened.
Note that

5
c

(u) = rk
D
(f) : D Fil
()
(Y ), Z D
+
and f
Y
v(u).
Note:

1
for each u
1
and x Fil
4
()
(Y,
1
) the set <
2
: f
x,
[c] is a well
dened function into u has cardinality < wlor(T
D
y
2
(u)), that is, f
x,
[c] :

x

2
) is a sequence of functions from Z
x
to u
1

1
, any two are
equal only on a set = mod D
x
2
(with choice it is
|Y |
[u[)), call this bound

|u|
.
Note

2
if u
1
u
2

2
then
() w(u
1
) w(u
2
) and v(u
1
) v(u
2
)
1
() c

(u
1
) c

(u
2
) for
+
, see Denition 0.15(4)
() if c

(u
1
) c

(u
2
) then c(v(u
1
)) c(v(u
2
))

3
if u
2
, [u[ < hrtg(Y ) then w(u) = f
y,
(z) : u, y Fil
4
()
(Y,
1
)
and z Z
y
is a subset of
1
of cardinality < hrtg([u[ Fil
4
()
(Y,
1
))
suphrtg() Fil
4
()
(Y,
1
)) : < hrtg(Y ) which is named
2,1

4
if u
1
and [u[ <
2,1
then c(v) : v [u]

is a subset of
1
of
cardinality < hrtg(
1
[u]

) sup
<2,1
hrtg(
1
[]

) which we call

2,2

5
if u
2
and [u[ < hrtg(Y ) then v(u) has cardinality <
2,2
.
[Why? By
3
and
3
and
4
.]

6
if u
2
and [u[ < hrtg(Y ) then c

(u)
2
and has cardinality <
2,3
which we call
2
.
4
we could have used {t Y : f,[c](t) c(v(u))} = mod D
y
2
)
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 19
[Why? Without loss of generality v(u) ,= . By
5
we have [c

(u)[ < hrtg(

Y
v(Y ))
Fil
()
(Y )) and by
5
the latter is suphrtg(
Y
Fil
()
(Y )) : <
2,2
=
2,3
recalling clause (e)() of the claim, so we are done.]

7
c

is a very weak closure operation on and has character (<

2
, Y ).
[Why? In Denition 0.15(1), clause (a) holds by the Denition of c

, clause (b)
holds by
6
and as for clause (c), 0 c

(u) by the denition of c

and u c

(u)
by the denition of c

.]
Now it is enough to prove

8
c

witnessed Ax
0
2,<2,Y
.
Recalling
7
, toward contradiction assume

U = U

: < ) is -decreasing,
U

[
1
]
<hrtg(Y )
and < U

c(U

). We dene =

: < ) by

= Min(U

c(U
+1
)).
As AC

follows from DC

, we can choose y

: < ) such that f

y,
[c] is well
dened for < .
We can choose z

: < ) such that z

Z
y
and

:= f
y,
(z

) / v(U
+1
).
[Why? By AC

it is enough to show that for each < there is such

. If there
is no such

then Rang(f
y,
) v(U
+1
) which by the denition of c

(U
+1
)
means that

(U
+1
).]
Let for <
u

: [, ).
So
()
1
u

[
1
]

[
1
]
<hrtg(Y )
.
[Why? By clause (a) of the assumption of 2.6.]
()
2
u

is -decreasing with .
[Why? By the denition.]
()
3

c(u
+1
) for < .
[Why?

by the denition of u

.]
Also u

v(U

) by the denition of v(U

).
Lastly,

/ v(U
+1
) by the choice of

. So u

: < ) contradict the

assumption on (
1
, c). From the above the conclusion should be clear.
2.6
{r.4A}
Claim 2.7. Assume
0
< = cf() < hence is regular of course, and D
is the club lter on and

=
i
: i < ) is increasing continuous with limit .
1)
+
rk
D
(f) : f

i<
+

+
i
.
2) Assume further c is a weak closure operation on of character (, ) then in
part (1) we can restrict ourselves to a function from

iZ

+
i
, Z .
(
8
3
5
)


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







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


20 SAHARON SHELAH
Proof. For each <
+
there is a one to one
5
function g from into [[ and
we let f :

i<

i
be
f(i) = otp( < : f() <
i
.
Let
F

= f : f is a function with domain satisfying i < f(i) <

+
i
such that for some one to one function g from into
for each i < we have f(i) = otp( < : f() <
i
).
Now
()
1
() F

,= for <
+
() F

: <
+
) exists as it is well dened
() if f F

then, for part (2), for some y Fil

4
()
(, c) we have
fZ[y] F
y,
.
[Why? For clause () let g : be one to one and so the f dened above
belongs to F

. For clause () see the denition of F

(for <
+
). Lastly, clause
() use 2.5(3).]
()
2
() if f F

, < <
+
then for some f

we have f

<
J
bd

f
() minrk
D
(f) : f F

: <
+
) is strictly increasing hence
minrk
D
(f) : f F

.
[Why? For clause (), let g witness f F

and dene the function f

by:
f

i<

+
i
by f

(i) = otp < : g() <

i
. So g witness f

, and
letting i() = mini : g() <
i
we have i [i(), ) f

(i) < f(i) hence

f

<
J
bd

f as promised. For clause () it follows.]

So we have proved 2.7.
2.7
{r.5}
Conclusion 2.8. 1) Assume
(a) Ax
0
,<,
(b) > cf() = (not really needed in (1)).
Then for some F

=: f : f a partial function from to we have
() every f

is a countable union of members of F

() F

is the union of [Fil

4
()
(, < )[ well ordered sets: F

y
: y Fil
4
()
(, )
() moreover there is a function giving for each y Fil
4
()
() a well ordering
of F

y
.
2) Assume in addition that hrtg(Fil
4
()
(, < )) < and hrtg(

) < then for

some y Fil
4
()
() we have [F
y
[ > .
5
but, of course, possibly there is no such sequence f : <
+

(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 21
Proof. 1) By the proof of 1.2.
2) Assume that this fails, let
i
: i < ) be increasing continuous with limit such
that
0
> hrtg(Fil
4
()
(, < )).
We choose f

by
f(i) = Min(f
y,
(i) : y Fil
4
()
(, )
f
y,
(i) is well dened, i.e.
i Z[y] and
y
and
otp(
y
) <
i
).
Now f(i) is well dened as the minimum is taken over a non-empty set, this holds as
we substruct from a set which as cardinality hrtg(Fil
4
()
(, )) +hrtg(

) +
i
which is < . But f contradicts part (1).
2.8
{r.6}
Conclusion 2.9. Assume Ax
0
,<,
.

+
is not measurable (which implies regular
6
) when
(a) > cf() = >
0
(b) > hrtg((Fil
4
()
(, )).
Proof. Naturally we x a witness c for Ax
0
,<,
. Let F
y
,
y
, f
y,
, F

y,
be dened
as in 2.5 so by claim 2.5, 2.7 we have
,y
: y Fil
4
()
()
+
.
Let y Fil
4
()
(, ) be such that [F
y
[ > , we can nd such y by 2.8, even
easier as without loss of generality we can assume
+
is regular (or even measurable,
toward contradiction). Let Z = Z[y]. So
y
is a set of ordinals of cardinality > .
For < otp(
y
) let

be the -th member of

y
, so f
y,

is well dened. Toward

contradiction let D be a (non-principal) ultralter on
+
which is
+
-complete.
For i Z let
i
< be the unique ordinal such that <
+
: f
y,

(i) = D.
As [Z[ <
+
and D is
+
-complete clearly :

iZ
f
y,

(i) =
i
D, so as
D is a non-principal ultralter, for some
1
<
2
, f
y,

1
= f
y,

2
, contradiction. So
there is no such D.
2.9
{r.7n}
Remark 2.10. Similarly if D is
+
-complete and weakly
+
-saturated and Ax
0

+
,<
,
see [Sh:F1078].
{r.9}
Claim 2.11. If Ax
0
,<,
, then we can nd

C such that:
(a)

C = C

: S)
(b) S = < : is a limit ordinal of conality ()
(c) C

is an unbounded subset of , even a club

(d) if S, cf() then [C

[ <
(e) if S, cf() > then [C

[ < hrtg( [cf()]

).
Remark 2.12. 1) Recall that if we have Ax
4

then trivially there is C

: < ,
cf() ), C

a club of of order type cf() as if <

well order []

we let C

:=
be the <

-minimal C which is a closed unbounded subset of of order type cf().

2) Ax
0
,<,
suces if < < .
6
the regular holds many times by 2.13
(
8
3
5
)


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







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


22 SAHARON SHELAH
Proof. The even a club is not serious as we can replace C

by its closure in .
Let c witness Ax
0
,<,
. For each S with cf() [(), ] we let
C

= c(C) : C a club of of order type cf().

Now

C

= C

: S and cf() [(), ]) is well dened and exist. Clearly C

is
a subset of .
For any club C of of order type cf() [(), ] clearly c(C) c(C) which
has cardinality < .
The main point is to show that C

is unbounded in , otherwise we can choose

by induction on < , a club C
,
of of order type cf(), decreasing with such
that C
,
c(C
,+1
), we use DC

. But this contradicts the choice of c recalling

Denition 2.2(1).
If < and cf() > we let
C

= c(u) : u C has cardinality :

C is a club of of order type cf().
A problem is a bound of [C

[. Clearly for C a club of of order type cf() the

order-type of the set c(v) : v C has cardinality is < hrtg([cf()]

).
As for C

is a club it is proved as above.

2.11
The following lemma gives the existence of a class of regular successor cardinals.
{s.2}
Lemma 2.13. 1) Assume
(a) is a limit ordinal <

with cf()
(b)

i
is a cardinal for i < increasing with i
(c)

i
: i <
(d)

i+1
hrtg(

i
)) for i < and () () where
() Ax
4

or
()

i+1
hrtg(Fil
4
()
(

i
, )) and hrtg([

i
]

i+1
(e) Ax
0
,<,
and <

0
(f) =
+

Then is a regular cardinal.

2) Assume Ax
4

, =
+

singular and <

hrtg(

then is
regular.
Remark 2.14. 1) This says the successor of a strong limit singular is regular.
2) We can separate the proof of
1
below to a claim.
Question 2.15. 1) Is hrtg(P(P(

i
))) hrtg(Fil
4
()
(

i
))?
2) Is [c(f B)[ hrtg([B]
<0
) for the natural c?
Proof. 1) We can replace by cf() so without loss of generality is a regular
cardinal.
So x c : []

P() a witness to Ax
0
,<,
and let C

[c] : < , cf() )

be as in the proof of 2.11, so < cf() < [C

[c][ < .
First, we shall use just >

( < )(cf() <

), a weakening of assumption
(f).
Now
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 23

1
for every i < and A of cardinality

i
, we can nd B of
cardinality

satisfying ( A)[ is limit cf()

i
=
sup( B)].
Why? By 2.11 the only problem is for Y := : A, > sup(A ), a limit
ordinal of conality < +
1
; so [Y [

i
. Note: if we assume Ax
4

this would be
immediate.
We dene D as the family of A Y such that

1
for some set C of ordinals, the set B
C
=: Rang(f
x,
) : x
Fil
4
()
(

i
) and C or for some C, we have

i
cf() > and
C

[c] satises Y A = sup( B

C
).
Clearly

2
(a) Y D
(b) D is upward closed
(c) D is closed under intersection of hence of < () sets.
[Why? For clause (a) use C = , for clause (b), note that if C witness a set A Y
belongs to D then it is a witness for any A

A. Lastly, for clause (c) if A

D
for < () <
+
, as we have AC

, there is a sequence C

: < ()) such that

C

witnesses A

D for < () <

+
, then C := C

: < () witnesses
A := A

: < () D and, again by AC

, we have [C[ .]

3
if D then we are done.
[Why? As B
C
is as required; its cardinality

i+1
by 2.11.]
So assume / D, so D is an
+
-complete lter on Y . As 1 [Y [

i
, let g be
a one to one function from [Y [

i
onto Y and let

3
(a) D
1
:= B

i
: g() : B [Y [ D
(b) := rk
D1
(g)
(c) D
2
:= B

i
: rk
D1+(

i
\B)
(g) > and B / D
1
of course.
So D
2
is an
+
-complete lter on

i
extending D
1
.
Let B

D
2
be such that (B

)[B

D
2
B

B c(Rang(g B

))
(Rang(g B)]. Let U = c(Range(g B

) : B

D
2
, so Rang(g B

) U .
Let h be the function with domain B

dened by B

h() = otp(g()
U ).
So x := (D
1
, D
2
, B

, h) Fil
4
()
(

i
) and for some we have g B

= f
x,
[c].
Now is a limit ordinal (as each g(), < [Y [, is) so it suces to consider the
following two subcases.
Subcase 1a: cf() > .
So C

[c] is well dened and let C := hence B

C
= Rang(f
x,
[c] :
C

[c] so C exemplies that the set A := Y : > sup( B

C
) belongs to
D hence A

= < [Y [ : g() A belongs to D

1
.
Now dene g

, a function from

i
to Ord by g

() = sup(g()B
C
)+1 if A

and g

() = 0 otherwise. Clearly g

< g mod D
1
hence rk
D1
(g

) < , hence there

is g

, g

<
D1
g

<
D1
g such that := rk
D1
(g

) C

[c].
Now for some y Fil
4
()
(

i
) we have D
y
= D
2
and g

= f
y,
mod D
y
2
.
(
8
3
5
)


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







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


24 SAHARON SHELAH
So B =: < [Y [ : g

() = f
y,
D
y
2
hence B D
+
2
. So B B

D
+
2
but if B B

then f
y,
() B
C
and f
y,
() (sup(B
C
g()), g()).
This gives contradiction.
Subcase 1b: cf() .
We choose a C of order type unbounded in and proceed as in subcase
1a.
As we have covered both subcases, we have proved
1
.
Recall we are assuming so

2
for every A of cardinality

there is B of cardinality

such
that:
(a) A B, [ + 1 A B] and [ A
0
cf() <

=
supp(B )].
[Why? Choose A
j
:< ) such that A = A
i
: i < and j < [A
j
[

j
,
possible as [A[

. For each j there

B
j
= B
j,i
: i < ) as in
1
so as AC

holds
(as ) there is a sequence

B
j
: j < ), each

B
j
as above.
Lastly, let B = B
j,i
: j, i < , it is as required.]

3
for every A of cardinality

we can nd B of cardinality

such that A B, [ + 1 B B] and [ B is a limit ordinal

cf() <

= sup(B )].
[Why? We choose B
i
by induction on i < such that [B
i
[

by B
0
=
B, B
2i+1
= : B
2i
or +1 B
2i+1
and B
2i+2
is chosen as B was chosen in

2
for i with B
2i+1
here in the role of A there. There is such B
i
: i < ) as DC
holds. So easily B = B
i
: i < is as required.]
Now return to our main case =
+

4

+

is regular.
[Why? Otherwise cf(
+

) <
+

hence cf(
+

, but

is singular so cf(
+

) <

hence there is a set A of cardinality cf(

+

) <

such that A
+

= sup(A). Now
choose B as in
3
. So [B[

, B is an unbounded subset of
+

, +1 B B
and if B is a limit ordinal then cf() [[

, but cf() is regular so cf() <

hence = sup(B ). But this trivially implies that B =

+

, but [B[

,
contradiction.]
2) Similar, just easier.
2.13
{sp.3}
Remark 2.16. Of course, if we assume Ax
4

then the proof of 2.13 is much simpler:

if <

is a well ordering of []

for < of conality let C

= the <

-rst
closed unbounded subset of of order type cf(), see 3.3.
{sp.4}
Claim 2.17. Assume
(a)
i
: i < ) is an increasing continuous sequence of cardinals >
(b) =

=
i
: i <
(c) = cf() >
(d) Ax
0
,<,
(e) hrtg(Fil
4
()
(, )) < and , <
0
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 25
(f) S := i < :
+
i
is a regular cardinal is a stationary subset of
(g) let D := D

+S where D

is the club lter on

(h) () = rk
D
(
+
i
: i < ).
Then () has conality > , so (, ()] Reg ,= .
Proof. Recall 2.7 which we shall use. Toward contradiction assume that cf(())

, but

is singular hence for some i() < , cf(())

i()
. Let c witness
Ax
0
,<,
.
Let B be an unbounded subset of () of order type cf(()) <
i()
. By renam-
ing without loss of generality i() = 0.
For < () let
U

= Rang(f
y,
) : f
y,
[c] is well dened
i
: i Z
y

and y Fil
4
()
() and D
y
1
= D.
Clearly U

is well dened by 2.7 and U

: < ()) exists and [U

[ hrtg(
Fil
4
()
(, )) = hrtg(Fil
4
()
(, )), even <. Let U = U

: B so [U [
hrtg(Fil
4
()
(, )) +[B[.
We dene f

i<

+
i
by
() f(i) is: sup(U
+
i
) + 1 if cf(
+
i
) > [U [
zero otherwise.
So
() f

i<

+
i
.
Clearly
() i < : f(i) = 0 = mod D moreover i < : f(i) is a limit ordinal D.
Let () = rk
D
(f), it is < rk
D
(
+
i
: i < )) = (), so by clause () there
is () B such that () < () < () hence for some g

i<

+
i
we have
rk
D
(g) = () and f < g mod D, so for some y Fil
4
()
(), D
y
1
= D

+ S and
g F
y,()
, hence f(i) < g(i) < f
y,()
(i) U
+
i
for every i Z
y
S.
So we get easy contradiction to the choice of g.
2.17
{sp.6}
Claim 2.18. Assume c witness Ax
0
,<
. The ordinals

, = 0, 1, 2 are nearly
equal see, i.e. below holds where:
(a)
0
= hrtg(
Y
), a cardinal
(b)
1
= ||
D
: D Fil
()
(Y )
(c)
2
= supotp(
y
[c] + 1 : y Fil
4
()
(Y )
()
2

1

0
()
0
is the union of Fil
4
()
(Y ) sets each of order type <
2
()
0
is the disjoint union of < hrtg(P(Fil
4
()
(Y ))) sets each of
order type <
2
(
8
3
5
)


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







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


26 SAHARON SHELAH
() if
0
> hrtg(P(Fil
4
()
(Y ))) and
0
[
0
[
+
then [
0
[ [
2
[
++
and cf([
2
[
+
) < hrtg(P(Fil
4
()
(Y ))).
Proof. Straight.
2.18
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 27
3. Concluding Remarks
In May 2010, David Aspero asked where it is true that I have results along the
following lines (or that it follows from such a result):
If GCH holds and is a singular cardinal of uncountable conality, then there
is a well-order of H (
+
) denable in (H (
+
), ) using a parameter.
The answer is yes by [Sh:497, 4.6,pg.117] but we elaborate this below somewhat
more generally.
Also though our original motivation was to look at consequences of Ax
4
, this
was shadowed by the try to use weaker relatives; see more in [Sh:F1078].
{c2}
Claim 3.1. [DC] Assume that is a limit singular of conality >
0
(no GCH
needed), the parameter X codes in particular the tree T =
>
and the set
P(P()), in particular from it well orderings of those sets are denable. Then
(with this paramter) we can dene a well ordering of the set of -branches of this
tree.
{c4}
Conclusion 3.2. [DC] Assume is a singular cardinal of uncountable conality
and H () is well orderable of cardinality and X code H () and a well
ordering of H (). Then we can (with this X as parameter) dene a well ordering
of P().
Proof. Proof of 3.1:
Let cd
i
: i < ) satises

1
cd
i
is a one-to-one from
i
into , (denable from X uniformly (in i))

2
let <

be a well ordering of Fil

4

() denable from X.
For

let f

: be dened by f

(i) = cd
i
(i), so

f = f

:

) is
well dened.
Let

F = F
y
: y Fil
4

()) be as in Theorem 1.2 with , here standing for

, Y there; there is such

F denable from X as X codes also []
0
.
So for every

there is y Fil
4

() such that fZ
y
F
y
so let y() be the
<

-rst such y. Now we dene a well ordering <

of

: for ,

let <

i rk
D1[y()]
(f

Z
y()
) < rk
D1(y())
(f

Z
y()
) or equality holds and y() < y().
This is O.K. because
() if ,=

then f

(i) ,= f

(i) for every large enough i < (i.e. i

minj : (j) ,= (j).

3.1
Proof. Proof of 3.2:
Let
i
: i < ) be an increasing sequence of cardinals < with limit .
Let cd

i
: i < ) satises

2
cd
i
is a one-to-one function from P(
i
) into , (denable uniformly from
X).
So cd

: P()

dened by (cd

(A))(i) = cd

i
(A
i
) for A , i < , is a
one-to-one function from P() into

. Now use 3.1.
3.2
We return to 2.13(2)
{c11}
Claim 3.3. [DC] 1) The cardinal
+
is regular when:
(
8
3
5
)


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







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


28 SAHARON SHELAH
(a) Ax
4

+
, i.e. [
+
]
0
is well orderable
(b) [[
0
< for <
(c) is singular.
2) Also there is e = e

: <
+
), e

= sup(e

), [e

[ cf()
0
.
Remark 3.4. Compare with 2.13. We use here more choice, but cover more cardi-
nals.
Proof. Let <

be a well ordering of the set [

+
]
0
.
Let F :

(
+
)
+
be such that there is no -decreasing sequence c
F
(u
n
) :
n < ) with u
n

+
. Let =
+
: a limit ordinal, <
+
cf() < ,
so
+
,
+
+ 1.
We dene e = e

: ) as follows.
Case 1: cf() =
0
, e

is the <

-minimal member of u : = sup(u) and

otp(u) = 0.
Case 2: cf() >
0
.
Let e

= c
F
(C) : C a club of .
So
()
1
e

is an unbounded subset of of order type < .

[Why? If cf() =
0
then e

has order type which is < by clause (b) of the

assumption.
If cf() >
0
then for some club C of , e

= c
F
(C) has otp(e

) [c
F
(C)[
(cf()
0
< . The last inequality holds as cf() as <
+
, cf() ,= as is
singular by clause (c) of the assumption, and lastly ((cf()
0
) < by clause (b) of
the assumption.]
This is enough for part (2). Now we shall dene a one-to-one function f

from
into by induction on as follows: pr

: is the pairing function

so one to one (can add onto ); if we succeed then f

+ cannot be well dened so

+
/ hence cf(
+
) , but is singular so cf(
+
) =
+
, i.e.
+
is not singular
so we shall be done proving part (1).
The inductive denition is:
(a) if then f

is the identity
(b) if = + 1 [,
+
) then for i < we let f

(i) be
1 +f

(i) if i <
0 if i =
(c) if so is a limit ordinal, e

= sup(e

), e

of cardinality
< and we let f

be dened by: for i < we let

f

(i) = pr

(f
min(e\(i+1))
(i), min(e

(i + 1))).

3.3
(
8
3
5
)


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







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


PCF WITHOUT CHOICE SH835 29
4. Private Appendix
Moved 2010.8.09 from Remark ??, pgs.10,11:
This works as well from weaker assumptions (see more later).
(b) we can dene functions i
n
: []
0
such that u = i
n
(u) : n <
(enough to dene it for the countable ordinals). This holds if
) [
1
]
0
is well ordered or just
() there is h

: [,
1
)) such that h

is one-to-one from onto .

Then it suces to apply F

just once.
[Why? For countable v , let F

(v) be i
n
(u
(v)
) : n is minimal such that
i
n
(u
(v)
) / v if there is such n and n = 0 otherwise. Why can we apply F

just
once? Toward contradiction assume U
n
: n < ) is strictly decreasing. So choose

n
U
n
U
n+1
. For some innite w ,
n
: n w) is strictly increasing and
included in some u

. Let = (v). Let

n
= i(u

) for n < hence

n
: n < )
list u

. Without loss of generality there is an increasing function h from w to

such that
n
= i
h(n)
(u

). Let n

< n

be the rst two members of u.

Let u =
n
: n

n w so u U
n
, n

1 and clearly
n

:
n

and we can prove by downward induction on n

that

U
n
because

= F

(u
k
: < n

, k ()), contradiction.]
Moved 2010.8.09 from Claim 2.3, pg.14:
2) If there is a -well founded [weak] closure operation on of character (, ) then
there is a well founded [weak] closure operation of character (, ).
3) Assume < hrtg(H
<1
()),
0
. There is a well founded weak closure operation
of character (,
0
) on and
2
there is a -well founded weak closure operation
of character (<
0
) on .
Moved 2010.8.09 from Proof, pgs.15:
2) Assume c : []

[]

witness the assumption; let c

: []

[]

be
c

(u) = c(v) : v u, [v[ u which is c(u) recalling Denition 0.15(1A)(d).

So c

witness the desired conclusion using proved part (4) below.

3) Let c : []

[]

witness the assumption and we dene c

: []

[]

by c

(u) = c(v) : v u, [v[ and continue as above.

Moved 2010.8.09 from Claim 2.7, pg.18:
Remark 4.1. 1) Assuming Ax
0
,,
but not AC

, still
+
is regular (see later) or
just has conality hrtg(Fil
4

(, ) we can nd the limit of the F

s.
2) See [Sh:497].
Moved 2010.8.09 from Remark 2.11, pg.21:
{r.9a}
Remark 4.2. In 2.11, we are given S

S such that cf() : S

) is constant then
we can choose C

: S

), C

is a club of of order type cf().

Moved 2010.8.09 from end of 2, pg.24:
(
8
3
5
)


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







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


30 SAHARON SHELAH
{sp.5}
Claim 4.3. Assume(a)-(e) from 2.17.
1) If a cardinal < and S := i < : cf(
+
i
) > is a stationary subset of
then cf(
+

) > .
2) If AC

and S is stationary then cf(

+

) < hrtg(

iS
cf(
+
i
)).
Proof. 1) Like 2.17.
2) Let D

be the club lter on and let D = D

+ S. By AC

we can choose

B = B
i
: i S), B
i
a club of
+
i
and otp(B
i
) = cf(
+
i
) and let B
i
= 0 for
i S. Let B = rk
D
(f) : f

iS
B
i
, it is an unbounded of
+
, so f rk
D
(f)
exemplify hrtg(

iS
B
i
) > cf(
+

), as S D so we are done.
4.3
Implicit above is
END OF LATEX REVISIONS
References
[AM95] Arthur Apter and Menahem Magidor, Instances of Dependent Choice and the Measura-
bility of
+1
, Annals of Pure and Applied Logic 74 (1995), 203219.
[DJ82] Tony Dodd and Ronald B. Jensen, The covering lemma for K, Ann. of Math Logic 22
(1982), 130.
[Eas70] William B. Easton, Powers of regular cardinals, Annals of Math. Logic 1 (1970), 139178.
[GH75] Fred Galvin and Andras Hajnal, Inequalities for cardinal powers, Annals of Mathematics
101 (1975), 491498.
[Sh:g] Saharon Shelah, Cardinal Arithmetic, Oxford Logic Guides, vol. 29, Oxford University
Press, 1994.
[Sh:E37] , What majority decisions are possible, math.CO/0303323.
[Sh:71] , A note on cardinal exponentiation, The Journal of Symbolic Logic 45 (1980),
5666.
[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.
[Sh:589] , Applications of PCF theory, Journal of Symbolic Logic 65 (2000), 16241674.
[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:955] , Pseudo pcf, preprint.
[Sh:F1039] , Bounds on pcf with weak choice using ranks and normal lters.
[Sh:F1078] , Pcf with narrow choice or ZF + DC +AX
4
.
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