4

4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Many simple cardinal invariants
November 1991
Martin Goldstern
1
Bar Ilan University
Saharon Shelah
2
Hebrew University of Jerusalem
Abstract: For g < f in

we dene c(f, g) be the least number of uniform

trees with g-splitting needed to cover a uniform tree with f-splitting. We
show that we can simultaneously force
1
many dierent values for dierent
functions (f, g). In the language of [Blass]: There may be
1
many distinct
uniform
0
1
characteristics.
0. Introduction
[Blass] dened a classication of certain cardinal invariants of the continuum, based on the Borel hierarchy.
For example, to every
0
1
formula (x, y) = nR(xn, yn) (R recursive) the cardinal

:= minB

: x

y B : (x, y)
is the uniform
0
1
characteristic associated to .
Blass proved structure theorems on simple cardinal invariants, e.g., that there is a smallest
0
1
characteristic
(namely, Cov(/), the smallest number of rst category sets needed to cover the reals), and also that the

0
2
-characteristics can behave quite chaotically. He asked whether the known uniform
0
1
characteristics
(c, d, r, Cov(/)) are the only ones or (since that is very unlikely) whether there could be a reasonable
classication of the uniform
0
1
characteristics say, a small list that contains all these invariants.
In this paper we give a strong negative answer to this question: For two
0
1
formulas
1
,
2
we say that

1
and
2
dene potentially nonequal characteristics if
1
,=
2
is consistent. We say that
1
and
2
dene actually dierent characteristics, if
1
,=
2
.
We will nd a family of
0
1
-formulas indexed by a real parameter (f, g), and we will show not only that
there is a perfect set of parameters which denes pairwise potentially nonequal
0
1
-characteristics, but we
produce a single universe in which (at least)
1
many cardinals appear as
0
1
-characteristics. (In fact it
1
supported by Israeli Academy of Sciences, Basic Research Fund
2
Publication 448. Supported partially by Israeli Academy Of Sciences, Basic Research Fund and by the
Edmund Landau Center for research in Mathematical Analysis (supported by the Minerva Foundation (Ger-
many))
1 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


is also possible to produce a universe where there is a perfect set of parameters dening pairwise actually
dierent
0
1
-characteristics. See [Shelah 448a]).
If we want more than countably many cardinals, we obviously have to use the boldface pointclass. But the
proof also produces many lightface uniform
0
1
characteristics.
For more information on cardinal invariants, see [Blass], [van Douwen], [Vaughan].
From another point of view, this paper is part of the program of nding consistency techniques for a large
continuum, i.e., we want 2
0
>
2
and have many values for cardinal invariants. We use a countable support
product of forcing notions with an axiom A structure.
We will use invariants that were implicitly introduced in [Shelah 326, 2], where it was proved that c(f, g)
and c(f

, g

) (see below) may be distinct.

0.1 Denition: If f

, we say that

B = B
k
: k ) is an f-slalom if for all k, [B
k
[ = f(k). We write
h

B for h

n
B
n
, i.e., nh(n) B
n
. (See gure 1) This is a
0
1
-formula in the variables h and

B.
Some authors call the set h : h

B a belt, or uniform tree.
For example,

n
f(n) is an f-slalom, because we identify the number f(n) with the set of predecessors,
0, . . . , f(n) 1.

3
2
1
B
B
0
B
B
Figure 1: A slalom
0.2 Denition: Assume f, g

. Assume that B is a family of g-slaloms, and

A = A
k
: k ) is an
f-slalom.
We say that B covers

A i:
() for all s

A there is

B B such that s

B
2 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


0.3 Denition: Assume f, g

. Then we dene the cardinal invariant c(f, g) to be the minimal number
of g-slaloms needed to cover an f-slalom.
(Clearly this makes sense only if k f(k), g(k) > 0, so we will assume that from now on.)
This is a uniform
0
1
-characteristic. (Strictly speaking, we are not working in

, but rather in

_
[]
<
_
,
but a trivial coding translates c(f, g) into a uniform
0
1
characteristic as dened above.)
Some relations between these cardinal invariants are provable in ZFC: For example, if g < g

< f

< f, then
c(f

, g

) c(f, g). Also, c(f

2
, g
2
) c(f, g).
We will show that if (f, g) is suciently dierent from (f

, g

), then the values of c(f, g) and c(f

, g

) are
quite independent, and moreover: if (f
i
, g
i
) : i <
1
) are pairwise suciently dierent, then almost any
assignment of the form c(f
i
, g
i
) =
i
will be consistent.
Similar results are possible for the dual version of c(f, g): c
d
(f, g) := the smallest family of g-slaloms

B
such that for every h bounded by f there are innitely many k with h(k) B
k
, and for the tree version
(a g-tree is a tree where every node in level k has g(k) many successors). See [Shelah 448a].
We thank Tomek Bartoszynski for pointing out the following known results about the cardinal characteristics
c(f, g):
For example, lemma 1.11 follows from Theorem 3.17 in [Comfort-Negrepontis]: Taking = = , = n,
and letting o n

be a family of -large oscillation, then no family of n1-slaloms of size < 2

0
can cover
o. Indeed, whenever F is a function on o such that for each s o, F(s) is a n1-slalom covering s, then
F has to be nite-to-one and in fact at most n1-to-one.
Also, since c(f, f1) is the size of the smallest family of functions below f which does not admit an innitely
equal function, i.e.,
c(f, f1) = min[G[ : G

n
f(n) & h

n
f(n) g G

n f(n) ,= g(n)
by [Miller] we have that the minimal value of c(f, f1) is the smallest size of a set of reals which does not
have strong measure zero.
Also, note that if r is a random real over V in

n
f(n), and if

n=1
1/f(n) = , then

n
(11/f(n)) = 0,
so r cannot be covered by any f1-slalom from V .
Conversely, if

n=1
1/f(n) < , then for any function h

n
f(n) V there is a condition forcing that h
is covered by the f1-slalom (0, . . . , f(k) 1 r(k) : k ).
Thus, if we add many random reals with the measure algebra, a easy density argument shows that in the
resulting model we have
c(f, f1) =
_
= 2
0
if

n=1
1/f(n) =

1
otherwise (use any
1
many of the random reals)
That already shows that we can have at least two distinct values of c(f, g) and c(f

, g

).
3 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Contents of the paper: In section 1 we prove results in ZFC of the form
If (f, g) is in relation . . . to (f

, g

), then c(f, g) c(f

, g

)
In section 2 we dene a forcing notion Q
f,g
that increases c(f, g). (I.e., in V
Q
f,g
, the g-slaloms from V do
not cover

n
f(n).) Informally speaking, elements of Q
f,g
are perfect trees in which the size of the splitting
is bounded by f, sometimes = 1, but often (i.e., on every branch), much bigger than g.
In section 3 we show that, assuming (f

, g

) : <
1
are suciently independent, a countable support
product

<1
Q

of such forcing notions will force c(f

, g

) =

.
We use the symbol

i
to denote the end of a proof, and we write

i
when we leave a proof to the reader.
1. Results in ZFC
1.1 Notation: Operations and relations on functions are understood to be pointwise, e.g., f/g, g

, g < f,
etc. x| is the greatest integer x. limf is lim
k
f(k).
We write f

g for nk n f(k) g(k).

First we state some obvious facts:
1.2 Fact:
(1) f g i c(f, g) = 1.
(2) f

g i c(f, g) nite.
(3) If A := k : g(k) < f(k) is innite then c(fA, gA) = c(f, g).
(4) If is a permutation of , then c(f , g ) = c(f, g).

i
1.2
(Strictly speaking, we dene c(f, g) only for functions f, g dened on all of , so (3) should be formally
rephrased as c(f h, g h) = c(f, g), where h is a 1-1 enumeration of A)
1.3 Convention: We will concentrate on the case where c(f, g) is innite, so we will wlog assume
that g < f. By (4), we may also wlog assume that g is nondecreasing.
In these cases we will have that c(f, g) is innite, and moreover an easy diagonal argument shows the
following fact:
1.4 Fact:
c(f, g) is uncountable.

i
1.4
Furthermore, we have the following properties:
1.5 Fact:
(1) (Monononicity) If f

, g

, then c(f, g) c(f

, g

).
(2) (Multiplicativity) c(f f

, g g

) c(f, g) c(f

, g

).
4 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


(3) (Transitivity) c(f, h) c(f, g) c(g, h).
(4) (Invariance) c(f, g) = c(f

, g

) (where f

is the function dened by f

(n) = f(n + 1).

(5) (Monotonicity II) If A is innite, then c(fA, gA) c(f, g).

i
1.5
1.6 Remark: (2) implies in particular c(f
n
, g
n
) c(f, g). See 3.4 for an example of c(f
2
, g
2
) < c(f, g).
The following inequalities need a little more work.
1.7 Lemma:
(1) c(f f/g| , f) = c(f, g).
(2) c(f f/g| , g) = c(f, g).
(3) c(f f/g|
m
, g) = c(f, g) for all m .
Proof: (2) follows from (1) using transitivity, and (3) follows from (2) by induction, so we only have to
prove (1).
Proof of (1): By monotonicity we only have to show . So let (N, ) be a reasonably closed model of a large
fragment of ZFC (say, (N, ) < (H(
+
), ), where = 2
c
) of size c(f, g) such

n
f(n) is covered by the set
of all g-slaloms from N.
Dene h by h(k) := f(k) f(k)/g(k)|. We can nd a family B
i
k
: i < f(k), k ) in N such that for all k,
0, . . . , h(k) 1 =

i<f(k)
B
i
k
, where [B
i
k
[ f(k)/g(k). We have to show that the set of f-slaloms from N
covers

k
h(k).
So let x be a function satisfying k x(k)

i<f(k)
B
i
k
. We can dene a function y

n
f(n) such that for
all k, x(k) B
y(k)
k
. So there is some g-slalom

C N such that for all k, y(k) C
k
.
Dene

A = A
k
: k ) by A
k
:=

iC
k
B
i
k
. Then [A
k
[ [C
k
[ [B
i
k
[ g(k) f(k)/g(k) = f(k), so

A is an
f-slalom in N, and for all k, x(k) A
k
.

i
1.7
1.8 Lemma: Assume f > g > 0. Assume that w
i
: i ) is a partition of into nite sets, and for each
i there are

H
i
= H
i
l
: l w
i
) satisfying (a)(c). Then c(f

, g

) c(f, g).
(a) domH
i
l
= f

(i) = 0, . . . , f

(i) 1
(b) rng H
i
l
f(l) = 0, . . . , f(l) 1
(c) Whenever u
l
: l w
i
) satises
u
l
f(l)
[u
l
[ g(l)
then n < f

(i) : l w
i
H
i
l
(n) u
l
has cardinality g

(i)
Proof: To any g-slalom

B = B
l
: l ) we can associate a g

-slalom

B

= B

i
: i ) by letting
B

i
:= n < f

(i) : l w
i
H
i
l
(n) w
l

5 Revised version, April 1992

4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Conversely, to any function x

i
f

(i) we can dene a function x

in

n
f(n) by
if l w
i
, then x

(l) = H
i
l
(x(i))
It is easy to check that if x

is in

B then x is in

B

. The result follows.

i
1.8
1.9 Corollary: Assume 0 = n
0
< n
1
< , and let
f

(i) := f(n
i
) f(n
i
+ 1) f(n
i+1
1)
g

(i) := g(n
i
) g(n
i
+ 1) g(n
i+1
1)
Then c(f

, g

) c(f, g).
Proof: Identify the set of numbers less than f(n
i
) f(n
i
+ 1) f(f
i+1
1) with the cartesian product

nik<ni+1
f(k), and let
H
i
l
:

nik<ni+1
f(k) f(l)
be the projection onto the l-coordinate. We leave the verication of 1.8(c) to the reader.

i
1.9
1.10 Lemma: If g is constant, f(k) 2
k
, then c(f, g) = c.
Proof: Let k g(k) = n, f(k) = 2
k
. Assume that

l
l
2 can be covered by < c many g-slaloms.
For any

2, the sequence := l : l ) is in

l
l
2. But any g-slalom can contain only n many such
, i.e. for any g-slalom

B = B
l
: l ) we have

_


2 : l l B
l
_

m
Since there are continuum many we need continuum many g-slaloms to cover

l
f(l) (or equivalently,

l
l
2).

i
1.10
1.11 Lemma: If f and g are constant with f > g, then c(f, g) = c.
Proof: Using monotonicity wlog we assume that f(k) = n + 1, g(k) = n for all k. We will use 1.8. Let
=

i
w
i
be a partition of where [w
i
[ = n
2
i
.
Let f

(i) = 2
i
, g

(i) = n, and let H

i
l
: l w
i
) enumerate all functions from 2
i
to n.
We plan to show c(f, g) c(f

, g

) (so c(f, g) = c by 1.10). We want to apply 1.8, so x a sequence

u
l
: l w
i
),where u
l
f(l) and [u
l
[ g(l).
To show that the hypotheses of 1.8 are satised, x i
0
and let
A := x < f

(i
0
) : l w
i0
H
i0
l
(x) u
l

and assume A has cardinality > g

(i
0
) = n. So let x
0
, . . . , x
n
be distinct elements of A. Let H : f

(i
0
) n+1
be a function satisfying
j n H(x
j
) = j
H is one of the functions H
i0
l
: l w
i0
, say H = H
i0
l0
. Let j
0
/ u
l0
, then also
x
j0
/ x < f

(i
0
) : H
i0
l0
(x) u
l0
A,
contradicting x
j0
A.

i
1.11
6 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


1.12 Corollary: If f > g, and liminf
k
g(k) < , then c(f, g) = c.
Proof: This follows from 1.11, using monotonicity and monotonicity II.

i
1.12
We can now extend 1.7 as follows:
1.13 Theorem: If for some > 0, g
1+
f, then for all n, c(f
n
, g) = c(f, g).
Proof: First we consider a special case: Assume that g
2
f. Then we get
c(f, g) c(f
2
, g) c(f
2
, f) c(f, g) c(f
2
, g
2
) c(f, g) = c(f, g)
Now we use this result on (f, g), then on (f
2
, g), etc, to get
c(f, g) = c(f
2
, g) = c(f
4
, g) = c(f
8
, g) =
and use monotonicity to get the general result under the assumption g
2
f.
Now we consider the general case g
1+
f:
If g does not diverge to innity, we have already (by 1.12) c(f, g) = c. Otherwise we can nd some > 0
such that for almost all k,
f(k)
g(k)
g(k)

+ 1,
so
_
f(k)
g(k)
_
g(k)

Now choose m such that m > 1. Then f(k)/g(k)|

m
g. By 1.7, c(f f/g|
m
, g) = c(f, g) and so by
monotonicity also c(f g, g) = c(f, g). Since g
2
f g, we can apply the result from the special case above
to get c(f, g) = c(f
n
g
n
, g) so in particular, c(f
n
, g) = c(f, g).

i
1.13
If f is not much bigger than g, the assumption in 1.7 and 1.13 may be false. For these cases, we can prove
the following:
1.14 Lemma:
(1) c(2f g, f) = c(f, g).
(2) c(2f g, g) = c(f, g).
(3) c(f +m(f g), g) = c(f, g) for all m .
Proof: The proof is similar to the proof of 1.7. Again we only have to show (1). Let (N, ) be a reasonably
closed model of a large fragment of ZFC (say, (N, ) (H(
+
), ), where = 2
c
) of size c(f, g) such

n
f(n) is covered by the set of all g-slaloms from N.
Dene h by h(k) := f(k) + f(k) g(k). We can nd a family B
i
k
: i < f(k), k ) in N such that for all
k, 0, . . . , h(k) 1 =

i<f(k)
B
i
k
, where [B
i
k
[ = 2 for l < f(k) g(k), and [B
i
k
[ = 1 otherwise. We have to
show that the set of f-slaloms from N covers

k
h(k).
So let x be a function satisfying k x(k)

i<f(k)
B
i
k
. We can dene a function y

n
f(n) such that for
all k, x(k) B
y(k)
k
. So there is some g-slalom

C N such that for all k, y(k) C
k
.
7 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Dene

A = A
k
: k ) by A
k
:=

iC
k
B
i
k
. Thus A
k
is the union of g(k) many sets, of which at most
f(k) g(k) are pairs, and the others singletons. Thus [A
k
[ g(k) +(f(k) g(k)) = f(k), so

A is an f-slalom
in N, and for all k, x(k) A
k
.

i
1.14
Similar to the proof of 1.13 we now get:
1.15 Lemma:
(1) If 2g f, then for all n, c(nf, g) = c(f, g).
(2) If for some > 0, (1 +)g f, then for all n, c(nf, g) = c(f, g).

i
1.15
2. The forcing notion Q
f,g
2.1 Denition: We x sequences n

k
: k ) and n
+
k
: k ) that increase very quickly and satisfy
n

0
n
+
0
n

1
n
+
1
. In particular, we demand
(1) For all k

j<k
n

j
n

k
(2) lim
k
log n
+
k
log n

k
= 0.
(3) n

k
n
+
k
< n

k+1
.
We will only consider functions f, g satisfying n

k
g(k) < f(k) n
+
k
. This is partly justied by 1.9, and
it also helps to keep the formulation of the main theorem relatively simple.
2.2 Denition: Let X ,= be nite, c, d . A (c, d)-complete norm on P(X) is a map
| | : P(X)
mapping any nonempty a X to a number |a| such that
whenever a = a
1
a
c
X, then for some i
1
, . . . , i
d
1, . . . , c, |a
i1
a
i
d
| |a|1.
([a[ is the cardinality of the set a)
A natural (c, d)-complete norm is given by |a| := log
c/d
[a[. c-complete means (c, 1)-complete.
2.3 Denition: We call (f, g, h) progressive, if f, g, h are functions in

, satisfying
(1) For all k, n

k
g(k) < f(k) n
+
k
(2) For all k, n

k
h(k)
(3) lim
k
log
f(k)
g(k)
_
log h(k) = .
We call (f, g) progressive, if there is a function h such that (f, g, h) is progressive(or equivalently, if (f, g, n

)
is progressive, where n

is the function dened by n

(k) = n

k
).
8 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


2.4 Remark: For example, if f and g satisfy (1), then (f, g, g) is progressive i log f/ log g .

i
2.4
In 2.6 we will dene a forcing notion Q
f,g,h
for any progressive (f, g, h). First we recall the following notation:
2.5 Notation:
<
=

n
n
2 is the set of nite sequences of natural numbers. For s
<
, [s[ is the length
of s.
A tree p is a nonempty subset of
<
with the properties
p k < [[ : k p
p : succ
p
() ,= , where
succ
p
() := p : , [[ + 1 = [[.
A branch b of p is a maximal linearly -ordered subset of p. Every branch b denes a function

b : by

b =

b. We usually identify b and

b, so we write bk (instead of (

b)k) for the kth element of b.

The set of all branches of p is written as [p].
For p, we let
p
[]
:= p : or
We let
split(p) := p : [succ
p
()[ > 1 (the splitting nodes of p)
split
n
(p) := split(p) : [ : split(p)[ = n (the n-th splitting level)
and we dene the stem of p to be the unique element of split
0
(p).
2.6 Denition: Assume f, g, h are as in 2.3. Then we dene for all k, and for all sets x
|x|
k
:=
_
log([x[/g(k))
log h(k)
_
and we dene the forcing notion Q
f,g
(or more accurately, Q
f,g,h
) to be the set of all p satisfying
(1) p is a perfect tree.
(2) p i dom() (i) < f(i).
(3) split
n
(p) |succ
p
()|
||
n.
We let p q (q extends p) i q p.
2.7 Remark: If we dene
p _
k
q i p q and split
k
(p) q
then Q
f,g,h
satises axiom A, and is in fact strongly

-bounding, i.e., for name of an ordinal,

, for any p
and for any n there is a nite set A and a condition q _
n
p, q

A. However, it will be more convenient

to use the relation
n
that is based on levels rather than splitting levels.
9 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


2.8 Denition: For p, q Q, n we dene
p
n
q i p q and p
n
q
2.9 Notation: We will usually write ||
p
instead of |succ
p
()|
||
.
2.10 Remark: This forcing is similar to the forcing in [Shelah 326], but note the following important
dierence: Whereas in [Shelah 326] all nodes above the stem have to be splitting points, we allow many
nodes to have only one successor, as long as there many nodes with high norm.
2.11 Remark:
(1) The norm ||
k
is h(k)-complete (hence also n

k
-complete).
(2) If c/d h(k), then the norm is (c, d)-complete.
(3) If |a|
k
> 0, then [a[ > g(k).
(4) |f(k)|
k
(so Q
f,g,h
is nonempty).

i
2.11
We will see in the next section that this forcing (and any countable support product of such forcings) is
proper and

-bounding. For the moment, we only show why this forcing is useful in connection with c(f, g):
2.12 Fact: Any generic lter G Q
f,g
denes a generic branch
r :=
_
pG
stem(p)
that avoids all g-slaloms from V .
Proof: Let

B = B
k
: k ) be a g-slalom in V , and let p Q
f,g
be a condition. Let p be a node
satisfying ||
p
> 0. Let k := [[. Then [succ
p
()[ > g(k) by 2.11(3), so there is i / B
k
,

i p. So
p
[

i]
r(k) = i / B
k
.

i
2.12
3. The construction
In this section we will prove the following theorem:
3.1 Theorem (CH): Assume that (f

, g

: <
1
) is a sequence of progressive functions, witnessed by
functions h

(see 2.3).
Let (

: <
1
) be a sequence of cardinals satisfying

such that whenever

<

, then
lim
k
min
_
f

(k)
g

(k)
,
f

(k)
g

(k)
_
h

(k)
_
= 0
(or informally: either f

, or f

/g

, or a combination of these two condition holds)

Then there is a proper forcing notion P not collapsing cardinals nor changing conalities such that

P
: c(f

, g

) =

10 Revised version, April 1992

4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


For the proof we use a countable support product of the forcing notions Q
f

,g

,h

described in the previous

section.
3.2 Remark: The theorem is of course also true (with the same proof) if we have countably or nitely
many functions to deal with.
If we are only interested in 2 cardinal invariants c(f

, g

), c(f, g), then we can phrase the theorem without

the auxiliary functions h as follows: If (f, g) and (f

, g

) are progressive, and satisfy

min
_
f

g
,
log(f/g)
log(f

/g

)
_
0
then c(f, g) < c(f

, g

) is consistent.
In particular, this shows that our result is quite sharp: For example, if for some function d we have limd = ,
f

= f
d
, g

= g
d
(and (f, g), (f

, g

) are progressive with the same n

k
, n
+
k
), then c(f, g) < c(f

, g

) is
consistent. On the other hand, c(f
n
, g
n
) c(f, g) for every xed n.
Proof: Choose h

such that log h

2 log(f/g) whenever
f

g

log(f/g)
log(f

/g

)
. (f

, g

, h

) is progressive, and the

assumptions of the theorem are satised. (Recall that (f, g) is progressive, hence log f/g log n

, so h

will satisfy h

(k) n

k
).

i
3.2
A similar simplied formulation of 3.1 is possible when we deal with only countably many functions.
3.3 Example: There is a family (f

, g

, g

: < c) of continuum many progressive functions such that for

any ,= , min
_
f

,
f

_
0. [In particular, under CH we may choose any family (

: <
1
) of cardinals
satisfying

and get an extension where c(f

, g

) =

.]
Proof: Let
k
:=
_
1
2
_
log
log n
+
k
log n

k
_
. (Here, log can be the logarithm to any (xed) base, say 2.) Then
lim
k

k
= , and by invariance (1.5(4)) we may assume
k
1 for all k.
Let T 2
<
be a perfect tree such that for all k we have [T 2
k
[ =
k
, say, T 2
k
= s
1
(k), . . . , s

k
(k).
For any x [T] (i.e., x 2

, k xk T) we now dene functions f

x
, g
x
, h
x
by:
If xk = s
i
(k), then
f
x
(k) =
_
n

k
_

2i
k
h
x
(k) = g
x
(k) =
_
n

k
_

2i1
k
We leave the verication that (f
x
, g
x
, h
x
) is indeed progressive to the reader. [Recall 2.4, and also note that
log log f
x
(k) 2
k
log
k
+ log log n

k
< log log n
+
k
. Finally, note that if x ,= y, then for almost all k we have
min
_
f
x
(k)
g
y
(k)
,
f
y
(k)
h
x
(k)
_

1
n

k
.]

i
3.3
3.4 Example: It is consistent to have c(f
2
, g
2
) < c(f, g) (for certain f, g).
Proof: Let
k
:=
_
1
6
log
n
+
k
n

k
_
. Assume
k
> 0 for all k. Then, letting
f(k) :=
_
n

k
_
3
k
g(k) :=
_
n

k
_
2
k
h(k) := n

k
11 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


We have that (f, g, h) and (f
2
, g
2
, h) are progressive, and lim
f
g
2
= 0, so we can apply the theorem.

i
3.4
3.5 Denition:
Let be a disjoint union =

<1
A

, where [A

[ =

.
For < , let Q

be the forcing Q
f

,g

,h

, if A

, and let P =

<
Q

be the countable support

product of the forcing notions Q

, i.e., elements of P are countable functions p with dom(p) , and

dom(p) p() Q

.
For A , we write PA := pA : p P. Clearly PA < P for any A. In particular, Q

< P.
We write r

for the Q

-name (or P-name) for the generic branch introduced by a generic lter on Q

.
We say that q strictly extends p, if q p and dom(q) = dom(p).
3.6 Facts: Assume CH. Then
(1) each Q

is proper and

-bounding.
(2) P is proper and

-bounding.
(3) P satises the
2
-cc.
(4) Neither cardinals nor conalities are changed by forcing with P.
Proof of (1), (2): See below (3.23, 3.24)
Proof of (3): A straightforward -system argument, using CH.
(4) follows from (2) and (3).

i
3.6
We plan to show that
P
c

for all <

1
.
3.7 Denition: If p P, k , we let the level k of p be
Level
k
(p) :=
_
: dom( ) = dom(p),
dom( ) : [ ()[ = k, () p()
_
We dene the set of active ordinals at level k as
active
k
(p) := dom(p) : [stem(p())[ k
3.8 Remark: Sometimes we identify the set Level
k
(p) with the set
: dom( ) = active
k
(p), dom( ) : [ ()[ = k
= active
k
(p) : Level
k
(p)
3.9 Denition: We say that the kth level is a splitting level of p (or k is a splitting level of p) i
dom(p) split(p()) : [[ = k
3.10 Denition: If Level
k
(p),

Level
k
(p), k < k

, then we say that

extends i for all

dom( ),

() extends (i.e., ) ().

12 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


3.11 Denition: For p, q P, k , we let
p
k
q i p q and dom(p) : p()
k
q() and active
k
(p) = active
k
(q)
That is, we allow dom(q) to be bigger than dom(p), but for all new dom(q) dom(p) we require that
[stem(q())[ > k.
3.12 Denition: Let A P. A set D P is
dense in A, if p Aq D : p q
strictly dense in A, if p Aq D : p q and dom(p) = dom(q)
open in A, if p Dq A: (p q implies q D)
almost open in A, if p Dq A: (p q and dom(p) = dom(q) implies q D)
These denitions can also be relativized to conditions above a given condition p
0
. If we omit A we mean A =
P.
3.13 Denition: If Level
k
(p), we let q = p
[ ]
be the condition dened by dom(q) = dom(p), and
dom(q) q() = p()
[ ()]
3.14 Denition: If p x

V , and Level
k
(p), we say that decides x

(or more accurately, p

[ ]
decides x

) if for some y V , p
[ ]
x

= y.
First we simplify the form of our conditions such that all levels are nite.
3.15 Fact: The set of all conditions p satisfying
I k [active
k
(p)[ < , and moreover:
II For any splitting level k there is exactly one pair (, ) such that [succ
p()
()[ > 1.
is dense in P.

i
3.15
3.16 Fact: If p is in the dense set given by (I) and (II), then the size of level k is n

k1
n
+
k1
< n

k
.
Proof: By induction.

i
3.16
From now on we will only work in the dense set of conditions satisfying (I) and (II).
3.17 Notation: For p satisfying (I)(II), we let k
l
= k
l
(p) be the lth splitting level. Let
l
=
l
(p) and

l
=
l
(p) be such that [
l
(p)[ = k
l
(p),
l
(p) split(p(
l
)). We let
l
=
l
(p) be such that
l
A

l
.
We write |p|
k
l
for |
l
|
p(
l
)
, i.e., for
_
_
succ
p(
l
)
(
l
)
_
_

l
,k
l
. (See gure 2)
3.18 Denition: If p is a condition, l ,

:=
l
(p),

:=
l
(p),

succ
p(

)
(

), we can dene a
stronger condition q by letting q() = p() for all ,=

, and
q(

) := p(

) : If

, then

In this case, we say that q was obtained from p by pruning the splitting node

.
To simplify the notation in the fusion arguments below, we will use the following game:
13 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9

0
succ( )
k
0
k
1
k
2

2
dom(p)
0
1
2

=
Figure 2: A condition satisfying (I) and (II)
3.19 Denition: For any condition p P, G(P, p) is the following two person game with perfect
information:
There are two players, the spendthrift and the accountant. A play in G(P, p) last many moves (starting
with move number 1) The accountant moves rst. We let p
0
:= p, i
0
:= 0.
In the n-th move, the accountant plays a pair (
n
,
n
) with
n
p
n1
(
n
), [
n
[ = i
n1
, and a number b
n
.
Player spendthrift responds by playing a condition p
n
and a nite sequence
n
(letting i
n
:= [
n
[ + 1)
satisfying the following: (See Figure 3)
(1) p
n

in1
p
n1
.
(2)
n
p
n
(
n
)
(3) |
n
|
pn(
n
)
> b
n
.
(4)
n

n
.
(5) For all dom(p
n
) dom(p
n1
), [stem(p
n
())[ > [
n
[.
(6) [Level
|
n
|
(p
n
)[ = [Level
|
n
|
(p
n
)[ = [Level
|
n
|
(p
n1
)[
(Remember that all conditions p
n
have to be in the dense set given by (I) and (II)) Player accountant
wins i after many moves there is a condition q such that for all n, p
n
q, or equivalently, if the
function q with domain

n
dom(p
n
), dened by
q() =
_
dom(pn)
p
n
()
is a condition. Note that we have
l
(q) =
l
,
l
(q) =
l
, since the only splitting points are the ones
chosen by spendthrift.
3.20 Fact: Player accountant has a winning strategy in G(P, p).
14 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


n
n
n
n+1

i
n-1
i
p
p
n
n-1
Figure 3: stage n
Proof: We leave the proof to the reader, after pointing out that a nitary bookkeeping will ensure that the
limit of the conditions p
n
is in fact a condition.
In particular, this shows that spendthrift has no winning strategy. Below we will dene various strategies
for the spendthrift, and use only the fact that there is a play in which the accountant wins.

i
3.20
The game gives us the following lemma:
3.21 Lemma: Assume that p is a condition satisfying (I)(II). For each l let ,= F

l
succ
p(
l
)
(
l
) be a
set of norm |F

l
|
k
l

_
_
succ
p(
l
)
(
l
)
_
_
/2.
Then there is a condition q p, dom(q) = dom(p) such that for all l:
() If
l
(p) q(
l
(p)), then succ
q(
l
(p))
(
l
(p)) F

l
Proof: The condition q can be constructed by playing the game. In the n-th move, spendthrift rst nds
a
n

n
satisfying
n
(i) F
i
whenever this is applicable, and
_
_
succ
pn1
(
n
)
_
_
> 2b
n
. Then spendthrift
obtains p
n
by pruning (see 3.18) all splitting nodes of p
n1
whose height is between [
n
[ and [
n
[ and further
thinning out the successors of
n
to satisfy succ
pn
(
n
) = F

n. (Note that F

n succ
pn1
(
n
) = succ
p0
(
n
).)
In the resulting condition q the only splitting nodes will be the nodes
n
, so () will be satised.

i
3.21
(Note that in general
l
(q) ,=
l
(p), and indeed k
l
(q) ,= k
l
(p), since many splitting levels of p are not splitting
levels in q anymore.)
3.22 Lemma: Assume

is a P-name of a function from to , or even from into ordinals. Then the

set of conditions satisfying (I)(III) is dense and almost open.
15 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


III Whenever k is a splitting level, then every in level k + 1 decides

k.
Proof of (III): We will use the game from 3.19. We will dene a strategy for the spendthrift ensuring that
the condition q the accountant produces at the end will satisfy (III).
In the n-th move, spendthrift nds a condition r
n

in1
p
n1
such that for every Level
in1
(r
n
) the
condition (p
n
)
[ ]
decides

i
n1
+ 10. Then spendthrift nds
n
r
n
(
n
) satisfying the rules and obtains
p
n
with
n
p
n
(
n
) from r
n
by pruning all splitting levels between i
n1
and [
n
[.

i
3.22
Since all levels of q are nite, it is thus possible to nd a nite sequence

B = B
k
: k ) in the ground
model that will cover

. (I.e. q

(k) B
k
). The rest of this section will be devoted to nding small
such sets B
k
.
3.23 Corollary: P is

-bounding and does not collapse
1
.

i
3.23
3.24 Remark: Although it does not literally follow from 3.22, the reader will have no diculty in showing
that P is actually -proper for any <
1
.

i
Indeed, using the partial orders _
n
from 2.7, it is possible
to carry out straightforward fusion arguments, without using the game 3.19 at all. However, the orderings

n
are more easy to handle, since in induction steps we only have to take care of a single
n
, instead of a
front.
3.25 Fact:
P


B , B countable, B V , and V [GB].
Proof: Let p be any condition and let

be a name for a real. There is a stronger condition q satisfying

(I), (II) and (III). Let B := dom(q). Clearly q

V [GB].

i
3.25
3.26 Corollary: If = [A[

, then
PA
2
0
.
Proof: For each countable subset B A,
PB
CH. Since every real in V [G] is in some such V [GB], the
result follows.

i
3.26
3.27 Fact and Notation: If p satises (II), then
(1) If (
l
) =
l
, and succ
p(
l
)
(
l
), then the requirement

+
(
l
) =
uniquely denes an extension
+
of in Level
k
l
+1
(p).
(2) If (
l
) ,=
l
, has a unique extension
+
Level
k
l
+1
(p). To simplify the notation in 3.33
below, we also dene for this case, for any succ
p(
l
)
(
l
),
+
:=
+
.
3.28 Fact: The set of conditions satisfying (IV) is strictly dense (but not almost open) in the set of
conditions satisfying (I)(II).
IV For all l:
[Level
k
l
(p)[ < min
_
|p|
k
l
2
, n

k
l
_
For the proof, note that [Level
k
l
(p)[ = [Level
k
l1
+1
(p)[.

i
3.28
16 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


3.29 Lemma: Assume

is a P-name of a function

, and
P
k

(k) < n
+
k
. Then the set of
conditions satisfying (V) is strictly dense and almost open in the set give by (I), (II), (III). where
V Whenever k is a splitting level, then every in level k decides

k.
Proof: Fix p satisfying (I), (II), (III), (IV).
Let k
l
:= k
l
(p), etc. Let m
l
:= [Level
k
l
[.
Proof: We will use 3.21. For each l , F

l
succ
p(
l
)
(
l
) will be dened as follows: Let m
l
:= [Level
k
l
(p)[,
and let
0
, . . . ,
m1
enumerate Level
k
(p). Find a sequence
succ
p(
l
)
(
l
) = F
0
F
1
F
m
i
_
_
F
i+1
_
_
k

_
_
F
i
_
_
k
1
such that for all i there exists x
i
such that for all F
i+1
we have p
[(
i
)
+
]

k = x. It is possible to
nd such F
i+1
since ||
k
is n

k
-complete, and there are only n
+
0
n
+
1
n
+
k1
< n

k
many possible values of

k.
Finally, let F

l
:= F
m
. Applying 3.21 will yield the desired result.

i
3.29
3.30 Remark: Note that (V) in particular implies
Va Whenever k is not a splitting level, then every in level k decides

(k).
3.31 Proof that
P
c(f

, g

: (This proof is essentially the same as 2.12.)

Recall that r

is the generic real added by the forcing Q

. Working in V [G], let B be a family of less than

many g

-slaloms. We will show that they cannot cover

f

, by nding an such that r

is forced not
to be covered.
There exists a set A V of size <

such that B V [GA]. Since [A[ <

there is A

A.
Assume that

B is a g

-slalom in V [GA] covering r

. So in V there are a PA-name

B

and a condition p
such that

PA

B

is a g-slalom
and
p
P

B

covers r

We can nd a node in p() with succ

p()
() having more than g([[) elements. Increase pA to decide
B

||
, then increase p() to make r

avoid this set.

i
3.31
3.32 Fact: Fix

. Then the set of conditions p satisfying

VI For all l: If

<

l
(p)
, then
min
_
f

l
(p)
(k
l
)
g

(k
l
)
,
f

(k
l
)
g

(k
l
)
_
h

l
(p)
(k
l
)
_
<
1
[Level
k
l
(p)[
is dense almost open.
17 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Proof: Write F

for the function min

_
f

,
f

_
h

_
. Recall that if

<

, then F

tends to 0.
Fix a condition p, We will use the game G(P, p). spendthrift will use the following strategy: Whenever

n
A

and

<

, then spendthrift rst nd m

0
such that for all m m
0
we have F

(m) <
1/[Level
hn1
(p
n1
)[. Now nd
n

n
of length > m
0
with a large enough norm, and play any condition
p
n
obeying the rules of the game. In particular, we must have

Level
|
n
|
(p
n
)

Level
|
n
|
(p
n
)

.
Clearly the condition resulting from the game satises the requirements.

i
3.32
3.33 Proof that
P
c(f

, g

: Fix . We will write f for f

, etc.
Let
A :=
_
A

.
We will show that the g-slaloms from V
PA
already cover

f. This is sucient, because
P
(2
0
)
V
PA

[A[ =

.
Let p
0
be an arbitrary condition. Let

be a name of a function < f. Find a condition p p

0
satisfying
(I)(VI).
For each l we now dene sets F

l
succ
p(
l
)
(
l
) as follows:
(1) If
l
A, then F

l
= succ
p(
l
)
(
l
).
(2) If f

l
(k
l
) g

(k
l
)/[Level
k
l
(p)[, then again F

l
= succ
p(
l
)
(
l
).
(3) Otherwise, we thin out the set succ
p(
l
)
(
l
) such that each in Level
k
l
(p) decides

(k
l
) up
to at most g(k
l
)/[Level
k
l
(p)[ many values.
Here is a more detailed description of case (3): Let k = k
l
, =
l
.
Note that if neither (1) nor (2) holds, then letting c := f

(k), d := g

(k)/[Level
k
(p)[, we have c/d h

(k).
Using (c, d)-completeness of the norm ||
,k
we dene a sequence
succ
p(
l
)
(
l
) = L(0) L(1) L([Level
k
(p)[)
as follows. Let
0
, . . . ,
|Level
k
(p)|1
be an enumeration of Level
k
(p).
Given L(i), we know that for each L(i) the sequence
+
i
, (i.e., the condition p
[
+
i
]
) decides

(k).
(See 3.27.) since there only c many possible values for

(k), we can use (c, d)-completeness to nd a set

L(i + 1) L(i) and a set C(i) such that
(a) |L(i + 1)| |L(i)| 1
(b) [C(i)[ d.
(c) For every L(i + 1), p
[
+
i
]

(k) C(i).
Now let F

l
be L([Level
k
(p)[), and let
() B
k
:=
_
i
C(i).
So [B
k
[ [Level
k
(p)[ d g(k).
18 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


Clearly |F

l
|

l
,k
l
|p|
k
l
[Level
k
l
(p)[ >
1
2
|p|
k
l
.
This completes the denition of the sets F

l
.
Let q p be the condition dened from p using the F

l
(see 3.21). We will nd a PA-name for a g-slalom

= B

k
: k ) such that
q

B

covers

.
If k is not a splitting level, then every in level k decides

(k) by (Va). So in this case we can let

B
k
:= i : Level
k
(p), p
[ ]

(k) = i
This set is of size [Level
k
(p)[ < g(k), and clearly q

(k) B
k
.
If k is a splitting level, k = k
l
, then there are three cases.
Case 1:
l
A: We dene B

k
to be a PA-name satisfying the following:

PA
B

k
= i : Level
k+1
(p), V [= p
[ ]

(k) = i, (
l
) r

Thus, we only admit those which agree with the generic real added by the forcing Q

l
. Clearly
PA
[B
k
[ Level
k
(p) < g(k), and p
P

(k) B
k
.
Case 2: f

l
(k) g

(k)/[Level
k
(p)[.
So we have [Level
k+1
(p)[ f

l
(k) [Level
k
(p)[ g(k), so we can let
B
k
:= i : Level
k+1
(p), p
[ ]

(k) = i
This set is of size [Level
k+1
(p)[ g(k), and again p

(k) B
k
.
Case 3: Otherwise. We have already dened B
k
l
in (). By condition (c) above, q

(k) B
k
.
So indeed q

B

= B

k
: k ) is a g-slalom covering

i
3.33
i
3.1
i
[GSh 448]
REFERENCES
[Blass] A. Blass, Simple cardinal invariants, preprint.
[van Douwen] E.K. van Douwen, The integers and Topology, in: Handbook of Set-Theoretic Topology, ed.
by K. Kunen and J.E. Vaughan, North-Holland, Amsterdam-New YorkOxford 1984
[Comfort-Negrepontis] Comfort and Negrepontis, Theory of ultralters, Springer Verlag, Berlin Heidelberg
New York, 1974.
[Miller] A. Miller, Some properties of measure and category, Transactions of the AMS 266.
19 Revised version, April 1992
4
4
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
9







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
9


[Shelah 326] S. Shelah, Vive la dierence!, to appear in: Proceedings of the MSRI Logic Year 1989/90,
ed. by H. Judah, W. Just, W. H. Woodin.
[Shelah 448a] S. Shelah, Notes on many cardinal invariants, May 1991.
[Vaughan] J.E. Vaughan, Small uncountable cardinals and topology, in: Open problems in Topology, ed.
by J. van Mill and G. Reeds.
Martin Goldstern Saharon Shelah
Dept of Mathematics Dept of Mathematics
Bar Ilan University Hebrew University
52900 Ramat Gan Givat Ram, Jerusalem
goldstrn@bimacs.cs.biu.ac.il shelah@math.huji.ac.il
20 Revised version, April 1992