7

2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS
HEIKE MILDENBERGER AND SAHARON SHELAH
Abstract. Given a binary relation R, we call a subset A of the range of R
R-adequate if for every x in the domain there is some y A such that
(x, y) R. Following Blass [4], we call a real needed for R if in every
R-adequate set we nd an element from which is Turing computable.
We show that every real needed for inclusion on the Lebesgue null sets,
Cof(N), is hyperarithmetic. Replacing R-adequate by R-adequate with
minimal cardinality we get the related notion of being weakly needed.
We show that it is consistent that the two notions do not coincide for the
reaping relation. (They coincide in many models.) We show that not all
hyperarithmetic reals are needed for the reaping relation. This answers
some questions asked by Blass at the Oberwolfach conference in December
1999 and in [4].
0. Introduction
We consider some aspects of the following notions:
Denition 0.1. (1) (Needed reals). Suppose that we have a cardinal char-
acteristic x of the reals of the following form: There are (in most cases:
Borel) sets A

, A
+
R and there is a (in most cases: Borel) relation
R A

A
+
such that
x = [[R[[ := min[Y [ : Y A
+
(x A

)(y Y )R(x, y).

We call [[R[[ the norm of R. A set Y A
+
is called R-adequate if
(x dom(R)) (y Y )R(x, y). We say that

2 is needed for R if
for every R-adequate set Y there is some y Y such that is Turing
reducible to y, abbreviated
Tur
y.
If A
+
, R but can be mapped continuously, injectively into R by
a mapping c which is, as a function on the digits, computable in both
directions, then we call the real needed for R and c if for any R-
adequate set Y A
+
there is some y Y such that
Tur
c(y). We
2000 Mathematics Subject Classication. 03E15, 03E17, 03E35, 03D65.
The rst author was supported by a Minerva fellowship and by grant 13983 of the Austrian
Fonds zur wissenschaftlichen Forderung.
The second authors research was partially supported by the Israel Science Foundation,
founded by the Israel Academy of Science and Humanities. This is the second authors work
number 725.
1
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


2 HEIKE MILDENBERGER AND SAHARON SHELAH
call such a function c a coding. In this situation, a real is called needed
for R, if it is needed for R and c for any coding c.
(2) (Weakly needed Reals). We call a real weakly needed for R if for any
R-adequate set Y of minimal cardinality there is some y Y such that

Tur
y.
Every needed real is weakly needed. Sections 4 to 7 will give some information
on the reverse direction. A very good motivation for the investigation of needed
reals is given in [4].
In the rest of this introduction, we describe briey what will be proved in the
sections. In Section 1 we prove that only hyperarithmetic reals are needed for
the conality relation on the ideal of Lebesgue null sets. In the second section
we prove the analogous statement for the slalom relation. In the third section
we extract from these two results sucient conditions for the property every
needed real for R is hyperarithmetic. In the fourth section we construct a
forcing extension such that all hyperarithmetic reals are weakly needed for the
reaping relation in the extension. This quite dicult model is further used in the
fth section, where we build a composed relation for which there are more weakly
needed reals than needed reals. In Section 6 we prove that all needed reals for
the reaping relation are in complexity less than 0

= x, y) : x, y , x 0
y
,
where 0
y
is the yth jump of the degree 0 of all recursive sets. Moreover, using
the model of Section 4 once more, we get that it is consistent that for the reaping
relation weakly needed and needed do not coincide. In the nal section we give
a sucient criterion for a relation R such that the two notions needed for R
and weakly needed for R coincide. From the proof in Section 1, we derive
one example of a relation for which the criterion is true. The denitions of the
mentioned relations will be recalled at their rst appearance.
1. Needed reals for Cof(^)
In this section we answer armatively Blass question whether only hyper-
arithmetic reals are needed for the conality relation on the ideal of Lebesgue
null sets.
In this section we work with two particular relations on the reals: For functions
f, g : we write f

g and say g eventually dominates f if (n < )(k

n)(f(k) g(k)). The dominating relation is
D = (f, g) : f, g

f

g,
and the conality relation for the ideal of sets of Lebesgue measure zero is
Cof(^) = (F, G) : F, G are G

-sets of Lebesgue measure 0 and F G.

We write cof (^) for [[Cof(^)[[.
Before stating our rst theorem, we review some notation: For s
>
2 = r :
(m )(r : m 2), we write lg(s) = dom(s). If r

2 and s

2, we
write r s if r = s lg(r). Let r s denote that r s and r ,= s. A subset
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 3
T
>
2 is called a tree if it is downward closed, i.e., if for all s T for all r s,
we have that r T. An element r T is a leaf if there is no s T such that
r s. For a tree T
>
2 and some n we set T n = T
n>
2. For t
n
2
set t = f

2 : f n t. The set of innite branches of T is denoted by
lim(T) = f

2 : (n)(f n T). The same notation applies to trees on
>
H for an arbitrary set H. We will consider trees whose nodes are not nite
sequences but nite sets of nite sequences.
Leb denotes the Lebesgue measure on the measurable subsets of

2, the product
space of copies of the space 0, 1 where each point has measure
1
2
.
We work with the Amoeba forcing in the representation of [2, 3.4B]:
Q =
_
p : p
>
2, p is a tree without leaves and Leb(lim(p)) >
1
2
_
.
For trees p, q Q, q is a stronger forcing condition than p, abbreviated q p, if
q p. In addition to the Jerusalem convention, that stronger conditions are the
greater or equal ones, we also follow the alphabetical convention [6]: letters later
in the alphabet or carrying more primes or stars are used for stronger conditions.
The weakest element in Q is
>
2, and we write 1 for it. We write G for some
Q-generic lter and T =

G for the generic tree. For x V [G], let x

denote a
name of x.
Denition 1.1. Q
1
is the set of conditions p Q that fulll:
(1.1) for all n < , s p
n
2, Leb(s lim(p)) ,=
1
2
.
Claim 1.2. Q
1
is dense in Q.
Proof. Let p Q and Leb(lim(p)) =
1
2
+ . Let s
n
, n < , be an enumeration
of

i<
T(
i
2). Now choose by induction on n p = p
0
p
1
p
2
. . . in Q
such that Leb(lim(p
n
))
1
2
+ (1

j<n
1
2
j+2
). We set
0
= . In step n,
we set
n
:= min(Leb(s
i
lim(p
n
))
1
2
: i < n Leb(s
i
lim(p
n
))
1
2
>
0
n1
) and choose p
n+1
p
n
such that Leb(s
n
lim(p
n+1
)) ,=
1
2
and such
that Leb(lim(p
n+1
)) Leb(lim(p
n
))
n
2
n+2
. Then automatically also Leb(s
i

lim(p
n+1
)) ,=
1
2
for i < n and once property (1.1) is true for a condition p
n
and s
i
, i < n, it holds also for all later p
k
because we chose the p
k
s such that
the dierences in their measures are suciently small. Then by the choices,
q =

n<
p
n
Q
1
.
The following denition is crucial for building an algorithm that uses the oracle
T already in V . For this purpose we require: incompatibility of a nite part of
T with a condition p can be read o a nite part of p (this is (b)), that measure
1
2
is forbidden in a preciser way than in equation (1.1) (this is (c)), and that the
convergence from above of
|p
k
2|
2
k
: k ) to Leb(lim(p)) is suciently fast (this
is (d)).
Denition 1.3. We say p obeys g if the following holds:
(a) p Q
1
, g

, (n)(n < g(n)).

7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


4 HEIKE MILDENBERGER AND SAHARON SHELAH
(b) If n < , s p
n
2 and Leb(slim(p)) <
1
2
, then
|{
g(n)
2 : ns,p}|
2
g(n)
<
1
2
.
(c) If n < , s p
n
2 and Leb(s lim(p))
1
2
, then Leb(s lim(p))
1
2
_
1 +
1
g(n)
_
.
(d) If n < , then Leb(lim(p))
|p
g(n)
2|
2
g(n)
_
1
1
2
2
n
_
.
The main part of the section will be the proof of
Theorem 1.4. Suppose that V

, that M V is a Turing machine and

that p obeys g and
(1.2) p
Q
M computes from T

.
Then is computable from g.
Proof. We x such an .
Fix for a while j . Since the statement for every j there is some j

such
that M computes (j) using T j

is forced, there is some stronger condition

r that forces a value for j

for the xed j. So [r j

] could serve as a condition

that describes enough of the oracle T in order to give the right computation of
(j).
Now the assigment j j

(say the minimal one), is an element of V

Q
, and
in general not in V . But since our computation is not allowed to use additional
information except for g, we will look, given j, at all possible rs and j

s simul-
taneously. The procedure to give a computation in V will be built upon guessing
nite parts of conditions r and nite parts of T that are already determined by
the same nite part of r. But, such an approximation, starting with trials of
size zero and successively increasing the size, could give a unique (and, of course,
halting) computation that gives the same outcome on all possibilities within the
guessed part and still be not the right guess because a too small part of r is used
and only a larger approximation would mirror correctly what happens in the
forcing process. However, fortunately from some approximation size onwards,
the outcome will not change any more. So we can remedy the problem of wrong
guesses by rst choosing a suitable n() and then looking into densely many
forcing conditions above p
n()
2 simultaneously, and search increasing in m for
an aproximation of size m. Starting from some m, all larger approximations will
give the same result. The search will be based upon g. And, from the denition
of p obeys g it follows that any eventually larger function could serve as well.
We assign some structure to the collection of nite initial segments of members
of Q
1
, that will allow us to work with nitely branching trees. These will be the
trees (T
n()
p,g
, ) from Denition 1.7. The procedure that computes relative to
g will rst search for a suciently large nite approximation of r, and then argue
that this approximation already determines the run of M with oracle T on the
given input j. The height of this approximation in T
n()
p,g
will be an appropriate
measure for being a suciently large approximation of r.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 5
Denition 1.5. A set t
m
2 is a subtree of
m
2 i t is not empty and closed
under initial segments and ( t
m>
2)(e 2)(e) t).
Denition 1.6. We say that a subtree t of
m
2 obeys g if the following holds:
(a) t is a subtree of
m
2, g

, (n)(n < g(n)).

(b) If n m, g(n) m, s t
n
2, and if Leb(st) <
1
2
, then
|{
g(n)
2 : ns,t}|
2
g(n)
<
1
2
.
(c) If n m, g(n) m, s t
n
2, and if Leb(s t)
1
2
, then Leb(s t)
1
2
_
1 +
1
g(n)
_
.
(d) If g(n) m, then
[t
g(n)
2[
2
g(n)

[t
m
2[
2
m

[t
g(n)
2[
2
g(n)
_
1
1
2
2
n
_
(note that the rst inequality holds trivially).
The properties t obeys g and p obeys g look similar. In order to decribe the
features of the similarities that will be useful later we make another denition:
Denition 1.7. Assume that p obeys g and n() < and that m n().
(1) Nb
n()
p,g
= q Q
1
: q (n() + 1) = p (n() + 1) and q obeys g.
(2)
m
T
n()
p,g
= ,= t
m
2 : t is a subtree of
m
2 and t obeys g and t
(n() + 1) = p (n() + 1).
(3) We let T
n()
p,g
=

0m<
m
T
n()
p,g
.
(4) If t T
n()
p,g
, then let m(t) be the minimal m such that t
m
2.
(5) The partial order on T
n()
p,g
is dened it as follows: s t i t
(m(s) + 1) = s.
Our rst claim about the trees T
n()
p,g
and the neighborhoods Nb
n()
p,g
concerns
the easier inclusion: members of the neighborhoods can be seen as branches of
the trees:
Claim 1.8. (1) T
n()
p,g
is a tree whose m-th level is
m
T
n()
p,g
. If 0 m m(t)
and t T
n()
p,g
then t (m + 1) t and t (m + 1)
m
T
n()
p,g
.
(2) If q Nb
n()
p,g
and m < then q (m + 1)
m
T
n()
p,g
.
In the next claim some easy and useful facts are listed.
Claim 1.9. (1) Every p Q
1
obeys some g.
(2) q lim(T
n()
p,g
) i (m)q (m + 1)
m
T
n()
p,g
.
(3) If p Q
1
obeys g and n() then Nb
n()
p,g
Q
1
.
(4) If g
1

g
2
then there is g
1
2
recursive in g
2
such that g
1
g
1
2
, where
is the pointwise order.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


6 HEIKE MILDENBERGER AND SAHARON SHELAH
(5) If g
1
g
2
and p obeys g
1
, then p obeys g
2
.
(6) Fixing p, n(), the function m
m
T
n()
p,g
is recursive in g.
The next claim will allow us to apply K onigs Lemma at an important step in
the proof of Claim 1.13.
Claim 1.10. Suppose that
(1.3) (k)(g(k) n() Leb(lim(p))
_
1
1
2
2
k1
_
>
1
2
).
Then Nb
n()
p,g
is the set of unions of -branches of the tree T
n()
p,g
.
Proof. Suppose that for all m , q (m + 1)
m
T
n()
p,g
. We prove that
q Nb
n()
p,g
. We rst prove that q Q. From the denition of subtree of
m
2
it follows that q has no leaves.
The main point is: Why is Leb(lim(q)) >
1
2
? For every m g(k), we have by
clause (d) of Denition 1.6
[q
g(k)
2[
2
g(k)

[q
m
2[
2
m

[q
g(k)
2[
2
g(k))
_
1
1
2
2
k
_
.
But as g(k) n() clearly q
g(k)
2 = p
g(k)
2. As p obeys g we have
Leb(lim(p))
[p
g(k)
2[
2
g(k)
_
1
1
2
2
k
_
.
Since the quotients are approaching the measure from above, the right side is
greater than or equal to Leb(lim(p))
_
1
1
2
2
k
_
. So
|q
m
2|
2
m
Leb(lim(p))
_
1
1
2
2
k
_
and this holds for every m . Hence Leb(lim(q)) Leb(lim(p))
_
1
1
2
2
k
_
and
the right hand side is strictly larger than
1
2
by equation (1.3).
Now we have to prove that q Q
1
. So let n and s q
n
2. Suppose that
Leb(s lim(q)) =
1
2
. Then for all m, Leb(s q
m
2)
1
2
. So by 1.6(b) for all
m, Leb(s q
m
2)
1
2
_
1 +
1
g(n)
_
. Hence also the limit is greater than or equal
to
1
2
_
1 +
1
g(n)
_
.
Now we have to prove that q obeys g. This follows from Denition 1.6 and the
nature of the limit process.
The reverse inclusion is Claim 1.8(2).

Denition 1.11. Assume that T is a Turing machine. We let

M,k
be the set of
nite partial characteristic functions h, dom(h)
>
2 such that if M runs with
the input k it uses only h as an oracle. So it answers h() =? for dom(h)
according to h and does not ask questions of the kind h() =? for , dom(h).
We let e
M,h
(k) be the result of such a run.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 7
Denition 1.12. For p Q set
(p) = h : (m)(h:
m
2 2 p , h , ch
T

).
For t
m
T
n()
p,g
let
(t) =
_
h : (n)
_
h:
n
2 2 g(n) m h
1
(1) is a tree
h
1
1 t
[
g(n)
2 : h( n) = 1 t[
2
g(n)
>
1
2
__
.
Claim 1.13. (1) For every p Q and nite u
>
2 there is some h (p)
whose domain is a superset of u.
(2) If p Q
1
obeys g and h:
n
2 2 then p
g(n)
2 determines the truth
value of h (p), in fact h (p) i h (p (g(n) + 1)).
Proof. (2) By looking at the relation
Q
we see: For h:
n
2 2, p h , ch
T

i (h
1
(1) , p or (h
1
(1) p and Leb(h
1
(1) lim(p))
1
2
)). But the
right hand side of the i-clause can be read o p (g(n) +1) by clause (b) of the
Denition of p obeys g.
Now we are ready to nish the proof of Theorem 1.4. So assume that p Q
1
,


2, p M computes from the oracle T

and p obeys g. We show that

is computable from g.
Step a) Let k > 0 be such that
1
2
< Leb(lim(p))
_
1
1
2
2
k
1
_
.
Step b) Let n() g(k).
Step c) Nowwe have for every q Nb
n()
p,g
and j < that (q)(p)
M,j
,= .
Why?
First, by the choice of n(), Leb(lim(p) lim(q)) >
1
2
. This follows from the
computation: lim(p) p
n()
2 p
g(k)
2 and the same holds for q. Since both
obey g we have lim(p) >
|p
g(k)
2|
2
g(k)
_
1
1
2
2
k
_
and the same holds for q. Hence
lim(p) and lim(q) each are missing only a part of p
g(k)
2 of measure less than
|p
g(k)
2|
2
g(k)+2
k
. The missing part of lim(p) lim(q) in p
g(k)
2 is at most twice this.
Hence Leb(lim(p) lim(q))
|p
g(k)
2|
2
g(k)
_
1
1
2
2
k
1
_
Leb(lim(p))
_
1
1
2
2
k
1
_
>
1
2
.
So there is some r Q that is above both. As r p, it forces that M running
on j and oracle T

gives (j) and the run uses only h = T

n
2 for some n.
Step d) If e
M,h
(j) is well-dened and h (p), then e
M,h
(j) = (j). This is
because V .
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


8 HEIKE MILDENBERGER AND SAHARON SHELAH
Step e) For every j there is some m, such that for all t
m
T
n()
p,g
we have
that (t) (p)
M,j
,= . Why? This follows by Claim 1.10 and step c) and
K onigs lemma applied to the nitely branching tree T
n()
p,g
.
Step f) For every j there are m and t
1

m
T
n()
p,g
such that for every t
2

m
T
n()
p,g
we have (t
1
)(t
2
)
M,j
,= . This holds by e): We can take t
1
= p (m+1).
Step g) For every j < and e 2 the following are equivalent
(i) (j) = e.
(ii) For some m < and t
1

m
T
n()
p,g
for every t
2

m
T
n()
p,g
there is h
(t
1
) (t
2
)
M,j
such that e
M,h
(j) = e.
(i) (ii) is step f).
(ii) (i): Let m, t
1
be as in (ii) Let t
2
= p (m + 1). By (ii) there is some
h (t
1
) (t
2
)
M,j
, e
M,h
(j) = e. By step b) h (p). So by step d) we
are done.
By Claim 1.9(6), the procedure indicated in (ii) of step g) is recursive.
So nally Theorem 1.4 is proved.
Corollary 1.14. Suppose that p Q and p
Q
is computable from T

. Then
is needed for the dominating relation.
Proof. Choose some q p, q Q
1
and some machine M such that q
Q
M computes from T

. Then choose g such that q obeys g. By Theorem 1.4,

is computable from g. Since q obeys also every g

g and since nite changes

in the oracle may only change the algorithm but not the fact whether a real is
computable from the oracle, is also computable from any g

g. Hence is
needed for the dominating relation.
The following equivalent formulation of neededness is useful to show that in a
generic extension that contains a real such that for all reals in the ground
model R, all needed reals in the ground model can be computed from such a
.
Fact 1.15. (Blass [3, following Denition 1]) An equivalent condition for is
needed for R is
(1.4) (x dom(R))(y range(R))(xRy
Tur
y).
Proof. Suppose that is needed for R and that there is no x as in (1). Then
(x dom(R)) (y range(R))(xRy ,
Tur
y). So we can build a R-adequate
set from all these ys, that shows that is not needed for R. For the other
implication: Fix x as in (1). Every R-adequate set has to contain some y such
that xRy and hence
Tur
y.
If R is a transitive relation and is a given R-adequate set, then x with the
property of equation (1.4) can be found in .
Theorem 1.16. Every needed real for Cof(^) is needed for the dominating
relation.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 9
Proof. : Let A
i
: i < be a Cof(^)-adequate set, such that each A
i
is a F

set. Let

2.
For each i choose a countable elementary submodel N
i
of (H(
3
), ) to which
and A
i
belong. We let G
i
be a subset of Q
Ni
that is generic over N
i
and let
T
i
= T

[G
i
]. Now let A

i
be
A

i
=

2 : no

2 which is almost equal to

(i.e. (n) =

(n) for every large enough n) belongs to lim(T

i
)
A

i
is a null set: Since it is a tail set, by the zero-one law it can only have
measure zero or one. Since it is disjoint from the set lim(T
i
), that has measure
one half, it is a null set.
By genericity of T
i
and because A
i
N
i
and because A
i
is a nullset in N
i
we have that A
i
lim(T
i
)
c
. The same argument shows that for all s
>
2 we
have that sf : s

([s

[ = [s[ s

f A
i
) is a subset of (lim(T
i
))
c
. Hence we
have that A
i
A

i
. Therefore also A

i
: i < is a Cof(^)-adequate set. We
choose i such that is recursive in A
i
and in all its supersets. Since Cof(^) is
transitive, such an A
i
exists by Fact 1.15 and the remark thereafter. Then is
also recursive in A

i
, because A

i
A
i
. If is recursive in A

i
it is also recursive
in T
i
. Since this holds for arbitrary G
i
, by Theorem 1.4, applied to some p Q
1
that obeys some g, the real is needed for dominating.
Fact 1.17. a) (Solovay [11]) Every real that is needed for the dominating relation
is hyperarithmetic.
b) (Jockusch, [8]) Every hyperarithmetic real is needed for the dominating re-
lation.
Blass [4, Theorem 6, Corollary 8] showed that every real that is needed for D
is also needed for Cof(^) and hence that all hyperarithmetic reals are needed
for Cof(^). So this gives the other inclusion in the following corollary:
Corollary 1.18. Exactly the hyperarithmetic reals are needed for the Cof(^)-
relation.
2. Needed reals for the slalom relation
In this section we deal with a forcing L which is closely related to the local-
ization forcing from [2, page 106]. Theorem 2.4 is analogous to Theorem 1.4,
but for the forcing L. Theorem 2.16 is analogous to Theorem 1.16 together with
Corollary 1.18.
For u = u

: ), m < , let u n = u

: < n).
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


10 HEIKE MILDENBERGER AND SAHARON SHELAH
Denition 2.1.
L = p : p = (n, u) = (n
p
, u
p
), u = u

: ), u

[]

,
[u

[ : ) is bounded,
p q
_

u
p

u
q

u
q
n
p
= u
p
n
p
n
p
n
q
_
.
Again we denote the weakest element (0, : i < )) of L by 1. For p L, we
write b(p) = max[u

[ : .
Notation 2.2. Let G

be a name for an L-generic element. Let S = S

G
=

u
p

n
p
: p = (n
p
, u
p
) G. We write S

for a L-name for S. We think of S as a

subset of and have its characteristic function ch
S
(, m) = 1 i m u

.
Denition 2.3. We say p = (n, u) L obeys (g, b) if ( < )( < g()), b ,
()(u
p

g()) and ()([u

p

[ b). Note that the condition 1 obeys every (g, b).

Theorem 2.4. Assume that M is a Turing machine and that

2. Suppose
that p L obeys (g, b) and
(2.1) p
L
M computes from S

.
Then is computable from g.
Proof. As in the proof of Theorem 1.4, we use (2.1) and the fact that p obeys
(g, b) in order to nd densely many conditions above p and nite approximations
of ch
S
and of the respective condition. We will keep the numbering of the claims
and of the denitions used in the proof of Theorem 2.4 parallel to the numbering
in the proof of Theorem 1.4, though many of them are much easier for L and will
be almost obsolete or empty. But this procedure will help to establish a general
scheme.
Denition 2.5. A tuple (n, u

: < m)) is a nite part of a condition i

n m and for all < m, u

is a non-empty nite set.

Denition 2.6. A nite part of a condition (n, u

: < m)) obeys (g, b) i

( < m)(u

g() [u

[ b).
Now, in the following we do not only number analogously but also use similar
names Nb
n()
p,g,b
and
m
T
n()
p,g,b
for the corresponding objects. We use g so that the
T
n()
p,g,b
will be nitely branching, and we use b to get the boundedness clause in
the denition of a condition.
Denition 2.7. Let p = (n(), u
p
) be a condition that obeys (g, b).
(1) Nb
n()
p,g,b
= q = (n
q
, u
q
) L : n
q
n() (i < n())u
p
i
= u
q
i
(i)u
q
i

g(i) (i)[u
q
i
[ b. As the trees and neighborhoods in Denition 1.7
used only p n(), also here the part of u
p
above n() is ignored. The
algorithm will depend on g and on a nite part of p.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 11
(2) If m n(), let
m
T
n()
p,g,b
= (n, u
i
: i < m)) : n n(), (i <
n())(u
i
= u
p
i
) (i < m)(u
i
g(i) [u
i
[ b). If m < n(), let
m
T
n()
p,g,b
= (m, u
p
m).
(3) T
n()
p,g,b
=

m<
m
T
n()
p,g,b
.
(4) For (n, u
i
: i < m))
m
T
n()
p,g,b
we let m(n, u
i
: i < m)) = m.
(5) For (n, u), (n

, v) T
n()
p,g,b
we write (n, u) (n

, v) i u is an initial
segment of v and u n = v n.
Claim 2.8. (1) T
n()
p,g,b
is a tree whose m-th level is
m
T
n()
p,g,b
. If 0 m m(t)
and t T
n()
p,g,b
then t (m+ 1) t and t (m + 1)
m
T
n()
p,g,b
.
(2) If q = (n, u) Nb
n()
p,g,b
and n m < then (n, u m)
m
T
n()
p,g,b
.
Claim 2.9. (1) Every p L obeys some (g, b).
(2) q = (n, u) lim(T
n()
p,g,b
) i (m)(min(m, n), u m)
m
T
n()
p,g,b
.
(3) If p = (n(), u) L obeys (g, b) and n() then Nb
n()
p,g,b
L.
(4) If g
1

g
2
then there is g
1
2
recursive in g
2
such that g
1
g
1
2
, where
is the pointwise order.
(5) If g
1
g
2
and p obeys (g
1
, b), then p obeys (g
2
, b).
(6) Fixing p, n(), the function m
m
T
n()
p,g,b
is recursive in g.
Claim 2.10. Suppose that p L obeys (g, b). Then Nb
n()
p,g,b
is compact as a subset
of T(
>
2), and is the set of unions of -branches of the tree T
n()
p,g,b
.
Proof. This is obvious.
Denition 2.11. Assume that T is a Turing machine. We let
M,k
be the set
of nite partial characteristic functions h, dom(h)

2 such that if M runs
with the input k it uses only h as an oracle. So it does not ask questions of the
kind h() =? for , dom(h). We let e
M,h
(k) be the result of such a run.
Denition 2.12. For p L set
(p) = h : (m)(h: mm 2 p , h , ch
S

).
For t = (n(), u
i
: i < m))
m
T
n()
p,g
let
(t) = h : h: mm 2 (i < n())h
1
(1) (i ) = i u
i
(i [n(), m))h
1
1 (i ) i u
i
.
Claim 2.13. (1) For every p L and nite z there is some h
(p) whose domain is a superset of z.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


12 HEIKE MILDENBERGER AND SAHARON SHELAH
(2) If p = (n(), u)) L obeys (g, b), n() m, and h: mm 2 then t =
(n(), u m) determines the truth value of h (p), in fact h (p)
i h (t).
Proof. (2) By looking at
L
we see: For h: m m 2, p h , ch
S

i
(h
1
(1) ,

i<m
i u
i
or (i < n())(h
1
(1) (i ) ,= i u
i
)).
Now we nish the proof of Theorem 2.4: Assume p = (n(), u
i
: i )) L,


2, p M computes from the oracle S

and p obeys (g, b). We show that

is recursive from g.
We name the steps in parallel to the steps in the end of the proof of Theo-
rem 1.4. Since n() is already given, the rst two steps are empty.
Step c) Now we have that for every q Nb
n()
p,g,b
and j < that (q) (p)

M,j
,= . Why? It is easy to see that p and q are compatible in L. So there
is some r Q that is above both. As r p, it forces that M running on j and
oracle S

gives (j) and the run uses only h := ch

S

mm
2 for some m.
Step d) If e
M,h
(j) is well-dened and h (p), then e
M,h
(j) =

(j).
Step e) For every j there is some m, such that for all t
m
T
n()
p,g,b
we have such
that (t)(p)
M,j
,= . Why? this follows by c) and K onigs lemma applied
to the nitely branching tree T
n()
p,g,b
.
Step f) For every j there are m and t
1

m
T
n()
p,g,b
such that for every t
2

m
T
n()
p,g
we have (t
1
) (t
2
)
M,j
,= . This holds by e): We can take
t
1
= (n(), u
i
: i < m)).
Step g) For every j < and e 2 the following are equivalent
(i) (j) = e.
(ii) For some m < and t
1

m
T
n()
p,g,b
for every t
2

m
T
n()
p,g,b
there is h
(t
1
) (t
2
)
M,j
such that e
M,h
(j) = e.
(i) (ii) is step f).
(ii) (i): Let m, t
1
be as in (ii) Let t
2
= (n(), u m). By (ii) there is some
h (t
1
) (t
2
)
M,j
, e
M,h
(j) = e. By step b) h (p). So by step d) we
are done.
By Claim 2.9(6), the procedure in (ii) of step g) is recursive.
Corollary 2.14. Suppose that p L and p
Q
is computable from S

. Then
is needed for the dominating relation.
Proof. Choose some q p, q L and some machine M such that q
Q
M
computes from S

. Then choose (g, b) such that q obeys (g, b). By Theorem 2.4,
is computable from g. Since q obeys also every (g

, b) for g

g and since nite

changes in the oracle may only change the algorithm but not the fact whether a
real is computable from the oracle is also computable from any g

g. Hence
is needed for the dominating relation.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 13
Denition 2.15. S

([]
<
) is called a slalom i for all n, [S(n)[ n.
Theorem 2.16. Exactly the hyperarithmetic reals are needed for the slalom re-
lation
SL = (f, S) : f

S is a slalom and (n )(f(n) S(n)).
Proof. First we show that only hyperarithmetic reals are needed for SL: Let
S
i
: i < be an SL-adequate set,

2 V be needed for SL. We take
N
i
(H(
3
), ) such that , S
i
N
i
. Then we let G
i
be L-generic over N
i
. Now
we set S

i
= S
Gi
=

u
p
n
p
: p G
i
. Then we have that for all but nitely
many n, S
i
(n) S

i
(n). Let be computable from S
i
. Then for all wider slaloms
S

i
than S
i
there is a (possibly even) wider slalom from which is computable as
well, because that collection of slaloms wider than S

i
is SL-adequate. Then by
density
Tur
S
Gi
for all generic G
i
. Hence we may choose p and M and (g, b)
such that Theorem 2.4 applies.
All hyperarithmetic reals are needed for SL, because all of then are needed
for D. Suppose that S
i
n
: n ) : i < is SL-adequate and that

2
is hyperarithmetic. Then T = max S
i
n
: n ) : i < is D-adequate and
hence there is some element max S
i
n
: n ) T from which is computable.
But then of course is also computable in S
i
n
: n ).
3. A general connection
In this section, we collect sucient conditions and give a general scheme for the
proofs of every real needed for R is hyperarithmetic. As in Theorems 1.4 and
2.4 we use a forcing that adds an R-dominating real . The rst step is to prove
that being computable in and being in V implies being hyperarithmetic. A
form of this step will be given in Theorem 3.1. The second step is to show that
every needed real for R is computable from any generic . We write a general
version of this step in Theorem 3.6.
Now we take (Q, R) instead of our two examples (Q, Cof (^)) and (L, SL). R
is a Borel binary relation on the reals, and Q is a notion of forcing adding some
element in the range of the extension of R. Since R is Borel, we can use its code
and thus get a unique extension of R to a larger model of ZFC. From the work
in the previous two sections we collect the following scheme:
Theorem 3.1. Assume that
(a) There is a notion p obeys g such that if g

g and p obeys g then p

obeys g

. For a dense subset Q

of Q we have (p Q

)(g)(p obeys g).

(b) T
p,g
is a recursive nitely branching tree whose nodes are nite functions.
(c) Q is a forcing notion, p Q, p obeys g, is a Q-name, and
p
Q
lim(T
p,g
).
(d) For a dense set of p
0
Q

there is some p p
0
such that the following
conditions are fullled:
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


14 HEIKE MILDENBERGER AND SAHARON SHELAH
() Let (p) = T
p,g
: p , , . This is a subtree of T
p,g
.
() Let S

p,g
=
_
t : for some leaess subtree T

of T
p,g
and some k,
t = T

: level
Tp,g
() k
_
, and order S

p,g
naturally.
() S
p,g
is a recursive subtree of S

p,g
such that
(i) T
p,g
is the union of an -branch of S
p,g
,
(ii) for every branch

t = t

: ) of S
p,g
there is q Q
such that q is compatible with p and T
q,g
=

.
(d)

2 or

Then the following holds: if p

Q
is recursive in

then is hyperarith-
metic.
Proof. For some p as in (c) and Turing machine M
p
Q
M computes from

.
Now we prove some intermediate facts, and the proof of 3.1 will be nished with
3.4.
Fact 3.2. For every -branch t
k
: k ) of S
p,g
and j for some (equiv-
alently every) large enough m for every t
m
level
k
(T
p,g
) if M runs on
input j and oracle it nishes (so we do not ask oracle questions outside the
domain) and gives the result (j) = k.
Proof. There is q such that T
q,g

n
t
n
. Let r q, and let G Q be
generic with r G. If M runs with

[G] m it gives (j), so for some T

p,g
,

[G]. Now we proceed as in 1.4. If there is some of height k that gives

another computation result, then it is incompatible with r. But then this is
witnessed by some initial segment of r. Take m larger than all these initial
segments.
Fact 3.3. For j , for every large enough m, for every t level
m
(S
p,g
) there
is t level
m
(T
p,g
) such that if M runs with as an oracle then it computes
(j).
Proof. By the previous fact and K onigs lemma applied to S
p,g
.
Crucial Fact 3.4. For j , k 2, the following are equivalent:
(i) (j) = k.
(ii) There are some m and some t
0
level
m
(S
p,g
) such that for every t
1

level
m
(S
p,g
) there is t
0
t
1
such that if we let M run with input
j and oracle then the run nishes and there are no questions to the
oracle that do not have an answer, and it gives answer k.
Proof. Analogous to the end of the proof of Theorem 1.4.
So we have proved 3.1.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 15
Remark 3.5. Usually, S
p,g
depends only on a nite part of p, so that we have
that Q =

k
Q
k
, and for all k we have S
p,g
as above being the same for
each p Q
k
.
Theorem 3.6. Suppose Q is a notion of forcing and is a Q-name and 1
(x)(xR). Then 1 every real in V that is needed for R is recursive in

.
Proof. Let p Q and V . Since is needed for R, by Fact 1.15 there is some
x in dom(R) that for any y such that xRy,
Tur
y. Now if p xR, then
p
Tur
.
4. Weakly needed reals for the reaping relation
In this section we show that for any ground model V there is a forcing extension
V [G] such that all hyperarithmetic reals from V are weakly needed in V [G] for the
reaping relation. The extension is necessarily a model where weakly needed and
needed are dierent and the CH fails, because of the following: In Section 6 we
shall prove in ZFC that not all hyperarithmetic reals are needed for the reaping
relation, answering another question from Blass work [4]. In a model of CH,
the notions needed real and weakly needed real coincide, and thus in such a
model not all hyperarithmetic reals in any submodel are weakly needed for the
reaping relation. If we take a ground model V with CH then from the coincidence
of needed and weakly needed and from the fact that there are so few needed reals,
we see that there are hyperarithmetic reals in V that are weakly needed in V [G]
but not weakly needed in V . So the model of this section, together with the
result from Section 6, gives an example for the fact that in contrast to the notion
of being needed, the notion of being weakly needed is not absolute.
Theorem 4.1. For any ground model V there is a generic extension V [G] by
some c.c.c. forcing such that in V [G] every hyperarithmetic real in V is weakly
needed for the reaping relation.
The proof of this theorem will occupy the whole section. First we recall the
denition of the reaping relation:
Denition 4.2. The relation
R = (f, X) : f

2, X

[] f X is constant
is called the reaping or the rening or the unsplitting relation. We say X renes
f if f X is constant. We say renes f if there is some X that renes
f. Finally we say renes F if for every f F we have that renes f.
The norm of this relation is called r, the reaping number or the rening number
or the unsplitting number.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


16 HEIKE MILDENBERGER AND SAHARON SHELAH
In this section we often use (nite) boolean combinations. For any nite set u
and
u
2 and A
i
, i u, we set
A

i
=
_
A
i
, if = 1,
A
i
, if = 0;
and

A
[]
=

iu
A
(i)
i
.
Denition 4.3. Let g

be strictly increasing and g(n) > n.
(1) We say A []

is g-slow if (

n)[A g(n)[ n.
(2)
T
g
= f : dom(f) []

, for i dom(f) we have that f(i) = (f

1
(i), f
2
(i))
and f
2
(i) [g(f
1
(i))]
f
1
(i)
and limsupf
1
(i) : i dom(f)) = .
(3) We say that a sequence

A = A
i
: i < ) of innite subsets of is
(g, )-o.k. if
if k < , f
0
, . . . , f
k1
T
g
,

k
dom(f

) = B []

and limsupminf
1

(i) : k : i B) = ,
then for some = (f

: < k))we have that:

For every u

[ ]
<
and

2 the set
n B : ( < k)(f
2

(n)

A
[

]
,= ) is innite.
Remarks: f T
g
implies that

idom(f)
f
2
(i) is g-slow. If g g

then T
g

T
g
.
Claim 4.4. We get an equivalent notion to

A is (g, )-o.k., if we modify the
Denition 4.3(3) as in (a) and/or as in (b), where
(a) We demand 4.3(3) only for f

T
g
that additionally satisfy dom(f
0
) =
= dom(f
k1
) = .
(b) We demand 4.3(3) only for f
0
, . . . , f
k1
T
g
such that minf
1

(i) : i <
k : i < B) is strictly increasing (we can even demand, increasing faster
than any given h), and for i B, maxf
1

(i) : < k < minf

1

(i +1) :
< k.
Proof. (a) Suppose the f
0
, . . . , f
k1
T
g
in the original sense, and that we have
required the analogue of 4.3(3) only for T
g
in the restricted sense. We suppose
that

<k
dom(f

) = B and take a strictly increasing enumeration b

r
: r
of B. Then we take

f

: []
<
,

f

(r) = f

(b
r
) for r . The analogue of
4.3(3) for the T
g
in the restricted sense gives and innite intersections in
4.3(3) for the

f

. The intersections are also innite for the original f

.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 17
(b) Suppose that k < , f
0
, . . . , f
k1
T
g
,

dom(f

) = B []

and
limsupminf
1

(i) : k : i B) = . Then we can thin out the domain B to

some innite B

, inductively on i such that the f

full all the requirements

from 4.4(b).
The following lemma describes the combinatorics that is used in the nal
model:
Lemma 4.5. Let g

. If r < = cf() and if there is some

A that is (g, )-
o.k., then every real that is computable in every function g

g is weakly needed
for the rening relation.
Proof. Let = B

: < [[ be a rening family of size r < . Since the

family

A is rened by , for every i < there are some
i
< [[ and (i) 0, 1
such that B
i
A
(i)
i
. Since is regular and since r < , there are some < 2
and some < [[ such that
Y = i < : (i) =
i
=
is unbounded. So B

iY
A

i
. We claim that B

is not g-slow. Why?

Otherwise we have C = n < : [B

g(n)[ n []

. We take a partial
function f = (f
1
, f
2
) with C = dom(f), f
1
(n) = n and f
2
(n) = B

g(n).
Then f T
g
. Now let be given. Then we take u
0
such that u
0
= ,
Y , > and
0
= (, 0) and

0
= (, 1). Then we do not have
(

n)f
2
(n) A
0

,= and (

n)f
2
(n) A
1

,= at the same time, because B

is
rening A

. So

A is not (g, )-o.k., in contrast to our assumption.
But now we can compute recursively from B

some g

g, for example we
may take g

(n) =(the nth element of B

) +1. Hence every real that is computable

in every function g

g is recursive in B

.
Now we show that there is a version of Lemma 4.5 that works simultaneously
for all hyperarithmetic reals in V .
Lemma 4.6. There is some g : such that every hyperarithmetic real is
computable in any g

g.
Proof. For any number e for a Turing machine take a real r
e
and a lower
bound g
e


such that for all g

g
e
, e computes r
e
with the oracle g

, if there
are such r
e
, g
e
. Now take g eventually dominating all the g
e
, e .
We will nd

A that is (g, )-o.k. in a forcing extension. However, the construc-
tion works only for g V . So the constellation in which we use Lemmata 4.5
and 4.6 is as follows:
Corollary 4.7. Let g V be as in Lemma 4.6 in V . If in V [G], r < = cf()
and there is some

A that is (g, )-o.k., then every hyperarithmetic real in V is in
V [G] weakly needed for the rening relation.
So, how do we get a c.c.c. forcing extension in which r < = cf() and in
which there is some

A that is (g, )-o.k.? The rest of this section will be devoted
to this issue. We consider the case = cf() >
1
and intend to show that for
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


18 HEIKE MILDENBERGER AND SAHARON SHELAH
every g it is consistent that r =
1
and there is some

A that is (g, )-o.k. The
construction works for any xed g V . It is open whether a statement like for
all g V [G], there is some

A
g
that is (g, )-o.k. and r < = cf() is consistent.
We give a sketch of the construction in the consistency proof. We rst add
Cohen reals to some ground model where there are at most reals. We show that
from these we get some

A that is (g, )-o.k. for all g simultaneously. The next
step is to extend further, in
1
steps, so that along this iteration a rening family
of size
1
is added. The lengthy work is to show that we can nd an extension
such that

A stays (g, )-o.k. for one chosen g. This is not trivial because T
g
is
enlarged.
Denition 4.8. (1) K
g
= K = (P,

A

) : P is a ccc forcing and

P


A

is
(g, )-o.k.. For a xed g, we often leave out the subscript.
(2) (P
1
,

A
1

)
K
(P
2
,

A
2

) i P
1
P
2
and

A
1

=

A
2

.
We really mean the same names

A
i

not just the same interpretations. Indeed

we think of a nite support iteration P

, Q

: <
1
,
1
) giving the P

s.
But we formulated 4.8 a bit more general, because also in the next claim the
Q

s do not appear.
Claim 4.9. (1) We have that K ,= . In fact, if P is the forcing adding
Cohen reals and

A

is the enumeration of the Cohen reals, then (P,

A

)
K
g
for any function g V .
(2) If (P

,

A

) K
g
for < , a limit cardinal, and P

: < ) is
increasing and continuous w.r.t. the complete embedding relation, and
P =

<
P

, and P has the c.c.c., then (P,

A

) K
g
and <
(P

,

A

)
K
(P,

A

).
Proof. (1) Suppose that f
0
, . . . , f
k1
V [G

] are injective functions. We take

such that f
0
, . . . , f
k1
V [G

] where G

is a generic lter for the rst Cohen

reals. Suppose that

2, u

. Now a density argument gives that

these

A
[

]
ip for innitely many n B to 0 or to 1 within f
2

(n) for every

< k.
(2) Now we show that
P


A

is (g, )-o.k.. Only the case of cf() = is

not so easy. We suppose that =

n
(n), 0 < (n) < (n + 1). Towards a
contradiction we assume that p

, and
p

P
B

, f

: < k) form a counterexample to

A

being (g, )-o.k.

For each n we nd q
n,i
: i ) such that
() q
n,i
P,
() q
n,0
= p

,
() P [= q
n,i
q
n,i+1
,
() for some b
n,i

, f
1
n,,i

, f
2
n,,i

P
(n)
-names we have
q
n,i
b
n,i

is the i-th member of B

, f

(b
n,i

) = (f
1
n,,i

, f
2
n,,i

),
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 19
() q
n,i
(n) = q
n,0
(n) = p

(n).
How do we choose these? Let n and (n) be given. Then we choose q

n,i
increasing in i such that q

n,i
P and b

n,i
, (f
1
)

n,i
, (f
2
)

n,,i
in V and
q

n,i


<k
the ith element of B

=

b

n,i
f

(

b

n,i
) = (

(f
1
)

n,,i
,

(f
2
)

n,,i
).
Then we take
b
n,i

= (b

n,i
, q

n,i
P
(n)
),
f
1
n,,i

= ((f
1
)

n,,i
, q

n,i
P
(n)
),
f
2
n,,i

= ((f
2
)

n,,i
, q

n,i
P
(n)
),
q
n,i
= p

(n) q

n,i
[(n), ).
Here, the restriction is any reduction function witnessing P

P (see [1]), and

in the general case, if P

is not the initial segment of length of some iteration,

the term q

n,i
[(n), ) has to be interpreted as some element from a quotient
forcing algebra.
Now for every n we dene P
(n)
-names
B

= b
n,i

: i < ,
f
,n

: B

V,
f
,n

(b
n,i

) = (f
1
,n

(b
n,i

), f
2
,n

(b
n,i

)) = (f
1
,n,i

, f
2
,n,i

).
Now we have that
p

[]
0
, f
,n

is a function with domain B

and
limsupf
1
,n

(b) : b B

) = and
f
2
,n,i

when dened is a subset of [0, g(f

1
,n,i

)) of cardinality f
1
,n,i

.
As (P
(n)
,

A

) is in K we have for every n

p

(n)
P
(n)
for some

< for every u

]
0
for every

2
_
b B

<k
f
2
,n
(b)

]
,=
_
is innite.
Let
n

< be such a P
(n)
-name. Since P
(n)
has the ccc, there is some

n
<
such that
P
(n)

n

<

n
< . Since is regular we have that

n
< .
It suces to prove that
p

is as required in the denition of (g, )-o.k.

If not, then there are counterexamples u

]
<0
,

2, q and b

such
that
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


20 HEIKE MILDENBERGER AND SAHARON SHELAH
p

q P = P

q
_
b B

: ( < k)(f
2

(b)

A
[

,= )
_
[0, b

].
(4.1)
For some n() < we have that q P
(n())
. Let G P be generic over V ,
and let q G
(n())
. So by the choice of
n()
<

we have that
q
P
(n())
C = b B

n()

: ( < k)(f
2
,n()

(b)

A
[

,= ) is innite.
Recall that B

n()

and f
,n()

(b) are P
(n())
-names and that

A
[

is a P
0
-name.
Now B

n()

= b
n(),i

: i < , so for some i we have that b

n(),i

[G] > b

. So
q
n(),i
G P
(n())
forces the i-th member of B

is b
n(),i

and f

(b
n(),i

) =
f
,n()

(b
n(),i

) = (f
1
,n(),i

, f
2
,n(),i

). Note that q
n(),i
(n()) = p

(n())
according to ), and hence q
n(),i
, q. So there is some r q and r q
n(),i
.
Such an r forces the contrary of the property forced in (4.1), and nally we
reached a contradiction.
The conclusion of the next claim is a strengthening of 4.3(3). Let D be an
ultralter on . Instead of innite we require being in D. Since ultralters
are closed under nite intersections we need to mention only one function in T
g
.
Claims 4.10 and 4.11 are like [10]. For h: we write lim
D
h(i) : i
) = if for all m < we have that i : h(i) > m D.
Claim 4.10. Assume that in V :
(a)

A is (g, )-o.k.
(b) = 2
0
is regular.
Then there is an ultralter D on such that
if f T
g
and dom(f) D and lim
D
f
1
(n) : n dom(f)) =
then for some
f
< for every u [
f
]
<0
and
u
2
we have that n dom(f) : f
2
(n)

A
[]
,= D.
(4.2)
Proof. The following is a mock forcing argument. We work with the partial order
/T, which is < -closed. We have to meet only dense sets. So, by taking one
union over conditions in the end, we nd a generic in V . Let T
g
= f
j
: j < .
Let /T be the set of tuples (D, i, ) such that
(i) D is a lter on containing the co-nite subsets, , D, i, < ,
(ii) D is generated by < members,
(iii) if k < and for < k, j

< i, and dom(f

j

) D and lim
D
f
1
j

(n) : n
dom(f
j

)) = and u

[ ]
<0
,

2, then
_
n

<k
dom(f
j

) :

<k
_
f
2
j

(n)

A
[]
,=
_
_
,= mod D.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 21
Let (D
1
, i
1
,
1
)
AP
(D
2
, i
2
,
2
) if both tuples are in /T and
() D
1
D
2
, i
1
i
2
,
1

2
, and
() if k < and j
0
, . . . , j
k1
i
1
, dom(f
j

) D
2
and lim
D2
f
1
j

(i) : i
dom(f
j

)) = and u

[
1
,
2
) is nite and

2 then
_
n

<k
dom(f
j

) :

<k
f
2
j

(n)


A
[

]
,=
_
D
2
.
Now we have that (/T,
AP
) is a non-empty partial order. Take i = = 0 and
D the lter of all conite subsets of . In (/T,
AP
) every increasing sequence
of length < has an upper bound, namely, take the lter generated by the union
in the rst coordinate and take the supremum in the second and in the third
coordinate.
Now we come to the rst kind of sets we want to meet: If B and (D, i, )
/T then there are some D

, i

such that (D

, i

)
AP
(D, i, ) and that
B D

or that B D

. Why? Try D

= the lter generated by D B

and the same i and . If this fails then we can nd k < , such that for < k
we have j

< i, such that dom(f

j

) D

and lim
D
f
1
j

(i) : i dom(f
j

)) = ,
u

[ ]
<0
,

2 and such that

_
n

<k
dom(f
j

) : f
2
j

(n)

A
[

]
,=
_
B = mod D.
Let

< be such that

and
_
<k
u

. Let D

be the lter generated

by
D
_
_
n

<k
dom(f
j

) : f
2
j

(n)

A
[

]
,=
_
:
k < , j

< i, u

]
<0
,

2
_
.
Then B D

, and (D

, i,

) /T.
Finally, there is a second useful kind of dense sets: If (D, i, ) /T then for
some D

we have that (D

, i + 1,

) /T.
Proof. Let M (H(), ) such that M , (D, i, ) M, T
g
M, and
[M[ < . Suppose that dom(f
i
) D and that lim
D
f
1
i
(k) : k dom(f
i
)) = .
Let

= M. Let D
1
be the lter in the boolean algebra in T()M generated
by
(D M)
_
_
n

<k
dom(f
j

) : f
2
j

(n)

A
[

]
,=
_
:
k < , j

i, u

]
<0
,

2
_
.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


22 HEIKE MILDENBERGER AND SAHARON SHELAH
Since in M,

A is (g, )-o.k., this has the innite intersection property. Let D

2
be an ultralter in M extending D
1
. Let D

be the lter on in V that D

2
generates.
Now we take an increasing chain (D
j
, i
j
,
j
) : j < ) in the partial order
(/T,
AP
) such that i
j
is unbounded in and D :=

j<
D
j
is an ultralter.
Then D fullls (4.2).
Now we use equation (4.2) of 4.7, which implies that

A is (g, )-o.k., to con-
struct an extension in which

A is still (g, )-o.k. The following preservation
theorem is a bit more general: it works also when the D

s do not coicide. In
our application, however, they will coincide.
Claim 4.11. Assume that
(a)

A is (g, )-o.k.
(b)

D = D

:
<
), D

= D, D is an ultralter on as in 4.7.
(c) Q
D
= T : T
<
is a subtree with exactly one -minimal element,
and for some T, T k : k T D

, ordered by
inverse inclusion. (The -minimal of this sort is called the trunk of
T, tr(T).)
Then
Q
D


A is (g, )-o.k..
Proof. We use the fact [10] that Q
D
has the pure decision property: Let
i
,
i , be countably many sentences of the Q
D
-forcing language. We think
of names f

, < k, for some elements of T

g
and
i
=
_
the i-th element of
B

=

<k
dom(f

)
_
=

b
i
and
_
<k
f

b
i
) = (

f
1
,i
,

f
2
,i
). The pure decision
property says:
p Q
D
q
tr
p r q i
_
r
i
(s
i
r)q
[si]

i
_
,
where we write
tr
for the pure extension: q
tr
r if r q and tr(q) = tr(r),
and q
[si]
= q : s
i
.
Towards a contradiction we assume that there is a counterexample. By Claim 4.4
(rst (b) and then (a)) we may assume that it is of the following form
p

: < k) are functions from to

and for i , maxf
1

(i) : < k < minf

1

(i + 1) : < k
and there is no < such that the statement
Denition 4.3(3) holds.
(4.3)
We nd q such that
() q P
() q
tr
p

,
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 23
() for all i for all f
1
,i
, f
2
,i
[0, g(f
1
,i
)) of size bigger than f
1
,i
we
have that
if r q, r f

i) = (

f
1
,i
,

f
2
,i
),
then also for some s
i
r, the condition q
[si]
forces the same.
We x such a q.
Now we set for q and < k
B
1
,
= i : some pure extension of q
[]
decides f

(i)

.
We say (, ) is 1-good if B
1
,
D. Let for i B
1
,
, h
,
(i) = (h
1

, h
2
,
) the
value of f

(i)

that is given by the pure decision. This is well-dened because any

two pure extensions are compatible. Of course, by the requirements we had put
on the counterexample, we have that lim
D
h
1
,
(i) : i B
1
,
) = .
We say that (, ) q k is 2-good, if it is not 1-good and we have for all
m that
M
,,m
= j : (i )(h
j,
(i)) is well-dened,
and h
1
j,
(i) > m)
_
D.
So, for 2-good but not 1-good (, ) we may dene for j M
,,m
,
g
,
(j) =h
j,
(i
j,
),
where i
j,
is such that h
j,
(i
j,
) is dened in h
1
j,
(i
j,
) > m
and if there is a maximal such i, then take this as i
j,
.
We show that there is M

,,m
D, M
,,m
M

,,m
such that for j M

,,m
there is a maximal such i: If h
j,
(i) is dened and i

< i then there is some

pure extension deciding h
j,
(i

) since there are only nitely many possibilities

for its values, by the third line of (4.3). Hence some pure extension decides the
value. Hence also h
j,
(i

) is dened. If h
j,
(i) is dened for all i, then (j, )
is 1-good. Hence , if (j, ) is 2-good but not 1-good, then there is a maximal i
witnessing j M
,,m
. If j : (j, ) is 1-good D, then by gluing together
suitable pure extensions r
j
of q
[j]
together we get a pure extension of q
[]
that
shows that (, ) is 1-good. Hence X = j : (j is 2-good and not 1-good D.
So we may take M

,,m
= M
,,m
X. In order to simplify notation, we assume
that M

,,m
= M
,,m
.
Also from the third line of equation (4.3) we get that for every q either for
all < k, (, ) is 1-good or no (, ) is 1-good. In the latter case there is some
i

, such that for all < k, dom(h

,
) = i

or dom(h
,
) = i

+ 1. Moreover, also
by (4.3) we get that if for some < k, for all m, M
,,m
D, then for all < k,
for all m, M
,,m
D. So if (, ) is 2-good, then all (,

) are 2-good. We call

i-good if there is some such that (, ) is i-good. We set M
,m
=

<k
M
,,m
.
We x some pseudo-intersection M

of M
,m
: m ), such that limi
j
:
j M

) = .
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


24 HEIKE MILDENBERGER AND SAHARON SHELAH
Then we also have that lim
D
ming
1
,
(j) : < k : j M

) = , because for
each z < , j : ming
1
,
(j) : < k < z is a nite set. Hence g
,
T
g
. By
combining with an enumeration of M

, we may assume that dom(g

,
) = D.
We will not write this enumeration, in order to prevent too clumsy notation, but
we shall later apply that D is as in 4.7 for T
g
, and therefore we need that the
domains are in D.
Now we take suciently large and N (H(), ) such that f

: < k)
N, B
1
,
, h
,
, g
,
: q, < k) N, q, D N. We take

= sup(N ). We
claim that q forces that

is as in the Denition 4.3(3).

If not, then there are counterexamples u

]
<0
and

2 and
r Q
D
, r q, and b

such that
r q, and
r
QD

<k
dom(f

) = and
(i ) maxf
1

(i) : < k < minf

1

(i + 1) : < k
and
_
b : ( < k)(f
2

(b)

A
[

,= )
_
[0, b

].
(4.4)
First case: There is some r with tr(r) such that all (, ) are 1-good.
Now we take for each t , some pure extension of q
[]
t
of r
[]
such that it forces
_
<k
(h
,
t = f

t). Since

A is (g, )-o.k., and since all is reected to N, and
by the choice of

, we have that I = n : ( < k)(h

2
,
(n)

A
[

]
,= is
innite. So we take t I such that t > b

. Now q
[]
t
contradicts (4.4).
Second case. There is some r such that all , < k are 2-good but not
1-good. We set g
,
(j) = h
j,
(i
j,
) as purely decided above q
[j]
. Fact:
Now g
,
: < k) is as required in the denition of

A being (g, )-o.k., because
= lim
D
g
1
(i
j
) : j ).
Now we take for each t , some pure extension q
[j]
t
of r
[j]
such that it
determines
_
<k
g
,
t. Since

A is (g, )-o.k., and since all is reected to N,
and by the choice of

, we have that J = n : ( < k)(g

2
,
(n)

A
[

]
,=
is innite. Then also

J = i
n
: n J is innite. So we take t > b

, t

J.
Now the gluing together of q
[j]
t
, j

<k
M
,,t
, contradicts (4.4) because we
have g
,
(j) = h
j,
(i
j,
) = f

(i
j
), if q
[]
t
G. Here we write f

for f

[G].
Third case: All r are neither 1-good nor 2-good. We shall prove something
stronger:
An end-segment of the generic

: there is some element q G

with trunk can be thinned out (so that still innitely many
points are left) and injected into an innite subset of n :
_
<k
f
2

[G](n) A
[

]
,= .
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 25
This is more than enough.
Let i
,
= max(B
1
,
) < , because (, ) is neither 1-good nor 2-good. Let
i

= dom(h
,
) such that i

= i

,
or i

= i

,
+ 1. By the premise (4.3),
there are such i

. There is r q with no r being 1-good or 2-good in N.

Without loss of generality, we take q like that. Now we try to shrink q purely.
Let
0
= tr(q).
First: We have that f

is decided by q. The range of i

j
: j q) is
bounded modulo D because is not 2-good. Hence we may assume that there
is just one value i

. So say (after shrinking q) that it is constant with value

i

.
Second we have that
0
q implies that q
[]
decides f

.
Third we have that if i [i

, i

] then lim
D
f
1
j,
(i) : j ) = by the
denition of i

and i

. So dene g
,,i
by g
,,i
(j) = h
j,
(i). So g
,,i
N is a
function of the right form.
We have by the denition of

, for all i [i

, i

) for all q for all u

that
A := b : ( < k)g
2
,,i
(b)

A
[

]
,= D.
Since the range of

: there is some element q G with trunk =:

is
eventually contained in every set in D, we now nd the following innite set: We
take
n
: n ) such that
n
range(

) A and such that i

n
< i

n+1
. We
set
n
=
n
[
n
1[. Then we have for almost all n such that
n
A and hence
for all i [i

n
, i

n
): g
n,,i
(
n
([
n
1[)) = h
nn(|n1|),
(i) = h
n,
(i) = f

(i).
So

n
[i

n
, i

n
]

b : ( < k)f
2

(b)

A
[

]
,= is innite.
Claim 4.12. Let = cf() >
1
. Let V [= 2

and let P
0
= C

be the
forcing adding Cohen reals. We x some function g V such that every
hyperarithmetic function is computable in every g

g. We let G
0
be P
0
-generic
over V and set V
1
= V [G
0
]. Let in V
1
,

A be the enumeration of the Cohen
reals.
(1) In V
1
, there is (P,

A) K
g
such that
P
r =
1
.
(2) For (P,

A) as in (1), we have that in V
1
,
P
every hyperarithmetic real
in V is weakly needed for the reaping relation.
Proof. (1) By 4.7 we have that

A is (g, )-o.k. in V
1
. According to 4.9, we may
choose in V
1
a -increasing sequence such that (P
i
,

A) K, P
i+1
= P
i
Q
D
i

and
P
j
, Q
i
: i <
1
, j
1
) is a nite support iteration and D
i

=

D
i

:
<
)

D
i

= D
i
V
Pi
enjoy the properties required there.
Then P forces that r =
1
: P consecutively adds (shoots)
1
reals through
the ultralters D

in the intermediate models V[G

], <
1
. Let f V
P
be
a real. Then, by the countable chain condition, there is some <
1
such that
f V
P
. Then f
1
[0] D

or f
1
[1] D

. Since Q
D
adds a real r

that
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


26 HEIKE MILDENBERGER AND SAHARON SHELAH
is almost a subset of every member of D

, we have that r

renes f. Hence in
V
P
, r

: <
1
witnesses r =
1
.
(2) Now by part (1) and by Lemma 4.6 the proof of (2) follows.
Finally, taking P = P
0
P

and G P-generic over V and

A as in 4.11, statement
(2) yields that in V [G] every hyperarithmetic real from V in weakly needed for
the reaping relation, and thus the proof of Theorem 4.1 is nished.
5. There may be more weakly needed reals than needed reals
Under CH, or if [[R[[ = 2
0
, the notions needed for R and weakly needed
for R coincide. In this section, we show that there is some quite simply dened
relation R and that there is some model of ZFC in which there are more weakly
needed reals for R than needed reals for R. The idea is to use the forcing model
from the previous section.
Claim 5.1. There is a simply dened relation R for which it is consistent that
the notions weakly needed and needed do not coincide. In fact, in the forcing
models V , V [G] from the previous section, in V [G] every needed real for R is
recursive (hence in V ), and all the hyperarithmetic (and possibly more) reals are
weakly needed for R.
Proof. Let R
0
be the ordinary reaping relation, which we write for functions on

2:
R
0
,

2 (

n)(n) = 1
1
1 is almost constant.
Let R
1
be as follows:
R
1
,

2 (

n)(n) = 1 (

n)(n) = 1
_
[
1
1
1
1 n[
[
1
1 n[
: n
_
converges to
1
2
.
We set R = R
0
R
1
and use V
P
from the previous section. There we have
that P = P
0
Q

, P
0
is the forcing adding Cohen reals, and

A

is an enumeration
of the names of these Cohen reals, and Q is the iteration described in 4.9. Then
in V
P
we have that [[R[[ [[R
0
[[ =
1
.
We rst show that every hyperarithmetic real V is weakly needed for R in
this model. We take some R-adequate set in V
P
of power
1
. We let
Y

= i < : (x )(A
i
R

x).
So, by the denition of adequate we have that Y
0
Y
1
= . If [Y
0
[ = , then by
the proof of 4.5, we get some x whose enumeration f with f(n) = m if m is
the nth element of x is so large in the eventual domination order that the real
is computable from it.
We now show that [Y
1
[ < . Then it follows that [Y
0
[ = . Towards a
contradiction, we assume that [Y
1
[ = . In the model from the previous section
we have that P =

i<1
P
i
, P
0
adds Cohen reals A

, < , P
i
increasing and
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 27
continuous, P
i+1
= P
i
Q
Di

as there, P = P
0
Q

. We work in V
P0
. We have
that for some p

Q and some Q-names

i

, i <
1
,
p

Q/P0
[Y
1

[ =

=
i

: i <
1
.
Let Y

= : p

, p

Q/P0
Y
1

. By the ccc of Q/P

0
, we have that
Y

[]

, and for Y

we choose p

such that p

Q/P0
Y
1

. So for
we have that A

R
1

i()
and hence for a large enough n

for many Y

(without loss of generality, for all Y

) we have (n n

)(
|
1
i()
{1}An|
|An|

_
1
4
,
3
4

). Moreover, there is a -system for the domp

[ 0]
<
whose root
is u

, and we assume that for Y

, i() = i

.
So we may assume that for j u

we have that p

(j) is an object with trunk

j
and not just a P
0
-name. By pure decidability for some

V
P0
we have:
For every Y

and m for some pure extension q of p

with the same domain

q
i

m =

m. By the choice of n

we get an easy contradiction: Suppose

p P
0
and
p
P0
Y

n n

tr
p

,
q

Q/P0

[

1
1 A

n()[
[A

n()[

_
1
4
,
3
4
_
.
This is impossible, because we may assume that

V (it needs only countably

many of the Cohen reals) and we may arrange all other A

s so that the quotient

will be arbitrary. The forcing P/P
0
does not change the value of the quotient.
Now we show that if a real is not recursive then it is not needed for R. If
is not recursive and x

2, let x, N (H(), ), N countable. Let
= (x, ) be random over N, and we claim
(5.1) ,
Tur
.
Proof of (5.1): Otherwise we would have that is recursive in the ground model.
This is proved in [5] and in [4, Proposition 14]. Since the proof is short, we repeat
it here: Suppose
(5.2) p
Random
M computes from the oracle

.
Then by the Lebesgue density theorem we nd s
<
2 such that above s, p
has Lebesgue measure >
99
100
Leb( : s . The we set
B
n
=

2 : s

and from

M computes (n) correctly.

From (5.2) we get that Leb(B
n
)
99
100
Leb( : s . So for every suciently
large m we have that
(5.3) 2
mlg(s)
[

m
2 : s

and from

M computes (n) correctly[.

So we can run a machine, that has s as a xed ingredient, and which, given input
n, increases m successively, and then computes (n) with all possible oracles
above s of length m lg(s) and decides with (5.3), when it is true for m (and
hence for all later m), which is the right value. So equation (5.1) is proved.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


28 HEIKE MILDENBERGER AND SAHARON SHELAH
Thus the collection (x, ) : x

2 is an R
1
-adequate family and hence an
R-adequate family. So, if is needed for R, there is some such that
Tur
,
and hence by equation (5.1) is recursive.
6. Needed reals for the reaping relation
In this section we prove in ZFC that for any submodel not all hyperarithmetic
reals in it are needed for the reaping relation. Since in the model V [G] from
Section 4 all hyperarithmetic reals in V are weakly needed for the reaping relation,
the model V [G] shows that also for the reaping relation it is consistent that weakly
needed and needed do not coincide. In contrast to the result on the relation R
from the previous section, we do not prove that only recursive reals are needed
for the reaping relation. It is open whether our result here is sharp.
Theorem 6.1. If is needed for the reaping relation, then is arithmetical.
So, applied to the pairs of models from Section 4, we get:
Conclusion 6.2. For any V , V [G], not all hyperarithmetic reals are needed for
the reaping relation in V [G].
Since there are non-arihmetical hyperarithmetic reals, 6.2 follows from 6.1.
Proof of 6.1: Suppose that is needed for the reaping relation. Then by
Fact 1.15 there is some B

such that:
(6.1) For all X, if X B

= B
1

or X B

= B
2

then is recursive in X.
For all X that rene B

, we have that is recursive in X. Note that equa-

tion (6.1) is similar to being hyperarithmetic: the dierence is that is com-
putable also in every innite subset of the complement of B

. Unless is recur-
sive, we have that B

in equation (6.1) is innite and coinnite.

Notation 6.3. Let (M
n
1
, M
n
2
, a
n
1
, a
n
2
) : n < ) be a recursive list of the quadru-
ples (M
1
, M
2
, a
1
, a
2
) such that
(i) M
1
, M
2
are Turing machines (with reference to an oracle),
(2) a
1
, a
2
are nite disjoint sets.
Without loss of generality, a
n
1
a
n
2
n and each quadruple appears innitely
often. (This will be used in 6.10.)
Denition 6.4. We say

E = E
n
: n ) is special if
(i) E
n
is an equivalence relation on n, and
(ii) for m < n, E
n
renes E
m
( n), i.e., every E
m
-class is the union of
some E
n
-classes plus some subset of n,
(iii) if A is an E
n
-equivalence class, then A (n+1) is divided by E
n+1
in at
most two equivalence classes, and E
0
has nitely many classes,
(iv) if
() A is an E
n
-equivalence class and
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 29
() there is a partition X
1
, X
2
of A (n+1) such that for all j < ,
Y
i
, i = 1, 2, (if a
n
i
Y
i
X
i
a
n
i
, h
i
< , and if the
machine M
n
i
running with input j and oracle Y
i
nishes its run
giving h
i
, then h
1
= h
2
),
then E
n+1
induces such a partition of A. (But note: it need not be the
same partition.)
Lemma 6.5. There is a special

E that has as a three place relation n, x, y) :
xE
n
y Turing degree 0

and such that if A is an E

n
-equivalence class then
A
Tur
0
n+1
.
Proof. We choose E
n
by induction on n.
n = 0. If for every m there is a partition (c
0
, c
1
) of m such that for i 1, 2
for every b
i
c
i
and j < n, if M
0
i
running with input j and oracle b
i
m and
gives the results k
i
then k
0
= k
1
, then we choose among these pairs (c
m
1
, c
m
2
)
such that c
m
1
is minimal in the lexicographical order. If (c
m
1
, c
m
2
) are dened for
every m, then we have that m
1
m
2
m
3
c
m2
1
m
1

lex
c
m3
1
m
1
. So
c
m
1
: m ) converges to some c
1
. Now we dene E
0
, having two classes: c
1
and c
1
. The relation E
0
is computable in 0
1
.
In the step from n to n + 1, the relation E
n+1
is dened similarly, with the
modication that we use the description of E
n
as a parameter and take partitions
(c
0
, c
1
) of (mn) C for each E
n
-class C and oracles b
i
a
n
i
. Clearly using 0
n+1
we can dene E
n+1
and it is in the degree 0
n+2
.
Note dierent successful computations have the same outcome.
6.5
Lemma 6.6. If is needed for reaping and

E that is special, then there are
n and an E
n
-class A such that is recursive in A.
Proof. We assume the contrary. Based on this assumption, forcing with Q
E,B
and absoluteness we will lead to a contradiction. The proof will be nished with
Claim 6.10.
Denition 6.7. For a special

E we dene Q = Q
E,B
as the following notion
of forcing:
(1) p Q has the form p = (n, A, b
1
, b
2
) = (n
p
, A
p
, b
p
1
, b
p
2
) such that
(i) n < ,
(ii) A is an E
n
-equivalence class,
(iii) A is innite,
(iv) b
1
, b
2
are disjoint subsets of n,
(v) b
1
B

, b
2
B

.
(2) p q i
(i) n
p
n
q
, A
p
A
q
, b
p
i
b
q
i
, for i = 1, 2,
(ii) (b
q
1
b
q
2
) (b
p
1
b
p
2
) A
p
.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


30 HEIKE MILDENBERGER AND SAHARON SHELAH
(3) B
i

=

b
p
i
: p G
Q

is a Q-name of a subset B
i
V [G] of B

i
for
i = 1, 2.
Q is equivalent to Cohen forcing and independent of

E and B

. Nevertheless
we keep the complicated conditions, because they t better to the investigation
of the s needed for the reaping relation. We shall show: is computable in the
generic (hence it is recursive) or it is computable in some E
n
-class A. So, for the
following three claims we assume that is not computable in any E
n
-class A.
Claim 6.8. For i = 1, 2 we have
(1)
Q
b
i

is an innite subset of B

i
.
(2) For densely many p

, p

Q

_
i=1,2
(M
n
p

i
computes with the oracle
b
i

).
Proof. (1) It is enough, to nd for a given p Q some q p, q Q such that for
i = 1, 2, b
p
i
,= b
q
i
. Now A
p
B

i
is innite, because of the hypothesis on B

and
because is not recursive in A
p
by the assumption. We may choose h A
p
B

i
,
h max(max(b
p
1
), max(b
p
2
)) + 2 and an innite E
h
-class A A
p
, which exists
because A
p
is innite and because E
h
has nitely many equivalence classes. We
dene q as n
q
= h + 2, A
q
= A, b
q
1
= b
p
1
h, b
q
3i
= b
p
3i
h + 1.
(2) The statement made in equation (6.1) on B

and on is
1
1
and holds
in V ; hence it holds in V [G] as well by Shoenelds absoluteness theorem [7,
Theorem 98, p. 530]. We arrange by possibly increasing n
p

that the machines

witnessing the recursiveness are M
n
p

i
. Now we apply it in V [G] to part (1) of
this claim.
We x p

, M
n
p

1
, M
n
p

2
as in part (2) of Claim 6.7.
Fact 6.9. There is some q p

such that for i = 1, 2, M

n
q
i
= M
n
p

i
and such
that b
q
i
= a
n
q
i
.
Proof. For some n
q
n
p

the quadruple (M
n
q
1
, M
n
q
2
, a
n
q
1
, a
n
q
2
) is equal to
(M
n
p

1
, M
n
p

2
, b
p

1
, b
p

2
). Let A be an innite E
n
p

-class which is a subset of A

p

.
So we take q = (n

, A, a
n
q
1
, a
n
q
2
).
Claim 6.10. For n

, A the demands () + () of clause (iv) of 6.4 hold, hence

the conclusion.
Proof. We work rst in V [G]. There, by 6.8, X
i
= AB

i
and A exemplify 6.4(iv).
But 6.4(iv) is a
1
2
-statement of the parameters (A, a
n
1
, a
n
2
), and therefore it holds
in V as well by Shoenelds absoluteness theorem.
Finally we nish the proof of Theorem 6.1: Let A
1
,= A
2
be the E
n

+1
-
equivalence classes which are A, with A
i
for M
i
as in 6.4(iv). So by 6.4(iv),
is computable in E
n

+1
.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


ON NEEDED REALS 31
7. Coincidence
In this section we give a condition on a relation R under which the notions
needed for R and weakly needed for R coincide and show that the condition
is fullled for the relation R
random
dened below.
Denition 7.1. The domain of the relation R
random
is T
<
2 with no leaves
and Leb(lim(T)) >
1
2
, i.e., the domain of the notion of forcing from Section 1.
The range of R
random
is

2. We set TR
random
i A
T
:=

2 : for
some

lim(T) we have that =

.
Denition 7.2. R is boring if
(a) R is a 2-place Borel relation on

2 and (x

2)(y

2)(xRy) , and
(b) for every x
1
, x
2

2, if x
2
is not recursive, there is x 2

such that
()
_
xR (x
1
R (x
2

Tur
))
_
.
Claim 7.3. (1) Assume that R is boring. Then the notions of being needed
for R and being weakly needed for R coincide and coincide with being
recursive.
(2) The relation R
random
is boring.
Proof. (1) We have show that every weakly needed real for R is recursive. Since
every recursive real is needed for R, and since weakly needed reals are needed,
this will complete the cycle of implications.
Suppose that x

2 is not recursive. We show that x

is not weakly needed

for R. Let Y be an R-adequate set of minimal cardinality. Let Y

= Y :
x

Tur
. Y

Y , and hence [Y

[ [Y [ = [[R[[. We show that Y

is also
R-adequate. Then, by the denition of Y

, x

is not needed for R, and the proof

is nished.
Let x
1


2 be given. We take x
2
= x

, and apply (b) of the denition of

boring. So we get x as there. Since Y is R-adequate we nd some Y such
that xR. Hence by Rs boringness we have that x
1
R x
2
,
Tur
. So Y

and x
1
R.
(2) Let x
1
, x
2
be given. We take N (H(
3
), ) such that x
1
, x
2
N. Let T
be Amoeba-generic over N. Then T = x is as claimed in Denition 7.2(b): Let
2

be such that A
T
. The closed set T is a subset of x
1
by the Amoeba
genericity of T. Hence x
1
R
random
. The set : x
2

Tur
is a tail set and
hence has measure zero or one. Since every real recursive in a generic for random
forcing is recursive (see the proof or equation (5.1) or [4, Proposition 14] or [5])
and since x
2
is not recursive, every generic real for the random forcing avoids the
set. Hence it has measure zero and is disjoint from A
T
, and therefore for A
T
,
(x
2

Tur
).
Conclusion 7.4. Needed reals for R
random
and weakly needed reals for R
random
coincide and are just all the recursive reals.
7
2
5


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
5
-
2
7







m
o
d
i
f
i
e
d
:
2
0
0
3
-
0
5
-
2
7


32 HEIKE MILDENBERGER AND SAHARON SHELAH
References
[1] Uri Abraham. Proper forcing. In Matthew Foreman, Akihiro Kanamori, and Menachem
Magidor, editors, Handbook of Set Theory. Kluwer, To appear.
[2] Tomek Bartoszy nski and Haim Judah. Set Theory, On the Structure of the Real Line. A
K Peters, Wellesley, Massachusetts, 1995.
[3] Andreas Blass. Needed reals. Talk at Oberwolfach December 1999.
[4] Andreas Blass. Needed reals and recursion in generic reals. Ann. Pure Appl. Logic, 109:77
88, 2001.
[5] K. de Leeuw, E.F. Moore, C.E. Shannon, and N. Shapiro. Computability by probabilistic
machines. In J. McCarthy and C.E. Shannon, editors, Automata Studies, pages 183 212.
Princeton University Press, 1956.
[6] Martin Goldstern. A taste of proper forcing. In Carlos Augusto di Prisco, Jean A. Larson,
Joan Bagaria, and A.R.D. Mathias, editors, Set Theory, pages 7182. Kluwer Academic
Publishers, 1998.
[7] Thomas Jech. Set Theory. Addison Wesley, 1978.
[8] Carl G. Jockusch Jr. Uniformly introreducible sets. J. Symbolic Logic, 33:521536, 1968.
[9] John Oxtoby. Measure and Category. Springer, second edition, 1980.
[10] Saharon Shelah. Long Iterations for the Continuum. Preprint [Sh:707], 2000.
[11] Robert Solovay. Hyperarithmetically computable sets. Trans. Amer. Math. Soc., 239:99
122, 1978.
Current address: Heike Mildenberger, Institut f ur formale Logik, Universit at
Wien, W ahringer Str. 25, 1090 Wien, Austria
Saharon Shelah, Institute of Mathematics, The Hebrew University of Jerusalem,
Givat Ram, 91904 Jerusalem, Israel
E-mail address: heike@logic.univie.ac.at
E-mail address: shelah@math.huji.ac.il