DEFINABLE FUNCTIONS ON DEGREES

THEODORE A. SLAMAN AND JOHN R. STEEL

§1. The work we shall describe was motivated by some attractive conjectures
of D. A. Martin. Let ≤T and ≡T be Turing reducibility and equivalence. Unless
otherwise specified, a degree is a Turing degree, that is, an ≡T -equivalence class;
deg(x) is the degree of x. A real is an element of 2, the Cantor space. A cone
of reals is a set of the form {y : x ≤T y} for some real x; a cone of degrees is
a set of the form {deg(y) : x ≤T y} for some real x. A property P of reals
(degrees) holds on a cone iff {x : P(x)} contains a cone of reals (degrees). A
function f : 2 → 2 is degree invariant iff ∀x, y(x ≡T y =⇒ f(x) ≡T f(y);
f is increasing on a cone iff x ≤T f(x) on a cone, order-preserving on a
cone iff ∃z∀x, y(z ≤T x ≤T y ⇒ f(x) ≤T f(y)), and constant on a cone iff
∃y(f(x) ≡T y on a cone). This last term is an abuse of language, since not
f, but the induced function on degrees, is literally constant on a cone. For
f, g : 2 → 2, let f ≤m g iff f(x) ≤T g(x) on a cone. The Turing jump of x
is x  .
Conjecture I (Martin). Assume ZF+AD+DC. Then if f is degree invariant
and not increasing on a cone, then f is constant on a cone.
Conjecture II (Martin). Assume ZF+AD+DC. Then ≤m prewellorders
the set of degree invariant functions which are increasing on a cone. If f has
≤m -rank α, then f  has ≤m -rank α + 1, where f  (x) = f(x) for all x.
A special case of Conjecture II was raised as a question by Sacks in [Sac67].
AD is the axiom of determinacy. The AD hypothesis in the conjectures
amounts to a definability restriction on the functions to be considered. If the
conjectures are true and all games in L(R) are determined, then Conjectures I
and II are true for all functions in L(R). Moreover, any proof of the conjectures
will almost certainly use determinacy “locally”, and so show, e.g., that Borel
determinacy implies Conjectures I and II for Borel functions. Our partial
results use determinacy locally.
One of the main reasons determinacy is useful in this area is a well known
theorem of Martin. Let X be the set of finite sequences from X ; a tree

Partially supported by an NSF grant.

Ordinal Definability and Recursion Theory: The Cabal Seminar, Volume III
Edited by A. S. Kechris, B. Löwe, J. R. Steel
Lecture Notes in Logic, 43
c 2016, Association for Symbolic Logic 458

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 459

on X is a subset of <X closed under initial segment. If T is a tree on X ,

then [T ] = {f ∈ X : ∀n ∈ (fn ∈ T )} is the set of infinite branches
of T . For s, t ∈ <X, s is compatible with t iff s ⊆ t or t ⊆ s; T is perfect iff
T = ∅ and every sequence in T has incompatible extensions in T . We shall
consider mainly trees on 2 = {0, 1}; a tree T on 2 is pointed iff T is perfect and
∀x ∈ [T ](T ≤T x).
Theorem 1.1 (Martin [Mar68]). Assume AD. Suppose P ⊆ 
2 and
∀x∃y ≥T x P(y). Then there is a pointed T such that [T ] ⊆ P.
If T is pointed then ∀x(T ≤T x =⇒ ∃y ∈ [T ](y ≡T x)). Thus if P ⊆ 2
is degree invariant, that is, (P(x) ∧ y ≡T x) =⇒ P(y), then assuming AD
either P or 2 − P contains a cone. A simple corollary is that, under AD, there
is no choice function on degrees. For if f : D → 2, where D is the set of
degrees, then ∀n ∈ (f(d)(n) is constant on a cone), so f(d) is constant on a
cone, so f(d) ∈ / d on a cone. (One can show there is no choice function on
degrees assuming only all sets have the Baire property; cf. §3.) Notice that one
can easily construct counterexamples to Conjectures I and II from a choice
function on degrees, so the fact that AD rules out such a function is reassuring.
We elaborate a bit on this theme in §3.
The only definable, degree invariant functions known are the constant
functions and the various jump operators (Turing jump, Δn1 jump, sharp,
etc.). The true intent of Conjectures I and II is to assert there are no others.
This assertion can be made more explicit as follows. Let {i}x be the ith
real recursive in x in a standard enumeration. We say x ≡T y via (i, j) iff
{i}x = y and {j}y = x. A degree invariant f : 2 → 2 is uniformly degree
invariant iff there is a pointed tree T and a function t :  ×  →  ×  such
that ∀x, y ∈ [T ](x ≡T y via (i, j) =⇒ f(x) ≡T f(y) via t(i, j)). We shall
show in §2 that any uniformly invariant function which is not increasing on
a cone is constant on a cone (assuming AD). Becker [Bec] shows that the
uniformly invariant functions which increase on a cone are precisely the jump
operators, in the following sense: for any such f there is a pointclass Γ such
that f(x) ≡T the universal Γ(x) subset of  on a cone. (It follows that any
uniformly invariant function is uniformly order preserving. Becker’s results
require AD.) Steel [Ste82A] shows that Conjecture II holds when restricted
to uniformly invariant functions (Lachlan proved a special case of this result
earlier in [Lac75]). Thus the following conjecture implies Martin’s conjectures,
and seems to capture their true intent.
Conjecture III (Steel [Ste82A]). Assume AD. Then every degree invariant
function is uniformly degree invariant.
Incidentally, Conjectures I and II imply the restriction of Conjecture III to
arithmetic functions. For if f is invariant, then Conjectures I and II imply f is
constant on a cone or f(x) ≡T x n (the nth jump of x) on a cone, so in either

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 460 THEODORE A. SLAMAN AND JOHN R. STEEL

case f ≡m g for some uniformly invariant g. But then by Martin’s theorem we

can fix a pointed T and an (i, j) such that ∀x ∈ [T ](f(x) ≡T g(x) via (i, j)),
so f is uniformly invariant.
In this paper we shall prove some theorems related to the foregoing con-
jectures. In §2 we prove Conjecture I for uniformly invariant functions. We
also prove Conjecture I, and hence Conjecture III, for functions f such that
f(x) <T x on a cone. Finally, we prove Conjecture III, and hence Conjec-
ture II, for Borel functions which are increasing and order-preserving on a
cone. In §3 we prove that, under AD, there is no function assigning to each
degree d a linear order of d of order type  ∗ + . This strengthens the theorem
that there is no choice function on degrees, and is important for the conjectures
since such a function would yield counterexamples to them. In a sequel to this
paper, [SS], we shall prove one further technical result related to conjectures:
we shall construct an r.e. operator x → Wex which acts like a counterexample
to Conjecture II on degrees r.e. in 0 .

§2. Our first theorem completes the work of [Ste82A] on the uniformly
invariant case of Conjectures I and II.

Theorem 2.1. Assume AD. Suppose f is uniformly invariant and not

increasing on a cone. Then f is constant on a cone.

Proof. Martin’s theorem implies x T f(x) on a cone. We show first

that f(x) <T x on a cone. Let T be a pointed tree and t :  ×  →  × 
witness the uniform invariance of f on [T ]. Consider the game from [Ste82A]
comparing f to the identity function: Player I plays x ∈  and player
II plays y ∈ , the players alternating moves as usual. Let x(0) code
(e, n) ∈  ×  and let x − (n) = x(n + 1). Then player I loses unless x − ∈ [T ]
and {e}x = y; if player I succeeds in this then player II loses unless x ≤T y
and f(x − )(n) = 0 ⇐⇒ y(y(0)) = 0.
Suppose s were a winning strategy for player I. We shall show z ≤T f(z)
on the cone above (T, s), a contradiction. So fix z ∈ 2 such that T, s ≤T z.
Let (e0 , n0 ) be the first move dictated by s. For m ∈ , let ym = mz, and
let xm ∈ [T ] be such that s(ym ) = (e0 , n0 )xm . Now there is a fixed (a, b)
such that for all m, xm ≡T xm+1 via (a, b). [For example, to compute xm+1
from xm , compute {e0 }xm = ym = mz. Then compute s from z. Then
compute xm+1 = s(m + 1z)− . This procedure is independent of m (since
e0 computes m from xm ).] But then f(xm ) ≡T f(xm+1 ) via t(a, b), for all m.
So f(xm ) : m <  is recursive in f(z). Since s is winning for player I, we
have z(m − 1) = 0 iff ym (ym (0)) = 0 iff f(xm )(n0 ) = 0. Thus z ≤T f(z).
So by AD we have a winning strategy s for player II. We shall show f(z) ≤T z
on the cone above (T, s). So let T, s ≤T z. To compute f(z) from z, proceed
as follows: given n, find effectively, using the recursion theorem relative to z,

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 461

an e such that s((e, n)z) = {e}x . Let y = s((e, n)z). Then f(z)(n) = 0
iff y(y(0)) = 0; otherwise f(z)(n) = 1.
Thus f(z) <T x on a cone. Theorem 2.2 to follow implies f is constant on
a cone, and with no further need for the uniformity hypothesis on f. Its proof,
however, works only for Turing degrees. We give here a different proof which
does make use of the uniformity of f, but which works for other notions of
degree.
For each e ∈  let Pe = {x ∈ [T ] : {e}x = f(x)}; then by Martin’s
theorem we can fix an e and a pointed T  ⊆ T such that [T  ] ⊆ Pe . Thus f is
recursive on [T  ]. The usual splitting argument gives a pointed U ⊆ T  such
that either f is constant on U or f is one-one on U . If f is constant on U we
are done, so assume toward a contradiction that f is T one-one on U .
By passing to a subtree we may assume U is uniformly pointed, that
is, there is a fixed i such that ∀x ∈ [U ](U = {i}x ). We may also assume
f(x) <T x for x ∈ [U ]. Finally, let U ∗ = {s ∈ U : s 0 ∈ U ∧
s 1 ∈ U }. We may assume
there is a map ϕ ≤T U , ϕ : U → <2, such
that ∀x ∈ [U ](f(x) = n< ϕ(xn)) and ∀s, t ∈ U ∗ (s is incompatible with
t =⇒ ϕ(s) is incompatible with ϕ(t)).
Let V = {v ∈ <2 : ∃u ∈ U (v ⊆ ϕ(u))}, so that V is a tree and [V ] =

f [U ]. Clearly we can find an x0 ∈ [U ] such that x0 ≤T U and U ≤T
(V, f(x0 )). It will be enough to show V ≤T f(x0 ), since then x0 ≤T f(x0 )
and x0 ∈ [U ], a contradiction.
Notice that there is a recursive, Lipschitz homeomorphism  of 2 so
that for any x, {n (x) : n ∈ } is dense in 2. [Define  : <2 → <2 by
induction on length. Set (∅) = ∅. Let (u i) = (u)i unless u = ∅
or ∀m ∈ dom u(u(m) = 0); in that case set (u i) = (u)1 − i. Finally,
let (x) = n< (xn).] Fix a recursive in U isomorphism  : (<2, ⊆
) ' (U ∗ , ⊆); we also use  for the induced homeomorphism from 2 to
[U ]. Let (y0 ) = x0 . Notice that there is a fixed a such that for all y,
(y) ≡T ((y)) via a (this makes use of the uniform pointedness of U .) Thus
f((y)) ≡T f(((y))) via t(a) for all y, so that f((n (y0 ))) : n < 
is recursive in f((y0 ) = f(x0 ). But {n (y0 ) : n < } is dense in 2, so its
image under  is dense in [U ] and its image under f ◦  is dense in [V ] thus
v ∈ V ⇐⇒ ∃n(v ⊆ f((n (y0 ))))
so that V is r.e. in f(x0 ).
Say that u n-splits on V iff u ∈ V and {k ∈ dom u : uk 0 ∈ V ∧
uk 1 ∈ V } has cardinality at least n. Since V is r.e. in f(x0 ), so is
{(u, n) : u n-splits on V }. But

v∈ / V ⇐⇒ ∃n∃u0 . . . ∃un u0 = un ∧ ∀i ≤ n[ui ⊆ f((i (y0 )))

∧ ui is incompatible with v ∧ ui (dom v+2)-splits on V ]

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 462 THEODORE A. SLAMAN AND JOHN R. STEEL

so that V is co-r.e. in f(x0 ), finishing the proof. The left-to-right direction

of the displayed equivalence is obvious. Now suppose u0 , . . . , un are as on
the right. Let ri = i (y0 ) dom v. Since ϕ((ri )) is compatible with ui and
does not (dom v+2)-split on V , ϕ((ri )) ⊆ ui . Thus r0 = rn , and then since 
is Lipschitz, k (r0 ) = j (r0 ) if k ≡ j (mod n). Thus ∀k∃i ≤ n (ϕ((ri )) ⊆
f((k (y0 )))). But ϕ((ri )) is incompatible with v since ϕ((ri )) ⊆ ui , ui is
incompatible with v, and dom ϕ((ri )) ≥ dom v. Thus ∀k(v  f((k (y0 )))),
so that v ∈
/ V. 
The arguments of [Ste82A] adapt easily to most notions of degree coarser
than ≤T , and our proof of Theorem 2.1 adapts similarly. For example,
let x ≡Δ1n y via (ϕ, ) iff ϕ and  are Σ1n formulae such that (x(k) =
 k, j)) and (y(k) = j ⇐⇒ (x, 
j ⇐⇒ ϕ(y, k, j)). Suppose f : 2 → 2
and t : HF → HF are such that x ≡T y via a =⇒ f(x) ≡Δ1n f(y) via t(a).
Suppose x Δ1n f(x) on a (Turing) cone. Then, granting AD,  ∃y(f(x) ≡ 1
Δn
y on a cone).Notice that it only strengthens the hypothesis on f to require 
that x ≡Δ1n y via a ⇒ f(x) ≡Δ1n f(y) via t(a). A similar result holds for

constructibility degree if we let“x ≡L y via (ϕ, )” mean that L[x] = L[y]
and (x(k) = j ⇐⇒ L[y] |= ϕ[y, k, j, c0y , . . . , cny ]) and (y(k) = j ⇐⇒ L[x] |=
ϕ[x, k, j, c0x , . . . , cnx ]), where cix = ciy is the ith canonical indiscernible of
L[x] = L[y]. Arithmetic equivalence is the one anomaly we know of in this
area; the uniformly invariant cases of both Conjectures I and II for arithmetic
degrees are open. Let x ≡a y via (ϕ, ) ⇐⇒ ϕ and  are Σ0n formulae, for
some n, and (x(k) = j ⇐⇒ ϕ(y, k, j)) and (y(k) = j ⇐⇒ ϕ(x, k, j)). The
trouble with generalizing our proof of Theorem 2.1, or the proofs of [Ste82A],
is that there are b and xm : m <  such that ∀m(xm ≡a xm+1 via b) but
xm : m <  a x0 ; take, for example, xm = 0m .

Question 1. Do the analogues for uniformly arithmetically invariant func-

tions of Conjectures I and II hold?

We pass now to the non-uniform case of Martin’s conjectures.

Theorem 2.2. Assume AD. Let f be degree invariant and f(x) <T x on a
cone. Then f is constant on a cone.

Proof. For e ∈  let Pe = {x : f(x) = {e}x }. By Martin’s theorem we

have a fixed e and a pointed tree U such that [U ] ⊆ Pe . Thus f is recursive
on [U ]. This is the only use of AD in the proof we are giving.
We can thin U to a pointed T such that either f is constant on [T ] or f
is one-one on [T ]. If f is constant on T we are done, so assume toward a
contradiction that f is one-one on T . By further thinning we may assume that
f(x) <T x for some x ∈ [T ], and that there is an f ∗ : T → <2 recursive in

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 463

T such that

∀u, v ∈ [T ][(u ⊆ v =⇒ f ∗ (u) ⊆ f ∗ (v))

∧ (u incompatible with v =⇒ f ∗ (u) incompatible with f ∗ (v))]

and ∀x ∈ [T ](f(x) = f ∗ (xn)).
n<

The crux of our proof is embodied in the next lemma, which implies that
any h :  →  is dominated by a function recursive in some f(x).
Lemma 2.3 (Delay Lemma). Let T, f and f ∗ be as described above. Let
h :  →  and h ≤T x, where x ∈ [T ]. Then there is an e ∈  such that,
letting
(0) = 0
and
(n)
(n + 1) = (number of steps to compute {e}f(x) (i)),
i=0

The function  is total and for all sufficiently large n, h((n)) < (n + 1).
Proof. We shall define by initial segments a y ∈ [T ] such that y ≡T x. We
shall be “h-slow” about committing ourselves to f(y), and thereby guarantee
that if the conclusion of the lemma fails for e, then {e}f(x) = f(y). Since
{e}f(x) = f(y) for some e, this is enough.
By thinning T we may assume x ≡T T . For any s ∈ T let u(s) and v(s) be
incompatible extensions of s on T of minimal length, and let m(s) be the least
i such that
f ∗ (u(s))(i) = f ∗ (v(s))(i).

We now define sn ∈ T by induction on n; we then set y = n< sn . As we
define the sn ’s we declare various e ∈  dead. We being by setting s0 = ∅.
Now suppose sn is given. Let u = u(sn ), v = v(sn ), and m = m(sn ).
Case 1. There is an e ≤ n such that e is not yet dead and {e}f(s) (m)
converges in max{h(i) : i ≤ m(u) + m(v)} steps.
In this case, let e0 be the least such e. Let sn+1 be the unique r ∈ {u, v} such
that f ∗ (r)(m) = {e0 }f(x) (m), and declare e0 dead.
Case 2. Otherwise.
Then set sn+1 = u.
Now clearly y = n sn ∈ [T ]. Since T is pointed, x ≤T y. On the other
hand, sn : n <  is recursive in x, T , and h, hence in x. Thus y ≡T x. So
pick e such that {e}f(x) = f(y). We claim that e is satisfied in the lemma.

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 464 THEODORE A. SLAMAN AND JOHN R. STEEL

Let  be defined from e as in the statement of the lemma. Let j ≥ e be large

enough that no e  ≤ e is declared dead at some step n ≥ j. We claim that
h((n)) < (n + 1) for n ≥ m(sj ). So fix n > m(sj ). Pick k such that
m(sk ) ≤ (n) < m(sk+1 ).
Now since m(sj ) ≤ n ≤ (n), j ≤ k. At step k no e  ≤ e is declared dead.
Thus the hypothesis of case 1 does not apply to e at step k. Since sk+1 is one
of u(sk ) and v(sk ), we have (n) < m(u(sk )) + m(u(sk )). The failure of the
case 1 hypothesis for e means that {e}f(x) (m(sk )) takes at least h((n)) steps
to compute. By the definition of  then, h((n)) < (n + 1). 
Notice that the Delay Lemma immediately gives 0 ≤ f(x) on a cone (and
more). For let h(n) be the least m such that 0 ∩ n is enumerated by step m.
Let x ∈ [T ] and h ≤T x. Let  be as in the claim. Then  ≤T f(x), and since
from  we can compute a function dominating h, 0 ≤T .
For s ∈ < we define us ∈ T as follows: u∅ = ∅. Given us , let us
i
be the v ⊇ us , v ∈ T , of minimal length such that v(dom v − 1) = 0 and
v(dom v − 1)1 ∈ T and
{k ∈ dom v − dom us : v(k) = 1 ∧ vk 0 ∈ T }
has cardinality i. For h :  → , let

xh = us .
s⊆h

So xh turns right on T h(0) times, then left, then right h(1) times, then left,
etc.. Since T is pointed, h ≤T xh .
Our plan now is to construct h1 , h2 ≤T T so that if 1 and 2 come from
applying the Delay Lemma to (h1 , xh1 ) and (h2 , xh2 ), then T ≤T 1 ⊕ 2 . Since
xh1 ≡T xh2 ≡T T , this will be enough. The only trouble comes in guessing the
e1 and e2 appropriate for the h1 and h2 we are constructing. But since ei is a
Borel function of hi , we can fix ei on a nonmeager set. This turns out to be
good enough.
So for any h :  →  let eh be the least e satisfying the conclusion of the
Delay Lemma for h, eh . Let
h (0) = 0,
h (n)
h (n + 1) = (number of steps to compute {eh }f(x) (i)).
i=0

Finally, let mh be the least m such that h(h (n)) < h (n + 1) for all n ≥ m.
Since the function h → (eh , mh ) is Borel, it is constant on a nonmeager set.
Thus we fix an r ∈ < and e, m ∈  so that for any s ⊇ r, s ∈ <, there is
an h ⊇ s such that e = eh and m = mh .

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 465

For s ∈ <, let

s (0) = 0

and
s (n)
∗
s (n + 1) = (number of steps to compute {e}f (us )
(i)),
i=0

where s (n + 1), and hence s (k) for k ≥ n + 1, are considered undefined if for
∗
some i ≤ s (n){e}f (us ) (i) does not converge within dom f ∗ (u
s ) steps. Notice
that s ⊆ t =⇒ s ⊆ t , and if r ⊆ h and e = eh , then h = n< hn . Also,
the function s → s is recursive in T . Also
s (i) = k =⇒ sk (i) = k. (∗)
This is true because if s (i) = k, then the e-computations which verify this only
use f ∗ (us )k, which is determined by us k and thus by sk. Notice finally
that
if r ⊆ t, i ≥ m, and t (i) ∈
/ dom t, then i + 1 ∈
/ dom t . (∗∗)
For suppose that t (i + 1) = k. Pick h ⊇ t such that e = eh and m = mh
and h(t (i)) > k. Then t ⊆ h , and as i ≥ m = mh , h(h (i)) < h (i + 1) a
contradiction.
Fix now a p ⊇ r and an m0 ≥ m such that m0 ∈ dom p and p (m0 ) = dom p.
[Take any h ⊇ r such that e = eh . Let m0 ≥ m be such that h (m0 ) ≥ dom r.
Let p = hh (m0 ) and apply (∗).] Fix also an x ∈ 2 such that x ≡T T . We
now define si , ti ∈ < by induction on i (and recursively in T ). Let
s0 = t0 = p.
Suppose now we are given si and ti such that m0 + i ∈ dom si ∩ dom ti and
dom si = si (m0 + i) and dom ti = ti (m0 + i). It follows by (∗∗) then that
m0 + i + 1 ∈
/ (dom si ∪ dom ti ).
Case 1. x(i) = 0.
Find recursively in T an s ⊇ si such that s (m0 + i + 1) is defined. Let
si+1 = ss (m0 + i + 1).
Say s (m0 +i +1) = k. Find recursively in T a t ⊇ ti k such that t (m0 +i +1)
is defined. Set
ti+1 = tt (m0 + i + 1).
Notice that by (∗) our induction hypotheses still hold of si+1 and ti+1 . Also
si+1 (m0 + i + 1) = k < ti+1 (m0 + i + 1)
since there is an h ⊇ ti+1 with e = eh and m = mh .
Case 2. x(i) = 1.

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 466 THEODORE A. SLAMAN AND JOHN R. STEEL

of si and ti .
Reverse the roles
Now let h = i< si and g = i< ti . By construction, g, h ≤T T . So
xh ≡T xg ≡T T . Let 1 = i< si and 2 = i< ti .
By construction 1 and 2 are total; by their definitions 1 ≤T f(xh ) and
2 ≤T f(xg ). By construction,
x(i) = 0 ⇐⇒ 1 (m0 + i) > 2 (m0 + i).
So T ≤T x ≤T f(xh ) ⊕ f(xg ). Since f(xh ) ≡T f(xg ), xh ≤T f(xh ),
contradicting the fact that xh ∈ [T ].  (Theorem 2.2)
Our proof of Theorem 2.2 seems to give no information about the analogous
question for other notions of degree.
Question 2. Let f : 2 → 2 be such that ∀x∀y(x ≡Δ11 y =⇒ f(x) ≡Δ11
 be a y such that
f(y)) and f(x) < 1 x on a cone. Granting AD, must there 
Δ1

f(x) ≡Δ11 y on a cone?

We turn now to Conjecture II and order-preserving functions. For x ∈ 2,
let 1x be the least ordinal not recursive in x. For α < 1x , let x α be the αth
Turing jump of x; that is x α = Hex where e is the least i ∈ Ox such that i is a
notation for α. We shall need the following slight generalization of a theorem
of Posner and Robinson ([PR81]; cf. also [JS85] and [SS89]).
Theorem 2.4 (Posner, Robinson). Let 1 ≤ α < 1x , and suppose x  <T y
for all  < α. Then there is a z ≥T x such that z α ≡T (y, z).
Theorem 2.5. Assume AD. Suppose f : 2 → 2 is order preserving and
increasing on a cone. Then either
(a) ∃α < 1 (f(x) ≡T x α ) on a cone, or
(b) ∀α < 1x (x α <T f(x)) on a cone.
Proof. Suppose that (b) is false. Martin has shown that if h : 2 → 1 and
x ≡T y =⇒ h(x) = h(y) and h(x) < 1x on a cone, then ∃α(h(x) = α on a
cone). (By Martin’s theorem we can fix an e and a pointed T such that {e}x
codes a wellorder of type h(x) for all x ∈ [T ]. By boundedness, h(x) < 1T
for all x ∈ [T ]. But 1T is countable, so applying Martin’s theorem again we
have a pointed S ⊆ T such that h is constant on [S].) So we can fix an α < 1
such that on a cone, x α <T f(x) but x  <T f(x) for all  < α. Pick any x in
this cone, and let y = f(x). The Posner–Robinson theorem yields a z ≥T x
such that z α ≡T (y, z). Now z ≤T f(z), and y = f(x) ≤T f(z) since f is
order preserving. Thus z α ≤T f(z), so that z α ≡T f(z) by our choice of α
and x. We have shown then that ∀x∃z ≥T x(f(z) ≡T z α ), so that f(z) ≡T z α
on a cone by Martin’s theorem. 
Corollary 2.6. Assume AD. Let f be increasing and order preserving, and
suppose f(x) ∈ Δ11 on a cone. Then f is uniformly invariant.


Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 467

Theorem 2.5 and its corollary beg for improvement. Woodin (unpublished)
has extended the Posner–Robinson theorem by showing that if x α <T y for
all α < 1x , then ∃z ≥T x(Oz ≡T (y, z)). It follows that we can strengthen
alternative (b) of Theorem 2.5 to read “Ox ≤T f(x) on a cone”. It is plausible
that these ideas will lead to a proof of the strengthening of the corollary obtained
by replacing “Δ11 (x)” by “L[x]”. The following proposition is relevant to the
attempt to do  this.
Proposition 2.7. Assume AD. Let g : 2 → Ord be such that x ≡T y =⇒
g(x) = g(y). Then for a cone of x, {y : g(x) = g((x, y))} is comeager.
Proof. For a cone of x, z ≥T x =⇒ g(z) ≥T g(x), as otherwise we get a
descending chain of ordinals. If for a cone of x, {y : g(x) = g((x, y))} is
nonmeager, then the Kuratowski–Ulam theorem gives a sequence yi : i < 
such that for some x, g((x, yi : n ≤ i < )) > g((x, yi : n + 1 ≤ i < )).
This again is a descending chain of ordinals. 
The proposition says that, under AD, Cohen forcing preserves all ordinal
assignments. (If Cohen had known enough about AD, this might have
motivated him!) It might be interesting to find analogues of the proposition for
other notions of forcing. The generalization of the Posner–Robinson theorem
needed to push Theorem 2.5 and its corollary through L[x] seems to require
this.
Of course, something more general, to which the Posner–Robinson theorem
seems irrelevant, must be true.
Question 3. Assume AD. Let F : 2 → 2 be increasing and order preserving
on a cone. Must f be uniformly invariant?
In view of Theorem 2.5 and its corollary, the answer is almost certainly “yes”.
We shall now attempt to clarify the relationships between Theorems 2.1,
2.2 and 2.5, and the conjectures. To construct a counterexample to any of
the conjectures is to construct a degree invariant function f. We can regard
such a construction as a family of constructions, one for each x ∈ 2, done
simultaneously; the x-construction has “coding requirements”, designed to
guarantee y ≡T x =⇒ f(y) ≡T f(x), and additional requirements designed
to guarantee whatever else we have in mind.
In this light, Theorem 2.1 and the results of [Ste82A] show that the x-coding
requirements must be satisfied in a way that is not uniform over [T ] for
any pointed T . It is at least superficially plausible that the injuries to these
requirements during the x-construction could produce this non-uniformity.
Theorem 2.2 shows that if the x-construction is to be recursive in x, then
the coding requirements become so difficult to guess that we can only meet
them uniformly. The reason is that the sets to be coded by f(x), that is, the
f(y) for y ≡T x, can appear very slowly. This fact, embodied in the Delay
Lemma, is the key to our proof. Unfortunately, there is no such limitation

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 468 THEODORE A. SLAMAN AND JOHN R. STEEL

on nonrecursive constructions. The following open question illustrates the

shortcomings of Theorem 2.2 in this regard. Let x ≡m y ⇐⇒ {n : x(n) = 0}
has the same many-one degree as {n : y(n) = 0}.
Question 4. Is there a continuous, one-one f : 2 → 2 such that for a cone
of x, ∀y(x ≡m y =⇒ f(x) ≡T f(y)) and 0 T f(x)?
Notice that, given such an f, the function g(x) = f(x  ) is a counterexample
to Conjecture I.
Theorem 2.5 shows that if we require f(y) ≤T f(x) for y ≤T x, then again
the coding requirements overwhelm everything else. The reason is that this
imposes a burden on the y-construction as well as the x-construction; the y-
construction must guarantee that f(y) is tame enough that f(x) can afford to
code it. Thus the y-construction must cooperate with 2ℵ0 other constructions,
a heavy burden indeed. Of course, if we want only f(x) ≡T f(y) for y ≡T x,
then every construction need only cooperate with countably many others, and
we have some hope. As a diluted form of cooperation within a degree, we offer
the following proposition. (We also mention this proposition because it rules
out one natural approach to proving Conjecture II.) Let Wex be the eth subset
of  r.e. in x.
Proposition 2.8. There is an e such that whenever x1 ≡T x2 ≡T · · · ≡T xn ,
then
x1 <T Wex1 ⊕ · · · ⊕ Wexn <T x1
One can obtain such an e by recursively associating to each x an infinite
injury construction of Wex , and arranging for the x and y constructions to
cooperate when x ≡T y. (One can also simply cite the relativization of a
theorem of Cooper and Yates; cf. [Mil81]: there is a nonrecursive r.e. A
such that for all r.e. B <T 0 , A ⊕ B <T 0 .) In contrast, notice that the
Posner–Robinson theorem implies that if x <T f(x) for all x, then there are y
and x such that y ≤T x and x  ≤T f(y) ⊕ f(x).
We close this section with a few remarks on arbitrary, not necessarily
definable, functions on degrees. So we assume AC for the duration of this and
the next paragraph. The proof of Theorem 2.5 gives: if f is order preserving
and x <T f(x) on a cone, then ∀y∃x ≥T y(x  ≤T f(x)). Thus, even with AC,
one cannot construct an order preserving g : D → D such that d < g(d) < d
for all d. The proof of Theorem 2.2 gives no information about undefinable
“pressdown” functions (functions f such that f(d) < d on a cone). Of course
the Friedberg jump inversion theorem implies there is a function f such that
f(d) = d for d ≥ 0 , and thus a pressdown function one-one on the cone above
0 . We know of one limitation on possible undefinable pressdown functions.
Proposition 2.9. Suppose f(d) < d on a cone. Then it is not the case that
∀c(c ≤ f(d) on a cone).

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 469

Proof. Suppose ∀c(c ≤ f(d) on a cone). Let d0 be arbitrary, and let di+1
be such that di ≤ f(d) whenever di+1 ≤ d. Now let u be a minimal upper
bound for {di : i ∈ } (cf. [Sac66]). By the definition of the di ’s f(u) is an
upper bound for {di : i ∈ }. Thus f(u) <T u. Since d0 was arbitrary, f is
not a pressdown function. 
We don’t know whether there are nontrivial order preserving pressdown
functions.
Question 5. (a) Is there a pressdown function which is order preserving
and one-one on a cone?
(b) Let I ⊆ D be a countable jump ideal (i.e., closed under join, Turing
jump, and downward under Turing reducibility). Let A ⊆ I be such that
∀b ∈ I (A  {c : c ≤ b}). Must there be an upper bound u for I such that no
v < u is an upper bound for A?
A positive answer to (b) would strengthen Sacks’ theorem on the existence
of minimal upper bounds (cf. [Sac66]). One could use this strengthening as in
the proposition above to obtain a negative answer to (a).

§3. In order to construct a degree invariant function we must assign to each

degree d a structure on d telling the x-constructions for x ∈ d how to cooperate
with one another. The structure easiest to use would be a distinguished element
of d, or equivalently, a wellorder of d; however, AD implies there is no function
assigning all d to such a structure. A somewhat weaker structure, which still
suffices to construct counterexamples to the conjectures, is a linear order of d
of order type  ∗ + . We now show that, assuming AD, there is no function
putting this weaker structure on an arbitrary degree.
The following reformulation will be useful. A resolution of an equivalence
E is a sequence En : n <  of equivalence relations such that
relation
E = n< En . We call a resolution En : n <  finite iff ∀x∀n([x]En is finite).
Since we are not assuming AC, the next lemma has content.
Lemma 3.1. Let E be an equivalence relation on 2. Then the following are
equivalent:
(a) There is a finite resolution of E,
(b) There is a function assigning to each E-equivalence class d a linear order
of d of order type  ∗ + .
Proof. (a) =⇒ (b): Let En : n <  be a finite resolution of E. Let <lex be
the usual lexicographic linear order of 2. Let d be a given E-equivalence class.
For x, y ∈ d let n be least such that xEn y. If n = 0, put x <d y ⇐⇒ x <lex y.
If n = m + 1, let w and z be the <lex -least elements of [x]Em and [y]Em
respectively, and put x <d y ⇐⇒ w <lex z.
It is easy to check that <d has order type  ∗ + , , or  ∗ . In the latter
cases we can pass canonically to a linear order of d of order type  ∗ + .

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 470 THEODORE A. SLAMAN AND JOHN R. STEEL

(b) =⇒ (a): Suppose that for all E-equivalence classes d, <d is a linear order
of d of order type  ∗ + . For xEy, let
D(x, y) = Card({z : x <d z <d y or y <d z <d x}).
For d an E-equivalence class, let
Td = {s ∈ <2 : ∃x ∈ d(s ⊆ x)}
and
xd = the leftmost branch of Td .
We define En ∩ (d × d) by cases on the relation of xd to <d .
Case 1. xd ∈ d. Then for y, z ∈ d
yEn z ⇐⇒ y = z ∨ (D(y, xd ) < n ∧ D(z, xd ) < n).
Case 2. xd ∈ / d and there is a k ∈  and y ∈ d such that z <d y =⇒
zk = xd k. Then let k be least such that some such y exists, and let yd be the
<d -largest such y for k. Set
yEn z ⇐⇒ y = z ∨ (D(y, yd ) < n ∧ D(z, yd ) < n).
Case 3. Cases 1 and 2 fail to hold, and there is a k ∈  and y ∈ d such that
z >d y =⇒ zk = xd k. Proceed as in Case 2.
Case 4. Otherwise. Then for y, z ∈ d, let
yEn z ⇐⇒ y = z ∨ ∀w[(y <d w ≤d z ∨ z <d w ≤d y) =⇒ wn = xd n].
It is not difficult to check that En : n <  is a finite resolution of E. 
The relation xEy ⇐⇒ y = z ∨∃m∀n ≥ m(x(n) = y(n)) is a simple example
of an equivalence relation admitting finite resolution. It is fairly easy to show
that if E admits a finite resolution, then there are pathological E-invariant
functions; for example, functions f such that xEy =⇒ f(x) ≡T f(y) and
x <T f(x) <T x  for all x and y.
We shall show AD implies ≡T admits no finite resolution. This phrasing
is misleading, however; it seems one must prove a stronger conclusion from
a weaker hypothesis. If G is a group of homeomorphisms of 2, then let
x ∼G y ⇐⇒ ∃ ∈ G((x) = y). If the homeomorphisms in G are recursive,
then ∼G refines ≡T . We shall show that if G is modestly complicated, then ∼G
admits no finite resolution. The relevant hypothesis is that all sets of reals are
Lebesgue measurable.
Consider an alphabet with symbols , −1 , , and  −1 . A reduced word is
a sequence a0 , . . . , ak  of symbols such that for no i < k do we have ai = 
and ai+1 = −1 , or ai = −1 and ai+1 = , or ai =  and ai+1 =  −1 , or
ai =  −1 and ai+1 = . Given homeomorphisms ¯ and ¯ of 2, let −1 = ¯ −1
and  −1 = ¯ −1 and for w = a0 · ak  a reduced word, let w̄ = āk ◦ · · · ◦ ān0 .
Notice that if ¯ and ¯ are Lipschitz, so that (x)
¯ = n< (xn)
¯ where  ¯ 2
is a permutation of n2 and similarly for , ¯ then w̄ is Lipschitz and w̄n2 is

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 471

determined by  ¯ n2. We call a pair {,

¯ n2 and  ¯ }
¯ of homeomorphisms
independent iff whenever w is a reduced word then w̄(x) = x for all x ∈ 2;
equivalently iff whenever s and v are distinct reduced words, then w̄(x) = v̄(x)
for all x ∈ 2.
Lemma 3.2. There is an independent pair of recursive, Lipschitz homeomor-
phisms of 2.
Proof. We define ¯ n2 and 
¯ n2 by induction on n. Let R be the set of
reduced words, and < a recursive wellorder of R × <2 in order type .
Step 1. Set
(0)
¯ = (0)
¯ = 1 and (1)
¯ = (1)
¯ = 0.
Step n + 1. Let w, s be the <-least pair such that dom s ≤ n, w̄(t) = t for
some t ∈ n2 such that s ⊆ t, and ū(s) = v̄(s) for all distinct initial segments u
and v of w. We call w, s the critical pair at n + 1. Let
F = {t ∈ n2 : s ⊆ t ∧ w̄(t) = t},
and for t ∈ F ,
Ct = {ū(t) : u is an initial segment of w}.
Thus t ∈ Ct and Ct has dom w distinct elements, all in n2. Notice that if r =  t
then Cr ∩ Ct = ∅. For otherwise ū(r) = v̄(t) for u, v initial segments of w,
and since s ⊆ r and s ⊆ t, and ū, v̄ preserve ⊆, ū(s) = v̄(s). Since w, s is
critical, u = v. Since ū(r) = ū(t) and ū is one-one, r = t.
Let w = a0 , . . . , ak . It is now reasonably easy to see that we can extend ¯

and ¯ to {u ∈ n+12 : un ∈ t∈F Ct } so that for t ∈ F and i ∈ {0, 1},
ā0 (t i) = ā0 (t)1 − i
and
āj (wj(t)i) = w(j + 1)(t)i for 1 ≤ j ≤ k.
(A problem might arise if a0 = ak−1 ; but then for t ∈ F , āk (ā0 (t)) = t =
āk (wk(t)), so ā0 (t) = wk(t), so ā0 (s) = wk(s), so a0 = wk, and
w = a0 a0−1 is not reduced.)
We complete step n + 1 by extending ¯ and ¯ arbitrarily to all of n+12. Notice
w̄(t) = t for all t ∈ n+12 such that s ⊆ t, so that w, s is never again critical.
This completes the definition of ¯ and . ¯ Suppose w̄(x) = x for some
reduced w and x ∈ 2. We may assume ū(x) = v̄(x) for all distinct initial
segments u and v of w by passing to a shorter word if necessary. Then there
is an n such that (w, xn) is eligible at all steps ≥ n + 1 to be critical. Since
every pair is critical at most once, (w, xn) is critical at some step, hence never
eligible to be critical again, a contradiction. 

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 472 THEODORE A. SLAMAN AND JOHN R. STEEL

It is just a matter of more notation to construct a family of 2ℵ0 Lipschitz

homeomorphisms which is independent in the obvious sense.
Theorem 3.3. Assume all sets of reals are Lebesgue measurable. Let G be a
group of homeomorphisms of 2 containing an independent pair of Lipschitz
homeomorphisms. Then ∼G admits no finite resolution.
Proof. Suppose En : n <  is a resolution of ∼G . Let {,
¯ }
¯ be an
independent pair of Lipschitz homeomorphisms from G. We assume ¯ and ¯
have been extended so as to act on <2.
For any x ∈ 2, let nx be the least m such that {(x),
¯ (x),
¯ ¯ −1 (x),
−1
¯ (x)} ⊆ [x]Em . Then we can fix an n and a C ⊆ 2 closed such that C


has measure ≥ 34 and nx < n for x ∈ C . Let T = {s ∈ <2 : ∃x ∈ C (s ⊆

x)}, so that T is a tree on {0, 1} and [T ] = C . Then for any symbol
a ∈ {, −1 , ,  −1 } we have
xEn ā(x), ∀x ∈ [T ].
We now define a tree U on {0, 1} × {, , −1 ,  −1 } as follows:

(s, w) ∈ U ⇐⇒ ∃k(s ∈ k2 ∩ T ∧ dom w = k ∧ w is a reduced word

∧ v̄(s) ∈ T for every initial segment v of w).
Clearly U is finitely branching. If U is infinite, then we have an (x, f) ∈ [U ].
Let wi = f(i + 1). Then x ∈ [T ] and w̄i ∈ [T ] for all i, so
xEn w̄0 (x)En w̄1 (x)En · · ·
On the other hand, since {, ¯ }
¯ is independent and each wi is a reduced
word, w̄j (x) = w̄i (x) for j = i. So [x]En is infinite, and we’re done.
So it is enough to show U is infinite. Let i be given; we seek a node of U on
level i. We may assume i is large enough that
∀a ∈ {, −1 , ,  −1 }∀s ∈ i 2(ā(s) = s)
and
∀a, b ∈ {, −1 , ,  −1 }∀s ∈ i 2(a = b =⇒ ā(s) = b̄(s)).
Consider the graph (that is, symmetric, irreflexive binary relation) E on i 2
defined by
(s, t) ∈ E ⇐⇒ ∃a ∈ {, −1 , ,  −1 } (ā(s) = t).
For each s there are exactly 4 t’s such that (s, t) ∈ E, so there are 4 · 2i pairs
in E. So the number of edges of E (that is, unordered pairs {s, t} such that
(s, t) and (t, s) are in E) is 2 · 2i . Now let F be the subgraph of E defined by
(s, t) ∈ F ⇐⇒ (s, t) ∈ E ∧ s ∈ T ∧ t ∈ T.

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 473

Since [T ] has measure ≥ 34 , i 2 − T has no more than 14 2i members. So in

passing from E to F we remove no more than 4( 14 2i ) edges from E. Thus F
has at least 2i edges.
A well known easy induction shows that any graph on k vertices with at least
k edges contains a circuit. (That is, a sequence v0 , . . . , vm  of vertices such
that v0 = vm , {vj , vj+1 } is an edge for j < m, and {vj , vj+1 } =  {v , v+1 } for
j,  < m such that j = .) Let s0 , . . . , sm  be a circuit in F . For j < m there is a
unique aj ∈ {, −1 , ,  −1 } such that āj (sj ) = sj+1 . Let v = a0 , . . . , am−1 .
Then v is reduced, as otherwise {sj , sj+1 } = {sj+1 , sj+2 } for some j, and
v 2 = v v, v 3 = v v v, etc., are reduced since otherwise {s0 , s1 } = {sm−1 , sm }.
Let w = v i i. Then (s0 , w) is on level i of U , and we are done.  (Theorem 3.3)
We say x, y ∈ 2 are recursively isomorphic if there is a recursive permutation
 of  such that ∀n(x(n) = 0 ⇐⇒ y((n)) = 0). If E and F are equivalence
relations, then F is coarser than E ⇐⇒ E ⊆ F .
Corollary 3.4. Assume all sets of reals are Lebesgue measurable. Then no
equivalence relation on 2 coarser than recursive isomorphism admits a finite
resolution.
Proof. If F is coarser than E, and F admits a finite resolution, so does E.
Let G be the group generated by an independent pair of recursive Lipschitz
homeomorphisms. Then ≡T is coarser than ∼G . Since ∼G admits no finite
resolution, neither does ≡T .
Now suppose En : n <  is a finite resolution of recursive isomorphism.
Let xFn y ⇐⇒ x  En y  . Then Fn : n <  is a finite resolution of ≡T , a
contradiction. The corollary follows. 
By the way, one can show that, assuming all sets are Lebesgue measurable
(or all sets have the Baire property) that if G is a group of homeomorphisms
of 2 such that every ∼G class is dense, then there is no function picking
a member from each ∼G class. [If such a function exists, then there is a
linear order <∗ of the ∼G classes. Let (s, t) ∈ <2 × <2 be such that, e.g.,
{(x, y) : [x]∼G <∗ [y]∼G } is comeager in the neighborhood determined by
(s, t). A symmetry argument gives a contradiction. This proof is due perhaps to
Sierpinski and Mycielski.] Letting G be generated by a single homeomorphism,
we get an example where ∼G admits a finite resolution (by Lemma 3.1), but no
choice function on equivalence classes. We would guess that there is a model
of ZF+DC in which ≡T admits a finite resolution, but no choice function on
equivalence classes.
The following two questions occur naturally if one attempts to push The-
orem 2.1 closer to a proof of Conjectures I, II and III. Call a resolution
En : n <  recursively finite iff ∀x∀n(there is no sequence yi : i <  ≤T x
of distinct elements of [x]En ).
Question 6. Assume AD. Does ≡T admit a recursively finite resolution?

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 474 THEODORE A. SLAMAN AND JOHN R. STEEL

Question 7. Let G be a group generated by an independent pair of recursive,

Lipschitz homeomorphisms of 2. Is there a continuous, one-one f : 2 → 2
such that for all x, y
x ∼G y =⇒ f(x) ≡T f(y)
and
0 T f(x)?
(If G is generated by a single homeomorphism, then there is an f as called
for in Question 7, essentially because ∼G admits a finite resolution.)
The next question seems unrelated to the conjectures, but is perhaps of some
interest.
Question 8. Assume AD. Suppose we have a function assigning to each
Turing degree d a linear order <d of d. Then
(a) must the rationals embed order preservingly in <d , for a cone of d’s?
(b) must there be a linear order < of 2 such that <d = < ∩ (d × d) on a
cone?
(A positive answer to (b) implies a positive answer to (a); if < is a linear
order of 2 then, assuming all sets have the Baire property, the rationals embed
into < ∩ (d × d) on a cone.) We cannot at the moment rule out a function
assigning to each d a linear order of d of order type ( ∗ + )( ∗ + ).

REFERENCES

Howard S. Becker
[Bec] Pointclass jumps, Manuscript.

Carl G. Jockusch and Richard A. Shore

[JS85] REA operators, r.e. degrees, and minimal covers, Proceedings of Symposia in Pure Mathe-
matics, vol. 42 (1985), pp. 3–11.

Alistair H. Lachlan
[Lac75] Uniform enumeration operations, The Journal of Symbolic Logic, vol. 40 (1975), no. 3,
pp. 401– 409.

Donald A. Martin
[Mar68] The axiom of determinateness and reduction principles in the analytical hierarchy, Bulletin
of the American Mathematical Society, vol. 74 (1968), pp. 687–689.

David P. Miller
[Mil81] High recursively enumerable degrees and the anticupping property, Logic Year 1979–80.
Proceedings of the Logic Year and of the Conference on Mathematical Logic, November 11–13,
1979, at Storrs (M. Lerman, J. H. Schmerl, , and R. I. Soare, editors), Lecture Notes in
Mathematics, vol. 859, Springer, 1981, pp. 230–245.

David B. Posner and Robert W. Robinson

[PR81] Degrees joining to 0 , The Journal of Symbolic Logic, vol. 46 (1981), no. 4, pp. 714–722.

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015
 DEFINABLE FUNCTIONS ON DEGREES 475

Gerald E. Sacks
[Sac66] Degrees of unsolvability, Princeton University Press, Princeton, N.J., 1966.
[Sac67] On a theorem of Lachlan and Martin, Proceedings of the American Mathematical Society,
vol. 18 (1967), pp. 140–141.

Theodore A. Slaman and John R. Steel

[SS] A construction degree invariant below 0 , manuscript.
[SS89] Complementation in the Turing degrees, The Journal of Symbolic Logic, vol. 54 (1989), no. 1,
pp. 160–176.

John R. Steel
[Ste82A] A classification of jump operators, The Journal of Symbolic Logic, vol. 47 (1982), no. 2,
pp. 347–358.

DEPARTMENT OF MATHEMATICS
UNIVERSITY OF CALIFORNIA
BERKELEY, CALIFORNIA 94720-3840, USA
E-mail: [email protected]
E-mail: [email protected]

Downloaded from https://www.cambridge.org/core. National University of Singapore (NUS), on 07 Sep 2020 at 03:04:13, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139519694.015