You are on page 1of 13

A Classification of Jump Operator

Author(s): John R. Steel


Source: The Journal of Symbolic Logic, Vol. 47, No. 2 (Jun., 1982), pp. 347-358
Published by: Association for Symbolic Logic
Stable URL: https://www.jstor.org/stable/2273146
Accessed: 07-09-2020 03:07 UTC

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2273146?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend
access to The Journal of Symbolic Logic

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
THE JOURNAL OF SYMBOLIC LOGIC
Volume 47, Number 2, June 1982

A CLASSIFICATION OF JUMP OPERATORS

JOHN R. STEEL

?0. Introduction. The structure (D, <) of the Turing degrees under Turing
reducibility is quite complicated. This is true even if we restrict our attention to
the substructure (R, <) of r.e. degrees. However, the theorems which imply that
these structures are complicated all involve ad hoc constructions of sets having the
desired reducibility relations. The complexity disappears when we turn to de-
grees occurring in nature. Of the degrees in R, only 0 and 0' seem natural. In
D, only 0, 0', 0", ..., O, Ow+1, ... (and on into the transfinite) seem natural.
Thus the natural degrees appear to be wellordered, with successors given by Turing
jump.
If this is true, one would like to prove it. Of course the first problem is to make
the concept of naturalness more precise. The following requirements seem plausi-
ble: a natural degree should be definable, its definition should relativise to an
arbitrary degree, and this relativisation should preserve reducibility relations
among natural degrees. Thus to each natural degree c is associated a definable
fc: D -- D so that fc(O) = c and Vd(d < fc(d)). Moreover, b < c iff Vd(fb
fc(d)). To be specific, let us take the definability offc to mean thatfc e L(R).
If P is a property of degrees, we say P holds almost everywhere (a.e.) iff
AcVd > c P(d). For , g: DU-D, let f <mg iff f(d) < g(d) a.e. Define ' by
f'(d) = f(d)', and let M = {f: D -DI f E L(R) A d < f(d) a.e.}. The following
conjecture is due to D. A. Martin:
CONJECTURE. M is prewellordered by < m. If f E M has rank ca in < me then f'
has rank a + 1.
Given the requirements on naturalness above, the conjecture implies that the
natural degrees are wellordered, with successors given by Turing jump.
Martin's conjecture is meant to be considered under the hypothesis that all
games in L(R) are determined (it easily implies that L(R) I= R is not wellorderable).
Assuming this, we shall prove the conjecture for a certain subclass J of M. J
consists of those functions in M representable by an F: ic) -I we which is uniform
Turing invariant (cf. ?1). Such a representation is characteristic of all the familiar
jump operators, and in fact these are the only functions we know of in J. We
conjecture that J = M.
The most concrete special case of Martin's conjecture appeared in [8]. There
Sacks asked whether there is a degree invariant solution to Post's problem, i.e. an

Received January 2, 1980.

347
? 1982, Association for Symbolic Logic
0022-48 12/82/4702-0008/$02.20

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
348 JOHN R. STEEL

e e co so that (i) Vx(x < T We <


Lachlan showed there is no unifo
(i) and (ii)'

3t(t is recursive A Vx, yVi(x T y via i =: We- T WY via t(i))).

The natural e such that Vx(Wx = x') satisfies (i) and (ii)', and so Lachlan's result
characterizes the Turing jump operator.
Our proof of Martin's conjecture for J will reprove the Lachlan result. In this
special case our proof requires the determinacy of XO games, making it much less
constructive than Lachlan's. On the other hand, our proof is much simpler.
Our use of determinacy is "local"; for example, to prove that < m prewellorders
the Borel functions in J we need Borel determinacy. For the remainder of the
paper we shall actually work in ZF + AD + DC, where AD is the false assertion
that all games on cl are determined. We leave it to the reader to reformulate our
results so that they follow from the amount of definable determinacy he is willing
to grant. The constructivist who will not grant X2 determinacy should read no
further. The reader who will not grant zIt determinacy should skip ?2.
In ? 1 we prove that < m prewellorders J, and give characterizations of a number
of jump operators analogous to Lachlan's characterization of the Turing jump
operator. We also define a prewellorder of theories in the language of second-
order arithmetic which is based upon <,m. In ?2 we give a "normal form" for
those f E J so that f(d) e L[d] a.e. We then compute the ordinal ranks in < m of
various jump operators.

?1. The prewellorder of jump operators. We identify subsets of cl with their


characteristic functions and thus with elements of wcw. For e E co and x E wco, fe}x
is the eth partial recursive function relative to x, Wx = dom({e}x), and x' =
fel e e Wx}. Let x?TY iff 3e(fe}Y = x), X--TY iff (x <Ty A y <Tx), and
dg(x) = fyi x Ty}. D = fdg(x)I x e @w} is the set of (Turing) degrees; D h
an order induced by < T. We say x-T y via (m, n) iff x = {m}Y and y = fn}x
We tacitly identify (on with cw and (,wc)n with cWW via recursive bijections.
We work henceforth in ZF + AD + DC. Let p = (A c DI 3cVd > c (d e A
By Martin [7], ,u is a countably complete ultrafilter on D. The abbreviation "a.e."
stands for "almost everywhere with respect to ,u".
DEFINITION. Let F: wI) - wI and z e wcI. We say F is uniformly degree invariant
above z iff 3x, y ? T ZtU t e woe((z ?<T t A t =T u via e) => (F(t), y) -T
(F(u), y) via x(e)). We say in this case that (x, y) witnesses the uniformity of F
above z.
DEFINITION. Letf: D -- D. Thenf e J iff d < f(d) a.e., and 3F: -I) wcw3x CI)
(F is uniformly degree invariant above x A Vy 2 T x (dg(F(y)) = f(dg(y))). We say
in this case that F representsf above x.
We call the elements of J jump operators. All the familiar jump operators (e.g.
the Turing jump operator, the co-jump operator, the Ak-jump operator, and the
sharp operator) are in J. In fact, we know of no f: D -+ D such that d < f(d) a.e.
andf? J.
Two further pieces of notation are convenient. For x e wc, let x- = {i[x(i + 2)].

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
A CLASSIFICATION OF JUMP OPERATORS 349

If or is a strategy in a game on wj, and x E Zw, let ov(x) E A


play when one's opponent plays x(O), x(1),.
LEMMA 1.Letf,g EJ. Thenf <?gorg <m f
PROOF. Let F and G represent f and g respectively, above x E Ecw. Consid
game rC'(F, G): Player I produces y E act and Player II produces z E ww, the players
alternating moves in the traditional way. Player I loses unless (y(l)}y- = z-. If
{y(l)}y- = z-, then II loses unless {z(l)}z- = y-. If neither player loses for these
reasons, then II wins iff (y(O) E F(y-) z(O) E G(z-)).
Suppose first that II has a winning strategy a in 0?(F, G). We claim that whenever
a, X <T Y then F(y) <T G(y), and thusf <m g. For suppose or, x <T Y. We give
an informal algorithm for deciding whether m E F(y) using G(y) as an oracle.
Let m be given. Using the recursion theorem relative to y, find effectively in y
(hence effectively in G(y)) an e E w so that {e}Y = or(<m>-<e>-y)-. This can be
done since a < T y. Let L(<m>-<e>-y) = <n>QKi>7z. So {e}y = z by the choice of
e, and {i}z = y since a is a winning strategy for II. Thus m E F(y) iff n E G(z). But
effectively in x (and hence G(y)) we can produce from (e, i) a number k so that
{k}G y) = G(z). Then m e F(y) iff n E {k}G(Y), and we have effectively in G(y)
reduced our question about F(y) to one about G(y).
If Player I has a winning strategy in 9(F, G) then a similar argument shows
that g < f [1
The game 0(F, G) used in Lemma 1 derives from games invented by W. Wadge
(cf. [11]) and D. Martin (cf. [7]).
Forf: D -- D, letf'(d) = f(d)'.
LEMMA 2. Let , g E J andf < M g. Then' ? g.
PROOF. Let F and G represent f and g respectively, above x e 6'w. Let Fl(y
F(y)', and F2(y) = l - F(y)'. Thus both F1 and F2 represent f' above x. Now
suppose f ' 4M g. By inspecting the proof of Lemma 1, we see that Player II h
winning strategies L1 in 9(G, F1) and L2 in 9(G, F2). Further, if l, x <T y then
as in Lemma 1 we can effectively in y produce from any m E w numbers k, n so
that m E G(y) iff n E ({k}FYW)'. Thus G(y) is r.e. in F(y) for li, x < T y. Similarly,
G(y) is co-r.e. in F(y) for a2, X <T y. Thus if L1, L2, X < T y then G(y) < T F(y),
and so g <m f ?
LEMMA 3. There is no sequence <f,, n < w> of jump operators
Vn(f,+, < m fn).
PROOF. Suppose that <(fl n < o> were such a sequence. We use the f
theorem of [10]: let P c Ecu x wW) be arithmetic and suppose <xl n <
that Vn(xn+? = the unique y so that P(x", y)). Then for some n, x,1 +i x",
Let <Fnl n < w> and x E act be so that Fn represents, above x, for e
Suppose also that x is chosen so that Fn+1(y)' < T Fj(y) whenever x
can be done by Lemma 2. Let (r., s.) witness the uniformity of F. above x
let Li, be a winning strategy for II in g(F,+,, F.)
Fix y E Ecu so that x, <(r., s,) I n E co>, and <Li,,j n < w> are all recu
(Here <(r,, sj)I n < w> and <aL,, n < wo> are regarded as coded by elem
in some natural way.) From the proof of Lemma 1 it follows that there i
so that (F,+,(y), y) = {z(n)} (F.(y),y) for each n. Thus there is an arith
0wc x wc so that for each n, (F,+,(y), <n + l>-y) is the unique t e W

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
350 JOHN R. STEEL

P((F"(y), <n>-y), t). But by the


<n>-y). This contradicts the theorem cited above. LI
We say < is a prewellorder of X and z has <-rank a iff 3,3 E Or3h: X 13
(h is onto A Vx, Y E X(x < y iff h(x) < h(y)) A h(z) = a). The lemmas above
yield
THEOREM 1. The relation < m is a prewellorder of J. Iff e J has < m-rank a, then
f ' e Jand has < -rank a + 1.
By Theorem 1, the Turing jump operator is the <.-least f e J such th
a.e. d, f(d) is not recursive in d. This extends Lachlan's characterization. The same
proof shows that if we set f(d) = the degree of the complete fll(d) subset of co,
then f is the < m least g e J such that for a.e. d, g(d) is not J' in d. The next lemma
gives similar results for selfdual jump operators. If Vn(x, e co), we define <xj
n < cw> ewc by: <xl n < c>((ij)) = xj(j).
LEMMA 4. Suppose that Vn e wo(fn E J and F. represents f, above x", as witnessed
by (to, us)). Let G(y) = <Fn(y)I n < cl>. Then G represents a <m least upper bound
of ff. I n < c} above <((x, to, uJ) n < wo>.
PROOF. Clearly G represents some g e J above <(x", to, UJ) n < Cl>, and
Vn(f <?m g). If h < m g and h is represented by H above s, then I has a winning
strategy a in 0(G, H) by the proof of Lemma 1. Let (i, j) = v(0) be l's first
move according to a. By the definition of G and the proof of Lemma 1, when-
ever s, O, xi < T y we have H(y) < T Fi(y). Thus h < m fi. LI
By Lemma 4, the c-jump operator f(d) = dw is the <m least g e J such that for
a.e. d, g(d) is not arithmetic in d. Similarly, the sharp operator f(d) = d# is the
< m-least g e J such that for a.e. d, g(d) 0 L[d]. (To apply Lemma 4 in this case,
let

F,(x) = {Fq(pvo ... v.)-'L~n?I (p [x, 41 *. On}


where rs(n is the Gddel number of so. Then Vx(F,(x) e L[x]), and F. represents
some fn E J above 0. But <F,(x)I n < cl> T xA, so Lemma 4 implies that the
sharp operator is the < m-least upper bound of ffI n < c}.)
There is a second Martin conjecture concerning functions from D to D which
complements the first, namely: If d 4 f(d) a.e., then 3c(f(d) = c a.e.). Little is
known about this conjecture. Martin did prove
LEMMA (MARTIN). Let f: D -+ D and suppose Vc(c < f(d) < d for a.e. d). Then
d < f(d)for a.e. d.
PROOF. Consider the game: I plays x, II plays y. Player I wins iffy < T X, {X(0)}X
is total, and f(dg(x)) = dg({x(0)}x). Since II can have no winning strategy, we
have a winning strategy or for I. Set F(y) = {e}U (Y), where e = v(0) is what a tells
I to play first. Since Vc(c < f(d) a.e.), F is not constant on any recursively pointed
perfect tree (cf. [4]). The usual splitting argument then gives a recursively pointed
perfect tree T 2 T 0 so that y < T (F(y), a) for all branches y of T. Thus if dg(a) <
f(d) and d = dg(y) for some branch y of T, then d < f(d). Since dg(a) < f(d) a.e.,
d < f(d) a.e.
We shall use Theorem 1 to give an ordinal measure of strength for theories in
the language of second-order arithmetic. We classify such theories according to
the closure conditions they impose on their natural inner models. As before,

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
A CLASSIFICATION OF JUMP OPERATORS 351

"natural" means that the inner model must be definable in a way which relativises
to an arbitrary degree. We also demand a uniformity.
DEFINITION. Let M: w2 -k P(w2), and
(i) Vx, y(X T y M(x) = M(y)),
(ii) Vx(M(x) is closed under join, Turing jump, and Turing reducibility),
(iii) Vx, y(y e M(x) -> M(y) c M(x)),
(iv) there is a relation W(x, y, z) so that Vx(W,, = {(y, z)l W(x, y, z)} is a
wellorder of M(x)),
(v) for a < o.t. W, and e E Co let
z if 3y(y T X via e and
Ze = z is the ath element of Wy),
0 otherwise.

Then <Zel e < co> E M(x). Then we say M and the associated map X(d) =
{yI 3x e d(y e M(x))} are inner model operators.
If X, X are inner model operators we say X < X iff X(d) c 4(d) for a.e. d.
THEOREM 2. < prewellorders the inner model operators.
PROOF. We begin with prelinearity. Let X and X be inner model operators
associated to M and N. Suppose X(d) ! .X(d) a.e. Let a(d) be the least ordinal
a so that for some x e d, the ath element of W,, is not in N(x), where Wx wellor
M(x) as in (iv) above. Let F(x) be the join of x" and {(e, i) I 3y(y -TX via e A i e
the a(d)th element of WY)}. Then F(x) e M(x) by (ii) and (v), while F(x) 0 N(x) by
the choice of a(d). Clearly F represents a jump operator f. Now if A(d) 4 X(d)
a.e. we can similarly define G(x) e N(x) - M(x) for dg(x) large, with G repre-
senting a jump operator g. But either g < mf off <? g, and in either case we have
a contradiction.
Suppose now <,&,,I n < wo> is a descending chain of inner model operators. As
in the last paragraph we have jump operators fn so that fn(d) = dg(y) for some
y e Xn(d) - A'1,1(d). But then <f.I n < cw> is a descending chain of jump opera-
tors, a contradiction. LI
Condition (iii) in the definition above was not used in the proof of Theorem 2.
Its role is to make the ordering of theories we now define more natural.
For T a theory in the language of second-order arithmetic, let 1T be the least
inner model operator such that <c( U XT(d), e, +, > I= T for a.e. d, if such an
operator exists. Let T1 < T2 iff XTi and XT2 exist, and XT1 < XT2
By a result of Moschovakis (cf. [1]), XT exists for every true T. It is easy to see
that JI1-CA < 41+1-CA for all n 2 1, and full CA < 2O-det < 2O -det <
< J, -det (by work of H. Friedman and D. Martin). For the theories T just
mentioned, XT(d) = Lr(d) [dJn f2 for some r(d). Of course, Hlj-det is the
least theory T such that XT is not of this form. We conjecture that I1 -
det < 1+1 - det for all n 2 1, that PD Tr where PD is the schema of
projective determinacy and Tr is the set of all true sentences, and that Tr is -
maximal among all theories.
The measure of strength defined above differs from the proof-theoretic measures.
For example, one can show that the schema asserting that every projectively de-
finable wellorder has a countable field has strength greater than that of H' - det;

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
352 JOHN R. STEEL

we conjecture its strength is that of PD. The reason is simply that our definition
eliminates "unnatural" models, such as forcing models, from consideration. There
is some methodological interest in this. Suppose one wants to find a natural inner
model so large that it admits no protectively definable wellorder. Then one will
probably need to assume PD, since PD is probably true in the smallest such
model. (Assuming PD, Moschovakis' M,, is such a model; cf. [1].)

?2. An initial segment of J. We shall analyze the class of jump operators f such
that f(d) e L[d] a.e. The analysis relies on Jensen's fine structure theory (cf. [3]).
So that we can simply cite results of [3], we use Jensen's Ja hierarchy for L.
According to [3], whenever (An+i(Ja) -n(Ja)) n Aw # 0, there is an element
of this set of largest Turing degree. We call such an element, and its degree, a
Jn+i(Ja) master code. The master code degrees are wellordered by the order of
construction of their members, a Jn(Ja) set being constructed before a properly
JA+i(Jh) set. This order coincides with Turing reducibility. For a < xf, let
Oa be the ath master code degree in L. If x is a Jn(Ja) master code, then x' is a
Jn+i(Jh) master code, so that 0r+1 = (Oa)'. These facts and definitions relativise
to an arbitrary degree d; thus da is the ath d-master code degree in L[d] for
a < xf[d], and da+1 = (da)'.
It is widely agreed that 0a is properly regarded as the ath iterate of the Tur
jump. The work of [2] and [4] on degree theoretic definitions of 02 in terms of
(OPt j < A} supports this belief. Under reasonable assumptions, the "only if"
direction of Theorem 3 is a proof of this belief. For suppose c e L is an iterate of
the Turing jump. Assume that c is definable in a way that relativises to yield a
jump operator fC with fc(O) = c. Assume further that Va < fL Vd(fc(d) < da
ifffc(O) < 0a). Then by Theorem 3, c = Oa for some a < xf.
R. Solovay suggested the "only if" direction of Theorem 3.
THEOREM 3. Let f(d) e L[d]for a.e. d. Then f is a jump operator ifff(d) is a d-master
codefor a.e. d.
PROOF. We begin with the "if" direction. So suppose f(d) is a d-master code a.e.,
and define a(d) so that f(d) is a master code for J. (d) [d] a.e. By countable additivity
fix n e cl so that f(d) is a Jn+, master code for Ja(d)[d] a.e. By [3], for a.e. d there
is a Jn+l(Ja (d) [d]) map of co onto J. (d) [d]. Consider the following game, whic
gives a uniform definition of such a map: I plays x and II plays y. Player I wins if
Y < T x and x(O) = to is a ,n+1 formula of 4 free variables so that for som
< a(d) (where d = dg(x)) we have
(i) {(i, t)I Ja (d) [d] I= [i, t, 3, x]} is a map of cl onto J. (d) [d], and
(ii) if r < 3, then no map of cl) onto Ja(d)[d] is Z+PW(Ja(d)[d]) definable fro
the parameters r, x alone.
Notice that a : satisfying (i) (for some so) and (ii) exists for all x e d, for a.e. d.
Moreover, 3 is uniquely determined by d, so that we can write 3 = :(d).
Since p(d) exists for a.e. d, II can have no winning strategy. Fix a winning
strategy a for I. Let so = v(0) be what a tells I to play first. For a <TY and
d = dg(y) let

hY= (i, t) Ja(d)[d] I= (q[i, t, ,8(d), (y)]}.

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
A CLASSIFICATION OF JUMP OPERATORS 353

Let <qil i < cl> enumerate the 1in formulae of < 5 f

F(y) = {(i, j, k)I i, j, k E Cl) and Ja (d) [d] F (i[j, h

Clearly F(y) is Jn+(Ja (d) [d]) for a < T y. Thus to show that dg(F(y)) =
f(dg(y)) for or < T y, it is enough to show that for or < T y and d = dg(y), every
Zn+l(Jad)[d]) subset of co is r.e. in F(y). Let a = (j e (I I Ja(d)[d] I= O[j, t]}, where
0 is Tn+1 and t e Ja(d)[d]. Since h. is onto, t is Tn+1 definable from p(d), a, and
y. Thus there is a 1in formula (pi so that

a = {]e OI Ja(d)[d] 1= 3 viq],[j, p(d), a, y]}.


But then

a =f e col 3k((i, j, k) eF(y)),


so a is r.e. in F(y).
To show the uniformity of F above a, notice that there are fixed Tn+1 formulae
O and p so that whenever a < T X and d = dg(x), 0 and p define F(x) and co - F(x)
over Ja (d) [d] from p(d), a, and x alone. But given e we can effectively produce
Tn+1 formulae O' and p' so that if x T y via e and a < T X, then O' and p' define
F(x) and co - F(x) from :(d), a, y alone (just replace x in 0, p by its definition from
y). From O' and p' we get effectively an e' so that {e'}F(Y) = F(x) by the method of
the preceding paragraph. (The only nonuniform step there, the choice of a Tn+1
definition of t, does not arise.) The map t(e) = e' is in fact recursive, and (t, a)
witnesses the uniformity of F above a.
For the "only if" direction, suppose f E J and f(d) E L[d] a.e. Let g(d) be the
least d-master code such that g(d) 4 f(d). We claim that g(d) is a successor in the
ordering of d-master codes a.e. If so, let g(d) = h(d)' where h(d) is a d-master code
a.e. Then g, h E J by the "if" part of this theorem. Also h < m f < m g, and h' =M
g. By Theorem 1, h =-m f, and sof(d) is a d-master code a.e.
So suppose g(d) is a limit in the ordering of d-master codes a.e. We shall derive
a contradiction, and thereby complete the proof.
Let a(d) be such that g(d) is a master code for Ja(d)[d] a.e., and fix n < co so that
g(d) is a Jn+l(Ja (d) [d]) master code a.e. Since g(d) is a limit a.e., every n(Ja (d) [d])
subset of co is a member of Ja (d) [d] a.e. Notice that f(d) < g(d) a.e. by Theorem 1.
Fix an F: Euo -e uo which representsf above some real. We now consider two cases.
Case 1. For a.e. d there is a Zn(Ja(d[d]) subset of co which is not a member of
Ja (d) [d]J
By Jensen [3] and the proof of the "if" direction above, we can in this case
find G: wco a cto such that G represents g above some real and for a.e. d, G(z) is
In(a(d) [d) for all z E d. We now make use of the nonselfduality of G(z) as in the
proof of Lemma 2. Since f < m g, the argument of that lemma shows that for a.e.
d, F(z) is Jn(Ja(d)[d]) for all z E d. But then f(d) E Ja(d)[d] a.e., and so for a.e. d
there is a d-master code c e Ja (d) [d] so that c 4 f(d). This contradicts our definition
of g.
Case 2. For a.e. d, every Zn(Ja (d) [d]) subset of co is a member of Ja (d) [d].
Let p(d) be the T. projectum of Ja(d)[d], so that co < p(d) < a(d) a.e. Let
,(d) be the least ordinal : so that some map from a subset of p(d) onto Ja (d) [d] is

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
354 JOHN R. STEEL

Sn(a(d) [d]) definable from


the I, formulae of < 3 free

Ay = {(i, r)I i < w A


Thus Ay is Sn(Ja(d) [d]),
by our choice of 13(d).
Now for a.e. d, every member of g(d) is An+l(Ja(d) [d]) and thus Zl(KJp(d) [d], Ay>)
for all y E d. By playing a game like that in the "if" direction of this theorem we can
find G: awc -k 2 representing g, Ti formulae 0 and so, and a E ac so that for a.e.
d, for all y E d

G(y) = {n I <Jp(d[d], Ay> I= 0[n, a, y]}


and

w - G(y) = {nI <Jp(d)[d], Ay> I p[n, a, y]}.


For 1 < p(d) let

G(y)p = {nI <Jp[d], Ay f 12> n [n, or,

and

(c - G(y))p = {nI <Jp[d], Ay n 32> - qp[n, a, y]J},

and

H(y)p = {(e, n, i)l 3x(x =TY via e A


((n E G(x)p A i = 0) V (n E (C - G(x))p A i =)))}.

Notice that given e we can effectively find a recursive te: l) - l) so that for a.e.
d, if y E dand x -T y via e, then H(x)p = t -'(H(y)p) for all 1 < p(d).
Let <KdIV < r(d)> enumerate in increasing order those 1 < p(d) so that for some
(equivalently, all) y E d, H(y)p+l - H(y)p # 0. Now for 3 < p(d), <H(y)pl 1 < (>
E ha (d) [d] (since {(e, i, r) 1 3x(x T Y via e A (i, r) E Af n2)} is In(Ja (d) [d]) and so
a member of Ja (d) [d]). Thus each v < r(d) is countable in Ja (d) [d], and since every
real in Ja (d) [d] is recursive in f(d) by the minimality of g(d), we have r(d) < c4(d).
Let H(y) = UP<P(d)HMY)p for y E d. Then G(y) < T H(y), and H(y) is approxi-
mated by a wellordered sequence of sets H(y)13, all recursive in F(y). We use a game
like that of Moschovakis' Coding lemma [5] to show H(y) < T F(y) for dg(y) large.
Consider the following "Friedman game". Player I pays e, then i, then plays out
x E Ecu. Player II plays k, then 1, then plays out y E Ecu. If {e}x # y, then I loses;
otherwise, if {k}Y # x then II loses. If x T Y via (k, e), then setting d = dg(y),
Player II wins iff
(i) WF(x) ! H(x), or
(ii) WF(Y) C H(y) A 3v < r(d) (H(x)ed ! WF(x) A H(y)p C WF(y)).
Let r be a winning strategy in this game. Let d be any degree so large that a,
and the reals witnessing the uniformity of F are recursive in d, and all statements
shown above to hold a.e. hold for d. We show g(d) < f(d). Since g(d) 4 f(d) a.e.
by assumption, we have the desired contradiction.

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
A CLASSIFICATION OF JUMP OPERATORS 355

Case A. X is a winning strategy for I.


Fix y e d. Let 1 e co be given. Using the recursion theorem we can effectively in
y produce ki so that if II plays <kk>1<l>^y and I responds by zr with <e>-<i>-x
(= <e1>i>x1), then {k1}Y = x. But then x--T y via (k1, e), and by the
uniformity of F we can effectively in y find no so that WF(x) = y). Since z is
winning, WF(y) C H(xi) and
WF (y = H(y)pd = H(X) C WF(y)
for all 1. Thus

H(y) = U t(*',>el) (Wni )


CEW

So H(y) is r.e. in F(y). Since y" < T F(y), H(y) < T F(y), so G(y) < T F(y).
Case B. r is a winning strategy for II.
Fix x E d. Let < * be a linear order of oi recursive in F(x) with a wellordred
initial segment of order type ,F(x). Choose < * so that the set of < *-limit points
and the < *-immediate successor function are recursive in F(x). We define a partial
cv: it- co recursive in F(x) by effective transfinite induction on < * (i.e. by the
recursion theorem).
(a) If n is < *-least, set 0(n) = k, where WF(x) = 0.
(b) If n is a < *-limit point, set 0(n) = k, where WF(x) = U< W((x).
(c) If n is the < *-immediate successor of m, find (using the recursion theorem) an
e so that if I plays <e>^<K(m)>-x and II responds with <k>-<l>^y by x, then {e}x
- y. Find I' s0 that Wf~(x) = Wf(). Then set 0(n) = i, where

W F~ - W F((,x) u t-blexFWFx).

Then 0(n) is defined for all n in the wellordered initial segment of < *. Also, if n
has < *-rank v < CvF(F), then H(x)dcd W,(x)) c: H(x). Now fix e so that x-TX
via e. Then Vm3n3i (n is in the wellordered initial segment of < * and (e, m, i)
E WV,). By boundedness, we have n in the wellordered part of <* so that
Vm3i ((e, m, i) e W4,,x))). But then both G(x) and a) - G(x) are r.e. in F(x), so
G(x) < TF(x).
Theorem 3 has been proved. Z
We now compute the <t-ranks of some familiar jump operators. It is useful
here to consider a similar but simpler prewellorder. Forf, g: D -+ Ord, letf < 0 g iff
f(d) < g(d) a.e. The countable additivity of Martin's measure on D and the fact that
the ordinals are wellordered implies <0 is a wellorder. For f: D -+ Ord, let If I be
the ? 0-rank off. ForfE J, let If I be the < m-rank off.
Let J1(d) be the least ordinal not the order type of a wellorder of co which is
il(x) for some x E d. The author has shown that IAddln(d)I = 821+1 all n > I
(cf. [111). Using this it is reasonably easy to see that if fn(d) = the Al-jump of d,
then 1f2nl = 42n+1* In particular, If21 = c We now identify the jump operator
with <,-rank s.
Let C be the class of uniform indiscernibles; that is, let ca E C iff Vx e wcv (a is a
canonical indiscernible for L[x]). Certainly every uncountable cardinal is in C, and
in fact (by AD) Nn is the nth element of C for n < cv (cf. [5]).

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
356 JOHN R. STEEL

LEMMA 5. Letf: D -+ Ord. Then


PROOF. If a is x#-admissible, then a is a canonical indiscernible of L[x]. Thus
If I E C iff If I is x-admissible for all x E Act. Given x E ww, we consider the ultra-
power DL[x]/p, where p is the Martin measure on D. Los' theorem holds for this
ultrapower since L[x] is wellorderable. Let Z: DL[x]lv L[x] be the transitive
collapse. Then r([f]) = If I. Thus If I is x-admissible iff f(d) is x-admissible for
a.e. d. The lemma follows at once. E]
When If I is small one can improve the "only if" direction of Lemma 4. Specifi-
cally, let 7y(d) be the least ordinal a such that whenever t E Lajd] and La[d] V 0b[t],
then 3, < a(t E Lj[d] and Lj[d] 1=- 0[t]). That is, 7y(d) is the least first-order re-
flecting ordinal, relative to d. Then we have
LEMMA 6. Suppose f(d) < 7r(d) a.e., and If I E C. Then for a.e. d, f(d) is either d-
admissible or a limit of d-admissibles.
PROOF. Let f be as in the hypotheses, and let g(d) = sup{a < f(d)l a is d-
admissible}. Suppose for a contradiction that g(d) < f(d) < g(d)+ a.e., where
g(d)+ is the next d-admissible after g(d).
Consider the following game. I plays p, x and II plays y. Player I wins iff y < T X
and g(dg(x)) is the least a so that La[x] I= 4x]. Since g(d) < 7y(d) a.e., II can have
no winning strategy in this game. (Here we use the fact that 7y(d) is the least ordinal
reflecting first-order sentences whose only parameter is some x E d.) Let a be a
winning strategy for I, and let (p = a(0) be I's first move by a.
Consider now a second game. I plays 0, x and II plays y. Player I wins iffy < T X
and 0 is a 2l formula of 3 free variables so that, setting d = dg(x), {(a, 3)lI a, , <
g(d) and Lg(d) [x] :- 0[a, A, x]} is a wellorder of g(d) of order type f(d). Again, II
can have no winning strategy. (Here we use that if a < 7)(d), then the next d-admis-
sible after a is the supremum of the order types of wellorders of a which are
27i(La[d]) definable from some parameter x E d alone.) Let r be a winning strategy
for I, and let 0 = r(0).
By Lemma 4 and the fact that If I E C, f(d) is (v, r)-admissible for a.e. d. For a
contradiction, let (v, r) < d and g(d) < f(d); we claim that f(d) is not (v, r)-
admissible. To see this, let

We S iff 3x E co((a, r) <T X A Lg(d)[ I(X)] k= o[PI(X)]


A V3 < g(d) (L5[a(x)] [ (p[a(x)]) A W = {(a, P) I Lg(d) [v(X)] 1= 0[a, , TW(x)]})
Notice that if x satisfies the right-hand side, then dg(x) = dg(u(x)) = dg(r(x)).
Also g(dg(x)) = g(d) and so W arises from a play according to r of the second
game, so that W is a wellorder of g(d) of order type f(dg(x)). Let p be the least
(v, r) admissible after g(d). Since S is a class of wellorders of g(d) which is 2l
definable over Lg(d)[(c, r)], we have sup{O.T.(W) I We S} < ,. (Proof By
forcing construct y E ace so that W(o "Y) - B Then S can be coded as a 21(a, a, y)
set of wellorders of co. The usual boundedness theorem now applies.) But if x E d
and W= {(a, P) I Lg(d [(X)] k= 0[a, , v(X)]} then x, W satisfy the right-hand
side so that O.T. (W) = f(d) and W E S. Thusf(d) < p, as desired. L
R. Solovay found an f such that f(d) < 61(d) a. e., If I E C, and the conclusion of
Lemma 5 fails forf.
Let Cl(d) be the nth d-admissible ordinal. Let hj(d) be the nth iterate of th

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
A CLASSIFICATION OF JUMP OPERATORS 357

hyperjump operation starting at d; that is, h1(d) = the J1 jum


hl(hn(d)).
THEOREM 4. For n > 1
(a) IMd.wn(d)l = I n
(b) IhnI = Nn.
PROOF. (a) Since IMd.wO(d)l E C and Nn is the nth element of C, Nn < IMd.w)(d)
But if 2d.wO(d) < 0 f <0 IMd.w0n+1(d)I then by Lemma 5, If I 0 C. Thus IMd.wc(d)l
< Nn.
(b) For g: D -+ Ord so that g(d) < $f[d] a.e., we define H(g): D -+ D by:
H(g)(d) = dg (d). By Theorem 3, H(g) E J, and clearly f < 0 g iff H(f) < mH(g). By
Theorem 3, iff E Jandf(d) E L[d] a.e., thenf -m H(g) for some g. Thus IgI = IH(g)I
for all g E dom H. Now by [4], dwn(d) = hn(d). So H(2d.wC(d)) = hn, and by (a)
IhnI = Nn. Z
We owe much of the computation of <0-ranks above to a conversation with
Leo Harrington. Our proof that IMd.wO(d)l = Nn generalizes Martin's original
proof that I2d.l(d)l = X1.
We shall state without proof some further facts about < 0-ranks. As mentioned
above, IMd.21(d)I = 62n+1. On the other hand, IMd.62n+1(d)l is not a cardinal. In
fact, let zr1(d) = sup{O.T.(R)I R is a wellfounded relation on co which is zl(x) for
some x E d}. Then 2z2n(d) = 621(d) and iC2"+1(d) > 621n+1(d) for all n, d. One can
show that IMd.lr(d)l = &n+1 for all n. Since 42n+2 is the successor cardinal of 2n+i1
we conclude that IMd.62n+1(d)l is not a cardinal. In fact one can characterize
do2n+(d) Ias the largest x < 42n+2 so that DV/, = C is a cardinal.
Our definition of "jump operator" has been the most liberal under which we can
prove Theorem 1. We believe that the nice properties of familiar jump operators
can be deduced from this definition. Theorem 3 provides some evidence of this.
Fox example, the familiar jump operators are all order preserving. We believe that
anyf E J is order preserving a.e., by which we mean that 3bVc, d(b < c < d = f(c)
< f(d)). When f(d) E L[d] a.e., this follows from Theorem 3. For notice that any
g: D -* Ord is order preserving a.e.: if for a.e. c there is d ? c so that g(d) < g(c),
then we easily get a descending chain of ordinals. Thus for a.e. c, d > c :- g(d) ?
g(c), and so g is order preserving a.e. But now by Theorem 3, when f(d) E L[d] a.e.
there is a g: D -+ Ord so that f(d) = dg (d) a.e. Since g is order preserving a.e.,f is
order preserving a.e.

REFERENCES

[1] H. BECKER, Partially playful universes, Cabal Seminar, 1976-1977, Lecture Notes in Mathe-
matics, vol. 689, Springer-Verlag, Berlin and New York, 1978, pp. 55-91.
[2] H. HODES, Jumping through the transfinite: the master-code hierarchy of Turing degrees, this
JOURNAL, vol. 45 (1980), pp. 204-220.
[3] R. B. JENSEN, The fine structure of the constructible hierarchy, Annals of Mathematical Logic,
vol. 4 (1972), pp. 229-308.
[4] C. JOCKUSCH and S. SIMPSON, A degree-theoretic definition of the ramified analytic hierarc
Annals of Mathematical Logic, vol. 10 (1976), pp. 1-32.
[5] A. S. KECHRIS, AD and projective ordinals, Cabal Seminar, 1976-1977, Lecture Notes in
Mathematics, vol. 689, Springer-Verlag, Berlin and New York, 1978, pp. 91-133.

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms
358 JOHN R. STEEL

[6] A. H. LACHLAN, Uniform enu


[7] D. A. MARTIN, The axiom of d
Bulletin of the American Mathe
[8] G. E. SACKS, On a theorem of Lachlan and Martin, Proceedings of the American Mathematical
Society, vol. 18 (1967), pp. 140-141.
[10] J. STEEL, Descending sequences of degrees, this JOURNAL, vol. 40 (1975), pp. 59-61.
[11 ] R. VAN WESEP, Wadge degrees and descriptive set theory, Cabal Seminar, 1976-1977, Lecture
Notes in Mathematics, vol. 689, Springer-Verlag, Berlin and New York, 1978, pp. 151-171.

UNIVERSITY OF CALIFORNIA AT LOS ANGELES

LOS ANGELES, CALIFORNIA 90024

This content downloaded from 137.132.123.69 on Mon, 07 Sep 2020 03:07:11 UTC
All use subject to https://about.jstor.org/terms

You might also like