Book
Book
Book
Alain Louveau
iii
Contents
iv
Contents
v
Contents
vi
Contents
4.5.2 Inductions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.5.3 Moschovakis’ theorem . . . . . . . . . . . . . . . . . . . . . . . . 99
4.5.4 Σ11 derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.5.5 Complexity of the stages of an induction . . . . . . . . . . . . . . 101
4.6 Reflection theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.6.1 First reflection theorem . . . . . . . . . . . . . . . . . . . . . . . 103
4.6.2 Second reflection theorem . . . . . . . . . . . . . . . . . . . . . . 104
4.6.3 Burgess’ reflection theorem . . . . . . . . . . . . . . . . . . . . . 106
4.6.4 Approximating Σ11 equivalence relations . . . . . . . . . . . . . . 106
5 Uniformizations 107
5.1 ∆11 points and uniformization by ∆11 -recursive functions . . . . . . . . . 107
5.1.1 Coding of ∆11 and B-sets . . . . . . . . . . . . . . . . . . . . . . . 107
5.1.2 Γ-recursive points . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.1.3 Coding of ∆11 -recursive points . . . . . . . . . . . . . . . . . . . . 110
5.1.4 Co-Suslin and closure under D∆ . . . . . . . . . . . . . . . . . . . 110
5.1.5 ∆11 -uniformization theorem . . . . . . . . . . . . . . . . . . . . . 111
5.1.6 Characterization of ∆11 -recursive functions . . . . . . . . . . . . . 112
5.1.7 One-to-one ∆11 -recursive functions in co-Suslin spaces . . . . . . . 112
5.1.8 Counterexamples in co-Suslin spaces . . . . . . . . . . . . . . . . 113
5.1.9 Complexity of α P ∆11 , α P ∆11 pβq . . . . . . . . . . . . . . . . . . 113
5.1.10 Antibasis, anti-uniformization . . . . . . . . . . . . . . . . . . . . 114
5.1.11 Spector-Gandy Theorem for Π11 . . . . . . . . . . . . . . . . . . . 115
5.1.12 One-to-one recursive images of Π01 sets in ω ω . . . . . . . . . . . 117
5.2 Coding Borel sets. HYP sets and the Suslin-Kleene Theorem . . . . . . 118
5.2.1 Coding of Borel sets . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.2.2 HYP sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2.3 The Suslin-Kleene theorem . . . . . . . . . . . . . . . . . . . . . 119
5.2.4 A “constructive” proof of Suslin’s Theorem 4.4.6 . . . . . . . . . . 120
5.2.5 The case of ω ω . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.2.6 The general case . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.2.7 B “ HYP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.2.8 The hyperarithmetical hierarchy . . . . . . . . . . . . . . . . . . 127
5.3 Complements on ω1CK and the ω1CK´α . . . . . . . . . . . . . . . . . . . 129
5.3.1 Complexity of WOωCK . . . . . . . . . . . . . . . . . . . . . . . . 129
1
5.3.2 Kleene’s O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.3.3 ω1CK as the domain of wellfoundedness of a recursive ordering . . 131
5.3.4 Complexity of ω1CK´α ď ω1CK´β . . . . . . . . . . . . . . . . . . . 132
5.3.5 Spector’s criterion . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.3.6 Basis result for Σ11 sets . . . . . . . . . . . . . . . . . . . . . . . . 133
5.4 Uniformization of Σ11 sets and Π11 sets . . . . . . . . . . . . . . . . . . . 135
5.4.1 BorpΣ11 q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.4.2 C-sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.4.3 The Jankov-Von Neumann theorem . . . . . . . . . . . . . . . . . 137
vii
Contents
viii
Contents
ix
Contents
Index 249
x
Liste des Corrections
Rectification of the proof? – R . . . . . . . . . . . . . . . . . . . . . . . . . . 21
fixing of recursive arguments – YP . . . . . . . . . . . . . . . . . . . . . . . . 36
strange to use this here and define it later on – R . . . . . . . . . . . . . . . . 38
a little bit too quick – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Here isn’t Hausdorff, or at least T1 , necessary? (See Kechris for instance) – R 40
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
by fixing of recursive argument – YP . . . . . . . . . . . . . . . . . . . . . . . 51
this seems too vague to me. – YP . . . . . . . . . . . . . . . . . . . . . . . . . 51
agreed – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Σ01 Ď Σ02 – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
This is because Γpαq is closed under fixing of recursive-in-α arguments in ω ω .
– YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
editing – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Notation to be fixed? – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
just let Apn, pq Ø DkBpn, p, kq with B recursive in ω 3 and let ϕpnq “
pminl Bpn, plq0 , plq1 qq0 – YP . . . . . . . . . . . . . . . . . . . . . . . . . 65
is it in the same sense as for a lightface class? – YP . . . . . . . . . . . . . . 66
cf Subsection 2.4.6 and proof of Theorem 2.4.4 – YP . . . . . . . . . . . . . . 67
recursively embedded? – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
rec.emb.? – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
def? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
def? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
L:ϕpα, xq ă ϕpα, yq – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
added – R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Pb with index, end of first line. – R . . . . . . . . . . . . . . . . . . . . . . . 102
Corrected until this point. – R . . . . . . . . . . . . . . . . . . . . . . . . . . 106
added – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
veut-il dire α recursive and α P D1 ? – YP . . . . . . . . . . . . . . . . . . . . 107
maladroit? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
il faut choisir entre Suslin-recursive et Suslin recursive. – YP . . . . . . . . . 110
Spector est souligné dans le manuscrit... – YP . . . . . . . . . . . . . . . . . . 111
ajoutÈ – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
L: is – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
ajoutÈ – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
ajoutÈ – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
we did not explain what antibasis means – YP . . . . . . . . . . . . . . . . . 114
ajoutÈ – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
xi
Liste des Corrections
xii
Liste des Corrections
ajout – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
L: ϕn : 2ω Ñ Cp2ω q – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
ne manque-t-il pas qqch ici? – YP . . . . . . . . . . . . . . . . . . . . . . . . 143
définition?? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
L: clopen – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
I am missing something here... and if Wα˚ “ H – YP . . . . . . . . . . . . . . 144
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
α ď β Ø α´1 p1q Ď β ´1 p1q – YP . . . . . . . . . . . . . . . . . . . . . . . . . . 144
oversight – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
encore? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
L: @η – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
ajout – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
where ¨ denotes the usual product in R – YP . . . . . . . . . . . . . . . . . . 148
L: consists to – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
L:Granting this – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
ajout – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
since txu is ∆11 pxq and included in Φ8 pAq there is ξ ă ω1CK´x with x P Φξ pAq.
– YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
take A Ď X ˆ ω to be H, then for n0 P ω we have Φ8 pHqn0 “ P is Π11 .
Now ψpx, nq “ inftξ : px, nq P Φξ pHqu is a Π11 -rank by Theorem 5.5.11,
hence so is ϕpxq “ ψpx, n0 q. – YP . . . . . . . . . . . . . . . . . . . . . . 151
pas encore dÈfini – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
regular tree again... pruned tree? – YP . . . . . . . . . . . . . . . . . . . . . . 153
regular again – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
regular! – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
un peu cavalier... – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
regular – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
regular – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
regular – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
really? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
regular? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
regular? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
sensé? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
je pense... – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
L: relative – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
regular? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
missing reference – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
bien jouÈ! – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
reference? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
regular? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
regular? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
regular – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
L: i ă l, i ă 2m – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
L: n – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
xiii
Liste des Corrections
regular – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
L: πn8 pAnn q – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
citations added – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Une autre Arlésienne – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
def? – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Reference missing – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Reference missing – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
rendu explicite l’utilisation de l’hypothèse – K . . . . . . . . . . . . . . . . . . 181
citation added – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Not sure about the reference – K . . . . . . . . . . . . . . . . . . . . . . . . . 186
I think this should be stated as a result on metrizable separable spaces. – YP 188
ref to Kuratowski’s book – YP . . . . . . . . . . . . . . . . . . . . . . . . . . 189
?? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
In fact, Theorem 5.1.8 only gives ∆11 , and one needs Suslin’s Theorem 4.4.6
to get Borel. – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
really? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
really? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
the sequence Wpαq1 , η 1 ă η, is not increasing in general, no? is it a problem?
ϕpη 1 q
– YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
It is not increasing in general indeed, but it’s not a problem, you take the
intersections. – K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
really? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
in fact under @ă – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
really? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
separable metrizable? – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
réf – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
the proof was in fact left to the reader – YP . . . . . . . . . . . . . . . . . . . 198
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
I can’t see that right now – YP . . . . . . . . . . . . . . . . . . . . . . . . . . 202
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
remark that the projection π : F Ñ X is a bijection and F is Polish! – YP . . 203
ref missing – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
added – YP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
weird, no? ∆11 -measurable, or simply bianalytic? – YP . . . . . . . . . . . . . 204
why not Suslin space? Alain wrote Suslin space and then rectified to Suslin
recursive space, but I am not sure why... – YP . . . . . . . . . . . . . . 204
again, ∆11 -measurable, or simply bianalytic function – YP . . . . . . . . . . . 206
Alain says ∆11 -uniformization, but? – YP . . . . . . . . . . . . . . . . . . . . 208
Polish space? I don’t understand how Alain chooses to state a theorem for
recursive spaces or just spaces... – YP . . . . . . . . . . . . . . . . . . . 210
xiv
Liste des Corrections
xv
1 Recursivity on ω
1.1 Recursive functions and sets
1.1.1 Definition of recursive functions
Throughout the book, ω “ t0, 1, 2, . . .u denotes the set of natural numbers, denoted by
letters i, j, k, l, n, m, p, q . . .. As usually in Set Theory, a natural number is the set of its
predecessors, i.e. n “ t0, 1, . . . , n ´ 1u; and in particular 0 “ H is the empty set. For
k ě 1, the Cartesian product ω k is identified with the set of functions from k to ω, or
sequences of natural numbers of length k, and we write its elements x̄ “ px0 , . . . , xk´1 q.
In this chapter we consider functions with domain one of the ω k ’s, k ě 1, and values
in ω.
Definition 1.1.1. The set Rec of recursive functions is defined as the least set of
functions (from some ω k to ω) satisfying the following clauses
(2) For n ă k, the projection function Pkn px̄q “ xn on the nth factor is recursive;
1
1 Recursivity on ω
In this definition, clauses (1), (2) and (3) correspond to choosing some very simple
functions at the beginning (we could have put more functions, and the set chosen
here is minimal, which will be useful in later work). And clauses (4), (5) and (6)
correspond to closing under some operations on functions. The intuitive content is that
of computability: these three operations allow to compute the value of the new function
from (a finite set of) values of the old ones. So all recursive functions are computable.
The converse, that computable functions are recursive, is known as Church’s thesis.
There is now a lot of evidence for this metamathematical statement: Till now all
mathematical definition which have been proposed to capture the notion of computable
function have led to the same class Rec.
In this book, our point of view will be very far from computability. For us, the
preceding definition simply isolates a nice countable set of functions, with remarkable
closure properties, rich enough to account for, or encode, most mathematical construc-
tions.
1.1.2 Examples
In the sequel, we will need to know that many functions are recursive. Some of them
are listed below.
Proposition 1.1.2.
(i) If x̄ P ω k , ȳ P ω l and f : ω k Ñ ω is recursive, gpx̄, ȳq “ f px̄q is recursive.
(ii) If π is a permutation of k and f : ω k Ñ ω is recursive,
gpx0 , . . . , xk´1 q “ f pxπp0q , . . . , xπpk´1q q
is recursive.
(iii) Addition : ω 2 Ñ ω, Multiplication : ω 2 Ñ ω and Exponentiation : ω 2 Ñ ω (with
say 00 “ 1) are recursive.
(iv) The predecessor function
#
0 if n “ 0
n´1“
n´1 if n ą 0
the characteristic function
#
1 if n “ 0
1t0u pnq “
0 if n ą 0
the subtraction function : ω 2 Ñ ω
#
0 if n ă k
n´k “
n´k if n ě k
2
1.1 Recursive functions and sets
(iv)
#
0´1“0
n`1´1“n
#
n´0“n
n ´ pm ` 1q “ pn ´ mq ´ 1
#
1t0u p0q “ 1
1t0u pn ` 1q “ 0
and
1ď pm, nq “ 1t0u pm ´ nq
1ă pm, nq “ 1 ´ 1ď pn, mq
1“ pm, nq “ 1ď pm, nq ¨ 1ď pn, mq. %
px, nq P Dď A ÐÑ Dm ď n px, mq P A
ď
px, nq P @ A ÐÑ @m ď n px, mq P A
and
x P D0 A ÐÑ Dm px, mq P A
0
xP@ A ÐÑ @m px, mq P A.
3
1 Recursivity on ω
Γ ^ Γ1 “ tA ^ B : A P Γ, B P Γ1 u
Γ “ t A : A P Γu
D Γ “ tD0 A : A P Γu, etc.
0
Proposition 1.1.3.
(i) For all k ě 1, ∆01 contains H, ω k , the finite subsets of ω k , and in ω 2 the sets ď,
ă, “.
is recursive too.
Proof. We did the case of ď, ă, and “ in Proposition 1.1.2. For H and ω k , the
characteristic functions are constants. We will prove it for finite sets ate the end.
For (ii), note that 1A^B “ 1A ¨ 1B , 1A_B “ 1A ` 1B ´ 1A^B and 1 A “ 1 ´ 1A .
We prove the closure under Dď . Let A Ď ω k ˆ ω be recursive. Note that
px̄, nq P Dď A ÐÑ Dm ď n Apx̄, mq
ÐÑ Dm ď n p1 ´ 1A qpx̄, mq “ 0.
` ˘
Let then ϕpx̄, nq “ least p 1ď pp, nq ¨ p1 ´ 1A qpx̄, pq “ 0 . As for each x̄, n there is such
a p (for instance p “ n ` 1 works!), ϕ is recursive by minimization, and one gets
so Dď A is recursive too.
4
1.1 Recursive functions and sets
which is recursive by closure under ^, as “ is recursive, and we can fix integers argu-
ments — a particular case of (iii). %
so f is recursive, by minimization. %
5
1 Recursivity on ω
Proof. There are many of them. Pick a simple one between ω 2 and ω, e.g.
ϕ2 pn, pq “ 2n p2p ` 1q ´ 1
with inverses
so that
k
ψk`1 ppq “ ψ21 ppq
i
and ψk`1 ppq “ ψki pψ20 ppqq.
From now on, we will fix once and for all such bijections between ω k and ω, and
to simplify notations, we will omit indexes and denote them ambiguously with brackets
(x y : ω k Ñ ω) and parentheses, so that pxn0 , . . . , nk´1 yqi “ ni , and n “ xpnq0 , . . . , pnqk´1 y.
So in expressions like xn0 , xn1 , n2 y , n3 y the brackets have different meanings, but this
should not be a problem.
Note that using these bijections, all recursive notions of sets and functions can be
transferred to the set ω, by noting that P Ď ω k is recursive iff tn : P ppnq0 , . . . , pnqk´1 qu
is recursive, and that f : ω k Ñ ω is recursive iff g : ω Ñ ω defined by gpnq “
f ppnq0 , . . . , pnqk´1 q is recursive.
6
1.1 Recursive functions and sets
1.1.7 Codings
If one wants to extend recursivity notions to other countable sets, one can use the
method of coding. Is D is a countable set, a coding of D (by ω) is a function ρ : ω D
surjective onto D. If d P D, any n P ω with ρpnq “ d is called a code for d. If now A
is a subset of Dk , its set of codes is
and A is recursive in the codes if this set is recursive in ω k — and similarly with
functions, by considering their graphs.
Of course when dealing with natural structures, we want the coding ρ to be chosen
carefully, so that the relations and functions defined on D become recursive in the
codes.
Two countable structures play an important role in this book: the field Q of rational
numbers, and the structure ω ăω of finite sequences of numbers. So we fix once and for
all nice codings for them.
1.1.8 Coding of Q
We define a map q : ω Ñ Q by
pnq1
qpnq “ qn “ p´1qpnq0 .
pnq2 ` 1
Clearly q is onto Q, so is a coding of Q; and by writing down explicitly the definitions,
the reader can easily check the following
Proposition 1.1.7. The relations of equality, the natural ordering, the singletons, the
closed and the open intervals in Q are all recursive in the codes. Similarly addition,
multiplication, subtraction, absolute value and division (we say pq “ 0 if q “ 0) are all
recursive in the codes.
Remark 1.1.8. We may also need codings of subsets #of Q, e.g. of Q X r0, 1s. As
0 if qn R r0, 1s
tn : qn P r0, 1su is recursive, we can e.g. take q1n “ so that the
qn otherwise
restrictions of the preceding relations and functions will again be recursive in the codes.
1.1.9 Coding of ω ăω
For any set X, X ăω denotes, the set of finite sequences from X, i.e. the set of functions
from some natural number k into X. The domain k is called the length lhpsq of the se-
quence s, and the value at i ă k is denoted spiq or si . So we denote s “ psp0q, . . . , spk ´
1qq or ps0 , . . . , sk´1 q, with the convention that for k “ 0, this notation represents the
empty sequence (denoted by H). For i ď lhpsq, si “ psp0q, . . . , spi ´ 1qq and we denote
by sa x “ psp0q, . . . , spk ´ 1q, xq and sa t “ psp0q, . . . , spk ´ 1q, tp0q, . . . , tpl ´ 1qq the
concatenation operations. And we say that s Ď t, or t extends s, if s “ tlhpsq .
7
1 Recursivity on ω
i.e. by considering first n as a pair ppnq0 , pnq1 q and then pnq1 as an pnq0 -tuple, using
the appropriate brackets. So any n “ x0, ky codes the empty sequence, x1, ky codes
pkq, and for p ě 2 xp, ky codes ppkq0 , . . . , pkqp´1 q. We can also define a bracket function
: ω ăω Ñ ω by xHy “ x0, 0y, xpkqy “ x1, ky, and xpk0 , . . . , kp´1 qy “ xp, xk0 , . . . , kp´1 yy
which satisfies sxty “ t for all t P ω ăω . So s is a coding of ω ăω .
Proposition 1.1.9. The following relations are recursive in the codes
(i) s Ď t, s “ t;
(ii) lhpsq “ k, i ă lhpsq ^ spiq “ k, i ď lhpsq ^ t “ si ;
(iii) t “ sa m, u “ sa t, s P 2ăω .
Proof. is very easy, and left to the reader. %
Note that the remark at the end of Subsection 1.1.8 can be made here too. For
example, if we want a coding of 2ăω , we can define s2 : ω Ñ 2ăω by
#
2 spnq if spnq P 2ăω
s pnq “ .
H otherwise
Then the same relations as in the proposition, when restricted to 2ăω are recursive in
the codes for this coding.
Then f is recursive.
(ii) Let f0 , g0 : ω k Ñ ω and h0 , h1 : ω k`3 Ñ ω be recursive, and define by simultane-
ous recursion
8
1.2 Σ01 sets
and
Proof.
and define gpn, x̄q “ xf p0, x̄q, . . . , f pn ´ 1, x̄qy. Then g is defined by primitive
recursion by
(ii) Define ϕpn, x̄q “ xf pn, x̄q, gpn, x̄qy. Then ϕ is defined by
#
ϕp0, x̄q “ xf0 px̄q, g0 px̄qy
ϕpn ` 1, x̄q “ xh0 pn, pϕpn, x̄qq0 , pϕpn, x̄qq1 , x̄q, h1 pn, pϕpn, x̄qq0 , pϕpn, x̄qq1 , x̄qy
9
1 Recursivity on ω
(So compared with the closure properties of ∆01 , note that D0 has been added, but
disappeared).
x̄ P A ÐÑ Dn Bpx̄, nq
and x̄ R A ÐÑ Dn Cpx̄, nq.
The set B YC is recursive, and moreover for any x̄ P ω k there is an n with px̄, nq P
B Y C so ϕpx̄q “ least n ppx̄, nq P B Y Cq is recursive, and x̄ P A Ø Bpx̄, ϕpx̄qq so
A is ∆01 .
10
1.3 Kleene’s Enumeration Theorem
This proposition explains why the Σ01 subsets of ω are also called the recursively
enumerable sets. Note that if A is infinite, the enumeration can be chosen one-to-one,
but not monotone unless A is ∆01 .
nPA ÐÑ peA , nq P G.
(So the Σ01 subsets of ω are exactly the sections of G. Such a set G is called universal
for Σ01 subsets of ω).
11
1 Recursivity on ω
Then
xPA ÐÑ Dn px, nq P B
ÐÑ Dn peA , xx, n, 1yq P G‹
ÐÑ peA , xq P G
as desired.
Definition 1.3.2. The set C of codes is the least set of numbers satisfying
(1) If
(1a) lh sn “ 3 and sn p0q “ 1, sn p1q ě 1
then n P C. [n “ x1, k, py codes Ckp : ω k Ñ ω]
(2) If
(2a) lh sn “ 3 and sn p0q “ 2, sn p2q ă sn p1q
then n P C. [n “ x2, k, py codes Ppk : ω k Ñ ω]
(3) If
12
1.3 Kleene’s Enumeration Theorem
(4) If both
(4a) lh sn ą 3, sn p0q “ 4, and @p p1 ď p ă lh sn Ñ lh ssn ppq ě 2q, ssn p2q p1q “
lh sn ´ 3 and @p p2 ă p ă lh sn Ñ ssn ppq p1q “ sn p1qq
(4b) @p p2 ď p ă lh sn Ñ sn ppq P Cq
then n P C. [we wrote formally that if sn “ p4, k, p, q0 , . . . , ql´1 q and p already
codes f : ω l Ñ ω, and q0 , . . . , ql´1 code g0 , . . . , gl´1 : ω k Ñ ω, then n codes a
function : ω k Ñ ω — we of course want it to code f pg0 , . . . , gl´1 q]
(5) If both
(5a) lh sn “ 4 and sn p0q “ 5 and lh sn p2q ě 2 and lh sn p3q ě 2 and sn p1q “
ssn p2q p1q ` 1 and ssn p3q p1q “ ssn p2q p1q ` 2
`
(5b) sn p3q P C and ˘ either sn p2q P C or lh sn p2q “ 3 and ssn p2q p0q “ 0 and
ssn p2q p1q “ 0
then n P C. [again we wrote formally that if sn “ p5, k ` 1, p, qq and q already
encoding a function g : ω k`2 Ñ ω and p either of form x0, 0, n0 y or already
encoding a function f0 : ω k Ñ ω, then codes a function f : ω k`1 Ñ ω — we have
in mind the function obtained by primitive recursion from pn0 , gq or pf0 , gq]
(6) If both
(6a) lh sn “ 3 and sn p0q “ 6 and lh sn p2q ě 2 and ssn p2q p1q “ sn p1q ` 1 ě 2
(6b) sn p2q P C
then n P C. [if sn “ p6, k, pq and p already codes g : ω k`1 Ñ ω, then n codes
f : ω k Ñ ω. Here we have in mind the function obtained by minimization from
g, if it does exist.]
Proof. Say that a number n is well-formed if it satisfies one of the clauses (1a), (2a),
(3a), (4a), (5a) or (6a) of Definition 1.3.2. Clearly the set of all well formed numbers
is recursive. If now n is well formed, define its set of predecessors by
$
’
’ H if sn p0q “ 1, 2 or 3,
’
’
&tsn ppq : 2 ď p ď lh sn u if sn p0q “ 4,
’
’
P pnq “ tsn p3qu if sn p0q “ 5 and ssn p2q p0q “ 0,
’
tsn p2q, sn p3qu if sn p0q “ 5 and ssn p2q p0q ě 1,
’
’
’
’
’
if sn p0q “ 6.
%
tsn p2qu
13
1 Recursivity on ω
14
1.3 Kleene’s Enumeration Theorem
The preceding definition is a bit informal; one should define a set of pairs pn, G˚n q
as the least with the preceding properties, and check that indeed for each n there is
exactly one G˚n , given by the above.
Lemma 1.3.4.
15
1 Recursivity on ω
Proof. Fact (ii) is easy: Consider the family of recursive functions which satisfy the
conclusion of (ii). It clearly contains the constants, the projections and the successor
function. And it is easy to verify, by the very form of Subsection 1.3.5, that it is closed
under composition, primitive recursion and minimization. So it contains all recursive
functions.
The proof of fact (i) is quite analogous to the proof given in Subsection 1.3.4 that
C is Σ01 . So we only sketch it.
Suppose we are given a pair pn, mq with n P C, and some number q. We define a
notion of predecessor P pn, m, qq as follows:
– If sn p0q “ 1, 2 or 3, P pn, m, qq “ H.
– If sn “ p6, k, pq, P pn, m, qq is ω 2 if one of the pqqi ’s, for i ă pmqk , is 0, and
otherwise consists of the pairs
From this definition, one easily checks that n P C^pi, jq P P pn, m, qq is Σ01 in ω 5 . Define
then a computation c as a number coding a sequence sc “ pxn0 , m0 y , . . . , xnk , mk yq of
length ě 1, with the property that for j ď k nj P C and there is a q such that for all ele-
ments pr, sq P P pnj , mj , qq, the number xr, sy is among xn0 , m0 y , xn1 , m1 y , . . . , xnj´1 , mj´1 y.
The set of computations is again clearly Σ01 , so to finish the proof, it is enough to prove
that
16
1.4 Hierarchy and recursion theorems
For direction Ñ, one consider the set of pairs pn, G˚n q for which n P C and the right-
hand side is satisfied, and easily show that it satisfies (1)–(6) of Subsection 1.3.5, by
concatenation of the computations.
For direction Ð, one argues by induction on the length of the sequence sc , and
by examination of conditions (1)–(6) of Subsection 1.3.5, one easily checks that the
hypothesis on the predecessor set is exactly what is needed to get them. Details are
left to the reader. %
Proof. One uses a method known as “Cantor’s diagonal method”. Let G be the Σ01
subset of ω 2 universal for all Σ01 subsets of ω, and define
∆ “ tn : pn, nq P Gu.
nP ∆ ÐÑ Gpe, nq,
Theorem 1.4.2. For each k ě 1 there exists a Σ01 set W k Ď ω k`1 such that if A Ď ω k`1
is Σ01 , there exists a recursive function fA : ω Ñ ω such that for all n and x̄ P ω k
17
1 Recursivity on ω
Proof. Note that indeed a set W k with this property is universal for Σ01 subsets of ω k ,
for if B Ď ω k`1 is Σ01 , consider Apn, x̄q Ø Bpx̄q, and apply the property to A.
To get W k , we start with the universal set G, and define
This W k is clearly Σ01 . We check that it works. Suppose A is Σ01 in ω k`1 , and define
as desired. %
From now on, we fix once and for all the codings of Σ01 subsets of the ω k ’s as given
by the good universal sets W k , and denote Wnk (Wn for k “ 1) the nth Σ01 set in this
enumeration.
1.4.3 Uniformities
Theorem 1.4.2, despite its simplicity, is the key to many important results, for it allows
to infer uniform results from a priori non uniform ones. As an example, let us prove
Wnk X Wm
k
“ Wfk^ pn,mq ,
Wnk Y Wm
k
“ Wfk_ pn,mq .
18
1.4 Hierarchy and recursion theorems
Proof. Define
App, ȳq ÐÑ W m`n pppq0 , ppq1 , . . . , ppqm , ȳq.
A is Σ01 in ω m`1 , so for some f : ω Ñ ω
Proof. This is an elaboration on the diagonal method, which looks rather magic.
Let S1k : ω 2 Ñ ω be the S ´ 1 ´ k function of Theorem 1.4.4, and define for x̄ P ω k
19
1 Recursivity on ω
Proof. is magic again. We consider the S-2-k function S2k , which satisfies W k`2 pm, n, p, x̄q Ø
W k pS2k pm, n, pq, x̄q, and define B Ď ω k`2 by
Define ϕpnq “ S2k peB , n, eB q. This is recursive, and we check that it works: If n is a
code for A Ď ω k`1 , one has
20
1.4 Hierarchy and recursion theorems
(iii) Φ8 pHq is Σ01 in ω, and ΦpΦ8 pHqq “ Φ8 pHq, so that Φ admits a fixed-point;
and moreover Φ8 pHq is the least fixed-point of Φ.
Proof. (i) Let A and B be Σ01 , with A Ď B. Let k P ΦpAq. We want to show that
k P ΦpBq.
Define a Σ01 set C in ω 2 by
pp, qq P C ÐÑ q P A _ pq P B ^ k P ΦpWp qq
Set pm, pq P C Ø p P A and (if q is least with p P Apqq , m R B pqq ). One has
´ ¯
pm, pq P C ÐÑ Dq p P Apqq ^ @r ă q pp R Aprq _ m R B pqq q
but this way for all m we have A Ď C pmq , indeed for all p P A it is enough to
choose q minimal such that p P Apqq . This will ruin the end of the proof.
Instead if one goes simply for
21
1 Recursivity on ω
Set ´ ¯
pm, pq P C ÐÑ Dq p P Apqq ^ m R B pqq
p P Ce ÐÑ Dq pp P Apqq q, i.e. Ce “ A.
(iii) Let An “ Φn pHq. AsŤH “ Φ0 pHq Ď Φ1 pHq, one gets by induction using (i) that
An Ď An`1 Ď A8 “ n An .
We first show that the relation k P An is Σ01 . To see this, note that as k P ΦpWn q
is Σ01 in ω 2 , we can find by Theorem 1.4.2 a recursive function ϕ : ω Ñ ω such
that ΦpWn q “ Wϕpnq . Let n0 P ω be such that Wn0 “ H, and define f : ω Ñ ω
by
f p0q “n0
f pn ` 1q “ϕpf pnqq.
1.5 Relativization
In this section we introduce the easy but important notion of relativization of the re-
cursive notions we introduced in the previous sections, in order to include any function
(or set of functions) we want to study. As we said in Section 1.1 our goal is not Re-
cursion Theory, but Descriptive Set Theory, and we want a machinery which enables
us to study all functions and all subsets of the ω k ’s (and much more, of course).
22
1.5 Relativization
and
Σ01 pΦq “ D0 ∆01 pΦq,
all results proved in Sections 1.1 and 1.2 for Rec, ∆01 and Σ01 are valid for RecpΦq,
∆01 pΦq and Σ01 pΦq for any family Φ. (Again for Φ “ tαu, one usually writes ∆01 pαq and
Σ01 pαq instead of ∆01 ptαuq and Σ01 ptαuq).
Theorem 1.5.1 (Kleene). Let α : ω Ñ ω be given. Then there exists a Σ01 pαq set
G Ď ω 2 which is universal for all Σ01 pαq subsets of ω.
It follows from this result that all results proved in Sections 1.3 and 1.4 for Rec,
∆01 and Σ01 are true for Recpαq, ∆01 pαq and Σ01 pαq for all α — we will say that they
relativize to α.
The proof of Theorem 1.5.1 is a trivial modification of the proof presented in Sec-
tion 1.3: One just adds a seventh clause which correspond to the function α, and is
treated the way we treated the function S.
23
1 Recursivity on ω
1.5.4 A counterexample
We now give an example of a countable family Φ for which the enumeration theorem
fails for Σ01 pΦq.
Define a sequence pfn q of functions : ω 2 Ñ ω, inductively, by
24
2 Metrizable separable and recursive
spaces
In this chapter we introduce the spaces which will serve as frame for developing De-
scriptive Set Theory, both from the topological and recursivity points of view.
(1) dpx, yq “ 0 ÐÑ x “ y;
The structure pX, dq is called a metric space, and is canonically given a topology,
namely the one generated by the open balls Bpx, rq “ ty P X : dpx, yq ă ru, where x
varies over X and r over R` .
A topological space is called metrizable if some distance on X define its topology.
Such a distance is said to be compatible with the topological space X.
Note that if d is a distance on X, so is d1 px, yq “ inftdpx, yq, 1u, so that a metrizable
space always has a compatible bounded by 1 distance.
2.1.2 Separability
A topological space X is separable if it admits a countable dense subset, and is second-
countable if its topology has a countable basis. Although these notions are different in
general, they coincide for metrizable spaces: Clearly any topological second-countable
space is separable (by picking a point of X in each set of the basis). Conversely if X
is metrizable, d is a compatible distance and D is a countable dense set, the countable
family of open balls with center in D and rational radius is a basis for the topology.
Metrizable (or equivalently second-countable) spaces will be our topological frame
in this book (although some of the results could be generalized in a wider context—
usually depending on the result).
25
2 Metrizable separable and recursive spaces
(i) Every subspace of a metrizable separable space is metrizable separable (with the
induced topology).
ś
(ii) If pXn qnPω is a sequence of metrizable separable spaces, its product X “ n Xn
(with the product topology) is metrizable separable.
One easily check that d is a distance on X which defines the product topology.
Moreover if Cn is countable dense in Xn , and cn is a fixed member of Cn (we
assume here that the Xn ’s are non empty), the set C of sequences x̄ “ pxn q such
that
– @n xn P Cn , and
– for all but finitely many n’s xn “ cn ,
forms a countable dense set in X. This proves (ii). %
lim dpxn , xm q “ 0.
n,mÑ8
The metric space pX, dq is complete if every Cauchy sequence converges to some
(necessarily unique) point of X.
Any metric space can be embedded in a (unique up to isometry) complete metric
space Xp as a dense subspace. Xp is called the completion of X.
A metrizable separable space X is called Polish if for some compatible distance d on
X, pX, dq is metric complete.
Familiar examples of Polish spaces are ω (with the discrete topology — a complete
compatible distance is dpn, mq “ 0 if n “ m, and 1 if n ‰ m); R and r0, 1s with the
usual topology (the usual distance is complete).
26
2.2 Recursively presented space
So in particular the Cantor space 2ω of all sequences of 0’s and 1’s, the Baire space
ωω ,the Hilbert cube r0, 1sω , and Rω , are all Polish spaces for their usual product
topologies.
27
2 Metrizable separable and recursive spaces
28
2.2 Recursively presented space
Proposition 2.2.1. Let pXi , di , ri qiPω be r.p. (resp. Polish r.p.) spaces with associated
relation Ai . Assume
is Σ01 in ω 5 . ś
Then if we define on X “ i Xi a distance d by
ÿ
dpx̄, ȳq “ 2´i ¨ inft1, di pxi , yi qu
i
Proof. By Propositions 2.1.1 and 2.1.2 we already know that pX, dq is metric (resp.
complete metric), moreover as r enumerate all elements x̄ in X such that @i xi P ri pωq
and for all but finitely many i’s xi “ ri p0q, rpωq is dense in X. So it remains to
compute the complexity of the relation Apm, n, k, lq associated to pX, d, rq. But one
29
2 Metrizable separable and recursive spaces
x P Vm X Vn ÐÑ Dp px P Vp ^ Rpm, n, pqq.
In the sequel, we will most of the time speak of “the basic space X”, the basis being
understood, and denoted pVnX q or simply pVn q if there is no risk of confusion. We will
also refer to the Σ01 relation R as the “witness” that X is basic.
Note that we didn’t say anything about the emptiness or non-emptiness of the Vn ’s.
It will be crucial in the sequel to make no such a priori commitment.
2.3.2 Subspaces
A basic subspace of a basic space X is a structure pY, pVnY qnPω q where Y is a subset
of X with the induced topology and for each n VnY “ VnX X Y . This is indeed a basic
space, as witnessed by the same relation R.
This possibility of considering substructures is the main difference between this ap-
proach and the one in Section 2.2, based on enumerating countable dense sets.
30
2.3 Basic spaces, and recursivity in basic spaces
2.3.3 Products
Proposition 2.3.1. ś(i) Let pX0 , pVn0 qnPω q, . . . , pXk´1 , pVnk´1 qnPω q be basic spaces.
We endow X “ k´1 i“0 Xi with a structure of basic space by defining
k´1
ź
VnX “ i
Vpnqi
.
i“0
(ii) Similarly, if pXp , pVnp qnPω q is a sequence of basic spaces, as witnessed by Rp , and
if moreover the relation
Spp, m, n, kq ÐÑ Rp pm, n, kq
This is done in particular for each X ω , where X is a basic space, as in this case
the condition above is automatically fulfilled.
Proof. For (i), let R0 , . . . , Rk´1 witness that X0 , . . . , Xk´1 are basic. One has
x̄ P VnX X VmX i
ÐÑ @i ă k xi P Vpnqi
i
X Vpmqi
X
`
ÐÑ Dp x̄ P Vp ^ Rppnq0 , pmq0 , ppq0 q
˘
^ . . . ^ Rppnqk´1 , pmqk´1 , ppqk´1 q ,
R˚ pm, n, pq ÐÑ plh sm ď lh sn “ lh sp q
^ @i ă lh sm Spi, sm piq, sn piq, sp piqq
` ˘
^ @i lh sm ď i ă lh sn Ñ sp piq “ sn piq
0 ˚ ˚
ś Σ1 relation Rpm, n, pq Ø R pm, n, pq_R pn, m, pq witnesses that the product
then the
X “ Xp is basic. %
31
2 Metrizable separable and recursive spaces
We let Σ01 pXq be the (countable) family of all effectively open subsets of X, and
refer to it as its effective topology.
Note that the effective topology of the basic space ω, by the definition above, is
exactly the family of Σ01 sets as defined in Chapter 1, so that our notation is not
ambiguous for this space. The same remark applies for the product spaces ω k , viewed
as basic spaces.
(ii) Σ01 is closed under Dď , @ď and D0 , i.e. for every basic space X, and A Σ01 in
X ˆ ω (with its product structure) the subsets Dď A, @ď A of X ˆ ω and D0 A of
X, defined respectively by
px, nq P Dď A ÐÑ Dm ď n px, mq P A,
ď
px, nq P @ A ÐÑ @m ď n px, mq P A,
0
and xPD A ÐÑ Dn px, nq P A,
Proof. (i) Let A0 , A1 be Σ01 in X, so that for some Σ01 sets A˚0 , A˚1
x P A0 ÐÑ Dn px P Vn ^ A˚0 pnqq
and x P A1 ÐÑ Dn px P Vn ^ A˚1 pnqq .
32
2.3 Basic spaces, and recursivity in basic spaces
which give the result, by the closure properties of the Σ01 subsets of ω.
For @ď , note that using the witnessing relation R for X, one can write
@m ă lh sn x P VsX X
` ˘ ` ˘
n pmq ÐÑ Dp x P Vp ^ Bpp, nq
where Bpp, nq is the Σ01 relation defined by
Bpp, nq ÐÑ lh sn “ 0
_ plh sn “ 1 ^ sn p0q “ pq
´ “
_ lh sn ě 2 ^ Dl lh sl “ lh sn ´ 1 ^ Rpsn p0q, sn p1q, sl p0qq
^ @i ă lh sl ´ 1 Rpsl piq, sn pi ` 2q, sl pi ` 1qq
‰¯
^ p “ sl plh sl ´ 1q .
33
2 Metrizable separable and recursive spaces
Proof. (ii)
´ follows immediately from
¯ (i). For (i), note that if A˚ is such that px, yq P
A Ø Dk px, yq P VkXˆY ^ A˚ pkq then by definition of VkXˆY
34
2.3 Basic spaces, and recursivity in basic spaces
and
X
ś
Xi X0
f pxq P Vn ÐÑ p0 ˝ f pxq P Vpnq 0
^ ¨ ¨ ¨ ^ pk´1 ˝ f pxq P Vpnqk´1
k´1
.
These equivalences easily imply the result.
For (ii) one has
´ ω
¯
pi ˝ f pxq P VnY ÐÑ Dk lh sk ą i and sk piq “ n and f pxq P VkY
and
ω
f pxq P VnY ÐÑ @i ă sn gpx, iq P VsYn piq
` ˘
35
2 Metrizable separable and recursive spaces
fixing of recurs
Could be useful for the sequel to note now that the constant functions cα : X Ñ ω ω
arguments – YP
with cα pxq “ α for α P ω ω which are recursive are precisely the ones for which α : ω Ñ ω
is recursive. Indeed using (ii), cα is recursive iff gα px, nq “ αpnq is recursive from X ˆω
to ω iff the relation αpnq P tku “ Vkω is Σ01 in ω 2 . Now a function is recursive iff its
graph is Σ01 by Proposition 1.2.2. Also cα is recursive-in-α for each α P ω ω .
It follows that Σ01 sets are closed under fixing of recursive arguments in ω ω : If
α : ω Ñ ω is recursive and A Ď ω ω ˆ X is Σ01 then its section Aα is Σ01 in X. Indeed
x P Aα ÐÑ pα, xq P A
ÐÑ pcα pxq, xq P A
ÐÑ x P pcα ˆ IdX q´1 pAq.
and we get a space pω, pWnω qq, which is in fact basic. Now we claim that the identity
: pω, pVnω qq Ñ pω, pWnω qq is a recursive isomorphism. This is because
and
So with our convention, we will not distinguish between the basic spaces pω, pVnω qq
and pω, pWnω qq, and continue referring to them as the basic space ω.
36
2.4 Recursive spaces
This gives an enumeration of all balls centered at some rpkq and with rational radius,
so is a countable basis for the topology of X.
Moreover, if we set Spp, nq Ø dprpppq0 q, rppnq0 qq ` qppq1 ă qpnq1 then S is Σ01 in ω 2 ,
and the relation
Rpm, n, pq ÐÑ Spp, mq ^ Spp, nq
witnesses that pX, pVnX qq is basic, as one easily checks using the triangular inequality.
Definition 2.4.1.
37
2 Metrizable separable and recursive spaces
structure may not be an interesting one. For example, there are only countably many
types of Polish recursive spaces. So there are Polish spaces which admit recursive
structures, but no structure for which they become Polish recursive.
It follows from the definition that recursive spaces are closed under subspaces, prod-
ucts, and infinite powers, and Polish recursive spaces are closed under products and
infinite powers. And ω, R, r0, 1s, ω ω , 2ω , r0, 1sω , Rω are all Polish recursive for their
usual basic structure.
Pi,j “ tx P X : @k pT pi, j, kq Ñ x R Vk qu ,
Proof. Suppose first pX, pVn qq comes from a recursive presentation pd, rq. Define
` ˘
Spm, nq ÐÑ d rppmq0 q, rppnq0 q ă qpnq1 ´ qpmq1
and
` ˘
T pm, n, kq ÐÑ Spm, nq ^ d rppmq0 q, rppkq0 q ą qpmq1 ` qpkq1 .
Clearly both S and T are Σ01 . We claim that they witness that X is recursively regular.
First if x P Vn , pick ε “ ql ą 0 small enough so that
` ˘
d x, rppnq0 q ă qpnq1 ´ 2ε,
38
2.4 Recursive spaces
and choose k so that rpkq P Bpx, εq. By the triangular inequality, one gets that if
m “ xk, ly, then x P Vm ^ Spm, nq. So condition (1) is fulfilled.
For (2) we assume Spm, nq. Let P “ Pm,n . We have to show that Vm Ď P and
P Ď Vn . For the first, let x P Vm zP . Then for some k x P Vm X Vk and T pm, n, kq.
But T pm, n, kq, by the triangular inequality, implies Vm X Vk “ H, a contradiction. To
prove P Ď Vn , note first that Spm, nq not only implies that Vm Ď Vn , but also that the
closed ball Vm˚ with center rppmq0 q and radius qpmq1 is contained in Vn . So it is enough
to prove that P Ď Vm˚ . But if x P X satisfies dpx, rppmq0 qq ą qpmq1 , pick qk ą 0 so
that ` ˘
d x, rppmq0 q ą qpmq1 ` 2qk ,
and l so that rplq P Bpx, qk q. Then x P Vxl,ky and T pm, n, xl, kyq holds, so x R P . This
proves P Ď Vm˚ and finishes the proof in the r.p. case.
As a second step, note that if X is recursively regular and Y is a subspace of X,
then Y is also recursively regular, as witnessed by the same relations S and T .
As a final step assume that X is recursively regular, with witnesses S and T , and
let ϕ : X Ñ Y be a recursive isomorphism, so that for some Σ01 relations A, B on ω 2
and
Define ` ˘
S ˚ pm, nq ÐÑ Dk Dl Bpm, kq ^ Spk, lq ^ Apl, nq
and
` ˘
T ˚ pm, n, iq ÐÑ Dk Dl Dj Bpm, kq ^ Spk, lq ^ Apl, nq ^ T pk, l, jq ^ Bpi, jq .
We claim that the (clearly Σ01 ) relations S ˚ and T ˚ are witnesses that Y is recursively
regular.
For (1), note that
ÐÑ y P VnY .
For (2) fix pm, nq P S ˚ , and k, l such that Bpm, kq, Spk, lq and Apl, nq hold. Then
VmY Ď ϕpVkX q Ď ϕpVlX q Ď VnY and moreover VkX Ď Pk,l Ď VlX .
˚
Let Pm,n “ ty P Y : @ipT ˚ pm, n, iq Ñ y R ViY qu. Note that if j is such that T pk, l, jq
holds, then
Di y P ViY ^ Bpi, jq ÐÑ y P ϕpVjX q
` ˘
39
2 Metrizable separable and recursive spaces
so we get
Dj T pk, l, jq ^ Di py P ViY ^ Bpi, jqq
` ˘
ÐÑ y R ϕpPk,l q
˚
Ş
and hence Pm,n “ tϕpPk,l q : Bpm, kq ^ Spk, lq ^ Apn, lqu.
For any such k, l, one has VmY Ď ϕpVkX q Ď ϕpPk,l q so VmY Ď Pm,n˚ .
And as pm, nq P S ˚ , there is such a pk, lq, so Pm,n˚ Ď ϕpPk,l q Ď ϕpVlX q Ď VnY , and
we are done. %
This result gives a characterization of recursive spaces purely in terms of the basis,
hence of the effective topology. Of course, it is a recursive strengthening of the following
topological version, which is due to Urysohn and Tychonoff and can be obtained from
it by a relativization argument.
Notice that it follows from the proof of Theorem 2.4.4 that every recursive space
recursively embeds as a subspace of the space r0, 1sω . This is a transfer theorem to
which we will refer later on (namely in Subsection 3.5.7).
(i) If P, Q are two disjoint Π01 subsets of X, one can find two disjoint Σ01 sets P ˚
and Q˚ with P Ď P ˚ and Q Ď Q˚ .
(ii) Property (i) holds uniformly, in the sense that there are recursive functions ϕ :
ω 2 Ñ ω and ψ : ω 2 Ñ ω such that, denoting by W a good universal set for Σ01
subsets of ω, if
(
P “Xz x : Dp px P Vp ^ p P Wn q ,
(
Q “Xz x : Dp px P Vp ^ p P Wm q
40
2.4 Recursive spaces
xRP ÐÑ Dp px P Vp ^ p P Aq
and x R Q ÐÑ Dp px P Vp ^ p P Bq .
Let A1 ppq Ø Dq pApqq ^ Spp, qqq, B 1 ppq Ø Dq pBpqq ^ Spp, qqq where pS, T q witnesses
that X is recursively regular.
One still has
` ˘
x R P ÐÑ Dp x P Vp ^ p P A1
` ˘
and x R Q ÐÑ Dp x P Vp ^ p P B 1 .
and
”
(2) x P Q˚ ÐÑ Dp Dm x P Vp ^ A˚ pp, mq^
`
@ xq, ny ď xp, my B ˚ pq, nq Ñ
˘ı
Di Dj pBpiq ^ Spq, iq ^ T pq, i, jq ^ x P Vj q
Clearly P ˚ and Q˚ are Σ01 . To see they work, let us show first that P Ď P ˚ (Q Ď Q˚
is similar). If x P P , then x R Q, hence for some p, x P Vp and Bppq. So for some q, m,
x P Vq and Spq, pq and B ˚ pq, mq. Pick now any xr, ny ă xq, my (in fact this works for
any xr, ny), and suppose A˚ pr, nq. Then for some i, Apiq and Spr, iq holds, so Vr Ď Vi
and Vi X P “ H. In particular x R Vi , so x R Pr,i and for some j x P Vj and T pr, i, jq.
This says exactly that x P P ˚ .
To prove that P ˚ X Q˚ “ H, argue by contradiction and suppose x P P ˚ X Q˚ .
Then for some p, m and q, n, B ˚ pp, mq and A˚ pq, nq hold. Suppose xq, ny ă xp, my (the
case xp, my ď xq, ny is similar). Then as above one gets a pair i, j with Spq, iq and
41
2 Metrizable separable and recursive spaces
”
(1) C ˚ px, n, mq ÐÑ Dp Dq x P Vp ^ W ˚ pm, p, qq^
`
@ xr, sy ă xp, qy W ˚ pn, r, sq Ñ
˘ı
Di DjpW pn, iq ^ Spr, iq ^ T pr, i, jq ^ x P Vj q
and
”
(2) D˚ px, n, mq ÐÑ Dp Dq x P Vp ^ W ˚ pn, p, qq^
`
@ xr, sy ď xp, qy W ˚ pm, r, sq Ñ
˘ı
Di DjpW pm, iq ^ Spr, iq ^ T pr, i, jq ^ x P Vj q
` ˘
C ˚ px, n, mq ÐÑ Dp x P Vp ^ Cpn, m, pq
` ˘
and D˚ px, n, mq ÐÑ Dp x P Vp ^ Dpn, m, pq .
since W 2 is a good universal set. Finally we can use the S-m-n Theorem 1.4.4: there
are recursive functions ϕ : ω 2 Ñ ω and ψ : ω 2 Ñ ω such that
(i) Let P, Q be two disjoint Π01 subsets of X. Then there exists a recursive function
f : X Ñ r0, 1s which is 0 on P and 1 on Q.
42
2.4 Recursive spaces
then for some recursive function fm,n : X Ñ r0, 1s, An,m “ Dfn,m , and fm,n is 0
on P and 1 on Q.
(iii) One can moreover impose, in (i) and (ii), that f ´1 p0q “ P and f ´1 p1q “ Q.
´ ¯˚ ´ ¯˚
U 2k`1 “ P k and P 2k`1 “ Xz XzU k`1 ;
2n`1 2n 2n`1 n 2
using the functions ϕ and ψ, this can be done uniformly so that the relations (in
X ˆ ω ˆ ω)
x R P kn and x P U kn
2 2
43
2 Metrizable separable and recursive spaces
44
2.4 Recursive spaces
Then clearly pX, pVn qq and pX, pVn˚ qq have the same Σ01 sets, namely the Vn˚ ’s. And
moreover by the S-m-n Theorem 1.4.4,
` ˘(
(‹) there is a recursive function ϕ such that if Ak “ x P X : Dp x P Vp˚ ^ W pk, pq
is the Σ01 set in pX, pVn˚ qq coded by Wk , then Ak “ Vϕpkq
˚ .
So instead of working with arbitrary bases, we could have worked only with bases
satisfying (‹) — i.e. directly with (enumeration of) recursive topologies. This how-
ever would have two defects: First we still would have to identify spaces for which
the effective topology is the same, but the enumeration differ (as long as one can go
recursively from one to the other). And secondly one would have to work with very un-
natural bases all the time (e.g. for ω with the basis Wn given by Kleene’s Enumeration
Theorem 1.3.1). This is why we introduced arbitrary, more manageable, bases.
45
3 Borel and analytic sets. The Kleene
classes.
3.1 Borel and analytic sets in metrizable separable spaces
3.1.1 Borel sets
Let X be a metrizable separable space. The family of Borel subset of X, denoted by
BorelpXq, is the smallest family of subsets of X containing the open sets and closed
under countable union and complementation (and hence under countable intersection
too).
Corollary 3.1.2. The family of Borel sets is the least family containing the open sets
(resp. the closed sets) and closed under unions and countable intersections.
47
3 Borel and analytic sets. The Kleene classes.
Proof. Let Φ be that least family. Clearly any element of Φ is Borel. Conversely,
consider the family of Borel sets which are in Φ, and with their complement in Φ,
and check that it contains the open sets and is closed under countable unions and
complements. %
48
3.1 Borel and analytic sets in metrizable separable spaces
Then
ď
xP An ÐÑ Dn Dα px, αq P Bn
n
ÐÑ Dβ px, βq P B
where
B “ px, βq : px, β1 q P Bβp0q
(
is closed in ω ω .
Similarly
č
xP An ÐÑ @n Dα px, αq P Bn
n
ÐÑ Dβ px, βq P C
where
C “ tpx, βq P X ˆ ω ω : @n px, pβqn q P Bn u
is closed in X ˆ ω ω .
This proves (i). Statement (ii) follows by Corollary 3.1.2. To prove (iii) write A as
D1 B with B closed in X ˆ ω ω ˆ ω ω , so that
x P D1 A ÐÑ Dα px, αq P A
ÐÑ Dα Dβ ppx, αq, βq P B
ÐÑ Dγ px, γq P C
where C “ tpx, γq : ppx, pγq0 q, pγq1 q P Bu is closed in X ˆ ω ω . %
Remark 3.1.6. The notions of analytic and coanalytic sets are often defined only for
subsets of Polish spaces X, for which the definition above is equivalent to many other
definitions (we will see later). However, one should be cautious that these other classical
definitions are not equivalent to the one we adopted in arbitrary metrizable separable
spaces.
49
3 Borel and analytic sets. The Kleene classes.
Definition 3.2.1. We let Π01 “ Σ01 . A set in Π01 is called a Π01 , or effectively closed ,
set. And by induction, we set
and
Π0n`1 “ Σ0n`1
i.e. a set A is Σ0n`1 in a recursive space X iff for some Π0n set B in X ˆ ω one has
xPA ÐÑ Dn px, nq P B
The notations used for these classes come from syntactical considerations: Particu-
larizing to the space ω, we can think of the recursive relations as the elementary ones,
and then a subset of ω k is Σ0n (resp Π0n ) if it can be defined using n (alternating)
quantifiers on ω, starting with an existential (resp. a universal) one, from elementary
relations. In these notations, Σ and Π refer to the nature of the first quantifier, n to
the number of quantifiers, and the upper index 0 to the nature of the quantification:
here over ω (“objects of type 0”. The elements of ω ω are “objects of type 1”, which
explains why projection along ω ω is denoted D1 ).
Note that in this notation, effective open sets are given index 1, whereas in the
notation for Borel classes, open sets are given class 0.
Clearly one has
50
3.2 The Kleene classes in recursive spaces
(i) Every Σ0n or Σ0n pαq set is Borel of additive class n ´ 1, and
(ii) Every Π0n or Π0n pαq set is Borel of multiplicative class n ´ 1.
So the Σ02 sets could be called the effective Fσ sets, the Π02 ones the effective Gδ sets,
etc.
Ť Note however that the arithmetical hierarchy has only ω levels. The reason is that
pΣ0 Y Π0 q is closed under both D0 and (and @0 ), as they are unary operations.
n n n
In order to extend this hierarchy in the transfinite, we need other operations. This is
more intricate, and will be done only in ??. Note also that anyway one cannot hope for
ref missing – YP
a reasonable effective analog for every Borel class, for this set of classes is uncountable,
and there should be only countably many effective objects.
Again, being Σ0n or Π0n is not an intrinsic property, but heavily depends on the
recursive space X in which we consider the set. And if X is a recursive subspace of a
recursive space Y , a set is Σ0n , resp. Π0n , in X iff it is the intersection with X of a Σ0n ,
resp. Π0n , subset of Y .
We also define the ambiguous class ∆0n “ Σ0n X Π0n and ∆0n pαq “ Σ0n pαq X Π0n pαq. So
in particular ∆01 is the class of effectively clopen or recursive sets. Note that in many
spaces, like e.g. R, the class ∆01 consists only of H and R itself. But in ω, it consists
exactly, by Proposition 1.2.2, of all recursive sets, and is very rich.
51
3 Borel and analytic sets. The Kleene classes.
Proof. As for Borel classes, (i) implies (ii), by an easy induction. To prove (i), recall
the Σ01 relations S and T which witness that X is a recursively regular space, and
satisfy for i, j
x P Vj ÐÑ Di pSpi, jq and x P Vj q
and if for pi, jq P S
Pi,j “ tx : @k pT pi, j, kq Ñ x R Vk qu
then Vi Ď Pi,j Ď Vj .
Let A be any Σ01 set in X, so that
xPA ÐÑ Dp px P Vp ^ A˚ ppqq
is Σ02 in X. – YP %
52
3.3 Universal sets
thus obtaining the effective counterpart of the projective hierarchy, called Kleene’s
analytical hierarchy but we will be interested in this book only in its first level, Σ11 , Π11
and ∆11 .
One also defines accordingly the relativized classes Σ11 pαq, Π11 pαq and ∆11 pαq, for
α P ω ω , by
Σ11 pαq “ D1 Π01 pαq, Π11 pαq “ Σ11 pαq and ∆11 pαq “ Σ11 pαq X Π11 pαq.
Every Σ11 pαq set is analytic, every Π11 pαq set is coanalytic and every ∆11 pαq set is
bianalytic, for all α P ω ω .
We omit the easy proof of this proposition, which follows from the existence of
recursive bijections between ω ω and ω ω ˆ ω, ω ω ˆ ω ω and pω ω qω .
As before if X is a recursive subspace of Y , the Σ11 or Π11 sets in X are the intersection
with X of the Σ11 or Π11 subsets of Y . And this may not be true for ∆11 sets — we will
come back to this question in Section 3.4.
53
3 Borel and analytic sets. The Kleene classes.
Proof. Any open set is the union of a subfamily of the basis pVn qnPω of X. But as it
may be that no Vn is empty, hence we need the empty union to get the empty set, we
define G by
pα, xq P G ÐÑ αp0q ‰ 0 and Dp ě 1 x P Vαppq .
G is clearly Σ01 in ω ω ˆ X. If αp0q “ 0, the section Gα is empty. So the empty set is
coded. If now U is a non empty open set, pick an α : ω Ñ ω which enumerates the
non empty set of n’s with Vn Ď U . Then clearly U “ Gα . %
(ii) There exists a Σ11 set G Ď ω ω ˆ X universal for all analytic subsets of X, and
hence a set A Ď X is analytic iff A is Σ11 (and similarly for Π11 ).
using our bijection between ω ω and pω ω qω . Clearly Gn`1 is Σ0n and universal for
sets of additive class n.
54
3.3 Universal sets
(ii) is similar: Let H 1 be Π01 and universal for closed subsets of X ˆ ω ω , by (i), and
define
pα, xq P G ÐÑ Dβ pα, x, βq P H 1 .
Clearly G is Σ11 and universal for analytic subsets of X.
boldface “ topological.
It explains why the classical theory, concerned with the topological notions, and the
modern theory of the effective notions, can be put in a unified theory. In fact the
effective (or also called lightface) results, once relativized, automatically give results
for their boldface counterparts. And as we will see, they usually give much more.
In the sequel, we will use the boldface notations Σ0n , Π0n , ∆0n and Σ11 , Π11 , ∆11 for the
topological notions — even in arbitrary metrizable separable spaces.
And by analogy, for ξ ě ω, we will use Σ0ξ for the Borel sets of additive class ξ, Π0ξ
for the Borel sets of multiplicative class ξ, and ∆0ξ for the ambiguous sets of class ξ,
so that Borel “ ξ Σ0ξ “ ξ Π0ξ .
Ť Ť
(i) There is, for each n ě 1, a Σ0n subset of ω ω which is not Π0n . So in particular the
arithmetical hierarchy and the finite Borel hierarchy are both strict hierarchies on
the space ω ω .
(ii) There is a Σ11 subset of ω ω which is not Π11 , and (by taking complements) a Π11
subset of ω ω which is not Σ11 . In particular Σ11 ‰ Π11 and Σ11 ‰ Π11 in the space
ωω .
Proof. We apply Cantor’s diagonal method. Let us prove (ii), (i) being similar. Let
G Ď ω ω ˆ ω ω be Σ11 and universal for Σ11 subsets of ω ω , by Theorem 3.3.2, and let
αPH ÐÑ pα, αq P G
Clearly H is Σ11 in ω ω . Now H is not Π11 , otherwise one would have, for some α P ω ω
βRH ÐÑ pα, βq P G
55
3 Borel and analytic sets. The Kleene classes.
and in particular
Theorem 3.3.4.
(i) For each (recursive) space X, there is a Σ0ξ set G in ω ω ˆ X universal for Σ0ξ
subsets of X (and similarly for Π0ξ ).
(ii) The Borel hierarchy pΣ0ξ , Π0ξ qξăω1 is strict on the space ω ω .
Proof. (ii) follows from (i) by the diagonal method used in the proof of Theorem 3.3.3.
And (i) is proved by induction on ξ. We already proved it for finite n (in a stronger
version in Theorem 3.3.2). Suppose we know the result for all η ă ξ, and let H η P Π0η
be universal for Π0η subsets of X. Let ϕ : ω ξ be a surjection, and define
The set is clearly Σ0ξ , and we leave to the reader the easy checking that it is universal
for Σ0ξ subsets of X. One defines H ξ “ Gξ , and the proof is finished. %
Theorem 3.3.5. Let X be a recursive space, and Γ one of the Kleene classes Σ0n , Π0n , Σ11
or Π11 .
xPA ÐÑ pε, xq P G.
56
3.3 Universal sets
Proof. (i) We prove it first for Σ01 . By Kleene’s Enumeration Theorem 1.3.1, there
is a Σ01 set W Ď ω 2 which is universal for Σ01 subsets of ω. Define then
Gpn, xq ÐÑ Dp px P VpX ^ W pn, pqq
Clearly G is Σ01 in ω ˆ X, and universal for Σ01 subsets of X as desired.
To prove it for all Kleene classes, it is enough to note that
(a) if G Ď ω ˆ X is universal for Γ, its complement is universal for the dual
class Γ.
(b) if G Ď ω ˆ pω ˆ Xq is universal for Γ subsets of ω ˆ X, then H defined by
pn, xq P H ÐÑ Dm pn, pm, xqq P G
is universal for D0 Γ subsets of X, and
(c) if G Ď ω ˆ pω ω ˆ Xq is universal for Γ subsets of ω ω ˆ X, then H “ D1 G
defined by
pn, xq P H ÐÑ Dα P ω ω pn, pα, xqq P G
is universal for D1 Γ.
Facts (a) to (c) are obvious, and prove (i).
The proof of (ii) is entirely similar. The only point here is to start with a uniform-
in-α version of Kleene’s enumeration theorem: There exists a Σ01 subset W of
ω ω ˆ ω 2 such that for each α P ω ω , its section Wα is universal for Σ01 pαq subsets
of ω. But this is exactly what the (relativized) proof of Kleene’s enumeration
theorem described in Subsection 1.5.3 gives, as the reader may check. %
3.3.7 Relativization
Remark 3.3.6. The relativized Kleene classes Γpαq have been defined using the notion
of recursive-in-α functions from ω k Ñ ω. But the preceding result implies that in any
recursive space X, a subset of X is in Γpαq iff it is the section at α of a Γ subset of
ω ω ˆ X, so the relativized classes could be defined using sections.
This is because Γpαq
is closed under fixing
3.3.8 The hierarchy theorem on ω of recursive-in-α
arguments in ω ω . –
Theorem 3.3.7 (the hierachy theorem for Kleene classes). YP
(i) There exists for each n a Σ0n subset of ω which is not Π0n (and conversely, using
its complement).
(ii) There exists a Σ11 subset of ω which is not Π11 (and conversely).
Proof. Apply the diagonal method used in the proof of Theorem 3.3.3 to the universal
Γ set for Γ subsets of ω given by 3.3.5. %
This result says that although the topological notions are of no interest for the
space ω — every subset is clopen in it, the effective notions provide an interesting
classification of the complexity of subsets of ω.
57
3 Borel and analytic sets. The Kleene classes.
Theorem 3.3.8 (Existence of good universal sets). Let X be a recursive space, Γ one
of the Kleene classes.
pn, xq P A ÐÑ pf pnq, xq P W Γ .
pα, xq P A ÐÑ pf pαq, xq P WΓ .
n0 , pn, xq P GΓ
` ˘
pn, xq P A ÐÑ
ÐÑ xn0 , ny , x P W Γ
` ˘
α0 , pα, xq P GΓ
` ˘
pα, xq P A ÐÑ
ÐÑ xα0 , αy , x P WΓ
` ˘
so that the continuous function f pαq “ xα0 , αy works. Note that this function is
in fact recursive-in-α. %
Sets W Γ , WΓ are called good universal sets for Γ, Γ respectively. From now on,
we fix one of them for each Kleene class Γ and recursive space X (the superscript Γ
will often be omitted — as we already did for the space X! — if it is clear from the
context).
58
3.4 Borel and Γ-recursive functions
pε, xq P A ÐÑ pε, xq P W Γ .
pε, xq P A ÐÑ pε, xq P WΓ .
59
3 Borel and analytic sets. The Kleene classes.
Proposition 3.4.2. A function is Σ11 -measurable iff it is Π11 -measurable iff it is ∆11 -
measurable.
Proof. Note that if f is Γ-measurable, then for any Gδ — hence any closed — set G,
f ´1 pGq is a countable intersection of sets in Γ. So if f is Σ11 -measurable, f ´1 p U q “
f ´1 pU q is in Σ11 too for any open set U , i.e. f is ∆11 -measurable. And similarly for
Π11 -measurable. %
(i) Let f : X Ñ Y be bianalytic. If A Ď Y is Σ11 (resp. Π11 , ∆11 ) then f ´1 pAq is Σ11
(resp. Π11 , ∆11 ).
Proof. It is clearly enough to prove (i) for A P Σ11 . Pick B closed in X ˆ ω ω such that
A “ D1 B, and consider
(So we are asking that each f ´1 pVnY q is in Γ, and that this holds uniformly in n.)
Note that for Γ “ Σ01 , this corresponds to our earlier definition of a recursive function
f : X Ñ Y , i.e. Σ01 -recursive = recursive.
60
3.4 Borel and Γ-recursive functions
(ii) Σ11 , Π11 and ∆11 are closed under substitution by ∆11 -recursive functions.
Proof. (i) Recall the Σ01 relations S and T which witness that Y is recursively reg-
ular. And note that
It follows that if Df is Σ11 , it is also Π11 , and if it is Π11 , it is also Σ11 . This proves
(i).
61
3 Borel and analytic sets. The Kleene classes.
f pxq P A ÐÑ Dα pf pxq, αq P B.
and finally
ω
ÐÑ Dα @p @n f pxq R VpY _ α R Vnω _ P pp, nq
` ˘
f pxq P A
3.4.8 Uniformities
Remarks 3.4.6. (1) As usual, the closure properties proved in the preceding propo-
sition are uniform in the codes: E.g. if f : X Ñ Y is ∆11 -recursive, there is
a recursive function g : ω Ñ ω which, for any code n of a Σ11 subset A of Y ,
produces a code gpnq of the Σ11 subset f ´1 pAq of X. To see this, it is enough to
note that the relation f pxq P WnY where W Y is the coding for Σ11 subsets of Y ,
is Σ11 in X ˆ ω, and apply Theorem 3.3.8 to it and the coding W X of Σ11 subsets
of X.
(2) A function is ∆11 -measurable iff it is ∆11 pαq-recursive for some α P ω ω (For if
f is ∆11 -measurable, Df is clearly ∆11 in X ˆ ω). So the relativized version of
Proposition 3.4.5 gives Propositions 3.4.2 and 3.4.3 again.
Proof. Composing with recursive isomorphisms, we may as well assume that Z is re-
cursively presented, by d and r. Denote by R, S, and T the canonical Σ01 relations
witnessing that Z is recursive.
By the hypothesis on f , Df is Σ0n in X ˆ ω, hence there is a Σ0n subset A of Y ˆ ω
such that
py, nq P Df ÐÑ y P X ^ Apy, nq.
62
3.4 Borel and Γ-recursive functions
! )
Define X
r “ y P Y : Dz P Z tn : Apy, nqu “ tn : z P VnZ u . Clearly X Ď X r as
for x P X tn : Apy, nqu “ tn : f pxq P VnZ u. Moreover if y P X r there is a unique z
˜
witnessing this fact, and this defines a function f : X Ñ Z, which extends f . And
r
ÐÑ y P X
r ^ Apy, nq
A˚ “tx P Ar : g̃ ˝ f˜pxq “ xu
r : f˜ ˝ g̃pyq “ yu.
and B ˚ “ty P B
63
3 Borel and analytic sets. The Kleene classes.
A˚ and B ˚ are Π02 sets too, and one easily checks that f˜A˚ is a recursive isomorphism
between A˚ and B ˚ , with inverse g̃B ˚ , which extends f . %
Proof. Let i be a recursive embedding of the Polish recursive space X into a recursive
space Y , which we can consider as a subspace of Polish recursive space Z. It is enough
to show that the image ipXq is Π02 in Z — for then the same will hold in Y .
Apply Lavrentiev’s Theorem 3.4.8 to the spaces Z, X and the recursive isomorphism
i´1 : ipXq Ñ X. By this theorem, i´1 extends to a recursive isomorphism between
64
3.5 Special recursive spaces
two Π02 sets of Z and X. But as i´1 has no proper extension, this means that ipXq is
Π02 in Z. %
Lemma 3.5.3.
Proof. For (i), write Df px, nq Ø Dp px P VpX ^ Apn, pqq with some Σ01 set A, and notice
that
For (ii), it suffices by (i) to prove it for Polish recursively presented spaces. But for
those spaces
VnX ‰ H ÐÑ qpnq1 ą 0
and we are done. %
Lemma 3.5.4. Suppose X is closed in the Polish r.p. space pZ, d, rq, and DX is Σ01 .
Then X admits a recursive presentation pd, r1 q with the same distance.
Proof. For the restriction of d, X is clearly metric and complete. We may assume that
DX ‰ H (otherwise X “ H!). Let f : ω Ñ ω be a recursive enumeration of DX .
Consider
1
Apn, pq ÐÑ Spf ppq, f pnqq ^ qpf ppqq1 ď qpf pnqq1
2
where S is the usual Σ01 relation witnessing Z is recursively regular. Clearly A is Σ01 ,
and moreover for every n, as VfZpnq ‰ H, we can find q with VqZ X X ‰ H, Spq, f pnqq,
and diam VqZ ď 12 diam VnZ , i.e. we can find p so that Apn, pq holds.
By uniformization, there is a recursive ϕ : ω Ñ ω such that @n Apn, ϕpnqq. Let
just let Apn, pq Ø
ψpk, nq “ ϕpkq pnq. The function ψ is recursive, and moreover for each n the sequence DkBpn, p, kq with B
Z
pV f pψpk,nqq : k P ωq converges in Z to some point of X, we denote it by r1 pnq. We claim recursive in ω 3 and
let ϕpnq “
pminl Bpn, plq0 , plq1 qq0
– YP
65
3 Borel and analytic sets. The Kleene classes.
that pd, r1 q works. It is enough to compute the complexity of ql ă dpr1 pnq, r1 pmqq ă qk .
But e.g.
` ˘
d r1 pnq, r1 pmq ă qk ÐÑ Di Dj Dl qf pψpi,nqq1 ă 2´l
and qf pψpj,mqq1 ă 2´l
´ ¯
and d rpf pψpi, nqq0 q, rpf pψpj, mqq0 q ă qk ´ 2l`1
Lemma 3.5.5. Let X be Π02 in the Polish r.p. space Z. Then X is recursively iso-
morphic to a closed subspace of Z ˆ Rω .
Define then ϕ : X Ñ Z ˆ Rω by
ˆ ´ ¯˙
1
ϕpxq “ x, :pPω .
f px, pq
66
3.5 Special recursive spaces
required. %
3.5.7 Compactness
To finish with this section on Polish recursive spaces, let us mention two transfer results,
which enable to reduce the general situation to specific ones.
We have already seen such a transfer theorem: Every recursive space can be consid-
ered as a subspace of the space r0, 1sω , which has the advantage of being compact. In
cf Subsection 2.4.6
fact, using Theorem 3.5.1 and Lemma 3.5.4, one gets and proof of
Theorem 2.4.4 – YP
Theorem 3.5.7. Every Polish recursive space admits a Polish recursive completion
which is a compact space.
Proof. We may assume X is a subspace of r0, 1sω , and by Theorem 3.5.1, X is Π02 and
DX is Σ01 in ω. Consider the closure X of X in r0, 1sω . As Vn XX ‰ H Ø Vn XX ‰ H,
DX “ DX is Σ01 . By Lemma 3.5.4, X is Polish recursive, and of course compact and
a completion of X. %
˘ı
_ spnqi “ 1 ^ x P VnX
`
67
3 Borel and analytic sets. The Kleene classes.
p is a Π0 subset of ω ω .
Proposition 3.5.8. If X is Polish recursively presented, X 2
Proof. In the same way as we computed the complexity of X r in the proof of Theo-
rem 3.4.7 one easily checks that
α P 2ω ^ @n αpnq “ 1 Ñ qpnq1 ą 0 ^
` ˘
αPX p ÐÑ
”` ˘
piq @k @m @n αpnq “ 1 ^ αpmq “ 1 Ñ
` ˘ı
Dp αppq “ 1 ^ Rpm, n, pq ^ 0 ă qppq1 ď 2´k
” ` ˘ ı
^ piiq @n Dp Spp, nq ^ αppq “ 1 Ñ αpnq “ 1 .
where R, S, T are the usual Σ01 relations which witness X is recursively presented. This
gives a Π02 definition of X.
p %
Proof. Let A Ď X be Σ01 , so that for some Σ01 subset A˚ of ω one has x P A Ø
Dp px P Vp ^ A˚ ppqq. Write A˚ ppq Ø Dm Bpp, mq with B ∆01 in ω 2 , so that
´ ` ˘¯
x P A ÐÑ Dn x P Vpnq0 ^ B pnq0 , pnq1 .
` ˘
By the hypothesis the relation x P Vpnq0 ^ B pnq0 , pnq1 is ∆01 in X ˆ ω, as desired. %
68
3.5 Special recursive spaces
ωω
` ˘
α P A ÐÑ @n Dm α P Vpmq 0
^ Bpm, nq .
(iii) Σ11 is closed under DX , i.e. if Y is any space and A Ď Y ˆX is Σ11 , then B “ DX A
defined by
y P B ÐÑ Dx P X py, xq P A
is Σ11 in Y .
69
3 Borel and analytic sets. The Kleene classes.
p “ iX pXq is Σ1 in ω ω .
(iv) The set X 1
y P DX A ÐÑ Dx P X py, xq P A
ÐÑ Dα P ω ω α P Zp ^ py, jZ pαqq P A
` ˘
y P DX A ÐÑ Dα P X
p py, jX pαqq P A.
So it remains to prove that (iii)Ñ(iv) and (iii)Ñ(i). This follows from the following
more general lemma, applied to iX in the first case, and a recursive embedding i in the
second case. %
Proof. Write:
y P f pXq ÐÑ Dx P X y “ f pxq
” ` ˘ı
ÐÑ Dx P X @n y P VnY Ø Df px, nq .
70
3.5 Special recursive spaces
hkkαp0q
ikkj hkkαp1q
ikkj hkkαp2q
ikkj
0¨¨¨0 1 0¨¨¨0 1 0¨¨¨0 1¨¨¨ ,
or by using Lemma 3.5.9. %
It follows from Corollary 3.5.13 that although a ∆11 -recursive image of a Luzin-
recursive space can be any Suslin-recursive space, it may not be a Luzin-recursive
space. However we will see later that a ∆11 -recursive one-to-one image of a Luzin-
recursive space is still a Luzin-recursive space, one of the deep result of the theory.
71
3 Borel and analytic sets. The Kleene classes.
(ii) A set A Ď X is a B-set iff A is the intersection with X of a ∆11 -absolute set.
Proof. (i) is immediate: if A is a ∆11 subset of the Luzin-recursive space X Ď Y ,
then as X is ∆11 in Y , A is also ∆11 in Y .
(ii) If A is a B-set in X, then A is the intersection with X of some ∆11 set B in some
Polish recursive completion of X, and B is Luzin-recursive by Theorem 3.5.14.
For the converse, let A “ X X A r where A r is ∆1 in the Polish recursive space Y ,
1
and let i be a recursive embedding of X into Z, which we may assume Polish
recursive. Let ı̃ be a recursive extension of the embedding i to two Π02 subsets
X 1 Ě X and Z 1 of Y and Z. A r X X 1 is ∆1 in X 1 , hence ı̃pAr X X 1 q is ∆1 in
1 1
ı̃pX 1 q “ Z 1 , hence in Z too, as Z 1 is Π02 in Z. And clearly ı̃pAXX
r 1 q XipXq “ ipAq
as desired. %
72
3.5 Special recursive spaces
(1) We will later extend Definition 3.5.15 (i) to Suslin spaces. And we will show that
in general, in fact even for some Π11 -absolute spaces, the classes B and ∆11 do not
coincide.
(2) One can easily introduce relativized classes Bpαq and B. Note that as Borel sets
of say additive class ξ can always be extended by Borel sets of the same class,
one has for any space X
We will again see later (cf Corollary 4.4.7) that in fact Borel “ B in any space
X, which explains our terminology of B-(and B-)sets.
73
4 Structural properties of Σ11 and Π11
sets in recursive spaces
4.1 Suslin’s operation A
4.1.1 Definition
For any set X, a Suslin scheme on X is a sequence pAs qsPωăω of subsets of X, indexed
by the set ω ăω of finite sequences of integers. The Suslin scheme pAs qsPωăω is regular
if s Ď t implies At Ď As .
The result of operation A on the Suslin scheme pAs qsPωăω , sometimes called its
kernel, in notation AppAs qq is defined by:
ď č
AppAs qq “ Aαn
αPω ω n
i.e.
AppAs qq “ x P X : Dα P ω ω @n x P Aαn
(
For Γ a class of sets, we denote as usual ApΓq the class of all AppAs qq, for pAs qsPωăω
a Suslin scheme with values in Γ.
It follows
Ş immediately from the definition that if pAs qsPω is any Suslin scheme,
ăω
and Bs “ tĎs At , then pBs qsPωăω is a regular Suslin scheme and moreover
AppAs qq “ AppBs qq
so that for classes of set closed under finite intersections, one can consider only regular
Suslin schemes.
And hence,
75
4 Structural properties of Σ11 and Π11
Proof. (i) Ť Note that for As “ A a constant scheme, ApAs q “ A; so Γ Ď ApΓq. Now
if A “ nPω An with An P Γ, define
#
X if s “ H;
Bs “
Asp0q otherwise.
Ş
Then AppBs qq “ A; so Γσ Ď ApΓq. Similarly if A “ nPω An with An P Γ,
define Cs “ Alh s . Then AppCs qq “ A; so Γδ Ď ApΓq.
(ii) Let A “ AppAs qq, with each As , s P ω ăω , in ApΓq, i.e. As “ AppBts qq, where for
each t P ω ăω , Bts P Γ. By definition:
α
xPA ÐÑ Dα P ω ω @n Dβ P ω ω @p x P Bβ n
p
ω n ω ω α
(4.1) ÐÑ Dα P ω Dpβ q P pω q @n @p x P Bβ nn .
p
and by (‹) and (‹‹), pγq0 pnq0 and pγqn0 `1 pnq1 depend only on γn , from which
(ii) immediately follows. %
4.1.3 Trees
A tree on a set X is a subset T of the set X ăω of finite sequences from X which is
closed under restrictions, i.e. if s P T and t Ď s, then t P T . A tree T is pruned if
every sequence s P T admits a strict extension in T .
A branch through a tree T is a sequence f P X ω such that for all n, f n P T . The
set of all branches through T is denoted rT s.
Let us give X the discrete topology, and X ω the product topology (this is the usual
topology for X “ ω or X “ 2), so that a basis of open (in fact clopen) sets consists of
the sets
Vs “ tf P X ω : s Ď f u ,
76
4.1 Suslin’s operation A
f R rT s ÐÑ Ds R T f P Vs ,
the set rT s is closed in X ω . Conversely, every closed subset F of X ω is the set of all
branches through some tree T , in fact through a unique pruned tree TF on X.
Indeed, let TF “ ts P X ăω : F X Vs ‰ Hu. One immediately checks that TF is a
pruned tree on X. Clearly, if f P F and s Ď f , then s P TF , so F Ď rTF s1 . Now if
f R F , there is an s P X ăω with s Ď f and Vs X F “ H, because F is closed. So
f R rTF s. Finally the uniqueness of the pruned tree T with rT s “ F follows from the
fact that for any pruned tree T and s P T , Vs X rT s ‰ H.
B “ px, αq P X ˆ ω ω : @n x P Bαn
(
“ tpx, αq P X ˆ ω ω : α P rTx su
1
None of this needs that F is closed.
77
4 Structural properties of Σ11 and Π11
4.1.6 Approximations of A
We can use Subsection 4.1.4 and the pruning method of Subsection 4.1.5 “sectionwise”
to analyze operation A.
Proposition 4.1.2. Let Γ be a class of sets, and BpΓq the least family containing Γ
and closed under countable unions and complements. For every A P ApΓq there exist
transfinite sequences pBξ qξăω1 and pCξ qξăω1 of sets in BpΓq, with pBξ q increasing and
pCξ q decreasing such that ď č
A“ Bξ “ Cξ .
ξPω1 ξPω1
Proof. We may as well assume Γ is closed under finite intersections, and let pAs qsPωăω
be a regular Suslin scheme with A “ AppAs qq.
Define inductively on ξ ă ω1 a transfinite sequence pAξs qsPωăω by:
A0s “ As
ď ξ
Aξ`1
s “ Asa n
nPω
č
Aλs “ Aξs if λ is limit.
ξăλ
Clearly each pAξs qsPωăω is a regular Suslin scheme of sets in BpΓq. Moreover if x P X
and Txξ “ ts P ω ăω : x P Aξs u, we get Tx0 “ Tx , Txξ`1 “ dpTxξ q and Txλ “ ξăλ Txξ
Ş
for limit λ, i.e. pTxξ qξăω1 is the inductive transfinite sequence associated to Tx in
Subsection 4.1.5.
So if we let
! )
Bξ “ x : Txξ “ Txξ`1 and Txξ ‰ H
! )
and Cξ “ x : Txξ ‰ H
we get by Subsections 4.1.4 and 4.1.5
x P AppAs qq ÐÑ rTx s ‰ H
ÐÑ @ξ ă ω1 Txξ ‰ H
ÐÑ Dξ ă ω1 Txξ “ Txξ`1 and Txξ ‰ H ,
` ˘
Ť Ş
i.e. A “ AppAs qq “ ξăω1 Bξ “ ξPω1 Cξ .
Clearly the Bξ ’s are increasing and the Cξ ’s are decreasing. So it remains to prove
that they are in BpΓq. But
x P Cξ ÐÑ Txξ ‰ H
ÐÑ H P Txξ
ÐÑ x P AξH
´ ¯
and x P Bξ Ø x P Cξ ^ @s P ω ăω x P Aξs Ñ x P Aξ`1
s . So Cξ “ AξH is in BpΓq, and
Bξ “ AξH X sPωăω Aξ`1 Y pXzAξs q is in BpΓq too.
Ş ` ˘
s %
78
4.1 Suslin’s operation A
(1) A Ď A,
p
Ats “Ata s
and let At “AppAts qsPωăω q.
a
“ A. Moreover At “ nPω At n , as
Ť
Clearly AH
x P At ÐÑ Dα P ω ω @n x P Atαn
ÐÑ Dn0 P ω Dβ P ω ω @n x P Atn0 a β
n
a
ÐÑ Dn0 P ω Dβ P ω @n x Pω
Atβ n0
n
an
ÐÑ Dn0 P ω x P At 0
.
A“A y H zpA
y H zAq P C too.
79
4 Structural properties of Σ11 and Π11
4.1.9 Operation A
We now introduce a lightface analog of operation A — which will of course be a unary
operation.
Definition 4.1.6. For any set X, and B a subset of X ˆ ω ăω , we define
ApBq “ tx P X : Dα P ω ω @n px, αn q P Bu.
So if Bs “ tx : px, sq P Bu are the sections of B, we get ApBq “ ApBs q. This means
that there is nothing really new in this definition, except that we have turned A into
a unary operation.
80
4.1 Suslin’s operation A
Σ11 “ ApΠ01 q.
Moreover if A is Σ11 in X:
px, sq P B ^ t Ď s Ñ px, tq P B.
is Π01 in X ˆ ω ω , as
ω
Dn α P Vnω ^ px, sn q R B
` ˘
px, αq R B
r ÐÑ
ω
As ApBq “ Dω B, r ApBq is Σ1 in X.
1 ω
Conversely, let A be Σ11 in X, so that for some Π01 set C Ď X ˆ ω ω , A “ Dω C. By
Proposition 2.3.4 (i), there is a Σ01 relation R Ď ω 2 such that
ω
px, αq R C ÐÑ DkDl x P VkX ^ α P Vlω ^ Rpk, lq .
` ˘
Let then
Dk x P VkX ^ Rpk, xsyq .
` ˘
px, sq P B ÐÑ
B is a Π01 subset of X ˆ ω ăω , and one checks easily that
px, sq P B 1 ÐÑ @t Ď spx, tq P B
81
4 Structural properties of Σ11 and Π11
which is Π01 , satisfies (i), and is such that C “ tpx, αq : @n px, αn q P B 1 u.
To prove (ii) we may as well assume X “ ω ω , and the proof will follow if we can
show that every Π01 subset of ω ω ˆ ω ω is of the form rT s for some recursive tree T on
ω ˆ ω.
(Note that there is a slight abuse of notation here: by definition, a tree on ω ˆ ω is
a subset of pω ˆ ωqăω , but we view it as a subset of ω ăω ˆ ω ăω , consisting of pairs of
the same length, by identifying a sequence of pairs
` ˘
pu0 , v0 q, pu1 , v1 q, . . . , puk´1 , vk´1 q
` ˘
with the pair pu0 , . . . , uk´1 q, pv0 , . . . , vk´1 q ; and similarly for infinite sequences.)
We do it for Π01 subsets of ω ω , the case of ω ω ˆ ω ω (or more generally pω ω qn ) being
the same. So let B be Π01 in ω ω . By definition, there is a recursive R Ď ω 2 such that
ω
α R B ÐÑ Dk α P Vkω ^ Dn Rpn, kq .
` ˘
Define ! )
T “ s P ω ăω : @k ď lh s @n ď lh s Rpn, xsk yq .
The ordinal f pZq “ tf pzq : z P Zu is called the length of the wellordering, and de-
noted by lhpZq. One also writes f pzq “ lhpz, Zq, the length, or rank of z in Z. Note
that if Zz “ ty P Z : y ă zu is ordered by the restriction of ď, then Zz is a wellorder-
ing, and by (4.2):
82
4.2 Coding the ordinals
4.2.2 LO and WO
For α P ω ω , we define a structure pDα , ďα q by:
Dα “ tn P ω : αpxn, nyq “ 1u ,
and for n, m in Dα :
n ďα m ÐÑ αpxn, myq “ 1
where x , y is our usual bijection between ω 2 and ω.
We set:
LO “ tα P ω ω : pDα , ďα q is a linear orderingu
and
WO “ tα P ω ω : pDα , ďα q is a wellorderingu .
Proposition 4.2.1. (i) LO is a Π01 subset of ω ω ;
(ii) WO is a Π11 subset of ω ω ;
(iii) the function α ÞÑ |α| : WO Ñ On defined by
|α| “ lhppDα , ďα qq
is onto ω1 , i.e. is a coding of all countable ordinals.
Proof. The relations n P Dα and n ďα m are recursive in ω ˆ ω ω and ω 2 ˆ ω ω respec-
tively as:
n P Dα ÐÑ αpxn, nyq “ 1
and
n ďα m ÐÑ αpxn, nyq “ 1 and αpxm, myq “ 1 and αpxn, myq “ 1.
Now
“
α P LO ÐÑ @n@m@p ppn P Dα ^ m P Dα q Ñ pn ďα m _ m ďα nqq^
ppn P Dα ^ m P Dα ^ n ďα m ^ m ďα nq Ñ m “ nq^
ppn P Dα ^ m P Dα ^ p P Dα ^ n ďα m ^ m ďα pq
‰
Ñ n ďα pq
so LO is Π01 in ω ω .
Similarly:
α P WO ÐÑ α P LO ^ @β rp@n βpnq P Dα q Ñ Dnpβpnq ďα βpn ` 1qqs
and hence WO is a Π11 subset of ω ω .
To prove (iii), let ξ be a countable ordinal, and ϕ some bijection between a subset
D of ω and ξ. Set
#
1 if n, m P D and ϕpnq ď ϕpmq;
αpxn, myq “
0 otherwise.
Then clearly Dα “ D, and ϕ is strictly increasing from ďα into ξ so that ďα is a
wellordering and lhppDα , ďα qq “ ξ %
83
4 Structural properties of Σ11 and Π11
4.2.3 Complexity of ď on WO
Now that we have a coding for ω1 , we can analyze the complexity of various structures
on ω1 , in the codes. We start with the natural ordering on ω1 , i.e. membership (or
inclusion). It turns out that it is ∆11 in the codes on WO ˆ WO, and in fact a bit
better, as given by the following.
are Π11 in ω ω ˆ ω ω .
Proof. Let
´
Rpα, β, γq ÐÑ @n P Dα γpnq P Dβ ^
@m, n pm ďα n Ø γpmq ďβ γpnqq ^
@m, nppm P Dβ ^ n P Dα ^ m ďβ γpnqq Ñ
¯
Dp P Dα m “ γppqq .
The relation R is Π02 in pω ω q3 , and expresses the fact, for α, β in LO, that ďα is
embedded, via γ in ďβ as an initial segment.
Now (ii) is equivalent to
α P WO ^ Dγ Rpβ, α, γq.
Similarly, if
α P WO ^ Dγ R1 pβ, α, γq.
The computation given by the preceding proposition will be important in the sequel.
It of course implies that |α| ď |β| (and |α| ă |β|) are ∆11 on WO ˆ WO, but is quite
stronger.
84
4.2 Coding the ordinals
85
4 Structural properties of Σ11 and Π11
s ďBK t ÐÑ t Ď s
or for the least k with spkq ‰ tpkq, spkq ă tpkq.
So ďBK is the lexicographical ordering, except that a longer word is considered less
than a smaller one.
Proof. If α is a branch through T , then the sequence pαn qnPω is strictly decreasing in
pT, ďBK q. So (ii) implies (i).
Suppose conversely that psn qnPω is strictly decreasing in pT, ďBK q. Note that, except
maybe for s0 , lhpsn q ě 1; and the sequence psn p0qqně1 must be decreasing (at large) in
ω, hence is constant from some n0 on. But then except maybe for sn0 (which could be
psn0 p0qq), the length of sn is greater or equal to 2 from this point on, and the sequence
psn p1qqněn0 must be decreasing at large in ω, hence stabilize from some n1 on, etc.
So we can easily construct an increasing sequence nk of integer such that for n ą nk ,
lhpsn q ě k ` 2 and sn pkq “ snk pkq. But then α defined by αpkq “ snk pkq is a branch
through T . %
(ii) for α P T :
86
4.3 The representation theorem for Π11 sets
xPA ÐÑ f pxq P WO
xPA ÐÑ f pxq P WO
If moreover X is a dim0 recursive space, we can apply the second part of Theorem 4.1.7,
as X is recursively isomorphic to a subspace of ω ω (cf Subsection 3.5.9). The function
ψ, defined as before, is then recursive and f is too. %
Corollary 4.3.2. The set WO of codes of ordinals is a Π11 non Σ11 set.
Proof. Otherwise by the preceding theorem any Π11 set could be Σ11 , contradicting
Theorem 3.3.7 (ii). %
The above method for proving that a given set is Π11 non Σ11 is called the completeness
method — and any Π11 set satisfying the analog of the theorem is called a complete
set. We will see later on other applications of this method.
87
4 Structural properties of Σ11 and Π11
(1) x P A ^ py R A _ x ď yq,
(2) x P A ^ py R A _ x ă yq,
Theorem 4.3.4 (The Prewellordering Theorem). Any Π11 set A in some recursive
space admits a Π11 -prewellordering, i.e. there is a prewellordering ďA on A such that
pA, ďA q is a Π11 -prewellordering (and similarly for Π11 pαq, Π11 ).
Proof. Let A be Π11 in X, and by Theorem 4.3.1 let f : X Ñ LO be ∆02 -recursive with
f pxq P WO ÐÑ x P A.
Define then
x ďA y ÐÑ |f pxq| ď |f pyq| .
Clearly ďA is a prewellordering. And the relations
x P A ^ py R A _ x ďA yq
and x P A ^ py R A _ x ăA yq
88
4.3 The representation theorem for Π11 sets
Π11 pαq, Π11 ) rank. The Π11 -ranks that arise from the Representation Theorem 4.3.1 as
in Theorem 4.3.4, are often called Luzin-Sierpinski indices.
Of course there may be many natural ranks on a given set. For example, we already
have given two ordinal assignments to wellfounded trees: One is lhpT, ďBK q, as in
Subsection 4.2.5, which is a Π11 -rank — in fact a Luzin-Sierpinski index — by Corol-
lary 4.2.6. Another one is the least ordinal ξ with Tξ “ H — where pTξ qξPω1 is the
transfinite sequence of trees defined by the pruning operation in Subsection 4.1.5. That
this is also a Π11 -rank is not as easy and will follow from results in Subsection 5.5.12,
see Subsection 5.5.14.
The previous result easily relativizes to the classes Π11 pαq and Π11 (and resp. ∆11 pαq-
recursive and ∆11 -recursive uniformizing functions). The boldface version is sometimes
rephrased in a slightly different way, which is worth putting down:
Corollary 4.3.6 (Kuratowski’s “second reduction theorem”). Let pAn qnPω be a se-
quence of Π11 subsets of some metrizable separable space X. There exists a sequence
pA˚n q of Π11 sets with the following properties:
(i) A˚n Ď An ,
Ť Ť
(ii) n A˚n “ n An ,
Proof. Apply the relativized version of the ω-Uniformization Theorem to the set A Ď
X ˆ ω defined by px, nq P A Ø x P An . Let f : n An Ñ ω be a ∆11 -uniformizing
Ť
function. Then A˚n “ f ´1 pnq clearly works. %
89
4 Structural properties of Σ11 and Π11
90
4.4 Boundedness theorems
(ii) can be proved by a similar argument, using a good universal Π11 set W Ď ω ω ˆ ω ω
for Π11 subsets of ω ω this time. %
Note that if A and B are two Π11 subsets if ω which are inseparable by a ∆11 set,
then X “ A Y B is a (co-Suslin) recursive space in which a ∆11 set, namely A, cannot
def? – YP
be extended to any ∆11 subset of ω. Similarly there exists in ω ω a (co-Suslin) recursive
def? – YP
space X and a ∆11 subset of X which cannot be extended to any ∆11 subset of ω ω .
These examples explain why we introduced the notion of B-sets in Subsection 3.5.13.
(2) The ordinal ω1CK , called “the Church-Kleene ω1 ”, (resp. ω1CK´α ) is defined as the
least non recursive (resp. non recursive in α) ordinal.
(iii) ω1CK is “recursively regular”, i.e. if α is recursive and @n pαqn P WO, then
supn |pαqn | ă ω1CK .
And similarly for the ω1CK´α ’s.
91
4 Structural properties of Σ11 and Π11
supt|α| : α P Au ă ω1 .
supt|α| : α P Au ă ω1CK .
β R WO _ pα P WO ^ |α| ď |β|q
β P WO ^ pα R WO _ |α| ă |β|q.
(4.5) supt|α| : α P Au “ ω1 .
We then get
” ` ˘ı
(4.6) α P WO ÐÑ Dβ pβ P Aq ^ β R WO _ pα P WO ^ |α| ď |β|q .
Indeed if α P WO, there is by (4.5) a β P A with |α| ď |β|, and the right-hand side of
(4.6) holds. Conversely if the right-hand side of (4.6) holds for some α and some β,
we get β P WO as A Ď WO, hence necessarily α P WO (and |α| ď |β|). But this gives
a Σ11 definition of WO, contradicting Corollary 4.3.2.
The proof of (ii) is quite similar: Suppose A Ď WO is Σ11 , and supt|α| : α P Au ě
CK
ω1 . We want to argue that this contradicts the Representation Theorem 4.3.1. To
do so, pick a Π11 subset C of ω, and by Theorem 4.3.1 let f be a recursive : ω Ñ ω ω
with
(4.7) f pnq P WO ÐÑ n P C.
Now note that for every n f pnq is a recursive element of ω ω , hence if f pnq P WO,
|f pnq| ă ω1CK . So we get
which gives a Σ11 definition of C. As there is in ω a Π11 non Σ11 set C by Theorem 3.3.7,
we get a contradiction and (ii) is proved. %
92
4.4 Boundedness theorems
Remark 4.4.5. In this corollary, the fact that X is Suslin recursive and A is Σ11 (resp.
Σ11 ) in X are essential. For example, if one takes X “ WO, and A “ tα P X :
|α| is successoru we get, as
αPA ÐÑ α P X ^ pDn P Dα @m P Dα m ďα nq
Corollary 4.4.7.
(i) Suslin’s separation Theorem: If X is a Suslin space, A, B are two disjoint Σ11
sets in X, there is a Borel C in X which separates A from B, i.e. A Ď C and
C X B “ H.
93
4 Structural properties of Σ11 and Π11
Proof. Both (i) and (ii) are in fact ways to rephrase Suslin’s Theorem. For (i), apply
the theorem to the (Suslin) space A Y B, and the ∆11 subset A of it.
For (ii), recall that by Proposition 3.5.16 (ii) and Theorem 3.5.14 a subset A of X is
a B-set iff A is the trace on X of a ∆11 set in some completion of X — where Suslin’s
Theorem applies. %
The function ht : Z Ñ On is onto the ordinal htpRq. Moreover if one defines Rz , for
z P Z, as the restriction of R to ty P Z : y R zu, one gets
htpz, Rq “ htpRz q
and htpy, Rz q “ htpy, Rq for every y in the domain of Rz .
Suppose (4.13) is known for all t with hptq ă ξ, and let s P Z be such that hpsq “ ξ. If
t is a strict extension of s in Z and η “ hptq, then t P Zη zZη`1 , hence s P dpZη q “ Zη`1
and η ă ξ. Conversely if η ă ξ, then s P Zη`1 , and for some strict extension t of s in
Z, t P Zη . So htptq ě η. If htptq “ η we are done. If htptq ą η, then by the induction
94
4.4 Boundedness theorems
(ii) if W is a good universal set for Π11 subsets of X and ϕ is a Π11 -rank on W , there
is a recursive ε0 such that
DpRq “ Wε0
and for x, y in DpRq, y R x implies ϕpε0 , yq ă ϕpε0 , xq.
Proof. Clearly it is enough to prove (ii). Define a set Q Ď ω ω ˆ X by
” ` ˘ı
Qpα, xq ÐÑ @y P X y R x Ñ pα, yq P W ^ ppα, xq R W _ ϕpα, yq ă ϕpα, xqq
Suppose this fails for some x P DpRq, and pick x R-minimal for which it fails
(using that R is wellfounded on DpRq).
First suppose x R Wε0 . So x R Qε0 , hence there is a y such that y R x and
y R Wε0 . But y P DpRq, contradicting the R-minimality of x.
So x P Wε0 , hence x P Qε0 and x is not a counterexample. This contradiction
proves (b).
And (a) and (b) together give (ii). %
95
4 Structural properties of Σ11 and Π11
we easily get that DpRq “ WO, and htpα, RDpRq q “ |α|. The next result shows that
this cannot happen if R itself is wellfounded.
Theorem 4.4.9 (the Kunen-Martin Theorem for Σ11 wellfounded relations). Let X
be a Suslin recursive space. If R is a Σ11 (resp. Σ11 ) wellfounded relation on X, then
htpRq ă ω1CK (resp. htpRq ă ω1 ).
Proof. By the assumption, DpRq “ X. Let f : ω ω ˆ X Ñ ω ω be ∆02 -recursive with
where W is a good universal set for Π11 subsets of X. If R is Σ11 , we get by Theorem 4.4.8
a recursive ε0 with
But then by an immediate induction htpx, Rq ď |f pε0 , xq| for all x. Finally, Wε0 is
Σ11 in X ˆ ω ω and as X is Suslin, Corollary 4.4.4 applies and supx |f pε0 , xq| ă ω1CK .
The proof for R P Σ11 is similar, using the relativized version of Theorem 4.4.8 and of
Corollary 4.4.4. %
(ii) If moreover P is Π11 , ϕ is a regular Π11 -rank and A Ď P is Σ11 , then suptϕpxq :
x P Au ă ω1CK .
Proof. For (i) we may assume that ϕ is regular — for if ϕ1 is the unique regular rank
with ϕpxq ă ϕpyq Ø ϕ1 pxq ă ϕ1 pyq, then the wellorderings tϕ1 pxq : x P Au and
tϕpxq : x P Au are isomorphic and hence have the same cardinality.
Let B “ tx P X : Dy P A px P P ^ ϕpxq ă ϕpyqqu. Clearly B is Σ11 , B Ď P and
supB ϕ “ supA ϕ. So it is enough to show that supB ϕ ă ω1 .
Let
Rpx, yq ÐÑ x P B ^ y P B ^ ϕpxq ă ϕpyq.
Clearly R is Σ11 , and wellfounded. So by Kunen-Martin’s Theorem 4.4.9 htpRq ă
ω1 . But by definition of ht, as B is an initial segment of P for the prewellordering
96
4.5 Π11 monotone inductions and Σ11 derivations
97
4 Structural properties of Σ11 and Π11
Lemma 4.5.1. Let Φ be an operator : PpXq Ñ PpY q. Then Φ is Π11 on Π11 if and
only if for every recursive space Z and any Π11 subset A of Z ˆ Y , the set ΦZ pAq “
tpz, yq : y P ΦpAz qu is Π11 in Z ˆ Y . (and similarly for Π11 on Π11 operators).
In particular, if Φ is Π11 on Π11 , so are all operators ΦZ , for Z a recursive space.
Proof. Clearly tpα, yq : y P ΦpWα qu “ Φωω pW q, where W is a good universal Π11 set
for Π11 subsets of X, so one direction is obvious. Conversely suppose Φ is Π11 on Π11 ,
and let A Ď Z ˆ X be Π11 . Let iZ : Z Ñ ω ω be the canonical ∆02 embedding of Z into
ω ω , with recursive inverse jZ , and set Zr “ ipZq Ď ω ω . Define
pα, xq P C ÐÑ pf pαq, xq P W.
We then get
Theorem 4.5.2. Let Φ be Π11 on Π11 : PpXq Ñ PpY q. Then Φ is monotone on Π11
sets.
Proof. Let A, B be two Π11 sets with A Ď B, and fix y P ΦpAq towards showing
y P ΦpBq.
Consider C Ď ω ω ˆ X defined by
pα, xq P C ÐÑ x P A _ px P B ^ y P ΦpWα qq
where W is the good universal set for Π11 subsets of X. Clearly C is Π11 , hence
by Kleene’s Fixed Point Theorem 3.3.9 there is α0 P ω ω with Cα0 “ Wα0 . Now if
y R ΦpWα0 q then by definition of C, Cα0 “ A; and this contradicts the assumption
that y P ΦpAq. So y P ΦpWα0 q, and by definition of C, Cα0 “ B. So y P ΦpBq and we
are done. %
98
4.5 Π11 monotone inductions and Σ11 derivations
4.5.2 Inductions
Let Φ be a monotone inductive operator on X. We define inductively for every ordinal
ξ an operator Φξ on X by
Φ0 pAq “A
Φ1 pAq “ΦpAq
and for ξ ą 1,
´ď ¯
Φξ pAq “Φ Φη pAq .
ηăξ
We also set
ď
Φăξ pAq “ Φη pAq
ηăξ
ď
and Φ 8
“ Φξ pAq.
ξ
One immediately checks by induction that all operators Φξ are monotone inductive,
and that for each A the transfinite sequence pΦξ pAqqξ is increasing, so that it must
stabilize at some ordinal. The least such ordinal ξ0 pAq is called the closure ordinal
of Φ on A. We have that Φ8 pAq “ Φξ0 pAq pAq is the least fixed point of Φ containing
A: For clearly ΦpΦ8 pAqq “ Φξ0 pAq`1 pAq “ Φξ0 pAq pAq “ Φ8 pAq, so Φ8 pAq is a fixed
point of Φ. And if B is any fixed point of Φ with A Ď B, then easily by monotonicity
Φξ pAq Ď B for all ξ, hence Φ8 pAq Ď B.
(iii) if A is Π11 (resp. Π11 ) in X, B is Σ11 (resp. Σ11 ) and B Ď Φ8 pAq, then there is
ξ ă ω1CK (resp. ξ ă ω1 ) with B Ď Φξ pAq.
Proof. In order to prove (i) and (ii), we will establish for any Π11 subset A of X that
Φăω1 pAq is Π11 and that Φω1 pAq “ Φăω1 pAq. This is enough, by Lemma 4.5.1, to show
that Φω1 “ Φ8 is Π11 on Π11 , applying the fact above to the operators ΦZ and noting
that pΦZ qξ “ ΦξZ for all ξ.
Fix a Π11 set A Ď X, and a regular Π11 rank ϕ on the good universal set W Ď ω ω ˆX.
99
4 Structural properties of Σ11 and Π11
´ ` ˘¯
Set pα, x, yq P B Ø py P Aq _ pα, yq P W ^ pα, xq R W _ ϕpα, yq ă ϕpα, xq . The
L:ϕpα, xq ă ϕpα
set B is Π11 in ω ω ˆ X ˆ X, hence by Lemma 4.5.1, so is C defined by R
pα, x, yq P C ÐÑ y P ΦpBα,x q
Finally by Kleene’s Fixed Point Theorem 3.3.9 pick a recursive α0 such that Dα0 “
Wα0 . We claim that
From (a) we get Φ8 pAq Ď Wα0 , and from (b) Wα0 Ď Φăω1 pAq so that
is Π11 , as desired.
x P ΦpAq Ď ΦpBα0 ,x q,
Proof of (b). Suppose first x P Wα0 and ϕpα0 , xq “ 0. Then Bα0 ,x “ A, and x P
ΦpBα0 ,x q “ ΦpAq. So (b) holds for ξ “ 0. Now by induction let ξ ą 0, ξ ă ω1 ,
Ť suppose (b) holds for η ă ξ. Let x P Wα0 with ϕpα0 , xq “ ξ. Then Bα0 ,x “
and
ηăξ ty P Wα0 : ϕpα0 , yq ď ηu, hence by the induction hypothesis
ď
Bα0 ,x Ď Φ1`η pAq
ηăξ
`Ť ˘
1`η pAq “ Φ1`ξ pAq.
and x P ΦpBα0 ,x q Ď Φ ηăξ Φ
It remains to prove (iii). Let us do the lightface case, the other being obtained by
relativization. Let A be as before and let B be Σ11 , B Ď Φ8 pAq. Then tα0 u ˆ B Ď W ,
and is Σ11 in the Suslin recursive space X. By the Boundedness Theorem 4.4.10, ϕ is
bounded on tα0 u ˆ B by a recursive ordinal ξ. And by (b) above, B Ď Φ1`ξ pAq. %
100
4.5 Π11 monotone inductions and Σ11 derivations
One also often consider the dual notion of a Σ11 on Σ11 derivation, so it is worth
putting down the corresponding dual result.
A derivation D on a recursive space X is Σ11 on Σ11 if the relation y P DpWα q is Σ11 ,
where W is now a good universal set for Σ11 subsets of X.
Clearly D is a Σ11 on Σ11 derivation iff its dual Dˇ is a Π11 on Π11 monotone inductive
operator.
One
Ş then defines a sequence Dξ , ξ an ordinal, 0
Ş ξ by D pAq “ A and D pAq “
ξ
η 8
Dp ηăξ D pAqq for ξ ą 0, and set D pAq “ ξ D pAq.
Clearly D8 pAq is the largest fixed point of D contained in A. And Theorem 4.5.3,
applied to Dˇ, now states
(iii) if A is Σ11 (resp. Σ11 ) in X, B is Π11 (resp. Π11 ) and D8 pAq Ď B, then there
exists ξ ă ω1CK (resp. ξ ă ω1 ) with Dξ pAq Ď B.
is Π11 in ω ω ˆ X.
101
4 Structural properties of Σ11 and Π11
as desired.
We now mix Ψ0 and Φ into an operator Ψ1 on ωˆω ω ˆX, defined for B Ď ωˆω ω ˆX
by $
’
’
’ either pn, α, xq P B,
’
&or n “ 0 and α P Ψ0 pB0,x q,
pn, α, xq P Ψ1 pBq ÐÑ
’
’
’ or n “ 1 and α P Ψ0 pB0,x q
’ `Ť ˘
% and x P Φ p B1,pαqp .
An easy computation shows that Ψ1 is Π11 on Π11 . For A a Π11 subset of X, let
(
A1 “ pn, α, xq : pn “ 0 and α P A0 q _ pn “ 1 ^ α P A0 ^ x P Aq .
A1 is Π11 in ω ˆ ω ω ˆ X, so the proof is finished if we can prove that
(a) p0, α, xq P Ψ8
1 pA1 q Ø α P WO
and
´ď ¯ ˆ ď´ ď ¯ ˙
p1, α, xq P Ψ1 Ψη1 pAq ÐÑ α P WOξ ^ x P Φ Ψη1 pAq
1,α|p
ηăξ p ηăξ
´ď ¯
ÐÑ α P WOξ ^ x P Φ Φη pAq
ηăξ
as desired. %
Pb with index, end of
first line. – R
102
4.6 Reflection theorems
Theorem 4.6.1 (The First Reflection Theorem). Let G be a Π11 in the codes family of
Π11 sets in X.
(iii) In fact (ii) holds uniformly, i.e. there are recursive functions ϕ0 and ϕ1 : ω Ñ ω
such that if Wn P G then Wϕ0 pnq “ XzWϕ1 pnq is a ∆11 subset of Wn which is in
G.
pn, xq P W ÐÑ fW pn, xq P WO
again E is Π11 . Let W 1 be the good universal set for Π11 subsets of ω ˆ X. There
is a recursive f : ω Ñ ω such that pm, n, xq P E Ø pm, xq P Wf1 pnq . Now the
103
4 Structural properties of Σ11 and Π11
by the uniform version of The Fixed Point Theorem 3.3.9 there is a recursive
function ψ : ω Ñ ω such that if A “ Wk1 then ψpkq satisfies Aψpkq “ Wψpkq . Now
ϕ0 “ ψ ˝ f is such that for all n,
pϕ0 pnq, n, xq P E ÐÑ pψ ˝ f pnq, xq P Wf1 pnq ÐÑ x P Wϕ0 pnq .
By the proof in (ii), if Wn P G the set Wϕ0 pnq is a ∆11 subset of Wn which is in G.
Now if we set
Remarks 4.6.2. (1) The proof of (i) above is very similar to the proof of monotonicity
of Π11 on Π11 operators (Theorem 4.5.2). This is more than an analogy: One can
also define, for a family G of Π11 sets, what it means to be Π11 in the codes (that
the relation Wα P G is Π11 , for a good universal W for Π11 subsets of X). And
this notion is a particular case of Π11 on Π11 operators, by taking Y “ t0u and
Φ : PpXq Ñ PpY q defined by
#
H if A R G
ΦpAq “
t0u if A P G.
(2) Reflection results are sometimes called separation results. This because, at least
for Suslin recursive X, the Separation Theorem 4.3.8 is a particular case of reflec-
tion. To see this, let A, B be two disjoint Σ11 sets in X, and let G “ tC : A Ď Cu.
G is Π11 in the codes as
` ˘
Wn P G ÐÑ @x x P A Ñ pn, xq P W
[here we use that Π11 is closed under @X , i.e. that X is Suslin recursive]. Now by
hypothesis B is a Π11 set in G, so there is by reflection a ∆11 set C with C Ď B
and C P G, i.e. a ∆11 set C with A Ď C and C X B “ H, as desired.
104
4.6 Reflection theorems
Theorem 4.6.3 (The Second Reflection Theorem). Let X be Suslin recursive and let
G Ď PpXq2 be monotonic upwards, continuous downwards and Π11 on Π11 . Let A Ď X
be Π11 with pA, Aq P G. Then there exists a ∆11 subset B of A with pB, Bq P G.
Proof. Let W be a good universal set for Π11 subsets of X. We claim that there are
recursive functions ψ0 and ψ1 : ω 2 Ñ ω such that if pn, mq satisfies
Wn “ XzWm and Wn Ď A
then
To see this, let Gm “ tB : pB, Wm q P Gu. Each Gm is Π11 on Π11 , in fact uniformly in m
and if A Ď Wm , then A P Gm by monotonicity of G in the second variable. So by (a
uniform version of) Theorem 4.6.1, we can find recursive ϕ0 : ω Ñ ω and ϕ1 : ω Ñ ω
such that
Wϕ0 pmq “ XzWϕ1 pmq , Wϕ0 pmq Ď A and Wϕ0 pmq P Gm .
Let (as W is a good universal set for Π11 sets) ψ0 and ψ1 be chosen so that
Then if Wm “ Wn and Wn Ď A, we get by the preceding facts Wψ0 pn,mq “ Wψ1 pn,mq ,
Wn Ď Wψ0 pn,mq Ď A and pWψ0 pn,mq , Wn q P G by monotonicity, as desired.
Pick n0 , m0 with Wn0 “ H, Wm0 “ X, and define by induction
$
’
& gp0q “ n0 , hp0q “ m0
gpn ` 1q “ ψ0 pgpnq, hpnqq
’
hpn ` 1q “ ψ1 pgpnq, hpnqq.
%
105
4 Structural properties of Σ11 and Π11
106
5 Uniformizations
5.1 ∆11 points and uniformization by ∆11 -recursive functions
5.1.1 Coding of ∆11 and B-sets
Let X be a recursive space. The easiest way to encode ∆11 (and ∆11 ) subsets of X
is to use a good universal set W Ď ω ˆ X for Π11 (resp. W Ď ω ω ˆ X for Π11 )
added – YP
subsets of X, and to say that a pair of integers pn0 , n1 q codes a ∆11 set A Ď X if
Wn0 “ A and Wn1 “ XzA (and similarly for ∆11 sets). This is the coding we used
in Theorem 4.6.3. Note that this coding is good in a rather strong sense: If A Ď ω
is Π11 , and B Ď A ˆ X is ∆11 in A ˆ X (so in particular if it is ∆11 in ω ˆ X, but
the hypothesis is weaker), then there is a recursive function f : ω Ñ ω such that for
each n P A, ppf pnqq0 , pf pnqq1 q is a code of the ∆11 set Bn . To see this just note that
n P A ^ pn, xq P B and n P A ^ pn, xq R B are both Π11 in ω ˆ X by the hypothesis, and
use the fact that W is good. A similar remark can be made for the coding of ∆11 sets.
On the other hand, the set of codes, i.e. the set of pairs pn0 , n1 q with Wn0 “ XzWn1 ,
is complicated. And one can get a better coding if one considers only Suslin recursive
spaces.
Proposition 5.1.1. Let X be Suslin recursive. There is a coding pD1 , W1 q of ∆11
subsets of X which satisfies
(i) D1 is Π11 in ω, W1 is ∆11 in D1 ˆ X,
(ii) tWn1 : n P D1 u is the set of ∆11 subsets of X,
(iii) if A Ď ω is Π11 and B Ď A ˆ X is ∆11 in A ˆ X, there is a recursive function
f : ω Ñ ω such that for all n P A, f pnq P D1 and Wf1 pnq “ Bn .
107
5 Uniformizations
D1 “ n P ω : @x P X pn, xq P U 0 _ pn, xq P U 1 .
` ˘(
W1 “ tpn, xq : n P D1 ^ pn, xq P U 0 u.
Clearly W1 is ∆11 in D1 ˆ X, as
n P D1 ^ pn, xq R W1 ÐÑ n P D1 ^ pn, xq P U 1
by definition of D1 .
As (ii) is a particular case of (iii), suppose A Ď ω is Π11 and B is ∆11 in A ˆ X. Then
B “ tpn, xq P ωˆX : n P A^pn, xq P Bu and B 1 “ tpn, xq P ωˆX : n P A^pn, xq R Bu
0
so that Vf0pnq X Vf1pnq “ H and Vf0pnq Y Vf1pnq “ X. But this implies that Uf0pnq “ Vf0pnq
and Uf1pnq “ Vf1pnq , so that Uf0pnq Y Uf1pnq “ X, i.e. f pnq P D1 , and moreover
as desired. %
108
5.1 ∆11 points
Σ11 pxq, Π11 pxq as respectively ∆11 pαx q, Σ11 pαx q, Π11 pαx q where αx “ iX pxq P ω ω is the
characteristic function of tn : x P VnX u.
So in particular, a point y P Y is ∆11 pxq-recursive if it is ∆11 pαx q-recursive.
One could define also the relativized classes Γpxq, for x P X, for all classes Γ. However
this notion is much less interesting, unless X is a dim0 recursive space, for the classes
Σ0n and Π0n . The reason is that the transfer function iX : X Ñ ω ω is ∆02 -recursive. This
does not matter for the classes ∆11 , Σ11 and Π11 , which are preserved, so that e.g. one
has trivially that a set is Σ11 pxq iff it is the section at x of a set in Σ11 : but it matters
for the low level Kleene classes, which are not preserved.
We will often say that a point x is in Γ instead of saying that x is Γ-recursive. But
the terminology Γ-recursive stresses the analogy between this notion and the notion
of Γ-recursive function. In fact these two notions are closely related, as we will see in
maladroit? – YP
a moment. Of course we did not introduce a boldface analog, because the boldface
notions trivialize for the space ω, and so does the notion of Γ-recursive point.
Proposition 5.1.2.
(i) A point α P ω ω is ∆11 -recursive iff α : ω Ñ ω is a ∆11 -recursive function. In
particular, α P 2ω is ∆11 -recursive iff tn : αpnq “ 1u P ∆11 .
(ii) A point x P X is ∆11 -recursive iff for every recursive Y the constant function
: Y Ñ X with value x is ∆11 -recursive.
(iii) A point x P X is ∆11 -recursive iff it is Σ11 -recursive iff it is Π11 -recursive.
109
5 Uniformizations
Y,X
Moreover if X is co-Suslin recursive, DX
1 and D1 can be chosen Π11 in ω and Y ˆ ω,
respectively.
Proof. We claim that it is enough to prove the result for X “ Y “ ω ω (and Π11 sets
D1 and D1 ). For if X and Y are subspaces of ω ω , one just restrict the functions to
Y,X
DX1 “ tn : d1 pnq P Xu and D1 “ tpy, nq : y P Y ^ d1 py, nq P Xu, respectively (which
1
are still Π1 if X is co-Suslin recursive). And finally for arbitrary X and Y , we just
compose with the transfer functions iX , iY and their inverses. Using Proposition 5.1.2,
one immediately gets the result.
Let us do (i) for X “ ω ω . Note that by Proposition 5.1.2 α P ω ω is a ∆11 point iff the
graph Gf is ∆11 in ω 2 . So let pD1 , W1 q be the good coding of ∆11 subsets of ω 2 given
by Proposition 5.1.1, and set
and for n P D1 , let d1 pnq be that function with graph Wn1 . Clearly td1 pnq : n P D1 u is
the set of ∆11 points in ω ω . It remains to show that D1 is Π11 . But D1 is Π11 , and
As W1 is ∆11 in D1 ˆ ω 2 , D1 is Π11 .
The proof for (ii) is entirely similar, starting with D1 and W1 , and we skip it. %
110
5.1 ∆11 points
Theorem 5.1.4. Suppose Y is a co-Suslin recursive space. Then Π11 is closed under
D∆ pY q, i.e. if X is recursive and A Ď X ˆ Y is Π11 , then D∆ A is Π11 in X.
which gives a Π11 definition of D∆ A, as Π11 is closed under substitution by ∆11 -recursive
functions (Proposition 3.4.5). %
(iii) X is Π11 -absolute, i.e. is Π11 in any recursive space in which it is embedded,
111
5 Uniformizations
Proof. (ii) is easy, as if f : B Ñ Y is ∆11 recursive, and x P B, then f pxq is a ∆11 pxq-
recursive point of Ax , hence x P D∆ A.
So we prove (i). We use again the coding DX,Y1 “ D1 and dX,Y
1 “ d1 : X ˆ ω Ñ Y
of Theorem 5.1.3, and set
By the properties of the coding and since Y is co-Suslin recursive, A˚ is Π11 . And
ajoutÈ – YP
x P D∆ A Ø x P D0 A˚ . Now by the ω-Uniformization Theorem 4.3.5 there is a ∆11 -
recursive function g : D0 A˚ Ñ ω such that for all x P D0 A˚ px, gpxqq P A˚ . And if we
set for x P D∆ A “ D0 A˚
f pxq “ d1 px, gpxqq,
f is a ∆11 -recursive function which clearly works. %
(i) A point x P X is ∆11 -recursive iff txu is Σ11 in X iff txu is ∆11 in X.
@n x P VnX Ñ y P VnX
` ˘
y P txu ÐÑ
112
5.1 ∆11 points
(ii) To prove that f pXq is Luzin, it is clearly enough to prove that f pXq is ∆11 in Y
(as Y is arbitrary). Clearly f pXq is Σ11 , by closure of Σ11 under DX .
Fix now y P f pXq. The set tx P X : f pxq “ yu is the section at y of the
∆11 set Graphpf q in Y ˆ X, hence is ∆11 pyq. But as f is one-to-one, this set is
a singleton. So applying (the relativized version of) part (i), we get that the
unique x for which y “ f pxq is ∆11 pyq.
But then we get y P f pXq Ø Dx P ∆11 pyq y “ f pxq hence f pXq is Π11 , by closure
of Π11 under D∆ (Theorem 5.1.4). So f pXq is ∆11 in Y .
And to prove that f ´1 : f pXq Ñ X is ∆11 -recursive, we can apply Corollary 5.1.7:
Graphpf ´1 q is ∆11 , and as we saw above, f ´1 pyq is ∆11 pyq for all y P f pXq, so
Corollary 5.1.7 applies and f ´1 is ∆11 -recursive. %
Proposition 5.1.10.
ω
(i) The set tα P ω ω : α is ∆11 -recursiveu is a Π11 non ∆11 subset of ω ω , as is D1 “ Dω1
in ω.
(ii) The set tpα, βq : α is ∆11 pβq-recursiveu is a Π11 non ∆11 set in ω ω ˆ ω ω , as is
ω ω
D1 “ D1ω ,ω in ω ω ˆ ω.
113
5 Uniformizations
so the set A of ∆11 points is Π11 in ω ω . To see that D1 is not ∆11 , we argue by
contradiction: Define a real α0 by
#
0 if n R D1
α0 pnq “
d1 pnqpnq ` 1 if n P D1 .
(ii) is similar. We already know that D1 is Π11 , hence so is B “ tpα, βq : α P ∆11 pβqu,
for
pα, βq P B ÐÑ Dn ppβ, nq P D1 ^ d1 pβ, nq “ αq .
If D1 were Borel, so would be B by this equivalence. And B is not Borel, for
otherwise it would be ∆11 pβ0 q for some β0 P ω ω , hence tα : α P ∆11 pβ0 qu would be
∆11 pβ0 q too, contradicting the relativized version of part (i). %
(ii) There exists a Π01 subset of ω ω ˆ ω ω , with first projection all of ω ω , which cannot
be uniformized by any Borel function f : ω ω Ñ ω ω .
Proof. (i) Pick any Σ11 subset of ω which is not ∆11 (exists by Theorem 3.3.3 (ii)),
ajoutÈ – YP
and let B be Π01 in ω ˆ ω ω with A “ D1 B. If for each n P A the Π01 set Bn has a
∆11 member, we get A “ D∆ B, hence by Theorem 5.1.4 A is Π11 , a contradiction.
So one of the Bn ’s, n P A, is a non empty Π01 subset of ω ω with no ∆11 member.
114
5.1 ∆11 points
Π11 “ D∆ Π01
i.e. for any recursive space X and any Π11 subset A of X, there is a Π01 subset B of
X ˆ ω ω such that
x P A ÐÑ Dβ P ∆11 pxq px, βq P B.
Proof. We may as well assume that X is Polish recursive (considering traces if neces-
sary).
First we prove
Define
“ ‰
βPB ÐÑ @n A˚ ppβq0 , pβq1 pnq, nq ^ @i ă pβq1 pnq A˚ ppβq0 , i, nq .
pα, β, γq P S ÐÑ α P LO ^ β P LO ^ @n P Dα γpnq P Dβ
^ γ : Dα Ñ Dβ is pďα , ďβ q strictly increasing and
onto a (non necessarily proper) initial segment of ďβ .
@n P Dα @m P Dα n ďα m Ñ γpnq ďβ γpmq
` ˘
and @n P Dα @m P Dβ m ďβ γpnq Ñ Dp P Dα m “ γppq .
We claim that R works. For suppose α P LO and β P WO. If |α| ď |β|, there
is clearly a unique γ such that Spα, β, γq holds, hence a unique xγ, δy such that
Rpα, β, xγ, δyq holds. Conversely, if Rpα, β, xγ, δyq, holds then Spα, β, γq holds,
but this implies pDα , ďα q is a wellordering, i.e. α P WO, and |α| ď |β|, as desired.
From facts 1 and 2, we now prove the theorem. Let X be Polish recursive, and
A Ď X be Π11 . By the Representation Theorem 4.3.1, there exists a ∆02 -recursive
f : X Ñ ω ω such that @x f pxq P LO and
As the relation α P ∆11 pxq is Π11 in X ˆω ω by Theorem 5.1.3, there is also a ∆02 -recursive
ajoutÈ – YP
g : X ˆ ω ω Ñ ω ω with
Define
px, αq P B ÐÑ pf pxq, gpx, pαq0 , pαq1 qq P R
where R is the Π01 relation of fact 2. We claim that
116
5.1 ∆11 points
Suppose first x P A. Then f pxq P WO, and by the relativized version of Theo-
rem 4.4.3 (ii), as tf pxqu is ∆11 pxq, |f pxq| ă ω1CK´x “ suptξ : ξ is recursive in xu.
Now note that ω1CK´x “ supt|gpx, αq| : α P ∆11 pxqu, for otherwise tα : α P ∆11 pxqu
would be ∆11 pxq, contradicting Proposition 5.1.10. So there is an α0 P ∆11 pxq such that
utilise-t-il la
rÈciproque de |f pxq| ď |gpx, α0 q|. Then by the properties of R, there is an α1 P ∆11 pxf pxq, gpx, α0 yq,
rem 4.4.3, mais hence a fortiori α1 P ∆11 pxq, such that Rpf pxq, gpx, α0 q, α1 q holds. The pair xα0 , α1 y is
on ne l’a pas ∆11 pxq, and px, xα0 , α1 yq P B, as desired.
ntrÈe..., a-t-on: Conversely, suppose that for some α P ∆11 pxq px, αq P B, i.e.
ery ξ ă ω1CK´α
that
“ tβ | |β| ď ξu Rpf pxq, gpx, pαq0 q, pαq1 q
is ∆11 pαq? by
ction? ce serait holds. As pαq0 P ∆11 pxq, gpx, pαq0 q is in WO. And f pxq P LO by choice of f . By the
relativisÈ de la property of R we know that f pxq P WO and |f pxq| ď |gpx, pαq0 q|, hence in particular
uve du Thm de x P A, as desired.
Suslin
em 4.4.6. – YP
This almost finishes the proof. However, as f, g are ∆02 -recursive, the set B defined
above is only Π02 in X ˆ ω ω . So we use again fact 1: There is a Π01 C Ď X ˆ ω ω ˆ ω ω
with
B “ D1 C “ D∆ C “ D!C
and if we set px, αq P D Ø px, pαq0 , pαq1 q P C, D is a Π01 set which clearly works. %
Theorem 5.1.13. Any Luzin-recursive space is the 1–1 recursive image of a Π01 subset
of ω ω . Similarly any Luzin space is the 1–1 continuous image of a closed subset of ω ω .
Proof. We prove the lightface version, the second one being obtained by relativization.
Again, without loss of generality, we may assume that the Luzin-recursive space A
is a ∆11 -subset of ω ω , by Theorem 3.5.14. Let f : ω ω Ñ ω ω be recursive, with α P A Ø
f pαq P WO, by Theorem 4.3.1. As f pAq is Σ11 in WO, there is by Theorem 4.4.3 a
recursive α0 P WO such that
αPA ÐÑ Dβ Rpα, α0 , βq
ÐÑ D!β Rpα, α0 , βq.
117
5 Uniformizations
(3) If α “ 2a β and for all n, pβqn P BC, where as usual pβqn pmq “ βpxn, myq, then
α P BC.
(1) If α P BC0 ,
#
H if αp1q “ 1,
Bα “
VnX if n is the unique integer with αpn ` 2q “ 1.
(2) If α “ 1a α` , Bα “ XzBα` .
(3) If α “ 2a α` , Bα “ n Bpα` qn .
Ť
One immediately checks that this indeed defines a function : BC Ñ BorelpXq. And
it is onto BorelpXq, as the set of coded sets contain the basic open sets, and is closed
under countable unions and complements (the only point to note is that by the form of
the coding, we only encode non empty unions, so that we have to encode H separately).
We will refer to the coding function π for X as π X , and to the set coded in X by
α P BC as BαX , if it is necessary to avoid confusion.
118
5.2 Coding Borel sets
R0 pα, xq ÐÑ α P BC ^ x P Bα
1
and R pα, xq ÐÑ α P BC ^ x R Bα
x P Bα ÐÑ R0 pα, xq ÐÑ R1 pα, xq
and using the three Π11 on Π11 operators Ψ, Φ0 and Φ1 defined by:
I changed the
0 1 a ` ` operators so that
pα, xq P ΨpS, T , T q Ø pα P Sq _ pα “ 1 α ^ α P Sq
they are inductive
_ pα “ 2a α` ^ @n pα` qn P Sq and the notations to
facilitate (?) the
pα, xq P Φ0 pS, T 0 , T 1 q Ø pα, xq P T 0 reading. – YP
_ pα “ 1a α` ^ α` P S ^ pα` , xq P T 1 q
_ pα “ 2a α` ^ @n pα` qn P S ^ Dn ppα` qn , xq P T 0 q
pα, xq P Φ1 pS, T 0 , T 1 q Ø pα, xq P T 1
_ pα “ 1a α` ^ α` P S ^ pα` , xq P T 0 q
_ pα “ 2a α` ^ @n pα` qn P S ^ @n ppα` qn , xq P T 1 q
119
5 Uniformizations
HYP “ ∆11 .
We will break the proof of this result in steps. Note that the second statement is an
easy consequence of Proposition 5.2.2, and the first statement of the theorem. For if
X is Suslin-recursive and B Ď X is ∆11 , we can apply the first statement to A0 “ B
and A1 “ XzB, which are Σ11 in X, so that B, which is the only set separating A0
from A1 , must be in HYP.
Cu,v,w,n,m,p,q “ tα P ω ω : ua n Ď αu,
Cu,v,w,n,m,p,q “ H,
120
5.2 Coding Borel sets
As S is wellfounded, the sets Cu,v,w,n,m,p,q are defined: For if they are not defined for
some pu, v, wq and n, m, p, q, we have pu, v, wq P S and n “ m by (1) and (2). Pick
then a least pu, v, wq in S (for reverse inclusion) for which Cu,v,w,n,m,p,q is not defined.
Then (3) defines it.
Define for u, v, w of the same length
Au,v ω
(
0 “ α P ω : u Ď α ^ Dβ pv Ď β ^ pα, βq P rT0 sq ,
and similarly Au,w “ α P ω ω : u Ď α ^ Dβ pw Ď β ^ pα, βq P rT1 sq ,
(
1
ď č
and Bu,v,w “ Cu,v,w,n,m,p,q
n,pPω m,qPω
(so that (3), for pua n, v a p, wa qq P S, can be rewritten Cu,v,w,n,m,p,q “ Bua n,va p,wa q ).
We claim that
a n,v a p a m,w a q
(i) for all u, v, w, n, m, p, q, Cu,v,w,n,m,p,q separates A0u from Au1 ,
(ii) for pu, v, wq P S, Bu,v,w separates Au,v u,w
0 from A1 .
a a a a
(i) is clear if n ‰ m, for then Au0 n,v p Ď tα : ua n Ď αu which is disjoint from A1v n,w q ,
which is contained in tα : ua m Ď αu.
a a
Suppose now n “ m, and pua n, v a p, wa qq R S. If pua n, wa qq R T1 , Au1 n,w q “ H,
a a
hence tα : ua n Ď αu separates A0u n,v p from it.
So it remains to prove (i) when pua n, v a p, wa qq P S, i.e. to prove that for all
pu, v, wq P S, (ii) holds. This done by induction on S: We may assume that (i) has
been proved for u, v, w and all n, m, p, q. Now note that
ď ď ua n,va p ď ď ua m,wa q
Au,v
0 “ A0 and Au,w1 “ A1 .
n p m q
a a
By the induction hypothesis, for all m, q Au0 n,v p Ď Cu,v,w,n,m,p,q hence
a a č
Au0 n,v p Ď Cu,v,w,n,m,p,q
m,q
and
ďč
Au,v
0 Ď Cu,v,w,n,m,p,q “ Bu,v,w .
n,p m,q
a m,w a q
Similarly for all n, p Au1 X Cu,v,w,n,m,p,q “ H, hence
č
Au,w
1 X Cu,v,w,n,m,p,q “ H
m,q
and hence
ďč
Au,w X Cu,v,w,n,m,p,q “ H.
n,p m,q
121
5 Uniformizations
Lemma 5.2.4.
(ii) There is a recursive ϕ2 : ω Ñ ω such that if W 3 is our good universal set for Σ01
subsets of ω 3 and n P ω is such that for all m Wn,m
3 is the graph of function αn,m
in BC, then ϕ2 pnq P bc and
ď
Btϕ2 pnqu “ Bαn,m .
m
122
5.2 Coding Borel sets
Bn “ Wϕ22 pnq .
3
Now if for all m, Wn,m is the graph of a function αm , then Bn is the graph of the
a
function 2 α, where αpxk, lyq “ αk plq. This easily shows that ϕ2 satisfies (ii).
(iii) is easy using ϕ1 and ϕ2 : By the S-m-n Theorem 1.4.4, pick first f1 recursive
this is S72 – YP
: ω 8 Ñ ω such that
and set
and set
Finally set ϕ3 “ ϕ2 ˝ f3 . The proof that ϕ3 works is immediate form the defini-
tions, and left to the reader. %
From this lemma, the proof of the Suslin-Kleene Theorem for ω ω is easy.
Recall that for u, n, p P ω the relation su a n “ sp is recursive (cf. Subsection 1.1.9).
rÈf ajoutÈ – YP
123
5 Uniformizations
Let T0 and T1 be recursive (in the codes) trees on ω ˆ ω. Following the definition of
Cu,v,w,n,m,p,q in Subsection 5.2.4 set, for pe, u, v, w, n, m, p, q, k, lq P ω 10
pe, u, v, w, n, m, p, q, k, lq P A ÐÑ
either (1) pn ‰ m _ psu a n, sw a qq R T1 q
^ Di su a n “ si ^ pk ‰ i ` 2 ^ l “ 0q _ pk “ i ` 2 ^ l “ 1q
` ˘
^ pϕ3 pe, i0 , i1 , i2 q, k, lq P W 2 .
‰
Ae0 “ We90 .
And by the definition of A and the choice of ϕ3 , it is a simple matter to show that
if D1 rT0 s X D1 rT1 s “ H then for each pu, v, w, m, n, p, qq the set Ae0 ,u,v,w,m,n,p,q is the
graph of a function αu,v,w,m,n,p,q P ω ω which is in BC, and moreover codes the Borel set
Csu ,sv ,sw ,m,n,p,q of Subsection 5.2.4, by induction on the wellfounded tree S associated
to T0 and T1 .
Then ϕ3 pe0 , u, v, wq P bc and codes a function αu,v,w P BC with Bαu,v,w “ Bsu ,sv ,sw
of Subsection 5.2.4, so that finally as s0 “ H, the number ϕ3 pe0 , 0, 0, 0q P bc and
codes a recursive function tϕ3 pe0 , 0, 0, 0qu in BC which codes a HYP set separating
A0 “ D1 rT0 s from A1 “ D1 rT1 s.
Note that the preceding construction is clearly uniform in T0 and T1 , i.e. there is
a recursive function f : ω ˆ ω Ñ ω such that if Wn2 and Wm 2 are, respectively, the
sets tpk, lq : psk , sl q P T0 u and tpk, lq : psk , sl q P T1 u, and D1 rT0 s X D1 rT1 s “ H, then
f pn, mq P bc and codes the corresponding separating HYP set. To see this, one just
uses the uniform version of the uniform version of the Recursion Theorem, namely
Theorem 1.4.6.
ajoutÈ – YP 124
5.2 Coding Borel sets
(ii) If f : X Ñ Y is ∆0k -recursive, then for any HYP set B Ď Y , f ´1 pBq is HYP
in X. In fact there is a recursive ψ : ω Ñ ω such that if n P bc, ψpnq P bc and
X
Btψpnqu Y q.
“ f ´1 pBtnu
Define
“
pn, k, p, qq P B ÐÑ either A˚ pn, pkq1 q ^ pp ‰ pkq0 ` 2 ^ q “ 0q
‰
_ pp “ pkq0 ` 2 ^ q “ 1q
“
or A˚ pn, pkq1 q ^ pp ‰ 1 ^ q “ 0q
‰
_ pp “ 1 ^ q “ 1q .
B is Σ01 in ω 4 , and for each pn, kq Bn,k is the graph of a function in BC0 which
codes in X the set Vpkq X if A˚ pn, pk qq holds, and H otherwise.
0
1
pn, xq P A ÐÑ Dkpxk, ny , xq P B.
ϕ1 “ ϕ2 ˝ g
where ϕ2 is the recursive function of Lemma 5.2.4 (ii). It should be clear that ϕ1
works.
125
5 Uniformizations
Applying part (i), we get a recursive ϕ : ω Ñ ω such that for all n, ϕpnq P bc
X
and Btϕpnqu “ f ´1 pVnY q.
Define pn, p, qq P B0 Ø pn, p ` 1, qq P W 2 , and pick g0 recursive with pn, p, qq P
B0 Ø pg0 pnq, p, qq P W 2 . Then clearly if n codes a function α, g0 pnq codes α` .
Similarly define pm, n, p, qq P B1 Ø pn, xm, py ` 1, qq P W 2 , and pick g1 recursive
with pg1 pm, nq, p, qq P W 2 Ø pm, n, p, qq P B1 . Clearly if n codes a function α,
then for all m, g1 pm, nq codes pα` qm .
Define now
pe, n, p, qq P A ÐÑ
(1) pn, 0, 0q P W 2 ^ pn, 1, 1q P W 2 ^
“ ‰
pp ‰ 1 ^ q “ 0q _ pp “ 1 ^ q “ 1q
or (2) pn, 0, 0q P W 2 ^
Dk pn, k ` 2, 1q P W 2 ^ pϕpkq, p, qq
“ ‰
”
or (3) pn, 0, 1q P W 2 ^ pp “ 0 ^ q “ 1q
ı
3
_ pp ą 0 ^ pe, g0 pnq, p ´ 1, qq P W q
”
or (4) pn, 0, 2q P W 2 ^ pp “ 0 ^ q “ 2q
ı
_ pp ą 0 ^ pe, g1 ppp ´ 1q0 , nq, pp ´ 1q1 , qq P W 3 q .
Ae0 “ We30 .
126
5.2 Coding Borel sets
(3) tnup0q “ 1. Then by choice of g0 , tnu codes Btg Y . And by the induction
0 pnqu
hypothesis Ae0 ,g0 pnq is the graph of a function α which codes f ´1 pBtnu Y q. But
as desired.
(4) is similar: If tnup0q “ 2, then by choice of g1
ď
Y Y
Btnu “ Btg 1 pm,nqu
m
and by induction hypothesis We30 ,g1 pm,nq “ Ae0 ,g1 pm,nq is the graph of a
` Y ˘
function αm which codes f ´1 Btg 1 pm,nqu
. But then by condition (4) above,
Ae0 ,n is the graph of the function α with αp0q “ 2 and pα` qm “ αm for all
m, i.e. is in BC and codes
ď ´ď ¯
Y
f ´1 pBtg1 pm,nqu q “ f ´1 Btg 1 pm,nqu
“ f ´1 pBtnu q
m m
as desired. %
5.2.7 B “ HYP
Corollary 5.2.6. In any recursive space X,
B “ HYP.
127
5 Uniformizations
BCΠ
0 “ BC0
´ď ¯
BCΣ
ξ “ Φ Σ
BCΠ
η
ηăξ
and BCΠ
ξ “ Φ pBCΣ
Π
ξ q.
BCΣ BCΠ
Ť Ť
Clearly BC “ ξăω1 ξ “ ξăω1 ξ , and by Theorem 4.5.5 the relations
β P WO ^ α P BCΣ
|β|
and β P WO ^ α P BCΠ
|β|
(ii) A Ď X is Σ0n (resp. Π0n ) iff it admits a recursive Borel code in BCΣ Π
n (resp. BCn ).
Proof. (i) is by induction on ξ. For ξ “ 0, the sets which are coded in BCΣ 1 are all
unions of basic sets VnX , including the empty set, i.e. all the open sets. Assuming
the result is known for BCΣ 1`η for all η ă ξ, we immediately get it for BC1`η
Π
Σ
(for η ă ξ), and then by definition of BC1`ξ , we get that a set admits a Borel
code in BCΣ 1`ξ iff it is a countable union of sets which admit a Borel code in
Ť Π
ηăξ BC1`η . This clearly gives the result.
(ii) is similar. First if A P Σ0k , we have shown that A is a HYP set in Lemma 5.2.5 (i),
by computing a recursive Borel code for it. Going back to that proof, one easily
checks that the code computed there is in fact in BCΣ k.
For the converse, it is enough to show for all k that if f : ω Ñ ω is recursive
with values in BCΣ X 0
k , the set tpn, xq : x P Bf pnq u is Σk in ω ˆ X. This is easy
for k “ 1. And assuming it for k, and letting f : ω Ñ BCΣ k`1 be recursive, the
function g : xm, ny ÞÑ ppf pnq qm q is recursive, and takes value in BCΣ
` `
k , so that
! )
X
B “ pm, n, xq : x P Bgpxm,nyq
is Σ0k , and as
x P BfXpnq ÐÑ Dm pm, n, xq R B,
we get the result. %
128
5.3 Complements on ω1CK and the ω1CK´α
additive (resp. multiplicative) Borel class ξ corresponds to Σ0ξ (resp Π0ξ ). This is the
notation we will use from now on to denote the Borel classes.
The operator ΦΣ and ΦΠ used to build the sets BCΣ Π
ξ and BCξ are not inductive — in
fact these sets, for ξ ă ω1 , are all disjoint. However the operator Φ of Subsection 5.2.1
used to build BC is inductive, and clearly ΦΣ pSq Ď ΦpSq and ΦΠ pSq Ď ΦpSq for all S.
trop rapide pour moi!
It follows, by Moschovakis’ Theorem 4.5.3, that if α P BCΣ ξ is recursive , then – YP
CK Σ 1
ξ ă ω1 . Now if α P BCξ is recursive for some ξ, then tαu is ∆1 and tαu Ď Φ8 pBC0 q.
I’d at least say: – YP
It follows by Moschovakis’ Theorem 4.5.3 that there is some η ă ω1CK with x P Φη pBC0 q.
Therefore ξ ď η and ξ ă ω1CK . – YP So the hierarchy pΣ0ξ , Π0ξ , ∆0ξ q is interesting only
for ξ ă ω1CK , and similarly pΣ0ξ pαq, Π0ξ pαq, ∆0ξ pαqq for ξ ă ω1CK´α , the first ordinal not
recursive in α.
The classes Σ0ξ , Π0ξ for ξ ă ω1CK , share many properties of the classes Σ0n , Π0n of
the arithmetical hierarchy, e.g the closure properties stated in Proposition 3.2.2, the
inclusion diagram of Theorem 3.2.3 and the existence of (good) universal sets.
Proofs are straightforward but rather tedious — as one usually needs heavy uses of
the recursion theorem, as we did e.g. in Lemma 5.2.5. We leave them as exercises (see
rÈf manquante – YP
??).
Proof. (i) If α P WO is ∆11 -recursive, then tαu is ∆11 in ω ω , hence by the Boundedness
Theorem 4.4.3 |α| ă ω1CK . This proves (i).
129
5 Uniformizations
5.3.2 Kleene’s O
Theorem 5.3.2.
(i) There is in ω ˆ ω a Π11 relation O such that
(a) O is a wellordering of dom O “ tn : pn, nq P Ou of length ω1CK ,
(b) O is ∆11 in dom O ˆ ω.
Define
` ˘(
A˚ “ n P A : @m ă n m R A _ |f pmq| ă |f pnq| _ |f pnq| ă |f pmq| .
n P A˚ ^ pm R A˚ _ |f pnq| ă |f pmq|q
` ˘
ÐÑ n P A˚ ^ m R A˚ _ p|f pnq| ď |f pmq| ^ n ‰ mq
130
5.3 Complements on ω1CK and the ω1CK´α
where pn, αq ÞÑ α|n is the recursive function of Proposition 4.2.3 (i). Clearly
ajoutÈ – YP
β P ∆11 , hence A˚ “ tm : Dn P Dα m “ βpnqu is ∆11 too. But this contradicts
ω1CK “ supt|f pnq| : n P A˚ u.
(ii) is similar, starting with a Π11 set A Ď ω ω ˆ ω such that for each α P ω ω , Aα is a
comment
Π11 pαq non Σ11 pαq subset of ω. We skip the details. % montre-t-on qu’un tel
ensemble existe? –
Note. A variant of the wellordering O is known in the literature as “Kleene’s O”. YP
Theorem 5.3.3.
Proof. (i) Let T be a recursive tree on ω such that A “ rT s is a non empty Π01 set
with no ∆11 member (such a tree exists by Theorem 5.1.11 (i) and the proof of
ajoutÈ,
Theorem 4.1.7). Define correspondance
fermÈs-arbres – YP
sn R T ^ sm R T ^ n ă m
Rpn, mq ÐÑ or sn P T ^ sm R T
or sn P T ^ sm P T ^ sn ďBK sm .
131
5 Uniformizations
where pn, αq ÞÑ α|n is the recursive function of Proposition 4.2.3 (i). This β is
ajoutÈ – YP
∆11 , so that A “ tm : Dn P Dα βpnq “ mu is also ∆11 . But then ωzDpRq is also
L: RzDpRq – YP
∆11 , and by definition of DpRq,
(ii) The proof of (ii) is entirely similar, starting from a recursive tree T on ω ˆ ω
such that for each α, rTα s, where Tα “ ts P ω ăω : pαlh s , sq P T u, is a non empty
ajoutÈ – YP
Π01 pαq set with no ∆11 pαq member, as given by Theorem 5.1.11 (ii). %
132
5.3 Complements on ω1CK and the ω1CK´α
Proof. We use the sets O and R given by Theorems 5.3.2 and 5.3.3. The claim is that
” `
ω1CK´α ď ω1CK´β ÐÑ Dγ @n P dom Oα @m P dom Oα
ppα, n, mq P O Ñ pβ, γpnq, γpmqq P Rq
˘
^ pn ‰ m Ñ γpnq ‰ γpmqq
^ @n P dom Oα @p pβ, p, γpnqq P R Ñ Dm
`
This will finish the proof, for the right hand side is clearly Σ11 (remember that the
j’ai un doute – YP
relation n P dom Oα ^ pm R dom Oα _ pα, n, mq P Oq is Π11 by Theorem 5.3.2).
Now the right hand side just says that there is a strictly increasing function from
pdom Oα , Oα q onto an initial segment of pω, Rβ q which by the properties of O and R
exactly means that ω1CK´α ď ω1CK´β . %
Proof. One direction is obvious: if Oα P ∆11 pβq, then its length ω1CK´α is a ∆11 pβq-
recursive ordinal, hence recursive in β by Proposition 5.3.1, and so ω1CK´α ă ω1CK´β .
ajoutÈ – YP
Conversely suppose α P ∆11 pβq and ω1CK´α ă ω1CK´β . Let f be recursive in α, such
that
pm, nq P Oα ÐÑ f pm, nq P WO
ÐÑ f pm, nq P WO ^ |f pm, nq| ă ω1CK´α .
As α P ∆11 pβq, f is ∆11 pβq-recursive, so the above equivalence gives a ∆11 pβq definition
of Oα . %
133
5 Uniformizations
Now Theorem 5.1.11 is an anti-basis result, which says that even for the simple case
of Π01 sets, the family of ∆11 -recursive points is not a basis. The next theorem gives the
best bases for the class Π01 — and hence for the class Σ11 . We will see the corresponding
uniformization result in the next Section.
Define #
1 if ppnq0 , pnq1 q P O,
βO pnq “
0 otherwise,
and similarly for Oα .
Theorem 5.3.7. Let X be a Suslin recursive space, and A Ď X a non empty Σ11 (resp.
Σ11 pαq) set.
(i) A contains a point x which is ∆11 pβO q (resp. ∆11 pβOα q).
(ii) (Gandy’s Basis Theorem) A contains a point x with ω1CK´x “ ω1CK (resp. ω1CK´x “
ω1CK´α ).
Proof. By transfer, it is clearly enough to prove it for X “ ω ω .
(i) We can assume that A is Π01 , as the projection of a ∆11 pβO q-point in ω ω ˆ ω ω is
∆11 pβO q.
Let T be a recursive tree with A “ rT s, and let α be the left-most branch of T ,
i.e. α is defined inductively by
´ ¯
αpnq “ least p Dβ αn a p Ď β ^ β P rT s .
so that f pnq P LO for all n, and f pnq P WO iff the tree Tsn is wellfounded, by
Proposition 4.2.5. By definition of the left-most branch, we get
ajoutÈ “code de” –
YP
α is the left-most branch ÐÑ @i @k ă αpiq f pxαi a kyq P WOωCK
`
@ D 1 ˘
^ f p αi`1 q R WOωCK .
1
134
5.4 Uniformization of Σ11 sets and Π11 sets
(1) The class BorpΣ11 q is defined, in any metrizable separable space, as the least σ-
algebra containing all Σ11 sets, i.e. the least family containing the Σ11 sets and
closed under countable unions and complements.
(2) A function f : X Ñ Y is BorpΣ11 q-measurable if for each open set U in Y the set
f ´1 pU q is in BorpΣ11 q.
5.4.2 C-sets
Clearly for f : X Ñ Y to be BorpΣ11 q-measurable, it is enough to satisfy f ´1 pVnY q P
BorpΣ11 q for all n. And it implies f ´1 pBq P BorpΣ11 q for any Borel set B in Y , as one
easily checks by induction on the Borel class of B. So in particular if f : X Ñ Y
is BorpΣ11 q-measurable and g : Y Ñ Z is Borel, the composition g ˝ f : X Ñ Z is
BorpΣ11 q-measurable.
On the other hand, if f is BorpΣ11 q-measurable, f ´1 pAq may not be in BorpΣ11 q for
A in BorpΣ11 q, even for A in Σ11 .
So we introduce a hierarchy pCξ qξăω1 of sets, called the hierarchy of C-sets of Seli-
navovski, by ˆ ´ď ¯˙
1
C0 “ Σ1 , and for ξ ě 1 Cξ “ A Cη ,
ηăξ
Ť
and we let C “ ξăω1 Cξ .
135
5 Uniformizations
(i) C is the least σ-algebra of subsets of X containing the VnX ’s and closed under
operation A.
Proof. (i) It is clear from the definition that any σ-algebra containing the open sets
and closed under A must contain all C-sets.
For the other direction, note that the family pCξ qξăω1 is increasing. This is
immediate by induction once we know that C0 Ď C1 . But C0 “ Σ11 “ ApΠ01 q,
and C1 “ ApΣ11 q, so this follows from the inclusion Π01 Ď Σ11 .
Suppose now pAs qsPωăω is a Suslin scheme of C-sets. Then by the discussion
above, the sets As are in Cξ for some ξ ă ω1 . Then As P Cξ`1 and finally
ApAs q P Cξ`2 is a C-set. So C is closed under A and . And as a countable
union is a particular case of A, C is also a σ-algebra. This proves (i).
(ii) We already noticed that Cξ Ď Cξ`1 . Also clearly Cξ Ď Cξ`1 . Now by Propo-
sition 4.1.1 A is idempotent, and hence Cξ`1 is closed under operation A. But
then, as countable union and countable intersection are particular cases of oper-
ation A, Cξ`1 is closed under these two operations. To prove (ii), it is enough to
check that BorpCξ q is the least family Φ containing Cξ Y Cξ and closed under
countable unions and intersections. But one immediately verifies that the family
of sets which are in Φ X Φ is closed under and countable union and contains
Cξ , so that Φ “ Φ X Φ “ BorpCξ q. This proves (ii).
The importance of the class C comes from the fact (that we will prove later) that
C-measurable functions are very regular with respect to measure and category, and
from the next uniformization theorem.
136
5.4 Uniformization of Σ11 sets and Π11 sets
py, αq P A ÐÑ Dβ py, α, βq P B.
psa n Ď αy ^ ta p Ď βy q ÐÑ ps Ď αy ^ t Ď βy q^
@m @q xm, qy ă xn, py Ñ psa m, ta qq R Ty
` ˘
^ psa n, ta pq P Ty .
But by induction hypothesis the first conjunct on the right hand side is in BorpΣ11 q,
and as ty : ps, tq P Ty u is Σ11 by definition of Ty , the second conjunct is Π11 and the
third Σ11 . This finishes the proof. %
Proof. Using the transfer result in Corollary 5.1.5, we may assume that X “ ω ω . rÈf modifiÈe – YP
Recall that we have defined in Proposition 4.2.3 (i) a recursive function pn, αq ÞÑ α|n
from ω ˆ ω ω into ω ω such that for α P LO, α|n P LO and codes the order ďα restricted
to the predecessors of n for ďα .
137
5 Uniformizations
Let A be a non empty Π11 set, and let f : ω ω Ñ LO be such that α P A Ø f pαq P WO.
Let (
A˚ “ α P A : |f pαq| “ inft|f pβq| : β P Au .
We define inductively sets An and A˚n by
(
A0 “ α P A˚ : αp0q “ inftβp0q : β P A˚ u
(
A˚0 “ α P A0 : |f pαq|0| “ inft|f pβq|0| : β P A0 u
and
(
An`1 “ α P A˚n : αpn ` 1q “ inftβpn ` 1q : β P A˚n u
(
A˚n`1 “ α P An`1 : |f pαq|n ` 1| “ inft|f pβq|n ` 1| : β P An`1 u
This shows that B is Π11 . And as any two points in An coincide on r0, ns, B contains
at most one point. So the main thing is to show that B ‰ H. To do this pick αn P A˚n .
Clearly as αn`1 n`1 “ αn n , the sequence αn converges to a point α P ω ω . We claim
that α P B — so that B “ tαu. To see this, let ξ be the value of |f pβq| for β P A˚ , ξn
the value of |f pβq|n| for β P A˚n , so that ξn ă ξ. Note that as f and gpβ, nq “ β |n are
recursive, hence continuous, one has
We claim that for n, m P Df pαq with n ăf pαq m, ξn ă ξm . To see this, note that for large
enough p ą n, m, n, m P Df pαp q , and n ăf pαp q m. But then |f pαp q|n| ă |f pαp q|m|, i.e.
ξn ă ξm as desired.
This proves that ďf pαq is a wellordering on Df pαq , i.e. f pαq P WO and moreover (as
n ÞÑ |f pαq|n| is the length function on Df pαq ) that for n P Df pαq , |f pαq|n| ď ξn and
finally |f pαq| ď ξ. But this immediately implies that α P B, as claimed. %
Note that this basis result, together with the antibasis result in Theorem 5.1.11,
gives the existence in ω ω of Π11 singletons which are not ∆11 -recursive points, and hence
furnishes the counter examples discussed in Subsection 5.1.8.
138
5.5 Natural Π11 sets, natural Π11 ranks
Proof. Using the transfer result in Corollary 5.1.5, we may assume that X “ Y “ ω ω . pas clair pour moi –
YP
The idea is to work on each section Aα the way we did in the proof of Theorem 5.4.4,
using a recursive function f : ω ω Ñ ω ω Ñ LO with pα, βq P A Ø f pα, βq P WO.
This clearly will define the graph of a uniformizing function, by the proof of Theo-
rem 5.4.4, and one just has to check that the corresponding sets pAα q˚n are such that
the relation
β P pAα q˚n is Π11 in α, β and n.
This is easy to check, and left to the reader. %
5.4.6 Limitations
The deep uniformization result of Kondo-Addison led to the notion of a scale, intro-
duced by Moschovakis (we won’t study this notion here, but it happens to be central in
understanding the structure of the more complicated classes of the projective hierarchy
(and beyond)). We refer the reader to [Mos80].
On the other hand, it cannot be proved from the usual axioms of set theory that
a function with Π11 graph has nice regularity properties, like measurability or Baire
property (see [Mos80]). This means that in order to make the Kondo-Addison theorem
useful in Analysis, one has to accept to work with strong set-theoretic axioms.
139
5 Uniformizations
(1) If W Ď ω ω ˆ ω ω is any universal set for Π11 subsets of ω ω , W is not Borel (as its
diagonal is not, by Theorem 3.3.3).
(2) The set tpα, βq : α P ∆11 pβqu is Π11 non Borel (by Proposition 5.1.10) — the proof
uses, uniformly, Cantor’s diagonal method for ω.
(3) The set D1 (cf. Proposition 5.1.1 and Proposition 5.1.10) and the set BC of Borel
codes (cf. Subsection 5.2.1) are not Borel (otherwise, they would give a Borel
set universal for Borel subsets of ω ω , which is clearly impossible by considering
again its diagonal).
The diagonal method is not flexible enough to be applied to most Π11 sets. We will
present in this section two other techniques for proving non-Borelness, the completeness
method, and the rank method. However, these two methods rely on the existence of Π11
non Borel sets, hence ultimately incorporate a diagonal argument.
140
5.5 Natural Π11 sets, natural Π11 ranks
5.5.3 @ universal
In Subsections 5.5.3 to 5.5.7, we will give applications of the completeness method
which are based on the possibility of nicely representing open sets of ω ω — or as it
turns out, Σ02 subsets of 2ω .
Proof. Pick some complete Π11 set in ω ω , e.g. by Corollary 4.3.2 the set WO, and let
A be Σ01 in ω ω ˆ ω ω with
α P WO ÐÑ @β pα, βq P A.
141
5 Uniformizations
Ş Ť
One has P “ n lh s“n Vs . And if one set
#
1 if α P Vsn
f pαqpnq “
0 otherwise,
where n ÞÑ sn is our nice coding of ω ăω , then (i) and (ii) are clear. %
ajout – YP
142
5.5 Natural Π11 sets, natural Π11 ranks
is a good universal set for Σ02 subsets of 2ω . It is one of the rather few known
“natural” good universal sets.
One can show that each Ef¯ is Π03 . Here we claim that again any Σ02 subset of
2ω is of the form Ef¯, uniformly: There is a continuous ψ : ω ω Ñ Cp2ω qω with
Wα “ Eψpαq .
To see this we use the same f as above, coming from Proposition 5.5.3, and now
set
ψn pαqpβq “ f pα, βqpnq.
As in the proof of (i), ψ is continuous. If pα, βq P W then f pα, βq is eventually
0, hence ψn pαqpβq Ñ 0. If pα, βq R W , then (as f pα, βq is never eventually 1)
ψn pαqpβq does not converge. So Wα “ Eψpαq as desired.
As in the discussion above, we get that ψ reduces @1 W to
143
5 Uniformizations
Proof. (i) Let W be the good universal set for Σ02 subsets of 2ω , and viewing ω ω as
a Π02 subset of 2ω , let
W ˚ “ tpα, βq P 2ω ˆ 2ω : pα, βq P W _ α R ω ω u.
pα, βq P W ˚ ÐÑ f pα, βq P D.
s P TKα ÐÑ Dβ P 2ω s Ď f pα, βq
Clearly ϕ is continuous, as
We now give a nice application of this theorem, due to Bourgain, to a result of Szlenk
in Banach Space Theory. It illustrates the power of the non-Borelness results.
Corollary 5.5.6 (Szlenk; Bourgain). There is no separable Banach space with sep-
oversight – YP
arable dual which is universal for separable Banach spaces with separable dual, i.e.
no space X which is a separable Banach space with separable dual and contains an
isomorphic copy of every such space.
144
5.5 Natural Π11 sets, natural Π11 ranks
5.5.8 WF
The sets KpDq and Kω p2ω q of Theorem 5.5.5 are quite commonly used in completeness
arguments. But the most commonly used complete Π11 set is the set WF of wellfounded
ăω
trees on ω, viewed as a subset of 22 — by identifying a tree T with its characteristic
function 1T .
ăω
Proposition 5.5.7. WF is complete Π11 in 22 .
Proof. WF is Π11 , as
145
5 Uniformizations
ăω
(In fact if we use our bijection between ω and 2ăω to turn 22 into a Polish recursive
space, WF is actually Π11 in it.)
ăω
To show WF is complete, we reduce WO to it. Define ϕ : LO Ñ 22 by
Theorem 5.5.8 (Mazurkiewicz). In Cpr0, 1sq (with the sup-norm topology), the set D
of differentiable functions is complete Π11 .
146
5.5 Natural Π11 sets, natural Π11 ranks
For each T , let DT “ tx P r0, 1s : fT is not differentiable at xu. We claim that for
each tree T , DT “ txα : α P rT su. This will finish the proof, for then fT P D Ø T P
WF.
To prove the claim, we fix T . Suppose first x “ xα for some α P rT s. Then for all n,
l . Let y “ x r
x P Jα r
n
n Jα and zn “ max Jαn .
n
As x R Jsr , fT pxq “ 0. Similarly fT pyn q “ |Jα
r | and f pz q “ 0. Also |y ´ x| ď
n
n n
3 r
2 |Jαn |. So we get for all n
ˇ ˇ ˇ ˇ
ˇ fT pzn q ´ fT pxq ˇ ˇ fT pyn q ´ fT pxq ˇ 3
ˇ ˇ “ 0 and ˇ ˇě
ˇ zn´x ˇ ˇ yn´x ˇ 2
where fTN1 “
ř
năN1 fsn is a continuously differentiable function. It follows immedi-
sn PT
ately that fT is differentiable at x, and the claim is proved. %
147
5 Uniformizations
and if λ is the Lebesgue Measure on r0, 1s, set µT “ ϕT pλq, i.e. for any f P Cp2ω q,
µT pf q “ λpf ˝ ϕT q. This defines a function T ÞÑ µT . First we show this function is
continuous. To see this fix f P Cp2ω q and ε ą 0, and pick N large enough so that if
αN “ βN , then |f pαq´f pβq| ď ε (by uniform continuity of f ). Suppose now T, T 1 are
trees with si P T Ø si P T 1 for i ă N . We then show that |µT pf q´µT 1 pf q| ď ε. For the
hypothesis implies that for all x P r0, 1s ϕT pxqN “ ϕT 1 pxqN , hence |f ˝ϕT ´f ˝ϕT | ď ε
on r0, 1s, so
|µT pf q ´ µT 1 pf q| “ |λpf ˝ ϕT ´ f ˝ ϕT 1 q| ď ε,
as desired.
We now show that for a tree T
T P WF ÐÑ µT P M
So µT R M .
Conversely, suppose µT R M , and let A be infinite, with µT pKA q ą 0. As before,
one has ϕT pxq P KA Ø @n`ps Şn P T and ˘ x P Esn q, 1hence the hypothesis implies that
for all n P A sn P TŞ , and λ Ş sn ą 0. Let T be the subtree of T 1 generated by
nPA E
tsn : n P Au. Then nPA Esn “ sPT 1 Es has positive λ measure, and T is infinite.
The main claim is that T 1 is finitely branching, i.e. for all s P T 1 the set tn : sa n P T 1 u
is finite. By a classical argument known as König’s Lemma (see below), it follows that
rT 1 s, hence rT s, is non empty, so T R WF and we are done.
To prove the claim, it is clearly enough to prove that if mn is an increasing sequence
of numbers and a ă 1, then
148
5.5 Natural Π11 sets, natural Π11 ranks
To see this, let Bka “ tx P r0, 1s : kx P r0, as mod 1u. Bka consists exactly of the union
of the intervals r nk , n`a
k s for 0 ď n ď k ´ 1. So if I is an interval and k is large enough,
at least kλpIq ´ 2 intervals are contained in I. We then get
kλpIq ´ 2 kλpIq 2
¨ a ď λpI X Bka q ď ¨a`
k k k
a
so that λpI X Bk q Ñ aλpIq as k Ñ ` Ş8. As athis
˘ holds also for finite unions of intervals,
N as k , . . . k
one easily gets for all N that λ B
iăN ki Ñ a 0 N ´1 Ñ 8. And this
clearly proves the claim. %
For the sake of completeness, we prove
Lemma 5.5.10 (König). If T is a tree on ω which is infinite and finitely branching,
rT s ‰ H.
Proof. For each s let Ts “ ts1 P ω ăω : sa s1 P T u. As tn : pnq P T u is finite and T
is infinite, there is n0 such that Tn0 is infinite. By the same argument, there is n1
such that pTn0 qn1 “ Tpn0 ,n1 q is infinite, and so on. This clearly allows one to build
α “ pn0 , n1 , . . .q with for each i Tαi infinite, in particular Tαi ‰ H, i.e. αi P T . So
α P rT s and we are done. %
149
5 Uniformizations
are Π11 , where pD1 , W1 q is the nice coding of ∆11 subsets of X, given by Proposi-
tion 5.1.1.
Theorem 5.5.11. Let Φ be Π11 on Π11 and ∆11 on ∆11 on X, A Ď X some ∆11 set,
and A “ ξăω1 Φ pAq. Then the natural assignment ϕpxq “ inftξ : x P Φξ pAqu on
ξ
8
Ť
A8 is a Π11 -rank.
Proof. We assume that Φ is Π11 on Π11 and ∆11 on ∆11 , and A P ∆11 , the general case
being obtained by relativization. By the properties of pD1 , W1 q, there is a recursive
function f : ω ω Ñ ω ω such that for α P D1 , f pαq P D1 and Wf1 pαq “ ΦpWα1 q.
Similarly, there is a recursive g : ω ω Ñ ω ω such that if @n pαqn P D1 , then gpαq P ω ω
1 1
Ť
and Wgpαq “ n Wpαq n
.
We claim that the relations
α P WO ^ x P Φ|α| pAq
and α P WO ^ x R Φ|α| pAq
and
` ˘
x P Φ8 pAq ^ y R Φ8 pAq _ ϕpxq ď ϕpyq
ÐÑ Dα P ∆11 pxq α P WO ^ x P Φ|α| pAq ^ @n y R Φ|α n| pAq
“ ‰|
which are true by Moschovakis’ Theorem 4.5.3,. which implies that for x P Φ8 pAq,
since txu is ∆11 pxq
and included in
ϕpxq is an ordinal ă ω1CK´x .
Φ8 pAq there is We now prove the claim. Let αA P D1 be recursive, with Wα1 A “ A, and similarly
ξ ă ω1CK´x with for αH . Define two relations R` and R´ by
x P Φξ pAq. – YP
R` px, αq ÐÑ
˜
”
α P LO ^ Dβ P ω ω
` ˘
@n @m n R Dα Ñ pβqn “ αH
` ˘
^ n is ďα -minimal Ñ pβqn “ αA
` ˘
^ n is the successor of m in ďα Ñ pβqn “ f ppβqm q
` ˘ı
^ n is limit in ďα Ñ pβqn “ f ˝ gpxpβqm : m ăα nyq
¸
´ ¯
1 1
^ β R D _ x P Wf ˝gpβq
150
5.5 Natural Π11 sets, natural Π11 ranks
Clearly R` and R´ are Σ11 relations in ω ω ˆ X. And if α P WO, one easily checks by
induction on α that if β is a witness that px, αq P R` (resp. R´ ) then @n pβqn P D1
1
and Wpβq “ Φ|α n| pAq, so that
|
Then P is Π11 and the rank ϕpxq “ inftξ : n0 P Φξx pHqu is a Π11 -rank on P .
Proof. Consider the operator Φ on X ˆ ω defined by
ΦpAq “ tpx, nq : n P Φx pAx qu .
As f is Borel, one easily checks that for W the good universal set for Π11 subsets of
X ˆ ω, and pD1 , W1 q the corresponding coding of ∆11 subsets of X ˆ ω, one has
` ˘
px, nq P ΦpWα q ÐÑ @B Ď ω @m pα, x, mq P W Ñ m P B Ñ f px, 1B qpnq “ 1,
so Φ is Π11 on Π11 , and
α P D1 ^ px, nq P ΦpWα1 q ÐÑ
α P D ^ @B Ď ω @m pα, x, mq P W1 Ñ m P B Ñ f px, 1B qpnq “ 1,
1
` ˘
whereas
α P D1 ^ px, nq R ΦpWα1 q ÐÑ
α P D ^ @B Ď ω @m m P B Ñ pα, x, mq P W1 Ñ f px, 1B qpnq “ 0,
1
` ˘
(1) Let WF be the set of wellfounded trees, and consider h : WF Ñ ω1 , the height
pas encore dÈfini – ăω
YP
function. Then h is a Π11 (in fact Π11 ) rank on WF. To see this defines f : 22 ˆ2ω Ñ 2ω
by
f p1T , 1S q “ txsy : s R T _ @n sa n P Su.
This is clearly Borel, and one can check that
ΦξT pHq “ txsy : s R Tξ u,
where Tξ is the ξ th stage of the pruning of T (see Subsection 4.1.5). So T P WF Ø
xHy P Φ8 1
T pHq, and for all T P WF, hpT q “ ϕpT q is a Π1 -rank by Theorem 5.5.12.
@α ‰ β ps Ď α ^ s Ď β Ñ
´ ` *
˘ ` ˘
Dn 1S xαn y “ 1 _ 1S xβn y “ 1 .
152
6 Infinite Games
6.1 Determinacy of games
6.1.1 Games
Let X be some set. A game is given by two objects: A non empty regular tree T on
regular tree again...
X, and a payoff set A Ď rT s. In a run of the game GpT, Aq, we imagine two players, pruned tree? – YP
usually named players I and II, alternatively choosing elements xi P X, I playing first:
I: x0 x2
¨¨¨
II : x1 x3
tx : ua x P Su “ tx : ua x P T u,
i.e. S does not restrict the possible plays of II. Heuristically, player I can force player
II to play in S if he wants. This explain the terminology.
A strategy for player I is a function σ from the set of positions of even length in T
into X, and such that for u P T of even length, ua σpuq P T .
Equivalently, we can consider the set Sσ of positions played according to σ, defined
by
u P Sσ ÐÑ @v Ă u of even length v a σpvq Ď u.
Then clearly Sσ is a quasi-strategy for player I, and moreover for each u P Sσ of even
length, tx : ua x P Sσ u is reduced to the singleton tσpuqu. And conversely if a quasi-
strategy S is such that for every u of even length in S, tx : ua xu is a singleton tσpuqu,
then extending σ to the positions u of even length in T zS by σpuq “ xu , any point in
tx : ua x P T u, on gets a strategy σ for player I such that Sσ “ S. So we will frequently
identify the strategy σ (as a function) with the corresponding quasi-strategy Sσ (as a
tree of positions). It will be also convenient, especially when the tree T is X ăω , to
un peu cavalier... –
YP
153
6 Infinite Games
view σ as a function of the sequences of plays of player II, as these define entirely the
corresponding tree Sσ of positions played according to σ.
The notions of quasi-strategy and strategy for player II are defined similarly, changing
the roles of I and II. A quasi-strategy S, or a strategy σ, for player I (resp. player II)
is winning in the game GpT, Aq if rSs Ď A (resp. rSs X A “ H), i.e. if any run played
according to the quasi-strategy is a winning run for the player.
Of course if a quasi-strategy S is winning, any sub-quasi-strategy, in particular any
finer strategy, is obviously winning. The point in looking at quasi-strategies is that in
order to define a finer strategy, one may need the axiom of choice. Also, as we will see
in a moment, quasi-strategies may be very canonical, whereas finer strategies may not.
Finally, we say that the game GpT, Aq is determined if one of the two players has a
winning strategy in the game. (If T is understood, we will also say that A is determined
in case GpT, Aq is).
Heuristically, the notion of existence of winning strategies is an attempt to give a
meaning to expressions like
` ˘
Dx0 @x1 Dx2 @x3 ¨ ¨ ¨ A px0 , x1 , . . .q ,
Tu “ts : ua s P T u
and Au “tf P rTu s : ua f P Au.
Define
154
6.1 Determinacy of games
Case 2. Otherwise. We claim that S is a winning quasi-strategy for player I in GpT, Aq.
First we have to show that S is a quasi-strategy for player I. Clearly the plays of
II are unrestricted in S, by its very definition. Also S ‰ H, as H P S. So we must
show S is regular, i.e. for any u P SI and any x with ua x P T , there is a y with
regular – YP
ua xa y P SI . Suppose this fails for some u and x. Then for each y with ua xa y P T
we can pick a winning strategy Sy for II in the game GpTua xa y , Aua xa y q, by the
axiom of choice. Define
ď
Su “ tpxqu Y tpx, yq : ua xa y P T u Y txa y a s : s P Sy u.
ua xa yPT
One easily checks that Su is a winning strategy for player II in the game GpTu , Au q,
contradicting that u P SI .
This shows S is a quasi-strategy for player I. Note that we haven’t yet used
any property of A, i.e. for any A, if II has no winning strategy in GpT, Aq,
the preceding Sdefines a quasi-strategy for I (we could call the “non-loosing”
quasi-strategy).
Now if A is closed, we claim that S is winning for I., i.e. that rSs Ď A. Suppose
not. As A is closed, this means that there exists a position s P S such that for
any f P rT s, s Ď f implies f R A. We may assume s is of even length. But then
any strategy for II is winning in GpTs , As q, as As “ H. So s R SI a contradiction
which proves the claim. If one wants a winning strategy for I, one just refines S,
using the axiom of choice. %
Theorem 6.1.2 (the strategic basis theorem). Let T be a ∆11 regular tree on ω, and
regular – YP
A Ď rT s a closed Σ11 set in ω ω . If player II has a winning strategy in GpT, Aq, he has
a ∆11 (in the codes) winning strategy (and similarly, by duality, with an open Π11 set
A Ď rT s and player I).
SI1 “ tu P T of even length : II has no ∆11 winning strategy in the game GpTu , Au qu
155
6 Infinite Games
To finish the proof, it is enough to show that S 1 is a winning quasi-strategy for player
I in GpT, Aq.
There are two new ingredients: First suppose rS 1 s * A, then as before Du P SI1
with @α P rT s u Ď α Ñ α R A. To get a contradiction we need to see that II
has a ∆11 winning strategy in GpTu , Hq. But this is simple, as Tu is ∆11 : The set
B “ tn : sn P Tu is of even lengthu is ∆11 , and for each n P B there is an m such that
sn a m P Tu . As this relation is ∆11 , by the ω-Uniformization Theorem 4.3.5, there is a
∆11 -recursive function f : B Ñ ω which satisfies sn a f pnq P Tu for n P B, i.e. gives a
∆11 winning strategy in GpTu , Hq.
The second thing is that we have to prove that for any u P SI1 and any n with ua n P T
there is an m such that ua na m P SI1 . By the argument in Theorem 6.1.1, this will
finish the proof. Suppose not, for some u P SI1 and some n with ua n P T . This means
that for every m with ua na m P T there exists α P ∆11 in ω ω which codes a winning
strategy for player II in GpTua na m , Aua na m q, i.e. satisfies
Now this property is Π11 in m and α, so we can apply the ∆11 -Uniformization Theo-
rem 5.1.6 to get a ∆11 -recursive g : ω Ñ ω ω which associates to m such that ua na m P T
such a (code for a) winning strategy gpmq. But then easily the strategy in GpTu , Au q
defined by σpHq “ n, and if I has played m, by the following the winning strategy gpmq
is both ∆11 in the codes, and winning for II in GpTu , Au q, contradicting u P SI1 . %
Note that the previous result is asymmetric. In the case of winning strategies for
player I in the above game GpT, Aq, one can say that
is Σ11 , as one easily checks. So Gandy’s Theorem 5.3.7 applies and gives a strategy
really? – YP
σ recursive in O, or with ω1CK´σ “ ω1 . But it is not always true that there are ∆11
winning strategies, even if T “ ω ăω and A is Π01 . For example, let A be any non Π01
subset of ω ω , and let A˚ “ tα P ω ω : pαq0 P Au where as usual pαq0 pkq “ αp2kq. If A
is non empty, I clearly has a a winning strategy in Gpω ăω , A˚ q, he plays some α0 P A,
whatever II does, and wins. On the other hand, any ∆11 -recursive in the codes winning
strategy for I in Gpω ăω , A˚ q clearly defines a ∆11 -recursive element of A, and there are
non empty Π01 sets in ω ω with non ∆11 member by Theorem 5.1.11.
156
6.1 Determinacy of games
Pas claire cette définition... Je ne pense pas que ça définisse UNE fonction, si? – K sensé? – K
J’ai changé le deuxième ď ξ par ă ξ. Je pense qu’il faut comprendre: je pense... – YP
fA puq is defined and fA puq “ η if and only if η is the minimal ξ such that
´ ´ ¯¯
@xDy ua x R T _ ua xa y P T ^ fA pua xa yq is defined ^ fA pua xa yq ă ξ .
– YP
Clearly the definition of fA depends only on pTA qu , hence if u E v, then u P dom fA
if and only if v P dom fA , and if u, v P dom fA , fA puq “ fA pvq.
Suppose H P dom fA and consider the quasi-strategy S for II which consists in
staying in dom fA and strictly decreasing fA if fA is positive, and remaining in a
position with fA “ 0 otherwise. By the definition of fA , this is a quasi-strategy, and
with no memory by the remark above. And it is winning, for in any run f played this
way, value 0 is obtained after a while, and hence f R A. Now if H R dom fA , then
by definition of fA the quasi-strategy for I which consists in staying outside dom fA is
indeed a quasi-strategy, clearly with no memory, and contained in TA , hence winning
for I. (This proof shows that in fact it is the canonical non losing winning quasi-strategy
defined in Theorem 6.1.1.) %
It can be shown, using canonical representation of ∆02 sets as difference of closed sets
(see Subsection 7.1.3), that the previous result extends to ∆02 games. However, it fails
for Σ02 and Π02 ones for both players, by considering e.g. T “ 2ăω and the following
Σ02 sets:
A0 “ tα P 2ω : α ends up with 01010101 . . .u ,
157
6 Infinite Games
and
A1 “ tα P 2ω : pαq0 ends up with 01010101 . . .u .
In GpT, A0 q, II clearly wins - by e.g. playing the same thing as I. In GpT, A1 q, I wins,
by e.g. playing 010101 . . .. But in both cases, all u’s of the same length are equivalent,
by homogeneity of A0 and A1 , hence a no-memory strategy must be constant - and
this is clearly not winning.
However, by relaxing a bit the no-memory condition, one can get a positive result
for Σ02 games. Given T and a sequence pAn q of closed subsets of rT s, let us say that
a quasi-strategy S is a no-memory quasi-strategy relatively to the sequence pAn q if for
L: relative – K
all u, v of the same length in T with Tu “ Tv and for all n pAn qu “ pAn qv , one has
Su “ Sv .
Theorem 6.1.4 (Gurevich - Harrington). Let A “ n An be a Σ02 set in rT s, with An
Ť
closed. Then in GpT, Aq, one of the player has a winning strategy which is no-memory
relatively to the sequence pAn q. So in particular (Wolfe), every Σ02 (and every Π02 )
game is determined.
Proof. Without loss of generality, we may assume the sets An are increasing. Let T n
be the regular tree for An , and let E be the equivalence u E v if and only if u and v
regular? – K
are two positions of the same length in T with Tu “ Tv and for all n, Tun “ Tvn . We
also set ! )
Eu “ pv, wq P T ˆ T : ua vEua w ,
so that clearly if u E v, then Eu “ Ev .
Say that a quasi-strategy S respects E if u E v Ñ Su “ Sv , and similarly for Eu . So
we want to find a winning quasi-strategy in GpT, Aq which respects E. We first prove
the following fact: let S be a quasi-strategy for II in GpT, Aq which respects E, and
consider:
D “ tu P S of even length :
I has a winning strategy in GpSu , Au X rSu sq which respects Eu u.
Let
D1 “ tu P S of even length :
I has a winning strategy on Su to reach a position in Du u.
158
6.1 Determinacy of games
(c) u P D. Then pick the least (for the wellordering) v “ vpuq P D such that some
extension w of v in Spvq satisfies uEw, and put ua x P S ˚ if wa x P Spvq. This
is well defined, for Spvq respects Ev , hence the definition does not depend on w,
only on v. Moreover, if u E u1 , the corresponding v’s will be the same, so this
definition respects E.
Finally, let u P D1 , and f a run in rSu˚ s. Then after a while, f 2n P D. Now by the
definition, the sequence vpf 2n q is decreasing, at large, for the well-ordering on D. So it
must stabilize at some f 2k , and from that point on, the run is according to a winning
strategy in GpSua f 2k , Aua f 2k X rSua f 2k sq. So f P Au , as desired. This proves the
fact.
We now come back to GpT, Aq. Consider
(b) Otherwise, pick the least n such that Anu ‰ H, and let fn,u be the function,
given by Lemma 6.1.3, corresponding to the closed game GpSu , Anu X rSu sq. The
claim is that H P dom fn,u , for otherwise by Lemma 6.1.3, I wins in this game
by a strategy respecting Eu , so u P D a contradiction. Then put ua xa y P S ˚ if
fn,u pxa yq ă fn,u pHq.
By weakening again the notion of no-memory strategies to include some (fixed) finite
amount of memory, Gurevich and Harrington [GH82] have extended the previous result
to include sets which are in the Boolean algebra generated by Σ02 and Π02 . For this,
and in relationships with automata theory see missing reference.
missing reference – K
bien jouÈ! – YP
6.1.5 Martin’s theorem
The determinacy of Σ03 and Π03 games was proved in 1963 by M. Davis ([Dav64]). Many
attempts were made to prove determinacy of higher levels Borel games on ω, but in
1970, H. Friedman [Fri71] proved a remarkable metamathematical result: over some
weak system of axioms of set theory, the determinacy of all Σ04 games on ω implies
159
6 Infinite Games
the consistency of the existence of ℵ1 . This result was later extended by Martin to
show that the determinacy of Σ0ξ games on ω implies the existence of ℵξ (reference?).
reference? – K
This means that there cannot be any elementary proof of the determinacy of all Borel
games on ω. Still this is a theorem of ZFC, as proved by D.A Martin in 1975 [Mar75].
In the proof we present below, which is a later proof of Martin [Mar85], we will use for
Σ03`ξ games on ω the existence of the 3 ` ξ th iterate of the power set of ω. This can
be reduced to the existence of ℵξ by logical means, so that the obstruction above gives
the exact level of complexity of this statement.
Theorem 6.1.5 (Martin). Let X be any set, T a regular tree on X. Then all games
regular? – K
GpT, Aq, with A Borel in rT s are determined.
– π is a function from T ˚ to T which respects the length and the extension order,
i.e. if u˚ P T ˚ , then lhpπu˚ q “ lh u˚ , and u˚ Ď v ˚ in T ˚ implies πu˚ Ď πv ˚ . So
π extends to a continuous function, we still denote by π : rT ˚ s Ñ rT s.
Condition (‹) easily implies that if A Ď rT s, and A˚ “ π ´1 pAq is such that GpT ˚ , A˚ q
is determined, then GpT, Aq is determined too. In fact, if σ ˚ is winning for I (resp.
II) in GpT ˚ , A˚ q, then σ “ ϕpσ ˚ q is winning for I (resp. II) in GpT, Aq. For otherwise
there is a β P rT s played according to σ with β R A (resp β P A), and then the run
unraveledβ ˚ P rT ˚ s given by (‹) shows that σ ˚ is not winning.
Given k P ω, let us say that pT ˚ , π, ϕq is a k-unraveling if it is an unraveling, and
moreover the sequences in T and T ˚ of length less or equal to 2k are the same, and π
is the identity on these sequences. If now A Ď rT s, we say that A can be unraveled if
for each k P ω, there is a k-unraveling pT ˚ , π, ϕq of T such that π ´1 pAq “ A˚ is clopen
in rT ˚ s. By the previous remark, if A can be unraveled, GpT, Aq is determined. So
Martin’s Theorem follows from:
160
6.1 Determinacy of games
161
6 Infinite Games
as in case 1. He then does as in case 1, i.e. plays according to the winning strategy till
a position u of even length 2p in ΣI zpTA qpx0 ,...,x2k q and then plays according to answers
by σ ˚ to positions of form
m ě p. In both cases, this clearly defines for each run β played this way some run β ˚
on T ˚ , played accordingly to σ ˚ , with β “ πβ ˚ , as desired.
Note that the run β ˚ is not defined continuously in β (this is the key to the proof),
for as long as we are in case 2, we do not know whether at the end β ˚ p2k ` 1q will be
of type px2k`1 , p0, uqq or px2k`1 , p1, ΣII qq. We now fix a strategy τ ˚ for II on T ˚ , and
define τ as follows: in the first 2k moves, II follows τ ˚ . Let then I play x2k . Consider:
(
S “ ΣI : ΣI is a strategy for I in Tpx0 ,...,x2k q
and
U “ u P T of even length :
(
for some ΣI P S, τ ˚ answers some px2k`1 , p0, uqq to px0 , . . . , x2k`1 , px2k , ΣI qq .
Consider in Tpx0 ,...,x2k q the game consisting for II to reach a position px2k`1 , . . . , x2l´1 q
with px0 , . . . , x2l´1 q P U. This is an open game for II.
Case I: II has a winning strategy in this game. He then plays it till such a u “
px0 , . . . , x2l´1 q is reached, with corresponding Σu P S. He then plays according to
answers by τ ˚ to positions of form
with m ą l. Clearly this gives for each run β played this way a β ˚ P rT ˚ s played
according to τ ˚ with β “ πβ ˚ as desired.
Case II: Otherwise. Then player I has his canonical winning quasi-strategy ΣI in
the above game. Now note that the answer by τ ˚ to px2k , ΣI q cannot be of the form
px2k`1 , p0, uqq, for otherwise, as ΣI P S, u P U. But u P ΣI , and this implies ΣI is not
winning for I in the above game. So τ ˚ gives px2k`1 , 1, ΣII q, and II answers in T by
the answers by τ ˚ to positions of form
with l ą k, as long as he can, i.e. as long as px2k`2 , . . . , x2l q P ΣII . Suppose now some
position px0 , . . . , x2l q is reached in T with px2k`2 , . . . , x2l q R ΣII . ΣII is a quasi-strategy
for II in a game on pΣI qpx2k`1 q , hence I cannot play outside ΣII without getting outside
pΣI qpx2k`1 q . So px2k`1 , . . . , x2l q R ΣI . But then II has a winning strategy from this
point on to get into U, and we are back in case I. In both cases, this gives for any run
β played this way some β ˚ played according to τ ˚ with β “ πβ ˚ , as desired. %
162
6.1 Determinacy of games
163
6 Infinite Games
cardinal axiom, and follows from the existence of a measurable cardinal (Martin 1970
citations added – K
[Mar70], and Harrington 1978 [Har78]). The proof of various instances of determinacy
from large cardinals assumptions is the central subject of D.A. Martin’s forthcoming
monograph r??s. And its applications to the study of properties of general definable
Une autre Arlésienne
–K subsets of Polish spaces is covered in Moschovakis [Mos80].
Proof. (1)ñ(2). If A is meager, its interior is also meager, hence is empty by (1).
164
6.2 Baire spaces and the Baire property
I: V0 V1
¨¨¨
II : U0 U1
Ş Ş
subject to the condition Vi`1 Ď Ui Ď Vi . Player II wins if n Un “ n Vn is non
empty.
Theorem 6.2.2 (Saint-Raymond). X is a Baire space if and only if Player I has no
winning strategy in the Choquet game on X.
Proof. Suppose first that X is not a Baire space, and pickŞa non empty open V0 which
is meager in X, and dense open sets pWn qnPω with V0 X n Wn “ H. We describe a
winning strategy for I: he first plays V0 , and at step n answers
Ş to Un by VŞn`1 “ Un XWn ,
which is non empty as Wn is dense. Clearly at the end, n Vn Ď V0 X n Wn “ H, so
this strategy is winning.
Conversely, suppose σ is a winning strategy for I, and let V0 “ σpHq. We claim
that V0 is meager. First, we pick a maximal family W1 of pairwiseŤdisjoint `open sets ˘
of form σpU q, for U a non empty open subset of V0 . Then W1 “ W1 Y Ť XzV 0 is
dense in X, for otherwise there is a non empty open U Ď V0 with U X W1 “ H.
But then σpU q can be added to W1 , contradicting its maximality. For each V1 P
W1 , we also pick U0 Ď V0 with σpU0 q “ V1 . Now for each V1 P W1 , we choose a
maximal family W1 pV1 q of pairwise disjoint sets of form σpU0 , U q, for U non empty
open subsetsŤof ŤV1 , and for each V2 P W1 pV1 q an U1 Ď V1 with σpU0 , U1 q “ V2 ;
again W2 “ t W1 pV1 q : V1 P W1 u Y pXzV0 q is dense. Continuing this way, we get
Ť pWn q of families of pairwise disjoint non empty open subsets of V0 with
a sequence
Wn “ Wn Y pXzV0 q dense in X, and such that for each Vn P Wn corresponds a
unique Vi P Wi , for each i ă n, and a unique Ui , for each i ă n, such that pV0 ,Ş
U0 , Vn q
is a position in the Choquet game and Vn “ σpU0 , . . . , Un´1 q. So if x P V0 X n Wn ,
there is a unique sequence pVn , Un q with Vn P Wn , such that for all n, x P Vn . But
this sequence isŞ a run of the Choquet game where Player I has followedŞ his winning
strategy, hence n Vn “ H, a contradiction. This shows that V0 X n Wn “ H, and
hence V0 is meager in X %
I: pV0 , x0 q pV1 , x1 q
¨¨¨
II : U0 U1
165
6 Infinite Games
As before, I and II alternate playing non-empty open sets with Vi`1 Ď Ui Ď Vi . But
player I also plays a point xi P Vi and II must answer with Ui containing xi .
Definition 6.2.3. A space X is a (strong) Choquet space if player II has a winning
strategy in the (strong) Choquet game on X.
Clearly every string Choquet space is a Choquet space, and every Choquet space is
a Baire space, by Subsection 6.2.2.
Theorem 6.2.4. (i) Every locally compact space, and every complete metrizable
space, is a strong Choquet space.
(ii) Any product of (strong) Choquet spaces is a (strong) Choquet space.
(iii) If X is a Choquet space and Y is a Baire space, X ˆ Y is a Baire space.
(iv) If X is a strong Choquet space, and Y Ď X is a Gδ set, Y is a strong Choquet
space.
Proof. (i) If X is locally compact, II can play as follows: if at step n player I has
played pVn , xn q, II answers by Un with xn P Un , Un compact, Un Ď Vn . Then, by
compactness, č č
Vn “ Un ‰ H.
nPω nPω
If X is complete metrizable and d is a complete compatible metric on X, II
answers to pVn , xn q by some Un with xn P Un , Un Ď Vn and diampUn q ď 2´n .
Then by completeness of d,
č č
Vn “ Un ‰ H.
nPω nPω
where, for all i RŤJn , Wni “ Xi , for a certain finite set Jn Ď J which we may
assume contains kăn Jk . Then he answers by
ź
Un “ Uni ,
iPJ
where, for i R Jn , Uni “ Xi , and for i P Jn , Uni is the answer by σi to the play
pWni 0 , Ť
Wni 0 `1 , . . . , Wni q, where n0 is least with i P Jn0 . At the end of a run, let
J8 “ nPω Jn . Then ˜ ¸
ź č č
Uni Ď Un ,
iPω nPω nPω
166
6.2 Baire spaces and the Baire property
i
Ş
with nPω UŞ n “ Xi if i R J8 , and a winning run of the Choquet game on Xi if
i P J8 . So nPω Un ‰ H as desired. The case of the strong Choquet games is
similar, and we omit it.
(iii) Let σ be a strategy for II in the Choquet game on X, and suppose τ is a winning
strategy for I in the game on X ˆ Y . We will then define a winning strategy θ
for I in the game on Y , a contradiction which will prove (iii).
First let θpHq “ V0XˆY . Pick V0X ˆ V0Y ‰ H such that V0X ˆ V0Y Ď V0XˆY , and
set θpHq “ V0Y . Suppose II plays U0Y Ď V0Y . Let U0X “ σpV0X q, and consider
U0XˆY “ U0X ˆ U0Y . Let V1XˆY “ θpU0XˆY q, and pick V1X ˆ V1Y ‰ H such that
V1X ˆ V1Y Ď V1XˆY , and set`θpU0Y q “˘ V`1Y . Continuing
˘ ` this way, we ˘set in a
X X Y Y
run followed by θ sequences Un , Vn , Un , Vn and Un XˆY , VnXˆY played
according to σ, θ and τ respectively, with UnXˆY X Y
“ Un ˆ Un . But
č č č
UnXˆY “ UnX ˆ UnY “ H,
nPω nPω nPω
(iv) Let σ Şbe a winning strategy for II in the strong Choquet game on X, and let
A “ nPω Gn be a Gδ set in X, with pGn qnPω a sequence of decreasing open
sets. We describe a winning strategy for II in the strong Choquet game on A as
follows.
Suppose I plays pV0A , x0 q. Player II first chooses V0X open in X with V0A “ V0X XA
and w.l.o.g. V0X Ď G0 , and plays
U0A “ A X σpV0X , x0 q.
The point is that as x0 P A, the set U0A is non-empty, hence is playable in the
game on A. Similarly at stage n, if I plays pVnA , xn q, II chooses VnX Ď Gn open
in X with VnA “ VnX X A, and answers
167
6 Infinite Games
(ii) X is a strong Choquet space if and only if X is Gδ in X̂, hence if and only if X
is Polish.
(ii) The proof is very similar. If X is Gδ in X̂, X is a strong Choquet space either by
(iv) or – as X is Polish by Theorem 3.5.6, by (i).
For the converse, we need the following fact: if U is any family of open sets in
X̂, there is a point-finite refinement V of U, i.e. a family of open sets with same
union, such that every V P V is contained in some U P U, and for all x P X̂,
tV P V : x P V u is finite. Moreover for any given integer k, we may insure that
for V P V, diampV q ď 2´k . To prove the Ť fact, pick a countable subfamily pUn qnPω
of U with same union, and write Un “ pPω Unp , with Unp increasing, open, and
with Unp Ď Un . Let then ď
Vn “ Un z Unp .
păn
One easily checks that V “ tVn : n P ωu works. If moreover one replaces at the
beginning the Un ’s by open subsets with same union and diameter less than 2´k ,
one gets the last conclusion.
168
6.2 Baire spaces and the Baire property
Let now σ be a winning strategy for player II in the strong Choquet game on X.
For each x P X, let Ux be open in X̂ with X X Ux “ σpX, xq. By the fact above,
let W1 be a point-finite refinement of U “ tUx : x P Xu, and for each Ť V1 P ŤW1 ,
pick some x0 P X such that V1 X X Ď σpX, x0 q. Note that W1 “ W1 “ U
is open and contains X, as x P Ux for all x P X. At step 2, for each V1 in
W1 with associated x0 and for each x P X X V1 , pick Ux1 open in X̂ such that
Ux1 X X “ σppX, x0 q, pX X V1 , xqq, and let W2 be a point finite refinement of
U1 “ Ux1 : x P V1 , V1 P W1 ,
(
consisting of sets of diameter less than 2´1 . For each V2 P W2 , pick some V1 P W1
and some x1 P V1 with V2Ť X X Ď σppX, x0 q, pV1 , x1 qq. Continuing this way, we
obtainŞa sequence Wn “ Wn of open sets in X̂ which all contain X, hence
A “ nPω Wn is a Gδ set on X̂ which contains X. The claim is that in fact,
A “ X. To see this, let x P A, and consider the set T of all sequences of
form ppV0 , x0 q, . . . , pVn´1 , xn´1 q, pVn , xqq where Vi P Wi , V0 “ X, x P Vn and
ppV0 , x0 q, . . . , pVn´1 , xn´1 qq is the unique sequence associated to Vn P Wn in the
construction. This set is a tree, and is infinite as it contains sequences of all
lengths. Also, every sequence in T has only finitely many successors in T , as
each Wn is point-finite. This implies, by König’s lemma, that T has an infinite
branch pVn , xn qnPω . By the construction, pVn X X, xn qnPω corresponds to a run
in the strong Choquet game on X, with II playing his winning strategy. Hence
˜ ¸
č
Vn X X ‰ H.
nPω
Ş
But by the condition on diameters, nPω Vn “ txu. So x P X as desired.
%
Remark 6.2.6. This theorem shows in particular that the notions of Choquet space
and strong Choquet space differ, even for metrizable separable spaces. And using the
axiom of choice, it is easy to build a dense non-comeager set in R which is a Baire
space, so that the three notions are different.
Reference missing –
K
6.2.6 The Gandy-Harrington topology
We now introduce a tool which has proved to be extremely useful in the Descriptive Set
Theory of Σ11 and Π11 sets. It was introduced first by Gandy (in equivalent but different
terms), and became used mainly after Harrington’s work on equivalence relations.
Reference missing –
Let X be a Suslin recursive space. The Gandy-Harrington topology τ “ τX on X is K
the topology generated by all Σ11 subsets of X. This topology has a countable basis,
and is finer than the original topology on X, as all VnX are Σ11 . In general, this topology
is not metrizable, or what is the same, not regular: for X “ ω ω , we know that there
are Σ11 that are not Π11 . Now each τ -open set being a countable union of Σ11 sets is Σ11 ,
hence each τ -closed, and each τ ´ Fσ set must be Π11 . So there are τ -open sets which
are not τ ´ Fσ , and τ is not regular.
169
6 Infinite Games
Theorem 6.2.7. Let X be a Suslin recursive space. Then pX, τ q, the space X endowed
with the Gandy-Harrington topology, is a strong Choquet space.
Proof. By embedding X as a Σ11 subset of ω ω , we may assume X “ ω ω . We describe
a winning strategy for II in the strong Choquet game on pω ω , τ q. Let I play pV0 , x0 q.
As V0 is τ -open, II can choose a Σ11 set A0 with x0 P A0 Ď V0 . Player II also chooses
a Π01 set F0 Ď ω ω ˆ ω ω with
α P A0 Ø Dβ pα, βq P F0 .
I now plays V1 Ď A0 and α1 P V1 , and II first chooses β01 such that pα1 , β01 q P F0 .
Consider the set
Observe that A10 is Σ11 , and α1 P A10 , so II can play a Σ11 set A1 with α1 P A1 and
A1 Ď A10 X V1 . He also picks F1 in Π01 pω ω ˆ ω ω q with
α P A1 Ø Dβ pα, βq P F1 .
Let I play V2 Ď A1 , and α2 P V2 . II first picks β02 and β12 with β02 “ β01 , pα2 , β02 q P F0
and pα2 , β12 q P F1 . Let
and
A21 “ α P ω ω : α2 “ α2 2 ^ Dβpβ2 “ β12 2 ^ pα, βq P F1 q .
(
These sets are Σ11 , and α2 P A20 X A21 X V2 , so II can plays some Σ11 set A2 with α2 P A2
and A2 Ď A20 X A21 X V2 , and pick some F2 in Π01 pω ω ˆ ω ω q with
α P A2 Ø Dβ pα, βq P F2 .
Continuing this way, II has built, at the end of a run, a sequence of Σ11 sets pAn qnPω ,
a sequence of Π01 pω ω ˆ ω ω q sets pFn qnPω , Σ11 sets pAni qiăn , and elements pβin qiăn corre-
sponding to the αn ’s played by I, in such a way that
(a) α P An Ø Dβ pα, βq P Fn , pαn , βin q P F , and
n
Ş
(b) An Ď Vn X iăn Ai .
(c) The sequence pαn qnPω converges to some α, and the sequence pβin qnPω converges
to some βi for each i P ω.
We claim that α P nPω An “ nPω Vn . To see this, note that, in ω ω ˆ ω ω and for
Ş Ş
each integer i, the sequence pαn , βin qnPω converges to pα, βi q. And as, for all integer
n, pαn , βin q P Fi and Fi is closed, pα, βi q P Fi . But then α P Ai , for all integer i, as
desired. %
170
6.2 Baire spaces and the Baire property
This result, together with (ii), shows that pX, τ q is very closed to being Polish. This
can be made even more precise, as follows. For each Σ11 set A in X, consider its closure
τ
A for the topology τ . By Theorem 4.3.8, one has
τ
x R A Ø DB P Σ11 pA X B “ H ^ x P Bq
Ø DB P ∆11 pA X B “ H ^ x P Bq
Ø Dn n P D1 ^ A X Wn1 “ H ^ x P Wn1 ,
` ˘
τ
where pD1 , W 1 q is the coding of ∆11 subsets of X defined in Proposition 5.1.1. So A
τ
is Π11 , and A is Σ11 , hence is open for τ . Let
ď τ
A zA : A P Σ11 pXq .
(
X ˚ “ Xz
τ
As each A zA is a τ -closed τ -rare set, X ˚ is a τ -dense Gδ set in X, hence pX ˚ , τ q
τ
is a strong Choquet space. But the basis A X X ˚ “ A X X ˚ , A P Σ11 of pX ˚ , τ q,
consists of clopen sets, hence pX ˚ , τ q is regular, and hence metrizable separable. By
Theorem 6.2.5, it is in fact Polish. So pX, τ q contains a dense Gδ which is Polish for
the induced topology.
Of course we can also consider the relativized topologies τ pαq, for α P ω ω , generated
on X by the Σ11 pαq sets. And by relativizing the previous discussion, we get for
pX, τ pαqq the same properties as for pX, τ q.
(i) BPpXq is a σ-field of subsets of X, and hence every Borel subset of X has the
Baire property.
(ii) BPpXq is a family which admits envelops. Hence by Theorem 4.1.3, it is closed
under operation A. In particular, if X is metrizable separable, any Σ11 set, and
even any C-set, in X has the Baire property for any finer topology on X.
Proof. (i) As the family of meager sets in X is a σ-ideal, BPpXq is clearly closed under
countable unions. Also any comeager set has the Baire property, so to show that
BPpXq is closed under complementation, it is enough to show that closed sets
are in BPpXq. But if F is closed and F̊ is its interior, F “ F̊ Y pF zF̊ q, and
pF zF̊ q is rare. So BPpXq is a σ-field. As it contains the open sets, it contains
all Borel sets in X.
171
6 Infinite Games
(ii) Let A Ď X. We want to find a set A˚ P BPpXq, A Ď A˚ , such that for any
B P BPpXq with A Ď B, A˚ zB is meager in X.
Consider a maximal family Ť W of pairwise disjoint open sets W with A X W
meager in X. Let W0 “ W. First we claim that W0 X A is meager in X. To
see this, it is enough to show that if B Ď W0 is such that, for all W P W, B X W
is rare, then B is rare in X. Now note that
ď (
BĎ B X W X W : W P W Y pW0 zW0 q,
Ť
so if x P B XW0 , there is a W P W with x P W , but then x R W ‰W 1 PW pB X W 1q
as the sets in W are pairwise disjoints; so x P B X W X W .
So if U is open and non-empty, U Ď B, there is W P W with U X W ‰ H, hence
U X W Ď B X W contradicting the fact that B X W is rare. This proves our
claim: W0 X A is meager in X. Let A˚ “ pXzW0 q Y pW0 X Aq, then A˚ P BPpXq
and A˚ contains A. We show that A˚ is an envelop for A. Let B P BPpXq be
such that A Ď B. As A˚ zB P BPpXq, we can write it as U 4M , with U open and
M meager in X. So U X B Ď M , and a fortiori U X A Ď M . This implies that
U zW0 is empty, for otherwise it could be added to the family W, contradicting
its maximality. Now
But U XpXzW0 q is empty, and the three last sets are meager in X. So U is meager,
and finally A˚ zB too. So BPpXq admits envelops, and by Spilrajn-Marczewski’s
Theorem 4.1.3, is closed under operation A.
Finally, if X is metrizable separable, the family C is the least σ-field of sets
closed under operation A and containing the open sets in X. If τ is a finer
topology on X, the open sets in X are all τ open, hence by the result above,
C Ď BPpX, τ q. %
172
6.2 Baire spaces and the Baire property
Ť
Conversely, suppose A is meager, so that A Ď nPω Fn , with Fn closed and rare in
X ˆ Y . Each pFn qx is closed, and as Y is Baire, pFn qx is meager if and only if pFn qx
is rare. So if pVn qnPω is a basis of non-empty sets of the topology of Y , one gets
ď
tx P X : pFn qx is non-meageru “ tx P X : Vk Ď pFn qx u .
kPω
Now for all n and k, the set Mnk “ tx P X : Vk Ď pFn qx u is closed, for if x R Mnk , there
exists y P Vk such that px, yq R Fn , and this still holds in a neighborhood of x, as Fn
is closed. Also Mnk is rare, for otherwise for some non-empty U in X we would have
U ˆVk Ď Fn , which is impossible as Fn is rare. So the set tx P X : pFn qx is non-meageru
is meager, and finally
ď
tx P X : Ax is non-meager in Y u Ď tx P X : pFn qx is non-meageru
nPω
is meager too. %
– tx P X : Ax is comeageru is Π0ξ ;
– tx P X : Ax is meageru is Π0ξ`1 .
Dually, if B Ď X ˆ Y is Σ0ξ ,
– tx P X : Bx is comeageru is Π0ξ`1 ;
– tx P X : Bx is meageru is Π0ξ .
Proof. The main point in the computation is that for a set E Ď Y with the Baire
property, E is non meager if and only if for some U ‰ H open in Y , U zE is meager.
This is clear as E “ U 4 M with U open and M meager, and as Y is a Baire space,
E is non meager if and only if U is non-empty.
The result is proved by induction, For A open,
tx P X : Ax is meageru “ tx P X : Ax “ Hu
tx P X : Ax is meageru “ tx P X : Ax is rareu
č! )
“ x P X : Vn X A{x ‰ H
nPω
173
6 Infinite Games
Observe that pX ˆ Vn qzA is Σ0ξ , hence Ax is not meager is Σ0ξ`1 , and Ax is meager is
Π0ξ`1 as desired. The final computation is again by duality. %
Proof. For each Vn in the basis of Y , An “ f ´1 pVn q has the Baire property,
Ť hence can
be written Un 4 Mn , with Un open and Mn meager in X. Let A “ Xz nPω Mn . A is
comeager in X, and f A is continuous, as f ´1 pVn q X A “ Un X A is open. %
This proposition can be applied to Borel functions, and also to the C-measurable
uniformizing functions for Σ11 sets given by the Jankov - Von Neumann uniformization
theorem in Subsection 5.4.3.
I: s0 s2
¨¨¨
II : s1 s3
174
6.2 Baire spaces and the Baire property
Lemma 6.2.12. (i) II has a winning strategy in G˚˚ pAq if and only if A is meager.
(ii) I has a winning strategy in G˚˚ pAq if and only if A is comeager on some non-
empty open set U , i.e. U zA is meager.
Proof.Ť(i) Suppose first A is meager in ω ω , and let Fn be closed rare sets with A Ď
nPω Fn . let Tn be a tree on ω with Fn “ rTn s. For every s P Tn , the open set
tα P ω ω : s Ď αu is not contained in Fn , hence there is a finite sequence t such
that both s Ď t and t R Tn . So II wins by playing as follows: I first plays s0 ,
and II answers s1 so that s0 a s1 R T0 . Then I plays s2 , and II answers s3 so that
s0 a s1 a s2 a s3 R T1 , and so on and soŤforth. Clearly at the end α “ s0 a s1 a . . . is
a branch through no Tn , hence α R nPω Fn and a fortiori α R A.
Conversely, let σ be a winning strategy for player II in the game G˚˚ pAq. For
α P ω ω , say that u “ ps0 , s1 , . . . , s2k´1 q of even length is critical for α if
(a) s0 a s1 a . . . s2k´1 Ď α,
(b) for all i ă k, up2i ` 1q “ σpu2i`1 q, and
(c) for any s P ω ăω non-empty,
First, we claim that for any α P A there is a u which is critical for α. For other-
wise, one can inductively build a sequence psn q which is a run of the game and in
which player II follows σ, such that α “ s0 a s1 a . . . a sn a . . .. But this contradicts
the fact that σ is winning for II. Let S “ tu P ω ăω : lhpuq is even, u satisfies (b)u,
ω
Ť each u P S, let Fu “ tα P ω : u is critical for αu. By the fact above,
and for
A Ď uPS Fu , and as S is countable, it is enough to prove that each Fu is a rare
closed set.
Suppose αn P Fu , and αn Ñ α. Then (a) is true for α. So if u is not critical,
there is an extension ua ps, s1 q still satisfying (a) and (b) for α. But then for large
enough n, ua ps, s1 q satisfies (a) for αn , and u is not critical for αn . This contra-
diction proves that α P Fu , hence Fu is closed. If now t is such that tα P ω ω :
t Ď αu Ď Fu , then without lost of generality t is of the form s0 a s1 a . . . a s2k´1 a s
with s non-empty. But then if α extends ta σps0 , s1 , . . . , s2k´1 , sq, u is not critical
for α, contradicting that tα P ω ω : t Ď αu Ď Fu . This shows that Fu is rare, as
desired.
(ii) Suppose A is comeager on U , and pick s0 non-empty with Us0 “ tα P ω ω : s0 Ď
αu Ď U . Then Us0 zA is meager, so by (i) player II has a winning strategy τ in
G˚˚ pUs0 zAq. But then I wins by first playing s0 , and then applying the strategy
τ.
Ť
Conversely, if σ is a winning strategy for I in G˚˚ pAq, let s0 “ σpHq, and s0 “
tα P ω ω : s0 Ď αu. Then σ is a winning strategy for II in the game G˚˚ pUs0 zAq,
and by part (i) the set Us0 zA is meager. %
175
6 Infinite Games
The game G˚˚ pAq is very close to the Choquet games. In fact, if one restricts the
sequences played in the game to belong to the regular tree associated to ω ω zA, one
gets the Choquet game on ω ω zA. So the previous result is a kind of extension of
Theorem 6.2.5(i).
This result can be used, together with determinacy of games, to prove that sets in
certain classes have the Baire Property, e.g. from Martin’s Theorem 6.1.5, one gets that
all Borel sets are in BPpω ω q, and similarly with more complicated sets, like projective
ones, under determinacy hypotheses (see Moschovakis [Mos80]). Here we will use a
variant of this game to prove effective results.
Theorem 6.2.13. Let X be a Polish recursive space, and A Ď X be a meager Σ11 set.
Then there exists a ∆11 -recursive α0 P ω ω and a Π01 pα0 q relation
Ť x P Fn in X ˆ ω, such
that for each integer n, Fn is a closed rare set and A Ď nPω Fn .
Proof. We first prove the result if X “ rT0 s, where T0 is a ∆11 regular tree on ω, and
A Ď X is a Σ11 set meager in X. Let T1 be a recursive tree on ω ˆ ω with
I: s0 , m0 s2 , m1
¨¨¨
II : s1 s3
176
6.2 Baire spaces and the Baire property
For each pu, vq P S, let Fu,v “ tα P rT0 s : pu, vq is critical for αu. Then as in the proof
od the lemma above, each Fu,v is closed and rare in rT0 s.
Suppose now α P A. Then for some pu, vq P S, pu, vq is critical for α, that is
α P Fu,v . Otherwise, pick β P ω ω with pα, βq P rT1 s. As pH, βp0qq P S is not critical
for α, there is s0 Ď α such that the answer s1 “ σps0 , βp0qq satisfies s0 a s1 Ď α. As
pps0 , s1 q, β2 q P S is not critical for α, there is s2 with s0 a s1 a s2 Ď α such that the
answer s3 “ σps0 , βp0q, s2 , βp1qq satisfies s0 a s1 a s2 a s3 Ď α, and so on. One builds
inductively a run in the game where II follows σ, and which produces pα, βq. But then
this run is a loss for II, a contradiction.
The end of the proof is a simple coding. Define ps, u, vq P T if and only if s P T0 ,
u “ ps0 , . . . , s2k´1 q, v “ pm0,...,mk q, pu, vq P S, s and s0 a s1 a ¨ ¨ ¨ a s2k´1 are compatible,
and for all non-empty s1 P ω ăω with s0 a s1 a ¨ ¨ ¨ a s2k´1 a s1 Ď s, if there are any,
One checks that for each pu, vq P S, Tu,v is a tree on ω, and Fu,v “ rTu,v s, whereas if
pu, vq R S then Tu,v “ H. So let
”
pn, xm, pyq P T ˚ ÐÑ Dk lhpsm q “ 2k and lhpsp q “ k ` 1 and
ı
psn , pssm p0q , . . . , ssm p2k´1q q, sp q P T .
Then X ˚ is a dense Gδ set in X, is Π03 and is Π02 pα0 q for any α0 P ω ω in which DX is
recursive, as
Now we will use the fact that if X is a Polish space and G Ď X is a dense Gδ set in X,
then for any A Ď X, A is comeager in X if and only if G X A is comeager in G, and
hence A is meager in X if and only if G X A is meager in G. To see this, note that if
A is comeager in X, it contains a Gδ set G1 dense in X, and G1 X G is a dense Gδ of
G contained in G X A. Conversely if G1 is a dense Gδ of G contained in G X A, it is
also a dense Gδ in X contained in A.
So, if we start with a Σ11 set A meager in X, A X X ˚ is a Σ11 set which is meager
˚
in X ˚ , as VnX “ X ˚ X VnX “ X ˚ X VnX is clopen in X ˚ . So it is a homeomorphism
177
6 Infinite Games
x P i´1 ˚ ˚
X pFn q ÐÑ x P X ^ x P x P Fn .
Finally set #
n “ 2m and x P Fn˚ , or
x P Hn ÐÑ
n “ 2m ` 1 and x P VnX zVnX .
Clearly x P Hn is Π01 pα0 q, and A Ď nPω Hn . Finally the sets VmX zVmX are rare, and
Ť
so are the sets Fn˚ , for
Ø Y ˚ X Fn is meager in Y ˚
Ø Fn is rare in F. %
(i) If A is Σ0n
– tx : Ax is meager in Y u is Π0n , and
– tx : Ax is comeager in Y u is Π0n`1 .
Dually, for A is Π0n
– tx : Ax is meager in Y u is Π0n , and
– tx : Ax is comeager in Y u is Π0n`1 .
(ii) If A is Σ11
– tx : Ax is meager in Y u is Π11 , and
– tx : Ax is comeager in Y u is Σ11 .
Dually, for A is Π11
– tx : Ax is meager in Y u is Π11 , and
– tx : Ax is comeager in Y u is Σ11 ,
and similarly for the relativized results for Σ11 , Π11 and Σ11 pαq, Π11 pαq.
178
6.2 Baire spaces and the Baire property
Proof. (i) It is the effective analog of Theorem 6.2.10, and is proved similarly, by
induction on n. The only point is at the beginning for A P Π01 . One has to show
that the relation
VnY Ď Ax
is Π01 in X ˆ ω. Write
so VnY Ď Ax is Π01 in X ˆ ω. We skip the easy induction to get the result for all
positive integer n.
(ii) Let A be Σ11 in X ˆ Y , and let W Ď ω ω ˆ Y be a good universal for all Π01 subsets
of Y . Then by Theorem 6.2.13:
´
Ax is meager Ø Dα0 P ∆11 pxq Dβ0 P ∆11 pxq Wα0 ,β0 pnq is rare in Y for all integer n and
ď ¯
Ax Ď Wα0 ,β0 pnq .
nPω
One direction, from right to left, is obvious. For the other one, if α0 is given by
Theorem 6.2.13 relativized, and if y P Fn is the corresponding Π01 pα0 q sequence
of rare sets, one can use that W is a good universal to find a recursive in α0 ,
hence ∆11 pxq, β0 : ω Ñ ω with Fn “ Wα0 ,β0 pnq . Now the relation Wα,k is rare in
Π02 by (i), and
ď ` ˘
Ax Ď Wα0 ,β0 pnq Ø @y px, yq R A _ Dn py P Wα0 ,β0 pnq q
nPω
is also Π11 , and Ax is comeager in some non-empty open set in Y , that is:
is also Π11 . But then its complement Ax is meager is Σ11 , and finally by duality,
if A is Σ11 , Ax is comeager is Σ11 . %
179
6 Infinite Games
is a singleton txu by the property of the diameters, and so x is a ∆11 point in X. But
x P VnX0 , and x R Fn for each integer n. So x P A, as desired %
tα P ω ω : ω1α “ ω1 u
is comeager.
Proof. This set is Σ11 , by Proposition 5.3.5. If its complement were not meager, then
by Theorem 6.2.15, it would contain a ∆11 point α0 . But for ∆11 points, ω1α “ ω1 . This
contradiction finishes the proof. %
180
6.3 The perfect set property
Proof. By induction on lhpsq, s P 2ăω , one builds open sets pUs qsP2ăω in A with
Usa i Ď Us for i “ 0, 1, Usa 0 XUsa 1 “ H and diampUs q ď 2´ lhpsq , starting with UH “ A,
where diameter refers to some complete distance on X (hence on A). This is immediate
to do, by noticing that, since A is perfect, each Us contains at least two points. For
rendu explicite
l’utilisation de each α P 2ω , the sequence pUαn qnPω decreases to some point ϕpαq P A, by completeness
’hypothèse – K of d. Clearly, ϕ : 2ω Ñ A is continuous, for if αn “ βn , dpϕpαq, ϕpβqq ď 2´n . And if
α ‰ β, and n is the least integer with αpnq ‰ βpnq, then ϕpαq and ϕpβq are respectively
in the two disjoint open sets Uαn `1 and Uβn `1 , hence ϕpαq ‰ ϕpβq. %
The set N pAq can be obtained in another way, which was Cantor’s original method:
for each closed set A in X, let dpAq “ Az tx P A : x is isolated in Au. The set dpAq is
closed in X, so this derivation, called the Cantor-Bendixson derivation, can be iterated,
so that one gets a decreasing sequence pdξ pAqqξăω1 of closed sets in X. This sequence
must stabilize at some countable ordinal ξpAq. Then one has N pAq “ dξpAq pAq, for
clearly dξpAq pAq is perfect, hence dξpAq pAq Ď N pAq. On the other hand N pAq Ď A and
dpN pAqq “ N pAq, hence easily by induction N pAq Ď dξ pAq for all countable ξ.
Thus one gets:
A is countable ÐÑ Dξ ă ω1 dξ pAq “ H
as claimed in Subsection 5.5.14. And by the result in Subsection 5.5.14, one also
gets that Kω p2ω q “ tK Ď 2ω : K is closed countableu is Π11 on Kp2ω q, as claimed in
Subsection 5.5.7.
181
6 Infinite Games
tion 6.3.1 a non-empty perfect subset. So Subsection 6.3.2 says that any closed set
in X has the perfect set property. This was extended by Alexandroff [Ale16] to Borel
sets, and then by Suslin [Luz17] to all Σ11 sets. Here we will prove a deep effective
refinement of this result, due to Harrison [Har66].
Theorem 6.3.3 (The perfect set theorem). Let A Ď X be a Σ11 set. Then either A
consists of ∆11 points, or A contains a non-empty perfect subset. In particular, every
Σ11 set in X has the perfect set property.
Proof. We will se two proofs of this result. Here we give a proof based on the Gandy-
Harrington topology τ on X (see Subsection 6.2.6). Note that the τ -isolated points in (
X are exactly the Σ11 -singletons, i.e. the ∆11 points in X. Let X0 “ x P X : x P ∆11 .
This is a Π11 set, so if A is not contained in X0 , then AzX0 is a non-empty Σ11 subset of
X. Now there is a dense Gδ subset X ˚ of Xwhich is Polish for τ , by Subsection 6.2.6,
and pAzX0 q X X is a non-empty open subset of X ˚ which has no isolated point. As it
is Polish, Subsection 6.3.1 applies and gives a τ -closed subset F of pAzX0 q X X, hence
of A, homeomorphic to 2ω . But 2ω is compact, hence F is closed in X and we are
done.
For the last assertion, use the relativized version of the firs. As for any α, the set
tx P X : x P ∆11 pαqu is countable, we are done. %
182
6.3 The perfect set property
Now this relation is Π11 , hence there is a ∆11 -recursive g : B Ñ ω such that for
py, xq P B
py, gpy, xqq P D1 and d1 py, gpy, xqq “ x.
The set
A0 “ tpy, gpy, xqq : py, xq P Bu
is Σ11 , and contained in the Π11 set
Hence for some ∆11 set A2 one has A0 Ď A2 Ď A1 , by separation. Pick a ∆11 -
recursive uniformization f0 : DX B Ñ X for B, and set
I: s0 s1 s2
¨¨¨
II : n0 n1 n2
183
6 Infinite Games
As in Lemma 6.2.12, the previous lemma is not very interesting for Σ11 sets, as the
game G˚ pAq is then Σ11 . However, it is possible to unfold it. Let A be Σ11 in 2ω , and
T a recursive tree on 2 ˆ ω with
α P A Ø Dβ P ω ω @npαn , βn q P T.
Consider the following game G˚1 pAq
with si P ω ăω 2, ni P t0, 1u and βpiq P ω. Player I wins the game if for all integer i
´ ¯
a a a
ps0 n0 ¨ ¨ ¨ ni´1 qi , pβp0q, . . . , βpi ´ 1qq P T.
184
6.3 The perfect set property
This game is Π01 , hence determined. Moreover, if II wins it, he has a ∆11 winning
strategy.
Theorem 6.3.6. If I wins G˚1 pAq, A contains a non-empty perfect subset. If II wins
G˚1 pAq, every point in A is ∆11 . In particular, Theorem 6.3.3 for X “ 2ω follows.
Proof. If I wins G˚1 pAq, I also wins G˚ pAq, which is easier. So by Lemma 6.3.5, A
contains a non-empty perfect subset. Suppose now II wins by τ . As in Lemma 6.3.5,
say that for u “ ps0 , βp0q, n0 , . . . , sk´1 , βpk ´ 1q, nk´1 q and m P ω, pu, mq is critical for
α P 2ω if s0 a n0 a ¨ ¨ ¨ a sk´1 a nk´1 Ď α, and u has been played according to τ , but for no
extension ua psk , m, nk q played according to τ , s0 a n0 a ¨ ¨ ¨ a sk a nk Ď α. Again if α P A,
there must be some pu, mq critical for α, and for each pu, mq there is exactly one αu,m
for which it is critical, defined by αu,m N “ s0 a n0 a ¨ ¨ ¨ a sk´1 a nk´1 , where N is the
length of s0 a n0 a ¨ ¨ ¨ a sk´1 a nk´1 , and for M ě N by the recursion:
6.3.6 C1
One cannot prove in ZFC that all Π11 sets have the perfect set property. In fact, this
statement is equivalent to a mild large cardinal hypothesis. However, as every Π11
set in a Polish space is a union of ω1 Borel sets, one gets that if A is a Π11 set with
no non-empty perfect subset, A is of cardinality at most ℵ1 , as its Borel subsets are
countable.
If one looks only at Π11 sets, one has the following.
Theorem 6.3.7. Let X be a Polish recursive space. There is a largest Π11 subset of X
with no non-empty perfect subset. For X “ ω ω , this set is usually called C1 .
Proof. Let W Ď ω ˆX be Π11 and universal for Π11 subsets of X, and let ϕ be a Π11 -rank
on W . Consider
and let C1 “ tx P X : Dnpn, xq P Cu. We show that C1 is the set we are looking for.
First we show that any Π11 subset A Ď X with no non-empty subset is contained
in C1 . For pick n with A “ Wn . By the hypothesis on A and Theorem 6.3.3, the set
ty P X : ϕpn, yq ď ϕpn, xqu is a Borel subset of A, hence countable, for every x P A.
So A “ Cn too, and A Ď C1 .
Secondly, C, and hence C1 , is Π11 . For let R be the Σ11 relation
Then
pn, xq P C Ø pn, xq P W ^ Rn,x is countable.
185
6 Infinite Games
186
7 Borel Hierarchies
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
7.1.1 Σ0ξ -ranks on Σ0ξ
We defined the Borel hierarchy in Subsection 3.1.2, and proved the existence of (good)
universal sets for the classes Σ0ξ and Π0ξ in Subsections 3.3.5 and 3.3.9, from which the
strictness of the hierarchy follows in Theorem 3.3.4. Here, we pursue the study of the
structural properties of these classes. As for the class Π11 , most results come from the
existence of nice ranks.
Given a class Γ of sets (in recursive spaces), let us recall from Subsection 4.3.3 that
for a set A Ď X, an ordinal assignment ϕ : A Ñ On is a Γ-rank on A if the relations
` ˘
x P A ^ y R A _ ϕpxq ď ϕpyq
` ˘
and x P A ^ y R A ^ ϕpxq ă ϕpyq
are both in Γ. For Γ “ Π11 (or Π11 ), this is the definition we took for the notion of a
Π11 -rank.
Theorem 7.1.1. Let ξ ě 2. Then any Σ0ξ set A in some space X admits a Σ0ξ -rank
with values in ω.
The same is true for ξ “ 1, i.e. for open sets, if X is 0-dimensional.
Remarks 7.1.2. (1) It should be clear from the proof above that a similar result holds
for the lightface classes: Every Σ0n set admits a Σ0n -rank, for n ě 2, and for n “ 1
if X is dim0-recursive. One can even extend it to the Σ0ξ classes, ξ ă ω1CK as
defined in Subsection 5.2.2. One just needs the fact that Σ0ξ is closed under D0
and bounded quantifications @ď , which is easy to establish.
187
7 Borel Hierarchies
(2) The Σ0ξ -rank given in the above proof is not any Σ0ξ -rank: If we fix n P ω, the
set tx P A : ϕpxq ď nu is not only ∆0ξ but in fact – as it is mďn Am – it is a Π0η
Ť
set for some η ă ξ.
From the existence of nice ranks, we can infer, as in Subsections 4.3.4 to 4.3.7,
reduction and separation results.
(ii) (reduction) If A0 and A1 are two Σ0ξ sets, there are two disjoint Σ0ξ sets B0 and
B1 with B0 Ď A0 , B1 Ď A1 and B0 Y B1 “ A0 Y A1 .
(iii) (separation of Π0ξ sets) If A0 , A1 are two disjoint Π0ξ sets, there is a ∆0ξ set B
which separates A0 from A1 , i.e. such that A0 Ď B and B X A1 “ H.
(iv) There are in ω ω two disjoint Σ0ξ sets which cannot be separated by a ∆0ξ set.
Proof. It is just an adaptation of the proof in Subsections 4.3.4 to 4.3.7, and is left to
the reader. %
One may think that boundedness theorems also extend to this context. But note
that the results in Section 4.4 used not only the existence of Γ-ranks (for Γ “ Π11 ), but
also the closure of the dual class (Σ11 ) under DX . Now for ξ ě 2, the classes Π0ξ are not
closed under DX , even in the case where X “ ω. And in fact there are no interesting
boundedness results. For ξ “ 1, there is one case where we can apply the methods of
Section 4.4: if X is a compact dim0-recursive space, then Π01 is closed under DX – as
projections of compact sets in X ˆ X are compact. But the boundedness result one
gets from this, that given a Σ01 -rank ϕ on a Σ01 set A Ď X, and a Π01 subset B of A,
the rank ϕ is bounded on B, is trivial – this is just the compactness of the Π01 set B.
7.1.2 Differences
Any ordinal ξ can be written in a unique way as ξ “ λ ` n, with λ limit (or 0) and
n P ω. By definition the parity, parpξq, of ξ is the parity of the corresponding integer
(so limit ordinals are even).
Let xA ˘ an increasing sequence of sets, with ξ ě 1. We define a set
` η : η ă ξy be
A “ Dξ xAη : η ă ξy , the difference set of the sequence xAη : η ă ξy, by
188
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
So
D1 pxA0 yq “ A0
D2 pxA0 , A1 yq “ A1 zA0
ď
D3 pxA0 , A1 , A2 yq “ pA2 zA1 q A0
¨¨¨
` ˘ ď
Dω xAn : n P ωy “ pA2n`1 zA2n q
n
` ˘ ´ ď ¯ ď
Dω`1 xAn : n ď ωy “ Aω z An Y pA2n`2 zA2n`1 q Y A0
n n
etc.
` ˘
If Γ is any family of sets, we denote as usual by Dξ pΓq the family of all Dξ xAη : η ă ξy ,
where xAη : η ă ξy is an increasing sequence of length ξ of sets in Γ.
In the sequel, we will be mainly interested in the classes Dξ pΣ0ξ q, ξ, η are countable
ordinals (ě 1), and their dual classes Ďη pΣ0ξ q “ tA : A P Dη pΣ0ξ qu, in recursive
spaces. Note that (i) these classes are preserved under continuous preimages, and
(ii) if X is a recursive subspace of Y , a set is Dη pΣ0ξ q in X iff it is the trace on X of a
@ D
Dη pΣ0ξ q subset of Y . This is almost immediate. The only point is that if Aη1 : η 1 ă η
is an increasing sequence of Σ0ξ sets in X, and Bη1 is Σ0ξ in Y with Aη1 “ X X Bη1 , then
the sets Bη‹1 “ η2 ďη1 Bη2 are Σ0ξ in Y , increasing, and with Aη1 “ X X Bη‹1 . Fact (ii)
Ť
easily follows from this.
There is another standard way of introducing differences, as done in Kuratowski’s
book, by considering decreasing sequences of sets xBη : η ă ξy and the set
ref to Kuratowski’s
˚
` ˘ ď ` ˘ book – YP
Dξ xBη : η ă ξy “ Bη zBη`1
η even
ηăξ
So that we also get, for any class Γ with dual class Γˇ the formulas:
#
Dξ pΓq if ξ is even,
D˚ξ pΓˇq “
Ďξ pΓq if ξ is odd,
189
7 Borel Hierarchies
which allow to correlate Kuratowski’s presentation with ours. Our choice of Dξ instead
of D˚ξ is motivated mainly by the fact that all the Dη pΣ0ξ q classes share the same
structural properties, as we will see in ref missing.
?? – YP
F0 “ A0
F1 “ A0 X A1
F2 “ A0 X A1 X A0
etc.
The sequence pFξ qξPOn is decreasing, hence there exists a leastŞordinal ξ “ ξpA0 , A1 q,
the Hausdorff ordinal of pA0 , A1 q which satisfies Fξ “ Fξ`1 “ η Fη , and ξpA0 , A1 q ă
ω1 . Set F8 pA0 , A1 q “ FξpA0 ,A1 q pA0 , A1 q and Gξ pA0 , A1 q “ XzFξ pA0 , A1 q, G8 pA0 , A1 q “
XzF8 pA0 , A1 q.
(i) F8 pA0 , A1 q is the largest closed subset F of X with the property that A0 X F and
A1 X F are both dense in F .
(ii) G8 pA0 , A1 q is the largest open subset U of X such that for some ξ ă ω1 there
is a Dξ pΣ01 q set C which separates U X Aparpξq from U X A1´parpξq , i.e. satisfies
U X Aparpξq Ď C and U X A1´parpξq “ H. Moreover, ξpA0 , A1 q is the least ordinal
ξ for which the preceding statement holds for G8 pA0 , A1 q.
Ş
Proof. (i) Fix ξ ă ω1 . As Fξ “ Aparpξq X ηăξ Fη , we get
č
Aparpξq X Fξ Ě Aparpξq X Fη .
ηăξ
Ş
Hence in fact Aparpξq X Fξ “ Aparpξq X ηăξ Fη is dense in Fξ .
So in particular both Aparpξq and A1´parpξq are dense in Fξ for ξ ě ξpA0 , A1 q, i.e.
in F8 pA0 , A1 q.
190
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
(i) there exists ξ ă ω1 and a Dξ pΣ01 q set C which separates Aparpξq from A1´parpξq .
Moreover the least such ξ is ξpA0 , A1 q.
Proof. (i) By the above proposition, it is enough to prove that if F is closed and
non-empty, F X A0 and F X A1 cannot be both dense in F . But as X is Polish,
F is a Baire space, and F X A0 and F X A1 are Gδ in F . So as they are disjoint,
they cannot be both dense.
(ii) If A is ∆02 , we can apply (i) to A0 “ A and A1 “ XzA. We get that for
ξ “ ξpA0 , A1 q, some Dξ pΣ01 q set C separates Aparpξq from A1´parpξq “ XzAparpξq ,
191
7 Borel Hierarchies
Finally define f : X Ñ ω ω by
192
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
hence π ´1 pAn,p q is clopen in F “ Graphpf q for all n, p and π ´1 pAn q is then Σ01 in F
for all n.
It remains to show that F is closed in X ˆ ω ω . So suppose xi Ñ x in X, and
f pxi q Ñ α in ω ω . Then for all xn, py the sequence f pxi qpxn, pyq is eventually constant,
equal to αpxn, pyq. But then all but finitely many xi ’s are in An,p or in some An,p k ,
which are closed. Hence x is also in this set, i.e. f pxqpxn, pyq “ αpxn, pyq. As this
holds for all xn, py, f pxq “ α, i.e. px, αq P F and F is closed.
In case X is not 0-dimensional, Ť the proof is slightlyn,pmore complicated. Set now
B n,p “ XzAn,p and write B n,p “ k An,p k , where the Ak ’s are closed, and increasing
in k. Set Bk “ Ak zAk´1 (with An,p
n,p n,p n,p
´1 “Ť H). Each Bkn,p is now D2 pΣ01 q, hence
in particular Σ02 , and we can write Bkn,p “ l An,p n,p
k,l with Ak,l closed and increasing,
n,p
and set Bk,l “ An,p n,p
k,l zAk,l´1 , etc. So inductively we build As
n,p
and Bsn,p for s P ω ăω ,
n,p n,p n,p
` n,pA˘s closed and
with Bs P D2 pΣ01 q such that Bs is the increasing union of the sets
Asa k kPω , and Bsk “ An,p
n,p
sa k
zAn,p
sa k´1
(with the convention An,p sa ´1
“ H). Note that
n,p
for lhpsq “ l, the sets Bs are pairwise disjoint, and have XzAn,p as union. Also by
construction we have Bsn,p n,p n,p
a k Ď As a k Ď B s .
Define now f : X Ñ ω ω by
where
$
&0
’ if x P An,p ,
fn,p,l pxq “ k ` 1 if x R An,p and sk is the unique element of ω ăω
such that lh sk “ l ` 1 and x P Bsn,p
’
k .
%
´1
First f is Baire class 1. To see this, note that fn,p,l p0q “ An,p is Π01 and that
#
´1 H if lhpsk q ‰ l ` 1,
fn,p,l pk ` 1q “
Bsn,p
k if lhpsk q “ l ` 1
is D2 pΣ01 q. So f is Baire class 1 as desired. (ii) is clear as before. We now prove that
F “ Graphpf q is closed in X ˆ ω ω . Let xi Ñ x and f pxi q Ñ α. We want to prove
that for all n, p, l fn,p,l pxq “ αpxn, p, lyq. As before, this is clear if αpxn, p, lyq “ 0.
So suppose that for some s of length l ` 1 αpxn, p, lyq “ xsy ` 1, and hence all but
finitely many points xi are in Bsn,p . We want to infer that x P Bsn,p too. Now note that
Bsn,p X An,p “ H, hence x R An,p . So if we consider αpxn, p, l ` 1yq, its value cannot
be 0, for otherwise all but finitely many xi ’s, hence x, would be in An,p . So for some
t of length l ` 2, αpxn, p, l ` 1yq “ xty ` 1, and all but finitely many xi ’s are in Btn,p .
But then Btn,p X Bsn,p ‰ H, hence for some k t “ sa k. But then x P Btn,p Ď Bsn,p and
fn,p,l pxq “ xsy ` 1, as desired. This finishes the proof for ξ “ 1.
Note that by this proof, we may assume that F is 0-dimensional, by putting, among
the sets An,p , the sets XzVnX , for n P ω.
193
7 Borel Hierarchies
Let now ξ ě 2, and assume the result is known for all η, 1 ď η ă ξ. By the
hypothesis each An is Σ1`ξ , XzAn “ p An,p where An,p is Π01`ηpn,pq , for some ηpn, pq
0
Ť
So for all n, p A˚n,p “ tpx, gpxqq : x P An,p u is Π01 in G. We now apply case ξ “ 1, for
the space G and the sequence A˚n,p of Π01 , with complements written as the disjoint
union of the clopen sets A˚n,p,k .
By the above proof, there is a Baire class 1 function h : G Ñ ω ω with H “ Graphphq
closed in G ˆ ω ω , and such that for all xn, py, the function hxn,py : G Ñ ω defined by
hxn,py px, αq “ hpx, αqpxn, pyq satisfies
h´1 ˚
xn,py p0q “An,p
´1
and hxn,py pk ` 1q “A˚n,p,k .
Moreover
tpx, αq P F : x P An,p u “ tpx, αq P F : pαq1 pxn, pyq “ 0u
is ∆01 in F , hence for each n tpx, αq : x P An u is Σ01 as desired. It remains to prove
(iii). So fix k, and consider fk pxq “ f pxqpkq. If k is even, say k “ 2 xn, py, then
fk pxq “ gxn,py pxq, which is by construction of class ηpn, pq. So for each i P ω fk´1 piq is
∆01`ηpkq with ηpkq ă ξ. If k is odd, say k “ 2 xn, py`1, then fk pxq “ hxn,py px, gpxqq and
by the choice of h fk´1 p0q “ An,p is Π01`ηpn,pq , whereas fk´1 pi ` 1q “ An,p,i is ∆01`ηpn,pq .
This proves (iii), and finishes the proof. %
194
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
We claim that ϕ is a Dη pΣ0ξ q-rank on A. To see this, define for every η 1 ă η a Σ0ξ set
Bη1 by
– if η 1 R P , px, yq P Bη1 ÐÑ x P Aη 1 ,
195
7 Borel Hierarchies
– if η 1 P P ,
ď
px, yq P Bη1 ÐÑ x P Aη2 _
η 2 ăη 1
˜ ˆ ˙¸
ď
Dn ηn “ η 1 ^ x P Aηn ,kn ^ @m ă n y R Aηm ,km z Aη 2
η 2 ăηm
One easily checks that these sets are Σ0ξ . Moreover for all η 1 ă η
ď
pAη2 ˆ Xq Ď Bη1 Ď Aη1 ˆ X,
η 2 ăη 1
As before, one easily gets from the existence of Dη pΣ0ξ q-ranks the property of
reduction for Dη pΣ0ξ q sets, and the separation property for Ďη pΣ0ξ q sets. The ω-
uniformization property is less clear (in Theorem 4.3.5, the proof uses closure under
finite intersections!), but still holds, by a better choice the the above ranks (depending
in fact under @ă –
YP on the parity and finiteness or non-finiteness of η). For this and related results, we
refer the reader to Louveau–Saint Raymond ??.
ref missing – YP
196
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
at least for ξ ě 2 or in dim0 recursive spaces, this is the same as saying that there is a
family pAn qn of Σ0ξ sets covering A such that A X An is in Γ in the space An – for by
reduction such sets can always be supposed disjoint.
Given now λ a limit ordinal ă ω1 , define a sequence SUλξ , for ξ ă ω1 , as follows.
First recall that any ordinal ξ ă ω1 can be written uniquely as ξ “ λ ¨ ξ0 ` Ťξ1 , with
ξ0 ă ω1 and ξ1 ă λ. We define SUλξ by induction on ξ0 . If ξ0 “ 0, let Γ0 “ ηăλ Σ0η ,
and set SUξ1 “ SU1`ξ1 pΓ0 q. Suppose SUλξ has been defined for ξ0 ă η, ξ1 ă λ. Let
then Γη “ ξ0 ăη SUλξ and set SUλ¨η`ξ1 “ SU1`ξ1 pΓη q. Clearly ξăω1 SUλξ is the
Ť Ť
least family containing all Σ0ξ , ξ ă λ, and closed under all operations SUξ , ξ ă λ.
Theorem 7.1.9 (Wadge). Let X be a Polish recursive space. Then for each λ ă ω1
really? – YP
ď
∆0λ “ SUλξ .
ξăω1
` ˘
Proof. If ξ ă λ, then Σ0ξ Ď ∆0λ . Also if A “ SU pAn qn , pBn qn with An P Σ0ξ for some
ξ ă λ and Bn P ∆0λ , then An X Bn P ∆0λ , hence A is Σ0λ . But also
´ ď ¯ ` ˘
XzA “ Xz An Y SU pAn qn , pXzBn qn
n
is in Σ0λ , hence A P ∆0λ . This proves that ξăω1 SUλξ Ď ∆0λ (in any recursive space).
Ť
separable metrizable?
For the converse, we will use Theorem 7.1.6. So let A be ∆0λ in X Polish, and apply – YP
Theorem 7.1.6 to A and its complement. We get a function f : X Ñ ω ω such that
F “ Graphpf q is closed in X ˆ ω ω , A˚ “ tpx, f pxqq : x P Au is ∆01 in F , and for each
k there is an ηk ă λ such that for all i P ω, tx P X : f pxqpkq “ iu is ∆0ηk `1 . And
without loss of generality, we may assume that Vn˚ “ pVnX ˆ ω ω q X F is clopen in F .
Let d be a complete distance on F , and define a tree S on ω ˆ ω by induction on the
length of ps, tq P ω ăω by pH, Hq P S, and if s “ pn0 , . . . , nk´1 q, t “ pm0 , . . . , mk´1 q
are such that ps, tq P S, then
The main point is that the tree S is wellfounded. Ş This because any branch pα, βq
through S defines a unique point px, βq P F with x P n Vαpnq X . Now pα, βq is either in
A˚ or in F zA˚ , which are both open, hence for some n the set Vαpnq ˚ X tpx, γq : γn “
˚ ˚
βn u is included either in A or in F zA , contradicting the fact that pαn`1 , βn`1 q P S.
View S as a subtree of
˚ ˚ ˚
(
T “ ps, tq : @i ă lh sVspiq Ď Vspi´1q and diampVspiq q ď 2´i .
197
7 Borel Hierarchies
then for some ξ ă ω1 , A X As,t is in SUλξ in As,t . This is clearly true if ps, tq R S, for
then on As,t , A is either H or As,t . Now as S is wellfounded, it is enough to show that
if ps, tq P S and for all n, m with psa n, ta mq P T there is ξn,m ă λ such that A is in
SUλξn,m onŤAsa n,ta m , then for some ξ ă λ, A is in SUλξ on As,t . Let k “ lh s. Now
note that psa n,ta mqPT Asa n,ta m “ As,t , and
is by our choice of f , Σ0ηk `1 in As,t . So if ξ0 is minimal with λξ0 ą ξn,m for all
psa n, ta mq P T , we get that A is in SUλλ¨ξ0 `ηk `1 on As,t This proves the claim. And
applying this claim to pH, Hq yields the result. %
One can again prove for the classes SUλξ structural properties analogous to those of
the classes Dη pΣ0ξ q. We refer the reader to Louveau–Saint Raymond ??.
réf – YP
One can also ask: Can one get similar results for the classes
is strict on the space ω, where the Borel hierarchy collapses to ∆01 – as the topology
is discrete. So there seems to be no relationship between the two hierarchies. This is
true at the recursive level, but not at all at the ∆11 level.
Let us define, in any recursive space X, for ξ ă ω1CK :
ď
Σ0ξ p∆11 q “ Σ0ξ pαq,
αP∆11
(i) If A0 , A1 are two disjoint Σ11 sets and there is a Σ0ξ set separating A0 from A1 ,
then there is a set C in Σ0ξ p∆11 q separating A0 from A1 .
In particular,
198
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
(‹‹) If A Ď X is Σ0ξ , there exists a τξ -open set A˚ such that A4A˚ “ pAzA˚ qYpA˚ zAq
is τ -meager.
Case ξ “ 1: (‹‹) is trivial in this case, as we can take A˚ “ A. For (‹), one has
1
Dn x P VnX ^ VnX X A “ H
` ˘
xRA ÐÑ
which is clearly Π11 . So we prove (i). Note that the hypothesis on A0 , A1 just
1 ` ˘
means A0 X A1 “ H, i.e. for each x P A0 , Dn x P VnX ^ VnX X A “ H . Hence
by Theorem 4.3.5 there exists a ∆11 -recursive f : X Ñ ω such that for every
x P A0 , x P VfXpxq ^ VfXpxq X A1 “ H. So the set B0 “ tn : Dx P A0 n “ f pxqu
is Σ11 , and contained in B1 “ tn : VnX X A1 “ Hu, which is Π11 . By separation
(Theorem 4.3.8), there exists a ∆11 set C with B0 Ď C Ď B1 .
Define now
#
0 if p ‰ 1,
αn ppq “ if n R C, and
1 if p “ 1,
#
0 if p ‰ n ` 2,
αn ppq “ if n P C,
1 if p “ n ` 2,
and set α “ 2a xαn : n P ωy. By definitionŤof a Borel code, we get that α P BCΣ1,
and clearly α is ∆11 -recursive. So BαX “ nPC VnX is Σ01 p∆11 q, and separates A0
from A1 .
General case: We assume (i), (‹) and (‹‹) for all 1 ď η ă ξ, and prove these statements
for ξ.
199
7 Borel Hierarchies
ξ
(a) Proof of (‹). Let A Ď X be Σ11 . Note that if x R A , there exists 1 ď η ă ξ
and a Σ11 X Π0η set B such that x P B and A X B “ H. But then A
is separable from B by a Σ0η set (namely XzB), hence by the induction
hypothesis there exists a Σ0η p∆11 q set C with A Ď C and B X C “ H, so
that x R C. Conversely if there is a Σ0η p∆11 q set C with A Ď C and x R C,
ξ
as C is also in Σ0η we clearly have x R A . So we get, if α0 is a recursive
code for ξ, α0 P WO,
´ ¯
ξ
xRA ÐÑ Dn Dα P ∆11 α P BCΠ |α0 n| ^ A Ď|
B X
α ^ x R B X
α
ξ
which gives a Π11 definition of XzA . This proves (‹).
(b) Proof of (‹‹). Let A Ď X be Σ0ξ , and write A “ n An , where for some
Ť
Each set An zA˚n is τ -meager, so it is enough to study A˚n zA˚ . Now A˚n is
τηn -closed, and A˚ is τξ -open. As τηn Ď τξ Ď τ , A˚n zA˚ is τ -closed, and the
proof will be complete if we can show that it is τ -rare.
So let B be Σ11 , B Ď A˚n zA˚ , towards proving B “ H. Consider the τηn -
ηn
closure B of B. By the induction hypothesis for (‹), it is a Σ11 set, and as
clearly τηn -open sets are Σ0ηn , it is also a Π0ηn set. Also as A˚n is τηn -closed,
ηn ηn
B Ď A˚n , hence B zA Ď A˚n zA Ď A˚n zAn is τ -meager. By definition of
ηn
A˚ , it follows that B Ď A˚ , and a fortiori B Ď A˚ . But by hypothesis
B Ď A˚n zA˚ . So B “ H, and (‹‹) is proved.
(c) Proof of (i). Let A0 , A1 be Σ11 sets in X and A some Σ0ξ set with A0 Ď A
and A1 X A “ H. Using (‹‹), let A˚ be τξ -open with A˚ 4 A τ -meager. As
A1 X A “ H, A1 X A˚ is τ -meager. But A1 is Σ11 , and A˚ is τ -open, hence
A1 XA˚ is τ -open. By the Baire category theorem for τ (see Subsections 6.2.2
and 6.2.6), A1 X A˚ is empty.
ξ ξ
As A˚ is τξ -open, it follows that A1 XA˚ “ H, and hence A1 XA0 Ď AzA˚
ξ ξ
is τ -meager. Now A1 is Σ11 by (‹), hence A1 X A0 “ H, again by the Baire
category theorem for the Gandy–Harrington topology.
The end of the proof is now as in case ξ “ 1: For each x P A0 there is α P ∆11
and n such that, if α0 a recursive code for ξ,α P BCΠ X
|α0 n| ^Bα XA1 “ H^x P
|
BαX .
200
7.1 The Borel hierarchy Σ0ξ , Π0ξ , ∆0ξ
Let pD1 , d1 q be the Π11 coding of ∆11 points in ω ω given by Theorem 5.1.3,
and note that
” ı
@x P A0 Dm P D1 Dnd1 pnq P BCΠ |α0 n|
|
^ B X
d1 pnq X A 1 “ H ^ x P B X
d1 pnq .
B1 “ tm P D1 : Dn d1 pmq P BCΠ X
|α0 n| ^ Bd1 pmq X A1 “ Hu.
|
clearly separates A0 from A1 , so (i) is proved. This finishes the proof of the
theorem. %
The preceding theorem is not specific to the Borel classes. Once lightface versions of
the classes Dη pΣ0ξ q and SUλη (for η, ξ, λ ă ω1CK ) have been defined, analogous results
are true for these classes. The same results can even be obtained for the more general
Wadge classes, that we will study in Section 7.3. See Louveau [??].
ref missing – YP
201
7 Borel Hierarchies
(ii) If A0 , A1 are Σ11 (resp. Σ11 ) in X ˆ Y , and ξ ă ω1CK (resp. ξ ă ω1 ) is such that
every section pA0 qx of A0 is separable from the section pA1 qx by a Σ0ξ set, then
there is some ∆11 (resp. Borel) set B with Σ0ξ sections such that A0 Ď B and
A1 X B “ H.
Another corollary deals with a structural characterization of Borel sets with Σ0ξ
sections. As for the previous (boldface) corollary, case ξ “ 1 is due to Azsenin and
Kunugui, ξ “ 2 to Saint Raymond and ξ “ 3 to Bourgain (see Louveau [??]).
ref missing – YP
Let X, Y be Polish recursive spaces, and define sequences of classes pBξΣ , BξΠ qξăω1 in
X ˆ Y , by
BξΠ “ BξΣ .
Theorem 7.1.12 (Louveau). Let X, Y be Polish recursive spaces. The following are
equivalent for A Ď X ˆ Y and 1 ď ξ ă ω1 :
(ii) A P BξΣ .
(iii) There is a Polish topology on X finer than the original one, such that A is Σ0ξ
in X ˆ Y , where X is equipped with this topology and X ˆ Y with the product
topology.
Proof. (iii)Ñ(i): The identity from X ˆ Y with the new topology to X ˆ Y with the
old one is continuous, and one-to-one, so the image of the Σ0ξ set A is Borel in
X ˆ Y with the original topology. And as the topology on Y is the same, the
sections of A are still Σ0ξ .
202
7.2 Complements on low-level Borel classes
(ii)Ñ(iii): Clearly each set A in BξΣ is built up as a Σ0ξ set, starting from countably
many sets of the form Bn ˆ VnY , with Bn Borel in X. So it is enough to show
that for every sequence pBn qnPω of Borel sets in X there is a finer Polish topology
on X for which each Bn is clopen. To see this let η ă ω1 be such that each Bn
is ∆01`η in X and apply Theorem 7.1.6 to the sets Bn and their complements.
One gets a Borel (class η) function f : X Ñ ω ω with closed graph F and such
that π ´1 pBn q is clopen in F for each n. The topology on X obtained by transfer
remark that the
from that of F is then Polish, and each Bn is ∆01 for it, as desired. projection π : F Ñ X
is a bijection and F
(i)Ñ(ii): By induction on ξ, it is clearly enough to prove that any Borel set A Ď X ˆ Y is Polish! – YP
with Σ0ξ sections is a countable union of Borel sets An with sections in Π0ηn
for some ηn ă ξ (resp., for ξ “ 1 with sections of the form VnY ). And by
relativization, it is enough to prove the result when A is ∆11 in X ˆ Y , and
ξ ă ω1CK . By the relativized version of Theorem 7.1.10, one then gets
As the relation in parentheses is Π11 , one gets by the ∆11 -Uniformization The-
orem 5.1.6 a ∆11 -recursive function f : X Ñ ω ω such that for all x P X,
f pxq P BCΣ X
ξ ^ Ax “ Bf pxq . Now for ξ “ 1 set
Y
px, yq P An ÐÑ y P Bpf pxq` qn ,
One immediately
Ť checks that each An is ∆11 with Π0ηn sections for ηn “ |α0 |n|,
and A “ n An , as desired. %
203
7 Borel Hierarchies
In fact, we have already presented the main idea of these results when we discussed
in Subsections 3.5.2 to 3.5.5 the characterization of Polish recursive spaces: We proved
in Subsection 3.5.4 that if X is a non-empty closed subset of a Polish recursively
presented space and its diagram DX “ tn : VnX ‰ Hu is Σ01 , then X admits a recursive
presentation with the induced distance. And the main point in the proof is to show,
under the hypotheses on X, that the set of recursive points in X is dense. Relativizing
this to the ∆11 level will eventually give the Kuratowski–Ryll–Nardjewski Theorem.
Lemma 7.2.1. Let X be Polish recursive, and A Ď X a closed subset. Then the
following are equivalent:
(i) A is ∆11 , and tx P A : x P ∆11 u is dense in A,
Dx P X x P A ^ x P VnX
` ˘
n P DA ÐÑ
Dx P ∆11 x P A ^ x P VnX ,
` ˘
and n P DA ÐÑ
x P A ÐÑ @n x P VnX Ñ n P DA ,
` ˘
hence A is ∆11 too. For the second statement, we use the relativized version of
Subsection 3.5.4: Let α0 be the characteristic function of DA . Then α0 is ∆11 in
ω ω , and DA is Σ01 pα0 q. By Subsection 3.5.4, tx P A : x P ∆01 pα0 qu is dense in A.
As ∆01 pα0 q Ď ∆11 , we are done. %
204
7.2 Complements on low-level Borel classes
Φ´1 pU q “ tx P E : Φpxq X U ‰ Hu
Now notice that if α “ f pyq, then A˚α “ Φpyq. This is because if f pyq “ f py 1 q, then
tn : y P En u “ tn : y 1 P En u, i.e. tn : Φpyq X VnX ‰ Hu “ tn : Φpy 1 q X VnX ‰ Hu, and
hence Φpyq “ Φpy 1 q as Φ has values in the the closed subsets of X.
So A˚α is closed, and moreover
ÐÑ Dy P En f pyq “ α
ÐÑ αpnq “ 1.
205
7 Borel Hierarchies
Note that we could replace the space X by VnX in the preceding considerations, for
each n P ω, and get a Borel uniformizing function on En for Φn pxq “ Φpxq X VnX .
This gives, under the Ťsame hypotheses, the existence of a sequence of E-measurable
functions ϕn on E 1 “ n En such that for each x P E 1 the set tϕn pxq : n P ωu is dense
in Φpxq, a result due to Castaing.
is ∆11 . Suppose that for each y the set Ay is comeager in Ay (this is always satisfied if
the sections are Π02 ).
Then DX A is ∆11 in Y and A can be uniformized on DX A by a ∆11 -recursive function.
again,
∆11 -measurable, or Proof. We prove the case where A and the relation y P An are ∆11 , the relativized result
simply bianalytic
being similar. And again, it is enough to show (by relativization and Theorem 5.1.6)
function – YP
that if a non-empty set B Ď X is ∆11 , has a ∆11 diagram DB “ tn : B X VnX ‰ Hu, and
is comeager in its closure, then it has a ∆11 member. As B X VnX ‰ H Ø B X VnX ‰ H,
DB “ DB is ∆11 , hence B is Polish recursive in some ∆11 point α0 by Subsection 3.5.4.
And as B is ∆11 and comeager in B, Theorem 6.2.15 gives a ∆11 point in B. %
206
7.2 Complements on low-level Borel classes
Proof. First we show that there exists a ∆11 point α0 P ω ω such that XzK “ X
Ť
n Vα0 pnq
and VαX0 pnq X K “ H for all n (We assume here that VnX0 “ H for some n0 , which is no
real restriction).
Hu. As K is ∆11 , A is Π11 , and nPA VnX “ XzK. Now
Ť
Consider A “ tn : VnX X K “Ť
the property, for a set C that nPAXC VnX “ŤXzK is Π11 on Π11 , hence by reflection
(Theorem 4.6.1) there is a ∆11 set B Ď A with nPB VnX “ XzK. We can set α0 pnq “ n
if n P B, and n0 otherwise.Ť
Let P “ tn P ω : K Ď iălh sn VsX n piq
u. The set P is a Π11 subset of ω. Define then
Q Ď ω 3 by
pp, q, nq P Q ÐÑ p P P ^ q P P ^ @i ă lh sq diam VsX
` ˘
q piq
ď 2´n
X
` ˘
^ @i ă lh sq Dj ă lh sp VsX q piq Ď V sp pjq
^ @i ă lh sq VsX X
` ˘
q piq
X V α0 pnq “ H .
207
7 Borel Hierarchies
Theorem 7.2.6. Let X be Polish recursive, and K Ď X compact. Then the following
are equivalent:
(i) K is ∆11 in X,
(ii) DK “ tn P ω : K X VnX ‰ Hu is ∆11 in ω.
In particular every non-empty ∆11 compact set has a ∆11 member.
Proof. By Lemma 7.2.1, if (ii) holds then K is ∆11 in X and if non-empty has a ∆11
member. So it remains to show (i)Ñ(ii).
Let T be associated to K by the preceding lemma. Note that by Lemma 7.2.1 it is
enough to show that the ∆11 points are dense in K. Moreover if K 1 is any compact and
∆11 set, then so is K 1 X VnX for every n. It is therefore enough to show that if K ‰ H,
then it has a ∆11 member. NowŞif K is non-empty, the tree T has a branch. And if α is
a ∆11 branch of T , then txu “ n VαpnqX is a ∆1 -singleton, hence x is a ∆1 point which
1 1
belongs to K. So it is enough to show that if T is a ∆11 tree on ω which is finitely
branching and infinite, it has a ∆11 branch – the effective version of König’s lemma.
To see this define
nPP ÐÑ sn P T ^ tt : sn a t P T u is infinite.
The set P is ∆11 in ω, and by our hypotheses 0 P P (s0 “ H P T and T is infinite),
and clearly for each n P P there is m with xsn a my P P . By ω-uniformization (The-
Alain says
∆11 -uniformization, orem 4.3.5), there is a ∆11 -recursive f : ω Ñ ω such that @n P P xsn a f pnqy P P .
but? – YP But then α defined inductively by αpnq “ f pxαn yq is ∆11 and a branch through T , as
desired. %
Corollary 7.2.7. Let X, Y be Polish recursive, and A Ď X ˆ Y a ∆11 (resp. Borel) set
with compact sections. Then DY A is ∆11 (resp. Borel) in X and A can be uniformized
by a ∆11 -recursive function (resp. Borel) function.
Proof. Use the previous basis theorem and the ∆11 -Uniformization Theorem 5.1.6. %
208
7.2 Complements on low-level Borel classes
Lemma 7.2.8. Let X be Polish recursive, and A Ď X be a ∆11 set and σ-compact set.
Ť
(i) There is a sequence pAn qnPω of compact sets such that A “ n An , and moreover
the relation x P An is ∆11 in X ˆ ω (so in particular each An is ∆11 ).
Proof. (i) As A is Kσ and ∆11 , A is Σ02 and ∆11 , hence is Σ02 p∆11 q by Theorem 7.1.10,
and hence is a ∆11 union of Π01 p∆11 q sets. However, this is not enough to conclude,
for these Π01 p∆11 q sets may not be compact in X (although they are certainly Kσ
again). To overcome this difficulty, note that by Theorem 3.5.7 X admits a
completion X p which is Polish recursive and compact. Now being compact, resp.
Kσ , in X is the same as being closed, resp. Σ02 , in X. p And as X is Π0 in X,
2
p
1
being ∆1 in X and being ∆1 in X1 p are equivalent for a subset of X. So A is Σ0
2
and ∆11 in X,p hence by Theorem 7.1.10 there is a sequence An of Π0 sets in X
1
p
with A “ n An and x P An a ∆11 relation in X
Ť p ˆ ω. And these An ’s do work.
(ii) If A is non-empty, one of the An ’s must be non-empty too, and Lemma 7.2.5
applies to give a ∆11 member of An , hence of A. %
As usual, these results can be turned into boldface results in product spaces.
Theorem 7.2.9. Let X, Y be Polish spaces, and B Ď X ˆY a Borel set with σ-compact
sections.
We now turn to two results which are specific to the class Π02 (in Polish spaces), i.e.
to the property of being Polish. The first result, due to Hurewicz, is a characterization
of Polish spaces among co-Suslin spaces. The second result, due to Saint Raymond,
gives a sufficient condition for the continuous image of a Polish space (i.e. a Suslin
space) to still be Polish.
209
7 Borel Hierarchies
The preceding result is asymmetric. In fact, it cannot be proved in ZFC that any
Suslin space which is not Polish contains a relatively closed subset homeomorphic to
D0 . (Note that this statement easily implies the perfect set property for Π11 sets, which
cannot be proved in ZFC either).
Hurewicz’ theorem gives a characterization, among all Π11 sets (in Polish spaces),
of the Borel class Π02 . To get a generalization to other Borel classes, we need, by the
preceding discussion, to restrict ourselves to Borel sets. We will see in Subsection 7.3.10
verify this ref – YP
that with this restriction, there are results analogous to Hurewicz’ theorem at all levels
of the Borel hierarchy.
Proof. We will use here Theorem 6.2.5. By this result, it is enough to describe a
winning strategy for Player II in the strong Choquet game on Y . To do this we will
use the following:
Fact. Let U be open in Y , and y P U . Let also pWn qnPω be a sequence of open sets
in X covering f ´1 pU q. Then there exists anŤ open subset V of Y and k P ω such that
y P V and the restriction of f to f ´1 pV q X kn“0 Wn is semi-proper (with values in V ).
Assume the Fact is false, and let Vn “ Bpy, n1 q X U where the ball Bpy, n1 q refers
to some distance on Y . Then we get that for each n there is a countable
Ťn compact
set Kn Ď Vn such that no compact set L contained in f ´1 pV q X W satisfies
Ť n i“0 i
f rLs “ K. Let then K “ n Kn Y tyu. The set K is countable and compact, hence for
211
7 Borel Hierarchies
some compact L, Ťnf rLs “ K. As K Ď U , L is covered by the sets pWi q, hence for some
n we have L Ď ´1
Ť i“0 Wi . But then Ln “ L X f pKn q is also compact and contained
in f ´1 pVn q X ni“0 Wi , and f rLn s “ Kn , a contradiction which proves the fact.
We now define inductively the winning strategy for Player II as follows. At each step
n, if I has played Un and yn P Un , II plays Vn containing yn and a sequence pWin qki“0 n
,
for some kn P ω, of open sets in X such that:
(a) f : f ´1 pVn q X ki“0 Win Ñ Vn is semi-proper,
Ť n
(b) each Win is non-empty with diameter (for some complete distance on X) ď 2´n ,
and
(c) @i ď kn`1 Dj ď kn Win`1 Ď Wjn .
This can clearly be done, using at level n`1 the above Fact, applied to f f ´1 pVn qXŤkn Win
,
i“0
the open set Un`1 , yn`1 , and a sequence pWin qiPω
of open sets covering f ´1 pUn`1 q X
` ´1 Ťkn n
˘
f pVn q X i“0 Wi , of small enough diameters, and with closure contained in one
of the Win ’s.
Ş strategy is winning for II, i.e. that in any run pUn , yn , Vn qnPω
It remains to show that this
played this way, we have n Vn ‰ H. Consider the associated open sets pWin qiďkn ,
n P ω, in X. Let T be the tree on ω defined by
i`1 i
sPT ÐÑ @i ă lh s ´ 1 Wspi`1q Ď Wspiq .
212
7.3 The Wadge hierarchy
so that studying ďW is the same as studying the inclusion relation between Wadge
classes.
First, an easy lemma.
Lemma 7.3.1. (i) The class tHu, its dual tHuˇ and ∆01 are Wadge classes. In fact
if A Ď X is ∆01 and is neither H nor X itself, then A is ∆01 -complete.
(iv) If Γ is a Wadge class, and A Ď X is locally in Γ, i.e. is such that for every x P X
there is a clopen set V containing x with A X V P ΓpV q, then A is in Γ.
Proof. (i) Trivially tHu is a pointclass, hence a Wadge class. And in X ‰ H, clearly
X R tHu, i.e. tHu is non-self-dual. Note also that if A Ď X is different from
X and x0 P XzA, then the constant function with value x0 reduces H to A,
213
7 Borel Hierarchies
214
7.3 The Wadge hierarchy
So for any Borel set A, the only Borel sets which are not ďW -comparable to A
are, in case ΓrAs is non-self-dual, the sets which are equivalent to Ǎ.
The key for proving these two results is a device due to Wadge. If X, Y are Π01
subsets of ω ω , with associated canonical trees TX , TY , and A Ď X, B Ď Y , consider
the following Wadge game WpA, Bq:
I: n0 n1 n2
¨¨¨
II : m0 m1 m2
Proof of Theorem 7.3.2. (i) If A, B are Borel, the game WpA, Bq is Borel too, hence
by Theorem 6.1.5 is determined. So we get by the preceding discussion that
either A ďW B or B ďW Ǎ. If A ıW B and A ıW B̌, the only remaining
possibilities are A ăW B, or B ăW Ǎ. To finish the proof, note that if A ăW B,
then also A ăW B̌. This is because otherwise, by the previous result, B ďW A.
Finally if A and B are incomparable, we must have A ďW B̌ and B ďW Ǎ, i.e.
B ”W Ǎ, and moreover A ıW Ǎ. This proves the last statement in (i).
(ii) We argue by contradiction, and assume that pAn qnPω is a ăW -decreasing sequence
of Borel sets, which we may assume are in ω ω , by Subsection 7.3.1 (iii).
By the preceding discussion, player I has a winning strategy, call it σn0 , in
WpAn , An`1 q, as An W An`1 . Also, as An`1 ăW Aˇn , player I has a winning
ˇ q.
strategy σn1 in WpAn , An`1
Given ε P 2ω , consider the following sequence of runs in the Wadge games:
215
7 Borel Hierarchies
α0 α1 α1 α2 α2
εpnq
Player I plays in all games following his strategies σn . II answers in the nth
game by copying I’s moves in the n ` 1st game. This uniquely defines, for each ε,
a sequence pαn qnPω of points in ω ω such that for all n the point αn is the answer
εpnq
by I following σn to αn`1 played by II. Moreover the function ε ÞÑ pαn qpεq is
continuous.
So B “ tε P 2ω : α0 pεq P A0 u is a Borel set in 2ω , hence has the Baire property in
2ω . So there exists a finite sequence s P 2ăω , of length say k, such that B is either
meager or comeager on Vs “ tε P 2ω : s Ď εu. Let ϕ : Vs Ñ Vs be the involution
which associates to ε the image ϕpεq defined by ϕpεqpnq “ εpnq if n ‰ k, and
ϕpεqpkq “ 1 ´ εpkq. It is an homeomorphism, so B is meager in Vs if and only if
ϕpBq is meager in Vs . To get a contradiction, it is enough to check that for ε P B,
ϕpεq P 2ω zB, i.e. α0 pϕpεqq R A0 . Now note that αn pεq, n ą k, depends only on
the values of εppq, p ą k, hence αn pεq “ αn pϕpεqq for n ą k. Now if εpkq “ 0,
αk pεq P Ak Ø αk`1 pεq R Ak`1 , and if εpkq “ 1, αk pεq P Ak Ø αk`1 pεq P Ak`1 .
So we get αk pεq P Ak Ø αk pϕpεqq R Ak ; and easily as εk “ ϕpεqk , we get for
i ď k, αi pεq P Ai Ø αi pϕpεqq R Ai , so that α0 pεq P A0 Ø α0 pϕpεqq R A0 . This
gives the contradiction, and finishes the proof. %
216
7.3 The Wadge hierarchy
ăω
with ω ω , consider
ăω
B “ pσ, αq P ω ω ˆ ω ω : σ is a strategy for II in a Wadge game played
on ω ăω and ω ăω , and the answer σ ‹ α
(
when I plays α is in A
The previous results heavily rests on Martin’s theorem that Borel games are deter-
mined. In fact deep results of Martin and Harrington show that the determinacy of all
Wadge games WpA, Bq for A, B in Π11 , or even the consequence that every true Π11 set
is Π11 -complete, are equivalent to the determinacy of all Π11 games on ω, a statement
which is known to be a large cardinal hypothesis (see Martin [Mar] for a discussion of
this and related results).
On the other hand, in the case of Borel sets, the use of Borel determinacy (which,
although provable in ZFC, is not an elementary statement) can be avoided to prove
both the determinacy of Wadge games WpA, Bq for A, B Borel, and its consequences,
as proved by Louveau and Saint Raymond [LS87, LS12].
217
7 Borel Hierarchies
fsa 0 “fs ˝ fn
and fsa 1 “fs ˝ gn .
` ˘
Consider now, for ε P 2ω , the sequence fεn pαq nPω . By construction, we get for m ě n
(iii) Conversely, every self-dual Wadge class of Borel sets is of form (i) or (ii).
218
7.3 The Wadge hierarchy
(iii) We now prove the converse. Let Γ be a self-dual Wadge class of Borel sets, and
A Ď ω ω be Γ-complete. We first claim that there is a partition of ω ω into clopen
sets Wn such that A X Wn generates on Wn a non-self-dual Wadge class Γn . As,
by Theorem 7.3.2, ďW is wellfounded on the Borel sets, we may prove this fact
by induction on ďW , i.e. assume the result is true for all self-dual Wadge class
Γ1 ( Γ. As Γ is self-dual, we know by the previous lemma that there exists a
partition of ω ω into clopen sets Vn such that A X Vn generates a class Γ1n ( Γ on
Vn . If Γ1n is non-self-dual, Vn is fine. If Γ1 is self-dual, the induction hypothesis
gives a further partition of Vn into clopen sets Wnp such that on Wnp , A X Wnp
generates a non-self-dual Wadge class. Taking these clopen sets together gives
the desired partition.
Fix such a partition Wn for A, and Ť let Γn be the non-self-dual Wadge class
generated on Wn by A X Wn . Clearly n Γn Ď Γ. There are two cases:
Ť
(a) There is an n0 such that for allŤn Γn Ď Γn0 Y Γ̌n0 . Then Γn0 Ď n Γn Ď
Γn0 Γ̌n0 . But one cannot haveŤ n Γn “ Γn0 , for then by (iv) A P Γn0 and
Γ “ Γn0 is non-self-dual. So n Γn “ Γn0 Y Γ̌n0 , and Γn0 Y Γ̌n0 Ď Γ Ď Γ`
n0 .
`
This implies, by part (i), that Γ “ Γn0 .
(b) Otherwise we can inductively define a strictly increasing sequence nk by
n0 “ 0 and nk`1 “ least m ą nk with Ť Γ * Γnk YŤΓ̌nk . By Theorem 7.3.2
this is equivalent to Γnk ( Γm , hence
Ť n Γn “ k Γnk and the sequence
Γnk is strictly increasing. Finally n Γn Ď Γ Ď xΓnk y` , hence by part (ii)
Γ “ xΓnk y` and we are done. %
Remark 7.3.6. Define an equivalence relation « between Wadge classes of Borel sets by
Γ « Γ1 if Γ Y Γ̌ “ Γ1 Y Γ̌1 . Then Theorem 7.3.2 says that up to «, the Wadge ordering
219
7 Borel Hierarchies
(So in particular PU0 pΓq is the class of sets which are locally in Γ).
The notation PU stands for “partitioned union”, and the notion is originally due to
Myers.
Proof. Clearly for any pointclass Γ, PUξ pΓq is a pointclass. And PUξ pΓ̌q “ pPUξ pΓqqˇ.
If Γ ‰ tHu, tHuˇ, then tHu` Ď Γ. But easily tHu` “ ∆01 , so ∆01 Ď Γ. It follows that
PUξ p∆01 q Ď PUξ pΓq. Now every set in PUξ p∆01 q is clearly in ∆0ξ , and conversely any
∆0ξ set is in PUξ p∆01 q, in fact in PUξ ptHu Y tHuˇq.
Fix Γ, and set A “ tη ă ω1 : PUη pΓq “ Γu. By (iv) 0 P A, and clearly A is
an initial segment of ω1 . By the previous remark, if η P A then ∆0η Ď Γ, so that
A cannot be all of ω1 . So it remains to show that A is a successor ordinal, i.e. it
220
7.3 The Wadge hierarchy
sets Wnp X Vnp are disjoint in ηăλ¨ξ0 SUλη , and An P Γ. By the induction hypothesis,
Ť
A X Wn “ p pA X Wnp X Vnp q is also in Γ, for all n. But the sets Wn are in Σ01`ξ1 , and
Ť
Note that in the previous proof, the constructed f is any transfer function which
transforms certain Σ01`ξ sets into open sets. So we can ask that f simultaneously works
for countably many Γpξq sets, and for other Σ01`ξ sets as well. This flexibility is the
key for the next results.
Theorem 7.3.9.
(i) Let Γ be a non-self-dual Wadge class of Borel sets, and ξ ă ω1 . Then Γpξq is also
a non-self-dual Wadge class of Borel sets, and if ∆01 Ď Γ, then lpΓpξq q ě ξ, and
lpΓpξq q ą ξ if lpΓq ą 0.
221
7 Borel Hierarchies
Proof. (i) We first build a set H Ď ω ω ˆ ω ω which is universal for Γpξq subsets of ω ω .
To do this, let G be universal for Σ01`ξ subsets of ω ω , and let L be universal for
Γ subsets of ω ω ˆ ω ω – it exists by Corollary 7.3.3.
Define for each n P ω, pα, βq P Gn Ø ppαqn , βq P G, and by Theorem 7.1.6 let
F : ω ω ˆω ω Ñ ω ω be of Baire class ξ, with Y “ GraphpF q closed in ω ω ˆω ω ˆω ω ,
and with G˚n “ tpα, β, F pα, βqq : pα, βq P Gn u Σ01 in Y .
Define then pα, βq P H Ø ppαq0 , β, F ppαq1 , βqq P L. Clearly H is in Γpξq , as F is
of Baire class ξ and L is in Γ. We now show that H is universal. So let A Ď ω ω
be in Γpξq , and B Ď ω ω , B P Γ, and g : ω ω Ñ ω ω of Baire class ξ, be such that
ω
A “ g ´1 pBq. For each n, g ´1 pVnω q “ Bn is Σ01`ξ , hence for some γn P ω ω Bn “
Gγn . So if γ “ xγn : n P ωy, Bn “ pGn qγ . Set C “ tpβ, F pγ, βqq : gpβq P Bu.
Then if Y “ GraphpF q, C Ď Yγ . Now g ˝ π : Yγ Ñ ω ω , pβ, F pγ, βqq ÞÑ gpβq
ω
is continuous since pg ˝ πq´1 pVnω q “ G˚n X Yγ , hence C “ pg ˝ πq´1 pBq P Γ in
Yγ . By (iii), there exists D Ď ω ω , D P Γ, with C “ Yγ X D. Let δ be such that
D “ Lδ , and set α “ xδ, γy. One then gets
i.e. A “ Hα , as desired.
This proves that Γpξq is a non-self-dual Wadge class – and clearly of Borel sets. Of
course one has tHupξq “ tHu for all ξ. Ť So suppose ∆01 Ď Γ. We want to show that
Γpξq is closed under PUξ . So let A “ n pWn X An q, with An P Γpξq and pWn q a
partition of X into ∆01`ξ sets. Then by the remark following Lemma 7.3.8, there
is an f : X Ñ ω ω of Baire class ξ and closed graph Y such that A˚n “ tpα, f pαqq :
α P An u is in Γ, for all n, and such that the Wn˚ “ tpα, f pαqq : α P Wn u form
a partition of Y into ∆01 sets (by applying the transfer to the Wn ’s and their
complements too). But then A˚ “ tpα, f pαqq : α P Au is in PU0 pΓq in Y , hence
A is in Γpξq , as desired.
The proof for the last statementŤ is similar. Suppose lpΓq ą 0, so that PU1 pΓq “
Γ. Let A Ď X be of the form n pWn X An q with An P Γpξq and Wn in ∆01`ξ`1
with W n,p , Wn,p in Σ01`ξ ,
Ť Ť 1 1
now. Writing Wn “ p Wn,p and XzWn “ p Wn,p
ω
we can find, as before, a Baire class ξ function f : X Ñ ω with closed graph Y ,
which transforms the sets An into Γ sets A˚n in Y , and the Wn,p , Wn,p
1 for n, p P ω,
0 0
into Σ1 sets. Then the Wn ’s are transformed into ∆2 sets, and A is transformed
222
7.3 The Wadge hierarchy
into a set in PU1 pΓq, hence in Γ. So A P Γpξq . This shows PUξ`1 pΓpξq q Ď Γpξq ,
i.e. lpΓpξq q ě ξ ` 1, as desired.
This case does not
(ii) The case of Γ “ tHu is clear, so we assume that ∆01 Ď Γ. It is immediate to check make sense, does it?
that if Γ Ď Γ1 , then Γpξq Ď Γ1pξq and Γ̌pξq “ pΓpξq qˇ. So if Γ ( Γ1 , Γpξq ( Γ1pξq . – YP
By part (i) we also know (for Wadge classes of Borel sets) that Γ ‰ Γ̌ implies I don’t see why. This
follows for nsd
Γpξq ‰ pΓpξq qˇ. So the map Γ ÞÑ Γpξq preserves the Wadge ordering. It follows
pξq Wadge classes from
that if Γ “ Γ0 , the class Γ0 is unique. And by the last assertion of part (i), if the next assertion –
Γ is of level ξ, Γ0 must be of level 0. So the only thing to prove is the existence YP
of Γ0 . Set
´ď` ˘ ¯
pξq
Ǎ P PUξ Γn0 Ď Γ.
n
223
7 Borel Hierarchies
is defined by
˜ ¸
ď ´ ď ¯ ´ ď ¯
A“ An,ξ X Vn,ξ z Vm,ξ1 Y A˚ z Vn,ξ
ξăη ξ 1 ăξ ξăη
nPω mPω nPω
Proof. (i) To check that SDη p∆, Γ˚ q is a non-self-dual Wadge class, it is enough to
exhibit a universal set in ω ω ˆ ω ω . Say that ∆ “ Γ Y Γ̌. Let G be a universal
for Σ01 subsets of ω ω , H for Γ subsets of ω ω and H ˚ for Γ˚ subsets of ω ω . Let
ϕ : η ˆ ω Ñ ω be a bijection, and set for ξ ă η, n P ω,
` ˘
pα, βq P Gξ,n ÐÑ pαqϕpξ,nq , β P G
` ˘
and pα, βq P Hξ,n ÐÑ pαqϕpξ,nq , β P H.
By ω-reduction of Σ01 sets (Corollary 7.1.3), let for each ξ ă η pG˚ξ,n qnPω reduce
pGξ,n qnPω . We leave to the reader the easy checking that the set L Ď ω ω defined
below is universal for SDη p∆, Γ˚ q:
224
7.3 The Wadge hierarchy
225
7 Borel Hierarchies
It is worth noting that in the previous analysis, all classes Γ, pΓn qnPω and Γ˚ are
non-self-dual Wadge classes which are contained in ∆pΓq, so that one gets:
Corollary 7.3.11. The family W of all non-self-dual Wadge classes of Borel sets is
the least family satisfying
226
7.3 The Wadge hierarchy
Theorem 7.3.12. A pointclass Γ is a non-self-dual Wadge class of Borel sets iff there
is a Borel set D Ď 2ω with Γ “ HD pΣ01 q.
Moreover, a non-self-dual Wadge class of Borel sets is of level ě ξ iff there is a Borel
set D Ď 2ω with Γ “ HD pΣ0ξ q.
Proof. We first prove the easy direction. First note that for@ any D, Dany ϕ : X Ñ Y
and any sequence pYn q in Y one has ϕ´1 pHD pxYn yqq “ HD p ϕ´1 pYn q q. Hence if Γ is
any pointclass, HD pΓq is also a pointclass.
If now D is Borel and pAn q is a sequence of Borel sets in X, the function f : X Ñ 2ω
defined by f pxqpnq “ 1An pxq is Borel, and HD pxAn yq “ f ´1 pDq. So HD pxAn yq is Borel
too. So if D is Borel and Γ is a pointclass of Borel sets, HD pΓq is a pointclass of Borel
sets.
Suppose now that Γ is a Wadge class admitting a universal set G for Γ subsets of
ω ω . Set
pα, βq P Gn ÐÑ ppαqn , βq P G,
and let H “ HD pxGn yq.
227
7 Borel Hierarchies
We now show that for every ξ ă ω1 there is Dξ Ď 2ω with Σ01`ξ “ HDξ pΣ01 q. This
is done by induction on ξ. It is clear for ξ “ 0 (taking e.g. D0 “ 2ω zt0ω u). Suppose
ξ ą 0 and that it is known for all η ă ξ. Let Dη correspond to Σ01`η , η ă ξ; let ϕ be
˚ pmq “ 1 ´ αpϕpm, n, ηqq. Let then
a bijection between ω ˆ ω ˆ ξ and ω, and set αη,n
Dξ “ α P 2ω : Dn Dη ă ξ αη,n ˚
(
P Dη .
so that
HDpξq pΣ01 q “ HD HDξ pΣ01 q “ HD pΣ01`ξ q “ Γpξq ,
` ˘
as desired.
It remains to show that if 1 ď η ă ω1 , Γ˚ “ HD˚ pΣ01 q, Γ “ HDŤ pΣ01 q, lpΓq ą 0
0
and ∆ “ Γ Y Γ̌, or for all n, Γn “ HDn pΣ1 q, lpΓn q ą 0 and ∆ “ n Γn , then for
Λ “ SDη p∆, Γ˚ q, Λ “ HE pΣ01 q for some E Ď 2ω . Let us do the case ∆ “ Λ Y Λ̌, the
other case being entirely similar.
There is a little subtlety in the proof. It is due to the fact that in the definition of
the operation SDη , we use Σ01 sets Wξ,n for ξ ă η and n P ω with the property that
for n ‰ m Wξ,n X Wξ,m “ H. So, as written, SDη is not a Hausdorff operation. We
must split the proof in two cases.
Case 1: ∆ “ tHu Y tHuˇ. In this case, one has Λ “ Dξ pΣ01 q, or Λ “ Ďξ pΣ01 q for some
ξ ă ω1 , ξ ě 1. In the first case, fix a bijection between ξ and a subset A of ω,
and set
Ěξ “ tα : α˚ P Eξ u.
228
7.3 The Wadge hierarchy
Conversely any set in SDη p∆, Γ˚ q is of the form (F), trivially. And it is only a
question of coding to get a Hausdorff operation from the expression (F): Let ϕ
be a bijection between ω ˆ η and ω, and set:
α “ xα0 , α1 , α2 , α3 y P E ÐÑ
– either Dξ ă η such that Dn α0 pϕpξ, nqq “ 1 and for the least
such ξ there is a unique n such that α0 pϕpξ, nqq “ 1
and if α1 pϕpξ, nqq “ 0, pα2 qϕpξ,nq P D,
and if α1 pϕpξ, nqq “ 1, pα2 q˚ϕpξ,nq P D,
– or @ξ ă η @n P ω α0 pϕpξ, nqq and α3 P D˚ .
We leave to the reader the checking that E works, i.e. that HE pΣ01 q “ SDη p∆, Γ˚ q.
It remains to prove the last assertion. Note that we have already seen that if Λ “
Γpξq for some non-self-dual Wadge class of Borel sets Γ, and if HD pΣ01 q “ Γ, then
HD pΣ01`ξ q “ Λ. And conversely that HD pΣ01`ξ q “ pHD pΣ01 qqpξq , hence must have
level ě ξ. So it remains to show that any non-self-dual Wadge class Λ of level ξ 1 ě
ξ can be written as Γpξq for some non-self-dual Γ. We already know this for ξ by
Theorem 7.3.9 (ii), so the result follow from the equality pΓpηq qpξq “ Γpξ`ηq . One
direction of this equality is immediate, for if g : X Ñ Y has Baire class ξ, and
h : Y Ñ Z has Baire class η, h ˝ g has Baire class ξ ` η. For the converse, let A Ď X
be of the form f ´1 pBq for some B Ď Y in Γ and f of class ξ ` η. Then easily there
are Σ01`ξ sets An,m in X such that for all n P ω f ´1 pVnY q “ HDη pxAn,m : m P ωyq
(where Dη is defined earlier in the proof). By Kuratowski’s transfer Theorem 7.1.6,
let h : X Ñ ω ω be a Baire class ξ function with closed graph
` Z such
˘ that for all n, m
A˚n,m “ tpα, f pαqq : α P An,m u is Σ01 in Z. If we set g pα, hpαqq “ f pαq, we get a
@ D
function g : Z Ñ Y such that for all n A˚n “ g ´1 pVnY q “ HDη p A˚n,m : m P ω q is Σ01`η .
So g has Baire class η and B ˚ “ g ´1 pBq is in Γpηq . As h̃pαq “ pα, hpαqq : X Ñ Z has
class ξ, A “ h̃´1 pB ˚ q is in pΓpηq qpξq , as desired. %
229
7 Borel Hierarchies
Note that the previous result gives a different characterization of the level of a non-
self-dual Wadge class Γ of Borel sets: It is the largest ordinal (which exists) ξ such
that for some (Borel) D Ď 2ω , Γ “ HD pΣ01`ξ q.
7.3.8 Ramifications
For Borel sets A and B, the Wadge game WpA, Bq is simple to describe, but its
complexity is roughly that of A and B, so that although it is determined by Martin’s
Theorem 6.1.5, it does not give much information, in particular about the complexity
of the reducing maps.
Here we introduce games, due to Louveau and Saint Raymond, which the same job
as the Wadge games (and are some kind of unfoldings of these games), and are open,
so that we know the complexity of their winning strategies. On the other hand, this is
done at the expense of simplicity.
The main tool is a way of transforming open games which is called a ramification in
Louveau–Saint Raymond [??] that we now describe.
réf manquante – YP
In the games we will consider, Player I plays at each step i “ 0 or 1, and player II
plays integer. We define a function r “ pr0 , r1 q which to each position pu, vq P p2ˆωqăω
associates a new position u˚ “ r0 puq, v ˚ “ r1 pu, vq in pu, vq P p2 ˆ ωqăω , and is defined
as follows. We view u as being played on ω 2 :
0 1 2 ¨¨¨ k
1 up1q up4q
2 up3q
..
.
n upxk, nyq
(So ρ0 puq takes eventually value 1.) Intuitively, ρ0 puq is the expected output in the
original game that player I guesses in the new game when playing u. The value u˚ “
r0 puq is a beginning of ρ0 puq, and pu˚ , v ˚ q are defined inductively as follows: First set
r0 pHq “ H, r1 pH, Hq “ H. Suppose r0 puq and r1 pu, vq have been defined, in defining
r0 pua iq and r1 pua i, v a kq, we consider two cases:
230
7.3 The Wadge hierarchy
Case 1: ρ0 pua iq extends r0 puq – so that player I has no reason to change his guess.
Let then r0 pua iq “ ρ0 pua iqlh r0 puq`1 and r1 pua i, v a kq “ r1 pu, vqa k.
Case 2: Otherwise. Note that this can happen only if i has been played on row j “
plh uq0 with j ă lhpr0 puqq, and changes the value of ρ0 puqpjq, i.e. if i “ 1, and it
is the first time a 1 is played on row j, so that ρ0 pjq “ 1 but ρ0 pua 1qpjq “ 0. In
this case we let
r0 pua 1q “ r0 puqj a 0
(i.e. player I erases his wrong guess as much as needed, and start the new one)
and
r1 pua 1, v a kq “ r1 pu, vqj a k.
Suppose now J is a subtree of p2ˆωqăω , which defines an open game GpJq for player
I, by : II wins the run pε, βq in GpJq if pε, βq P rJs. We can use the function r to define
the ramification rpJq by
` ˘
pu, vq P rpJq ÐÑ rpu, vq “ r0 puq, r1 pu, vq P J.
In other words, player II wins the run pε, βq in the ramified game if all guessed positions
rpεk , βk q, for k P ω, are in J.
Lemma 7.3.13. For every pε, βq P 2ω ˆω ω , the set Tε,β “ trpεk , βk q : k P ωu is a sub-
tree of p2ˆωqăω . Moreover rTε,β s is a singleton we denote by ρpε, βq “ pρ0 pεq, ρ1 pε, βqq.
The function ρ : 2ω ˆ ω ω Ñ 2ω ˆ ω ω has Baire class 1, and ρ0 is given by
#
0 if Dn εpxk, nyq “ 1,
ρ0 pεqpkq “
1 otherwise,
from which it follows that the inverse image of any basic clopen set is ∆02 .
The claim is proved by induction on n: Note that r0 pεk qp0q is always defined for
k ą 0, and takes value 1 as long as εk takes only value 0 on row 0, and jumps to 0
only once, in the event εk takes a value 1 on row 0, in which case it remains 0 from
231
7 Borel Hierarchies
that point on. And after that, r0 pεk qp1q is defined, takes value 1 as long εk takes
only value 0 on row 1, and may change to value 0 only once, if εk takes the value 1 on
row 1, etc. And inductively if k is large enough so that r0 pεp qn is constant from k on,
then r0 pεp qpnq is defined for p ą k, and may change only once, hence stabilizes, and
clearly to the value of ρ0 pεqpnq. Moreover by the definition of r1 pεp , βp q, its value at
n stabilizes exactly at the same moment k (with value the corresponding βpkq). This
proves the claim, and hence the lemma. %
The previous lemma implies that it is harder for player II to win rpJq than to win J
– he has to insure that in any run pε, βq we have Tε,β Ď J, not only that ρpε, βq P rJs.
ăω
But we now show that player I has no easy win in rpJq. To see this, let Σ “ 2pω q ,
so that σ P Σ can be viewed as a strategy for player I in our games, and let J be
the set of all subtrees of p2 ˆ ωqăω . We now associate to each game J P J and each
σ P Σ, viewed a strategy for I in rpJq, a strategy σ ˚ “ F pJ, σq for player I in J, defined
informally as follows, together with a function f “ f pJ, σq : ω ăω Ñ 2ăω ˆ ω ăω .
Step 0: Let
One has puH a σpvH q, vH a n0 q P Lxn0 y . Again there are two cases:
(i) If for all pu, vq P Lxn0 y r0 pua σpvqqp1q is constant, we set uxn0 y “ uH a σpvH q,
vxn0 y “ vH a n0 , f pxn0 yq “ puxn0 y , vxn0 y q, and σ ˚ pxn0 yq “ r0 puxn0 y a σpvxn0 y qqp1q.
(ii) If the value r0 pua σpvqqp1q is not constant on Lxn0 y , it means that it is
1 for uH a σpvH qa σpvH a n0 q but becomes 0 for some pu, vq P Lxn0 y . Let
then f pxn0 yq “ puxn0 y , vxn0 y q be such a pair of minimal length, and set
σ ˚ pxn0 yq “ 0.
232
7.3 The Wadge hierarchy
General case: Suppose f pwq “ puw , vw q and σ ˚ pwq have been defined for w P ω n ,
such that if pσ ˚ pwq, wq P J, then rpuw , vw q “ pσ ˚ pwq, wq, uw is the answer to
vw by σ, r0 puw a σpvw qq is a one-step extension of σ ˚ pwq, and such that for all
pu, vq extending puw , vw q, with u the answer to v by σ, and with pu, vq P rpJq,
rpu, vq is an extension of rpuw , vw q. We define then f pwa kq and σ ˚ pwa kq in
any way if pr0 puw qa σ ˚ pwq, wa kq R J. And if pr0 puw qa σ ˚ pwq, wa kq P J, then
puw a σpvw q, vw a kq is in
One easily checks that the induction hypotheses are still satisfied with this choice.
Lemma 7.3.14. (i) Let σ, J be fixed, and σ ˚ “ F pJ, σq, f “ pJ, σq be defined as
above. If pξ, βq is a winning
Ť run for player II in the game J with player I following
˚ ˚ ˚
σ , then pε , β q “ k f pβk q is a winning run for player II in the game rpJq,
with player I following σ, and moreover ρpε˚ , β ˚ q “ pε, βq.
In particular if σ is a winning strategy for player I in rpJq, σ ˚ is a winning
strategy for player I in J.
Proof. (ii) is an easy computation, which is left to the reader (note that the these
functions are not continuous, a priori, as one decides at each step which case to
apply by a condition which is closed in J and σ, but not necessarily clopen).
(i) By the way f has been defined, we know that if pεk , βk q P J and εk is the answer
to βk by σ ˚ , then puβk , vβk q “ f pβk q satisfies rpuβk , vβk q “ pεŤ
k , βk q, uβk
is the answer to vβk by σ. Moreover f preserves extension, hence k f pβk q is
a play pε˚ , β ˚ q, and ε˚ is the answer to β ˚ by σ. And as puβk , vβk q P rpJq for
all k, pε˚ , β ˚ q P rrpJqs, i.e. is a winning play for player II in rpJq against σ. And
finally as rpuβk , vβk q “ pεk , βk q, ρpε˚ , β ˚ q “ pε, βq as desired.
The final assertion is the immediate, for if σ ˚ is not winning for I in J, it can
be beaten by some β P ω ω , with answer ε, and the corresponding play pε˚, β ˚ q
beats the strategy σ in rpJq. %
233
7 Borel Hierarchies
Then we get rn`1 J “ rn prJq “ rprn Jq. We can also define the nth iterate ρn of
ρ : 2ω ˆ ω ω Ñ 2ω ˆ ω ω , and define by induction F n : J ˆ Σ Ñ Σ by F 1 “ F , and
234
7.3 The Wadge hierarchy
that these games are determined. Here we will prove this in case A is a true Π0n set,
for n P ω.
First we study the case n “ 1. Let T0 and T1 be two trees on ω, and define an open
game J1 “ J1 pT0 , T1 q P J as follows:
I: εp0q P 2 εp1q P 2
¨¨¨
II : βp0q P ω βp1q P ω
As usual the open game J1 is defined by II wins the run pε, βq in the game J1 if for
all k pεk , βk q P J1 . So II wins a run pε, βq if either I plays only 0’s and II produces a
branch through T0 , or at some point I plays a 1 and II produces, from that point on,
a branch through T1 .
Let T be the set of all subtrees of ω ăω . Define H1 “ t0ω u Ď 2ω , and C1 Ď T ˆ T ˆ Σ
by ` ˘
pT0 , T1 , σq P C1 ÐÑ Du P T0 @i ă lh u σpui q “ 0 ^ σpuq “ 1 .
Proof. (i) Let TA be some subtree of TX with rTA s “ A. Let player II answer 0 to
u P TX as long as u P TA , then 1 when u R TA . This clearly is winning in
WpA, H1 q.
(ii) is immediate.
235
7 Borel Hierarchies
Let now B0 and B1 be two disjoint Σ11 subsets of Y Ď ω ω , and fix trees T0 and T1
on ω ˆ ω (in fact in TY ˆ ω ăω ), such that or i “ 0, 1 Bi “ tα P Y : Dβ pα, βq P rTi su.
For each α P Y , we let Ti pαq “ tu P ω ăω : pαlh u , uq P Ti u, so that Ti pαq is a tree on
ω, for i “ 0, 1.
Let n P ω, and define a game Gn`1 pB0 , B1 q as follows:
where I plays 0’s and 1’s, II plays the first coordinates in TY and the second ones in
ω ăω . Let pε, α, βq be the
` result of a run.
˘ Then player II wins this run if pε, βq is a win
n
for II in the game r J1 T0 pαq, T1 pαq .
Clearly the game Gn`1 pB0 , B1 q is closed, hence determined. Set Hn`1 “ pρn0 q´1 pH1 q.
If σ is a strategy for I in Gn`1 pB0 , B1 q and α P Y , let σpαq be the induced strategy in
Σ, defined for u P ω ăω by σpαqpuq “ σpαlh u , uq. Define then
(
Cn`1 pσq “ α P Y : pT0 pαq, T1 pαq, σpαqq P C1 .
Theorem 7.3.17. (i) The set Hn`1 Ď 2ω is Π0n`1 . And for every A Ď X in Π0n`1 ,
player II wins WpA, Hn`1 q.
(ii) For each strategy σ for I in Gn`1 pB0 , B1 q, the set Cn`1 pσq is Σ0n`1 , in fact
Σ0n`1 pσ, T0 , T1 q.
(iii) Suppose player II has a winning strategy τ in Gn`1 pB0 , B1 q. Then the strategy ob-
tained by deleting the β-moves is a winning strategy for player II in W* pHn`1 ; B0 , B1 q.
(iv) Suppose player I has a winning strategy σ in Gn`1 pB0 , B1 q. Then the set Cn`1 pσq Ď
Y separates B0 from B1 .
Proof. (i) As H1 is Π01 and ρn has Baire class n, Hn`1 is Π0n`1 in 2ω . We prove that
player II wins WpA, Hn`1 q, for A P Π0n`1 , by induction on n. This part (i) of the
previousŤlemma for n “ 0. Suppose we know it for n ´ 1 ě 0. If A P Π0n`1 , write
XzA “ k Ak where Ak Ď X is Π0n . Let σk be a winning strategy for player II
in WpAk , Hn q. If I plays u P TX , with lh u “ xk, ly, define σpuq “ σk pul q. We
claim σ is winning in WpA, Hn`1 q. Note that if α P Y and β is the answer by σ
to α, then for each k, pβqk is the answer by σk to α. So we get
αPA ÐÑ @k α R Ak ÐÑ @k pβqk R Hn .
Now by definition of the Hn ’s, Hn`1 “ ρ´10 pHn q, and as ρ0 pβqpkq “ 1 Ø pβqk “
0ω , one easily checks by induction that β P Hn`1 Ø @k pβqk R Hn . This proves
(i).
236
7.3 The Wadge hierarchy
` ˘
(ii) As F n is a Baire class n function, so is the composition function α ÞÑ F n J1 pT0 pαq, T1 pαqq, σpαq .
And C1 is clearly open, so Cn`1 pσq is Σ0n`1 . To see that it is Σ0n`1 pσ, T0 , T1 q one
has to check that C1 is in fact Σ01 , and F n is Σ0n`1 -recursive. This is easy, but
tedious, and left to the reader.
(iii) Let τ be winning for II in Gn`1 pB0 , B1 q, and for ε P 2ω let pα, βq be the answer
by II following the strategy τ . Then by definition of Gn`1 the run pε, βq is a win
for II in rn J1 pT0 pαq, T1 pαqq hence by Theorem 7.3.15 ρn pε, βq is a win for II in
J1 pT0 pαq, T1 pαqq. Suppose that ε P Hn`1 . Then ρn0 pεq “ 0ω P H1 , i.e. ρn1 pε, βq is
a branch through T0 pαq by Lemma 7.3.16, hence α P B0 . If ε R Hn`1 , ρn0 pεq P H1 ,
and by Lemma 7.3.16 again, ρn1 pε, βq is a branch through T1 pαq, hence α P B1 .
This proves (iii).
(iv) Let now σ be winning for I in Gn`1 pB0 , B1 q. Then by Theorem 7.3.15 for each α
the strategy F n pJ1 pT0 pαq, T1 pαqqq is winning for I in J1 pT0 pαq, T1 pαqq. But then
by Lemma 7.3.16 if α P B0 , i.e. rT0 pαqs ‰ H, then α P Cn`1 pσq. And if α P B1 ,
i.e. rT1 pαqs ‰ H, then α R Cn`1 pσq. This proves (iv), and finishes the proof of
the theorem. %
(i) For every pair B0 , B1 of disjoint Σ11 sets in Y , the separation game W* pA; B0 , B1 q
is determined.
Proof. (ii) is a consequence of (i). To prove (i), suppose first that no Σ0n set sepa-
rates B0 from B1 . Then by the previous theorem, player II has a winning strategy in
Gn`1 pB0 , B1 q, hence in W* pHn ; B0 , B1 q. As A is Π0n`1 , II also has a winning strat-
egy in WpA, Hn q. Composing these strategies, one gets a winning strategy for II in
WpA; B0 , B1 q. Suppose now C is Σ0n`1 and separates B0 from B1 . Then II wins
WpČ, Hn q. For all k let Ak “ tα P ω ω : k a α P Au in Xk “ tα P ω ω : k a α P Xu. If
each Ak is Σ0n`1 in Xk , so is A in X. So pick k0 such that Ak0 R Σ0n`1 . Then by the
first argument, player II has a winning strategy in W* pHn , Ak0 , Aˇk0 q “ WpHn , Ak0 q,
hence by composition in WpČ, Ak0 q. Let then I play in W* pA; B0 , B1 q by first playing
k0 , and then following II’s winning strategy in WpČ, Ak0 q. One easily checks that this
strategy is winning for I in W* pA; B0 , B1 q. %
Remarks 7.3.19. (1) In the previous proof, we haven’t used Borel determinacy. So
this gives a different, and more elementary, proof of the statement: If B is Borel
and A is a true Π0n set, WpA, Bq is determined. The method of proof is extended
in [??] to all Wadge games WpA, Bq with A, B Borel, and in fact to all games
ref missing – YP
W* pA; B0 , B1 q with A Borel and B0 , B1 in Σ11 .
237
7 Borel Hierarchies
(2) It is not hard to check that if B0 , B1 are Σ11 (in a Polish dim0-recursive space
Y ), the game Gn`1 pB0 , B1 q is ∆11 . So if I wins it, he has a winning strategy σ
which is ∆11 -recursive by Theorem 6.1.2. But then Cn`1 pσq is Σ0n`1 p∆11 q, hence
one gets as a corollary the instances of Louveau’s Theorem 7.1.10 corresponding
to the classes Σ0n and Π0n . Again extending these techniques to all non-self-dual
Wadge classes of Borel sets gives ([??],[??]) analogous lightface results for all
ref missing – YP
Wadge classes.
Proof. The last statements follow easily from the first, by considering the pairs pB, XzBq
and pXzB, Bq, respectively. Suppose first n “ 1. We cannot apply directly Theo-
rem 7.2.10, as the sets B0 and B1 are Σ11 . But by Theorem 7.3.17, the hypothesis im-
plies the existence of a continuous g : 2ω Ñ X with gr2ω s Ď B0 Y B1 and g ´1 pB0 q “ Ď.
Now gr2ω s “ Y is compact, and B0 XY is Borel in Y (as its complement is B1 XY ), and
is still not Σ02 as g ´1 pB0 X Y q “ Ď is complete Π02 . So we can apply Theorem 7.2.10
to B0 X Y , which gives the one-to-one function f as desired.
Assume now n ą 1. We define a variant of the game Gn`1 pB0 , B1 q of Subsec-
tion 7.3.9. Fix i “ 0, 1 a tree Ti on ω ˆ ω with Bi “ tα P ω ω : Dβ P ω ω pα, βq
inTi u, and consider the game G˚n`1 pB0 , B1 q:
I: εp0q P 2 εp1q
¨¨¨
II : s0 P ω ăω βp0q, s1 P ω ăω βp1q, s2 P ω ăω
(so that II starts this game). At the end of a run, I has produced ε P 2ω , and II
some β P ω ω ; moreover we set α “ s0 a εp0qa s1 a εp1qa . . . P ω ω (so that α is played
as in a ˚-game of Subsection 6.3.5). And we say the run is winning for II if pε, βq
is a win for II in rn J1 pT0 pαq, T1 pαqq, as for the Gn`1 pB0 , B1 q game (so the only dif-
ference between the two games is the way α is produced). Again G˚n`1 pB0 , B1 q is
238
7.4 The Baire hierarchy of functions
an open game for player I, hence is determined. Suppose first that II has a win-
ning strategy τ in this game. Let f : 2ω Ñ ω ω be the continuous function de-
fined by: f pεq is the α corresponding to the run where I plays ε and II follows his
strategy τ . For each ε, we know that if β is the other play of II, pε, βq is a win
in rn J1 pT0 pf pεqq, T1 pf pεqqq. By the same proof as for Gn`1 pB0 , B1 q, this gives that
ε P Hn`1 Ñ f pεq P B0 and ε R Hn`1 Ñ f pεq P B1 . And finally f is one-to-one,
for if ε and ε1 differ at n0 for the first time, the images f pεq and f pε1 q will have the
same beginning of the form s0 a εp0qa s1 a ¨ ¨ ¨ a sn0 ´1 , and then differ at the next value.
This gives the theorem. So it remains to show that I cannot win G˚n`1 pB0 , B1 q. So
suppose σ is winning for I in this game. We now argue as for the game G˚1 pAq of Theo-
rem 6.3.6: Say that for a position u “ ps0 , εp0q, pβp0q, s1 q, εp1q, pβp1q, s2 q, . . . , εpk ´ 1qq
and m P ω, pu, mq is critical for α P ω ω if u is a beginning where I follows his strategy
σ, s0 a εp0qa s1 a εp1qa ¨ ¨ ¨ a sk´1 a εpk ´ 1q Ď α, and for no extension ua ppm, sk q, εpkqq
where I follows σ one has s0 a εp0qa s1 a ¨ ¨ ¨ a sk´1 a εpk ´ 1qa sk a εpkq Ď α. As in the
proof of Theorem 6.3.6 if pu, mq is critical for α, pu, mq together with σ entirely de-
termines α. So the set D0 “ tα : Dpu, mq pu, mq is critical for αu is countable. Let
H “ ω ω zD. For each α P H, we can define a strategy σpαq P Σ as follows: Pick
first s0 of minimal length such that the answer εp0q by σ satisfies s0 a εp0q Ď α. If II
plays βp0q, pick s1 of minimal length such that the answer εp1q to ps0 , εp0q, pβp0q, s1 qq
satisfies s0 a εp0qa s1 a εp1q Ď α, etc. This is defined because α P H, and σpαq is clearly
continuous in α. Moreover the strategy σpαq is winning for I in rn J1 pT0 pαq, T1 pαqq,
hence as in Theorem 7.3.17 the set
(
C “ α P H : pT0 pαq, T1 pαq, σpαqq P C1
separates B0 X H from B1 X H. This set is Σ0n`1 in H, hence for some Σ0n`1 set
C 1 Ď ω ω , C “ C 1 X H. Define C 2 “ C 1 Y pB0 zHqzpB1 qzHq. Then C 2 separates B0
from C1 – and so does C 2 X X. Now B0 zH and B1 zH are countable, hence Σ02 in ω ω ,
and by our hypothesis that n ą 1, C 2 is also Σ0n`1 in ω ω , hence C 2 X X is Σ0n`1 in X.
As this contradicts our assumption on pB0 , B1 q, we are done. %
239
7 Borel Hierarchies
which is immediate (using the regularity of Y ). Note however that the converse is far
from being true: For each ξ, there are by Kuratowski’s transfer Theorem 7.1.6 Baire
class ξ functions ω ω Ñ ω ω with Π01 graph, which are not of Baire class η for any η ă ξ.
To see this, pick any A Ď ω ω which is a true Σ01`ξ , and by Theorem 7.1.6 an f : ω ω Ñ
ω ω of Baire class ξ, with closed graph Y and such that A˚ “ tpα, f pαqq : α P Au
is Σ01 in Y . If f were of Baire class η for some η ă ξ, one would get A in Σ01`η , a
contradiction.
Proof. As fn converges uniformly, let for each k, nk bu such that dpfn pxq, f pxqq ă 2´k
for all x and n ě nk . If A is closed in Y , set Ak “ ty P Y : Dz P A dpy, zq ă 2´k u.
Each Ak is open in Y , so the proposition will follow immediately from the equivalence:
Proposition 7.4.2. If X, Y are recursive spaces, and pfn qnPω is a sequence of Baire
class ξ functions : X Ñ Y which pointwise converges to f : X Ñ Y , the limit f has
Baire class ξ ` 1.
240
7.4 The Baire hierarchy of functions
for then, as fn´1 pAk q is Σ01`ξ , we get f ´1 pAq P Π01`ξ`1 . For direction Ñ, note that for
each k there is nk pxq such that dpfn pxq, f pxqq ă 2´k for n ě nk pxq. This easily gives
the implication. Conversely if the right-hand side holds, we get for infinitely
Ş many n’s
fn pxq P Ak , hence f pxq P Ak , for all k. And again f pxq P A as A “ k Ak . %
The next two paragraphs will be devoted to a proof of the converse to this propo-
sition, the Lebesgue–Hausdorff theorem. However, one has to be a bit cautious: If
X “ R and Y “ 2 “ t0, 1u, the only continuous functions : X Ñ Y are the constants,
which are closed under pointwise limits, so the converse is not always true. But we
will see that this is essentially the only case where it fails.
Fix n. The sets Ai,n , for i “ 1, . . . , k ar Π01`ξ and disjoint, hence by the separation
property for Π01`ξ , which holds by Corollary 7.1.3, if ξ ě 1 or if ξ ě 0 in dim0 spaces,
there is a partition pA˚i,n qi“1,...k of X in ∆01`ξ sets with Ai,n Ď A˚i,n . Define fn : X Ñ Y
by fn A˚ “ yi . Clearly fn is of class ξ, and it remains to check that fn pointwise
i,n
converges to f to get (i) and (iv). But if x P X, say x P Ai , then eventually x P Ai,n ,
hence eventually x P A˚i,n , i.e. eventually fn pxq “ f pxq.
´1
Ť To get (ii), suppose now ξ is limit. Let Ai “ 0f pyi q as 0
before, and write Ai “
A with pAi,n q an increasing sequence of Π1`ξ “ Πξ sets, and Ǎi “ XzAi “
Ťn i,n
with pBi,n q an increasing sequence of Π0ξ sets. So we can write Ai,n “ j Ai,n,j ,
Ş
n Bi,n Ş
B “ j Bi,n,j with decreasing sequences pAi,n,j qj and pBi,n,j qj of sets in Π0ăξ “
Ťi,n 0
ηăξ Πη . Set, for i “ 1, . . . , k and j P ω:
ď` ˘
Ci,j “ Ai,n,j zBi,n,j .
nďj
241
7 Borel Hierarchies
dpf pxq, gl pxqq ďdpf pxq, fk pxqq ` dpfk pxq, flk pxqq ` dpflk , fll pxqq
l´1
ÿ
ď4´k ` 2 ¨ 4´i ` dpfk pxq, flk pxqq,
i“k
242
7.4 The Baire hierarchy of functions
which, as flk pxq Ñ fk pxq as l Ñ 8, can be made as small as we want for large enough
l. This finishes the proof of (i), (ii) and (iv). And (iii) follows immediately from these
results together with Proposition 7.4.2. %
Proof. We first reduce to the case of Y “ r0, 1s. Clearly the result for R implies the
result for Rn , n ď ω, by considering the ` projections.
˘ Let ϕ be a homeomorphism
: r0, 1s Ñ R “ R Y t´8, `8u with ϕ s0, 1r “ R. For f a Baire class 1 function
: X Ñ R, let gpxq “ ϕ´1 ˝ f pxq, so that g : X Ñ r0, 1s is also Baire class 1. Assuming
the result is known for g, we get a sequence gn : X Ñ r0, 1s of continuous functions
which pointwise converges to g. Then gn1 pxq “ 2´n ` p1 ´ 2n qgn pxq is also continuous
: X Ñ r0, 1s, and still pointwise converges to g. Moreover as gn1 takes value in s0, 1r,
fn “ ϕ ˝ gn is a continuous function : X Ñ R. And by continuity of ϕ, fn pointwise
converges to f , as desired.
So it is enough to prove the result for Y “ r0, 1s. Let f : X Ñ Y be Baire class
1. As in the proof of the above theorem, we can find a sequence opfk q of Baire class
1 functions : X Ñ r0, 1s with |fk pxq ´ f pxq| ď 4´k , and fk taking only finitely many
values. We claim that it is enough to get the result for functions taking only finitely
many values. This is because we can find continuous functions glk : X Ñ r´1, 1s
converging pointwise to f0 for k “ 0, and to gk “ fk ´ fk´1 for k ą 0, andřsatisfying
|gk pxq| ď 2¨4´k . Then as in the proof of the theorem, the sequence hk pxq “ lďk glk pxq
is continuous, and pointwise converges to f .
So it remains to prove the case where f : X Ñ r0, 1s takes only finitely many
´1 0
values,Ťsay y1 , . . . , yk . Let Ai “ f pyi q. Each Ai is Σ2 in X, hence can be written
Ai “ n Ai,n with pAi,n q an increasing sequence of closed sets. This is where the proof
departs from the theorem, as we cannot use the separation property for Π01 , which
is false unless X has dimension 0. But instead, we can use Urysohn’s Lemma 2.4.7:
Fix n, the sets Ai,n , i “ 1, . . . , k are disjoint closed sets, so we can Ť find for each i
a continuous function ϕi : X Ñ r0, 1s which is 1 on Ai,n and 0 on j‰i Aj,n . But
` řk ˘
then fn “ min i“1 yi ϕi , 1 is continuous X Ñ r0, 1s, and takes value yi on Ai,n for
i “ 1, . . . , k. It follows, as in the proof of the theorem, that fn pointwise converges to
f , as desired. %
243
7 Borel Hierarchies
244
List of Symbols
The main symbols used throughout the book are listed below together with a short
description and the number of the page where they first appear.
A Suslin’s Operation 75
A effective Suslin’s Operation 80
αpn a coding of the initial segment of α P LO below n 85
245
List of Symbols
O, O Kleene’s O 130
ω1CK , ω1CK´α the least non recursive (resp. in α) ordinal 91
On the class of ordinals 83
246
List of Symbols
247
Index
The main concepts of the book with the pages where they are introduced and treated...
adequate class 51
Cauchy sequence 26
code 7
recursive in the codes 7
derivation 97
Σ11 on Σ11 101
difference set 188
distance 25
compatible 25
effective topology 32
function
B-function 72
Γ-measurable function 60
Γ-recursive function 60
Baire class ξ function 59
bianalytic function 60
Borel function 59
partial recursive function 12
recursive function 1
recursive isomorphism 36
retraction 214
semi-proper 211
game
Choquet 165
Wadge game 215
Γ-recursive point 108
hierarchy
Borel hierarchy 47
Kleene arithmetical hierarchy 50
249
Index
Kleene classes 53
operator 97
Π11 on Π11 97
Σ01 in the codes 21
monotone inductive 97
uniformly Σ01 21
ordinal
ω1CK , the Church-Kleene ω1 91
recursive ordinal 91
preordering
Γ-prewellordering 88
prewellordering 88
set
B-set 72, 91
Fσ -set 47
Gδ -set 47
Γ-complete set 213
additive class ξ (of) 47
250
Index
theorem
Kleene’s S-m-n 19
Kleene’s Fixed Point 59
Kuratowski’s “second reduction theorem” 89
Kuratowski’s Reduction Theorem for Π11 90
Recursion Theorem 19
uniform version 20
Sulin-Kleene separation theorem 119
The ∆11 -uniformization Theorem 111
The Prewellordering Theorem 88
The Separation Theorem for Σ11 sets 90
Transfer Theorem to ω ω 67
Urysohn-Tychonoff 40
effective Urysohn-Tychonoff 40
topological space
0-dimensional 68
basic space 30
recursively regular 38
metrizable 25
Polish space 26
Baire space 27
251
Index
Cantor space 27
Hilbert cube 27
Polish recursive space 37
Polish recursively presented space 28
second-countable 25
separable 25
tree 76
branch 76
pruned 76
universal set
good universal set for Σ01 subsets of ω k 17
good universal set for a Kleene class 58
good universal set for a Topological class 58
universal set for Σ01 subsets of ω 11
252
Bibliography
[Ale16] Pavel Alexandroff. Sur la puissance des ensembles mesurables B. Comptes
rendus hebdomadaires des séances de l’Académie des sciences, Paris, 162:323–
325, 1916. 182
[Fri71] Harvey M. Friedman. Higher set theory and mathematical practice. Annals
of Mathematical Logic, 2:325–357, 1971. 159
[GH82] Yuri Gurevich and Leo Harrington. Trees, automata, and games. In Proceed-
ings of the Fourteenth Annual ACM Symposium on Theory of Computing,
STOC ’82, pages 60–65. ACM, 1982. 156, 159
[KLS12] Alexander S. Kechris, Benedikt Löwe, and John R. Steel, editors. Wadge
Degrees and Projective Ordinals: The Cabal Seminar, Volume II. Cambridge
University Press, 2012. 253
[LS87] Alain Louveau and Jean Saint Raymond. Borel classes and closed games:
Wadge-type and Hurewicz-type results. Transactions of the American Math-
ematical Society, 304(2):431–467, 1987. 217
[LS12] Alain Louveau and Jean Saint Raymond. The Strength of Borel Wadge De-
terminacy. In Kechris et al. [KLS12], pages 74–101. First published in 1988.
217
253
Bibliography
[Mos80] Yannis N. Moschovakis. Descriptive set theory. North-Holland, 1980. 28, 50,
88, 139, 140, 164, 176
[Sus17] Mikhail Y. Suslin. Sur une définition des ensembles mesurables B sans nom-
bres transfinis. Comptes rendus hebdomadaires des séances de l’Académie des
sciences, Paris, 164:88–91, 1917. 80
[Wol55] Philip Wolfe. The strict determinateness of certain infinite games. Pacific
Journal of Mathematics, 5:841–847, 1955. 156
254