Descriptive Set Theory

and
Forcing:

How to prove theorems about Borel sets

the hard way.

Arnold W. Miller
Department of Mathematics
480 Lincoln Dr.
Van Vleck Hall
University of Wisconsin
Madison, WI. 53706
[email protected]
http://www.math.wisc.edu/∼miller
This page left blank.
 Note to the readers

Departing from the usual author’s statement-I would like to say that I
am not responsible for any of the mistakes in this document. Any mistakes
here are the responsibility of the reader. If anybody wants to point out a
mistake to me, I promise to respond by saying “but you know what I meant
to say, don’t you?”
These are lecture notes from a course I gave at the University of Wis-
consin during the Spring semester of 1993. Some knowledge of forcing is
assumed as well as a modicum of elementary Mathematical Logic, for exam-
ple, the Lowenheim-Skolem Theorem. The students in my class had a one
semester course, introduction to mathematical logic covering the complete-
ness theorem and incompleteness theorem, a set theory course using Kunen
[56], and a model theory course using Chang and Keisler [17]. Another good
reference for set theory is Jech [44]. Oxtoby [90] is a good reference for the
basic material concerning measure and category on the real line. Kuratowski
[59] and Kuratowski and Mostowski [60] are excellent references for classical
descriptive set theory. Moschovakis [89] and Kechris [54] are more modern
treatments of descriptive set theory.
The first part is devoted to the general area of Borel hierarchies, a subject
which has always interested me. The results in section 14 and 15 are new and
answer questions from my thesis. I have also included (without permission)
an unpublished result of Fremlin (Theorem 13.4).
Part II is devoted to results concerning the low projective hierarchy. It
ends with a theorem of Harrington from his thesis that is consistent to have
Π12 sets of arbitrary size.
e The general aim of part III and IV is to get to Louveau’s theorem. Along
the way many of the classical theorems of descriptive set theory are presented
“just-in-time” for when they are needed. This technology allows the reader
to keep from overfilling his or her memory storage device. I think the proof
given of Louveau’s Theorem 33.1 is also a little different. 1
Questions like “Who proved what?” always interest me, so I have included
my best guess here. Hopefully, I have managed to offend a large number of
1
In a randomly infinite Universe, any event occurring here and now with finite probabil-
ity must be occurring simultaneously at an infinite number of other sites in the Universe.
It is hard to evaluate this idea any further, but one thing is certain: if it is true then it is
certainly not original!– The Anthropic Cosmological Principle, by John Barrow and Frank
Tipler.
mathematicians.
AWM April 1995

Added April 2001: Several brave readers ignored my silly joke in the first
paragraph and sent me corrections and comments. Since no kind act should
go unpunished, let me say that any mistakes introduced into the text are
their fault.
Contents
1 What are the reals, anyway? 1

I On the length of Borel hierarchies 4

2 Borel Hierarchy 4

3 Abstract Borel hierarchies 10

4 Characteristic function of a sequence 13

5 Martin’s Axiom 16

6 Generic Gδ 18

7 α-forcing 22

8 Boolean algebras 28

9 Borel order of a field of sets 33

10 CH and orders of separable metric spaces 35

11 Martin-Solovay Theorem 38

12 Boolean algebra of order ω1 43

13 Luzin sets 48

14 Cohen real model 52

15 The random real model 65

16 Covering number of an ideal 73

II Analytic sets 78
17 Analytic sets 78
18 Constructible well-orderings 82

19 Hereditarily countable sets 84

20 Shoenfield Absoluteness 86

21 Mansfield-Solovay Theorem 88

22 Uniformity and Scales 90

23 Martin’s axiom and Constructibility 95

24 Σ12 well-orderings 97

25 Large Π12 sets 98

III Classical Separation Theorems 102

26 Souslin-Luzin Separation Theorem 102

27 Kleene Separation Theorem 104

28 Π11 -Reduction 107

29 ∆11 -codes 109

IV Gandy Forcing 113

30 Π11 equivalence relations 113

31 Borel metric spaces and lines in the plane 119

32 Σ11 equivalence relations 125

33 Louveau’s Theorem 130

34 Proof of Louveau’s Theorem 137

References 142
Index 152
1 WHAT ARE THE REALS, ANYWAY? 1

1 What are the reals, anyway?

Let ω = {0, 1, . . .} and let ω ω (Baire space) be the set of functions from
ω to ω. Let ω <ω be the set of all finite sequences of elements of ω. |s| is
the length of s, hi is the empty sequence, and for s ∈ ω <ω and n ∈ ω let
sˆn denote the sequence which starts out with s and has one more element
n concatenated onto the end. The basic open sets of ω ω are the sets of the
form:
[s] = {x ∈ ω ω : s ⊆ x}
for s ∈ ω <ω . A subset of ω ω is open iff it is the union of basic open subsets.
It is separable (has a countable dense subset) since it is second countable
(has a countable basis). The following defines a complete metric on ω ω :

0 if x = y
d(x, y) = 1
n+1
if x  n = y  n and x(n) 6= y(n)

Cantor space 2ω is the subspace of ω ω consisting of all functions from ω to

2 = {0, 1}. It is compact.

Theorem 1.1 (Baire [4]) ω ω is homeomorphic to the irrationals P.

Proof:
First replace ω by the integers Z. We will construct a mapping from Zω
to P. Enumerate the rationals Q = {qn : n ∈ ω}. Inductively construct a
sequence of open intervals hIs : s ∈ Z<ω i satisfying the following:

1. Ihi = R, and for s 6= hi each Is is a nontrivial open interval in R with

rational endpoints,

2. for every s ∈ Z<ω and n ∈ Z Isˆn ⊆ Is ,

3. the right end point of Isˆn is the left end point of Isˆn+1 ,

4. {Isˆn : n ∈ Z} covers all of Is except for their endpoints,

1
5. the length of Is is less than |s|
for s 6= hi, and

6. the nth rational qn is an endpoint of It for some |t| ≤ n + 1.

Descriptive Set Theory and Forcing 2

Define the function f : Zω → P as follows. Given x ∈ Zω the set

\
Ixn
n∈ω

must consist of a singleton irrational. It is nonempty because

closure(Ixn+1 ) ⊆ Ixn .

It is a singleton because their diameters shrink to zero.

So we can define f by
\
{f (x)} = Ixn .
n∈ω

The function f is one-to-one because if s and t are incomparable then Is and

It are disjoint. It is onto since for every u ∈ P and n ∈ ω there is a unique s
of length n with u ∈ Is . It is a homeomorphism because

f ([s]) = Is ∩ P

and the sets of the form Is ∩ P form a basis for P.


Note that the map given is also an order isomorphism from Zω with the
lexicographical order to P with it’s usual order.
We can identify 2ω with P (ω), the set of all subsets of ω, by identifying
a subset with its characteristic function. Let F = {x ∈ 2ω : ∀∞ n x(n) = 0}
(the quantifier ∀∞ stands for “for all but finitely many n”). F corresponds
to the finite sets and so 2ω \ F corresponds to the infinite subsets of ω which
we write as [ω]ω .

Theorem 1.2 ω ω is homeomorphic to [ω]ω .

Proof:
Let f ∈ ω ω and define F (f ) ∈ 2ω to be the sequence of 0’s and 1’s
determined by:
F (f ) = 0f (0) ˆ1ˆ0f (1) ˆ1ˆ0f (2) ˆ1ˆ · · ·
where 0f (n) refers to a string of length f (n) of zeros. The function F is a
one-to-one onto map from ω ω to 2ω \ F . It is a homeomorphism because
F ([s]) = [t] where t = 0s(0) ˆ1ˆ0s(1) ˆ1ˆ0s(2) ˆ1ˆ · · · ˆ0s(n) ˆ1 where |s| = n + 1.
Descriptive Set Theory and Forcing 3

Note that sets of the form [t] where t is a finite sequence ending in a one
form a basis for 2ω \ F .

I wonder why ω ω is called Baire space? The earliest mention of this I have
seen is in Sierpiński [99] where he refers to ω ω as the 0-dimensional space of
Baire. Sierpiński also says that Frechet was the first to describe the metric
d given above. Unfortunately, Sierpiński [99] gives very few references.2
The classical proof of Theorem 1.1 is to use “continued fractions” to
get the correspondence. Euler [19] proved that every rational number gives
rise to a finite continued fraction and every irrational number gives rise to
an infinite continued fraction. Brezinski [13] has more on the history of
continued fractions.
My proof of Theorem 1.1 allows me to remain blissfully ignorant3 of even
the elementary theory of continued fractions.
Cantor space, 2ω , is clearly named so because it is homeomorphic to
Cantor’s middle two thirds set.

2
I am indebted to John C. Morgan II for supplying the following reference and comment.
“Baire introduced his space in Baire [3]. Just as coefficients of linear equations evolved into
matrices the sequences of natural numbers in continued fraction developments of irrational
numbers were liberated by Baire’s mind to live in their own world.”
3
It is impossible for a man to learn what he thinks he already knows.-Epictetus
 4

Part I
On the length of Borel
hierarchies
2 Borel Hierarchy

Definitions. For X a topological space define Σ01 to be the open subsets of

X. For α > 1 define A ∈ Σ0α iff there exists a sequence
e hBn : n ∈ ωi
0
with each Bn ∈ Σβn for some βn < α such that
e
e
[
A= ∼ Bn
n∈ω

where ∼ B is the complement of B in X, i.e., ∼ B = X \ B. Define

Π0α = {∼ B : B ∈ Σ0α }
e e
and
∆0α = Σ0α ∩ Π0α .
e e e
The Borel subsets of X are defined by Borel(X) = α<ω1 Σ0α (X). It is clearly
S
the smallest family of sets containing the open subsets of X
e and closed under
countable unions and complementation.

Theorem 2.1 Σ0α is closed under countable unions, Π0α is closed under
countable intersections,
e and ∆0α is closed under complements.
e For any α,
e
Π0α (X) ⊆ Σ0α+1 (X) and Σ0α (X) ⊆ Π0α+1 (X).
e e e e
Proof:
That Σ0α is closed under countable unions is clear from its definition.
It follows efrom DeMorgan’s laws by taking complements that Π0α is closed
under countable intersections. e


Theorem 2.2 If f : X → Y is continuous and A ∈ Σ0α (Y ), then f −1 (A) is

in Σ0α (X). e
e
2 BOREL HIERARCHY 5

This is an easy induction since it is true for open sets (Σ01 ) and f −1 passes
over complements and unions. e

Theorem 2.2 is also, of course, true for Π0α or ∆0α in place of Σ0α .
e e e
Theorem 2.3 Suppose X is a subspace of Y , then

Σ0α (X) = {A ∩ X : A ∈ Σ0α (Y )}.

e e
Proof:
For Σ01 it follows from the definition of subspace. For α > 1 it is an easy
induction.
e

The class of sets Σ02 is also referred to as Fσ and the class Π02 as Gδ .
Theorem 2.3 is truee for Π0 in place of Σ0 , but not in general
e for ∆0α .
α α
For example, let X be the rationals
e in [0, 1] and
e Y be [0,1]. Then since X e is
0 0
countable every subset of X is Σ2 in X and hence ∆2 in X. If Z contained
in X is dense and codense thene Z is ∆02 in X (every e subset of X is), but
0
there is no ∆2 set Q in Y = [0, 1] whose e intersection with X is Z. (If Q
is Gδ and Fσ and contains Z then its comeager, but a comeager Fσ in [0, 1]
e
contains an interval.)

Theorem 2.4 For X a topological space and Π01 (X) ⊆ Π02 (X) (i.e., closed
sets are Gδ ), then e e

1. Π0α (X) ⊆ Π0α+1 (X),

e e
2. Σ0α (X) ⊆ Σ0α+1 (X), and hence
e e
3. Πα (X) ∪ Σ0α (X) ⊆ ∆0α+1 (X)
0
e e e
0
4. Σα is closed under finite intersections,
e
5. Π0α is closed under finite unions, and
e
6. ∆0α is closed under finite intersections, finite unions, and complements.
e
Proof:
2 BOREL HIERARCHY 6

Induction on α. For example, to see that Σ0α is closed under finite inter-
sections, use that e
[ [ [
( Pn ) ∩ ( Qn ) = (Pn ∩ Qm )
n∈ω n∈ω n,m∈ω

It follows by DeMorgan’s laws that Π0α is closed under finite unions. ∆0α is
closed under finite intersections, finite
e unions, and complements sincee it is
the intersection of the two classes.

In metric spaces closed sets are Gδ , since
\ 1
C= {x : ∃y ∈ C d(x, y) < }
n∈ω
n+1

for C a closed set.

The assumption that closed sets are Gδ is necessary since if

X = ω1 + 1

with the order topology, then the closed set consisting of the singleton point
{ω1 } is not Gδ ; in fact, it is not in the σ-δ-lattice generated by the open
sets (the smallest family containing the open sets and closed under countable
intersections and countable unions).
Williard [112] gives an example which is a second countable Hausdorff
space. Let X ⊆ 2ω be any nonBorel set. Let 2ω∗ be the space 2ω with the
smallest topology containing the usual topology and X as an open set. The
family of all sets of the form (B ∩ X) ∪ C where B, C are (ordinary) Borel
subsets of 2ω is the σ-δ-lattice generated by the open subsets of 2ω∗ , because:
\ \ \
(Bn ∩ X) ∪ Cn = ( ( Bn ∪ Cn ) ∩ X) ∪ Cn
n n n
[ [ [
(Bn ∩ X) ∪ Cn = (( Bn ) ∩ X) ∪ Cn .
n n n

Note that ∼ X is not in this σ-δ-lattice.

M.Laczkovich has pointed out to me that the class Π03 (X) where for the
ordered space X = ω1 + 1 is not closed under finite unions:
e
The elements of Π03 are of the form ∩∞ n=1 An where each An is either open
0
or Fσ . This implies that Π3 contains the open sets and the closed sets.
e
e
2 BOREL HIERARCHY 7

However, the union of an open set and a closed set is not necessarily in Π03 .
Let A be the set of isolated points of ω1 + 1 and let B = {ω1 }. Then Ae is
open and B is closed. But A ∪ B ∈ / Π03 . Suppose A ∪ B = ∩∞ n=1 An , where
each An is either open or Fσ . If An is open then ω1 ∈ An implies that An
e
contains an unbounded closed subset of ω1 . If An is Fσ then A ⊆ An implies
the same. Therefore ∩n An also contains an unbounded closed subset of ω1 .
Thus A ∩ B contains a countable limit point, which is impossible.

Theorem 2.5 (Lebesgue [63]) For every α with 1 ≤ α < ω1

Σ0α (2ω ) 6= Π0α (2ω ).

e e
The proof of this is a diagonalization argument applied to a universal
set. We will need the following lemma.

Lemma 2.6 Suppose X is second countable (i.e. has a countable base), then
for every α with 1 ≤ α < ω1 there exists a universal Σ0α set U ⊆ 2ω × X,
i.e., a set U which is Σ0α (2ω × X) such that for every A e∈ Σ0α (X) there exists
x ∈ 2ω such that A =eUx where Ux = {y ∈ X : (x, y) ∈ U }.e

Proof:
The proof is by induction on α. Let {Bn : n ∈ ω} be a countable base for
X. For α = 1 let
[
U = {(x, y) : ∃n (x(n) = 1 ∧ y ∈ Bn )} = ({x : x(n) = 1} × Bn ).
n

For α > 1 let βn be a sequence which sups up to α if α a limit, or equals

α − 1 if α is a successor. Let Un be a universal Σ0βn set. Let
e
n
hn, mi = 2 (2m + 1) − 1

be the usual pairing function which gives a recursive bijection between ω 2

and ω. For any n the map gn : 2ω × X → 2ω × X is defined by (x, y) 7→
(xn , y) where xn (m) = x(hn, mi). This map is continuous so if we define
Un∗ = gn−1 (Un ), then Un∗ is Σ0βn , and because the map x 7→ xn is onto it is also
a universal Σ0βn set. Now e define U by:
e
[
U= ∼ Un∗ .
n
2 BOREL HIERARCHY 8

U is universal for Σ0α because given any sequence Bn ∈ Σ0βn for n ∈ ω there
exists x ∈ 2ω suchethat for every n ∈ ω we have that Bne = (Un∗ )x = (Un )xn
(this is because the map x 7→ hxn : n < ωi takes 2ω onto (2ω )ω .) But then
[ [ [
Ux = ( ∼ Un∗ )x = ∼ (Un∗ )x = ∼ (Bn ).
n n n


Proof of Theorem 2.5:
Let U ⊆ 2ω × 2ω be a universal Σ0α set. Let
e
D = {x : hx, xi ∈ U }.

D is the continuous preimage of U under the map x 7→ hx, xi, so it is Σ0α ,

but it cannot be Π0α because if it were, then there would be x ∈ 2ω with
e
∼ D = Ux and thene x ∈ D iff hx, xi ∈ U iff x ∈ Ux iff x ∈∼ D.

Define ord(X) to be the least α such that Borel(X) = Σ0α (X). Lebesgue’s
theorem says that ord(X) = ω1 . Note that ord(X) = 1e if X is a discrete
space and that ord(Q) = 2.

Corollary 2.7 For any space X which contains a homeomorphic copy of 2ω

(i.e., a perfect set) we have that ord(X) = ω1 , consequently ω ω , R, and any
uncountable complete separable metric space have ord = ω1 .

Proof:
If the Borel hierarchy on X collapses, then by Theorem 2.3 it also collapses
on all subspaces of X. Every uncountable complete separable metric space
contains a perfect set (homeomorphic copy of 2ω ). To see this suppose X
is an uncountable complete separable metric space. Construct a family of
open sets hUs : s ∈ 2<ω i such that

1. Us is uncountable,

2. cl(Usˆ0 ) ∩ cl(Usˆ1 ) = ∅,

3. cl(Usˆi ) ⊆ Us for i=0,1, and

4. diameter of Us less than 1/|s|

2 BOREL HIERARCHY 9

Then the map f : 2ω → X defined so that

\
{f (x)} = Uxn
n∈ω

gives an embedding of 2ω into X.


Lebesgue [63] used universal functions instead of sets, but the proof is
much the same. Corollary 33.5 of Louveau’s Theorem shows that there can be
no Borel set which is universal for all ∆0α sets. Miller [82] contains examples
from model theory of Borel sets of arbitrary
e high rank.
0 0
The notation Σα , Πβ was first popularized by Addison [1]. I don’t know
if the “bold face” and “light face” notation is such a good idea, some copy
machines wipe it out. Consequently, I use

Σ0α
e
which is blackboard boldface.
3 ABSTRACT BOREL HIERARCHIES 10

3 Abstract Borel hierarchies

Suppose F ⊆ P (X) is a family of sets. Most of the time we would like to
think of F as a countable field of sets (i.e. closed under complements and
finite intersections) and so analogous to the family of clopen subsets of some
space.
We define the classes Π0α (F ) analogously. Let Π00 (F ) = F and for every
e there exists Bn ∈ Π0 for
α > 0 define A ∈ Π0α (F ) iff βn
e some βn < α such that
e \ e
A= ∼ Bn .
n∈ω

Define
• Σ0α (F ) = {∼ B : B ∈ Π0α (F )},
e e
• ∆0α (F ) = Π0α (F ) ∩ Σ0α (F ),
e eS e
• Borel(F ) = α<ω1 Σ0α (F ), and
e
• let ord(F ) be the least α such that Borel(F ) = Σ0α (F ).
e
Theorem 3.1 (Bing, Bledsoe, Mauldin [12]) Suppose F ⊆ P (2ω ) is a count-
able family such that Borel(2ω ) ⊆ Borel(F ). Then ord(F ) = ω1 .
Corollary 3.2 Suppose X is any space containing a perfect set and F is a
countable family of subsets of X with Borel(X) ⊆ Borel(F ). Then ord(F ) =
ω1 .
Proof:
Suppose 2ω ⊆ X and let F̂ = {A ∩ 2ω : A ∈ F }. By Theorem 2.3 we have
that Borel(2ω ) ⊆ Borel(F̂ ) and so by Theorem 3.1 we know ord(F̂ ) = ω1 .
But this implies ord(F ) = ω1 .

The proof of Theorem 3.1 is a generalization of Lebesgue’s universal set
argument. We need to prove the following two lemmas.
Lemma 3.3 (Universal sets) Suppose H ⊆ P (X) is countable and define
R = {A × B : A ⊆ 2ω is clopen and B ∈ H}.
Then for every α with 1 ≤ α < ω1 there exists U ⊆ 2ω × X with U ∈ Π0α (R)
such that for every A ∈ Π0α (H) there exists x ∈ 2ω with A = Ux . e
e
Descriptive Set Theory and Forcing 11

Proof:
This is proved exactly as Theorem 2.6, replacing the basis for X with H.
Note that when we replace Un by Un∗ it is necessary to prove by induction on
β that for every set A ∈ Π0β (R) and n ∈ ω that the set
e
A∗ = {(x, y) : (xn , y) ∈ A}

is also in Π0β (R).

 e

Lemma 3.4 Suppose H ⊆ P (2ω ), R is defined as in Lemma 3.3, and

Borel(2ω ) ⊆ Borel(H).

Then for every set A ∈ Borel(R) the set D = {x : (x, x) ∈ A} is in Borel(H).

Proof:
If A = B × C where B is clopen and C ∈ H, then D = B ∩ C which is
in Borel(H) by assumption. Note that
\ \
{x : (x, x) ∈ An } = {x : (x, x) ∈ An }
n n

and
{x : (x, x) ∈∼ A} =∼ {x : (x, x) ∈ A},
so the result follows by induction.

Proof of Theorem 3.1:
Suppose Borel(H) = Π0α (H) and let U ⊆ 2ω × 2ω be universal for Π0α (H)
given by Lemma 3.3. By e Lemma 3.4 the set D = {x : (x, x) ∈ Ue} is in
Borel(H) and hence its complement is in Borel(H) = Π0α (H). Hence we get
the same old contradiction: if Ux =∼ D, then x ∈ D iffe x ∈
/ D.


Theorem 3.5 (Reclaw) If X is a second countable space and X can be

mapped continuously onto the unit interval, [0, 1], then ord(X) = ω1 .

Proof:
Let f : X → [0, 1] be continuous and onto. Let B be a countable base for
X and let H = {f (B) : B ∈ B}. Since the preimage of an open subset of [0, 1]
Descriptive Set Theory and Forcing 12

is open in X it is clear that Borel([0, 1]) ⊆ Borel(H). So by Corollary 3.2

it follows that ord(H) = ω1 . But f maps the Borel hierarchy of X directly
over to the hierarchy generated by H, so ord(X) = ω1 .

Note that if X is a discrete space of cardinality the continuum then there
is a continuous map of X onto [0, 1] but ord(X) = 1.
The Cantor space 2ω can be mapped continuously onto [0, 1] via the map
∞
X x(n)
x 7→ .
n=0
2n+1

This map is even one-to-one except at countably many points where it is

two-to-one. It is also easy to see that R can be mapped continuously onto
[0, 1] and ω ω can be mapped onto 2ω . It follows that in Theorem 3.5 we may
replace [0, 1] by 2ω , ω ω , or R.
Myrna Dzamonja points out that any completely regular space Y which
contains a perfect set can be mapped onto [0, 1]. This is true because if
P ⊆ Y is perfect, then there is a continuous map f from P onto [0, 1]. But
since Y is completely regular this map extends to Y .
Reclaw did not publish his result, but I did, see Miller [86] and [87].
4 CHARACTERISTIC FUNCTION OF A SEQUENCE 13

4 Characteristic function of a sequence

The idea of a characteristic function of a sequence of sets is due to Ku-
ratowski and generalized the notion of a characteristic function of a set in-
troduced by de le Vallée-Poussin. The general notion was introduced by
Szpilrajn [106]. He also exploited it in Szpilrajn [107]. Szpilrajn changed his
name to Marczewski soon after the outbreak of World War II most likely to
hide from the Nazis. He kept the name Marczewski for the rest of his life.
Suppose F ⊆ P (X) is a countable field of sets (i.e. F is a family of
sets which is closed under complements in X and finite intersections). Let
F = {An : n ∈ ω}. Define c : X → 2ω by

1 if x ∈ An
c(x)(n) =
0 if x ∈/ An

Let Y = c(X), then there is a direct correspondence between F and

{C ∩ Y : C ⊆ 2ω clopen }.

In general, c maps X into 2|F | .

Theorem 4.1 (Szpilrajn [107]) If F ⊆ P (X) is a countable field of sets,

then there exists a subspace Y ⊆ 2ω such that ord(F ) = ord(Y ).

Proof:
If we define x ≈ y iff ∀n (x ∈ An iff y ∈ An ), then we see that members
of Borel(F ) respect ≈. The preimages of points of Y under c are exactly the
equivalence classes of ≈. The map c induces a bijection between X/ ≈ and
Y which takes the family F exactly to a clopen basis for the topology on Y .
Hence ord(F ) = ord(Y ).

The following theorem says that bounded Borel hierarchies must have a
top.

Theorem 4.2 (Miller [75]) Suppose F ⊆ P (X) is a field of sets and

ord(F ) = λ

where λ is a countable limit ordinal. Then there exists B ∈ Borel(F ) which

is not in Π0α (F ) for any α < λ.
e
Descriptive Set Theory and Forcing 14

Proof:
By the characteristic function of a sequence of sets argument we may
assume without loss of generality that

F = {C ∩ Y : C ⊆ 2κ clopen }.

A set C ⊆ 2κ is clopen iff it is a finite union of sets of the form

[s] = {x ∈ 2κ : s ⊆ x}

where s : D → 2 is a map with D ∈ [κ]<ω (i.e. D is a finite subset of κ). Note

that by induction for every A ∈ Borel(F ) there exists an S ∈ [κ]ω (called a
support of A) with the property that for every x, y ∈ 2κ if x  S = y  S
then (x ∈ A iff y ∈ A). That is to say, membership in A is determined by
restrictions to S.

Lemma 4.3 There exists a countable S ⊆ κ with the properties that α < λ
and s : D → 2 with D ∈ [S]<ω if ord(Y ∩ [s]) > α then there exists A in
Σ0α (F ) but not in ∆0α (F ) such that A ⊆ [s] and A is supported by S.
e e
Proof:
This is proved by a Lowenheim-Skolem kind of an argument.

By permuting κ around we may assume without loss of generality that
S = ω. Define
T = {s ∈ ω <ω : ord(Y ∩ [s]) = λ}.
Note that T is a tree, i.e., s ⊆ t ∈ T implies s ∈ T . Also for any s ∈ T either
sˆ0 ∈ T or sˆ1 ∈ T , because

[s] = [sˆ0] ∪ [sˆ1].

Since hi ∈ T it must be that T has an infinite branch. Let x : ω → 2 be such

that x  n ∈ T for all n < ω. For each n define

tn = (x  n)ˆ(1 − x(n))

and note that [

2κ = [x] ∪ [tn ]
n∈ω
Descriptive Set Theory and Forcing 15

is a partition of 2κ into clopen sets and one closed set [x].

Claim: For every α < λ and n ∈ ω there exists m > n with

ord(Y ∩ [tm ]) > α.

Proof:
Suppose not and let α and n witness this. Note that
[
[x  n] = [x] ∪ [tm ].
n≤m<ω

Since ord([x  n] ∩ Y ) = λ we know there exists A ∈ Σ0α+1 (F ) \ ∆0α+1 (F )

such that A ⊆ [x  n] and A is supported by S = ω. Since e A is supported
e by
ω either [x] ⊆ A or A is disjoint from [x]. But if ord([tm ] ∩ Y ) ≤ α for each
m > n, then [
A0 = (A ∩ [tm ])
n≤m<ω

is Σ0α (F ) and A = A0 or A = A0 ∪ [x] either of which is Σ0α (F ) (as long as

α> e 1). This proves the Claim. e

The claim allows us to construct a set which is not at a level below λ as
follows. Let αn < λ be a sequence unbounded in λ and let kn be a distinct
sequence with ord([tkn ] ∩ Y ) ≥ αn . Let An ⊆ [tkn ] be in Borel(F ) \ ∆0αn (F ).
Then ∪n An is not at any level bounded below λ. e


Question 4.4 Suppose R ⊆ P (X) is a ring of sets, i.e., closed under finite
unions and finite intersections. Let R∞ be the σ-ring generated by R, i.e., the
smallest family containing R and closed under countable unions and countable
intersections. For n ∈ ω define Rn as follows. R0 = R and let Rn+1 be the
family of countable unions (if n even)
S or family of countable intersections (if
n odd) of sets from Rn . If R∞ = n<ω Rn , then must there be n < ω such
that R∞ = Rn ?
5 MARTIN’S AXIOM 16

5 Martin’s Axiom
The following result is due to Rothberger [94] and Solovay [45][74]. The
forcing we use is due to Silver. However, it is probably just another view of
Solovay’s ‘almost disjoint sets forcing’.

Theorem 5.1 Assuming Martin’s Axiom if X is any second countable Haus-

dorff space of cardinality less than the continuum, then ord(X) ≤ 2 and, in
fact, every subset of X is Gδ .

Proof:
Let A ⊆ X be arbitrary and let B be a countable base for the topology on
X. The partial order P is defined as follows. p ∈ P iff p is a finite consistent
set of sentences of the form
◦
1. “x ∈
/ U n ” where x ∈ X \ A or
◦
2. “B ⊆U n ” where B ∈ B and n ∈ ω.
◦ ◦
Consistent means that there is not a pair of sentences “x ∈ / U n ”, “B ⊆U n ”
in p where x ∈ B. The ordering on P is reverse containment, i.e. p is stronger
◦
than q, p ≤ q iff p ⊇ q. The circle in the notation U n ’s means that it is the
name for the set Un which will be determined by the generic filter. For an
element x of the ground model we should use x̌ to denote the canonical name
of x, however to make it more readable we often just write x. For standard
references on forcing see Kunen [56] or Jech [44].
We call this forcing Silver forcing.

Claim: P satisfies the ccc.

Proof:
Note that since B is countable there are only countably many sentences
◦
of the type “B ⊆U n ”. Also if p and q have exactly the same sentences of
this type then p ∪ q ∈ P and hence p and q are compatible. It follows that
P is the countable union of filters and hence we cannot find an uncountable
set of pairwise incompatible conditions.

For x ∈ X \ A define
◦
Dx = {p ∈ P : ∃n “x ∈
/ U n ” ∈ p}.
Descriptive Set Theory and Forcing 17

For x ∈ A and n ∈ ω define

◦
Exn = {p ∈ P : ∃B ∈ B x ∈ B and “B ⊆U n ” ∈ p}.

Claim: Dx is dense for each x ∈ X \ A and Exn is dense for each x ∈ A and
n ∈ ω.
Proof:
To see that Dx is dense let p ∈ P be arbitrary. Choose n large enough so
◦ ◦
that Un is not mentioned in p, then (p ∪ {“x ∈ / U n ”}) ∈ P.
To see that Exn is dense let p be arbitrary and let Y ⊆ X \ A be the set
of elements of X \ A mentioned by p. Since x ∈ A and X is Hausdorff there
◦
exists B ∈ B with B ∩ Y = ∅ and x ∈ B. Then q = (p ∪ {“B ⊆U n ”}) ∈ P
and q ∈ Exn .

Since the cardinality of X is less than the continuum we can find a P-filter
G with the property that G meets each Dx for x ∈ X \ A and each Exn for
x ∈ A and n ∈ ω. Now define
[ ◦
Un = {B : “B ⊆U n ” ∈ G}.
T
Note that A = n∈ω Un and so A is Gδ in X.

Spaces X in which every subset is Gδ are called Q-sets.
The following question was raised during an email correspondence with
Zhou.
Question 5.2 Suppose every set of reals of cardinality ℵ1 is a Q-set. Then
is p > ω1 , i.e., is it true that for every family F ⊆ [ω]ω of size ω1 with the
finite intersection property there exists an X ∈ [ω]ω with X ⊆∗ Y for all
Y ∈ F?
It is a theorem of Bell [11] that p is the first cardinal for which MA for
σ-centered forcing fails. Another result along this line due to Alan Taylor is
that p is the cardinality of the smallest set of reals which is not a γ-set, see
Galvin and Miller [30].
Fleissner and Miller [23] show it is consistent to have a Q-set whose union
with the rationals is not a Q-set.
For more information on Martin’s Axiom see Fremlin [27]. For more on
Q-sets, see Fleissner [24] [25], Miller [83] [87], Przymusinski [92], Judah and
Shelah [46] [47], and Balogh [5].
6 GENERIC Gδ 18

6 Generic Gδ
It is natural4 to ask
“What are the possibly lengths of Borel hierarchies?”
In this section we present a way of forcing a generic Gδ .
Let X be a Hausdorff space with a countable base B. Consider the fol-
lowing forcing notion.
p ∈ P iff it is a finite consistent set of sentences of the form:
◦
1. “B ⊆U n ” where B ∈ B and n ∈ ω, or
◦
/ U n ” where x ∈ X and n ∈ ω, or
2. “x ∈
T ◦
3. “x ∈ n<ω U n ” where x ∈ X.
◦ ◦
Consistency means that we cannot say that both “B ⊆U n ” and “x ∈ / U n ” if it
◦ T ◦
happens that x ∈ B and we cannot say both “x ∈ / U n ” and “x ∈ n<ω U n ”.
The ordering is reverse inclusion. A P filter G determines a Gδ set U as
follows: Let [ ◦
Un = {B ∈ B : “B ⊆U n ” ∈ G}.
T
Let U = n Un . If G is P-generic over V , a density argument shows that for
every x ∈ X we have that
\ ◦
x ∈ U iff “x ∈ U n ” ∈ G.
n<ω

Note that U is not in V (as long as X is infinite). For suppose p ∈ P and

A ⊆ X is in V is such that
◦
p |`U = Ǎ.
Since X is infinite there exist x ∈ X which is not mentioned in p. Note that
T ◦ ◦
p0 = p ∪ {“x ∈ n<ω U n ”} is consistent and also p1 = p ∪ {“x ∈ / U n ”} is
consistent for all sufficiently large n (i.e. certainly for Un not mentioned in
◦ ◦
p.) But p0 |` x ∈U and p1 |` x ∈
/ U , and since x is either in A or not in A
we arrive at a contradiction.
4
‘Gentlemen, the great thing about this, like most of the demonstrations of the higher
mathematics, is that it can be of no earthly use to anybody.’ -Baron Kelvin
Descriptive Set Theory and Forcing 19

In fact, U is not Fσ in the extension (assuming X is uncountable). To

see this we will first need to prove that P has ccc.

Lemma 6.1 P has ccc.

Proof:
Note that p and q are compatible iff (p ∪ q) ∈ P iff (p ∪ q) is a consistent
set of sentences. Recall that there are three types of sentences:
◦
1. B ⊆U n
◦
2. x ∈
/U n
T ◦
3. x ∈ n<ω Un
where B ∈ B, n ∈ ω, and x ∈ X. Now if for contradiction A were an
uncountable antichain, then since there are only countably many sentences
of type 1 above we may assume that all p ∈ A have the same set of type
1 sentences. Consequently for each distinct pair p, q ∈ A there must be an
◦ T ◦
x ∈ X and n such that either “x ∈ / U n ” ∈ p and “x ∈ n<ω U n ” ∈ q
or vice-versa. For each p ∈ A let Dp be the finitely many elements of X
mentioned by p and let sp : Dp → ω be defined by
( T ◦
0 if “x ∈ n<ω U n ” ∈ p
sp (x) = ◦
n + 1 if “x ∈/U n ” ∈ p

But now {sp : p ∈ A} is an uncountable family of pairwise incompatible finite

partial functions from X into ω which is impossible. (FIN(X, ω) has the ccc,
see Kunen [56].)

If V [G] is a generic extension of a model V which contains a topological
space X, then we let X also refer to the space in V [G] whose topology is
generated by the open subsets of X which are in V .

Theorem 6.2 (Miller [75]) Suppose X in V is an uncountable Hausdorff

space with countable base B and G is P-generic over V . Then in V [G] the
Gδ set U is not Fσ .
Descriptive Set Theory and Forcing 20

Proof:
We call this argument the old switcheroo. Suppose for contradiction
\ ◦ [ ◦ ◦
p |` U n= C n where C n are closed in X .
n∈ω n∈ω

For Y ⊆ X let P(Y ) be the elements of P which only mention y ∈ Y in type

2 or 3 statements. Let Y ⊆ X be countable such that
1. p ∈ P(Y ) and
2. for every n and B ∈ B there exists a maximal antichain A ⊆ P(Y )
◦
which decides the statement “B∩ C n = ∅”.
Since X is uncountable there exists x ∈ X \ Y . Let
\ ◦
q = p ∪ {“x ∈ U n ”}.
n∈ω

Since q extends p, clearly

[ ◦
q |` x ∈ Cn
n∈ω

so there exists r ≤ q and n ∈ ω so that

◦
r |` x ∈C n .

Let \ ◦
r = r0 ∪ {“x ∈ U n ”}
n∈ω

where r0 does not mention x. Now we do the switch. Let

◦
t = r0 ∪ {“x ∈
/ U m ”}

where m is chosen sufficiently large so that t is a consistent condition. Since

\ ◦
t |` x ∈
/ Un
n∈ω

we know that ◦
t |` x ∈
/Cn .
Consequently there exist s ∈ P(Y ) and B ∈ B such that
Descriptive Set Theory and Forcing 21

1. s and t are compatible,

◦
2. s |` B∩ C n = ∅, and

3. x ∈ B.

But s and r are compatible, because s does not mention x. This is a contra-
◦ ◦
diction since s ∪ r |` x ∈C n and s ∪ r |` x ∈
/ C n.

7 α-FORCING 22

7 α-forcing
In this section we generalize the forcing which produced a generic Gδ to
arbitrarily high levels of the Borel hierarchy. Before doing so we must prove
some elementary facts about well-founded trees.
Let OR denote the class of all ordinals. Define T ⊆ Q<ω to be a tree iff
s ⊆ t ∈ T implies s ∈ T . Define the rank function r : T → OR ∪ {∞} of
T as follows:

1. r(s) ≥ 0 iff s ∈ T ,

2. r(s) ≥ α + 1 iff ∃q ∈ Q r(sˆq) ≥ α,

3. r(s) ≥ λ (for λ a limit ordinal) iff r(s) ≥ α for every α < λ.

Now define r(s) = α iff r(s) ≥ α but not r(s) ≥ α + 1 and r(s) = ∞ iff
r(s) ≥ α for every ordinal α.
Define [T ] = {x ∈ Qω : ∀n x  n ∈ T }. We say that T is well-founded
iff [T ] = ∅.

Theorem 7.1 T is well-founded iff r(hi) ∈ OR.

Proof:
It follows easily from the definition that if r(s) is an ordinal, then

r(s) = sup{r(sˆq) + 1 : q ∈ Q}.

Hence, if r(hi) = α ∈ OR and x ∈ [T ], then

r(x  (n + 1)) < r(x  n)

is a descending sequence of ordinals.

On the other hand, if r(s) = ∞ then for some q ∈ Q we must have
r(sˆq) = ∞. So if r(hi) = ∞ we can construct (using the axiom of choice) a
sequence sn ∈ T with r(sn ) = ∞ and sn+1 = sn ˆx(n). Hence x ∈ [T ].

Definition. T is a nice α-tree iff

1. T ⊆ ω <ω is a tree,

2. r : T → (α + 1) is its rank function (r(hi) = α),

Descriptive Set Theory and Forcing 23

3. if r(s) > 0, then for every n ∈ ω sˆn ∈ T ,

4. if r(s) = β is a successor ordinal, then for every n ∈ ω r(sˆn) = β − 1,
and
5. if r(s) = λ is a limit ordinal, then r(sˆ0) ≥ 2 and r(sˆn) increases to
λ as n → ∞.
It is easy to see that for every α < ω1 nice α-trees exist. For X a Hausdorff
space with countable base, B, and T a nice α-tree (α ≥ 2), define the partial
order P = P(X, B, T ) which we call α-forcing as follows:
p ∈ P iff p = (t, F ) where
1. t : D → B where D ⊆ T 0 = {s ∈ T : r(s) = 0} is finite,
2. F ⊆ T >0 × X is finite where
T >0 = T \ T 0 = {s ∈ T : r(s) > 0},

3. if (s, x), (sˆn, y) ∈ F , then x 6= y, and

4. if (s, x) ∈ F and t(sˆn) = B, then x ∈
/ B.
The ordering on P is given by p ≤ q iff tp ⊇ tq and Fp ⊇ Fq .
Lemma 7.2 P has ccc.
Proof:
Suppose A is uncountable antichain. Since there are only countably many
different tp without loss we may assume that there exists t such that tp = t
for all p ∈ A. Consequently for p, q ∈ A the only thing that can keep p ∪ q
from being a condition is that there must be an x ∈ X and an s, sˆn ∈ T >0
such that
(s, x), (sˆn, x) ∈ (Fp ∪ Fq ).
But now for each p ∈ A let Hp : X → [T >0 ]<ω be the finite partial function
defined by
Hp (x) = {s ∈ T >0 : (s, x) ∈ Fp }
where domain Hp is {x : ∃s ∈ T >0 (s, x) ∈ Fp }. Then {Hp : p ∈ A} is
an uncountable antichain in the order of finite partial functions from X to
[T >0 ]<ω , a countable set.

Define for G a P-filter the set Us ⊆ X for s ∈ T as follows:
Descriptive Set Theory and Forcing 24

1. for s ∈ T 0 let Us = B iff ∃p ∈ G such that tp (s) = B and

2. for s ∈ T >0 let Us = n∈ω ∼ U sˆn

T

Note that Us is a Π0β (X)-set where r(s) = β.

e
Lemma 7.3 If G is P-generic over V then in V [G] we have that for every
x ∈ X and s ∈ T >0

x ∈ Us ⇐⇒ ∃p ∈ G (s, x) ∈ Fp .

Proof:
First suppose that r(s) = 1 and note that the following set is dense:

D = {p ∈ P : (s, x) ∈ Fp or ∃n∃B ∈ B x ∈ B and tp (sˆn) = B}.

To see this let p ∈ P be arbitrary. If (s, x) ∈ Fp then p ∈ D and we are

already done. If (s, x) ∈
/ Fp then let

Y = {y : (s, y) ∈ Fp }.

Choose B ∈ B with x ∈ B and Y disjoint from B. Choose sˆn not in the

domain of tp , and let q = (tq , Fp ) be defined by tq = tp ∪ (sˆn, B). So q ≤ p
and q ∈ D. Hence D is dense. T
Now by definition x ∈ Us iff x ∈ n∈ω ∼ U sˆn . So let G be a generic
filter and p ∈ G ∩ D. If (s, x) ∈ Fp then we know that for every q ∈ G and
for every n, if tq (sˆn) = B then x ∈ / B. Consequently, x ∈ Us . On the other
hand if tp (sˆn) = B where x ∈ B, then x ∈ / Us and for every q ∈ G it must
be that (s, x) ∈/ Fq (since otherwise p and q would be incompatible).

Now suppose r(s) > 1. In this case note that the following set is dense:

E = {p ∈ P : (s, x) ∈ Fp or ∃n (sˆn, x) ∈ Fp }.

To see this let p ∈ P be arbitrary. Then either (s, x) ∈ Fp and already

p ∈ E or by choosing n large enough q = (tp , Fp ∪ {(sˆn, x)}) ∈ E. (Note
r(sˆn) > 0.)
Now assume the result is true for all Usˆn . Let p ∈ G ∩ E. If (s, x) ∈ Fp
then for every q ∈ G and n we have (sˆn, x) ∈ / Fq and so by induction
x ∈/ Usˆn and so x ∈ Us . On the other hand if (sˆn, x) ∈ Fp , then by
Descriptive Set Theory and Forcing 25

induction x ∈ Usˆn and so x ∈ / Us , and so again for every q ∈ G we have

(s, x) ∈
/ Fq .

The following lemma is the heart of the old switcheroo argument used
in Theorem 6.2. Given any Q ⊂ X define the rank(p, Q) as follows:

rank(p, Q) = max{r(s) : (s, x) ∈ Fp for some x ∈ X \ Q}.

Lemma 7.4 (Rank Lemma). For any β ≥ 1 and p ∈ P there exists p̂

compatible with p such that

1. rank(p̂, Q) < β + 1 and

2. for any q ∈ P if rank(q, Q) < β, then

p̂ and q compatible implies p and q compatible.

Proof:
Let p0 ≤ p be any extension which satisfies: for any (s, x) ∈ Fp and n ∈ ω,
if r(s) = λ > β is a limit ordinal and r(sˆn) < β + 1, then there exist m ∈ ω
such that (sˆnˆm, x) ∈ Fp0 . Note that since r(sˆn) is increasing to λ there
are only finitely many (s, x) and sˆn to worry about. Also r(sˆnˆm) > 0 so
this is possible to do.
Now let p̂ be defined as follows:

tp̂ = tp

and
Fp̂ = {(s, x) ∈ Fp0 : x ∈ Q or r(s) < β + 1}.
Suppose for contradiction that there exists q such that rank(q, Q) < β, p̂
and q compatible, but p and q incompatible. Since p and q are incompatible
either

1. there exists (s, x) ∈ Fq and tp (sˆn) = B with x ∈ B, or

2. there exists (s, x) ∈ Fp and tq (sˆn) = B with x ∈ B, or

3. there exists (s, x) ∈ Fp and (sˆn, x) ∈ Fq , or

4. there exists (s, x) ∈ Fq and (sˆn, x) ∈ Fp .

Descriptive Set Theory and Forcing 26

(1) cannot happen since tp̂ = tp and so p̂, q would be incompatible. (2) cannot
happen since r(s) = 1 and β ≥ 1 means that (s, x) ∈ Fp̂ and so again p̂ and
q are incompatible. If (3) or (4) happens for x ∈ Q then again (in case 3)
(s, x) ∈ Fp̂ or (in case 4) (sˆn, x) ∈ Fp̂ and so p̂, q incompatible.
So assume x ∈ / Q. In case (3) by the definition of rank(q, Q) < β we know
that r(sˆn) < β. Now since T is a nice tree we know that either r(s) ≤ β
and so (s, x) ∈ Fp̂ or r(s) = λ a limit ordinal. Now if λ ≤ β then (s, x) ∈ Fp̂ .
If λ > β then by our construction of p0 there exist m with (sˆnˆm, x) ∈ Fp̂
and so p̂, q are incompatible. Finally in case (4) since x ∈ / Q and so r(s) < β
we have that r(sˆn) < β and so (sˆn, x) ∈ Fp̂ and so p̂, q are incompatible.

Intuitively, it should be that statements of small rank are forced by con-
ditions of small rank. The next lemma will make this more precise. Let
L∞ (Pα : α < κ) be the infinitary propositional logic with {Pα : α < κ} as
the atomic sentences. Let Π0 -sentences be the atomic ones, S {Pα : α < κ}.
For any β > 0 let θ be a Πβ -sentence iff there exists Γ ⊆ δ<β Πδ -sentences
and ^^
θ= ¬ψ.
ψ∈Γ

Models for this propositional language can naturally be regarded as subsets

Y of κ where we define

1. Y |= Pα iff α ∈ Y ,

2. Y |= ¬θ iff not Y |= θ, and

VV
3. Y |= Γ iff Y |= θ for every θ ∈ Γ.

Lemma 7.5 (Rank and Forcing Lemma) Suppose rank : P → OR is any

◦
function on a poset P which satisfies the Rank Lemma 7.4. Suppose |`P Y ⊂ κ
and for every p ∈ P and α < κ if
◦
p |` α ∈Y

then there exist p̂ compatible with p such that rank(p̂) = 0 and

◦
p̂ |` α ∈Y .
Descriptive Set Theory and Forcing 27

Then for every Πβ -sentence θ (in the ground model) and every p ∈ P, if
◦
p |` “ Y |= θ”

then there exists p̂ compatible with p such that rank(p̂) ≤ β and

◦
p̂ |` “ Y |= θ”.

Proof:
This is one of those lemmas whose statement is longer than its proof. The
proof is induction on β andVVfor β = 0 the conclusion
S is true by assumption.
So suppose β > 0 and θ = ψ∈Γ ¬ψ where Γ ⊆ δ<β Πδ -sentences. By the
rank lemma there exists p̂ compatible with p such that rank(p̂) ≤ β and for
every q ∈ P with rank(q) < β if p̂, q compatible then p, q compatible. We
claim that ◦
p̂ |` “ Y |= θ”.
Suppose not. Then there exists r ≤ p̂ and ψ ∈ Γ such that
◦
r |` “ Y |= ψ”.

By inductive assumption there exists r̂ compatible with r such that

rank(r̂) < β

such that ◦
r̂ |` “ Y |= ψ”.
But r̂, p̂ compatible implies r̂, p compatible, which is a contradiction because
θ implies ¬ψ and so
◦
p |` “ Y |= ¬ψ”.


Some earlier uses of rank in forcing arguments occur in Steel’s forcing,
see Steel [108], Friedman [29], and Harrington [36]. It also occurs in Silver’s
analysis of the collapsing algebra, see Silver [101].
In Miller [77] α-forcing for all α is used to construct generic Souslin sets.
8 BOOLEAN ALGEBRAS 28

8 Boolean algebras
In this section we consider the length of Borel hierarchies generated by a
subset of a complete boolean algebra. We find that the generators of the
complete boolean algebra associated with α-forcing generate it in exactly
α + 1 steps. We start by presenting some background information.
Let B be a cBa, i.e, complete boolean algebra. This means that in
addition to being a boolean algebra, infinite
P sums and products, also exist;
i.e., for any C ⊆ B there exists b (denoted C) such that
1. c ≤ b for every c ∈ C and
2. for every d ∈ B if c ≤ d for every c ∈ C, then b ≤ d.
Q P
Similarly we define C = − c∈C −c where −c denotes the complement of
c in B.
A partial order P is separative iff for any p, q ∈ P we have
p ≤ q iff ∀r ∈ P(r ≤ p implies q, r compatible).
Theorem 8.1 (Scott, Solovay see [44]) A partial order P is separative iff
there exists a cBa B such that P ⊆ B is dense in B, i.e. for every b ∈ B if
b > 0 then there exists p ∈ P with p ≤ b.
It is easy to check that the α-forcing P is separative (as long as B is
infinite): If p 6≤ q then either
1. tp does not extend tq , so there exists s such that tq (s) = B and either
s not in the domain of tp or tp (s) = C where C 6= B and so in either
case we can find r ≤ p with r, q incompatible, or
2. Fp does not contain Fq , so there exists (s, x) ∈ (Fq \ Fp ) and we can
either add (sˆn, x) for sufficiently large n or add tr (sˆn) = B for some
sufficiently large n and some B ∈ B with x ∈ B and get r ≤ p which is
incompatible with q.
The elegant (but as far as I am concerned mysterious) approach to forcing
using complete boolean algebras contains the following facts:
1. for any sentence θ in the forcing language
X X
[| θ |] = {b ∈ B : b |` θ} = {p ∈ P : p |` θ}
where P is any dense subset of B,
Descriptive Set Theory and Forcing 29

2. p |` θ iff p ≤ [| θ |],

3. [| ¬θ |] = −[| θ |],

4. [| θ ∧ ψ |] = [| θ |] ∧[| ψ |],

5. [| θ ∨ ψ |] = [| θ |] ∨[| ψ |],

6. for any set X in the ground model,

Y
[| ∀x ∈ X̌ θ(x) |] = [| θ(x̌ |].
x∈X

Definitions. For B a cBa and C ⊆ B define:

Π00 (C) = CQand
Πe 0 (C) = { Γ : Γ ⊆ {−c : c ∈ S 0
α β<α Πβ (C)}} for α > 0.
e e
ord(B) = min{α : ∃C ⊆ B countable with Π0α (C) = B}.
e
Theorem 8.2 (Miller [75]) For every α ≤ ω1 there exists a countably gen-
erated ccc cBa B with ord(B) = α.

Proof:
Let P be α-forcing and B be the cBa given by the Scott-Solovay Theo-
rem 8.1. We will show that ord(B) = α + 1.
Let
C = {p ∈ P : Fp = ∅}.
C is countable and we claim that P ⊆ Π0α (C). Since B = Σ01 (P) this will
imply that B = Σ0α+1 (C) and so ord(B) ≤e α + 1. e
e for any s ∈ T with r(s) = 0 and x ∈ X,
First note that
X
[| x ∈ Us |] = {p ∈ C : ∃B ∈ B tp (s) = B and x ∈ B}.

By Lemma 7.3 we know for generic filters G that for every x ∈ X and
s ∈ T >0
x ∈ Us ⇐⇒ ∃p ∈ G (s, x) ∈ Fp .
Hence [| x ∈ Us |] = h∅, {(s, x)}i since if they are not equal, then

b = [| x ∈ Us |] ∆ h∅, {(s, x)}i > 0,

Descriptive Set Theory and Forcing 30

but letting G be a generic ultrafilter with b in it would lead to a contradiction.

We get that for r(s) > 0:
\ Y
h∅, {(s, x)}i = [| x ∈ Us |] = [| x ∈ ∼ U sˆn |] = −[| x ∈ Usˆn |].
n∈ω n∈ω

Remembering that for r(sˆn) = 0 we have [| x ∈ Usˆn |] ∈ Σ01 (C), we see by

induction that for every s ∈ T >0 if r(s) = β then e

h∅, {(s, x)}i ∈ Π0β (C).

e
For any p ∈ P Y
p = htp , ∅i ∧ h∅, {(s, x)}i.
(s,x)∈Fp

So we have that p ∈ Π0α (C).

e
Now we will see that ord(B) > α. We use the following Lemmas.
B+ are the nonzero elements of B.

Lemma 8.3 If r : P → OR is a rank function, i.e. it satisfies the Rank

Lemma 7.4 and in addition p ≤ q implies r(p) ≤ r(q), then if P is dense in
the cBa B then r extends to r∗ on B+ :
X
r∗ (b) = min{β ∈ OR : ∃C ⊆ P : b = C and ∀p ∈ C r(p) ≤ β}

and still satisfies the Rank Lemma.

Proof:
Easy induction.


Lemma 8.4 If r : B+ → ord is a rank function and E ⊆ B is a countable

collection of rank zero elements, then for any a ∈ Π0γ (E) and a 6= 0 there
exists b ≤ a with r(b) ≤ γ. e

Proof: ◦
To see this let E = {en : n ∈ ω} and let Y be a name for the set in the
generic extension
Y = {n ∈ ω : en ∈ G}.
Descriptive Set Theory and Forcing 31

◦
Note that en = [| n ∈Y |]. For elements b of B in the complete subalgebra
generated by E let us associate sentences θb of the infinitary propositional
logic L∞ (Pn : n ∈ ω) as follows:

θen = Pn

θ−b = ¬θb
^^
θQ R = θr
r∈R

Note that [| Y |= θb |] = b and if b ∈ Π0γ (E) then θb is a Πγ -sentence. The

Rank and Forcing Lemma 7.5 gives us,e by translating
p |` Y |= θb into p ≤ [| Y |= θb |] = b) that:

For any γ ≥ 1 and p ≤ b ∈ Π0γ (E) there exists a p̂ compatible

with p such that p̂ ≤ b and r(p̂)
e ≤ γ.


Now we use the lemmas to see that ord(B) > α.
Given any countable E ⊆ B, let Q ⊆ X be P countable so that for any
e ∈ E there exists H ⊆ P countable so that e = H and for every p ∈ H
we have rank(p, Q) = 0. Let x ∈ X \ Q be arbitrary; then we claim:

/ Σ0α (E).
[| x ∈ Uhi |] ∈
e
We have chosen Q so that r(p) = rank(p, Q) = 0 for any p ∈ E so the
hypothesis of Lemma 8.4 is satisfied. Suppose for contradiction that

[| x ∈ Uhi |] = b ∈ Σ0α (E).

e
0
P
Let b = n∈ω bn where each bn is Πγn (C) for some γn < α. For some n and
p ∈ P we would have p ≤ bn . Bye Lemma 8.4 we have that there exists p̂
with p̂ ≤ bn ≤ b = [| x ∈ Uhi |] and rank(p̂, Q) ≤ γn . But by the definition of
rank(p̂, Q) the pair (hi, x) is not in Fp̂ , but this contradicts

p̂ ≤ bn ≤ b = [| x ∈ Uhi |] = h∅, {(hi, x)}i.

This takes care of all countable successor ordinals. (We leave the case
of α = 0, 1 for the reader to contemplate.) For λ a limit ordinal take αn
Descriptive Set Theory and Forcing 32

P
increasing to λ and let P = n<ω Pαn be the direct sum, where Pαn is αn -
forcing. Another way to describe essentially the same thing is as follows: Let
Pλ be λ-forcing. Then take P to be the subposet of Pλ such that hi doesn’t
occur, i.e.,
P = {p ∈ Pλ : ¬∃x ∈ X (hi, x) ∈ Fp }.
Now if P is dense in the cBa B, then ord(B) = λ. This is easy to see,
because
S for each p ∈ P there exists β < λ with p ∈ Π0β (C). Consequently,
P ⊆ β<λ Π0β (C) and so since B = Σ01 (P) we get B = Σe0λ (C). Similarly to the
other argument
e we see that for anye countable E we ecan choose a countable
Q ⊆ X such for any s ∈ T with 2 ≤ r(s) = β < λ (so s 6= hi) we have that
[| x ∈ Us |] is not Σ0β (E). Hence ord(B) = λ.
For ord(B) = eω1 we postpone until section 12.

9 BOREL ORDER OF A FIELD OF SETS 33

9 Borel order of a field of sets

In this section we use the Sikorski-Loomis representation theorem to transfer
the abstract Borel hierarchy on a complete boolean algebra into a field of
sets.
A family F ⊆ P (X) is a σ-field iff it contains the empty set and is closed
under countable unions and complements in X. I ⊆ F is a σ-ideal in F iff

1. I contains the empty set,

2. I is closed under countable unions,

3. A ⊆ B ∈ I and A ∈ F implies A ∈ I, and

4. X ∈
/ I.

F/I is the countably complete boolean algebra formed by taking F and

modding out by I, i.e. A ≈ B iff A∆B ∈ I. For A ∈ F we use [A] or [A]I to
denote the equivalence class of A modulo I.

Theorem 9.1 (Sikorski,Loomis, see [100] section 29) For any countably
complete boolean algebra B there exists a σ-field F and a σ-ideal I such
that B is isomorphic to F/I.

Proof:
Recall that the Stone space of B, stone(B), is the space of ultrafilters u
on B with the topology generated by the clopen sets of the form:

[b] = {u ∈ stone(B) : b ∈ u}.

This space is a compact Hausdorff space in which the field of clopen sets
exactly corresponds to B. B is countably complete means that for any se-
quence
{bn : n < ω} in B
P
there exists b ∈ B such that b = n∈ω bn . This translates to the fact that
given any countable family of clopen
S sets {Cn : n ∈ ω} in stone(B)S there
exists a clopen set C such that n∈ω Cn ⊆ C and the closed set C \ n∈ω Cn
cannot contain a clopen set, hence it has no interior, so it is nowhere dense.
Let F be the σ-field generated by the clopen subsets of stone(B). Let I be
the σ-ideal generated by the closed nowhere dense subsets of F (i.e. the ideal
Descriptive Set Theory and Forcing 34

of meager sets). The Baire category theorem implies that no nonempty open
subset of a compact Hausdorff space is meager, so st(B) ∈ / I and the same
holds for any nonempty clopen subset of stone(B). Since the countable union
of clopen sets is equivalent to a clopen set modulo I it follows that the map
C 7→ [C] is an isomorphism taking the clopen algebra of stone(B) onto F/I.


Shortly after I gave a talk about my boolean algebra result (Theorem
8.2), Kunen pointed out the following result.

Theorem 9.2 (Kunen see [75]) For every α ≤ ω1 there exists a field of sets
H such that ord(H) = α.

Proof:
Clearly we only have to worry about α with 2 < α < ω1 . Let B be the
complete boolean algebra given by Theorem 8.2. Let B ' F/I where F is a
σ-field of sets and I a σ-ideal. Let C ⊆ F/I be a countable set of generators.
Define
H = {A ∈ F : [A]I ∈ C}.
By induction on β it is easy to prove that for any Q ∈ F :

Q ∈ Σ0β (H) iff [Q]I ∈ Σ0β (C).

e e
From which it follows that ord(H) = α.

Note that there is no claim that the family H is countable. In fact, it
is consistent (Miller [75]) that either ord(H) ≤ 2 or ord(H) = ω1 for every
countable H.
10 CH AND ORDERS OF SEPARABLE METRIC SPACES 35

10 CH and orders of separable metric spaces

In this section we prove that assuming CH that there exists countable field
of sets of all possible Borel orders, which we know is equivalent to existence
of separable metric spaces of all possible orders. We will need a sharper form
of the representation theorem.
Theorem 10.1 (Sikorski, see [100] section 31) B is a countably generated
ccc cBa iff there exists a ccc σ-ideal I in Borel(2ω ) such that B ' Borel(2ω )/I.
Furthermore if B is generated by the countable set C ⊆ B, then this isomor-
phism can be taken so as to map the clopen sets mod I onto C.
Proof:
To see that Borel(2ω )/I is countably generated is trivial since the clopen
sets modulo I generate it. A general theorem of Tarski is that any κ-complete
κ-cc boolean algebra is complete.
For the other direction, we may assume by using the Sikorski-Loomis
Theorem, that B is F/J where F is a σ-field and J a σ-ideal in F . Since B is
countably generated there exists Cn ∈ F for n ∈ ω such that {[Cn ] : n ∈ ω}
generates F/J where [C] denotes the equivalence class of C modulo J. Now
let h : X → 2ω be defined by

1 if x ∈ Cn
h(x)(n) =
0 if x ∈
/ Cn
and define φ : Borel(2ω ) → F by
φ(A) = h−1 (A).
Define I = {A ∈ Borel(2ω ) : φ(A) ∈ J}. Finally, we claim that
φ̂ : Borel(2ω )/I → F/I defined by φ̂([A]I ) = [φ(A)]J
is an isomorphism of the two boolean algebras.

For I a σ-ideal in Borel(2ω ) we say that X ⊆ 2ω is an I-Luzin set
iff for every A ∈ I we have that X ∩ A is countable. We say that X is
super-I-Luzin iff X is I-Luzin and for every B ∈ Borel(2ω ) \ I we have that
B ∩ X 6= ∅. The following Theorem was first proved by Mahlo [70] and later
by Luzin [69] for the ideal of meager subsets of the real line. Apparently,
Mahlo’s paper was overlooked and hence these kinds of sets have always been
referred to as Luzin sets.
Descriptive Set Theory and Forcing 36

Theorem 10.2 (Mahlo [70]) CH. Suppose I is a σ-ideal in Borel(2ω ) con-

taining all the singletons. Then there exists a super-I-Luzin set.

Proof:
Let
I = {Aα : α < ω1 }
and let
Borel(2ω ) \ I = {Bα : α < ω1 }.
Inductively choose xα ∈ 2ω so that
[
xα ∈ Bα \ ({xβ : β < α} ∪ Aα ).
β<α

Then X = {xα : α < ω1 } is a super-I-Luzin set.



Theorem 10.3 (Kunen see [75]) Suppose B = Borel(2ω )/I is a cBa, C ⊆ B

are the clopen mod I sets, ord(C) = α > 2, and X is super-I-Luzin. Then
ord(X) = α.

Proof:
Note that the ord(X) is the minimum α such that for every B ∈ Borel(2ω )
there exists A ∈ Π0α (2ω ) with A ∩ X = B ∩ X.
= α we know that given any Borel set B there exists a Π0α
Since ord(C) e
set A such that A∆B ∈ I. Since X is Luzin we know that X ∩ (A∆B)e is
countable. Hence there exist countable sets F0 , F1 such that

X ∩ B = X ∩ ((A \ F0 ) ∪ F1 ).

But since α > 2 we have that ((A\F0 )∪F1 ) is also Π0α and hence ord(X) ≤ α.
On the other hand for any β < α we know there e exists a Borel set B such
0
that for every Πβ set A we have B∆A ∈ / I (since ord(C) > β). But since X
is super-I-Luzin we have that for every Π0β set A that X ∩ (B∆A) 6= ∅ and
e
hence X ∩ B 6= X ∩ A. Consequently, ord(X)
e > β.


Corollary 10.4 (CH) For every α ≤ ω1 there exists a separable metric space
X such that ord(X) = α.
Descriptive Set Theory and Forcing 37

While a graduate student at Berkeley I had obtained the result that it

was consistent with any cardinal arithmetic to assume that for every α ≤ ω1
there exists a separable metric space X such that ord(X) = α. It never
occurred to me at the time to ask what CH implied. In fact, my way of
thinking at the time was that proving something from CH is practically the
same as just showing it is consistent. I found out in the real world (outside
of Berkeley) that they are considered very differently.
In Miller [75] it is shown that for every α < ω1 it is consistent there exists
a separable metric space of order β iff α < β ≤ ω1 . But the general question
is open.

Question 10.5 For what C ⊆ ω1 is it consistent that

C = {ord(X) : X separable metric }?

11 MARTIN-SOLOVAY THEOREM 38

11 Martin-Solovay Theorem
In this section we the theorem below. The technique of proof will be used in
the next section to produce a boolean algebra of order ω1 .

Theorem 11.1 (Martin-Solovay [74]) The following are equivalent for an

infinite cardinal κ:

1. MAκ , i.e., for any poset P which is ccc and family D of dense subsets
of P with |D| < κ there exists a P-filter G with G∩D 6= ∅ for all D ∈ D

2. For any ccc σ-ideal I in Borel(2ω ) and I ⊂ I with |I| < κ we have that
[
2ω \ I = 6 ∅.

Lemma 11.2 Let B = Borel(2ω )/I for some ccc σ-ideal I and let P = B \
{0}. The following are equivalent for an infinite cardinal κ:

1. for any family D of dense subsets of P with |D| < κ there exists a
P-filter G with G ∩ D 6= ∅ for all D ∈ D

2. for any family F ⊆ Bω with |F| < κ there exists an ultrafilter U on B

which is F-complete, i.e., for every hbn : n ∈ ωi ∈ F
X
bn ∈ U iff ∃n bn ∈ U
n∈ω

3. for any I ⊂ I with |I| < κ

[
2ω \ I=
6 ∅

Proof:
To see that (1) implies (2) note that for any hbn : n ∈ ωi ∈ Bω the set
X
D = {p ∈ P : p ≤ − bn or ∃n p ≤ bn }
n

is dense. Note also that any filter extends to an ultrafilter.

Descriptive Set Theory and Forcing 39

To see that (2) implies (3) do as follows. Let Hγ stand for the family of
sets whose transitive closure has cardinality less than the regular cardinal γ,
i.e. they are hereditarily of cardinality less than γ. The set Hγ is a natural
model of all the axioms of set theory except possibly the power set axiom,
see Kunen [56]. Let M be an elementary substructure of Hγ for sufficiently
large γ with |M | < κ, I ∈ M , I ⊆ M .
Let F be all the ω-sequences of Borel sets which are in M . Since |F| < κ
we know there exists U an F-complete ultrafilter on B. Define x ∈ 2ω by the
rule:
x(n) = i iff [{y ∈ 2ω : y(n) = i}] ∈ U.

Claim: For every Borel set B ∈ M :

x ∈ B iff [B] ∈ U.

Proof:
This is true for subbasic clopen sets by definition. Inductive steps just
use that U is an M-complete ultrafilter.


To see that (3) implies (1), let M be an elementary substructure of Hγ

for sufficiently large γ with |M | < κ, I ∈ M , D ⊆ M . Let

I = M ∩ I.

By (3) there exists [

x ∈ 2ω \ I.
Let BM = B ∩ M . Then define

G = {[B] ∈ BM : x ∈ B}.

Check G is a P filter which meets every D ∈ D.


This proves Lemma 11.2.

To prove the theorem it necessary to do a two step iteration. Let P be

◦
a poset and Q∈ V P be the P-name of a poset, i.e.,

|`P Q is a poset.
Descriptive Set Theory and Forcing 40

Then we form the poset

◦ ◦ ◦ ◦
P∗ Q= {(p, q ) : p |`q ∈Q}

ordered by (p̂, q̂) ≤ (p, q) iff p̂ ≤ p and p̂ |` q̂ ≤ q. In general there are

◦
two problems with this. First, P∗ Q is a class. Second, it does not satisfy
antisymmetry: x ≤ y and y ≤ x implies x = y. These can be solved by
cutting down to a sufficiently large set of nice names and modding out by
the appropriate equivalence relation. Three of the main theorems are:

Theorem 11.3 If G is P-generic over V and H is QG -generic over V [G],

then ◦
G ∗ H = {(p, q) ∈ P∗ Q: p ∈ G, q G ∈ H}.
◦
is a P∗ Q filter generic over V .
◦
Theorem 11.4 If K is a P∗ Q-filter generic over V , then

G = {p : ∃q (p, q) ∈ K}

is P-generic over V and

H = {q G : ∃p (p, q) ∈ K}

is QG -generic over V [G].

◦
Theorem 11.5 (Solovay-Tennenbaum [104]) If P is ccc and |`P “Q is ccc”,
◦
then P∗ Q is ccc.

For proofs of these results, see Kunen [56] or Jech [44].

Finally we prove Theorem 11.1. (1) implies (2) follows immediately from
Lemma 11.2. To see (2) implies (1) proceed as follows.
Note that κ ≤ c, since (1) fails for FIN(c+ , 2). We may also assume
that the ccc poset P has cardinality less than κ. Use a Lowenheim-Skolem
argument to obtain a set Q ⊆ P with the properties that |Q| < κ, D ∩ Q is
dense in Q for every D ∈ D, and for every p, q ∈ Q if p and q are compatible
(in P) then there exists r ∈ Q with r ≤ p and r ≤ q. Now replace P by Q.
The last condition on Q guarantees that Q has the ccc.
Descriptive Set Theory and Forcing 41

Choose X = {xp : p ∈ P} ⊆ 2ω distinct elements of 2ω . T If G is P-filter

generic over V let Q be Silver’s forcing for forcing a Gδ -set, n∈ω Un , in X
such that \
G = {p ∈ P : xp ∈ Un }.
n∈ω
◦
Let B ∈ V be a countable base for X. A simple description of P∗ Q can be
given by:
◦
(p, q) ∈ P∗ Q
iff p ∈ P and q ∈ V is a finite set of consistent sentences of the form:
◦
/ U n ” where x ∈ X or
1. “x ∈
◦
2. “B ⊆U n ” where B ∈ B and n ∈ ω.
◦
with the additional requirement that whenever the sentence “x ∈
/ U n ” is in
q and x = xr , then p and r are incompatible (so p |` r ∈
/ G).
◦
Note that if D ⊆ P is dense in P, then D is predense in P∗ Q, i.e., every
◦
r ∈ P∗ Q is compatible with an element of D. Consequently, it is enough
◦
to find sufficiently generic filters for P∗ Q. By Lemma 11.2 and Sikorski’s
◦
Theorem 10.1 it is enough to see that if P∗ Q⊆ B is dense in the ccc cBa
algebra B, then B is countably generated. Let

C = {[| B ⊆ Un |] : B ∈ B, n ∈ ω}.

We claim that C generates B. To see this, note that for each p ∈ P

Y
[| xp ∈ ∩n Un |] = [| xp ∈ Un |]
n∈ω
X
[| xp ∈ Un |] = [| B ⊆ Un |]
B∈B,xp ∈B

furthermore
(p, ∅) = [| xp ∈ ∩n Un |]
Descriptive Set Theory and Forcing 42

◦
and so it follows that every element of P∗ Q is in the boolean algebra gen-
◦
erated by C and so since P∗ Q is dense in B it follows that C generates B.


Define X ⊆ 2ω to be a generalized I-Luzin set for an ideal I in the Borel
sets iff |X| = c and |X ∩ A| < c for every A ∈ I. It follows from the
Martin-Solovay Theorem 11.1 that (assuming that the continuum is regular)
MA is equivalent to
for every ccc ideal I in the Borel subsets of 2ω there exists a generalized
I-Luzin set.
Miller and Prikry [84] show that it is necessary to assume the continuum
is regular in the above observation.
12 BOOLEAN ALGEBRA OF ORDER ω1 43

12 Boolean algebra of order ω1

Now we use the Martin-Solovay technique to produce a countably generated
ccc cBa with order ω1 . Before doing so we introduce a countable version of
α-forcing which will be useful for other results also. It is similar to one used
in Miller [76] to give a simple proof about generating sets in the category
algebra.
Let T be a nice tree of rank α (2 ≤ α < ω1 ). Define
Pα = {p : D → ω : D ∈ [ω]<ω , ∀s, sˆn ∈ D p(s) 6= p(sˆn)}.
This is ordered by p ≤ q iff p ⊇ q. For p ∈ Pα define
rank(p) = max{rT (s) : s ∈ domain(p)}
where rT is the rank function on T .
Lemma 12.1 rank : Pα → α+1 satisfies the Rank Lemma 7.4, i.e, for every
p ∈ Pα and β ≥ 1 there exists p̂ ∈ Pα such that
1. p̂ is compatible with p,
2. rank(p̂) ≤ β, and
3. for any q ∈ Pα if rank(q) < β and p̂ and q are compatible, then p and
q are compatible.
Proof:
First let p0 ≤ p be such that for every s ∈ domain(p) and n ∈ ω if
rT (sˆn) < β < λ = rT (s)
then there exists m ∈ ω with p0 (sˆnˆm) = p(s). Note that
rT (sˆn) < β < λ = rT (s)
can happen only when λ is a limit ordinal and for any such s there can be
at most finitely many n (because T is a nice tree).
Now let
E = {s ∈ domain(p0 ) : rT (s) ≤ β}
and define p̂ = p0  E. It is compatible with p since p0 is stronger than both.
From its definition it has rank ≤ β. So let q ∈ Pα have rank(q) < β and be
incompatible with p. We need to show it is incompatible with p̂. There are
only three ways for q and p to be incompatible:
Descriptive Set Theory and Forcing 44

1. ∃s ∈ domain(p) ∩ domain(q) p(s) 6= q(s),

2. ∃s ∈ domain(q) ∃sˆn ∈ domain(p) q(s) = p(sˆ n), or
3. ∃s ∈ domain(p) ∃sˆn ∈ domain(q) p(s) = q(sˆ n).
For (1) since rank(q) < β we know rT (s) < β and hence by construction s is
in the domain of p̂ and so q and p̂ are incompatible. For (2) since
rT (sˆn) < rT (s) < β
we get the same conclusion. For (3) since sˆn ∈ domain(q) we know
rT (sˆn) < β.
If rT (s) = β, then s ∈ domain(p̂) and so q and p̂ are incompatible. Otherwise
since T is a nice tree,
rT (sˆn) < β < rT (s) = λ
a limit ordinal. In this case we have arranged p̂ so that there exists m with
p(s) = p̂(sˆnˆm) and so again q and p̂ are incompatible.

Lemma 12.2 There exists a countable family D of dense subsets of Pα such
that for every G a Pα -filter which meets each dense set in D the filter G
determines a map x : T → ω by p ∈ G iff p ⊆ x. This map has the property
that for every s ∈ T >0 the value of x(s) is the unique element of ω not in
{x(sˆn) : n ∈ ω}.
Proof:
For each s ∈ T the set
Ds = {p : s ∈ domain(p)}
is dense. Also for each s ∈ T >0 and k ∈ ω the set
Esk = {p : p(s) = k or ∃n p(sˆn) = k}
is dense.

The poset Pα is separative, since if p 6≤ q then either p and q are incom-
patible or there exists s ∈ domain(q) \ domain(p) in which case we can find
p̂ ≤ p with p̂(s) 6= q(s).
Descriptive Set Theory and Forcing 45

Now if Pα ⊆ B is dense in the cBa B, it follows that for each p ∈ Pα

p = [| p ⊆ x |]

and for any s ∈ T >0 and k

Y
[| x(s) = k |] = [| x(sˆ m) 6= k |].
m∈ω

Consequently if
C = {p ∈ Pα : domain(p) ⊆ T 0 }
then C ⊆ BP has the property that ord(C) = α + 1.
Now let α<ω1 Pα be the direct sum, i.e., p = hpα : α < ω1 i with pα ∈ Pα
and pα = 1α = ∅ for all but finitely many α. This forcing is equivalent to
adding ω1 Cohen reals, so the usual delta-lemma argument shows that it is
ccc. Let
X = {xα,s,n ∈ 2ω : α < ω1 , s ∈ Tα0 , n ∈ ω}
be distinct elements of 2ω . For G = hGα : α < ω1 i which is α<ω1 Pα -generic
P
over V , use X and Silver forcing to code the rank zero parts of each Gα , i.e.,
P ◦ P ◦
define ( α<ω1 Pα )∗ Q by (p, q) ∈ ( α<ω1 Pα )∗ Q
iff P
p ∈ α<ω1 Pα and q is a finite set of consistent sentences of the form:
◦
1. “x ∈
/ U n ” where x ∈ X or
◦
2. “B ⊆U n ” where B is clopen and n ∈ ω.
◦
with the additional proviso that whenever “xα,s,n ∈ / U n ” ∈ q then s is in the
domain of pα and pα (s) 6= n. This is a little stronger than saying p |` q̌ ∈ Q,
but would be true for a dense set of conditions.
The rank function
X ◦
rank : ( Pα )∗ Q:→ ω1
α<ω1

is defined by

rank(hpα : α < ω1 i, q) = max{rank(pα ) : α < ω1 }

which means we ignore q entirely.

Descriptive Set Theory and Forcing 46

P ◦
Lemma 12.3 For every p ∈ ( α<ω1 Pα )∗ Q and β ≥ 1 there exists p̂ in the
P ◦
poset ( α<ω1 Pα )∗ Q such that
1. p̂ is compatible with p,
2. rank(p̂) ≤ β, and
P ◦
3. for any q ∈ ( α<ω1 Pα )∗ Q if rank(q) < β and p̂ and q are compatible,
then p and q are compatible.
Proof:
Apply Lemma 12.1 to each pα to obtain pˆα and then let
p̂ = (hpˆα : α < ω1 i, q).
This is still a condition because pˆα retains all the rank zero part of pα which
◦
is needed to force q ∈Q.
 ◦
P
Let ( α<ω1 Pα )∗ Q⊆ B be a dense subset of the ccc cBa B. We show
that B is countably generated and ord(B) = ω1 . A strange thing about ω1
is that if one countable set of generators has order ω1 , then all countable
sets of generators have order ω1 . This is because any countable set will be
generated by a countable stage.
One set of generators for B is
◦
C = {[| B̌ ⊆Un |] : B clopen , n ∈ ω}.
Note that
Y YX ◦
[| x ∈ ∩n∈ω Un |] = [| x ∈ Un |] = {[| B̌ ⊆Un |] : x ∈ B}
n∈ω n∈ω

and also each Pα is generated by

{p ∈ Pα : domain(p) ⊆ Tα0 }.
We know that for each α < ω1 , s ∈ Tα0 and n ∈ ω if p = (hpα : α < ω1 i, q) is
the condition for which pα is the function with domain {s}, and pα (s) = n,
and the rest of p is the trivial condition, then
\ ◦
p = [| x̌α,s,n ∈ Un |].
n∈ω
Descriptive Set Theory and Forcing 47

From these facts it follows that C generates B.

It follows from Lemma 8.4 that the order of C is ω1 . For any β < ω1 let
b = (hpα : α < ω1 i, q) be the condition all of whose components are trivial
except for pβ , and pβ any the function with domain hi. Then b ∈ / Σ0β (C).
Otherwise by Lemma 8.4, there would be some a ≤ b with rank(a, C) e < β,
but then paβ would not have hi in its domain.
This proves the ω1 case of Theorem 8.2.
13 LUZIN SETS 48

13 Luzin sets
In this section we use Luzin sets and generalized I-Luzin sets to construct
separable metric spaces of various Borel orders. Before doing so we review
some standard material on the property of Baire.
Given a topological space X the σ-ideal of meager sets is defined as fol-
lows. Y ⊆ X is nowhere dense iff the interior of the closure of Y is empty,
i.e, int(cl(Y )) = ∅. A subset of X is meager iff it is the countable union
of nowhere dense sets. The Baire category Theorem is the statement that
nonempty open subsets of compact Hausdorff spaces or completely metriz-
able spaces are not meager. A subset B of X has the Baire property iff there
exists U open such that B∆U is meager.

Theorem 13.1 (Baire) The family of sets with the Baire property forms a
σ-field.

Proof:
If B∆U is meager where U is open, then

B∆cl(U ) = (B \ cl(U )) ∪ (cl(U ) \ B)

and (B \ cl(U )) ⊆ B \ U is meager and (cl(U ) \ B) ⊆ (U \ B) ∪ (cl(U ) \ U )

is meager because cl(U ) \ U is nowhere dense. Therefore,

∼ B ∆ ∼ cl(U ) = B∆cl(U )

is meager.
If Bn ∆Un is meager for each n, then
[ [ [
( Bn )∆( Un ) ⊆ Bn ∆Un
n∈ω n∈ω n∈ω

is meager.

Hence every Borel set has the property of Baire.

Theorem 13.2 Suppose that every nonempty open subset of X is nonmea-

ger, then B = Borel(X)/meager(X) is a cBa.
Descriptive Set Theory and Forcing 49

Proof:
It is enough to show that it is complete. Suppose Γ ⊆ B is arbitrary. Let
U be a family of open sets such that

Γ = {[U ]meager(X) : U ∈ U}.

S
Let V = U and we claim that [V ] is the minimal upper bound of Γ in B.
Clearly it is an upper bound. Suppose [W ] is any upper bound for Γ with W
open. So U \ W is meager for every U ∈ U. We need to show that V \ W is
meager (so [V ] ≤ [W ]). V \ W ⊆ cl(V ) \ W and if the latter is not nowhere
dense,Sthen there exists P a nonempty open set with P ⊆ cl(V ) \ W . Since
V = U we may assume that there exists U ∈ U with P ⊆ U . But P is
a nonempty open set and [P ] ≤ [W ] so it is impossible for P to be disjoint
from W .

We say that X ⊆ 2ω is a super Luzin set iff for every Borel set B the
set X ∩ B is countable iff B is meager. (This is equivalent to super-I-Luzin
where I is the ideal of meager sets.) It is easy to see that if X is an ordinary
Luzin set, then in some basic clopen set C it is a super Luzin set relative to
C. Also since ω ω can be obtained by deleting countably many points from
2ω it is clear that having a Luzin set for one is equivalent to having it for the
other. With a little more work it can be seen that it is equivalent to having
one for any completely metrizable separable metric space without isolated
points.
The generic set of Cohen reals in the Cohen real model is a Luzin set.
Let FIN(κ, 2) be the partial order of finite partial functions from κ into
2. If G is FIN(κ, 2)-generic over V and for each α < κ we define xα by
xα (n) = G(ω ∗ α + n), then X = {xα : α < κ} is a Luzin set in V [G].

Theorem 13.3 (Miller [75]) If there exists a Luzin set in ω ω , then for every
α with 3 ≤ α < ω1 there exists Y ⊆ ω ω with ord(Y ) = α.

Proof:
Let Tα be the nice α-tree used in the definition of

Pα = {p : p : D → ω, D ∈ [Tα ]<ω , ∀s, sˆn ∈ D p(s) 6= p(sˆn)}.

Let Qα be the closed subspace of ω Tα

Qα = {x ∈ ω Tα : ∀s, sˆn ∈ Tα x(s) 6= x(sˆn).

Descriptive Set Theory and Forcing 50

For each p ∈ Pα we get a basic clopen set

[p] = {x ∈ Qα : p ⊆ x}.
It easy to check that Qα is homeomorphic to ω ω . Hence there exists a super
0
Luzin set X ⊆ Qα . Consider the map r : Qα → ω Tα defined by restriction,
i.e., r(x) = x  Tα0 . Note that by Lemma 12.2 there exists a countable
sequence
T T subsets of Qα , hDn : n ∈ ωi, such that r is one-to-one
of dense open
on n∈ω Dn . Since n∈ω Dn is a comeager set in Qα and X is Luzin we may
assume that \
X⊆ Dn .
n∈ω

Y is just the image of X under r. (So, in fact, Y is the one-to-one

continuous image of a Luzin set.) An equivalent way to view Y is just to
imagine X with the topology given by
B = {[p] : p ∈ Pα , domain(p) ⊆ Tα0 }.
We know by Lemma 8.4 that
ord{[B] : B ∈ B} = α + 1
as a subset of Borel(Qα )/meager(Qα ) which means that:
α+1 is minimal such that for every B ∈ Borel(Qα ) there exists a Σ0α+1 (B)
set A such that such that B∆A is meager in Qα . e
This translates (since X is super-Luzin) to:
α+1 is minimal such that for every B ∈ Borel(Qα ) there exists a Σ0α+1 (B)
set A such that such that (B∆A) ∩ X is countable. e
Which means for Y that:
α + 1 is minimal such that for every B ∈ Borel(Y ) there exists a Σ0α+1 (Y )
set A such that such that B∆A is countable. e
But since countable subsets of Y are Σ02 and α > 2, this means ord(Y ) =
α + 1. e
To get Y of order λ for a limit λ < ω1 just take a clopen separated union
of sets whose order increases to λ.

Now we clean up a loose end from Miller [75]. In that paper we had shown
that assuming MA for every α < ω1 there exists a separable metric space X
with α ≤ ord(X) ≤ α + 2 or something silly like that. Shortly afterwards,
Fremlin supplied the missing arguments to show the following.
Descriptive Set Theory and Forcing 51

Theorem 13.4 (Fremlin [26]) MA implies that for every α with 2 ≤ α ≤ ω1

there exists a second countable Hausdorff space X with ord(X) = α.

Proof:
Since the union of less than continuum many meager sets is meager, the
Mahlo construction 10.2 gives us a set X ⊆ Qα of cardinality c such that
for every Borel set B ∈ Borel(Qα ) we have that B is meager iff B ∩ X has
cardinality less than c.
Letting B be defined as in the proof of Theorem 13.3 we see that:
α+1 is minimal such that for every B ∈ Borel(Qα ) there exists a Σ0α+1 (B)
set A such that such that (B∆A) ∩ X has cardinality less than c. e
What we need to see to complete the proof is that:
for every Z ⊆ X of cardinality less than c there exists a Σ02 (B) set F such
that F ∩ X = Z. e

Lemma 13.5 (MA) For any Z ⊆ Qα of cardinality less than c, there exists
hDn : n ∈ ωi such that:
1. Dn is predense in Pα ,
2. p ∈ Dn implies domain(p) ⊆ Tα0 , and
T S
3. Z ∩ n∈ω s∈Dn [s] = ∅.

Proof:
Force with the following poset
P = {(F, hpn : n < N i) : F ∈ [Z]<ω , N < ω, domain(p) ∈ [Tα0 ]<ω }
where (F, hpn : n < N i) ≤ (H, hqn : n < M i) iff F ⊇ H, N ≥ M , pn = qn
for n < M , and for each x ∈ H and M ≤ n < N we have x ∈ / [pn ]. Since
this forcing is ccc we can apply MA with the appropriate choice of family of
dense sets to get Dn = {pm : m > n} to do the job.

By applying the Lemma we get that for every Z ⊆ X of cardinality less
than c there exists a Σ02 (B) set F which is meager in Qα and such that
Z ⊆ F ∩ X. But since F e is meager we know F ∩ X has cardinality less than
c. By Theorem 5.1 every subset of r(F ∩ X) is a relative Σ02 in Y , so there
exists an F0 a Σ02 (B) set such that Z = (F ∩ X) ∩ F0 . Thise proves Theorem
13.4. e

14 COHEN REAL MODEL 52

14 Cohen real model

I have long wondered if there exists an uncountable separable metric space
of order 2 in the Cohen real model. I thought there weren’t any. We already
know from Theorem 13.3 that since there is an uncountable Luzin set in
Cohen real model that for every α with 3 ≤ α ≤ ω1 there is an uncountable
separable metric space X with ord(X) = α.

Theorem 14.1 Suppose G is FIN(κ, 2)-generic over V where κ ≥ ω1 . Then

in V [G] there is a separable metric space X of cardinality ω1 with ord(X) = 2.

Proof:
We may assume that κ = ω1 . This is because FIN(κ, 2) × FIN(ω1 , 2) is
isomorphic to FIN(κ, 2) and so by the product lemma we may replace V by
V [H] where (H, G) is FIN(κ, 2) × FIN(ω1 , 2)-generic over V .
We are going to use the fact that forcing with FIN(ω1 , 2) is equivalent
to any finite support ω1 iteration of countable posets. The main idea of the
proof is to construct an Aronszajn tree of perfect sets, a technique first used
by Todorcevic (see Galvin and Miller [30]). We construct an Aronszajn
tree (A, ) and a family of perfect sets ([Ts ] : s ∈ A) such that ⊇ is the
same order as . We will then show that if X = {xs : s ∈ A} is such that
xs ∈ [Ts ], then the order of X is 2.
In order to insure the construction can keep going at limit ordinals we
will need to use a fusion argument. Recall that a perfect set corresponds to
the infinite branches [T ] of a perfect tree T ⊆ 2<ω , i.e., a tree with the
property that for every s ∈ T there exist a t ∈ T such that both tˆ0 ∈ T
and tˆ1 ∈ T . Such a T is called a splitting node of T . There is a natural
correspondence of the splitting nodes of a perfect tree T and 2<ω .
Given two perfect trees T and T 0 and n ∈ ω define T ≤n T 0 iff T ⊆ T 0
and the first 2<n splitting nodes of T remain in T 0 .

Lemma 14.2 (Fusion) Suppose (Tn : n ∈ ω) is aTsequence of perfect sets

such that Tn+1 ≤n Tn for every n ∈ ω. Then T = n∈ω Tn is a perfect tree
and T ≤n Tn for every n ∈ ω.

Proof:
Descriptive Set Theory and Forcing 53

If T = n<ω Tn , then T is a perfect tree because the first 2<n splitting

T
nodes of Tn are in Tm for every m > n and thus in T .
 P
By identifying FIN(ω1 , 2) with α<ω1 FIN(ω, 2) we may assume that

G = hGα : α < ω1 i

where Gβ is FIN(ω, 2)-generic over V [Gα : α < β] for each β < ω1 .

Given an Aronszajn tree A we let Aα be the nodes of A at level α, i.e.

Aα = {s ∈ A : {t ∈ A : t
s} has order type α}

and [
A<α = Aβ .
β<α

We use hGα : α < ω1 i to construct an Aronszajn tree (A, ) and a family of

perfect sets ([Ts ] : s ∈ A) such that

1. s  t implies Ts ⊇ Tt ,

2. if s and t are distinct elements of Aα , then [Ts ] and [Tt ] are disjoint,

3. every s ∈ Aα has infinitely many distinct extensions in Aα+1 ,

4. for each s ∈ A<α and n < ω there exists t ∈ Aα such that Tt ≤n Ts ,

5. for each s ∈ Aα and t ∈ Aα+1 with s
t, we have that [Tt ] is a generic

perfect subset of [Ts ] obtained by using Gα (explained below in Case
2), and

6. {Ts : s ∈ A<α } ∈ V [Gβ : β < α].

The first three items simply say that {[Ts ] : s ∈ A} and its ordering by
⊆ determines (A, ), so what we really have here is an Aronszajn tree of
perfect sets. Item (4) is there in order to allow the construction to proceed
at limits levels.
Item (5) is what we do a successor levels and guarantees the set we are
building has order 2. Item (6) is a consequence of the construction and would
be true for a closed unbounded set of ordinals no matter what we did anyway.
Here are the details of our construction.
Descriptive Set Theory and Forcing 54

Case 1. α a limit ordinal.

The construction is done uniformly enough so that we already have that

{Ts : s ∈ A<α } ∈ V [Gβ : β < α].

Working in V [Gβ : β < α] choose a sequence αn for n ∈ ω which strictly

increases to α. Given any sn ∈ Aαn we can choose by inductive hypothesis a
sequence sm ∈ Aαm for m ≥ n such that

Tsm+1 ≤m Tsm .
T
If T = m>n Tsm , then by Lemma 14.2 we have that T ≤n Tsn . Now let
{Tt : t ∈ Aα } be a countable collection of perfect trees so that for every n
and s ∈ Aαn there exists t ∈ Aα with Tt ≤n Ts . This implies item (4) because
for any s ∈ A<α and n < ω there exists some m ≥ n with s ∈ A<αm hence by
inductive hypothesis there exists ŝ ∈ Aαm with Tŝ ≤n Ts and by construction
there exists t ∈ Aα with Tt ≤m Tŝ and so Tt ≤n Ts as desired.

Case 2. Successor stages.

Suppose we already have constructed

{Ts : s ∈ A<α+1 } ∈ V [Gβ : β < α + 1].

Given a perfect tree T ⊆ 2<ω define the countable partial order P(T ) as
follows. p ∈ P(T ) iff p is a finite subtree of T and p ≤ q iff p ⊇ q and p is an
end extension of q, i.e., every new node of p extends a terminal node of q. It
is easy to see that if G is P(T )-generic over a model M , then
[
TG = {p : p ∈ G}

is a perfect subtree of T . Furthermore, for any D ⊆ [T ] dense open in [T ]

and coded in M , [TG ] ⊆ D. i.e., the branches of TG are Cohen reals (relative
to T ) over M . This means that for any Borel set B ⊆ [T ] coded in M , there
exists an clopen set C ∈ M such that

C ∩ [TG ] = B ∩ [TG ].

To see why this is true let p ∈ P(T ) and B Borel. Since B has the Baire
property relative to [T ] by extending each terminal node of p, if necessary,
we can obtain q ≥ p such that for every terminal node s of q either [s] ∩ B is
Descriptive Set Theory and Forcing 55

meager in [T ] or [s] ∩ B is comeager in [T ] ∩ [s]. If we let C be union of all

[s] for s a terminal node of q such that [s] ∩ B is comeager in [T ] ∩ [s], then

q |` B ∩ TG = C ∩ TG .

To get TG ≤n T we could instead force with

P(T, n) = {p ∈ P(T ) : p end extends the first 2<n splitting nodes of T }.

Finally to determine Aα+1 consider

X
{P(Ts , m) : s ∈ Aα , m ∈ ω}.

This poset is countable and hence Gα+1 determines a sequence

hTs,m : s ∈ Aα , m ∈ ωi

of generic perfect trees such that Ts,m ≤m Ts . Note that genericity also
guarantees that corresponding perfect sets will be disjoint. We define Aα+1
to be this set of generic trees.
This ends the construction.
By taking generic perfect sets at successor steps we have guaranteed the
following. For any Borel set B coded in V [Gβ : β < α+1] and Tt for t ∈ Aα+1
there exists a clopen set Ct such that

Ct ∩ [Tt ] = B ∩ [Tt ].

Suppose X = {xs : s ∈ A} is such that xs ∈ [Ts ] for every s ∈ A. Then X

has order 2. To verify this, let B ⊆ 2ω be any Borel set. By ccc there exists
a countable α such that B is coded in V [Gβ : β < α + 1]. Hence,
[ [
B∩ [Tt ] = (Ct ∩ [Tt ]).
t∈Aα+1 t∈Aα+1

Hence B ∩ X is equal to a Σ02 set intersected X:

e
[
X∩ (Ct ∩ [Tt ])
t∈Aα+1
Descriptive Set Theory and Forcing 56

union a countable set: [

(B ∩ X) \ [Tt ]
t∈Aα+1

and therefore B ∩ X is Σ02 in X.

 e
Another way to get a space of order 2 is to use the following argument. If
the ground model satisfies CH, then there exists a Sierpiński set. Such a set
has order 2 (see Theorem 15.1) in V and therefore by the next theorem it has
order 2 in V [G]. It also follows from the next theorem that if X = 2ω ∩ V ,
then X has order ω1 in V [G]. Consequently, in what I think of as “the Cohen
real model”, i.e. the model obtained by adding ω2 Cohen reals to a model of
CH, there are separable metric spaces of cardinality ω1 and order α for every
α with 2 ≤ α ≤ ω1 .

Theorem 14.3 Suppose G is FIN(κ, 2)-generic over V and

V |= ord(X) = α

Then
V [G] |= ord(X) = α

By the usual ccc arguments it is clearly enough to prove the Theorem for
FIN(ω, 2). To prove it we will need the following lemma.

Lemma 14.4 (Kunen, see [57]) Suppose p ∈ FIN(ω, 2), X is a second count-
◦
able Hausdorff space in V , and B is a name such that
◦
p |`B ⊆ X̌ is a Π0α -set.
e
Then the set ◦
{x ∈ X : p |` x̌ ∈B }
is a Π0α -set in X.
e
Proof:
This is proved by induction on α.
For α = 1 let B ∈ V be a countable base for the closed subsets of X,
◦
i.e., every closed set is the intersection of elements of B. Suppose p |`“B is
Descriptive Set Theory and Forcing 57

◦
a closed set in X̌”. Then for every x ∈ X p |`“x̌ ∈B ” iff for every q ≤ p and
◦
for every C ∈ B if q |`“B ⊆ Č”, then x ∈ C. But
◦
{x ∈ X : ∀q ≤ p ∀B ∈ B (q |` “ B ⊆ Č” → x ∈ C)}

is closed in X. ◦
Now suppose α > 1 and p |`B ∈ Π0α (X). Let βn be a sequence which
is either constantly α − 1 if α is a successor
e or which is unbounded in α
if α is a limit. By the usual forcing facts there exists a sequence of names
hBn : n ∈ ωi such that for each n,

p |` Bn ∈ Π0βn ,
e
and \
p |` B = ∼ Bn .
n<ω

Then for every x ∈ X

p |` x̌ ∈ B
iff
∀n ∈ ω p |` x ∈∼ Bn
iff
∀n ∈ ω ∀q ≤ p q 6|` x ∈ Bn .
Consequently,
◦ \\ ◦
{x ∈ X : p |` x ∈B } = ∼ {x : q |` x̌ ∈Bm }.
n∈ω q≤p


Now let us prove the Theorem. Suppose V |=“ord(X) = α”. Then in
V [G] for any Borel set B ∈ Borel(X)
[ ◦
B= {x ∈ X : p |` x̌ ∈B }.
p∈G

◦
By the lemma, each of the sets {x ∈ X : p |` x̌ ∈B } is a Borel set in V , and
since ord(X) = α, it is a Σ0α set. Hence, it follows that B is a Σ0α set. So,
e e
Descriptive Set Theory and Forcing 58

V [G] |= ord(X) ≤ α. To see that ord(X) ≥ α let β < α and suppose in V

the set A ⊆ X is Σ0β but not Π0β . This must remain true in V [G] otherwise
there exists a p ∈ eG such thate

p |` “Ǎ is Π0β ”
e
but by the lemma
{x ∈ X : p |` x̌ ∈ Ǎ} = A
is Π0β which is a contradiction.
e
Part of this argument is similar to one used by Judah and Shelah [46]
who showed that it is consistent to have a Q-set which does not have strong
measure zero.
It is natural to ask if there are spaces of order 2 of higher cardinality.

Theorem 14.5 Suppose G is FIN(κ, 2)-generic over V where V is a model

of CH and κ ≥ ω2 . Then in V [G] for every separable metric space X with
|X| > ω1 , we have ord(X) ≥ 3.

Proof:
This will follow easily from the next lemma.

Lemma 14.6 (Miller [81]) Suppose G is FIN(κ, 2)-generic over V where V

is a model of CH and κ ≥ ω2 . Then V [G] models that for every X ⊆ 2ω
with |X| = ω2 there exists a Luzin set Y ⊆ 2ω and a one-to-one continuous
function f : Y → X.

Proof:
Let hτα : α < ω2 i be a sequence of names for distinct elements of X. For
each α and n choose a maximal antichain Aαn ∪ Bnα such that

p |` τα (n) = 0 for each p ∈ Aαn and

p |` τα (n) = 1 for each p ∈ Bnα .

Let Xα ⊆ κ be union of domains of elements from n∈ω Aαn ∪ Bnα . Since each
S
Xα is countable we may as well assume that the Xα ’s form a ∆-system, i.e.
there exists R such that Xα ∩ Xβ = R for every α 6= β. We can assume that
R is the empty set. The reason is we can just replace Aαn by

Âαn = {p  (∼ R) : p ∈ Aαn and p  R ∈ G}

Descriptive Set Theory and Forcing 59

and similarly for Bnα . Then let V [G  R] be the new ground model.
Let
hjα : Xα → ω : α < ω2 i
be a sequence of bijections in the ground model. Each jα extends to an
order preserving map from FIN(Xα , 2) to FIN(ω, 2). By CH, we may as well
assume that there exists a single sequence, h(An , Bn ) : n ∈ ωi, such that
every jα maps hAαn , Bnα : n ∈ ωi to h(An , Bn ) : n ∈ ωi.
The Luzin set is Y = {yα : α < ω2 } where yα (n) = G(jα−1 (n)). The
continuous function, f , is the map determined by h(An , Bn ) : n ∈ ωi:

f (x)(n) = 0 iff ∃m x  m ∈ An .

This proves the Lemma.


If f : Y → X is one-to-one continuous function from a Luzin set Y ,
then ord(X) ≥ 3. To see this assume that Y is dense and let D ⊆ Y be
a countable dense subset of Y . Then D is not Gδ in Y . This is because
any Gδ set containing D is comeager and therefore must meet Y in an un-
countable set. But note that f (D) is a countable set which cannot be Gδ in
X, because f −1 (f (D)) would be Gδ in Y and since f is one-to-one we have
D = f −1 (f (D)). This proves the Theorem.

It is natural to ask about the cardinalities of sets of various orders in
this model. But note that there is a trivial way to get a large set of order
β. Take a clopen separated union of a large Luzin set (which has order 3)
and a set of size ω1 with order β. One possible way to strengthen the notion
of order is to say that a space X of cardinality κ has essential order β iff
every nonempty open subset of X has order β and cardinality κ. But this is
also open to a simple trick of combining a small set of order β with a large
set of small order. For example, let X ⊆ 2ω be a clopen separated union of
a Luzin set of cardinality κ and set of cardinality ω1 of order β ≥ 3. Let
hPn : n ∈ ωi be a sequence of disjoint nowhere dense perfect subsets of 2ω
with the property that for every s ∈ 2<ω there exists n with PS n ⊆ [s]. Let
Xn ⊆ Pn be a homeomorphic copy of X for each n < ω. Then n∈ω Xn is a
set of cardinality κ which has essential order β.
With this cheat in mind let us define a stronger notion of order. A
separable metric space X has hereditary order β iff every uncountable
Y ⊆ X has order β. We begin with a stronger version of Theorem 13.3.
Descriptive Set Theory and Forcing 60

Theorem 14.7 If there exists a Luzin set X of cardinality κ, then for every
α with 2 < α < ω1 there exists a separable metric space Y of cardinality κ
which is hereditarily of order α.

Proof:
This is a slight modification of the proof of Theorem 13.3. Let Qα be the
following partial order. Let hαn : n ∈ ωi be a sequence such that if α is a
limit ordinal, then αn is a cofinal increasing sequence in α and if α = β + 1
then αn = β for every n.
The rest of the proof is same except we use Qα+1 instead of Pα for suc-
cessors and Qα for limit α instead of taking a clopen separated union. By
using the direct sum (even in the successor case) we get a stronger property
for the order. Let Y
Q̂α = Qαn
be the closed subspace of Y
ω T αn
n∈ω

and let B be the collection of clopen subsets of Qα whichQare given by rank

zero conditions of Q(α), i.e., all rectangles of the form n∈ω [pn ] such that
pn ∈ Qαn with domain(p) ⊆ Tα0 and pn the trivial condition for all but finitely
many n.
As in the proof of Theorem 13.3 we get that the order of {[B] : B ∈ B}
as a subset of Borel(Q̂α )/meager(Q̂α ) is α. Because we took the direct sum
we get the stronger property that for any nonempty clopen set C in Q̂α the
order of {[B ∩ C] : B ∈ B} is α.
But know given X a Luzin set in Q̂α we know that for any uncountable
Y ⊆ X there is a nonempty clopen set C ⊆ Q̂α such that Y ∩ C is a super-
Luzin set relative to C. (The accumulation points of Y , the set of all
points all of whose neighborhoods contain uncountably many points of Y , is
closed and uncountable, therefore must have nonempty interior.) If C is a
nonempty clopen set in the interior of the accumulation points of Y , then
since {[B ∩ C] : B ∈ B} is α, we have by the proof of Theorem 13.3, that the
order of Y is α.


Theorem 14.8 Suppose that in V there is a separable metric space, X, with

hereditary order β for some β ≤ ω1 . Let G be FIN(κ, 2)-generic over V for
any κ ≥ ω. Then in V [G] the space X has hereditary order β.
Descriptive Set Theory and Forcing 61

Proof:
In V [G] let Y ⊆ X be uncountable. For contradiction, suppose that
◦ ◦
p |` ord(Y ) ≤ α and | Y | = ω1

for some p ∈ FIN(κ, 2) and α < β. Working in V by the usual ∆-system

argument we can get q ≤ p and

hpx : x ∈ Ai

for some A ∈ [X]ω1 such that and px ≤ q and

◦
px |` x̌ ∈Y

for each x ∈ A and

dom(px ) ∩ dom(py ) = dom(q)

for distinct x and y in A. Since A is uncountable we know that in V the

order of A is ω1 . Consequently, there exists R ⊆ A which is Σ0α (A) but not
Π0α (A). We claim that in V [G] the set R ∩ Y is not Π0α (Y ).e If not, there
e ◦ e
exists r ≤ q and S such that
◦ ◦ ◦ ◦
r |` “ Y ∩R =Y ∩ S and S ∈ Π0α (A)”.
e
Since Borel sets are coded by reals there exists Γ ∈ [κ]ω ∩ V such that for
◦
any x ∈ A the statement “x̌ ∈S ” is decided by conditions in FIN(Γ, 2) and
also let Γ be large enough to contain the domain of r.
Define ◦
T = {x ∈ A : q |` x̌ ∈S }.
According to Lemma 14.4 the set T is Π0α (A). Consequently, (assuming
α ≥ 3) there are uncountably many x ∈ Ae with x ∈ R∆T . Choose such an
x which also has the property that dom(px ) \ dom(q) is disjoint from Γ. This
can be done since the px form a ∆ system. But now, if x ∈ T \ R, then
◦ ◦ ◦
r ∪ px |` “x̌ ∈Y ∩ S and x ∈
/ Y ∩Ř”.

On the other hand, if x ∈ R \ T , then there exists r̂ ≤ r in FIN(Γ, 2) such

that ◦
r̂ |` x̌ ∈
/S
Descriptive Set Theory and Forcing 62

and consequently,
◦ ◦ ◦
r̂ ∪ px |` “x̌ ∈
/ Y ∩ S and x ∈Y ∩Ř”.

Either way we get a contradiction and the result is proved.



Theorem 14.9 (CH) There exists X ⊆ 2ω such that X has hereditary order
ω1 .

Proof:
By Theorem 8.2 there exists a countably generated ccc cBa B which has
order ω1 . For any b ∈ B with b 6= 0 let ord(b) be the order of the boolean
algebra you get by looking only at {c ∈ B : c ≤ b}. Note that in fact B has
the property that for every b ∈ B we have ord(b) = ω1 . Alternatively, it easy
to show that any ccc cBa of order ω1 would have to contain an element b
such that every c ≤ b has order ω1 .
By the proof of the Sikorski-Loomis Theorem 9.1 we know that B is
isomorphic to Borel(Q)/meager(Q) where Q is a ccc compact Hausdorff space
with a basis of cardinality continuum.
Since Q has ccc, every open dense set contains an open dense set which
is a countable union of basic open sets. Consequently, by using CH, there
exists a family F of meager sets with |F| = ω1 such that every meager set
is a subset of one in F. Also note that for any nonmeager Borel set B in Q
there exists a basic open set C and F ∈ F with C \F ⊆ B. Hence by Mahlo’s
construction (Theorem 10.2) there exists a set X ⊆ Q with the property that
for any Borel subset B of Q

|B ∩ X| ≤ ω iff B meager.

Let B be a countable field of clopen subsets of Q such that

{[B]meager(Q) : B ∈ B}

generates B. Let
R = {X ∩ B : B ∈ B}.
If X̃ ⊆ 2ω is the image of X under the characteristic function of the sequence
B (see Theorem 4.1), then X̃ has hereditary order ω1 . Of course X̃ is really
just the same as X but retopologized using B as a family of basic open sets.
Descriptive Set Theory and Forcing 63

Let Y ∈ [X]ω1 . Since ord(p) = ω1 for any basic clopen set the following claim
shows that the order of Y (or rather the image of Y under the characteristic
function of the sequence B) is ω1 .

Claim: There exists a basic clopen p in Q such that for every Borel B ⊆ p,

|B ∩ Y | ≤ ω iff B meager.

Proof:
Let p and q stand for nonempty basic clopen sets. Obviously if B is
meager then B ∩ Y is countable, since B ∩ X is countable. To prove the
other direction, suppose for contradiction that for every p there exists q ≤ p
and Borel Bq ⊆ q such that Bq is comeager in q and Bq ∩ Y is countable.
By using ccc there exists a countable dense family Σ and Bq for q ∈ Σ with
Bq ⊆ q Borel and comeager in q such that Bq ∩ Y is countable. But
[
B = {Bq : q ∈ Σ}

is a comeager Borel set which meets Y in a countable set. This implies that
Y is countable since X is contained in B except for countable many points.


Theorem 14.10 Suppose G is FIN(κ, 2)-generic over V where V is a model

of CH and κ ≥ ω. Then in V [G] there exists a separable metric space X with
|X| = ω1 and hereditarily of order ω1 .

Proof:
Immediate from Theorem 14.8 and 14.9.

Finally, we show that there are no large spaces of hereditary order ω1 in
the Cohen real model.

Theorem 14.11 Suppose G is FIN(κ, 2)-generic over V where V is a model

of CH and κ ≥ ω2 . Then in V [G] for every separable metric space X with
|X| = ω2 there exists Y ∈ [X]ω2 with ord(Y ) < ω1 .

Proof:
By the argument used in the proof of Lemma 14.6 we can find

hGα : α < ω2 i ∈ V [G]

Descriptive Set Theory and Forcing 64

P
which is α<ω2 FIN(ω, 2)-generic over V and a FIN(ω, 2)-name τ for an
element of 2ω such that Y = {τ Gα : α < ω2 } is subset of X. We claim that
ord(Y ) < ω1 . Let
F = {[| τ ∈ C |] : C ⊆ 2ω clopen }
where boolean values are in the unique complete boolean algebra B in which
FIN(ω, 2) is dense. Let F be the complete subalgebra of B which is generated
by F. First note that the order of F in F is less than ω1 . This is because F
contains a countable dense set:
Y
D = { {c ∈ F : p ≤ c} : p ∈ FIN(ω, 2)}.
Since D is countable and Σ01 (D) = F, it follows that the order of F is
countable. e
I claim that the order of Y is essentially less than or equal to the order
of F in F.
Lemma 14.12 Let B be a cBa, τ a B-name for an element of 2ω , and
F = {[| τ ∈ C |] : C ⊆ 2ω clopen }.
Then for each B ⊆ 2ω a Π0α set coded in V the boolean value [| τ ∈ B̌ |] is
Π0α (F) and conversely, fore every c ∈ Π0α (F) there exists a B ⊆ 2ω a Π0α set
coded
e in V such that c = [| τ ∈ B̌ |]. e e

Proof:
Both directions of the lemma are simple inductions.

Now suppose the order of F in F is α. Let B ⊆ 2ω be any Borel set coded
in V [G]. By ccc there exists H = G  Σ where Σ ⊆ κ is countable set in V
such that B is coded in V [H]. Consequently, since we could replace V with
V [H] and delete countably many elements of Y we may as well assume that
B is coded in the ground model. Since the order of F is α we have by the
lemma that there exists a Π0α set A such that
e
[| τ ∈ Ǎ |] = [| τ ∈ B̌ |].
It follows that
Y ∩A=Y ∩B
and hence order of Y is less than or equal to α (or three since we neglected
countably many elements of Y ).

15 THE RANDOM REAL MODEL 65

15 The random real model

In this section we consider the question of Borel orders in the random real
model. We conclude with a few remarks about perfect set forcing.
A set X ⊆ 2ω is a Sierpiński set iff X is uncountable and for every
measure zero set Z we have X ∩ Z countable. Note that by Mahlo’s Theorem
10.2 we know that under CH Sierpiński sets exists. Also it is easy to see that
in the random real model, the set of reals given by the generic filter is a
Sierpiński set.

Theorem 15.1 (Poprougenko [91]) If X is Sierpiński, then ord(X) = 2.

Proof:
For any Borel set B ⊆ 2ω there exists an Fσ set with F ⊆ B and B \ F
measure zero. Since X is Sierpiński (B \ F ) ∩ X = F0 is countable, hence
Fσ . So
B ∩ X = (F ∪ F0 ) ∩ X.


I had been rather hoping that every uncountable separable metric space
in the random real model has order either 2 or ω1 . The following result shows
that this is definitely not the case.

Theorem 15.2 Suppose X ∈ V is a subspace of 2ω of order α and G is

measure algebra 2κ -generic over V , i.e. adjoin κ many random reals.
Then V [G] |= α ≤ ord(X) ≤ α + 1.

The result will easily follow from the next two lemmas.
Presumably, ord(X) = α in V [G], but I haven’t been able to prove this.
Fremlin’s proof (Theorem 13.4) having filled up one such missing gap, leaving
this gap here restores a certain cosmic balance of ignorance.5
Clearly, by the usual ccc arguments, we may assume that κ = ω and
G is just a random real. In the following lemmas boolean values [| θ |] will
be computed in the measure algebra B on 2ω . Let µ be the usual product
measure on 2ω .
5
All things I thought I knew; but now confess, the more I know I know, I know the
less.- John Owen (1560-1622)
Descriptive Set Theory and Forcing 66

◦
Lemma 15.3 Suppose  a real, b ∈ B, and U the name of a Π0α subset of 2ω
in V [G]. Then the set e
◦
{x ∈ 2ω : µ(b ∧[| x̌ ∈U |]) ≥ }

is Π0α in V .
e
Proof:
The proof is by induction on α.

◦
Case α = 1.
If U is a name for a closed set, then
◦ Y ◦
[| x̌ ∈U |] = [| [x  n]∩ U 6= ∅ |].
n∈ω

Consequently,
◦
µ(b ∧[| x̌ ∈U |]) ≥ 
iff ◦
∀n ∈ ω µ(b ∧[| [x  n]∩ U 6= ∅ |]) ≥ 
and the set is closed.

Case α > 1.
◦ ◦ ◦
Suppose U = n∈ω ∼ Un where each Un is a name for a Π0αn set for some
T
αn < α. We can assume that the sequence ∼ Un is descending.
e Consequently,

µ(b ∧[| x̌ ∈ Ǔ |]) ≥ 

iff \ ◦
µ(b ∧[| x̌ ∈ ∼ Un |]) ≥ 
n∈ω

iff ◦
∀n ∈ ω µ(b ∧[| x̌ ∈∼ Un |]) ≥ 
iff ◦
∀n ∈ ω not µ(b ∧[| x̌ ∈ Un |]) > µ(b) − .
iff ◦
∀n ∈ ω not ∃m ∈ ω µ(b ∧[| x̌ ∈ Un |]) ≥ µ(b) −  + 1/m
Descriptive Set Theory and Forcing 67

By induction, each of the sets

◦
{x ∈ 2ω : µ(b ∧[| x̌ ∈ Un |]) ≥ µ(b) −  + 1/m}

is Π0αn and so the result is proved.

e
It follows from this lemma that if X ⊆ 2ω and V |=“ord(X) > α”, then
V [G] |=“ord(X) > α”. For suppose F ⊆ 2ω is Σ0α such that for every H ⊆ 2ω
which is Π0α we have F ∩ X 6= H ∩ X. Suppose e for contradiction that
e
◦ ◦
b |` “ U ∩X̌ = F̌ ∩ X̌ and U is Π0α ”.
e
But then
{x ∈ 2ω : µ(b ∧[| x̌ ∈ Ǔ |]) = µ(b)}
is a Π0α set which must be equal to F on X, which is a contradiction.
To
e prove the other direction of the inequality we will use the following
lemma.

Lemma 15.4 Let G be B-generic (where B is the measure algebra on 2ω )

and r ∈ 2ω is the associated random real. Then for any b ∈ B
3
b ∈ G iff ∀∞ n µ([r  n] ∧ b) ≥ µ([r  n]).
4
Proof:
Since G is an ultrafilter it is enough to show that b ∈ G implies
3
∀∞ n µ([r  n] ∧ b) ≥ µ([r  n]).
4
Let B+ be the nonzero elements of B. To prove this it suffices to show:

Claim: For any b ∈ B+ and for every d ≤ b in B+ there exists a tree T ⊆ 2<ω
with [T ] of positive measure, [T ] ≤ d, and
3
µ([s] ∩ b) ≥ µ([s])
4
for all but finitely many s ∈ T .
Proof:
Descriptive Set Theory and Forcing 68

Without loss we may assume that d is a closed set and let Td be a tree
such that d = [Td ]. Let t0 ∈ Td be such that
9
µ([t0 ] ∩ [Td ]) ≥ µ([t0 ]).
10
Define a subtree T ⊆ Td by r ∈ T iff r ⊆ t0 or t0 ⊆ r and
3
∀t (t0 ⊆ t ⊆ r implies µ([t] ∩ b) ≥ µ([t]) ).
4
So we only need to see that [T ] has positive measure. So suppose for contra-
diction that µ([T ]) = 0. Then for some sufficiently large N
[ 1
µ( [s]) ≤ µ([t0 ]).
10
s∈T ∩2N

For every s ∈ Td ∩ 2N with t0 ⊆ s, if s ∈

/ T then there exists t with t0 ⊆ t ⊆ s
and µ([t] ∩ b) < 43 µ([t]). Let Σ be a maximal antichain of t like this. But
note that [ [
[t0 ] ∩ [Td ] ⊆ [s] ∪ ([t] ∩ b).
s∈2N ∩T t∈Σ

By choice of Σ
[ 3
µ( [s] ∩ b) ≤ µ([t0 ])
s∈Σ
4
and by choice of N
[ 1
µ( [s]) ≤ µ([t0 ])
10
s∈2N ∩T

which contradicts the choice of t0 :

1 3 17 9
µ([t0 ] ∩ [Td ]) ≤ ( + )µ([t0 ]) = 6 µ([t0 ]) < µ([t0 ]).
10 4 20 10
This proves the claim and the lemma.

In effect, what we have done in Lemma 15.4 is reprove the Lebesgue
density theorem, see Oxtoby [90].
6
Trust me on this, I have been teaching a lot of Math 99 “College Fractions”.
Descriptive Set Theory and Forcing 69

So now suppose that the order of X in V is ≤ α. We show that it is ≤ α+1

◦
in V [G]. Let U be any name for a Borel subset of X in the extension. Then
◦
we know that x ∈ U G iff [| x̌ ∈U |] ∈ G. By Lemma 15.3 we know that for
any s ∈ 2<ω the set
◦ 3
Bs = {x ∈ X : µ([s] ∩ [| x̌ ∈U |]) ≥ µ([s])}
4
is a Borel subset of X in the ground model and hence is Π0α (X). By Lemma
15.4 we have that for any x ∈ X e

x ∈ U iff ∀∞ n x ∈ Brn

and so U is Σ0α+1 (X) in V[G].

This concludes
e the proof of Theorem 15.2.

Note that this result does allow us to get sets of order λ for any countable
limit ordinal λ by taking a clopen separated union of a sequence of sets whose
order goes up λ.
Also a Luzin set X from the ground model has order 3 in the random
real extension. Since (ord(X) = 3)V we know that (3 ≤ ord(X) ≤ 4)V [G] .
To see that (ord(X) ≤ 3)V [G] suppose that B ⊆ X is Borel in V [G]. The
above proof shows that there exists Borel sets Bn each coded in V (but the
sequence may not be in V ) such that

x ∈ B iff ∀∞ n x ∈ Bn .

For each Bn there exists an open set Un ⊆ X such that Bn ∆Un is countable.
If we let [ \
C= Um
n∈ω m>n

then C is Σ03 (X)and B∆C is countable. Since subtracting and adding a

countable set from a Σ03 (X) is still Σ03 (X) we have that B is Σ03 (X) and so
e
the order of X is ≤ 3 ein V [G]. e e

Theorem 15.5 Suppose V models CH and G is measure algebra on 2κ -

generic over V for some κ ≥ ω2 . Then in V [G] for every X ⊆ 2ω of
cardinality ω2 there exists Y ∈ [X]ω2 with ord(Y ) = 2.
Descriptive Set Theory and Forcing 70

Proof:
Using the same argument as in the proof of Theorem 14.11 we can get a
Sierpiński set S ⊆ 2ω of cardinality ω2 and a term τ for any element of 2ω
such that Y = {τ r : r ∈ S} is a set of distinct elements of X. This Sierpiński
set has two additional properties: every element of it is random over the
ground model and it meets every set of positive measure, i.e. it is a super
Sierpiński set.
We will show that the order of Y is 2.

Lemma 15.6 Let F ⊆ B be any subset of a measure algebra B closed under

finite conjunctions. Then Π02 (F) = Σ02 (F), i.e. F has order ≤ 2.
e e
Proof:
Let µ be the measure on B.
(1) For any b ∈ Π01 (F) and real  > 0 there exists a ∈ F with b ≤ a and
µ(a − b) < . Q e
pf:7 b = n∈ω an . Let a = n<N an for some sufficiently large N .
Q

(2) For any b ∈ Σ02 (F) and real  > 0 there exists a ∈ Σ01 (F) with b ≤ a and
µ(a − b) < P
. e e
pf: b = n<ω bn . Applying (1) we get an ∈ F with bn ≤ an and

µ(an − bn ) < .
2n
P
Then let a = n∈ω an .

Now suppose b ∈ Σ02 (F). Then by applying (2) there Qexists an ∈ Σ01 (F)
with b ≤ an and µ(an e− b) < 1/n. Consequently, if a = n∈ω an , theneb ≤ a
and µ(a − b) = 0 and so a = b.

Let
F = {[| τ ∈ C |] : C ⊆ 2ω clopen }
where boolean values are in the measure algebra B on 2ω . Let F be the
complete subalgebra of F which is generated by F.
Since the order of F is 2, by Lemma 14.12 we have that for any Borel
set B ⊆ Y there exists a Σ02 (Y ) set F such that y ∈ B iff y ∈ F for all but
e
7
Pronounced ‘puff’.
Descriptive Set Theory and Forcing 71

countably many y ∈ Y . Thus we see that the order of Y is ≤ 3. To get it

down to 2 we use the following lemma. If B = (F \ F0 ) ∪ F1 where F0 and
F1 are countable and F is Σ02 , then by the lemma F0 would be Π02 and thus
B would be Σ02 . e e
e
Lemma 15.7 Every countable subset of Y is Π02 (Y ).
e
Proof:
It suffices to show that every countable subset of Y can be covered by a
countable Π02 (Y ) since one can always subtract a countable set from a Π02 (Y )
and remaine Π02 (Y ). e
For any se∈ 2<ω define
bs = [| s ⊆ τ |].
Working in the ground model let Bs be a Borel set with [Bs ]B = bs . Since
the Sierpiński set consists only of reals random over the ground model we
know that for every r ∈ S

r ∈ Bs iff s ⊆ τ r .

Also since the Sierpiński set meets

T every Borel set of positive measure we
know that for any z ∈ Y the set n<ω Bzn has measure zero. Now let Z =
{zn : n < ω} ⊆ Y be arbitrary but listed with infinitely many repetitions.
For each n choose m so that if sn = zn  m, then µ(Bsn ) < 1/2n . Now for
every r ∈ S we have that
\ [ \ [
r∈ Bsm iff τ r ∈ [sm ].
n<ω m>n n<ω m>n

0
T S
The set H = n<ω m>n [sm ]Tcovers
S Z and is Π 2 . It has countable intersec-
tion with Y because the set n<ω m>n Bsm has measure zero.
e
This proves the Lemma and Theorem 15.5.


Perfect Set Forcing

Descriptive Set Theory and Forcing 72

In the iterated Sack’s real model the continuum is ω2 and every set
X ⊆ 2ω of cardinality ω2 can be mapped continuously onto 2ω (Miller [81]).
It follows from Reclaw’s Theorem 3.5 that in this model every separable
metric space of cardinality ω2 has order ω1 . On the other hand this forcing
(and any other with the Sack’s property) has the property that every meager
set in the extension is covered by a meager set in the ground model and every
measure set in the extension is covered by a measure zero set in the ground
model (see Miller [78]). Consequently, in this model there are Sierpiński sets
and Luzin sets of cardinality ω1 . Therefore in the iterated Sacks real model
there are separable metric spaces of cardinality ω1 of every order α with
2 ≤ α < ω1 . I do not know if there is an uncountable separable metric space
which is hereditarily of order ω1 in this model.
Another way to obtain the same orders is to use the construction of
Theorem 22 of Miller [75]. What was done there implies the following:

For any model V there exists a ccc extension V [G] in which every
uncountable separable metric space has order ω1 .

If we apply this result ω1 times with a finite support extension,

we get a model, V [Gα : α < ω1 ], where there are separable metric spaces
of all orders of cardinality ω1 , but every separable metric space of cardinality
ω2 has order ω1 .
To see the first fact note that ω1 length finite support iteration always
adds a Luzin set. Consequently, by Theorem 14.7, for each α with 2 < α < ω1
there exists a separable metric space of cardinality ω1 which is hereditarily of
order α. Also there is such an X of order 2 by the argument used in Theorem
14.1.
On the other hand if X has cardinality ω2 in V [Gα : α < ω1 ], then for
some β < ω1 there exists and uncountable Y ∈ V [Gα : α < β] with Y ⊆ X.
Hence Y will have order ω1 in V [Gα : α < β + 1] and by examining the proof
it is easily seen that it remains of order ω1 in V [Gα : α < ω1 ].
16 COVERING NUMBER OF AN IDEAL 73

16 Covering number of an ideal

This section is a small diversion.8 It is motivated by Theorem 11.1 of Martin
and Solovay.
Define for any ideal I in Borel(2ω )
[
cov(I) = min{|I| : I ⊆ I, I = 2ω }.

The following theorem is well-known.

Theorem 16.1 For any cardinal κ the following are equivalent:

1. MAκ (ctbl), i.e. for any countable poset, P, and family D of dense
subsets of P with |D| < κ there exists a P-filter G with G ∩ D 6= ∅ for
every D ∈ D, and

2. cov(meager(2ω )) ≥ κ.

Proof:
MAκ (ctbl) implies cov(meager(2ω )) ≥ κ, is easy because if U ⊆ 2ω is a
dense open set, then
D = {s ∈ 2<ω : [s] ⊆ U }
is dense in 2<ω .
cov(meager(2ω )) ≥ κ implies MAκ (ctbl) follows from the fact that any
countable poset, P, either contains a dense copy of 2<ω or contains a p such
that every two extensions of p are compatible.


Theorem 16.2 (Miller [79]) cof(cov(meager(2ω ))) > ω, e.g., it is impossi-

ble to have cov(meager(2ω )) = ℵω .

Proof:
Suppose for contradiction that κ = cov(meager(2ω )) has countable cofi-
nality and let κn for n ∈ ω be a cofinal sequence in κ. Let hCα : α < κi be
a family of closed nowhere dense sets which cover 2ω . We will construct a
sequence Pn ⊆ 2ω of perfect sets with the properties that

1. Pn+1 ⊆ Pn ,
8
All men’s gains are the fruit of venturing. Herodotus BC 484-425.
Descriptive Set Theory and Forcing 74

S
2. Pn ∩ {Cα : α < κn } = ∅, and

3. ∀α < κ Cα ∩ Pn is nowhere dense in Pn .

T
This easily gives a contradiction, since n<ω Pn is nonempty and disjoint
from all Cα , contradicting the fact that the Cα ’s cover 2ω .
We show how to obtain P0 , since the argument easily relativizes to show
how to obtain Pn+1 given Pn . Since cov(meager(2ω )) > κn there exists a
countable sequence
D = {xn : n ∈ ω} ⊆ 2ω
such that D is dense and for every n
[
xn ∈
/ Cα .
α<κn

Consider the following forcing notion P.

P = {(H, n) : n ∈ ω and H ∈ [D]<ω }

This is ordered by (H, n) ≤ (K, m) iff

1. H ⊇ K,

2. n ≥ m, and

3. for every x ∈ H there exists y ∈ K with x  m = y  m.

Note that P is countable.

For each n ∈ ω define En ⊆ P by (H, m) ∈ En iff

1. m > n and

2. ∀x ∈ H∃y ∈ H x  n = y  n but x  m 6= y  m.

and for each α < κ0 let

Fα = {(H, m) ∈ P : ∀x ∈ H [x  m] ∩ Cα = ∅}.

For G a P-filter, define X ⊆ D by

[
X = {H : ∃n (H, n) ∈ G}
Descriptive Set Theory and Forcing 75

and let P = cl(X). It easy to check that the En ’s are dense and if G meets
each one of them, then P is perfect (i.e. has no isolated points). The Fα for
α < κ0 are dense in P. This is because D ∩ Cα = ∅ so given (H, n) ∈ P there
exists m ≥ n such that for every x ∈ H we have [x  m] ∩ Cα = ∅ and thus
(H, m) ∈ Fα . Note that if G ∩ Fα 6= ∅, then P ∩ Cα = ∅. Consequently, by
Theorem 16.1, there exists a P-filter G such that G meets each En and all
Fα for α < κ0 . Hence P = cl(X) is a perfect set which is disjoint from each
Cα for α < κ0 . Note also that for every α < κ we have that Cα ∩ D is finite
and hence Cα ∩ X is finite and therefore Cα ∩ P is nowhere dense in P . This
ends the construction of P = P0 and since the Pn can be obtained with a
similar argument, this proves the Theorem.


Question 16.3 (Fremlin) Is the same true for the measure zero ideal in
place of the ideal of meager sets?

Some partial results are known (see Bartoszynski, Judah, Shelah [7][8][9]).

Theorem 16.4 (Miller [79]) It is consistent that cov(meager(2ω1 )) = ℵω .

Proof:
In fact, this holds in the model obtained by forcing with FIN(ℵω , 2) over
a model of GCH.

cov(meager(2ω1 )) ≥ ℵω : Suppose for contradiction that

{Cα : α < ωn } ∈ V [G]

is a family of closed nowhere dense sets covering 2ω1 . Define

Eα = {s ∈ FIN(ω1 , 2) : [s] ∩ Cα = ∅}.

Using ccc, there exists Σ ∈ [ℵω ]ωn in V with

{Eα : α < ωn } ∈ V [G  Σ].

Let X ⊆ ℵω be a set in V of cardinality ω1 which is disjoint from Σ. By the

product lemma G  X is FIN(X, 2)-generic over V [G  Σ]. Consequently, if
H : ω1 → 2 corresponds to G via an isomorphism of X and ω1 , then H ∈ / Cα
for every α < ωn .
Descriptive Set Theory and Forcing 76

cov(meager(2ω1 )) ≤ ℵω : Note that for every uncountable X ⊆ ω1 such

that X ∈ V [G] there exists

n ∈ ω and Z ∈ [ω1 ]ω1 ∩ V [G  ωn ]

with Z ⊆ X. To see this, note that for every α ∈ X there exists p ∈ G such
that p |` α ∈ X and p ∈ FIN(ωn , 2) for some n ∈ ω. Consequently, by ccc,
some n works for uncountably many α.
Consider the family of all closed nowhere dense sets C ⊆ 2ω1 which are
coded in some V [G  ωn ] for some n. We claim that these cover 2ω1 . This
follows from above, because for any Z ⊆ ω1 which is infinite the set

C = {x ∈ 2ω1 : ∀α ∈ Z x(α) = 1}

is nowhere dense.


Theorem 16.5 (Miller [79]) It is consistent that there exists a ccc σ-ideal
I in Borel(2ω ) such that cov(I) = ℵω .

Proof: ◦ ◦
Let P = FIN(ω1 , 2)∗ Q where Q is a name for the Silver forcing which
codesQup generic filter for FIN(ω1 , 2) just like in the proof of Theorem 11.1.
Let α<ℵω P be the direct sum (i.e. finite support product) of ℵω copies of
P. Forcing with the direct sum adds a filter G = hGα : α < ℵω i where each
Gα is P-generic. In general, a direct sum is ccc iff every finite subproduct is
ccc. This follows by a delta-system argument. Every finite product of P has
ccc, because P is σ-centered, i.e., it is the countable union
Q of centered sets.
Let V be a model of GCH and G = hGα : α < ℵω i be α<ℵω P generic over
V . We claimQ that in V [G] if I is the σ-ideal given byωSikorski’s Theorem 9.1
such that α<ℵω P is densely embedded into Borel(2 )/I then cov(I) = ℵω .
First define, mP , to be the cardinality of the minimal failure of MA for
P, i.e., the least κ such that there exists a family |D| = κ of dense subsets of
P such that there is no P-filter meeting all the D ∈ D.

Lemma 16.6 In V [hGα : α < ℵω i] we have that mP = ℵω .

Proof:
Note that for any set D ⊂ P there exists a set Σ ∈ [ℵω ]ω1 in V with
D ∈ V [hGα : α ∈ Σi]. So if |D| = ωn then there exists Σ ∈ [ℵω ]ωn in V with
Descriptive Set Theory and Forcing 77

D ∈ V [hGα : α ∈ Σi]. Letting α ∈ ℵω \ Σ we get Gα a P-filter meeting every

D ∈ D. Hence mP ≥ ℵω .
On the other hand:

Claim: For every X ∈ [ω1 ]ω1 ∩ V [hGα : α < ℵω i] there exists n ∈ ω and
Y ∈ [ω1 ]ω1 ∩ V [hGα : α < ℵn i] with Y ⊆ X.
Proof: ◦
For every α ∈ X there exist p ∈ G and n < ω such that p |` α̌ ∈X
and domain(p) ⊆ ℵn . Since X is uncountable there is one n which works for
uncountably many α ∈ X.

It follows from the Claim that there is no H which is FIN(ω1 , 2) generic
over all the models V [hGα : α < ℵn i], but forcing with P would add such an
H and so mP ≤ ℵω and the Lemma is proved.


Lemma 16.7 If P is ccc and dense in the cBa Borel(2ω )/I, then mP =
cov(I).

Proof:
This is the same as Lemma 11.2 equivalence of (1) and (3), except you
have to check that m is the same for both P and Borel(2ω )/I.

Kunen [58] showed that least cardinal for which MA fails can be a singular
cardinal of cofinality ω1 , although it is impossible for it to have cofinality ω
(see Fremlin [27]). It is still open whether it can be a singular cardinal
of cofinality greater than ω1 (see Landver [61]). Landver [62] generalizes
Theorem 16.2 to the space 2κ with basic clopen sets of the form [s] for
s ∈ 2<κ . He uses a generalization of a characterization of cov(meager(2ω ))
due to Bartoszynski [6] and Miller [80].
 78

Part II
Analytic sets
17 Analytic sets
Analytic sets were discovered by Souslin when he encountered a mistake of
Lebesgue. Lebesgue had erroneously proved that the Borel sets were closed
under projection. I think the mistake he made was to think that the count-
able intersection commuted with projection. A good reference is the volume
devoted to analytic sets edited by Rogers [93]. For the more classical view-
point of operation-A, see Kuratowski [59]. For the whole area of descriptive
set theory and its history, see Moschovakis [89].

Definition. A set A ⊆ ω ω is Σ11 iff there exists a recursive

[
R⊆ (ω n × ω n )
n∈ω

such that for all x ∈ ω ω

x ∈ A iff ∃y ∈ ω ω ∀n ∈ ω R(x  n, y  n).

A similar definition applies for A ⊆ ω and also for A ⊆ ω × ω ω and so forth.

For example, A ⊆ ω is Σ11 iff there exists a recursive R ⊆ ω × ω <ω such that
for all m ∈ ω
m ∈ A iff ∃y ∈ ω ω ∀n ∈ ω R(m, y  n).
A set C ⊆ ω ω × ω ω is Π01 iff there exists a recursive predicate
[
R⊆ (ω n × ω n )
n∈ω

such that
C = {(x, y) : ∀n R(x  n, y  n)}.
That means basically that C is a recursive closed set.
The Π classes are the complements of the Σ’s and ∆ is the class of sets
which are both Π and Σ. The relativized classes, e.g. Σ11 (x) are obtained by
allowing R to be recursive in x, i.e., R ≤T x. The boldface classes, e.g., Σ11 ,
Π11 , are obtained by taking arbitrary R’s. e
e
17 ANALYTIC SETS 79

Lemma 17.1 A ⊆ ω ω is Σ11 iff there exists set C ⊆ ω ω × ω ω which is Π01

and
A = {x ∈ ω ω : ∃y ∈ ω ω (x, y) ∈ C}.

Lemma 17.2 The following are all true:

1. For every s ∈ ω <ω the basic clopen set [s] = {x ∈ ω ω : s ⊆ x} is Σ11 ,

2. if A ⊆ ω ω × ω ω is Σ11 , then so is

B = {x ∈ ω ω : ∃y ∈ ω (x, y) ∈ A},

3. if A ⊆ ω × ω ω is Σ11 , then so is

B = {x ∈ ω ω : ∃n ∈ ω (n, x) ∈ A},

4. if A ⊆ ω × ω ω is Σ11 , then so is

B = {x ∈ ω ω : ∀n ∈ ω (n, x) ∈ A},

5. if hAn : n ∈ ωi is sequence of Σ11 sets given by the recursive predicates

Rn and hRn : n ∈ ωi is (uniformly) recursive, then both
[ \
An and An are Σ11 .
n∈ω n∈ω

6. if the graph of f : ω ω → ω ω is Σ11 and A ⊆ ω ω is Σ11 , then f −1 (A) is

Σ11 .

Of course, the above lemma is true with ω or ω × ω ω , etc., in place of ω ω .

It also relativizes to any class Σ11 (x). It follows from the Lemma that every
Borel subset of ω ω is Σ11 and that the continuous pre-image of Σ11 set is Σ11 .
e e e
1 ω ω
Theorem 17.3 There exists a Σ1 set U ⊆ ω × ω which is universal for
all Σ11 sets, i.e., for every Σ11 set A ⊆ ω ω there exists x ∈ ω ω with
e e
A = {y : (x, y) ∈ U }.
17 ANALYTIC SETS 80

Proof:
There exists C ⊆ ω ω × ω ω × ω ω a Π01 set which is universal for Π01 subsets
of ω ω × ω ω . Let U be the projection of C on its second coordinate. e

Similarly we can get Σ11 sets contained in ω × ω (or ω × ω ω ) which are
universal for Σ11 subsets of ω (or ω ω ).
The usual diagonal argument shows that there are Σ11 subsets of ω ω which
are not Π11 and Σ11 subsets of ω which are not Π11 .
e
Theorem 17.4 (Normal form) A set A ⊆ ω ω is Σ11 iff there exists a recursive
map
<ω
ω ω → 2ω x 7→ Tx
such that Tx ⊆ ω <ω is a tree for every x ∈ ω ω , and x ∈ A iff Tx is ill-founded.
By recursive map we mean that there is a Turing machine {e} such that for
x ∈ ω ω the machine e computing with an oracle for x, {e}x computes the
characteristic function of Tx .

Proof:
Suppose
x ∈ A iff ∃y ∈ ω ω ∀n ∈ ω R(x  n, y  n).
Define
Tx = {s ∈ ω <ω : ∀i ≤ |s| R(x  i, s  i)}.


A similar thing is true for A ⊆ ω, i.e., A is Σ11 iff there is a uniformly
recursive list of recursive trees hTn : n < ωi such that n ∈ A iff Tn is ill-
founded.
The connection between Σ11 and well-founded trees, gives us the following:

Theorem 17.5 (Mostowski’s Absoluteness) Suppose M ⊆ N are two tran-

sitive models of ZFC∗ and θ is Σ11 sentence with parameters in M . Then
e
M |= θ iff N |= θ.

Proof:
17 ANALYTIC SETS 81

ZFC∗ is a nice enough finite fragment of ZFC to know that trees are well-
founded iff they have rank functions (Theorem 7.1). θ is Σ11 sentence with
parameters in M means there exists R in M such that e

θ = ∃x ∈ ω ω ∀n R(x  n).

This means that for some tree T ⊆ ω <ω in M θ is equivalent to “T has an

infinite branch”. So if M |= θ then N |= θ since a branch T exists in M . On
the other hand if M |= ¬θ, then

M |= ∃r : T → OR a rank function”

and then for this same r ∈ M

N |= r : T → OR is a rank function”

and so N |= ¬θ.

18 CONSTRUCTIBLE WELL-ORDERINGS 82

18 Constructible well-orderings
Gödel proved the axiom of choice relatively consistent with ZF by producing a
definable well-order of the constructible universe. He announced in Gödel [32]
that if V=L, then there exists an uncountable Π11 set without perfect subsets.
Kuratowski wrote down a proof of the theorem below but the manuscript was
lost during World War II (see Addison [2]).
A set is Σ12 iff it is the projection of a Π11 set.

Theorem 18.1 [V=L] There exists a ∆12 well-ordering of ω ω .

Proof:
S Recall the definition of Gödel’s Constructible sets L. L0 = ∅, Lλ =
α<λ Lα for λ a limit ordinal, and Lα+1 is theSdefinable subsets of Lα . De-
finable means with parameters from Lα . L = α∈OR Lα .
The set x is constructed before y, (x <c y) iff the least α such that x ∈ Lα
is less than the least β such that y ∈ Lβ , or α = β and the “least” defining
formula for x is less than the one for y. Here “least” basically boils down
to lexicographical order. Whatever the exact formulation of x <c y is it
satisfies:
x <c y iff Lα |= x <c y
where x, y ∈ Lα and Lα |=ZFC∗ where ZFC∗ is a sufficiently large finite
fragment of ZFC. (Actually, it is probably enough for α to be a limit ordinal.)
Assuming V = L, for x, y ∈ ω ω we have that x <c y iff there exists E ⊆ ω ×ω
◦ ◦
and x, y ∈ ω such that letting M = (ω, E) then

1. E is extensional and well-founded,

2. M |=ZFC∗ + V=L
◦ ◦
3. M |=x<c y ,
◦ ◦ ◦
4. for all n, m ∈ ω (x(n) = m iff M |=x (n) =m), and
◦ ◦ ◦
5. for all n, m ∈ ω (y(n) = m iff M |=y (n) =m).

The first clause guarantees (by the Mostowski collapsing lemma) that M is
isomorphic to a transitive set. The second, that this transitive set will be
Descriptive Set Theory and Forcing 83

of the form Lα . The last two clauses guarantee that the image under the
◦ ◦
collapse of x is x and y is y.
Well-foundedness of E is Π11 . The remaining clauses are all Π0n for some
n ∈ ω. Hence, we have given a Σ12 definition of <c . But a total ordering <
which is Σ1n is ∆1n , since x 6< y iff y = x or y < x. It follows that <c is also
Π12 and hence ∆12 .

19 HEREDITARILY COUNTABLE SETS 84

19 Hereditarily countable sets

HC is the set consisting of all hereditarily countable sets. There is a
close connection between the projective hierarchy above level 2 and a natural
hierarchy on the subsets of HC. A formula of set theory is ∆0 iff it is in
the smallest family of formulas containing the atomic formulas of the form
“x ∈ y” or “x = y”, and closed under conjunction, θ ∧ φ, negation, ¬θ, and
bounded quantification, ∀x ∈ y or ∃x ∈ y. A formula θ of set theory is Σ1
iff it of the form ∃u1 , . . . , un ψ where ψ is ∆0 .

Theorem 19.1 A set A ⊆ ω ω is Σ12 iff there exists a Σ1 formula θ(·) of set
theory such that
A = {x ∈ ω ω : HC |= θ(x)}.

Proof:
We note that ∆0 formulas are absolute between transitive sets, i.e., if
ψ(· · ·) is ∆0 formula, M a transitive set and y a finite sequence of elements
of M , then M |= ψ(y) iff V |= ψ(y). Suppose that θ(·) is a Σ1 formula of
set theory. Then for every x ∈ ω ω we have that HC |= θ(x) iff there exists a
countable transitive set M ∈ HC with x ∈ M such that M |= θ(x). Hence,
◦
HC |= θ(x) iff there exists E ⊆ ω ×ω and x∈ ω such that letting M = (ω, E)
then

1. E is extensional and well-founded,

2. M |=ZFC∗ , (or just that ω exists)

◦
3. M |= θ(x),
◦ ◦ ◦
4. for all n, m ∈ ω x(n) = m iff M |=x (n) =m.

Therefore, {x ∈ ω ω : HC |= θ(x)} is a Σ12 set. On the other hand given a Σ12

set A there exists a Π11 formula θ(x, y) such that A = {x : ∃yθ(x, y)}. But
then by Mostowski absoluteness (Theorem 17.5) we have that x ∈ A iff there
exists a countable transitive set M with x ∈ M and there exists y ∈ M such
that M |=ZFC∗ and M |= θ(x, y). But this is a Σ1 formula for HC.

The theorem says that Σ12 = ΣHC 1 HC
1 . Similarly, Σn+1 = Σn . Let us
illustrate this with an example construction.
Descriptive Set Theory and Forcing 85

Theorem 19.2 If V=L, then there exists a ∆12 Luzin set X ⊆ ω ω .

Proof:
Let {Tα : α < ω1 } (ordered by <c ) be all subtrees T of ω <ω whose
branches [T ] are a closed nowhere dense subset of ω ω . Define xα to be the
least constructed (<c ) element of ω ω which is not in
[
[Tβ ] ∪ {xβ : β < α}.
β<α

Define X = {xα : α < ω1 }. So X is a Luzin set.

To see that X is ΣHC1 note that x ∈ X iff there exists a transitive count-
able M which models ZFC∗ +V=L such that M |=“x ∈ X”(i.e. M models
the first paragraph of this proof).
To see that X is ΠHC 1 note that x ∈ X iff for all M if M is a transitive
countable model of ZFC∗ +V=L with x ∈ M and M |=“∃y ∈ X x <c y”,
then M |=“x ∈ X”. This is true because the nature of the construction is
such that if you put a real into X which is constructed after x, then x will
never get put into X after this. So x will be in X iff it is already in X.

20 SHOENFIELD ABSOLUTENESS 86

20 Shoenfield Absoluteness
κn × ω n define
S
For a tree T ⊆ n<ω

p[T ] = {y ∈ ω ω : ∃x ∈ κω ∀n(x  n, y  n) ∈ T }.

A set defined this way is called κ-Souslin. Thus Σ11 sets are precisely the
ω-Souslin sets. Note that if A ⊆ ω ω × ω ω and A =ep[T ] then the projection
of A, {y : ∃x ∈ ω ω (x, y) ∈ A} is κ-Souslin. To see this let <, >: κ × ω → κ
be a pairing function. For s ∈ κn let s0 ∈ κn and s1 ∈ ω n be defined by
s(i) =< s0 (i), s1 (i) >. Let T ∗ be the tree defined by
[
T∗ = {(s, t) ∈ κn × ω n : (s0 , s1 , t) ∈ T }.
n∈ω

Then p[T ∗ ] = {y : ∃x ∈ ω ω (x, y) ∈ A}.

Theorem 20.1 (Shoenfield [98]) If A is a Σ12 set, then A is ω1 -Souslin set

coded in L, i.e. A = p[T ] where T ∈ L.

Proof:
From the construction of T ∗ it is clear that is enough to see this for A
which is Π11 .
We know that a countable tree is well-founded iff there exists a rank
function r : T → ω1 . Suppose

x ∈ A iff ∀y∃n (x  n, y  n) ∈
/T

where T is a recursive tree. So defining Tx = {t : (x  |t|, t) ∈ T } we have

that x ∈ A iff Tx is well-founded (Theorem 17.4).
The ω1 tree T̂ is just the tree of partial rank functions. Let {sn : n ∈ ω}
be a recursive listing of ω <ω with |sn | ≤ n. Then for every N < ω, and
(r, s) ∈ ω1N × ω N we have (r, t) ∈ T̂ iff

∀n, m < N [(t, sn ), (t, sm ) ∈ T and sn ⊂ sm ] implies r(n) > r(m).

Then A = p[T̂ ]. To see this, note that if x ∈ A, then Tx is well-founded and

so it has a rank function and therefore there exists r with (x, r) ∈ [T̂ ] and
so x ∈ p[T̂ ]. On the other hand if (x, r) ∈ [T̂ ], then r determines a rank
function on Tx and so Tx is well-founded and hence x ∈ A.

Descriptive Set Theory and Forcing 87

Theorem 20.2 (Shoenfield Absoluteness [98]) If M ⊆ N are transitive mod-

els of ZFC∗ and ω1 N ⊆ M , then for any Σ12 (x) sentence θ with parameter
x∈M
M |= θ iff N |= θ.

Proof:
If M |= θ, then N |= θ, because Π11 sentences are absolute. On the other
hand suppose N |= θ. Working in Neusing the proof of Theorem 20.1 we get
a tree T ⊆ ω1<ω with T ∈ L[x] such that T is ill-founded, i.e., there exists
r ∈ [T ]. Note that r codes a witness to a Π11 (x) predicate and a rank function
showing the tree corresponding to this predicate is well-founded. Since for
some α < ω1 , r ∈ αω we see that

Tα = T ∩ α<ω

is ill-founded. But Tα ∈ M (since by assumption (ω1 )N ⊆ M ) and so by the

absoluteness of well-founded trees, M thinks that Tα is ill-founded. But a
branch thru [T ] gives a witness and a rank function showing that θ is true,
and consequently, M |= θ.

21 MANSFIELD-SOLOVAY THEOREM 88

21 Mansfield-Solovay Theorem
Theorem 21.1 (Mansfield [72], Solovay [103]) If A ⊆ ω ω is a Σ12 set with
constructible parameter which contains a nonconstructible element
e of ω ω ,
then A contains a perfect set which is coded in L.

Proof:
BySShoenfield’s Theorem 20.1, we may assume A = p[T ] where T ∈ L and
T ⊆ n<ω ω1n × ω n . Working in L define the following decreasing sequence
of subtrees as follows.
T0 = T
T,
Tλ = β<λ Tβ , if λ a limit ordinal, and
Tα+1 = {(r, s) ∈ Tα : ∃(r0 , s0 ), (r1 , s1 ) ∈ Tα such that (r0 , s0 ), (r1 , s1 )
extend (r, s), and s0 and s1 are incompatible}.
Each Tα is tree, and for α < β we have Tβ ⊆ Tα . Thus there exists some
α0 such that Tα0 +1 = Tα0 .

Claim: [Tα0 ] is nonempty.

Proof:
Let (x, y) ∈ [T ] be any pair with y not constructible. Since A = p[T ] and
A is not a subset of L, such a pair must exist. Prove by induction on α that
(x, y) ∈ [Tα ]. This is easy for α a limit ordinal. So suppose (x, y) ∈ [Tα ] but
(x, y) ∈
/ [Tα+1 ]. By the definition it must be that there exists n < ω such
that (x  n, y  n) = (r, s) ∈
/ Tα+1 . But in L we can define the tree:

Tα(r,s) = {(r̂, ŝ) ∈ Tα : (r̂, ŝ) ⊆ (r, s) or (r, s) ⊆ (r̂, ŝ)}

(r,s)
which has the property that p[Tα ] = {y}. But by absoluteness of well-
(r,s)
founded trees, it must be that there exists (u, y0 ) ∈ [Tα ] with (u, y0 ) ∈ L.
But then y0 = y ∈ L which is a contradiction. This proves the claim.

Since Tα0 +1 = Tα0 , it follows that for every (r, s) ∈ Tα0 there exist

(r0 , s0 ), (r1 , s1 ) ∈ Tα0

such that (r0 , s0 ), (r1 , s1 ) extend (r, s) and s0 and s1 are incompatible. This
allows us to build by induction (working in L):

h(rσ , sσ ) : σ ∈ 2<ω i
Descriptive Set Theory and Forcing 89

with (rσ , sσ ) ∈ Tα0 and for each σ ∈ 2<ω (rσ0 , sσ0 ), (rσ1 , sσ1 ) extend (rσ , sσ )
and sσ0 and sσ1 are incompatible. For any q ∈ 2ω define
[ [
xq = rqn and yq = sqn .
n<ω n<ω

Then we have that (xq , yq ) ∈ [Tα0 ] and therefore P = {yq : q ∈ 2ω } is a

perfect set such that
P ⊆ p[Tα0 ] ⊆ p[T ] = A
and P is coded in L.

This proof is due to Mansfield. Solovay’s proof used forcing. Thus we
have departed9 from our theme of giving forcing proofs.

9
“Consistency is the hobgoblin of little minds. With consistency a great soul has simply
nothing to do.” Ralph Waldo Emerson.
22 UNIFORMITY AND SCALES 90

22 Uniformity and Scales

Given R ⊆ X × Y we say that S ⊆ X × Y uniformizes R iff

1. S ⊆ R,

2. for all x ∈ X if there exists y ∈ Y such that R(x, y), then there exists
y ∈ Y such that S(x, y), and

3. for all x ∈ X and y, z ∈ Y if S(x, y) and S(x, z), then y = z.

Another way to say the same thing is that S is a subset of R which is the
graph of a function whose domain is the same as R’s.
The Π11 sets have the uniformization property.

Theorem 22.1 (Kondo [49]) Every Π11 set R can be uniformized by a Π11
set S.

Here, X and Y can be taken to be either ω or ω ω or even a singleton {0}.

In this last case, this amounts to saying for any nonempty Π11 set A ⊆ ω ω
there exists a Π11 set B ⊆ A such that B is a singleton, i.e., |B| = 1. The
proof of this Theorem is to use a property which has become known as the
scale property.

Lemma 22.2 (scale property) For any Π11 set A there exists hφi : i < ωi
such that

1. each φi : A → OR,

2. for all i and x, y ∈ A if φi+1 (x) ≤ φi+1 (y), then φi (x) ≤ φi (y),

3. for every x, y ∈ A if ∀i φi (x) = φi (y), then x = y,

4. for all hxn : n < ωi ∈ Aω and hαi : i < ωi ∈ ORω if for every i and for
all but finitely many n φi (xn ) = αi , then there exists x ∈ A such that
limn→∞ xn = x and for each i φi (x) ≤ αi ,

5. there exists P a Π11 set such that for all x, y ∈ A and i

P (i, x, y) iff φi (x) ≤ φi (y)

and for all x ∈ A, y ∈

/ A, i ∈ ω P (i, x, y), and
Descriptive Set Theory and Forcing 91

6. there exists S a Σ11 set such that for all x, y ∈ A and i

S(i, x, y) iff φi (x) ≤ φi (y)

and for all y ∈ A, x, i ∈ ω if S(i, x, y), then x ∈ A.

Another way to view a scale is from the point of view of the relations on
A defined by x ≤i y iff φi (x) ≤ φi (y). These are called prewellorderings.
They are well orderings if we mod out by x ≡i y which is defined by

x ≡i y iff x ≤i y and y ≤i x.

The second item says that these relations get finer and finer as i increases.
The third item says that in the “limit” we get a linear order. The fourth
item is some sort of continuity condition. And the last two items are the
definability properties of the scale.
Before proving the lemma, let us deduce uniformity from it. We do not
use the last item in the lemma. First let us show that for any nonempty Π11
set A ⊆ ω ω there exists a Π11 singleton B ⊆ A. Define

x ∈ B iff x ∈ A and ∀n∀y P (n, x, y).

Since P is Π11 the set B is Π11 . Clearly B ⊆ A, and also by item (2) of the
lemma, B can have at most one element. So it remains to show that B is
nonempty. Define αi = min{φi (x) : x ∈ A}. For each i choose xi ∈ A such
that φi (xi ) = αi .

Claim: If n > i then φi (xn ) = αi .

Proof:
By choice of xn for every y ∈ A we have φn (xn ) ≤ φn (y). By item
(2) in the lemma, for every y ∈ A we have that φi (xn ) ≤ φi (y) and hence
φi (xn ) = αi .

By item (4) there exists x ∈ A such that limn→∞ xn = x and φi (x) ≤ αi
all i. By the minimality of αi it must be that φi (x) = αi . So x ∈ B and we
are done.
Now to prove a more general case of uniformity suppose that R ⊆ ω ω ×ω ω
is Π11 . Let φi : R → OR be scale given by the lemma and

P ⊆ ω × (ω ω × ω ω ) × (ω ω × ω ω )
Descriptive Set Theory and Forcing 92

be the Π11 predicate given by item (5). Then define the Π11 set S ⊆ ω ω × ω ω
by
(x, y) ∈ S iff (x, y) ∈ R and ∀z∀n P (n, (x, y), (x, z)).
The same proof shows that S uniformizes R.
The proof of the Lemma will need the following two elementary facts
about well-founded trees. For T, T̂ subtrees of Q<ω we say that σ : T → T̂
is a tree embedding iff for all s, t ∈ T if s ⊂ t then σ(s) ⊂ σ(t). Note
that s ⊂ t means that s is a proper initial segment of t. Also note that tree
embeddings need not be one-to-one. We write T  T̂ iff there exists a tree
embedding from T into T̂ . We write T ≺ T̂ iff there is a tree embedding which
takes the root node of T to a nonroot node of T̂ . Recall that r : T → OR is a
rank function iff for all s, t ∈ T if s ⊂ t then r(s) > r(t). Also the rank of T
is the minimal ordinal α such that there exists a rank function r : T → α + 1.

Lemma 22.3 Suppose T  T̂ and T̂ is well-founded, i.e., [T̂ ] = ∅, then T

is well-founded and rank of T is less than or equal to rank of T̂ .

Proof:
Let σ : T → T̂ be a tree embedding and r : T̂ → OR a rank function.
Then r ◦ σ is a rank function on T .


Lemma 22.4 Suppose T and T̂ are well founded trees and rank of T is less
than or equal rank of T̂ , then T  T̂ .

Let rT and rT̂ be the canonical rank functions on T and T̂ (see Theorem 7.1).
Inductively define σ : T ∩ Qn → T̂ ∩ Qn , so as to satisfy rT (s) ≤ rT̂ (σ(s)).

Now we a ready to prove the existence of scales (Lemma 22.2). Let

ω1− = {−1} ∪ ω1

be well-ordered in the obvious way. Given a well-founded tree T ⊆ ω <ω with

rank function rT extend rT to all of ω <ω by defining rT (s) = −1 if s ∈ / T.
ω 1
Now suppose A ⊆ ω is Π1 and x ∈ A iff Tx is well-founded (see Theorem
17.4). Let {sn : n < ω} be a recursive listing of ω <ω with s0 = hi. For each
n < ω define ψn : A → ω1− × ω × · · · ω1− × ω by

ψn (x) = hrTx (s0 ), x(0), rTx (s1 ), x(1), . . . , rTx (sn ), x(n)i.
Descriptive Set Theory and Forcing 93

The set ω1− × ω × · · · ω1− × ω is well-ordered by the lexicographical order. The

scale φi is just obtained by mapping the range of ψi order isomorphically
to the ordinals. (Remark: by choosing s0 = hi, we guarantee that the first
coordinate is always the largest coordinate, and so the range of ψi is less than
or equal to ω1 .) Now we verify the properties.
For item (2): if ψi+1 (x) ≤lex ψi+1 (y), then ψi (x) ≤lex ψi (y). This is true
because we are just taking the lexicographical order of a longer sequence.
For item (3): if ∀i ψi (x) = ψi (y), then x = y. This is true, because

ψi (x) = ψi (y) implies x  i = y  i.

For item (4): Suppose hxn : n < ωi ∈ Aω and for every i and for all but
finitely many n ψi (xn ) = ti . Then since ψi (xn ) contains xn  i there must
be x ∈ ω ω such that limn→∞ xn = x. Note that since {sn : n ∈ ω} lists every
element of ω <ω , we have that for every s ∈ ω <ω there exists r(s) ∈ OR such
that rTxn (s) = r(s) for all but finitely many n. Using this and

lim Txn = Tx
n→∞

it follows that r is a rank function on Tx . Consequently x ∈ A. Now since

rTx (s) ≤ r(s), it follows that ψi (x) ≤lex ti .
For item (5),(6): The following set is Σ11 :

{(T, T̂ ) : T, T̂ are subtrees of ω <ω , T  T̂ }.

Consequently, assuming that T, T̂ are well-founded, to say that rT (s) ≤ rT̂ (s)
is equivalent to saying there exists a tree embedding which takes s to s. Note
that this is Σ11 . This shows that it is possible to define a Σ11 set S ⊆ ω×ω ω ×ω ω
such that for every x, y ∈ A we have (n, x, y) ∈ S iff

hrTx (s0 ), x(0), rTx (s1 ), x(1), . . . , rTx (sn ), x(n)i

is lexicographically less than or equal to

hrTy (s0 ), y(0), rTy (s1 ), y(1), . . . , rTy (sn ), y(n)i.

Note that if (n, x, y) ∈ S and y ∈ A then x ∈ A, since Ty is a well-founded

tree and S implies Tx  Ty , so Tx is well-founded and so x ∈ A.
To get the Π11 relation P (item (5)), instead of saying T can be embedded
into T̂ we say that T̂ cannot be embedded properly into T ,i.e., T̂ 6≺ T or
Descriptive Set Theory and Forcing 94

in other words, there does not exists a tree embedding σ : T̂ → T such

that σ(hi) 6= hi. This is a Π11 statement. For T and T̂ well-founded trees
saying that rank of T is less than or equal to T̂ is equivalent to saying rank
of T̂ is not strictly smaller than the rank of T . But by Lemma 22.4 this is
equivalent to the nonexistence of such an embedding. Note also that if x ∈ A
and y ∈ / A, then we will have P (n, x, y) for every n. This is because Ty is not
well-founded and so cannot be embedded into the well-founded tree Tx .
This finishes the proof of the Scale Lemma 22.2.

The scale property was invented by Moschovakis [88] to show how deter-
minacy could be used to get uniformity properties10 in the higher projective
classes. He was building on earlier ideas of Blackwell, Addison, and Martin.
The 500 page book by Kuratowski and Mostowski [60] ends with a proof of
the uniformization theorem.

10
I have yet to see any problem, however complicated, which, when you looked at it in
the right way, did not become still more complicated. Poul Anderson
23 MARTIN’S AXIOM AND CONSTRUCTIBILITY 95

23 Martin’s axiom and Constructibility

Theorem 23.1 (Gödel see Solovay [103]) If V=L, there exists uncountable
Π11 set A ⊆ ω ω which contains no perfect subsets.

Proof:
Let X be any uncountable Σ12 set containing no perfect subsets. For
example, a ∆12 Luzin set would do (Theorem 18.1). Let R ⊂ ω ω × ω ω be Π11
such that x ∈ X iff ∃y R(x, y). Use Π11 uniformization (Theorem 22.1) to
get S ⊆ R with the property that X is the one-to-one image of S via the
projection map π(x, y) = x. Then S is an uncountable Π11 set which contains
no perfect subset. This is because if P ⊆ S is perfect, then π(P ) is a perfect
subset of X.

Note that it is sufficient to assume that ω1 = (ω1 )L . Suppose A ∈ L is
defined by the Π11 formula θ. Then let B be the set which is defined by θ in
V . So by Π11 absoluteness A = B ∩ L. The set B cannot contain a perfect
set since the sentence:

∃T T is a perfect tree and ∀x (x ∈ [T ] implies θ(x))

is a Σ12 and false in L and so by Shoenfield absoluteness (Theorem 20.2) must

be false in V . It follows then by the Mansfield-Solovay Theorem 21.1 that B
cannot contain a nonconstructible real and so A = B.
Actually, by tracing thru the actual definition of X one can see that
the elements of the uniformizing set S (which is what A is) consist of pairs
(x, y) where y is isomorphic to some Lα and x ∈ Lα . These pairs are reals
which witness their own constructibility, so one can avoid using the Solovay-
Mansfield Theorem.

Corollary 23.2 If ω1 = ω1L , then there exists a Π11 set of constructible reals
which contains no perfect set.

Theorem 23.3 (Martin-Solovay [74]) Suppose MA + ¬CH + ω1 = (ω1 )L .

Then every A ⊆ 2ω of cardinality ω1 is Π11 .
e
Proof:
Let A ⊆ 2ω be a uncountable Π11 set of constructible reals and let B be an
arbitrary subset of 2ω of cardinality ω1 . Arbitrarily well-order the two sets,
A = {aα : α < ω1 } and B = {bα : α < ω1 }.
Descriptive Set Theory and Forcing 96

By Theorem 5.1 there exists two sequences of Gδ sets hUn : n < ωi and
hVn : n < ωi such that for every α < ω1 for every n < ω

aα (n) = 1 iff bα ∈ Un

and
bα (n) = 1 iff aα ∈ Vn .
This is because the set {aα : bα (n) = 1}, although it is an arbitrary subset
of A, is relatively Gδ by Theorem 5.1.
But note that b ∈ B iff ∀a ∈ 2ω

[∀n (a(n) = 1 iff b ∈ Un )] implies [a ∈ A and ∀n (b(n) = 1 iff a ∈ Un )].

Since A is Π11 this definition of B has the form:

∀a([Π03 ] implies [Π11 and Π03 ])

e e
So B is Π11 .
 e
Note that if every set of reals of size ω1 is Π11 then every ω1 union of Borel
sets is Σ12 . To see this let hBα : α < ω1 i be any
e sequence of Borel sets. Let
U be a euniversal Π11 set and let hxα : α < ω1 i be a sequence such that
e
Bα = {y : (xα , y) ∈ U }.

Then [
y∈ Bα iff ∃x x ∈ {xα : α < ω1 } ∧(x, y) ∈ U.
α<ω1

But {xα : α < ω1 } is Π11 and so the union is Σ12 .

e e
24 Σ12 WELL-ORDERINGS 97

24 Σ12 well-orderings
Theorem 24.1 (Mansfield [71]) If (F,
) is a Σ12 well-ordering, i.e.,
F ⊆ ω ω and
⊆ F 2
are both Σ12 , then F is a subset of L.
Proof:
We will use the following:
Lemma 24.2 Assume there exists z ∈ 2ω such that z ∈ / L. Suppose f : P →
F is a 1-1 continuous function from the perfect set P and both f and P are
coded in L, then there exists Q ⊆ P perfect and g : Q → F 1-1 continuous so
that both g and Q are coded in L and for every x ∈ Q we have g(x)
f (x).
Proof:
(Kechris [52]) First note that there exists σ : P → P an autohomeomor-
phism coded in L such that for every x ∈ P we have σ(x) 6= x but σ 2 (x) = x.
To get this let c : 2ω → 2ω be the complement function, i.e., c(x)(n) = 1−x(n)
which just switches 0 and 1. Then c(x) 6= x but c2 (x) = x. Now if h : P → 2ω
is a homeomorphism coded in L, then σ = h−1 ◦ c ◦ h works.
Now let A = {x ∈ P : f (σ(x))
f (x)}. The set A is a Σ12 set with code
in L. Now since P is coded in L there must be a z ∈ P such e that z ∈
/ L.
Note that σ(z) ∈ / L also. But either
f (σ(z))
f (z) or f (z) = f (σ 2 (z))
f (σ(z))
and so either z ∈ A or σ(z) ∈ A. In either case A has a nonconstructible
member and so by the Mansfield-Solovay Theorem 21.1 the set A contains a
perfect set Q coded in L. Let g = f ◦ σ.

Assume there exists z ∈ F such that z ∈ / L. By the Mansfield-Solovay
Theorem there exists a perfect set P coded in L such that P ⊆ F . Let
P0 = P and f0 be the identity function. Repeatedly apply the Lemma to
obtain fn : Pn → F so that for every T n and Pn+1 ⊆ Pn , for every x ∈ Pn+1
fn+1 (x)
fn (x). But then if x ∈ n<ω the sequence hfn (x) : n < ωi is a
descending
sequence with contradicts the fact that
is a well-ordering.

Friedman [28] proved the weaker result that if there is a Σ12 well-ordering
of the real line, then ω ω ⊆ L[g] for some g ∈ ω ω . e
25 LARGE Π12 SETS 98

25 Large Π12 sets

A set is Π12 iff it is the complement of a Σ12 set. Unlike Σ12 sets which cannot
have sizeestrictly in between ω1 and the econtinuum (Theorem
e 21.1), Π12 sets
11
can be practically anything. e

Theorem 25.1 (Harrington [35]) Suppose V is a model of set theory which

satisfies ω1 = ω1L and B is arbitrary subset of ω ω in V . Then there exists a
ccc extension of V , V [G], in which B is a Π12 set.
e
Proof:
Let PB be the following poset. p ∈ PB iff p is a finite consistent set of
sentences of the form:
◦
1. “[s]∩ C n = ∅”, or
◦
2. “x ∈C n , where x ∈ B.

This partial order is isomorphic to Silver’s view of almost

S disjoint sets forcing
(Theorem 5.1). So forcing with PB creates an Fσ set n∈ω Cn so that
[
∀x ∈ ω ω ∩ V (x ∈ B iff x ∈ Cn ).
n<ω
Q
Forcing with the direct sum of ω1 copies of PB , α<ω1 PB , we have that
\
∀x ∈ ω ω ∩ V [hGα : α < ω1 i](x ∈ B iff x ∈ ∪n<ω Cnα ).
α<ω1

One way to see this is as follows. Note that in any case

\
B⊆ ∪n<ω Cnα .
α<ω1

So it is the other implication which needs to be proved. By ccc, for any

x ∈ V [hGα : α < ω1 i] there exists β < ω1 with x ∈ V [hGα : α < βi]. But
considering V [hGα : α < βi] as the new ground model, then Gβ would be PB -
generic over V [hGα : α < βi] and hence if x ∈ / ∪n<ω Cnβ .
/ B we would have x ∈
Another argument will be given in the proof of the next lemma.
11
It’s life Jim, but not as we know it.- Spock of Vulcan
Descriptive Set Theory and Forcing 99

Lemma 25.2 Suppose hcα : α < ω1 be a sequence in V of elements of ω ω

and haα : α < ω1 i is a sequence in V [hGα : α < ω1 i] of elements of 2ω . Using
Silver’s forcing add a sequence of Π02 sets hUn : n < ωi such that
e
∀n ∈ ω∀α < ω1 (aα (n) = 1 iff cα ∈ Un ).

Then
\
V [hGα : α < ω1 i][hUn : n < ωi] |= ∀x ∈ ω ω (x ∈ B iff x ∈ ∪n<ω Cnα ).
α<ω1

Proof:
The lemma is not completely trivial, since adding the hUn : n < ωi adds
new elements of ω ω which may somehow sneak into the ω1 intersection.
Working in V define p ∈ Q iff p is a finite set of consistent sentences of
the form:
1. “[s] ⊆ Un,m ” where s ∈ ω <ω , or

2. “cα ∈ Un,m ”.
Here we intend that Un = ∩m∈ω Un,m . Since the c’s are in V it is clear that
the partial order Q is too. Define
Y
P = {(p, q) ∈ ( PB ) × Q : if “cα ∈ Un,m ”∈ q, then p |` aα (n) = 1}.
α<ω1

Note that P is a semi-lower-lattice, i.e., if (p0 , q0 ) and (p1 , q1 ) are compatible

elements of P, then (p0 ∪ p1 , q0 ∪ q1 ) is their greatest lower bound. This is
another way to view the iteration, i.e, P is dense in the usual iteration. Not
every iteration has this property, one which Harrington calls “innocuous”.
Now to prove the lemma, suppose for contradiction that
◦ \ ◦
(p, q) |`x∈ ∪n<ω Cnα and x∈ / B.
α<ω1

To simplify the notation, assume (p, q) = (∅, ∅). Since P has the ccc a se-
quence of Working in V let iAn : n ∈ ωh be a sequence of maximal antichains
◦
of P which decide x, i.e. for (p, q) ∈ An there exists s ∈ ω n such that
◦
(p, q) |`x n = š.
Descriptive Set Theory and Forcing 100

Since P has the ccc, the An are countable and we can find S an α < ω which
does not occur in the support of any p for any (p, q) in n∈ω An . Since x is
forced to be in ∪n<ω Cnα there exists (p, q) and n ∈ ω such that
◦
(p, q) |`x∈ Cnα .

Let “xi ∈ Cnα ” for i < N be all the sentences of this type which occur in
p(α). Since we are assuming x is being forced not in B it must be different
than all the xi , so there must be an m, (p̂, q̂) ∈ Am , and s ∈ ω m , such that

1. (p̂, q̂) and (p, q) are compatible,

◦
2. (p̂, q̂) |`x m = š, and

3. xi  m 6= s for every i < N .

((To get (p̂, q̂) and s let G be a generic filter containing (p, q), then since
xG 6= xi for every i < N there must be m < ω and s ∈ ω m such that
xG  m = s and s 6= xi  m for every i < N . Let (p̂, q̂) ∈ G ∩ Am .))
Now consider (p ∪ p̂, q ∪ q̂) ∈ P. Since α was not in the support of p̂,

(p ∪ p̂)(α) = p(α).

Since s was chosen so that xi ∈

/ [s] for every i < N ,

p(α) ∪ {[s] ∩ Cnα = ∅}

is a consistent set of sentences, hence an element of PB . This is a contradic-

tion, the condition
(p ∪ p̂ ∪ {[s] ∩ Cnα = ∅}, q ∪ q̂)
/ Cnα .
forces x ∈ Cnα and also x ∈

Let F be a universal Σ02 set coded in V and let haα ∈ 2ω : α < ω1 i be
such that [
Faα = Cnα .
n∈ω

Let C = hcα : α < ω1 i be a Π11 set in V . Such a set exists since ω1 = ω1L .

Lemma 25.3 In V [hGα : α < ω1 i][hUn : n < ωi] the set B is Π12 .
e
Descriptive Set Theory and Forcing 101

Proof:
x ∈ B iff x ∈ α<ω1 n∈ω Cnα iff x ∈ α<ω1 Faα iff
T S T
∀a, c if c ∈ C and ∀n (a(n) = 1 iff c ∈ Un ), then (a, x) ∈ F , i.e. (x ∈ Fa ).

Note that

• “c ∈ C” is Π11 ,

• “∀n (a(n) = 1 iff c ∈ Un )” is Borel, and

• “(a, x) ∈ F ” is Borel,

and so this final definition for B has the form:

∀((Π11 ∧ Borel)) → Borel)

Therefore B is Π12 .
 e
Harrington [35] also shows how to choose B so that the generic extension
has a ∆13 well-ordering of ω ω . He also shows how to take a further innocuous
extensions
e to make B a ∆13 set and to get a ∆13 well-ordering.
 102

Part III
Classical Separation Theorems
26 Souslin-Luzin Separation Theorem
Define A ⊆ ω ω to be κ-Souslin iff there exists a tree T ⊆ n
× ωn)
S
n<ω (κ
such that
y ∈ A iff ∃x ∈ κω ∀n < ω (x  n, y  n) ∈ T.
In this case we write A = p[T ], the projection of the infinite branches of the
tree T . Note that ω-Souslin is the same as Σ11 .
e family of subsets of ω ω con-
Define the κ-Borel sets to be the smallest
taining the usual Borel sets and closed under intersections or unions of size
κ and complements.

Theorem 26.1 Suppose A and B are disjoint κ-Souslin subsets of ω ω . Then

there exists a κ-Borel set C which separates A and B, i.e., A ⊆ C and
C ∩ B = ∅.

Proof:
Let A = p[TA ] and B = p[TB ]. Given a tree T ⊆ n<ω (κn × ω n ), and
S
s ∈ κ<ω , t ∈ ω <ω (possibly of different lengths), define

T s,t = {(ŝ, t̂) ∈ T : (s ⊆ ŝ or ŝ ⊆ s) and (t ⊆ t̂ or t̂ ⊆ t)}.

Lemma 26.2 Suppose p[TAs,t ] cannot be separated from p[TBr,t ] by a κ-Borel

set. Then for some α < κ the set
p[TAsˆα,t ] cannot be separated from p[TBr,t ] by a κ-Borel set.

Proof:
Note that p[TAs,t ] = α<κ p[TAsˆα,t ]. If there were no such α, then for every
S
α we would have a κ-Borel set Cα with

p[TAsˆα,t ] ⊆ Cα and Cα ∩ p[TBr,t ] = ∅.

Cα is a κ-Borel set separating p[TAs,t ] and p[TBr,t ].

S
But then α<κ

26 SOUSLIN-LUZIN SEPARATION THEOREM 103

Lemma 26.3 Suppose p[TAs,t ] cannot be separated from p[TBr,t ] by a κ-Borel

set. Then for some β < κ
p[TAs,t ] cannot be separated from p[TBrˆβ,t ] by a κ-Borel set.

Proof:
Since p[TBr,t ] = β<κ p[TBrˆβ,t ], if there were no such β then for every β we
S
would have κ-Borel set Cβ with

p[TAs,t ] ⊆ Cβ and Cβ ∩ p[TBrˆβ,t ] = ∅.

Cβ is a κ-Borel set separating p[TAs,t ] and p[TBr,t ].

T
But then β<κ


Lemma 26.4 Suppose p[TAs,t ] cannot be separated from p[TBr,t ] by a κ-Borel

set. Then for some n < ω
p[TAs,tˆn ] cannot be separated from p[TBr,tˆn ] by a κ-Borel set.

Proof:
Note that
p[TAs,tˆn ] = p[TAs,t ] ∩ [tˆn]
and
p[TBr,tˆn ] = p[TBr,t ] ∩ [tˆn].
Thus if Cn ⊆ [tˆn] were to separate p[TAs,tˆn ] and p[TBr,tˆn ] for each n, then
S s,t r,t
n<ω Cn would separate p[TA ] from p[TB ].

To prove the separation theorem apply the lemmas iteratively in rotation
to obtain, u, v ∈ κω and x ∈ ω ω so that for every n, p[TAun,xn ] cannot be
separated from p[TBvn,xn ]. But necessarily, for every n

(u  n, x  n) ∈ TA and (v  n, x  n) ∈ TB

otherwise either p[TAun,xn ] = ∅ or p[TBvn,xn ] = ∅ and they could be separated.

But this means that x ∈ p[TA ] = A and x ∈ p[TB ] = B contradicting the fact
that they are disjoint.

27 KLEENE SEPARATION THEOREM 104

27 Kleene Separation Theorem

We begin by defining the hyperarithmetic subsets of ω ω . We continue
with our view of Borel sets as well-founded trees with little dohickey’s (basic
clopen sets) attached to its terminal nodes.
A code for a hyperarithmetic set is a triple (T, p, q) where T is a
recursive well-founded subtree of ω <ω , p : T >0 → 2 is recursive, and q :
T 0 → B is a recursive map, where B is the set of basic clopen subsets of ω ω
including the empty set. Given a code (T, p, q) we define hCs : s ∈ T i as
follows.

• if s is a terminal node of T , then

Cs = q(s)

• if s is a not a terminal node and p(s) = 0, then

[
Cs = {Csˆn : sˆn ∈ T },

and
• if s is a not a terminal node and p(s) = 1, then
\
Cs = {Csˆn : sˆn ∈ T }.

Here we are being a little more flexible by allowing unions and intersections
at various nodes.
Finally, the set C coded by (T, p, q) is the set Chi . A set C ⊆ ω ω is
hyperarithmetic iff it is coded by some recursive (T, p, q).

Theorem 27.1 (Kleene [55]) Suppose A and B are disjoint Σ11 subsets of
ω ω . Then there exists a hyperarithmetic set C which separates them, i.e.,
A ⊆ C and C ∩ B = ∅.

Proof:
This amounts basically to a constructive proof of the classical Separation
Theorem 26.1.
S Let An
= p[TA ] and B = p[TB ] where TA and TB are recursive subtrees of
n
n∈ω (ω × ω ), and

p[TA ] = {y : ∃x∀n (x  n, y  n) ∈ TA }
Descriptive Set Theory and Forcing 105

and similarly for p[TB ]. Now define the tree

T = {(u, v, t) : (u, t) ∈ TA and (v, t) ∈ TB }.

Notice that T is recursive tree which is well-founded. Any infinite branch thru
T would give a point in the intersection of A and B which would contradict
the fact that they are disjoint.
Let T + be the tree of all nodes which are either “in” or “just out” of T ,
i.e., (u, v, t) ∈ T + iff (u  n, v  n, t  n) ∈ T where |u| = |v| = |t| = n + 1.
Now we define the family of sets

hC(u,v,t) : (u, v, t) ∈ T + i

as follows.
Suppose (u, v, t) ∈ T + is a terminal node of T + . Then since (u, v, t) ∈
/T
either (u, t) ∈ / TA in which case we define C(u,v,t) = ∅ or (u, t) ∈ TA and
(v, t) ∈/ TB in which case we define C(u,v,t) = [t]. Note that in either case
C(u,v,t) ⊆ [t] separates p[TAu,t ] from p[TBv,t ].

Lemma 27.2 Suppose hAn : n < ωi, hBm : m < ωi hCnm : n, m < ωi
are
S suchT that for every
T n and
S m Cnm separatesS An from BmS. Then both
n<ω C
m<ω nm and m<ω C
n<ω nm separate n<ω An from m<ω Bm .

Proof:
Left to reader.

It follows from the Lemma that if we let
[ \ [
C(u,v,t) = C(uˆn,vˆm,tˆk)
k<ω m<ω n<ω

(or any other permutation12 of

T S
and ), then by induction on rank of
(u, v, t) in T + that C(u,v,t) ⊆ [t] separates p[TAu,t ] from p[TBv,t ]. Hence, C =
C(hi,hi,hi) separates A = p[TA ] from B = p[TB ].
To get a hyperarithmetic code use the tree consisting of all subsequences
of sequences of the form,

ht(0), v(0), u(0), . . . , t(n), v(n), u(n)i

12
Algebraic symbols are used when you do not know what you are talking about
(Philippe Schnoebelen).
Descriptive Set Theory and Forcing 106

where (u, v, t) ∈ T + . Details are left to the reader.


The theorem also holds for A and B disjoint Σ11 subsets of ω. One way
to see this is to identify ω with the constant functions in ω ω . The definition
of hyperarithmetic code (T, p, q) is changed only by letting q map into the
finite subsets of ω.

Theorem 27.3 If C is a hyperarithmetic set, then C is ∆11 .

Proof:
This is true whether C is a subset of ω ω or ω. We just do the case C ⊆ ω ω .
Let (T, p, q) be a hyperarithmetic code for C. Then x ∈ C iff
there exists a function in : T → {0, 1} such that

1. if s a terminal node of T , then in(s) = 1 iff x ∈ q(s),

2. if s ∈ T and not terminal and p(s) = 0, then in(s) = 1 iff there exists
n with sˆ n ∈ T and in(sˆ n) = 1,

3. if s ∈ T and not terminal and p(s) = 1, then

in(s) = 1 iff for all n with sˆ n ∈ T we have in(sˆ n) = 1, and finally,

4. in(hi) = 1.

Note that (1) thru (4) are all ∆11 (being a terminal node in a recursive tree
is Π01 , etc). It is clear that in is just coding up whether or not x ∈ Cs for
s ∈ T . Consequently, C is Σ11 . To see that ∼ C is Σ11 note that x ∈/ C iff
there exists in : T → {0, 1} such that (1), (2), (3), and (4)0 where
40 in(hi) = 0.


Corollary 27.4 A set is ∆11 iff it is hyperarithmetic.

Corollary 27.5 If A and B are disjoint Σ11 sets, then there exists a ∆11 set
which separates them.

For more on the effective Borel hierarchy, see Hinman [40]. See Barwise
[10] for a model theoretic or admissible sets approach to the hyperarithmetic
hierarchy.
28 Π11 -REDUCTION 107

28 Π11-Reduction
We say that A0 ,B0 reduce A,B iff
1. A0 ⊆ A and B0 ⊆ B,

2. A0 ∪ B0 = A ∪ B, and

3. A0 ∩ B0 = ∅.
Π11 -reduction is the property that every pair of Π11 sets can be reduced by
a pair of Π11 sets. The sets can be either subsets of ω or of ω ω .

Theorem 28.1 Π11 -uniformity implies Π11 -reduction.

Proof:
Suppose A, B ⊆ X are Π11 where X = ω or X = ω ω . Let

P = (A × {0}) ∪ (B × {1}).

Then P is a Π11 subset of X × ω ω and so by Π11 -uniformity (Theorem 22.1)

there exists Q ⊆ P which is Π11 and for every x ∈ X, if there exists i ∈ {0, 1}
such that (x, i) ∈ P , then there exists a unique i ∈ {0, 1} such that (x, i) ∈ Q.
Hence, letting
A0 = {x ∈ X : (x, 0) ∈ Q}
and
B0 = {x ∈ X : (x, 1) ∈ Q}
gives a pair of Π11 sets which reduce A and B.

There is also a proof of reduction using the prewellordering property,
which is a weakening of the scale property used in the proof of Π11 -uniformity.
So, for example, suppose A and B are Π11 subsets of ω ω . Then we know there
are maps from ω ω to trees,

x 7→ Txa and y 7→ Tyb

which are “recursive” and

x ∈ A iff Txa is well-founded and
y ∈ B iff Tyb is well-founded.
Now define
Descriptive Set Theory and Forcing 108

1. x ∈ A0 iff x ∈ A and not (Txb ≺ Txa ), and

2. x ∈ B0 iff x ∈ B and not (Txa  Txb ).

Since ≺ and  are both Σ11 it is clear, that A0 and B0 are Π11 subsets of A
and B respectively. If x ∈ A and x ∈ / B, then Txa is well-founded and Txb is
b a
ill-founded and so not (Tx ≺ Tx ) and a ∈ A0 . Similarly, if x ∈ B and x ∈ / A,
b a
then x ∈ B0 . If x ∈ A ∩ B, then both Tx and Tx are well-founded and either
Txa  Txb , in which case x ∈ A0 and x ∈/ B0 , or Txb ≺ Txa , in which case x ∈ B0
and x ∈ / A0 .

Theorem 28.2 Π11 -reduction implies Σ11 -separation, i.e., for any two dis-
joint Σ11 sets A and B there exists a ∆11 -set C which separates them. i.e.,
A ⊆ C and C ∩ B = ∅.

Proof:
Note that ∼ A ∪ ∼ B = X. If A0 and B0 are Π11 sets reducing ∼ A and
∼ B, then ∼ A0 = B0 , so they are both ∆11 . If we set C = B0 , then

C = B0 =∼ A0 ⊆∼ A

so C ⊆∼ A and therefore A ⊆ C. On the other hand C = B0 ⊆∼ B implies

C ∩ B = ∅.

29 ∆11 -CODES 109

29 ∆11-codes
Using Π11 -reduction and universal sets it is possible to get codes for ∆11 subsets
of ω and ω ω .
Here is what we mean by ∆11 codes for subsets of X where X = ω or
X = ωω .
There exists a Π11 sets C ⊆ ω × ω ω and P ⊆ ω × ω ω × X and a Σ11 set
S ⊆ ω × ω ω × X such that

• for any (e, u) ∈ C

{x ∈ X : (e, u, x) ∈ P } = {x ∈ X : (e, u, x) ∈ S}

• for any u ∈ ω ω and ∆11 (u) set D ⊆ X there exists a (e, u) ∈ C such
that

D = {x ∈ X : (e, u, x) ∈ P } = {x ∈ X : (e, u, x) ∈ S}.

From now on we will write

“e is a ∆11 (u)-code for a subset of X”
to mean (e, u) ∈ C and remember that it is a Π11 predicate.
We also write “D is the ∆11 (u) set coded by e” if “e is a ∆11 (u)-code for a
subset of X” and

D = {x ∈ X : (e, x) ∈ P } = {x ∈ X : (e, x) ∈ S}.

Note that x ∈ D can be said in either a Σ11 (u) way or Π11 (u) way, using either
S or P .

Theorem 29.1 (Spector-Gandy Theorem [105],[31] ) ∆11 codes exist.

Proof:
Let U ⊆ ω × ω ω × X be a Π11 set which is universal for all Π11 (u) sets,
i.e., for every u ∈ ω ω and A ∈ Π11 (u) with A ⊆ X there exists e ∈ ω such
that A = {x ∈ X : (e, u, x) ∈ U }. For example, to get such a U proceed
as follows. Let {e}u be the partial function you get by using the eth Turing
machine with oracle u. Then
S define (e, u, x) ∈ U iff {e}u is the characteristic
function of a tree T ⊆ n<ω (ω n × ω n ) and Tx = {s : (s, x  |s|) ∈ T } is
well-founded.
Descriptive Set Theory and Forcing 110

Now get a doubly universal pair. Let e 7→ (e0 , e1 ) be the usual recursive
unpairing function from ω to ω × ω and define

U 0 = {(e, u, x) : (e0 , u, x) ∈ U }

and
U 1 = {(e, u, x) : (e1 , u, x) ∈ U }.
The pair of sets U 0 and U 1 are Π11 and doubly universal, i.e., for any u ∈ ω ω
and A and B which are Π11 (u) subsets of X there exists e ∈ ω such that

A = {x : (e, u, x) ∈ U 0 }

and
B = {x : (e, u, x) ∈ U 1 }.
Now apply reduction to obtain P 0 ⊆ U 0 and P 1 ⊆ U 1 which are Π11 sets.
0 1 0
Note that the by the nature of taking cross sections, Pe,u and Pe,u reduce Ue,u
1
and Ue,u . Now we define
0
• “e is a ∆11 (u) code” iff ∀x ∈ X(x ∈ Pe,u 1
or x ∈ Pe,u ), and

• P = P 0 and S =∼ P 1 .

Note that e is a ∆11 (u) code is a Π11 statement in (e, u). Also if e is a ∆11 (u)
code, then P(e,u) = Se,u and so its a ∆11 (u) set. Furthermore if D ⊆ X is a
∆11 (u) set, then since U 0 and U 1 were a doubly universal pair, there exists e
0 1 0 0
such that Ue,u = D and Ue,u =∼ D. For this e it must be that Ue,u = Pe,u
1 1
and Ue,u = Pe,u since the P ’s reduce the U ’s. So this e is a ∆11 (u) code which
codes the set D.


Corollary 29.2 {(x, u) ∈ P (ω) × ω ω : x ∈ ∆11 (u)} is Π11 .

Proof:
x ∈ ∆11 (u) iff ∃e ∈ ω such that

1. e is a ∆11 (u) code,

2. ∀n if n ∈ x, then n is in the ∆11 (u)-set coded by e, and

3. ∀n if n is the ∆11 (u)-set coded by e, then n ∈ x.

Descriptive Set Theory and Forcing 111

Note that clause (1) is Π11 . Clause (2) is Π11 if we use that (e, u, n) ∈ P is
equivalent to “n is in the ∆11 (u)-set coded by e”. While clause (3) is Π11 if we
use that (e, u, n) ∈ S is equivalent to “n is in the ∆11 (u)-set coded by e”.

We say that y ∈ ω ω is ∆11 (u) iff its graph {(n, m) : y(n) = m} is ∆11 (u).
Since being the graph a function is a Π02 property it is easy to see how to
obtain ∆11 (u) codes for functions y ∈ ω ω .

Corollary 29.3 Suppose θ(x, y, z) is a Π11 formula, then

ψ(y, z) = ∃x ∈ ∆11 (y) θ(x, y, z)

is a Π11 formula.

Proof:
ψ(y, z) iff
∃e ∈ ω such that

1. e is a ∆11 (y) code, and

2. ∀x if x is the set coded by (e, y), then θ(x, y, z).

This will be Π11 just in case the clause “x is the set coded by (e, y)” is Σ11 .
But this is ∆11 provided that e is a ∆11 (y) code, e.g., for x ⊆ ω we just say:
∀n ∈ ω

1. if n ∈ x then (e, y, n) ∈ S and

2. if (e, y, n) ∈ P , then n ∈ x.

Both of these clauses are Σ11 since S is Σ11 and P is Π11 . A similar argument
works for x ∈ ω ω .

The method of this corollary also works for the quantifier

∃D ⊆ ω ω such that D ∈ ∆(y) θ(D, y, z).

It is equivalent to say ∃e ∈ ω such that e is a ∆11 (y) code for a subset of

ω ω and θ(. . . , y, z) where occurrences of the “q ∈ D” in the formula θ have
been replaced by either (e, y, q) ∈ P or (e, y, q) ∈ S, whichever is necessary
to makes θ come out Π11 .
Descriptive Set Theory and Forcing 112

Corollary 29.4 Suppose f : ω ω → ω ω is Borel, B ⊆ ω ω is Borel, and f is

one-to-one on B. Then the image of B under f , f (B), is Borel.

Proof:
By relativizing the following argument to an arbitrary parameter we may
assume that the graph of f and the set B are ∆11 . Define

R = {(x, y) : f (x) = y and x ∈ B}.

Then for any y the set

{x : R(x, y)}
is a ∆11 (y) singleton (or empty). Consequently, its unique element is ∆11 in
y. It follows that

y ∈ f (B) iff ∃x R(x, y) iff ∃x ∈ ∆11 (y) R(x, y)

and so f (B) is both Σ11 and Π11 .


Many applications of the Gandy-Spector Theorem exist. For example, it
is shown (assuming V=L in all three cases) that

1. there exists an uncountable Π11 set which is concentrated on the ratio-

nals (Erdos, Kunen, and Mauldin [21]),

2. there exists a Π11 Hamel basis (Miller [85]), and

3. there exists a topologically rigid Π11 set (Van Engelen, Miller, and Steel
[18]).
 113

Part IV
Gandy Forcing
30 Π11 equivalence relations
Theorem 30.1 (Silver [101]) Suppose (X, E) is a Π11 equivalence relation,
i.e. X is a Borel set and E ⊆ X 2 is a Π11 equivalence
e relation on X. Then
either E has countably many equivalencee classes or there exists a perfect set
of pairwise inequivalent elements.

Before giving the proof consider the following example. Let W O be the
set of all characteristic functions of well-orderings of ω. This is a Π11 subset
of 2ω×ω . Now define x ' y iff there exists an isomorphism taking x to y or
/ W O. Note that (2ω×ω , ') is a Σ11 equivalence relation with exactly ω1
x, y ∈
equivalence classes. Furthermore, if we restrict ' to W O, then (W O, ') is
a Π11 equivalence relation (since well-orderings are isomorphic iff neither is
isomorphic to an initial segment of the other). Consequently, Silver’s theorem
is the best possible.
The proof we are going to give is due to Harrington [33], see also Kechris
and Martin [53], Mansfield and Weitkamp [73] and Louveau [64]. A model
theoretic proof is given in Harrington and Shelah [38].
We can assume that X is ∆11 and E is Π11 , since the proof readily relativizes
to an arbitrary parameter. Also, without loss, we may assume that X = ω ω
since we just make the complement of X into one more equivalence class.
Let P be the partial order of nonempty Σ11 subsets of ω ω ordered by
inclusion. This is known as Gandy forcing. Note that there are many
trivial generic filters corresponding to Σ11 singletons.

Lemma 30.2 If G is P-genericTover V , then there exists a ∈ ω ω such that

G = {p ∈ P : a ∈ p} and {a} = G.

Proof:
For every n an easy density argument shows that there exists a unique
s ∈ ω n such that [s] ∈ G whereT[s] = {x ∈ ω ω : s ⊆ x}. Define a ∈ ω ω by
[a  n] ∈ G for each n. Clearly, G ⊆ {a}.
Now suppose B ∈ G, we need to show a ∈ B. Let B = p[T ].
30 Π11 EQUIVALENCE RELATIONS 114

Claim: There exists x ∈ ω ω such that p[T xn,an ] ∈ G for every n ∈ ω.

Proof:
This is by induction on n. Suppose p[T xn,an ] ∈ G. Then

p[T xn,an+1 ] ∈ G

since
p[T xn,an+1 ] = [a  n + 1] ∩ p[T xn,an ]
and both of these are in G. But note that
[
p[T xn,an+1 ] = p[T xnˆm,an+1 ]
m∈ω

and so by a density argument there exists m = x(n) such that

p[T xnˆm,an+1 ] ∈ G.

This proves the Claim.


∈ [T ] (since elements of P are nonempty)
By the Claim we have that (x, a) T
and so a ∈ p[T ] = B. Consequently, G = {a}. Now suppose that a ∈ p ∈ P
and p ∈
/ G. Then since

{q ∈ P : q ≤ p or q ∩ p = ∅}

is dense there must be q ∈ G with q ∩ p = ∅. But this is impossible, because

a ∈ q ∩ p, but q ∩ p = ∅ is a Π11 sentence and hence absolute.

We say that a ∈ ω ω is P-generic over V iff G = {p ∈ P : a ∈ p} is
P-generic over V .

Lemma 30.3 If a is P-generic over V and a = ha0 , a1 i (where h, i is the

standard pairing function), then a0 and a1 are both P-generic over V .

Proof:
The proof is symmetric so we just do it for a0 . Note that we are not
claiming that they are product generic only that each is separately generic.
Suppose D ⊆ P is dense open. Let

E = {p ∈ P : {x0 : x ∈ p} ∈ D}.
30 Π11 EQUIVALENCE RELATIONS 115

To see that E is dense let q ∈ P be arbitrary. Define

q0 = {x0 : x ∈ q}.

Since q0 is a nonempty Σ11 set and D is dense, there exists r0 ≤ q0 with

r0 ∈ D. Let
r = {x ∈ q : x0 ∈ r0 }.
Then r ∈ E and r ≤ q.
Since E is dense we have that there exists p ∈ E with a ∈ p and conse-
quently,
a0 ∈ p0 = {x0 : x ∈ p} ∈ D.

Lemma 30.4 Suppose B ⊆ ω ω is Π11 and for every x, y ∈ B we have that

xEy. Then there exists a ∆11 set D with B ⊆ D ⊆ ω ω and such that for
every x, y ∈ D we have that xEy.

Proof:
Let A = {x ∈ ω ω : ∀y y ∈ B → xEy}. Then A is a Π11 set which
contains the Σ11 set B, consequently by the Separation Theorem 27.5 or 28.2
there exists a ∆11 set D with B ⊆ D ⊆ A. Since all elements of B are
equivalent, so are all elements of A and hence D is as required.

Now we come to the heart of Harrington’s proof. Let B be the union of
all ∆11 subsets of ω ω which meet only one equivalence class of E, i.e.
[
B = {D ⊆ ω ω : D ∈ ∆11 and ∀x, y ∈ D xEy}.

Since E is Π11 we know that by using ∆11 codes that this union is Π11 , i.e.,
z ∈ B iff ∃e ∈ ω such that

1. e is a ∆11 code for a subset of ω ω ,

2. ∀x, y in the set coded by e we have xEy, and

3. z is in the set coded by e.

30 Π11 EQUIVALENCE RELATIONS 116

Note that item (1) is Π11 and (2) and (3) are both ∆11 (see Theorem 29.1).

If B = ω ω , then since there are only countably many ∆11 sets, there would
only be countably many E equivalence classes and we are done. So assume
A =∼ B is a nonempty Σ11 set and in this case we will prove that there is a
perfect set of E-inequivalent reals.

Lemma 30.5 Suppose c ∈ ω ω ∩ V . Then

◦
A |`P č E
6 a
◦
where a is a name for the generic real (Lemma 30.2).

Proof:
Suppose not, and let C ⊆ A be a nonempty Σ11 set such that C |` cEa.
We know that there must exists c0 , c1 ∈ C with c0 6 Ec1 . Otherwise there
would exists a ∆11 superset of C which meets only one equivalence class
(Lemma 30.4). But we these are all disjoint from A. Let
Q = {c : c0 ∈ C, c1 ∈ C, and c0 E
6 c1 }.
Note that Q is a nonempty Σ11 set. Let a ∈ Q be P-generic over V . Then by
Lemma 30.3 we have that both a0 and a1 are P-generic over V and a0 ∈ C,
a1 ∈ C, and a0 E
6 a1 . But ai ∈ C and C |` ai Ec means that
6 a1 .
a0 Ec, a1 Ec, and a0 E
This contradicts the fact that E is an equivalence relation.
Note that “E is an equivalence relation” is a Π11 statement hence it is
absolute. Note also that we don’t need to assume that there are a which are
P-generic over V . To see this replace V by a countable transitive model M
of ZFC∗ (a sufficiently large fragment of ZFC) and use absoluteness.

Note that the lemma implies that if (a0 , a1 ) is P × P-generic over V and
a1 ∈ A, then a0 6 E a1 . This is because a1 is P-generic over V [a0 ] and so a0
can be regarded as an element of the ground model.

Lemma 30.6 Suppose M is a countable transitive model of ZFC∗ and P is

a partially ordered set in M . Then there exists {Gx : x ∈ 2ω }, a “perfect” set
of P-filters, such that for every x 6= y we have that (Gx , Gy ) is P × P-generic
over M .
30 Π11 EQUIVALENCE RELATIONS 117

Proof:
Let Dn for n < ω list all dense open subsets of P × P which are in M .
Construct hps : s ∈ 2<ω i by induction on the length of s so that

1. s ⊆ t implies pt ≤ ps and

2. if |s| = |t| = n + 1 and s and t are distinct, then (ps , pt ) ∈ Dn .

Now define for any x ∈ 2ω

Gx = {p ∈ P : ∃n pxn ≤ p}.


Finally to prove Theorem 30.1 let M be a countable transitive set iso-
morphic to an elementary substructure of (Vκ , ∈) for some sufficiently large
κ. Let
{Gx : x ∈ 2ω }
be given by Lemma 30.6 with A ∈ Gx for all x and let

P = {ax : x ∈ 2ω }

be the corresponding generic reals. By Lemma 30.5 we know that for every
x 6= y ∈ 2ω we have that ax 6 E ay . Note also that P is perfect because the
map x 7→ ax is continuous. This is because for any n ∈ ω there exists m < ω
such that every ps with s ∈ 2m decides a  n.


Corollary 30.7 Every Σ11 set which contains a real which is not ∆11 contains
a perfect subset.

Proof:
Let A ⊆ ω ω be a Σ11 set. Define xEy iff x, y ∈/ A or x=y. Then is E is a
Π11 equivalence relation. A ∆11 singleton is a ∆11 real, hence Harrington’s set
B in the above proof must be nonempty. Any perfect set of E-inequivalent
elements can contain at most one element of ∼ A.


Corollary 30.8 Every uncountable Σ11 set contains a perfect subset.

e
30 Π11 EQUIVALENCE RELATIONS 118

Perhaps this is not such a farfetched way of proving this result, since one
of the usual proofs looks like a combination of Lemma 30.2 and 30.6.

V.Kanovei has pointed out to me (email, see also Kanovei [48]) that there
is a shorter proof of

If (a, b) is a P × P-generic over V pair of reals a, b ∈ A then a E

6 b.

which avoids Lemma 30.5 and absoluteness:

1. Assume not. Then there exist conditions X, Y ⊆ A such that X ×Y |`
aEb.
2. Thus if (a, b) ∈ X × Y is P × P -generic then aEb.
3. It follows that aEa0 for any two P-generic a, a0 ∈ X. Indeed take b ∈ Y
which is P-generic over V [a, a0 ]. [Or over V [a] ∪ V [a0 ] if you see difficulties
with V [a, a0 ] when the pair (a, a0 ) is not generic over V .] Then both (a, b)
and (a0 , b) are P × P-generic, and use item 2.
4. Similarly bEb0 for any pair of P-generic b, b0 ∈ Y .
5. Therefore aEb for any pair of P-generic a ∈ X and b ∈ Y .
6. Finally aEb for all a ∈ X and b ∈ Y . Indeed otherwise the nonempty
set
Q = {(a, b) ∈ X × Y : a E 6 b}
produces (by Lemma 30.3) a pair (a, b) ∈ Q such that a ∈ X and b ∈ Y are
P-generic. Contradiction with item 5.
31 BOREL METRIC SPACES AND LINES IN THE PLANE 119

31 Borel metric spaces and lines in the plane

We give two applications of Harrington’s technique of using Gandy forcing.
First let us begin by isolating a principal which we call overflow. It is an
easy consequence of the Separation Theorem.

Lemma 31.1 (Overflow) Suppose θ(x1 , x2 , . . . , xn ) is a Π11 formula and A

is a Σ11 set such that

∀x1 , . . . , xn ∈ A θ(x1 , . . . , xn ).

Then there exists a ∆11 set D ⊇ A such that

∀x1 , . . . , xn ∈ D θ(x1 , . . . , xn ).

Proof:
For n = 1 this is just the Separation Theorem 27.5.
For n = 2 define

B = {x : ∀y(y ∈ A → θ(x, y))}.

Then B is Π11 set which contains A. Hence by separation there exists a ∆11
set E with A ⊆ E ⊆ B. Now define

C = {x : ∀y(y ∈ E → θ(x, y)}.

Then C is a Π11 set which also contains A. By applying separation again we

get a ∆11 set F with A ⊆ F ⊆ C. Letting D = E ∩ F does the job. The
proof for n > 2 is similar.

We say that (B, δ) is a Borel metric space iff B is Borel, δ is a metric
on B, and for every  ∈ Q the set

{(x, y) ∈ B 2 : δ(x, y) ≤ }

is Borel.

Theorem 31.2 (Harrington [39]) If (B, δ) is a Borel metric space, then

either (B, δ) is separable (i.e., contains a countable dense set) or for some
 > 0 there exists a perfect set P ⊆ B such that δ(x, y) >  for every distinct
x, y ∈ P .
Descriptive Set Theory and Forcing 120

Proof:
By relativizing the proof to an arbitrary parameter we may assume that
B and the sets {(x, y) ∈ B 2 : δ(x, y) ≤ } are ∆11 .

Lemma 31.3 For any  ∈ Q+ if A ⊆ B is Σ11 and the diameter of A is less

than , then there exists a ∆11 set D with diameter less than  and A ⊆ D ⊆ B.

Proof:
This follows from Lemma 31.1, since

θ(x, y) iff δ(x, y) <  and x, y ∈ B

is a Π11 formula.

For any  ∈ Q+ look at
[
Q = {D ∈ ∆11 : D ⊆ B and diam(D) < }.

Note that Q is a Π11 set. If for every  ∈ Q+ Q = B, then since there are
only countably many ∆11 sets, (B, δ) is separable and we are done. On the
other hand suppose for some  ∈ Q+ we have that

P = B \ Q 6= ∅.

Lemma 31.4 For every c ∈ V ∩ B

◦
P |` δ(a, č) > /3
◦
where |` is Gandy forcing and a is a name for the generic real (see Lemma
30.2).

Proof:
Suppose not. Then there exists P ≤ P such that

P |` δ(a, c) ≤ /3.

Since P is disjoint from Q by Lemma 31.3 we know that the diameter of P

is ≥ . Let

R = {(a0 , a1 ) : a0 , a1 ∈ P and δ(a0 , a1 ) > (2/3)}.

Descriptive Set Theory and Forcing 121

Then R is in P and by Lemma 30.3, if a is P-generic over V with a ∈ R, then

a0 and a1 are each separately P-generic over V . But a0 ∈ R and a1 ∈ R means
that δ(a0 , c) ≤ /3 and δ(a1 , c) ≤ /3. But by absoluteness δ(a0 , a1 ) > (2/3).
This contradicts the fact that δ must remain a metric by absoluteness.

Using this lemma and Lemma 30.6 is now easy to get a perfect set P ⊆ B
such that δ(x, y) > /3 for each distinct x, y ∈ P . This proves Theorem 31.2.


Theorem 31.5 (van Engelen, Kunen, Miller [20]) For any Σ11 set A in the
plane, either A can be covered by countably many lines or thereeexists a perfect
set P ⊆ A such that no three points of P are collinear.

Proof:
This existence of this proof was pointed out to me by Dougherty, Jackson,
and Kechris. The proof in [20] is more elementary.
By relativizing the proof we may as well assume that A is Σ11 .

Lemma 31.6 Suppose B is a Σ11 set lying on a line in the plane. Then there
exists a ∆11 set D with B ⊆ D such that all points of D are collinear.

Proof:
This follows from Lemma 31.1 since

θ(x, y, z) iff x, y, and z are collinear

is Π11 (even Π01 ).


Define
[
∼ P = {D ⊆ R2 : D ∈ ∆11 and all points of D are collinear}.

It is clear that ∼ P is Π11 and therefore P is Σ11 . If P ∩ A = ∅, then A can

be covered by countably many lines.
So assume that
Q = P ∩ A 6= ∅.
For any two distinct points in the plane, p and q, let line(p, q) be the unique
line on which they lie.
Descriptive Set Theory and Forcing 122

Lemma 31.7 For any two distinct points in the plane, p and q, with p, q ∈ V
◦
Q |`a∈
/ line(p̌, q̌).

Proof:
Suppose for contradiction that there exists R ≤ Q such that
◦
R |`a∈ line(p̌, q̌).

Since R is disjoint from

[
{D ⊆ R2 : D ∈ ∆11 and all points of D are collinear}

it follows from Lemma 31.6 that not all triples of points from R are collinear.
Define the nonempty Σ11 set

S = {a : a0 , a1 , a2 ∈ R and a0 , a1 , a2 are not collinear}

where a = (a0 , a1 , a2 ) via some standard tripling function. Then S ∈ P and

by the obvious generalization of Lemma 30.3 each of the ai is P-generic if
a is. But this is a contradiction since all ai ∈ line(p, q) which makes them
collinear.

The following Lemma is an easy generalization of Lemma 30.6 so we leave
the proof to the reader.

Lemma 31.8 Suppose M is a countable transitive model of ZFC∗ and P is

a partially ordered set in M . Then there exists {Gx : x ∈ 2ω }, a “perfect”
set of P-filters, such that for every x, y, z distinct, we have that (Gx , Gy , Gz )
is P × P × P-generic over M .

Using Lemma 31.7 and 31.8 it is easy to get (just as in the proof of
Theorem 30.1) a perfect set of triply generic points in the plane, hence no
three of which are collinear. This proves Theorem 31.5.

Obvious generalizations of Theorem 31.5 are:

1. Any Σ11 subset of Rn which cannot be covered by countably many lines

contains
e a perfect set all of whose points are collinear.
Descriptive Set Theory and Forcing 123

2. Any Σ11 subset of R2 which cannot be covered by countably many circles

contains
e a perfect set which does not contain four points on the same
circle.

3. Any Σ11 subset of R2 which cannot be covered by countably many

parabolas
e contains a perfect set which does not contain four points on
the same parabola.

4. For any n any Σ11 subset of R2 which cannot be covered by countably

many polynomials
e of degree < n contains a perfect set which does not
contain n + 1 points on the same polynomial of degree < n.

5. Higher dimensional version of the above involving spheres or other sur-

faces.

A very general statement of this type is due to Solecki [102]. Given any
Polish space X, family of closed sets Q in X, and analytic A ⊆ X; either
A can be covered by countably many elements of Q or there exists a Gδ set
B ⊆ A such that B cannot be covered by countably many elements of Q.
Solecki deduces Theorem 31.5 from this.
Another result of this type is known as the Borel-Dilworth Theorem.
It is due to Harrington [39]. It says that if P is a Borel partially ordered set,
then either P is the union of countably many chains or there exist a perfect
set P of pairwise incomparable elements. One of the early Lemmas from [39]
is the following:

Lemma 31.9 Suppose A is a Σ11 chain in a ∆11 poset P. Then there exists
a ∆11 superset D ⊇ A which is a chain.

Proof:
Suppose P = (P, ≤) where P and ≤ are ∆11 . Then

θ(x, y) iff x, y ∈ P and (x ≤ y or y ≤ x)

is Π11 and so the result follows by Lemma 31.1.


For more on Borel linear orders, see Louveau [67]. Louveau [68] is a survey
paper on Borel equivalence relations, linear orders, and partial orders.
Q.Feng [22] has shown that given an open partition of the two element
subsets of ω ω , that either ω ω is the union of countably many 0−homogenous
Descriptive Set Theory and Forcing 124

sets or there exists a perfect 1−homogeneous set. Todorcevic [111] has given
an example showing that this is false for Borel partitions (even replacing
open by closed).
32 Σ11 EQUIVALENCE RELATIONS 125

32 Σ11 equivalence relations

Theorem 32.1 (Burgess [14]) Suppose E is a Σ11 equivalence relation. Then
either E has ≤ ω1 equivalence classes or there exists
e a perfect set of pairwise
E-inequivalent reals.

Proof:
We will need to prove the boundedness theorem for this result. Define

W F = {T ⊆ ω <ω : T is a well-founded tree}.

For α < ω1 define W F<α to the subset of W F of all well-founded trees of

rank < α. W F is a complete Π11 set, i.e., for every B ⊆ ω ω which is Π11 there
exists a continuous map f such that f −1 (W F ) = B (see Theorem e 17.4).
Consequently, W F is not Borel. On the other hand each of the W F<α are
Borel.

Lemma 32.2 For each α < ω1 the set W F<α is Borel.

Proof:
Define for s ∈ ω <ω and α < ω1
s
W F<α = {T ⊆ ω <ω : T is a tree, s ∈ T, rT (s) < α}.
s
The fact that W F<α is Borel is proved by induction on α. The set of trees
0
is Π1 . For λ a limit [
s s
W F<λ = W F<α .
α<λ

For a successor α + 1
s
T ∈ W F<α+1 iff s ∈ T and ∀n (sˆ n ∈ T → T ∈ W F<α
sˆn
).


Another way to prove this is take a tree T of rank α and note that

W F<α = {T̂ : T̂ ≺ T }

and this set is ∆11 and hence Borel by Theorem 26.1.

e
Descriptive Set Theory and Forcing 126

Lemma 32.3 Boundedness Theorem If A ⊆ W F is Σ11 , then there exists

α < ω1 such that A ⊆ W Fα . e

Proof:
Suppose no such α exists. Then

T ∈ W F iff there exists T̂ ∈ A such that T  T̂ .

But this would give a Σ11 definition of W F , contradiction.

 e
There is also a lightface version of the boundedness theorem, i.e., if A is
a Σ11 subset of W F , then there exists a recursive ordinal α < ω1CK such that
A ⊆ W F<α . Otherwise,

{e ∈ ω : e is the code of a recursive well-founded tree }

would be Σ11 .

Now suppose that E is a Σ11 equivalence relation. By the Normal Form

Theorem 17.4 we know there exists
e a continuous mapping (x, y) 7→ Txy such
that Txy is always a tree and

xEy iff Txy ∈

/ WF.

Define
xEα y iff Txy ∈
/ W F<α .
By Lemma 32.2 we know that the binary relation Eα is Borel. Note that Eα
refines Eβ for α > β. Clearly,
\
E= Eα
α<ω1

and for any limit ordinal λ \

Eλ = Eα .
α<λ

While there is no reason to expect that any of the Eα are equivalence

relations, we use the boundedness theorem to show that many are.

Lemma 32.4 For unboundedly many α < ω1 the binary relation Eα is an

equivalence relation.
Descriptive Set Theory and Forcing 127

Proof:
T Note that every Eα must be reflexive, since E is reflexive and E =
α<ω1 Eα .
The following claim will allow us to handle symmetry.

Claim: For every α < ω1 there exists β < ω1 such that for every x, y
6 α x, then x E
if xEα y and y E 6 β y.
Proof:
Let
A = {Txy : xEα y and y E
6 α x}.
6 α x implies y E
Then A is a Borel set. Since y E 6 x and hence x E
6 y, it follows
that A ⊆ W F . By the Boundedness Theorem 32.3 there exists β < ω1 such
that A ⊆ W F<β .

The next claim is to take care of transitivity.

Claim: For every α < ω1 there exists β < ω1 such that for every x, y, z
6 α z, then either x E
if xEα y and yEα z, and x E 6 β y or y E
6 β z.
Proof:
Let
B = {Txy ⊕ Tyz : xEα y, yEα z, and x E
6 α z}.
The operation ⊕ on a pair of trees T0 and T1 is defined by

T0 ⊕ T1 = {(s, t) : s ∈ T0 , t ∈ T1 , and |s| = |t|}.

Note that the rank of T0 ⊕ T1 is the minimum of the rank of T0 and the rank
of T1 . (Define the rank function on T0 ⊕ T1 by taking the minimum of the
rank functions on the two trees.)
The set B is Borel because the relation Eα is. Note also that since
x 6 Eα z implies x 6 Ez and E is an equivalence relation, then either x 6 Ey
6 z. It follows that either Txy ∈ W F or Tyz ∈ W F and so in either case
or y E
Txy ⊕ Tyz ∈ W F and so B ⊆ W F . Again, by the Boundedness Theorem
there is a β < ω1 such that B ⊆ W F<β and this proves the Claim.

Now we use the Claims to prove the Lemma. Using the usual Lowenheim-
Skolem argument we can find arbitrarily large countable ordinals λ such that
Descriptive Set Theory and Forcing 128

for every α < λ there is a β < λ which satisfies both Claims for α. But this
means that T Eλ is an equivalence relation. For suppose xEλ y and y E 6 λ x. Then
since Eλ = α<λ Eα there must be α < λ such that xEα y and y E 6 α x. But by
the Claim there exist β < λ such that x E 6 β y and hence x E 6 λ y, a contradiction.
A similar argument using the second Claim works for transitivity.

Let G be any generic filter over V with the property that it collapses ω1
but not ω2 . For example, Levy forcing with finite partial functions from ω
V [G]
to ω1 (see Kunen [56] or Jech [44]). Then ω1 = ω2V . By absoluteness, E is
still an equivalence relation and for any α if Eα was an equivalence relation
in V , then it still is one in V [G]. Since
\
Eω1V = Eα
α<ω1V

and the intersection of equivalence relations is an equivalence relation, it

follows that the Borel relation Eω1V is an equivalence relation. So now suppose
that E had more than ω2 equivalence classes in V . Let Q be a set of size ω2
in V of pairwise E-inequivalent reals. Then Q has cardinality ω1 in V [G] and
for every x 6= y ∈ Q there exists α < ω1V with x E 6 α y. Hence it must be that
the elements of Q are in different Eω1V equivalence classes. Consequently,
by Silver’s Theorem 30.1 there exists a perfect set P of Eω1V -inequivalent
reals. Since in V [G] the equivalence relation E refines Eω1V , it must be that
the elements of P are pairwise E-inequivalent also. The following is a Σ12
statement: e

V [G] |= ∃P perfect ∀x∀y (x, y ∈ P and x 6= y) → x E

6 y.
Hence, by Shoenfield Absoluteness 20.2, V must think that there is a perfect
set of E-inequivalent reals.
A way to avoid taking a generic extension of the universe is to suppose
Burgess’s Theorem is false. Then let M be the transitive collapse of an
elementary substructure of some sufficiently large Vκ (at least large enough
to know about absoluteness and Silver’s Theorem). Let M [G] be obtained as
in the above proof by Levy collapsing ω1M . Then we can conclude as above
that M thinks E has a perfect set of inequivalent elements, which contradicts
the assumption that M thought Burgess’s Theorem was false.

By Harrington’s Theorem 25.1 it is consistent to have Π12 sets of arbitrary
e a Π1 set B with
cardinality, e.g it is possible to have c = ω23 and there exists 2
e
Descriptive Set Theory and Forcing 129

|B| = ω17 . Hence, if we define

xEy iff x = y or x, y ∈
/B

then we get Σ12 equivalence relation with exactly ω17 equivalence classes, but
since the continuum
e is ω23 there is no perfect set of E-inequivalent reals.
See Burgess [15] [16] and Hjorth [41] for more results on analytic equiva-
lence relations. For further results concerning projective equivalence relations
see Harrington and Sami [37], Sami [96], Stern [109] [110], Kechris [51], Har-
rington and Shelah [38], Shelah [97], and Harrington, Marker, and Shelah
[39].
33 LOUVEAU’S THEOREM 130

33 Louveau’s Theorem
Let us define codes for Borel sets in our usual way of thinking of them as
trees with basic clopen sets attached to the terminal nodes.
Definitions

1. Define (T, q) is an α-code iff T ⊆ ω <ω is a tree of rank ≤ α and

q : T 0 → B is a map from the terminal nodes, T 0 , of T (i.e. rank zero
nodes) to a nice base, B, for the clopen sets of ω ω , say all sets of the
form [s] for s ∈ ω <ω plus the empty set.

2. Define S s (T, q) and P s (T, q) for s ∈ T by induction on the rank of s as

follows. For s ∈ T 0 define

P s (T, q) = q(s) and S s (T, q) =∼ q(s).

For s ∈ T >0 define

[
P s (T, q) = {S(T, q)sˆm : sˆm ∈ T } and S s (T, q) =∼ P s (T, q).

3. Define
P (T, q) = P hi (T, q) and S(T, q) = S hi (T, q)
the Π0α set and the Σ0α set coded by (T, q), respectively. (S is short for
Sigma and P is short for Pi.)

4. Define C ⊆ ω ω is Π0α (hyp) iff it has an α-code which is hyperarithmetic.

5. ω1CK is the first nonrecursive ordinal.

Theorem 33.1 (Louveau [65]) If A, B ⊆ ω ω are Σ11 sets, α < ω1CK , and
A and B can be separated by Π0α set, then A and B can be separated by a
Π0α (hyp)-set. e

Corollary 33.2 ∆11 ∩ Π0α = Π0α (hyp)

e
Corollary 33.3 (Section Problem) If B ⊆ ω ω × ω ω is Borel and α < ω1
is such that Bx ∈ Σ0α for every x ∈ ω ω , then
e
B ∈ Σ0α ({D × C : D ∈ Borel(ω ω ) and Cis clopen}).
e
Descriptive Set Theory and Forcing 131

Note that the converse is trivial.

This result was proved by Dellecherie for α = 1 who conjectured it in gen-
eral. Saint-Raymond proved it for α = 2 and Louveau and Saint-Raymond
independently proved it for α = 3 and then Louveau proved it in general. In
their paper [66] Louveau and Saint-Raymond give a different proof of it. We
will need the following lemma.

Lemma 33.4 For α < ω1CK the following sets are ∆11 :
{y : y is a β-code for some β < α},
{(x, y) : y is a β-code for some β < α and x ∈ P (T, q)} , and
{(x, y) : y is a β-code for some β < α and x ∈ S(T, q)}.

Proof:
For the first set it is enough to see that W F<α the set of trees of rank < α
is ∆11 . Let T̂ be a recursive tree of rank α. Then T ∈ W F<α iff T ≺ T̂ shows
that W F<α is Σ11 . But since T̂ is well-founded T ≺ T̂ iff ¬(T̂  T ) and so
it is Π11 . For the second set just use an argument similar to Theorem 27.3.
The third set is just the complement of the second one.

Now we prove Corollary 33.3 by induction on α. By relativizing the proof
to a parameter we may assume α < ω1CK and that B is ∆11 . By taking
complements we may assume that the result holds for Π0β for all β < α.
Define

R(x, (T, q)) iff (T, q) ∈ ∆11 (x), (T, q) is an α-code, and P (T, q) = Bx .

where P (T, q) is the Π0α set coded by (T, q). Note that by the relativized
version of Louveau’s Theorem for every x there exists a (T, q) such that
R(x, (T, q)). By Π11 -uniformization (Theorem 22.1) there exist a Π11 set R̂ ⊆ R
such that for every x there exists a unique (T, q) such that R̂(x, (T, q)). Fix
β < α and n < ω and define
Bβ,n (x, z) iff there exists (T, q) ∈ ∆11 (x) such that

1. R̂(x, (T, q)),

2. rankT (hni) = β and

3. z ∈ P hni (T, q).

Descriptive Set Theory and Forcing 132

Since quantification over ∆11 (x) preserves Π11 (Theorem 29.3), R̂ is Π11 , and the
rest is ∆11 by Lemma 33.4, we see that Bβ,n is Π11 . But note that ¬Bβ,n (x, z)
iff there exists (T, q) ∈ ∆11 (x) such that

1. R̂(x, (T, q)),

2. rankT (hni) 6= β, or

3. z ∈ S hni (T, q).

and consequently, ∼ Bβ,n is Π11 and therefore Bβ,n is ∆11 . Note that every
cross section of Bβ,n is a Π0β set and so by induction (in case α > 1)
e
Bβ,n ∈ Πα ({D × C : D ∈ Borel(ω ω ) and Cis clopen}).
0
e
But then [
B= Bβ,n
n<ω,β<α

and so
B ∈ Σ0α ({D × C : D ∈ Borel(ω ω ) and Cis clopen}).
e
Now to do the case for α = 1, define for every n ∈ ω and s ∈ ω <ω
Bs,n (x, z) iff there exists (T, q) ∈ ∆11 (x) such that

1. R̂(x, (T, q)),

2. rankT (hni) = 0,

3. q(hni) = s, and

4. z ∈ [s].

As in the other case Bs,n is ∆11 . Let z0 ∈ [s] be arbitrary, then define the
Borel set Cs,n = {x : (x, z0 ) ∈ Bs,n }. Then Bs,n = Cs,n × [s] where But now
[
B= Bs,n
n<ω,s∈ω <ω

and so
B ∈ Σ01 ({D × C : D ∈ Borel(ω ω ) and C clopen}).
e
Descriptive Set Theory and Forcing 133


Note that for every α < ω1 there exists a Π11 set U which is universal for
all ∆0α sets, i.e., every cross section of U is ∆e0 0
α and every ∆α set occurs as a
cross section of U . To see this, let V be a Πα set which is universal for Π0α
e e 0 e
sets. Now put e

(x, y) ∈ U iff y ∈ Vx0 and ∀z(z ∈ Vx0 iff z ∈

/ Vx1 )
where x = (x0 , x1 ) is some standard pairing function. Note also that the
complement of U is also universal for all ∆0α sets, so there is a Σ11 which is
universal for all ∆0α sets. Louveau’s Theorem
e implies that theree can be no
e for all ∆0 sets.
Borel set universal α
e
Corollary 33.5 There can be no Borel set universal for all ∆0α sets.
e
In order to prove this corollary we will need the following lemmas. A
space is Polish iff it is a separable complete metric space.

Lemma 33.6 If X is a 0-dimensional Polish space, then there exists a closed

set Y ⊆ ω ω such that X and Y are homeomorphic.

Proof:
Build a tree hCs : s ∈ T i of nonempty clopen sets indexed by a tree
T ⊆ ω <ω such that
1. Chi = X,
2. the diameter of Cs is less that 1/|s| for s 6= hi, and
3. for each s ∈ T the clopen set Cs is the disjoint union of the clopen sets
{Csˆn : sˆn ∈ T }.

If Y = [T ] (the infinite branch of T ), then X and Y are homeomorphic.


I am not sure who proved this first. I think the argument for the next
lemma comes from a theorem about Hausdorff that lifts the difference hier-
archy on the ∆02 -sets to the ∆0α -sets. This presentation is taken from Kechris
13
[54] mutatis mutandis.
e e
13
Latin for plagiarized.
Descriptive Set Theory and Forcing 134

Lemma 33.7 For any sequence hBn : n ∈ ωi of Borel subsets of ω ω there

exists 0-dimensional Polish topology, τ , which contains the standard topology
and each Bn is a clopen set in τ .

Proof:
This will follow easily from the next two claims.

Claim: Suppose (X, τ ) is a 0-dimensional Polish space and F ⊆ X is closed,

then there exists a 0-dimensional Polish topology σ ⊇ τ such that F is clopen
in (X, σ). (In fact, τ ∪ {F } is a subbase for σ.)
Proof:
Let X0 be F with the subspace topology given by τ and X1 be ∼ F with
the subspace topology. Since X0 is closed in X the complete metric on X is
complete when restricted to X0 . Since ∼ F is open there is another metric
which is complete on X1 . This is a special case of Alexandroff’s Theorem
which says that a Gδ set in a completely metrizable space is completely
metrizable in the subspace topology. In this case the complete metric dˆ on
∼ F would be defined by

ˆ y) = d(x, y) + 1 1
d(x, −
d(x, F ) d(y, F )

where d is a complete metric on X and d(x, F ) is the distance from x to the

closed set F .
Let
(X, σ) = X0 ⊕ X1
be the discrete topological sum, i.e., U is open iff U = U0 ∪U1 where U0 ⊆ X0
is open in X0 and U1 ⊆ X1 is open in X1 .


Claim: If (X, τ ) is a Hausdorff space and (X, τn ) for n ∈ ω are 0-dimensional

Polish topologies extending τ , then there exists a 0-dimensional
S Polish topol-
ogy (X, σ) such that τn ⊆ σ for every n. (In fact n<ω τn is a subbase for
σ.)
Proof:
Consider the 0-dimensional Polish space
Y
(X, τn ).
n∈ω
Descriptive Set Theory and Forcing 135

Q
Let f : X → n∈ω (X, τn ) be the embedding which takes each x ∈ X to the
constant Qsequence x (i.e., f (x) = hxn : n ∈ ωi where xn = x for every n).
Let D ⊆ n∈ω (X, τn ) be the range of f , the set of constant sequences. Note
that f : (X, τ ) → (D, τ ) is a homeomorphism. Let σ be the topology on X
defined by

U ∈ σ iff there exists V open in n∈ω (X, τn ) with U = f −1 (V ).

Q

Q
Since each τn extends τ we get that D is a closed subset ofQ n∈ω (X, τn ).
Consequently, D with the subspace topology inherited from n∈ω (X, τn ) is
Polish. It follows that σ is a Polish topology on X. To see that τn ⊆ σ for
every n let U ∈ τN and define
Y Y
V = X ×U × X.
n<N n>N

Then f −1 (V ) = U and so U ∈ σ.

We prove Lemma 33.7 by induction on the rank of the Borel sets. Note
that by the second Claim it is enough to prove it for
S one Borel set at a time.
So suppose B is a Σ0α subset of (X, τ ). Let B = n∈ω Bn where each Bn is
Π0β for some β < α. e By induction on α there exists a 0-dimensional Polish
topology
e τn extending τ in which each Bn is clopen. Applying the second
Claim gives us a 0-dimensional topology σ extending τ in which each Bn is
clopen and therefore B is open. Apply the first Claim to get a 0-dimensional
Polish topology in which B is clopen.

Proof:
(of Corollary 33.5). The idea of this proof is to reduce it to the case of a
∆α set universal for ∆0α - sets, which is easily seen to be impossible by the
0

standard
e diagonal argument.
e
Suppose B is a Borel set which is universal for all ∆0α sets. Then by the
Corollary 33.3 e

B ∈ ∆0α ({D × C : D ∈ Borel(ω ω ) and C is clopen}).

By Lemma 33.7 there exists a 0-dimensional Polish topology τ such that if

X = (ω ω , τ )
Descriptive Set Theory and Forcing 136

then B is ∆0α (X × ω ω ). Now by Lemma 33.6 there exists a closed set Y ⊆ ω ω

and a homeomorphism
e h : X → Y . Consider

C = {(x, y) ∈ X × X : (x, h(y)) ∈ B}.

The set C is ∆0α in X × X because it is the continuous preimage of the set B

under the map e (x, y) 7→ (x, h(y)). The set C is also universal for ∆0 subsets
α
of X because the set Y is closed. To see this for α > 1 if H ∈ ∆e0α (Y ), then
H ∈ ∆0α (ω ω ), consequently there exists x ∈ X with Bx = H. For e α = 1 just
ω
use that disjoint closed subsets of ω can be separated by clopen sets.
e
Finally, the set C gives a contradiction by the usual diagonal argument:

D = {(x, x) : x ∈
/ C}

would be ∆0α in X but cannot be a cross section of C.

 e

Question 33.8 (Mauldin) Does there exists a Π11 set which is universal for
all Π11 sets which are not Borel?14
e
We could also ask for the complexity of a set which is universal for Σ0α \∆0α
sets. e e

14
This was answered by Greg Hjorth [42], who showed it is independent.
34 PROOF OF LOUVEAU’S THEOREM 137

34 Proof of Louveau’s Theorem

Finally, we arrive at our last section. The following summarizes how I feel
now.

You are walking down the street minding your own business and
someone stops you and asks directions. Where’s xxx hall? You
don’t know and you say you don’t know. Then they point at the
next street and say: Is that xxx street? Well by this time you feel
kind of stupid so you say, yea yea that’s xxx street, even though
you haven’t got the slightest idea whether it is or not. After all,
who wants to admit they don’t know where they are going or
where they are.

For α < ω1CK define D ⊆ ω ω is Σ0α (semihyp) iff there exists S a Π11 set
of hyperarithmetic reals such that every element of S is a β-code for some
β < α and [
D = {P (T, q) : (T, q) ∈ S}.
A set is Π0α (semihyp) iff it is the complement of a Σ0α (semihyp) set. The
Π00 (semihyp) sets are just the usual clopen basis ([s] for s ∈ ω <ω together
with the empty set) and Σ00 (semihyp) sets are their complements.

Lemma 34.1 Σ0α (semihyp) sets are Π11 and consequently Π0α (semihyp) sets
are Σ11 .

Proof: S
x ∈ {P (T, q) : (T, q) ∈ S} iff there exists (T, q) ∈ ∆11 such that (T, q) ∈
S and x ∈ P (T, q). Quantification over ∆11 preserves Π11 ( see Corollary 29.3
) and Lemma 33.4 implies that “x ∈ P (T, q)” is ∆11 .

We will need the following reflection principle in order to prove the Main
Lemma 34.3.
A predicate Φ ⊆ P (ω) is called Π11 on Π11 iff for any Π11 set N ⊆ ω × ω
the set {e : Φ(Ne )} is Π11 (where Ne = {n : (e, n) ∈ N }).

Lemma 34.2 (Harrington [39] Kechris [50]) Π11 -Reflection. Suppose Φ(X)
is Π11 on Π11 and Q is a Π11 set.
If Φ(Q), then there exists a ∆11 set D ⊆ Q such that Φ(D).
Descriptive Set Theory and Forcing 138

Proof:
By the normal form theorem 17.4 there is a recursive mapping e → Te
such that e ∈ Q iff Te is well-founded. Define for e ∈ ω

Ne0 = {ê : Tê  Te }

Ne1 = {ê : ¬(Te ≺ Tê )}

then N 0 is Σ11 and N 1 is Π11 . For e ∈ Q we have Ne0 = Ne1 = De ⊆ Q is ∆11 ;
and for e ∈/ Q we have that Ne1 = Q. If we assume for contradiction that
¬Φ(Ne1 ) for all e ∈ Q, then

/ Q iff φ(Ne1 ).
e∈

But this would mean that Q is ∆11 and this proves the Lemma.

Note that a Π11 predicate need not be Π11 on Π11 since the predicate

Φ(X) = “0 ∈
/ X”

is ∆00 but not Π11 on Π11 . Some examples of Π11 on Π11 predicates Φ(X) are

Φ(X) iff ∀x ∈
/ X θ(x)

or
Φ(X) iff ∀x, y ∈
/ X θ(x, y)
where θ is a Π11 sentence.

Lemma 34.3 Suppose A is Σ11 and A ⊆ B ∈ Σ0α (semihyp), then there exists
C ∈ Σ0α (hyp) with A ⊆ C ⊆ B.

Proof:
Let B = {P (T, q) : (T, q) ∈ S} where S is a Π11 set of hyperarithmetic
S
< α-codes. Let Ŝ ⊆ ω be the Π11 set of ∆11 -codes for elements of S, i.e.

e ∈ Ŝ iff e is a ∆11 -code for (Te , qe ) and (Te , qe ) ∈ S.

Now define the predicate Φ(X) for X ⊆ ω as follows:

S
Φ(X) iff X ⊆ Ŝ and A ⊆ e∈X P (Te , qe ).
Descriptive Set Theory and Forcing 139

The predicate Φ(X) is Π11 on Π11 and Φ(Ŝ). Therefore by reflection

(Lemma 34.2) there exists a ∆11 set D ⊆ Ŝ such that Φ(D). Define (T, q) by

T = {eˆs : e ∈ D and s ∈ Te } q(eˆs) = qe (s) for e ∈ D and s ∈ Te0 .

Since D is ∆11 it is easy to check that (T, q) is ∆11 and hence hyperarithmetic.
Since Φ(D) holds it follows that C = S(T, q) the Σ0α (hyp) set coded by (T, q)
has the property that A ⊆ C and since D ⊆ Ŝ it follows that C ⊆ B.

Define for α < ω1CK the α-topology by taking for basic open sets the
family [
{Π0β (semihyp) : β < α}.
As usual, clα (A) denotes the closure of the set A in the α-topology.
The 1-topology is just the standard topology on ω ω . The α-topology has
its basis certain special Σ11 sets so it is intermediate between the standard
topology and the Gandy topology corresponding to Gandy forcing.

Lemma 34.4 If A is Σ11 , then clα (A) is Π0α (semihyp).

Proof:
Since the Σ0β (semihyp) sets for β < α form a basis for the α-closed sets,
\
clα (A) = {X ⊇ A : ∃β < α X ∈ Σ0β (semihyp)}.

By Lemma 34.3 this same intersection can be written:

\
clα (A) = {X ⊇ A : ∃β < α X ∈ Σ0β (hyp)}.

But now define (T, q) ∈ Q iff (T, q) ∈ ∆11 , (T, q) is a β-code for some β < α,
and A ⊆ S(T, q). Note that Q is a Π11 set and consequently, clα (A) is a
Π0α (semihyp) set, as desired.

Note that it follows from the Lemmas that for A a Σ11 set, clα (A) is a Σ11
set which is a basic open set in the β-topology for any β > α.
Let P be Gandy forcing, i.e., the partial order of all nonempty Σ11 subsets
◦
of ω ω and let a be a name for the real obtained by forcing with P, so that by
Lemma 30.2, for any G which is P-generic, we have that p ∈ G iff aG ∈ p.
Descriptive Set Theory and Forcing 140

Lemma 34.5 For any α < ω1CK , p ∈ P, and C ∈ Π0α (coded in V ) if

e
◦
p |`a∈ C,
then
◦
clα (p) |`a∈ C.
Proof:
This is proved by induction on α.
For α = 1 recall that the α-topology is the standard topology and C
◦
is a standard closed set. If p |`a∈ C, then it better be that p ⊆ C, else
there exists s ∈ ω <ω with q = p ∩ [s] nonempty and [s] ∩ C = ∅. But then
◦
q ≤ p and q |`a∈ / C. Hence p ⊆ C and since C is closed, cl(p) ⊆ C. Since
cl(p) |` a ∈ cl(p), it follows that cl(p) |` a ∈ C.
For α > 1 let \
C= ∼ Cn
n<ω

where each Cn is Π0β for some β < α. Suppose for contradiction that
e
◦
clα (p) 6|`a∈ C
Then for some n < ω and r ≤ clα (p) it must be that
◦
r |`a∈ Cn .
Suppose that Cn is Π0β for some β < α. Then by induction
e
◦
clβ (r) |`a∈ Cn .
But clβ (r) is a Π0β (semihyp) set by Lemma 34.4 and hence a basic open
set in the α-topology. Note that since they force contradictory information
◦ ◦
(clβ (r) |`a∈/ C and p |`a∈ C) it must be that clβ (r) ∩ p = ∅, (otherwise the
two conditions would be compatible in P). But since clβ (r) is α-open this
means that
clβ (r) ∩ clα (p) = ∅
which contradicts the fact that r ≤ clα (p).

Now we are ready to prove Louveau’s Theorem 33.1. Suppose A and B
are Σ11 sets and C is a Π0α set with A ⊆ C and C ∩ B = ∅. Since A ⊆ C it
follows that e
◦
A |`a∈ C.
Descriptive Set Theory and Forcing 141

By Lemma 34.5 it follows that

◦
clα (A) |`a∈ C.

Now it must be that clα (A) ∩ B = ∅, otherwise letting p = clα (A) ∩ B would
be a condition of P such that
◦
p |`a∈ C

and
◦
p |`a∈ B
which would imply that B ∩ C 6= ∅ in the generic extension. But by ab-
soluteness B and C must remain disjoint. So clα (A) is a Πα (semihyp)-set
(Lemma 34.4) which is disjoint from the set B and thus by applying Lemma
34.3 to its complement there exists a Π0α (hyp)-set C with clα (A) ⊆ C and
C ∩ B = ∅.

The argument presented here is partially from Harrington [34], but con-
tains even more simplification brought about by using forcing and abso-
luteness. Louveau’s Theorem is also proved in Sacks [95], Mansfield and
Weitkamp [73] and Kanovei [48]. For a generalization to higher levels of the
projective hierarchy using determinacy, see Hjorth [43].

Elephant Sandwiches

A man walks by a restaurant. Splashed all over are signs saying “Order
any sandwich”, “Just ask us, we have it”, and “All kinds of sandwiches”.
Intrigued, he walks in and says to the proprietor, “I would like an elephant
sandwich.”
The proprietor responds “Sorry, but you can’t have an elephant sand-
wich.”
“What do you mean?” says the man, “All your signs say to order any
sandwich. And here the first thing I ask for, you don’t have.”
Says the proprietor “Oh we have elephant. Its just that here it is 5pm
already and I just don’t want to start another elephant.”
REFERENCES 142

References
[1] J.W.Addison, Separation principles in the hierarchies of classical and
effective descriptive set theory, Fundamenta Mathematicae, 46(1959),
123-135. Cited on page: 9

[2] J.W.Addison, Some consequences of the axiom of constructibility, Fun-

damenta Mathematicae, 46(1957), 337-357. Cited on page: 82

[3] R.Baire, Sur la théorie des ensembles, Comptes Rendus Academie Sci-
ence Paris, 129(1899), 946-949. Cited on page: 3

[4] R.Baire, Sur la représentation des fonctions discontinues (2me partie),

Acta Mathematica, 32(1909), 97-176. Cited on page: 1

[5] Z.Balogh, There is a Q-set space in ZFC, Proceedings of the American

Mathematical Society, 113(1991), 557-561. Cited on page: 17

[6] T.Bartoszynski, Combinatorial aspects of measure and category, Fun-

damenta Mathematicae, 127(1987), 225-239. Cited on page: 77

[7] T.Bartoszynski, On covering of real line by null sets, Pacific Journal of

Mathematics, 131(1988),1-12. Cited on page: 75

[8] T.Bartoszynski, H.Judah, S.Shelah, The cofinality of cardinal invariants

related to measure and category, Journal of Symbolic Logic, 54(1989),
719-726. Cited on page: 75

[9] T.Bartoszynski, H.Judah, On the cofinality of the smallest covering of

the real line by meager sets, Journal of Symbolic Logic, 54(1989), 828-
832. Cited on page: 75

[10] J.Barwise, Admissible sets and structures, Springer-Verlag, 1975.

Cited on page: 106

[11] M.Bell, On the combinatorial principal P(c), Fundamenta Mathemati-

cae, 114(1981), 149-157. Cited on page: 17

[12] R.H.Bing, W.W.Bledsoe, R.D.Mauldin, Sets generated by rectangles,

Pacific J. Math., 51(1974, 27-36. Cited on page: 10
REFERENCES 143

[13] C.Brezinski, History of Continued Fractions and Padé Approxi-

mants, Springer-Verlag, 1991. Cited on page: 3
[14] J.Burgess, Infinitary languages and descriptive set theory, PhD Univer-
sity of California, Berkeley, 1974. Cited on page: 125
[15] J.Burgess, Equivalences generated by families of Borel sets, Proceed-
ings of the American Mathematical Society, 69(1978), 323-326. Cited on
page: 129
[16] J.Burgess, Effective enumeration of classes in a Σ11 equivalence relation,
Indiana University Mathematics Journal, 28(1979), 353-364. Cited on
page: 129
[17] C.C.Chang, H.J.Keisler, Model Theory, North Holland, 1973. Cited
on page: 0
[18] F.van Engelen, A.Miller, J.Steel, Rigid Borel sets and better quasiorder
theory, Contemporary Mathematics (American Mathematical Society),
65(1987), 199-222. Cited on page: 112
[19] L.Euler, De fractionibus continuis dissertatio, Comm. Acad. Sc. Imp.
St. Pétersbourg, 9(1737), 98-137. Cited on page: 3
[20] F.van Engelen, K.Kunen, A.Miller, Two remarks about analytic sets,
Lecture Notes in Mathematics, Springer-Verlag, 1401(1989), 68-72.
Cited on page: 121
[21] P.Erdos, K.Kunen, R.Mauldin, Some additive properties of sets of real
numbers, Fundamenta Mathematicae, 113(1981), 187-199. Cited on
page: 112
[22] Q.Feng, Homogeneity for open partitions of pairs of reals, Transactions
of the American Mathematical Society, 339(1993), 659-684. Cited on
page: 123
[23] W.Fleissner, A.Miller, On Q-sets, Proceedings of the American Mathe-
matical Society, 78(1980), 280-284. Cited on page: 17
[24] W.Fleissner, Current research on Q sets, in Topology, Vol. I (Proc.
Fourth Colloq., Budapest, 1978), Colloq. Math. Soc. Janos Bolyai,
23(1980), 413-431. Cited on page: 17
REFERENCES 144

[25] W.Fleissner, Squares of Q sets, Fundamenta Mathematicae, 118(1983),

223-231. Cited on page: 17

[26] D.H.Fremlin, On the Baire order problem, Note of 29.4.82. Cited on

page: 51

[27] D.H.Fremlin, Consequences of Martin’s Axiom, Cambridge Univer-

sity Press, 1984. Cited on page: 17 77

[28] H.Friedman, PCA well-orderings of the line, Journal of Symbolic Logic,

39(1974), 79-80. Cited on page: 97

[29] S.Friedman, Steel Forcing and Barwise compactness, Annals of Mathe-

matical Logic, 22(1982), 31-46. Cited on page: 27

[30] F.Galvin, A.Miller, γ-sets and other singular sets of real numbers, Topol-
ogy and Its Applications, 17(1984), 145-155. Cited on page: 17 52

[31] R.O.Gandy, Proof of Mostowski’s conjecture, Bulletin de L’Academie

Polonaise des Science, 8 (1960), 571-575. Cited on page: 109

[32] K.Gödel, The consistency of the axiom of choice and of the generalized
continuum hypothesis, Proceedings of the National Academy of Science,
24(1938), 556-557. Cited on page: 82

[33] L.Harrington, A powerless proof of a Theorem of Silver, handwritten

note dated 11-76. Cited on page: 113

[34] L.Harrington, Theorem (A. Louveau), handwritten note undated, circa

1976. Cited on page: 141

[35] L.Harrington, Long projective well-orderings, Annals of Mathematical

Logic, 12(1977), 1-24. Cited on page: 98 101

[36] L.Harrington, Analytic determinacy and 0# , Journal of Symbolic Logic,

43(1978), 685-693. Cited on page: 27

[37] L.Harrington, R.Sami, Equivalence relations, projective and beyond,

Logic Colloquium 78, ed Boffa et al, North Holland 1979, 247-263. Cited
on page: 129
REFERENCES 145

[38] L.Harrington and S.Shelah, Counting equivalence classes for co-κ-

Souslin equivalence relations, in Logic Colloquium 80, North Holland,
1982, 147-152. Cited on page: 113 129
[39] L.Harrington, D.Marker, S.Shelah, Borel orderings, Transactions of the
American Mathematical Society, 310(1988), 293-302. Cited on page: 119
123 129 137
[40] P.Hinman, Recursion-Theoretic Hierarchies, Springer-Verlag, 1978.
Cited on page: 106
[41] G.Hjorth, Thin equivalence relations and effective decompositions, Jour-
nal of Symbolic Logic, 58(1993), 1153-1164. Cited on page: 129
[42] G.Hjorth, Universal co-analytic sets, Proceedings of the American Math-
ematical Society, 124(1996), 3867-3873. Cited on page: 136
[43] G.Hjorth, Two applications of inner model theory to the study of Σ12
sets, Bulletin of Symbolic Logic, 2(1996), 94-107. Cited on page: 141
[44] T.Jech, Set Theory, Academic Press, 1978. Cited on page: 0 16 28 40
128
[45] R.B.Jensen, R.M.Solovay, Some applications of almost disjoint sets, in
Mathematical Logic and Foundations of Set Theory, ed by Y.Bar
Hillel, 1970, North Holland, 84-104. Cited on page: 16
[46] H.Judah, S.Shelah, Q-sets do not necessarily have strong measure zero,
Proceedings of the American Mathematical Society, 102 (1988), 681-683.
Cited on page: 17 58
[47] H.Judah, S.Shelah, Q-sets, Sierpiński sets, and rapid filters, Proceedings
of the American Mathematical Society, 111(1991), 821-832. Cited on
page: 17
[48] V.G.Kanoveı̆, Topologies generated by effectively Suslin sets and their
applications in descriptive set theory, Russian Mathematical Surveys,
51(1996), 385-417. Cited on page: 118 141
[49] M.Kondǒ, Sur l’uniformisation des complémentaires analytiques et les
ensembles projectifs de la seconde classe, Japan J. Math. 15(1939), 197-
230. Cited on page: 90
REFERENCES 146

[50] A. Kechris, Spector second order classes and reflection, in Generalized

recursion theory, II , North Holland 1978, Studies Logic Foundations
Math., 94(1978), 147-183. Cited on page: 137

[51] A.Kechris, On transfinite sequences of projective sets with an appli-

cation to Σ12 equivalence relations, in Logic Colloquium ’77 ed by
Macintyre et al, North Holland, 1978, 155-160. Cited on page: 129

[52] A. Kechris, The perfect set theorem and definable wellorderings of the
continuum, Journal of Symbolic Logic, 43(1978), 630-634. Cited on page:
97

[53] A.S.Kechris, D.A.Martin, Infinite games and effective descriptive set

theory, in Analytic sets, ed by C.A.Rogers, Academic Press, 1980.
Cited on page: 113

[54] A.S.Kechris, Classical descriptive set theory, Graduate Texts in

Mathematics, 156, Springer-Verlag, New York, 1995. Cited on page: 0
133

[55] S.Kleene, Hierarchies of number-theoretic predicates, Bulletin of the

American Mathematical Society, 63(1955), 193-213. Cited on page: 104

[56] K.Kunen, Set Theory , North Holland, 1980. Cited on page: 0 16 19

39 40 128

[57] K. Kunen, A.Miller, Borel and projective sets from the point of view of
compact sets, Mathematical Proceedings of the Cambridge Philosophical
Society, 94(1983), 399-409. Cited on page: 56

[58] K.Kunen, Where MA first fails, Journal of Symbolic Logic, 53(1988),

429-433. Cited on page: 77

[59] K.Kuratowski, Topology, vol 1, Academic Press, 1966. Cited on page:

0 78

[60] K.Kuratowski, A.Mostowski, Set theory, with an introduction to

descriptive set theory, North Holland, 1976. Cited on page: 0 94

[61] A.Landver, Singular σ-dense trees, Journal of Symbolic Logic, 57(1992),

1403-1416. Cited on page: 77
REFERENCES 147

[62] A.Landver, Baire numbers, uncountable Cohen sets and perfect-set forc-
ing, Journal of Symbolic Logic, 57(1992), 1086-1107. Cited on page: 77
[63] H.Lebesgue, Sur les fonctions représentables analytiquement, Journal de
Mathématiques Pures et Appliqués, 1(1905), 139-216. Cited on page: 7
9
[64] A. Louveau, Relations d’equivalence dans les espaces polonais, Collec-
tion: General logic seminar, (Paris, 1982-83), 113-122, Publ. Math.
Univ. Paris VII. Cited on page: 113
[65] A.Louveau, A separation theorem for Σ11 sets, Transactions of the Amer-
ican Mathematical Society, 260(1980), 363-378. Cited on page: 130
[66] A.Louveau, J.Saint Raymond, Borel classes and closed games: Wadge-
type and Hurewicz-type results, Transactions of the American Mathe-
matical Society, 304(1987), 431-467. Cited on page: 131
[67] A. Louveau, Two results on Borel orders, Journal of Symbolic Logic,
54(1989), 865-874. Cited on page: 123
[68] A.Louveau, Classifying Borel Structures, in Set Theory of the Con-
tinuum, ed by Judah et al., Springer-Verlag, 1992, 103-112. Cited on
page: 123
[69] N.Luzin, Sur un problème de M. Baire, Comptes Rendus Hebdomad-
daires Seances Academie Science Paris, 158(1914), 1258-1261. Cited on
page: 35
[70] P.Mahlo, Über Teilmengen des Kontinuums von dessen Machtigkeit,
Sitzungsberichte der Sachsischen Akademie der Wissenschaften zu Leip-
siz, Mathematisch Naturwissenschaftliche Klasse 65, (1913), 283-315.
Cited on page: 35 36
[71] R.Mansfield, The non-existence of Σ12 well-ordering of the Baire space,
Fundamenta Mathematicae, 86(1975), 279-282. Cited on page: 97
[72] R.Mansfield, Perfect subsets of definable sets of real numbers, Pacific
Journal of Mathematics, 35(1970), 451-457. Cited on page: 88
[73] R.Mansfield, G.Weitkamp, Recursive Aspects of Descriptive Set
Theory, Oxford University Press (1985). Cited on page: 113 141
REFERENCES 148

[74] D.A.Martin, R.M.Solovay, Internal Cohen extensions, Annals of Math-

ematical Logic, 2(1970), 143-178. Cited on page: 16 38 95
[75] A.Miller, On the length of Borel hierarchies, Annals of Mathematical
Logic, 16(1979), 233-267. Cited on page: 13 19 29 34 36 37 49 50 72
[76] A.Miller, On generating the category algebra and the Baire order prob-
lem, Bulletin de L’Academie Polonaise des Science, 27(1979), 751-755.
Cited on page: 43
[77] A.Miller, Generic Souslin sets, Pacific Journal of Mathematics, 97(1981),
171-181. Cited on page: 27
[78] A.Miller, Some properties of measure and category, Transactions of the
American Mathematical Society, 266(1981), 93-114. Corrections and ad-
ditions to some properties of measure and category, Transactions of the
American Mathematical Society, 271(1982), 347-348. Cited on page: 72
[79] A.Miller, The Baire category theorem and cardinals of countable cofi-
nality, Journal of Symbolic Logic, 47(1982), 275-288. Cited on page: 73
75 76
[80] A.Miller, A characterization of the least cardinal for which the Baire cat-
egory theorem fails, Proceedings of the American Mathematical Society,
86(1982), 498-502. Cited on page: 77
[81] A.Miller, Mapping a set of reals onto the reals, Journal of Symbolic
Logic, 48(1983), 575-584. Cited on page: 58 72
[82] A.Miller, The Borel classification of the isomorphism class of a countable
model, Notre Dame Journal of Formal Logic, 24(1983), 22-34. Cited on
page: 9
[83] A.Miller, Special subsets of the real line, in Handbook of set-
theoretic topology, North Holland, 1984, 201-233. Cited on page:
17
[84] A.Miller, K. Prikry, When the continuum has cofinality ω1 , Pacific Jour-
nal of Mathematics, 115(1984), 399-407. Cited on page: 42
[85] A.Miller, Infinite combinatorics and definability, Annals of Pure and
Applied Mathematical Logic, 41(1989), 179-203. Cited on page: 112
REFERENCES 149

[86] A.Miller, Projective subsets of separable metric spaces, Annals of Pure

and Applied Logic, 50(1990), 53-69. Cited on page: 12

[87] A.Miller, Special sets of reals, in Set Theory of the Reals, ed Haim
Judah, Israel Mathematical Conference Proceedings, vol 6(1993), 415-
432, American Math Society. Cited on page: 12 17

[88] Y.Moschovakis, Uniformization in a playful universe, Bulletin of the

American Mathematical Society, 77(1971), 731-736. Cited on page: 94

[89] Y.Moschovakis, Descriptive set theory, North Holland 1980. Cited

on page: 0 78

[90] J.C.Oxtoby, Measure and category, Springer-Verlag, 1971. Cited on

page: 0 68

[91] G.Poprougenko, Sur un probl/‘eme de M. Mazurkiewicz, Fundamenta

Mathematicae, 15(1930), 284-286. Cited on page: 65

[92] T.Przymusinski, The existence of Q-sets is equivalent to the existence

of strong Q-sets, Proceedings of the American Mathematical Society,
79(1980), 626-628. Cited on page: 17

[93] C.A.Rogers et al, editors, Analytic Sets, Academic Press, 1980. Cited
on page: 78

[94] F.Rothberger, On some problems of Hausdorff and of Sierpiński, Fun-

damenta Mathematicae, 35(1948), 29-46. Cited on page: 16

[95] G.Sacks, Higher Recursion Theory, Springer-Verlag, 1990. Cited on

page: 141

[96] R.Sami, On Σ11 equivalence relations with Borel classes of bounded rank,
Journal of Symbolic Logic, 49(1984), 1273-1283. Cited on page: 129

[97] S. Shelah, On co-κ-Souslin relations, Israel Journal of Mathematics,

47(1984), 139-153. Cited on page: 129

[98] J.Shoenfield, The problem of predicativity, Essays on the foundations of

mathematics, Magnes press, Hebrew University, Jerusalem, 1961, 132-
139. Cited on page: 86 87
REFERENCES 150

[99] W.Sierpiński, Introduction to general topology, University of

Toronto Press, 1934, translated from original Polish version of 1928.
Cited on page: 3

[100] R. Sikorski, Boolean Algebras, Springer-Verlag New York, 1969, 3rd

edition. Cited on page: 33 35

[101] J.Silver, Counting the number of equivalence classes of Borel and co-
analytic equivalence relations, Annals of Mathematical Logic, 18(1980),
1-28. Cited on page: 27 113

[102] S.Solecki, Covering analytic sets by families of closed sets, Covering an-
alytic sets by families of closed sets, Journal of Symbolic Logic, 59(1994),
1022-1031. Cited on page: 123

[103] R.M.Solovay, On the cardinality of Σ12 sets of reals, in Foundations of

Mathematics, Biulof et al eds., Springer-Verlag (1969), 58-73. Cited
on page: 88 95

[104] R.M.Solovay, S.Tennenbaum, Iterated Cohen extensions and Souslin’s

problem, Annals of Mathematics, 94(1971), 201-245. Cited on page: 40

[105] C.Spector, Hyperarithmetical quantifiers, Fundamenta Mathematicae,

48(1960), 313-320. Cited on page: 109

[106] E.Szpilrajn, Sur l’équivence des suites d’ensembles et l’équivence des

fonctions, Fundamenta Mathematicae, 26(1936), 302-326. Cited on page:
13

[107] E.Szpilrajn, The characteristic function of a sequence of sets and some

of its applications, Fundamenta Mathematicae, 31(1938), 207-223. Cited
on page: 13

[108] J.R.Steel, Forcing with tagged trees, Annals of Mathematical Logic,

15(1978), 55-74. Cited on page: 27

[109] J.Stern, Perfect set theorems for analytic and coanalytic equivalence
relations, in Logic Colloquium ’77 ed by Macintyre et al, North Hol-
land, 1978, 277-284. Cited on page: 129
REFERENCES 151

[110] J. Stern, Effective partitions of the real line into Borel sets of bounded
rank, Annals of Mathematical Logic, 18 (1980), 29-60. Cited on page:
129

[111] S.Todorcevic, Two examples of Borel partially ordered sets with the
countable chain condition, Proceedings of the American Mathematical
Society, 112(1991), 1125-1128. Cited on page: 124

[112] S.Williard, General Topology, Addison-Wesley, 1970. Cited on page:

6
Index 152

◦
P
( α<ω1 Pα )∗ Q 45 Σ0α (semihyp) 137
2ω 1 Σ11 equivalence relations 125
Aα 53 Σ11 78
A<α 53 Σ11 (x) 78
F/I 33 Σ12 82
Fσ 5 Σ12 = ΣHC1 84
Gδ 5 Σ1 -formula 84
I-Luzin set 35 α-code 130
L∞ (Pα : α < κ) 26 α-forcing 23
P (T, q) 130 α-topology 139
Qα 49 ∆0α -universal set 133
e0
∆ 4
S(T, q) 130 α
e 0 (F )
∆ 10
T  T̂ 92 α
e+ 30
T ≤n T 0 52 B
T ≺ T̂ 92 Π0α 4
e 0 (F )
Π 10
T0 23 α
e1
Π 78
T >0 23 1
e1
Π 98
Tα 49 2
e0
Σ 4
WF 125 α
e 0 (F )
Σ 10
W F<α 125 α
e1
Σ 78
WO 113 1
e 1 equivalence relation
Σ 129
[A]I 33 2
[T ] 22 x̌e 16
[ω]ω 2 clα (A) 139
[s] 1 FIN(X, ω) 19
∆11 -codes 109 FIN(ℵω , 2) 75
∆12 well-ordering 82 FIN(c+ , 2) 40
∆0 -formulas 84 FIN(κ, 2) 49
Π0α (hyp) 130 Q̂α 60
Π0α (semihyp) 137 κ-Borel 102
Π11 equivalence relations 113 κ-Souslin 102
Π11 singleton 91 κ-Souslin 86
Π11 -Reduction 107 µ 65
◦
Π11 -Reflection 137 Un 16
Π01 78 ω <ω 1
Π11 on Π11 137 ωω 1
Πβ -sentence 26 ω1CK 130
Index 153

ord(B) 29 Kleene Separation Theorem 104

◦
P∗ Q 40 Louveau’s Theorem 130
P(T ) 54 MAκ (ctbl) 73
Pα 43 Mansfield-Solovay Theorem 88
Qα 60 Martin’s Axiom 16
σ-δ-lattice 6 Martin-Solovay Theorem 38
σ-field 33 Mostowski’s Absoluteness
1
80
σ-ideal 33 Normal form for Σ1 80
∼
PB 4 Q-sets 17
α<ω1 Pα 45 Sack’s real model 72
mP 76 Section Problem 130
sˆn 1 Shoenfield Absoluteness 87
s⊂t 92 Sierpiński set 65
x <c y 82 Silver forcing 16
pp 17 Souslin-Luzin Separation 102
Borel(F ) 10 Spector-Gandy Theorem 109
Borel(X) 4 V=L∗ 82
Borel(X)/meager(X) 48 ZFC 81
MAκ 38 accumulation points 60
OR 22 cBa 28
cov(I) 73 characteristic function 13
cov(meager(2ω )) 73 code for a hyperarithmetic set 104
meager(X) 49 collinear points 121
ord(X) 8 direct sum 45
rank(p) 43 field of sets 10
stone(B) 33 hereditarily countable sets 84
|s| 1 hereditary order 59
Aronszajn tree 52 hyperarithmetic subsets 104
Baire space 1 nice α-tree 22
Borel metric space 119 perfect set forcing 72
Borel-Dilworth Theorem 123 perfect set 8
Boundedness Theorem 126 perfect tree 52
Cantor space 1 prewellorderings 91
Fusion 52 prewellordering 107
Gandy forcing 113 rank function 22
Hγ 39 scale property 90
HC 84 second countable 1
separable 1
Index 154

separative 28
splitting node 52
super Luzin set 49
super-I-Luzin 35
switcheroo 25
tree embedding 92
tree 22
two step iteration 39
uniformization property 90
universal for Σ11 sets 79
universal set e 7
well-founded 22