Tel Aviv University, 2013 Measure and category 1

1 Basic ideas

1a From Cantor’s uncountability theorem to Baire

category theorem . . . . . . . . . . . . . . . . . . 1
1b From Cantor’s uncountability theorem to null sets 2
1c Two approaches to small sets and typical objects 2
1d Compact metrizable spaces; sequence spaces . . 3
1e “Almost all” versus “quasi all”: first examples . 5
1f Digits of a typical number . . . . . . . . . . . . . 6
Hints to exercises . . . . . . . . . . . . . . . . . . . . . . . 7
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1a From Cantor’s uncountability theorem to Baire cat-

egory theorem
By the famous Cantor’s uncountability theorem, R is not countable. Here is
one of the proofs. Let a1 , a2 , · · · ∈ R; we need x ∈ R such that ∀n x 6= an .
To this end we first take b1 < c1 such that a1 ∈ / [b1 , c1 ]. Then we take b2 < c2
such that [b2 , c2 ] ⊂ [b1 , c1 ] and a2 ∈/ [b2 , c2 ]. And so on; [b1 , c1 ] ⊃ [b2 , c2 ] ⊃
[b3 , c3 ] ⊃ . . . Their intersection is not empty, and contains no an .
Can we generalize it to some sets A1 , A2 , · · · ⊂ R proving that ∪n An 6= R?
Yes, provided that these sets satisfy the following.
1a1 Definition. A set A ⊂ R is nowhere dense if every nonempty open
interval contains some nonempty open subinterval that does not intersect A.
1a2 Exercise. A set A ⊂ R is nowhere dense if and only if Int(Cl(A)) = ∅.
Prove it. (Here “Int” stands for interior, and “Cl” for closure.)
1a3 Theorem (Baire). If A1 , A2 , · · · ⊂ R are nowhere dense then Int(∪n An ) =
∅.
1a4 Exercise. Prove the theorem.
Equivalently: R \ ∪n An is dense; that is, Cl(R \ ∪n An ) = R.
In particular, ∪n An 6= R.
Clearly, a singleton is nowhere dense; therefore Cantor’s uncountability
theorem follows from Baire category theorem.
Tel Aviv University, 2013 Measure and category 2

1b From Cantor’s uncountability theorem to null sets

Here is another proof of Cantor’s uncountability theorem. Let a1 , a2 , · · · ∈ R;
P x ∈ R such that ∀n x 6= an . To this end we take ε1 , ε2 , · · · > 0 such
we need
that n εn < 1/2 and consider open intervals (an − εn , an + εn ). A finite
number of these intervals cannot cover [0, 1] since their total length is less
than 1. (Take the Riemann integral of the sum of indicators. . . ) By the
Heine-Borel theorem, the infinite sequence of these intervals still does not
cover [0, 1].

1b1 Definition. A set A ⊂ R is a null set if for every ε > 0 there exist
, ε2 , · · · > 0 and a1 , a2 , · · · ∈ R such that A ⊂ ∪n (an − εn , an + εn ) and
ε1P
2 n εn ≤ ε.

1b2 Theorem. If A1 , A2 , · · · ⊂ R are null sets then Int(∪n An ) = ∅.

1b3 Exercise. (a) Prove that ∪n An is also a null set.

(b) Prove the theorem.

1c Two approaches to small sets and typical objects

1c1 Definition. Given a set X, a set N of subsets of X is called
(a) an ideal 1 (on X), if

(A ⊂ B ∧ B ∈ N ) =⇒ A ∈ N ;
A, B ∈ N =⇒ A ∪ B ∈ N ;
∅∈N.

(b) a σ-ideal (on X), if it is an ideal and

A1 , A2 , · · · ∈ N =⇒ ∪n An ∈ N .

An ideal (or σ-ideal) N on X is proper if X ∈

/ N.

Clearly, null sets are a proper σ-ideal on R.

The complement of a null set is called a set of full measure.

1c2 Definition. A set A ⊂ R is meager 2 if A ⊂ ∪n An for some nowhere

dense sets A1 , A2 , · · · ⊂ R.
1
This notion of set theory is different from (but related to) ideals in ring theory, order
theory etc.
2
Or of the first category.
Tel Aviv University, 2013 Measure and category 3

Clearly, meager sets are a proper σ-ideal on R.

The complement of a meager set is called comeager.1
When a property holds off a null set (in other words, on a set of full
measure), one says that it holds almost everywhere or for almost all elements.
Dealing with a probability measure one also says almost sure(ly).
When a property holds off a meager set (in other words, on a comea-
ger set), one says that it holds quasi-everywhere or for quasi all elements.
One also says that this property holds generically, for a generic element,
or for most of elements. Sometimes the word “typical” is used rather than
“generic”.

1d Compact metrizable spaces; sequence spaces

1d1 Definition. (a) A metric space is a pair (X, ρ) of a set X and a metric ρ
on X, that is, a function ρ : X×X → [0, ∞) such that ρ(x, y) = 0 ⇐⇒ x = y,
ρ(x, y) = ρ(y, x), ρ(x, z) ≤ ρ(x, y) + ρ(y, z) for all x, y, z ∈ X.
(b) Let ρ1 , ρ2 be two metrics on X; ρ2 is stronger than ρ1 if
ρ2 (xn , x) → 0 =⇒ ρ1 (xn , x) → 0 for all x, x1 , x2 , · · · ∈ X;2 further, ρ1 , ρ2
are equivalent, if ρ1 (xn , x) → 0 ⇐⇒ ρ2 (xn , x) → 0 for all x, x1 , x2 , · · · ∈ X.
(c) A metrizable space 3 is a pair (X, R) where X is a set and R is an
equivalence class of metrics on X (metrizable topology; metrics of R are
called compatible).
(d) A metrizable space (as well as its metrizable topology) is compact 4 if
every sequence has a convergent subsequence.

Every subset of R is a metric space with the metric ρ(x, y) = |x − y|.

This space is compact if and only if the set is closed and bounded.
The Cantor set C ⊂ [0, 1] may be defined as consisting of all numbers of
the form ∞
X 2x(k)
ϕ(x) =
k=1
3k
for x ∈ {0, 1}∞ , that is x : {1, 2, . . . } → {0, 1}.

1d2 Exercise. (a) ϕ : {0, 1}∞ → C is a bijection;

1
Or residual.
2
However, a Cauchy sequence in (X, ρ2 ) need not be Cauchy in (X, ρ1 ).
3
Equivalently, and usually, a metrizable space is defined as a special case of a topological
space; but here we do not need the notion of general (not just metrizable) topological space.
4
Equivalently (for metrizable spaces), and usually, a compact space is defined by the
Heine-Borel property: every open cover has a finite subcover.
Tel Aviv University, 2013 Measure and category 4

(b) if x, x1 , x2 , · · · ∈ {0, 1}∞ then

 
ϕ(xn ) −−−→ ϕ(x) ⇐⇒ ∀k xn (k) −−−→ x(k) .
n→∞ n→∞

Prove it.
The metric ρ(x, y) = |ϕ(x) − ϕ(y)| is not invariant under permutations
of coordinates on {0, 1}∞ , but its equivalence class R is (see 1d2(b)). Thus,
we have a compact metrizable space {0, 1}∞ , and moreover, the compact
metrizable space {0, 1}S is well-defined for an arbitrary countable set S (ir-
respective of its enumeration). The space {0, 1}S may also be thought of as
the space of all subsets of S.
1d3 Definition. A set A in a metrizable space X is nowhere dense if ev-
ery nonempty open set contains some nonempty open subset that does not
intersect A.
Still, A is nowhere dense if and only if Int(Cl(A)) = ∅.
1d4 Exercise. (a) Prove that nowhere dense sets are an ideal (on a metriz-
able space).
(b) On R, prove that they are not a σ-ideal.
1d5 Exercise. A set A ⊂ {0, 1}∞ is nowhere dense if and only if for all m
and t1 , . . . , tm ∈ {0, 1} there exist n > m and tm+1 , . . . , tn ∈ {0, 1} such that
all sequences that start with t1 , . . . , tn do not belong to A.
Prove it.
1d6 Theorem (Baire). Let X be a compact metrizable space. If A1 , A2 , · · · ⊂
X are nowhere dense then Int(∪n An ) = ∅.
1d7 Exercise. (a) Prove the theorem.
(b) Find an example of a non-compact metrizable space such that the
σ-ideal of meager sets is not proper.
Thus, the proper σ-ideal of meager sets is well-defined on every compact
metrizable space, in particular, on {0, 1}∞ , and we may speak about generic
elements, quasi-everywhere etc. Now, what about null sets? Can we transfer
Lebesgue measure from R to {0, 1}∞ by ϕ−1 ? No, we cannot, since the
Cantor set is itself a null set. But on the other hand, endless coin tossing
should provide a useful probability measure on {0, 1}∞ ; and binary digits
can be thought of as endless coin tossing over Lebesgue measure!
We consider the map ψ : [0, 1) → {0, 1}∞ ,


ψ(u) = b1 (u), b2 (u), . . . ,
Tel Aviv University, 2013 Measure and category 5

where bk (u) are the binary digits of u, that is,

∞
X bk (u)
bk (u) ∈ {0, 1} , = u, lim inf bk (u) = 0 .
k=1
2k k

True, ψ is not a bijection, but do not bother: the countable set {x : lim inf k x(k) =
1} is anyway a null set, and outside it ψ is a bijection,
∞
−1
X x(k)
ψ (x) = .
k=1
2k

We transfer Lebesgue measure to {0, 1}∞ by ψ. That is, a set A ⊂ {0, 1}∞ is
measurable if ψ −1 (A) is Lebesgue measurable, and then µ(A) is equal to the
Lebesgue measure of ψ −1 (A). This probability measure µ is sometimes called
Lebesgue measure on {0, 1}∞ .1 It is invariant under permutations of coordi-
nates on {0, 1}∞ . Thus, we have a probability space {0, 1}∞ , and moreover,
the probability space {0, 1}S is well-defined for an arbitrary countable set S
(irrespective of its enumeration). It gives us the proper σ-ideal of null sets
on such space, and we may speak about almost all elements etc.

1e “Almost all” versus “quasi all”: first examples

1e1 Example. The famous strong law of large numbers (SLLN) states that
n
1X 1
(a) lim x(k) = for almost all x ∈ {0, 1}∞ .
n n 2
k=1

In contrast,
(b)
n n
1X 1X
lim inf x(k) = 0 , lim sup x(k) = 1 for quasi all x ∈ {0, 1}∞ ,
n n k=1 n n k=1

as we will see soon.

1e2 Example. Consider sets
An = x : x(1) = x(n + 1), x(2) = x(n + 2), . . . , x(n) = x(2n) ⊂ {0, 1}∞ .


Clearly, µ(An ) = 2−n , thus

P
n µ(An ) < ∞; by the first Borel-Cantelli
lemma,


(a1) µ lim sup An = 0 .
n
1
It is in fact the Haar measure on the topological group (Z2 )∞ .
Tel Aviv University, 2013 Measure and category 6

In other words, almost every x belongs to An only for finitely many n. Equiv-
alently,1
X
(a2) 1An (x) < ∞ for almost all x ∈ {0, 1}∞
n

(1A being the indicator of A). In contrast,

X
(b) 1An (x) = ∞ for quasi all x ∈ {0, 1}∞ ,
n

as we will see soon. That is, quasi every x belongs to An for infinitely many
n. (Of course, the infinite set of n depends on x.)

1e3 Exercise. Denote by Bn the complement of An , and by Cn the set

Bn ∩ Bn+1 ∩ . . . Prove that
(a) Cn is closed;
(b) Cn is nowhere dense.

Thus, C = ∪n Cn is meager, and its complement ∩n (An ∪ An+1 ∪ . . . ) =

lim supn An is comeager, which proves 1e2(b).

1e4 Exercise. Now consider sets An = {x : x(n) = x(n+1) = · · · = x(n2 ) =

0}. Prove that
(a) the set limPsupn An is comeager;
(b) lim inf n n nk=1 x(k) = 0 for all x ∈ lim supn An .
1

A half of 1e1(b) is thus proved; the other half is similar.

1f Digits of a typical number

We return to the map ψ : [0, 1) → {0, 1}∞ , ψ(u) = b1 (u), b2 (u), . . . where

bk (u) are the binary digits of u. Of course, ψ is discontinuous; and neverthe-

less. . .

1f1 Exercise. Prove that

(a) If A ⊂ {0, 1}∞ is nowhere dense then ψ −1 (A) ⊂ [0, 1) is nowhere
dense.
(b) If A ⊂ {0, 1}∞ is meager then ψ −1 (A) ⊂ [0, 1) is meager.
(c) If A ⊂ {0, 1}∞ is comeager then ψ −1 (A) ⊂ [0, 1) is comeager.
1
The sum of the indicators is integrable, therefore, finite almost everywhere. (This is
the proof of the first Borel-Cantelli lemma.)
Tel Aviv University, 2013 Measure and category 7

1f2 Exercise. Let A ⊂ {0, 1}∞ . Prove or disprove:

(a) If ψ −1 (A) is nowhere dense then A is nowhere dense.
(b) If ψ −1 (A) is meager then A is meager.

1f3 Remark. A map satisfying the equivalent conditions 1f1(b,c) (but not
necessarily (a)) may be called genericity preserving.1 Informally, such map
transforms a generic element of the first space into a generic element of the
second space.

Combining 1f1 with 1e1(b) and 1e2(b) we see that quasi all u ∈ [0, 1)
satisfy
n n
1X 1X
lim inf bk (u) = 0 , lim sup bk (u) = 1 ,
n n k=1 n n k=1
and the relation
b1 (u) = bn+1 (u), . . . , bn (u) = b2n (u)
holds for infinitely many n.
All said about {0, 1}∞ and binary digits generalizes readily to {0, 1, . . . , 9}∞
and decimal digits, as well as any other basis. Given comeager sets Ap ⊂
{0, . . . , p − 1}∞ , we observe for a generic number u ∈ [0, 1) the following
property: for every basis p = 2, 3, . . . the corresponding digits of u are a
sequence that belongs to Ap .

Hints to exercises
1a4: [b1 , c1 ] ⊃ [b2 , c2 ] ⊃ . . .
2 2
1d2: if x(1) = y(1), . . . , x(n) = y(n) then |ϕ(x) − ϕ(y)| ≤ 3n+1 + 3n+2 + ...;
otherwise |ϕ(x) − ϕ(y)| ≥ 32n − 3n+1 2 2
− 3n+2 − ...
1d4: (a) [b1 , c1 ] ⊃ [b2 , c2 ] ⊃ [b3 , c3 ]; (b) the union can be dense.
1d7: (a) similar to 1a4 with balls rather than intervals; (b) try a dense
countable set.
1e3: (b) use 1d5.
1e4: (b) try n ∈ {1, 4, 9, 16, . . . }
1f1: (a) by 1d5 every binary interval [ 2kn , k+1 2n
) contains a binary subinterval
such that. . . (b), (c) follow from (a).
1f2: consider {0, 1}∞ \ ψ([0, 1)).

1
According to Melleray and Tsankov, a continuous map with this property is called
category-preserving; see arXiv:1201.4447, Def. 2.7.
Tel Aviv University, 2013 Measure and category 8

Index
σ-ideal, 2 metric, 3
metric space, 3
almost, 3, 5 metrizable space, 3
metrizable topology, 3
Baire theorem, 1, 4
binary digits, 5 nowhere dense, 1, 4
null set, 2
Cantor set, 3
comeager, 3 proper, 2
compact, 3
compatible metric, 3 quasi, 3, 4

equivalent metrics, 3 stronger metric, 3

full measure, 2 typical, 3

generic, 3, 4 {0, 1}S , 4

genericity preserving, 7 {0, 1}∞ , 3
1A , 6
ideal, 2 bk (·), 5
Cl, 1
Lebesgue measure on {0, 1}∞ , 5 Int, 1
ϕ(·), 3
meager, 2 ψ(·), 4