Lecture notes

Course “Analysis for PhD students”

Uppsala University, Spring 2018
Rostyslav Kozhan

Chapter 1

Measures

1.1 Some prerequisites

I will follow closely the textbook “Real analysis: Modern Techniques and Their Applications” by Gerald
B. Folland (2nd edition). Throughout these notes, [F] will refer to this book, and (occasionally) a number
afterwards will refer to the page.
The following books were also used: “Comprehensive Course in Analysis, Part 1: Real Analysis” by Barry
Simon, “Real Analysis and Probability” by R. M. Dudley, “Real and Complex Analysis” by Walter Rudin,
“Measure Theory” by Paul R. Halmos.
I will assume the knowledge of:
• Calculus, including Riemann integration;
• Basics of set theory;
• Basics of point-set topology;
• Basics of metric spaces, including compactness.
[F, Chapt 0] should be sufficient.

1.2 Introduction
Intuition. We want to define the notion of measure (“size”) for various subsets of a space X. Ideally, every
subset should be assigned a measure, but that turns out to be unrealistic in practice: see the Banach–Tarski
paradox below. Conclusion: “sick” sets should not have a measure.

Example 1.1. Banach–Tarski Paradox : Let n ≥ 3 and let BS and D be anyStwo balls in Rn . There exists an
m m
m ∈ N and disjoint sets {Bj }m m
j=1 , {Dj }j=1 such that B = j=1 Bj , D = j=1 Dj , such that for each j, Bj
n
can be moved to Dj using only rotations and translations in R .

1
1.3 Algebras and σ-algebras
Intuition. From the previous section, we decided that we are able to measure only nice enough sets, which we
will start calling “measurable”. It is natural to demand that complement and unions of two measurable sets
is again measurable. This leads us to the notion of algebra of sets. Since we want to do analysis and take
limits, we will also want the countable union of measurable sets to also be measurable. That is why we will
demand that our measurable sets form in fact a σ-algebra.

Definition 1.2. For X be a non-empty set. A collection A of subsets of X is called an algebra if

(i) ∅ ∈ A;

(ii) If E ∈ A, then its complement E c is also in A;

Sn
(iii) If Ej ∈ A, then j=1 Ej ∈ A for any (finite) n ∈ N.
Tn
Remark 1.3. From (ii) and (iii) it follows that if Ej ∈ A then j=1 Ej is also in A for any (finite) n ∈ N.

Definition 1.4. For X be a non-empty set. A collection M of subsets of X is called an σ-algebra if

(i) ∅ ∈ M;

(ii) If E ∈ M, then its complement E c is also in M;

S∞
(iii) If Ej ∈ M, then j=1 Ej ∈ M.
T∞
Remarks 1.5. (a) From (ii) and (iii) it follows that if Ej ∈ M then j=1 Ej is also in M.
(b) If A, B ∈ M then A \ B ∈ M. Indeed, A \ B = A ∩ B c .
Example 1.6. (a) Denote P(X) = {E : E ⊆ X} to be the set of all subsets of X (“the power-set of X”).
P(X) is an example of a σ-algebra on X. Another trivial example of a σ-algebra is {∅, X}. We will see less
trivial examples below.
Proposition 1.7. (i) Intersection of any family of σ-algebras on X is a σ-algebra.
(ii) If F is any collection of subsets of X, then there exists the smallest σ-algebra σ(F) on X containing F.

Proof. Part (i) follows directly from the definitions.

Part (ii) follows from (i) since σ(F) is clearly the intersection of the family of all possible σ-algebras
containing F (this family is non-empty as P(X) contains F).
Definition 1.8. We call σ(F) in Proposition 1.7 the σ-algebra generated by F.
Qn
Example 1.9. Let X = j=1 Xj , where {Xj }nj=1 is a (finite or countable) collection of sets (n ∈ N ∪ ∞), each
nQ o
n
with a chosen σ-algebra Mj . Then clearly F = j=1 Ej : Ej ∈ Mj is not a σ-algebra on X. Instead we
define the product σ-algebra on X to be the σ-algebra generated by F, denoted by ⊗nj=1 Mj .

2
1.4 Borel σ-algebra
Intuition. We want the majority of our every-day sets to be measurable. Typically this includes all of the
open sets. This leads to the notion of Borel σ-algebra.
Definition 1.10. For any topological space X, let τ be the collection of all open sets in X (the topology). We
call σ(τ ) the Borel σ-algebra on X, denoted by B(X). Elements of B(X) are called the Borel sets of X.
Examples 1.11. (a) Every open or closed set is a Borel set.
(b) From (ii) and (iii) of Definition 1.4, any countable intersection of open sets is a Borel set. Such sets
are called Gδ ’s (δ stands for intersection (“durchschnitt” in German); G somehow stands for open sets).
(c) Similarly, any countable union of closed sets is a Borel set. Such sets are called Fσ ’s (σ stands for
union (“summe” in German); F somehow stands for closed sets).
(d) In R1 , every half-open interval is an Fσ ; is a Gδ ; is a Borel set on R1 .
(e) In R1 , the set of all half-open finite intervals generate B(R1 ). Similarly, B(R1 ) is generated by all open
intervals; by all closed intervals; by all open rays; by all closed rays ([F, Prop 1.2]).
Exercise 1.12. Let n ∈ N. Show that B(Rn ) is the same σ-algebra as ⊗nj=1 B(R1 ) (Note: the equality
Qn
⊗nj=1 B(Xj ) = B( j=1 Xj ) may fail in general).

1.5 Measures
Definition 1.13. Let X be a set and M be any σ-algebra on X. Then we call (X, M) a measurable space.
Sets in M are called measurable sets.

Example 1.14. Any topological space X with its Borel σ-algebra B = B(X) can therefore be regarded as a
measurable space (X, B). In what follows we will mostly restrict ourselves to these measurable spaces.
Definition 1.15. Let X be a set and M be a σ-algebra on X. A (positive) measure on (X, M) is a function
µ : M → [0, ∞] such that

(i) µ(∅) = 0;
(ii) if {Ej }∞
j=1 is a countable collection of disjoint elements of M, then
 
∞
[ ∞
X
µ Ej  = µ(Ej ). (1.5.1)
j=1 j=1

< ∞ (i.e., µ(E) < ∞ for any measurable set E), then we call µ finite.
If µ(X)S
∞
If X = j=1 Ej , where µ(Ej ) < ∞ for all j, then we call µ σ-finite.
Remarks 1.16. (a) A signed measure is a function ν : M → [−∞, ∞) or ν : M → (−∞, ∞] that satisfies
(i) and (ii) above, where the convergence of the right-hand side of (1.5.1) is absolute if the left-hand side is
finite.
(b) A complex measure is a function µ : M → C that satisfies (i) and (ii) above, where the convergence
of the right-hand side of (1.5.1) is absolute. Note that infinite value is not allowed, so a finite positive measure
is a complex measure, but a non-finite positive measure is not a complex measure. Quite annoying.
Definition 1.17. If µ is a (positive) measure on (X, M), then we call (X, M, µ) a measure space.

3
Examples 1.18. (a) Let (X, M) be any measurable space. Let us fix a point x0 ∈ X. For any set E ∈ M,
define µ(E) to be 1 if x0 ∈ E and µ(E) = 0 otherwise. Then µ is a measure called point mass concentrated
at x0 or the Dirac measure at x0 , usually notated by δx0 .
(b) For any set E ⊆ X define µ(E) to be the number of points in E (with ∞ allowed). Such µ is a measure
on (X, P(X)), and is called the counting measure on X. P
(c) More generally, suppose we are given a function f : X → [0, ∞]. Then the formula µ(E) = x∈E f (x)
defines a measure on (X, P(X)). The above two examples are the special cases.
(d) Our “common” Lebesgue measure dx on R1 will take some work to carefully construct: we do this in
Sections 1.8–1.11.
(e) Any measure µ on (X, B(X)) is called a Borel measure.

1.6 Basic properties of measures

Theorem 1.19. Let (X, M, µ) be a measure space. Then:

(i) (Monotonicity) If E, F ∈ M and E ⊆ F then µ(E) ≤ µ(F );

(ii) (Subadditivity) For any (not necessarily disjoint) elements Aj ∈ M,
 
∞
[ ∞
X
µ Aj  ≤ µ(Aj ). (1.6.1)
j=1 j=1

S∞
(iii) (Continuity from below) If Ej ∈ M for each j and E1 ⊆ E2 ⊆ E3 ⊆ . . ., then µ( j=1 Ej ) =
limj→∞ µ(Ej );
T∞
(iv) (Continuity from above) If Ej ∈ M for each j, E1 ⊇ E2 ⊇ E3 ⊇ . . ., and µ(E1 ) < ∞, then µ( j=1 Ej ) =
limj→∞ µ(Ej ).

Remarks 1.20. (a) The condition µ(E1 ) < ∞ in (iv) is required: e.g., take counting measure on R and
Ej = {j, j + 1, j + 2, . . .}.
(b) If µ(E) = 0 and F ⊆ E then µ(F ) = 0 by (i), but only if F ∈ M. This motivates the definition below.
Proof. More details: [F26].
(i) is immediate from F = E ∪ (F \ E) and additive property.
(ii) is simple by removing overlaps: F1 = E1 , Fk = Ek \ ∪k−1
j=1 EPj.
S∞ P∞ n
(iii) follows from µ( j=1 Ej ) = j=1 µ(Ej \ Ej−1 ) = limn→∞ j=1 µ(Ej \ Ej−1 ) = limn→∞ En .
(iv) follows by taking complements: define Fj = E1 \ Ej , apply part (iii) to get µ(∪Fj ) = lim µ(Fj ) =
µ(E1 ) − lim µ(Ej ), then rewrite µ(∪Fj ) = µ(E1 \ ∩Ej ) = µ(E1 ) − µ(∩Ej ).

1.7 Completion
Definition 1.21. Let (X, M, µ) be a measure space.
(i) If E ∈ M and µ(E) = 0 then we call E a null set (or µ-null set).

4
 (ii) If some property of points x ∈ X holds except on a null-set, then we say that the property holds almost
everywhere (or µ-almost everywhere). We abbreviate it as “a.e.”.
(iii) Measure space (X, M, µ) is called complete if every subset of every µ-null set is measurable. It is very
common to just say that µ is complete.

(iv) Given a measure µ on (X, M), let J be the set of all possible subsets of µ-null sets. Define M :=
{E ∪ F : E ∈ M, F ∈ J }. Then we define the completion of µ to be the measure µ on (X, M) defined
by µ(E ∪ F ) = µ(E) for E ∈ M, F ∈ J .
Remarks 1.22. (a) It is easy to see that µ is complete if and only if µ = µ.
(b) It is clear that µ = µ for any measure µ.

1.8 Outer measures

Intuition. Before we go any further into the properties of measures, let us now try to carefully construct our
usual length and area measures – what we will call the Lebesgue measures on R1 and R2 , respectively. Our
construction will actually be quite general and allow to construct a wide selection of other measures.
The idea is as follows. We first define “measure” of some elementary sets (think of length of intervals on
R1 or area of rectangles on R2 , etc.), and then we hope that this will be extended to measures on the whole
σ-algebra, consistently with the property (ii) of Definition 1.15. Now how do we extend this? Through the
notion of the outer measure.
For the intuition, imagine we want to define µ ((a, b)) = b−a andnhope to extend this to the whole Borel σ- o
P∞ S∞
algebra B(R). The “outer measure” of any set E will be µ∗ (A) := inf j=1 µ((aj , bj )) : j=1 (aj , bj ) cover A .
Similarly we can approximate E from inside by elementary sets and taking the supremum of their total length.
If these two “outer” and “inner” numbers coincide, we will call E to be “a nice set”, and we will simply define
this common value to be the measure µ(E) of E. Then we just cross our fingers in hope that all such nice
sets form a σ-algebra and that µ will satisfy the properties of Definition 1.15.

Definition 1.23. Let X be a set and P(X) be its power-set (the set of all subsets). An outer measure on
X is a function µ∗ : P(X) → [0, ∞] such that
(i) µ∗ (∅) = 0;

(ii) µ∗ (A) ≤ µ∗ (B) if A ⊆ B;

S  P
∞ ∞
(iii) µ∗ j=1 j ≤
A ∗
j=1 µ (Aj ).

Remark 1.24. “Outer measure” is not a measure, typically!

Proposition 1.25. Let S be a family of subsets of X such that ∅ ∈ S and X is a countable union of sets in
S. Let µ0 : S → [0, ∞] be any function satisfying µ0 (∅) = 0. Then for any set A ∈ P(X),
 
X∞ ∞
[ 
µ∗ (A) := inf µ0 (Ej ) : Ej ∈ S, A ⊆ Ej
 
j=1 j=1

defines an outer measure on X.

5
Proof. [F29], [S681]
µ∗ ≥ 0 and µ∗ (∅) = 0 are obvious.
Property (ii) follows easily since any cover of B is also a cover of A. P∞
Property (iii): fix ε > 0, and for each a cover {Sj,k }∞
j, choose P k=1 ofPAj such that k=1 µ0 (Sj,k ) − 2εj ≤
∞ ∞
µ∗ (Aj ). Then j=1 Aj is covered by j,k Sj,k with j,k µ0 (Sj,k ) ≤ ε + j=1 µ∗ (Aj ). This implies (iii) since
S S
ε was arbitrary.

Examples 1.26. (a) Taking X = R1 and µ0 ((a, b]) = b − a, and S to be the set of all finite half-intervals, we
obtain an outer measure µ∗ called the Lebesgue outer measure.
(b) More generally, we can take µ((a, b]) = F (b) − F (a) for some non-decreasing, right-continuous function
F . This leads to the outer measure µ∗F which is called the Lebesgue–Stieltjes outer measure. More on this,
see Section 1.11 below.

1.9 µ∗ -measurable sets

Intuition. µ∗ -measurable sets are basically those “nice” sets for which the inner and outer measures coincide.
This condition can be restated as (1.9.1) so that we don’t in fact need to define the inner measure.

Definition 1.27. Given a set X and an outer measure µ∗ , we say that a set A ⊆ X is µ∗ -measurable if

µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) for all E ⊆ X. (1.9.1)

Remarks 1.28. (a) If E is an elementary set containing A, then the outer measure of A is µ∗ (A) = µ∗ (E ∩ A),
and the inner measure of A is µ∗ (E) − µ∗ (E ∩ Ac ), so (1.9.1) is indeed the right condition.
(b) Inequality ≤ in (1.9.1) is automatic by subadditivity (iii) of Definition 1.23, so A is µ∗ -measurable if
and only if the inequality ≥ holds in (1.9.1).

1.10 Carathéodory theorem: construction of a measure from an

outer measure
Theorem 1.29 (Carathéodory). Let µ∗ be an outer measure on X, and M be the collection of all µ∗ -
measurable sets. Then M is a σ-algebra, and the restriction of µ∗ to M is a (complete) measure.
Proof. [F29–30]
(i): ∅ ∈ M is trivial: µ∗ (E) = µ∗ (∅) + µ∗ (E).
(ii): A ∈ M implies Ac ∈ M is immediate from the definition since (1.9.1) is symmetric in A and Ac .
(iii), algebra: applying measurability condition (1.9.1) to A and then B we get µ∗ (E) = µ∗ (E ∩ A ∩ B) +
µ (E ∩A∩B c )+µ∗ (E ∩Ac ∩B)+µ∗ (E ∩Ac ∩B c ) for any set E. Now using A∪B = (Ac ∩B)∪(A∩B c )∪(A∩B)
∗

and subadditivity of µ∗ , we get µ∗ (E ∩(A∪B)) ≤ µ∗ (E ∩A∩B)+µ∗ (E ∩A∩B c )+µ∗ (E ∩Ac ∩B). Combining
them, we get µ∗ (E) ≥ µ∗ (E ∩ (A ∪ B)) + µ∗ (E ∩ Ac ∩ B c ) which means A ∪ B ∈ M by definition of µ∗ -
measurability.
Sn (iii), σ-algebra:Snote that it suffices to take disjoint unions. So let Aj ∈ M be disjoint. Define Bn =
∞ ∗ ∗ ∗ c
A
j=1 j and B = j=1 j . We want to show that for any E, we have µ (E) ≥ µ (E ∩ B) + µ (E ∩ B ).
A

6
 By µ∗ -measurability P
of An , we get that for any E, µ∗ (E ∩ Bn ) = µ∗ (E ∩ An ) + µ∗ (E ∩ Bn−1 ), so by
n
induction µ (E ∩ Bn ) = j=1 µ∗ (E ∩ Aj ). Since Bn ∈ M (finite union), we get
∗

n
X
µ∗ (E) = µ∗ (E ∩ Bn ) + µ∗ (E ∩ Bnc ) ≥ µ∗ (E ∩ Aj ) + µ∗ (E ∩ B c ). (1.10.1)
j=1
S∞
Take n → ∞ and apply subadditivity: we get µ∗ (E) ≥ µ∗ ( j=1 (E∩Aj ))+µ∗ (E∩B c ) = µ∗ (E∩B)+µ∗ (E∩B c ),
∗
which is what we need for µ -measurability of B.
This shows that M is a σ-algebra.
Now let us show that µ∗ restricted to M is a measure.
µ∗ (∅) = 0 is given since µ∗ is an outer measure.
P∞To show countable additivity for disjoint unions, take E = B and n → ∞ in (1.10.1): we get µ∗ (B) ≥
∗
j=1 µ (Aj ). The reverse inequality is immediate from subadditivity of outer measures, so we get the required
equality.
Finally, showing that µ∗ on M is complete is easy: if µ∗ (A) = 0 for some set A, then also µ∗ (E ∩ A) = 0
by monotonicity, so µ∗ -measurability of A is equivalent to µ∗ (E) ≥ µ∗ (E ∩ Ac ) which holds by monotonicity
again.

1.11 Lebesgue–Stieltjes measures on R: construction

Intuition. Given a finite Borel measure µ on R, we can form a function F : R → [0, ∞) given by F (x) =
µ((−∞, x]). Such a function is non-decreasing (by (i) of Theorem 1.19) and right-continuous (by (iv) of The-
orem 1.19). Notice then that µ((a, b]) = F (b) − F (a). Now in the converse direction, we use the Carathéodory
theorem to construct a measure µF corresponding to any non-decreasing, right-continuous function F . In
particular F (x) = x corresponds to the Lebesgue measure on R (“length”).

Theorem 1.30. Given a non-decreasing, right-continuous function F : R → R, there is a unique Borel

measure µF on R such that
µF ((a, b]) = F (b) − F (a).
Any Borel measure on R that is finite on all bounded Borel sets is of the form µF for some non-decreasing,
right-continuous function F .
Remarks 1.31. (a) If F (x) = x, we call µF (the completion of (R, B(R), µF )) the Lebesgue measure on R,
denoted by m.
(b) For a general F , we call µF (the completion of (R, B(R), µF )) the Lebesgue–Stieltjes measure
associated to F .
Exercise 1.32. What will be different in the statement of the theorem if we start off with a left-continuous
function F instead?

Proof. [F31–35]
Let S be the set of all finite half-intervals (a, b]. We can apply proposition 1.25 with µ0 ((a, b]) = F (b)−F (a)
to obtain the outer measure µ∗0 (the Lebesgue–Stieltjes outer measure). Then we apply the Carathéodory
Theorem 1.29 to obtain a complete measure µ on the σ-algebra of µ∗0 -measurable sets. We are left with
showing three things: 1) that µ∗ and µ0 agree on half-intervals; 2) that any Borel set in R is µ∗0 -measurable;
3) uniqueness.

7
 3): suppose µ1 and µ2 are two measures that agree on all intervals (a, b]. Note that the collection of sets
where µ1 and µ2 agree forms a σ-algebra. Therefore µ1 and µ2 agree on the minimal σ-algebra generated by
intervals (a, b], that is they agree on all of the Borel sets.
We show (1) and (2) in steps below:
Step 1: µ0 is countably additive on S.
Proof: finite additiveness (for disjoint unions) is obvious from telescoping summation. Finite sub-additiveness
(for non-disjoint unions) is also clear too by subdividing intervals into disjoint or coinciding intervals. Now
to prove countable additivity,P let I = (a, b] be equal to the countable
P∞ disjoint union of Ij = (aj , bj ]. By finite
n
additivity, for any n, µ0 (I) ≥ j=1 µ0 (Ij ), which gives µ0 (I) ≥ j=1 µ0 (Ij ). To prove the converse, fix ε > 0,
and for every n ≥ 0, use the right-continuity of F to get δn > 0 such that F (bn + δn ) < F (bn ) + 2εn , and
similarly δ > 0 such that F (a + δ) < F (a) + ε. Now the compact interval [a + δ, b] is covered by open intervals
(an , bn + δn ) (n ∈ N), we can choose a finite subcover (with indices Λn ⊂ N). By finite subadditivity, we get
∞
X X ε
µ0 (I) − ε ≤ F (b) − F (a + δ) ≤ (F (bn + δn ) − F (an )) ≤ (F (bn ) − F (an ) + ).
n=1
2n
Λn
P∞
Taking ε → ∞ gives µ0 (I) ≤ n=1 µ0 (In ).
Step 2: µ0 and µ∗ agree on S.
Proof: Let I ∈SS. µ∗ (I) ≤ µ0 (I) is obvious from the definition of µ∗ since we can just cover I with I.
∞
Now suppose I ⊆ j=1 Ij , Ij ∈ S. Define Bn = I ∩ (In \ ∪n−1 Then Bn ’s are disjoint, in S, with union
j=1 Ij ). P
P∞ ∞
equal I, so by countable additivity of µ0 , µ0 (I) = j=1 µ0 (Bn ) ≤ j=1 µ0 (In ). This holds for any cover of
∗
I, so µ0 (I) ≤ µ (I). This proves their equality.
Step 3: Every half-interval A := (a, b] is µ∗ -measurable.
Choose any E ⊆ X P with µ∗ (E) < ∞. We need ≥ in (1.9.1). Given ε > 0, choose countable Ij ∈ S so
∗ c c
that ∪j Ij cover E and j µ0 (Ij ) ≤ µP (E) + ε. Then ∪Pj (Ij ∩ A) cover EP∩ A and ∪j (Ij ∩ A ) cover E ∩ A .
Therefore µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ≤ j µ0 (Ij ∩ A) + j µ0 (Ij ∩ Ac ) = j µ0 (Ij ) by countable additivity of
µ0 on S. Thus we showed µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ≤ µ∗ (E) + ε. Now we take ε → 0.
Step 4: Since half-intervals generate B(R), and the set of µ∗ -measurable sets is a σ-algebra, we obtain that
every Borel set is µ∗ -measurable.
Remark: One can show that Borel sets are always µ∗ -measurable in Theorem 1.29 as long as µ∗ is a
so-called “metric outer measure”, see, e.g., [F, Prop 11.16].

1.12 Regular sets and regular measures

Intuition. Measurable sets are often not easy to understand, and it would be nice to be able to approximate
them with the sets that we understand better: for example open or compact sets.
Definition 1.33. Let X be a topological space, and (X, M, µ) be a measure space.
(i) We say that a set E ∈ M is outer regular if

µ(E) = inf {µ(U ) : E ⊇ E and U is open and measurable}

(ii) We say that a set E ∈ M is inner regular if

µ(E) = sup {µ(K) : K ⊆ E and K is compact and measurable}

(iii) We say that the measure µ is outer regular if every set in M is outer regular.

8
(iv) We say that the measure µ is inner regular if every set in M is inner regular.
(v) We say that the measure µ is regular if it is both outer and inner regular.
Remark 1.34. There are slight variations in terminology in the literature for (inner/outer) regular measures,
so keep this mind when reading books!
Theorem 1.35. Let (X, M, µ) be a measure space with µ σ-finite, and M ⊇ B(X). If µ is outer regular
then:
(i) For any E ∈ M and ε > 0, there exists a closed set C and an open set U such that C ⊆ E ⊆ U and
µ(U \ C) < ε.
(ii) For any E ∈ M, there exists a Gδ set G and an Fσ set F such that F ⊆ E ⊆ G and µ(G \ F ) = 0.
Remarks 1.36. (a) Definitions of Gδ and Fσ are in Section 1.4.
(b) This is closely related to [F, 1.19–1.20], but more general, borrowed from [R, 2.17].
Proof. (i) If µ(E) < ∞, then use outer regularity
S∞ gives an open set U such that U ⊇ E with µ(U \ E) < ε/2.
If µ(E) = ∞, then by σ-finiteness, X = j=1 Xj with µ(Xj ) < ∞, so µ(E ∩ Xj ) < ∞, and by outer
j+1
S∞
regularitySwe can find S Uj ⊇ E ∩ Xj with µ(Uj \ (E ∩ Xj )) < ε/2 . Then let U = j=1 Uj : we get
U \ E = Uj \ E ⊆ Uj \ (E ∩ Xj ), so that µ(U \ E) < ε/2.
Now applying this to E c instead of E, we get a closed C ⊆ E with µ(E \ C) < ε/2, which completes the
proof of (i).
(ii) Apply (i)Swith ε = 1/n to get closedTset Cn and open sets Un with Cn ⊆ E ⊆ Un and µ(Un \Cn ) < 1/n.
Then take F = Cj (an Fσ set) and G = Uj (a Gδ set). Then F ⊆ E ⊆ G and µ(G\F ) ≤ µ(Un \Cn ) < 1/n
for all n.

1.13 Lebesgue–Stieltjes measures on R: regularity

Theorem 1.37. For any F (see Theorem 1.30), the Lebesgue–Stieltjes measure µF is regular.
Remark 1.38. One can extend this to Rn : any Borel measure on Rn that is finite on compacts is necessarily
regular.
Proof. [F36]
Let us use µ instead of µF .
Outer regularity: µ(E) ≤ infSis obvious from inclusion
P E ⊂ U . For the other direction, given ε > 0, by
∞
definition of µ∗ , we can choose j=1 (aj , bj ] ⊇ E with µ((aj , bj ]) ≤Pµ(E) + ε. Then choose δj > 0 with
F (bj +δj )−F (δj ) < 2εj (right-continuity). Then E ⊆ (aj , bj +δj ) with µ((aj , bj +δk )) ≤
S P
µ((aj , bj ])+ε ≤
µ(E) + 2ε.
Inner regularity: Suppose E is bounded. Need to show µ(E) ≤ sup. If E is closed, then it’s obvious.
If not, then by (i), choose an open U ⊇ E \ E such that µ(U ) ≤ µ(E \ E) + ε (note: E \ E is Borel so µ-
measurable). Then we take K = E \U = E \U (draw a pic), compact and ⊆ E, get µ(K) = µ(E)−µ(E ∩U ) =
µ(E) − [µ(U ) − µ(U \ E)] ≥ µ(E) − µ(U ) + µ(E \ E) ≥ µ(E) − ε. Finally, if E is unbounded, divide E into
countable union of disjoint bounded sets, e.g., Ej = E ∩ (j, j + 1], and then approximate each of them by a
compact Kj , then take finite union.
Remark 1.39. Combining with the results of the previous section, we obtain that any Borel set (or more
generally, any Lebesgue set) is an Fσ -set plus a set of µ-measure zero (or a Gδ set minus a set of µ-measure
zero).

9
1.14 Lebesgue measure on R: translations and dilations
Theorem 1.40. Let m be the Lebesgue measure on R and L be the σ-algebra of the Lebesgue measurable sets.
(i) If E ∈ L then for any s ∈ R, {x + s : x ∈ E} =: E + s is also in L and m(E + s) = m(E).
(ii) If E ∈ L then for any r ∈ R, {rx : x ∈ E} =: rE is also in L and m(rE) = |r| m(E).
Remarks 1.41. (a) More generally, the Lebesgue measure mn on Rn (defined, e.g., through volumes of Rn -
parallelepipeds and then extended via Carathéodory theorem; alternatively, it can be defined through the
product of one-dimensional measures – to be discussed later). This measure also behaves well under transla-
tions, dilations, as well as rotations or more generally under any linear transformations, see [F70–76].
(b) It is not hard to see that, up to a normalization, mn is a unique translation invariant Borel measure
on Rn .
Proof. [F37]
Collection of open intervals are translation and dilation invariant, so the same is true for Borel sets. Three
measures m(E), ms (E) := m(E + s), mr (E) = m(rE)/|r| all agree on intervals, and then by σ-additivity,
also on B(R). Finally their completions must also be the same measure.

1.15 Lebesgue measure on R: measurable and non-measurable sets

Examples 1.42. (a) Any point, and therefore any countable set (e.g., Q), is clearly a Lebesgue measurable set
of measure 0.
(b) Does there exist a Lebesgue measurable set that isn’t Borel? Yes (many!). One way to see it to
compare their cardinality (2c and c, respectively), see [F39].
(c) Does there exist a Lebesgue non-measurable set? The classical example is a Vitali set: subdivide R
into disjoint sets of the form r + Q (put it another way, into elements of the quotient group R/Q). Choose
one point in [0, 1] from each of the disjoint sets (using the axiom of choice). Denote this set S∞V . Enumerate
Q ∪ [−1, 1] = {qj }∞
j=1 , and note that V j = V + qj are pairwise disjoint sets satisfying [0, 1] ⊆ j=1 Vj ⊆ [−1, 2].
P∞
If V were Lebesgue measurable, then every Vk has equal measure, and we get 1 ≤ j=1 m(Vk ) ≤ 3 which is
a contradiction.

1.16 The Cantor set

The Cantor set is obtained by repeatedly removing one thirds of the interval from [0, 1]. Namely, first we
remove the set U1,1 = ( 31 , 23 ) (one open interval of length 31 ). Then remove one thirds of the remaining two
intervals: U2,1 = ( 19 , 29 ) and U2,2 = ( 79 , 89 ) (two open intervals of total length 92 ). We continue in the same
fashion: on the n-th step we remove Un,1 = ( 31n , 32n ), . . . , Un,2n−1 = (1 − 32n , 1 − 31n ) (2n−1 disjoint open
n−1
intervals of total length 2 3n ). Define
n−1
n 2[
[
Cn = [0, 1] \ Uk,j
k=1 j=1

and the Cantor set to be

n−1
∞ 2[
[ ∞
\
C = [0, 1] \ Uk,j = Cn .
k=1 j=1 n=1

10
Exercise 1.43. Show that
(i) C is compact, nowhere dense (i.e., its closure has empty interior), totally disconnected (the only con-
nected subset are single points), perfect (non-empty, closed, with every point being an accumulation
point), with no isolated points.

(ii) m(C) = 0.
(iii) card(C) = c.
Equivalently, the Cantor set can also be defined as the set of all points in [0, 1] whose base-three expansion
has digits 0 or 2 only, see [F38].
Let us also define the Cantor function c(x) as follows: c(x) = 21 on U1,1 ; c(x) = 41 on U2,1 , c(x) = 34 on
U2,2 ; and so on: c(x) = 2j−1
2k
on Uk,j . This defines c(x) on [0, 1] \ C (note that this set is dense everywhere
on [0, 1]). Then for x ∈ C we define
∞ ∞
X aj X aj /2
c(x) = c( j
) =
j=1
3 j=1
3j

(each aj is 0 or 2 here by the discussion above).

Exercise 1.44. Show that

(i) c defined above is continuous on [0, 1].

(ii) c is C 1 on [0, 1] \ C with c0 (x) = 0 there.
c(x+ε)−c(x−ε)
(iii) For x ∈ C, limε↓0 2ε = ∞.
Remarks 1.45. (a) Note that c(1) − c(0) = 1, even though c0 (x) = 0 Lebesgue-a.e.!
(b) One can form the so-called generalized Cantor sets by removing not always 1/3’s of the intervals, but
varying amounts. These Cantor sets can have positive Lebesgue measure [F39].

11