2017 Fall Real Analysis I 1

Suggested Solutions to Exercise 3

Standard notations are in force.

(1) Prove the conclusion of Lebsegue’s dominated convergence theorem still holds
when the condition “{fk } converges to f a.e.” is replaced by the condition “
{fk } converges to f in measure”.
Z
Solution. Suppose on the contrary that |fk − f |dµ does not tend to zero.
By considering the limit supremum of the sequence, we can find a positive
constant M and a subsequence {fnj } such that
Z
|fnj − f |dµ ≥ M

for all j. By Prop 1.17, a subsequence {gk } of {fnj } converges to f a.e.. By

Lebsegue’s dominated convergence theorem, we have
Z
0 = lim |gk − f |dµ ≥ M > 0,
k→∞

contradiction holds.

(2) Find an example in each of the following cases.

(a) A sequence which converges in measure but not at every point.

(b) A sequence which converges pointwisely but not in measure.

(c) A sequence which converges in measure but not in L1 .

Solution.

(a) Take X = [0, 1] and the Lebesgue measure. Basically, let f1 = χ[0,1/2] , f2 =
χ[1/2,1] , f3 = χ[0,1/22 ] , f4 = χ[1/22 ,1/2] , f5 = χ[1/2,3/22 ] , f6 = χ[3/22 ,1] , · · · .
2017 Fall Real Analysis I 2

(b) According to Proposition 1.18, the measure cannot be finite. Take X =

R and the Lebesgue measure. Consider gn = χ[n,n+1] . We have gn (x) →
0 for all x but not in measure.

(c) Take X = R and the Lebesgue measure. Consider hn = n2 χ[0,1/n] . Then

hn → 0 in measure but not in L1 .

(3) Let fn , n ≥ 1, and f be real-valued measurable functions in a finite mea-

sure space. Show that {fn } converges to f in measure if and only if each
subsequence of {fn } has a subsubsequence that converges to f a.e..

Solution. Let µ be the measure in a finite measure space. If {fn } converges

to f in measure, then every subsequence {fnk } also converges to f in measure.
By Prop 1.17, the subsequence {fnk } has a sub-subsequence converging to f
a.e..

Now suppose that each subsequence of {fn } has a sub-subsequence that con-
verges to f a.e.. Assume that fn does not converge to f in measure. By
considering the limit supremum, there are positive ρ , M and subsequence
{fnj } such that, µ({x : |fnj (x) − f (x)| ≥ ρ}) ≥ M , for all j. However, a sub-
sequence {gk } of {fnj } converges to f a.e. and by Prop 1.18, {gk } converges
to f in measure and

0 = lim µ({x : |gk (x) − f (x)| ≥ ρ}) ≥ M > 0,

k→∞

which is impossible. Hence {fn } converges to f in measure.

(4) Let (X, M, µ) be a measure space. Let M

f contain all sets E such that there

exist A, B ∈ M, A ⊂ E ⊂ B, µ(B\A) = 0. Show that M

f is a σ-algebra

containing M and if we set µ

e(E) = µ(A), then (X, M,
f µe) is a complete
measure space.

Solution. We see that M

f contains M by taking E = A = B for any E ∈

M. Suppose Ei ∈ M,
f Bi ⊆ Ei ⊆ Ai where Bi , Ai ∈ M and µ(Ai \Bi ) = 0,
2017 Fall Real Analysis I 3

then
∞
\ ∞
\ ∞
\
Bi ⊆ Ei ⊆ Ai
i=1 i=1 i=1

and
∞
\ ∞
\ ∞
[ ∞
X
µ( Ai \ Bi ) ≤ µ( Ai \Bi ) ≤ µ(Ai \Bi ) = 0.
i=1 i=1 i=1 i=1

∞
\
We have Ei is in M.
f If A ⊇ E ⊇ B, then
i=1

X\A ⊆ X\E ⊆ X\B

and
µ((X\B)\(X\A)) = µ(A\B).

Hence X\E is in M
f and M
f is a σ algebra. We check that µ
e is a measure
on M
f. Obviously µ
e(φ) = 0. Let Ei be mutually disjoint µ
e measurable set,
∃Bi , Ai µ measurable s.t
Ai ⊆ Ei ⊆ Bi

and
µ(Bi \ Ai ) = 0.

∞
[ ∞
[
Using above argument, we have µ( Bi \ Ai ) = 0, And Ai are mutually
i=1 i=1
disjoint,
∞
[ ∞
[ ∞
X ∞
X
µ
e( Ei ) = µ( Ai ) = µ(Ai ) = µ
e(Ei ).
i=1 i=1 i=1 i=1

So µ
e is a measure on M
f.

Finally, we check that µ

e is a complete measure, let E be a µ
e measurable and
null set, for all subset C ⊆ E, we have ∃A, B ∈ M s.t.A ⊆ E ⊆ B and
2017 Fall Real Analysis I 4

µ(A) = µ(B) = 0. Therefore

φ⊆C⊆B

and
µ(B) = 0.

We have C ∈ M.
f

(5) Show that M

f in the previous problem is the σ-algebra generated by M and

all subsets of measure zero sets in M.

Solution. Let M1 be the σ-algebra generated by M and all subsets of

measure zeros sets in M.

By definition, M
f contains all the sets in M and all subsets of measure zero

sets in M. Since M1 is the smallest such σ-algebra, we have M1 ⊂ M.

To prove that M
f ⊂ N , we let E ∈ M.
f Then there exist A, B ∈ M such that

A ⊂ E ⊂ B and µ(B \ A) = 0. Then E \ A ⊂ B \ A is a subset of a measure

zero set. Now E = A ∪ (E \ A) is a union of a set in M and a subset of a
measure zero set. Hence, E ∈ M1 .

(6) Here we consider an application of Caratheodory’s construction. An algebra

A on a set X is a subset of PX that contains the empty set and is closed under
taking complement and finite union. A premeasure µ : A → [0, ∞] is a finitely
additive function which satisfies: µ(φ) = 0 and µ( ∞
S P∞
k=1 Ek ) = k=1 µ(Ek )
S∞
whenever Ek are disjoint and k=1 Ek ∈ A. Show that the premeasure µ can
be extended to a measure on the σ-algebra generated by A. Hint: Define the
outer measure

X [
µ(E) = inf µ(Ek ) : E ⊂ Ek , Ek ∈ A .
k k
2017 Fall Real Analysis I 5

This is called Hahn-Kolmogorov theorem.

Solution. We follow the proof from Terence Tao’s book “Introduction To

Measure Theory”. Define µ as above and obviously µ is an outer measure on
power set of X. By Caratheodory’s construction, we get a measure defined
on a σ- algebra M . We claim that A ⊆ M , let E ∈ A and C ⊆ X such that
µ(C) < ∞, for all ε > 0, there is {Ei }∞
i=1 ⊆ A covering C such that

∞
X
µ(C) + ε > µ(Ei ).
i=1

As {Ei ∩ E}∞ ∞
i=1 and {Ei \ E}i=1 are subset of A and cover C ∩ E and C \ E

respectively, we have,

∞
X
µ(C ∩ E) ≤ µ(Ei ∩ E),
i=1

and
∞
X
µ(C \ E) ≤ µ(Ei \ E).
i=1

Using the fact that µ is a premeasure, µ(Ei ∩ E) + µ(Ei \ E) = µ(Ei ). Sum-

ming over i, we know that E is in M and M contains the σ algebra generated
by A. Now we try to show that the measure induced extends µ, obviously
by definition µ(E) ≤ µ(E). Let {Ei }∞
i=1 ⊆ A covering E. Without affect-

ing the countable union, we may make {Ei }∞ ∞

i=1 disjoint and obtain {Bi }i=1 .

Furthermore, by taking intersection with E, we have

∞
[
Bi ∩ E = E
i=1

and
∞
X ∞
X ∞
X
µ(Ei ) ≥ µ(Bi ) ≥ µ(Bi ∩ E) = µ(E),
i=1 i=1 i=1

where the last equality follows from the condition of µ. Hence µ(E) ≥ µ(E)
2017 Fall Real Analysis I 6

and the measure extends µ.

The following problems are concerned with the Lebesgue measure. Let R =
I1 ×I2 ×· · ·×In , Ij bounded intervals (open, closed or neither), be a rectangle
in Rn . More properties of the Lebesgue measure can be found in Exercise 4.

(7) For a rectangle R in Rn , define its “volume” to be

|R| = (b1 − a1 ) × (b2 − a2 ) × · · · × (bn − an )

where bi , ai are the right and left endpoints of Ij . Show that

N
[
(a) if R = Rk where Rk are almost disjoint, then
k=1

N
X
|R| = |Rk |.
k=1

N
[
(b) If R ⊂ Rk , then
k=1
N
X
|R| ≤ |Rk |.
k=1

Solution.

(a) Take n = 2 for simplicity. Each rectangle is of the form [a, b] × [c, d]. We
order the endpoints of the x-coordinates of all rectangles R1 , · · · , RN into
a1 < a2 < · · · < an and y-coordinates into b1 < b2 < · · · < bm . This division
breaks R into an almost disjoint union of subrectangles Rj,k = [aj , aj+1 ] ×
[bk , bk+1 ]. Note that each Rj,k is contained in exactly one Rl and each Rl is
2017 Fall Real Analysis I 7

an almost disjoint union of subrectangles from this division. We have

|R| = (an − a1 )(bm − b1 )

= (an − an−1 + an−1 − an−2 + · · · + a2 − a1 )(bm − bm−1 + bm−1 + · · · + b2 − b1 )

X
= |Rj,k |
j,k
X X
= |Rj,k |
l Rj,k ⊆Rl
X
= |Rl |,
l

(b) The proof is similar to that of (a), but now we need to subdivide Rj
N
[
and R together. Now, R ⊆ Rj . We order all x-coordinates of Rj , R into
j=1
a1 < a2 < · · · < aN and and y-coordinates into b1 < b2 < · · · < bM . Then R
is the union of parts of Rk,j

X
|R| = |Rj,k |
Rj,k ⊆R
X
≤ |Rj,k |
j,k
X
≤ |Rj |,
j

where the last inequality follows from the fact that each Rk,j is contained in
some Rj .

(8) Let R be the collection of all closed cubes in Rn . A closed cube is of the form
I × · · · × I where I is a closed, bounded interval.

(a) Show that (R, | · |) forms a gauge, and thus it determines a complete
measure Ln on Rn called the Lebesgue measure.

(b) Ln (R) = |R| where R is a cube, closed or open.

2017 Fall Real Analysis I 8

(c) For any set E and x ∈ Rn , Ln (E + x) = Ln (E). Thus the Lebsegue

measure is translational invariant.

Solution.

(a) We clearly have

[
inf |R| = 0 and R = Rn .
R∈R
R∈R

Hence, R forms a gauge.

Define µ by (∞ )
X
µ(A) = inf |Gj | : Gj ∈ R .
j=1

We check that µ is an outer measure.

• Clearly, µ(φ) = 0.
S
• Suppose {Ej : j ∈ N } are given and write E = j∈N Ej . Let ε > 0.
Choose Gjk ∈ R such that

X
µ(Ej ) + 2−j ε > |Gjk |.
k∈N

Then {Gjk : j, k ∈ N} is a countable cover for E. We have

X X X
µ(E) ≤ |Gjk | < µ(Ej ) + 2−j ε = µ(Ej ) + ε.
j,k∈N j∈N j∈N

Taking ε → 0, we have

X
µ(E) ≤ µ(Ej ).
j∈N

Following the Carathéodory’s construction, we obtain a complete mea-

sure.
2017 Fall Real Analysis I 9

∞
[ ∞
X
(b) By (9)(b), it suffices to show that R ⊆ Rj ⇒ |R| ≤ |Rj |. We
replace Rj = [aj , bj ] × [cj , dj ] by Ŕj = (aj − 2εj , bj + 2εj ) × (cj − 2ε , dj + 2ε ).
Since {Ŕj } is an open cover of R and R is compact, there exists a finite
subcover R´j1 , · · · , R´jM . By (9)(b)

M
X X X
|R| ≤ |R´jk | ≤ |Rj,k | ≤ |Rj | + Cε,
k=1 j,k j

C depends on n only. Let ε → 0,

X
|R| ≤ |Rj |,
j

which shows that

∞
X ∞
[
|R| = inf{ |Rj | : R ⊆ Rj , Rj closed cube}.
1 j=1

Therefore,
Ln (R) = |R|

where R is a closed cube.

In order to show it also holds for any cube, it suffices to show Ln (F ) = 0
whenever F is a face of R. First, Ln (F ) ≤ Ln (R) < ∞. Let N be
any number> 1 and x1 = (a1 , 0), · · · , xN = (aN , 0) be distinct point.
Consider F + xj . For small aj , F + xj can be chosen to sit inside R, then
[ [
(F + xj ) ⊂ R. By (c), N Ln (F ) = Ln (F + xj ) = Ln ( (F + xj )) ≤
P

n n Ln (R)
L (R) ⇒ L (F ) ≤ → 0 as N → ∞.
N
(c) Result follows directly from definition.
2017 Fall Real Analysis I 10

(9) This problem is optional. Use Hahn-Kolmogorov theorem to construct the

Lebesgue measure instead of Problems 7 and 8.

Solution. Look up Wiki.