National Taiwan University Real Analysis I

Graduate Institute of Mathematics MATH 7201

Midterm
Room 101, October 28, 2024, Monday 10:20 - 12:20

Notes: You can freely use your notes and other course materials while you are taking the midterm,
but electronic devices are not allowed during the exam. Your final grade will be normalized with respect
to the highest score and surely in the interval [0, 100]. There are 6 + 1 problems printed in 2 pages.
You can enjoy writing down all creative solutions within 2 hours. Good luck everyone!

Problem 1 (15 pts): Let q ≥ 2 be an integer. Construct a Cauchy sequence of rational numbers
√ √
that converges to 3 q. The notion 3 q must not appear in your defined sequence!

Problem 2 (15R pts): Suppose (X, A, µ) is a measure space.

R Let h : X → R be an A-measurable
function with |h|dµ < ∞ and define ν : A → R by ν(E) = E hdµ. Prove that
∞
[  ∞
X
ν Ei = ν(Ei ) for every disjoint sequence E1 , E2 , · · · of sets in A.
i=1 i=1

Problem 3 (15R pts): Suppose λ is Lebesgue measure on R and f : R → R is a Borel measurable

function with |f |dλ < ∞. Define g : R → R by
R
Z
g(x) = f dλ.
(−∞,x)

Prove that g is uniformly continuous on R.

Problem 4 (15 pts): a) Let (fn ) and (gn ) be sequences of real-valued continuous functions defined
on the bounded interval [a, b] with −∞ < a < b < ∞. Suppose that (fn ) and (gn ) converge uniformly
on [a, b] to f and g respectively. Prove that (fn gn ) converges uniformly on [a, b] to f g.

b) Does the statement in a) still hold true if the bounded interval [a, b] is replaced by R? Prove your
assertion or provide a counterexample.

Problem 5 (20 pts): For −∞ ≤ a < b ≤ ∞, we denote

Z
1
L ((a, b)) = {f : (a, b) → R Lebesgue mesurable and |f |dλ < ∞},
(a,b)
Rb
where λ is Lebesgue measure on R. If f ∈ L1 ((a, b)), we write
R
(a,b) f dλ = a f (x)dx.

a) Let f (x) = xs sin(x2 ). Find all s ∈ R for which f ∈ L1 ((0, ∞)).

b) Let f ∈ L1 ((0, 1)) and define fn (x) = f (x) sin(nx) for all n ∈ N. Prove that
Z
fn ∈ L1 ((0, 1)) and lim fn dλ = 0.
n→∞ (0,1)

1
Problem 6 (20 pts): Let (R, B(R), µ) be a measure space where

B(R) is the σ-algebra of Borel sets
on R and µ : B(R) → [0, ∞] is a given measure on R, B(R) . Let h : R → R be a measurable map
and define the image measure generated by h by

h∗ µ(B) = µ h−1 (B) for every B ∈ B(R).

a) Prove that h∗ µ is a measure on R, B(R) . R
b) Let g : R → R be an integrable function with respect to the measure h∗ µ, i.e. |g|d(h∗ µ) < ∞.
R
Prove that Z Z
gd(h∗ µ) = (g◦ h)dµ, where g◦ h(x) = g(h(x)).
R R
c) Let h : R → R be a measurable map which is strictly monotonically increasing, continuous differ-

entiable and surjective. Let (R, B(R), µh ) be a measure space, where µh is a measure on R, B(R)
defined by Z
µh (B) = h′ (x)dx for every B ∈ B(R).
B
Verify that h∗ µh = λ, where λ is Lebesgue measure and the following formula holds
Z Z
g(y)dy = g(h(x))h′ (x)dx.
R R

Problem 7 (bonus 10 pts): Suppose f is a bounded, measurable function defined on a set of finite
measure E. Define the upper and lower Lebesgue integrals, respectively, as
Z nZ o
f dµ = inf ψdµ : ψ simple measurable and f ≤ ψ on E ,
E E
Z nZ o
f dµ = sup ϕdµ : ϕ simple measurable and ϕ ≤ f on E .
E E
R R
Prove that E f dµ = E f dµ. What if the function f is allowed to be unbounded or defined on a set of
infinite measure?

2
Sketch of solutions
Problem 1: The basic idea is to use Newton’s method by looking for the positive root of the function
g(x) = x3 − q. Consider the recursive sequence
g(xn ) x3n − q 2 q 
xn+1 = xn − = x n − = x n + , x0 = q.
g ′ (xn ) 3x2n 3 2x2n
Clearly, (xn )n∈N is a sequence of rational numbers whenever the initial guess x0 is integer. From
here, we can prove that xn converges to some limit by several ways. The limit, say L, satisfies
2 q
 √
L = 3 L + 2L2 , which gives L = 3 q. Note that every convergent sequence is a Cauchy sequence,
so we just need to check that the sequence (xn )n∈N converges. For that, we show that (xn )n∈N is
monotone and bounded. √
We first observe by induction that xn > 0 for all n ≥ 1. From the inequality a+b+c
3 ≥ 3 abc for
any a, b, c ≥ 0, we have
√
r
2 q  1 q  q
xn+1 = xn + 2 = xn + xn + 2 ≥ 3 xn xn 2 = 3 q,
3 2xn 3 xn xn

hence, (xn )n∈N is bounded from below, in particular, x3n ≥ q for all n ≥ 1. We now compute
2 q  1
xn + 2 = 2 x3n − q ≥ 0.


xn − xn+1 = xn −
3 2xn 3xn
Hence, (xn )n∈N is monotonically decreasing. Since the sequence is bounded from below and decreasing,
it converges.
|xn+1 −xn |
You may also estimate the ratio |x n −xn−1 |
≤ c for some constant c ∈ (0, 1). You can guess the
g(x) g ′ (x)
constant c from the derivative estimate of the map T (x) = x − g ′ (x) , that is T ′ (x) = 1 − g ′ (x) +
g(x)g ′′ (x) (x3 −q)6x x3 −q
= 9x4 = 32 x3 , hence, |T ′ (x)| < 23 , thus, you can check the ratio with c = 23 . From this
(g ′ (x))2
ratio estimate, you can simply assert that (xn )n∈N is a Cauchy sequence (analog to the contraction
mapping principle). Since R is complete, every Cauchy sequence converges. The limit is L solving
2 q
 √
L = 3 L + 2L2 which gives L = 3 q.

Problem 2: Suppose E1 , E2 , · · · is a sequence of disjoint sets in A. By definition of ν, we write

∞
[ Z X∞


ν Ei = χEi hdµ.
i=1 i=1
Pn 1
Note that i=1 χEi (x)h(x) ≤ |h(x)|, and since h ∈ L (µ), the dominated convergence theorem
allows us to interchange the limit and the integral, that is
Z X ∞ Z Xn X∞ Z ∞
X
χEi hdµ = lim χEi hdµ = χEi hdµ = ν(Ei ).
n→∞
i=1 i=1 i=1 i=1

1
Pn 3: Since f ∈ L (λ), the function |f | can be approximate by a positive step function, say
Problem
ϕ = i=1 ci χEi with λ(Ei ) < ∞ for all i = 1, · · · , n, such that
Z
|f | − ϕ dλ < ϵ,

with ϵ > 0 arbitrary. We now write from the definition for all x, y ∈ R with x < y,
Z Z


|g(x) − g(y)| = f dλ ≤ ϕ + |f | − ϕ dλ ≤ M λ((x, y)) + ϵλ((x, y)),
(x,y) (x,y)

3
where M = max{|c1 |, · · · , |cn |}. We can check the uniform continuity of g by definition: for every
ϵ > 0, choose δ = 2(Mϵ+ϵ) , then for any x, y ∈ R with |x − y| < δ, we have

ϵ
|g(x) − g(y)| < (M + ϵ)δ = < ϵ.
2

Problem 4: a) You know how to do it in your homework. The boundedness of (fn ) and (gn ) should
be used in the proof.
b) The uniform convergence of (hn ) and (gn ) does not guarantee that (fn gn ) converges uniformly to
(f g) on unbounded set, say R. For example, consider fn = x + n1 which converges uniformly on R to
f (x) = x, however, (fn2 ) does not converge uniformly to x2 on R. Indeed, we compute
2x 1
|fn2 (x) − f 2 (x)| = + 2 ,
n n
hence,
1
sup fn2 (x) − f 2 (x) ≥ fn2 (n) − f 2 (n) = 2 + > 2.
x∈R n2

Problem 5: a) We can split the integral over (0, ∞) R 1 into (0, 1) and [1, ∞). On (0, 1), we note that
2 2 2
c1 x ≤ sin x ≤ c2 x , so c1 x s+2 ≤ f (x) R≤ c2 x s+2 s+2
and 0 x dx < ∞ if and only if s > −3. On [1, ∞),
∞
we have |f (x)| ≤ xs and the integral 1 xs dx < ∞ if s < −1. We now check that f ∈ L1 ([1, ∞))
implies p < −1. To see this, we consider the measurable simple function
1 s
hp p i
ϕ(x) = [(2n + 1/3)π] 2 if x ∈ (2n + 1/3)π, (2n + 2/3)π and ϕ(x) = 0 elsewhere,
2
which satisfies 0 < ϕ ≤ |f | on [1, ∞). We then compute
Z ∞ ∞ ∞ ∞
X 1 p s p  X s
− 12
X s−1
ϕ(x)dx = [(2n + 1/3)π] (2n + 2/3)π − (2n + 1/3)π ≤ c3
2 n n =
2 n 2 .
1 2
n=1 n=1 n=1

This series converges if and only if (s − 1)/2 < −1, or s < −1. We then conclude that f ∈ L1 ((0, ∞))
if and only if s ∈ (−3, −1).

b)R You can check fn ∈ L1 ((0, 1)) by using |fn | ≤ |f | on (0, 1) and f ∈ L1 ((0, 1)). To check
1
lim fn = 0, we use the fact that f ∈ L1 ((0, 1)) can be approximated by a step function, say
n→∞ 0
R1
ψ(x) = m
P
k=1 ck χ(xk−1 ,xk ) with x0 = 0 and xm = 1 such that 0 |f − ϕ| < ϵ for arbitrary small ϵ > 0.
R1 R1 R1
We then write 0 fn = 0 (f − ψ) sin(nx)dx + 0 ψ sin(nx)dx, where we compute explicitly
Z 1 m m
1X h i 2X
ψ sin(nx) ≤ |ck | | cos(nxk )| + | cos(nxk−1 )| ≤ |ck |.
0 n n
k=1 k=1

The triangle inequality yields for n ≥ N with N large enough,

Z 1 Z 1 Z 1 m
2X
fn ≤ |f − ψ|| sin(nx)|dx + ψ sin(nx) < ϵ + |ck | < 2ϵ.
0 0 0 n
k=1
R1
Since ϵ is arbitrary, the integral 0 fn must go to 0 as n → ∞.

4
Problem 6: a) We need to check three axioms: i) for every B ∈ B(R), h∗ µ(B) = µ(h−1 (B)) ≥ 0
since µ is a measure; ii) h∗ µ(∅) = µ(h−1 (∅)) = µ(∅) = 0; iii) For every sequence of disjoint sets
E1 , E2 , · · · in B(R), we write
∞
[   ∞ ∞ ∞ ∞
[
 [  X
X
h∗ µ Ei = µ h−1 Ei = µ h−1 (Ei ) = µ h−1 (Ei ) = h∗ µ(Ei ).
i=1 i=1 i=1 i=1 i=1

b) Since a general function can be decomposed into positive and negative parts, we may assume g is
nonnegative and start with the simplest case when g is a characteristic function, say g = χE for some
measurable set E ⊂ R. By definition, we write
Z Z
−1


χE d(h∗ µ) = h∗ µ(E) = µ h (E) = χh−1 (E) dµ.
R R

We also have Z Z Z
χE ◦ hdµ = χE (h(x))dµ(x) = χh−1 (E) dµ,
R R R

where we used the fact that χE (h(x)) = 1 if x ∈ h−1 (E) and χE (h(x)) = 0 if x ̸∈ h−1 (E). Therefore,
the formula holds for the simplest case of characteristic functions. By the linearity
Pn of Lebesgue integral,
we can also conclude the same for the case of simple functions, say g = i=1 ci χEi . For the general
case, we write by the definition of Lebesgue integral
Z Z
gd(h∗ µ) = sup{ ϕd(h∗ µ) : ϕ : R → [0, ∞) simple, ϕ ≤ g}.
R R

We note that ϕ(y) ≤ g(y) can be written as ϕ(h(x)) ≤ g(h(x)) or ϕ◦ h(x) ≤ g◦ h(x), and ϕ◦ h is a simple
function. Therefore,
Z Z
gd(h∗ µ) = sup{ ϕd(h∗ µ) : ϕ : R → [0, ∞) simple, ϕ ≤ g}
R ZR
= sup{ ϕ◦ hdµ : ϕ : R → [0, ∞) simple, ϕ◦ h ≤ g◦ h}
ZR
= sup{ ψdµ : ψ : R → [0, ∞) simple, ψ ≤ g◦ h}
Z R

= (g◦ h)dµ.
R

c) We write by definition and the fact that h is a nice function (surjective, strictly monotonically
increasing and continuous differentiable),


(h∗ µh ) [a, b) = µh h−1 ([a, b)) = µh h−1 (a), h−1 (b)

Z h−1 (b) Z b
′
= h (x)dx = dy = λ([a, b)).
h−1 (a) a

This computation works for all intervals, so works for all Borel sets of R. Hence, h∗ µh = λ with λ
is Lebesgue measure on R. Since, h∗ µh is Lebesgue measure on R, we can write from b) and the
definition of µh ,
Z Z Z Z
g(y)dy = gd(h∗ µh ) = (g◦ h)dµh = g(h(x))h′ (x)dx.
R R R R

5
Problem 7: a) Fix ϵ > 0, by the approximation lemma, there exist simple measurable functions ψϵ
ϵ
and ϕϵ such that ϕϵ ≤ f ≤ ψϵ and ψϵ − ϕϵ < µ(E) for all x ∈ E. We then compute
nZ o
0 ≤ inf ψdµ : ψ simple measurable and f ≤ ψ on E
E
nZ o
− sup ϕdµ : ϕ simple measurable and ϕ ≤ f on E
Z EZ Z Z
ϵ
≤ ψϵ dµ − ϕϵ dµ = (ψϵ − ϕϵ )dµ < dµ = ϵ.
E E E E µ(E)

Since ϵ is arbitrary, the conclusion follows.

b) If f is defined
R ∞ −2on a set of infinite measure, for example f = x12 defined on [1, ∞). We know that
the integral 1 x dx < ∞, however the upper Lebesgue integral is infinite. Indeed, take g a simple
function such thatR g ≥ f and note that g(x) ≥ η > 0 for all x ≥ 1, where η is the smallest non zero
value of g. Thus, (1,∞) gdµ ≥ ηµ([1, ∞)) = ∞. If f is allowed to be unbounded, the upper Lebesgue
integral may be infinite while the lower Lebesgue integral is finite, for example, the function f (x) = √1x
defined on (0, 1].