Math 210 - Introduction to Analysis

Fall 2024
Solutions to problem set 5

Exercise 1. We say that a differentiable function f : R → R is

convex if the function f ′ is increasing (possibly, non-strictly).
Assume that f is a differentiable convex function.
(a) Show that f (x) + ax + b is convex for any choice of a, b ∈ R.
(b) Suppose that f (c) = f ′ (c) = 0 for some c ∈ R. Show that f is
non-negative. (Geometrically, the graph of f lies above its tangent
line at x = c.)
(c) Now, do not make any assumptions on f (c), f ′ (c). Show that
the graph of f lies above its tangent line at x = c. (Hint: use (a)
with a = −f ′ (c), b = −f (c) + cf ′ (c), and then apply (b))
(d) Prove that either f is non-increasing, or limx→∞ f (x) = +∞.

Solution. (a) Putting g(x) = f (x) + ax + b, we have g ′ (x) =

f ′ (x) + a. Since f ′ is increasing, it follows that g ′ is also increasing,
hence g is convex.
(b) Let x ∈ R. Suppose first that x > c. Since f is differentiable
on (c, x) and continuous on [c, x], we can apply the mean value
theorem: there exists d ∈ (c, x) such that f ′ (d) = f (x)−fx−c
(c)
, or f (x) =
f (c) + f ′ (d)(x − c) = f ′ (d)(x − c). Furthermore, since d > c and
f ′ is increasing, we have that f ′ (d) ≥ f ′ (c) = 0: but then f (x) ≥
0 · (x − c) ≥ 0. If x < c, we can use a similar argument: there exists
d ∈ (x, c) such that f ′ (d) = f (c)−f
c−x
(x)
, or f (x) = −f ′ (d)(c − x). In
′ ′
this case f (d) ≤ f (c) = 0 (because d < c), and so again f (x) ≥
0 · (c − x) ≥ 0.
(c) Define h(x) = f (x) − f ′ (c)x + cf ′ (c) − f (c) = f (x) − f (c) +
f ′ (c)(c − x). From part (a) follows that h is convex. Furthermore,
h(c) = 0, and since h′ (x) = f ′ (x) − f ′ (c), we also have h′ (c) = 0.
Using part (b), we deduce that h(x) ≥ 0 for all x ∈ R, that is

f (x) − f (c) + f ′ (c)(c − x) ≥ 0 ⇔ f (x) ≥ f (c) + f ′ (c)(x − c).

The equation y = f (c) + f ′ (c)(x − c) describes the line tangent to
the graph of f at x = c: the inequality above means exactly that
the graph {y = g(x)} lies above that line.
(d) Suppose that f is not non-increasing. Then there exist x1 <
x2 ∈ R such that f (x1 ) < f (x2 ). Applying the mean value theorem,
we find c ∈ (x1 , x2 ) such that f ′ (c) = f (xx22)−f
−x1
(x1 )
> 0. From part (c)
then follows that f (x) ≥ f (c) + f ′ (c)(x − c) for all x ∈ R. On the
other hand, limx→+∞ (f (c)+f ′ (c)(x−c)) = +∞ (because f ′ (c) > 0),
and so limx→+∞ f (x) = +∞.

Exercise 2.
(a) Let f, g : R → R be differentiable functions s.t. f (x) ≤ g(x),
f (x0 ) = g(x0 ). Show that f ′ (x0 ) = g ′ (x0 ).
(b) Suppose that f, g are as in (a), and h : R → R is a function
such that f ≤ h ≤ g. Show that h is differentiable at x0 , and
h′ (x0 ) = f ′ (x0 ) = g ′ (x0 ).
Define now f (x) = x2 sin(1/x2 ) for x ̸= 0, f (0) = 0.
(c) Show that f is differentiable on R (distinguish the cases x ̸= 0,
x = 0).
(d) Verify that f ′ is unbounded on any neighborhood of x = 0.
Can the fundamental theorem of calculus (in the form seen in class)
be applied to f on [−1, 1]?

Solution. (a) Let h = g − f . It follows from the assumptions that

h is differentiable, non-negative and h(x0 ) = 0. Since x0 is a local
minimum for h, we have h′ (x0 ) = 0, that is f ′ (x0 ) = g ′ (x0 ).
(b) The assumption implies that h(x0 ) = f (x0 ) = g(x0 ). More-
over, for x > x0 we can write
f (x) − f (x0 ) h(x) − h(x0 ) g(x) − g(x0 )
≤ ≤
x − x0 x − x0 x − x0
and for x < x0 ,
g(x) − g(x0 ) h(x) − h(x0 ) f (x) − f (x0 )
≤ ≤ .
x − x0 x − x0 x − x0
Put limx→x0 f (x)−f
x−x0
(x0 )
= ℓ = limx→x0 g(x)−g(x
x−x0
0)
. We can take the
h(x)−h(x0 )
right and left limits of x−x0
as x → x0 : using the sandwich
theorem, we obtain
h(x) − h(x0 ) h(x) − h(x0 )
lim− = ℓ = lim+
x→x0 x − x0 x→x0 x − x0
which shows that h is differentiable at x0 , and h′ (x0 ) = f ′ (x0 ) =
g ′ (x0 ).
(c) On the open intervals (−∞, 0) and (0, +∞), the function f
is the composition of differentiable functions, and thus it is differen-
tiable. Furthermore, note that −x2 ≤ f (x) ≤ x2 : the differentiabil-
ity of f at x = 0 is then a consequence of part (b), and f ′ (0) = 0.
(d) We can compute the derivative for x ̸= 0 explicitly:
2
f ′ (x) = 2x sin(1/x2 ) − cos(1/x2 ).
x
The term 2x sin(1/x2 ) is estimated by 2|x|, and thus it is bounded
on bounded neighborhoods of 0. On the other hand, the√term
2
x
cos(1/x2 ) is equal to x2 whenever cos(1/x2 ) = 1 (i.e. x = ±1/ 2πn),
and therefore is unbounded around 0. It follows that f ′ is unbounded
on bounded neighborhoods of 0: in particular, it is not Riemann
integrable. Thus, despite the fact that f ′ exists everywhere, the
fundamental theorem of calculus does not apply on any bounded
interval containing 0, in particular it does not apply on [−1, 1].

Exercise 3. Let f : [a, +∞) → R be a continuous function. We

define the improper integral of f on [a, +∞) as
Z +∞ Z b
f (t)dt = lim f (t)dt
a b→+∞ a

provided that the limit exists (in that case we say that the integral
converges). (Note: lim g(b) = ℓ ⇔ ∀ϵ > 0 ∃N ∈ R : ∀b ≥
b→+∞
N, |g(b) − ℓ| < ϵ. The
R +∞characterization via sequences also holds).
(a) Prove that a f (t)dt converges if and only if the following
Cauchy criterion is satisfied: for all ϵ > 0 there exists M ≥ a such
Rx
that, for all x2 > x1 ≥ M , x12 f (t)dt < ϵ.
R +∞
(b) Show that 1 sin(t)t2
dt is convergent.
Rb
Solution. (a) Suppose first that limb→+∞ f (t)dt = ℓ exists. Let
a
Rb
ϵ > 0: there exists M ≥ a such that, for all b ≥ M , a f (t)dt − ℓ <
ϵ/2. Then for all b2 ≥ b1 ≥ M , we also have
Z b2 Z b2 Z b1
f (t)dt = f (t)dt − f (t)dt =
b1 a a
Z b2  Z b1 
= f (t)dt − ℓ − f (t)dt − ℓ ≤
a a
Z b2 Z b1
ϵ ϵ
≤ f (t)dt − ℓ + f (t)dt − ℓ <
+ = ϵ.
a a 2 2
Vice versa, suppose that the Cauchy condition is satisfied. Let
Rx R x′′
F (x) = a f (t)dt, so that F (x′′ ) − F (x′ ) = x′ f (t)dt, and let {xn }
be any sequence diverging to infinity. Fixed ϵ > 0, let M ≥ a be as
provided by the Cauchy condition: then, there exists N ∈ N such
that xn > M for all n ≥ N , and therefore |F (xm ) − F (xn )| < ϵ
for all m, n ≥ N . This implies that the sequence F (xn ) is Cauchy,
and therefore converges to a finite limit. However, to apply the
characterization in terms of sequences, we still need to show that
this limit is the same for any choice of {xn }. Let then {x′n }, {x′′n } be
any two sequences diverging to infinity, and define {x′′′ ′′′ ′′
n } as x2n = xn ,
x′′′ ′ ′′′
2n+1 = xn . It is easy to show that {xn } diverges to infinity, and
thus by the argument above F (xn ) → ℓ ∈ R. But F (x′n ), F (x′′n ) are
′′′

subsequences of F (x′′′ n ), and thus converge to the same limit ℓ.

(b) For all 1 ≤ x1 < x2 , we have
Z x2 Z x2 Z x2
sin(t) sin(t) 1 1 1
dt ≤ dt ≤ dt = − .
x1 t2 x1 t2 x1 t
2 x1 x2
Fixed ϵ > 0, take M = max{1, 1ϵ }. Then for all x2 > x1 ≥ M
Z x2
sin(t) 1 1 1 1
2
dt ≤ − < ≤ < ϵ;
x1 t x1 x2 x1 M
R +∞
applying the Cauchy criterion, it follows that 1 sin(t)
t2
dt converges.

Exercise 4. Define fn (x) = (x + n)/n2 for −n < x < 0, fn (x) =

(−x + n)/n2 for 0 ≤ x < n, fn (x) = 0 otherwise.
(a) Verify that fn is continuous for all n ∈ N.
(b) Show that {fnR} converges uniformly
R to 0 as n → ∞.
(c) Is it true that R fn (x)dx → R 0dx as n → ∞?

Solution. (a) The function fn is defined as a (linear) polynomial on

the open intervals (−∞, −n),(−n, 0), (0, n), (n, +∞), and therefore
it is continuous on those intervals. One only needs to check the
points x = −n, 0, n. Using well-known limits, we can write
x+n
lim − fn (x) = lim − 0 = 0 = fn (−n) = lim + = lim + fn (x),
x→−n x→−n x→−n n2 x→−n

−x + n
lim− fn (x) = lim− = 0 = fn (n) = lim+ 0 = lim + fn (x),
x→n x→n n2 x→n x→−n
x+n 1 −x + n
lim− fn (x) = lim− 2
= = fn (0) = lim+ = lim+ fn (x).
x→0 x→0 n n x→0 n2 x→0
It follows that fn is also continuous at x = −n, 0, n, and thus is
continuous everywhere on R.
(b) It is clear that supx∈(−∞,−n) |fn (x)| = 0 = supx∈(n,+∞) |fn (x)|.
On the other intervals we have supx∈[−n,0] |fn (x)| = fn (0) = n1
(note that fn is positive and increasing on the interval [−n, 0]),
supx∈[0,n] |fn (x)| = fn (0) = n1 (fn is positive and decreasing on the
interval [0, n]). It follows that
1
d(fn , 0) = sup |fn (x) − 0| = sup |fn (x)| = .
x∈R x∈R n
It immediately follows that d(fn , 0) → 0 as n → ∞, which means
that fn → 0 uniformly as n → ∞.
(c) We can compute the integral of fn directly. Since fn (x) = 0
for |x| > n,R we are reduced to an integral over a finite interval (in
particular R fn (t)dt converges):
Z n Z 0 Z n
−t + n
Z
t+n 1 1
fn (t)dt = f (t)dt = 2
dt + 2
dt = + = 1
R −n −n n 0 n 2 2
R
R all n ∈ N. It follows that R fn (t)dt does not converge to
for
R
0dt = 0 as n → ∞, despite the fact that fn → 0 uniformly
as n → ∞. (Note that the theorem about convergence of integrals
does not apply here, since the domain of integration is not compact).

Exercise 5. Let A ⊂ R, and consider sequences of continuous

functions fn , gn : A → R such that fn → f , gn → g uniformly on A.
(a) Show that fn + gn → f + g uniformly on A.
(b) Is it always true that fn · gn → f · g uniformly on A?
(c) Same question as (b), assuming that A is compact.

Solution. (a) Let ϵ > 0: by assumption, there exist N1 , N2 > 0

such that ∥fn − f ∥∞ < ϵ/2 for n ≥ N1 , ∥gn − g∥∞ < ϵ/2 for n ≥ N2 .
Therefore, putting N = max{N1 , N2 }, for n ≥ N we can write
∥fn + gn − (f + g)∥∞ = ∥(fn − f ) + (gn − g)∥∞ ≤
ϵ ϵ
≤ ∥fn − f ∥∞ + ∥gn − g∥∞ < + = ϵ.
2 2
This means that fn + gn → f + g as n → ∞, uniformly on A.
(b) Define fn , gn : R → R as fn (x) = gn (x) = x + n1 , and let
h(x) = x. Clearly, fn , gn → h uniformly on R as n → ∞. However,
for all x ∈ R we have
2 1 2 1
|fn (x)gn (x) − h2 (x)| = x2 + x + 2 − x2 = x + 2
n n n n
therefore, for all n ∈ N, we have
2 1
∥fn gn − h2 ∥∞ = sup |fn (x)gn (x) − h2 (x)| = sup x + 2 = +∞
x∈R x∈R n n
and thus clearly fn gn does not converge to h2 uniformly over R.
(c) For all x ∈ A, we have
|fn (x)gn (x)−f (x)g(x)| = |(fn (x)−f (x))gn (x)+f (x)(gn (x)−g(x))| ≤
 ≤ |fn (x) − f (x)| · |gn (x)| + |f (x)| · |gn (x) − g(x)|.
Next observe that, since A is compact,
∥f ∥∞ = sup |f (x)| < ∞, ∥g∥∞ = sup |g(x)| < ∞
x∈A x∈A

because |f |, |g| are continuous functions on the compact set A. Fur-

thermore, there exists N ∈ N such that ∥gn −g∥∞ < 1 for all n ≥ N ,
and thus
∥gn ∥∞ = ∥gn − g + g∥∞ ≤ ∥gn − g∥∞ + ∥g∥∞ < 1 + ∥g∥∞
for n ≥ N . Taking the supremum in the estimates above,
∥fn gn − f g∥∞ ≤ ∥fn − f ∥∞ ∥gn ∥∞ + ∥f ∥∞ ∥gn − g∥∞ ≤
≤ ∥fn − f ∥∞ (1 + ∥g∥∞ ) + ∥f ∥∞ ∥gn − g∥∞ .
Since ∥fn − f ∥∞ → 0, ∥gn − g∥∞ → 0 as n → ∞, it follows that
∥fn gn − f g∥∞ → 0 as n → ∞, i.e. fn gn → f g uniformly on A.

Exercise 6.* Let ∞ n

P
n=0 an x = f (x) be a power
P∞ series with radius
of convergence 1. Furthermore, assume thatP n=0 an = ℓ is conver-
gent. We denotePby sn the partial sums of an , and by tn (x) the
n
partial sums of x .
(a) Show that, for |x| < 1, f (x) − ℓ = (x − 1) ∞
P
n=0 an tn−1 (x).
(b) Using the formula of summation by parts, verify that
N
X N
X −1
an tn−1 (x) = − sn xn + sN tN −1 (x)
n=0 n=0

and deduce that, for |x| < 1,

∞ ∞ ∞
X X ℓn
X
an tn−1 (x) = − sn x + = (ℓ − sn )xn .
n=0 n=0
1 − x n=0

(c) FixedPϵ > 0, show that there exists k ∈ N such that, for all
0 < x < 1, | ∞ n
n=k (ℓ − sn )x | < ϵ/(1 − x).
(d) Using (a,b,c), show that limx→1− f (x) = ℓ (Abel’s theorem).
(−1)n+1
(e) Prove that ∞
P
n=1 n
= log(2).

Solution. (a) For all x ∈ R with |x| < 1, we have

∞
X ∞
X ∞
X
f (x) − ℓ = an x n − an = an (xn − 1) =
n=0 n=0 n=0
 ∞
X ∞
X
n−1 n−2
= an (x − 1)(x +x + . . . + x + 1) = (x − 1) an tn−1 (x).
n=0 n=0
(b) Putting s−1 = 0, we can write (re-proving the summation by
parts formula)
N
X N
X N
X N
X
an tn−1 (x) = (sn −sn−1 )tn−1 (x) = sn tn−1 (x)− sn−1 tn−1 (x) =
n=0 n=0 n=0 n=0
N
X N
X −1
= sn tn−1 (x) − sn tn (x) =
n=0 n=−1
N
X −1 N
X −1
= sn tn−1 (x) − sn tn (x) − s−1 t−1 (x) + sN tN −1 (x) =
n=0 n=0
N
X −1 N
X −1
=− sn (tn (x) − tn−1 (x)) + sN tN −1 (x) = − sn xn + sN tN −1 (x).
n=0 n=0
Since the resulting series are convergent whenever |x| < 1 (note
p that
sn is bounded,
√ |sn | ≤ C for all n ∈ N, hence lim supn→∞ |sn | ≤
n

n
lim supn→∞ C = 1, meaning that the radius of convergence of
PN −1 n
n=0 sn x is at least 1) we can take the limit as N → ∞:
∞
X ∞
X
an tn−1 (x) = − sn xn + lim sN tN −1 (x) =
N →∞
n=0 n=0
∞ ∞ ∞ ∞
X
n1 X
n
X
n
X
=− sn x + ℓ · =− sn x + ℓ x = (ℓ − sn )xn .
n=0
1 − x n=0 n=0 n=0
(c) Let ϵ > 0. Then there exists k ∈ N such that |sn − ℓ| < ϵ for
all n ≥ k. Using this fact, we can estimate the tail of the series as
follows:
∞ ∞ ∞ ∞
X X X X ϵ
(ℓ − sn )xn ≤ |ℓ − sn |xn < ϵ xk ≤ ϵ xk = .
n=k n=k n=k n=0
1 − x
(d) Using (a) and (b), for 0 < x < 1 we can rewrite |f (x) − ℓ| as
∞
X ∞
X
|f (x) − ℓ| = |x − 1| an tn−1 (x) = (1 − x) (ℓ − sn )xn .
n=0 n=0

Given ϵ > 0, let k ∈ N be as in part (c). Then

k−1
X ∞
X
|f (x) − ℓ| = (1 − x) (ℓ − sn )xn + (ℓ − sn )xn ≤
n=0 n=k
 k−1
X ∞
X k−1
X
≤ (1−x) (ℓ − sn )xn +(1−x) (ℓ − sn )xn < (1−x) (ℓ − sn )xn +ϵ.
n=0 n=k n=0
Pk−1
Now, the finite sum n=0 (ℓ − sn )xn is just a polynomial, and there-
fore continuous on [−1, 1]: in particular it is bounded on that inter-
val, i.e. there exists D > 0 such that | k−1 n
P
n=0 (ℓ − sn )x | ≤ D for all
|x| ≤ 1. Let δ = ϵ/D. Then, for all 1 − δ < x < 1, we have
k−1
X
|f (x) − ℓ| < (1 − x) (ℓ − sn )xn + ϵ < δD + ϵ = ϵ + ϵ = 2ϵ.
n=0

By definition, it follows that limx→1− f (x) = ℓ.

(e) Recall the power series expansion of the logarithm:
∞
x2 x3 X (−1)n+1
log(1 + x) = x − + − ... = xn for |x| < 1.
2 3 n=1
n

(−1)n+1
Furthermore, the alternating series ∞
P
n=1 n
is convergent. We
have thus the right to apply the result of part (d):
∞
X (−1)n+1
= lim− log(1 + x) = log(2)
n=1
n x→1

since log(1 + x) is continuous at x = 1.