18.100C. Problem Set 8.

Solutions

Problem 1: This problem constructs an example of a continuous function

which is nowhere dierentiable. For a real number x, let {x} denote the
distance of x to the nearest integer. Consider the function f : R R, given
by the formula

{10n x}
f (x) = .
10n
n=0
(a) Show that the series converges for every x R (and therefore, f is
well-dened).
(b) Show that f is continuous at all x R.
(c) Prove that for every x R, f is not dierentiable at x, by showing
that the limit
f (x + h) f (x)
lim
h0 h
does not exist. (Hint: Consider the decimal expansion of x and take h m =
10m depending on the m-th digit after the decimal point in the expan
sion.)

a) For all y R, 0 {y} < 1, so {10

n x} 1
10n < 101n , and since n=0 10n
converges, f (x) converges uniformly by Weierstrass M-test.
b) The function x {x} is continuous on R (and periodic with period 1).
Then x {10n x} is continuous being the composition of two continuous
functions. This shows that the series is one of continuous functions, and
since it converges uniformly, the sum of the series, f (x), is continuous.
c) Fix x = a.a1 a2 . . . am . . . . Set hm = 10m where the choice for + or
will be made later. Then


f (x + hm ) f (x) m1
{10n x + 10n hm } {10n x}
= 10 m
.

hm
n=0
10n

This is because, if n m, 10n h n n

m is an integer, so {10 (x + hm )} = {10 x}.
n n n
For simplicity, denote n = {10 x + 10 hm } {10 x}. For n < m 1,
|n | = |.an . . . am .an . . . (am 1) . . . | 10(mn1) . For n = m 1,
|m1 | = |{.(am 1) . . . } {.am . . . }| = 101 , if we make the choice of + or
so that both .(am 1) . . . and .am . . . are on the same side of 12 .
In conclusion,

f (x + hm ) f (x)
10m (1 101 102 10m+1 ) > 10m (1 1 ),
hm 9
and this shows that the sequence diverges, and therefore the derivative does
not exist.
Problem 2: Rudin: Chapter 6, ex. 13.
1
2

(a) This is a straightforward calculation. We just record the result

2
cos x2 cos(x + 1)2 1 (x+1) cos u
x+1
2
f (x) = sin(t )dt = 3 du.
x 2x 2(x + 1) 4 x2 u2
Since
(x+1)2 3
2
u 2 du = ,

x(x + 1)

x2

we conclude that |f (x)| < 12 ( x1 + x+1

1 1
+ x(x+1) ) = x1 .
(b) Multiply by 2x in the previous formula and nd that 2xf (x) =
cos(x2 ) cos(x + 1)2 + r(x), where r(x) = x1 .
(c) The formula in (b), since limx r(x) = 0, immediately implies that
lim supx xf (x) 1, and lim inf x xf (x) 1. In fact equality holds
in both these formulas.
(d) The integral converges (see the note about computing the value of
this integral).

y To decide convergence, one
y can proceed as in (a), but with
the integral x sin(t2 )dt and show that | x sin(t2 )dt| < x1 .
Problem 3: Rudin: Chapter 7, ex. 4.
(a) Note rst that the series is undened when x is of the form x = m12 ,
for some m Z. Also, it is clear that for x = 0, the series diverges.
If x > 0, 1
n=1 1+n2 x = 1
x
1 1
n=1 n2 + 1 . Since n2 + 1 < n2 , and
1 1
n=1 n2
x x
1
converges, the series n=1 1+n2 x converges (in fact absolutely, as it has
positive terms anyway).
If x < 1, 1
n=1 |1+n2 x | = |x|
1 1
n=1 |n2 + 1 | = |x|
1 1
n=1 n2 1 . Since x <
|x|
x
1, for n 2, n2 1 1 < n211 . The series 2
n=2 1n 1 converges, and so by
|x|
the comparison test, 1

n=1 1+n2 x converges absolutely when x < 1.
It remains the case 1 < x < 0. There exists m Z, such that m12 <
1 1 1
x < (m+1) 2 . Then n=m+1 |1+n2 x| = n=m+1 n2 |x|1 . A similar argument
to the other two cases shows that the series converges absolutely in this case
as well.
(b) If x a > 0, then 1+1n2 x 1+n 1 1
2 a . Since 1+n2 a
converges, by
1
Weierstrass M-test, 1+n2 x converges unformly on every interval [a, b]
(0, ).
On intervals of the form (0, b] the series does not converge uniformly.
Assume by way of contradiction that it does. Then Cauchys test implies
that there exists N > 0 such that m n=N 1+n2 x
1
< 1
2 , for all x (0, b]. But if
we set x = N12 we get a contradiction.
1
If x b < 1, | 1+1n2 x | = n2 |x|1 < n2 |b1|1 , so similarly the series converges
uniformly on all intervals [a, b] (, 1). But if one considers intervals of
the form [a, 1), a similar argument to the case x > 0 shows that the series
does not converge uniformly.
 3

The case 1 < x < 0 is similar. As long as the interval is [a, b]

( m12 , (m+1)
1
2 ), the series converges uniformly, but if one of the endpoints

is of the form m12 , it does not.

(c) Since all the terms of the series are continuous functions whenever they
are dened, f (x) is continuous on all intervals where it converges uniformly.
Every point of convergence (see a)) can be put in an interval [a, b] as in
b) where the series converges uniformly. Therefore, f is continuous for all
values of x for which is converges.
(d) The function f is unbounded though. For example, lim x0+ f (x) =
.