Solutions: Problem Set 4

Math 201B, Winter 2007

Problem 1. (a) Define f : R → R by

 −1/2
x if 0 < x < 1,
f (x) =
0 otherwise.
Use the monotone convergence theorem to show that f ∈ L1 (R).
(b) Suppose that {rn ∈ Q | n ∈ N} is an enumeration of the rational numbers.
Define g : R → R by
∞
X 1
g(x) = n
f (x − rn ),
n=1
2
where f is the function defined in (a). Show that g ∈ L1 (R), even though it
is unbounded on every interval.

Solution.
• Define fn : R → R by
x−1/2 if 1/n < x < 1 − 1/n,

fn (x) =
0 otherwise.
Then (fn ) is a monotone increasing sequence of nonnegative, measur-
able functions (since f −1 ((a, ∞)) is open and therefore measurable for
every a ∈ R) which converges pointwise to f on R (so f is measurable).
• Each fn is Riemann integrable on [0, 1], and
Z Z 1−1/n
fn dx = x−1/2 dx
R 1/n
1−1/n
= 2x1/2 1/n


→ 2 as n → ∞.

• The monotone convergence theorem implies that

Z Z
f dx = lim fn dx = 2 < ∞,
R n→∞ R
1
so f ∈ L (R).
 • (b) Let
N
X 1
gN (x) = n
f (x − rn ).
n=1
2
Then (gN ) is a monotone increasing sequence, since f ≥ 0, that con-
verges pointwise to g. By the linearity of the integral and the transla-
tion invariance of Lebesgue measure,
Z N Z
X 1
gN dx = f (x − rn ) dx
R n=1
2n R
N
X 1
= 2
n=1
2n
→ 2 as N → ∞.

Hence, by the monotone convergence theorem

Z
g dx = 2,
R

so g is integrable. (In particular, the series defining g diverges to ∞ on

at most a set of measure zero.)

• The function g is unbounded in any neighborhood of rn ∈ Q, and

therefore on any open interval since the rationals are dense in R.

Remark. This example illustrates that integrable functions are not neces-
sarily as ‘nice’ as one might imagine; in particular they do not necessarily
approach 0 at infinity.
Problem 2. If f ∈ L1 (R), prove that
Z n
1
lim f dx = 0.
n→∞ 2n −n

Give an example to show that this result need not be true if f is not integrable
on R.

Solution.
• Let
1
fn = χ[−n,n] f,
2n
where χ[−n,n] is the characteristic function of the interval [−n, n]. Then
Z Z n
1
fn dx = f dx.
2n −n

• We have fn (x) → 0 as n → ∞ whenever f (x) 6= ±∞, so fn → 0

pointwise a.e. on R. Also, for n ≥ 1,
1
|fn | ≤ |f | ∈ L1 (R).
2

• The Lebesgue dominated convergence theorem implies that

Z Z Z
lim fn dx = lim fn dx = 0 dx = 0,
n→∞ n→∞

which proves the result

• If f = 1, then Z n
1
lim f dx = 1.
n→∞ 2n −n
In this case the sequence
1
fn = χ[−n,n]
2n
converges pointwise (and even uniformly) to 0 on R as n → ∞, but the
integrals do not. Note that the convergence is not monotone and the
sequence (fn ) is not dominated by any integrable function.
Problem 3. Define f : R2 → R by

 1/x2 if 0 < y < x < 1,
f (x, y) = −1/y 2 if 0 < x < y < 1,
0 otherwise.


Compute the following integrals:

Z Z Z  Z Z 
|f (x, y)| dxdy; f (x, y) dx dy; f (x, y) dy dx.
R2 R R R R

Are your results consistent with Fubini’s theorem?

Solution.

• If y ≤ 0 or y ≥ 1, then f (x, y) = 0, and

Z
f (x, y) dx = 0.
R

If 0 < y < 1, then

Z Z y Z 1
1 1
f (x, y) dx = − 2 dx + 2
dx
R 0 y y x
 x=y  x=1
x 1
= − 2 + −
y x=0 x x=y
1 1
= − −1+
y y
= −1.

It follows that
Z Z  Z 1
f (x, y) dx dy = −1 dy = −1.
R R 0

• If x ≤ 0 or x ≥ 1, then f (x, y) = 0, and

Z
f (x, y) dy = 0.
R
 If 0 < x < 1, then
Z Z x Z 1
1 1
f (x, y) dy = 2
dy + − 2 dy
R 0 x x y
h y iy=x  1 y=1
= +
x2 y=0 y y=x
1 1
= +1−
x x
= 1.

It follows that
Z Z  Z 1
f (x, y) dy dx = 1 dx = 1.
R R 0

• According to Fubini’s theorem, we can evaluate the integral of |f | ≥ 0

as an iterated integral (in either order):
Z Z Z 
|f (x, y)| dxdy = |f (x, y)| dx dy
R2 R R
Z 1 Z y Z 1 
1 1
= 2
dx + 2
dx dy
0 0 y y x
Z 1  x=y  x=1 !
x 1
= + − dy
0 y 2 x=0 x x=y
Z 1 
2
= − 1 dy
0 y
Z 1  
2
= lim − 1 dy
n→∞ 1/n y
= lim [2 log y − y]11/n
n→∞
= ∞.

• This example shows that if one drops the assumption that f ∈ L1 in

Fubini’s theorem then the iterated integrals
R with different
R orders need
not be equal.
R Also note that both f+ dxdy and f− dxdy are equal
to ∞, so f dxdy is undefined.
Problem 4. Define f : (0, ∞) × (0, ∞) → R by

f (x, y) = xe−x (1+y ) .

2 2

Compute the iterated integrals with respect to x, y and y, x, and use Fubini’s
theorem to show that Z ∞ √
−t2 π
e dt = .
0 2

Solution.
• Integrating with respect to x followed by y, and using the monotone
convergence theorem, we get
Z ∞ Z ∞  Z ∞ Z n 
−x2 (1+y 2 )
f (x, y) dx dy = lim xe dx dy
0 0 0 n→∞ 0
" #x=n !
e−x (1+y )
2 2
1 ∞
Z
= lim − dy
2 0 n→∞ 1 + y2
x=0
1 ∞ 1
Z
= dy
2 0 1 + y2
Z n
1 1
= lim 2
dy
2 n→∞ 0 1+y
1 n
lim tan−1 y 0

=
2 n→∞
π
= .
4
• Integrating with respect to y followed by x, using the monotone con-
vergence theorem, and making the change of variables t = xy, we get
Z ∞ Z ∞  Z ∞ Z n 
−x2 (1+y 2 )
f (x, y) dy dx = lim xe dy dx
0 0 0 n→∞ 0
Z ∞ Z nx 
−(x2 +t2 )
= lim e dt dx
0 n→∞ 0
Z ∞  Z ∞ 
−x2 −t2
= e dx e dt
0 0
Z ∞ 2
−t2
= e dt .
0
• The function f is non-negative, so Fubini’s theorem implies that the
iterated integrals of f are equal, and both are finite if one is finite.
Equating the two iterated integrals and taking the square-root, we get
the result.
Problem 5. In a normed space X, let

Br (a) = {x ∈ X | kx − ak < r} .

be the ball of radius r > 0 centered at a ∈ X. Lebesgue measure m on Rd

(with the Euclidean norm, say) has the properties that every ball is mea-
surable with finite, nonzero measure, and the measure of a ball is invariant
under translations. That is, for every 0 < r < ∞ and a ∈ X

0 < m (Br (a)) < ∞, m (Br (a)) = m (Br (0)) .

Prove that it is not possible to define a measure with these properties on an

infinite-dimensional Hilbert space.

Solution.

• Since H is an infinite-dimensional Hilbert space, it contains a countably

infinite orthonormal set {en | n ∈ N}.

• If k 6= n then the Pythagorean theorem implies that

p √
kek − en k = kek k2 + ken k2 = 2.

It follows that the balls B√2/2 (ek ) and B√2/2 (en ) are disjoint.

• Suppose that  
m B√2/2 (0) = .
Then, by translation invariance,
 
m B√2/2 (en ) =  for every n ∈ N.

• Since the balls are disjoint, the countable additivity of m implies that
if  > 0
!
[ X  
m B√2/2 (en ) = m B√2/2 (en )
n∈N n∈N
X
= 
n∈N
= ∞.
• If x ∈ B√2/2 (en ), then
√
2
kxk ≤ ken k + kx − en k < 1 + .
2
It follows that [
B√2/2 (en ) ⊂ B1+√2/2 (0).
n∈N

• The additivity and non-negativity of m implies that

!
[  
m B√2/2 (en ) ≤ m B1+√2/2 (0) ,
n∈N
 
so if m B√2/2 (0) > 0 then
 
m B1+ √ = ∞.
2/2 (0)

√
• Thus, if the measure of the smaller ball, with radius
√ 2/2, is nonzero,
the measure of the larger ball, with radius (1 + 2/2), is infinite, so
there is no translation invariant measure that assigns a non-zero, finite
measure to every ball.