Folland: Real Analysis, Chapter 2

Sebastien Picard
Problem 2.3
If {f
n
} is a sequence of measurable functions on X, then {x : limf
n
(x) exists} is a measurable set.
Solution:
Dene h = limsupf
n
, g = liminff
n
. By Proposition 2.7, h, g are measurable. Let
E

n=1
g
1
(n, ) h
1
(n, )
E

n=1
g
1
(, n) h
1
(, n)
.
It is clear that both E

and E

are measurable sets. Next, dene

.
Problem 2.9
Let f : [0, 1] [0, 1] be the Cantor function (1.5), and let g(x) = f(x) + x.
a. g is a bijection from [0, 1] to [0, 2] and h = g
1
is continuous from [0, 2] to [0, 1].
b. If C is the Cantor set, m(g(C)) = 1.
c. By Exercise 29 of Chapter 1, g(C) contains a Lebesgue nonmeasurable set A. Let B = g
1
(A).
Then B is Lebesgue measurable but not Borel.
d. There exists a Lebesgue measurable function F and a continuous function G on R such that
F G is not Lebesgue measurable.
Solution:
(a) We know that f is an increasing function, and therefore f(x) +x is a strictly increasing function.
Also, since f is continuous, then f(x) + x is continuous. Since g(0) = 0 and g(1) = 2, and g is
continuous and strictly increasing on [0, 1], then g is a bijection from [0, 1] to [0, 2]. It follows that
g
1
exists.
1
Since g is continuous map from a compact set to a Hausdor space, then g
1
is continuous:
indeed, g(K) is compact for any compact set K, and compacts sets are closed in a Hausdor space,
so (g
1
)
1
(C) is closed for any closed set C in the domain of g.
(b) Let
[0, 1] C
c
=

_
n=1
E
n
where E
n
are countably many disjoint intervals. Since f is constant on a given E
n
, then g(E
n
) is
a translate of E
n
, so m(g(E
n
)) = m(E
n
). Therefore
m(g([0, 1] C
c
)) = m(

_
m=1
g(E
n
)) = m(

_
m=1
E
n
) = m([0, 1] C
c
) = 1
.
Since 2 = m([0, 2]) = m(g(C)) + m(g([0, 1] C
c
)), we have m(g(C)) = 1.
(c) Suppose B is a Borel set. Since g
1
is continuous, (g
1
)
1
(B) = A is Borel measurable. But A
is not even Lebesgue measurable, which is a contradiction.
However, B C. Since C is a null set, B is Lebesgue measurable by completeness of Lebesgue
measure.
(d) Dene G and F as follows:
G(x) =
_
_
_
g
1
(x) if x [0, 2]
x if x 0
1 if x 2
F(x) =
_
x if x B
69 else
G is continuous by the Pasting Lemma from elementary topology. F is measurable since F
1
(a, )
is either empty, the whole real line, or a subset of B (which is measurable since B has measure zero).
F G is not Lebesgue measurable since G
1
(F
1
(1, )) = G
1
(B) = A.
Problem 2.14
If f L
+
, let (E) =
_
E
fd for E M. Then is a measure on M, and for any g L
+
,
_
gd =
_
fgd. (First suppose that g is simple.)
Solution:
We will show that is a measure on M. It is clear that () = 0. Let {A
k
} be a countable
sequence of disjoint sets, and dene A =

k=1
A
k
.
2
(A
k
) =
_
A
fd =
_
A
(

k=1

A
k
f)d =

k=1
_
A

A
k
fd =

k=1
(A
k
).
The summation can be taken out from the integral by Theorem 2.15.
Let g L
+
be a simple function. Then g =

n
i=1
a
i

A
i
where A
i
are measurable sets. Therefore
_
gd =
n

i=1
a
i
(A
i
) =
n

i=1
a
i
_
f
A
i
d =
_
n

i=1
a
i

A
i
fd =
_
fgd.
Now take any g L
+
. By Theorem 2.10 there exists a sequence {
n
} of simple functions that
converges to g pointwise such that 0
1

2
g. Since {
n
} and {f
n
} are monotone
increasing, we can apply the monotone convergence theorem twice to obtain
_
gd =
_
lim
n
d = lim
_

n
d = lim
_

n
fd =
_
fgd.
Problem 2.16
If f L
+
and
_
f < , for every > 0 there exists E Msuch that (E) < and
_
E
f > (
_
f).
Solution:
Let f L
+
such that
_
f < . Let > 0. By denition of
_
f, there exists a simple function
=

n
i=1
a
i

E
i
such that 0 f and
_
f <
_
.
Let E =
n
i=1
E
i
. Since the E
i
are in M, then E M. Also,
_

_
f < , so for each E
i
we
have (E
i
) < and therefore (E) < . Hence we have
_
f <
_
E

_
E
f.
Problem 2.20
(A generalized Dominated Convergence Theorem) If f
n
, g
n
, f, g L
1
, f
n
f and g
n
g a.e.,
|f
n
| g
n
, and
_
g
n

_
g, then
_
f
n

_
f. (Rework the proof of the dominated convergence theo-
rem.)
Solution:
Since f
n
+ g
n
0, we can apply Fatous lemma:
_
f + g =
_
lim(f
n
+ g
n
) liminf
_
f
n
+ g
n
=
_
g + liminf
_
f
n
.
The same process can be repeated for g f
n
:
_
g f =
_
lim(g
n
f
n
) liminf
_
g
n
f
n
=
_
g limsup
_
f
n
.
3
Since
limsup
_
f
n

_
f liminf
_
f
n
we can conclude that
_
f
n

_
f.
Problem 2.21
Suppose f
n
, f L
1
and f
n
f a.e. Then
_
|f
n
f| 0 i
_
|f
n
|
_
|f|. (Use Exercise 20.)
Solution:
Suppose
_
|f
n
f| 0. Then
limsup
_
|f
n
| limsup
_
|f
n
f| +|f| =
_
|f|.
On the other hand, we can apply Fatous lemma to obtain
_
|f| liminf
_
|f
n
|.
Therefore
limsup
_
|f
n
|
_
|f| liminf
_
|f
n
|
and hence
_
|f
n
|
_
|f|.
Next we prove the converse: suppose
_
|f
n
|
_
|f|. Dene g
n
= |f
n
| + |f|. Then |f
n
f| g
n
and
_
g
n
2
_
|f| L
1
. By applying the generalized Dominated Convergence Theorem (Problem
20) we obtain
lim
_
|f
n
f| =
_
lim|f
n
f| = 0.
Problem 2.25
Let f(x) = x
1/2
if 0 < x < 1, f(x) = 0 otherwise. Let {r
n
}

1
be an enumeration of the rationals,
and set g(x) =

1
2
n
f(x r
n
).
a. g L
1
(m), and in particular g < a.e.
b. g is discontinuous at every point and unbounded on every interval, and it remains so after
any modication on a Lebesgue null set.
c. g
2
< , but g
2
is not integrable on any interval.
Solution:
(a) We need to show
_
|g| < . Using Theorem 2.15 and the change of variables z = x r
n
, we can
compute
4
_

n=1
2
n
f(x r
n
)dx =

n=1
_

2
n
f(x r
n
)dx
=

n=1
2
n
_
1
0
f(z)dz
=

n=1
2
n
_
1
0
z
1/2
dz
=

n=1
2
1n
= 2(

n=0
2
n
1) = 2 < .
(b) Let x R, M > 0, > 0. It will be shown that there exists a y (x , x + ) such that
g(y) > M. This will prove that g is unbounded on each interval and that g is discontinuous at every
point.
There exists a rational number r
n
(x , x + ). Furthermore, there exists a y (x , x + )
such that
0 < y r
n
<
2
2n
4M
2
.
Then
g(y) 2
n
f(y r
n
) 2
n
(2
n
(2M)) = 2M > M.
This proof does not fail after redening g on any Lebesgue null set, since one can still nd an
irrational y with the desired properties.
(c) Since g < a.e., it follows that g
2
< a.e. However,
g
2

n=1
2
2n
f
2
(x r
n
)dx =

n=1
2
2n
_
1
0
f
2
(z)dz
1
4
_
1
0
dz
z
> .
Problem 2.27
Let f
n
(x) = ae
nax
be
nbx
where 0 < a < b.
a.

1
_

0
|f
n
(x)|dx = .
b.

1
_

0
f
n
(x)dx = 0.
c.

1
f
n
L
1
([0, ), m), and
_

1
f
n
(x)dx = log(b/a).
Solution:
5
(a) Since f
n
is the dierence of two exponential functions, we can nd a point c R such that f
n
< 0
on (0, c) and f
n
> 0 on (c, ). In order to nd this point c, we solve
be
nbc
= ae
nac
log(b/a) = nc(b a)
c = log(b/a)
1
n(b a)
We can now split up the integral in order to integrate |f
n
|:
_

0
|f
n
| =
_
c
o
_
ae
nax
be
nbx
_
dx +
_

c
_
ae
nax
be
nbx
_
dx
=
_
e
nax
n

e
nbx
n
_

c
0
+
_
e
nbx
n

e
nax
n
_

c
=
2
n
_
e
nac
e
nbc
_
=
2
n
_
e

a
ba
log(b/a)
e

b
ba
log(b/a)
_
.
Therefore
_

0
|f
n
| is proportional to (1/n), hence

1
_

0
|f
n
| = .
(b)
_

0
f
n
(x)dx =
_

o
ae
nax
dx
_

0
be
nbx
dx
=
e
nax
n

0
+
e
nbx
n

0
=
1
n
(1 + 1) = 0.
Therefore

1
_

0
f
n
(x)dx = 0.
(c)

n=1
f
n
=

n=1
_
ae
nax
be
nbx
_
= a
_

n=0
e
nax
1
_
b
_

n=0
e
nbx
1
_
= a
_
1
1 e
ax
1
_
b
_
1
1 e
bx
1
_
=
a
1 e
ax

b
1 e
bx
+ (b a).
6
We can see that

n=1
f
n
is positive on (0, ), so to show that

n=1
f
n
L
1
we must compute
_

n=1
f
n
:
_

o

n=1
f
n
=
_

0
_
a
1 e
ax

b
1 e
bx
+ (b a)
_
dx
=
_
log(1 e
ax
) log(1 e
bx
) + (b a)x
_

0
= log
_
(1 e
ax
)(e
(ba)x
)
1 e
bx
_

0
= lim
z0
lim
x
log
_
(1 e
ax
)(1 e
bz
)e
bx
e
ax
(1 e
bx
)(1 e
az
)e
bz
e
az
_
= lim
z0
lim
x
log
_
(1 e
bz
e
ax
+ e
ax+bz
)e
bx
e
az
(1 e
az
e
bx
+ e
bx+az
)e
bz
e
ax
_
= lim
z0
lim
x
log
_
e
ax
(e
ax
e
bzax
1 + e
bz
)e
bx
e
az
e
bx
(e
bx
e
azbx
1 + e
az
)e
bz
e
ax
_
= lim
z0
lim
x
log
_
(e
ax
e
bzax
1 + e
bz
)e
az
(e
bx
e
azbx
1 + e
az
)e
bz
_
= lim
z0
log
_
(e
bz
1)e
az
(e
az
1)e
bz
_
= lim
z0
log
_
1 e
bz
1 e
az
_
= log(b/a).
The last step follows from lHopitals rule:
lim
z0
1 e
bz
1 e
az
= lim
z0
be
bz
ae
az
=
b
a
.
Problem 2.28
Compute the following limits and justify the calculations:
a. lim
n
_

0
(1 + (x/n))
n
sin(x/n)dx.
b. lim
n
_
1
0
(1 + nx
2
)(1 + x
2
)
n
dx.
c. lim
n
_

0
nsin(x/n)[x(1 + x
2
)]
1
dx.
d. lim
n
_

a
n(1 + n
2
x
2
)
1
dx. (The answer depends on whether a > 0, a = 0, or a < 0. How
does this accord with the various convergence theorems?
Solution:
(a) Denote
f
n
=
sin(x/n)
(1 + (x/n))
n
.
First, notice that by the binomial theorem, when n 2, for all x [0, ) we have
7
(1 + (x/n))
n
1 + x +
_
n
2
_
x
2
n
2
1 + x +
x
2
4
.
Thus we can bound f
n
on [0, ) by
|f
n
| =

sin(x/n)
(1 + (x/n))
n

1
(1 + (x/n))
n

1
1 + x + x
2
/4
.
A rough computation shows that the following integral is bounded:
_

0
dx
1 + x + x
2
/4
=
_
1
0
dx
1 + x + x
2
/4
+
_

1
dx
1 + x + x
2
/4

_
1
0
dx
1 + 0 + 0
+
_

1
dx
0 + 0 + x
2
/4
1 + 4 < .
By the Dominated Convergence Theorem, we can throw the limit under the integral.
lim
n

_

0
f
n
dx

_

0
lim
n
|f
n
|dx =
_

0
lim
n

sin(x/n)
(1 + (x/n))
n

dx =
_

0
sin(0)e
x
dx = 0.
(b) Denote
f
n
=
1 + nx
2
(1 + x
2
)
n
.
By the binomial theorem, we know
(1 + x
2
)
n
1 + nx
2
.
Therefore f
n
1, and
_
1
0
1 < , so by the Dominated Convergence Theorem we can conclude
lim
n
_
1
0
f
n
dx =
_
1
0
_
lim
n
1 + nx
2
(1 + x
2
)
n
_
dx = 0.
(c) Denote
f
n
=
nsin(x/n)
x(1 + x
2
)
.
Since | sin(x/n)| (x/n) for x > 0, we can bound f
n
on (0, ) by

nsin(x/n)
x(1 + x
2
)

1
1 + x
2
.
Next we verify that
8
_

0
dx
1 + x
2
= arctan(x)

0
= /2 < .
By the Dominated Convergence Theorem we can conclude
lim
n
_

0
f
n
dx =
_

0
_
lim
n
sin(x/n)
x/n
(1 + x
2
)
1
_
dx =
_

0
dx
1 + x
2
= /2.
(d) In this case we can evaluate the integral directly:
lim
n
_

a
n
1 + n
2
x
2
dx = lim
n
_

na
dy
1 + y
2
= lim
n
arctan(y)

na
= /2 lim
n
arctan(na).
Hence
lim
n
_

a
n
1 + n
2
x
2
dx =
_
_
_
0 a > 0
/2 a = 0
a < 0
This agrees with the fact that one cannot apply the Dominated Convergence Theorem unless a > 0
since there is no way to bound f
n
(0).
Problem 2.32
Suppose (X) < . If f and g are complex-valued measurable functions on X, dene
(f, g) =
_
|f g|
1 +|f g|
d.
Then is a metric on the space of measurable functions if we identify functions that are equal a.e.,
and f
n
f with respect to this metric i f
n
f in measure.
Solution:
We will rst show that is a metric on the space of measurable functions if we identify functions
that are equal a.e.
It is clear that (f, g) = (g, f) and that (f, g) 0. We can also see that
(f, g) = 0
_
|f g|
1 +|f g|
d = 0

|f g|
1 +|f g|
= 0 a.e. (by Proposition 2.16)
f = g a.e.
It only remains to show the triangle inequality. Let x, y, z X. First, suppose |x z| |x y|
and |z y| |x y|. Then we have
9
|x y|
1 +|x y|

|x z|
1 +|x y|
+
|z y|
1 +|x y|

|x z|
1 +|x z|
+
|z y|
1 +|z y|
.
On the other hand, suppose |x z| |x y|. Then
|x y|
1 +|x y|

|x y|
1 +|x y|
+
|z y|
1 +|z y|

|x z|
1 +|x z|
+
|z y|
1 +|z y|
.
The above argument can be repeated when |z y| |x y|. Hence for all x, y, z X,
|x y|
1 +|x y|

|x z|
1 +|x z|
+
|z y|
1 +|z y|
.
Using basic properties of the integral (Proposition 2.13), we can conclude that (f, g) (f, h) +
(h, g) for any measurable functions f, g, h. This completes the proof that is a metric.
Suppose f
n
f with respect to . Let E
n,
= {x : |f
n
(x) f(x)| }.
_
|f
n
f|
1 +|f
n
f|
d
_
En,
|f
n
f|
1 +|f
n
f|
d

1 +
(E
n,
).
It follows that
(E
n,
)
1 +

_
|f g|
1 +|f g|
d 0.
Conversely, suppose f
n
f in measure. Let > 0. Choose N N such that for all integers
n > N, we have

_
_
x : |f
n
(x) f(x)|

2(X)
_
_
<

2
.
Let A = {x : |f
n
(x) f(x)|

2(X)
}. Then the following holds for all n > N:
(f
n
, f) =
_
|f g|
1 +|f g|
d =
_
A
|f g|
1 +|f g|
d +
_
X\A
|f g|
1 +|f g|
d
(A) +

2(X)
(X\A)
< /2 + /2 = .
Problem 2.40
In Egoros theorem, the hypothesis (X) < can be replaced by |f
n
| g for all n, where
g L
1
().
Solution:
Suppose f
1
, f
2
, . . . and f are measurable complex-valued functions on X such that f
n
f a.e.
and |f
n
| g for all n, where g L
1
(). We will follow the proof of Theorem 2.33 and make some
minor adjustments.
10
Without loss of generality, assume that f
n
f everywhere on X. For k, n N, let
E
n
(k) =

_
m=n
{x : |f
m
(x) f(x)| k
1
}.
For xed k, E
n
(k) decreases as n increases and

n=1
E
n
(k) = . To apply continuity of measure
from above, we need (E
1
) < . Since |f
n
f| 2|g|, we observe that
E
1
(k) A(k) := {x : 2|g(x)| k
1
}.
We can use the fact that
> 2
_
X
|g|
_
A(k)
2|g| k
1
(A(k)),
in order to conclude
(E
1
(k)) (A(k)) < .
Therefore, by continuity of measure from above, (E
n
(k)) 0 as n . Given > 0 and k N,
there exists a positive integer n
k
such that (E
n
k
(k)) < 2
k
.
If we dene E =

k=1
E
n
k
(k), then (E) < and f
n
f uniformly on E
c
.
Problem 2.49
Prove Theorem 2.39 by using Theorem 2.37 and Proposition 2.12 together with the following lemmas.
a. If E MN and (E) = 0, then (E
x
) = (E
y
) = 0 for a.e. x and y.
b. If f is L-measurable and f = 0 -a.e., then f
x
and f
y
are integrable for a.e. x and y, and
_
f
x
d =
_
f
y
d = 0 for a.e. x and y. (Here the completeness of and is needed.)
Solution:
(a) Suppose E M N and (E) = 0. Dene f =
E
. Then f
x
=
Ex
and f
y
=
E
y . Apply
Fubinis theorem:
0 =
_
fd( ) =
_
_
_
f
x
d(y)
_
d(x) =
_
_
_
f
y
d(x)
_
d(y).
It follows that
_

Ex
d = 0 -a.e. and (E
x
) = 0 -a.e., and similarly
_

E
y d = 0 -a.e. and
(E
y
) = 0 -a.e..
(b) Suppose f is L-measurable and f = 0 -a.e.. Dene
A = {(x, y) M N : f(x, y) = 0}.
Then A E for some E MN such that (E) = 0. By part (a), (E
x
) = 0 and (E
y
) = 0
for a.e. x and y. Since A
x
E
x
and A
y
E
y
, we have (A
x
) = 0 and (A
y
) = 0. Therefore
_
|f
x
|d =
_

Ax
|f
x
|d = 0 for -a.e. x and
_
|f
y
|d =
_

A
y |f
y
|d = 0 for -a.e. y.
11
We now prove Theorem 2.39. Suppose f is L-measurable and either (a) f 0 or (b) f L
1
(). By
Proposition 2.12, there exists a M N-measurable function g such that f = g -almost everywhere.
By Proposition 2.34, g
x
is N-measurable and g
y
is M-measurable. Dene h = g f. Then h = 0
-a.e.. By lemma (b), h
x
is N-measurable for a.e. x and h
y
is M-measurable for almost every y,
hence f
x
is N-measurable for a.e. x and f
y
is M-measurable for almost every y
In case (b), g L
1
( ), so by lemma (b), h
x
= g
x
f
x
is integrable for a.e. x. By Fubinis
theorem, g
x
is integrable for a.e. x, hence f
x
is integrable for a.e. x. Similarly f
y
is integrable for a.e.
y.
By applying lemma (b) on the function h, we can see that for a.e. x we have
_
(g
x
f
x
)d = 0
hence
_
g
x
d =
_
f
x
d. Similarly for a.e. y we have
_
g
y
d =
_
f
y
d. In case (a), g L
+
(M N),
so by Tonellis theorem x
_
g
x
d =
_
f
x
d is measurable, and y
_
g
y
d =
_
f
y
d is measurable.
In case (b), g L
1
(M N), so by Fubinis theorem x
_
g
x
d =
_
f
x
d is integrable, and
y
_
g
y
d =
_
f
y
d is integrable.
Since
_
gd =
_
fd, after applying Tonellis theorem (case a) or Fubinis theorem (case b) and
using the fact that
_
g
y
d =
_
f
y
d and
_
g
x
d =
_
f
x
d for almost every x and y, we obtain
_
fd =
_ _
f(x, y)dd =
_ _
f(x, y)dd.
Problem 2.55
Let E = [0, 1] [0, 1]. Investigate the existence and equality of
_
E
fdm
2
,
_
1
0
_
1
0
f(x, y) dx dy, and
_
1
0
_
1
0
f(x, y) dy dx for the following f.
a. f(x, y) = (x
2
y
2
)(x
2
+ y
2
)
2
.
b. f(x, y) = (1 xy)
a
(a > 0).
c. f(x, y) = (x 1/2)
3
if 0 < y < |x 1/2|, f(x, y) = 0 otherwise.
Solution:
(a) First, we evaluate
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dx.
Using the substitution x = y tan, dx = y sec
2
d, we obtain
12
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dx =
_
arctan(1/y)
0
(tan
2
1)y
3
sec
2
d
(y
2
(tan
2
+ 1))
2
=
_
arctan(1/y)
0
(tan
2
1)d
y sec
2

=
1
y
_
arctan(1/y)
0
(sin
2
cos
2
)d
=
1
y
_
arctan(1/y)
0
(1 2 cos
2
)d
=
1
y
_
arctan(1/y)
0
cos 2d
=
1
2y
sin(2 arctan(1/y))
=
1
y
sin(arctan(1/y)) cos(arctan(1/y))
=
1
y
sin(arctan(1/y))
cos(arctan(1/y))
cos(arctan(1/y))
1
cos(arctan(1/y))
=
1
y
tan(arctan(1/y))
1 + tan
2
(arctan(1/y))
=
1
1 + y
2
Therefore,
_
1
0
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dxdy =
_
1
0
dy
1 + y
2
= /4.
We observe that
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dx =
_
1
0
y
2
x
2
(x
2
+ y
2
)
2
dx =
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dy.
Hence
_
1
0
_
1
0
x
2
y
2
(x
2
+ y
2
)
2
dydx =
_
1
0
dx
1 + x
2
= /4.
By Fubinis theorem,
_
E
fdm
2
is not dened.
(b) Since f is non-negative on [0, 1] [0, 1], f L
+
(E), so by Tonellis theorem
_
E
fdm
2
=
_
1
0
_
1
0
f(x, y)dxdy =
_
1
0
_
1
0
f(x, y)dydx. The integral may be innite for some values of a... I havent
had time to do this computation yet.
13
(c) First, we compute
_
1
0
_
1
0
f(x, y)dydx =
_
1
0
_
|x0.5|
0
(x
1
2
)
3
dydx =
_
1
0
(x
1
2
)dx
|x
1
2
|
3
.
The function
(x
1
2
)dx
|x
1
2
|
3
is not integrable on [0, 1]:
_
1
0
|x
1
2
|dx
|x
1
2
|
3
=
_
1
0
dx
|x
1
2
|
2
= .
Therefore, the integral
_
1
0
_
1
0
f(x, y)dydx does not exist, and hence
_
E
fdm
2
does not exist.
However,
_
1
0
_
1
0
f(x, y)dxdy = 0:
_
1
0
f(x, y)dx =
_
1
0

{y<|x1/2|}
(x
1
2
)
3
dx
=
_
1/2
1/2

{|z|>y}
dz
z
3
=
_
y
1/2
dz
z
3
+
_
1/2
y
dz
z
3
= 2 +
1
2y
2

1
2y
2
+ 2 = 0.
Problem 2.57
Show that
_

0
e
sx
x
1
sin x dx = arctan(s
1
) for s > 0 by integrating e
sxy
sin x with respect to x and
y. (It may be useful to recall that tan(/2 ) = (tan)
1
. Cf. Exercise 31d.)
Solution:
We will investigate the integral
_

0
_

1
e
sxy
sin xdydx.
We want to apply Fubinis theorem, so rst we verify that
_

0
_

1
|e
sxy
sin x|dydx
_

0
_

1
e
sxy
xdydx =
_

0
e
sx
s
dx < .
Next, we observe that
_

0
_

1
e
sxy
sin xdydx =
1
s
_

0
e
sx
x
1
sin xdx.
This integral can be computed by switching the order of integration. The rst step is to use
integration by parts to compute
14
I =
_

0
e
sxy
sin xdx =
_

0
(sy)e
sxy
cos xdx e
sxy
cos x

0
=
_

0
(sy)e
sxy
cos xdx + 1
=
_

0
(s
2
y
2
)e
sxy
sin xdx sye
syx
sin x

0
+ 1
=
_

0
(s
2
y
2
)e
sxy
sin xdx + 1.
I = s
2
y
2
I + 1.
Therefore
I =
1
1 + s
2
y
2
.
Thus
_

1
_

0
e
sxy
sin xdxdy =
_

1
dy
1 + s
2
y
2
=
1
s
_

s
dz
1 + z
2
=
1
s
_

2
arctan(s)
_
.
It is given that
tan(

2
) = (tan )
1
which implies
tan
_

2
arctan(s)
_
=
1
s
.
Therefore

2
arctan(s) = arctan(s
1
).
By Fubinis theorem, we conclude
1
s
_

0
e
sx
x
1
sin xdx =
1
s
arctan(s
1
).
_

0
e
sx
x
1
sin xdx = arctan(s
1
).
Problem 2.58
Show that
_
e
sx
x
1
sin
2
x dx =
1
4
log(1 + 4s
2
) for s > 0 by integrating e
sx
sin 2xy with respect to x
and y.
Solution:
We will investigate the integral
15
_

0
_
1
0
e
sx
sin(2xy)dydx.
We want to apply Fubinis theorem, so rst we verify that
_

0
_
1
0
|e
sx
sin(2xy)|dydx
_

0
_
1
0
e
sxy
2xydydx =
_

0
e
sx
xdx < .
Next, we observe that
_

0
_
1
0
e
sx
sin(2xy)dydx =
_

0
e
sx
x
_
1
2

1
2
cos 2x
_
=
_

0
e
sx
sin
2
xdx
x
.
This integral can be computed by switching the order of integration. The rst step is to use
integration by parts to compute
I =
_

0
e
sx
sin(2xy)dx =
_

0
e
sx
s
2y cos(2yx)dx
e
sx
s
sin(2yx)

0
=
_

0
e
sx
s
2y cos(2yx)dx
=
_

0
(2y)
2
e
sx
s
2
sin(2yx)dx
2y
s
2
e
sx
cos(2yx)

0
=
_

0
(2y)
2
e
sx
s
2
sin(2yx)dx +
2y
s
2
.
I =
4y
2
I
s
2
+
2y
s
2
.
I =
2y
s
2
+ 4y
2
.
Therefore
_
1
0
_

0
e
sx
sin(2xy)dxdy =
_
1
0
2y
s
2
+ 4y
2
dy =
1
4
log(s
2
+ 4)
1
4
log(s
2
) =
1
4
log(1 + 4s
2
).
By Fubinis theorem, this proves that
_

0
e
sx
sin
2
xdx
x
=
1
4
log(1 + 4s
2
).
Problem 2.60
(x)(y)/(x + y) =
_
1
0
t
x1
(1 t)
y1
dt for x, y > 0. (Recall that was dened in 2.3. Write
(x)(y) as a double integral and use the argument of the exponential as a new variable of integra-
tion.)
Solution:
16
By denition, we have
(x)(y) =
_

0
_

0
t
x1
s
y1
e
ts
dsdt.
We perform the change of variables s = u uv, t = uv. The Jacobian is
(s, t)
(u, v)
=
s
u
t
v

s
v
t
u
= (1 v)u (uv) = u.
The change of variables formula for multiple integrals yields
(x)(y) =
_
1
0
_

0
(uv)
x1
(u uv)
y1
e
u
ududv =
_
1
0
_

0
v
x1
(1 v)
y1
e
u
u
x+y1
dudv.
By denition of (x + y) and Fubinis theorem, this can be rewritten as
(x)(y) =
_
_
1
0
v
x1
(1 v)
y1
dv
_
(x + y).
This proves that
(x)(y)
(x + y)
=
_
1
0
t
x1
(1 t)
y1
dt.
Problem 2.63
The technique used to prove Proposition 2.54 can also be used to integrate any polynomial over S
n1
.
In fact, suppose f(x) =

n
1
x

j
j
(
j
N{0}) is a monomial. Then
_
fd = 0 if any
j
is odd, and
if all
j
s are even,
_
fd =
2(
1
) (
n
)
(
1
+ +
n
, where
j
=

j
+ 1
2
.
Solution:
By Theorem 2.49, we know that
_
R
n
e
|x|
2
n

j=1
x

j
j
dx =
_

0
_
S
n1
e
r
2
n

j=1
_
x
j
|x
j
|
_

j
r

j
r
n1
ddr.
First we compute the right-hand side:
_

0
_
S
n1
e
r
2
n

j=1
_
x
j
|x
j
|
_

j
r

j
r
n1
ddr =
_
_
S
n1
fd
__
_

0
e
r
2
r
n1+

j
dr
_
=
_
_
S
n1
fd
__
_

0
e
s
2
s

j
2
s
n1
2
s

1
2
ds
_
=
_
_
S
n1
fd
_
1
2
(

j
+ 1
2
).
17
If any
i
is odd, then
_

e
x
2
i
x

i
i
dx
i
= 0 by symmetry, hence then left-hand side is zero:
_
R
n
e
|x|
2
n

j=1
x

j
j
dx =
n

j=1
_

e
x
2
j
x

j
j
dx
j
= 0.
Therefore
_
S
n1
fd = 0.
If every
i
is even, then by symmetry
_

e
x
2
i
x

i
i
dx
i
= 2
_

0
e
x
2
i
x

i
i
dx
i
. The left-hand side is
then
_
R
n
e
|x|
2
n

j=1
x

j
j
dx = 2
n
n

j=1
_

0
e
x
2
j
x

j
j
dx
j
=
n

j=1
_

0
e
s
s

j
2

1
2
ds =
n

j=1
(

j
+ 1
2
).
By combining the identities of the left-hand side and right-hand side, we obtain
_
S
n1
fd =
2(
1
) (
n
)
(
1
+ +
n
)
where
j
=

j
+1
2
.
18