Exercises in Classical Real Analysis

Themis Mitsis

Contents
Chapter 1.

Numbers

Chapter 2.

Sequences, Series and Limits

11

Chapter 3.

Topology

23

Chapter 4.

Measure and Integration

29

CHAPTER 1

Numbers
E 1.1. Let a, b, c, d be rational numbers and x an irrational number such that
cx + d , 0. Prove that (ax + b)/(cx + d) is irrational if and only if ad , bc.
S. Suppose that (ax + b)/(cx + d) = p/q, where p, q Z. Then
(aq cp)x = d p bq, and so we must have d p bq = aq cp = 0, since x is irrational. It
follows that ad = bc. Conversely, if ad = bc then (ax + b)/(cx + d) = b/d Q.

E 1.2. Let a1 a2 an and b1 b2 bn be real numbers. Prove
that

n n
n
X
X X
ai b j n
ak bk

i=1

j=1

k=1

and that equality obtains if and only if either a1 = an or b1 = bn .

S. Since {ai }ni=1 and {bi }ni=1 are both increasing, we have

n n
n
X
X
X X
0
(ai a j )(bi b j ) = 2n
ak bk 2 ai b j .
1i, jn

k=1

i=1

j=1

If we have equality then the above implies (ai a j )(bi b j ) = 0 for all i, j. In particular
(a1 an )(b1 bn ) = 0, and so either a1 = an or b1 = bn .

E 1.3. (a) If a1 , a2 , . . . , an are all positive, then
n n

X X 1
ai
n2
a
i
i=1
i=1
and equality obtains if and only if a1 = a2 = = an .
(b) If a, b, c are positive and a + b + c = 1, then
(1/a 1)(1/b 1)(1/c 1) 8
and equality obtains if and only if a = b = c = 1/3.
S. (a) By the Cauchy-Schwarz inequality we have
n
1/2
!1/2 X
n
X
n 1/2 X

1
1/2 1
.
n=
ai
ai
ai
a
i=1 i
i=1
i=1
(b) Since a + b + c = 1, (a) implies 1/a + 1/b + 1/c 9 and therefore
(1/a 1)(1/b 1)(1/c 1) = 1/a + 1/b + 1/c 1 8.

5

CHAPTER 1. NUMBERS
E 1.4. Prove that for all n N we have
1 3 5
2n 1
1


2 4 6
2n
3n + 1
end equality obtains if and only if n = 1.
S. Note that

2k 1
3k 2

2k
3k + 1
and therefore the product telescopes.

E 1.5. (a) For all n N we have

1
n + 1 n < < n n 1.
n
(b) If n N ad n > 1 then
n
X

1
2 n+12<
< 2 n 1.
k
k=1

S. (a) We have

n+1 n=

n n1=

1
< ,
n+1+ n 2 n
1

1
1
> .

n+ n1 2 n

(b) Sum inequalities (a) for k = 2, 3, . . . , n.

E 1.6. Let n N and x R. Then

(a) 1 < x < 0 implies (1 + x)n 1 + nx + (n(n 1)/2)x2 .
(b) x > 0 implies (1 + x)n 1 + nx + (n(n 1)/2)x2 .


S. Induction on n.
E 1.7. If n N, then n! ((n + 1)/2)n .
S. In the Geometric-Arithmetic Means Inequality, take ak = k.

E 1.8. If b1 , b2 , . . . , bn are positive real numbers, then

n
(b1 b2 bn )1/n .
1
1
1
+
+

b1
b2
bn
S. In the Geometric-Arithmetic Means Inequality, take ak = 1/bk .
E 1.9. If x, y R and n N, then
(a) [x + y] [x] + [y],
(b) [[x]/n] = [x/n],
n1
P
(c) [x + k/n] = [nx].
k=0

CHAPTER 1. NUMBERS
S. (a) [x] + [y] is an integer and satisfies [x] + [y] x + y, therefore [x] + [y]
[x + y].
(b) We claim that [x/n] [x]/n. Indeed, if this were not the case we would have
[x]/n < [x/n] ([x] + )/n, for some 0  < 1. Therefore [x] < n[x/n] [x] + ,
a contradiction since n[x/n] is an integer. It follows that [x/n] [[x]/n]. The converse
inequality is obvious.
(c) Let
n1
X
f (x) =
[x + k/n] [nx].
k=0

Then f is periodic with period 1/n and vanishes on the interval [0, 1/n]. So, f = 0 identically.

E 1.10. (a) If a, b, c are positive real numbers then
!2
1
1
1
1
1
1
a + b + c a2 + b2 + c2
2
3
6
2
3
6
with equality if and only if a = b = c.
(b) If a1 , . . . , an and w1 , . . . , wn are positive real numbers with

n
P

wi = 1 then

i=1

2
n
n
X

X
a2i wi
ai wi
i=1

i=1

with equality if and only if a1 = a2 = = an .

S. (a),(b) Cauchy-Schwarz inequality.
E 1.11. If n N, then
 
n  2
P
n
(a)
= 2n
k
n .
(b)

k=1
2n
P

(1)k

k=1

2n2
k

= (1)n

2n
n .

S. (a) By the Binomial Theorem we have

!
2n
X
2n k
2n
(1 + x) =
x.
k
k=0
But
2n

(1 + x)

= (1 + x)

X n!
=
i
i, j

n
!
! n
X n i X
n j

x
x
(1 + x) =
i j=0 j
i=0
! !
!
2n
n i+ j X k X n n
.
x
x =
i j
j
i+ j=k
k=0
n

Equating the coefficients of xn we get

!
! X
!2
! X ! ! X
n
n
n n
n n
n
2n
=
=
.
=
i j
i ni
i
n
i+ j=n
i=0
i=0
7

CHAPTER 1. NUMBERS
(b) As in (a) we have
(1 x2 )2n =
and
(1 x2 )2n

!
2n
X
2n
(1)k x2k
k
k=0

2n
2n

X 2n!
X 2n!
= (1 x)2n (1 + x)2n =
(1)i xi
x j
i
j
i=0
j=0
! !
4n
X
X 2n! 2n!
X
2n 2n
(1)i
=
(1)i xi+ j =
xk
.
i
j
i
j
i, j
i+ j=k
k=0

Equating the coefficients of x2n we get

!
! ! X
!2
2n
X
n 2n
i 2n 2n
i 2n
(1)
=
(1)
=
(1)
.
n
i
j
i
i+ j=2n
i=0

E 1.12. If m, n N, then 1 +


m 
P
n+k
k

k=1

S.
1+

n+m+1
m

!
!
!!
!
m
m
X
X
n+k
n+k+1
n+k
n+m+1
=1+

=
.
k
k
k1
m
k=1
k=1


E 1.13. Prove Lagranges inequality for real numbers

n
2 n
n
X

X 2 X

X
2
(ak b j a j bk )2 .
ak bk = ak bk
k=1

k=1

1k< jn

k=1

S. We have
X
X
(ak b j a j bk )2 =
(a2k b2j + a2j b2k 2ak b j a j bk ).
1k< jn

But

1k< jn

a2k b2j

1k< jn

and

a2j b2k

n
n
n
X
X 2 X
2

a2k b2k
= ak bk
k=1

1k< jn

X
1k< jn

k=1

k=1

n
2
n
X
X

a2k b2k .
2ak b j a j bk = ak bk
k=1

k=1

The result follows.

E 1.14. Given a real x and an integer N > 1, prove that there exist integers p
and q with 0 < q N such that |qx p| < 1/N.
N
S. For k = 0, 1, . . . , N let ak = kx [kx]. Then {ak }k=0
[0, 1), and therefore
there exist 0 k1 , k2 N such that |ak1 ak2 | < 1/N.


E 1.15. If x is irrational prove that there are infinitely many rational numbers
p/q with q > 0 and such that |x p/q| < 1/q2 .
8

CHAPTER 1. NUMBERS
S. Assume there are finitely many, say, p1 /q1 , . . . , pn /qn . Then, by the preceding exercise, there exists p/q such that |x p/q| < 1/(qN) with q N and 1/N <
min{|x pi /q1 | : 1 i n}. (The minimum is positive because x is irrational.)


CHAPTER 2

Sequences, Series and Limits

E 2.1. Evaluate lim

n
Q

n k=0

(1 + a2 ) where a C.

S. If a , 1, then for all n N we have

n+1
n
Y
1 a2
k
(1 + a2 ) =
.
1a
k=0
Therefore the sequence converges to 1/(1 a) for |a| < 1. It diverges for |a| > 1 or a = 1.
The limit does not exist if |a| = 1 and a , 1.

E 2.2. Evaluate lim

n
P

n k=1

1 .
n2 +k

S. Note that
n

n2 + n

n
X
k=1

1
1.

2
n +k

Therefore the sum converges to 1.

E 2.3. Let x = 2 + 2 and y = 2 2. Then n N implies

(a) xn + yn N and xn + yn = [xn ] + 1.
(b) lim (xn [xn ]) = 1.

S. (a) By the Binomial Theorem, we have

!
n
X n!
X
nk
n k+ nk
nk
n
n
2
x +y =
2
(1 + (1) ) =
2k+1+ 2 N.
k
k
k=0
0kn
nk even

Since x + y 1 < x < x + y , we conclude that [xn ] = xn + yn 1.

(b) By (a), xn [xn ] = 1 yn 0 as n .
n

E 2.4. If {xn }
n=1 R, {yn }n=1 (0, ) and {xn /yn }n=1 is monotone, then the

sequence {zn }n=1 defined by

x1 + + xn
zn =
y1 + + yn
is also monotone.

S. Assume that {xn /yn }

n=1 is increasing and prove inductively that zn zn+1
xn+1 /yn+1 using the fact
a a+c
c
a c

.
b d
b b+d d

11

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

E 2.5. Let 0 < a < b < . Define

x1 = a, x2 = b, x2n+1 = x2n x2n1 ,

x2n+2 =

x2n + x2n1
.
2

Then the sequence {xn }

n=1 converges.
S. Note that [x2n+1 , x2n+2 ] [x2n1 , x2n ] and
x2n x2n1
x2 x1
n1 0.
x2n+2 x2n+1
2
2
Therefore the sequence converges and

\
lim xn =
[x2n1 , x2n ].
n

n=1


E 2.6. Let 0 < a < b < . Define
xn + xn+1
.
2
converges and determine its limit.

x1 = a,
Prove that the sequence {xn }
n=1

x2 = b,

xn+2 =

S. Note that xn+1 xn = (1/2)n1 (x2 x1 ). Therefore

n2
X
1
2 a + 2b
xn = x1 + (x2 x1 ) ( )k a + (b a) =
.
2
3
3
k=0

E 2.7. Let {xn }
n=1 R satisfy 0 < xn < 1 and 4xn+1 (1 xn ) 1 for all n N.
Show that lim xn = 1/2.
n

S. Note that

1
xn .
4(1 xn )
Therefore the sequence is increasing. Since it is bounded, it converges to a limit l which
must satisfy 4l(1 l) 1. We conclude that l = 1/2.

xn+1

E
2.8. Let 1 < a < , x = 1, and xn+1 = a(1 + xn )/(a + xn ). Show that

xn a.
by

S. Prove inductively that the sequence is decreasing and bounded from below
a.


E 2.9. Define x0 = 0, x1 = 1, and

1
n
xn+1 =
xn1 +
xn .
n+1
n+1
Prove that {xn }
n=1 converges and determine its limit.
S. Note that xn+1 xn = (1)n /(n + 1)!, and so
n1
X
1
(1)k
.
xn =
(k + 1)!
e
k=0

E 2.10. Let a R, a < {0, 1, 2} and define x1 = a, xn+1 = 2 2/xn for n N.
Find the limit points of the sequence {xn }
n=1 .
12

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

S. Note that xn+4 = xn for all n N. Therefore the sequence takes on the
values {x1 , x2 , x3 , x4 } only.

E 2.11. For n N, write n = 2 j1 (2k 1) where j, k N and write
1 1
Sn = + .
j k
Find all limit points of the sequence {S n }
n=1 . Evaluate lim S n and lim S n .
S. Let A be the set of limit points of {S n }
n=1 . We claim that A = {0} {1/n :
n N}. Indeed, let nk = 2k1 (2k 1) and m p,k = 2 p1 (2k 1).Then
2
1
1 1
0, S m p,k = +
as k .
k
p k
p
Hence A {0} {1/n : n N}. Now take l A, l , 0. Then there exists a subsequence
jm 1
{S nm }
(2km 1). Note that at least one of the
m=1 such that S nm l. Write nm = 2
sets { jm : m N}, {km : m N} is unbounded, and so we may assume, without loss of
generality, that there exists { jmi }
i=1 with jmi . Then, since S nmi l, we have kmi
1/l. Therefore {kmi }
i=1 is eventually constant and l {1/kmi : i N}. lim S n = inf A = 0,
lim S n = sup A = 1.

S nk =

E 2.12. Prove that (n/e)n < n! for all n N.

S. Induction on n. It is clearly true for n = 1. Assuming (n/e)n < n! we have
!n+1
!n
n+1
n+1
1  n n n + 1
<
=
1+
e n! = (n + 1)!.
e
e
n
e
e

E 2.13. Evaluate
(a) lim ((2n)!/(n!)2 )1/n ,
n

(b) lim (1/n)[(n + 1)(n + 2) (n + n)]1/n ,

n

(c) lim [(2/1)(3/2)2 (4/3)3 ((n + 1)/n)n ]1/n .

n

S. Let
an =

(n + 1)(n + 2) (n + n)
,
nn
! !2 !3
!n
2 3
4
n+1
cn =

.
1 2
3
n

(2n)!
,
(n!)2

Then

bn =

bn+1  n n (2n + 1)(2n + 2)

4
=
,
bn
n+1
e
(n + 1)2
!n+1
1
= 1+
e.
n+1

an+1 (2n + 1)(2n + 2)

=
4,
an
(n + 1)2
cn+1
cn
Therefore

an 4,

pn
4
bn ,
e
n

E 2.14. Evaluate lim n ( n 1)n .

13

n
cn e.


CHAPTER 2. SEQUENCES, SERIES AND LIMITS

S. Since n n 1, there exists n0 N such that 0 < n n 1 < 1/2 for all
n

n n0 , and so 0 < ( n 1)n < (1/2)n . Therefore 0 lim( n n 1)n lim( n n)n 0. We

conclude that lim ( n n 1)n = 0.


n

1/n
E 2.15. If {xn }
x.
n=1 (0, ) and xn x, then (x1 xn )

S. By the Harmonic-Geometric-Arithmetic Means Inequality we have

n
x1 + + xn
(x1 xn )1/n
.
1
1
n
+

+
x
x
1

Therefore (x1 xn )

1/n

x.

E 2.16. (a) Let S n =


n
P

1/k for n N. Then lim |S n+p S n | = 0 for all p N,

n

k=1

but {S n }
n=1 diverges to .
(b) Find a divergent sequence {xn }
n=1 in R such that lim |xn2 xn | = 0.
n

S. (a) |S n+p S n | = 1/(n + 1) + + 1/(n + p) p/(n + 1) 0

k(n)
k(n)+1
(b) For n 4 let k(n) be the unique integer such that 22 n < 22
and define
k(n)
P
1/ j. Note that k(n) and k(n2 ) = k(n) + 1. Therefore xn and |xn2 xn | =
xn =
j=1

1/(k(n) + 1) 0.


P

E 2.17. There exist two divergent series an and bn of positive terms with
P
a1 a2 and b1 b2 such that if cn = min{an , bn }, then cn converges.
S. Let
ak = 1/2k ,

bk = 1/2n

if 2n k < 2n+1 , n even

ak = 1/2n ,

bk = 1/2k

if 2n k < 2n+1 , n odd.

and

E 2.18. Evaluate the sums

P
(a)
1/(n(n + 1)(n + 2)),
(b)

n=1

(n 1)!/(n + p)!, where p N is fixed.

n=1

S. (a) Note that

"
#
1
1
1
1
=

.
n(n + 1)(n + 2) 2 n(n + 1) (n + 1)(n + 2)
Consequently
"
#
n
X
1 1
1
1
1
=

.
k(k + 1)(k + 2) 2 2 (n + 1)(n + 2)
4
k=1
(b) We have
"
#
(n 1)!
1
1
1
1
=
=

.
(n + p)! n (n + p)
p n (n + p 1) (n + 1) (n + p)
Therefore
"
#
n
X
1 1
1
1
(k 1)!
=

.
(k + p)!
p p! (n + 1) (n + p)
p p!
k=1
14

CHAPTER 2. SEQUENCES, SERIES AND LIMITS


P

E 2.19. Let an be a convergent series of nonnegative terms. Then

(a) lim nan = 0,
(b) possibly lim nan > 0,
(c) if an an+1 for all n > n0 , then lim nan = 0.
S. (a) Suppose that lim nan > c > 0 for some c. Then there exists n0 N such
that nan > c for n n0 . Consequently,
N
X

an > c

n=n0

N
X
1

n
n=n

as n ,

a contradiction.
(b) Let ak = 1/2k if k , 2n and ak = 1/2n if k = 2n . Then
N
N

X
X
X
X
X 1
X
1
1
ak =
ak +
ak
+
<

2
n
k
k
2
2
2
n:2n N
k=1
k,2n
k=2n
k=1
k=1
and lim 2n a2n = 1.
n
(c) Note that
2n
X

na2n

ak 0 and

na2n+1

k=n+1

2n+1
X

ak 0.

k=n+2

Therefore lim 2na2n = lim (2n + 1)a2n+1 = 0. We conclude that lim nan = 0.
n

E 2.20. If {cm }
m=1 [0, ] and
bn =
then

n
X
1
mcm ,
n(n + 1) m=1

bn =

n=1

S. Define
am,n
Then

mcm

n(n+1)
=

am,n =

n=1 m=1

cm .

m=1

if 1 m n,
if m > n.

X
X
1
mcm =
bn .
n(n + 1) m=1
n=1

n=1

On the other hand

X

am,n =

m=1 n=1

X
1
=
cm .
n(n + 1) m=1
n=m

mcm
m=1


E 2.21. (a) Prove that

1/n2 < 2.

n=1

(b) Prove that

X
X

= .

2
(m + n)
m=1 n=1

15

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

S. (a) We have
!

X
X
X
X
1
1
1
1
1
=1+

= 2.
=1+
<1+
n(n 1)
n1 n
n2
n2
n=1
n=2
n=2
n=2
(b) We have
X

X
m=1 n=1

!
X

X
X
1
1
1
1
=

(m + n)2 m=1 n=m+1 n2 m=1 n=m+1 n n + 1

=

1
= .
m+1
m=1


E 2.22. Let b be an integer > 1 and let d be a digit (0 d < b). Let A denote
the set of all k N such that the b-adic expansion of k fails to contain the digit d.

P
(a) If ak = 1/k for k A and ak = 0 otherwise, then
ak < .
k=1

(b) For n N let A(n) denote the number of elements of A that are n. Then
lim (A(n)/n) = 0.

S. Let
An = {k : k is an n-digit number and does not contain the digit d}
= {k : bn1 k < bn } A.
Note that |An | = (b 2)(b 1)n1 .
(a) We have

ak =

k=1

X
X
n=1 kAn

(b) If b , 2 then

A(n)

!n1

X
X
|An |
b1
ak
= (b 2)
< .
b
bn1
n=1
n=1

|Ak | = (b 2)

k:bk1 n

(b 1)k n1/ logb1 b 1.

k:bk n

If b = 2 then A(n) = |{k : 2k n}| log2 n. Therefore lim (A(n)/n) = 0.

n

E 2.23. Let 0 < x < 1. Then x has a terminating decimal expansion if and only
if there exist nonnegative integers m and n such that 2m 5n x is an integer.
S. If x has a terminating decimal expansion, then x = p/10k = p/(2k 5k ).
Conversely, if 2m 5n x = N N for some, say, m n, then x = 2nm N/10n .

E 2.24. Evaluate lim (n!e [n!e]).
n

S. Let S n =

n
P

1/k!. Then, using the error estimate for the tail, we have

k=0

0 < n!e n!S n < 1/n. We conclude that [n!e] = n!S n and therefore n!e [n!e] 0.
E 2.25. Show that lim n sin(2en!) = 2.
n

16

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

S. Since lim (en! [en!]) = 0 we have
n

sin(2en! 2[en!])
sin(2en!)
= 1 lim
= 2.
n en! [en!]
2en! 2[en!]
Note that the error estimate for the Maclaurin series expansion of e implies
1/(n + 1) < en! [en!] < 1/n, and so lim n(en! [en!]) = 1. It follows that
lim

n sin(2en!) = n(en! [en!])

sin(2en!)
2.
en! [en!]


E 2.26. Find the sum of the series

X
1
.

n=1 (n + 1) n + n n + 1
S.

X
n=1

X
n
n + 1

= 1.
=

n+1
(n + 1) n + n n + 1 n=1 n
1


E 2.27. Let an > 0 for each n N. Then

P
P
(a)
an < implies
an an+1 < ,
n=1

n=1

(b) the converse of (a) is false,

P
P
1 1
(c)
an < implies (a1
< ,
n + an+1 )
n=1

n=1

(d) the converse of (c) is false.

S. By the Harmonic-Geometric-Arithmetic Means Inequality, we have

1
1 1
an an+1 (an + an+1 ),
2(a1
n + an+1 )
2
proving (a) and (c). For (b) and (d), let an = 1/n if n is even and an = 1/n3 if n is odd.
E 2.28. Suppose that dn > 0 for all n N and

= . What can be said of

n=1

the following series?

P
(a)
dn /(1 + dn ),
(b)
(c)

n=1

P
n=1

P
n=1

dn /(1 + ndn ),
dn /(1 + dn2 ).

S. (a) If {dn }
n=1 is bounded then 1/(1 + dn ) is bounded from below, therefore

X
n=1

X
dn
C
dn = .
1 + dn
n=1

If {dn }
n=1 is unbounded then there exists a subsequence {dkn }n=1 with dkn . Therefore

P
there exists n0 such that dkn /(1+dkn ) > 1/2 for all n n0 . Consequently
dn /(1+dn ) = .
n=1

17

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

(b) Let dn = 1 for all n N. Then

X
X
dn =
n=1

n=1

dn
= .
1 + ndn

Let dk = 1/2k if k , 2n and dk = 2n if k = 2n . Then

k+2
dk

k
=

2n
1 + kdk 1+4
n
Therefore

dn = and

n=1

if k , 2n ,
if k = 2n .

dn /(1 + ndn ) < .

n=1

(c) Let dn = 1 for all n. Then

n=1

dn /(1 + dn2 ) = . Let dn = n2 . Then

P
n=1

dn /(1 + dn2 ) <

E 2.29. Let 0 < a < b < and define x1 = a, x2 = b, and xn+2 =

n N. Find lim xn .

xn xn+1 for

S. Let yn = log xn and use Exercise 2.6.


2(xn1

1
y1
n ) ,

E 2.30. Let 0 < a < b < and define x1 = a, y1 = b, xn+1 =

+

and yn+1 = xn yn . Then {xn }

and
{y
}
both
converge
and
have
the
same
limit.
n n=1
n=1

S. Prove inductively, using the Harmonic-Geometric Means Inequality, that

1
a < xn xn+1 yn+1 yn < b and yn+1 xn+1 (yn xn ).
2
E 2.31. Show that if
x (0, 1) there is a subseries

P
k=1

ak = 1 and 0 < an

k=1

, n = 1, 2, . . . , then for every

k=n+1

ank whose sum is x.

S. Note that, since the sum of the series is 1 and x (0, 1), there exists n1 N
such that

X
X
ak > x and
ak x
k=n1 +1

k=n1

implying

ak > x an1

and an1 x.

k=n1 +1

Therefore there exists n2 > n1 such that

X
ak > x an1

X
and

x an1 .

k=n2 +1

k=n2

Continuing this way, we can find a sequence of integers n1 < n2 < such that
m

X
X
0 x
ank <
ak .
k=nm +1

k=1

Letting m , we conclude that

P
k=1

ank = x.
18

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

E 2.32. Show that if an , bn R, (an +bn )bn , 0, n = 1, 2, . . . , and both
and

(an /bn )2 converge, then

n=1

an /bn

n=1

an /(an + bn ) converges.

n=1

S. Choose k0 N such that |1 + ak /bk | 1/2 for all k k0 . Then

1
2
.

|ak bk + b2k | |bk |2

Note that

n
X
k=k0

and

n
n
X
a2k
ak X
ak
=

ak + bk k=k bk k=k ak bk + b2k

0
0
n 
X



k=k0

We conclude that


2
n 
X

 ak  .

2

 bk 
ak bk + b2k 
k=k0
a2k

an /(an + bn ) converges.

n=1

E 2.33. Show that if bn & 0 and

R such that an /bn 1 as n and

P
n=1

bn = , then there is a sequence {an }

n=1

(1)n an diverges.

n=1

S. Let
Sn =

n
X

bk ,

an = bn + (1)n

k=1

bn
.
Sn

Note that an > 0 for large n and

m
m
m
X
X
X
bn
(1)n an =
(1)n bn +
.
S
n=1
n=1
n=1 n
The first series in the above sum converges, being alternating, while the second diverges by

P
Abels Theorem. Therefore
an diverges. On the other hand, an /bn = 1 + (1)n /S n 1
n=1

as n .

E 2.34. Show that if n 2, then

(1 (1 2k )n ) ' log n.

k=1

S. Note that
!m
Z 1
Z 11/2k+1

X
X
1
1
1
=
xm dx =
xm dx
1

k
m+1
2k
2k
0
k=0 11/2
k=1
and similarly

!m

X
1
1
2
.
1

k
k
m
+1
2
2
k=1

Therefore

X
k=1

1
1 1 k
2

!n !

!m X
!m
X
n1
n1 X

X
1
1
1
1
=
1 k =
1 k
2k
2
2k
2
m=0 k=1
k=1 m=0
'

n
X
1
' log n.
m
m=1

19

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

R

E 2.35. Show that if rn R, then lim

n 0

S.


ex (sin(x + rn ))n dx = 0.

|ex (sin(x + rn ))n |dx =

ex | sin(x + rn mod 2)|n dx

Z
rn mod 2
ex | sin(x)|n dx
=e
rn mod 2
Z
2
e
ex | sin(x)|n dx.

Note Rthat | sin(x)|n 0 almost everywhere, and so, by the Dominated Convergence Theo
rem, 0 ex | sin(x)|n dx 0.

E 2.36. Let f : [0, 1] R be defined by
x log x

if 0 < x < 1,

x1
f (x) =
0
if x = 0,

1
if x = 1.
Show that

f (x)dx = 1
0

1
.
1)

n2 (n

n=2

S. Note that

x log x X x(1 x)n

=
x1
n+1
n=0

and the convergence is uniform on [0, 1] by Weierstrass M-test. Therefore

Z 1
Z 1

X
X
1
1
x(1 x)n dx =
f (x)dx =
n+1 0
(n + 1)2 (n + 2)
0
n=0
n=0
=1

1
.
n2 (n 1)

n=2


E 2.37. Show that
lim

Conclude that

j=n

= log k.

X
(1) j+1
j=1

S. Note that

j+1
j

Therefore

kn
X
1

Z
n

kn+1

dx 1

x
j

= log 2.
Z

dx X 1

x
j
j=n
kn

20

j+1
j

Z
n

dx
.
x1

kn+1

dx
,
x1

CHAPTER 2. SEQUENCES, SERIES AND LIMITS

and consequently
log k +

! X
!
kn
1
1
k

log k +
.
n
j
n1
j=n

Taking the limit as n , we obtain the first assertion. To prove the second assertion,
note that
2n
n
2n
2n
2n
X
X
X
1
1 X1 1
(1) j+1 X 1
=
2
=
=
log 2, as n .
j
j
2 j j=n+1 j
j n
j=n
j=1
j=1
j=1
P

j+1
P
converges by Leibniz, we conclude that j=1 (1)j = log 2.

R

2
2
E 2.38. Show that e x /2 x et /2 dt is a decreasing function of x on [0, ) and
that its limit as x is 0.

Since

j=1

(1) j+1
j

S. By LHospitals Rule we have

R 2
2
et /2 dt
ex
1
x
lim
= lim
= lim = 0.
x xex2 /2
x x
x
ex2 /2
Now let
Z
Z
2
ex /2
2
x2 /2
t2 /2
g(x) = e
e
dt and h(x) =

et /2 dt.
x
x
x
Then
Z
2
ex /2
2
2
g0 (x) = xe x /2
et /2 dt 1 and h0 (x) = 2 < 0.
x
x
Hence h is strictly decreasing. Note that lim h(x) = 0. therefore h(x) > 0 and consequently
x
g0 (x) < 0.


21

CHAPTER 3

Topology
E 3.1. Let X be a 2nd countable space. Show that if {Gi }iI is an arbitrary
S
family of open sets in X then there exists a countable subset J I such that iI Gi =
S
iJ G i .
S. Suppose {Uk }kN is a basis for the topology of X. Let
K = {k N : i(k) I such that Uk Gi(k) }
and put J = {i(k) : k K}.

E 3.2. Let X be a 2nd countable space, and let A X be an uncountable set.

Prove that A has at least one condensation point.
S. Suppose that for each x A there is an open set U x X with x U x and
S
|A U x | 0 . Since X is 2nd countable there exists {xn }
xA U x =
n=1 A such that
S
S
(U

A)
and
therefore
U

A
must
be
uncountable
for some
U
.
Hence
A
=
x
x
x
n
n0
n
n=1
n=1
n0 , a contradiction.

E 3.3. If X is a 2nd countable space and A is a closed subset of X, then there
exist a perfect set P and a countable set N, such that A = P N. Conclude that any subset
of a 2nd countable space can have only countably many isolated points.
S. Let P = {x X : for each nbd U x of x, U x A is uncountable}. Using the
preceding exercise, P is perfect and A \ P is countable.

E 3.4. Prove the following assertions.
(a) If A is nonempty perfect subset of a complete metric space then A is uncountable.
(b) Any countable closed subset of a complete metric space has infinitely many isolated points.
(c) There exists a countable closed subset of R having infinitely many limit.points.
S. Suppose X is a complete metric space.
(a) Note that since A is a closed subset of X, it is complete as a metric space. If A is
countable then by the Baire category theorem, at least one of its points must be isolated.
(b) Assume that there exists a countable closed subset of X with finitely many isolated
points. Removing these points results in a countable perfect set, contradicting (a).
(c) Take infinite copies of a convergent sequence together with its limit.

E 3.5. It is impossible to express [0, 1] as a union of disjoint closed nondegenerate intervals of length < 1.
S
S. Suppose [0, 1] = iI [xi , yi ], where {[xi , yi ]}iI is disjoint. Note that I must
be countable. Then the set of endpoints ({xi : i I} {yi : i I}) \ {0, 1} is a countable
perfect set, a contradiction by the preceding exercise.

23

CHAPTER 3. TOPOLOGY
E 3.6. It is impossible to express [0, 1] as a countable union of disjoint closed
sets.
S
S. Suppose [0, 1] =
n=1 F n with the F n s closed and pairwise disjoint. Since
F1 F2 = , we can find a closed interval I1 such that I1 F1 = , I1 F2 , , I1 \ F2 , .
We repeat the same procedure inside I1 with I1 F2 playing the role of F1 and I1 Fk
playing the role of F2 , where Fk is the first set in the sequence {Fn }
n=3 intersecting I1 . We
thereby construct a decreasing sequence of closed intervals {In }
n=1 such that In F n = , a
contradiction.

E 3.7. Let A be a bounded subset of R which is not closed. Construct explicitly
an open cover of A that has no finite subcover.
S. Let x R \ A be a point such that (x , x + ) A , for all  > 0. For
each n choose xn (x 1/n, x + 1/n) A. Without loss of generality we may assume that

{xn }
n=1 is monotone. If x1 < < xn < x, consider the cover {(, xn )}n=1 {(x, )}.


If x1 > > xn > x, then take the covering {(xn , )}n=1 {(, x)}.
E 3.8. Let (X, ) be a metric space and A, B X disjoint closed sets. Show
that there exists a continuous function f : X R such that f | A = 0 and f | B = 1.
S. Let

(x, A)
.
(x, A) + (x, B)
Then f is well-defined and has the required properties.
f (x) =

E 3.9. If X is a connected metric space with at least two points, then X is

uncountable.
S. Let x, y X be two distinct points. By the preceding exercise, there exists
a continuous function f : X R with f (x) = 0 and f (y) = 1. Since X is connected,
f has the intermediate value property. Therefore [0, 1] f (X). We conclude that X is
uncountable.

E 3.10. Let S be a nonempty closed subset of R and let f : S R be continuous. Then there exists a continuous g : R R such that f (x) = g(x) for all x S and
sup xR |g(x)| = sup xS | f (x)|. This is false for every nonclosed S R.
S
S. Write R \ S =
n=1 In , where the In s are disjoint open intervals and extend
f on each In linearly (if (, a) or (a, ) appear among the In s take f to be constant
on these intervals). If S is not closed we can find a point x < S and, say, an increasing
sequence x1 < < xn < < x of points in S such that limn xn = x. Any continuous
function f on R \ {x}, and therefore on S , with f (xn ) = n cannot be extended to the whole
line.

E 3.11. Let X be a topological space, Y a metric space, f : X Y an arbitrary
function and define A f = {x X : f is continuous at x}.
(a) Prove that A f is a G set.
(b) Assume that there exists a set D X such that D and X \ D are both dense in
X. Prove that for any G set G X there exists a function f : X R such that
A f = G.
(c) Show that there is no function f : R R which is continuous at each rational
and discontinuous at each irrational.
24

CHAPTER 3. TOPOLOGY
(d) Construct explicitly a function f : R R which is continuous at each irrational
and discontinuous at each rational.
P. For any point x X define the oscilation of f at x by
w f (x) = inf{diam( f (U)) : U is a nbd of x}.
(a) Note that x A f if and only if w f (x) = 0. Therefore

\
Af =
{x X : w f (x) < 1/n}.
n=1

The sets in the intersection are open, hence A f is G .

T
(b) Write G =
n=1 G n where each G n is open and X = G 1 G 2 . Define
f : X R by

0
if x G,

f (x) =
.
1/n
if x D (Gn \ Gn+1 ),

1/n if x (X \ D) (Gn \ Gn+1 )

(c) If such a function existed, Q would be G by (a).
(d)

1/n if x = m/n, (m, n) = 1,

.
f (x) =

0
if x is irrational

E 3.12. Construct a strictly increasing function that is continuous at each irrational and discontinuous at each rational.
S. Let {rn : n N} be an enumeration of the rationals and define f : R R
P
by f (x) = rn <x 1/2n . Note that f (rn ) = f (rn ) = f (rn +) 1/2n and f (x) = f (x) = f (x+)
for all x + inR \ Q.

E 3.13. Let X be a topological space and (Y, ) a metric space. Suppose that
{ fn }
n=1 is a sequence of continuous functions from X into Y and that f : X Y is some
function such that limn fn (x) = f (x) for all x X.
(a) Show that there exists a set E X that is of 1st category in X such that f is
continuous at each point of X \ E. In particular, if X is a complete metric space,
then f is continuous at every point of a dense subset of X.
(b) f 1 (V) is an F set in X for every open V Y.
(c) There is no sequence { fn }
n=1 of continuous real functions on R such that fn (x)
1 for x Q and fn (x) 0 for x R \ Q.
(d) Show that Q , the characteristic function of Q, is the pointwise limit of a sequence of functions, so that each of them is the pointwise limit of a sequence of
continuous functions.
S. (a) Let Ak,m = {x X : ( fm (x), fn (x)) 1/k for all n m}. Then each Ak,m
is closed, and so Ak,m \ Ak,m is nowhere dense. Now let
G=

Ak,m , E =

k=1 m=1

[
k=1 m=1

(Ak,m \ Ak,m ).

S
Then E is of 1st category, X \ G E (since X =
m=1 Ak,m for all k), and each x G is a
point of continuity of f .
S S
(b) f 1 (V) =
m=1 {x X : ( fn (x), Y \ V) 1/k for all n m}.
k=1
25

CHAPTER 3. TOPOLOGY
(c) If such a sequence existed then the characteristic function of Q would be continuous
at some point by (a).
(d) Let : R R be defined by (x) = |2x 2k 1| for x [k, k + 1], k Z. Then


lim lim (m!x)n = Q (x) for all x R.
m n


E 3.14. Every compact metric space X is the continuous image of the Cantor
space {0, 1}N .
S. Construct inductively a family of nonempty closed sets {Bs } s{0,1}< such
that
lim diam(Bk ) = 0 for all {0, 1}N ,

Bs = X for all n N,

Bs = Bsa 0 Bsa 1 for all s {0, 1}< .

|s|=n

We give the first step. Using compactness, we can find a number N and a covering
{F1 , . . . , F2N } of X by closed sets such that diam(Fi ) 1/2diam(X) for all i. From these
sets construct all Bt s with |t| N. Repeat the same procedure inside each compact space
Bs with |s| = N. Now define f : {0, 1}N X by

\
f () =
Bn .
n=1


E 3.15. Construct an example of a two-to-one function f : [0, 1] R. Prove
that no such f can be continuous on [0, 1].
S. Let {rn : n N} be an enumeration of the rationals in [0, 1] and define
f : [0, 1] R by

|2x 1| if x is irrational,

f (x) =
r2k1
if x = r2k1 ,

r2k1
if x = r2k .
Suppose now that f is a continuous two-to-one function. We can then assume that its, say,
minimum is attained at the points x1 < x2 , and x2 is not an endpoint. Choose disjoint
closed intervals [a1 , b1 ], [a2 , b2 ] with x1 [a1 , b1 ], x1 , b1 and x2 (a2 , b2 ). Then the
intermediate value theorem implies that a value r with min{ f (b1 ), f (a2 ), f (b2 )} > r > min f
is taken on at least three times.

E 3.16. Suppose that f : [a, b] R satisfies f 1 ({y}) is closed for all y R
and f ([c, d]) is connected for all [c, d] [a, b]. Prove that f is continuous.
S. Let x [a, b] and take {xn }
n=1 [a, b] such that xn x. Then I =
f
([x
,
x])
is
an
interval
containing
f
(x).
We claim that I = { f (x)} and therefore
n
n=1
f (xn ) f (x). Indeed, take f (y) I. Then there exist tn [xn , x] such that f (tn ) =
f (y). Hence tn x and tn f 1 ({ f (y)}). Since f 1 ({ f (y)}) is closed, it follows that
x f 1 ({ f (y)}), and so f (x) = f (y).


E 3.17. Let (X, ) be a metric space. Then there exists a continuous f : X R

that is not uniformly continuous on X if and only if there exist two nonempty disjoint closed
sets A and B such that dist(A, B) = 0.
26

CHAPTER 3. TOPOLOGY
S. Suppose that A and B are disjoint closed sets with dist(A, B) = 0. Define
f : X R by
(x, A)
.
f (x) =
(x, A) + (x, B)
Then f is continuous but not uniformly continuous. Now if f is a real continuous function
on X which is not uniformly continuous, then we can inductively choose points xn , yn X

such that (xn , yn ) < 1/n, | f (xn ) f (yn )| 0 , for a certain 0 , and {xn }
n=1 {yn }n=1 = .
The sets {xn : n N} and {yn : n N} have the required properties.

E 3.18. Let f : R R be continuous and satisfy | f (x) f (y)| c|x y| for all
x, y R, where c does not depend on x and y. Then f (R) = R.
S. Note that f is one-to-one and that
| lim f (x)| = | lim f (x)| = .
x


E 3.19. Let f : R R be arbitrary. Show that the set E of x R such that f
has a simple discontinuity at x is at most countable.
S. Suppose that E is uncountable. Then at least one of the sets A = {x :
f (x+) , f (x)} and B = {x : f (x+) = f (x), f (x) , f (x+)} must be uncountable.
Without loss of generality, we may assume that A is uncountable, and so there exists a
number 0 such that the set {x : | f (x+) f (x)| > 0 } is uncountable and therefore has

a point of accumulation a. Then we can find two sequences {xn }

n=1 and {yn }n=1 such that
xn a, yn a and | f (xn ) f (yn )| 0 /2, contradicting the fact that lim xa f (x) exists. 
E 3.20. If f : R R has a local maximum at each x R, then f (R) is
countable.
S. For every a f (R) choose xa R with f (xa ) = a and an open interval Ia
with rational endpoints such that xa Ia and for each x Ia , f (x) f (xa ) = a. Then the
function
f (R) 3 a 7 Ia {(p, q) : p, q Q}
is one-to-one.

E 3.21. Let R \ Q. Then the set A = {m + n : m, n Z} is dense in R.
S. Note that all the elements of A are distinct since is irrational. So, the
set {m [m] : m N} is an infinite subset of [0, 1] and therefore has a limit point.
Consequently, there exists {rn } A with 0 < rn 0. Now let x > 0,  > 0. Choose n N
with rn <  and let m be the smallest integer such that mrn > x. Then (m 1)rn x and so,
0 < mrn x rn < .


27

CHAPTER 4

Measure and Integration

R

1
E
4.1.
Let
{
}
be
an
approximate
identity
in
L
(R)
(that
is,

0,
n =
n
n
n=1
R
1, limn |t| n (t)dt = 0 for all > 0). Show that limn kn k p = for all p > 1.
S. Let M > 0. Then there exists n0 N such that for all n n0
Z

Z
3/4

n (t)dt
|t|1/(8M)

n (t)dt +

1/4 +

n (t)dt

[n M]

{t:|t|1/(8M)}[n M]

n (t)dt.

[n M]

It follows that
Z
n (t)dt 1/2
[n M]

and therefore
Z

Z
np (t)dt

Z
np1 (t)n (t)dt

[n M]

p1

n (t)dt 1/2M p1 .

[n M]

We conclude that kn k p .

E 4.2. Let A R be a measurable set with |A| > 0. Then for any n N, A
contains arithmetic progressions of length n.
S. Let x0 be a point of density of A. Choose 0 > 0 such that n0 < 1/8. Then
there exists l > 0 such that |(x0 l0 , x0 + l0 ) A| 2(1 0 )l0 for all 0 < l0 l. Now choose
 > 0 such that n2  < 1/8l. Then for k = 0, 1, . . . , n 1 we have
|(x0 l, x0 + l) k + A| = |k + (x0 l k, x0 + l k) A|
|k + (x0 l + n, x0 + l n) A|
= |(x0 l + n, x0 + l n) A|
2(1 0 )(l n).
29

CHAPTER 4. MEASURE AND INTEGRATION

Hence



 
n1
n1
\


 [
(x0 l, x0 + l) \
k + A =  (x0 l, x0 + l) \ k + A

  k=0

k=0

n1
X

|(x0 l, x0 + l) \ k + A|

k=0

n1
X

(2l |(x0 l, x0 + l) k + A|)

k=0

n1
X

(2l 2(1 0 )(l n))

k=0

< 2n2 + 2n0 l < l/4 + l/4 = l/2.

T
Tn1
Therefore | n1
k=0 k + A| > 0. In particular, there exists x
k=0 k + A, and so, x, x
, . . . , x (n 1) A.

E 4.3. Let A be a measurable set of reals with arbitrarily small periods (there
exist positive numbers pn with pn 0 so that pn + A = A for all n). Then either A or its
complement has measure zero.
S. Suppose that |A| > 0 and |A{ | > 0. Let x1 be a point of density of A and x2
a point of density of A{ with x1 < x2 . Then there exists > 0 such that
|(x1 , x1 + ) A| 3/2,

|(x2 , x2 + ) A{ | > 3/2.

It follows that
|(x2 , x2 + ) x2 x1 + A| = |(x2 x1 ) + (x1 , x1 + ) A|
= |(x1 , x1 + ) A| 3/2.
Consider the function : [0, ) R defined by
(x) = |(x2 , x2 + ) x + A|.
Then is continuous and therefore constant, since it is constant on the dense set {mpn :
m, n N}. Therefore
|(x2 , x2 + ) A| = |(x2 , x2 + ) x2 x1 + A| 3/2.
But this is impossible since |(x2 , x2 + ) A{ | 3/2.

E 4.4. Let f : R R be a measurable function with periods s and t whose

quotient is irrational. Prove that f is constant a.e.
S. Note that since s/t is irrational, the set {ns + mt : m, n Z} is dense in R.
Therefore the set f 1 ([a, b]) has arbitralily small periods and hence has either full or zero
measure for all a < b. If it has zero measure for all a < b then f = + or f = almost
everywhere. Suppose that f 1 (I1 ) has full measure for some interval I1 . Divide I1 into two
subintervals of equal length. Then the inverse image of one of these subintervals must have
full measure. Call this interval I2 . Continuing this way we obtain a decreasing sequence
T
I1 I2 of closed intervals whose length tends to zero. Let {r} =
n=1 In . Then the
T 1
1
set f ({r}) = n=1 f (In ) has full measure and therefore f = r almost everywhere.

30

CHAPTER 4. MEASURE AND INTEGRATION

E 4.5. Let A, B R be measurable sets of positive measure. Show that A B
contains an interval.
S. Let x1 be a point of density of A and x2 a point of density of B. Then there
exists > 0 such that
|(x1 , x1 + ) A| 3/2,

|(x2 , x2 + ) B| 3/2.

It follows that
|(x2 , x2 + ) x2 x1 + A| = |(x2 x1 ) + (x1 , x1 + ) A|
= |(x1 , x1 + ) A| 3/2.
Therefore |(x2 , x2 + ) (x2 x1 + A) B| > 0. Now consider the function
(x) = |(x2 , x2 + ) (x + A) B|.
Then is continuous and (x2 x1 ) > 0. Hence there is an interval I such that (x) > 0
for all x I. It follows that (x + A) B , for all x I and so, I B A.

E 4.6. Suppose (X, R) is a -finite measure space and let f : X C be a
measurable function such that | f g| < for all g L p (X). Show that f Lq (X) where q
is the exponent conjugate to p.
S
R S. Write X = k=1 Ak with Ak disjoint and (Ak ) < . Suppose that f 0,
f q = and let Bn = [2n f < 2n+1 ], n Z. Then
Z
Z
X
Z
X
X
q
q
=
f =
f =
fq
.

n= k=1 Bn Ak

X
qN(i)

Bn

n=

2qn (Bn Ak ) =

n= k=1

(BN(i) A M(i) ).

i=1

Let
Sn =

n
X

2qN(k) (BN(k) A M(k) )

k=1

and
g=

X
2qN(i)/p
i=1

Then

Z
gp =

S N(i)

BN(i) AM(i) .

X
2qN(i)

p (BN(i) A M(i) ) <

S N(i)
by Abels Theorem. On the other hand
Z
Z

X
X
2qN(i)/p
2qN(i)/p 2N(i)
fg =
f
(BN(i) A M(i) )
S N(i) BN(i) AM(i)
S N(i)
i=1
i=1
i=1

X
2qN(i)
i=1

S N(i)

(BN(i) A M(i) ) =


By Abels Theorem again.

31