July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Chapter 1

Sequences and Limits

{ }
• Let a n be a sequence of real or complex numbers. A necessary and sufficient
condition for the sequence to converge is that for any  > 0 there exists an
integer N > 0 such that

| a p − aq | < 
holds for all integers p and q greater than N. This is called the Cauchy criterion.
• Any monotone bounded sequence is convergent.
{ }
• For any sequence a n the inferior limit and the superior limit are defined by
the limits of monotone sequences
{ }
lim inf a n = lim inf a n , a n+1 , ...
n→∞ n→∞

and
{ }
lim sup a n = lim sup a n , a n+1 , ...
n→∞ n→∞

respectively. Note that the inferior and superior limits always exist if we adopt
±∞ as limits.
{ }
• A bounded sequence a n converges if and only if the inferior limit coincides
with the superior limit.

1
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

2 Problems and Solutions in Real Analysis

Problem 1. 1
Prove that n sin (2n!eπ) converges to 2π as n → ∞.

Problem 1. 2
Prove that the sequence
( )n ( )n ( )n
1 2 n
+ + ··· +
n n n
converges to e/(e − 1) as n → ∞.

Problem 1. 3
Prove that the sequence
( )
e n/4 n −(n+1)/2 11 · 2 2 · · · n n 1/n
converges to 1 as n → ∞.

This was proposed by Cesàro (1888) and solved by Pólya (1911).

Problem 1. 4
Suppose that a n and bn converge to α and β as n → ∞ respectively. Show
that the sequence
a 0 bn + a1 bn−1 + · · · + a n b 0
n
converges to αβ as n → ∞.

Problem 1. 5
{ }
Suppose that a n n≥0 is a non-negative sequence satisfying

a m+n ≤ a m + a n + C
for all positive integers m, n and some non-negative constant C. Show that
a n /n converges as n → ∞.

This is essentially due to Fekete (1923). In various places we encounter this

useful lemma in deducing the existence of limits.
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 3

Problem 1. 6
{ }
For any positive sequence a n n≥1 show that
( )n
a1 + a n+1
>e
an
for infinitely many n’s, where e is base of the natural logarithm. Prove more-
over that the constant e on the right-hand side cannot in general be replaced
by any larger number.

Problem 1. 7
For any 0 < θ < π and any positive integer n show the inequality
sin 2θ sin nθ
sin θ + + ··· + > 0.
2 n

This was conjectured by Fejér and proved by Jackson (1911) and by Gron-
wall (1912) independently. Landau (1934) gave a shorter (maybe the shortest)
elegant proof. See also Problem 5. 9. Note that
∞
sin nθ π − θ
∑ n
=
2
n=1

for 0 < θ < 2π, which is shown in Solution 7. 10.

Problem 1. 8
For any real number θ and any positive integer n show the inequality
cos θ cos 2θ cos nθ 1
+ + ··· + ≥− .
2 3 n+1 2

This was shown by Rogosinski and Szegö (1928). Verblunsky (1945) gave
another proof. Koumandos (2001) obtained the lower bound −41/96 for n ≥ 2.
Note that ( )
∞
cos nθ π − θ θ
∑ n + 1 = 2 sin θ − cos θ log 2 sin 2
n=0

for 0 < θ < 2π.

For the simpler cosine sum Young (1912) showed that
m
cos nθ
∑ n
> −1
n=1
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

4 Problems and Solutions in Real Analysis

for any θ and positive integer m ≥ 2. Brown and Koumandos (1997) improved
this by replacing −1 by −5/6.

Problem 1. 9
{ } √ √
Given a positive sequence a n n≥0 satisfying a 1 ≥ a 0 + 1 and
a n2
a n+1 − ≤1
a n−1
for any positive integer n, show that
a n+1
an
converges as n → ∞. Show moreover that a n θ −n converges as n → ∞, where
θ is the limit of the sequence a n+1 /a n .

This is due to Boyd (1969).

Problem 1. 10
Let E be any bounded closed set in the complex plane containing an infinite
number of points, and let Mn be the maximum of | V(x1 , ..., x n ) | as the points
x1 , ..., x n run through the set E, where
∏
V(x1 , ..., x n ) = (x i − x j )
1≤i< j≤n

is the Vandermonde determinant. Show that Mn2/(n(n−1)) converges as n → ∞.

This is due to Fekete (1923) and the limit

τ(E ) = lim Mn2/(n(n−1))

n→∞

is called the transfinite diameter of E. See Problem 15. 9.

July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 5

Solutions for Chapter 1

¤ ¡
£Solution 1. 1 ¢
Let rn and  n be the integral and fractional parts of the number n!e respectively.
Using the expansion
1 1 1
e=1+ + + ··· + + ··· ,
1! 2! n!
we have
 ( )



1 1 1

 r = n! 1 + + + · · · +



n
1! 2! n!





 1 1

 n = + + ··· ,
n + 1 (n + 1)(n + 2)
since
1 1 1 1
< n < + + ··· = .
n+1 n + 1 (n + 1) 2 n
Thus sin (2n!eπ) = sin (2π n ). Note that this implies the irrationality of e.
Since n n converges to 1 as n → ∞, we have
sin (2π n )
lim n sin (2π n ) = lim = 2π.
n→∞ n→∞ n
Hence n sin (2n!eπ) converges to 2π as n → ∞. ¤

Remark. More precisely one gets

( )
1 1 1
n = − 3 +O 4 ;
n n n
hence we have
4π3 3 ( )
n sin (2n!eπ) = 2πn n + n n + O n n5
3
( )
2π(2π 2 − 3) 1
= 2π + 2
+O 3
3n n
as n → ∞.
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

6 Problems and Solutions in Real Analysis

¤ ¡
£Solution 1. 2 ¢
{ }
Let d n be any monotone increasing sequence of positive integers diverging
to ∞ and satisfying d n < n for n > 1. We divide the sum into two parts as follows.
 ( )n ( )n ( )n


 1 2 n − 1 − dn

 an = + + ··· + ,


 n n n



 ( )n ( )n


 n − dn n


 bn = + ··· + .
n n

First the sum a n is roughly estimated above by

∫ ( )n
1 n−d n
(n − d n ) n+1 dn
x dx =
n
< 1− .
nn 0 (n + 1)n n n

Now using the inequality log(1 − x) + x < 0 valid for 0 < x < 1 we obtain

0 < a n < e n log(1− dn /n) < e −dn ,

which converges to 0 as n → ∞.
Next by using Taylor’s formula for log(1 − x) we can take a positive constant
c1 such that the inequality

| log(1 − x) + x | ≤ c1 x 2

holds for any | x | ≤ 1/2. Thus for any integer n satisfying d n /n ≤ 1/2 we get
( )
k c1 k 2 c1 d n2
n log 1 − +k ≤ ≤
n n n

for 0 ≤ k ≤ d n . Suppose further that d n2 /n converges to 0 as n → ∞. For exam-

[ ]
ple d n = n 1/3 satisfies all the conditions imposed above. Next take a positive
constant c2 satisfying

| e x − 1 | ≤ c2 | x |

for any | x | ≤ 1. Since c1 d n2 /n ≤ 1 for all sufficiently large n, we have

( )n
k
e 1−
k
− 1 = e n log(1−k/n)+k − 1
n
c1 c2 d n2
≤ .
n
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 7

Dividing both sides by e k and summing from k = 0 to d n , we get

dn ( )n
k c1 c2 d n2 dn −k
∑ 1 −
n
− e −k ≤
n k=0
∑e .
k=0

Hence
dn
ec1 c2 d n2
bn − ∑ e −k <
(e − 1)n
,
k=0

which implies
( )
e e c1 c2 d n2
bn − ≤ + e −dn .
e−1 e−1 n
Therefore a n + bn converges to e/(e − 1) as n → ∞. ¤
¤ ¡
£Solution 1. 3 ¢
One can easily verify that the function f (x) = x log x satisfies all the con-
ditions stated in Problem 5.7. Therefore the logarithm of the given sequence
converges to
f (1) − f (0+)
= 0,
2
hence the limit is 1. ¤
¤ ¡
£Solution 1. 4 ¢
Let M be an upper bound of the two convergent sequences | a n | and | bn |. For
any  > 0 we can take a positive integer N satisfying | a n − α | <  and | bn − β | < 
for all integers n greater than N. If n is greater than N 2 , then

| a k bn−k − αβ | ≤ | (a k − α)bn−k + α(bn−k − β ) |

≤ (M + | α | ) 
[√ √ ]
for any integer k in the interval n , n − n . Therefore
n
1 1
n
∑ ak bn−k − αβ ≤
n √
∑ √
| a k bn−k − αβ |
k=0 n ≤k≤n− n
( ) [√ n ] + 1
+ 2 | αβ | + M 2
n
( ) √n + 1
≤ (M + | α | ) + 2 | αβ | + M 2 .
n
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

8 Problems and Solutions in Real Analysis

We can take n so large that the last expression is less than (M + | α | + 1). ¤
¤ ¡
£Solution 1. 5 ¢
For an arbitrary fixed positive integer k we put n = qk + r with 0 ≤ r < k.
Since a n = a qk+r ≤ q(a k + C) + a r , we have
an ak + C ar
≤ + .
n k n
Taking the limit as n → ∞, we get
an ak + C
lim sup ≤ .
n→∞ n k
The sequence a n /n is therefore bounded. Since k is arbitrary, we may conclude
that
an ak
lim sup ≤ lim inf ,
n→∞ n k→∞ k

which means the convergence of a n /n. ¤

¤ ¡
£Solution 1. 6 ¢
Without loss of generality we may put a 1 = 1. Suppose, contrary to the con-
clusion, that there is an integer N satisfying
( )n
1 + a n+1
≤e
an
for all n ≥ N. Put
( )
1 1
sj,k = exp + ··· +
j k
for any integers j ≤ k. Since 0 < a n+1 ≤ e 1/n a n − 1, we get successively



 0 < a n+1 ≤ s n,n a n − 1,






 0 < a n+2 ≤ s n,n+1 a n − s n+1,n+1 − 1,



 ..


 .




 0 < a n+k+1 ≤ s n,n+k a n − s n+1,n+k − · · · − s n+k,n+k − 1

for any non-negative integer k. Hence it follows that

1 1 1
an > + + ··· + .
s n,n s n,n+1 s n,n+k
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 9

On the other hand, using the inequality

( ∫ n+ j )
1 dx n−1
> exp − = ,
s n,n+ j n−1 x n+ j
we get
k
n−1
an > ∑ n+ j
,
j=0

which is a contradiction, since the right-hand side diverges to ∞ as k → ∞.

To see that the bound e cannot be replaced by any larger number, consider the
case a n = n log n for n ≥ 2. Then
( )n ( ( ( 1 )))
a 1 + (n + 1) log(n + 1) 1
= exp n log 1 + + O
n log n n n log n
( ( 1 ))
= exp 1 + O ,
log n
which converges to e as n → ∞. ¤
¤ ¡
£Solution 1. 7 ¢
Denote by s n (θ) the left-hand side of the inequality to be shown. Write θ for
2ϑ for brevity. Since
( )
sn0 (θ) = < e iθ + e 2iθ + · · · + e niθ
cos (n + 1)ϑ sin nϑ
= ,
sin ϑ
we obtain the candidates for extreme points of s n (θ) on the interval (0, π] by solv-
ing the equations cos (n + 1)ϑ = 0 and sin nϑ = 0, as follows:
π 2π 3π 4π
, , , , ...
n+1 n n+1 n
where the last two candidates are (n − 1)π/(n + 1) and π if n is even, and (n − 1)π/n
and nπ/(n + 1) if n is odd. In any case sn0 (θ) vanishes at least at n points in the
interval (0, π).
Since sn0 (θ) can be expressed as a polynomial in cos θ of degree n and cos θ
maps the interval [0, π] onto [−1, 1] homeomorphically, this polynomial possesses
at most n real roots in [−1, 1]. Therefore all these roots must be simple and give
the actual extreme points of s n (θ) except for θ = π. Clearly s n (θ) is positive in
the right neighborhood of the origin, and the maximal and minimal points stand
in line alternately from left to right. Thus s n (θ) attains its minimal values at the
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

10 Problems and Solutions in Real Analysis

points 2`π/n ∈ (0, π) when n ≥ 3. In the cases n = 1 and n = 2, however, s n (θ)

has no minimal points in (0, π).
Now we will show that s n (θ) is positive on the interval (0, π) by induction on
n. This is clear for n = 1 and n = 2 since s 1 (θ) = sin θ and s 2 (θ) = (1+cos θ ) sin θ.
Suppose that s n−1 (θ) > 0 for some n ≥ 3. Then the minimal values of s n (θ) are
certainly attained at some points 2`π/n in (0, π), whose values are
( ) ( )
2`π 2`π sin 2`π
sn = s n−1 +
n n n
( )
2`π
= s n−1 > 0.
n
Therefore s n (θ) > 0 on the interval (0, π). ¤

Remark. Landau (1934) gave the following elegant shorter proof using mathe-
matical induction on n. Suppose that s n−1 (θ) > 0 on (0, π). If s n attains the
non-positive minimum at some point, say θ ∗ , then sn0 (θ ∗ ) = 0 implies
( )
1 ∗ θ∗
sin n + θ = sin
2 2
and hence
( )
1 ∗ θ∗
cos n + θ = ± cos .
2 2
Since
( ) ( )
∗ 1 ∗ θ∗ 1 ∗ θ∗
sin nθ = sin n + θ cos − cos n + θ sin
2 2 2 2
∗ ∗ ∗ ∗
θ θ θ θ
= sin cos ± cos sin ,
2 2 2 2
being equal either to 0 or sin θ ∗ ≥ 0 according to the sign. We are led to a contra-
diction.
¤ ¡
£Solution 1. 8 ¢
The proof is substantially based on Verblunsky (1945). Write θ for 2ϑ for
brevity. Let cn (ϑ) be the left-hand side of the inequality to be shown. It suffices to
confine ourselves to the interval [0, π/2]. Clearly c1 (ϑ) = cos ϑ/2 ≥ −1/2 and
2 1 1 41
c 2 (ϑ) = cos2 ϑ + cos ϑ − ≥ − ,
3 2 3 96
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 11

and we assume that n ≥ 3. Note that

sin (2n + 1)ϑ − sin (2n − 1)ϑ
cos nθ =
2 sin ϑ
sin2 (n + 1)ϑ − 2 sin2 nϑ + sin2 (n − 1)ϑ
= ,
2 sin2 ϑ
{ }
whose numerator is the second difference of the positive sequence sin 2 nϑ . Us-
ing this formula we get

1 n
sin2 (k + 1)ϑ − 2 sin2 kϑ + sin2 (k − 1)ϑ
cn (ϑ) = ∑
2 sin2 ϑ k=1 k+1
,

which can be written as

(
1 2 sin2 ϑ sin2 2ϑ 2 sin2 (n − 1)ϑ
− + + ··· +
2 sin ϑ
2 3 12 n(n 2 − 1)
)
(n − 1) sin2 nϑ sin2 (n + 1)ϑ
− + .
n(n + 1) n+1
Hence we obtain
1 cos2 ϑ sin2 (n + 1)ϑ − sin2 nϑ
cn (ϑ) ≥ − + +
3 6 2(n + 1) sin2 ϑ
1 sin2 ϑ sin (2n + 1)ϑ
=− − + .
6 6 2(n + 1) sin ϑ
For any ϑ satisfying sin (2n + 1)ϑ ≥ 0 we obviously have cn (ϑ) ≥ −1/3. More-
over if ϑ belongs to the interval (3π/(2n + 1), π/2), then using Jordan’s inequality
sin ϑ ≥ 2ϑ/π,
1 1
cn (ϑ) ≥ − −
3 2(n + 1) sin (3π/(2n + 1))
1 2n + 1 1
≥− − >− .
3 12(n + 1) 2
Thus it suffices to consider the interval [π/(2n + 1), 2π/(2n + 1)].
In general, we consider an interval of the form
[ απ βπ ]
, .
2n + 1 2n + 1
For any ϑ satisfying sin (2n + 1)ϑ ≤ c on this interval it follows that

1 sin2 ϑ c
cn (ϑ) ≥ − − − .
6 6 2(n + 1) sin ϑ
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

12 Problems and Solutions in Real Analysis

Now the right-hand side can be written as −1/6−ϕ(sin ϑ), where ϕ(x) is a concave
function; hence, the maximum of ϕ is attained at an end point of that interval. By
using
απ
απ sin ϑ ≥ 7ϑ sin ,
7
we get
( απ ) 1 απ c
ϕ sin = sin2 +
2n + 1 6 2n + 1 2(n + 1) sin (απ/(2n + 1))
(απ) 2 c 2n + 1
≤ + · .
6(2n + 1) 2 2(n + 1) 7 sin (απ/7)
Since n ≥ 3, the last expression is less than

(απ) 2 c
+ .
294 7 sin (απ/7)
Similarly we get an estimate for another end point.
√
For α = 1 and β = 4/3 we can take c = 3/2 so that the value of ϕ at the
corresponding end point is less than 0.319 and 0.28 respectively. Similarly for α =
4/3 and β = 2 we can take c = 1 so that the value of ϕ is less than 0.314 and 0.318
respectively. Therefore the maximum of ϕ on the interval [π/(2n+1), 2π/(2n+1)]
is less than 1/3, which implies that cn (ϑ) > −1/2. ¤

¤ ¡
£Solution 1. 9 ¢
We first show that
a n+1 1
>1+ √ (1. 1)
an a0
√
by induction on n. When n = 0 this holds by the assumption. Put α = 1 + 1/ a 0
for brevity. Suppose that (1. 1) holds for n ≤ m. We then have a k > α k a 0 for
1 ≤ k ≤ m + 1. Thus
m+1 m+1
a m+2 a 1 a k+1 ak 1
− ≤ ∑ − ≤ ∑ ,
a m+1 a 0 k=1
a k a k−1 k=1
a k

which is less than

m+1
1 1 1
a0
∑ α −k < a0 (α − 1) = √ .
a0
k=1
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

Sequences and Limits 13

Therefore
a m+2 a 1 1 1
> −√ >1+ √ ;
a m+1 a 0 a0 a0
thus (1. 1) holds also for n = m + 1.
Let p > q be any positive integers. In the same way,
p p
a p+1 a q+1 a k+1 ak 1
− ≤ ∑ − ≤ ∑ ,
ap aq k=q+1
a k a k−1 a
k=q+1 k

which is less than

p−q √
1 1 a0
∑
a q k=1 α k
<
aq
.

{ }
This means that the sequence a n+1 /a n satisfies the Cauchy criterion since a q
diverges to ∞ as q → ∞. Letting p → ∞ in the above inequalities, we get
√
a q+1 a0
−θ ≤ .
aq aq

Multiplying both sides by a q /θ q+1 , we have

√
a q+1 a q a0
− ≤ ,
θ q+1 θ q θ q+1
{ }
which shows that the sequence a n /θ n also satisfies the Cauchy criterion. ¤
¤ ¡
£Solution 1. 10 ¢
Let ξ 1 , ..., ξ n+1 be the points at which | V(x 1 , ..., x n+1 ) | attains its maximum
M n+1 . Since
V(ξ 1 , ..., ξ n+1 )
= (ξ 1 − ξ n+1 ) · · · (ξ n − ξ n+1 ),
V(ξ 1 , ..., ξ n )
we have
M n+1
≤ | ξ 1 − ξ n+1 | · · · | ξ n − ξ n+1 |.
Mn
Applying the same argument to each point ξ 1 , ..., ξ n , we get n + 1 similar inequal-
ities whose product gives
( )n+1 ∏
M n+1
≤ | ξ i − ξ j | = M n+1
2
.
Mn i, j
July 31, 2007 13:35 WSPC/Book Trim Size for 9in x 6in real-analysis

14 Problems and Solutions in Real Analysis

Hence the sequence M n2/(n(n−1)) is monotone decreasing. ¤