You are on page 1of 58

Chapter 3, Numerical Sequences and Series

Exercise 1. Prove that convergence of {sn } implies convergence of {|sn |}. Is the converse true?

Hints and Steps:

• Use inequality ||a| − kbk| ≤ ka − bkshown in exercise 1.13

Solution 1. Suppose s = lim sn , and let ε > 0. Then there is an integer N such that n ≥ N implies |s − sn | < ε. By Exercise
13 Chapter 1, if n ≥ N then
||s| − |sn || ≤ |s − sn | < ε.

Hence {|sn |} converges to |s|. The converse does not hold, for instance if sn = (−1)n then {sn } has two distinct constant
subsequences, s2n → 1 and s2n+1 → −1, hence {sn } does not converge. However |sn | = 1 is constant, so {|sn |} converges to 1.

Remarks:

• As we saw, the reciprocal of this exercise is not true and the result can be generalized to metric spaces, as follows, Let
(X, d) be a metric space, then convergence of {pn } implies convergence of {d (pn , q). For all q ∈ X, the proof is essentially
the same as the one given here. Geometrically speaking, the result of this exercise can be stated in the following way, if a
sequence of points is convergent, then the sequence of the distances from these points to any fixed point is also convergent.

2.5 s1
s2
2 s3
s4
s
1.5 sn

0.5

−0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5


−0.5

Figure 1: The length of the red arrows tends to the length of the black arrow when the points sn tends to s

1
2

Exercise 2. Calculate limn→∞ ( n2 + n − n).

Hints and Steps:

• Multiply and divide by the conjugate radical to obtain an equivalent form that eliminates the indetermination ∞ − ∞

Solution 2. Note that for n = 1, 2, 3, . . .



p p  n2 + n + n n 1
n2 +n−n= n2 +n−n × √ =√ =p .
2
n +n+n 2
n +n+n 1 + 1/n + 1

For these n we have r !2 r


1 1 1 1
1+ −1 ≤ =⇒ 1 + − 1 ≤ √ .

n n n n
p √
Let ε > 0, and let N be an integer so that N ε2 > 1. Then whenever n ≥ N , 1 + 1/n − 1 ≤ 1/ n < ε. It follows that

p
1 + 1/n → 1. By the laws for limits:
p 1 1
lim ( n2 + n − n) = lim p = .
n→∞ n→∞ 1 + 1/n + 1 2

Remarks:

• This exercise serves to exemplify how by means of algebraic manipulations we can solve indeterminate limits , but it
is good to note that many times if we are modeling a discrete process and the numerical values that we obtain are

f (n) = n2 + n − n then most likely from the numerical values we first arrive at the formula f (n) = √ 1 . rather
1+1/n+1

than a n2 + n − n to model the process.

3

Exercise 3. If s1 = 2, and

q
sn+1 = 2+ sn (n = 1, 2, 3, . . .),

prove that {sn } converges, and that sn < 2 for n = 1, 2, 3 . . . .

Hints and Steps:

• Use induction to prove that sn is increasing monotone and bounded by 2, then apply Theorem 3.14
p √ √
Solution 3. Note s2 = 2 + 2 > 2 = s1 . If sn+1 > sn , then
√ √
q q
sn+2 = 2 + sn+1 > 2 + sn = sn+1 .

By induction {sn } is increasing, and hence bounded below by s1 . We have s21 = 2 < 4, so s1 < 2. If sn < 2, then
√ √
s2n+1 = 2 + sn < 2 + 2 < 4,

hence sn+1 < 2. By induction sn < 2 for n = 1, 2, 3, . . . . Therefore {sn } is monotonic and bounded, so it converges.

Remarks:

• This exercise is an example of how to use Theorem 3.14 to show the convergence of a given monotone sequence.

4
Exercise 4. Find the upper and lower limits of the sequence {sn } defined by
s2m−1 1
s1 = 0; s2m = ; s2m+1 = + s2m .
2 2
Hints and Steps:

• Use induction and definition 3.15and Theorem 3.1 to show first that the subsequence {s2m−1 } converges, by the law limits
show that limm→∞ s2m−1 = 1 and limm→∞ s2m = 12 , and prove that these values are the only subsequential limits of {sn }.

Solution 4. Consider the subsequence {s2m−1 }. We have


1 1 + s2m−1
s2m+1 = + s2m = .
2 2
Note s1 < 1, and if s2m−1 < 1 then
1 + s2m−1
s2m+1 = < 1.
2
1
By induction {s2m−1 } is bounded above by 1. Also s3 = 2 > 0 = s1 , and if s2m+1 > s2m−1 then

1 + s2m+1 1 + s2m−1
s2m+3 = > = s2m+1 .
2 2
By induction {s2m−1 } is increasing. Hence {s2m−1 } is bounded and monotonic, so it converges. By applying the limit laws to
2s2m+1 = 1 + s2m−1 , we see
2 lim s2m−1 = lim 2s2m+1 = 1 + lim s2m−1
m→∞ m→∞ m→∞

So
lim s2m−1 = 1.
m→∞
s2m−1
Since s2m = 2 , by the limit laws, the subsequence {s2m } converges to 21 . We claim that 1
2 and 1 are the only subsequential
limits of {sn }.
Let {snk } be a subsequence of {sn }. If infinitely many of the nk are even, and infinitely many are odd, there is a subsequence
1
of {snk } converging to 1, and another converging to 2, so that {snk } does not converge. If all but finitely many of the nk are
odd then snk → 1, and if all but finitely many are even then snk → 12 .
1
Hence 2 are 1 are the only subsequential limits of {sn }, and so lim sup sn = 1 and lim inf sn = 21 .

5
Remarks:

• In this exercise we have a sequence {sn } with infinite range such that lim sup sn 6= lim inf sn and only two subsequential
limits.

6
Exercise 5. For any two real sequences {an }, {bn }, prove that

lim sup(an + bn ) ≤ lim sup an + lim sup bn ,


n→∞ n→∞ n→∞

provided the sum on the right is not of the form ∞ − ∞.

Hints and Steps:

• Use Definition 3.16 and Theorems 3.3(a), 3.17(a)

Solution 5. Let
s = lim sup(an + bn ) a = lim sup an and b = lim sup bn .
n→∞ n→∞ n→∞

Suppose that lim supn→∞ an + lim supn→∞ bn is not of the form ∞ − ∞.

• Case (i): When a = +∞ or b = +∞. The inequality holds trivially in this case.

• Case (ii): When a < +∞ or b < +∞. By Theorem 3.17(a), there exists a subsequence {ank + bnk } such that
s = limn→∞ (ank + bnk ).

1. If any one of {ank } and {bnk } converges, say {ank } , then the equation bnk = ank + bnk − ank implies that {bnk } also
converges and it follows from Theorem 3.3(a) and Definition 3.16 that

s = lim (ank + bnk ) = lim ank + lim bnk ≤ a + b.


k→∞ k→∞ k→∞

2. If both {ank } and {bnk } diverge but {ank + bnk } converges, then we consider α = lim supk→∞ ank and β =
lim supk→∞ bnk . By Definition 3.16, we have

α = lim sup ank ≤ lim sup an = a and β=lim sup bnk ≤ lim sup bn = b.
k→∞ n→∞ k→∞ n→∞

n o
so αand β are finite. By Theorem 3.17(a) again, there exists a subsequence ankj of the divergent sequence {ank }
such that α = limj→∞ ankj .
By Definition 3.5 we have that s == limj→∞ ankj + bnkj .

7
n o
Thus, as in proof of item 1 above, we have that the subsequence bnkj of the divergent sequence {bnk } is convergent.
Hence we obtain from the definition of s and Theorem 3.3(a) that

s = lim ankj + bnkj = lim ankj + lim bnkj ≤ lim sup ank + lim sup bnk ≤ a + b.
j→∞ j→∞ j→∞ k→∞ k→∞

Remarks:

• In algebraic terms the result of this exercise tell us that the upper limits has the property known as sublinearity.

8
P
Exercise 6. Investigate the behavior (convergence or divergence) of an if
√ √
1. an = n + 1 − n;
√ √
n+1− n
(a) an = ;
√ n
(b) an = ( n n − 1)n ;
1
(c) an = , for complex values of z.
1 + zn
Hints and Steps:

• Use the convergence tests show in the book, from Theorem 3.25 to Theorem 3.33
√ √
Solution 6. (a) Since n + 1 + n 6= 0, we have
√ √  √ √ 
√ √ n+1− n n+1+ n 1 1
an = n + 1 − n = √ √ =√ √ ≥ 1 ≥ 0.
n+1+ n n+1+ n 2(n + 1) 2
By Theorem 3.28,
∞ ∞
X 1X 11
1 1 =
n=1 2(n + 1)
2 k2 2
k=2
P∞ √ √
diverges. So by Theorem 3.25 (Comparison Test), n=1 n + 1 − n diverges.
(b) By item (a) we have √ √
n+1− n 1 1
an = = √ √ ≤ 3 .
n n n+1+ n 2n 2
By Theorem 3.28,
∞ ∞
X 1 1X 1
3 =
n=1 2n
2 2 n=1 n 32
P∞ √ √
n+1− n
converges, so the comparison test shows that n=1 n converges.
√ √ √ √
(c) By Theorem 3.2 (c) we know that n n − 1 ≥ 0 for all positive integer n and limn→∞ n n − 1 = 0, since n an = n n − 1 and

the sequence { n n − 1} converges, then
p √  √ 
lim sup n |an | = lim sup n n − 1 = lim n n − 1 = 0 < 1,
n→∞ n→∞ n→∞

9
P∞ √
it follows from Theorem 3.33 (Root Test) that n=1 (
n
n − 1)n converges.
(d) Note that if |z| ≤ 1, then the triangle inequality implies that
n
|1 + z n | ≤ 1 + |z| ≤ 2,

hence
1 1
|an | = ≥
|1 + z n | 2
P∞ 1
Thus, by Theorem 3.23, n=1 diverges if |z| ≤ 1,.
1 + zn
k
On the other hand, fix |z| > 1,, then |z| > 1 for all positive integer k and the triangle inequality implies that
n
|z| = |z n | = |(1 + z n ) − 1| ≤ |1 + z n | + 1

so
n
|z| − 1 ≤ |1 + z n |
n−1
Since − |z| < −1, then  
n 1 n n−1 n
0 < |z| 1− = |z| − |z| |z| − 1 ≤ |1 + z n |
|z|
Hence
1 1 |z| 1
0< < =
|z| − 1 |z|n

|1 + z n | n
|z| 1 − 1
|z|

So
∞ ∞  n  2
X |z| 1 |z| X 1 |z|
= =
n=1
|z| − 1 |z|n |z| − 1 n=1 |z| |z| − 1
P∞   n
1
converges, because we can applied Theorem 3.26 to the series n=1 |z| . Hence it follows from Theorem 3.25 (Comparison
P∞ 1
Test) that n=1 1+zn converges.

Remarks:

• With these exercises the student is expected to develop skills to determine the convergence of a given series, using the
different tests demonstrated in the Rudin’s book.

10
P
Exercise 7. Prove that the convergence of an implies the convergence of
X √an
n
if an ≥ 0.

Hints and Steps:


√ P √a n
• Expand 1
2
an − n and and deduce a relationship between the term of the series n and other convergent series that
allows you to use Theorem 3.25 (Comparison Test).
P
Solution 7. Suppose that an converges
Note that   √
√ 1 2 an 1
0≤ an − = an − 2 + 2
n n n
So √  
an 1 1
≤ an + 2
n 2 n
1
P P1 1

By Theorem 3.28, converges (it is a p-series with p = 2). Hence by Theorem 3.47 the series
n2 2 an + n2 converges,
P √an
Since an ≥ 0, we can use Theorem 3.25 (Comparison Test) and conclude that n converges.

Remarks:

• This exercise give us other criteria of convergence of series.

11
P P
Exercise 8. If an converges, and if {bn } is monotonic and bounded, prove that an bn converges.

Hints and Steps:

• Define b = lim bn , which exists by Theorem 3.14, and construct a sequence {cn } related to {bn } such that hypothesis of
Theorem 3.42 are fulfill.

Solution 8. Since ( {bn } is monotonic and bounded, by Theorem 3.14 we have that b = lim bn exists, Let {cn } the sequence
b − bn if {bn }is increasing
defined by by cn = . So we have
bn − b if {bn }is decreasing
P
1. an converges

2. cn is a monotonously decreasing sequence whose limit is 0


P P
Hence we can apply Theorem 3.42 and we get that an cn converges. Also note that Theorem 3.47 implies that ban converges.
Finally note that ( P P
P P P ban + an cn if {bn }is decreasing
an bn = ban + an (bn − b) = P P
ban − an cn if {bn }is increasing
P
Since both summands in the right side converges in both cases, then Theorem . 3.47 implies that an bn converges too.

Remarks:

• This criteria of convergence of the series


P
an bn is known as Abel’s Test.

12
Exercise 9. Find the radius of convergence of each of the following power series:

n3 z n ,
P
1.
P 2n n
2. z ,
n!
P 2n n
3. z ,
n2
P n3 n
4. z .
3n
Hints and Steps:

• Use the definition of radius of convergence ( Note in Theorem 3.39), when lim aan+1
n
exists use this and Theorem 3.37

Solution 9. (a) Using item (c) in Theorems 3.3 and 3.20, we have

n √ 3
 √ 3
lim n3 = lim n
n = lim n
n = 13 = 1
n→∞ n→∞ n→∞

So
p
n

n

n
α = lim sup |n3 | = lim sup n3 = lim n3 = 1.
n→∞ n→∞ n→∞

1
n3 z n is R =
P
Hence we have that the radius of convergence of the series α = 1.
(b) Note that
2n+1
an+1 (n + 1)! 2
lim = lim = lim = 0,
n→∞ an n→∞ 2n n→∞ n + 1
n!
then by Theorem 3.37, we have

s r 2n+1 2n+1
2n n 2
n
(n + 1)! (n + 1)!
0 ≤ α = lim sup n = lim sup ≤ lim sup = lim = 0.
n→∞ n! n→∞ n! n→∞ 2n n→∞ 2n
n! n!

13
P 2n n
Therefore the radius of convergence of the series z is R = α1 = ∞.
n!
(c) Using item (d) in Theorem 3.3 and item (c) in Theorem 3.20, we have
r
n
n 2 2 2
lim = lim √ 2 = √ 2 =2
n→∞ n2 n→∞ n n (limn→∞ n n)

So s r r
n
n 2 n 2
n
n 2
n
α = lim sup 2 == lim sup = lim = 2.
n→∞ n n→∞ n2 n→∞ n2
P 2n n 1
Hence we have that the radius of convergence of the series z is R = α = 12 .
n2
(d) Using item (c) in Theorem 3.20, we have
r
3
√ 3 √ 3
n n n (limn→∞ n n) 1
n

lim = lim = =
n→∞ 3n n→∞ 3 3 3
So s r r
3
n n n n
3
n n
3
1
α = lim sup n == lim sup n
= lim n
= .
n→∞ 3 n→∞ 3 n→∞ 3 3
P n3 n 1
Hence we have that the radius of convergence of the series z is R = α =3
3n
Remarks:
With this exercise the student is expected to improve his understanding of the definition of the radius of convergence of a
power series and to be able to use the properties of limits to calculate it in particular cases.

14
an z n are integers, infinitely many of which are distinct from
P
Exercise 10. Suppose that the coefficients of the power series
zero. Prove that the radius of convergence is at most 1.

Hints and Steps:

• Suppose that contrary and obtain a contradiction about the convergence of the series
P
an .

an z n has radius of convergence greater than 1, them taking z = 1 the


P
Solution 10. Suppose the contrary, i.e., the series
P
series an converges, but since infinitely many an are integers distinct from zero, then limn→∞ an 6= 0, and this fact contradicts
P
that an converges.

Remarks:

• This Exercise provides an upper bound for the radius of convergence of power series with integer coefficients.

15
P
Exercise 11. Suppose an > 0, sn = a1 + · · · + an , and an diverges.
P an
1. Prove that diverges.
1 + an
2. Prove that
aN +1 aN +k sN
+ ··· + ≥1−
sN +1 sN +k sN +k
P an
and deduce that diverges.
sn
3. Prove that
an 1 1
2
≤ −
sn sn−1 sn
P an
and deduce that converges.
s2n
4. What can be said about
X an X an
and ?
1 + nan 1 + n2 an
Hints and Steps:

• For (a). Suppose an


P
converges and show that {an } is bounded, use comparison test to obtain a contradiction about
1+an
P
the divergence of the series an

• For (b). Use that {Sn } is monotonically increasing in the first part, then use that {Sn } is unbounded to obtain that the
P an
series does not satisfy Theorem 3.22
sn
P an
• For (c). Related the series with a telescopic convergent series
s2n
• For (d). Use Comparison test to show that
P an
1+n2 an converges, construct examples of sequences {an } and {bn } that fulfill
P an P bn
the hypothesis and so that 1+nan diverges and 1+nbn converges
P an
Solution 11. (a) Assume that 1+an converges, since an > 0 for all n ∈ N, then by Theorem 3.23 we have

an 1
0 = lim = lim 1
n→∞ 1 + an n→∞
an +1

16
the limits at the right side in this equation is zero if and only if limn→∞ an = 0, hence in this situation we have that the sequence
{an } is bounded, let M such that M > an > 0, then 1 + M > 1 + an > 0, so
an an
>
1 + an 1+M
for all n, and by our assumption and comparison test the series
1 X X an
an =
1+M 1+M
P P an
converges, but this fact contradicts that an diverges. So we need have 1+an diverges
(b) Since an > 0 for all n ∈ N, then {Sn } is monotonically increasing, so for all N, k ∈ N with k > 1, we have

SN +1 < SN +2 < · · · < SN +k−1 < SN +k

Hence
aN +1 aN +k aN +1 aN +k
+ ··· + > + ··· +
sN +1 sN +k sN +k sN +k
1
= (aN +1 + · · · + aN +k )
sN +k
1
= (SN +k − SN )
sN +k
sN
=1−
sN +k

sN
Now, since {Sn } is increasing and unbounded, for each N ∈ N it is possible to choose k such that sN +k is arbitrarily close to
sN
0, in particular sN +k > 12 , for this N and k, we have

aN +1 aN +k sN 1 1
+ ··· + >1− >1− =
sN +1 sN +k sN +k 2 2

1
does not exists N ∈ N so that asN + · · · + asN < 12 for all k > 1, hence by Theorem 3.22 the series
N +1 N +k
Thus, taking ε = 2 +1 +k
P an
diverges.
sn

17
(c) Since {Sn } is monotonically increasing, we have for all n> 1 that Sn2 ≥ Sn−1 Sn , so
an an Sn − Sn−1 1 1
≤ = = −
s2n Sn−1 Sn Sn−1 Sn Sn−1 Sn
Hence
N N  
X an a1 X 1 1 a1 1 1 a1 1
2
≤ 2+ − = 2+ − < 2+
s
n=1 n
S1 n=1 Sn−1 Sn S1 S1 SN S1 S1
P an
So the partial sums of the series form an increasing and bounded sequence, hence by Theorem 3.24 convergent, thus
s2n
P an
the series converges.
s2n
(d) Since an > 0, we have
an 1 1
= 1 < 2
1 + n2 an an + n2 n
1
P
and note that the series n2 converges, because it is a p-series with p = 2, so by Theorem 3.28 (Comparison test) the series
P an
1+n2 an converges.
P P an
There exists sequences {an } with an > 0 and an diverges, so that the series 1+nan diverges, for example, an = 1 for ail
P an P 1
n, in this case 1+nan = 1+n which diverges.
P P bn
On the other hand, also there exists sequences {bn } with bn > 0 and bn diverges, so that the series 1+nbn converges,

n n = 2k
for example, consider bn = . And note that
0 otherwise


X ∞
X
bn = 2k
n=1 k=1
N ∞
X bn X 2k
=
n=1
1 + nbn 1 + 22k
k=1
P∞ k
2 1
Clearly n=1 bn
diverges because its terms does not tends to zero, and we have 1+22k < 2k for all integer positive number k,
P 1 P 2k
and by Theorem 2.26 the series 2k
converges, so by Theorem 2.28 the series 1+22k
also converges.

Remarks:

18
• The items in this Exercise allow us to construct, from a given divergent series of positive terms, criteria to determine the
convergence of some series related to the given series

19
P
Exercise 12. Suppose an > 0 and an converges. Put

X
rn = am .
m=n

1. Prove that
am an rn
+ ··· + >1−
rm rn rm
P an
if m < n, and deduce that diverges.
rn
(a) Prove that
an √ √
√ < 2( rn − rn+1 )
rn
P an
and deduce that √ converges.
rn

Hints and Steps:

• For (a) imitate the proof of (b) in Exercise 11

• For (b) imitate the proof of (c) in Exercise 11

Solution 12. (a) The definition of rn imply that this sequence is monotonically decreasing, since rn is the ’tail’ of a convergent
series then limn→∞ rn = 0, So given any m ∈ N we have
rm+k
lim 1 − = 1,
k→∞ rm
rm+k 1
hence it is possible to choose k such that rm < 2 , for this m and k, we have

k k k
X am+j X am+j 1 X
> = am+j
j=0
rm+j j=0
rm rm j=0
rm –rm+k rm+k 1
= = 1– >
rm rm 2

20
1
P
k am+j

1
P an
Thus, taking ε = 2 does not exists m ∈ N so that j=0 rm+j < 2 for all k > 1, hence by Theorem 3.22 the series
rn
diverges.
(b) Since the sequence is monotonically decreasing we have
√ √  √ √  √ √ √ 
an = rn − rn+1 = rn + rn+1 rn − rn+1 ≤ 2 rn , rn − rn+1

so
an √ √ 
√ < 2 rn − rn+1
rn
Since
N N
X an X √ √  √ √ √
√ < 2 rn − rn+1 = 2( r1 – rn+1 ) < r1
n=1
rn n=1
P an
So the partial sums of the series √ form an increasing and bounded sequence, hence by Theorem 3.24 convergent, thus the
rn
P an
series converges..
s2n
Remarks:

• As in Exercise 11, the items in this Exercise allow us to build, from a given convergent series of positive terms, criteria to
determine the convergence of some series related to the given series

21
Exercise 13. Prove that the Cauchy product of two absolutely convergent series converges absolutely.

Hints and Steps:

• Use Theorem 3.50 to show the convergence of the Cauchy product of the series
P P
|an | and
|bn |, then use triangle
P P
inequality and comparison test to show the absolute convergence of the Cauchy product of the series an and bn
P P P
Solution 13. Since |an | and |bn | converges, by Theorem 3.50 also converges their Cauchy product Cn , where
n
X n
X
Cn = |ak | |bn−k | = |ak bn−k |
k=0 k=0
Pn
Let cn be the Cauchy product of an and bn . So cn = k=0 ak bn−k , hence by triangle inequality we have

Xn X n
|cn | = ak bn−k ≤ |ak bn−k | = Cn


k=0 k=0
P P P
thus, the term of the series |cn |are majored by a convergent series, and Theorem 2.8 assure that |cn |converges. i.e., cn
converges absolutely.

Remarks:

• This results in particular say that the convergence of two series with positive terms, always implies the convergence of its
Cauchy product.

22
Exercise 14. If {sn } is a complex sequence, define its arithmetic means σn by
s0 + s1 + · · · + sn
σn = (n = 0, 1, 2, . . .).
n+1
1. If lim sn = s, prove that lim σn = s.

(a) Construct a sequence {sn } which does not converge, although lim σn = 0.
(b) Can it happen that sn > 0 for all n and that lim sup sn = ∞, although lim σn = 0?
(c) Put an = sn − sn−1 , for n ≥ 1. Show that
n
1 X
sn − σn = kak .
n+1
k=1

Assume that lim(nan ) = 0 and that {σn } converges. Prove that {sn } converges. [This gives a converse of (a), but
under the additional assumption that nan → 0.]
(d) Derive the last conclusion from a weaker hypothesis: Assume M < ∞, |nan | ≤ M for all n, and lim σn = σ. Prove
that lim sn = σ, by completing the following outline:

If m < n, then
n
m+1 1 X
sn − σn = (σn − σm ) + (sn − si ).
n−m n − m i=m+1
For these i,
(n − i)M (n − m − 1)M
|sn − si | ≤ ≤ .
i+1 m+2
Fix ε > 0 and associate with each n the integer m that satisfies
n−ε
m≤ < m + 1.
1+ε
Then (m + 1)/(n − m) ≤ 1/ε and |sn − si | < M ε. Hence

lim sup|sn − σ| ≤ M ε.
n→∞

Since ε was arbitrary, lim sn = σ.

23
Hints and Steps:

Solution 14. (a) Since sn converges, then {sn } is a bounded set, let M so that |sn | < M and note that |s| < M .
Now using the triangle inequality, for all k ∈ N we have

|s0 –s| + · · · + |sk−1 − s| < 2M k

ε
Let ε > 0 arbitrary and choose N ∈ N such that |sn –s| < 2 for all n > N .
ε
Consider n > N . so that 2M N < (n + 1) 2, Hence


s0 + · · · + sn |s0 –s + · · · + sn − s|
|σn − s| =
− s =
n+1 n+1
|s0 –s| + · · · + |sN −1 − s| |sN –s| + · · · + |sn − s|
≤ +
n+1 n+1
2M N (n − N + 1) ε
< +
n+1 n+1 2
ε ε
< + =ε
2 2
This shows that (σn − s) converges to 0, i.e. that σn → s. 
0 n is even
(b) Let sn = (−1)n . Then sn does not converge, but σn = which converges to 0.
1

n+1 n is odd

k if n = k 3 for some k ∈ N
(c) the answer to this item is yes. Indeed, Let sn = . Then sn > 0, and the sequence sn is
1 otherwise
n

24
3
unbounded, so lim sup sn = ∞. Note that if k 3 ≤ n < (k + 1) , then we have
n k
1 X1 1 X
0 < σn < + i
n + 1 i=0 n n + 1 i=1
n
1 X1 1 k (k + 1)
= +
n + 1 i=0 n n + 1 2
n
1 X1 1 k (k + 1)
< +
n + 1 i=0 n k 3 2
n
1 X1 1 1
= + +
n + 1 i=0 n 2k 2k 2
1
by (a) apply to the sequence n , we have
n
1 X1 1
lim = lim = 0,
n→∞ n + 1 n n→∞ n
i=0

also we have that if n → ∞, then k → ∞,. So


   
1 1 1 1
lim + = lim + = 0,
n→∞ 2k 2k 2 k→∞ 2k 2k 2
In consequence,
n
! n  
1 X1 1 1 1 X1 1 1
0 < lim σn < lim + + 2 = lim + lim + 2 = 0,
n→∞ n→∞ n + 1 i=0 n 2k 2k n→∞ n + 1
i=0
n k→∞ 2k 2k

which shows that σn → 0.


(d) Note that
n
X n
X
ksk = ksk
k=1 k=0
n
X n−1
X
ksk−1 = (k + 1)sk
k=1 k=0

25
So

n
X n
X n
X n
X
kak = k(sk –sk−1 ) = ksk – ksk−1
k=1 k=1 k=1 k=1
n
X n−1
X n−1
X n−1
X
= ksk – (k + 1)sk = nsn + ksk − (k + 1)sk
k=0 k=0 k=0 k=0
n−1
X n
X
= nsn − sk = (n + 1) sn − sk
k=0 k=0

= (n + 1) (sn − σn )

Therefore
n n
1 X 1 X
sn − σn = kak = kak
n+1 n+1
k=1 k=0

Note that the right hand side is the (n + 1)-th arithmetic mean of the sequence {kak }k=0 , which by assumption and item (a)
converges. Therefore sn − σn is convergent, and since σn is convergent by assumption, then (a) in Theorem 3.3 imply that
sn = (sn –σn ) + σn is also convergent.
(e) If n > m, we have
n
" n
#
m+1 1 X 1 X
(σn − σm ) + (sn − si ) = (m + 1) (σn − σm ) + (sn − si )
n−m n − m i=m+1 n−m i=m+1
" m
! n
#
1 1 X X
= (m + 1) σn − si + (n − m) sn − si
n−m m + 1 i=0 i=m+1
" m n
#
1 X X
= (m + 1) σn − si + (n − m) sn − si
n−m i=0 i=m+1
" n
#
1 X
= (m + 1) σn + (n − m) sn − si
n−m i=0
1
= [(n − m) (sn − σn )] = sn − σn
n−m

26
Also note that for m + 1 ≤ i < n, we have m + 2 ≤ i + 1 and n − i < n − (m + 1) = n − m − 1, hence

n−i n−m−1

i+1 m+2
On the other hand, for m + 1 ≤ i < n, applying triangular inequality and the assumption |nan | ≤ M, we have

|sn − si | = |sn − sn−1 + sn−1 − sn−2 + sn−2 − · · · − si+1 + si+1 − si |


≤ |sn − sn−1 | + |sn−1 − sn−2 | + · · · + |si+1 − si |
= |an | + |an−1 | + · · · + |ai+1 |
M M M
≤ + + ··· +
n n−1 i+1
(n − i)M (n − m − 1)M
≤ ≤ .
i+1 m+2
n−ε
Let 1 > ε > 0, for each n we have 1+ε > 0 and by Archimedean property there exist N ∈ N so that

n−ε
< N,
1+ε
define m + 1 as the infimum of N ∈ N the that satisfies the above inequality, then
n−ε
m≤ <m+1 (1)
1+ε
and note that if n → ∞, then m → ∞, because n < m + 1 + mε + 2ε < 2m + 3
Also note that
n−ε m+1 1
m≤ ⇐⇒ m + mε ≤ n − ε ⇐⇒ (m + 1) ε ≤ n − m ⇐⇒ ≤
1+ε n−m ε
and
n−ε n−m−1
< m + 1 ⇐⇒ n − ε < m + 1 + (m + 1) ε ⇐⇒ n − m − 1 < (m + 2) ε ⇐⇒ <ε
1+ε m+2
Therefore,
(n − m − 1)M
|sn − si | ≤ < Mε
m+2

27
By assumption the sequence {σn } converges, hence this sequence is a Cauchy sequence so there exists N , so that if n, m > N ,
n−3
then |σn − σm | < M ε2 . Take n so that 2 > N and consider m defined in inequality (1), then n, m > N and
n
m+1 1 X
|sn − σn | = | (σn − σm ) + (sn − si )|.
n−m n − m i=m+1
n
m+1 1 X
≤ |σn − σm | + |sn − si |
n−m n − m i=m+1
n
1 1 X
≤ M ε2 + M ε = 2M ε
ε n − m i=m+1

Since ε was arbitrary, we have lim|sn − σn | = 0. In consequence the sequence {sn − σn } also converges to zero, since σn is
convergent by assumption, then (a) in Theorem 3.3 imply that sn = (sn –σn ) + σn is also convergent and

lim sn = lim sn –σn + lim σn = 0 + σ = σ.

Remarks:
This exercise shows relationships between a convergent series and the series of arithmetic means of its terms, these
relationships are widely used especially in the measure theory and probability.

28
Exercise 15. Definition 3.21 can be extended to the case in which the an lie in some fixed rearrangements of . Absolute
P
convergence is defined as convergence of |an |. Show that Theorems 3.22, 3.23, 3.25(a), 3.33, 3.34, 3.42, 3.45, 3.47, and 3.55
are true in this more general setting. (Only slight modifications are required in any of the proofs.)

Hints and Steps:

• Use the Theorems or its proofs that want generalize.

Solution 15.
P
Theorem. (Generalized Theorem 3.22) an converges if and only if for every ε > 0 there is an integer N ∈ N such that
m
X
ai ≤ ε



i=n

if m ≥ n ≥ N .

Proof. First note that if b1 , · · · , bk are non negative numbers, then

k k k
!2
X X X X
b2j ≤ b2i ≤ b2i +2 bi bl = bi
i=1 i=1 i6=l i=1

for all j = 1, ..., k, hence


k
! 21 k
X X
bj ≤ b2i ≤ bi
i=1 i=1

Thus, if a = (a1 , · · · , ak ), then


k
! 12 k
X 2
X
|aj | ≤ |a| = |ai | ≤ |ai | (2)
i=1 i=1

Recall that theorem 3.4 states that the sequence {xn } converges if and only if each sequence of components {xn,j } converges,
for j = 1, ..., k.

29
P P
Suppose that an converges and let ε > 0 , then by the say above, an,j converges for j = 1, ..., k. By theorem 3.22 for
each j = 1, ..., k, there existNj ∈ N such that m
X ε
ai,j ≤

k


i=n

if m ≥ n ≥ Nj .
Let N = max {Nj, : j = 1, ..., k}, then by the right inequality (2), if m ≥ n ≥ N , we have
m m
k X
k
X X X ε
ai ≤ ai,j ≤ =ε

k


i=n j=1 i=n j=1

And reciprocally, if for every ε > 0 there is an integer N ∈ N such that


m
X
ai ≤ ε



i=n

if m ≥ n ≥ N . Then for each j = 1, ..., k the left inequality (2) imply


m m
X X
ai,j ≤ ai ≤ ε



i=n i=n
P P
if m ≥ n ≥ N . Hence by theorem 3.22 an,j converges for each j = 1, ..., k, in consequence also converges the series an 
P
Theorem. (Generalized Theorem 3.23) If an converges, then limn→∞ an = 0.

Proof. Taking m = n in the previous theorem we have that for for every ε > 0 there is an integer N ∈ N such that
n
X
|an − 0| = |an | = ai ≤ ε


i=n

if n > N , which tell us that limn→∞ |an − 0| = 0, i.e., limn→∞ an = 0. 


P
Theorem. (Generalized Theorem 3.25 (a)) If |an | ≤ cn for n ≥ N0 , where N0 is some fixed integer and if cn converges, then
P
an converges.

30
P
Proof. Note first that all cn ≥ 0, for n ≥ N0 . Since cn converges, by Cauchy criterion, given ε > 0, there exists N ≥ N0 such
that if m ≥ n ≥ N , implies m
m
X X
ci = ci ≤ ε


i=n i=n

Hence, by triangular inequality we have m


X X m Xm
ai ≤ |ai | ≤ ci ≤ ε



i=n i=n i=n
P
if m ≥ n ≥ N , and by Generalized Theorem 3.22 an converges. 
P P
Theorem. (Generalized Theorem 3.45) If an converges absolutely, then an converges.
P
Proof. If an converges absolutely, applying Generalized Theorem 3.25 where we consider cn = |an |we obtain that required
P
thesis, i.e., an converges. 
P P P
Theorem. (Generalized Theorem 3.47) If an = A converges and bn = B, then (an + bn ) = A + B and Σcan = cA for
any fired c.

Proof. By definition we have


∞ ∞ ∞ ∞ ∞ ∞
! !
X X X X X X X X
an bn = ai,1 , ai,2 , · · · , ai,k = A, = bi,1 , bi,2 , · · · , bi,k =B
i=1 i=1 i=1 i=1 i=1 i=1

so
∞ ∞ ∞
!
X X X X
(an + bn ) = (ai,1 + bi,1 ) , (ai,2 + bi,2 ) , · · · , (ai,k + bi,k )
i=1 i=1 i=1
∞ ∞ ∞ ∞ ∞ ∞
!
X X X X X X
= ai,1 + bi,1 , ai,2 + bi,1 , · · · , ai,k + bi,k
i=1 i=1 i=1 i=1 i=1 i=1
∞ ∞ ∞ ∞ ∞ ∞
! !
X X X X X X
= ai,1 , ai,2 , · · · , ai,k + bi,1 , bi,2 , · · · , bi,k
i=1 i=1 i=1 i=1 i=1 i=1

=A+B

31
and
∞ ∞ ∞
!
X X X X
can = cai,1 , cai,2 , · · · , cai,k
i=1 i=1 i=1
∞ ∞ ∞
!
X X X
= c ai,1 , c cai,2 , · · · , c ai,k
i=1 i=1 i=1
∞ ∞ ∞
!
X X X
=c ai,1 , ai,2 , · · · , ai,k
i=1 i=1 i=1

= cA


P p
Theorem. (Generalized Theorem 3.33) Given an , put α = lim sup n |an |. Then
P n→∞
(a) if α < 1, an converges,-
P
(b) if α > 1, an diverges;
(c) if α = 1, the test gives no information.
P P
Proof. (a) if α < 1, then by Theorem 3.33 |an | converges, so by Generalized Theorem 3.45 an converges.
p
(b) Theorem 3.17 tell us that α = lim nk
|ank |, for some subsequence {ank } . Thus, given 1 < β < α, there exists N ∈ N
such that if k ≥ N , implies
p
nk
|ank | ≥ β
Hence
|ank | ≥ β nk > 1
if k ≥ N .
P
Assumes α > 1 and suppose that an converges, then by Generalized Theorem 3.23

lim |an | = 0
n→∞

in particular
lim |ank | = 0
k→∞

32
P
but this fact contradicts |ank | > 1 if k ≥ N . Therefore, the an series diverges.
(c) The examples given in the book are examples in this case too, taking k = 1 
P
Theorem. (Generalized Theorem 3.34) The series an
(a) converges if lim sup |a|an+1
n|
|
<1
n→∞
(b) diverges if |a|an+1
n|
|
≥ 1 for n ≥ n0 , where n0 is some fixed integer.

Proof. (a.) Theorem 3.34 and the inequality implies that Σ |an | converges. Therefore by Generalized Theorem 3.45 the series
P
an also converges.
|an +1 |
(b) If |a|an+1 |
≥ 1 for n ≥ n0 , then |an+1 | ≥ |an | for n ≥ n0 , Since a0 ≥ 1 we have |an0 | > 0 and note that
n| | n0 |
lim |an | ≥ |an0 | > 0

This inequality implies that an does not tend to zero, so that the series must diverge. 

Theorem. (Generalized Theorem 3.42) Suppose


P
(a) The partial sums An of an form a bounded sequence, i.e., there exist M > 0 such that

Xk
an ≤ M



n=1

for all k ∈ N ;
(b) b0 ≥ b1 ≥ b2 ≥ · · · ;

(c) limn→∞ bn = 0,

Theorem. Then Σbn an converges.

Proof. First note that hypothesis (b) implies |bk − bk+1 | = bk − bk+1 and by (c) we have bk ≥ 0
We show that Σbn an satisfy hypothesis of generalized Theorem 3.22. Indeed, Since ak = Ak − Ak−1 , we have

33

Xm Pm
bk ak =| k=n bk (Ak − Ak−1 )|



k=n
Pm Pm
= | k=n bk Ak − k=n bk Ak−1 |
P
m Pm−1
= k=n bk Ak − k=n−1 bk+1 Ak

P
m−1 Pm−1
= k=n bk Ak − k=n bk+1 Ak − bn An−1 + bm Am

P
m−1
= k=n (bk − bk+1 ) Ak − bn An−1 + bm Am

P 
m−1
≤M k=n |b k − bk+1 | + |bn | + |bm |
P 
m−1
=M k=n kb − bk+1 + bn + b m

≤ 2M bn

ε
Since limn→∞ bn = 0, given ε > 0 choose N so that bn < 2M for all n > N . Thus if m ≥ n > N , we have

Xm
bk ak ≤ 2M bn < ε



k=n

This proves that Σbn an satisfy hypothesis of generalized Theorem 3.22. In consequence, Σbn an converges. 
P
Theorem. (Generalized Theorem 3.55) If an is a series of vectors which converges absolutely, then every rearrangement of
P
an converges, and they all converge to the same sum,

Proof. By the left side in inequality (2), we have


|an,j | ≤ |an |
P
for all j = 1, ..., k, so by hypothesis and comparison test we have that |an,j | converges for all j = 1, ..., k, i.e., each component
P P
series an,j converges absolutely, by Theorem 3.55 we obtain that the every rearrangement of an,j converges, and they all
P
converge to the same sum, let cj = an,j these common values for j = 1, ..., k.

34
P P
Since any rearrangement of aσ(n) induces rearrangements of aσ(n),j for j = 1, ..., k and these rearrangements converges,
P P P P P
then aσ(n) , since aσ(n),j = an,j = cj , then aσ(n) = (c1 , ..., ck ), Thus, every rearrangement of an converges, and
they all converge to the same sum. 

Remarks:

• This exercise shown that some results about numerical series are holds for series in Rk , really this results remain holds in
the general context of finite dimensional vector metric space.

• The series
P P
an that satisfies the thesis in Theorem 3.55, i.e., every rearrangement of an converges are known as
k
unconditionally convergent series, in R is also true the converse of Theorem 3.55.

• Theorem 3.54 also has a generalization, that says that the rearrangements of series that not converges absolutely in Rk
form a affine subspace of Rk , but the proof is not simpler.

35

Exercise 16. Fix a positive number α. Choose x1 > α, and define x2 , x3 , x4 , . . . , by the recursion formula
 
1 α
xn+1 = xn + .
2 xn

[(a)]

1. Prove that {xn } decreases monotonically and that lim xn = α.

2. Put εn = xn − α and show that
ε2n ε2
εn+1 = < √n
2xn 2 α

so that, setting β = 2 α,
 2n
ε1
εn+1 < β (n = 1, 2, 3, . . .).
β

3. This is a good algorithm for computing square roots, since the recursion formula is simple and the convergence is extremely
1
rapid. For example, if α = 3 and x1 = 2, show that ε1 /β < 10 and that therefore

ε5 < 4 · 10−16 , ε6 < 4 · 10−32 .

Hints and Steps:



• For (a) show inductively that xn > α for all positive integers n and use this fact to show that {xn } decreases monotonically,
next use Theorem 3.14

• For (b) use induction and item (a)

• For (c) only do calculations.


2
Solution 16. (a) Note first that expand the inequality(a − b) ≥ 0 we obtain the inequality

a2 + b2
≥ ab (3)
2

36
√ √
Since α > 0, and assumption x1 > α we can be shown by induction that xn > α > 0 for all positive integers n. Indeed
√ √
the inequality (3) when a = xn and b = √xαn , implies


 
1 α √ α
xn+1 = xn + ≥ xn √ = α
2 xn xn

So
x2 + α x2n − α
 
xn − xn+1 = xn − n −= ≥0
2xn 2xn

Thus, {xn } decreases monotonically. Since α < xn ≤ x1 for all positive integers n, the sequence {xn } is bounded and Theorem
3.14 (Monotone Convergence Theorem) implies that {xn } converges. Let x = lim xn . Then Theorem 3.3 implies that

 
1 α
lim xn+1 = lim xn +
n→∞ n→∞ 2 xn
1 α
x = x+
2√ x
x = ± α

Since xn > 0 for all positive integers n, we have x > 0 and hence x = α, as desired.
(b) We have


 
1 α
εn+1 = xn+1 − α = xn +
2 xn
x2n + α √
= − α
2xn
√ 2
(xn − α) ε2
= = n
2xn xn 2xn
√ √
Since we see in item (a) that xn > α > 0 for all positive integers n, we have for β = 2 α that

ε2n ε2 ε2
εn+1 = < √n = n .
2xn 2 α β

37
2 2k
ε21
 
ε1 ε1
The above inequality say for n = 1 that ε2 < β =β β , and assume inductively that εk+1 < β β for some positive
integer k. For n = k + 1, we have
"   #2  2k+1
2k
ε2k+1 1 ε1 ε1
εk+2 < < β =β .
β β β β
Hence it follows from induction that
 2n
ε1
εn+1 < β
β
for n = 1, 2, ....
√ √ √ √
(c) Taking x1 = 2, α = 3, Since 2 = x1 > α= 3, clearly we have β = 2 α = 2 3 < 4. And, since 25 < 27, we have
√ √
5 < 3 3 and so20 < 12 3 or equivalently
√ √
10(2 − 3) < 2 3
√ √ √
. Since ε1 = x1 − 3=2− 3 and β = 2 3, we have
10ε1 < β.

hence
ε1 1
< .
β 10
By part (b), we have

√ √
ε5 < 2 3 · 10−16 < 4 · 10−16 and ε6 < 2 3 · 10−32 < 4 · 10−32 .

Remarks:

• In the chapter 5 the recursive formula will be obtained as application of the Newton’s method to find roots.

38

Exercise 17. Fix α > 1. Take x1 > α, and define

α + xn α − x2n
xn+1 = = xn + .
1 + xn 1 + xn
[(a)]

1. Prove that x1 > x3 > x5 > · · · .

2. Prove that x2 < x4 < x6 < · · · .



3. Prove that lim xn = α.

4. Compare the rapidity of convergence of this process with the one described in Exercise 16.

Hints and Steps:

• Use the recursive formula to find a recursive formula for xn+1 in function of xn−1 and use this to show that the sequence

{xn } oscillate around α, deduce of these facts (a) and (b).

• For (c) estimate |x2n+1 − α| and show that the sequences {x2n } and {x2n−1 } both converges to α.

• For (d) use the inequality obtained in (c) to find upper bounds of the rapidity of convergence of this process

Solution 17. First note that

 2 2 2
α + x1 (α + x1 ) − α (1 + x1 )
x22 −α= −α= 2
1 + x1 (1 + x1 )

α + x21 (1 − α)
= 2 <0
(1 + x1 )
because α > 1
Also note that
α + xn α + α+xn−1
1+xn−1 2α + (α + 1) xn−1
xn+1 = = α+xn−1 = (4)
1 + xn 1 + 1+xn−1 (α + 1) + 2xn−1

39
So

2
(2α + (α + 1) xn−1 )
x2n+1 − α = 2 −α
((α + 1) + 2xn−1 )
2 2
(2α + (α + 1) xn−1 ) − ((α + 1) + 2xn−1 ) α
= 2
((α + 1) + 2xn−1 )
   
2 2
4α + 4α (α + 1) xn−1 + (α + 1) x2n−1 − α (α + 1) + 4α (α + 1) xn−1 + 4αx2n−1
2

= 2
((α + 1) + 2xn−1 )
2
 2 2
 2 
4α α − xn−1 + (α + 1) xn−1 − α (α − 1) x2n−1 − α
= 2 = 2 (5)
((α + 1) + 2xn−1 ) ((α + 1) + 2xn−1 )
and

2α + (α + 1) xn−1
xn+1 − xn−1 = − xn−1
(α + 1) + 2xn−1
(2α + (α + 1) xn−1 ) − ((α + 1) + 2xn−1 ) xn−1
=
(α + 1) + 2xn−1
2

2 α − xn−1
= (6)
(α + 1) + 2xn−1
√ √ √ √ √
Equation (5) tell is that if xn−1 < α, then xn+1 < α, and if xn−1 > α, then xn+1 > α. Thus, x2n−1 > α, and

x2n < α for all n ∈ N . Therefore, by equation (6)

2 α − x2n−1
x2n+1 − x2n−1 = <0
(α + 1) + 2xn−1
hence x2n+1 < x2n−1 and (a) is true.
And 
2 α − x22n
x2n+2 − x2n = >0
(α + 1) + 2x2n
hence x2n < x2n+2 and (b) is true.

40
(c) Equation (5), also imply

|x2n+1 − α| < r2 x2n−1 − α ,


α−1
where r = α+1 < 1. So

2
xn+2k − α ≤ r2k x2n−1 − α

Hence

lim x2n+2k − α ≤ lim r2k x2n−1 − α = 0.



k→∞ k→∞

Therefore,

lim xn+2k = α,
k→∞

whether n is odd or even, and so


lim xn = α.
n→∞

(d) Let εn = |xn − α|. the by the recursive definition we have
√ √ √ √ √


α + xn − α − αxn (xn − α) (1 − α) (1 − α)
εn+1 = |xn+1 − α| = = =
1 + xn εn .

1 + xn 1 + xn
√ √ 2
(1 − α) (1 − α)
= εn = εn−1
1 + xn (1 + xn ) (1 + xn−1 )
α+xn−1
Since xn = 1+xn−1 , then
1 + α + 2xn−1
1 + xn =
1 + xn−1
So (1 + xn )(1 + xn−1 ) = 1 + α + 2xn−1 > 1 + α. Thus,
√ √ √
(1 − α)2 (1 − α)2 α−1
εn+1 = εn−1 < εn−1 = √ εn−1
(1 + xn )(1 + xn−1 ) 1+α α+1

41
which implies that
√ n √ n
α−1 α−1
ε2n+1 < √ ε1 and ε2n+2 < √ ε2 (7)
α+1 α+1
where n is a positive integer.
We now compare the rapidity of convergence of the process with the same data as in Problem 3.16, i.e., α = 3 and x1 = 2.
√ √ √
Then we have √3−1
3+1
3
< 10 3
, ε1 = |2 − 3| < 10 and ε2 = | 35 − 3| < 10
1
. these values and the inequalities (7), implies
 n+1  n
3 1 3
ε2n+1 < and ε2n+2 < · ,
10 10 10

where n is a positive integer. Put n = 2 in the above estimates, we have


 3  2
3 1 3
ε5 < and ε6 < · .
10 10 10

These upper bounds are more greater that the obtain in Exercise 3.16 so the algorithm described in Exercise 3.16 is much faster
than the one in this exercise.

Remarks:
Here we have other algorithm to find the root of a positive number, the recursive formula of this exercise is similar to one
that Rudin’s book uses in chapter 1 to obtain a sequence of rational numbers that approximates the root of 2.

42
Exercise 18. Replace the recursion formula of Exercise 16 by
p−1 α
xn+1 = xn + x−p+1
p p n

where p is a fixed positive integer, and describe the behavior of the resulting sequences {xn }.

Hints and Steps:


1 1
• First show inductively that if x1 > α p , then xn > α p for all p, next show that the sequence {xn } is is monotone decreasing
1
and bounded below and note that the limit of this sequence is α p .
1
Solution 18. This exercise is a generalization of Exercise 16 which is the case p = 2. Assume as Exercise 16 that x1 > α p , and
1 1
suppose inductively that xn > α p , we see now that xn+1 > α p . Indeed,
p
First we show that if p is an integer number with p > 1, and 0 < x < 1, then (1 − x) > 1 − px
for p = 2, we have
2
(1 − x) = 1 − 2x + x2 > 1 − 2x
p
Suppose inductively that (1 − x) > 1 − px. Since (1 − x) > 0 , then for p + 1 we have

p+1 p
(1 − x) = (1 − x) (1 − x) > (1 − px) (1 − x) = 1 − (p + 1) x + px2 > 1 − (p + 1) x

which complete the inductive step.


1 1
αp αp p
By inductive hypothesis we have 1 > xn > 0, so taking 1 − x = xn , in the inequality (1 − x) > 1 − px, we have

1
! 1
!
α αp xn − α p
p >1−p 1− =1−p .
xn xn xn
So 1
!
xn − α p α
p >1−
xn xpn
or equivalently
1 xn α
xn − α p > − p−1
p pxn

43
hence
p−1 α xn α 1
xn+1 = xn + p−1 = xn − + p−1 > α p
p pxn p pxn
The above inequality also imply

xn α 1
xn+1 − xn = − + p−1 = p−1 (α − xpn ) < 0
p pxk pxk
1
for every positive integer n. Thus, the sequence {xn } is monotone decreasing and bounded below by α p , hence this sequence
converges to some z, taking limits in the recursive formula we have
 
p−1 α p−1 α
z = lim xn+1 = lim xn + p−1 = z + p−1 ,
n→∞ n→∞ p pxn p pz

hence
1 α − zp
0 == ,
p pz p−1
1
thus z = α p
To analyze the convergence rapidity, first note that

(1 − x) x + x2 + · · · + xp−1 = x − xp


1
αp
In particular if x = xn we obtain

1
! 1
! 1
!p−1  1 1
!p
αp  α
p αp = α −
p αp
1− + ··· +
xn xn xn xn xn

or equivalently 
1
! 1
!p−1 
 α αp α
 1
 p 1
xn − α p + ··· +  = αp − .
xn xn xp−1
n

Thus,

44
1 p−1 α 1 p−1 α 1 1 1 1 1
xn+1 − α p = xn + x−p+1 − αp = xn + x−p+1 − αp + αp − αp
p p n p p n p p
p−1 1
 1 α 1

= xn − α p + − α p
p p xnp−1
  ! !p−1 
1 1
p−1 1
 1  1
 α p α p
= xn − α p −  xn − α p  + ··· + 
p p xn xn
  ! !p−1 
1 1
 1
 p − 1 1 α p α p
= xn − α p  −  + ··· + 
p p xn xn
  !p−1 !p−1 !p−1 
1 1 1
 1
 p − 1 1 α p α p α p
< xn − α p  −  + ··· + 
p p xn xn xn
 !p−1 
1
 1
 p − 1 α p
= xn − α p 1 − 
p xn
 1
 p − 1   1 p−1 
p−1
= xn − α p x n − αp
pxp−1
n
 1
2 p − 1 h 1 p−2
i
= xn − α p · p−1 · xp−2 n + x p−3 p
n α + · · · + α p

pxn
 1
2 p−1 
xn − α p · p−1 · xnp−2 + xp−2 + · · · + xnp−2

< n
pxn
 1
2 p − 1 
xn − α p · p−1 · (p − 1) xp−2

= n
pxn
2
 1
2 (p − 1)
= xn − α p ·
pxn
 1
2 (p − 1)2
< xn − α p · 1
pα p
1
1
pα p
Therefore, we can guarantee quadratic-geometric convergence, starting with ε1 = x1 − α p < β = (p−1)2
. In that case we

45
obtain the same inequalities as in Exercise 16.

Remarks:

• The recursive formula of this exercise give us an algorithm to find the n-root of a positive number, the recursive formula of
this exercise is similar to the arguments that Rudin’s book uses in chapter 1 in the proof of Theorem 1.21. This recursive
formula also can be obtained as application of the Newton’s methods.

46
Exercise 19. Associate to each sequence a = {αn }, in which αn is 0 or 2, the real number

X αn
x(a) = .
n=1
3n

Prove that the set of all x(a) is precisely the Cantor set described in Sec. 2.44.

Hints and Steps:

• Draw a figure with the steps to construct the Cantor set, and note that the third middle removed are the points in the
[0, 1] with 1’s in its ternary representation.

Solution 19. Recall that the Cantor set P is defined as



\
P = En ,
n=0

where E0 = [0, 1]. Removing the the middle third interval of this interval we obtain E1 = E0 − ( 31 , 23 ), so E1 is [0, 31 ] ∪ [ 23 , 1].
Then, we remove the middle third intervals of these intervals and we obtain E2 , continuing in this way, we obtain a sequence of
compact sets En , such that
E0 ⊃ E1 ⊃ E2 ⊃ · · · ,

and En is the union of 2n intervals, each one have length 3−n . The figure below sketch the construction of the Cantor set

E0
0 1 2 1
3 3

E1
E2
E3

Figure 2: Construction of the Cantor set, firts four steps

Now we show that P can be identified with the set of the sequences (α0 α1 . . . αn . . .) where αn = 0 or αn = 2. In other words,
if we write the Cantor set in ternary we can see that at each step, n = 1, 2, 3, . . . of the construction we remove numbers with

47
1
ternary representations that include a 1 in the nth decimal place. Since we can write 3 as .1 or .0222 . . . Taking the .0222 . . .
representation, we see that every number in the set P has a ternary representation of only 0 and 2 (any number with a ternary
expansion with 1 is removed).

Remarks:
Here we obtain another characterization of the Cantor, using this characterization we also can prove that there exist a
P∞ αn P∞ βn
n=1 3n . →
bijection between the Cantor set and the interval [0, 1] , the map n=1 2n . where βn = 0 if αn = 0 and βn = 1 if
αn = 2 assigns to each element in the Cantor set a element of [0, 1] . This also show that the Cantor set is uncountable.

Exercise 20. Suppose {pn } is a Cauchy sequence in a metric space X, and some subsequence {pni } converges to a point p ∈ X.
Prove that the full sequence {pn } converges to p.

Hints and Steps:

• Only use the definition of Cauchy sequence and triangle inequality

Solution 20. Let ε > 0. Since {pn } is a Cauchy sequence, choose N1 so that
ε
d(pm , pn , ) <
2
if m, n ≥ N1
Now let {pnk } the subsequence of {pn } that converges, then choose N ≥ N1 so that
ε
d(pnk , p) <
2
if nk ≥ N . Thus, if n > N taking nk ≥ N , so nk , n ≥ N ≥ N1 ,hence by triangle inequality we have

ε ε
d(pn , p) ≤ d (pn , pnk ) + d (pnk , p) < + =ε
2 2
Hence the full sequence {pn } converges to p.

Remarks:
This result tell us that in the case of Cauchy sequences is enough to find the limit of some subsequence to find the limit of
the entire sequence.

48
Exercise 21. Prove the following analogue of Theorem 3.10(b): If {En } is a sequence of closed nonempty and bounded sets in
a complete metric space X, if En ⊃ En+1 , and if
lim diam En = 0,
n→∞
T∞
then 1 En consists of exactly one point.

Hints and Steps:



\
• Show that for all sequence {xn } with xn ∈ En , then {xn } is a Cauchy sequence in.show that the limit of En =
1
{limn→∞ xn }

\
Solution 21. Here we need suppose that En 6= ∅, because if En = ∅ for some n ∈ N, then En = ∅.
1
Since limn→∞ diamEn = 0, then given ε > 0, there is an integer N such that diamEn < ε for n ≥ N , Suppose that

\
x, y ∈ En , then x, y ∈ En for all n ∈ N, so d(x, y) ≤ diamEn for all n ∈ N and by the noted above d(x, y) ≤ ε for all ε > 0, in
1

\
consequence d(x, y) = 0 and x = y. So En has at most a point.
1
For each n ∈ N, consider xn ∈ En , if ε, N are as the previous paragraph, let n, m ≥ N , since En ⊃ EN , Em ⊃ EN , then

d(xn , xm ) ≤ diamEN ≤ ε.

So {xn } is a Cauchy sequence in X and since X is a complete metric space this sequence converges to some x in X, i.e.,

lim xn = x,
n→∞

Also note that the hypothesis En ⊃ En+1 implies that the subsequence {xk }k≥n ⊂ En , since

lim xk = x,
k≥n

\ ∞
\
and En is closed, then x ∈ En for all n ∈ N, hence x ∈ En , and by the shown above {x} = En .
1 1

Remarks:
This results is a similar to Corollary of Theorem 2.36 for nested sequences of compact sets.

49
Exercise 22. Suppose X is a nonempty complete metric space, and {Gn } is a sequence of dense open subsets of X. Prove
T∞
Baire’s Theorem, namely, that 1 Gn is not empty. (In fact, it is dense in X.) Hint: Find a shrinking sequence of neighborhoods
En such that En ⊂ Gn , and apply Exercise 21.

Hints and Steps:

• Follow the given hint. Note that N r1 (x) = {y : d (y, x) ≤ r1 } ⊂ Nr (x), if r1 < r and N r1 (x) are closed sets.

Solution 22. We claim that if G is dense in X and x ∈ X, then for any the open ball Nr (x) there exists x1 ∈ G and r1 so that

x1 ∈ N r1 (x1 ) ⊂ G ∩ Nr (x)

and r1 < 2r . Indeed, Since G is a dense subset of X, there exists x1 ∈ G such that x1 ∈ Nr (x). Since G1 and Nr (x) are open in
X also is open G ∩ Nr (x), so there exists s1 so that x1 ∈ Ns1 (x1 ) ⊂ G ∩ Nr (x). Note that

s1 ≤ d (x1 , x) ≤ r

s1
Now choose r1 < 2 , then
x1 ∈ N r1 (x1 ) ⊂ Ns1 (x1 ) ⊂ G1 ∩ Nr (x)

Here N r1 (x1 ) = {y : d (y, x1 ) ≤ r1 }, see the figure below to have an idea of as choose r1 .

50
N r1 (x1 )
Nr (x)
x1
x
Ns (x1 )

Figure 3: Construction of N r1 (x1 )

Since G1 is open and dense, fix an open ball Nr (x), by the say above there exist x1 ∈ G1 and r1 so that

x1 ∈ N r1 (x1 ) ⊂ G1 ∩ Nr (x)

and r1 < 2r .
Considering G1 and Nr1 (x1 ), then there exist x2 ∈ G2 and r2 so that

x2 ∈ N r2 (x2 ) ⊂ G2 ∩ Nr1 (x1 )

51
r1 r
and r2 < 2 < 22 . Continuing in this way for each k ∈ N we can get xk ∈ Gk and rk so that

xk ∈ N rk (xk ) ⊂ Gk ∩ Nrk−1 (xk−1 )

r
and rk < 2k
. Since Ek = N rk (xk ) are closed and bounded sets and Ek ⊃ Ek+1
r
lim diamEk ≤ lim =0
k→∞ k→∞ 2k−1

\
Since X is a nonempty complete metric space, Problem 3.21 implies that En is non empty. Since En ⊆ Gn for each positive
n=1

\
integer k, we have Gn is non empty, completing the proof.
n=1

Remarks:
For a set A ⊆ X we say that A is nowhere dense in X if the interior of its closure is empty, A is of first category or meagre
in X if it is a union of countably many nowhere dense subsets and A is of second category or nonmeagre in X if it is not of first
category in X.
The Baire’s category theorem say that complete metric spaces are of second categories.

52
Exercise 23. Suppose {pn } and {qn } are Cauchy sequences in a metric space X. Show that the sequence {d(pn , qn )} converges.
Hint: For any m, n,
d(pn , qn ) ≤ d(pn , pm ) + d(pm , qm ) + d(qm , qn );
it follows that
|d(pn , qn ) − d(pm , qm )|
is small if m and n are large.

Hints and Steps:

• Only follow the given hints

Solution 23. Applying twice the the triangle inequality, we have .

d(pn , qn ) ≤ d(pn , pm ) + d(pm , qn ) ≤ d(pn , pm ) + d(pm , qm ) + d(qm , qn )


from this inequality we obtain
d(pn , qn ) − d(pm , qm ) ≤ d(pn , pm ) + d(qm , qn )
and since the same inequality holds interchanging m and n, it follows that

|d(pn , qn ) − d(pm , qm )| ≤ (pn , pm ) + d(qm , qn ).


ε
Since {pn } and {qn } are Cauchy sequences in a metric space X, given ε > 0, choose N1 and N2 so that d(pn , pm ) < 2 if
ε
m > N1 , n > N1 , and d(qn , qm ) < 2 if m > N2 , n > N2 . Then taking N = max(N1 , N2 ). We have

ε ε
|d(pn , qn ) − d(pm , qm )| ≤ (pn , pm ) + d(qm , qn ) <
+ =ε
2 2
if m > N and n > N . So the {d(pn , qn )} is a Cauchy sequence in the complete metric space of all real numbers, so {d(pn , qn )}
converges.

Remarks:
This result will allow us to show in the case of a complete metric space (X, d), that the distance d : X ×X → R is a continuous
function.

53
Exercise 24. Let X be a metric space.

1. Call two Cauchy sequences {pn }, {qn } in X equivalent if

lim d(pn , qn ) = 0.
n→∞

Prove that this is an equivalence relation.

2. Let X ∗ be the set of all equivalence classes so obtained. If P ∈ X ∗ , Q ∈ X ∗ , {pn } ∈ P , {qn } ∈ Q, define

∆(P, Q) = lim d(pn , qn );


n→∞

by Exercise 23, this limit exists. Show that the number ∆(P, Q) is unchanged if {pn } and {qn } are replaced by equivalent
sequences, and hence that ∆ is a distance function in X ∗ .

3. Prove that the resulting metric space X ∗ is complete.

4. For each p ∈ X, there is a Cauchy sequence all of whose terms are p; let Pp be the element of X ∗ which contains this
sequence. Prove that
∆(Pp , Pq ) = d(p, q)

for all p, q ∈ X. In other words, the mapping ϕ defined by ϕ(p) = Pp is an isometry (i.e., a distance-preserving mapping)
of X into X ∗ .

5. Prove that ϕ(X) is dense in X ∗ , and that ϕ(X) = X ∗ if X is complete. By (d), we may identify X and ϕ(X) and thus
regard X as embedded in the complete metric space X ∗ . We call X ∗ the completion of X.

Hints and Steps:

• For (a) only check the property of equivalence relation.

• For (b) check that ∆ satisfy the property of distances.

• Show first items (d) and (e) and then show that X ∗ is complete.

54
Solution 24. (a) We need to show that:
1) {pn } is equivalent to itself. Indeed, since d(pn , pn ) = 0, then lim d(pn , pn ) = 0
n→∞
2) If {pn } is equivalent to {qn }, then {qn } is equivalent to {pn }, this follow because d(pn , qn ) = d(qn , pn ), hence if
lim d(pn , qn ) = 0, we have
n→∞
lim d(qn , pn ) = lim d(pn , qn ) = 0
n→∞ n→∞

and
3) If {pn } is equivalent to {qn } and {qn } is equivalent to {rn }, then {pn } is equivalent to {rn }. Indeed, from the triangle
inequality, we have d(pn , rn ) ≤ (pn , qn ) + d(qn , rn ). So if lim d(pn , qn ) = 0 and lim d(qn , rn ) = 0, then
n→∞ n→∞

0 ≤ lim d(pn , rn ) ≤ lim d(pn , qn ) + lim d(pn , qn ) = 0.


n→∞ n→∞ n→∞

So lim d(pn , rn ) = 0.
n→∞
(b) Let {pn } be equivalent to {un } and {qn } equivalent to {vn }, Then applying twice the the triangle inequality, we have

d(pn , qn ) ≤ d(pn , un ) + d(un , qn ) ≤ d(pn , un ) + d(un , vn ) + d(vn , qn )

from this inequality we obtain


d(pn , qn ) − d(un , vn ) ≤ d(pn , un ) + d(vn , qn )

Interchanging pn with un and qn with vn , it follows that

|d(pn , qn ) − d(un , vn )| ≤ d(pn , un ) + d(vn , qn ).

we know by exercise 3,23 that all the limits exist, Thus

0 ≤ lim |d(pn , qn ) − d(un , vn )| ≤ lim d(pn , un ) + lim d(vn , qn ) = 0


n→∞ n→∞ n→∞

So lim |d(pn , qn ) − d(un , vn )| = 0 , which imply lim d(pn , qn ) = lim d(un , vn ), so ∆(P, Q) is well defined.
n→∞ n→∞ n→∞
Now we check ∆ is a metric in X ∗ , clearly ∆(P, Q) ≥ 0, and by definition ∆(P, Q) = 0 means P = Q,

55
Since d(pn , qn ) = d(qn , pn ) for every {pn } ∈ P and {qn } ∈ Q, we have

∆(P, Q) = lim d(pn , qn ) = lim d(qn , pn )∆(Q, P ).


n→∞ n→∞

Now, the triangle inequality tell us that d(pn , qn ) ≤ d(pn , rn ) + d(rn , qn ) for every {pn } ∈ P , {qn } ∈ Q and {rn } ∈ R, so

∆(P, Q) = lim d(pn , qn ) ≤ lim d(pn , rn ) + lim d(rn , qn ) = ∆(P, R) + ∆(R, Q)


n→∞ n→∞ n→∞

so ∆ satisfy the definition of metric.


Before showing that X ∗ is complete, we will first demonstrate items (d) and (e)
(d) Let p, q ∈ X and consider the constant sequences (p, ṕ, p, p, ...) ∈ Pp and (q, q, q, q, ...) ∈ Pq , then

∆(P, Q) = lim d(p, q) = d(p, q).


n→∞

(e) We will now show that ϕ (X) is dense in X ∗ . To do this, let P ∈ X ∗ and let (pn ) be a representative of class P . Let
ε
ε > 0. Since {pn } is a Cauchy sequence in X, then there is an index N such that ifn, m ≥ N , then d(pn , pm ) < 2. Let us
consider the constant sequence (pN , pN , pN , ...), whose equivalence class is ϕ (pN ). So,
ε
∆(P, ϕ (pN )) = lim d(pn , pN ) ≤ <ε
n→∞ 2
Thus, ϕ (X) is dense in X ∗ .
Now, suppose that X is complete. Let P ∈ X ∗ and {pn } ∈ P . Since {pn } is a Cauchy sequence in the complete metric space
X, we have pn → p for some p ∈ X. Thus we have

∆(P, Pp ) = lim d(pn , p) = 0.


k→∞

Hence we have P = Pp and then ϕ(X) = X ∗ .


(c) Now, we are going to show that X ∗ is complete. To do this, let us first note that if {pn } is a Cauchy sequence in X, and
{pn } ∈ P then
lim ∆(P, ϕ (pk )) = lim lim d(pn , pk ) = 0
k→∞ k→∞ n→∞

So the sequence {ϕ (pk )}is a Cauchy sequence that converges to P in X ∗

56
Suppose {Pk }∞ ∗ ∗
k=1 is a Cauchy sequence in X and let ε > 0, for each k let {pk,n } ∈ Pk . Since ϕ (X) is dense in X we can
choose qk in X such that
ε
∆ (Pk , ϕ (qk )) = lim d(pk,n , qk ) < ,
n→∞ 3k
.
for k = 1, 2, ...
Since{Pk }∞ ∗
k=1 is a Cauchy sequence in X , there exists a positive integer N , such that
ε
∆ (Pk , Pl ) = lim d(pk,n , pl,n ) <
n→∞ 3
if k, l ≥ N . Let and define pk = pk,Nk . so
.
By the triangle inequality, for any k, l, and n we have

d(qk , ql ) ≤ d(qk , pk,n ) + d(pk,n , pl,n ) + d(pl,n , ql )


Hence, taking limits when n tends to ∞ and assuming k, l > N , we obtain

ε ε ε
d(qk , ql ) ≤ lim (d(qk , pk,n ) + d(pk,n , pl,n ) + d(pl,n , ql )) <
+ + <ε
n→∞ 3k 3 3l
.then we have that {qk } is a Cauchy sequence in X. Let P be the element of X ∗ such that {qk } ∈ P . We claim Pk → P in X ∗ .
Indeed,

ε
∆(Pk , P ) = lim d(pk,n , qn ) <
n→∞ 3k
So
ε
0 ≤ lim ∆(Pk , P ) ≤ lim =0
k→∞ k→∞ 3k

So any Cauchy sequence in X converges, hence X ∗ is complete.


Remarks:
This results is known as the Completion Theorem for Metric Spaces.

57
Exercise 25. Let X be the metric space whose points are the rational numbers, with the metric d(x, y) = |x − y|. What is the
completion of this space? (Compare Exercise 24.)

Hints and Steps:

• Only recall that Q is dense in R and use exercise 24.

Solution 25. By the Exercise 24, the completion of a metric space X is any complete metric space containing a dense subset
isometric to the space X . Since the real numbers have this property (recall that the rationals numbers are dense in the real
numbers see remark of Theorem 1.20), then the completion of the rational numbers is the real numbers.

Remarks:
There exists other completions of Q distinct of R, but we need consider another metrics in Q.

58

You might also like