Lecture Notes On Advanced Calculus II

Jie Wu
Department of Mathematics
National University of Singapore
 Contents

Chapter 1. Sequences of Real Numbers 5

1. Sequences 5
2. Limits of Sequences 5
3. Sequences which tend to ∞ 8
4. Techniques For Computing Limits 9
5. The Least Upper Bounds and the Completeness Property of R 12
6. Monotone Sequences 15
7. Subsequences 16
8. The Limit Superior and Inferior of a Sequence 17
9. Cauchy Sequences and the Completeness of R 25
Chapter 2. Series of Real Numbers 31
1. Series 31
2. Tests for Positive Series 34
3. The Dirichlet Test and Alternating Series 43
4. Absolute and Conditional Convergence 49
5. Remarks on the various tests for convergence/divergence of series 50
Chapter 3. Sequences and Series of Functions 53
1. Pointwise Convergence 53
2. Uniform Convergence 57
3. Uniform Convergence of {Fn } and Continuity 63
4. Uniform Convergence and Integration 65
5. Uniform Convergence and Differentiation 70
6. Power Series 75
7. Differentiation of Power Series 83
8. Taylor Series 87
Bibliography 97

3
 CHAPTER 1

Sequences of Real Numbers

1. Sequences
A sequence is an ordered list of numbers. For example,
1, 2, 3, 4, 5, 6
The order of the sequence is important. For example,
2, 1, 4, 3, 6, 5
is different from above sequence. An infinite sequence is a list which does not end.
For example,
1, 1/2, 1/3, 1/4, 1/5, · · ·
We are going to study infinite sequences. We denote by {an } the sequence
a1 , a2 , a3 , · · · , an , · · ·
Example 1.1. Here are some examples of infinite sequences.
1 1 1
(1). 1, , , · · · , , · · ·
2 3 n
1 1 1 1
(2). , , , ,···
3 32 33 34
(3). 1, −2, 3, −4, 5, · · ·

Can you find a formula for each of the above sequences?

Answer: (1). an = 1/n. (2). an = 1/3n . (3). (−1)n−1 n.

2. Limits of Sequences
Definition 2.1. The limit of {an } is A, and is written as
lim an = A,
n→∞

if for any  > 0, there is a natural number N such that for every n > N , we have
|an − A| < .
Remark. 1. Some sequences do not satisfy the above. We call such sequences
divergent.
2. Sequences which satisfy the above definition, i.e. A exists and is finite, are called
convergent sequences.

Example 2.2. Prove the following limits by using  − N definition

1
1) lim = 0.
n→∞ n
5
6 1. SEQUENCES OF REAL NUMBERS
r
n2
2) lim 2
= 1.
n→∞
 nn+ 1
3
3) lim = 0.
n→∞ 4
 
1 
Solution. (1). Given any  > 0, we want to find N such that  − 0 <  for
n
1 1
n > N , i.e., n > for n > N . Choose N to be the smallest integer such that N ≥ .
   
1 1  1
(N is found now!) When n > N , then n > N ≥ or  − 0 < . Thus lim = 0.
 nr  n→∞ n
 n2 
(2). Given any  > 0, we want to find N such that  − 1 <  for n > N .
 
 n +1 2 
Now
 n − √n 2 + 1 
r     
 n 2   n 
− 1 <  ⇔  √ − 1 <  ⇔  √ <
   
 n2 + 1
 2 2
 n +1  n +1 
n2 − (n2 + 1)
 
  1
⇔ √
 √ <⇔ √ √ <
2
n + 1(n + n + 1) 2  n + 1(n + n2 + 1)
2

√ √ 1
⇔ n2 + 1(n + n2 + 1) >

Observe that √ √
n2 + 1(n + n2 + 1) > n
for n ≥ 1. Choose N to be the smallest integer such that N ≥ 1 . Then, for n > N ,
√ √ √ √ 1
n2 + 1(n + n2 + 1) > N 2 + 1(N + N 2 + 1) > N ≥

r 
 n2 
or  − 1  < . Thus N is found and hence the result.
 
2
 n +1 
  n 
 3 
(3). Given any  > 0, we want to find N such that  − 0 <  for n > N .
4
Observe that  n  
3 3 ln()
<  ⇔ n ln < ln() ⇔ n >
4 4 ln(3/4)
(Note. ln(3/4) < 0!!) Choose N to be the smallest positive integer such that
ln()
N≥ . When n > N , then
ln(3/4)
ln()
n>N ≥
ln(3/4)
  n 
 3 
or  − 0 < . The proof is finished. 
4

Theorem 2.3. If {an } has a limit, then the limit is unique.

 2. LIMITS OF SEQUENCES 7

Proof. Let A and B be limits of {an }. Suppose that A 6= B. Choose  =

|A − B|
. Then  > 0 because A 6= B. By definition, there exists N1 and N2 such
2
that |an − A| <  for n > N1 and |an − B| <  for n > N2 . For n > max{N1 , N2 }, we
have
|A − B|
|A − B| = |(A − an ) + (an − B)| ≤ |A − an | + |an − B| < 2 = 2 = |A − B|,
2
which is a contradiction. Thus A = B. 
Theorem 2.4 (Squeeze or Sandwich Theorem). Given 3 sequences
{an }, {bn }, {cn }
such that
(i) an ≤ b n ≤ c n for every n and
(ii) lim an = A = lim cn ,
n→∞ n→∞
then lim bn = A.
n→∞

Proof. For any  > 0, there exists N1 and N2 such that |cn − A| <  for n > N1
and |an − A| <  for n > N2 . Let N = max{N1 , N2 }. For n > N , we have
− < cn − A <  and −  < an − A < 
A −  < cn < A +  and A −  < an < A + .
Thus
A −  < an ≤ bn ≤ cn < A +  or |bn − A| < .
By definition, we have lim bn = A and hence the result. 
n→∞

Remark. The above theorem is still applicable if the inequality

an ≤ b n ≤ c n
is true eventually.

Example 2.5. Find limits

1 + sin n
1) lim .
n→∞
 n
n
3n − 1
2) .
4n + 1
Solution. (1). Since
1 + sin n 2
0≤ ≤
n n
2
and lim = lim 0 = 0, we have
n→∞ n n→∞
1 + sin n
lim = 0.
n→∞ n
(2). Since  n  n
3n − 1 3
0≤ ≤
4n + 1 4
8 1. SEQUENCES OF REAL NUMBERS
 n
3
and lim = lim 0 = 0, we have
n→∞ 4 n→∞
 n
3n − 1
lim = 0.
n→∞ 4n + 1


3. Sequences which tend to ∞

Definition 3.1. {an } tends to +∞ if for each positive number k, there is an N
such that
an > k for all n > N.

Remark. For such sequences, we write as an → +∞ as n → ∞ or

lim an = +∞.
n→∞

Example 3.2. The following sequences tend to +∞

√
1) an = ln n.
2) an = (3/2)n .
The sequences − ln n, −n2 and etc then tend to −∞.
Theorem 3.3 (Reciprocal Rule). Consider a sequence {an }.
1
(i) If an > 0 for all n and lim = 0, then
n→∞ an

lim an = +∞.
n→∞

1
(ii) If lim an = ±∞, then lim = 0.
n→∞ n→∞ an

Proof. We only prove (i). For each positive integer k, there exists N such that
 
1
 − 0 < 1

 an  k

1
for n > N because lim = 0. Then, for n > N , an > k because an > 0. This
n→∞ an
finishes the proof. 

1 √
Example 3.4. Since lim √ = 0, we have lim n = ∞. Similarly, since
n→∞ n n→∞
√ 1
lim n = +∞, we have lim √ = 0.
n→∞ n→∞ n
 4. TECHNIQUES FOR COMPUTING LIMITS 9

4. Techniques For Computing Limits

Theorem 4.1. Let f be a continuous function. Then
lim f (an ) = f ( lim an ).
n→∞ n→∞

Idea of Proof. By the definition of continuity, when x → x0 , f (x) → f (x0 ).

Now lim an = A means that an → A when n → ∞. Thus f (an ) → f (A) when
n→∞
n → ∞, that is,
lim f (an ) = f (A) = f ( lim an ).
n→∞ n→∞

Example 4.2.
   
nπ π π 
lim sin = lim = sin = 1.
n→∞ 2n + 1 n→∞ 2 + 1/n 2
f (n)
Theorem 4.3 (L’Hopital’s Rule). Suppose an = f (n), bn = g(n). If lim is
n→∞ g(n)
∞ 0
of the form or , then
∞ 0
f (n) f 0 (n)
lim = lim 0 .
n→∞ g(n) n→∞ g (n)

History Remark. Although the theorem is named after Marquis de l’Hospital

(1661-1704), it should be called Bernoulli’s rule. The story is that in 1691, l’Hopital
asked Johann Bernoulli (1667-1748) to provide, for a fee, lectures on the new sub-
ject of calculus. L’Hopital subsequently incorporated these lectures into the first
calculus text, L’Analyse des infiniment petis (Analysis of infinitely small quantities),
published in 1696. The initial version of what is now known as l’Hopital’s rule first
appeared in this text.
 x n
Example 4.4. Show that lim 1 + = ex .
n→∞ n
Proof. h x n i  x
lim ln 1 + = lim n ln 1 +
n→∞ n n→∞ n
1 x


x · −


ln 1 + n 1+ nx
n2 x
= lim = lim = lim = x.
n→∞ 1/n n→∞ 1
− n2 n→∞ 1 + x
n
 x n
Thus lim 1 + = ex . 
n→∞ n
Theorem 4.5. If lim an and lim bn exist, then
n→∞ n→∞
(1). lim (an + bn ) = lim an + lim bn ,
n→∞ n→∞ n→∞
(2). lim kan = k lim an ,
n→∞ n→∞
(3). lim an bn = lim an lim bn ,
n→∞ n→∞ n→∞
an lim a n
(4). lim = n→∞ , provided bn 6= 0 and lim bn 6= 0.
n→∞ bn lim bn n→∞
n→∞
10 1. SEQUENCES OF REAL NUMBERS

Proof. omitted. 
2
   
n + 3n + 2 1
Example 4.6. Find the limit of ln + cos √ .
2 + 4n + 2n2 n
Solution.   2   
n + 3n + 2 1
lim ln + cos √
n→∞ 2 + 4n + 2n2 n
  2
(n + 3n + 2)/n2
  
1
= lim ln + cos √
n→∞ (2 + 4n + 2n2 )/n2 n
2
    
1 + 3/n + 2/n 1
= lim ln + cos √
n→∞ 2/n2 + 4/n + 2 n
   
1+0+0 1
= ln + cos 0 = ln + 1 = 1 − ln 2.
0+0+2 2

Theorem 4.7 (Some Standard Limits). Some standard limits are given as follows.
1
1. lim p = 0 for any fixed p > 0.
n→∞ n
2. lim cn = 0 for any fixed c where |c| < 1.
n→∞
1
3. lim c n = 1 for any fixed c > 0.
n→∞ √
4. lim n n = 1.
n→∞
np
5. lim n = 0 for any fixed p and c > 1.
n→∞ c
cn
6. lim = 0 for any fixed c.
n→∞ n!
x n
7. lim 1 + = ex for any fixed x.
n→∞ n
(ln n)p
8. lim = 0 for any fixed k > 0.
n→∞ nk
Proof. Assertion 7 was proved in Example 4.4.
1.
1  1 p
lim
= lim = 0p = 0.
n→∞ np n→∞ n
2. Case 1: When c = 0, the statement is obvious.
Case 2: When c > 0, we have
 
ln lim c = lim ln cn = lim n ln c = −∞.
n
n→∞ n→∞ n→∞
n
Thus, lim c = 0.
n→∞
Case 3: When c < 0, we have −|c|n ≤ cn ≤ |c|n for all n. By Case 2, we have
lim (−|c|n ) = 0 = lim |c|n . Hence by Squeeze theorem, we also have lim cn = 0.
n→∞ n→∞ n→∞
1 1
limn→∞ n 0
3. lim c = cn = c = 1.
n→∞
4.  √  √ ln n
ln lim n = lim ln n n = lim
n
= 0 (by L’Hopital’s rule).
n→∞ n→∞ n→∞ n
 4. TECHNIQUES FOR COMPUTING LIMITS 11
√
Thus, lim n n = e0 = 1.
n→∞
5. Let k be a fixed positive integer such that p − k < 0. Then
np pnp−1 p(p − 1)np−2
lim= lim = lim = ···
n→∞ cn n→∞ cn ln c cn (ln c)2

p(p − 1) · · · (p − k + 1)np−k p(p − 1) · · · (p − k + 1)

= lim n k
= lim =0
n→∞ c (ln c) n→∞ cn nk−p (ln c)k
by L’Hopital’s rule.
cn c · c · ··· · c
6. Let an = = . Now fix an integer M > c. Then for any n > M ,
n! n(n − 1) · · · · · 1
c · c · ··· · c c
0 < an = aM < aM .
n(n − 1) · · · · · (M + 1) n
c
Note that aM is a fixed number because M is fixed. Since lim 0 = 0 = lim aM ,
n→∞ n→∞ n
by the Squeeze theorem, lim an = 0.
n→∞
8. Let m = ln n. Then n = em . By (5),
(ln n)p mp mp
lim = lim = lim m = 0,
n→∞ nk m→∞ ekm m→∞ (ek )

where ek > 1 because k > 0. 

Strategy: One can find the limits of many sequences from those of the standard
sequences.

Example 4.8. Find the limits

8n + (ln n)10 + n!
1) lim 6
.
n→∞
 n − n! 3n
1
2) lim 1 − .
n→∞ 2n + 1
Solution. (1).
8n + (ln n)10 + n! 8n /n! + (ln n)10 /n! + 1 0+0+1
lim 6
= lim 6
= = −1.
n→∞ n − n! n→∞ n /n! − 1 0−1
(2).
" 3n
 3n 2n+1 # 2n+1
1 −1
lim 1− = lim 1+
n→∞ 2n + 1 n→∞ 2n + 1
" 3
2n+1 # 2+1/n
−1
23 1
= lim 1+ = e−1 = √
n→∞ 2n + 1 e e

12 1. SEQUENCES OF REAL NUMBERS

5. The Least Upper Bounds and the Completeness Property of R

5.1. From Natural Numbers to Real Numbers. Starting with natural num-
bers N = {1, 2, 3, · · · }, we obtain real numbers R by adding more and more new
numbers in the following steps:
Step 1. By adding 0 and negative numbers, we  have integers Z = {0,  ±1, ±2, ±3, · · · }.
p 
Step 2. Then we have rational numbers Q =  p, q ∈ Z, q 6= 0 .
q
Step 3. Then, by adding irrational numbers, we have all real numbers.
Below we give some examples of irrational numbers. Recall that a natural num-
ber p > 1 is called prime if p is NOT divisible by any natural numbers other than
p and 1. For instance, 2, 3, 5, 7, 11, · · · are primes. Every natural number n > 1
admits a unique (prime) factorization
n = p1 · p2 · · · pk ,
where each pi is prime. For instance, 20 = 2 · 2 · 5 and 66 = 2 · 3 · 11.
Example 5.1.√If n is a natural number, and there is no natural
√ √number
√ √whose
square is n, then n is NOT a rational number. In particular, 2, 3, 5, 6 are
irrational numbers.
√ √ a
Proof. Suppose that n is a rational number. We can write n as , where
b
a, b ∈ N and b 6= 0. Then
√ a a2
n= ⇔ n= 2 ⇔ a2 = b2 n.
b b
Any prime occurring in the (unique) factorization of a will occur an even number of
times in the factorization of a2 ; similarly for b and b2 . By a2 = b2 n, any prime that
occurs in the factorization of n must occur an even number of times, since all primes
occurring in the factorization of b2 n are exactly those occurring in the factorization
of a2 .
Thus n can be written as
n = (p1 · p2 · · · pk )2 ,
where, of course, the pi ’s need not be distinct. This, however, is a contradiction to
the hypothesis, since n is expressed as the square of a natural number. 

5.2. Bounded Sets.

Definition 5.2. A set of real numbers S is bounded above if there exists a finite
real number M such that
x≤M ∀x ∈ S.
M is called an upper bound of S.
Definition 5.3. A set of real numbers S is bounded below if there exists a finite
real number m such that
m≤x ∀x ∈ S.
m is called a lower bound of S.
 5. THE LEAST UPPER BOUNDS AND THE COMPLETENESS PROPERTY OF R 13

Definition 5.4. A set which is both bounded above and below is called a bounded
set.
Remark.
1. Upper bounds and lower bounds are not unique.
2. Some sets only have upper bounds but not lower bounds.
3. Some sets have only lower bounds but not upper bounds.
4. A set which is not bounded is called an unbounded set.
√
Example 5.5. Let S = {r | r is a rational number with r < 2}. Then S is
bounded above.
Theorem 5.6. Every convergent sequence is bounded.
Proof. Let {an } be a sequence convergent to A. For  = 1, there exists N such
that |an − A| < 1 or A − 1 < an < A + 1 for n > N . Choose M and m to be the
largest and smallest number of the finite numbers
a1 , a2 , . . . , aN , A + 1, A − 1,
respectively. When n ≤ N , we have m ≤ an ≤ M because M (m) is the largest
(smallest) number of the above finite set. When n > N , we have
m ≤ A − 1 < an < A + 1 ≤ M.
Thus, for all n, we have m ≤ an ≤ M and so {an } is bounded. The proof is
finished. 
Corollary 5.7 (Test for divergence). If {an } is unbounded, then {an } diverges.
Remark.
1. The converse may not be true, i.e., divergent sequence need not be unbounded.
2. The inverse may not be true, i.e., a bounded sequence may not be convergent.
Example. The sequence {1, −1, 1, −1, · · · } is bounded but NOT convergent.
5.3. Infimum and Supremum. Recall that any finite set of real numbers has
a greatest element (maximum) and a least element (minimum).
Example 5.8. {−2.5, 3.1, −4.4, 4.5, 5}
However, this property does not necessarily hold for infinite sets.
Example 5.9. {1, 2, 3, 4, · · · , }.
Definition 5.10. A real number M (6= ±∞) is called the least upper bound or
supremum of a set E if
(i) M is an upper bound of E, i.e., x ≤ M for every x ∈ E, and
(ii) if M 0 < M , then M 0 is not an upper bound of E (i.e., there is an x ∈
E such that M 0 < x).
We write M = sup E.
Remark.
(i) sup E is unique whenever it exists.
(ii) The main difference between sup E and max E is that sup E may not be an
element of E, whereas max E must be an element of E if it does exist).
(iii) If E has a maximum, then sup E = max E.
14 1. SEQUENCES OF REAL NUMBERS
√ √
Example 5.11. 1. Let√ E = {r ∈ Q | 0 ≤ r ≤ 2}. Then sup E = 2 but max E
does not exist because 2 is not a rational number, that is, sup E 6∈ E.
2. Let E = {1/2, 2/3, 3/4, 4/5, 5/6, · · · }. Then sup E = 1 and max E does not exist.
3. Let E = {1, 1/2, 1/3, 1/4, 1/5, · · · }. Then max E = 1 = sup E.
Definition 5.12. A real number m (6= ±∞) is called the greatest lower bound
or infimum of a set E if
(i) m is a lower bound of E, i.e., m ≤ x for every x ∈ E, and
(ii) if m0 > m, then m0 is not a lower bound of E (i.e., there exists an
x ∈ E such that x < m0 ).
We write m = inf E.
Remark.
(i) inf E is unique whenever it exists.
(ii) The main difference between inf E and min E is that inf E may not be an
element of E, whereas min E must be an element of E if it does exist.
(iii) If E has a minimum, then inf E = min E.
Example 5.13. 1. Let E = {1, 1/2, 1/3, 1/4, · · · , }. Then inf E = 0 but min E
does not exist. √
2. Let E = {r ∈ Q | 0 ≤ r ≤ 2}. Then min E = inf E = 0.
5.4. The Completeness of R. Consider the set E = {r√∈ Q | r2 < 2}. Then
E is a bounded subset of rational numbers, but sup E = 2 is NOT a rational
number. For considering sup and inf of bounded subsets of rational numbers, we
may obtain irrational numbers. For bounded subsets of real numbers, sup and inf
are still real numbers. This is called completeness property of R. In details, we
have the following.
Theorem 5.14 (Completeness Axiom of R). The following statement hold for
subsets of real numbers:
(i) If E is bounded above, then sup E exists.
(ii) If E is bounded below, then inf E exists.
Remark. For assertion (i), it just means that if a subset of real numbers E is
bounded above, then sup E exists as a real number. Compare with rational case: if
a subset of rational numbers E is bounded above, then sup E exists only as a real
number, but it need not be a rational number.
Remark. Richard Dedekind (a German mathematician, 1831-1916), in 1872, used
algebraic techniques to construct real number system R from Q. His basic ideas are
as follows.
Given a rational number r, we can construct two sets U = {x ∈ Q | x ≥ r}
and L = {x ∈ Q | x < r}. (One can also construct U = {x ∈ Q | x > r} and
L = {x ∈ Q | x ≤ r}.) The sets U and L have the property that
(1). U and L are subsets of Q;
(2). U ∪ L = Q;
(3). U 6= ∅;
(4). L 6= ∅;
(5). U ∩ L = ∅; and
 6. MONOTONE SEQUENCES 15

(6). every element in U is greater than every element in L.

Such a paring (U, L) is called a Dedekind cut. Then we can use inf U (or sup L) to
define a new number. This is Dedekind’s idea to construct all real numbers by using
rational numbers. For instance,
√ let U = {x ∈ Q | x2 > 2} and L = {x ∈ Q | x2 > 2}.
Then inf U = sup L = 2. Another way to construct real numbers using rational
numbers was introduced by Georg Cantor (1845-1917). We will explain Cantor’s
ideas in the section of Cauchy sequences.
Recall that a set E is bounded if and only if it is bounded above and bounded
below. Thus the Completeness Axiom leads to
Corollary 5.15. If E is bounded, then both sup E and inf E exist.

6. Monotone Sequences
Definition 6.1. {an } is called monotone increasing (decreasing) if
an ≤ (≥) an+1
for every n, that is,
a1 ≤ a2 ≤ a3 ≤ a4 ≤ · · ·
(a1 ≥ a2 ≥ a3 ≥ · · · ).
Example 6.2. 1. The sequence {1/n} is monotone decreasing.
2. The sequence {1/2, 2/3, 3/4, 4/5, 5/6, · · · } is monotone increasing.
Proposition 6.3. A monotone increasing (decreasing) sequence is bounded below
(above).
Proof. Let {an } be a monotone increasing sequence, that is,
a1 ≤ a2 ≤ a3 ≤ · · · .
Then a1 is a lower bound for {an } and hence the result. 
Theorem 6.4 (Monotone Convergence Theorem). Let {an } be a sequence.
(i) If {an } is monotone increasing and bounded above, then {an } is convergent
and
lim an = sup an .
n→∞ n
(ii) If {an } is monotone decreasing and bounded below, then {an } is convergent
and
lim an = inf an .
n→∞ n

Proof. (i). Suppose {an } is monotone increasing and bounded above. Then
by the Completeness Axiom of R, sup an exists (finite). Now, given  > 0, since
n
sup an −  < sup an , it follows that sup an −  is not an upper bound of {an }. In other
n n n
words, there exists an integer N such that aN > sup an − . Then for all n > N , we
n
have
sup an −  < aN ≤ an ≤ sup an < sup an +  (since n > N ).
n n n
16 1. SEQUENCES OF REAL NUMBERS
 
 
Equivalently, an − sup an  <  for all n > N and so lim an = sup an (exists).

n n→∞ n
The proof of (ii) is similar. 
n
Example 6.5. Let an = , that is, {an } = {1/2, 2/3, 3/4, · · · }. Then an is
n+1
monotone increasing and bounded above. Thus
sup an = lim an = 1.
n n→∞

Corollary 6.6. If {an } is monotone increasing (decreasing), then either

(i) {an } is convergent or
(ii) lim an = +∞(−∞).
n→∞

Proof. Suppose {an } is monotone increasing, then either {an } is bounded above
or not bounded above.
Case (a): If {an } is bounded above, then by the Monotone Convergence Theorem,
{an } converges.
Case (b): If {an } is not bounded above, then {an } has no upper bounds. Thus for
any given k > 0, k is not an upper bound of {an }. In other words, there exists N
such that
aN > k.
Since {an } is monotone increasing, it follows that for all n > N ,
an ≥ aN > k.
Therefore, lim an = +∞.
n→∞
The proof for the case when {an } is monotone decreasing is similar. 
7. Subsequences
Example 7.1. The following are the subsequences of {an } = {1, −1, 1, −1, 1, −1, · · · }.
{a2n−1 } = {1, 1, 1, · · · }
{a2n } = {−1, −1, −1, · · · }.
In general, subsequences of {an } are of the form {ank }, k = 1, 2, 3, ..., with
n1 < n 2 < n 3 < · · · .
Note. The rule is that we should choose an1 first and then an2 with n2 > n1 and
then an3 with n3 > n2 , so far and so on (up to infinite). Thus n1 is at least 1, n2 is
at least 2, n3 is at least 3, · · · .
Theorem 7.2. Suppose lim an = A. Then every subsequence of {an } also con-
n→∞
verges to A, that is,
lim ank = A.
k→∞

Proof. For any given  > 0, since lim an = A, there exists N such that
n→∞

|an − A| <  for all n > N.

Then for all k > N , we have
nk ≥ k > N.
 8. THE LIMIT SUPERIOR AND INFERIOR OF A SEQUENCE 17

Hence
|ank − A| <  for all k > N.
Therefore, lim ank = A. 
k→∞

Corollary 7.3. Suppose that {an } has two subsequences that converge to differ-
ent limits. Then {an } is divergent. 
Example 7.4. The sequence {1, −1, 1, −1, · · · } is divergent because {a2n−1 } =
{1, 1, · · · } converges to 1 and {a2n } = {−1, −1, · · · } converges to −1.

8. The Limit Superior and Inferior of a Sequence

Given a sequence {an }, we can form another sequence {bn } given by
bn = sup ak = sup{an , an+1 , an+2 , · · · }.
k≥n

Example 8.1. Let {an } = {1, −1, 1, −1, · · · }. Then

bn = sup ak = sup{±1, ∓1, ±1, ∓1, · · · } = 1.
k≥n

Proposition 8.2. For any sequence {an }, the associated sequence {bn } = {sup ak }
k≥n
is always monotone decreasing.
Proof. For each n,
bn = sup{an , an+1 , an+2 , · · · } ≥ sup{an+1 , an+2 , · · · } = bn+1 .

Definition 8.3. The limit superior of {an }, denoted by lim sup an or lim sup an
n→∞
or lim an is defined to be lim bn , i.e.
n→∞ n→∞

lim an = lim bn = lim sup ak .

n→∞ n→∞ n→∞ k≥n

Example 8.4. 1. Let {an } = {1, −1, 1, −1, 1, −1, · · · }.

lim an = lim bn = lim 1 = 1.
n→∞ n→∞ n→∞

2. Let {an } = {1, 2, 3, · · · }. Then

bn = sup ak = sup{n, n + 1, · · · } = +∞
k≥n

and so lim an = lim bn = +∞.

n→∞ n→∞
3. Let {an } = {−1, −2, −3, · · · }. Then
bn = sup ak = sup{−n, −n − 1, · · · } = −n
k≥n

and so lim an = lim bn = −∞.

n→∞ n→∞

Theorem 8.5. Given any sequence {an }, either

(1). lim an exists (finite), or
n→∞
18 1. SEQUENCES OF REAL NUMBERS

(2). lim an = +∞, or

n→∞
(3). lim an = −∞.
n→∞

Proof. If {an } is not bounded above, then each bn is +∞, and thus
lim an = lim bn = +∞.
n→∞ n→∞

If {an } is bounded above, then each bn is finite. Since {bn } is monotone decreasing,
by Corollary 1.7.3, {an } converges (to a finite limit), or limn→∞ bn = −∞. 
Similarly, given any sequence {an }, we can form another sequence {cn } given by
cn = inf ak = inf{an , an+1 , an+2 , · · · }.
k≥n

Definition 8.6. The limit inferior of {an }, denoted by lim inf an or lim inf an
n→∞
or lim an is defined to be lim cn , i.e.
n→∞ n→∞

lim an = lim cn = lim inf ak .

n→∞ n→∞ n→∞ k≥n

Example 8.7. 1. Let {an } = {1, −1, 1, −1, 1, −1, · · · }.

lim an = lim cn = lim (inf{±1, ∓1, ±1, ∓1, · · · ) = lim −1 = −1.
n→∞ n→∞ n→∞ n→∞

2. Let {an } = {1, 2, 3, · · · }. Then

cn = inf ak = inf{n, n + 1, · · · } = n
k≥n

and so lim an = lim cn = +∞.

n→∞ n→∞
3. Let {an } = {−1, −2, −3, · · · }. Then
cn = inf ak = inf{−n, −n − 1, · · · } = −∞
k≥n

and so lim an = lim cn = −∞.

n→∞ n→∞

Proposition 8.8. (i). As in Proposition 8.2, for any given sequence {an }, the
associated sequence {cn } = {inf ak } is always monotone increasing.
k≥n
(ii). As in Theorem 8.5, for any given {an }, lim an either exists (finite), or +∞,
n→∞
or −∞).
Remark. We always have
lim an ≤ lim an
n→∞ n→∞

because cn ≤ bn .
Proposition 8.9. (i). If lim an = B with B 6= −∞, then given  > 0, there
n→∞
exists N such that an < B +  for all n > N .
(ii). lim an = C with C 6= +∞, then given  > 0, there exists N such that an >
n→∞
C −  for all n > N .
 8. THE LIMIT SUPERIOR AND INFERIOR OF A SEQUENCE 19

Proof. (i). If B = +∞, the assertion is obvious and so we assume that B is

finite. Since lim an = B, given any  > 0, there exists N such that for all n > N ,
n→∞

|bn − B| <  ⇒ bn < B +  ⇒ sup{an , an+1 , · · · } < B + ,

i.e. an , an + 1, · · · < B +  for all n > N . Proof of (ii) is similar. 
Warning!! Given a sequence {an }, lim an is a different concept from sup an . From
n→∞ n
the definition, we have
b1 = sup an = sup{a1 , a2 , · · · }
n

bn = sup{an , an+1 , an+2 , · · · }

with b1 ≥ b2 ≥ b3 ≥ · · · and lim an = lim bn . Thus we have the relation
n→∞ n→∞

lim an ≤ sup an = b1 ,
n→∞ n

but lim an need not be equal to sup an in general. Similarly,

n→∞ n

lim an ≥ inf an = c1 ,
n→∞ n

but need not be equal to in general.

The correct understanding is that lim is the largest subsequential limit of
convergent subsequences, including those possible subsequences tending to +∞ or
−∞. Similarly, lim is the smallest subsequential limit. This is described in the
following theorem.
Theorem 8.10. Let {an } be any sequence. Let B = lim an and let C = lim an .
n→∞ n→∞
(i) Let {ank } be any subsequence of {an } such that lim ank exists, +∞, or
k→∞
−∞. Then
C = lim an ≤ lim ank ≤ lim an = B.
n→∞ k→∞ n→∞

(ii) There exists a subsequence {ank } of {an } such that

lim ank = B.
k→∞

(iii) There exists a subsequence {amk } of {an } such that

lim amk = C.
k→∞

Proof. Let bn = sup{an , an+1 , · · · } and let cn = inf{an , an+1 , · · · }.

(i). Since nk ≥ k, we have
ck = inf{ak , ak+1 , · · · } ≤ ank ≤ bk = sup{ak , ak+1 , · · · }
and so
C = lim ck ≤ lim ank ≤ lim bk = B.
k→∞ k→∞ k→∞
(ii). We consider three cases B = +∞, −∞ or finite.
Case I. B = −∞.
20 1. SEQUENCES OF REAL NUMBERS

Since bn = sup{an , an+1 , an+2 , · · · } ≥ an and lim bn = B = −∞, we have

n→∞
lim an = −∞ = B. In this case, we can choose {an } itself as a subsequence with
n→∞
the desired property.
Case II. B = +∞. In this case we are going to construct a subsequence {ank } of
{an } such that lim ank = B = +∞.
k→∞
Since
b1 ≥ b2 ≥ · · · ≥ bn ≥ · · · ≥ B = lim bn = +∞,
n→∞
we have
b1 = b2 = · · · = +∞,
that is bn = +∞ for all n. Since b1 = sup{a1 , a2 , · · · } = +∞, there exists n1 such
that an1 > 1 because 1 is NOT an upper bound of {a1 , a2 , · · · }. Since
bn1 +1 = sup{an1 +1 , an1 +2 , an1 +3 · · · } = +∞,
there exists an2 such that n2 > n1 and an2 > 2 because 2 is NOT an upper bound
of {an1 +1 , an1 +2 , an1 +3 · · · }. Now, by induction, suppose that we have constructed
an1 , an2 , · · · , ank such that n1 < n2 < · · · < nk and
ans > s
for 1 ≤ s ≤ k. Since
bnk +1 = sup{ank +1 , ank +2 , · · · } = +∞,
there exists ank+1 such that nk+1 > nk and
ank+1 > k + 1
because k + 1 is NOT an upper bound of {ank +1 , ank +2 , ank +3 · · · }. The induction is
finished and so we obtain a subsequence {an1 , an2 , · · · } with the property that
ank > k
for any k. Since lim k = +∞, we have
k→∞

lim ank = +∞ = B.
k→∞

Case III. B is finite. We are going to construct a subsequence {ank } of {an } such
that lim ank = B.
k→∞
Since
b1 = sup{a1 , a2 , · · · },
b1 − 1 is not an upper bound of {a1 , a2 , · · · } and so there exists an1 such that
an1 > b1 − 1.
Since
bn1 +1 = sup{an1 +1 , an1 +2 , · · · },
1
bn1 +1 − is not an upper bound of {an1 +1 , an1 +2 , · · · } and so there exists an2 such
2
that n2 > n1 and
1
an2 > bn1 +1 − .
2
 8. THE LIMIT SUPERIOR AND INFERIOR OF A SEQUENCE 21

Now, by induction, suppose that we have constructed an1 , an2 , · · · , ank such that
n1 < n2 < · · · < nk and
1
ans > bns−1 +1 −
s
for 1 ≤ s ≤ k. Since
bnk +1 = sup{ank +1 , ank +2 , · · · },
1
bnk +1 − is not an upper bound of {ank +1 , ank +2 , · · · } and so there exists ank+1
k+1
such that nk+1 > nk and
1
ank+1 > bnk +1 − .
k+1
The induction is finished and so we obtain a subsequence {an1 , an2 , · · · } with the
property that
1
ank > bnk−1 +1 −
k
for any k. Consider the inequality
1
bnk−1 +1 − < ank ≤ bnk .
k
Since {bn } is convergent, we have
lim bnk = lim bnk−1 +1 = lim bn = B
k→∞ k→∞ n→∞

and
1
lim (bnk−1 +1 − ) = B − 0 = B.
k→∞ k
Thus, by the Squeeze theorem, we have
lim ank = B = lim an .
k→∞ n→∞

(iii). The proof is similar to that of (ii). We leave it to you as a tutorial question. 
Example 8.11. Find the limit inferior and limit superior of the following se-
quences
1 − 2(−1)n n
 
i) ,
n 3n + 2 nπ o
ii) (1 + (−1)n ) sin ,
4
iii) {[1.5 + (−1)n ]n } .
Solution. (i). Note that
1 − 2 · 2k 1/k − 4 4 2
a2k = lim a2k = lim =− =−
3 · 2k + 2 k→∞ k→∞ 6 + 2/k 6 3
1 + 2 · (2k − 1) 1/k + 2 · (2 − 1/k) 4 2
a2k−1 = lim a2k−1 = lim = =
3 · (2k − 1) + 2 k→∞ k→∞ 3 · (2 − 1/k) + 2/k 6 3
2
Thus the subsequential limits are ± and so
3
2 2
lim an = and lim an = − .
n→∞ 3 n→∞ 3
22 1. SEQUENCES OF REAL NUMBERS

(ii). The sequence


n
n nπ o 2π 4π 6π
(1 + (−1) ) sin = 0, 2 sin = 2, 0, 2 sin = 0, 0, 2 sin = −2,
4 4 4 4

8π 10π
0, 2 sin = 0, 0, 2 sin = 2, · · · .
4 4
The subsequential limits are −2, 0 and 2. Thus
lim an = 2 and lim an = −2.
n→∞ n→∞

(iii). Note that

2
a2k = 2.52k lim a2k = lim 2.5k = +∞
k→∞ k→∞
"  #(2k−1)/k "  #2−1/k
k k
1 1
a2k−1 = 0.52k−1 lim a2k−1 = lim = lim = 02 = 0.
k→∞ k→∞ 2 k→∞ 2
The subsequential limits are 0 and +∞. Thus
lim an = +∞ and lim an = 0.
n→∞ n→∞


The following theorem was originally proved by Bernhard Bolzano (1781-1848)
and modified slightly by Karl Weierstrass (1815-1897).
Corollary 8.12 (Bolzano-Weierstrass). Every bounded sequence has a conver-
gent subsequence.
Proof. Let {an } be a bounded sequence. Since {an } is bounded,
−∞ < inf {a1 , a2 , · · · } = c1 ≤ lim an ≤ lim an ≤ b1 = sup{a1 , a2 , · · · } < +∞.
n→∞ n→∞

Thus lim an is finite. By Part (ii) of Theorem 8.10, there exists a convergent subse-
n→∞
quence {ank } of {an } such that
lim ank = lim an .
k→∞ n→∞


Theorem 8.13. lim an = lim an (finite, +∞, −∞) if and only if lim an exists
n→∞ n→∞ n→∞
(finite), +∞, or −∞.
Proof. Suppose that lim an = lim an = A. Let bn = sup{an , an+1 , · · · } and let
n→∞ n→∞
cn = inf{an , an+1 , · · · }. Then
cn = inf{an , an+1 , · · · } ≤ an ≤ bn = sup{an , an+1 , · · · }.
By the assumption, we have
lim cn = lim an = lim an = lim bn .
n→∞ n→∞ n→∞ n→∞
 8. THE LIMIT SUPERIOR AND INFERIOR OF A SEQUENCE 23

By the Squeeze theorem, the sequence {an } converges and

lim an = lim an = lim an .
n→∞ n→∞ n→∞

Conversely suppose that {an } converges, tends to +∞, or tends to −∞. Let
A = lim an and let {ank } be any subsequence of {an }. By Theorem 7.2, we have
n→∞
lim ank = A. Thus the only subsequential limit of {an } is A. By Theorem 8.10, we
k→∞
have
lim an = lim an = A = lim an .
n→∞ n→∞ n→∞

Remark. This theorem means that
1) If lim an = lim an , then
n→∞ n→∞

lim an = lim an = lim an .

n→∞ n→∞ n→∞

2) If lim an exists, +∞ or −∞, then

n→∞

lim an = lim an = lim an .

n→∞ n→∞ n→∞

Example 8.14. Let {an } be a sequence. Show that lim |an | = 0 if and only if
lim |an | = 0, if and only if lim an = 0
n→∞ n→∞

Proof. Suppose that lim |an | = 0. From |an | ≥ 0, we have

0 = lim 0 = lim 0 ≤ lim |an | ≤ lim |an | = 0.
n→∞
Thus
lim |an | = lim |an | = 0
and so lim |an | exists and
n→∞

lim |an | = lim |an | = lim |an | = 0

n→∞

Conversely suppose that lim |an | = 0. Then

n→∞

lim |an | = lim |an | = lim |an | = 0

n→∞

because {an } is convergent. Hence lim |an | = 0 if and only if lim |an | = 0.
n→∞
Now suppose that lim |an | = 0. From
n→∞

−|an | ≤ an ≤ |an |,
we have lim an = 0 by the Squeeze theorem.
n→∞
Conversely suppose that lim an = 0. Then
n→∞

lim |an | = | lim an | = |0| = 0

n→∞ n→∞

because the function f (x) = |x| is continuous. Hence

lim |an | = 0 ⇐⇒ lim an = 0.
n→∞ n→∞
24 1. SEQUENCES OF REAL NUMBERS


√ an+1
Example 8.15. Let an > 0 for all n. Prove that lim n
an ≤ lim .
an
an+1 √
Proof. Let B = lim . If B = +∞, clearly lim n an ≤ B = +∞. So we may
an
an+1
assume that B < +∞. Since an > 0, we have > 0 and so B ≥ 0. Thus B is a
an
finite nonnegative number. By Proposition 8.9, given any  > 0, there exists N such
that
an+1
<B+
an
for n > N . Fixed any n with n > N , we have
an+1 an+2 an+3
0< , , ,··· < B + 
an an+1 an+2
and so, for any k ≥ 1, we have
an+1 an+2 an+3 an+k an+k
0< · · ··· = ≤ (B + )k
an an+1 an+2 an+k−1 an
1 1 k
=⇒ an+k ≤ an (B + )k =⇒ n+k
an+k ≤ ann+k · (B + ) n+k for any k ≥ 1.
Let k tends to ∞. We have
1 k 1 1
lim ann+k · (B + ) n+k = lim ann+k · lim (B + ) n/k+1 = a0n · (B + ) = B + .
k→∞ k→∞ k→∞

Thus 1 1
1 k
n+k
lim am
m
= lim an+k ≤ lim ann+k · (B + ) n+k
m→∞ k→∞ k→∞
1 k
n+k
= lim an · (B + ) n+k = B + .
k→∞
In other words,
√
lim n
an ≤ B + 
n→∞
for any  > 0 and so
√  √ 
lim n
an = lim lim n
an ≤ lim(B + ) = B,
n→∞ →0 n→∞ →0

that is,
√ an+1
lim n
an ≤ lim .
n→∞ n→∞ an

an+1 √
Exercise 8.1. Let an > 0 for all n. Prove that lim ≤ lim n an .
an
By Example 8.15 and Exercise 8.1, we have
an+1 √ √ an+1
lim ≤ lim n an ≤ lim n an ≤ lim .
an an
an+1
Exercise 8.2. Let an > 0 for all n. Suppose that the limit lim exists or
n→∞ an
√ √ an+1
+∞. Prove that lim n an exists and lim n an = lim .
n→∞ n→∞ n→∞ an
 9. CAUCHY SEQUENCES AND THE COMPLETENESS OF R 25

For instance,
√
n (n + 1)!
lim n! = lim = lim (n + 1) = +∞.
n→∞ n→∞ n! n→∞

9. Cauchy Sequences and the Completeness of R

9.1. Cauchy Sequences.
Definition 9.1. {an } is called a Cauchy sequence if given any  > 0, there exists
a natural number N such that for all m, n > N , we have
|an − am | < .
Remark. Roughly speaking, a sequence is Cauchy if the width of its tail → 0 as
n → ∞.
Proposition 9.2. Every Cauchy sequence is bounded.
Proof. Let {an } be a Cauchy sequence. Choose  = 1. There exists N such that
|an − am | < 1 for n, m > N . In particular, |an − aN +1 | < 1 or
aN +1 − 1 < an < aN +1 + 1
for n > N . Let
M = max{a1 , a2 , · · · , aN , aN +1 + 1}
m = min{a1 , a2 , · · · , aN , aN +1 − 1}.
For n ≤ N , we have m ≤ an ≤ M , and, for n > N , we have
m ≤ aN +1 − 1 < an < aN +1 + 1 ≤ M.
Thus, for all n, we have m ≤ an ≤ M and so {an } is bounded. 
9.2. Completeness of R. The following Criterion was formulated by Augustin-
Louis Cauchy (1789-1857).
Theorem 9.3 (Cauchy’s criterion). A sequence is a convergent sequence if and
only if it is a Cauchy sequence.
Proof. =⇒, i.e., every convergent sequence is Cauchy.
Given that {an } is convergent, say lim an = A. Then for any given  > 0, there
n→∞
exists N such that

|an − A| <
2
for all n > N. Now for any m, n > N ,
 
|an − am | = |(an − A) − (am − A)| ≤ |an − A| + |am − A| < + = 
2 2
since both m, n > N . Therefore, {an } is a Cauchy sequence.
⇐=, i.e., every Cauchy sequence is convergent.
Given that {an } is Cauchy. By Proposition 9.2, {an } is bounded. By the Bolzano-
Weierstrass Theorem (Corollary 8.12), there exists a convergent subsequence {ank }
of {an }. Let A = limk→∞ ank . Given any  > 0, since {an } is Cauchy, there exists N1
such that

|an − am | < for all n, m > N1 .
2
26 1. SEQUENCES OF REAL NUMBERS

Since {ank } converges to A, there exists K such that


|ank − A| < for all k > K.
2
Let N = max{K, N1 }. Choose an nk such that k > N , for instance, choose nk to be
nN +1 . When n > N , by triangular inequality,
 
|an − A| = |(an − ank + (ank − A)| ≤ |an − ank | + |ank − A| < + = 
2 2
because n > N ≥ N1 , nk ≥ k > N ≥ N1 and k > N ≥ K. Therefore {an } converges
to A by the definition. 
Remark. The statement that every Cauchy sequence in R converges is often ex-
pressed by saying that R is complete. Note that our proof of the Cauchy Criterion
used the Bolzano-Weierstrass Theorem and the proof of that one used the complete-
ness axiom of R (Theorem 5.14). We did not prove Theorem 5.14, namely we treated
Theorem 5.14 as an axiom. (Dedekind and Cantor proved the completeness axiom
in 1872 independently.) Conversely, we can treat the Cauchy Criterion as an axiom,
and prove Theorem 5.14. In this sense, both Theorem 5.14 and the Cauchy Criterion
are regarded as the completeness of R.
Cantor constructed real numbers from rational numbers by using Cauchy se-
quences. His ideas are as follows. Consider all of the Cauchy sequences {an } with
an ∈ Q. By the Cauchy criterion, {an } converges to a real number. Then he proved
that all real numbers can be obtained as the limits of all of the Cauchy sequences
{an } with an ∈ Q. One can think that, by using our standard base√10 number sys-
tem, any real number admits a (decimal) expansion. For instance, 2 = 1.4142 · · · ,
we can define a sequence of rational numbers
√
a1 = 1, a2 = 1.4, a3 = 1.41, a4 = 1.414, a5 = 1.4142, · · · with lim an = 2.
n→∞

1 1 1
Example 9.4. Let sn = 1 + 2
+ 2 + · · · + 2 . Show that {sn } is convergent.
2 3 n
Proof. For each k ≥ 1, we have
   
 1 1 1 1 1 1 
|sn+k −sn | =  1 + 2 + · · · + 2 + + · · · + − 1 + + · · · + 
2 n (n + 1)2 (n + k)2 22 n2 
1 1 1
= 2
+ 2
+ ··· +
(n + 1) (n + 2) (n + k)2
1 1 1 1
≤ + + + ··· +
n(n + 1) (n + 1)(n + 2) (n + 2)(n + 3) (n + k − 1)(n + k)
     
1 1 1 1 1 1
= − + − + ··· + −
n n+1 n+1 n+2 n+k−1 n+k
1 1 1
= − < .
n n+k n
1 1
Given any  > 0, choose N such that N < , that is, N > . When m > n > N ,

from the above,
1 1
|sm − sn | < < < .
n N
 9. CAUCHY SEQUENCES AND THE COMPLETENESS OF R 27

Thus {sn } is a Cauchy sequence and hence {sn } converges by the Cauchy Criterion.

9.3. Contractive Sequences. In this subsection, we give an application of the
Cauchy Criterion.
A sequence {an } is called contractive if there exists b, 0 < b < 1, such that
(1) |an+1 − an | ≤ b|an − an−1 |
for all n ≥ 2.
Theorem 9.5 (Contractive Theorem). Every contractive sequence converges. Fur-
thermore, if {an } is contractive and A = lim an , then
n→∞
bn−1
(a). |A − an | ≤ |a2 − a1 |, and
1−b
b
(b). |A − an | ≤ |an − an−1 |,
1−b
where b is the constant in above definition.
Proof. From Inequality (1), we have
|an+1 − an | ≤ b|an − an−1 | ≤ b2 |an−1 − an−2 | ≤ b3 |an−2 − an−3 | ≤ · · · ≤ bn−1 |a2 − a1 |
For each k ≥ 1, we have
|an+k − an | = |(an+k − an+k−1 ) + (an+k−1 − an+k−2 ) + · · · + (an+1 − an )|
≤ |an+k − an+k−1 | + |an+k−1 − an+k−2 | + · · · + |an+1 − an |
≤ bn+k−1 |a2 − a1 | + bn+k−2 |a2 − a1 | + · · · + bn−1 |a2 − a1 |
= (bn+k−1 + bn+k−2 + · · · + bn−1 )|a2 − a1 |
bn−1 · (1 − bk+1 ) bn−1
= bn−1 (1 + b + b2 + · · · + bk )|a2 − a1 | = |a2 − a1 | < |a2 − a1 |.
1−b 1−b
Since 0 < b < 1, we have limn→∞ bn−1 = 0 by the Standard limits and so
bn−1
lim |a2 − a1 | = 0.
n→∞ 1 − b

Therefore, given any  > 0, there exists N such that

bn−1
|a2 − a1 | < 
1−b
for n > N and so, for all n > N and k ≥ 1,
bn−1
|an+k − an | < |a2 − a1 | < .
1−b
It follows that {an } is Cauchy and so it is convergent.
Let A = lim an . From the inequality
n→∞

bn−1
|an+k − an | < |a2 − a1 |,
1−b
we have
  bn−1
|A − an | =  lim an+k − an  = lim |an+k − an | ≤ |a2 − a1 |.
 
k→∞ k→∞ 1−b
28 1. SEQUENCES OF REAL NUMBERS

This is Inequality (a).

For each k ≥ 1, we have
|an+k − an | = |(an+k − an+k−1 ) + (an+k−1 − an+k−2 ) + · · · + (an+1 − an )|
≤ |an+k − an+k−1 | + |an+k−1 − an+k−2 | + · · · + |an+1 − an |
= |an+1 − an | + |an+2 − an+1 | + · · · + |an+k − an+k−1 |
≤ b|an − an−1 | + b2 |an − an−1 | + · · · + bk |an − an−1 |
= (b + b2 + · · · + bk )|an − an−1 |
b · (1 − bk ) b
= b(1 + b + b2 + · · · + bk−1 )|an − an−1 | = |an − an−1 | < |an − an−1 |.
1−b 1−b
Thus
  b
|A − an | =  lim an+k − an  = lim |an+k − an | ≤ |an − an−1 |.
 
k→∞ k→∞ 1−b
This is Inequality (b). 
1
Example 9.6. Let c1 ∈ (0, 1) be arbitrary, and for n ≥ 1 set cn+1 = (c2n + 1).
3
1) Prove that the sequence {cn } is contractive.
2) Show that if c = limn→∞ cn , then c is a root of the polynomial x2 − 3x + 1.
1
3) Let c1 = . Determine a value of n such that |c − cn | < 10−3 .
2
Proof. (1). First we show that 0 < cn < 1 for all n by induction. By the
assumption 0 < c1 < 1. Suppose that 0 < cn < 1. Then
1 1 2
0 < cn+1 = (c2n + 1) < (1 + 1) = < 1.
3 3 3
The induction is finished and so 0 < cn < 1 for all n. Now from
 
1 2 1 2  1
|cn+1 − cn | =  (cn + 1) − (cn−1 + 1) = |c2n − c2n−1 |

3 3 3
1 1 2
= |cn − cn−1 ||cn + cn−1 | ≤ |cn − cn−1 |(1 + 1) = |cn − cn−1 |,
3 3 3
the sequence {cn } is contractive with b = 23 .
(2). From
1
cn+1 = (c2n + 1),
3
1 2 1  1
c = lim cn+1 = lim (cn + 1) = lim c2n + 1 = (c2 + 1).
n→∞ n→∞ 3 3 n→∞ 3
2 2
Thus c − 3c + 1 = 0, that is, c is a root of x − 3x + 1.
(3). We use Inequality (a) of the Contractive Theorem.
Note that  
1 2 1 1 5
c2 = (c1 + 1) = + 1 = .
3 3 22 12
2 n−1 


bn−1

5 1
|c2 − c1 | = 3 2  −  < 10−3

1−b 1 − 3 12 2
 9. CAUCHY SEQUENCES AND THE COMPLETENESS OF R 29
 n−1  n−1
2 1 1 2 4
=⇒ 3 · < 3 =⇒ < 3
3 12 10 3 10
 
2
=⇒ (n − 1) · ln < ln 4 − 3 ln 10
3
3 ln 10 − ln 4
=⇒ (n − 1) > = 13.62.
ln 3 − ln 2
Thus when n = 15, we have
bn−1
|c − c15 | < |c2 − c1 | < 10−3 .
1−b

Note. From x2 − 3x + 1 = 0, we have
√ √
3 ± 32 − 4 3± 5
x= = .
2 2
√
3− 5
Since 0 ≤ c ≤ 1, c = .
2
 CHAPTER 2

Series of Real Numbers

1. Series
The expression
a1 + a2 + a3 + · · ·
∞
X
written alternatively as ak is called an infinite series.
k=1

Example 1.1. (1). 1 + 2 + 3 + 4 + · · · .

(2). 1 + 1/2 + 1/3 + 1/4 + · · ·
(3). 1 + 1/22 + 1/32 + 1/42 + · · · .
(4). 1 + 0 + 1 + 0 + 1 + 0 + · · · .
∞
X
Definition 1.2. Given a series ak , its nth partial sum Sn is given by
k=1
n
X
Sn = ak = a1 + a2 + · · · + an .
k=1
∞
X
The sequence {Sn } is called the sequence of partial sums of the series ak .
k=1

Example 1.3. Consider the series 1−1+1−1+1−1+1−1+· · · . The S2n−1 = 1

and S2n = 0.
∞
X
Definition 1.4. Consider the sequence of partial sums {Sn } of the series ak .
k=1
∞
X
If this sequence converges to a number S, we say that the series ak converges
k=1
to S and write
∞
X
ak = lim Sn = S.
n→∞
k=1
∞
X
If {Sn } diverges, then we say an diverges.
k=1

Example 1.5 (Geometric Series). Let a 6= 0. Consider the series

∞
X
arn = a + ar + ar2 + ar3 + · · · .
n=0
31
32 2. SERIES OF REAL NUMBERS

Then the partial sum

1 − rn

 a
 r 6= 1
Sn = a + ar + ar2 + · · · + arn−1 n−1
= a(1 + r + · · · + r ) = 1 − r


a(n + 1) r = 1
a
When −1 < r < 1, Sn → as n → ∞.
1−r
When r > 1, Sn diverges because rn → +∞ as n → ∞.
When r = 1, Sn = a(n + 1) diverges.
a[1 − (−1)n ]
When r = −1, Sn = diverges.
2
When r < −1, Sn diverges because rn → ±∞.
X∞
Thus the geometric series arn converges if and only if −1 < r < 1,
n=0
and,
∞
X a
arn =
n=0
1−r
for −1 < r < 1.
X∞ ∞
X
Remark. If ak and bk converges, then one always has
k=1 k=1
∞
X ∞
X ∞
X
(i) (ak + bk ) = ak + bk .
k=1 k=1 k=1
∞
X ∞
X
(ii) cak = c ak .
k=1 k=1

Example 1.6.
∞  n  n  X ∞  n ∞  n
X 1 1 1 X 1
+ = +
n=1
4 5 n=1
4 n=1
5
"  2 # "  2 #
1 1 1 1
= + + ··· + + + ···
4 4 5 5
"  2 # "  2 #
1 1 1 1 1 1
= 1+ + + ··· + 1+ + + ···
4 4 4 5 5 5
1 1
1 1
= +
1
4 5 1
1− 1−
4 5
1 1 1 1 7
= + = + = .
3 4 3 4 12
4· 5·
4 5
∞
X
Theorem 1.7. If ak converges, then lim ak = 0.
k→∞
k=1
 1. SERIES 33

Proof. Recall that partial sum Sk = a1 + a2 + · · · + ak . We have

Sk − Sk−1 = (a1 + a2 + · · · + ak−1 + ak ) − (a1 + a2 + · · · + ak−1 ) = ak .
∞
X
Since the series ak converges, the sequence {Sk } converges. Let S = lim Sk .
k→∞
k=1
Then
lim ak = lim (Sk − Sk−1 ) = lim Sk − lim Sk−1 = S − S = 0.
k→∞ k→∞ k→∞ k→∞

∞
X
Corollary 1.8 (Divergence Test). If lim an 6= 0 (or does not exist), then an
n→∞
n=1
diverges. 
∞
X
Example 1.9. (1). The series (−1)n is divergent because the limit of the n-th
n=1
term (−1)n does not exist.
∞
X n!
(2). The series is divergent because
n=1
n2
n! 1 1
lim = 2 = 6 0.
= +∞ =
n→∞ n2 n 0
lim
n→∞ n!
∞
X 2n + 1
(3). The series is divergent because
n=1
3n + 2
2n + 1 2 + 1/n 2
lim = lim = 6= 0.
n→∞ 3n + 2 n→∞ 3 + 2/n 3
Remark. The divergence test is a “one-way” test, i.e., lim an = 0 does NOT imply
n→∞
∞
X
an converges.
n=1
∞
X
Theorem 1.10 (Cauchy Criterion). The series ak converges if and only if
k=1
given any  > 0, there exists N such that
 
 Xm 
ak  < 
 

 
k=n+1

for all m > n > N .

∞
X
Proof. The series ak converges if and only if the sequence of its partial sums
k=1
{Sn } converges, (by definition), if and only if {Sn } is Cauchy. The result follows from
|Sm − Sn | = |(a1 + a2 + · · · + an + an+1 + · · · + am ) − (a1 + a2 + · · · + an )|
34 2. SERIES OF REAL NUMBERS
 
m
 X 
= |an + an+1 + · · · + am | =  ak  .
 
 
k=n+1

∞
X 1
Example 1.11 (Harmonic Series). Show that the series diverges.
n=1
n
Proof. Note that
   
 1 1 1 1 1 1 
|S2n − Sn | =  1 + + · · · + + + ··· + − 1 + + ··· + 
2 n n+1 2n 2 n 
1 1 1 1 1 1 1 1
= + + ··· + ≥ + + ··· + =n· = .
n+1 n+2 2n 2n 2n 2n 2n 2
∞
X 1
Suppose that the series converges. Then {Sn } is Cauchy. Given  = 21 . There
n=1
n
1
exists N such that |Sm − Sn | < for all m > n > N . This contradicts to the above
2
1
fact that |S2n − Sn | ≥ , where m is chosen to be 2n. 
2
Note. The divergence of the harmonic series appears to(have been ) established by
n
X 1
Nicole Oresme (1323?-1382) by showing that the sequence is NOT bounded.
k=1
k
∞
X
Theorem 1.12. Suppose that eventually ak ≥ 0. Then ak converges if and
k=1
only if {Sn } is bounded above.
Proof. We may assume that ak ≥ 0 for all k. Since
Sn+1 − Sn = an+1 ≥ 0,
X ∞
the sequence {Sn } is monotone increasing. Thus ak converges if and only if
k=1
{Sn } converges, if and only if {Sn } is bounded above (by the Monotone Convergence
Theorem). 
Note. Suppose that eventually ak ≥ 0. This theorem means that the following.
∞
X
1) If {Sn } is bounded above, then ak converges.
k=1
∞
X
2) If {Sn } is NOT bounded above, then ak diverges.
k=1

2. Tests for Positive Series

∞
X
A series ak is called a (eventually) positive series if every term ak is (even-
k=1
tually) positive.
 2. TESTS FOR POSITIVE SERIES 35

2.1. Comparison Test.

∞
X ∞
X
Theorem 2.1 (Comparison Test). Consider 2 positive series ak and bk .
k=1 k=1
Suppose that eventually 0 ≤ ak ≤ bk .
∞
X ∞
X
(i) If bk converges, then ak converges.
k=1 k=1
∞
X ∞
X
(ii) If ak diverges, then bk diverges.
k=1 k=1
n
X
Proof. Assertion (ii) follows immediately from (i). Let An = ak , and Bn =
k=1
n
X ∞
X ∞
X
bk . Then An ≤ Bn for all n. Suppose that bk converges, that is, bk is a
k=1 k=1 k=1
(finite) number. Then
∞
X
An ≤ Bn ≤ bk
k=1
∞
X
for all n and so An is bounded above. By Theorem 1.12, ak is convergent. 
k=1
∞  k
X 2k − 1
Example 2.2. The series converges because
k=1
3k + 2
 k  k
2k − 1 2
≤
3k + 2 3
∞  k
X 2
and the geometric series converges.
k=1
3
∞
X
Remark. 1. Suppose 0 ≤ an ≤ bn and bn diverges. Then NO conclusion can be
n=1
drawn. ∞
X
2. Similarly, suppose 0 ≤ an ≤ bn and an converges. Then NO conclusion can be
n=1
drawn.
Example 2.3.  n
1
0≤ ≤ 2n ≤ 3n .
2
X∞ ∞
X  1 n X∞
n
The series 3 diverges. Now converges and 2n diverges.
n=1 n=1
2 n=1
36 2. SERIES OF REAL NUMBERS

∞
X ∞
X
Corollary 2.4 (Limit Comparison Test). Suppose that ak and bk are
k=1 k=1
(eventually) positive series.
∞ ∞
ak X X
(a). If lim = L with 0 < L < ∞, then ak converges if and only if bk
k→∞ bk
k=1 k=1
∞
X ∞
X
converges. (So ak diverges if and only if bk diverges.)
k=1 k=1
∞ ∞
ak X X
(b). If lim = 0, (that is, ak << bk ), and bk converges, then ak
k→∞ bk
k=1 k=1
converges.
∞ ∞
ak X X
(c). If lim = 0, (that is, ak << bk ), and ak diverges, then bk di-
k→∞ bk
k=1 k=1
verges.
an
Remark. If lim = +∞, interchange an and bn , and then apply assertions (b)
n→∞ bn
and (c).

Proof. We may assume that ∞

P P∞
k=1 ak and k=1 bk are positive series.
(a). Since
ak
lim = L (6= 0, 6= ∞),
k→∞ bk
 
ak
is bounded above, say, by M . Thus
bk
0 ≤ ak ≤ M b k
 
bk 1 bk
for all k. Similarly, since lim = (6= 0, 6= ∞), is bounded above, say, by
k→∞ ak L ak
M 0 . Thus 0 ≤ bk ≤ M 0 ak for all k.
∞
X X∞ X∞
If ak is convergent, then M 0 ak = M 0 ak is also convergent. By the
k=1 k=1 k=1
∞
X
comparison test, it follows that bk is also convergent because bk ≤ M 0 ak .
k=1
∞
X ∞
X ∞
X
If bk is convergent, then M bk = M bk is also convergent. By the
k=1 k=1 k=1
∞
X
comparison test, it follows that ak is also convergent because ak ≤ M bk . Hence
k=1
∞
X ∞
X ∞
X
ak is convergent if and only if bk is convergent. Hence ak is divergent if
k=1 k=1 k=1
∞
X
and only if bk is divergent.
k=1
 2. TESTS FOR POSITIVE SERIES 37

ak
Now we prove assertions (b) and (c). lim = 0 implies for every  > 0, there
k→∞ bk
is an N such that  
 ak 
 − 0 <  ∀k > N.
 bk 
We choose  = 1. Then the above inequality is
ak < b k ∀k > N.
We get the result by applying the comparison test. 
Standard series used in comparison and limit comparison tests.
1. The Geometric Series:

X∞  converges if |r| < 1,
n−1
ar =
diverges if |r| ≥ 1.

n=1

2. The p-series: for a fixed p,


∞
X 1  converges if p > 1,
=
n=1
np  diverges if p ≤ 1.
To be proved in the subsection on Integral Test.
Example 2.5. Determine the convergence or divergence:
∞
X 1 + cos n
1)
n=1
n2
∞
X ln n + n3 + 8
2)
n=1
n4 − 2n + 3

Solution. (1). It is convergent, by the comparison test, because

1 + cos n 2
0≤ 2
≤ 2
n n
∞
X 1
and the p-series 2
is convergent.
n=1
n
(2). It is divergent, by the limit comparison test, because
ln n + n3 + 8 ln n 1
n ln n + n 4
+ 8n +1+ 3
4
lim n − 2n + 3 = lim = lim n
3 n = 0+1+0 =1
n→∞ 1 n→∞ 4
n − 2n + 3 n→∞ 2 3 1−0+0
1− 3 + 4
n n n
∞
X 1
and the harmonic series diverges. 
n=1
n
∞
X 1
Example 2.6. Determine convergence or divergence of , where k is a
n=2
(ln n)k
constant.
38 2. SERIES OF REAL NUMBERS

1 1
Solution. Let bn = and let a n = . Then
(ln n)k n

1
an n (ln n)k
lim = lim = lim = 0.
n→∞ bn n→∞ 1 n→∞ n
(ln n)k

∞ ∞
X 1 X 1
Since the harmonic series is divergent, the series is divergent for
n=2
n n=2
(ln n)k
any k. 

2.2. Integral Test. Let f (x) be a real-valued function on [a, +∞) such that
Z b
f (x) is Riemann integrable, that is, the integral f (x) dx exists for every b > a.
a
The following theorem is useful.

Theorem 2.7. Let f (x) be a real-valued function on [a, b].

(a). If f (x) is continuous on [a, b], then f (x) is Riemann integrable on [a, b].
(b). If f (x) is monotone on [a, b], then f (x) is Riemann integrable on [a, b].

The improper integral is defined by

Z ∞ Z b
f (x) dx = lim f (x) dx = area under f (x) over [a, ∞).
a b→∞ a

Z ∞ Z b
Here we say that f (x) dx converges if the limit lim f (x) dx exists (finite),
a b→∞ a
i.e., the area under f (x) over [a, ∞) is finite.
Z ∞ Z b
We also say that f (x) dx diverges if the limit lim f (x) dx does not exist.
a b→∞ a

Theorem 2.8 (Integral Test). Let f (x) be an (eventually) positive monotone

X∞
decreasing function on [1, +∞). Suppose we have a series ak such that ak =
k=1
∞
X Z ∞
f (k), then the series ak and the integral f (x)dx either both converge or both
k=1 1
diverge.

Proof. We may assume that f (x) is positive monotone decreasing on [1, +∞).
Let an = f (n) for all n.
 2. TESTS FOR POSITIVE SERIES 39

From the graph, we see that

area of the rectangles ≤ area under f (x) over [1, n], i.e.,
n
X Z n Z ∞
f (k) ≤ f (x)dx ≤ f (x)dx.
k=2 1 1
Z ∞
Thus, if f (x) dx < ∞, then
1
n
X n
X Z ∞
ak = f (k) ≤ f (x) dx < ∞,
k=2 k=2 1

n
X Z ∞
i.e., for all n, ak is bounded above by the finite number f (x) dx. Since we
k=2 1
∞
X ∞
X
also have ak ≥ 0, it follows from Theorem 1.12 that ak converges, and thus ak
k=2 k=1
also converges.
Next we consider the following graph:

From the graph, it is easy to see that

n−1
X Z n
area under the rectangles ≥ area under f (x) over [1, n], i.e., f (k) ≥ f (x) dx.
k=1 1

∞
X
Thus, if ak < ∞, then
k=1

∞
X n−1
X n−1
X Z n
∞> ak ≥ ak = f (k) ≥ f (x) dx,
k=1 k=1 k=1 1

Z n ∞
X
i.e., f (x) dx is bounded above by the finite number ak < ∞. Letting n → ∞,
1 k=1
it follows that we have
Z ∞ ∞
X
f (x) dx ≤ ak < ∞.
1 k=1
40 2. SERIES OF REAL NUMBERS

∞
X Z ∞
In conclusion, we have ak converges if and only if f (x) dx converges, which
k=1 1
∞
X Z ∞
also means that ak diverges if and only if f (x) dx diverges. 
k=1 1

Example 2.9. Show that

∞
X 1
1) the series converges if and only if p > 1.
n=1
np
∞
X 1
2) the series converges if and only if k > 1.
n=2
n(ln n)k
∞
1 X 1
(1). If p ≤ 0, then p does not tend to 0 and so, by divergence test, p
diverges.
n n=1
n
1
Assume that p > 0. Let f (x) = p on [1, +∞). Then f (x) is positive monotone
x
decreasing. Now
 +∞
1 −p+1 

x p 6= 1
Z ∞ Z ∞ 


f (x)dx = x−p dx = −p + 1 
1
1 1 

ln(+∞) − ln 1 p=1


Z ∞ ∞
X 1
Thus f (x) dx converges if and only if p > 1 and so p
converges if and only
1 n=1
n
if p > 1.
1
(2). Let f (x) = on [2, +∞). Then f (x) is positive. We check that f (x) is
x(ln x)k
eventually monotone decreasing. From

0 1
f 0 (x) = x−1 (ln x)−k = −x−2 (ln x)−k −kx−1 (ln x)−k−1 = −x−2 (ln x)−k−1 (ln x+k),
x
we have f 0 (x) ≤ 0 when ln x > −k. Thus f (x) is monotone decreasing when ln x > −k
and so f (x) is eventually monotone decreasing. Now
 ∞
1 −k

y k 6= 1
Z ∞ Z ∞ Z ∞ 
 
1 y = ln x 1 
−k + 1 
f (x) dx = dx ======== dy = ln 2
2 2 x(ln x)k 1 ln 2 y
k 
dy = dx

ln(+∞) − ln(ln 2) k = 1

x
Z ∞ ∞
X 1
Thus f (x) dx converges if and only if k > 1 and so the series con-
2 n=2
n(ln n)k
verges if and only if k > 1.

2.3. Ratio Test.

 2. TESTS FOR POSITIVE SERIES 41

∞
X
Theorem 2.10 (Ratio Test). Consider the positive series an . Suppose
n=1

an+1
(2) lim = `.
n→∞ an

∞
X
(i) If 0 ≤ ` < 1, then an converges.
n=1
X∞
(ii) If 1 < ` ≤ ∞, then an diverges.
n=1
(iii) If ` = 1, then the test is inconclusive.

Proof. We will prove (i) and (ii). Given any  > 0, it follows from (1) that there
exists N such that for all n > N ,
 
 an+1
 <  or ` −  < an+1 < ` + .


 an − ` an

By repeating using the above inequalities, it follows that for all m > 0,

(3) aN +1 (` − )m < aN +1+m < aN +1 (` + )m .

∞
X
(i). If ` < 1, choose  > 0 such that ` +  < 1, then aN +1 (` + )m converges
m=1
(since it is a geometric series with common ratio satisfying |r| = ` +  < 1). Together
X∞
with the right-hand-side of (3), it follows from the comparison test that aN +1+m
m=1
∞
X
converges, and thus an converges.
n=1
(ii). If ` > 1, choose  > 0 such that ` −  > 1, then by the left-hand-side of (3), we
have, for all m > 0,

aN +1+m ≥ aN +1 (` − )m > aN +1 > 0.

∞
X
In particular, lim an 6= 0 or does not exist. By the divergence test, an diverges.
n→∞
n=1


Example 2.11. Determine convergence or divergence.

∞
X n!
1) n
.
n=1
n
∞
X (n!)2
2) .
n=1
(2n)!
42 2. SERIES OF REAL NUMBERS

(1).
(n + 1)!
an+1 (n + 1)n+1 (n + 1)! 1 1
lim = lim = lim n = lim  n = < 1.
n→∞ an n→∞ n! n→∞ (n + 1) n→∞ 1 e
n
n! · n
· (n + 1) 1+
n n n
Thus the series converges.
(2).
[(n + 1)!]2
an+1 (2n + 2)! (n + 1)2
lim = lim = lim
n→∞ an n→∞ (n!)2 n→∞ (2n + 2)(2n + 1)

(2n)!
(n + 1)2 /n2 1 + 2/n + 1/n2 1+0+0 1
= lim 2
= lim = = < 1.
n→∞ (2n + 2)(2n + 1)/n n→∞ (2 + 2/n)(2 + 1/n) (2 + 0) · (2 + 0) 4
Thus the series converges.
2.4. Root Test.
∞
X
Theorem 2.12. Consider the series an with each
n=1
an ≥ 0, and let
√
(4) ` = lim n
an .
n→∞
∞
X
(i) If 0 ≤ ` < 1, then an converges.
n=1
X∞
(ii) If 1 < ` ≤ ∞, then an diverges.
n=1
(iii) If ` = 1, then the test is inconclusive.
Proof. We will prove (i) and (ii).
(i) Suppose that ` < 1. Then for all given  > 0, it follows from (4) and Proposi-
√
tion 8.9 of chapter 1, that there exists an N such that n an < ` +  for all n > N .
Now choose  > 0 s.t. ` +  < 1. Then
(5) 0 ≤ an < (` + )n for all n > N.
∞
X
Since (` + )n converges (as it is a geometric series with common ratio satisfying
n=1
∞
X
|r| = ` +  < 1), it follows from (5) and the comparison test that an converges.
n=1
(ii). We are going to prove (ii) by contradiction. Given that ` > 1. Suppose that
X∞
an converges. Then by Theorem 1.7, we have lim an = 0. In particular, there
n→∞
n=1
 3. THE DIRICHLET TEST AND ALTERNATING SERIES 43
√
exists N such that 0 ≤ an < 1 for all n > N . Hence an < 1 for all n > N , and it
n

X∞
follows that we must have ` ≤ 1, which is a contradiction. Hence an diverges. 
n=1
∞
X
Corollary 2.13 (Simplified Root Test). Consider the series an with each
√ n=1
an ≥ 0. Suppose that lim n an = `.
n→∞
∞
X
(i) If 0 ≤ ` < 1, then an converges.
n=1
X∞
(ii) If 1 < ` ≤ ∞, then an diverges.
n=1
(iii) If ` = 1, then the test is inconclusive.
Proof. We will prove (i) and (ii). Recall from Theorem 8.13 of chapter 1 that
√ √ √
if lim n an exists, then lim n an = lim n an . Then the Corollary follows from
n→∞ n→∞ n→∞
Theorem 2.12. 
Example 2.14. Determine convergence or divergence of the series
∞  n2
X
n 1
2 1− .
n=1
n

1
" n2 # n
n
 
√ n 1 1 2
lim n
an = lim 2 1− = lim 2 1 − = < 1.
n→∞ n→∞ n n→∞ n e
Thus the series converges.
Example 2.15. Determine convergence or divergence of the series
∞  n2
X
n 2
(3 + sin n) 1 − .
n=1
n

1
" n2 #
n

√ 2
lim n
an = lim (3 + sin n)n 1 −
n→∞ n→∞ n
 n  n
2 2 4
= lim (3 + sin n) 1 − ≤ lim 4 1 − = < 1.
n→∞ n n→∞ n e2
Thus the series converges.

3. The Dirichlet Test and Alternating Series

3.1. The Dirichlet Test. The following theorem is due to Neils Abel (1802-
1829).
44 2. SERIES OF REAL NUMBERS

Theorem 3.1 (Abel Partial Summation Formula). Let {ak } and {bk } be se-
quences, and let {Ak }k≥0 be a sequence such that
Ak − Ak−1 = ak
for each k ≥ 1. Then if 1 ≤ p ≤ q,
q q−1
X X
ak bk = Aq bq − Ap−1 bp + Ak (bk − bk+1 ).
k=p k=p

Proof. Since ak = Ak − Ak−1 ,

ap bp + ap+1 bp+1 + · · · + aq bq

= (Ap − Ap−1 )bp + (Ap+1 − Ap )bp+1 + (Ap+2 − Ap+1 )bp+2 + · · · + (Aq − Aq−1 )bq
= −Ap−1 bp + Ap (bp − bp+1 ) + Ap+1 (bp+1 − bp+2 ) + · · · + Aq−1 (bq−1 − bq ) + Aq bq
q−1
X
= −Ap−1 bp + Aq bq + Ak (bk − bk+1 ).
k=p


Remark. There are two canonical choices of {An }
n
X
(1). A0 = 0, An = ak . Then An − An−1 = an for all n.
k=1
∞
X
(2). Suppose that ak converges. Then we can also choose {An } by letting
k=1
n
X X∞ ∞
X P∞
An = ak − ak = − ak , n ≥ 1, A0 = − k=1 ak . In this case, we
k=1 k=1 k=n+1
also have
∞
! ∞
!
X X
An − An−1 = − ak − − ak = an .
k=n+1 k=n

An application is to give the following theorem of Peter Lejeune Dirichlet (1805-

1859).
Theorem 3.2 (Dirichlet Test). Suppose that {ak } and {bk } are sequences of real
numbers satisfying the following:
n
X
(i). the sequence of the partial sums An = ak is bounded,
k=1
(ii). b1 ≥ b2 ≥ b3 ≥ · · · ≥ 0, and
(iii). limk→∞ bk = 0.
X∞
Then the series ak bk converges.
k=1
 3. THE DIRICHLET TEST AND ALTERNATING SERIES 45

Proof. Since {An } is bounded, there exists a positive number M > 0 such that
|An | < M for all n. Also, since lim bn = 0, given  > 0, there exists N such that
n→∞

bn = |bn − 0| < for all n > N . Now, for m > n > N , by Abel partial summation
2M
formula,    
 X m   m−1
X 
ak bk  = Am bm − An bn+1 + Ak (bk − bk+1 )
   

   
k=n+1 k=n+1
m−1
X
≤ |Am | · |bm | + |An | · |bn+1 | + |Ak | · |bk − bk+1 |
k=n+1
m−1
X
≤ M ·bm +M ·bn+1 + M ·(bk −bk+1 ) (because bk ≥ bk+1 ≥ 0, |Ak | ≤ M )
k=n+1
= M · [bm + bn+1 + (bn+1 − bn+2 ) + (bn+2 − bn+3 ) + · · · + (bm−2 − bm−1 ) + (bm−1 − bm )]

= 2M bn+1 < 2M · = .
2M
X∞
Hence by the Cauchy Criterion, the series ak bk converges. 
k=1

3.2. Alternating Series Test. An alternating series is of the form

X∞
(−1)n+1 an = a1 − a2 + a3 − a4 + · · · , or
n=1
∞
X
(−1)n an = −a1 + a2 − a3 + a4 − · · ·
n=1
with each an > 0.
Example 3.3.
1 − 1 + 1 − 1 + 1 − 1 + 1 − 1 + ···
1 1 1
1 − + − + ···
2 3 4
−1 + 2 − 3 + 4 − 5 + 6 − · · ·
Theorem 3.4 (The Alternating Series test). If {bn } is a sequence satisfying
(i) b1 ≥ b2 ≥ b3 ≥ · · · ≥ 0, and
(ii) lim bn = 0,
n→∞
∞
X ∞
X
then (−1)n+1 bn ( and (−1)n bn ) converge.
n=1 n=1
n
X
k+1
Proof. Let ak = (−1) and let An = ak . Then
k=1

  0 when n even
|An | = 1 − 1 + 1 − 1 + · · · + (−1)n + (−1)n+1  =
1 when n odd
Thus |An | ≤ 1 for all n, and the Dirichlet test applies. 
46 2. SERIES OF REAL NUMBERS

Example 3.5. Show that convergence or divergence of the series


∞  convergence p > 0
n 1
X
(−1) p =
n=1
n 
divergence p ≤ 0.
1
Proof. If p ≤ 0, then the n-th term (−1)n p does not tend to 0. Thus the series
n
diverges in this case by the divergence test.
1
Assume that p > 0. Let an = p . Then an > 0, monotone decreasing and
n
lim an = 0. Thus the alternating series converges in this case.
n→∞
∞
X 1
In conclusion, we have that the series (−1)n p converges when p > 0 and
n=1
n
diverges when p ≤ 0. 
Now we are going to give a theorem providing an estimate on the sum of (certain)
∞
X 1
series. For instance, let S = (−1)n+1 2 . By taking the partial sum, we have
n=1
n
1 1 1
S ≈ S1 = 1, S ≈ S2 = 1 − 2
= 0.75, S ≈ S3 = 1 − 2 + 2 ≈ 0.861, . . . .
2 2 3
By using computer program, we are able to compute much more, say S1000000 . A
mathematical problem is then what is the ‘error’ for estimating S by using the partial
sum Sn . In other words, how to estimate the remainder
Rn = |S − Sn | = |an+1 + an+2 + · · · |.
Theorem 3.6 (Alternating Series Estimation). Let {bn } be a sequence satisfying
(i) b1 ≥ b2 ≥ b3 ≥ · · · ≥ 0, and
(ii) lim bn = 0.
n→∞
Let
n
X ∞
X
k+1
Sn = (−1) bk and S= (−1)k+1 bk
k=1 k=1
Then the remainder Rn = |S − Sn | ≤ bn+1 for all n.
Proof.  
Rn = (−1)n+2 bn+1 + (−1)n+3 bn+2 + · · · 
= |bn+1 − bn+2 + bn+3 − bn+4 + · · · | .
Since
bn+1 − bn+2 + bn+3 − bn+4 + · · ·
= bn+1 − (bn+2 − bn+3 ) − (bn+4 − bn+5 ) − (bn+6 − bn+7 ) − · · · ≤ bn+1
and
bn+1 − bn+2 + bn+3 − bn+4 + · · · = (bn+1 − bn+2 ) + (bn+3 − bn+4 ) + · · · ≥ 0,
we have Rn ≤ an+1 . 
 3. THE DIRICHLET TEST AND ALTERNATING SERIES 47

∞
X 1
Example 3.7. Estimate (−1)n+1 with error within 0.001.
n=1
n4
1
Solution. From ≤ 10−3 , we have n + 1 ≥ 6 or n ≥ 5. Thus
(n + 1)4
∞
X 1 1 1 1 1
(−1)n+1 4 ≈ 1 − 4 + 4 − 4 + 4
n=1
n 2 3 4 5
with error within 0.001. 
3.3. Trigonometric Series. Another application of Dirichlet test is to conver-
gence of trigonometric series. These series will be studied in much detail in the course
on Fourier series with a lot of applications in physics and other sciences. (Joseph
Fourier, 1768-1830.) We require the following two identities.
Lemma 3.8. For t 6= 2pπ, p ∈ Z,
 
1 1
n cos t − cos n + t
X 2 2
sin kt =
1
k=1 2 sin t
2
 
1 1
n sin n + t − sin t
X 2 2
cos kt =
1
k=1 2 sin t
2
Proof. We prove the second identity. A proof of the first identity can be found
from text book [2, pp.297].
Using the trigonometric identity
1
sin x cos y = (sin(x + y) + sin(x − y)) ,
2
we obtain
n n n     
1 X X 1 1X t t
sin t cos kt = sin t cos kt = sin + kt + sin − kt
2 k=1 k=1
2 2 k=1
2 2
n      
1X 1 1
= sin k + t − sin k − t
2 k=1 2 2
           
1 1 1 1 1
= sin 1 + t − sin 1 − t + sin 2 + t − sin 2 − t
2 2 2 2 2
           
1 1 1 1
+ sin 3 + t − sin 3 − t + · · · + sin n + t − sin n − t
2 2 2 2
   
1 1 1
= sin n + t − sin t .
2 2 2

Theorem 3.9 (Trigonometric Series Test). Let {bn } be a sequence satisfying
48 2. SERIES OF REAL NUMBERS

(i) b1 ≥ b2 ≥ b3 ≥ · · · ≥ 0, and
(ii) lim bn = 0.
n→∞
Then
∞
X
(a). The series bk sin kt converges for all t ∈ R, and
k=1
X∞
(b). The series bk cos kt converges for all t ∈ R, except perhaps t = 2pπ,
k=1
p ∈ Z.
n
X
Proof. (a). Let ak = sin kt and let An = ak . If t = 2pπ, then ak =
k=1
sin 2kpπ = 0 and so An = 0 for t = 2pπ with p ∈ Z. If t 6= 2pπ, then
  
  cos 1 t − cos n + 1 t 


n
X   2 2 
|An | =  sin kt = 
 
1 
2 sin t
  
k=1 
 2 
     
cos 1 t + cos n + 1 t
   
 2   2  1+1 1
≤ 
 1 
 ≤ 
 1 
=
 1 .

2 sin t
  2 sin t
 
sin 2 t
 
2 2
Thus {|An |} is bounded above, and the Dirichlet test applies.
n
X
(b). Let ak = cos kt and let An = ak . The Dirichlet test applies because
k=1
  
  sin n + 1 t − sin 1 t 


n
X   2 2 
|An | =  cos kt = 
 
1 
2 sin t
  
k=1 
 2 
     
sin n + 1 t + sin 1 t
   
 2   2  1+1 1
≤ 
 1 
 ≤ 
 1 
=
 1 .

2 sin t
  2 sin t
 
sin 2 t
 
2 2

∞
X cos kt
Example 3.10. Determine convergence or divergence of the series ,
k=1
kq
t ∈ R, q > 0.
Solution. When t = 2pπ with p ∈ Z, the series
∞ ∞
X cos kt X 1
=
k=1
kq k=1
kq
and so it converges for q > 1 and diverges for q ≤ 1.
 4. ABSOLUTE AND CONDITIONAL CONVERGENCE 49

When t 6= 2pπ, then the series converges by the trigonometric series test because,
1 1
for q > 0, the sequence q is positive, monotone decreasing and lim q = 0. In
k k→∞ k
conclusion, the series

∞  convergent q > 1, t ∈ R
X cos kt
q
is convergent 0 < q ≤ 1, t ∈ R, t 6= 2pπ, for any p ∈ Z
k=1
k 
divergent 0 < q ≤ 1, t = 2pπ for some p ∈ Z

4. Absolute and Conditional Convergence

∞
X ∞
X
Definition 4.1. A series an is called absolutely convergent if |an |
n=1 n=1
converges.

Theorem 4.2. Every absolutely convergent series is convergent.

∞
X ∞
X
Proof. Suppose that an converges absolutely, that is, |an | converges by
n=1 n=1
n
X n
X
the definition. Let Tn = |ak |, Sn = ak . Since {Tn } converges, {Tn } is Cauchy.
k=1 k=1
Thus, for any  > 0, there is a N such that |Tn − Tm | <  for all n, m > N . For any
n, m > N , we may assume that m ≥ n, say m = n + p (as one of them should be
greater than another). Then

|Sn − Sm | = |Sn − (Sn + an+1 + an+2 + · · · + an+p )| = |an+1 + an+2 + · · · + an+p |

≤ |an+1 | + |an+2 | + · · · + |an+p | = Tm − Tn = |Tn − Tm | < .

∞
X
Thus {Sn } is a Cauchy sequence and so it converges. Thus the series an converges
n=1
and hence the result. 

Example 4.3. Determine convergence or divergence of the series

1
∞ sin n +
2.
X

n=2
n(ln n)2

Solution. Since
1 1 1
 
 sin n +  | sin n| +

1+
 2  ≤ 2 ≤ 2
 n(ln n)2  n(ln n)2 n(ln n)2

 
50 2. SERIES OF REAL NUMBERS

1 1
 

∞ 1+ 3X
∞
1
∞  sin n + 
2 2
X X
and = converges by Example 2.9, the series

n(ln n) 2 2 n=2 n(ln n)2 2
 n(ln n) 
 
n=2 n=2  
1
∞ sin n +
2 converges.
X
converges. Thus the series 2

n=2
n(ln n)

Remark. If you are testing for absolute convergence, all the techniques for the
positive series are applicable.

Q: Is the converse of the Corollary true? I.e., if a series is convergent, will it be

absolutely convergent?
A: No, it is not necessarily true.
∞
X 1
Example 4.4. The series (−1)n converges by Example 3.5, but it is NOT
n=1
n
absolutely convergent by the p-series.
∞
X
Definition 4.5. A series an is said to be conditionally convergent if
n=1
∞
X ∞
X
an converges but |an | diverges.
n=1 n=1
∞
X 1
Example 4.6. The series (−1)n √ is conditionally convergent.
n=1
n

Remark 4.7. Every series is either absolutely convergent, conditionally conver-

gent or divergent. 
Example 4.8. The series


 absolutely convergence p>1
∞


X 1 
(−1)n p = conditionally convergence 0 < p < 1
n=1
n 



 divergence p≤0

5. Remarks on the various tests for convergence/divergence of series

1. n-th term test for divergence:
- a test for divergence ONLY, and it works for series with positive and negative
X∞
terms, e.g. (−1)n .
n=1
2. Comparison test/Limit Comparison test:
- when applying these tests, one usually compares the given series with a geometric
series or a p-series.
- generally works for series which look like the geometric series or the p-series,
 5. REMARKS ON THE VARIOUS TESTS FOR CONVERGENCE/DIVERGENCE OF SERIES 51

∞ ∞ 1
X 2 + (−1)n X 2n
e.g. , .
n=1
4n n=1
n2
∞
X 2 + (−1)n
- when an oscillating factor/term appears, e.g. , try the Comparison
n=1
3n
test rather than the Limit Comparison test.
3. Integral test:
∞
X 1
e.g. .
n=2
n(ln n)2
4. Ratio test:
- generally works for series which look like the geometric series, series with n!, and
certain series defined recursively,
∞ ∞
X n2 X (2n)!
e.g. n
, n · n!
,
n=1
3 n=1
4
∞
X 1 1
an , where a1 = 1, an = ( + )an−1 , n = 2, 3, · · · .
n=1
2 n
5. (Simplified) Root test:
- generally works for series where an involves a high power such as the n-th power,
∞ ∞
X n X n 1 n2
e.g. , 2 1− .
n=1
3n n=1 n
6. Alternating Series test: - works for alternating series only,
∞
X ln n
e.g. (−1)n .
n=2
n
X∞
Remark. In general, Tests 2 - 5 works only for an , where an ≥ 0.
n=1
 CHAPTER 3

Sequences and Series of Functions

1. Pointwise Convergence
1.1. Sequences of Functions. Let I be a (nonempty) subset of R, e.g. (−1, 1), [0, 1],
etc. For each n ∈ N, let Fn : I → R be a function. Then we say {Fn } forms a se-
quence of functions on I.
Example 1.1. 1. Fn (x) = xn , 0 < x < 1. Then {Fn } forms a sequence of
functions on (0,1).
n x n o
2. 1+ forms a sequence of functions on (−∞, ∞).
n
Write out some terms:
 x 2  x 3
F1 (x) = 1 + x F2 (x) = 1 + F3 (x) = 1 +
2 3
If we fix the x, and let n → ∞,
 x n
lim Fn (x) = lim 1 + = ex .
n→∞ n→∞ n
So for each x, we can define
F (x) = lim Fn (x).
n→∞

Definition 1.2. A sequence {Fn } is said to converge pointwise to a function

F on I if
lim Fn (x) = F (x) for each x ∈ I,
n→∞
i.e., for each x ∈ I and given any  > 0, there exists an N (which depends on x and
) such that
|Fn (x) − F (x)| <  ∀n > N.
The function F is called the limiting function of {Fn }.
Remark. The limiting function is necessarily unique.
Example 1.3. The sequence {xn } converges pointwise on I = (0, 1) because the
limit F (x) = lim xn = 0 exists for any 0 < x < 1.
n→∞
The sequence {xn } does NOT converge on [−1, 1] because the limit lim xn does
n→∞
not exist when x = −1 ∈ [−1, 1].
1.2. Series of Functions. A series of functions on a set I is of the form
X∞
fn (x) = f1 (x) + f2 (x) + · · · ,
n=1

where each fn is a function on I.

53
54 3. SEQUENCES AND SERIES OF FUNCTIONS

Example 1.4. Below are some examples.

X∞
1. xn−1 = 1 + x + x2 + x3 + · · ·
n=1
∞
X sin kx sin x sin 2x sin 3x
2. = + + + ··· , 0 ≤ x ≤ 1.
k=1
k+x 1+x 2+x 3+x

As in chapter 2, we may form the partial sums

n
X
Sn (x) = fk (x) = f1 (x) + f2 (x) + · · · + fn (x).
k=1

Then {Sn } forms a sequence of functions on I.

∞
X
Definition 1.5. The series fn is said to converge pointwise (to a function
n=1
S) on I if {Sn } converges pointwise (to S) on I, (i.e. lim Sn (x) = S(x) for each
n→∞
x ∈ I.)
∞
X
Example 1.6. What is the pointwise limit of xn−1 , where x ∈ (−1, 1)? Does
n=1
∞
X
xn−1 converge pointwise on [−1, 1)
n=1

Solution. Consider the partial sum

n
X 1 − xn
Sn (x) = xi−1 = 1 + x + · · · + xn−1 =
i=1
1−x
for −1 < x < 1. Thus
∞
X
n−1 1 − xn 1
x = lim Sn (x) = lim =
n→∞ n→∞ 1 − x 1−x
n=1

for x ∈ (−1, 1).

∞
X ∞
X
Since the series xn−1 diverges when x = −1, the series of functions xn−1
n=1 n=1
does NOT converge pointwise on [−1, 1). 

1.3. Some Questions on Pointwise Convergence. Suppose a sequence of

functions {Fn } converges pointwise to a function F on the interval [a, b]. Also sup-
∞
X
pose a series of functions fn (x) converges pointwise to a function S(x) on [a, b].
n=1
Among the questions we want to consider the following. Some of these questions were
incorrectly believed to be true by many mathematicians prior to nineteenth century,
including the famous Cauchy. Cauchy in his text Cours d’Analyse “proved” a theo-
rem to the effect that the limit of a convergent sequence of continuous functions was
again continuous. As we will see, this result is false!
 1. POINTWISE CONVERGENCE 55

Question (a). If each Fn is continuous at p ∈ [a, b], is F necessarily continuous at

p? Recall that F is continuous at p if and only if
lim F (t) = F (p).
t→p

Since F (x) = limn→∞ Fn (x) for every x ∈ [a, b], what we really asking is does
   
lim lim Fn (t) = lim lim Fn (t) ?
t→p n→∞ n→∞ t→p

Question (a’). For series of functions, we can ask similar question. If each fn (x)
∞
X
is continuous at p, is S(x) = fn (x) necessarily continuous at p? Again what we
n=1
really asking is does
∞
X ∞
X
lim fn (t) = lim fn (t) ?
t→p t→p
n=1 n=1
Question (b). If each Fn is differentiable on [a, b], is F necessarily differentiable on
[a, b]? If so, does
d d
lim Fn (x) = lim Fn (x) ?
dx n→∞ n→∞ dx
X∞
Question (b’). If each fn is differentiable on [a, b], is S(x) = fn (x) necessarily
n=1
differentiable on [a, b]? If so, does
∞ ∞
d X X d
fn (x) = fn (x) ?
dx n=1 n=1
dx
Question (c). If each Fn is Riemann integrable on [a, b], is F necessarily Riemann
integrable on [a, b]? If so, does
Z b Z b
lim Fn (x) dx = lim Fn (x) dx ?
a n→∞ n→∞ a
∞
X
Question (c’). If each fn is Riemann integrable on [a, b], is S(x) = fn (x)
n=1
necessarily Riemann integrable on [a, b]? If so, does
Z bX∞ X∞ Z b
fn (x) dx = fn (x) dx ?
a n=1 n=1 a

Without additional hypothesis the answer to all of these questions is generally

no. This additional hypothesis is so-called uniform convergence, introduced by
Weierstrass in 1850. For his many contributions to the subject area, Weierstrass
is often referred to as the father of modern analysis. The subsequent study on the
subject together with questions from geometry and physics also lead to a new area
called topology in the end of 19th century. Poincaré is often referred to as the father
of topology. Topology together with Riemann geometry provide the mathematics
foundation for Einstein’s relativity theory. You may learn some basic knowledge of
topology and modern geometry in 4000 and 5000 modules.
56 3. SEQUENCES AND SERIES OF FUNCTIONS

Below we only give counter-examples to Questions (a) and (c). You may read the
text book [2] for more examples.
Example 1.7 (Counter-example to Question (a)). Consider the functions
Fn (x) = xn , x ∈ [0, 1].
For each fixed x ∈ [0, 1), we have
lim Fn (x) = lim xn = 0 (since |x| < 1).
n→∞ n→∞
At x = 1, we have
lim Fn (1) = lim 1n = 1.
n→∞ n→∞
Thus {Fn } converges pointwise to the function F on the interval [0, 1] given by

 0, for x ∈ [0, 1),
F (x) =
1, x = 1.


Each Fn is continuous on the whole interval [0, 1], but F is not continuous at x = 1.
This simple example gives a counter example to Cauchy’s (false) statement that
the limit of continuous functions is continuous, namely the limit of continuous func-
tions need not be continuous.
Example 1.8 (Counter-example to Question (c)). Consider the functions


 n2 x, 0 < x < n1 ,



Fn (x) = 2n − n2 x, n1 ≤ x < n2 ,



2

 0, n
≤ x < 1.
For each fixed x ∈ (0, 1], one sees that Fn (x) = 0 whenever n ≥ x2 , and hence
lim Fn (x) = lim 0 = 0.
n→∞ n→∞
Also, at x = 0, we have
lim Fn (0) = lim n2 · 0 = 0.
n→∞ n→∞
Thus, {Fn } converges pointwise to the zero function F (x) ≡ 0 on the interval [0, 1].
For each n ≥ 1, we have
Z 1 Z 1/n Z 2/n Z 1
2 2
Fn (x) dx = n x dx + (2n − n x) dx + 0 dx
0 0 1/n 2/n
1   2
n2 x2  n n2 x2  n
= + 2nx − 1 + 0
2 0 2
n

1 1
= + (4 − 2) − (2 − ) + 0 = 1.
2 2
Thus we have
Z 1 Z 1
lim Fn (x) dx = lim 1 = 1 6= 0 = F (x) dx, i.e.
n→∞ 0 n→∞ 0
 2. UNIFORM CONVERGENCE 57
Z 1 Z 1  
lim Fn (x) dx 6= lim Fn (x) dx.
n→∞ 0 0 n→∞

Reason: At different x, Fn (x) converges to F (x) at different pace (more specifically,

in the definition of pointwise convergence, the choice of N depends on both  and x).

2. Uniform Convergence
We define a slightly different concept of convergence.
2.1. Uniform Convergence of Sequences of Functions.
Definition 2.1. {Fn } is said to converge uniformly to a function F on a set
I if for every  > 0, there exists an N (which depends only on ) such that
|Fn (x) − F (x)| < 
for ALL x ∈ I whenever n > N .
Remark 2.2. If {Fn } converges uniformly to F on I, then {Fn } converges point-
wise to F on I. Conversely, if {Fn } converges pointwise to F on I, then {Fn } need
not converge uniformly to F on I.
The inequality in the definition can be expressed as
F (x) −  < Fn (x) < F (x) + 
for all x ∈ I and n > N . The geometric interpretation is that for n > N the graph
of y = Fn (x) lies in the band spanned by the curves y = F (x) −  and y = F (x) + ,
that is, given any  > 0 the graph of y = Fn (x) eventually lies in the band spanned
by the curves y = F (x) −  and y = F (x) + .
Example 2.3. The sequence of function Fn = xn converges pointwise to 0 on
[0, 1). But from the graph we can see that the graph of Fn = xn DOES NOT
eventually lie in the band spanned by y = − and y = +. The geometric reason also
tells us that {xn } does not converge uniformly on [0, 1).
2.2. Two Criteria for Uniform Convergence of {Fn }. The following the-
orem is useful (computationally) in determining whether a sequence of functions
converges uniformly or not.
Theorem 2.4 (T -test). Suppose {Fn } is a sequence of functions converging point-
wise to a function F on a set I, and let
Tn = sup |Fn (x) − F (x)|.
x∈I

Then {Fn } converges uniformly to F on I if and only if lim Tn = 0.

n→0

Proof. First we prove the ‘only if’ part. Suppose that {Fn } converges uniformly
to F on I. Then for any given  > 0, there exists N such that

|Fn (x) − F (x)| < for all n > N and x ∈ I
2

⇒ Tn = sup |Fn (x) − F (x)| ≤ <  for all n > N
x∈I 2
58 3. SEQUENCES AND SERIES OF FUNCTIONS

⇒ |Tn − 0| = Tn <  for all n > N.

Hence we have lim Tn = 0.
n→0
Next we prove the ‘if’ part. Suppose that lim Tn = 0. Then for any given  > 0,
n→0
there exists N such that
|Tn − 0| = Tn <  for all n > N

⇒ sup |Fn (x) − F (x)| <  for all n > N

x∈I

⇒ |Fn (x) − F (x)| <  for all n > N and x ∈ I.

Hence {Fn } converges uniformly to F on I. This finishes the proof of the theorem. 
Theorem 2.5 (Cauchy’s Criterion). A sequence of functions {Fn } converges uni-
formly on a set I if and only if given any  > 0, there exists a natural number N
such that
(6) |Fn (x) − Fm (x)| <  for all x ∈ I and all m, n > N.
Remark: Here N does not depend on x.

Proof. First we prove the ‘only if’ part. Suppose that {Fn } converges uniformly
to the function F on I. Then given any  > 0, there exists N such that

|Fn (x) − F (x)| < for all x ∈ I and all n > N.
2
Then for all x ∈ I and m, n > N ,
|Fn (x) − Fm (x)| = |(Fn (x) − F (x)) − (Fm (x) − F (x))|
 
≤ |Fn (x) − F (x)| + |Fm (x) − F (x)| < + = .
2 2
This finishes the proof of the ‘only if’ part.
Next we prove the ‘if’ part. Suppose that equation 6 holds. Then for each fixed
point x ∈ I, {Fn (x)} is a Cauchy sequence of real numbers, and thus by Cauchy’s
criterion for sequences, the sequence of real numbers {Fn (x)} converges. For each
x ∈ I, we denote the limit by F (x) = lim Fn (x). Then {F (x)}x∈I forms a function
n→∞
on I, which we denote by F . Given any  > 0, by equation 6, there exists N such
that

|Fn (x) − Fm (x)| < for all x ∈ I and all m, n > N.
2
Then for each fixed x ∈ I and n > N , we have
|Fn (x) − F (x)| = |Fn (x) − lim Fm (x)|
m→∞

 
= lim |Fn (x) − Fm (x)| ≤ lim = < .
m→∞ m→∞ 2 2
Thus {Fn } converges uniformly to F , and this finishes the proof of the ‘if’ part. 
 2. UNIFORM CONVERGENCE 59

2.3. Examples.
sin2 x
Example 2.6. Show that Fn (x) = , x ∈ (−∞, +∞), converges uniformly.
n
Proof. The limiting function F (x) is
sin2 x
F (x) = lim =0
n→∞ n
for all x ∈ (−∞, +∞). Since
 2 
 sin x  1
Tn = sup |Fn (x) − F (x)| = sup  ≤ →0
x∈(−∞,+∞) x∈(−∞,+∞) n  n
as n → ∞, thus the sequence of functions {Fn (x} converges uniformly on (−∞, +∞).

Example 2.7. Determine whether the following sequences of functions converge
uniformly on the indicated interval.
n2 ln x
(a) fn (x) = , x ∈ [1, +∞);
xn
n2 ln x
(b) fn (x) = , x ∈ [2, +∞).
xn
Solution. Let f (x) = lim fn (x) = 0 for x ≥ 1.
n→∞
n2 ln x
(a). Tn = sup |fn (x) − 0| = sup = sup fn (x). From
x≥1 x≥1 xn x≥1

1 −n n2 − n3 ln x
fn0 (x) = n2 · x − n3 ln x · x−n−1 = = 0,
x xn+1
1
we have n2 − n3 ln x = 0 or x = e n . Observe that fn (x) is monotone increasing for
1 1
1 ≤ x ≤ e n and monotone decreasing for x ≥ e n . Thus
1 n2 · 1 n
Tn = max fn (x) = fn (e n ) =  1 nn = 6→ 0
x≥1
en e

as n → ∞ and so {fn (x)} does NOT converge uniformly.

1
(b). Since e n ≤ 2 for n ≥ 2, the function fn (x) is monotone decreasing on
n2 ln 2
[2, +∞) for n ≥ 2 and so Tn = sup |fn (x) − f (x)| = fn (2) = for n ≥ 2. Since
x≥2 2n
lim Tn = 0, {fn } converges uniformly on [2, +∞). 
n→∞

n2 ln x sin nx
Example 2.8. Show that Fn (x) = converges uniformly on [2, +∞).
xn
Solution. F (x) = lim Fn (x) = 0 for x ≥ 2. Observe
n→∞

n2 ln x| sin nx| n2 ln 2
Tn = sup |Fn (x) − F (x)| = sup ≤
x≥2 x≥2 xn 2n
for n ≥ 2. Since lim Tn = 0, {Fn } converges uniformly. 
n→∞
60 3. SEQUENCES AND SERIES OF FUNCTIONS

Remark. Let {Fn } be a sequence of functions on an interval I. To see whether {Fn }

is uniformly convergent, we may try to do by the following steps.
(1). Determine the limiting function F (x) = lim Fn (x).
n→∞
(2). Determine Tn = sup |Fn (x) − F (x)|.
x∈I
(3). Check whether lim Tn = 0.
n→∞
If Tn is difficult to be determined, then we may try to estimate an upper bound of
Tn ( a lower bound of Tn if we guess that the sequence of functions might not be
uniformly convergent).

2.4. Uniform Convergence of Series of Functions.

∞
X
Definition 2.9. fn is said to converge uniformly (to S) on I if the sequence
n=1
of its partial sums {Sn } converges uniformly (to S) on I.
∞
X
Theorem 2.10 (T -test for Series of Functions). Suppose fn (x) is a series of
n=1
functions converging pointwise on a set I, and let
 
 X∞ 
Tn = sup  fk (x) .
 
x∈I  
k=n+1
∞
X
Then fn (x) converges uniformly on I if and only if lim Tn = 0.
n→∞
n=1
n
X ∞
X
Proof. Let Sn (x) = fk (x) and S(x) = fk (x). Then
k=1 k=1
 
∞
 X 
|Sn (x) − S(x)| =  fk (x)
 
 
k=n+1

and so the T -test, Theorem 2.4, applies. 

∞
X
Theorem 2.11 (Cauchy Criterion). A series of functions fn converges uni-
n=1
formly on a set I if and only if given any  > 0, there exists a natural number N
such that  
m
 X 
f (x) <  for all x ∈ I and all m > n > N.
 
 k
 
k=n+1

Remark: Here N does not depend on x.

Proof. The proof follows by applying the Cauchy Criterion to the partial sums
n
X
Sn (x) = fk (x). 
k=1
 2. UNIFORM CONVERGENCE 61

The following test is very useful in verifying that certain series of functions con-
verge uniformly to some functions on an interval.
∞
X
Theorem 2.12 (Weierstrass M -test). Consider a series of functions fk on a
k=1
set I. Suppose that
(i) |fk (x)| ≤ Mk for all x ∈ I, k = 1, 2, · · · , and
X∞
(ii) Mk converges.
k=1
∞
X
Then fk converges uniformly (to some function) on I.
k=1

Remark. Weierstrass M -test only states that if a series of functions satisfies condi-
tions (i) and (ii), it converges uniformly on I. If a series of functions does not satisfy
these two conditions, the test fails, namely, no conclusion that you can claim from
this test.
∞
X
Proof. Since Mk converges (by (ii)), by the Cauchy Criterion, given any
k=1
 > 0, there exists N such that
 
Xm m
 X 
Mk =  Mk  <  for all m > n > N
 
 
k=n+1 k=n+1

because Mk ≥ 0. Then for all x ∈ I, we have

 
 Xm  Xm m
X
fk (x) ≤ |fk (x)| ≤ Mk < .
 

 
k=n+1 k=n+1 k=n+1
∞
X
Thus, by the Cauchy criterion, fk converges uniformly on I. 
k=1

2.5. Examples.
∞
X cosn x
Example 2.13. Show that converges uniformly on (0, ∞).
n=1
n2 + x

Proof. Since
 cosn x 
 
1
 n2 + x  ≤ n2
 

∞
X 1
for all x ∈ (0, ∞) and the series is convergent by the p-series, the series of
n=1
n2
∞
X cosn x
functions converges uniformly by the Weierstrass M -test and hence the
n=1
n2 + x
result. 
62 3. SEQUENCES AND SERIES OF FUNCTIONS

Example 2.14. Does the series of functions

X∞
n2 xn sin nx
n=1

converge uniformly on the interval [0, 12 ]? Justify your answer.

Solution. Note that   n2
 2 n
n x sin nx ≤ n

2
for x ∈ [0, 12 ]. Since
(n + 1)2
n+1 (n + 1)2 1
lim 2 2 = lim 2
= < 1,
n→∞ n n→∞ 2n 2
2n
∞
Xn 2 X∞
the series n
converges and so the series of functions n2 xn sin nx converges
n=1
2 n=1
uniformly on [0, 21 ] by the Weierstrass M -test. 
∞
X xn
Example 2.15. Show that the series (−1)n+1 converges uniformly on [0, 1].
n=1
n
n
 
n+1 x  1

Note. The M -test fails for this example because sup (−1)  = but
x∈[0,1] n  n
∞
X 1
diverges. In this case, we use T -test or Cauchy Criterion.
n=1
n
xn
Proof. Let an (x) = . Then, for 0 ≤ x ≤ 1, we have
n
a1 (x) ≥ a2 (x) ≥ · · · ≥ 0 and lim an (x) = 0.
n→∞
∞ n
X x
Thus the series (−1)n+1 converges pointwise on [0, 1] by the Alternating Series
n=1
n
Test. By the Alternating Series Estimation,
 
∞ k
 X x  xn+1 1
Tn = sup  (−1)k+1  ≤ sup = .

0≤x≤1 
k=n+1
k  0≤x≤1 n + 1 n+1
1 xn
Since lim = 0, the series (−1)n+1 converges uniformly by the T -test. 
n→∞ n + 1 n
∞
X xn
Example 2.16. Show that the series does not converge uniformly on [0, 1).
n=1
n
Proof. Observe that  ∞ 
 X xk  ∞
X xk
Tn = sup |Sn (x) − S(x)| = sup   = sup .
 
0≤x<1 0≤x<1 
k=n+1
k  0<x<1 k=n+1 k
 3. UNIFORM CONVERGENCE OF {Fn } AND CONTINUITY 63

xn+1 xn+2 xn+1 xn+2 x2n

   
= sup + + ··· ≥ sup + + ··· +
0<x<1 n+1 n+2 0<x<1 n+1 n+2 2n
1 1 1 xk
= + + ··· + because monotone increasing on (0, 1)
n+1 n+2 2n k
1 1 1 1 1
≥ + + ··· + =n· = .
2n 2n 2n 2n 2
∞
X xk
Thus the sequence {Tn } does not tend to 0 and so the series of functions does
k=1
k
NOT converge uniformly on [0, 1) by the T -test. 

3. Uniform Convergence of {Fn } and Continuity

In this section we will prove that the limit of uniformly convergent sequence of
continuous functions is again continuous.
Theorem 3.1. Let {Fn } be a sequence of continuous functions on an interval I.
Suppose that {Fn } converges uniformly to a function F on I. Then F is continuous
on I.
Proof. Fix any point x0 ∈ I. We are going to show that F is continuous at the
point x0 . Given any  > 0, since {Fn } converges uniformly to F on I, there exists N
such that

|Fn (x) − F (x)| < for all x ∈ I and all n > N.
3
Next we fix an n > N (say, n = [N ] + 1). Since Fn is continuous at x0 , there exists
δ > 0 (here δ depends on x0 and ) such that for all x satisfying |x − x0 | < δ, we have

|Fn (x) − Fn (x0 )| < .
3
Then for all x satisfying |x − x0 | < δ, we have
|F (x) − F (x0 )| = |F (x) − Fn (x) + Fn (x) − Fn (x0 ) + Fn (x0 ) − F (x0 )|
≤ |F (x) − Fn (x)| + |Fn (x) − Fn (x0 )| + |Fn (x0 ) − F (x0 )|
  
< + + = .
3 3 3
Thus F is continuous at x0 . Since x0 is arbitrary, it follows that F is continuous on I.
This finishes the proof of the theorem. 
Example 3.2. Find the pointwise limit F of the sequence
x2n
Fn (x) = , x ∈ [0, 1].
1 + x2n
Show using Theorem 3.1 that the convergence is not uniform.
Solution. If 0 ≤ x < 1, we have
x2n 0
lim Fn (x) = lim = = 0.
n→∞ n→∞ 1 + x2n 1+0
64 3. SEQUENCES AND SERIES OF FUNCTIONS

1
If x = 1, then Fn (1) = . Thus the limiting function F (x) is
2

 0 0≤x<1
F (x) =
 1
2
x=1
Because each Fn (x) is continuous but F (x) is not, the sequence of functions {Fn (x)}
does not converge uniformly on [0, 1] by Theorem 3.1. 
It is possible that each Fn and F (x) = lim Fn (x) are continuous, but {Fn } does
n→∞
not converge uniformly to F (x).
2
Example 3.3. Let Fn (x) = nxe−nx . Show that
1) Each Fn (x) is continuous on I = [0, 1].
2) {Fn (x)} converges pointwise to a continuous function on I.
3) {Fn (x)} does not converge uniformly on I.
Proof. Each Fn (x) is continuous because it is a well-defined elementary function
on [0, 1]. Let F (x) = lim Fn (x) on [0, 1]. When x = 0, Fn (0) = 0 for each n and
n→∞
F (0) = limn→∞ 0 = 0, when x 6= 0 (fixed),
2 nx x 1
F (x) = lim Fn (x) = lim nxe−nx = lim 2 = lim 2 = lim = 0.
n→∞ n→∞ enx n→∞
n→∞ nx
e ·x 2 n→∞ xenx2
Thus F (x) = 0 is continuous on [0, 1]. Now Tn = sup |Fn (x) − F (x)| = sup Fn (x).
0≤x≤1 0≤x≤1
From
2 2 2
Fn0 (x) = ne−nx + nxe−nx · (−2nx) = ne−nx (1 − 2nx2 ) = 0,
r
1
we have x = ± . Since Fn (0) = 0 and Fn (1) = ne−n ,
2n
  r
n 1 1
−n· 2n n
sup Fn (x) = max Fn (x) = max 0, n , n · √ · e = .
0≤x≤1 0≤x≤1 e 2n 2e
r
n
Thus Tn = . Since lim Tn = +∞, the sequence {Fn } does not converge uni-
2e n→∞
formly to F (x) on [0, 1]. 
∞
X
Corollary 3.4. Suppose that fk converges uniformly to a function S on an
k=1
interval I. Suppose that each fk is continuous on I. Then S is also continuous on I.
Proof. Consider the sequence of partial sums {Sn } on I, where we have Sn =
n
X
fk . Then {Sn } converges uniformly to S on I. If each fk is continuous on I, then
k=1
each Sn is also continuous on I. Then by Theorem 3.1, S is also continuous on I. 
∞
X x
Example 3.5. Is , x ∈ (0, ∞), a continuous function?
n=1
n2 enx
 4. UNIFORM CONVERGENCE AND INTEGRATION 65

x
Solution. Let fn (x) = . We find an upper bound of fn (x). Observe that
n2 enx
 
−nx 1
 x 0  xe−nx 0 e−nx − nxe−nx e −x
0 n
fn (x) = = = = .
n2 enx n2 n2 n
We obtain that fn0 (x) > 0 for 0 < x < n1 and fn0 (x) < 0 for x > n1 . It follows that
fn (x) is monotone increasing on (0, n1 ] and monotone decreasing on [ n1 , +∞). Thus
  1
1 n 1
sup fn (x) = max fn (x) = fn = 1 = .
0<x<+∞ 0<x<+∞ n 2
ne nn· en3
∞
1 X
Let Mn = . Then |fn (x)| ≤ M n for x ∈ (0, ∞). Since Mn converges by
en3 n=1
∞
X x
the p-series, the series of functions 2 enx
converges uniformly by Weierstrass M -
n=1
n
∞
X x
test on (0, ∞). According to Corollary 3.4, the function 2 nx
is continuous on
n=1
n e
(0, ∞). 
∞
X ∞
X
It is possible that each fn and fn are continuous, but fn does not converge
n=1 n=1
uniformly.
Example 3.6. Consider the geometric series
∞
X 1
xn = 1 + x + x2 + · · · =
n=0
1−x
∞
X
for −1 < x < 1. In this case, each x and n
xn are continuous on I = (0, 1). We
n=0
∞
X
show that xn does not converge uniformly by T -test. Since
n=0
   
Tn = sup xn+1 + xn+2 + · · ·  ≥ sup xn+1 + xn+2 + · · · 
−1<x<1 0≤x<1

= sup xn+1 + xn+2 + · · · ≥ sup xn+1 = 1,

0≤x<1 0≤x<1
∞
X
the sequence {Tn } does not tend to 0 and so the series xn does not converge
n=0
uniformly on (−1, 1) by the T -test. 

4. Uniform Convergence and Integration

Before we go on, we first recall some facts about Riemann integrals (Reference:
Our Text Book [2, Chapter 6, pp. 208-216]).
66 3. SEQUENCES AND SERIES OF FUNCTIONS

4.1. Review of Riemann Integration. Let f be a bounded function on a

finite interval [a, b]. A partition P of [a, b] is a set of points {x0 , x1 , · · · , xn } such
that
a = x0 < x1 < x2 < · · · < xn = b.
For such a partition P and i = 1, 2, · · · n, we denote
Mi (f ) = sup f (x)
x∈[xi−1 ,xi ]

mi (f ) = inf f (x).
x∈[xi−1 ,xi ]

The upper (Riemann) sum of f with respect to the partition P is defined to be

n
X
U (P, f ) := Mi (f )∆xi .
i=1

Here ∆xi = xi − xi−1 . Similarly the lower (Riemann) sum of f with respect to P
is defined to be
n
X
L(P, f ) := mi (f )∆xi .
i=1
Z b Z b
The upper and lower integrals of f , denoted by f (x), dx and f (x), dx,
a a
respectively, are defined by
Z b
f (x) dx = inf{U (P, f ) | P is a partition of [a, b]},
a
Z b
f (x) dx = sup{L(P, f ) | P is a partition of [a, b]}.
a

Remark. If f is bounded, then

Z b Z b
f (x) dx ≤ f (x) dx,
a a

see [2, Theorem 6.1.4, p.211].

A function f is said to be Riemann integrable on [a, b] if
Z b Z b
f (x) dx = f (x) dx.
a a

The common value is called the Riemann integral of f over [a, b], and it is denoted
Z b Z b Z b
by f (x) dx = f (x) dx = f (x) dx.
a a a
Remark. In the following, (1) and (2) are from [2, Theorem 6.1.8].
1) Any continuous function on a finite interval [a, b] is Riemann integrable.
2) Any monotone function on a finite interval [a, b] is Riemann integrable.
3) Any elementary function is continuous on its domain.
 4. UNIFORM CONVERGENCE AND INTEGRATION 67

The elementary functions are the polynomials, rational functions, power functions
(xa ), exponential functions (ax ), logarithmic functions, trigonometric and inverse
trigonometric functions, hyperbolic and inverse hyperbolic functions, and all func-
tions that can be obtained from these by five operations of addition, subtraction,
multiplication, division, and composition.

4.2. The Theorems.

Theorem 4.1. Let {Fn } be a sequence of Riemann integrable functions on a
finite interval [a, b]. Suppose that {Fn } converges uniformly to a function F on [a, b].
Then F is Riemann integrable on [a, b], and
Z b Z b Z b Z b 
lim Fn (x) dx = F (x) dx i.e. lim Fn (x) dx = lim Fn (x) dx.
n→∞ a a n→∞ a a n→∞

Proof. By definition, given any  > 0, there exists N such that


|Fn (x) − F (x)| <
2(b − a)
for all n > N and x ∈ [a, b]. Thus
 
Fn (x) − < F (x) < Fn (x) +
2(b − a) 2(b − a)
for all x ∈ [a, b] and n > N . It follows that
(7)
Z b  Z b Z b Z b 
 
Fn (x) − dx ≤ F (x) dx ≤ F (x) dx ≤ Fn (x) + dx
a 2(b − a) a a a 2(b − a)
for n > N . Since Fn (x) is Riemann integrable, we have
Z b  Z b 
 
Fn (x) − dx = Fn (x) − dx
a 2(b − a) a 2(b − a)
Z b Z b
 
= Fn (x) dx − · (b − a) = Fn (x) dx − ,
a 2(b − a) a 2
Z b  Z b 
 
Fn (x) + dx = Fn (x) + dx
a 2(b − a) a 2(b − a)
Z b Z b
 
= Fn (x) dx + · (b − a) = Fn (x) + .
a 2(b − a) a 2
Together with Inequality (7), we obtain
Z b Z b Z b Z b
 
(8) Fn (x) dx − ≤ F (x) dx ≤ F (x) dx ≤ Fn (x) dx +
a 2 a a a 2
for n > N . It follows that
Z b Z b
0≤ F (x) dx − F (x) dx ≤ .
a a
68 3. SEQUENCES AND SERIES OF FUNCTIONS

Let  → 0, we have
Z b Z b
F (x) dx − F (x) dx = 0
a a

and so F (x) is Riemann integrable. From Inequality (8), we have

Z b Z b Z b
 
Fn (x) dx − ≤ F (x) dx ≤ Fn (x) dx + , that is,
a 2 a a 2
Z b Z b 
  

 Fn (x) dx − F (x) dx ≤ <
 2
a a
for n > N and hence Z b Z b
lim Fn (x) dx = F (x) dx.
n→∞ a a

Example 4.2. Compute, justifying your answer,
Z 1
sin nx
lim dx.
n→∞ 0 n + x2

sin nx
Solution. Let Fn (x) = . Then the limiting function
n + x2
sin nx
F (x) = lim Fn (x) = lim =0
n→∞ n→∞ n + x2

for any given 0 ≤ x ≤ 1, by the Squeeze theorem, because

1 sin nx 1
− ≤ 2
≤
n n+x n
1 1
and lim = − lim = 0. Since
n→∞ n n→∞ n
 
 sin nx  1
Tn = sup |Fn (x) − F (x)| = sup  2
≤ ,
0≤x≤1 0≤x≤1 n + x n
lim Tn = 0 by the Squeeze Theorem and so the sequence of functions {Fn } converges
n→∞
uniformly to F (x). Thus
Z 1 Z 1 Z 1
sin nx sin nx
lim dx = lim dx = 0 dx = 0.
n→∞ 0 n + x2 0 n→∞ n + x
2
0

∞
X
Corollary 4.3. Suppose that fk converges uniformly to a function S on an
k=1
interval [a, b]. Suppose that each fk is a Riemann integrable (bounded) function on
[a, b]. Then S is also Riemann integrable on [a, b], and
Z b X∞ Z b ∞
Z bX ∞ Z b
X
S(x) dx = fk (x) dx, i.e. fk (x) dx = fk (x) dx.
a k=1 a a k=1 k=1 a
 4. UNIFORM CONVERGENCE AND INTEGRATION 69

n
X
Proof. Consider the sequence of partial sums {Sn } on [a, b], where Sn = fk .
k=1
Then {Sn } converges uniformly to S on [a, b]. If each fk is Riemann integrable on
[a, b], then each Sn is also Riemann integrable on [a, b]. Then by Theorem 4.1, S is
also Riemann integrable on [a, b], and
Z b Z b Z bX n
S(x) dx = lim Sn (x) dx = lim fk (x) dx
a n→∞ a n→∞ a k=1

n Z
X b ∞ Z
X b
= lim fk (x) dx = fk (x) dx.
n→∞ a a
k=1 k=1

By using this theorem, we have the following amazing formula.
Example 4.4. Show that
∞
X 1 1 1 1
ln 2 = n
= + 2
+ + ···
n=1
n·2 2 2·2 3 · 23
 n−1 ∞  n−1
  1 1 X 1
Proof. Since xn−1  ≤ for 0 ≤ x ≤ and the geometric series
2 2 n=0
2
∞
X 1
converges, the series xn−1 converges uniformly to on [0, 12 ] by the Weier-
n=1
1−x
strass M -test. Thus we have
Z 1 Z 1 ∞
2 1 2 X
= xn−1 dx
0 1−x 0 n=1
∞ Z 1 ∞ 1/2 X ∞
X 2
n−1
X xn  1
= x dx = = n
.
n=1 0 n=1
n 0 
n=1
n · 2
Since 1
Z
2 1 1 
1
  
1
2
= − ln(1 − x) = − ln 1 − = − ln = ln 2,
0 1−x 0 2 2
we obtain the formula
∞
X 1 1 1 1
ln 2 = n
= + 2
+ 3
+ ··· .
n=1
n · 2 2 2 · 2 3 · 2

Remark 4.5. Example 4.4 gives a way to estimate the number ln 2 because the
remainder
∞
X 1 1 1
Rn = k
= n+1
+ + ···
k=n+1
k·2 (n + 1)2 (n + 2)2n+2
1 1 1
< n+1
+ n+2
+ + ···
(n + 1)2 (n + 1)2 (n + 1)2n+3
70 3. SEQUENCES AND SERIES OF FUNCTIONS
 
1 1 1 1 1 1
= 1 + + 2 + ··· = · 1 = .
(n + 1)2n+1 2 2 (n + 1)2n+1 1− 2
(n + 1)2n
For instance,
1 1 1 1
ln 2 ≈+ + + · · · +
2 2 · 22 3 · 23 10 · 210
1 1
with error less than 10
= .
11 · 2 11264
4.3. Remarks on Theorem 4.1. For completeness we include the following
result, which does not require uniform convergence. The proof requires new theory
called Lebesgue integration. There are applications in the area of probability and
statistics.
Theorem 4.6 (Bounded Convergence Theorem). Let {Fn } be a sequence of Rie-
mann integrable functions converging pointwise to F (x) on [a, b]. Suppose that
(i) F (x) is Riemann integrable, and
(ii) there exists a positive constant M such that
|Fn (x)| ≤ M
for all x ∈ [a, b] and n ∈ N.
Then Z b Z b Z b
lim Fn (x) dx = F (x) dx = lim Fn (x) dx
n→∞ a a a n→∞

5. Uniform Convergence and Differentiation

Theorem 5.1. Let {Fn } be a sequence of functions on [a, b] such that
(i) each Fn0 exists and is continuous on [a, b],
(ii) {Fn } converges pointwise to a function F on [a, b], and
(iii) {Fn0 } converges uniformly on [a, b].
Then F is differentiable on [a, b], and for all x ∈ [a, b],
 
0 0 d   d
F (x) = lim Fn (x), i.e. lim Fn (x) = lim Fn (x) .
n→∞ dx n→∞ n→∞ dx
Remark. Here the differentiability and continuity at the endpoints a and b refer to
the one sided derivatives and limits respectively.
Proof. By (iii), there exists a function g such that {Fn0 } converges uniformly
to g on [a, b]. In particular, lim Fn0 (x) = g(x) for all x ∈ [a, b]. By (i), since each Fn0
n→∞
is continuous on [a, b], Fn0 is also Riemann integrable on [a, b], and by the fundamental
theorem of calculus,
Z x
Fn0 (t) dt = Fn (x) − Fn (a) for all x ∈ [a, b].
a

Letting n → ∞, we have, for all x ∈ [a, b],

Z x
Fn0 (t) dt = lim Fn (x) − Fn (a) = F (x) − F (a).


(9) lim
n→∞ a n→∞
 5. UNIFORM CONVERGENCE AND DIFFERENTIATION 71

On the other hand, since {Fn0 } converges uniformly to g on [a, b], it follows from
Theorem 4.1 that
Z x Z x Z x
0 0


lim Fn (t) dt = lim Fn (t) dt = g(t) dt.
n→∞ a a n→∞ a

Together with equation 9, it follows that

Z x
F (x) − F (a) = g(t) dt for all x ∈ [a, b].
a

By (i) and Theorem 3.1, g is continuous on [a, b]. Then by the fundamental theorem
of calculus, we have
Z x
d
g(t) dt = g(x).
dx a
Together with equation 9, it follows that F is also differentiable on [a, b], and for all
x ∈ [a, b],
Z x
d d
g(t) dt = g(x), i.e.F 0 (x) = g(x), i.e.


F (x) − F (a) =
dx dx a

d
d

lim Fn (x) = lim Fn (x) .
dx n→∞ n→∞ dx

The proof is finished. 

Remark 5.2. By inspecting the proof, Theorem 5.1 still holds when the closed
interval [a, b] is replaced by (a, b), (a, b] or [a, b).
This theorem can be generalized as follows.
Theorem 5.3. Let {Fn } be a sequence of differentiable functions on [a, b] such
that
(a) {Fn (x0 )} converges for some x0 ∈ [a, b], and
(b) {Fn0 } converges uniformly on [a, b].
Then {Fn } converges uniformly to a function F (x) on [a, b] with,
d d
F 0 (x) = lim Fn0 (x), i.e.



lim Fn (x) = lim Fn (x) .
n→∞ dx n→∞ n→∞ dx

The proof of this theorem is omitted, see [2, Theorem 8.5.1, pp.340-341].
∞
X
Corollary 5.4. Let fk be a series of differentiable functions on [a, b] such
k=1
that
∞
X
(a) fk (x0 ) converges for some x0 ∈ [a, b], and
k=1
∞
X
(b) fk0 converges uniformly on [a, b].
k=1
72 3. SEQUENCES AND SERIES OF FUNCTIONS

∞
X
Then fk converges uniformly to a function S(x) on [a, b], and for all x ∈ [a, b],
k=1
∞ ∞ ∞
0
X d X  X d
S (x) = fk0 (x), i.e. fk (x) = fk (x).
k=1
dx k=1 k=1
dx
n
X
Proof. Consider the partial sums Sn (x) = fk (x) on [a, b]. By (a), {Sn (x0 )}
k=1
converges, and, by (b), {Sn0 (x)} converges uniformly on [a, b]. By Theorem 5.3,
∞
X
{Sn (x)} converges uniformly to a function S(x) on [a, b] and so fk (x) converges
k=1
uniformly to S(x) on [a, b] by the definition. Furthermore, by Theorem 5.3, for all
x ∈ [a, b], S 0 (x) = lim Sn0 (x), that is,
n→∞
∞ n 0 n ∞
d X  X X X d
fk (x) = lim fk (x) = lim fk0 (x) = fk (x).
dx k=1 n→∞
k=1
n→∞
k=1 k=1
dx

As an application, we give a proof of binomial series. Let a be any real number.
The binomial number, a chooses n, is defined by
 
a a(a − 1)(a − 2) · · · (a − n + 1)
=
n n!
1 1
1 1

1 2
· 2
− 1 1
for positive integers n. For instance, 2 = , 2 = =− ,
1 2 2 2! 8
1 1
1
1
1 1

3


· −1 2 −2 · −2 · −2 1
2 = 2 2 = 2 = .
3 3! 6 16
 
a
We also use the convention that = 1 for any a.
0
Theorem 5.5 (Binomial Series). Let a be any real constant. Then
∞   ∞  
a a(a − 1) 2 X a n X a n
(1 + x) = 1 + ax + x + ··· = 1 + x = x
2! n=1
n n=0
n
for |x| < 1.
Proof. Let ρ be any fixed positive number with 0 < ρ < 1 and let I = [−ρ, ρ].
Consider the series of functions
∞  
X a n
1+ x
n=1
n
converges when x = 0 ∈ [−ρ, +ρ]. So it satisfies condition (a) of Corollary 5.4. We
check condition (b) of Corollary 5.4, namely, the series of functions
∞  
X a
nxn−1
n=1
n
 5. UNIFORM CONVERGENCE AND DIFFERENTIATION 73

converges uniformly on [−ρ, +ρ]. Note that

    
 a   a  n−1
n−1 
nx  ≤  n  nρ
  
 n
 
 a  n−1
for x ∈ [−ρ, +ρ]. Let Mn =   nρ . Then
n 
|a| · |a − 1| · · · |a − n|
 a 

 · (n + 1) · ρn
Mn+1 
n+1
(n + 1)ρn (n + 1)!
lim = lim  a
 = lim
n→∞ Mn n→∞   nρn−1 n→∞ |a| · |a − 1| · · · |a − n + 1|
n · n · ρn−1
n!
|a| · |a − 1| · · · |a − n| · n! · (n + 1) · ρ
= lim
n→∞ (n + 1)! · |a| · |a − 1| · · · |a − n + 1| · n

|a − n| · ρ
= lim = lim |a/n − 1| · ρ = ρ < 1.
n→∞ n n→∞
∞ ∞  
X X a
Thus the series Mn converges and so the series of functions nxn−1 con-
n=1 n=1
n
verges uniformly on [−ρ, ρ], by the Weierstrass M -test.
∞  
X a n
By Corollary 5.4, the series 1 + x converges uniformly to a function f (x)
n=1
n
on [−ρ, ρ] with
∞  
0
X a
f (x) = nxn−1 .
n=1
n
Next we are going to set up a differential equation that
(1 + x)f 0 (x) = af (x).
(Note. Since the goal is to show that f (x) = (1 + x)a , this equation is observed from
that, if y = (1 + x)a , then y 0 = a(1 + x)a−1 and so (1 + x)y 0 = (1 + x)a = y.)
Now
∞   ∞
0
X a n−1
X a(a − 1) · · · (a − n + 1) · n n−1
f (x) = nx = x
n=1
n n=1
n!
∞ ∞  
X a(a − 1) · · · (a − n + 1) n−1
X a−1
= x =a xn−1 and
n=1
(n − 1)! n=1
n−1
∞   ∞  
0 0 0
X a−1 n−1
X a−1
(1 + x)f (x) = f (x) + xf (x) = a x +a xn
n=1
n−1 n=1
n−1
 ∞   ∞ 
a−1 n X a−1 n X
=a x +a x
n=0
n n=1
n−1
( ∞     ) " ∞  
#
X a−1 a−1 X a
=a 1+ + xn = a 1 + xn = af (x),
n=1
n n − 1 n=1
n
74 3. SEQUENCES AND SERIES OF FUNCTIONS

where    
a−1 a−1
+
n n−1
(a − 1)(a − 2) · · · (a − 1 − n + 1) (a − 1)(a − 2) · · · (a − 1 − n + 2)
= +
n! (n − 1)!
(a − 1)(a − 2) · · · (a − 1 − n + 2)
= (a − 1 − n + 1 + n)
n!
 
a(a − 1) · · · (a − n + 1) a
= = .
n! n
Let y = f (x). Then we obtain the differential equation
dy dy a dx
(1 + x) = ay =
dx y 1+x
Z Z
dy a dx
=⇒ =
y 1+x
=⇒ ln |y| = a ln |1 + x| + A = ln |1 + x|a + A
=⇒ |y| = eln |y| = eA |1 + x|a
=⇒ y = C|1 + x|a ,
where C = ±eA is a constant. By putting x = 0,
∞  
a
X a n
C = (1 + 0) = y(0) = 1 + 0 = 1.
n=1
n
Thus y = |1 + x|a or
∞  
a
X a
(1 + x) = 1 + xn
n=1
n
for |x| ≤ ρ because 1+x > 0 when |x| ≤ ρ < 1. Since ρ is any number with 0 < ρ < 1,
the formula
∞  
a
X a n
(1 + x) = 1 + x
n=1
n
holds for all x ∈ (−1, 1). 
For instance,
∞ 1 1
− 12 2



√ 1
X
k 1 2
1 + x = (1 + x) =2 2 x =1+ x+ x + ···
k=0
k 2 2!
∞ 1
√ 1
− 12 6



3 3 12
X
2 3 k

1 3 2
1 − x = (1 − x ) = −x =1− x + x + ···
k=0
k 2 2!
 12 1 ∞ 1  k
√
 
1 1 2 X
2
1
4.1 = 4 + =2 1+ =2 .
10 40 k=0
k 40
√
Example 5.6. Evaluate 4.1 with error less than 0.001.
 6. POWER SERIES 75

Solution.
 12 1 ∞ 1  k
√
 
1 1 2 X
2
1
4.1 = 4+ =2 1+ =2 .
10 40 k=0
k 40
∞  1
   k
X
2
1
=2+2 .
k=1
k 40
Now
1 1 1 1 1
· · · k − 12 − 1
1




2 2
· 2
− 1 ··· 2
−k+1 k+1 2 · 1− 2
= = (−1)
k k! k!
for k ≥ 2. Let
 k
1 1
· · · k − 12 − 1


2
· 1− 2 1
bk = .
k! 40
Then, for k ≥ 2, we have bk ≥ 0,
1 1

 k+1
· · · k − 12 − 1 · k − 12



2
· 1− 2 1
bk+1 (k + 1)! 40
= 1 1

1

 k
bk 2
· 1 − 2 ··· k − 2 − 1 1
k! 40
k − 12
= ≤ 1,
40(k + 1)
that is, b2 ≥ b3 ≥ · · · ≥ 0, and lim bk = 0 by the Squeeze Theorem because
k→∞
1 k
· 1 · 2 · · · (k − 1)

2 1 1
0 ≤ bk ≤ =
k! 40 2k · 40k
1
for k ≥ 2 and lim = 0. By the alternating series estimation, from
k→∞ 2k · 40k
   k+1 
1

2
1
2·  < 0.001,
 
 k+1 40 
1
we have k ≥ 1, because 2b2 = < 0.001, and so
6400
√
1
1
4.1 ≈ 2 + 2 2 = 2.025
1 40
with error less than 0.001. 

6. Power Series
6.1. Power Series.
Definition 6.1. A power series in x is of the form
X∞
an x n = a0 + a1 x + a2 x 2 + · · ·
n=0
76 3. SEQUENCES AND SERIES OF FUNCTIONS

Example 6.2. Below are some examples

X∞
1. (n + 1)xn = 1 + 2x + 3x2 + 4x3 + 5x4 + · · ·
n=0
∞
X xn x2 x3
2. =1+x+ + + ···
n=0
n! 2! 3!

Definition 6.3. A power series in x − x0 is of the form

∞
X
an (x − x0 )n = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · ·
n=0

Example 6.4. Here are some examples.

X∞
1. (x − 1)n = 1 + (x − 1) + (x − 1)2 + · · · .
n=0
∞
X
2. n2 (x + 2)n = (x + 2) + 22 (x + 2)2 + 32 (x + 2)3 + · · · .
n=1

∞
X
Warning. Don’t expand out the terms an (x−x0 )n in the power series an (x−x0 )n
n=0
because, when you rearrange terms in an (infinite) series, you may get different values.
(For partial sums, you can expand out, if it is necessary, because there are only finitely
many terms.)
X∞
Question: Given a power series an (x − x0 )n , when does it converge and when
n=0
∞
X
does it diverge? In other words, what is the domain of the function an (x − x0 )n .
n=0
We are going to answer this question.

6.2. Radius of Convergence.

∞
X
Definition 6.5. Given a power series an (x − x0 )n , the radius of conver-
n=0
gence R is defined by
1
R= p
n
.
lim |an |
n→∞
p
n
p
n
If lim |an | = ∞, we take R = 0, and if lim |an | = 0, we set R = ∞. If
n→∞ n→∞
|an+1 |
lim exists, R is also given by
n→∞ |an |

1
R=
|an+1 |
lim
n→∞ |an |
 6. POWER SERIES 77

Remark. Recall that

|an+1 | p p |an+1 |
lim ≤ lim n |an | ≤ lim n |an | ≤ lim .
|an | |an |
|an+1 |
If lim exists, then
n→∞ |an |
|an+1 | |an+1 |
lim = lim
|an | |an |
p
n
and so lim |an | exists and
n→∞
p
n |an+1 |
lim |an | = lim .
n→∞ n→∞ |an |
Example 6.6. What is the radius of convergence for the series
x x2 x3 x4 x5 x6
1+ + + 3 + 4 + 5 + 6 + ···
3 42 3 4 3 4
Solution. Since  1
 n = 2k
42k


an =
 1
n = 2k − 1,


32k−1
we have  1

 4
 n = 2k
p
n
|an | =
 1 n = 2k − 1.


3
p
n 1
Thus lim |an | = and so the radius of convergence
n→∞ 3
1 1
R= p
n
= 1 = 3.
lim |an | 3
n→∞


∞
X (4x + 3)n
Example 6.7. Find the radius of convergence of the power series
n=0
n3
Solution. Observe that
∞ ∞ n
(4x + 3)n 4n

X X 3
= · x+ .
n=1
n3 n=1
n 3 4
Thus
1 1 1 1
R= = = = .
|an+1 | 4n+1
·n 3 4 4
lim lim lim 3
n→∞ |an | 1


n→∞ (n + 1)3 · 4n n→∞ 1 +
n

78 3. SEQUENCES AND SERIES OF FUNCTIONS

∞
X
Theorem 6.8. Given any power series ak (x − x0 )k with radius of convergence
k=0
∞
X
R, 0 ≤ R ≤ ∞, then the series ak (x − x0 )k
k=0
(i) converges absolutely for all x with |x − x0 | < R, and
(ii) diverges for all x with |x − x0 | > R.
1
Proof. By definition, the radius of convergence R = 1 . Assertion (i)
lim |ak | k
follows from the root test because, from
 1 1 1 1 1
lim ak (x − x0 )k  k = lim |ak | k · |x − x0 | = |x − x0 | · lim |ak | k = |x − x0 | · < R · < 1,
R R
∞
X 
the series ak (x − x0 )k  converges.
k=0
∞
X
Next we are going to prove (ii) by contradiction. Suppose that ak (x − x0 )k
k=0
converges at a point x with |x − x0 | > R. Then by Theorem 1.7, we have
lim ak (x − x0 )k = 0.
k→∞

Let  = 1. Then there exists N such that

1
|ak (x − x0 )k − 0| < 1 for all k > N ⇒ |ak (x − x0 )k | k < 1 for all k > N
1 1 1 1
⇒ |ak | k < for all k > N ⇒ sup |ak | k ≤ for all n > N
|x − x0 | n≥k |x − x0 |
1 1 1 1 1
⇒ lim |ak | k ≤ ≤ ⇒ < ,
R|x − x0 |
|x − x0 | R
X∞
which is a contradiction. Hence we must have ak (x − x0 )k diverges at each x
k=0
satisfying |x − x0 | > R. 

6.3. Interval of convergence. In view of Theorem 6.8, for a power series

∞
X ∞
X
k
ak (x−x0 ) with radius of convergence R, the set of points at which ak (x−x0 )k
k=0 k=0
is convergent form an interval called the interval of convergence, which must be
either
(x0 − R, x0 + R), (x0 − R, x0 + R],
[x0 − R, x0 + R) or [x0 − R, x0 + R].
Example 6.9. Find the interval of convergence of the power series.
∞ ∞ ∞
X (x − 2)n X (x − 2)n X
(i) (ii) (iii) n(x − 2)n
n=1
n2 n=1
n n=1
 6. POWER SERIES 79

Solution. (i). First we find the radius of convergence

1 1 1
R= = = = 1.
|an+1 | n 2 1
lim lim lim
n→∞ |an | n→∞ (n + 1)2 n→∞ (1 + 1 )2
n

Next we check the ending-points x0 ± R = 2 ± 1 = 1, 3. When x = 1, the series

∞ ∞
X (−1)n X 1
is 2
, which is convergent by Example 3.5. When x = 3, the series is ,
n=1
n n=1
n2
which is convergent by the p-series. Thus the interval of convergence is [1, 3].
(ii). The radius of convergence is
1 1 1
R= = n = = 1.
|an+1 | lim 1
lim n→∞ (n + 1)
lim
n→∞ |an | n→∞ (1 + 1 )
n
∞
X (−1)n
Now we check the ending-points x0 ± R = 1, 3. When x = 1, the series is ,
n=1
n
∞
X 1
which is convergent by Example 3.5. When x = 3, the series is , which is
n=1
n
divergent by the p-series. Thus the interval of convergence is [1, 3).
(iii). The radius of convergence is
1 1 1
R= = =  = 1.
|an+1 | n+1

1
lim lim lim 1 +
n→∞ |an | n→∞ n n→∞ n
∞
X
Now we check the ending-points x0 ± R = 1, 3. When x = 1, the series is n(−1)n ,
n=1
∞
X
and when x = 3, the series is n. Both of these series are divergent by the
n=1
divergence test. Thus the interval of convergence is (1, 3). 

6.4. Uniform Convergence of Power Series.

∞
X
Theorem 6.10 (Abel Theorem). Let an (x − x0 )n be a power series, and let
n=0
R ≥ 0.
∞
X ∞
X
n
(1). If an R converges, then the series of functions an (x − x0 )n converges
n=0 n=0
uniformly on [x0 , x0 + R].
∞
X ∞
X
n
(2). If an (−R) converges, then the series of functions an (x − x0 )n con-
n=0 n=0
verges uniformly on [x0 − R, x0 ].
Proof. We only prove assertion (1). The proof of assertion (2) is similar to.
80 3. SEQUENCES AND SERIES OF FUNCTIONS

x − x0
We may assume that R > 0. Let t = . Then
R
∞
X X∞
an (x − x0 )n = an Rn tn .
n=0 n=0

We are going to show that this series converges uniformly on 0 ≤ t ≤ 1, that is,
∞
X ∞
X
n
x0 ≤ x ≤ x0 +R. Write ān for an R . By the assumption, the series ān = an R n
n=0 n=0
n
X ∞
X ∞
X
converges. Let An = āk − āk = − āk . Then
k=0 k=0 k=n+1

∞
! ∞
!
X X
An − An−1 = − āk − − āk = ān .
k=n+1 k=n

By Abel Partial Summation Formula, for 0 ≤ n < m, 0 ≤ t ≤ 1,

   
m
 X   m−1
X 
āk tk  = Am tm − An tn+1 + Ak (tk − tk+1 )
   

   
k=n+1 k=n+1

m−1
X
m n+1
≤ |Am |t + |An |t + |Ak | · tk (1 − t).
k=n+1
∞
X ∞
X
Since ān converges, the remainders āk = −An tends to 0 and so lim An = 0.
n→∞
n=0 k=n+1

Given  > 0, there exists N such that |An | < for n > N . Now, for m > n > N and
2
0 ≤ t ≤ 1,
 
m
 X  m−1
X
k m n+1
āk t  ≤≤ |Am |t + |An |t + |Ak | · tk (1 − t)


 
k=n+1 k=n+1

m−1
 m  n+1 X 
< t + t + · tk · (1 − t)
2 2 k=n+1
2
 m
t + tn+1 + (1 − t)(tn+1 + tn+2 + · · · + tm−1 )


=
2
 m

= t + tn+1 + (tn+1 + tn+2 + · · · + tm−1 ) − (tn+2 + tn+3 + · · · + tm ) = ·2tn+1 ≤ .
2 2
X∞ ∞
X
By the Cauchy Criterion, the series of functions ān tn = an (x − x0 )n converges
n=0 n=0
uniformly on 0 ≤ t ≤ 1, or on x0 ≤ x ≤ x0 + R. 
∞
X
Theorem 6.11 (Uniform Convergence Theorem). Let an (x − x0 )n be a power
n=0
series of radius of convergence R > 0. Let I be the interval of convergence. Then
 6. POWER SERIES 81

∞
X
the series of functions an (x − x0 )n converges uniformly on any closed interval
n=0
[c, d] ⊆ I.
Remark. (1). I = [x0 −R, x0 +R], [x0 −R, x0 +R), (x0 −R, x0 +R], or (x0 −R, x0 +R).
If I = [x0 − R, x0 + R], then the power series converges on I. In other cases, since I
is not closed, the theorem says that the power series converges on any closed sub-
interval of I. For instance, if x0 = 0, R = 1, and I = [−1, 1), then the power series
converges uniformly on [−1, 0.9], [0, 0.9] and etc, but it need not converge uniformly
on [−1, 1) or [0, 1).
(2). In any of the four cases, the power series converges uniformly on any closed
sub-interval of (x0 − R, x0 + R).

Proof. There are three cases: (i). c ≤ x0 ≤ d, (ii). x0 < c ≤ d, or (iii).

c ≤ d < x0 .
∞
X ∞
X
n
Case (i). c ≤ x0 ≤ d. Since an (d − x0 ) and an (c − x0 )n (because c, d ∈ I),
n=0 n=0
∞
X
where d − x0 ≥ 0 and c − x0 ≤ 0, the power series an (x − x0 )n converges uniformly
n=0
on c ≤ x ≤ x0 ] and x0 ≤ x ≤ d by the Abel theorem and so on the union [c, d] =
[c, x0 ] ∪ [x0 , d].
∞
X ∞
X
n
Case (ii). x0 < c ≤ d. Since an (d − x0 ) converges, the power series an (x −
n=0 n=0
x0 )n converges uniformly on [x0 , d] by the Abel Theorem and so on the sub-interval
[c, d] ⊆ [x0 , d].
∞
X ∞
X
n
Case (iii). c ≤ d < x0 . Since an (c − x0 ) converges, the power series an (x −
n=0 n=0
x0 )n converges uniformly on [c, x0 ] and so on [c, d] ⊆ [c, x0 ]. 
∞
X
Corollary 6.12. Let an (x − x0 )n be a power series of radius of convergence
n=0
R > 0, and let I be the interval of convergence. Suppose that [c, d] is a closed sub-
interval of I. Then
Z dX∞ ∞
n
X (d − x0 )n+1 − (c − x0 )n+1
an (x − x0 ) dx = an .
c n=0 n=0
n + 1
∞
X
Proof. Since an (x − x0 )n converges uniformly on [c, d],
n=0
∞
dX ∞ Z d ∞
(d − x0 )n+1 − (c − x0 )n+1
Z X X
n
an (x − x0 ) dx = an (x − x0 )n dx = an .
c n=0 n=0 c n=0
n+1


82 3. SEQUENCES AND SERIES OF FUNCTIONS

∞
X
Corollary 6.13 (Abel). Let an (x − x0 )n be a power series of radius of con-
n=0
vergence R > 0.
∞
X
(a). If an Rn converges, then
n=0
∞
X ∞
X
n
lim an (x − x0 ) = an R n .
x→(x0 +R)−
n=0 n=0
∞
X
(b). If an (−R)n converges, then
n=0
∞
X ∞
X
n
lim an (x − x0 ) = an (−R)n .
x→(x0 −R)+
n=0 n=0
∞
X
Proof. (a). By the Abel theorem, the power series an (x − x0 )n converges
n=0
∞
X
uniformly on [x0 , x0 + R] and so the function f (x) = an (x − x0 )n continuous on
n=0
[x0 , x0 + R]. Hence
∞
X ∞
X
an Rn = f (x0 + R) = lim −
f (x) = lim an (x − x0 )n .
x→(x0 +R) x→(x0 +R)−
n=0 n=0

The proof of (b) is similar to that of (a). 

∞
X 1
Example 6.14. From the geometric series xn = for |x| < 1, we have
n=0
1−x
∞
1 X
= (−1)n tn
1+t n=0

by letting x = −t. For any x ∈ (−1, 1), we have

Z x ∞ Z x ∞ n+1
1 X
n n
X
n x
ln(1 + x) = dt = (−1) t dt = (−1) .
0 1+t n=0 0 n=0
n+1
∞
X xn+1
Since (−1)n converges when x = 1, we have
n=0
n+1
∞ ∞
X
n1 X xn+1
(−1) = lim− (−1)n = lim ln(1 + x) = ln 2.
n=0
n + 1 x→1 n=0 n + 1 x→1−
In other words,
1 1 1
ln 2 = 1 − + − + ··· .
2 3 4
 7. DIFFERENTIATION OF POWER SERIES 83

∞
X 1
Example 6.15. From the geometric series xn = for |x| < 1, we have
n=0
1−x
∞
1 X
= (−1)n t2n
1 + t2 n=0

by letting x = −t2 . For any x ∈ (−1, 1), we have

Z x ∞ Z x ∞ 2n+1
1 X
n 2n
X
n x
arctan x = 2
dt = (−1) t dt = (−1) .
0 1+t n=0 0 n=0
2n + 1
∞
X x2n+1
Since (−1)n converges when x = ±1, it converges uniformly on [−1, 1] and
n=0
2n + 1
so
∞
X x2n+1
arctan x = (−1)n for all |x| ≤ 1.
n=0
2n + 1
In particular,
∞
π X 1 1 1 1
= arctan 1 = (−1)n = 1 − + − + ··· .
4 n=0
2n + 1 3 5 7
Example 6.16. From
∞
X x2n+1
arctan x = (−1)n for all |x| ≤ 1,
n=0
2n + 1
we have
∞ 2(2n+1) ∞
2
X
nx
X x4n+2
arctan x = (−1) = (−1)n
n=0
2n + 1 n=0
2n + 1
for |x| ≤ 1 and so
Z 1 ∞ 1 ∞
x4n+2
Z
2
X
n
X 1
arctan x dx = (−1) dx = (−1)n .
0 n=0 0 2n + 1 n=0
(2n + 1)(4n + 3)

7. Differentiation of Power Series

Lemma 7.1. Let {an } and {bn } be sequences such that an ≥ 0, lim bn exists with
n→∞
limn→∞ bn 6= 0. Then lim an bn = lim an · lim bn .
n→∞

Proof. Let B = lim bn .Given any  > 0, there exists N such that |bn − B| < 
n→∞
for n > N , that is,
B −  < bn < B +  for n > N.
Thus, since an ≥ 0,
an (B − ) < an bn < an (B + ) for n > N
and so
(B − ) lim an = lim an (B − ) ≤ lim an bn ≤ lim an (B + ) = (B + ) lim an .
84 3. SEQUENCES AND SERIES OF FUNCTIONS

Now, by letting  tend to 0, we have

B · lim an ≤ lim an bn ≤ B · lim an .
Thus
lim an bn = B · lim an = lim an · lim bn .
n→∞

∞
X
Theorem 7.2. Suppose that an (x − x0 )n has radius of convergence R > 0,
n=0
and
∞
X
f (x) = an (x − x0 )n , |x − x0 | < R.
n=0
Then
∞
X
(a). The power series nan (x − x0 )n−1 has radius of convergence R, and
n=1
∞
X
(b). f 0 (x) = nan (x − x0 )n−1 for |x − x0 | < R.
n=1

Proof. (a). By Lemma 7.1,

1 1 1 1 √
lim |nan | n = lim |an | n · n n = lim |an | n · lim n
n
n→∞ n→∞ n→∞ n→∞

1 1 1
lim |an | n · 1 = lim |an | n = .
n→∞ n→∞ R
Thus the power series
∞
X ∞
X ∞
X
nan (x − x0 )n = nan (x − x0 )n = (x − x0 ) · nan (x − x0 )n−1
n=0 n=1 n=1
∞
X
has radius of convergence R and so has nan (x − x0 )n−1 .
n=1
∞
X
(b). For any ρ with 0 < ρ < R, the series of functions nan (x−x0 )n−1 converges
n=1
uniformly on |x − x0 | ≤ ρ by the Uniform Convergence Theorem because the closed
interval [x0 − ρ, x0 + ρ] ⊆ (x0 − R, x0 + R). The result follows from Theorem 5.3. 
∞
X
Remark. The formula f 0 (x) = nan (x − x0 )n−1 need not hold at the end points
n=1
x = x0 ± R in general even if the interval of convergence of ∞ n
P
n=0 an (x − x0 ) is
[x0 − R, x0 + R].
Example 7.3. From Example 6.15,
∞
X x2n+1
arctan x = (−1)n for all |x| ≤ 1.
n=0
2n + 1
 7. DIFFERENTIATION OF POWER SERIES 85

But
∞
1 0
X
= (arctan x) = (−1)n x2n
1 + x2 n=0
only holds for |x| < 1 because when x = ±1, the right hand side diverges (and the
1
left hand side = ).
2
X∞
Corollary 7.4. Suppose that ak (x − x0 )k has radius of convergence R > 0
k=0
∞
X
with pointwise limiting function f (x) on |x − x0 | < R (i.e. f (x) = ak (x − x0 )k
k=0
on |x − x0 | < R), then f (x) has derivatives of all orders on
|x − x0 | < R, and for each n,
∞
X
(10) f (n) (x) = k(k − 1)(k − 2) · · · (k − n + 1)ak (x − x0 )k−n .
k=n

In particular,
f (k) (x0 )
(11) ak = for all k.
k!
∞
X f (k) (x0 )
(i.e. we have f (x) = (x − x0 )k .)
k=0
k!
Proof. The result is obtained by successively applying the previous theorem to
f, f 0 , f 00 , and etc. Equation follows by setting x = x0 in Equation 10, that is,
f (n) (x) = n!an + (n + 1)n · · · 2an+1 (x − x0 ) + (n + 2)(n + 1) · · · 3an+2 (x − x0 )2 + · · · .

Example 7.5. From the geometric series
∞
1 X
= xn = 1 + x + x2 + x3 + · · · |x| < 1,
1 − x n=0
we have
 0 ∞
1 1 X
= = nxn−1 = 1 + 2x + 3x2 + 4x3 + · · ·
(1 − x)2 1−x n=1

for |x| < 1, and so

∞
x X
= nxn = x + 2x2 + 3x3 + 4x4 + · · · |x| < 1.
(1 − x)2 n=1
1
By letting x = , we have
2
∞ 1
X n 2
=
=2
2n 1 2
n=1 1 − 2
86 3. SEQUENCES AND SERIES OF FUNCTIONS

Example 7.6. Consider the function

∞
X xn x2 x3
y= =1+x+ + + ··· .
n=0
n! 2! 3!
The radius of convergence
1 1 1 1
R= = = = = ∞.
|an+1 | n! 1 0
lim lim lim
n→∞ |an | n→∞ (n + 1)! n→∞ n + 1

Thus for any x ∈ (−∞, +∞),

x2 x3
y0 = 0 + 1 + x +
+ + ··· = y
2! 3!
Z Z
dy dy dy
=⇒ =y =⇒ = dx =⇒ = dx =⇒ ln |y| = x + A
dx y y
=⇒ |y| = ex+A =⇒ y = Cex , C = ±eA constant.
Let x = 0.
C = Ce0 = y(0) = 1 + 0 + 0 + · · · = 1.
Hence we obtain the formula
∞
X xn x2 x3
ex = =1+x+ + + ··· .
n=0
n! 2! 3!
Definition 7.7. A real-valued function f defined on an open interval I is said
to be infinitely differentiable on I if all (higher) derivatives f (n) (x), n ≥ 1, exist.
The set of infinitely differentiable functions on I is denoted by C ∞ (I).
X∞
As a consequence, the functions f (x) = an (x−x0 )n are infinitely differentiable
n=0
on (x0 − R, x0 + R) if R > 0.
∞
X ∞
X
n
Corollary 7.8 (Uniqueness Theorem). Suppose that an (x−x0 ) and bn (x−
n=0 n=0
x0 )n are two power series which converge for |x − x0 | < R with R > 0. Then
X∞ ∞
X
an (x − x0 )n = bn (x − x0 )n for |x − x0 | < R
n=0 n=0
if and only if ak = bk for all k = 0, 1, 2, 3, · · ·
∞
X
Proof. Suppose that ak = bk for all k = 0, 1, 2, 3 · · · . Then an (x − x0 )n =
n=0
∞
X
bn (x − x0 )n .
n=0
∞
X ∞
X
Conversely, suppose that an (x−x0 )n = bn (x−x0 )n = f (x) for |x−x0 | < R.
n=0 n=0
f (n) (x0 ) f (n) (x0 )
Then an = and bn = . Thus an = bn for all n = 0, 1, 2, 3, · · · . 
n! n!
 8. TAYLOR SERIES 87

8. Taylor Series
8.1. History Remarks. The study of sequences and series of functions has its
origins in the study of power series representation of functions. The power series
of ln(1 + x) was known to Nicolaus Mercator (1620-1687) by 1668, and the power
series of many other functions such as arctan x, arcsin x, and etc, were discovered
around 1670 by James Gregory (1625-1683). All these series were obtained without
any reference to calculus. The first discoveries of Issac Newton (1642-1727), dating
back to the early months of 1665, resulted from his ability to express functions in
terms of power series. His treatise on calculus, published in 1737, was appropriately
entitled A treatise of the methods of fluxions and infinite series. Among his many
accomplishments, Newton derived the power series expansion of (1 + x)m/n using
algebraic techniques. This series and the geometric series were crucial in many of
his computations. Newton also displayed the power of his calculus by deriving the
power series expansion of ln(1+x) using term-by-term integration of the expansion of
1/(1 + x). Colin Maclaurin (1698-1746) and Brooks Taylor (1685-1731) were among
the first mathematicians to use Newton’s calculus in determining the coefficients in
the power series expansion of a function. Both realized that if a function f (x) had
X∞
a power series expansion an (x − x0 )n , then the coefficients an had to be given by
n=0
f (n) (x0 )
.
n!
8.2. Taylor Polynomials and Taylor Series.
Definition 8.1. Let f (x) be a function defined on an open interval I, and let
x0 ∈ I and n ≥ 1. Suppose that f (n) (x) exists for all x ∈ I. The polynomial
n
X f (k) (x0 )
Tn (f, x0 )(x) = (x − x0 )k
k=0
k!
is called the Taylor polynomial of order n of f at the point x0 . If f is infinitely
differentiable on I, the power series
∞
X f (n) (x0 )
T (f, x0 )(x) = (x − x0 )n
n=0
n!
is called the Taylor series of f and x0 .
For the special case x0 = 0, the Taylor series of a function f is often referred to
as Maclaurin series. The first few Taylor polynomials are as follows:
T0 (f, x0 )(x) = f (x0 ),
T1 (f, x0 )(x) = f (x0 ) + f 0 (x0 )(x − x0 ),
f 00 (x0 )
T2 (f, x0 )(x) = f (x0 ) + f 0 (x0 )(x − x0 ) + (x − x0 )2 ,
2!
f 00 (x0 ) f 000 (x0 )
T3 (f, x0 )(x) = f (x0 ) + f 0 (x0 )(x − x0 ) + (x − x0 )2 + (x − x0 )3 .
2! 3!
The Taylor polynomial T1 (f, x0 ) is the linear approximation of f at x0 , that is
the tangent line passing through (x0 , f (x0 )) with slope f 0 (x0 ).
88 3. SEQUENCES AND SERIES OF FUNCTIONS

In general, the Taylor polynomial Tn of f is a polynomial of degree less than or

equal to n that satisfies the conditions

Tn(k) (f, x0 )(x0 ) = f (k) (x0 )

for 0 ≤ k ≤ n. Since f (n) (x0 ) might be zero, Tn could very well be a polynomial of
degree strictly less than n.

Example 8.2. Find the Maclaurin series of ex .

Solution. Let f (x) = ex . Then f (n) (x) = ex . Thus f (n) (0) = 1 and so the
Maclaurin series of ex is
x2
1+x+ + ··· .
2!


Example 8.3. Find the Taylor series of f (x) = sin x at x0 = π.

Solution.

f (x) = sin x f 0 (x) = cos x f 00 (x) = − sin x f 000 (x) = − cos x ···

f (π) = 0 f 0 (π) = −1 f 00 (π) = 0 f 000 (π) = 1, ···

1
T3 (f, π)(x) = 0 − (x − π) + 0 + (x − π)3 .
3!
∞
1 1 X 1
T (f, π)(x) = −(x−π)+ (x−π)3 − (x−π)5 +· · · = (−1)k+1 (x−π)2k+1 .
3! 5! k=0
(2k + 1)!


1
Example 8.4. Find the Taylor series of f (x) = at x0 = 3.
x
Proof.
1 1 1 1
= = ·
x 3 − (3 − x) 3 1 − 3−x
3

∞ n ∞ ∞
1 X (−1)n X (−1)n

1X 3−x n
= = n
(x − 3) = n+1
(x − 3)n .
3 n=0 3 3 n=0 3 n=0
3
1
Thus the Taylor series of at x0 = 3 is
x
∞
X (−1)n
(x − 3)n .
n=0
3n+1


 8. TAYLOR SERIES 89

8.3. Taylor Theorem. In view of Theorem 7.4, we may ask the following ques-
tion:
Question: Given a infinitely differentiable function f (x), does the equality
∞
X f (k) (x0 )
(12) f (x) = T (f, x0 )(x) = (x − x0 )k
k=0
k!
hold for |x − x0 | < R?
(Here R is the radius of the convergence of the Taylor series.)
It turns out that in general, the answer is NO. (See Example 8.5 for an example of a
function such that equation 12 does not hold.)
However, as we have seen, the above equality does hold for some elementary
functions such as ex , sin x, cos x, ln(1 + x), (1 + x)a , arctan x, and etc. We are going
to give certain hypothesis such that the above equality holds for some functions.
Example 8.5 (Counter-example to the Question). Consider the function
1

 e− x2 x 6= 0,
f (x) =
0 x = 0.


Then we will show that f (x) 6= its Taylor series at x0 = 0.

1
Proof. First we compute f 0 (0). By substituting y =
x2
1
0 f (x) − f (0) e− x2 − 0 1 1
f (0) = lim = lim = lim 1 = lim 1 · x
x→0 x−0 x→0 x x→0
xe x2 x→0 2 x2
xe
y
= lim y · lim x = 0 · 0 (by L’Hopital’s rule) = 0.
y→∞ e x→0
For x 6= 0,
d − 12
1
f 0 (x) = e x = 2x−3 e− x2 .
dx
Thus,
1

 2x−3 e− x2 x 6= 0,
f 0 (x) =
0 x = 0.

1
Next we compute f 00 (x). By substituting y = 2 ,
x
1
00 f 0 (x) − f 0 (0) 2x−3 e− x2 − 0 1
f (0) = lim = lim = 2 lim 1
x→0 x−0 x→0 x x→0 4 x2
xe
y2
= 2 lim = 0 (by L’Hopital’s rule).
y→∞ ey
Again, for x 6= 0,
d 1 1
f 00 (x) = 2x−3 e− x2 ) = (−6x−4 + 4x−6 )e− x2 .
dx
Similar calculations will lead to
f (0) = f 0 (0) = f 00 (0) = f (3) (0) = f (4) (0) = · · · = 0.
90 3. SEQUENCES AND SERIES OF FUNCTIONS

Thus we have
∞ ∞
X f (k) (0) k
X 0 k
x = x = 0 + 0x + 0x2 + · · · = 0.
k=0
k! k=0
k!
∞
− 1 X f (k) (0)
Clearly, at any x 6= 0, f (x) = e x2 6= 0. Therefore, f (x) 6= xk . 
k=0
k!

The remainder or error function between f (x) and Tn (f, x0 ) is defined by

Rn (f, x0 )(x) = f (x) − Tn (f, x0 )(x).
Clearly
n ∞
X f (k) (x0 ) k
X f (k) (x0 )
f (x) = lim Tn (f, x0 )(x) = lim (x − x0 ) = (x − x0 )k
n→∞ n→∞
k=0
k! k=0
k!
if and only if
lim Rn (f, x0 )(x) = 0.
n→∞
To emphasize this fact, we state it as a theorem
Theorem 8.6. Suppose that f is an infinitely differentiable function on an open
interval I and x0 ∈ I. Then, for x ∈ I,
∞
X f (n) (x0 )
f (x) = (x − x0 )n
n=0
n!
if and only if lim Rn (f, x0 )(x) = 0. 
n→∞

The remainder Rn (f, x0 ) has been studied much and there are various forms of
Rn (f, x0 ). We only provide one result called Lagrange Form of the Remainder,
attributed by Joseph Lagrange (1736-1813). But this result sometimes also referred
to as Taylor’s theorem.
Theorem 8.7 (Taylor Theorem). Let f be a function on an open interval I,
x0 ∈ I and n ∈ N. If f (n+1) (t) exists for every t ∈ I, then for any x ∈ I, there exists
a ξ between x0 and x such that
f (n+1) (ξ)
(13) Rn (f, x0 )(x) = (x − x0 )n+1 .
(n + 1)!
Thus
f 00 (x0 ) f (n) (x0 ) f (n+1) (ξ)
f (x) = f (x0 )+f 0 (x0 )(x−x0 )+ (x−x0 )2 +· · ·+ (x−x0 )n + (x−x0 )n+1 ,
2! n! (n + 1)!
for some ξ between x and x0 .
Proof. Recall that
n
X f (k) (x0 )
Rn (f, x0 ) = f (x) − (x − x0 )k .
k=0
k!
 8. TAYLOR SERIES 91

Fixed x ∈ I, let M be defined by Rn (f, x0 ) = M (x − x0 )n+1 , that is,

n
X f (k) (x0 )
f (x) = (x − x0 )k + M (x − x0 )n+1 .
k=0
k!

f n+1 (ξ)
(Note. M depends on x.) Our goal is to show that M = for some ξ between
(n + 1)!
x0 and x.
We construct a function
n
X f (k) (x0 )
g(t) = f (t) − (t − x0 )k − M (t − x0 )n+1
k=0
k!

f (n) (x0 )
 
0
= f (t) − f (x0 ) + f (x0 )(t − x0 ) + · · · + n
(t − x0 ) − M (t − x0 )n+1 .
n!
By taking derivatives, we have
g(x0 ) = g 0 (x0 ) = g 00 (x0 ) = · · · = g (n) (x0 ) = 0.
and
(14) g (n+1) (t) = f (n+1) (t) − (n + 1)!M.
For convenience, let’s assume x > x0 . By the choice of M , g(x) = 0. By applying
the mean value theorem to g on the interval [x0 , x], there exists c1 , x0 < c1 < x, such
that
0 = g(x) − g(x0 ) = g 0 (c1 )(x − x0 ) =⇒ g 0 (c1 ) = 0.
Since g 0 (x0 ) = g 0 (c1 ) = 0, by applying the mean value theorem to g 0 on the interval
[x0 , c1 ], there exists c2 , x0 < c2 < c1 , such that
0 = g 0 (c1 ) − g 0 (x0 ) = g 00 (c2 )(c1 − x0 ) =⇒ g 00 (c2 ) = 0.
Continuing this manner, we obtain points c1 , c2 , · · · , cn , x0 < cn < cn−1 < · · · < c2 <
c1 < x, such that g 0 (c1 ) = 0, g 00 (c2 ) = 0, g 000 (c3 ) = 0, · · · , g (n) (cn ) = 0. By applying
the mean value theorem once more to g (n) on [x0 , cn ], there exists ξ, x0 < ξ < cn ,
such that
0 = g (n) (cn ) − g (n) (x0 ) = g (n+1) (ξ)(cn − x0 ) =⇒ g (n+1) (ξ) = 0.
From Equation (14),
0 = g (n+1) (ξ) = f (n+1) (ξ) − (n + 1)!M,
f (n+1) (ξ)
that is, M = for some ξ between x0 and x (because x0 < ξ < cn < x). 
(n + 1)!
Example 8.8. Show that
∞
X (−1)n x2n+1 x3 x5 x7
sin x = =x− + − + ··· (|x| < ∞).
n=0
(2n + 1)! 3! 5! 7!
92 3. SEQUENCES AND SERIES OF FUNCTIONS

Proof. First the right hand side is the Maclaurin series of f (x) = sin x because
f (x) = sin x f 0 (x) = cos x f 00 (x) = − sin x f 000 (x) = − cos x ···
0 00 000
f (0) = 0 f (0) = 1 f (0) = 0 f (0) = −1, ···
 (−1)n x2n+1 
 
Next let an =  . By the ratio test
(2n + 1)! 
 (−1)n+1 x2n+3 
 
 
an+1  (2n + 3)! 
lim = lim  
n→∞ an n→∞ (−1)n x2n+1 
 (2n+1)! 
x2
= lim =0<1
n→∞ (2n + 3)(2n + 2)
for all x. Thus R = +∞.
In the last step we show that the remainder tends to 0. Since f (x) = sin x, the
higher derivatives f (n+1) (x) is either ± sin x or ± cos x. Thus |f (n+1) (x)| ≤ 1 for all x
and all n. By the Taylor Theorem,
 (n+1)
|x|n+1

f (ξ) n+1

|Rn (f, 0)(x)| =  x  ≤ .
(n + 1) (n + 1)!
|x|n+1
Since lim = 0, we have lim |Rn (f, 0)(x)| = 0, by the Squeeze Theorem, or
n→∞ (n + 1)! n→∞
lim Rn (f, 0)(x) = 0 for all x. Hence
n→∞
∞
X (−1)n x2n+1 x3 x5 x7
sin x = =x− + − + ··· (|x| < ∞).
n=0
(2n + 1)! 3! 5! 7!

8.4. Some Standard Power Series. Below is a list of Maclaurin series of some
elementary functions.
∞
x
X xn x2 x3
e = =1+x+ + + ··· (|x| < ∞)
n=0
n! 2! 3!
∞
X (−1)n x2n+1 x3 x5 x7
sin x = =x− + − + ··· (|x| < ∞).
n=0
(2n + 1)! 3! 5! 7!
∞
X (−1)n x2n x2 x4 x6
cos x = =1− + − + ··· (|x| < ∞).
n=0
(2n)! 2! 4! 6!
∞
X (−1)n+1 xn x2 x3
ln(1 + x) = =x− + − ··· (−1 < x ≤ 1).
n=1
n 2 3
∞
1 X
= xn = 1 + x + x2 + x3 + · · · (|x| < 1).
1 − x n=0
∞
1 X
= (−1)n xn = 1 − x + x2 − x3 + · · · (|x| < 1).
1 + x n=0
 8. TAYLOR SERIES 93

∞
X x2n+1 x3 x5
arctan x = (−1)n =x− + − ··· (−1 ≤ x ≤ 1)
n=0
2n + 1 3 5
∞        
a
X a n a a 2 a 3
(1 + x) = x =1+ x+ x + x + ··· (|x| < 1),
n=0
n 1 2 3
where  
a a · (a − 1) · (a − 2) · · · (a − k + 1)
=
k k!
 
a
for any real number a and integers k ≥ 1, and =1
0
Remark. From these power series, we can obtain Maclaurin series of various more
complicated functions by using operations such as substitution, addition, subtraction,
multiplication, division, integrals, derivatives and etc. For the Maclaurin series of
sin x2 can be obtained by replacing x by x2 in the Maclaurin series of sin x. By using
multiplication, we can obtain the Maclaurin series of ex ·sin x. By using long division,
sin x
we can obtain the Maclaurin series of tan x = . By taking integral, we can obtain
cos x Z x
the Maclaurin series of non-elementary functions, for instance f (x) = sin t2 dt.
0
As an application, we are going to compute number π.
Computation of π
Step 1. Find the Maclaurin series of arcsin x for |x| < 1.
For |x| < 1,
Z x Z x Z xX∞  1
1
1
2 −2 −2
arcsin x = √ dt = 1 + (−t ) dt = (−t2 )k dt
0 1 − t2
0 0 k=0 k

∞ Z x ∞ ∞
− 21
   1  2k+1  1  2k+1
k −2 x k −2 x
X X X
k 2k
= (−1) t dt = (−1) =x+ (−1)
k=0 0 k k=0
k 2k + 1 k=1
k 2k + 1
Note that
           
1 1 1 1 3 2k − 1
− · − − 1 ··· − − k + 1 − · − ··· −
− 12
 
2 2 2 2 2 2
= =
k k! k!

1 · 3 · · · · · (2k − 1)
= (−1)k
k! · 2k
for k ≥ 1. Thus
(15)
∞ ∞
k 1 · 3 · · · · · (2k − 1) 2k+1 1 · 3 · · · · · (2k − 1) 2k+1
X X
k
arcsin x = x+ (−1) (−1) k
x = x+ x
k=1
2 · k! · (2k + 1) k=1
2k · k! · (2k + 1)

for |x| < 1.

π π 1
Step 2. Find a series expansion of using = arcsin .
6 6 2
94 3. SEQUENCES AND SERIES OF FUNCTIONS

From Equation (15), we obtain the following formula.

∞
π 1 X 1 · 3 · · · · · (2k − 1)
(16) = + .
6 2 k=1 k! · (2k + 1) · 23k+1
Step 3. Estimate the remainder.
The remainder of the formula 16 can be estimated as follows.
∞ ∞
X 1 · 3 · · · · · (2k − 1) X 1
Rn = |S − Sn | = <
k=n+1
2k · k! · (2k + 1) · 22k+1 k=n+1 (2k + 1)22k+1
∞  
X 1 1 1 1
< 2k+1
= 1 + + 2 + ···
k=n+1
(2n + 3)2 (2n + 3)22n+3 4 4
1 1
= 1

=
(2n + 3)22n+3 1 − 4
3(2n + 3)22n+1
For instance, let n = 10, we have
10
!
1 X (1 · 3 · · · · · (2k − 1)
π≈6 +
2 k=1 k! · (2k + 1) · 23k+1
with error less than
1 1 1
6· 21
= 20
= .
3 · 23 · 2 23 · 2 24117248
If we choose n = 20, we have
20
!
1 X (2k − 1)!!
π≈6 +
2 k=1 k! · (2k + 1) · 23k+1
with error less than
1 1 1
6· 41
= 40
= < 10−13 .
3 · 43 · 2 43 · 2 47278999994368
1
If n = 40, the error is less than < 10−26 .
100340843028014221500612608
Remark. There are several other methods for computing π. For instance, we can
also use
π 1 1 1
= arctan 1 = 1 − + − + · · · ,
4 3 5 7
but one needs a huge number of terms to get enough accuracy. (So this method is
no good for computational purpose!) Another method is to use the formula of John
Machin (1680-1751):
1 1 π
4 arctan − arctan = ,
5 239 4
see our text book [1, Problem 7, p.813] for details. Machin used his method in 1706
to find π correct to 100 decimal places. In 1995 Jonathan and Peter Borwein of Simon
Fraster University and Yasumasa Kanada of the University of Tokyo calculated the
value of π to 4, 294, 967, 286 decimal places!
 8. TAYLOR SERIES 95

Another story on computing π is a Chinese mathematician and astronomer Tsu

355
Ch’ung Chi (430-501). He gave the rational approximation to π which is correct
113
to 6 decimal places. He also proved that
3.1415926 < π < 3.1415927
a remarkable result (Note. He was a person lived 1500 years ago!), on which it would
be nice to have more details but Tsu Ch’ung Chi’s book, written with his son, is lost.
(His method is to cut off the circle by equal pieces to get his approximation to π.)
Tsu’s astronomical achievements include the making of a new calendar in 463 which
never came into use. (According to the article of J J O’Connor and E F Robertson
in http://www-groups.dcs.st-and.ac.uk/ history/Mathematicians/Tsu.html.)
Remark. Those, who are interested in more applications of Taylor series, can try to
finish the applied project, Radiation from the Stars, in our text book [1, pp.808-809].
 Bibliography

[1] James Stewart, Calculus, 4th Edition, Brooks/Cole Publishing Company press, 1999. (Text
Book)
[2] Manfred Stoll, Introduction to Real Analysis, 2nd Edition, Addison Wesley Longman, Inc.
press, 2001. (Text Book)
[3] Watson Fulks, Advanced Calculus, 3rd Edition, John Wiley & Sons, Inc. press, 1961.
[4] Wilfred Kaplan, Advanced Calculus, 3rd Edition, Addison-Wesley Publishing Company
press, 1984.
[5] William R. Parzynski and Philip W. Zipse, Introduction to mathematical analysis, Inter-
national Edition, McGraw-Hill Book Company press,1987.
[6] G. B. Thomas, Jr. and Ross L. Finney, Calculus and analytic geometry, 9th Edition,
International Student Edition, Addison-Wesley Longman Inc. press, 1996.

97