Calculus

Min Yan
Department of Mathematics
Hong Kong University of Science and Technology

May 13, 2020

2
Contents

1 Limit 7
1.1 Limit of Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Arithmetic Rule . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Sandwich Rule . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.3 Some Basic Limits . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.4 Order Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.1.5 Subsequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.2 Rigorous Definition of Sequence Limit . . . . . . . . . . . . . . . . . . 25
1.2.1 Rigorous Definition . . . . . . . . . . . . . . . . . . . . . . . . 26
1.2.2 The Art of Estimation . . . . . . . . . . . . . . . . . . . . . . 28
1.2.3 Rigorous Proof of Limits . . . . . . . . . . . . . . . . . . . . . 31
1.2.4 Rigorous Proof of Limit Properties . . . . . . . . . . . . . . . 33
1.3 Criterion for Convergence . . . . . . . . . . . . . . . . . . . . . . . . 37
1.3.1 Monotone Sequence . . . . . . . . . . . . . . . . . . . . . . . . 38
1.3.2 Application of Monotone Sequence . . . . . . . . . . . . . . . 42
1.3.3 Cauchy Criterion . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.4 Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.4.1 Divergence to Infinity . . . . . . . . . . . . . . . . . . . . . . . 48
1.4.2 Arithmetic Rule for Infinity . . . . . . . . . . . . . . . . . . . 50
1.4.3 Unbounded Monotone Sequence . . . . . . . . . . . . . . . . . 52
1.5 Limit of Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.5.1 Properties of Function Limit . . . . . . . . . . . . . . . . . . . 53
1.5.2 Limit of Composition Function . . . . . . . . . . . . . . . . . 56
1.5.3 One Sided Limit . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.5.4 Limit of Trigonometric Function . . . . . . . . . . . . . . . . . 63
1.6 Rigorous Definition of Function Limit . . . . . . . . . . . . . . . . . . 66
1.6.1 Rigorous Proof of Basic Limits . . . . . . . . . . . . . . . . . 67
1.6.2 Rigorous Proof of Properties of Limit . . . . . . . . . . . . . . 70
1.6.3 Relation to Sequence Limit . . . . . . . . . . . . . . . . . . . 73
1.6.4 More Properties of Function Limit . . . . . . . . . . . . . . . 78
1.7 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
1.7.1 Meaning of Continuity . . . . . . . . . . . . . . . . . . . . . . 80
1.7.2 Intermediate Value Theorem . . . . . . . . . . . . . . . . . . . 82

3
4 CONTENTS

1.7.3 Continuous Inverse Function . . . . . . . . . . . . . . . . . . . 84

1.7.4 Continuous Change of Variable . . . . . . . . . . . . . . . . . 88

2 Differentiation 93
2.1 Linear Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.1.1 Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.1.2 Basic Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.1.3 Constant Approximation . . . . . . . . . . . . . . . . . . . . . 98
2.1.4 One Sided Derivative . . . . . . . . . . . . . . . . . . . . . . . 100
2.2 Property of Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2.2.1 Arithmetic Combination of Linear Approximation . . . . . . . 101
2.2.2 Composition of Linear Approximation . . . . . . . . . . . . . 102
2.2.3 Implicit Linear Approximation . . . . . . . . . . . . . . . . . . 109
2.3 Application of Linear Approximation . . . . . . . . . . . . . . . . . . 113
2.3.1 Monotone Property and Extrema . . . . . . . . . . . . . . . . 113
2.3.2 Detect the Monotone Property . . . . . . . . . . . . . . . . . 115
2.3.3 Compare Functions . . . . . . . . . . . . . . . . . . . . . . . . 118
2.3.4 First Derivative Test . . . . . . . . . . . . . . . . . . . . . . . 120
2.3.5 Optimization Problem . . . . . . . . . . . . . . . . . . . . . . 122
2.4 Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.4.1 Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . 125
2.4.2 Criterion for Constant Function . . . . . . . . . . . . . . . . . 127
2.4.3 L’Hospital’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . 129
2.5 High Order Approximation . . . . . . . . . . . . . . . . . . . . . . . . 133
2.5.1 Taylor Expansion . . . . . . . . . . . . . . . . . . . . . . . . . 136
2.5.2 High Order Approximation by Substitution . . . . . . . . . . . 139
2.5.3 Combination of High Order Approximations . . . . . . . . . . 144
2.5.4 Implicit High Order Differentiation . . . . . . . . . . . . . . . 148
2.5.5 Two Theoretical Examples . . . . . . . . . . . . . . . . . . . . 150
2.6 Application of High Order Approximation . . . . . . . . . . . . . . . 151
2.6.1 High Derivative Test . . . . . . . . . . . . . . . . . . . . . . . 151
2.6.2 Convex Function . . . . . . . . . . . . . . . . . . . . . . . . . 154
2.6.3 Sketch of Graph . . . . . . . . . . . . . . . . . . . . . . . . . . 158
2.7 Numerical Application . . . . . . . . . . . . . . . . . . . . . . . . . . 163
2.7.1 Remainder Formula . . . . . . . . . . . . . . . . . . . . . . . . 164
2.7.2 Newton’s Method . . . . . . . . . . . . . . . . . . . . . . . . . 166

3 Integration 171
3.1 Area and Definite Integral . . . . . . . . . . . . . . . . . . . . . . . . 171
3.1.1 Area below Non-negative Function . . . . . . . . . . . . . . . 171
3.1.2 Definite Integral of Continuous Function . . . . . . . . . . . . 174
3.1.3 Property of Area and Definite Integral . . . . . . . . . . . . . 178
3.2 Rigorous Definition of Integral . . . . . . . . . . . . . . . . . . . . . . 180
CONTENTS 5

3.2.1 What is Area? . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

3.2.2 Darboux Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
3.2.3 Riemann Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
3.3 Numerical Calculation of Integral . . . . . . . . . . . . . . . . . . . . 187
3.3.1 Left and Right Rule . . . . . . . . . . . . . . . . . . . . . . . 187
3.3.2 Midpoint Rule and Trapezoidal Rule . . . . . . . . . . . . . . 189
3.3.3 Simpson’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.4 Indefinite Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
3.4.1 Fundamental Theorem of Calculus . . . . . . . . . . . . . . . 194
3.4.2 Indefinite Integral . . . . . . . . . . . . . . . . . . . . . . . . . 198
3.5 Properties of Integration . . . . . . . . . . . . . . . . . . . . . . . . . 201
3.5.1 Linear Property . . . . . . . . . . . . . . . . . . . . . . . . . . 202
3.5.2 Integration by Parts . . . . . . . . . . . . . . . . . . . . . . . 207
3.5.3 Change of Variable . . . . . . . . . . . . . . . . . . . . . . . . 217
3.6 Integration of Rational Function . . . . . . . . . . . . . . . . . . . . . 231
3.6.1 Rational Function . . r . . . . . . . . . . . . . . . . . . . . . . . 231
n ax + b
3.6.2 Rational Function of . . . . . . . . . . . . . . . . . . 238
cx + d
3.6.3 Rational Function of sin x and cos x . . . . . . . . . . . . . . . 241
3.7 Improper Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
3.7.1 Definition and Property . . . . . . . . . . . . . . . . . . . . . 243
3.7.2 Comparison Test . . . . . . . . . . . . . . . . . . . . . . . . . 248
3.7.3 Conditional Convergence . . . . . . . . . . . . . . . . . . . . . 252
3.8 Application to Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 256
3.8.1 Length of Curve . . . . . . . . . . . . . . . . . . . . . . . . . . 256
3.8.2 Area of Region . . . . . . . . . . . . . . . . . . . . . . . . . . 261
3.8.3 Surface of Revolution . . . . . . . . . . . . . . . . . . . . . . . 267
3.8.4 Solid of Revolution . . . . . . . . . . . . . . . . . . . . . . . . 271
3.8.5 Cavalieri’s Principle . . . . . . . . . . . . . . . . . . . . . . . . 279
3.9 Polar Coordinate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
3.9.1 Curves in Polar Coordinate . . . . . . . . . . . . . . . . . . . 286
3.9.2 Geometry in Polar Coordinate . . . . . . . . . . . . . . . . . . 290
3.10 Application to Physics . . . . . . . . . . . . . . . . . . . . . . . . . . 294
3.10.1 Work and Pressure . . . . . . . . . . . . . . . . . . . . . . . . 294
3.10.2 Center of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . 297

4 Series 301
4.1 Series of Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
4.1.1 Sum of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
4.1.2 Convergence of Series . . . . . . . . . . . . . . . . . . . . . . . 304
4.2 Comparison Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
4.2.1 Integral Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
4.2.2 Comparison Test . . . . . . . . . . . . . . . . . . . . . . . . . 309
6 CONTENTS

4.2.3 Special Comparison Test . . . . . . . . . . . . . . . . . . . . . 313

4.3 Conditional Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 317
4.3.1 Test for Conditional Convergence . . . . . . . . . . . . . . . . 317
4.3.2 Absolute v.s. Conditional . . . . . . . . . . . . . . . . . . . . 321
4.4 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
4.4.1 Convergence of Taylor Series . . . . . . . . . . . . . . . . . . . 324
4.4.2 Radius of Convergence . . . . . . . . . . . . . . . . . . . . . . 326
4.4.3 Function Defined by Power Series . . . . . . . . . . . . . . . . 330
4.5 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
4.5.1 Fourier Coefficient . . . . . . . . . . . . . . . . . . . . . . . . 333
4.5.2 Complex Form of Fourier Series . . . . . . . . . . . . . . . . . 340
4.5.3 Derivative and Integration of Fourier Series . . . . . . . . . . . 342
4.5.4 Sum of Fourier Series . . . . . . . . . . . . . . . . . . . . . . . 345
4.5.5 Parseval’s Identity . . . . . . . . . . . . . . . . . . . . . . . . 347
4.5.6 Isoperimetric Inequality . . . . . . . . . . . . . . . . . . . . . 349
Chapter 1

Limit

1.1 Limit of Sequence

A sequence is an infinite list

x1 , x2 , . . . , xn , . . . .

The n-th term of the sequence is xn , and n ∈ Z is the index of the term. In this
course, we will always assume that all the terms xn ∈ R are real numbers. Here are
some examples

xn = n : 1, 2, 3, . . . , n, . . . ;
yn = 2 : 2, 2, 2, . . . , 2, . . . ;
1 1 1
zn = : 1, , . . . , , . . . ;
n 2 n
un = (−1)n : 1, −1, 1, . . . , (−1)n , . . . ;
vn = sin n : sin 1, sin 2, sin 3, . . . , sin n, . . . .

Note that the index does not have to start from 1. For example, the sequence
un actually starts from n = 0 (or any even integer). What we are interested is only
the behaviour when n is large. Moreover, a sequence does not have to be given by
a formula. For example, the decimal expansions of π give a sequence

wn : 3, 3.1, 3.14, 3.141, 3.1415, 3.14159, 3.141592, . . . .

If n is the number of digits after the decimal point, then the sequence wn starts at
n = 0. Alternatively, we can introduce the notation called the integral part of n:

[n] := biggest integer less than or equal to n.

Then we can express

[10n π]
wn = .
10n
7
8 CHAPTER 1. LIMIT

xn

wn

yn

vn
zn n
un

Figure 1.1.1: Sequences.

Now we look at the trend of the examples above as n gets bigger. We find that
xn gets bigger and can become as big as we want. On the other hand, yn remains
constant, zn gets smaller and can become as small as we want. This means that
yn approaches 2 and zn approaches 0. Moreover, un and vn jump around and do
not approach anything. Finally, wn is equal to π up to the n-th decimal place, and
therefore approaches π.

Definition 1.1.1 (Intuitive). If xn approaches a finite number l when n gets bigger

and bigger, then we say that the sequence xn converges to the limit l and write

lim xn = l.
n→∞

A sequence diverges if it does not approach a specific finite number when n gets
bigger.

The sequences yn , zn , wn converge respectively to 2, 0 and π. The sequences

xn , un , vn diverge. Since the limit describes the behavior when n gets very big, we
have the following property.

Proposition 1.1.2. If yn is obtained from xn by adding, deleting, or changing finitely

many terms, then limn→∞ xn = limn→∞ yn .

The equality in the proposition means that xn converges if and only if yn con-
verges. Moreover, the two limits have equal value when both converge.

1 1
Example 1.1.1. The sequence √ is obtained from √ by deleting the first
n+2 n
1 1
two terms. By limn→∞ √ = 0 and Proposition 1.1.2, we get limn→∞ √ =
n n
1
limn→∞ √ = 0.
n+2
1.1. LIMIT OF SEQUENCE 9

In general, we have limn→∞ xn+k = limn→∞ xn for any integer k.

1
The example assumes limn→∞ √ = 0, which is supposed to be intuitively obvi-
n
ous. Although mathematics is inspired by intuition, a critical feature of mathematics
is rigorous logic. This means that we need to be clear what basic facts are assumed
in any argument. For the moment, we will always assume that we already know
1
limn→∞ c = c and limn→∞ p = 0 for p > 0. After the two limits are rigorously
n
established in Examples 1.2.2 and 1.2.3, the conclusions based on the two limits
become solid.

1.1.1 Arithmetic Rule

Intuitively, if x is close to 3 and y is close to 5, then the arithmetic combinations

x+y and xy are close to 3+5 = 8 and 3·5 = 15. The intuition leads to the following
property of limit.

Proposition 1.1.3 (Arithmetic Rule). Suppose limn→∞ xn = l and limn→∞ yn = k.

Then

xn l
lim (xn ± yn ) = l ± k, lim cxn = cl, lim xn yn = kl, lim = ,
n→∞ n→∞ n→∞ n→∞ yn k

where c is a constant and k 6= 0 in the last equality.

The proposition says limn→∞ (xn + yn ) = limn→∞ xn + limn→∞ yn . However, the

equality is of different nature from the equality in Proposition 1.1.2, because the
convergence of the limits on two sides are not equivalent: If the two limits on the
right converge, then the limit on the left also converges and the two sides are equal.
However, for xn = (−1)n and yn = (−1)n+1 , the limit limn→∞ (xn + yn ) = 0 on the
left converges, but both limits on the right diverge.

Exercise 1.1.1. Explain that limn→∞ xn = l if and only if limn→∞ (xn − l) = 0.

Exercise 1.1.2. Suppose xn and yn converge. Explain that limn→∞ xn yn = 0 implies either
limn→∞ xn = 0 or limn→∞ yn = 0. Moreover, explain that the conclusion fails if xn and
yn are not assumed to converge.
10 CHAPTER 1. LIMIT

Example 1.1.2. We have

 
1 1
2+ limn→∞ 2 +
2n2 + n n n
lim = lim =
1 1
 
n→∞ n2 − n + 1 n→∞ 1 1
1− + 2 limn→∞ 1 − + 2
n n n n
1
limn→∞ 2 + limn→∞
= n
1 1 1
limn→∞ 1 − limn→∞ + limn→∞ · limn→∞
n n n
2+0
= = 2.
1−0+0·0
The arithmetic rule is used in the second and third equalities. The limits limn→∞ c =
1
c and limn→∞ = 0 are used in the fourth equality.
n

Exercise 1.1.3. Find the limits.

n+2 n3 + 4n2 − 2 (n2 + 1)(n + 2)
1. . 4. . 7. .
n−3 2n3 − n + 3 (n + 1)(n2 + 2)
n+2 (n + 1)(n + 2) (2 − n)3
2. . 5. . 8. .
n2 − 3 2n2 − 1 2n3 + 3n − 1

2n2 − 3n + 2 2n2 − 1 (n2 + 3)3

3. . 6. . 9. .
3n2 − 4n + 1 (n + 1)(n + 2) (n3 − 2)2

Exercise 1.1.4. Find the limits.

√ √ √ √
n+2 n+4 n−2
3
( n + 1)(n + 2)
1. √ . 4. √ . 7. √ .
n−3 23n−n+3 (n + 1)( n + 2)
√ √ √ √
n+2 ( n + 1)( n + 2) (2 − 3 n)3
2. . 5. . 8. √ .
n−3 2n − 1 2 3 n + 3n − 1
√ √
2 n − 3n + 2 2n − 1 ( 3 n + 3)3
3. √ . 6. √ √ . 9. √ .
3 n − 4n + 1 ( n + 1)( n + 2) ( n − 2)2

Exercise 1.1.5. Find the limits.

√
n+a n+a an3 + b
1. . 4. √ . 7. √ .
n+b n+b n+c (c n + d)6
√ √ √ √
n+a ( n + a)( n + b) (a 3 n + b)2
2. . 5. . 8. √ .
n+b cn + d (c n + d)3
√
n+a cn + d (a n + b)2
3. . 6. √ √ . 9. √ .
2
n + bn + c ( n + a)( n + b) (c 3 n + d)3
1.1. LIMIT OF SEQUENCE 11

Exercise 1.1.6. Show that


np
ap + ap−1 np−1 + · · · + a1 n + a0 0, if 0 < p < q,
lim = ap
n→∞ bq nq + bq−1 nq−1 + · · · + b1 n + b0  , if 0 < p = q and bq 6= 0.
bq

Exercise 1.1.7. Find the limits.

√
1010 n 55 (2n + 1)2 − 1010 55 (2 n + 1)2 − 1010
1. 2 . 2. . 3. .
n − 10 10n2 − 5 10n − 5

Exercise 1.1.8. Find the limits.

n n n+a n+c n3 + a n3 + c
1. − . 4. − . 7. − .
n+1 n−1 n+b n+d n2 + b n2 + d
n2 n2 n2 + a n2 + c n2 + a n2 + c
2. − . 5. − . 8. − .
n+1 n−1 n+b n+d n3 + b n3 + d
√ √
n n n+a n+c n+a n+c
3. √ −√ . 6. √ −√ . 9. √ −√ .
n+1 n−1 n+b n+d 3
n+b 3
n+d

Exercise 1.1.9. Find the limits.

n2 + a1 n + a0 n2 + c1 n + c0
 2  2
n+a n+c
1. − . 3. − .
n+b n+d n+b n+d
2 2
n2 + a n2 + c
 
n2 + a1 n + a0 n2 + c1 n + c0
2. 2 − 2 . 4. − .
n + b1 n + b0 n + d1 n + d0 n+b n+d

Exercise 1.1.10. Find the limits, p, q > 0.

np + a np + a np + c
1. . 3. − q .
nq + b nq + b n +d
anp + bnq + c n2p + a1 np + a2
2. . 4. .
anq + bnp + c n2q + b1 nq + b2

1.1.2 Sandwich Rule

The following property reflects the intuition that if x and z are close to 3, then
anything between x and z should also be close to 3.

Proposition 1.1.4 (Sandwich Rule). Suppose xn ≤ yn ≤ zn for sufficiently big n. If

limn→∞ xn = limn→∞ zn = l, then limn→∞ yn = l.

Sandwich Rule is also known as the Squeeze Theorem.

Note that something holds for sufficiently big n is the same as something fails
for only finitely many n.
12 CHAPTER 1. LIMIT

Example 1.1.3. By 2n − 3 > n for sufficiently big n (in fact, n > 3 is enough), we
have
1 1
0< √ <√ .
2n − 3 n
1 1
Then by limn→∞ 0 = limn→∞ √ = 0 and the sandwich rule, we get limn→∞ √ =
n 2n − 3
0.
n
On the other hand, for sufficiently big n, we have n + 1 < 2n and n − 1 > ,
2
and therefore √ √ √
n+1 2n 2 2
0< < n = √ .
n−1 n
2
√
2 2 √ 1
By limn→∞ √ = 2 2 limn→∞ √ = 0 (arithmetic rule used) and the sandwich
n √ n
n+1
rule, we get limn→∞ = 0.
n−1

Example 1.1.4. By −1 ≤ sin n ≤ 1, we have

1 sin n 1 1 1 1
− ≤ ≤ , √ ≤√ ≤√ .
n n n n+1 n + sin n n−1
1 1 sin n
By limn→∞ = 0, − limn→∞ = 0 and the sandwich rule, we get limn→∞ = 0.
n n n
1 1
Moreover, by limn→∞ √ = limn→∞ √ = 0 (see argument in Example
n+1 n−1
1
1.1.1) and the sandwich rule, we get limn→∞ √ = 0.
n + sin n

Exercise 1.1.11. Prove that limn→∞ |xn | = 0 implies limn→∞ xn = 0.

Exercise 1.1.12. Find the limits, a > 0.

√ √
1 2n + 3 1 an + b
1. √ . 2. . 3. √ . 4. .
3n − 4 4n − 1 an + b cn + d

Exercise 1.1.13. Find the limits.

cos n cos n cos n
1. . 4. √ . 7. p .
n n−2 n + (−1)n 2

(−1)n 1 cos n
2. . 5. . 8. p √ .
n n + (−1)n n + sin n

√ cos n (−1)n
sin n 6. . 9. p .
3. . n + (−1)n n + (−1)n
n
1.1. LIMIT OF SEQUENCE 13
√ √
2 + (−1)n 3 3 n+2 n + sin n
10. √ . 13. . 16. √ .
3
n2 − 2 cos n 2n + (−1)n 3 n − cos n
√
sin n + (−1)n cos n n sin n + cos n (−1)n (n + 1)
11. √ . 14. . 17. .
n + (−1)n n−1 n2 + (−1)n+1
√
| sin n + cos n| n + sin n (−1)n (n + 10)2 − 1010
12. . 15. . 18. .
n n + cos 2n 10(−1)n n2 − 5

Exercise 1.1.14. Find the limits.

√ √
n+a n + (−1)n a cos n + a
1. . 4. . 7. .
n + (−1)n b n + (−1)n b n + b sin n
√
1 (−1)n (an + b) cos n + a
2. √ . 5. . 8. √ .
3
n2 + an + b n2 + c(−1)n+1 n + d n + b sin n
√
n+c+d (−1)n (an + b)2 + c an + b sin n
3. √ . 6. . 9. .
3
n2 + an + b (−1)n n2 + d cn + d sin n

Exercise 1.1.15. Find the limits, p > 0.

√
sin n sin(n + 1) a sin n + b a cos(sin n)
1. . 2. p . 3. . 4. .
n p n + (−1)n np + c np − b sin n
√ √
Example 1.1.5. For a > 0, the sequence n + a − n satisfies
√ √ √ √
√ √ ( n + a − n)( n + a + n) a a
0< n+a− n= √ √ =√ √ <√ .
n+a+ n n+a+ n n
a √ √
By limn→∞ √ = 0 and the sandwich rule, we get limn→∞ ( n + a − n) = 0. Similar
n
argument also shows the limit for a < 0.
r
n+2
Example 1.1.6. The sequence satisfies
n
r
n+2 n+2 1
1< < =1+2 .
n n n
  r
1 n+2
By limn→∞ 1+2 = 1 + 2 · 0 = 1 and the sandwich rule, we get limn→∞ = 1.
n n
√ √
Exercise 1.1.16. Show that limn→∞ ( n + a − n) = 0 for a < 0.
r
n+2
Exercise 1.1.17. Use the idea of Example 1.1.5 to estimate − 1 and then find
r n
n+2
limn→∞ .
n
14 CHAPTER 1. LIMIT
r
n+a
Exercise 1.1.18. Show that limn→∞ = 1. You may need separate argument for
n+b
a > b and a < b.

Exercise 1.1.19. Find the limits.

√ √ r
n
1. n + a − n + b. 9. .
√ n2 +n+1
n+a
2. √ √ . r
n+c+ n+d n+a
√ √ 10. .
n2 + bn + c
n+a+ n+b
3. √ √ . √ √
n+c+ n+d 11. n2 + an + b − n2 + cn + d.
√ √
n+a n+b √ √ √ √
4. √ √ . 12. n + a n + b − n + c n + d.
n+c n+d
√ n
n+a+b 13. √ .
5. √ . 2
n +n+1
n+c+d
√ √ √ n+a
6. n( n + a − n + b). 14. √ .
√ √ √ n2 + bn + c
7. n + c( n + a − n + b). r
√ √ √ n2 + an + b
8. n + a + n + b − 2 n + c. 15. .
n2 + cn + d

Exercise 1.1.20. Find the limits.

√ √ p √ √
1. n + a sin n − n + b cos n. 5. n + (−1)n ( n + a − n + b).
√ √
r
n + a sin n
2. . 6. n2 + an + sin n − n2 + bn + cos n.
n + b cos n
r
n2 + an + sin n
s
n + (−1)n a 7. .
3. . n2 + bn + cos n
n + (−1)n b
√ √
n + a + sin n n2 + an + b
4. √ . 8. .
n + c + (−1)n n + (−1)n c

Exercise 1.1.21. Find the limits.

√ √
3
√
3 √ √
1. 3
n+a− n + b. 3. n2 ( 3 n + a − 3 n + b).
r
n+a √ p√ p√
2. 3 . 4. 3
n( 3 n + a − 3 n + b).
n+b

Exercise 1.1.22. Find the limits.

1.1. LIMIT OF SEQUENCE 15
 5  5.4 −√2 
n+a
p
n−2 n−2

1. . 2. . n−2 4. .
n+1 n+1 3. . n+b
n+1

Exercise 1.1.23. Find the limits.

√ p  2 p  p
n + a sin n n + an + b n+a
1. √ . 2. . 3. 2
.
n + b cos 2n n2 + (−1)n c n + bn + c

Exercise 1.1.24. Suppose limn→∞ xn = 1. Use the arithmetic rule and the sandwich rule
to prove that, if xn ≤ 1, then limn→∞ xpn = 1. Of course we expect the condition xn ≤ 1
to be unnecessary. See Example 1.1.21.

1.1.3 Some Basic Limits

1
Using limn→∞ = 0 and the sandwich rule, we may establish some basic limits.
n

Example 1.1.7. We show that

√
n
lim a = 1, for a > 0.
n→∞
√
First assume a ≥ 1. Then xn = n
a − 1 ≥ 0, and
n(n − 1) 2
a = (1 + xn )n = 1 + nxn + xn + · · · + xnn > nxn .
2
This implies
a
0 ≤ xn < .
n
a
By the sandwich rule and limn→∞ = 0, we get limn→∞ xn = 0. Then by the
n
arithmetic rule, this further implies
√
lim n a = lim (xn + 1) = lim xn + 1 = 1.
n→∞ n→∞ n→∞

1
For the case 0 < a ≤ 1, let b = ≥ 1. Then by the arithmetic rule,
a
√ 1 1 1
lim n a = lim √n
= √
n
= = 1.
n→∞ n→∞ b limn→∞ b 1

Example 1.1.8. Example 1.1.7 can be extended to

√
lim n n = 1.
n→∞
√
Let xn = n n − 1. Then we have xn > 0 for sufficiently large n (in fact, n ≥ 2 is
enough), and
n(n − 1) 2 n(n − 1) 2
n = (1 + xn )n = 1 + nxn + xn + · · · + xnn > xn .
2 2
16 CHAPTER 1. LIMIT

This implies √
2
0 ≤ xn < √ .
n−1
√
2
By limn→∞ √ = 0 (see Example 1.1.1 or 1.1.3) and the sandwich rule, we get
n−1
limn→∞ xn = 0. This further implies
√
lim n n = lim xn + 1 = 1.
n→∞ n→∞

Example 1.1.9. The following “n-th root type” limits can be compared with the
limits in Examples 1.1.7 and 1.1.8
√ √
n
√
n √
1 < n n + 1 < 2n = 2 n n,
1 √
1 < n n+1 < n n,
n n √
1 < (n2 − n) n2 −1 < (n2 ) n2 /2 = ( n n)4 .

By Examples 1.1.7, 1.1.8 and the arithmetic rule, the sequences on the right converge
to 1. Then by the sandwich rule, we get
√ 1 n
lim n n + 1 = lim n n+1 = lim (n2 − n) n2 −1 = 1.
n→∞ n→∞ n→∞

Example 1.1.10. We have

√
n
√ √ √
n
3= 3n < n 2n + 3n < n 3n + 3n = 3 2.
√
n
By Example
√ 1.1.7, we have lim n→∞ 3 2 = 3. Then by the sandwich rule, we get
limn→∞ n 2n + 3n = 3.
For another example, we have

3n > 3n − 2n = 3n−1 + 2 · 3n−1 − 2 · 2n−1 > 3n−1 .

Taking the n-th root, we get

√
n 1
3> 3n − 2n > 3 √
n
.
3
1 √
By limn→∞ 3 √
n
= 3 and the sandwich rule, we get limn→∞
n
3n − 2n = 3.
3

Exercise 1.1.25. Prove that if a ≤ xn ≤ b for some constants a, b > 0 and sufficiently big
√
n, then limn→∞ n xn = 1.

Exercise 1.1.26. Find the limits, a > 0.

1.1. LIMIT OF SEQUENCE 17

1 c c
1. n 2n . 5. (an + b) n . 9. (an + b) n+d .
2 c cn
2. n n . 6. (an2 + b) n . 10. (an + b) n2 +dn+e .
c √ c c
3. n n . 7. ( n + 1) n . 11. (an2 + b) n+d .
c 1 cn+d
4. (n + 1) n . 8. (n − 2) n+3 . 12. (an2 + b) n2 +en+f .

Exercise 1.1.27. Find the limits, a > 0.

√ p 1
1. n n + sin n. 3. n n + (−1)n sin n. 5. (n − cos n) n+sin n .
√ p n
2. n an + b sin n. 4. n an + (−1)n b sin n. 6. (an + b sin n) n2 +c cos n .

Exercise 1.1.28. Find the limits, p, q > 0.

√ √ √ √
1. n np + sin n. 2. n np + nq . 3. n+2
np + nq . 4. n−2
np + nq .

Exercise 1.1.29. Find the limits.

√ √ 1
1. n 5n − 4n . 4. n 5n − 3 · 4n − 2n . 7. (5n − 4n ) n+1 .
√ √ 1
2. n
5n − 3 · 4n . 5. n
42n−1 − 5n . 8. (5n − 4n ) n−2 .
√ p n+1
3. n
5n − 3 · 4n + 2n . 6. n
42n−1 + (−1)n 5n . 9. (5n − 4n ) n2 +1 .

Exercise 1.1.30. Find the limits, a > b > 0.

√ √
n n 2n+1
n
1. n an + bn . 4. a b . 7. (an + bn ) n2 −1 .
√ √ n
2. n an − bn . 5. n+2 an + bn . 8. (an − bn ) n2 −1 .
p √ n
3. n an + (−1)n bn . 6. n−2 an − bn . 9. (an − (−1)n bn ) n2 −1 .
√
n
Exercise 1.1.31. For a, b, c > 0, find limn→∞ an + bn + cn .
√
Exercise 1.1.32. For a ≥ 1, prove limn→∞ n a = 1 by using
√ √ √ √
a − 1 = ( n a − 1) ( n a)n−1 + ( n a)n−2 + · · · + n a + 1 .

Example 1.1.11. We show that

lim an = 0, for |a| < 1.
n→∞
1
First assume 0 < a < 1 and write a = . Then b > 0 and
1+b
1 1 1
0 < an = n
= < .
(1 + b) n(n − 1) 2 n nb
1 + nb + b + ··· + b
2
18 CHAPTER 1. LIMIT

1
By limn→∞ = 0 and the sandwich rule, we get limn→∞ an = 0.
nb
If −1 < a < 0, then 0 < |a| < 1 and limn→∞ |an | = limn→∞ |a|n = 0. By Exercise
1.1.11, we get limn→∞ an = 0.

Example 1.1.12. Example 1.1.11 can be extended to

lim nan = 0, for |a| < 1.

n→∞

This follows from

n n n 2
0 < nan = = < = ,
(1 + b)n n(n − 1) 2 n
n(n − 1) 2 (n − 1)b2
1 + nb + b + ··· + b b
2 2
2
the limit limn→∞ = 0 and the sandwich rule.
(n − 1)b2
Exercise 1.1.34 gives further extension of the limit.

Example 1.1.13. We show that

an
lim = 0, for any a
n→∞ n!

for the special case a = 4. For n > 4, we have

4n 4·4·4·4 4 4 4 4·4·4·4 4 45 1
0< = · · ··· ≤ · = .
n! 1·2·3·4 5 6 n 1·2·3·4 n 4! n
45 1 4n
By limn→∞ = 0 and the sandwich rule, we get limn→∞ = 0.
4! n n!
Exercise 1.1.35 suggests how to show the limit in general.

Exercise 1.1.33. Show that limn→∞ n2 an = 0 for |a| < 1 in two ways. The first is by using
the ideas from Examples 1.1.11 and 1.1.12. The second is by using limn→∞ nan = 0 for
|a| < 1.

Exercise 1.1.34. Show that limn→∞ n5.4 an = 0 for |a| < 1. What about limn→∞ n−5.4 an ?
What about limn→∞ np an ?

an
Exercise 1.1.35. Show that limn→∞ = 0 for a = 5.4 and a = −5.4.
n!

an
Exercise 1.1.36. Show that limn→∞ √ = 0 for a = 5.4 and a = −5.4.
n!

n!an
Exercise 1.1.37. Show that limn→∞ = 0 for any a.
(2n)!

Exercise 1.1.38. Find the limits.

1.1. LIMIT OF SEQUENCE 19

n+1 3. n99 0.99n . n + 2n n3n

1. . 5. . 7. .
2n 3n (1 + 2n )2

n2 (n2 + 1)1001 n2n + (−3)n 5n − n6n+1

2. . 4. . 6. . 8. .
2n 1.001n−2 4n 32n−1 − 23n+1

Exercise 1.1.39. Find the limits. Some convergence depends on a and p. You may try
some special values of a and p first.

1. np an . 1 np np an
4. . 7. . 10. √ .
np an n!an 3
n!
an np np n!an
2. . 5. . 8. √ . 11. .
np n! n! (2n)!

np np an np an n!np an
3. n . 6. . 9. √ . 12. .
a n! n! (2n)!

Exercise 1.1.40. Find the limits.

n2 + 3n + 5n n2 + n3n + 5! n2 + n! + (n − 1)!
1. . 3. . 5. .
n! n! 3n − n! + (n − 1)!

n2 + 3n + 5n n2 3n+5 + 5 · (n − 1)! 2n n! + 3n (n − 1)!

2. . 4. . 6. .
n! − n2 + 2n (n + 1)! 4n (2n − 1)! − 5n n!
√
r
n n 1
Exercise 1.1.41. Prove n! > . Then use this to prove limn→∞ √
n
= 0.
2 n!

n! 1 (n!)2 1
Exercise 1.1.42. Prove n
< and < for n > 2. Then use this to prove
n n (2n)! n+1
n! (n!)2 (n!)k
limn→∞ = limn→∞ = 0. What about limn→∞ where k ≥ 2 is an integer?
nn (2n)! (kn)!

1.1.4 Order Rule

The following property reflects the intuition that bigger sequence should have bigger
limit.

Proposition 1.1.5 (Order Rule). Suppose limn→∞ xn = k and limn→∞ yn = l.

1. If xn ≤ yn for sufficiently big n, then k ≤ l.

2. If k < l, then xn < yn for sufficiently big n.

By taking yn = l, we get the following special cases of the property for a con-
verging sequence xn .
1. If xn ≤ l for sufficiently big n, then limn→∞ xn ≤ l.
20 CHAPTER 1. LIMIT

2. If limn→∞ xn < l, then xn < l for sufficiently big n.

Similar statements with reversed inequalities also hold (see Exercise 1.1.43).
Note the non-strict inequality in the first statement of Proposition 1.1.5 and the
1 1
strict inequality the second statement. For example, we have xn = 2 < yn = ,
n n
but limn→∞ xn 6< limn→∞ yn . The example also satisfies limn→∞ xn ≥ limn→∞ yn
but xn 6≥ yn , even for sufficiently big n.

Exercise 1.1.43. Explain how to get the following special cases of the order rule.
1. If xn ≥ l for sufficiently big n, then limn→∞ xn ≥ l.

2. If limn→∞ xn > l, then xn > l for sufficiently big n.

2n2 + n
Example 1.1.14. By limn→∞ = 2 and the order rule, we know 1 <
n2 − n + 1 r
2n2 + n n 2n2 + n √
n
2
< 3 for sufficiently big n. This implies 1 < 2
< 3 for
n −n+1 √ n −n+1
sufficiently big n. By limn→∞ n 3 = 1 and the sandwich rule, we get
s
n 2n2 + n
lim = 1.
n→∞ n2 − n + 1
√
Example 1.1.15. We showed limn→∞ n 3n − 2n = 3 in Example 1.1.10. Here we use
a different method, with
s thehelp of the order rule. s
n  n
√n n n
2 2
By 3 − 2 = 3 1 − n
, we only need to find limn→∞ 1 −
n
. By
3 3
  n 
2
Example 1.1.11, we have limn→∞ 1 − = 1. By the order rule, therefore,
3
we have  n
1 2
<1− <2
2 3
for sufficiently big n. This implies that
s  n
1 n 2 √
n
√
n
< 1 − < 2
2 3

for sufficiently
s big n. Then by Example 1.1.7 and the sandwich rule, we get
 n
2
limn→∞ n 1 − = 1, and we conclude that
3
s  n
√n n 2
n n
lim 3 − 2 = 3 lim 1− = 3.
n→∞ n→∞ 3
1.1. LIMIT OF SEQUENCE 21

√
Exercise 1.1.44. Prove that if limn→∞ xn = l > 0, then limn→∞ n xn = 1. Moreover, find
√
a sequence satisfying limn→∞ xn = 0 and limn→∞ n xn = 1. Can we have xn converging
√
to 0 and n xn converging to 0.32?

Exercise 1.1.45. Find the limits.

√ √ n−1
1. n
5n − n4n . 3. n−2
5n − n4n . 5. (5n − n4n ) n2 +1 .
√ p n+(−1)n
2. n+2
5n − n4n . 4. n
5n − (−1)n n4n . 6. (5n −(−1)n n4n ) n2 +1 .

Exercise 1.1.46. Find the limits.

r n−1 r
1 4. (n2 42n−1 − 5n ) n2 +1 . n − 1 2n−1
1. n 5n − n4n . 7. 23n + 2
n−2
3 .
n n +1
r  n−1
r 3n−1 
32n−1 n2
n 1 n
n
5. n2 n + 4n+1 + . 3n
2. 5 − (−1)n n4n . n 8. n2 + .
n n2
1
32n−1 n2
r r 
n+2 1 n n−2 32n−1 3n
9. n2 + .
3. 5 − n4n sin n. 6. n23n + .
n n2 n2

Exercise 1.1.47. Find the limits, a > b.

√ √ n+1
1. n an+1 + bn . 6. n
a + bn . 10. (an + bn ) n2 +1 .
p √
2. n an+1 + (−1)n bn . 7. n
an + bn .
n

p 11. ((n + 1)an + bn ) n2 −1 .

3. n−2 an+1 + (−1)n bn . p
8. n−2 nan + (n2 + 1)bn . 1
√ 12. (an + bn ) n2 −1 .
4. n 4an − 5bn . r
√ n+2 1 n (−1)n
5. n 4an + 5b2n+1 . 9. a + nbn+1 . 13. (an + (−1)n bn ) n2 −1 .
n

Exercise 1.1.48. Find the limits, a, b, c > 0.

√ p
1. n n2 an + nbn + 2cn . 3. n
(n + sin n)an + bn + n2 cn .
p p
2. n an (bn + 1) + ncn . 4. n an (n + bn (1 + ncn )).

Exercise 1.1.49. Suppose a polynomial p(n) = ap np + an−1 np−1 + · · · + a1 n + a0 has leading

coefficient ap > 0. Prove that p(n) > 0 for sufficiently big n.

Exercise 1.1.50. Suppose a,p b, c > 0, and p, q, r are polynomials with positive leading coef-
ficients. Find the limit of n p(n)an + q(n)bn + r(n)cn .

Exercise 1.1.51. Find the limits, a, b, p, q > 0.

22 CHAPTER 1. LIMIT
√
n
√
n+2
√
n2
1. anp + b sin n. 3. anp + bnq . 5. anp + bnq .
√ √ 1
2. n
anp + bnq . 4. n−2
anp + bnq . 6. (anp + bnq ) n2 −1 .

3n (n!)2
Example 1.1.16. The sequence xn = satisfies
(2n)!

xn 3n2 3
lim = lim = = 0.75.
n→∞ xn−1 n→∞ 2n(2n − 1) 4
xn
By the order rule, we have < 0.8 for sufficiently big n, say for n > N (in fact,
xn−1
N = 8 is enough). Then for n > N , we have
xn xn−1 xN +1
0 < xn = ··· xN < 0.8n−N xN = C · 0.8n , C = 0.8−N xN .
xn−1 xn−2 xN
By Example 1.1.11, we have limn→∞ 0.8n = 0. Since C is a constant, by the sandwich
rule, we get limn→∞ xn = 0.
Exercises 1.1.52 and 1.1.53 summarise the idea of the example.
 
 xn 
Exercise 1.1.52. Prove that if 
  ≤ c for a constant c < 1, then xn converges to 0.
xn−1 

xn
Exercise 1.1.53. Prove that if limn→∞ = l and |l| < 1, then xn converges to 0.
xn−1

Exercise 1.1.54. Find a such that the sequence converges to 0, p, q > 0.

n2 nq an
p
(2n)! n (2n)! a
1. a . 4. n
a . 7. √ . 10. .
(n!)2 n! n! (n!)p

(n!)2 n √ 2
(n!)p n
2. a . 5. n!an . 8. (n!)p an . 11. a .
(3n)! ((2n)!)q

(n!)3 n an
2
an n5 (n!)p n
3. a . 6. . 9. . 12. a .
(3n)! n! (n!)p ((2n)!)q

1.1.5 Subsequence
A subsequence is obtained by choosing infinitely many terms from a sequence. We
denote a subsequence by

xnk : xn1 , xn2 , . . . , xnk , . . . ,

where the indices satisfy

n1 < n2 < · · · < nk < · · · .

1.1. LIMIT OF SEQUENCE 23

The following are some examples

x2k : x2 , x4 , x6 , x8 , . . . , x2k , . . . ,
x2k−1 : x1 , x3 , x5 , x7 , . . . , x2k−1 , . . . ,
x2k : x2 , x4 , x8 , x16 , . . . , x2k , . . . ,
xk! : x1 , x2 , x6 , x24 , . . . , xk! , . . . .

If xn starts at n = 1, then n1 ≥ 1, which further implies nk ≥ k for all k.

Proposition 1.1.6. If a sequence converges to l, then any subsequence converges

to l. Conversely, if a sequence is the union of finitely many subsequences that all
converge to the same limit l, then the whole sequence converges to l.

1 1 1 1
Example 1.1.17. Since 2
is a subsequence of , limn→∞ = 0 implies limn→∞ 2 =
n n n n
1 1
0. We also know limn→∞ √ = 0 implies limn→∞ = 0 but not vice versa.
n n

n + (−1)n 3
Example 1.1.18. The sequence is the union of the odd subsequence
n − (−1)n 2
(2k − 1) − 3 2k − 4 2k + 3
= and the even subsequence . Both subsequences con-
(2k − 1) + 2 2k + 1 2k − 2
verge to 1, either by direct computation, or by regarding them also as subsequences
n−4 n+3 n + (−1)n 3
of and , which converge to 1. Then we conclude limn→∞ = 1.
n+1 n−2 n − (−1)n 2

Example 1.1.19. The sequence (−1)n has one subsequence (−1)2k = 1 converging to
1 and another subsequence (−1)2k−1 = −1 converging to −1. Since the two limits
are different, by Proposition 1.1.6, the sequence (−1)n diverges.

Example 1.1.20. The sequence sin na converges to 0 when a is an integer multiple of

π. Now assume 0 < a < π. For any natural number k, the interval [kπ, (k + 1)π]
of length π contains the following interval of length a (both intervals have the same
middle point)  
π−a π−a
[ak , bk ] = kπ + , (k + 1)π − .
2 2
 
π−a a
For even k, we have sin x ≥ sin = cos > 0 on [ak , bk ]. For odd k, we
2 2
a
have sin x ≤ − cos < 0 on [ak , bk ].
2
Since the arithmetic sequence a, 2a, 3a, . . . has increment a, which is the length
of [ak , bk ], we must have nk a ∈ [ak , bk ] for some natural number nk . Then sin n2k a is a
a
subsequence of sin na satisfying sin n2k ≥ cos > 0, and sin n2k+1 a is a subsequence
2
24 CHAPTER 1. LIMIT

a
satisfying sin n2k ≤ − cos . Therefore the two subsequences cannot converge to the
2
same limit. As a result, the sequence sin na diverges.
Now for general a that is not an integer multiple of π, we have a = 2N π ± b for
an integer N and b satisfying 0 < b < π. Then we have sin na = ± sin nb. We have
shown that sin nb diverges, so that sin na diverges.
We conclude that sin na converges if and only if a is an integer multiple of π.

Exercise 1.1.55. Find the limit.

√ √ 1 1
1. n! + 1 − n! − 1. 3. ((n + 1)!) n! . 5. (n!) (n+1)! .
1 1 2 −1 2 1
2. (n!) n! . 4. ((n + (−1)n )!) n! . 6. (2n + 3 n ) n2 .

Exercise 1.1.56. Explain convergence or divergence.

n √ p √  nπ nπ
1. 2(−1) . 6. n n + (−1)n − n . 11. sin cos .
2 3
(−1)n
2. n n . n 1 nπ
7. (2(−1) n + 3n ) n . n sin
n 3 .
3. n(−1) . n (−1) nn 1
12. nπ
8. (2 + 3 )n . n cos +2
(−1)n n + 3 2
4. . nπ
n − (−1)n 2 9. tan . nπ
3 n − sin
13. 3 .
(−1)n n2 nπ nπ
5. . 10. (−1)n sin . n + 2 cos
n3 − 1 3 2

Exercise 1.1.57. Find all a such that the sequence cos na converges.

Example 1.1.21. We prove that limn→∞ xn = 1 implies limn→∞ xpn = 1. Exercises

1.1.58 and 1.1.59 extend the result.
The sequence xn is the union of two subsequences x0k and x00k (short for xmk and
xnk ) satisfying all x0k ≥ 1 and all x00k ≤ 1. By Proposition 1.1.6, the assumption
limn→∞ xn = 1 implies that limk→∞ x0k = limk→∞ x00k = 1.
Pick integers M and N satisfying M < p < N . Then x0k ≥ 1 implies x0k M ≤
x0k ≤ x0k N . By the arithmetic rule, we have limk→∞ x0k M = (limk→∞ x0k )M = 1M = 1
p

and similarly limk→∞ x0k N = 1. Then by the sandwich rule, we get limk→∞ x0k p = 1.
Similar proof shows that limk→∞ x00k p = 1. Since the sequence xpn is the union of
two subsequences x0k p and x00k p , by Proposition 1.1.6 again, we get limn→∞ xpn = 1.

Exercise 1.1.58. Suppose limn→∞ xn = 1 and yn is bounded. Prove that limn→∞ xynn = 1.

Exercise 1.1.59. Suppose limn→∞ xn = l > 0. By applying Example 1.1.21 to the sequence
xn
, prove that limn→∞ xpn = lp .
l
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 25

1.2 Rigorous Definition of Sequence Limit

1 1
The statement limn→∞ = 0 means that gets smaller and smaller as n gets
n n
bigger and bigger. To make the statement rigorous, we need to be more specific
about smaller and bigger.
Is 1000 big? The answer depends on the context. A village of 1000 people is
big, and a city of 1000 people is small (even tiny). Similarly, a rope of diameter less
than one millimeter is considered thin. But the hair is considered thin only if the
diameter is less than 0.05 millimeter.
So big or small makes sense only when compared with some reference quantity.
We say n is in the thousands if n > 1000 and in the millions if n > 10000000.
The reference quantities 1000 and 1000000 give a sense of the scale of bigness. In
1
this spirit, the statement limn→∞ = 0 means the following list of infinitely many
n
implications
 
1 
n > 1 =⇒  − 0 < 1,
n
 
1 
n > 10 =⇒  − 0 < 0.1,

n
 
1 
n > 100 =⇒  − 0 < 0.01,

n
..
.
 
1 
n > 1000000 =⇒  − 0 < 0.000001,

n
..
.

2n
For another example, limn→∞ = 0 means the following implications
n!
 n 
2 
n > 10 =⇒  − 0 < 0.0003,
n!
 n 
2 
n > 20 =⇒  − 0 < 0.0000000000005,
n!
..
.

So the general shape of the implications is

n > N =⇒ |xn − l| < .

Note that the relation between N (measuring the bigness of n) and  (measuring
the smallness of |xn − l|) may be different for different limits.
26 CHAPTER 1. LIMIT

The problem with infinitely many implications is that our language is finite.
In practice, we cannot verify all the implications one by one. Even if we have
verified the truth of the first one million implications, there is no guarantee that
the one million and the first implication is true. To mathematically establish the
truth of all implications, we have to formulate one finite statement that includes the
consideration for all N and all .

1.2.1 Rigorous Definition

Definition 1.2.1 (Rigorous). A sequence xn converges to a finite number l, and
denoted limn→∞ xn = l, if for any  > 0, there is N , such that n > N implies
|xn − l| < .

In case N is a natural number (which can always be arranged if needed), the

definition means that, for any given horizontal -band around l, we can find N , such
that all the terms xN +1 , xN +2 , xN +3 , . . . after N lie in the shaded area in Figure
1.2.1.

xN +2 xN +3 l+
x4
x2 xN +1 xn x l
n+1
x3 l−
x1

n
N

Figure 1.2.1: n > N implies |xn − l| < .

1
Example 1.2.1. For any  > 0, choose N = . Then

 
1  1 1
n > N =⇒  − 0 = <
 = .
n n N
1
This verifies the rigorous definition of limn→∞ = 0.
n
By applying the rigorous definition to  = 0.1, 0.01, . . . , we recover the infinitely
many implications we wish to achieve. This justifies the rigorous definition of limit.

Example 1.2.2. For the constant sequence xn = c, we rigorously prove

lim c = c.
n→∞

For any  > 0, choose N = 0. Then

n > 0 =⇒ |xn − c| = |c − c| = 0 < .
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 27

In fact, the right side is always true, regardless of the left side.

Example 1.2.3. We rigorously prove

1
lim = 0, for p > 0.
n→∞ np

1
For any  > 0, choose N = 1 . Then
p
 
1  1 1
n > N =⇒  p − 0 = p < p = .

n n N

We need to be more specific on the logical foundation for the arguments. We

will assume the basic knowledge of real numbers, which are the four arithmetic
x
operations x + y, x − y, xy, , the exponential operation xy (for x > 0), the order
y
x < y (or y > x, and x ≤ y means x < y or x = y), and the properties for these
1 1
operations. For example, we assume that we already know x > y > 0 implies <
x y
and xp > y p for p > 0. These properties are used in the example above.
More important about the knowledge assumed above is the knowledge that are
not assumed and therefore cannot be used until after the knowledge is established.
In particular, we do not assume any knowledge about the logarithm. The logarithm
and its properties will be rigorously established in Example 1.7.15 as the inverse of
exponential.

n2 − 1
Example 1.2.4. To rigorously prove limn→∞ = 1, for any  > 0, we have
n2 + 1
r  2 
2 n − 1  2 2 2
n>N = − 1 =⇒  2 − 1 = 2 < 2 =  = .
 n +1 n +1 N +1 2
−1 +1

Therefore the sequence converges
r to 1.  2 
2 n − 1 
How did we choose N = − 1? We want to achieve  2
 − 1 < . Since
 n +r
1
2 2
this is equivalent to 2 < , which we can solve to get n > − 1, choosing
r n +1 
2
N= − 1 should work.


Example 1.2.5. To rigorously prove the limit in Example 1.1.5, we estimate the
difference between the sequence and the expected limit
√ √ (n + 2) − n 2
| n + 2 − n − 0| = √ √ <√ .
n+2+ n n
28 CHAPTER 1. LIMIT

2 4
This shows that for any  > 0, it is sufficient to have √ < , or n > 2 . In other
n 
4
words, we should choose N = 2 .

The discussion above is the analysis of the problem, which you may write on
your scratch paper. The formal rigorous argument you are supposed to present is
4
the following: For any  > 0, choose N = 2 . Then

√ √ 2 2 2
n > N =⇒ | n + 2 − n − 0| = √ √ < √ < √ = .
n+2+ n n N

1.2.2 The Art of Estimation

 2 
n − 1 
In Examples 1.2.4, the formula for N is obtained by solving  2 − 1 <  in
n +1
exact way. However, this may not be so easy in general. For example, for the limit
in Example 1.1.2, we need to solve
 2n2 + n
 
 = 3n − 2 < .


 n2 − n + 1 − 2 n2 − n + 1

While the exact solution can be found, the formula for N is rather complicated. For
more complicated example, it may not even be possible to find the formula for the
exact solution.
We note that finding the exact solution of |xn − l| <  is the same as finding
N = N (), such that
n > N ⇐⇒ |xn − l| < .
However, in order to rigorously prove the limit, only =⇒ direction is needed. The
weaker goal can often be achieved in much simpler way.

Example 1.2.6. Consider the limit in Example 1.1.2. For n > 1, we have
 2n2 + n
 
 3n − 2 3n 3
 n2 − n + 1 − 2 = n2 − n + 1 < n2 − n = n − 1 .
 

 2n2 + n
 
3  3
Since <  implies  2 − 2 < , and <  is equivalent to
n−1 n −n+1 n−1
3 3
n > + 1, we find that choosing N = + 1 is sufficient
 
 2n2 + n
 
3  3n − 2 3 3
n > N = + 1 =⇒  2  − 2 = 2 < < = .
 n −n+1 n −n+1 n−1 N −1
 2
n2 − 1

n − 1  2
Exercise 1.2.1. Show that  2
 − 1 < and then rigorously prove limn→∞ 2 = 1.
n +1 n n +1
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 29

The key for the rigorous proof of limits is to find a simple and good enough
estimation. We emphasize that there is no need to find the best estimation. Any
estimation that can fulfill the rigorous definition of limit is good enough.
Everyday life is full of good enough estimations. Mastering the art of such
estimations is very useful for not just learning calculus, but also for making smart
judgement in real life.

Example 1.2.7. If a bottle is 20% bigger in size than another bottle, how much bigger
is in volume?
The exact formula is the cube of the comparison in size
(1 + 0.2)3 = 1 + 3 · 0.2 + 3 · 0.22 + 0.23 .
Since 3 · 0.2 = 0.6, 3 · 0.22 = 0.12, and 0.23 is much smaller than 0.1, the bottle is a
little more than 72% bigger in volume.

Example 1.2.8. The 2013 GDP per capita is 9,800USD for China and 53,100USD for
the United States, in terms of PPP (purchasing power parity). The percentage of
the annual GDP growth for the three years up to 2013 are 9.3, 7.7, 7.7 for China
and 1.8, 2.8, 1.9 for the United States. What do we expect the number of years for
China to catch up to the United States?
First we need to estimate how much faster is the Chinese GDP growing compared
to the United States. The comparison for 2013 is
1 + 0.077
≈ 1 + (0.077 − 0.019) = 1 + 0.058.
1 + 0.019
Similarly, we get the (approximate) comparisons 1 + 0.075 and 1 + 0.049 for the
other two years. Among the three comparisons, we may choose a more conservative
1 + 0.05. This means that we assume Chinese GDP per capita grows 5% faster than
the United States for the next many years.
Based on the assumption of 5%, the number of years n for China to catch up to
the United States is obtained exactly by solving
n(n − 1) 53, 100
(1 + 0.05)n = 1 + n 0.05 + 0.052 + · · · + 0.05n = ≈ 5.5.
2 9, 800
5.5 − 1
If we use 1 + n 0.05 to approximate (1 + 0.05)n , then we get n ≈ = 90.
0.05
n(n − 1)
However, 90 years is too pessimistic because for n = 90, the third term 0.052
2 n
is quite sizable, so that 1 + n 0.05 is not a good approximation of (1 + 0.05) .
An an exercise for the art of estimation, we try to avoid using calculator in
53, 100
getting better estimation. By ≈ 2.32 , we may solve
9, 800
m(m − 1)
n = 2m, (1 + 0.05)m = 1 + m 0.05 + 0.052 + · · · + 0.05m ≈ 2.3.
2
30 CHAPTER 1. LIMIT

2.3 − 1 m(m − 1)
We get n ≈ 2 · = 52. Since 0.052 is still sizable for m = 26 (but
0.05 2
giving much better approximation than n = 90), the actual n should be somewhat
smaller than 52. We try n = 40 and estimate (1 + 0.05)n by the first three terms
40 · 39
(1 + 0.05)40 ≈ 1 + 40 · 0.05 + 0.052 ≈ 5.
2
So it looks like somewhere between 40 and 45 is a good estimation.
We conclude that, if Chinese GDP per capita growth is 5% (a very optimistic
assumption) faster than the United States in the next 50 years, then China will
catch up to the United States in 40 some years.

Exercise 1.2.2. I wish to paint a wall measuring 3 meters tall and 6 meters wide, give or
take 10% in each direction. If the cost of paint is $13.5 per square meters, how much
should I pay for the paint?

Exercise 1.2.3. In a supermarket, I bought four items at $5.95, $6.35, $15.50, $7.20. The
sales tax is 8%. The final bill is around $38. Is the bill correct?

Exercise 1.2.4. In 1900, Argentina and Canada had the same GDP per capita. In 2000,
the GDP per capita is 9,300USD for Argentina and 24,000USD for Canada. On average,
how much faster is Canadian GDP growing annually compared with Argentina in the 20th
century?

Next we leave real life estimations and try some examples in calculus.

Example 1.2.9. If x is close to 3 and y is close to 5, then 2x−3y is close to 2·3−3·5 =

−9. We wish to be more precise about the statement, say, we want to find a tolerance
for x and y, such that 2x − 3y is within ±0.2 of −9.
We have
|(2x − 3y) − (−9)| = |2(x − 3) − 3(y − 5)| ≤ 2|x − 3| + 3|y − 5|.
For the difference to be within ±0.2, we only need to make sure 2|x − 3| + 3|y − 5| <
0.2
0.2. This can be easily achieved by |x − 3| < 2+3 = 0.04 and |y − 5| < 0.04.

Example 1.2.10. Again we assume x and y are close to 3 and 5. Now we want to
find the percentage of tolerance, such that 2x − 3y is within ±0.2 of −9.
We can certainly use the answer in Example 1.2.9 and find the percentage 0.04
3
≈
0.04
1.33% for x and 5 ≈ 0.8% for y. This implies that, if both x and y are within
0.8% of 3 and 5, then 2x − 3y is within ±0.2 of −9.
The better (or more honest) way is to directly solve the problem. Let δ1 and δ2
be the percentage of tolerance for x and y. Then x = 3(1 + δ1 ) and y = 5(1 + δ2 ),
and
|(2x−3y)−(−9)| = |2(x−3)−3(y−5)| ≤ |2·3δ1 −3·5δ2 | ≤ 21δ, δ = max{|δ1 |, |δ2 |}.
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 31

To get 21|δ| to be within our target of 0.2, we may take our tolerance δ = 0.9% <
0.2
21
≈ 0.0095.

Example 1.2.11. Assume x and y are close to 3 and 5. We want to find the tolerance
for x and y, such that xy is within ±0.2 of 3 · 5 = 15. This means finding δ > 0,
such that
|x − 3| < δ, |y − 5| < δ =⇒ |xy − 15| < 0.2.
Under the assumptions |x − 3| < δ and |y − 5| < δ, we have
|xy − 15| ≤ |xy − 3y| + |3y − 15| ≤ |x − 3||y| + 3|y − 5| ≤ (|y| + 3)δ.
We also note that, if we postulate δ ≤ 1, then |y − 5| < δ implies 4 < y < 6, so that
|xy − 15| ≤ (|y| + 3)δ ≤ (6 + 3)δ = 9δ.
To get 9|δ| to be within our target of 0.2, we may take our tolerance δ = 0.02 < 0.2
9
.
Since this indeed satisfies δ ≤ 1, we conclude that we can take δ = 0.02.
If the targeted error ±0.2 is changed to some other amount ±, then the same
argument shows that we can take the tolerance to be δ = 10 . Strictly speaking,
since we also use δ ≤ 1 in the argument above, we should take δ = min{ 10 , 1}.

Exercise 1.2.5. Find a tolerance for x, y, z near −2, 3, 5, such that 5x − 3y + 4z is within
± of 1.

Exercise 1.2.6. Find a tolerance for x and y near 2 and 2, such that xy is within ± of 4.

Exercise 1.2.7. Find a tolerance for x near 2, such that x2 is within ± of 4.

1
Exercise 1.2.8. Find a percentage of tolerance for x near 2, such that is within ±0.1 of
x
0.5.

1.2.3 Rigorous Proof of Limits

We revisit the limits derived before and make the argument rigorous.
√ √
Example 1.2.12. In Example 1.1.5, we argued that limn→∞ ( n + a − n) = 0. To
make the argument rigorous, we use the estimation in the earlier example. In fact,
regardless of the sign of a, we always have
 √ √ √ √ 
√ √  ( n + a − n)( n + a + n)  |a| |a|
| n + a − n − 0| =  √ √ = √ √ <√ .
n+a+ n  n+a+ n n
|a| a2
For the right side √ < , it is sufficient to have n > 2 . Then we can easily get
n 
a2 √ √
n> =⇒ | n + a − n − 0| < .
2
32 CHAPTER 1. LIMIT
√ √
This gives the rigorous proof of limn→∞ ( n + a − n) = 0.
r
n+2
Example 1.2.13. In Example 1.1.6, we argued that limn→∞ = 1. To make
n
the argument rigorous, we use the estimation in the earlier example. The estimation
1
suggested that it is sufficient to have 2 < . Thus we get the following rigorous
n
argument for the limit

r r 
2 n+2 n+2 2  n+2 
n< =⇒ 0 < −1< − 1 = <  =⇒  − 1 < .
 
 n n n  n 

Exercise 1.2.9. Rigorously prove the limits.

√ √
n+2 n+2 sin n
1. . 5. √ . 10. .
n−3 n−3 n
√ √
n+a cos n + a
n−2 6. . 11. .
2. . n+b n + b sin n
n+3
n n √ √
7. − . 12. n + a − n + b.
n+a n+1 n−1
3. . n+a n+c r
n+b 8. − . n+a
n+b n+d 13. .
n+b
2n2 − 3n + 2 1 √ √
4. . 9. √ . 14. 3
n + 1 − 3 n.
3n2 − 4n + 1 an + b

Exercise 1.2.10. Rigorously prove the limits, p > 0.

a np + a a sin n + b
1. √ . 2. . 3. .
np +b np + b np + c
√ a
Example 1.2.14. The estimation in Example 1.1.7 tells us that | n a − 1| < for
n
a
a > 1. This suggests that for any  > 0, we may choose N = . Then

√ a a
n > N =⇒ | n a − 1| < < = .
n N
√
This rigorously proves that limn→∞ n
a = 1 in case a ≥ 1.

Example 1.2.15. We try to rigorously prove limn→∞ n2 an = 0 for |a| < 1.

1
Using the idea of Example 1.1.11, we write |a| = . Then |a| < 1 implies
1+b
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 33

b > 0, and for n ≥ 3, we have

n2
|n2 an − 0| = n2 |a|n =
(1 + b)n
n2
=
n(n − 1) 2 n(n − 1)(n − 2) 3
1 + nb + b + b + · · · + bn
2! 3!
n2 3!n 3!n 3!22
< = < nn 3 = .
n(n − 1)(n − 2) 3 (n − 1)(n − 2)b3 b nb3
b 22
3!
3!22 3!22
Since <  is the same as n > , we have
nb3 b3 
3!22 2 n 3!22
n> and n ≥ 3 =⇒ |n a − 0| < < .
b3  nb3
 2 
3!2
This shows that we may choose N = max ,3 .
b3 
It is clear from the proof that we generally have

lim np an = 0, for any p and |a| < 1.

n→∞

an
Example 1.2.16. We rigorously prove limn→∞ = 0 in Example 1.1.13.
n!
Choose a natural number M satisfying |a| < M . Then for n > M , we have
 n
a  Mn M · M ···M M M M MM M M M +1 1
 < = · · · · · ≤ · = · .
 n!  n! 1 · 2···M M +1 M +2 n M! n M! n
Therefore for any  > 0, we have
 M +1  n
 M M +1 1 M M +1
 
M a 1
n > max , M =⇒  − 0 < · < · M +1 = .
M ! n! M! n M! M
M !

Exercise 1.2.11. Rigorously prove the limits.

√ np np an
1. n
n. 3. . 5. . 7. np an , |a| < 1.
n! n!
n5.4 n5.4 3n n!
2. . 4. . 6. .
n! n! nn

1.2.4 Rigorous Proof of Limit Properties

The rigorous definition of limit allows us to rigorously prove some limit properties.
34 CHAPTER 1. LIMIT

√ √
Example 1.2.17. Suppose limn→∞ xn = l > 0. We prove that limn→∞ xn = l.
First we clarify the problem. The limit limn→∞ xn = l means the implication

For any  > 0, there is N , such that n > N =⇒ |xn − l| < .

√ √
The limit limn→∞ xn = l means the implication
√ √
For any  > 0, there is N , such that n > N =⇒ | xn − l| < .

We need to argue is that the first implication implies the second implication.
We have
√ √ √ √
√ √ |( xn − l)( xn + l)| |xn − l| |x − l|
| xn − l| = √ √ =√ √ ≤ n√ .
xn + l xn + l l

|xn − l|
Therefore for any given  > 0, the second implication will hold as long as √ < ,
√ √ l
or |xn − l| < l. The inequality |xn − l| < l can√be achieved from the first
implication, provided we apply the first implication to l in place of .
The analysis above leads to the√following formal proof. Let  > 0. By applying
the definition of limn→∞ xn = l to l > 0, there is N , such that
√
n > N =⇒ |xn − l| < l.

Then
√
n > N =⇒ |xn − l| < l
√ √ √ √
√ √ |( xn − l)( xn + l)| |xn − l| |x − l|
=⇒ | xn − l| = √ √ =√ √ ≤ n√ < .
xn + l xn + l l

In the argument, we take advantage

√ of the fact that the definition of limit can
be applied to any positive number, l for example, instead of the given positive
number .

Example 1.2.18. We prove the arithmetic rule limn→∞ (xn + yn ) = limn→∞ xn +

limn→∞ yn in Proposition 1.1.3. The concrete Example 1.2.9 provides idea of the
proof.
Let limn→∞ xn = l and limn→∞ yn = k. Then for any 1 > 0, 2 > 0, there are
N1 , N2 , such that

n > N1 =⇒ |xn − l| < 1 ,

n > N2 =⇒ |yn − k| < 2 .

We expect to choose 1 , 2 as some modification of , as demonstrated in Example

1.2.17.
1.2. RIGOROUS DEFINITION OF SEQUENCE LIMIT 35

Let N = max{N1 , N2 }. Then

n > N =⇒ n > N1 , n > N2

=⇒ |xn − l| < 1 , |yn − k| < 2
=⇒ |(xn + yn ) − (l + k)| ≤ |xn − l| + |yn − k| < 1 + 2 .

If 1 + 2 ≤ , then this rigorously proves limn→∞ (xn + yn ) = l + k. Of course this


means that we may choose 1 = 2 = at the beginning of the argument.
2
The analysis above leads to the following formal proof. For any  > 0, apply the

definition of limn→∞ xn = l and limn→∞ yn = k to > 0. We find N1 and N2 , such
2
that

n > N1 =⇒ |xn − l| < ,
2

n > N2 =⇒ |yn − k| < .
2

Then
 
n > max{N1 , N2 } =⇒ |xn − l| < , |yn − k| <
2 2
 
=⇒ |(xn + yn ) − (l + k)| ≤ |xn − l| + |yn − k| < + = .
2 2

Example 1.2.19. The arithmetic rule limn→∞ xn yn = limn→∞ xn limn→∞ yn in Propo-

sition 1.1.3 means that, if we know the approximate values of the width and height
of a rectangle, then multiplying the width and height approximates the area of the
rectangle. The rigorous proof requires us to estimate how the approximation of the
area is affected by the approximations of the width and height. Example 1.2.11
gives the key idea for such estimation.
area= |x − l||y|

k y

l
area= |l||y − k|
x

Figure 1.2.2: The error in product.

36 CHAPTER 1. LIMIT

Let limn→∞ xn = l and limn→∞ yn = k. Then for any 1 > 0, 2 > 0, there are
N1 , N2 , such that
n > N1 =⇒ |xn − l| < 1 ,
n > N2 =⇒ |yn − k| < 2 .
Then for n > N = max{N1 , N2 }, we have (see Figure 1.2.2)
|xn yn − lk| = |(xn − l)yn + l(yn − k)|
≤ |xn − l||yn | + |l||yn − k|
< 1 (|k| + 2 ) + |l|2 ,
where we use |yn − k| < 2 implying |yn | < |k| + 2 . The proof of limn→∞ xn yn = lk
will be complete if, for any  > 0, we can choose 1 > 0 and 2 > 0, such that
1 (|k| + 2 ) + |l|2 ≤ .
This can be achieved by choosing 1 , 2 satisfying
 
2 ≤ 1, 1 (|k| + 1) ≤ , |l|2 ≤ .
2 2
In other words, if we choose
 
 
1 = , 2 = min 1,
2(|k| + 1) 2|l|
at the very beginning of the proof, then we get a rigorous proof of the arithmetic
rule. The formal writing of the proof is left to the reader.

Example 1.2.20. The sandwich rule in Proposition 1.1.4 reflects the intuition that, if
x and z are within  of 5, then any number y between x and z is also within  of 5
|x − 5| < , |z − 5| < , x ≤ y ≤ z =⇒ |y − 5| < .
Geometrically, this means that if x and z lies inside an interval, say (5 − , 5 + ),
then any number y between x and z also lies in the interval.
Suppose xn ≤ yn ≤ zn and limn→∞ xn = limn→∞ zn = l. For any  > 0, there are
N1 and N2 , such that
n > N1 =⇒ |xn − l| < ,
n > N2 =⇒ |zn − l| < .
Then
n > N = max{N1 , N2 } =⇒ |xn − l| < , |zn − l| < 
=⇒ l −  < xn , zn < l + 
=⇒ l −  < xn ≤ yn ≤ zn < l + 
⇐⇒ |yn − l| < .
1.3. CRITERION FOR CONVERGENCE 37

Example 1.2.21. The order rule in Proposition 1.1.5 reflects the intuition that, if x is
very close to 3 and y is very close to 5, then x must be less than y. More specifically,
we know x < y when x and y are within ±1 of 3 and 5. Here 1 is half of the distance
between 3 and 5.
Suppose xn ≤ yn , limn→∞ xn = l, limn→∞ yn = k. For any  > 0, there is N ,
such that (you should know from earlier examples how to find this N )

n > N =⇒ |xn − l| < , |yn − k| < .

Picking any n > N , we get

l −  < xn ≤ yn < k + .

Therefore we proved that l −  < k +  for any  > 0. It is easy to see that the
property is the same as l ≤ k.
Conversely, we assume limn→∞ xn = l, limn→∞ yn = k, and l < k. For any  > 0,
there is N , such that n > N implies |xn − l| <  and |yn − k| < . Then

n > N =⇒ xn < l + , yn > k −  =⇒ yn − xn > (k − ) − (l + ) = k − l − 2.

k−l
By choosing  = > 0 at the beginning of the argument, we conclude that
2
yn > xn for n > N .

Exercise 1.2.12. Prove that if limn→∞ xn = l, then limn→∞ |xn | = |l|.

Exercise 1.2.13. Prove that limn→∞ |xn − l| = 0 if and only if limn→∞ xn = l.

Exercise 1.2.14. Prove that if limn→∞ xn = l, then limn→∞ cxn = cl.

Exercise 1.2.15. Prove that a sequence xn converges if and only if the subsequences x2n
and x2n+1 converge to the same limit. This is a special case of Proposition 1.1.6.

Exercise 1.2.16. Suppose xn ≥ 0 for sufficiently big n and limn→∞ xn = 0. Prove that
limn→∞ xpn = 0 for any p > 0.

Exercise 1.2.17. Suppose xn ≥ 0 for sufficiently big n and limn→∞ xn = 0. Suppose yn ≥ c

for sufficiently big n and some constant c > 0. Prove that limn→∞ xynn = 0.

1.3 Criterion for Convergence

Any number close to 3 must be between 2 and 4, and in particular have the absolute
value no more than 4. The intuition leads to the following result.

Theorem 1.3.1. If xn converges, then |xn | ≤ B for a constant B and all n.

38 CHAPTER 1. LIMIT

The theorem basically says that any convergent sequence is bounded. The num-
ber B is a bound for the sequence.
If xn ≤ B for all n, then we say xn is bounded above, and B is an upper bound.
If xn ≥ B for all n, then we say xn is bounded below, and B is a lower bound. A
sequence is bounded if and only if it is bounded above and bounded below.
n2 + (−1)n
The sequences n, diverge because they are not bounded. On the
n+1
other hand, the sequence 1, −1, 1, −1, . . . is bounded but diverges. Therefore the
converse of Theorem 1.3.1 is not true in general.

Exercise 1.3.1. Prove that if xn is bounded for sufficiently big n, i.e., |xn | ≤ B for n ≥ N ,
then xn is still bounded.

Exercise 1.3.2. Suppose xn is the union of two subsequences x0k and x00k . Prove that xn is
bounded if and only if both x0k and x00k are bounded.

1.3.1 Monotone Sequence

The converse of Theorem 1.3.1 holds under some additional assumption. A sequence
xn is increasing if
x1 ≤ x2 ≤ x3 ≤ · · · ≤ xn ≤ xn+1 ≤ · · · .
It is strictly increasing if
x1 < x2 < x3 < · · · < xn < xn+1 < · · · .
The concepts of decreasing and strictly decreasing can be similarly defined. More-
over, a sequence is monotone if it is either increasing or decreasing.
1 1 √ 1
The sequences , n , n 2 are (strictly) decreasing. The sequences − , n are
n 2 n
increasing.

Theorem 1.3.2. A monotone sequence converges if and only if it is bounded.

An increasing sequence xn is always bounded below by its first term x1 . Therefore

xn is bounded if and only if it is bounded above. Similarly, a decreasing sequence is
bounded if and only if it is bounded below.
The world record for 100 meter dash is a decreasing sequence bounded below by
0. The proposition reflects the intuition that there is a limit on how fast human
being can run. We note that the proposition does not tells us the exact value of the
limit, just like we do not know the exact limit of the human ability.

Example 1.3.1. Consider the sequence

1 1 1 1
xn = 2
+ 2 + 2 + ··· + 2.
1 2 3 n
1.3. CRITERION FOR CONVERGENCE 39

The sequence is clearly increasing. Moreover, the sequence is bounded above by

1 1 1
xn ≤ 1 + + + ··· +
1·2 2·3 (n − 1)n
     
1 1 1 1 1
=1+ 1− + − + ··· + −
2 2 3 n−1 n
1
= 2 − < 2.
n
Therefore the sequence converges.
The limit of the sequence is the sum of the infinite series
∞
X 1 1 1 1
2
= 1 + 2 + 2 + ··· + 2 + ··· .
n=1
n 2 3 n

π2
We will see that the sum is actually .
6

Exercise 1.3.3. Show the convergence of sequences.

1 1 1 1
1. xn = 3
+ 3 + 3 + ··· + 3.
1 2 3 n
1 1 1 1
2. xn = + + + ··· + .
12.4 22.4 32.4 n2.4
1 1 1 1
3. xn = + + + ··· + .
1·3 3·5 5·7 (2n − 1)(2n + 1)
1 1 1
4. xn = + + ··· + .
1! 2! n!

√
q p
Example 1.3.2. The number 2 + 2 + 2 + · · · is the limit of the sequence xn
inductively given by √ √
x1 = 2, xn+1 = 2 + xn .
After trying first couple of terms, we expect
p the√sequence to√be increasing. This
can be verified by induction. We have x2 = 2 + 2 > x1 = 2. Moreover, if we
assume xn > xn−1 , then
√ p
xn+1 = 2 + xn > 2 + xn−1 = xn .

This proves inductively that xn is indeed increasing.

Next we claim that xn is bounded above. For an increasing sequence, we expect
its limit to be the upper bound. So we find the hypothetical limit first. Taking the
limit on both sides of the equality x2n+1 = 2 + xn and applying the arithmetic rule,
we get l2 = 2 + l. The solution is l = 2 or −1. Since xn > 0, by the order rule, we
must have l ≥ 0. Therefore we conclude that l = 2.
40 CHAPTER 1. LIMIT

The hypothetical limit value suggests

√ that xn < 2 for all n. Again we verify this
by induction. We already have x1 = 2 < 2. If we assume xn < 2, then
√ √
xn+1 = 2 + xn < 2 + 2 = 2.

This proves inductively that xn < 2 for all n.

We conclude that xn is increasing and bounded above. By Theorem 1.3.2, the
sequence converges, and the hypothetical limit value 2 is the real limit value.
Figure 1.3.1 suggests that our conclusion actually depends only on the general
shape
√ of the graph of the function, and has little to do with the exact formula
2 + x.

y=x

y = f (x)
f (x3 )
f (x2 )

f (x1 )

x1 x2 x3 x4 l

Figure 1.3.1: Limit of inductively defined sequence.

√
Exercise 1.3.4. Suppose a sequence xn satisfies xn+1 = 2 + xn .

1. Prove that if −2 < x1 < 2, then xn is increasing and converges to 2.

2. Prove that if x1 > 2, then xn is decreasing and converges to 2.

Exercise 1.3.5. For the three functions f (x) in Figure 1.3.2, study the convergence of the
sequences xn defined by xn+1 = f (xn ). Your answer depends on the initial value x1 .

1
Exercise 1.3.6. Suppose a sequence xn satisfies xn+1 = (x2n + xn ). Prove the following
2
statements.

1. If x1 > 1, then the sequence is increasing and diverges.

2. If 0 < x1 < 1, then the sequence is decreasing and converges to 0.

3. If −1 < x1 < 0, then the sequence is increasing and converges to 0.

4. If −2 < x1 < −1, then the sequence is decreasing for n ≥ 2 and converges to 0.
1.3. CRITERION FOR CONVERGENCE 41

Figure 1.3.2: Three functions

5. If x1 < −2, then the sequence is increasing for n ≥ 2 and diverges.

Exercise 1.3.7. Determine the convergence of inductively defined sequences. Your answer
may depend on the initial value x1 .

1. xn+1 = x2n . 3. xn+1 = 2x2n − 1. 1

5. xn+1 = 1 + .
xn
x2n + 1 1 1
2. xn+1 = . 4. xn+1 = . 6. xn+1 = 2 − .
2 xn xn

Exercise 1.3.8. Determine the convergence of inductively defined sequences, a > 0. In

some cases, the sequence may not be defined after certain number of terms.
√ √ √
1. xn+1 = a + xn . 3. xn+1 = a − xn . 5. xn+1 = 3 xn − a.
√ √ √
2. xn+1 = xn − a. 4. xn+1 = 3 a + xn . 6. xn+1 = 3 a − xn .

Exercise 1.3.9. Explain the continued fraction expansion

√ 1
2=1+ .
1
2+
1
2+
2 + ···
What if 2 on the right side is changed to some other positive number?

Exercise 1.3.10. For any a, b > 0, define a sequence by

xn−1 + xn−2
x1 = a, x2 = b, xn = .
2
Prove that the sequence converges.

a+b √
Exercise 1.3.11. The arithmetic and the geometric means of a, b > 0 are and ab.
2
By repeating the process, we get two sequences defined by
xn + yn √
x1 = a, y1 = b, xn+1 = , yn+1 = xn yn .
2
42 CHAPTER 1. LIMIT

Prove that xn ≥ xn+1 ≥ yn+1 ≥ yn for n ≥ 2, and the two sequences converge to the same
limit.

Exercise 1.3.12. The Fibonacci sequence

1, 1, 2, 3, 5, 8, 13, 21, 34, . . .

xn+1
is defined by x0 = x1 = 1 and xn+1 = xn + xn−1 . Consider the sequence yn = .
xn
1. Find the relation between yn+1 and yn .

2. Assume yn converges, find the limit l.

3. Use the relation between yn+2 and yn to prove that l is the upper bound of y2k and
the lower bound of y2k+1 .

4. Prove that the subsequence y2k is increasing and the subsequence y2k+1 is decreasing.

5. Prove that the sequence yn converges to l.

√ √
Exercise 1.3.13. To find a for a > 0, we start with a guess x1 > 0 of the value of a.
a √
Noting that x1 and are on the two sides of a, it is reasonable to choose the average
  1x
1 a
x2 = x1 + as the next guess. This leads to the inductive formula
2 x1
 
1 a
xn+1 = xn +
2 xn
√
as a way of numerically computing better and better approximate values of a.
√
1. Prove that limn→∞ xn = a.
 
1 a
2. We may also use weighted average xn+1 = xn + 2 as the next guess. Do we
√ 3 xn
still have limn→∞ xn = a for the weighted average?

3. Compare the two methods for specific values of a and b (say a = 4, b = 1). Which
way is faster?
√
4. Can you come up with a similar scheme for numerically computing 3 a? What
choice of the weight gives you the fastest method?

1.3.2 Application of Monotone Sequence

We use Theorem 1.3.2 to prove some limits and define a special number e.

Example 1.3.3. We give another argument for limn→∞ an = 0 in Example 1.1.11.

1.3. CRITERION FOR CONVERGENCE 43

First assume 0 < a < 1. Then the sequence an is decreasing and satisfies
0 < an < 1. Therefore the sequence converges to a limit l. By the remark in
Example 1.1.1, we also have limn→∞ an−1 = l. Then by the arithmetic rule, we have

l = lim an = lim a · an−1 = a lim an−1 = al.

n→∞ n→∞ n→∞

Since a 6= 1, we get l = 0.
For the case −1 < a < 0, we may consider the even and odd subsequences
of an and apply Proposition 1.1.6. Another way is to apply the sandwich rule to
−|a|n ≤ an ≤ |a|n .

3n (n!)2
Example 1.3.4. We give another argument that the sequence xn = in Exam-
(2n)!
xn
ple 1.1.16 converges to 0. By limn→∞ = 0.75 < 1 and the order rule, we have
xn−1
xn
< 1 for sufficiently big n. Since xn is always positive, we have xn < xn−1 for
xn−1
sufficiently big n. Therefore after finitely many terms, the sequence is decreasing.
Moreover, 0 is the lower bound of the sequence, so that the sequence converges.
Let limn→∞ xn = l. Then we also have limn→∞ xn−1 = l. If l 6= 0, then

xn limn→∞ xn l
lim = = = 1.
n→∞ xn−1 limn→∞ xn−1 l

But the limit is actually 0.75. The contradiction shows that l = 0.

Exercise 1.3.14. Extend Example 1.3.3 to a proof of limn→∞ nan = 0 for |a| < 1.

|xn |
Exercise 1.3.15. Extend Example 1.3.4 to prove that, if limn→∞ = l < 1, then
|xn−1 |
limn→∞ xn = 0.

 n
1
Example 1.3.5. For the sequence 1 + , we compare two consecutive terms by
n
44 CHAPTER 1. LIMIT

their binomial expansions

 n
1 1 n(n − 1) 1 n(n − 1) · · · 3 · 2 · 1 1
1+ =1+n + 2
+ ··· +
n n 2! n n! nn
 
1 1 1
=1+ + 1− + ···
1! 2! n
    
1 1 2 n−1
+ 1− 1− ··· 1 − ,
n! n n n
 n+1  
1 1 1 1
1+ =1+ + 1− + ···
n+1 1! 2! n+1
    
1 1 2 n−1
+ 1− 1− ··· 1 −
n! n+1 n+1 n+1
    
1 1 2 n
+ 1− 1− ··· 1 − .
(n + 1)! n+1 n+1 n+1

A close examination shows that the sequence is increasing. Moreover, by the com-
putation in Example 1.3.1, the first expansion gives
 n
1 1 1 1
1+ < 1 + + + ··· +
n 1! 2! n!
1 1 1
<1+1+ + + ··· + < 3.
1·2 2·3 (n − 1)n

By Theorem 1.3.2, the sequence converges. We denote the limit by e

 n
1
e = lim 1 + = 2.71828182845904 · · · .
n→∞ n

Exercise 1.3.16. Find the limit.

1 n 1 n
n+1      n
2n + 1

n+1
1. . 2. 1 − . 3. 1 + . 4. .
n n 2n 2n − 1

1 n+1
 
Exercise 1.3.17. Let xn = 1 + .
n

1. Use induction to prove (1 + x)n ≥ 1 + nx for x > −1 and any natural number n.
xn−1
2. Use the first part to prove > 1. This shows that xn is decreasing.
xn
3. Prove that limn→∞ xn = e.

1 n
 
4. Prove that 1 − is increasing and converges to e−1 .
n
1.3. CRITERION FOR CONVERGENCE 45

Exercise 1.3.18. Prove that for n > k, we have

1 n
        
1 1 1 1 1 2 k
1+ ≥1+ + 1− + ··· + 1− 1− ··· 1 − .
n 1! 2! n k! n n n

Then use Proposition 1.1.5 to show that

 k
1 1 1 1
e ≥ 1 + + + ··· + ≥ 1+ .
1! 2! k! k

Finally, prove  
1 1 1
lim 1 + + + ··· + = e.
n→∞ 1! 2! n!

1.3.3 Cauchy Criterion

Theorem 1.3.2 gives a special case that we know the convergence of a sequence
without knowing the actual limit value. Note that the definition of limit makes
explicit use of the limit value and therefore cannot be used to derive the convergence
here. The following provides the criterion for the convergence in general, again
without referring to the actual limit value.

Theorem 1.3.3 (Cauchy Criterion). A sequence xn converges if and only if for any
 > 0, there is N , such that

m, n > N =⇒ |xm − xn | < .

Sequences satisfying the property in the theorem are called Cauchy sequences.
The theorem says that a sequence converges if and only if it is a Cauchy sequence.
The necessity is easy to see. If limn→∞ xn = l, then for big m, n, both xm and
xn are very close to l (say within 2 ). This implies that xm and xn are very close
(within 2 + 2 = ).
The proof of sufficiency is much more difficult and relies on the following deep
result that touches the essential difference between the real and rational numbers.

Theorem 1.3.4 (Bolzano-Weierstrass). Any bounded sequence has a convergent sub-

sequence.

Using the theorem, the converse may be proved by the following steps.

1. A Cauchy sequence is bounded.

2. By Bolzano-Weierstrass Theorem, the sequence has a convergent subsequence.

3. If a Cauchy sequence has a subsequence converging to l, then the whole se-

quence converges to l.
46 CHAPTER 1. LIMIT

Example 1.3.6. In Example 1.1.19, we argued that the sequence (−1)n diverges be-
cause two subsequences converge to different limits. Alternatively, we may apply
the Cauchy criterion. For  = 1 and any N , we pick any n > N and pick m = n + 1.
Then m, n > N and |xm − xn | = |(−1)n+1 − (−1)n | = 2 > . This means that the
Cauchy criterion fails, and therefore the sequence diverges.

Example 1.3.7. In Example 1.3.1, we argued the convergence of

1 1 1
xn = 1 + 2
+ 2 + ··· + 2
2 3 n
by increasing and bounded property. Alternatively, for m > n, we have
1 1 1
|xm − xn | = 2
+ 2
+ ··· + 2
(n + 1) (n + 2) m
1 1 1
≤ + + ··· +
n(n + 1) (n + 1)(n + 2) (m − 1)m
     
1 1 1 1 1 1
= − + − + ··· + −
n n+1 n+1 n+2 m−1 m
1 1
= − .
n m
Therefore for any , we have
1 1 1
m>n>N = =⇒ |xn − xm | < < = .
 n N
By the Cauchy criterion, xn converges.
We note that the same argument can be used to show the convergence of
1 1 1
xn = 1 − 2
+ 2 − · · · + (−1)n 2 .
2 3 n
The method in Example 1.3.7 cannot be used here.

Example 1.3.8. The following is the partial sum of the harmonic series
1 1 1
xn = 1 + + + ··· + .
2 3 n
For any n, we have
1 1 1 1 1 1 1
x2n − xn = + + ··· + ≥ + + ··· + = .
n+1 n+2 2n 2n 2n 2n 2
1
For  = and any N , we choose a natural number n > N and also choose m =
2
2n > N . Then
|xm − xn | = x2n − xn ≥ .
This shows that the sequence fails the Cauchy criterion and diverges.
1.3. CRITERION FOR CONVERGENCE 47

Exercise 1.3.19. If xn is a Cauchy sequence, is |xn | also a Cauchy sequence? What about
the converse?

1
Exercise 1.3.20. Suppose |xn+1 − xn | ≤ . Prove that xn converges.
n2

Exercise 1.3.21. Suppose cn is bounded and |r| < 1. Prove that the sequence

xn = c0 + c1 r + c2 r2 + · · · + cn rn

converges.

Exercise 1.3.22. Use Cauchy criterion to determine convergence.

1 1 1 1 1 1
1. 1 + √ + √ + · · · + √ . 4. 1 + + + ··· + .
2 3 n 3 5 2n + 1
1 1 (−1)n+1 1 1 1
2. 1 − + − · · · + . 5. 1 + + + ··· + .
23 33 n3 2! 3! n!
1 2 3 n−1 2 3 n
3. + + + ··· + . 6. 1 + + + ··· + .
2 3 4 n 1! 2! (n − 1)!

Example 1.3.9. Theorem 1.3.4 shows the existence of converging subsequences of a

bounded sequence. How about the limit values of such subsequences?
Let us list all finite decimal expressions in (0, 1) as a sequence

xn : 0.1, 0.2, . . . , 0.9, 0.01, 0.02, . . . , 0.99, 0.001, 0.002, . . . , 0.999, . . . .

The number 0.318309 · · · is the limit of the following subsequence

0.3, 0.31, 0.318, 0.3183, 0.31830, 0.318309, . . . .

It is easy to see that any number in [0, 1] is the limit of a convergent subsequence
of xn .

Exercise 1.3.23. Construct a sequence such that the limits of convergent subsequences are
1
exactly , n ∈ N and 0.
n

Exercise 1.3.24. Construct a sequence such that any number is the limit of some convergent
subsequence.

Exercise 1.3.25. Use any suitable method or theorem to determine convergence.

1 2 2 3 n n+1 1 2 2 3 n n+1
1. , , , , ..., , , .... 2. , − , , − , ..., , − , ....
2 1 3 2 n+1 n 2 1 3 2 n+1 n
48 CHAPTER 1. LIMIT

1 2 n−1 1 · 3 · 5 · · · (2n − 1)
3. 1 + 2
+ 2 + ··· + . 7. .
2 3 n2 2 · 4 · 6 · · · (2n)

1 1 1 1 · 3 · 5 · · · (2n − 1)
4. + + ··· + . 8. .
1·2 3·4 (2n − 1)2n n!
v v
u u s
u r
1 1 1 u u1 1 1
5. − +· · ·+(−1)n+1 . 9. t 1+ t + + ··· + .
1·2 3·4 (2n − 1)2n 2 3 n
r
2 3 n q
√
+ ··· +
p
6. + . 10. 1+ 2 + 3 + · · · + n.
1·3 2·4 (n − 1)(n + 1)

1.4 Infinity
1.4.1 Divergence to Infinity
A sequence may diverge for various reasons. For example, the sequence n diverges
because it can become arbitrarily big. On the other hand, the bounded sequence
(−1)n diverges because it has two subsequences with different limits. The first
example may be summarized by the following definition.

Definition 1.4.1. A sequence diverges to infinity, denoted limn→∞ xn = ∞, if for

any B, there is N , such that n > N implies |xn | > B.

In the definition, the infinity means that the absolute value (or the magnitude)
of the sequence can become arbitrarily big. If we further take into account of the
signs, then we get the following definitions.

Definition 1.4.2. A sequence diverges to +∞, denoted limn→∞ xn = +∞, if for

any B, there is N , such that n > N implies xn > B. A sequence diverges to −∞,
denoted limn→∞ xn = −∞, if for any B, there is N , such that n > N implies xn < B.

The meaning of limn→∞ xn = +∞ is illustrated in Figure 1.4.1. For example, we

have limn→∞ n = +∞. We also note that, in the definition of limn→∞ xn = +∞, we
may additionally assume B > 0 (or B > 100) without loss of generality.

Example 1.4.1. We have

lim np = +∞, for p > 0.
n→∞

For the rigorous proof, for any B > 0, we have

1
n > B p =⇒ np > B.
1.4. INFINITY 49

xN +1
x2 x xn
x4 xN +2 N +3
B
xn+1
x1 x3

n
N

Figure 1.4.1: n > N implies xn > B.

Example 1.4.2. Example 1.1.11 may be extended to show that limn→∞ an = ∞ for
|a| > 1. Specifically, let |a| = 1 + b. Then |a| > 1 implies b > 0, and we have
n(n − 1) 2
|an | = (1 + b)n = 1 + nb + b + · · · + bn > nb.
2
For any B, we then have
B
n> =⇒ |an | > nb > B.
b
This proves limn→∞ an = ∞ for |a| > 1. If we take the sign into account, this also
proves limn→∞ an = +∞ for a > 1.

1
Example 1.4.3. Suppose xn 6= 0. We prove that limn→∞ xn = 0 implies limn→∞ =
xn
∞. Actually the converse is also true and the proof is left to the reader.
If limn→∞ xn = 0, then for any B > 0, we apply the definition of the limit to
1
> 0 to get N , such that
B
 
1 1
n > N =⇒ |xn | < =⇒   > B.
B xn
1
This proves limn→∞ = ∞.
xn
Applying what we just proved to the limit in Example 1.1.11, we get another
proof of limn→∞ an = ∞ for |a| > 1.

1
Exercise 1.4.1. Let xn 6= 0. Prove that limn→∞ xn = 0 if and only if limn→∞ = ∞.
xn

Exercise 1.4.2. Prove that limn→∞ xn = +∞ if and only if xn > 0 for sufficiently big n
1
and limn→∞ = 0.
xn

Exercise 1.4.3. Rigorously prove divergence to infinity. Determine ±∞ if possible.

50 CHAPTER 1. LIMIT

n2 − n + 1 an n!
1. . 3. , |a| > 1. 5. .
n+1 n 4n
n
2. √ . 4. np an , |a| > 1. 6. n!an .
n+1

1.4.2 Arithmetic Rule for Infinity

A sequence xn is an infinitesimal if limn→∞ xn = 0. Example 1.4.3 and Exercise 1.4.1
show that a sequence is an infinetesimal if and only if its reciprocal is an infinity.
Many properties of the finite limit can be extended to infinity. For example, we
l
have = ∞ for l 6= 0. This means that, if limn→∞ xn = l 6= 0 and limn→∞ yn = 0,
0
xn
then limn→∞ = ∞. For example,
yn
 
1 1
2+ 2 limn→∞ 2 + 2
2n2 + 1 n n 2
lim = lim =  = = ∞.
1 1

n→∞ n − 1 n→∞ 1 1 0
− limn→∞ − 2
n n2 n n
l
Note that l in can represent any sequence converging to l, and is not necessarily
0
a constant.
The following are more extensions of the arithmetic rules to infinity. The rules
are symbolically denoted by “arithmetic equalities”, and the exact meaning of the
rules are also given.
l xn
• = 0: If limn→∞ xn = l and limn→∞ yn = ∞, then limn→∞ = 0.
∞ yn
• (+∞) + (+∞) = +∞: If limn→∞ xn = +∞ and limn→∞ yn = +∞, then
limn→∞ (xn + yn ) = +∞.
• (−∞) + l = −∞: If limn→∞ xn = −∞ and limn→∞ yn = l, then limn→∞ (xn +
yn ) = −∞.
• (+∞) · l = −∞ for l < 0: If limn→∞ xn = +∞ and limn→∞ yn = l < 0, then
limn→∞ xn yn = −∞.
l
• = +∞ for l > 0: If limn→∞ xn = l > 0, limn→∞ yn = 0 and yn > 0, then
0+
xn
limn→∞ = +∞.
yn
On the other hand, we should always be cautious not to overextend the arithmetic
properties. For example, the following “arithmetic rules” are actually wrong
+∞
∞ + ∞ = ∞, = −1, 0 · ∞ = 0, 0 · ∞ = ∞.
−∞
1.4. INFINITY 51

A counterexample for the first equality is xn = n and yn = −n, for which we have
limn→∞ xn = ∞, limn→∞ yn = ∞ and limn→∞ (xn + yn ) = 0. In general, one needs to
use common sense to decide whether certain extended arithmetic rules make sense.

Example 1.4.4. By Example 1.4.1 and the extended arithmetic rule, we have
 
3 3 3 1
lim (n − 3n + 1) = lim n 1 − 2 + 3 = (+∞) · 1 = +∞.
n→∞ n→∞ n n

In general, any non-constant polynomial of n diverges to ∞, and for rational func-

tions, we have



 +∞, if p > q, ap bq > 0,

ap np + ap−1 np−1 + · · · + a1 n + a0 −∞, if p > q, ap bq < 0,

lim = ap
n→∞ bq nq + bq−1 nq−1 + · · · + b1 n + b0 , if p = q, bq 6= 0,
bq




if p < q, bq 6= 0.

0,

Exercise 1.4.4. Prove the extended arithmetic rules

l l
= ∞, l + (+∞) = +∞, (+∞) · (−∞) = −∞, = −∞ for l > 0.
0 0−

Exercise 1.4.5. Construct sequences xn and yn , such that both diverge to infinity, but
xn + yn can have any of the following behaviors.

1. limn→∞ (xn + yn ) = ∞.

2. limn→∞ (xn + yn ) = 2.

3. xn + yn is bounded but does not converge.

4. xn + yn is not bounded and does not diverge to infinity.

The exercise shows that ∞ + ∞ has no definite meaning.

Exercise 1.4.6. Prove that if p > 0, then limn→∞ xn = +∞ implies limn→∞ xpn = +∞.
What about the case p < 0?

Exercise 1.4.7. Prove the extended sandwich rule: If xn ≤ yn for sufficiently big n, then
limn→∞ xn = +∞ implies limn→∞ yn = +∞.

Exercise 1.4.8. Prove the extended order rule: If limn→∞ xn = l is finite and limn→∞ yn =
+∞, then xn < yn for sufficiently big n.

Exercise 1.4.9. Suppose limn→∞ xn = l > 1. Prove that limn→∞ xnn = +∞.
52 CHAPTER 1. LIMIT

xn
Exercise 1.4.10. Prove that if limn→∞ = l and |l| > 1, then xn diverges to infinity.
xn−1

Exercise 1.4.11. Explain the infinities. Determine the sign of infinity if possible.
n + sin 2n 1 3n − 2n
1. √ . 4. √ √ . 7. .
n − cos n n
n − n 2n n
n! √ √ 3n − 2n
2. n , a + b 6= 0. 5. n( n + 2 − n). 8. .
a + bn n3 + n2
 2
1 n

1 (−1)n n2
3. √ . 6. . 9. 1 + .
n
n−1 n−1 n

1.4.3 Unbounded Monotone Sequence

The following complements Theorem 1.3.2.

Theorem 1.4.3. Any unbounded monotone sequence diverges to infinity.

If an increasing sequence xn is bounded, then by Theorem 1.3.2, it converges to

a finite limit. If the sequence is not bounded, then it is not bounded above. This
means that any number B is not an upper bound, or some xN > B. Then by xn
increasing, we have
n > N =⇒ xn ≥ xN > B.
This proves that limn→∞ xn = +∞. Similarly, an unbounded decreasing sequence
diverges to −∞.

Example 1.4.5. In Example 1.3.8, we showed that the increasing sequence

1 1 1
xn = 1 + + + ··· +
2 3 n
diverges. By Theorem 1.4.3, we know that it diverges to +∞. Therefore the sum of
the harmonic series
1 1 1
1 + + + · · · + + · · · = +∞.
2 3 n

Example 1.4.6. For a > 1, the sequence an is increasing. If the sequence converges
to a finite limit l, then

l = lim an = a lim an−1 = al.

n→∞ n→∞

Since the sequence is increasing, we have l ≥ a > 1, which contradicts to l = al. By

Theorem 1.4.3, therefore, we conclude that limn→∞ an = +∞.

an
Exercise 1.4.12. Prove limn→∞ = +∞ for a > 1.
n2
1.5. LIMIT OF FUNCTION 53

1.5 Limit of Function

Similar to the sequence limit, we say a function f (x) converges to l at a, and write

lim f (x) = l,
x→a

if f (x) approaches l when x approaches a

x → a, x 6= a =⇒ f (x) → l.

Note that we include x 6= a because f (x) is not required to be defined at a.

The definition allows a or l to be ∞ (or ±∞ if the sign can be determined).
When l = ∞, we should say that f (x) diverges to ∞ at a.
1
In Figure 1.5.1, as x approaches 0, we see that x2 approaches 0, gets arbitrarily
x
1
big, and sin swings between −1 and 1, never approaching any one specific finite
x
1 1
number. We write limx→0 x2 = 0, limx→0 = ∞, and say that limx→0 sin diverges.
x x

1 1
y = x2 y= y = sin
x x

Figure 1.5.1: Behavior of functions near 0.

On the other hand, as x approaches the infinity, we find that x2 gets arbitrar-
1 1 1
ily big, and and sin approach 0. Therefore limx→∞ x2 = ∞, limx→∞ =
x x x
1
limx→∞ sin = 0. Moreover, sin x swings between −1 and 1, and limx→∞ sin x
x
diverges.

1.5.1 Properties of Function Limit

The function limit shares similar properties with the sequence limit.

Proposition 1.5.1. If f (x) = g(x) for x sufficiently near a and x 6= a, then

limx→a f (x) = limx→a g(x).
54 CHAPTER 1. LIMIT

Proposition 1.5.2 (Arithmetic Rule). Suppose limx→a f (x) = l and limx→a g(x) = k.
Then
f (x) l
lim (f (x) + g(x)) = l + k, lim cf (x) = cl, lim f (x)g(x) = lk, lim = ,
x→a x→a x→a x→a g(x) k
where c is a constant and k 6= 0 in the last equality.

Proposition 1.5.3 (Sandwich Rule). If f (x) ≤ g(x) ≤ h(x) and limx→a f (x) =
limx→a h(x) = l, then limx→a g(x) = l.

Proposition 1.5.4 (Order Rule). Suppose limx→a f (x) = l and limx→a g(x) = k.
1. If f (x) ≤ g(x) for x near a and x 6= a, then l ≤ k.
2. If l < k, then f (x) < g(x) for x near a and x 6= a.

Proposition 1.1.6 will be extended to Proposition 1.5.5 (composition rule) and

Proposition 1.6.2 (relation between sequence and function limits).
Since
x → a, x 6= a =⇒ c → c,
and
x → a, x 6= a =⇒ x → a,
we have
lim c = c, lim x = a.
x→a x→a
It is also intuitively clear that
1
lim |x| = 0, lim = 0 for p > 0.
x→0 x→∞ |x|p

The subsequent examples are based on these limits.

Example 1.5.1. For a > 0, we have |x| = x near a. By Proposition 1.5.1, we have
limx→a |x| = limx→a x = a = |a|.
For a < 0, we have |x| = −x near a. By Propositions 1.5.1 and 1.5.2, we have
limx→a |x| = limx→a −x = − limx→a x = −a = |a|.
Combining the two cases with limx→0 |x| = 0, we get limx→a |x| = |a|.

Example 1.5.2. Let (

x, if x 6= 0,
f (x) =
A, if x = 0.
Then f (x) = x for x 6= 0, and by Proposition 1.5.1, we have limx→0 f (x) =
limx→0 x = 0. Note that the limit is independent of the value f (0) = A. In fact, the
limit does not even require the function to be defined at 0.
1.5. LIMIT OF FUNCTION 55

x3 − 1
Example 1.5.3. The rational function is not defined at x = 1. Yet the function
x−1
converges at 1

x3 − 1  2
lim = lim (x2 + x + 1) = lim x + lim x + lim 1 = 12 + 1 + 1 = 3.
x→1 x − 1 x→1 x→1 x→1 x→1

Example 1.5.4. By the arithmetic rule, we have

 3
lim (x3 − 3x + 1) = lim x − lim 3 lim x + lim 1 = a3 − 3a + 1.
x→a x→a x→a x→a x→a

More generally, for any polynomial p(x) = cn xn + cn−1 xn−1 + · · · + c1 x + c0 and finite
a, we have
lim p(x) = p(a).
x→a

p(x)
A rational function r(x) = is the quotient of two polynomials p(x) and
q(x)
q(x), and is defined at a if q(a) 6= 0. Further by the arithmetic rule, we also have

limx→a p(x) p(a)

lim r(x) = = = r(a)
x→a limx→a q(x) q(a)

whenever r(x) is defined at a.

Example 1.5.5. By the arithmetic rule, we have

1 1
2x2 + x 2+ 2 + limx→∞
lim 2 = lim x = x = 2.
x→∞ x − x + 1 x→∞ 1 1 1 1 1
1− + 2 1 − limx→∞ + limx→∞ · limx→∞
x x x x x
This is comparable to Example 1.1.2. In general, Example 1.4.4 can be extended to
the function limit

∞, if p > q,

ap xp + ap−1 xp−1 + · · · + a1 x + a0  ap , if p = q, b 6= 0,
lim = q
x→∞ bq xq + bq−1 xq−1 + · · · + b1 x + b0 
 bq
0, if p < q, bq 6= 0.


p p
Example 1.5.6. Similar to Example 1.1.5, the function |x| + 2 − |x| satisfies
p p p p
p p ( |x| + 2 − |x|)( |x| + 2 + |x|)
0 < |x| + 2 − |x| = p p
|x| + 2 + |x|
1 2
=p p <p .
|x| + 2 + |x| |x|
56 CHAPTER 1. LIMIT

2 1
By limx→∞ p = 2 limx→∞ p = 0 and the sandwich rule, we get
|x| |x|
p p
lim ( |x| + 2 − |x|) = 0.
x→+∞

Example 1.5.7. The following computation

1 1
lim x sin = lim x lim sin
x→0 x x→0 x→0 x
is wrong because limx→0 sin x1 diverges. However, if we use
1
−|x| ≤ x sin ≤ |x|, lim |x| = 0,
x x→0

1
and the sandwich rule, then we get limx→0 x sin = 0.
x

Exercise 1.5.1. Explain that limx→a f (x) = l if and only if limx→a (f (x) − l) = 0.

Exercise 1.5.2. Use the sandwich rule to prove that limx→a |f (x)| = 0 implies limx→a f (x) =
0.

Exercise 1.5.3. Find the limits.

x x2 − 1
1. limx→∞ . 4. limx→∞ .
x+2 x2 + x − 2
x x2 − 1
2. limx→2 . 5. limx→0 .
x+2 x2 + x − 2
x x2 − 1
3. limx→−2 . 6. limx→1 .
x+2 x2 + x − 2

Exercise 1.5.4. Find the limits.

1 p
4. limx→∞ ( (x + a)(x + b) − x).
1. limx→0 x cos .
x2 r
x+a
1 5. limx→∞ .
2. limx→∞ p sin x2 . x+b
|x| √ √ !
3 3
x2 x2
p p 6. limx→∞ √ −√ .
3. limx→∞ ( |x| + a − |x| + b). 3
x+a 3
x+b

1.5.2 Limit of Composition Function

Suppose three variables x, y, z are related by
f g
x→
− y = f (x) →
− z = g(y).
1.5. LIMIT OF FUNCTION 57

Then z and x are related by z = g(f (x)), the composition of two functions. Suppose
both f and g have limits

lim f (x) = b, lim g(y) = c,

x→a y→b

where b is the value of the first limit as well as the location of the second limit. The
problem is whether the composition has limit

lim g(f (x)) = c.

x→a

What we want is to combine two implications

x → a, x 6= a =⇒ y → b,
y → b, y =
6 b =⇒ z → c,

to get the third implication

x → a, x 6= a =⇒ z → c.

However, the two implications cannot be combined as is, because “y → b” does not
imply “y → b, y 6= b”. There are two ways to save this. The first is to strengthen
the first implication to

x → a, x 6= a =⇒ y → b, y 6= b.

Here the extra condition is f (x) 6= b for x near a and x 6= a. The second is to
strengthen the second implication to

y → b =⇒ z → c.

Here the extra condition is y = b =⇒ z → c. Since y = b implies z = g(b), the

extra condition is simply c = g(b), so the strengthened second implication becomes

y → b =⇒ z → c = g(b).

Proposition 1.5.5 (Composition Rule). Suppose limx→a f (x) = b and limy→b g(y) =
c. Then we have limx→a g(f (x)) = c, provided one of the following extra conditions
is satisfied
1. f (x) 6= b for x near a and x 6= a.

2. c = g(b).

Note that the second condition means

lim g(y) = g(b).

y→b
58 CHAPTER 1. LIMIT

Later on, this will become the definition of the continuity of g(y) at b. Moreover,
the composition rule in this case means
 
lim g(f (x)) = c = g(b) = g lim f (x) .
x→a x→a

So the continuity of g(y) is the same as the exchangeability of the function g and
the limit.
The composition rule extends Proposition 1.1.6 because a sequence xn can be
considered as a function
xn : n 7→ x(n) = xn .
Then a subsequence can be considered as a composition with a function nk : N → N
that satisfies limk→∞ nk = ∞

k 7→ n(k) = nk 7→ x(n(k)) = xnk .

Example 1.5.8. We have

√ |x − 1|
0 ≤ | x − 1| = √ ≤ |x − 1|.
x+1

Note that |x − 1| is the composition of z = |y| and y = x − 1. By limx→1 (x − 1) =

1 − 1 = 0, limy→0 |y| = 0 = |0|, and the composition rule (both extra conditions
√ are
satisfied), we get limx→1 |x−1| =
√ 0. Then by the sandwich rule, we get limx→1 | x−
1| = 0. This implies limx→1 (√ x − 1) = 0 (see Exercise 1.5.2). Finally, by the
arithmetic rule, we get limx→1 x = 1.

Example 1.5.9. We have limx→1 (3x3 − 2) = 1 from the arithmetic rule. We also know
√
limy→1 y = 1 from Example 1.5.8. The composition
√ √
x 7→ y = f (x) = 3x3 − 2 7→ z = g(y) = y = g(f (x)) = 3x3 − 2
√
should give us limx→1 3x3 − 2 = 1.
We need to verify one of the extra conditions in the composition rule. If x is
close to 1 and x < 1, then we have x3 < 13 = 1, so that 3x3 − 2 < 1. Similarly, if x
is close to 1 and x > 1, then 3x3 − 2 > 1. Therefore for x close to 1 and 6= 1, we
indeed have 3x3 − 2 6= 1. This verifies the first condition.
Although the validity of the first condition already
√ allows us to apply the com-
√
position rule, the second condition limy→1 y = 1 = 1 is also valid.
Note that it is rather tempting to write
√ √ √
lim 3x3 − 2 = 3lim 3x3 − 2 = lim y.
x→1 3x −2→1 y→1

In other words, the composition rule appears simply as a change of variable. How-
ever, one needs to be careful because hidden in the definition of limx→1 is x 6= 1.
1.5. LIMIT OF FUNCTION 59

Similarly, the assumption 3x3 − 2 6= 1 is implicit in writing lim3x3 −2→1 , and the first
equality above actually requires you to establish

x → 1, x 6= 1 ⇐⇒ 3x3 − 2 → 1, 3x3 − 2 6= 1.

This turns out to be true in our specific example, but might fail for other examples.
√ √
Example 1.5.10. Example 1.5.8 can be extended to show that √ limx→a x = a for
any a > 0 (see Exercise 1.5.5). This actually means that x is continuous, and
therefore we may apply the composition rule to get

√ √
q r 
lim 3
3x − 2 + 7x = lim 3x3 − 2 + 7x
x→1 x→1
√
r
= lim 3x3 − 2 + lim 7x
x→1 x→1
rq
= lim (3x3 − 2) + lim 7x
x→1 x→1
√
q
= 3 · 12 + 1 + 7 · 1.

Here is the detailed reason. The last equality is by√the arithmetic rule. The third
equality makes use of the continuity of the function x and the composition rule to
move the limit from outside the square root to inside the√square root. The second
equality is by the arithmetic rule. Once we know limx→1 ( 3x3 − 2 + 7x) converges
√
to a positive number, the first equality then follows from the continuity of x and
the composition rule.
√ √
Exercise 1.5.5. Show that limx→a x= a for any a > 0.
√ √
Exercise 1.5.6. Show that limx→a 3
x = 3 a for any a 6= 0.

Exercise 1.5.7. Find the limits, a, b > 0.

√ r r 
1. limx→0 a + x. a+x a
3. limx→0 − .
a a+x
r r 
1 √ √ 1 a+x a
2. limx→0 ( a + x − a − x). 4. limx→0 − .
x x b b+x

Example 1.5.11. A change of variable can often be applied to limits. For example,
we have
lim f (x2 ) = lim2 f (y), for a 6= 0.
x→a y→a

The equality means that the two limits are equivalent.

60 CHAPTER 1. LIMIT

Suppose limy→a2 f (y) = l. Then f (x2 ) is the composition of f (y) with y = x2 ,

and by a 6= 0, the first condition is satisfied
x → a, x 6= a =⇒ y → a2 , y 6= a2 .
Therefore the composition rule can be applied to give limx→a f (x2 ) = l.
Conversely, suppose limx→a f (x2 ) = l. For x near a > 0, f (y) is the composition
√
of f (x2 ) with x = y, the first condition is satisfied
y → a2 , y 6= a2 =⇒ x → a, x 6= a.
Therefore the composition rule can be applied to give limy→a2 f (y) = l. For the case
√
a < 0, the similar argument with x = − y works, and the composition rule still
gives limy→a2 f (y) = l.
Note that if a = 0, the function y = x2 is not invertible near x = 0, so you
cannot solve y for x.
We note that we cannot verify the second condition in the problem above because
not much is assumed about f . In particular, f is not assumed to be continuous.
Here are more examples of equivalent limits
 
3 1
lim f (x) = lim f (y − a), lim f (x ) = lim3 f (y), lim f (x) = lim f ,
x→a y→0 x→a y→a x→∞ y→0 y
where the first condition for the composition rule is satisfied in both directions
x → a, x 6= a ⇐⇒ x − a → 0, x − a 6= 0;
x → a, x =6 a ⇐⇒ x3 → a3 , x3 6= a3 ;
1
x → ∞ ⇐⇒ → 0.
x
1
In the last equivalence, we automatically have x 6= ∞ and 6= 0.
x

Example 1.5.12. The composition rule fails when neither conditions are satisfied,
which means that f (x) = b for some x 6= a arbitrarily close to a, and c 6= g(b).
For a concrete example, consider
(
1 y, if y 6= 0,
f (x) = x sin , g(y) =
x A, if y = 0.
We have limx→0 f (x) = 0 (see Example 1.5.7) and limy→0 g(y) = 0 (see Example
1.5.2). This means a = b = c = 0. However, the composition is

x sin 1 , if x 6= (nπ)−1 ,
g(f (x)) = x
A, if x = (nπ)−1 .
and limx→0 g(f (x)) converges if and only if g(0) = A = 0 = c.

Exercise 1.5.8. Rewrite the limits as limx→c f (x) for suitable c.

1.5. LIMIT OF FUNCTION 61

1. limx→b f (x − a). 3. limx→0 f (ax + b). 5. limx→−1 f (x2 + x).

√
2. limx→∞ f (x + a). 4. limx→1 f ( x). 6. limx→a f (x2 + x).

1.5.3 One Sided Limit

In the sequence limit limn→∞ xn , the subscript n has only the positive infinity
direction to go. The function limit can have various directions. For example,
limx→+∞ f (x) = l means

x → ∞, x > 0 =⇒ f (x) → l.

Moreover, the right limit limx→a+ f (x) = l means

x → a, x > a =⇒ f (x) → l.

Similarly, the left limit limx→a− f (x) = l means

x → a, x < a =⇒ f (x) → l.

All the properties of the usual (two sided) limits still hold for one sided limits.
Moreover, we have the following relation (for a = ∞, a± means ±∞).

Proposition 1.5.6. limx→a f (x) = l if and only if limx→a+ f (x) = limx→a− f (x) = l.

Example 1.5.13. For the sign function


1,
 if x > 0,
f (x) = 0, if x = 0,

−1, if x < 0,


we have limx→0+ f (x) = 1 and limx→0− f (x) = −1. Since the two limits are not
equal, limx→0 f (x) diverges.

Figure 1.5.2: Sign function.

62 CHAPTER 1. LIMIT

|x + 4|
Example 1.5.14. To find limx→∞ , we consider the limit at +∞ and −∞. For
x
|x + 4| x+4 1
x > 0, we have = = 1 + 4 . By the arithmetic rule, we have
x x x
 
|x + 4| 1 1
lim = lim 1 + 4 = 1 + 4 lim = 1.
x→+∞ x x→+∞ x x→+∞ x
|x + 4| x+4 1
For x < −4, we have =− = −1 − 4 . By the arithmetic rule, we have
x x x
 
|x + 4| 1 1
lim = lim −1 − 4 = −1 − 4 lim = −1.
x→−∞ x x→−∞ x x→−∞ x

|x + 4|
Since the two limits are different, limx→∞ diverges.
x

Example 1.5.15. If we apply the argument in Example 1.5.6 to x > 0, then we get
√ √
lim ( x + 2 − x) = 0.
x→+∞

If we apply the argument to x < 0, then we get

√ √
lim ( 2 − x − −x) = 0.
x→−∞

Exercise 1.5.9. Find the limits at 0.

( ( (
1, if x < 0, 1, if x 6= 0, x, if x < 0,
1. . 2. . 3. .
2, if x > 0. 2, if x = 0. −x2 , if x > 0.

Exercise 1.5.10. Find the limits.

1 √ √ √
1. limx→+∞ √ sin x2 . 3. limx→+∞ x( x + a − x + b).
x+a √ √ √
√ √ x− a+ x−a
2. limx→+∞ ( x + a − x + b). 4. limx→a+ √ , a > 0.
x2 − a2

Example 1.5.16. The composition rule can also be applied to one sided limit. For
√
example, we have limx→0+ f (x2 ) = limx→0+ f (x) by introducing y = x2 and x = y,
and the first condition for the composition rule is satisfied in both directions
x → 0, x > 0 ⇐⇒ y → 0, y > 0.
We also have limx→0− f (x2 ) = limx→0+ f (x) by
√
x = − y → 0, x < 0 ⇐⇒ y = x2 → 0, y > 0.
Then we may use Proposition 1.5.6 to conclude that limx→0 f (x2 ) = limx→0+ f (x).

Exercise 1.5.11. Rewrite the limits as limx→c f (x) for suitable c.

1.5. LIMIT OF FUNCTION 63

1. limx→b+ f (a − x). 3. limx→0+ f (ax + b). 5. limx→(−1)− f (x2 + x).

√
2. limx→−∞ f (x + a). 4. limx→0+ f ( x). 6. limx→a− f (x2 + x).

Example 1.5.17. We show that

lim xp = ap , for any a > 0 and any p.

x→a

First consider the special case a = 1. We find integers M and N satisfying

M < p < N . Then for x > 1, we have xM < xp < xN . By the arithmetic rule, we
have limx→1+ xM = (limx→1+ x)M = 1M = 1. Similarly, we have limx→1+ xN = 1.
Then by the sandwich rule, we get limx→1+ xp = 1.
For 0 < x < 1, we have xM > xp > xN . Again, we have limx→1− xM =
limx→1− xN = 1 by the arithmetic rule. Then we get limx→1− xp = 1 by the sandwich
rule.
Combining limx→1+ xp = 1 and limx→1− xp = 1, we get limx→1 xp = 1.
x
For general a > 0, we move the problem from a to 1 by introducing y =
a
lim xp = lim ap y p = ap lim y p = ap 1 = ap .
x→a y→1 y→1

Specifically, we first use limx→1 xp = 1, which we just proved, in the third equality.
Then we use the arithmetic rule to get the second equality. Finally, the first equality
is obtained by a change of variable, which is essentially the composition rule.

Exercise 1.5.12. Suppose limx→a f (x) = l > 0. Prove that limx→a f (x)p = lp . This extends
Exercise 1.1.35 to function limit.

1.5.4 Limit of Trigonometric Function

π
The sine and tangent functions are defined for 0 < x < by Figure 1.5.3. The
2
lengths of line AB, arc BC, and line CD are respectively sin x, x, and tan x. Since
the length of line AB is smaller than the length of line BC, which is further smaller
than the length of arc BC, we get
π
0 < sin x < x, for 0 < x < .
2
By limx→0+ x = 0 and the sandwich rule, we get

lim sin x = 0.
x→0+

Changing x to y = −x (i.e., applying the composition rule) gives the left limit

lim sin x = lim+ sin(−y) = lim+ (− sin y) = − lim+ sin y = 0.

x→0− y→0 y→0 y→0
64 CHAPTER 1. LIMIT

Combining the two one sided limits at 0, we conclude

lim sin x = 0.
x→0

Then we further get

 x 2
 x 2
lim cos x = lim 1 − 2 sin = 1 − 2 lim sin = 1 − 2 · 02 = 1.
x→0 x→0 2 x→0 2

1
D
B

tan x
sin x
x
O A C

Figure 1.5.3: Trigonometric function.

1 1
Note that the area of fan OBC is x, and the area of triangle ODC is tan x.
2 2
1 1
Since the fan is contained in the triangle, we get x < tan x, which is the same
2 2
sin x
as cos x < . Combined with 0 < sin x < x obtained before, we get
x
sin x π
cos x < < 1, for 0 < x < .
x 2
sin x
By limx→0+ cos x = 1 and the sandwich rule, we get limx→0+ = 1. Then
x
changing x to y = −x gives the left limit
sin x sin(−y) sin y
lim− = lim+ = lim+ = 1.
x→0 x y→0 −y y→0 y
Therefore we conclude
sin x
lim = 1.
x→0 x
This further implies
x
cos x − 1 −2 sin2 −2 sin2 y 1

sin y
2
1
lim = lim 2 = lim = − lim =− ,
x→0 x 2 x→0 x 2 y→0 (2y) 2 2 y→0 y 2
cos x − 1 cos x − 1 1
lim = lim 2
x = − · 0 = 0,
x→0 x x→0 x 2
tan x sin x 1 1
lim = lim = 1 · = 1.
x→0 x x→0 x cos x 1
1.5. LIMIT OF FUNCTION 65

We may get the limit of trigonometric functions at any a by using y = x + a to

move the limit to be at 0.
lim sin x = lim sin(a + y) = lim (sin a cos y + cos a sin y)
x→a y→0 y→0

= sin a lim cos y + cos a lim sin y = (sin a)1 + (cos a)0 = sin a,
y→0 y→0

lim cos x = lim cos(a + y) = lim (cos a cos y − sin a sin y)

x→a y→0 y→0

= cos a lim cos y − sin a lim sin y = (cos a)1 + (sin a)0 = cos a,
y→0 y→0
limx→a sin x sin a
lim tan x = = = tan a, if cos a 6= 0.
x→a limx→a cos x cos a

Example 1.5.18. By the arithmetic rule and the composition rule, we have
 π
cos y +
limπ
cos x
= lim 2 = lim − sin y = −1,
x→ 2
π y→0 y y→0 y
x−
2
1 1
lim x sin = lim sin y = 1,
x→∞ x y→0 y
sin x − sin a sin(a + y) − sin a
lim = lim
x→a x−a y→0 y
 
cos y − 1 sin y
= lim sin a + cos a = cos a.
y→0 y y

Exercise 1.5.13. Find the limit.

sin πx tan x − tan a
1. limx→−1 . 8. limx→a .
x+1 x−a
tan x − 1
2. limx→ π4 . cos(2x + 1) − cos(2x − 1)
4x − π 9. limx→0 .
x2
sin x − cos x
3. limx→ π4 . √ √
4x − π 10. limx→+∞ x(sin x + 2 − sin x).
√
sin πx − 3 cos πx
4. limx→ 1 . tan(sin x)
3 3x − 1 11. limx→0 .
1 x
5. limx→∞ x tan .
x tan(sin x)

1
 12. limx→0 .
6. limx→∞ x cos − 1 . sin(tan x)
x
cos x − cos a tan(sin x)
7. limx→a . 13. limx→0+ √ .
x−a sin x

Exercise 1.5.14. Study the limit of the sequences sin(sin(sin . . . a))) and cos(cos(cos . . . a))),
where the trigonometric functions are applied n times.
66 CHAPTER 1. LIMIT

1.6 Rigorous Definition of Function Limit

The sequence limit is the behavior as n approaches infinity, which is described by
n > N (N is the measurement of bigness). The function limit is the behavior as x
approaches a but not equal to a, which may be described by 0 < |x − a| < δ (δ is
the measurement of smallness). By replacing n > N with 0 < |x − a| < δ in the
rigorous definition of limn→∞ xn , we get the rigorous definition of limx→a f (x) = l
for finite a and l.

Definition 1.6.1. A function f (x) converges to a finite number l at a if for any

 > 0, there is δ > 0, such that 0 < |x − a| < δ implies |f (x) − l| < .

The meaning of the definition is given by Figure 1.6. The other limits can be
similarly defined. For example, limx→a+ f (x) = l means that for any  > 0, there is
δ > 0, such that
0 < x − a < δ =⇒ |f (x) − l| < .

The limit limx→∞ f (x) = l means that for any  > 0, there is N , such that

|x| > N =⇒ |f (x) − l| < .

Moreover, limx→a− f (x) = +∞ means that for any B, there is δ > 0, such that

−δ < x − a < 0 =⇒ f (x) > B.

l+
l
l−

a−δ a a+δ

Figure 1.6.1: 0 < |x − a| < δ implies |f (x) − l| < .

Exercise 1.6.1. Write down the rigorous definitions of limx→a f (x) = −∞, limx→a− f (x) =
l, and limx→+∞ f (x) = −∞.
1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 67

1.6.1 Rigorous Proof of Basic Limits

Example 1.6.1. Here is the rigorous reason for limx→a c = c. For any  > 0, take
δ = 1. Then
0 < |x − a| < δ = 1 =⇒ |c − c| = 0 < .
Here is the rigorous reason for limx→a x = a. For any  > 0, take δ = . Then

0 < |x − a| < δ =  =⇒ |x − a| < .

Example 1.6.2. To prove limx→1 x2 = 1 rigorously means that, for any  > 0, we need
to find suitable δ > 0, such that

0 < |x − 1| < δ =⇒ |x2 − 1| < .

We have

0 < |x − 1| < δ =⇒ |x2 − 1| = |x + 1||x − 1| ≤ |x + 1|δ.

Moreover, when x is close to 1, we expect x + 1 to be close to 2. Such intuition can

be made rigorous by

0 < |x − 1| < 1 =⇒ |x + 1| ≤ |x − 1| + 2 < 3.

Therefore

0 < |x − 1| < δ, 0 < |x − 1| < 1 =⇒ |x2 − 1| ≤ |x + 1|δ < 3δ.

To complete the rigorous proof, we only need to make sure δ ≤ 1 and 3δ ≤ .

The analysis
n  o above suggests the following rigorous proof. For any  > 0, choose
δ = min 1, . Then
3
0 < |x − 1| < δ =⇒ |x − 1| < 1, 3|x − 1| < 
=⇒ |x + 1| ≤ |x − 1| + 2 < 3, 3|x − 1| < 
=⇒ |x2 − 1| = |(x + 1)(x − 1)| ≤ 3|x − 1| < .

1
Example 1.6.3. To rigorously prove limx→1 = 1, we note
x
 
1
 − 1 = 1 |x − 1|.

x  |x|

When x is close to 1, we know |x − 1| is very small. We also know that |x| is close
1
to 1, so that can be controlled by a specific bound. Combining the two facts,
|x|
1
we see that |x − 1| can be very small. Of course concrete and specific estimation
|x|
68 CHAPTER 1. LIMIT

1 1 1
is needed to get a rigorous proof. If |x − 1| < , then x > and < 2. Therefore
2 2 |x|
 1
if we also have |x − 1| < , then we get |x − 1| < .
2 |x|
The analysis
  above suggests the following rigorous proof. For any  > 0, choose
1 
δ = min , . Then
2 2

1 
0 < |x − 1| < δ =⇒ |x − 1| < , |x − 1| <
2 2
1 1 
=⇒ x > 1 − = , |x − 1| <
  2 2 2
1  1
=⇒  − 1 = |x − 1| ≤ 2|x − 1| < .
x |x|

Example 1.6.4. We prove

lim xp = 0, for any p > 0.

x→0+

1
For any  > 0, choose δ =  p . Then by p > 0, we have

0 < x < δ =⇒ |xp − 0| = xp < δ p = .

We also prove
lim xp = +∞, for any p > 0.
x→+∞
1
For any B > 0, choose N = B p . Then by p > 0, we have

x > N =⇒ xp > N p = B.

Combined with Example 1.5.17, we get

(
ap , if p > 0, 0 < a < +∞ or a = 0+ ,
lim xp =
x→a +∞, if p > 0, a = +∞.

1
Changing x to , we get the similar conclusions for p < 0
x

p
a ,
 if p < 0, 0 < a < +∞,
p
lim x = +∞, if p < 0, a = 0+ ,
x→a 
0, if p < 0, a = +∞.

1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 69

Example 1.6.5. We try to rigorously prove the limit in Example 1.5.5. Example 1.2.6
gives the rigorous proof for the limit of the similar sequence. Instead of copying the
proof, we make a slightly different estimation of |f (x) − l|
 2x2 + x
   
  3x − 2  4|x| 12
 x2 − x + 1 − 2 =  x2 − x + 1  < x2 = |x| .
   

3
The crucial inequality is based on the intuition that |3x−2| < 4|x| and |x2 −x+1| >
x2
for sufficiently big x. The first inequality is satisfied when 2 < |x|, and the second
3
x2 x2 x2
is satisfied when |x| < and 1 < . It is easy to see that 2 < |x|, |x| < and
3 3 3
2
x 12
1< are all satisfied when |x| > 4. Therefore |f (x) − l| < when |x| > 4.
3   |x|
12
Formally, for any  > 0, choose N = max 4, . Then

12
|x| > N =⇒ |x| > 4, |x| >

x2 12
=⇒ |3x − 2| < 4|x|, |x2 − x + 1| > , <
3 |x|
2
   
 2x + x   3x − 2  4|x| 12
=⇒  2 − 2 =  2 <
2 = < .
x −x+1 x − x + 1 x |x|
3

Exercise 1.6.2. Extend the proof of limx→a c = c and limx→a x = a to the case a = ±∞.

Exercise 1.6.3. Rigorously prove the limits.

√ x+1
1. limx→4 x = 2. 4. limx→−∞ √ = −1.
x2 + 1
1 1
2. limx→2 = . x + sin x
x 2 5. limx→∞ = 1.
x + cos x
3x2 − 2x + 1 sin x
3. limx→∞ = 3. 6. limx→∞ = 0.
x2 + 3x − 1 x

Exercise 1.6.4. Rigorously prove the limits. For the first three limits, give direct proof
instead of using Example 1.5.17.
x
1. limx→a x2 = a2 . 4. limx→∞ = 1.
x+a
√ √ √ √
2. limx→a x = a, a > 0. 5. limx→+∞ ( x + a − x + b) = 0.
1 1 r
x+a
3. limx→a = , a 6= 0. 6. limx→∞ = 1.
x a x+b
70 CHAPTER 1. LIMIT

Exercise 1.6.5. Suppose f (x) ≥ 0 for x near a and limx→a f (x) = 0. Suppose g(x) ≥ c
for x near a and constant c > 0. Prove that limx→a f (x)g(x) = 0. This extends Exercise
1.1.34 to the function limit.

1.6.2 Rigorous Proof of Properties of Limit

Example 1.6.6. The arithmetic rule limx→a (f (x) + g(x)) = limx→a f (x) + limx→a g(x)
in Proposition 1.5.2 can be proved in the same way as Example 1.2.18. For any

 > 0, apply the definition of limx→a f (x) = k and limx→a g(x) = l to > 0. We
2
find δ1 and δ2 , such that

0 < |x − a| < δ1 =⇒ |f (x) − l| < ,
2

0 < |x − a| < δ2 =⇒ |g(x) − k| < .
2
Then

0 < |x − a| < δ = min{δ1 , δ2 }

 
=⇒ |f (x) − l| < , |g(x) − k| <
2 2
 
=⇒ |(f (x) + g(x)) − (l + k)| ≤ |f (x) − l| + |g(x) − k| < + = .
2 2

Example 1.6.7. We prove the extended arithmetic rule (+∞)·l = +∞ for l > 0. This
means that, if limx→a f (x) = +∞ and limx→a g(x) = l > 0, then limx→a f (x)g(x) =
+∞.
For any B, there is δ1 > 0, such that
2
0 < |x − a| < δ1 =⇒ f (x) > B.
l
l
For  = > 0, there is δ2 > 0, such that
2
l l
0 < |x − a| < δ2 =⇒ |g(x) − l| < =⇒ g(x) > .
2 2
Combining the two implications, we get
2 l
0 < |x − a| < δ = min{δ1 , δ2 } =⇒ f (x)g(x) > B · = B.
l 2
This completes the proof of (+∞) · l = +∞.
As an application of the extended arithmetic rule, we have
 
3 3 3 1
lim (x − 3x + 1) = lim x lim 1 − 2 + 3 = (+∞) · 1 = +∞.
x→+∞ x→+∞ x→+∞ x x
1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 71

In general, we have
(
+∞, if an > 0,
lim (an xn + an−1 xn−1 + · · · + a1 x + a0 ) =
x→+∞ −∞, if an < 0.

Example 1.6.8. We prove Proposition 1.5.6 about one sided limits.

Assume limx→a f (x) = l. Then for any  > 0, there is δ > 0, such that
0 < |x − a| < δ =⇒ |f (x) − l| < .
The implication is the same as the following two implications
0 < x − a < δ =⇒ |f (x) − l| < ,
−δ < x − a < 0 =⇒ |f (x) − l| < .
These are exactly the definitions of limx→a+ f (x) = l and limx→a− f (x) = l.
Conversely, assume limx→a+ f (x) = limx→a− f (x) = l. Then for any  > 0, there
are δ+ , δ− > 0, such that
0 < x − a < δ+ =⇒ |f (x) − l| < ,
−δ− < x − a < 0 =⇒ |f (x) − l| < .
Therefore
0 < |x − a| < min{δ+ , δ− } =⇒ 0 < x − a < δ+ or − δ− < x − a < 0
=⇒ |f (x) − l| < .
This proves that limx→a f (x) = l.

Example 1.6.9. We prove the second case of the composition rule in Proposition 1.5.5.
In other words, limx→a f (x) = b and limy→b g(y) = g(b) imply limx→a g(f (x)) = g(b).
By limy→b g(y) = g(b), for any  > 0, there is µ > 0, such that
0 < |b − y| < µ =⇒ |g(y) − g(b)| < .
Since the right side also holds when y = b, we actually have
|y − b| < µ =⇒ |g(y) − g(b)| < . (1.6.1)
On the other hand, by limx→a f (x) = b, for the µ > 0 just found above, there is
δ > 0, such that
0 < |x − a| < δ =⇒ |f (x) − b| < µ.
Then we get
0 < |x − a| < δ =⇒ |f (x) − b| < µ
=⇒ |g(f (x)) − g(b)| < .
In the second step, we apply the implication (1.6.1) to y = f (x). This completes
the proof that limx→a g(f (x)) = g(b).
72 CHAPTER 1. LIMIT

Example 1.6.10. In Example 1.5.16, we argued that limx→0 f (x2 ) = limx→0+ f (x).
Now we give rigorous proof.
Suppose limx→0 f (x2 ) = l. Then for any  > 0, there is δ > 0, such that

0 < |x| < δ =⇒ |f (x2 ) − l| < .

√
Now for any y > 0, let x = y. Then
√
0 < y < δ 2 =⇒ 0 < x = y < δ =⇒ |f (y) − l| = |f (x2 ) − l| < .

This proves limy→0+ f (y) = l.

On the other hand, suppose limx→0+ f (x) = l. Then for any  > 0, there is δ > 0,
such that
0 < x < δ =⇒ |f (x) − l| < .
Now for any y, let x = y 2 . Then
√
0 < |y| < δ =⇒ 0 < x = y 2 < δ =⇒ |f (y 2 ) − l| = |f (x) − l| < .

This proves limy→0 f (y 2 ) = l.

Exercise 1.6.6. Prove the arithmetic rule limx→a f (x)g(x) = limx→a f (x) limx→a g(x) in
Proposition 1.5.2.

Exercise 1.6.7. Prove the sandwich rule in Proposition 1.5.3.

Exercise 1.6.8. Prove the order rule in Proposition 1.5.4.

Exercise 1.6.9. Prove the first case of the composition rule in Proposition 1.5.5.

l
Exercise 1.6.10. Prove the extended arithmetic rules (−∞) + l = −∞ and = 0 for
∞
function limit.

Exercise 1.6.11. For a subset A of R, define limx∈A,x→a f (x) = l if for any  > 0, there is
δ > 0, such that
x ∈ A, 0 < |x − a| < δ =⇒ |f (x) − l| < .
Suppose A ∪ B contains all the points near a and 6= a. Prove that limx→a f (x) = l if and
only if limx∈A,x→a f (x) = l = limx∈B,x→a f (x).

Exercise 1.6.12. Prove that limx→a f (x) = l implies limx→a max{f (x), l} = l and
limx→a min{f (x), l} = l. Can you state and prove the sequence version of the result?

Exercise 1.6.13. Suppose f (x) ≤ 1 for x near a and limx→a f (x) = 1. Suppose g(x) is
bounded near a. Prove that limx→a f (x)g(x) = 1. What about the case f (x) ≥ 1?
1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 73

Exercise 1.6.14. Use Exercises 1.6.12, 1.6.13 and the sandwich rule to prove that, if limx→a f (x) =
1 and g(x) is bounded near a, then limx→a f (x)g(x) = 1. This is the function version of
Exercise 1.1.34. An alternative method for doing the exercise is by extending Proposition
1.1.6.

1.6.3 Relation to Sequence Limit

The sequence limit and the function limit are related.

Proposition 1.6.2. limx→a f (x) = l if and only if

lim xn = a, xn 6= a =⇒ lim f (xn ) = l.

n→∞ n→∞

The necessary direction means that if the whole function converges to l at a,

then the restriction of the function to any sequence converging to (but not equal to)
a also converges to l. The sufficiency direction means that if all such restrictions
converge to l, then the original function converges to l.
The necessary direction can also be considered as a version of the composition
rule because xn can be considered as a function x(n) with n as variable, and f (xn )
is a composition
n 7→ x = xn 7→ f (x) = f (xn ).
By analogy with the first case of Proposition 1.5.5, we have

lim xn = a, xn 6= a (at least for big n), and lim f (x) = l =⇒ lim f (xn ) = l.
n→∞ x→a n→∞

Example 1.6.11. By taking a = +∞ in Proposition 1.6.2, the limit in Example 1.5.5

implies the limit in Example 1.1.2.

sin x 1 1
Example 1.6.12. From limx→0 = 1 and limn→∞ = limn→∞ √ = 0, we get
x n n

1 1
sin sin √
1 n = 1, √ 1 n
lim n sin = lim lim n sin √ = lim = 1.
n→∞ n n→∞ 1 n→∞ n n→∞ √1
n n

Example 1.6.13. A consequence of Proposition 1.6.2 is that, if the restrictions of

f (x) to two sequences converging to a converge to different limits, then limx→a f (x)
diverges.
The Dirichlet function is
(
1, if x ∈ Q,
D(x) =
0, if x ∈
/ Q.
74 CHAPTER 1. LIMIT

For any a, we have a rational sequence xn converging to a, and the restriction of

D(x) to the sequence is D(xn ) = 1, converging to 1. We also have an irrational
sequence converging to a, and the restriction of D(x) converges to 0. Therefore
D(x) diverges everywhere.

Q Q

R−Q R−Q

y = D(x) y = xD(x)

Figure 1.6.2: Dirichlet function.

By the same reason, the function xD(x) diverges at everywhere except 0. On

the other hand, we have −|x| ≤ |xD(x)| ≤ |x|. By limx→0 |x| = 0 and the sandwich
rule, we have limx→0 xD(x) = 0. So xD(x) converges only at 0.

If limn→∞ f (xn ) = l for one (instead of all) sequence xn converging to a, then

it suggests (but does not necessarily imply) that limx→+∞ f (x) = l. Sometimes we
can “fill the gap” between xn and derive limx→+∞ f (x) = l in general.

Example 1.6.14. The limit limn→∞ an = 0 in Example 1.1.11 suggests

lim ax = 0, for 0 ≤ a < 1.
x→+∞

Here we do not consider −1 < a < 0 because ax is not always defined.

For rigorous proof, we compare ax with an for a natural number n near x. Specif-
ically, for any x > 1, we have n ≤ x ≤ n + 1 for some natural number n. Then
0 ≤ a < 1 implies
0 ≤ ax < an .
By limn→∞ an = 0, the sandwich rule should imply limx→+∞ ax = 0. However, we
cannot quote the sandwich rule directly because we have a function sandwiched by
a sequence. We have to repeat the proof of the sandwich rule.
For any  > 0, by limn→∞ an = 0, there is N , such that
n > N =⇒ an < .
For x > N + 1, let n be a natural number satisfying n ≤ x ≤ n + 1. Then
x > N + 1 =⇒ n ≥ x − 1 > N =⇒ 0 ≤ ax ≤ an < .
1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 75

This rigorously proves limx→+∞ ax = 0.

1
For a > 1, let b = . Then 0 < b < 1 and by the arithmetic rule, we have
a
1 1
lim ax = = = +∞.
x→+∞ limx→+∞ bx 0+
Thus we have 
+∞, if a > 1,

x
lim a = 1, if a = 1,
x→+∞ 
0, if 0 < a < 1.


By using the composition rule to change x to −x, we also get


0,
 if a > 1,
x
lim a = 1, if a = 1,
x→−∞ 
+∞, if 0 < a < 1.


We emphasize that we cannot use the following argument at the moment

log 
x>N = = loga  =⇒ 0 < ax < .
log a
The reason is that our logical foundation only assumes the arithmetic operations
and the exponential operation of real numbers. The concept of logarithm must be
defined as the inverse operation of the exponential, and will be developed only after
we have a theory of inverse functions. See Example 1.7.15.

The above example suggests the following extension of the sandwich rule:

Proposition 1.6.3 (Extended Squeeze Theorem). Suppose yn ≤ f (x) ≤ zn for n ≤

x ≤ n + 1. If limn→∞ yn = limn→∞ zn = l, then limx→+∞ f (x) = l.

The proof is exactly the same as the above example. For any  > 0, there exists
N 0 such that n > N 0 implies
|yn − l| < , |zn − l| < 
Then if x > N = N 0 + 1, there exists n satisfying n ≤ x ≤ n + 1 and n ≥ x − 1 > N 0 ,
such that
l −  < yn ≤ f (x) ≤ zn < l + 
in other words, for x > N ,
|f (x) − l| < .

Exercise 1.6.15. Use the limit in Example 1.2.15 and the idea of Example 1.6.14 to prove
that limx→+∞ x2 ax = 0 for 0 < a < 1. Then use the sandwich rule to prove that
limx→+∞ xp ax = 0 for any p and 0 < a < 1.
76 CHAPTER 1. LIMIT

Exercise 1.6.16. State and prove the extended exponential rule

(
+∞, if a > 1,
a+∞ =
0, if 0 < a < 1.

√
Example 1.6.15. The limit limn→∞ n
n = 1 in Example 1.1.8 suggests that
1
lim x x = 1.
x→+∞

By the composition rule, this is the same as

lim xx = 1.
x→0+

For n ≤ x ≤ n + 1, we have the inequality

1 1 1 1 1
n n+1 ≤ n x ≤ x x ≤ (n + 1) x ≤ (n + 1) n
By Example 1.1.9 (not quite Example 1.1.8), we know
1 1
lim (n + 1) n = lim n n+1 = 1.
n→∞ n→∞

Hence by the Extended Squeeze Theorem, we conclude that

1
lim x x = lim+ xx = 1.
x→+∞ x→0

Example 1.6.16. We show that

lim ax = ab .
x→b
For the
√ special case b = 0, the limit is limx→0 ax = 1, and is closely related to
limn→∞ n a = 1 in Example 1.1.7. This suggests us to prove the special case using
the idea in Example 1.6.15. See Exercise 1.6.17. Alternatively, we may prove the
special case by comparing with limx→0+ xx = 1 in Example 1.6.15. For sufficiently
1
small x > 0, we have x < a < . This implies
x
1
xx < ax < x for sufficiently small x > 0.
x
1
Since Example 1.6.15 tells us that limx→0+ xx = 1 = limx→0+ x , by the sandwich
x
rule, we get limx→0+ ax = 1. For x < 0, we may use the change of variable y = −x
to get
1
lim− ax = lim+ a−y = = 1.
x→0 y→0 limy→0+ ay
Combing the left and right limits gives the special case limx→0 ax = 1.
The general case can be derived from the special case by introducing x = y + b
lim ax = lim ay+b = lim ay ab = ab lim ay = ab · 1 = ab .
x→b y→0 y→0 y→0
1.6. RIGOROUS DEFINITION OF FUNCTION LIMIT 77
√
Exercise 1.6.17. Prove limx→0 ax = 1 by comparing with limn→∞ n
a = 1.

Exercise 1.6.18. Prove that limn→∞ xn = l implies limn→∞ axn = al .

Exercise 1.6.19. Use Exercises 1.1.34 and 1.6.18 to prove that limn→∞ xn = l > 0 and
limn→∞ yn = k imply limn→∞ xynn = lk .

Exercise 1.6.20. Prove that limx→a f (x) = l implies limx→a bf (x) = bl .

Exercise 1.6.21. Use Exercises 1.6.14 and 1.6.20 to prove that limx→a f (x) = l > 0 and
limx→a g(x) = k imply limx→a f (x)g(x) = lk .

Example 1.6.17. Example 1.3.5 suggests that

1 x
 
lim 1 + = e.
x→∞ x
This is the same as
1
lim (1 + x) x = e.
x→0
Suppose n ≤ x ≤ n + 1. Then
n 
1 n 1 x 1 n+1 1 n+1
       
1
1+ ≤ 1+ ≤ 1+ ≤ 1+ ≤ 1+ .
n+1 x x x n
By Example 1.3.5, we have
 n+1
1
 n limn→∞ 1 +
1 n+1
lim 1 + =   = e,
n→∞ n+1 1
limn→∞ 1 +
n+1
 n+1  n  
1 1 1
lim 1 + = lim 1 + lim 1 + = e.
n→∞ n n→∞ n n→∞ n
1 x
 
Therefore by the Extended Squeeze Theorem, we have limx→+∞ 1 + = e.
x
Then we use the composition rule to further get
1 x 1 −x
     x
x
lim 1+ = lim 1− = lim
x→−∞ x x→+∞ x x→+∞ x − 1
 x−1
1 x
= lim 1+
x→+∞ x−1 x−1
 x−1
1 x
= lim 1+ lim
x→+∞ x−1 x→+∞ x−1
1 x
 
x
= lim 1+ lim
x→+∞ x x→+∞ x − 1
= e.
78 CHAPTER 1. LIMIT

Exercise 1.6.22. Find the limits.

 a x  2
1 x

1. limx→∞ 1 + .
x 4. limx→+∞ 1 + .
x
a bx  2
1 x
 
2. limx→∞ 1 + .
x 5. limx→−∞ 1 − .
x
1 x x+a x
   
3. limx→∞ 1 + 2 . 6. limx→∞ .
x x+b

Exercise 1.6.23. Find the limits.

1 n
   n+1  n+1
n n+1
1. 1 − . 6. 1 + 2 . 11. .
n n +1 n
n+1 (−1)n n
1 n
 
n n
 
2. 1 + . 7. 1 + 2 . 12. .
2n n + (−1)n n + (−1)n
n2 n+2 n
 
 n  
2 1 n+1
13. .
3. 1+ . 8. 1+ . n−2
n n
n2
n2
 
 a n   n−1 n+(−1)n

4. 1+ . 2 n−1 14. .
n 9. 1+ . n
n
n+1 n  n2 +(−1)n n
(−1)n
  
n n−1 n+1
5. 1+ 2 . 10. 1+ 2 . 15. .
n −1 n −1 n+2

1.6.4 More Properties of Function Limit

The following is the function version of Theorem 1.3.1.

Proposition 1.6.4. If f (x) converges at a, then |f (x)| ≤ B for a constant B and

all x near a.

1
Like the sequence limit, the example of sin at 0 shows that a bounded function
x
does not necessarily converge. However, Theorem 1.3.2 suggests that a bounded
monotone function should converge.

Definition 1.6.5. A function f (x) is increasing if

x1 < x2 =⇒ f (x1 ) ≤ f (x2 ).

It is strictly increasing if

x1 < x2 =⇒ f (x1 ) < f (x2 ).

1.7. CONTINUITY 79

The concepts of decreasing and strictly decreasing are similar, and a function is
(strictly) monotone if it is either (strictly) increasing or (strictly) decreasing.

Theorem 1.6.6. If f (x) is monotone and bounded on (a, a + δ), then limx→a+ f (x)
converges.

The theorem also holds for the left limit. The theorem can be proved by choosing
a decreasing sequence xn converging to a. Then f (xn ) is a bounded decreasing
sequence. By Theorem 1.3.2, we have limn→∞ f (xn ) = l. Then limx→a+ f (x) = l
can be proved by comparing with the sequence limit, similar to (actually simpler
than) Examples 1.6.15 and 1.6.17.
The Cauchy criterion in Theorem 1.3.3 can also be extended to functions.

Theorem 1.6.7 (Cauchy Criterion). The limit limx→a f (x) converges if and only if
for any  > 0, there is δ > 0, such that

0 < |x − a| < δ, 0 < |y − a| < δ =⇒ |f (x) − f (y)| < .

The criterion also holds for one sided limit. Again the proof starts by choosing a
sequence xn converging to a, then applying the sequence version of Cauchy criterion
(Theorem 1.3.3) to f (xn ), and then comparing limn→∞ f (xn ) and limx→a f (x).

Exercise 1.6.24. If f (x) is monotone and bounded on (a−δ, a)∪(a, a+δ), does limx→a f (x)
converge?

Exercise 1.6.25. Prove that if limx→a f (x) converges, then f (x) satisfies the Cauchy crite-
rion.

1.7 Continuity
A function is continuous at a if its graph is “not broken” at a. For example, the
function in Figure 1.7.1 is continuous at a2 and a5 , and we have limx→a2 f (x) = f (a2 )
and limx→a5 f (x) = f (a5 ). It is not continuous at the other ai for various reasons.
The function diverges at a1 and a7 because it has different left and right limits. The
function converges at a3 but the limit is not f (a3 ). The function diverges to infinity
at a4 . The function diverges at a6 because the left limit diverges.

Definition 1.7.1. A function f is continuous at a if limx→a f (x) = f (a).

The left side limx→a f (x) implies that the function should be defined for x near
a and x 6= a. The right side f (a) implies that the function is also defined at
a. Therefore the concept of continuity can only be applied to functions that are
80 CHAPTER 1. LIMIT

a1 a2 a3 a4 a5 a6 a7

Figure 1.7.1: Continuity and discontinuity.

defined near a and including a, which means all x satisfying |x − a| < δ for some
δ > 0. Then by the definition of limx→a f (x), it is easy to see that f (x) is continuous
at a if and only if for any  > 0, there is δ > 0, such that

|x − a| < δ =⇒ |f (x) − f (a)| < .

A function f (x) is right continuous at a if limx→a+ f (x) = f (a). The definition

can be applied to functions defined for x satisfying a ≤ x < a + δ for some δ > 0.
Similar remark can be made for the left continuity. A function is continuous at a if
and only if it is left and right continuous at a.
The function in Figure 1.7.1 is left continuous at a1 and right continuous at a6 ,
but is not continuous at the two points.
A function is continuous on an interval if it is continuous at every point of the
interval. For example, a function is continuous on [0, 1) if it is continuous at every
0 < a < 1 and is also right continuous at 0.

1.7.1 Meaning of Continuity

By Example 1.5.4, polynomials are continuous and rational functions are continuous
wherever it is defined. By Examples 1.5.17, the power function xp is continuous on
(0, ∞) for all p and, by Example 1.6.4, is right continuous at 0 for p > 0. In Section
1.5.4, we find that all trigonometric functions are continuous wherever they are
defined. By Example 1.6.16, the exponential function ax is continuous.
By the properties of limit, we know that the arithmetic combinations and compo-
sitions of continuous functions are continuous, wherever the new function is defined.
As remarked after Proposition 1.5.5, if f (x) is continuous at b and limx→a g(x) =
b, then  
lim f (g(x)) = f (b) = f lim g(x) .
x→a x→a

In other words, the continuity of f means that the limit and the evaluation of f can
be exchanged. By using Proposition 1.6.2 (another variant of the composition rule),
1.7. CONTINUITY 81

the same remark can be applied to a sequence limit limn→∞ xn = b instead of the
function limit limx→a g(x) = b, and we get
 
lim f (xn ) = f (b) = f lim xn .
n→∞ n→∞

Example 1.7.1. The sign function in Example 1.5.13 is continuous everywhere except
at 0. The Dirichlet function D(x) in Example 1.6.13 is not continuous anywhere.
The function xD(x) is continuous at 0 and not continuous at all the other places.

x3 − 1
Example 1.7.2. The function in Example 1.5.3 is not defined at x = 1, and we
x−1
cannot talk about its continuity at the point. In order to make the function contin-
x3 − 1
uous at 1, we need to assign the value of the function at 1 to be limx→1 = 3,
x−1
and we get a continuous function
 3
x − 1
, if x 6= 1
f (x) = x − 1 = x2 + x + 1.
3, if x = 1


√
Example 1.7.3. By the composition rule and the continuity of x, ax and sin x, we
have
√ q
lim x3 + 1 = lim x3 + 1 = 3,
x→2 x→2
n n
lim
lim a = an→∞ n2 −1 = 1,
n2 −1
n→∞
√ √
  q
lim− sin 1 − x = sin lim− 1 − x = sin lim− (1 − x2 ) = 0.
2 2
x→1 x→1 x→1

√
Example 1.7.4. √The continuity of x implies that, if limn→∞ xn = l > 0, then
√
limn→∞ xn = l. This is Example 1.2.17.
The continuity of ax implies
√ that, if limn→∞ xn = l, then limn→∞ axn = al . This
implies the limit limn→∞ a = 1 in√Example 1.2.14, and also implies the limits such
n

n+1
√1 √
as limn→∞ a n = 1 and limn→∞ a n−sin n = a.

Exercise 1.7.1. Determine the intervals on which the function is continuous. Is it possible
to extend to a continuous function at more points?
x2 − 3x + 2 3. sign(x). 5. xx .
1. .
x2 − 1
x2 − 1 1 cos x
2. . 4. x sin . 6. .
x−1 x 2x − π

Exercise 1.7.2. Find a function on R that is continuous at 1, 2, 3 and is not continuous at

all the other places.
82 CHAPTER 1. LIMIT

Exercise 1.7.3. Find a function on R that is not continuous at 1, 2, 3 and is continuous at

all the other places.

1 + f (0)g(x)
Exercise 1.7.4. Find two continuous functions f (x) and g(x), such that limx→0
1 + f (x)g(0)
converges but the value is not 1.

1.7.2 Intermediate Value Theorem

If we start at the sea level and climb to the mountain top of 1000 meters, then we
will be at 500 meters somewhere along the way, and will be at 700 meter some other
place. This is the intuition behind the following result.

Theorem 1.7.2 (Intermediate Value Theorem). If f (x) is continuous on [a, b], then
for any number γ between f (a) and f (b), there is c ∈ [a, b] satisfying f (c) = γ.

x
a c b

Figure 1.7.2: Intermediate value theorem.

Example 1.7.5. The polynomial f (x) = x3 − 3x + 1 is continuous and satisfies f (0) =

1, f (1) = −1. Therefore f (x) must attain value 0 somewhere on the interval [0, 1].
In other words, the polynomial has at least one root on (0, 1).
To find more precise location of the root, we may try to evaluate the function
at 0.1, 0.2, . . . , 0.9 and find f (0.3) = 0.727, f (0.4) = −0.136. This tells us that f (x)
has a root on (0.3, 0.4).
The discussion can be summarized as follows: If f is a continuous function on
[a, b], such that f (a) and f (b) have opposite signs, then f has at least one root in
(a, b).

Example 1.7.6. By Example 1.6.7, we have limx→+∞ (x3 − 3x + 1) = +∞ and

limx→−∞ (x3 − 3x + 1) = −∞. Therefore for sufficiently big b > 0, we have f (b) > 0
and f (−b) < 0. By the Intermediate Value Theorem, the polynomial has a root on
(−b, b).
In general, any odd order polynomial has at least one real root.
1.7. CONTINUITY 83

Back to f (x) = x3 −3x+1. In Example 1.7.5, we actually already know that f (x)
has at least one root on (0, 1). In fact, by f (−b) < 0, f (0) > 0, f (1) < 0, f (b) > 0,
we know f has at least one root on each of the intervals (−b, 0), (0, 1), (1, b). Since
a polynomial of order 3 has at most three roots, we conclude that f (x) has exactly
one root on each of the three intervals.

Example 1.7.7. We know

limx→ π − sin x 1
2
lim tan x = = = +∞,
π−
x→ 2 limx→ π − cos x 0+
2

limx→ π + sin x −1
limπ + tan x = 2
= = −∞.
x→− 2 limx→ π + cos x 0+
2

π π
Therefore for any number γ, we can find a > − and very close to − , such that
2 2
π π
tan a < γ. We can also find b < and very close to , such that tan b > γ. Then
2 2
tan x is continuous on [a, b] and tan a < γ < tan b. This implies that γ = tan c for
π
some c ∈ (a, b). Therefore any number is the tangent of some angle between −
2
π
and .
2
The example shows that, if f (x) is continuous on (a, b) satisfies limx→a+ f (x) =
−∞ and limx→b− f (x) = +∞, then f (x) can take any number as value on (a, b).
Note that the interval (a, b) here does not even have to be bounded.

Example 1.7.8. The function

(
x, if − 1 ≤ x ≤ 0,
f (x) = 2
x + 1, if 0 < x ≤ 1,

satisfies f (−1) = −1, f (1) = 2, but does not take any number in (0, 1] as value. The
problem is that the function is not continuous at 0, where a jump in value misses the
numbers in (0, 1]. Therefore the Intermediate Value Theorem cannot be applied.

Exercise 1.7.5. Let f (x) : [0, 1] → [0, 1] be a continuous function. Prove that there exists
at least one c ∈ [0, 1] satisfying f (c) = c.

Exercise 1.7.6. Find all the possible values.

x2 − 3x + 2 4. sin x.
1. .
x2 − 1 1
5. sin .
2. xx . x
(
2x , if − 1 ≤ x ≤ 0,
3. ex . 6.
x2 + 3, if 0 < x ≤ 1.
84 CHAPTER 1. LIMIT
√ √ √ √ √
Exercise 1.7.7. limx→+∞ cos( x + 2 + x) and limx→+∞ x(sin x + 2 − sin x) diverge.

1.7.3 Continuous Inverse Function

Given a function f (x), its inverse function f −1 (y) is obtained by solving f (x) = y
for x.

Example 1.7.9. To find the inverse of f (x) = 3x − 2, we solve 3x − 2 = y and get

1 2 1 2
x = y + . Therefore the inverse function is f −1 (y) = y + .
3 3 3 3

Example 1.7.10. To find the inverse of f (x) = x2 , we try to solve x2 = y.

The problem is that the equation has no solution for y < 0 and has two solutions
for y > 0. The ambiguity on which of the two solutions to choose can be removed if
we additionally specify x ≥ 0 or x ≤ 0. In other words, in order to unambiguously
specify an inverse of f (x) = x2 , we must specify the ranges for x and for y.
If we consider
f1 (x) = x2 : [0, +∞) → [0, +∞),
which means that we specify x ≥ 0 and y ≥ 0, then x2 = y always has unique
√
non-negative solution, which is usually denoted x = f1−1 (y) = y.
If instead we consider

f2 (x) = x2 : (−∞, 0] → [0, +∞),

then x2 = y also has a unique non-positive solution and gives the inverse
√
f2−1 (y) = − y : [0, +∞) → (−∞, 0].

As a more elaborate example, the function

f3 (x) = x2 : (−∞, −1] ∪ [0, 1) → [0, +∞)

has inverse
(√
y, if 0 ≤ y < 1
f3−1 (y) = √ : [0, +∞) → (−∞, −1] ∪ [0, 1).
− y, if y ≥ 1

The examples show that, for the concept of inverse function to be unambiguous,
we have to specify the ranges for the variable and the value. In this regard, if two
functions have the same formula but different ranges, then we should really think
of them as different functions.
In general, if a function f (x) is defined for all x ∈ D, then D is the domain of
the function, and all the values of f (x) is the range

R = {f (x) : x ∈ D}.
1.7. CONTINUITY 85

With the domain and range explicitly specified, we express the function as a map

f (x) : D → R.

Now the equation f (x) = y has solution only when y ∈ R. Moreover, we need to
make sure that the solution is unique in order for the inverse to be unambiguous.
This means that the function is one-to-one

x1 , x2 ∈ D, x1 6= x2 =⇒ f (x1 ) 6= f (x2 ).

The condition is the same as

x1 , x2 ∈ D, f (x1 ) = f (x2 ) =⇒ x1 = x2 .

Example 1.7.11. Consider the function f (x) = x5 + 3x3 + 1 defined for all x. By the
remark in Example 1.7.7, together with (see Example 1.6.7)

lim f (x) = −∞, lim f (x) = +∞,

x→−∞ x→+∞

we see that any number can be the value of f (x). This shows that the range of f is
R.
Is the function one-to-one? This can be established as follows. If x1 6= x2 , then
either x1 < x2 or x1 > x2 . In the first case, we have

x1 < x2 =⇒ x51 < x52 , x31 < x32

=⇒ f (x1 ) = x51 + 3x31 + 1 < f (x2 ) = x52 + 3x32 + 1.

By switching the roles of x1 and x2 , we get x1 > x2 implying f (x1 ) > f (x2 ). Either
way, we get x1 6= x2 implying f (x1 ) 6= f (x2 ).
We conclude that f (x) = x5 + 3x3 + 1 : R → R is invertible.

The argument in the example can be generalized. Suppose f (x) is a continuous

function defined on an interval domain D. Then the Intermediate Value Theorem
implies that the range R is also an interval. Moreover, the uniqueness of the solution
to f (x) = y can be obtained if f (x) is strictly increasing or strictly decreasing.

Theorem 1.7.3. A strictly increasing and continuous function on an interval is

invertible, and the inverse function is also strictly increasing and continuous. The
same holds for strictly decreasing and continuous functions.

It is remarkable that, according to the theorem, we get the continuity of the

inverse function for free. For example, although it is impossible to find the formula
for the solution of x5 + 3x3 + 1 = y, we already know the solution is a continuous
function of y,
86 CHAPTER 1. LIMIT

Example 1.7.12. The sine function can take any value in [−1, 1]. To make sure it is
one-to-one, we specify the domain and range
h π πi
sin x : − , → [−1, 1].
2 2
This is strictly increasing and continuous, and takes any number in [−1, 1] as value.
Therefore we get the inverse sine function
h π πi
arcsin y : [−1, 1] → − , .
2 2
The inverse sine function is also strictly increasing and continuous.

tan x
arcsin x
− π2
1 sin x π
2 arctan x
− π2 −1 − π2
−1 π π
2 2
− π2
−1
− π2

Figure 1.7.3: Inverse trigonometric functions.

Example 1.7.13. The cosine function

cos x : [0, π] → [−1, 1]
is strictly decreasing, continuous, and takes any number in [−1, 1] as value. There-
fore we get the inverse cosine function
arccos x : [−1, 1] → [0, π],
which is also strictly decreasing and continuous.
However, the inverse cosine function is not much a new function. If x = arccos y,
x ∈ [0, π], y ∈ [−1, 1], then
π  π h π πi
sin − x = cos x = y, −x∈ − ,
2 2 2 2
π
shows that − x = arcsin y. Therefore we have the equality
2
π
arcsin y + arccos y = .
2
1.7. CONTINUITY 87

Example 1.7.14. The tangent function

 π π
tan x : − , → (−∞, +∞)
2 2
is strictly increasing, continuous, and by Example 1.7.7, takes any number as the
value. Therefore we have the inverse tangent function
 π π
arctan y : (−∞, +∞) → − , ,
2 2
which is also strictly increasing and continuous.
We claim that
π π
lim arctan y = , lim arctan y = − .
y→+∞ 2 y→−∞ 2
π  π  π
For any >  > 0, let N = tan −  . Then −  = arctan N , and the strict
2 2 2
increasing property of arctan implies
π π  π 
y > N =⇒ > arctan y > arctan N = −  =⇒ arctan y −  < .

2 2 2
This proves the first limit. The second limit can be proved similarly.

Example 1.7.15. For a > 1, the exponential function

ax : R → (0, +∞)
is strictly increasing and continuous. Moreover, by Example 1.6.14, we have
lim ax = 0, lim ax = +∞.
x→−∞ x→+∞

Then by an argument similar to Example 1.7.7, any number in (0, +∞) is a value
of ax . Therefore the exponential function has an inverse
loga x : (0, +∞) → R,
called the logarithmic function with base a. Like the exponential function, the
logarithmic function is strictly increasing and continuous. We can also show
lim loga x = +∞, lim loga x = −∞, for a > 1
x→+∞ x→0+
by method similar to Example 1.7.13.
The logarithm loge x based on the special value e is called the natural logarithm.
We will denote the natural logarithm simply by log x.
For 0 < a < 1, the exponential ax is strictly decreasing and continuous. The
corresponding logarithm can be similarly defined and is also strictly decreasing and
continuous. Moreover, we have
lim loga x = −∞, lim loga x = +∞, for 0 < a < 1.
x→+∞ x→0+

Exercise 1.7.8. Let f be a strictly increasing function. Show that its inverse is also strictly
increasing.
88 CHAPTER 1. LIMIT

ex

1 log x

Figure 1.7.4: Exponential and logarithm.

1.7.4 Continuous Change of Variable

Suppose f (x) is continuous and strictly increasing near a. Then by Theorem 1.7.3,
f (x) can be inverted near a, and the inverse f −1 (y) is also continuous and strictly
increasing near b = f (a). The continuity of f and f −1 means that

x → a ⇐⇒ y → b.

The one-to-one property implies that

x 6= a ⇐⇒ y 6= b.

Therefore the composition rule can be applied in both directions, and we have

lim g(f (x)) = lim g(y).

x→a y→b

Here the equality means that the convergence of both sides are equivalent, and the
limits have the same value.

Example 1.7.16. Since sin x is strictly increasing and continuous near 0, we have
arcsin y x
lim = lim = 1.
y→0 y x→0 sin x

The same argument also tells us

arctan y x
lim = lim = 1.
y→0 y x→0 tan x

Example 1.7.17. In Examples 1.5.11, 1.5.16, 1.6.10, we find limx→a f (x2 ) = limx→a2 f (x)
for a 6= 0 and limx→0 f (x2 ) = limx→0+ f (x). Here we explain the two equalities from
the viewpoint of continuous change of variable.
The function
x2 : [0, +∞) → [0, +∞)
1.7. CONTINUITY 89

is strictly increasing and continuous, and is therefore invertible, with strictly in-
creasing and continuous inverse
√
x : [0, +∞) → [0, +∞).

The continuous change of variable implies that limx→a f (x2 ) = limx→a2 f (x) for
a > 0 and limx→0+ f (x2 ) = limx→0
√ f (x). Note that the second equality makes use
+

of the right continuity of x2 and x at 0.

Similarly, the function

x2 : (−∞, 0] → [0, +∞)

is strictly decreasing and continuous, and is therefore invertible, with strictly de-
creasing and continuous inverse. This implies that limx→a f (x2 ) = limx→a2 f (x) for
a < 0 and limx→0− f (x2 ) = limx→0+ f (x).
By
lim+ f (x2 ) = lim+ f (x) = lim− f (x2 ),
x→0 x→0 x→0
2
we also conclude that limx→0 f (x ) = limx→0+ f (x).

Example 1.7.18. Since the natural logarithm log x is continuous, we may move the
limit from outside the logarithm to inside the logarithm

log(x + 1) 1
 1

lim = lim log(1 + x) x = log lim (1 + x) x = log e = 1.
x→0 x x→0 x→0

Here the third equality follows from Example 1.6.17.

Exercise 1.7.9. Find the limits.

loga (x + 1) log x − log a logb x − logb a
1. limx→0 . 2. limx→a . 3. limx→a .
x x−a x−a

Exercise 1.7.10. Find the limits.

log(x2 − 2x + 1) log(x2 − 2x + 1) log(x2 − 2x + 1)

1. limx→2 . 2. limx→1 . 3. limx→0 .
x2 − 4 x2 − 1 x

Exercise 1.7.11. Find the limits.

1 ax + c ax + b
1. limx→0 log . 4. limx→∞ x log .
x bx + c ax + c
1 ax + b ax + b
2. limx→0 log . 5. limx→∞ x log .
x cx + d cx + d
1 a2 x2 + a1 x + a0 a2 x2 + a1 x + a0
3. limx→0 log . 6. limx→∞ x log .
x b2 x2 + b1 x + b0 b2 x2 + b1 x + b0
90 CHAPTER 1. LIMIT

Exercise 1.7.12. Find the limits.

x log(x2 + 3x + 2) − log 2
1. limx→0 . 3. limx→0 .
log(ax + 1) x
x2 − 1 log x
2. limx→1 . 4. limx→1 .
log x sin πx

Example 1.7.19. Since y = ex − 1 is strictly increasing and continuous, with inverse

x = log(y + 1), we may change the variable and use Example 1.7.18 to get

ex − 1 y
lim = lim = 1.
x→0 x y→0 log(y + 1)

Exercise 1.7.13. Find the limits.

e2x − 1 ax − ab eax − ebx

1. limx→0 . 4. limx→b . 7. limx→0 .
x x−b x
ax − 1 ex − 1 eax − ebx
2. limx→0 . 5. limx→∞ . 8. limx→+∞ .
x x x
ex − ea ax − bx ax − bx
3. limx→a . 6. limx→0 . 9. limx→0 .
x−a x cx − dx

Example 1.7.20. The continuity of the logarithm implies the continuity of the func-
tion xx = ex log x on (0, ∞). In general, if f (x) and g(x) are continuous and f (x) > 0,
then f (x)g(x) = eg(x) log f (x) is continuous.
The continuity of the exponential and the logarithm can also be used to prove
that
lim an = l > 0, lim bn = k =⇒ lim abnn = lk .
n→∞ n→∞ n→∞

The reason is that the continuity of log implies limn→∞ log an = log l. Then the
arithmetic rule implies limn→∞ bn log an = k log l. Finally, the continuity of the
exponential implies

lim abnn = lim ebn log an = elimn→∞ bn log an = ek log l = lk .

n→∞ n→∞

In Exercises 1.6.19 and 1.6.21, we took a number of steps to prove the same
property, without using the logarithmic function.

Example 1.7.21. For p 6= 0, y = p log(x + 1) is strictly increasing and continuous.

1.7. CONTINUITY 91

Using change the variable and Examples 1.7.18 and 1.7.19, we get

(x + 1)p − 1 ep log(x+1) − 1 ep log(x+1) − 1 log(x + 1)

lim = lim = lim p
x→0 x x→0 x x→0 p log(x + 1) x
p log(x+1)
e −1 log(x + 1)
= p lim lim
x→0 p log(x + 1) x→0 x
y
e −1 log(x + 1)
= p lim lim = p · 1 · 1 = p.
y→0 y x→0 x

Exercise 1.7.14. Find the limits.

xp − 1 xp − ap sin xp − sin ap
1. limx→1 . 3. limx→a . 5. limx→a .
x−1 x−a x−a
p p
xp − 1 xp − ap ex − ea
2. limx→1 . 4. limx→a . 6. limx→a .
xq − 1 xq − aq bx − ba

Exercise 1.7.15. Let

1 1 1 1
xn = 1 + + · · · + − log n, + · · · + − log(n + 1).
yn = 1 +
2 n 2 n
 
1 1 1
1. Use Exercise 1.3.17 to prove that < log 1 + < .
1+n n n
2. Prove that xn is strictly decreasing and yn is strictly increasing.

3. Prove that both xn and yn converge to the same limit

 
1 1 1
lim 1 + + + · · · + − log n = 0.577215669015328 · · · .
n→∞ 2 3 n

The number is called the Euler1 -Mascheroni2 constant.

1
Leonhard Paul Euler, born 1707 in Basel (Switzerland), died 1783 in St. Petersburg (Russia).
Euler is one of the greatest mathematicians of all time. He made important discoveries in almost
all areas of mathematics. Many theorems, quantities, and equations are named after Euler. He
also introduced much of√the modern mathematical terminology and notation, including f (x), e, Σ
(for summation), i (for −1), and modern notations for trigonometric functions.
2
Lorenzo Mascheroni, born 1750 in Lombardo-Veneto (now Italy), died 1800 in Paris (France).
The Euler-Mascheroni constant first appeared in a paper by Euler in 1735. Euler calculated the
constant to 6 decimal places in 1734, and to 16 decimal places in 1736. Mascheroni calculated the
constant to 20 decimal places in 1790.
92 CHAPTER 1. LIMIT
Chapter 2

Differentiation

2.1 Linear Approximation

The basic idea of differentiation is solving problems by using simple functions to ap-
proximate general complicated functions. The simplest functions are the constant
functions, which are usually too primitive to be useful. More effective approxima-
tions are given by linear functions A + Bx.

Definition 2.1.1. A linear approximation of a function f (x) at x0 is a linear function

L(x) = a + b(x − x0 ), such that for any  > 0, there is δ > 0, such that
|x − x0 | < δ =⇒ |f (x) − L(x)| = |f (x) − a − b(x − x0 )| ≤ |x − x0 |.
A function is differentiable if it has a linear approximation.

The differentiability at x0 requires the function to be defined on a neighborhood

of x0 , and the linear approximation depends only on the function near x0 .
In everyday life, we use approximations all the time. For example, when we
measure certain distance and get 7 meters and 5 centimeters, we really mean give or
take some millimeters. So the real distance might be 7.052 meters or 7.046 meters.
The function f (x) is like the real distance (7.052 meters or 7.046 meters), and the
linear function L(x) is like the reading (7.05 meters) from the ruler.
The accuracy of the measurement depends on how refined the ruler is. We often
use the rulers with two units m and cm. The centimeter cm is smaller among the
two units and is therefore the “basic unit” that gives the accuracy of the ruler. The
error |7m5.2cm − 7m5cm| = 0.2cm between the real distance and the measurement
should be significantly smaller than the basic unit 1cm.
Analogously, the linear function L(x) = a · 1 + b · (x − x0 ) is a combination of
two units 1 and x − x0 . Since the approximation happens for x near x0 , x − x0 is
much smaller than 1 and is therefore the “basic unit”. The error |f (x) − L(x)| of
the approximation should be significantly smaller than the size |x − x0 | of the basic
unit, which exactly means ≤ |x − x0 | on the right side of the definition.

93
94 CHAPTER 2. DIFFERENTIATION

2.1.1 Derivative
Geometrically, a function may be represented by its graph. The graph of a linear
function is a straight line. Therefore a linear approximation at x0 is a straight line
that “best fits” the graph of the given function near x0 . This is the tangent line of
the function.

LP Q
Q tangent L
P

x0 x

Figure 2.1.1: The linear approximation is the tangent line.

Specifically, the point P in Figure 2.1.1 is the point (x0 , f (x0 )) on the graph
of f (x). We pick a nearby point Q = (x, f (x)) on the graph, for x near x0 . The
straight line connecting P and Q is the linear function (the variable in LP Q is t
because x is already used for Q)

f (x) − f (x0 )
LP Q (t) = f (x0 ) + (t − x0 ).
x − x0

As x → x0 , Q approaches P , and the linear function approaches L(t) = a + b(t − x0 ).

Therefore we have a = f (x0 ), and b is given below.

Definition 2.1.2. The derivative of a function f (x) at x0 is

f (x) − f (x0 ) f (x0 + h) − f (x0 )

f 0 (x0 ) = lim = lim .
x→x0 x − x0 h→0 h

We emphasize that the linear approximation is the concept. As the coefficient b

of the first order term, the derivative f 0 (x0 ) is the computation of the concept. The
following says that the concept and its computation are equivalent.

Proposition 2.1.3. A function f (x) is differentiable at x0 if and only if the derivative

f 0 (x0 ) exists. Moreover, the linear approximation is given by f (x0 ) + f 0 (x0 )(x − x0 ).

The notation f 0 for the derivative is due to Joseph Louis Lagrange. It is simple
and convenient, but could become ambiguous when there are several variables related
df
in more complicated ways. Another notation , due to Gottfried Wilhelm Leibniz,
dx
2.1. LINEAR APPROXIMATION 95

is less ambiguous. So we also write


df  f (x) − f (x0 ) ∆f
= lim = lim , ∆f = f (x) − f (x0 ), ∆x = x − x0 .
dx x=x0
 x→x 0 x − x0 ∆x→0 ∆x
We emphasize that Leibniz’s notation is not the “quotient” of two quantities df and
dx. It is an integrated notation that alludes to the fact that the derivative is the
∆f
limit of the quotient of differences.
∆x

Example 2.1.1. The function f (x) = 3x − 2 is already linear. So its linear approx-
imation must be L(x) = f (x) = 3x − 2. This reflects the intuition that, if the
distance is exactly 7m5cm, then the measure by the ruler in centimeters should be
d(3x − 2)
7m5cm. In particular, the derivative f 0 (x) = 3, or = 3. In general, we
dx
have (A + Bx)0 = B.

Example 2.1.2. To find the linear approximation of x2 at 1, we rewrite the function

in terms of x − 1
x2 = (1 + (x − 1))2 = 1 + 2(x − 1) + (x − 1)2 .
Note that L(x) = 1 + 2(x − 1) is linear, and the error |x2 − L(x)| = (x − 1)2 is
significantly smaller than |x − 1| when x is near 1
|x − 1| < δ =  =⇒ |x2 − L(x)| ≤ |x − 1|.
Therefore 1 + 2(x − 1) is the linear approximation of x2 at 1, and the derivative
d(x2 ) 
(x2 )0 |x=1 = = 2.
dx x=1

Exercise 2.1.1. Find the linear approximations and then the derivatives.

1. 5x + 3 at x0 . 3. x2 at x0 . 5. xn at 1.

2. x3 − 2x + 1 at 1. 4. x3 at x0 . 6. xn at x0 .

Exercise 2.1.2. Interpret the limits as derivatives.

(1 + x)p − 1 cos x arcsin x
1. limx→0 . 3. limx→ π2 . 5. limx→0 .
x x − π2 x
√
x+9−3 sin x 1 x
2. limx→0 . 4. limx→π . 6. limx→2 log .
x x−π x−2 2

2.1.2 Basic Derivative

We derive the derivatives of the power function, the exponential function, the loga-
rithmic function, and the trigonometric functions.
96 CHAPTER 2. DIFFERENTIATION

Example 2.1.3. For x0 6= 0, we have

1 1
  −
d 1 
 x x0 x0 − x 1 1
= lim = lim = lim − = − 2.
dx x x=x0 x→x0 x − x0 x→x 0 x0 x(x − x0 ) x→x x0 x x0
 0

1 1 1
Therefore is differentiable at x0 , and the linear approximation is − 2 (x − x0 ).
x x0 x0
We express the derivative as  0
1 1
= − 2.
x x

Example 2.1.4. For x0 > 0, we have

√  √ √ √ √
d x  x − x0 x − x0 1
= lim = lim √ √ √ √ = √ .
dx x=x0 x→x0 x − x0
 x→x0 ( x − x0 )( x + x0 ) 2 x0

√ √ 1
Therefore x is differentiable, and the linear approximation is x0 − √ (x − x0 ).
2 x0
We express the derivative as
√
√ 0 d x 1
( x) = = √ .
dx 2 x

Example 2.1.5. By Example 1.7.21, we have

d(xp )  (1 + h)p − 1

= lim = p.
dx x=1 h→0 h

Therefore xp is differentiable at 1 and has linear approximation 1 + p(x − 1).

1
Examples 2.1.3 and 2.1.4 are the derivatives of xp for p = −1 and at general
2
x0 > 0. For general p, we take h = x0 y and get

d(xp )  (x0 + h)p − xp0 (x0 + x0 y)p − xp0


= lim = lim
dx x=x0 h→0 h y→0 x0 y
(1 + y)p − 1
= lim xp−1
0 = pxp−1
0 .
y→0 y
We express the derivative as
(xp )0 = pxp−1 .

Example 2.1.6. By Example 1.7.18, we have


d log x  log x − log 1 log(y + 1)
= lim = lim = 1.
dx x=1 x→1
 x−1 y→0 y
2.1. LINEAR APPROXIMATION 97

Therefore log x is differentiable at 1 and has linear approximation x − 1.

In general, at any x0 > 0, by taking h = x0 y, we have

d log x  log(x0 + h) − log x0 log(y + 1) 1
= lim = lim = .
dx x=x0 h→0
 h y→0 x0 y x0

We express the derivative as

1
(log x)0 = .
x

Example 2.1.7. By Example 1.7.19, we have

dex  ex0 +h − ex0 h


x0 e − 1
= lim = lim e = e x0 .
dx x=x0 h→0 h h→0 h

We express the derivative as

(ex )0 = ex .

Example 2.1.8. In Section 1.5.4, we find

sin x − sin 0 sin x
lim = lim = 1,
x→0 x−0 x→0 x
Therefore sin x is differentiable at x = 0, and the linear approximation at 0 is x.
More generally, we have

d sin x  sin(x0 + h) − sin x0
= lim
dx x=x0 h→0
 h
sin h cos x0 + cos h sin x0 − sin x0
= lim
h→0
 h 
cos h − 1 sin h
= lim sin x0 + cos x0
h→0 h h
= 0 · sin x0 + 1 · cos x0 = cos x0 .

We express the result as

(sin x)0 = cos x.
By similar method, we have
(cos x)0 = − sin x.

Example 2.1.9. In Section 1.5.4, we find

sin x − sin 0 sin x
lim = lim = 1,
x→0 x−0 x→0 x
Therefore sin x is differentiable at x = 0, and the linear approximation at 0 is x.
98 CHAPTER 2. DIFFERENTIATION

More generally, we have


d sin x  sin(x0 + h) − sin x0
= lim
dx x=x0 h→0
 h
sin h cos x0 + cos h sin x0 − sin x0
= lim
h→0
 h 
cos h − 1 sin h
= lim sin x0 + cos x0
h→0 h h
= 0 · sin x0 + 1 · cos x0 = cos x0 .

We express the result as

(sin x)0 = cos x.
By similar method, we have
(cos x)0 = − sin x.

Exercise 2.1.3. Find the derivatives and then the linear approximations.
√
1. 3
x at 1. 3. cos x2 at 0. 5. arcsin x at 0. 7. sin sin x at 0.

2. (log x)2 at 1. 4. tan x at 0. 6. arctan x at 0. 8. x2 D(x) at 0.

Exercise 2.1.4. Find the derivatives, a > 0.

1. loga x. 2. ax . 3. tan x. 4. arcsin x.

Exercise 2.1.5. We have log |x| = log(−x) for x < 0. Show that the derivative of log(−x)
1
at x0 < 0 is . The interpret your result as
x0
1
(log |x|)0 = .
x

Exercise 2.1.6. What is the derivative of loga |x|?

Exercise 2.1.7. Suppose p is an odd integer. Then xp is defined for x < 0. Do we still have
(xp )0 = pxp−1 for x < 0?

2.1.3 Constant Approximation

If a measurement of distance by a ruler in centimeters gives 7 meters and 5 centime-
ters, then the measurement by another (more primitive) ruler in meters should give
7 meters.
Analogously, if a + b(x − x0 ) is a linear approximation of f (x) at x0 , then a is a
constant approximation of f (x) at x0 . Since the “basic unit” for constant functions
2.1. LINEAR APPROXIMATION 99

is 1, the constant approximation means that, for any  > 0, there is δ > 0, such that

|x − x0 | < δ =⇒ |f (x) − a| ≤ .

This means exactly that f (x) is continuous at x0 , and the approximating constant
is a = f (x0 ). Therefore the fact of linear approximation implying constant approx-
imation means the following.

Theorem 2.1.4. If a function is differentiable at a, then it is continuous at a.

We do not expect the continuity to imply differentiability, because we do not

expect the measurement in meters can tell us the measurement in centimeters.

Example 2.1.10. The sign function


1,
 if x > 0,
sign(x) = 0, if x = 0,

−1, if x < 0,


is not continuous at 0, and is therefore not differentiable at 0. Of course, we have

(sign(x))0 = 0 away from 0.

Example 2.1.11. The absolute value function |x| is continuous everywhere. Yet the
derivative
|x| − |0|
(|x|)0 |x=0 = lim = lim sign(x)
x→0 x − 0 x→0

diverges. Therefore the continuous function is not differentiable at 0.

Example 2.1.12. The Dirichlet function D(x) in Example 1.6.13 is not continuous
anywhere and is therefore not differentiable anywhere.
On the other hand, the function xD(x) is continuous at 0. Yet the derivative

xD(x)
(xD(x))0 |x=0 = lim = lim D(x)
x→0 x x→0

diverges. Therefore xD(x) is not differentiable at 0, despite the continuity.

Exercise 2.1.8. Find the derivative of |x| at x0 6= 0.

Exercise 2.1.9. Determine the differentiability of |x|p at 0.

Exercise 2.1.10. Is xD(x) differentiable at x0 6= 0?

Exercise 2.1.11. Determine the differentiability of |x|p D(x) at 0.

100 CHAPTER 2. DIFFERENTIATION

Exercise 2.1.12. Determine the differentiability of


|x|p sin 1 , if x 6= 0,
f (x) = x
0, if x = 0,

at 0.

Exercise 2.1.13. Let [x] be the greatest integer ≤ x. Study the differentiability of [x].

2.1.4 One Sided Derivative

Like one sided limits, we have one sided derivatives

f (x) − f (x0 ) f (x) − f (x0 )

f+0 (x0 ) = lim+ , f−0 (x0 ) = lim− .
x→x0 x − x0 x→x0 x − x0

The derivative f 0 (x0 ) exists if and only if both f+0 (x0 ) and f−0 (x0 ) exist and are equal.

Example 2.1.13. We have

|x| |x|
(|x|)0 |at 0+ = lim+ = 1, (|x|)0 |at 0− = lim− = −1.
x→0 x x→0 x
Therefore |x| has left and right derivatives. Since the two one sided derivatives are
different, the function is not differentiable at 0.

Example 2.1.14. Consider the function

(
ex , if x ≥ 0,
f (x) =
x + 1, if x < 0.

We have (note that f (0) = e0 = 0 + 1)

f+0 (0) = (ex )0 |x=0 = 1, f−0 (0) = (x + 1)0 |x=0 = 1.

Therefore f 0 (0) = 1 and has linear approximation 1 + x at 0.

Exercise 2.1.14. Determine the differentiability.

1. |x2 − 3x + 2| at 0, 1, 2. 3. | sin x| at 0.
√
2. 1 − cos x at 0. 4. |π 2 − x2 | sin x at π.

Exercise 2.1.15. Determine the differentiability at 0.

2.2. PROPERTY OF DERIVATIVE 101
1
( (
x2 , if x ≥ 0, xe− x , if x ≥ 0,
1. 3.
x, if x < 0. 0, if x < 0.

 1 , if x ≥ 0, (
2. x+1 log(1 + x), if x ≥ 0,
4.
x, if x < 0. ex − 1, if x < 0.

Exercise 2.1.16. Determine the differentiability, p, q > 0.

( ( 2
(x − a)p (b − x)q , if a ≤ x ≤ b, x2 e−x , if |x| ≤ 1,
1. 3.
0, otherwise. e−1 , if |x| > 1.
( (
arctan x, if |x| ≤ 1, log |x|, if |x| ≥ 1,
2. π 4.
x, if |x| > 1. x, if |x| < 1.
4

Exercise 2.1.17. Find a, b, c, such that the function

(a
, if x > 1,
f (x) = x
bx + c, if x ≤ 1,

is differentiable at 1.

Exercise 2.1.18. For p ≥ 0, xp is defined on [0, δ). What is the right derivative of xp at 0?

Exercise 2.1.19. For some p (see Exercises 2.1.7 and 2.1.18), xp is defined on (−δ, δ). What
is the derivative of xp at 0?

Exercise 2.1.20. Suppose g(x) is continuous at x0 . Show that f (x) = |x − x0 |g(x) is

differentiable at x0 if and only if g(x0 ) = 0.

2.2 Property of Derivative

2.2.1 Arithmetic Combination of Linear Approximation
Suppose f (x) and g(x) are linearly approximated respectively by a + b(x − x0 ) and
c + d(x − x0 ) at x0 . Then f (x) + g(x) is approximated by

(a + b(x − x0 )) + (c + d(x − x0 )) = (a + c) + (b + d)(x − x0 ).

Therefore f + g is differentiable and (f + g)0 (x0 ) = b + d. Since b = f 0 (x0 ) and

d = g 0 (x0 ), we conclude that

d(f + g) df dg
(f + g)0 (x) = f 0 (x) + g 0 (x), or = + .
dx dx dx
102 CHAPTER 2. DIFFERENTIATION

Similarly, Cf (x) is approximated by

C(a + b(x − x0 )) = Ca + Cb(x − x0 ).

Therefore Cf (x) is differentiable and (Cf )0 (x0 ) = Cb, which means

d(Cf ) df
(Cf )0 (x) = Cf 0 (x), or =C .
dx dx
We also have f (x)g(x) approximated by

(a + b(x − x0 ))(c + d(x − x0 )) = ac + (bc + ad)(x − x0 ) + bd(x − x0 )2 .

Although the approximation is not linear, the square unit (x − x0 )2 is much smaller
than x − x0 when x is close to x0 . Therefore f (x)g(x) is differentiable and has
linear approximation ac + (bc + ad)(x − x0 ), and we get (f g)0 (x0 ) = bc + ad. By
a = f (x0 ), b = f 0 (x0 ), c = g(x0 ), d = g 0 (x0 ), we get the Leibniz rule
d(f g) df dg
(f g)0 (x) = f 0 (x)g(x) + f (x)g 0 (x), or = g+f .
dx dx dx
The explanation above on the derivatives of arithmetic combinations are analo-
gous to the arithmetic properties of limits.

Exercise 2.2.1. Find the derivative of the polynomial p(x) = cn xn + · · · + c1 x + c0 .

Exercise 2.2.2. Compute the derivatives.

1. ex sin x. 4. sin2 x cos x. 7. (x − 1)ex . 10. 2x2 log x − x2 .

2. sin2 x. 5. sin x − x cos x. 8. x2 ex . 11. xex cos x.

3. e2x . 6. cos x + x sin x. 9. x log x − x. 12. xex cos x log x.

Exercise 2.2.3. Find a polynomial p(x), such that (p(x)ex )0 = x2 ex . In general, suppose
(pn (x)ex )0 = xn ex . Find the relation between polynomials pn (x).

Exercise 2.2.4. Find polynomials p(x) and q(x), such that (p(x) sin x + q(x) cos x)0 =
x2 sin x. Moreover, find a function with derivative x2 cos x?

Exercise 2.2.5. Find constants A and B, such that (Aex sin x+Bex cos x)0 = ex sin x. What
about (Aex sin x + Bex cos x)0 = ex cos x?

2.2.2 Composition of Linear Approximation

Consider a composition g ◦ f

x 7→ y = f (x) 7→ z = g(y) = g(f (x)).

2.2. PROPERTY OF DERIVATIVE 103

Suppose f (x) is linearly approximated by a + b(x − x0 ) at x0 and g(y) is linearly

approximated by c + d(y − y0 ) at y0 = f (x0 ). Then
a = f (x0 ) = y0 , b = f 0 (x0 ), c = g(y0 ) = g(f (x0 )), d = g 0 (y0 ) = g 0 (f (x0 )),
and the composition g ◦ f is approximated by the composition of linear approxima-
tions (recall a = y0 )
c + d[(a + b(x − x0 )) − y0 ] = c + db(x − x0 ).
Therefore the composition is also differentiable, with
(g ◦ f )0 (x0 ) = db = g 0 (y0 )f 0 (x0 ) = g 0 (f (x0 ))f 0 (x0 )
This gives us the chain rule
(g(f (x)))0 = (g ◦ f )0 (x) = g 0 (f (x))f 0 (x) = g 0 (y)|y=f (x) f 0 (x),
or
dz dz dy
= .
dx dy dx

1
Example 2.2.1. We know (log x)0 = for x > 0. For x < 0, we have
x

0 0 0 1  1 1
(log(−x)) = (log y) |y=−x (−x) =  (−1) = (−1) = .
y y=−x −x x

Therefore we conclude
1
(log |x|)0 = , for x 6= 0.
x

Example 2.2.2. In Example 2.1.5, we use the definition to derive xp = pxp−1 . Alter-
natively, we may also derive the derivative of xp at general x0 > 0 from the derivative
(xp )0x=1 = p at a special place.
x
To move from x0 to 1, we introduce y = . Then xp is the composition
x0
x
x 7→ y = 7→ z = xp = xp0 y p .
x0
Then x = x0 corresponds to y = 1, and we have
d(xp )  d(xp0 y p ) 
     
dz  dy  d x 
= =
dx x=x0 dy y=1 dx x=x0 dy y=1 dx x0 x=x0
d(y p ) 

1 1
= xp0 · · = xp0 · p · = pxp−1
0 .
dy  x0y=1 x0

Exercise 2.2.6. Use the derivative at a special place to find the derivative at other places.
104 CHAPTER 2. DIFFERENTIATION

1. log x. 2. ex . 3. sin x. 4. cos x.

π 
Exercise 2.2.7. Use cos x = sin − x and the derivative of sine to derive the derivative
2
of cosine. Use the similar method to find the derivatives of cot x and csc x.

Exercise 2.2.8. A function f (x) is odd if f (−x) = −f (x), and is even if f (−x) = f (x).
What can you say about the derivative of an odd function and the derivative of an even
function?

Example 2.2.3. By Example 2.1.3 and the chain rule, we have

0  0 
f 0 (x)
 
1 1  0 1  0
= f (x) = − f (x) = − .
f (x) y y=f (x) y 2 y=f (x) f (x)2

Then we may use the Leibniz rule to get the derivative of quotient
 0  0  0
f (x) 1 0 1 1
= f (x) = f (x) + f (x)
g(x) g(x) g(x) g(x)
0
1 g (x) f (x)g(x) − f (x)g 0 (x)
0
= f 0 (x) − f (x) = .
g(x) g(x)2 g(x)2

Exercise 2.2.9. Derive the derivatives.

(tan x)0 = sec2 x, (sec x)0 = sec x tan x, (e−x )0 = −e−x .

Exercise 2.2.10. Compute the derivatives.

1 x2 − x + 1 1 x
1. . 3. . 5. . 7. .
x+2 x3 + 1 ax + b x2 + ax + b

x+1 x3 + 1 ax + b 1
2. . 4. . 6. . 8. .
x−2 x2 −x+1 cx + d (x + a)(x + b)

Exercise 2.2.11. Compute the derivatives.

log x log x xp ex
1. . 2. . 3. . 4. .
x xp log x x log x

Exercise 2.2.12. Compute the derivatives.

sin x 1 1 + x tan x cos x + x sin x

1. . 2. . 3. . 4. .
a + cos x a + tan x tan x − x sin x − x cos x

Exercise 2.2.13. The hyperbolic trigonometric functions are

ex − e−x ex + e−x
sinh x = , cosh x = ,
2 2
2.2. PROPERTY OF DERIVATIVE 105

and
sinh x cosh x 1 1
tanh x = , coth x = , sechx = , cschx = .
cosh x sinh x cosh x sinh x
Find their derivatives and express them in hyperbolic trigonometric functions.

Example 2.2.4. The function (x2 − 1)10 is the composition of z = y 10 and y = x2 − 1.

Therefore

0 d(y 10 ) d(x2 − 1)
(x2 − 1)10 = = 10y 9 · 2x = 20x(x2 − 1)9 .
dy dx

Example 2.2.5. The function ax = ebx , b = log a, is the composition of z = ey and

y = bx. Therefore

(ax )0 = (ey )0 |y=bx (bx)0 = ebx b = ax log a.

Exercise 2.2.14. Compute the derivatives.

1. (1 − x)10 . 4. (1 + (1 − x2 )10 )9 . (x + 1)9

7. .
(3x + 5)8
2. (3x + 2)10 . 5. ((x3 −1)8 +(1−x2 )10 )9 .
x(x + 1)
3. (x3 − 1)10 (1 − x2 )9 . 6. ((1 − (3x + 2)3 )8 + 1)9 . 8. .
(x + 2)(x + 3)

Exercise 2.2.15. Compute the derivatives.

√
1. cos(x5 + 3x2 + 1). 6. sin x + cos x. sin2 x
10. .
sin x2
2. tan10 x(x + 1)9 .

 3 10
sin x
√ 7. . sin 2x + 2 cos 2x
3. sin( x + 3). cos4 x 11. .
2 sin x − cos 2x
√
4. sin( x − 2 + 3). 8. sin(cos x). √
sin8 x
5. (sin x + cos x)10 . 9. sin(cos(tan x)). 12. .
1 + cos10 (sin9 x)

Exercise 2.2.16. Compute the derivatives.

2
1. ex . 5. logx e. 9. log | cos x|.

2. (x2 − 1)ex .
2 6. log(log x). 10. log | tan x|.
 
x 1 11. log | sec x − tan x|.
3. e(e ) . 7. log .
log x
1 − sin x
4. elog x . 8. log(log(log x)). 12. log .
1 + sin x

Exercise 2.2.17. Compute the derivatives.

106 CHAPTER 2. DIFFERENTIATION

1. (ax + b)p . 4. eax . 7. cos(ax + b).

2. (ax2 + bx + c)p . 5. log(ax + b). 8. tan(ax + b).

3. (a + (bx2 + c)p )q . 6. sin(ax + b). 9. sec(ax + b).

Exercise 2.2.18. Compute the derivatives.

1 1 x
1. √ . 3. √ . 5. √ .
a − x2
2 x − a2
2 a2 + x2
1 x x
2. √ . 4. √ . 6. √ .
a2 + x2 a2 − x2 x2 − a2

Exercise 2.2.19. Compute the derivatives.

√ √ √ 10
x+1
p 
1. 1 + x. 8. √ . 1+ x
14. √ .
(1 − x + x)3 1− x
1 √
2. p √ . ( x + 1)4 √ √
1+ x 9. √ .
(1 − x + x)3 x+1− x−1
15. √ √ .
q
√ r x+1+ x−1
1 + x2
p
3. 1 + 1 + x. 10. .
1 − x2 √ √
1+x− 1−x
√ 16. √ √ .
q r
2
p
4. x + x + x. 3 1 + x 1+x+ 1−x
11. .
1 − x2
√
5. (1 + 2 x + 1)10 . p √  10
1+ x 1 1
12. p √ . 17. √ + √ .
√ 1− x 1+ x 1− x
6. (1 + 2 x + 1)−10 .
√
s √
1 + x
r
x+1 13. √ . 1
7. . 1− x 18. 1+ √ .
x−2 2
x +1

Exercise 2.2.20. Compute the derivatives.

1. |x2 (x + 2)3 |. 2. | sin3 x|. 3. |x(ex − 1)|. 4. |(x − 1)2 log x|.

Exercise 2.2.21. Compute the derivatives.

√ √ 1
1. x − log | x + a|. 5. log(a2 + x2 ).
2
√ √
b 1
p
2. + log |ax + b|. 6. x(a + x) − a log( x + x + a).
a2 (ax + b) a2 √ √
1 ax + b − b
7. √ log √ √ .
1 ax + b b ax + b + b
3. − log . √ √
b x 8. x log(x + x2 + a) − x2 + a.
√
1 1 ax + b 1 a + a2 + x2
4. − 2 log . 9. − log .
b(ax + b) b x a x
2.2. PROPERTY OF DERIVATIVE 107
√
a2
 
1 a+ a2 − x2 1 1
10. − log . 12. − x2 + x2 − 2 log(a2 − b2 x2 ).
a x 2 2 b
b2
 
b 1 2 1 2
11. x− x + x − 2 log(ax + b).
2a 4 2 a

Example 2.2.6. By the chain rule, we have

  √ 0 √
log x + x2 + a = (log |y|)0 |y=x+√x2 +a x0 + ( z)0 |z=x2 +a (x2 + a)0
   
 
1 1 1
= √ 1+ √ 2x = √ .
x + x2 + a 2 x2 + a x2 + a
Now suppose we wish to find a function f (x) with derivative
1
f 0 (x) = √ .
x2 + ax + b
By
1 1 a a2
f (x) = r =p , y =x+ , c=b− ,
a 2
 a2 y2 + c 2 4
x+ +b−
2 4
a a2
we may substitute x by x + and substitute a by c = b − . Then we get
2 4
a √ 2

 0 0 a 0
      p   
log x + + x + ax + b = log y + y + c 
  2   x+
2 y=x+ a2 2

1  1
= p =√ .

2
 2
y + c y=x+ a x + ax + b
2

Exercise 2.2.22. Find constants A and B, such that (Aeax sin bx+Beax cos bx)0 = eax cos bx.
What about (Aeax sin bx + Beax cos bx)0 = eax sin bx?

Exercise 2.2.23. Use Example 2.2.6 to compute the derivatives.

√  √ 
1. log(ex + 1 + e2x ). 2. log x − x2 + a.
  3. log(tan x + sec x).

x
Exercise 2.2.24. Compute the derivative of log . Then find a function with derivative
x+1
1 1
. In case a2 ≥ 4b, can you find a function with derivative 2 ?
(x + a)(x + b) x + ax + b
√ √
Exercise 2.2.25. Compute√the derivative of x x2 + a + a log(x + x2 + a). Then find a
function with derivative x2 + ax + b.

Exercise 2.2.26. Use Example 2.2.6 to compute the derivatives.

108 CHAPTER 2. DIFFERENTIATION
√ √
1. log(ex + 1 + e2x ). 2. log(x − x2 + a). 3. log(tan x + sec x).

Example 2.2.7. By viewing xp = ep log x as a composition of z = ey and y = p log x,

we have
d(xp ) (ep log x ) d(ey ) 

d(p log x) p p
= = = ep log x = xp = pxp−1 .
dx dx dy y=p log x
 dx x x

This derives the derivative of xp by using the derivatives of ex and log x.

Example 2.2.8. Suppose u(x) and v(x) are differentiable and u(x) > 0. Then
u(x)v(x) = eu(x) log v(x) , and

(u(x)v(x) )0 = (ev(x) log u(x) )0 = (ey )0 |y=v(x) log u(x) (v(x) log u(x))0 .

By
(ey )0 |y=v(x) log u(x) = ey |y=v(x) log u(x) = ev(x) log u(x) = u(x)v(x) ,
and
v(x)u0 (x)
(v(x) log u(x))0 = v 0 (x) log u(x)+v(x)(log u)0 |u=u(x) u0 (x) = v 0 (x) log u(x)+ ,
u(x)
We get

v(x)u0 (x)
 
v(x) 0 v(x) 0
(u(x) ) = u(x) v (x) log u(x) +
u(x)
= u(x)v(x)−1 (u(x)v 0 (x) log u(x) + u0 (x)v(x)).

Exercise 2.2.27. Compute the derivatives.

a
1. xx . 4. (ax )x . 7. (xx )x . 10. x(x ) .
2 x x
2. xx . 5. (xa )x . 8. a(x ) . 11. x(x ) .
x x
3. (x2 )x . 6. (xx )a . 9. x(a ) . 12. (xx )(x ) .

Exercise 2.2.28. Compute the derivatives.

√
1. xsin x . 3. (sin x)cos x . 5. x
log x. 7. (ex + e−x )x .
√
2. (sin x)x . 4. x x. 6. xlog x . 8. (log |x2 − 1|)x .

Exercise 2.2.29. Let f (x) = u(x)v(x) . Then log f (x) = u(x) log v(x). By taking the deriva-
tive on both sides of the equality, derive the formula for f 0 (x).

Exercise 2.2.30. Use the idea of Exercise 2.2.29 to compute the derivatives.
2.2. PROPERTY OF DERIVATIVE 109

x+a (x + c)(x + d) (x2 + x + 1)7

1. . 3. . 5. .
x+b (x + a)(x + b) (x2 − x + 1)3
√ 2 √
1 (x + 3)7 2x − 1 ex +1 sin x
2. . 4. . 6. .
(x + a)(x + b) (2x + 1)3 (x2 − x + 1)3 log x

2.2.3 Implicit Linear Approximation

The chain rule can be used to compute the derivatives of functions that are “implic-
itly” given. Such functions often do not have explicit formula expressions.
Strictly speaking, we need to know that implicitly given functions are differen-
tiable before taking their derivatives. There are general theorems confirming such
differentiability. In the subsequent examples, we will always assume the differentia-
bility of implicitly defined functions.

2 2
Example 2.2.9. The unit
√ circle x + y = √
1 on the plane is made up of the graphs
of two functions y = 1 − x and y = − 1 − x2 . We may certainly compute the
2

derivative of each one explicitly

√ 1 1 −x √ x
( 1 − x2 )0 = (1 − x2 )− 2 (−2x) = √ , (− 1 − x2 )0 = √ .
2 1 − x2 1 − x2
On the other hand, we may use the fact that both functions y = y(x) satisfy
the equation x2 + y(x)2 = 1. Taking the derivatives in x of both sides, we get
2x + 2yy 0 = 0. Solving the equation, we get
x
y0 = − .
y
This is consistent with the two derivatives computed above.
There is yet another way of computing the derivative y 0 (x). The circle can be
parametrized as x = cos t, y = sin t. In this view, the function y = y(x) satisfies
sin t = y(cos t). By the chain rule, we have

cos t = y 0 (x)(− sin t).

Therefore
cos t x
y 0 (x) = − =− .
sin t y
In general, the derivative of a function y = y(x) given by a parametrized curve
x = x(t), y = y(t) is
y 0 (t)
y 0 (x) = 0 .
x (t)
Note that the formula is ambiguous, in that y 0 (x) = − cot t and y 0 (t) = cos t are not
d d
the same functions. The primes in the two functions refer to and respectively.
dx dt
110 CHAPTER 2. DIFFERENTIATION

So it is better to keep track of the variables by using Leibniz’s notation. The formula
above becomes
dy
dy
= dt .
dx dx
dt
dy dy dx
This is just another way of expressing the chain rule = .
dt dx dt

Exercise 2.2.31. Compute the derivatives of the functions y = y(x) given by curves.

1. x = sin2 t, y = cos2 t.

2. x = a(t − sin t), y = a(1 − cos t).

3. x = et cos 2t, y = et sin 2t.

4. x = (1 + cos t) cos t, y = (1 + cos t) sin t.

Example 2.2.10. Like the unit circle, the equation 2y−2x2 −sin y+1 = 0 is a curve on
the plane, made up of the graphs of several functions y = y(x). Although we cannot
find an explicit formula for the functions, we can still compute their derivatives.
Taking the derivative of both sides of the equation 2y − 2x2 − sin y + 1 = 0 with
respect to x and keeping in mind that y is a function of x, we get 2y 0 −4x−y 0 cos y = 0.
Therefore
4x
y0 = .
2 − cos y
r 
π π
The point P = , satisfies the equation and lies on the curve. The
2 2
tangent line of the curve at the point has slope
r
π
4 √
2
y 0 |P = π = 2π.
2 − cos
2
Therefore the tangent line at P is given by the equation

π √
 r 
π
y − = 2π x − ,
2 2
or √ π
y= 2πx − .
2
2.2. PROPERTY OF DERIVATIVE 111

Example 2.2.11. The equations x2 + y 2 + z 2 = 2 and x + y + z = 0 specify a circle

in the Euclidean space R3 and define functions y = y(x) and z = z(x). To find the
derivatives of the functions, we take the derivatives of the two equations in x
2x + 2yy 0 + 2zz 0 = 0, 1 + y 0 + z 0 = 0.
Solving for y 0 and z 0 , we get
z−x y−x
y0 = , z0 = .
y−z z−y

Exercise 2.2.32. Compute the derivatives of implicitly defined functions.

√ √ √
1. y 2 + 3y 3 + 1 = x. 4. x + y = a.

2. sin y = x. 5. ex+y = xy.

x2 y 2
3. + 2 = 1. 6. x2 + 2xy − y 2 − 2x = 0.
a2 b

Exercise 2.2.33. Find the derivative of the implicitly defined functions of x.

1. xp + y p = 2 at x = 1, y = 1.
2. xy = sin(x + y) at x = 0, y = π.
x+y y+z z+x
3. = = at x = y = z = 1.
z x y

Exercise 2.2.34. If f (sin x) = x, what can you say about the derivative of f (x)? What if
sin f (x) = x?

Example 2.2.12. In Example 1.7.11, we argued that the function f (x) = x5 + 3x3 + 1
is invertible. The inverse g(x) satisfies g(x)5 + 3g(x)3 + 1 = x. Taking the derivative
in x on both sides, we get 5g(x)4 g 0 (x) + 9g(x)2 g 0 (x) = 1. This implies
1
g 0 (x) = .
5g(x) + 9g(x)2

Example 2.2.13. In Example 2.2.12, we interpreted the derivative of an inverse

function as an implicit differentiation problem. In general, the inverse function
g(x) = f −1 (x) satisfies f (g(x)) = x. Taking the derivative of both sides, we get
f 0 (g(x))g 0 (x) = 1. Therefore
1 1
(g(x))0 = , or (f −1 (x))0 = .
f 0 (g(x)) f 0 (f −1 (x))
For example, the derivative of arcsin x is
1 1 1
(arcsin x)0 = = =√ .
(sin y)0 |y=arcsin x (cos y)|x=sin y 1 − x2
112 CHAPTER 2. DIFFERENTIATION
h π πi
In the last step, we have positive square root because y ∈ − , .
2 2
The computation can also be explained by considering two variables related by
y = y(x) and x = x(y), with x(y) being the inverse function of y(x). The chain rule
dx dy dx
tells us = = 1. Then we get
dy dx dx

dx 1 1 1
= , or x0 (y) = 0 = 0 .
dy dy y (x)|x=x(y) y (x(y))
dx
Specifically, for y = arcsin x, we have x = sin y. then

d arcsin x dy 1 1 1 1 1
= = = = =p = √ .
dx dx dx (sin y)0 cos y 1 − sin2 y 1 − x2
dy

Exercise 2.2.35. Derive the derivatives of the inverse trigonometric functions

1 1 1
(arctan x)0 = , (arccos x)0 = − √ , (arcsecx)0 = √ .
1 + x2 1 − x2 x x2 − 1

Exercise 2.2.36. Compute the derivatives.

√ √
1. arcsin x. arccos x 1 1 − 1 − x2
4. + log √ .
x 2 1 + 1 − x2
√ arcsin x 1
2. arcsin 1 − x2 . 5. √ + log(1 − x2 ).
1−x 2 2
1+x 1−x
3. arctan . 6. arctan √ .
1−x 2x − x2

Exercise 2.2.37. Compute the derivatives.

r p
x−a p x(a − x)
1. 2 arcsin . 5. − x(a − x) − a arctan .
b−a x−a
1 a
2. arcsin .
a x 1 √ 2 1 x
6. x a − x2 + a2 arctan √ .
1 x 2 2 a − x2
2
3. arcsec .
a a
2 2ax + b x
4. √ arctan √ . 7. x log(x2 + a2 ) + 2a arctan − 2x.
4ac − b2 4ac − b2 a
x
Exercise 2.2.38. Compute the derivative of arcsin . Then use the idea and result of
a
1
Example 2.2.6 to find a function with derivative √ .
ax2 + bx + c
2.3. APPLICATION OF LINEAR APPROXIMATION 113
√ x
Exercise 2.2.39. Compute the derivative of x a2 − x2 + a2 arcsin . Then combine with
√ a
the result of Exercise 2.2.25 to find a function with derivative ax2 + bx + c.

1 x
Exercise 2.2.40. Compute the derivative of arctan . Then for the case a2 ≤ 4b, find a
a a
1
function with derivative 2 . This complements Exercise 2.2.24.
x + ax + b
√
x+1−1 √
Exercise 2.2.41. Compute the derivative of log √ and arctan x. Then find a
x+1+1
1
function with derivative √ .
x ax + b

Exercise 2.2.42. Suppose

 f (x) is invertible,
 with f (1) = 1, f 0 (1) = a. Find the derivative

 1  −1 −1
of the functions f    and f (f (x)) at 1.

 x
f −1
f (x)

Exercise 2.2.43. Explain the formula for the derivative of the inverse function by consid-
ering the inverse of the linear approximation.

Exercise 2.2.44. Find the place on the curve y = x2 where the tangent line is parallel to
the straight line x + y = 1.

Exercise 2.2.45. Show that the area enclosed by the tangent line on the curve xy = a2 and
the coordinate axes is a constant.

Exercise 2.2.46. Let P be a point on the curve y = x3 . The tangent at P meets the curve
again at Q. Prove that the slope of the curve at Q is four times the slope at P .

2.3 Application of Linear Approximation

The linear approximation can be used to determine behaviors of functions. The
idea is that, if the linear approximation of a function has certain behavior, then the
function is likely to have the similar behavior.

2.3.1 Monotone Property and Extrema

We say a function f (x) has local maximum at x0 , if there exists δ > 0 such that
x ∈ domain, |x − x0 | < δ =⇒ f (x) ≤ f (x0 ).
Similarly, f (x) has local minimum at x0 , if
x ∈ domain, |x − x0 | < δ =⇒ f (x) ≥ f (x0 ).
114 CHAPTER 2. DIFFERENTIATION

The function has a (global) maximum at x0 if f (x0 ) ≥ f (x) for all x in the
domain
x ∈ domain =⇒ f (x) ≤ f (x0 ).
The concepts of (global) minimum can be similarly defined. The maximum and
minimum are extrema of the function. A global extreme is also a local extreme.
The local maxima are like the peaks in a mountain, and the global maximum is
like the highest peak.

max

loc max

loc min
min
x
a b

Figure 2.3.1: Local and global extrema.

The following result shows the existence of global extrema in certain case.

Theorem 2.3.1 (Extreme value theorem). Any continuous function on a bounded

closed interval has global maximum and global minimum.

If a function f is increasing on (x0 −δ, x0 ] (i.e., on the left of x0 and including x0 ),

then f (x0 ) is the biggest value on (x0 −δ, x0 ]. If f is also decreasing on [x0 , x0 +δ) (i.e.,
on the right of x0 and including x0 ), then f (x0 ) is the biggest value on [x0 , x0 + δ).
In other words, if f changes from increasing to decreasing as we pass x0 from left to
right, then x0 is a local maximum of f . Similarly, if f changes from decreasing to
increasing at x0 , then x0 is a local minimum.

Example 2.3.1. The square function x2 is strictly decreasing on (−∞, 0] because

x1 < x2 ≤ 0 =⇒ x21 > x22 .

By the same reason, the function is strictly increasing on [0, +∞). This leads to
the local minimum at 0. In fact, by x2 ≥ 0 = |0| for all x, we know x2 has a
global minimum at 0. The function has no local maximum and therefore no global
maximum on R.
On the other hand, if we restrict x2 to [−1, 1], then x2 has global minimum at
0 and global maxima at −1 and 1. If we restrict to [−1, 2], then x2 has global
minimum at 0, global maximum at 2, and local (but not global) maximum at −1. If
we restrict to (−1, 2), then x2 has global minimum at 0, and has no local maximum.
2.3. APPLICATION OF LINEAR APPROXIMATION 115
h π πi
Example 2.3.2. The sine function is strictly increasing on 2nπ − , 2nπ + and
  2 2
π 3π π
is strictly decreasing on 2nπ + , 2nπ + . This implies that 2nπ + are local
2 2 2
π  π
maxima and 2nπ − are local minima. In fact, by sin 2nπ − = −1 ≤ sin x ≤
2 2
 π 
1 = sin 2nπ + , these local extrema are also global extrema.
2

Exercise 2.3.1. Determine the monotone property and find the extrema for |x|

1. on [−1, 1]. 2. on (−1, 1]. 3. on [−2, 1]. 4. on (−∞, 1].

Exercise 2.3.2. Determine the monotone property and find the extrema on R.

1. |x|. 5. x6 . 1 13. ex .
9. .
1 x2 +1
2. x2 + 2x. 6. . 14. e−x .
x 10. cos x.
p
3. |x2 + 2x|. 7. |x|. 11. sin2 x. 15. log x.
1
4. x3 . 8. x + . 12. sin x2 . 16. xx .
x

Exercise 2.3.3. How are the extrema of the function related to the extrema of f (x)?

1. f (x) + a. 2. af (x). 3. f (x)a . 4. af (x) .

Exercise 2.3.4. How are the extrema of the function related to the extrema of f (x)?

1. f (x + a). 2. f (ax). 3. f (x2 ). 4. f (sin x).

Exercise 2.3.5. Is local maximum always the place where the function changes from in-
creasing to decreasing? In other words, can you construct a function f (x) with local
maximum at 0, but f (x) is not increasing on (−δ, 0] for any δ > 0?

Exercise 2.3.6. Compare the global extrema on various intervals in Example 2.3.1 with
Theorem 2.3.1.

2.3.2 Detect the Monotone Property

Suppose f (x) is approximated by the linear function L(x) = f (x0 ) + f 0 (x0 )(x − x0 )
near x0 . The linear function L(x) is increasing if and only if the slope f 0 (x0 ) ≥
0. Since L(x) is very close to f (x), we expect f (x) to be also increasing. The
expectation is true if the linear approximation is increasing everywhere.

Theorem 2.3.2. If f 0 (x) ≥ 0 on an interval, then f (x) is increasing on the interval.

If f 0 (x) > 0 on the interval, then f (x) is strictly increasing on the interval.
116 CHAPTER 2. DIFFERENTIATION

Similar statements hold for decreasing functions. Moreover, for a function on

a closed interval [a, b], we just need the derivative criterion to be satisfied on (a, b)
and the function to be continuous on [a, b].

Example 2.3.3. We have (x2 )0 = 2x < 0 on (−∞, 0) and x2 continuous on (−∞, 0].
Therefore x2 is strictly decreasing on (−∞, 0]. By the similar reason, x2 is strictly
increasing on [0, +∞). This implies that 0 is a local minimum. The conclusion is
consistent with the observation in Example 2.3.1 obtained by direct inspection.

x (−∞, 0) 0 (0, +∞)

f = x2 & loc min 0 %
f 0 = 2x − 0 +

Example 2.3.4. The function f (x) = x3 −3x+1 has derivative f 0 (x) = 3(x+1)(x−1).
The sign of the derivative implies that the function is strictly increasing on (−∞, −1]
and [1, +∞), and is strictly decreasing on [−1, 1]. This implies that −1 is a local
maximum and 1 is a local minimum.

x (−∞, −1) −1 (−1, 1) 1 (1, +∞)

3
f = x − 3x + 1 % loc max 3 & loc min −1 %
f 0 = 3(x + 1)(x − 1) + 0 − 0 +

Example 2.3.5. The function f (x) = sin x − x cos x has derivative f 0 (x) = x sin x.
The sign of the derivative determines the strict monotone property on the interval
[−5, 5] as described in the picture. The strict monotone property implies that −π, 5
are local minima, and −5, π are local maxima.

x −5 (−5, −π) −π (−π, 0) 0 (0, π) π (π, 5) 5

f max & min % % max & min
f0 − 0 + 0 + 0 −

√ (5x + 2)
x2 (x + 1) has derivative f 0 (x) =
3
Example 2.3.6. The function f (x) = √
33x
for x 6= 0. Using the sign of the derivative and the continuity, we get the strict
2
monotone property of the function, which implies that − is a local maximum, and
5
0 is a local minimum.

x (−∞, − 25 ) − 25 (− 25 , 0) 0 (0, +∞)

f % max & min %
f0 + 0 − no +
2.3. APPLICATION OF LINEAR APPROXIMATION 117

local max
local max π
2.377

−π 5
−5 π

−2.377
local min
−π
local min

Figure 2.3.2: Graph of sin x − x cos x.

x3 0 x2 (x2 − 3)
Example 2.3.7. The function f (x) = has derivative f (x) = for
x2 − 1 (x2 − 1)2
x 6= ±1. The sign of the derivative determines the strict
√ monotone property away
√ ±1. The strict monotone property implies that − 3 is a local minimum, and
from
3 is a local maximum.
√ √
x − 3 −1 0 1 3
f % max & & & & min %
f0 + 0 − no − 0 − no − 0 +

Exercise 2.3.7. Determine the monotone property and find extrema.

p
1. x3 − 3x + 2 on R. 4. |x3 − 3x + 2| on R. 7. |x3 − 3x + 2|.
p
2. x3 − 3x + 2 on [−1, 2]. 5. |x3 −3x+2| on [−1, 2]. 8. 7 (x3 − 3x + 2)2 .
1
3. x3 − 3x + 2 on (−1, 2). 6. |x3 −3x+2| on (−1, 2). 9. .
x3 − 3x + 2

Exercise 2.3.8. Determine the monotone property and find extrema.

x 1 + x2
1. on [−1, 1]. 4. on [−1, 0) ∪ (0, 1].
1 + x2 x
sin x h π   πi
2. on [0, 2π]. 5. tan x + cot x on − , 0 ∪ 0, .
1 + sin2 x 4 4
cos x
3. on [0, 2π]. 6. ex + e−x on R.
1 + cos2 x

Exercise 2.3.9. Determine the monotone property and find extrema.

118 CHAPTER 2. DIFFERENTIATION

1. x4 on [−1, 1]. 2. cos4 x on R. 3. sin2 x on R.

Exercise 2.3.10. Determine the monotone property and find extrema.

1. −x4 + 2x2 − 1 on [−2, 2]. 10. x3 + 3 log x on (0, +∞).

√
2. 3 + 2x − x2 on (−1, 3]. 11. x − log(1 + x) on (−1, +∞).
3. |x|p (x + 1) on R. 12. e−x sin x on R.
4. x2 ex on R.
13. x − sin x on [0, 2π].
5. xp ax on (0, +∞).
14. |x − sin x| on [−π, π].
6. |x|e−|x−1| on [−2, 2]. h π πi
15. |x − sin x| on − , .
7. x log x on (0, +∞). 2 2

8. x2 log3 x on (0, +∞). 16. 2 sin x + sin 2x on [0, 2π].

9. xp log x on (0, +∞). 17. 2x − 4 sin x + sin 2x on [0, π].

Exercise 2.3.11. Show that 2x + sin x = c has only one solution. Show that x4 + x = c has
at most two solutions.

Exercise 2.3.12. If f is differentiable and has 9 roots on (a, b), how many roots does f 0
have on (a, b)? If f also has second order derivative, how many roots does f 00 have on
(a, b)?
√
Exercise 2.3.13. Find smallest A > 0, such that log x ≤ A x. Find smallest B > 0, such
B
that log x ≥ − √ .
x

Exercise 2.3.14. A quantity is measured n times, yielding the measurements x1 , . . . , xn .

Find the estimate value x̂ of x that minimizes the squared error (x − x1 )2 + · · · + (x − xn )2 .
√
n
Exercise 2.3.15. Find the biggest term in the sequence n.

2.3.3 Compare Functions

If we apply Theorem 2.3.2 to f (x) − g(x), then we get the following comparison of
two functions.

Theorem 2.3.3. Suppose f (x) and g(x) are continuous for x ≥ a and differentiable
for x > a. If f (a) ≥ g(a) and f 0 (x) ≥ g 0 (x) for x > a, then f (x) ≥ g(x) for x > a.
If f (a) ≥ g(a) and f 0 (x) > g 0 (x) for x > a, then f (x) > g(x) for x > a.

There is a similar statement for the case x < a.

2.3. APPLICATION OF LINEAR APPROXIMATION 119

Example 2.3.8. We have ex > 1 for x > 0 and ex < 1 for x < 0. This is the
comparison of ex with the constant term of the Taylor expansion (or the 0th Taylor
expansion) in Example 2.5.5. How do we compare ex with the first order Taylor
expansion 1 + x?
We have e0 = 1+0. For x > 0, we have (ex )0 = ex > (1+x)0 = 1. Therefore we get
e > 1+x for x > 0. On the other hand, for x < 0, we have (ex )0 = ex < (1+x)0 = 1.
x

Therefore we also get ex > 1 + x for x < 0. We conclude that

ex > 1 + x for x 6= 0.

Example 2.3.9. We claim that

x
< log(1 + x) < x for x > −1, x 6= 0.
1+x
The three functions have the same value 0 at 0. Then we compare their derivatives
 0
x 1 1
= 2
, (log(1 + x))0 = , (x)0 = 1.
1+x (1 + x) 1+x
We have
1 1
< < 1 for x > 0,
(1 + x)2 1+x
and
1 1
2
> > 1 for − 1 < x < 0.
(1 + x) 1+x
The inequalities then follow from Theorem 2.3.3.

Example 2.3.10. For a, b > 0 and p > q > 0, we claim that

1 1
(ap + bp ) p < (aq + bq ) q .
b
By symmetry, we may assume a ≤ b. Then c = > 1, and the inequality means
1
a
that f (x) = (1 + cx ) x is strictly decreasing for x > 0.
By Example 2.2.8, we have
1 + cx
 
0 x x1 −1 x x 1
f (x) = (1 + c ) − 2 log(1 + c ) + c (log c)
x x
1
(1 + cx ) x −1 x
= (c log cx − (1 + cx ) log(1 + cx )).
x2
So we study the monotone property of the function g(t) = t log t. By
g 0 (t) = log t + 1 > 0, for t > e−1 ,
we see that g(t) is increasing for t > e−1 . Since c ≥ 1 and x > 0 implies 1 + cx >
cx ≥ 1 > e−1 , we get cx log cx ≤ (1 + cx ) log(1 + cx ). Therefore f 0 (x) < 0 and f (x)
is decreasing.
120 CHAPTER 2. DIFFERENTIATION

Exercise 2.3.16. State Theorem 2.3.3 for the case x < a.

Exercise 2.3.17. Prove the inequality.

2 π
1. sin x > x, for 0 < x < .
π 2
1
2. ≤ xp + (1 − x)p ≤ 1, for 0 ≤ x ≤ 1, p > 1.
2p−1
√ √
3 1+x 3
3. √ ≤ 2
≤ √ .
6+2 3 2+x 6−2 3

1 x 1 x+1
   
4. 1 + <e< 1+ , for x > 0.
x x
x−y
5. arctan x − arctan y ≤ 2 arctan , for x > y > 0.
2
x2 x2
6. < x − log(1 + x) < , for x > 0. What about −1 < x < 0?
2(1 + x) 2

Exercise 2.3.18. For natural number n and 0 < a < 1, prove that the equation

x2 xn
1+x+ + ··· + = aex
2! n!
has only one solution on (0, +∞).

2.3.4 First Derivative Test

We saw that the local extrema are often the places where the function changes
between increasing and decreasing. If the function is differentiable, then these are the
places where the derivative changes the sign. In particular, we expect the derivatives
at these places to become 0. This leads to the following criterion for the candidates
of local extrema.

Theorem 2.3.4. If f (x) is differentiable at a local extreme x0 , then f 0 (x0 ) = 0.

If f 0 (x0 ) > 0, then the linear approximation L(x) = f (x0 ) + f 0 (x0 )(x − x0 ) of
f near x0 is strictly increasing. This means that L(x) < L(x0 ) for x < x0 and
L(x) > L(x0 ) for x > x0 . Since L is very close to f near x0 , we expect that f is also
“lower” on the left of x0 and “higher” on the right of x0 . In particular, this implies
that x0 is not a local extreme of f . By the similar argument, if f 0 (x0 ) < 0, the x0
is also not a local extreme. This is the reason behind the theorem.
Since our reason makes explicit use of both the left and right sides, the criterion
does not work for one sided derivatives. Therefore for a function f defined on an
interval, the candidates for the local extrema must be one of the following cases:
2.3. APPLICATION OF LINEAR APPROXIMATION 121

f (x) > f (x0 )

L(x) > L(x0 )

L(x0 ) = f (x0 )

f (x) < f (x0 )

x0
L(x) < L(x0 )

Figure 2.3.3: What happens near x0 when f 0 (x0 ) > 0.

1. End points of the interval.

2. Points inside the interval where f is not differentiable.

3. Points inside the interval where f is differentiable and has derivative 0.

Example 2.3.11. The derivative (x2 )0 = 2x vanishes only at 0. Therefore the only
candidate for the local extrema of x2 on R is 0. By x2 ≥ 02 for all x, 0 is a minimum.
If we restrict x2 to the closed interval [−1, 2], then the end points −1 and 2 are
also candidates for the local extrema. By x2 ≤ (−1)2 on [−1, 0] and x2 ≤ 22 on
[−1, 2], −1 is a local maximum and 2 is a global maximum.
On the other hand, the restriction of x2 on the open interval (−1, 2) has no other
candidates for local extrema besides 0. The function has global minimum at 0 and
has no local maximum on (−1, 2).

Example 2.3.12. Consider the function

(
x2 , if x 6= 0,
f (x) =
2, if x = 0,

that modifies the square function by reassigning the value at 0. The function is not
differentiable at 0 and has nonzero derivative away from 0. Therefore on [−1, 2],
the candidates for the local extrema are 0 and the end points −1 and 2. The end
points are also local maxima, like the unmodified x2 . By x2 < f (0) = 2 on [−1, 1],
0 is a local maximum. The modified square function f (x) has no local minimum on
[−1, 2].

Example 2.3.13. The function f (x) = x3 − 3x + 1 in Example 2.3.4 has derivative

f 0 (x) = 3(x + 1)(x − 1). The possible local extrema on R are ±1. These are not the
global extrema on the whole line because limx→−∞ f (x) = −∞ and limx→+∞ f (x) =
+∞.
122 CHAPTER 2. DIFFERENTIATION

If we restrict the function to [−2, 2], then ±2 are also possible local extrema. By
comparing the values

f (−2) = −1, f (−1) = 3, f (1) = −1, f (2) = 3,

we get global minima at −2, 1, and global maxima at 2, −1.

Example 2.3.14. By (x3 )0 = 3x2 , 0 is the only candidate for the local extreme of x3 .
However, we have x3 < 03 for x < 0 and x3 > 03 for x > 0. Therefore 0 is actually
not a local extreme.
The example shows that the converse of Theorem 2.3.4 is not true.

Example 2.3.15. The function f (x) = xe−x has derivative f 0 (x) = (x − 1)ex . The
only possible local extreme on R is at 1. We have limx→−∞ f (x) = −∞ and
limx→+∞ f (x) = 0. We claim that the limits imply that f (1) = e−1 is a global
maximum.
Since the limits at both infinity are < f (1), there is N , such that f (x) < f (1)
for |x| ≥ N . In particular, we have f (±N ) < f (1). Then consider the function
on [−N, N ]. On the bounded and closed interval, Theorem 2.3.1 says that the
continuous function must reach its maximum, and the candidates for the maximum
on [−N, N ] are −N, 1, N . Since f (±N ) < f (1), we see that f (1) is the maximum
on [−N, N ]. Combined with f (x) < f (1) for |x| ≥ N , we conclude that f (1) is the
maximum on the whole real line.

Exercise 2.3.19. Find the global extrema.

√
1. x2 (x − 1)3 on R. 6. sin x2 on [−1, π].

2. x2 (x − 1)3 on [−1, 1]. 7. x log x on (0, +∞).

3. |x2 − 1| on [−2, 1]. 8. x log x on (0, 1].

4. x2 + bx + c on R. 9. xx on (0, 1].

5. 1 − x4 + x5 on R. 10. (x2 + 1)ex on R.

2.3.5 Optimization Problem

Example 2.3.16. Given the circumference a of a rectangle, which rectangle has the
largest area?
a
Let one side of the rectangle be x. Then the other side is − x, and the area
2
a 
A(x) = x −x .
2
h ai
The problem is to find the maximum of A(x) on 0, .
2
2.3. APPLICATION OF LINEAR APPROXIMATION 123

a a a
By A0 (x) = − 2x, the candidates for the local extrema are 0, , . The values
2 4 2
a2 a
of A at the three points are 0, , 0. Therefore the maximum is reached when x = ,
16 4
which means the rectangle is a square.

Example 2.3.17. The distance from a point P = (x0 , y0 ) on the plane to a straight
line ax + by + c = 0 is the minimum of the distance from P to a point (x, y) on the
line. The distance is minimum when the square of the distance

f (x) = (x − x0 )2 + (y − y0 )2

is minimum. Note that when b 6= 0, y is a function of x given by the equation

ax + by + c = 0 and satisfies a + by 0 = 0.

P (x0 , y0 )

ax + by + c = 0

(x, y)

Figure 2.3.4: Distance from a point to a straight line.

From
2
f 0 (x) = 2(x − x0 ) + 2(y − y0 )y 0 = (b(x − x0 ) − a(y − y0 )),
b
we know that f (x) is minimized when

b(x − x0 ) − a(y − y0 ) = 0.

Moreover, recall that (x, y) must also be on the straight line

ax + by + c = 0.

Solving the system of two linear equations, we get

a(ax0 + by0 + c) b(ax0 + by0 + c)
x − x0 = − , y − y0 = − .
a2 + b 2 a2 + b2
The minimum distance is
p |ax0 + by0 + c|
(x − x0 )2 + (y − y0 )2 = √ .
a2 + b 2
On the other hand, when b = 0, it is obvious that the distance is just x0 .
124 CHAPTER 2. DIFFERENTIATION

A
spee
d u
a
x
Q
L
P l y

sp
ee
b

d
v
B

Figure 2.3.5: Snell’s law.

Example 2.3.18. Consider light traveling from a point A in one medium to point B
in another medium. Fermat’s principle says that the path taken by the light is the
path of shortest traveling time.
Let u and v be the speed of light in the respective medium. Let L be the place
where two media meet. Draw lines AP and BQ perpendicular to L. Let the length
of AP, BP, P Q be a, b, l. Let x be the angle by which the light from A hits L. Let
y be the angle by which the light leaves L and reaches B.
The angles x and y are related by
a tan x + b tan y = l.
This can be considered as an equation that implicitly defines y as a function of x.
The derivative of y = y(x) can be obtained by implicit differentiation
a sec2 x
y 0 (x) = − .
b sec2 y
The time it takes for the light to travel from A to B is
a sec x b sec y
T = + .
u v
By thinking of y as a function of x, the time T becomes a function of x. The time
will be shortest when
dT a sec x tan x b sec y tan y 0
0= = + y
dx u v
a sec x tan x b sec y tan y a sec2 x
= −
u v b sec2 y
 
sin x sin y
= a sec2 x − .
u v
This means that the ratio between the sine of the angles x and y is the same as the
ratio between the speeds of light
sin x u
= .
sin y v
2.4. MEAN VALUE THEOREM 125

This is Snell’s law of refraction.

Exercise 2.3.20. A rectangle is inscribed in an isosceles triangle. Show that the biggest
area possible is half of the area of the triangle.

Exercise 2.3.21. Among all the rectangles with area A, which one has the smallest perime-
ter?

Exercise 2.3.22. Among all the rectangles with perimeter L, which one has the biggest
area?

Exercise 2.3.23. A rectangle is inscribed in a circle of radius R. When does the rectangle
have the biggest area?

Exercise 2.3.24. Determine the dimensions of the biggest rectangle inscribed in the ellipse
x2 y 2
+ 2 = 1.
a2 b

Exercise 2.3.25. Find the volume of the biggest right circular cone with a given slant height
l.

Exercise 2.3.26. What is the shortest distance from the point (2, 1) to the parabola y =
2x2 ?

2.4 Mean Value Theorem

If one travels between two cities at the average speed of 100 kilometers per hour,
then we expect that the speed reaches exactly 100 kilometers per hour somewhere
during the trip. Mathematically, let f (t) be the distance traveled by the time t.
Then the average speed from the time a to time b is
f (b) − f (a)
.
b−a
Our expectation can be interpreted as
f (b) − f (a)
= f 0 (c) for some c ∈ (a, b).
b−a

2.4.1 Mean Value Theorem

Theorem 2.4.1 (Mean Value Theorem). If f (x) is continuous on [a, b] and differen-
tiable on (a, b), then there is c ∈ (a, b), such that
f (b) − f (a)
= f 0 (c).
b−a
126 CHAPTER 2. DIFFERENTIATION

The conclusion can also be expressed as

f (b) − f (a) = f 0 (c)(b − a), for some a < c < b,

or
f (a + h) − f (a) = f 0 (a + θh)h, for some 0 < θ < 1.
We also note that the conclusion is symmetric in a, b. Therefore there is no need to
insist a < b.

f (x) f (b)

f (a) L(x)

a c c c b

Figure 2.4.1: Mean value theorem.

Geometrically, the Mean Value Theorem means that the straight line L connect-
ing the two ends (a, f (a)) and (b, f (b)) of the graph of f is parallel to the tangent
of the function somewhere. Figure 2.4.1 suggests that c in the Mean Value Theo-
rem is the place where the the distance between the graphs of f and L has local
extrema. Since such local extrema for the distance f (x) − L(x) always exists by
Theorem 2.3.1, we get (f − L)0 (c) = 0 for some c by Theorem 2.3.4. Therefore
f (b) − f (a)
f 0 (c) = L0 (c) = .
b−a

Example 2.4.1. We try to verify the Mean Value Theorem for f (x) = x3 − 3x + 1 on
[−1, 1]. This means finding c, such that

f (1) − f (−1) −1 − 3
f 0 (c) = 3(c2 − 1) = = = −2.
1 − (−1) 1 − (−1)
1
We get c = ± √ .
3

Example 2.4.2. By the Mean Value Theorem, we have

1
log(1 + x) = log(1 + x) − log 1 = x for some 0 < θ < 1.
1 + θx
Since
x 1
≤ x ≤ x for x > −1,
1+x 1 + θx
2.4. MEAN VALUE THEOREM 127

we conclude that
x
≤ log(1 + x) ≤ x.
1+x
The inequality already appeared in Example 2.3.9.

Example 2.4.3. For the function |x| on [−1, 1], there is no c ∈ (−1, 1) satisfying

f (1) − f (−1) = 0 = f 0 (c)(1 − (−1)).

The Mean Value Theorem does not apply because |x| is not differentiable at 0.

Exercise 2.4.1. Is the conclusion of the Mean Value Theorem true? If true, find c. If not,
explain why.

1. x3 on [−1, 1]. 4. |x3 −3x+1| on [−1, 1]. 7. log x on [1, 2].

2. 2x on [0, 1].
p
5. |x| on [−1, 1]. 8. log |x| on [−1, 1].
1
3. on [1, 2]. 6. cos x on [−a, a]. 9. arcsin x on [0, 1].
x

Exercise 2.4.2. Suppose f (1) = 2 and f 0 (x) ≤ 3 on R. How large and how small can f (4)
be? What happens when the largest or the smallest value is reached? How about f (−4)?

Exercise 2.4.3. Prove inequality.

1. | sin x − sin y| ≤ |x − y|.
x−y x x−y
2. < log < , for x > y > 0.
x y y
3. | arctan x − arctan y| ≤ |x − y|.

Exercise 2.4.4. Find the biggest interval on which |ex −ey | > |x−y|? What about |ex −ey | <
|x − y|?

Exercise 2.4.5. Suppose f (x) is continuous at x0 and differentiable on (x0 −δ, x0 )∪(x0 , x0 +
δ). Prove that if limx→x0 f 0 (x) = l converges, then f (x) is differentiable at x0 and f 0 (x0 ) =
l.

2.4.2 Criterion for Constant Function

The Mean Value Theorem can be used to prove Theorem 2.3.2: If f 0 ≥ 0 on an
interval, then for any x1 < x2 on the interval, by the Mean Value Theorem, we have

f (x2 ) − f (x1 ) = f 0 (c)(x2 − x1 ) ≥ 0.

For the special case f 0 = 0 throughout the interval, the argument gives the following
result.
128 CHAPTER 2. DIFFERENTIATION

Theorem 2.4.2. If f 0 (x) = 0 for all x ∈ (a, b), then f (x) is a constant on (a, b).

Applying the theorem to f (x) − g(x), we get the following result.

Theorem 2.4.3. If f 0 (x) = g 0 (x) for all x ∈ (a, b), then there is a constant C, such
that f (x) = g(x) + C on (a, b).

Example 2.4.4. The function ex satisfies f (x)0 = f (x). Are there any other functions
satisfying the equation?
If f (x)0 = f (x). Then

(e−x f (x))0 = (e−x )0 f (x) + e−x (f (x))0 = ex (−f (x) + f 0 (x)) = 0.

Therefore e−x f (x) = C is a constant, and f (x) = Cex .

0
x2

0 0
Example 2.4.5. Suppose f (x) = x and f (1) = 2. Then f (x) = implies
2
x2 1
f (x) = + C for some constant C. By taking x = 1, we get 2 = + C. Therefore
2 2
3 x2 + 3
C = and f (x) = .
2 2

Example 2.4.6. By
1 1
(arcsin x)0 = √ , (arccos x)0 = − √ .
1 − x2 1 − x2
we have (arcsin x+arccos x)0 = 0, and we have arcsin x+arccos x = C. The constant
can be determined by taking a special value x = 0
π π
C = arcsin 0 + arccos 0 = 0 + = .
2 2
Therefore we have
π
arcsin x + arccos x = .
2

Exercise 2.4.6. Prove that a differentiable function is linear on an interval if and only if its
derivative is a constant.

Exercise 2.4.7. Find all functions on an interval satisfying the following equations.

1. f 0 (x) = −2f (x). 3. f 0 (x) = f (x)2

2. f 0 (x) = xf (x). 4. f 0 (x)f (x) = 1.

Exercise 2.4.8. Prove equality.

2.4. MEAN VALUE THEOREM 129

π
1. arctan x + arctan x−1 = , for x 6= 0.
2
1
2. 3 arccos x − arccos(3x − 4x3 ) = π, for |x| ≤ .
2
x+a
3. arctan − arctan x = arctan a, for ax < 1.
1 − ax
x+a
4. arctan − arctan x = arctan a − π, for ax > 1.
1 − ax

2.4.3 L’Hospital’s Rule

The following limits cannot be computed by simple arithmetic rules.

0 x3 − 1 sin x log(1 + x)
: lim , lim , lim ;
0 x→1 x−1 x→0 x x→0 x
∞ log x log x x2
: lim −1 , lim , lim x ;
∞ x→0 x  x→∞ x
x x→∞ e
1
1∞ : lim 1 + , lim (1 + sin x)log x ;
x→∞ x x→0
 
1 1
∞ − ∞ : lim − x .
x→0 x e −1

We say these limits are indeterminate. Other indeterminates include 0 · ∞, ∞ + ∞,

00 , ∞0 . The derivative can help us computing such limits.

Theorem 2.4.4 (L’Hospital’s Rule). Suppose f (x) and g(x) are differentiable func-
tions on (a, b), with g 0 (x) 6= 0. Suppose

1. Either limx→a+ f (x) = limx→a+ g(x) = 0 or limx→a+ f (x) = limx→a+ g(x) = ∞.

f 0 (x)
2. limx→a+ = l converges.
g 0 (x)
f (x)
Then limx→a+ = l.
g(x)

0 ∞
The theorem computes the limits of the indeterminates of type or . The
0 ∞
conclusion is the equality

f (x) f 0 (x)
lim+ = lim+ 0
x→a g(x) x→a g (x)

whenever the right side converges. It is possible that the left side converges but the
right side diverges.
130 CHAPTER 2. DIFFERENTIATION

The theorem also has a similar left sided version, and the left and right sided
versions may be combined to give the two sided version. Moreover, l’Hospital’s rule
also allows a or l to be any kind of infinity.
The reason behind l’Hospital’s rule is the following version of the Mean Value
Theorem, which can be proved similar to the Mean Value Theorem.

Theorem 2.4.5 (Cauchy’s Mean Value Theorem). If f (x) and g(x) are continuous
on [a, b] and differentiable on (a, b), such that g 0 (x) 6= 0 on (a, b), then there is
c ∈ (a, b), such that
f (b) − f (a) f 0 (c)
= 0 .
g(b) − g(a) g (c)

Consider the parametrized curve (g(t), f (t)) for t ∈ [a, b]. The theorem says that
the straight line connecting the two ends (g(a), f (a)) and (g(b), f (b)) of the curve is
f 0 (c)
parallel to the tangent of the curve somewhere. The slope of the tangent is 0 .
g (c)

f 0 (c)
g 0 (c)

(g(a), f (a))

(g(b), f (b))

Figure 2.4.2: Cauchy’s Mean Value Theorem.

For the case limx→a+ f (x) = limx→a+ g(x) = 0 of l’Hospital’s rule, we may extend
f and g to continuous functions on [a, b) by assigning f (a) = g(a) = 0. Then for
any a < x < b, we may apply Cauchy’s Mean Value Theorem to the functions on
[a, x] and get
f (x) f (x) − f (a) f 0 (c)
= = 0 for some c ∈ (a, x).
g(x) g(x) − g(a) g (c)
f 0 (c) f (x)
Since x → a+ implies c → a+ , we conclude that limc→a+ 0
= l implies limx→a+ =
g (c) g(x)
l.
1
By changing the variable x to , it is not difficult to extend the proof to the case
x
∞
a = ±∞. The proof for the type is must more complicated and is omitted here.
∞

Example 2.4.7. In Example 1.2.15, we proved that limx→+∞ ax = 0 for 0 < a < 1.
Exercise 1.6.15 extended the limit to limx→+∞ x2 ax = 0. We derive the second limit
from the first one by using l’Hospital’s rule.
2.4. MEAN VALUE THEOREM 131

1
We have b = > 1, and limx→+∞ ax = 0 is the same as limx→+∞ bx = ∞. We
a
x2 ∞
also have limx→+∞ x2 = ∞. Therefore limx→+∞ x2 ax = limx→+∞ x is of type ,
b ∞
and we may apply l’Hospital’s rule (twice)

x2 (x2 )0 2x
lim x
=(3) lim x 0
= lim x
x→+∞ b x→+∞ (b ) x→+∞ b log b
0
(2x) 2
=(2) lim x 0
= lim x =(1) 0.
x→+∞ (b log b) x→+∞ b (log b)2

Here is the precise reason behind the computation. The equality =(1) is from
Example 1.2.15. Then by l’Hospital’s rule, the convergence of the right side of =(2)
implies the convergence of the left side of =(2) and the equality =(2) itself. The left of
=(2) is the same as the right side of =(3) . By l’Hospital’s rule gain, the convergence of
the right side of =(3) implies the convergence of the left side of =(3) and the equality
=(3) .

sin x 0
Example 2.4.8. Applying l’Hospital’s rule to the limit limx→0 of type , we get
x 0

sin x (sin x)0

lim = lim = lim cos x = 1.
x→0 x x→0 (x)0 x→0

However, this argument is logically circular because it makes use of the formula
(sin x)0 = cos x. A special case of this formula is

sin x
lim = (sin x)0 |x=0 = 1,
x→0 x

which is exactly the conclusion we try to get.

Exercise 2.4.9. Are the application of l’Hospital’s rule logically circular?

x cos x − 1 x2 − 1
1. limx→0 . 3. limx→0 . 5. limx→1 .
sin x x2 x3 − 1
cos x − 1 x2 − 1 log(1 + x)
2. limx→0 . 4. limx→1 . 6. limx→0 .
x x−1 x
132 CHAPTER 2. DIFFERENTIATION

Example 2.4.9. By blindly using l’Hospital’s rule four times, we have

sin2 x − sin x2 2 sin x cos x − 2x cos x2

lim = lim
x→0 x4 x→0 4x3
cos x − sin2 x − cos x2 + 2x2 sin x2
2
= lim
x→0 6x2
−4 cos x sin x + 6x sin x2 + 4x3 cos x2
= lim
x→0 12x
2
4 sin x − 4 cos x + 6 sin x2 + 24x2 cos x2 − 8x4 sin x2
2
= lim
x→0 12
1
=− .
3
We find that it is increasingly difficult to calculate the derivatives. The following
compute the the limit after calculating the derivatives twice.

sin2 x − sin x2 2 sin x cos x − 2x cos x2

lim = lim
x→0 x4 x→0 4x3
cos x − sin2 x − cos x2 + 2x2 sin x2
2
= lim
x→0 6x2 !
2
1 − cos x2 1

1 sin x 2
= lim − + + sin x
x→0 3 x 6x2 3
1 1 1
= − · 12 + 0 + · 0 = − .
3 3 3
In fact, the smartest way is not to calculate the derivatives at all. See Example
2.5.11.

Exercise 2.4.10. Use l’Hospital’s rule to compute the limits.

sin x − x 1 − cos x2
1. limx→0 . 5. limx→0 .
x3 x3 sin x
sin x − tan x xx − x
2. limx→0 . 6. limx→1 .
x3 log x − x + 1
ex − esin x x − tan x
3. limx→0 . 7. limx→0 .
x3 x − sin x
cos(sin x) − cos x (x − 1) log x
4. limx→0 . 8. limx→1 .
x4 1 + cos πx
∞
Example 2.4.10. The limit limx→0+ x log x is of type 0 · ∞. We convert into type
∞
and apply l’Hospital’s rule (in the second equality)

log x x−1
lim+ x log x = lim+ = lim = lim+ −x = 0.
x→0 x→0 x−1 x→0+ −x−2 x→0
2.5. HIGH ORDER APPROXIMATION 133

The similar argument gives

lim xp log x = 0, for p > 0.

x→0+

Taking a positive power of the limit, we further get

lim xp (− log x)q = 0, for p, q > 0.

x→0+

1
By converting x to , we also have
x
(log x)q
lim = 0, for p, q > 0.
x→+∞ xp

Example 2.4.11. We compute the limit in Example 2.5.13 by first converting it to

0
type and then applying l’Hospital’s rule
0
ex − 1 − x ex − 1 ex
 
1 1 1
lim − = lim = lim x = lim x = .
x→0 x ex − 1 x→0 x(ex − 1) x→0 e − 1 + xex x→0 2e + xex 2

x + sin x
Example 2.4.12. If we apply l’Hospital’s rule to the limit limx→∞ of type
x
∞
, then we get
∞
x + sin x
lim = lim (1 + cos x).
x→∞ x x→∞

We find that the left converges and the right diverges. The reason for the l’Hospital’s
rule to fail is that the second condition is not satisfied.

2.5 High Order Approximation

Linear approximations can be used to solve many problems. When linear approxi-
mations are not enough, however, we may use high order approximations.

Definition 2.5.1. An n-th order approximation of f (x) at x0 is a degree n polynomial

P (x) = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · + an (x − x0 )n ,

such that for any  > 0, there is δ > 0, such that

|x − x0 | < δ =⇒ |f (x) − P (x)| ≤ |x − x0 |n .

A function is n-th order differentiable if it has n-order approximation.

134 CHAPTER 2. DIFFERENTIATION

The error Rn (x) = f (x) − P (x) of the approximation is called the remainder.
The definition means that
f (x) = P (x) + Rn (x)
= a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · + an (x − x0 )n + o((x − x0 )n ),
where the “small o” notation means that the remainder term satisfies
f (x) − P (x) Rn (x)
lim n
= lim = 0.
x→x0 (x − x0 ) x→x0 (x − x0 )n

The n-th order approximation of a function is unique. See Exercise 2.5.7. More-
over, if m < n, then the truncation a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · + am (x − x0 )m
is the m-th order approximation of f at x0 . After all, if we have the 10th order
approximation, then we should also have the 5th order approximation.

Example 2.5.1. We have

x4 = (1 + (x − 1))4 = 1 + 4(x − 1) + 6(x − 1)2 + 4(x − 1)3 + (x − 1)4 .
n o
For any  > 0, we have |x − 1| < δ = min 1, implying
5
|x4 − 1 − 4(x − 1) − 6(x − 1)2 | = |4(x − 1)3 + (x − 1)4 | ≤ (4 + |x − 1|)|x − 1|3

≤ (4 + 1) |x − 1|2 = |x − 1|2 .
5
Therefore 1 + 4(x − 1) + 6(x − 1)2 is the quadratic approximation of x4 at 1. By
similar argument, we get approximations of other orders.
linear : 1 + 4(x − 1),
quadratic : 1 + 4(x − 1) + 6(x − 1)2 ,
cubic : 1 + 4(x − 1) + 6(x − 1)2 + 4(x − 1)3 ,
quartic : 1 + 4(x − 1) + 6(x − 1)2 + 4(x − 1)3 + (x − 1)4 ,
quintic : 1 + 4(x − 1) + 6(x − 1)2 + 4(x − 1)3 + (x − 1)4 ,
..
.

Example 2.5.2. The limit

cos x − 1 1
lim 2
=−
x→0 x 2
in Example 1.5.18 can be interpreted as
1
cos x − 1 + x2
lim 2 = 0.
x→0 x 2

This means that cos x is second order differentiable at 0, with quadratic approxima-
1
tion 1 − x2 .
2
2.5. HIGH ORDER APPROXIMATION 135

1 xn+1
Exercise 2.5.1. Prove that = 1 + x + x2 + · · · + xn + . What does this tell you
1−x 1−x
1
about the differentiability of at 0?
1−x

1 1
Exercise 2.5.2. Show that and are differentiable of arbitrary order. What
1+x 1 + x2
are their high order approximations?

Exercise 2.5.3. What is the n-th order approximation of 1 + 2x + 3x2 + · · · + 100x100 at 0?

Exercise 2.5.4. Use l’Hospital’s rule to compute the limits. Then interpret your results as
high order approximations.

sin x − x sin x − tan x

1. limx→0 . 5. limx→0 .
x3 x3
 
sin x2 − x2 1 x − 1 − x − 1 x2 .
2. limx→0 . 6. limx→0 e
x6 x3 2

sin2 x − x2
 
1 1 2
3. limx→0 . 7. limx→0 log(1 + x) − x + x .
x4 x3 2
 
1 1 3 2 log x + (x − 1)(x − 3)
4. limx→0 sin x − x + x . 8. limx→1 .
x5 6 (x − 1)3

Exercise 2.5.5. For what choice of a, b, c is the function

(
x4 , if x ≥ 1,
2
a + bx + cx , if x < 1,

second order differentiable at 1? Is it possible for the function to be third order differen-
tiable?

Exercise 2.5.6. Suppose f (x) is second order differentiable at 0. Show that

3f (x) − 3f (2x) + f (3x) = f (0) + o(x2 ).

P (x)
Exercise 2.5.7. Suppose P (x) = a0 + a1 (x − a) + a2 (x − a)2 satisfies limx→a = 0.
(x − a)2
Prove that a0 = a1 = a2 = 0. Then explain that the result means the uniqueness of
quadratic approximation. Moreover, extend the result to high order approximation.

Exercise 2.5.8. Suppose P (x) is the n-order approximation of f (x). What is the n-order
approximation of f (−x)? Then use Exercise 2.5.7 to explain that the high order ap-
proximation of an even function has not odd power terms. What about the high order
approximation of an odd function?
136 CHAPTER 2. DIFFERENTIATION

2.5.1 Taylor Expansion

The linear approximation may be computed by the derivative. The high order
approximation may be computed by repeatedly taking the derivative. The idea is
suggested by the following example. By applying l’Hospital’s rule three times, we
get a more precise limit than the one in Example 2.5.2
1
cos x − 1 + x2
lim 2 = lim − sin x − x = lim − cos x − 1 = lim sin x = 0.
x→0 x 3 x→0 3x2 x→0 3 · 2x x→0 3 · 2 · 1

Each application of the l’Hospital’s rule means taking derivative once. Therefore we
get the third order approximation of cos x at 0 by taking derivative three times.
If f (x) is differentiable everywhere on an open interval, then the derivative f 0 (x)
is a function on the open interval. If the derivative function f 0 (x) is also differ-
entiable, then we get the second order derivative f 00 (x) = (f 0 (x))0 . If the function
f 00 (x) is yet again differentiable, then taking the derivative one more time gives the
third order derivative f 000 (x) = (f 00 (x))0 . The process may continue and we have
the n-th order derivative f (n) (x). The Leibniz notation for the high order derivative
dn f
f (n) (x) is .
dxn
1
Let f (x) = cos x and P (x) = 1 − x2 . The key to the repeated application of
2
the l’Hospital’s rule is that the numerator is always 0 at x0 = 0. This means that

f (x0 ) = P (x0 ), f 0 (x0 ) = P 0 (x0 ), f 00 (x0 ) = P 00 (x0 ), f 000 (x0 ) = P 000 (x0 ).

In general, if P (x) = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + a3 (x − x0 )3 , then the above

equalities become

f (x0 ) = a0 , f 0 (x0 ) = a1 , f 00 (x0 ) = 2a2 , f 000 (x0 ) = 3 · 2a3 .

Theorem 2.5.2. If f (x) has n-th order derivative at x0 , then f is n-th order differ-
entiable, with n-th order approximation

f 00 (x0 ) f (n) (x0 )

Tn (x) = f (x0 ) + f 0 (x0 )(x − x0 ) + (x − x0 )2 + · · · + (x − x0 )n .
2! n!

The polynomial Tn is called the n-th order Taylor expansion of f .

Note that the existence of the derivative f (n) (x0 ) implicitly assumes that f (k) (x)
exists for all x near x0 and all k < n. The theorem gives one way (but not the only
way!) to compute the high order approximation in case the function has high order
derivative. However, we will show in Example 2.5.8 that it is possible to have high
order approximation without the existence of the high order derivative. Here the
concept (high order differentiability) is strictly weaker than the computation (high
order derivative).
2.5. HIGH ORDER APPROXIMATION 137

Example 2.5.3. The high order derivatives of the power function xp are
(xp )0 = pxp−1 ,
(xp )00 = p(p − 1)xp−2 ,
..
.
(xp )(n) = p(p − 1) · · · (p − n + 1)xp−n .
More generally, we have
((a + bx)p )(n) = p(p − 1) · · · (p − n + 1)bn (a + bx)p−n .
For a = b = 1, we get the Taylor expansion at 0
p(p − 1) 2 p(p − 1) · · · (p − n + 1) n
(1 + x)p = 1 + px + x + ··· + x + o(xn ).
2! n!
For a = 1, b = −1 and p = −1, we get the Taylor expansion at 0
1
= 1 + x + x2 + · · · + xn + o(xn ).
1−x
You may compare with Exercise 2.5.1.

Example 2.5.4. By (log x)0 = x−1 and the derivatives from Example 2.5.3, we have
(n − 1)!
(log x)(n) = (−1)n−1 . This gives the Taylor expansion at 1
xn
1 1 1
log x = (x − 1) − (x − 1)2 + (x − 1)3 − · · · + (−1)n+1 (x − 1)n + o((x − 1)n ).
2 3 n
This can also be expressed as a Taylor expansion at 0
1 1 1
log(1 + x) = x − x2 + x3 − · · · + (−1)n+1 xn + o(xn ).
2 3 n

Example 2.5.5. By (ex )0 = ex , it is easy to see that (ex )(n) = ex for all n. This gives
the Taylor expansion at 0
1 1 1
ex = 1 + x + x2 + · · · + xn + o(xn ).
1! 2! n!

Example 2.5.6. The high order derivatives of sin x and cos x are 4-periodic in the
sense that sin(n+4) x = sin(n) x and cos(n+4) x = cos(n) x, and are given by
(sin x)0 = cos x, (cos x)0 = − sin x,
(sin x)00 = − sin x, (cos x)00 = − cos x,
(sin x)000 = − cos x, (cos x)000 = sin x,
(sin x)0000 = sin x, (cos x)0000 = cos x.
138 CHAPTER 2. DIFFERENTIATION

This gives the Taylor expansions at 0

1 3 1 5 1
sin x = x − x + x − · · · + (−1)n+1 x2n−1 + o(x2n ),
3! 5! (2n − 1)!
1 1 1 2n
cos x = 1 − x2 + x4 − · · · + (−1)n x + o(x2n+1 ).
2! 4! (2n)!

Note that we have o(x2n ) for sin x at the end, which is more accurate than o(x2n−1 ).
The reason is that the 2n-th term 0 · x2n is omitted from the expression, so that the
approximation is actually of 2n-th order. The similar remark applies to cos x.
We also note that the Taylor expansions of ex , sin x, cos x are related by the
equality √
eix = cos x + i sin x, i = −1.

Exercise 2.5.9. Prove the following properties of high order derivative

(f + g)(n) = f (n) + g (n) ,

(cf )(n) = cf (n) ,
n
(n)
X n!
(f g) = f (i) g (j) .
i!(n − i)!
i=0

Exercise 2.5.10. Prove the chain rule for second order derivative

(g(f (x)))00 = g 00 (f (x))f 0 (x)2 + g 0 (f (x))f 00 (x).

Exercise 2.5.11. Compute derivatives of all order.

1. ax . 4. cos(ax + b). ax + b
7. .
cx + d
2. eax+b . 5. log(ax + b).
ax + b 1
3. sin(ax + b). 6. log . 8. .
cx + d (ax + b)(cx + d)

Exercise 2.5.12. Compute high order derivatives.

x
d2

1. (tan x)000 . 3. (sin x2 )000 . 5. (arctan x)00 . 1
7. 1+ .
dx2 x
2. (sec x)000 . 4. (arcsin x)00 . 6. (xx )00 .

Exercise 2.5.13. Use high order derivatives to find high order approximations.

1. ax , n = 5, at 0. 4. sin2 x, n = 6, at π. 7. x3 ex , n = 5, at 0.
2
2. ax , n = 5, at 1. 5. ex , n = 6, at 0. 8. x3 ex , n = 5, at 1.
2
3. sin2 x, n = 6, at 0. 6. ex , n = 6, at 1. 9. ex sin x, n = 5 at 1.
2.5. HIGH ORDER APPROXIMATION 139

Exercise 2.5.14. Compute high order derivatives.

1. (x2 + 1)ex . 2. (x2 + 1) sin x. 3. x2 (x − 1)p . 4. x log x.

1 (−1)n 1
Exercise 2.5.15. Prove (xn−1 e x )(n) = ex .
xn+1
n
Exercise 2.5.16. Prove (eax sin(bx + c))(n) = (a2 + b2 ) 2 eax sin(bx + c + nθ), where sin θ =
b
√ . What is the similar formula for (eax cos(bx + c))(n) ?
2
a +b 2

Exercise 2.5.17. Suppose f (x) has second order derivative near x0 . Prove that

f (x0 + h) + f (x0 − h) − 2f (x0 )

f 00 (x0 ) = lim .
h→0 h2

Exercise 2.5.18. Compare ex , sin x, cos x, log(1 + x) with their Taylor expansions. For ex-
x2 x3 xn
ample, is ex bigger than or smaller than 1 + x + + + ··· + ?
2! 3! n!

2.5.2 High Order Approximation by Substitution

The functions (and their variations) in Examples 2.5.3 through 2.5.11 are the only
ones that we can compute all the high order derivative functions. These give the ba-
sic examples of high order approximations. We get other high order approximations
by combining the basic ones.

b
Example 2.5.7. Substituting x by x in the Taylor expansion of (1 + x)p , we get
a
 p
p p b
(a + bx) = a 1 + x
a
p(p − 1) b2 2

b
= ap 1 + p x + x + ···
a 2! a2
p(p − 1) · · · (p − n + 1) bn n
 n 
b n
+ n
x +o x
n! a an
p(p − 1) p−2 2 2
= ap + pap−1 bx + a b x + ···
2!
p(p − 1) · · · (p − n + 1) p−n n n
+ a b x + o(xn ).
n!
 n 
b n
Note that we used ap o n
x = o(xn ) in the computation. The reason is that
 n  a  n 
b n b n
o n
x really means a function R x , where R(x) is the remainder of the
a an
140 CHAPTER 2. DIFFERENTIATION

R(x)
n-th order Taylor expansion of (1 + x)p . Since limx→0 = 0, we get
xn
bn n
 
p
aR x
an ap R(y) R(y)
lim = lim n = ap−n bn lim n = 0.
x→0 xn y→0 a y→0 y
n
yn
b
n
 
b
This means ap R xn = o(xn ).
an
Further substitution of a, b, x by x0 , 1, x − x0 gives the Taylor expansion of xp at
x0

xp = (x0 + (x − x0 ))p
p(p − 1) p−2
= xp0 + pxp−1
0 (x − x0 ) + x0 (x − x0 )2 + · · ·
2!
p(p − 1) · · · (p − n + 1) p−n
+ x0 (x − x0 )n + o((x − x0 )n ).
n!
The Taylor expansion can also be obtained from the high order derivative in Example
2.5.3.

1
Example 2.5.8. The Taylor expansion of at 0 in Example 2.5.3 induces the
1−x
following approximations
1
= 1 − x + x2 − · · · + (−1)n xn + o(xn ),
1+x
1
2
= 1 − x2 + x4 − · · · + (−1)n x2n + o(x2n ).
1+x
Similar to the Taylor expansions of sin x and cos x, we expect that the odd power
1
terms vanish in the Taylor expansion of . Therefore the remainder should be
1 + x2
improved to o(x2n+1 ). To get the improved remainder, we consider the 2(n + 1)-th
1
order Taylor expansion of
1 + x2
1
2
= 1 − x2 + x4 − · · · + (−1)n x2n + (−1)n+1 x2(n+1) + o(x2(n+1) ).
1+x

This shows that the remainder of the 2n-th order Taylor expansion is (−1)n+1 x2(n+1) +
R(x)
R(x), where R(x) satisfies limx→0 2(n+1) = 0. By
x
(−1)n+1 x2(n+1) + R(x)
 
n+1 R(x)
lim = 0 = lim (−1) x + 2(n+1) x = 0,
x→0 x2n+1 x→0 x
2.5. HIGH ORDER APPROXIMATION 141

we get
(−1)n+1 x2(n+1) + o(x2(n+1) ) = o(x2n+1 ),

and the improved approximation

1
= 1 − x2 + x4 − · · · + (−1)n x2n + o(x2n+1 ).
1 + x2

1
Finally, it is easy to see that has derivative of any order. From the
1 + x2
coefficients in the Taylor expansion, we get
(
dn 
  
1 0, if n = 2k − 1,
n 2
= n!an = k
dx x=0 1 + x
 (−1) (2k)!, if n = 2k.

It is practically impossible to get this by directly computing the high order deriva-
tives (i.e., by repeatedly taking derivatives).

Exercise 2.5.19. Explain and justify the following claims about remainders.

1. o(x5 ) = o(x3 ). 3. o(x4 ) + o(x5 ) = o(x3 ). 5. o(x3 )o(x5 ) = o(x8 ).

2. o(x5 ) + o(x5 ) = o(x5 ). 4. x3 o(x5 ) = o(x8 ). 6. o(x3 ) + x5 = o(x3 ).

1
Exercise 2.5.20. Find the Taylor expansion of at 0, and the high order derivatives
1 − x3
of the function at 0.

Exercise 2.5.21. Use the high order derivatives in Example 2.5.8 to find the Taylor expan-
sion of arctan x at 0.

1
Exercise 2.5.22. Find the Taylor expansion of √ at 0. Find the high order derivatives
1 − x2
of the function at 0. Then find the Taylor expansion of arcsin x at 0.

Example 2.5.9. The Taylor expansion of ex at 0 induces

1 1 1
e−x = 1 − x + x2 − · · · + (−1)n xn + o(xn ),
1! 2! n!
1 1
ex = ex0 ex−x0 = ex0 − ex0 (x − x0 ) + ex0 (x − x0 )2 − · · ·
1! 2!
1
+ (−1)n ex0 (x − x0 )n + o((x − x0 )n ),
n!
2 1 2 1 4 1
ex = 1 + x + x + · · · + x2n + o(x2n+1 ).
1! 2! n!
142 CHAPTER 2. DIFFERENTIATION

2
Note that we have the more accurate remainder o(x2n+1 ) for ex for the reason similar
2
to Example 2.5.8. Moreover, the Taylor expansion of ex gives


2
0, if n = 2k − 1,
(ex )(n) |x=0 = (2k)!
 , if n = 2k.
k!

Example 2.5.10. The high order approximation of x2 ex at 0 is

 
2 x 2 1 1 2 1 n n
x e = x 1 + x + x + · · · + x + o(x )
1! 2! n!
1 1 1
= x2 + x3 + x4 + · · · + xn+2 + o(xn+2 ).
1! 2! n!

Here we use x2 o(xn ) = o(xn+2 ). The n-th order approximation is

1 3 1 4 1
x2 e x = x2 + x + x + ··· + xn + o(xn ).
1! 2! (n − 2)!

Can you find the n-th order derivative of x2 ex at 0?

On the other hand, to find the high order approximation of x2 ex at 1, we express

2.5. HIGH ORDER APPROXIMATION 143

the variables in terms of x − 1 and get

x2 ex = ((x − 1) + 1)2 e(x−1)+1 = e((x − 1)2 + 2(x − 1) + 1)ex−1

n
!
X 1
= e((x − 1)2 + 2(x − 1) + 1) (x − 1)i + o((x − 1)n )
i=0
i!
n
X 1
=e (x − 1)i+2 + o((x − 1)n+2 )
i=0
i!
n
X 1
+ 2e (x − 1)i+1 + o((x − 1)n+1 )
i=0
i!
n
X 1
+e (x − 1)i + o((x − 1)n )
i=0
i!
n
X 1
=e (x − 1)i + o((x − 1)n )
i=2
(i − 2)!
n
X 1
+ 2e(x − 1) + 2e (x − 1)i + o((x − 1)n )
i=2
(i − 1)!
n
1
X
+ e + e(x − 1) + e (x − 1)i + o((x − 1)n )
i=2
i!
n  
X 1 2 1
= e + 3e(x − 1) + e + + (x − 1)i + o((x − 1)n )
i=2
(i − 2)! (i − 1)! i!
n
X i2 + i + 1
= e + 3e(x − 1) + e (x − 1)i + o((x − 1)n ).
i=2
i!

Exercise 2.5.23. Use the basic Taylor expansions to find the high order approximations
and derivatives of functions in Exercise 2.5.11.

Exercise 2.5.24. Use the basic Taylor expansions to find the high order approximations
and derivatives at 0.
1 5. log(1 + 3x + 2x2 ). 9. sin x cos 2x.
1. .
x(x + 1)(x + 2)
√ 1 + x2 10. sin x cos 2x sin 3x.
2. 1 − x2 . 6. log .
1 − x3
√
3. 1 + x3 . 7. e2x . 11. sin x2 .
2
4. log(1 + x2 ). 8. ax . 12. sin2 x.

Exercise 2.5.25. Use the basic Taylor expansions to find high order approximations and
high order derivatives.
144 CHAPTER 2. DIFFERENTIATION

1. x3 + 5x − 1 at 1. 5. e−2x at 4. 9. sin x at π.
2. xp at −3. 6. log x at 2. 10. cos x at π.
x+3 π
3. at 1. 7. log(3 − x) at 2. 11. sin 2x at .
x+1 4
√ π
4. x + 1 at 1. 8. sin x at . 12. sin2 x at π.
2

Exercise 2.5.26. Use the basic Taylor expansions to find high order approximations and
high order derivatives at x0 .

1. ax . 3. log x. 5. sin x cos 2x.

2. x2 ex . 4. sin x. 6. sin2 x.

2.5.3 Combination of High Order Approximations

So far we only used simple substitutions to get new approximations. In the subse-
quent examples, we compute more sophisticated combinations of approximations.

Example 2.5.11. By the Taylor expansion of sin x, we have

 2  
2 2 1 3 1 5 6 2 1 6 2 4
sin x − sin x = x − x + x + o(x ) − x − x + o((x ) )
6 120 6
 
2 1 4 1 6 1 6 6 2 6
= x − x + x + x + o(x ) + 2xo(x ) + · · ·
3 36 60
 
1
− x2 − x6 + o(x8 )
6
1 19
= − x4 + x6 + o(x7 ).
3 90

The term o(x7 ) at the end comes from

R(x) R(x)2 xR(x)

lim = 0 =⇒ lim = 0, lim = 0, . . . .
x→0 x6 x→0 x7 x→0 x7

In particular, we get the limit in Example 2.4.9

sin2 x − sin x2 1
lim = − .
x→0 x4 3

We also get the high order derivatives of f (x) = sin2 x − sin x2 at 0

f 0 (0) = f 00 (0) = f 000 (0) = f (5) (0) = 0, f (4) (0) = −8, f (6) (0) = 152.
2.5. HIGH ORDER APPROXIMATION 145

Example 2.5.12. We may compute the Taylor expansions of tan x and sec x from the
1
Taylor expansions of sin x, cos x and
1−x
1 1
sec x = =
cos x 1 1
1 − x2 + x4 + o(x5 )
2 24
   2
1 2 1 4 5 1 2 1 4 5
=1+ x − x + o(x ) + x − x + o(x )
2 24 2 24
 3
1 2 1
+ x − x4 + o(x5 ) + o(x6 )
2 24
1 2 1 1
= 1 + x − x4 + x4 + o(x5 )
2 24 4
1 2 5 4
= 1 + x + x + o(x5 ),
2 24  
1 3 1 5 6 1 2 5 4 5
tan x = sin x sec x = x − x + x + o(x ) 1 + x + x + o(x )
6 120 2 24
1 3 1 5 1 3 1 5 5 5
=x− x + x + x − x + x + o(x6 )
6 120 2 12 24
1 3 2 5 6
= x + x + x + o(x ).
3 15
(4) (5)
The expansions give (sec x)x=0 = 5 and (tan x)x=0 = 16.
 
1 1
Example 2.5.13. We computed limx→0 − x by using l’Hospital’s rule in
x e −1
1
Example 2.4.11. Alternatively,, we use the Taylor expansions of ex and
1−x
 
1 1 1 1 1 1
− x = − 2 1− = x

x e −1 x x 2
x 1 + + o(x)
x+ + o(x ) 2
2
1   x   x  1 o(x)
= 1−1+ + o(x) + o + o(x) = + .
x 2 2 2 x
1
This implies that the limit is .
2
 x
1
Example 2.5.14. We know limx→∞ 1 + = e from Example 1.6.17. The next
 xx
1
question is what the difference 1 + − e looks like. As x goes to infinity, does
x
1
the difference approach 0 like ?
x
146 CHAPTER 2. DIFFERENTIATION

1
The question is the same as the behavior of (1 + x) x − e near 0. By the Taylor
expansions of log(1 + x) and ex at 0, we get
1 1
(1 + x) x − e = e x log(1+x) − e
2
 
1
x− x2 +o(x2 )
=e x
−e
− x2 +o(x)


=e e −1
h x   x i
= e − + o(x) + o − + o(x)
2 2
e
= − x + o(x).
2
Translated back into x approaching infinity, we have
 x  
1 e 1
1+ −e=− +o .
x 2x x

Exercise 2.5.27. Find the 5-th order approximations at 0.

1. ex sin x. 3. (1 + x)x . 5. log cos x.

√ 2 sin x x
2. x + 1ex tan x. 4. log . 6. .
x ex −1

Exercise 2.5.28. Use approximations to compute limits.

1 − cos x2 ex − esin x
1. limx→0 . 7. limx→0 .
x3 sin x x3
sin x − tan x cos ax
2. limx→0 . 8. limx→0 log .
x3 cos bx

1 1
 x − tan x
3. limx→0 − . 9. limx→0 .
sin2 x x2 x − sin x
1 x
   

1 1

2 1
4. limx→0 − . 10. limx→∞ x e − − 1 + .
x2 tan2 x x x
 

x 1

2 1
5. limx→1 − . 11. limx→∞ x log x sin .
x − 1 log x x

1 (x − 1) log x
6. limx→0 (cos x + sin x) x(x+1) . 12. limx→1 .
1 + cos πx

Exercise 2.5.29. Use whatever method you prefer to compute limits, p, q > 0.
q
1. limx→0+ xp e−x . 4. limx→1+ (x − 1)p log x.
q
2. limx→+∞ xp e−x . 5. limx→1+ (x − 1)p (log x)q .
q
3. limx→+∞ xp log x. 6. limx→+∞ xp e−x log x.
2.5. HIGH ORDER APPROXIMATION 147

7. limx→+∞ xp log(log x). 9. limx→e+ (x − e)p log(log x).

8. limx→+∞ (log x)p (log(log x))q .

Exercise 2.5.30. Use whatever method you prefer to compute limits, p, q > 0.
tanp x − xp tan x − cot x
1. limx→0+ . 3. limx→ π4 .
sinp x − xp 4x − π
sinp x − tanp x a tan bx − b tan ax
2. limx→0+ . 4. limx→0 .
xq a sin bx − b sin ax

Exercise 2.5.31. Use whatever method you prefer to compute limits.

   
1 1 2 1 1
1. limx→∞ x3 sin − sin . 3. limx→0 − .
x 2 x log(1 + x) x
 
1 1
 
1 2 1
2. limx→0 − . 4. limx→0 √ − .
x x(2 + x) ex − 1 log(x + 1 + x2 ) log(1 + x)

Exercise 2.5.32. Use whatever method you prefer to compute limits.

x
1. limx→1− log x log(1 − x). x(x ) − x
4. limx→0+ .
x2 (log x)2
xx − 1
2. limx→0+ . xlog x − 1
x log x 5. limx→1 .
(log x)2
xx − 1 − x log x xx − x
3. limx→0+ . 6. limx→1 .
x2 (log x)2 log x − x + 1

Exercise 2.5.33. Use whatever method you prefer to compute limits.

(1 + ax)b − (1 + bx)a cos(sin x) − cos x

1. limx→0 . 8. limx→0 .
x2 x4
(1 + ax + cx2 )b − (1 + bx + dx2 )a 9. lim arcsin 2x − 2 arcsin x
2. limx→0 . x→0 .
x 2 x3
ax − xa 1 − cos x2
3. limx→a . 10. limx→0 .
x−a x2 sin x2
(a + x)x − ax 1 − 2 sin x
4. limx→0 . 11. limx→ π6 .
x2 cos 3x
ax − asin x πx
5. limx→0+ , a > 0. 12. limx→1+ log x tan .
x3 2
x −1
log(sin ax) 13. limx→0+ xx .
6. limx→0+ , a, b > 0.
log(sin bx)
14. limx→0 xsin x .
log(cos ax)
7. limx→0 . 15. limx→0+ (− log x)x .
log(cos bx)
148 CHAPTER 2. DIFFERENTIATION

 1
1  1
16. limx→0 e−1 (1 + x) x .
x 2 x
18. limx→0 arccos x .
π
 1
1
cos x x

17. limx→0 (x−1 arcsin x) x2 . 19. limx→0 .

1 + sin x

Exercise 2.5.34. In Example 2.4.2, we applied the Mean Value Theorem to get log(1 + x) =
x
for some 0 < θ < 1.
1 + θx
1. Find explicit formula for θ = θ(x).

2. Compute limx→0 θ by using l’Hospital’s rule.

3. Compute limx→0 θ by using high order approximation.

You may try the same for other functions such as ex − 1 = eθx . What can you say about
limx→0 θ in general?

Exercise 2.5.35. Show that the limits converge but cannot be computed by L’Hospital’s
rule.

1 x − sin x
x2 sin 2. limx→∞ .
1. limx→0 x. x + sin x
sin x

Exercise 2.5.36. Find a, b so that the following hold.

1. x − (a + b cos x) sin x = o(x4 ). 2. x − a sin x − b tan x = o(x4 ).

2.5.4 Implicit High Order Differentiation

Example 2.5.15. In Example 2.2.9, the function y = y(x) implicitly given by the unit circle
x
x2 + y 2 = 1 has derivative y 0 = − . Then
y
x
y+x
y− xy 0 y x2 + y 2 1
y 00 = − =− =− = − 3.
y2 y2 y3 y

You
√ may verify the result by directly computing the second order derivative of y =
± 1 − x2 .

Example 2.5.16. In Example 2.2.10, we computed the derivative of the function y = y(x)
equation2y − 2x2 − sin y + 1 = 0 and then obtained the linear
implicitly given by ther
π π
approximation at P = , . We can certainly continue finding the formula for the
2 2
second order derivative of y(x) and then get the quadratic approximation at P .
2.5. HIGH ORDER APPROXIMATION 149

Alternatively, we may compute the quadratic approximation at P by postulating the

approximation to be
r
π π
y = + a1 ∆x + a2 ∆x2 + o(∆x2 ), ∆x = x − .
2 2

Substituting into the equation, we get

π r 2
2 2
 π
0=2 + a1 ∆x + a2 ∆x + o(∆x ) − 2 + ∆x
2 2
π 
− sin + a1 ∆x + a2 ∆x2 + o(∆x2 ) + 1
2 r
2 π
= 2a1 ∆x + 2a2 ∆x − 4 ∆x − 2∆x2 + o(∆x2 ) − cos(a1 ∆x + a2 ∆x2 + o(∆x2 )) + 1
2
√ 1
= 2a1 ∆x + 2a2 ∆x2 − 2 2π∆x − 2∆x2 + (a1 ∆x + a2 ∆x2 )2 + o(∆x2 )
2

The coefficients of ∆x and ∆x2 on the right must vanish. Therefore

√ 1
2a1 − 2 2π = 0, 2a2 − 2 + a21 = 0.
2
√ 2−π
The solution is a1 = , and the quadratic approximation is
2π, a2 =
2

π √
r
2−π 2 2 π
y(x) = + 2π∆x + ∆x + o(∆x ), ∆x = x − .
2 2 2

Exercise 2.5.37. Compute quadratic approximations of implicitly defined functions.

√ √ √
1. y 2 + 3y 3 + 1 = x. 4. x+ y= a.

2. sin y = x. 5. ex+y = xy.

x2 y 2
3. + 2 = 1. 6. x2 + 2xy − y 2 − 2x = 0.
a2 b

Exercise 2.5.38. Compute quadratic approximations of functions y = y(x) given by the

curves.

1. x = sin2 t, y = cos2 t.

2. x = a(t − sin t), y = a(1 − cos t).

3. x = et cos 2t, y = et sin 2t.

4. x = (1 + cos t) cos t, y = (1 + cos t) sin t.

150 CHAPTER 2. DIFFERENTIATION

2.5.5 Two Theoretical Examples

Theorem 2.5.2 tells us that the existence of high order derivative implies the high
order differentiability. The following example shows that the converse is not true.

Example 2.5.17. The function x3 D(x) satisfies

|x| < δ =  =⇒ |x3 D(x) − 0| ≤ |x|3 ≤ |x|2 .

Therefore the function is second order differentiable, with 0 = 0 + 0x + 0x2 as the

quadratic approximation.
On the other hand, we have (x3 D(x))0 |x=0 = 0 and x3 D(x) is not differentiable
(because not even continuous) away from 0. Therefore the second order derivative
is not defined.

Exercise 2.5.39. Show that for any n, there is a function that is n-th order differentiable
at 0 but has no second order derivative at 0.

Exercise 2.5.40. The lack of high order derivatives for the function in Example 2.5.17 is
due to discontinuity away from 0. Can you find a function with the following properties?

1. f that has first order derivative everywhere on (−1, 1).

2. f has no second order derivative at 0.

3. f is second order differentiable at 0.

The next example deals with the following intuition from everyday life. Suppose
we try to measure a length by more and more refined rulers. If our readings from
meter ruler, centimeter ruler, millimeter ruler, micrometer ruler, etc, are all 0, then
the real length should be 0. Similarly, the Taylor expansion of a function at 0 is the
measurement by “xn -ruler”. The following example shows that, even if the readings
by all the “xn -ruler” are 0, the function does not have to be 0.

Example 2.5.18. The function

( 1
e− |x| , if x 6= 0,
f (x) =
0, if x = 0,

has derivative
1 − |x|

1

 e , if x > 0,
 x2

f 0 (x) = − 1 e− |x|
1
, if x < 0,


 x2
0, if x = 0.
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 151

The derivative at x 6= 0 is computed by the usual chain rule, and the derivative at
0 is computed directly
1 1 y
f 0 (0) = lim e− |x| = lim |y| = 0.
x→0 x y→∞ e

In general, it can be inductively proved that

  
1 −1

 p e x, if x > 0,
 x 



f (n) (x) = (−1)n p − 1 e− x1 , if x < 0,
x




0, if x = 0,


where p(t) is a polynomial depending only on n.

The function has the special property that the derivative of any order vanishes
at 0. Therefore the function is differentiable of any order, and all the high order
approximations are 0

f (x) = 0 + 0x + 0x2 + · · · + 0xn + o(xn ).

However, the function is not 0.

The example can be understood in two ways. The first is that, for some functions,
even more refined ruler is needed in order to measure “beyond all orders”. The
second is that the function above is not “measurable by polynomials”. The functions
that are measurable by polynomials are call analytic, and the function above is not
analytic.

Exercise 2.5.41. Directly show (i.e., without calculating the high order derivatives) that
the function in Example 2.5.18 is differentiable of any order, with 0 as the approximation
of any order.

2.6 Application of High Order Approximation

2.6.1 High Derivative Test
Theorem 2.3.4 gives the first order derivative condition for a (two sided) differen-
tiable function to have local extreme. As the subsequent examples show, the theorem
only provides candidates for the local extrema. To find out whether such candidates
are indeed local extrema, high order approximations are needed.
Let us consider the first non-trivial high order approximation at x0

f (x) = f (x0 ) + c(x − x0 )n + o((x − x0 )n ) = f (x0 ) + c(x − x0 )n (1 + o(1)), c 6= 0.

152 CHAPTER 2. DIFFERENTIATION

When x is close to x0 , we have 1 + o(1) > 0 and therefore f (x) > f (x0 ) when
c(x − x0 )n > 0 and f (x) < f (x0 ) when c(x − x0 )n < 0. Specifically, we have the
following signs of c(x − x0 )n for various cases.

• If n is odd and c > 0, then c(x − x0 )n > 0 for x > x0 and c(x − x0 )n < 0 for
x < x0 .

• If n is odd and c < 0, then c(x − x0 )n < 0 for x > x0 and c(x − x0 )n > 0 for
x < x0 .

• If n is even and c > 0, then c(x − x0 )n > 0 for x 6= x0 .

• If n is even and c < 0, then c(x − x0 )n < 0 for x 6= x0 .

The sign of c(x−x0 )n then further determines whether f (x) < f (x0 ) or f (x) > f (x0 ),
and we get the following result.

Theorem 2.6.1. Suppose f (x) has high order approximation f (x0 ) + c(x − x0 )n at
x0 .

1. If n is odd and c 6= 0, then x0 is not a local extreme.

2. If n is even and c > 0, then x0 is a local minimum.

3. If n is even and c < 0, then x0 is a local maximum.

If f has n-th order derivative at x0 , the condition of the theorem means

f 0 (x0 ) = f 00 (x0 ) = · · · = f (n−1) (x0 ) = 0, f (n) (x0 ) = n!c 6= 0.

The special case n = 1 is Theorem 2.3.4. For the special case n = 2, the theorem
gives the second derivative test: Suppose f 0 (x0 ) = 0 (i.e., the criterion in Theorem
2.3.4 is satisfied), and f has second order derivative at x0 .

1. If f 00 (x0 ) > 0, then x0 is a local minimum.

2. If f 00 (x0 ) < 0, then x0 is a local maximum.

Example 2.6.1. In Example 2.3.13, we found the candidates ±1 for the local extrema
of f (x) = x3 − 3x + 1. The second order derivative f 00 (x) = 6x at the two candidates
are
f 00 (1) = 6 > 0, f 00 (−1) = −6 < 0.
Therefore 1 is a local minimum and −1 is a local maximum.
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 153

Example 2.6.2. Consider the function y = y(x) implicitly defined in Example 2.2.10.
4x
By y 0 (x) = , we find a candidate x = 0 for the local extreme of y(x). Then
2 − cos y
we have  
00 4 d 1
y0.

y (x) = + 4x 
2 − cos y dy 2 − cos y y=y(x)
4
At the candidate x = 0, we already have y 0 (0) = 0. Therefore y 00 (0) = >
2 − cos y(0)
0. This shows that x = 0 is a local minimum of the implicitly defined function.

Example 2.6.3. The function f (x) = x2 − x3 D(x) has no second order derivative
at 0, but still has the quadratic approximation f (x) = x2 + o(x2 ). The quadratic
approximation tells us that 0 is a local minimum of f (x).

sin x 1
Example 2.6.4. Let f (x) = 3
for x 6= 0 and f (0) = . Then for x 6= 0 close
6x − x 6
to 0, we have

x3 x5 x2 x4
   
1 6 1 5
f (x) = x− + + o(x ) = 1− + + o(x )
6x − x3 6 120 6 − x2 6 120
x2 x 4 x2 x4 x4
  
1 5 5 1
= 1+ + + o(x ) 1− + + o(x ) = + + o(x5 ).
6 6 36 6 120 6 120

1
We note that by f (0) = , the 4-th order approximation also holds for x = 0. Then
6
by Theorem 2.6.1, we find that x = 0 is a local minimum.
Alternatively, we may directly use the idea leading to Theorem 2.6.1. For x 6= 0
close to 0, we have

x3 x5 x4 (1 + o(x))
 
1 6 1
f (x) = x − + + o(x ) = + .
6x − x3 6 120 6 (6 − x2 ) · 120

x4 (1 + o(x)) 1
For small x 6= 0, we have 2
> 0, which further implies f (x) > = f (0).
(6 − x ) · 120 6
Therefore 0 is a (strict) local minimum.

Exercise 2.6.1. Find the local extrema by using quadratic approximations.

1. x3 − 3x + 1 on R. 3. x log x on (0, +∞).

2. xe−x on R. 4. (x2 + 1)ex on R.

Exercise 2.6.2. Find the local extrema for the function y = y(x) implicitly given by x3 +
y 3 = 6xy.
154 CHAPTER 2. DIFFERENTIATION

Exercise 2.6.3. For p > 1, determine whether 0 is a local extreme for the function

x2 + |x|p sin x sin 1 , if x =
6 0,
x
0, if x = 0.

Exercise 2.6.4. Determine whether 0 is a local extreme.

 
1. x3 + x4 . 1 2 x
4. 1 − x + x e .
2!
2. sin x − x cos x.  
1 2 1 3 x
3. tan x − sin x. 5. 1 − x + x − x e .
2! 3!

Exercise 2.6.5. Let f (0) = 1 and let f (x) be given by the following for x 6= 0. Determine
whether 0 is a local extreme.
sin x sin x sin x
1. . 2. . 3. .
x + ax2 x + bx3 x + ax2 + bx3

2.6.2 Convex Function

A function f is convex on an interval if for any x, y in the interval, the straight line
Lx,y connecting points (x, f (x)) and (y, f (y)) lies above the part of the graph of f
over [x, y]. This means
f (y) − f (x)
Lx,y (z) = f (x) + (z − x) ≥ f (z) for any x ≤ z ≤ y.
y−x
The function is concave if Lx,y lies below the graph of f , which means changing
≥ f (z) above to ≤ f (z). A function f is convex if and only if −f is concave.

Lx,y (z)

f (z)

x z y

Figure 2.6.1: Convex function.

By the geometric intuition illustrated in Figure 2.6.2, the following are equivalent
convexity conditions, for any x ≤ z ≤ y,
1. Lx,y (z) ≥ f (z).

2. slope(Lx,z ) ≤ slope(Lx,y ).
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 155

3. slope(Lz,y ) ≥ slope(Lx,y ).

4. slope(Lx,z ) ≤ slope(Lz,y ).

Algebraically, it is not difficult to verify the equivalence.

Lz,y
Lx,z Lx,y

x z y

Figure 2.6.2: Convexity by comparing slopes.

If a convex function f is differentiable, then we may take y → x+ in the inequality

slope(Lx,z ) ≤ slope(Lx,y ), and get f 0 (x) ≤ slope(Lx,y ). Similarly, we may take
z → y − in the inequality slope(Lz,y ) ≥ slope(Lx,y ), and get f 0 (y) ≥ slope(Lx,y ).
Combining the two inequalities, we get

x < y =⇒ f 0 (x) ≤ f 0 (y).

It turns out that the converse is also true.

Theorem 2.6.2. A differentiable function f on an interval is convex if and only if

f 0 is increasing. If f has second order derivative, then this is equivalent to f 00 ≥ 0.

Similarly, a function f is concave if and only if f 0 is decreasing, and is also

equivalent to f 00 ≤ 0 in case f 00 exists.
The converse of Theorem 2.6.2 is explained by Figure 2.6.2. If f 0 is increasing,
then by the Mean Value Theorem (for the two equalities), we have

slope(Lx,z ) = f 0 (c) ≤ f 0 (d) = slope(Lz,y ).

A major application of the convexity is another interpretation of the convexity.

In the setup above, a number z between x and y means z = λx + (1 − λ)y for some
λ ∈ [0, 1]. Then the linear function Lx,y (z) = az + b preserves the linear relation

Lx,y (z) = a(λx + (1 − λ)y) + b

= λ(ax + b) + (1 − λ)(ay + b)
= λLx,y (x) + (1 − λ)Lx,y (y)
= λf (x) + (1 − λ)f (y).
156 CHAPTER 2. DIFFERENTIATION

Lz,y
Lx,z

x c z d y

Figure 2.6.3: Increasing f 0 (x) implies convexity.

Therefore the convexity means

f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y) for any 0 ≤ λ ≤ 1.

The following generalizes the inequality.

Theorem 2.6.3 (Jensen’s inequality). If f is convex, and

λ1 + λ2 + · · · + λn = 1, 0 ≤ λ1 , λ2 , . . . , λn ≤ 1,

then

f (λ1 x1 + λ2 x2 + · · · + λn xn ) ≤ λ1 f (x1 ) + λ2 f (x2 ) + · · · + λn f (xn ).

By reversing the direction of inequality, we also get Jensen’s inequality for con-
cave functions.
In all the discussions about convexity, we may also consider the strict inequalities.
So we have a concept of strict convexity, and a differentiable function is strictly
convex on an interval if and only if its derivative is strictly increasing. Jensen’s
inequality can also be extended to the strict case.

Example 2.6.5. By (xp )00 = p(p − 1)xp−2 , we know xp is convex on (0, +∞) when
p ≥ 1 or p < 0, and is concave when 0 < p ≤ 1.
For p ≥ 1, Jensen’s inequality means that

(λ1 x1 + λ2 x2 + · · · + λn xn )p ≤ λ1 xp1 + λ2 xp2 + · · · + λn xpn .

1
In particular, if all λi = , then we get
n
p
xp1 + xp2 + · · · + xpn

x1 + x2 + · · · + xn
≤ , for p ≥ 1, xi ≥ 0.
n n

This means that the p-th power of the average is smaller than the average of the
p-th power.
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 157

We note that (xp )00 > 0 for p > 1. Therefore all the inequalities are strict,
provided some xi > 0 and 0 < λi < 1.
p
By replacing p with and replacing xi with xqi , we get
q
 1q  p1
xq1 + xq2 + · · · + xqn xp1 + xp2 + · · · + xpn
 
≤ , for p > q > 0.
n n

x2
x
1
x2

1
x− 2
x−1
x−2
1

Figure 2.6.4: xp for various p.

1
Example 2.6.6. By (log x)00 = − 2 < 0, the logarithmic function is concave. Then
x
Jensen’s inequality tells us that

log(λ1 x1 + λ2 x2 + · · · + λn xn ) ≥ λ1 log x1 + λ2 log x2 + · · · + λn log xn .

This is the same as

λ1 x1 + λ2 x2 + · · · + λn xn ≥ xλ1 1 xλ2 2 · · · xλnn .

1
For the special case λi = , we get
n
x1 + x2 + · · · + xn √
≥ n x 1 x2 · · · xn .
n
In other words, the arithmetic mean is bigger than the geometric mean.

Exercise 2.6.6. Study the convexity.

158 CHAPTER 2. DIFFERENTIATION

1. x3 + ax + b. 4. (x2 + 1)ex . 1 10. xx .

7. .
x2 +1
5
2. x + x 3 . 5. e−x .
2
x2 − 1 11. x + 2 sin x.
8. .
x2 + 1
√
3. x + xp . 6. x2 + 1. 9. log(x2 + 1). 12. x sin(log x).

Exercise 2.6.7. Find the condition on A and B so that the function

(
Axp , if x ≥ 0,
f (x) = q
B(−x) , if x < 0,

is convex. Is the function necessarily differentiable at 0?

1 1
Exercise 2.6.8. Let p, q > 0 satisfy + = 1. Use the concavity of log x to prove Young’s
p q
inequality
1 p 1 q
x + y ≥ xy.
p q

Exercise 2.6.9. For the case λ1 6= 1, Find suitable µ2 , . . . , µn satisfying

λ1 x1 + λ2 x2 + · · · + λn xn = λ1 x1 + (1 − λ1 )(µ2 x2 + · · · + µn xn ).

Then prove Jensen’s inequality by induction.

Exercise 2.6.10. Use the concavity of log x to prove that, for xi > 0, we have

x1 log x1 + x2 log x2 + · · · + xn log xn x2 + x22 + · · · + x2n

≤ log 1 ≤ log(x1 + x2 + · · · + xn ).
x1 + x2 + · · · + xn x1 + x2 + · · · + xn

1
Exercise 2.6.11. Use Exercise 2.6.10 to show that f (p) = log(xp1 + xp2 + · · · + xpn ) p is
decreasing. Then explain
1 1
(xq1 + xq2 + · · · + xqn ) q ≥ (xp1 + xp2 + · · · + xpn ) p , for p > q > 0.

Note that the similar inequality in Example 2.6.5 is in reverse direction.

Exercise 2.6.12. Verify the convexity of x log x and then use the property to prove the
inequality (x + y)x+y ≤ (2x)x (2y)y . Can you extend the inequality to more variables?

2.6.3 Sketch of Graph

We have learned the increasing and decreasing properties, and the convex and con-
cave properties. We may also pay attention to the symmetry properties such as even
or odd function, and the periodic property. Moreover, we should pay attention to
the following special points.
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 159

1. Intercepts, where the graph of function intersects the axes.

2. Disruptions, where the functions are not continuous, or not differentiable.

3. Local extrema, which is often (but not restricted to) the places where the
function changes between increasing and decreasing.

4. Points of inflection, which is the place where the function changes between
convex and concave.

5. Infinity, including the finite places where the function tends to infinity, and
the behavior of the function at the infinity.

One infinity behavior is the asymptotes of a function. If a linear function a + bx

satisfies
lim (f (x) − a − bx) = 0,
x→+∞

then the linear function is an asymptote at +∞. If b = 0, then the line is a horizontal
asymptote. We also have similar asymptote at −∞ (perhaps with different a and
b). Moreover, if limx→x0 f (x) = ∞ at a finite x0 , then the line x = x0 is a vertical
asymptote.
In subsequent examples, we sketch the graphs of functions and try to indicate
the characteristics listed above as much as possible.

Example 2.6.7. In Example 2.3.4, we determined the monotone property and the
local extrema of f (x) = x3 − 3x + 1. The second order derivative f 00 (x) = 6x also
tells us that f (x) is concave on (−∞, 0] and convex on [0, +∞), which makes 0 into
a point of inflection. Moreover, the function has no asymptotes. The function is
also symmetric with respect to the point (0, 1) (f (x) − 1 is an odd function). Based
on these information, we may sketch the graph.

x (−∞, −1) −1 (−1, 0) 0 (0, 1) 1 (1, +∞)

3
f = x − 3x + 1 −∞ ← 3 0 −1 → +∞
+ 0 − 0 +
f 0 = 3(x + 1)(x − 1)
% max & min %
+ 0 −
f 00 = 6x
infl

Example 2.6.8. In Example 2.3.6, we determined the monotone property of f (x) =

√ 2(5x − 1)
x2 (x+1). The second order derivative f 00 (x) =
3
√
3
implies that the function
9 4
   x 
1 1 1
is concave on (−∞, 0) and on 0, , convex on , +∞ , with as a point of
5 5 5
inflection. Moreover, we have limx→−∞ f (x) = −∞, limx→+∞ f (x) = +∞. The
160 CHAPTER 2. DIFFERENTIATION

max
3

infl

−3 −2 −1 0 1 2 3
1
min

Figure 2.6.5: Graph of x3 − 3x + 1.

f (x)
function has no asymptote and no symmetry. We also note that limx→0 =
x
x+1
limx→0 √ = ∞. Therefore the tangent of f (x) at 0 is vertical.
3
x
x (−∞, − 25 ) − 52 (− 25 , 0) 0 (0, 15 ) 1
5
( 15 , +∞)
f −∞ ← 0.1518 0 0.1073 → +∞
+ 0 − no +
f0
% max & min %
− no − 0 +
f 00
infl

max infl

−0.4 min 0.2

√
3
Figure 2.6.6: Graph of x2 (x + 1).

Example 2.6.9. In Example 2.3.15, we determined the monotone property of f (x) =

xe−x . From f 00 (x) = (x − 2)e−x , we also know the function is concave on (−∞, 2]
and convex on [2, +∞), with f (2) = e−2 as a point of inflection. Moreover, we
have limx→−∞ f (x) = −∞, limx→+∞ f (x) = 0, so that the x-axis is a horizontal
asymptote. The function has no symmetry.
2.6. APPLICATION OF HIGH ORDER APPROXIMATION 161

max e−1 2e−2infl

1 2

Figure 2.6.7: Graph of xe−x .

Example 2.6.10. In Example 2.3.7, we determined the monotone property and the
x3
local extrema of f (x) = 2 . The function is not defined at ±1, and has limits
x −1
lim f (x) = ±∞, lim f (x) = ∓∞.
x→1± x→−1±

These give vertical asymptotes at ±1. Moreover, we have

lim (f (x) − x) = 0.
x→∞

Therefore y = x is a slant asymptote at ∞.

00 2x(x2 + 3)
The second order derivative f (x) = shows that the function is convex
(x2 − 1)3
on (−1, 0), (1, +∞), and is concave at the other two intervals. Therefore 0 is a point
of inflection.
We also know the function is odd. So the graph is symmetric with respect to
the origin.

Exercise 2.6.13. Use the graph of x3 − 3x + 1 in Example 2.6.7 to sketch the graphs.

1. −x3 + 3x2 − 1. 3. |x3 − 3x + 1|. 5. x3 − bx, b > 0.

2. x3 − 3x + 2. 4. x3 − 3x. 6. ax3 − bx + c, ab > 0.

Exercise 2.6.14. Use the graph of xe−x in Example 2.6.9 to sketch the graphs.

1. xex . 3. (x + 1)ex . 5. xax .

2. xe2x . 4. (ax + b)ex . 6. |x|e−|x−1| .

Exercise 2.6.15. Sketch the graphs.

162 CHAPTER 2. DIFFERENTIATION

min

√
− 3 infl
√
3

max

x3
Figure 2.6.8: Graph of .
x2 − 1

1. x − sin x. 3. x − cos x. 5. x + cos x.

1
2. |x − sin x|. 4. x + sin x. 6. + sin x cos x.
2

Exercise 2.6.16. Sketch the graphs.

1 1 1 1
1. . 2. . 3. . 4. .
x2 x2 + 1 x2 − 1 x2 + bx + c

Exercise 2.6.17. Sketch the graphs.

√ √
1. (x2 + 1)p . 2. (x2 − 1)p . 3. 1 − x2 . 4. ax2 + bx + c.

ax + b 1
Exercise 2.6.18. Sketch the graph of by using the graph of .
cx + d x

Exercise 2.6.19. Sketch the graphs on the natural domains of definitions.

√ 1
1. x4 − 2x2 + 1. 5. x x − 1. 10. e x .
√ 11. x log x.
1 6. x2 + 1 − x.
2. x + .
x
1 2 12. x − log(1 + x).
x 7. x 3 (1 − x) 3 .
3. 2 . 13. log(1 − log x).
x +1
8. x2 e−x .
14. log(1 + x4 ).
x3 1
4. . 9. . 15. e−x sin x.
(x − 1)2 1 + ex
2.7. NUMERICAL APPLICATION 163

16. x tan x. 1 18. 2 sin x + sin 2x.

17. .
1 + sin2 x
19. 2x − 4 sin x + sin 2x.

2.7 Numerical Application

The linear approximation can be used to find approximate values of functions.
√
Example 2.7.1. The linear approximation of x at x = 4 is
√ √ 1
L(x) = 4 + ( x)0 |x=4 (x − 4) = 2 + (x − 4).
4
Therefore the value of the square root near 4 can be approximately computed
√ 1 √ 1
3.96 ≈ 2 + (−0.04) = 1.99, 4.05 ≈ 2 + (0.05) = 2.0125.
4 4

Example 2.7.2. Assume some metal balls of radius r = 10 are selected to make a
ball bearing. If the radius is allowed to have 1% relative error, what is the maximal
relative error of the weight?
The weight of the ball is
4
W = ρπr3 .
3
where ρ is the density. The error ∆W of the weight caused by the error ∆r of the
radius is
dW
∆W ≈ ∆r = 4ρπr2 ∆r.
dr
Therefore the relative error is
∆W ∆r
≈3 .
W r
 
 ∆r 
Given the relative error of the radius is no more than 1%, we have   ≤ 1%, so
  r
 ∆W 
that the relative error of the weight is   ≤ 3%.
W 

Example 2.7.3. In Example 2.2.11, we computed the derivatives of the functions

y = y(x) and z = z(x) given by the equations x2 + y 2 + z 2 = 2 and x + y + z = 0,
which is really a circle in R3 . The point P = (1, 0, −1) lies on the circle, where we
have

1 − (−1) 1−0
y(1) = 0, z(1) = −1, y 0 (1) = = −2, z 0 (1) = = 1.
(−1) − 0 0 − (−1)
164 CHAPTER 2. DIFFERENTIATION

Therefore
y(1.01) ≈ 0 − 2 · 0.01 = −0.02, z(1.01) ≈ −1 + 1 · 0.01 = −0.99,
y(0.98) ≈ 0 − 2 · (−0.02) = 0.04, z(0.98) ≈ −1 + 1 · (−0.02) = −1.02.
In other words, the points (1.01, −0.01, −0.99) and (0.98, 0.04, −1.02) are near (1, 0, −1)
and almost on the circle.

Exercise 2.7.1. For a > 0 and small x, derive

√
n n x
a +x≈a+ .
nan−1
Then find the approximate values.
√ √ √ √
1. 4 15. 2. 46. 3. 5
39. 4. 7
127.
r
L
Exercise 2.7.2. The period of a pendulum is T = 2π , where L the length of the
g
pendulum and g is the gravitational constant. If the length of the pendulum is increased
by 0.4%, what is the change in the period?

2.7.1 Remainder Formula

We may get more accurate values by using high order approximations. On the other
hand, we have more confidence on the estimated values if we also know what the
error is. The following result gives a formula for the error.

Theorem 2.7.1 (Lagrange Form of the Remainder). If f (x) has (n + 1)-st order
derivative on (a, x), then the remainder of the n-th order Taylor expansion of f (x)
at a is
f (n+1) (c)
Rn (x) = (x − a)n+1 for some c ∈ (a, x).
(n + 1)!

We only illustrate the argument for the case n = 2. We know that the remainder
satisfies R(a) = R0 (a) = R00 (a) = 0. Therefore by Cauchy’s Means Value Theorem
(Theorem 2.4.5), we have
R2 (x) R2 (x) − R2 (a) R20 (c1 )
= = (a < c1 < x)
(x − a)3 (x − a)3 − (a − a)3 3(c1 − a)2
R20 (c1 ) − R20 (a) R200 (c2 )
= = (a < c2 < c1 )
3[(c1 − a)2 − (a − a)2 ] 3 · 2(c2 − a)
00 00 000
R (c2 ) − R2 (a) R (c3 ) f 000 (c3 )
= 2 = 2 = . (a < c3 < c2 )
3 · 2(c2 − a) 3·2·1 3!
In the last step, we use R2000 = f 000 because the f − R2 is a quadratic function and
has vanishing third order derivative.
2.7. NUMERICAL APPLICATION 165

A slight modification of the proof above actually gives a proof that the Taylor
expansion is high order approximation (Theorem 2.5.2).

Example 2.7.4. The error for the linear approximation in Example 2.7.1 can be esti-
mated by
1
3
4c 1
R1 (x) = − 2 ∆x2 = 3 ∆x3 .
2! 8c 2
For both approximate values, we have
1
|R1 | ≤ 3 0.052 = 0.0000390625 < 4 × 10−5 .
8·4 2

If we use the quadratic approximation at 4

√ 1 1
x ≈ 2 + (x − 4) − (x − 4)2 ,
4 64
then we get better estimated values
√ 1 1
3.96 ≈ 2 + (−0.04) − (−0.04)2 = 1.989975,
4 64
√ 1 1
4.05 ≈ 2 + (0.05) − (0.05)2 = 2.0124609375.
4 64
The error can be estimated by
3
5
1
R2 (x) = 8c 2 ∆x3 = 3
5 ∆x .
3! 16c 2
For both computations, we have
1
|R2 | ≤ 5 0.053 = 0.00000025 = 2.5 × 10−7 .
16 · 4 2
√ √
The true values are 3.96 = 1.989974874213 · · · and 4.05 = 2.01246117975 · · · .

Example 2.7.5. The Taylor expansion of ex tells us

1 1 1
e = e1 = 1 + + + ··· + + Rn (1).
1! 2! n!
By
ec e
|Rn (1)| = 1n+1 ≤ , 0 < c < 1,
(n + 1)! (n + 1)!
we know
|R13 (1)| ≤ 0.000000000035 = 3.5 × 10−11 .
166 CHAPTER 2. DIFFERENTIATION

On the other hand,

1 1 1
1+ + + ··· + = 2.718281828447 · · · ,
1! 2! 13!
Therefore e = 2.7182818284 · · · .

Exercise 2.7.3. Find approximate values by using Taylor expansions and estimate the er-
rors.

1. sin 1, 3rd order approximation. 3. e−1 , 5th order approximation.

2. log 2, 5th order approximation. 4. arctan 2, 3rd order approximation.

Exercise 2.7.4. Find approximate values accurate up to the 10-th digit.

√
1. sin 1. 2. 4.05. 3. e−1 . 4. tan 46◦ .

Exercise 2.7.5. Find the approximate value of tan 1 accurate up to the 10-th digit by using
the Taylor expansions of sin x and cos x.

Exercise 2.7.6. If we use the Taylor expansion to calculate e accurate up to the 100-th
digit, what is the order of the Taylor expansion we should use?

2.7.2 Newton’s Method

The linear approximation can also be used to find approximate solutions of equa-
tions. To solve f (x) = 0, we start with a rough estimation x0 and consider the linear
approximation at x0
L0 (x) = f (x0 ) + f 0 (x0 )(x − x0 ).
We expect the solution of the linear equation L0 (x) = 0 to be very close to the
solution of f (x) = 0. The solution of the linear equation is easy to find
f (x0 )
x1 = x0 − .
f 0 (x0 )
Although x1 is not the exact solution of f (x) = 0, chances are it is an improvement
of the initial estimation x0 .
To get an even better approximate solution, we repeat the process and consider
the linear approximation at x1

L1 (x) = f (x1 ) + f 0 (x1 )(x − x1 )

The solution of the linear equation L1 (x) = 0

f (x1 )
x2 = x1 −
f 0 (x1 )
2.7. NUMERICAL APPLICATION 167

solution

x0 x1 x2 x3
L0 L1 L2 L3

Figure 2.7.1: Newton’s method.

is an even better estimation than x1 . The idea leads to an inductively constructed

sequence
f (xn )
xn+1 = xn − .
f 0 (xn )

We expect the sequence to rapidly converge to the exact solution of f (x) = 0.

The scheme above for finding the approximate solution is called the Newton’s
method. The method may fail for various reasons. However, if the function is rea-
sonably good and the initial estimation x0 is sufficiently close to the exact solution,
then the method indeed produces a sequence that rapidly converges to the exact
solution. In fact, the error between xn and the exact solution c satisfies

|xn+1 − c| ≤ M |xn − c|2

for some constant M that depends only on the function.

Example 2.7.6. By Example 1.7.5, the equation x3 −3x+1 = 0 should have a solution
on (0.3, 0.4). By Example 1.7.6, the equation should also have a second solution > 1
and a third solution < 0. More precisely, by f (−2) = −1, f (−1) = 3, f (1) = −1,
f (2) = 3, the second solution is on (−2, −1) and the third solution is on (1, 2).
Taking −2, 0.3, 2 as initial estimations, we apply Newton’s method and compute
the sequence
x3n − 3xn + 1 2 2xn − 1
xn+1 = xn − = x n + .
3(x2n − 1) 3 3(x2n − 1)

We find the three solutions

−1.87938524157182 · · · , 0.347296355333861 · · · , 1.53208888623796 · · · .

168 CHAPTER 2. DIFFERENTIATION

n x0 = −2 x0 = 0.3 x0 = 2
1 -1.88888888888889 0.346520146520147 1.66666666666667
2 -1.87945156695157 0.347296117887934 1.54861111111111
3 -1.87938524483667 0.347296355333838 1.53239016186538
4 -1.87938524157182 0.347296355333861 1.53208898939722
5 -1.87938524157182 0.347296355333861 1.53208888623797
6 1.53208888623796
7 1.53208888623796

Note that the initial estimation cannot be 1 or −1, because the derivative van-
ishes at the points. Moreover, if we start from 0.88, 0.89, 0.90, we get very different
sequences that respectively converge to the three sequences. We see that Newton’s
method can be very sensitive to the initial estimation, especially when the estimation
is close to where the derivative vanishes.

n x0 = 0.88 x0 = 0.89 x0 = 0.90

1 -0.5362647754137 -0.657267917268 -0.80350877192983
2 0.6122033746535 0.920119732577 1.91655789116111
3 0.2884916149262 -1.212642459862 1.63097998546252
4 0.3461342508923 -3.235117846394 1.54150263235725
5 0.3472958236620 -2.419800571908 1.53218794505509
6 0.3472963553337 -2.014098301161 1.53208889739446
7 0.3472963553339 -1.891076746708 1.53208888623796
8 -1.879485375060
9 -1.879385249013
10 -1.879385241572

Example 2.7.7. We solve the equation sin x + x cos x = 0 by starting with the estima-
tion x0 = 1. After five steps, we find the exact solution should be 0.325639452727856 · · · .

n xn
0 1.000000000000000
1 0.471924667505487
2 0.330968826345873
3 0.325645312076542
4 0.325639452734876
5 0.325639452727856
6 0.325639452727856

Exercise 2.7.7. Applying Newton’s method to solve x3 − x − 1 = 0 with the initial estima-
tions 1, 0.6 and 0.57. What lesson can you draw from the conclusion?

Exercise 2.7.8. Use Newton’s method to find the unique positive root of f (x) = ex − x − 2.
2.7. NUMERICAL APPLICATION 169

Exercise 2.7.9. Use Newton’s method to find all the solutions of x2 − cos x = 0.
√
Exercise 2.7.10. Use Newton’s method to find the approximate values of 4.05 and e−1
accurate up to the 10-th digit.

Exercise 2.7.11. Use Newton’s method to find all solutions accurate up to the 6-th digit.

1. x4 = x + 3. 2. ex = 3 − 2x. 3. cos2 x = x. 4. x + tan x = 1.

Note that one may rewrite the equation into another equivalent form and derive a
simpler recursive relation.

Exercise 2.7.12. The ancient Babylonians used the recursive relation

 
1 a
xn+1 = xn +
2 xn
√
to get more an more accurate approximate values of a. Explain the scheme by Newton’s
method.

Exercise 2.7.13. What approximate values does the recursive relation xn+1 = 2xn − ax2n
give you? Explain by Newton’s method.

Exercise 2.7.14. Explain why Newton’s method does not work if we try to solve x3 −3x+1 =
0 by starting at the estimation 1.

Exercise 2.7.15. Newton’s method fails to solve the following equations by starting at any
x0 6= 0. Why?
√ p
1. 3 x = 0. 2. sign(x) |x| = 0.
170 CHAPTER 2. DIFFERENTIATION
Chapter 3

Integration

3.1 Area and Definite Integral

3.1.1 Area below Non-negative Function
Let f (x) be a non-negative function on [a, b]. We wish to find the area of the region
G[a,b] (f ) = {(x, y) : a ≤ x ≤ b, 0 ≤ y ≤ f (x)}
between the graph of the function and the x-axis.

G[a,b] (f )

a b

Figure 3.1.1: The region between the function and the x-axis.

Our strategy is the following. For any x ∈ [a, b], let A(x) be the area of G[a,x] (f ),
which is part of the region over [a, x]. We will find how the function A(x) changes
and recover A(x) from its change. The area we wish to find is then the value A(b).
The subsequent argument assumes that f (x) is continuous.
Consider an interval [x, x + h] ⊂ [a, b], which implicitly assumes h > 0. Then the
change A(x+h)−A(x) is the area of G[x,x+h] (f ). Note that G[x,x+h] (f ) is sandwiched
between two rectangles
[x, x + h] × [0, m] ⊂ G[x,x+h] (f ) ⊂ [x, x + h] × [0, M ],

171
172 CHAPTER 3. INTEGRATION

where
m = min f, M = max f.
[x,x+h] [x,x+h]

Since bigger region should have bigger area, we have

mh ≤ A(x + h) − A(x) ≤ M h. (3.1.1)
By h > 0, this is the same as
A(x + h) − A(x)
m≤ ≤ M. (3.1.2)
h
Since the (right) continuity of f (x) implies
lim m = f (x) = lim+ M,
h→0+ h→0

by the sandwich rule, we further get the right derivative

A(x + h) − A(x)
A0+ (x) = lim+ = f (x). (3.1.3)
h→0 h

m
G[x,x+h] (f )

a x x+h
b

Figure 3.1.2: Estimate the change of area.

The argument above assumes h > 0. For h < 0, we consider [x + h, x] ⊂ [a, b].
Then the change A(x + h) − A(x) is the negative of the area of G[x+h,x] (f ), and the
interval [x + h, x] has length −h. By the same reason as before, we get
m(−h) ≤ −(A(x + h) − A(x)) ≤ M (−h).
By −h > 0, we still get the inequality (3.1.2), and further application of the sandwich
rule gives the left derivative
A(x + h) − A(x)
A0− (x) = lim− = f (x).
h→0 h
We conclude that, for non-negative and continuous f (x), we have
A0 (x) = f (x). (3.1.4)
3.1. AREA AND DEFINITE INTEGRAL 173

Example 3.1.1. To find the area of the region below f (x) = c over [a, b], by (3.1.4),
we have A0 (x) = c = (cx)0 . Then by Theorem 2.4.3, we get A(x) = cx + C. Further
by A(a) = 0, we get C = −ca and A(x) = c(x − a). Therefore the area of the region
is A(b) = c(b − a).
The region G[a,b] (c) is actually a rectangle of base b − a and height c. The
computation of the area is consistent with the common sense.

Example 3.1.2. To find the

0 area of the region below f (x) = x over [0, a], we start
1 1
with A0 (x) = x = x2 . This implies A(x) = x2 + C. By A(0) = 0, we further
2 2
1 2 1
get C = 0 and A(x) = x . Therefore the region has area A(a) = a2 .
2 2
The region is actually a triangle, more precisely half of the square of side length
a. The computation of the area is consistent with the common sense.

The pattern we see from the examples above is that, to find the area below a non-
negative function and over an interval [a, b], we first find a function F (x) satisfying
f (x) = F 0 (x). Then by Theorem 2.4.3, A0 (x) = F 0 (x) implies A(x) = F (x) + C.
Further, by A(a) = 0, we get C = −F (a). Therefore A(x) = F (x) − F (a), and the
area we wish to find is

Area(G[a,b] (f )) = F (b) − F (a).

This is the Newton-Leibniz formula. The function F is naturally called an an-

tiderivative of f .

2
Example
 0 3.1.3. To find the area of the region below x and over [0, a], we use
1 3 1 1 1
x = x2 . The area is a3 − 03 = a3 .
3 3 3 3  0
1 p
More generally, for any p 6= −1 and 0 < a < b, by x = xp , the area of
p+1
the region below xp and over [a, b] is
1
(bp+1 − ap+1 ).
p+1
√
For example, the area of the region below x and over [1, 2] is
x=2
2 3  2 3 3 2 √
x2  = (2 2 − 1 2 ) = (2 2 − 1).
3 x=1 3 3

Exercise 3.1.1. Find the area of the region below the function over the given interval.
174 CHAPTER 3. INTEGRATION

√
x

1 2

2 √
Figure 3.1.3: Area below parabola is (2 2 − 1).
3
√
1. xp on [0, 1], p > 0. 3. ex on [0, a]. 5. 1 + x on [1, 2].
1
2. sin x on [0, π2 ]. 4. on [1, a]. 6. log x on [1, a].
x

Exercise 3.1.2. Find the area of the region bounded by 1 − x2 and the x-axis.

3.1.2 Definite Integral of Continuous Function

What do we get if we apply the Newton-Leibniz formula to general continuous
functions, which might become negative somewhere? The answer is the signed area.
This means that we count the region between the non-negative part of f and the
x-axis as having positive area and count the region between the non-positive part
of f and the x-axis as having negative area. See Figure 3.1.4.

−
f

Figure 3.1.4: Computation by the Newton-Leibniz formula gives signed area.

To justify our claim, let A(x) be the signed area for f (x) over [a, x]. What we
are really concerned with is the change of A(x) where f is negative. So we consider
[x, x + h] ⊂ [a, b], with h > 0 and f < 0 on [x, x + h]. The change A(x + h) − A(x)
is the negative of the positive, “unsigned” area of the region

G[x,x+h] (f ) = {(t, y) : x ≤ t ≤ x + h, 0 ≥ y ≥ f (t)}

between f and the x-axis along the interval [x, x + h]. We have the similar inclusion
3.1. AREA AND DEFINITE INTEGRAL 175

(see Figure 3.1.5, and note that [M, 0] ⊂ [m, 0] because 0 ≥ M ≥ m)

[x, x + h] × [M, 0] ⊂ G[x,x+h] (f ) ⊂ [x, x + h] × [m, 0], m = min f, M = max f.

[x,x+h] [x,x+h]

The heights of the rectangles are respectively −M and −m, and we get (beware of
the signs)
(−M )h ≤ −(A(x + h) − A(x)) ≤ (−m)h.
Again we get the inequality (3.1.2) and subsequently the limit (3.1.3).
The discussion for the case h < 0 is similar, and we conclude that A0 (x) = f (x).

a xx+h
G(f ) b
M
m

Figure 3.1.5: Estimate the change of negative area.

Z b
The signed area is the definite integral of f (x) and is denoted f (x)dx. The
a
function f (x) is called the integrand and the ends a, b of the interval are called
the upper limit and lower limit. The argument above and the explanation before
Example 3.1.3 show that the definite integral can be computed by the Newton-
Leibniz formula
Z b
f (x)dx = F (b) − F (a), where F 0 (x) = f (x).
a

 0
0 1 2
Example 3.1.4. By (x) = 1 and x = x, we get
2
Z b Z b
1
dx = b − a, xdx = (b2 − a2 ).
a a 2
In general, for any integer n 6= −1, we have
Z b
1
xn dx = (bn+1 − an+1 ).
a n+1
However, for n < 0, a and b need to have the same sign. The reason is that we
derived the Newton-Leibniz formula under the assumption that the integrand is
176 CHAPTER 3. INTEGRATION

continuous. If a and b have different sign, then 0 ∈ [a, b], and xn is not continuous
on [a, b] for n < 0.
On the other hand, for any p, xp is defined for x > 0. Then for b ≥ a > 0, we
have Z b
1
xp dx = (bp+1 − ap+1 ).
a p + 1
We note that xp is also defined at 0 for p ≥ 0, and the formula above holds for
p ≥ 0 and b ≥ a ≥ 0. The reason is that xp is right continuous at 0, and we derived
Newton-Leibniz formula by one-sided derivatives.

1
Example 3.1.5. By (ex )0 = ex and (log x)0 = , we get
x
Z b Z b
x b a dx b
e dx = e − e , = log b − log a = log .
a a x a
Note that the second integral requires b ≥ a > 0.

Example 3.1.6. By (sin x)0 = cos x and (cos x)0 = − sin x, we get
Z b Z b
cos xdx = sin b − sin a, sin xdx = cos a − cos b.
a a

For example, we have

Z 0
sin xdx = cos(−π) − cos 0 = −2.
−π

Example 3.1.7. From the derivatives of arcsin x and arctan x, we get

Z b Z b
dx dx
√ = arcsin b − arcsin a, 2
= arctan b − arctan a.
a 1 − x2 a 1+x

Of course, we need |a|, |b| < 1 in the first equality because the integrand is defined
1
only on the open interval (−1, 1). In particular, the area of the region below
1 + x2
and over [0, 1] is Z 1
dx π
2
= arctan 1 − arctan 0 = .
0 1+x 4
We also note that
Z b
dx π  π
lim = lim arctan b − lim arctan a = − − = π.
a→−∞ a 1 + x2 b→+∞ a→−∞ 2 2
b→+∞

1
So the area of the unbounded region between and the x-axis is π.
1 + x2

Exercise 3.1.3. Use the area meaning of definite integral to directly find the value.
3.1. AREA AND DEFINITE INTEGRAL 177
Z 1 p Z 3 Z b
1. 1 − x2 dx. 2. (x − 2)dx. 3. |x − 1|dx.
−1 0 a

Exercise 3.1.4. Compute and compare

Z 2 Z 2 Z 2 Z 2 Z 2
3 5 3 5
x dx, x dx, 4x dx, 6x dx, (4x3 + 6x5 )dx.
0 0 0 0 0

What can you observe about the relation between

Z b Z b Z b Z b
f (x)dx, g(x)dx, cf (x)dx, (f (x) + g(x))dx.
a a a a

Exercise 3.1.5. Compute definite integral.

Z 2 Z 8√ Z b
1. (x2 − 3x − 4)dx. 4. 3x + 1dx. 7. ex+a dx.
−1 0 0
Z 2
Z 1 √
Z 1
2. (3x + 1)2 dx. 5. (3 + x x)dx. 8. (e−x + sin πx)dx.
0 0 0
2 2 π
Z 2 Z
1
Z
4
3. (3x + 1)(x − 3)dx. 6. x+ dx. 9. sec x tan xdx.
0 1 x 0

Exercise 3.1.6. For non-negative integers m and n, prove that

Z 2π
cos mx sin nxdx = 0;
0

Z 2π 0,
 if m 6= n,
cos mx cos nxdx = π, if m = n 6= 0,
0 
2π, if m = n = 0;

Z 2π (
0, if m 6= n or m = n = 0,
sin mx sin nxdx =
0 π, if m = n 6= 0.

Z b √
Z b
1
Exercise 3.1.7. Compute 3
xdx and √
3
dx. Explain for what range of a, b are the
a a x
formulae valid.

Exercise 3.1.8. What is wrong with the equality?

π
1 x=1
Z
1
Z 
1 2. sec2 xdx = tan π − tan 0 = 0.
1. dx = −  = 2.
−1 x2 x x=−1 0

1
Exercise 3.1.9. What is the area of the unbounded region between √ and the x-axis,
1 − x2
over the interval (−1, 1)?
178 CHAPTER 3. INTEGRATION

3.1.3 Property of Area and Definite Integral

Since the definite integral is the signed area, the usual properties of area is reflected
as properties of the definite integral. An important property of area is the additivity.
Specifically, if X ∩ Y has zero area, then the area of X ∪ Y should be the area of X
plus the area of Y . Translated to definite integral, this means
Z c Z b Z c
f (x)dx = f (x)dx + f (x)dx. (3.1.5)
a a b

The equality can be used to calculate the definite integral of “piecewise continuous”
functions.

Example 3.1.8. The definite integral of the function (which is not continuous at 0)
(
−2x, if − 1 ≤ x < 0,
f (x) =
ex , if 0 ≤ x ≤ 1,

on [−1, 1] is
Z 1 Z 0 Z 1 Z 0 Z 1
f (x)dx = f (x)dx + f (x)dx = (−2x)dx + ex dx
−1 −1 0 −1 0
= −x2 |0−1 + ex |10 = e.

ex
−2x

−1 1

Figure 3.1.6: Definite integral of a piecewise continuous function.

Z 0
We note that the computation of f (x)dx actually reassigns the value f (0) = 0
−1
to make the function continuous on [−1, 0]. The modification happens inside the
vertical line x = 0. Since the vertical line has zero area, this does not affect the
whole integral.
In general, changing the value of the integrand at finitely many places does not
affect the integral.
3.1. AREA AND DEFINITE INTEGRAL 179
Z b
Presumably, the definite integral f (x)dx is defined only for the case a ≤ b,
a
and the equality (3.1.5) implicitly assumes a ≤ b ≤ c. However, by
Z a
f (x)dx = 0,
a

and by taking c = a in (3.1.5), we get

Z a Z b
f (x)dx = − f (x)dx.
b a

This extends the definite integral to the case the upper limit is smaller than the lower
limit. With such extension, the equality (3.1.5) holds for any combination of a, b, c.
Moreover, the extended definite integral is still computed by the antiderivative as
before.
Another important property of area is positivity. Translated into definite inte-
gral, this means
Z b
f ≥ 0 =⇒ f (x)dx ≥ 0, for a < b. (3.1.6)
a
Z b
Note that if a > b, then f (x)dx ≤ 0. The positivity is further extended to
a
monotonicity in Example 3.5.5.
If we shift the graph under f (x) over [a, b] by d, we get the graph under f (x − d)
over [a + d, b + d]. Since the area is not changed by shifting, we get
Z b+d Z b
f (x − d)dx = f (x)dx. (3.1.7)
a+d a

See Exercise for more examples of properties of area implying properties of definite
integral.
In Section 3.5, we will introduce more properties from the the viewpoint of
computation (i.e., Newton-Leibniz formula). Some of these properties cannot be
easily explained by properties of area.
Z 2 Z 4 Z 0 Z 4
Exercise 3.1.10. Suppose f (x)dx = 3, f (x)dx = 2, f (x)dx = 0. Find f (x)dx.
0 5 5 2

Exercise 3.1.11. Use area to explain the equalities.

Z −a Z b
1. f (−x)dx = f (x)dx.
−b a
Z b Z b
2. cf (x)dx = c f (x)dx.
a a
180 CHAPTER 3. INTEGRATION
Z b Z cb
−1
3. f (cx)dx = c f (x)dx.
a ca
Z b
Then express f (Ax + B)dx as some multiple of the definite integral of f (x) over some
a
interval.

Exercise 3.1.12. Compute the integrals.

Z b Z 2 Z 2
2
1. x
a dx. 4. x sign(x)dx. 7. |x2 − 3x + 2|dx.
0 −1 0
Z b Z 2 Z 2π
2. (1 + 2x)n dx. 5. |x|dx. 8. | sin x|dx.
a −1 0
Z 2 Z −1 Z 0
3. sign(x)dx. 6. |x|dx. 9. | sin x|dx.
−1 2 2π

Exercise 3.1.13. Compute the integrals.

Z 2 (
x2 , if x < 1,
1. f (x)dx, f (x) = −2
0 x , if x ≥ 2.
Z 2 (
e2|x| , if |x| < 1,
2. f (x)dx, f (x) =
−2 e−x , if |x| ≥ 1.
(
sin x, if x < π2 ,
Z π
3. f (x)dx, f (x) =
0 cos x, if x ≥ π2 .
(
sin x, if |x| < π2 ,
Z π
4. f (x)dx, f (x) =
−π cos x, if |x| ≥ π2 .

3.2 Rigorous Definition of Integral

3.2.1 What is Area?
The definite integral is defined as the signed area. Therefore the rigorous definition
of integral relies on the rigorous definition of area. Any reasonable definition of area
should have the following three properties (the area of a subset X ⊂ R2 is denoted
µ(X)):

1. Bigger subsets have bigger area: X ⊂ Y implies µ(X) ≤ µ(Y ).

2. Areas can be added: If µ(X ∩ Y ) = 0, then µ(X ∪ Y ) = µ(X) + µ(Y ).

3. Rectangles have the usual area: µ(ha, bi × hc, di) = (b − a)(d − c).
3.2. RIGOROUS DEFINITION OF INTEGRAL 181

Here ha, bi can mean any of [a, b], (a, b), (a, b], or [a, b). A carefully review of the
argument in Section 3.1 shows that nothing beyond the three properties are used.
Suppose a plane region A ⊂ R2 is a union of finitely many rectangles, A = ∪ni=1 Ii ,
such that the intersections between Ii are at most lines. Since lines have zero area
by the third property, we may use the second property to further define µ(A) =
P n
i=1 µ(Ii ). We give such a plane region the temporary name “good region”, since
we have definite idea
Pnabout the area of a good region. (Strictly speaking, we still
n
need to argue that i=1 µ(Ii ) is independent of the decomposition A = ∪i=1 Ii .)

Figure 3.2.1: Good region.

For any (bounded) subset X ⊂ R2 , we may try to approximate X by good

regions, from inside as well as from outside. In other words, we consider good
regions A and B satisfying A ⊂ X ⊂ B. Then by the first property of area, we must
have
µ(A) ⊂ µ(X) ⊂ µ(B).
Note that µ(A) and µ(B) have been defined, and µ(X) is yet to be defined. So we
introduce the inner area (the maximum should really be the supremum)

µ∗ (X) = max{µ(A) : A ⊂ X, A is a good region},

as the lower bound for µ(X), and the outer area (the minimum should really be the
infimum)
µ∗ (X) = min{µ(B) : B ⊃ X, B is a good region},
as the upper bound for µ(X). We say that the subset X has area (or Jordan
measurable) if µ∗ (X) = µ∗ (X), and the common value is the area µ(X) of X. If
µ∗ (X) 6= µ∗ (X), then we say X has no area.
The subset X has area if and only if for any  > 0, there are good regions A
and B, such that A ⊂ X ⊂ B and µ(B) − µ(A) < . In other words, we can
find good inner and outer approximations, such that the difference between the
approximations can be arbitrarily small.
182 CHAPTER 3. INTEGRATION

A
X

Figure 3.2.2: Approximation by good regions.

Example 3.2.1. A point can be considered as a reduced rectangle and has area 0. If X
consists of finitely many points, then we can take B = X be the union of all “point
rectangles” in X. Since µ(B) = 0, we get µ∗ (X) = 0. By 0 ≤ µ∗ (X) ≤ µ∗ (X), we
also have µ∗ (X) = 0. Therefore finitely many points has area 0.

Example 3.2.2. Consider the triangle with vertices (0, 0), (1, 0) and (1, 1). We parti-
tion the interval [0, 1] into n parts of equal length
 
n i−1 i
[0, 1] = ∪i=1 , .
n n

Correspondingly, we have the inner and outer approximations of the triangle

       
n i−1 i i−1 n i−1 i i
An = ∪i=1 , × 0, , Bn = ∪i=1 , × 0, .
n n n n n n

They have area

n n
X 1i−1 1 X 1i 1
µ(An ) = = (n − 1), µ(Bn ) = = (n + 1).
i=1
n n 2n i=1
nn 2n

1
By taking sufficiently big n, the difference µ(Bn ) − µ(An ) = can be arbitrarily
n
small. Therefore the triangle has area, and the area is given by limn→∞ µ(An ) =
1
limn→∞ µ(Bn ) = . This justifies the conclusion of Example 3.1.2 for the case a = 1.
2

Example 3.2.3. For an example of subsets without area, i.e., satisfying µ∗ (X) 6=
µ∗ (X), let us consider the subset X = (Q ∩ [0, 1])2 of all rational pairs in the unit
square.
3.2. RIGOROUS DEFINITION OF INTEGRAL 183

1 1
inner approx. outer approx.

Figure 3.2.3: Approximating triangle.

Since the only rectangles contained in X are single points, any good region
A ⊂ X must be finitely many points. Therefore µ(A) = 0 for any good region
A ⊂ X, and µ∗ (X) = 0.
On the other hand, if B is a good region containing X, then B must almost
contain the whole square [0, 1]2 , with the only exception of finitely many horizontal
or vertical (irrational) line segments. Therefore we have µ(B) ≥ µ([0, 1]2 ) = 1. This
implies µ∗ (X) ≥ 1 (show that µ∗ (X) = 1!).

Exercise 3.2.1. Use inner and outer approximations to explain that any rectangle has area
given by the multiplication of two sides. This justifies Example 3.1.1.

Exercise 3.2.2. Explain that a (not necessarily horizontal or vertical) straight line segment
has area 0.

Exercise 3.2.3. Explain that the region between y = x and the x-axis over [0, a] has area
1 2
a . This fully justifies the computation in Example 3.1.2.
2

Exercise 3.2.4. Explain that the subset X = (Q ∩ [0, 1]) × [0, 1] of all vertical rational lines
in the unit square has no area.

Exercise 3.2.5. Show that if X ⊂ Y , then µ∗ (X) ≤ µ∗ (Y ) and µ∗ (X) ≤ µ∗ (Y ). In

particular, we have µ(X) ≤ µ(Y ) in case both X and Y have areas. The property is used
in deriving (3.1.2).

3.2.2 Darboux Sum

After the rigorous definition of area, we can give the rigorous definition of definite
integral.

Definition 3.2.1. A function f (x) is Riemann integrable if the region

G[a,b] (f ) = {(x, y) : a ≤ x ≤ b, y is between 0 and f (x)}

184 CHAPTER 3. INTEGRATION

has area. Moreover, the Riemann integral is

Z b



f (x)dx = µ G[a,b] (f ) ∩ H+ − µ G[a,b] (f ) ∩ H− ,
a
where
H+ = {(x, y) : y ≥ 0}, H− = {(x, y) : y ≤ 0}
are the upper and lower half planes.

Suppose f ≥ 0 on [a, b]. As indicated by Figure 3.2.4, for any inner approxima-
tion of G[a,b] (f ), we can always choose “full vertical strips” to get a better approxima-
tion for G[a,b] (f ). Here better means closer to the expected value of µ(G[a,b] (f )). The
outer approximations have similar improvements by full vertical strips. Therefore
we only need to consider the approximations by full vertical strips.

mi

a b a xi−1 xi b

Figure 3.2.4: Better inner approximations by vertical strips.

An approximation by full vertical strips is determined by a partition of the in-

terval
P : a = x0 < x1 < · · · < xn = b.
On the i-th interval [xi−1 , xi ], the inner strip has height mi = min[xi−1 ,xi ] f (the mini-
mum should really be the infimum), and the outer strip has height Mi = max[xi−1 ,xi ] f
(the maximum should really be the supremum). Therefore the inner and outer ap-
proximations are
AP = ∪ni=1 [xi−1 , xi ] × [0, mi ) ⊂ X ⊂ BP = ∪ni=1 [xi−1 , xi ] × [0, Mi ].
The areas of the two approximations are the lower and upper Darboux sums
n
X
L(P, f ) = µ(AP ) = mi (xi − xi−1 ),
i=1
n
X
U (P, f ) = µ(BP ) = Mi (xi − xi−1 ).
i=1

The Riemann integrability of f means that G[a,b] (f ) has area, which further
means that the difference between inner and outer approximations can be arbitrarily
small. Therefore we get the following criterion for the integrability.
3.2. RIGOROUS DEFINITION OF INTEGRAL 185

Theorem 3.2.2 (Riemann Criterion). A bounded function f on [a, b] is Riemann

integrable, if and only if for any  > 0, there is a partition P , such that
n 
X 
U (P, f ) − L(P, f ) = max f − min f (xi − xi−1 ) < .
[xi−1 ,xi ] [xi−1 ,xi ]
i=1

The quantity
ω[xi−1 ,xi ] f = max f − min f
[xi−1 ,xi ] [xi−1 ,xi ]

measures how much the value of f fluctuates on [xi−1 , xi ] and is called the oscillation
of the function on the interval. Since the continuity of a function can imply that such
oscillations are uniformly small, continuous functions are always Riemann integrable.
The criterion also implies that monotone functions are Riemann integrable. On the
other hand, there are functions that are not Riemann integrable.

Example 3.2.4. Consider the Dirichlet function D(x) on [0, 1]. We always have

min D = 0, max D = 1.
[xi−1 ,xi ] [xi−1 ,xi ]

Therefore U (P, f ) − L(P, f ) = 1 cannot be arbitrarily small. We conclude that the

Dirichlet function is not Riemann integrable.
The example is closely related to Example 3.2.3. See Exercise 3.2.4.

We note that U (P, f ) − L(P, f ) is the area of the good subset BP − AP =

∪ni=1 [xi−1 , xi ]×[mi , Mi ].
If we consider all the partitions P , the all such good subsets
are essentially all the outer approximations of the graph {(x, f (x)) : a ≤ x ≤ b} of
f (a curve, not including the part below f ). Therefore Theorem 3.2.2 basically says
that a function is Riemann integrable if and only if the graph curve of the function
has zero area.
The graph curve is part of the boundary of G(f ). In this viewpoint, Theorem
3.2.2 is a special case of the following.

Theorem 3.2.3. A bounded subset X ⊂ R2 has area if and only if its boundary ∂X
has zero area.

We remark that the theory of area can be easily extended to the theory of
volume for subsets in Rn . We may then get the rigorous definition of multivariable
Riemann integrals on subsets of Euclidean spaces, where the subsets should have
volume themselves. The high dimensional versions of Theorems 3.2.2 and 3.2.3 are
still valid.
Further extension of the area theory introduces countably many in place of
finitely many. The result is the modern theory of Lebesgue measure and Lebesgue
integral.
186 CHAPTER 3. INTEGRATION

3.2.3 Riemann Sum

Suppose f is Riemann integrable. Intuitively, when the partition gets more and
more refined, the upper and lower Darboux sums, as the
Z areas of the outer and
b
inner approximations, will become closer to the integral f (x)dx.
a
For each interval [xi−1 , xi ] of the partition P , let us choose a sample point x∗i ∈
[xi−1 , xi ] so that
mi ≤ f (x∗i ) ≤ Mi .
Then we get the Riemann sum
n
X
S(P, f ) = f (x∗i )(xi − xi−1 ), x∗i ∈ [xi−1 , xi ].
i=1

sandwiched between the two Darboux sums

L(P, f ) ≤ S(P, f ) ≤ U (P, f ).

Z b
We conclude that S(P, f ) will also become closer to the integral f (x)dx.
a
The data (P, {x∗i }) of a partition P together with the choice of sample points
{x∗i }
is called a tagged partition.

Theorem 3.2.4. Suppose f is Riemann integrable on [a, b]. Then for any  > 0,
there is a tagged partition P0 of [a, b], such that for any refinement P of P0 , we have
 Z b 
 
S(P, f ) − f (x)dx < .

a

Here a refinement of P0 is a tagged partition P obtained by adding more partition

points to P0 and choosing additional sample points x∗i if necessary.
The statement above is very similar to the limit of sequences and functions. (It
is known as the limit of a net in topology). The subtlety here is that refinement of
partitions replaces n > N or |x − a| < δ.

For practical purposes (see the next section), we usually take the partition to
be regular, i.e. the subintervals are spaced evenly. However, the definitions of both
Darboux and Riemann sums allow arbitrary partitions, and usually √ it is not possible
to obtain a refinement which is also regular (e.g. take P : 1 ≤ 2 ≤ 2).
Luckily, we have the following result (which is equivalent to both Theorem 3.2.2
and Theorem 3.2.4), that allows us to use any partition as long as the subinterval
widths are small enough.
3.3. NUMERICAL CALCULATION OF INTEGRAL 187
Z b
Theorem 3.2.5. Suppose f is Riemann integrable on [a, b] with s = f (x)dx.
a
Then for any  > 0, there exists δ > 0 such that for any partition P with kP k < δ
and any sample points x∗i for S(P, f ), we have
|L(P, f ) − s| < , |S(P, f ) − s| < , |U (P, f ) − s| < ,
where kP k = max (xi − xi−1 ) is the largest subinterval width (“mesh size”).
1≤i≤n
Equivalently,
s −  < L(P, f ) ≤ S(P, f ) ≤ U (P, f ) < s + .

In particular, this justisfies all the calculations using regular partitions Pn in

the next section. As the number of partition points n increases to infinity, the
subinterval width tends to zero, which implies that the limit of the Riemann sums
will converge to the Riemann integrals.

3.3 Numerical Calculation of Integral

Although Riemann integrals of an integrable function f can be computed by the
Newton-Leibniz formula, it is often impossible to find the exact formula of a function
F satisfying F 0 = f . Moreover, even if we can find a formula for F , it might be too
complicated to evaluate. For many practical applications, it is sufficient to find an
approximate value of the integration. Many efficient numerical schemes have been
invented for this purpose.
All the schemes are the extensions of the Riemann sum in Section 3.2.3. As
explained in the last part of Section 3.2.3, it suffices to consider partitions that
evenly divides the interval
Pn : a = x0 < x1 = a + h < · · · < xi = a + ih < · · · < xn = b = a + nh,
where
b−a
h= = xi − xi−1
n
is the step size of the partition.

3.3.1 Left and Right Rule

By taking all the sample points to be the left of the partition intervals, we get
x∗i = xi−1 = a + (i − 1)h and the left rule
Ln = h(f (x0 ) + f (x1 ) + · · · + f (xn−1 )).
By taking all the sample points to be the right of the partition interval, we get
x∗i = xi = a + ih and the right rule
Rn = h(f (x1 ) + f (x2 ) + · · · + f (xn )).
188 CHAPTER 3. INTEGRATION

a b a b

Figure 3.3.1: Left and right rules.

Example 3.3.1. For f (x) = x on [0, 1], we have

 
1 0 1 n−1
Ln = + + ··· +
n n n n
1 1 1 n−1
= 2 (0 + 1 + · · · + (n − 1)) = 2 (n − 1)n = ,
n  n 2 2n
1 1 2 n
Rn = + + ··· +
n n n n
1 1 1 n+1
= 2 (1 + 2 + · · · + n) = 2 n(n + 1) = .
n n 2 2n
Z 1
1
Both converge to = xdx as n → ∞.
2 0

1
Example 3.3.2. To compute the integral of f (x) = on [0, 1], we take n = 4.
1 + x2
The partition is
P4 : 0 < 0.25 < 0.5 < 0.75 < 1, h = 0.25.
The values of f (x) at the five partition points are

1.000000, 0.941176, 0.800000,

0.640000, 0.500000.
Z 1
dx
Then we get the following approximate values of 2
0 1+x

L4 = 0.25 × (1.000000 + 0.941176 + 0.800000 + 0.640000) ≈ 0.845294,

R4 = 0.25 × (0.941176 + 0.800000 + 0.640000 + 0.500000) ≈ 0.720294.
π
By Example 3.1.7, the actual value is = 0.7853981634 · · · .
4

Exercise 3.3.1. Find Ln and Rn and confirm the value of related integral.
3.3. NUMERICAL CALCULATION OF INTEGRAL 189

1. f (x) = x on [a, b]. 3. f (x) = 2x on [0, 1].

2. f (x) = x2 on [0, 1]. 4. f (x) = ax on [0, 1].

Exercise 3.3.2. Explain the identity.

  Z 1
1 2 n−1
1. limn→∞ + + ··· + = xdx
n2 n2 n2 0
  Z 1
1 1 1 dx
2. limn→∞ + + ··· + = .
n+1 n+2 2n 0 1+x
  Z 1
n n n dx
3. limn→∞ 2 2
+ 2 2
+ ··· + 2 2
= 2
.
n +1 n +2 n +n 0 1+x
  Z 1
1 π 2π (n − 1)π
4. limn→∞ sin + sin + · · · + sin = sin πxdx.
n n n n 0

Exercise 3.3.3. Interpret the limit as integration.

1p + 2p + · · · + np
1. limn→∞ .
np+1
 
1 Pn b−a
2. limn→∞ f a+k .
n k=1 n
r
Z 2
1 n (2n)!
Exercise 3.3.4. By interpreting log xdx, find limn→∞ .
1 n n!

3.3.2 Midpoint Rule and Trapezoidal Rule

The left and right rules are quite primitive approximations of the integral. A better
choice is the middle points
xi−1 + xi 2i − 1
x̄i = =a+ h
2 2
and the corresponding Riemann sum

Mn = h(f (x̄1 ) + f (x̄2 ) + · · · + f (x̄n )).

Another choice is the average of the Riemann sums using the left and right points.
Ln + Rn h
Tn = = (f (x0 ) + 2f (x1 ) + 2f (x2 ) + · · · + 2f (xn−1 ) + f (xn )).
2 2
The two approximation schemes are the midpoint rule and the trapezoidal rule.
Since Mn , Ln , Rn are Riemann sums, they converge to the integral of f (x). Conse-
quently Tn also converges.
190 CHAPTER 3. INTEGRATION

a b a b

Figure 3.3.2: Midpoint and trapezoidal rules.

Example 3.3.3. For f (x) = x on [0, 1], we have

 
1 1 3 2n − 1
Mn = + + ··· +
n 2n 2n 2n
1 1 1
= 2 (1 + 3 + · · · + (2n − 1)) = 2 n2 = ,
2n  2n 2
1 n−1 n+1 1
Tn = + = .
2 2n 2n 2
Z 1
1
Both happen to be equal to = xdx.
2 0

Example 3.3.4. For f (x) = x2 on [0, 1], we have

1 12 32 (2n − 1)2
 
Mn = + + ··· +
n 4n2 4n2 4n2
1
= 3 [(12 + 22 + · · · + (2n)2 ) − 22 (12 + 23 + · · · + n2 )]
4n 
4n2 − 1

1 1 4
= 3 2n(2n + 1)(4n + 1) − n(n + 1)(2n + 1) = ,
4n 6 6 12n2
1 02 12 22 (n − 1)2 n2
 
Tn = + 2 2 + 2 2 + ··· + 2 + 2
2n n2 n n n2 n
1
= 3 [2(12 + 22 + · · · + n2 ) − n2 ]
2n 
2n2 + 1

1 1 2
= 3 2 n(n + 1)(2n + 1) − n = .
2n 6 6n2
Z 1
1 1 1
Compared with the actual value x2 dx = , the errors are 2
and .
0 3 12n 6n2
Z 1
dx
Example 3.3.5. We apply the midpoint and trapezoidal rules to . For
0 1 + x2
3.3. NUMERICAL CALCULATION OF INTEGRAL 191

n = 4, we have

M4 = 0.25 × (0.984615 + 0.876712 + 0.719101 + 0.566372) ≈ 0.786700,

0.25
T4 = × (1.000000 + 2 × 0.941176 + 2 × 0.800000 + 2 × 0.640000 + 0.500000)
2
≈ 0.782794.

For n = 8, we have h = 0.125 and the following values.

1 1
i xi x̄i
1 + x2i 1 + x̄2i
0 0 1.000000
1 0.125 0.984615 0.0625 0.996109
2 0.25 0.941176 0.1875 0.966038
3 0.325 0.876712 0.3125 0.911032
4 0.5 0.800000 0.4375 0.839344
5 0.625 0.719101 0.5625 0.759644
6 0.75 0.640000 0.6875 0.679045
7 0.875 0.566372 0.8125 0.602353
8 1 0.500000 0.9375 0.532225

Then we get the approximations

M8 ≈ 0.785721, T8 ≈ 0.784747.
Z 1
dx π
Compared with the actual value 2
= = 0.7853981634 · · · , the follow-
0 1+x 4
ing are the errors of various schemes.

error n=4 n=8

|Rn − I| 0.065104 0.031901
|Ln − I| 0.059896 0.030599
|Mn − I| 0.001302 0.000323
|Tn − I| 0.002604 0.000651

We observe that the midpoint and trapezoidal rules are much more accurate,
and the error for the midpoint rule is about half of the error for the trapezoidal rule.
Moreover, doubling the number of partition points improves the error by a factor of
4 for the two rules. The following gives an estimation of the errors.

Theorem 3.3.1. Suppose f 00 (x) is continuous and bounded by K2 on [a, b], then
Z b Z b
 K2 (b − a)3  K2 (b − a)3
 
 
 f (x)dx − Mn
≤ ,  f (x)dx − Tn
≤ .

a
 24n2 
a
 12n2
192 CHAPTER 3. INTEGRATION

The estimations are derived in Exercises 3.5.9 and 3.5.17.

Z b
Exercise 3.3.5. To compute the integral x2 dx, for any partition of [a, b], we take x∗i =
r a
1 2
(x + xi−1 xi + x2i ) ∈ [xi−1 , xi ]. Show that the Riemann sum is exactly the value of
3 i−1 Z b
the integral. How can you generalize this to xn dx?
a

Z b
Exercise 3.3.6. For the integral x2 dx, we take any partition of [a, b], in which the
a
intervals may not have the same length. Estimate the error of the various schemes in
terms of the size δ = maxni=1 (xi − xi−1 ) of the partition.

Exercise 3.3.7. Apply the midpoint and trapezoidal rules to the integral and compare with
the actual value.
Z 2
Z π
dx
1. , n = 6, 12. 2. sin xdx, n = 4, 12.
1 x 0

Exercise 3.3.8. Apply the midpoint and trapezoidal rules to the integral. Moreover, esti-
mate the number of partition points needed for the approximation to be accurate up to
10−6 .
Z π Z 1
cos x2 dx, n = 5, 10.
2
1. 4. ex dx, n = 10.
0 0
Z π 2
sin x
Z
1
2. dx, n = 5, 10. 5. e x dx, n = 10.
0 x 1
Z 2 Z 2
1 log x
3. √ dx, n = 5, 10. 6. dx, n = 10.
0 1 + x3 1 1+x
1
Exercise 3.3.9. Show that T2n = (Mn + Tn ).
2
Z b
Exercise 3.3.10. Prove that if f is a concave positive function, then Tn ≤ f (x)dx < Mn .
a

3.3.3 Simpson’s Rule

The midpoint rule is based on the constant approximation, and the trapezoidal rule
is based on linear approximation (actually not quite, as average of two constant
approximations). We may expect better approximation by using quadratic curves.
Since a quadratic curve is determined by three points, we try to approximate f (x)
on the interval [xi−1 , xi+1 ] by the quadratic function Q(x) = A(x−xi )2 +B(x−xi )+C
3.3. NUMERICAL CALCULATION OF INTEGRAL 193

satisfying

f (xi−1 ) = Q(xi−1 ) = Ah2 − Bh + C,

f (xi ) = Q(xi ) = C,
f (xi+1 ) = Q(xi+1 ) = Ah2 + Bh + C.
Z xi+1
Then f (x)dx is approximated by (the first equality uses (3.1.7))
xi−1
Z xi+1
2 h
Q(x)dx = Ah3 + 2Ch = (f (xi−1 ) + 4f (xi ) + f (xi+1 )).
xi−1 3 3

Suppose n is even. Then we may apply the quadratic approximations to [x0 , x2 ],

[x2 , x4 ], . . ., [xn−2 , xn ]. Adding such approximations together, we get an approxima-
Z b
tion of f (x)dx
a

h
Sn = (f (x0 )+4f (x1 )+2f (x2 )+4f (x3 )+2f (x4 )+· · ·+2f (xn−2 )+4f (xn−1 )+f (xn )).
3
1
This is Simpson’s rule. Observe that Sn = (2Tn + M n2 ) is the weighted average
3
of the trapezoidal (with step size h) and midpoint (with step size 2h) rules. As a
result, it also converges to the integral of f (x).
The errors in Simpson’s rule can be estimated by the bound on the fourth order
derivative.

Theorem 3.3.2. Suppose f (4) (x) is continuous and bounded by K4 on [a, b], then
Z b
 K4 (b − a)5


 f (x)dx − S n
≤ .

a
 180n4

A consequence of the theorem is that doubling the partition improves the error
by a factor of 16!
Z 1
dx
Example 3.3.6. Applying Simpson’s rule to for n = 4, we use the same
0 1 + x2
data from Example 3.3.1 to get
0.25
S4 = × (1.000000 + 4 × 0.941176 + 2 × 0.800000 + 4 × 0.640000 + 0.500000)
3
≈ 0.785392.

The error |Sn − I| = 0.000540 is comparable to the midpoint and trepezoidal rule
for n = 8.
194 CHAPTER 3. INTEGRATION

How many partition points are needed in order to get the approximate value
accurate up to the 6-th digit? To answer the question, we compute the derivatives

24(5x4 − 10x2 + 1) 240x(x2 − 3)(3x2 − 1)

f (4) (x) = , f (5) (x) = − .
(1 + x2 )5 (1 + x2 )6

1
From f (5) (x), the extrema of f (4) (x) on [0, 1] can only be at 0, √ or 1. By
3
  
(4)
 (4) 1  81
|f (0)| = 24, f  √  = , |f (4) (1)| = 3,
3 8

we get K4 = 24. Then the question becomes

24
≤ 10−6 .
180n4
Therefore we need n ≥ 19.1. Since n should be an even integer, this means n ≥ 20.
We may carry out the similar estimation for the midpoint and trapezoidal rules.
We find K2 = |f 00 (0)| = 2, so that the estimations become

2 2
≤ 10−6 , ≤ 10−6 .
24n2 12n2
The answers are respectively n ≥ 289 and n ≥ 409.

Exercise 3.3.11. Repeat Exercise 3.3.7 for the Simpson’s rule.

Exercise 3.3.12. Repeat Exercise 3.3.8 for the Simpson’s rule.

Z 1
dx
Exercise 3.3.13. If we apply Simpson’s rule to 2
to get an approximate value for
0 1+x
π accurate up to 10−6 , how many partition points do we need?

Exercise 3.3.14. Simpson’s 3/8 rule is obtained by using cubic instead of quadratic approx-
imation. Derive the formula for this rule.

3.4 Indefinite Integral

3.4.1 Fundamental Theorem of Calculus
Z x
0
The Newton-Leibniz formula is derived from A (x) = f (x), where A(x) = f (t)dt
a
is the signed area over [a, x]. The equality is summarized below.
3.4. INDEFINITE INTEGRAL 195

Theorem 3.4.1 (Fundamental Theorem of Calculus). If f (x) is continuous on [a, b],

then A(x) is continuous on [a, b] and differentiable on (a, b), with
Z x
d
f (t)dt = f (x), x ∈ (a, b).
dx a

Note that the continuity was used critically in our argument for A0 (x) = f (x).
Note also that if f (x) is only integrable (hence bounded), then A(x) is still
continuous on [a, b] by the same argument as in(3.1.1).
Z x2
Example 3.4.1. Let f (x) be a continuous function. To find the derivative of f (t)dt,
a
we note that the integral is a composition
Z x2 Z x
2
f (t)dt = A(x ), A(x) = f (t)dt.
a a

By the chain rule and the Fundamental Theorem of Calculus, we have

x2
dA(x2 )
Z
d
f (t)dt = = A0 (x2 )2x = 2xf (x2 ).
dx a dx

The Fundamental Theorem also implies the following derivatives

Z x
d
f (t2 )dt = f (x2 ),
dx a
Z x
d
f (t)2 dt = f (x)2 ,
dx a
Z x 2 Z x
d
f (t)dt = 2f (x) f (t)dt.
dx a a

Z x
2
Example 3.4.2. The function f (x) = et dt cannot be expressed as combinations
0
2
of the usual elementary functions. Still, we know f 0 (x) = ex . We also have
Z x2 Z x2 Z x !
d 2 d 2 2 4 2
et dt = et dt − et dt = 2xex − ex .
dx x dx 0 0

Example 3.4.3. For the sign function


1,
 if x > 0,
sign(x) = 0, if x = 0,

−1, if x < 0,

196 CHAPTER 3. INTEGRATION

we have 
Z x 1,
 if x > 0,
0
A(x) = sign(x)dx = |x|, A (x) = no, if x = 0,
0 
−1, if x < 0.


We note that A(x) is not differentiable at 0, exactly the place where the sign function
is not continuous. The example shows that the continuity assumption cannot be
dropped from the Fundamental Theorem.

Example 3.4.4. The sine integral function is

Z x
sin t
Si(x) = dt.
0 t
Since the integrand can be made continuous by assigning value 1 at 0, we know

 sin x
0 , if x 6= 0,
Si (x) = x
1, if x = 0.

Therefore the function Si(x) is strictly increasing on the following intervals

. . . , [−5π, −4π], [−3π, −2π], [−π, π], [2π, 3π], [4π, 5π], . . . ,

and is strictly decreasing on the following intervals

. . . , [−4π, −3π], [−2π, −π], [π, 2π], [3π, 4π], . . . .

This implies that Si(x) has local maxima at . . . , −6π, −4π, −2π, π, 3π, 5π, . . . , and
has local minima at . . . , −5π, −3π, −π, 2π, 4π, 6π, . . . . Moreover, we can also calcu-
late the second order derivative

 x cos x − sin x
00 , if x 6= 0,
Si (x) = x2
0, if x = 0,

and find the convexity property of Si(x).

Example 3.4.5. Suppose f is a continuous function satisfying

Z x Z x
2 tf (t)dt = x f (t)dt.
0 0

Then by taking derivative on both sides, we get

Z x
2xf (x) = xf (x) + f (t)dt.
0
3.4. INDEFINITE INTEGRAL 197
Z x
Let F (x) = f (t)dt. Then the equation is the same as xF 0 (x) = F (x). This
0
implies 0
xF 0 (x) − F (x)

F (x)
= = 0.
x x2
Z x
Therefore f (t)dt = F (x) = Cx for a constant C, and we further find that
0
f (x) = C is a constant.

Exercise 3.4.1. Find the derivative of function.

Z x Z x2 Z x3 Z x3
1. t3 dt. 2. t3 dt. 3. 2
t dt. 4. t2 dt.
0 0 0 x2

Exercise 3.4.2. Find the derivative of function.

Z x Z π Z π
dt
1. 3
. 3. cos t2 dt. 5. arctan tdt.
1 1 + t x tan x
Z x2 Z 1q √ Z cot x 3
2. log(1 + t2 )dt. 4. 1 + tdt. 6. (1 + t2 ) 2 dt.
1 x2 tan x

Exercise 3.4.3. Let f (x) be a continuous function. Find the derivative.

Z b Z x Z b Z f (x)
1. f (t)dt. 3. f (t2 )dt. 5. f (sin t)dt. 7. f (t)dt.
x2 a x 0
Z x2 Z x Z cos x Z f −1 (x)
2. f (t)dt. 4. ef (t) dt. 6. f (t)dt. 8. f (t)dt.
x a sin x 0

In the 7-th problem, f is differentiable. In the 8-th problem, f is invertible and

differentiable.

Exercise 3.4.4. Study the monotone and convex properties, including the extrema and the
points of inflection.
Z x Z x Z x
dt πt2 2 2
1. 2
. 2. sin dt. 3. √ e−t dt.
0 1 + t + t 0 2 π 0

Exercise 3.4.5. Find continuous functions f (x) satisfying the equality.

Z x Z 1
1. f (t)dt = f (t)dt on [0, 1].
0 x
Z x Z x
2. A tf (t)dt = x f (t)dt on (0, +∞).
0 0
Z x
3. (f (x))2 = 2 f (t)dt on (−∞, +∞).
0
198 CHAPTER 3. INTEGRATION
Z x Z x
4. f (t)dt = (2x − 1)e 2x
+ e−t f (t)dt on (−∞, +∞).
0 0

Exercise 3.4.6. Find the limit.

1 x sin t 2
Z Z x
1. limx→0 dt. t2
x 0 t e dt
Z x 3. limx→+∞ Z0 x .
1 e2t dt
2
2. limx→0 3 sin t2 dt.
x 0 0

Exercise 3.4.7. Prove that for a positive continuous function f (x) on (0, +∞), the function
Z x 2
tf (t)dt
0
g(x) = Z x
f (t)dt
0
is strictly increasing on (0, +∞).
Z x
Exercise 3.4.8. Discuss where f (x) is not continuous and where f (t)dt is not differen-
0
tiable.
(  0
x, if x 6= 0,  x2 sin 1 ,

if x 6= 0,
1. f (x) = .
1, if x = 0, 3. f (x) = x .

0,
( if x = 0,
x, if x > 0,
2. f (x) = .
1, if x ≤ 0,

3.4.2 Indefinite Integral

A function F is an antiderivative of f if F 0 = f . By Theorem 2.4.3, the antiderivative
is unique up to adding constants. Therefore we denote the set of all antiderivatives
of f by
Z
f (x)dx = F (x) + C, for some F (x) satisfying F 0 (x) = f (x).

This is the indefinite integral of f .

The Newton-Leibniz formula says that the definite integral of a continuous func-
tion can be calculated from the antiderivative
Z b
f (x)dx = F (b) − F (a).
a

The Fundamental Theorem of Calculus says that the signed area gives one an-
tiderivative, and can be used as F (x) above
Z Z x
f (x)dx = f (t)dt + C.
a
3.4. INDEFINITE INTEGRAL 199

Example 3.4.6. By
1
(xp+1 )0 = (p + 1)xp , (log |x|)0 = , (ex )0 = ex ,
x
we get  p+1
Z x + C, for p 6= −1,
Z
p
x dx = p + 1 ex dx = ex + C.
log |x| + C, for p = −1;


More generally, we have

(ax + b)p+1

Z 
 + C, for p 6= −1, Z
ax
(ax + b)p dx (p + 1)a ax dx = + C.
 1 log |ax + b| + C, for p = −1;
 log a
a

Example 3.4.7. The antiderivative of the logarithmic function is more complicated

Z
log |x|dx = x log |x| − x + C.

The equality can be verified by taking the derivative

1
(x log |x| − x)0 = log |x| + x − 1 = log |x|.
x
Z
Example 3.5.9 gives the systematic way of deriving log |x|dx.

Example 3.4.8. The derivatives of the trigonometric functions give us

Z Z
cos xdx = sin x + C, sin xdx = − cos x + C,
Z Z
sec2 xdx = tan x + C, sec x tan xdx = sec x + C.

The antiderivatives of tan x and sec x are more complicated, and are given in Ex-
ample 3.5.27.

Example 3.4.9. The derivative of the inverse sine function gives

Z Z
dx dx x
√ = arcsin x + C, √ = arcsin + C, a > 0.
1−x 2 2
a −x 2 a
Similarly, the derivative of the inverse tangent function gives
Z Z
dx dx 1 x
= arctan x + C, = arctan + C.
x2 + 1 x 2 + a2 a a
Z Z
dx dx
The similar integrals 2 2
and √ are given in Exercise 3.4.10 and
x −a x2 + a
Examples 3.5.2, 3.5.31, 3.5.32.
200 CHAPTER 3. INTEGRATION

Exercise 3.4.9. Compute the integrals.

√
Z Z Z
1. 4
1 − xdx. 3. ax dx. 5. csc x cot xdx.
Z
1
Z Z
2 dx
2. √ dx. 4. csc xdx. 6. .
3
2x + 1 cos2 x

Exercise 3.4.10. Verify the antiderivatives.

log |x|
Z
1
1. dx = (log |x|)2 + C.
x 2
eax
Z
2. eax cos bxdx = 2 (a cos bx + b sin bx) + C.
a + b2
Z
1
3. cos(ax + b)dx = sin(ax + b) + C.
a

a2
Z p
x 1 p
4. a2 − x2 dx = arcsin + x a2 − x2 + C, a > 0.
2 a 2
 
x − a
Z
dx 1
5. = log   + C.
x2 − a2 2a x + a
Z
dx  p 
6. √ = log x + x2 + a + C.
 
2
x +a
Z p
1 p a  p 
7. x2 + adx = x x2 + a + log x + x2 + a + C.
 
2 2
Z Z
Exercise 3.4.11. If f (x)dx = F (x) + C, then what is f (ax + b)dx? Apply your
conclusion to compute the integrals.
Z Z Z
dx
1. log |a + bx|dx. 3. sec2 (3x − 1)dx. 5. p .
x(1 − x)
Z
Z dx Z
dx
2. sin(ax + b)dx. 4. p . 6. .
1 − (x − 1)2 x2 + 2x + 2

Exercise 3.4.12. Find the antiderivative of x(ax2 + b)p . Then compute the integrals.
Z p Z Z
xdx xdx
1. x x2 + 3dx. 2. 2
. 3. √ .
x +1 4 − x2

Exercise 3.4.13. Compute the integrals.

Z Z Z
2
1. x sin(ax2 + b)dx. 2. xeax +b dx. 3. x2 (ax3 + b)p dx.
3.5. PROPERTIES OF INTEGRATION 201

One should not just mindlessly compute the antiderivative. Sometimes we need
to consider the meaning of antiderivative and question whether the answer makes
sense.

Example 3.4.10. Without much thinking, we may write


Z  1 x2 + C,

if x ≥ 0,
|x|dx = 2 1
− x2 + C, if x < 0.

2
However, the constant C in the two cases cannot be independently chosen because
the antiderivative must be differentiable and is therefore continuous at 0. The more
sensible answer is 
Z  1 x2 ,

if x ≥ 0
|x|dx = 2 1 + C.
− x2 , if x < 0

2
In other words, the constant C is two cases must be equal.
For another example, instead of
( (
ex , if x ≥ 0, ex + C, if x ≥ 0,
Z
f (x) = f (x)dx =
1, if x < 0, x + C, if x < 0,
we should have (
ex , if x ≥ 0
Z
f (x)dx = + C.
x + 1, if x < 0

Exercise 3.4.14. Compute indefinite integral.

( (
x2 , if x ≤ 0, 1 − x2 , if |x| ≤ 1,
1. . 2. .
sin x, if x > 0. sin(1 − |x|), if |x| > 1.

Exercise 3.4.15. Does the sign function


1,
 if x > 0,
sign(x) = 0, if x = 0,

−1, if x < 0,


have antiderivative? Does the function have definite integral? What do you learn from
the example?

3.5 Properties of Integration

Computationally, integration is the reverse of differentiation. Therefore properties
of differentiation have corresponding properties of integration.
202 CHAPTER 3. INTEGRATION

3.5.1 Linear Property

Suppose F (x) and G(x) are antiderivatives of f (x) and g(x). Then the linear prop-
erty of the derivative

(F (x) + G(x))0 = F 0 (x) + G0 (x) = f (x) + g(x), (cF (x))0 = cF 0 (x) = cf (x),

implies the linear property of the antiderivative

Z Z Z Z Z
(f (x) + g(x))dx = f (x)dx + g(x)dx, cf (x)dx = c f (x)dx.

By the Newton-Leibniz formula, we get the linear property for the definite integral
Z b
(f (x) + g(x))dx = (F (b) + G(b)) − (F (a) + G(a))
a
Z b Z b
= (F (b) − F (a)) + (G(b) − G(a)) = f (x)dx + g(x)dx,
a a
Z b Z b
cf (x)dx = cF (b) − cF (a) = c(F (b) − F (a)) = c f (x)dx.
a a

Example 3.5.1. We have

Z Z Z Z Z
x(1 + x) dx = (x + 2x + x )dx = xdx + 2 x dx + x3 dx
2 2 3 2

1 2 1
= x2 + x3 + x4 + C.
2 3 4
This furthers gives the definite integral
Z 1
1 2 1 17
x(1 + x)2 dx = (12 − 02 ) + (13 − 03 ) + (14 − 04 ) = .
0 2 3 4 12
Z
On the other hand, it would be very complicated to compute x(x + 1)10 dx by
the binomial expansion of (1 + x)10 . The following is much simpler
Z Z Z Z
x(x + 1) dx = ((x + 1) − 1)(x + 1) dx = (x + 1) dx − (x + 1)10 dx
10 10 11

1 1 1
= (x + 1)12 − (x + 1)11 + C = (11x − 1)(x + 1)11 + C.
12 11 12 · 11

Exercise 3.5.1. Compute the integrals.

3.5. PROPERTIES OF INTEGRATION 203

√
Z Z Z  2
1. x x + 1dx. 5. (x − 1)(x + 1)p dx. x−1
9. dx.
x
Z
x 2
Z Z 
2. p
x(ax + b) dx. 6. dx. x−1
(x + 1)10 10. dx.
x2

x2 − x + 1
Z Z Z  2
2 p x−1
3. x (ax + b) dx. 7. dx. 11. dx.
(x + 1)10 x+1

x−1 (x − 1)2
Z Z Z
4
4. (x − 1)(x + 1) 3 dx. 8. √ dx. 12. dx.
x (x + 1)4

Exercise 3.5.2. Find A, B satisfying

ax + b B
=A+ .
cx + d cx + d

ax + b 2
 
ax + b
Then compute the antiderivatives of and .
cx + d cx + d

Exercise 3.5.3. Compute the integrals.

Z Z
2x+1 − 3x−1
Z
x −x 2
1. (e − e ) dx. 2. (2x + 3x )2 dx. 3. dx.
6x
1
Example 3.5.2. To find the antiderivative of , we use
x2
− a2
 
1 1 1 1 1
= = −
x 2 − a2 (x − a)(x + a) 2a x − a x + a

to get
 
x − a
Z
dx 1 1 1
= log |x − a| − log |x + a| + C = log   + C.
x 2 − a2 2a 2a 2a x + a

This furthers gives the definite integral

Z 3    
1 1 3 − 1 1 2 − 1 1
dx = log   − log   = log 3 .
2
2 x −1 2 3+1 2 2 + 1 2 2
 

As noted in Example 3.1.4, we should not blindly use the Newton-Leibniz formula
in competing the definite integral. For example, we cannot get
Z 2    
1 1 2 − 1 1 0 − 1
dx = log   − log   = − 1 log 3,
2
0 x −1 2 2+1 2 0 + 1 2
 

because the interval [0, 2] contains 1, where the integrand approaches infinity.
204 CHAPTER 3. INTEGRATION

Example 3.5.3. The idea in Example 3.5.2 can be extended

Z 
Z  Z  
dx 1 1 1 1 1
= − dx = − dx
x(x + 1)(x + 2) x x+1 x+2 x(x + 1) x(x + 2)
Z    
1 1 1 1 1
= − − − dx
x x+1 2 x x+2
1 1
= log |x| − log |x + 1| + log |x + 2| + C
2   2
1  x(x + 2) 
= log  + C,
2 (x + 1)2 
Z Z  2
dx 1 1 1
= − dx
(x2 − 1)2 4 x−1 x+1
Z  
1 1 1 2
= + − dx
4 (x − 1)2 (x + 1)2 (x + 1)(x − 1)
Z   
1 1 1 1 1
= + − − dx
4 (x − 1)2 (x + 1)2 x−1 x+1
   
1 1 1 x + 1
= − − + log 
  +C
4 x−1 x+1 x − 1
 
x 1 x + 1
=− + log    + C.
2(x2 − 1) 4 x − 1

Exercise 3.5.4. Compute the integrals.

x2 dx
Z
xdx
Z
(2x + 1)dx
Z
1. . 4. . 7. .
x2 − 1 x2 + 3x + 2 x2 + 1
1
x2 dx
Z
dx
Z
(2x + 1)dx
Z
2. . 5. . 8. .
x2 − 1 0 x2 + 3x + 2 (x2 + 1)(x2 + 4)

x2 dx x2 dx
Z Z Z
dx
3. . 6. . 9. .
2
x + 3x + 2 x2 + 3x + 2 (x2 + 1)(x2 + 4)

Exercise 3.5.5. Compute the integrals.

Z Z Z
dx xdx dx
1. . 2. . 3. .
(x + a)(x + b) (x + a)(x + b) (x + a)(x + b)(x + c)

Example 3.5.4. By trigonometric formula, we have

π π
1 π 1 π
Z Z Z Z
2 1
sin xdx = (1 − cos 2x)dx = xdx − cos 2xdx
0 0 2 2 0 2 0
1 1 1
= (π 2 − 02 ) − (sin 2π − sin 0) = π 2 .
4 4 4
3.5. PROPERTIES OF INTEGRATION 205

Similar idea gives

Z Z
2 1 1 1
sin xdx = (1 − cos 2x)dx = x − sin 2x + C,
2 2 4
Z Z
1 1 1
sin x cos 2xdx = (sin 3x − sin x)dx = − cos 3x + cos x + C,
2 6 2
Z Z
tan2 xdx = (sec2 x − 1)dx = tan x − x + C.

Exercise 3.5.6. Compute the integrals.

Z Z Z π

sin3 xdx.
2
1. cos x sin xdx. 4. 7. sin x cos 2xdx.
0
Z π Z Z π
2. sin2 x cos xdx. 5. cos x sin 2xdx. 8. | sin x cos 2x|dx.
0 0
Z Z Z π
2 2
3. cos xdx. 6. cot xdx. 9. | sin x − cos x|dx.
0

Example 3.5.5. For f ≥ g and a ≤ b, by the inequality (3.1.6) and the linearity of
definite integral, we have
Z b Z b Z b
f (x)dx − g(x)dx = (f (x) − g(x))dx ≥ 0.
a a a

Therefore we have
Z b Z b
f ≥ g =⇒ f (x)dx ≥ g(x)dx, for a < b.
a a

The inequality corresponds to Theorem 2.3.3 that uses the derivatives to compare
functions. However, it is more direct to get the inequality by using the non-negativity
of area.
If we apply the inequality to −|f | ≤ f ≤ |f |, then we get
Z b Z b 
 
|f (x)|dx ≥  f (x)dx , for a < b.
a a

Z b
1
Example 3.5.6 (Average). The average of a function f on [a, b] is f (x)dx.
b−a a
If m ≤ f ≤ M on [a, b], then
Z b Z b Z b
m(b − a) = mdx ≤ f (x)dx ≤ M dx = M (b − a).
a a a

This implies that the average of f lies between m and M , which is consistent with
our intuition.
206 CHAPTER 3. INTEGRATION

For continuous f , we may take m and M to be the minimum and maximum of

f on [a, b]. By the Intermediate Value Theorem, any value between m and M can
be reached by the function. Therefore the average
Z b
1
f (x)dx = f (c), for some c ∈ (a, b).
b−a a
This conclusion is the Integral Mean Value Theorem.
x
sin t2
Z
Example 3.5.7. Consider the function F (x) = dt. The 4-th order Taylor
0 t
x3
expansion T (x) = x − of sin x means that, for any  > 0, there is δ > 0, such
6
that
|x| < δ =⇒ | sin x − T (x)| ≤ |x|4 .
Then for t between 0 and x, we have
√
|x| < δ =⇒ |t2 | < δ =⇒ | sin t2 − T (t2 )| ≤ |t|8 ,

so that
√  sin t2 5  sin t2 1
   
t 
|x| < δ =⇒  − t +  =  − T (t ) ≤ |t|7 .
2
t 6 t t
Therefore
√
 2
 Z x 
6 2 5
  Z x
 x x  sin t t  
|x| < δ =⇒ F (x) −
 + =
  −t+ dt ≤ 
 |t|7 dt = |x|8 .
2 36 0 t 6 0 8

This means exactly the 7-th order approximation of F (x)

1 1
F (x) = x2 − x6 + o(x8 ).
2 36

The same argument in this Example can be generalized to show that

Theorem 3.5.1. If f (x) = o(xn ) is integrable for some n ≥ 0, then

Z x
F (x) = f (t)dt = o(xn+1 ).
0

Example 3.5.8. Suppose f (x) has second order derivative on [a, b]. We may take the
a+b
linear approximation at the middle point c = . By the Lagrange form of the
2
remainder (Theorem 2.7.1), we get

f 00 (x̄)
f (x) = f (c) + f 0 (c)(x − c) + (x − c)2 ,
2
3.5. PROPERTIES OF INTEGRATION 207

where x̄ depends on x and lies between x and c. Then we have

Z b  Z b 00 
 (f (x) − f (c) − f 0 (c)(x − c))dx = 
   f (x̄) 2

(x − c) dx 

a
 
a 2 

By the linear property of the integral, we have the left side

Z b Z b Z b Z b
0 0
(f (x) − f (c) − f (c)(x − c))dx = f (x)dx − f (c) dx − f (c) (x − c)dx
a a a a
Z b
= f (x)dx − f (c)(b − a).
a

Let K2 be the bound for the second order derivative. In other words, |f 00 | ≤ K2 on
[a, b]. Then the right side
Z b 00  Z b 00
|f (x̄)| K2 b
Z
 f (x̄) 2
 2 K2
 (x − c) dx ≤
 (x − c) dx ≤ (x − c)2 dx = (b − a)3 .

a 2 a 2 2 a 24

We conclude the inequality

Z b 
  K2
 f (x)dx − f (c)(b − a)≤ (b − a)3 .

a
 24

Exercise 3.5.7. Show that the integration of n-th order approximation is (n + 1)-st order
approximation. Specifically, find high order approximation of function at 0.
Z x
cos t − 1
Z 0 t
e −1
1. dt, order 5. 3. dt, order 5.
0 t −x2 t
Z √x Z x
sin t − t log(1 + t)
2. 2
dt, order 4. 4. dt, order 7.
0 t −x t
Z b 
 
Exercise 3.5.8. Derive an estimation for 
 f (x)dx − f (a)(b − a) in terms of the bound
a
K1 of f on [a, b].

Exercise 3.5.9. Apply the estimation in Example 3.5.8 to each interval of a partition and
derive the error formula for the midpoint rule in Theorem 3.3.1.

3.5.2 Integration by Parts

The Leibniz rule says that, if F (x) and G(x) are antiderivatives of f (x) and g(x),
then
(F (x)G(x))0 = F 0 (x)G(x) + F (x)G0 (x) = f (x)G(x) + F (x)g(x).
208 CHAPTER 3. INTEGRATION

In other words, F (x)G(x) is an antiderivative of f (x)G(x) + F (x)g(x), or

Z Z
F (x)G(x) + C = f (x)G(x)dx + F (x)g(x)dx.

If we use the differential notation

dF (x) = F 0 (x)dx = f (x)dx, dG(x) = G0 (x)dx = g(x)dx,

then the equality becomes

Z Z
F (x)dG(x) = F (x)G(x) − G(x)dF (x).
Z
The equality can be used in the following way. To compute an integral h(x)dx,
we separate the integrand
Z into a product h(x) = F (x)g(x)
Z of two partsZand integrate
the second part to get g(x)dx = G(x) + C. Then h(x)dx = F (x)dG(x),
which
Z by the equality above is converted into the computation of another integral
G(x)dF (x) that exchanges F (x) and G(x). This method of computing the integral
is called the integration by parts.
By Newton-Leibniz formula, the integration by parts for indefinite integral im-
plies the method for definite integral
Z b Z b
F (x)dG(x) = F (b)G(b) − F (a)G(a) − G(x)dF (x).
a a

The use of Newton-Leibniz formula requires that f = F 0 and g = G0 to be con-

tinuous. Then it is not hard to extend the equality to the case that F and G are
continuous on [a, b] and have continuous derivatives at all but finitely many points
on [a, b].

Example 3.5.9. The antiderivative of the logarithmic function in Example 3.4.7 may
be derived by using the integration by parts (taking F (x) = log |x| and G(x) = x)
Z Z Z
log |x|dx = x log |x| − xd log |x| = x log |x| − x(log |x|)0 dx
Z
= x log |x| − dx = x log |x| − x + C.

The antiderivative x log |x| − x just obtained can be further used

Z Z Z
x log |x|dx = xd(x log |x| − x) = x(x log |x| − x) − (x log |x| − x)dx
Z
1
= x log x − x − x log |x|dx + x2 .
2 2
2
3.5. PROPERTIES OF INTEGRATION 209

Solving the equation, we get

Z
1 1
x log |x|dx = x2 log x − x2 + C.
2 4
The following is an alternative way of applying the integration by parts to the same
integral
Z Z Z
1 2 1 2 1
x log |x|dx = log |x|d(x ) = x − x2 d(log |x|)
2 2 2
Z
1 2 1 1 1 1
= x log |x| − x2 dx = x2 log |x| − x2 + C.
2 2 x 2 4
Z
We may compute xp log xdx by the similar idea.

Example 3.5.10. The integral in Example 3.5.1 can also be computed by using the
integration by parts
Z Z
10 1
x(x + 1) dx = xd(x + 1)11 (integrate (x + 1)10 part)
11
Z
1 11 1
= x(x + 1) − (x + 1)11 dx (exchange two parts)
11 11
1 1
= x(x + 1)11 − (x + 1)12 + C.
11 12 · 11

Example 3.5.11. Using integration by parts, we have

Z Z Z
x e dx = − x de = −x e + 2 xe−x dx
2 −x 2 −x 2 −x

Z Z
= −x e − 2 xde = −x e − 2xe + 2 e−x dx
2 −x −x 2 −x −x

= −(x2 + 2x + 2)e−x + C.

In general, we have the recursive formula

Z Z
n x 1 n x n
x a dx = x a − xn−1 ax dx.
log a log a

Exercise 3.5.10. Compute the integral.

Z Z Z
p
1. x(ax + b) dx. 2. x2 (ax + b)p dx. 3. (x − 1)(x + 1)p dx.

Exercise 3.5.11. Compute the integral.

210 CHAPTER 3. INTEGRATION

xex dx
Z Z Z
2 x x 2
1. (x − 1)a dx. 2. (x + a ) dx. 3. .
(x + 1)2

Exercise 3.5.12. Compute the integral.

Z Z Z
2
1. x log |x|dx. 3. (log |x|)2 dx. 5. x log(x + 1)dx.
Z Z Z
p p 2 1+x
2. x log xdx. 4. x (log |x|) dx. dx. 6. x log
1−x
Z Z
Exercise 3.5.13. Derive the recursive formula for (log |x|)n dx. How about xp (log x)n dx?

Z 1
Exercise 3.5.14. Compute xn ax dx.
0

Z 1
m!n!
Exercise 3.5.15. For natural numbers m, n, show that xm (1 − x)n dx = .
0 (m + n + 1)!

Exercise 3.5.16. Compute the integral.

√ √
Z Z
x p
1. log( x + a + x − a)dx. 4. √ log(x + 1 + x2 )dx.
1 + x2
√
√ √
Z
x log(x + 1 + x2 )
Z
2. log( a + x − a − x)dx. 5. √ dx.
1 + x2
Z  
log(x + a) log(x + b)
Z  p 2
3. + dx. 6. log(x + 1 + x2 ) dx.
x+b x+a

Exercise 3.5.17. Let f have second order derivative on [a, b].

1. Show that for any constants A and B, we have
Z b  b
1 2 0
f (x)dx = (x + A)f (x) − ((x + A) + B)f (x)
a 2 a
Z b
1
+ ((x + A)2 + B)f 00 (x)dx.
2 a

2. By choosing suitable A and B in the first part, show that

Z b
1 b
Z
f (a) + f (b)
f (x)dx = (b − a) + ((x + A)2 + B)f 00 (x)dx,
a 2 2 a
and Z b
1
|(x + A)2 + 2B|dx = (b − a)3 .
a 12
3. Use the second part to derive the error formula for the trapezoidal rule in Theorem
3.3.1.
3.5. PROPERTIES OF INTEGRATION 211

Example 3.5.12. Using the integration by parts, we have

Z Z Z
x cos xdx = xd sin x = x sin x − sin xdx = x sin x + cos x + C.

The idea can be extended to product of xn , sin ax and cos bx for various a and b
Z Z Z  
1 1 1
x sin x sin 2xdx = x(cos 3x − cos x)dx = xd sin 3x − sin x
2 2 3
Z
1 1
= x(sin 3x − 3 sin x) − (sin 3x − 3 sin x)dx
6 6
1 1 1 1
= x sin 3x − x sin x + cos 3x − cos x + C.
6 2 18 2
An example of the definite integral is
Z π Z π  π 2 Z π
2
2
2
2 π 2
2
x sin xdx = − x d cos x = − cos + 0 cos 0 + 2x cos xdx
0 0 2 2 0
Z π Z π
2 π π 2
=2 xd sin x = 2 sin − 2 · 0 sin 0 − 2 sin xdx
0 2 2 0
π
= π + 2 cos − 2 cos 0 = π − 2.
2

Example 3.5.13. Let

Z Z
ax
I0 = e cos bxdx, J0 = eax sin bxdx.

We have
Z Z
−1 ax −1 ax −1
I0 = a cos bxde = a e cos bx − a eax d cos bx

= a−1 eax cos bx + a−1 bJ0 ,

Z Z
−1 ax −1 ax −1
J0 = a sin bxde = a e sin bx − a eax d sin bx

= a−1 eax sin bx − a−1 bI0 .

Solving the system for I0 and J0 , we get

eax
Z
eax cos bxdx = 2 (a cos bx + b sin bx) + C,
a + b2
eax
Z
eax sin bxdx = 2 (−b cos bx + a sin bx) + C.
a + b2
Let Z Z
x
I1 = xe cos xdx, J1 = xex sin xdx.
212 CHAPTER 3. INTEGRATION

Using the earlier computation of I0 and J0 , we have

Z Z
I1 = x cos xde = xe cos x − (cos x − x sin x)ex dx
x x

1
= xex cos x − ex (cos x + sin x) + J1 .
2
Similarly, we have

1
J1 = xex sin x − ex (− cos x + sin x) − I1 .
2
Solving the two equations, we get
Z
1
xex cos xdx = xex (cos x + sin x) − ex sin x + C,
2
Z
1
xex sin xdx = xex (− cos x + sin x) + ex cos x + C.
2

Example 3.5.14. Let Z

Im,n = cosm x sinn xdx.

If n 6= 0, then we may integrate a copy of sin x to get

Z
Im,n = − cosm x sinn−1 xd cos x

= − cosm+1 x sinn−1 x
Z
+ (−m cosm−1 x sinn x + (n − 1) cosm+1 x sinn−2 x) cos xdx

= − cosm+1 x sinn−1 x
Z
+ (−m cosm x sinn x + (n − 1) cosm x(1 − sin2 x) sinn−2 x)dx

= − cosm+1 x sinn−1 x − (m + n − 1)Im,n − (n − 1)Im,n−2 .

Therefore (the formula can be directly verified for n = 0)

1 n−1
Im,n = − cosm+1 x sinn−1 x + Im,n−2 , m + n 6= 0.
m+n m+n

The formula reduces the power of sine by 2. If we first integrate a copy of cos x,
then we get another recursive relation that reduces the power of cosine by 2

1 m−1
Im,n = cosm−1 x sinn+1 x + Im−2,n , m + n 6= 0.
m+n m+n
3.5. PROPERTIES OF INTEGRATION 213

On the other hand, we can also express Im,n−2 and Im−2,n in terms of Im,n . After
substituting n by n + 2, we get recursive relations that increase the power by 2
1 m+n+2
Im,n = cosm+1 x sinn+1 x + Im,n+2 , n 6= −1,
n+1 n+1
1 m+n+2
=− cosm+1 x sinn+1 x + Im+2,n , m 6= −1.
m+1 m+1
Here is a concrete example of using the recursive relation
6−1
Z
1
cos4 x sin6 xdx = I4,6 = − cos4+1 x sin6−1 x + I4,6−2
4+6 4+6
1 5
= − cos5 x sin5 x + I4,4
10 10  
1 5 5 5 1 5 3 3
= − cos x sin x + − cos x sin x + I4,2
10 10 8 8
 
1 5
= − cos5 x sin5 x + sin3 x
10 10 · 8
 
5·3 1 5 1
+ − cos x sin x + I4,0
10 · 8 6 6
 
5 1 5 5 3 5·3
= − cos x sin x + sin x + sin x
10 10 · 8 10 · 8 · 6
 
5·3·1 1 3 3
+ cos x sin x + I2,0
10 · 8 · 6 4 4
 
5 1 5 5 3 5·3
= − cos x sin x + sin x + sin x
10 10 · 8 10 · 8 · 6
 
5·3·1 1 3 5·3·1 3 1 1
+ cos x sin x + + cos x sin x + I0,0
10 · 8 · 6 4 10 · 8 · 6 4 2 2
 
1 5 5·3
= − cos5 x sin5 x + sin3 x + sin x
10 10 · 8 10 · 8 · 6
 
5·3·1 1 3 3·1 5·3·1 3·1
+ cos x + cos x sin x + + x + C.
10 · 8 · 6 4 4·2 10 · 8 · 6 4 · 2
Here is another example that requires increasing the power
sin2 x −4 + 2 + 2
Z
1
4
dx = I−4,2 = − cos−4+1 x sin2+1 x + I−4+2,2
cos x −4 + 1 −4 + 1
1 sin3 x
= + C.
3 cos3 x
Applying the recursive relation to the definite integral, we have
Z π π Z π Z π
2
n 1 n−1 
2 n − 1 2 n−2 n−1 2
sin xdx = − cos x sin x + sin xdx = sinn−2 xdx.
0 n 0 n 0 n 0
214 CHAPTER 3. INTEGRATION

Then we get

Z π
2 n−1n−3
Z π
2
 (n − 1)!! π , if n is even,

sinn xdx = ··· sin0 or 1
xdx = (n −n!! 2
0 n n−2 0
1)!!
, if n is odd.


n!!
Here sin0 or 1 x takes power 0 for even n and takes power 1 for odd n. Moreover, we
used the double factorial
(
2k(2k − 2)(2k − 4) · · · 4 · 2, if n = 2k,
n!! = n(n − 2)(n − 4) · · · =
(2k + 1)(2k − 1)(2k − 3) · · · 3 · 1, if n = 2k + 1.
Z Z
Exercise 3.5.18. Find the recursive relations for xp cos axdx and xp sin axdx. Then
compute the integral.
Z Z Z π
1. x cos2 xdx. 3. x cos2 x sin 2xdx. 5. x6 cos xdx.
0
Z Z Z π
2
2. 3 2
x cos xdx. 4. 3 2
x cos x sin 2xdx. 6. x5 sin 2xdx.
0
Z Z
p ax
Exercise 3.5.19. Find the recursive relations for x e cos bxdx and xp eax sin bxdx.
Then compute the integral.
Z Z Z
1. x2 e−x sin 3xdx. 2. 2 x
x 2 cos xdx. 3. x3 ex cos2 xdx.

Exercise 3.5.20. Compute the integral.

Z Z Z
6 dx
1. sin xdx. 4. cos3 x sin2 xdx. 7. .
cos6 x
Z Z Z
dx
2. cos8 xdx. 5. cos3 x sin5 xdx. 8. .
2
sin x cos2 x
Z Z
3. cos8 x sin6 xdx. 6. cos−2 x sin2 xdx.

Z π
2 (2m)!(2n)!
Exercise 3.5.21. Show that sin2m x cos2n xdx = for natural
22m+2n+1 m!n!(m + n)!
Z0 π
2
numbers m, n. Can you find sinm x cosn xdx?
0

Exercise 3.5.22. Use (tan x)0 = sec2 x = tan2 x + 1 to derive the recursive formula for
Z Z π
4
secm x tann xdx similar to Example 3.5.14 and then find the value of tan2n xdx.
0
3.5. PROPERTIES OF INTEGRATION 215

Example 3.5.15. Let

Z
Ip = (ax2 + bx + c)p dx, a 6= 0, b2 6= 4ac.

We have
Z
2 p
Ip = x(ax + bx + c) − xd(ax2 + bx + c)p
Z
2 p
= x(ax + bx + c) − px(2ax + b)(ax2 + bx + c)p−1 dx.

We try to express px(2ax + b) as a combination of (ax2 + bx + c) and (ax2 + bx + c)0 ,

up to adding a constant

2pax2 + pbx = A(ax2 + bx + c) + B(ax2 + bx + c)0 + C.

pb p(b2 − 4ac)
We get A = 2p, B = − ,C= . Then
2a 2a
Z
Ip = x(ax + bx + c) − A (ax2 + bx + c)p dx
2 p

Z Z
0
− B (ax + bx + c) (ax + bx + c) dx − C (ax2 + bx + c)p−1 dx
2 p−1 2

B
= x(ax2 + bx + c)p − AIp − (ax2 + bx + c)p − CIp−1 .
p
This gives us the recursive relation
1 2 p p(b2 − 4ac) 1
Ip = (2ax + b)(ax + bx + c) − Ip−1 , p 6= − .
(2p + 1)2a (2p + 1)2a 2
On the other hand, we may also express Ip−1 in terms of Ip . After substituting p by
p + 1, we get
1 2 p+1 (2p + 3)2a
Ip = (2ax + b)(ax + bx + c) − Ip+1 , p 6= −1.
(p + 1)(b2 − 4ac) (p + 1)(b2 − 4ac)
For the special case
Z
Ip = (ax2 + b)p dx, a, b 6= 0,

the recursive relations become

1 2pb 1
Ip = x(ax2 + b)p + Ip−1 , p 6= − ;
2p + 1 2p + 1 2
1 2p + 3
Ip = − x(ax2 + b)p+1 + Ip+1 , p 6= −1.
2(p + 1)b 2(p + 1)b
216 CHAPTER 3. INTEGRATION

1
For the special cases of p = − , −1, Ip is given by Exercise 3.4.10 (and will be
2
derived in Examples 3.5.30, 3.5.31, 3.5.32)
Z
dx x
√ = arcsin + C, a > 0,
2 2 a
Z a −x √
dx  
√ 2
= log x + x + a + C,
 
2
x +a  
x − a
Z
dx 1
= log 
  + C,
x 2 − a2 2a x + a
Z
dx 1 x
2 2
= arctan + C.
x +a a a
Then the recursive relations can be used to compute Ip when p is an integer or a
half integer. For example, we have
Z √
2 2
1 2 2 12 2 · 12 a2
a − x dx = I 1 = x(a − x ) + I1
2 2 · 12 + 1 2 · 12 + 1 2 −1
1 √ a2 x
= x a2 − x2 + arcsin + C,
2 2 a
2 · 32 a2
Z
3 1 2 32
(a2 − x2 ) 2 dx = I 3 = x(a 2
− x ) + I3
2 2 · 32 + 1 2 · 32 + 1 2 −1
3a2 1 √ 2 2
 
1 2 2 32 a x
= x(a − x ) + x a − x2 + arcsin +C
4 4 2 2 a
1 √ 3a4 x
= − x(2x2 − 5a2 ) a2 − x2 + arcsin + C,
Z 8 8 a
dx 1 2(−2) + 3
2 2 2
= I−2 = − 2
x(x2 + a2 )−2+1 + I−2+1
(x + a ) 2(−2 + 1)a 2(−2 + 1)a2
x 1 x
= 2 2 2
+ 3 arctan + C.
2a (x + a ) 2a a

Exercise 3.5.23. Compute the integral.

Z
x2 dx
Z
dx
p Z
2 2 2
1. x a − x dx. 4. √ . 7. 5 .
a2 − x2 (a2 − x2 ) 2
Z
Z p dx Z
dx
2. 2
x a − x dx. 2 5. 3 . 8. .
(a − x2 ) 2
2 (a2 − x2 )3
Z
Z p xdx Z
xdx
3. 2
(x+b) a2 − x2 dx. 6. 3 . 9. .
(a − x2 ) 2
2 (a2 − x2 )3

Exercise 3.5.24. Compute the integral.

3.5. PROPERTIES OF INTEGRATION 217
Z Z Z
p
2 xdx dx
1. x x2 + adx. 4. √ . 7. .
x2 + a (x2 + a2 )3
Z Z Z
p
2 3 xdx
2. x x2 + adx. 5. (x + a) dx. 2 8. .
(x2 + a2 )3
Z
Z
x2 dx dx Z
x2 dx
3. √ . 6. 3 . 9. .
x2 + a (x2 + a) 2 (x2 + a2 )3

Exercise 3.5.25. Combine the ideas of Exercise 3.4.11 and Example 3.5.15 to compute the
integral.
Z Z Z p
dx dx
1. . 3. 3 . 5. x(1 − x)dx.
(x2 + 2x + 2)2 (x2 + 2x + 2) 2
Z
Z
dx
Z
3
dx
2. √ . 4. (x 2
+ 2x + 2) 2 dx. 6. 3 .
x2 + 2x + 2 (x(1 − x)) 2

3.5.3 Change of Variable

Z
The chain rule says that, if f (y)dy = F (y) + C is the indefinite integral of f (y),
and φ(x) is a differentiable function, then

F (φ(x))0 = F 0 (φ(x))φ0 (x) = f (φ(x))φ0 (x).

In other words, F (φ(x)) is the antiderivative of f (φ(x))φ0 (x), or

Z Z 
0

f (φ(x))φ (x)dx = F (φ(x)) + C = f (y)dy  .
y=φ(x)

If we use the differential notation dφ(x) = φ0 (x)dx, then the equality becomes
Z Z 

f (φ(x))dφ(x) = f (y)dy  .
y=φ(x)

The right side means computing the antiderivative of the function of y first, and
then substituting y = φ(x) into the antiderivative. This is the change of variable
formula. By Newton-Leibniz formula, we further get the change of variable formula
for definite integral
Z b Z b Z φ(b)
0
f (φ(x))φ (x)dx = f (φ(x))dφ(x) = f (y)dy.
a a φ(a)

Z
Example 3.5.16. If f (y)dy = F (y) + C, then by letting y = ax + b, we have
Z Z
1 1
f (ax + b)dx = f (ax + b)d(ax + b) = F (ax + b) + C.
a a
218 CHAPTER 3. INTEGRATION

For example,

y p+1 (2x + 1)p+1

Z Z
p 1
(2x + 1) dx =y=2x+1 y p dy = +C = + C,
2 2(p + 1) 2(p + 1)
Z Z
1 1 1
sin(3x − 2)dx =y=3x−2 sin ydy = − cos y + C = − cos(3x − 2) + C,
3 3 3
Z Z Z
dx dx dx
2
= 2
=y=x+1 2
x + 2x + 2 (x + 1) + 1 y +1
= arctan y + C = arctan(x + 1) + C.

Example 3.5.17. The following is a simple change of variable

Z Z
x2 1 1 1 2
xe dx =y=x2 ey dy = ey + C = ex + C.
2 2 2

The idea is a “mini-integration” of xdx that can be expressed more clearly by writing
Z Z
x2 2 1 1 2
xe dx = ex d(x2 ) = ex + C.
2 2

After the mini-integration, we view x2 as the new variable.

The following are more examples following the mini-integration idea
Z Z   Z
dx 1 dx 1
= d(log x) = log | log x| + C,
=
x log x log x x log x
d xa
Z Z Z

dx d 1 x
=   =   = arctan + C,
x 2 + a2 a2 xa + 1

2
a xa + 1

2 a a

d(x2 ) x2
Z Z
xdx 1 1
= = arctan + C.
x 4 + a4 2 (x2 )2 + a4 2a2 a2

Example 3.5.18. The integral in Example 3.5.1 was computed in Example 3.5.10
again by using the integration by parts. The integral can also be computed by
change of variable.
Let y = x + 1. Then
Z Z Z
10 10 1 11 1
x(x + 1) dx = (y − 1)y dy = (y 10 − y 11 )dy = y − y 12 + C.
11 12

Substituting y = x + 1 back, we get

Z
1 1
x(x + 1)10 dx = (x + 1)11 − (x + 1)12 + C.
11 12
3.5. PROPERTIES OF INTEGRATION 219

By the same change, we have

4
(y − 2)4
Z 
x−1
Z
dx =y=x+1 dy
x+1 y4
Z
= (1 − 4 · 2y −1 + 6 · 22 y −2 − 4 · 23 y −3 + 24 y −4 )dy

= x + 1 − 8 log |x + 1|
16
− 24(x + 1)−1 + 16(x + 1)−2 − (x + 1)−3 + C
3
8(9x2 + 12x + 5)
=x− − 8 log |x + 1| + C.
3(x + 1)3
Note that the second C is the first C plus 1.
Compare the above with the computation of definite integral
Z 1 4 Z 2
x−1 (y − 2)4
dx =y=x+1 dy
0 x+1 1 y4
Z 2
= (1 − 4 · 2y −1 + 6 · 22 y −2 − 4 · 23 y −3 + 24 y −4 )dy
1
 2
−1 −2 16 −3
= y − 8 log y − 24y + 16y − y
3
   1  
1 1 16 1
= 1 − 8 log 2 − 24 − 1 + 16 −1 − −1
2 4 3 8
17
= − 8 log 2.
3
Note that the evaluation is done by using the new variable y instead of the old x.

Example 3.5.19. The integrals of inverse trigonometric functions can also be com-
puted by combining integration by parts and change of variable
d(1 − x2 )
Z Z Z
xdx 1
arcsin xdx = x arcsin x − √ = x arcsin x + √
1 − x2 2 1 − x2
√
= x arcsin x + 1 − x2 + C.
Alternatively, we may simply introduce the trigonometric function as the new vari-
able. For example, by y = arcsin x, x = sin y, we have
Z Z Z
arcsin xdx = yd(sin y) = y sin y − sin ydy
√
= y sin y + cos y + C = x arcsin x + 1 − x2 + C.
√
Note that cos y = 1 − x2 is non-negative because x ∈ [−1, 1] and y ∈ [− π2 , π2 ]. The
integration by parts used in both computations are essentially the same.
220 CHAPTER 3. INTEGRATION
Z
The idea for integrating arcsin x can also be used to compute xm (arcsin x)n dx
Z
and xm (arctan x)n dx.

Z
dx
Example 3.5.20. To compute √ , we introduce
ex + a
√
y= ex + a, y 2 = ex + a, 2ydy = ex dx.

Then
Z Z 2y dy Z
dx ex 2dy
√ x = = 2
.
e +a y y −a
By Examples 3.4.9 and 3.5.2, we have
  √  √ x √ 
1  y − −a  1  e + a − a
 √ log  √  + C = √ log  √ x √  + C, if a > 0,
Z 

dx a y + −a a e r x a
+ a +
√ x =
e +a  2 y 2 e
√ arctan √ +C = √ arctan − − 1 + C, if a < 0.

−a −a −a a

Example 3.5.21. To compute

Z π
x sin x
I= dx,
0 1 + cos2 x
we introduce y = π − x. Then
Z 0 Z π Z π
(π − y) sin y sin y y sin y
I=− 2
dy = π 2
dy − 2
dy
π 1 + cos y 0 1 + cos y 0 1 + cos y
Z π
sin y
=π 2
dy − I.
0 1 + cos y

Therefore
π π sin y π cos π 1
Z Z
I= dy = − dz
2 0 1 + cos2 y 2 cos 0 1 + z 2
π 1 dz π2
Z
π
= = (arctan 1 − arctan(−1)) = .
2 −1 1 + z 2 2 4

Note that the computation of the definite integral makes use of the new variable z
only. There is no need to go back to the original variable x.

Exercise 3.5.26. Compute the integral.

3.5. PROPERTIES OF INTEGRATION 221
Z Z
2 10
1. (x + 1)(2x − 1) dx. 4. cos(2x − 1)dx.
Z Z
2. log(2x − 1)dx. 5. e3x cos(2x − 1)dx.
Z Z
3. x log(2x − 1)dx. 6. sin(2x + 1) cos(2x − 1)dx.

Exercise 3.5.27. Compute the integral.

Z Z Z
x x p
1. dx. 3. dx. 5. x3 x2 + 1dx.
x2 + 1 (x2 + 1)p
x3 dx
Z Z Z
bx + c
2. dx. 4. x(x2 + a2 )p dx. 6. √
3
.
x2 + a2 x2 + a2

Exercise 3.5.28. Compute the integral.

Z Z Z
1 p
1. sin x sin(cos x)dx. 5. tan(log x)dx. 9. sin 2x a + cos2 xdx.
x
√
Z
cos xdx
Z Z
3
2. x sin(1 + x 2 )dx. 6. sin(log x)dx. 10. √ .
a + cos2 x
Z
sin x cos xdx
Z Z
sin xdx
3. cot xdx. 7. . 11. p .
a + cos2 x a sin2 x + b2 cos2 x
2
Z Z Z
2 sin 2xdx sin x cos xdx
4. x tan x dx. 8. . 12. .
a + cos2 x sin4 x + cos4 x

Exercise 3.5.29. Compute the integral.

Z Z Z
dx
1. arccos xdx. 5. √ . 9. (arccos x)2 dx.
1 − x2 arccos x
Z Z
arcsin xdx x3 arccos x
Z
2. √ . 6. √ dx. 10. x arctan xdx.
1 − x2 1 − x2
(arctan x)2
Z p Z Z
3. 1 − x2 arcsin xdx. 2 11. dx.
7. x arccos xdx.
1 + x2
Z
dx
Z Z
p arcsin x
4. x 1 − x2 arcsin xdx. 8. dx. 12. .
x2 (1 + x2 ) arctan x

Exercise 3.5.30. Compute the integral.

Z Z −√x Z
1. 3 x2
x e dx. e 5. ecos x sin xdx.
3. √ dx.
x

√ √ √ ex dx
Z Z Z
x x
2. e dx. 4. xe dx. 6. .
1 + ex
222 CHAPTER 3. INTEGRATION

√
Z Z Z
dx dx
7. . 8. −x
. 9. ex + adx.
1 + ex x
e +e
Z Z
5
Exercise 3.5.31. Find a recursive relation for (e + a) dx. Then compute (ex + a) 2 dx
x p
Z
dx
and .
(e + a)3
x

Exercise 3.5.32. Compute the integral.

Z Z
dx 1 x+1
1. √ . 3. log dx.
x 1 + log x x2 −1 x−1
log(x + 1) − log x
Z Z
dx
2. . 4. dx.
x log x log(log x) x(x + 1)

Exercise 3.5.33. Compute the integral.

Z 0
f 0 (x)
Z Z
f (x)
1. dx. 2. dx. 3. 2f (x) f 0 (x)dx.
f (x)p 1 + f (x)2

Exercise 3.5.34. Prove the equalties

Z π Z π Z π Z π
2 2 π
1. f (sin x)dx = f (cos x)dx. 2. xf (sin x)dx = f (sin x)dx.
0 0 0 2 0

1
Exercise 3.5.35. Explain why we cannot use the change of variable y = to compute the
Z 1 x
dx
integral 2
.
−1 1 + x

Exercise 3.5.36. Suppose f is continuous on an open interval containing [a, b]. Find the
d b
Z
derivative f (x + t)dx.
dt a

Exercise 3.5.37. Explain the equalities in Exercise 3.1.11 by change of variable.

Z a Z a
Exercise 3.5.38. Prove that f (x)dx = 2 f (x)dx for even function f . Prove that
Z a −a 0

f (x)dx = 0 for odd function f .

−a

√
Z
dx
Example 3.5.22. To compute √ , we simply let y = x − 1. Then x =
1+ x−1
y 2 + 1, dx = 2ydy, and
Z Z Z  
dx 2ydy 1
√ = =2 1− dy
1+ x−1 1+y 1+y
√ √
= 2y − 2 log(1 + y) + C = 2 x − 1 − 2 log(1 + x − 1) + C.
3.5. PROPERTIES OF INTEGRATION 223

Example 3.5.23. By taking y = x6 , we get rid of the square root and cube root at
the same time.
6y 5 dy
Z Z Z 3
dx y dy
√ √ =x=y6 3 2
=6
x+ x3
y +y y+1
3 Z  
(y + 1) − 1dy
Z
2 1
=6 =6 y −y+1− dy
y+1 1+y
= 2y 3 − 3y 2 + 6y − 6 log(1 + y) + C
√ √ √ √
= 2 x − 3 3 x + 6 6 x − 6 log(1 + 6 x) + C.
Z
dx
Example 3.5.24. To compute √ √ , we introduce
x+1+ x+1
√ √
y = x + 1 + x.

Then
1 √ √ 1 √ 1 √
= x + 1 − x, y+ = 2 x + 1, y− = 2 x,
y y y
and  2  
1 1 1 1
x= y− , dx = y − 3 dy.
4 y 2 y
Therefore
 
1 1
Z
dx
Z
2
y− y31
Z dy
(y − 1)(y + 1)(y 2 + 1)
√ √ = = dy
x+1+ x+1 y+1 2 (y + 1)y 3
Z  
1 1 1 1
= 1 − + 2 − 3 dy
2 y y y
 
1 1 1
= y − log |y| − + 2 + C
2 y 2y
1 √ √ √ 1 √ √
= − log( x + 1 + x) + x + ( x + 1 + x)2 + C.
2 4

Example 3.5.25. The change of variable

√ in Example√ 3.5.24 can be used for in-
tegratingrother functions involving x + a and x + b. For example, to com-
Z
x+1
pute dx, which makes sense for x > 1 or x ≤ −1, we introduce y =
√ x−1
√
x + 1 + x − 1 for x > 1. Then
2 √ √ 2 √ 2 √
= x + 1 − x − 1, y+ = 2 x + 1, y− = 2 x − 1,
y y y
and  2  
1 2 1 4
x= y− + 1, dx = 1− 4 ydy.
4 y 2 y
224 CHAPTER 3. INTEGRATION

Therefore by x > 1, we have

Z r Z y+2   Z  
x+1 y 1 4 1 4 4
dx = 1 − 4 ydy = y + + 3 dy
x−1 y − y2 2 y 2 y y
  
1 2 1 1 2 2
= y + 2 log |y| − 2 + C = 2 log |y| + y+ y− +C
4 y 4 y y
1 √ √ √
= log( x + 1 + x − 1) + x2 − 1 + C.
4
√ √
For x ≤ −1, we may introduce y = −x + 1 − −x − 1. Then we have
Z r Z     
x+1 1 4 4 1 2 2
dx = y + + 3 dy = 2 log |y| + y+ y− +C
x−1 2 y y 4 y y
1 √ √ √
= log( −x − 1 − −x + 1) − x2 − 1 + C.
4
Z r
x
Example 3.5.26. The integral dx is comparable to the integral in Example
1−x
3.5.25. Yet the similar change of variable does not work, due to the requirement
0 ≤ x < 1. So we introduce

y2
r
x 2y
y= , x= 2
, dx = dy,
1−x 1+y (1 + y 2 )2

and get
Z r Z Z  
x 2y 1 1
dx = y dy = 2 − dx
1−x (1 + y 2 )2 1 + y 2 (1 + y 2 )2
r
y p x
=− + arctan y + C = − x(1 − x) + arctan + C.
1 + y2 1−x

The last computation in Example 3.5.15 is used here.

Note that theridea here can also be applied to the integral in Example 3.5.25, by
x+1
introducing y = . The advantage of the approach is that we do not need to
x−1
distinguish x > 1 and x ≤ 1.

Exercise 3.5.39. Compute the integrals in Example 3.5.24 and 3.5.25 by using change of
variable similar to Example 3.5.26.

Exercise 3.5.40. Compute the integral.

3.5. PROPERTIES OF INTEGRATION 225
Z
√
Z
√ √ Z √
1. (1 + p
x) dx. 3. ( x + x + 1)p dx. 1 + x2
5. dx.
x
√
√
Z Z
1+ x dx
Z
3 p 4. √ dx. 6. √ .
2. (1 + x) dx.
1+ 3x x 1 + x2

Example 3.5.27. We can use (cos x)0 = − sin x, (sin x)0 = cos x and cos2 x+sin2 x = 1
to calculate the antiderivative of cosm x sinn x, in which either m or n is odd.
Z Z Z
sin xdx = − sin xd cos x = (cos2 x − 1)d cos x
3 2

1
= cos3 x − cos x + C,
Z 3Z Z
4 5 4 4
cos x sin xdx = − cos x sin xd cos x = − cos4 x(1 − cos2 x)2 d cos x
Z
=− (cos4 x − 2 cos6 x + cos8 x)d cos x
1 2 1
= − cos5 x + cos7 x − cos9 x + C,
Z Z5 7 Z 9
sin x d cos x
tan xdx = dx = − = − log | cos x| + C,
cos x cos x
Z Z Z Z
dx d sin x d sin x
sec xdx = = =
cos x cos2 x 1 − sin2 x
1 1 + sin x 1 (1 + sin x)2 1 + sin x
= log + C = log 2 + C = log +C
2 1 − sin x 2 1 − sin x | cos x|
= log | sec x + tan x| + C.

Example 3.5.28. Similar to Example 3.5.27, we can also use (tan x)0 = sec2 x, (sec x)0 =
sec x tan x and sec2 x = 1 + tan2 x to calculate the antiderivative of secm x tann x.
Z Z Z
1
sec x tan xdx = tan xd sec x = (sec2 x − 1)d sec x = sec3 x − sec x + C,
3 2
3
Z Z
1
sec4 xdx = (tan2 x + 1)d tan x = tan3 x + tan x + C,
3
Z Z Z
tan4 xdx = (sec2 x − 1)2 dx = (sec4 x − 2 sec2 x + 1)dx
 
1 3
= tan x + tan x − 2 tan x + x + C
3
1
= tan3 x − tan x + x + C,
3
The following is computed in Example 3.5.14 by more complicated method
sin2 x
Z Z Z
1
4
dx = sec x tan xdx = tan2 xd tan x = tan3 x + C.
2 2
cos x 3
226 CHAPTER 3. INTEGRATION

Example 3.5.29. The method of Example 3.5.28 cannot be directly applied to the
antiderivatives of secn x and tann x for odd n. Instead, the idea of Example 3.5.27
can be used.
Using the integration by parts, we have
Z Z Z
3
sec xdx = sec xd tan x = sec x tan x − tan xd sec x
Z Z
= sec x tan x − tan x sec xdx = sec x tan x − (sec2 x − 1) sec xdx
2

Z Z
3
= sec x tan x − sec xdx + sec xdx.

Then with the help of Example 3.5.27, we get

Z Z
3 1 1
sec xdx = sec x tan x + sec xdx
2 2
1 1
= sec x tan x + log | sec x + tan x| + C.
2 2
Z
In fact, the integral sec3 xdx is I−3,0 in Example 3.5.14, and the expression above
Z
in terms of sec xdx is the expression of I−3,0 in terms of I−1,0 .
Example 3.5.27 also gives
Z Z Z Z
3 2
tan xdx = tan x(sec x − 1)dx = tan xd tan x − tan xdx
1
= tan2 x + log | cos x| + C.
2

Exercise 3.5.41. Compute the integral.

Z Z Z
dx dx
1. cos3 x sin2 xdx. 3. . 5. .
cos6 x sin x cos x
Z Z Z
dx
2. 3 5
cos x sin xdx. 4. 2 . 6. cos−3 x sin5 xdx.
sin x cos2 x

Exercise 3.5.42. Compute the integral.

Z Z Z
1. csc xdx. 4. tann x sec4 xdx. 7. tan3 x sec xdx.

Z Z Z
3 6 4
2. tan xdx. 5. cot x csc xdx. 8. tan5 x sec7 xdx.

Z Z Z
3. tan6 x sec4 xdx. 6. tan2 x sec xdx. 9. tan3 x cos2 xdx.
3.5. PROPERTIES OF INTEGRATION 227
Z Z Z
5 4 4 6
10. cot x sin xdx. 11. csc x cot xdx. 12. x tan x sec xdx.

Exercise 3.5.43. Compute the integral.

Z Z Z
3 −x2 2 dx dx
1. x e cos x dx. 6. . 11. 2 .
a sin x + b cos x a sin x + b cos2 x
Z Z
A sin x + B cos x
Z
1 + sin x
2. x sin(log x)dx. 7. dx. 12. dx.
1 + cos x a sin2 x + b cos2 x
Z
dx
Z Z
p 3 dx
3. (sin x) cos xdx. 8. . 13. √ .
1 + cos x 2 + sin x + cos x
Z Z
sin xdx
Z
cos 2xdx dx
4. . 9. . 14. √ .
sin2 x cos2 x a + tan x 2 + sin x + cos x
Z Z Z
a sin x + b cos x A sin x + B cos x xdx
5. dx. 10. dx. 15. .
sin 2x a sin x + b cos x cos2 x

Exercise 3.5.44. Show that there are constants An , Bn , Cn , such that

Z Z
dx An sin x + Bn cos x dx
n
= n−1
+ Cn .
(a sin x + b cos x) (a sin x + b cos x) (a sin x + b cos x)n−2

Exercise 3.5.45. For |a| =

6 |b|, show that there are constants An , Bn , Cn , such that
Z Z Z
dx A sin x dx dx
n
= n−1
+ B n−1
+ C .
(a + b cos x) (a + b cos x) (a + b cos x) (a + b cos x)n−2
Z
(A sin x + B cos x + C)dx
Exercise 3.5.46. How to calculate ?
(a sin x + b cos x + c)n
Z
dx
Exercise 3.5.47. Compute by using
cos(x + a) cos(x + b)
sin(a − b)
tan(x + a) − tan(x + b) = .
cos(x + a) cos(x + b)
Use the similar idea to compute the following integral.
Z Z
dx dx
1. . 3. .
sin(x + a) cos(x + b) sin x − sin a
Z Z
dx
2. tan(x + a) tan(x + b)dx. 4. .
cos x + cos a
√
Example 3.5.30. To integrate a function of a2 − x2 , with a > 0, we may introduce
x = a sin y, dx = a cos ydy. Note that the hfunction makes sense only for |x| ≤ a.
x π πi
Correspondingly, we take y = arcsin ∈ − , . This implies cos y ≥ 0, and
√ a 2 2
a2 − x2 = a cos y.
228 CHAPTER 3. INTEGRATION

The following example is also computed in Example 3.5.15.

Z √
a2
Z Z
2 2 2 2
a − x dx = a cos ydy = (1 + cos 2y)dy
2
a2
 
2 1 1
=a y + sin 2y + C = (y + sin y cos y) + C
2 4 2
a x 1 √
= arcsin + x a2 − x2 + C.
2 a 2
We may also use x = a cos y instead of x = a sin y.
sin y(1 − sin y)dy
Z Z Z
dx a sin ydy
√ = =
a+ a −x 2 2 a + a sin y (1 + sin y)(1 − sin y)
Z  2
 Z
sin y sin y
= − dy = (sec y tan y − sec2 y + 1)dy
cos2 y cos2 y
r
a a x2 x
= sec y − tan y + y + C = − 1 − 2 + arccos + C.
x x a a
The following is an example of definite integral.
Z 1 Z 0 Z π
2
2 p 2 p
(1 − x ) dx = − (1 − cos y) sin ydy = sin2p+1 ydy.
π
0 2
0

By Example 3.5.15, we know the specific value when 2p + 1 is a natural number.

Z 1 Z 1
2 n (2n)!! 1 (2n − 1)!! π
(1 − x ) dx = , (1 − x2 )n− 2 dx = .
0 (2n + 1)!! 0 (2n)!! 2
√
Example 3.5.31. To integrate a function of x2 + a2 , withh a > 0,iwe may introduce
x π π √
x = a tan y, dx = a sec2 ydy. We have y = arctan ∈ − , and x2 + a2 =
a 2 2
a sec y.
With the help of Example 3.5.27, we have
a sec2 ydy
Z Z Z
dx
√ = = sec ydy
x 2 + a2 a sec y
√
= log | sec y + tan y| + C = log( x2 + a2 + x) + C.

With the help of Example 3.5.29, we have

Z √
a2 a2
Z
x + a xdx = a2 sec3 ydy =
2 2 sec y tan y + log | sec y + tan y| + C
2 2
1 √ a 2 √
= x x2 + a2 + log( x2 + a2 + x) + C.
2 2
One may verify that the two integrals satisfy the recursive relation in Example
3.5.15.
3.5. PROPERTIES OF INTEGRATION 229
√
Example 3.5.32. To integrate a function of x2 − a2 , with a > 0, we may introduce
x a
x = a sec y, dx = a sec y tan ydy. We have y = arcsec = arccos ∈ [0, π] and
√ a x
x2 − a2 = ±a tan y, where the sign depends on whether x ≥ a or x ≤ −a.
With the help of Example 3.5.28, we have
Z Z Z
dx a sec y tan ydy
√ = = sec ydy
x 2 − a2 ±a tan y
 √ 
= ± log | sec y + tan y| + C = ± log x ± x2 − a2  + C
 
 √ 
2 2
= log x + x − a  + C,
 
Z √ Z Z
2
2 2
x − a dx = (±a tan y)a sec y tan ydy = ±a (sec3 y − sec y)dy
a2 a2
=± sec y tan y ∓ log | sec y + tan y| + C
2 2
1 √ 2 2
 √
2 2

= x x − a − log x + x − a  + C.
 
2
Combing with Example 3.5.31, for positive as well as negative a, we have
Z
dx  √ 
√ 2
= log x + x + a + C,
 
2
Z √ x +a
2
1 √ 2 a  √
2

x + adx = x x + b + log x + x + a + C.
 
2 2

Example 3.5.33. By completing the square, a quadratic function ax2 + bx + c can be

b 4ac − b2
changed to a(y 2 + d), where y = x + and d = . For example, if b2 < 4ac,
2a 4a2
then
Z Z
dx dy 1 y
2
= √ = √ arctan √ + C
ax + bx + c 2
a(y + ( d) )2 a d d
2 2ax + b
=√ arctan √ + C.
4ac − b2 4ac − b2
Using the recursive relation in Example 3.5.15, we further get
Z
dx 2ax + b 4a 2ax + b
2 2
= 2 2
− 3 arctan √ + C.
(ax + bx + c) (4ac − b )(ax + bx + c) (4ac − b ) 2
2 4ac − b2
If b2 ≥ 4ac, then by the similar idea, we may get
 √ 
 y − −d 
Z Z
dx dy 1
= √ = √ log  √ +C
ax2 + bx + c a(y 2 − ( −d)2 ) 2a −d y + −d 
 √ 
1  2ax + b − b2 − 4ac 
=√ arctan  √  + C.
b2 − 4ac 2ax + b + b2 − 4ac 
230 CHAPTER 3. INTEGRATION

In fact, the quadratic function has two real roots, and it is more direct to calculate
the integral by using
√
2 −b ± b2 − 4ac
ax + bx + c = a(x − x1 )(x − x2 ), x1 , x2 = ,
2a
and the idea of Examples 3.5.2 and 3.5.3.

Example 3.5.34. We further

√ use the idea of Example 3.5.33 to convert the antideriva-
2
tives of functions of ax + bx + c to the computations in Examples 3.5.30, 3.5.31
and 3.5.32. For example, by
   2
1 2 1 1 1 1 1
x(1 − x) = 2 − x − 2 x + 2 = 2 − x − = 2 (1 − (2x − 1)2 ),
2 2 2 2 2 2

we let y = 2x − 1 and get

Z Z
dx dy
p = p = arcsin y + C = arcsin(2x − 1) + C.
x(1 − x) 1 − y2

Moreover, for 0 ≤ x < 1, we get

Z y+1
d(1 − y 2 )
Z r
dy
Z Z Z
x xdx 2 dy
dx = p = p =− p + p
1−x x(1 − x) 1−y 2 4 1−y 2 2 1 − y2
1p 1 1p 1
= 1 − y 2 + arcsin y + C = x(1 − x) + arcsin(2x − 1) + C.
4 2 2 2
The reader is left to verify that the result is the same as the one in Example 3.5.26.

Exercise 3.5.48. Compute the integral.

Z Z Z
xdx 3 dx
1. √ . 6. (x2 + a2 ) 2 dx. 11. √ .
1−x 2 1 − 1 − x2
(ax2 + bx + c)dx
Z
dx
Z Z
7. dx
2. √ . 3 . 12. √ √ .
1−x 2 x(x + a2 ) 2
2 1+ x2 + 1 − x2
Z
dx Z
x3 dx
Z
xdx
3. √ . 8. 13. √ √ .
x 1−x 2 3 .
(x2 + a2 ) 2 1 + x2 + 1 − x2
x2 dx
Z Z
4.
Z dx
3 .
3
2 2 9. (x(x + 1)) 2 dx. 14. √ .
(a − x ) 2 a + x2 + a2
Z
dx Z p Z
xdx
5. 3 . 10. x 2
1 − x2 dx. 15. √ .
x(a2 − x2 ) 2 a + x2 + a2

Exercise 3.5.49. Compute the integral.

3.6. INTEGRATION OF RATIONAL FUNCTION 231
Z Z
log(sin x)
1. sin x log tan xdx. 3. dx.
sin2 x
 
√
Z
1
Z
sin x
2. arctan xdx. 4. e cos2 x + dx.
cos2 x

Exercise 3.5.50. Derive the formula

2ax + b √ 2
  
1
 √ log √ + ax + bx + c + C, if a > 0,
Z 

dx a a
√ =
ax2 + bx + c  1 −2ax − b
√
 arcsin √ + C, if a < 0.
−a b2 − 4ac

1
Exercise 3.5.51. Use the change of variable y = x ± to compute the integral.
x
x2 + 1
Z
dx
Z
1. dx. 3. 4
dx.
x4 + 1 x +1
Z 2 
Z
x2 − 1 1 1
2. dx. 4. 1+x− ex+ x dx.
x4 + 1 1 x
2

3.6 Integration of Rational Function

A rational function is the quotient of two polynomials. Examples 3.5.2, 3.5.3, 3.5.18,
3.5.33 are some typical examples of integrating rational functions. In this section,
we systematically study how to integrate rational functions and how to convert some
integrations into the integration of rational functions.

3.6.1 Rational Function

Example 3.6.1. The idea in Example 3.5.2 can be extended to the integral of rational
functions whose denominator is a product of linear functions. For example, to
x2 − 2x + 3
integrate , we postulate
x(x + 1)(x + 2)

x2 − 2x + 3 A B C
= + + .
x(x + 1)(x + 2) x x+1 x+2

The equality is the same as

x2 − 2x + 3 = A(x + 1)(x + 2) + Bx(x + 2) + Cx(x + 1).

Taking x = 0, −1, −2, we get

3 = 2A, 6 = −B, 11 = −2C.

232 CHAPTER 3. INTEGRATION

Therefore
x2 − 2x + 3
Z Z  
3 6 11
dx = − − dx
x(x + 1)(x + 2) 2x x + 1 2(x + 2)
3 11
= log |x| − 6 log |x + 1| − log |x + 2| + C.
2 2

Example 3.6.2. If some real root of the denominator has multiplicity, then we need
x2 − 2x + 3
more sophisticated postulation. For example, to integrate , we postulate
x(x + 1)3
x2 − 2x + 3 A B C D
3
= + + 2
+ .
x(x + 1) x x + 1 (x + 1) (x + 1)3
This is the same as
x2 − 2x + 3 = A(x + 1)3 + Bx(x + 1)2 + Cx(x + 1) + Dx.
Taking various values, we get
x = 0 : 3 = A, x = −1 : 6 = −D,

d 
(coefficient of) x3 : 0 = A + B, : − 4 = D − C.
dx x=−1
Therefore A = 3, B = −3, C = −2, D = −6, and (C below means the general
constant, and is different from the coefficient C = −2 above)
Z 2 Z  
x − 2x + 3 3 3 2 6
dx = − − − dx
x(x + 1)3 x x + 1 (x + 1)2 (x + 1)3
 
 x 
= 3 log 
 + 2 + 3
+C
x+1  x + 1 (x + 1)2
 
 x 
= 3 log 
  + 2x + 5 + C.
x + 1  (x + 1)2

Example 3.6.3. In Examples 3.5.2, 3.6.1, 3.6.2, the numerator has lower degree than
the denominator. In general, we need to divide polynomials for this to happen.
x5
For example, to integrate , we first divide x5 by (x + 1)2 (x − 1) =
(x + 1)2 (x − 1)
x3 + x2 − x − 1.
x2 − x + 2


x3 + x2 − x − 1 x5
− x5 − x4 + x3 + x2
− x4 + x3 + x2
x4 + x3 − x2 − x
2x3 −x
− 2x3 − 2x2 + 2x + 2
− 2x2 + x + 2
3.6. INTEGRATION OF RATIONAL FUNCTION 233

Then

x5 2 −2x2 + x + 2
=x −x+2+
(x + 1)2 (x − 1) (x + 1)2 (x − 1)
9 1 1
= x2 − x + 2 − + 2
+ ,
4(x + 1) 2(x + 1) 4(x − 1)

and

x5 x3 x2
Z
9 1 1
2
dx = − + 2x − log |x + 1| − + log |x − 1| + C.
(x + 1) (x − 1) 3 2 4 2(x + 1) 4

Exercise 3.6.1. Compute the integral.

x2 dx x5 dx
Z
dx
Z Z
1. . 4. . 7. .
1+x x2 + x − 2 x(1 + x)(2 + x)
(2 − x)2 dx
Z
dx
Z Z
dx
2. . 5. . 8. .
2
x +x−2 2 − x2 x2 (1
+ x)
x4 dx
Z
dx
Z Z
xdx
3. . 6. . 9. .
2
x +x−2 1 − x2 (x + a)2 (x + b)2

The examples above illustrate how to integrate rational functions of the form
bm xm + bm−1 xm−1 + · · · + b1 x + b0
. This means exactly that all the roots of the
(x − a1 )n1 (x − a2 )n2 · · · (x − ak )nk
denominator are real. In general, however, a real polynomial may have complex
roots, and a conjugate pair of complex roots corresponds to a real quadratic factor.

1
Example 3.6.4. To integrate , we note that x3 = (x − 1)(x2 + x + 1), where
−1 x3
x2 + x + 1 has a conjugate pair of complex roots. We postulate

1 A Bx + C
= + 2 .
x3 −1 x−1 x +x+1

This means 1 = A(x2 + x + 1) + (Bx + C)(x − 1) and gives

x = 0 : 1 = A − C; x = 1 : 1 = 3A; x2 : 0 = A + B.
234 CHAPTER 3. INTEGRATION

1 1 2
Therefore A = , B = − , C = − , and
3 3 3
−x − 2
Z Z Z
dx 1 dx 1
3
= + 2
dx
x −1 3 x−1 3 x +x+1
d(x2 + x + 1) 1
Z Z
1 1 dx
= log |x − 1| − 2
+ 2
3 6 x +x+1 2 x +x+1
Z
1 1 1 dx
= log |x − 1| − log(x2 + x + 1) −
2  √ 2
3 6 2
x + 12 + 23
1 (x − 1)2 1 1 x + 12
= log 2 − √ arctan √ + C
6 x +x+1 2 3 3
2 2
2
1 (x − 1) 1 2x + 1
= log 2 − √ arctan √ + C.
6 x +x+1 3 3

x2
Example 3.6.5. To integrate , we postulate
(x4 − 1)2

x2 x2
=
(x4 − 1)2 (x − 1)2 (x + 1)2 (x2 + 1)2
A1 A2 B1 B2 C1 x + D1 C2 x + D2
= + + + + + 2 .
x − 1 (x − 1)2 x + 1 (x + 1)2 x2 + 1 (x + 1)2

Since changing x to −x does not change the left side, we see that A1 = −B1 ,
A2 = B2 , C1 = C2 = 0, and the equality becomes

x2 2A1 x2 + 1 D1 D2
4 2
= 2
+ 2A2 2 2
+ 2 + 2 .
(x − 1) x −1 (x − 1) x + 1 (x + 1)2

1 1 1
It is then easy to find A1 = − , A2 = , D1 = 0, D2 = − . Therefore with the
16 16 4
help of Example 3.5.15,

x2
Z Z 
1 1
4 2
dx = − +
(x − 1) 16(x − 1) 16(x − 1)2

1 1 1
+ + − dx
16(x + 1) 16(x + 1)2 4(x2 + 1)2
 
1 x + 1
− 1
= log 

16 x−1  16(x − 1)
1 x 1
− − 2
− arctan x + C
16(x + 1) 8(x + 1) 8
x3
 
1 x + 1 1
=− + log 
  − arctan x + C.
4(x4 − 1) 16 x − 1 8
3.6. INTEGRATION OF RATIONAL FUNCTION 235

(x2 + 1)4
Example 3.6.6. To integrate the rational function , we first notice that the
(x3 − 1)2
degree of the numerator is higher. Therefore we divide (x2 + 1)4 by (x3 − 1)2 .

x2 + 4


x6 − 2x3 + 1 x8 + 4x6 + 6x4 + 4x2 + 1
8 5
−x + 2x − x2
4x6 + 2x5 + 6x4 + 3x2 + 1
6 3
− 4x + 8x −4
2x5 + 6x4 + 8x3 + 3x2 − 3

This means that

(x2 + 1)4 2 2x5 + 6x4 + 8x3 + 3x2 − 3

= x + 4 + .
(x3 − 1)2 (x3 − 1)2

Since x3 − 1 = (x − 1)(x2 + x + 1), we postulate

2x5 + 6x4 + 8x3 + 3x2 − 3 A1 A2 B1 x + C1 B2 x + C2

3 2
= + 2
+ 2 + 2 .
(x − 1) x − 1 (x − 1) x + x + 1 (x + x + 1)2

This can be interpreted as an expression for 2x5 + 6x4 + 8x3 + 3x2 − 3, which gives

x = 0: − 3 = −A1 + A2 + C1 + C2 ,
x = 1: 16 = 9A2 ,
x = −1 : − 4 = −2A1 + A2 + 4(−B1 + C1 ) + 4(−B2 + C2 ),
x5 : 2 = A1 + B1 ,
x4 : 6 = A1 + A2 − B1 + C1 ,

d 
: 32 = 9A1 + 18A2 .
dx  x=1

Solving the system, we get

32 16 14 8 1
A2 = , A1 = , B1 = − , C1 = − , B2 = 0, C2 = − .
9 9 9 9 3
236 CHAPTER 3. INTEGRATION

Then
(x2 + 1)4
Z Z Z
1 3 32 16
3 2
dx = x + 4x + dx + dx
(x − 1) 3 9(x − 1) 9(x − 1)2
Z Z
14x + 8 1
− 2
dx − dx
9(x + x + 1) 3(x + x + 1)2
2

1 32 16
= x3 + 4x + ln |x − 1| −
3 9 9(x − 1)
Z 2 Z Z
7d(x + x + 1) dx dx
− 2
− 2
−
9(x + x + 1) 9(x + x + 1) 3(x + x + 1)2
2

1 32 16 7
= x3 + 4x + log |x − 1| − − log(x2 + x + 1)
3 9 9(x − 1) 9
Z Z
dx dx
− 2
− .
9(x + x + 1) 3(x + x + 1)2
2

By
 2 √ !2 1
2 1 3 x+ 2 2x + 1
x +x+1= x+ + , √ = √ ,
2 2 3 3
2

and Example 3.5.15, we have

Z
dx 2 2x + 1
2
= √ arctan √ + C,
x +x+1 3 3

and
1
Z
dx x+ 1 2x + 1
=  √ 2 2 +  √ 3 arctan √ +C
(x + x + 1)2
2
3 2 3 3
2 2
(x + x + 1) 2 2
2x + 1 4 2x + 1
= + √ arctan √ + C.
3(x2 + x + 1) 3 3 3

Combining everything together, we get

(x2 + 1)4 6x2 + 5x + 5 13

Z
1 3 7
3 2
dx = x + 4x − 3
+ log |x − 1| − log |x3 − 1|
(x − 1) 3 3(x − 1) 3 9
2 2x + 1
− √ arctan √ + C.
3 3 3

In general, the numerator of a rational function is a product of (x + a)m and

(x2 + bx + c)n , where b2 < 4c so that the factor x2 + bx + c has no real root. After
dividing the numerator by the denominator, we can make sure the numerator has
lower degree than the denominator. When the numerator has lower degree, the
3.6. INTEGRATION OF RATIONAL FUNCTION 237

rational function can then be expressed as a sum: For each factor (x + a)m of the
denominator, we have terms
A1 A2 Am
+ 2
+ ··· + ,
x + a (x + a) (x + a)m

and for each factor (x2 + bx + c)n of the denominator, we have terms

B1 x + C1 B2 x + C2 Bn x + Cn
2
+ 2 2
+ ··· + 2 .
x + bx + c (x + bx + c) (x + bx + c)n

The computation is then reduced to the integration of the terms.

We have
1

Z
A − + C, if m > 1,
(m − 1)(x + a) m−1
dx =
(x + a)m 
log |x + a| + C, if m = 1.

The quadratic term can be split into two parts

d(x2 + bx + c)
Z Z  Z
Bx + C B B dx
2 n
dx = 2 n
+ C− .
(x + bx + c) 2 (x + bx + c) 2 (x + bx + c)n
2

The first part is easy to compute

1

d(x + bx + c) −
2 + C, if n > 1,
Z
2 n
= (n − 1)(x + bx + c)n−1
2
(x + bx + c) 
log |x2 + bx + c| + C, if n = 1.

The second part can be computed by the recursive relation in Example 3.5.15
Z
dx
(x + bx + c)n
2
 Z 
1 2x + b dx
= + 2(2n − 3) .
(4c − b2 )(n − 1) (x2 + bx + c)n−1 (x2 + bx + c)n−1

For n = 1, 2, we have
Z
dx 2 2x + b
= √ arctan √ + C,
x2 + bx + c 4c − b 2 4c − b 2
Z
dx 2x + b 4 2x + b
2 2
= 2 2
+ 3 arctan √ + C.
(x + bx + c) (4c − b )(x + bx + c) (4c − b2 ) 2 4c − b2

Exercise 3.6.2. Compute the integral.

238 CHAPTER 3. INTEGRATION

(1 + x)2 dx x2 dx
Z Z
1. . 10. .
1 + x2 (x2 + 4x + 6)2
(2x2 + 3)dx
Z
dx
Z
2. . 11. .
x3 + x2 − 2 x4 −1
x2 dx
Z
dx
Z
3. 3
. 12. .
x −1 x4 − 1
(x + 1)3 dx
Z Z
dx
4. . 13. .
x3 + 1 (x4 − 1)2
Z
dx
Z
dx
5. . 14. .
(x + 1)(x2 + 1) x4 +4
Z
dx
Z
dx
6. . 15. .
x(x + 1)(x2 + x + 1) x4 + x2 + 1
Z Z
dx dx
7. . 16. .
(x + 1)(x + 1)(x3 + 1)
2 x6 +1
Z Z
dx xdx
8. . 17. .
(x + a )(x2 + b2 )
2 2 (x − 1)2 (x2
+ 2x + 2)

x3 dx (x4 + 4x3 + 4x2 + 4x + 4)dx

Z Z
9. . 18. .
(x2 + 1)2 x(x + 2)(x2 + 2x + 2)2
r
n ax + b
3.6.2 Rational Function of
cx + d
Using suitable changes of variables, some integrals can be changed to integrals of
rational functions.
r
x−2
Example 3.6.7. To integrate , we introduce
x−1
r
x−2 y2 − 2 2y
y= , x= 2 , dx = 2 dy.
x−1 y −1 (y − 1)2
Then for x ≥ 2, we have
Z r Z  
x−2
Z
2y 1 1 1 1 1
dx = y 2 dy = − + +
x−1 (y − 1)2 2 y − 1 y + 1 (y − 1)2 (y + 1)2
 
1 y − 1
= log  − y +C
2 y + 1  y2 − 1
√ √
q
 x−2
1  x − 2 − x − 1 x−1
= log  √
 √  − x−2 +C
2 x−2+ x−1 
x−1
−1
√ √ p
= log( x − 1 − x − 2) + (x − 1)(x − 2) + C.
3.6. INTEGRATION OF RATIONAL FUNCTION 239
√ √ p
For x ≤ 1, the answer is log( 2 − x − 1 − x) − (2 − x)(1 − x) + C.
The same substituting can be used to compute that, for x > 2,
Z r
x − 2 dx
Z Z
dx dx
√ = p =
2
x x − 3x + 2 x (x − 1)(x − 2) x − 1 x(x − 2)
Z r
x − 2 dx
Z
dy
= =− 2
x − 1 x(x − 2) y −2
 √  √ √
1 y + 2
 1 2x − 2 + x − 2
= √ log  √  + C = √ log √ √ + C.
2 2 y − 2 2 2 2x − 2 − x − 2
√ √
1 2 − 2x + 2 − x
For the case x < 1, the answer is − √ log √ √ + C.
2 2 2 − 2x − 2 − x
r
x 1 1
Example 3.6.8. For y = 3
, we have 3 = 1 + and
x+1 y x
 2
3 1 1
− 4 dy = − 2 dx = − 1 − 3 dx.
y x y

Therefore by Example 3.6.4, we have

− y34 dy
Z Z r Z   Z
dx 1 x 1 3dy
√ = 3
dx = 1− 3 y 2 = 3
3
x + x2
3 x x+1 y y −1

− 1 − y13
1 (y − 1)2 √ 2y + 1
= log 2 − 3 arctan √ + C
2 y +y+1 3
1 (y − 1)3 √ 2y + 1
= log 3 − 3 arctan √ + C
2 y −1 3
√ √ √
 r 
3 3 1 x
= log( x + 1 − x) − 3 arctan √ 2
3 3
+ 1 + C.
2 3 x+1
r
n ax + b
In general, a function involving can be integrated by introducing
cx + d
r
n ax + b dy n − b y n−1
y= , x= , dx = n(ad − bc) dy.
cx + d −cy n + a (cy n − a)2

Exercise 3.6.3. Compute the integral.

Z Z √
dx dx x32+x
Z
1. √ . 2. √ . 3. √ dx.
1+ x x(1 + x) x+ 32+x
240 CHAPTER 3. INTEGRATION
Z p
x3
Z r
1−x
Z
1 16. (x − a)(x − b)dx.
4. √
3
dx. 10. 3
dx.
x2 + 1 x2 1+x
Z
dx
Z
dx
5. √ √ dx. √
Z p
11. √ .
3
x+ 4x 2 x+ x+1+1 17. (x − a)(b − x)dx.
Z
dx
Z
dx
6. p . 12. √ √ .
4
x3 (a − x)
Z p
x+a+ x+b+c
Z √ √ √ 18. x (x − a)(b − x)dx.
x+1+ x−1
Z
1+ x+a
7. √ √ dx. 13. √ dx.
x+1− x−1 1+ x+b Z r
x−a
Z r
1−x
Z r
x−a 19. x dx.
8. dx. 14. dx. x−b
1+x x−b
Z
xdx
r Z r
1 1−x x−a
Z
9. dx. 15. dx. 20. p .
x2 1 + x b−x (x − a)(x − b)

Exercise 3.6.4. Compute the integral.

√
Z Z Z
dx dx
1. 1+ ex dx. 2. √ √ . 3. √ .
1 + e + 1 − ex
x ax + b

Exercise 3.6.5. Suppose R is a rational function. Suppose r, s are rational

Z numbers such
that r + s is an integer. Find a suitable change of variable, such that R(x, (ax + b)r (cx +
d)s )dx is changed into the antiderivative of a rational function.

Exercise 3.6.6. Suppose r, s, t are rational

Z numbers. For each of the following cases, find
a suitable change of variable, such that xr (a + bxs )t dx is changed into the integral of a
rational function.

1. t is an integer.

r+1
2. is an integer.
s

r+1
3. + t is an integer.
s

A theorem by Chebyshev1 says that these are the only cases that the antiderivative can
be changed to the integral of a rational function.

1
Pafnuty Lvovich Chebyshev, born 1821 in Okatovo (Russia), died 1894 in St Petersburg (Rus-
sia). Chebyshev’s work touches many fields of mathematics, including analysis, probability, number
theory and mechanics. Chebyshev introduced his famous polynomials in 1854 and later generalized
to the concept of orthogonal polynomials.
3.6. INTEGRATION OF RATIONAL FUNCTION 241

3.6.3 Rational Function of sin x and cos x

A rational function of sin x and cos x can be integrated by introducing
x 2y 1 − y2 2
y = tan , sin x = , cos x = , dx = dy.
2 1 + y2 1 + y2 1 + y2

Example 3.6.9. For a 6= 0, we have

Z Z Z
dx 2dy 2dy
=   = .
a + sin x a+ 2y
(1 + y 2 ) ay 2 + 2y + a
1+y 2

If |a| > 1, then

y+1 a tan x + 1
Z
dx 1 2
= q arctan q a +C = √ arctan √ 2 +C.
a + sin x a 1− 1
1 − a12 sign(a) a2 − 1 a2 − 1
a2

If |a| < 1, then

 q 
1 1
 
Z
dx 1  y + a
− a 2 − 1 
= log  +C
 
1
q
a + sin x
q
a a12 − 1  y + + a12 − 1 
 
a
 a tan x + 1 − √1 − a2 

1 2
= √ log  √  + C.
 
2 x 2
sign(a) 1 − a  a tan 2 + 1 + 1 − a 
If |a| = 1, then
−2 −2
Z
dx
= +C = + C.
a + sin x ay + 1 a tan x2 + 1
Z
dx
The example can be extended to . We have
a + b sin x + c cos x
√
b sin x + c cos x = b2 + c2 sin(x + θ),
b c
where θ is any fixed angle satisfying sin θ = √ and cos θ = √ . Then
b2 +c 2 b + c2
2
Z Z
dx dy
= √ , y = x + θ.
a + b sin x + c cos x a + b + c2 sin y
2

Example 3.6.10. Rational functions of sin x and cos x can be integrated by a simpler
substitution if it has additional property. For example, to integrate the function
sin x
, we introduce
sin x + cos x
dy
y = tan x, x = arctan y, dx = .
y2+1
242 CHAPTER 3. INTEGRATION

Here we use the tangent of the full angle x, instead of half the angle in Example
3.6.9. Then
Z y dy
y2 + 1
Z Z
sin xdx tan xdx y 2 +1 1 1
= = = log + arctan y + C
sin x + cos x tan x + 1 y+1 4 (y + 1)2 2
1 1
= x − log | sin x + cos x| + C.
2 2
The key point here is that the integrand is a rational function R(sin x, cos x) that
satisfies R(−u, −v) = R(u, v). In this case, the integrad can always be written as a
rational function of tan x, and the change of variable can be applied.
cos x
Example 3.6.11. Note that rational function of sin x and cos x is
cos x sin x + sin3 xm
odd in the sin x variable. This is comparable to the function cos x sinn x for the case
n is odd. We may introduce the same change of variable y = cos x, dy = − sin xdx
like the earlier example and get
−ydy
Z Z Z
cos xdx cos x sin xdx
3 = 2 2 =
cos x sin x + sin x cos x sin x + (sin x) 2 y(1 − y ) + (1 − y 2 )2
2
 √ 
1  y − 1+ 5  1 
y − 1

2 
= √ log  √ + log  +C
 
5  y − 1− 5  2 y + 1
2
 √ 
1  2 cos x − 1 − 5  1 1 − cos x

= √ log  √  + log + C.
5  2 cos x − 1 + 5  2 1 + cos x

Similarly, a rational function of sin x and cos x that is odd in the cos x variable
can be integrated by introducingZ x = sin y. If R(−u, −v) = R(u, v), then we may
introduce y = tan x to compute R(sin x, cos x)dx.

Exercise 3.6.7. Compute the integral.

1 − r2
Z
dx
Z
1. dx, |r| < 1. 6. .
1 − 2r cos x + r2 2 sin x + sin 2x

sin2 x
Z Z
dx
2. . 7. dx.
a − cos 2x 1 + sin2 x
Z
dx
Z
dx
3. . 8. .
a + tan x sin(x + a) sin(x + b)
Z
dx
Z
dx
4. . 9. .
cos x + tan x (1 + cos2 x)(2 + sin2 x)
Z
(1 + sin x)dx
Z
dx
5. . 10. .
sin x + tan x sin x(1 + cos x)
3.7. IMPROPER INTEGRAL 243
Z Z
dx dx
11. . 14. .
(a + cos x) sin x a2 sin2 x + b2 cos2 x
1 − tan x
Z Z
(sin x + cos x)dx
12. . 15. dx.
sin x(sin x − cos x) 1 + tan x
Z Z √
dx
13. . 16. tan xdx.
(a + cos2 x) sin x

3.7 Improper Integral

The definition of Riemann integral requires both the function and the interval to
be bounded. If either the function or the interval is unbounded, then the integral
is improper. We may still make sense of an improper integral if it can be viewed as
the limit of usual integral of bounded function on bounded interval.
Throughout this section, we assume that our functions are integrable on closed
interval of their domains.

3.7.1 Definition and Property

Example 3.7.1. The function e−x is bounded on the unbounded interval [0, +∞). To
Z +∞
make sense of the improper integral e−x dx, we consider the integral on any
0
bounded interval Z b
e−x dx = 1 − e−b .
0

As the bounded interval approaches [0, +∞), we get

Z b
lim e−x dx = lim (1 − e−b ) = 1.
b→+∞ 0 b→+∞

Z +∞
Therefore the improper integral e−x dx has value 1. Geometrically, this means
0
that the area of the unbounded region under the graph of the function e−x and over
the interval [0, +∞) is 1.

Example 3.7.2.
Z The function log x is unbounded on the bounded interval (0, 1]. Since
1
the integral log xdx is improper at 0+ , we consider the integral over [, 1] for  > 0
0

Z 1 x=1
log xdx = (x log x − x) = −1 −  log  + .

 x=
244 CHAPTER 3. INTEGRATION

1
log x
e−x

Figure 3.7.1: The unbounded region has area 1.

Since the right side converges to −1 as  → 0+ , the improper integral converges and
has value Z 1 Z 1
log xdx = lim+ log xdx = −1.
0 →0 

The unbounded region measured by this improper integral is actually the same as
the one in Example 3.7.1, up to a rotation.
Z +∞
dx
Example 3.7.3. Consider the improper integral , where a > 0. We have
a xp
 1−p
b b − a1−p
, if p 6= 1,
Z
dx 
= 1 − p
a xp
log b − log a, if p = 1.


As b → +∞, we get
 1−p
Z +∞ a
dx  , if p > 1,
p
= p−1
a x
diverge, if p ≤ 1.


Z 1
dx
Example 3.7.4. The integral is improper at 0+ for p > 0. For  > 0, we have
0 xp

1 − 1−p

1
, if p 6= 1,
Z
dx 
p
= 1−p
 x
− log , if p = 1.

3.7. IMPROPER INTEGRAL 245

As  → 0+ , we get
1

1
, if p < 1,
Z
dx 
= 1 − p
0 xp diverge, if p ≥ 1.
Z b
By the same argument, for a < b, the improper integrals (x − a)p dx and
a
Z b
(b − x)p dx converge if and only if p > −1.
a

Z +∞
dx
Example 3.7.5. The integral is improper at +∞ and −∞. The integral
−∞ x2 + 1
on a bounded interval is
Z b
dx
= arctan b − arctan a.
a x2 +1

Then we get
Z +∞ Z b
dx dx π  π
2
= lim = lim arctan b − lim arctan a = − − = π.
−∞ x + 1 a→−∞ a x2 + 1 b→+∞ a→−∞ 2 2
b→+∞

Example 3.7.6. Since

Z b
cos xdx = sin b
0
Z +∞
diverges as b → +∞, the improper integral cos xdx diverges.
0

In general, an integral may be improper at several places. We may divide the

interval into several parts, such that each part contains exactly one improperness.
If an integral has one improperness at +∞ or −∞, then we study the limit of the
integral on bounded intervals. If an integral has one improperness at a+ or a− , then
we study the limit of the integral on intervals [a + , b] or [b, a − ].

Example 3.7.7. A naive application of the Newton-Leibniz formula would tell us

Z 1
dx x=1
= (log |x|) = log 1 − log 1 = 0.

−1 x x=−1

1
However, the computation is wrong since the integrand is not continuous on
Z 1 x
dx
[−1, 1]. In fact, the integral is improper on both sides of 0, and we need
−1 x
246 CHAPTER 3. INTEGRATION
Z 0Z 1
dx dx
both improper integrals and to converge and then get
−1 x 0 x
Z 1 Z 0 Z 1
dx dx dx
= + .
−1 x −1 x 0 x
Since Z 1 Z 1
dx dx
= lim+ = lim+ − log  = +∞
0 x →0  x →0
Z 1 Z 1
dx dx
diverges, the improper integral diverges, so that also diverges.
0 x −1 x

Z 0
Example 3.7.8. To compute the improper integral xex dx, we start with integra-
−∞
tion by parts on a bounded interval
Z 0 Z 0 Z 0
x x b
xe dx = xde = −be − ex dx = −beb − 1 + eb .
b b b

Taking b → −∞ on both sides, we get

Z 0
xex dx = −1.
−∞

The example shows that the integration by parts can be extended to improper
integrals, simply by taking the limit of the integration by parts formula for the usual
proper integrals.
Z +∞
dx
Example 3.7.9. For a > 1, consider the improper integral . We have
a x(log x)p
Z b Z b Z log b
dx d(log x) dy
= = .
a x(log x)p a (log x)p log a yp
Taking b → +∞ on both sides, we get
Z +∞ Z +∞
dx dy
= .
a x(log x)p log a y
p

The equality means that the improper integral on the left converges if and only if
the improper integral on the right converges, and the two values are the same. By
Z +∞
dx
Example 3.7.3, we see that the improper integral converges if and
a x(log x)p
only if p < 1, and
Z +∞
dx (log a)p+1
= − , if p < 1.
a x(log x)p p+1
3.7. IMPROPER INTEGRAL 247

The example shows that the change of variable can also be extended to improper
integrals, simply by taking the limit of the change of variable formula for the usual
proper intervals.

Exercise 3.7.1. Determine the convergence of improper integrals and evaluate the conver-
gent ones.

Z +∞ Z 0 Z 1
dx
1. xp dx. 4. ax dx. 7. √ .
0 −∞ −1 1 − x2
Z 1 Z 1 π
dx dx
Z
2
2. . 5. . 8. tan xdx.
0 x(− log x)p −1 1 − x2 0
Z +∞ Z +∞ Z π
x dx
3. a dx. 6. . 9. sec xdx.
0 2 1 − x2 0

Exercise 3.7.2. Determine the convergence of improper integrals and evaluate the conver-
gent ones.

Z +∞ Z +∞ Z +∞
dx dx log x
1. . 7. √ . 13. dx.
1 x+1 0 x(1 + x) 1 x2
Z +∞ 9 1
xdx
Z Z
dx log x
2. . 8. √ . 14. √ dx.
−∞ x2 + 1 1
3
x−9 0 x
Z +∞ Z +∞ Z +∞
dx x arctan x
3. √ . 9. xex dx. 15. dx.
1
3
x+1 0 0 (1 + x2 )2
+∞ +∞ +∞
x2 dx
Z Z Z
2
4. . 10. xe−x dx. 16. e−ax cos bxdx.
0 x3 + 1 0 0
+∞ +∞
x2 dx √ +∞
Z Z Z
5. . 11. e− x
dx. 17. e−ax sin bxdx.
0 (x3 + 1) 2
0 0
Z +∞ Z 1 Z +∞
dx
6. . 12. x log xdx. 18. e−x | sin x|dx.
0 x(x + 1)(x + 2) 0 0

Exercise 3.7.3. Prove that

  Z 1
1 1 2 n
lim log + log + · · · + log = log xdx.
n→∞ n n n n 0

Note that the left side is a “Riemann sum” for the right side. However, since the integral is
improper, we cannot directly use the fact that
√ the Riemann sum converges to the integral.
n
n!
Moreover, the limit is the same as limn→∞ = e−1 .
n
248 CHAPTER 3. INTEGRATION

3.7.2 Comparison Test

The improper integral is defined by taking limit. Therefore there is always the
problem of convergence. Z +∞
The convergence of the improper integral f (x)dx means the convergence
Z b a

of the function I(b) = f (x)dx as b → +∞. The Cauchy criterion for the conver-
a
gence is that, for any  > 0, there is N , such that
Z c 
 
b, c > N =⇒ |I(c) − I(b)| =   f (x)dx < .
b

The Cauchy criterion for the convergence

Z b of other types of improper integrals is
similar. For example, if the integral f (x)dx is improper at a+ , then the integral
a
converges if and only if for any  > 0, there is δ > 0, such that
Z d 
 
c, d ∈ (a, a + δ) =⇒  f (x)dx < .
c

The Cauchy criterion shows that the convergence of an improper integral depends
only on the behavior of the function near the improper place. Moreover, the Cauchy
criterion also implies the following test for convergence.
Z b
Theorem 3.7.1 (Comparison Test). If |f (x)| ≤ g(x) on (a, b) and the integral g(x)dx
Z b a

converges, then f (x)dx also converges.

a

Note that if |f | ≤ g, then | |f | | ≤ g. Therefore whenever we use the comparison

Z b
test, we may always conclude that |f (x)|dx also converges.
Z b a Z b
We say that f (x)dx absolutely converges if |f (x)|dx converges. The com-
a a
parison test tells us that absolute convergence impliesZconvergence.
+∞
Here we justify the comparison test for the integral f (x)dx that is improper
Z +∞ a

at +∞. The convergence of g(x)dx implies that for any  > 0, there is N ,
a
such that Z c
c > b > N =⇒ g(x)dx < .
b
The assumption |f (x)| ≤ g(x) further implies that for c > b,
Z c  Z c Z c
 

 f (x)dx ≤
 |f (x)|dx ≤ g(x)dx.
b b b
3.7. IMPROPER INTEGRAL 249

Combining the two implications, we get

Z c  Z c
 
c > b > N =⇒  f (x)dx ≤ g(x)dx < .
b b
Z +∞
This verifies the Cauchy criterion for the convergence of f (x)dx.
a

Z +∞
1 − 23 3
Example 3.7.10. We know √ < x for x ≥ 1. Since x− 2 dx converges,
x3 + 1 Z 1
+∞
dx
by the comparison test, we know √ also converges. We note that
x 3+1
Z +∞ 1
dx
√ converges too because only the behavior of the function for big x
0 x3 + 1
(i.e., near +∞) is involved.
Z +∞
2 2
Example 3.7.11. To determine the convergence of e−x dx, we use 0 < e−x ≤
0 Z +∞
2
−x
e for x ≥ 1. By Example 3.7.1 and the comparison test, we know e−x dx
Z 1 Z +∞ 1
−x2 −x2
converges. Since e dx is a proper integral, we know e dx also con-
0 0
verges.
Z +∞
log x
Example 3.7.12. To determine the convergence of dx, p > 0, we use the
1 xp Z +∞
log x 1 1
comparison p
≥ p > 0 for x ≥ e. For p ≤ 1, the divergence of dx
x x Z 1 xp
+∞
log x
implies the divergence of dx.
1 xp Z +∞
1
For p > 1, although we also know the convergence of p
dx, the comparison
Z1 +∞ x
log x
above cannot be used to conclude the convergence of dx. Instead, we
1 xp
choose q satisfying p > q > 1. Then by
log x
p log x
lim x = lim p−q = 0,
x→+∞ 1 x→+∞ x
x q
  Z +∞
 log x  1 dx
we have  p  ≤ q for sufficiently large x. By the convergence of q
,
x x Z +∞ 1 x
log x
therefore, we know the converges of dx.
1 xp
250 CHAPTER 3. INTEGRATION
Z +∞
log x
We conclude that dx converges if and only if p > 1.
1 xp
Z +∞ Z +∞
Exercise 3.7.4. Compare the integrals I = f (x)dx and J = g(x)dx that
a a
are improper at +∞.
f (x)
1. Prove that if limx→+∞ converges, g(x) ≥ 0 for sufficiently large x, and J
g(x)
converges, then I also converges.
f (x)
2. Prove that if limx→+∞ converges to a nonzero number, and g(x) ≥ 0 for
g(x)
sufficiently large x, then I converges if and only if J converges.

This is known as the Limit Comparison Test.

Z 1
dx
Example 3.7.13. The integral p is improper at 0+ and 1− . By applying
0 x(1 − x)
the idea of Exercise 3.7.4 to
1 1
p p
x(1 − x) x(1 − x)
lim+ = 1, lim− = 1,
x→0 1 x→1 1
√ √
x 1−x
Z 1 Z 1 Z 1
dx dx dx
and the convergence of √ and √ , we know that p con-
0 x 0 1−x 0 x(1 − x)
verges.
Z +∞
Exercise 3.7.5. Suppose f ≥ 0, prove that the improper integral f (x)dx converges if
Z b a Z b
and only if f (x)dx, for all b ∈ [a, +∞), is bounded. What about the integral f (x)dx
a a
that is improper at a+ .
Z +∞ Z +∞ Z +∞
2 2
Exercise 3.7.6. Suppose f dx and g dx converge. Prove that f gdx and
Z +∞ a a a

(f + g)2 dx converge.
a

Exercise 3.7.7. Determine convergence.

Z +∞ Z 2 Z 1
dx dx dx
1. p q
. 2. p q
. 3. .
2 x (log x) 1 x (log x) 0 xp | log x|q
3.7. IMPROPER INTEGRAL 251

1 1 +∞
xp dx
Z Z Z
dx
4. . 7. √ , q > 0. 10. xp ax dx, a > 0.
−∞ |x|p | log x|q 0 1 − xq 1
Z 1
Z +∞
dx dx Z +∞
5. . 8. , q > 11. xp log(1 + xq )dx.
xp + (log x)q 0 xp (1 − xq )r
2 1
0.
+∞ Z 1 1
xp dx
Z Z
6. . 9. xp ax dx, a > 0. 12. xp log(1 + xq )dx.
0 1 + xq 0 0

Exercise 3.7.8. Determine convergence.

Z +∞ Z +∞ Z +∞
xdx x sin xdx x arctan xdx
1. √ . 3. √ . 5. √ .
5 2
x − 2x + 1 x5 − 2x2 + 1 x5 − 2x2 + 1
2 2 2
Z 1 Z 1 Z 1
xdx x sin xdx x arctan xdx
2. √ . 4. √ . 6. √ .
5 2
x − 2x + 1 x5 − 2x2 + 1 x5 − 2x2 + 1
0 0 0

Exercise 3.7.9. Determine convergence.

Z +∞ Z 1 Z 1
x+1 dx
1. dx. 4. (1 − x)p | log x|q dx. 7. .
3 − 2x + 3 √
q
1 x 0 0
p
x+ x+ x
Z +∞
Z 10
dx
Z 2
dx dx
2. . 5. . 8. .
√
p q p
0 |x2 − 4x + 3| 1 (3x − 2 − x2 )p 1 x+ x+ x
Z +∞
+∞ +∞
dx
x2 )dx 9. .
Z Z
log x log(1 + q
√
3. √ p dx. 6. . 0
p
1+ 1+ x
1 x +1 0 1 + xq

Exercise 3.7.10. Determine convergence.

Z π Z π Z +∞
2 dx 4 dx
1. p
. 6. p
. 11. eax cos bxdx.
0 cos x 0 | sin x − cos x| 0
Z π Z π Z +∞ √
2 dx 2 dx
2. q . 7. . 12. e− x
cos bx2 dx.
p
0 x sin x π (1 − sin x)p 0
4
Z π Z π Z +∞
2 dx 2 x + cos x
3. p qx
. 8. (− log sin x)p dx. 13.
x2 − sin x
dx.
0 sin x cos 0 1
Z π π Z +∞
log(x + cos x)
3
Z
2
p
4. tan xdx. 9. xp log sin xdx. 14. dx.
0 0 1 x2 − sin x
π π
Z Z Z +∞
3 2 log x + cos x
5. tanp x logq xdx. 10. | log tan x|p dx. 15.dx.
0 0 1 x2 − sin x
Z +∞  
1 a
Exercise 3.7.11. Find a constant a, such that √ + dx converges.
0 x2 + 1 x + 1
Moreover, evaluate the integral for this a.
252 CHAPTER 3. INTEGRATION

3.7.3 Conditional Convergence

Although the comparison test is very effective, some improper integrals needs to be
modified before the comparison test can be applied.
Z +∞
sin x
Example 3.7.14. By the comparison test, we know dx converges for p > 1.
1 xp Z +∞
sin x
However, the argument fails for the case p = 1. We will show that dx
1 x
still
Z +∞converges.
  We will also show that, after taking the absolute value, the integral
 sin x 
 x  dx actually diverges. This means that the comparison test cannot be
 
1 Z +∞
sin x
directly applied to dx.
1 x
Using integration by parts, we have
Z b Z b Z b
sin x 1 cos b cos x
dx = − d cos x = − + cos 1 − dx.
1 x 1 x b 1 x2
Z +∞
cos x
By the comparison test, the improper integral dx converges. Therefore
1 x2 Z
+∞
sin x
the right side converges as b → +∞, and we conclude that dx converges.
1 x
On the other hand, we have
Z nπ   Z nπ   n Z kπ   n Z kπ
 sin x   sin x  X  sin x  X 1
  dx ≥   dx =   dx ≥ | sin x|dx
 x   x   x  kπ
1 π k=2 (k−1)π k=2 (k−1)π
n  
X 1 1 1 1 1
≥ ≥ + + ··· + .
k=2
kπ π 2 3 n

+∞
Z  
 sin x 
By Example 1.3.8, the right side diverges to +∞. Therefore  x  dx diverges.
 
1

Example 3.7.15. By a change of variable, we have

Z +∞ Z +∞ +∞
√
Z
2 sin y
sin x dx = sin yd( y) = √ dy.
0 0 0 2 y

The integral on the right is proper at 0+ and improper at +∞.

Z +∞ It converges by an
argument similar to Example 3.7.14. Therefore the integral sin x2 dx converges.
0
We first used the change of variable, then used the integration
Z +∞by parts, and
finally used the comparison test to conclude the convergence of sin x2 dx.
0
3.7. IMPROPER INTEGRAL 253
Z +∞
The reader can further use the idea of Example 3.7.14 to show that | sin x2 |dx
0
diverges.
Z b
Example 3.7.14 shows that it is possible for an improper integral f (x)dx to
Z b a

converge but the corresponding absolute improper integral |f (x)|dx to diverge.

a Z b
In this case, the integral converges but not absolutely, and we say f (x)dx con-
a
ditionally converges.
The idea in Example 3.7.14 can be elaborated to get the following useful tests.
Z b
Theorem 3.7.2 (Dirichlet Test). Suppose f (x)dx is bounded for all b ∈ [a, +∞).
Z +∞ a

Suppose g(x) is monotonic and limx→+∞ g(x) = 0. Then f (x)g(x)dx con-

a
verges.
Z +∞
Theorem 3.7.3 (Abel Test). Suppose f (x)dx converges. Suppose g(x) is mono-
Za +∞
tonic and bounded on [a, +∞). Then f (x)g(x)dx converges.
a

1
The tests basically replaces sin x and in the example by f (x) and g(x). In case
x
f (x) is continuous and g(x) is continuously differentiable,
Z we can justify the tests x
by repeating the argument in the example. Let F (x) = f (t)dt. Then F (a) = 0,
a
and Z b Z b Z b
f (x)g(x)dx = g(x)dF (x) = g(b)F (b) − F (x)g 0 (x)dx.
a a a
Under the assumption of the Dirichlet test, we have limb→+∞ g(b)F (b) = 0, and
|F (x)| < M for some constant M and all x ≥ a. Assume the monotonic function
g(x) is increasing. Then g 0 (x) ≥ 0, and
|F (x)g 0 (x)| ≤ M g 0 (x).
Since Z +∞ Z b
0
g (x)dx = lim g 0 (x)dx = lim (g(b) − g(a)) = −g(a)
a b→+∞ a b→+∞

converges, by the comparison test, the improper integral

Z +∞ Z b
0
F (x)g (x)dx = lim F (x)g 0 (x)dx
a b→+∞ a
254 CHAPTER 3. INTEGRATION
Z +∞ Z b
converges. Therefore f (x)g(x)dx = lim f (x)g(x)dx also converges. The
a b→+∞ a
proof for decreasing g(x) is similar.
Under the assumption of the Abel test, we know both F (b) and g(b) converge
as b → +∞. Therefore F (x) is bounded, and we may apply the comparison test
as
Z before. Moreover, the convergence Zof limb→+∞ g(b) implies the convergence of
+∞ +∞
0
g (x)dx. We conclude again that f (x)g(x)dx converges.
a a

Exercise 3.7.12. Determine convergence. Is the convergence absolute or conditional?

Z +∞ Z 1 Z +∞
sin xq 1 1 sin2 x
1. dx. 4. sin dx. 7. dx.
0 xp 0 x
p x 0 x
Z +∞ Z +∞ Z +∞
cos xq cos ax sin3 x
2. p
dx. 5. dx. 8. dx.
0 x 0 1 + xp 0 x
Z 1 Z +∞ p Z +∞
1 1 x sin ax sin x arctan x
3. sin dx. 6. q
dx. 9. dx.
0 x x 0 1 + x 1 xp
Z +∞
Exercise 3.7.13. Construct a function f (x) such that |f (x)| = 1 and f (x)dx con-
0
verges.

Finally, we show some examples of using the integration by parts and change of
variable to compute improper integrals. We note that the convergence needs to be
verified before applying the properties of integration.
Z +∞
2
Example 3.7.16. In Example 3.7.11, we know the convergence of e−x dx. By the
2 Z +∞ 0
xp e−x 2
similar idea, especially limx→+∞ −x = 0, we know that xp e−x dx converges
e 0
for any p ≥ 0.
Let Z +∞
2
In = xn e−x dx.
0
Then we may apply the integration by parts to get
1 +∞ n−1 −x2
Z
In = − x de
2 0
x=+∞
n − 1 +∞ n−2 −x2 n−1
Z
1 n−1 −x2 
=− x e  + x e dx = In−2 .
2 x=0 2 0 2
It is known (by using integration of two variable function, for example) that
Z +∞ √
−x2 π
I0 = e dx = .
0 2
3.7. IMPROPER INTEGRAL 255

We can also apply the change of variable to get

Z +∞
1 +∞ −x
Z
−x2 1
I1 = xe dx = e dx = .
0 2 0 2
Then we can use the recursive relation to compute In for all natural number n.
Z π
2
Example 3.7.17. The integral log sin xdx is improper at 0. By L’Hospital’s rule,
0
we have
cos x
log sin x x cos x
lim+ = lim+ sin x = lim+ = 1.
x→0 log x x→0 1 x→0 sin x
x
Z 1 Z 1
By the convergence of | log x|dx = − log xdx in Example 3.7.2 and the com-
0Z π 0
2
parison test, we see that log sin xdx converges.
0
The value of the improper integral can be computed as follows
Z π Z π Z π
2 4 2
log sin xdx = log sin xdx + log sin xdx
π
0 0 4
π
Z
4
Z 0
= log sin xdx − log cos xdx
π
0 4
Z π Z π  
4 4 1
= (log sin x + log cos x)dx = log sin 2x dx
0 0 2
Z π Z π
4 π 1 2 π
= log sin 2xdx − log 2 = log sin xdx − log 2.
0 4 2 0 4
Note that all the deductions are legitimate because all the improper integrals in-
volved converge. Therefore we conclude that
Z π
2 π
log sin xdx = − log 2.
0 2

Exercise 3.7.14. Compute improper integral.

Z 1 Z +∞ π
dx
Z
2
1. (log x)n dx. 3. . 5. log cos xdx.
0 0 (1 + x2 )n 0
Z +∞ Z 1
xn dx
2. xn e−x dx. 4. √ .
0 0 1 − x2

Exercise 3.7.15. The Gamma function is

Z +∞
Γ(x) = tx−1 e−t dt.
0
256 CHAPTER 3. INTEGRATION

1. Show that the function is defined for x > 0.

2. Show the other formulae for the Gamma function

Z ∞ Z ∞
2
Γ(x) = 2 t2x−1 e−t dt = ax tx−1 e−at dt.
0 0

3. Show that Γ(x + 1) = xΓ(x) and Γ(n) = (n − 1)!.

3.8 Application to Geometry

3.8.1 Length of Curve
Curves in a Euclidean space are often presented by parametrization. For example,
the unit circle centered at the origin of R2 may be parametrized by the angle

x = cos θ, y = sin θ, 0 ≤ θ ≤ 2π.

The helix in R3
h
x = r cos θ, y = r sin θ, z= θ, 0 ≤ θ ≤ 2π,
2π
moves along the circle of radius r from the viewpoint of the (x, y)-coordinates, and
moves up in the z-direction in constant speed, such that each round moves up by
height h.
In general, a parametrized curve in R2 is given by

x = x(t), y = y(t), a ≤ t ≤ b.

The initial point of the curve is (x(a), y(a)), and the end point is (x(b), y(b)). To
compute the length of the curve between the two points, we consider the length s(t)
from the initial point (x(a), y(a)) to the point (x(t), y(t)). We find the change s0 (t)
and then integrate the change to get s(t). The length of the whole curve is s(b).
Similar to the argument for the area of the region G[a,b] (f ), we need to be careful
about the sign. In the subsequent discussion, we pretend everything is positive
(which at least gives you the right derivative), and further argument about the
negative case is omitted. Moreover, we restrict the argument to the case x(t) and y(t)
are nice. In fact, we will assume the two functions are continuously differentiable.
In general, we may break the curve into finitely many continuously differentiable
pieces and add the lengths of the pieces together.
As the parameter t is changed by ∆t, the change ∆s = s(t + ∆t) − s(t) of the
length is the length of the curve segment from (x(t), y(t)) to (x(t + ∆t), y(t + ∆t)).
The curve segment is approximated by the straight line connecting the two points.
Therefore the length of the curve is approximated by the length of the straight line
p p
∆s ≈ (x(t + ∆t) − x(t))2 + (y(t + ∆t) − y(t))2 = (∆x)2 + (∆y)2 .
3.8. APPLICATION TO GEOMETRY 257

(x(t + ∆t), y(t + ∆t))

(x(t), y(t))

t=b
t=a

Figure 3.8.1: Length of curve.

Dividing the change ∆t of parameter, we get

s 2  2
∆s ∆x ∆y
≈ + .
∆t ∆t ∆t
The approximation gets more refined as ∆t → 0. By taking the limit as ∆t → 0,
the approximation becomes an equality
s 2  2
0 ∆s ∆x ∆y p
s (t) = lim = lim + = x0 (t)2 + y 0 (t)2 .
∆t→0 ∆t ∆t→0 ∆t ∆t
p
Therefore the length function s(t) is the antiderivative of x0 (t)2 + y 0 (t)2 , or
p
ds = x0 (t)2 + y 0 (t)2 dt,
and we have Z bp
length of curve = s(b) = x0 (t)2 + y 0 (t)2 dt.
a

Example 3.8.1. The length of the unit circle is

Z 2π p Z 2π
2 2
(− sin θ) + (cos θ) dθ = dθ = 2π.
0 0

x2 y 2
More generally, an ellipse + 2 = 1 can be parametrized as
a2 b
x = a cos θ, y = b sin θ, 0 ≤ θ ≤ 2π.
The length of the ellipse is the so called elliptic integral
Z 2π p Z 2π √
2 2
b2
(−a sin θ) + (b cos θ) dθ = a 1 + K cos2 θdθ, K= − 1.
0 0 a2
The integral cannot be computed as an elementary expression if a 6= b.
258 CHAPTER 3. INTEGRATION

Example 3.8.2. The graph of a function f (x) on [a, b] is a curve

x = t, y = f (t), t ∈ [a, b].

The length of the graph is Z bp

1 + f 0 (x)2 dx,
a
where the variable t is substituted to the more familiar x.
For example, the parabola y = x2 is cut by the diagonal y = x. With the help
of Example 3.5.31, the finite segment corresponding to x ∈ [0, 1] has length
Z 1p Z 2√
1
1 + (2x)2 dx = 1 + x2 dx
0 2 0
1 √ √ 2 1 √ 1 √
= x 1 + x2 + log( 1 + x2 + x) = 5 + log( 5 + 2).
4 0 2 4

x2

Figure 3.8.2: Parabola x2 cut by the diagonal.

2 2
Example 3.8.3. The astroid x 3 + y 3 = 1 can be parametrized as

x = cos3 t, y = sin3 t, t ∈ [0, 2π].

Note that the range [0, 2π] for t corresponds to moving around the astroid exactly
once. Therefore the perimeter is
Z 2π q Z 2π
2
(−3 cos2 t sin t)2 + (3a sin t cos t)2 dt = 3| sin t cos t|dt = 6.
0 0

Example 3.8.4. The argument about the length of curves also applies to curves in
R3 and leads to
Z bp
length of curve = x0 (t)2 + y 0 (t)2 + z 0 (t)2 dt.
a
3.8. APPLICATION TO GEOMETRY 259

Figure 3.8.3: Astroid.

For example, the length of one round of the helix is

s  2 s  2
Z 2π Z 2π
2 2
h 2
h
(−r sin θ) + (r cos θ) + dθ = r + dθ
0 2π 0 2π
s  2
2
h p
= 2π r + = (2πr)2 + h2 .
2π

The result has a simple geometrical interpretation: By cutting along a vertical line,
the cylinder can be “flattened” into a plane. Then the helix becomes the hypotenuse
of a right triangle with horizontal length 2πr and vertical length h.

Example 3.8.5. When a circle rolls along a straight line, the track of one point on
the circle is the cycloid. Let r be the radius of the circle, and assume the point is
at the bottom at the beginning. After rotating angle t, the center of the circle is at
(rt, r), and the point is at (rt, r) + r(− cos(t − π2 ), sin(t − π2 )). Therefore the cycloid
is parameterized by
x = rt − r sin t, y = r − r cos t.
As the circle makes one complete rotation, we get one period of the cycloid,
corresponding to t ∈ [0, 2π]. The length of this one period is
Z 2π p Z 2π p Z 2π  
2 2
 t 
(r − r cos t) + (r sin t) dt = r 2(1 − cos t)dt = r 2 sin  dt = 8r.

0 0 0 2

Exercise 3.8.1. Compute length.

1. y 2 = 2x, x ∈ [0, a]. 3. y = ex , x ∈ [0, a].

2. x2 = 2py, x ∈ [0, a]. 4. y = log x, x ∈ [1, a].

260 CHAPTER 3. INTEGRATION

Figure 3.8.4: Cycloid.

5. y = log(4 − x2 ), x ∈ [−1, 1]. 9. y 2 = x3 , x ∈ [0, a].

h πi
6. y = log cos x, x ∈ 0, . 10. y 4 = x3 , x ∈ [0, a].
4
h πi ex + 1
7. y = log sec x, x ∈ 0, . 11. y = log , x ∈ [1, 2].
4 ex − 1
Z xp
ex + e−x 12. y = t3 − 1dt, x ∈ [1, 4].
8. y = , x ∈ [−a, a].
2 0

Exercise 3.8.2. Compute length.

√ √
1. x + y = 1, x, y ≥ 0.
2. x = t, y = log t, t ∈ [1, 2].
3. x = et − t, y = et + t, t ∈ [0, 1].
4. x = et cos t, y = et sin t, t ∈ [0, π].
5. x = cos2 t, y = cos t sin t, t ∈ [0, π].
6. x = 3 cos t − cos 3t, y = 3 sin t − sin 3t, t ∈ [0, π].
t hπ π i
7. x = cos t + log tan , y = sin t, t ∈ , .
2 4 2
8. x = cos t + t sin t, y = sin t − t cos t, z = t2 , t ∈ [0, 2π].
√
9. x = a2 et , y = b2 e−t , z = 2abt, t ∈ [0, 1].

Exercise 3.8.3. Compute the length of Cornu’s spiral

Z t Z t
πu2 πu2
x= cos du, y = sin du.
0 2 0 2

Exercise 3.8.4. Think of the rolling circle that produces the cycloid as a disk. What is the
track of a point on the disk that is not necessarily on the circle (i.e., the boundary of the
disk)? Find the formula for computing the length of this track.

Exercise 3.8.5. Suppose a line is wrapped around a circle. When the line is unwrapped
from the circle, the track of one point on the line is the involute of the circle. Let r be the
radius of the circle and let t be the unwrapped angle.
3.8. APPLICATION TO GEOMETRY 261

1. Find the parameterized formula for the involute.

2. Find the length of the involute as the line is unwrapped by half of the circle.

Figure 3.8.5: Involute of circle.

3.8.2 Area of Region

Being defined as area, the integration is naturally adapted to the computation of
area. We start with the area of region bounded by two functions.

Example 3.8.6. The curve y = x2 and the straight line y = x enclose a region over
0 ≤ x ≤ 1. The area of the region is the area below x subtracting the area below
x2 , which is
Z 1 Z 1 Z 1
2 1
xdx − x dx = (x − x2 )dx = .
0 0 0 6

x2 x

Figure 3.8.6: Region between x and x2 .

262 CHAPTER 3. INTEGRATION

Example 3.8.7. To compute the area of the region bounded by y = x2 −2x and y = x.
We denote the (positive) areas of the four indicated regions by A1 , A2 , A3 , A4 . Then
Z 2 Z 2 Z 3 Z 3
2
xdx = A1 , (x −2x)dx = −A2 , xdx = A3 +A4 , (x2 −2x)dx = A4 .
0 0 2 2

The area we are interested in is

A1 + A2 + A3 = A1 − (−A2 ) + (A3 + A4 ) − A4
Z 2 Z 2 Z 3 Z 3
2
= xdx − (x − 2x)dx + xdx − (x2 − 2x)dx
Z0 3 Z0 3 Z2 3 2
9
= xdx − (x2 − 2x)dx = [x − (x2 − 2x)]dx = .
0 0 0 2

x
x2 − 2x

A3

A4
A1

2 3
A2

Figure 3.8.7: Region between x and x2 − 2x.

The examples suggest that, if f (x) ≥ g(x) on [a, b], then the area of the region
between f and g over [a, b] is
Z b Z b Z b
f (x)dx − g(x)dx = (f (x) − g(x))dx.
a a a

In general, we can divide [a, b] into some intervals, such that on each interval, one of
the following happens: f (x) ≥ 0 ≥ g(x), f (x) ≥ g(x) ≥ 0, 0 ≥ f (x) ≥ g(x). Then
an argument similar to Example 3.8.7 shows that the total area is indeed given by
the formula above.

Example 3.8.8. The functions sin x and cos x intersect at many
 places and enclose
π 5π
many regions. One such region is over the interval , , on which we have
4 4
sin x ≥ cos x. The area of the region is
Z 5π √
4
(sin x − cos x)dx = 2 2.
π
4
3.8. APPLICATION TO GEOMETRY 263

0≥f ≥g f ≥0≥g f ≥g≥0

Z b
Figure 3.8.8: The area is (f (x) − g(x))dx.
a

5π
π 4 cos x
0 π π
4 2

sin x

Figure 3.8.9: Region between sin x and cos x.

Example 3.8.9. The region between the parabola y 2 − x = 1 and the straight line
x + y = 1 is between the functions
(√
x + 1, if − 1 ≤ x ≤ 0, √
f (x) = g(x) = − x + 1.
1 − x, if 0 ≤ x ≤ 3,

The area is
√ √
Z 3 Z 3 Z 0 Z 3 Z 3
9
f (x)dx − g(x)dx = x + 1dx + (1 − x)dx − (− x + 1)dx = .
−1 −1 −1 0 −1 2

Note that the region is obtained by rotating the region in Example 3.8.7. Natu-
rally the results are the same. The previous example actually suggests another way
of computing the area, by exchanging the roles of x and y.

Example 3.8.10. Consider f (x) = x5 + 2x2 − 2x − 3 and g(x) = x5 − x3 + x2 − 3. We

have
f (x) − g(x) = x3 + x2 − 2x = x(x − 1)(x + 2).
Therefore the two functions intersect at x = −2, 0, 1 and enclose two regions. The
first region is over [−2, 0], on which f (x) ≥ g(x). The second region is over [0, 1],
264 CHAPTER 3. INTEGRATION

x+y =1 y2 − x = 1

−1 3

−2

Figure 3.8.10: Region between a parabola and a straight line.

on which f (x) ≤ g(x). The areas of the regions are

Z 0
8
Area over [−2, 0] = (f (x) − g(x))dx = ,
−2 3
Z 1
5
Area over [0, 1] = (f (x) − g(x))dx = .
0 12
2
Note that we may change the functions to f (x) = x5 + 2x2 − 2x − 3 + ex and
2
g(x) = x5 − x3 + x2 − 3 + ex and get the same result. Although it is hard (actually
Z b Z b
impossible) to compute the exact values of f (x)dx and g(x)dx. Yet we can
a a
still compute the area.

Exercise 3.8.6. Compute area of the region with the given bounds.
√
1. y = x, y-axis, y = 1. 6. y = ex , y = x2 − 1, on [−1, 1].

2. y = ex , y = x, on [0, 1]. 7. y = log x, y 2 = x + 2, y = −1, y = 1.

3. y = log x, y = x, y = 0, y = 1. 8. x = y 2 − 4y, x = 2y − y 2 .

4. y = x2 , y = 2x − x2 . 9. y = 2x − x2 , x + y = 0.

5. y = sin x, y = cos x, on [0, π4 ]. 10. y = x, y = x + sin2 x, on [0, π].

Exercise 3.8.7. Explain that, if 0 ≥ f ≥ g on [a, b], then the area of the region between the
Z b
graphs of f and g over [a, b] is (f (x) − g(x))dx.
a

Exercise 3.8.8. Explain that, the area of the region between the graphs of f and g over
Z b
[a, b] is |f (x) − g(x)|dx, even when we may have f > g some place and f < g some
a
other place.
3.8. APPLICATION TO GEOMETRY 265

In practise, a region is often enclosed by a closed boundary curve (or several

closed curves if the region has holes). It is often more convenient to describe curves
by their parameterisations. For example, the unit disk is enclosed by the unit circle,
which can be conveniently parameterised as x = cos t, y = sin t, t ∈ [0, 2π].
A parameterisation of a curve can be considered as a movement along the curve,
and therefore imposes a direction on the curve. We will always make the standard
assumption that, as we move along a parameterised boundary curve, the region is
always on the left of curve. Figure 3.8.2 illustrates the meaning of the assumption.
For a region without holes, this means that the curve has counterclockwise direction.
The unit circle parameterisation above is such an example. If the region has holes,
then the “inside boundary components” should have clockwise direction.

C C1

C2

Figure 3.8.11: The region is always on the left of the boundary curve.

Consider a simple region in Figure 3.8.2, such that the boundary curve can be
divided into the graphs of two functions y = y1 (x) and y = y2 (x) for x ∈ [α, β].
Suppose the boundary curve has parameterisation φ(t) = (x(t), y(t)), t ∈ [a, b], such
that y1 and y2 correspond respectively to t ∈ [a, c] and t ∈ [c, b]. The direction of
the parameterisation satisfies our assumption.

y1 C

t=a
X t=b
t=c

y2

α β

Figure 3.8.12: Calculate the area by integrating along the boundary curve.

Z β
The area of the region is (y1 (x) − y2 (x))dx. We may use the parameterisation
α
of the boundary curve as the change of variable to get the following formula for the
266 CHAPTER 3. INTEGRATION

area
Z β Z β Z β
(y1 (x) − y2 (x))dx = y1 (x)dx − y2 (x)dx
α α α
Z a Z b
= y(t)dx(t) −
y(t)dx(t)
c c
Z b Z
=− y(t)dx(t) = − ydx.
a C

We note that the negative sign comes from our assumption

Z β of the direction. The
upper part y1 is supposed to positively contribute y1 (x)dx to the area of X.
α
However, the direction of y1 is leftward, opposite to the x-direction (the direction of
dx). This introduces a negative sign. Similarly, the lower part y2 should negatively
contribute to the area, and yet has the rightward direction, the same as the x-
direction. This also introduces a negative sign.
We may further use the integration by parts to get another formula for the area
Z b Z b Z
− y(t)dx(t) = −y(b)x(b) + y(a)x(a) + x(t)dy(t) = xdy.
a a C

Here we have x(a) = x(b) and y(a) = y(b) because the boundary curve is closed.
The positive sign on the right can be explained as follows. The area is supposed to
be the contribution from the right boundary part subtracting the contribution from
the left boundary part. From the picture, we see that the direction of the right part
is upward, the same as the y-direction (the direction of dy), and the direction of the
left part is downward, opposite to the y-direction.

Example 3.8.11. The boundary circle of the unit disk is parameterised by x(t) = cos t,
y = sin t, t ∈ [0, 2π]. Since the parameterisation satisfies our assumption, we may
use it to calculate the area of the unit disk
Z 2π Z 2π
− y(t)dx(t) = sin2 tdt = π.
0 0

Example 3.8.12. Consider the region enclosed by the Archimedean spiral x = t cos t,
y = t sin t, t ∈ [0, π], and the x-axis. The boundary of the region consists of the
spiral and the interval [−π, 0] on the x-axis. After checking that the direction of the
boundary satisfies the assumption, we get the area
Z π Z 0 Z π
0 1
− (t sin t)(t cos t) dt − 0dx = − (t sin t cos t − t2 sin2 t)dt = π 3 .
0 −π 0 6
3.8. APPLICATION TO GEOMETRY 267

t=π t=0

Figure 3.8.13: Archimedean spiral.

Exercise 3.8.9. Explain

Z that the area of the region on the left of Figure 3.8.2 may be
calculated by − ydx. You may need to break the boundary into four parts y1 , y2 , y3 , y4 .
C Z
Moreover, show that the area may also be calculated by xdy.
C

Exercise 3.8.10. ZExplain that

Z the area of
Z the region
Z on the right of Figure 3.8.2 may be
calculated by − ydx − ydx and xdy + xdy.
C1 C2 C1 C2

Exercise 3.8.11. Explain that,Zif the direction of the boundary curve C is opposite to our
assumption, then the area is ydx.
C

Exercise 3.8.12. Compute the areas of the regions enclosed by the curves.
x2 y 2
1. Ellipse + 2 = 1.
a2 b
2 2
2. Astroid x 3 + y 3 = 1.
p p
3. |x| + |y| = 1.

4. Hyperbola x2 − y 2 = 1 and x = a (a > 1).

5. Sprial x = et cos t, y = et sin t, t ∈ [0, π], and the x-axis.

6. One period of the cycloid in Example 3.8.5 and the x-axis.

3.8.3 Surface of Revolution

If we revolve a curve on the plane with respect to a straight line, we get a surface.
For example, the sphere is obtained by revolving a circle around any straight line
passing through the center of the circle, and the torus is obtained by revolving a
circle around any straight line not intersecting the circle.
Let (x(t), y(t)), t ∈ [a, b], be a parametrized curve in the upper half of the (x, y)-
plane (this means y(t) ≥ 0). To find the area of the surface obtained by revolving
268 CHAPTER 3. INTEGRATION

the curve around the x-axis, we let A(t) be the area of the surface obtained by
revolving the [a, t] segment of the curve around the x-axis. Again the subsequent
argument ignores the sign.
As the parameter is changed by ∆t, the change ∆A = A(t+∆t)−A(t) of the area
is the area of surface obtained by revolving the curve segment from (x(t), y(t)) to
(x(t + ∆t), y(t + ∆t)) = (x, y) + (∆x, ∆y). Since the curve segment is approximated
by the straight line connecting the two points, the area ∆A is approximated by the
area of the revolution of the straight line.

2
y
∆
2 +
t=b p ∆
x

y + ∆y
y

t=a

2π(y + ∆y)
2πy
x x + ∆x

Figure 3.8.14: Area of surface of revolution.

The revolution of the straight

p line can be expanded to lie on the plane. It is part
of the annulus of thickness ∆x + ∆y 2 . Moreover, the inner arc has length 2πy(t)
2

and the outer arc has length 2πy(t + ∆t). Therefore the area of the partial annulus
gives the approximation
1 p p
∆A ≈ (2πy(t) + 2πy(t + ∆t)) ∆x2 + ∆y 2 = π(y(t) + y(t + ∆t)) ∆x2 + ∆y 2 .
2
Dividing the change ∆t of the parameter, we get
s 2  2
∆A ∆x ∆y
≈ π(y(t) + y(t + ∆t)) + .
∆t ∆t ∆t

The approximation gets more refined as ∆t → 0. By taking the limit as ∆t → 0,

the approximation becomes an equality
s 2  2
0 ∆A ∆x ∆y
A (t) = lim = lim π(y(t) + y(t + ∆t)) +
∆t→0 ∆t ∆t→0 ∆t ∆t
p
= 2πy(t) x0 (t)2 + y 0 (t)2 .
3.8. APPLICATION TO GEOMETRY 269

This leads to
Z b p
area of surface of revolution = A(b) = 2π y(t) x0 (t)2 + y 0 (t)2 dt.
a
p
We note that ds = x0 (t)2 + y 0 (t)2 dt is used for computing the length of curve, and
we can write Z b
area of surface of revolution = 2π y(t)ds.
a

Here y is really the distance from the curve to the axis of revolution.

Example 3.8.13. The 2-dimensional sphere of radius r is obtained by revolving the

half circle
x = r cos θ, y = r sin θ, θ ∈ [0, π]
around the x-axis. Since the length of circular arc is given by ds = rdθ, the area of
the sphere is Z π
2π (r sin θ)rdθ = 4πr2 .
0

Example 3.8.14. The torus is obtained by revolving a circle on the upper half plane
around the x-axis. Let the radius of the circle be a and let the center of the circle
be (0, b). Then a < b and the circle may be parametrized as

x = a cos θ, y = a sin θ + b, θ ∈ [0, 2π].

The length is given by ds = adθ, so that the area of the torus is

Z 2π
2π (a sin θ + b)adθ = 4π 2 ab.
0

(a cos θ, a sin θ + b)
a
θ
b

Figure 3.8.15: Torus.

270 CHAPTER 3. INTEGRATION

Example 3.8.15. Take the segment y = x2 , x ∈ [0, 1] of the parabola √ in Example

3.8.2. If we revolve the parabola around the x-axis, then ds = 1 + 4x2 dx, the
distance from the curve to the axis of rotation (i.e., the x-axis) is x2 . Therefore the
area of the surface of revolution is
Z 1 √
2π x2 1 + 4x2 dx
0

With the help of Example 3.5.15 and the computation in Example 3.8.2, we have
Z 1 √
1 2 2√ 1 2
Z Z 
3 1
2 2
x 1 + 4x dx = 2
x 1 + x dx = (1 + x2 ) 2 − (1 + x2 ) 2 dx
0 8 0 8 0
2  !
3 Z 2
1 1 3 2 · 1
= x(1 + x2 ) 2  + 2
−1 (1 + x2 ) 2 dx
8 2 · 32 + 1 0
2 · 3
2
+ 1 0
√ √ √
  
1 2 3 2 1 1 1
= 5 −
2 5 + log(2 + 5) =√ − log(2 + 5).
8 5 5 2 5 40
2π π √
So the area is √ − log(2 + 5).
5 20

x−x 2
√
2

x
x2
x x x

Figure 3.8.16: Revolving a parabola segment around different axes.

√ If we revolve around the y-axis, then we get a paraboloid. We still have ds =

1 + 4x2 dx, but the distance from the curve to the axis of rotation (i.e., the y-axis)
is now x. Therefore the area of the paraboloid is
1
π √
Z 1 √
2
2 2 32 

2π x 1 + 4x dx = 2π (1 + 4x )  = (5 5 − 1).
0 3·8 0 6
Finally, if we revolve around the diagonal y = x, then the distance from the
x − x2
curve to the axis of rotation is √ , and the area is
2
x − x2 √ π √ √
Z 1  
1 2π π
2π √ 1 + 4x dxdx = √
2 (5 5 − 1) − √ + log(2 + 5)
0 2 2 6 5 20
√ 13 √ √
 
1 1
= 2π 5− − log(2 + 5) .
60 12 40
3.8. APPLICATION TO GEOMETRY 271

Example 3.8.15 shows how to adapt the formula for the area of the surface
of revolution to the more general case of any parametrized curve (x(t), y(t)) with
respect to a straight line αx + βy + γ = 0. Assume the curve is on the “positive
side” of the straight line

αx(t) + βy(t) + γ ≥ 0, for all t ∈ [a, b].

αx(t) + βy(t) + γ
Then the distance y(t) should be replaced by p . We still have
α 2 + β2
p
ds = x0 (t)2 + y 0 (t)2 dt. Therefore we get the general formula
b
p
(αx(t) + βy(t) + γ) x0 (t)2 + y 0 (t)2
Z
area of surface of revolution = 2π p dt.
a α2 + β 2

Exercise 3.8.13. Find the formula for the area of the surface of revolution of the graph of
a function y = f (x) around the x-axis. What about revolving around the y-axis? What
about revolving around the line x = a?

Exercise 3.8.14. Find the area of the surface of revolution.

1. y = x3 , x ∈ [0, 2], around x-axis. ex + e−x

7. y 2 = , x ∈ [−a, a], around
2
2. x2 = 2py, x ∈ [0, 1], around y-axis. x-axis.

3. y = ex , x ∈ [0, 1], around x-axis. 8. y 2 = x3 , x ∈ [0, 1], around x-axis.

4. y = ex , x ∈ [0, 1], around y-axis. x2 y 2
9. Ellipse + 2 = 1, around x-axis.
5. y = ex , x ∈ [0, 1], around y = 1. a2 b
2 2
6. y = tan x, x ∈ [0, π4 ], around x-axis. 10. Astroid x 3 + y 3 = 1, around x-axis.

Exercise 3.8.15. Find the area of the surface obtained by revolving one period of the cycloid
in Example 3.8.5 around the x-axis.

Exercise 3.8.16. Find the area of the surface obtained by revolving the involute of the circle
in Example 3.8.5 around the x-axis.

3.8.4 Solid of Revolution

If we revolve a region in the plane with respect to a straight line, we get a solid. For
example, the ball is obtained by revolving a disk around any straight line passing
through the center of the disk, and the solid torus is obtained by revolving a disk
around any straight line not intersecting the disk.
Consider the region G[a,b] (f ) for a function f (x) ≥ 0 over the interval [a, b]. To
find the volume of the solid obtained by revolving the region around the x-axis, we
272 CHAPTER 3. INTEGRATION

let V (x) be the volume of the part of solid obtained by revolving G[a,x] (f ) around
the x-axis. Then the change ∆V = V (x + ∆x) − V (x) is the volume of the solid
obtained by revolving G[x,x+∆x] (f ).

M
m
f (b)
∆x
f (a)

Figure 3.8.17: Volume of solid of revolution.

Let m = min[x,x+∆x] f and M = max[x,x+∆x] f . Then G[x,x+∆x] (f ) is sandwiched

between the rectangles [x, x + ∆x] × [0, m] and [x, x + ∆x] × [0, M ]. Therefore
the revolution of G[x,x+∆x] (f ) is sandwiched between the revolutions of the two
rectangles. The revolutions of rectangles are cylinders and have volumes πm2 ∆x
and πM 2 ∆x. Therefore we get
πm2 ∆x ≤ ∆V ≤ πM 2 ∆x.
This implies
∆V
πm2 ≤ ≤ πM 2 .
∆x
If f is continuous, then lim∆x→0 m = lim∆x→0 M = f (x). By the sandwich rule, we
get
∆V
V 0 (x) = lim = πf (x)2 .
∆x→0 ∆x
This leads to
Z b
volume of solid of revolution = V (b) = π f (x)2 dx.
a

Example 3.8.16. The 3-dimensional ball of radius r is obtained

√ by revolving the half
disk around the x-axis. The half disk is the region between r2 − x2 and the x-axis
over [−r, r]. Therefore the volume of the ball is
Z 1 √
4
π ( r2 − x2 )2 dx = πr3 .
−1 3
3.8. APPLICATION TO GEOMETRY 273

Example 3.8.17. The solid torus is obtained by revolving a disk in the upper half
plane around the x-axis. Let the radius of the disk be a and let the center√ of the disk
2 2
√ a < b and the disk is the region between y1 (x) = b + a − x and
be (0, b). Then
2 2
y2 (x) = b − a − x over the interval [−a, a]. The torus is the solid obtained by re-
volving G[−a,a] (y1 ) subtracting the solid obtained by revolving G[−a,a] (y2 ). Therefore
the volume of the torus is the volume of the first solid subtracting the second
Z a Z a Z a
2 2
π y1 (x) dx − π y2 (x) dx = π (y1 (x)2 − y2 (x)2 )dx
−a −a
Z−aa √ √
=π ((b + a2 − x2 )2 − (b − a2 − x2 )2 )dx
Z−aa √
=π 4b a2 − x2 dx
−a
Z π
2
= 4πb a2 cos2 tdt = 2π 2 a2 b.
− π2

annulus section

Figure 3.8.18: Solid of revolution of the region between two functions.

Example 3.8.17 shows that, if f ≥ g ≥ 0 on [a, b], then the volume of the solid of
revolution obtained by revolving the region between f and g over [a, b] around the
x-axis is
Z b
π (f (x)2 − g(x)2 )dx.
a

Now we extend the discussion before Example 3.8.11 about calculating the area
of a plan region by the integrating along the boundary curve. Suppose the region X
in Figure 3.8.2 lies in the upper half plane. Then similar to the earlier discussion,
274 CHAPTER 3. INTEGRATION

the volume of the solid obtained by rotating X around the x-axis is

Z β Z β Z β
2 2 2
π (y1 (x) − y2 (x) )dx = π y1 (x) dx − π y2 (x)2 dx
α α α
Za Z b
=π y(t)2 dx(t) − π
y(t)2 dx(t)
c c
Z b Z
2
= −π y(t) dx(t) = −π y 2 dx.
a C

So all the earlier discussion about the area can be applied to the volume of the solid
of revolution.

Example 3.8.18. The volume of the 3-dimensional ball of radius r is obtained by

revolving the half disk around the x-axis. The boundary of the half disk consists
of the half circle x = cos t, y = cos t, t ∈ [0, π], and the interval [−1, 1] on the x-
axis. Moreover, the parameterisation of the boundary curve satisfies our assumption.
Therefore the volume of the ball is
Z π Z 1 Z π
2 2 3 4
π (r sin t) d(r cos t) + π 0 dx = πr (1 − cos2 t)d(cos t) = πr3 .
0 −1 0 3

Example 3.8.19. We use the parameterisation x = a cos t, y = b + a sin t, t ∈ [0, 2π]

of the circle to calculate the volume of the torus in Example 3.8.17
Z 2π Z 2π
2
−π (b + a sin t) d(a cos t) = πa (b2 sin t + 2ab sin2 t + a2 sin2 t)dt
0 0
= 2π 2 a2 b.

Example 3.8.20. Consider the region enclosed by the Archimedean spiral and the
x-axis in Example 3.8.20. The volume of the solid obtained by revolving the region
around the x-axis is
Z π Z π
2 2
− (t sin t) d(t cos t) = − (t2 sin2 t cos t − t3 sin3 t)dt = π 3 − 4π.
0 0 3

Exercise 3.8.17. Find volume of the solid obtained by revolving the region in Exercise
3.8.12 around the x-axis.

Exercise 3.8.18. Use integration by

Z parts to explain that the volume of the solid of revolu-
tion can also be calculated by 2π xydy. Then use this formula to calculate the volumes
C
of the ball, the torus, and the solids obtained by revolving the regions in Exercise 3.8.12
around the x-axis.
The formula will be the “shell method” in Example 3.8.27.
3.8. APPLICATION TO GEOMETRY 275

Exercise 3.8.19. Explain that, if the direction of the boundary curve C is opposite
Z to our
assumption, then the volume of the solid of revolution around the x-axis is π y 2 dx.
C

Exercise 3.8.20. For the solid obtained by

Z revolving a region in the lower half plane around
the x-axis, how should the formula −π y 2 dx for the volume be modified?
C

Next we consider the general case of revolving a region X around a straight line
L : αx + βy + γ = 0. We assume X is on the “positive side” of L in the sense that
the parameterisation (x(t), y(t)) of the boundary curve C of X satisfies

αx(t) + βy(t) + γ ≥ 0, for all t ∈ [a, b].

Moreover, we still
Z assume that the direction of C satisfies our assumption. Then in
αx(t) + βy(t) + γ
the formula −π y 2 dx, y should be understood as the distance p
C α2 + β 2
from C to L, and dx should be understood as the progression

βdx − αdy βx0 (t) − αy 0 (t)

p = p dt
α2 + β 2 α2 + β 2

(β, −α)
along the direction p of L. Therefore the volume of the solid of revolution
α2 + β 2
is Z b
(αx(t) + βy(t) + γ)2 (βx0 (t) − αy 0 (t))
−π p dt
a ( α2 + β 2 )3
We note that the negative sign is due to the mismatch (See Figure 3.8.4) of the
direction of the boundary curve and the direction of the progression along L. In
general, we may determine the sign by comparing the direction of the parameter
and the direction of progression.

(α, β)
X L : αx + βy + γ = 0

progression along L

Figure 3.8.19: Revolving a region X around a line.

276 CHAPTER 3. INTEGRATION

For the special case that X is above the horizontal lineZ y = b ((α, β) = (0, 1)),
the volume of the solid of revolution around the line is −π (y − b)2 dx. If X is on
C
the right of the y-axis (i.e., the line x = 0, with (α, β) = (1, 0)), then the volume of
the solid of revolution around the y-axis is
Z b Z
2 0
−π x(t) (−y (t))dt = π x2 dy.
a C
0
The negative sign in front of y comes from the fact that the progression for the line
x = 0 goes downwards, the opposite of the y-direction. If X is on the right of the
vertical line
Z x = a, then the volume of the solid of revolution around the vertical
line is π (x − a)2 dy.
C

Example 3.8.21. Take the segment y = x2 , x ∈ [0, 1], of the parabola in Example
3.8.2. If we revolve the region X between the parabola and the x-axis around the
x-axis, then the volume of the solid is
Z x=1 Z 1
2 1
π y dx = π (x2 )2 dx = π.
x=0 0 5
If we revolve X around the y-axis, then the volume of the solid is
Z x=1 Z y=1
2 1
π x dy = π ydy = π.
x=0 y=0 2
Let Y be the region between the parabola and the vertical line x = 1. If we
revolve Y around the vertical line x = 1, then the volume of the solid is
Z x=1 Z 1
2 7
π (1 − x) dy = π (1 − x)2 d(x2 ) = π.
x=0 0 6
If we revolve Y around the y-axis instead, then the volume of the solid is
Z x=1 Z y=1
2 2 1
π (1 − x )dy = π (1 − y)dy = π.
x=0 y=0 2
Let Z be the region between the parabola y = x2 and the diagonal y = x. If we
revolve Z around the x-axis, then the volume of the solid is
Z x=1
2
π (x2 − (x2 )2 )dx = π.
x=0 15
If we revolve Z around the line x = −1, then the volume of the solid is
Z x=1 Z 1
2 2 2
π ((x + 1) d(x ) − (x + 1) dx) = π (x + 1)2 (2x − 1)dx
x=0
Z0 2
1
=π z 2 (2z − 3)dz = π.
1 2
3.8. APPLICATION TO GEOMETRY 277

If we revolve Z around the diagonal y = x, then to make sure the region is on

the positive side of the diagonal, we should write the diagonal as x − y = 0, with
x − x2
(α, β) = (1, −1). The distance between the parabola and the diagonal is √ .
2
(β, −α) (−1, −1)
The progression of the parabola in the direction p = √ of the line
α2 + β 2 2
is
−dx − dy −dx − d(x2 ) −(1 + 2x)
√ = √ = √ dx.
2 2 2
Therefore the volume of the solid is
Z x=1  2
x − x2 −(1 + 2x) 1
−π √ √ dx = √ π.
x=0 2 2 30 2

If we revolve Z around the line y = x − 1, we should write the line as −x + y + 1 = 0

for Z to be on the positive side, with (α, β) = (−1, 1). The distance from the
1 1 x − x2
diagonal and the parabola to the line are √ and √ − √ . The progressions
2 2 2
(β, −α) (1, 1)
of the diagonal and the parabola in the direction p = √ of the line are
α2 + β 2 2

dx + dx √ dx + d(x2 ) 1 + 2x
√ = 2dx and √ = √ dx.
2 2 2

Therefore the volume of the solid is

2 2 !
x=1 √ x − x2
Z  
1 1 1 + 2x 3
π √ 2dx − √ − √ √ dx = √ π.
x=0 2 2 2 2 10 2

Exercise 3.8.21. Let A ≤ f (x) ≤ B for x ∈ [a, b]. Find the formula for the volume of the
solid of revolution of the region between the graph of function f and y = A around the
line y = C, where C 6∈ (A, B).

Exercise 3.8.22. Find the formula for the volume of the solid obtained by revolving a region
X for which the parameterised boundary has the right direction.

1. X is on the left of the y-axis, around the y axis.

2. X is on the left of x = a, around x = a.

3. X is below y = b, around y = b.

4. X is on the negative side of x + y = 0, around x + y = 0.

5. X is on the negative side of x + y = 1, around x + y = 1.

278 CHAPTER 3. INTEGRATION

Exercise 3.8.23. Find the volume of the solid obtained by revolving the region between the
curve and the axis of revolution.
1. y = x3 , x ∈ [0, 2], around x-axis.
2. x2 = 2py, x ∈ [0, 1], around y-axis.
3. y = ex , x ∈ [0, 1], around x-axis.
4. y = ex , x ∈ [0, 1], around y-axis.
5. y = ex , x ∈ [0, 1], around x = 1.
h πi
6. y = tan x, x ∈ 0, , around x-axis.
4
ex + e−x
7. y 2 = , x ∈ [−a, a], around x-axis.
2
8. y 2 = x3 , x ∈ [0, 1], around x-axis.
x2 y 2
9. Ellipse + 2 ≤ 1, around x-axis.
a2 b
2 2
10. Astroid x 3 + y 3 ≤ 1, around x-axis.

Exercise 3.8.24. Find the volume of the solid of revolution.

1. Region bounded by y = x and y = x2 , around x-axis.
2. Region bounded by y = x and y = x2 , around x = 2.
3. Region bounded by y = x and y = x2 , around x = y.
4. Region bounded by y = x and y = x2 , around x + y = 0.
5. Region bounded by y 2 = x + 1 and x + y = 1, around x + y = 1.
6. Region bounded by y 2 = x + 1 and x + y = 1, around x = 3.
7. Region bounded by y 2 = x + 1 and x + y = 1, around y = 1.
8. Region bounded by y = log x, y = 0, y = 1, and y-axis, around y-axis.
9. Region bounded by y = cos x and y = sin x, around y = 1.
10. Triangle with vertices (0, 0), (1, 2), (2, 1), around x-axis.
x2 y 2
11. Ellipse + 2 ≤ 1, around y = b.
a2 b
x2 y 2
12. Ellipse + 2 ≤ 1, around bx + ay = 2ab.
a2 b
1
13. Region bounded by y = and the x-axis, around x-axis.
1 + |x|
14. Region bounded by y = e−|x| and the x-axis, around x-axis.
3.8. APPLICATION TO GEOMETRY 279

3.8.5 Cavalieri’s Principle

The formulae for the area of surface of revolution and the volume of solid of revolu-
tion follow from a more general principle.
In general, an n-dimensional solid X has n-dimensional size. For n = 1, X is a
curve and the size is the length. For n = 2, X is a region in R2 or more generally a
surface, and the size is the area. For n = 3, X is typically a region in R3 but can
also be a “3-dimensional surface” such as the 3-dimensional sphere in R4 , and the
size is the volume.
To find the size of an n-dimensional solid X, we may decompose X into sections
Xt of one lower dimension (i.e., Xt has dimension n − 1). For 2-dimensional X,
this means that X is decomposed into a one parameter family of curves. For 3-
dimensional X, this means that X is decomposed into a one parameter family of
surfaces. The decomposition is equidistant if the distance between two nearby pieces
does not depend on the location where the distance is measured. In this case, we
have the distance function s(t), such that the distance between the sections Xt and
Xt+∆t is ∆s = s(t + ∆t) − s(t). If X spans from distance s = a to distance s = b,
then
Z t=b Z b
size of X = size(Xt )ds = size(Xt )s0 (t)dt.
t=a a

A consequence of the formula is the following principle of Cavalieri: If two solids

X and Y have equidistant decompositions Xt and Yt , such that Xt and Yt have the
same size, and the distance between Xt and Xt0 is the same as the distance between
Yt and Yt0 , then X and Y have the same size.

Example 3.8.22. Let X be a region inside the plane. We decompose X by intersecting

with vertical lines Xx = X ∩ x × R. The decomposition is equidistant, with the x-
Z b
coordinate as the distance. Thus the area of X is length(Xx )dx. In the special
a
case X is the region between f (x) and g(x), where f (x) ≥ g(x), the section Xx is
the interval [g(x), f (x)] and has length f (x) − g(x). Then we recover the formula
Z b
(f (x) − g(x))dx in Section 3.8.2.
a

Example 3.8.23. Let X be a region inside R3 . We decompose X by intersecting

with vertical planes Xx = X ∩ x × R2 . The decomposition
Z is equidistant, with
b
the x-coordinate as the distance. Thus the volume of X is area(Xx )dx. In the
a
special case X is obtained by rotating the region between f (x) and g(x), where
f (x) ≥ g(x) ≥ 0, around the x-axis, the section Xx is the annulus with outer radius
f (x) and inner radius g(x). The section has area π(f (x)2 − g(x)2 ), and we get the
280 CHAPTER 3. INTEGRATION

Xx

a x b

Figure 3.8.20: Area of a region in R2 .

Z b
formula π (f (x)2 − g(x)2 )dx for the solid of revolution.
a

Xx

a x b

Figure 3.8.21: Volume of a solid in R3 .

Example 3.8.24. Let R be a region in the plane. Let P be a point not in the plane.
Connecting P to all points in R by straight lines produces the pyramid X with base
R and apex P .
We may put R on the (x, y)-plane in R3 and assume that P = (0, 0, h) lies in the
positive z-axis, where h is the distance from P to the plane. Let A be the area of R.
We decompose the pyramid by the horizontal planes, so that z is the distance. The
section Xz is similar to R, so that the area of Xz is 
proportional
2 to the square of its
h−z
distance h − z to P . We find the area of Xz to be A, and the volume of
h
the pyramid is
Z h 2
h−z 1
Adx = hA.
0 h 3

Example 3.8.25. Let X be the intersection of two round solid cylinders of radius 1 in
3.8. APPLICATION TO GEOMETRY 281

P
h
 2
h−z
z Xz area h
A

area A
0 R

Figure 3.8.22: Pyramid.

orthogonal position. We put the two cylinders in R3 , by assuming the two cylinders
to be x2 + y 2 ≤ 1 and x2 + z 2 ≤ 1. Then we decompose the solid by intersecting with
the
√ planes perpendicular to the x-axis. 2The section Xx is a square of side length
2 1 − x2 and therefore has area 4(1 − x ). The volume of the intersection solid X
is Z 1
16
4(1 − x2 )dx = .
−1 3
1 − x2
p

x −1
1 x −1

Figure 3.8.23: Orthogonal intersection of two cylinders.

Exercise 3.8.25. Explain the formula in Section 3.8.3 for the area of surface of revolution
by using suitable equidistant decomposition.

Exercise 3.8.26. Explain that if a solid is stretched by a factor A in the x-direction, by B

in the y-direction, and by C in the z-direction, then the volume of the solid is multiplied
by the factor ABC.

Exercise 3.8.27. Find the volume of solid.

282 CHAPTER 3. INTEGRATION

x2 y 2 z 2
1. Ellipsoid + 2 + 2 ≤ 1.
a2 b c
x2 y 2 z 2
2. Solid bounded by + 2 − 2 = 1 and z = ±c.
a2 b c
3. Intersection of the sphere x2 + y 2 + z 2 ≤ 1 and the cylinder x2 + y 2 ≤ x.

4. Solid bounded by x + y + z 2 = 1 and inside the first quadrant.

Exercise 3.8.28. Find the volume of solid.

1. A solid with a disk as the base, and the parallel sections perpendicular with the
base are equilateral triangles.

2. A solid with a disk as the base, and the parallel sections perpendicular with the
base are squares.

3. Cylinder cut by two planes, one is perpendicular to the cylinder and the other form
angle α with the cylinder. The two planes do not intersect inside the cylinder.

4. Cylinder cut by two planes forming respective angles α and β with the cylinder.
The two planes do not intersect inside the cylinder.

5. A wedge cut out of a cylinder, by two planes forming respective angles α and β with
the cylinder, such that the intersection of two planes is a diameter of the cylinder.

So far we used parallel lines and planes to construct the decomposition. We may
also use equidistant curves and surfaces to construct the decomposition.

Example 3.8.26. We may decompose a region X in the plane by concentric circles.

The decomposition is equidistant, with the radius r of the circles as the distance. The
section Xr consists of the points in X of distance r from the origin and is typically
an arc from angle φ(r) to angle ψ(r). The length of the arc Xr is (ψ(r) − φ(r))r, so
Z b
that the area of X is (ψ(r) − φ(r))rdr.
a

b
r
ψ(r)
a

Xr
φ(r)
O

Figure 3.8.24: Equidistant decomposition by concentric circles.

3.8. APPLICATION TO GEOMETRY 283

For example, for the disk centered at (1, 0) and of radius 1, we have φ =
r r
− arccos and ψ = arccos , r ∈ [0, 2]. The area of the disk of radius 1 is (taking
2 2
r
t = arccos , r = 2 cos t)
2
Z 2 Z 0 Z π
r 2
2r arccos dr = 2t(2 cos t)d(2 cos t) = 8 t cos t sin tdt
0 2 π
2
0
Z π Z π
2
=4 t sin 2tdt = u sin udu = π.
0 0

Example 3.8.27. Let X be a region in the right plane (i.e., the right side of y-axis).
Let Y be the solid obtained by revolving X around the y-axis. We may use the
cylinders centered at the y-axis to decompose Y . The decomposition is equidistant,
with x as the distance. Let Xx be the intersection of X with the vertical line
x × R. Then the section Yx is the cylinder obtained by revolving Xx around the
y-axis. The area of the section is 2πx(length of Xx ). Therefore the volume of the
Z b
solid of revolution is 2π x(length of Xx )dx. In particular, if X is the region
a
between functions f (x) and g(x), where f (x) ≥ g(x) on [a, b], then the volume is
Z b
2π x(f (x) − g(x))dx.
a

Yx Xx
x
X

Figure 3.8.25: Equidistant decomposition by concentric cylinders.

For example,
p consider the solid toruspin Example 3.8.17. The disk is the region
between x = a2 − (y − b)2 and x = − a2 − (y − b)2 , for y ∈ [b − a, b + a]. If we
use the formula above (note that x and y are exchanged), we get the volume of the
solid torus
Z b+a p Z a √
2π 2 2
y(2 a − (y − b) )dy = 4π (t + b) a2 − t2 )dt
b−a
Z−aa √
= 8π b a2 − t2 dt = 4π 2 a2 b.
0
284 CHAPTER 3. INTEGRATION

Exercise 3.8.29. Compute the volumes of the solids of revolution in Example 3.8.21 by
using the formula in Example 3.8.27.

Exercise 3.8.30. Compute the volumes of the solids of revolution in Exercise 3.8.23 by
using the formula in Example 3.8.27.

Exercise 3.8.31. Compute the volumes of the solids of revolution in Exercise 3.8.24 by
using the formula in Example 3.8.27.

Exercise 3.8.32. After Example 3.8.21, we presented the formula for computing the volume
of a solid obtained by revolving a region in R2 bounded by a parameterized curve. Can
you derive the similar formula by using the idea from Example 3.8.27?

Exercise 3.8.33. In Section 3.8.4 and Example 3.8.27, we have two ways of computing the
volume of a solid of revolution. For the following simple case, explain that the two ways
give the same result: Let f (x) be and invertible non-negative function on [0, a], such that
f (a) = 0 and both f (x) and f −1 (y) are continuously differentiable. The solid is obtained
by revolving the region between the graph of f and the two axis.

Example 3.8.28. Finally, we compute the size of high dimensional objects. Let αn
be the volume of the n-dimensional sphere S n of radius 1. Then

α0 = 2, α1 = 2π, α2 = 4π.

Moreover, the n-dimensional sphere of radius r has volume αn rn .

To compute αn , we decompose S n by intersecting with “horizontal hyperplanes”.
The hyperplanes are indexed by the angle t. The section at angle t is the (n − 1)-
dimensional sphere S n−1 of radius cos t, and form an equidistant decomposition. In
fact, the angle t can be used to measure the distance between h the sections. Since
n−1 π πi
the section at t has volume αn−1 cos t and the range of t is − , , we conclude
2 2
that Z π
2
αn = αn−1 cosn−1 tdt = 2αn−1 In−1 ,
− π2
Z π Z π
2 2
n−1
where In−1 = cos tdt = sinn−1 tdt has been computed in Example 3.5.14
0 0

(2k)! 22k (k!)2

I2k = 2k+1
π, I2k+1 = .
2 (k!)2 (2k + 1)!

Thus
π 2π
αn = 2αn−1 In−1 = 4αn−2 In−1 In−2 = 4αn−2 = αn−2 .
2(n − 1) n−1
3.9. POLAR COORDINATE 285

By the values of α1 and α2 , we conclude that

 n
(2π) 2
, if n is even,


2 · 4 · · · (n − 2)

αn = n−1
 2(2π) 2

 , if n is odd.
1 · 3 · · · (n − 2)

S n−1 of radius cos t

cos t

t
1

Figure 3.8.26: Decomposing n-dimensional sphere of radius 1.

Exercise 3.8.34. Let βn be the volume of the ball B n of radius 1.

1. Similar to Example 3.8.28, use the intersection with horizontal hyperplanes to derive
the relation between βn and βn−1 . Then use the special values β1 and β2 , and
Example 3.5.14 to compute βn .

2. Use the decomposition of B n by concentric (n − 1)-dimensional spheres to derive

the relation between βn and αn−1 . Then use Example 3.8.28 to find βn .
The two methods should give the same result.

Exercise 3.8.35. Suppose R is a region in Rn−1 with volume. Suppose P is a point in Rn

of distance h from Rn−1 . By connecting P to all points of R by straight lines, we get a
pyramid X with base R and apex P . Find the relation between the volumes of X and R.

3.9 Polar Coordinate

The polar coordinate locates a point on the plane by its distance r to the origin
and the angle θ indicating the direction from the viewpoint of origin. It is roughly
related to the cartesian coordinates (x, y) by
p y
x = r cos θ, y = r sin θ; r = x2 + y 2 , θ = arctan .
x
We say “roughly” because the relation between (x, y) and (r, θ) is not a one-to-one
correspondence. For example, the last formula literally restricts θ, as the value of
286 CHAPTER 3. INTEGRATION
 π π
inverse tangent function, to be within − , . In fact, the angle for a point in the
2 2
plane is unique only up to adding an integer multiple of 2π, and is more precisely
determined by
(x, y)
(cos θ, sin θ) = p .
x2 + y 2
Another way to say this is that θ is unique if we restrict to [0, 2π) (or [−π, π), etc.).
For the convenience of presenting polar equations, we also allow r to be negative,
by specifying that (−r, θ) and (r, θ + π) represent the same point. In other words,
(−r, θ) and (r, θ) are symmetric with respect to the origin. The cost of such extension
is more ambiguity in the polar coordinates of a point because all the following
represent the same point
(r, θ), (−r, θ ± π), (r, θ ± 2π), (−r, θ ± 3π), . . . .

3.9.1 Curves in Polar Coordinate

Example 3.9.1. The equation r = c is the circle of radius |c| centered at the origin.
The equation θ = c is a straight line passing through the origin.

r=c θ=c

d
r
α
θ

Figure 3.9.1: r = c, θ = c, and polar equation for general straight line.

The equation for a general straight line is

d
r= .
cos(α − θ)
Moreover, r = a cos θ is the circle of diameter a passing through the origin.

θ
r = cos θ

θ a r
O π 3π
0 2 2 2π

Figure 3.9.2: Circle r = a cos θ.

Exercise 3.9.1. Find the cartesian equation.

3.9. POLAR COORDINATE 287

1. r = 2. 3. r = sin θ. 5. r = tan θ sec θ.

2. r = −2. 4. r sin θ = 1. 6. r = cos θ + sin θ.

Exercise 3.9.2. Find the polar equation.

1. x = 1. 3. x + y = 1. 5. x2 + y 2 = x.

2. y = −1. 4. x = y 2 . 6. xy = 1.

Exercise 3.9.3. What is the polar equation of the curve obtained by flipping r = f (θ) with
respect to the origin? Then use your conclusion to find the curve r = − cos θ.

Exercise 3.9.4. What is the relation between the curves r = f (θ) and r = −f (θ + π)?

Exercise 3.9.5. What is the polar equation of the curve obtained by rotating r = f (θ) by
angle α? Then use your conclusion to answer the following.
1. What is the curve r = sin θ?

2. Find the polar equation for a general circle passing through the origin.

Exercise 3.9.6. Find the polar equation of a general circle.

Example 3.9.2. The Archimedean spiral is r = θ. Note that r < 0 when θ < 0, so
that a flipping with respect to the origin is needed when we draw the part of the
spiral corresponding to θ < 0. The symmetry with respect to the y-axis is due to
the fact that if (r, θ) satisfies r = θ, then (−r, −θ) also satisfies r = θ.
The Fermat’s spiral is r2 = θ. The symmetry with respect to the origin is due
to the fact that if (r, θ) satisfies r2 = θ, then (−r, θ) also satisfies r2 = θ.

(−r, −θ) (r, θ)

(−r, θ) (r, −θ)

Figure 3.9.3: Spirals r = θ and r2 = θ.

Example 3.9.3. The curve r = 1 + cos θ is a cardioid. Its clockwise rotation by 90◦ is
another cardioid r = 1 + sin θ. More generally, the curve r = a + cos θ is a limaçon.
288 CHAPTER 3. INTEGRATION

The curve intersects itself when |a| < 1 and does not intersect itself when |a| > 1.
The symmetry with respect to the x-axis is due to the fact that if (r, θ) satisfies
r = a + cos θ, then (r, −θ) also satisfies the equation.

1.5
π 1
2
0.4

π θ=0

3π
2
r = 1 + cos θ r = 1 + sin θ

Figure 3.9.4: Cardioids and limaçons r = a + cos θ, a = 0.4, 1, 1.5.

The cardioid originates from the following geometrical construction. Consider a

circle C of diameter 1 rolling outside of a circle A of equal diameter 1. This is the
same as the circle rolling inside a big circle B of diameter 3. The track traced by a
point on C is the cardioid. Note that the origin O of the polar coordinate should
be a point on A, not the center of A.

1 θ

sθ
co θ
θ θ
B θ
O 1 2

A C

Figure 3.9.5: Origin of the cardioid.

If we imagine the rolling circle C as part of a rolling disk D, and we fix a point
in D of distance d from the center of C. Then the track tranced by the point is the
limaçon r = 1 + 2d cos θ, with the origin of the polar coordinate being a point of
distance d form the center of A.

Example 3.9.4. The curve r = cos 2θ is the four-leaved rose, and r = cos 3θ is the
three-leaved rose. The circle r = cos θ can be considered as the one-leaved rose.
3.9. POLAR COORDINATE 289

Inh general, ithe curve r = cos nθ can be described as follows. For θ in the arc
π π
I = − , , the value of r goes from 0 to 1 and then back to 0, so that the
2n 2n
π
corresponding curve is one leaf occupying angle of the whole circle. This is the
h π πi h π πi n π
leaf in − , for n = 2 and in − , for n = 3. For θ in the second arc I + ,
4 4 6 6 n
π
we need to rotate this first leaf by angle and then flipping with respect to the
n
π
origin (because r becomes negative), which gives a leaf occupying I + + π. This
    n
5π 7π 7π 9π
is the leaf in , for n = 2 and in , for n = 3. For θ in the third arc
4 4 6 6
2π 2π
I+ , we get the leaf obtained by rotating the first leaf by angle (no flipping
n n
needed now because r becomes  non-negative
 again), which
 gives a leaf occupying
2π 3π 5π 3π 5π
I+ . This is the leaf in , for n = 2 and in , for n = 3. Keep
n 4 4 6 6
going, we see two distinct patterns depending on the parity of n.

3π
θ= 2 2π
θ= 3

4 3
θ=π 3 1 θ=0 1 θ=0

2 2

π
π
θ= 3
θ= 2

Figure 3.9.6: Four-leaved rose r = cos 2θ and three-leaved rose r = cos 3θ.

More
 generally,
 we may consider r = cos pθ. Again we get first leaf occupying
π π π 2π
I= − , , the second leaf occupying I + +π, the third leaf occupying I + ,
2p 2p p p
etc. The pattern could be very complicated, depending on whether p is rational or
irrational, and in case p is rational, the parity of the numerator and denominator of
p.
π
Finally, r = sin pθ is obtained by rotating r = cos 2θ by . We also get many
2p
leaved roses by other rotations.

Exercise 3.9.7. Describe the curve.

290 CHAPTER 3. INTEGRATION

Figure 3.9.7: Many leaved roses r = sin 2θ and r = sin 32 θ.

1. r = −θ. 11. r = 2 − cos θ. 1

21. r = cos θ.
3
2. r = θ + π. 12. r = cos θ + sin θ.
2
13. r = 1 + cos θ + sin θ. 22. r = cos θ.
3. r = 2θ. 3
4. r2 = −θ. 14. r = cos 4θ. 2 2
23. r = cos θ + sin θ.
15. r = 2 sin 5θ. 3 3
5. r2 = 4θ.
16. r = −3 sin 6θ. 24. r2 = sin 2θ.
6. r2 = θ + π.
17. r = sin 2θ − cos 2θ.
7. er = θ. 25. r2 = − cos 4θ.
18. r = sin 2θ + 2 cos 2θ.
8. rθ = 1. 1
4 26. r = 1 + 2 cos θ.
19. r = cos θ. 2
9. r = 2 + cos θ. 3
5 1
10. r = 2 + 3 cos θ. 20. r = sin θ. 27. r = 2 + cos θ.
3 2

3.9.2 Geometry in Polar Coordinate

The curve r = f (θ) for θ ∈ [α, β] is the parameterized curve
x = f (θ) cos θ, y = f (θ) sin θ, θ ∈ [α, β]
in the cartesian coordinate. The length of the curve is
Z βp Z βp
02 02
(f (θ) cos θ) + (f (θ) sin θ) dθ = f 2 + f 02 dθ.
α α

For the area in terms of polar coordinate, assume f ≥ 0 and consider the region
X[α,β] (f ) bounded by r = f (θ), θ ∈ [α, β], and the rays θ = α and θ = β. Using the
idea of Section 3.1.1, let A(θ) be the area of the region X[α,θ] (f ). Then the change
A(θ + h) − A(θ) is the area of X[θ,θ+h] (f ). Since X[θ,θ+h] (f ) is sandwiched between
fans of angle between θ, θ + h and radii m = min[θ,θ+h] f , M = max[θ,θ+h] f , we get
1 2 1
m h ≤ A(θ + h) − A(θ) ≤ M 2 h.
2 2
3.9. POLAR COORDINATE 291

Here the left and right sides are the known areas of the fans. The inequality is the
same as
1 2 A(θ + h) − A(θ) 1
m ≤ ≤ M 2.
2 h 2

θ+h
M θ
m

Figure 3.9.8: Estimate the change of area.

If f is continuous, then limh→0 m = limh→0 M = f (θ). By the sandwich rule, we

get
1
A0 (θ) = f (θ)2 .
2
Therefore the area of X[α,β] (f ) is
Z β
1
A(β) = f (θ)2 dθ.
2 α

Example 3.9.5. The cardioid r = 1 + sin θ has length

Z 2π p Z 2π p
2 02
(1 + sin θ) + (1 + sin θ) dθ = 2(1 + sin θ)dθ
0 0
Z − 34 π p π 
= 2(1 + cos 2t)d − 2t
1
4
π 2
1
Z
4
π Z π
=4 | cos t|dt = 4 cos tdt = 8.
− 34 π 0

The region enclosed by the cardioid has area

Z 2π Z 2π
1 2 1 3
(1 + sin θ) dθ = (1 + 2 sin θ + sin2 θ)dθ = π.
2 0 2 0 2

1
Example 3.9.6. Let p > . Then one leaf of the rose r = cos pθ is from the angle
2
292 CHAPTER 3. INTEGRATION

π π
− to the angle . The length of the leaf is
2p 2p
Z π p Z π q
2p 2p
2 02
(cos pθ) + (cos pθ) dθ = cos2 pθ + p2 sin2 pθdθ
π π
− 2p − 2p
Z π q
2
= p−2 cos2 t + sin2 tdt
− π2
Z 2π
1 p
= 1 + (p−2 − 1) cos2 tdt.
2 0

This is the elliptic integral in Example 3.8.1. Moreover, the area of the leaf is
Z π
1 2p π
(cos pθ)2 dθ = .
2 − 2p
π 4p

Example 3.9.7. The cardioid r = 1 + cos θ and the circle r = 3 cos θ intersect at
π
θ = ± . The area of the region outside the cardioid and inside the circle is
3
Z π
1 3
((3 cos θ)2 − (1 + cos θ)2 )dθ = π.
2 − π3

π
3

− π3

Figure 3.9.9: Ourside cardioid r = 1 + cos θ and inside circle r = 3 cos θ.

Example 3.9.8. We try to find the volume of the solid of revolution obtained by
revolving the region between the two leaves of the limaçon r = a + cos θ, 0 < a < 1,
around the x-axis. In the cartesian coordinate, the curve is parameterized by

x = (a + cos θ) cos θ, y = (a + cos θ) sin θ, θ ∈ [0, π].

Let θ = α at the origin O. Then the volume we are looking for is the volume of
the solid of revolution from θ = 0 to θ = α, subtracting the volume of the solid of
3.9. POLAR COORDINATE 293

revolution from θ = α to θ = π. As θ goes from 0 to α, we are moving opposite to

Z θ=α
the direction of the x-axis. Therefore the first volume is −π y 2 dx. As θ goes
θ=0
from α to π,Z we are moving in the direction of the x-axis. Therefore the second
θ=π
volume is π y 2 dx. We conclude that the volume we are looking for is
θ=α
Z θ=α Z θ=π Z θ=π
2 2
−π y dx − π y dx = −π y 2 dx
θ=0 θ=α
Zθ=0
π
= −π (a + cos θ)2 sin2 θd[(a + cos θ) cos θ]
Z0 π
= −π (a + cos θ)2 (1 − cos2 θ)(a + 2 cos θ)d(cos θ)
Z0 −1
4
= −π (a + t)2 (1 − t2 )(a + 2t)dt = πa(a2 + 1).
1 3

O
θ=π θ=0
θ=α

Figure 3.9.10: Revolving the region between two leaves of a limaçon.

Exercise 3.9.8. What is the length of lemiçon? What is the area of the region enclosed by
lemiçon? Note that for |c| > 1, we have two parts of the lemiçon and two regions.

Exercise 3.9.9. Find length of the part of the cardioid r = 1 + cos θ in the first quadrant.
Moreover, find the area of the region enclosed by this part and the two axes.

Exercise 3.9.10. Find the area of the region enclosed by strophoid r = 2 cos θ − sec θ.

Exercise 3.9.11. Find length.

1. r = θ, θ ∈ [0, π]. 2. r = θ2 , θ ∈ [0, π]. 3. r = eθ , θ ∈ [0, 2π].

Exercise 3.9.12. Find area.

1. Bounded by r = θ, θ ∈ [0, π] and the x-axis.
294 CHAPTER 3. INTEGRATION

2. Outside r = 1 and inside r = 2 cos θ.

3. Inside r = 1 and outside r = 2 cos θ.

4. Inside both r = 1 and r = 1 + cos θ.

5. Outside r = 3 sin θ and inside r = 2 − sin θ.

6. Inside both r = cos 2θ and r = sin 2θ.

7. Inside both r = 1 + c cos θ and r = 1 + c sin θ, |c| < 1.

8. Inside both r = 1 + c cos θ and r = 1 − c cos θ, |c| < 1.

9. Inside both r2 = cos 2θ and r2 = sin 2θ.

10. Outside r = 1 and inside r = 2 cos 3θ.

11. Between the two loops of r = 1 + 2 cos 3θ.

3.10 Application to Physics

3.10.1 Work and Pressure
Integration is also widely used to compute physical quantities. If under a constant
force F , an object moves by a distance d in the direction of the force, then the work
done by the force is F d. In general, however, the force may vary. For simplicity,
assume the object moves along the x-axis, from x = a to x = b, and a horizontal
force F (x) is applied when the object is at location x. Then the work done by
the force when the objects moves a little bit from x to x + ∆x is approximately
∆W ≈ F (x)∆x.
Similar to the earlier argument, let W (x) be the work done by the force when
the object moves from a to x. Since the work is additive, we have ∆W = W (x +
∆W
∆x) − W (x). The approximation ≈ F (x) becomes more accurate as ∆x → 0,
∆x
and we get an equality after taking the limit

W (x + ∆x) − W (x) ∆W
W 0 (x) = lim = lim = F (x).
∆x→0 ∆x ∆x→0 ∆x

This implies that the work done for the whole trip from a to b is
Z b
W (b) = F (x)dx.
a

Example 3.10.1. Suppose one end of spring is fixed and the other end is attached
to an object. In the natural position, when the spring is neither stretched nor
compressed, no force is exercised on the object. When the position of the object
3.10. APPLICATION TO PHYSICS 295

deviate from the natural position by x, however, Hooke’s law says that the spring
exercises a force F (x) = −kx on the object. Here k is the spring constant, and the
negative sign indicates that the direction of the force is opposite to the direction of
the deviation.
If the object starts at distance a from its natural position, then the work done
by the spring in pulling the object to its natural position is
Z a
k
kxdx = a2 .
0 2

Here we use the positive sign because the direction of movement is the same as the
direction of the force.

The argument about the work done by a force is quite typical. In general, if a
quantity is additive, then the quantity can be decomposed into small pieces. The
estimation of each small piece tells us the change of the quantity. The whole quantity
is then the integration of the change.
In the subsequent examples, we will only analyze a small piece of an additive
quantity. We will omit the limit part of the argument and directly write down the
corresponding integration.

Example 3.10.2. We want to find the work it takes to pump a bucket of liquid out
of the top of the bucket.

x
H
h ∆x

Figure 3.10.1: Bucket of liquid.

Suppose the bucket has base diameter r, top diameter R, and height H. Suppose
the liquid has density ρ and depth h. We decompose liquid into horizontal sections.
At distance x from the top, the section is a disk of radius r(x) satisfying

r(x) − r H −x
= .
R−r H
The liquid of thickness ∆x and at distance x from the top has (approximate) weight
gρπr(x)2 ∆x (g is the gravitational constant). The work it takes to lift this piece
296 CHAPTER 3. INTEGRATION

of liquid to the top of bucket is ∆W ≈ (gρπr(x)2 ∆x)x = πgρxr(x)2 ∆x. Since the
liquid spans from x = H − h to x = H, the total work needed is
Z H
πgρ H
Z
2
W = πgρ xr(x) dx = 2 x [(R − r)(H − x) + rH]2 dx
H−h H H−h
 
2 2 2 1 2 1 3 1 2 4
= πgρH R a b + a(2a − 3b)b + (1 − a)(1 − 3a)b − (1 − a) b ,
2 3 4

r h
where a = and b = .
R H

Example 3.10.3. We want to find the force exercised by water on a dam.

Let ρ be the density of water. At the depth x, the pressure of water is ρx per unit
area. Now suppose the dam is a vertical trapezoid with base length l, top length L,
and height H. We decompose dam into horizontal sections. At distance x from the
top, the section is a strip of height ∆x and length l(x) satisfying

l(x) − l H −x
= .
L−l H
The force exercised on the strip is ∆F ≈ (ρx)l(x)∆x. Since the water spans from
x = 0 to x = H, the total force
Z H Z H  
L−l 1
F =ρ xl(x)dx = ρ x L− x dx = ρH 2 (L + 2l).
0 0 H 6

x
H l(x)
∆x

Figure 3.10.2: Hydraulic dam.

Exercise 3.10.1. A spring has natural length a. If the force F is needed to stretch the
spring to length b, how muck work is needed to stretch the spring from the natural length
to the length b?

Exercise 3.10.2. A ball of radius R is full of liquid of density ρ. Due to the gravity, the
liquid leaks out of a hole at the bottom of the ball. How much work is done by the gravity
in draining all the liquid?
3.10. APPLICATION TO PHYSICS 297

Exercise 3.10.3. A circular disk of radius r is fully submerged in liquid of density ρ, such
that the center of the disk is at depth h. What is the force exercised by the liquid on one
side of the plate? Note that the plate may be inclined at some angle.

Exercise 3.10.4. A ball of radius r is fully submerged in liquid of density ρ, such that the
center of the disk is at depth h. What is the force exercised by the liquid on the ball?

Exercise 3.10.5. A cable of mass m and length l has a mass M tied to the lower end. How
much word is done in using the cable to lift the mass M to the top end of the cable?

Exercise 3.10.6. Newton’s law of gravitation says that two bodies with masses m and M
gmM
attract each other with a force F = , where d is the distance between the bodies.
d2
Suppose the radius of the earth is R and the mass is M . How much work is needed to
launch a satellite of mass m vertically to a circular orbit of height H? What is the minimal
initial velocity needed for the satellite to escape the earth’s gravity?

3.10.2 Center of Mass

Consider n masses m1 , m2 , . . . , mn distributed at the locations x1 , x2 , . . . , xn along
a straight line. The center of mass is
m1 x2 + m2 x2 + · · · + mn xn
x̄ = .
m1 + m2 + · · · + mn
The center has the physical meaning that the total moment of the system with
respect to x̄ is zero, or the system is balanced with respect to x̄.
Now suppose we have masses distributed throughout an interval [a, b], with the
density ρ(x) at location x. We partition the interval into small pieces

P : a = x0 < x1 < x2 < · · · < xn = b.

Then the system is decomposed into n pieces. The i-th piece can be approximately
considered as a mass mi = ρ(x∗i )(xi − xi−1 ) located at x∗i , for some x∗i ∈ [xi−1 , xi ].
The whole system is approximated by the system of n pieces, and has approximate
center of mass Pn
ρ(x∗i )(xi − xi−1 )x∗i
x̄P = Pi=1 n ∗
.
i=1 ρ(xi )(xi − xi−1 )
The denominator is the Riemann sum of the function xρ(x) and the numerator is the
Riemann sum of the function ρ(x) (see the beginning of Section 3.3.1). Therefore
as the partition gets more and more refined, the limit becomes the center of mass
Z b
xρ(x)dx
a
x̄ = Z b .
ρ(x)dx
a
298 CHAPTER 3. INTEGRATION

The center of mass can be extended to higher dimensions, simply by considering

each coordinate separately. For example, the system of n masses m1 , m2 , . . . , mn at
(x1 , y1 ), (x2 , y2 ), . . . , (xn , yn ) in the plane has the center of mass (x̄, ȳ) given by
P P
m i xi mi yi
x̄ = P , ȳ = P .
mi mi

Now consider masses distributed along a curve (x(t), y(t)), t ∈ [a, b], with the density
ρ(t) at location t. Take a partition P of [a, b]. The curve is approximated by
straight line segments connectingp(x(ti−1 ), y(ti−1 )) to (x(ti ), y(ti )). The i-th straight
line segment has length ∆si = (x(ti ) − x(ti−1 ))2 + (y(ti ) − y(ti−1 ))2 and can be
approximately considered as a mass mi = ρ(t∗i )∆si located at (x(t∗i ), y(t∗i )), for some
t∗i ∈ [ti−1 , ti ]. The whole system is approximated by the system of n pieces, and has
approximate center of mass
Pn Pn
i=1 (ρ(t∗i )∆si )x(t∗i ) i=1 (ρ(t∗i )∆si )y(t∗i )
x̄P = Pn ∗
, ȳ P = P n ∗
.
i=1 ρ(ti )∆si i=1 ρ(ti )∆si

As the partition gets more and more refined, the limit becomes the center of mass
Z b Z b
x(t)ρ(t)ds y(t)ρ(t)ds p
a a
x̄ = Z b
, ȳ = Z b
, ds = x0 (t)2 + y 0 (t)2 dt.
ρ(t)ds ρ(t)ds
a a

Example 3.10.4. For constant density ρ(x) = ρ distributed on the interval, the center
of mass is the middle point
Z b
xρdx 1
ρ (b2 − a2 ) a+b
x̄ = Za b
= 2 = .
ρ(b − a) 2
ρdx
a

If the density is ρ(x) = λ + µx, which is linearly increasing, then the center of mass
is Z b
x(λ + µx)dx
a 3λ(a + b) + 2µ(a2 + ab + b2 )
x̄ = Z b = .
3(2λ + µ(a + b))
(λ + µx)dx
a

Example 3.10.5. Consider the semi-circular curve of radius r and constant density ρ.
We have
x = r cos θ, y = r sin θ, ds = rdθ, 0 ≤ θ ≤ π,
3.10. APPLICATION TO PHYSICS 299

and the center of mass is

Z π Z π
(r cos θ)ρrdθ (r sin θ)ρrdθ
0 Z 0 2r
x̄ = π = 0, ȳ = Z π = .
π
ρrdθ ρrdθ
0 0

Exercise 3.10.7. Find the center of mass of the parabola y = x2 , x ∈ [0, 2], of constant
density.

Exercise 3.10.8.√ Find the center of mass of a triangle of constant density and with vertices
at (−1, 0), (0, 15) and (7, 0).

Exercise 3.10.9. Let m[a,b] and x̄[a,b] be the mass and the center of mass of a distribution
of masses on [a, b] with the density ρ(x). Let [a, b] = [a, c] ∪ [c, b] and similarly introduce
m[a,c] , m[c,b] , x̄[a,c] , x̄[c,b] . Show that the center of mass has the distribution property

m[a,c] x̄[a,c] + m[c,b] x̄[c,b]

m[a,b] = m[a,c] + m[c,b] , x̄[a,b] = .
m[a,c] + m[c,b]

Does the property extend to curves in R2 ?

300 CHAPTER 3. INTEGRATION
Chapter 4

Series

4.1 Series of Numbers

A series is an infinite sum

∞
X
an = a1 + a2 + a3 + · · · .
n=1

The following are some examples.

∞
X
rn = 1 + r + r2 + r3 + · · · + rn + · · · ,
n=0
∞
X
n = 1 + 2 + 3 + ··· + n + ··· ,
n=1
∞
X 1 1 1 1 1
= + + + ··· + + ··· ,
n=1
n(n + 1) 1·2 2·3 3·4 n(n + 1)
∞
X 1 1 1 1
p
= 1 + p + p + ··· + p + ··· ,
n=1
n 2 3 n
∞
X (−1)n+1 1 1 (−1)n+1
=1− + − ··· + + ··· ,
n=1
n 2 3 n
∞
X 1 1 1 1
= 1 + + + ··· + + ··· .
n=0
n! 1! 2! n!

Like sequences, series do not havePto start at n = 1. For example, it is more

convenient for the geometric series ∞ n
n=0 r to start at n = 0.

301
302 CHAPTER 4. SERIES

4.1.1 Sum of Series

P∞
Definition 4.1.1. The partial sum of a series n=1 an is
n
X
sn = ai = a1 + a2 + · · · + an .
i=1

If the partial sum converges, then the series converges and has sum (or value)
∞
X
an = lim sn .
n→∞
n=1

If the partial sum diverges, then the series diverges.

If finitely many terms in a series are modified, added or dropped, then the new
partial sum s0n and the original partial sum sn are related by s0n = sn+n0 + C for
some constants n0 and C. This implies that the convergence of series is not affected,
although the sum may be affected.
The arithmetic properties of the sequence limit implies
X X X X X
(an + bn ) = an + bn , can = c an .
P P an
However, there is no formula for an bn or .
bn

Example 4.1.1. Let sn = 1 + r + r2 + · · · + rn be the partial sum of geometric series

P ∞ n
n=0 r . Then

(1 − r)sn = (1 + r + r2 + · · · + rn ) − (r + r2 + r3 + · · · + rn+1 ) = 1 − rn+1 .

1 − rn+1
Therefore sn = and
1−r

∞
X  1 , if |r| < 1,
n
r = 1−r
n=0
diverges, if |r| ≥ 1.

P 1
Example 4.1.2. The computation in Example 1.3.1 gives the partial sum of .
n(n + 1)
     
1 1 1 1 1 1 1 1 1
+ +· · ·+ = 1− + − +· · ·+ − = 1− .
1·2 2·3 n(n + 1) 2 2 3 n n+1 n+1
Therefore ∞  
X 1 1
= lim 1 − = 1.
n=1
n(n + 1) n→∞ n+1
4.1. SERIES OF NUMBERS 303

P 1
Example 4.1.3. Example 2.7.5 shows that the partial sum of satisfies |sn − e| =
n!
e 1
, which implies ∞
P
|Rn (1)| ≤ n=0 = e. Exercise 1.3.18 gives an alternative
(n + 1)! n!
argument. Of course the argument, which uses the Lagrange form of the remainder
P xn
(Theorem 2.7.1), can be extended to the series . The partial sum satisfies
n!
ec e|x|
|sn − ex | = |Rn (x)| = |x|n+1 ≤ |x|n+1 , |c| < |x|.
(n + 1)! (n + 1)!
Since for fixed x, the right side converges to 0 as n → ∞, we conclude that
P∞ x n
n=0 = ex .
n!
n P
Exercise 4.1.1. Suppose the partial sum sn = . Find the series an and its sum.
2n + 1

Exercise 4.1.2. Decimal expressions for rational numbers have repeating patterns. For
example, we have
34 34 34
1.234 = 1.2343434 · · · = 1.2 + + + + ···
1000 100000 10000000
∞
34 X 1 34 1 611
= 1.2 + = 1.2 + = .
1000 100 n 1000 1 495
n=0 1−
100
1. Find rational expressions for 1.23, 1.230, 1.023.
5 43
2. Final the decimal based series representing the rational numbers , .
12 35

Exercise 4.1.3. What is the total area of infinitely many disks?

Exercise 4.1.4. The Sierpinski carpet is obtained from the unit square by successively
deleting “one third squares”. Find the area of the carpet.
304 CHAPTER 4. SERIES

Exercise 4.1.5. Two lines L and L0 form an angle θ at P . A boy starts on L at distance a
from P and walk to L0 along shortest path. After reaching L0 , he walks back to L along
shortest path. Then he walks to L0 again along shortest path, and keeps walking back and
forth. What is the total length of his trip?

Exercise 4.1.6. Find the area between curves y = xn and y = xn+1 and use this to conclude
1
that ∞
P
n=1 = 1.
n(n + 1)

Exercise 4.1.7. Compute the partial sum and the sum of series.

1.
P∞ n. P∞ 1
n=1 nr 4. n=2 .
n(n + 1)(n + 2)
P 2n + (−1)n 3n−1 
1

2. . P∞
5n+1 5. n=2 log 1 − 2 .
n
P∞ 1 P∞ n
3. n=0 . 6. n=1 .
(a + nd)(a + (n + 1)d) (n + 1)!
P∞ xn
Exercise 4.1.8. Suppose xn > 0. Compute n=1 .
(1 + x1 )(1 + x2 ) · · · (1 + xn )

Exercise 4.1.9. The Fibonacci sequence 1, 1, 2, 3, 5, . . . is defined recursively by a0 = a1 = 1,

an = an−1 + an−2 . Prove the following
∞ ∞
1 1 1 X 1 X an
= − , = 1, = 2.
an−1 an+1 an−1 an an an+1 an−1 an+1 an−1 an+1
n=2 n=2

Exercise 4.1.10. Use the Lagrange form of the remainder to show that the Taylor series
1
for converges for |x| < 1.
1−x

Exercise 4.1.11. Use the Lagrange form of the remainder to show that the Taylor series
for cos x and sin x converge for any x
∞ ∞
X x2n X x2n+1
(−1)n = cos x, (−1)n = sin x.
(2n)! (2n + 1)!
n=0 n=0

4.1.2 Convergence of Series

P
Theorem 4.1.2. If an converges, then limn→∞ an = 0.

This is a consequence of

lim an = lim (sn − sn−1 ) = lim sn − lim sn−1 = s − s = 0.

n→∞ n→∞ n→∞ n→∞
4.1. SERIES OF NUMBERS 305

P n
(−1)n ,
P P P
By the theorem, the series 1, n, diverge. By Example
P n+1
1.1.20, the series sin na converges if and only if a is an integer multiple of π.
If an ≥ 0 for sufficiently large n, then the partial sum sequence is increasing for
large n, and Theorem 1.3.2 becomes the following.
P
Theorem 4.1.3. If an ≥ 0, then an converges if and only if the partial sums are
bounded.
P 1
Example 4.1.4. The terms in the series = p are positive. Therefore the con-
n
vergence is equivalent to the boundedness of the partial sum. For p ≥ 2, we have
1 1
p
< and the following bound from Examples 1.3.1 and 4.1.2
n (n − 1)n
1 1 1 1 1 1 1
p
+ p + ··· + p ≤ 1 + + + ··· + = 2 − < 2.
1 2 n 1·2 2·3 (n − 1)n n

By Theorem 4.1.3, therefore, the series converges for p ≥ 2.

For p = 1, we used Cauchy criterion in Example 1.3.8 to show that the harmonic
P1
series diverges.
n

P (−1)n+1
Example 4.1.5. The even partial sum of the series is the partial sum of
n
the series
X 1 1
 
1
 
1 1

− = 1− + − + ··· .
2n − 1 2n 2 3 4
The terms of the series above are positive, and the partial sum has upper bound
   
1 1 1 1 1 1 1 1 1
1− + −· · ·+ − = 1− − −· · ·− − − < 1.
2 3 2n − 1 2n 2 3 2n − 2 2n − 1 2n
 
P 1 1
By Theorem 4.1.3, − converges. This means that the even partial
2n − 1 2n
P (−1)n+1 1
sum s2n of converges. By s2n+1 = s2n + , the odd partial sum
n 2n + 1
P (−1)n+1
s2n+1 converges to the same limit. Therefore converges.
n
P 1 P n
Exercise 4.1.12. Show the divergence of √ and .
n
a 2n − 1

Exercise 4.1.13. Determine convergence.

306 CHAPTER 4. SERIES

1 1 1 1 1 1 1 1
1. 1 + − + + − + ···. 3. 1 + 2
+ 3 + 4 + ···.
2 3 4 5 6 2 3 4
1 2 1 1 2 1 1 1
2. 1 + − + + − + ···. 4. 1 + √ +√ +√ + ···.
2 3 4 5 6 1·2 3·4 5·6

rn for 0 ≤ r < 1
P
Exercise 4.1.14. Use Theorem 4.1.3 to argue about the convergence of
P 1
and .
n!

Exercise 4.1.15. What is wrong with the following calculation?

0 = 0 + 0 + 0 + ···
= (1 − 1) + (1 − 1) + (1 − 1) + · · ·
= 1 − 1 + 1 − 1 + 1 − 1 + ···
= 1 + (−1 + 1) + (−1 + 1) + (−1 + 1) + · · ·
= 1 + 0 + 0 + 0 + · · · = 1.

We used the Cauchy criterion (Theorem 1.3.3) for the divergence of harmonic
P1
series . In general, applying the Cauchy criterion to the partial sum shows that
P n
an converges if and only if for any  ≥ 0, there is N , such that (since |sm − sn | is
symmetric in m and n, we may always assume n > m)

n > m > N =⇒ |sm − sn | = |am+1 + am+2 + · · · + an | < .

We may further modify the criterion by taking m + 1 to be m.

P
Theorem 4.1.4 (Cauchy Criterion). A series an converges if and only if for any
 > 0, there is N , such that

n ≥ m > N =⇒ |am + am+1 + · · · + an | < .

Theorem 4.1.2 is a special case of the Cauchy criterion by taking m = n.

4.2 Comparison Test

Z +∞
P
Series an are very much like improper integrals f (x)dx. The two can be
a
compared in two aspects. First the convergence of the two can
Z be compared, through
+∞
the integral test. Second all the convergence theorems for f (x)dx, such as the
a
comparison test, Dirichlet test and Abel test, have parallels for the convergence of
series.
4.2. COMPARISON TEST 307

4.2.1 Integral Test

Theorem 4.2.1 (Integral Test). Suppose f (x) is a decreasing function on [1, +∞)
satisfying limx→+∞ f (x) = 0. Then
Z n
f (1) + f (2) + · · · + f (n) = f (x)dx + γ + n ,
1

for a constant 0 ≤ γ ≤ f (1) and a decreasing sequence n converging to Z0. In partic-

+∞
P
ular, the series f (n) converges if and only if the improper integral f (x)dx
a
converges.

Let Z n
xn = f (1) + f (2) + · · · + f (n) − f (x)dx.
1
By f decreasing, we get
Z n
xn − xn−1 = f (n) − f (x)dx ≤ 0,
n−1
Z n
xn − f (n) = f (1) + · · · + f (n − 1) − f (x)dx ≥ 0
1
n−1 
X Z k+1 
= f (k) − f (x)dx ≥ 0.
k=1 k

The first inequality implies xn is decreasing, and the second inequality implies xn ≥
f (n) ≥ 0. Therefore lim xn = γ converges, and the theorem follows. We have
0 ≤ γ ≤ x1 = f (1).

1
Example 4.2.1. For p > 0, the function is decreasing and converges to 0 as
xp
P 1
x → +∞. By Theorem 4.2.1, therefore, the series converges if and only if the
Z +∞ np
dx
improper integral converges. By Example 3.7.3, this happens if and only
1 xp
if p > 1.
P1
Although the harmonic series diverges, Theorem 4.2.1 estimates the partial
n
sum
1 1 1
1 + + + · · · + = log n + γ + n ,
2 3 n
where n decreases and converges to 0, and

γ = 0.577215664901532860606512090082 · · ·

is the Euler-Mascheroni constant.

308 CHAPTER 4. SERIES

Example 4.2.2. For p > 0 and x > e, the integral test can be applied to the function
1 P 1
p
. We conclude that converges if and only if the improper
x(log x) Z n(log n)p
+∞
dx
integral converges. By Example 3.7.9, this means p > 1.
a x(log x)p

1 π2
Example 4.2.3. We will show that ∞
P
n=1 2 = in Example 4.5.13. In fact, for
n 6
1
even k, ∞ k
P
n=1 k can be calculated as a rational multiple of π . However, very little
n
is known about the sum for odd k. Still, we may use the idea of Theorem 4.2.1 to
estimate the remainder
Z +∞ ∞ Z k+1
X X∞ ∞ Z k+1
X Z +∞
f (x)dx = f (x)dx ≤ f (k) ≤ f (x)dx = f (x)dx.
n+1 k=n+1 k k=n+1 k=n k n

P∞ 1
For example, the 10-th partial sum of n=1 is
n3
1 1 1
s10 = 3
+ 3 + · · · + 3 = 1.197532 · · · .
1 2 10
By Z ∞ Z ∞
dx 1 dx 1
= = 0.005, = = 0.004132 · · · ,
10 x3 2(10)2 11 x3 2(11)2
we get

1.201664 · · · = 1.197532 · · · + 0.004132 · · ·

∞
X 1
≤ ≤ 1.197532 · · · + 0.005 = 1.202532 · · · .
n=1
n3

1
If we want to get the approximate value of ∞
P
n=1 3 up to the 6-th digit, then
n
we my try to find n satisfying
Z n+1
dx 2n + 1 1
= < < 0.000001.
n x3 2n2 (n + 1)2 n3

So we may take n = 100 and get

100 Z ∞ ∞ 100 Z ∞
X 1 dx X 1 X 1 dx
3
+ 3
≤ 3
≤ 3
+ 3
.
n=1
n 101 x n=1
n n=1
n 100 x

P 1
Exercise 4.2.1. Determine the convergence of .
n(log n)(log(log n))p
4.2. COMPARISON TEST 309

1 1
Exercise 4.2.2. Find suitable function f (n), such that the sequence 1+ √ +· · ·+ √ −f (n)
2 n
P∞ n 1
converges to a limit γ. Then express the sum of the series n=0 (−1) √ in terms of γ.
n

P∞ 1
Exercise 4.2.3. Estimate n=1 3 to within 0.01.
n2

4.2.2 Comparison Test

Theorem 4.2.2 (Comparison
P P Test). Suppose |an | ≤ bn for sufficiently large n. If
bn converges, then an also converges.

The test is completely parallel to the similar test (Theorem 3.7.1) for the con-
vergence of improper integrals, and can be proved similarly by using the Cauchy
criterion (Theorem 4.1.4). P P
For the special case bn = |an |, the test says that if |an | converges, then an
converges. In other words, absolute convergence implies convergence. We note that
the conclusion of the comparison test is always absolute convergence.

P log n log n 1
Example 4.2.4. Consider the series . If p ≤ 1, then ≥ . By the
np np n
P1 P log n
comparison test, the divergence of implies the divergence of .
n np
If p > 1, then choose q satisfying p > q > 1. We have

log n log n 1 1
= < for large n.
np np−q nq nq
log n
Here the inequality is due to the fact that p − q > 0 implies lim p−q = 0. By
n
P 1
Example 4.2.1, converges. Then by the comparison test, we conclude that
nq
P log n
converges.
np

log n 1
The key idea of the example above is to compare an = p
with bn = q by
n n
using the limit of their quotient. By

an log n
lim = lim p−q = 0,
bn n
an
we get < 1 for sufficiently large n. Since both an and bn are positive, we may
bn P
apply the comparison
P test to conclude that the convergence of bn implies the
convergence of an .
310 CHAPTER 4. SERIES

an
In general, if an , bn > 0 and lim = l converges, then by the comparison test,
P bn P
the convergence of bn implies the convergence of an . Moreover, if l 6= 0, then
bn 1 P P
we also have lim = , and we conclude that an converges if and only if bn
an l
converges.

P n + sin n
Example 4.2.5. For , we make the following comparison
n3 + n + 2
n + sin n
3
lim n + n + 2 = 1.
n→∞ 1
n2
P 1 P n + sin n
By the convergence of , we get the convergence of .
n2 n3 + n + 2
Similarly, by the comparison
2n + n2
√ √
5n−1 − n4 3n
lim n = 5,
n→∞ 2
√
5
n
2n + n2

P 2 P
and the convergence of √ , the series √ converges.
5 5n−1 − n4 3n
 x  
1 e 1
Example 4.2.6. By Example 2.5.14, we know 1 + −e = − +o . This
 n x 2x x
1
implies that for sufficiently large n, 1 + − e is negative and comparable to
n  n 
1 P1 P 1
. Since the harmonic series diverges, we conclude that 1+ −e
n n n
diverges.
+∞
| sin x|
Z
Example 4.2.7. By Example 3.7.14, we know that dx converges if and
1 xp Z +∞
| sin ax|
only if p > 1. By a change of variable, we also know that, for a 6= 0, dx
1 xp
converges if and only if p > 1. However, we cannot use the integral test (Theorem
P | sin na| | sin ax|
4.2.1) to get the similar conclusion for p
. The problem is that is
n xp
not a decreasing function.
| sin na| 1 P | sin na|
By p
≤ p and the comparison test, we know converges for
n n np
p > 1. The series also converges if a is a multiple of π, because all the terms are 0.
It remains to consider the case p ≤ 1 and a is not a multiple of π.
4.2. COMPARISON TEST 311
 
π π 3π
First assume 0 < a ≤ . For any natural number k, the interval kπ + , kπ +
2 4 4
π
has length and therefore must contain nk a for some natural number nk . Then
2
1
| sin nk a| ≥ √ , and for p ≤ 1,
2
∞ ∞ ∞ ∞
X | sin na| X | sin na| X | sin nk a| 1 X 1
≥ ≥ ≥√ .
n=1
np n=1
n k=1
nk 2 k=1 nk

3π 1 4 a a P1
By nk ≤ kπ + , we get ≥ > . Then by = +∞, we get
4 nk 4k + 3 π 4k k
P 1 P | sin na|
= +∞ and = +∞.
nk np
π
In general, if a is not an integer multiple of π, then there is b, such that 0 < b ≤
2
and either a + b or a − b is an integer multiple of π. Then we have | sin na| = | sin nb|,
P | sin na|
and we still conclude that diverges for p ≤ 1.
np

a2n converges. Moreover,

P P
Exercise 4.2.4. Show that if an > 0 and an converges, then
show that the converse is not true.
P an
a2n , then
P
Exercise 4.2.5. Show that if converges.
n

a2n and b2n converge, then (an + bn )2

P P P P
Exercise 4.2.6. Show that if an bn and
converge.

Exercise 4.2.7. Determine the convergence.

√
P 4n5 + 5n4 P 3n2 − 2n3 P 3n2 + (−1)n 2n3
1. . 2. √ . 3. .
3n2 − 2n3 4n5 + 5n4 4n5 + 5n4

Exercise 4.2.8. Determine the convergence, p, q, r, s > 0.

P 1 P nr + (log n)s P nr + (log n)s
1. . 3. . 5. .
n + (log n)q
p
np + (log n)q np (log n)q
P 1 P nr (log n)s P 1
2. . 4. . 6. .
n (log n)q
p np + (log n)q np (log n)q (log(log n))r

Exercise 4.2.9. Determine the convergence, b, d, p, q > 0.

P 1 P (c + nd)q P 1
1. . 2. . 3. .
(a + nb)p (a + nb)p (a + nb)p (c + nd)q
312 CHAPTER 4. SERIES

P (log(c + nd))q P 1 P (log(c + nd))q

4. . 5. . 6. .
(a + nb)p (a + nb)p (log(c + nd))q (a + nb)p

Exercise 4.2.10. Determine the convergence, p, q > 0.

q
np + a
P p  
1. ((n + a)r − (np + b)r ). 2.
P
−1 .
np + b

Exercise 4.2.11. Determine the convergence.

P 1 P 1 n2 P (log n)n
1. √ . 3. .
P
1+ log1 n 5. . 7. .
n n n (log n)n nn
P 1 P 1 P 1 P nlog n
2. 1 . 4. . 6. √ . 8. .
n 1+ n (log n)n n
log n (log n)n

Exercise 4.2.12. Determine the convergence, p, q > 0.

1 P n2 − n sin n
 
P 1
1. sin .
P
3. . 5. cos p − 1 .
n n3 + cos n n
P 1 1 P n2 − n sin n 1 P 1 1
2. sin q . 4. sin . 6. cos sin q .
n p n n3 + cos n n n p n

Exercise 4.2.13. Determine the convergence.

P 1 3n+1 P 5n−1 − n2 2n
1. .
P
5n − 1 2. . 3. .
5n−1 − n2 2n 3n+1

Exercise 4.2.14. Determine the convergence, a, b > 0.

P√ 1 n2
1. an + bn .
P
3. .
P
an + bn 5. .
nan + bn
P 1 P 1
2. √ . 4.
P
(an + bn )p . 6. √
n
.
an + bn an + bn

Exercise 4.2.15. Determine the convergence.

2 √ 2
xn . n. n2 xn .
P P P
1. 3. x 5.
2 √ q
nxn . n. np xn .
P P P
2. 4. nx 6.

Exercise 4.2.16. Determine the convergence.

P np P  1p   n
1. a . 4. nn − 1 . P an + b
6. .
P 1 cn + d
2. (a n − 1).
a log n n
n
a + (−1)n
    
1 1
n3
P P P
3. en − 1 − . 5. 1+ . 7. .
n n b + (−1)n
4.2. COMPARISON TEST 313

n2 a n2
r !
1 n+1 P P
8.
P
√ − log . 11.  . 14. cos .
1 n

n n n
a+
n 
P 1 p

 2 15. − log cos .
1 n n

2n
P
9. 1− . P n
n 12. .  a
log np sin q .
P
(n + a) (n + b)n+a 16.
n+b
n
2
1 2n−n
 
P P √
 √ √  P | cos na|
10. 1+ . 13. 2na− nb− nc . 17. .
n np

Exercise 4.2.17. Determine the convergence.

Z n+1 √ Z 1 Z 1
P
e− x n log x n
1. sin xdx. | sin x|p dx.
P P
3. dx. 5.
n 1 xp 0
n+1

1
n+1 1
xp
Z Z Z
sin x n
sin xn dx.
P P P
2. dx. 4. dx. 6.
n xp 0 1 + x2 0

P1
Exercise 4.2.18. Suppose an is a bounded sequence. Show that (an − an+1 ) converges.
n

Exercise 4.2.19. The decimal representations of positive real numbers are actually the sum
of series. For example,

π = 3.1415926 · · · = 3 + 0.1 + 0.04 + 0.001 + 0.0005 + 0.00009 + 0.000002 + 0.0000006 + · · · .

Explain why the expression always converges.

4.2.3 Special Comparison Test

We compare a series an with the geometric series rn , which we know converges
P P
if andPonly if |r| < 1. If |an | ≤ rn for some r < 1, then the comparison test implies
n
that n converges. We note that the condition |an | ≤ r for some r < 1 is the
ap
same as |an | ≤ r < 1.
n

TheoremP4.2.3 (Root Test). Suppose |an | ≤ rn for some r < 1 and sufficiently large
n. Then an converges.

4.2.8. To determine the convergence of (n5 + 2n + 3)xn , we note that

P
Example p
limn→∞ n |(n5p + 2n + 3)xn | = |x|. If |x| < 1, then we can pick r satisfying
p |x| < r <
5 n
1. By limn→∞ |(n + 2n + 3)x | < r and the order rule, we get P
n n
|(n + 2n + 3)xn | <
5

r for sufficiently large n. Then by the root test, we conclude that (n5 + 2n + 3)xn
converges for |x| < 1.
If |x| ≥ 1, then the term (n5 + 2n + 3)xn of the series does not converge to 0. By
Theorem 4.1.2, the series diverges for |x| ≥ 1.
314 CHAPTER 4. SERIES

The example suggests that, in practice, it is p

often more convenient to use the
limit version
p of the root test. Suppose limn→∞ |an | <
n
p 1. Then fix r satisfying
limn→∞ |an | < r < 1. By the order rule,
n
P we have
n
|an | < r for sufficiently
large n. pThen the root test showspthat an converges. On the other hand, if
limn→∞ n |an |P> 1, then we have n |an | > 1 for sufficiently large n. This implies
|an | > 1, and an diverges by Theorem 4.1.2.

Exercise 4.2.20. Determine the convergence, a, b > 0.

P (log n)n 6.
P q
np xn . P a 2n−n2
1. . 10. 1+ .
n2 n
 n
P np P an + b
2. . 7. .
(log n)n cn + d P np
11.  .
b n

1 a+
a n2
P
3. . P n
an + bn 8. 1+ .
P n n
4. (a + bn )p . n
a + (−1)n

P p n P a −n2 12.
P
n3 .
5. n x . 9. 1+ . b + (−1)n
n

Next we turn to another way of comparing series. Theorem 2.3.3 compares two
functions by comparing their derivatives (i.e., the changes of functions). Similarly,
we may compare two sequences an and bn by either comparing the differences an+1 −
an+1 bn+1
an and bn+1 − bn , or the ratios and . The comparison of ratio is especially
an bn
suitable for the comparison of series.
an+1 bn+1 aN
Suppose an , bn > 0, and ≤ for n ≥ N . Then for c = , the two
an bn bN
an+1 cbn+1
sequences an and cbn are equal at n = N , and ≤ implies that the second
an cbn
sequence has bigger change than the first one, at least for n ≥ N . This should imply
an ≤ cbn for n ≥ N . The following is the rigorous argument

aN +1 aN +2 an bN +1 bN +2 bn
an = aN ··· ≤ cbN ··· = cbn .
aN aN +1 an−1 bN bN +1 bn−1
P P
By the comparison test, if bn converges, then an converges.
 
 an+1 
Theorem 4.2.4 (Ratio Test). Suppose   ≤ bn+1 for sufficiently large n. If
P P an  bn
bn converges, then an converges.

We note that the assumption implies that the terms bn have the same sign for
sufficiently large n. By changing all bn to −bn if necessary, we may assume that
bn > 0 for sufficiently large n.
4.2. COMPARISON TEST 315

P (2n)! n
Example 4.2.9. The series x satisfies
(n!)2
(2n)! n
 
 
   x 
 = lim 2n(2n − 1) |x| = 4|x|.
 an   (n!)2 
lim 
  = lim 
n→∞ an−1 n→∞
 (2n − 2)! xn−1  n→∞ n2
  
 ((n − 1)!)2 

If 4|x| < 1, then we fix r satisfying 4|x| < r < 1. By the order rule, we have
n
 
 an 
  < r = r for large n.
 an−1  rn−1
bn = rn , Theorem 4.2.4 implies that
P P P
By comparing with the power series an
converges. If 4|x| > 1, then we get
 
 an 
 an−1  > 1 for large n.
 

P
Therefore |an | is increasing and does not converge to 0. By Theorem 4.1.2, an
diverges.
P (2n)! n 1 1
We conclude that 2
x converges for |x| < and diverges for |x| > . For
(n!) 4 4
1 P n
x = , we cannot compare with the geometric series r . Instead, we may try to
4
P 1 (2n)!
compare with p
. The terms an = > 0, and
n (n!)2 4n
1  
an 2n(2n − 1) 1 np p 1
= =1− , =1− +o .
an−1 4n 2 2n 1 n n
(n − 1)p
1 P 1 P
So we expect an to be comparable to 1 . Since 1 diverges, we expect an
n 2 n 2
diverges. For a rigorous argument, we wish to show that
1
an np
≥ for some p ≤ 1 and large n.
an−1 1
(n − 1)p
1
Of course this holds for p = 1 > . Therefore by the ratio test, we conclude that
2
P (2n)!
diverges.
(n!)2 4n
1 (2n)!
(−1)n
P
We will show in Example 4.3.4 that, for r = − , the series
4 (n!)2 4n
converges.
316 CHAPTER 4. SERIES

There are several generalisations

P wePcann make from the example. First, if we
apply the ratio test to the case bn = r is the geometric series, we find that
 
 an+1  X
  ≤ r < 1 for large n =⇒ an converges.
 an 

The limit version of this specialised ratio test is

 
 an+1  X
lim   < 1 =⇒ an converges.
n→∞ an 

On the other hand, the example also shows that

 
 an+1  X
lim 
  > 1 =⇒ an diverges.
n→∞ an 

Second, when the comparison with the geometric series does not work, we may
P 1
compare with . Suppose
np
 
 an+1  p
 an  ≤ 1 − n for some p > 1 and large n. (4.2.1)
 

We find q satisfying p > q > 1. Then the property above implies

    1
 an+1  q 1 nq
 an  ≤ 1 − n + o n = for large n.
 
1
(n − 1)q
1 P
By applying the ratio test to bn = p
, we conclude that an converges. The use
n
of criterion (4.2.1) for the convergence of series is the Raabe test.
The Raabe test also has the limit version. We note that (4.2.1) is equivalent to
  
 an+1 
n 1 −   ≥ p > 1 for large n.
an 

This will be satisfied if we can verify

  
 an+1 
lim n 1 − 
  > 1.
n→∞ an 

Exercise 4.2.21. Determine convergence.

4.3. CONDITIONAL CONVERGENCE 317

4 4 · 7 4 · 7 · 10 2 2·5 2·5·8
1. + + + ···. 3. + + + ···.
2 2 · 6 2 · 6 · 10 4 4 · 7 4 · 7 · 10
2 2 · 6 2 · 6 · 10 2 2·5 2·5·8
2. + + + ···. 4. + + + ···.
4 4 · 7 4 · 7 · 10 4 · 7 4 · 7 · 10 4 · 7 · 10 · 13

Exercise 4.2.22. Determine convergence.

P a(a + 1) · · · (a + n) P (a + c)(a + 2c)2 · · · (a + nc)n
1. . 4. .
b(b + 1) · · · (b + n) (b + d)(b + 2d)2 · · · (b + nd)n
P a(a + 1p ) · · · (a + np ) P (a1 + b1 1 + c1 12 ) · · · (a1 + b1 n + c1 n2 )
2. . 5. .
b(b + 1p ) · · · (b + np ) (a2 + b2 1 + c2 12 ) · · · (a2 + b2 n + c2 n2 )
P ap (a + c)p · · · (a + nc)p 1 P a(a + 1) · · · (a + n) c(c + 1) · · · (c + n)
3. . 6. .
bq (b + d)q · · · (b + nd)q nr b(b + 1) · · · (b + n) d(d + 1) · · · (d + n)

Exercise 4.2.23. Determine convergence. There might be come special values of r for which
you cannot yet make conclusion.
P (n!)2 n P n!(2n)! n P n! n P nn+1 n
1. r . 3. r . 5. r . 7. r .
(2n)! (3n)! nn (n + 1)!
P (3n)! n P nn n n! P (2n)! n
rn .
P
2. r . 4. r . 6. 8. r .
(n!)3 n! (n + 1)n n2n

Exercise 4.2.24. Prove the divergent part of the Raabe test.

an+1 1 P
1. If ≥ 1 − for sufficiently large n, then an diverges.
an n
  
 an+1 
2. If an > 0 and limn→∞ n 1 −    < 1, then P an diverges.
an 

4.3 Conditional Convergence

Like improper integrals, the comparison test implies that a series can have three
mutually exclusive possibilities:
P P
• Absolute Convergence: |an | converges ( =⇒ an converges).
P P
• Conditional Convergence: |an | diverges and an converges.
P P
• Divergence: an diverges ( =⇒ |an | diverges).

4.3.1 Test for Conditional Convergence

P (−1)n+1
The series in Example 4.1.5 is a typical conditionally convergent series.
n
P1
Its absolute value series is the harmonic series , which we know diverges. We
n
318 CHAPTER 4. SERIES

cannot apply the comparison test to the whole series because the conclusion of
comparison test is always absolute convergence. In fact, we applied Theorem 4.1.3
to the even partial sum of the series in Example 4.1.5. The following is an elaboration
of the idea of Example 4.1.5.

Proposition 4.3.1 (Leibniz Test). If an is decreasing for sufficiently large n and

limn→∞ an = 0, then (−1)n an converges.
P

The series
∞
X
(−1)n an = a0 − a1 + a2 − a3 + · · ·
n=0

is called an alternating series. If an is decreasing (for all n), then the odd partial
sum

s2n+1 = (a0 − a1 ) + (a2 − a3 ) + (a4 − a5 ) + · · · + (a2n − a2n+1 )

= a0 − (a1 − a2 ) − (a3 − a4 ) − · · · − (a2n−1 − a2n ) − a2n+1

is increasing and has upper bound a0 . Therefore lim s2n+1 converges. By s2n =
s2n+1 − a2n+1 and lim a2n+1 = 0, we have lim s2n = lim s2n+1 and therefore the whole
partial sum sequence converges.

Example 4.3.1. By the Leibniz test, the series

∞
X (−1)n+1 1 1
=1− + − ···
n=1
np 2p 3p

converges for p > 0. By Example 4.2.1, the series absolutely converges for p > 1,
conditionally converges for 0 < p ≤ 1, and diverges for p ≤ 0.
P a n p
Example 4.3.2.
P a n Consider the series n b . By limn→∞
n
|na bn | = |b| and the root
test, n b absolutely converges Pfora |b| < 1 and diverges for |b| > 1.
P is n na , which converges if and only if a < −1. If
If b = 1, then the series
b = −1, then the series is (−1) n , which by Example 4.3.1 converges if and only
if a < 0.
In conclusion, the series na bn absolutely converges for either |b| < 1, or a < −1
P
and |b| = 1, conditionally converges for −1 ≤ a < 0 and b = −1, and diverges
otherwise.

n2 + a
(−1)n 3
P
Example 4.3.3. Consider the alternating series . The corresponding
n +b
absolute value series is comparable to the harmonic series and therefore diverges.
x2 + a
If we can show that f (x) = 3 is decreasing, therefore, then the Leibniz test
x +b
4.3. CONDITIONAL CONVERGENCE 319

implies the conditional convergence. By

0 −x4 − 3ax2 + 2xb

f (x) = ,
(x3 + b)2

the function indeed decreases for sufficiently large x.

P (2n)! n
Example 4.3.4. In Example 4.2.9, we determined the convergence of x for
(n!)2
1 1
all x except x = − . For x = − , the series is alternating, and |an | is decreasing
  4 4
 an  1
by   =1− < 1. If we can show that an converges to 0, then we can apply
an−1  2n
the Leibniz test.
1
We compare with the ratio of |an | with the ratio of p
n

  1  
 an  1 np p 1
 an−1  = 1 − 2n ≤ =1− +o .
 
1 n n
(n − 1)p

This happens if we pick p = 0.4 and n is sufficiently large. The comparison of

c
the ratio implies |an | < 0.4 for a constant a and sufficiently large n. This further
n
(2n)!
implies that lim |an | = 0. By the Leibniz test, we conclude that (−1)n
P
(n!)2 4n
converges.
P (2n)! n
Combined with Examples 4.2.9, we conclude that x absolutely con-
(n!)2
1 1
verges for |x| < , conditionally converges for x = − , and diverges otherwise.
4 4
 
an p 1
Exercise 4.3.1. Suppose an > 0 and = 1− +o for some p > 0. Prove that
an−1 n n
lim an = 0 and (−1)n an converges.
P

Exercise 4.3.2. Determine absolute or conditional convergence.

4 4 · 7 4 · 7 · 10 2 2·5 2·5·8
1. − + − ···. 3. − + − ···.
2 2 · 6 2 · 6 · 10 4 4 · 7 4 · 7 · 10
2 2 · 6 2 · 6 · 10 2 2·5 2·5·8
2. − + − ···. 4. − + − ···.
4 4 · 7 4 · 7 · 10 4 · 7 4 · 7 · 10 4 · 7 · 10 · 13

Exercise 4.3.3. Determine absolute or conditional convergence.

320 CHAPTER 4. SERIES

a(a + 1r ) · · · (a + nr ) (a + 1)(a + 2)2 · · · (a + n)n

(−1)n
P
(−1)n
P
1. . 2. .
b(b + 1r ) · · · (b + nr ) (b + 1)(b + 2)2 · · · (b + n)n

Exercise 4.3.4. Determine absolute or conditional convergence.

P (−1)n n2 P rn 7.
P p
(−1)n nq an .
1. . 4. .
n3 + n + 2 np
P (−1)n
n2 + sin n rn 8. .
p+ logq n
P P
2. . 5. . n
(−1)n n3 + n + 2 np (log n)q
P (−1)n n(n−1) 1
P (−1)n nn+p
3. . 6.
P
(−1) 2 . 9. n .
(a + nb)p np (an2 + bn + c) 2 +q

Exercise 4.3.5. Determine the absolute or conditional convergence for the undecided cases
in Exercise 4.2.23.

Like the convergence of improper integrals, we also have the analogues of the
Dirichlet and Abel tests.
P
Proposition 4.3.2 (Dirichlet Test). Suppose the partial
P sum of an is bounded.
Suppose bn is monotonic and limn→∞ bn = 0. Then an bn converges.
P
Proposition 4.3.3 (Abel
P Test). Suppose an converges. Suppose bn is monotonic
and bounded. Then an bn converges.

P | sin na|
Example 4.3.5. In Example 4.2.7, we showed that diverges when a is
np
not an integer multiple of π. Then Example 3.7.14 suggests that the series should
converge conditionally.
By the Dirichlet test, if we can show that the partial sum

sn = sin a + sin 2a + · · · + sin na

is bounded, then the series converges. By

a   a  a    a  a 
2sn sin = cos a − − cos a + + cos 2a − − cos 2a +
2  2 2  2 2
a  a 
+ · · · + cos na − − cos na +
2 2
a  a
= cos − cos na + ,
2 2
we get
1
|sn | ≤  a  .
sin 
2
The right side is a bound for the partial sums in case a is not a multiple of π.
4.3. CONDITIONAL CONVERGENCE 321

Exercise 4.3.6. Derive the Leibniz test and the Abel test from the Dirichlet test.

P an P an
Exercise 4.3.7. Prove that if p
converges, then converges for any q > p.
n nq

Exercise 4.3.8. Determine the absolute and conditional convergence.

P cos na sin na P sin3 na

1. .
P
(n + b)p 3. . 5. .
np (log n)q np (log n)q
cos na sin2 na n(n−1) sin na
(−1)n
P
(−1)n
P P
2. . 4. . 6. (−1) 2 .
n+b np np

Exercise 4.3.9. Determine absolute or conditional convergence.

P 1 (−1)n (−1)n
1. .
P P
3. . 5. .
n + (−1)n n2 (n + (−1)n )p np + (−1)n
n(n−1)
P (−1)n P (−1) 2
2. √ . 4. √ .
( n + (−1)n )p n + (−1)n

Exercise 4.3.10. Determine the convergence.

√
a log n n an + b n
   
sin n2 + aπ.
P
1. 2.
P n
(−1) 1 − . 3.
P
log .
n cn + d
√
P (−1)[ n]
Exercise 4.3.11. Let [x] be the biggest integer ≤ x. Determine the convergence of
np
P (−1) [log n]
and .
np

4.3.2 Absolute v.s. Conditional

The distinction between absolute and conditional convergence has implications on
how we can manipulate series. For example, we have a + b + c + d = c + b + a + d.
However, we need to be more careful in rearranging orders in an infinite sum.

Theorem 4.3.4. The sum of an absolutely convergent series does not depend on the
order. On the other hand, given any conditionally convergent series and any number
s, it is possible to rearrange the order so that the sum of the rearranged series is s.

∞
X (−1)n+1
Example 4.3.6. We know from Example 4.3.1 that the series converges
n=1
n
conditionally. The partial sum can be estimated from the partial sum of the har-
322 CHAPTER 4. SERIES

monic series in Example 4.2.1

1 1 1 1 1
1 − + − + ··· + −
2 3 4 2n − 1 2n   
1 1 1 1 1 1 1 1 1
= 1 + + + + ··· + + −2 + + + ··· +
2 3 4 2n − 1 2n 2 4 6 2n
= (log 2n + γ + 2n ) − (log n + γ + n ) = log 2 + (2n − n ).

This implies
1 1 1
1− + − + · · · = log 2.
2 3 4
If the terms are rearranged, so that one positive term is followed by two negative
terms, then the partial sum is

1 1 1 1 1 1 1 1
1 − − + − − + ··· + − −
2 4 3 6 8 2n − 1 4n − 2 4n
1 1 1 1 1
= 1 + + + + ··· + +
2 3 4 2n − 1 2n
   
1 1 1 1 1 1 1 1
− + + + ··· + − + + + ··· +
2 4 6 2n 2 4 6 4n
1 1
= (log 2n + γ + 2n ) − (log n + γ + n ) − (log 2n + γ + 2n )
2 2
1 1
= log 2 + (2n − n ).
2 2
If two positive terms are followed by one negative term, then the partial sum is

1 1 1 1 1 1 1 1
1 + − + + − + ··· + + −
3 2 5 7 4 4n − 1 4n − 3 2n
1 1 1 1
= 1 + + + + ··· +
2 3 4 4n
   
1 1 1 1 1 1 1 1
− + + + ··· + − + + + ··· +
2 4 6 4n 2 4 6 2n
1 1
= (log 4n + γ + 4n ) − (log 2n + γ + 2n ) − (log n + γ + n )
2 2
3 1
= log 2 + (4n − 2n − n ).
2 2
We get

1 1 1 1 1 1
1− − + − − + ··· = log 2,
2 4 3 6 8 2
1 1 1 1 1 3
1 + − + + − + ··· = log 2.
3 2 5 7 4 2
4.3. CONDITIONAL CONVERGENCE 323

1 1 1
Exercise 4.3.12. Rearrange the series 1 − + − + · · · so that p positive terms are
2 3 4
followed by q negative terms and the pattern repeated. Show that the sum of new series
1 p
is log 2 + log .
2 q

1 1 1 1 1 1
Exercise 4.3.13. Show that 1 − √ + √ − √ + · · · converges, but 1 − √ − √ + √ −
2 3 4 2 4 3
1 1 1 1 1
√ − √ + √ − √ − √ + · · · diverges.
6 8 5 10 12

Another distinction between absolute and conditional convergence is reflected on

the product of two series.
∞
X ∞
X ∞
X
Theorem 4.3.5. Suppose an and bn converge absolutely. Then ai bj also
n=1 n=1 i,j=1
converge absolutely, and
∞ ∞
! ∞
!
X X X
ai bj = an bn .
i,j=1 n=1 n=1

∞
X
Note that the infinite sum ai bj is a “double series” with two indices i and j.
i,j=1
There are many ways of arranging this series into a single series. For example, the
following is the “diagonal arrangement”
X
(ab)k = a1 b1 + a1 b2 + a2 b1 + · · ·
+ a1 bn−1 + a2 bn−2 + · · · + an−1 b1 + · · · ,
and the following is the “square arrangement”
X
(ab)k = a1 b1 + a1 b2 + a2 b2 + a2 b1 + · · ·
+ a1 bn + a2 bn + · · · + an bn−1 + an bn + an bn−1 + · · · + an b1 + · · · .
Under the condition of the theorem, the series is supposed to converge absolutely.
Then by Theorem 4.3.4, all arrangements give the same sum.

xn
Example 4.3.7. We know from Example 4.1.4 that ∞
P
n=0 absolutely converges to
x
n!
e . By Theorem 4.3.5, we have
∞
! ∞ ! ∞ ∞
n X yn
X x X xi y j X X xi y j
ex ey = = =
n=0
n! n=0
n! i,j=0
i!j! n=0 i+j=n
i!j!
∞ n ∞
X 1 X n! X 1
= xi y j = (x + y)n = ex+y .
n=0
n! i=0 i!(n − i)! n=0
n!
324 CHAPTER 4. SERIES

n n

(ab)7 (ab)10 (ab)11 (ab)12 (ab)13

(ab)4 (ab)8 (ab)5 (ab)6 (ab)7 (ab)14

(ab)2 (ab)5 (ab)9 (ab)2 (ab)3 (ab)8 (ab)15

(ab)1 (ab)3 (ab)6 (ab)10 (ab)1 (ab)4 (ab)9 (ab)16

m m

Figure 4.3.1: Diagonal and square arrangements.

In the second to the last equality, we used the binomial expansion.

Exercise 4.3.14. If you take the product of a geometric series with itself, what conclusion
can you make?

P (−1)n
Exercise 4.3.15. Suppose √ = l. Show that the square arrangement of the product
n
of the series with itself converges to l2 . What is the sum of the diagonal arrangement?

4.4 Power Series

A power series at x0 is
∞
X
an (x − x0 )n .
n=0

By a simple change of variable, it is sufficient to consider power series at 0

∞
X
an x n .
n=0

4.4.1 Convergence of Taylor Series

If f (x) has derivatives of arbitrary order at x0 , then the high order approximations
of the function gives us the Taylor series
∞
X f (n) (x0 ) f 00 (x0 ) f (n) (x0 )
(x−x0 )n = f (x0 )+f 0 (x0 )(x−x0 )+ (x−x0 )2 +· · ·+ (x−x0 )n +· · · .
n=0
n! 2! n!

The partial sum of the series is the n-th order Taylor expansion Tn (x) of f (x) at x0 .
4.4. POWER SERIES 325

In Example 4.1.3, we used the Lagrange form of the remainder Rn (x) = f (x) −
Tn (x) (Theorem 2.7.1) to show that the Taylor series of ex converges to ex . Exer-
cise 4.1.11 further showed that the Taylor series of sin x and cos x converge to the
trigonometric functions.

P∞ (−1)n+1 n
Example 4.4.1. Consider the Taylor series n=0 x of log(1 + x). We have
n
dn (−1)n−1 (n − 1)!
log(1 + x) = ,
dxn (1 + x)n

and the Lagrange form of the remainder gives

1 n! n+1 |x|n+1
|Rn (x)| = |x| = ,
(n + 1)! |1 + c|n+1 (n + 1)|1 + c|n+1

1
where c lies between 0 and x. If − ≤ x ≤ 1, then |x| ≤ |1 + c|, and we get
2
1
|Rn (x)| < . Therefore limn→∞ Rn (x) = 0, and the Taylor series converges to
n+1
log(1 + x).
The Taylor series is the harmonic series at x = −1 and therefore diverges. For
|x| > 1, the terms of the Taylor series diverges to ∞, and therefore the Taylor series
1
also diverges. The remaining case is −1 < x < − .
2
In Exercise 4.4.2, a new form of the remainder is used to show that the Taylor
series actually converges to log(1 + x) for all −1 < x ≤ 1.

Example 4.4.2. The Taylor series of (1 − x)−1 is the geometric series ∞ n

P
n=0 x . Ex-
ample 4.1.1 shows that the Taylor series converges to the function for |x| < 1 and
diverges for |x| ≥ 1.
p(p − 1) · · · (p − n + 1) n
The Taylor series of (1 + x)p is ∞
P
n=0 x . The Lagrange
n!
form of the remainder gives

|p(p − 1) · · · (p − n)|
|Rn (x)| = |1 + c|p−n−1 |x|n
(n + 1)!
|p(p − 1) · · · (p − n)| |1 + c|p−1 |x|n
= ,
(n + 1)! |1 + c|n

1
where c lies between 0 and x. For − < x ≤ 1, we have |x| ≤ |1 + c| and |1 + c|p−1
2
is bounded. Moreover, by Exercise 4.3.1, for p > −1, we have

|p(p − 1) · · · (p − n)|
lim = 0.
n→∞ (n + 1)!
326 CHAPTER 4. SERIES

1
Then we conclude that limn→∞ Rn (x) = 0 for − < x ≤ 1 and p > −1, and the
2
Taylor series converges to (1 + x)p .
The Taylor series diverges for |x| > 1. Using a new form of the remainder,
Exercise 4.4.2 shows that the Taylor series actually converges to (1 + x)p for |x| < 1
and any p.

Example 4.4.3. In Example 2.5.18, we showed that the function

( 1
e− |x| , if x 6= 0,
f (x) =
0, if x = 0,
has
P all the high derivatives equal to 0. Therefore the Taylor series of the function is
0 = 0, although the function is not 0.

A smooth function is analytic if it is always equal to its Taylor series. As pointed

out after Example 2.5.18, the analytic property means that the function can be
measured by polynomials.

Exercise 4.4.1. Use the Lagrange form of the remainder to show that Cauchy form of the
remainder is
f (n+1) (c)
Rn (x) = (x − c)n (x − x0 ),
n!
where c lies between x0 and x. Use this to show that the Taylor series of log(1 + x) and
(1 + x)p at x0 = 0 converge to the respective functions for any |x| < 1.

Exercise 4.4.2. The Cauchy form of the remainder is

f (n+1) (c)
Rn (x) = (x − c)n (x − x0 ),
n!
where c lies between x0 and x. Use this to show that the Taylor series of log(1 + x) and
(1 + x)p at x0 = 0 converge to the respective functions for any |x| < 1.

4.4.2 Radius of Convergence

an xn as a function with variable x. The domain of the
P
We regard a power series
function consists of those x, such that the series converges. p
We may use the root test to find the domain. Suppose limn→∞ n |an | converges.
Then p p
lim n |an xn | = |x| lim n |an |.
n→∞ n→∞
Let
1
R= p .
limn→∞ n |an |
By the limit version of the root test (see the discussion after Theorem 4.2.3), the
power series converges for |x| < R and diverges for |x| > R.
4.4. POWER SERIES 327
p
Example 4.4.4. For the geometric series xn = 1+x+x2 +· · · , we have limn→∞ n |1| =
P
1. Therefore the geometric series converges for |x| < 1 and diverges for |x| > 1.
Moreover, the series also diverges for |x| = 1. s
P∞ (−1)n+1 n  (−1)n+1 

n 
For the Taylor series n=0 x of log(1+x), we have limn→∞  =
n n 
1
limn→∞ √ = 1. Therefore the series converges for |x| < 1 and diverges for |x| > 1.
n
n
We also know that the series converges at x = 1 and diverges at x = −1.
P 2n
Example 4.4.5. p The series x has an = 1 for even n and an p = 0 for odd n.
The sequence
p
n
|an | diverges because it has two limits: limeven n |an | = 1 and
limodd n |an | = 0. P 2n
x = f (x2 ), where f (x) =
P n
On the other hand, as a function, we have x
2
is the geometric series. Since the domain of f is |x| < 1, the domain of f (x ) is
2
P 2n
|x | < 1, which is equivalent to |x| < 1. Therefore x converges for |x| < 1 and
diverges for |x| > 1.

an xn , there is R ≥ 0, such that the series

P
Theorem 4.4.1. For any power series
absolutely converges for |x| < R and diverges for |x| > R.
p
Using the root test, we have proved the theorem in case limn→∞ n |an | converges.
Example 4.4.11 suggests that Pthe theorem is true in general. TheP theorem is a con-
sequence of the fact that, if an rn converges and |x| < |r|, then an xn absolutely
an rn implies limn→∞ an rn = 0. This
P
converges. Specifically, the convergence of
n
further implies that |an r | < 1 for sufficiently big n. Then for any fixed x satisfying
|x| < |r|, we have  x n  x n
n n
|an x | = |an r | ·   ≤   .
   
r r
x P  x n
Since   < 1 implies the convergence of   , by the comparison test, the series
 
P n
r r
an x absolutely converges.
The number R is the radius of convergence of power series. If R = 0, then the
power series converges only for x = 0. If R = +∞, then the power series converges
for all x.
an (x − x0 )n . The power series
P
The same radius of convergence applies to
converges on (x0 − R, x0 + R), and diverges p on (−∞, x0 − R) and on (x0 + R, +∞).
We had the formula for radius in case |an | converges. In general, the sequence
n

mayphave many possible limits (for various subsequences). Let the upper limit
lim n |an | be the maximum of all the possible limits. Then the radius is
1
R= p .
limn→∞ n
|an |
One can verify the formula for Example 4.4.11.
328 CHAPTER 4. SERIES
√
n
Example 4.4.6. For any p, we have
P p n limn→∞ np = 1. Therefore the radius of conver-
gence for the power series n x is 1. The example already appeared in Example
4.3.2.
√ P n
Example 4.4.7. By limn→∞ n 2n + 3n = 3, the series (2 + 3n )xn converges for
1 1 1
|x| < and diverges for |x| > . The series also diverges for |x| = because the
3 3 3
terms do not converge to 0.
1 P n n 1 P n n
We note that the radius of convergence is for 2 x and for 3 x . The
2 3
radius for the sum of the two series is the smaller one.
√ P n n
Example 4.4.8. By limn→∞ n nn = +∞, the series n x diverges for all x 6= 0.
r n
1 P (−1) n
By limn→∞ n n = 0, the series x converges for all x.
n nn

Example 4.4.9. In Example 4.2.9, we use the ratio test to show that the radius of
P (2n)! n 1
convergence of 2
x is . The idea can be used to show that the radius of
(n!) 4  
P n
 an 
convergence for an x is R = limn→∞  , provided that the limit converges.
an+1 
For example, by
p
np

n
lim = lim = 1,
n→∞ (n + 1)p n→∞ n + 1

np xn is 1. Moreover, the radius of convergence for

P
the radius of convergence for
the Taylor series of ex is

1
lim n! = lim (n + 1) = +∞.
n→∞ 1 n→∞
(n + 1)!

Example 4.4.10. The Bessel function of order 0 is

∞
X (−1)n x2n
J0 (x) = .
n=0
22n (n!)2

The radius of convergence is the square root of the radius of convergence of the
series
∞
X (−1)n xn
.
n=0
22n (n!)2
4.4. POWER SERIES 329

By
(−1)n
 
 
 
 22n (n!)2 
lim   = lim 4(n + 1)2 = +∞,
n+1
n→∞
 (−1)  n→∞

 22n+2 ((n + 1)!)2 

the later series converges for all x. Therefore

 the Bessel function is defined for all x.
 an 
Note that we cannot calculate limn→∞   directly for the Bessel function. In
an+1 
fact, the limit diverges.

Exercise 4.4.3. Suppose an can be divided into two subsequences an0 and an00 . Suppose
0 00 1
limn0 →∞ n |an0 | = l0 and limn00 →∞ n |an00 | = l00 converge. Prove that
p p
is the
max{l0 , l00 }
an xn .
P
radius of convergence for
 
 an 
Exercise 4.4.4. Suppose limn→∞ 
  = R converges. Prove that R is the radius of
an+1

n
P
convergence for an x .

Exercise 4.4.5. Determine the radius of convergence.

√
xn 7.
P
n n xn . 12.
P 2
an xn .
(−1)n
P
1. .
np
n!xn .
P
2.
P p
n (x − 1)n . 8. 2 2
an xn .
P
13.
3.
P p
n (2x − 1)n . P (−1)n+1 n
9. √ x . 2 −1
2n xn
P
P p n! 14. .
4. n (2x + 3)n .
P an P (n!)2 n
bn
 
(2 + (−1)n )n xn .
P
5. + 2 xn . 10. x . 15.
n n (2n)!

xn P (3n)! n P (2 + (−1)n )n n
11. x . 16. x .
P
6. .
an + bn n!(2n)! log n

Exercise 4.4.6. Find the radius of convergence.

P n+1 n n  2  2
 
P n + a n n+2 n + 1 n n2
 
1. x . 3. x . 5.
P
(−1)n x .
n n+b n
n2 n2
P an + b n n−2
   
n+1 n+1
xn . (−1)n xn .
P P
2. 4. 6. x .
n n cn + d

Exercise 4.4.7. Find the domain of the Bessel function of order 1

∞
X (−1)n x2n+1
J1 (x) = .
n!(n + 1)!22n+1
n=0
330 CHAPTER 4. SERIES

Exercise 4.4.8. Find the domain of the Airy function

x3 x6 x9
A(x) = 1 + + + + ··· .
2·3 2·3·5·6 2·3·5·6·8·9

an xn and bn xn are R and R0 .

P P
Exercise 4.4.9. Suppose the radii of convergence for
What can you say about the radii of convergence for the following power series?
X X X X
(an + bn )xn , (an − bn )xn , (−1)n an xn , an (2x − 1)n ,
X X X X
an x2n , an xn+2 , a2n xn , an+2 xn ,
2 2
X X X X
an xn , an2 xn , a2n x2n , an2 xn .

4.4.3 Function Defined by Power Series

Examples 2.5.18 and 4.4.3 suggest that if a function is the sum of a power series,
then the function is particularly nice. In fact, the function should be nicer than
functions with derivatives of any order.
Because power series converge absolutely within the radius of convergence, by
Theorem 4.3.5, we can multiply two power series together within the common radius
of convergence.

an xn and g(x) = bn xn have radii of conver-

P P
Theorem 4.4.2. Suppose f (x) =
gence R and R0 . Then for |x| < min{R, R0 }, we have
X
f (x)g(x) = cn xn , cn = a0 bn + a1 bn−1 + · · · + an b0 .

The product should be the sum of ai bj xi+j . We get the power series cn xn by
P
gathering all the terms with power xn .
The power series can also be differentiated or integrated term by term within
the radius of convergence.
P∞
Theorem 4.4.3. Suppose f (x) = n=0 an xn for |x| < R. Then
∞
X
0
f (x) = (an xn )0 = a1 + 2a2 x + 3a3 x2 + · · · + nan xn−1 + · · · ,
n=1

and
Z x ∞ Z x
X a1 2 a2 3 an n+1
f (t)dt = an tn dt = a0 x + x + x + ··· + x + ···
0 n=0 0 2 3 n+1

for |x| < R.

4.4. POWER SERIES 331

1
= ∞ n
P
Example 4.4.11. Taking the derivative of n=0 x , we get
1−x
1
= 1 + 2x + 3x2 + 4x3 + · · · + nxn−1 + · · · ,
(1 − x)2
2
3
= 2 · 1 + 3 · 2x + 4 · 3x2 + 5 · 4x3 + · · · + n(n − 1)xn−2 + · · · .
(1 − x)

Therefore
∞
X ∞
X ∞
X
2 2 2 2 n 2 n n−1 2
1 x + 2 x + ··· + n x + ··· = n x =x nx +x n(n − 1)xn−2
n=1 n=1 n=2
1 2 x(1 + x)
=x 2
+ x2 3
= .
(1 − x) (1 − x) (1 − x)3

If we integrate instead, then we get

Z x
dx x2 x3 xn
log(1 − x) = − = −x − − − ··· − − ··· , for |x| < 1.
0 1−x 2 3 n!

Substituting −x for x, we get the Taylor series of log(1 + x)

x2 x3 xn
log(1 + x) = x − + − · · · + (−1)n+1 + · · · , for |x| < 1.
2 3 n!
Note that in Example 4.4.1, by estimating the remainder, we were able to prove the
1
equality rigorously only for − < x < 1. Here by using term wise integration, we
2
get the equality for all x within the radius of convergence.

1
= ∞ n 2n
P
Example 4.4.12. By integrating 2 n=0 (−1) x , we get the Taylor series of
1+x
arctan x
∞
X (−1)n 2n+1 x3 x5 x7 (−1)n 2n+1
arctan x = x = x− + − +· · ·+ x + · · · for |x| < 1.
n=0
2n + 1 3 5 7 2n + 1

Exercise 4.4.10. Use the product of power series to verify the identity sin 2x = 2 sin x cos x.

Exercise 4.4.11. Find Taylor series and determine the radius of convergence.
1 4. sin2 x, at 0. 7. arcsin x, at 0.
1. , at 0.
(x − 1)(x − 2)
π
√ 5. sin x, at . 8. arctan x, at 0.
2. x, at x = 2. 2 Z x
π sin t
3. sin x2 , at 0. 6. sin 2x, at . 9. dt, at 0.
2 0 t
332 CHAPTER 4. SERIES

P∞ n f (x)
Exercise 4.4.12. Given the Taylor series n=0 an x of f (x), find the Taylor series of .
1+x

P∞ xn
Exercise 4.4.13. Show that the function f (x) = n=0 satisfies xf 00 + f 0 − f = 0.
(n!)2

Exercise 4.4.14. Show that the Airy function in Exercise 4.4.8 satisfies f 00 − xf = 0.

Exercise 4.4.15. Show that the Bessel functions in Example 4.4.10 and Exercise 4.4.7 satisfy

xJ000 + J00 + xJ0 = 0, x2 J100 + xJ10 + (x2 − 1)J1 = 0.

A power series may or may not converge at the radius of convergence (i.e., at
±R). If it converges, then the following gives the value of the sum.
P∞ n
Theorem
P∞ 4.4.4. Suppose n=0 an x converges for |x| < R and x = R. Then
n
P ∞ n
n=0 an R = limx→R− n=0 an x .

The theorem says that, if f (x) = ∞ n

P
n=0 an x is also defined at R, then f (x) is
left continuous at R. We also have the similar statement at the other end −R.

(−1)n+1 n Pn
Example 4.4.13. By Examples 4.4.11 and 4.3.1, we know log(1 + x) = n=1 x
n
converges for |x| < 1, and the series converges at x = 1. Since log(1 + x) is continuous at
x = 1, by Theorem 4.4.4, we get
n n
X (−1)n+1 X (−1)n+1
= lim xn = lim log(1 + x) = log(1 + 1) = log 2.
n x→1− n x→1−
n=1 n=1

We computed the sum in Example 4.3.6 by another way.

Exercise 4.4.16. Find the sum. Discuss what happens at the radius of convergence.

1.
P∞ 2 xn . (x − 1)n x2n+1
n=1 n
P∞ P∞
3. n=2 . 5. n=0 .
n(n − 1) 2n + 1
P∞ xn P∞ xn
2.
P∞
n3 xn . 4. n=1 . 6. n=0 .
n=1 n(n + 1)(n + 2) 2n + 1

Exercise 4.4.17. Find the sum. Discuss what happens at the radius of convergence.

P∞ x 2n+1 P∞ x2n P∞ xn
1. (−1)n . 2. (−1)n . 3. n=1 .
n=1
n! n=1
(2n + 1)! 2n (2n − 1)!

Exercise 4.4.18. Find the Taylor series of the function and the radius of convergence. Then
explain why the sum of the Taylor series is the given function.
4.5. FOURIER SERIES 333
√
1. arcsin x. 3. arctan x. 5. log(x + 1 + x2 ).
Z x Z x Z x
sin t 2 log(1 − t)
2. dt. 4. e−t dt. 6. dt.
0 t 0 0 t

4.5 Fourier Series

If f (x + p) = f (x) for all x and a constant p, then we say f (x) is a periodic function
of period p. For example, the functions sin x and cos x have period 2π, and tan x
has period π.
A periodic function of period p is also a periodic function of period kp for any
2π
integer k. For example, cos nx and sin nx have the period p = as well as the
n
period np = 2π.
If f (x) has period p, then f (x + a) still has period p, and f (ax) is periodic with
p
period .
a
A combination of periodic functions of the √same2 period p is still periodic of period
2
p. For example, sin x + cos x, sin 3x cos x, 2 sin x + cos4 x are periodic of period
2π.
The Taylor series approximates a function near a point by linear combinations
of power functions. Similarly, we wish to approximate a periodic function by linear
combinations of simple periodic functions such as sine and cosine. Specifically, we
wish a periodic function f (x) of period 2π to be approximated as
∞
a0 X
f (x) ∼ + (an cos nx + bn sin nx)
2 n=1
a0
= + a1 cos x + b1 sin x + a2 cos 2x + b2 sin 2x + · · · .
2

Note that there is no b0 because sin 0x = 0. The 12 factor for a0 allows us to have a
unify formula for an later. Moreover, like the Taylor series, we use ∼ instead of =
to indicate that the equality is yet to be established.
The approximation of a periodic function by (linear combinations of) trigono-
metric functions is not measured by the values at single points, but rather the overall
approximation in terms of the integral of the difference function. This means that we
can only expect that the sum of the trigonometric series to be equal to the function
“almost everywhere”.

4.5.1 Fourier Coefficient

Let f (x) be a periodic function of period 2π. Our first problem is to find the
coefficients an and bn in the trigonometric series. By Exercise 3.1.6, the trigonometric
334 CHAPTER 4. SERIES

functions are “orthogonal” in the sense that

Z 2π
cos mx sin nxdx = 0;
0

Z 2π 0,
 if m 6= n,
cos mx cos nxdx = π, if m = n 6= 0,
0 
2π, if m = n = 0;

Z 2π (
0, if m 6= n or m = n = 0,
sin mx sin nxdx =
0 π, if m = n 6= 0.

We expect the sum of the trigonometric series to be equal to f (x) as far as integra-
tions are concerned. We also assume that the integration of infinite series can be
calculated term by term. Then we get
Z 2π ∞
a0 2π
Z X Z 2π
f (x) cos nxdx = cos nxdx + am cos mx cos nxdx
0 2 0 k=1 0
∞
X Z 2π
+ bm sin mx cos nxdx = πan
k=1 0
Z 2π Z 2π ∞ Z 2π
a0 X
f (x) sin nxdx = sin nxdx + am cos mx sin nxdx
0 2 0 m=1 0
∞
X Z 2π
+ bm sin mx sin nxdx = πbn .
m=1 0

Definition 4.5.1. The Fourier series of a periodic function f (x) of period 2π is

∞
a0 X
f (x) ∼ + (an cos nx + bn sin nx),
2 n=1

with the Fourier coefficients

1 2π
Z
an = f (x) cos nxdx, n ≥ 0,
π 0
1 2π
Z
bn = f (x) sin nxdx, n > 0.
π 0

Example 4.5.1. Similar to Example 2.5.1, by

 
4 1 2 1 1
sin x = (1 − cos 2x) = 1 − 2 cos 2x + (1 − cos 4x)
4 4 2
3 1 1
= − cos 2x + cos 4x,
8 2 8
4.5. FOURIER SERIES 335

the right side is the Fourier series of sin4 x. The coefficients give
Z 2π Z 2π Z 2π
4 3π 4 π π
sin xdx = , sin x cos 2xdx = − , sin4 x cos 4xdx = .
0 4 0 2 0 8

Example 4.5.2. A periodic function is determined by its value on one interval of

period length. For example, if f (x) is a periodic function of period 2π and satisfies
(
1, if 0 ≤ x < a,
f (x) =
0, if a ≤ x < 2π,

then (
0, if 2kπ ≤ x < 2kπ + a,
f (x) =
1, if 2kπ + a ≤ x < 2(k + 1)π.

−4π −2π a 2π 4π

The Fourier coefficients are

1 a
Z
a
a0 = 1dx = ,
π 0 π
1 a
Z
sin na
an = cos nxdx = ,
π 0 nπ
1 a 1 − cos na
Z
bn = sin nxdx = .
π 0 nπ
and the Fourier series is
∞
a X 1
f (x) ∼ + (sin na cos nx + (1 − cos na) sin nx).
2π n=1 nπ

Example 4.5.3. Let f (x) be the even periodic function of period 2π satisfying
(
1, if |x| ≤ a,
f (x) =
0, if a < |x| ≤ π.

−4π −2π −π a π 2π 4π
336 CHAPTER 4. SERIES

We have
1 2π 1 π 2 π 2 a
Z Z Z Z
2a
a0 = f (x)dx = f (x)dx = f (x)dx = 1dx = ,
π 0 π −π π 0 π 0 π
Z 2π Z π Z a
1 2 2 2 sin na
an = f (x) cos nxdx = f (x) cos nxdx = cos nxdx = .
π 0 π 0 π 0 nπ
Z p Z a+p
The calculation used the fact that f (x)dx = f (x)dx for any periodic func-
0 a
tion of period p. We may also calculate bn and find bn = 0. In fact, for even function,
we expect that all the odd terms bn sin nx to vanish.

Exercise 4.5.1. Suppose f (x) is an even periodic function of period 2π. Prove that
2 π
Z
an = f (x) cos nxdx, bn = 0.
π 0

Exercise 4.5.2. Suppose f (x) is an odd periodic function of period 2π. Prove that
2 π
Z
an = 0, bn = f (x) sin nxdx.
π 0
 π
Exercise 4.5.3. Extend f (x) on 0, to a periodic function of period 2π, such that its
P∞ 2
Fourier series is of the form n=1 an cos(2n − 1)x? How about ∞
P
n=1 bn sin(2n − 1)x?

Exercise 4.5.4. Given the Fourier series of f (x) and g(x), what is the Fourier series of
af (x) + bg(x)? Use the idea and Example 4.5.2 to find the Fourier series of the periodic
function f (x) of period 2π satisfying
(
1, if a ≤ x < b,
f (x) =
0, if 0 ≤ x < a or b ≤ x < 2π.

Exercise 4.5.5. Suppose f (x) is a periodic function of period 2π. What is the relation
between the Fourier series of f (x) and f (x + a)? Use the idea and Example 4.5.3 to derive
Example 4.5.2.

 Ap periodic
 function f (x) of period p may be converted
 p  to a periodic function
f x of period 2π. Then the Fourier series of f x gives the Fourier series
2π 2π
of f (x). Alternatively, the basic periodic functions cos nx and sin nx of period 2π
2nπ 2nπ
give the basic periodic functions cos x and sin x of period p, and we expect
p p
the Fourier series of f (x) to be
∞  
a0 X 2nπ 2nπ
f (x) ∼ + an cos x + bn sin x .
2 n=1
p p
4.5. FOURIER SERIES 337

By an argument similar to the case of period 2π, we get the Fourier coefficients
2 p
Z
2nπ
an = f (x) cos xdx, n ≥ 0,
p 0 p
2 p
Z
2nπ
bn = f (x) sin xdx, n > 0.
p 0 p

Exercise 4.5.6. Suppose f (x) is a periodic function of period p.

 p 
1. Write down the Fourier series for f x , together with the formulae for its coef-
2π
ficients.

2. Convert the first part to statements about the original f (x).

Exercise 4.5.7. Use Example 4.5.2 to derive the Fourier series of the periodic function of
period 1 satisfying (
1, if 0 ≤ x < a,
f (x) =
0, if a ≤ x < 1.

Exercise 4.5.8. Derive the formulae for the Fourier coefficients of periodic even or odd
functions of period p, similar to Exercises 4.5.1 and 4.5.2.

Example 4.5.4. The function x on (0, 1) extends to a periodic function of period 1

f (x) = x − k, k < x < k + 1.

−4 −3 −2 −1 1 2 3 4

Note that we do not care about the value at the integer points because it does
not affect the Fourier coefficients, which are
2 1
Z
a0 = xdx = 1,
1 0
2 1
Z Z 1 Z 1
1 1
an = x cos 2nπxdx = xd sin 2nπx = − sin 2nπxdx = 0,
1 0 nπ 0 nπ 0
2 1
Z Z 1
1
bn = x sin 2nπxdx = − xd cos 2nπx
1 0 nπ 0
 Z 1 
1 1
=− 1− cos 2nπxdx = − .
nπ 0 nπ
1
We note that the reason for an = 0 for n 6= 0 is that f (x) − is an odd function.
2
338 CHAPTER 4. SERIES

The Fourier series is

∞
1 1X1
x∼ − sin 2nπx, x ∈ (0, 1).
2 π n=1 n

We indicate x ∈ (0, 1) because the function equals x only on the interval. The
function is x − 1 instead of x on the interval (1, 2).

Example 4.5.5. The function x on (0, 1) extends to an even periodic function of

period 2
f (x) = |x − 2k|, 2k − 1 < x < 2k + 1.
This is also the extension of the function |x| on (−1, 1) to a periodic function of
period 2.

−4 −3 −2 −1 1 2 3 4

Using Exercises 4.5.1 and 4.5.8, we get bn = 0 and

2 1
Z Z 1
a0 = 2 · f (x)dx = 2 xdx = 1,
2 0 0 
2
Z 1 Z 1 − 4 , if n is odd,
an = f (x) cos nπxdx = 2 x cos nπxdx = n2 π 2
1 0 0 0, if n is even.

Replacing n by 2n + 1, the Fourier series is

∞
1 4 X 1
|x| ∼ − 2 cos(2n + 1)πx, x ∈ (−1, 1).
2 π n=0 (2n + 1)2

We may also extend the function to an odd periodic function of period 2. This
is also the extension of the function x on (−1, 1) to a periodic function of period 2.
The Fourier coefficients an = 0 for the odd function, and
1
(−1)n+1 2
Z
bn = 2 x sin nπxdx = .
0 nπ

The Fourier series is

∞
X (−1)n+1 2
x∼ sin nπx, x ∈ (−1, 1).
n=1
nπ
4.5. FOURIER SERIES 339

−4 −3 −2 −1 1 2 3 4

Example 4.5.6. Let f (x) be the periodic function of period 1 extending the function
x2 on (0, 1). Then
Z 1
2
a0 = 2 x2 dx = ,
3
Z0 1
1
an = 2 x2 cos 2nπxdx = 2 2 ,
nπ
Z0 1
1
bn = 2 x2 sin 2nπxdx = − .
0 nπ
The Fourier series is
∞  
21 X 1 1
x ∼ + cos 2nπx − sin 2nπx , x ∈ (0, 1).
3 n=1 n2 π 2 nπ

−4 −3 −2 −1 1 2 3 4

Exercise 4.5.9. The periodic function is given on one interval of period length. Find the
Fourier series.

1. sin2 x on (−π, π). 4. sin x on (0, p). 7. x on (a, b).

2. sin x on (0, π). 5. cos x on (0, p). 8. x sin x on (−π, π).

3. | sin x| on (0, 2π). 6. |x| on (−p, p). 9. ex on (0, 1).

Exercise 4.5.10. Use the Fourier coefficient to calculate the integral.

Z π Z 2π Z 2π
1. sin2 x cos 2xdx. 3. sin x cos 2x sin 3xdx. 5. sin3 x sin 3xdx.
0 0 0
Z 2π Z 2π Z 2π
2. sin6 xdx. 4. sin x cos 2x cos 3xdx. 6. sin3 x cos 3xdx.
0 0 0

Exercise 4.5.11. Write the formula for the Fourier coefficients of the even periodic (of
period 2p) extension of a function f (x) on (0, p). What about the odd extension?
340 CHAPTER 4. SERIES

Exercise 4.5.12. Extend the function on (0, p) to even and odd functions of period 2p and
compute the Fourier series.

1. x2 on (0, 1). 2. sin x on (0, π). 3. cos x on (0, p).

Exercise 4.5.13. Given the Fourier series of functions f (x) and g(x) of period p. Find the
Fourier series of the following periodic functions.

1. f (ax). 2πx 1 x+h

Z
3. f (x) sin . 5. f (t)dt.
p h x
Z x+h Z p
2πx 1
2. f (x) cos . 4. f (t)dt. 6. f (t)g(x − t)dt.
p 2h x−h 0

Exercise 4.5.14. Use Examples 4.5.4 and 4.5.5 to find the Fourier series.

1. x on (0, p). 4. 0 on (−1, 0) and x on (0, 1).

2. −x on (−p, 0). 5. ax on (−1, 0) and bx on (0, 1).

3. |x| on (−p, p). 6. ax on (−p, 0) and bx on (0, p).

4.5.2 Complex Form of Fourier Series

The sine and cosine functions are related to the exponential function via the use of
complex numbers

eix + e−ix eix − e−ix √

eix = cos x + i sin x, cos x = , sin x = , i= −1.
2 2i

Correspondingly, the Fourier series may be rewritten

∞
X a0 an − ibn an + ibn
f (x) = cn einx , c0 = , cn = , c−n = ,
n=−∞
2 2 2

where complex Fourier coefficients are

Z 2π
1
cn = f (x)e−inx dx.
2π 0

In this course, f (x) is a real valued function, and an , bn are real numbers. There-
fore c−n ei(−n)x is the complex conjugation of cn einx , and the usual Fourier series is
given by
a0 = 2c0 , an cos nx + bn sin nx = 2Re(cn einx ).
4.5. FOURIER SERIES 341

Example 4.5.7. For the function in Example 4.5.2, the complex Fourier coefficient is
Z a
1 a
c0 = dx = ,
2π 0 2π
Z a
1 1 i
cn = e−inx dx = e−inx |a0 = (e−ina − 1).
2π 0 −2inπ 2nπ
The complex Fourier series is
a X i
f (x) ∼ + (e−ina − 1)einx , x ∈ (0, 2π).
2π n6=0 2nπ

By
   
i i
2Re (e−ina − 1)einx = Re ((cos na − 1) − i sin na)(cos nx + i sin nx)
2nπ nπ
1
= (sin na cos nx − (cos na − 1) sin nx),
nπ
we recover the Fourier series in terms of trigonometric functions in Example 4.5.2.

Example 4.5.8. Consider the periodic function of period 2π given by ex on (0, 2π).
The complex Fourier coefficient is (recall that e2ikπ = 1)
Z 2π
1 1 e2π − 1
cn = ex e−inx dx = e(1−in)x |2π
0 = .
2π 0 2π(1 − in) 2nπ(1 − in)
The complex Fourier series is
∞
e2π − 1 X
x 1
e ∼ einx , x ∈ (0, 2π).
2π n=−∞ 1 − in

By
   
1 1 + in cos nx − n sin nx
Re einx = Re 2
(cos nx + i sin nx) = ,
1 − in 1+n 1 + n2
we get the Fourier series in terms of trigonometric functions
∞
!
2π
e − 1 1 X cos nx − n sin nx
ex ∼ + , x ∈ (0, 2π).
π 2 n=1 1 + n2

By changing x to 2πx, we get the Fourier series for the periodic function of period
1 given by e2πx on (0, 1)
∞
!
2π
e − 1 1 X cos 2nπx − n sin 2nπx
e2πx ∼ + , x ∈ (0, 1).
π 2 n=1 1 + n2
342 CHAPTER 4. SERIES

a sin x
Example 4.5.9. Consider the function f (x) = , with |a| < 1. We
1 − 2a cos x + a2
rewrite the function and take the Taylor expansion in terms of einx = z n , z = eix ,

eix − e−ix
a sin x a
= 2i
1 − 2a cos x + a2 eix + e−ix
1 − 2a + a2
2
eix − e−ix
 
a 1 1 1
= = −
2i (1 − aeix )(1 − ae−ix ) 2i 1 − aeix 1 − ae−ix
∞ ∞
!
1 X ix n X −ix n 1 X |n| inx
= (ae ) − (ae ) = a e
2i n=0 n=0
2i n6=0
∞ ∞
1 X n inx X
= a (e − e−inx ) = an sin nx.
2i n=1 n=1

Note that we have geometric series because |aeix | = |ae−ix | = |a| < 1. Moreover,
the Fourier series is actually equal to the function. The example also shows the
connection between the Fourier series and the power series.

Exercise 4.5.15. Find the complex form of the Fourier series of ax on (0, 2π). Then by
1
taking a = e 2π , derive the Fourier series of the function ex on (0, 1).

ix
Exercise 4.5.16. Find the complex form of the Fourier series of e 2 on (0, 2π). Then derive
the Fourier series of cos x and sin x on (0, π) by taking the real and imaginary parts.

Exercise 4.5.17. What is the complex form of the Fourier series for a periodic function of
period p?

Exercise 4.5.18. Find complex form of Fourier series.

1. sin2 x on (−π, π). 3. |x| on (−p, p). 5. x sin x on (−π, π).

2. eix on (0, p). 4. x2 on (0, 1). 6. x cos x on (−π, π).

Exercise 4.5.19. Find Fourier series, where |a| < 1.

1 − a cos x 1 − a2 3. log(1 − 2a cos x + a2 ).

1. . 2. .
1 − 2a cos x + a2 1 − 2a cos x + a2

4.5.3 Derivative and Integration of Fourier Series

The derivative of a periodic function is still periodic. We may use f (p) = f (0)
(which is the periodic property) and the integration by parts to compute the Fourier
4.5. FOURIER SERIES 343

coefficients An , Bn of f 0 (x)
2 p 0
Z
2
A0 = f (x)dx = (f (p) − f (0)) = 0,
p 0 p
Z p
2 p
Z
2 0 2nπ 2nπ
An = f (x) cos xdx = cos xdf (x)
p 0 p p 0 p
 Z p 
2 2nπ 2nπ 2nπ
= f (p) − f (0) + f (x) sin xdx = bn ,
p 0 p p p
2 p 0 2 p
Z Z
2nπ 2nπ
Bn = f (x) sin xdx = sin xdf (x)
p 0 p p 0 p
2 p
Z
2nπ 2nπ 2nπ
=− f (x) cos xdx = − an .
p 0 p p p
This shows that we may differentiate the Fourier series term by term.

Proposition 4.5.2. Suppose f (x) is a periodic function of period p that is continuous

and piecewise continuously differentiable. If the Fourier coefficients of f (x) are
an , bn , then the Fourier series of f 0 (x) is
∞  
0 2π X 2nπ 2nπ
f (x) ∼ n bn cos x − an sin x .
p n=1 p p
Z x
The integration F (x) = f (t)dt of a periodic function of period p is still
Z p 0
2
periodic if a0 = f (x)dx = 0. Then for n 6= 0, the other Fourier coefficients of
p 0
F (x) are
2 p
Z Z p
2nπ 1 2nπ
An = F (x) cos xdx = F (x)d sin x
p 0 p nπ 0 p
Z p
1 2nπ p
=− f (x) sin xdx = − bn ,
nπ 0 p 2nπ
2 p
Z Z p
2nπ 1 2nπ
Bn = F (x) sin xdx = − F (x)d cos x
p 0 p nπ 0 p
Z p Z p 
1 2nπ p
=− f (x)dx − f (x) cos xdx = an .
nπ 0 0 p 2nπ
The 0-th coefficient is
2 p
Z p
2 p
Z   Z
2 x=p
A0 = F (x)dx = xF (x)|x=0 − xf (x)dx = − xf (x)dx.
p 0 p 0 p 0
This shows that, in case a0 = 0, we may almost integrate the Fourier series term by
term.
344 CHAPTER 4. SERIES

Proposition 4.5.3. Suppose f (x) is a periodic function of period p. If the Fourier co-
Z x
efficients of f (x) are an , bn and a0 = 0, then the Fourier series of F (x) = f (t)dt
0
is ∞
Z x  
A0 p X1 2nπ 2nπ
f (t)dt ∼ + −bn cos x + an sin x .
0 2 2π n=1 n p p

Example 4.5.10. In Example 4.5.4, we found the Fourier series

∞
1 1X1
x∼ − sin 2nπx, x ∈ (0, 1).
2 π n=1 n
We wish to integrate to get the Fourier series of x2 on (0, 1). However, to satisfy
1
the condition of Proposition 4.5.3, we should consider the Fourier series of x − ,
Z x  2
1 1 2 1
which has vanishing 0-th coefficient. Then F (x) = t− dt = x − x has
0 2 2 2
Fourier series
∞
1 2 1 1 1 X 1
x − x∼− + 2 cos 2nπx, x ∈ (0, 1).
2 2 12 2π n=1 n2
Here the 0-th coefficient is
Z 1 
A0 1 2 1 1
= x − x dx = − .
2 0 2 2 12
 
1 2 1
Then the Fourier series of x2 = 2 x − x + x on (0, 1) is
2 2
∞
! ∞
!
1 1 X 1 1 1 X 1
x2 ∼ 2 − + 2 cos 2n + − sin 2nπx
12 2π n=1 n2 2 π n=1 n
∞  
1 X 1 1
= + cos 2nπx − sin 2nπx , x ∈ (0, 1).
3 n=1 n2 π 2 nπ
We obtained this Fourier series in Example 4.5.6 by direct computation.

Exercise 4.5.20. Derive the Fourier series of x3 on (0, 1) from the Fourier series of x2 .

Exercise 4.5.21. Derive the Fourier series of |x| on (−1, 1) from the Fourier series of its
derivative.

Exercise 4.5.22. Suppose f (x) is continuously differentiable on [0, p], with perhaps different
f (0+ ) and f (p− ). Then we can extend both f (x) and f 0 (x) to periodic functions of period
p. If the Fourier coefficients of the extended f (x) are an , bn , prove that
∞ 
f (p− ) − f (0+ ) 2 X

0 − + 2nπ 2nπ
f (x) ∼ + (f (p ) − f (0 ) + nπbn ) cos x + nπan sin x .
p p p p
n=1
4.5. FOURIER SERIES 345

4.5.4 Sum of Fourier Series

Like the Taylor series, the Fourier series may not always converge to the function.
The following is one good case when the Fourier series converges.

Theorem 4.5.4. Suppose f (x) is a periodic function. If f (x) has one sided limits
f (x− +
0 ) and f (x0 ) at x0 , and there is M , such that

x < x0 and close to x0 =⇒ |f (x) − f (x−

0 )| ≤ M |x − x0 |,

and
x > x0 and close to x0 =⇒ |f (x) − f (x+
0 )| ≤ M |x − x0 |.
−
f (x+
0 ) + f (x0 )
Then the Fourier series of f (x) converges to at x = x0 .
2

The condition means that the value of f (x) lies in two “corners” on the two sides
of x0 .

f (x+
0)

f (x−
0)

x0

Figure 4.5.1: Condition for the convergence of Fourier series.

Example 4.5.11. In Example 4.5.4, we find the Fourier series of x on (0, 1). The function
satisfies the condition of Theorem 4.5.4 at x0 = 0. We conclude that
∞
0+1 f (0− ) + f (0+ ) 1 1X1
= = − sin 2nπ0.
2 2 2 π n
n=1

1
This is trivially true. We get similar trivial equalities at x0 = 1 and x0 = .
2
1
If we take x0 = , then
3
∞  
1 1 1X 1 2(3k + 1)π 1 2(3k + 2)π 1 2(3k + 3)π
= − sin + sin + sin
3 2 π 3k + 1 3 3k + 2 3 3k + 3 3
k=1
∞ √ √ !
1 1X 1 3 1 3
= − − .
2 π 3k + 1 2 3k + 2 2
k=1
346 CHAPTER 4. SERIES

This means that

1 1 1 1 1 1 1 1 1 π
1− + − + − + − + − + ··· = √ .
2 4 5 7 8 10 11 13 14 3 3

1 1
Similarly, by evaluating the Fourier series at x0 = and x0 = , we get
4 8
1 1 1 1 1 1 1 1 1 π
1− + − + − + − + − + ··· = ,
3 5 7 9 11 13 15 17 19 4
1 1 1 1 1 1 1 1 1 π
1+ − − + + − − + + − ··· = √ .
3 5 7 9 11 13 15 17 19 2 2

0 1

Figure 4.5.2: Partial sums of the Fourier series for x on (0, 1).

Example 4.5.12. The periodic function of period 2 given by |x| on (−1, 1) satisfies the
condition of Theorem 4.5.4 everywhere. Evaluating the Fourier series in Example 4.5.4 at
x = 0, we get
∞
1 4 X 1
0= − 2 .
2 π (2n + 1)2
n=0

This means that

∞
X 1 1 1 1 π2
= 1 + + + + · · · = .
(2n + 1)2 32 52 72 8
n=0

If the sum also includes the even terms, then we have

∞ ∞ ∞ ∞ ∞
X 1 X 1 X 1 X 1 1X 1
= + = + .
n2 (2n + 1)2 (2n)2 (2n + 1)2 4 n2
n=1 n=0 n=1 n=0 n=1

Therefore
∞ ∞
X 1 1 1 1 4X 1 π2
= 1 + + + + · · · = = .
n2 22 32 42 3 (2n + 1)2 6
n=1 n=0
4.5. FOURIER SERIES 347

1
If we evaluate at x = , then we get
4
 
1 1 4 1 1 1 1 1 1 1 1
= −√ + − − + + − − + ··· .
4 2 2π 2 12 32 52 72 92 112 132 152
Combined with the sum above, we get
 
1 1 1 1 1 1 1 1
+ + + + + + ··· = + √ π2,
12 32 92 112 172 192 12 16 2
 
1 1 1 1 1 1 1 1
+ + + + + + ··· = − √ π2.
52 72 132 152 212 232 12 16 2

P∞ 1
Exercise 4.5.23. Use the Fourier series of x2 on (0, 1) to compute n=1 .
n2

Exercise 4.5.24. Use the Fourier series of x2 on (−1, 1) to get the Fourier series of x3 and
1
x4 on (−1, 1). Then evaluate the Fourier series of x4 to get ∞
P
n=1 4 .
n

4.5.5 Parseval’s Identity

In Example 4.5.1, we had the Fourier series
3 1 1
sin4 x = − cos 2x + cos 4x.
8 2 8
Then 2
Z 2π Z 2π 
8 3 1 1
sin xdx = − cos 2x + cos 4x dx.
0 0 8 2 8
1
We may expand the square on the right. We get square terms such as 2 cos2 2x
2
1 1
and cross terms such as 2 · cos 2x cos 4x. The square term has nonzero integral
2 8
Z 2π
cos2 2xdx = π.
0

The cross terms has vanishing integral

Z 2π
cos 2x cos 4xdx = 0.
0

Therefore  2  2  2
Z 2π
8 3 1 1 35
sin xdx = 2π + π+ π = π.
0 8 2 8 64
The idea (which is essentially Pythagorean theorem) leads to the following formula.
348 CHAPTER 4. SERIES

Theorem 4.5.5 (Parseval’s Identity). Suppose f (x) is a periodic function of period

p. Then its Fourier coefficients satisfy
∞ p
a20 X 2
Z
2
+ (an + b2n ) = |f (x)|2 dx.
2 n=1
p 0

The identity means that the Fourier series, considered as a conversion between
periodic functions and sequences of numbers, preserves the “Euclidean length”. The
complex form of Parseval’s identity is
∞ Z 2π
X
2 1
|cn | = |f (x)|2 dx.
n=−∞
2π 0

The proof of this identity requires the notion of uniform convergence to justify the
interchange of summation and integration, and is outside the scope of this course.

Example 4.5.13. Applying Parseval’s identity to the Fourier series of x on (0, 1), we
get
∞ Z 1
1 X 1 2
+ 2 2
=2 x2 dx = .
2 n=1 n π 0 3
This is the same as
1 1 1 1 π2
+ + + + · · · = .
12 22 32 42 6
Applying the identity to the Fourier series of x2 on (0, 1), we get
∞ ∞ Z 1
2 X 1 X 1 2
+ + =2 x4 dx = .
9 n=1 n4 π 4 n=1 n2 π 2 0 5

P∞ 1 π2
Using n=1 = , we get
n2 6
1 1 1 1 π4
+ + + + · · · = .
14 24 34 44 90

Example 4.5.14. Applying Parseval’s identity to the complex form of the Fourier
series of ex on (0, 2π) in Example 4.5.8, we get
∞ Z 2π
(e2π − 1)2 X 1 1 2x e4π − 1
= e dx = .
4π 2 n=−∞
1 + n 2 2π 0 4π

The equality leads to

∞
X 1 e2π + 1
1+2 = π,
n=1
1 + n2 e2π − 1
4.5. FOURIER SERIES 349

or ∞
X 1 1 1 1 1 1 π(e2π + 1) 1
= + + + + + · · · = − .
n=1
1 + n2 2 5 10 17 26 2(e2π − 1) 2

Exercise 4.5.25. Apply Parseval’s identity to the Fourier series in Example 4.5.2 to find
P∞ sin2 na P∞ cos2 na
n=1 and n=1 .
n2 n2

Exercise 4.5.26. Apply Parseval’s identiy to the Fourier series of x3 and x4 on (0, 1) to find
P∞ 1 P∞ 1
n=1 6 and n=1 8 .
n n

P∞ 1
Exercise 4.5.27. Find n=1 by evaluating the Fourier series in Example 4.5.8.
1 + n2

The Parseval’s Identity immediate implies the following important properties of the
Fourier coefficients:

Theorem 4.5.6 (Riemann-Lebesgue Lemma). If f (x) is integrable on [0, p], then

lim an = lim bn = lim cn = 0

n→∞ n→∞ n→∞

4.5.6 Isoperimetric Inequality

As an application of the Parseval’s identity, we prove the following classical result from
geometry.

Theorem 4.5.7 (Isoperimetric Inequality). If a differentiable simple closed curve has perime-
ter L, then the area
L2
A≤
4π
where the maximum is achieved when the curve is a circle.

Recall from Section 3.8.2 that if

r(t) = hx(t), y(t)i 0 ≤ t ≤ 2π

is the parametric equation of our simple closed curve, then the area is given by
Z Z 2π
A= xdy = x(t)y 0 (t)dt
C 0

First assume the curve is parametrized by arc length, i.e.

|r0 (t)|2 = x0 (t)2 + y 0 (t)2 = 1

so
R 2πthat it has
R 2πperimeter L = 2π. By shifting the picture, we can also assume that
0 x(t)dt = 0 y(t)dt = 0.
350 CHAPTER 4. SERIES

Writing down the Fourier series for x(t) and x0 (t) respectively:
∞
X
x(t) = cn eint , cn = 0
n=−∞
X∞
x0 (t) = incn eint
n=−∞

by Parseval’s Identity, we have

Z 2π ∞ ∞ Z 2π
1 X X 1
x(t)2 dt = |cn |2 ≤ n2 |cn |2 = x0 (t)2 dt.
2π 0 n=−∞ n=−∞
2π 0

Therefore, our area is

Z 2π
A= x(t)y 0 (t)dt
0
1 2π
Z
(2ab≤a2 +b2 ) ≤ x(t)2 + y 0 (t)2 dt
2 0
1 2π 0 2
Z
(P arseval) ≤ x (t) + y 0 (t)2 dt
2 0
1 2π (2π)2
Z
(|r0 (t)|=1) = dt = π =
2 0 4π

During the use of Parseval’ Identity, note that equality holds when we have cn = 0 for
|n| ≥ 2. This means
x(t) = a cos t + b sin t
During the use of 2ab ≤ a2 + b2 , equality holds when a = b, i.e. y 0 (t) = x(t), so that

y(t) = a sin t − b cos t

Therefore equality holds when the curve is a circle.

Finally, if we rescale our picture by a factor of r, then L = 2πr while

L2
A ≤ πr2 = .
4π
L2 L2
(Alternatively, the ratio 4πA is dimensionless, so it is invariant under rescaling and 1 ≤ 4πA
holds for all curves.)