Calculus and Applications I

D. T. Papageorgiou
Department of Mathematics
Imperial College London
London SW7 2AZ, UK

©Demetrios T. Papageorgiou (2023) These notes are provided for the personal
study of students taking this module. The distribution of copies in part or whole
is not permitted.
2
Contents

I Fundamentals of differential calculus 7

1 Limits of Functions, continuity 9

2 Derivative of a Function 19
2.1 Definition with limits, examples . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 General rules, chain rule, rates of change . . . . . . . . . . . . . . . . . . . 22
2.3 Implicit differentiation, related rates of change . . . . . . . . . . . . . . . 25

3 Mean Value and Intermediate Value Theorems 29

4 Inverse Functions 35

5 Exponentials and Logarithms 43

5.1 Geometrical Definition, Derivative . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Exponential as Inverse of log x . . . . . . . . . . . . . . . . . . . . . . . . 47
5.3 Function Estimates for Small and Large Arguments . . . . . . . . . . . . 49
5.4 Logarithmic Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.5 L’Hôpital’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

II Integration 57
6 Anti-derivatives and Geometrical Interpretation 59
6.1 Area under a curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2 Application to a first order differential equation . . . . . . . . . . . . . . . 62

7 The Riemann Sum 65

8 Properties of the Definite Integral; Fundamental Theorem of Calculus 71

9 Improper Integrals 73

10 Mean Value Theorem for Integrals 77

11 Techniques of Integration 79

12 Applications of Integration: Lengths, surfaces and volumes of revolu-

tion 83
12.1 Length of curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
12.2 Volumes and Volumes of Revolution . . . . . . . . . . . . . . . . . . . . . 86

3
4 CONTENTS

12.3 Surface Areas of Revolution . . . . . . . . . . . . . . . . . . . . . . . . . . 90

12.4 Centres of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
12.5 Moment of Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
12.6 Length of curves and areas using polar coordinates . . . . . . . . . . . . . 100

III Series, Power Series and Taylor’s Theorem 105

13 Series 107
13.1 Partial sums and geometric series . . . . . . . . . . . . . . . . . . . . . . . 108
13.2 Cauchy sequences and convergence of series . . . . . . . . . . . . . . . . . 109
13.3 Convergence tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

14 Power Series 121

14.1 Convergence tests and radius of convergence . . . . . . . . . . . . . . . . . 121
14.2 Differentiation and integration of power series . . . . . . . . . . . . . . . . 124

15 Power series solutions of differential equations 129

15.1 First order equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
15.2 Second order equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

16 Taylor Series 141

16.1 Taylor’s theorem with remainder . . . . . . . . . . . . . . . . . . . . . . . 142
16.2 Examples, bounding the remainder, estimates . . . . . . . . . . . . . . . . 144
16.3 Exponentials and logarithms. Binomial theorem . . . . . . . . . . . . . . 145

IV Fourier Series 149

17 Orthogonal and orthonormal function spaces 151

18 Periodic functions and periodic extensions 153

19 Trigonometric polynomials 157

19.1 Euler’s relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
19.2 Complex notation for trigonometric polynomials . . . . . . . . . . . . . . 158

20 Fourier series 161

20.1 Fourier series theorem, Riemann-Lebesgue Lemma . . . . . . . . . . . . . 162
20.2 Examples, sine and cosine series . . . . . . . . . . . . . . . . . . . . . . . . 166
20.3 Complex form of Fourier series . . . . . . . . . . . . . . . . . . . . . . . . 170
20.4 Fourier series on 2L−periodic domains . . . . . . . . . . . . . . . . . . . . 171
20.5 Parseval’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
20.6 Fourier transforms as limits of Fourier series . . . . . . . . . . . . . . . . . 173

V Further Applications in Physics and Geometry 175

21 Theory of plane curves 177
21.1 Parametric representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
21.2 Change of parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
CONTENTS 5

21.3 Motion along a curve with time as a parameter . . . . . . . . . . . . . . . 179

21.4 Arc length representation of curves . . . . . . . . . . . . . . . . . . . . . . 184
21.5 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

22 Laplace transforms and applications, the brachistochrone and Abel’s

mechanical problem 189
22.1 Definition and properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
22.2 Applications to differential equations . . . . . . . . . . . . . . . . . . . . . 191
22.3 Derivatives and integrals of Laplace transforms . . . . . . . . . . . . . . . 193
22.4 Convolution theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
22.5 The brachistochrone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
22.6 Abel’s mechanical problem and Abel’s integral equation . . . . . . . . . . 201

23 Newton’s law of gravitation, planetary motion and Kepler’s laws 205

23.1 Motion in a plane under the action of an arbitrary force . . . . . . . . . . 205
23.2 Special case I: Central forces . . . . . . . . . . . . . . . . . . . . . . . . . 206
23.3 Newton’s law of gravitation - central gravitational forces . . . . . . . . . . 207

VI Miscellaneous Problems 211

6 CONTENTS

Preface - how to read these notes

This set of notes is not extensive (still not exhaustive) but not everything in them
will be worked through in the face-to-face lectures. The student should strive to work
through the notes and also to solve problems assigned in the notes and also in the seven
problem sheets (one for each week of lectures).

A brief roadmap and guide of use is in order. There are six Parts. Part I - Fundamen-
tals of Differential Calculus - is mostly familiar from mathematics courses at A-level/IB.
The rigorous ε − δ definitions of limits are included for completeness and are covered in
detail in Analysis I. I include the brief discussion here to set the stage for the terminology
that will be used later. Part I will be covered very quickly due to its familiarity. Read
through and complete exercises. Part II is concerned with integration including im-
proper integrals and methods to test convergence, in addition to simple applications on
volumes and surfaces of revolution, centres of mass and moments of inertia (again a lot
of the material is probably familiar). Part III covers sequences and series (convergence
tests) from a practical viewpoint; theorems will be stated and proved for completeness
but again details are left for Analysis courses. Part III also covers power series and
Taylor series - applications to series solutions of differential equations are also included.
Part IV considered trigonometric approximation of single-variable functions and Fourier
series approximations of periodic functions. Part V has the general title Further Appli-
cations in Geometry and Physics and contains sections on the theory of plane curves,
Laplace transforms and applications (integral equations will be encountered here for the
first time), and a self-contained section on Newton’s laws of gravitation and their us in
describing planetary motion.

I will assume throughout that students have mastered the material in Maths and
Further Maths (or equivalent). In particular things like: differentiation of standard func-
tions (including product and chain rule); curve sketching (maxima/minima, inflection
points); integration of standard functions; techniques of integration (including integra-
tion by parts, integration by substitution, partial fractions); elementary mechanics. We
will be using all of these and I will assume full and masterful knowledge.

At the end of the notes I have part VI called Miscellaneous Problems. These problems
are mostly related to Parts I to IV and are additional to problems appearing in the notes
and in the seven problem sheets. You can figure out which part of the notes they refer to
but the line breaks sort of indicate approximate groupings. Some of these problems are
designed to be a bit more straightforward in an effort to allow the student to practice
before attacking more challenging problems.
 Part I

Fundamentals of differential
calculus

7
Chapter 1

Limits of Functions, continuity

All this material will be covered in a lot of detail in Analysis. I am including some basic
results that are useful for us going forward, and will not spend any time on going over
the material. Please go over this in your own time.
Given a function f (x) we are concerned with the behaviour near a point x = x0 , and
in particular the meaning of the statement limx→x0 f (x) = `, i.e. the limit of f (x) as x
tends to x0 , exists and is equal to `. The precise ε − δ definition is the following.

Definition 1. ε − δ Definition of Limit

Let f be a function defined at all points near x0 , except possibly at x0 , and let
` be a real number. We say that ` is the limit of f (x) as x approaches x0 , if for
every ε > 0 there exists δ > 0 such that |f (x) − `| < ε whenever |x − x0 | < δ and
x 6= x0 . We write limx→x0 f (x) = `.

Note: The definition implies that limx→x0 is unambiguous regarding taking limits
from above or below. If you see limx→x0 then it means that we must take the limit both
from above, i.e. limx→x0 + , and from below, i.e. limx→x0 − .
√ √
Example Prove that limx→2 x = 2. Before doing the ε − δ proof, we should note
√
that what we are asked to show appears trivially √ obvious. The function f (x) = x near
x = 2 is perfectly well-behaved and so f (2) = 2 is immediate. I am using the example
to illustrate the rigorous definition in anticipation of its use in less obvious situations.
√ √
Solution: Here f (x) = x, x0 = 2√and we want to prove that ` = 2. Given ε > 0 we
√
need to
√ √ find δ > 0 so√that |√ x − 2| < ε whenever |x − 2| < δ. For all x > 0 we have
x − 2 = (x − 2)/( x + 2), hence
√ √ |x − 2| |x − 2|
| x − 2| = √ √ ≤ √ .
x+ 2 2
√
Hence picking δ = 2 ε will do.

In practice we do not want to be doing ε − δ proofs for every limit we encounter.

9
10 CHAPTER 1. LIMITS OF FUNCTIONS, CONTINUITY

Instead we use the following laws of limits which can be proven easily using the ε − δ
definition.

Basic Properties of Limits

Assume that limx→x0 f (x) and limx→x0 g(x) BOTH EXIST. Then

(i) Sum rule:

lim [f (x) + g(x)] = lim f (x) + lim g(x). (1.1)
x→x0 x→x0 x→x0

(ii) Product rule:

lim [f (x) g(x)] = lim f (x) lim g(x). (1.2)
x→x0 x→x0 x→x0

(iii) Reciprocal rule: If limx→x0 f (x) 6= 0 then

lim [1/f (x)] = 1/ lim f (x). (1.3)

x→x0 x→x0

(iii)0 Quotient rule: If limx→x0 g(x) 6= 0 then

lim [f (x)/g(x)] = lim f (x)/ lim g(x). (1.4)

x→x0 x→x0 x→x0

This follows immediately from (ii) and (iii).

(iv) Composite function rule: If h(x) is continuous at limx→x0 f (x), then

 
lim h(f (x)) = h lim f (x) . (1.5)
x→x0 x→x0

Example 1  
Calculate limx→1 √x−1
x−1
.
Solution. Of the form “0/0”. Rationalise, i.e.
   √    √x + 1) 
x−1 (x − 1)( x + 1) −1)(
(x
lim √ = lim √ √ = lim = 2.

x→1 x−1 x→1 ( x − 1)( x + 1) x→1  −1)
(x 

Example 2
Sketch the function f (x) = x/|x|. Do this by considering x > 0 and x < 0 separately.
What happens when x = 0?
Solution. If x > 0, then f (x) = 1. If x < 0, then f (x) = |x| x
= −|x|
|x| = −1.
At x = 0 the function is undefined. Hence this function is defined everywhere except
at x = 0.A plot is given in Figure 1.1. The function is discontinuous at x = 0 as can be
seen from the jump in going from left to right (it is called a “step function” and can be
written as a linear combination of Heaviside functions - see below).
We have a problem, however. How do we define the function at x = 0? The answer
is that we can define it to be anything we want, and irrespective of our choice we can
 11

1.5

0.5
f(x)=x/|x|

-0.5

-1

-1.5

-2
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
x

x
Figure 1.1: The function f (x) = |x| .

never make the function continuous there! In figure 1.1 it is chosen to be zero (i.e. the
average of the values as I approach from the right and the left - we will encounter this
later when we do Fourier series). Here is a summary of what figure 1.1 depicts:
 x
|x| x 6= 0
f (x) = (1.6)
0 x=0
Note that:
1. limx→0+ f (x) = 1
2. limx→0− f (x) = −1
where x → 0± means take the limit as x tends to 0 from above/below. Of course
limx→0 f (x) 6= f (0) for this function, and so the function is not continuous. The defini-
tion of continuity is summarised below in Definition 5.

The properties given above also hold as x becomes large and positive or negative.
For example if f (x) = 1/x then we know that limx→±∞ f (x) = 0. Lets make this precise.

Definition 2. The ε − A definition of limx→∞ f (x) = `.

Let f (x) be defined on a domain containing the interval (a, ∞). A real number
` is the limit of f (x) as x approaches ∞ if, for every ε > 0 there exists a A > a,
such that |f (x) − `| < ε whenever x > A. We write limx→∞ f (x) = `. [Similarly
for limx→−∞ f (x) = `.]

NOTE: The limit properties (1.1)-(1.5) hold for limits of f (x) as x → ±∞, when the
limits are defined.

Consider next the limits limx→0 sin(1/x) and limx→0 (1/x2 ). The limits do not exist (I
cannot plug x = 0 into the functions). Sketch them and determine that they behave
12 CHAPTER 1. LIMITS OF FUNCTIONS, CONTINUITY

differently: the former is bounded, the latter is unbounded. In fact limx→0 (1/x2 ) = ∞.
More precisely we have:

Definition 3. ε − B definition of limx→x0 f (x) = ∞.

Let f (x) be a function defined in an interval containing x0 , except possibly at

x = x0 . We say that f (x) approaches ∞ as x approaches x0 if given any real
number B > 0, there exists ε > 0, so that whenever |x − x0 | < ε and x 6= x0 , we
have f (x) > B. We write limx→x0 f (x) = ∞. [Definition of limx→x0 f (x) = −∞
totally analogous.]

In the example f (x) = 1/x2 we found that as x → 0, f (x) → ∞ - the function is even,
so it does not matter if I approach the limit from the right (i.e. through positive values
of x) or the left (through negative x values). What about f (x) = 1/x? It is not hard to
see that as x tends to 0 through positive values then f → +∞, whereas as x tends to 0
through negative values we have f → −∞.
Hence, we need to define one-sided limits.

Definition 4. One-Sided Limits:

Let f (x) be defined for all x in an interval (x0 , a). We say that f (x) approaches
` as x approaches x0 from the right if, for any ε > 0, there exists a δ > 0, such
that for all x0 < x < x0 + δ we have |f (x) − `| < ε. We write limx→x0 + f (x) = `.
[Analogous definition for the left-sided limit, i.e. limx→x0 − f (x) = `.]

Note: If limx→x0 f (x) = +∞ or −∞, then the line x = x0 is a vertical asymptote.

Analogously, if limx→±∞ = `± then the lines y = `± are horizontal asymptotes.
1
As an Example consider f (x) = (x−1)(x−2) 2 . There are vertical asymptotes at x = 1

and x = 2 and a horizontal asymptote y = 0. Sketch the graph without using the
differentiation methods of finding critical points etc., that you are familiar with. Use
intuition and estimation.

In addition to the basic properties (1.1)-(1.5), there is another powerful test which is
very useful in calculations:

Comparison Test for Limits (a.k.a. Squeezing Property)

1. If limx→x0 f (x) = 0 and |g(x)| ≤ |f (x)| for all x near x0 with x 6= x0 , then
limx→x0 g(x) = 0.

2. If limx→∞ f (x) = 0 and |g(x)| ≤ |f (x)| for all large enough x, then
limx→∞ g(x) = 0.

Example
 13

Figure 1.2: Geometrical construction in the proof of the trigonometric limits (1.7)-(1.8).

(i) Establish Comparison Test 1 using the ε − δ definition of a limit.

(ii) Show that limx→0 x sin x1 = 0.

Solution
(i) Since limx→x0 f (x) = 0, then given ε > 0, there exists a δ > 0 such that |f (x)| < ε
when |x − x0 | < δ. For the same ε and δ, we also have |g(x)| < ε when |x − x0 | < δ,
since |g(x)| ≤ |f (x)|. Hence limx→x0 g(x) = 0 also.
(ii) Take g(x) = x sin(1/x) and f (x) = x. Then |g(x)| ≤ |x| for all x 6= 0, so the
comparison test applies. Clearly limx→0 x = 0, hence the result follows.

Two Basic Trigonometric Limits

We will need the following results in finding derivatives of sin x and cos x from
first principles.
sin h
lim =1 (1.7)
h→0 h
cos h − 1
lim =0 (1.8)
h→0 h

The proof of the former is geometrical and the construction is given in Figure 1.2.
OBC is the sector of a circle of radius 1 with subtended angle h. The two triangles
OAB and OCD are constructed as shown with BD the extension of OB. Considering
triangles OAB and OCD we have
AB DC
sin h = = s, tan h = = t.
OB OC
From geometry we have the following inequality
area of triangle OAB < area of sector OCB < area of triangle OCD,
which in turn provides
1 h 1
sin h cos h < < tan h.
2 2 2
14 CHAPTER 1. LIMITS OF FUNCTIONS, CONTINUITY

The middle quantity follows by noting that the area of the sector OCB is equal to h/2π
times the area of a circle of unit radius which is π. Considering the first inequality (after
canceling the 1/2 factor throughout) we have

sin h 1
sin h cos h < h ⇒ < .
h cos h
The above is fine since h and cos h are positive and non-zero so I can divide by them.
The second inequality gives

sin h sin h
h< ⇒ cos h < .
cos h h
Putting these together gives
sin h 1
cos h < < .
h cos h
As h tends to zero cos h tends to 1, hence sin h/h is squeezed between two numbers that
tend to 1. By the Squeezing Property we get the desired result.
To prove the second result we write

cos h − 1 cos h − 1 cos h + 1 cos2 h − 1

= =
h h cos h + 1 h(cos h + 1)
2  
− sin h sin h sin h
= = − .
h(cos h + 1) h cos h + 1

Using the product rule for limits, it follows immediately that

   : −1
  *0 
 : 1/2

sin h sin h −sinh  1
lim − = lim
  lim sin
 h lim   = 0.
h→0 h cos h + 1 h→0 h 
h→0
cos h + 1
h→0

Continuity
Looking back at Definition 1, the ε − δ definition of a limit, we can see that it is
equivalent to the statement

lim f (x0 + h) = `.
h→0

We can then define what we mean by continuity of a function f (x) at a point x0 .

Definition 5. Continuity

We say that f is continuous at x0 if limh→0 f (x0 + h) = f (x0 ). Equivalently

limx→x0 f (x) = f (x0 ).
A totally equivalent definition is: f (x) is continuous at a point x0 if for every
ε > 0 there exists a number δ > 0 such that |f (x) − f (x0 )| < ε for all x in the
domain of f for which |x − x0 | < δ.
 15

We have seen examples of functions that are not continuous, e.g.

x2
 
1 x>0 x 6= 0
f (x) = g(x) =
0 x≤0 1 x=0

are both not continuous at x = 0. If I exclude x = 0, then the limits as x → 0 exist,

limh→0+ f (h) = 1, limh→0− f (h) = 0, and limh→0 g(h) = 0. (Note that limh→0+ =
limh→0,h>0 and limh→0− = limh→0,h<0 .)

Removable discontinuities
We will encounter functions that are perfectly nice everywhere except at a point or
points. Here are some examples.

 
1
f (x) = sin (1.9)
x
 
1
f (x) = x sin (1.10)
x

Both of these functions are well behaved everywhere except at x = 0. We say that the
functions are not defined at x = 0.
The question is: Can we define f (0) by a number of our choice so as to make the
functions (1.9) and (1.10) continuous?
For the former function this is impossible - the function oscillates infinitely many
times as x → 0 and takes on all values between −1 and 1. We can never satisfy the ε − δ
definition of continuity.
However, for the function in (1.10) we can remove the discontinuity by picking f (0) =
0, i.e.

1


x sin x x 6= 0
f (x) = (1.11)
0 x=0

Continuity at x = 0 is now guaranteed because

|x sin(1/x) − 0| ≤ |x| < ε ∀ |x| < ε.

(Note that here δ(ε) = ε exactly.)

Plots of the two functions (1.9) and (1.10) are given in Figure 1.3.
16 CHAPTER 1. LIMITS OF FUNCTIONS, CONTINUITY

f(x)=sin(1/x)
1

0.5

-0.5

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
f(x)=x sin(1/x)
0.8
0.6
0.4
0.2
0
-0.2
-0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Figure 1.3: The functions sin(1/x) and x sin(1/x). Only x > 0 is shown, the functions
are odd and even, respectively, with respect to x.

Miscellaneous Solved Examples

√ 
1. Find limx→∞ x2 + 1 − x , and interpret the result geometrically by considering
a right angled triangle with base of length x and unit height.
We calculate
√ 
p  p  x2 + 1 + x 1
x2 + 1 − x = x2 + 1 − x √ =√ .
2
x +1+x
x2 + 1 + x
√
As x becomes arbitrarily large 2
√  then 1/( x + 1 + x) becomes arbitrarily small, and
hence limx→∞ x2 + 1 − x = 0. [Can you prove this using the ε − A definition
of limit?]
Geometrical√picture for you to do. Hint: The right angled triangle suggested has
hypotenuse x2 + 1. Consider a circle of radius
√ x whose arc cuts the hypotenuse
2
at a point, and figure out what the quantity x + 1 − x represents geometrically.
√

2. Now consider limx→∞ x − x + 1 . Find the limit in this case.
Don’t√need to do much here. Main thing is to notice√ that
x is much much bigger
than x + 1 when x is large. Hence, limx→∞ x − x + 1 = ∞.
A precise definition in this case (for a general function f (x)) would be: We say
limx→∞ f (x) = ∞, if given an arbitrarily large A > 0, there exists a number
M > 0, so that f (x) > A for all x > M .
√
In our particular example where f (M ) = M − M + 1 it is easy to see that taking
M = A2 will do the trick. Of course it can be proven for smaller M but we are
not looking for anything sharper than a proof.
 17

1 1−x2
3. Find (a) limx→1 (x−1)2
, and (b) limx→∞ x3/2
.
For (a) as x → 1 (from above or below), then (x − 1)2 becomes arbitrarily small.
1
Its inverse becomes arbitrarily large, so limx→1 (x−1)2 = ∞

Intuitive answer is: For (b) as x becomes very large then x2  x3/2 , hence the
limit is −∞. Can formalize as follows

1 − x2
lim = lim (x−3/2 − x1/2 ) = −∞.
x→∞ x3/2 x→∞

Problems
1. For the function
x
x 6= 0

f (x) = |x|
c x=0

where c ∈ R, use the ε − δ definition to prove that the function is not continuous
at x = 0.
2. Consider the function

x1/2

x≥0
f (x) =
x2 x<0

Clearly f (0) = 0 and the function is continuous there. Prove continuity using ε−δ.
[Hint: The task here is, given a ε to find one δ that will work for both x > 0 and
x < 0.]
18 CHAPTER 1. LIMITS OF FUNCTIONS, CONTINUITY
Chapter 2

Derivative of a Function

2.1 Definition with limits, examples

Consider graphs of functions y = f (x). We need to define the derivative or slope of the
curve at a given point P .

Figure 2.1: Slope of f at P is the slope of the line QP as Q tends to P . Note: The Qs
are to the right of P , the definition is the same is Q1 , Q2 etc are to the left of P .

Definition of Differentiability

The function f (x) is differentiable at x if ‘Newton’s quotient’,

f (x + h) − f (x)
lim
h→0 h
exists. We call this f 0 (x), the derivative of f at point x.

19
20 CHAPTER 2. DERIVATIVE OF A FUNCTION

Examples

(i) Is f (x) = x2 differentiable everywhere?

(x + h)2 − x2 x2 + 2xh + h2 − x2
f 0 (x) = lim = lim
h→0 h h→0 h
= lim (2x + h) = 2x ⇒ YES
h→0

(ii) Is f (x) = |x| differentiable at x = 0? Draw a picture.

Need to check if the limit exists and the values are equal as we approach 0 from
above or below.
(a)
f (0 + h) − f (0) h−0
lim = lim =1
h→0, h>0 h h→0, h>0 h
(b)
f (0 + h) − f (0) −h − 0
lim = lim = −1
h→0, h<0 h h→0, h<0 h
Right and left derivatives exist but are not equal.

A function is differentiable at x if right and left derivatives exist and are

equal.

Exercise: Sketch the derivative of f (x) = |x|.

f (x) = |x| is continuous at x = 0, but not differentiable there. Geometrically we can
see this - there is a ‘corner’ in the graph.
Note: f (x) is not continuous at x = x0 if the limit limx→x0 f (x) does not exist. We
already gave the ε − δ definitions.
2.1. DEFINITION WITH LIMITS, EXAMPLES 21

Example: Consider (
x if 0 < x ≤ 1
f (x) =
x−1 if 1 < x ≤ 2
Here is the graph:

What is the derivative at x = 1?

(a) Left derivative at x = 1

f (1 + h) − f (1) 1+h−1
lim = lim =1
h→0, h<0 h h→0, h<0 h

(b) Right derivative at x = 1

f (1 + h) − f (1) (1 + h − 1) − 1
lim = lim as f (1) = 1
h→0, h>0 h h→0, h>0 h
1
= lim (1 − )
h→0, h>0 h

which does not exist. In fact, → −∞. Function has no right derivative.

2.1.1 Polynomials

Theorem 1
Let n be an integer ≥ 1 and let f (x) = xn . Then

df
f 0 (x) = = nxn−1 .
dx
22 CHAPTER 2. DERIVATIVE OF A FUNCTION

Proof.
f (x + h) = (x + h)n = xn + nxn−1 h + h2 g(x, h)
where g(x, h) involves powers of x and h with some numerical coefficients. We don’t
care what it is exactly but limh→0 g(x, h) = some number. Then
df xn + nxn−1 h + h2 g − xn
= lim = nxn−1 .
dx h→0 h

Theorem 2
Let f (x) = xa , where a is any real number and x > 0. Then f 0 (x) = axa−1 .
If a is a negative integer then this is easy. General case is different from proof
above.

2.2 General rules, chain rule, rates of change

2.2.1 General rules
(i) If c is a constant, (cf )0 (x) = cf 0 (x).
(ii) If f (x), g(x) are given functions and f 0 (x), g 0 (x) exist, then
(f + g)0 (x) = f 0 (x) + g 0 (x).
(iii) (f g)0 (x) = f 0 (x)g(x) + f (x)g 0 (x) (product rule)
(iv) Let g(x) be a function that has a derivative g 0 (x) and such that g(x) 6= 0.
Then
g 0 (x)
 
d 1
=−
dx g(x) (g(x))2
(iv)*
g(x)f 0 (x) − f (x)g 0 (x)
 
d f (x)
= .
dx g(x) (g(x))2

Proof. (iii) - (do the rest yourselves as an exercise)

f (x + h)g(x + h) − f (x)g(x)
(f g)0 = lim
h→0 h
=0
z }| {
f (x + h)g(x + h) − f (x)g(x) −f (x)g(x + h) + f (x)g(x + h)
= lim
h→0 h
(f (x + h) − f (x)) g(x + h) + (g(x + h) − g(x)) f (x)
= lim
h→0 h
= g(x)f 0 (x) + f (x)g 0 (x)
2.2. GENERAL RULES, CHAIN RULE, RATES OF CHANGE 23

2.2.2 The Chain Rule

Composition (f ◦ g)(x) = f (g(x)) is a function constructed as follows: Take a number
x, find g(x) and then take the value of f at g(x).
√
e.g. f (x) = x2 g(x) = x defined for x > 0.
√
Then (f ◦ g)(x) = ( x)2 = x,
p
and (g ◦ f )(x) = (x2 ) = x.
Discussion: In the example above we had (f ◦ g)(x) = (g ◦ f )(x). Is this generally true?
If yes prove it, if not provide a counterexample.

Theorem 3: Chain Rule.

Let f , g be two functions having derivatives and such that f is defined for all
numbers that are values of g. Then
d
(f ◦ g)(x) = (f ◦ g)0 (x) = f 0 (g(x))g 0 (x)
dx

Why is this useful?

d
 3
(x + 9x2 + π)51

Example: dx
Last thing you want to do is multiply out the 51 factors and then differentiate. Of
course a symbolic manipulator in Matlab, Mathematica, Python etc., would do it, but
let’s go green and save some computer energy!
With the chain rule we identify
f (x) = x51 g(x) = (x3 + 9x2 + π).
So that
(x3 + 9x2 + π)51 = (f ◦ g)(x)
d 3
Then (x + 9x2 + π)51 = 51(x3 + 9x2 + π)50 (3x2 + 18x)
dx

Proof.
f (g(x + h)) − f (g(x))
(f ◦ g)0 (x) = lim
h→0 h
f (g(x + h)) − f (g(x)) g(x + h) − g(x)
= lim ·
h→0 g(x + h) − g(x) h
Let k = g(x + h) − g(x), (if h 6= 0, k 6= 0), and write u = g(x). Then
f (u + k) − f (u) g(x + h) − g(x)
(f ◦ g)0 (x) = lim ·
h→0
 k  h 
f (u + k) − f (u) g(x + h) − g(x)
= lim lim
h→0 k h→0 h
0 0 0 0
= f (u)g (x) = f (g(x))g (x)
24 CHAPTER 2. DERIVATIVE OF A FUNCTION

Analogous definition of the derivative f 0 (x) is the following:

f (y) − f (x)
f 0 (x) = lim .
y→x y−x
To see equivalence write y = x + h.

Application: Particle motion (rectilinear for the moment).

Let the position of a particle at time t be s = f (t), say. If the particle moved from
P1 at t = t1 to P2 at t = t2 , then its

f (t2 ) − f (t1 )
average speed = .
t2 − t1
So instantaneous speed at any time t is

f (t) − f (t0 )
f 0 (t) = lim rate of change.
t→t0 t − t0

f 0 (t) is also a function, call it v(t). If it is differentiable then

d2 f
v 0 (t) = is the acceleration.
dt2
Can define higher derivatives (if they exist) by continuing this process.

Theorem 4
If f (x) is differentiable at x = x0 , then it is also continuous there.

Question: Is the converse true?

Proof.
 
f (x) − f (x0 )
lim (f (x) − f (x0 )) = lim · x − x0
x→x0 x→x0 x − x0
 
f (x) − f (x0 )
= lim · lim (x − x0 )
x→x0 x − x0 x→x0
0
= f (x0 ) · 0 = 0
2.3. IMPLICIT DIFFERENTIATION, RELATED RATES OF CHANGE 25

2.3 Implicit differentiation, related rates of change

Recall: We saw that if n is an integer then
d n d −n
x = nxn−1 , (x ) = −nx−(n+1) .
dx dx
This also holds if,
(i) y = x1/n where n is an integer (and x > 0, why the restriction?),
(ii) y = xr where r is a rational number; i.e. r = pq , with p, q integers.

Can prove these using implicit differentiation. Start with (i) y = x1/n , n integer.
Assume x1/n is defined.
Then
d n d
yn = x ⇒ (y ) = (x) ⇒
dx dx
dy dy 1 1 1−n 1 1
ny n−1 =1 ⇒ = y 1−n = x n = x n −1 .
dx dx n n n
e.g.
d 1 1 4
x 5 = x− 5
dx 5

p 1
(ii) y = x q . Let g(x) = x q where q is an integer. Then y = (g(x))p with p an integer.
Use chain rule.
dy 1 1 −1 p p −1
= pg p−1 x q = x q
dx q q
These of course generalize to powers of the function.
e.g.
d
(f (x))r = rf r−1 f 0 , r rational.
dx

Example of implicit differentiation:

Find the equation of the tangent line to the curve 2x6 + y 4 = 9xy at the point (1, 2).
Solution: Note that we cannot solve for y as a function of x. Hence implicit differenti-
ation is very powerful here. Calculate the derivative
dy dy
12x5 + 4y 3 = 9y + 9x .
dx dx
Substitute point (1, 2)
dy dy
12 + 32 = 18 + 9
dx dx
dy 6
⇒ = is the slope of the tangent line
dx 23
26 CHAPTER 2. DERIVATIVE OF A FUNCTION

6
Its equation is y − 2 = 23 (x − 1).

Can also use implicit differentiation to obtain related rates of change.

If x and y are both functions of a parameter t, then we can differentiate implicitly
with respect to t
e.g. x = cos(t), y = sin(t), t ≥ 0

Equation is x2 + y 2 = 1, where x = x(t), y = y(t). Differentiate implicitly with

respect to t.

dx dy dy x dx
2x + 2y =0⇒ =−
dt dt dt y dt
dy/dt x
i.e =− .
dx/dt y

dy
√
This is the derivative of dx if we think of y = ± 1 − x2 as a function of x. (Another
dy dy dy/dt
way is x2 + y 2 = 1 → 2x + 2y dx = 0 → dx = − xy = dx/dt .)

In general, if x = f (t) and y = g(t), describe a curve in the plane called a parametric
curve, then the slope of the tangent line to the curve is given by

dy dy/dt dx
= if 6= 0.
dx dx/dt dt

To prove this, note that the curve can be defined (piecewise) as the graph of a
function y = h(x) or x = H(y). Chain rule gives dy dy dx
dt = dx dt as required.

Example
The surface area of a cube is growing at a constant rate of 4cm2 /s. How fast is the
length of a side growing when the cube sides are 2cm long? Find the side length when
the rate of change of the volume exceeds that of the area.
2.3. IMPLICIT DIFFERENTIATION, RELATED RATES OF CHANGE 27

Solution
dA dx dx 1 dA 4
A = 6x2 ⇒ = 12x ⇒ = =
dt dt dt 12x dt 12x
dx 1
If x = 2 = cm s−1
dt 6
dV dx 3x2 dA
V = x3 ⇒ = 3x2 = .
dt dt 12x dt
dV dA
So if x > 4, dt > dt in numerical value.
28 CHAPTER 2. DERIVATIVE OF A FUNCTION
Chapter 3

Mean Value and Intermediate

Value Theorems

For a function f (x) which is defined at a point c, we say that c is a maximum of f if

f (c) ≥ f (x) ∀x where f is defined.

For a minimum we have

f (c) ≤ f (x).

e.g.

Here is a result when f is differentiable:

Theorem 1
Let f be a function which is defined and differentiable on the open interval (a, b).
Let c be a number in the interval which is a maximum for the function.

Then f 0 (c) = 0. f 0 (c) = 0 also, if c is a minimum of f .

Proof. Obvious, here is a geometrical interpretation.

29
30 CHAPTER 3. MEAN VALUE AND INTERMEDIATE VALUE THEOREMS

As h → 0, the slope → 0, ⇒ f 0 (c) = 0.

Detailed proof:
f (c) ≥ f (c + h) ⇒ f (c + h) − f (c) ≤ 0.

f (c + h) − f (c)
i.e. lim ≤0
h→0, h>0 h

Similarly for left limit

f (c) − f (c − h)
lim ≥ 0.
h→0, h>0 h

As h → 0 these can only be equal if f 0 (c) = 0 since the function is differentiable.

All points c such that f 0 (c) = 0 are called critical points.

Definition
f (x) is said to be continuous on an interval [a, b] if limx→x0 f (x) = f (x0 ) for all
x0 in [a, b]. Analogously, limh→0 f (x0 + h) = f (x0 ) a ≤ x0 ≤ b.
 31

Theorem 2
Let f (x) be continuous on the closed interval [a, b]. Then f (x) has a maximum
and a minimum on this interval. I.e there exist c1 and c2 so that f (c1 ) ≥ f (x)
and f (c2 ) ≤ f (x) for all x in [a, b].

e.g

Theorem 3 (Combines Theorems 1 and 2)

Let f (x) be continuous over the closed interval a ≤ x ≤ b and differentiable on
the open interval a < x < b. Assume also that f (a) = f (b) = 0.
Then there exists a point c, a < c < b, such that f 0 (c) = 0.

Proof. If f (x) = 0 then there is nothing to prove. If f (x) 6= 0 then it must have at least
one maximum or minimum or both. Denote by c such points. By Theorem 1 we have
f 0 (c) = 0.

Example:

Here there are 3 such points.

32 CHAPTER 3. MEAN VALUE AND INTERMEDIATE VALUE THEOREMS

Theorem 4
Suppose f is continuous on [a, b] and differentiable on (a, b).
Then there exists a < c < b such that
f (b) − f (a)
f 0 (c) =
b−a

Geometrical interpretation:

f (b)−f (a)
The point x = c is where the tangent has the slope b−a .

Proof. The straight line joining (a, f (a)) and (b, f (b)) has equation
y − f (a) x−a f (b) − f (a)
= ⇒ y= (x − a) + f (a).
f (b) − f (a) b−a b−a
Consider g(x) = f (x) − f (b)−f b−a
(a)
(x − a) − f (a). Then g(a) = 0, g(b) = 0, and by
Theorem 3 there exists a c with a < c < b such that g 0 (c) = 0.
But g 0 (x) = f 0 (x) − f (b)−f
b−a
(a)
, and the result follows.

Definition
We say that a continuous function f (x) is increasing over a given interval if
given x1 , x2 in the interval with x1 ≤ x2 , we have f (x1 ) ≤ f (x2 ).

Strictly increasing if f (x1 ) < f (x2 ) when x1 < x2 .

Strictly decreasing if f (x1 ) > f (x2 ) when x1 < x2 .
 33

Theorem 5
Let f (x) be continuous in the closed interval [a, b] and differentiable in the open
interval I = (a, b).
If f 0 (x) = 0 in I, then f is constant.
If f 0 (x) > 0 in I, then f is strictly increasing.
If f 0 (x) < 0 in I, then f is strictly decreasing.

Proof. Use the mean value theorem.

Let x1 , x2 be points in the interval with x1 < x2 . Then there exists x1 < c < x2
such that

f (x2 ) − f (x1 )
f 0 (c) = =⇒ f (x2 ) − f (x1 ) = (x2 − x1 )f 0 (c)
x2 − x2

If f 0 (x) = 0 in the interval, f 0 (c) = 0 and f (x2 ) = f (x1 ) i.e f is constant. If f 0 (x) > 0
then f 0 (c) > 0 and f (x2 ) > f (x1 ), i.e strictly increasing. If f 0 (x) < 0 then f 0 (c) < 0 and
f (x2 ) < f (x1 ), i.e strictly decreasing.

Example 1 (Do yourself ). Determine the region of increase and decrease of the function
f (x) = x3 − 2x + 1.

Example 2 Prove that sin(x) ≤ x for x ≥ 0.

Solution Let f (x) = x − sin(x). Then f (0) = 0.
f 0 (x) = 1 − cos(x) ≥ 0 for all x. Hence f (x) is an increasing function ⇒ f (x) ≥ 0
for all x.

Theorem 6 - Intermediate value theorem

Let f be continuous on the closed interval a ≤ x ≤ b. Given any number y ∗
between f (a) and f (b), there exists a point x∗ between a and b such that f (x∗ ) =
y∗.

Picture where it works:

34 CHAPTER 3. MEAN VALUE AND INTERMEDIATE VALUE THEOREMS

Note: there can be more than one x∗ .

Picture where is does not work because the function is not continuous:
Chapter 4

Inverse Functions

Given y as a function of x, when can we express x as a function of y? Here is an easy

case:
1
y = 3x + 1 ⇒ x = (y − 1)
3
Usually we do not have a formula like this, but we can say a lot about the function
x = g(y).

Definition
Let y = f (x) be defined on some interval. Given any y0 in the range of f , if we
can find a unique value x0 in its domain such that f (x0 ) = y0 , then we can
define the inverse function

x = g(y) (sometimes written x = f −1 (y) )

Clearly we have
f (g(y)) = y and g(f (x)) = x
or f (f −1 (y)) = y and f −1 (f (x)) = x

35
36 CHAPTER 4. INVERSE FUNCTIONS

Question: When can we be certain an inverse function exists?

Theorem 1
Let f (x) be strictly increasing or strictly decreasing. Then the inverse function
exists.

Proof. Obvious from definition of strictly increasing/decreasing.

Theorem 2
If f (x) is continuous on [a, b] and is strictly increasing (or decreasing), and f (a) =
ya and f (b) = yb , then x = g(y) is defined on [ya , yb ].

Proof. Easy by the intermediate value theorem.

Here is what goes wrong if we drop continuity

4.0.1 Derivative of inverse functions

Theorem 3
Let f (x) be differentiable on (a, b) and f 0 (x) > 0 or f 0 (x) < 0 for all x in (a, b).
Then the inverse function exists and we have

d −1 1
g 0 (y) = f (y) = 0 .
dy f (x)

Proof. Need to find

g(y + k) − g(y)
lim
k→0 k
where we have y = f (x). Here is a useful picture:
 37

Going from y to y + k, increases x to x + h. The intermediate value theorem ensures

that this is true for all k in fact.
Hence f (x + h) = y + k ⇒ g(y + k) = x + h. Since f (x) = y ⇒ g(y) = x.
Back to the limit
g(y + k) − g(y)
lim
k→0 k
h 1
lim = = 0
h→0 f (x + h) − f (x) f (x)

Chain rule way: g(y) = x where y = f (x). (In fact x = g(y), iff y = f (x)). i.e

dg 0 1
f (x) = 1 g 0 = 0
dy f (x)

Example: Consider
y = x4 + 3x3 + x − 5, x > 0.
Find g 0 (0) - i.e d
dy f −1 (y) y=0
. Note: We will not even attempt to find x = g(y).
dg
Solution: Theorem says dy = g 0 (y) = f 01(x) where y = f (x). If y = 0 then need to
solve 0 = f (x) by inspection f (1) = 0 ⇒ f 0 (1) = 14 ⇒ g 0 (0) = 14
1
.
The solution above implicitly assumed that x = 1 is the only point where f (x) = 0.
Need to justify this. The key is the observation that f 0 (x) = 4x3 + 9x2 + 1 which is > 0
for all x > 0. Hence the function is strictly increasing in the given interval and since
f (0) = −5 < 0 and limx→+∞ = +∞, by the intermediate value theorem there must be
a unique x s.t. f (x) = 0. Of course this happens to be x = 1 by inspection.
Here are some additional things for you to consider: Suppose x < 0 now.
(i) Prove that there is at least one x < 0 such that f (x) = 0. Let this point(s) be
x = ξ, i.e. f (ξ) = 0, ξ < 0.
38 CHAPTER 4. INVERSE FUNCTIONS

(ii) Prove that there is at least one point ξ < c < 0 where f 0 (c) = 0.

Note: Very useful in solving problems.

dx 1
x = g(y) i.e. some function of y. Our theorem really says dy = dy/dx .

4.0.2 Some special inverse functions

(i) The arcsin, or sin−1 .
Consider y = sin(x) (shown below)

So given any −1 ≤ y ≤ 1 there are an infinite number of x such that y = sin(x). If

we restrict our domain to regions where sin(x) is strictly increasing or decreasing,
then we can find inverse functions - we have a theorem. By convention we take
− π2 ≤ x ≤ π2 for the domain (the range is of course [−1, 1]).
d
Now dx sin(x) > 0 if − π2 < x < π2 , and d
dx sin(x) = 0 at x = ± π2 .
Let the inverse function be g(y) = x. Then

1 1 π π
g 0 (y) = = > 0 x ∈ (− , ).
f 0 (x) (sin(x))0 2 2
 39

dx 1 1 1
g 0 (y) = = =p 2
=p .
dy cos(x) 1 − sin (x) 1 − y2

Think of y as a dummy variable now. Then arcsin(y) = sin−1 (y) is a function with
domain [−1, 1] and range [− π2 , π2 ]. Instead of y use x, i.e. y = f (x) = sin−1 (x).
Then
dy d 1
= sin−1 (x) = √ .
dx dx 1 − x2

Once you have identified the domain and range where the inverse function exists,
there is an easier way (equivalent) to find derivatives.
Start with y = sin−1 (x) i.e y is the angle whose sin is x. Then sin(y) = x. By the
dy dy 1 1
chain rule, (cos(y)) dx = 1 ⇒ dx = cos(y) = √1−x 2
as shown above.

(ii) arctan or tan−1

Consider y = tan(x)

Can define tan(x) on (− π2 , π2 ), dx

d
tan(x) = 1+tan2 (x) > 0. So x = g(y) the inverse
function has domain (−∞, ∞) and range (− π2 , π2 ).
Derivative of tan−1 :

y = tan−1 (x)
tan(y) = x
dy dy 1
(1 + tan2 (y)) =1 ⇒ =
dx dx 1 + x2

Here is the graph of y = tan−1 (x)

40 CHAPTER 4. INVERSE FUNCTIONS

Note for later material: we showed that dx d

tan−1 (x) = 1+x
1
2. In other words,
the anti-derivative of 1+x2 is tan−1 (x) + c where c is a constant. Similarly, the
1
1
anti-derivative of √1−x 2
is sin−1 (x) + c.

Miscellaneous Problems

1. For the following functions construct specific ε−δ definitions of continuity at x = 0.

In other words given a ε you need to find δ(ε).

x for x ≥ 0
f (x) =
x2 for x < 0

x for x ≥ 0
g(x) =
|x|1/2 for x < 0

2. Consider the function

x2

x≥0
f (x) =
−x2 x<0

(a) Is f (x) a continuous function?

(b) Show that f 0 (0) exists and find its value.
(c) Define g(x) = f 0 (x), x 6= 0, and g(0) = f 0 (0). Determine whether g(x) is
differentiable or not.
(d) If instead of x2 in the definition of f (x) we had xn where n is a positive
integer. How many derivatives of f (x) would exist in this case?
3. A spherical balloon is being blown up by injecting air into it at 1 liter per second.
When its radius is 1 m, find the rate at which its area is increasing (pay attention
to the units).
4. Sand is being piled onto a conical pile at a constant rate of R cm3 /s. As the pile
grows, frictional forces between sand particles constrain the height of the pile to
be equal to the radius of its base.
(i) When the height equals 1 cm, find the rate at which it is increasing.
 41

(ii) If the height at time t is h(t), find an explicit expression for it. What happens
to its rate of change as t becomes large? Explain physically/intuitively.
5. (a) Determine the regions of increase and decrease of the function f (x) = x3 −
2x + 1.
(b) Sketch functions for which the intermediate value theorem holds and:
(i) For a chosen y ∗ there are at most two values of x∗ .
(ii) For a chosen y ∗ I can choose an interval [a, b] to have as many x∗ as I
want. [Any guess as to what function this is?]
(iii) For a chosen y ∗ there does not exist a x∗ , i.e. Theorem 6 does not hold.

6. Find tan−1 (tan 3π

4 ) and arctan(tan 2π). (No computers/calculators!)
42 CHAPTER 4. INVERSE FUNCTIONS
Chapter 5

Exponentials and Logarithms

Summary of what we know:

If a > 0 is any real number and r any rational number, we know how to define ar .
What about ax where x is any real number (including irrationals)? The quantity ax is
also a continuous function of x.
Intuitively, we can imagine that any irrational x can
√ be approximated by rationals
and accurate values obtained.
√ is an example: 2 3 . A decimal approximation of
√ Here17321
1732
3 ≈ 1.732050808 ⇒ 1000 < 3 < 10000 . Now, 2x is an increasing function, therefore

1732
√ 17321
√
3 3
2 1000 < 2 < 2 10000 i.e. 2 ≈ 3.322 correct to 3 decimal places.

There is a way of defining ax and deriving all its properties by using properties of
real numbers - though this is technical and we do not have time to do it.
We will do it in a more intuitive way that may seem a bit unnatural at first, namely by
defining a new function (the logarithm) and then defining the exponential as the inverse
function of the logarithm. Advantage of this - intuitive, simple and clear arguments.

5.1 Geometrical Definition, Derivative

The following leads to the natural logarithm.

Definition
The quantity log(x) is the area under the curve x1 between 1 and x if x ≥ 1; and
negative the area under the curve x1 between 1 and x if 0 < x < 1. In
particular, log(1) = 0.

43
44 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

Hence log(x) ≥ 0 if x ≥ 1; and log(x) < 0, 0 < x < 1.

Theorem 1
d
log(x) is differentiable and dx log(x) = x1 .

Proof. Need to consider the Newton quotient and prove

log(x + h) − log(x) 1
lim = .
h→0 h x
Start with h > 0 and consider the area under the curve between x and x + h.

log(x + h) − log(x) is the shaded area above. From geometry we have (or from the
fact that x1 is a decreasing function):
5.1. GEOMETRICAL DEFINITION, DERIVATIVE 45

1 1
· h < log(x + h) − log(x) < ·h
x+h x
1 log(x + h) − log(x) 1
⇒ < <
x+h h x
(here h > 0 so we can divide by h as done above.) As h → 0 we use the squeezing
theorem to get the required limit.
If h < 0 the picture is:

So
1 1
−h · < log(x) − log(x + h) < −h ·
x x+h
1 log(x + h) − log(x) 1
⇒ < <
x+h h x
1
Hence limit is x as h → 0.
d
log(x) is a function defined for x > 0, has log(1) = 0 and dx log(x) = x1 . (*)
We will use (*) alone in what follows. If g(x) is another such function then g(x) =
log(x) uniquely. The condition log(1) = 0 fixes this.

Theorem 2
If a, b > 0, then log(ab) = log(a) + log(b).

Proof. Let f (x) = log(ax), x > 0 and a as above.

df 1 1
= ·a= by the chain rule.
dx ax x
i.e. same derivative as log(x) ⇒ they differ by a constant.
46 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

⇒ log(ax) = log(x) + K ∀x > 0

Put x = 1 ⇒ K = log(a). Put x = b ⇒ log(ab) = log(b) + log(a).

Theorem 3
log(x) is strictly increasing for all x > 0. Its range is (−∞, ∞).

Proof.
d 1
log(x) = > 0 for all x > 0 ⇒ strictly increasing.
dx x
To prove that it takes on arbitrarily large values, note that since it is strictly increas-
ing and log(1) = 0, we must have, for example, log(2) > 0.
From Theorem 2,

n terms
z }| {
n
log(2 ) = log(2 · 2 · · · · · 2) = log(2) + log(2) + · · · + log(2) = n log(2).

This holds for any positive integer and log(2) > 0 ⇒ as n becomes large, so does
log(2n ). To prove it takes on arbitrarily large negative values, note that
   
1 1
0 = log(1) = log 2 · = log(2) + log
2 2
 
1
⇒ log = − log(2)
2

Hence
 
1
log n = −n log(2) by Theorem 2
2
→ −∞ as n → ∞

Theorem 4
If n is an integer (positive or negative) then log(an ) = n log(a) for all a > 0.

Proof. As above - simple use of Theorem 2.

5.2. EXPONENTIAL AS INVERSE OF log X 47

Domain (0, ∞), range (−∞, ∞), strictly increasing.

5.2 Exponential as Inverse of log x

Define the exponential as the inverse function of log(x). We know it exists - write it as
exp(x). Since 0 = log(1) by inverse we have exp(0) = 1.

Theorem 5
If x1 , x2 are two numbers, then exp(x1 + x2 ) = exp(x1 ) · exp(x2 ).

Proof. Let a = exp(x1 ), b = exp(x2 ). By inverses, x1 = log(a), x2 = log(b).

By Theorem 2
x1 + x2 = log(a) + log(b) = log(ab)
⇒ ab = exp(x1 + x2 ) as required.
Define the number e to be exp(1), i.e. log(e) = 1, or exp(1) = e.
Geometric interpretation of e shown below:
48 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

Can show easily (by induction for example) that

exp(n) = en for every positive integer n.

Since 1 = exp(0) = exp(m − m) = exp(m) exp(−m) by Theorem 5.

1
exp(−m) = = e−m m a positive integer.
em

Theorem 6
d
exp(x) is differentiable and dx exp(x) = exp(x).

d
Proof. Think of dx exp(x) as the derivative of the inverse function to log(x). Hence it is
differentiable.
dy 1
We have proved that if g(y) is the inverse function of y = f (x), then dx = dx/dy i.e.
f 0 (x) = g0 (y) or g 0 (y) = f 0 (x) . Let y = exp(x), then the inverse function is x = log(y).
1 1

1 dy dy dy 1
Chain rule 1 = y dx ⇒ dx = y = exp(x). (Equivalently dx = dx/dy ).

Use ex instead of exp(x) - completely analogous. Derivative for definition:

ex+h − ex
 h 
d x x e −1
e = lim = e lim .
dx h→0 h h→0 h
eh −1
By Theorem 6, we have limh→0 h = 1.1

Generally we may have the exponential function

y = ax a > 0.

Can write this as y = exp(x log(a)) = ex log(a) . All the usual properties ax+y = ax ay
etc. all hold.

1
This suggests that limh→0 (1 + h)1/h = e. Prove by (1 + h)1/h = exp 1


h
log(1 + h) =
 
exp log(1+h)−log(1)
h
→ e.
5.3. FUNCTION ESTIMATES FOR SMALL AND LARGE ARGUMENTS 49

Theorem 7
d x
a = ax (log(a))
dx

Proof.

d x d
a = exp(x log(a)) = exp(x log(a)) · log(a) by the chain rule
dx dx

Corollary:
ah − 1
lim = log(a) for a > 0
h→0 h

Finally, the general power function.

Theorem 8
Let a be any number and let f (x) = xa for x > 0. Then f 0 (x) exists and
f 0 (x) = axa−1 .

Proof.

a
f (x) = xa = elog(x ) = ea log(x)
a
⇒ f 0 (x) = ea log(x) · by chain rule
x
= axa−1

5.3 Function Estimates for Small and Large Arguments

Start with showing 2 < e < 4. We know that log(e) = 1 i.e. the area between x = 1
and x = e of y = 1/x, is 1.
50 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

Since log(2) < 1, i.e. area A1 < 1, so we must have e > 2. Now
1
log(4) = 2 log(2) > 2 × = 1.
2
This implies e < 4 ⇒ 2 < e < 4. Accurate calculation later using Taylor’s theorem.

Theorem 9
(1+a)n
Let a be any positive number. Then n → ∞ as n → ∞. [Analogously,
n
limn→∞ (1+a) n = 0.]

n(n−1) 2
Proof. Write (1 + a)n = 1 + na + 2 a + b where b ≥ 0 is some number.
(1 + a)n 1 n−1 2 b
⇒ = +a+ a +
n n 2 n
b (1+a)n
n ≥ 0, and so for large n n becomes arbitrarily large.

en
Corollary: n → ∞ as n → ∞, since e = 1 + a for some a > 0.

Theorem 10
ex
The function f (x) = x is strictly increasing for x > 1 and limx→∞ f (x) = ∞.
exp beats x.

Proof.
xex − ex ex
f 0 (x) = = (x − 1) > 0 for x > 1
x2 x2
limn→∞ f (n) = ∞, hence result follows.
5.3. FUNCTION ESTIMATES FOR SMALL AND LARGE ARGUMENTS 51

Corollary 1.
The function x − log(x) becomes arbitrarily large as x becomes arbitrarily large.
x beats log.

Proof.
ex
 
log = x − log(x) > 0 for x large enough
x
Since log(t) becomes large for t large.

Corollary 2
x
The function log(x) becomes large as x becomes large. x beats log.

x y
Proof. Let y = log(x), then x = ey . So log(x) = ey . log(x) becomes large as x becomes
large, hence y also becomes large. By Theorem 10 the result follows.

Corollary 3
As x becomes large, x1/x approaches the limit 1.

Proof.
1/x log(x)
x1/x = elog(x ) = e x
log(x) 1
Now = → 0 as x becomes large by Corollary 2
x x/ log(x)
⇒ lim x1/x = 1
x→∞

Note Corollary 3 is used many times for integers. i.e. n1/n → 1 for n → ∞ being
integers.

Theorem 11 - exp(x) beats any power of x.

x
Let a be a positive number. Then the function f (x) = xe a is strictly increasing for
x > a and becomes arbitrarily large as x becomes arbitrarily large.

Proof.
ex
f (x) = = ex−a log(x)
ea log(x)
 a
f 0 (x) = ex−a log(x) 1 − > 0 if x > a
x  
x
log(f (x)) = x − a log(x) = (log(x)) −a
log(x)
By Corollary 2, log(f (x)) → ∞ ⇒ f (x) → ∞.
52 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

5.4 Logarithmic Differentiation

Examples (Do yourselves)
√
(i) Differentiate y = xx x

(2x+1)1/2
(ii) Differentiate y = (x2 +1)1/4

5.5 L’Hôpital’s Rule

Theorem 11
If f , g, are differentiable on an open interval containing x0 , g(x0 ) = f (x0 ) = 0,
and g 0 (x0 ) 6= 0, then

f (x) f 0 (x0 )
lim = 0
x→x0 g(x) g (x0 )

Proof. Write
f (x)−f (x0 )
f (x) f (x) − f (x0 ) x−x0
= = g(x)−g(x0 )
g(x) g(x) − g(x0 )
x−x0
f (x)−f (x0 )
f (x) x−x0 f 0 (x0 )
⇒ lim = lim =
x→x0 g(x) x→x0 g(x)−g(x0 ) g 0 (x0 )
x−x0

Example 1

1 − cos(x) 0
lim of form
sin(x)
x→0 0
sin(x)
⇒ = lim =0
x→0 cos(x)

Example 2

0/0
z }| {
sin(x) − x cos(x) − 1
lim 3
= lim 2
= . . . carry on differentiating . . .
x→0 x x→0
| {z3x }
also 0/0

f 0 (x0 ) 0
Problem is that we need to know that if limx→x0 g 0 (x0 ) which is of the form 0 exists,
then it is equal to the limx→x0 fg(x)
(x)
. This is the useful form of L’Hôpital’s rule.
5.5. L’HÔPITAL’S RULE 53

Theorem 12
Let f (x) and g(x) be differentiable on an open interval containing x0 . Assume
that g(x) 6= 0 and g 0 (x) 6= 0 for x in an interval about x0 but with x 6= x0 . Assume
0 (x)
also that f , g are continuous at x0 with f (x0 ) = g(x0 ) = 0, and limx→x0 fg0 (x) = l.
Then also:
f (x)
lim =l
x→x0 g(x)

Example 1:
1 − cos(x) sin(x)
lim = lim if latter limit exists
x→0 x2 x→0 2x

Can use L’Hôpital’s rule again to show (since limit of form 00 .)

sin(x) cos(x) 1
lim = lim =
2x
x→0 x→0 2 2
1 − cos(x) 1
⇒ lim = by Theorem 2
x→0 x2 2
Note: Need to check that every time we apply L’Hôpital’s rule the limit is of form 0/0.

Example 2:
x2 + 1
 
2x
lim 6= lim = 0.
x→0 x x→0 1
Not of form 0/0, i.e. not indeterminate.
Example 3:
sin(x)−x
Show that limx→0 tan(x)−x = − 21 .

Practical note: L’Hôpital’s rule holds for:

(i) One-sided limits
(ii) Limits as x → ∞.
∞
(iii) Indeterminate forms ∞.
Let us prove the rule for the form 0/0 as x → ∞.
0
Proof. By assumption we have limx→∞ fg0 (x)
(x)
= l exists. If y = x1 , then y = 0+ as x → ∞,
i.e.
f 0 (x) f 0 (1/y) −y 2 f 0 (1/y)
lim 0 = lim 0 = lim
x→∞ g (x) y→0+ g (1/y) y→0+ −y 2 g 0 (1/y)
d
dy [f (1/y)] f (1/y)
= lim d
= lim by L’Hôpital
y→0+ [g(1/y)] y→0 g(1/y)
dy
f (x)
= lim
x→+∞ g(x)
54 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

f (x) ∞ 0
What about limx→x0 g(x) when of the form ∞? Would think that casting into 0 form
f (x) 1/g(x) 0
would help, but it doesn’t. i.e. g(x) = 1/f (x) i.e this is now of form 0. But

1/g −g 0 /g 2
lim = lim
x→x0 1/f |{z} x→x+0 −f 0 /f 2
L’Hôp

does not help! The proof is more technical, see later.

Theorem - L’Hôpital’s Rule - general case.

To find limx→x0 fg(x)
(x)
when limx→x0 f (x) and limx→x0 g(x) are both zero or both
infinite, differentiate numerator and denominator and take the limit of the new
function. Repeat as many times as needed as long as L’Hôpital’s rule applies at
each stage. We then have

f (x) f 0 (x)
lim = lim 0
x→x0 g(x) x→x0 g (x)
Note that x0 may be replaced by ±∞ or x0 ±

Example 1

log(x) ∞
lim p
, p > 0, form
x→∞ x ∞
1/x
= lim = 0 since p > 0
x→∞ pxp−1

Example 2

lim x log(x) of form 0 × ∞ so need to rewrite as

x→0+
log(x) ∞
lim which is of form
x→0+ 1/x ∞
1/x
⇒ lim x log(x) = lim = 0.
x→0+ x→0+ −1/x2

Example 3
(a) Find limx→0+ xx
Of form 00 (indeterminate).

(b) limx→1 x−1/(1−x)

Of form 1∞ , again indeterminate.

Solutions
(a) xx = ex log(x) . Have shown limx→0+ x log(x) = 0 and since exp is continuous,
limx→0+ exp(x log(x)) = exp limx→0+ (x log(x) = e0 = 1.
5.5. L’HÔPITAL’S RULE 55

(b) Again use logs, namely

 
−1/(1−x) 1
x = exp log(x)
x−1
log(x) 0
lim is of form ⇒ by L’Hôp
x→1 x−1 0
1/x
= lim =1
x→1 1
  
1/(x−1) log(x)
⇒ lim x = exp lim
x→1 x→1 x−1
= e by continuity of exp(x).

Note: If we set x = 1+ n1 with n an integer, then we have shown limn→∞ (1 + n1 )n = e.

Example 4  
1 1
Find limx→0 x sin(x) − x2
. (The answer is 1/6 and you need to apply L’Hôpital’s
rule 3 times!)

Proof of L’Hôpital’s rule when limx→x0 f (x) = limx→x0 g(x) = 0 i.e. f (x0 ) = g(x0 ) =
0. The case x0 → ∞ is left as an exercise below. We need the following Theorem:

Theorem (Cauchy Mean Value Theorem)

Let f, g be continuous on [a, b] and differentiable on (a, b) with g(a) 6= g(b), and
assume that g 0 does not vanish on (a, b). Then there exists c in (a, b) such that

f (b) − f (a)
g 0 (c) = f 0 (c).
g(b) − g(a)

Proof. Let
f (b) − f (a)
h(x) = f (a) + (g(x) − g(a)) .
g(b) − g(a)
Then h(a) = f (a) and h(b) = f (b). For the function φ(x) = h(x) − f (x) we have
φ(a) = φ(b) = 0 and φ is differentiable on (a, b). Hence by the MVT, there exists a
c ∈ (a, b) such that φ0 (c) = 0. i.e. h0 (c) = f 0 (c), and the theorem is proved.
Proof. L’Hôpital’s Theorem
Here we prove the 0/0 version, i.e. f (x0 ) = g(x0 ) = 0.
f (x) f (x) − f (x0 ) f 0 (c)
= = 0
g(x) g(x) − g(x0 ) g (c)
where c is a number between x and x0 . As x → x0 , c → x0 also. By hypothesis
 0
f 0 (c)

f (x)
lim = ` ⇒ lim =`
x→x0 g 0 (x) x→x0 g 0 (c)

f (x)
⇒ lim = ` as required
x→x0 g(x)
56 CHAPTER 5. EXPONENTIALS AND LOGARITHMS

Exercise: Prove L’Hôpital’s rule when x0 → ∞, i.e. show that if limx→∞ f (x) =
limx→∞ g(x) = 0, then limx→∞ [f (x)/g(x)] = limx→∞ [f 0 (x)/g 0 (x)] = `, say.
Proof. Here are some steps to follow:
(i) First show using Cauchy’s Mean Value Theorem that given ε > 0 there exists M
such that for all y > x > M

f (y) − f (x)
− ` < ε.
g(y) − g(x)

(ii) Note that fg(x)

(x)
= limx→∞ f (x)−f (y)
g(x)−g(y) . Take y sufficiently large in the result in part
(i) to conclude that
f (x)
− ` < ε,
g(x)
and hence complete the proof.
 Part II

Integration

57
Chapter 6

Anti-derivatives and Geometrical

Interpretation

The anti-derivative or integral of a function f (x).

Given f (x) defined over some interval, then if we can find a function F (x) defined
over the same interval such that
F 0 (x) = f (x),
R
then F (x) is the indefinite integral of f ⇒ F = f (x)dx. This is not unique. Let G
be another indefinite integral, i.e. G0 (x) = f (x). Then

d
(F − G) = 0 ⇒ F (x) = G(x) + constant.
dx

6.1 Area under a curve

Suppose f (x) ≥ 0 in some given interval [a, b] and it is also continuous on [a, b], (a < b).
Define by F (x) the area under the curve between x = a and some x.

By definition, F (a) = 0.

59
60 CHAPTER 6. ANTI-DERIVATIVES AND GEOMETRICAL INTERPRETATION

Theorem 1
The function F (x) is Rdifferentiable and its derivative is equal to f (x). Another
d x
way to state this is dx a f (t)dt = f (x).

F (x+h)−F (x)
Proof. Start with the Newton quotient h .
Suppose x 6= b and also h > 0. F (x + h) − F (x) is the area under the graph between
x and x + h.

Figure 6.1:

Since f (x) is continuous on [x, x + h] and is defined there, it must have a maximum
at some point x+ , and minimum at some point x− . Hence, for all t ∈ [x, x + h]

f (x− ) ≤ f (t) ≤ f (x+ ).

Can also bound the area using the rectangles shown in Figure 6.1.

h · f (x− ) ≤ F (x + h) − F (x) ≤ h · f (x+ )

F (x + h) − F (x)
i.e. f (x− ) ≤ ≤ f (x+ )
h

Since x+ and x− are contained in [x, x + h], as h → 0, x− , x+ → x and by the

squeezing theorem we have limh→0 F (x+h)−F
h
(x)
= f (x), i.e. F 0 (x) = f (x). Hence the
anti-derivative is connected to area under the curve.
6.1. AREA UNDER A CURVE 61

The constant is fixed by F (a) = 0. In other words, if I can guess a function G(x)
whose derivative is f (x) (e.g. guess log(x) for the anti-derivative of x1 ), then since F
and G differ by a constant I have

F (x) = G(x) + K.

But F (a) = 0 ⇒ −G(a) = K ⇒ F (x) = G(x) − G(a). Hence

Z b
f (x)dx = F (b) = G(b) − G(a).
a

This is the familiar definite integral.

Example 1
2 2
x3
Z 
2 8 1
x dx = = − .
1 3 1 3 3

x3
Here f (x) = x2 , G(x) = 3 is the guessed anti-derivative.

Definition: signed area

If f (x) < 0 then the area is below the x-axis. Define F (x) to be minus the area.
(All very familiar). This leads to the definite integral.

Example Drawing of an example function f (x)

Figure 6.2:

Z b
f (x)dx = F (b) − F (a) All negative areas are accounted for.
a
62 CHAPTER 6. ANTI-DERIVATIVES AND GEOMETRICAL INTERPRETATION

6.2 Application to a first order differential equation

We can use these results to solve the following general linear first order differential
equation:
dy
+ f (x)y = g(x), (6.1)
dx
where the functions f (x) and g(x) are given and are assumed to be integrable. An initial
condition at x = x0 is also prescribed, namely
y(x0 ) = y0 , (6.2)
where y0 is a given constant. The method is known as that of the integrating factor.
The key to solving (6.2) is the following observation (the integrating factor)
d  Rxx f (t)dt  Rx
f (t)dt
e 0 = f (x) e x0 , (6.3)
dx
a result that follows readily from
R using the
 chain rule and Theorem 1 in this chapter.
x
Next, multiplying (6.2) by exp x0 f (t)dt and using (6.3), casts the equation into
 Z x   Z x 
d
exp f (t)dt y = exp f (t)dt g(x). (6.4)
dx x0 x0

We can readily integrate this once more to find

 Z x  Z x Z s    Z x 
y(x) = exp − f (t)dt exp f (t)dt g(s)ds + exp − f (t)dt y0 .
x0 x0 x0 x0
(6.5)
This is a fairly general solution and hence looks a bit cumbersome. Let us integrate the
following equation to illustrate the method (there is no point remembering the formula
(6.5), the method is what is needed).
Example: Solve the differential equation
 
dy 1 1
+ + 1 y = 2, y(1) = 1, x > 1, (6.6)
dx x x
and find limx→∞ y(x).
Solution: I will solve this by carrying out indefinite integrations and imposing the
initial condition at the end. Try it also by doing what I calculated above with definite
integrals.
The integrating factor is
Z   
1
exp + 1 dx = exp (log x + x) = x exp(x). (6.7)
x
Hence,
d ex
(xex y) = , (6.8)
dx x
6.2. APPLICATION TO A FIRST ORDER DIFFERENTIAL EQUATION 63

which can be integrated to give

ex
Z
x
xe y = dx + K, (6.9)
x
where K is a constant. Now I have a choice of what to do with the integral’s limits. The
most natural thing to do is to pick the lower limit to be x = 1 where we are given that
y(1) = 1, and then will adjust K to satisfy this. Hence
Z x t
x e
xe y = dt + K ⇒ e = K. (6.10)
1 t

The solution becomes, then

Z x
1 −x 1 t 1
y(x) = e e dt + e−x+1 . (6.11)
x 1 t x
Rx
The integral 1 (et /t)dt cannot be done explicitly. The second term e−x+1 /x clearly
tends to zero as x → ∞. It remains to estimate the first term as follows (the integrand
is positive so no need for absolute values)
Z x Z x
1 t 1 1 1 1
e dt < et dt = ex − e ⇒ 0 < y(x) < − e−x+1 + e−x+1 = , (6.12)
1 t 1 x x x x

hence limx→0 y(x) = 0.

64 CHAPTER 6. ANTI-DERIVATIVES AND GEOMETRICAL INTERPRETATION
Chapter 7

The Riemann Sum

Given f (x), a ≤ x ≤ b, take a partition of the interval [a, b] to be

b−a
xi = a + ih i = 0, 1, . . . , n, h=
n

Figure 7.1: Riemann sum partition.

Note: The selected partition has regular spacing. Can generalise this to have a partition
defined by a sequence {xk }k=0, ... ,n and in the limit we require max |xk −xk−1 | → 0. We
k
will avoid this technical issue; it can be dealt with in a direct way and is not necessary
for what we want to do.
Take any sub-interval [xi−1 , xi ] and let x∗i ∈ [xi−1 , xi ]. Then the Riemann sum is
defined to be
Xn
RS ∗ := f (x∗i )h.
i=1

Three particularly useful ways of picking x∗ :

(i) x∗i = xi “right-hand” RS (Riemann Sum)

(ii) x∗i = xi−1 “left-hand” RS

(iii) x∗i = 21 (xi + xi−1 ) midpoint RS

65
66 CHAPTER 7. THE RIEMANN SUM

Figure 7.2: The three canonical choices of x∗i .

Now in the limit n → ∞, h → 0, we can prove

n
X Z b
lim f (x∗i )h = f (x)dx.
n→∞ a
i=1

Sketch of the proof:

Pn
Lower Riemann sum = i=1 f (xi−1 )h := Ln
Pn
Upper Riemann sum = i=1 f (xi )h := Un
By geometry:
Z b
Ln ≤ f (x)dx ≤ Un .
a
In the limit it gets squeezed, if the limit exists then it is the value of the integral. We say
the function is Riemann integrable. Generalisations in important applications such as
probability theory, require different approaches that you will encounter in more advanced
courses with Lesbegue integration.
Example 1:
f (x) = x 0≤x≤1

R1
Figure 7.3: Riemann sum for 0 f (x)dx.
 67

1 i
a = 0, b = 1 ⇒ h= , xi = ih =
n n

n n n
X 1 X i 1 1 X
Upper RS = f (xi ) = · = 2 i
n n n n
i=1 i=1 i=1
1 n(n + 1) 1 1 1
= 2· = + → as n → ∞
n 2 2 2n 2

Aside: The sum of the first n integers:

Pn 1
i=1 i = 2 n(n + 1) := S.
(i) Consider

1 2 3 ... n−1 n
+ n n−1 n−2 ... 2 1

Add 2S = (1 + n) + (1 + n) + · · · + (1 + n) = n(1 + n)
| {z }
n times
1
S = n(n + 1).
2

(ii) Consider (i + 1)2 − i2 = 2i + 1.

n
X n
X n
X
⇒ (i + 1)2 − i2 = 2 i+ 1
i=1 i=1 i=1
2
i.e. 2@
@ − 12 + @2
3@ −@2
2@ +H· · + (n + 1)2 − @
·H 2
n@ = 2S + n
| {z }
“telescoping series”
1
S = n(n + 1)
2
Exercise: Re-do Example 1 but with (a) lower Riemann Sum, (b) midpoint Rie-
mann Sum.

Example 2 Z 1
ex dx
0

n
X 1
Upper Riemann Sum Un = ei/n
n
i=1
Z 1 n
1 X  1/n i
ex dx = lim e =e−1
0 n→∞ n
i=1
| {z }
geometric series
68 CHAPTER 7. THE RIEMANN SUM

The last part is left as problem 2 on Problem Sheet 4. It uses the result for a geometric
series which easily follows from the algebraic identity

(1+x+x2 +. . .+xn−1 )(1−x) = 1−x+x−x2 +x2 +. . .−xn−1 +xn−1 −xn = 1−xn ⇒

1 − xn
1 + x + x2 + . . . + xn−1 = .
1−x

7.0.1 Comparison between upper Riemann sum and midpoint RS

Defining the upper Riemann sum by Un and the midpoint RS by Mn we have

n n  
X X xi−1 + xi
Un = f (xi )h Mn = f h
2
i=1 i=1

Rb R1
We know that limn→∞ Un = limn→∞ Mn = a f (x)dx. Example for 0 ex dx = e − 1 ≈
1.71828183 := I correct to 8 decimal places.

Here are some calculations

n h Un |I − Un | Mn I − Mn
1 1 2.7183 1.0000 1.6487 0.0696
2 0.5 2.1835 0.4652 1.7005 0.0178
4 0.25 1.9420 0.2237 1.7138 0.0045

Conclusions: If h decreases by a factor of two, then the error |I − Un | decreases by

1/2, but |I − Mn | decreased by 1/4.

Midpoint is far superior. Why? (Geometrical explanation.)

Question: Can you think of a better way still? We have to go beyond the Riemann
sum definition - Numerical Analysis Answer: Better approximation of the function, e.g.
instead of linear segments use three points and fit parabolas; other ways are cubic splines,
etc.

A method analogous to the midpoint rule is to approximate the function by piecewise

linear functions. Geometry - (trapezium rule).
 69

Approximate f (x), x ∈ [xi−1 , xi ]

f (xi ) − f (xi−1 )
by li (x) = (x − xi−1 ) + f (xi−1 ).
xi − xi−1

We will be able to calculate the accuracy of such schemes using Taylor’s Theorem - see
later.
70 CHAPTER 7. THE RIEMANN SUM
Chapter 8

Properties of the Definite

Integral; Fundamental Theorem
of Calculus
Rb Rb
1) a cf (x)dx = c a f (x)dx c constant.
Rb Rb Rb
2) a (f (x) + g(x))dx = a f (x)dx + a g(x)dx

3) If c ∈ (a, b) (and here a < b), then

Z b Z c Z b
f (x)dx = f (x)dx + f (x)dx
a a c

4) If f (x) ≤ g(x) for x ∈ [a, b] then

Z b Z b
f (x)dx ≤ g(x)dx
a a

Rb Rb Rb Rb
Hence a f (x)dx ≤ a |f (x)|dx and a f (x)dx ≤ a |f (x)|dx
Rb Ra
5) a f (x)dx = − b f (x)dx.

Proofs follow easily from RS definitions and the use of signed areas.

Theorem 1
Suppose g(x) is defined for all x ∈ [a, b] and is differentiable on [a, b]. Then
Z b
g 0 (x)dx = g(b) − g(a)
a

Proof. (Sketch)

71
72CHAPTER 8. PROPERTIES OF THE DEFINITE INTEGRAL; FUNDAMENTAL THEOREM OF

b−a
Let xi = a + ih, h = n .

n n
X X g(xi + h) − g(xi )
Upper RS g 0 (xi )h ≈ ·S
h = g(b + h) − g(a + h)
h{z
S
i=1 i=1 | }
telescoping series

As h → 0 result follows.

Theorem: Fundamental Theorem of Calculus

Suppose F is differentiable on [a, b] and F 0 is integrable on [a, b]. Then
Z b
F 0 (x)dx = F (b) − F (a).
a

If f is integrable on [a, b] and has anti-derivative F , then

Z b
f (x)dx = F (b) − F (a).
a

Useful Theorem:
Z g(x)
d
f (t)dt = f (g(x)) · g 0 (x).
dx a

Rx
Proof. Let F (x) = a f (t)dt. Then F 0 (x) = f (x) - already proved.
R g(x)
Now a f (t)dt = F (g(x)) by the definition of F .
Z g(x) !
d d
⇒ f (t)dt = F (g(x) = F 0 (g(x)) · g 0 (x)
dx a dx
= f (g(x)) · g 0 (x).

Example
!
Z x2
d t 2
e dt = ex · 2x
dx a
Z x2
x2 2
or et dt = et a
= ex − ea same as before
a
Chapter 9

Improper Integrals

Definition
Rb
a f (x)dx is an improper integral if

(i) a = −∞ and/or b = ∞ (ii) f (x) → ±∞ in (a, b)

To find improper integrals we take the limit of proper integrals. If the limit is finite,
the integral converges, otherwise it diverges.
Example 1
Z ∞ Z b  
1 1 1
dx = lim dx = lim − + 1 = 1
1 x2 b→∞ 1 x2 b→∞ b
Example 2 Z ∞
dx
= lim log(b) = ∞ i.e. diverges.
1 x b→∞

Geometrically:

In general
b
br+1 − 1
Z  
lim xr dx = lim r 6= −1
b→∞ 1 b→∞ r+1
So need r +1 < 0 for convergence, i.e. r < −1, (r = −1 is divergent - see log example
earlier).

73
74 CHAPTER 9. IMPROPER INTEGRALS

9.0.1 Comparison Theorem/Test

Suppose f and g satisfy:
(i) |f (x)| ≤ g(x) for all x ≥ a
Rb Rb
(ii) a f (x)dx and a g(x)dx exist for every b > a.
Then
R∞ R∞
(a) If a g(x)dx is convergent, so is a f (x)dx
R∞ R∞
(b) If a f (x)dx is divergent, so is a g(x)dx
Rb R∞
Similarly for −∞ f (x)dx and −∞ f (x)dx.

Intuitive “proof”: If f , g both positive then the picture is

Comparison test is useful if we cannot carry out the integral exactly. It will tell us
if it exists, then we can find it numerically etc. e.g
Z ∞
sin(x)
converges
0 (1 + x)2
 75

R∞ dx
R∞ dx
First thing to note is that 0 (1+x)2
converges by comparison to 1 x2
. Why? If
1 1
x ≥ 1, (1+x)2
< x2
. By the comparison test

∞
| sin(x)|
Z
sin(x) 1 1
dx converges since ≤ < 2 for x ≥ 1 (9.1)
0 (1 + x)2 (1 + x)2 (1 + x)2 x
e.g.
Z ∞
dx
√ is divergent
1 1 + x2
Z b Z b Z b
dx dx dx
√ ≥ √ = √
1+x 2 2
x +x 2 2x
1
Z ∞1 1
Z ∞
dx dx
Now diverges ⇒ so does √
1 x 1 1 + x2
R∞ dx
e.g. 1
√
x
diverges. (Already saw this, and we can do it directly).

Here is a proof using the comparison theorem.

Z b Z b
1 1 1 dx
If x ≥ 1, ≤√ ⇒ dx < √ , b>1
x x 1 x 1 x
Z ∞
1
and dx diverges
1 x

9.0.2 Improper integrals of unbounded functions

Without loss of generality, consider the situation where |f (x)| → ∞ as x → 0. Again,
take limits of bounded integrals. e.g.
Z 1 (
1 converges if p < 1
p
dx
0 x diverges if p ≥ 1

Proof. Left as an exercise.

Example 1
Z 1
log(x)dx exists
0
Z 1  Z 1 
1 1
= lim log(x)dx = lim [x log(x)] − x dx
→0  →0  x
= lim [− log() − 1 + ] = −1
→0

Example 2
R∞ −x
e√
Show that the improper integral I = 0 x
dx converges.
76 CHAPTER 9. IMPROPER INTEGRALS

∞ −x
1
e−x
Z Z
e
Write I = I1 + I2 where I1 = I2 = √ dx,√ dx
0 1x x
Z 1 −x Z 1
e 1
I1 = √ < √ dx which is convergent
0 x 0 x
Z ∞ −x Z ∞
e
√ dx < e−x dx which is also convergent
1 x 1

Example 3 √
Find the length of the curve y = 1 − x2 for x ∈ [−1, 1].
Z 1 Z 1
0 2 1/2 dx
Length L = (1 + (y ) ) dx = √
improper at both ends
−1 −1 1 − x2
Z 0 Z 0
dx dx  π π
= lim sin−1 (0) − sin−1 (p) = 0 − −
 
√ = lim √ =
−1 1 − x2 p→−1 p 1 − x2 p→−1 2 2
Z 1 Z 1
dx  π
lim sin−1 (1) − sin −1(p) =

√ = lim ⇒ π is the length
0 1 − x2 p→1 0 p→0 2
Chapter 10

Mean Value Theorem for

Integrals

Given a function f that is integrable on [a, b], we define its average hf i[a,b] by the formula.

Z b
1
hf (x)i[a,b] = f (x)dx.
b−a a

Since hf i[a,b] is a number (constant), then we have

Z b Z b
f (x)dx = hf i[a,b] dx
a a

Geometrically, the area of the shaded rectangle is equal to the area under y = f (x).

77
78 CHAPTER 10. MEAN VALUE THEOREM FOR INTEGRALS

Theorem 1 Let f be continuous on [a, b]. Then there exists a point x0 ∈ (a, b)
such that Z b
1
f (x0 ) = f (x)dx.
b−a a

Rx
Proof. Define F (x) = a f (t)dt. By the Fundamental Theorem of Calculus we have
F 0 (x) = f (x) for all x ∈ (a, b). F is continuous at a and b. By MVT we have

F (b) − F (a)
F 0 (x0 ) =
b−a
i.e. Rb Ra Z b
a f (t)dt − a f (t)dt 1
f (x0 ) = = f (t)dt
b−a b−a a
Chapter 11

Techniques of Integration

Will assume familiarity with substitution and integration by parts.

11.0.1 Trigonometric Integrals

Z
sinm (x) cosn (x)dx m, n integers

For:
Z Z
n=1 substitute u = sin(x) ⇒ sin (x) cos(x)dx = um du
m

Z
cos(x)
n = 1, m = −1 dx = log | sin(x)| + c
sin(x)
m, n 6= 1 Write in terms of sinp (x) cos(x) or similar, or use double angle formulae

Important - use trig formulas! - a reminder of double angle formulas:

cos(2x) = 2 cos2 (x) − 1 = 1 − 2 sin2 (x)

sin(2x) = 2 sin(x) cos(x)
sin(x ± y) = sin(x) cos(y) ± cos(x) sin(y)
cos(x ± y) = cos(x) cos(y) ∓ sin(x) sin(y)

From these we have

1
sin(x) cos(y) = [sin(x + y) + sin(x − y)]
2
1
sin(x) sin(y) = [cos(x − y) − cos(x + y)]
2
1
cos(x) cos(y) = [cos(x + y) + cos(x − y)]
2
Example 1
Z Z Z
sin (x) cos (x)dx = sin2 (x)(1 − sin2 (x)) cos(x)d
2 3
⇒
|{z} u2 (1 − u2 )du
u=sin(x)

79
80 CHAPTER 11. TECHNIQUES OF INTEGRATION

Example 2
sin2 (2x) 1 1 − cos(4x)
Z Z Z
2 2
sin (x) cos (x) = = dx
4 4 2
Example 3 Z
I= tan3 (θ) sec3 (θ)dθ

sin3 (θ) 1 − cos2 (θ) 1 − u2

Z Z Z
i) Write dθ = sin(θ)dθ ⇒ =− du
cos6 (θ) cos6 (θ) |{z} u6
u=cos(θ)
d
ii) Notice sec(θ) = tan(θ) sec(θ)
Z dθ Z Z
tan (θ) sec (θ)dθ = tan(θ) sec(θ)(sec2 (θ) − 1) sec2 (θ)dθ
3 3
⇒
|{z} = (u2 − 1)u2 du
u=sec(θ)

Example 4 Z Z
1
cos(3x) cos(5x)dx = (cos(8x) + cos(2x))dx etc
2
Trigonometric substitutions
√ √
(1) If a2 − x2 appears in an integral, try x = a sin(θ), dx = a cos(θ)dθ, a2 − x2 =
a cos(θ) (a > 0, θ acute).

√ √
(2) If x2 − a2 occurs, try x = a sec(θ), dx = a tan(θ) sec(θ)dθ and x2 − a2 =
a tan(θ).

√ √
(3) If a2 + x2 or a2 + x2 occur, try x = a tan(θ), 2 2 2
√ dx = a sec (θ)dθ, a + x =
2 2
a sec(θ). Also x = a sinh(θ), dx = a cosh(θ)dθ, a + x = a cosh(θ).
 81

Example 5
x2
Z
dx x = tan(θ) dx = sec2 (θ)dθ
(1 + x2 )3/2
tan2 (θ) sec2 (θ)
Z Z
⇒ dθ = tan(θ) sin(θ)dθ
sec3 (θ)
Better by parts:
Z Z
x 2 −1/2 1
2 3/2
· xdx = −(1 + x ) ·x+ √ dx
(1 + x ) 1 + x2
x
= −√ + sinh−1 (x) + c
1 + x2

11.0.2 Recursion formulas

Z
Let In = sinn (x)dx.
1 n−1
Then In = − sinn−1 (x) cos(x) + In−2
n n
Proof.
Z
In = sinn−1 x sin(x)dxintegrate by parts
Z
= − cos(x) sinn−1 (x) + (n − 1) sinn−2 (x) cos2 (x)dx
Z
In = − cos(x) sinn−1 (x) + (n − 1) (− sinn (x) + sinn−2 (x))dx

⇒ In = − cos(x)(sin(x))n−1 − (n − 1)In + (n + 1)In−2

1 n−1
In = − sinn−1 (x) cos(x) + In−2
n n

R π/2 8
Example 6 Show that 0 sin5 (x) = 15

Solution: Use recursion above, and keep track of the limits.

π/2
1 X 4
I5 = − Xsin4 (x)
XX cos(x) + I5−2
5 | {z } 5
0 0
 
π/2
4  1 XX2 2 
= − sin (x)
X cos(x)
X + I1 
5 3 | {z } 3
0 0
Z π/2
4 2 8
= · sin(x)dx =
5 3 0 15
82 CHAPTER 11. TECHNIQUES OF INTEGRATION
Chapter 12

Applications of Integration:
Lengths, surfaces and volumes of
revolution

12.1 Length of curves

Start with something we have already seen. Given y = f (x), find the length of the graph
of the function. Do something similar to Riemann sums but for the length. Partition
(x0 , x1 , . . . , xn ), x0 = a, xn = b.

For the ∆ABC

p (AB)2 = (xi − xi−1 )2 + (f (xi ) − f (xi−1 ))2 . But length of curve
segment AB ≈ (xi − xi−1 )2 + (f (xi ) − f (xi−1 ))2 .

n p
X
Total length ≈ (xi − xi−1 )2 + (f (xi ) − f (xi−1 ))2
i=1
s
n  2
X f (xi ) − f (xi−1 )
= (xi − xi−1 ) 1 + .
xi − xi−1
i=1

83
84CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

b−a
Now let xi − xi−1 = h = n := ∆x.

s  2
X f (xi ) − f (xi−1 )
Total length = lim ∆x 1+
n→∞, (h→0, ∆x→0) xi − xi−1
Z b 1/2
L= 1 + (f 0 (x))2 dx
a

In parametric form this is

Z t1
" 2  2 #1/2
dx dy
L= + dt
t0 dt dt

Example Find the length of the parabola y = x2 , 0 ≤ x ≤ 1.

√
Clearly L > 2 = 1.4142 . . . , y = f (x) = x2 , f 0 (x) = 2x.

Z 1
L= (1 + 4x2 )1/2 dx
0
(
x = 0, θ = 0
Substitution 2x = tan(θ)
x = 1, θ = a tan(2)
2dx = sec2 (θ)dθ see Fig 12.1 below
Z a tan(2)
1
L= sec3 (θ)dθ
0 2
12.1. LENGTH OF CURVES 85

Figure 12.1:

Method 1
Z Z Z
3 cos(θ) cos(θ)
sec (θ)dθ = dθ = dθ
4
cos (θ) (1 − sin2 (θ))2
Z
du
u = sin(θ) = partial fractions
(1 − u2 )2
A B C D
= + 2
+ +
1 + u (1 + u) (1 − u) (1 − u2 )
= etc . . . etc
or write in terms of x again.
Method 2
Z Z
sec3 (θ)dθ = sec(θ) sec2 (θ)dθ integrate by parts
d
(sec(θ))
Z z dθ }| {
= tan(θ) sec(θ) − tan(θ) sec(θ) tan(θ) dθ
Z Z
3 2
= tan(θ) sec(θ) − sec (θ)(1 − cos (θ))dθ move − sec3 (θ)dθ to other side
Z Z
3
2 sec (θ)dθ = tan(θ) sec(θ) + sec(θ)dθ)
Z
1
i.e. sec3 (θ)dθ = [tan(θ) sec(θ) + log(sec(θ) + tan(θ))] (see HW3)
2
Z 1
1 a tan(2)
(1 + 4x2 )1/2 dx = [tan(θ) sec(θ) + log(sec(θ) + tan(θ))]0
0 4
1h p  p i1 1 h √  √ i
= 2 × 1 + 4x2 + log 2x + 1 + 4x2 = 2 5 + log 2 + 5 = 1.4789
4 0 4
Method 3 - the easiest one! Rθ
Let 2x = sinh θ so that 2dx = cosh θ dθ and L = 0 1 (1/2) cosh2 θdθ, and sinh θ1 = 2.
Now use cosh(2θ) = cosh2 θ + sinh2 θ = 2 cosh2 θ − 1 to find
Z θ1
1 + cosh 2θ θ1 1
L= dθ = + sinh 2θ1
0 4 4 8
86CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

12.2 Volumes and Volumes of Revolution

A plane cuts a solid V - cross sectional area is Rx say. Then the volume of a slice is
Rx dx. So if Px is a family of parallel planes with common axis x, and the area of V cut
by Px is A(x), then the volume of V is

Z b
A(x)dx
a

where the solid V lies between planes Pa and Pb .

Example 1 Volume of a sphere of radius r. Pick planes along the x-axis.

12.2. VOLUMES AND VOLUMES OF REVOLUTION 87


OA
 =r
∆OAB OB =x

AB = (r2 − x2 )1/2

Z r
4
So A(x) =π(r2 − x2 ), V = π(r2 − x2 )dx = πr3
−r 3

Example 2 Volume of conical solids, with circular base of radius r and height h.

Take x to be upwards as shown, so Px cuts the cone in circular areas.

Geometry - ∆ABB 0 and ∆OAB.

Similar ∆s - we are after DE, the radius of the circle cut by Px .

88CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

DE AE AF DE h−x
= = → i.e. → =
OB AB AC r h
h−x
⇒ DE = r
h
r2
A(x) = π 2 (h − x)2
h
Z h 2
r
Volume = π 2 (h − x)2 dx
0 h
h
r2 (h − x)3 1
= −π = πr2 h
h2 3 0 3

Example 3 A sphere of radius r is cut into 3 pieces with the two cuts symmetrically
placed about the centre. Where should the cuts be in order to get three equal volumes?
Complete for homework.

12.2.1 Volumes of revolution

Given a area bounded by x = a, x = b, y = f (x), y = 0.

The volume of the solid produced by revolving y = f about the x-axis (as shown) is
given by
Z b
V = π(f (x))2 dx.
a

This follows immediately by the slice method seen earlier.

Example 4 The region between the graphs of sin(x) and x for [0, π2 ], is revolved about
the x-axis. Sketch the resulting solid and find its volume.
12.2. VOLUMES AND VOLUMES OF REVOLUTION 89

π/2
π4 π2
Z
V = π(x2 − sin2 (x))dx = − show this
0 24 4
If we revolve about the y-axis, what is the volume? Consider a non-negative function
f (x) on [a, b].

Revolving the element about the y-axis gives a shell of volume

(b)
z}|{
|{z} (x) |{z}
2πx f dx
(a) (c)
Z b
⇒ V = 2πxf (x)dx.
a

Where (a) - circumference of cylindrical shell, (b) - radius of shell, (c) - thickness of
shell. Note that this can be done by the slice method but with planes along the y-axis
and parallel to the x-axis.
90CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

12.3 Surface Areas of Revolution

As we revolve about the x-axis, the area of the surface area swept out is a strip of length
≈ 2πf (xi ) and thickness ∆li . Now
1/2
∆li = (xi − xi−1 )2 + (f (xi ) − f (xi−1 ))2

1/2
≈ 1 + (f 0 (xi ))2

∆x as seen earlier.

Therefore, in the limit, area S is

Z b p
S= 2πf (x) 1 + (f 0 (x))2 dx
a

Example 5 Sphere radius r

12.3. SURFACE AREAS OF REVOLUTION 91

y = (r2 − x2 )1/2 i.e. f (x) = (r2 − x2 )1/2

Z r 1/2
x2

0 x 2 2 1/2
f (x) = − 2 ⇒ S= 2π(r − x ) · 1 + 2 dx
(r − x2 )1/2 −r (r − x2 )
Z r
= 2πr dx = 4πr2
−r

Example 6 Torus of cross-sectional radius r and radius a > r.

Revolve about the x-axis to get a torus.

p
Circle equation is x2 + (y − a)2 = r2 , i.e. y = a ± r 2 − x2
p
Upper semi-circle: f+ (x) = a + r2 − x2
p
Lower semi-circle: f− (x) = a − r2 − x2
Z r
1/2
⇒ S = S+ + S− = 2πf+ (x) 1 + (f+0 (x))2 dx
−r
Z r

1/2
+ 2πf− (x) 1 + (f−0 (x))2 dx
−r

x
f+0 = − √ = −f−0
−r2 − x2
r2
⇒ 1 + f+02 = 1 + f−02 =
(r2 − x2 )

Put together - don’t integrate separately!

92CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

Z r h p p i r
S = 2π a + r 2 − x2 + a − r 2 − x2 · dx
r (r2 − x2 )1/2
Z r
dx
= 4πar √
−r r2− x2

Rr
Put x = r sin θ and show that √ dx = π.
−r r2 −x2

⇒ Storus = 4π 2 ar = (2πa) × (2πr)

12.4 Centres of Mass

1D case - straightforward.

If centre of mass is at x = x̄, then we must have a zero total moment. i.e.

Pn
k=1 mk xk
X
mk (x̄ − xk ) = 0 i.e. x̄ = P n
k=1 mk

2D case - discrete masses.

12.4. CENTRES OF MASS 93

n masses of mass mk and coordinates (xk , yk ). Find the center of mass, assume it
is (x̄, ȳ). There are two degrees of freedom, so without loss of generality we need to
have zero moments about the x-axis and the y-axis. What do I mean by this? Here is a
schematic.

For balance I need:

P )
(1) mi (x̄ − xi ) = 0
P P 
mi xi mi yi
⇒ (x̄, ȳ) = P , P
mi mi
P
(2) mi (ȳ − yi ) = 0

Note: If the masses are placed symmetrically and are equal, centre of mass is on
the line of symmetry. e.g.
94CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

Exercise: if m1 = m2 = m3 = m, find the centre of mass. What happens if

m1 = m2 6= m3 ?

Now consider a continuous mass distribution, i.e. a plate of a certain spatial density.

General theory - divide it into small rectangles

12.4. CENTRES OF MASS 95

Moment about y-axis isxi ρ(xi , yi )∆x∆y

XX
Add all of them up (like a Riemann sum) xi ρ(xi , yi )∆x∆y
i j
ZZ
Moment about the whole plate about the y-axis x̄ ρ(x, y)dxdy
ZZ Area ZZ
In the limit xρ(x, y)dxdy = x̄ ρdxdy
Z ZA ZZA
Similarly yρ(x, y)dxdy = ȳ ρ(x, y)dxdy
A A

You will see how to work with double integrals in term 2 and further. For the
moment, we will consider the centre of mass of regions bounded by one or more graphs
y = f (x).
Case 1 Region {(x, y) : a ≤ x ≤ b, 0 ≤ y ≤ f (x)}

Consider a partition of [a, b] as shown. For rectangle Ri , the centre of mass is (by
symmetry):
1 1
x∗i = (xi−1 + xi ) yi∗ = f (x∗i )
2 2
96CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

Moment of Ri about y-axis My (Ri ) = ρf (x∗i )∆x · x∗i

| {z } |{z}
mass distance
1
Moment of Ri about x-axis Mx (Ri ) = ρf (x∗i )∆x
· f (x∗i )
2
Xn
Physics M (R1 ∪ R2 ∪ R3 · · · ∪ Rn ) = M (Ri )
i=1
Moment of the union of rectangles = sum of moments of the individual rectangles (Archimedes)

Z b
⇒ My = lim ρx∗i f (x∗i )∆x
ρxf (x)dx
=
n→∞ a
1 b
Z
1
Mx = lim ρ f (x∗i )2 ∆x = ρ(f (x))2 dx.
n→∞ 2 2 a
Rb
Now for a balance of moments, if the total mass of R is m (note m = a ρf (x)dx).
Then
Rb 1 b
R 2
a xf (x)dx 2 a (f (x)) dx
x̄ = R b ȳ = R b .
a f (x)dx a f (x)dx

Note: density ρ cancels out, so take ρ = 1 w.l.o.g.

Example 7 Half disk

By symmetry x̄ = 0.

1
R1
2 −1 (1 − x2 )dx 1 2 4
ȳ = = (2 − ) = ≈ 0.424
π/2 π 3 3π

1
(Note: If we found it to be > 2 then we know it’s wrong! Why?)
12.4. CENTRES OF MASS 97

Case 2: R = {(x, y) : a ≤ x ≤ b, g(x) ≤ y ≤ f (x)}

The centre of mass of “rectangle” Ri is (x∗i , 21 (f (x∗i ) + g(x∗i ))) as before. The area of
the rectangle is approximately equal to (f (x∗i ) − g(x∗i ))∆x. So as before we find
Rb 1 b
R 2 2
a x(f (x) − g(x))dx 2 a (f (x) − g(x) )dx
x̄ = R b ȳ R b
a (f (x) − g(x))dx a (f (x) − g(x))dx
Example 8 Region between y = x and y = x2 , 0 ≤ x ≤ 1. Find centre of mass

R1
x(x − x2 )dx 1/12 1
x̄ = R0 1 = =
2 1/6 2
0 (x − x )dx
1 1 2
(x − x4 )dx
R
1/15 2
ȳ = 2 0 = =
1/6 1/6 5
98CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O
)
1 f ( 12 ) =1 1− 2 3
5 = 5
At x= , ⇒ closer to top curve
2 g( 12 ) = 1
4
2 1 3
5 − 4 = 20

Is this expected or not?

Note: If f (x) = xm , g(x) = xn , for some m, n, the centre of mass could be outside the
region. (This is ok.)

Theorem of Pappus
Let R be a region that lies on one side of a line l.

A = area of R
V = Volume obtained by rotating about l
d = distance travelled by the centre of mass when R is rotated about l

Then V = Ad

Example 9 Volume of a cylinder radius r. Take the function y = r, 0 ≤ x ≤ l.

12.5. MOMENT OF INERTIA 99

Rotate about x-axis.

1
Area = rl ȳ = r (by symmetry)
2
1
⇒ d = r · 2π = πr
2
V = rl · πr =πr2 l as known

12.5 Moment of Inertia

These are important in problems involving the rotation of objects.
Consider a particle of mass m which is at a distance y from the x−axis. Suppose the
particle rotates about the x−axis with a constant angular velocity ω (i.e. in unit time
it sweeps out an angle ω).
The velocity of the particle is yω and hence its kinetic energy is

1
KE = m(yω)2 .
2

The coefficient of 12 ω 2 is defined to be the moment of inertia. Hence for the single
particle considered here we define the moment of inertia I to be

I = my 2 .

If we have a collection of masses m1 , . . . , mn at positions y1 , . . . , yn , then the moment of

inertia is
Xn
I= mi yi2 . (12.1)
i=1

The usefulness of (12.1) is that under rigid body rotation (i.e. all the masses rotate at
the same angular velocity), the kinetic energy is

1
KE = ω 2 I.
2
The moment of inertia only has information about the masses, not their velocity.
Consider next the curve y = f (x) in the x − y plane in the interval x0 ≤ x ≤ x1 , e.g.
a wire of the given shape. Let the density of the curve (wire) be ρ(x) per unit length.
Then we can write down the moment of inertia Ix about the x−axis, and Iy about the
y−axis by using (12.1) and Riemann sums based on a partition of the curve into element
pieces. We have
Z x1 p Z x1 p
2
Ix = 02
ρ(x)y 1 + y dx = ρ(x)f (x)2 1 + f 02 dx (12.2)
x0 x0
Z x1 p Z x1 p
2
Iy = 02
ρ(x)x 1 + y dx = ρ(x)x2 1 + f 02 dx. (12.3)
x0 x0

For a uniform density wire we have ρ = ρ0 , a constant.

100CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

12.6 Length of curves and areas using polar coordinates

Recall

Z b " 2  2 #1/2
dx dy
L= + dt for parametric curved (x(t), y(t)).
a dt dt

Now in polar coordinates we have curves r = f (θ) so we use θ as a parameter.

x = r cos(θ) = f (θ) cos(θ) y = r sin(θ) = f (θ) sin(θ)

Z β  0 1/2
L= (f cos(θ) − f sin(θ))2 + (f 0 sin(θ) + f cos(θ))2 dθ
θ=α
Zβp
= (f 0 (θ))2 + (f (θ))2 dθ
α
" 2 #1/2
Z β
dr
= + r2 dθ
α dθ

Another way using infinitesimals.

12.6. LENGTH OF CURVES AND AREAS USING POLAR COORDINATES 101

(dr)2 + r2 (dθ)2 = (ds)2

Pythagoras:
"  #1/2
dr 2
ds = + r2 dθ
dθ
Z β " 2 #1/2
dr
⇒ L= + r2 dθ
α dθ

Example 10 Find the length of the cardioid r = 1 + cos(θ), 0 ≤ θ ≤ 2π.

Z q 2π Z 2π p
2 2
L= (1 + cos(θ)) + sin )θ)dθ = 2 + 2 cos(θ)dθ
0 0
   
2 θ 2 θ
Now cos(θ) = 2 cos − 1 ⇒ (1 + cos(θ)) = 2 cos
2 2
 
p θ
⇒ 2(1 + cos(θ) = 2 cos
2
102CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

We need to do this because 1 + cos(θ) can be positive but cos 2θ is negative, e.g. for


π ≤ θ ≤ 2π. So need to write this integral as
Z π π  Z 2π   
θ
L= 2 cos dθ + −2 cos dθ = 8
0 2 π 2
Z 2π  
θ
Otherwise 2 cos dθ = 0 which is absurd!
0 2
Area in polar coordinates in a region inside the graph of f (θ) on [α, β].

Use segments of angles ∆θi , and f constant.

Approximate by r = f (θi ) − constant. We need the area of a segment of a circle of

radius f (θi ) and subtending an angle of ∆θ radians. If the area is ∆A then we have
∆θ ∆A
= .
2π π(f (θi ))2
Hence
1
∆A = (f (θi ))2 ∆θi
2
1 β 1 β 2
Z Z
⇒ A= f (θ)2 dθ = r dθ
2 α 2 α
12.6. LENGTH OF CURVES AND AREAS USING POLAR COORDINATES 103

In Multi-variable Calculus you will see a more general construction.

Note:

Figure 12.2:

Example 11 Find the area enclosed by the four-petaled rose r = cos(2θ).

104CHAPTER 12. APPLICATIONS OF INTEGRATION: LENGTHS, SURFACES AND VOLUMES O

π π π π
r ≥ 0 ⇒ − ≤ 2θ ≤ − ≤θ≤
2 2 4 4
Z π Z π
1 4 1 4 1 + cos(4θ)
A= cos2 (2θ)dθ = dθ
2 − π 2 −π 2
4 4
1π π
= =
42 8
 Part III

Series, Power Series and Taylor’s

Theorem

105
Chapter 13

Series

Definition
Given a sequence {an }n≥1 of real numbers, define the sequence of partial sums
by
N
X
SN = a1 + a2 + · · · + aN = an .
n=1

If SN → S as N → ∞, we say the series converges to the sum S. Write

N
X ∞
X
S = lim an = an .
N →∞
n=1 n=1

Example 1
The geometric series ∞ n
P
n=0 x . (x 6= 1)

SN = 1+(x + · · · + xN )
xSN = (x + · · · + xN ) + xN +1

1 − xN +1
Subtract SN =
1−x
∞
1 X
If |x| < 1, lim SN = = xn ,
N →∞ 1−x
n=1
hence the series converges. If x ≥ 1, the series diverges.
Example 2
∞ N N  
X 1 X 1 X 1 1
SN = = −
n(n + 1) n(n + 1) n n+1
n=1 n=1 n=1

1
Telescoping series ⇒ SN = 1 − N +1 → 1 as N → ∞. Series converges to 1.

107
108 CHAPTER 13. SERIES

13.1 Partial sums and geometric series

13.1.1 Series of positive terms
Note negative terms have the same theory.
Since terms are positive, the sequence SN is an increasing sequence of numbers.
Hence if the sequence of partial sums is P
bounded above then the series converges. If the
sequence SN is unbounded above, then → ∞.

Theorem P 1
The series ∞n=1
1
n diverges to +∞.

Proof. It is enough to prove that the partial sums are not bounded above.
Consider
1 1 1
S2K = 1 + + + · · · + K
2  3 2    
1 1 1 1 1 1 1 1 1
=1+ + + + + + + + ··· + + ··· + K
2 3 4 5 6 7 8 2K−1 + 1 2
     
1 1 1 1 1 1 1 1 1
≥1+ + + + + + + + ··· + + ··· + K
2 4 4 8 8 8 8 2K 2
1 2 4 2K−1 1
= 1 + + + + ··· + K = 1 + K
2 4 8 2 2
Partial sums unbounded for K large ⇒ series diverges.

Theorem 2
If α > 1 is a rational number, then
∞
X 1
converges.
nα
n=1

Proof. Partial sums are increasing, so enough to prove that they are bounded above.
Compare SN ≤ S2N −1 , note N ≤ 2N − 1.

1 1 1
SN ≤ S2N −1 = 1 + α
+ α + ··· + N
2 3 (2 − 1)α
     
1 1 1 1 1 1 1 1
=1+ + + + + + + ··· + + ··· + N
2α 3α 4α 5α 6α 7α 2(N −1)α (2 − 1)α
2 4 2N −1
≤ 1 + α + α + · · · + (N −1)α
2 4 2
 2  N −1
1 1 1
= 1 + α−1 + + ··· +
2 2α−1 2α−1
1

N
1 − 2α−1 1 1
= 1

≤ 1 if α−1
<1
1 − 2α−1 1 − 2α−1 2
13.2. CAUCHY SEQUENCES AND CONVERGENCE OF SERIES 109

i.e. α > 1, which is the assumption of the theorem.

Note: We will see a much easier proof later.

13.1.2 Elementary algebraic rules for series

P∞ P∞ P∞
If n=1 an and n=1 bn converge, then n=1 (αan + βbn ) also converges for any con-
stants α, β.

Theorem 3P - Necessary condition for convergence.

If the series ∞ n=1 an converges, then an → 0 as n → ∞.

P∞ PN
Proof. Let the sum be S, i.e.
PN −1 1 an = S. Then SN = n=1 an → S as N → ∞, and
SN −1 = n=1 an → S as N → ∞. Now aN = SN − SN −1 → S − S = 0 as N → ∞.

Example 3
∞
X
(−1)n = −1 + 1 − 1 + 1 . . . diverges by the theorem above, an 6→ 0 as n → ∞.
n=1

P∞ 1
Note: Theorem 3 provides a necessary but not sufficient condition, e.g. n=1 n has
an → 0 but diverges.

Proposition: (follows from what we have shown). P

If n=1 an converges, then for every N the series ∞
P∞
n=N an → 0 as N → ∞.
Intuitively, the “tail” of the series must go to zero if the series converges.

13.2 Cauchy sequences and convergence of series

Definition - Cauchy sequence

We say that the sequence of numbers Sk , k = 1, 2, . . . is a Cauchy sequence, if
given any  > 0 we can find an N such that for any m > N and n > N

|Sm − Sn | < 

Intuition: as k increases, Sm and Sn get arbitrarily close. Cauchy sequences do not

require all positive terms or any other special assumptions.
Connection with series: we have the following results for Cauchy sequences.

(1) Any convergent sequence is a Cauchy sequence.

(2) Any Cauchy sequence is bounded.

110 CHAPTER 13. SERIES

Theorem 4
Every Cauchy sequence converges. (Proof by use of the Bolzano-Weierstrass the-
orem seen in Analysis).

Theorem 5 (The alternating series test)

Suppose {an }n≥1 is a decreasing sequence of positive numbers with an → 0 as
n → ∞. Then the series ∞ n−1 a = a − a + a − a + . . . converges.
P
n=1 (−1) n 1 2 3 4

Proof. We will show that the sequence Sk of partial sums is a Cauchy sequence, i.e.
given any  > 0 we need to find N such that for all n > m > N , |Sn − Sm | < .
Consider any n > m. Then since an is decreasing

0 ≤ am+1 − am+2 + am+3 − · · · + an ≤ am+1

Since an → 0 as n → ∞, given , I can find N such that for any n > N , an < .
Now for any n > m > N .

|Sn − Sm | = |(a1 − a2 + a3 − a4 + . . . an ) − (a1 − a2 + . . . am )|

= |am+1 − am+2 + am+3 − . . . an |
≤ am+1 <  since m > N
⇒ Sk is a Cauchy sequence and the Theorem follows.

Example 4

∞
X (−1)n−1 1
converges since |an | = → 0 and it is an alternating series.
n n
n=1
∞
X (−1)n−1 1 1 1
In fact, = 1 − + − + · · · = log(2) (we will see this later)
n 2 3 4
n=1

13.3 Convergence tests

Theorem
P∞ 6 (Comparison test)
P∞ n=1 bn be convergent with bn non-negative. If |an | ≤ bn (n = 1, 2, . . . ), then
Let
n=1 an converges.
13.3. CONVERGENCE TESTS 111

Pk P
Proof. Let sk = j=1 aj , i.e. k-partial sum of an . For n > m we have

|sn − sm | = |am+1 + am+2 + · · · + an |

≤ |am+1 | + |am+2 | + · · · + |an | triangle inequality
≤ bm+1 + bm+2 + · · · + bn by assumption
X∞ X
≤ bi <  for m large enough since bi converges
i=m+1

P∞
(More precisely, given  > 0, there is N such that i=m+1 bi <  for all m > N ).

Hence {sk } is Cauchy ⇒ ∞

P
i=1 an converges.

Example 5 Show that

∞
X (−1)n
n3n+1
n=1

converges.
(−1)n P∞ 1
With an = n3n , compare with n=1 bn with bn = 3n .

1 1
n3n+1 > 3n ⇒ |an | = n
< n
n3 3
P∞
By comparison test n=1 an converges.
Example 6 Prove that if α is any positive number and |x| < 1, then the series

∞
X
nα xn converges.
n=1

First we note that nα xn → 0 as n → ∞. In fact,

|nα xn | = nα |x|n = nα en log |x|

and since log |x| < 0, the exponential decay term dominates over any power of n.
Hence nα+2 xn → 0 as n → ∞ and the sequence {nα+2 xn } is bounded. Hence there
exists a constant C such that

C
|nα+2 xn | ≤ C i.e. |nα xn | ≤ n≥1
n2

But ∞ 1
P P α n
n=1 n2 converges, hence so does n x by the comparison test. Note: it is
much easier to use the Ratio Test (below).
112 CHAPTER 13. SERIES

13.3.1 Absolute and conditional convergence

A series n=1 an is said to be absolutely convergent if the series ∞
P∞ P
n=1 |an | is conver-
gent. A series that converges but does not do so absolutely is said to be conditionally
convergent.

Theorem 7
Every absolutely convergent series is convergent.

Proof. Comparison test with bn = |an |.

Example 7
∞
X (−1)n−1
n
n=1

is conditionally convergent.
Example 8 Discuss the convergence of
∞ √
X (−1)n n
.
n+4
n=1
√
n 1 √
Series is not absolutely convergent since n+4 = √
n+4/ n
. Now for n ≥ 1,

1 1
√ √ ≥ √
n + 4/ n 5 n

√1
P
and since n
diverges, we are done.
√
n
For the alternating series test to apply we need to show that n+4 is decreasing.
√
x 4−x
If f (x) = ⇒ f 0 (x) = √ < 0 for x > 4.
x+4 2 x(x + 4)2
So series terms decrease for n > 4. We only care about what happens beyond the 1st
three terms - all the action is in the tail. Hence the series converges by the alternating
series test, but it is not absolutely convergent.

13.3.2 The Integral Test

Theorem 8
Let f (x) be a function which is defined for all x ≥ 1, and is positive and decreasing.
Then the series
∞
X
f (n)
n=1
R∞
converges if and only if the improper integral 1 f (x)dx converges.

To see this, consider the diagram:

13.3. CONVERGENCE TESTS 113

For the partial sums f (2) + f (3) + · · · + f (n) we have

Z 2 Z 3
f (2) ≤ f (3) ≤
f (x)dx, f (x)dx etc.
1 2
Z n
⇒ f (2) + f (3) + · · · + f (n) ≤ f (x)dx
1
Rn R∞
By assumption, i.e. that limn→∞ 1 f (x)dx converges, nk=2 f (k) ≤ 1 f (x)dx, i.e the
P
partial sums sk are
R ∞bounded and so the series converges. Have proved this for one of the
“ifs”, i.e. when 1 f (x) < ∞. I can add the f (1) term at the end to put the result in
the form stated by the Theorem. This is fine since I am just adding a number.
Conversely, assume that f (1) + · · · + f (n) approaches a limit for large n. Consider
the dashed rectangles in the diagram above.

Z 2 Z 3
f (1) ≥ f (x)dx, f (2) ≥
f (x)dx etc
1 2
Z n
⇒f (1) + f (2) + · · · + f (n − 1) ≥ f (x)dx
1

P∞So if the partial sums are bounded by L say, (we know this is true since by assumption
n=1 f (n) converges) we have
Z n
f (x)dx ≤ L (*)
1
R∞
Claim that this implies that 1 f (x)dx exists. Give me any number b, however large
you wish. Then I can find an integer n > b so that
Z b Z n
f (x)dx ≤ f (x)dx ≤ L by (*)
1 1
Rb
Hence 1 f (x)dx is bounded above for all b. Now send b to infinity.
114 CHAPTER 13. SERIES

Example 9 Show that

1 1
+ · · · + ≥ log(n + 1)
1+
2 n
and so obtain a new way of showing that ∞ 1
P
n=1 n diverges.

1
Solution Take f (x) = x in the integral test. Hence
Z n+1
1 1 dx
1 + + ··· + ≥ = log(n + 1)
2 n 1 x
Z n+1 ∞
dx X 1
⇒ lim diverges, and by the integral test, so does .
n→∞ 1 x n
1

Example 10 For what values of p do the series

∞
X 1
np
n=1

converge/diverge?
1
Solution Let f (x) = xp and consider

n
n1−p
Z
dx 1
p
= −
1 x 1−p 1−p

(have shown this already when we did improper integrals.)

Z n
dx
Hence lim exists if p > 1 and diverges if p ≤ 1.
n→∞ 1 xp
∞
X 1
Hence converges for p > 1, and diverges otherwise.
np
n=1

P∞ P∞
Example 11 Show that n=2 n
√1 diverges but 1
n=2 n(log(n))2 converges.
log(n)

Solution
Use the integral test by considering
Z ∞ Z b Z b
dx dx 1
p = lim p = lim (log(x))−1/2 dx
2 x log(x) b→∞ 2 x log(x) b→∞ 2 x
Z
d 1
(of the form f 0 (g(x))g 0 (x)dx since (log(x)) = )
dx x
h ib h p i
= lim 2(log(x))1/2 = lim 2(log(b))1/2 − 2 log(2) = ∞
b→∞ 2 b→∞
13.3. CONVERGENCE TESTS 115

Note that we can do this generally, i.e.

b b
(log(x))1−p
Z 
dx
lim = lim
b→∞ 2 x(log(x))p b→∞ (1 − p) 2
1−p (log(2))1−p
 
(log(b))
= lim −
b→∞ 1−p 1−p

converges if p > 1, so for p = 2 it converges. p = 1 must be done separately, i.e. consider.

Z b
dx
lim = lim [log(log(x))]b2 −→ ∞
b→∞ 2 x log(x) b→∞

Hence by the integral test, the series

∞
X 1
n(log(n))p
n=2

converges for p > 1 and diverges otherwise.

13.3.3 The Ratio Test

Theorem 9
Let ∞
P
n=1 n be a series satisfying
a

an+1
lim = L.
n→∞ an

Then:

1. If L < 1 the series converges absolutely.

2. If L > 1 the series diverges.

3. If L = 1 the test is inconclusive.

Example 12 Prove that the series

∞
X xn
n!
n=0

converges for all values of x. By the ratio test:

xn an+1 |x|n+1 n!
an = , lim = lim
n! n→∞ an n→∞ (n + 1)! |x|n

|x|
= lim =0
n→∞ n + 1
P∞ xn
In fact we will see that n=0 n! = ex .
116 CHAPTER 13. SERIES

Proof. By assumption an+1

an is close to L for large n.
Case (1), L < 1. Then pick  > 0 so small that L +  < 1, and for sufficiently large
N we have
an+1
< (L + ) if n > N.
an
Now start with aN +K . By the above bound

|aN +K | < (L + )|aN +K−1 | < (L + )2 |aN +K−2 | < · · · < (L + )K |aN |.
∞
X ∞
X
Hence, |aN +j | < |aN | (L + )j
j=1 j=1
P∞
which is bounded since the series j=1 (L + ) is a geometric series ( (L + ) < 1. Hence

|aN +1 | + |aN +2 | + . . . converges

⇒ |a1 | + · · · + |aN | + |aN +1 | + |aN +2 | + . . .
∞
X
= |an | also converges
n=1

(We only added a finite number of terms). This proves absolute convergence in case
(1) L < 1.
In case (2), L > 1, pick L +  > 1 now and we have |aN +K | > (L + )N |aN | which
diverges now as a geometric series. To prove that if L = 1 the test is inconclusive, it is
sufficient to pick an example, i.e.
∞  p  p
X 1 an+1 n 1
, = =
np an n+1 1 + 1/n
n=1
 p
1
Since p is fixed, we have limn→∞ 1+1/n = 1p = 1. But if p > 1 we have convergence
but if p ≤ 1 divergence. Hence, test is inconclusive.
Using this proof, we have a practical way of estimating errors in truncating series.
Suppose
an
< r < 1 for n > N.
an−1
Then
∞
X N
X ∞
X
an − an = an
n=1 n=1 n=N +1

is the error made in approximating the infinite series by a N −terms finite series. But
|aN +1 | < r|aN | and generally |aN +K | < |aN |rK . So
∞ ∞ ∞
X X X r
|an | = |aN +K | ≤ |aN | rk = |aN | .
1−r
n=N +1 k=1 k=1

r
So the error is ≤ |aN | 1−r .
13.3. CONVERGENCE TESTS 117

Example 12 What is the error made in approximating

∞ 4
X 1 X 1
by
n! n!
n=1 n=1

an
Solution: We have an−1 = n1 .
an
In the example, the truncation is N = 4, so if n > 4, an−1 < 15 .
1/5 1 1 1
The error is ≤ |a4 | · 1−1/5 = 4! · 4 = 96 < 0.0105

13.3.4 The Root Test

Theorem 10
For the given series ∞ 1/n = L. Then
P
n=1 an , suppose that limn→∞ |an |

1. If L < 1 the series converges absolutely.

2. If L > 1 the series diverges.

3. If L = 1 the test is inconclusive.

Proof. Similar to that for the ratio test.

Case (1), pick  > 0 so that L +  < 1 and for large N , |an |1/n < (L + ) < 1 for
n > N . Hence
|an | < (L + )n for n > N.
Now compare
∞
X ∞
X
|an | with (L + )n .
n=N +1 n=N +1

ThePlatter converges (geometric series with L +  < 1) hence ∞

P
n=N +1 |an | converges
∞
⇒ n=1 |an | converges.
Case (2) is simply ∞ |a | > ∞ n
P P
P∞n=N +1 n n=N +1 (L+) where now L+ > 1, i.e. diverges.
Case (3), consider n=1 an with an = n. Clearly series diverges but limn→∞ n1/n = 1.
(Why!?)
118 CHAPTER 13. SERIES

Examples:

1.
∞
X 1 1
converges, |an |1/n = → 0.
nn n
n=1

2.

∞
X 3n
diverges. (We already know this by other methods.)
n2
n=1
3n
i.e. an = 6→ 0 as n→∞
n2
3
|an |1/n = →3 as n → ∞.
(n1/n )2

3.

∞
X nn
(again we have seen this before in HW3)
n!
n=1
(n + 1)n+1 n! 1 n
 
an+1
Use ratio test = = 1+
an (n + 1)! nn n
an+1
lim = e > 1 diverges.
n→∞ an

4.

∞
1
X
.
− log(n) n2
n=1
X 1
Intuition: large n series ≈ <∞
n2
1 1
2
can be bounded above by with 0 < α < 1
n − log(n) αn2
n2 − log n = αn2 + (1 − α)n2 − log n > αn2
since 0 < α < 1 and (1 − α)x2 − log x ≥ (1 − α) > 0 when x ≥ 1
∞ ∞
X 1 X 1
⇒ 2
< <∞
n − log n αn2
1 1
13.3. CONVERGENCE TESTS 119
P∞
13.3.5 Testing convergence for n=1 an
120 CHAPTER 13. SERIES
Chapter 14

Power Series

An infinite series of the form

∞
X
an xn = a0 + a1 x + a2 x2 + · · · , (14.1)
n=0

is called a power series in x. The series (14.1) can be generalised to a power series in
x − x0 by
X∞
an (x − x0 )n = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · , (14.2)
n=0

and it is clear that (14.2) can be transformed into (14.1) by replacing x − x0 in (14.2) by
x to give (14.1). Unless stated otherwise we will be considering power series in x, and
it is understood that there is no loss of generality in doing so. We have the following
formal definition.

14.1 Convergence tests and radius of convergence

Definition
and {an }n≥0 be a
Let x be a real number (can extend to complex numbers also) P
sequence of numbers. Then we can form the power series ∞ n
n=0 an x . The
PN n
partial sums sN = n=1 an x are degree N polynomials.

A series of the form (14.1) is said to converge at the point x if

N
X
lim an xn ,
N →∞
n=0

exists with value equal to the sum of the infinite series. Obviously (14.1) converges at
x = 0. The question is for what values of x does a power series exist? Here are three
canonical examples of power series with different behaviour:
P∞ 2 2 3
1. n=0 n! x = 1 + x + 2! x + 3! x + · · ·

121
122 CHAPTER 14. POWER SERIES

P∞ xn x2 x3
2. n=0 n! =1+x+ 2! + 3! + ···
P∞ n
3. n=0 x = 1 + x + x2 + x3 + · · ·

Considering the convergence of each of these series using the ratio test, we find that
series 1 diverges for all x 6= 0, series 2 converges for all x, and series 3 converges for
|x| < 1. In fact series 3 for |x| < 1 is a geometric series and we can conclude that
∞
1 X
= xn , if |x| < 1,
1−x
n=0

hence the power series in this case is equal to a recognisable function. This is not
always the case, in fact it is seldom the case, but what is important is that a convergent
power series defines some function f (x) where it converges. This brings us to the crucial
question “for what values of x does a given power series converge?”

n
P
Theorem 1 Assume that there is a number P∞ R >n 0 such that n=0 |an |R con-
verges. Then for all |x| < R, the series n=0 an x converges absolutely.

Proof.
|an | |x|n ≤ |an |Rn
Hence, we obtain absolute convergence by using the comparison test with the given
series.

Definition
The greatest value
Pof R fornwhich we get convergence is called the radius of
convergence and ∞ n=0 an x converges absolutely if |x| < R. Note that x = ±R
must be tested separately.

One tool we will use for testing the convergence of power series is the ratio test as
we detail below.

Theorem 2 - Ratio Test for power series

Let ∞ n an+1
P
n=0 an x be a power series and assume that limn→∞ an = L exists.
1
Let R = L. (If L = 0 let R = ∞, if L = ∞ let R = 0.)
Then

(i) If |x| < R the series converges absolutely.

(ii) If |x| > R the power series diverges.

(iii) If x = ±R, the series may converge or diverge.

14.1. CONVERGENCE TESTS AND RADIUS OF CONVERGENCE 123

Proof. Use the ratio test for the power series:

an+1 xn+1 an+1

n
= |x| −−−→ L|x| by hypothesis.
an x an n→∞

1
For convergence, L|x| < 1, ⇒ |x| < L = R, where R is the radius of convergence
defined earlier.

Just as with the usual infinite series we used the root test, we also use it for power
series as detailed below.

Theorem 3 - Root test for power series

Let ∞ n 1/n = L exists.
P
n=0 n x be a power series, and assume that limn→∞ |an |
a
Then the radius of convergence of the power series is R = 1/L

P∞ n
Proof. For n=0 an x use the root test.
1
lim |an |1/n |x| = L|x| ⇒ |x| < =R for convergence.
n→∞ L

Examples:
(i) Determine the radius of convergence of
∞
X np
xn where p > 0 is given.
(n + 1)!
n=0

Solution: Ratio test

(n + 1)p (n + 1)! 1 p 1
 
an+1
= = 1+ → 0 as n → ∞.
an (n + 2)! np n n+2

np
Hence L = limn→∞ an+1 = 0 and the radius of convergence R = ∞, i.e. ∞ p
P
an n=0 (n+1)! x
converges for all x.
P∞ xn
(ii) n=0 n Ratio test - (will do it directly now, without the L intermediate step.)

xn+1 n n
lim = lim |x| = |x|.
n→∞ (n + 1) xn n→∞ n + 1

Hence convergence if |x| < 1. Radius of convergence is R = 1. The points x = ±1

must be done separately:
X1
x=1 series is hence diverges
n
X (−1)n
x = −1 series is hence converges by the alternating series test
n
124 CHAPTER 14. POWER SERIES

P∞ n,
Note - also mentioned earlier: Instead of n=0 an x could define power series
centered at points other than 0, i.e.
∞
X
an (x − x0 )n .
n=0

Everything is the same, simply substitute x − x0 = y.

(iii) Find the interval of convergence of the following power series (i.e. find for what
values of x it converges):
∞
X 4n
√ (x + 5)n .
n=0
2n + 5
Use ratio test,
( s )
an+1 2n + 5
lim |x + 5| = lim 4|x + 5|
n→∞ an n→∞ 2(n + 1) + 5
= 4|x + 5|.

Convergence if |x + 5| < 14 . i.e.

1 1
− <x+5< ,
4 4
21 19
− < x <− .
4 4
P∞
At the point x + 5 = 1/4 the series becomes √ 1
n=0 2n+5 which diverges (e.g.
comparison test or integral test).
(−1)n
On the other hand at x+5 = −1/4 the series becomes ∞
P
n=0 2n+5 which converges
√

by the alternating series test.

Note that in general each power series within its radius of convergence defines a
function f (x). Usually we do not have a standard or recognisable function. In fact we
seek solutions of differential equations in terms of power series - it is the only way to
obtain analytical information. We will do this later as an application.
It is crucial to know whether power series can be differentiated or integrated term
by term. This is discussed next.

14.2 Differentiation and integration of power series

For polynomials of degree m, i.e. a0 + a1 x + · · · + am xm := fm (x), we can differentiate
or integrate so that
m m
an xn+1
Z
dfm X X
= nan xn−1 and fm (x)dx = .
dx n+1
n=1 n=0
14.2. DIFFERENTIATION AND INTEGRATION OF POWER SERIES 125

The question is, can we do this for power series? The answer is YES if |x| < R,
i.e. we are within the radius of convergence. We have the following very important
theorems.

Theorem 4 Let f (x) = ∞ n

P
n=0 an x be a power series which converges absolutely
for |x| < R. Then f (x) is differentiable for |x| < R, and
∞
X
f 0 (x) = nan xn−1 .
n=1

Theorem 5 Let f (x) = ∞ n

P
n=0 an x be a power series that converges absolutely
for |x| < R. Then in the interval |x| < R, we have
∞
an xn+1
Z X
f (x)dx = .
n+1
n=0

Conclusion: For a power series within its radius of convergence, we can differentiate or
integrate term by term.

Note f (x) = ∞ n
P
n=0 an x can be differentiated an infinite number of times as long as
|x| < R, and the derivatives will exist. The function is smooth.
The way to show this is to consider each differentiated series as a new power series.
For example

dk X n
 X
an x = n(n − 1) . . . (n − (k − 1))xn−k an
dxk
X n!
= an xn−k .
(n − k)!

Ratio test

(n + 1)! (n − k)! an+1

lim |x|
n→∞ (n + 1 − k)! n! an
 
n+1 an+1
= lim |x| = L|x|
n→∞ n + 1 − k an

i.e. the same radius of convergence as the undifferentiated power series. The integer k
is arbitrary so the power series can be differentiated as many times as we want and the
derivatives will exist as long as |x| < R. Similarly, we can integrate as many times as
needed.
126 CHAPTER 14. POWER SERIES

Definition
If a function f (x) has a power series expansion f (x) = ∞ n
P
n=0 an (x − x0 ) in the
neighbourhood of the point x = x0 , then it is said to be analytic at x0 . In
(n)
addition an = f n!(x0 ) . The power series is called the Taylor series of f (x) at the
point x0 .

Example
Write down power series for

x
log 1 + x2 .


and
1 + x2

Solution
1
Recall the geometric series formula 1 + r + r2 + · · · = 1−r for |r| < 1. If r = −x2 we
have
1
(1 − x2 + x4 − x6 + . . . ) · x = · x.
1 + x2

Hence
∞
x X
= (−1)n x2n+1 , |x| < 1 for convergence.
1 + x2
n=0

Now for the log 1 + x2 we observe that

Z
d 2x xdx
log 1 + x2 = 2



so log 1 + x =2 .
dx 1 + x2 1 + x2

If |x| < 1 we can integrate term by term using Theorem 5, i.e.

Z
2
(x − x3 + x5 . . . )dx


log 1 + x =2

x4 x6 x8
= x2 − + − + ...
2 3 4

(Constant of integration is zero since log 1 = 0). We have convergence for |x| < 1 and
also at x = ±1 since we have an alternating series now. In fact, evaluating at x2 = 1 we
obtain the formula log(2) = 1 − 12 + 31 − 14 + . . . .
14.2. DIFFERENTIATION AND INTEGRATION OF POWER SERIES 127

P∞ n
Theorem 6 - Algebraic operations. Let f (x)
P∞= n=0 an x be a power series
with radius of convergence R1 , and g(x) = n=0 bn xn be another power series
with radius of convergence R2 . Let

R = min(R1 , R2 ).

Then

(1) f (x) + g(x) = ∞ n

P
n=0 (an + bn )x for |x| < R.

(2) cf (c) = ∞ n
P
n=0 can x for |x| < R1 (c 6= 0).

(3) f (x)g(x) = ∞
P Pn n
n=0 ( m=0 am bn−m ) x for |x| < R.

P∞Some nmore details of property (3) are in order. Start with writing f (x)g(x) =
n=0 cn x , i.e.
∞
! ∞ ! ∞
X X X
n n
an x bn x = cn xn , i.e.
n=0 n=0 n=0

a0 + a1 x + a2 x2 + · · · b0 + b1 x + b2 x2 + · · · = (c0 + c1 x + c2 x2 + · · · ).

I used n as a summation variable in all series. Now we ask what terms need to be
multiplied on the left to give me a term xn on the right. Here it goes: a0 with bn xn , or
a1 with bn−1 xn−1 , or · · · , to give me precisely the sum of n products with
n
X
cn = a0 bn + a1 bn−1 + · · · + an−1 b1 + an b0 = am bn−m , (14.3)
m=0

which is the form that appears in property (3) above. There is an alternative form of
P0
Pn that arises from putting k = n − m in the series in (14.3) to give k=n an−k bk =
(14.3)
k=0 an−k bk . Hence
X n Xn
cn = am bn−m = an−m bm .
m=0 m=0
128 CHAPTER 14. POWER SERIES
Chapter 15

Power series solutions of

differential equations

The objective of this chapter is to show how the power series theory developed can be
applied to construct solutions of differential equations. We will be exclusively concerned
with linear equations of first and second order 1 :
y 0 = F (x)y, y 00 + P (x)y 0 + Q(x)y = 0, (15.1)
where the functions F (x), P (x) and Q(x) are given. The objective is to find general
power series solutions of these equations once the form of F, P, Q is known. Of course,
the equation y 0 = F (x)y can be solved by separation of variables and integration, so
a development of power series solutions is only illustrative in this case. For 2nd order
equations with non-constant coefficients as is the case when P (x), Q(x) are not constant,
the situation is seldom that easy. In fact closed form solutions in terms of recognisable
elementary functions are not available. The applications of such equations are vast and
include diffraction theory, wave propagation, and quantum mechanics to mention a few.
We begin by some formal calculations to set the stage for the theory to follow.

15.1 First order equations

We begin with an example. Consider the equation
y 0 = y. (15.2)
We assume that a power series solution exists, i.e.
∞
X
y(x) = an xn = a0 + a1 x + a2 x2 + · · · + an xn + · · · , (15.3)
n=0

converges for |x| < R for some positive radius of convergence R > 0. This assumption is
equivalent to assuming that y(x) is analytic at x = 0. Because of our assumptions, we
can differentiate the power series term by term to find
∞
X
0 2 n
y = a1 + 2a2 x + 3a3 x + · · · + (n + 1)an+1 x + · · · = nan xn−1 . (15.4)
n=1
dk y
1
The notation y (k) (x) = dxk
will be used with y (1) = y 0 , y (2) = y 00 , etc.

129
130CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS

Hence equation (15.2) is satisfied if the power series (15.3) and (15.4) are equal, i.e. the
coefficients of different powers of x must match. By inspection

a1 = a0 , 2a2 = a1 , 3a3 = a2 , · · · , (n + 1)an+1 = an , · · ·

which can be used to express all the an , n ≥ 1 in terms of a0

a1 a0 a2 a0 a0
a1 = a0 , a2 = = , a3 = = , · · · , an = .
2 2 3 2·3 n!
A more direct way to get the key recursion formula (n + 1)an+1 = an is to rewrite
the sum in (15.4) so that it looks like the one in (15.3). Here is how to do it with a
substitution m = n − 1. Since n ≥ 1 in the sum, then m ≥ 0:
∞
X ∞
X ∞
X
nan xn−1 = (m + 1)am+1 xm = (n + 1)an+1 xn , (15.5)
n=1 m=0 n=0

where the last equality just switches the summation variable from m to n. Now we are
in a position to write equation (15.3) into the more useful power series form
∞
X an
[(n + 1)an+1 − an ] xn = 0 ⇒ an+1 = , (15.6)
n+1
n=0

a0
from which it follows readily that an+1 = (n+1)! , i.e. an = an!0 . Substituting this result
into the power series solution (15.3) gives

x2 xn
 
y(x) = a0 1 + x + + ··· + + ··· , (15.7)
2! n!

where a0 is an arbitrary constant. This would be fixed by a condition y(0) = a0 .

We know well from separation of variables that the solution to (15.3) is

y(x) = a0 ex . (15.8)

Hence, the power series in the square bracket of (15.7) is equal to ex as long as we are
within the radius of convergence. It remains to prove that our formal calculation of
(15.7) is valid by calculating R. Using the ratio test in (15.7) we find

|xn+1 | n! |x|
lim = lim = 0, (15.9)
n→∞ (n + 1)! |xn | n→∞ n + 1

and so R = ∞, hence the power series converges for all x and ex is analytic everywhere,
a result we already know.
Equation (15.3) had constant coefficients. The next example is

(1 + x)y 0 = py, y(0) = 1, (15.10)

where p is an arbitrary constant. Again, by separation of variables and integration we

find
y = (1 + x)p . (15.11)
15.1. FIRST ORDER EQUATIONS 131

The question is to find the power series expansion of (1 + x)p for arbitrary p, not just
integers (in fact p could be negative, irrational etc.). This is the Binomial Theorem 2
that we will encounter later also. We will do it by Psolving (15.10) assuming there exists
a power series solution with R > 0. Writing y = ∞ a
n=0 n x n , and picking the coefficient
of xn for each of the terms y 0 , xy 0 and py, we find that equation (15.11) is satisfied if
(n + 1)an+1 + nan = pan , n = 0, 1, 2, · · · (15.12)
hence, noting that a0 = 1 from the initial condition
a1 (p − 1) p(p − 1) a2 (p − 2) p(p − 1)(p − 2)
a1 = p, a2 = = , a3 = = ,
2 2 3 2·3
an−1 (p − n + 1) p(p − 1)(p − 2) · · · (p − n + 1)
an = = .
n n!
From these formulas we see that if p is a positive integer the series will terminate after
p + 1 terms as seen from the Binomial Theorem in footnote 2.
It remains to show that the power series
p(p − 1) 2 p(p − 1)(p − 2) · · · (p − n + 1) n
y = 1 + px + x + ··· + x + ··· , (15.13)
2! n!
has a positive radius of convergence. Using the ratio test it is easy to show that R = 1.
Additional Problems

1. Show that the power series (15.13) has radius of convergence R = 1.

2. Solve the equation
y 0 = x − y,
y(0) = 0, (∗)
P∞
by assuming a power series solution of the form n=0 an xn .
Does your solution represent a familiar function? Verify by solving (*) directly
using elementary methods.
3. The equation
y0 = 1 + y2, y(0) = 0, (∗∗)
can be solved using separation of variables and integration to find y = tan x.
(a) Find a power series solution of (**) and hence show that
1 2
tan x = x + x3 + x5 + . . .
3 15
[Note: you will need to use the earlier formulas for multiplication of two
infinite power series.]
2
From algebra we know that if n is a positive integer, then
n  
X n
(1 + x)n = xm ,
m
m=0

where  
n n! n(n − 1) · · · (n − m + 1)
= = .
m m!(n − m)! m!
132CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS

(b) Now get the result above by repeated differentiation of (**) and use of the
f (n) (0)
formula an = n! .

15.2 Second order equations

In this section we will consider differential equations of order higher than 1. To illustrate
matters we begin with a familiar equation, that of the simple harmonic oscillator
d2 y
+ ω 2 y = 0, (15.14)
dx2
where the parameter ω denotes the frequency of oscillation or wave number in wave
phenomena - this would be fixed by the physical problem. The general solution (15.14)
is
y = A sin ωx + B cos ωx, (15.15)
where A and B are arbitrary constants. We can easily verify this by substitution of
(15.15) into (15.14). Equation (15.14) arises in numerous applications (mechanics, chem-
ical reactions, electrical circuits, etc).
The purpose of this section is to show how we can develop solutions of equations
such as (15.14) by using a power series expansion. To simplify the notation take ω = 1
(this can be done without loss of generality since a time transformation X = ωx turns
d2 y
(15.14) into dX 2 + y = 0; now revert back to x instead of X). To this end, we suppose
that there exists a solution of (15.14) of the form
∞
X
y(x) = an xn , (15.16)
n=0

where the constants an are to be found. Taking derivatives of (15.16) gives

∞ ∞
dy X d2 y X
= nan xn−1 , 2
= n(n − 1)an xn−2 , (15.17)
dx dx
n=1 n=2

hence the differential equation becomes

∞
X ∞
X
n(n − 1)an xn−2 + an xn = 0. (15.18)
n=2 n=0

The first sum starts from n = 2 while the second one starts from n = 0. Hence, we
introduce a new variable k = n − 2 in the first sum as follows
∞ ∞
n=k+2
X X
n(n − 1)an xn−2 = (k + 1)(k + 2)ak+2 xk . (15.19)
n=2 k=0

Now the fact that there is a k in the sum in (15.19) is immaterial, I can change it back
to an n since it is a summation variable. Doing this and substituting into (15.18) gives
∞
X
[(n + 2)(n + 1)an+2 + an ] xn = 0. (15.20)
n=0
15.2. SECOND ORDER EQUATIONS 133

Equation (15.20) is equivalent to the differential equation (15.14). We know from the
theory of vector spaces that 1, x, x2 , · · · , are a linearly independent basis for polynomi-
als.3 Hence each coefficient must be zero, yielding

(n + 2)(n + 1)an+2 + an = 0, n = 0, 1, 2, · · · (15.21)

Considering n even, starting from n = 0, we have

(2 · 1)a2 = −a0 , (4 · 3)a4 = −a2 , (6 · 5)a6 = −a4 , ... ⇒

1 1 1
a2 = − a0 , a4 = a0 , a6 = − a0 , . . .
2! 4! 6!
Next consider n being odd and starting from n = 1, we find that everything can be
expressed in terms of a1 :
1 1 1
a3 = − a1 , a5 = a1 , a7 = − a1 , ...
3! 5! 7!
Hence, we can write the power series solution as

x2 x4 x3 x5
   
y(x) = a0 1 − + − . . . + a1 x − + − ... (15.22)
2! 4! 3! 5!
∞ ∞
X x2n X x2n+1
= a0 (−1)n + a1 (−1)n .
(2n)! (2n + 1)!
n=0 n=0

It will be shown in Chapter 16 that the first bracket/sum is the Taylor series of cos x
while the second bracket is that for sin x. In fact these series converge absolutely for
all x and hence the power series solution (15.22) is absolutely convergent. This is not
always the case of course. To show that the radius of convergence is R = ∞ we use the
ratio test as before. (Do this as an exercise.)
Equation (15.14) has constant coefficients and hence produced elementary solutions.
This kind of outcome is the exception rather than the rule! In applications (e.g. potential
theory, heat flow, electrostatics etc.) the equations are not as simple.

15.2.1 Ordinary points

We start with the homogeneous equation

y 00 + P (x)y 0 + Q(x)y = 0. (15.23)

For general P (x), Q(x), elementary solutions are not available. As expected the solutions
will depend on the form and behaviour of the functions P and Q. Equation (15.14) is
the simplest we could have hoped for - it has P = 0 and Q = const. In this section we
concentrate on situations where solutions are sought near a point x = x0 and in addition
we take P (x) and Q(x) to be nice well behaved functions near x0 , i.e. they are analytic
at x = x0 (all their derivatives exist and are bounded). Such points are called ordinary
points of the differential equation.
3
If you have not seen this yet, reason as follows: the representation (15.20) is assumed to be valid
for all x, and so the only way the left hand side is zero is if the coefficients of x0 , x1 , etc are identically
equal to 0.
134CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS

To illustrate we consider the Legendre equation

(1 − x2 )y 00 − 2xy 0 + p(p + 1)y = 0, (15.24)

where p is a given constant. Equation (15.24) is of the form of (15.23) with

2x p(p + 1)
P (x) = − , Q(x) = .
1 − x2 1 − x2

We see that P (x) and Q(x) are analytic near x = 0, and so x = 0 is an ordinary point.
In fact they are not analytic at x = ±1, they become infinite there. Hence, we cannot
expect our series solution to have an infinite radius of convergence - a good guess would
be that it is R = 1, but we will prove this. P
Seeking a series solution of the form y = ∞ n
n=0 an x , substituting this into (15.24)
and collecting all terms contributing to the coefficient of xn (see previous detailed cal-
culations) gives

(n + 1)(n + 2)an+2 − (n − 1)nan − 2nan + p(p + 1)an = 0, (15.25)

which can be re-arranged to give the recursion formula

(p − n)(p + n + 1)
an+2 = − an . (15.26)
(n + 1)(n + 2)

It can be seen that all even index terms a2 , a4 , · · · are expressible in terms of a0 , and all
odd ones a3 , a5 , · · · are expressible in terms of a1 . Since a0 and a1 are arbitrary constants,
the construction gives us two linearly independent solutions as desired. Calculating gives

p(p + 1) (p − 1)(p + 2)
a2 = − a0 , a3 = − a1 ,
1·2 2·3

(p − 2)(p + 3) p(p − 2)(p + 1)(p + 3)

a4 = − a2 = a0 ,
3·4 4!
(p − 3)(p + 4) (p − 1)(p − 3)(p + 2)(p + 4)
a5 = − a3 = a1 ,
4·5 5!
(p − 4)(p + 5) p(p − 2)(p − 4)(p + 1)(p + 3)(p + 5)
a6 = − a4 = − a0 ,
5·6 6!
(p − 5)(p + 6) (p − 1)(p − 3)(p − 5)(p + 2)(p + 4)(p + 6)
a7 = − a5 = − a1 ,
6·7 7!
and so on. Hence, the assumed series solution is formally given by
 
p(p + 1) 2 p(p − 2)(p + 1)(p + 3) 4 p(p − 2)(p − 4)(p + 1)(p + 3)(p + 5) 6
y(x) = a0 1 − x + x − x + ···
2! 4! 6!

(p − 1)(p + 2) 3 (p − 1)(p − 3)(p + 2)(p + 4) 5
+a1 x − x + x
3! 5!

(p − 1)(p − 3)(p − 5)(p + 2)(p + 4)(p + 6) 7
− x + ··· . (15.27)
7!
15.2. SECOND ORDER EQUATIONS 135

If p is not an integer then the radius of convergence of each series in the brackets is
R = 1. The easiest way to prove this is to use the recursion formula (15.26) and the
ratio test. Considering the even terms first, we can replace n in (15.26) by 2n to find

a2n+2 x2n+2 (p − 2n)(p + 2n + 1)

2n
= − |x|2 → |x|2 as n → ∞,
a2n x (2n + 1)(2n + 2)

hence we have convergence for |x| < 1, i.e. R = 1. The calculation for the odd terms is
completely analogous and is left as an exercise. This justifies the assumption of a power
series and the solution is analytic as long as |x| < 1.
Note: The solutions (15.27) for general p are not elementary. If p is a non-negative
integer, then one of the series in (15.27) terminates after a finite number of terms and
hence is a polynomial. The first series terminates if p is even and the second one if p is
odd. In either case the remaining series remains infinite and non-elementary. The result-
ing polynomials are called Legendre polynomials and play a huge role in mathematical
physics.
We have the following important theorem that connects the properties of P and Q
to those of the solutions.

Theorem
Let x0 be an ordinary point of

y 00 + P (x)y 0 + Q(x)y = 0, (15.28)

and let a0 , a1 be arbitrary constants. Then there exists a unique function y(x) that
is analytic at x0 and solves the differential equation (15.28) with initial conditions
y(x0 ) = a0 , y 0 (x0 ) = a1 . Furthermore, if the power series expansions of P (x) and
Q(x) are valid for |x − x0 | < R with R > 0, then the power series expansion of
y(x) is also valid on the same interval.

The proof, albeit straightforward, is beyond the scope of this course and is omitted.
The important aspect of the theorem is that once we know the radius of convergence for
P and Q we know that for the solution with no further calculation. Recall the Legendre
equation solved earlier. Since P and Q contain 1/(1 − x2 ) = 1 + x2 + x4 + · · · , we
immediately have that R = 1 since the series is geometric. Hence the series solution
(15.27) also has radius of convergence R = 1 without the need of the ratio test that was
carried out.

15.2.2 Regular singular points Optional Reading

Here we consider equations when the functions P (x) and Q(x) fail to be analytic at a
point x = x0 where a solution is sought. To see what we mean, reconsider the Legendre
equation (15.24)
2x 0 p(p + 1)
y 00 − y + y = 0. (15.29)
1 − x2 1 − x2
In the previous section we constructed power series solutions near x = 0 where P (0) and
Q(0) exist. What would happen if we want solutions near x = ±1? In such cases we
136CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS

have
lim |P (x)| = ∞ = lim |Q(x)|,
x→±1 x→±1
and so we have a fundamentally different problem.
Here is another example from mathematical physics, the Bessel equation of order p
1 0 x2 − p2
x2 y 00 + xy 0 + (x2 − p2 ) = 0 ⇒ y 00 + y + y = 0, (15.30)
x x2
which clearly has (for p 6= 0)
lim |P (x)| = ∞ = lim |Q(x)|.
x→0 x→0

Here is a crucial definition:

Definition - Regular singular points

A singular point x0 of equation (15.23) is said to be regular if the functions
(x − x0 )P (x) and (x − x0 )2 Q(x) are analytic. Otherwise the point x0 is irreg-
ular.

For equation (15.29) we have near x = 1

2(x − 1) 1 (x − 1)2 p(p + 1) (1 − x)p(p + 1)
(x−1)P (x) = − 2
= , (x−1)2 Q(x) = 2
= ,
1−x 1+x 1−x 1+x
which are clearly analytic at x = 1, with a similar calculation near x = −1.
For the Bessel equation (15.30) we have
1 x2 − p 2
xP (x) = x = 1, x2 Q(x) = x2 = x2 − p2 ,
x x2
so again we see that x = 0 is a regular singular point according to our definition. In
solving problems, this is the first thing we need to verify.
What does the definition really say? Let us consider, without loss of generality, the
singular point being at the origin x = 0. The functions P (x) and Q(x) are singular at
x = 0 but xP (x) and x2 Q(x) are analytic. This means, then, that they have power series
expansions, i.e.
b−1
xP (x) = b−1 + b0 x + b1 x2 + · · · ⇒ P (x) = + b0 + b1 x + · · ·
x
c−2 c−1
x2 Q(x) = c−2 + c−1 x + c0 x2 + c1 x3 + · · · ⇒ Q(x) = 2 + + c0 + c1 x + · · ·
x x
Here is a key step that will allow us to find the solutions. Use these power series
expansions for P (x) and Q(x) in (15.23) to find
 
00 b−1 c
−2 c−1 
y + + b0 + b1 x + · · · y 0 + + + c0 + c1 x + · · · y = 0, (15.31)
x x2 x
and on multiplying through by x2 we obtain
   
x2 y 00 + x b−1 + b0 x + b1 x2 + · · · y 0 + c−2 + c−1 x + c0 x2 + · · · y = 0. (15.32)
15.2. SECOND ORDER EQUATIONS 137

Notice that the leading order terms have been underlined. If I write down an equation
that involves only the underlined terms then this is

x2 y 00 + b−1 xy 0 + c−2 y = 0, (15.33)

and we drop the −1 index from the constants for simplicity to write

x2 y 00 + bxy 0 + cy = 0. (15.34)

Equations such as (15.34) are called Euler differential equations. They have non-constant
coefficients but have very simple solutions that can be found by making the transforma-
tion z = log x, assuming x > 0 for the moment. Using the chain rule we have

dy dy dz dy 1
= = ,
dx dz dx dz x

d2 y 1 d2 y
   
d 1 dy 1 dy 1 d dy 1 dy
2
= =− 2 + =− 2 + 2 2.
dx dx x dz x dz x dx dz x dz x dz
Substituting these expressions into (15.34) gives

d2 y dy
2
+ (b − 1) + cy = 0, (15.35)
dz dz

which can be solved by looking for a solution of the form y = eλz so that the indicial
equation is
λ2 + (b − 1)λ + c = 0. (15.36)
If the roots are λ1 and λ2 then we have the following linearly independent solutions

e λ1 z and eλ2 z λ2 6= λ1 ,
e λ1 z and zeλ1 z λ2 = λ1 .

Since z = log x, i.e. x = ez , then the solutions are

xλ1 and xλ2 λ2 6= λ1 , (15.37)

λ1 λ1
x and x log x λ2 = λ1 . (15.38)

Note that if we look for solutions in x < 0, it is enough to put t = −x and then work
with t > 0 so that the results above hold. Note also that λ1 and/or λ2 can be negative.
The Euler solutions (15.37) or (15.38) are suggestive of the solutions when x0 is a
regular singular point. The leading singular part(s) are multiplied by power series of the
form
y = xλ (a0 + a1 x + a2 x2 + · · · ), (15.39)
where λ must be found and if the roots are equal a second solution (see (15.38)) of the
form
y = xλ log x(a0 + a1 x + a2 x2 + · · · ). (15.40)
Such series solutions are known as Frobenius solutions.
A worked example:
138CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS

To illustrate the method of Frobenius consider the equation

(1/2) + x (−1/2)
2x2 y 00 + x(2x + 1)y 0 − y = 0 or y 00 + + y = 0. (15.41)
x x2
Since xP (x) = (1/2) + x and x2 Q(x) = −1/2, the point x = 0 is a regular singular point.
Hence, we look for a Frobenius solution of the form

y = xλ (a0 + a1 x + a2 x2 + · · · ), (15.42)

and substitute this into the first of equations (15.41). A common factor xλ−2 can be
cancelled to yield

a0 λ(λ − 1) + a1 (λ + 1)λx + a2 (λ + 2)(λ + 1)x2 + · · ·

 
1
+ x a0 λ + a1 (λ + 1)x + a2 (λ + 2)x2 + · · ·
 
+
2
1
− (a0 + a1 x + a2 x2 + · · · ) = 0. (15.43)
2
Next, collect powers of x and set their coefficients to zero. The first three equations from
this procedure, at x0 , x and x2 , are,
 
λ 1
a0 λ(λ − 1) + − = 0, (15.44)
2 2
 
1 1
a1 (λ + 1)λ + (λ + 1) − + a0 λ = 0, (15.45)
2 2
 
1 1
a2 (λ + 2)(λ + 1) + (λ + 2) − + a1 (λ + 1) = 0, (15.46)
2 2
···

The constant a0 can be taken to be non-zero - if it is, then the series solution (15.39)
has a common factor x which can be taken out to leave xλ+1 (a1 + a2 x + · · · ), which on
redefinition of λ+1 → λ and the constants ak → ak−1 brings us back to the form (15.39).
Hence a0 6= 0 will be assumed throughout. Hence, the indicial equation becomes

λ 1 1
λ(λ − 1) + − =0 ⇒ λ1 = 1, λ2 = − .
2 2 2
The two roots are distinct and one is negative, hence the solution will be singular. The
Frobenius method enables us to find the precise structure of the singularity. The two
independent series solutions are constructed by picking each of the indicial roots and
working out the coefficients using equations (15.44), (15.45), etc.
Starting with λ = 1 we find
a0 2
a1 = − 1 1 = − a0 ,
2·1+ 2 ·2− 2
5

2a1 4
a2 = − = a0 ,
3 · 2 + 21 · 3 − 1
2
35
15.2. SECOND ORDER EQUATIONS 139

and so on.
For λ = −1/2 we find
a0
2
a1 = 1 1

1 1 1 = −a0 ,
2 −2 + 2 · 2 − 2
a1
2 1
a2 = 3 1 1 3 1 = a0 ,
2 · 2 + 2 · 2 − 2
2
and so on.
Even though both independent solutions have been calculated in terms of a0 , it
should be understood that a0 is an arbitrary constant in each case, it is not the same.
Hence the solutions can be written as (picking c1 and c2 for a0 in each independent
solution)
   
2 4 1
y = c1 x 1 − x + x2 + · · · + c2 x−1/2 1 − x + x2 + · · · . (15.47)
5 35 2

We will conclude this section with a theorem that determines the radius of conver-
gence for the series solutions.

Theorem
Assume x = 0 is a singular point x0 of equation (15.23), and the power series
expansions of xP (x) and x2 Q(x) have radius of convergence R > 0, i.e. the power
series are valid on the interval |x| < R. If the roots of the indicial equation are
real and λ2 ≤ λ1 , then equation (15.23) has at least one solution
∞
X
λ1
y1 = x an xn (a0 6= 0),
n=0

on the interval 0 < x P < R with the coefficients an determined by recursion

formulas and the series ∞ n
n=0 an x converges for |x| < R. If λ1 − λ2 is not zero
or a positive integer, then equation (15.23) has a second independent solution
∞
X
y2 = xλ2 an xn (a0 6= 0),
n=0

where
P∞ then coefficients an are again found from the recursion formulas and
n=0 an x converges for |x| < R.
140CHAPTER 15. POWER SERIES SOLUTIONS OF DIFFERENTIAL EQUATIONS
Chapter 16

Taylor Series

This is a power series that represents a function f (x) by using its derivatives at a single
point. Intuitive construction: Assume the power series exists and identify the coefficients.
Take a fixed point x = x0 . If
∞
X
f (x) = an (x − x0 )n
n=0

converges for |x − x0 | small enough we can find the coefficients as follows:

Re-write as:

f (x) = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + a3 (x − x0 )3 + . . .
f 0 (x) = a1 + 2a2 (x − x0 ) + 3a3 (x − x0 )2 + . . .
f 00 (x) = 2a2 + 3 · 2a3 (x − x0 ) + . . .
f 000 (x) = 3 · 2 · 1a3 + . . .
dk f
f (k) (x) := = k!ak + O(x − x0 ).
dxk | {z }
(*)

(*) - This means “terms of order (x − x0 )” or smaller for (x − x0 ) small.

So we can see immediately that by putting x = x0 in the formula of f (k) (x) we find

1 (k)
ak = f (x0 ) ⇒
k!
∞
X f (n) (x0 )
f (x) = (x − x0 )n (Note 0! = 1).
n!
n=0

This is the Taylor series about the point x = x0 . If x0 = 0 we get the Maclaurin
series
∞
X f (n) (0) n
f (x) = x .
n!
n=0

141
142 CHAPTER 16. TAYLOR SERIES

In both formulas we have assumed that f (x) is infinitely differentiable on some

interval containing the point x = x0 . This of course can happen for many functions, e.g.
f (x) = sin(x), f (x) = ex , f (x) = cos(x) ...
Example 1 Maclaurin series for f (x) = sin(x) ⇒ f (0) = 0

f 0 (x) = cos(x) f 0 (0) = 1

f (2) (x) = − sin(x) f (2) (0) = 0
f (3) (x) = − cos(x) f (3) (0) = −1
f (4) (x) = sin(x) f (4) (0) = 0
repeats

y
x3 x5 x7
f (x) = sin(x) = x − + − + ...
3! 5! 7!
∞
X x2n+1
= (−1)n .
(2n + 1)!
n=0

Example 2

f (x) = ex ⇒ f (k) (x) = ex i.e. f (k) (0) = 1

∞
X xn x2 x3
ex = =1+x+ + + ...
n! 2! 3!
n=0

16.1 Taylor’s theorem with remainder

We have left an important detail out! In sending the sum to ∞, we assume that there
is convergence, and the convergence is to the function f (x). We have the following:

Theorem 1 (Taylor’s) Let f be a function defined on a closed interval between

two numbers x0 and x. Assume that the function has n + 1 derivatives on the
interval and that they are all continuous. Then

f (2) (x0 )
f (x) = f (x0 )+f 0 (x0 )(x − x0 ) + (x − x0 ) + . . .
2!
f (n) (x0 )
+ (x − x0 )n + Rn
n!
where the remainder Rn is given by
Z x
(x − t)n (n+1)
Rn = f (t) dt.
x0 n!

Proof. Use integration by parts. From the fundamental theorem of calculus we have
Z x
f (x) = f (x0 ) + f 0 (t)dt. (16.1)
x0
16.1. TAYLOR’S THEOREM WITH REMAINDER 143

Now Z x Z x
0
f (t)dt = f 0 (t) d(−(x − t)) (16.2)
x0 x0

and use integration by parts.

Z x Z x
0 0
x
(t − x)f 00 (t)dt

f (t)dt = +(t − x)f (t) x0 −
x0 x0
Z x
= (x − x0 )f 0 (x0 ) + (x − t)f (2) (t)dt.
x0

One more
x Z x
(x − t)2 (2) x (x − t)2 (3)
Z
(x − t)f (2) (t)dt = − f (t)|x0 + f (t)dt
x0 2 x0 2
Z x
(x − x0 )2 (2) (x − t)2 (3)
= f (x0 ) + f (t)dt.
2 x0 2

Repeat n times to get the result.

The choice in the integration by parts giving (16.2) may seem a little strange at first.
To motivate the choice and to understand why it is the only way to get something useful,
let us start with (16.1) and integrate by parts as we normally would, i.e. think of 1 as
the differential of t. This is what we would obtain
Z x Z x Z x
0
 0 x 0 0 0
f (t)dt = tf (t) x0 − tf (t)dt = xf (x) − x0 f (x0 ) − tf 0 (t)dt, (16.3)
x0 x0 x0

and combining (16.3) with (16.1) we get

Z x
f (x) = xf 0 (x) − x0 f 0 (x0 ) − tf 00 (t)dt. (16.4)
x0

The identity (16.4) is of course perfectly valid but it is not useful. The reason is that we
want to generate a power series whose coefficients are known in terms of the function and
its derivatives at x = x0 (see the statement of the Theorem). Going one step further, I
can also think of 1 as the differential with respect to t of (t − α) where α is any number
independent of t. Using this to carry out the integration by parts casts (16.1) into (verify
this)
Z x
0 0
f (x) = f (x0 ) + (x − α)f (x) − (x0 − α)f (x0 ) − (t − α)f 00 (t)dt. (16.5)
x0

We can see now that picking α = x will remove the second term on the RHS and lay the
ground for the development of the known power series and the proof developed above.

Alternative form of the remainder: The remainder

Z x
(x − t)n (n+1)
Rn = f (t)dt,
x0 n!
144 CHAPTER 16. TAYLOR SERIES

is in integral form and in many cases it is useful to have it as a pointwise expression.

We can do this by using the Integral Mean Value Theorem - Problem 10, sheet 3.
(x − t)n
Since x − t ≥ 0, g(t) := ≥0
Z x n! n
(x − t)
⇒ Rn = f (n+1) (c) dt,
x0 n!
f (n+1) (c)
⇒ Rn = (x − x0 )n+1 ,
(n + 1)!
where c is a number between x0 and x.
Finally, we observe that we will get convergence to f (x) if Rn → 0 as n → ∞. In
such cases the function f (x) is analytic since its (infinite) power series converges. By
analytic we mean that derivatives of any arbitrary order exist and are continuous.

16.1.1 Summary and link with power series

P∞ n
(1) If f (x) = n=0 an (x − x0 ) is a convergent power series on an open interval
centered at x0 , then f (x) is infinitely differentiable and
f (n) (x0 )
an = .
n!
i.e. We get Taylor’s formula.
(2) If f is infinitely differentiable on an open interval centered at x0 , and if Rn → 0 as
n → ∞ for all x in the interval, then the Taylor series of f converges and equals
the function, i.e.
∞
X f (n) (x0 )
f (x) = (x − x0 )n .
n!
n=0

Here is a very useful alternative of Taylor’s theorem that we use in Numerical Anal-
ysis. Put x = x0 + h (and after that x0 → x if you want)
h2 hn (n)
f (x0 + h) = f (x0 ) + hf 0 (x0 ) + f (2) (x0 ) + · · · + f (x0 ) + Rn (x0 , h)
2! n!
Z x0 +h
(x0 + h − t)n (n+1)
Rn = f (t)dt
x0 n!
h(n+1) (n+1)
= f (c)
(n + 1)!
where c is between x0 and x0 + h.

16.2 Examples, bounding the remainder, estimates

Will do this with x0 = 0 (Maclaurin), other cases follow. Use form
f (n+1) (c) n+1
Rn = x
(n + 1)!
16.3. EXPONENTIALS AND LOGARITHMS. BINOMIAL THEOREM 145

with c is a number between 0 and x. If |f (n+1) (x0 )| ≤ Mn+1 for all x0 between 0 and x.
Then
|x|n+1
|Rn | ≤ Mn+1
(n + 1)!
Example 3
x3 x2n+1
f (x) = sin(x) = x − + · · · + (−1)n + R2n+3
3! (2n + 1)!
f (2n+3) (c) 2n+3 |x|2n+3
where |R2n+3 | = x ≤ .
(2n + 3)! (2n + 3)!

10−3
Hence sin(0.1) ≈ 0.1 − 6 with an error which is less than

(0.1)7 10−7
= < 10−10 .
7! 5040

π
+ 0.2 to an accuracy of 10−4 .


Example 4 Compute sin 6

Solution 3 5
Even though sin(x) = x − x3! + x5! + . . . will converge if we take enough terms, since
π
6 + 0.2 is not small, we will need a lot πof terms to get to the required accuracy.
It is much better to expand about 6 using the formula

h2 00 hn (n)
f (x + h) = f (x) + hf 0 (x) + f (x) + · · · + f (x) + Rn
2! n!
f (n+1) (c) n+1
where Rn = h with c between x and x + h.
(n + 1)!

Now h = 0.2, f (n+1) = sin or cos.

(0.2)n+1 0.24 16 × 10−4
⇒ |Rn | ≤ ⇒ R3 ≤ = < 10−4
(n + 1)! 24 24
π  π   π  (0.2)2   π  (0.2)3   π 
⇒ sin + 0.2 ≈ sin + 0.2 cos + − sin + − cos .
6 6 6 2! 6 3! 6

16.3 Exponentials and logarithms. Binomial theorem

16.3.1 The exponential ex
x2 xn ec
ex = 1 + x + + ··· + + xn+1
2! n! (n + 1)!
| {z }
remainder
|x|n+1
If x < 0 then c < 0 and |Rn | ≤ (n+1)! .
If x > 0 and such that x ≤ b, say. Then
bn+1
|Rn | ≤ eb →0 as n → ∞.
(n + 1)!
146 CHAPTER 16. TAYLOR SERIES

Example 5
Compute e to 3 decimals. Showed earlier (see chapter on logarithms) that 2 < e < 4.
e 4
From results above, |Rn | ≤ (n+1)! ≤ (n+1)! .

Need 4
(n+1)! to be less than 10−3 .

4 1 1
Try n = 4, 5, 6 → = = > 10−3
5! 5×3×2 30
4 1 1
= = > 10−3
6! 6×5×3×2 180
4 1 1
|R6 | ≤ = = < 10−3 .
7! 7×6×5×3×2 1260
So
1 1 1 1 1
e≈1+1+ + + + + + R6 .
2 3! 4! 5! 6! |{z}
<10−3

16.3.2 The Logarithm

We have the following expansion for log(1 + x):

x2 x3 xn
log(1 + x) = x − + − · · · + (−1)n−1 + Rn+1 .
2 3 n
Need to show this. One way is Taylor’s theorem - exercise. Another way is to use
the identity (telescoping product)

(1 − t + t2 − · · · + (−1)n−1 tn−1 )(1 + t) = 1 + (−1)n−1 tn

1 tn
⇒ = (1 − t + t2 − · · · + (−1)n−1 tn−1 ) + (−1)n . (*)
1+t (1 + t)

In the interval −1 < x ≤ 1 (why?), integrate (*) between 0 and x.

x2 x3 xn
log(1 + x) = x − + − · · · + (−1)n−1 + Rn
2 3 Z x n n
t
where Rn = (−1)n dt.
0 1+t

Need to estimate this:

tn
(i) Consider 0 ≤ x ≤ a ≤ 1. Here 1 + t ≥ 1 ⇒ 1+t ≤ tn , and this implies that

xn+1 an+1
|Rn | ≤ ≤ →0 as n → ∞.
n+1 n+1

(ii) Now take −1 < a < 0 and consider x in the interval a ≤ x ≤ 0. Hence
1 + t ≥ 1 + a > 0 since t is in the interval (x, 0).
16.3. EXPONENTIALS AND LOGARITHMS. BINOMIAL THEOREM 147

tn (−t)n
≤ (t ≤ 0 remember)
1+t 1+a
0
(−t)n (−x)n+1 |a|n+1
Z
So |Rn | ≤ dt = ≤ →0 as n → ∞.
x 1+a (n + 1)(1 + a) (n + 1)(1 + a)

Hence, we have shown that when −1 < x ≤ 1 we have Rn → 0 as n → ∞ and hence we

have the convergent power series
∞
X xn
log(1 + x) = (−1)n−1 .
n
n=1

Exercise: Calculate log(1.1) to 3 decimals.

16.3.3 Binomial Expansion

If |x| < 1 we have (for any real α)

α(α − 1) 2
(1 + x)α = 1 + αx + x + ...
2!
∞
X α(α − 1) . . . (α − n + 1) n
= x .
n!
n=0

Can prove |Rn | → 0 as n → ∞ for |x| < 1, hence convergence.

16.3.4 Alternatives to L’Hôpital

Sometimes it is easier to use Taylor’s theorem instead of differentiating many times.
Example 6

sin(x) − x sin(x) cos(x) − x cos(x)

lim = lim
x→0 tan(x) − x x→0 sin(x) − x cos(x)
 3
 2
  2

x − x6 + . . . 1 − x2 + . . . − x 1 − x2 + . . .
= lim 3

2

x→0
x − x6 + · · · − x 1 − x2 + . . .
3
−x + . . . 1
= lim 1 63 =− .
x→0 x + . . .
3
2
148 CHAPTER 16. TAYLOR SERIES

Example 7

log(x)
lim put x = 1 + y
x→1 ex − e
2 3
log(1 + y) y − y2 + y3 + . . .
= lim = lim
y→0 e(ey − 1) y→0 e(1 + y + · · · − 1)
1
= .
e

16.3.5 L’Hôpital’s Rule derived from Taylor’s Theorem

Consider F (x) = fg(x)
(x)
and consider limx→a F (x) in cases where f (a) = g(a) = 0. Assume
that the first (k − 1) derivatives of f and g also vanish at x = a. i.e.

f (i) (a) =g (i) (a) = 0 i = 1, . . . , k − 1

hk−1 hk
Then f (a + h) = f (a) + f 0 (a)h + · · · + f (k−1) (a) + f (k) (c1 )
(k − 1)! k!
h k−1 hk
g(a + h) = g(a) + g 0 (a)h + · · · + g (k−1) (a) + g (k) (c1 )
(k − 1)! k!

where c1 , c2 are numbers between a and a + h.

f (k) (c1 )
⇒ F (a + h) = , send h → 0, c1 , c2 → a, so we get the required result.
g (k) (c1 )

An example of a function that does not have a Maclaurin series. Consider

( 2
e−1/x x = 6 0
f (x) =
0 x = 0.

f (x) and all its derivatives are continuous everywhere. (You have shown this in the
problem sheets). In addition, f (n) (0) = 0 for all n. So the (Taylor) Maclaurin expansion
is
f 00 (0) 2
f (0) + f 0 (0)x + x + ...
2!
The approximating polynomial is

f (n) (0) n
Pn (x) = f (0) + xf 0 (0) + · · · + x .
n!
But this is exactly zero for all

n. Hence the remainder Rn cannot go to zero. In fact it
2
must be equal to exp −1/x2 . Reason: e−1/z with z a complex number is not analytic.
2
In fact, z = iy gives f = e1/y → ∞ as y → 0. You will see more in Complex Analysis.
 Part IV

Fourier Series

149
Chapter 17

Orthogonal and orthonormal

function spaces

Will see how to represent fairly arbitrary functions (e.g. they can be discontinuous) with
approximations of smooth functions.

Definition 1
If f , g are real valued functions that are Riemann integrable on [a, b], then we
define the inner product of f and g, denoted by (f, g), by
Z b
(f, g) := f (x)g(x)dx
a
Z b 1/2
1/2 2
Note (f, f ) = f dx := ||f || ≥ 0.
a

Definition 2
Let S = {φ0 , φ1 , φ2 , . . . } be a collection of functions that are Riemann integrable
on [a, b]. If

(φn , φm ) = 0 whenever m 6= n

then S is an orthogonal system on [a, b]. If in addition, ||φn || = 1, i.e.

Rb 2
a φn dx = 1, then S is said to be orthonormal on [a, b].

φn
Note: Can easily go from orthogonal to orthonormal by considering ||φn || . The or-
thonormal trigonometric system will be used

S = {φ0 , φ1 , φ2 , . . . } where
1 cos(nx) sin(nx)
φ0 (x) = √ , φ2n−1 = √ , φ2n = √ (n = 1, 2, . . . )
2π π π

151
152 CHAPTER 17. ORTHOGONAL AND ORTHONORMAL FUNCTION SPACES

i.e. the system is

 
1 cos(x) sin(x) cos(2x) sin(2x)
√ , √ , √ , √ , √ ,... .
2π π π π π

S defined above is orthonormal on any interval of length 2π, e.g. [0, 2π], [−π, π] etc..
(You have already shown this in HW sheet 3).
Chapter 18

Periodic functions and periodic

extensions

A function f (x) is periodic with period T if

f (x + T ) = f (x)
for all values of x. It follows that a T periodic function is also mT periodic for any
integer m, i.e.
f (x ± mT ) = f (x).
e.g. sin(x) is 2π-periodic but also 4π, 6π etc. Geometrically f (x) is T periodic if a shift
by T units reproduces the shape of the function.

Start with any continuous function f (x) in an interval a ≤ x < b. Can extend this
periodically to have period T = b − a.

Figure 18.1: Function on [a, b) extended periodically and the new function is discontin-
uous at x = a + mT for any integer m.

153
154 CHAPTER 18. PERIODIC FUNCTIONS AND PERIODIC EXTENSIONS

Example 1 Extend periodically f (x) = x defined on [0, 1).

This is known as f (x) = x − [x], where [x] denotes the integer part of x (e.g. [1.25] =
1).

Example 2 Extend periodically f (x) = cos(x) on − π2 , π2 .

 

From the examples, we see that the function can be discontinuous at some points
x = ξ, i.e.

lim f (x) 6= lim f (x).

x→ξ+ x→ξ−

Definition
At points of discontinuity define
1
f (ξ) = [f (ξ+ ) + f (ξ− )] .
2
1
e.g. for f (x) = x − [x], f (n) = 2 for all integers n.

Conclusion: Given a function defined on a closed interval [a, b], extend it periodically
and at points of discontinuity x = ξ prescribe the value f (ξ) = 12 (f (ξ− ) + f (ξ+ )).

e.g. y = x 0≤x≤1
 155

18.0.1 Integrals over a period

For a periodic function f (x) of period T and for arbitrary values of a, we have
Z T −a Z T
f (x)dx = f (x)dx.
−a 0

In fact,
Z β Z β+T
f (x)dx = f (x)dx.
α α+T

Proof.
Z β Z β+T Z β+T Z β+T
f (x)dx = f (y − T )dy = f (y)dy = f (x)dx
|α {z } α+T α+T α+T
sub x = y − T
156 CHAPTER 18. PERIODIC FUNCTIONS AND PERIODIC EXTENSIONS
Chapter 19

Trigonometric polynomials

Start with an oscillation sin(ωx). Here ω is the frequency. The period is T = 2π

ω . This is
a “pure” harmonic oscillation. Signals - e.g. sound, electromagnetic waves, water waves,
are not pure oscillations, they contain higher harmonics.
Lets add another oscillation of frequency 2ω, i.e. sin(2ωx), whose period is T2 =
2π
2ω = ωπ . This is called the 1st harmonic.
Signal could be
S2 (x) = A1 sin(ωx) + A2 sin(2ωx).
S2 (x) has period T = 2π T
ω overall. 1st harmonic has period 2 . Can add more and more
higher frequencies and in fact can produce a wave (oscillation) that is a trigonometric
polynomial defined by
n
1 X
Sn (x) = a0 + [ak cos(kωx) + bk sin(kωx)] .
2
k=1

The constant 21 a0 is included ( 12 is useful as we will see later.)

Note: Went from ω1 = ω to ω2 = 2ω etc. i.e. all the frequencies have ratios that are
rational. If ωω21 is irrational, we get quasi-periodic oscillations.

19.1 Euler’s relation

Useful to use Euler’s relation

cos(θ) + i sin(θ) = eiθ

and since

cos(θ) − i sin(θ) = e−iθ (take complex conjugate)

1  iθ 
cos(θ) = e + e−iθ
2
1  iθ 
sin(θ) = e − e−iθ .
2i

157
158 CHAPTER 19. TRIGONOMETRIC POLYNOMIALS

So, can represent everything as complex and find real expressions by taking real or
imaginary parts. e.g.
d iω(x−φ)
ae = aiωeiω(x−φ) .
dx
Can also integrate
Z Z  
inx sin(nx) i cos(nx)
e dx = (cos(nx) + i sin(nx)) dx = −
n n
1
= einx .
in

19.1.1 Orthogonality
(
π
n 6= 0
Z
0
einx dx = .
−π 2π n=0
For any integers m, n we have
Z π (
inx −imx 0 n 6= m
e e dx = (easier than HW3.)
−π 2π n=m

19.2 Complex notation for trigonometric polynomials

Start with the polynomial (have set ω = 1).
n
1 X
Sn (x) = a0 + (ak cos(kx) + bk sin(kx)) .
2
k=1

Use the relations found earlier

n
eikx + e−ikx eikx − e−ikx
   
1 X
Sn (x) = a0 + ak + bk
2 2 2i
k=1
n n
1 X 1 X1
= a0 + (ak − ibk ) eikx + (ak + ibk ) e−ikx .
2 2 2
k=1 k=1

Can now write this as a single complex series as follows

n
X
Sn (x) = γk eikx (19.1)
k=−n

where
= 21 a0

γ0 

1
γk = 2 (ak − ibk ) k = 1, 2, . . . , n.

= 12 (ak + ibk )

γ−k
Notice that γk = ∗ ,
γ−k (or γk∗ = γ−k ), where ∗ denotes complex conjugate. This is not
accidental. Sn (x) in equation (19.1) is real. Hence it must equal its complex conjugate.
Calculate
19.2. COMPLEX NOTATION FOR TRIGONOMETRIC POLYNOMIALS 159

n
X n
X
Sn (x) = γk eikx , Sn (x)∗ = γk∗ e−ikx .
k=−n k=−n

Change indexing, put k = −l to find

−n
X n
X n
X
Sn (x)∗ = ∗ ilx
γ−l e = ∗ ilx
γ−l e = ∗ ikx
γ−k e .
l=n l=−n k=−n

In the last step above, I just changed the summation variable l to k. Comparing the
two, we see that they are equal iff
∗
γk = γ−k i.e. γk∗ = γ−k , identical statements.

Conversely, if we are given a complex trigonometric form

n
X
f (x) = γk eikx ,
k=−n

∗ , i.e.
then f (x) is real if and only if γk = γ−k

γk + γ−k = γk + γk∗ = real

and γk − γ−k = γk − γk∗ = pure imaginary.

1
Example 1 Take Sn (x) = cos(x) + 2 sin(x) + 3 cos(2x). Express as a complex trigono-
metric series
1 ix
i ix
3 2ix
e + e−ix − e − e−ix + e + e−2ix


Sn =
2
  4  2
1 i 1 i 3 3
= − eix + + e−ix + e2ix + e−2ix
2 4 2 4 2 2
2
X
= γk eikx
k=−2

where
1 i 1 i
γ1 = − γ−1 = + = γ1∗
γ0 = 0 2 4 2 4
3 3
γ2 = γ−2 = = γ2∗ .
2 2
160 CHAPTER 19. TRIGONOMETRIC POLYNOMIALS
Chapter 20

Fourier series

Consider the trigonometric polynomial

N
1 X
f (x) = SN (x) = a0 + an cos(nx) + bn sin(nx).
2
n=1

There are 2N + 1 coefficients to determine. Use orthogonality on the interval [−π, π] for
sin(mx), cos(nx) etc.

Find for all n (including n = 0) (See HW3)

1 π
Z
an = f (x) cos(nx)dx
π −π
1 π
Z
bn = f (x) sin(nx)dx.
π −π

Big question is: starting with fairy arbitrary functions f (x) (e.g. they are discontinuous),
can we represent them by SN by letting N → ∞?

Orthogonality properties:
If m, n are integers, then
(
π π
m 6= n
Z Z
0
sin(mx) sin(nx)dx = cos(mx) cos(nx)dx =
−π −π π if m = n 6= 0
Z π
sin(mx) cos(nx) = 0.
−π

Complex form (
π
m 6= n
Z
imx −inx 0
e e dx =
−π 2π m = n.

161
162 CHAPTER 20. FOURIER SERIES

Theorem 1
The Fourier series 21 a0 + ∞
P P∞ inx ),
n=1 (an cos(nx) + bR
n sin(nx)) (or n=−∞ αn e
1 π
formed by the Fourier coefficients an = π −π f (x) cos(nx)dx and bn =
1 π
R
π −π f (x) sin(nx)dx converges to the value f (x) for any piecewise continuous
function f (x) of period 2π which has piecewise continuous derivatives of first and
second order.a At any discontinuities, the value of the function must be defined
by f (x) = 21 [f (x+ ) + f (x− )].
Can relax the assumption of the second derivative. It is enough to have f 0 (x) be piecewise
a

continuous, i.e. the function is piecewise smooth. If f (x) is continuous, the convergence is
absolute and uniform. If it is discontinuous, absolute and uniform convergence everywhere except
at the discontinuity.

For the proof we will need some additional Lemmas.

20.1 Fourier series theorem, Riemann-Lebesgue Lemma

20.1.1 A trigonometric formula
We will prove the following - needed later
1
cn (x) = + cos(x) + cos(2x) + · · · + cos(nx)
2
sin n + 21 x


=
.
2 sin 21 x

Clearly 12 x 6= 0, ±π, ±2π, . . . i.e. x 6= 0, ±2π, ±4π, . . . If we define cn (x) at these

points by n + 21 , then the function is continuous everywhere (show this?).

Use cos(kx) = 12 eikx + e−ikx to re-write

n
1 X ikx 1 −inx
+ eix e−inx + (eix )2 e−inx + · · · + e−inx e2inx


cn (x) = e = e
2 2
k=−n

i.e. a geometric progression with ratio r = eix = cos(x) + i sin(x). Now r = 1 only if
x = 0, ±2π, . . . , i.e. the exceptional points that we excluded (treated separately). Sum
it up to find
1 1 − r2n+1 1 h −inx i 1
cn (x) = e−inx = e − ei(n+1)x .
2 1−r 2 1 − eix
1
Multiply top and bottom by e− 2 ix
h i
−i(n+ 12 )x i(n+ 12 )x
1 e − e
⇒ cn (x) = 1 1
2 e− 2 ix − e 2 ix
sin (n + 12 )x


⇒ cn (x) =
.
2 sin 12 x
20.1. FOURIER SERIES THEOREM, RIEMANN-LEBESGUE LEMMA 163

Integrate from 0 to π we find

n
!
sin (n + 12 )t
Z π

Z π
1 X π

dt = + cos(kt) dt = .
0 2 sin 12 t 0 2
k=1
2

Lemma 1 (Riemann-Lebesgue)
If the function g(x) is integrable on [a, b], (e.g. it is piecewise continuous), then
Z b
Iλ = g(x) sin(λx)dx
a

tends to zero as λ → ∞.
Proof. (Will do it when g 0 is also piecewise continuous. For the general case see HW
problems).
Can use integration by parts
Z b  b Z b
cos(λx) cos(λx) 0
Iλ = g(x) sin(λx)dx = − g(x) + g (x)dx
a λ a a λ
 Z b 
1
= g(a) cos(λa) − g(b) cos(λb) + cos(λx)g 0 (x)dx
λ a
1
⇒ |Iλ | ≤ M
λ
for some constant M , and the result follows.
Lemma 2
Z ∞
sin(z) π
dz = .
0 z 2

Proof. Show improper integral exists, i.e.

Z M
sin(z)
I = lim dz exists.
M →∞ 0 z

(Note z = 0 is not a problem. Why?) Consider 0 < M < N and calculate

Z N N Z N
sin(z) cos(z) cos(z)
IN − IM = dz = − −
dz
M z z M M z2
Z N
cos(M ) cos(N ) cos(z)
= − − dz.
M N M z2

Z N
1 1 dz 2
⇒ |IN − IM | ≤ + + 2
=
M N M z M

hence convergence since |IN − IM | can be made arbitrarily small (Cauchy).

164 CHAPTER 20. FOURIER SERIES

2
In fact, letting N → ∞, we see that |I − IM | ≤ M so IM approaches its limit
algebraically. Now take p > 0 arbitrary and pick M = λp.

Z λp z=λx Z p
sin(z) z}|{ sin(λx)
IM = Iλp = dz = (λdx)
z λx
Z 0p 0
= sin(λx)
dx
0 x

where we have now fixed the integration range to [0, p]. As M → ∞, λp → ∞ i.e.
λ → ∞, and by the estimate above
Z p
sin(λx) 2 2
I− dx ≤ =
x M λp
Z0 p
sin(λx)
i.e. lim dx = I (20.1)
λ→∞ 0 x

for all p sufficiently big. Cannot apply Riemann-Lebesgue directly. Consider the function
(
1 1
x − 2 sin(x/2) x 6= 0
h(x) =
0 x = 0.

Fact: h(x) is continuous and also has a continuous first derivative for 0 ≤ x < 2π. (Proof
see HW5). Now we use the Riemann-Lebesgue Lemma 1 to see that for 0 ≤ p < 2π
Z p  
1 1
sin(λx) − dx → 0 as λ → 0.
0 x 2 sin(x/2)

Note: The convergence is uniform for 0 ≤ p ≤ π since |h(x)| and |h0 (x)| are both
bounded in this interval. From (20.1) we have immediately,
Z p
sin(λx)
lim dx = I.
λ→∞ 0 2 sin(x/2)

1
Pick λ = n + 2 and p = π, we have shown already that

sin (n + 21 )x
Z π


π
dx =
0 2 sin(x/2) 2

independent of n. Hence we have proved:

Z ∞
sin(z) π
I= dz = .
0 z 2
20.1. FOURIER SERIES THEOREM, RIEMANN-LEBESGUE LEMMA 165

20.1.2 Proof of Theorem 1

Start with nth “Fourier polynomial”
n
1 X
Sn (x) = a0 + (ak cos(kx) + bk sin(kx))
2
k=1

and substitute the formulas for ak , bk , change order of summation and integration (finite
sum, so ok), to find
n
" #
1 π
Z
1 X
Sn (x) = f (t) + (cos(kt) cos(kx) + sin(kt) sin(kx)) dt
π −π 2
k=1
n
" #
1 π
Z
1 X
= f (t) + cos(k(t − x)) dt
π −π 2
k=1
sin (n + 12 )(t − x)
 
1 π
Z
= f (t)
dt
π −π 2 sin 12 (t − x)
1 π−x f (x + ξ) sin (n + 12 )ξ
Z

substitute ξ = t − x ⇒ = dξ
2 sin 21 ξ


π −π−x
sin (n + 12 )ξ


1 π
Z
= f (x + ξ)
dξ (20.2)
π −π 2 sin 21 ξ

by using properties of integrals of periodic functions discussed earlier. Note that x is a

fixed number. We will prove that (and this proves the Theorem):

sin (n + 12 )ξ


1 π
Z
lim f (x + ξ)
dξ = f (x).
n→∞ π −π 2 sin 12 ξ

At all points x ∈ [−π, π], even points of discontinuity, we have

1
f (x+ ) + f (x− ) .

f (x) =
2
We have proven already that

sin (n + 12 )t
Z π


π
1

dt =
0 2 sin 2 t 2

and by change of variables t = −t0 we also find

sin (n + 21 )t0
Z 0


π
1 0

dt0 = .
−π 2 sin 2 t 2

Hence
sin (n + 12 )t sin (n + 12 )t
π



1 0
Z Z
1 + −
f (x) = f (x )
dt + f (x )
dt.
π 0 2 sin 12 t π −π 2 sin 12 t
166 CHAPTER 20. FOURIER SERIES

Using this identity gives

 sin (n + 12 )ξ
Z π


1  +
Sn (x) − f (x) = f (x + ξ) − f (x )
dξ
π 0 2 sin 12 ξ
 sin (n + 12 )ξ


1 0 
Z
−
+ f (x + ξ) − f (x )
dξ
π −π 2 sin 12 ξ
What is left to do is to prove the following: (see HW)
(i) Prove
f (x + ξ) − f (x+ )
sin 12 ξ

is piecewise continuous and so is the 1st derivative, on 0 ≤ ξ ≤ π.

(ii) Prove
f (x + ξ) − f (x− )
sin 12 ξ

is piecewise constant along with its 1st derivative on −π ≤ ξ ≤ 0.

Then by Riemann-Lemma, Sn → f (x) as n → ∞, i.e. convergence (uniform away from
discontinuities).

20.2 Examples, sine and cosine series

Will consider f (x) to be 2π-periodic.
(i) If f (x) is even, i.e. f (−x) = f (x), then
1 π
Z
f (x) sin(nx)dx = bn = 0.
π −π
So f (x) has only a cosine series. If f (x) is odd, f (−x) = −f (x) then an = 0 and
f (x) has a sine series.
(ii) If a function is defined on [0, π] by an expression f (x), then it can be extended as
an even or odd function on [−π, π]. e.g.
f (x) = π − x 0≤x≤π
(A) Extend to an even function
(
π−x 0≤x≤π
f (x) =
π+x −π ≤ x ≤ 0.
20.2. EXAMPLES, SINE AND COSINE SERIES 167

(B) Extend to an odd function, f (−x) = −f (x)


π − x
 0<x<π
f (x) = −π − x −π < x < 0

0 x = 0, π, −π.


Consider the second function on [−π, π]. It is


π − x
 x>0
f (x) = 0 x=0 (20.3)

−π − x x > 0.


Here it is periodically extended:

Aside: General result for odd functions f (x):

Z π Z π
1 2
an = 0 bn = f (x) sin(nx)dx = f (x) sin(nx)dx (show this!)
π −π π 0

Similarly for f (x) even:

Z π
2
bn = 0 an = f (x) cos(nx)dx.
π 0
168 CHAPTER 20. FOURIER SERIES

Hence for (20.3),

2 π
Z
bn = (π − x) sin(nx)dx
π 0
cos(nx) π
   Z π 
2 cos(nx)
= (π − x) − − dx
π n 0 0 n
cos(nx) sin(nx) π
  
2
= −(π − x) −
π n n2 0
2 π
h i 2
= = convergence is uniform as long as  < |x| ≤ π
π n n 
sin(2x) sin(3x)
⇒ f (x) = 2 sin(x) + + + ... .
2 3
This now gives us for free the Fourier series of the function
φ(x) = x − π < x < π.

This is also odd of course, but we get φ(x) from f (x) by (i) shifting the latter to the
right by π (ii) reflecting about x = 0.
⇒ φ(x) = f (−(x − π)) = f (π − x)
 
sin(2(π − x)) sin(3(π − x))
= 2 sin(π − x) + + + ...
2 3
 
sin(2x) sin(3x)
= 2 + sin(x) − + − ...
2 3
∞
X sin(kx)
=2 (−1)k+1 .
k
k=1

Convergence is uniform as long as |x| < π −  for any small  > 0. In particular, putting
x = π2 we recover the Leibnitz series
 
π 1 1 1
= 2 1 − + − + ...
2 3 5 7
d P∞
Note: Cannot differentiate dx 2 k=1 (−1)k+1 sin(kx)
k and get a convergent series. Rea-
son: derivative of φ(x) does not satisfy conditions of Fourier Theorem.
20.2. EXAMPLES, SINE AND COSINE SERIES 169

Example 2:
f (x) = |x| −π ≤x≤π

Function is now continuous but has discontinuous derivatives at a set of finite points
±nπ.
1 X
Even function ⇒ f (x) = a0 + an cos(nx)
2
n=1

and

2 π
Z
an = x cos(nx)dx
π 0
sin(nx) π cos(nx) π
 
2
= x +
π n 0 n2 0
2 1
= (cos(nπ) − 1) 2
π
( n
0 n even n 6= 0
⇒ an = 4
− πn2 n odd
2 π
Z
a0 = x dx = π
π 0
 
1 4 cos(3x) cos(5x)
⇒ |x| = π − cos(x) + + + . . . .
2 π 32 52

Convergence is uniform at all x. Put x = 0 we find a formula for π 2 , i.e.

∞
π2 X 1 1 1
= 2
= 1 + 2 + 2 + ...
8 (2n + 1) 3 5
n=0

Example 3

−1 for − π < x < 0

f (x) = sgn(x) = 0 x=0

+1 0 < x < π

170 CHAPTER 20. FOURIER SERIES

P
Clearly f (x) is odd ⇒ f (x) = bn sin(nx)
π
cos(nx) π
Z  
2 2
bn = −
sin(nx)dx =
0 π π n 0
(
2 (1 − cos(nπ)) 0 n even
= = 4
π n nπ n odd
 
4 sin(3x)
⇒ sgn(x) = sin(x) + + ... .
π 3
π
Check: function f (x) = 0 at x = nπ, uniform convergence elsewhere. Putting x = 2
again gives  
π 1 1 1
= 2 1 − + − + . . . . Leibnitz
2 3 5 7
d
Note: dx |x| = sgn(x) two series agree everywhere except at discontinuities. Ques-
tion: why can I differentiate |x| series but not φ(x)? Former is sectionally or piecewise
continuous.

20.3 Complex form of Fourier series

Have already shown that for f (x) real
∞
1 X
f (x) = a0 + (an cos(nx) + bn sin(nx))
2
n=1
∞
X
= γn einx , −π < x < π
n=−∞

where )
1
γn = 2 (an − ibn )
1
for n = 1, 2, . . .
γ−n = 2 (an + ibn )

1 1 π
Z
1
γn = (an − ibn ) = · (f (x) cos(nx) − if (x) sin(nx)) dx
2 2 π −π
Z π
1
= f (x)e−inx dx.
2π −π
20.4. FOURIER SERIES ON 2L−PERIODIC DOMAINS 171

Similarly Z π
1
γ−n = f (x)e+inx dx.
2π −π
(Clearly γn∗ = γ−n since f (x) is real).
Hence,
∞
X
f (x) = γn einx −π <x<π
n=−∞
where Z π
1
γn = f (x)e−inx dx n = 0, ±1, ±2, . . .
2π −π
Note: If the period is 2L instead of 2π
∞ Z L
X 1
f (x) = γn e inπx/L
, |x| < L, γn = f (x)e−inπx/L dx, n = 0, ±1, ±2, . . .
−∞
2L −L

20.4 Fourier series on 2L−periodic domains

The set of functions
1 1  nπx  1
√ , √ cos , √ n = 1, 2, . . .
2L L L L
are orthonormal on [−L, L] (and in fact on any interval [a, a + 2L] since the function is
periodic). In addition,
Z L  nπx   mπx  Z L  nπx   mπx 
sin sin d= cos cos dx
−L L L L L
(−L
L m=n
=
0 m 6= n
Z L  nπx   mπx 
and sin cos dx = 0
−L L L

∞ h
1 X  nπx   nπx i
⇒ f (x) = a0 + an cos + bn sin
2 L L
n=1
1 L
Z  nπx 
where an = f (x) cos dx
L −L L
1 L
Z  nπx 
bn = f (x) sin dx.
L −L L
The complex form is
∞
X
f (x) = γn einπx/L |x| ≤ L (20.4)
n=−∞
Z L
1
γn = f (x)e−inπx/L dx n = 0, ±1, ±2, . . . (20.5)
2L −L
172 CHAPTER 20. FOURIER SERIES

20.5 Parseval’s theorem

If f (x) is represented by its Fourier series
∞
1 X
f (x) = a0 + an cos(nx) + bn sin(nx), −π ≤ x ≤ π
2
n=1

then we have Z π ∞
1 1 X
f 2 dx = a20 + (a2n + b2n ).
π −π 2
n=1

Proof. Easier to use complex notation

1
γn = (an − ibn )
∞
2
X
inx 1
f (x) = γn e where γ−n = (an + ibn ) = γn∗
n=−∞
2
1
γ0 = a0
2!
∞ ∞
!
X X
(f (x))2 = γn einx γm eimx .
n=−∞ m=−∞

Integrate and use orthogonality - see earlier

∞ ∞ ∞
" #
Z π X X X
2 2 2 2 2
[f (x)] dx = 2π γn γ−n = 2π |γn | = 2π |γ0 | + (|γn | + |γ−n | )
−π −∞ −∞ n=1
Z π ∞
1 1 X
⇒ f dx = a20 +
2
(a2n + b2n )
π −π 2
n=1

as needed.
Example 4 Compute the Fourier series of cos(x/2) over (−π, π]. Use Parseval’s theorem
to deduce the value of
∞
X 1
.
(4n − 1)2
2
n=1
20.6. FOURIER TRANSFORMS AS LIMITS OF FOURIER SERIES 173

Function is even ⇒
x 1 X∞
cos = a0 + an cos(nx) − π ≤ x ≤ π
2 2
n=1
2 π
Z x 4
a0 = cos dx =
π 0 2 π
Z π
2  x 
an = cos cos(nx)dx
π 0 2
2 π1
Z     
1 1
= cos (n + )x + cos (n − )x dx
π 0 2 2 2
"
#
1 1


1 sin (n + 2 )π sin (n − 2 )π
= 1 +
π n+ 2 n − 12
" #
(−1)n
 
1 cos(nπ) cos(nπ) 2 2
= − = −
π n + 12 n − 12 π 2n + 1 2n − 1
(−1)n −4
= .
π 4n2 − 1
By Parsevel’s theorem
∞
1 π
Z
x 1 X
cos2 ( )dx = a20 + a2n
π −π 2 2
n=1
∞
8 16 X 1
= +
π2 π2 (4n2 − 1)2
1
∞
X 1 π2 −8
LHS = 1 ⇒ = .
(4n2 − 1)2 16
n=1

20.6 Fourier transforms as limits of Fourier series

We discussed 2π periodic functions in detail. Consider now f (x) periodic on [−L, L]
with L arbitrary. We have shown that
∞
X
f (x) = γn einπx/L −L≤x≤L
n=−∞
Z L
1
where γn = f (t)e−inπt/L dt n = 0, ±1, ±2, . . .
2L −L
Put γn into the sum to find
∞  Z L 
X 1 −inπt/L
f (x) = f (t)e dt einπx/L .
n=−∞
2L −L

This is exact, we want to send L → ∞.

∞ Z L 
1 X
f (x) = h f (t)e−inht dt einhx
2π n=−∞ −L
174 CHAPTER 20. FOURIER SERIES

π
where h = L. In the limit L → ∞, h → 0 but nh := ωn = O(1).
∞ Z L 
1 X −iωn t
f (x) = h f (t)e dt eiωn x .
2π n=−∞ −L

P∞
This is of the form n=−∞ G(ωn )h. Now h = ωn+1 − ωn = (n + 1)h − nh := δω
∞
X Z ∞
⇒ Riemann sum G(ωn )δω → G(ω)dω.
n=−∞ −∞

This gives, sending L → ∞,

Z ∞ Z ∞ 
1 −iωt
f (x) = f (t)e dt eiωx dω
2π −∞ −∞

where f (x) is defined on R.

This gives the Fourier Transform pair
Z ∞
1
f (x) = fˆ(k)eikx dk
2π −∞
Z ∞
fˆ(K) = f (x)e−ikx dx
−∞

Very useful in many applications. You will use them a lot to solve differential equa-
tions.
 Part V

Further Applications in Physics

and Geometry

175
Chapter 21

Theory of plane curves

21.1 Parametric representation

We have already seen simple examples of parametric curves, e.g. x = cos t, y = sin t that
describes a unit circle as t varies. The reason we want such parametrisations is because
the equation y = f (x) may assign two or more points to y for each x. For example a
circle of unit radius can be represented by two functions
p p
y = 1 − x2 , y = − 1 − x2 , −1 ≤ x ≤ 1. (21.1)

There is a general way of representing curves in the plane as the zero level curves of
a function of several variables. You will encounter a lot more of this in Multivariable
Calculus, but for our needs we can see that the function

f (x, y) = 0, (21.2)

gives an implicit representation of y in terms of x. For example f (x, y) = x2 + y 2 − 1 = 0

gives the circle (21.1) above. Solving (21.2) is typically hard. This leads us to the
parametric representation of the curve.
We introduce a third independent variable t and a curve C is represented by

x = φ(t) = x(t), y = ψ(t) = y(t), (21.3)

where we will use the shorter notation x(t), y(t) without confusion. We will assume
throughout that φ(t) and ψ(t) possess continuous derivatives.
Given an interval t0 ≤ t ≤ t1 , equation (21.3) represents a curve C in the plane. This
curve can cross itself, can be multivalued etc., see the examples of a circle (21.1) and an
ellipse whose parametrisation is

x(t) = a cos t, y(t) = b sin t. (21.4)

21.2 Change of parameters

The parameter t maps to different points on the curve C. It is a type of “label” for
different points. The parametrisation is not unique. In fact we can derive many different
parametrisations as follows.

177
178 CHAPTER 21. THEORY OF PLANE CURVES

Take an arbitrary function τ = h(t) that is monotonic and continuous in the t−interval
of interest. We are interested in using τ in the parametrisation rather than t. The in-
verse of h exists, hence we can write t = σ(τ ) in a corresponding τ −interval. Then the
parametrisation of (21.3) becomes

x = φ(σ(τ )) = α(τ ), y = ψ(σ(τ )) = β(τ ), (21.5)

with α(τ ), β(τ ) also continuous. Since σ is monotonic, for each t there exists a unique
τ . The role of τ is to “re-label” the points on the curve; this is very useful in Numerical
Analysis where we may want to cluster nodes in a given region of the curve.
Example 1 For the line y = x we have an obvious parametrisation
x = t, y = t, −∞ < t < ∞. (21.6)

But I can also put τ = t3 , in which case t = τ 1/3 and

x = τ 1/3 , y = τ 1/3 , −∞ < τ < ∞. (21.7)

Regularly spaced points in t give regularly spaced points on the curve (21.6). However,
regularly spaced points in τ will give points that are further apart on the line (21.7).
This is illustrated in Figure 21.1 where the marked points are ti = 0.02i, i = 0, . . . 50 on
the left panel, and mapped according to (21.7) on the right panel.
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x

Figure 21.1: Left panel parametrisation (21.6). Right panel parametrisation (21.7).
21.3. MOTION ALONG A CURVE WITH TIME AS A PARAMETER 179

Example 2 Consider the parametrisation for the ellipse (21.4), i.e.

x(t) = a cos t, y(t) = b sin t, 0 ≤ t < 2π.

The chosen interval for t describes the ellipse exactly once in the positive (anti-clockwise)
direction. The transformation

τ = tan(t/2), −∞ < τ < ∞,

gives the rational parameter representation of the ellipse

a(1 − τ 2 ) 2bτ
x= , y= . (21.8)
1 + τ2 1 + τ2

All points of the ellipse are recovered except the point at (−a, 0) which is recovered
asymptotically close as τ → ±∞.
Show (21.8) as an exercise. Use trigonometric identities, e.g. cos t = 2 cos2 (t/2) − 1,
etc., and write sin(t/2), cos(t/2) in terms of τ .
Example 3 We can remove singularities in a curve by an appropriate parametrisation.
Consider the function y = x2/3 . There is a cusp at x = 0, since dy/dx = (2/3)x−1/3
which is unbounded at x = 0. We can parametrise the curve by the smooth functions

x = t3 , y = t2 , −∞ < t < ∞.

Of course dy/dx = (dy/dt)/(dx/dt) = (2/3t) = (2/3x1/3 ) as before, hence still singular,

but the parametric functions are smooth.

21.3 Motion along a curve with time as a parameter

These are curves traced out by the trajectory of the point (x(t), y(t)) where t denotes
time. As an example we will study in detail the cycloid. This is generated as follows:

1. Place a circle of radius a on a flat surface y = 0.

2. Define the origin O to be the point of contact with the circle.

3. Label the point of the circle at the origin by P and define this time as t = 0

4. Now roll the circle to the right (without any sliding) with unit angular velocity so
that at time t the circle has turned from its original orientation by an angle t.

5. This means that the centre M of the circle moves with constant speed a to the
right.

6. The cycloid is the curve traced by the point P as the circle moves to the right. A
sketch is given in Figure 21.2.
180 CHAPTER 21. THEORY OF PLANE CURVES

4.5

3.5

3
y

2.5

1.5
M
1 a
P t
a
0.5

0
0 1 N 2 3 4 5 6
x
Figure 21.2: Construction of the cycloid by a rolling circle on a flat plane.

The coordinates of M are (at, a), and noting that the angle P M N is t, we immedi-
ately have from simple trigonometry
x = a(t − sin t), y = a(1 − cos t), t ≥ 0. (21.9)
The blue branch of the cycloid drawn in Figure 21.2 corresponds to 0 ≤ t ≤ 2π. As t
increases the figure repeats identically - it is a periodic function of period 2π.
Expression (22.45) is compact and elegant, unlike the x − y version. The latter can
be calculated by noting that cos t = a−y
a , i.e. t = cos
−1 a−y . Follow this through to show
a
that
a−y
x = a cos−1
p
∓ y(2a − y). (21.10)
a
Note that (21.10) is an equation for x in terms of y unlike the usual y in terms of x. The
multivalued nature of the cos−1 function is what gives a countable infinity of x values
for a given 0 < y < 2a.
Note: If we take the solution (22.45) and send y → −y and also shift up by a units, we
obtain the curve
x = a(t − sin t), y = a cos t, t ≥ 0. (21.11)
This is now a convex curve and we can study what happens if particles are released and
slide under gravity in the bowl type region of the curve. Such problems are taken up
later.
21.3. MOTION ALONG A CURVE WITH TIME AS A PARAMETER 181

Exercise 1: Find dy/dx at x = 2nπ, n = 0, 1, . . .. Do it two ways, first by using the

parametric form (22.45), and then by using the x − y form (21.10).

Exercise 2: Sketch the curve x = t2 − 1, y = t3 − t, for −∞ < t < ∞ and indicate with
arrows the direction of increasing t starting from −∞. To help you along with minimal
calculations, consider the first time the curve passes through the origin, the next time it
again passes through the origin, and what happens between these times.

Challenge Problem

(a) Consider a fixed circle Ca of radius a with its centre at (0, 0) in the usual Cartesian
coordinates. A second circle Cb of radius b is rolling at a uniform speed along the
circumference and outside of the first circle. Suppose that the the circle Cb is
rolling in such a way that its centre has rotated about the origin to an angle t in
time t. A point P is fixed on Cb and at t = 0 the coordinates of P are at the
contact point (a, 0).
Show that the position of P as the circle Cb rolls over the circle Ca is given by the
parametric equations
 
a+b
x(t) = (a + b) cos t − b cos t ,
b
 
a+b
y(t) = (a + b) sin t − b sin t .
b

In the special case a = b = 1, sketch the trajectory of P and analyse what happens
in the neighbourhood of (1, 0).

(b) Now take b < a but allow the circle Cb to roll as above but on the interior circum-
ference of Ca . Again start with the point P at the position t = 0, and show that
its trajectory is
 
a−b
x(t) = (a − b) cos t + b cos t ,
b
 
a−b
y(t) = (a − b) sin t − b sin t .
b

What happens geometrically when b = 21 a, in other words what kind of trajectory

do we obtain? Can this be a useful mechanical device?
Now take b = a/3. Show that

5 4
x2 + y 2 = a2 + a2 cos(3t),
9 9

and use this result to show that the trajectory meets the circle at exactly three
points. Sketch the resulting curve.
182 CHAPTER 21. THEORY OF PLANE CURVES

21.3.1 Parametric derivatives, tangents and normals

Recall (21.3)
x = φ(t) = x(t), y = ψ(t) = y(t),
the derivatives with respect to t are denoted by ẋ(t), ẏ(t), using Newton’s notation.
Physically, these are the velocity components of the coordinates of the point P as it
moves along the curve C.
If ẋ(t) 6= 0 over some t−interval, then C can be represented by a function y = f (x).
To see how this is done, invert x = φ(t) - we can do this as long as φ̇(t) 6= 0 - to find
t = φ−1 (x). Then substitute into the y expression to find y = ψ(φ−1 (x)) ≡ f (x). By
the chain rule the slope of the function is then
dy
dy dy dt dt ẏ
= = dx
= , (21.12)
dx dt dx dt
ẋ

and equivalently dx/dy = ẋ/ẏ holds if ẏ 6= 0. There is one detail I skipped, namely the
result
dt 1
= dx .
dx dt
This is not because of the properties of fractions. It is due to the derivative of inverse
functions that we have seen before. In fact using the notation above
dt d −1 1 1
= φ (x) = 0 = .
dx dx φ (t) ẋ(t)
Recall that in taking the derivative of the inverse function and writing it in terms of the
original function, we evaluate the latter in its domain, i.e. the domain of φ−1 is x and
that of φ is t. See previous chapters.
The tangent to the curve always exists as long as ẋ and ẏ do not vanish at the same
point, i.e.
ẋ2 + ẏ 2 6= 0, (21.13)
If ẋ = 0, ẏ 6= 0, then the tangent is vertical, and if ẋ 6= 0, ẏ = 0 the tangent is horizontal.

Example: For the cycloid curve (22.45) derived earlier, we have

dy ẏ
ẋ = a(1 − cos t), ẏ = a sin t ⇒ = = cot(t/2). (21.14)
dx ẋ
From (21.14) we see that ẋ = ẏ = 0, and hence ẋ2 +ẏ 2 = 0, whenever t = 0, ±2π, ±4π, . . ..
The curve at these points has a cusp with dy/dx becoming infinite there. At ±π, ±3π, . . .
we have dy/dx = 0 as seen in Figure 21.2 for t = π.
To see the local cusplike behaviour near x = 0, say, we would need to find y = f (x) as
described above, i.e. start with x = a(t − sin t) and invert this in an appropriate interval
to find t ≡ t(x) (this is the φ−1 (x) function introduced above), to then substitute into
y = a(1 − cos t) to find f (x). Such a calculation is impossible to do in closed form,
hence we proceed asymptotically. We use the following results that will be proved in
later chapters. For the moment take them as given. If t is small, then
t3 t2
sin t = t − + ..., cos t = 1 − + .... (21.15)
3! 2!
21.3. MOTION ALONG A CURVE WITH TIME AS A PARAMETER 183

Near x = 0 we know that t is small and hence we have

t3 t2
   
a 3 a
x = a t − t + + . . . ≈ t , y = a(1 − cos t) = a 1 − 1 + + . . . ≈ t2 .
3! 6 2! 2

These expressions can now be inverted to give t = (6x/a)1/3 and hence

 2/3
a 6
y≈ x2/3 . (21.16)
2 a

Clearly this is a cusp since dy/dx ∼ x−1/3 .

Challenge Problem
Show the cusp structure (21.16) found above parametrically BUT starting from the
implicit equation of y as a function of x given by (21.10). You will only need the result
2
cos ξ = 1 − ξ2! + . . . given above when ξ is small.

Next consider a point on the parametric curve that corresponds to t = t0 , i.e. consider
the point P0 = (x0 , y0 ) where x0 = φ(t0 ) and y0 = ψ(t0 ). We wish to find the equations
for the tangent and normal directions to the curve at P0 . The equation for the tangent
is
dy
y − y0 = (x − x0 ). (21.17)
dx P0

But dy/dx = ẏ/ẋ at t = t0 , hence the equation of the tangent is

(y − y0 )ẋ − (x − x0 )ẏ = 0. (21.18)

At points where ẋ2 + ẏ 2 = 0 (i.e. ẋ and ẏ are zero simultaneously), equation (21.18) does
not make sense. Geometrically we would be trying to find a tangent at a point such as
a corner or a cusp.
The normal to the curve is a straight line perpendicular to the tangent, i.e. it has
slope −ẋ/ẏ at P0 . Hence the equation is

(y − y0 )ẏ + (x − x0 )ẋ = 0. (21.19)

As an example, consider Exercise 2 in section 21.3, i.e. the curve x = t2 −1, y = t3 −t,
for −∞ < t < ∞. We see that when t = −1 and t = +1, we obtain x = y = 0, that is
the curve crosses itself at the origin. The tangent directions are different at t = −1 and
t = +1 as can be seen by calculation:
t = −1 x+y =0
t = +1 x − y = 0.
Finally note that if α is the angle the tangent makes with the horizontal, then
dy ẏ
= = tan α. (21.20)
dx ẋ
184 CHAPTER 21. THEORY OF PLANE CURVES

It follows from this that

ẋ ẏ
cos α = p , sin α = p , (21.21)
ẋ + ẏ 2
2 ẋ2 + ẏ 2
for α > 0, i.e.when the slope is positive. If α < 0, i.e. the slope is negative, then we have
ẋ ẏ
cos α = − p , sin α = − p . (21.22)
ẋ2 + ẏ 2 ẋ2 + ẏ 2

21.4 Arc length representation of curves

We will consider rectifiable curves, i.e. curves that have finite length. They can be
piecewise smooth, in which case all the constructions that follow can be done in a
piecewise manner.
Start with an arbitrary parametric representation of a curce C,
x = x(t), y = y(t), α ≤ t ≤ β. (21.23)
Starting from a point P0 where t0 ∈ [α, β), we have already seen that the length of the
curve between P0 and an arbitrary point with parameter t ∈ (t0 , β] is given by
Z tp
s = s(t) = ẋ2 (t0 ) + ẏ 2 (t0 ) dt0 , (21.24)
t0

and on differentiation with respect to t

s 
dx 2
 2
ds dy
= + , (21.25)
dt dt dt
which suggests the shorthand notation
p
ds = dx2 + dy 2 .
If t is time then x(t), y(t) are the coordinates of a moving particle that traces out the
curve C. The speed of the particle is equal to the rate of change, with respect to t, of
the distance travelled, i.e.
ds s(t + h) − s(t)
ṡ = = lim . (21.26)
dt h→0 h
• As a special case, let the speed be constant. Then ds/dt is constant and hence s(t)
is a linear function of t. Physically, the particle still moves on a curve but in such
a way that the speed is constant. Think of a car travelling at 50 kmh over a hilly
road.
Assuming, as usual, that ẋ2 + ẏ 2 6= 0, then ds/dt 6= 0 and we can introduce s as the
parameter instead of t. Using the chain rule we have
dx dx dt ẋ
= =p = cos α, (21.27)
ds dt ds ẋ + ẏ 2
2

dy dy dt ẏ
= =p = sin α. (21.28)
ds dt ds ẋ + ẏ 2
2
21.5. CURVATURE 185

From (21.27)-(21.28) it follows that

 2  2
dx dy
+ = 1, (21.29)
ds ds

which must be satisfied if s is arc length along the curve.

21.5 Curvature

The curvature at a point of a curve is defined to be the rate at which the angle of
inclination α (the angle that the tangent at the point makes with the horizontal) changes
with arc length. This can be expressed as

dα
κ= . (21.30)
ds

The geometry of the situation is given in Figure 21.3.

To calculate (21.30) for general parametric curves x = x(t), y = y(t), we assume

that first and second derivatives exist and are continuous, and also ẋ2 + ẏ 2 6= 0. As we
can see from the figure - see also (21.20) - the quantity tan α = ẏ/ẋ is defined unless
ẋ = 0. If ẋ = 0 (geometrically this is a vertical tangent), then the function cot α = ẋ/ẏ is
defined, so we can use whichever is appropriate. This is a technical point which does not
affect our final result. For completeness, however, what we need to do is the following.
Given an interval t ∈ [t0 , t1 ] that contains the point P where we wish to calculate the
curvature, we define α in terms of ẋ and ẏ as follows:

ẏ
tan α = if ẋ 6= 0, and (21.31)
ẋ
ẋ
cot α = if ẏ 6= 0. (21.32)
ẏ
186 CHAPTER 21. THEORY OF PLANE CURVES

K =
IF
.

Figure 21.3: Curvature at a point.

We can differentiate either of (21.31), (21.32) and the result is the same. Will do it
for (21.31) with implicit differentiation and use of the quotient rule, i.e.

dα ẋ ÿ − ẍ ẏ ẋ ÿ − ẍ ẏ ẋ2 ẋ ÿ − ẍ ẏ ẋ ÿ − ẍ ẏ
sec2 α = ⇒ α̇ = cos2 α = = 2 .
dt ẋ2 ẋ2 2
ẋ + ẏ 2 ẋ 2 ẋ + ẏ 2

Using the formula for ṡ in equation (21.25) and the result above, we have

dα α̇ ẋ ÿ − ẍ ẏ
κ= = = 2 . (21.33)
ds ṡ (ẋ + ẏ 2 )3/2

If the arc length s is the parameter, then the dot means an s−derivative, and on use of
(21.29) we have
κ = xs yss − ys xss .
Note: The sign of κ is positive if α increases as s increases and negative if α decreases
with increasing s. Hence, κ > 0 if the function is convex and κ < 0 if it is concave. See
Figure 21.4.
Next, consider the curvature of the curve given by the function y = f (x). In this
case x is the parameter and so the expression (21.33) gives

y 00
κ= . (21.34)
(1 + y 02 )3/2

This formula shows that κ > 0 if y 00 > 0 and such curves are called convex. On the other
hand κ < 0 if y 00 < 0 and we define such curves as concave. At a point where y 00 = 0 we
have a change in sign of κ, i.e. a point of inflexion that is already familiar from earlier
work.
21.5. CURVATURE 187

T÷÷¥÷÷÷¥ I •X
at
.

CENCI
it •

Figure 21.4: Positive κ and negative κ for convex functions (left) and concave functions
(right).

Problems:

1. Find the curvature of a circle x = a cos t, y = a sin t.

2 2
2. Repeat for the ellipse xa2 + yb2 = 1 by (i) using its parametric form, (ii) working
directly with y as a function of x. If a > b, find the points where the curvature has
its maximum magnitude. Could you have anticipated the result geometrically?
3. Find the curvature of y = x3 and discuss your findings geometrically.
188 CHAPTER 21. THEORY OF PLANE CURVES
Chapter 22

Laplace transforms and

applications, the brachistochrone
and Abel’s mechanical problem

22.1 Definition and properties

The Laplace transform of a function f (x) is a linear operator L[·] defined as follows:
Z ∞
L[f (x)] = e−px f (x)dx. (22.1)
0

We will use the notation

F (p) := L[f (x)]. (22.2)
In what follows a capital letter will be used for the Laplace transform of a function, e.g.
L[y(x)] = Y (p), L[g(x)] = G(p) etc.
Rb
This makes sense only if the integral exists, i.e. limb→∞ 0 e−px f (x)dx exists. The
answer is a function of the parameter p. Generally p must be constrained for the integral
to exist.
Here are some examples. All of these can be carried out using elementary methods
of integration - do them as an exercise.
R∞
1. f (x) = 1, F (p) = 0 e−px dx = p1 . Converges for p > 0.
R∞
2. f (x) = x, F (p) = 0 e−px x dx = p12 . Converges for p > 0.
R∞
3. f (x) = xn , F (p) = 0 e−px xn dx = pn+1n!
. Converges for p > 0. (Here n is
a positive integer.)
R∞
4. f (x) = eax , F (p) = 0 e−px eax dx = p−a1
. Converges for p > a.
R∞
5. f (x) = sin ax, F (p) = 0 e−px sin ax dx = p2 +a
a
2. Converges for p > 0.
R∞ p
6. f (x) = cos ax, F (p) = 0 e−px cos ax dx = p2 +a 2. Converges for p > 0.
R∞
7. f (x) = sinh ax, F (p) = 0 e−px sinh ax dx = p2 −a
a
2. Converges for p > a.

189
190CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

R∞ p
8. f (x) = cosh ax, F (p) = 0 e−px cosh ax dx = p2 −a2
. Converges for p > a.
Note that we can use the linearity of L and the results above to write down Laplace
transforms of simple functions. Here is an example:
 
1 1 1 1 1 p
L cos2 ωx = L
 
+ cos 2ωx = L [1] + L [cos 2ωx] = + . (22.3)
2 2 2 2 2p 2(p2 + 4ω 2 )
Problem: Use the result (22.3) and without integrations calculate L[sin2 ωx].
Solution: Since sin2 ωx + cos2 ωx = 1, we have
1 1 p 1 p
L[sin2 ωx] = L[1 − cos2 ωx] = − − = − .
p 2p 2(p2 + 4ω 2 ) 2p 2(p2 + 4ω 2 )

Remark: Piecewise continuous functions that are also of exponential order (see below
for the definition), have Laplace transforms. The intuitive reason is that integrals of
piecewise continuous functions over finite domains exist. Therefore, we can deal with
numerous functions found in applications, e.g. the Heaviside function (or step function),
the square function found in electrical circuits etc. See problems for these.
R∞
The existence of L[f (x)] centres on the existence of the improper integral 0 e−px f (x)dx.
This has been discussed in general already and essentially we need to consider the inte-
grand e−px f (x) for large x. If f (x) is of exponential order then the integral exists as we
see next.

Definition
f (x), x > 0, is said to be of exponential order if there exist constants M and c
such that

|f (x)| ≤ M ecx . (22.4)

As examples, we note that a bounded function is of exponential order with c = 0, eax

is of exponential order with c = a, and xn of exponential order with any c > 0.
Now if f (x) satisfies (22.4) then we have
Z ∞ Z ∞ Z ∞
−px −px M
|F (p)| = e f (x)dx ≤ |e f (x)|dx ≤ M e−(p−c)x dx = , p > c.
0 0 0 p−c
Hence, F (p) → 0 as p → ∞. This is in fact true whenever F (p) exists. Importantly,
if a function G(p) does not satisfy G(∞) = 0, then it cannot be the Laplace transform
of any function g(x) (even piecewise continuous functions).
What we have shown above is that any piecewise continuous function of exponential
order has a Laplace transform. These are sufficient conditions for the existence of F (p)
but they are not necessary. The following example (that will be useful later) illustrates
this. Rb
Consider f (x) = x−1/2 . This has an infinite discontinuity at x = 0 but 0 x−1/2 dx
exists as we have seen before. It is bounded for x large, hence its Laplace transform
exists. In particular for p > 0 we have
Z ∞
−1/2
L[x ]= e−px x−1/2 dx.
0
22.2. APPLICATIONS TO DIFFERENTIAL EQUATIONS 191

A series of substitutions px = t and then t = s2 gives

Z ∞ Z ∞ r
−1/2 −1/2 −t −1/2 t=s2 −1/2 −s2 π
L[x ]=p e t dt = 2p e ds = .
0 0 p

Problems

1. Sketch each of the following functions and find their Laplace transforms (it is
understood that in all cases x > 0):
(i) f (x) = H(x − a) where a > 0 and the Heaviside function H(x) is defined by

0 x<0
H(x) =
1 x≥0

(ii) f (x) = [x] where [x] is the integer part of x (e.g. [π] = 3).
(iii) 
sin x 0 ≤ x ≤ π
f (x) =
0 x>π

2. Discuss the existence of L[x−1 ].

3. Let  be a positive number and

1/ 0≤x≤
f (x) =
0 x>
R∞
Show that 0 f (x)dx = 1. Find L[f (x)] and show that lim→0 L[f (x)] = 1.
R∞
[This suggests that 0 δ(x)dx = 1 and L[δ(x)] = 1, where δ(x) is the Dirac delta
function.]

22.2 Applications to differential equations

We will discuss solutions to inhomogeneous equations of the form

y 00 + ay 0 + by = f (x), (22.5)

with initial conditions

y(0) = y0 , y 0 (0) = y1 . (22.6)
In addition, a, b are constants and f (x) is piecewise continuous and of exponential or-
der. You will see general methods for solving such equations in the second part of the
module, but here we consider how to attack these problems with Laplace transforms. It
is important to note that these general methods are hard to apply when f (x) is discon-
tinuous with numerous points of discontinuity, hence the motivation of using Laplace
transforms.
Due to linearity of L, taking the Laplace transform of equation (22.5) gives

L[y 00 ] + aL[y 0 ] + bL[y] = L[f (x)] (22.7)

192CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

Using integration by parts we calculate

Z ∞ Z ∞
∞
L[y 0 ] = e−px y 0 dx = ye−px 0
+p e−px ydx = pL[y] − y0 . (22.8)
0 0

Using this result (or performing two integrations by parts) we find

L[y 00 ] = L[(y 0 )0 ] = pL[y 0 ] − y1 ,
hence
L[y 00 ] = p2 L[y] − py0 − y1 . (22.9)
The differential equation (22.5) transforms into an algebraic one for L[y] := F (p) that
can be solved to give
L[f (x)] + (p + a)y0 + y1
L[y] = . (22.10)
p2 + ap + b

Example Find the solution of

y 00 + 4y = 4x, y(0) = 1, y 0 (0) = 5.
Taking the Laplace transform and using L[x] = 1/p2 and the results above transforms
the problem into
4
p2 Y (p) − p − 5 + 4Y (p) = 2 ,
p
hence
p 5 4 p 4 1
Y (p) = + + = 2 + + , (22.11)
p2 + 4 p2 + 4 p2 (p2 + 4) p + 4 p2 + 4 p2
where partial fractions were used in the last step. Refer to the results of section 22.1 to
recognise that (22.11) is
L[y] = Y (p) = L[cos 2x + 2 sin 2x + x] ⇒ y = cos 2x + 2 sin 2x + x,
a result that could have been anticipated with using elementary methods.
We have used the notion of the inverse Laplace transform. For example we had
L[g(x)] = p2p+4 and asked the question what g(x) is; we recognised g(x) = cos 2x as one
of the standard results of section 22.1. In fact we introduced another linear operator
L−1 say, whereby L−1 [p/(p2 + 4)] = cos 2x. Generally, then, we have the pair

L[g(x)] = G(p) L−1 [G(p)] = g(x).

Note: There are general ways of finding inverse transforms using contour integration in
the complex plane rather than relying on recognising the inverse transforms or looking
them up in tables. This is beyond the scope of this course but will be encountered in
Year 2. However for completeness, I will make the connection to Fourier transforms and
derive the form of the integral that needs to be carried out to compute inverse Laplace
transforms.
This section is concluded by deriving the following shifting formula and give examples
of its use. The formula says that given a function f (x) whose Laplace transform is
F (p) = L[f (x)], then
L[eax f (x)] = F (p − a). (22.12)
22.3. DERIVATIVES AND INTEGRALS OF LAPLACE TRANSFORMS 193

To show this we observe

Z ∞ Z ∞
L[eax f (x)] = e−px eax f (x)dx = e−(p−a)x f (x)dx = F (p − a).
0 0

Here is an example. Using L−1 [1/p2 ] = x we have L−1 [1/(p + 1)2 ] = e−x x.
Problems
6 p+3 1
1. Find the inverse Laplace transforms of (i) (p+2)2 +9
, (ii) p2 +2p+5
, (iii) (p+3)4
.

2. Show that (using our notation L[f (x)] = F (p))

Z x 
F (p)
L f (t)dt = .
0 p
h i
Verify this formula by finding L−1 p(1+p)1
in two ways.

22.3 Derivatives and integrals of Laplace transforms

Start with our general formula for the Laplace transform
Z ∞
F (p) = e−px f (x)dx,
0

and differentiate with respect to p under the integral sign (permissible since the integral
is absolutely convergent) to find
Z ∞
0
F (p) = e−px (−xf (x))dx ⇒ L[−xf (x)] = F 0 (p).
0

Differentiating this again gives

L[x2 f (x)] = F 00 (p),
and differentiating n times gives
L[(−1)n xn f (x)] = F (n) (p).

Solved Examples
1. Find L[x sin x]. We use L[sin x] = 1/(p2 + 1) and the result above to find
 
d 1 2p
L[x sin x] = − 2
= 2 .
dp p + 1 (p + 1)2

2. We saw in Section 22.1 that L[x−1/2 ] =

p
π/p. Let’s use this result to find L[x1/2 ].
Calculate
r  r
−1/2 d d π 1 π
L[x 1/2
] = L[x(x )] = − L[x−1/2 ] = − = .
dp dp p 2p p
Here is another
R x −1/2way using the result in Problem 2 in Section 22.2. Observe that
1/2
x =2 0 t dt
194CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

Next we consider the use of Laplace transforms in solving differential equations that
do not have constant coefficients. The methods will be illustrated with Bessel’s1 equation
of order zero, namely
xy 00 + y 0 + xy = 0. (22.13)
Equation (22.13) is second order and so two independent solutions exist. One of them
can be taken to satisfy y(0) = 1 (i.e. y0 = 1 in our notation) and we will try to find it.
Taking the Laplace transform of (22.13) we need to calculate

d d 2 d 2
L[xy 00 ] = − L[y 00 ] = −



p Y − py0 − y1 = − p Y − py0 , (22.14)
dp dp dp
d d
L[xy 0 ] = − L[y 0 ] = − (pY ), (22.15)
dp dp
d dY
L[xy] = − L[y] = − . (22.16)
dp dp

Putting all these together gives

dY dY p dp
(p2 + 1) = −pY ⇒ =− 2 ,
dp Y p +1

which can be integrated to give

Y = k (p2 + 1)−1/2 , (22.17)

where k is a constant. The transform (22.17) is not one that we recognise! You will
learn contour integration techniques that can deal with such inverses (they include branch
cuts), but for our purposes we wish to use elementary techniques to get a solution. One
way to do this is to re-write (22.17) in the form

1 −1/2
 
k
Y = 1+ 2 , (22.18)
p p

and use the Binomial theorem to expand in powers of 1/p2 to find

∞
(−1)n (2n)!
 
k 1 1 1 3 1 1 X
Y (p) = 1 − · 2 + · · · 4 − ··· = k . (22.19)
p 2 p 2 2 2! p 22n (n!)2 p2n+1
n=0

We have shown that L[xn ] = pn+1n!

(and hence L[x2n ] = p(2n)!
2n+1 ), hence we can proceed

formally and take the inverse transform of (22.19) term-by-term. The result is
∞
(−1)n 2n x2 x4 x6
X  
y(x) = k x = k 1 − 2 + 2 2 − 3 2 2 + ··· . (22.20)
22n (n!)2 2 4 ·2 4 ·3 ·2
n=0
1
Bessel’s equation of order α is x2 y 00 + xy 0 + (x2 − α2 )y = 0, and clearly α = 0 gives (22.13). Bessel
functions arise in the theory of oscillations of hanging chains (Daniel Bernoulli), vibrations of circular
membranes (Euler), planetary motion (Bessel), as well as numerous modern applications such as wave
propagation, elasticity, fluid dynamics, potential theory and diffusion, as well in some pure mathematics
problems.
22.3. DERIVATIVES AND INTEGRALS OF LAPLACE TRANSFORMS 195

Since y(0) = 1 we have k = 1 and so we have found the power series of the zeroth order
Bessel function of the first kind, namely,
x2 x4 x6
 
J0 (x) = 1 − 2 + 2 2 − 3 2 2 + · · · . (22.21)
2 4 ·2 4 ·3 ·2
The function J0 (x) is plotted in figure 22.1. It is seen that it oscillates infinitely often
and decays to zero for large x. The zeros of the function form an infinite sequence
{z1 , z2 , z3 , · · · } and it can be shown that limn→∞ (zn+1 − zn ) = π, i.e. the difference
between consecutive zeros tends to π. This is similar to sin x and cos x, but in addition
we have a decay of J0 (x) as x → ∞. The radius of convergence of the power series
(22.19) and (22.21) are left as an exercise - see Problem 1 below.

0.5
J 0(x)

-0.5
0 20 40 60 80 100
x
Figure 22.1: Zero order Bessel function of the first kind.

A final remark is that the solution (22.21) can be found using a power series solution
without going in Laplace space. The alternative calculation presented here is more
straightforward algebraically but is of course fully equivalent.
Problems
1. Find the radius of convergence of the function Y (p) in equation (22.19) and of the
function J0 (x) in equation (22.21).
2. This problem is concerned with showing the result
  Z ∞
f (x)
L = F (p0 )dp0 , F (p) = L[f (x)], (22.22)
x p
196CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

as long as the integral on the right exists.

Do this as follows. Let L[f (x)/x] = F(p). Next, calculate dF
dp using the definition
dF
of L[f (x)/x] to show that dp = −F (p). Integrate and justify your choice of the
integration constant and any assumptions that are needed to prove the required
result.
3. Use the result (22.22) to show that
Z ∞
sin x π
dx = .
0 x 2
R∞ R∞
4. Show formally that 0 J0 (x)dx = 1. [Hint: Use the result L[f (x)/x] = p F (p)dp.]

5. If f (x) is periodic of period a (i.e. f (x + a) = f (x)), show that F (p) = L[f (x)] is
given by Z a
1
F (p) = e−px f (x)dx.
1 − e−ap 0
Find F (p) is f (x) = 1 in the intervals 0 to 1, 2 to 3, 4 to 5, etc., and f (x) = 0 in
the remaining intervals, i.e. 1 to 2, 3 to 4, etc.

22.4 Convolution theorem

The convolution theorem for Laplace transforms states
Z x 
L f (x − t)g(t)dt = F (p)G(p). (22.23)
0

This is very useful as we will see below, because given a product of Rtwo recognisable
x
Laplace transforms, their inverse transform is the convolution integral 0 f (x − t)g(t)dt.
And vice vera.
We want to prove this result. Start with
Z ∞  Z ∞ 
−ps −pt
F (p)G(p) = e f (s)ds e g(t)dt
0 0
Z ∞ Z ∞ 
−p(s+t)
= e f (s)ds g(t)dt. (22.24)
0 0

The integration is in the first quadrant of the st−plane, i.e. s ≥ 0, t ≥ 0. We make a

change of variables by letting s + t = x in order to eliminate s, and now the integral
(22.24) becomes
Z ∞ Z ∞  Z ∞Z ∞
−px
F (p)G(p) = e f (x − t)dx g(t)dt = e−px f (x − t)g(t)dx dt.
0 t 0 t
(22.25)
Looking at the integral (22.25) we notice that the x−integration is done first in the range
t ≤ x ≤ ∞, followed by the t−integration in the range 0 ≤ t ≤ ∞. We complete the
proof by changing the order of integration, i.e. we do the t integration first followed by
the x integration. The integral is done over the triangular region x ≥ t, t ≥ 0 of the
22.5. THE BRACHISTOCHRONE 197

xt−plane. Hence for a given x doing the t integration first we have a range of integration
0 ≤ t ≤ x, while the x integration has a range x ≥ 0. The result follows since (22.25)
becomes
Z ∞ Z x  Z x 
F (p)G(p) = e−px f (x − t)g(t)dt dx = L f (x − t)g(t)dt . (22.26)
0 0 0

The result (22.23) is useful in solving integral equations. These are equations that
involve the unknown as well as its weighted integral. Here is a fairly general example of
an integral equation of the first kind. Given functions f (x) and K(x) we need to find
y(x) that satisfies Z x
y(x) + K(x − t)y(t)dt = f (x). (22.27)
0
In most integral equations arising from physical problems an added complication is that
K(x) is singular at x = 0 - we will see such an example later. Note also that it makes
no sense to differentiate (22.27) with respect to x since the integral term will not go
away2 (even assuming that K(0) < ∞). However, the convolution theorem proved
above becomes a useful tool to find solutions as the following example illustrates.
Solved Example We will find the solution of the integral equation
Z x
3
y(x) = x + sin(x − t) y(t) dt. (22.28)
0

Take the Laplace transform using the result (22.23). We find

3! 1 3! 3!
Y (p) = 4
+ Y (p) ⇒ Y (p) = + 6 (22.29)
p 1 + p2 p4 p

Taking the inverse Laplace transform is straightforward and yields

3! 5
y(x) = x3 + x . (22.30)
5!
The answer can be verified by plugging into (22.28).

22.5 The brachistochrone

The statement of the brachistochrone is the following. Imagine a point A connected by
a wire to a lower point B. A bead is allowed to slide under gravity and without friction
from A to B. What should the smooth shape of the wire be so that the time it takes for
the bead to reach B is the smallest it can possibly be?3
I will present Bernoulli’s solution using the optics analogy. Consider a ray of light
going trough a layered medium that has stratified density, i.e. it is made of horizontal
layers of different density. Physics tells us that the speed of light will change according
2
As an exercise, assume that K(0) 6= 0 and differentiate (22.27) with respect to x to convince
yourselves that nothing is gained by doing so.
3
The answer is not a straight line; intuitively we can see why since a vertical fall is faster than a
constrained one. It is also not the arc of a circle - Galileo thought it was but was wrong! Johann
Bernoulli solved it using a beautiful analogy with optics and Snell’s law.
198CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

to the material it is travelling through (e.g. water, air, glass etc.). Consider initially,
then, a ray of light entering one layer of thickness a at a point A1 , travelling along a
straight line (since the layer properties are constant) with speed v1 to reach point A2 ,
from where it enters a second layer of thickness b with different properties and travelling
along a straight line with speed v2 to reach the layer’s bottom surface at a point A3 .
The constant speeds v1 and v2 are different. A schematic of this scenario is given in
figure 22.2.

As

r
-

·
e e
a

I
⑧
M
layer 2 ix2
~ ~
!
⑭ I As
P
i
-
-
i

Figure 22.2: Schematic of Snell’s law for two layers.

Next we calculate the time t it takes for the light to travel from A1 to A3 . Given the
geometry in figure 22.2 we have
√ p
a2 + x2 b2 + (c − x)2
t= + . (22.31)
v1 v2
Now we impose Fermat’s principle of least time which says that when light travels from
one point to another in layered media, it picks the path that gives the least time of travel.
For expression (22.31) this requires us to find where the point A2 should be to minimise
travel time, i.e. we need to find x and for a minimum we need to solve dt/dx = 0. This
gives
x c−x
√ = p , (22.32)
v1 a2 + x2 v2 b2 + (c − x)2
Inspection of the triangles A1 A2 P1 and A2 A3 P2 and trigonometry shows that equation
(22.32) is
sin α1 sin α2
= . (22.33)
v1 v2
This is Snell’s law named after Willebrod Snell (1591-1626) who was a Dutch mathe-
matician and astronomer.
22.5. THE BRACHISTOCHRONE 199

In a medium that has continuously variable density we use calculus to split it into
layers of smaller and smaller thickness. If we have n layers then Snell’s law says that

sin α1 sin α2 sin αn

= = ··· = , (22.34)
v1 v2 vn

and in the limit as n → ∞ we have a continuous ray path. At any point on the path the
tangent is at an angle α to the vertical and the speed is v such that

sin α
= const. (22.35)
v

This limiting process is shown in figure 22.3.

n
VI

limit
!
"
y a

-
-

" i wo ⑧
I
-

!
V

Figure 22.3: Schematic of Snell’s law for n layers and the limit n → ∞.

Back to the brachistochrone. Introduce a Cartesian coordinate system with origin at

the starting point A and y measured downwards. A schematic is given in figure 22.4 and
we assume that the path is such that the bead descends in the shortest possible time.
By the arguments from Snell’s law we have

sin α
= const., (22.36)
v

but we can also find v from the principle of conservation of energy: the bead has de-
scended a distance y and assuming it started from rest (and since there are no frictional
forces) we have
1
mgy = mv 2
p
⇒ v = 2gy. (22.37)
2
200CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

A
⑧ >x

m
P

·
⑰
y~ ⑧
B

Figure 22.4: Schematic of the brachistochrone.

π
Looking at figure 22.4 we know that the tangent to the curve makes an angle 2 −α
to the horizontal, and hence

cos α 1 1
y 0 = tan(π/2 − α) = i.e. tan α = ⇒ sin α = p . (22.38)
sin α y0 1 + (y 0 )2

Substituting (22.37) and (22.38) into (22.36) provides a single equation for y, namely

y 1 + (y 0 )2 = k,


(22.39)

k−y
where k is a constant. To solve (22.39) we note that (y 0 )2 = y and separating variables
we have √
y
√ dy = dx. (22.40)
k−y
Change variables by introducing
√
y
√ = tan φ, (22.41)
k−y

and solving we find that y = k sin2 φ, hence dy = 2k sin φ cos φ dφ, casting (22.40) into

dx = tan φ dy = 2k sin2 φ dφ = k(1 − cos 2φ)dφ. (22.42)

Integrate to find  
sin 2φ
x=k φ− + `. (22.43)
2
22.6. ABEL’S MECHANICAL PROBLEM AND ABEL’S INTEGRAL EQUATION201

The curve passes through the origin x = y = 0 by construction, hence x = 0 when φ = 0

is the right condition. Hence, ` = 0 in (22.43) and the solution can be written as

k k
x= (2φ − sin 2φ), y = k sin2 φ = (1 − cos 2φ). (22.44)
2 2

Introducing a = k/2 and θ = 2φ gives the parametric equations for a cycloid

x = a(θ − sin θ), y = a(1 − cos θ). (22.45)

We have encountered the cycloid in a different context in Section 21.3 where it was
constructed as the locus of a point on the circumference of a circle that is rolling in
a straight line. The connection to the brachistochrone is that given the point B, the
constant a is chosen to obtain a cycloid (its first arch) that passes through B. This can
always be done.

22.6 Abel’s mechanical problem and Abel’s integral equa-

tion

In section 22.5 we asked the question: What should the shape of a wire be for a particle to
descend a given vertical height in the smallest possible time. Abel’s mechanical problem
asks a different question: For a given height specify the time of descend (without loss
of generality to the origin y = 0), and find the shape of the wire that will achieve this.
This is a very different problem as we will see, and belongs to a class of inverse problems
that do not always have solutions.

To analyse this problem begin with defining the starting point to be (x, y) and the
end point to be the origin (0, 0). What we are given is the time T (y) and we need to
find the shape of the wire y ≡ y(x) that achieves this time of descend. Think of T (y) as
a prescribed function - it is not a constant, different y give different times.

Let any point on the curve between the starting and end points, have coordinates
(ξ, η). A schematic is given in figure 22.5. As shown in the figure we define s to be the
arc length of the curve y(x) measured from the origin (or end point).
202CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

1.8

1.6

1.4

1.2

1 (x,y)
y

o
0.8

0.6
( , )
0.4 o

0.2 s

0
0 0.5 1 1.5 2 2.5 3 3.5
x
Figure 22.5: Schematic of Abel’s mechanical problem.

Use conservation of energy. Since the particle of mass m, say, starts from rest, when
it gets to the point (ξ, η) it will have a kinetic energy equal to the potential energy lost
by the descend. This implies
 2
1 ds
m = mg(y − η), (22.46)
2 dt
and on taking a square root and picking the minus sign (since the particle is moving
from left to right) we have
ds p ds
= − 2g(y − η) ⇒ dt = − p . (22.47)
dt 2g(y − η)
We need to integrate this equation between 0 and y; the integration variable is ξ. Hence
Z y Z y ds dη
ds 1 dη
T (y) = p =√ √ . (22.48)
0 2g(y − η) 2g 0 y−η

The next step is to express ds/dη in terms of the shape

p of the wire, i.e. the function y(x).
We know that (ds)2 = (dx)2 + (dy)2 , hence ds = dy 1 + (dx/dy)2 , and integrating this
gives s
dx 0 2 0
Z y  
s(y) = 1+ (y ) dy , (22.49)
0 dy
22.6. ABEL’S MECHANICAL PROBLEM AND ABEL’S INTEGRAL EQUATION203

where y 0 is a dummy variable. Just as easily we have

s
dx 0 2 0
Z η  
s(η) = 1+ (y ) dy (22.50)
0 dy

and on differentiation we see

s  2
dx
s0 (η) := f (η) = 1+ (η) , (22.51)
dy

and equation (22.51) should be understood as defining a function f (η) or f (y) or what-
ever independent variable I choose to use.
Putting these elements together into (22.48) gives the equation
Z y
1 f (η)dη
T (y) = √ √ . (22.52)
2g 0 y−η

The easy problem is to find T (y) when y(x) is given. This immediately gives us f (η)
and the problem becomes one of carrying out a singular integral that exists.
To go the other way, that is to find y(x) given T (y) in (22.52), we use Laplace
transforms and the results we found in earlier sections. The first thing to note is that
the integral on the rhs of equation (22.52) is the convolution of the functions y −1/2 and
f (y). Taking the Laplace transform of (22.52) and using the convolution Theorem we
obtain √
1 −1/2 1 π
L[T (y)] = √ L[y ]L[f (y)] = √ √ L[f (y)], (22.53)
2g 2g p
and keeping track of what we want to calculate, namely y(x), we re-write the equation
above as r
2g √
L[f (y)] = p L[T (y)]. (22.54)
π
In principle, given the function T (y) we can find its Laplace transform and then by
inversion find f (y). With f (y) known we next need to solve the nonlinear differential
equations (22.51). All this cannot be done in general - numerical methods are needed.
We will solve, however, another classical problem that arises when T (y) = T0 where
T0 is a constant. What this says physically, is that we want to find the shape of the wire
so that the time taken to descend to the origin from any starting point is the same! This
curve has a name, the tautochrone. In this case (22.54) becomes
r
2g −1/2
L[f (y)] = T0 p , (22.55)
π
√
and since the inverse transform of p−1/2 is πy −1/2 (we did this calculation earlier), we
find √
f (y) = by −1/2 , where b = 2gT02 /π 2 . (22.56)
This in turn implies from (22.51) that
 2 Z  1/2
dx b b−y
1+ = ⇒ x= dy, (22.57)
dy y y
204CHAPTER 22. LAPLACE TRANSFORMS AND APPLICATIONS, THE BRACHISTOCHRONE A

where the integral has been left as indefinite for now. Substituting y = b sin2 φ turns the
integral into Z
b
x = 2b cos2 φ dφ = (2φ + sin 2φ) + c, (22.58)
2
where c is a constant. Hence the parametric form of the curve is
b b
x = (2φ + sin 2φ) + c, y = b sin2 φ = (1 − cos 2φ). (22.59)
2 2
We have chosen the curve to pass through the origin (0, 0) and this can happen when
φ = 0 and c = 0. Putting 2φ = θ and a = (b/2) = gT02 /π 2 gives the familiar cycloid
again
x = a(θ + sin θ), y = a(1 − cos θ). (22.60)
In fact in figure 22.5 I plotted half a branch of the cycloid for a = 1. Note that any
starting point in space can be found by choosing a appropriately.

Problems

1. Check the Abel integral equation (22.52) for the special case of finding the time of
descend of a particle of mass m from the point (2, 1) to the origin along a straight
wire joining the two.
√
2. If T (y) = k y where k is a constant, compute the curve of descend.
3. Start with equation (22.54) and show that it can be written as
√ Z y
2g d T (χ)dχ
f (y) = √ .
π dy 0 y−χ

Hence verify that when T (y) = T0 , a constant, we obtain (22.56).

4. Find the solutions of the following integral equations using Laplace transform
methods:
Rx
(i) y(x) = 1 − 0 (x − x0 )y(x0 )dx0 .
Rx
(ii) e−x = y(x) + 2 0 cos(x − x0 )y(x0 )dx0 .
h Rx 0
i
(iii) y(x) = ex 1 + 0 e−x y(x0 )dx0 .

5. Consider the differential equation

y 00 + a2 y = f (x), y(0) = y 0 (0) = 0,

where a is a real constant. Show using Laplace transform methods that the solution
is
1 x
Z
y(x) = f (x0 ) sin[a(x − x0 )]dx0 .
a 0
Chapter 23

Newton’s law of gravitation,

planetary motion and Kepler’s
laws

23.1 Motion in a plane under the action of an arbitrary

force
We will begin with a fairly general situation before considering planetary motion. As-
sume that a particle of mass m moves in the x − y plane due to a force F acting on it.
Our first assumption here is the motion in a plane and without loss of generality we take
it to be the usual x − y plane. Secondly, and looking ahead to planetary motion, we
assume that m moves around a point fixed at the origin where we will eventually place
the sun with mass M . For the moment, proceed with the motion of m relative to the
orgigin O.
We use a polar coordinates system (r, θ). Any point P in the plane is at a distance
r from O. The position vector of P is
r = r er , (23.1)
where er is a unit vector along the line OP . The corresponding unit vector perpendicular
to er and pointing in the increasing θ direction is denoted by eθ . A schematic of the
geometry along with the force F causing the motion, is given in figure ???. In terms of
the Cartesian unit vectors i and j we have
er = cos θ i + sin θ j, eθ = − sin θ i + cos θ j, (23.2)
and on differentiation with respect to θ we have
der deθ
= eθ , = −er . (23.3)
dθ dθ
To write down Newton’s Second Law of motion we need the velocity and acceleration
of the particle P . It is clear from the geometry that as the particle moves in the plane,
the variables r and θ and the vectors er and eθ , are functions of the time t. The velocity
is given by v = dr/dt, and with r given by (23.1) we calculate to find
der dr der dθ dr dθ dr
v=r + er = r + er = r eθ + er , (23.4)
dt dt dθ dt dt dt dt

205
206CHAPTER 23. NEWTON’S LAW OF GRAVITATION, PLANETARY MOTION AND KEPLER’S

and differentiating once more we find the acceleration a = dr/dt to be

"  2 #
d2 r
 2
d2 r

dv d θ dr dθ dθ
a = 2 = r 2 +2 eθ + 2
−r er . (23.5)
dt dt dt dt dt dt dt

The force acting on the particle can be resolved into components in the r and θ directions
as follows
F = Fr er + Fθ eθ . (23.6)
From Newton’s Second Law of motion we have

ma = F,

hence the differential equations of motion for P are given by

 2 
d θ dr dθ
m r 2 +2 = Fθ , (23.7)
dt dt dt
"  2 #
d2 r dθ
m 2
−r = Fr . (23.8)
dt dt

Equations (23.7)-(23.8) are general in the sense that the nature of F has not been
specified. Some additional physics is needed to specialise these to planetary motion as
we do next. Note also that the equations are nonlinear for the unknowns r(t) and θ(t);
if a solution can be found, then the trajectory of P under the action of the force F will
have been determined.

23.2 Special case I: Central forces

The force F introduced above is called a central force if it acts along the radial direction
alone, i.e. Fθ = 0. This in turn implies that equation (23.7) becomes

d2 θ
 
dr dθ d 2 dθ
0=r 2 +2 = r , (23.9)
dt dt dt dt dt
and on integration we have
dθ
r2 = h, (23.10)
dt
where h is a constant. If the mass m moves in the positive θ−direction (i.e. counter-
clockwise), then h > 0.
From equation (23.10) we can put Kepler’s Second Law that was based on obser-
vations on a sound mathematical footing. Kepler’s Second Law states that as a planet
moves around the sun (in our notation the origin O), the radius vector r sweeps out
equal areas in equal intervals of time. This in turn implies that when a planet is closer
to the sun it will travel faster but the precise speed is dictated by Kepler’s Second Law.
Consider, then, the point P being at the position (r, θ) at a given time t. After a
small time dt it sweeps through an angle dθ to the new position (r + dr, θ + dθ). As we
saw in Section 12.6, the swept area dA is given by
1 1 dθ 1
dA ≈ r2 dθ = r2 dt = h dt, (23.11)
2 2 dt 2
23.3. NEWTON’S LAW OF GRAVITATION - CENTRAL GRAVITATIONAL FORCES207

which immediately implies that on integration from t = t1 to t = t2 we obtain

1
A(t2 ) − A(t1 ) = h(t2 − t1 ). (23.12)
2
Equation (23.12) is exactly Kepler’s Second Law. Kepler’s First Law states that the
motion of a planet about the sun traces out an ellipse, with the sun at one of the
ellipse’s eccentricity. We will derive this mathematically also. 1

23.3 Newton’s law of gravitation - central gravitational

forces
In our quest to solve the system (23.7)-(23.8), we build gravitation into our model of
central attractive forces. According to Newton’s law of gravitation the attractive force
between two masses M and m is (i) directly proportional to the product of the two
masses, and (ii) inversely proportional to the square of the distance between them. In
our notation we write
GM m km
Fr = − 2
=− 2 , k = GM, (23.13)
r r

where G is the gravitational constant - in fact G = 6.6743 × 10−11 m3 kg−1 s−2 . In

what follows we introduce another constant k = GM and in the case of the sun where
M = 1.989 × 1030 kg we have k = 1.328 × 1020 m3 s−2 . We continue with all calculations
in dimensional variables.2
Equation (23.8) now reads
2
d2 r

dθ k
−r =− . (23.14)
dt2 dt r2

Equation (23.14) is nonlinear but we can turn it into a linear problem by introducing a
new variable z = 1/r and also using θ as the independent variable instead of t. From
(23.10) we have dθ/dt = hz 2 , and the chain rule is used to compute

dr d 1 dz 1 dz dθ dz
= (1/z) = − 2 =− 2 = −h , (23.15)
dt dt z dt z dθ dt dθ
d2 r d2 z dθ 2 2 d2z
= −h = −h z . (23.16)
dt2 dθ2 dt dθ2
1
Historical note: Observational astronomy went on for thousands of years until Kepler finally
quantifies the hidden beauty of planetary motion that had occupied so many scientists from ancient
times onwards. Kepler (1571-1630) was the assistant of the astronomer Tycho Brahe who died in 1601,
and so he had access to huge amounts of raw observational data. It took him twenty years of constant
work to extract his three elegant laws from a zoo of data. It is probably the first and largest achievement
of physics-informed data science but with no computers.
2
To get an idea we can estimate |Fr | in (23.13); the mass of the earth is m = 5.972 × 1024 kg, and
the average distance of the earth from the sun is about r = 1.5 × 1011 m, hence |Fr | ≈ 3.5 × 1022 kg m s−2
(note that 1 Newton is 1 kg m s−2 . For comparison, if you go to a depth of 99 m of water, the pressure
you feel is 10 bar which is 107 N m−2 , so |Fr | is about 1015 times larger than the force exerted by the
100 m of water on an area of 1 square meter.
208CHAPTER 23. NEWTON’S LAW OF GRAVITATION, PLANETARY MOTION AND KEPLER’S

Substituting these expressions into (23.14) leads to a cancellation of the nonlinearities

to provide
d2 z k
+ z = 2, (23.17)
dθ2 h
whose solution is
k
z = A1 sin θ + B cos θ + 2 , (23.18)
h
where A and B are additional constants.
Recall that the distance of the mass m from the origin is r = 1/z. Hence, we will
select axes so that when θ = 0 the mass is closest to the origin, or equivalently that
z is maximised at θ = 0. The function z(θ) is maximised at θ = 0 when dz/dθ = 0
and d2 z/dθ2 < 0. Using these conditions in (23.18) gives A1 = 0 and B > 0 and since
z = 1/r we have
(h2 /k)
r= . (23.19)
1 + (Bh2 /k) cos θ
On defining a positive constant ε = Bh2 /k we obtain the orbit r(θ) to be
h2 /k
r= . (23.20)
1 + ε cos θ
This is the equation of a conic section.3 The conic section is an ellipse if ε < 1, a
parabola if ε = 1 and a hyperbola if ε > 1. Since planets move around the sun (which is
at the focus) in closed orbits, then their orbits are ellipses as stated by Keppler’s First
Law. The crucial parameter ε has a physical meaning as we explain next.

23.3.1 The physical meaning of the eccentricity ε

From the expression for the velocity vector v given by (23.4), we can express the kinetic
energy of the planet by
"    2 #
2
1 2 1 2 dθ dr
KE = v|v| = m r + . (23.21)
2 2 dt dt

The next component in the energy equation is the potential energy stored by the mass
m due to its presence in the gravitational field of the sun. Now at infinity (i.e. very
far away from the sun at the focus) the potential energy is zero. To find the potential
energy at a finite distance r from the focus we need to calculate the work done to move
the mass from r to infinity. The attractive gravitational force is given Fr = −km/r2 (see
(23.13)), and so the work done against this field is the potential energy and is given by
Z ∞
km km
PE = − 2
dr = − . (23.22)
r r r
By the principle of conservation of energy E = KE + P E is a constant, therefore we
have "    2 #
2
1 dθ dr km
m r2 + − = E. (23.23)
2 dt dt r
3
The polar curve r = 1+εpεcos θ is a conic section with eccentricity ε and distance p between the focus
and the directrix. Hence in our case we have ε = Bh2 /k and p = 1/B.
23.3. NEWTON’S LAW OF GRAVITATION - CENTRAL GRAVITATIONAL FORCES209

Our objective now is to connect the energy E, a constant for a given trajectory, with the
eccentricity ε. Evaluating (23.23) at the instant of the trajectory when θ = 0 and the
planet is at its closest approach to the sun, we immediately see that dr/dt = 0 there.
We also have the fundamental expression (23.10); using this gives r2 (dθ/dt)2 = h2 /r2 ,
and hence

mh2 km (h2 /k)

− = E, where r= , (23.24)
2r2 r 1+ε

and this can be readily solved (luckily the linear term in ε cancels) to give the eccentricity

r
2h2
ε= 1+ E, (23.25)
mk 2

to be substituted into the trajectory (23.20).

We can conclude, therefore, that ε < 1 only if E < 0, and so this is the condition for
an ellipse. If E = 0 we have ε = 1 and hence a parabolic trajectory, and if E > 0 we get
a hyperbolic trajectory since ε > 1. Planets with large enough energy E would escape
the solar system.

23.3.2 The periods of revolution and Kepler’s Third Law

To find the period of a planet, that is the time it takes to complete one full elliptical
orbit, we will use Kepler’s second law that we gave as equation (23.12). If the period is
T , then (23.12) says that hT
2 = A where A is the area of the ellipse. Writing the ellipse
in Cartesian coordinates we have

x2 y 2
+ 2 = 1. (23.26)
a2 b

From analytic geometry we have the following facts connecting the eccentricity ε and
the distance c of the focus from the Cartesian origin:

c a2 − b2
c2 = a2 − b2 , ε= ⇒ ε2 = , b2 = a2 (1 − ε2 ). (23.27)
a a2

A schematic with these quantities is provided in Figure 23.1.

210CHAPTER 23. NEWTON’S LAW OF GRAVITATION, PLANETARY MOTION AND KEPLER’S

nY

-m
b

↓
--
·
O
>

·
F i

- a-

Figure 23.1: Geometry of the ellipse.

We need to express a and b in terms of h and k. From the figure we see that
1 1
a = (rmin + rmax ) = [r(0) + r(π)]. (23.28)
2 2
The quantity a is termed the mean distance of the planet. Using the solution (23.20) we
calculate
1 h2 /k h2 /k h2 h2 a2
 
a= + = 2
= , (23.29)
2 1+ε 1−ε k(1 − ε ) kb2
where the last equality follows from (23.27). Solving for b we find
h2 a
b2 = . (23.30)
k
We have shown that T = 2A/h and given that for our ellipse A = πab we have
2πab 4π 2 a2 b2 4π 2 3
T = ⇒ T2 = = a , (23.31)
h h2 k
where the last equality is due to (23.30).
In the formula (23.31) the constant k = GM is the same for all planets since it only
depends on the mass of the sun (the central mass) alone. Formula (23.31) is Kepler’s
Third Law that says that the squares of the periods of the planets are proportional to
the cubes of their mean distances.

Problems
1. Show that the speed v of a planet at any point of its orbit is given by
 
2 2 1
v=h k − .
r a
 Part VI

Miscellaneous Problems

211
 213

1. For the following functions construct specific ε−δ definitions of continuity at x = 0.

In other words given a ε you need to find δ(ε).

x for x ≥ 0
f (x) =
x2 for x < 0

x for x ≥ 0
g(x) =
|x|1/2 for x < 0

2. Consider the function

x2

x≥0
f (x) =
−x2 x<0

(a) Is f (x) a continuous function?

(b) Show that f 0 (0) exists and find its value.
(c) Define g(x) = f 0 (x), x 6= 0, and g(0) = f 0 (0). Determine whether g(x) is
differentiable or not.
(d) If instead of x2 in the definition of f (x) we had xn where n is a positive
integer. How many derivatives of f (x) would exist in this case?
3. A spherical balloon is being blown up by injecting air into it at 1 liter per second.
When its radius is 1 m, find the rate at which its area is increasing (pay attention
to the units).
4. Sand is being piled onto a conical pile at a constant rate of R cm3 /s. As the pile
grows, frictional forces between sand particles constrain the height of the pile to
be equal to the radius of its base.
(i) When the height equals 1 cm, find the rate at which it is increasing.
(ii) If the height at time t is h(t), find an explicit expression for it. What happens
to its rate of change as t becomes large? Explain physically/intuitively.
5. (a) Determine the regions of increase and decrease of the function f (x) = x3 −
2x + 1.
(b) Sketch functions for which the intermediate value theorem holds and:
(i) For a chosen y ∗ there are at most two values of x∗ .
(ii) For a chosen y ∗ I can choose an interval [a, b] to have as many x∗ as I
want. [Any guess as to what function this is?]
(iii) For a chosen y ∗ there does not exist a x∗ , i.e. Theorem 6 does not hold.

6. Find tan−1 (tan 3π

4 ) and arctan(tan 2π). (No computers/calculators!)

7. (a) Let f (x) = xa for x > 0 and a any real number. Prove that f 0 (x) exists and
is equal to axa−1 . [Hint: Write as an exponential function.]
d
(b) We know that dx log x = x1 . Show that d
dx loga x = 1
x log a .
[Hint: Let y = loga x so that x = ay .]
8. Find the equation of the tangent line at (0, log 3) to the graph of the curve defined
implicitly by ey − 3 + log(x + 1) cos y = 0.
214

9. What is wrong with the following solution using L’Hôpital’s rule:

x2 + 1 2x
lim = lim =0
x→0 x x→0 1

10. Consider the step-ladder function f (x) = 1 + [x] for x ≥ 0, where [x] denotes the
integer part of x. (For example [0.9] = 0, [2.1] = 2, etc.)
(a) Sketch f (x).
Rx
(b) Calculate and sketch F (x) = 0 f (t)dt.
R1
11. Need to calculate 0 ex dx. (Of course the answer is e − 1.)
The Upper Riemann sum is Un = ni=1 ei/n n1 .
P

Show that limn→∞ n1 ni=1 (e1/n )i = e − 1.

P
n+1
[Recall: Geometric series ni=1 ri = r−r
P
1−r .]

12. Prove that Z log 2

1
e−x cos x2 dx ≤ .
0 2

13. An inverted conical tank of base radius 6m and height 10m is full and is to be
pumped empty (from the top) by a pump that has power output of 105 joules per
hour (this is approximately 27.77 watts).
(a) What is the water level at the end of 6 minutes of pumping?
(b) How fast is the water level dropping this time?
R ∞ dx
14. Show that 0 √1+x 8
is convergent, by comparison with 1/x4 .
2
15. (a) Calculate the integral (1+xx2 )3/2 dx using (i) a trigonometric substitution, (ii)
R

integration by parts.
R π/2
(b) Calculate 0 cos7 xdx.

16. An object moves from left to right along the curve y = x3/2 at constant speed. If
the object is at (0, 0) at noon and at (1, 1) at 1:00 pm, where is it at 1:30pm?
17. The region between the graphs of sin x and x over the interval 0 ≤ x ≤ π/2, is
revolved about the y−axis. Sketch the resulting solid and find its volume.
18. Find the area of the surface of the solid formed by rotating the curve y = x2 ,
0 ≤ x ≤ 1, about (i) the x− axis, and (ii) the y-axis.

19. (i) Consider the particular case of the centre of mass of a plate of area A that
has unit density per unit area. Write down formulas for the centre of mass in
terms of double integrals.
(ii) Now take A to be the region 0 ≤ x ≤ 1, 0 ≤ y ≤ 1 (unit density also).
Compute the double integrals to show that the centre of mass is (1/2, 1/2) as
expected by intuition.
 215

(iii) Now take A as in part (ii) but assume that the density in half the square
between 0 ≤ x ≤ 1/2 is 1, and the density in the remaining half of the square
is ρ0 . Compute the centre of mass in this case.
(iv) Check that your formula in (iii) agrees with your answer in (ii) when ρ0 = 1.
What happens as ρ0 tends to 0 and ∞? Explain using physical intuition.
20. (a) Consider the functions f (x) = xm and g(x) = xn where 0 < m < n. Find a
condition satisfied by m and n that guarantees that the centre of mass of the
region between the graphs y = f (x) and y = g(x) is outside the region.
(b) We know from Example 2 from Recording 20 that m = 1, n = 2 has the
centre of mass within the region. What are the smallest integer pair m, n for
which the centre of mass is outside the region?
21. Use the results developed in Recording 21 to show that a circle of radius a has
perimeter 2πa and area πa2 .
eαn
22. (a) For what real values of α (if any at all) does the series ∞
P
n=1 n converge?
(b) Calculate ∞ 1
P
n=100 n1/100 .

(c) Let P = K
P n
PK n
n=1 3 and Q = n=1 (1/3) . Find P , Q and P Q.
Which ones converge as K → ∞? Could you have anticipated this without
calculations?
(−1)n log n
23. (a) Does the series ∞
P
n=1 n2
converge?
P∞ log n
(b) Does the series n=1 n2 converge?
(−1)n log n
(c) Is the series ∞
P
n=1 n convergent? Is it absolutely convergent?
P∞ 1
24. (a) Show that n=1 n2 +1 converges by using (i) the comparison test, (ii) the
integral test.
√
(b) Does the series ∞ 3n+ n
P
n=1 2n +2 converge or diverge?
3/2

(c) For what values of p > 0 does the series ∞ 1

P
n=2 n(log n)p converge?
P∞ 1
25. Discuss the convergence of the series n=1 n2 −log n .

(−1)n x2n+1
26. Find the radius of convergence of the power series ∞
P
n=0 2n+1 .

27. (a) Prove that ( ∞

P n
P∞ n
P∞ Pn n
n=0 an x )( n=0 bn x ) = n=0 ( m=0 am bn−m )x .
log(1+x2 )
(b) Find the power series of 1+x2
.

28. Use Taylor’s theorem to find a second order accurate formula for the finite differ-
ence approximation of f (2) (x) using information from the points f (x), f (x ± h).
29. Use Taylor series to compute 1/e to three decimal accuracy. Compare the number
of terms needed with example 3 in Recording 29. Explain.
 √ 
30. Find the first four non-zero terms in the Maclaurin expansion of log 1 + sin x .
216

31. The Chebyshev4 polynomials of the first kind are defined by Tn = cos n cos−1 x ,


where n = 0, 1, 2, . . ..
(a) Show that

T0 = 1, T1 (x) = x, T2 (x) = 2x2 − 1, T3 (x) = 4x3 − 3x

(b) Show that the Chebyshev polynomials Tn (x) are orthogonal on the interval
1
[−1, 1] with respect to the weight function w(x) = √1−x 2
.
(c) Verify that Tn (x), n = 0, 1, 2, . . ., satisfy the Chebyshev differential equation

d2 y dy
(1 − x2 ) 2
−x + n2 y = 0.
dx dx
32. Find and sketch the periodic extensions of
(a) f (x) = x2 , defined on x ∈ [−π, π]. Call this function g1 (x).
(b) f (x) = ex , defined on x ∈ [0, 1]. Call this function g2 (x).
R 7π R 21/2
(c) Calculate 0 g1 (x)dx and 1/2 g2 (x)dx.

33. Express f (x) = sin(x + π/4) + sin2 x − cos2 x as a complex trigonometric series.
34. Give examples of functions f (x) that are periodic of period 2π and are:
(i)
Smooth (i.e. infinitely differentiable).
(ii)
Piecewise continuous.
(iii)
Piecewise continuous with piecewise continuous first derivative.
(iv)Piecewise continuous with piecewise continuous first derivative but not second
derivative.
(v) As nasty a function as you can think of whose Fourier series exists.
35. Find directly the Fourier series of φ(x) = x, −π < x < π without recourse to shifts
and reflections used in class.
36. (a) For the function f (x) = ex defined on [0, π], find the cosine series and the
sine series. Call them Sc (x) and Ss (x).
(b) Discuss convergence of Sc and Ss for all values of x ∈ [−π, π].
d d
(c) Now consider dx (Sc ) and dx (Ss ). Which one (if any) converges and why?
37. Sketch the function f (x) given below and find its Fourier series:


 0 −3 ≤ x < −2
−x − 1 −2 < x ≤ 1



f (x) = 0 −1 ≤ x < 1

 1 − (x − 1)2 1<x≤2


0 2≤x≤3

4
Pafnuti Chebyshev (1821-1894), Russian mathematician. Chebyshev polynomials are used widely in
solutions of partial differential equations, including aerodynamics, biofluiddynamics, materials, waves,
...
 217

38. Suppose u(x) is a smooth 2π−periodic function defined on −π ≤ x ≤ π. Use

Parseval’s theorem to find
Z π Z π
1 1
u2x (x)dx, u2 (x)dx,
2π −π 2π −π xx
1
Rπ 2
in terms of kuk2 := 2π −π u (x)dx.
Hence show that if 0 < m < n are integers, then
π 2 π 2
dm u dn u
Z  Z 
dx ≤ dx.
−π dxm −π dxn