Sequences, Series and Taylor Approximation

(MA2712b, MA2730)

Level 2 Teaching Team

Current curator: Simon Shaw

November 20, 2015

Contents

0 Introduction, Overview 6

1 Taylor Polynomials 10
1.1 Lecture 1: Taylor Polynomials, Definition . . . . . . . . . . . . . . . . . . . . . 10
1.1.1 Reminder from Level 1 about Differentiable Functions . . . . . . . . . . 11
1.1.2 Definition of Taylor Polynomials . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Lectures 2 and 3: Taylor Polynomials, Examples . . . . . . . . . . . . . . . . . 13
1.2.1 Example: Compute and plot Tn f for f (x) = ex . . . . . . . . . . . . 13
1.2.2 Example: Find the Maclaurin polynomials of f (x) = sin x . . . . . . 14
2
1.2.3 Find the Maclaurin polynomial T11 f for f (x) = sin(x ) . . . . . . . 15
1.2.4 Questions for Chapter 6: Error Estimates . . . . . . . . . . . . . . . . 15
1.3 Lecture 4 and 5: Calculus of Taylor Polynomials . . . . . . . . . . . . . . . . . 17
1.3.1 General Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Lecture 6: Various Applications of Taylor Polynomials . . . . . . . . . . . . . 22
1.4.1 Relative Extrema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4.2 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.3 How to Calculate Complicated Taylor Polynomials? . . . . . . . . . . . 26
1.5 Exercise Sheet 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5.1 Exercise Sheet 1a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5.2 Feedback for Sheet 1a . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2 Real Sequences 40
2.1 Lecture 7: Definitions, Limit of a Sequence . . . . . . . . . . . . . . . . . . . . 40
2.1.1 Definition of a Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.1.2 Limit of a Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.1.3 Graphic Representations of Sequences . . . . . . . . . . . . . . . . . . . 43
2.2 Lecture 8: Algebra of Limits, Special Sequences . . . . . . . . . . . . . . . . . 44
2.2.1 Infinite Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

1
 2.2.2 Algebra of Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.2.3 Some Standard Convergent Sequences . . . . . . . . . . . . . . . . . . . 46
2.3 Lecture 9: Bounded and Monotone Sequences . . . . . . . . . . . . . . . . . . 48
2.3.1 Bounded Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 Convergent Sequences and Closed Bounded Intervals . . . . . . . . . . 48
2.4 Lecture 10: Monotone Sequences . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4.1 Convergence of Monotone, Bounded Sequences . . . . . . . . . . . . . . 50
2.5 Exercise Sheet 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.5.1 Exercise Sheet 2a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.5.2 Feedback for Sheet 2a . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3 A flipped classroom approach to Improper Integrals 58

3.1 Self-study for Lecture 11: Improper Integrals — Type 1 . . . . . . . . . . . . . 58
3.2 Self-study for Lecture 11: Improper Integrals — Type 2 . . . . . . . . . . . . . 61
3.3 Homework for improper integrals, Types 1 and 2 . . . . . . . . . . . . . . . . . 63
3.4 Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4.1 Feedback on in-class exercises . . . . . . . . . . . . . . . . . . . . . . . 64
3.4.2 Homework feedback for improper integrals . . . . . . . . . . . . . . . . 66

4 Real Series 78
4.1 Lecture 12: Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.1.1 A Tale of a Rabbit and a Turtle following Zeno’s . . . . . . . . . . . . 78
4.1.2 Definition of a Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1.3 Convergent Series, Geometric Series . . . . . . . . . . . . . . . . . . . . 80
4.2 Lecture 13: Important Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.1 A Criterion for Divergence . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.2 Telescopic Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.3 Harmonic Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2.4 Algebra of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.3 Lecture 14: Test for Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3.1 Comparison Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.3.2 Integral Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.4 Lecture 15: Further Tests for Convergence . . . . . . . . . . . . . . . . . . . . 92
4.4.1 Comparing a Series with a Geometric Series: the Root and Ratio Tests 92
4.5 Exercise Sheet 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5.1 Exercise Sheet 3a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5.2 Additional Exercise Sheet 3b . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5.3 Feedback for Sheet 3a . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

2
5 Deeper Results on Sequences and Series 103
5.1 Lecture 16: (ǫ, N )-Definition of Limits . . . . . . . . . . . . . . . . . . . . . . 103
5.1.1 Practical Aspects of Estimates of Convergent Sequences . . . . . . . . . 104
5.1.2 Divergent Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.2 Extra curricular material: Error Estimates from the Tests . . . . . . . . . . . . 108
5.2.1 Error Estimates from the Ratio and Root Tests . . . . . . . . . . . . . 108
5.2.2 Error Estimates for the Integral Test . . . . . . . . . . . . . . . . . . . 108
5.3 Lecture 17: Absolute Convergence of Series and the Leibnitz Criterion of Con-
vergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.3.1 Alternating Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.4 Exercise Sheet 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.4.1 Exercise Sheet 4a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.4.2 Additional Exercise Sheet 4b . . . . . . . . . . . . . . . . . . . . . . . . 117
5.4.3 Short Feedback for Exercise Sheet 4a . . . . . . . . . . . . . . . . . . . 117
5.4.4 Short Feedback for the Additional Exercise Sheet 4b . . . . . . . . . . 118
5.4.5 Feedback for Exercise Sheet 4a . . . . . . . . . . . . . . . . . . . . . . 119

6 Approximation with Taylor Polynomials 123

6.1 Lecture 18: Taylor’s theorem and error estimates . . . . . . . . . . . . . . . . 123
6.2 Taylor Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.3 Estimates Using Taylor Polynomial . . . . . . . . . . . . . . . . . . . . . . . . 128
6.3.1 How to compute e in a few decimal places? . . . . . . . . . . . . . . . . 128
6.3.2 How good is the approximation of T4 f for f (x) = cos(x)? . . . . . . 129
6.3.3 Error in the Approximation sin x ≈ x . . . . . . . . . . . . . . . . . . 130
6.4 Estimating Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.5 Extra curricular material: Estimating n for Taylor Polynomials . . . . . . . . 135
6.6 Exercise Sheet 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.6.1 Exercise Sheet 5a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.6.2 Exercise Sheet 5b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.6.3 Miscellaneous Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.6.4 Feedback for Exercise Sheet 5a . . . . . . . . . . . . . . . . . . . . . . 144

3
7 Power and Taylor Series 147
7.1 Lecture 19: About Taylor and Maclaurin Series . . . . . . . . . . . . . . . . . 147
7.1.1 Some Special Examples of Maclaurin Series . . . . . . . . . . . . . . . 147
7.1.2 Convergence Issues about Taylor Series . . . . . . . . . . . . . . . . . . 148
7.1.3 Valid Taylor Series Expansions . . . . . . . . . . . . . . . . . . . . . . 149
7.2 Lecture 20: Power Series, Radius of Convergence . . . . . . . . . . . . . . . . . 151
7.2.1 Behaviour at the Boundary of the Interval of Convergence . . . . . . . 153
7.2.2 Elementary Calculus of Taylor Series . . . . . . . . . . . . . . . . . . . 154
7.3 Lecture 21: More on Power and Taylor Series . . . . . . . . . . . . . . . . . . 155
7.3.1 The Binomial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.4 Extra curricular material: General Theorem About Taylor Series . . . . . . . . 159
7.4.1 Examples of Power Series Revisited . . . . . . . . . . . . . . . . . . . . 160
7.4.2 Leibniz’ Formulas For ln 2 and π/4 . . . . . . . . . . . . . . . . . . . . 161
7.5 Extra curricular material: Taylor Series and Fibonacci Numbers . . . . . . . . 163
7.5.1 Taylor Series in Number Theory . . . . . . . . . . . . . . . . . . . . . . 163
7.5.2 Taylor’s Formula and Fibonacci Numbers . . . . . . . . . . . . . . . . . 164
7.5.3 More about the Fibonacci Numbers . . . . . . . . . . . . . . . . . . . . 165
7.6 Extra curricular Christmas treat: Series of Functions Can Be Difficult Objects 167
7.6.1 What Can Go ‘Wrong’ with Taylor Approximation? . . . . . . . . . . . 167
7.6.2 The Day That All Chemistry Stood Still . . . . . . . . . . . . . . . . . 168
7.6.3 Series Can Define Bizarre Functions: Continuous but Nowhere Differ-
entiable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
7.7 Exercise Sheet 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.7.1 Exercise Sheet 6a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.7.2 Feedback for Exercise Sheet 6a . . . . . . . . . . . . . . . . . . . . . . 175
7.7.3 Additional Exercise Sheet 6b . . . . . . . . . . . . . . . . . . . . . . . . 180
7.7.4 Exercise Sheet 0c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.7.5 Feedback for Exercise Sheet 0c . . . . . . . . . . . . . . . . . . . . . . . 183

4
List of Figures

1.1 The Maclaurin polynomials of degree 0, 1 and 2 of ex . . . . . . . . . . . . . . 13

1.2 Maclaurin polynomials of ex and perturbations . . . . . . . . . . . . . . . . . . 14
1.3 Maclaurin polynomials of sin x . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
x3
1.4 The graph near x = 0 of g(x) = cos(x2 ) − esin x − ln(1 − x) − . . . . . . . . 24
3
(−1)n
2.1 The first few values of a sequence alternating about the limit, like 1 + n
. . 43
2.2 The first few values of a sequence converging in a random fashion to the limiting
value, like 1 + cosn n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 The first few values of a sequence increasing to 1, like (1 − n1 ). . . . . . . . . 51
2.4 The first few values of a sequence decreasing to 1, like (1 + n1 ). . . . . . . . . 51

4.1 The area under the curve is less than the area under the line y = f (k − 1) and
it is greater than the area under the line y = f (k). . . . . . . . . . . . . . . . 90

7.1 An innocent looking function with an unexpected Taylor series. . . . . . . . . 169

7.2 Example (7.10) of a continuous nowhere differentiable real function . . . . . . 170

5
Chapter 0

Introduction, Overview

The first 12 lectures (Chapters 1-3) contribute to the study blocks MA2730 (for M, FM, MSM
and MCS) and MA2712b (for MMS and MCC). Those blocks feed into the assessment blocks
MA2812 and MA2815 (for M, FM, MSM and MCS) or MA2810 (for MMS and MCC). The
purpose of those lectures is to make you familiar with important concepts in Calculus and
Analysis, namely those of sequences and series as well as Taylor1 polynomials and series.
In the first three chapters, you shall be introduced to elementary ideas about these concepts,
so you could apprehend them as well as follow and perform relevant calculations. Students
studying the full Analysis study block (MA2730) will continue, revisiting, broadening and
deepening these concepts. In particular, you will be given the means to use more formal
definitions and prove the results stated in the following first set of lectures.
Most of Calculus (MA1711) and Fundamentals of Mathematics (MA1712) may be used in this
set of lectures. For quick reference, we have put some essential background material of Level 1
in a revision section on Blackboard. You will have two lectures a week with one seminar (the
class is split in four seminar groups). The 24 lectures are split into 6 chapters; each section
of those chapters corresponding to a lecture. There will be one set of exercise sheets per
chapter, obviously most of these sheets lasting for a few weeks. In Exercise Sheet number Na,
N = 1, . . . , 6, we expect to give you some short feedback first (to check your answers), finally
rather detailed feedback. Additional exercises are given in Exercise Sheets Nb, N = 1, . . . , 6.
Hence there will only be short feedback for them.

Chapter 1 Taylor Polynomials (5 Lectures)

Taylor polynomials are approximating a given differentiable function f , say, in a neighbour-
hood of a given point x = a, say, by a polynomial of degree n, say. This means that the
notation Tna f for such polynomial looks cumbersome, but it encodes the full information we
need to know about them. Because polynomials are often easier to manipulate than more
complex functions, we can replace those by one of their appropriate Taylor polynomial
1
B. Taylor (1685-1731) was an English mathematician who worked on analysis, geometry and mechanics
(vibrating string). He introduced the ‘calculus of finite differences’. In that context, he invented integration
by parts and gave his version of what is known as Taylor’s Theorem, although the theorem was already known.
He got embroiled in the disputes between English and continental mathematicians about following Newton’s
or Leibnitz’s versions of calculus. He worked also on the foundation of descriptive and projective geometry
but did not elaborate although he gave some deep results.

6
 1. to evaluate limits,

2. to determine the type of a degenerate critical point or, even,

3. to approximate integrals of complicated functions.

In the first two operations, the result of the replacement will be an exact value, in the last it
will only be an approximation.
In the five lectures of this chapter, we shall first define Taylor polynomials, then give examples
of calculation of those Taylor polynomial. This involves the calculation of many derivatives.
You should pay attention to the Product, Quotient and Chain Rules for differentiation.
In the second lecture we shall also give a first idea about the error Rna f = f − Tna f made by
replacing a function f by one of its Taylor polynomial Tna f . In practical term, the calculation
of a Taylor polynomial of a complex function can be simplified by using calculus rules to obtain
the calculus of Taylor polynomials. To justify those calculations we need the information
we got in the second lecture about the error term. Those are the topics of the third lecture.
We end the chapter with two lectures on the application of Taylor polynomials to calculate
limits and to study the type of degenerate critical points of real functions.

Chapter 2: Real Sequences (3 Lectures)

In the second chapter, we consider sequences (of real numbers) and their limits. Sequences
are ordered countable sets a1 , a2 , . . . of real numbers. In this chapter we shall concentrate
on special sequences defined by real functions such that an = f (n), for all n ∈ N, where
f : [1, ∞) → R. This will allow us to use what you did with limits of functions at Level 1.
This is not a significant restriction because, in practice, we shall use sequences of this type. In
lecture six, the limit of such sequence will then be defined as equal to limx→∞ f (x). In lecture
seven, we use the algebra of limits of functions to determine an essentially equivalent
algebra of limits of sequences as well as establishing some important limits. Those
results are important because, when they hold, they free us to have to go back to the initial
definition. The chapter ends with lecture eight where we discuss some important qualitative
properties of sequences, in particular the fundamental theorem stating that bounded AND
monotone sequences converge.

Chapter 3: Improper Integrals (2 Lectures)

This material has been pulled through from Level 1 in order to leave more room behind.
It deals with the evaluation of definite integrals in two cases. Type 1: where the interval
of integration if infinite; and, Type 2: where the integrand becomes infinite at a vertical
asymptote.

7
Chapter 4: Real Series (4 Lectures)
The first half of the course finishes in Chapter 3 with the study of series of real numbers.
X∞
Series are infinite sums of sequences, say an . You have already seen that a geometric
n=1
series converges if and only if its common ratio is strictly smaller than 1 in modulus. In
lecture nine, we give the definition of the convergence of a series, using the convergence of
the sequence of partial sums. In the next lecture ten, we look at other important series
and show that there exists an algebra of convergent series. This leads to the discussion
of many tests to assess the convergence or divergence of series without calculating their
sum. In the last two lectures of the chapter, eleven and twelve, we discuss the Comparison,
Integral, Ratio and Root Tests, usually by comparing with known convergent or divergent
series, like the geometric or harmonic series. In this theory, it means that we establish first
the convergence or divergence of a series, then establish its exact sum (when possible), or give
estimates for it.
This finishes the first half of the block. We now move to deepen our understanding of the first
12 lectures.

Chapter 5: Additional, Deeper Results on Sequences and

Series (2 Lectures)
In this chapter we use a more general definition of limit of sequences that works for any
real sequence: the (ǫ, N)-definition. In lecture thirteen you shall learn how to use it, and show
that it is equal to our previous definition in the context of Chapter 2. In lectures fourteen
and fifteen, we use the more subtle definition to study the convergence of alternating series,
series of real numbers that converge but, with a divergent series of the modulus of the terms.
We shall also see (mainly in exercises) that such series behave in an unexpected manner.

Chapter 6: Approximation with Taylor Polynomials

(3 Lectures)
Chapter 5 is entirely devoted to the estimate of errors made by replacing functions by their
Taylor polynomials. In lecture sixteen we state and prove Taylor’s Theorem, giving an estimate
of the error term Rna f with greater precision than in Chapter 1, so that we can estimate other
error terms, in particular in lecture eighteen, where we discuss approximation of integrals. All
those estimates lead to the next, and final, chapter.

Chapter 7: Power and Taylor Series (4 Lectures)

This final chapter is about functions defined as series (with an infinite sum depending on
∞
X xn
powers of the variable x, like ). This series happens to converge to ex and its finite
n=0
n!

8
initial segments correspond to the Taylor polynomials Tn ex of ex. From the other side, as we
let n tend to ∞, what happens to Tna f (x)? How far is it from f (x)? Those questions are
studied in the last 5 lectures. In lectures twenty one and twenty two, we derive fundamental
properties shared by all power series. We finish in the last two lectures by applying those
results to Taylor series.

9
Chapter 1

Taylor Polynomials

1.1 Lecture 1: Taylor Polynomials, Definition

We start with an example.

Example. What we can do to determine the behaviour of

p(x) = 1 + x + x2 + x3

around x = 3, is to recast p(x) such that the behaviour of p near x = 3 is more transparent.
Set x = 3 + y, y = x − 3, and calculate:

1 + x + x2 + x3 = 1 + (3 + y) + (3 + y)2 + (3 + y)3
= 40 + 34y + 10y 2 + y 3
= 40 + 34(x − 3) + 10(x − 3)2 + (x − 3)3. (1.1)

This last expression allows us to see more clearly what happens when we are close to x = 3,
that is, when |y| = |x − 3| < 1, because then the powers of x − 3 decrease in value.
There is a question: the final form (1.1) of p involved some calculations (that we skipped).
How is it possible to avoid them? The answer is to look at the derivatives of p at
x = 3. We have

f ′ (x) = 1 + 2x + 3x2 , f ′′ (x) = 2 + 6x and f ′′′ (x) = 6.

Therefore, f (3) = 40, f ′ (3) = 34, f ′′ (3) = 20, f ′′′ (3) = 6, and thus
20 6
p(x) = 40 + 34(x − 3) + (x − 3)2 + (x − 3)3 = 40 + 34(x − 3) + 10(x − 3)2 + (x − 3)3.
2! 3!
Both sides in the above equation are third degree polynomials, and their derivatives of order
0, 1, 2 and 3 are the same at x = 3, so they must be the same polynomial.

We shall generalise that idea to any function f : [a, b] → R. Later in MA2712 and MA2715,
you shall see that those ideas can also be generalised to any function of several variables.

10
1.1.1 Reminder from Level 1 about Differentiable Functions
But first some reminder of Level 1 material about calculus, in particular differentiable func-
tions. A real valued function f in one dimension is a triple f : I ⊆ R → J ⊆ R where

1. I is a subset of R, called the domain of f , often denoted by D(f ),

2. J is a subset of R, called the co-domain of f ,
3. f (x) is the rule defining the value f (x) ∈ J for every x ∈ I.

Remark 1.1. When the domain and co-domains are clear from the context, we often use only
the rule f (x) to define f : I → J, x 7→ f (x). An example is f (x) = sin(x) or f (x) = ex when
I = J = R (for instance). This is not precise, but often convenient.

The first, second or third derivatives of f are denoted by f ′ , f ′′ or f ′′′ . For higher derivatives,
we use the notation f (k) to mean the k-th derivative of f . By definition, f (0) = f . For
k ∈ N, the notation k! means k-factorial, which by definition is
k! = 1 · 2 . . . · (k − 1) · k = Πkj=1 j.
Recall that we had also defined 0! = 1! = 1. The following two theorems are fundamental.
Theorem 1.2 (Rolle’s1 Theorem). Let f : [a, b] → R be continuous on [a, b] and differentiable
on (a, b) such that f (a) = f (b) = 0, then there exists a < c < b such that f ′ (c) = 0.

The theoretical underpinning for these facts about Taylor polynomials is the Mean Value
Theorem, which itself depends upon some fairly subtle properties of the real numbers.
Theorem 1.3 (Mean Value Theorem). Let f : [a, b] → R be continuous on [a, b] and dif-
ferentiable on (a, b). For every x ∈ (a, b), there exists a < ξ < x, ξ dependent on x, such
that
f (x) − f (a)
f ′ (ξ) = .
x−a

1.1.2 Definition of Taylor Polynomials

Suppose you need to do some computation with a complicated function f , and suppose that
the only values of x you care about are close to some constant x = a. Since polynomials are
simpler than most other functions, you could then look for a polynomial p which somehow
‘matches’ your function f for values of x close to a. And you could then replace your function
f with the polynomial p, hoping that the error you make is not too big.
Which polynomial you will choose depends on when you think a polynomial ‘matches’ a
function. In this chapter we will say that a polynomial p of degree n matches a function
f at x = a if p has the same value and the same derivatives of order 1, 2, . . . , n at x = a as
the function f . The polynomial which matches a given function at some point x = a is given
in the following formula.
1
M. Rolle (1652-1719) was a French mathematician. Largely self-taught, he worked on analysis, geometry √
and arithmetic. He is best remembered for the theorem bearing his name. He also invented the notation n x
and adopted the notion that if a > b then −b > −a in opposition to various senior mathematicians of his time.

11
Definition 1.4. Let I ⊆ R be an open interval, f : I → R be a n-times differentiable function
at a ∈ I. The Taylor polynomial Tna f of degree n at a point a of f is the polynomial
f ′′ (a) f (n) (a)
Tna f (x) = f (a) + f ′ (a)(x − a) + (x − a)2 + · · · + (x − a)n. (1.2)
2! n!
In the following lemma we show that Tna f is indeed the polynomial we seek.
Theorem 1.5. Let I ⊆ R be an open interval, f : I → R be a n-times differentiable function
at a ∈ I. The Taylor polynomial Tna f is the only polynomial p of degree n for which
p(a) = f (a), p′ (a) = f ′ (a), p′′ (a) = f ′′ (a), ..., p(n) (a) = f (n) (a)
holds.

Proof. The result depends from a lemma about polynomial that will be proved in the last
exercise in Sheet 1a. Namely, given any polynomial p of degree n, say, and a ∈ R, there exist
ai ∈ R, 0 ≤ i ≤ n, such that
n
X
p(x) = ai (x − a)i = a0 + a1 (x − a) + . . . + an (x − a)n. (1.3)
i=0

Clearly, taking the derivatives of p in the shape of (1.3) and evaluating at x = a, we find
p(a) = a0 , p′ (0) = a1 , p′′ (a) = 2a2 , p(3) (a) = 2 · 3 a3 ...
and, in general,
p(k) (a) = k! ak , 0 ≤ k ≤ n.
Therefore, p is unique and its coefficients in (1.3) are given by
f (k) (a)
ak = , 0 ≤ k ≤ n.
k!

Remark 1.6. Note that the zeroth order Taylor polynomial is just a constant,
T0a f (x) = f (a),
while the first order Taylor polynomial is
T1a f (x) = f (a) + f ′ (a)(x − a).
This is exactly the linear approximation of f for x close to a which was derived in Level 1.
Taylor polynomial generalizes this first order approximation by providing ‘higher
order approximations’ to f .

The particular case when a = 0 is an important one, so it has a special name.

Definition 1.7. When a = 0, Tn0 f is called the Maclaurin2 polynomial of degree n of f ,
and is denoted by Tn f (for simplicity).

2
C. Maclaurin (1698-1746) was a Scottish mathematician. He got his degree at 14 years of age. He spread
Newton’s theory of fluxions (Newton’s version of calculus), developed the integral test for series and worked
on geometry. He is the founder of actuarial studies.

12
1.2 Lectures 2 and 3: Taylor Polynomials, Examples
We look here at a few examples of Taylor and Maclaurin polynomials.

1.2.1 Example: Compute and plot Tn f for f (x) = ex

All the derivatives of f are identical, f (x) = f (n) (x), ∀n ∈ N, so that f (n) (0) = 1. Therefore
the first three Taylor polynomials of ex at a = 0 (or simply Maclaurin polynomials of ex ) are

T0 f (x) = 1,
T1 f (x) = 1 + x,
1
T2 f (x) = 1 + x + x2.
2

y = f (x) y = f (x) y = f (x)

y = T1 f (x)

y = T0 f (x) y = T2 f (x)

Figure 1.1: The Maclaurin polynomials of degree 0, 1 and 2 of ex

Contemplating the graphs in Figure 1.1, the following comments are of order:

1. The zeroth order Taylor polynomial T0 f has the right value at x = 0 but it does not
know whether or not the function f is increasing at x = 0. T0 f (x) is close to ex for
small x, by virtue of the continuity of ex, because ex does not change very much if x
stays close to x = 0.

2. T1 f (x) = 1 + x corresponds to the tangent line to the graph of y = ex at x = 0, and so it

also captures the fact that the function f is increasing near x = 0, but it does not see if
the graph of f is curved up or down at x = 0. Clearly T1 f (x) is a better approximation
to ex than T0 f (x).

3. The graphs of both T0 f and T1 f are straight lines, while the graph of y = ex is curved
(in fact, convex). The graph of T2 f is a parabola and, since it has the same first and
second derivative at x = 0, it captures this convexity. It has the right curvature at
x = 0. So it seems that
y = T2 f (x) = 1 + x + x2 /2
is an approximation to y = ex which beats both T0 f and T1 f .

In Figure 1.2, the top edge of the shaded region is the graph of y = ex. The graphs are of the
functions y = 1 + x + Cx2 for various values of C. These graphs all are tangent at x = 0, but
one of the parabolas matches the graph of y = ex better than any of the others.

13
 y = 1 + x + x2

y = 1 + x + 32 x2

y = 1 + x + 21 x2

x
y=e
x
+
1
=

y = 1 + x − 12 x2
y

Figure 1.2: Maclaurin polynomials of ex and perturbations

1.2.2 Example: Find the Maclaurin polynomials of f (x) = sin x

When you start computing the derivatives of sin x you find

f (x) = sin x, f ′ (x) = cos x, f ′′ (x) = − sin x, f (3) (x) = − cos x, (1.4)

and thus
f (4) (x) = sin x.
So, after four derivatives, you are back to where you started, and the sequence of derivatives
of sin x cycles through the pattern (1.4). At x = 0, you then get the following values for the
derivatives

f (4j) (0) = f (4j) (0) = 0, f (4j+1) (0) = −f (4j+3) (0) = 1, j ∈ N.

This gives the following Maclaurin polynomials

T0 f (x) = 0,
T1 f (x) = x,
T2 f (x) = T1 f (x),
x3
T3 f (x) = x − = T4 f (x),
3!
x3 x5
T5 f (x) = x − + = T6 f (x) . . .
3! 5!

Note that the Maclaurin polynomial Tn f of any function is a polynomial of degree at most n,
and this example shows that the degree can sometimes be strictly less.

14
 T1 f (x) T5 f (x) T9 f (x)

y = sin x
π 2π
−2π −π

T3 f (x) T7 f (x) T11 f (x)

Figure 1.3: Maclaurin polynomials of sin x

1.2.3 Find the Maclaurin polynomial T11 f for f (x) = sin(x2 )

Note, if we tried to do this in the same way as our other examples it would be lengthy. For the
first derivative we would need the Chain Rule, after that we would need the Product Rule and
we would get two terms. After that we would need the Chain and Product Rules again and
we would need three terms (though, after some algebra, it combined into two again). After
that, it just keeps getting worse, though, most of the terms vanish at x = 0. We would find
that
1 1
T11 f (x) = x2 − x6 + x10,
3! 5!
2
basically replacing x by x into the Maclaurin polynomial of degree 5 of sin(x):

T11 f (x) = T10 f (x) = T5 sin x(x2 ).

We shall see in Lecture 1.2 that this is to be expected and will be a useful way not to have to
calculate an enormous number of derivatives.

1.2.4 Questions for Chapter 6: Error Estimates

In this subsection we hint at some issues we have ignored but shall come back to in Chapter
6 when we will have the technical tools to approach it.
We have seen how Tna f approximate f and its derivatives at x = a. What about in an
interval around x = a? Many mathematical operations need the values of a function on an
interval, not only at one given point x = a. Having a polynomial approximation that works all
along an interval is much more substantive than an evaluation at a single point. The Taylor
polynomial Tna f (x) is almost never exactly equal to f (x), but often it is a good approximation,
especially if |x − a| is small.
Given an interval I around a, a tolerance ǫ > 0 and the order n of the approximation, here
are the big questions:

1. Given Tna f . Within what tolerance does Tna f approximate f on I?

2. Given Tna f and ǫ. On how large an interval I ∋ a does Tna f achieve that tolerance?

3. Given f , x = a, I ∋ a and ǫ. Find how many terms n must be used in Tna f to

approximate f to within ǫ on I.

15
To be able to be precise about those statements we shall introduce the following idea.

Definition 1.8. Let I ⊆ R be an open interval, a ∈ I and f : I → R be a n-times differentiable

function, then
Ran f (x) = f (x) − Tna f (x) (1.5)
is called the n-th order remainder (or error) term of the Taylor polynomial Tna f of f .

In general we shall see that there are formula for the n-th order remainder we can estimate to
give answers to the previous questions. But this is for the future. At present we concentrate to
the calculations and some use of the Taylor polynomials without detailed justification. They
are for a revisit later.

Proposition 1.9 (Peano3 ’s Form). Let I ⊆ R be an open interval, a ∈ I and f : I → R be a

n-times differentiable function, then there exists a neighbourhood of a ∈ J ⊆ I and a function
hn : J → R such that the error term Rna f satisfies

Rna f (x) = hn (x) (x − a)n (1.6)

with limx→a hn (x) = 0.

Proof. The proof is not really difficult. We leave it to the exercise sheet. It boils down to
using l’Hôpital Rule n-times on

Rna f (x)
hn (x) = , x 6= a.
(x − a)n

3
G. Peano (1858-1932) was an Italian analyst and a founder of mathematical logic. He worked on differential
equations, set theory and axiomatic. Peano had a great skill in seeing that theorems were incorrect by spotting
exceptions. At the end of his life he became interested in universal languages, both between humans and for
teaching and working in mathematics.

16
1.3 Lecture 4 and 5: Calculus of Taylor Polynomials
The obvious question to ask about Taylor polynomials is ‘What are the first so-many terms in
the Taylor polynomial of some function expanded at some point?’. The most straightforward
way to deal with this is just to do what is indicated by the formula (1.2): take however high
order derivatives you need and plug in x = a. However, very often this is not at all the most
efficient method. Especially in a situation where we are interested in a composite function f
of the form f (xn ) or, more generally, f (p(x)), where p is a polynomial, even f (g(x)), where g
is a function, all with ‘familiar’ Taylor polynomial expansions.
The fundamental principles of those calculations follow from the following results. Recall that
the degree of a polynomial is the degree of the highest non zero monomial appearing in
p. It is denoted deg(p).
Lemma 1.10. 1. Let p and q two polynomials. Then deg(p · q) = deg(p) + deg(q) and
deg(p(q)) = deg(p) · deg(q).
2. When the Taylor polynomial Tna f of a function f is in the form (1.2), we obtain easily
Tka f , k ≤ n, by ignoring the terms of powers (x − a)l, with l > k, in Tna f .

The proof of this lemma will be provided in Chapter 6, but you can attempt it, it is not
difficult.

1.3.1 General Results

Using Lemma 1.10, we can show how to get the Taylor polynomials of composed functions.
Theorem 1.11. 1. The Taylor polynomial of the sum h of two functions f and g is given
by the sum of their Taylor polynomials:
Tna h = Tna f + Tna g. (1.7)

2. The Taylor polynomial of the product of two functions f and g is given by the terms of
degree up to n of the product of their Taylor polynomials. Namely, the terms of degree
up to n of:
Tna h = Tna (Tna f · Tna g). (1.8)

Examples.

1. Consider f (x) = ex and g(x) = sin(x). To illustrate (1.7) and (1.8), we calculate the
Maclaurin polynomials of order 3 of h1 (x) = f (x) + g(x) and of h2 (x) = ex sin(x). For
h1 ,
x2 x3 x3 x2
   
T3 h1 (x) = T3 f (x) + T3 g(x) = 1 + x + + + x− = 1 + 2x + .
2 3! 3! 2

The idea from (1.8) is just that you multiply the polynomials for ex and sin(x). And so,
x2 x3 x3
 
1 1 1
T3 f (x) = T3 (1 + x + + )(x − ) = x+x2 +(− + )x3 = x+x2 + x3. (1.9)
2 3! 3! 3! 2 3

17
 2. As a more complicated example of (1.8), let f (x) = e2x /(1 + 3x). We exploit our
knowledge of the Maclaurin polynomial of both factors e2x and 1/(1+x). As an example,
what is T3 f ? Let
2x 22 x2 23 x3 4
T3 e = 1 + 2x + + = 1 + 2x + 2x2 + x3,
2! 3! 3
and
1
T3 ( ) = 1 − 3x + 9x2 − 27x3.
1 + 3x
Then multiply these two
41 3
T3 f (x) = 1 − x + 5x2 − x.
3

The composition of functions is more subtle.

Theorem 1.12. The Taylor polynomial of the composition h of two functions f and g, h(x) =
f (g(x)) is given by the terms of degree up to n of the composition of their Taylor polynomials.
Namely,
Tna h = Tna Tng(a) f (Tna g) .


(1.10)

Remark 1.13. Pay attention to the fact that, with the composition of function, we need the
Taylor polynomial of the second function f about the value of the first one g(a). In the
expression h(x) = f (g(x)) we apply first g then f . Remark that we often use, in Theorem
1.12, functions g such that g(0) = 0 so that (1.10) becomes simply
Tn h = Tn (Tn f (Tn g)). (1.11)

Example. As a first example of (1.10), let f (x) = 1/(1 + x2 ). What is T12 f ?

The direct approach is how NOT to compute it. Diligently computing derivatives one by one,
you find
1
f (x) = so f (0) = 1,
1 + x2
−2x
f ′ (x) = so f ′ (0) = 0,
(1 + x2 )2
6x2 − 2
f ′′ (x) = so f ′′ (0) = −2,
(1 + x2 )3
x − x3
f (3) (x) = 24 so f (3) (0) = 0,
(1 + x2 )4
1 − 10x2 + 5x4
f (4) (x) = 24 so f (4) (0) = 24 = 4!,
(1 + x2 )5
−3x + 10x3 − 3x5
f (5) (x) = 240 so f (4) (0) = 0,
(1 + x2 )6
(6) −1 + 21x2 − 35x4 + 7x6
f (x) = −720 so f (4) (0) = 720 = 6!.
(1 + x2 )7
..
.

18
I am getting tired of differentiating - can you find f (12) (x)? After a lot of work we give up at
the sixth derivative, and all we have found is

T6 f (x) = 1 − x2 + x4 − x6.

By the way,

1 − 78 x2 + 715 x4 − 1716 x6 + 1287 x8 − 286 x10 + 13 x12

f (12) (x) = 479001600
(1 + x2 )13

and 479001600 = 12!.

The right approach to finding T12 f is to use (1.10). We have seen that if g(t) = 1/(1 − t) then

T6 g(t) = 1 + t + t2 + t3 + t4 + t6.

Following (1.10), because t = 0 when x = 0, substitute t = −x2 in this limit, we get

T12 f (x) = 1 − x2 + x4 − x6 + · · · + x12.

Note that, in general, we get easily T2n f = T2n+1 f from Tn g as

T2n f (x) = Tn g(−x2 ) = 1 − x2 + x4 − x6 + · · · + (−1)n x2n. (1.12)

A more complicated one is (1.13) in the next section.

Example. To find the Maclaurin polynomial T4 f for f (x) = exp(sin(x)), we use

x3
T4 sin(x) = x −
6
and
x2 x3 x4
T4 exp(x) = 1 + x + + + .
2 6 24
sin(0)
Note that because sin(0) = 0, T4 exp(x) = T40 exp(x) = T4 exp(x). And so, from (1.10),

T4 f (x) = 1 + (x − x3 /6) + (x − x3 /6)2 /2 + x3 /6 + x4 /24

x2 x4
= 1+x+ − . (1.13)
2 8

Differentiation and integration are straightforward.

Theorem 1.14. 1. The Taylor polynomial of the derivative of f is given by the derivative
term per term of the Taylor polynomial, that is,

′
Tna f ′ (x) = Tn+1
a
f (x) . (1.14)

19
 2. The Taylor polynomial of the anti-derivative (primitive/integral) F of f is given by
integrating term per term the Taylor polynomial, that is,
Z
Tn+1 F (x) = C + Tna f (x) dx.
a
(1.15)

Example. To find the Maclaurin polynomials for f (x) = tan−1 (x) = arctan(x), we start with
an example we already know. Thinking about derivatives and anti-derivatives, we see that f
1
is the anti-derivative of f ′ (x) = . So, we would have a plan if we knew the Maclaurin
1 + x2
1
polynomial for . Well, we know it. So the plan is this:
1 + x2
T2n f ′ (x) = 1 − x2 + x4 − x6 + x8 + . . . + (−1)n x2n.

And so, using (1.14) and (1.15),

Z
1 − x2 + x4 − x6 + x8 + . . . + (−1)n x2n dx


T2n+1 f (x) =
x3 x5 x7 x2n+1
= C +x− + − − · · · + (−1)n .
3 5 7 2n + 1
Because f (0) = arctan(0) = 0, C = 0 and so
n
X x2k+1
k
T2n+1 f (x) = (−1) . (1.16)
k=0
2k + 1

When a = 0, Theorems 1.11, 1.12 and 1.14 lead to the following classical, useful examples
about Maclaurin polynomials.

Corollary 1.15. Let p be a polynomial such that p(0) = 0.

1. If h(x) = exp(p(x)), then

n
!
X 1
Tn h(x) = Tn (p(x))k .
k!
k=0

1
2. If h(x) = , then
1 − p(x) !
n
X
Tn h(x) = Tn (p(x))k .
k=0

3. If h(x) = sin(p(x)), then

n
!
X 1
Tn h(x) = Tn (−1)k (p(x))2k+1 .
k=0
(2k + 1)!

20
 4. If h(x) = cos(p(x)), then
n
!
X 1
Tn h(x) = Tn (−1)k (p(x))2k .
k=0
(2k)!

5. If h(x) = ln(1 + p(x)), then

n
!
X 1
Tn h(x) = Tn (−1)k+1 (p(x))k .
k
k=1

We shall illustrate these results in the next few lectures.

21
1.4 Lecture 6: Various Applications of Taylor Polyno-
mials
We shall look at two areas of use of Taylor approximations:

1. generalised criteria using derivatives to determine the type of a critical point (maximum,
minimum or saddle points),

2. generalised l’Hôpital4 Rule for limits.

1.4.1 Relative Extrema

At Level 1 you have seen that if a function f : I → R has a critical point at a ∈ I, that is, if
f ′ (a) = 0, then f has

1. a (local) maximum at x = a if f ′′ (a) < 0,

2. a (local) minimum at x = a if f ′′ (a) > 0,

3. an degenerate critical point at x = a if f ′′ (a) = 0.

In that last case, what can we say using Taylor expansions of f at x = a?

Theorem 1.16 (Relative Extrema). Let I ⊆ R be an open set, a ∈ I, and f : I ⊆ R → R be
n-times continuously differentiable at x = a, n ≥ 2. Suppose

f ′ (a) = . . . = f (n−1) (a) = 0, f (n) (a) 6= 0.

Then,

1. if n is odd, a is not a (local) extrema, it is a saddle point,

2. if n is even,

(i) a is a local maximum if f (n) (a) < 0,

(ii) a is a local minimum if f (n) (a) > 0.

Proof. For x close to a, we have

f (n) (a)
f (x) = Tna f (x) + Rna f (x) = f (a) + f ′ (a)(x − a) + . . . + (x − a)n + Rna f (x)
n!
f (n) (a)
= f (a) + (x − a)n + Rna f (x)
n!
 (n) 
f (a)
= f (a) + + hn (x) (x − a)n ,
n!
4
G.F.A. marquis de l’Hôpital (1661-1704) was a French mathematician. He wrote the first textbook of
Calculus where he gave the rule having his name.

22
where hn appears in the Peano form of the remainder.
f (n) (a) f (n) (a)
Define g1 (x) = − hn (x), and denote by g2 the constant function g2 (x) = .
n! n!
f (n) (a)
The function g1 = g1 (x) is close to the constant function g2 (x) = when x is close to
n!
a. Therefore, we can deduce that for x close to a the value g1 (x) has the same sign as the
f (n) (a)
constant .
n!
Thus, it is enough to study the local extrema of the approximation

f (n) (a)
f (a) + (x − a)n.
n!
It is now simple calculus to show that our conclusions are correct.

Examples.

1. Let f (x) = sin(x2 ) − x2 cos(x). Determine the type of the critical point x = 0.
The first non zero terms of the Maclaurin polynomials are

x6
sin(x2 ) = x2 − + ...,
6
x2 x4 x4 x6
 
2 2
x cos(x) = x 1 − + = x2 − + + ....
2 24 2 24

x4 5 6
And so, f (x) = − x + . . .. Hence, x = 0 is a (local) minimum.
2 24
2. Let g(x) = cos(x2 ) − esin x − ln(1 − x) + ax3 where a ∈ R is a parameter. Discuss the
type of the critical point x = 0 as a function of a.
The first non zero terms of the Maclaurin polynomials are

x4 x8
cos(x2 ) = 1 − + + ...,
2 24
2 3
x3 x3 x3
   
sin(x) 1 1
e = 1+ x− + x− + x− + ...
6 2 6 6 6
x2 x4
= 1+x+ − + ...,
2 8
x2 x3 x4
− ln(1 − x) = x + + + + ....
2 3 4
And so,
x4
 
1
g(x) = + a x3 − + ....
3 8
Therefore, when a 6= −1/3, x = 0 is a saddle point, but when a = −1/3, x = 0 is a
(local) maximum. This local maximum is indeed ‘illustrated’ in Figure 1.4.

23
 -0.1 -0.05 0.05 0.1
-7
-2·10
-7
-4·10
-7
-6·10
-7
-8·10
-6
-1·10
-6
-1.2·10

x3
Figure 1.4: The graph near x = 0 of g(x) = cos(x2 ) − esin x − ln(1 − x) − .
3

1.4.2 Limits
f (x)
At Level 1 you have seen l’Hôpital’s Rule. Let h(x) = and suppose that there exists an
g(x)
f ′ (a)
a such that f (a) = g(a) = 0 and g ′ (a) 6= 0, then lim h(x) = ′ . But if g ′(a) = 0 this is of
x→a g (a)
no help. We now prove an extension of that result using Taylor polynomials.
Theorem 1.17 (Generalised l’Hôpital Rule). Let I ⊆ R be an open set, a ∈ I, and f, g : I ⊆
R → R be n + 1, resp. m + 1,-times continuously differentiable at x = a. Suppose
0 = f (a) = f ′ (a) = . . . = f (n−1) (a) = 0, f (n) (a) 6= 0,
0 = g(a) = g ′ (a) = . . . = g (m−1) (a) = 0, g (m) (a) 6= 0.
f (x)
Then, h : I → R, defined by h(x) = , has the following limits as x → a:
g(x)
1. if n > m, limx→a h(x) = 0,
f (n) (a)
2. if m = n, lim h(x) = ,
x→a g (n) (a)
3. if n < m, the limit limx→a h(x) is unbounded.

Proof. For x close to a,

f (n) (a)
f (x) = f (a) + f ′ (a)(x − a) + . . . + (x − a)n + Rna f (x),
n!
f (n) (a)
 (n) 
n a f (a)
= (x − a) + Rn f (x) = + hn (x) (x − a)n,
n! n!

24
where hn appears in the Peano’s form of the remainder. Similarly for g. For x close to a,

g (m) (a)
g(x) = g(a) + g ′(a)(x − a) + . . . + (x − a)m + Rm a
g(x),
m!
g (m) (a) g (m) (a)
 
= (x − a)m + Rm
a
g(x) = + hm (x) (x − a)m.
m! m!

Now, define
f (n) (a)
!
f (x) + hn (x)
h(x) = = (x − a)n−m n!
g (m) (a)
.
g(x) + hm (x)
m!
Note that the big bracket is bounded when x → a. Now we consider our three cases:

1. when n > m, clearly limx→a h(x) = 0;

2. when n = m, h simplifies into

f (n) (a)
n!
+ hn (x)
h(x) = g (n) (a)
,
n!
+ hn (x)

and the conclusion about the limit follows;

3. when n < m, clearly limx→a h(x) = ±∞ because (x − a)n−m will blow-up.

Example. Discuss the limit as x → 0 of

sin(x2 ) − x2 cos(x)
h(x) =
cos(x2 ) − esin x + ln(1 − x) + ax3

as a function of a. Using the previous results, we get

x4 /2

0, a 6= −1/3;
lim h(x) = lim =
x→0 x→0 (1/3 + a)x3 − x4 /8 −4, a = −1/3.

Remark 1.18. In Chapter 6, with a more precise expression for the error term, we shall look
also at the use of Taylor polynomials approximations to estimate integral, for instance
Z b Z b Z b Z b
a b
f (x) dx ≈ Tn f (x) dx ≈ Tn f (x) dx ≈ Tnc f (x) dx,
a a a a

a+b
where c = . Note that those are only approximations. In Chapter 6 we shall be able
2
to estimate the maximum error of the different formulas and so determine which one is more
appropriate depending on what we want to achieve.

25
1.4.3 How to Calculate Complicated Taylor Polynomials?
We consider a few examples illustrating Corollary 1.15 and showing how to handle the figuring
out of Maclaurin polynomials of arbitrary orders (when such formulae exists). The question
is to find the Maclaurin polynomial of order n of the following functions.

Examples.

x+2
1. Let f (x) = . Simplify
3 + 2x
(x + 2) 1
f (x) = .
3 (1 + (2x)/3)
We know that   n  k
1 X
k 2x
Tn = (−1) .
(1 + (2x)/3) k=0
3
(x + 2)
Multiplying by and collecting the terms up to power xn we find
3
n
2 X 2k−1
Tn f (x) = + (−1)k k+1 xk.
3 k=1 3

1
2. Let f (x) = √ . You can use directly the Binomial Theorem (from Level 1) or
1−x
calculate the derivatives of f . We get
1 · 3 · 5 · · · (2k − 1)
f (k) (0) = .
2k
Recall that 2 · 4 · 6 · · · 2k = 2k (k!). And so by multiplying by the product of even terms,
(2k)!
top and bottom, we get f (k) (0) = k . Therefore
4 (k!)
n
X (2k)! k
Tn f (x) = k (k!)2
x.
k=0
4
√
Remark 1.19. Note that f is equal to (−2) times the derivative of g(x) = 1 − x. So,
if we know Tn g, Tn f = −2(Tn g)′ .
√ 4
3. Let 4
√ f (x) = 1 + x . We replace z = x into the Maclaurin polynomials of g(z) =
1 + z. Recall that
n
X (2k)!
Tn g(z) = 1 + (−1)k+1 z k.
4k (k!)2 (2k − 1)
k=1

And so,
⌊n/4⌋
X (2k)!
Tn f (x) = 1 + (−1)k+1 x4k.
k=1
4k (k!)2 (2k − 1)

26
 r !
1+x
4. Let f (x) = ln . We simplify
1−x

1 1
f (x) = ln(1 + x) − ln(1 − x).
2 2
Recall the Maclaurin polynomials
n
X xk
Tn ln(1 + x) = (−1)k+1 ,
k=1
k
n
X xk
Tn ln(1 − x) = − .
k=1
k

Hence,
⌊(n−1)/2⌋
X x2k+1
Tn f (x) = .
k=0
2k + 1

1
5. Let f (x) = arctan(x). Recall that f ′ (x) = . Use Theorem 1.14, integrating the
1 + x2
Maclaurin polynomial of f ′ :
⌊n/2⌋
X
′
Tn f (x) = (−1)k x2k.
k=0

We get that
⌊(n−1)/2⌋
X x2k+1
Tn f (x) = (−1)k .
k=0
2k + 1

27
Summary of Chapter 1
We have seen:

• the definition of a Taylor polynomial and the error term;

• the calculus of Taylor polynomials and its use to determine Taylor polynomials of
complicated functions using known expansions;
1
• known expansions for functions like ;
1−x
• how to use Taylor polynomials to calculate limits and determine the type of a
degenerate critical point.

28
1.5 Exercise Sheet 1

1.5.1 Exercise Sheet 1a

1. (i) Find a second order polynomial q such that q(7) = 43, q ′ (7) = 19 and q ′′ (7) = 11.
(ii) Find a third order polynomial p such that p(2) = 3, p′ (2) = 8, and p′′ (2) = −1.
How many possibilities are there?
(iii) Find all third order polynomials p satisfying p(0) = 1, p′ (0) = −3, p′′ (0) = −8 and
p′′′ (0) = 24.
√
2. Let f (x) = x + 25.

(i) Find the largest domain for the rule f (x).

(ii) Find the polynomial p(x) of degree three such that p(k) (0) = f (k) (0) for k = 0, 1, 2, 3.

3. Compute T0a f (x), T1a f (x) and T2a f (x) for the following functions.

(i) f (x) = x3 , a = 0; then for a = 1 and a = 2.

1
(ii) f (x) = , a = 1. Also do a = 2.
x
√
(iii) f (x) = x, a = 1.
(iv) f (x) = ln x, a = 1. Also do a = e2 .
√
(v) f (x) = ln( x), a = 1.
(vi) f (x) = sin(2x), a = 0, also do a = π/4.
(vii) f (x) = cos(x), a = π.
(viii) f (x) = (x − 1)2 , a = 0, and also do a = 1.
1
(ix) f (x) = x , a = 0.
e
4. Determine the Maclaurin polynomial of order 4 for the following rules:

(i) f (x) = exp(x2 + x).

1
(ii) f (x) = ln( ).
1−x
(iii) f (x) = ecos x.
(iv) f (x) = e−x cos x.

5. Find the Maclaurin polynomials Tn f of any order n for the following rules. (i) f (t) =
2
ekt, for some constant k. (ii) f (t) = e1+t. (iii) f (t) = e−t . (iv) f (t) = cos(t5 ). (v) f (t) =
1+t 1 1 et 1
. (vi) f (t) = . (vii) f (t) = . (viii) f (t) = . (ix) f (t) = √ .
1−t 1 + 2t 2−t 1−t 1−t
1
(x) f (t) = 2
(xi) f (t) = ln(1 − t2 ). (xii) f (t) = sin t cos t.
2−t−t
6. Let f : R → R be defined as f (x) = |x|3.

(i) Determine whether f has Taylor polynomials of any order n at

29
 [a] a = 0?
[b] a = 1?
[c] a = −1?
(ii) When f has Taylor polynomials, determine them (in the three cases).

7. Determine the following limits.

x(1 + cos x) − 2 tan x
(i) lim ;
x→0 2x − sin x − tan x
1 − x + ln x
(ii) lim √ ;
x→1 1 − 2x − x2
 
1 1
(iii) lim − ;
x→0 x ln(1 + x)
 
1 1
(iv) lim − ;
x→0 x (sin x)2
(v) Determine α ∈ R such that the following limit is finite:

xe− sin(x) − sin(x) + αx2

lim ,
x→0 x − sinh(x)
and calculate that limit.

8. Determine the type of critical points for the following functions.

(i) f (x) = sin(x2 ) − x2 cos(x) at x = 0;

x + ax3
(ii) f (x) = sin x − at x = 0; discuss in terms of a and b and determine the
1 + bx2
leading order term.

9. Prove Proposition 1.9.

∗
10. Show that given any polynomial p of degree n, say, and a ∈ R, there exist ai ∈ R,
0 ≤ i ≤ n, such that
n
X
p(x) = ai (x − a)i = a0 + a1 (x − a) + . . . + an (x − a)n.
i=0

Short Feedback for Sheet 1a

1. Use Taylor’s formula.

11 2 359
(i) q(x) = x − 58x + .
2 2
13
(ii) p(x) = x3 − x2 + 22x − 23. There is an infinite number determined by
2
1
p(x) = − x2 + 10x − 15 + c(x − 2)3,
2
where c ∈ R.

30
 (iii) p(x) = 4x3 − 4x2 − 3x + 1.
2. (i) [−25, ∞).
x x2 x3
(ii) 5 + − + .
10 1000 5 · 104
3. T2a f (x) is only given (think why?).
(i) T2 f (x) = 0, T21 f (x) = 1+3(x−1)+3(x−1)2 and T22 f (x) = 8+12(x−2)+6(x−2)2.
1
(ii) T21 f (x) = 1 − (x − 1) + (x − 1)2 and T22 f (x) = 2
− 41 (x − 2) + 18 (x − 2)2.
(iii) T21 f (x) = 1 + 21 (x − 1) − 18 (x − 1)2.
2
(iv) T21 f (x) = (x − 1) − 12 (x − 1)2 and T2e f (x) = 2 + e−2 (x − e2 ) − e−4
2
(x − e2 )2.
√
(v) Hint: note that ln( x) = 12 ln(x), then use the answer of (iv)!
π/4
(vi) T2 f (x) = 2x and T2 f (x) = 1 − 2(x − π/4)2.
(vii) T2π f (x) = −1 + 21 (x − π)2.
(viii) T2 f (x) = 1 − 2x + x2 and T21 f (x) = (x − 1)2.
x2
(ix) T2 f (x) = 1 − x + 2
.
3 2 7 3 25 4
4. (i) T4 f (x) = 1 + x + x + x + x.
2 6 24
z2 z3 z4
Hint: use T4 ez = 1 + z + + + in part 1. of Corollary 1.15 because
2! 3! 4!
p(x) = x2 + x satisfies p(0) = 0.
x2 x3 x4
(ii) T4 f (x) = x + + + .
2 3 4
Hint: recall ln(1/a) = − ln(a). Then use part 5. of Corollary 1.15 with p(x) = −x.
e e
(iii) T4 f (x) = e − x2 + x4.
2 6
Hint: this is the more difficult. Note cos(0) = 1, so use the Taylor polynomial of
z x2 x4
e near z = 1 with z = T4 cos x = 1 − + .
2 24
x3 x4
(iv) T4 f (x) = 1 − x + − .
3 6
Hint: look at (1.9).
5. Recall that ⌊n⌋ denotes the largest integer smaller than n. This will be useful to stop the
sums so that the powers of the variable do not exceed n. Unless indicated, use Corollary
1.15 with the right p (check, arrange that p(0) = 0 if needed).
n
X kj
(i) Tn f (t) = tj.
j=0
j!
n
X e k
(ii) Tn f (t) = t.
k!
k=0
⌊n/2⌋
X (−1)k
(iii) Tn f (t) = t2k.
k=0
k!

31
 ⌊n/10⌋
X (−1)k
(iv) Tn f (t) = t10k.
k=0
2k!
(v) Tn f (t) = 1 + 2 nk=1 tk.
P

Hint: decompose f in partial fractions.

(vi) Tn f (t) = nk=0(−2)k tk.
P
n
X 1 k
(vii) Tn f (t) = k+1
t.
k=0
2
Hint: to use part 2. of Corollary 1.15 you need to write 2 − t = 2(1 − t/2).
(viii) Tn f (t) = nk=0 (Tk exp(1)) tk.
P
n
X 2k! k
(ix) Tn f (t) = t.
k=0
(k!2k )2
Hint: for this problem, use the Binomial Theorem (see last year) or differentiate
many times and find a ‘general form’ of the derivative of f at 0.
n
(−1)k k
 
X 1
(x) Tn f (t) = 1 + k+1 t .
k=0
3 2
Hint: decompose f in partial fractions.
⌊n/2⌋
X −1
(xi) Tn f (t) = t2k.
k=0
k
⌊(n−1)/2⌋
X (−1)k 4k 2k+1
(xii) Tn f (t) = t .
k=0
2k + 1!
Hint: use a trigonometric formula.

6. (i) T2 f = 0, Tn f does not exist for n ≥ 3.

(ii) T31 f = 1 + 3(x − 1) + 3(x − 1)2 + (x − 1)3.
(iii) T3−1 f = 1 − 3(x + 1) + 3(x + 1)2 − (x + 1)3.
7. (i) 7.
(ii) -1.
(iii) −1/2.
(iv) −∞.
(v) α = 1 and the limit is −2.
8. (i) a minimum.
(ii) a saddle point anyway (think about it). When b 6= a + 1/6, the leading term is
quintic, when b = a + 1/6 and a 6= −7/60, the leading term is cubic, and, when
(a, b) = (−7/60, 1/20), the leading term is of order 7.
9. Use induction with l’Hôpital Rule.
10. Use the fact that if a polynomial p has a root at x−a, it is uniquely divisible by x−a, that
is, there exists an unique polynomial r of degree deg(p) − 1 such that p(x) = r(x)(x − a).

32
1.5.2 Feedback for Sheet 1a
1. We use Theorem 1.5.

(i) We have q(x) = 43+19(x−7)+ 11 2

(x−7)2. Expanding and re-ordering the coefficients
11 359
we get q(x) = x2 − 58x + .
2 2
(ii) Note that we have three conditions for a polynomial with four coefficients, so one
coefficient is free. Clearly, using Theorem 1.5, the first three terms of p are fixed,
and we can choose whatever we like for the cubic term. Thus we have an infinite
number of cubic polynomials
1
p(x) = 3 + 8(x − 2) − (x − 2)2 + c(x − 2)3,
2
for any non-zero constant c ∈ R. For the first part we choose c = 1 and get
13 2
p(x) = x3 − x + 22x − 23.
2

(iii) The final polynomial is clearly

p(x) = 4x3 − 4x2 − 3x + 1.

2. (i) To determine the largest domain of a rule f , you need to get the x’s such that
f (x) make sense. In this case, x + 25 must be non negative, so x + 25 ≥ 0, hence
D(f ) = [−25, ∞).
(ii) From Theorem 1.5, p is the Taylor polynomial T30 f of f . Straightforward differen-
tiation get you p(0) = f (0) = 5, p′ (0) = f ′ (0) = 1/10, p′′ (0) = f ′′ (0) = −1/500 and
p′′′ (0) = f ′′′ (0) = 6/50000. Hence

x x2 x3
p(x) = 5 + − + .
10 1000 5 · 104

3. You are being asked to provide the Taylor polynomials of f up to order 2. Note that if
you keep the Taylor polynomial is the shape (1.2), the Taylor polynomial of lower order,
say Tka f , can be found from the Taylor polynomial of higher order, say Tla f , by cutting
out all the terms (x − a)j of order k < j ≤ l, from Tla f (though, this must be at the
same a). Hence we only calculate and give the answer as T2a f (x) and in the shape of
(1.2).

(i) T2 f (x) = 0, T21 f (x) = 1+3(x−1)+3(x−1)2 and T22 f (x) = 8+12(x−2)+6(x−2)2.

1
(ii) T21 f (x) = 1 − (x − 1) + (x − 1)2 and T22 f (x) = 2
− 41 (x − 2) + 18 (x − 2)2.
(iii) T21 f (x) = 1 + 21 (x − 1) − 18 (x − 1)2.
2
(iv) T21 f (x) = (x − 1) − 12 (x − 1)2 and T2e f (x) = 2 + e−2 (x − e2 ) − e 2 (x − e2 )2.
−4

√
(v) Let g(x) = ln(x). Because ln( x) = 12 ln(x), then T21 f = 21 T21 g (think about it).
Using the answer of (iv), T21 f (x) = 21 (x − 1) − 41 (x − 1)2.
π/4
(vi) T2 f (x) = 2x and T2 f (x) = 1 − 2(x − π/4)2.

33
 (vii) T2π f (x) = −1 + 21 (x − π)2.
(viii) T2 f (x) = 1 − 2x + x2 and T21 f (x) = (x − 1)2.
x2
(ix) T2 f (x) = 1 − x + .
2
4. In the following you do not need to evaluate all powers of (x − a), only the ones up to
order 4 (because we want the Taylor polynomials of order 4).

(i) Use part 1. of Corollary 1.15 with p(x) = x + x2. Because p(0) = 0, we can
replace it into the fourth order Maclaurin polynomial of ez (around z = 0), namely
z2 z3 z4
T4 ez = 1 + z + + + . We get
2! 3! 4!
(x + x2 )2 (x + x2 )3 (x + x2 )4
 
2
T4 f (x) = T4 1 + (x + x ) + + +
2! 3! 4!
3 7 25 4
= 1 + x + x2 + x3 + x.
2 6 24

(ii) Recall that ln(1/a) = − ln(a). Use part 5. of Corollary 1.15 with p(x) = −x. Then

x2 x3 x4
T4 f (x) = x + + + .
2 3 4

(iii) Because cos(0) = 1, we need to use the Taylor polynomial of ez around z = 1, that
is,
1 z (z − 1)2 (z − 1)3 (z − 1)4
T4 e = e + e(z − 1) + e +e +e .
2 3 4
Then we replace z − 1 by the Maclaurin expansion of order 4 of

x2 x4
cos x − 1 = − + .
2 24
Therefore,
e 2 e 4
T4 f (x) = e − x + x.
2 6
(iv) Use (1.9) in Theorem 1.11. We multiply together the Maclaurin polynomials of
order 4 of e−x and cos(x), namely,
1 1 1
T4 e−x = 1 + (−x) + (−x)2 + (−x)3 + (−x)4,
2 6 24
1 2 1 4
T4 cos(x) = 1 − x + x.
2 24
Hence, collecting all terms up to order 4,

x3 x4
T4 f (x) = T4 T4 e−x · T4 cos(x) = 1 − x +


− .
3 6

5. Recall that ⌊n⌋ denotes the largest integer smaller than n. This will be useful to stop
the sums so that the powers of the variable do not exceed n.

34
 n n
X (kt)j X kj
(i) Use part 1. of Corollary 1.15 with p(t) = kt. We get Tn f (t) = = tj.
j=0
j! j=0
j!
n
X e k
(ii) Because e1+t = e · et, so Tn f (t) = t.
k=0
k!
(iii) Use part 1. of Corollary 1.15 with p(t) = −t2. Because we need to stop at degree
n = 2k, we only need to add the terms up to k = ⌊n/2⌋. Finally, we get Tn f (t) =
⌊n/2⌋
X (−1)k
t2k.
k=0
k!
(iv) Use part 4. of Corollary 1.15 with p(t) = t5. Because we need to stop at degree
⌊n/10⌋
X (−1)k
n = 10k, we get Tn f (t) = t10k.
2k!
k=0
1+t 2
(v) Decompose f in partial fractions: f (t) = 1−t
= −1 + 1−t
. Using the GP formula,
you get Tn f (t) = 1 + 2 nk=1 tk.
P
Pn k k
(vi) Substitute u = −2t in the GP formula for 1/(1−u). You get Tn f (t) = k=0 (−2) t.
(vii) To use part 2. of Corollary 1.15, you need to write 2 − t = 2(1 − t/2). And so,
n  k n
1X t X 1 k
Tn f (t) = = k+1
t.
2 k=0 2 k=0
2
(viii) Let g(t) = et and h(t) = 1/(1 − t). Then,
n n
X tk X
Tn g(t) = , Tn h(t) = tk.
k=0
k! k=0

Multiplying together we see that the term of degree k is

(1 + 1 + 2!1 + · · · + k!1 tk = (Tk exp(1)) tk,

and so n
X
Tn f (t) = (Tk exp(1)) tk.
k=0

(ix) Calculate the derivatives of (1 − t)−1/2, being careful with the minus signs (alterna-
tively, use the Binomial Theorem). The term of order k in the Taylor polynomial
is
1 · 3 · · · (2k − 1)
.
2 · 4 · · · 2k
To simplify it, note that the product of even terms

2 · 4 · . . . · 2n = 2n (n!).

Hence, if you multiply top and bottom by the missing even terms, you get
n
X 2k! k
Tn f (t) = t.
k=0
(k!2k )2

35
 (x) Decompose in partial fraction:
1 1 1 1 1
2
= + = + .
(2 − t − t ) 3(2 + t) 3(1 − t) 6(1 + (t/2)) 3(1 − t)
Using the Taylor polynomials of a geometric progression, we get
n n
! !
1 X 1 X
Tn f (t) = (−t/2)k + tk
6 k=0 3 k=0
n
(−1)k k
 
X 1
= 1 + k+1 t .
k=0
3 2

(xi) Using part 5. of Corollary 1.15 with p(t) = −t2, we get

⌊n/2⌋
X −1
Tn f (t) = t2k.
k
k=0

1
(xii) Recall that sin t cos t = 2
sin(2t). Using part 3. of Corollary 1.15 with p(t) = 2t,
we get
⌊(n−1)/2⌋
X (−1)k 4k 2k+1
Tn f (t) = t .
k=0
2k + 1!

x3,

x≥0
6. Recall that from Level 1, the rule f (x) can be written as f (x) = 3 .
−x , x < 0.
It is clear that all the derivatives of f exist at x = 1 or x = −1. At x = 0, only
f (0) = f ′ (0) = f ′′ (0) = 0 exist, the others do not. And so, we have the following Taylor
polynomials.

(i) T2 f = 0, Tn f does not exist for n ≥ 3.

(ii) T31 f = 1 + 3(x − 1) + 3(x − 1)2 + (x − 1)3.
(iii) T3−1 f = 1 − 3(x + 1) + 3(x + 1)2 − (x + 1)3.

7. (i) At the limit, the fraction is of type 0/0. To eliminate the ambiguity, it is quick
to see that we need to go to order at least 3 (maybe higher, but let’s start at 3).
Recall that T3 sin x = x − x3 /6, T2 cos x = 1 − x2 /2 and so

x3 x2 x3
    
T3 sin x
T3 tan x = T3 = x− 1+ = x+ .
T3 cos x 6 2 3
We can now replace everything and get, up to third order,

x3 2x3 7
T3 (x(1 + cos x) − 2 tan x) = 2x − − 2x − = − x3,
2 3 3
3 3
x x 1
T3 (2x − sin x − tan x) = 2x − x + −x− = − x3.
6 3 6
And so, the quotient, hence the limit, is 7.
Note that we would need to differentiate 3 times to use l’Hôpital Rule (try it?).

36
(ii) We need to expand the functions around x = 1 to identify the leading order term.
We get
1
Tn1 ln x = (x − 1) − (x − 1)2 + . . . .
2
1
And so, for the numerator, Tn1 (1 − x + ln x) = − (x − 1)2 + . . .. The denominator
2
can also be written as
√ p
1 − 2x − x2 = 1 − 1 − (x − 1)2

which brings in nicely the expansion around x = 1. Recall that, around z = 0,

√ z
Tn 1 − z = 1 − + . . . ,
2
and so our denominator has leading term
(x − 1)2 (x − 1)2
 
1− 1− = .
2 2
Finally, we get the limit as the quotient of the two leading terms, here equal to -1.
Note that l’Ĥopital Rule would have needed two derivatives. There is also another
way to derive the leading order term of the denominator:
p p
p (1 − 1 − (x − 1) 2 )(1 + 1 − (x − 1)2 )
1 − 1 − (x − 1)2 = p
1 + 1 − (x − 1)2
1 − (1 − (x − 1)2 ) (x − 1)2
= p ≈ + ...
1 + 1 − (x − 1)2 2

(iii) The two fractions tend to +∞ as x tends to 0. It is easier to recast the expression
ln(1 + x) − x
as one fraction, namely and evaluate the leading order terms. We
x ln(1 + x)
x2
have Tn ln(1 + x) = x − + . . ., and so the elements of the fraction become
2
x2
Tn (ln(1 + x) − x) = − ,
2
x2
Tn (x ln(1 + x)) = ,
2
and so the limit is −1/2.
(iv) We again have the difference ‘∞ − ∞’. Again, recast the expression as a fraction:
(sin x)2 − x
.
x(sin x)2
The leading order terms are

Tn ((sin x)2 − x) = −x + . . . ,
Tn (x(sin x)2 ) = x3 + . . . ,

and so the overall limit is −∞.

37
 (v) We start by establishing the leading order therm of the denominator around x = 0.
Recall that
2
2Tn sinh x = Tn (ex − e−x ) = Tn ex − Tn e−x = 2x + x3 + . . . ,
3
x3
and so x − sinh x = − + . . .. Therefore we need to look at the numerator and fix
3
α such that its leading term is at least cubic. The difficult term to deal with is

x3
 
− sin x 1 2
= xT3 1 − sin x + sin x = x − x2 + .


T3 xe
2 2

And so, up to cubic order, the numerator is equal to

x3 x3 2
T3 xe− sin x − sin x + αx2 = x − x2 + + αx2 = (α − 1) x2 + x3.


−x+
2 6 3
Therefore we need to set α = 1 and the limit is −2.

8. (i) We need to determine the leading order terms around x = 0 of

x6
Tn sin(x2 ) = x2 − + ...,
6
x2 x4
Tn (x2 cos x) = x2 (1 − + . . .) = x2 − ,
2 2
and so the leading order terms of the function are

x4 x6
− + ...
2 6
and so x = 0 is a minimum.
(ii) The rule f (x) is odd, f (−x) = −f (x), and so the critical point x = 0 can only be
a saddle point. We need to get the leading term of f (x) at x = 0. We know that

x3 x5 x7
Tn sin x = x − + − = ....
6 120 7!
For the fraction, using that (1 + z)−1 = 1 − z + z 2 − z 3 + . . ., we have

(x + ax3 )(1 + bx2 )−1 = (x + ax3 )(1 − bx2 + b2 x4 − b3 x6 + . . .),

and so the expansion of f (x) is

(−1/6 + b − a)x3 + (1/120 − b2 + ab)x5 + (−1/7! + b3 − ab2 )x7 + . . . .

When b 6= a + 1/6, the leading term is cubic, when b = a + 1/6 and a 6= −7/60,
the leading term is quintic, and, when (a, b) = (−7/60, 1/20), the leading term is
of order 7.

38
 g(x)
9. Consider g(x) = f (x) − Tna f (x). Then hx (x) = (x−a) n . Now, the n − 1-th derivative of

g and of (x − a)n are f (n−1) (x) − f (n−1) (a) − f (n) (x − a) and (n!)(x − a), respectively.
Clearly they are both 0 at x = a, so, using l’Hôpital Rule,

g (n−1) (x) f (n) (a) − f (n) (a)

lim = = 0.
x→a (n!)(x − a) n!

We can now re-iterate the process backward until we get

g ′ (x)
lim hn (x) = lim = 0.
x→a x→a n(x − a)(n−1)

10. Set a0 = p(a). Let q1 (x) = p(x) − a0 , so deg(q) = deg(p) and q1 (a) = 0. Hence,
there exists a unique polynomial r1 (x) of degree deg(q) − 1 = deg(p) − 1 such that
q1 (x) = r1 (x)(x − a). Therefore,

p(x) = a0 + r1 (x)(x − a).

Now, let a1 = r1 (a) and let q2 (x) = r1 (x) − a1 . Again, deg(q2 ) = deg(r1 ) = deg(p) − 1
and q2 (a) = 0. So, there exists an unique polynomial r2 of degree deg(r1 )−1 = deg(p)−2
such that q2 (x) = r2 (x)(x − a). Hence, r1 (x) = a1 + r2 (x)(x − a) and

p(x) = a0 + (a1 + r2 (x)(x − a))(x − a) = a0 + a1 (x − a) + r2 (x)(x − a)2.

The process can be re-iterated until the degree of rn is 0 and the conclusion follows.

39
Chapter 2

Real Sequences

2.1 Lecture 7: Definitions, Limit of a Sequence

Reference: Stewart Chapter 12.1, Pages 710–720. (Edition?)
We will define sequences properly soon, but for the moment think of a sequence as just as
an infinitely long list of real numbers in a definite order. For instance,

1, 2, 3, 4, . . . 1, 4, 9, 16, . . . − 1, 1, −1, 1, . . . 1/2, 1/4, 1/8, . . . 1/4, 1/2, 1/16, 1/8, . . . (2.1)

are all sequences. We will be interested in the limits of sequences, i.e. whether they get
closer and closer to a particular value as we take more and more terms. In the examples above,
providing we make reasonable assumptions about how the tail of the sequence behaves, the
first two sequences are tending to infinity, the third oscillates and so does not tend to a
limit and the fourth and fifth tend to 0. Determining whether a sequence converges can
be extremely difficult, so we will concentrate on a few situations and tools we can use. In
particular, we will see how to link a sequence with a real function and use the connection
with limits of functions to determine its convergence. You have seen and practiced limits
of functions at Level 1. Further, more general information, will be the subject of Chapter 5.

Cautionary Tale 2.1. The limit behaviour of a sequence is really determined by its (infinite)
tail, not by its first initial numbers. We shall come back later to this point.

2.1.1 Definition of a Sequence

Usually we use a more concise notation to define a sequence, by specifying a general term.
A general sequence
a1 , a2 , a3 , . . . , an , . . .
will be denoted by {an }∞
n=1 . The second sequence (2.1), i.e. 1, 4, 9, 16, . . . corresponding to
the function f (n) = n can be denoted by {n2 }∞
2
n=1 .

Definition 2.2. A sequence (of real numbers) is the infinite ordered list of real numbers
f (1), f (2), f (3), . . . , obtained from a function f : N → R.

40
Conversely, given a general sequence of real numbers a1 , a2 , a3 , a4 , . . ., we can define a function
f : N → R by setting f (n) = an . So, a sequence of real numbers and a function
f : N → R are the same thing.
Cautionary Tale 2.3. In principle, the function f defining a sequence has no need to
have a formula f (n) for its rule. For instance, a sequence starting by −π, e2, 8, −4.67, . . .
is unlikely to have a ‘reasonable rule’, although it will still define a function.

But, in the following examples, we have such rules.

Examples.

1. The first example of (2.1) is obtained from f : N → R given by f (n) = n.

(
27, if n is odd,
2. The function f : N → R, f (n) =
n2, if n is even,
determines the sequence 27, 4, 27, 16, 27, 36, . . ..
3. Revisiting the other examples of (2.1), we get for f : N → R
(
2 n −n 2n+1, if n is odd,
(i) f (n) = n , (ii) f (n) = (−1) , (iii) f (n) = 2 , (iv) f (n) =
2n−1, if n is even.

Remark 2.4. Some authors use the notation (n2 )∞ n=1 to emphasise that the order of the
terms in the sequence matters (Curly brackets are used with sets where the order of the terms
is irrelevant). Because the type with curly brackets is more common, we make this choice.

Our definition requires sequences to start with terms from n = 1. However, it can be sometimes
convenient to start with a different value of n.

Examples. For instance the sequence 2, 3, . . . could be denoted by {n}∞ ∞

n=2 or {n + 1}n=1 . The
2 ∞ 2 ∞
sequence {n + n}n=1 is 2, 6, 12 . . .. It can also be written as {n − n}n=2 . If we
 want to ignore
∞
2 ∞ ln(x − 4)
the first few terms we could write {n + n}n=4 or, with another sequence, .
n2 n=5
All of that has no bearing on limits.

2.1.2 Limit of a Sequence

We are often interested in what happens to a sequence as n becomes large, in particular,
whether it tends to a limit as n → ∞. In Chapter 5 we shall define limit for any sequence,
but it is quite complicated. Because you already know from Level 1 the notion of limit for
functions, we shall use it now for an easier definition, BUT working only for a reduced
class of sequences where we can calculate associated limits of functions.
Definition 2.5. Let {an }∞
n=1 be obtained from a function f : [1, ∞) ⊆ R → R, an = f (n)
whenever n ∈ N, AND suppose that lim f (x) = l, l ∈ R. Then the sequence is called
x→∞
convergent with limit lim an = lim f (x) = l. If one cannot find a limit, that is no
n→∞ x→∞
such function f exists, the sequence is called divergent.

41
Examples.

1 1
1. Let an = 2
. To show that lim an = 0, consider f : [1, ∞) → R, f (x) = 2 . Now
n n→∞ x
1
an = f (n) and lim 2 = 0. Then, by definition, lim an = 0.
x→∞ x n→∞

n2 ∞ n2
2. To determine whether the sequence { } is convergent, set an =
3n2 + 2n + 1 n=1 3n2 + 2n + 1
x2
and f : [1, ∞) → R, f (x) = so that f (n) = an when n ∈ N. Because
3x2 + 2x + 1
x2 1 1
lim f (x) = lim 2
= lim 2
= ,
x→∞ x→∞ 3x + 2x + 1 x→∞ 3 + 2/x + 1/x 3
1
lim an = lim f (x) = .
n→∞ x→∞ 3

Definition 2.5 is typical of definitions in mathematics where we use an explicit second object
(like the function f ) to define a property of a first one (the limit of the sequence {an }∞
n=1 ). For
such definitions to make sense we need to show that for ANY choice of the second object
we arrive at the same property. In our case we need the following result.

Lemma 2.6. Let {an }∞ n=1 be a convergent sequence with two functions f, g : [1, ∞) → R
such that an = f (n) = g(n) and limx→∞ f (x) = l1 , limx→∞ g(x) = l2 , l1 , l2 ∈ R. Then l1 = l2 .

Proof. By contradiction, suppose l1 6= l2 . Take ǫ = |l1 − l2 |/3 > 0 so that the intervals
(li − ǫ, l1 + ǫ) and (l2 − ǫ, l2 + ǫ) do not intersect. By definition of the limit of functions as
x → ∞, there exists m1 , m2 > 0 such that |f (x) − l1 | < ǫ when x > m1 and |g(x) − l2 | < ǫ
when x > m2 . Now, when x > max(m1 , m2 ),

3ǫ = |l1 − l2 | ≤ |l1 − f (x)| + |f (x) − g(x)| + |l2 − g(x)| < 2ǫ + |f (x) − g(x)|.

And so, ǫ < |f (x) − g(x)| for all x > max(m1 , m2 ). This is a contradiction because f (n) =
an = g(n) for all integer n > max(m1 , m2 ).

Cautionary Tale 2.7. In the definition of limit it is important that f such that f (n) = an
has a limit as x → ∞. The following example shows why. The sequence {0}∞ n=1 is clearly
convergent to 0. Indeed, using the definition, let an = 0 and f : [1, ∞) → R, f (x) = 0, then
limn→∞ an = limx→∞ f (x) = 0.
But, if we made another, ‘strange’, choice of function, things could go wrong. Choose f2 :
R → R, f2 (x) = sin(2πx), then f2 (n) = 0 if n ∈ N. However, limx→∞ f2 (x) does not exist.
This is not a contradiction. One important hypothesis of the definition is that limx→∞ f (x)
exists for (at least) one function f . That is the case for our first choice f , hence we can
conclude, although it is not the case for our second choice f2 . Note that, because it has no
limit, f2 tells us nothing about whether the sequence converges or not.

42
2.1.3 Graphic Representations of Sequences
n
In the following Figures 2.1 and 2.2, we portray the sequences {1+ (−1)
n
}∞ cos n ∞
n=1 and {1+ n }n=1 .

A sequence alternating and converging to 1

2

1.5

0.5
0 5 10 15 20 25 30

(−1)n
Figure 2.1: The first few values of a sequence alternating about the limit, like 1 + n
.

cos(πx)
They both converge to 1. We can take f1 (x) = 1 + for the first sequence and
x
cos x
f2 (x) = 1 + . Clearly,
x
 
cos(πx)  cos x 
1 = lim 1 + = lim 1 + .
x→∞ x x→∞ x

The first one alternates up and down as it goes to 1, the second oscillates in no organised way
towards 1.
A sequence converging to 1 in a random fashion
1.5

0.5

0
0 5 10 15 20 25 30

Figure 2.2: The first few values of a sequence converging in a random fashion to the limiting
value, like 1 + cosn n .

43
2.2 Lecture 8: Algebra of Limits, Special Sequences
In the previous lecture the limit of a sequence was defined, linking the limit of a sequence to
the limit of an associated function. We can use further this association and establish results
that are very useful to evaluate explicitly limits of convergent sequences. In practice, using
the algebra of limits, we can split limits into components we can evaluate, and put them
together to find the final limit. In this lecture we also introduce the notion of infinite limit.

2.2.1 Infinite Limits

As with limit of functions, we can define what we mean by lim an = +∞ or −∞.
n→∞

Definition 2.8. Suppose that f : [1, ∞) ⊆ R → R and f (n) = an whenever n ∈ N. Further-

more, suppose that limx→∞ f (x) = ±∞, respectively. Then, limn→∞ an = ±∞, respectively.

Example. The sequence {2n }∞ x

n=1 tends to (positive) infinity because limx→∞ 2 = +∞ (as
you have seen in Level 1).

Cautionary Tale 2.9. Recall that ±∞ are not real numbers. In particular, sequences with
limits +∞ or −∞ are divergent.

Remark 2.10. A divergent sequence does not need to grow unbounded. It can oscillate
between two values, for instance {(−1)n }∞
n=1 (see Exercise 2(k) in Sheet 2a). Therefore a
sequence either converges or diverges, but not both.

2.2.2 Algebra of Limits

We now have two theorems on limits of sequences which are very similar to theorems on limits
of functions which you have seen earlier at Level 1. The first theorem is the algebra of limits
for sequences and will be used to calculate limits of the type

cos(n2 )
 
lim 2 + + n ln(1 − 2/n) , (2.2)
n→∞ 3n
or
lim cos (n sin(1/n) − 1). (2.3)
n→∞

Theorem 2.11 (Algebra of Limits). Let {an }∞ ∞

n=1 and {bn }n=1 be convergent sequences and
c be a real constant. Then:

1. lim (an + bn ) = lim an + lim bn ;

n→∞ n→∞ n→∞
 
2. lim (can ) = c lim an ;
n→∞ n→∞
   
3. lim (an bn ) = lim an · lim bn ;
n→∞ n→∞ n→∞

44
 a  limn→∞ an
n
4. lim = , providing bn 6= 0 for any natural n and lim bn 6= 0.
n→∞ bn limn→∞ bn n→∞

5. Let f be a continuous function around a = limn→∞ an , then the sequence {f (an )}∞
n=1
converges to f (a).

Proof. Those results are left as Exercises 7 and 8 in Sheet 2a. They follow directly from the
definition of a limit and the same results for limits of functions you have seen at Level 1.

This result is not enough to get the previous limit (2.2). We need one more result. The second
theorem is the Sandwich Rule for sequences.
Theorem 2.12 (Squeeze Theorem, Sandwich Rule). Let {an }∞ ∞ ∞
n=1 , {bn }n=1 and {cn }n=1 be
sequences such that lim an = lim cn = l and for all n, an ≤ bn ≤ cn . Then lim bn = l.
n→∞ n→∞ n→∞

Proof. This result is left as Exercise 9 in Sheet 2a. It follows from the definition of a limit
and the Squeeze Theorem for functions you have seen at Level 1.
Cautionary Tale 2.13. The results of Theorem 2.11 do not hold if the limits are not finite.
For instance, for the product of limits, consider an = n2, bn = 1/n and cn = 1/n3. Then,
lim (an bn ) = +∞ =
6 0 = lim (an cn ),
n→∞ n→∞

even if both limn→∞ bn = limn→∞ cn = 0.

Examples. We can now revisit the previous examples to see how we can use Theorems 2.11
and 2.12.
1. For (2.2), we know that
−| cos(n2 )| cos(n2 ) | cos(n2 )|
lim ≤ lim ≤ lim .
n→∞ 3n n→∞ 3n n→∞ 3n
| cos(n2 )| 1
We know that lim ≤ lim = 0, and so the limit of this fraction is 0. For
 n→∞
 3n n→∞ 3n
2
lim n ln 1 − , we can use a Taylor polynomial expansion
n→∞ n
   
2 2 −2
lim n ln 1 − = lim x ln 1 − = lim x · = −2.
n→∞ n x→∞ x x→∞ x
And so we can collect everything together and we find the final limit as
cos(n2 )
 
lim 2 + + n ln(1 − 2/n) = 2 + 0 − 2 = 0.
n→∞ 3n
2. For the second limit, using a Taylor polynomial expansion,
lim n sin(1/n) = lim x · (1/x) = 1.
n→∞ x→∞

And so,
lim cos (n sin(1/n) − 1) = lim cos (x sin(1/x) − 1)
n→∞ x→∞
 
= cos lim x sin(1/x) − 1 = cos(0) = 1.
x→∞

45
2.2.3 Some Standard Convergent Sequences
We state next a few standard sequences with ‘well known’ limits. We can use the previous
theorems to get from their behaviour the limits of more complicated sequences. The proofs
are not always trivial!
Proposition 2.14. 1. Let p > 0, limn→∞ 1/np = 0.
2. The limit limn→∞ r n = 0 when |r| < 1. It diverges when |r| > 1.
3. The limit limn→∞ n1/n = 1.
4. If c > 0 then limn→∞ c1/n = 1.
5. This is a very important limit:
 n
1
lim 1 + = e,
n→∞ n
and has an equally important generalisation for a ∈ R:
 a n
lim 1 + = ea.
n→∞ n
np
6. For all a, p > 0, lim = 0.
n→∞ ean

ln n
7. For all p > 0, lim p = 0.
n→∞ n

Proof. The proofs will mainly follow from the application of l’Hospital’s Rule or the following
idea. If f is a positive function such that limx→∞ ln(f (x)) exists, because exp x and ln x are
continuous on their domain of definition, the following limit makes sense
 
lim f (x) = exp lim ln f (x) . (2.4)
x→∞ x→∞

It is also straightforward to check that when limx→∞ ln(f (x)) = −∞,

lim f (x) = lim exp (ln f (x)) = 0. (2.5)
x→∞ x→∞

1. Using (2.4),
   
lim n−p = exp lim ln x−p = exp −p lim ln x = 0.
n→∞ x→∞ x→∞

2. Again, we shall be using (2.4) and the Squeeze Theorem. When r > 0,

−∞, r < 1;
lim x ln r =
x→∞ ∞, r > 1.
And so, when 0 ≤ r < 1,
 
n
lim r = exp lim x ln r = 0
n→∞ x→∞

and, when r > 1,  

lim r n = exp lim x ln r = ∞.
n→∞ x→∞
Moreover, when −1 < r < 0, −|r| ≤ r ≤ 0 and so, from the Squeeze Theorem,
limn→∞ r n = 0 and, obviously, r n diverges when r < −1.

46
3. Using (2.4),
 
1/x

1/x
 ln x
lim x = exp lim ln x = exp lim = exp(0) = 1,
x→∞ x→∞ x→∞ x

ln x 1
having used l’Hospital Rule for the limit lim = lim = 0.
x→∞ x x→∞ x

4. Using (2.4),
 
1/x

1/x
 ln c
lim c = exp lim ln c = exp lim = exp(0) = 1.
x→∞ x→∞ x→∞ x

5. Using (2.4), and Taylor series expansion for the last limit, we have for all a ∈ R,
 a n  a   a
lim 1 + = exp lim x ln(1 + ) = exp lim x · = ea.
n→∞ n x→∞ x x→∞ x

6. Use the Squeeze Theorem because 0 ≤ np ≤ nN for any integer N > p. Then, use N
times l’Hospital Rule of the type ‘∞/∞’:

xN N!
lim = lim = 0.
x→∞ eax x→∞ aN eax

7. Use l’Hospital Rule of the type ‘∞/∞’ once. We get

ln x 1/x 1
lim = lim = lim = 0.
x→∞ xp x→∞ pxp−1 x→∞ pxp

47
2.3 Lecture 9: Bounded and Monotone Sequences
We defined the convergence of a sequence and found some ways of evaluating limits. It may
often be difficult to calculate the explicit limit of a sequence, so we split the problem into two
stages:

1. show that a limit exists, then

2. calculate the limit (when it exists), even using numerical means.

It is only the first part that interests us here. We are going to determine sufficient conditions
for the convergence of sequences.

2.3.1 Bounded Sequences

Convergence implies that only a finite number of terms of the sequence lie outside of a neigh-
bourhood of this limit. Hence, the sequence is bounded, with the following formal definitions.

Definition 2.15. A sequence {an }∞

n=1 is bounded below (bounded above) if there is
m ∈ R (M ∈ R) such that an ≥ m (an ≤ M) for all n ≥ 1. We say that a sequence is
bounded if it is bounded below AND above, that is, if m ≤ an ≤ M for all n ≥ 1.

Examples. The sequence {en + cos n} satisfies 0 ≤ e − 1 ≤ en + cos n and so is bounded

below (m = e − 1 or 0). Also, −π/2 ≤ arctan n ≤ π/2 and so {arctan n} is bounded.

The sequence {(−1)n }∞

n=1 is bounded BUT not convergent. The converse is however true
which we next prove.

Theorem 2.16 (Convergence implies bounded). A convergent sequence is bounded.

Proof. Let {an }∞

n=1 be a convergent sequence with limit a, induced by f : [1, ∞) → R such
that an = f (n) and limx→∞ f (x) = a. Take ǫ = 1. There exists an integer N such that
|f (x) − a| < ǫ = 1 for all x ≥ N. Define m = min{ f (1), f (2), . . . , f (N), a − 1 } and
M = max{ f (1), f (2), . . . , f (N), a + 1 }, then m ≤ f (n) ≤ M for all n ∈ N.

From the contrapositive statement (see Level 1) of Theorem 2.16, it follows that ‘not
bounded’ implies ‘not convergent’: an unbounded sequence is divergent.

2.3.2 Convergent Sequences and Closed Bounded Intervals

Theorem 2.17 (Non-negative sequences). If {xn }∞
n=1 is a convergent sequence of real numbers
and if xn > 0 for all n ∈ N then
l = lim xn > 0.
n→∞

Similarly, if xn ≤ 0, then l ≤ 0.

48
Proof. We will defer this until a later lecture when we meet the ǫ — N definition of conver-
gence.

What follows is an important property characterising the limit of a convergent sequence re-
maining in a closed bounded interval.

Corollary 2.18 (Closed bounded intervals contain their limits). If {xn }∞ n=1 is a convergent
sequence in the closed bounded interval [a, b] then the limit l ∈ [a, b].

Proof. We show that l ≥ a. A similar argument shows that l ≤ b. Being a closed interval, it
means that l ∈ [a, b]. Consider the sequence {yn }∞
n=1 where yn = xn − a. Then yn ≥ 0 and so
limn→∞ yn = l − a ≥ 0.

2.4 Lecture 10: Monotone Sequences

As with functions, we can use the information that a sequence is increasing or decreasing.

Definition 2.19. A sequence {an }∞ n=1 is increasing (or decreasing) if an ≤ an+1 (or
an ≥ an+1 ) for all n ≥ 1 or, in other words, if a1 ≤ a2 ≤ a3 ≤ . . . (increasing) or if
a1 ≥ a2 ≥ a3 ≥ . . . (decreasing). A sequence that is either increasing or decreasing is called
monotone.

Remark 2.20. From our definitions, a sequence can be both increasing and decreasing. In
that case it must be of constant value, say c. Formally such a sequence {c}∞
n=1 is monotone.

Examples.
 
n
1. To determine whether the sequence is increasing or decreasing, we could simply
n+1
n
calculate the sign of the difference an+1 − an where an = . We get
n+1
n+1 n (n + 1)2 − n(n + 2) 1
an+1 − an = − = = > 0.
n+2 n+1 (n + 1)(n + 2) (n + 1)(n + 2)

Hence an+1 > an and so an ≤ an+1 and the sequence is increasing.

 
n n
2. For the sequence 2
, let an = 2 . We consider
n +1 n +1

n+1 n (n + 1)(n2 + 1) − n((n + 1)2 + 1)

an+1 − an = − =
(n + 1)2 + 1 n2 + 1 (n2 + 1)((n + 1)2 + 1)
(n3 + n2 + n + 1) − (n3 + 2n2 + n + 1) n2
= = − < 0.
(n2 + 1)((n + 1)2 + 1) (n2 + 1)((n + 1)2 + 1)

Hence an+1 < an and so an ≥ an+1 and the sequence is decreasing.

49
When a sequence is obtained from a continuously differentiable function f : [1, ∞) → R, we
can use Level 1 results about the monotonicity of functions to get the following result.
Proposition 2.21. Let f : [1, ∞) ⊆ R → R be continuously differentiable and {an }∞ n=1 be the
sequence obtained from f such that an = f (n), n ∈ N. If f (x) ≥ 0, x ∈ [1, ∞), then {an }∞
′
n=1
is increasing, if f ′ (x) ≤ 0, x ∈ [1, ∞), then {an }∞
n=1 is decreasing.

Proof. Observe that, from the Mean Value Theorem, and with (n + 1) − n = 1,
f (n + 1) − f (n)
xn+1 − xn = = f ′ (ξn )
(n + 1) − n
where ξn ∈ (n, n + 1). If the derivative of f is non-negative then the sequence is increasing, if
it is non-positive, the sequence is decreasing.

We can revisit the two previous examples.

Examples.
 ′
x ′ 1 1
1. Let f (x) = . Then f (x) = 1 − = ≥ 0 for x ≥ 1. and so, as
x+1 x+1 (x + 1)2
expected, the sequence {f (n)}∞
n=1 is increasing.

x ′ 1 − x2
2. Let f (x) = . Then f (x) = ≤ 0 for x ≥ 1. And so, the sequence
x2 + 1 (x2 + 1)2
{f (n)}∞
n=1 is decreasing.

Remark 2.22. As a comment here, a minority of texts define ‘increasing’ as x1 < x2 <
· · · and ‘decreasing’ as x1 > x2 > · · · which will be defined in Analysis later as strictly
increasing and strictly decreasing. Then, those texts use the terms ‘non decreasing’ for
what we define here as increasing and ‘non increasing’ for decreasing. This minor difference
in terminology makes no difference to any of the results covered in this study block.

2.4.1 Convergence of Monotone, Bounded Sequences

To be sure that a bounded sequence is convergent we need to add one more condition. The
main result that we will consider about such sequences is as follows.
Theorem 2.23 (Monotone Convergence Theorem). A monotone sequence converges if and
only if it is bounded. More precisely, a bounded above and increasing sequence OR a bounded
below and decreasing sequence are convergent.

The proof of that theorem will have to wait until Analysis II (MA2731) in Term 2.
Remark 2.24. We shall also see in Term 2 that the limit of a monotone convergent sequence
{sn }∞
n=1 can be determined by, respectively,

lim xn = sup{ xn : n ∈ N } (xn increasing), lim xn = inf{ xn : n ∈ N } (xn decreasing).

n→∞ n→∞

We shall not give here precise definitions of the sup and the inf of a set. This will happen in
MA2731 next term. Simply take them as

50
 1. the ‘lowest upper bound’ for the sup, like in

sup{ 1 − 1/n : n ∈ N} = sup{ 0, 1/2, 2/3, 3/4 · · · } = 1,

2. the ‘highest lower bound’ for the inf, like in

inf{ 1/n : n ∈ N} = inf{ 1, 1/2, 1/3, 1/4 · · · } = 0.

This result indicates that we have a precise description of the limit of a monotone sequence,
provided we know the definition and to calculate the sup and inf of a set. We illustrate what
happens in Figures 2.3 and 2.4.
A sequence increasing to 1
1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30

Figure 2.3: The first few values of a sequence increasing to 1, like (1 − n1 ).

A sequence decreasing to 1
2

1.9

1.8

1.7

1.6

1.5

1.4

1.3

1.2

1.1

1
0 5 10 15 20 25 30

Figure 2.4: The first few values of a sequence decreasing to 1, like (1 + n1 ).

We now finish by looking at examples. We can calculate the limits using our previous tech-
niques, but we show that they satisfy Theorem 2.23.

Examples. Consider the following sequences {xn }∞

n=1 .

1
1. The sequence {an }∞
n=1 where an = is convergent because it is
2n + 3
(i) bounded, as clearly 0 ≤ an and an ≤ 1 (since 2n + 3 > 1) for all n > 1;

51
 (ii) monotone decreasing because

1 1 (2n + 1) − (2n + 3) 2
an+1 −an = − = =− < 0.
2(n + 1) + 3 2n + 1 (2n + 1)(2n + 3) (2n + 1)(2n + 3)
√ √
n ∞ n
2. The sequence { }n=1 is convergent because it is bounded by 0 ≤ ≤ 1 and,
n+1 √ n+1
x
defining f : [1, ∞) by f (x) = , its derivative is negative when x ≥ 1:
x+1
1 √ √ √
′
√
2 x
(x + 1) − x (x + 1) − 2 x x 1−x
f (x) = 2
= √ = √ ≤ 0.
(x + 1) 2 x(x + 1)2 2 x(x + 1)2

And so the sequence is decreasing and bounded (hence convergent).

√ √
3. The sequence { n + 1− n}∞ n=1 is the difference of two divergent sequences and thus the
result about combining convergent sequences does √ not apply.
√ However, it is decreasing
because, either we could show that for f (x) = x + 1 − x, f ′ (x) < 0 for x ≥ 1, or we
could see that f (x) simplifies (when multiplying the top and bottom by the conjugate
√ √ 1
quantity n + 1 + n) to f (x) = √ √ . This function is clearly decreasing.
n+1+ n
2n2 − 3n ∞
4. The sequence { } is an example of a divergent sequence. It is increasing
n + 1 n=1
2x2 − 3x
because f ′ (x) > 0 when x > 1 for f (x) = , but also unbounded because of the
x+1
following estimate

2n2 − 3n 2(n + 1)2 − 7n − 2 7n + 7 − 5

xn = = = 2(n + 1) −
n+1 n+1 n+1
5
= 2(n + 1) − 7 + ≥ 2n − 5.
n+1

52
Summary of Chapter 2
We have seen:

• the definition of a sequence and of a convergent/divergent sequence;

• how to calculate limits of many types of sequences;

• the meaning of monotone (increasing/decreasing) and bounded sequences, and

that all bounded monotone sequences converge.

• To know the definition and properties of monotone bounded sequences.

53
2.5 Exercise Sheet 2

2.5.1 Exercise Sheet 2a

1. Fully revise the Level 1 material on limits of functions.

2. Determine which of the following sequences converge, and for those that do, find the
limit.
n n o∞ n 1 o∞ n  1 o∞
(a) ; (b) 1 − n ; (c) cos ;
nn + 2n on=1
(−1) ∞ n  12 on=1
∞ n n on=1
 nπ ∞
(d) 2
; (e) ln ; (f) sin ;
n n no∞ n=1 n n 1 n=1
n o∞ n n 2o n=1
(g) ; (h) 1− ; (i) ;
2n on=1
n ln n n=1
n 2n +n1o
n n o (−1)
(j) ; (k) (−1)n ; (l) .
n n

3. Discuss the convergence of the following sequences {xn }∞

n=1 with the general terms:

(i) xn = (sin n)(sin(1/n)),

√
(ii) xn = −n + n2 + 3n,
5n3 + 3n + 1
(iii) xn = ,
15n3 + n2 + 2
sin(n2 + 1)
(iv) xn = ,
n2 + 1
√ √
n+2− n+1
(v) xn = √ √ ,
n+1− n
7n4 + n2 − 2
(vi) xn = ,
14n4 + 5n − 4
n3 + 3n2
(vii) xn = − n2,
n+1
 3
n+1
(viii) xn = − n3,
n
and calculate their limit when it exists.

4. (i) Show that a polygon

π with n equal sides inscribed in a circle of radius r has perimeter
pn = 2rn sin .
n
(ii) By finding the limit of the sequence {pn }∞n=1 , derive the formula for the circumfer-
ence of a circle. (Hint: put m = π/n and use l’Hôpital’s rule.)
n! 1 n!
5. Show that 0 ≤ n
≤ and hence compute lim n .
n n n→∞ n
6. Calculate the limit (when it exists) of the sequences with the following general terms.
n3 + 2n2 + 1
(i) xn = .
6n3 + n + 4

54
 n2 + n + 1
(ii) xn = .
3n2 + 4
√
(iii) xn = n4 + n2 − n2.
 4
n+1
(iv) xn = − n4.
n2
√
(v) xn = −n + n2 + n.
sin n √ √
(vi) xn = + ( n + 1 − n).
n
n2 + 500n + 1
(vii) xn = .
5n2 + 3
p p
(viii) xn = n(n + 1) − n(n − 1).

7. Prove Statements 1-4 of Theorem 2.11.

8. Prove Statement 5 of Theorem 2.11.

9. Prove the Squeeze Theorem 2.12.

∗
10. We can prove a simplified version of the Contraction Mapping Theorem that some of
you will see next term in MA2731. Let φ : R → R be a function such that φ′ (x) > 0 for
all x ∈ R. Given x0 ∈ R, define a sequence {xn }∞
n=1 by xn+1 = φ(xn ), n ≥ 0. Sequences
of this type are often monotone. More precisely, show that the sequence is increasing if
φ(x0 ) > x0 and decreasing if φ(x0 ) < x0 .

Short Feedback for Sheet 2a

1. Bookwork.

2. (a) 1; (b) 1; (c) 1;

(d) 0; (e) divergent, −∞; (f) divergent;
(g) 0; (h) 1/e; (i) 1/2;
(j) 0; (k) divergent; (l) 0.

3. The limits are

(i) convergent.
(ii) 3/2.
(iii) 1/3.
(iv) 0.
(v) 1.
(vi) 1/2.
(vii) ∞.
(viii) −∞.

4. (i) Divide the polygon into triangles.

(ii) 2πr.

55
 5. 0.

6. The limits are

(i) 1/6.
(ii) 1/3.
(iii) 1/2.
(iv) −∞.
(v) 1/2.
(vi) 0.
(vii) 1/5.
(viii) 1.

7. Use the equivalent results you have seen at Level 1 for the limits of functions.

8. Use the fact that f is continuous and that you have seen at Level 1 a similar result for
limits of functions.

9. To use the Squeeze Theorem for limit of functions, you need to use a continuous function
obtained by linking by straight lines two consecutive points of the sequence.
∗
10. Compare x1 − x0 in both cases.

2.5.2 Feedback for Sheet 2a

1. Bookwork.

2. See ‘short feedback’.

3. (a) Let xn = (sin n)(sin(1/n)). We know that | sin n| 6 1 and that | sin x| 6 |x| for all
x ∈ R, therefore

lim |xn | = lim |(sin n)(sin(1/n))| 6 lim | sin(1/n)| 6 lim 1/n = 0.

n→∞ n→∞ n→∞ n→∞

4. .

5. .

6. .

7. .

8. .

9. .

56
 ∗
10. Observe that by the Mean Value Theorem (a result you have met in Calculus at Level
1)
xn+1 − xn = φ(xn ) − φ(xn−1 ) = φ′ (ξn )(xn − xn−1 )
where ξn is some point between xn−1 and xn . If the starting point x0 is such that
x1 = φ(x0 ) > x0 then this successively implies that x2 − x1 > 0, x3 − x2 > 0, · · · ,
xn+1 − xn > 0. That is xn+1 > xn > · · · > x2 > x1 > x0 , i.e. the sequence is increasing.
Similar reasoning shows that if instead x1 < x0 then the sequence is decreasing.

57
Chapter 3

A flipped classroom approach to

Improper Integrals

3.1 Self-study for Lecture 11: Improper Integrals —

Type 1
Reference: Stewart Chapter 8.8, Pages 544–547.
In this part we will see how to integrate functions over infinite intervals. In the next part we
will see that we can sometimes integrate functions even when they are not defined at one or
more points covered by the range of integration.
Let t
t 
1 1 1
Z
A(t) = 2
dx = − =1− .
1 x x 1 t
Thus however large we take t, we have A(t) < 1 so it is always finite. Also lim A(t) = 1, so
t→∞
it seems reasonable to say that Z ∞
1
dx = 1.
1 x2

However, we do need to be careful here. We cannot just deal with the ∞ as if it were a
number, because it’s not a number. There is another reason why we cannot just do this.
Usually when we integrate, we use the Fundamental Theorem of Calculus to compute the
integral asZ the difference between two evaluations of an antiderivative. For example, we
b h ib
might say f (x) dx = F (x) where F is an antiderivative for f . We have only proved the
a a
Fundamental Theorem of Calculus for finite intervals, so we can’t just plug in infinity and
hope for the best.
Definition 3.1 (Improper Integrals of Type 1).
Z t
1. If f (x) dx exists (i.e. can be computed and is finite) for all t ≥ a, then we define
a
Z ∞ Z t
f (x) dx = lim f (x) dx,
a t→∞ a

58
 provided this limit exists.
Z b
2. If f (x) dx exists (i.e. can be computed and is finite) for all t ≤ b, then we define
t
Z b Z b
f (x) dx = lim f (x) dx,
−∞ t→−∞ t

provided this limit exists.

Z ∞ Z a
3. If for some value of a, f (x) dx and f (x) dx both exist then we define
a −∞
Z ∞ Z a Z ∞
f (x) dx = f (x) dx + f (x) dx.
−∞ −∞ a

If an improper integral exists we say that the integral is convergent. Otherwise we say that it
is divergent.
Z ∞
1
Example. Determine whether dx is convergent and if so evaluate it.
1 x

Solution. We have
∞ t
1 1
Z Z
 t
dx = lim dx = lim ln x 1 = lim ln t = ∞.
1 x t→∞ 1 x t→∞ t→∞

Since the limit does not exist, the integral is divergent.

In the lecture notes our next example was

Z ∞
1
dx.
−∞ 1 + x2

Here though we give a harder example, you should complete this yourself — for practice.
Z ∞
e−x
Example. Determine −2x
dx.
−∞ 1 + e

Solution. We have
∞ 0 ∞
e−x e−x e−x
Z Z Z
dx = dx + dx,
−∞ 1 + e−2x −∞ 1 + e−2x 0 1 + e−2x

provided both of these integrals are convergent.

Z ∞ Z t
e−x e−x
We compute each of them separately. dx = lim dx. To compute
0 1 + e−2x t→∞ 0 1 + e−2x

this integral, we substitute u = e−x . Then du = −e−x and so “− du = e−x dx”. When x = 0,
dx
u = 1 and when x = t, u = e−t .

59
Consequently
t e−t
e−x −1
Z Z
e−t
du = lim − tan−1 u 1 = lim (− tan−1 (e−t )+tan−1 1).

lim dx = lim
t→∞ 0 1 + e−2x t→∞ 1 1+u 2 t→∞ t→∞

Because tan is continuous on (−π/2, π/2), the function tan−1 is continuous and so

lim tan−1 (e−t ) = tan−1 lim e−t = tan−1 (0) = 0.

t→∞ t→∞

t
e−x
Z
Hence lim dx = π/4.
t→∞ 0 1 + e−2x
Z 0 Z 0
e−x e−x
On the other hand −2x
dx = lim −2x
dx = lim (− tan−1 1 + tan−1 (e−t )).
−∞ 1 + e t→−∞ t 1+e t→−∞

Now as t → −∞, e−t → ∞ and since tan−1 is continuous, lim tan−1 (e−t ) = lim tan−1 y = π/2.
t→−∞ y→∞
Z 0 −x
e
Hence lim dx = π/2 − π/4 = π/4. Consequently
t→−∞ t 1 + e−2x

Z ∞
e−x
−2x
dx = π/4 + π/4 = π/2.
−∞ 1 + e

Exercise 3.2. Determine whether the following integrals are convergent and if so evaluate
them.
∞ 0 ∞ ∞
1 1
Z Z Z Z
x
(a) dx; (b) e dx; (c) dx; (d) x3 dx.
1 x3 −∞ −∞ 1 + x2 −∞

Summary
We have seen:

• that when we integrate over an infinite range, we cannot just treat ∞ like a number;

• how to evaluate integrals over an infinite range using a limit.

60
3.2 Self-study for Lecture 11: Improper Integrals —
Type 2
Reference: Stewart Chapter 8.4, Pages 547–550.
So far we have only integrated functions that are continuous or at least consist of a number
of continuous pieces. This is because whenever we compute a definite integral, we are using
the Fundamental Theorem of Calculus and this only applies to continuous functions. In this
session we will see how we can integrate functions that are discontinuous. In some cases the
functions may not even be defined at all points in the range of integration.
Suppose that f is continuous on the interval [a, b) but has a vertical asymptote at x = b, that
is, f is undefined at x = b. We can still try to find the area beneath the graph of f . Let
Z t
A(t) = f (x) dx.
a

Suppose that A(t) tends to a finite limit as t → b− , then it would be reasonable to define the
area under f to be lim− A(t) and say that
t→b

Z b
f (x) dx = lim− f (x) dx.
a t→b

Definition 3.3 (Improper Integrals of Type 2).

1. Suppose that f is continuous on [a, b) and discontinuous at b. (f does not even have to
be defined at x = b.) Then we define
Z b Z t
f (x) dx = lim− f (x) dx,
a t→b a

providing this limit exists.

2. Suppose that f is continuous on (a, b] and discontinuous at a. (f does not even have to
be defined at x = a.) Then we define
Z b Z b
f (x) dx = lim+ f (x) dx,
a t→a t

providing this limit exists.

3. Suppose that f is continuous on [a, b] except at c ∈ (a, b) where it may not even be
Z c Z b
defined and that f (x) dx and f (x) dx both exist. Then we define
a c
Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx.
a a c

61
Again we describe integrals as convergent or divergent depending on whether they exist.
1
x
Z
Example. Determine whether √ dx is convergent and if so evaluate it.
−3 9 − x2

Solution. We have
 √ √ √
Z 1 Z 1
x x 1
√ dx = lim+ √ dx = lim+ − 9 − x2 t = lim+ ( 9 − t2 − 8).
−3 9 − x2 t→−3 t 9 − x2 t→−3 t→−3

√
Z 1
x √
+ 2 + 2 √
As t → −3 , 9 − t → 0 and so 9 − t → 0. Hence dx = − 8.
−3 9 − x2

3
1
Z
Example. Determine whether dx is convergent and if so evaluate it.
−1 x2

Solution. We have
3 Z 0 Z 3
1 1 1
Z
2
dx = 2
dx + 2
dx,
−1 x −1 x 0 x
providing both of these integrals are convergent. We will try to evaluate the first one.
Z 0 Z t
1 1 h 1 it  1 
2
dx = lim dx = lim − = lim − − 1 .
−1 x t→0− −1 x2 t→0− x −1 t→0− t
This limit does not exist, so the original integral is divergent.

Here is a completely incorrect solution.

1
Bad Solution: Suppose we fail to notice that the function 2 is not continuous at 0. If we
x
just use the Fundamental Theorem of Calculus we get
Z 3
1 h 1 i3 4
2
dx = − = − .
−1 x x −1 3

This cannot possibly be correct because we have integrated a positive function and obtained
a negative answer.
Exercise 3.4. Find:
Z 2 Z 1 4 ∞
1 1 1 1
Z Z
(a) dx; (b) √ dx; (c) dx; (d) √ dx.
1 1−x 0 1−x 1 (x − 2)2/3 0 x

Summary
We have seen:

• that more care is required to integrate functions when they are not continuous;
• how to evaluate integrals when the integrand is not continuous / undefined at certain
points using a limit.

62
3.3 Homework for improper integrals, Types 1 and 2
1. Finish off any remaining in-class exercises.

2. Determine which of the following integrals exist and evaluate them when possible.
Z ∞ Z ∞ Z ∞
1 1
(a) dx; (b) dx; (c) cos x dx;
1 (3x + 1)2 0 x2 + 5x + 6 0
Z ∞ Z 2 Z π/2
−x 2
(d) e cos x dx; (e) x ln x dx; (f) tan x dx;
0 0 0
1 ∞ ∞
e1/x 1 1
Z Z Z
(g) dx; (h) dx; (i) √ dx;
0 x2 −∞
2
x +4 0 x(x + 1)

[Hint: It should be helpful to know that for any k > 0, lim+ xk ln x = 0. How do we
x→0
prove this?]

3. Find the values of p for which the following integrals exist and evaluate the integrals
when they exist.
1 ∞ ∞
1
Z Z Z
p p
(a) x dx; (b) x dx; (c) dx.
0 1 e x(ln x)p
2
4. (a) Show that e−x ≤ e−x for x ≥ 1.
Z ∞
(b) Determine e−x dx.
1
Z ∞
2
(c) What can you deduce about e−x dx?
1

5. If f is continuous
Z ∞on [0, ∞), then its Laplace transform is a function F for which the
rule is F (s) = f (t)e−st dt, and the domain of F consists of those s for which the
0
integral converges.
Find the Laplace transform of:

(a) f (t) = 1; (b) f (t) = et ; (c) f (t) = t.

6. The definitions of the trigonometric functions and hyperbolic functions can be extended
to all complex numbers. Use Euler’s theorem concerning eix and the definition of sinh
and cosh to show that sinh x = −i sin(ix) and cosh x = cos ix.

7. In this question we have a random mixture of integrals. You have to work out the
method required. A couple of them are not too hard, most of them are slightly tricky
and some are very tricky. Evaluate:

63
 1 1 2
1 ex
Z Z Z
x − x2  dx;
 
(a) √ dx; (b) √ dx; (c)
0 x2 + 4 0 e2x + 1 −1
π/4 π/3
x4
Z Z Z
2 2
(d) dx; (e) cos t tan t dt; (f) sin(4x) cos(3x) dx;
x10 + 16 0 π/4

1 1 1
Z Z Z
(g) √ dx; (h) dx; (i) √ √ dx;
x2 + 4x + 8 e − e−x
x
x+1− x
1
etan (y)
Z −1 Z Z
2 x
(j) dy; (k) x sin x dx; (l) ex+e dx;
−1 1 + y2
3
1
Z Z Z
2
(m) dx; (n) ln(x − 1) dx; (o) r 4 ln r dr.
1 + ex 1

8. Challenge Question: Find the value of the constant C for which the integral
Z ∞ 
x C
− dx
0 x2 + 1 x + 1
exists. Evaluate the integral for this value of C.

3.4 Feedback

3.4.1 Feedback on in-class exercises

Lecture 1: Improper Integrals — Type 1

Feedback on In-Class Exercises

Exercise 1:

(a)
∞ t
1 1 h 1 it  1 1 1
Z Z
dx = lim dx = lim − = lim − + = .
1 x3 t→∞ 1 x3 t→∞ 2x2 1 t→∞ 2t2 2 2

(b)
Z 0 Z 0  x 0
x
e dx = lim ex dx = lim e t = lim (1 − et ) = 1.
−∞ t→−∞ t t→−∞ t→−∞

(c)
∞ 0 ∞
1 1 1
Z Z Z
dx = dx + dx,
−∞ 1 + x2 −∞ 1 + x2 0 1 + x2
providing both of these integrals converge. Now
Z ∞ Z t
1 1 t π
dx = lim tan−1 x 0 = lim tan−1 t = .

2
dx = lim 2
0 1+x t→∞ 0 1 + x t→∞ t→∞ 2

64
 Similarly
Z 0 0
1 1 π
Z
−1
0
= lim (− tan−1 t) = .

dx = lim dx = lim tan x
−∞ 1 + x2 t→−∞ t 1+x2 t→−∞ t t→−∞ 2

Hence
∞ 0 ∞
1 1 1
Z Z Z
dx = dx + dx = π/2 + π/2 = π.
−∞ 1 + x2 −∞ 1 + x2 0 1 + x2

(d) Z ∞ Z 0 Z ∞
3 3
x dx = x dx + x3 dx,
−∞ −∞ 0

providing both of these integrals converge. Now

Z ∞ Z t h x4 it
3 3 t4
x dx = lim x dx = lim = lim = ∞.
0 t→∞ 0 t→∞ 4 0 t→∞ 4

Z ∞
So at least one of the two integrals diverges and hence x3 dx diverges.
−∞

Lecture 2: Improper Integrals — Type 2

Feedback on In-Class Exercises

Exercise 1:

1
(a) is undefined at x = 1 but is continuous on (1, 2]. Hence
1−x
Z 2 Z 2
1 1  2
dx = lim+ dx = lim+ − ln |1 − x| t = lim+ ln(t − 1) = −∞.
1 1−x t→1 t 1−x t→1 t→1

So the integral diverges.

1
(b) √ is undefined at x = 1 but is continuous on [0, 1). Hence
1−x
Z 1 Z t
1 1  √ t √
√ dx = lim− √ dx = lim− −2 1 − x 0 = lim− (2 − 2 1 − t) = 2.
0 1−x t→1 0 1−x t→1 t→1

1
(c) is undefined at x = 2 but is continuous on [1, 4] \ {2}. Hence
(x − 2)2/3
Z 4 Z 2 Z 4
1 1 1
2/3
dx = 2/3
dx + 2/3
dx,
1 (x − 2) 1 (x − 2) 2 (x − 2)

65
 providing both of these integrals converge. Now
Z 2 Z t
1 1  1/3 t

2/3
dx = lim dx = lim 3(x − 2)
1 (x − 2) t→2− 1 (x − 2)2/3 t→2− 1

= lim− (3(t − 2)1/3 + 3) = 3

t→2

and
4 Z 4
1 1
Z
1/3 4
 
dx = lim dx = lim 3(x − 2) t
2 (x − 2)2/3 t→2+ t (x − 2)2/3 t→2+

= lim+ (3 × 21/3 − 3(t − 2)1/3 ) = 3 × 21/3 .

t→2

Hence
4 2 4
1 1 1
Z Z Z
dx = dx + dx = 3 + 3 × 21/3 .
1 (x − 2)2/3 1 (x − 2)2/3 2 (x − 2) 2/3

(d) This integral is improper for two reasons. First, the range of integration is infinite and
1 1
second, √ is undefined at 0. However √ is continuous on (0, ∞). So
x x
Z ∞ Z 1 Z ∞
1 1 1
√ dx = √ dx + √ dx
0 x 0 x 1 x

provided both of these integrals converge. We have

Z ∞
1
Z t
1  √ t √
√ dx = lim √ dx = lim 2 x 1 = lim 2 t − 2 = ∞.
1 x t→∞ 1 x t→∞ t→∞

Z ∞
1
So, at least one of the two integrals diverges and hence √ dx does not converge.
0 x

3.4.2 Homework feedback for improper integrals

Homework solutions for Improper Integrals, Types 1

and 2
2. (a)
∞ t
1 1 1
Z Z h it
dx = lim dx = lim −
1 (3x + 1)2 t→∞ 1 (3x + 1)
2 t→∞ 3(3x + 1) 1
h 1 1 i 1
= lim − + = .
t→∞ 3(3t + 1) 12 12

66
(b) We have
∞ t
1 1
Z Z
2
dx = lim dx.
0 x + 5x + 6 t→∞ 0 x2 + 5x + 6
t
1
Z
We will determine dx using the method of partial fractions.
0 + 5x + 6 x2
Factorising gives x2 + 5x + 6 = (x + 2)(x + 3) and so
1 A B
= + ,
x2 + 5x + 6 x+2 x+3
where A and B are constants to be determined.
Standard calculations yield A = 1 and B = −1. So
Z t Z t Z t
1 1 1  t  t
2
dx = dx − dx = ln |x + 2| 0
− ln |x + 3| 0
0 x + 5x + 6 0 x+2 0 x+3
t + 2 3
= ln + ln .
t+3 2
Hence t
1 t + 2 3
Z
lim dx = lim ln + ln .
t→∞ 0 x2 + 5x + 6 t→∞ t+3 2
t+2
As t → ∞, → 1 and since ln is continuous at 1, we can use the result from
t+3
Continuity III which allows us to say that
t + 2 t+2
lim ln = ln lim = ln 1 = 0.
t→∞ t+3 t→∞ t+3
(We will use this argument lots of times on thisZ sheet, so it is really important that
∞
1 3
you try to understand it fully.) Consequently dx = ln .
0 x2 + 5x + 6 2
(c) Z ∞ Z t h it
cos x dx = lim cos x dx = lim sin x = lim sin t.
0 t→∞ 0 t→∞ 0 t→∞

But this limit does not exist, because sin keeps oscillating between −1 and 1. Hence
the integral is divergent.
(d) We have Z ∞ Z t
−x
e cos x dx = lim e−x cos x dx.
0 t→∞ 0
Z t
We will determine I(t) = e−x cos x dx using integration by parts.
0
Let f (x) = e−x and g ′ (x) = cos x. Then f ′ (x) = −e−x and g(x) = sin x. Hence
Z t
 −x t
I(t) = e sin x 0 + e−x sin x dx.
0

Now we evaluate the second integral by integrating by parts again. Let h(x) = e−x
and k ′ (x) = sin x. Then h′ (x) = −e−x and k(x) = − cos x. So
Z t
 −x t  −x t
I(t) = e sin x 0 + −e cos x 0 − e−x cos x dx.
0

67
 Consequently t  t
2I(t) = e−x sin x 0 + −e−x cos x 0


and so
1
I(t) = (e−t sin t − e−t cos t + 1).
2
Now for any t, −e−t ≤ e−t sin t ≤ e−t and −e−t ≤ e−t cos t ≤ e−t . Furthermore
lim e−t = 0, so we can apply the Sandwich Theorem to see that
t→∞

lim e−t sin t = lim e−t cos t = 0.

t→∞ t→∞

1
Hence lim I(t) = and so
t→∞ 2
Z ∞
1
e−x cos x dx = lim I(t) = .
0 t→∞ 2

(e) x2 ln x is undefined at 0 but continuous on (0, 2]. So

Z 2 Z 2
2
x ln x dx = lim+ x2 ln x dx.
0 t→0 t
Z 2
We will determine x2 ln x dx using integration by parts. Let f (x) = ln x and
t
1 x3
g ′ (x) = x2 . Then f ′ (x) = and g(x) = . So
x 3
Z 2 h x3 2 2 h x3 i2 h x3 i2 8
x t3 8 t3
i2 Z
x2 ln x dx = ln x − dx = ln x − = ln 2− ln t− + .
t 3 t t 3 3 t 9 t 3 3 9 9
  t3 t3
Using the hint we know that lim+ ln t = 0 and we know that lim+ = 0.
t→0 3 t→0 9
Hence Z 2 Z 2
2 8 8
x ln x dx = lim+ x2 ln x dx = ln 2 − .
0 t→0 t 3 9

(f) tan x is undefined at x = π/2 but continuous on [0, π/2), so

Z π/2 Z t  t
tan x dx = lim − tan x dx = lim − ln | sec x| 0 = lim − ln | sec t|.
0 t→π/2 0 t→π/2 t→π/2

As t → π/2− , cos t → 0 but cos t > 0, whenever 0 ≤ t < π/2. Hence as t → π/2− ,
sec t → ∞ and so ln | sec t| → ∞. Consequently the integral is divergent.
e1/x
(g) 2 is undefined at 0, but continuous on (0, 1]. Hence,
x
Z 1 1/x Z 1 1/x
e e
dx = lim dx.
0 x2 t→0+ t x2

68
 1
e1/x 1 1
Z
To determine dx, we substitute u = 1/x. du = − 2 , so “ 2 dx = −du”.
t x 2 dx x x
1
When x = t, u = , and when x = 1, u = 1. Thus
t
Z 1 1/x Z 1
e  1
2
dx = − eu du = − eu 1/t = e1/t − e.
t x 1/t

As t → 0+ , 1/t → ∞. Since exp is continuous, as t → 0+ , e1/t → ∞. Consequently

Z 1 1/x
e
lim+ dx = ∞ and so the integral is divergent.
t→0 t x2
(h) The range of this integral is infinite at both ends, so following the definition of such
an integral, we must split it into two integrals. We have
Z ∞ Z 0 Z ∞
1 1 1
2
dx = 2
dx + 2
dx,
−∞ x + 4 −∞ x + 4 0 x +4
providing both of these integrals exist. We have
Z ∞ Z t
1 1 h1
−1 x
 it 1   π
−1 t
dx = lim dx = lim tan = lim tan = .
0 x2 + 4 t→∞ 0 x2 + 4 t→∞ 2 2 0 t→∞ 2 2 4
Furthermore
Z 0 0
1 1 h1
−1 x
Z  i0
2
dx = lim 2
dx = lim tan
−∞ x +4 t→−∞ t x +4 t→−∞ 2 2 t
 1  t  π
= lim − tan−1 = .
t→−∞ 2 2 4
Hence ∞
1 π π π
Z
dx = + = .
−∞ x2 +4 4 4 2

(i) This integral is improper in two ways. First, the range of integration is infinite and
1
second, √ is undefined at 0. We have
x(x + 1)
Z ∞ Z 1 Z ∞
1 1 1
√ dx = √ dx + √ dx,
0 x(x + 1) 0 x(x + 1) 1 x(x + 1)
providing both of these integrals exist.
Now Z 1 Z 1
1 1
√ dx = lim+ √ dx.
0 x(x + 1) t→0 t x(x + 1)
Z 1
1 √
We evaluate √ dx using integration by substitution. Let u = x.
t x(x + 1)
1 1 √
Then du = √ and so “ √ dx = 2 du”. When x = t, u = t and when x = 1,
dx 2 x x
u = 1. Hence
√
Z 1 Z 1
1 1 1
du = 2 tan−1 u √t = π/2 − 2 tan−1 t.

√ dx = 2 √ 2
t x(x + 1) t u +1

69
 √
As t → 0+ , t → 0. Since tan−1 is continuous,
√ √

lim+ tan−1 t = tan−1 lim+ t = tan−1 0 = 0.
t→0 t→0

Hence
1 1
1 1
Z Z
√ dx = lim+ √ dx = π/2.
0 x(x + 1) t→0 t x(x + 1)
Similarly
∞
1 1 t √t
Z Z
−1

√ dx = lim √ dx = lim 2 tan u 1
1 x(x + 1) t→∞ 1 x(x + 1) t→∞
−1
√
= lim 2 tan t − π/2.
t→∞
√
As t → ∞, t → ∞ and because tan−1 is continuous we have
√
lim tan−1 ( t) = lim tan−1 u = π/2.
t→∞ u→∞

Thus ∞ t
1 1
Z Z
√ dx = lim √ dx = π/2.
1 x(x + 1) t→∞ 1 x(x + 1)
Consequently
∞
1
Z
√ dx = π/2 + π/2 = π.
0 x(x + 1)

3. How should we tackle a problem like this with a parameter? The usual idea is just to
work through the problem treating the parameter as if it were a number. At some point
we might see that different things will happen for different values of the parameter, so
we might end up with a number of different cases.
(a) Clearly there is no problem if p ≥ 0. If p < 0 then xp is not defined at x = 0. In
that case Z 1 Z 1
p
x dx = lim+ xp dx.
0 t→0 t
Now if p 6= −1,
1 h xp+1 i1  1 tp+1 
Z
lim+ xp dx = lim+ = lim+ − .
t→0 t t→0 p+1 t t→0 p+1 p+1
Whether or not this limit exists depends on whether or not lim+ tp+1 exists. From
t→0
lectures on limits we know that this limit exists (and equals zero) if and only if
p + 1 ≥ 0, or equivalently if and only if p ≥ −1. However we are not currently
considering the case p = −1 because in that case the integral has a different form.
So we know that the integral exists if p > −1 and does not exist if p < −1.
If p = −1, Z 1
1
xp dx = lim+ ln x t = lim+ (− ln t) = ∞.

lim+
t→0 t t→0 t→0

Hence the integral is divergent if p = −1.

Consequently the integral is convergent if and only if p > −1 when it is equal to
1
.
p+1

70
 (b) When p 6= −1 we have
Z ∞ Z t h xp+1 it  tp+1
p p 1 
x dx = lim x dx = lim = lim − .
1 t→∞ 1 t→∞ p + 1 1 t→∞ p + 1 p+1
Using results from lectures on limits, we know that this limit exists (and equals
zero) if and only if p + 1 < 0, or equivalently if and only if p < −1. Again we are
not considering the case when p = −1 at this point, so we have shown that the
integral converges if p < −1 and diverges if p > −1.
When p = −1, Z t
t
xp dx = lim ln x 1 = lim ln t = ∞.

lim
t→∞ 1 t→∞ t→∞

Consequently the integral is convergent if and only if p < −1 when it is equal to

1
− .
p+1
(c) We have Z ∞ Z t
1 1
p
dx = lim dx.
e x(ln x) t→∞ e x(ln x)p
Z t
1 1 1
We evaluate dx by substituting u = ln x. Then du = and so “ dx =
e x(ln x)
p dx x x
du”. When x = e, u = 1 and when x = t, u = ln t. Hence
Z t Z ln t
1 1
p
dx = dx.
e x(ln x) 1 xp

If p 6= 1 then
ln t h x1−p iln t (ln t)1−p
1 1
Z
p
dx = = − .
1 x 1−p 1 1−p 1−p
As t → ∞, ln t → ∞ and so lim (ln t)1−p exists if and only if 1 − p < 0, or
t→∞
equivalently if p > 1. So we have shown that the integral converges if p > 1 and
diverges if p < 1.
If p = 1 then
Z ln t
1  ln t
lim dx = lim ln x 1
= lim ln ln t = ∞.
t→∞ 1 xp t→∞ t→∞

So the integral diverges if p = 1.

1
Consequently the integral converges if and only if p > 1, when it is equal to .
p−1
2 2
4. (a) If x ≥ 1, x2 ≥ x and since e > 1, we have ex ≥ ex . Consequently e−x ≤ e−x .
(b) We have
Z ∞ Z t
−x
t
e−x dx = lim −e−x 1 = lim e−1 − e−t ) = e−1 .

e dx = lim
1 t→∞ 1 t→∞ t→∞

71
 Z ∞
2
(c) We can’t say anything about the precise value of e−x dx, but it should seem
1
2
reasonable that the integral will converge because 0 ≤ e−x ≤ e−x for x ≥ 1.
5. (a) We have
Z ∞ Z y h e−st iy  1 e−sy 
−st
F (s) = e dt = lim e−st dt = lim − = lim − .
0 y→∞ 0 y→∞ s 0 y→∞ s s
Z ∞
This derivation is only valid for s 6= 0. If s = 0 then we have F (0) = 1 dt and
0
it is easy to see that this is not convergent.
So for the integral to converge we need lim e−sy to exist. From our results on
y→∞
limits, we know that this limit exists if and only if s ≥ 0. But s = 0 is already
excluded for convergence and so F (s) exists if and only if s > 0, in which case it
 1 e−sy  1
equals lim − = .
y→∞ s s s
(b) We have
Z ∞ Z y h et(1−s) iy  ey(1−s)
t −st 1 
F (s) = e e dt = lim et(1−s) dt = lim = lim − .
0 y→∞ 0 y→∞ 1 − s 0 y→∞ 1−s 1−s
This derivation is only valid for s 6= 1. If s = 1 then we have
Z ∞ Z ∞
t −t
F (1) = e e dt = 1 dt
0 0

and this is divergent.

For this integral to converge we need lim ey(1−s) to exist. From our results on
y→∞
limits, we know that this limit exists if and only if 1 − s ≤ 0 or equivalently s ≥ 1.
But s = 1 is already excluded for convergence, so F (s) exists if and only if s > 1,
 ey(1−s) 1  1
in which case it equals lim − = .
y→∞ 1−s 1−s s−1
(c) We have Z ∞ Z y
−st
F (s) = te dt = lim te−st dt.
y→∞
Z 0∞ 0

When s = 0, F (s) = F (0) = t dt which is easily shown to be divergent.

0 Z y
Suppose that s 6= 0. We will compute te−st dt using integration by parts. Let
0
′ −st ′ e−st
g(t) = t and h (t) = e . Then g (t) = 1 and h(t) = − . So
s
Z y h te−st iy Z y e−st h te−st iy h e−st iy
−st ye−sy e−sy 1
te dt = − + dt = − − = − − 2 + 2.
0 s 0 0 s s 0 s2 0 s s s
ye−sy e−sy
Now if s < 0 then lim = ∞ and lim = ∞.
y→∞ s2 y→∞ s
 ye−ys e−sy 1 1
If s > 0 then lim − − + 2 = 2.
y→∞ s s s s
1
Hence F (s) exists if and only if s > 0, in which case it equals .
s2

72
 eiθ + e−iθ
6. We have eiθ = cos θ + i sin θ and e−iθ = cos θ − i sin θ. Consequently cos θ =
2
eiθ − e−iθ
and i sin θ = . Now let θ = ix to get
2
2x 2x
ei + e−i ex + e−x
cos(ix) = = = cosh x
2 2
and
2x 2x
ei − e−i ex − e−x
−i sin(ix) = − = = sinh x.
2 2

7. (a) This is a standard integral, we have

Z 1
1 h
−1 x
 i1
−1 1
  1 p 
2
dx = sinh = sinh = ln + 5/4 .
0 x +4 2 0 2 2

√
(b) We substitute u = ex . If we do this then the denominator becomes u2 + 1 and
so we can be hopeful of turning it into one of our standard integrals. We have
du
= ex and hence “du = ex dx”. When x = 0, u = 1 and when x = 1, u = e.
dx
Consequently
Z 1
ex
Z e
1 h ie  e + √e2 + 1 
√ dx = √ du = sinh−1 (u) = ln √ .
e2x + 1 1 + u 2 1 1+ 2
0 1

(c) The function we are integrating contains an absolute value. Our first aim should
be to remove this by setting up a piecewise definition. We have
(
2 2
x − x2  = x − x if x − x ≥ 0,
 
x2 − x if x − x2 < 0.

In order to make use of this piecewise definition, we must determine when x−x2 ≥ 0.
First we determine when x − x2 = 0. We have x − x2 = x(1 − x) so x − x2 = 0 if
x = 0 or x = 1. We split up the range [−1, 2] into intervals bounded by the roots
of x − x2 = 0 to get the intervals [−1, 0), (0, 1) and (1, 2]. On [−1, 0), x − x2 < 0,
on (0, 1), x − x2 > 0 and on (1, 2], x − x2 < 0. We also have x − x2 = 0 when x = 0
or when x = 1. Hence
(
2
x − x  = x − x if x ∈ [0, 1],
2
 
x2 − x if x ∈ [−1, 0) ∪ (1, 2].

We split the range of integration into three parts to get

Z 2 Z 0 Z 1 Z 2
x − x2  dx = 2 2
(x2 − x) dx
 
(x − x) dx + (x − x ) dx +
−1 −1 0 1
h x3 x 2 i0 h x2 3 i1
x h x3 x2 i2 11
= − + − + − = .
3 2 −1 2 3 0 3 2 1 6

73
(d) We substitute u = x5 . Why should we do this? First, notice that the denominator
of the integral is x10 + 16 so this substitution will change this to u2 + 16 which is
a familiar function to have in the denominator. Also the numerator is a constant
du
times which means that it will disappear when we do the substitution. We have
dx
du 1
= 5x4 and so “x4 dx = du”. Hence
dx 5
x4 1 1 1 1 −1 u 1  5
−1 x
Z Z  
dx = du = × tan + c = tan + c,
x10 + 16 5 u2 + 16 5 4 4 20 4
where c is a constant.
(e) We have
π/4 π/4 π/4
1
Z Z Z
2 2 2
cos t tan t dt = sin t dt = (1 − cos(2t)) dt
0 0 0 2
ht sin 2t iπ/4 π 1
= − = − .
2 4 0 8 4

(f) Since we have a product where it is simple to either integrate or differentiate both
parts of the product, we try to integrate by parts. Let f (x) = sin(4x) and let
g ′ (x) = cos(3x). (Doing this the other way round should work too.) Then f ′ (x) =
sin(3x)
4 cos(4x) and g(x) = and so
3
Z π/3
sin 3x iπ/3 4 π/3
h Z
sin(4x) cos(3x) dx = sin(4x) − cos(4x) sin(3x) dx
π/4 3 π/4 3 π/4
4 π/3
Z
=− cos(4x) sin(3x) dx.
3 π/4

Now we integrate by parts again, setting f (x) = cos(4x) and g ′ (x) = sin(3x). (This
time we have to be careful to choose f and g ′ this way round, if we made the initial
choice as we have done here, else we end up back where we started.) We have
cos(3x)
f ′ (x) = −4 sin(4x) and g(x) = − and so
3

4 π/3 4h cos(3x) iπ/3 16 π/3

Z Z
− cos(4x) sin(3x) dx = cos(4x) + sin(4x) cos(3x) dx.
3 π/4 3 3 π/4 9 π/4

The final term is just the integral we started with so rearranging gives
π/3
7 4
Z
π/3
sin(4x) cos(3x) dx = − cos(4x) cos(3x) π/4 .
9 π/4 9

So
π/3
2 √
Z
sin(4x) cos(3x) dx = ( 2 − 1).
π/4 7

74
 1 1
Z Z
(g) We have √ dx = p dx. Substituting u = x + 2 yields
x2 + 4x + 8 (x + 2)2 + 4
1 1 u x + 2
Z Z
p dx = √ du = sinh−1 + c = sinh−1 + c,
(x + 2)2 + 4 u2 + 4 2 2
where c is a constant.
(h) We multiply the top and bottom of the fraction by ex to get
1 ex
Z Z
dx = dx.
ex − e−x e2x − 1
The denominator of the integral is a composition f (g(x)) where g(x) = ex and
du
f (x) = x2 − 1. So we substitute u = ex . Then = ex . Hence “ex dx = du”. So
dx
ex 1
Z Z
2x
dx = 2
du.
e −1 u −1
In the last integral the denominator factorises as (u − 1)(u + 1), so we can use
partial fractions to compute the integral. We have
1 1 A B
= = + ,
u2 −1 (u − 1)(u + 1) u−1 u+1
where A and B are constants to be determined. Multiplying out gives
1 = A(u + 1) + B(u − 1).
1
By substituting u = 1, we get 1 = 2A and so A = . Similarly substituting u = −1
2
1
gives 1 = −2B and so B = − . Hence
2
1 1 1
= − .
u2 −1 2(u − 1) 2(u + 1)
So
1 1 1  1 1
Z Z 
2
du = − = ln |u − 1| − ln |u + 1| + c
u −1 2(u − 1) 2(u + 1) 2 2
1  u − 1  1  ex − 1 
= ln   + c = ln  x  + c,
2 u+1 2 e +1
where c is a constant.
(i) We have
√ √ √ √
1 1 x+1+ x x+1+ x
Z Z Z
√ √ dx = √ √ √ √ dx = dx
x+1− x x+1− x x+1+ x x+1−x
√ √  2 2 3/2
Z 
= x + 1 + x dx = (x + 1)3/2 + x +c
3 3
2 
= (x + 1)3/2 + x3/2 + c,
3
where c is a constant.

75
 (j) One part of this function is a composition because etan (y) = f (g(y)) where g(y) =
−1

tan−1 y and f (y) = ey . So we should try to substitute u = tan−1 y.

1 1
We have du = and “du = dy” which looks good.
dy 1 + y2 1 + y2
Hence Z tan−1 (y)
e
Z
dy = eu du = eu + c = etan (y) + c,
−1

1+y 2

where c is a constant.
(k) We have
x x 1
Z Z Z Z
2
x sin x dx = (1 − cos(2x)) dx = dx − x cos(2x) dx.
2 2 2
We evaluate the second integral using integration by parts with f (x) = x and
sin(2x)
g ′ (x) = cos(2x). Hence f ′ (x) = 1 and g(x) = . Consequently
2
x2 x sin(2x) 1 x2 x sin(2x) cos(2x)
Z Z
2
x sin x dx = − + sin(2x) dx = − − + c,
4 4 4 4 4 8
where c isZa constant. Z
x x
(l) We have ex+e dx = ex ee dx. Again we have a composition so we substitute

for the “inner” part of the composition, i.e. we substitute u = ex . Then du = ex

dx
and so “du = ex dx”.
Consequently
Z Z Z
x+ex x ex x
e dx = e e dx = eu du = eu + c = ee + c,

where c is a constant.
(m) There are lots of ways in which this integral can be written as a composition but
it looks a reasonable idea to try to substitute u = ex and see what happens. We
have du = ex and hence “du = ex dx” So we have
dx
1 ex 1
Z Z Z
x
dx = x x
dx = du.
1+e e (1 + e ) u(u + 1)
This last integral is found using partial fractions. We have
1 A B
= + .
u(u + 1) u u+1
Multiplying out gives
1 = A(u + 1) + Bu.

Setting u = 0 gives A = 1 and setting u = −1 gives B = −1.

Hence
1 1 1 1
Z Z Z Z
x
dx = du = du − du
1+e u(u + 1) u u+1
= ln |u| − ln |u + 1| + c = x − ln |1 + ex | + c,
where c is a constant.

76
 Z Z Z
2
(n) We have ln(x − 1) dx =
ln(x − 1) dx + ln(x + 1) dx.
Z
Now we remember how to find ln x dx using integration by parts with f (x) = ln x
Z
′
and g (x) = 1. We get ln x dx = x ln x − x + c, where c is a constant.
Now return to the original integral. Substitute u = x − 1 in the first integral and
y = x + 1 in the second integral to give
Z Z Z
2
ln(x − 1) dx = ln u du + ln y dy = u ln u − u + y ln y − y + c

= (x − 1) ln(x − 1) + (x + 1) ln(x + 1) − 2x + c,

where c is a constant.
1
(o) We will integrate by parts with f (r) = ln r and g ′(r) = r 4 . Then f ′ (r) = and
r
r5
g(r) = . Hence
5
3 h r 5 ln r i3 3
r4 h r 5 ln r i3 h r 5 i3 243 ln 3 242
Z Z
4
r ln r dr = − dr = − = − .
1 5 1 1 5 5 1 25 1 5 25

8. We have
Z ∞ Z t
x C  x C  h1
2
it
− dx = lim − dx = lim ln(x + 1) − C ln |x + 1|
0 x2 + 1 x + 1 t→∞ 0 x2 + 1 x + 1 t→∞ 2 0
1   √t2 + 1 
= lim ln(t2 + 1) − C ln(t + 1) = lim ln .
t→∞ 2 t→∞ (t + 1)C
 √t2 + 1   √t2 + 1 
We need to establish lim . If this is finite and non-zero then lim ln
t→∞ (t + 1)C t→∞ (t + 1)C
√
 t2 + 1  √
 t2 + 1 
exists, if it is zero then lim ln = −∞ and if it is ∞ then lim ln = ∞.
t→∞ (t + 1)C t→∞ (t + 1)C
We have 
√ p 1 if C = 1,

1 + t2 1 + 1/t2
= → 0 if C > 1, as t → ∞.
(t + 1)C (1 + 1/t)(t + 1)C−1 
∞ if C < 1

Z ∞
x C 
Hence − dx = 0 if C = 1 and otherwise fails to converge.
0 x2 + 1 x + 1

77
Chapter 4

Real Series

4.1 Lecture 12: Series

Reference: Stewart Chapter 12.2, Pages 723–730. (Edition?)
In this chapter we learn about one very special type of sequence, derived from another sequence
by adding together the terms one by one. This new sequence is called a series.

In this first release of this

Chapter it is suggested that you
ignore the contents of
Subsection 4.1.1.
Start at Sunsection 4.1.2.

4.1.1 A Tale of a Rabbit and a Turtle following Zeno’s

A snapping turtle flies the scene of its attack on a wee bonnie kit outside the cave of Caer-
bannog. Five minutes later, coming out of its cave, the killer rabbit, having seen its injured
offspring, give chase to the turtle. Will the mighty battle between the killer rabbit and the
snapping turtle take place? If so, when and where? Zeno1 pointed out that it is possible to
‘show’ that the rabbit will never catch the turtle OR that the rabbit will catch the turtle.
Mathematics helps to reconcile the point of view, though the final word is still debated today.
We assume that the rabbit pursue the turtle in a straight line (1-D problem) and that both
animals are traveling at constant speed, say vr for the rabbit, vt for the turtle. We let the
rabbit loose when the tortoise is at l0 distance (vt for 5’) and start time when it gives chase.
Using continuous time and space, if there is a meeting, we can write the following system
1
Zeno of Elea (ca.490BC-ca.430BC) was a Greek Philosopher. He lived in the Greek ‘colony’ of Elea in
Southern Italy (modern Velia). Although we know that his contributions to Dialectic were important, none
of his writing survived intact to our times. He is best known for his paradoxes, still debated nowadays.

78
of equations in terms of the place of the meeting L and its time T. The rabbit will reach
L = T · vr and the turtle L = l0 + T · vt . This is a linear system of equations,

L − vr · T = 0,
L − vt · T = l + 0.

Solving the system we see that there is a unique solution

vr l0 l0
L= , T = .
vr − vt vr − vt
Another way to look at the problem is to create a discrete process. So the rabbit will reach
x = l0 at time t1 = l0 /vr . At that time, the turtle will have traveled a further l1 = t1 · vt . At
the next stage, t2 = l1 /vr and l2 = t2 · vt , and so one. We get sequences of time tn = ln−1 /vr
and length ln = tn · vt . Using induction, we find

ln = l0 · (vt /vr )n, tn = l0 /vr · (vt /vr )n−1, n ≥ 1.

This process is infinite and a ‘paradoxical’ conclusion is that the rabbit will never catch the
turtle. That point of view was used to discuss discrete versus continuous space and time. We
leave the metaphysics aside and shall see how to reconcile both point of view using geometric
∞
X 1
series. From school, recall that rn = (we shall soon revisit that formula). Thus, we
n=0
1 − r
find that the total time is
∞
X l0 1 l0
T = (lo /vr ) (vt /vr )n = · = .
k=0
vr 1 − vt /vr vr − vt

The rabbit travels

∞ ∞
!
X X 1 vr l0
Lr = l k = l0 1+ (vt /vr )k = l0 · = ,
k=0 k=1
1 − vt /vr vr − vt

and the turtle travels ∞ ∞

X X vt l0
Lt = lk = l0 · (vt /vr )k = .
k=1 k=1
vr − vt
Using infinite series we see that we recover the continuous time formulae.

4.1.2 Definition of a Series

If we add together the first terms of an infinite sequence {an }∞
n=1 , we get another sequence
{sn }∞
n=1 whose terms s n are

s1 = a1 , s2 = a1 + a2 , s3 = a1 + a2 + a3 , . . . , sn = a1 + a2 + · · · + an , . . .

This very special sort of sequence is what we will work with in this chapter.

79
Definition 4.1. Given a sequence {an }∞ ∞
n=1 , the corresponding series is the sequence {sn }n=1
of partial sums for which the n-th term is
n
X
sn = a1 + a2 + · · · an = ak .
k=1

∞
X
We denote the full series by ak .
k=1

Examples.
∞
1 1 1 X1 1
1. The infinite sum 1 + + + + . . . = is a series for the sequence { }∞
n=1 .
2 3 4 n=1
n n
∞ ∞
X (−1)n X cos(nπ)
2. A series can have alternating positive and negative terms like = ,
n=1
n n=1
n
∞
X cos(n)
or ‘random’ changes of sign like in (no π in the cosine).
n=1
n2

4.1.3 Convergent Series, Geometric Series

∞
X
We will be interested in whether the series ak converges. To do this we have to determine
k=1
whether the sequence {sn }∞ of partial sums sn = nk=1 ak converges.
P
n=1

∞
X
Definition 4.2. Given a series ak , let sn denotes the n-th partial sum. If {sn }∞
n=1 is con-
k=1
∞
X ∞
X
vergent, the series ak is convergent and the sum of the series is given by ak = lim sn .
n→∞
k=1 k=1
If the series does not converge, it is called divergent.

Remark 4.3. If the value of the sum depends on all the terms {an }∞ n=1 involved, the
convergence or divergence of P the series depends only on the entries in the tail of the
∞ ∞ P∞
sequence {an }n=1 . For example, if n=100 an converges then n=1 an also converges with the
sums related by ! !
X∞ X99 X∞
an = an + an .
n=1 n=1 n=100

As we will see, series where all the terms are positive (or all the terms are negative) can be
dealt with much more easily than sequences where some of the terms are positive and some
are negative. Weird things can happen with these series as the following example shows.

80
Cautionary Tale 4.4. Suppose we try to compute
1 1 1 1 1
1−
+ − + − + ...
2 3 4 5 6
by grouping together adjacent terms. So
∞
1 1 1  1 1 1 1 1 1 1 1 X 1
1− + − +· · · = 1− + − + − +· · · = + + +· · · = .
2 3 4 2 3 4 5 6 2 12 30 k=1
2k(2k − 1)
Alternatively since addition is commutative, we might hope to rearrange the terms to get
 1 1 1 1 1 1 1 1 ∞
5 13 7 X 8k − 3
1+ − + + − + + − +· · · = + + +· · · = .
3 2 5 7 4 9 11 6 6 140 198 k=1
2k(4k − 1)(4k + 3)
So, we apparently get the same sum being equal to two different series. The terms in the
second sum are all bigger than the corresponding terms in the first sum because
8k − 3 1
> , k ≥ 1.
2k(4k − 1)(4k + 3) 2k(2k − 1)
It can be shown that both series converge but the second one converges to a limit that is strictly
greater than the first one. This shows that if we rearrange the order of adding terms
in an infinite series with positive and negative terms then we may change the value
of the sum. When we are dealing with infinite sums, the rule that addition is commutative
is no longer always valid and extreme care is required.

We have seen that the following important series is actually one of the only ‘easily’ computable
series.
Definition 4.5. Let a 6= 0 and r be real numbers. They determine a geometric series (of
common ratio r) equal to
∞
X
2 n−1
a + ar + ar + · · · + ar +··· = ar n−1.
n=1

∞
X
Note that this series if often written as ar n, starting at n = 0, to simplify the notation.
n=0

Theorem 4.6. If |r| < 1, then

∞
X a
ar k−1 = .
k=1
1−r
∞
X
If |r| ≥ 1, then ar k−1 is divergent.
k=1

Proof. The proof of this result is basically a calculation P

that needs to be known. When r = 1,
the series is obviously divergent. When r 6= 1, let sn = nk=1 ar k−1. Then,
n n
!
X X
rsn = r ar k−1 = ar k = sn − a + ar n = sn − a(1 − r n ),
k=1 k=1
n
a(1 − r )
so, re-arranging to sn , sn = . We can then conclude taking the limit as n → ∞.
1−r

81
Examples.
X
1. To see if the series 22k 31−k converge or diverge, we can write it as the geometric
∞  k k=1
X 4
series 3 . From Theorem 4.6, the series diverges.
k=1
3
∞
X xk
2. For what values of x does the series converge? Here x is variable, but, again,
k=1
2k
x
this series is a geometric series with common ratio . The series converges if and only
x 2
2x
if   < 1, i.e. if |x| < 2 or, equivalently, if −2 < x < 2. In that case its sum is .
 
2 2−x

82
4.2 Lecture 13: Important Series
Reference: Stewart Chapter 12.2, Pages 727–728, Chapter 12.3, Pages 733–736, Chapter 12.6,
Pages 750–754. (Edition?)

4.2.1 A Criterion for Divergence

There is an easy test every convergent sequence must satisfy. This is a necessary test for
convergence (or, when it fails, a sufficient test for divergence).
∞
X
Theorem 4.7 (Divergence Criterion). If ak is convergent, then lim an = 0, that is, it is
n→∞
k=1
necessary for a convergent series that the sequence {an }∞
n=1 converges to 0.

Proof.PAs has already been stated, ∞ ∞

P
n=1 an converges means that {sn }n=1 converges where
sn = nk=1 ak . If s = limn→∞ sn , note that an = sn − sn−1 , then

lim an = lim (sn − sn−1 ) = s − s = 0.

n→∞ n→∞

∞
X
The converse of the theorem fails. It is not necessarily true that if lim an = 0, then ak is
n→∞
k=1
convergent. We shall see examples later where this happens. But it is a cheap test.

Examples.

1 P∞ (−1)n 1
1. The series ∞ and ∞
P P
n=1 , n=1 n=1 2 all satisfy the criterion, but we shall see
n n n
that the first one diverges, but the second and third will converge.
P∞
2. It is not clear if the series n=1 sin(n) converges or not. It sums positive and neg-
ative numbers in a random fashion. But we know it must be divergent because
limn→∞ sin(n) is not determined (hence not 0).

3. In Theorem 4.6, it is clear that if |r| ≥ 1, then the series is divergent as limn→∞ |r|n 6= 0,
and may be convergent when |r| < 1 because, in this case, limn→∞ r n = 0.
n+3
. Then limn→∞ an = 1 6= 0 and so the series ∞
P
4. Let an = n=1 an diverges.
n+7
5. Let
n+3
3 −n2 +n
an = (−1)n .
n+7
it changes sign like (−1)n ). But, anyway, an
In that case, an changes sign (note that P
does not converge to 0, and so the series ∞ n=1 an diverges.

83
4.2.2 Telescopic Series
Let first consider an explicit example that will speak (hopefully) by itself.
∞
X 1
Example. How can we check if the series is convergent and find its sum?
k=1
k(k + 1)
n
X 1 1
Let sn = . We can rewrite using partial fractions. We have
k=1
k(k + 1) k(k + 1)

1 A B
= + ,
k(k + 1) k k+1

where A and B are constants to be determined. Multiplying out gives

1 = A(k + 1) + Bk.

Setting k = 0 gives 1 = A. Setting k = −1 gives 1 = −B so B = −1. Hence

1 1 −1
= + .
k(k + 1) k k+1

So,
n n
X 1 X 1 1 
sn = = −
k=1
k(k + 1) k=1 k k + 1
1 1 1 1 1 1 1 1 1 1
=1− + − + − + ...+ − + − =1− ,
2 2 3 3 4 n−1 n n n+1 n+1
hence lim sn = 1.
n→∞

This sort of sum is an example of what is called a telescoping sum because its terms cancel
each other, collapsing into almost nothing.
Pn
Definition 4.8. A finite sum sn = k=1 ak in which subsequent terms cancel each other,
leaving only the initial and final terms is called a telescopic sum.

Examples.

1. Given a finite set {ak }nk=1 of real numbers, the sum

n−1
X
sn = (ai − ai+1 ) = a1 − an
k=1

is telescopic.

84
 ∞
X
2. We have seen before that the series sin(n) is divergent. Remarkably we can use
n=1
telescoping sums to show such result. Using trigonometric identities, we have
N N
X X 1
sin(n) = csc(1/2) (2 sin(1/2) sin(n))
n=1 n=1
2
N  
1 X 2n − 1 2n + 1
= csc(1/2) cos( ) − cos( )
2 n=1
2 2
 
1 1 2N + 1
= csc(1/2) cos( ) − cos( ) .
2 2 2

We can then see that such sum has no limit as N → ∞.

4.2.3 Harmonic Series

∞
X 1 1 1 1
Theorem 4.9. The Harmonic Series =1+ + + + . . . diverges.
n=1
n 2 3 4

1
Proof. Let an = . Consider the following inequalities
n
a1 = 1 ≥ 1,
1 1
a1 + a2 = 1 + ≥ 1 + ,
2 2
1 1 1 1 1 1 2
a1 + a2 + a3 + a4 = 1 + + + ≥ 1 + + + =1+ ,
2 3 4 2 4 4 2
1 1 1 1 1 1 1 1 1 1 3
a1 + · · · + a8 = 1 + + . . . + + ≥ 1 + + + + + + + =1+ ,
2 7 8 2 4 4 8 8 8 8 2
··· ≥ ··· ,
1 1 1 1 1 n
a1 + · · · + a2n ≥ 1 + + + +···+ n +···+ n = 1+ .
2 4 4 2 2 2
n
X 1
This shows that we can make as large as we like by choosing n large enough. Conse-
k=1
k
quently the series cannot converge.

Cautionary Tale 4.10. Note that, even though lim an = 0, this series diverges. This is
n→∞
the standard example of a series that behaves in this way.

A similar technique to the above can be used to show that the series of the inverse of squares
of integers converges.

Lemma 4.11. The series ∞ 2

P
n=1 1/n converges and its sum is less than 2.

85
Proof. Let
2(n+1)
X−1      
1 1 1 1 1 1 1
tn = = 1+ 2 + 2+ +···+ 2 +···+ + · · · + (n+1)
k=1
k2 2 3 42 7 22n (2 − 1)2
2n
 
2 4 1 1
6 1+ 2 + 2 − − + · · · + 2n
2 4 4 9 2
1 1 1 5
= 1+ + 2 +···+ n −( )
2 2 2 36
(n+1)
1 − (1/2) 5 67 1
= −( )= − n < 2,
1 − (1/2) 36 36 2

where in the above we have used the formula for summing a geometric series. The sequence of
67
partial sums is hence increasing and bounded above by 36 ≈ 1.861 and so, by the monotone
convergence theorem it converges to a limit smaller than 2.

Remark 4.12. We shall see that, as a consequence of our analysis of Fourier series (see
MA2712), we can calculate the sum of the series:
∞
X 1 π2
= ≈ 1.645 . . . < 2.
n=1
n2 6

4.2.4 Algebra of Series

We now see that we can combine series like we have combined functions and sequences etc.
However, there are not so many ways in which we can combine series.
∞
X ∞
X
Theorem 4.13 (Algebra of Limits for Series). Suppose ak converges to a and bk
k=1 k=1
converges to b and c ∈ R is a constant. Then,
∞
X
1. (ak + bk ) converges to a + b;
k=1

∞
X
2. (ak − bk ) converges to a − b;
k=1

∞
X
3. cak converges to ca.
k=1

Proof. We will prove the first part. The other parts are left as Exercise 7 of Sheet 3a.
Let us define the sequences {sn }∞ ∞ ∞
n=1 , {tn }n=1 and {un }n=1 by

n
X n
X n
X
sn = ak , tn = bk , un = (ak + bk ).
k=1 k=1 k=1

86
Then un = sn + tn . Since limn→∞ sn = a and limn→∞ tn = b, using Theorem 4.13, we have

lim (sn + tn ) = lim sn + lim tn = a + b.

n→∞ n→∞ n→∞

Consequently,
∞
X
(ak + bk ) = lim un = lim (sn + tn ) = a + b.
n→∞ n→∞
k=1
∞
X
Therefore (ak + bk ) converges to a + b.
k=1

This result will be used to split complicated formula for ak into manageable parts.

87
4.3 Lecture 14: Test for Convergence
Reference: Stewart Chapter 12.3, Pages 733-739, Chapter 12.4, Pages 741-744 and Chapter
12.6, Pages 752-754. (Edition?)
Many series are impossible to sum exactly. Second best is to determine its convergence (or
divergence). There are tests that allow us to determine whether a series converges, but they
will not tell us anything about their sum. We shall see in Chapter 7 that many functions
can be represented by a series of monomial terms. Knowing that a series converges will allow
us to perform some manipulations on the terms of the series and hence on the function. For
instance, we should be able to integrate and differentiate these series and so to solve some
differential equations (see MA3610 in Level 3). We will look at four tests. Some tests may be
inconclusive on a given series. In that case we should use them one by one until a conclusion
is reached (all four might fail though).
Before we look into those tests, we prove an important lemma. Recall that bounded monotone
sequences converge. If the terms are positive, the sequence of partial sums of a series must
be increasing, and so if we can bound them a priori, the series must be convergent.
This leads to the following useful result where we can ignore the change of signs of some series.
∞
X ∞
X
Lemma 4.14. Suppose that the series |an | converges. Then, the series an converges.
n=1 n=1

n
X
Proof. Let bk = |ak | + ak . Note that 0 ≤ bk ≤ 2|ak |. Let sn = bk . Then, {sn }∞
n=1 is an
k=1
increasing sequence (because bk ≥ 0) and bounded because
n
X n
X ∞
X
0 ≤ sn = bk ≤ 2 |ak | ≤ 2 |ak |.
k=1 k=1 k=1
∞
X
Consequently, from Theorem 2.23, it converges. Now, ak = bk − |ak | and both bk and
k=1
∞
X ∞
X
|ak | converge. Therefore ak converges from Theorem 4.13.
k=1 k=1
P∞
Remark 4.15. We shall see in Section 5.3 that a convergent series n=1 an such that
P∞
n=1 |an | does not converge behaves quite extraordinarily.

Examples.
∞   ∞
X cos(n)  ≤ 1 and the series
 cos(n)  X 1
1. The series 2
converges because 
2 2 2
converges
n=1
n  n  n n=1
n
(Lemma 4.14 and the Comparison Test). But its sum is difficult to calculate.
∞
X cos(nπ)
2. We cannot apply Lemma 4.14 to because the harmonic series diverge. This
n=1
n
 
 cos(nπ) 
n
follows because cos(nπ) = (−1) and so    = 1 . We shall see that this series
n  n
has some interesting, although unexpected, properties.

88
4.3.1 Comparison Test
Finally, we can use our results about monotone sequences to give the following existence result.
Proposition 4.16. Suppose that {an }∞ n=1 is a sequence of positive numbers, then the series
P ∞
n=1 an converges if and only if the sequence of partial sums is bounded.

Proof. If the terms of the series are positive, its sequence of partial sums is monotonically
increasing, and so, from the Monotone Convergence Theorem 2.23, it has a limit if and only
if it is bounded.

Given the above comments we restrict attention to non-negative sequences {an }∞n=1 . Hence we
establish convergence or divergence by testing whether or not the sequence of partial sums
{sn }∞n=1 is bounded. This is usually done by comparing the given series with a ‘simpler’
series which is known to converge or diverge. We make this precise in the following result.
Theorem 4.17 (Comparison Test). Let {an }∞ ∞
n=1 and {bn }n=1 be real sequences of numbers.
Suppose that there exists an N > 0 such that 0 6 an 6 bn for all n ≥ N. Then,

(i) if ∞
P P∞
n=1 bn converges, then n=1 an converges,
P∞ P∞
(ii) if n=1 an diverges, then n=1 bn diverges.

Proof. Recall that convergence or divergence do not depend

P∞ on the first N terms. Without
limiting the generality
P∞ we can assume that N = 1. If n=1 Pb∞
n converges then the sequence of
partial sums of n=1 an is bounded, hence it converges. If n=1 an diverges then the sequence
of partial sums of {an }∞
n=1 is unbounded.
P∞From the inequality the sequence of partial sums of
∞
{bn }n=1 is also unbounded and hence n=1 bn diverges.

Examples.
n+3 X
1. Let an = . We can compare the series an with the harmonic series because
n2 + 7 n→∞
1 X
an ≥ for n ≥ 3. Hence the series an diverges because the harmonic series diverges.
n n→∞

n+3 1 2
2. Let now an = 3 . We cannot compare an with 2 directly, but an ≤ 2 for n ≥ 3.
n +X7 n n
P∞ 2
And so the series an converges by comparison with n=1 (2/n ).
n→∞

The next result is an equivalent test, but often easier to use.

∞
X ∞
X
Theorem 4.18 (Limit Comparison Test). Suppose that an and bn are series of positive
n=1 n=1
an
terms and lim = l where l > 0 (and is finite). Then, either both series converge, or both
n→∞ bn
diverge together.

This test follows from Lemma 4.14 and Theorem 2.23. We shall look at its proof in Chapter
5.

89
4.3.2 Integral Test
We now consider a test which can be used for series of the type
∞
X 1
n=1
np

1 1
for fixed p > 0. When p ≤ 1, we see that np ≤ n, ≤ p , and so the series diverges using
n n
2 p 1 1
the Comparison Test. When p ≥ 2, we have n ≤ n , p ≤ 2 , and so the series converges
n n
using the Comparison Test. When 1 < p < 2 we cannot conclude. P We need another
∞
idea. The next test that we will consider applies to series of the form n=1 f (n) where
f : [1, ∞) → [0, ∞) is a positive decreasing continuous function with f (x) → 0 as x → ∞. In
the above case the function f (x) = 1/xp has this property.

Theorem 4.19 (Integral Test). If f : [1, ∞) → [0, ∞) is a positive decreasing positive con-
tinuous function then the following quantities
∞
X Z ∞
s= f (n) and I = f (x) dx
n=1 1

either both converge or both diverge.

y = f (k − 1)

y = f (x)

y = f (k)

x=k−1 x=k

Figure 4.1: The area under the curve is less than the area under the line y = f (k − 1) and it
is greater than the area under the line y = f (k).

Proof. If f is decreasing then, for x ∈ [k − 1, k], we have

ak = f (k) 6 f (x) 6 f (k − 1) = ak−1 .

90
By considering the area under the curves of y = f (k), y = f (x) and y = f (k − 1), when
k − 1 6 x 6 k, we have
Z k Z k Z k
f (k) = f (k) dx 6 f (x) dx 6 f (k − 1) dx = f (k − 1),
k−1 k−1 k−1

see Figure 4.1. Note that both {sn }∞ ∞

n=1 and {In }n=1 are increasing sequences and

n
X Z 2 Z 3 Z n  n
X n−1
X
sn − f (1) = f (k) 6 + +···+ f (x) dx 6 f (k − 1) = f (k) = sn−1 .
k=2 1 2 n−1 k=2 k=1

That is,
sn − f (1) 6 In 6 sn−1 .
Hence, if {In }∞ ∞ ∞ ∞
n=1 converges then {sn }n=1 converges and if {In }n=1 diverges then {sn }n=1
diverges from the Comparison Test.

Remarks 4.20. As before, the convergent result will remain true if the hypothesis are true
for n, x on an interval [N, ∞) for some N large enough (see Exercise 6 of Sheet 3a). More-
over, integrating functions can be very difficult but, using the Limit Comparison Test, we can
simplify f (n) into an expression that can be integrated.

∞
X 1 1
Example. Check for which p > 0, is convergent. For the Integral Test, let an =
n=1
np np
1
and let f : [1, ∞) → R, f (x) = . Then f is continuous on [1, ∞), positive and decreasing.
xp
Furthermore an = f (n) and so we may apply the test. We have
 1−p
Z ∞
1
Z t
1  t /(1 − p), 0 < p < 1;
dx = lim dx = lim ln(t),
p = 1;
1 xp t→∞ 1 xp t→∞  1
1− 1
, p > 1. p−1 tp−1

The first two limit are infinity, but the last one is 0. Consequently the integral, and hence the
series, is divergent when p ≤ 1 and is convergent when p > 1.
∞
X
The next test can be used for series of the type f (n) where f is integrable basically on
n=1
the positive real line.

Cautionary Tale 4.21. Note that the sum of the series is not necessarily equal to the improper
∞
X 1 1 1
integral. After all, the series 2
starts 1 + + + · · · and consists of positive terms, so
n=1
n 4 9
its sum is larger than 1.5 (suming the first 10 terms). But the integral test gives
Z ∞ Z t
1 1 h 1 it  1
dx = lim dx = lim − = lim 1 − = 1.
1 x2 t→∞ 1 x2 t→∞ x 1 t→∞ t

91
4.4 Lecture 15: Further Tests for Convergence

4.4.1 Comparing a Series with a Geometric Series: the Root and

Ratio Tests
There are two tests for convergence involving comparing a given series with a geometric series.
First a theorem that is mainly due to d’Alembert2 .

Theorem 4.22 (d’Alembert’s Ratio Test). Let {an }∞

n=1 be a real sequence such that

lim |an+1 /an | = ρ1 .

n→∞

∞
X
1. If ρ1 < 1, the series an converges.
n=1

∞
X
2. If ρ1 > 1, the series an diverges.
n=1

3. If ρ1 = 1, we cannot conclude.

The second test is mainly due to Cauchy 3 .

Theorem 4.23 (Cauchy’s Root Test). Let {an }∞

n=1 be a real sequence such that

lim |an |1/n = lim n |an | = ρ2 .

p
n→∞ n→∞

∞
X
1. If ρ2 < 1, the series an converges.
n=1

∞
X
2. If ρ2 > 1, the series an diverges.
n=1

3. If ρ = 1, we cannot conclude.

Those two results are corollaries of the following lemma about the convergence/divergence of
series with general terms.

Lemma 4.24. Let {an }∞

n=1 be a sequence of terms such that there exists an N with,
2
J.L.R. d’Alembert(1717-1783) was a French mathematician. He was at the forefront of the mathematics
of his time, in particular in calculus and mathematical methods for physics where he was one of the first to
use PDEs. He worked on the Encyclopédie. He was a difficult man and fell out with many contemporaries.
At the end of his life, he devoted more time to philosophy and literature and became the Perpetual Secretary
of the Académie Française.
3
A.L. Cauchy (1789-1857) was a French mathematician who made analysis what it is today via his work on
convergence. The first analysis course mathematician are following today are still broadly based on his own
at the École Polytechnique in 1821. Cauchy was also very productive in all areas of analysis and differential
equations. His full life work covers 27 volumes.

92
 1. either
|an+1 /an | ≤ ρ1 < 1, n ≥ N,
or
|an |1/n ≤ ρ2 < 1, n ≥ N,
then the series converges.
2. If, either
|an+1 /an | ≥ ρ1 > 1, n ≥ N,
or
|an |1/n ≥ ρ2 > 1, n ≥ N,
then the series diverges.

Proof. This proof has been removed. It had a flaw. See the lecture and seminar notes.

∞
X xn
Example. We can determine the values of x for which is convergent. If the sum
n=1
n!
depends on a variable x like this one does, it is often a good idea to use the Ratio Test. We
xn
have an = and we check the hypotheses of the test. First, we need an to be non-zero
n!
an+1
and this is ok providing x 6= 0, so let’s assume x 6= 0 for the moment. Second, =
an
xn+1 /(n + 1)! x
= . So, we have
xn /n! n+1
a  |x|
 n+1 
lim   = lim = 0.
n→∞ an n→∞ n + 1

Using the Ratio Test, we see that the series converges providing x 6= 0. But the case when
x = 0 is easy because then every term of the series is zero and so it converges. Consequently
the series converges for all x.

Cautionary Tale 4.25. Notice that if l = 1 then the Ratio Test tells us nothing. Typical
examples are when we have rational functions (i.e. ratio of polynomials).

∞
X 1
Example. Check for which p > 0, is convergent applying the last two tests.
n=1
np

1. If we try to use the Ratio Test, we have

p
np
a  
 n+1  n
= = .
an (n + 1)p (n + 1)


Hence, because p is a constant (independent of n), the Ratio Test is inconclusive:

a   p
 n+1  an+1
lim   = lim = 1p = 1.
n→∞ an n→∞ an

93
 2. For the Root Test, consider
1 ln n
bn = ln(|an |1/n ) = ln(n−p ) = −p ,
n n
and so limn→∞ bn = 0 and the test is inconclusive because

lim |an |1/n = exp( lim bn ) = 1.

n→∞ n→∞

Remark 4.26. How do we know whether one of these tests will tell us about convergence?
We don’t for sure. However, if the general term of the series looks like something we could
integrate, then we should try the integral test. On the other hand, if the ratio of consecutive
terms of the series is simple enough, we should try the ratio test.

94
Summary of Chapter 3
We have seen:

• the definition of a series;

• the geometric and harmonic series;

• how to sum telescoping sums;

• an example of a series where the terms tend to zero but the series diverges;

• four tests for determining whether a series is convergent;

• that we must check carefully that it is appropriate to use either of the tests;
∞
X 1
• that the series p
, p ≥ 0, converges for p > 1 and diverges for p ≤ 1.
k=1
n

95
4.5 Exercise Sheet 3

4.5.1 Exercise Sheet 3a

1. Determine whether the following series converge. If they do, find the limit.
∞ ∞ ∞ ∞
X 3k X 1 X
3k
X 3
(a) 3k
. (b) . (c) (−1) . (d) .
2 (k + 1)(k + 2) 2k
k=1 k=1 k=1 k=3

2. In each of the following cases determine whether or not the series converges.
∞ ∞ ∞ √ ∞
X 1 X 4n2 − n + 3 X n+ n X 2
(a) n
. (b) 3
. (c) 3
. (d) n4 e−n .
n=1
2 +1 n=1
n + 2n n=1
2n − 1 n=1

3. For each of the following series, determine the values of x ∈ R such that the given series
converges.
∞
X xk
(a) .
k!
k=0

(b) In the following α ∈ R is not an integer.

∞  
X α(α − 1) · · · (α − k + 1) α(α − 1) 2
xk = 1 + αx + x +···
k! 2
k=0

∞
X k 3 xk
(c) .
k=0
3k
X∞
(d) k k xk.
k=0
X∞
(e) ak xk = 1 + 2x + x2 + 2x3 + x4 + · · · , i.e. with a2k = 1 and a2k+1 = 2 for
k=0
k = 0, 1, 2, · · · .
∞ √ 2
X x + k − |x|
(f) .
k=1
k2
∞  
X cos kx sin kx
(g) +3 2 .
k=1
k3 k

4. Show that
1 1 1 1
+ + +··· = .
1·3 3·5 5·7 2
5. Use the Integral Test to test the convergence of the following series for c > 0:
∞ ∞
X 1 X 1
(a) c
. (b) .
n=2
n(ln n) n=2
n(ln n)(ln ln n)c

6. Show that the Integral Test remain valid if its hypotheses on f are only true for some
interval [N, ∞), N ∈ N.

96
 7. Prove Parts 2 and 3 of Theorem 4.13.

8. Determine the values of x for which the following series converge. Find the sum of the
∞ ∞ ∞
X xk X
k
X
series for those values of x. (a) k
. (b) (x − 4) . (c) tank x.
k=1
3 k=1 k=1

9. Show that the Ratio Test is always inconclusive for series whose terms are rational
p(n)
functions, that is a ratio of polynomials an = for p, q some polynomials. Deduce
Pq(n)
by other means the convergence of the series ∞ n=1 an .

10. Get the details of the proof of the Root Test when ρ2 < 1.

4.5.2 Additional Exercise Sheet 3b

1. Determine whether or not the following series are convergent.
∞ ∞ ∞ ∞ ∞
X 1 X n X n+4 X 1 X (−3)n
(a) √ ; (b) ; (c) ; (d) ; (e) .
n=1
n+2 n=1
en n=1
n+1 n=2
n ln n n=1
n!

2. Determine which of the following series converge and for those series that do converge
find ∞
the sum. ∞ ∞ ∞ ∞
X 3 X 1 X 1 X k2 + 2 X 3k + 2k
(a) k
; (b) ; (c) 2 + 3k − 2
; (d) 3+3
; (e) k
.
k=1
5 k=1
(k + 2)(k + 3) k=1
9k k=1
k k=1
6
∞
X 7(n + 1)
3. Determine whether converges or diverges.
n=1
2n n
∞
X n4 + 3n2
4. Determine whether converges or diverges.
n=1
n9/2 + n

Short Feedback for Sheet 3a

1. (a) 3/5.
(b) 1/2.
(c) divergent.
(d) 3/4.

2. (a) convergent.
(b) divergent.
(c) convergent.
(d) convergent.

3. (a) x ∈ R.
(b) |x| < 1.
(c) |x| < 3.

97
 (d) x = 0.
(e) |x| < 1.
(f) x ∈ R. Compare with a power of k.
(g) x ∈ R.

4. The sum is telescopic.

5. (a) convergent when c > 1.

(b) convergent when c > 1.

6. The first N − 1 terms do not influence convergence of the full series.

7. Use the algebra of the limits of sequences.

x
8. (a) Converges when |x| < 3 to .
3−x
x−4
(b) Converges when 3 < x < 5 to .
5−x
π tan(x)
(c) Converges when x 6= ± + kπ, k ∈ Z, to .
4 1 − tan(x)
p(n + 1)
9. Show that for any polynomial p(x), lim = 1 and use the Comparison or
n→∞ p(n)
Integral Tests.

10. Use a similar comparison than for the Ratio Test.

Short Feedback for the Additional Exercise Sheet 3b

1. (a) Divergent;
(b) convergent;
(c) divergent;
(d) divergent;
(e) convergent.

2. (a) 3/4;
(b) telescopic;
(c) telescopic;
(d) divergent;
(e) 3/2.

3.

98
4.5.3 Feedback for Sheet 3a
3/8 3
1. It is a geometric series of ratio 3/8 and so it converges to its limit = .
1 − 3/8 5
We have a telescopic series. We decompose
1 a b 1 1
= + = − ,
(k + 1)(k + 2) k+1 k+2 k+1 k+2

because a(k + 2) + b(k + 1) = 1 for all k ∈ N. Therefore the terms from n = 3

cancel out and so the sum is 1/2.
(a) The series is divergent because (−1)k does not converge to 0.
(b)
∞ ∞
X 3 3 X 1 3 1/2 3
(c) We have k
= · k
= · = .
2 8 2 8 1 − 1/2 4
k=3 k=0

2. (a) We could show convergence here by using the Ratio or Root Tests or more simply
by using the Comparison Test by noting that
1 1
06 6 n.
2n +1 2
The upper bound is a term from a convergent geometric series.
4n2 − n + 3
(b) This is divergent. Using the Limit Comparison Test with an = and
n3 + 2n
bn = 1/n, we get

an n(4n2 − n + 3)
lim = lim nan = lim = 4,
n→∞ bn n→∞ n→∞ n3 + 2n
and so both series diverge together.
(c) This converges. Using the simple Comparison Theorem we see that
√
n+ n 2n
≤
2n3 − 1 2n3 − 1
2n
and, using the Limit Comparison Test with an = and bn = 1/n2, we get
2n3 − 1
an 2n3
lim= lim n2 an = lim = 1,
n→∞ bn n→∞ n→∞ 2n3 − 1

and so all series converge together.

2
(d) Using the Root Test with an = n4 e−n , we have
 2 1/n
lim a1/n
n = lim (n 1/n 4
) e−n = lim (n1/n )4 e−n = 0.
n→∞ n→∞ n→∞

Here the results is as a consequence of limn→∞ n1/n = 1 and limn→∞ e−n = 0. And
so, the series converges.

99
3. (a) Let ak = xk /k! and use the Ratio Test. We have
k+1
 
 ak+1 
lim 
  = lim |x| /(k + 1)! = lim |x| = 0.
k→∞ ak  k→∞ |x|k /k! k→∞ k + 1

From the Ratio Test, the series actually converges for all x ∈ R.
(b) Let ak = α(α − 1) · · · (α − k + 1)xk /k!. Using the Ratio Test
 
 ak+1 
lim   = lim α − k |x| = lim α/k − 1 |x| = |x|.
k→∞  ak  k→∞ k + 1 k→∞ 1 + 1/k

Thus the series converges if |x| < 1. If |x| > 1 then the terms of the series are
unbounded and thus the series diverges. What happens when x = −1 or x = 1
needs more refined tests to determine if the series converges or diverges and the
outcome depends on α. This will not be considered further here.
(c) The Root Test is the easiest test to use here. With ak = k 3 xk /3k we have

(k 1/k )3 |x| |x|

lim |ak |1/k = lim = .
k→∞ k→∞ 3 3
From the Root Test, the series converges if |x| < 3. It diverges if |x| > 3. If |x| = 3
then |ak | = k 3 and since these terms become unbounded, it follows that the series
diverges when x = ±3.
(d) The Root Test is the easiest test to use here. With ak = k k xk , we have
1/k
|ak | = |kx|.

This only converges if x = 0 and is unbounded for x 6= 0. Hence the series only
converges when x = 0.
(e) Let bk = ak xk . The Ratio Test does not give any information here as ak+1 /ak does
not have a limit as k → ∞. However we can still use the Root Test. Since
1/k
1 6 ak 6 2, 1 6 lim ak 6 lim 21/k = 1.
k→∞ k→∞

Thus,
1/k
lim |bk |1/k = lim ak |x| = |x|.
k→∞ k→∞

The series converges if |x| < 1 and diverges if |x| > 1. By inspection the series
diverges if x = 1 as the terms of the series do not tend to 0 as k → ∞. It can be
shown that the series also diverges when x = −1.
(f) Since x2 ≥ 0,
√
x2 + k − |x| (x2 + k) − x2 1 1
ak = = √ = √ 6 ,
k2 ( x2 + k + |x|)k 2 ( x2 + k + |x|)k k 3/2

6 1/k 3/2 , the series ∞

P
for all x ∈ R. Since 0 6 akP k=1 ak converges by comparison
∞ −3/2
with the convergent series k=1 k for all x ∈ R.

100
 (g) Let ak denotes the k-th term of the series. From the triangle inequality and the
fact that | sin kx| 6 1 and | cos kx| 6 1, we have
1 1
|ak | 63
+3· 2
k k
for all x ∈ R.PSince k=1 1/k 3 and ∞
P∞ P 2
k=1 1/k are standard convergent series, it
follows that ∞ k=1 |ak | converges from the Comparison Test. Hence the original
series converges for all x.

4. The series is telescopic. Indeed

1 1/2 1/2
= − ,
(2k − 1)(2k + 1) 2k − 1 2k + 1

and so the series is equal to

∞ ∞ ∞
X 1 1X 1 1X 1 1
= − = .
k=1
(2k − 1)(2k + 1) 2 k=1 2k − 1 2 k=1 2k + 1 2

5. (a) We use the Integral Test. We have to evaluate the convergence of the integral
Z ∞
dx
2 x(ln x)c

and check that the integrand is decreasing. The last claims follows readily from
1
the derivative of :
x(ln x)c
 ′
1 1 c
c
=− 2 c
− 2 .
x(ln x) x (ln x) x (ln x)c+1

It is always negative, so the integrand is decreasing. For the integral, let u = ln x,

then Z ∞ Z ∞
dx du
c
= c
.
2 x(ln x) ln 2 u

We know that the integral converges if and only if c > 1. Hence the series converges
if c > 1 and diverges otherwise.
(b) We proceed in the same way. The derivative of the integrand is
 ′
1 −1 −1 −c
c
= 2 c
+ 2 2 c
+ 2 ,
x(ln x)(ln ln x) x (ln x)(ln ln x) x (ln x) (ln ln x) x (ln x) (ln ln x)c+1
2

that is negative. The convergence of the integral is measured by

Z ∞ Z ∞ Z ∞
dx du dv
c
= c
= c
,
2 x(ln x)(ln ln x) ln 2 u(ln u) ln(ln 2) v

using a second substitution v = ln u. As before, the series converges if c > 1 and

diverges otherwise.

101
 6. Because N is finite,
∞
X N
X −1 ∞
X
an = an + an .
n=1 n=1 n=N

Let bn = aN +n−1 with g : [1, ∞) → R defined by g(x) = f (N + x − 1). Then,

∞
X ∞
X ∞
X
bn = f (N + n − 1) = g(n)
n=1 n=1 n=1

with g that satisfies the hypotheses of the Integral Test and so all integrals and all series
converge or diverge together.

7. As in the proof of Part 1., let un = nk=1 (ak − bk ). Then, un = sn − tn and so

P

lim un = lim sn − lim tn = a − b.

n→∞ n→∞ n→∞

Finally, let
n
X n
X
un = (cak ) = c( ak ) = c · sn .
k=1 k=1

And so, limn→∞ un = c(limn→∞ sn ) = ca.

8. Those are geometric series. Recall that they converge if and only if their ratio r is
smaller than 1 (in modulus). And so,
x
(a) r = x/3, the series converges if |x| < 3, that is, x ∈ (−3, 3) to ;
3−x
x−4
(b) r = x − 4, the series converges if |x − 4| < 1, that is, x ∈ (3, 5) to ;
5−x
(c) r = tan(x), the series converges if | tan(x)| < 1, that is, x ∈ (−π/4 + kπ, π/4 + kπ),
tan x sin x
k ∈ Z, to = .
1 − tan x cos x − sin x
     
 an + 1   p(n + 1)   q(n) 
9. The Ratio Test take the form lim   = lim  · . Note that
n→∞ an  n→∞  p(n)   q(n + 1) 
the leading
 coefficients
 of p(n + 1) and p(n) are the same for any polynomial, and so
 p(n + 1) 
lim   = 1, as well as for the ratio of q, therefore the Ratio Test is always
n→∞  p(n) 
inconclusive. Because the degrees of p and q are non negative integers, using the Limit
Comparison Test, we see that the series converges if and only if the degree of q is larger
or equal to the degree of p plus 2.

10. We know that |an | ≤ ρn2 for n ≥ N. And so,

∞ N −1 ∞ N −1 ∞ N −1
X X X X X X ρN
2
|an | = |an | + |an | ≤ |an | + ρn2 ≤ |an | + < ∞,
n=1 n=1 n=N n=1 n=N n=1
1 − ρ2

meaning that the series converges.

102
Chapter 5

Deeper Results on Sequences and

Series

5.1 Lecture 16: (ǫ, N )-Definition of Limits

In this lecture we give a more general definition of limit that does not need the existence
of a function f : [1, ∞) → R defining an . It is more precise, and somewhat simpler, than
Definition 2.5, although it looks more involved.

Definition 5.1. A sequence {an }∞ n=1 has limit a if for every ǫ > 0, there is an integer N(ǫ)
(usually depending on ǫ) such that, whenever n > N(ǫ), |an − a| < ǫ. If {an }∞
n=1 has limit a we
∞
write lim an = a. If lim an exists, we say that the sequence {an }n=1 converges. Otherwise
n→∞ n→∞
we say it diverges.

This definition of convergence is an extension of Definition 2.5.

Theorem 5.2. Suppose f : R → R and f (n) = an whenever n ∈ N. Furthermore, suppose

that lim f (x) = l. Then lim an = l.
x→∞ n→∞

Proof. Given any ǫ > 0, we know from the definition of lim f (x) that there exists M such
x→∞
that, whenever x > M, we have |f (x) − l| < ǫ. So take N = ⌈M⌉, that is, M rounded up
to the nearest integer (see Fundamentals at Level 1). Then, if n > N, we have |an − l| =
|f (n) − l| < ǫ because n > N ≥ M.

Remark 5.3. Because the limit of a sequence depends only on the (N)-tail of a sequence
{xn }∞
n=1 , defined as { xn : n > N }. Those results are valid for any tail being positive, or
bounded in [a, b].

As with limit of functions, we can define what it means to have limit ±∞.

Definition 5.4. We say that lim an = +∞ if for every M ∈ R there exists an integer N
n→∞
such that for all n > N , an ≥ M. We say that lim an = −∞ if for every M ∈ R there
n→∞
exists an integer N such that for all n > N , an ≤ M.

103
Note that the idea in the definition is that M can be as large as we wish. Conversely, in
the following, M is as large as we wish in modulus but negative.

Example. The sequence {2n }∞ n

n=1 tends to infinity because 2 can be made as large as we like
for n large enough. Basically, for any real M > 0, 2n > M when n > log2 M.

5.1.1 Practical Aspects of Estimates of Convergent Sequences

It is important to have a good grasp of the above definition of convergence. It implies that
however small we make the interval around l, all but a finite number of terms of the sequence
will lie outside of the interval. To state this in another way: the ‘tail of the sequence’ will lie
inside the interval. The smaller we make ǫ, the larger N = N(ǫ) has to be but, when we have
convergence, there is always such a value N(ǫ).

Remarks 5.5. 1. You should also note that N(ǫ) is not unique since, if we have an N(ǫ)
corresponding to a given ǫ > 0, then any natural number larger that this N will also
suffice to satisfy the definition. This will be useful later.

2. The definition also implies that any number of the initial values of xn are irrelevant
for convergence.

Today we shall concentrate on understanding this definition, in particular estimating the

number N(ǫ). What it means is that, whatever (small) number ǫ is given, when n > N(ǫ) we
know that |xn − l| < ǫ. To give an estimate to N(ǫ) we must:

1. Have a candidate for l. That candidate could come from calculating, guessing etc.

2. Find a simple upper bound for |xn − l| that is simple in N.

3. Set the upper bound to be smaller than ǫ. If the bound is simple, then we can get easily
the value of N(ǫ).

Examples.

1
1. A very simple example is the sequence {xn }∞n=1 with xn = . We believe, or calculated
n
using our knowledge of Level 1, that l = 0. We can use the definition to check if this is
correct. We need to evaluate (recall n ∈ N)
1 1
|xn − l| = | − 0| = < ǫ.
n n
And so, for n > N(ǫ) = ⌊1/ǫ⌋, where ⌊x⌋ represents the integer part of x (the largest
1
integer smaller than x), < ǫ.
n
Note that, as expected, N (ǫ) increases (to infinity) as ǫ tends to 0.

104
 1
2. Using the sequence {xn }∞
n=1 with xn = , we can again estimate N(ǫ). The
n + n2
candidate for the limit is again l = 0. And so,
1 1
|xn − l| = 2
< < ǫ,
n+n n
because n2 ≥ 0. We could use the previous value for N(ǫ) = ⌊1/ǫ⌋. This value for N(ǫ)
is far from optimal as we are going to see now.
3. To get a more concrete understanding of what N(ǫ) does, consider the previous example
1 1
with xn = 2
. Take ǫ = 0.1, say. The estimate n ≥ is easy to calculate, basically,
n+n ǫ
−1 1
if n > (0.1) = 10, < 0.1, but it is not optimal. Indeed, it is enough to choose
n + n2
1 1 1
n ≥ 3 to get 2
≤ < . Actually, in the exercises, we are going to see that,
n+n 12 10
from the analysis of quadratic inequalities you did early in Level 1, if
$ r %  
1 4+ǫ 2
n > N(ǫ) = − + = √ √ √ , (5.1)
2 4ǫ ǫ( 4 + ǫ + ǫ)
1
then n+n 2 < ǫ. Again, note that N (ǫ) increases as ǫ tends to 0. But, clearly,
−1
(5.1) is very
√ complicated, but gives the first ‘exact’ estimate of N(ǫ). For ǫ = 10 , you
get n > [ 10.25 − 0.5] = 2.
cos n
4. Let xn = 1 + n
.
The candidate for the limit is l = 1. Then,
 cos n  | cos n| 1
|xn − 1| =  = ≤ < ǫ,
 
n n n
and so we can choose N(ǫ) = 1/ǫ.
5. For a more complicated example, consider the sequence whose general term is
103n2 − 8
xn = .
4n2 + 99n − 3
The sequence converges because it is a combination of standard convergent sequences.
We have
103n2 − 8 103 − 8/n2 103
lim xn = lim 2
= lim 2
= .
n→∞ n→∞ 4n + 99n − 3 n→∞ 4 + 99/n − 3/n 4

103
To show, using the definition, that l = is indeed the limit, we estimate
4
103n2 − 8 103 99 · 103n − 277 100 · 104n 25 · 26 700
2
− = 2
≤ 2
≤ ≤ ,
4n + 99n − 3 4 4(4n + 99n − 3) 4(4n ) n n
700
because 99n − 3 ≥ 0 for n ≥ 1. Therefore |xn − l| < ǫ if n > = N(ǫ).
ǫ
Cautionary Tale 5.6. Note that this is highly not unique. Each of us will usually have
their own expression. But, some estimates are correct, some others are not because of
errors, not of choice.

105
5.1.2 Divergent Sequences
Sequences that do not converge satisfy the following result.

Proposition 5.7. A sequence {xn }∞ n=1 is not convergent if and only if for all l ∈ R, there
exists an ǫ(l) > 0 such that for all N ∈ N, there exists n̄(N) > N such that

|xn̄ − l| ≥ ǫ(l).

Proof. To show that a sequence is divergent, we need to state the contraposition of the defi-
nition of a limit. For that we need to look back at some propositional logic. The contra-
position of statements with the quantifiers ∃ or ∀ are as follows:

1. ¬(∀x, P (x))) = ∃x, ¬P (x),

2. ¬(∃x, P (x))) = ∀x, ¬P (x).

We use those two properties to state the contrary of the definition of a limit by exchanging ∀
and ∃. Now, in a compact notation, a sequence is convergent if

∃l, ∀ǫ, ∃N(ǫ), ∀n > N(ǫ), |xn − l| < ǫ.

The contraposition is thus

∀l, ∃ǫ(l), ∀N, ∃n̄(N), |xn − l| ≥ ǫ(l).

Example. Proposition 5.7 shows that the sequence defined by xn = (−1)n is divergent.
Indeed, let l ∈ R be any candidate for its limit. Take ǫ(l) = 1. Then, for all N, either

1. l ≥ 0, and |x2N +1 − l| = | − 1 − l| = 1 + l ≥ 1,

2. l < 0, and |x2N − l| = |1 − l| = 1 + |l| ≥ 1.

In both cases, n̄(N) = 2N + 1, l ≥ 0, or 2N, l < 0, is such that |xn̄(N ) − l| ≥ 1.

Using the same type of argument we can also show the following uniqueness result showing
that we can talk about THE LIMIT of a convergent sequence.

Theorem 5.8 (The limit is unique). If the sequence {xn }∞

n=1 converges then its limit is unique.

Proof. We use a proof by contradiction and thus we start by assuming that the converse is
true. That is, we assume that limn→∞ xn = x and limn→∞ xn = y with x 6= y. That is we
have |x − y| > 0. Now, by the triangle inequality, we have

|x − y| = |(x − xn ) + (xn − y)| 6 |xn − x| + |xn − y|.

106
Because x and y are limits of {xn }∞ n=1 , for every ǫ > 0 there exists N1 (ǫ) and N2 (ǫ) such that
|xn − x| < ǫ, for all n > N1 (ǫ), and |y − xn | < ǫ, for all n > N2 (ǫ). We want both of these to
be true and thus take n > N(ǫ) = max{N1 (ǫ), N2 (ǫ)} and we get

0 < |x − y| < 2ǫ.

This is true for every ǫ > 0 and hence, it is true for ǫ = |x − y|/4, which immediately gives us
a contradiction since we cannot have
2 1
0 < |x − y| 6 |x − y| = |x − y|,
4 2
when |x − y| =
6 0.

107
5.2 Extra curricular material: Error Estimates from the
Tests
In this extra curricular material we study various error estimates of the sum of series we can
deduce from the work on the tests for convergence

5.2.1 Error Estimates from the Ratio and Root Tests

We can obtain error estimates from the tests in Lemma 4.24 and Theorem 4.19.
Lemma 5.9. Let {an }∞
n=1 be a convergent sequence of terms such that there exists an N with,
either
|an+1 /an | ≤ ρ1 < 1, n ≥ N,
or
|an |1/n ≤ ρ2 < 1, n ≥ N.
We can estimate the error between the sum s of the series and its partial sum sn by
ρ1
|s − sn | ≤ |an |, n ≥ N,
1 − ρ1
in the first case, or by
ρn+1
2
|s − sn | ≤ , n ≥ N,
1 − ρ2
in the second.

Proof. Using the notation and calculations of Lemma 4.24, we can evaluate the difference
between s and sn when n ≥ N. We have
∞
X ρ1
|s − sn | ≤ |an+k | ≤ |an |.
k=1
1 − ρ1

For the second case, as before, comparing with the convergent geometric series we deduce that
the series converges and that the error s − sn is bounded by
∞ ∞
X X ρn+1
|s − sn | ≤ |an+k | ≤ ρ2n+k ≤ 2 .
k=1 k=1
1 − ρ2

5.2.2 Error Estimates for the Integral Test

Theorem 5.10 (Integral Test ErrorZ Estimates). 1. If f : [1, ∞) → [0, ∞) is a decreasing
∞
f (x) dx < ∞, so that s = ∞
P
continuous function with I = n=1 f (n) converges. Then
1
s ∈ [I, I + f (1)]. The ‘speed’ of convergence of sn to s is evaluated by
Z ∞
0 6 s − sn 6 f (x) dx. (5.2)
n

108
 2. In the situation of Theorem 4.19, when both I and s diverge, there exists γ ≤ f (1) such
that
lim (sn−1 − In ) = γ.
n→∞

That is, s and I both diverge at a rate comparable with a constant less or equal to f (1).

Proof. Following the proof of Theorem 4.19, we can relate the limits of {sn }∞ ∞
n=1 and {In }n=1
in the convergent case. We have that if sn → s and In → I as n → ∞ then

s − f (1) 6 I 6 s

and thus the sum of the series s satisfies s ∈ [I, I + f (1)]. We also have
n+j Z n+j
X
0 6 sn+j − sn = f (k) 6 f (x) dx.
k=n+1 n

Letting j → ∞ gives Z ∞
0 6 s − sn 6 f (x) dx.
n
This gives a bound on how rapidly the sequence converges to s. In general, recall that
Z k
f (k) 6 f (x) dx 6 f (k − 1).
k−1

And so,  Z k 
06 f (k − 1) − f (x) dx 6 f (k − 1) − f (k).
k−1

Summing from k = 2 to n gives

0 6 sn−1 − In 6 f (1) − f (n) < f (1)

where  Z 2   Z n 
sn−1 − In = f (1) − f (x) dx + · · · + f (n − 1) − f (x) dx .
1 n−1

Since {sn−1 − In }∞
n=1 is increasing with n and bounded by f (1) we know that it converges with

lim (sn−1 − In ) = γ 6 f (1).

n→∞

Thus the difference between the two sequences converges even in the case when the sequences
themselves diverge.

Examples.

1. When f (x) = 1/x we obtain the result that

n−1
!
X 1
lim − ln n = γ < 1.
n→∞
k=1
k

109
 In this case it can be shown that γ = 0.577.. which is known as the Euler1 constant.
Because
1
lim (ln(n + 1) − ln n) = lim ln(1 + ) = ln 1 = 0,
n→∞ n→∞ n
we can actually write γ in a more convenient form:
n
!
X 1
γ = lim − ln n .
n→∞
k=1
k

P∞
Thus n=1 1/n diverges as slowly as ln n, that is, for all ǫ > 0, there exists N(ǫ) such
that n
X 1
ln(n) + γ − ǫ ≤ ≤ ln(n) + γ + ǫ, n ≥ N(ǫ).
k=1
k

2. In the case f (x) = 1/xp , p > 1, we have

∞
dx
Z
0 6 s − sn 6 .
n xp
Now,
∞ M  
dx dx 1 1
Z Z
= lim p = .
n xp M →∞ n x p−1 np−1
When p = 2, we have
∞
X 1 1
06 2 6 .
k=n+1
k n

Hence to obtain the sum accurate to 10−3 we need a thousand (103 ) terms and to
obtain the sum accurate to 10−6 we need a million (106 ) terms. The series converges
rather slowly.

1
L. Euler (1707-1783) was a Swiss mathematician who spent most of his career between St-Petersburg
and Berlin. He was one of the towering lights of 18th century mathematics. He has been the most prolific
mathematician of his exceptional caliber and was nearly blind for many years still producing papers on all
areas of mathematics. Euler claimed that he made some of his greatest mathematical discoveries while holding
a baby in his arms with other children playing round his feet (he had 13 of them, although only 5 did not die
in infancy).

110
5.3 Lecture 17: Absolute Convergence of Series and the
Leibnitz Criterion of Convergence
Coming back to Lemma 4.14, we can introduce some terminology connected with this result.
P∞
Definition
P∞ 5.11. The series n=1 an is said to be absolutely convergent if the series
P∞ P∞
n=1 |an | converges.
P∞ The series a
n=1 n is said to be conditionally convergent if n=1 an
converges but n=1 |an | diverges.

We can thus rephrase Lemma 4.14: absolute convergence implies convergence. We

would like here to look at what happens to a series that is not absolutely convergent.

5.3.1 Alternating Sequences

∞
We have seen before that
P∞ the fact that the sequence {an }n=1 tends to 0 is not sufficient to
ensure that the series n=1 an converges. Here to consider sequences when this is true when
it is monotonic decreasing.
Definition 5.12. The sequence {an }∞ n ∞
n=1 is alternating if the sequence {(−1) an }n=1 is of
one sign, that is, all positive or all negative. Such sequences generate an alternating series
P ∞
n=1 an .

An example is the alternating harmonic series

∞
X (−1)n+1 1 1 1
=1− + − +··· .
n=1
n 2 3 4

Theorem 5.13 (Leibnitz Test). Let {an }∞ such that {|an |}∞
n=1 be an alternating sequence P n=1
is strictly monotonically decreasing with limn→∞ an = 0. Then the series ∞ n=1 an converges.
Moreover, the rate of convergence of the partial sum sn can be evaluated by

|s − sn | ≤ |an+1 |, n ≥ 1.

Proof. We assume that the first term a1 of the series is positive. When it is negative
P∞ we
multiply everything by (−1) and follow the same procedure. So, the series
P2n−1 is n=1 n =
a
a1 − |a2 | + a3 − |a4 | + . .. Consider the odd partial sums s2n−1 = i=1 ai and the even
P. 2n
partial sums s2n = i=1 ai . We claim that s2n−1 is monotone decreasing because

s2n+1 = s2(n+1)−1 = s2n−1 − |a2n | + a2n+1 < s2n−1 ,

and s2n is monotone increasing because

s2n+2 = s2(n+1) = s2n + a2n+1 − |a2n+2 | > s2n .

Therefore we have the inequalities

a1 − |a2 | = s2 < s2n < s2n+1 < s1 = a1 . (5.3)

111
The sequences of odd and even partial sums are monotone and bounded, hence they both
converge. Their limits must be equal because

lim (s2n+1 − s2n ) = lim a2n+1 = 0.

n→∞ n→∞

Therefore s = limn→∞ s2n+1 = limn→∞ s2n = limn→∞ sn .

The relation (5.3) means that the partial sums of an alternating series also alternates above
and below the final limit s. More precisely, when there are odd (even) number of terms, i.e.
the last term is a plus (minus) term, then the partial sum is above (below) the final limit.
This leads immediately to the error bound of partial sums, shown below. We have

|s2n+1 − s| ≤ s2n+1 − s ≤ s2n+1 − s2n+2 = |a2n+2 |,

|s2n − s| ≤ s − s2n ≤ s2n+1 − s2n = a2n+1 .

And so we can summarise both inequalities as |s − sn | ≤ |an+1 |, n ≥ 1.

Example. The Leibnitz result can be useful to study the behaviour of power series at the of
convergence because we get an alternating series at x = −R.

A series is determined by the sequence of its partial sums. So the order in which we sum its
terms is in principle important. Absolutely convergent series have the remarkable property
that their sum is independent of the order of summation, hence their name.
Definition 5.14. A series ∞
P P∞
n=1 bn is a re-arrangement of a series n=1 an if there exists
a bijection φ : N → N such that bn = aφ(n) .

Theorem 5.15 (Re-arrangement Theorem). Every re-arrangement of an absolutely conver-

gent series is absolutely convergent and all have the same sum.

Proof. We proceed in two steps. First we show that the result is true for the re-arrangement
of a series with positive terms. Let {an }∞ n=1 be a sequence with Ppositive terms andP {bn }∞
n=1
be a re-arrangement given by a bijection φ : N → N. Let sn = k=1 ak and tn = nk=1 bk .
n

For each n, choose m(n) ≥ max(φ(1),

P∞ · · · , φ(n)). Then tm(n) ≥ sn because all the terms in sn
appear in tm(n) If the series n=1 bn exists,
∞
X ∞
X
an = lim sn ≤ lim tm(n) = bn .
n→∞ n→∞
n=1 n=1

Note that {an }∞ ∞ −1

n=1 is also a re-arrangement of {bn }n=1 via φ . And so, both sums exist and
are equal because
X∞ X∞ X∞
an ≤ bn ≤ an .
n=1 n=1 n=1
P∞
In a second
P∞ part we consider
Pnan absolutely convergent
Pnseries s = n=1 an and a re-arrangement
+ −
t = n=1 bn . Let sn = k=1, ak >0 ak and sn = k=1, ak <0 (−ak ) be the (positive) partial
sums of the series of positive, respectively negative, terms of the original series. Similarly,
let {t+ ∞ − ∞
n }n=1 and {tn }n=1 be the (positive) partial sums of the series of positive, respectively

112
negative, terms of the re-arranged series. Because the original series is absolutely convergent,
s+ −
n and sn converge and have positive terms. It is clear that any re-arrangement of the sequence
{an }∞
n=1 does not alter the sign of its terms and so induces re-arrangements of the sequences of
positive, respectively, negative, terms. From the first part we know that {s+ ∞ + ∞
n }n=1 and {tn }n=1 ,
respectively {s− ∞ − ∞
n }n=1 and {tn }n=1 , converge to the same limit. And so the conclusion because

s = lim sn = lim (s+ − + −

n − sn ) = lim (tn − tn ) = lim tn = t.
n→∞ n→∞ n→∞ t→∞

Conditionally convergent sequences have a remarkable property: they are formed of entwined
divergent series of positive, resp. negative, terms.

Lemma 5.16. If a series ∞

P
n=1 an is conditionally convergent, then the series of its positive,
respectively negative terms, are divergent.

Pn Pn
Proof. Let s+n =
−
k=1, ak >0 ak and sn = (−ak ) be the (positive) partial sums of the
k=1, ak <0P
series of positive, respectively negative, terms of ∞ + −
n=1 ak . Recall that sn = sn − sn .We are
going to show that if one of the sequences {s+ ∞ − ∞
n }n=1 or {sn }n=1 converges then the other must
also converge, hence the original series must have been absolutely converging. Suppose that
limn→∞ s+ n = s
+
< ∞. Define s− = s+ − s where s is the sum of the original series. We
evaluate

|s− − s− + − + − + + + +
n | ≤ |s − s − sn | ≤ | − s + sn − sn | + |s − sn | = |s − sn | + |s − sn |.

From the convergence of the sequences {sn }∞ + ∞ − ∞

n=1 and {sn }n=1 we can conclude that {sn }n=1
is also convergent, a contradiction with the fact that the original series was not absolutely
convergent.

Finally, a remarkable result dues to Riemann2 .

P∞
Theorem 5.17. If a seriesP n=1 an converges conditionally, then for any number s ∈ R there
exists a re-arrangement ∞ n=1 bn converging to s.

Proof. The basis of the proof is that the partial sums of the positive and negative numbers
must both diverge as was shown in Lemma 5.16. Then we use the trick shown in the lectures
to show the result for the alternating harmonic series. Suppose c ≥ 0 (the proof follows exactly
the same line if c < 0). We take positive terms of the series until we go over c for the first
time. Then we take negative terms to get for the first time under c. Then we retake positive
terms to go over c again, etc. At each step we go over or under c by a strictly smaller amount,
hence the final convergence to c of the re-arrangement of the series.
2
G.F.D. Riemann (1826-1866) was a German mathematician of great originality. He developed complex
analysis and the relation between geometry and analysis. In his exceptional PhD thesis, he already showed
his style: great insight but somewhat lacking of the necessary rigour. On the other hand, the work of other
mathematicians to fill in the gaps in his ideas proved very fertile in the development of mathematics. One of
the very important conjectures in mathematics is about the structure of the complex zeros of the Riemann’s
ζ function.

113
Example

Conditionally convergent series are difficult objects to manipulate. For instance, let
∞ ∞ 
(−1)i+1

X X 1 1 1
s = = − − (by re-arrangement)
i=1
i i=1
2i − 1 2(2i − 1) 4i
∞  ∞ ∞
!
(−1)i+1 1 X (−1)i−1
 X
X 1 1 1
= − = = = s.
i=1
2(2i − 1) 4i i=1
2i 2 i=1 i 2

Hence s = 0. But s > 0 (in Exercise 5 of Sheet 4a we show that it is ln 2). This contradiction
occurs because we have kept the equal sign with the re-ordering of the series which is not
possible unless the series are absolutely convergent.

114
Summary of Chapter 4
We have seen:

• The (ǫ, N)-definition of a limit and to be able to evaluate N(ǫ) in simple cases;

• To know how to apply the Ratio, Root, Integral and Leibniz Tests for convergence of
series.

• further tests for determining whether a series is convergent;

• that we must check carefully that it is appropriate to use either of the tests;

• To know the definitions and to understand the difference between absolute and condi-
tional convergence of series.

115
5.4 Exercise Sheet 4

5.4.1 Exercise Sheet 4a

1. You have already calculated most of the following limits in Sheet 2a. Show using the
(ǫ, N)-definition that those limits are actually correct. To be precise,

(a) Determine, when it exists, a ‘candidate’ for the limit l for the following sequences
{xn }∞
n=1 .

(b) When such ‘candidate’ limit l exists. Determine an N(ǫ) such that the definition
of the limit of a sequence gives you that |l − xn | < ǫ, for all n > N(ǫ). Recall that
N(ǫ) does not need to be the smallest value (although you might try to have fun
to get a small one).
(c) When such ‘candidate’ limit l does not exists, show that the sequence has no limit
using the definition (more precisely its converse).

√ 5n3 + 3n + 1 sin(n2 + 1)
(a) xn = −n + n2 + 3n. (b) xn = . (c) x n = .
√ √ 15n3 + n2 + 2 n2 + 1
n+2− n+1 7n4 + n2 − 2 n3 + 3n2
(d) xn = √ √ . (e) xn = . (f) xn = − n2.
n+1− n 14n4 + 5n − 4 n+1
 3
n+1
(g) xn = − n3.
n
1
2. (a) Consider the sequence {xn }∞
n=1 with xn = . We would like to estimate the
n + n2
exact value of N(ǫ) such that
1
|xn − l| = < ǫ. (5.4)
n + n2
Note that rewriting (5.4), we would like to find N(ǫ) such that, if n > N(ǫ),
1
n2 + n − > 0.
ǫ
Use the analysis of quadratic inequalities you did early in Level 1, to find that
$ r %  
1 4+ǫ 2
N(ǫ) = − + = √ √ √ .
2 4ǫ ǫ( 4 + ǫ + ǫ)

(b) Show that N(ǫ) increases as ǫ tends to 0.

3. Prove the Limit Comparison Theorem, Theorem 4.18.

4. Let 0 < a < 1. Consider the series

a2 + a + a4 + a3 + · · · + a2n + a2n+1 + · · · .

Show that the Root Test applies but not the Ratio Test.

116
 ∗
Pn 1

5. Let cn = k=1 k − ln n. Show that (c ) is a decreasing sequence of positive numbers.
P2nn
Let γ be its limit. Show that if bn = k=1 (−1)k+1 k1 , then lim bn = ln 2.
n→∞

6. Using the (ǫ, N)-definition of a limit, prove that

(a) a convergent sequence is bounded;

(b) If {xn }∞
n=1 is a convergent sequence of real numbers and if xn > 0 for all n ∈ N
then
l = lim xn > 0.
n→∞

Similarly, if xn ≤ 0, then l ≤ 0.

7. Using the (ǫ, N)-definition of a limit, prove that the algebra of limit holds. Namely, let
{xn }∞ ∞
n=1 and {yn }n=1 be convergent sequences in R with limits x and y, respectively,
and let α ∈ R. Then,

(a) limn→∞ (xn + yn ) = x + y,

(b) limn→∞ (αxn ) = αx,
(c) limn→∞ (xn yn ) = xy,
(d) limn→∞ (xn /yn ) = x/y provided yn 6= 0 and y 6= 0;

8. Suppose that {xn }∞ ∞ ∞

n=1 , {yn }n=1 and {zn }n=1 are sequences of real numbers such that

xn 6 yn 6 zn ,

for all n ∈ N, and that

lim xn = lim zn .
n→∞ n→∞

Using the (ǫ, N)-definition of a limit, show that the ‘intermediate’ sequence {yn }∞
n=1 is
convergent and
lim xn = lim yn = lim zn .
n→∞ n→∞ n→∞

5.4.2 Additional Exercise Sheet 4b

1. When the following limits exist, determine an N(ǫ) satisfying the (ǫ, N)-definition of
n3 + 2n2 + 1 n2 + n + 1 √
limit. (a) xn = 3 +n+4
. (b) xn = 2 +4
. (c) x n = n4 + n2 − n2.
6n
4 3n
√ √ √

n+1 4 2 + n. (f) x =
sin n
(d) xn = 2
− n . (e) x n = −n + n n + ( n + 1 − n).
n n
n2 + 500n + 1 p p
(g) xn = . (h) xn = n(n + 1) − n(n − 1).
5n2 + 3

5.4.3 Short Feedback for Exercise Sheet 4a

1. (a) l = 3/2,
(b) l = 1/3,
(c) l = 0,

117
 (d) l = 1,
(e) l = 1/2,
(f) l = ∞ (divergent),
(g) l = −∞ (divergent).

2. (a) Consider 1/ǫ as a parameter perturbing the parabola y(x) = x2 + x.

(b) Show that the derivative of the denominator is monotone.

3. Show that the limit condition means that for any ǫ there exists N(ǫ) such that

(l − ǫ) bn < an < (l + ǫ) bn , ∀n > N(ǫ).

∞
X ∞
X
Then use that relation to show that the tails of the series an and bn converge
n=N (ǫ) n=N (ǫ)
or diverge together.

4. The coefficient ρ2 in the Root Test has a limit, but ρ in the Ratio Test has not.

5.4.4 Short Feedback for the Additional Exercise Sheet 4b

1. (a) l = 1/6,
(b) l = 1/3,
(c) l = 1/2,
(d) l = −∞, (divergent),
(e) l = 1/2,
(f) l = 0,
(g) l = 1/5,
(h) l = 1/2.

This result follows from the following more sophisticated test, but whose proof must be delayed
until next term. Note that our results are valid if the tail of the series start to be alternating
for n ≥ N. Those series form a good source of conditionally convergent series if they are
not absolutely convergent. The Leibnitz Test is the consequence of the more sophisticated
Dirichlet Test.
Abel3 Test.
∞
Proposition 5.18 (Dirichlet Test). Let {bP
n }n=1 be a decreasing sequenceP
such that limn→∞ bn =
0 and {an }n=1 a sequence such that m ≤ i=1 ai ≤ M for all n. Then ∞
∞ n
k ak bk converges.

Proof. See MA2731, Analysis 2.

3
N.H. Abel (1802-1829) was a brilliant Norvegian mathematician that died young of illness (mainly due to
poverty). He was the first to prove that there is no general formula for the solution of a quintic equation.

118
5.4.5 Feedback for Exercise Sheet 4a
Recall that for positive numbers
a a+p a a
≤ , ≤
b b b+p b

a2 − b2
and a − b = .
a+b

1. In that exercise, calculate the limit, when it exists, using the usual tools. Then, estimate
|xn − l| trying to get a simple expression that obviously tends to 0 as n is big. Then
we can find N(ǫ) by estimating how big n should be for the estimate of |xn − l| to be
smaller than ǫ. Note that this is highly not unique. Everybody will usually
have their own expression, but some estimates are correct, some others are
not, this is the difference between being correct or wrong.

(a) The limit can be calculated multiplying by the conjugate quantity

√ 3n 3
lim xn = lim (−n + n2 + 3n) = lim √ = .
n→∞ n→∞ n→∞ n + n2 + 3n 2

The estimate
  √
3 3n  3 2 + 3n − n
|xn − l| =  − √ = · n √
2 n + n2 + 3n  2 n + n2 + 3n
3 3n 9n 9 2
≤ · √ ≤ 2
= ≤ .
2 (n + n2 + 3n)2 2(2n) 8n n

2
Clearly, |xn − l| < ǫ when n > .
ǫ
1
(b) As in (i), we show that limn→∞ xn = . To find an estimate for N(ǫ),
3
| − n2 + 9n + 1| |n2 − 9n − 1| 1
|xn − l| = 3 2
≤ 3
≤ ,
3(15n + n + 2) 3(15n ) 45n
1
because |n2 − 9n − 1| ≥ 1 when n ≥ 3. And so, |xn − l| < ǫ when n > .
45ǫ
(c) Here, we can estimate directly
 sin(n2 + 1) 
 
|xn | ≤  ≤ 1.
n + 1  n2
2

1
This provides that limn→∞ xn = 0 and an estimate for N(ǫ) = √ .
ǫ
(d) Applying the identity a−b = (a2 −b2 )/(a+b) to the numerator and the denominator
gives √ √
n+1+ n
lim xn = lim √ √ = 1.
n→∞ n→∞ n+2+ n+1

119
 We then estimate
√ √
| n + 2 − n|
|xn − 1| = √ √
| n + 2 + n + 1|
2 2 1
= √ √ √ √ = √ 2 = ,
| n + 2 + n| · | n + 2 + n + 1| (2 n) 2n

1
and so, a possibility for N(ǫ) = .
2ǫ
(e) Clearly limn→∞ xn = 12 . And so, when n ≥ 2,

2n2 − 5n 2
 
 = 2n ≤ 1 ≤ 1 ,
 
|xn − l| = 
2(14n4 + 15n − 4)  28n4 14n2 9n2
1
and so we can choose N(ǫ) = √ .
3 ǫ
2n2 2n2 2n2
(f) Simplifying xn , we find that xn = . Clearly then, xn = ≥ = n,
n+1 n+1 2n
and so the sequence diverges as xn > M if n > M.
 3
n+1
(g) The term is bounded below by 1 and above by 8. This follows from
n
n+1 1
1≤ ≤ 1 + ≤ 2.
n n
And so xn is unbounded, with limn→∞ xn = −∞. Given an arbitrary positive
number M, we can estimate for which n is xn < −M. We have

xn ≤ 8 − n3 < −M
√
3
when n > 8 + M.

2.

3.
an+1
4. The Ratio Test gives = a3 < 1, if n is even, or a−1 > 1, if n is odd. So we cannot
an
conclude.
On the other hand, note that if n is odd, an = an+1 , and if n is even, an = an−1 . The
Root Test gives (an )1/n = a1+1/n if n is even and a1−1/n if n is odd. So the limit of
1
the root (an )1/n as n tend to ∞ is a < 1 in both cases as lim = 0. Thus the series
n→∞ n
converges.
∗
5. That cn > 0 follows from the estimates in Theorem 4.19 and Example 2 after it. That
it is decreasing follows from the following estimate:
 
1 1 1
cn+1 − cn = − ln(n + 1) + ln n = + ln 1 − .
n+1 n+1 n+1

120
 1
Let x = . So as n tends to ∞, x tends to 0+ . Now, in terms of x,
n+1
cn+1 − cn = x + ln(1 − x) = ln(ex (1 − x)).

The function f (x) = ex (1 − x) is strictly smaller than 1 when x tends to 0+ because

f (0) = 1 and f ′ (x) = −ex < 0. And so cn+1 − cn = ln(f (x)) < 0. Therefore {cn }∞
n=1
tends to a limit which we called γ, the Euler constant. Note that

bn = c2n − cn + ln 2

because
2n n
! !
X 1 X 1
c2n − cn + ln 2 = − ln(2n) − − ln n + ln 2
k k
k=1 k=1
2n n
X 1 1 X
= − − ln 2 − ln n + ln n + ln 2
k k
k=1 k=1
2n n 2n 2n
! !
X 1 X 1 X 1 X 1
= −2 = −2
k 2k k j=1,even
j
k=1 k=1 k=1
2n
X 1
= (−1)k+1
k=1
k
= bn .

Hence,
lim bn = lim (c2n − cn + ln 2) = γ − γ + ln 2 = ln 2.
n→∞ n→∞

6. (a)
(b) We use a proof by contradiction. That is, we start with the assumption that l < 0.
But the elements of the tail of the sequence can be arbitrarily close to l and in
particular this implies that for sufficiently large n, xn < 0. In terms of ǫ’s, we take
ǫ = −x/2 = |l|/2 > 0. The convergence means that there exists an N ∈ N such
that |xn − l| < ǫ, that is, l < xn − l < −l or 2l < xn < 0 for all n > N. This
contradicts xn > 0 for all n ∈ N.

7. We will only prove the first and third assertions. The others are proved in the same
spirit.
Let N1 (ǫ), N2 (ǫ), be such that |xn − x| < ǫ, |yn − y| < ǫ, for n > N1 (ǫ), n > N2 (ǫ),
respectively. For the first assertion we evaluate

|(x + y) − (xn + yn )| ≤ |x − xn | + |y − yn | < ǫ

ǫ ǫ
for n > max{N1 ( ), N2 ( )}.
2 2
For the third assertion, the key to the proof is to re-write the difference xn yn − xy in a
form where we can compare xn with x and yn with y. We have

xn yn − xy = xn (yn − y) + y(xn − x).

121
 Then by considering the absolute value, and using the triangle inequality, we have

|xn yn − xy| 6 |xn | · |yn − x| + |y| · |xn − x|.

Next we use a previous result that {xn }∞

n=1 being convergent implies that |xn | is bounded
by a number M > 0. Hence

|xn yn − xy| 6 |xn | · |yn − y| + |y| · |xn − x|

6 M · |yn − y| + |y| · |xn − x|.
ǫ ǫ
Now take n > max{N1 ( ), N2 ( )}, then
2|y| 2M

|xn yn − xy| < ǫ.

8. This is a direct proof. Let

w = lim xn = lim zn .
n→∞ n→∞

The convergence of {xn }∞and

n=1 {zn }∞
to w implies that given any ǫ > 0 there exists
n=1
N(ǫ) ∈ N (the maximum of the N’s for the two sequences) such that

|xn − w| < ǫ and |zn − w| < ǫ for all n > N(ǫ).

Now the ordering of the elements xn 6 yn 6 zn implies

xn − w 6 yn − w 6 zn − w

and hence, for n > N(ǫ),

−ǫ < xn − w 6 yn − w 6 zn − w < ǫ,

and so |yn − w| < ǫ. Thus limn→∞ yn = w.

122
Chapter 6

Approximation with Taylor

Polynomials

6.1 Lecture 18: Taylor’s theorem and error estimates

It is now time to put all of these results to work. In Lectures 18 and 19 we begin to draw
all of our results together and return to Taylor and Maclaurin poynomials. Our fundamental
question, which we have dodged until now, is:
When is it true that the degree of the Taylor or Maclaurin polynomial can be
taken to the limit and so produce legitimate Taylor or Maclaurin series of the
form,
∞
X (x − a)n (n)
f (x) = f (a)?
n=0
n!

This chapter and the next diverge somewhat from the content of Lectures 18 and 19 by
providing an extensive set of worked examples. Our strategy is to return to the remainder
term for Taylor and Maclaurin polynomials that was introduced much earlier in Section 5. We
derive the precise form of this remainder and are then able to use it for two primary purposes.

• By terminating the Taylor or Maclaurin polynomial at some finite degree, we can esti-
mate the size of the remainder to determine a bound on the accuracy with which the
polynomial, Tna f (x) approximates f (x) at x. By inverting the question, we can then
determine how many terms we need to reach a specified accuracy.

• By examining the limiting behaviour of the remainder as the polynomial degree tends
to infinity we can ascertain whether or not f (x) = limn→∞ Tna f (x). If this limit holds
a
then we will have rigorously proven that f (x) = T∞ f (x).

The second item is the focus of Lectures 18 and 19. It is without a doubt harder to grasp and
so is better presented ‘live’. Once the structure of the remainder term is understood though,
the first item can provide ample opportunity for investigation. The first item is therefore
supported mainly by this document.
Section 6.2 discusses the remainder term and then in Section 6.3 we show how this remainder
can be estimated.

123
Concrete examples on estimating the error in replacing some familiar functions by their
Maclaurin series are then detailed in Subsection 6.3.1, for computing e; Subsection 6.3.2,
for the error in T4 cos(x); and, Subsection 6.3.3, for sin(x)/x.
Furthermore, Section 6.4 discusses the possibility of approximating integrals by replacing
integrands with their Taylor or Maclaurin polynomials.
Once again, though, note that the main points will be covered in the lectures and detailed on
the lecture slides. This document supports that activity but does not replace it.
There is one important lemma that we will make use of in the lectures. The proof is boardwork
for Lecture 19.
Lemma 6.1.
xn
lim =0 ∀x ∈ R.
n→∞ n!

6.2 Taylor Theorem

Given a n-times differentiable function f : I → R, where I is an open interval and a ∈ I, we
have seen that its Taylor polynomial of degree n at a

f ′′ (a) f (n) (a)

Tna f (x) ′
= f (a) + f (a)(x − a) + 2
(x − a) + · · · + (x − a)n. (6.1)
2! n!
This notation encodes the data we need to know: the function f , the degree n and the point a.
We had seen that the function f and its Taylor polynomial Tna f share the same values at a as
well as their first n derivatives. What about in an interval around x = a? Many mathematical
operations need the values of a function on an interval, not only at one given point x = a.
Consider an interval I around a, a tolerance ǫ > 0 and the order n of the approximation, here
are the big questions:

1. Given Tna f and I. Within what tolerance does Tna f approximate f on I?

2. Given Tna f and ǫ. On how large an interval I does Tna f achieve that tolerance?

3. Given f , a ∈ I and ǫ. Find how many terms n must be used for Tna f to approximate
f to within ǫ on I.

Having a polynomial approximation that works all along an interval is a much more sub-
stantive property than evaluation at a single point. The Taylor polynomial Tna f (x) is almost
never exactly equal to f (x), but often it is a good approximation, especially if |x − a| is small.
Remark 6.2. It must be noted that there are also other ways to approach the issue of best
approximation by a polynomial on an interval. And beyond worry over approximating
the values of the function, we might also want the values of one or more of the derivatives
to be close, as well. The theory of splines is one approach to approximation which is very
important in practical applications. Also the approximation in the mean can be useful:
given a function f , which polynomial p of degree n minimises the error
Z b
(f (x) − p(x))2 dx?
a

124
To see how good is the approximation Tna f of f , we defined the ‘error term’ or, ‘remainder
term’.

Definition 6.3. Let I ⊂ R be an open interval, a ∈ I and f : I → R be a n-times differentiable

function, then
Ran f (x) = f (x) − Tna f (x) (6.2)
is called the n-th order remainder (or error) term of the Taylor polynomial Tna f of f .

In general, we cannot really do much with Definition 6.3 on its own.

Example. If f (x) = sin x then we have found that T3 f (x) = x − 16 x3, so that

R3 sin(x) = sin x − x + 61 x3.

Although correct, it is not a very useful formula as it stands.

But, there is an important example where we can (un-usually) compute Rn f (x).

Example. Consider the function f : (−1, 1) → R given by

1
f (x) = .
1−x
We have seen that the McLaurin polynomials (at a = 0) of f is
n
X 1 − xn+1 xn+1
Tn f (x) = xk = = f (x) − ,
k=0
1−x 1−x

a geometric progression. The remainder term therefore is

xn+1
Rn f (x) = f (x) − Tn f (x) = .
1−x

Our main theorem is as follows. It gives a formula for the remainder, formula we will be able
to use for estimates later.

Theorem 6.4 (Taylor’s Theorem). Let f be an n + 1 times differentiable function on some

open interval I containing a. Then, for every x in the interval I there is a ξ between a and x
such that
a f (n+1) (ξ)
Rn f (x) = (x − a)n+1 . (6.3)
(n + 1)!

Proof. Suppose a < x, otherwise interchange them. The full proof is not significantly harder,
but we are going to show the previous result for n = 1 only, that is,

f ′′ (ξ)
R2a f (x) = (x − a)2,
2!

125
for some a < ξ < x. Define the function in t, F : [a, x] → R by

F (t) = f (x) − f (t) − f ′ (t)(x − t).

Then F ′ (t) = −f ′′ (t)(x − t). Define G : [a, x] → R by

 2
x−t
G(t) = F (t) − F (a).
x−a

Then G(a) = G(x) = 0. Applying Rolle’s Theorem, there exists a < ξ < x such that

(x − ξ)
0 = G′ (ξ) = F ′ (ξ) + 2F (a) .
(x − a)2

Hence,
(x − a)2 ′ f ′′ (ξ)
F (a) = − F (ξ) = (x − a)2.
2(x − ξ) 2

The formula (6.3) is called the Lagrange Remainder Formula. Even though you usually
cannot compute the mystery point ξ precisely, Lagrange’s formula allows you to estimate it.
A couple of example of a type already done.
√
Example. Approximate
√ x by a Taylor polynomial of degree 2 at x = 4. Use this polynomial
to approximate 4.1 and use Taylor’s Theorem to estimate the error in this approximation.
√ 1 1 3
Let f : (0, ∞) → R, f (x) = x. Then f ′ (x) = √ , f ′′ (x) = − 3/2 and f ′′′ (x) = .
2 x 4x 8x5/2
The Taylor polynomial, T24 f (x), is
1 1 1
T24 f (x) = f (4) + f ′ (4)(x − 4) + f ′′ (4)(x − 4)2 = 2 + (x − 4) − (x − 4)2.
2 4 64
On substituting x = 4.1,
√ 1 1
4.1 ≈ 2 + (0.1) − (0.1)2 = 2.02484375.
4 64
Taylor’s Theorem gives
M(4.1 − 4)3
|R24 f (4.1)| ≤ ,
3!
3
where M is the absolute maximum of |f (3) (x)| on [4, 4.1]. Now f (3) (x) = x−5/2 . This is
8
(3) 3 −5/2
a decreasing positive function so the absolute maximum of |f (x)| = x occurs when
8
3 3
x = 4. Hence, M = = and it follows that
8 · 45/2 256
(3/256)(0.1)3
|R24 f (4.1)| ≤ ≈ 0.000002.
3!
√
Thus, our approximation to 4.1 differs from the exact value by 0.000002 or less!

126
 3.5

2.5

1.5

0.5

0
-2 0 2 4 6 8 10
-0.5

-1
√
The graph shows √ x in red and T24 f in green and shows clearly that the Taylor polynomial
T24 f approximates x extremely well but only for a certain range of x.

127
6.3 Estimates Using Taylor Polynomial
Here is the most common way to estimate the remainder.

Theorem 6.5 (Estimate of the Remainder Term). Let f be an n + 1 times differentiable

function on some open interval I containing a. If you have a constant M such that
 (n+1) 
f (ξ) ≤ M, (6.4)

for all ξ between a and x, then

M|x − a|n+1
|Rna f (x)| ≤ . (6.5)
(n + 1)!

Proof. We do not know what ξ is in Lagrange’s formula, but it doesn’t matter, for wherever
it is, it must lie between a and x so that our assumption (6.4) implies that |f (n+1) (ξ)| ≤ M.
Put that in Lagrange’s formula and you get the stated inequality.

Some final remark before we embark in using those results.

1. Note, usually we will find M by finding the absolute max and min of f (n+1) (x) on the
interval I. Sometimes, however, we can find a value for M without calculating absolute
max’s and min’s. For example, if f (x) equals sin(x), then we can always take M = 1.

2. Note that this theorem gives some idea why Maclaurin approximations get better by
M
using more terms. As n gets bigger, the fraction (n+1)! will often get smaller. Why?
M
Because (n + 1)! gets really big. Is it true then that (n+1)! always get smaller for very
large n? Well, to be rigorous, M might also increase with n. But those cases are rare
for functions we deal with in general.

3. Note, we are often interested in bounding |Rna f (x)| on an interval. In such a case we
replace |x − a|n+1 by its absolute max on the interval. In other words, if the interval is
I = [c, d], we shall replace |x − a|n+1 by |c − a|n+1 or |d − a|n+1 , whichever is bigger.

We consider here two simple cases of application of the error estimates.

6.3.1 How to compute e in a few decimal places?

Note that e = f (1) where f (x) = ex . We computed Tn f (x) before. If you set x = 1, then you
get e = f (1) = Tn f (1) + Rn f (1), and thus, taking n = 8,
1 1 1 1 1 1 1 1
e=1+ + + + + + + + + R8 (1).
1! 2! 3! 4! 5! 6! 7! 8!
By Lagrange’s formula there is a ξ between 0 and 1 such that

f (9) (ξ) 9 eξ
R8 (1) = 1 = .
9! 9!

128
We do not really know where ξ is, but, since it lies between 0 and 1, we know that 1 < eξ < e.
So, the remainder term R8 (1) is positive and no more than e/9!. Estimating e < 3, we find
1 3
< R8 (1) < ≈ (120000)−1.
9! 9!
Note that we have been able to estimate a lower bound and upper bound for R8 f (1).
Thus we see that
1 1 1 1 1 1 1 1 1 1 1 3
1+ + + +···+ + + < e < 1+ + + +···+ + +
1! 2! 3! 7! 8! 9! 1! 2! 3! 7! 8! 9!
or, in decimals,
2.718 281 . . . < e < 2.718 287 . . .

6.3.2 How good is the approximation of T4 f for f (x) = cos(x)?

2 4
For f (x) = cos(x), let’s look at the approximation T4 f (x) = 1 − x2 + x4! on the interval [− 12 , 21 ].
We might ask within what tolerance does T4 f approximate cos x on that interval? To answer
this, recall that the error term for T4 f is
− sin ξ 5
R4 f (x) = x.
5!
For x in the indicated interval, we want to know the worst-case scenario for the size of
this error. A sloppy but good and simple estimate on sin ξ is that | sin ξ| 6 1, regardless
of what ξ is. This is a very happy kind of estimate because it is not so bad and because it
does not depend at all upon x. The biggest that x5 can be is ( 12 )5 ≈ 0.03. Then the error is
estimated as
− sin ξ 5 1
|R4 f (x)| = | x |6 5 6 0.0003.
5! 2 · 5!
This is not so bad at all! We could have been a little clever here, taking advantage of the fact
that T4 f = T5 f . Thus, we are actually entitled to use the remainder term R5 f . This typically
will give a better outcome, namely
cos ξ 6
R5 f (x) = − x.
6!
Again, in the worst-case scenario | − cos ξ| 6 1. And still |x| 6 21 , so we have the error
estimate
cos ξ 6 1
|R5 f (x)| = | x |6 6 6 0.000022.
6! 2 · 6!
This is less than a tenth as much as in the first version.
But what happened here? Are there two different answers to the question of how well that
polynomial approximates the cosine function on that interval? Of course not. Rather, there
were two approaches taken by us to estimate how well it approximates cosine. In fact, we still
do not know the exact error! The point is that the second estimate (being a little wiser) is
closer to the truth than the first. The first estimate is true, but is a weaker assertion than
we are able to make if we try a little harder.
This already illustrates the point that ‘in real life’ there is often no single ‘right’ or ‘best’
estimate of an error, in the sense that the estimates that we can obtain by practical procedures
may not be perfect, but represent a trade-off between time, effort, cost, and other priorities.

129
6.3.3 Error in the Approximation sin x ≈ x
In many calculations involving sin x for small values of x one makes the simplifying approxi-
mation sin x ≈ x, justified by the known limit
sin x
lim = 1.
x→0 x
How big is the error in this approximation?
To answer this question, we use Lagrange’s formula for the remainder term. Let f : R → R,
f (x) = sin x. Its Maclaurin polynomial of order 1 is

T1 f (x) = x,

of remainder
f ′′ (ξ) 2
R1 f (x) = x = − 21 sin ξ · x2
2!
for some ξ between 0 and x. As always with Lagrange’s remainder term, we don’t know where
ξ is precisely, so we estimate the remainder term. As we did several time before, | sin ξ| ≤ 1,
∀ξ ∈ R. Hence, the remainder term is bounded by

|R1 f (x)| ≤ 12 x2, (6.6)

and so we find that

x − 12 x2 ≤ sin x ≤ x + 21 x2 . (6.7)

Example. Now we could fix the maximum error we are seeking and ask how small must
we choose x to be sure that the approximation sin x ≈ x is not off by more than,
say, 1%? If we want the error to be less than 1% of the estimate (it is a relative error, then
we should require 12 x2 to be less than 1% of |x|, that is |x| < 0.02.

Recall that because T1 f = T2 f , the error term R1 f can be estimated using R2 f , that is

|R2 f (x)| ≤ 61 |x|3

instead of 21 x2 as in (6.6). This means that (6.7) becomes

x − 61 |x|3 ≤ sin x ≤ x + 61 |x|3. (6.8)

Now, is that better? Think about it. It is actually better for ‘small’ x, but worse for ‘large’
x.

130
6.4 Estimating Integrals
Thinking simultaneously about the difficulty (or impossibility) of ‘direct’ symbolic integration
of complicated expressions, by contrast to the ease of integration of polynomials, we might
hope to get some mileage out of integrating and differentiating Taylor polynomials.

Example. We know that

T
dx
Z
= [− ln(1 − x)]T0 = − ln(1 − T ). (6.9)
0 1−x
1
On the other hand, the Maclaurin polynomial of f (x) = is
1−x

Tn f (x) = 1 + x + x2 + . . . + xn. (6.10)

Replacing into (6.9), we could obtain

T T
x2 xn+1 T2 T n+1
Z 
2 n
(1 + x + x + . . . + x ) dx = x + + ...+ =T + + ...+ .
0 2 (n + 1) 0 2 (n + 1)

For x < 1, let F (x) = − ln(1 − x). Putting the previous results together (and changing the
variable back to ‘x’), we get the ‘formula’:

x2 xn+1
Tn F (x) = x + + ...+ .
2 (n + 1)

For the moment we do not worry about what happens to the error term for the Taylor
polynomial. This little computation has several useful interpretations.

1. First, we obtained a Taylor polynomial for − ln(1 − T ) from that of the geometric series
in (6.10), without going to the trouble of recomputing derivatives.

2. Second, from a different perspective, we have an expression for the integral

Z T
dx
0 1−x

without necessarily mentioning the logarithm: that is,

Z T
dx T2 T n+1
≈T+ +···+
0 1−x 2 (n + 1)

for arbitrary large n.

Being a little more careful, we would like to keep track of the error term in the example so
we could be more sure of what we claim. We have
1 1 1
= 1 + x + x2 + . . . + xn + xn+1
1−x (n + 1) (1 − ξ)n+1

131
for some ξ between 0 and x (and also depending upon x and n). One way to avoid having
1
the term to ‘blow up’ on us, is to keep x itself in the range [0, 1) so that ξ is in the
(1 − ξ)n+1
range [0, x) ⊆ [0, 1), keeping ξ away from 1. To do this we might demand that 0 6 T < 1.
For simplicity, and to illustrate the point, we take 0 6 T 6 21 . Then, in the worst-case
scenario,
1 1
| |6 = 2n+1.
(1 − ξ) n+1 (1 − 21 )n+1
Thus, integrating the error term, we have
Z T Z T n+1 Z T
1 1 n+1 2 n+1 2n+1
| n+1
x dx| 6 x dx = xn+1 dx
0 n + 1 (1 − ξ) 0 n+1 n+1 0
n+1
 n+2 T
2 x 2n+1 T n+2
= = .
n+1 n+2 0 (n + 1)(n + 2)
Since we have required 0 6 T 6 12 , we actually have

2n+1 ( 21 )n+2
Z T
1 1 n+1 2n+1 T n+2 1
| n+1
x dx| 6 6 = .
0 n + 1 (1 − c) (n + 1)(n + 2) (n + 1)(n + 2) 2(n + 1)(n + 2)
That is, we have
T2 Tn 1
| − log(1 − T ) − [T + + ...+ ]| 6
2 n 2(n + 1)(n + 2)
for all T in the interval [0, 12 ]. Actually, we had obtained
T2 Tn 2n+1 T n+2
| − log(1 − T ) − [T + + ...+ ]| 6
2 n (n + 1)(n + 2)
and the latter expression shrinks rapidly as T approaches 0.

Next we return to another example from the previous lecture.

Example. Let f : R → R given by f (x) = cos x. We use T3a f about a = π/2 to estimate
Z π
cos x dx,
π/3

and compare the estimation with the actual answer. The polynomial is

π/2 (x − π/2)3
T3 f (x) = −(x − π/2) + .
3!
We approximate the integral by
Z π Z π 
(x − π/2)3  h (x − π/2)2 (x − π/2)4 iπ
cos x dx ≈ −(x − π/2) + dx = − + ≈ −0.846085.
π/3 π/3 3! 2 24 π/3

The exact answer is

π
√
3
Z h iπ
cos x dx = sin x =− ≈ −0.866025.
π/3 π/3 2

132
The approximate answer is quite close to the exact answer. But could we guarantee this? The
π/2 f (4) (ξ)
error could be evaluated from the integral of the remainder term R3 f = (x − π/2)4 .
4!
Because each derivative of cos is bounded by 1 (determine why?), we get
π π
|(x − π/2)|4 [(x − π/2)5]ππ/3 244π 5
Z Z
π/2
|R3 f (x)| dx ≤ dx ≤ ≤ ≈ 0.08.
π/3 π/3 4! 5! 933120

We see that our predicted error is about four times larger than the real value. This is not a
contraction, but we could have been smarter. Note that because cos an odd function about
π/2 π/2 π/2
x = π/2, T4 f (x) = T3 f (x), and so we could use the remainder R4 f (x). In that case,
we get the estimate

[(x − π/2)6 ]ππ/3

Z π Z π
π/2 |(x − π/2)|5 91π 6
|R4 f (x)| dx ≤ dx ≤ ≤ 6 ≈ 0.020835,
π/3 π/3 5! 6! 6 · 90

which is basically the error we actually got.

Of course, we would always compute this integral by working out the exact answer, since it
is much simpler than doing the approximation in this case. But, more generally, the first few
terms of a Taylor expansion can be used as an approximation and might help us approximately
determine integrals of functions which cannot be integrated exactly, for instance, in Question
9 in Exercise Sheet 5a, Z 0.5 √
1 + x4 dx.
0
There are no simple exponential or trigonometric functions giving the integral of that square
root. And so, to evaluate it, we need some approximation, either numerical (like the Trapezium
Rule) or using a Taylor polynomial (or series).

We now give an alternative Taylor estimate using an integral formulation for the error.
Theorem 6.6 (Taylor’s Theorem Revisited). Let f : [a, b] → R be n + 1 times differentiable
on [a, b]. Let Rna f (x) = f (x) − Tna f (x) for x ∈ [a, b], then,
 Z x f (n+1) (t)  M|x − a|n+1
a n
|Rn f (x)| =  (x − t) dt ≤ , (6.11)
 
a n! (n + 1)!

where M is the maximum value of |f (n+1) (t)| on [a, x].

Proof. For t ∈ [a, x], let

n
X f (k) (t)
S(t) = Rnt f (x) = f (x) − Tnt f (x) = f (x) − (x − t)k.
k!
k=0

We claim that
f (n+1) (t)
S ′ (t) = − (x − t)n . (6.12)
n!
We have
d f (k) (t) f (k+1) (t) f (k) (t)
(x − t)k = (x − t)k − k(x − t)k−1 .
dt k! k! k!

133
So,
n n
′
X f (k+1) (t) k
X f (k) (t)
S (t) = − (x − t) + k(x − t)k−1
k=0
k! k=0
k!
n n
X f (k+1) (t) k
X f (k) (t)
=− (x − t) + (x − t)k−1 .
k=0
k! k=1
(k − 1)!

Now we work with the second sum and manipulate it so that the index begins at 0. Set
j = k − 1. As k runs from 1 to n, j runs from 0 to n − 1. Substituting k = 1 + j gives
n n−1 (j+1)
′
X f (k+1) (t) k
X f (t)
S (t) = − (x − t) + (x − t)j.
k=0
k! j=0
j!

But now these two sums are the same except that the first one has an extra term. Consequently
most of the terms cancel and we are left with (6.12).
Because
n n
X f (k) (t) k (0)
X f (k) (t)
S(t) = f (x) − (x − t) = f (x) − f (t) − (x − t)k .
k=0
k! k=1
k!

Hence,
n
X f (k) (x)
S(x) = f (x) − f (x) − (x − x)k = 0,
k!
k=1

taking care of the initial condition. Now, for any function S,

Z x
x
S ′ (t) dt = S(t) a = S(x) − S(a).

a

But S(a) = Rna (x). Because S(x) = 0,

x
f (n+1) (t)
Z
Rna f (x) = (x − t)n dt.
a n!
So,
x
Z f (n+1) (t) 
|Rna f (x)| = n
(x − t) dt.
 
a n!
Now, since t ∈ [a, x], |f (n+1) (t)| ≤ M. So
x  Z x M(x − t)n 
Z f (n+1) (t) 
|Rna f (x)| = n
(x − t) dt ≤  dt
   
a n! a n!
h (x − t)n+1 ix  M|x − a|n+1
= M − = .
 
(n + 1)! a (n + 1)!

134
6.5 Extra curricular material: Estimating n for Taylor
Polynomials
In this section we look at the issue of determining n in our problems.

Example. Find n such that the Maclaurin polynomial Tn f (0.5) for f (x) = sin(x) would
approximate sin(0.5) with an error less than 10−5.
using the now familiar estimates for f , the error terms are bounded by
|x|n+1
|Rn f (x)| = .
(n + 1)!
Therefore we are looking to find n such that
1
(1/2)n+1 6 10−5.
(n + 1)!
Now, this is an inequality for n, that cannot be solved exactly. So, we have to guess and
check, more or less cleverly. We check only even values for n since there are no even terms in
sin(x)!
1
n=4: 5!
(.5)5 = 2.6 × 10−4,

1
n=6: 7!
(.5)7 = 1.55 × 10−6.
Thus T5 f (0.5) gives an error which is small enough.

Now we look at the most difficult question about accuracy:

Given a function, given a fixed point, given an interval around that fixed point, and given a re-
quired tolerance, find how many terms must be used in the Taylor expansion to approximate
the function to within the required tolerance on the given interval.

Example. For example, let’s get a Taylor polynomial approximation to f (x) = ex which is
within 0.001 on the interval [− 12 , + 21 ]. Recall that
x2 x3 xn ec
Tn f (x) = 1 + x + + + ...+ + xn+1
2! 3! n! (n + 1)!
for some ξ between 0 and x, and where we do not yet know what we want n to be. It is very
convenient here that the n-th derivative of ex is still just ex ! We are wanting to choose n
large-enough to guarantee that
eξ
| xn+1 | 6 0.001
(n + 1)!
for all x in that interval.
The error term is estimated as follows, by thinking about the worst-case scenario for the
sizes of the parts of that term: we know that the exponential function is increasing along the
whole real line, so in any event ξ lies in [− 21 , + 12 ] and

|eξ | 6 e1/2 6 2

135
(where we’ve not been too fussy about being accurate about how big the square root of e is!).
And for x in that interval we know that
1
|xn+1 | 6 ( )n+1
2
So we are wanting to choose n large-enough to guarantee that

eξ 1
| ( )n+1 | 6 0.001.
(n + 1)! 2

Since
eξ 1 2 1
| ( )n+1 | 6 ( )n+1 .
(n + 1)! 2 (n + 1)! 2
we can be confident of the desired inequality if we can be sure that
2 1
( )n+1 6 0.001.
(n + 1)! 2

That is, we want to ‘solve’ for n in the inequality

2 1
( )n+1 6 0.001.
(n + 1)! 2

There is no genuine formulaic way to ‘solve’ for n to accomplish this. Rather, we just evaluate
the left-hand side of the desired inequality for larger and larger values of n until (hopefully!)
we get something smaller than 0.001. So, trying n = 3, the expression is
2 1 1
( )3+1 =
(3 + 1)! 2 12 · 16

which is more like 0.01 than 0.001. So just try n = 4:

2 1 1
( )4+1 = 6 0.00052
(4 + 1)! 2 60 · 32

which is better than we need.

The conclusion is that we needed to take the Taylor polynomial of degree n = 4 to achieve
the desired tolerance along the whole interval indicated. Thus, the polynomial

x2 x3 x4
1+x+ + +
2 3 4
approximates ex to within 0.00052 for x in the interval [− 12 , 12 ].
Yes, such questions can easily become very difficult. And, as a reminder, there is no real or
genuine claim that this kind of approach to polynomial approximation is ‘the best’. It will
depend on what we want to do.

136
Summary of Chapter 5
We have seen:

• the definition of a series;

• the geometric and Harmonic series;

• how to sum telescoping sums;

• an example of a series where the terms tend to zero but the series diverges;

• three tests for determining whether a series is convergent;

• that we must check carefully that it is appropriate to use either of the tests.

137
6.6 Exercise Sheet 5

6.6.1 Exercise Sheet 5a

1. Let f (x) = cos(x), how well does T6 f approximate cos x on the intervals

(a) [−0.1, 0.1]?

(b) [−1, 1]?
(c) [− π2 , π2 ]?

2. (Computing the cube root of 9)√The cube root of 8√ is 2, and 9 is ‘only’ one more than 8.
So you could try to compute 3 9 by viewing it as 3 8 + 1.
√ √
(a) Let f (x) = 3 8 + x. Find T2 f , and estimate the error | 3 9 − T2 f (1)|.
(b) Repeat part (i) for T3 f and compare.
√
3. Follow the method of Question 2 to compute 10 within an error of 10−3 using a
polynomial approximation.
1 √
4. For what range of x > 25 does 5 + (x − 25) approximate x to within 0.001?
10
5. (a) Let f (t) = sin t. Compute T2 f and give an upper bound for R2 f for t ∈ [0, 0.5].
R 0.5
(b) Use part (i) to approximate 0 sin(x2 ) dx, and give an upper bound for the error
in your approximation.

6. Let f (t) = et.

(a) Find T2a f for a = 0, a = 1/2 and a = 1.

(b) Use each of the previous Taylor polynomials approximation of f to give estimates
for the integral Z 1
2
ex dx, (6.13)
0
and compare them.
(c) Suppose instead we used T5a f for the same values of a. Estimate the integral (6.13)
and give an upper bound for the errors.
R 0.1
7. (a) Approximate 0 arctan x dx and estimate the error in your approximation by an-
alyzing T2 f (t) and R2 f (t) where f (t) = arctan t.
R 0.1
(b) Estimate 0 arctan x dx with an error of less than 0.001.
R 0.1 2
8. Approximate 0 x2 e−x dx and estimate the error in your approximation by analyzing
T3 f (t) and R3 f (t) where f (t) = te−t.
R 0.5 √
9. Estimate 0 1 + x4 dx with an error of less than 10−4.

138
6.6.2 Exercise Sheet 5b
x3
1. For what range of values of x is x − within 0.01 of sin x?
6
2. Only consider −1 ≤ x ≤ 1. For what range of values of x inside this interval is the
polynomial 1 + x + x2 /2 within .01 of ex ?

3. On how large an interval around 0 is 1 − x within 0.01 of 1/(1 + x)?

x − 100 √
4. On how large an interval around 100 is 10 + within 0.01 of x?
20
5. Determine how many terms are needed in order to have the corresponding Taylor poly-
nomial approximate ex to within 0.001 on the interval [−1, +1].

6. Determine how many terms are needed in order to have the corresponding Taylor poly-
nomial approximate f (x) = cos x to within the tolerance ǫ on the interval [a, b] for the
following cases:

(a) ǫ = 0.001, [a, b] = [−1, +1];

π π
(b) ǫ = 0.001, [a, b] = [− , ];
2 2
(c) ǫ = 0.001, [a, b] = [−0.1, +0.1].
√
7. Approximate e to within 0.01 by using a Maclaurin polynomial with remainder term.
√
8. Approximate 101 to within 10−15 using a Taylor polynomial with remainder term.
Note: do NOT add up the finite sum you get! One point here is that most hand
calculators do not easily give 15 decimal places.

9. Find the 4-th degree Maclaurin polynomial for the function f (x) = sin x. Estimate the
error for |x| < 1.

10. Let f (x) = cos x.

(a) Find the fourth degree Maclaurin polynomial for f and estimate the error for
|x| < 1.
(b) Find the eighth degree Maclaurin polynomial for f and estimate the error for
|x| < 1.
(c) Now, find the ninth degree Maclaurin polynomial for f . What is the error for
|x| ≤ 1?

139
6.6.3 Miscellaneous Exercises
Question

1. If
Tn f (x) = a0 + a1 x + a2 x2 + · · · + an xn
is the Taylor polynomial of a function y = f (x), then what is the Taylor polynomial of
its derivative f ′ (x)?

Theorem 6.7. The Taylor polynomial of degree n − 1 of f ′ (x) is given by

Tn−1 {f ′(x)} = a1 + 2a2 x + · · · + nan xn−1 .

In other words, “the Taylor polynomial of the derivative is the derivative of the Taylor
polynomial.”

Proof. Let g(x) = f ′ (x). Then g (k) (0) = f (k+1) (0), so that

x2 xn−1
Tn−1 g(x) = g(0) + g ′ (0)x + g (2) (0) + · · · + g (n−1) (0)
2! (n − 1)!
2
x xn−1
= f ′ (0) + f (2) (0)x + f (3) (0) + · · · + f (n) (0) ($)
2! (n − 1)!

On the other hand, if Tn f (x) = a0 + a1 x + · · · + an xn , then ak = f (k) (0)/k!, so that

k (k) f (k) (0)

kak = f (0) = .
k! (k − 1)!

In other words,

f (3) (0)
1 · a1 = f ′ (0), 2a2 = f (2) (0), 3a3 = , etc.
2!
So, continuing from ($) you find that

Tn−1 {f ′(x)} = Tn−1 g(x) = a1 + 2a2 x + · · · + nan xn−1

as claimed.

2. For the following functions, find an approximation to f by a Taylor polynomial of degree

n about a number a.
Use Taylor’s Theorem to estimate the accuracy of the approximation and compare with
the exact value.
x4
(a) f (x) = x sin x, a = 0, n = 4, x = 1; The Taylor polynomial is x2 − .
6
When x = 1, the approximation to x sin x given by the Taylor polynomial is equal to
5
≈ 0.833. However the actual value of 1 × sin 1 is approximately 0.841 giving an error
6
of approximately 0.008.

140
 6
A bound on the error is = 0.05. We see that the actual error is much smaller than
120
this.
(full) To compute the Taylor polynomial of degree 4 and a bound on the remainder, we
will need the first 5 derivatives of x sin x. Let f (x) = x sin x. Then

f (1) (x) = sin x + x cos x, f (2) (x) = 2 cos x − x sin x, f (3) (x) = −3 sin x − x cos x,
f (4) (x) = −4 cos x + x sin x, f (5) (x) = 5 sin x + x cos x.

So the Taylor polynomial is

4
X f (n) (0) x4
xn = x2 − .
n=0
n! 6

When x = 1, the approximation to x sin x given by the Taylor polynomial is equal to

5
≈ 0.833. However the actual value of 1 × sin 1 is approximately 0.841 giving an error
6
of approximately 0.008.
M 5
The bound on the error given by Taylor’s Theorem is 1 where M is the absolute
5!
(5)
maximum of |f (t)| for t ∈ [0, 1]. This value is not easy to find exactly, because it is
not easy to maximise 5 sin x + x cos x. However since | sin x| ≤ 1 and | cos x| ≤ 1 for all
6
x, we know that M is at most 6. So a bound on the error is = 0.05. We see that
120
the actual error is much smaller than this.
2
(b) f (x) = ex , a = 0, n = 3, x = 0.5.
2
The Taylor polynomial is 1 + x2 . When x = 0.5, the approximation to ex given by the
5 2
Taylor polynomial is equal to = 1.25. However the actual value of e0.5 is approxi-
4
mately 1.284 giving an error of approximately 0.034.
32.101
A bound on the error is (0.5)4 ≈ 0.084. Of course, the actual error is no larger
24
than this.
(full) To compute the Taylor polynomial of degree 3 and a bound on the remainder, we
2 2
will need the first 4 derivatives of ex . Let f (x) = ex . Then
2 2
f (1) (x) = 2xex , f (2) (x) = (2 + 4x2 )ex ,
2 2
f (3) (x) = (12x + 8x3 )ex , f (4) (x) = (12 + 48x2 + 16x4 )ex .

So the Taylor polynomial is

3
X f (n) (0)
xn = 1 + x2 .
n=0
n!
2
When x = 0.5, the approximation to ex given by the Taylor polynomial is equal to
5 2
= 1.25. However the actual value of e0.5 is approximately 1.284 giving an error of
4
approximately 0.034.

141
 M
The bound on the error given by Taylor’s Theorem is (0.5)4 where M is the absolute
4!
maximum of |f (4) (t)| for t ∈ [0, 0.5]. Since f (4) is increasing on [0, ∞), we have M =
32.101
f (4) (0.5) ≈ 32.101. So a bound on the error is (0.5)4 ≈ 0.084. Of course, the
24
actual error is no larger than this.

Exercise 6.8. For the following functions, find an approximation to f by a Taylor polyno-
mial of degree n about a number a. Use Taylor’s Theorem to estimate the accuracy of the
approximation and compare with the exact value.

(a) f (x) = sin x, a = π/6, n = 4, x = π/3; (b) f (x) = ln(1 + x), a = 1, n = 3, x = 2.

Exercise 6.9. Determine the nth degree Taylor polynomial for ex about x = 0 and write down
an expression for the remainder.
1
Use this to determine n so that the nth degree Taylor polynomial for ex evaluated at x =
√ 2
gives an approximation to e for which the error is at most 0.0001.
Compute the approximation and check with the correct answer.

142
Short Feedback for Exercise Sheet 5a

1. (a) The error is smaller than 2.5 · 10−13.

(b) The error is smaller than 2.5 · 10−5.
(c) The error is smaller than 10−3.
1 1 2
2. (a) T2 f (x) = 2 + x− x and the error is smaller than 2.5 · 10−4.
12 288
1 1 2 5
(b) T3 f (x) = 2 + x − x + x3 and the error is smaller than 3.52 · 10−5.
12 288 10368
Compare with T2 f with improve by one digit.
√ 683 √
3. Take f (x) = 9 + x. Then T2 f (1) = approximate 10 within an error of 3 · 10−4 <
216
10−3 .

4. 25 ≤ x ≤ 26.
t3
5. (a) T2 f (t) = t and |R2 f (t)| ≤ for t ∈ [0, 0.5].
6
1
(b) The approximation of the integral is with an error of at most 1/5376 < 2 · 10−4.
24
t2 √ (t − 0.5)2
 
1/2
6. (a) T2 f (t) = 1 + t + , T2 f (t) = e 1 + (t − 0.5) + and T21 f (t) =
2 2
(t − 1)2
 
e 1 + (t − 1) + .
2
(b) Integrating over the interval [0, 1] the polynomial approximations, we find 1.4333,
1.4701 and the worse one 1.631.
(c) With T5 f , the estimate is 1.4625 with error less than 3 · 10−4.

7. (a) T2 f (t) = t, the integral is approximated by 5 · 10−3 and the error is less than
3.5 · 10−4.
(b) Nothing to do, the previous calculation is OK.

8. The value is 3.3134 · 10−4 with an error of less than 10−5.

9. Use T2 f , to find 0.5031.

143
6.6.4 Feedback for Exercise Sheet 5a
r8
1. The error on [−r, r] is bounded by .
8!
1
(a) Here, r = 1/10, and so the error is smaller than ≤ 2.5 10−13.
108 8!
1
(b) Here, r = 1, and so the error is smaller than ≤ 2.5 10−5.
8!
π8
(c) Here, r = π/2, and so the error is smaller than 8 ≤ 10−3.
2 8!
1 1
2. (a) The polynomial is T2 f (x) = 2 + x − x2. Thus, T2 f (1) ≈ 2.07986111. The
12 9 · 32
error is
√ 10 − 8 1
· 8 3 · < 2.5 · 10−4.
3
| 9 − T2 f (1)| ≤
27 3!
√3
(b) The 9 according to a computer is:
√
3
9 ≈ 2.08008382305.
√ √
3. (a) Use Taylor’s formula with f (x) = 9 + x, n = 1, to calculate 10 approximately.
The error is less than 1/216.
(b) Repeat with n = 2. The error is less than 0.0003.

4. One can easily check that the question ask for the larger
√ range 25 ≤ x ≤ c for which
T125 f approximate f within 0.001. Indeed, for f (x) = x, taking a = 25, we have
√ 1 1 1
T125 f (x) = f (a) + f ′ (a)(x − a) = 25 + √ (x − 25) = 5 + (x − 25).
2 25 10

f ′′ (ξ)
The corresponding remainder term is (x − a)2 for some ξ between a and x. Ex-
2!
plicitly we have
11 1 1 1
R125 f (x) = − 3/2
(x − 25)2 = − 3/2 (x − 25)2,
2! 4 (c) 8ξ

where 25 ≤ ξ ≤ x. Because the three-halves-power function is increasing, we have

253/2 6 ξ 3/2 6 x3/2

Taking inverses (with positive numbers) reverses the inequalities: we have

25−3/2 > ξ −3/2 > x−3/2

So, in the worst-case scenario, the value of ξ −3/2 is at most 25−3/2 = 1/125. Taking
the absolute value for the remainder term (in order to talk about error),
1 1 1 1
|R1 f 25 f (x)| = | 3/2
(x − 25)2| ≤ | (x − 25)2 |.
8ξ 8 125

144
 To give a range of values of x for which we can be sure that the error is never larger
than 0.001 based upon our estimate, we solve the inequality
1 1
| (x − 25)2 | 6 0.001
8 125
(with x > 25). Multiplying out by the denominator of 8 · 125 gives

|x − 25|2 6 1

so the solution is 25 6 x 6 26.

5. (a) The Taylor series is

sin(t) = t − t3 /6 + · · ·
and the order one and two Taylor polynomial is the same p(t) = t. For any t there
is a ζ between 0 and t with

f (3) (ζ) 3
sin(t) − p(t) = t
3!
When f (t) = sin(t), |f (n) (ζ)| ≤ 1 for any n and ζ. Consequently,

t3
| sin(t) − p(t)| ≤
3!
for nonnegative t.
(b) Hence
Z 1 Z 1  Z 1
 2 2
 2
2 2
sin(x ) dx − p(x ) dx ≤ | sin(x2 ) − p(x2 )| dx
 

 0 0  0
Z 1 6
2 x (1/2)7
≤ dx = =ǫ
0 3! 3! 7
1
(1/2)3
Z
2
Since p(x2 ) dx = = A (the approximate value) we have that
0 3
Z 1
2
A−ǫ≤ sin(x2 ) dx ≤ A + ǫ
0

t2 1/2
6. (a) Because all the derivatives of f are equal to f , T2 f (t) = 1 + t + , T2 f (t) =
2 
√ (t − 0.5)2 (t − 1)2
  
1
e 1 + (t − 0.5) + and T2 f (t) = e 1 + (t − 1) + .
2 2
x4 √ 5 x2 x4
 
x2 2
(b) The respective polynomial estimates for e are 1+x + , e + + and
2 8 2 2
e
(1+x4 ), respectively. Integrating over the interval [0, 1] the polynomial estimates,
2
43
we find = 1.4333, 1.4701 and the worse one 1.631.
30

145
 R1
(c) Note: You need not find p(t) or the integral 0
p(x2 ) dx. The error for T5 f is
3
smaller than .
6! · 13
7. (a)
(b)

8.

9.

Short Feedback for Exercise Sheet 5b

1.

For Question 10a:

The derivatives of f satisfy f (x) = f (4) (x) = cos x, f (1) (x) = f (5) (x) = − sin x, f (2) (x) =
− cos x and f (3) (x) = sin x. And so f (0) = f (4) (0) = 1, f (1) (0) = f (3) (0) = 0, f (2) (0) = −1
and hence the fourth degree Maclaurin polynomial of f is

x2 x4
T4 f (x) = 1 − + .
2! 4!
The error is
f (5) (ξ) 5 (− sin ξ) 5
R4 f (x) = x = x
5! 5!
for some ξ between 0 and x. As | sin ξ| 6 1, we have

|x5 | 1
|R4 (x)| 6 <
5! 5!
for |x| < 1. Remark that since the fourth and fifth order Maclaurin polynomial for f are the
same, R4 (x) = R5 (x). It follows that 6!1 is also an upperbound.

146
Chapter 7

Power and Taylor Series

7.1 Lecture 19: About Taylor and Maclaurin Series

Let I ⊆ R be an open interval and f : I → R be infinitely differentiable at x = a ∈ I. We
can construct the Taylor polynomials Tna f of f of any degree n.

Definition 7.1. Let I ⊂ R be an open interval and f : I → R be an infinitely differentiable

a
function at a ∈ I. The Taylor SERIES T∞ f at a point a of f is the series
∞
a
X f (n) (a)
T∞ f (x) = (x − a)n. (7.1)
n=0
n!

When a = 0, we denote the Taylor series by T∞ f and call it the Maclaurin series at a
point a of f .

7.1.1 Some Special Examples of Maclaurin Series

We now list expansions you should be aware of. These Maclaurin series can be computed
directly from the definition by repeatedly differentiating the rule:
∞
x
X xn x2 x3
T∞ e = = 1+x+ + + ...,
n=0
n! 2 6
∞
X x2n+1 x3 x5
T∞ sin(x) = (−1)n = x− + + ...,
n=0
(2n + 1)! 6 120
∞
X x2n x2 x4
T∞ cos(x) = (−1)n =1− + + ...,
n=0
(2n)! 2 24
∞
1 X
T∞ ( )= xn = 1 + x + x2 + x3 . . . , (Geometric Series),
1−x n=0
∞
X xn x2 x3
T∞ (ln(1 + x)) = (−1)n+1 =x− + + ....
n=1
n 2 3

147
Another function whose Maclaurin polynomial/series you should know is f (x) = (1 + x)α ,
where α ∈ R is a constant. We have already seen it at Level 1. You can compute Tn f (x)
directly from the definition, and, when you do this, you find

α(α − 1) 2 α(α − 1)(α − 2) 3

Tn f (x) = 1 + αx + x + x
1·2 1·2·3
α(α − 1) · · · (α − n + 1) n
+···+ x . (7.2)
1 · 2···n
This formula is called Newton’s binomial formula. The coefficient of xn is called a bino-
mial coefficient, and it is written
 
α α(α − 1) · · · (α − n + 1)
= . (7.3)
n n!

When α is an integer, αn is also called α choose n. When α is a positive integer, the

binomial coefficients are just the numbers in Pascal’s triangle. When α = −1, the binomial
formula is the Geometric Series.

7.1.2 Convergence Issues about Taylor Series

a
We have created infinite series T∞ f . And so, we have usual mathematical questions.

a
1. What sense can we make of the series T∞ f?
a
2. What is the relation between the series T∞ f and the limit of Tna f as n → ∞?

3. When they exist, what are the properties of the series and the limit?

To approach those questions, we first remind you of the error estimates for Taylor polynomials.

Reminder about Taylor Polynomials

Recall that, for x ∈ [a, b], we call Rna (x) = f (x) − Tna f (x), the remainder of order n at
x = a. We had the Lagrange formula for the remainder when f has n + 1 continuous
derivatives:
f (n+1) (ξ)
Rna f (x) = (x − a)n+1
(n + 1)!
for some ξ between a and x (ξ depending on x). We also derived an alternative formula:
x
f (n+1) (t)
Z
Rna (x) = (x − t)n dt.
a n!
When we work for many x simultaneously, the following estimate for the remainder will prove
useful:
 a  M|b − a|n+1
 
Rn (x) ≤ , ∀x ∈ [a, b],
(n + 1)!
where |f (n+1) (x)| ≤ M for x ∈ [a, b].

148
Convergence May or May Not Occur

We can now look at an explicit example to see that our initial questions are issues that are
not necessarily easy to ignore.

Example. ConsiderP f : nR\{1} → R, f (x) = 1/(1 − x). From its definition, the Maclaurin
expansion of f is ∞ n=0 x and, from the theory of convergence of series, such series converges
for |x| < 1. From the formula for sum of geometric progressions we have

1 1 − xn+1 + xn+1
f (x) = =
1−x 1−x
xn+1 xn+1
= 1 + x + x2 + · · · + xn + = Tn f (x) + .
1−x 1−x
The remainder term is
xn+1
Rn f (x) = ,
1−x
and, when |x| < 1, we have

|x|n+1 limn→∞ |x|n+1 0

lim |Rn f (x)| = lim = = = 0.
n→∞ n→∞ |1 − x| |1 − x| |1 − x|

Thus we have shown that

1. the Maclaurin series of f converges for all −1 < x < 1, but not for |x| ≥ 1, even if the
function f is defined for any other point but x = 1,

2. the remainder term Rn f (x) valued at any |x| < 1 tends to 0 as n → ∞,

3. the Maclaurin series converges to f on (−1, 1).

So, for x ∈ (−1, 1), we can ignore the Taylor series notation and write with confidence
∞
1 X
= xn.
1 − x n=0

7.1.3 Valid Taylor Series Expansions

From the previous example, we see that the following questions are worth asking. Let I ⊂ R
be an open interval, let f : I → R be a function, infinitely differentiable function at a ∈ I.

a
1. For which x ∈ I does T∞ f (x) exist (i.e. converge as a series in n, x fixed)?
a
2. For which x ∈ I does T∞ f (x) converge to f (x)?

3. What are the relations between f , f ′ and f and T∞ f , T∞ f ′ and T∞ f ?

R R

149
Remarks 7.2. 1. Examples later on will show that we can get converging Taylor series
expansion whose sum is not equal to the original function. However this is a fairly rare
occurrence for our examples and exercises.
2. At x = a, the Taylor series is always equal to f (a).
3. For many functions, we shall see that their Taylor series converges to f (x) for x in some
interval a − R < x < a + R around a.
Definition 7.3. We shall then say that a function f has a valid power series expansion
about a if limn→∞ Rna (x) = 0 for x in a neighbourhood of a, that is, when T∞
a
(x) = f (x).

There is a simple example showing that everything we can hope for can indeed occur.
Proposition 7.4. The exponential function f (x) = ex has a valid Maclaurin expansion,
converging for all x ∈ R.

∞
x
X xn
Proof. First, remember that T∞ e = . To prove that this really is a power series
n=0
n!
expansion of ex , valid for all x, we need to prove that lim Rn (x) = 0. Our aim will be to use
n→∞
the Sandwich Theorem to show that
lim |Rn (x)| = 0.
n→∞

We know that 0 ≤ |Rn (x)| so all we need to do is find an upper bound for |Rn (x)|.
Taylor’s Theorem says that
Mxn+1
|Rn (x)| ≤ ,
(n + 1)!
where M is the absolute maximum of |f (n+1) (t)| on the interval [0, x] or [x, 0] (depending on
the sign of x). Now, f (n+1) (t) = et ≤ e|t| , so the absolute maximum of e|t| on the interval [0, x]
e|x| xn+1
or [x, 0] is e|x| . Therefore |Rn (x)| ≤ .
(n + 1)!
xn+1 n xn+1 o∞
We claim that lim = 0. The quotient of successive terms in the sequence
n→∞ (n + 1)! (n + 1)! n=0
x
is , which means that the sequence in n is initially increasing, then its tail is monoton-
n+1
ically decreasing for n > x. The sequence is bounded below, so it is a simple exercise to show
that its limit must be 0 (see Exercise 4 of Sheet 6b). Hence,
e|x| xn+1
lim = lim e|x| lim xn+1 (n + 1)! = e|x| · 0 = 0.
n→∞ (n + 1)! n→∞ n→∞

So, using the Sandwich Theorem, lim Rn (x) = 0.

n→∞

Because f = f ′ =
R
f on R when f is the exponential function, we get the following result.
Corollary 7.5. Let f : R → R, defined as f (x) = ex. Then T∞ f , T∞ f ′ and T∞ ( f ) are
R

all valid on R and we obtain each of them from the other by differentiating or integrating the
relevant series term per term.

150
7.2 Lecture 20: Power Series, Radius of Convergence
Taylor or MacLaurin series belong to a special type of series called power series. These are
incredibly important in mathematics, both theoretically and for doing numerical calculations.

Definition 7.6. A power series is a series of the form

∞
X
cn (x − a)n,
n=0

where {cn }∞
n=0 is a sequence of real numbers.

We think of a and c0 , . . . P
, cn . . . as being constants and x as a variable. So, a power series is
a function rule f (x) = ∞ n=0 cn (x − a) .
n

Remark 7.7. Often we work with the special case where a = 0, i.e. we use power series of
the form
∞
X
cn xn.
n=0

To define a proper function, we need to define the domain of the power series. The natural
domain of a power series is the set of values of x for which f (x) converges. We can use
the Root or the Ratio Tests to check such convergence.
For the geometric series ∞ n
P
n=0 x , it was easy to find the domain and it had a simple answer:
(−1, 1). The following theorem shows that the domain of a power series, or equivalently the
values of x for which it converges, always has this form. Before, let start with a definition and
a lemma to indicate that the definition makes sense.
∞
X
Definition 7.8. Let cn (x − a)n be a power series and let
n=0
 
 cn+1 
ρ1 = lim  , and ρ2 = lim |cn |1/n. (7.4)
n→∞ cn  n→∞

The radius of convergence R of the series is equal to

1. R = 1/ρ2 when 0 < ρ2 < ∞,

2. R = 0 when ρ2 = ∞,

3. R = ∞ when ρ2 = 0.

The radius of convergence is defined from ρ2 that comes from the Root Test because it applies
to slightly more cases (see Exercise 10 of Sheet 6a), but they are in general equal.

Lemma 7.9. In (7.4), when ρ1 and ρ2 exist, they are equal.

Proof. This is quite tricky — it is not examinable.

151
Then, the domain of f is determined by its radius of convergence.
∞
X
Theorem 7.10 (Pointwise convergence of a power series). Given a power series cn (x − a)n,
n=0
either:
1. the series converges for all x when R = ∞;
2. the series converges just for x = a when R = 0;
3. the series converges for all x with |x − a| < R and we need to check what happens for
x = a + R and x = a − R.

Proof. Apply the Root Test to the power series. We need that ρ2 · |x| < 1 for convergence or
ρ2 · |x| > 1 for divergence. The proof is completed in Exercise 8 of Sheet 6a.

Examples.
∞
X (x − 1)n
1. To find the radius of convergence of the power series , we need to look at
n=1
n2n
1
cn = . We evaluate
n2n
1 1
ρ2 = lim |cn |1/n = lim n−1/n = ,
n→∞ 2 n→∞ 2
and so the radius of convergence is R = 1/ρ2 = 2. To test Lemma 7.9,
|cn+1 | n2n (n) 1
ρ1 = lim = lim n+1
= lim = = ρ2 .
n→∞ |cn | n→∞ (n + 1)2 n→∞ 2(n + 1) 2
So, the power series converges when |x − 1| < 2 and diverges when |x − 1| > 2. To
see what happens at the radius of convergence, we replace x = 3 and x = −1 in the
∞
X 1
series. In the first case, we get the divergent harmonic series , in the second
n=1
n
∞
X (−1)n
case alternating (convergent) harmonic series . Therefore the series is
n=1
n
well defined on [−1, 3).
∞
X xn
2. As we know, the power series converges for all x ∈ R. Indeed, its radius of
n=0
n!
convergence is R = ∞ because
|cn+1 | n! 1
ρ1 = lim = lim = lim = 0.
n→∞ |cn | n→∞ (n + 1)! n→∞ n + 1

3. The power series ∞ n

P
n=0 n!x converges only for x = 0 because
|cn+1 | (n + 1)!
ρ1 = lim = lim = lim (n + 1) = ∞.
n→∞ |cn | n→∞ n! n→∞

The coefficient cn tends so fast to infinity that no small (but non zero) amount of x can
keep the series converging.

152
7.2.1 Behaviour at the Boundary of the Interval of Convergence
When 0 < R < ∞ exists, the power series converges when |x − a| < R and diverges when
|x − a| > R. When |x − a| = R, that is, x = a ± R, anything can happen. It needs to be looked
X∞
at individually. The power series becomes (−1)n cn Rn. In particular it is quite clear that
n=1
the theory of alternating series is important there. We shall see in the next example (7.5)
thereafter that any behaviour can be found at the boundaries of the interval of convergence.

Examples. Let p ∈ R be a fixed parameter. Consider the power series

∞
X (x − 1)n
. (7.5)
n=1
np 2n

What is its radius of convergence (depending on p)? We have:

|cn+1 | np 2n 1 1
ρ1 = lim = lim p n+1
= lim (1 + 1/n)p = ,
n→∞ |cn | n→∞ (n + 1) 2 2 n→∞ 2

and so R = 2 (independently of p). What happens at the boundary of the interval of conver-
gence? This will depend on p. We get the series
∞
X 1
n=1
np

at x = 3, and
∞
X (−1)n
n=1
np
at x = −1. And so, mostly using the Integral Test,

1. when p > 1, we have convergence on both sides of the interval, the power series converges
on [−1, 3] and diverges outside;

2. when 0 < p ≤ 1, we have convergence at x = −1 only because the alternating series

∞
X (−1)n
converges for 0 < p ≤ 1; thus, the power series converges on [−1, 3) and
n=1
np
diverges outside;

3. when p ≤ 0, convergence does not occur at the boundary. The power series converges
on (−1, 3) and diverges outside.

153
7.2.2 Elementary Calculus of Taylor Series
Inside the radius of convergence, we shall see that we can differentiate and integrate a power
series term per term. But first, we look at simpler properties of Taylor series.
Lemma 7.11.
P Let f : In → R be an infinitely differentiable function with Taylor series
a
T∞ f (x) = ∞
n=0 cn (x − a) about a ∈ I.

1. The sum and product of Taylor series about x = a are the Taylor series of the sum and
product functions.
a
2. For any k ∈ N, the Taylor series T∞ g of g(x) = (x − a)k f (x) about a ∈ I is given by
∞
X
a
T∞ g(x) = (x − a)k T∞
a
f (x) = cn (x − a)n+k.
n=0

3. Let b ∈ R, k ∈ N, the Maclaurin series for g(x) = f (bxk ) is

∞
X
T∞ g(x) = (cn bn ) xkn.
n=0

4. The Taylor series of the derivative f ′ of f is given by the series of term per term
derivatives of the Taylor series of f and has the same radius of convergence.

Proof. See Exercise 9 of Sheet 6a for the details.

1. This follows from the similar properties for the Taylor polynomials about x = a.
2. This is true by computing the derivatives of (x − a)k f (x) at x = a.
3. This is true by a simple substitution.
4. By definition of T∞ f ′ (x),
∞ ∞
f (n+1)
 (k) 
′
X
n
X d f k
T∞ f (x) = (x − a) = (x − a) .
n=0
n! k=0
dx k!

To get the radius of convergence of T∞ f we need to evaluate ρ2 (f ) = limn→∞ cn (f )1/n

(n)
where cn (f ) = f n!(a) . Then

n/(n−1)
ρ2 (f ′ ) = lim n1/(n−1) cn (f )1/n = lim n1/(n−1) · lim (ρ2 (f ))n/(n−1) = ρ2 (f ).
n→∞ n→∞ n→∞

∞
X (−1)n
Examples. The Maclaurin series of cosine is T∞ cos(x) = x2n . The Maclaurin
n=0
(2n)!
series for f1 (x) = cos(2x), f2 (x) = x cos x and f3 (x) = cos(4x2 ), respectively, are
∞ ∞ ∞
X 22n 2n X (−1)n X 24n 4n
(−1)n x , x2n+1, (−1)n x .
n=0
(2n)! n=0
(2n)! n=0
(2n)!

154
7.3 Lecture 21: More on Power and Taylor Series
in this lecture we are concerned with the ‘inverse’ problem for Taylor series. Given a power
series, what sort of function does it define?

Definition 7.12. Let I be an open interval, a ∈ I and a ∈ J ⊂ I, an open subinterval of I.

∞
X
Let g(x) = cn (x − a)n be a power series. If a function f : I → R is equal to the power
n=0
series f (x) = g(x) for x ∈ J ⊂ I, we say that g : J → R is a power series expansion of f
about x = a.

Remark 7.13. We shall see later that if f has a power series expansion about x = a then the
power series is unique and must be its Taylor series (7.1), or, in the special case where
a = 0, its MacLaurin series.

Example. The series ∞ n 2 3

P
n=0 (x + 1) = 1 + (x + 1) + (x + 1) + (x + 1) + . . . is a power series,
a geometric series with common ratio x + 1. Providing |x + 1| < 1, that is, x ∈ (−2, 0), it
1
converges to ! In other words the two functions f : (−2, 0) → R and g : (−2, 0) → R given
|x|
by
∞
1 X
f (x) = and g(x) = (1 + x)n,
|x| n=0

are the same, that is f = g.

Earlier on we saw that if we started with a known power series, e.g. the geometric series, then
if we could compute the sum, we could write down a simpler function to which it is equal.
On the other hand, suppose we have a function and want to try to write it as a power series.
How do we find the right coefficients of the power series in order to do this?
X∞
Suppose we know that f (x) = cn (x − a)n but we do not know the numbers c0 , c1 , . . . cn , . . ..
n=0
Together, the following two theorems help us to compute them.
∞
X
Theorem 7.14 (Differentiating a power series). Suppose that f (x) = cn (x − a)n is a power
n=0
series with radius of convergence R. If |x − a| < R then
∞
X
′
f (x) = ncn (x − a)n−1.
n=0

Proof. The proof will be sketched in the next lecture.

So given a power series for f (x) we can compute a related power series for f ′ (x).

155
Perhaps this theorem looks like it is not saying very much. One way to look at it, is that it
says
∞ ∞ 
d X n
 X d n

cn (x − a) = cn (x − a) .
dx n=0 n=0
dx
So we are interchanging the order of summation and differentiation. If a sum is finite, we
know from the basic rules of differentiation that, to differentiate the sum, we can differentiate
each term separately and add up the answers. So this switch would be valid for a finite sum,
i.e.
k k
d X n
 X
cn (x − a) = ncn (x − a)n−1.
dx n=0 n=0

However, we have an infinite sum, and infinite sums are much more complicated than finite
sums. As we have already seen, results about finite sums e.g commutativity, do not necessarily
carry over to infinite sums. So, this theorem is really saying quite a lot.
Both an infinite sum and differentiation are defined in terms of limits. So the theorem is
saying that we can switch around the order of these limits without affecting the answer. This
sort of thing requires careful justification and proof.
∞
X
Theorem 7.15. Suppose that f (x) = cn (x − a)n is a power series with radius of conver-
n=0
gence R > 0. Then,
f (n) (a)
cn = .
n!

Proof. We will prove this by first proving by induction that, providing |x − a| < R, we have
∞
X
(k)
f (x) = cn+k (n + 1) · · · (n + k)(x − a)n .
n=0

Clearly this is true when k = 0, because the statement when k = 0 is just the definition of f .
Suppose that the result is true whenever k ≤ j. Then providing |x − a| < R, we have
∞
X
(j)
f (x) = cn+j (n + 1) · · · (n + j)(x − a)n .
n=0

We can apply the previous theorem to the power series f (j) to get
∞
X
(j+1)
f (x) = cn+j n(n + 1) · · · (n + j)(x − a)n−1 .
n=0

Hence the result is also true when k = j + 1. So our claim follows by induction.
So substituting x = a gives
∞
X
(k)
f (a) = cn+k (n + 1) · · · (n + k)(a − a)n = ck k!.
n=0

This is true because all of the terms in the sum disappear except for the term when n = 0.
This is exactly what we want.

156
Finally, as an immediate corollary, Theorem 7.14 shows that, in the domain (a − R, a + R), a
power series is infinitely differentiable and defines a unique function.
P∞ n
P∞ n
Corollary 7.16. Suppose that two power series n=0 an (x − x0 ) and n=0 bn (x − x0 )
converge on some interval (x0 − r, x0 + r) to the same function, then an = bn , n ≥ 0, and they
are both infinitesimally differentiable.

Proof. By repeated differentiation,

n!an = f (n) (x0 ) = n!bn , n ≥ 0,

and the conclusion.

7.3.1 The Binomial Theorem

We now turn our attention to an important result we have already seen without proof.

Theorem 7.17 (The Binomial Theorem). If α is any real number and |x| < 1, then
∞
α
X α(α − 1) · · · (α − (n − 1))
(1 + x) = xn. (7.6)
n=0
n!

Remark 7.18. If α is a positive integer then this just reduces to a special case of the more
familiar version of the Binomial Theorem for positive integer powers.

Theorem 7.19. If m ∈ N then

m  m
m
X n n
X m(m − 1) · · · (m − (n − 1))
(1 + x) = x = xn.
n=0
m n=0
n!

To see that this is a special case of the previous theorem just notice that if n − 1 ≥ m, that is,
n ≥ m + 1, the terms in the series are all zero, so we only have a finite sum. Also, we do
not need the condition on |x| because the sum is finite and so we do not have to
worry about convergence.

α(α − 1) · · · (α − (n − 1)) n
Proof. Let cn = x . Then,
n!
cn+1 (n + 1)!α(α − 1) · · · (α − (n − 1)) n+1
R = lim = = lim = 1.
n→∞ cn n!α(α − 1) · · · (α − n) n→∞ n − α

Hence the series converges if |x| < 1 and diverges if |x| > 1. We will not worry about showing
what happens when |x| = 1.
To calculate the convergence we will need to show that cn corresponds to the general term of
the Maclaurin series of f (x) = (1 + x)α and that the remainder tends to 0 for every |x| < 1.

157
It is a straightforward calculation to see that f (n) (x) = α(α − 1) · · · (α − (n − 1)) · (1 + x)α−n.
Hence, f (n) (0) = α(α − 1) · · · (α − (n − 1)). And so,
f (n+1) (ξ) n+1 α(α − 1) · · · (α − n)
Rn f (x) = x = · (1 + ξ)(α−n) · xn.
(n + 1)! (n + 1)!
A close look at this reveals that we can expect trouble if x < 0 since then 1 + ξ can become
arbitrarily close to zero and for large n, we’ll have α − n < 0.
It turns out that a reasonably straightforward proof of convergence can be given if we insist
that 0 6 x < 1 but that the proof requires a lot more effort if −1 < x 6 0. We do not give
the details here.

Examples.
1
1. To find the MacLaurin series for the function f (x) = √ and the values of x for
9−x
which it converges, we proceed as follows. We have
1 1 1 x −1/2
√ =√ p = 1− .
9−x 9 1 − x/9 3 9
We can compute this  using
 the Binomial expansion which will result in a valid MacLaurin
 x
series expansion if −  < 1, that is, if −9 < x < 9. We get
9
1  x −1/2 1 
  1  x   1  3  x 2
1− = 1+ − − + − − − +···
3 9 3 2 9 2 2 9
(−1/2)(−3/2) · · · (−1/2 − (n − 1))  x n
+ − +···
n! 9
∞ ∞
X 1 · 3 · · · · (2n − 1) n X (2n)!
= n
x = n 2
xn.
n=0
(18) n! n=0
(36) (n!)
1
2. To find the MacLaurin series for (2 + 3x)− 2, we use the Binomial Theorem noting that
p = 1/2. Rewrite the expression as
 − 12
− 21 − 21 3
(2 + 3x) = 2 1+ x .
2
Then, from (7.6):
p(p − 1) 2 p(p − 1)(p − 2) 3
(1 + u)p = 1 + pu + u + u + ..., |u| < 1,
2! 3!
3x
so, on substituting u = ,
2
"
3
 2
3
5
 3 #
1  1 1


− − − − − −

1 1 3 3 3
(2 + 3x)− 2 = √ 1 + 2
x + 2 2
x + 2 2 2
x + ...
2 1 2 1 · 2 2 1 · 2 · 3 2
 
1 3 27 2 135 3
= √ 1− x+ x − x + ...
2 4 32 128
 
which is convergent for  3 x < 1 that is |x| < 2 .
2 3

158
7.4 Extra curricular material: General Theorem About
Taylor Series
In this lecture we collect together the information we have on power series and add some
additional properties we are going to prove next term in Analysis 2 (MA2731).

Theorem 7.20 (Power and Taylor’s series). Let {an }∞ n=1 be a sequence of real numbers such
1
that ρ = limn→∞ |an |1/n is defined. When ρ > 0, let R = and, when ρ = 0, let R = ∞. The
ρ
power series
∞
X
f (x) = ak (x − x0 )k
k=0

1. defines the values f (x) of a function f : (x0 − R, x0 + R) → R.

2. The function f is infinitely differentiable (in particular continuous) for x ∈ (x0 −R, x0 +
R).

3. The values f (x) are not defined for |x − x0 | > R. What happens to x = x0 ± R needs to
be investigated individually.

4. For any n ∈ N, the derivatives f (n) of f have also a representation as power series
with the same radius of convergence R. The coefficients of the power series of the n-th
derivative are obtained by differentiating n-times each of the terms of the power series
for f . The function f has the Taylor’s series representation
∞
X f (n) (x0 )
f (x) = (x − x0 )n, x ∈ (x0 − R, x0 + R).
n=0
n!

5. Let x0 − R < a < b < x0 + R, the integral

Z b
f (x) dx
a

of f on [a, b] has a representation in an absolutely convergent series obtained by inte-

grating each term of the power series defining f .

159
7.4.1 Examples of Power Series Revisited
Now we derive consequences of Theorem 7.20.

1. The geometric series is

∞
1 X
= 1 + x + x2 + · · · = xk, |x| < 1.
1−x k=0

for x0 = 0 and R = 1. Hence we can differentiate term-by-term and we can integrate

term-by-term to get the following series
∞
1 2
X
2 = 1 + 2x + 3x + · · · = kxk−1, |x| < 1,
(1 − x) k=1
Z x 2 ∞
dt x x3 X xk+1
− ln(1 − x) = =x+ + +··· = , |x| < 1.
0 1−t 2 3 k=0
k + 1

The results that have been covered say nothing about the convergence or divergence at
any point x with |x| = 1. In the case of the series for 1/(1 − x)2 the terms do not tend
to 0 when |x| = 1 and the series diverges for all |x| = 1. In the case of the series for
− ln(1 − x), the series is the harmonic series when x = 1 and hence diverges when x = 1.
Notice also that
lim− − ln(1 − x) = ∞.
x→1

When x = −1, the left hand side is − ln(2) and the right hand side is the alternating
harmonic series. We shall see in the next subsection that the series converges so
1 1 1
ln(2) = 1 − + − +··· .
2 3 4

2. The geometric series

1 2 4 6
2 = 1−x +x −x +··· ,
1+x
converges for |x| < 1. Now from x = tan y, y = arctan x and

dx
= sec2 y = 1 + tan2 y = 1 + x2,
dy
we get
dy 1
= .
dx 1 + x2
We can integrate term-by-term to obtain
Z x
dt x3 x5 x7
arctan x = 2 = x− + − +··· .
0 1+t 3 5 7
Note that arctan is defined for all x ∈ R, but this representation as a power series is
only true for |x| < 1.

160
 3. Theorem 7.20 does not make any claims about the convergence of the power series, or
its derivatives, at |x − x0 | = R. Indeed, anything can happen,

(a) no convergence at all,

(b) convergence at one point, or
(c) convergence at both end points.

Also, the power series may converge, but not its derivative etc. For instance,
the series ∞
X xn
n=1
n2
converges at both end-points x = ±1, but its derivative converges (conditionally) only
at x = −1.

7.4.2 Leibniz’ Formulas For ln 2 and π/4

Leibniz1 showed that
1 1 1 1 1
− + − + − · · · = ln 2
1 2 3 4 5
and
1 1 1 1 1 π
− + − + −··· =
1 3 5 7 9 4
Both formulas arise by setting x = 1 in the Maclaurin series for

x2 x3 x4
ln(1 + x) = x − + + −···
2 3 4
x3 x5 x7
arctan x = x − + + −···
3 5 7
This is only justified if you show that the series actually converge at x = 1, which we’ll do
here for the first of these two formulas. The proof for the second formula is similar. The
following is not Leibniz’ original proof. Begin with the geometric sum

1 (−1)n+1 xn+1
1 − x + x2 − x3 + · · · + (−1)n xn = +
1+x 1+x
Then you integrate both sides from x = 0 to x = 1 and get
Z 1 Z 1 n+1
1 1 1 n 1 dx n+1 x dx
− + − · · · + (−1) = + (−1)
1 2 3 n+1 0 1+x 0 1+x
Z 1 n+1
x dx
= ln 2 + (−1)n+1
0 1+x
1
G.W.L. Leibniz (1646-1716) was a German mathematician and philosopher who contributed to many
different areas of human knowledge. He is famous for his version of calculus and the dispute with Newton.
Many of the notation and ideas where R made explicit by Leibnitz. His formalism was to prove vital in latter
development. He used the notation , gave the product rule for differentiation and the derivative of xr for
integral and fractional r, but he never thought of the derivative as a limit. Another great achievements in
mathematics were the development of the binary system of arithmetic, his work on determinants which arose
from his developing methods to solve systems of linear equations.

161
Instead of computing the last integral, you estimate it by noting that

xn+1
0≤ ≤ xn+1
1+x
implying that
1 1
xn+1 dx 1
Z Z
0≤ ≤ xn+1 dx = .
0 1+x 0 n+2
Hence,
1 1
xn+1 dx xn+1 dx
Z Z
n+1
lim (−1) = (−1)n+1 lim = 0,
n→∞ 0 1+x n→∞ 0 1+x
and we get
1
1 1 1 1 xn+1 dx
Z
lim − + − · · · + (−1)n = ln 2 + lim (−1)n+1 = ln 2.
n→∞ 1 2 3 n+1 n→∞ 0 1+x

As a final example, we can try to construct the sum of the series

∞
X
kxk,
k=1

using directly Theorem 7.20 by manipulating the series to obtain derivatives or integrals of
known series. The radius of convergence is R = 1 and x0 = 0. For |x| < 1, we write
∞ ∞ ∞ ∞
X
k
X
k
X
k
X
′ x
kx = (k + 1)x − x = xk+1 −
k=1 k=1 k=1 k=1
x−1
∞
!′ ′
x2

X
k x x x
= x x − = − = .
k=1
x−1 1−x x−1 (1 − x)2

162
7.5 Extra curricular material: Taylor Series and Fi-
bonacci Numbers
In this lecture we shall look at examples of use of Taylor/Maclaurin series in different areas
of Mathematics.

7.5.1 Taylor Series in Number Theory

First, an interesting application of Taylor’s Theorem and the integral form of the remainder.
Theorem 7.21. The number e is irrational.

Proof. Recall that e is irrational means that e cannot be written as a fraction of two integers.
Let f (x) = ex . We know that
n
X 1
ex = + Rn f (x).
k=0
k!
Take x > 0. Then, the revisited Taylor’s Theorem 6.6 tells us that the integral form of Rn f
is
Z x (n+1) Z x t
f (t) e ex x
Z
n n
Rn f (x) = (x − t) dt = (x − t) dt ≤ (x − t)n dt
0 n! 0 n! n! 0
ex h (x − t)n+1 ix ex xn+1
= − = .
n! n+1 0 (n + 1)!
Now take x = 1 and an integer N > e. Then,
n
X 1
e= + Rn f (1),
k!
k=0

where
N
0 < Rn f (1) < .
(n + 1)!
p
Suppose, for contradiction, that e is rational. Let e = where p and q are positive integers.
q
Choose n so that n > q and n > 3. Then,
p 1 1 1
= 1+1+ + +···+ + Rn f (1).
q 2! 3! n!
So,
pn! n! n!
= n! + n! + + + · · · + 1 + n!Rn f (1).
q 2! 3!
Every term in this equation other than possibly n!Rn f (1) is an integer, so n!Rn f (1) must be
an integer. However,
N
0 < n!Rn f (1) < < 1.
n+1
So, n!Rn f (1) is not an integer after all, giving a contradiction.

163
7.5.2 Taylor’s Formula and Fibonacci Numbers
Now, consider the function rule
1
f (x) = .
1 − x − x2
How can we calculate T∞ f ? Using the definition seems complicated because the derivatives
of f will involve complicated rational functions:

1 + 2x 2(2 + 3x + 3x2 )
f ′ (x) = , f ′′ (x) = ...
(1 − x − x2 )2 (1 − x − x2 )3

What else could we do? An idea is the following. Let

T∞ f (x) = c0 + c1 x + c2 x2 + c3 x3 + · · ·

be its Taylor series. Considering the Taylor polynomials Tn f for large n, due to Lagrange’s
Remainder Theorem, we have, for any n,
1
= c0 + c1 x + c2 x2 + c3 x3 + · · · + cn xn + Rn f (x).
1 − x − x2
Multiply both sides with 1 − x − x2 , ignoring the terms of degree larger than n, you get

1 = (1 − x − x2 ) · (c0 + c1 x + c2 x2 + · · · + cn xn )
= c0 + c1 x + c2 x2 + · · · + cn xn
− c0 x − c1 x2 − · · · − cn−1 xn
− c0 x2 − · · · − cn−2 xn
= c0 + (c1 − c0 )x + (c2 − c1 − c0 )x2 + · · · + (cn − cn−1 − cn−2 )xn.

Compare the coefficients of powers xk on both sides for k = 0, 1, . . . , n and you find

c0 = 1, c1 − c0 = 0, c2 − c1 − c0 = 0, ··· , cn − cn−1 − cn−2 = 0.

Therefore the coefficients of the Taylor series T∞ f (x) satisfy the (second order) recursion
relation
cn = cn−1 + cn−2 , n ≥ 2, (7.7)
with initial data
c0 = 1 = c1 = 1. (7.8)

The numbers satisfying (7.7) and (7.8) are known in other aspects mathematics, they are the
Fibonacci numbers Fn with F0 = F1 = 1. We can calculate them:

F2 = F1 + F0 = 1 + 1 = 2,
F3 = F2 + F1 = 2 + 1 = 3,
F4 = F3 + F2 = 3 + 2 = 5,
etc.

The first time they appeared was in the model problem of calculating the growth in a popu-
lation of ‘rabbits’ such that each pair has exactly one pair of offsprings each year but cannot

164
reproduce for one year (as juveniles). Since, they appeared in other area of biology (in a more
realistic way).
Since it is much easier to compute the Fibonacci numbers one by one than it is to compute
the derivatives of f (x) = 1/(1 − x − x2 ), this is a better way to compute the Taylor series of
f (x) than just directly from the definition. The Fibonacci numbers are encoded into
the Taylor series of f (x) = 1/(1 − x − x2 ). This appears in area of higher mathematics, for
instance in Probability and Statistics. The term is to say that f is the generating function
for the Fibonacci numbers.

7.5.3 More about the Fibonacci Numbers

In this section we are going to see a way to compute the Taylor series of any any rational
function. As a consequence we shall be able to find an explicit formula for the Fibonacci
numbers (7.7).
You already know the trick: find the partial fraction decomposition of the given rational
function. Ignoring the case that you have quadratic expressions in the denominator, this lets
you represent your rational function as a sum of terms of the form
A
.
(x − a)p
These are easy to differentiate any number of times, and thus they allow you to write their
Taylor series. If we apply this idea to f (x) = 1/(1 − x − x2 ), from the previous example 7.5.2,
we get the following.
First, factor the denominator. We have

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

where x± are the roots of the quadratic polynomial

x2 + x − 1 = 0.

Using the general formula for the roots of a quadratic equation, we get,
√
−1 ± 5
x± = .
2
Define the golden ratio φ as
√
1+ 5
φ= ≈ 1.618 033 988 749 89 . . . .
2
Note that √ √
1 2 2( 5 − 1) 5−1
=√ = = .
φ 5+1 5−1 2
1
√ 1
And so, φ + φ
= 5 and φ − φ
= 1. Then,

1
x− = −φ, x+ = .
φ

165
So,
1
1 − x − x2 = −(x − )(x + φ).
φ
Hence, f (x) can be written as

1 −1 A B
f (x) = 2
= 1 = 1 + ,
1−x−x (x − φ )(x + φ) x− φ
x+φ

where
−1 −1 1 1
A= 1 =√ , B= 1 =√ .
φ
+φ 5 φ
+φ 5
Therefore,
1 φ 1/φ
f (x) = 2
=√ +√ .
1−x−x 5(1 − xφ) 5(1 + φx )
Using the Binomial Theorem, we find the expansions of (1 − xφ)−1 and (1 + φx )−1 as

1
= 1 + (xφ) + (xφ)2 + · · · + (xφ)n + · · ·
1 − xφ
1 x x x
x = 1 − + ( )2 + · · · + (−1)n ( )n + · · ·
1+ φ φ φ φ

Collecting the terms together, we find

(−1)n
 
1 n+1
Fn = √ φ + n+1 . (7.9)
5 φ

Note. We can check that (7.9) gives F0 and F1 correctly. For that, recall that φ2 = 1 + φ.
And so,
  √
1 1 5
F0 = √ φ + = √ = 1,
5 φ 5
 4
φ2 − 1
  
1 2 1 φ φ −1
F1 = √ φ − 2 = = = 1.
5 φ 1 + φ2 φ2 φ

Remark 7.22. As a final remark, there are some problems where the initial numbers are
f0 = 0 and f1 = 1. In that case we can see that fn = Fn−1 , n ≥ 1, and their generating
x
function is now .
1 − x − x2

166
7.6 Extra curricular Christmas treat: Series of Func-
tions Can Be Difficult Objects
In this appendix we look at two phenomena illustrating difficult behaviour what can happen
with series of functions.

1. In the first example, we show that a Taylor series for a function f about x = a may
converge everywhere but never to f (x) outside of x = a.

2. In a second example, we show that Fourier series can converge to bizarre function.
We show an explicit example where the limit is a continuous function, but nowhere
differentiable, hence impossible to draw.

7.6.1 What Can Go ‘Wrong’ with Taylor Approximation?

Let f be the function ( 2
e−1/x , if x 6= 0,
f (x) =
0, if x = 0.
We are going to show that the Maclaurin series of f is T∞ f (x) = 0 even if f (x) > 0 for x 6= 0.
Therefore, even if a Taylor series of some function f converges, you cannot be sure
that it converges to f , it could converge to a different function.

Proposition 7.23. The Maclaurin series of f is T∞ f (x) = 0.

Proof. We are going to show that all derivatives of f are 0 at the origin. We proceed in a few
steps.

1. We need to prove that for all n ∈ N, providing x 6= 0,

2
(n) p(x)e−1/x
f (x) =
xk
where k ∈ N and p is a polynomial, both depending on n. We will prove this by induction
on n.
2
(1) 2e−1/x
For x 6= 0, we have f (x) = , so the statement is true when n = 1.
x3
2
p(x) e−1/x
Suppose the statement is true when n = m, that is for all x 6= 0, f (m) (x) =
xk
for some k ∈ N and polynomial p. Then, providing x 6= 0,
2 2 2
d (m) p′ (x)e−1/x xk + 2p(x)/x3 · xk e−1/x − p(x)e−1/x kxk−1
f (m+1) (x) = f (x) =
dx x2k
3 ′ 2 −1/x2
(x p (x) + 2p(x) − kx p(x))e
= ,
xk+3
which is of the required form. Hence the statement follows by induction.

167
 2. Now show that f (n) (0) = 0 using the rigorous definition of a derivative. To evaluate the
2
e−1/x
limit, you will need to find lim where k ∈ N. This can be done by taking logs
x→0 xk
and then using L’Hôpital’s Rule.
We prove this by induction too. This time we will show that for all n ≥ 0, f (n) (0) = 0.
This is clearly true for n = 0. Suppose it is true for n = m. Then
2 2
(m+1) f (m) (x) p(x)e−1/x e−1/x
f (0) = lim = lim = lim p(x) lim .
x→0 x x→0 xk+1 x→0 x→0 xk+1

The first limit is finite and we will show that the second limit is 0 using L’Hôpital’s
Rule. This will show that the limit we want is equal to zero and establish the result by
induction.
2
e−1/x
It will be enough to show that lim k+1 = 0.
x→0 |x|

 e−1/x2  1 1 + (k + 1)x2 ln |x|

ln =− − (k + 1) ln |x| = − .
|x|k+1 x2 x2

We will show that lim x2 ln |x| = 0. This can be done with a standard application of
x→0
L’Hôpital’s rule by writing
ln |x|
x2 ln |x| = .
1/x2
Therefore lim (1 + (k + 1)x2 ln |x|) = 1 and so
x→0

 1  1 + (k + 1)x2 ln |x|
lim − 2 − (k + 1) ln |x| = − lim = −∞.
x→0 x x→0 x2
Because exp is continuous, if g(x) → −∞ as x → 0 then exp(g(x)) → 0 as x → 0 so
2
e−1/x  1 
lim lim exp − 2 − (k + 1) ln |x| = 0.
x→0 |x|k+1 x→0 x

3. We are told that f (0) = 0 and have proved that f (n) (0) = 0 for all n ∈ N. So, all
the coefficients of the MacLaurin series are 0. Consequently the MacLaurin series is
identically zero.

7.6.2 The Day That All Chemistry Stood Still

The rate at which a chemical reaction ‘A → B’ proceeds depends among other things on
the temperature at which the reaction is taking place. This dependence is described by the
Arrhenius law which states that the rate at which a reaction takes place is proportional to
∆E
A(T ) = e− kT

168
where ∆E is the amount of energy involved in each reaction, k is Boltzmann’s constant and
T is the temperature in degrees Kelvin. If you ignore the constants ∆E and k (i.e. if you set
them equal to one by choosing the right units) then the reaction rate is proportional to

f (T ) = e−1/T.

If you have to deal with reactions at low temperatures you might be inclined to replace this
function with its Taylor series at T = 0, or at least the first non-zero term in this series. If you
were to do this, you would be in for a surprise. To see what happens, look at the following
function, (
e−1/x, x > 0,
f (x) =
0, x 6 0.
This function goes to zero very quickly as x → 0. In fact one has, setting t = 1/x,

f (x) e−1/x
lim = lim = lim tn e−t = 0.
x→0 xn x→0 xn t→∞

As x → 0, this function vanishes faster than any power of x.

The Taylor
series at this
point does not
converge to f y = e−1/x

1 2 3

Figure 7.1: An innocent looking function with an unexpected Taylor series.

If you try to compute the Taylor series of f you need its derivatives at x = 0 of all orders.
These can be computed (see last weeks example), and the result turns out to be that all
derivatives of f vanish at x = 0,

f (0) = f ′ (0) = f ′′ (0) = f (3) (0) = · · · = 0.

The Taylor series of f is therefore

x2 x3
T∞ f (x) = 0 + 0 · x + 0 · +0· + · · · = 0.
2! 3!
Clearly this series converges (all terms are zero, after all), but instead of converging to the
function f we started with, it converges to the function g = 0.
What does this mean for the chemical reaction rates and Arrhenius’ law? We wanted to
‘simplify’ the Arrhenius law by computing the Taylor series of f (T ) at T = 0, but we have
just seen that all terms in this series are zero. Therefore replacing the Arrhenius reaction rate
by its Taylor series at T = 0 has the effect of setting all reaction rates equal to zero.

169
7.6.3 Series Can Define Bizarre Functions: Continuous but Nowhere
Differentiable Functions
As a final remark about series, Karl Weierstrass in 1872 considered the function
∞
X 1 j
f : R → R, f (x) = fj (x), where fj (x) = j cos 3 x. (7.10)
j=0
2

For all x, |fj (x)| 6 1/2j and, since the geometric series ∞ −j
P
j=0 2 converges, the series converges
uniformly on R and consequently defines a continuous function. The particularly interesting
property of f that Weierstrass proved (which will not be shown here) is that f is not differen-
tiable at any point. The function f is an example of a continuous but nowhere differentiable
function.
In Figure 7.2 we represent the finite sums n−6+ nj=1 1j cos 3j x that are translations upwards
P
2
(by an amount of n − 3) of the graphs so that we can compare them. Note that as n increases
the graphs become more and more jaggy.

-3 -2 -1 1 2 3
-1

-2

-3

-4

-5

Figure 7.2: Example (7.10) of a continuous nowhere differentiable real function

170
Summary of Chapter 6
We have seen:

• To be able to calculate the radius of convergence of a power series and to evaluate the
convergence at the boundary of the convergence interval.

171
7.7 Exercise Sheet 6

7.7.1 Exercise Sheet 6a

1. Find the MacLaurin series of the following function rules:
√
3
(a) 1 + 2t + t2 .
(b) e−t sin 2t.
(c) arcsin t
1
(d) √ .
1 − t2
sin(x)
(
(e) f (x) = , if x 6= 0,
x
1, if x = 0.
ln(1 + x)
(f) .
x
(g) e−t cos t.

2. Determine the values of x for which the following function rules has a valid power series
expansion about 0.

(a) cos x,
(b) sin x,
(c) ln(1 + x).

3. Determine the radius of convergence of the following power series.

P∞ k 3 xk
(a) .
k=0
3k
P∞ k k
(b) k=1 k x .
P∞ 2k xk
(c) k=0 k! .
P∞ k
(d) k=0 k!x .
P∞  k 2  k
(e) x.
k=0
5k
P∞  k  k
(f) x.
k=1
2k
P∞  1  k
(g) x.
k=1
kk
xn+1 ∞
4. Finish the proof of Proposition 7.4. Show that the sequence { } has limit 0.
(n + 1)! n=1
5. Determine the exact sums of the power series in Exercises 3(i), 3(iii), 3(v) and 3(vi).
Z 1
2
6. Use Maclaurin series to find an expression of the integral e−x dx as a series.
0

172
 ∞
X
7. Let cn (x − a)n be a power series. Prove that, when they exist,
n=0
 
 cn+1 
ρ1 = lim  , and ρ2 = lim |cn |1/n,
n→∞ cn  n→∞

are equal.

8. Complete the proof of Theorem 7.10.

9. Complete the proof of Lemma 7.11.

10. Show that we cannot apply the Ratio Test to the power series ∞ n
P
n=1 an x where an = 1
if n is odd, an = 2 when n is even, but we can apply the Root Test. Find its radius of
convergence.

Short Feedback for Exercise Sheet 6a

1. (a) Hint: use the Binomial Formula.

(b) Hint: Use de Moivre’s Formula (see Level 1)

eix = cos x + i sin x. (7.11)

(c) Hint: Differentiate arcsin t.

(d) Hint: use the Binomial Formula.
(e) Hint: use the calculus of Taylor series.
(f) Hint: use the calculus of Taylor series.
(g) Hint: Use (7.11).

2. (a) x ∈ R,
(b) x ∈ R,
(c) x ∈ (−1, 1).

3. (a) R = 3.
(b) R = 0.
(c) R = ∞.
(d) R = 0.
(e) R = 5.
(f) R = 2.
(g) R = ∞.

4. Use the convergence of a monotone sequence to establish the existence of a limit, then
calculate it.

173
 ∞
X y
5. 3(i) You may need ky k = .
k=1
(1 − y)2
2x
3(iii) e .
∞
X
3(v) You may need to calculate k2 y k .
k=1
2x
3(vi) .
(2 − x)2
1 1 1 1
6. 1 − + − + − ....
3 5 · 2 7 · 3! 9 · 4!
7.

8.

9.

10. R = 1.

174
7.7.2 Feedback for Exercise Sheet 6a
√ p
1. (a) 3
1 + 2t + t2 = 3 (1 + t)2 = (1 + t)2/3 . The answer follows from Newton’s binomial
formula. The Maclaurin series of (1 + t)2/3 is
∞  
X 2/3
tn.
n=0
n

(b) Let f (t) = e−t sin(2t). To simplify calculation we are going to use a ‘neat trick’: to
write sin(2t) as a complex exponential:

e2it − e−2it i −2it

− e2it .


sin(2t) = = e
2i 2
Combining with e−t, we find
i −(1+2i)t
− e(−1+2i)t .


f (t) = e
2
Therefore we can use the Maclaurin expansion for the exponential:
∞ ∞
iX tn iX tn
T∞ f = (−1)n (1 + 2i)n − (−1 + 2i)n .
2 n=0 n! 2 n=0 n!

To be able to calculate in a straightforward manner the powers of the complex

numbers, we write them in polar form
√ √
1 + 2i = 5 eiα, −1 + 2i = − 5 e−iα

where tan α = 2. We can now go back to T∞ f :

∞ √ n tn ∞  √ n tn
iX n iα iX n −iα
T∞ f (x) = (−1) 5e − (−1) − 5 e
2 n=0 n! 2 n=0 n!
∞
5n/2 inα n 5n/2 −inα n
 
iX
= (−1)n e t − (−1)n e t
2 n=0 n! n!
∞
X 5n/2 i inα
(−1)n e − e−inα tn


=
n=0
n! 2
∞ n/2
n5
X
= (−1) sin(nα) tn.
n=0
n!

1
(c) The derivative of arcsin t is √ . So you need to integrate the result (7.12) of
1 − t2
the next exercise. We find
∞
! ∞
(2n)! 2n (2n)! t2n+1
Z X X
T∞ arcsin(t) = t = .
n=0
4n (n!)2 n=0
4n (n!)2 2n + 1

175
(d) We use the Binomial Theorem or direct differentiation on g(z) = (1 − z)−1/2 , then
replace z by t2. We find that the Maclaurin series expansion of g is
∞ ∞
X 1 · 3 · · · (2n − 1) n
X (2n)! n
z = n (n!)2
z
n=0
2 · 4 · · · 2n n=0
4

multiplying top and bottom by the product of the first n even numbers:

2 · 4 · · · 2n = 2n (n!).

Therefore, the Maclaurin series we seek is

∞
X (2n)! 2n
n (n!)2
t . (7.12)
n=0
4

∞
X x2n+1
(e) Because T∞ sin(x) = (−1)n+1 , we have
n=1
2n + 1!
∞
X x2n
(−1)n+1 .
n=1
2n + 1!

(f) Recall that

1
(ln(1 + x))′ = .
1+x
1
is ∞ n n
P
Because the Maclaurin expansion of n=0 (−1) x , integrating, we get the
1+x
Maclaurin expansion of ln(1 + x) as
∞ ∞
xn+1
X Z X
n n
(−1) x dx = (−1)n .
n=0 n=0
n+1

Therefore we obtain ∞
X xn
(−1)n .
n=0
n+1

(g) We can differentiate f (t) = e−t cos t. We get

f ′ (t) = −e−t (cos t + sin t),

f ′′ (t) = −2e−t sin t,
f ′′′ (t) = 2e−t (cos t − sin t),
f (4) (x) = −4f (x).

And so f (0) = 1, f ′ (0) = −1, f ′′ (0) = 0, f ′′′ (0) = 2 and f (4) (0) = −4f (0) = −4,
hence, for n ≥ 0,

f 4n (0) = (−4)n,
f 4n+1 (0) = −(−4)n,
f 4n+2 (0) = 0,
f 4n+3 (0) = 2(−4)n.

176
 So the Maclaurin series expansion is
∞
X t4n t4n+1 t4n+3
T∞ f (t) = f 4n (0) + f 4n+1 (0) + f 4n+3 (0)
n=0
4n! (4n + 1)! (4n + 3)!
∞ 4n
t4n+1 t4n+3
 
nt
X
n n
= (−4) − (−4) + 2(−4)
n=0
4n! (4n + 1)! (4n + 3)!
∞ 4n
2t3
 
nt t
X
= (−4) 1− +
n=0
4n! 4n + 1 (4n + 1)(4n + 2)(4n + 3)
∞
t4n t3
X  
n
= (−4) 4n + 1 − t + .
n=0
4n + 1! (2n + 1)(4n + 3)

(n)
2. (a) The remainder term Rn (x) is equal to f n!(ζn ) xn for some ζn . For either the cosine
n
or sine and any n and ζ, we have |f (n) (ζ)| ≤ 1. So, |Rn (x)| ≤ |x|
n!
. But we know
|x|n
limn→∞ n! = 0 and hence limn→∞ Rn (x) = 0.
∞
X (−1)n 2n+1
(b) We know that the MacLaurin expansion for sin x is x . To test
n=0
(2n + 1)!
when this converges to sin x, we need to test if the remainder |Rn,0 (x)| converges
to 0. We will use the Sandwich Theorem to show this. From Taylor’s Theorem we
know that  M 
n+1 
|Rn,0 (x)| ≤  x ,

(n + 1)!
where M is the absolute maximum of |f (n+1) (t)| on the interval [0, x] or [x, 0]
(depending on the sign of x). But all the derivatives of sin x are either ± sin x or
± cos x and we have | sin x| ≤ 1 and | cos x| ≤ 1 for all x. Hence, M ≤ 1 and
consequently
1
|Rn,0(x)| ≤ |x|n+1 .
(n + 1)!
1
From Exercise 4 we know that lim |x|n+1 = 0. Consequently, by the Sand-
n→∞ (n + 1)!
wich Theorem, lim |Rn,0(x)| = 0.
n→∞
∞
X
(c) We know that if f (x) = ln(1 + x) has a power series expansion ln(1 + x) = cn xn
n=0
f (n) (0) (−1)n−1 (n − 1)!
then cn = . We know that f (n) (x) = and so f (n) (0) = (−1)n−1 (n − 1)!.
n! (1 + x)n
∞
X (−1)n−1 xn
Hence, if ln(1+x) has a power series expansion, then it must be ln(1 + x) = .
n=0
n
First we will apply the ratio test to the series to see when this power series converges.
Clearly it converges when x = 0 because all the terms are zero. So assume x 6= 0
(−1)n xn
and apply the ratio test. Let an = . Then
n
an+1 (−1)n+1 xn+1 /(n + 1) nx
= n n
=− .
an (−1) x /n n+1

177
 So, a 
 n+1  n
lim   = lim |x| = |x|.
n→∞ an n→∞ n+1
Hence the series converges if |x| < 1, diverges if |x| > 1 and may or may not
converge if |x| = 1.
So ln(1 + x) cannot have a valid power series expansion when |x| > 1. We will
not worry about what happens when |x| = 1 and will now prove that it has a
valid power series expansion when |x| < 1. We need to show that if |x| < 1,
lim Rn,0 (x) = 0.
n→∞
Our aim will be to use the Sandwich Theorem to show this. We know that 0 ≤
|Rn,0 (x)| so all we need to do is find an upper bound for |Rn,0 (x)|. Taylor’s Theorem
says that
Mxn+1
|Rn,0(x)| ≤ ,
(n + 1)!
where M is the absolute maximum of |f (n+1) (t)| on the interval [0, x] or [x, 0]
(depending on the sign of x). Now,
 (−1)n n!  n!
|f (n+1) (t)| =  ≤ .
 
(1 + t)n+1 (1 − |t|)n+1


So the absolute maximum of |f (n+1) (t)| on the interval [0, x] or [x, 0] is at most
n!
. So, for 0 ≤ x < 1, the absolute maximum of |f (n+1) (t)| on the
(1 − |x|)n+1
interval [0, x] is n! and we have

Mxn+1
|Rn,0 (x) ≤ .
n+1
So if 0 ≤ x < 1 then |Rn,0 (x)| → 0 as n → 0.
3
k3 3k+1

k
3. (a) The radius R is R = lim k · = 3 lim = 3.
k→∞ 3 (k + 1)3 k→∞ k + 1

(b) The radius R is

k
kk

k 1
R = lim k+1
= lim ·
k→∞ (k + 1) k→∞ k + 1 (k + 1)
 k
1 1
= lim · lim
k→∞ (k + 1) k→∞ 1 + 1/k
1  −1 1 1
= lim · lim (1 + 1/k)k = · lim = 0.
k→∞ (k + 1) k→∞ e k→∞ (k + 1)

2k k + 1! k+1
(c) The radius R is R = lim · k+1 = lim = +∞.
k→∞ k! 2 k→∞ 2
k! 1
(d) The radius R is R = lim = lim = 0.
k→∞ k + 1! k→∞ k + 1
2
k2 5k+1

k
(e) The radius R is R = lim k · = 5 lim = 5.
k→∞ 5 (k + 1)2 k→∞ k + 1

178
 k 2k+1 k
(f) The radius R is R = lim k
· = 2 lim = 2.
k→∞ 2 (k + 1) k→∞ k + 1

(g) The radius R is

(k + 1)(k+1)
R = lim = lim (k + 1) · (1 + 1/k)k = e · lim (k + 1) = +∞.
k→∞ kk k→∞ k→∞

4.

5.
2 3
6. We start with the series for ex : namely 1 + x + x2 + x3! + . . . . Replacing x with −x2, we
obtain
2 x4 x6 x8
e−x = 1 − x2 + − + − ...
2! 3! 4!
Now we integrate the series term per term:
R 1 −x2 R1 4 6 8
0
e dx = 0
1 − x2 + x2! − x3! + x4! − . . .
1
x3 x5 x7 x9
= x − 3 + 5·2! − 7·3! + 9·4! − . . . 

0
1
= 1 − 13 + 5·2 1
− 7·3! 1
+ 9·4! − ...

If we add the first three terms here we get .7666. As a rough idea of how accurate this is,
suppose we added the next term, this would change the result to .7429. This isn’t much
of a change. If we added one more term, this would change it even less, the convergence
is very slow indeed.

7.

179
7.7.3 Additional Exercise Sheet 6b
1.

Short Feedback for the Additional Exercise Sheet 6b

1.

180
7.7.4 Exercise Sheet 0c
1. Compute the Maclaurin series T∞ f for the following rules defined on the maximal in-
terval containing 0 where they are defined.

(a) f (t) = sinh t.

(b) f (t) = cosh t.
3
(c) f (t) = .
(2 − t)2
(d) f (t) = ln(2 + 2t).
t
(e) f (t) = .
1 − t2
(f) f (t) = sin t + cos t.
2
(g) f (t) = 1 + t2 − t7.
3
√3
(h) f (t) = 1 + t.
t4
2. Let f (t) = 1+4t2
, what is f (10) (0)?

3. Let f : R → R defined by f (t) = tan t.

(a) What are the first 5 terms of T∞ f ?

(b) Can you find a general formula for the n-th term of T∞ f ?

Short Feedback for Exercise Sheet 0c

∞
X 1
1. (a) T∞ f (t) = t2k+1.
k=0
(2k + 1)!
∞
X 1 2k
(b) T∞ f (t) = t .
k=0
(2k)!
∞
X 3 · (n + 1) n
(c) T∞ f (t) = t.
n=0
2n+2
∞
X (−1)n+1 n
(d) T∞ f (t) = t.
n=1
n
P∞ 2k+1
(e) T∞ f (t) = k=0 t .
∞ 4m
t4m+1 t4m+2 t4m+3
 
X t
(f) T∞ f (t) = + − − .
m=0
(4m)! (4m + 1)! (4m + 2)! (4m + 3)!
2
(g) T∞ f (t) = 1 + t2 − t7.
3
∞
X (1/3)(1/3 − 1)(1/3 − 2) · · · (1/3 − n + 1)
(h) Use the Binomial Theorem, T∞ f (t) = 1+ tn.
n=1
n!

2. f (10) (0) = 64(10!).

181
3. (a) T5 f (t) = t + 31 t3 + 2 5
15
t.
(b) There is no simple general formula for the nth term in the Taylor series for f .

182
7.7.5 Feedback for Exercise Sheet 0c
∞
X 1
1. (a) T∞ f (t) = t2k+1.
k=0
(2k + 1)!
∞
X 1 2k
(b) T∞ f (t) = t .
k=0
(2k)!
∞
X 3 · (n + 1) n
(c) T∞ f (t) = t.
2n+2
k=0
∞
X (−1)n+1
(d) T∞ f (t) = tn.
k=1
n
t
(e) 1−t2
h i
t 1/2 1/2
T∞ 1−t 2 = T∞ 1−t
− 1+t
= t + t3 + t5 + · · · + t2k+1 + · · ·
(f) sin t + cos t
The pattern for the nth derivative repeats every time you increase n by 4. So we
indicate the the general terms for n = 4m, 4m + 1, 4m + 2 and 4m + 3:

1 1 1 t4m t4m+1 t4m+2 t4m+3

T∞ (sin t + cos t) = 1+t− t2 − t3 + t4 +· · ·+ + − − +· · ·
2! 3! 4! (4m)! (4m + 1)! (4m + 2)! (4m + 3)!

(g) 1 + t2 − 32 t7
A: T∞ 1 + t2 − 23 t7 = 1 + t2 − 32 t7
 
√
(h) 3 1 + t
√
A: T∞ 3 1 + t = 1 + 1/3
1!
t + (1/3)(1/3−1)
2!
t2 + · · · + (1/3)(1/3−1)(1/3−2)···(1/3−n+1) n
n!
t +···
x4
2. f (x) = 1+4x2
, what is f (10) (0)?
6
A: 10! · 2

3. tan t (3 terms only)

A: T3 tan t = t + 13 t3 . There is no simple general formula for the nth term in the Taylor
series for tan x.

183
Index

α choose n, 148 remainder term, 16, 125

Rolle’s Theorem, 11
Rna f , 7, 16, 125 rule (of a function), 11
T∞ f , 147
a
T∞ f , 147 sequence
a
Tn f , 6, 12 (ǫ, N)-definition of limit, 8
sequences, 7
alternating series, 8 bounded and monotone, 7
binomial coefficient, 148 limit, 7
series, 80
co-domain (of a function), 11 alternating, 8
critical point, 7 convergent, 80
divergent, 80
degree geometric, 81
polynomial, 17 common ratio, 81
differentiation sum, 81
Chain Rule, 6 Maclaurin, 147
critical point, 7 partial sum, 80
Product Rule, 6 power, 9
Quotient Rule, 6 sum, 80
domain (of a function), 11 Taylor, 9, 147
factorial, 11 Taylor polynomial, 12
function Taylor polynomials, 6
co-domain, 11 Taylor series, 9, 147
domain, 11 Theorem
rule, 11 Estimate of the Remainder Term, 128
Mean Value, 11
geometric progression, 125
Rolle’s, 11
Lagrange Remainder Formula, 125 Taylor’s, 125

Maclaurin polynomial, 12
Maclaurin series, 147
Mean Value Theorem, 11

Newton’s binomial formula, 148

polynomial
degree, 17
Maclaurin, 12
Taylor, 12
power series, 9

184